University of Southern California Dissertations and Theses

Comparative genomics of translational regulation

(USC Thesis Other)

Comparative genomics of translational regulation

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Comparative genomics of translational regulation by Saket Choudhary A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (COMPUTATIONAL BIOLOGY AND BIOINFORMATICS) August 2020 Copyright 2020 Saket Choudhary Dedication To my family and teachers One important idea is that science is a means whereby learning is achieved, not by mere theoretical speculation on the one hand, nor by the undirected accumulation of practical facts on the other, but rather by a motivated iteration between theory and practice. George Box ii Acknowledgements My journey through graduate school would not have been possible without the support of some wonderful people around me. I feel I have been privileged in multiple aspects, but of everything I have been specially privileged to have some really smart and caring people around me. I would first like to thank my advisor Dr. Andrew Smith for agreeing to serve as my advisor and giving me a lot of freedom to pursue my ideas. His questions have helped me shape up as a better scientist and I have picked up some really great ideas from him in the process. I would also like to thank my committee members, Dr. Remo Rohs, Dr. Mark Chaisson , Dr. Peter Calabrese, Dr. Zhiping Lu and Dr. Adam Maclean. Remo and Mark, in particular, were great mentors who helped me with their advice everytime I was stuck. I felt privileged reaching out to them whenever I felt like and I will forever be grateful to them . I would also like to specially thank Dr. Peter Calabrese for the wonderful time I had assisting him in teaching BISC-305 - I learned a lot about how to teach students engagingly and had lot of fun doing so. I was very fortunate to work with two great collaborators - Dr. David Kadosh and Dr. Luiz Penalva at UT Health San Antonio. A small project with Dr. Luiz Penalva played a fundamental role in shaping the rest of my PhD so I will forever be grateful to him. Working with Dr. David Kadosh has been one of the most enlightening parts of my PhD - I came across a number of iii interesting biological problems that required development of novel computational methods that eventually formed the core part of my PhD. I will also be grateful to David for the time he spent advising me on life after PhD and specially during my postdoc search . I was also fortunate to have a great set of lab members and colleagues. I would especially like to thank Dr. Benjamin Decato and Amal Thomas for being both a critical lab member and a great friend . Our discussions about Science and life otherwise have been really insightful. My colleague, lab mate, and friend Dr. Wenzheng Li played a great role in shaping up my research. Our way of challenging each other over Science and our discussions lead to a very productive research life. My special thanks to Dr. Meng Zhou for being a great office mate. I am also grateful to my lab member and friend Dr. Rishvanth Prabhakar for his critical reviews and for fun discussions over the numerous Friday night dinners. I am grateful to Guilherme and Liz for their critical reviews on my writings. I have been fortunate to have an extended family here in LA. Some relationships go beyond friendship and the only way to do justice to them is to leave them unnamed. Satish, Swati, and Hiteshi have been an integral part of my journey from the beginning. I cannot imagine myself living so far away from family had it not been the support of these three particularly and I will run out of words if I start thanking them in words - my heart fills with gratitude whenever I think about them . I also had the fortune to make some really great friends in these six years. Satya and Puja's company was always cheerful. Sheetal and Anirban were great friends, neighbors and my LA-food-exploring buddies. Nilkanth, my room-mate for being a constant source of fun, a great friend and for all the help he has continued to extend me over all these years In the later years of my journey, my friendship with Angana gave me a new light of confidence in myself and others. She taught me many a things about life, about people and became my perpetual coffee, then tea iv and finally hot-chocolate buddy. I keep wishing we meet again . Deepak and Soniya , Archith and Aarthy, Nilkanth and Payal, Debanjan , Bibek, Mukul, Gaurav, Karthik, Samprita, Karishma and Rimita for a beautiful friendship and listening to my rants and for having me over for numerous lovely dinners. Asif for being a wonderful-wonderful buddy who always listened to my constant blabbering, for inspiring me to do good in Science and in life. Vasantika for watching my back for me and extending a helping hand whenever I needed it. Nachikethas for being my MATH-buddy, a great friend and for teaching me how to be focused! Shweta Ramdas for constantly checking on me and for reviewing my writings even before we met each other physically for the first time. Madhura Mukhopadhyay for her words of wisdom during my postdoc search . Shailja for being a wonderful friend and for celebrating my little successes with me. Friends from my undergrad - Akshay Korde, Prateek Maheshwari, Pushkar Godbole and Ashlesh Sonje for constantly checking on me and for constatntly planning awesome trips some of which did see the light of day. Shekhar Mishra and Anirudh Subramanyam for inspiring me to be sane when the times were tough and for their book recommendations some of which changed part of my life (Siddharatha) . My close friends Seshadri and Shuchi for a wonderful friendship and for making me witness my first wedding in the US! Yamini Krishnan and Parijat Bandyopadhyay for the wonderful conversations on Science and beyond and for going through my writings multiple times. My alma mater, IIT Bombay for introducing me to a wonderful pool of both opportunities and friends! My cousins Sanju and Chintu bhaiyya, Puja and Neha bhabhi and my wonderful nephews and nieces for their unconditional love and support! Suddu mama and Jaya mami for always keeping me motivated. My aunt, Mitthu didi for all the love she has showered over all of us over the years and for always looking after me, right from my birth. My extended family: Jay and Thelma V auntie for showering me with lots of love and keeping me motivated during tough and not so tough times. Jay's insights have made me re-realize some simple but important things in life - 'A no response is also a response' and how to believe in the power of intuition, something that was particularly useful during my postdoc search . Conversations with auntie have made me realize the importance of realizing the value of things in life, something I hope will stick with me forever. I am here today because of some really wonderful people, my pillars - my parents. No words that I write here will ever do justice they mean to me and for what they have done for me. My Papa instilled in me the spirit of hard work, patience and perseverance and Maa has taught me the art of keeping calm . Their teaching helped me through graduate school and I believe they will keep doing so. My sister Khushi, who literally lives by her name and has been a source of happiness for me and everyone in her family. She was my only source of inspiration while growing up and continues to inspire me to think, about things, about people, how to really care for someone and in whose shadows I will always walk. vi Table of Contents Dedication Acknowledgements List Of Tables List Of Figures Abstract Chapter 1: Introduction 1.1 Contributions 1.2 Outline . . . . . . Chapter 2: Background: Translational regulation and Ribo-seq 2.1 Translation .......... . 2.1.1 Translation Initiation 2.1.2 Translation Elongation . 2.1.3 Translation Termination 2.1.4 Translation in eukaryotes versus prokaryotes 2.2 Factors affecting translation 2.2.1 m 7 Gcap .......... . 2.2.2 Sequence context . . . . . . 2.2.3 The relative position of AUG 2.2.4 Secondary Structure . . . . . 2.2.5 Leader length . . . . . . . . 2.3 Quantitative analysis of in-vivo translation by Ribo-seq . 2.3.0.1 Comparison to polysome profiling . . 2.3.1 Ribo-seq protocol . . . . . . . . . .. . . . . . 2.3.2 Using Ribo-seq to decipher translation regulation . ii iii xi xiii xx 1 6 7 9 9 10 10 11 12 12 12 13 13 15 15 15 16 17 18 Chapter 3: Accurate detection of short and long ORFs from Ribo-seq data 22 3.1 Introduction . . . . . . . . . . . . . . . . 22 3.2 Methods . . . . . . . . . . . . . . . . . . 26 3.2.1 Obtaining and pre-processing data 38 vii 3.2.2 Learning cutoff of phase score . . . . . . . . . . . . . . . . . . 42 3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 3.3.1 Ribotricer accurately detects translating ORFs at the exon level 43 3.3.2 Ribotricer accurately detects translating ORFs at the transcript level 53 3.4 Ribotricer's phase score remains stable on truncated ORFs . 61 3.4.1 Ribotricer can detect ORFs as short as 20 codons 65 3.4.2 Learning species-specific cutoffs . . . 65 3.4.3 Learning dataset-specific cutoffs . . 67 3.4.4 Running ribotricer on a new species . 72 3.5 Using ribotricer . . . . . . . . . . . . . . . 86 3.5.1 Preparing candidate ORFs list . . . . 86 3.5.2 Detecting actively translating ORFs using ribotricer 90 3.5.3 Filtering actively translating ORFs using multiple criteria . 92 3.5.4 Downstream ranking and filtering . . 93 3.5.5 Learning cutoff empirically from data . . . . . . . . . . . 95 Chapter 4: Translational landscape of morphological changes in Candida albicans 97 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 4.2.1 Deep RNA sequencing and Ribosome profiling of C. albicans under yeast and filamentous like growth conditions . 98 4.2.2 Changes in the transcriptional landscape . . . . . . . . 99 4.2.3 Translational landscape changes . . . . . . . . . . . . . 99 4.2.3.1 Re-annotating the C. albicans transcriptome . 106 4.2.4 Detecting novel exons and potential ORFs 110 4.2.5 Detecting pausing sites . . . . . . . 111 4.3 Methods . . . . . . . . . . . . . . . . . . . . . . 112 4.3.1 RNA-seq and Ribo-seq data analysis . . . 112 4.3.2 Differential expression and translational efficiency analysis 113 4.3.3 Detecting pausing sites . . . . . . . . . . . . . . . . . . 113 Chapter 5: Integrated analyses of early responses to radiation in glioblastoma 116 5.1 Introduction . . . . . . . . . . . . . . . . . 116 5.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 5.2.1 Cell culture and radiation treatment . . . . . . . . . . . . . . 118 5.2.2 RNA preparation , RNA-seq and Ribosome Profiling (Ribo-seq) 122 5.2.3 Overall strategy to identify gene expression alterations upon radiation . 122 5.2.4 Sequence data pre-processing and mapping . . 122 5.2.5 Differential gene expression analysis . . . . . . 123 5.2.6 Weighted gene co-expression network analysis 123 5.2.7 Translational efficiency analysis . . . . . . . . 124 5.2.8 Alternative splicing analysis . . . . . . . . . . 125 5.2.9 Gene ontology (GO) and pathway enrichment analysis 125 5.2.10 Expression correlation analysis . . 126 5.2.11 Availability of data and materials . . . . . . . . . . . 126 viii 5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 5.3.1 Changes in global transcriptome profile in response to radiation 128 5.3.2 Changes in lncRNA profile in response to radiation . 134 5.3.3 Effect of radiation on splicing . . . . . 134 5.3.4 Differential translational efficiency 137 5.3.5 Crosstalk between regulatory processes 139 5.4 Discussion . . . . . . . . . . . . . . . . . . . 139 5.4.1 E2Fl, E2F2, E2F8, and FOXMl as major drivers . 140 5.4.2 RNA processing and regulation as novel categories in radio-response 142 5.4.3 Potential new targets to increase radio-sensitivity and prevent relapse 142 Chapter 6: Ribopod: A database of uniformly processed Ribo-seq datasets 145 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . 145 6.2 Ribopod : a database of de-noised Ribo-seq datasets 146 6.2.1 Obtaining metadata 147 6.2.2 Workflow 148 6.3 Results . . 6.4 Conclusion . . . . Chapter 7: Conservation of uORF mediated regulation across species 7.1 Introduction ........................... . 7.2 Results .............................. . 7.2.1 uORFs have a repressive effect on translational efficiency 7.2.2 uORFs are present across species . .. . . . . . . . 7.2.3 More uORFs than dORFs show active translation 7.2.4 uuORFs - uORFs that are almost always translating 7.2.5 Conservation of uuORFs across species 7.3 Discussion . . . . . .. . . . . . . . . . .. . Chapter 8: Conclusions Bibliography 149 150 153 153 155 156 158 162 167 168 176 182 185 AppendixA pysradb: A Python package to query next-generation sequencing meta- data and data from NCBI Sequence Read Archive 229 A.1 Introduction . . . . . . 229 A.2 Methods . . . . . . . . 231 A.2.1 Implementation 231 A.2 .2 Operation 232 A.3 Use cases . . . . . . . . 232 A.3.1 Search . . . . . 233 A.3.2 Getting metadata for a SRA project 234 A.3.3 Getting SRPs from GSE . . . . . . . 238 A.3.4 Getting a list of GEO experiments for a GEO study 239 A.3.5 Getting SRR from GSM . . 242 A.3.6 Downloading SRA datasets . . . . . . . . . . . . . 242 ix A.4 Summary . . . . . . A.5 Data availability . . A.6 Software availability 243 243 244 X List Of Tables 3.1 Adapters trimmed from Ribo-seq and RNA-seq samples for each dataset. 39 3.2 Reference assemblies and GTF for each species . . . . . . . . . . . . . . 40 3.3 Ribo- and RNA-seq samples used for the benchmarking along with the read lengths and P-site offsets used for Ribo-seq samples. . . . . . . . . . . . . . . . . . . . 41 3.4 List of datasets. . . . . . . . . . . . . . . . 3.5 Datasets used to learn ribotricer phase score cutoffs. 3.6 Species specific recommended phase score cutoffs for ribotricer. A "#" indicates the cutoff for the species is taken to be the median phase score difference between CDS annotated Ribo-seq and RNA-seq profiles since they only had one dataset 44 45 each . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 3.7 Species wise mean , median and standard deviation of difference of Ribo-seq and RNA-seq phase scores. SD = Standard Deviation . A "# " indicates that the median phase score difference for these species is also considered as cutoff for ribotricer, since they only had one dataset each. . . . . . . . . . . . . . . . . . 73 3.8 Best and second to best performing methods at AUROC metric for each dataset. 74 3.9 Best and second to best performing methods at Fl score metric for each dataset using dataset-specific cutoff. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 3.10 Best and second to best performing methods at Fl score metric for each dataset- specific cutoff. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 3.11 Ribotricer's performance at Fl score when considering species-specific or dataset- specific cutoff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 xi 4.1 Genes transcriptionally induced during the C. albicans morphological transition and involved in pathogenesis and / or virulence-related properties show significantly reduced translational efficiency (TE) . . . . . . . . . . . . . . . . . . . . . . . 105 4.2 Selected genes showing significantly increased translational efficiency (TE) during the C. albicans morphological transition . . . . . . . . . . . . . . . . . . . . . . 108 5.1 List of oncogenic factors, genes whose high expression is associated with poor survival and genes previously associated with radio-resistance in GBM that showed increased expression upon radiation . Genes are listed according to molecular function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 5.2 Splicing regulators showing changes in expression 24 hours post-radiation . Factors showing an increase in the expression are shown in red, while factors showing a decrease in the expression are represented in blue. . . . . . . . . . . . . . . . . 136 7.1 7.2 7.3 7.4 7.5 ORF types and their description. . ..... Species-specific cutoffs for active translation. Thresholds for defining universal uORFs Sequence context of uuORFs Sequence context of CDS . . . 160 168 179 179 180 xii List Of Figures 2.1 Canonical model of translation: initiation, elongation and termination. 2.2 2.3 2.4 3.1 3.2 uORFs and their potential effects on the downstream CDS. Protocol of Ribo-seq . Ribo-seq protocol .. Length distribution of candidate ORFs for human and mouse. Methodology design of ribotricer. . . . . . . . . . . . . . . . 11 14 17 19 23 28 3.3 Read length distribution of Ribo-seq and RNA-seq samples from human datasets. 32 3.4 Read length distribution of Ribo-seq and RNA-seq samples from mouse datasets. 33 3.5 Metagene plots for representative read lengths for human Ribo-seq samples. 34 3.6 Metagene plots for representative read lengths for mouse Ribo-seq samples. 35 3.7 3.8 3.9 Distribution of the resulting vector angles for datasets in human. Distribution of the resulting vector angles for datasets in mouse . . Learning the cutoff for phase scores for human datasets. 3.10 Learning the cutoff for phase scores for mouse datasets. 36 37 46 47 3.11 ROC plots and Precision-Recall plots for human datasets for exon level classification . 49 3.12 ROC plots and Precision-Recall plots for mouse datasets for exon level classification . 50 xiii 3.13 Number of translating exons recovered when controlling the false positive rate to be the same. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 3.14 Effect of ORF length on output scores. 3.15 Comparison of performance on detecting translating exons. The performance of ribotricer is compared with that of RiboCode, Ribo Taper, and ORFscore. (A) The ROC and precision recall curves summarizing performance of ribotricer, RiboCode, Ribo Taper and ORFscore on one human and one mouse dataset. (B) The number of translating exons recovered when controlling the false positive rate to be the 52 same. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 3.16 ROC plots and Precision-Recall plots on transcript level for human datasets. 55 3.17 ROC plots and Precision-Recall plots on transcript level for mouse datasets. 56 3.18 Comparison of Fl score ( exon level) of ribotricer with RiboCode, Ribo Taper, and ORFscore. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 3.19 Comparison of sensitivity ( exon level) of ribotricer with RiboCode, Ribo Taper, and ORFscore. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 3.20 Comparison of specificity ( exon level) of ribotricer with RiboCode, Ribo Taper, and ORFscore. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 3.21 Comparison of precision ( exon level) of ribotricer with RiboCode, Ribo Taper, and ORFscore. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 3.22 Number of translating transcripts recovered when controlling the false positive rate to be the same. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 3.23 Performance of different methods on transcript level measured using Fl score. 61 3.24 Effect of number of codons on ribotricer's phase score in human dataset. 62 3.25 Effect of number of codons on ribotricer's phase score in mouse dataset. 63 3.26 Comparison of performance on detecting translating transcripts.The performance of ribotricer is compared with that of RibORF, RiboWave, and Rp-Bp. (A) The ROC and precision recall curves summarizing performance of ribotricer, RibORF, RiboWave and Rp-Bp on one human and one mouse dataset. (B) The number of translating transcripts recovered when controlling the false positive rate to be the same. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 xiv 3.27 Effect of truncating an ORF on ribotricer's phase score, RiboCode's p-values and ORFscore in human dataset. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 3.28 Effect of truncating an ORF on ribotricer's phase score, RiboCode's p-values and ORFscore in mouse dataset. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 3.29 Example of phase scores for an active and a non-active ORF. 68 3.30 Summarized median phase score for RNA-seq and Ribo-seq for all datasets. 69 3.31 Median phase score for RNA-seq and Ribo-seq and their differences across multiple species. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 3.32 Distribution of median phase scores for RNA-seq and Ribo-seq samples and their differences across multiple species. . . . . . . . . . . . . . . . . . . . . . . . . 70 3.33 Distribution of individual RNA-seq and Ribo-seq samples' phase scores across species ............... . 71 3.34 Distribution of median difference between Ribo-seq and RNA-seq sample as de- termined using only two datasets per species. . . . . . . . . . . . . . . . . . . . 73 3.35 Distribution of area under ROC (AUROC) across multiple species . . 3.36 Distribution of Fl scores across species using species-specific cutoff. . 3.37 Performance of ribotricer at AUROC and Fl scores metrics across species at different median phase scores of RNA-seq and Ribo-seq samples using species- 78 79 specific cutoff. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 3.38 Distribution of Fl scores across species using dataset-specific cutoff. 81 3.39 Difference in performance of ribotricer using species-specific or dataset-specific cutoffs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 3.40 Summarized performance of ribotricer using species-specific and dataset-specific strategies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 3.41 Distribution of ribotricer's Fl scores with respect to median phase score difference of Ribo-seq and RNA-seq, using species-specific and dataset-specific cuoffs. . . . 83 3.42 Effect of Ribo-seq and RNA-seq phase scores on species-specific and dataset- specific based Fl performance. . . . . . . . . . . . . . . . . . . . . . . . . . . 84 3.43 Distribution of area under ROC in the independent datasets . . 87 xv 3.44 Distribution of Fl scores in the independent datasets. . . . . . . . . . . . . . . 88 3.45 Distribution of ribotricer's phase scores for RNA-seq and Ribo-seq samples in the independent datasets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 4.1 Distribution of total reads segregated by read length in Ribo-seq and RNA-seq experiments. . . . . 100 4.2 Ribosome profiling in Candida albicans. . . . . . . . . . . . . . . . . . . . . . . 101 4.3 Correlation between replicates across conditions. . . . . . . . . . . . . . . . . . 102 4.4 Differential expression analysis between 37° C + Serum and 30° C samples. . . . 103 4.5 Differential translational efficiency between 37°C + Serum and 30°( samples. . 104 4.6 Gene ontology analysis of TE genes . . . . . . . . . . . . . . . . . . . . . . . . 107 4.7 Ribo-seq enables novel ORF discovery. . . . . . . . . . . . . . . . . . . . . . . 109 4.8 Pausing in Candida albicans. . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 4.9 RPF profile of PCSK9 gene from data in Lintner et al. [1] . . . . . . . . . . . . . 114 5.1 Experimental design and Principal Component Analysis of RNA-Seq and Ribo- Seq data from glioblastoma cell lines. A) Schematic representation of experimental protocol followed for radiation exposure of glioma cell lines, and sample preparation for sequencing the RNA and ribosome footprints. B) Principal component analyses performed on normalized log- transformed read counts of RNA-Seq and Ribo-Seq datasets. . . . . . . . . . . . . . . . . . . . . . . . . . 119 5.2 Fragment length distribution of ribosome footprints of glioblastoma cell lines. Fragment lengths distribution was obtained using ribotricer. . . . . . . . 120 5.3 The ribosome density profiles of glioblastoma cell lines. The metagene distribution was obtained using ribotricer. . . . . . . . . . . . . . . . . . . . . . 121 xvi 5.4 Global view of glioblastoma cell lines transcription and translation profiles after radiation. A) Differentially expressed genes (left) and the number of genes whose translation efficiency is differentially regulated (right) after radiation exposure at different time points. B) Volcano plots showing the expression and transition alterations of genes at 24 hr compared to 1 hr after radiation exposure. Blue dots indicate upregulated genes (adjusted p-value < 0.05, log 2 fold change < 0) , and orange dots indicate downregulated genes. (adjusted p-value < 0.05, log 2 fold change < 0) . C) Sizes of gene modules found in U251 and U343 cell lines. D) Preservation Median Rank and Zsummary for all modules. A lower median rank indicates the module is preserved, and the corresponding modules in U251 and U343 cell lines share a high number of genes. A Zsummary score of 2-10 indicates weak preservation , while a Zsummary > 10 indicates high preservation . 127 5.5 Characteristics of downregulated genes at 24 hours (T24) after radiation exposure in glioblastoma cell lines. A) Enriched gene ontology related to cell cycle, DNA replication, and repair among downregulated genes. B) RNA-related Gene Ontology (GO) terms enriched among downregulated genes summarized using REVIGO [2] . C) Protein-protein interaction network, according to STRING [3] showing downregulated genes associated with RNA-related functions. Gene clusters based on the strength of connection and gene function are identified by color. Lines colors indicate the type of association: light green indicates an association based on literature findings; blue indicates gene co-occurrence; magenta indicates experimental evidence. . . . . . . . . . . . . . . . . . . . . . 129 5.6 E2Fs and FOXMl in glioblastoma. A) Correlation of E2Fl, E2F2, E2F8, and FOXMl with target genes involved in cell cycle. B) Expression levels of E2Fl, E2F2, E2F8, and FOXMl in gliomas grades II, Ill, and IV in TCGA samples. C) E2Fl, E2F2, E2F8, and FOXMl expression correlation in glioblastoma (TCGA samples) using Gliovis [4] . *** p-value < 0.0001. . . . . . . . . . . . . . . . . 130 5.7 Global view of upregulated genes at T24 post-radiation in glioblastoma cells. A) Gene ontology analysis of upregulated genes B) Protein-protein inter action networks according to STRING [3] showing genes associated with extra cellular matrix organization and response to interferon. Gene clusters based on the strength of connection and gene function are identified by color. Lines colors indicate type of association : light green, association based on literature findings; blue indicates gene co-occurrence; magenta indicates experimental evidence. . . 131 xvii 5.8 Impact of radiation on the splicing profile of glioblastoma cells. A) GO enriched terms among genes showing changes in splicing profiles at T24. GO enriched terms are summarized using REVIGO [2]. B) Protein-protein interaction networks according to STRING [3] showing genes associated with RNA-related functions whose splicing profiles displayed alterations at T24. Gene clusters based on the strength of connection and gene function are identified by color. Lines color indicate type of association : light green , an association based on literature find- ings; blue indicates gene co-occurrence; magenta indicates experimental evidence. 135 5.9 Impact of radiation on the translation profile of glioblastoma cells. A) GO-enriched terms among genes showing changes in translation efficiency at T24. GO-enriched terms are summarized using REVIGO [2] . B) Protein-protein inter action network, according to STRING [3] showing genes whose translation effi ciency decreased at T24. Gene clusters based on the strength of connection and gene function are identified by color. Line colors indicate the type of associa- tion : light green, an association based on literature findings; blue indicates gene co-occurrence; magenta indicates experimental evidence. . . . . . . . . . . . . . 138 6.1 Ribopod's visualization interface. 149 6.2 Workflow of ribopod .. . . . . . 150 6.3 Phase score distribution for datasets present in ribopod. 151 7.1 Distribution of phase scores and the number of libraries 156 7.2 Higher number of actively translating uORFs result in down-regulation of the main ORF in (a) human (b) chimpanzee, and (c) macaque. (d) Number of differential translationally efficient genes in chimpanzee and macaque with respect to human. 158 7.3 Correlation between translational efficiency of CDS and uORF in orthologous regions of human , chimpanzee and macaque indicate the phylogenetic distances. Correlation values are indicated in the legend . . . . . . . . . . . . . . . . . . . 159 7.4 7.5 7.6 7.7 Distribution of number of known and potential ORFs. Distribution of number of different ORF types per annotated CDS. Distribution of lengths of different ORF types. . . . . . Distribution of phase scores in uORF, CDS and dORF . . 160 161 161 162 7.8 Distribution of percentage of uORFs and dORFs under active translation with respected to coding genes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 xviii 7.9 Distribution of uORFs and dORFs that are always translating with respect to coding genes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164 7.10 Correlation between uORF and CDS phase scores. 7.11 Correlation between dORF and CDS phase scores. 165 166 7.12 Distribution of phase scores of CDS regions that are above the minimum score required for active translation. . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 7.13 Determining thresholds for 'always active' translation . 7.14 Percentage of conserved uuORFs . . . . . . .. . . 7.15 Upstream sequence context of CDS in vertebrates. 7.16 Upstream sequence context of CDS in invertebrates. 169 170 171 172 7.17 Sequence similarity of upstream sequence contexts of always and never translating uORFs in vertebrates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 7.18 Sequence similarity of upstream sequence contexts of always and never translating uORFs in invertebrates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174 7.19 Sequence similarity of upstream sequence contexts of always and never translating CDS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 7.20 Sequence similarity of upstream sequence contexts of always and never translating CDS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 7.21 Sequence similarity of upstream sequence contexts of always and never translating CDS and uORFs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178 xix Abstract xx Abstract Any biological process is a manifestation of the protein levels. Protein are synthesized inside the body through the process of translation . Translation is one of the most energy-intensive pro cesses and hence is highly regulated to ensure the proteins get synthesized in the right quantity in the cell. The advent of next-generation sequencing has enabled transcriptome-wide monitoring of translation regulation . In particular, Ribo-seq has provided us with a high-throughput sequencing method to profile mRNA undergoing active translation at sub-codon resolution . However, the resulting dataset has inherent noise arising from the sequencing of non-ribosomal and non-active translation fragments. Through this work, I develop a method to identify regions under active translation using Ribo-seq data. The method enables more accurate identification of both short and long open reading frames. I develop a database of uniformly processed public Ribo-seq datasets which have been appropriately de-noised to recover only the actively translating frag ments. I also demonstrate the utility of our method as a quality control metric to asses the quality of Ribo-seq data generated in a variety of treatment and biological contexts. I utilized our method to understand the changes involved in translational landscape in two important biological contexts. First, I present a study on translational landscape changes involved in the irradiation of glioblastoma cell lines thereby providing evidence of radio-resistance in some target genes that can be likely candidates for increasing irradiation efficacy. Second , I provide xxi evidence of translational landscape changes involved in the morphological transition of Candida albicans, a fungal pathogen involved in inducing diseases as Candidasis in immunocompromised patients. These target genes can serve as potential targets for anti-fungal drugs. Finally, by applying our method to identify actively-translating regions on public Ribo-seq datasets, I discovered thousands of upstream open reading frames under active translation across various biological contexts in multiple species. These upstream open reading frames (uORFs) often act as a repressive element for the downstream protein coding regions. Using Ribo-seq datasets from multiple species across different conditions, I characterize the sequence properties of a subset of these uORFs which are almost always under active translation . Further, I analyze it in an evolutionary context providing evidence for the conservation of the upstream sequence context. xxii Chapter 1 Introduction The basic structural, functional and biological unit of all organisms is the cell. Cell, the smallest unit of life, is enclosed within a membrane and contains two key biomolecules : DNA and proteins. Since their discovery in 1869 and 1838 respectively, these have become widely studied and amongst the best-understood biological molecules. DNA has been known to biologists since 1869 when the young Swiss doctor Friedrich Miescher, working in the laboratory of Felix Hoppe-Seyler at the University of Tobingen, while performing experiments to determine the chemical composition of leukocytes isolated a precipitate that contained large amounts of phosphorus, lacked sulfur, and was resistant to protease digestion thus essentially ruling out proteins and lipids [5]. By itself, DNA is little more than a blueprint for life. Any biological process is a manifestation of the proteins and their abundance. Proteins were described in scientific literature as early as 1838 by the Dutch chemist Gerardus Johannes Mulder though the name itself was a creation of Swedish chemist Sons Jacob Berzeliuswith [6]. The name protein originates from the Greek word 1rpwrw:.; (proteios) - primary or leading in the front [7]. As such, proteins act as building blocks 1 for a large variety of large molecules and account for 44% of the human body's dry weight [8]. They also also act as enzymes that catalyze the biochemical reactions in all living beings. Francis Crick gave a lecture on 19 th September 1957 as part of the Society of Experimental Biology Symposium on the Biological Replication of Macromolecules at University College London . This lecture entitled " Protein Synthesis" that was later written up as " On Protein Synthesis" [9] was significant from multiple fronts. First, it laid down the foundation of what is now famously known as the central dogma of molecular biology. The central dogma states that once 'information ' has passed into the protein , it cannot be transferred back to either nucleic acid or protein . In other words, information can flow from nucleic acid to nucleic acid or from nucleic acid to protein but transfer of information from protein to protein or protein to nucleic acid is impossible. Second , Crick made a couple of predictions that were eventually proved to be true. First, he predicted the existence of adaptor molecules (later discovered in the same year as tRNA [10]) that work as mediators carrying amino acids to the site of protein synthesis And hence, Crick "permanently altered the logic of biology [11]". Second , he predicted that it would be possible in the future to compare sequencing data to explore the vast amount of evolutionary information between them. The single largest investment of energy by cells is in the process of translation of mRNAs into protein [12, 13]. Being a substantial energy investment, translation is highly regulated to ensure that the resulting proteins have the desired stoichiometry and are available at the right places within the cell. For example, in a rapidly growing yeast cell, nearly 200, 000 ribosomes occupy as much as 30- 40% of the cytoplasm and 2000 ribosomes are synthesized every minute absorbing around 60% of its transcriptional activity [14]. Translation regulation is especially important 2 in maintaining homeostasis and controlling cell proliferation and growth [15]. Dysregulation of translation has been implicated in a wide range of diseases [16]. The notion of translation regulation emerged only a few years after Crick formulated the central dogma of molecular biology. In 1961, Jacob and Monod proposed the idea of 'messenger' intermediate (mRNA) could be subject to different utilization depending on the context [17]. They perceived "that the synthesis of individual proteins may be provoked or suppressed within a cell, under the influence of specific external agents, and more generally that the relative rates at which different proteins are synthesized may be profoundly altered, depending on external conditions" [17] and that such a regulation is "is absolutely essential to the survival of the cell " [17]. This idea of mRNA being utilized differently to regulate the final protein drew very little attention at that time but has been one of the seminal ideas in molecular biology. The advent of next-generation DNA sequencing technologies and the development of assays such as RNA-seq [18] has facilitated relative ease of measuring mRNA abundance in a high throughput manner, thus, advancing our knowledge of how cells modulate their gene expression across different physiological and pathological processes. However, biological processes are driven by proteins, while mRNA concentration acts as a mere proxy to protein abundance. At a steady state, protein abundance is a function of the rates of four phenomena: transcription, translation , mRNA decay, and protein decay [19, 20]. Measuring mRNA abundance has been found to be a good predictor of protein levels in multiple contexts [21]. Measuring mRNA abundance though highly informative provides a glimpse into the early steps of a long chain of regulatory events. The correlation between mRNA and protein levels in some contexts has been found to be as low as 0.36 [22]. Such low correlations have been attributed to translational regulation and post-translational buffering. Deciphering the relative 3 contribution of phenomena other than transcription can help us deepen our understanding of biological regulation. The need to decipher translational regulation has motivated the development of experimental approaches to profile the translational landscape. With the advent of next-generation sequencing technologies, it is now possible to determine the sequence of large DNA and RNA molecules in a high-throughput manner [23]. Ribo-seq [24] is a deep-sequencing based technique that captures snapshots of ribosome protected fragments revealing the positions of the entire pool of ribosomes engaged in translation, hence, providing a global view of the translational process in-vivo. Ribo seq has been used to answer some key biological questions involving the prevalent and dynamic nature of translational regulation of the mammalian cell cycle [25], the discovery of alternative translation initiation sites [26, 27]. and translational pauses under induced stress [28]. Ribo-seq provides a rich resource for understanding translation and its regulation . However, ever since the inception of Ribo-seq , it has been known that not all fragments arising from a Ribo-seq experiment are reflective of active translation [29, 30, 31]. Furthermore, Ribo-seq data is highly heterogeneous and noisy making the task of separating "signal " from noise particularly challenging. Multiple approaches have been taken to tackle this shortcoming [32, 33, 34, 35, 36]. These methods can be broadly classified into two paradigms. The first paradigm of methods hypothesizes that the distribution of Ribo-seq fragments is different from a hypothetical null and use this to separate the ones actively-translation from the non-active ones. The second paradigm of methods uses the periodic pattern expected from Ribo-seq signal, since the ribosome moves three nucleotides at a time, to determine actively-translating fragments. 4 The number of Ribo-seq studies has grown over the years. These studies span several tissues and species. They have been performed under different biologically or technically diverse condi tions. However, every new Ribo-seq studywill overlap with previous studies with respect to some biological dimensions of interest. Hence, interpreting new data can leverage on existing public data to accelerate the process of deciphering all biological characteristics for the new experi ment. Systematically leveraging on public Ribo-seq data can help us answer some key biological questions involving translation regulation . A key mechanism of regulating translation during the initiating step is via upstream open reading frames (uORFs) located in the 5' untranslated regions (5' UTRs). uORFs are sequences located upstream of the main protein coding region charac terized by an initiation codon in-frame with the stop codon. uORFs regulate the translation of downstream protein coding region by multiple mechanisms: by inhibiting initiation at the down stream start codon (AUG) because of their secondary structure; by stimulating cap-independent translation through the internal ribosome entry sites (IRES) ; by inhibiting or promoting transla tion because of presence of potential RNA binding protein sites and by the upstream sequence and start codon context [37]. The literature on uORFs thus far has primarily relied on simply using the presence of a 'favorable' sequence context in the 5' UTRs. This favorable sequence context is most often char acterized by the presence of a start codon (AUG) and an in-frame stop codon (UAG/ UGA/ UGA). Ribo-seq, on the other hand , can provide direct evidence of active translation in uORFs as com pared to a mere potential arising because of a favorable sequence context. By providing a snapshot of active translation, Ribo-seq has enabled investigating the true coding potential of the uORFs. Given that uORFs mediate translation regulation and it has functional implication , we expect this 5 mechanism to be similar across species as "Nothing in Biology Makes Sense Except in the Light of Evolution [38]". 1.1 Contributions 1. A method for detecting active translation in both short and long ORFs: Though tools exist for identifying active translation in Ribo-seq data, they are not sensitive enough to detect active translation in short ORFs ( < 100 amino acids). I developed a method, ribotricer, that exploits the inherent periodicity in Ribo-seq data to identify active transla tion . It gives the highest accuracy and sensitivity compared to other existing approaches as evaluated on multiple datasets across species . 2. Characterization of translational landscape changes in C. a/bicans: C. albicans is a fungal pathogen that inhabits the mucosal surfaces of most healthy individuals as human commensals. Though mostly asymptomatic, they are opportunistic pathogens and are known to causes disease in individuals with a debilitated immune system or a disruption in the host's microbiome. Using ribotricer, I provide insights into changes that take place in its translational landscape and are responsible for its increased virulence of this pathogen . 3. A uniformly processed database of Ribo-seq studies: I develop a database, ribopod, of uniformly processed public Ribo-seq datasets across multiple species. Leveraging the accuracy of ribotricer, I provide access to de-noised Ribo-seq datasets that will serve as a rich resource for the translation regulation research community. 6 4. Conservation of uORF mediated regulation: uORFs are known to play a role in sup pression of translation. Using ribotricer and ribopod, I provide evidence of conservation of uORF mediated regulation across species. I describe universal uORFs (uuORFs) that are almost always translating across different physiological and pathological contexts within a species. Furthermore, I show that the sequence context of the uuORFs is conserved across species. 1.2 Outline In Chapter 2, I provide a background for translational regulation and Ribo-seq . I provide a history of translation regulation, the known mechanisms that regulate translation. Finally, I describe Ribo-seq and highlight some of the challenges associated with the analysis of Ribo-seq data. Chapter 3 introduces a new computational method , ribotricer, to identify actively-translating regions in Ribo-seq data. I first provide a review of the existing computational approaches and their limitations. Next, I introduce our method , ribotricer, that overcomes the shortcomings of previous approaches. I provide detailed validation of our approach on multiple datasets across species. In Chapter 4, I describe changes in the transcriptional and translational landscape of C. albicans, a fungal pathogen, that rapidly transitions between yeast-like and filamentous growth patterns where the latter pattern leads to increased virulence. I utilize ribotricer to discover hundreds of genes with altered translation during the morphological translation . Using RNA-seq 7 and Ribo-seq data jointly I re-annotated the partially incomplete transcriptome of C. albicans and discovered novel exons. In Chapter 5, with the aim of developing a better targeted therapy for glioblastoma , I charac terize the effect of radiation on glioblastoma. Using RNA-seq and Ribo-seq data, we conducted an integrated analysis in the glioblastoma cell lines to profile alterations in the gene expression at multiple levels including transcription, splicing, and translation . Our approach provides a com prehensive view of early response to radiation in and suggests new target options to increase radiation sensitivity and prevent relapse of glioblastoma. In Chapter 6, I provide a background of existing Ribo-seq datasets and highlight their short comings. Next, I try to overcome these shortcomings by developing a database, ribopod, of uniformly processed public Ribo-seq projects from multiple species. Ribopod provides ready ac cess to de-noised Ribo-seq projects that can be used to discover new biological mechanisms or as a resource to test a hypothesis in different biological and physiological contexts. In Chapter 7, I use ribopod database to investigate uORF mediated regulation across species. I describe uuORFs, a set of uORFs that are almost always expressed across physiological and pathological conditions withing a species. I discover that the sequence context associated with these uORFs is different from the context associated with protein coding regions yet is conserva tion across species. 8 Chapter 2 Background: Translational regulation and Ribo-seq Protein synthesis is a quintessential part of the gene expression pathway and is itself involved in its control. The rate of synthesis of protein depends on both the concentration of the mRNA and its translational efficiency [39]. Translational efficiency is determined by the density of active ribosomes translating the mRNA. Half-life, i.e., the time required for degradation of 50% mRNA molecules for a majority of mammalian mRNAs is > 2 hours [40]. and hence tight protein regulation requires is done via translational efficiency and protein decay rates [41]. 2.1 Translation The process of translation is one of the most energy-intensive processes in any cell. The energy requirements are high owing to the multiple GTP hydrolysis steps involved [42]. The process is precise with the frequency of translations errors reported to be as low as 0.1 % [43, 44] in incorporating an erroneous amino acid. Aberrations in the translation pathway are known to cause various diseases [45]. Complete translation of a protein involves three steps of initiation , elongation, and termination that we briefly discuss in the sections that follow. 9 2.1.1 Translation Initiation Translation initiation begins with recognizing the 5' -m 7 G cap by the protein initiation com plex. This protein initiation complex consists of the 405 (smaller) subunit, translation initiation factors, and tRNA conjugated to methionine and attached to the peptidyl transfer (P-) site of the 405 subunit. The eukaryotic initiation factor 4F (elF4F) , which consists of factors elF4E, elF4A, and elF4G bind to the m 7 GpppN cap structure located at the 5' end of pol-II transcribed mRNAs. The 405 subunit carries the tRNA-elF2-GTP complex that gets attached at or near the 5' end of the mRNA. The 40-S associated factor elF3 and elF4G aid this linking [46]. The pre-initiation complex scans for the start codon 'AUG' in the 5' to 3' direction along the mRNA. Once an AUG is encountered , the methionine tRNA canonically base pairs with the AUG start codon thus arresting scanning and releasing initiation-factors. It makes way for the more prominent subunit 605 to bind to the 405 at AUG resulting in the 805 ribosome assembly. The presence of m 7 G cap is quintessential for translation [47]. though alternate mechanisms such as the presence of internal ribosomal entry sites (IRES) in the 5' untranslated region (UTR) can also recruit the translation initiation machinery independent of m 7 G presence [48]. Initiation is the rate-limiting step of the three steps in protein synthesis [49]. 2.1.2 Translation Elongation The 805 assembly traverses the mRNA codon by codon. There are three sites inside the 805 monosome: A, P and E. The A (acceptor) site acts as a binding site for the aminoacyl tRNA, the P (peptidyl) site binds to the tRNA that holds the growing nascent polypeptide and the E ( exit) site serves as a threshold for releasing the tRNA devoid of its amino acid . At each codon, 10 @) - =40S • =60S (, = Protein • = Initiation Complex Figure 2.1 Canonical model of translation: initiation, elongation and termination. 40S subunit along with the pre-initiation complex scans for 'AUG'. On encountering AUG, the 60S subunit binds to 40S forming the 80S complex while the pre-initiation complex is let free. The 80S complex traverses the mRNA one codon at a time growing the peptide chain (green) till it encounters a stop codon. The two subunits are released and recycled. the corresponding aminoacyl-tRNA is base pairs with the codon at the A-site. It results in the formation of a peptide bond and the growing peptide chain (Figure 2.1) is translocated to the P-site. The tRNA-peptide is still base-paired to the mRNA at the P-site while the initiator tRNA now occupies the E-site. It leaves the A-site empty for the next codon to recognize its cognate aminoacyl-tRNA. The elongation process proceeds one codon at a time until it encounters an in-frame stop codon . 2.1.3 Translation Termination At the stop codon, the linkage between elongated peptide chain and the P-site tRNA site undergoes hydrolysis thus releasing the nascent protein, and the two subunits of the ribosome disassociate. The subunits are recycled and re-used for the translation of other mRNA fragments. 11 2.1.4 Translation in eukaryotes versus prokaryotes In prokaryotes, ribosomes engage in co-transcriptional regulation. As soon as the mRNA emerges from the RNA polymerase, ribosomes can recognize the start codon on these mRNA and start translating. On the other hand, the eukaryotic translation starts only after some quality control checks. The mRNA in eukaryotes gets processed in the nucleus before being exported to the cytoplasm where the ribosomes can translate it. The mRNA goes through capping and polyadenylation, among other processing steps in the nucleus. These processing steps ensure that the ribosomes engage with only mRNAs that are complete [50]. 2.2 Factors affecting translation There are five main contexts that can modulate translation in eukaryotes and more specifically in vertebrates [51]: 1. m 7 G cap 2. the sequence context surrounding the AUG start codon 3. the relative position of AUG start codon in the transcript ("first" versus others) 4. upstream and downstream secondary structure 5. leader length We briefly discuss the effects of each of these contexts in the sections that follow. 2.2.1 m 7 G cap The m 7 G cap stabilizes the mRNA by preventing it from degrading by the attack of phos phatases and other nucleases. The cap generally gets added to mRNA precursors synthesized by RNA polymerase 11 and to viral transcripts that are replicated in the nucleus. It increases translational efficiency [52] and is quintessential for translation except in viral mRNAs. 12 2.2.2 Sequence context Marilyn Kozak characterized GCC(A/G)CCAUGG as the optimal sequence context for translation initiation [47] at the AUG start codon in vertebrates. This context affects both fidelity and the efficiency of initiation. AUGs around sub-optimal contexts can cause some 40S subunits to bypass it and initiate translation at a downstream AUG. It is possible that the first AUG if lying in a sub-optimal context, is recognized inefficiently which is consistent with the leaky scanning model. The leaky scanning model proposes that the 40S subunit continues scanning the mRNA for AUGs even after encountering an AUG. This leaky scan ning model enables some viral mRNAs to produce two proteins by initiating at two AUGs ly ing close to each other [53]. Based on the two most crucial nucleotides -3 (A/ G) and +4G, the surrounding sequence context has been classified as 'optimal ', GCC(A/ G)CCAUGG; 'strong' NNN(A/ G)CCAUGG when the two important nucleotides are present; 'adequate', when either of the two important nucleotides are present: NNNRNNAUG(A/ C/ U) or NNN(C/ U)NNAUGG, and 'weak' NNN(C/ U)NNAUG(A/ C/ U) that lacks both of the two important nucleotides. This con- text was assumed to be universal amongst eukaryotes. Later, however, it was eastablished that sequence context might be species dependent. Amongst the characterized species, the sequence context is ACAACCAAAAUGGC for Drosophila, AAAAAAAAAAUGTC for Saccharomyces cere visiae, and UAAAT(A/ C)AACAUG(A/ G)C for other invertebrates [54]. 2.2.3 The relative position of AUG The scanning model of the ribosome hypothesizes that the ribosomes will initiate translation at the first encountered AUG as long as it lies in a 'optimal ' context [55, 56]. An experiment 13 / · --------~~---~~::::====]--- uORF. CDS Figure 2.2 - / repression - leaky scanning uORFs and their potential effects on the downstream CDS. uORF translation can result in short peptides while the free sub-units can re-initiate translation at the downstream CDS. Presence of uORFs can repress the production from downstream CDS. Leaky scanning by 40S subunit can skip uORFs and initiate translation at the downstream CDS. involving the insertion of out-of-frame AUG codons showed dramatic inhibition of translation [57]. Thus, efficient translation requires that no spurious AUG codons be present upstream. However, the presence of upstream AUGs in the 5' UTR regions may not lead to complete inhibition of translation if it is followed by a stop codon. The 80S ribosome may be released after the translation of this small region, but the 40S will remain bound and continue to scan the mRNA. It may then re-initiate translation at a downstream start codon (the original CDS), however, the re-initiation will happen efficiently only if the upstream ORF is short (Figure 2.2). 14 2.2.4 Secondary Structure Secondary structure can have both positive and negative effects on translation initiation. A small amount of secondary structure near the AUG can up-regulate its recognition by 40S. The secondary structure downstream tends to slow down the scanning by the 40S thus giving it more time to recognize the AUG. On the other hand , the presence of stem-and-loop secondary structure between the AUG codon and the 5' cap can inhibit translation . This stem-and-loop prevents the 40S subunit from binding if it inhibits the entry sites for ribosomes. If the secondary structure occurs far away from the cap such that binding of the 40S subunit is not impaired, the inhibition of translation depends on the stability of the stem-and-loop as a strong stable structure will inhibit 40S' movement. 2.2.5 Leader length If the AUG is located very close to the 5'cap, the recognition might be impaired. In an experiment with synthetic transcripts where the AUG was located in a favorable Kozak context too close to the cap, the initiation efficiency was significantly lower as compared to when the leader sequences were elongated independent of the sequence context. This can be attributed to the fact that longer 5' leaders might provide more avenues for loading 40S as compared to shorter sequences. 2.3 Quantitative analysis of in-vivo translation by Ribo-seq lngolia et al. [24] developed a high throughput sequencing-based approach for ribosome profiling called Ribo-seq . It has enabled a transcriptome-wide quantitative analysis of in-vivo 15 translation . The assay involves deep-sequencing of mRNA fragments protected by ribosomes which are believed to be undergoing active translation . It gives a codon level resolution for the dwell time of ribosomes [58] . The abundance of Ribo-seq fragments has been shown to be more correlated with the absolute protein abundances [24]. Ribosomes protect around 28 - 30 nucleotides of the mRNA from digestion [24, 59] while tightly packed ribosome pairs can protect around 58 - 62 nucleotides [60]. On the other hand , 48S pre-initiation complex can protect 50 - 70 nucleotides [61] because of the presence of elFs. 2.3.0.1 Comparison to polysome profiling Polysome profiling measures the constituent mRNA bound by different ribosome number fractions using microarrays [62]. Ribo-seq has some clear advantages over the microarray based approach . Ribo-seq can distinguish between ribosomes engaged at the protein-coding regions from those engaged in the upstream regulatory open reading frames (uORFs) [31]. mRNAs that undergo rapid degradation might not be profiled by polysome profiling [62]. Ribo-seq, on the other hand , relies only on nuclease footprint from single ribosomes and as such has lesser sensitivity to the integrity of mRNA. At the same point, there are certain key properties of polysome profiling that Ribo-seq is not able to capture. For instance, polysome profiling has a greater ability to measure differences in the translation of alternate isoforms, particularly so when they differ in their ribosomal occupancy in the 5' or 3' untranslated regions. Also, polysome profiling monitors the translation status of entire full length transcripts while Ribo-seq focuses on determining activity of individual ribosomes. As such , using polysome profiling it is possible to capture the difference arising from a uniform decrease in ribosomal occupancy on all copies of a transcript from the 16 = = - t Nuclease Digestion I Footprin! purification t Library preparation Figure 2.3 Protocol of Ribo-seq. A translating ribosome encloses 30 nucleotides of the mRNA and protects it from digestion. mRNA protected fragments are obtained by nuclease digestion which digests the unprotected mRNA. The protected fragments are then sequenced post ribosome separation. scenario where a sub-population of mRNAs is under complete repression, a scenario that has been observed in mouse embryonic stem cells [63]. 2.3.1 Ribo-seq protocol Ribo-seq protocol consists of five major steps: 1. Cell lysis and ribosome arrest 2. Nuclease footprinting (RN Ase digestion) 3. Isolation of ribosome footprints 4. Library preparation and high-throughput sequencing Cell lysis and ribosome arrest: In order to profile the translatome, the polysomes need to be first stabilized . This has traditionally been carried out by treating the cells with translation elongation inhibitors before lysing them. One of such inhibitors, cycloheximide has been shown to 17 be biased towards certain codons, altering the distribution of ribosomes on the mRNA, especially near the start codons [64, 65, 66, 67]. Based on these observations, flash-freezing of cells appears to be the most robust approach [68]. Nuclease footprinting (RNAse digestion): This step involves digesting the portion of mRNA that is not protected by ribosomes. RNAse I in eukaryotes and micrococcal nuclease (MNAse) has been used in bacterial cells to perform digestion step [69]. Isolation of Ribosome footprints: After RNA digestion, the ribosome protected fragments that are intact need to separated from the cell lysates. This was originally performed using a sucrose density gradient centrifugation but is now performed by passing the lysate through a sucrose cushion during ultracentrifugation to ensure purification ribosome-bound mRNA. Library preparation: The library preparation step encompasses rRNA deletion and linker lig ation . Each ribosome-footprint complex has several kilo-bases of rRNA while the mRNA protected fragments are only 28 bases 2.3.2 Using Ribo-seq to decipher translation regulation Ribo-seq data can be leveraged to answer the "how much-where-how" questions involving translation regulation. The simplest of these questions, " how much " , is answered by the profiled mRNA fragments serving as a proxy to the protein levels. On the other hand , proximity-specific ribosome profiling has enabled deciphering the "where" questions. For example, the majority of mitochondrial inner membrane proteins are co-translationally translocated except the proteins that are targeted to other mitochondrial sites [70]. Not all aspects of translational control cannot be emulated in-vitro. By enabling in-vivo measurements, Ribo-seq facilitates identification of 18 z C: n re UI (1) o' 2. ,:, :!. a s· IQ ;io 6' 0 UI 0 3 (1) ii! n ~ '< t Cell lysls )I, C: 0 a ·;: u UI C: f - GI I!! GI > GI 0:: C: 0 5'-0P03 / ) t II I I ? t :; A g ... ~ C: :J -oH-3' t - -oH-3' Footprint fragment purification Figure 2.4 c ) t c · . :::: : ) l - C: .2 '1ii N ·;: ca "S u ... c3 C: 0 ~ u ~ ii E <( 0:: 0 Q. l~ I UI 1 'iij C: ca ,:, C: ca Cl C: 'ij C: GI :::, CT GI "' Ribo-seq protocol The protocol encompasses four major steps. In the first step, the cell is lysed and harvested under conditions to ensure in-vivo positions of the ribosomes are unaffected. The cell lysate can then be digested using a nuclease which will digest all portions of the mRNA not being protected by the ribosome. The portion being protected by the ribosome called the ribosome footprints can then be purified and ligated to a single-stranded linker molecule that serves as a priming site for reverse transcription. The products of first-strand reverse transcription can be circularized to provide a second priming site flanking the original captured footprint sequence which is then used for PCR amplification of a deep-sequencing library. 19 translation mechanism that vary across cell states and organisms thus answering the "how" questions. For example, Ribo-seq has been used to understand the role of Dom34 in rescuing stalled ribosomes [71]. If multiple isoforms are co-expressed , translation initiated at the upstream initiation sites can obscure the strength of initiation signal downstream. However, treatment with certain drugs such as harringtonine [72], or lactimidomycin [26] which is capable of immobilizing the initiating ribosomes will result in an overabundance poof ribosome protected fragments (RPFs) at the initiation sites. Translation initiation inhibitors have been used to discover alternate translation initiation sites [73]. The number of footprints originating from a transcript directly correspond to the number of ribosomes engaged in translation . This is also equivalent to a) the amount of protein being pro duced and b) the time required to synthesize it. The time to translate an an ORF is proportional to its length. Ribo-seq has been used to provide empirical evidence that the speed dynamics of translation are consistent across different group of genes [74]. The number of ribosomes over a gene indicates how many ribosomes are translating it. Similarly, the number of footprints over a particular codon indicates how often do ribosomes hit that particular spot. If ribosomes pause at a particular location while translating a gene, then ribosomes will spend a longer duration of time at these codons which would lead to an over-abundance of footprints at this location. Ribo-seq data has been used to discover ribosomal pausing sites in bacteria [69], yeast [75] and mammals [76]. In the upcoming chapters we rely on Ribo-seq data to answer some key questions on mech anisms of translational regulation . In particular, we first present a new computational method 20 to identify short regions which are actively engaged by the ribosomes and apply this method to discover prevalent translation in the upstream leaders across species. 21 Chapter 3 Accurate detection of short and long ORFs from Ribo-seq data 3.1 Introduction The process of translating messenger RNA into protein is among the greatest investments of energy by cells [12]. Consequently, translation is highly regulated to ensure that each cell synthesizes the right amount of each protein . Our understanding of the mechanisms regulating the translational process remains limited , which has motivated the development of experimental approaches to profile the translation landscape globally. Ribo-seq [24] is a technology that uses deep-sequencing to identify ribosome-protected fragments, revealing the positions of the entire pool of ribosomes engaged in translation . Ribo-seq has led to the surprising discovery of prevalent translation through non-canonical ORFs [77]. The non-canonical ORFs include the upstream ORFs (uORFs) located in the 5' untranslated region (UTR) , the downstream ORFs (dORFs) located in the 3' UTR, and the ORFs within presumed non-coding genes [78]. 22 A human 2.5 2.0 >, - · u51 .5 C (]) 0 1.0 0.5 0.0 0 1 2 3 4 5 ORF length (log10) B - CCDS (transcript) - CCDS ( exon) - uORF - dORF - novel Figure 3.1 mouse 2.5 ----------- 2.0 ~ ·w 1.5 C (]) 0 1.0 0.5 0 1 2 3 4 5 ORF length (log10) Length distribution of candidate ORFs for human and mouse. The length distribution of uORFs, dORFs, and novel ORFs predicted from presumably non-coding genes compared with the CCDS exon and CCDS transcript lengths (CCDS = Canonical Coding Sequence; uORF = upstream ORF in 5' UTR; dORF = downstream ORF in 3' UTR; novel = candidate ORFs in annotated non-coding genes.) Transcriptome-wide searches for pairs of in-frame start and stop codons defining potential ORFs in human, and mouse genomes reveal that the sizes of such non-canonical ORFs are gen- erally 10-20 fold shorter [79] than the conventional coding sequences (CDS) (Figure 3.1). Their short size presents challenges in detecting the resulting peptides through proteomic approaches [ 80]. However, there is emerging evidence that these short ORFs, or the products of their transla tion , serve some function [81 , 58]. In particular, the role of uORFs in regulating the translation of downstream CDS has been well documented [ 82] for individual genes [ 83], and they are correlated with substantial (30%-80%) repression of protein production [79]. The same mechanism is also used to encode condition-specific activation : in integrated stress response, where the repressed state is the default, uORF-associated repression is released following the stress stimulus [84]. 23 The presence of amplification bias, non-ribosomal RNA-protein complexes or other non ribosomal contamination can often result in apparent RPFs that do not represent actively translating ribosomes. Some RNAs such as telomerase RNA, RNAse P, snRNAs, and snoRNAs that are known to be "classical" non-coding RNAs and are predominantly localized in the nucleus have also been reported as origin for RPFs [30]. This is an indication that not all RPFs rep resent actively-translating ribosomes. Such fragments could represent non-ribosomal protected regions such as those protected by RNA binding proteins. When drawing any conclusion about translational regulation from Ribo-seq data it is imperative to focus only on those fragments that represent actively-translating ribosomes. However, the presence of noise in the data makes the task of identifying actively translated regions challenging. A shorter translation unit means less total data on average for inference, so detection of short ORFs in Ribo-seq has remained especially difficult. Several methods exist for analyzing Ribo-seq data to determine the coding potential of the transcribed RNA. FLOSS [31], one of the earliest methods, identifies actively translating ORF by focusing on the read length distribution. The key assumption is that the distribution of sequenced fragments contains both RPFs and technical noise, and the true RPFs should exhibit a particular length distribution. FLOSS first learns a reference distribution of RPF lengths on a set of protein-coding genes likely to represent active translation, and then compares fragment lengths through the other regions in the transcriptome to this reference distribution. The idea of treating different fragment lengths separately has been adopted in several subsequent methods. Most other methods can be understood broadly through two paradigms. The first hypothesizes that the distribution of number of mapped fragments differs over actively translated regions, and 24 compares this distribution with some selected null model. The other general approach exploits the periodic pattern in the mapped fragment profiles to distinguish actively-translating regions. In the first paradigm of methods, ORFscore [85] compares the distribution of reads falling in the three frames to a uniform distribution. ORF-RATER [33] uses a combination of regression and random-forest based classification to predict actively-translating ORFs. It uses a non-negative least squares fit for regressing Ribo-seq read profile of the transcript against the profile obtained from known protein-coding genes. A random-forest classifier then uses these scores to predict the translational status of the ORF. RiboHMM [86], on the other hand, uses a hidden Markov model (HMM) to detect translating ORFs. It models the contribution of each fragment length separately and then combines them to increase sensitivity. The HMM learns the distributions of Ribo-seq coverage over the start/ stop codons and the translated CDS; the distributions are then used to predict translation status for candidate ORFs. Rp-Bp [36] uses probabilistic modeling to estimate if read counts at each position belong to an enriched model or a null uniform model. RiboCode [87] uses a modified Wilcoxon signed-rank test [88] to assess periodicity by testing for differential enrichment in one of the frames against the other two. The second paradigm typically leverages spectral approaches to examine the periodic pattern in Ribo-seq data. Mapping RPFs from Ribo-seq onto the mRNA is expected to reveal a "high low-low" pattern, owing to ribosome's movement over codons, resulting in a three-nucleotide periodicity. RiboTaper [35] uses multi-tapered windows for calculating a Fourier transform to as sess periodicity in the Ribo-seq signal. Based on related principles in signal processing, SPECtre [89] makes use of spectral coherence to correlate Ribo-seq signal with the expected "high-low-low" 25 pattern . RiboWave [90] uses a wavelet transform based method to denoise the RPF profile by ex tracting the three-nucleotide periodicity. This denoised RPF profile leads to a better performance when identifying active translation . Methods within both paradigms have enabled discovery of actively-translating ORFs. Each method makes assumptions about the data that are not always satisfied in practice, for different data sets or different data analysis goals. The detection of short ORFs is an example of the latter. However, these methods provide a conceptual foundation that we borrow from to design a simplified method that is more robust to varying statistical features across datasets, and that is capable of detecting both short and long ORFs. Our method , called ribotricer, directly assesses the three-nucleotide periodicity in Ribo-seq data. Ribotricer can account for read length specific P-site offsets and sparsity in Ribo-seq data. Its underlying model emphasizes consistency in the qualitative profile through each codon while down-weighting the influence of the magnitude of the individual values contributing to that profile. This approach helps ribotricer overcome the challenge of detecting short ORFs in regions of low signal to noise ratio. 3.2 Methods To detect actively-translating ORFs, ribotricer focuses on the characteristic three-nucleotide periodicity in Ribo-seq data. The workflow of ribotricer consists of five major steps. Ribotricer first prepares a candidate set of all potentially translatable ORFs by searching for pairs of start and stop codons genome-wide but inside annotated transcription units. This requires providing gene annotations and the reference genome but is only done once for each genome and gene annotation. Next, ribotricer partitions the mapped reads based on their length. The rationale 26 for processing reads by their length is that each length may be associated with a different P-site offset relative to the 5' end of the mapped fragment. For each read length, ribotricer generates a metagene profile using 5' ends of the mapped reads (accounting for strand as appropriate) . The metagene profiles are used to infer P-site offsets for different read lengths by choosing the offsets that maximize the cross-correlation of these profiles with the profile for the most abundant read length. The read profiles corresponding to different read-lengths can then be merged using the corresponding inferred P-site offsets, an approach taken previously by Calviello et al. for RiboTaper [35] and Xiao et al. for RiboCode [87]. The previous step produces a single RPF profile for each candidate ORF. In its final step, ribotricer assesses the periodicity of the merged RPF profile using a novel approach to predict its translation status. Our key contribution is a novel method for assessing the three-nucleotide periodicity of RPF profile based on 3D to 2D projection (Figure 3.2). Within each codon , we may observe reads with 5' ends at each of the three nucleotides, providing three unconstrained count values. These count values can be imagined as vectors in a three-dimensional space with each nucleotide position representing one dimension. Based on early observations, we hypothesized that in practice, using information related to total read abundance might obscure the signal. This would happen , for example, if codons that are assigned a higher total read count are also more likely to receive reads associated with non-active translation . Applying any spectral method would require that the profiles satisfy conditions to ensure stationarity. Instead, we rely on using the qualitative information at each codon in the form of " high-low-low" or related pattern . This approach discards much of the quantitative information associated with individual read counts but also simplifies the problem while eliminating the need to explicitly model random variation or systematic trend in total read counts along the RPF profile. 27 • • •••••••• ~GGOQ O GG0G 3 ij Figure 3.2 Methodology design of ribotricer. For a given ORF consisting of N codons, let xij denote the number of P-sites inferred from the reads of Ribo-seq experiment aligning to the i-th codon and j -th frame of the ORF, where i = 1, 2, ... , N and j = l , 2, 3. The RPF profile of the ORF can then be denoted as we perform the following transformation to convert it into a 2D unit vector </>i = (ai, bi?: where w = (: cos (-27f /3) sin (-21r/3) cos (- 47f /3)J . sin (- 41r/ 3) (3.1) 28 With this transformation , the three basis vectors {(1, 0, 0) , (0, 1, 0) , (0, 0, 1)} are mapped as (1, 0, 0) ---+ (1 , 0), (0, 1,0) ---+ (cos (-21r/3),sin(-21r/3)), (0,0, 1)---+ (cos(-41r/3),sin (-41r/3) ). The three mapped unit vectors lie 21r /3 away from each other to ensure the direction of the transformed vector c/\ is equally determined by reads of each frame. These can be replaced by any three unit vectors that are equally spaced on the unit circle, and the results would not change. For the transformation performed , the direction of the resulting vector is determined by the relative values of xi l • xi 2 , and xi 3 . For an actively-translating ORF, we expect to see a "high-low low" pattern for each codon. This is equivalent to observing x i 1 as the largest value consistently over all codons. If this holds, we expect the directions of the resulting unit vectors c/\ to be consistent across codons. As indicated above, the motivation behind unit normalization of each vector is to help ensure that each codon contributes equally to our assessment of translation status, avoiding bias from the fraction of codons with an over-abundance of reads. This trans formation disregards the total read counts at each of the three positions. For example, the two codon profiles (100, 20, 10) and (10, 2, 1) will result in the same unit vectors when applying equation Eq ., (3.1). While this discards quantitative information, it still captures the qualitative " high-low-low" pattern of the profile. This approach helps ribotricer handle the heterogeneous nature of Ribo-seq data where despite of pervasive active-translation , different codons could have completely different coverages either because of the actual difference in ribosome's dwell time or because of usage of drugs like cycloheximide which can alter codon-specific elongation rates [91]. 29 The Z 2 -norm of the mean vector of the transformed vectors can be used to assess the peri odicity of RPF profile. More consistent directions of the vectors would result in a larger Z 2 -norm. The mean vector of the transformed vectors is and its Z 2 -norm 11¢ 11 is - ✓( 1 N ) 2 ( 1 N ) 2 11 ¢ 11 = N L i=l ai + N L i=l bi , which falls in [O, 1], with a value of 1 if and only if in which case the directions for all vectors are the same. Besides heterogeneity arising from uneven distribution of read counts across codons [92], another key challenge in Ribo-seq data is sparsity leading to profiles with many empty codons, i.e., codons to which no reads map. Since both actively-translating and non-translating ORFs can have empty codons, they do not contribute any information about the translation status and hence need to be handled specially. For a particular data set with N codons, define the set V of non-empty codons as V = { i = 1, 2, ... , N I xi -=/= (0, 0, 0)}, 30 and let Nv = IVI. If we define </>* as the mean vector including only non-empty codons, the ratio between 11 ¢ 11 and 11¢* 11 is 11¢ 11 NV 11¢* 11 N . With the reasoning outlined above, we use 11¢* II as our measure for assessing the periodicity of the RPF profile of an ORF. This score describes how "aligned" all the vectors are, and is equivalent to measuring how similar the phases are, i.e., the angles created by the resulting vectors with respect to the abscissa. We will refer to this score as the "phase score" hereafter. Note that in theory, a high phase score may result from strong consistency of some pattern other than the anticipated "high-low-low". In designing our approach , we hypothesized that the only source of consistency in the signal would be an active translation . A consistent "low-high-low" or "low- low-high " pattern would most likely result from an inaccurate estimate of the P-site offsets, in which case our assumptions add a layer of robustness. The angles made by the resultant vectors when all the codons follow a "high-low-low" pattern should be concentrated around 0. The distribution we observe for the Ribo-seq data is centered around O (Figures 3.7 and 3.8), which confirms that most codons follow the "high-low-low" pattern . For the RNA-seq data, the resulting angles follow a multimodal distribution with the highest peaks at {-21r /3, 0, 21r /3} (Figures 3.7 and 3.8) which corresponds to the three unit vectors. To interpret the multimodal distribution observed in RNA-seq data, we simulated read counts using a Poisson distribution. To account for variation in total data between genes, we simulated means of the Poisson distribution using the per nucleotide coverage from the RNA-seq . 31 700.0 K ~ _gi 600.0 K Sl 500.0 K ;;. 400.0 K ill 300.0 K 6 200.0 K c9 100.0 K o:: 0.0 'l<c "' 2.4 M ~ ~ 2.0M ~ 1.6M :if 1.2M 6 800.0 K .o 400.0 K i:i: 0.0 'l<c 6.4M; _gi 5.6M Sl 4.8M ~ 4.0M :if 3.2 M 6 u~ c9 800.0 K ! 0.0 SRX118286 Total reads= 3.3 M ___ .... 11~ •.... _ _ __ _ 1111111111111111111111111111111 20 25 30 35 40 45 50 SRX345309 Total reads= 16.4 M -·•···•111111 ... _ __ ---- 1111111111111111111111111111111 20 25 30 35 40 45 50 SRX1254413 Total reads= 19.2 M ___ __ . ..iii._ I q11111111 'II 11 'II 1111111 'II 11 'I 20 25 30 35 40 45 50 SRX2536421 Total reads= 10.7 M 3.2M; I "' 2.8M ~ 2.4M ~ 2.0 M I ~ 1BM I ill 1.2 M 6 800.0 K c9 400.0 K I ! 0~ ---- --• ••--- --- 1111111111111111111111111111111 20 25 30 35 40 45 50 SRX2647167 Total reads= 3.3 M 1 ~~~ :~ ~ ~ 11 ;;. 450.0 K I ill 300.0 K _8 150.0 K I i:i: o~ I 'l<c ----- -•• ••-- - 1111111111111111111111111111111 20 25 30 35 40 45 50 _gi 1.1 M 1.2M; ro 900.0 K ~ 750.0 K :if 600.0 K "' 450.0 K <i: 300.0 K ~ 150.0 K 'l<c 0.0 "' 4.2M ~ ~ 3.6M ~ 3.0M ~ 2.4M ill 1.8M <i: 1.2 M z 600.0 K o:: 0.0 'l<c "'22.5M i -o 20.0 M Sl 17.5M ~ 15.0M :if 12.5M '? 10.0M <I'. 7.5 M ~ 5.0M 'l<c 2.5M 0.0 "' 2.8M ~ ~ 2.4M ~ 2.0M ~ 1.6M ill 1.2M <i: 800.0 K z 400.0 K o:: 0.0 'l<c "'9.0M i -o 8.0 M Sl 7.0M ~ 6.0M :if 5.0M '? 4.0M <I'. 3.0 M ~ 2.0M 'l<c 1.0M 0.0 Figure 3.3 SRX118285 Total reads= 12.3 M ---····11111111111 __ 1 I I I I I I I I I I I I I I I I I I I I I 20 25 30 35 40 SRX345311 Total reads = 54.9 M _____ ...... 11111111~111 •.. _ _ _ _ 1111111111111111111111111111111 20 25 30 35 40 45 50 I I 90 SRX426378 Total reads= 24.2 M I I I 95 _I I I II I I I 100 105 SRX2536426 Total reads = 26.1 M ____ _ .11111111 •... _ _ _ I I 110 1111111111111111111111111111111 20 25 30 35 40 45 50 I 45 SRX2647164 Total reads= 9.7 M I I I I 50 I I 55 I I 60 Read length distribution of Ribo-seq and RNA-seq samples from human datasets. SRA sample accession and total uniquely mapping reads are shown in individual subplots. 32 "' 2.4 M i " 2.1 M Sl 1.8M ~ 1.5M g 1.2M o/ 900.0 K _g 600.0 K ii: 300.0 K at 0.0 ~ 12.0 M "'14.0M ~ ~ 10.0 M CT 8.0 M ~ 6.0M 6 4.0M '9 2.0 M o:: 0.0 at i ;:~~ ~ ;;_ 4.5M ~ 3.0M _g 1.5 M a: 0.0 at ~ 4.8M "' 4.0M ~ 3.2M SRX026871 Total reads= 12.3 M _______ .1ll11L.I 1111111111111111111111111111111 20 25 30 35 40 45 50 SRX1149649 Total reads= 50.3 M 1111111111111111111111111111111 20 25 30 35 40 45 50 SRX1900396 Total reads= 47.1 M ......... 1111 .• 11 _ _ _ _ , 1111111111111111111111111111111 20 25 30 35 40 45 50 SRX2255510 Total reads= 30.4 M 5.6M ~ fil" 2.4 M I 6 1.6M I ,g 800.0 K I ~ 0 . 0 111111 "······· ......... ·~ .. . . ..... . ~ 28.0 M 32.0 M I "'24.0 M ~ 20.0 M CT 16.0 M ~ 12.0 M 6 8.0M ~ 4.0M at 0.0 1111111111111111111111111111111111111111111111111111111111111 20 25 30 35 40 45 50 55 60 65 70 75 80 SRX3110803 Total reads = 113.9 M ___ ______ ,1IIL.I __ 1111111111111111111111111111111 20 25 30 35 40 45 50 ~ 640.0 K 720.0K I "' 560.0 K ~ 480.0 K g 400.0 K o/ 320.0 K <C 240.0 K ~ 160.0 K at 80.0 K "'14.0M ~ " 12.0 M Sl 10.0 M ;;_ 8.0 M ~ 6.0 M J: 4.0 M z 2.0 M o:: 0.0 at "' 2.4 M; " 2.1 M Sl 1.8 M ~ 1.5M g 1.2M o/ 900.0 K <C 600.0 K ~ 300.0 K at 0.0 2.1M ~ ~ 1.8M Sl 1.5M ;;_ 1.2 M ~ 900.0 K J: 600.0 K z 300.0 K o:: 0.0 at 4.8M I ~ 4.2 M gi 3.6 M ~ 3.0 M g 2.4 M o/ 1.8 M <C 1.2 M ~ 600.0 K at 0.0 Figure 3.4 I I 20 SRX026872 Total reads= 9.5 M I I I I I I I I I I I I I 25 30 35 SRX1149668 Total reads= 56.2 M I I 40 _ _ _ .11111111 _ 1 1111111111111111111111111111111 20 25 30 35 40 45 50 SRX1900402 Total reads= 35.5 M ... 111111111111~11 .. _ _ _ , 1111111111111111111111111111111 20 25 30 35 40 45 50 SRX2255511 Total reads= 54.5 M ··· ·····""' 111111111111111, ....... . 1111111111111111111111111111111111111111111111111111111111111 20 25 30 35 40 45 50 55 60 65 70 75 80 SRX3110807 Total reads= 47.6 M ____ .1111111111. ___ _ 1111111111111111111111111111111 20 25 30 35 40 45 50 Read length distribution of Ribo-seq and RNA-seq samples from mouse datasets. SRA sample accession and total uniquely mapping reads are shown in individual subplots. 33 (/) cl "' ~ 10.0 C "' 7.5 Q) E cl 5.0 Q) _ !:,! "iii 2.5 E 0 0.0 z -20 (/) cl "' ~6 C "' ~4 cl Q) &i2 "' E ~o -20 (/) cl "' ~4 C "' Q) E ~2 _!:,! "iii E ~o -20 (/) cl m4 ~ 3 Q) E cl 2 l!l ro 1 E ~o 0 0 -20 0 (/) cl "' ~4 C IB 3 E ~2 _!:,! ~ 1 ~o -20 0 0 SRX118286 (28 nt reads) phase_score: 0.99 20 40 60 80 100 120 140 160 180 Distance from start codon (nt) SRX345309 (29 nt reads) phase_score: 0.97 20 40 60 80 100 120 140 160 180 Distance from start codon (nt) SRX1254413 (29 nt reads) phase_score: 1.0 20 40 60 80 100 120 140 160 180 Distance from start codon (nt) SRX2536421 (28 nt reads) phase_score: 1.0 20 40 60 80 100 120 140 160 180 Distance from start codon (nt) SRX2647167 (30 nt reads) phase_score: 0.99 20 40 60 80 100 120 140 160 180 Distance from start codon (nt) Figure 3.5 Metagene plots for representative read lengths for human Ribo-seq samples. SRA sample accession, read length and phase score are shown in individual subplots. 34 SRX026871 (30 nt reads) V, phase_score: 0.99 -g 6 ~ C IB4 E -c, -~ 2 <ii E 50 z -20 0 20 40 60 80 100 120 140 160 180 Distance from start codon (nt) SRX1148649 (29 nt reads) V, phase_score: 0.98 -c, "' ~ 10.0 C "' Q) 7.5 E -c, 5.0 Q) - ~ <ii 2.5 E 5 0.0 z -20 0 20 40 60 80 100 120 140 160 180 Distance from start codon (nt) SRX1900396 (30 nt reads) V, phase_score: 1.0 -c, ~6 C "' ~4 -c, Q) ~2 "' E ~o -20 0 20 40 60 80 100 120 140 160 180 Distance from start codon (nt) SRX2255510 (28 nt reads) V, phase_score: 0.99 -c, "' ~ 20 C IB 15 E ~ 10 -~ <ii 5 E 5 0 z -20 0 20 40 60 80 100 120 140 160 180 Distance from start codon (nt) SRX3110803 (30 nt reads) V, phase_score: 0.99 -c, "' ~8 C IB 6 E ~4 N ~2 E ~o -20 0 20 40 60 80 100 120 140 160 180 Distance from start codon (nt) Figure 3.6 Metagene plots for representative read lengths for mouse Ribo-seq samples. SRA sample accession, read length and phase score are shown in individual subplots. 35 Ribo-seq RNA-seq 10 0.8 8 6 0.6 4 0.4 2 0.2 1 I I 0 .L.....,--',l,_...,..,.. .......... ..,..._..J 0.0...._'--r-~~"-r'-~ .,...... 1.50 ~-----~ 0.8~-----~ 1.25 m ! ~: i 0.50 0.6 0.4 0.2 0.8 0.6 0.4 0.2 0.4 0.3 0.2 0.1 Poisson 0.25 0.00....._...,,__,_-"-.=~ .,,............,..,.__, ...._.-~~-~ ........ 0.0, ......... ~-~~~-r'-' 15 0.8 0.6 0.4 i 0.2 0 .L.....,-_,._ ............... ......... _, o .o ..._-~~--.--~.,...... 3 2 0.8 0.6 0.4 0.2 0 A.. . A oo .L.....,--',),~1!:,-1-....,._~.........., . ...._'--r--=--r=-~ "-r'-.......--=-.,...... t.-~~:~'? <:) ~'?'i~'? ~ Angle t.-~~:~'? <:) ~'?'i~'? ~ Angle Figure 3.7 0.6 0.4 0.2 0.8 0.6 0.4 0.2 t.-~~:~'? <:) ~'?'i~'? ~ Angle Distribution of the resulting vector angles for datasets in human. Angles are formed by projecting the CCDS 3D codon profiles to 2D unit vectors. The left sub-panel indicates the distribution for Ribo-seq sample; the center sub-panel shows the distribution for its corresponding RNA-seq sample; the right sub-panel shows the distribution of angles resulting from a RNA-seq profile simulated from a Poisson distribution with the mean parameter estimated from the RNA-seq data. 36 Ribo-seq 15 O.L.....,-... 1 _...,... . -..+-...._t-----l . 20 0 .t......,.._..._ , ,,_....,,..... __ _J 10 8 I I : 2 0.L.,. ........... M,a .............. .,.._J i I I . 5~-----~ 4 0.8 0.6 0.4 0.2 0.0 0.8 0.6 0.4 0.2 0.0 0.6 0.4 0.2 0.0 0.6 0.4 0.2 0.0 0.8 0.6 0.4 0.2 Poisson RNA-seq ~-----~ 1.0,~-----~ 0.8 0.6 0.4 1.0 0.8 0.6 0.4 0.2 -t'--..----,c----T--r--r---,,...., 0. 0 0.4 0.3 0.2 0.1 0.5 0.4 0.3 0.4 0.3 0.2 0.1 I\ I\J ,A /\I ,A IA o -'---,-..,.......,_,.,."+-'..,,..............., o.o ...._--.-----.----.----r---.--........ o. o,~ --.----.----r----.----r-....,.._. f:.-~~~,l'J ~ ~"J'l,~"J ~ f:.-~~~~"J ~ ~"J'l,~"J ~ f:.-~~~~"J ~ ~"J'l,~"J ~ Angle Angle Angle Figure 3.8 Distribution of the resulting vector angles for datasets in mouse. Angles are formed by projecting the CCDS 3D codon profiles to 2D unit vectors. The left sub-panel indicates the distribution for Ribo-seq sample; the center sub-panel shows the distribution for its corresponding RNA-seq sample; the right sub-panel shows the distribution of angles resulting from a RNA-seq profile simulated from a Poisson distribution with the mean parameter estimated from the RNA-seq data. 37 The resulting angle distribution of the simulated codon profiles is similar to that obtained from profiles of the RNA-seq data (Figures 3.7 and 3.8) which explains the observed multimodality. 3.2.1 Obtaining and pre-processing data We downloaded the raw data (Table 3.4) from NCBl 's Sequence Read Archive (SRA) using pysradb [93]. We used cutadapt [94] to perform adapter trimming. The specific adapters for each dataset are either obtained from the corresponding papers or were automatically inferred by checking for over-represented k-mers at the 3'end. Sequences of the adapters for each dataset is documented in Table 3.1. All the Ribo-seq and RNA-seq data were mapped using STAR [95] by allowing at most two mismatches (--outFilterMismatchNmax 2) and forcing end-to-end (--alignEndsType EndToEnd) read alignment. Only uniquely mapping reads were retained (-outFilterMultimapNmax 1). For human and mouse, we relied on the GENCODE [96] GTF for annotation. For all other species except C. albicans, we used ENSEMBL [97]. For C. albicans, both the FASTA and the GTF were obtained from the Candida Genomes database [98]. The assembly and GTF information is summarized in Table 3.2. FASTA is handled using the pyfaidx package [99]. The strand-specific protocol, either forward stranded , reverse stranded or unstranded, is in ferred by checking the first 20, 000 reads from the mapping results. Since most tools we compared with can only deal with forward stranded protocol , our ten datasets are all forward stranded for both RNA-seq and Ribo-seq samples. BAM files are processed using pysam , a python interface to samtools [100]. 38 Table 3.1 Adapters trimmed from Ribo-seq and RNA-seq samples for each dataset. SRA Accession SRP010679 SRP029589 SRP063852 SRP098789 SRP102021 SRP003554 SRP062407 SRP078005 SRP091889 SRP115915 SRP108862 SRP087624 SRP029587 SRP059391 SRP018118 SRP075766 SRP033499 SRP028614 SRP028552 SRP000637 SRP056647 SRP026198 SRP014427 SRP010374 SRP108999 SRP028243 SRP076919 SRP045475 SRP056012 SRP045777 ERP007231 SRP034750 SRP010040 SRP023492 SRP032814 SRP107240 SRP062129 Ribo-seq adapter CTGTAGGCAC CTGTAGGCACCATCAAT None CTGTAGGCACCATCAAT TCGTATGCCGTCTTCTGCTTG TCGTATG TGGAATTCTCGGGTGCCAAGG TGGAATTCTCGGGTGCCAAGG AGATCGGAAGAGCACACGTCT TGGAATTCTCGGGTGCCAAGG TGGAATTCTCGG AGATCGGAAGAGC TCGTATGCCGTCTTCTGCTTG TGGAATTCTCGG TGGAATTCTCGG AGATCGGAAGAGC AGATCGGAAGAGC AAAAAAAAAAA_AGATCGGAAGAGC AGATCGGAAGAGC AAAAAAAA_AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AAAAAAA_AGATCGGAAGAGC AGATCGGAAGAGC CTGTAGGCACCATCAAT AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGCACACGTCTGAACTCCAGTCA ATCTCGTATGCCGTCTTCTGCTTGAAAAAAAAA AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC TGGAATTCTCGG RNA-seq adapter CTGTAGGCAC CTGTAGGCACCATCAAT None CTGTAGGCACCATCAAT None TCGTATG TGGAATTCTCGGGTGCCAAGG TGGAATTCTCGGGTGCCAAGG AGATCGGAAGAGCACACGTCT TGGAATTCTCGGGTGCCAAGG AGATCGGAAGAGC AGATCGGAAGAGC TGGAATTCTCGGGTGCCAAGGAACTCCAGTCAC TGGAATTCTCGG TGGAATTCTCGG AGATCGGAAGAGC AGATCGGAAGAGC AAAAAAAAAAA_AGATCGGAAGAGC AGATCGGAAGAGC AAAAAAAA_AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC TGGAATTCTCGG AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGCACACGTCTGAACTCCAGTCA ATCTCGTATGCCGTCTTCTGCTTGAAAAAAAAA AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC AGATCGGAAGAGC 39 Table 3.2 Reference assemblies and GTF for each species Species Reference assembly GTF Human GRCh38 Gencode (v94) Mouse GRCm38 Gencode (v94) Arabidopsis TAIRlO ENSEMBL (v96) C.elegans WBcel235 ENSEMBL (v96) Drosophila BDGP6 ENSEMBL (v96) Rat Rnor6.0 ENSEMBL (v96) Zebrafish GRCzll ENSEMBL (v96) C. albicans SC5314 Candida Genomes Database (r27) 5. pombe ASM294v2 ENSEMBL (v96) Chimpanzee Pantro3 ENSEMBL (v96) Macaque Mmul8 ENSEMBL (v96) To create fragment length specific metagene profile, we counted the number of 5' end of reads at each nucleotide per fragment length . Figures S2 and S3 show the distribution of frag- ment lengths for Ribo-seq and RNA-seq samples across different datasets in human and mouse, respectively. Metagene plots for individual fragment lengths which were retained for downstream analysis for different datasets are shown in Figures 3.5 and 3.6 The specific Ribo-seq and RNA-seq samples used from each dataset for the benchmarking along with the read lengths and the corresponding P-site offsets used for the Ribo-seq samples can be found in Table 3.3. To evaluate the performance of ribotricer and other existing methods, acknowledging the heterogeneity and appreciable noise levels in Ribo-seq data, we selected five human and five mouse datasets for performance comparison (Table 3.4). This includes the human HEK293 cells dataset (SRA accession : SRP063852) [35], which was originally used as a benchmark dataset when Ribo Taper was introduced [35] and subsequently used in other studies. We followed the strategy previously established by Calviello et al. in assessing Ribo Taper [35] and Xiao et al. in assessing RiboCode [87]. For all the ten datasets, we obtained the RPF profiles for all the CCDS 40 Table 3.3 Ribo- and RNA-seq samples used for the benchmarking along with the read lengths and P-site offsets used for Ribo-seq samples. SRA Accession Ribo-seq sample Read lengths ( nt) P-site offsets ( nt) RNA-seq sample Species SRP010679 SRX118286 28,29,30 12,13,13 SRX118285 Human SRP029589 SRX345309 29,30,32 12, 12, 13 SRX345311 Human SRP063852 SRX1254413 28,29,30 12,12,12 SRX426378 Human SRP098789 SRX2536421 28,30 12,13 SRX2536426 Human SRP102021 SRX2647167 28,29,30,31 12, 12, 12, 12 SRX2647164 Human SRP003554 SRX026871 28,29,30 12,12,12 SRX026872 Mouse SRP062407 SRX1149649 28,29,30,31 12, 12, 12, 12 SRX1149668 Mouse SRP078005 SRX1900396 26,27,28,29,30 12, 12, 12, 12, 12 SRX1900402 Mouse SRP091889 SRX2255510 26,27,28,29,30 12, 12, 12, 12, 12 SRX2255511 Mouse SRP115915 SRX3110803 29,30,31,32,33,34 12, 12, 12, 13, 13, 13 SRX3110807 Mouse SRP108862 SRX2896566 23 12 SRX2896570 Arabidopsis SRP087624 SRX2148419 28,29,30,31,32 12, 12, 12, 12, 12 SRX2148418 Arabidopsis SRP029587 SRX345240 26,27 12,12 SRX345251 Arabidopsis SRP059391 SRX1056790 27,30 12,12 SRX1056791 Arabidopsis SRP018118 SRX219170 28,29,30,31 11,12,13,13 SRX347226 Arabidopsis SRP075766 SRX1801603 26,27,28 11,12,13 SRX1801650 Baker's Yeast SRP033499 SRX386988 29,30,31 12,12,12 SRX386983 Baker's Yeast SRP028614 SRX333052 28,29,30 12,13,13 SRX334053 Baker's Yeast SRP028552 SRX332185 28,29,30 11,12,12 SRX332188 Baker's Yeast SRP000637 SRX003187 28,29,30,31 12, 12, 12, 12 SRX003191 Baker's Yeast SRP056647 SRX971770 28,29,30,31,32 12, 12, 12, 12, 12 SRX971774 C. e/egans SRP026198 SRX311784 29,30,31,32 12, 12, 12, 12 SRX311777 C. elegans SRP014427 SRX160518 28,29,30,31,32 12, 12, 12, 12, 12 SRX160149 C. elegans SRP010374 SRX118118 28,29,30,31,32 12, 12, 12, 12, 12 SRX118116 C. elegans SRP108999 SRX2902857 29,30,31,32 12, 13, 10, 12 SRX2902867 Drosophila SRP028243 SRX327686 28,29,30,32,33,34 12, 12, 12, 12, 12, 13 SRX327688 Drosophila SRP076919 SRX1870218 34 12 SRX1870191 Drosophila SRP045475 SRX679371 28,29,30,31,32 12, 12, 12, 12, 12 SRX679372 Drosophila SRP056012 SRX915217 29,30,31,32 12, 12, 13, 13 SRX915210 Rat SRP045777 SRX686499 28,29,30,31 12, 12, 12, 13 SRX686500 Rat ERP007231 ERX609893 28,29,30,31,32 12, 12, 12, 12, 12 ERX609898 Rat SRP034750 SRX399800 28,29,30,31 12, 12, 12, 12 SRX399817 Zebrafish SRP010040 SRX113357 27,28,30,31,33,34 12, 12, 12, 12, 12, 12 SRX113344 Zebrafish SRP023492 SRX288475 28,29,30 12,12,12 SRX288474 Zebrafish SRP032814 SRX375317 28,29,30 12,12,12 SRX375318 C. albicans SRP107240 SRX2825796 28,29,30 12,13,13 SRX2825805 5 . pombe SRP062129 SRX1135820 28,29,30 12,12,12 SRX333018 (SRP028612) Chimpanzee SRP062129 SRX1135825 28,29,30 12,12,12 SRX333023 (SRP028612) Macaque 41 from the results generated by Ribo Taper and used the expressed CCDS profiles from Ribo-seq data as true positives and the corresponding CCDS profiles from RNA-seq data as true negatives. Since Ribo Taper was designed and benchmarked for detecting active translation at the exon level , we split the existing methods for active translation detection into two groups; those that support detection at the exon level and those that only allow detection at the transcript level. We compared the performance of ribotricer at both the exon and transcript levels. 3.2.2 Learning cutoff of phase score The phase score is indicative of how consistent the profile is through a defined region. We re quire some cutoff to distinguish phase scores that differentiate active from non-active translation , with the latter representing either some form of noise or inactive translation . Our approach is to learn this cutoff empirically using ten published datasets (Table 3.4) with an assumed ground truth set for regions of active translation and regions lacking active translation . Taking this strategy, we used RPF profiles of expressed Consensus Coding Sequence (CCDS) [101] exons from Ribo-seq data as the true positives, and mapped read profiles from RNA-seq data for a negative control , as previously described [35, 87]. In order to choose the best cutoff, we relied on maximizing the Fl score statistic. Fl score represents the harmonic mean of precision and recall and is considered a more realistic measure of a classifier's performance than precision or recall in isolation. The chosen cutoff should apply universally across all Ribo-seq datasets and hence should remain independent of our choice of the datasets. The ten datasets we chose in human and mouse (Table Sl) might not necessarily be representative of all Ribo-seq datasets. Hence, we performed a bootstrap [102] analysis with 50, 000 bootstraps using selected datasets, 42 defining the bootstrap statistic as the phase score that maximizes the Fl score resulting in a 95% confidence interval of (0.419, 0.469) (Figure 3.10). This resulted in a mean cutoff score of 0.444. However, for our benchmarking at the exon and the transcript level, we wanted to take the most conservative approach avoiding any bias that might arise from learning this cutoff on all the ten datasets. Hence, we used only two human and two mouse datasets (SRA accession: SRP010679 [103], SRP098789 [1], SRP003554 [104], and SRP115915 [105]), to learn the cutoff (0.428) and used the remaining six datasets to assess the performance. Note that this chosen cutoff still resides in the 95% confidence interval of our bootstrap analysis and hence can be applied universally for all other datasets. 3.3 Results 3.3.1 Ribotricer accurately detects translating ORFs at the exon level We evaluated the performances of methods that support exon-level detection of translation, including ORFscore [85], RiboTaper [35], and RiboCode [87], and compared their performance with that of ribotricer. We first compared the ability of each method to distinguish Ribo-seq profiles from RNA-seq using the area under the receiver operating characteristic (ROC) and precision-recall (PR) curve. For human HEK293 cells dataset (SRA accession : SRP063852) [35], ribotricer achieved an area under the ROC (AUROC) of 0.97. The second best one was achieved by RiboCode with an AUROC of 0.93. RiboTaper and ORFscore achieved an AUROC of 0.88 and 0.87, respectively (Figure 3.15A) . For the mouse liver tissue dataset (SRA accession : SRP078005) [108], ribotricer achieved an AUROC of 0.99 while RiboCode, RiboTaper, and ORFscore achieved AUROC of 0.97, 43 Table 3.4 List of datasets. SRA Accession Species Cell type Treatment Citation SRP010679 Human PC3 100 g/ ml cycloheximide [103] SRP029589 Human Hela cyclohexi m ide [25] SRP063852 Human HEK293 100 g/ ml cycloheximide [35] SRP098789 Human Hela 100 g/ ml cycloheximide [1] SRP102021 Human H1933 100 g/ ml cycloheximide [106] SRP003554 Mouse neutrophils 100 g/ ml cycloheximide [104] SRP062407 Mouse hippocampal neurons 100 g/ ml cycloheximide [107] SRP078005 Mouse liver 200 g/ ml cycloheximide [108] SRP091889 Mouse ESC cyclohexi m ide [109] SRP115915 Mouse liver 200 g/ ml cycloheximide [105] SRP108862 Arabidopsis inflorescences unavailable unpublished SRP087624 Arabidopsis leaf tissue 50 g/ ml cycloheximide [110] SRP029587 Arabidopsis whole seedlings 50 g/ ml cycloheximide [111] SRP059391 Arabidopsis leaf tissue 100 g/ ml cycloheximide [112] SRP018118 Arabidopsis etiolated seedling 100 g/ ml cycloheximide [113] SRP075766 Baker's Yeast strain by4743 100 g/ ml cycloheximide [114] SRP033499 Baker's Yeast strain : by4741 0.1 mg/ ml cycloheximide [115] SRP028614 Baker's Yeast strain : by4176 cyclohexi m ide [116] SRP028552 Baker's Yeast strain: s288 cyclohexi m ide [117] SRP000637 Baker's Yeast strain: by4741 100 g/ ml cycloheximide [118] SRP056647 C. elegans strain: n2 100 g/ ml cycloheximide [119] SRP026198 C. elegans strain : n2 100 g/ ml cycloheximide [120] SRP014427 C. elegans strain : n2 cyclohexi m ide [121] SRP010374 C. elegans strain : n2 cyclohexi m ide [122] SRP108999 Drosophila body wall muscle 100 g/ ml cycloheximide [123] SRP028243 Drosophila embryo 20 g/ ml emetine [124] SRP076919 Drosophila oocytes 100 g/ ml cycloheximide [125] SRP045475 Drosophila S2 cell 100 g/ ml cycloheximide [126] SRP056012 Rat PC12 Cells 100 g/ ml streptomycin [127] SRP045777 Rat Pheoch romocytoma streptomycin [128] ERP007231 Rat strain: bn/ shr 0.1 mg/ ml cycloheximide [129] SRP034750 Zebrafish strain: tuab 100 g/ ml cycloheximide [32] SRP010040 Zebrafish strain : tuab 100 g/ ml cycloheximide [130] SRP023492 Zebrafish strain : tuab 50 g/ ml cycloheximide [131] SRP032814 C. albicans strain: sc5314 10 g/ ml Blasticidin S [132] SRP107240 5. pombe strain: WT 0.15 g/ ml tunicamycin [133] SRP062129 Chimpanzee Lymphoblastoid cell line flash freezing [134] SRP062129 Macaque Lymphoblastoid cell line flash freezing [134] 44 Table 3.5 Datasets used to learn ribotricer phase score cutoffs. SRA Accession Species Used to learn cutoff SRP010679 Human Yes SRP029589 Human No SRP063852 Human No SRP098789 Human Yes SRP102021 Human No SRP003554 Mouse Yes SRP062407 Mouse No SRP078005 Mouse No SRP091889 Mouse No SRP115915 Mouse Yes SRP108862 Arabidopsis No SRP087624 Arabidopsis No SRP029587 Arabidopsis No SRP059391 Arabidopsis Yes SRP018118 Arabidopsis Yes SRP075766 Baker's Yeast Yes SRP033499 Baker's Yeast No SRP028614 Baker's Yeast No SRP028552 Baker's Yeast Yes SRP000637 Baker's Yeast No SRP056647 C. elegans No SRP026198 C. elegans Yes SRP014427 C. elegans No SRP010374 C. elegans Yes SRP108999 Drosophila Yes SRP028243 Drosophila Yes SRP076919 Drosophila No SRP045475 Drosophila No SRP056012 Rat Yes SRP045777 Rat No ERP007231 Rat Yes SRP034750 Zebrafish Yes SRP010040 Zebrafish Yes SRP023492 Zebrafish No 45 Q) .... ::::i Cl) ro Q) E LL cii u Q) 0:::: I C: 0 "iii ·u Q) .... Cl.. CX) ci (!) ci '<:I' ci N ci 0 ci 0.0 0.2 Human 0.441 I 0.4 Cutoff Figure 3.9 0.6 0.8 1.0 Learning the cutoff for phase scores for human datasets. The optimum cutoff for distinguishing actively translating regions from non-active translation was learned by maximizing the Fl score. The profiles from expressed CCDS exons in Ribo-seq data were treated as positives and corresponding profiles from RNA-seq were treated as negatives. Two datasets in human (SRA accession: SRP010679, SRP098789) were used for learning this cutoff. 46 Q) .... ::::i Cl) ro Q) E LL cii u Q) 0:::: I C: 0 "iii ·u Q) .... Cl.. CX) ci (!) ci '<:I' ci N ci 0 ci 0.0 0.2 0.418 I 0.4 Mouse Cutoff Figure 3.10 0.6 0.8 1.0 Learning the cutoff for phase scores for mouse datasets. The optimum cutoff for distinguishing actively translating regions from non-active translation was learned by maximizing the Fl score. The profiles from expressed CCDS exons in Ribo-seq data were treated as positives and corresponding profiles from RNA-seq were treated as negatives. Two datasets in mouse (SRA accession: SRP003554, and SRP115915)were used for learning this cutoff. 47 0.92 , and 0.92, respectively (Figure 3.15A) . Ribotricer also outperformed the other three methods consistently under the PR metric (Figure 3.15A). Ribotricer displayed the best performance on almost all the datasets at both ROC and PR metrics (Figures 3.35 and 3.36). Next, we compared the performance of ribotricer, ORFscore, Ribo Taper, and RiboCode by contrasting the number of true positives detected by each method while controlling the false positive rate at 0.1. We calibrated the cutoffs for each method so that the number of false positives reported by each method is 10% of the number of negatives. For human HEK293 cell dataset (SRA accession: SRP063852), ribotricer recovered 39, 517 truly translating exons, while RiboCode recovered 33,665. Ribo Taper, and ORFscore recovered 28,333 and 26, 486 translating exons, respectively (Figure 3.15B) . For mouse liver tissue dataset (SRA accession: SRP078005), ribotricer recovered 46,380 truly translating exons, RiboCode recovered 43,332, while Ribo Taper and ORFscore recovered 35, 746 and 36, 120 translating exons, respectively (Figure 3.15B) . We observed a similar performance for the other eight datasets where ribotricer consistently recovered more truly translating exons compared to the other three methods ( Figures S12). Longer ORFs have a higher chance of accumulating more ribosomes which leads to richer features when analyzing the associated RPF profiles. In order for ribotricer to be capable of detecting both short and long ORFs, the phase scores generated should be independent of the length of the ORF assessed . We investigated the effect of ORF length on the scores or the p-values generated by each method . The phase score generated by ribotricer is unaffected by the length of ORF while RiboCode, RiboTaper, and ORFscore generate a higher score or more significant P value as the ORF gets longer (Figure 3.14). Finally, we compared the performance of ribotricer with other methods in terms of Fl score using the default cutoff for each method. Since we learned the cutoff for ribotricer from four 48 C) C) 2 ~ § ~ co co Q) 6 C 6 > <D 0 <D ~ :;::; ·u; ·u; 6 6 0 ·u a. -st: AUG ~ -st: Q) 0 0.. 0 ::, N - 0.94 N - Ribotricer t= - 0.85 - RiboCode 6 - 0.80 6 - RiboTaper 0 - 0.79 0 - ORFscore 6 6 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall C) C) 2 ~ I ~ co co ~ 6 C 6 :;::; <D 0 <D ·u; 6 ·u; 6 0 ·u a. -st: AUG ~ -st: Q) 0 0.. 0 ::, - 0.85 - Ribotricer t= N - 0.70 N - RiboCode 6 - 0.67 6 - RiboT!iper 0 - 0.60 0 - ORFscore 6 6 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall C) 0 2 ~ - ' ~ ~ co co I Q) 6 C 6- > :;::; CC! 0 <D ·u; 0 ·u; 6- 0 ·u a. -st: AUG ~ st Q) 0 0.. 6- ::, - 0.97 - Ribotricer t= "! - 0.93 N - RiboCode 0 - 0.88 6- - RiboTaper C) - 0.87 0 - ORFscore 0 6- 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall 2 C) C) I ~ co co ~ ~ 6 C 6 :;::; CC! 0 CC! ·u; ·u; 0 0 ·u 0 a. -st: ~ -st: Q) 0 AUG 0.. 0 ::, - 0.97 - Ribotricer t= N - 0.91 N - RiboCode 6 - 0.88 6 - RiboTaper 0 - 0.88 0 - ORFscore 6 6 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall 2 C) C) ~ ~ co co ~ ~ 6 C 6 :;::; CC! 0 CC! ~ ·u; ·u; 0 0 ·u 0 a. st ~ st Q) 6 AUG 0.. 6 ::, - 0.96 - Ribotricer t= N - 0.93 N - RiboCode 6 - 0.90 6 - RiboTaper 0 - 0.88 0 - ORFscore 6 6 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall Figure 3.11 ROC plots and Precision-Recall plots for human datasets for exon level classification. Performance of ribotricer for detecting translating ORFs at exon level is compared with RiboCode, RiboTaper and ORFscore. The profiles of expressed CCDS exons in Ribo-seq data were treated as true positive and the corresponding RNA-seq profile as true negative. 49 C) C) 2 r ~ I ~ co co Q) 6 C 6 > <D 0 <D :;::; ·u; ·u; 6 6 0 ·u a. -st: AUG ~ -st: Q) 0 0.. 0 ::, N - 0.97 N - Ribotricer t= - 0.97 - RiboCode 6 - 0.97 6 - RiboTaper 0 - 0.92 0 - ORFscore 6 6 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall 2 C) C) I ~ co ~ co ~ 6 C 6 :;::; <D 0 <D ·u; 6 ·u; 6 0 ·u a. -st: AUG ~ -st: Q) 0 0.. 0 ::, - 0.98 - Ribotricer t= N - 0.99 N - RiboCode 6 - 0.98 6 - RiboT!iper 0 - 0.98 0 - ORFscore 6 6 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall C) 0 2 ~ - ~ ~ co co I Q) 6 C 6- > :;::; CC! 0 <D ·u; 0 ·u; 6- 0 ·u a. -st: AUG ~ -st- Q) 0 0.. 6- ::, - 0.99 - Ribotricer t= "! - 0.97 N - RiboCode 0 - 0.92 6- - RiboTaper C) - 0.92 0 - ORFscore 0 6- 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall 2 C) C) ~ ~ co co , ~ 6 C 6 I :;::; CC! 0 CC! ·u; ·u; 0 0 ·u 0 a. -st: ~ -st: Q) 0 AUG 0.. 0 ::, - 0.98 - Ribotricer t= N - 0.97 N - RiboCode 6 - 0.93 6 - RiboTaper 0 - 0.94 0 - ORFscore 6 6 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall 2 C) C) L!) ~ co co I ~ ~ ~ 6 C 6 ~ :;::; CC! 0 CC! ·u; ·u; 0 0 ·u 0 a. -st: ~ -st: Q) 0 AUG 0.. 0 ::, - 0.93 - Ribotricer t= N - 0.92 N - RiboCode 6 - 0.87 6 - RiboTaper 0 - 0.89 0 - ORFscore 6 6 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall Figure 3.12 ROC plots and Precision-Recall plots for mouse datasets for exon level classification. Performance of ribotricer for detecting translating ORFs at exon level is compared with RiboCode, RiboTaper and ORFscore. The profiles of expressed CCDS exons in Ribo-seq data were treated as true positive and the corresponding RNA-seq profile as true negative. 50 A X 10 4 SRP010679 B X 10 4 SRP003554 ~ 3 z VJ § 2 X Q) Cl) Cl 1 <) <) 0 X 10 4 SRP029589 X 10 4 SRP062407 VJ C ~ 2 Q) Cl) 8 1 <) X 10 4 SRP063852 X 10 4 SRP078005 E 4 :::, 3 z 3 VJ C 2 0 X 2 Q) Cl) 1 Cl <) <) 0 0 SRP098789 X10 4 SRP091889 X 10 4 SRP102021 X 10 4 SRP11 5915 Figure 3.13 Number of translating exons recovered when controlling the false positive rate to be the same. Performance of ribotricer is compared with RiboCode, RiboTaper, and ORFscore when the false positive rate is controlled to be 0.1. The number of truly translating exons are shown for both human {A) and mouse {B) datasets. The profiles of expressed CCDS exons in Ribo-seq data were treated as true positive and the corresponding RNA-seq profile as true negative. 51 Ribotricer RiboCode 1.0 35 ,,,,~ 30 0.8 ~ 25 Q)0.6 Ol .2 20 8 ....!...- +~~ Q) 15 (f) 0.4 :J 0.2 ~ 10 Q. 5 0.0 0 RiboTaper ORFSoore 10 10 o 8 8 ' ..- Ol ++! Q) 6 .Q 6 ....!...- 0 gi 4 c?.l 4 rn > 2 Q. 2 0 0 <:)<:) ~<:) ~<:) s:)<:) s:)<:) <:)<:) ~<:) s:)<:) s:)<:) ~<:) 1,-"'- "-(<o "<::s v.,<::s X)<::s I,,"' "<p "<::s v.,<::s X)<::s "<:) ~"-( s:)"-( '1 "<:) ~"-( ~"-( '1 <o "<::s <o "<::s CCDS exon length (nt) CCDS exon length (nt) Figure 3.14 Effect of ORF length on output scores. Distribution of scores generated by ribotricer and ORFscore, and the P-values generated by RiboCode and RiboTaper over different CCDS exon lengths. 52 real datasets, we summarized the performance of ribotricer on the remaining six datasets that were not used to learn the empirical cutoff ( Figure Sl4-Sl 7) . Notably, for human Hela cell dataset (SRA accession : SRP029589) [25], all methods achieved relatively low Fl score with the best one to be 0.67 achieved by ribotricer. We checked the angle distribution of the 3D to 2D projection described earlier for this dataset (Figure 3.7), and found that it displays high noise level compared to other datasets analyzed , which indicates low data quality. Consequently, we excluded this dataset from further analysis. For the other two human datasets, ribotricer achieved an average Fl score of 0.91, and RiboCode achieved an average Fl score of 0.84. Ribo Taper and ORFscore achieved an average Fl score of 0. 73 and 0.12, respectively. For the three mouse datasets, ribotricer achieved an average Fl score of 0.93, and RiboCode achieved an average Fl score of 0.90. Ribo Taper and ORFscore achieved an average Fl score of 0.85 and 0.55, respectively. 3.3.2 Ribotricer accurately detects translating ORFs at the transcript level ORF-RATER [33], RibORF [34], Rp-Bp [36] and RiboWave [90] only detect translating ORFs at the full transcript level. To evaluate ribotricer against these methods we use a similar to the comparison strategy as used for exon-level benchmarking. For transcript level comparison, we first used the area under ROC/ PR curves to assess the ability of different methods to distinguish Ribo seq profiles from those from RNA-seq data. For human HEK293 cell dataset (SRA accession: SRP063852), ribotricer correctly distinguished Ribo-seq profiles from the simulated RNA-seq profiles with an AUROC of 1.0, while both Rp-Bp and RibORF achieved an AUROC of 0.96. 53 A 0 ~- , 0 ·- SRP063852 AUG - 0.97 - 0.93 - 0.88 - 0.87 0.2 0.4 0.6 0.8 1.0 False positive rate - Ribotricer RiboCode RiboTaper ORFscore 0 -,--,~-~,-,--,~--, 0.0 0.2 0.4 0.6 0.8 1.0 Recall SRP078005 AUG - 0.99 - 0.97 - 0.92 - 0.92 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate - - - - Ribotricer - RiboCode - RiboTaper - ORFscore - -,--,~-~,-,--,~--, 0.0 0.2 0.4 0.6 0.8 1.0 Recall Figure 3.15 B E :::J z CJ) C 0 X Q) Cl) 0 u u E :::J z CJ) C 0 X Q) Cl) 0 u u 40000 30000 20000 10000 50000 40000 30000 20000 10000 SRP063852 SRP078005 Comparison of performance on detecting translating exons. The performance of ribotricer is compared with that of RiboCode, RiboTaper, and ORFscore. (A) The ROC and precision recall curves summarizing performance of ribotricer, RiboCode, RiboTaper and ORFscore on one human and one mouse dataset. (B) The number of translating exons recovered when controlling the false positive rate to be the same. 54 ~ CX) 2 ci ~ ~ C g? <D ci 0 ..... :;:; ·;:;; ! ·;:;; ·13 0 -sj- [I'. 0. ci a. Cl) 1.00 Ribotricer :, 0.87 RpBp ~ N ci 0.93 RibORF 0 0.72 RiboWave ci ~ F:: ==SJ CX) 2 ci I ~ Cl) <D C 0 > ci ·;:;; :;:; ·;:;; ·13 0 -sj- [I'. 0. ci a. Cl) 0.99 Ribotricer :, 0.84 RpBp ~ N ci 0.92 RibORF 0 0.62 RiboWave ci ~ 7 CX) 2 ci I ~ Cl) <D C > ci 0 :;:; ·;:;; ·;:;; ·13 0 -sj- [I'. 0. ci Cl) 1.00 a. Ribotricer :, 0.96 RpBp ~ N ci 0.96 RibORF 0 0.90 RiboWave ci ~ - Pl ~ CX) - 2 ci ~ ~ g? <D C 0- I :;:; ci ·;:;; ·;:;; ·13 0 -sj- Cl) 0. ci ct- Cl) 1.00 Ribotricer :, 0.91 RpBp ~ N ci 0.96 RibORF 0 0.83 RiboWave ci I I I I I I 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall Figure 3.16 ROC plots and Precision-Recall plots on transcript level for human datasets. Performance of ribotricer for detecting translating ORFs at transcript level is compared with RpBp, ribORF and RiboWave. The profiles of expressed CCDS transcripts in Ribo-seq data were treated as true positive and the corresponding RNA-seq profile as true negative. 55 N ci 0 ci N ci 0 ci N ci 0 ci N ci 0 ci C 0 ·u; ·c3 ~ 1.00 a. 0.90 1.00 0.85 C 0 ·u; ·c3 ~ 1.00 a. 0.95 0.99 0.90 C 0 ·u; ·c3 ~ 1.00 a. 0.97 0.99 0.76 C 0 ·u; ·c3 ~ 1.00 a. 0.93 0.99 0.88 Ribotricer RpBp RibORF RiboWave Ribotricer RpBp RibORF RiboWave Ribotricer RpBp RibORF RiboWave Ribotricer RpBp RibORF RiboWave 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate Recall Figure 3.17 ROC plots and Precision-Recall plots on transcript level for mouse datasets. Performance of ribotricer for detecting translating ORFs at transcript level is compared with RpBp, ribORF and RiboWave. The profiles of expressed CCDS transcripts in Ribo-seq data were treated as true positive and the corresponding RNA-seq profile as true negative. 56 RiboWave achieved an AUROC of 0.90 (Figure 3.26A) . For human Hela cell dataset (SRA accession: SRP098789) [1], ribotricer again perfectly distinguished Ribo-seq profiles from the simulated RNA-seq ones with an AUROC of 1.0, and Rp-Bp achieved an AUROC of 0.91. RibORF and RiboWave achieved an AUROC of 0.96 and 0.83, respectively (Figure 3.26A) . Ribotricer also consistently outperformed other methods under the PR metric (Figure 3.26A). The complete results for all human and mouse samples can be found in Figure 3.16 and 3.17. It is worth mentioning that RibORF [34] uses a classification based method which trains its model by selecting one-third of the CDS profiles as true positives which might give it an extra advantage in this comparison . Notably, here we excluded ORF-RATER from the comparison because it always reports around half the number of detected ORFs compared with other methods, as noticed by Xiao et al. previously [87]. A B SRP010679 SRP003554 SRP102021 SRP029589 SRP115915 SRP062407 - Ribotricer - RiboCode - RiboTaper - ORFscore Figure 3.18 Comparison of Fl score (exon level) of ribotricer with RiboCode, RiboTaper, and ORFscore. Performance of ribotricer is compared with RiboCode, RiboTaper, and ORFscore in terms of Fl score when the default threshold score is used for each tool. Results are shown for human (A) and mouse (B) datasets. The profiles of expressed CCDS exons in Ribo-seq data were treated as true positive and the corresponding RNA-seq profile as true negative. Next, we compared the performances of different methods by checking the number of truly translating transcripts recovered when controlling the false positive rate to be the same as 0.1. 57 A SRP010679 B SRP003554 SRP102021 SRP029589 SRP115915 SRP062407 - Ribotricer - RiboCode - RiboTaper - ORFscore Figure 3.19 Comparison of sensitivity (exon level) of ribotricer with RiboCode, RiboTaper, and ORFscore. Performance of ribotricer is compared with RiboCode, RiboTaper, and ORFscore in terms of sensitivity when the default threshold score is used for each tool. Results are shown for human (A) and mouse (B) datasets. The profiles of expressed CCDS exons in Ribo-seq data were treated as true positive and the corresponding RNA-seq profile as true negative. A B SRP010679 SRP003554 SRP102021 SRP029589 SRP115915 SRP062407 - Ribotricer - RiboCode - RiboTaper - ORFscore Figure 3.20 Comparison of specificity (exon level) of ribotricer with RiboCode, RiboTaper, and ORFscore. Performance of ribotricer is compared with RiboCode, RiboTaper, and ORFscore in terms of specificity when the default threshold score is used for each tool. Results are shown for human (A) and mouse (B) datasets. The profiles of expressed CCDS exons in Ribo-seq data were treated as true positive and the corresponding RNA-seq profile as true negative. 58 A B SRP010679 SRP003554 SRP102021 SRP029589 SRP115915 SRP062407 - Ribotricer - RiboCode - RiboTaper - ORFscore Figure 3.21 Comparison of precision (exon level) of ribotricer with RiboCode, RiboTaper, and ORFscore. Performance of ribotricer is compared with RiboCode, RiboTaper, and ORFscore in terms of precision when the default threshold score is used for each tool. Results are shown for human (A) and mouse (B) datasets. The profiles of expressed CCDS exons in Ribo-seq data were treated as true positive and the corresponding RNA-seq profile as true negative. For the human HEK293 cell dataset (SRA accession : SRP063852), ribotricer recovered 577 truly translating transcripts, while Rp-Bp, RibORF, and RiboWave recovered 508, 542, and 459 translating transcripts, respectively (Figure 3.26B). For the human Hela cell dataset (SRA acces sion : SRP098789) , ribotricer recovered 2, 251 truly translating transcripts, and Rp-Bp recovered 1, 730. RibORF and RiboWave recovered 2, 130 and 1, 308 truly translating transcripts, respec- tively (Figure 3.26 B) . Finally, we used the Fl score to assess the performance of ribotricer in detecting actively- translating transcripts in comparison with other tools. For the two human samples, ribotricer achieved an average Fl score of 0.99, and Rp-Bp achieved an average Fl score of 0.89. RibORF and RiboWave achieved an average Fl score of 0.91 and 0.75, respectively. For the three mouse samples, ribotricer achieved an average Fl score of 0.99, and Rp-Bp achieved an average Fl score of 0.87. RibORF and RiboWave achieved an average Fl score of 0.97 and 0.69, respectively (Figure 3.23). 59 A 4000 2 3500 . g- 3000 ~ 2500 _§ 2000 :i:t: 1500 1000 500 0 2500 li 2000 - ~ 1500 ! 1000 en 500 0 600 500 . 9c 400 u ~ 300 (U .;:, 200 =l:t: 100 0 2500 li 2000 - ~ 1500 ! 1000 500 0 SRP010679 SRP029589 SRP063852 SRP098789 B 2500 5000 8000 7000 6000 5000 4000 3000 2000 1000 0 4000 3500 3000 2500 2000 1500 1000 500 0 Figure 3.22 SRP003554 SRP062407 SRP078005 SRP091889 Number of translating transcripts recovered when controlling the false positive rate to be the same. Performance of ribotricer is compared with RpBp, ribORF, and RiboWave when the false positive rate is controlled to be 0.1. The number of truly translating transcripts are shown for both human {A) and mouse {B) datasets. The profiles of expressed CCDS transcripts in Ribo-seq data were treated as true positive and the corresponding RNA-seq profile as true negative. 60 SRP098789 SRP062407 � RpBp O RiboWave SRP063852 � RibORF D Ribotricer Figure 3.23 Performance of different methods on transcript level measured using Fl score. Performance of ribotricer is compared with RpBp, ribORF, and RiboWave in terms of Fl score when the default threshold score is used for each tool. The profiles of expressed CCDS transcripts in Ribo-seq data were treated as true positive and the corresponding RNA-seq profile as true negative. AUROC, Fl scores, and p-values for AUROC difference were calculated using the pROC [135] package in R. For calculating p-values, we used the bootstrap method and set alternative=' greater' . 3.4 Ribotricer's phase score remains stable on truncated ORFs In order to test the ability of ribotricer to correctly predict the translation status of an ORF whose length has been shortened due to truncation we performed a simulation where for all candidate ORFs which have atleast 50% of non-empty codons, i.e. codons with non-zero reads, we truncated it from 3' end such that the truncated length was 10-100% of the original length. For each such truncated ORF, we calculated ribotricer's phase score and compared it with the corresponding RiboCode generated p-value. It is worth mentioning, that among the tools of 61 L() ....... 0 0 (f) I 0 - ....... + C 0 ro Q) ~ L() 0 0 0 0 IIIII I I I I I I I 0 0 20 40 60 80 100 number of codons used Figure 3.24 Effect of number of codons on ribotricer's phase score in human dataset. Mean absolute difference and standard deviation between original phase score using all codons and the one with down-sampled number of codons. The plot was generated on human dataset (SRA accession: SRP063852) using 5K genes with at least 50% valid codons, the down-sampling is repeated 100 times for each gene. Similar trend is observed for other human datasets. 62 0 N 0 0 (f) I - 11 + 0 C ....... ro 0 Q) ~ 0 III!! I I I I 0 0 0 20 40 60 80 100 number of codons used Figure 3.25 Effect of number of codons on ribotricer's phase score in mouse dataset. Mean absolute difference and standard deviation between original phase score using all codons and the one with down-sampled number of codons. The plot was generated on mouse dataset (SRA accession: SRP003554) using 5K genes with at least 50% valid codons, the down-sampling is repeated 100 times for each gene. Similar trend is observed for other mouse datasets. 63 A SRP063852 ~ .,.... 2 cq ~ 0 Q) (0 > 0 :;:::: " ui 0 s::t: 0. 0 1.00 Q) :::, "! 0.96 i= 0 0.96 ~ 0.90 0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 False positive rate ~ .,.... cq 0 C <'! 0 " ui 0 · c:s s::t: Q) I... 0 Ribotricer Q.. N RpBp 0 RibORF 0 RiboWave 0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 Recall SRP098789 1.00 0.91 0.96 0.83 0.2 0.4 0.6 0.8 1.0 False positive rate Ribotricer RpBp RibORF RiboWave 0.2 0.4 0.6 0.8 1.0 Recall Figure 3.26 B 600 E 500 :::, z 400 (/) o. 300 ·c (.) ~ 200 ~ 1- 100 0 2500 E 2000 :::, ~ 1500 0. ·5 1000 (/) C ~ 500 l- o SRP063852 SRP098789 Comparison of performance on detecting translating transcripts. The performance of ribotricer is compared with that of RibORF, RiboWave, and Rp-Bp. (A) The ROC and precision recall curves summarizing performance of ribotricer, RibORF, RiboWave and Rp-Bp on one human and one mouse dataset. (B) The number of translating transcripts recovered when controlling the false positive rate to be the same. capable of performing exon level classification , we were able to benchmark ribotricer against only RiboCode and ORFscore as Ribo Taper requires barn files of both RNA-seq and Ribo-seq samples. Ribotricer's score for the truncated ORF is negligibly different from the original ORF with a maximum difference of ±0.05 (Figure 3.27 and 3.28) as demonstrated using a human (SRA accession: SRP063852) and a mouse dataset (SRA accession : SRP003554). On the other hand, the RiboCode generated p-values show a clear dependence on the ORF length with the deviation from original score being as high as ± 100. It is worth mentioning that the differences between 64 truncated and original profile for RiboCode are calculated on a log 10 scale as it outputs p-values, while for both ribotricer and ORFscore, the differences are calculated on the same scale as the scores. 3.4.1 Ribotricer can detect ORFs as short as 20 codons In order to determine the minimum length of ORF that can be detected by ribotricer we performed a simulation using the Ribo-seq profiles of genes with total codons > 100 and with at least 50% non-empty codons. We then randomly sampled 10-100 codons, without maintaining their order explicitly, and generated a "downsampled " profile. The mean absolute difference between the original phase score calculated using the full length profile versus the "downsampled " profile with 20 or more codons is smaller than 0.05 and does not change after increasing the number of codons (Figures 3.24 and 3.25). 3.4.2 Learning species-specific cutoffs Ribo-seq 's protocol was initially developed to profile the translational landscape in yeast [24], but it has been widely used to profile the translational status of ORFs in multiple species [136, 137]. We benchmarked ribotricer first using human and mouse datasets where we have access to CCDS annotated regions as a high confidence ground truth for known protein coding status (Figures 3.13-3.23). In order to further benchmark ribotricer against other methods, we used additional public Ribo-seq datasets from Arabidopsis, C. elegans, Drosophila, rat, yeast, and zebrafish ( Table 3.4) . Unlike human and mouse, CCDS annotations are not available for these species. Hence, for such species, we considered the Ribo-seq profile of annotated CDS regions as 65 ribotricer RiboCode ORFscore 0 8 0.1 0 0 -20 06 l/1 l/1 l/1 -.!... 0.0 -.!... -40 t4 + + C C C (0 (0 -60 (0 aJ -0.1 Q) ~2 ~ ~ -80 -0.2 0 -100 ##..-;fl. ~ ~ -;fl,'#.'#.#-.'#. #-. ';f.. #-.'#.-;ft ~ ~ -;fl.#..# ~ ';fl#-.'#.'#. -;fl, ~ ~ -;fl.# 000 0 0 00000 00000 0 0 000 0 00000 0 0 00 ...... N f"') s:t U') <.O ,..._ 0) O'> 0 ...... N f"') s:t U') <.O ,..._ 0) O'> 0 ...... N f"') s:t U') <.O ,..._ 0) O'> 0 ...... ...... ...... Percentage retained codons Percentage retained codons Percentage retained codons Figure 3.27 Effect of truncating an ORF on ribotricer's phase score, RiboCode's p-values and ORFscore in human dataset. Mean difference and standard deviation between original phase score using full length ORF and the ones after truncating it from the 3' end. The plot was generated on human dataset {SRA accession: SRP063852) using 5K genes with at least 50% valid codons and truncating it to have indicated percentage (X-axis) of codons. The differences between truncated and original profile for RiboCode are calculated on a log 10 scale as it outputs p-values, while for both ribotricer and ORFscore, the differences are calculated on the same scale as the scores. ribotricer RiboCode ORFscore 0 0.1 6 0 0 -20 0 l/1 0.0 l/1 l/1 t t -40 -.!... 4 + C C -60 C :lJ -0.1 (0 (0 Q) ~2 ~ ~ -80 -0.2 -100 0 ~~'#. ';!2, 0 ';!2, 0 '#. '#. ~ '#. ~ '#. ~ '#. ~ '#. ';!2, 0 ';!2, 0 '#.~~ ';!2, 0 '#. ~ ~ ~ '#. ';!2, 0 ';!2, 0 '#.'#. 000 0 0 00000 00000 0 0 000 0 00000 0 0 00 ...... N f"') s:t U') <.O ,..._ 0) O'> 0 ...... N f"') s:t U') <.O ,..._ 0) O'> 0 ...... N f"') s:t U') <.O ,..._ 0) O'> 0 ...... ...... ...... Percentage retained codons Percentage retained codons Percentage retained codons Figure 3.28 Effect of truncating an ORF on ribotricer's phase score, RiboCode's p-values and ORFscore in mouse dataset. Mean difference and standard deviation between original phase score using full length ORF and the ones after truncating it from the 3' end. The plot was generated on mouse dataset {SRA accession: SRP003554) using 5K genes with at least 50% valid codons and truncating it to have indicated percentage {X-axis) of codons. The differences between truncated and original profile for RiboCode are calculated on a log 10 scale as it outputs p-values, while for both ribotricer and ORFscore, the differences are calculated on the same scale as the scores. 66 the true positive and the corresponding RNA-seq profile as the true negative. In order to establish if we needed to re-adjust our phase score cutoff for each species separately, we summarized the phase scores for both Ribo-seq and RNA-seq samples from multiple public datasets (Figure 3.33) . We observed that phase scores of both RNA-seq and Ribo-seq samples vary across species (Figures 3.30, 3.31 , and 3.32) with higher variation arising from the Ribo-seq samples. The variation in phase scores for RNA-seq samples in the same species is limited, though it also exhibits a species related trend (Figure 3.32). Ribo-seq samples on the other hand exhibit higher intra-species and across-species heterogeneity. Hence, in order to capture this species-specific differences in RNA-seq and Ribo-seq scores, we learned cutoffs for each species separately (Table 3.5 and 3.6; Figure 3.34 ). It is worth noting that, human and mouse samples that we previously used for our benchmark exhibit similar variation in RNA-seq and Ribo-seq phase scores besides having higher Ribo-seq phase scores as compared to all other species. On the other hand , the difference between Ribo- and RNA-seq phase scores appears to be particularly low in Drosophila datasets (Figure 3.32). 3.4.3 Learning dataset-specific cutoffs In studies where both Ribo-seq and RNA-seq experiment are available, it is possible to fine tune the phase-score cutoff to be dataset-specific. The Ribo-seq and RNA-seq samples within the same species can show variation in terms of their phase score (Figure 3.32) and hence, it is possible that learning dataset-specific cutoffs leads to an overall better performance (Figures 3.38- 3.42). To learn the dataset-specific cutoffs, we calculated the median difference between phase scores of Ribo-seq and RNA-seq profiles for each dataset over only protein-coding regions. Using a 67 ~6---------------------------, IB phase score: 0.92 I... ~ 4 Q) E "O ~ 2 ·~ a JI lliil~ ~,~~~~~i~-~i~~~~i~ ~ ~ z -20 0 20 40 60 80 100 120 140 160 180 ~ "O IB 2.0 I... C IB 1.5 E "O 1.0 Q) N ro 0 .5 E I... Distance from start codon (nt) phase score: 0.078 ~ 0.0 --.----r----r----..-----..------.-----r-----r----r---~~ -20 0 20 40 60 80 100 120 140 160 180 Distance from start codon (nt) Figure 3.29 Example of phase scores for an active and a non-active ORF. Phase score generated by ribotricer for two different profiles. 68 0.8- .... 0.7- A ~ 0.6- A A 0 u A V) QJ A V) 0 5- A <U • .c Q. • C -~ 0.4- A "'O A QJ A E 0.3- # A A • A • • A ! , • ' ~ .. ' • • 0.2- •• • • • • • • 0.1- • ' I I I I I I I I V') .µ V') (0 C QJ .µ ..c V') V') C (0 V') (0 V') Q. (0 (0 ..c E ::J a::: i.+= 0 QJ 0) Q. 0 (0 "O >- QJ 0 ::J ~ I- V') QJ V') I ..0 ..0 -I- 0 QJ (0 QJ u I- N I- .::,,t. 0 <l: (0 OJ I • RNA-seq A Ribo-seq] Figure 3.30 Summarized median phase score for RNA-seq and Ribo-seq for all datasets. For each dataset, the median phase score was calculated for all the candidate ORFs for both Ribo-seq and the corresponding RNA-seq sample. 69 ~ 0 • • ~08 a:, • a: 0.8 0.8 Vl Vl • • • • Q) ~ • • ~ • 806 806 • • • ~06 I Vl Vl • • 6 • • Q) Q) • • Vl • Vl • a:, • • Jg 0.4 Jgo.4 • • • • • ~04 • • • • Q. Q. i • c C • , • Q) O" • O" u Q) , •• • • Q) C I • • • • ,i 0 2 • ' 6 0 2 ~02 • • • z .0 :::: • • • er: a: i3 ~ 0.0 -~ 0.0 0.0 • '6 -0 • Q) Q) :;;: :;;: Vl t, Vl ..!!1 C: Qi ~ .c Vl t, Vl ..!!1 C: Qi ~ .c Vl t, Vl ..!!1 C: Qi ~ .c "iii C: "' Vl "' Vl "iii C: "' Vl "' Vl "iii C: "' Vl "' Vl a. "' "' :c E ::, 0:: "' a. "' "' :c E ::, 0:: "' a. "' "' :c E ::, 0:: "' 0 Qi Cl a. 0 ~ 0 Qi Cl a. 0 ~ 0 Qi Cl a. 0 ~ "O >- Qi 0 ::, :;;: "O >- Qi 0 ::, :;;: "O >- Qi 0 ::, :;;: Vl ai Vl I .0 Vl ai Vl I .0 Vl ai Vl I .0 :0 -a; e Qi :0 -a; e Qi :0 -a; e Qi ~ u 0 N ~ u 0 N ~ u 0 N <( .:,(. <( .:,(. <( .:,(. "' "' "' al al al • Arabidopsis • C.elegans • Human • Rat • Baker's Yeast • Drosophila Mouse • Zebrafish Figure 3.31 Median phase score for RNA-seq and Ribo-seq and their differences across multiple species. For each dataset, the median phase score was calculated for all the candidate ORFs for both Ribo-seq and the corresponding RNA-seq sample. Same as Figure 3.30 except that the RNA- and Ribo-seq samples have been separated into individual panels. 1.0 1.0 ro C ... 0.8 0.8 6 .a t ·c: QJ 0.6 QJ 0.6 I l ... u I 0 C 1 u QJ l Vl ... QJ 0.4 QJ 0.4 ' Vl I:!:: ro -0 .c QJ 1 0. 0.2 ... 0.2 0 u Vl QJ 0.0 Vl 0.0 ro .c 0. -0.2 -0.2 Vl ..., Vl ~ C Q) ..., .J:: Vl ..., Vl ~ C Q) ..., .J:: ·v=; Vl C l1J Vl l1J Vl · v=; Vl C l1J Vl l1J Vl Q. l1J l1J .J:: E ::, er: ti::: Q. l1J l1J .J:: E ::, er: ti::: 0 Q) CJ\ Q. 0 !:' 0 Q) CJ\ Q. 0 !:' -0 >- Q) 0 ::, :;;: -0 >- Q) 0 ::, :;;: Vl ai Vl I .0 Vl ai Vl I .0 :0 1... e Q) :0 1... e Q) !:' Q) u N !:' Q) u N <( _,, 0 <( _,, 0 l1J l1J a:, a:, I- Ribo-seq - RNA-seq I Figure 3.32 Distribution of median phase scores for RNA-seq and Ribo-seq samples and their differences across multiple species. For each species, medians were calculated on the collection of merged datasets for that species. 70 -..J f,-l fJ) :::0 ("') =- "0 -· Q c: ro C o- en 3 D. -· 9 fJ) § rn ~ ~ "0 . ::::!, .cg§'T1g' "09:c.Q!:!'. ""' :::i 0 ("') s_oq 3 m :::i o::...,on O ~ro~c:::::.-, (0 0 2 ~ C. -· ~-.)> . Q,):::i o Q,) Ji :::o er e: ... :::i - -· < _, :::i \ \,I 0- fJ) -· c.o.coroc. :i' !"I' "0 I !"I' C: oq~a~-g.~ :::0 (0 ::::!'? .Q Q,) :::0 zc.mn~ z l> n :: n fJ) l> JiCC.>CnJi ro en fJ) en o ro .Q""'!"!' en.C:::!1 "0 ro en ~ fJ) Q,) oq ""'oq'"0<:::lC: 0 -· - c; ""' -oi:;?"2!roC.ro -· :::i \\,I (0 ""' cir fJ) c. fJ) ro :::0 w !"!'::C.l<na-:w enrofJ)~~9w C.lenq::::nV'l !"I' C: \\,I C: (0 C'O!"!'(O!"!'Qj°.Q c. en ""' !"I' fJ) Q,) Q,) ~ m ~ Q,) fJ) (ti oq !"I' 3 !"I' c. C.> roc.o...., "0 ""' !"I' - C: Q,) -· ""' (0 (0 ·~ < Q,) '" fJ) ,,, (0 fJ) - • :::i !"I' • !"I' = "0 (02'T1""'("')::::; oq (0 0 C: Q,) Q,) Q,) ""'C'O:::ilJ'J !:!'. "0 Q,) "0 C. (0 <o-o-c.·V'l (0 fJ) - . -· 0 fJ) Q,) ("') !"I' -· !"I' 0 -· !"I' !:!'. (0 ""' < =- < (0 (0 (0 (0 0 fJ) Q,) ""' Q,) :::lV'l§~("') c.~c.!'>a !"I' ("') !"I' fJ) =- -· =- 'T1 fJ) (0 (0 (0 0 fJ) ""' I 2J. er 0 V> <D .0 I ::0 z )> V> <D .0 0 Q) rt Q) Vl <D rt 0 0 0 l'J phase score 0 0 :i:s en 0 I-' co 0 SRP018118 (Arabidopsis) r-:;.;;;;::=••••;;;;:;===7 SRP029587 (Arabidopsis) SRP059391 (Arabidopsis) SRP087624 (Arabidopsis) SRP108862 (Arabidopsis) SRP000637 (Baker's Yeast) SRP028552 (Baker's Yeast) SRP028614 (Baker's Yeast) SRP033499 (Baker's Yeast) SRP075766 (Baker's Yeast) SRP010374 (C.elegans) SRP014427 (C.elegans) 5RP026198 (C.elegans) 5RP056647 (C.elegans) SRP028243 (Drosophila) SRP045475 (Drosophila) SRP076919 (Drosophila) SRP108999 (Drosophila) 5RP010679 (Human) 5RP029589 (Human) 5RP063852 (Human) 5RP098789 (Human) SRP102021 (Human) SRP003554 (Mouse) SRP062407 (Mouse) SRP078005 (Mouse) SRP091889 (Mouse) SRP115915 (Mouse) ERP007231 (Rat) SRP045777 (Rat) SRP056012 (Rat) SRP010040 (Zebrafish) SRP023492 (Zebrafish) 5RP034750 (Zebrafish) ~ I sampling strategy where a one-third fraction of protein-coding profiles were used to determine the median difference between Ribo-seq and RNA-seq profiles with replacement ( nbootstraps = 10000) [102], the dataset-specific cutoff was assigned to be the median of these differences. It is worth mentioning that this approach is only viable for studies where both Ribo-seq and RNA-seq samples are available. The dataset-specific cutoffs result in ribotricer achieving higher Fl scores in some but not all datasets (Tables 3.9-3.11; Figures 3.38-3.42). In all our datasets, a median difference of 0.25 or more between Ribo-seq and RNA-seq protein-coding profiles results in an Fl score greater than 0.73 (Figure 3.41 ). Given a set of Ribo-seq and RNA-seq mapped files (BAM), the dataset-specific cutoffs can be determined by using ribotricer learn-cutoff . Table 3.6 Species specific recommended phase score cutoffs for ribotricer. A " # " indicates the cutoff for the species is taken to be the median phase score difference between CDS annotated Ribo-seq and RNA-seq profiles since they only had one dataset each. Species Cutoff Arabidopsis 0.330 Baker's Yeast 0.318 C. elegans 0.249 Drosophila 0.181 Human 0.440 Mouse 0.418 Rat 0.453 Zebrafish 0.249 C. albicans# 0.228 5. pombe# 0.409 Chimpanzee# 0.334 Macaque# 0.321 3.4.4 Running ribotricer on a new species We provide a list of recommended phase score cutoffs (Table 3.6) for most species where there are at least three or more public Ribo-seq datasets (Table 3.4). The cutoffs for each species were 72 Table 3.7 Species wise mean, median and standard deviation of difference of Ribo-seq and RNA-seq phase scores. SD = Standard Deviation. A " # " indicates that the median phase score difference for these species is also considered as cutoff for ribotricer, since they only had one dataset each. species number of samples mean dif- ference phase score Arabidopsis 5 0.308 Baker's Yeast 5 0.309 C.elegans 4 0.232 Drosophila 4 0.048 Human 5 0.385 Mouse 5 0.468 Rat 3 0.260 Zebrafish 3 0.325 C. albicans# 1 0.228 5. pombe# 1 0.380 Chimpanzee# 1 0.328 Macaque# 1 0.285 Maximizinq Fl score ~ 0.60 !!l ~ 0 .60 ...., ' a o.45 11 ?f! 0.45 ~ ~ $ ~ 8 0 .30 0.30 Vl ~ ~ 0.15 0 .15 co ~0 .00 0 .00 l/l ...., l/l (0 C QJ ...., ..c l/l 'vi l/l C (0 l/l (0 l/l 'vi (0 c:: 0. QJ (0 ..c E ~ ;,;:: 0. 0 Ol 0. 0 (0 0 'O >- ~ 0 ~ :a '- 'O l/l l/l I .0 .0 - ..... QJ 0 QJ .0 (0 QJ u '- N (0 '- :,(, 0 '- <I'. (0 <I'. co Figure 3.34 median difference phase score 0.365 0.287 0.273 0.054 0.428 0.528 0.303 0.388 0.225 0.409 0.334 0.321 SD 0.252 0.225 0.235 0.221 0.240 0.230 0.253 0.309 0.151 0.176 0.233 0.218 Median Ribo-RNA difference ~ ~ ~ 4- ¢ ~ ...., l/l ~ C QJ ...., l/l C (0 l/l (0 (0 c:: QJ (0 ..c E ~ Ol 0. 0 >- ~ 0 ~ :a l/l l/l I - ..... QJ 0 QJ u '- :,(, 0 (0 co ~ ..c l/l ;,;:: (0 '- .0 QJ N Distribution of median difference between Ribo-seq and RNA-seq sample as determined using only two datasets per species. For each species all possible combinations of two datasets were chosen and median difference between phase scores of Ribo-seq and RNA-seq determined. 73 Table 3.8 Best and second to best performing methods at AU ROC metric for each dataset. The p-values were calculated using pROC [135] package using bootstrap method and alternative='greater'. AUROC (B) and AUROC (SB) denotes area under ROC for the best and the second to best methods respectively. A * indicates the dataset was later used to learn the ribotricer cutoffs by maximizing the Fl score. The AU ROC values however do not depend on any cutoff. SRP Species Best (B) Second Best (SB) AUROC (B) AUROC (SB) p-value SRP018118* Arabidopsis RiboCode 0.982 0.923 < 2.2 X 10- 16 SRP029587 Arabidopsis 0.897 0.594 < 2.2 X 10- 16 SRP059391 * Arabidopsis 0.690 0.632 < 2.2 X 10- 16 SRP087624 Arabidopsis 0.697 0.523 < 2.2 X 10- 16 SRP108862 Arabidopsis 0.732 0.607 < 2.2 X 10- 16 SRP000637 Baker's Yeast 0.921 0.837 < 2.2 X 10- 16 SRP028552* Baker's Yeast 0.986 0.951 < 2.2 X 10- 16 SRP028614 Baker's Yeast 0.966 0.846 < 2.2 X 10- 16 SRP033499 Baker's Yeast 0.947 0.783 < 2.2 X 10- 16 SRP075766* Baker's Yeast 0.996 0.962 < 2.2 X 10- 16 SRP010374* C. elegans 0.867 0.776 < 2.2 X 10- 16 SRP014427 C. elegans 0.927 0.920 3.774 X 10- 14 SRP026198* C. elegans ORFscore 0.956 0.908 < 2.2 X 10- 16 SRP056647 C. elegans RiboCode RiboTaper 0.745 0.745 0.247 SRP028243* Drosophila RiboCode 0.725 0.587 < 2.2 X 10- 16 SRP045475 Drosophila ORFscore RiboTaper 0.633 0.522 < 2.2 X 10- 16 SRP076919 Drosophila 0.638 0.465 0.317 SRP108999* Drosophila 0.884 0.727 0.068 SRP010679* Human 0.944 0.849 < 2.2 X 10- 16 SRP029589 Human 0.846 0.701 < 2.2 X 10- 16 SRP063852 Human 0.969 0.930 < 2.2 X 10- 16 SRP098789* Human 0.975 0.908 < 2.2 X 10- 16 SRP102021 Human 0.961 0.927 < 2.2 X 10- 16 SRP003554* Mouse 0.974 0.972 2.045 X 10- 6 SRP062407 Mouse RiboCode ORFscore 0.986 0.981 < 2.2 X 10- 16 SRP078005 Mouse RiboCode 0.989 0.968 < 2.2 X 10- 16 SRP091889 Mouse RiboCode 0.981 0.966 < 2.2 X 10- 16 SRP115915* Mouse RiboCode 0.926 0.923 1.095 X 10- ll ERP007231 * Rat Ribo Taper RiboCode 0.955 0.953 3.321 X 10- 9 SRP045777 Rat RiboCode 0.793 0.746 < 2.2 X 10- 16 SRP056012* Rat ORFscore RiboCode 0.971 0.872 < 2.2 X 10- 16 SRP010040* Zebrafish ORFscore 0.658 0.562 < 2.2 X 10- 16 SRP023492 Zebrafish ORFscore 0.970 0.958 < 2.2 X 10- 16 SRP034750* Zebrafish RiboCode 0.995 0.977 < 2.2 X 10- 16 SRP032814 C. albicans RiboCode 0.953 0.842 < 2.2 X 10- 16 SRP062129 Chimp ORFscore 0.918 0.883 < 2.2 X 10- 16 SRP107240 5. pombe RiboCode 0.972 0.939 < 2.2 X 10- 16 SRP062129 Macaque ORFscore 0.904 0.854 < 2.2 X 10- 16 Table 3.9 Best and second to best performing methods at Fl score metric for each dataset using dataset-specific cutoff. Fl (B) and Fl (SB) denotes the Fl scores for the best and the second to best methods respectively. An asterisk ( *) indicates that the dataset was used to learn the cutoffs by maximizing the Fl score. A # indicates the ribotricer phase score cutoff for the dataset is taken to be the median phase score difference between CDS annotated Ribo-seq and RNA-seq profiles. SRP Species Best (B) Second Best (SB) Fl (B) Fl (SB) SRP018118* Arabidopsis RiboCode 0.937 0.848 SRP029587 Arabidopsis RiboCode 0.645 0.176 SRP059391 * Arabidopsis RiboCode 0.562 0.361 SRP087624 Arabidopsis ORFscore 0.675 0.338 SRP108862 Arabidopsis RiboCode 0.628 0.333 SRP000637 Baker's Yeast RiboCode 0.680 0.503 SRP028552* Baker's Yeast RiboCode 0.964 0.859 SRP028614 Baker's Yeast RiboCode 0.855 0.738 SRP033499 Baker's Yeast RiboCode RiboTaper 0.747 0.705 SRP075766* Baker's Yeast RiboTaper 0.951 0.877 SRP010374 C. elegans RiboCode 0.799 0.517 SRP014427 C. elegans RiboCode 0.826 0.776 SRP026198* C. elegans RiboCode 0.828 0.636 SRP056647 C. elegans RiboCode 0.690 0.634 SRP028243* Drosophila RiboCode 0.693 0.562 SRP045475 Drosophila RiboCode 0.561 0.391 SRP076919 Drosophila RiboCode 0.667 0.125 SRP108999* Drosophila RiboCode 0.769 0.400 SRP010679* Human RiboCode 0.877 0.773 SRP029589 Human RiboCode 0.651 0.599 SRP063852 Human RiboCode 0.919 0.854 SRP098789* Human RiboCode 0.932 0.824 SRP102021 Human RiboCode 0.890 0.835 SRP003554* Mouse RiboTaper 0.901 0.899 SRP062407 Mouse RiboTaper 0.930 0.910 SRP078005 Mouse RiboCode 0.951 0.901 SRP091889 Mouse RiboCode 0.938 0.900 SRP115915* Mouse RiboCode 0.853 0.842 ERP007231 * Rat RiboTaper 0.879 0.874 SRP045777 Rat RiboCode 0.618 0.511 SRP056012* Rat RiboCode 0.787 0.786 SRP010040* Zebrafish RiboCode 0.670 0.377 SRP023492 Zebrafish RiboCode 0.838 0.826 SRP034750* Zebrafish RiboCode 0.920 0.894 SRP032814# C.albicans RiboCode 0.883 0.752 SRP062129# Chimp RiboCode 0.865 0.436 SRP062129# Macaque RiboCode 0.842 0.635 SRP107240# 5. pombe RiboCode 0.913 0.869 Table 3.10 Best and second to best performing methods at Fl score metric for each dataset-specific cutoff. Fl (B) and Fl (SB) denotes the Fl scores for the best and the second to best methods respectively. The cutoff was learned independently for each dataset as the median difference between Ribo-seq and RNA-seq phase scores over protein coding ORFs. SRP Species Best (B) Second Best (SB) Fl (B) Fl (SB) SRP018118 Arabidopsis RiboCode 0.920 0.848 SRP029587 Arabidopsis RiboCode 0.846 0.176 SRP059391 Arabidopsis RiboCode 0.678 0.361 SRP087624 Arabidopsis ORFscore 0.671 0.338 SRP108862 Arabidopsis RiboCode 0.695 0.333 SRP000637 Baker's Yeast RiboCode 0.850 0.680 SRP028552 Baker's Yeast RiboCode 0.928 0.859 SRP028614 Baker's Yeast RiboCode 0.923 0.738 SRP033499 Baker's Yeast RiboCode 0.904 0.747 SRP075766 Baker's Yeast RiboTaper 0.935 0.877 SRP010374 C.elegans RiboCode 0.798 0.517 SRP014427 C.elegans RiboCode 0.868 0.776 SRP026198 C.elegans RiboCode 0.846 0.636 SRP056647 C.elegans RiboCode 0.716 0.634 SRP028243 Drosophila RiboCode 0.679 0.562 SRP045475 Drosophila RiboCode 0.667 0.391 SRP076919 Drosophila RiboCode 0.667 0.125 SRP108999 Drosophila RiboCode 0.818 0.400 SRP010679 Human RiboCode 0.878 0.773 SRP029589 Human RiboCode 0.765 0.599 SRP063852 Human RiboCode 0.919 0.854 SRP098789 Human RiboCode 0.922 0.824 SRP102021 Human RiboCode 0.900 0.835 SRP003554 Mouse RiboTaper 0.919 0.901 SRP062407 Mouse RiboTaper 0.936 0.930 SRP078005 Mouse RiboCode 0.944 0.901 SRP091889 Mouse RiboCode 0.924 0.900 SRP115915 Mouse RiboCode 0.863 0.842 ERP007231 Rat Ribo Taper RiboCode 0.874 0.867 SRP045777 Rat RiboCode 0.722 0.618 SRP056012 Rat RiboCode 0.786 0.738 SRP010040 Zebrafish RiboCode 0.668 0.377 SRP023492 Zebrafish RiboCode 0.918 0.838 SRP034750 Zebrafish RiboCode 0.937 0.920 76 Table 3.11 Ribotricer's performance at Fl score when considering species-specific or dataset-specific cutoff Fl (SS) and Fl (DS) denotes the Fl scores for ribotricer when using species-specif and dataset-specific cutoffs respectively. Ribo-RNA indicates the median difference between phase score of protein coding ORFs in Ribo- and RNA-seq samples. 'sampled' indicates the median was calculated using 30% of protein coding ORFs per dataset with resampling (nbootstraps = 10000) while 'all' indicates the median was calculated using the complete list of protein coding ORFs. SRP species Fl (SS) Fl (DS) Ribo-RNA (sampled) Ribo-RNA ( all) SRP018118 Arabidopsis 0.937 0.920 0.455 0.447 SRP029587 Arabidopsis 0.645 0.846 0.206 0.191 SRP059391 Arabidopsis 0.562 0.678 0.109 0.104 SRP087624 Arabidopsis 0.675 0.671 0.233 0.145 SRP108862 Arabidopsis 0.628 0.695 0.181 0.154 SRP000637 Baker's Yeast 0.503 0.850 0.186 0.179 SRP028552 Baker's Yeast 0.964 0.928 0.383 0.382 SRP028614 Baker's Yeast 0.855 0.923 0.267 0.263 SRP033499 Baker's Yeast 0.573 0.904 0.204 0.194 SRP075766 Baker's Yeast 0.951 0.935 0.694 0.671 SRP010374 C.elegans 0.799 0.798 0.224 0.222 SRP014427 C.elegans 0.826 0.868 0.343 0.334 SRP026198 C.elegans 0.828 0.846 0.322 0.316 SRP056647 C.elegans 0.690 0.716 0.141 0.135 SRP028243 Drosophila 0.693 0.679 0.109 0.098 SRP045475 Drosophila 0.561 0.667 -0.019 -0.020 SRP076919 Drosophila 0.667 0.667 -0.025 -0.034 SRP108999 Drosophila 0.769 0.818 0.363 0.360 SRP010679 Human 0.878 0.878 0.421 0.404 SRP029589 Human 0.651 0.765 0.234 0.223 SRP063852 Human 0.919 0.919 0.522 0.498 SRP098789 Human 0.932 0.922 0.526 0.514 SRP102021 Human 0.891 0.900 0.427 0.417 SRP003554 Mouse 0.900 0.919 0.542 0.526 SRP062407 Mouse 0.910 0.936 0.588 0.568 SRP078005 Mouse 0.951 0.944 0.603 0.591 SRP091889 Mouse 0.939 0.924 0.509 0.497 SRP115915 Mouse 0.854 0.863 0.372 0.361 ERP007231 Rat 0.879 0.863 0.403 0.388 SRP045777 Rat 0.511 0.722 0.176 0.173 SRP056012 Rat 0.787 0.738 0.264 0.247 SRP010040 Zebrafish 0.670 0.668 0.136 0.108 SRP023942 Zebrafish 0.826 0.918 0.512 0.502 SRP034750 Zebrafish 0.894 0.937 0.660 0.649 77 0 Tool AUROC iii' ;o ;o 0 0 0 0 0 0 .... §: c' 0 " :.. "' 00 0 t""I' 0 c' ~ ~ 0 Q,) "" .;< n SRP018118 (Arabidopsis) :, ,;· "O 0 0 O" ~ ~ C. ;;; C. ro C: SRP029587 (Arabidopsis) ::c t""I' SRP018118 (Arabidopsis) n o· SRP059391 (Arabidopsis) Qj'" 2 :, SRP029587 (Arabidopsis) en )> SRP059391 (Arabidopsis) SRP087624 (Arabidopsis) en I 0 :::::;.; en ...., SRP108862 (Arabidopsis) r=;· C'D Q,) SRP087624 (Arabidopsis) Q,) .c "" SRP108862 (Arabidopsis) SRP000637 (Baker's Yeast) !:!'. "C C'D 0 Q,) Q,) SRP000637 (Baker's Yeast) SRP028552 (Baker's Yeast) :, -· C: SRP028614 (Baker's Yeast) "" :, SRP028552 (Baker's Yeast) C: :, C. !:!?. C'D SRP028614 (Baker's Yeast) SRP033499 (Baker's Yeast) :, Q,) "" 0 I SRP075766 (Baker's Yeast) oq SRP033499 (Baker's Yeast) - ~ C. ::c ~ Q,) 0 SRP075766 (Baker's Yeast) 0 SRP010374 (C.elegans) t""I' ~ O" Q,) n SRP010374 (C.elegans) SRP014427 (C.elegans) a~- 0 SRP014427 (C.elegans) 0 " SRP026198 (C.elegans) -w ::::. !""' )> 0 n Q.l C SRP026198 (C.elegans) I SRP056647 (C.elegans) ~ "" ::c 'Tl • C'D O -· SRP056647 (C.elegans) ;o SRP028243 (Drosophila) ::c Q,) n oq 0 o' SRP028243 (Drosophila) e: 0 SRP045475 (Drosophila) n -· C: C: 0 O" - "" SRP045475 (Drosophila) C. 0 SRP076919 (Drosophila) 0 :, Q,) C'D ro ~ ..... c. n w 0 SRP076919 (Drosophila) OJ Q,) C'D "" • I!> I l£ SRP108999 (Drosophila) "C "" 0 W ~ SRP108999 (Drosophila) 0 C'D ::C en u, ~ "' SRP010679 (Human) 0 ;o ~ 0 en SRP010679 (Human) o' ~ SRP029589 (Human) ::c n 3 SRP029589 (Human) "' "O SRP063852 (Human) -· C: ~ O" ~ !:!'. SRP063852 (Human) 0 0 Q,) ~ SRP098789 (Human) n en "C SRP098789 (Human) I 0 n en SRP102021 (Human) C. SRP102021 (Human) :,_ Q,) en C" SRP003554 (Mouse) C'D --c SRP003554 (Mouse) 0 ~ n C'D io ,;· Q,) =- n 0 ~ SRP062407 (Mouse) :, Q,) ii5' SRP062407 (Mouse) C. t""I' SRP078005 (Mouse) C'D !'> SRP078005 (Mouse) 0 C. SRP091889 (Mouse) ::c 'Tl SRP091889 (Mouse) ~ 0 SRP115915 (Mouse) 'Tl "" "" SRP115915 (Mouse) (/) ERP007231 (Rat) n C'D C'D ERP007231 (Rat) 0 X Q,) SRP045777 (Rat) "" 0 n SRP045777 (Rat) C'D :, :r SRP056012 (Rat) ~ SRP056012 (Rat) en SRP010040 (Zebrafish) SRP010040 (Zebrafish) < O" ~ 0 SRP023492 (Zebrafish) I en SRP034750 (Zebrafish) -..J C'D co .c -..J ~ 9. ~ t-t' ""I :E a,: 0 C: C. QJ t-t' QJ fJ) (0 t-t' fJ) ~ ""I (0 C: fJ) (0 C. !:!'. 0 :::::s 0 ...., 'Tl I-" fJ) n 0 ""I (0 fJ) QJ n t-t' ""I O 0 fJ) - fJ) (0 fJ) ~ "C :::::s (0 t-t' !:?. ::r (0 (0 fJ) n c: C: !!?. t-t' :::::s S,OQ ..., fJ) fJ) "C n C'O o n ""I - · (0 (0 fJ) 0 I ...., fJ) ""I "C - · (0 o- n 0 :::::;.: t-t' -· ""I n ;:;· n (0 C: ""I t-t' ~~ ""I • t-t' 'Tl ::r 0 QJ ""I t-t' (0 fJ) QJ "C n (0 ::r !:?. fJ) (0 "C fJ) (0 n ii>' fJ) 'Tl oci' C: ""I (0 w w O'I SRP018118 (Arabidopsis) SRP029587 (Arabidopsis) SRP059391 (Arabidopsis) SRP087624 (Arabidopsis) SRP108862 (Arabidopsis) SRP000637 (Baker's Yeast) SRP028552 (Baker's Yeast) SRP028614 (Baker's Yeast) SRP033499 (Baker's Yeast) SRP075766 (Baker's Yeast) SRP010374 (C.elegans) SRP014427 (C.elegans) SRP026198 (C.elegans) SRP05664 7 (C.elegans) SRP028243 (Drosophila) SRP045475 (Drosophila) ~ SRP076919 (Drosophila) or ii SRP108999 (Drosophila) SRP010679 (Human) SRP029589 (Human) SRP063852 (Human) SRP098789 (Human) SRP102021 (Human) SRP003554 (Mouse) SRP062407 (Mouse) SRP078005 (Mouse) SRP091889 (Mouse) SRP115915 (Mouse) SRP010040 (Zebrafish) SRP023492 (Zebrafish) "' w c' i 'E' w "' 0 iv "' 0 w w u ~ 0 0 .... 0) 0 ~ 0 w ... 0 0 0) DJl1 - . • I;: 0 0 ~ a - . 0) w 0 "' "' 0 - ~ 0 :j 0 0, 0) 0 "' V, 0 ~ 0 ~ 0 0, "' 0 "' 0 0 ::: 0 :5: 0 0 :") 0 :_., ~ 0 :_., ... 0 "' "' 0 "' 0 0 "' w 0 :j 0 ~ 0 "' 0 0 :_., 0) - 0 0, ... 0 ~ 0 :_., w 0 ~ 0 0 !" I " c' i I I I 0 0 SRP018118 (Arabidopsis) SRP029587 (Arabidopsis) ~ SRP059391 (Arabidopsis) SRP087624 (Arabidopsis) SRP108862 (Arabidopsis) ~ SRP000637 (Baker's Yeast) SRP028552 (Baker's Yeast) SRP028614 (Baker's Yeast) SRP033499 (Baker's Yeast) SRP075766 (Baker's Yeast) SRP010374 (C.elegans) = SRP014427 (C.elegans) SRP026198 (C.elegans) SRP056647 (C.elegans) SRP028243 (Drosophila) SRP045475 (Drosophila) i;:::- Fl score 0 0 :,. . SRP076919 (Drosophila)--1-•------ SRP108999 (Drosophila) SRP010679 (Human) SRP029589 (Human) SRP063852 (Human) SRP098789 (Human) SRP102021 (Human) SRP003554 (Mouse) SRP062407 (Mouse) SRP078005 (Mouse) SRP091889 (Mouse) SRP115915 (Mouse) ERP007231 (Rat) SRP045777 (Rat) SRP056012 (Rat) SRP010040 (Zebrafish) I:: SRP023492 (Zebrafish) SRP034750 (Zebrafish) !" ~ 0 1.0 •• •• ... , ., .. • ~ . 0.9 •• •• • • • • , . . , - ~ • • • •••• •• • • • u 0.8 • • • • 0 •• c::: •" • :::, 0.7 • • <( • • • • 0.6 0.5 • • • • • • •• • • • . , 0.9 ... . ., ••• • • •• • • • • f<e •• • Qi 0.8 • • • • • L • 0 u V) .-t 0. 7 ., • • LL • •• • • •• • • • • • 0.6 • • • . " 0.5 • • • • oi.noi.noi.noi.n stN0CO<OstN0 ~~NNf'"!f'"!~~ N(V)~~~~f'-:~ 00000000 00000000 Median RNA-seq phase score (RNA) Median Ribo-seq phase score (RIBO) • Arabidopsis • Baker's Yeast • C.elegans • Drosophila • Human Mouse Figure 3.37 • • • • • • •• . , ~ · · ' • •• \ • • • • " . • • • • • • • ' . , • • fe •• • . , ... • • I • • " •• • • ...... 0 ...... 0 0 0 I N(V)stlJ)<OI'- . . . . 0 0 0 0 0 0 Difference (RIBO-RNA) e Rat • Zebrafish Performance of ribotricer at AUROC and Fl scores metrics across species at different median phase scores of RNA-seq and Ribo-seq samples using species-specific cutoff. For each dataset, median phase score was calculated for both RNA-seq and Ribo-seq samples for the same list of candidate ORFs. 80 Tool ::!. ~ ~ c.. [ ~ ~ ~ C ~- i a C, -• , , m g- fJ) ~ SRP018118 (Arabidopsis) t"t' Cl) "'I :e t""'t" er SRP029587 (Arabidopsis) m t"t' ~ SRP0S9391 (Arabidopsis) :::r -· ~ ('t) 0 SRP087624 (Arabidopsis) ::i :::0 g O SRP108862 (Arabidopsis) C"' t"t' _, "O O O SRP000637 (Baker's Yeast) "'I I ::::i 'Tl 0 ~ i-,,. SRP028552 (Baker's Yeast) (ti .C ~ ~ SRP028614 (Baker's Yeast) ::i a, fJ) 0 n ~ -,: SRP033499 (Baker's Yeast) 0 0. Cl) Cl) """" 0J VI SRP075766 (Baker's Yeast) .... :::0 "'I a, ::i -, ::i n SRP010374 (C.elegans) ()q '- Cl) "'I )> 0. 0 SRP014427 (C.elegans) t"t' I fJ) @ ~ ~ fJl SRP026198 (C.elegans) ::i .c fJ) fJ) Q. "O 'Tl SRP0S6647 (C.elegans) n "O Cl) Cl) - · -,: -,: t""'t" n Q'ct SRP028243 (Drosophila) -o' 0 Cl) -• C f"'t' - ""'1 3 m --.it SRP045475 (Drosophila) fJ) - Cl) --, m C (..,J a;' SRP076919 (Drosophila) ,,.., ::i fJ) @" g- C"' ::i :i' (..,J !£ SRP108999 (Drosophila) ~ '<oqoq co Ul ,,, SRP010679 (Human) C, v, t"t' 0. ~ a, 3 :::::i C, SRP029589 (Human) Cl) t"t' 11 "2. 3 ~ SRP063852 (Human) f--' ::i Cl) ~ SRP098789 (Human) O oq 0. I 0 ~• 00 SRP102021 (Human) 0 0 - "O 0 ::i :;i Cl) SRP003554 (Mouse) t"t' Cl) "O !:? • _ 3 , ~ ::r ~ SRP062407 (Mouse) :::r a, n Cl) fJ) SRP078005 (Mouse) fJl "'I Cl) n Q. fJ) C SRP091889 (Mouse) o n t"t' -, 0 ~ SRP115915 (Mouse) ;. CD ' ERP007231 (Rat) Cl) Q. 'TIO SRP045777 (Rat) t"t' -· a ~ ; SRP0S6012 (Rat) C, CD ~ SRP010040 (Zebrafish) ::i r') ::r SRP023492 (Zebrafish) Cl) 00 SRP034750 (Zebrafish) f--' 0 :,. "' "' I " c' f I 0 !1 " g i I a;' ~ ~ SRP018118 (Arabidopsis) SRP029587 (Arabidopsis) SRP059391 (Arabidopsis) SRP087624 (Arabidopsis) SRP108862 (Arabidopsis) SRP000637 (Baker's Yeast) SRP028552 (Baker's Yeast) SRP028614 (Baker's Yeast SRP033499 (Baker's Yeast SRP075766 (Baker's Yeast SRP010374 (C.elegans SRP014427 (C.elegans SRP026198 (C.elegans SRP056647 (C.elegans SRP028243 (Drosophila SRP045475 (Drosophila SRP076919 (Drosophila SRP108999 (Drosophila SRP010679 (Human SRP029589 (Human SRP063852 (Human SRP098789 (Human SRP102021 (Human SRP003554 (Mouse SRP062407 (Mouse SRP078005 (Mouse SRP091889 (Mouse SRP115915 (Mouse ERP007231 (Ra SRP045777 (Ra SRP056012 (Ra SRP010040 (Zebrafish SRP023492 (Zebrafish SRP034750 (Zebrafish 0 ~ Fl score 0 :,. ~ ~ 1.0 0,8 ~ 0.6 u Vl ~ 0.4 0,2 0.0 - Vl Vl ·.,; ·.,; Q. Q. 0 0 "O "O ,5 ,5 ~ ~ :'!, :'!, ~ r-- a) r< U'l ~ a, N 0 0 Q. Q. a: a: <Jl <Jl n Vl Vl Vl i ·.,; ·.,; ·.,; ro Q. Q. Q. 0 0 0 "' "O "O "O >- Vl ,5 ,5 ,5 ~ ~ ~ ~ :'!, :'!, :'!, ro r< " N ~ a, N \0 "' \0 a) r-- a, r-- a) "' U'l a) 0 \0 0 0 r< 0 0 Q. Q. Q. 0 a: a: a: Q. <Jl <Jl <Jl a: <Jl n n t; t; i i Vl Vl Vl ro ro ro ro C C C ro ro ro "' "' "' "' 0, 0, 0, >- >- >- >- "' "' "' Vl Vl Vl Vl v v v ~ ~ ~ ~ ~ ~ ~ -"' -"' ro ro ro ro " r-- a) ~ ~ ~ ~ r-- N ::: "' " N ;"; a, \0 0 " \0 U'l a, \0 .., .., N U'l \0 " r-- 0 0 0 a) a) "' U'l Q. Q. Q. N N "' r-- a: a: a: 0 0 0 0 <Jl <Jl <Jl Q. Q. Q. Q. a: a: a: a: <Jl <Jl <Jl <Jl 1 - r, Vl ~ ~ ~ ~ '2 '2 '2 '2 '2 ~ ~ ~ ~ ~ ro ro ro ~ -5i -5i C :c :c :c :c ro ro ro ro ro ro E E E E E :, :, :, :, :, ~ ~ ~ e: e: e: 0, Q. Q. Q. Q. 0 0 0 0 0 ~ ~ ~ "' ~ ~ ~ ~ :, :, :, :, :, ~ ~ ~ ~ ~ r< r-- N v !, !, !, !, !, "' r-- .., .0 .0 .0 e e e e N r-- 0 "' "' "' ~ " r-- U'l a, U'l t, t, t, e e e e a, a, N a, .., U'l 0 0 a) .., r-- U'l \0 r-- a) U'l a) N U'l st 0 a) a, 0 " U'l r-- "' U'l a, \0 U'l a) r-- 0 "' N a) .., U'l 0 0 0 0 N 0 " ::: 0 a, "' a) N 0 \0 r-- a, ~ Q. Q. Q. " a, U'l \0 " r-- a, .., N \0 a, 0 0 0 0 0 a: a: a: 0 " r-- \0 N " a, a, 0 0 0 0 .., Q. Q. Q. Q. Q. w <Jl <Jl 0 "' st U'l a) U'l \0 a) Q. Q. Q. Q. Q. a: a: a: a: a: .., N "' 0 N " r-- ;: a: a: a: a: a: <Jl <Jl <Jl <Jl <Jl 0 0 0 Q. 0 0 0 Q. <Jl <Jl <Jl <Jl <Jl Q. Q. Q. a: Q. Q. Q. a: a: a: <Jl a: a: a: a: <Jl <Jl <Jl <Jl <Jl <Jl <Jl Dataset dataset-specific - species-specific I Figure 3.39 Difference in performance of ribotricer using species-specific or dataset-specific cutoffs. Species-specific cutoffs were learned by maximizing the Fl scores for two datasets per species while dataset-specific cutoffs were learned per dataset using the median difference of phase 5 ecies-s ecific RiboTaper ORFscore 0 .8 RiboCode ribotricer 0.0 0.2 0.4 0.6 0.8 1.0 Fl score score of Ribo-seq and RNA-seq protein coding profiles. Dataset-s ecific - ribotricer 0 .0 0.2 0.4 0 .6 0.8 1.0 Fl score Figure 3.40 0.0 0.2 0 .4 0.6 0.8 1.0 Fl score Summarized performance of ribotricer using species-specific and dataset-specific strategies. Species-specific cutoffs were learned by maximizing the Fl scores for two datasets per species while dataset-specific cutoffs were learned per dataset using the median difference of phase score of Ribo-seq and RNA-seq protein coding profiles. Species-specific and dataset-specific cutoffs only apply to ribotricer. 82 Q) ,_ 0.95 0 .90 0.85 0.80 8 0.75 V) ~ 0 .70 0.65 0 .60 0.55 0 .50 s + + i i i i i i i i , . I i i ·~ I i ecific cutoff • • . . ~ • . , • •• • + e!.- • • + t- • ·-·-·- ·-·-·- ·- ·- ·-·-· j ·- ·-·-·- ·- ·- ·-·-·- ·-·-·- ·-·-·- ·- ·- ·- + • • I i · 1 = -~ i i • • i f I , : + 0.95 0 .90 0.85 0.80 0.75 0 .70 0.65 0 .60 0.55 0.50 Dataset-s ecific cutoff i i i I • • I- + i i ·~ i i i •j · + i • I • • • t- . • • • • • t- ..... · -· -·-· -· -· -· -·· .... ,• ·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·- . : •• •i 0 .0 0 .1 0.2 0 .3 0.4 0.5 0 .6 0.7 0 .0 0.1 0.2 0 .3 0.4 0.5 0 .6 0 .7 Median phase score difference (Ribo-RNA) Median phase score difference (Ribo-RNA) • Arabidopsis • C.elegans • Human e Rat • Baker's Yeast • Drosophila • Mouse • Zebrafish Figure 3.41 Distribution of ribotricer's Fl scores with respect to median phase score difference of Ribo-seq and RNA-seq, using species-specific and dataset-specific cuoffs. Species-specific cutoffs were learned by maximizing the Fl scores for two datasets per species while dataset-specific cutoffs were learned per dataset using the median difference of phase score of Ribo-seq and RNA-seq protein coding profiles. The dashed red lines indicate a median difference of 0.25 between Ribo-seq and RNA-seq phase scores results in a Fl score of 0. 73 and above. 83 0.3 .--.. ~ 0.2 I V"l 0 ~ 0.1 0.0 Q) u C Q) 0.3 I,... Q) tt:: "O 0.2 .--.. <( 0.1 z er:: I _g 0.0 er:: .__. -0.1 C (Cl ~ -0.2 Q) ~ I -0.3 tt:: .8 -0.4 ::::; u V"l V"l i i • i • i i i i i i i i i • i i i • i i , . i • i i • • : . . . ~ ! ."Y ·+ ·-·-·-· -·• ..... � . · -·-·-•-· -· -· -·-·-·-·-· -· -· -· -·-·- : • ., ,t • i • • i i I I I I I 0 N ~ ~ 00 0 0 0 0 0 Median phase score RNA-seq i • • i a • i I I i • • • • • ,. . 0.3 .--.. ~ 0.2 I V"l 0 ~ 0.1 0 .0 0.3- 0.2- .--.. <( z er:: I 0.1- i i i i i i i i i i i i i i i i i i i • • • • • • i • • • i , ! • • ! ~ ·+ ·-·-·-·-·-·-·-·-·-· -·• • . ·-• ·-·-·-· : . .. i • • i • • • 0 N ~ ~ 00 0 0 0 0 0 Median phase score Ribo-seq i i i i i i •• • • • i ••• : , ... • i • • • • ·- • -·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·- I 0 .0 ! • • 0 . 0 - ·-·-: ·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·-·- ·-·-·-·-·-·-·-·- •' 1'\ ~ ! J , . I i • i i i e! :• i 0 .0 0.2 0.4 0.6 Fl (DS-SS) • Arabidopsis • Baker's Yeast 0.8 ii: .__. -0.1- c (Cl ~ -0.2- Q.) ~ I -0.3- tt:: .8 -0.4- ::::; u V"l V"l i i • • ! • I i i i i i i i i i i i i • • •• • • •• • • 0 N ~ ~ 00 0 0 0 0 0 Median phase score Ribo-seq • C.elegans • Human e Rat • Drosophila • Mouse • Zebrafish Figure 3.42 Effect of Ribo-seq and RNA-seq phase scores on species-specific and dataset-specific based Fl performance. Fl {DS-SS) indicates difference in Fl scores using species-specific (SS) or dataset-specific (DS) cutoff. Each single data point represents one dataset. Median phase scores were calculated using all the candidate ORF profiler of either RNA-seq or Ribo-seq sample. 84 learned empirically by using Ribo-seq and RNA-seq samples from two datasets and maximizing the Fl score by treating the Ribo-seq profiles of CCDS/ CDS regions as ground true positive and the corresponding RNA-seq profiles as true negatives (Figure 3.34; Table 3.5). However, this approach is only best suited for species where there are multiple datasets available. For a new species where there are only few or none datasets available and hence the cutoff cannot be learned empirically, we recommend using the median score difference between the profiles of annotated CDS regions of a Ribo-seq and the corresponding RNA-seq sample. This strategy is also used by RibORF [34] which tunes the parameters of its model by selecting one-third of the CDS profiles as true positives. We followed this strategy of using the median phase score difference as the phase score cutoff for each of the four species: C. albicans, chimpanzee, macaque and 5. pombe. Except for 5. pombe, all other species have only one public dataset available to the best of our knowledge (Table Sl) . We first generated candidate ORF list for each species using ribotricer over transcripts with annotated CDS regions. Phase scores were then calculated for each RNA-seq and Ribo-seq sample over these CDS annotated candidate ORFs (Figure 3.45). The median differences in Ribo-seq and RNA-seq phase scores for C. albicans, chimpanzee, macaque and 5. pombe is summarized at the end of Table 3.7. We used these differences as species-specific cutoffs for benchmarking ribotricer against other methods. Ribotricer results in the best AUROC for all the four species with the difference between ribotricer and the second best method statistically significant in all the cases (Figure 3.43; Table 3.8). It is worth mentioning that the AUROC metric is not dependent on the choice of the learned cutoff. Furthermore, ribotricer is also the best method using the Fl score metric (Figure 3.44; Table 3.9). 85 We recommend using the species-specific cutoffs for all the species as listed in Table 3.6. For any new species, we recommend using median phase score differences on ribotricer generated candidate ORFs over CDS annotated transcripts between Ribo-seq and RNA-seq samples (Figure 3.45). This can be determined by ribotricer itself, using the learn-cutoff subcommand . 3.5 Using ribotricer In order to use ribotricer, the following three files are required : GTF: genome annotation file in GTF format (ENSEMBL/ Gencode/ others) • FASTA: reference genome file in FASTA format • BAM: alignment file in BAM format Henceforth, we use the boldface acronyms above to refer to these files as such . 3.5.1 Preparing candidate ORFs list ribotricer prepares a candidate list of ORFs given a GTF and FASTA file. For any species, given a reference and a fixed version of GTF, this step only needs to be done once. Ribotricer by default searches for ORFs defined by an 'AUG' start and an in-frame stop codon ('UAG', 'UAA', and 'UGA') and are a minimum of 60 nucleotides long. It is possible to expand the definition of ORF by supplying a list of all start codons using the --start_codons parameter. It is also possible to change the minimum length of an ORF by using the --min_orf_length option . If multiple potential in-frame start codons exist upstream of a stop codon , we always choose AUG if it exists, otherwise, we take the most upstream one as the start codon . 86 1.0 0.8 g 0.6 0:: :J <( 0.4 0.2 0.0 vi' C (0 u :.0 (0 u s:i- .-I co N (Y) 0 a.. a:: l/l [ - ORFscore RiboTaper ribotricer Q. a; E ::, C, ..c (0 u u (0 0) ~ N .-I 0) N N \0 .-I 0 N a.. \0 a:: 0 l/l a.. a:: l/l Dataset - RiboCode - RiboTaper 0.66 0.72 0.78 0.84 0 .90 0 .96 I ..:. '..=:::::::::::::1 ___ _ 'v, a: a, ..., C E :::, 1/l C1l CJ' C1l E Q) u C1l >- :0 ~ u C1l C C1l O'l ~ 0 ~ N 'iii ...... O'l 1/l s;- N N U::: ...... 0.0 ...... 00 0 N 0 N 0.. 0.0 s;- ('(\ a: 0 N 0 (jJ 0.. r-- 0.. a: 0 a: (jJ ...... (jJ 0.. a: (jJ Dataset Figure 3.43 .., l/l (0 a; > c 0 l/l l/l LL 0 s:i N r-- 0 .-I a.. a:: l/l - ribotricer J Distribution of area under ROC in the independent datasets. For each Ribo-seq ari&' RNA-seq pair in a dataset, area under ROC was calculated for exon level classification using Ribotricer, Ribotaper, RiboCode and ORFScore. 0.8 OJ 0.6 I... 0 u Vl u: 0.4 0.2 0.0 Vl c.. O.I C E :::J Ct) cr- u .c Ct) :.c u u Ct) fl'! 0) ~ u N ,-i 0) tj' N N ,-i \.0 ,-i ro 0 N N a.. \.0 (Y) c::: 0 0 l/) a.. a.. c::: c::: l/) l/) Dataset [ - ORFscore - RiboCode - RiboTaper 0 0 I- 0.15 0.30 0.45 0.60 0.75 0.90 I I ORFscore- 0.22 0.31 0.17 RiboTaper 0.14 0.35 ribotricer I I v, ci w z C E :::; Ul ro O' ro QJ u ..c ro >--- :a ~ u ro C ro CJ'> 2 0 Li N 'vi rl CJ'> Ul <:I' N N u::: rl l.O rl co 0 N 0 N c.. l.O <:I' ('() 0:: 0 N 0 V) c.. r-- c.. 0:: 0 0:: V) rl V) c.. 0:: V) Dataset .:;- Vl Ct) ~ C ,Q Vl - ~ LL 0 tj' N r--- 0 ,-i a.. c::: l/) - ribotricer J Figure 3.44 Distribution of Fl scores in the independent datasets. For each Ribo-seq and 88 RNA-seq pair in a dataset, Fl score was calculated for exon level classification using Ribotricer, Ribotaper, RiboCode and ORFScore. 1.0 0.8 QJ 0 0.6 u V) QJ V) 10.4 a. 0.2 0.0 Vl C C1l .!::! .0 C1l u ""' ,-t a:, N (Y) 0 a.. a::: Vl c.. E .s= u 0) N ,-t N I.O 0 a.. a::: Vl Dataset :;:,- Vl C1l (1/ >- C 0 Vl Vl LL 0 ""' N r-- 0 ,-t a.. a::: Vl (1/ ::::, O" C1l u C1l :a 0) N ,-t N I.O 0 a.. a::: Vl I- Ribo-seq - RNA-seq J Figure 3.45 Distribution of ribotricer's phase scores for RNA-seq and Ribo-seq samples in the independent datasets. For each dataset phase scores were calculated for all candidate ORFs. For human and mouse, Ribo-seq CCDS profiles were treated as true positive and the corresponding RNA-seq profile was treated as true negative. For all other species Ribo-seq profile of annotated CDS regions were treated as true positive and the corresponding RNA-seq profile treated as true negative. 89 ribotricer prepare-orfs --gtf {GTF} \ --fasta {FASTA} \ --prefix {RIB0TRICER_INDEX} The command above will create a list of candidate ORFs at the RIB0TRICER INDEX location . For this study, we used a total of ten codons with a maximum of one nucleotide difference from "ATG" as potential start codons including ATA, ATC, ATT, AAG, ACG , AGG, ATG, CTG, GTG , TTG. Note that we use 'T' as a nucleotide here instead of 'U' as the reference FASTA almost always contains DNA sequences. 3.5.2 Detecting actively translating ORFs using ribotricer Ribotricer's ORF list as created above can then be used along with the BAM to define the translation status of these ORFs: ribotricer detect-orfs --barn {BAM}\ --ribotricer index {RIB0TRICER_INDEX}_candidate_0RFs.tsv \ --prefix {0UT_PREFIX} For each ORF in the candidate ORFs list, ribotricer calculates the phase score on the read profiles after performing read length appropriate offset shifts. These offsets are determined by maximizing the cross-correlation of these profiles with the profile for the most abundant read length . Additionally, ribotricer automatically infers the sequencing protocol (forward/ reverse) and only uses unique mapping reads that conform to the strand orientation in the GTF. For example, a read uniquely mapping to a gene defined on the negative strand for a forward stranded protocol , will be discarded . 90 In order to assign 'non-translating' or 'translating' status, ribotricer, by default, uses a cutoff threshold of 0.428. ORFs with phase score above 0.428 are marked as translating as long as they have at least five codons with non-zero read count. Ribotricer does not take coverage into account for predicting an ORF to be translating or not-translating. Apart from these two criteria , there is no other requirement for an ORF to be active. Though, a region with higher overall coverage as defined by number of reads per unit codon might be a more confident 'hit' for active translation, our method is designed to find evidence of active translation based on the qualitative pattern of "high-low-low" and hence our rankings are purely based on phase scores. The default cutoff (0.428) was learned using public human and mouse Ribo-seq datasets, where the gap between Ribo- and RNA-seq phase scores is the highest amongst other species (Table 3.7) and hence, it is a conservative cutoff for detecting active translation . We provide a list of species-specific recommended cutoffs (Table 3.6), optimized for Fl score based performance. The main output of the above command is a tab separated file consisting for each candidate ORF, its translation status, the corresponding transcript and gene and the ORF type. Different ORF types defined by ribotricer are described below: annotated: CDS annotated in the provided GTF file super_uORF: upstream ORF of the annotated CDS, not overlapping with any CDS of the same gene • super_dORF: downstream ORF of the annotated CDS, not overlapping with any CDS of the same gene uORF: upstream ORF of the annotated CDS, not overlapping with the main CDS 91 dORF: downstream ORF of the annotated CDS, not overlapping with the main CDS overlap_uORF: upstream ORF of the annotated CDS, overlapping with the main CDS overlap_dORF: downstream ORF of the annotated CDS, overlapping with the main CDS novel: ORF in non-coding genes or in non-coding transcripts of coding genes 3.5.3 Filtering actively translating ORFs using multiple criteria In order to assign 'non-translating' or 'translating' status, ribotricer by default uses a cutoff threshold of '0.428'. ORFs with phase score above '0.428' are marked as translating as long as they have at least five codons with non-zero read count. By default, ribotricer does not take coverage or count information explicitly into account for predicting an ORF to be translating or not-translating. However, this behavior can be changed by following filters: --min_ valid_ codons ( default=5): Minimum number of codons with non-zero reads for determining active translation --min_valid_codons_ratio (default=0): Minimum ratio of codons with non-zero reads to total codons for determining active translation --min_reads_per_codon (default=0) : Minimum number of reads per codon for deter mining active translation --min_read_density (default=0.0): Minimum read density (total reads/ length) over an ORF total codons for determining active translation For each of the above filters, an ORF failing any of the filters is marked as 'non-translating'. 92 For example, to ensure that each ORF has at least 3/ 4 of its codons non-empty, we can specify --min valid codons ratio to be 0.75: ribotricer detect-orfs --barn {BAM}\ --ribotricer index {RIB0TRICER_INDEX}_candidate_0RFs.tsv \ --prefix {0UTPUT_PREFIX} --min valid codons ratio 0.75 - - - It might also often be desired to have some minimum density of reads over an ORF. The read density here is defined as the ratio of total number of reads over an ORF to its length. For example to ensure that each 'translating' ORF has at least a read density of 10, we will specify --min_read_densi ty to be 10. ribotricer detect-orfs --barn {BAM}\ --ribotricer index {RIB0TRICER_INDEX}_candidate_0RFs.tsv \ --prefix {0UTPUT_PREFIX} --min_read_density 10.0 The above filters can be combined to give ORFs that have high read density as well as have have reads present over most of the codons in the profile. Note that increasing the value of any of the four filters will usually result in a smaller list of ORFs marked 'translating'. 3.5.4 Downstream ranking and filtering It is also possible to filter actively-translating ORFs after running ribotricer. Ribotricer pro duces a tab separated file with columns that include read-density, number and ratio of valid 93 codons to total codons in the ORF besides the phase score. As such , filtering can be performed downstream using awk or any other programming language. Here we provide an example of filtering and sorting the output of a ribotricer run using Python using the pandas library: Listing 3.1 Filtering ORFs using python. The function returns a filtered list of translating ORFs which have a read density of at least 2.5; a total read count of atleast 50; and the ratio of non-empty codons to total codons atleast 0. 75. import pandas as pd def filtered_df (df) : df filtered df filtered d f. Io c [ d f. status= ' trans I at in g ' ] df. loc [( df [ ' read_density ' ] > =2.5) & \ ( df [ ' read_count ' ] > =50) & \ (df[ ' valid_codons_ratio ' ] > =0 .75)] df sorted df_filtered . sort_values(by=[' phase_score' , 'read_density '] , ascending=[False , False]) return df sorted # read ribotricer output ribotricer_output_df =pd . read_csv( '/ path / to / translating_ORFs . tsv' , sep=' \ t # filter and sort ribotricer output 94 ribotricer filtered df filtered_df ( ribotricer_output_df) 3.5.5 Learning cutoff empirically from data Ribotricer can learn cutoff empirically from the data. Given at least one Ribo-seq and one RNA-seq BAM file , ribotricer learns the cutoff by running one iteration of the algorithm on the provided files with a pre-specified cutoff(--phase_score_cutoff, default: 0.428) and then uses the generated output to find the median difference between Ribo-seq and RNA-seq phase scores of only candidate ORFs with transcript_ type annotated as protein_ coding: ribotricer learn-cutoff --ribo barns ribo_barn1.barn , ribo_barn2.barn \ --rna barns rna_1.barn \ --prefix ribo_rna_prefix \ --ribotricer_index {RIBOTRICER_ANNOTATION} By default, ribotricer searches for ORFs that are at least 60 nucleotides or 20 codons long to build the candidate list but this minimum length can be set to a user-defined value. We arrived at the default value of 20 codons by performing a simulation using the Ribo-seq profiles of genes with total codons > 100 and with at least 50% non-empty codons. In the simulation , we randomly sampled 10 - 100 codons and generated a "downsampled " profile. The mean absolute difference between the original phase score calculated using the full length profile versus the "downsampled " profile with 20 or more codons is smaller than 0.05 and does not change after increasing the number of codons (Figs. 3.24 and 3.25). Ribotricer enables discovery of both short and long ORFs that will deepen our understanding of translational regulation across various biological contexts. We envision ribotricer's phase score 95 to become a commonly used quality control metric for assessing the quality of Ribo-seq datasets, especially for novel datasets in species where no prior datasets exist. 96 Chapter 4 Translational landscape of morphological changes in Candida albicans 4.1 Introduction Environmental human pathogens have evolved the ability to survive in diverse environmental conditions. Fungal diseases have remained an important public health problem since the early 1980s [138]. Candidemia and other forms of invasive candidasis remain one of the most prevalent mycoses [139] and Candida albicans ( C. albicans) remains the most common species respon sible for it [140]. C. albicansis a fungal pathogen that inhabits the mucosa I surfaces of most healthy individuals as human commensals. Though mostly asymptomatic, C. albicansis an op portunistic pathogen known to causes disease in individuals with a debilitated immune system or a disruption in the host's microbiome [141]. It normally manifests as a commensal in the human gastrointestinal tract, oral and vaginal cavities. lmmunocompromised individuals such as organ transplant patients, patients undergoing cytotoxic or immunosupressive therapies, and HIV/ AIDS patients are particularly at higher risk of infection [142]. Candida infections are associated with 97 considerable mortality and morbidity [139, 143, 144]. C. albicans's virulence arises because of its ability to 1) form true hyphae 2) resist phagocytosis 3) adhere to the host surfaces, and 4) secrete proteinase [141]. C. albicans are aerobic yeasts yet can grow anaerobically [145] and have a diploid genome. It adheres to the host cells epithelial surface via the molecular adhesins located in its cellular enve lope. They are capable of undergoing a reversible morphological transition from single budding yeasts to a continuously branching filaments via a transitory psuedohyphal state. This ability of C. albicans to undergo reversible filamentous transition is one of the key reasons responsible for its virulence [146]. The yeast to filament transition in C. albicans occurs in a variety of conditions, but most importantly in the presence of serum at body temperature (37°(). Understanding the transcriptional and translational mechanisms behind this morphological transition can provide a better understanding of its pathogenesis. It can also help in the development of new anti-fungal drugs that can prevent infections in immunocompromised patients. Here, we provide a compre hensive study of the changes in transcriptional and translational landscape using deep sequencing Ribo-seq and RNA-seq in yeast and filamentous like growth conditions. 4.2 Results 4.2.1 Deep RNA sequencing and Ribosome profiling of C. albicans under yeast and filamentous like growth conditions In order to study the transcriptional and translational landscape changes involved in yeast to filamentous transition, we performed deep sequencing of mRNA and ribosome bound mRNA 98 fragments in C. albicans cells growing at 30° C and 37°( in the presence of serum (Figure 4.2) . While cells at 30°( undergo yeast like growth, the cells at 37°( in the presence of serum show filamentous growth. From all experiments, we obtained a total of 246.2 million reads with a median of 6.25 million reads for Ribo-seq and 20.5 million reads for RNA-seq experiments (Figure 4.1 ). 4.2.2 Changes in the transcriptional landscape In order to study changes that occur at the transcriptional level during yeast to filamen tous transition , we performed differential expression analysis between the two conditions. Using DESeq2 [147], we identified a total of 498 genes that are up-regulated (log 2 fold change > 1; adjusted p < 0.05) and 385 that are down-regulated (log 2 fold change < -1 ; adjusted p < 0.05). A gene ontology analysis reveals, that while the up-regulated genes show enrichment in filamen tous growth , nuclear periphery, and cellular bud neck besides other terms (Figure 4.4A) . On the other hand, the down-regulated genes are involved in translation , ribosome and rRNA synthesis (Figure 4.4B). Table 4.1 highlights some of the genes that are transcriptionally induced dur ing the morphological transition , have virulence-related properties and show significantly reduced translational efficiency (TE) . 4.2.3 Translational landscape changes In order to identify changes at the translational landscape, we sequenced the ribosome pro tected fragments (RPFs) in the two conditions following the protocol of lngolia et al. [148]. RPFs are enriched in the range of 28-32 nucleotides as expected (Figure 4.2A) . There is high correlation 99 Ribo 30°C #1 5.7 M 1.4 M--------,r-_._-~--, 1.2 M 1.1 M 900.0 K 750.0 K 600.0 K 450.0 K 300.0 K 150.0 K 0.0 ...................... 900.0 K 750.0 K 600.0 K 450.0 K 300.0 K 150.0 K 3.6M 3.2M 2.8M 2.4M 2.0M 1.6 M 1.2 M 800.0 K 400.0 K 0.0 ........ TTTT.,.,..,.,""' 2.7M 2.4M 2.1 M 1.8 M 1.5M 1.2 M 900.0 K 600.0 K 300.0 K o.o.J,..~.,...,....,.., 1.1 M 900.0 K 750.0 K 600.0 K 450.0 K 300.0 K 150.0 K 9.0M 8.0M 7.0M 6.0M 5.0M 4.0M 3.0M 2.0M 1.0 M Ribo 30°C #2 6.8 M Ribo 37°C + Serum #2 63.2 M 0.0 ............... - .. 1.4M 1.2 M 1.1 M 900.0 K 750.0 K 600.0 K 450.0 K 300.0 K 150.0 K 0.0 .J,..TTTi ......... 1.4M 1.2 M 1.1 M 900.0 K 750.0 K 600.0 K 450.0 K 300.0 K 150.0 K RNA 37°C + Serum #2 20.6 M o.o.J,.........,,...., 1.8 M 1.6 M 1.4 M 1.2 M 1.0 M 800.0 K 600.0 K 400.0 K 200.0 K 900.0K 750.0 K 600.0 K 450.0 K 300.0 K 150.0 K 1.4M 1.2 M 1.1 M 900.0K 750.0 K 600.0 K 450.0 K 300.0 K 150.0 K 1.8 M 1.6 M 1.4M 1.2 M 1.0 M 800.0 K 600.0 K 400.0 K 200.0 K Ribo 30°C #3 9.3 M 15 20 25 30 35 40 45 50 15 20 25 30 35 40 45 50 15 20 25 30 35 40 45 50 Figure 4.1 Distribution of total reads segregated by read length in Ribo-seq and RNA-seq experiments. Total read counts are displayed in parenthesis. 100 A B 1.2M 1.1M 900K 750K 600K 450K 300K 150K I .. ___ _ _ _ _ _ _ - Mapped read length (nt) C D 12 ----- R 2 :0.93 10 10 -5 -10 4 6 8 10 12 log2(1+TPM) RNA-seq RPFs - Frame1 - Frame2 - Frame3 ;~~~~ 0 M~~~~~N~~g~~~~~~~~~g~$m~~~~~~g~i~ Distance from start codon (nt) • • • • -20 -10 PC1: 57% variance Figure 4.2 • 10 • • 30' C 20 • 37°C+Serum • RNA-seq A Ribo-seq Ribosome profiling in Candida albicans. A) Read length distribution in a 30 ° C sample. B) Principle component analysis of Ribo-seq and RNA-seq samples in 30° C and 37° + Serum samples. C) Correlation between read counts measured as mapped transcripts per million (TPM) between a RNA-seq and the corresponding Ribo-seq sample. D) Metagene plot of a ribosome protected fragments (RPF) and RNA-seq fragments. 101 A 1 5 N ----- R 2 = 0.95 M M - - --- R 2 = 0.92 .. .. .. p p p 0 0 10 0 M 1 0 M M 10 0 0 0 .c .c .c ii: ii: ii: ~ ~ ~ D. D. D. I- 5 I- 5 I- 5 + + + ~ ~ ~ C) C) C) .2 .2 .2 0 0 0 0 1 0 5 5 log, (1+TPM) Ribo 30°C #1 log, (1+TPM) Ribo 30°C #1 log, (1+TPM) Ribo 30°C #2 N 1 5 ----- R 2 = 0.93 M 15 ----- R 2 = 0.91 M 1 5 R 2 = 0.89 .. .. .. ~ ~ ~ Q) Q) Q) "' "' "' + + + p p p 10 .... 1 0 .... 10 .... M M M 0 0 0 .c .c .c ii: ii: ii: ~ 5 ~ 5 ~ 5 D. D. D. I- I- I- + + + ~ ~ ~ C) C) C) .2 0 .2 0 .2 0 0 10 0 10 0 1 0 log 2 (1+TPM) Ribo 37°C + Ser #1 log, (1+TPM) Ribo 37°C + Ser #1 log 2 (1+TPM) Ribo 37C 0 + Ser #2 B ----- R 2 = 0.97 ----- R 2 = 0.96 ----- R 2 = 0.99 N M M .. .. .. p 10 p 10 p 10 0 0 0 M M M <( <( <( z z z a:: a:: a:: ~ ~ ~ D. 5 D. 5 D. 5 I- I- I- + + + ~ ~ ~ C) C) C) .2 .2 .2 0 0 0 I I I I I 0 5 10 0 5 10 log, (1+TPM) RNA 30°C #1 log, (1+TPM) RNA 30°C #1 log, (1+TPM) RNA 30°C #2 N M M .. ----- R 2 = 0.98 .. ----- R 2 = 0.94 .. ~ ai ai Q) "' "' "' + + + u 1 0 p 10 u 1 0 1-- .... 1-- M M M <( <( <( z z z a:: a:: a:: ~ 5 ~ 5 ~ 5 D. D. D. I- I- I- + + + ~ ~ ~ C) C) C) .2 .2 .2 0 0 0 I I I I I I I I I I I 0 5 10 0 10 log 2 (1+TPM) RNA 37°C + Ser #1 log 2 (1+TPM) RNA 37°C + Ser #1 log 2 (1+TPM) RNA 37°C + Ser#2 Figure 4.3 Correlation between replicates across conditions. Each row indicates pairwise correlation between the three replicates belonging to A) 30°C and B) 37° C + Serl'(W conditions across RNA-seq and Ribo-seq. A nucleus RNA metabolic process ribosome biogenesis mitochondrion nucleolus RNA binding structural molecule activity ribosome translation • • • Reduced DE • • • • • • Count • 10 e 40 e 60 e 120 B Increased DE 0 0 0 0 Count 0 0 15 0 0 40 0 70 0 0 100 0 0 translation regulator activity ~-~-~--~ response to stress filamentous growth interspecies interaction between organisms carbohydrate metabolic process pathogenesis cell wall extracellular region biofilm formation cell wall organization cell adhesion ~~~-~- 0.05 0.1 0.15 0.2 Gene Ratio translation regulator activity 0.25 0.3 Count • 10 e 40 e 60 e 120 Figure 4.4 0.05 biofilm fonnation pathogenesis response to stress 0.1 0.15 0.2 Gene Ratio extracellular region cell wall organization 0 Count 0 15 0 40 0 70 0 100 Differential expression analysis between 37° C + Serum and 30° C samples. Gene ontology of A} down-regulated and B} up-regulated genes. 103 A 3 a. 2 0 di 0 ""j" 1 0 -6 • reduced TE • NS • increased TE -3 ... .. ... . • :-•1 .: • ~SN3 • ~~DB 0 3 log 2 fold change Figure 4.5 6 B - -I O 0 m m m :b :b tr z 0 )> log 2 fold change 8 6 4 2 0 -2 -4 Differential translational efficiency between 37°C + Serum and 30°C samples. A) Volcano plot showing the fold change in translational efficiency (TE) and the associated p-value. Differential TE genes are defined with the criteria log 2 fold change > 1 (red, increased TE) or < -1 (green, reduced TE) and P < 0.05. B) Heatmap showing fold changes at the TE, RNA-seq and Ribo-seq levels for the genes highlighted in A. 104 Gene name Ref. # Description Fold Fold change change (mRNA)(TE) C3_07980C orf19.6192 protein of unknown function ; Piel regulated 7.6 TLOl orf19.7544 C3_05990C orf19.7380 member of TLO gene family protein of unknown function member of TLO gene family candidalysin, cytolytic peptide 15.0 17.8 21.4 388.0 109.1 -22.8 -13.9 TLO34 orf19.2661 -7.3 EC El orf19.3374 -5.5 HWPl SKN7 DDR48 HXKl RBTl SFUl SFL2 SLK19 toxin essential for mucosal infec tion orf19.1321 hyphal wall protein , adhesin , host 362.0 -5.4 transglutaminase substrate mimic orf19.971 Putative response regulator in 3.7 -4.9 phophorelay signal transduction; required for H2O2 resistance orf19.4082 immunogenic stress-associated pro- 22.2 -4.5 tein orf19.2154 GlcNAc kinase; required for hyphal 4.2 -4.3 growth and virulence orf19.1327 cell wall protein similar to Hwpl; 64.4 -2.7 required for virulence orf19.4869 GATA-type transcriptional regula- 2.0 -2.3 tor of iron-responsive genes; pro- motes gastrointestinal commensal- ism orf19.3969 transcriptional regulator of mor- 2.0 -2.3 phogenesis orf19.6763 Alkaline-induce plasma membrane 2.2 -2.1 protein important for cell wall ; re- quired for virulence Table 4.1 Genes transcriptionally induced during the C. albicans morphological transition and involved in pathogenesis and/or virulence-related properties show significantly reduced translational efficiency (TE) 105 (> 0.8, p-value < 0.05) between Ribo-seq and RNA-seq normalized read counts (Figure 4.2B and 4.3) while the metagene plots are periodic indicating the samples are representative of actively translation (Figure 4.2D) . We used ribotricer [149] to filter out actively translating fragments for all Ribo-seq exper iments. Ribotricer exploits the periodicity information in Ribo-seq data to separate out the actively-translating from non-actively translating fragments. In order to identify changes in the translational landscape, we focused on genes showing differential translational efficiency. Trans lational efficiency is a measure of the rate of mRNA translation into proteins and can be approx imated as the ratio of Ribo-seq to the corresponding RNA-seq normalized read counts. We used riborex [150] to identify genes that are differentially translationally efficient during the morpho logical transition . Gene ontology analysis reveals the genes showing higher translational efficiency are involved in pathogenesis and hyphael growth (Figure 4.5) . Some of the key genes showing increased translational efficiency are highlighted in Table 4.2 while Table 4.1 highlights some key genes that are transcriptionally induced but show reduced translational efficiency. 4.2.3.1 Re-annotating the C. albicans transcriptome The transcriptome annotation made available through the Candida Genomes Database (CGD) (http://www . candidagenome . org/) only contains the genomic coordinates of exons and coding domain sequences. The 5' and 3' untranslated regions (UTRs) are currently not annotated in the annotation file (GTF) made available through CGD website. However, Bruno et al. [151] provide coordinates for both 5' UTRs and 3' UTRs for Assembly 21. We expanded the current annotation of exons and coding sequences by also adding UTR coordinates made available by Bruno et al. [151]. Since the annotation were made using 106 A nucleus regulation of biological process cytoplasm organelle organization DNA binding response to stress response to chemical chromosome filamentous growth membrane RNA metabolic process interspecies interaction between organisms • () • • cell cycle pathogenesis hydrolase activity mitochondrion DNA metabolic process protein binding nucleolus transferase activity o 0 0.1 mitochondrion Reduced TE • • 0 • 0 • e Gt • () 0 • 0.2 0.3 Gene Ratio Count O 5 9 10 .. 20 0 40 Count • 5 0 10 20 () 40 B Increased TE cytoplasm membrane transferase activity 0 nucleus 0 transport 0 mitochondrion 0 regulation of biological process 0 organelle organization 0 endomembrane system 0 RNA metabolic process 0 hydrolase activity 0 cellular protein modification process 0 mitochondrial envelope 0 response to chemical O response to stress 0 oxidoreductase activity 0 transporter activity 0 lipid metabolic process 0 plasma membrane 0 generation of precursor metabolites and energy 0 0 0.1 0.2 0.3 Gene Ratio 0 Count 0 10 0 20 0 30 O so O •• Count 0 10 0 20 0 30 O so response to stress hydrolase act1v,ty Q 65 / 0 0 response to chemical a I \ generation of precursor metabolites and energy regulation of biological proce ~ s ~ i cl ondrion 0 oxidoreductase activity nucleus.Q " 'tochondrial envelope organelle organiz~tion \ / cytoplasm RNA metaboHc 2~ase activity / '( /~ ::::sporter activity / ··-··- endomembranesystem .__() ..,,. O cellular protein modification process transpo lipid metab~ic process plasma membrane Figure 4.6 Gene ontology analysis of TE genes A) Genes with reduced TE B) Genes with increased TE 107 Gene name Ref. # Description Fold change (TE) AA Pl orf19.2810 putative amino acid permease; fungal-specific multicopper oxidase 40.5 33.8 32.7 26.2 22.5 13.3 FET3 orf19.4213 SAP98 orf19.852 GPl-anchored aspartic endopeptidase CDC14 orf19.4192 Protein involved in exit from mitosis and morphogenesis pepstatin A-insensitive secreted aspartyl protease putative glucose transporter of the major facilitator su perfamily SAP7 orf19.756 HGT20 orf19.1587 MED8 orf19.4497 ortholog of RNA Polymerase II Mediator complex com- 12.3 GNP3 UECl MAL31 SSUl LIP6 BMT8 ALG6 HGT5 PSA2 MNN24 CHS7 HYRl HMSl TPO5 OPTS CEKl PTC8 RPD3 BEMl RAX2 SMF12 orf19.7565 orf19.4646 orf19.3981 orf19.7313 orf19.4823 orf19.860 orf19.1843 orf19.6005 orf19.4943 orf19.1995 ponent putative high-affinity glutamine permease; fungal- 9.8 specific protein required for oral epithelial cell damage, hyphal 8.7 growth and stress resistance putative high-affinity maltose transporter 8.5 protein similar to S. cerevisiae Ssul sulfite transporter; 5.9 important for filamentous growth secreted lipase 5.8 putative ,8-mannosyltransferase 5.7 putative glucosyltransferase involved in cell wall mannan 4.8 biosynthesis putative glucose transporter of the major facilitator su- 4. 7 perfamily mannose-1-phosphate guanyltransferase 4.6 a -1 ,2-mannosyltransferase; required for normal cell wall 4.3 mannan content orf19.2444 Protein required for chitin synthase Ill activity 4.1 orf19.4975 GPl-anchored hyphal cell wall protein ; important for re- 4.1 sistance to killing by neutrophils, azoles orf19.921 hlh domain Myc-type transcription factor required for 3.6 morphogenesis orf19.151 putative polyamine transporter 3.5 orf19.5770 oligopeptide transporter 3.5 orf19.2886 ERK-family protein kinase; required for yeast-hyphal 3.5 switch , mating efficiency and virulence orf19.4698 predicted type 2C protein phosphatase; required for hy- 3.1 phal growth orf19.2834 histone deacetylase; regulates white-opaque switching 3.1 orf19.4645 protein required for budding, hyphal growth and viru- 2.6 lence orf19.3765 plasma membrane protein involved in establishment of 2.4 bud sites and linear direction of hyphal growth orf19.2270 manganese transporter 2.4 Table 4.2 Selected genes showing significantly increased translational efficiency (TE) during the C. albicans morphological transition. 108 A B [EJ-- S'UTR 4658 De novo transcriptome assembly cos 6210 (String Tie) ! 3'UTR 2812 Rlbo and RNA-seq samples Potential 59 novel I Create consensus GTF I 1000 2000 3000 4000 5000 6000 Count I C D CGDGTF Bruno, eta/, 12.5 0.25 ----- Potenllal novel coordinates --1 Merge UTRs 1-- coordinates - Annotated (CDS) (UTRs) - 10 t 0.2 I = ·io.1s g' 7.5 !I 5 ~ 0.1 ~ i 2.5 -------------...... ______ S'UTR cos 3'UTR Potential 0 2.5 5 7.5 10 12.5 F novel 1092 Distance (nt) E MSTRG.9377.6 F MSTRG.9632.1 Ribo 30°C Ribo 37°C + Serum Ribo 30°C Ribo 37°C + Serum 2 :i. 1 , , • . l, J . , I 2 :1.-.,u~~ • . .. ,~.~~ » tJ i ] di. , "1,ll u : .. , , , , I :J1•4~•1'4,, 1il, I q ,,, , , , , I : lQ,~ . -i;; ,,J i: ~. i ij : i M,•-R ·~ - ~~ii ,,i . ~ I ~ : Uiiiij~iii~ L , ,. , .,I ] •• ~kiil!~ :T. : 11 .. ,, J 200 400 600 800 200 400 600 800 200 400 600 800 1000 200 400 600 800 1000 Location (nt) Location fnt) Location (nt) Location (nt) Figure 4.7 Ribo-seq enables novel ORF discovery. A) Workflow for re-annotating transcriptome of Candida a/bicans. B) Distribution of regions in the final annotation as obtained from workflow in A. Novel indicates exons that were not annotated in the original annotation. C) Distribution of sizes of 5' UTR, 3'UTR, CDS and novel exons. D) Distribution of distances of novel exons (purple, dashed) to known exons and known exons to other known exons (red). E) A novel exon with active transcription and translation in both 30°C and 37°C + Serum samples. F) A novel exon with active transcription with a strong Ribo-seq signal at a stop codon. Assembly 21 while the current Assembly is 22, we lifted over the coordinates to the latest assembly using the liftOver [152] tool. The chain file was obtained from CGD. The lifted over candidates were further filtered to ensure there was no overlap between the existing coding domain sequences. This resulted in a total of 9084 5' UTRs, 5464 3' UTRs for a total of 12389 coding domain sequences for the diploid assembly (Figure 4.7B). 109 4.2.4 Detecting novel exons and potential ORFs We hypothesized that using our deep-sequenced Ribo-seq and RNA-seq samples we could potentially discover "novel" exons and open reading frames given that the transcriptome anno tation of C. albicans is currently incomplete. In order to discover novel exons, we performed a guided denovo assembly using our RNA-seq and Ribo-seq samples across the two conditions (Figure 4.7 A) . In particular, we used String Tie [153] along with the annotation file from CGD (r27 GTF) as the guide annotation (-G ref .GTF) . For each input, StringTie outputs a new GTF which represents a super-set of transcripts and exons annotated in the guide GTF consisting of additional exons that are unannotated in the guide GTF. We then created a consensus catalogue of novel exons by looking for overlapping regions of the novel exons from all the new GTFs. Using our strategy we were able to discover additional 71 exons (Figure 4.7) that were previously unannotated. The novel exons have smaller size distribution as to the previously annotated with the median length of novel exons being 315 nucleotides while that of previously annotated exons around 1187 nucleotides (Figure 4.7C) . It is possible that the novel exons remained unannotated because they are smaller in size and hence harder to detect through any technique that relies on relative abundance of transcripts for annotating regions. It was also likely that these novel exons were essentially an extension of the previously known exons which could happen if the previously annotated exons were shorter than the actual exons. In order to rule out such a scenario, we calculated the distance of these novel exons to the previously unannotated exons. The distribution of these distances is similar to the distance of any known exon to any other known exon (Figure 4.7D). 110 Next, we used ribotricer to detect translation in novel exons. All of the 71 exons found have a potential open reading frame (ORF). Though all the ORFs show an accumulation of RPFs, ribotricer is able to find evidence of active translation in only one of them (Figure 4.8D) . 4.2.5 Detecting pausing sites Translation elongation is the most dynamic stage of translation when the ribosome scans one codon at a time and decodes it, resulting in a new amino acid that gets added onto the nascent peptide chain . Translation progresses at the speed of ~ 6 amino acids per second [154, 74], but this speed is not constant. Ribosomes may slow down or speed up at certain codons during elongation to regulate protein synthesis. There is emerging evidence for the elongation process to be regulatory. In particular, it has been shown to be critical for early development [155], functioning of neurons [156, 157] and cancer [158]. Ribo-seq [24] provides nucleotide level resolution of the ribosomes attached to mRNA. It has been used to demonstrate the slow elongation rates at proline [74, 64] , translation inhibition caused by a drug compound inducing sequence dependent stalling [1] and heat shock induced pausing [28]. We developed a new method to identify transcriptome-wide stalling sites using Ribo-seq data. Briefly, the method relies on smoothing the Ribo-seq profile and then locates peak pileups in the smooth profile (See methods). Applying this method to our data revealed stalling in two genes: an adhesion protein ALSl and the 60S ribosomal protein RPLll (Figure 4.8). Both the pausing sites are reproducible across all the three replicates of both the conditions. 111 A 30°c B 30°c 750- ALS1 .h.i] . , I L .. ,JI .JJhJ,.J 6000- RPL11 - 500- 4000- Ill 250- Ill 2000- 'O ' d .• L, .. 'O II L ,I Ill Ill Q) 0 Q) 0 ... ... .... I I I I I I .... I I I I I I 0 0 ... 37°C + Serum ... 37°C + Serum Q) Q) ~ 2000 ALS1 .c RPL11 .. 1111 E 6000- :I - :I z z 4000- 1000 2000- , 1i1.lli. I 0 0 I l11,1111 ,l .. 1 11,,,,,,,1 ,, 1 .111 11 I I I I I I I 0 200 400 600 800 1000 1200 0 25 50 75 100 125 150 175 Location (nt) Location (nt) Figure 4.8 Pausing in Candida albicans. A) A reproducible pausing site in both 30°( and 37°( + Serum conditions. A) ALSl and B) RPLll. 4.3 Methods 4.3.1 RNA-seq and Ribo-seq data analysis The quality of raw sequences reads from RNA-Seq and Ribo-Seq datasets were assessed using FastQC [159]. Adaptor sequences and low-quality score (phred quality score < 5) bases were trimmed from RNA-Seq and Ribo-Seq datasets with TrimGalore (v0.4.3) [160]. Trimmed sequences from RNA-seq and Ribo-seq were both mapped using STAR (v.2.5.2b) [95] using Assembly 22 fasta as the reference and GTF r27 from CGD allowing a mismatch of at most two positions. All the reads mapping to rRNA and tRNA sequences were filtered out before downstream analysis. 112 4.3.2 Differential expression and translational efficiency analysis Differential expression analysis was performed using DESeq2 [147] . We filtered out genes with at least one read per replicate before doing the library size normalization and running the moderated t-statistic test. Genes are said to be differentially expressed if their absolute fold change on log 2 scale is at least 1 and the FDR adjusted p-value is at least 0.05. Gene ontology analysis was performed using clusterProfiler [161] using GO slim ontology file available from CGD. Differential translational efficiency was performed using riborex [150]. Only genes that had a a read count of at least one per replicate were used as input to riborex. We define genes to be exhibiting differential translational efficiency if their absolute fold-change on log 2 scale is at least 1 and the non-adjusted p-value is less than 0.05. 4.3.3 Detecting pausing sites In Ribo-seq data, stalling site appears as sharp peaks [58]. However, the heterogeneity and the sparsity in data deems this task particularly challenging. A naive approach of identifying pausing sites using Z-scores [60] results in too many high-positives. In scenarios where ribosomal pausing leads to queuing up of ribosomes near the start codon site, the downstream profile might end up with fewer ribosomes due to some ribosomes getting dropped off at the paused site [28]. The Z-score approach ignores the trend effects that could arise from such scenarios besides ignoring the heterogeneity of the data. Moreover, the thresholds are arbitrarily defined (25 in [7 4] and 2 in [162]). In our experiments with real data, the Z-score distribution of the normalized RPF counts appears to be long-tailed (Figure 4.9b). 113 c ENSG00000169174 10 min (treated) ::, 0 u 150 u.. c.. a:: 100 "O Q.) N 50 ro E I.... 0 z 0 0 200 400 600 (a) RPF profile of PCSK9 Figure 4.9 1.2 1.0 0.8 0.6 0.4 0.2 0.0 0 10 Z score (b) Z-scores 15 RPF profile of PCSK9 gene from data in Lintner et al. [l ]. (a) The pausing signal observed at around 65 th codon in the form of a sharp peak is a true pausing site. (B) The associated Z- score has a long tailed distribution, thus arbitrary thresholds leads to lot of false positives. Given the heterogeneous nature of read counts in Ribo-seq profile, we considered smoothing the read counts based on adjacent read counts, the motivation being reducing the heterogeneity by pooling observations. The smoothed counts can then be passed onto a peak finding method such as even the Z-score based approach , provided the resulting distribution is near gaussian. We propose using a de-noising approach using Savitzky-Golay filter [163] that has been applied to a wide range of problems involving signals of similar nature [164]. Savitzky-Golay filter finds a low-degree polynomial fit over adjacent points by the method of linear least squares. Post filtering, we use Z-scores to identify peaks at sub-codon resolution. Savitzky-Golay filter is a digital filter that acts as a low-pass filter for smoothing the data. It increases the signal to noise ratio without distorting the signal overall. This is achieved by convolution, where in sub-sets of adjacent data points are fitted with a low-degree polynomial by the method of linear least squares. An analytical solution exists for finding the solution to least squares problem if the data points are equally spaced in the form of "convolution coefficients". 114 Let Y 1 ,r = {y 1 , y 2 ... , Yr } represent the profile of read counts over T codons. On treating them with m convolution coefficients the transformed points are rn- 1 - 2 - y ; = L CjYt+i j = 1 2 m , m-1 m-1 -- < t <T--- 2 - - 2 The coefficients Cj are analytically derived as follows. Consider a modified variable z such - - (T+l) that given codons {1 , 2, ... , T} such that z = t - t where t = 2 . Fitting a polynomial of degree k , The coefficients a i are solved using least-squares approximation, where i th row of J is given by (1, zi, z;, ... , zf ). Hence the coefficients Cj above are given by For each possible ORF and the corresponding Ribo-seq profile as obtained from ribotricer, we apply the Savitzky-Golay filter to this profile. Peaks are then called such that the called site has a signal to noise ratio above 2.5 where the noise is estimated by fitting a single variance parameter for the entire profile. For each such peak the corresponding p-value is calculated a gaussian distribution whose mean and variance are empirically estimated from the given profile. The p-values are corrected for multiple testing using Benjamini-Hochberg procedure [165]. 115 Chapter 5 Integrated analyses of early responses to radiation in glioblastoma 5.1 Introduction Glioblastoma is the most common intracranial malignant brain tumor with an aggressive clinical course. Standard of care entails maximally safe resection followed by radiotherapy with concomitant and adjuvant temozolomide. Nonetheless, the median overall survival remains ap proximately 16 months [166, 167], and the recent addition of tumor-treating fields to the standard of care has only increased median overall survival to 20.5 months [166]. Recurrence occurs in part because glioblastoma uses sophisticated cellular mechanisms to repair DNA damage from double-stranded breaks caused by ionizing radiation, specifically homologous recombination and non-homologous end-joining. Thus, the repair machinery confers a mechanism for resistance to radiation therapy. Ionizing radiation can also cause base damage and single-strand breaks, which are repaired by base excision and single-strand break repair mechanisms, respectively [168]. A 116 comprehensive analysis of molecular mechanisms driving resistance to chemotherapy and radiation is required to surpass major barriers and advance treatments for glioblastoma. The Cancer Genome Atlas (TCGA) was instrumental in improving the classification and iden tification of tumor drivers [169], but its datasets provide limited opportunities to investigate radiation response. Thus, studies using cell and murine models are still the best alternatives to evaluate radiation response at the genomic level. The list of biomarkers associated with radi ation resistance in glioblastoma is still relatively small. Among the most relevant are FOXMl [170, 171], STAT3 [171], LlCAM [172], NOTCHl [173], RAD51 [174], EZH2 [175], CHKl/ ATR [176], COX-2 [177], and XIAP [178]. Dissecting how gene expression is altered by ionizing radia tion is critical to identify possible genes and pathways that could increase radio-sensitivity. A few genomic studies [179, 180, 181] have explored this question, but these analyses were restricted to describing changes in transcription . Gene expression is regulated at multiple levels, and RNA-mediated mechanisms such as splic ing and translation are particularly relevant in cancer biology. A growing number of inhibitors against regulators of splicing and translation are being identified [182]. Splicing alterations are a common feature across cancer types and affect all hallmarks of cancer [183]. Numerous splicing regulators display altered expression in glioblastoma ( e.g. PTBPl, hnRNPH, and RBM14) and function as oncogenic factors [184]. Importantly, a genome-wide study using patient-derived models revealed that transformation-specific depended on RNA splicing machinery. The SF3b complex protein PHF5A was required for glioblastoma cells to survive, but not neural stem cells (NSCs) . Moreover, genome-wide splicing alterations after PHF5A loss appear only in glioblas toma cells [185]. Translation regulation also plays a critical role in glioblastoma development. Many translation regulators such as elF4E, eEF2, Musashil, HuR, IGF2BP3, and CPEBl promote 117 oncogenic activation in glioblastoma , and pathways linked to translation regulation ( e.g., m TOR) promote cancer phenotypes [186]. To elucidate expression responses to radiation , we conducted an integrated study in U251 and U343 glioblastoma cell lines covering transcription (mRNAs and lncRNAs) , splicing, and translation. We determined that the downregulation of FOXMl and members of the E2F family are likely the major drivers of observed alterations in cell cycle and DNA replication genes upon radiation exposure. Genes involved in RNA regulatory mechanisms were particularly affected at the transcription, splicing, and translation levels. In addition, we identified several oncogenic factors and genes associated with poor survival in glioblastoma that displayed increased expression upon radiation exposure. Importantly, many have been implicated in radio-resistance, and therefore, their inhibition in combination with radiation could increase therapy efficacy. 5.2 Methods 5.2.1 Cell culture and radiation treatment U251 and U343 cells were obtained from the University of Uppsala (Sweden) and main tained in Dulbecco's Modified Eagle Medium (DMEM, Hyclone) supplemented with 10% fetal bovine serum, 1 % Penicillin/ Streptomycin at 37°C in 5% COrhumidified incubators and were sub-cultured twice a week. Cells were plated after appropriate dilution, and ionizing radiation treatment was performed on the next day at a dose of 5 Gray (Gy). A cabinet X-ray system (CP-160 Cabinet X-Radiator; Faxitron X-Ray Corp., Tucson , AZ) was used. After exposure to ionizing radiation, cells were cultured for 1 and 24 hours (hrs). 118 A 5Gyioo0iog ~ TO T1 I I I I -0-0-0-0 Total RNA Ribosome Total RNA Ribosome footprints footprints I I ~ I I -0-0 Total RNA Ribosome footprints 5Gyionizingradiation @ I I~ 6v ~ ~ I T24 I I Tl I ' TO I -0-0-0-0-0-0 Total RNA Ribosome Total RNA Ribosome Total RNA Ribosome footpriots footprints footprints RNA-Seq Ribo-Seq RNA-Soq Ribo-Seq RNA-Seq Ribo-Soq RNA-Seq Ribo-Seq RNA-Seq Ribo-Seq RNA-Seq Ribo-Seq Gene co-expression analysis Aternativesplicing analysis i Differential gene expression analysis i Differential translational efficiency analysis Figure 5.1 B g ~ ~ ! ~ RNA-seq . 20 . Time • To 10 . e n e T24 Cellline ,. • U251 - A U343 -10 • ' t -,0 " PC1:89% Variance Ribo-seq . 20 . . Time 10 0 TO O T1 e T24 0 Cellline . .. • U251 1 • A U343 -10 . . -60 -30 0 30 60 PC1: 88% Variance Experimental design and Principal Component Analysis of RNA-Seq and Ribo- Seq data from glioblastoma cell lines. A) Schematic representation of experimental protocol followed for radiation exposure of glioma cell lines, and sample preparation for sequencing the RNA and ribosome footprints. B) Principal component analyses performed on normalized log- transformed read counts of RNA-Seq and Ribo-Seq datasets. 119 ~ LL a. "' .,: z "' E 4.0 M 3.2M 2.4 M 1.6 M ~~ ~ l o 800.0 k '1t: 0.0 3.0M 2.5M 2.0 M 1.5M 1.0M il~l 0 500.0 K '1t: 0.0 !f. ;;~l ~ 2.5M <( 2.0 M Z 1.SM ! so~:6~ '1t: 0.0 ~ LL a. "' .,: z "' E 2.SM 2.0M 1.5M 1.0M ;;~l ~ 50�.i.~ I m j c::: 1.0M E 750.0M 0 ~gg:g~ '1t: 0.0 ~ LL a. "' .,: z "' E 3.0 M 2.5 M 2.0 M 1.SM 1.0M ;:~l 0 500.0 K '1t: 0.0 /f_ 1.8M l a.. 1.SM C::: 1.2 M ~ 75~--~ ~ E 500.0 K o 250.0 K '1t: 0.0 Ribo-seq fragment length distribution II U251_L1_5GY_T24_S6 ___ ..• I I. _ _ ~5 3~ 3~ 4~ I U251_L2_5GY_T1_S8 ----····' I .. ___ _ _ I 20 20 20 I 25 25 25 I 30 30 30 I 35 35 35 I 40 40 40 I U343_L1_5GY_T1_S11 -·•····· 1 ••. __________ _ I 45 45 45 20 25 30 35 40 45 U343_L2_5GY _ T24_S15 ---·•···111 •. ___ _ 20 25 30 35 40 45 I U343_L3_5GY_T1_S17 .......• 1 •-------- 20 25 30 35 40 45 I 50 50 50 s 1 o 50 50 50 4.0 M 3.2 M 2.4M 1.6 M m l 0 800.0K '1t;; 0.0 2.5 M 2.0 M 1.5M 1.0 M ;; ~ l o 500.0K lt;; 0.0 ~ LL a. "' .,: z "' E 3.6 M 3.0 M 2.4 M 1.8 M 1.2 M :~~ ! 0 600.0 K =tt: 0.0 ~ I m 1 er: 1s6:2 ~ E 500.0 K o 250.0 K '1t: 0.0 ~ ~ m ! 0:: 2.4 M <C 2.0 M Z 1.6M i ao6·.% ~ o 400.0 K '1t: 0.0 ~ LL a. "' .,: z "' E 3.0 M 2.5 M 2.0 M 1.5 M 1.0 M t:~ I o 500.0 K '1t: 0.0 ~ 28 M l :t 2:4 M ; ~:~ ~ c::: 1.2 M E 800.0 K 0 400.0 K '1t: 0.0 !f. ~ :~~l i ;:;~ E 1.6M 0 800.0K ~ 0.0 Figure 5.2 ______ ...111.__ U251_L2_5GY _T1_S5 20 25 30 35 I I I I 20 25 30 35 40 I 40 45 I 45 ____ __ ..• 111. __ ~"-'"'-'"' - ''-~ 20 25 30 35 40 45 II U343_L1_0GY_T1_S10 ---·····' '•--------- 20 25 30 35 40 45 I U343_L3_5GY_T1_S14 --•·····I I. _______ _ 20 25 30 35 40 45 I U343_L3_0GY _ T0_S16 ___ ...• 11 I .. ______ _ 20 25 30 35 40 45 I U343_L3_5GY_T24_S18 ......•• 1 I. ________ _ 20 25 30 35 40 45 50 I 50 50 50 5~ 50 50 50 Fragment length distribution of ribosome footprints of glioblastoma cell lines. Fragment lengths distribution was obtained using ribotricer. 120 n U343_L2_5GY _ T24_S15 ·i r =•- •_ - _ w _ ~ l; ~ ... , .... 1.-,,,. •• lo -:i'o ; ;o s'o ~o 1;0 1~ 1~ i10 2~0 2 1 10 3~0 35a 3~0 3'oo 4 1 20 Ao~· !]~ ~ -.::::... lo -io ~ ;o io ~o 1;0 1~ 1~ i10 2~0 2 1 10 3~0 Jo 3~0 3'oo 4 1 20 ~ ~ • U251_L3_5GY_T1_S8 ..... ,1111:•-~-- ·----············· 30 60 90 120 150 180 210 240 270 300 330 360 390 420 450 480 -60 -30 0 30 60 90 120 150 180 210 240 270 300 330 360 390 420 450 480 r ~-l cto11••11•:::: .. ,.. o I ~ -3~ ~ :i'o ~ :o 1~0 1~ ,:a 2~0 2~0 2~0 3~0 3io 3:o 3 to 4;0 4;0 4:o • 7 z:- 6 l' U343_L1_5GY_T1_S11 ! ~ ,...., ~•u 1111• •• 11 ••11••• • ,., i I I -60 -30 0 • • • • • • • • • • • • • • I I•• 30 60 90 120 150 180 210 240 270 300 330 360 390 420 450 480 8 -~ ~ L ' U343_L2_5GY_T1_S14 ~ 5 Cl 4 lt 3 "' 2 ~ r-, tA ll AWi I l;Jt1 I b l f 11• Ii- .... ·····------------···· ~~ow~oow~w==m=~=~=== l , U343_L1_0GY_T0_S10 z:- 4 ] 3 ~ 2 "' 1 - .ll111MM•Nt•111nsh1UIW 11.i o f . I I I I -60 -30 0 5 I I I I I I I I I I I I I I I I •• 30 60 90 120 150 180 210 240 270 300 330 360 390 420 450 480 8 7 z:- 6 ~ 5 c3 4 lt 3 "' 2 1 0 U343_L3_0GY_T0_S16 II Pllll 7' 11 Jlll"ir5 US * t f 1 I I • • • • I I I I I I I I I I I•• •~owoooow~w==m~=~=-== U343_L3_5GY_T1_S17 • • • • • • • • I I I I I I I I I•• -60 -30 0 30 60 90 120 150 180 210 240 270 300 330 360 390 420 450 480 I I I I I I I I I I I I I I I•• •~omoooow~w==m~=~==== I I I I I I I I I I I I I I I•• -60 -30 0 30 60 90 120 150 180 210 240 270 300 330 360 390 420 450 480 Figure 5.3 The ribosome density profiles of glioblastoma cell lines. The metagene distribution was obtained using ribotricer. 121 5.2.2 RNA preparation, RNA-seq and Ribosome Profiling (Ribo-seq) RNA was purified using a GeneJet RNA kit from Thermo Scientific. The TruSeq Ribo Profile (Mammalian) kit from lllumina was used to prepare material for ribosome profiling (Ribo-seq). RNA-seq and Ribo-seq samples were prepared according to lllumina protocols and sequenced at UTHSCSA Genome Sequencing Facility. 5.2.3 Overall strategy to identify gene expression alterations upon radiation To identify the most relevant expression alterations in the early response to radiation, we analyzed samples from U251 and U343 cells collected at 0 (TO), 1 (Tl) , and 24 (T24) hours post-radiation . To capture the progressive dynamics of expression alterations, we compared TO to Tl samples and Tl to T24 samples. Our strategy to identify the most relevant alterations in expression with maximal statistical power was to combine all samples and use a design matrix with cell type defined as a covariate with time points (Figure 51) . 5.2.4 Sequence data pre-processing and mapping The quality of raw sequences reads from RNA-Seq and Ribo-Seq datasets were assessed using FastQC [159]. Adaptor sequences and low-quality score (phred quality score < 5) bases were trimmed from RNA-Seq and Ribo-Seq datasets with TrimGalore (v0.4.3) [160]. The trimmed reads were then aligned to the human reference genome sequence (Ensembl GRCh38.p7) using STAR aligner (v.2.5.2b) [95] with GENCODE [187] v25 as a guided reference annotation, allowing a mismatch of at most two positions. All the reads mapping to rRNA and tRNA sequences were filtered out before downstream analysis. Most reads in the Ribo-seq samples mapped to the 122 coding domain sequence (CDS) . The distribution of fragment lengths for ribosome footprints was enriched in the 28-30 nucleotides range, as expected (Figure S2) . The ribosome density profiles exhibit high periodicity as within the CDS, as expected since ribosomes traverse three nucleotides at a time (Figure S3) . The periodicity analysis was performed using ribotricer [149]. The number of reads assigned to annotated genes included in the reference genome was obtained by htseq-count [188]. 5.2.5 Differential gene expression analysis For differential expression analysis, we performed counting over exons for the RNA-seq sam ples. For translational efficiency analyses, counting was restricted to the CDS. A Principal Com ponent Analysis (PCA) was then performed on RNA-Seq and Ribo-Seq data from U251 and U343 cells. Most variation was explained by the cell type along the first principal component, and radiation time-related changes were captured along the second principal component (Sup plementary Figure SlB). Differential gene expression analysis was performed by employing the DESeq2 package [147], with read counts from both U251 and U343 cell samples as inputs. We adjusted p-values controlling for the false discovery rate (adjusted p-value) using the Benjamini and Hochberg (BH) procedure [165]. Differentially expressed genes were defined with an adjusted p-value < 0.05 5.2.6 Weighted gene co-expression network analysis Weighted Gene Co-expression Network Analysis (WGCNA) [189] uses pairwise correlations on expression values to identify genes significantly 123 co-expressed across samples. We used this approach to identify gene modules with significant co-expression variations as an effect of radiation. The entire set of expressed genes, defined here as those with one or higher transcripts per million higher (TPM) , followed by variance stabilization) from U251 and U343 samples were clustered separately using the signed network strategy. We used the Z summary [190] statistic as a measure of calculating the degree of module preservation between U251 and U343 cells. Z summary is a composite statistic defined as the average of the density and connectivity based statistic. Thus, both density and connectivity are considered for defining the preservation of a module. Modules with Z summary > 5 were considered as significantly preserved . The expression profile of all genes in each co-expression module can be summarized as one "eigengene". We used the eigengene- based connectivity (kME) defined as the correlation of a gene with the corresponding module eigengene to assess the connectivity of genes in a module. The intramodular hub genes were then defined as genes with the highest module membership values (kME > 0.9) . All analysis was performed using the R package WGCNA. The protein-coding hub genes were then selected for gene ontology enrichment analysis. 5.2. 7 Translational efficiency analysis We used Riborex [150] to perform differential translational efficiency analysis. The underlying engine selected was DESeq2 [147]. DESeq2 estimates a single dispersion parameter per gene. However, RNA-Seq and Ribo-Seq libraries can have different dispersion parameters owing to different protocols. We estimated the dispersion parameters for RNA-Seq and Ribo-Seq samples separately and found them to be significantly different (mean difference = 0.04, p-value < 2.2e-16). This leads to a skew in translational-efficiency p-value distribution since the estimated 124 null model variance for the Wald test is underestimated. To address this issue, we performed a p-value correction using fdrtool [191] that re-estimates the variance using an empirical bayes approach . 5.2.8 Alternative splicing analysis Alternative splicing analysis was performed using rMATS [192]. All reads were trimmed using cutadapt [193] with parameters (-u -13 -U -13) to ensure trimmed reads had equal lengths (138 bp) . rMATs was run with default parameters in paired end mode (-t paired) and read length set to 138 bp (-len 138) using GENCODE GTF (v25) and STAR index for GRCh38. 5.2. 9 Gene ontology (GO) and pathway enrichment analysis To classify the functions of differentially enriched genes, we performed GO enrichment, and the Reactome pathway [194] analysis using Panther [195]. For both analyses, we considered terms to be significant if BH adjusted p-values weree < 0.05, and fold enrichment is > 2.0. Further, we used REVIGO [2] to reduce redundancy of the enriched GO terms and visualize the semantic clustering of the identified top-scoring terms. We used STRING database (vlO) [3] to construct protein-protein interaction networks and determine associations among genes in a given dataset. The interactions are based on experimental evidence procured from high-throughput experiments text-mining, and co-occurrence. Only high-confidence (0.70) nodes were retained . 125 5.2.10 Expression correlation analysis Gene expression correlation analysis was done using Gliovis [4] using glioblastoma samples (RNAseq) from the TCGA. To select correlated genes, we used Pearson correlation , R > 0.3, and p-value < 0.05. A list of genes affecting survival in glioblastoma was downloaded from GEPIA [196]. A list of long non-coding RNAs (lncRNAs) implicated in glioma development was obtained from Lnc2Cancer [197]. Drug-gene interactions were identified using the Drug-Gene Interaction Database [198]. 5.2.11 Availability of data and materials The processed data of read abundance matrices is available through GEO accession GSE141013. Scripts for differential expression analysis and translational efficiency analysis are available at https://github . com/saketkc/2019_radiation_gbm Table S1 • Table S2 Table S3 Table S4 Table S5 Table S6 • Table S7 Table S8 126 Table 59 Table 510 5.3 Results A C U251_L1_T:U u2s,_u_1:u U251_U_T2' U251_L2_T1 U251_L1_TO 200 U251_L2_T0 150 U251_L1_T1 U251_L3_T1 100 U343_L1_TO 50 U343_L1_T1 U343_L2_T0 O U343_L2_T1 U3-l3_U_T0 U3-l3_U_T1 U343_L1_T2' U3-l3_L2_T2' U3-l3_U_T2' 3000 - l ·- I::"" ! 15 & ~ 2000 I!! 50 0 • .! l 100:'--- -- - -~-.---------J i 25 -- . -- U251_L3_SGY_T1 U251_L2_0GY_TO U251_L2_5GY_T1 UZ111_TC1_0GY_T0 120 l ~=:-:~1;:v~; ~~ um=rc~_l!G~_w loo U251_L3_SGY_T24 40 U3'3_L1_SGY_T24 20 U3'3_L2_SGY_T24 0 U3'3_L3_5GY_T24 U3'3_L1_0GY_TO U30_L1_5GY_T1 U3O_L2_0GY_T0 U30_L2_5GY_T1 U3O_L3_0GY_TO U30_L3_5GY_T1 B D Figure 5.4 60~-~-~ - I:-" 20 ,;_ , J.,., . _ Jog, (T24/Tl) DE 10.0~-~-~ : ·.·.·. 2.6 . . /·.,. : .. !,:·. 0.0 ~ - - ~ _ : : ··:~~t ~~~ -- -1.6 -1.0 � .6 0.0 0.5 1.0 log, (T24/Tl) TE ~ ! ~ l ~ " Contrast i - Global view of glioblastoma cell lines transcription and translation profiles after radiation. A) Differentially expressed genes (left) and the number of genes whose translation efficiency is differentially regulated (right) after radiation exposure at different time points. B) Volcano plots showing the expression and transition alterations of genes at 24 hr compared to 1 hr after radiation exposure. Blue dots indicate upregulated genes (adjusted p-value < 0.05, log 2 fold change < 0), and orange dots indicate down regulated genes. ( adjusted p-value < 0.05, log 2 fold change < 0). C) Sizes of gene modules found in U251 and U343 cell lines. D) Preservation Median Rank and Zsummary for all modules. A lower median rank indicates the module is preserved, and the corresponding modules in U251 and U343 cell lines share a high number of genes. A Zsummary score of 2-10 indicates weak preservation, while a Zsummary > 10 indicates high preservation. 127 5.3.1 Changes in global transcriptome profile in response to radiation We first conducted an integrated analysis to evaluate the early impact of radiation [1 hour (Tl) and 24 hours (T24)] on the expression profile of U251 and U343 GBM lines. A relatively small number of genes displayed altered expression at Tl (Supplementary Figures 54A and 54B) . Downregulated genes are mainly involved in transcription regulation and include 18 zinc finger transcription factors displaying high expression correlation in glioblastoma samples from TCGA (Table 51) . Upregulated sets contain genes implicated in cell cycle arrest, apoptosis, and stress such as ZFP36, FBXW7, SMAD7, BTG2, and PLK3 (Table 51) . Since many alterations were observed when comparing the Tl vs. T24 time points (Table 51 ) , we opted to focus on genes showing the most marked changes (log 2 fold-change > 1.0 or < -1.0 and adjusted p-value < 0.05) to identify biological processes and pathways most affected at the T24 time point. Top enriched GO terms and pathways among downregulated genes include chromatin remodeling, cell cycle, DNA replication , and repair (Figure 5.5A) . Additionally, we identified several GO terms associated with mRNA metabolism , decay, translation , and ncRNA processing, suggesting active participation of RNA-mediated processes in radio-response (Figure 5.5B). Network analysis indicated the set of genes in these categories is highly interconnected (Figure 5.5C and Table 52) . To expand the expression analysis, we employed WGCNA [189] to identify gene modules with significant co-expression variation as an effect of radiation. All identified modules, along with the complete list of genes in each module, are shown in Supplementary Figures 54( and 54D and Table 53. Seven modules were identified (Zsummary > 5) as tightly regulated, independent 128 A CENP-A containing chromatin organization (GO:0061641) chromatin remodeling at centromere (GO:0031055) chromosome organization (GO:0051276) DNA strand elongation (GO:0022616) interstrand cross-link repair (GO:0036297) signal transduction in response to DNA damage (GO:0042770) ONA synthesis involved in DNA repair (G0:0000731) postreplication repair (GO:0006301) chromosome segregation (GO:0007059) cell division (GO:0051301) cell cycle process (GO:0022402) cell cycle (GO:0007049) signal transduction by p53 class mediator (GO:0072331) regulation of signal transduction by p53 class mediator (G0:1901796) protein localization to chromosome (GO:0034502) B U4 snRNA 3'-end processing (GO:0034475) snoRNA processing (GO:0043144) nuclear RNA surveillance (GO:0071027) rRNA processing (GO:0006364) RNA 3'-end processing (GO:0031123) ribonucleoprotein complex localization (GO:0071166) RNA export from nucleus (GO:0006405) ribosome biogenesis (GO:0042254) RNA localization (GO:0006403) ncRNA processing (GO:0034470) ribonucleoprotein complex biogenesis (GO:0022613) ncRNA metabolic process (GO:0034660) -log10(pvalue) fold-enrichment - 5 10152025 0 5101520 - "' C: ::, "' iii E > .r:. .S . !::? 0 ~ ~ C: O> "' 0 ' - ,:, ·~ C Figure 5.5 Characteristics of downregulated genes at 24 hours (T24) after radiation exposure in glioblastoma cell lines. A) Enriched gene ontology related to cell cycle, DNA replication, and repair among downregulated genes. B) RNA-related Gene Ontology (GO) terms enriched among downregulated genes summarized using REVIGO [2]. C) Protein-protein interaction network, according to STRING [3] showing downregulated genes associated with RNA-related functions. Gene clusters based on the strength of connection and gene function are identified by color. Lines colors indicate the type of association: light green indicates an association based on literature findings; blue indicates gene co-occurrence; magenta indicates experimental evidence. 129 of the cell line (Figure S4D) . Among modules with the highest significant correlation (0.8, p value< l e-7), module 2 contains genes downregulated in T24, with many involved in cell cycle, metabolism mRNA metabolism , processing, splicing, and transport (Table S3) , corroborating results described above. A FOXM1 E2F1 E2F2 E2F8 al iii u: N iii u. :, N N :, al w w al :; >:: 0 <( (.) "' <O ... N "' "' .., (.) al ... (.) N N (.) C 0:: C (.) (.) C (.) (.) C C (.) (.) (.) 8 E2F1 E2F2 12 10.0 N° C) g_ 10 7.6 C: 0 ·;;; "' f 8 Q, 6.0 )( " <( z 2.5 0:: 6 E 0.0 . ,, ,,, ~ ,, ,,, ~ E2F8 FOXM1 ' N° C) 12 0 ::::. 6 C: 0 10 ·;;; e 4 Q, )( ; 2 z 0:: E o \\ ,,, ~ \\ ,,, ~ Grade Grade .... " N (.) <( C w z (.) :c (.) (.) (.) C 0.96 0.88 0.80 iii N " N :; N .., .., "' <O .... " <( ;::: - 0.56 w >:: :. :. :. :. :. :. z z z C C ll. (.) (.) (.) (.) (.) (.) ..J (.) > (.) (.) (.) (.) rn :. :. :. :. :. :. ll. ll. :. (.) (.) w >:: ll. Fif7F9F9 \]lLij L::J L:J FOXM1 0.732 ... m N 3! :~LuBB ::(?l~llllq o~~~~ ::~ ., . ·. - ~---~ =·=~- _ .. · .. rncl s,T• ••• • •' • i . . . 6 • • • .,., ... 6 8 10 2.5 5.0 7.5 10 2 4 6 8 10 6 8 10 12 Figure 5.6 E2Fs and FOXMl in glioblastoma. A) Correlation of E2Fl, E2F2, E2F8, and FOXMl with target genes involved in cell cycle. B) Expression levels of E2Fl, E2F2, E2F8, and FOXMl in gliomas grades II, Ill, and IV in TCGA samples. C) E2Fl, E2F2, E2F8, and FOXMl expression correlation in glioblastoma (TCGA samples) using Gliovis [4]. *** p-value < 0.0001. Next, we investigated downregulated genes with the gene set enrichment analysis (GSEA) tool Enrichr [199] and conducted expression correlation analysis with Gliovis [4]. Based on their genomic binding profiles and effect of gene expression, FOXMl and the E2F family of transcription factors emerged as potential regulators of a large group of cell cycle/ DNA replication-related 130 genes in the affected set (Figure 5.6A, Table 54) . In agreement, E2Fl , E2F2, E2F8, and FOXMl displayed a significant decrease upon radiation . FOXMl and E2F factors have been previously implicated in chromatin remodeling, cell cycle regulation , DNA repair, and radio-resistance [200, 201]. All four factors are highly expressed in glioblastoma with respect to low-grade glioma. Importantly, they display high expression correlation with a large set of downregulated genes implicated in cell cycle and DNA replication and among themselves in glioblastoma samples in TCGA (Figure 5.6B-C). A extracellular structure organization (GO:0043062) extracellular matrix organization (GO:0030198) axonogenesis (GO:0007409) gland morphogenesis (GO:0022612) regulation of cell size (GO:0008361) negative regulation of cell projection organization (GO:0031345) sensory perception of mechanical stimulus (GO:0050954) cell morphogenesis involved in neuron differentiation (GO:0048667) defense response to virus (GO:0051607) response to type I interferon (GO:0034340) type I interferon signaling pathway (GO:0060337) negative regulation of cellular component movement (GO:0051271) negative regulation of viral genome replication (GO:0045071) modulation of chemical synaptic transmission (GO:0050804) B Figure 5.7 Global view of upregulated genes at T24 post-radiation in glioblastoma cells. A} Gene ontology analysis of upregulated genes B} Protein-protein interaction networks according to STRING [3] showing genes associated with extracellular matrix organization and response to interferon. Gene clusters based on the strength of connection and gene function are identified by color. Lines colors indicate type of association: light green, association based on literature findings; blue indicates gene co-occurrence; magenta indicates experimental evidence. Upregulated genes at T24 are preferentially associated with the extracellular matrix recep- tor interaction pathway, extracellular matrix organization , axonogenesis, and response to type I interferon (Figure 5.7A, and Table 52). With respect to the extracellular matrix, we observed changes in the expression levels of several collagens (types II, IV, V, and XI), glycoproteins of 131 the laminin family (subunits a, /3, and')'), and also integrins (subunits a, and /3) (Figure 5.7B, and Table Sl). Collagen type IV is highly expressed in glioblastoma and implicated in tumor progression [202]. In addition , it has been observed that the activation of two integrins, ITGB3 and ITGB5, contributes to radio-resistance [203]. Radiation treatment also induced the expression of genes involved in neuronal differenti ation and axonogenesis. Some key genes in these categories include SRC, VEGFA, EPHA4, DLG4, MAPK3, BMP4, and several semaphorins. These genes can have very different effects on glioblastoma development, with some factors activating oncogenic programs and others behaving as tumor suppressors. Similarly, type I interferon's effects on treatment are varied . For instance, interferon inhibited proliferation of glioma stem cells and their sphere-forming capacity and in duced STAT3 activation [204]. On the other hand , chronic activation of type I IFN signaling has been linked to adaptive resistance to therapy in many tumor types [205]. Activation of oncogenic signals post-radiation could counteract treatment effects and later contribute to relapse. We searched the set of highly up-regulated genes post-radiation for previ ously identified radio-resistance genes in glioblastoma, oncogenic factors and genes whose high expression is associated with poor prognosis (Table S5) . In Table 5.1 , we list these genes accord ing to their molecular function . Since several of these genes have never been characterized in the context of glioblastoma, our results open new opportunities to prevent radio-resistance and increase treatment efficiency. Importantly, there are inhibitors available against several of these proteins (Table S4). 132 Table 5.1 List of oncogenic factors, genes whose high expression is associated with poor survival and genes previously associated with radio-resistance in GBM that showed increased expression upon radiation. Genes are listed according to molecular function. Function Genes Membrane protein AQPl, ARHGEF2, BAALC, CSFlR, CSPG4, EPS8L2, ERBB3, FGFR4, FYN, GPM6A, ITGB3, JUP Protein kinase ANKKl , CDKNlA, CSFlR, ERBB3, FAM20C, FGFR4, FYN, IKBKE, MERTK, PDGFRB, SRC, TEC Gene expression regula- ARID3A, ASAHl, BCL3, BCL6, CBX7, CEBPB, ELF3, tion FAM20C, FEZFl, HOXAl, HOXB9, JUP, KDM5B, LMOl, LMO2, MACCl, MAF, MSll, MUCl, NKX2-1, PML, PRDM6, RORC, SATBl, SREBFl, TP53BP1 , ZMYM2 Enzymatic activity ACSS2, AGAP2, APIP, ARHGEF2, ClR, CARD16, CD24, CDKNlA, CEBPB, CSFlR, CSPG4, CTSZ, CUL7, CYTH4, EPS8L2, ERBB3, FAM20C, FGFR4, FTHl, FUCAl, FYN, GHDC, 1DO1, IKBKE, ITGB3, JUP, KDM5B, MCF2, MERTK, MFNG, MRAS, NKX2- 1,PDE6G,PDGFRB, PML, QPRT, RRM2B, SERPINA5, SFN , SGSH, SRC, SREBFl, TEC, TGFBl , ZMYM2 Phosphotransferase CSFlR, ERBB3, FAM20C, FYN, IKBKE, MERTK, PDGFRB, SREBFl, TEC Cell surface receptor BMP7, CSFlR, ERBB3, FGFR4, FYN, ITGB3, ITGB5, LRIG2, MCF2, MERTK, MFNG, PDGFRB, PRDM6, SRC, TEC, TRPM8 Metabolism regulation ACSS2, APIP, BCL3, BCL6, BTG2, CEBPB,CRTCl , CSPG4, CTSZ, ELF3, FUCAl, 1DO1, ITGB3, JUP, MAF, MFNG, NKX2-1, PARP3, PRDM6, PTGES, QPRT, RRM2B, SGSH, TGFBl , TP53BP1, TRPM8, USP9X, VEGFA, ZMYM2 133 5.3.2 Changes in lncRNA profile in response to radiation lncRNAs have been implicated in the progression of glioblastoma [206], but their role in re sponse to ionizing radiation is still poorly understood. We identified 161 lncRNAs with expression alterations in Tl vs. T24 comparisons. Analysis of this set with LnC2Cancer [197] identifiedden tified several lncRNAs aberrantly expressed in cancer and with relevance to prognosis (Table S1 ) . We also detected significant downregulation of MIR155HG, whose high expression is associated with glioma progression and poor survival [207]. Another downregulated lncRNA with relevance to prognosis is linc000152, whose increased expression has been observed in multiple tumor types [208, 208]. On the other hand, we observed a significant upregulation of two "oncogenic" lncR NAs, NEATl and FTX. NEATl is associated with tumor growth , grade, and recurrence rate in gliomas [209], while FTX promotes cell proliferation and invasion through negatively regu lating miR-342-3p [210]. Thus, if further studies corroborate NEATl and FTX as players in radio-resistance, targeting these lncRNAs should be considered to improve treatment response. 5.3.3 Effect of radiation on splicing Alternative splicing impacts genes implicated in all hallmarks of cancer [211] and is an impor tant component of changes in expression triggered by ionizing radiation [212]. All types of splicing events ( exon skipping, alternative donor, and acceptor splice sites, multiple exclusive exons, and intron retention) were affected similarly upon exposure to radiation (Table S7) . At T24, we observed that transcripts associated with RNA-related functions ( especially translation), showed the most splicing alterations. Affected transcripts encode ribosomal proteins, translation initia tion factors, regulators of translation, and genes involved in tRNA processing and endoplasmic 134 A viral transcription (GO:0019083) translational initiation (GO:0006413) positive regulation of endopeptidase activity (GO:0010950) nucleobase-containing compound catabolic process (GO:0034655) histone modification (GO:0016570) ncRNA processing (GO:0034470) mRNA metabolic process (GO:0016071) RNA metabolic process (GO:0016070) RNA processing (GO:0006396) cellular nitrogen compound biosynthetic process (GO:0044271) regulation of cell cycle process (GO:0010564) cellular response to ONA damage stimulus {GO:0006974) protein localization to organelle (GO:0033365) j I 0 "' 0 .,. B Figure 5.8 '~ • F TIF ? 1 S15 MT11 Impact of radiation on the splicing profile of glioblastoma cells. A) GO-enriched terms among genes showing changes in splicing profiles at T24. GO-enriched terms are summarized using REVIGO [2]. B) Protein-protein interaction networks according to STRING [3] showing genes associated with RNA-related functions whose splicing profiles displayed alterations at T24. Gene clusters based on the strength of connection and gene function are identified by color. Lines color indicate type of association: light green, an association based on literature findings; blue indicates gene co-occurrence; magenta indicates experimental evidence. 135 reticulum . Other enriched GO terms include mRNA and ncRNA processing, mRNA degradation, and modification. Catabolism is another process associated with several enriched terms, sug- gesting that splicing alterations in genes involved in catabolic routes could ultimately contribute to apoptosis (Figure 5.8A-B, and Table S7). Changes in the splicing profile are likely driven by an alteration in the expression of splicing regulators. In Table 5.2, we show a list of splicing factors displaying strong expression alterations. Among those previously connected to glioblas toma development upon radiation, LGALS3 is the most extensively characterized. LGALS3 is a galactosidase-binding lectin and non-classic RNA binding protein implicated in pre-mRNA splicing and regulation of proliferation , adhesion , and apoptosis; LGALS3 also is a marker of the early stage of glioma [213]. Table 5.2 Splicing regulators showing changes in expression 24 hours post-radiation. Factors showing an increase in the expression are shown in red, while factors showing a decrease in the expression are represented in blue. Gene ID Gene name Function AHNAK2 Protein AHNAK2 splicing regulation ESRPl Epithelial splicing regulatory protein 1 regulation of mRNA splicing LGALS3 Galectin-3 signaling receptor binding NOVA2 RNA-binding protein Nova-2 alternative splicing regulation SNRPN Small nuclear ribonucleoprotein N spliceosomal snRNP assembly ALYREF THO complex subunit 4 RNA binding DDX39A ATP-dependent RNA helicase DDX39A RNA helicase GEMIN4 Gem-associated protein 4 rRNA processing HNRNPL Heterogeneous nuclear ribonucleoprotein L alternative splicing regulation LSM2 U6 snRNA-associated Sm-like protein LSm2 U6 snRNA-associated Sm-like protein MAGOHB Mago nashi homolog 2 exon-exon junction complex PPIH Peptidyl-prolyl cis-trans isomerase H ribonucleoprotein complex binding RBMX RNA-binding motif protein, X chromosome regulation of mRNA splicing SNRPDl Small nuclear ribonucleoprotein Sm D1 spliceosomal snRNP assembly SNRPE Small nuclear ribonucleoprotein E spliceosomal snRNP assembly SRSF2 Serine/ arginine-rich splicing factor 2 regulation of mRNA splicing SRSF3 Serine/ arginine-rich splicing factor 3 regulation of mRNA splicing TTF2 Transcription termination factor 2 transcription regulation 136 5.3.4 Differential translational efficiency We used Ribo-seq [31] to identify changes in translation efficiency triggered by radiation. Translation , protein localization , and metabolism appear as top enriched terms among downregu lated genes in Tl vs. T24 comparisons (Tables S8-S9). In particular, several ribosomal proteins, along with translation initiation factors and m TOR, showed a significant decrease in translation efficiency (Figure 5.9A-B) . Overall , these results indicate repression of the translation machin ery post-radiation exposure and its strong auto-regulation . Since changes in components of the translation machinery are occurring at all levels (transcription , splicing, and translation) at T24, we expect that major translational alterations take place in later stages of post-radiation . In the upregulated set, we highlight three genes FTHl, APIP, and LRIG2 that could potentially counteract the impact of radiation (Table S10) . FTHl encodes the heavy subunit of ferritin , an essential component of iron homeostasis [214]. Pang et al., 2016 [215] reported that H ferritin plays an important role in radio-resistance in glioblastoma by reducing oxidative stress and activating DNA repair mechanisms. The depletion of ferritin causes down-regulation of ATM, leading to increased DNA sensitivity towards radiation. APIP is involved in the methionine salvage pathway and has a key role in various cell death processes. It can inhibit mitochondria mediated apoptosis by directly binding to APAF-1 [216]. LRIG2 is a member of the leucine-rich and immunoglobulin-like domain family [217], and its expression levels are positively correlated with the glioma grade and poor survival. LRIG2 promotes proliferation and inhibits apoptosis of glioblastoma cells through activation of EGFR and Pl3K/ Akt pathway [218]. 137 A CENP-A containing chromatin organization (GO:0061641) chromatin remodeling at centromere (GO:0031055) chromosome organization (GO:0051276) DNA strand elongation (GO:0022616) interstrand cross-link repair (GO:0036297) signal transduction in response to DNA damage (GO:0042770) ONA synthesis involved in DNA repair (G0:0000731) postreplication repair (GO:0006301) chromosome segregation (GO:0007059) cell division (GO:0051301) cell cycle process (GO:0022402) cell cycle (GO:0007049) signal transduction by p53 class mediator (GO:0072331) regulation of signal transduction by p53 class mediator (G0:1901796) protein localization to chromosome (GO:0034502) B U4 snRNA 3'-end processing (G0:0034475) snoRNA processing (GO:0043144) nuclear RNA surveillance (GO:0071027) rRNA processing (GO:0006364) RNA 3'-end processing (GO:0031123) ribonucleoprotein complex localization (GO:0071166) RNA export from nucleus (GO:0006405) ribosome biogenesis (GO:0042254) RNA localization (GO:0006403) ncRNA processing (GO:0034470) ribonucteoprotein complex biogenesis (GO:0022613) ncRNA metabolic process (GO:0034660) -log10(pvalue) fold~nrichment - 5 10152025 0 5 10 15 20 -- .. C: :, .. iii E > .c: a. (J Q 'i: - C: "' .. 0 ' -~ C Figure 5.9 Impact of radiation on the translation profile of glioblastoma cells. A) GO-enriched terms among genes showing changes in translation efficiency at T24. GO-enriched terms are summarized using REVIGO [2]. B) Protein-protein interaction network, according to STRING [3] showing genes whose translation efficiency decreased at T24. Gene clusters based on the strength of connection and gene function are identified by color. Line colors indicate the type of association: light green, an association based on literature findings; blue indicates gene co-occurrence; magenta indicates experimental evidence. 138 5.3.5 Crosstalk between regulatory processes Parallel analyses of transcription, splicing, and translation alterations in the early response to radiation provided an opportunity to identify crosstalk between different regulatory processes. The datasets showed little overlap, with just a few genes showing alterations in two different regulatory processes. However, we identified several shared GO terms when comparing the results of alternative splicing, mRNA levels, and translation efficiency (Table S10). These terms show two main groups of biological processes. The first group indicates that the expression of genes involved in DNA and RNA synthesis and metabolism is particularly compromised . The second group is related to translation initiation. Ribosomal proteins were particularly affected (Figures 5.8 and 5.9). There is growing support for the concept of specialized ribosomes. According to this model , variations in the composition of the ribosome due to the presence or absence of certain ribosomal proteins or alternative isoforms could ultimately dictate which mRNAs get preferentially translated [219]. Therefore, these alterations could later lead to translation changes of a specific set of genes. 5.4 Discussion We performed the first integrated analysis to define global changes associated with the early response to radiation in glioblastoma. Our approach allowed the identification of "conserved " alterations at the transcription, splicing and translation levels and defined possible crosstalk between different regulatory processes. Alterations at the level of transcription were dominant, but changes affecting genes implicated in RNA mediated regulation were ubiquitous; they indicate 139 that these processes are important components in radio-response and suggest that more robust changes in splicing and translation might take place later. 5.4.1 E2Fl, E2F2, E2F8, and FOXMl as major drivers We observed marked changes in the mRNA levels of genes implicated in cell cycle, DNA replication , and repair 24 hours (T24) after radiation . Downregulation of several transcription factors, most of them members of the zinc finger family, was observed at one hour post-radiation . This group displays high correlations in expression within glioblastoma samples from TCGA, suggesting that they might work together to regulate gene expression . Unfortunately, most are poorly characterized, and the lack of information has prevented establishing further connections to changes in the cell cycle and DNA replication that we observed at T24. GSEA and expression correlation analysis suggested that the downregulation of members of the E2F family is likely responsible for several of the expression changes we observed at T24. E2Fs have been defined as major transcriptional regulators of the cell cycle. The family has eight members that could act as activators or repressors depending on the context, and are known to regulate one another. They are upregulated in many tumors due to overexpression of cyclin dependent kinases (CDKs) , inactivation of CDK inhibitors, or RB Transcriptional Corepressor 1 (RBl) and are linked to poor prognosis. Alterations in E2F genes can induce cancer in mice [220, 221]. Specifically, we found that three E2F members showed decreased expression upon radiation : E2Fl , E2F2, and E2F8, all of which have been previously implicated in glioblastoma development. 140 E2Fl is probably the best-characterized member of the E2F family. Besides its known ef fect on cell cycle regulation and DNA replication, it is also a positive regulator of telomerase activity, binding the TERT promoter [222]. Recent studies show that lncRNAs and miRNAs function in an antagonistic fashion to regulate E2Fl expression , ultimately affecting cell prolif eration, glioblastoma growth, and response to therapy [223, 224, 225]. E2F2 has been linked to the maintenance of glioma stem cell phenotypes and cell transformation [226, 227]. Several tumor suppressor miRNAs (let7b, miR-125b, miR-218, and miR-138) decrease the proliferation and growth of glioblastoma cells by targeting E2F2 [226, 228, 229, 230]. Although still poorly characterized in the context of glioblastoma , E2F8 drives an oncogenic phenotype in glioblas toma. Its expression is modulated by HOXD-ASl, which serves as a sponge and prevents the binding of miR-130a to E2F8 transcripts [231]. FOXMl is another potential regulator of the group of cell cycle and DNA replication genes affected by radiation . FOXMl is established as an important player in chemo- and radio-resistance and a contributor to glioma stem cell phenotypes [170, 171, 232, 233, 234, 235, 236, 237]. FOXMl and E2F protein have a close relationship and share target genes [238]. Additionally, FOXMl- and E2F2-mediated cell cycle transitions are implicated in the malignant progression of IDHl mutant glioma [239]. E2F and FOXMl targeting could be considered as an option to increase radio-sensitivity. Since the development of transcription factor inhibitors is very challenging, an alternative to be considered is the use of BET (bromodomain and external) inhibitors. BET is a family of proteins that function as readers for histone acetylation and modulates the transcription of oncogenic programs [240]. Recent studies in glioblastoma with a new BET inhibitor, dBET6, showed promising results and established that its effect on cancer phenotypes comes via disruption of the transcriptional program regulated by E2Fl [241]. 141 5.4.2 RNA processing and regulation as novel categories in radio-response Besides the expected changes in expression of cell cycle, DNA replication and repair genes, radiation affected preferentially the expression of genes implicated in RNA processing and regu lation. Additionally, we identified a co-expression module containing multiple genes associated with translation initiation , rRNA and snoRNA processing, RNA localization, and ribonucleoprotein complex biogenesis. Many regulators of RNA processing are implicated in glioblastoma development, and splicing alterations affect all hallmarks of cancer [242]. Radiation-induced changes in the splicing patterns of oncogenic factors and tumor suppressors such as CDHll, CHNl, CIC, EIF4A2, FGFRl, HN RNPA2B1 , MDM2, NCOAl, NUMAl, RPL22, SRSF3, TPM3, APC, CBLB, FAS, PTCHl, and SETD2. We also observed changes in expression of four RNA processing regulators previously identified in genomic/ functional screening for RNA binding proteins contributing to glioblastoma phenotypes: MAGOH , PPIH , ALYREF, and SNRPE [243]. 5.4.3 Potential new targets to mcrease radio-sensitivity and prevent relapse Activation of oncogenic signals is an undesirable effect of radiation that could influence treat ment response and contribute to relapse. We observed increased expression or translation and splicing alterations of a number of pro-oncogenic factors, genes whose high expression is associ ated with poor survival and genes previously implicated in radio-resistance. Among genes with the most marked increase in expression upon radiation, we identified members of the Notch pathway (HES2, NOTCH 3, MFNG, and JAG2). Notch activation has been 142 linked to radio-resistance in glioblastoma, and Notch targeting improves the results of radiation treatment [244, 245]. We also identified several genes associated with the Pl3K-Akt, Ras, and Rapl signaling pathways that increased expression levels upon radiation exposure. Targeting these pathways has been explored as a therapeutic option in glioblastoma [246, 244]. Other oncogenic factors relevant to glioblastoma that had increased expression after radiation exposure include SRC, MUCl, LMO2, PML, PDGFR,B, BCL3, and BCL6. Anti-apoptotic genes (BCL6, RRM2B, and IDOl) also showed increased expression upon radiation. BCL6 is a member of the ZBTB family of transcription factors, which functions as a p53 pathway repressor. The blockage of the interaction between BCL6 and its cofactors has been established as a novel therapeutic route to treat glioblastoma [247]. RRM2B is an enzyme essential for DNA synthesis and participates in DNA repair, cell cycle arrest, and mitochondrial homeostasis. The depletion of RRM2B resulted in ADR-induced apoptosis, growth inhibition , and enhanced sensitivity to chemo- and radiotherapy [248]. IDOl is a rate-limiting metabolic enzyme involved in tryptophan metabolism that is highly expressed in numerous tumor types [249]. The combination of radiation therapy and IDOl inhibition enhanced therapeutic response [250]. Among genes whose high expression correlates with decreased survival in glioblastoma, we identified several components of the "matrisome" and associated factors (FAM20C, SEMA3F, ADAMTSL4, ADAMTS14, SERPINA5, and CRELDl). The core of the " matrisome" con tains ECM proteins, while associated proteins include ECM-modifying enzymes and ECM-binding growth factors. This complex of proteins assembles and modifies extracellular matrices, contribut ing to cell survival, proliferation , differentiation, morphology, and migration [251]. In addition, several genes of the proteinase inhibitor SERPIN family (SERPINA3, SERPINA12, SERPINA5, 143 and SERPINll) implicated in ECM regulation [252] were among those with high levels of expres- sion upon radiation . In conclusion , our results generated a list of candidates for combination therapy. Contracting the effect of oncogenic factors and genes linked to poor survival could increase radio-sensitivity and treatment efficiency. Importantly, there are known inhibitors against several of these pro teins (Table S5) . Moreover, RNA processing and translation were determined to be important components of radio-response. These additional vulnerable points could be explored in therapy, as many inhibitors against components of the RNA processing and translation machinery have been identified [253, 254]. 144 Chapter 6 Ribopod: A database of uniformly processed Ribo-seq datasets 6.1 Introduction Even after 20 years of the Human Genome Project, there appears to be a no consensus on the number of protein coding genes [255]. The ENSEMBL [256] gene prediction process is based on alignments of protein and cDNA sequences, which is claimed to produce a low rate of false positives. The consensus coding sequence project (CCDS) [101] represents a consensus model which is defined as protein-coding regions that agree at the start codon , stop codon and splice junctions such that the prediction meets certain quality assurance benchmarks. These benchmarks involve finding consensus between CCDS regions and SWISS-PROT [257] proteins and ensuring CCDS regions satisfy genomic conservation criterion. Ribo-seq data provided exact information on regions that are being actively translated. The availability of public Ribo-seq datasets across different physiological and pathological contexts across different species has made it possible to learn new biology, particularly of mechanisms that are shared across the conditions. This has 145 motivated the development of Ribo-seq databases which provide access to processed Ribo-seq data. There have been multiple attempts at creating a database for Ribo-seq studies including GWIPS-Viz [258], RPFdb [259], and HRPDViewer [260]. GWIPS-Viz [258] provides a database of Ribo-seq coverages across multiple species. It provides global aggregates of analyzed public datasets as UCSC genome browser tracks. HRPDViewer [260] allows visualization of transcript level profiles for different datasets. RPF-db [259] provides a visualization interface for read count summaries across multiple r egions of the transcriptome as a summary statistic for quality. While largely useful, these databases do not systematically handle the intricacies of Ribo-seq data. Not all the fragments in Ribo-seq data represent active translation [85, 68] and hence any data downloaded from these datasets needs to be pre-processed to obtain actively translating fragments before performing any downstream analysis. We aim to bridge this gap by developing a new database, ribopod , that provides a visualization interface for assessing the quality scores of a public Ribo-seq dataset besides providing access to de-noised Ribo-seq profiles. Ribopod aims bridge the gap in leveraging the existing public datasets to understand translational regulation. 6.2 Ribopod: a database of de-noised Ribo-seq datasets We developed a database of uniformly processed public Ribo-seq data while paying attention to the intricacies of Ribo-seq data. We use our method ribotricer to filter out non-active fragments and pay attention to the library protocols, track and store cell/ tissue type, treatment and, other related metadata as obtained from NCBl's Sequence Reach Archive (SRA) or Gene Expression Omnibus (GEO). 146 6.2.1 Obtaining metadata The NCBI Sequence Read Archive (SRA) is the primary archive of next-generation sequencing datasets including Ribo-seq datasets. However, methods to programmatically access this data are limited . We needed a way to automatically extract all metadata of each Ribo-seq project to minimize manual curation. We developed pysradb package [93] that provided a simple and user friendly command-line interface for querying metadata and downloading datasets from . Metadata for all SRA projects in ribopod is obtained using pysradb. Retrieving metadata for any SRA ac- cession is done using: $ pysradb metadata SRP010679 --desc --expand 147 ... [truncated] run_accession cell - line sample_type source_name treatment SRR403882 pc3 polya rna pc3 human prostate cancer cells vehicle SRR403883 pc3 ribosome protected rna pc3 human prostate cancer cells vehicle SRR403884 pc3 polya rna pc3 human prostate cancer cells rapamycin SRR403885 pc3 ribosome protected rna pc3 human prostate cancer cells rapamycin SRR403886 pc3 polya rna pc3 human prostate cancer cells pp242 SRR403887 pc3 ribosome protected rna pc3 human prostate cancer cells pp242 SRR403888 pc3 polya rna pc3 human prostate cancer cells vehicle SRR403889 pc3 ribosome protected rna pc3 human prostate cancer cells vehicle SRR403890 pc3 polya rna pc3 human prostate cancer cells rapamycin SRR403891 pc3 ribosome protected rna pc3 human prostate cancer cells rapamycin SRR403892 pc3 polya rna pc3 human prostate cancer cells pp242 SRR403893 pc3 ribosome protected rna pc3 human prostate cancer cells pp242 Ribopod' provides a visualization interface for visualizing the metagene plots, fragment length distribution and the phase scores generated by ribotricer. Each sample in an SRA accession is described separately on all the three metrics (Figure 6.1 ). 6.2.2 Workflow After obtaining metadata and raw data from SRA, the raw sequences are mapped to the transcriptome using STAR [95] using a transcriptome annotation from ENSEMBL [256]. We 148 Read length distribution SRX021972-GSM546926 , Footprint mock 12hr runl (28 - 32nt e : 0.02 (pval : -9 . 28@ -• 01) )- SRX021974-GSM546928 : Footprint miR-155 12hr runl 128-32nt e : 0 . 86 (pval-;- 2 .87e-0ll) - _I .. Metagene distribution SRX021972 - GSM546926 : Footprint mock 12hr runl fram@ O/total : 0 . 48 (pval : 3 . 82@ -U) I coh@ -r@ nc@ --; 0 . 7-3 (pval : 7 . 72@-09) SRX021974-GSMS46928 : Footprint miR-155 12hr runl fr"meO/total : 0 . 47 (pval : l . 29e-12) I cohei-ence , -0 . 64-(pval : 7 . 72e-091 ' ·: .. .,1ro1~ ~ 1 " ,, l l , ,ut~ ~ " ''1 u 1 . ,u, 1 1 ~ 1 . , 1.,,. , 1 .~ ,, 1 , 1 1 , 1 1.u, 1 1 .~ ., t~:11. ,I , 1 , , 1 ., 1., 1 ~ 1.11i, 1 J, 1.,t1 M ,~ ·'' ~ . 1 1 .11. Figure 6.1 = Ribopod's visualization interface. The top panel shows a heatmap of ribotricer phase scores for each fragment length. The bottom left shows the fragment length distribution plots for each sample with the corresponding metagene plot on the right. See http://ribopod.usc.edu. allow for a maximum of two bases mismatches while mapping and only uniquely mapping reads are retained . Next, the uniquely mapped reads are processed through ribotricer to assess them for periodicity. Ribotricer outputs de-noised profiles retaining only the actively translating Ribo-seq reads (Figure 6.2). 6.3 Results Ribotricer's phase score [149] can be used as a quality control metric to asses the quality of Ribo-sseq data. We processed multiple Ribo-seq projects from different species including human, mouse. rat, zebrafish, fruitfly, C. elegans, 5. cerevisiae. and 5. pombe. A detailed list of datasets and the associated data is available on this link: https : //bit . ly /ribopod-datasets. The code used to analyze the datasets is available at https: / /gi thub. com/saketkc/re-ribo-smk and the code for the database is available at https: / /gi thub. com/saketkc/ribopod. 149 Sequence Read Archi.ve Sped.es FASTA Sped.es Ri.botri.cer Index e------+i ri.botri.cer ri.bopod Figure 6.2 Workflow of ribopod 0 ~~~ 0 ~:'lg~~sgiii~~§~ Distaneefromstartcodon _____ J111J We analyzed hundreds of libraries from different species and assessed their phase score as calculated using ribotricer. As seen in Figure 6.3, the phase score of Ribo-seq datasets are highly variable. They also appear to be dependent on the species. The species-specific effect have two likely sources or origin : 1) biological difference 2) technological difference. However, it is difficulty to quantify their relative contribution . One striking observation of this effect is in Drosophila datasets in which all the Ribo-seq libraries seem to have a low phase-score overall. This effect is not necessarily related to a biological difference arising from Drosophila samples, since the C. elegans and Zebrafish, the two nearby species seem to have higher phase scores. 6.4 Conclusion Ribopod provides a readily accessible database of processed Ribo-seq datasets that have been processed to filter only actively-translating fragments. Since each dataset is accompanied 150 500 u, 400 Cl) ·;: f! .0 :: 300 0 ... Cl) .0 E :i 200 100 0 C Cl) Ill Cl) E N :::J C Ill J: C. E :.c (.) Cl) Cl) - s::. ... :::J U) Ill U) .2! 0::: tr :::J Ii= U) Ill 0 Ill Ill CJ ::E ... Cl Ill .0 0 ::E Cl) N C Ill ai E c:i U) Cl) Cl) U) C Ill .0 'iii Ill " iii E C. Cl '> 0 0 Cl) ai 2! C. "O (/) :a 0 Cl) CJ Ill (/) ... < 'cii 0 CJ :::J CJ ... u.i .0 ..,: 0.7 0.6 0.5 2! 0 CJ : 0.4 U) Ill s::. o.. 0.3 0.2 0.1 Figure 6.3 .~ C Cl) Cl) Cl) - s::. ... U) Cl) Cl) U) 'cii Ill Cl) :::J U) Ill U) .2! C Ill .0 'iii 0::: CJ E N tr :::J Ii= U) Ill " iii E C. :::J :::J C Ill 0 Ill Ill Cl '> 0 0 ... Ill CJ ::E ... Cl Cl) .0 J: C. Ill .0 0 ai 2! C. "O ..,: E ::E Cl) (/) :a N C 0 Cl) :.c Ill CJ Ill ai (/) ... (.) E < c:i Phase score distribution for datasets present in ribopod. Phase scores were calculated over annotated protein-coding regions only. 0 CJ u.i 151 with an associated phase score, ribopod can be used to perform novel analysis on high quality datasets. The availability of de-noised Ribo-seq profiles across both protein-coding and upstream open reading frames (uORFs) can serve as the key resource for discovering the role of uORFs in different physiological and pathological contexts. 152 Chapter 7 Conservation of uORF mediated regulation across species 7 .1 Introduction The term expression in "gene expression " , a colloquial term in biology, refers to the process by which the gene synthesizes functional products, most often in the form of proteins. Gene expres sion is modulated by both transcriptional and post-transcriptional regulation. In all organisms, the synthesis of protein requires two key steps: transcription and translation . The expression itself is regulated at multiple steps spanning transcription , post-transcription , translation and post-translation [19, 20]. Upstream open reading frames (uORFs) are a major regulatory element located in the 5' leader sequences that play a role in regulating translation . uORFs are defined by a start codon and an in-frame stop codon and are located upstream of the main coding sequence in the 5' untranslated region (5' UTR) [261 , 262, 27]. uORFs have been suggested to conform to Kozak sequence rules [57]. Kozak sequence rules define a 'favorable' context required for initiating translation. However, this hypothesis in the context of uORFs has also been contested 153 [263]. uORFs have been hypothesized to play a role of down-regulating the expression of corre sponding CDS [264, 265, 266]. Around 50% transcripts in human, mouse, and Zebrafish have been hypothesized to have potential uORFs [267, 261 , 262, 29]. In eukaryotes, the process of translation may itself regulate gene expression independent of the encoded peptide [268]. The uORFs are unlikely to code for functional peptide products be cause their peptide sequences are not conserved, even if the presence in 5' region itself might be conserved [265]. But they play dual roles of i) allowing downstream re-initiation and ii) capturing some fraction of the pre-initiation complex thus reducing the translation of protein-coding region. A well studied example of the first role is demonstrated by the first uORF in activating transcrip tion factor 4 (ATF4) transcript that is constitutively translated, and then ribosomes re-initiate at either the second uORF or the CDS [79, 269]. Stress induces phosphorylation of the a subunit of eukaryotic translation initiation factor 2 (e1F2) . The phosphorylated elF2 inhibits the action of elF2B which in turn attenuates the formation of elF2-GTP-tRNA~et ternary complex. This prevents recycling of elF2 between different runs of the protein synthesis cycle and overall global translation initiation thus leading to an overall decrease in protein levels [270]. Paradoxically, in ATF4, the stress-induced phosphorylation of elF2 increases the translation which in turn activates the integrated stress response (a pro-survival pathway) . ATF4 has two conserved uORFs, uORFl and uORF2. In unstressed condition, lower levels of elF2a phosphorylation favor the ribosomes to re-initiate translation at uORF2. When phoshorylated , the formation of elF2-GTP-tRNA~et is sequestered which reduces the probability that the ribosomes terminating at uORFl will be able to acquire enough concentration of this complex to re-initiate translation at uORF2. However, they they may acquire them in the process of scanning the downstream AUG corresponding to the CDS, thus up-regulating the translation . For the second example, where uORFs are inhibitory to 154 the main CDS. uORF translation can thus affect the expression of downstream CDS in a global [266] or a local [271] setting. uORFs have been known to play a role in myriad of human diseases [272, 273, 274, 79, 275]. In particular, loss of function mutations in uORF of kinases is known to be associated with malignancies [276], uORFs serve as a source of tumor antigen [277] and in neurodegenerative diseases [278]. Preotemics based analyses have also linked uORFs to lower protein levels [279]. While much is known about the regulatory role of uORFs in the context of individual genes [273, 274], a global analysis on their translation status across different physiological and pathological contexts is currently missing. We utilize public Ribo-seq studies across eight species including human, mouse, rat, zebrafish, fruitfly, C. elegans, 5. cerevisiae, and 5. pombe across different physiological and pathological contexts to discover uORFs that are always under active translation and analyze their sequence context and its conservation across the species. 7.2 Results In order to decipher the regulatory role of uORFs and to asses if it is conserved across species we analyzed multiple public Ribo-seq datasets as made available through ribopod (Chapter 6) . Our results are organized in two parts. In the first part we establish the regulatory activity of uORFs across three species (human, chimp, and macaque) using matched Ribo-seq and RNA seq data. In the second part we analyze datasets from eight species including four vertebrates: human, mouse, rat, and zebrafish and four invertebrates: fruitfly, C. elegans, baker's yeast, and 5. pombe. to establish the conservation of uORFs across species. The datasets spanning these species have been performed in a variety of physiological and pathological contexts. The total 155 0.7 500 0.6 u, 400 ' Cl) 0.5 ·;: I'll ~ ... :e 0 u 1S 300 : 0.4 ... U) Cl) I'll .0 .c: E ll. 0.3 ;i 200 0.2 100 0.1 0 C Cl) Cl) Cl) - .c: ... U) Cl) Cl) U) " iii 0 C Cl) Cl) Cl) - .c: ... U) Cl) Cl) U) " iii 0 I'll Cl) ::I U) I'll U) .2! C I'll .0 " iii I'll Cl) ::I U) I'll U) .2! C I'll .0 "iii a: u a: u E N tr ::I i;:: U) I'll " iii E 0. ::I u E N tr ::I i;:: U) I'll " iii E 0. ::I u ::I C I'll 0 E I'll 0) " > 0 0 ... u.i ::I C I'll 0 I'll I'll 0) " > 0 0 ... u.i I'll u :E 0) Cl) .0 I'll u :E ... 0) Cl) .0 ::c 0. I'll .0 0 ai ~ 0. "C ..,: ::c 0. I'll .0 0 ai ~ 0. "C ..,: E :E Cl) en :a E :E Cl) en :a N C 0 Cl) N C 0 Cl) :E I'll u E :E I'll u I'll ai en ai en ... 0 E <C 0 E <C c:i c:i Figure 7.1 Distribution of phase scores and the number of libraries number of libraries and their corresponding ribotricer generated phase scores [149] are summarised in Figure 7.1. 7.2.1 uORFs have a repressive effect on translational efficiency To first characterize the regulatory role of uORFs, we analyzed Ribo-seq data available from Wang et al. [280] (SRA accession SRP062129) . The dataset consists of Ribo-seq performed in lymphoblastoids in human, macaque and chimpanzee. A matched RNA-seq study is available for this dataset from Khan et al. [281] (SRA accession SRP028612) which we utilize for calculating translational efficiency. Translational efficiency of a gene is reflective of the density of ribosomes attached to the mRNA for every transcript. We processed mapped Ribo-seq datasets through ribotricer and only retained actively-translating fragments. Translational efficiency was calculated 156 as the ratio of Ribo-seq to RNA-seq counts over a gene. uORFs were also searched using ribotricer (See Methods). To asses the regulatory impact of uORFs on translational efficiency, we compared the trans lational efficiency of all genes with different number of actively translating uORFs (0, 1, 2, 3, and above) and plotted its cumulative distribution across all the three species (Figure 7.2). Genes with higher number of uORFs tend to have reduced translational efficiency as compared to transcripts with lesser number of uORFs across human, macaque and chimpanzee. We tested using a Mann Whitney test. Genes with no actively translating uORFs have significantly higher translational efficiency as compared to the genes with 1 or more ORFs. Among the orthologous genes, human and chimpanzee have 180 translationally efficient genes (p < 0.05) while 475 genes are translationally efficient between human and macaque. These numbers are in line with our expectation based on phylogenetic relationship between the three species. We also wanted to assess if the uORF mediated regulatory activity is conserved across these species. To this end, we compared the distribution of uORF translational efficiency to coding domain sequence (CDS) translational efficiency between human and chimpanzee and between human and macaque using one to one orthologous genes. The uORF activity is correlated across species with the correlation between human and chimpanzee as 0.45 (Figure 7.3a) and between human and macaque as 0.33 (Figure 7.3b) , thus capturing the phylogenetic relationship between the two species. 157 0.75 IL go.so 0.25 (a) -5 0 5 10 CDS TE (log2) Human 0.75 IL go.so 0.25 (c) - 5 0 5 10 CDS TE (log2) Macaque - 0 ORFs (n=8393) - 0 ORFs (n=8122) - 1 ORFs (n=1537, p=1 .03e-14) 0.75 - 1 ORFs (n=988, p=7.80e--02) - 2 ORFs (n=710, p=1 .44e--01) IL - 2 ORFs (n=432, p=2.09e--02) - 3+ ORFs (n=910, p=1.28e-04)8 0.50 - 3+ ORFs (n=385, p=3.05e-01) 0.25 -5 0 5 10 CDS TE (log2) (b) Chimpanzee - 0 ORFs (n=8324) chimpanzee ~-- -- -- -- ... ----~ - 1 ORFs (n=886, p=2e--09) :ii - 2 ORFs (n=304, p=2.13e--01) -~ - 3+ ORFs (n=287, p=6.13e-02) 5l- macaque-t.,===:;:::==::::::;::~ 0 200 400 # TE genes (d) Number of TE orthologous genes in chimpanzee and macaque Figure 7.2 Higher number of actively translating uORFs result in down-regulation of the main ORF in (a) human (b) chimpanzee, and (c) macaque. (d) Number of differential translationally efficient genes in chimpanzee and macaque with respect to human. 7.2.2 uORFs are present across species We characterized the prevalence of uORFs, by searching for "complete" ORFs in the 5' UTRs. A "complete" ORF is defined by the presence of a start codon (AUG) with an in-frame stop codon (UAG, UAA, and UGA). There is also evidence that translation can be initiated at non-AUG start codons in the uORFs can initiate at non-AUG start codons [282, 283]. However, unless the Ribo-seq experiment was prepared with a translation initiation inhibitor, it is difficult to determine the translation initiation site using Ribo-seq data alone. Thus, we took the more conservative approach of using only AUG start codons. We defined 'dORFs' (downstream ORFs) to be analogous of uORFs that are located in the 3' UTR. Not much is known about the regulatory role of the dORFs. 158 L= - A - N N C) C) 0 7.5 r = 0.33 0 4 - • - p=Se-15 - • Cl) 0. ::::J 5.0 E 2 CT • • cu •• .c: (,) • (,) cu 2.5 - E C: 0 - cu C: E cu 0.0 ::::J E .c: -2 ::::J w .c: t- • • w -2.5 • r = 0.46 t- en • -4 • ti) C p = 2.4e-2 -5.0 0 C 0 -2.5 0.0 2.5 5.0 -5 0 5 uORF TE human/chimp (log2) uORF TE human/macaque (log2) (a) (b) Chimpanzee Macaque Figure 7.3 Correlation between translational efficiency of CDS and uORF in orthologous regions of human, chimpanzee and macaque indicate the phylogenetic distances. Correlation values are indicated in the legend We searched for start codons (AUG) in the 5' UTR and the first in-frame stop-codon to define the uORFs. Analogously, we searched for AUGs and the first in-frame stop codon in the 3' UTRs to define the dORFs. Since such uORFs and dORFs can overlap with coding domain sequences (CDS) , we categorized them into different categories based on their overlap status (Table 7.1). For the rest of our discussion, we use 'uORFs' to refer to 'super_uORFs' which do not overlap with any CDS of any of the possible isoforms. The relative distribution of dORFs is higher as compared to uORFs across species (Figure 7.4). Most genes have just one most upstream uORF or most downstream dORF present (Figure 7.5). The annotated CDS appear to be the longest regions, while the size distribution of uORFs and dORFs are similar and are around four times shorter than the annotated CDS (Figure 7.6). 159 Human uORF Mouse super_uORF Rat super_dORF Zebrafish overlap_ uORF Fruitfly overlap_ dORF C.elegans Yeast dORF S.pombe annotated 0 20000 40000 60000 80000 100000 Figure 7.4 Distribution of number of known and potential ORFs. ORFs were searched using ribotricer [149]. super_uORF and super_dORF describe uORFs and dORFs that do not overlap with any CDS in any of the isoforms. The description of each ORF is given in Table 7.1. ORF type annotated super_uORF super_dORF uORF dORF overlap_uORF overlap_dORF novel Description CDS annotated in the provided GTF file upstream ORF of the annotated CDS, not overlapping with any CDS of the same gene downstream ORF of the annotated CDS, not overlapping with any CDS of the same gene upstream ORF of the annotated CDS, not overlapping with the main CDS downstream ORF of the annotated CDS, not overlapping with the main CDS upstream ORF of the annotated CDS , overlapping with the main CDS downstream ORF of the annotated CDS, overlapping with the main CDS ORF in non-coding genes or in non-coding transcripts of coding genes Table 7.1 ORF types and their description. ORFs for each species were determined using ribotricer [149] 160 Zebrafish 39760) super_uORF uORF overlap_uORF overlap_ dORF dORF super_dORF 1 2 3 4 5 2 3 4 5 1 2 3 4 5 1 2 3 4 5 Number of regions Figure 7.5 Distribution of number of different ORF types per annotated CDS. For each annotated CDS number uORFs and dORFs were searched using ribotricer by looking for a start codon and an in-frame stop codon in the 5'UTR and 3'UTR regions respectively. 0.5 ~o 0 ·en 5 10 15 C 0) 0 0.5 0.5 0 0 5 10 15 5 0.5 0.25 0 5 10 15 5 C.ele ans ~ 1 0.5 10 15 1092 length Figure 7.6 0.5 10 0.5 - annotated - super_dORF 5 10 15 - overlap_dORF 7.5 10 12.5 - dORF super_uORF - overlap_uORF uORF Distribution of lengths of different ORF types. Kernel density estimates of the length of each ORF type on log 2 scale across species. 161 • - uORF - CDS 0.7 - dORF 0.6 • • 0.5 ~ : 0 (.) ~0.4 , • • "' ct1 ' .c Q. 0.3 0.2 • 0.1 • 0 C Q) 16 .c >, "' 1ii Q) "' .ai 0 ct1 "' a::: "' 5 C ct1 .0 ·oo (.) E ::::, q: ·2 ct1 Q) E a. 2 (.) 0 ~ Ol u.i ::::, ~ Q) >- 0 0 .0 I .0 u. ai a. -0 i-: Q) cti :0 N c_j ~ Figure 7.7 Distribution of phase scores in uORF, CDS and dORF. Phase scores were calculated using ribotricer across all the datasets available on bit. ly/ribopod-datasets. 7 .2.3 More uORFs than dORFs show active translation We calculated the phase score of each ORF generated through ribotricer [149]. The phase score is a reflective of the active-translation status of an ORF being translated based on the periodicity in its Ribo-seq profile. A higher phase score indicates active-translation . Across all the species, CDS has the highest phase scores followed by uORFs and dORFs (Figure 7.7). Though the abundance of dORFs is higher as compared to uORFs (Figure 7.4), the uORFs seem to be translated more often. There is a modest positive correlation (r = 0.33, p < l e - 6) between phase scores of CDS and uORFs implying a uORF translation does not necessarily shutdown the translation of the downstream CDS and a modest negative correlation (r = -0.26, p < le - 6) between phase scores of CDS and dORFs (Figures 7.10 and 7.10). 162 Human I Mouse I Rat I Zebrafish I Fruitfly I C.elegans I dORF uORF Yeast S.pombe Arabidopsis T.brucei- E.coli- 0 I I 20 I I I I I 40 60 80 100 Percentage of protein-coding genes Figure 7.8 Distribution of percentage of uORFs and dORFs under active translation with respected to coding genes. Proportion of uORFs and dORFs with respect to annotated CDS regions that are almost always translating. 163 Human Mouse Rat Zebrafish Fruitfly C.elegans Yeast S.pombe Arabidopsis T.brucei E.coli 0 20 40 60 Percent of genes Figure 7.9 80 100 dORF CDS uORF Distribution of uORFs and dORFs that are always translating with respect to coding genes. Percentages of uORFs and dORFs that are always translating as percentage of protein coding genes. 164 0.4 0.35 0.3 [0.25 0 2- ~ 8 : 0.2 (/) ro .c 0. C ro (I) ~ 0.15 0.1 0.05 • • • • Human Mouse T.brucei Arabidopsis • • • • C.elegans Zebrafish Macaque • •• • • • • Chimpanzee S.pombe Rat • •• ' .. ·~ . • ••• • • •• • • •• • • • , ... • : •• • • • ••• . ~· • • • • ••• • • . , .. I • • • • • • S.cerevisiae • D.melanogaster • E.coli [ r=0.36 (p=9.63e-40)] • • <• • • ·~· 0 ... -___,,• ___,, •---,.-· ...-... • ---,.-...-...-...-...--.-...................................................... --.---.---.----------.......................................... ..... 0 0.08 0.16 0.24 0.32 0.4 0.48 0.56 0.64 0.72 Mean phase score (CDS) Figure 7.10 Correlation between uORF and CDS phase scores. 165 0.4 0.35 0.3 U:- a:: ~0.25 ~ 8 (/) Q) ~ 0.2 a. C "' Q) :::!: 0.15 0.1 0.05 0 • • • • --- 0 Human Mouse T.brucei Arabidopsis • --- 0.08 0.16 • • ., • • C.elegans • Chimpanzee • Zebrafish • S.pombe • Macaque • Rat •• • • • • • • • • • • • • • 0.24 0.32 0.4 0.48 Mean phase score (CDS) Figure 7.11 0.56 • S.cerevisiae • D.melanogaster • E.coli 1 r--0.26 (p=9.46e-22) 1 • ,. . , . • e • • 0.64 0.72 Correlation between dORF and CDS phase scores. 166 7.2.4 uuORFs - uORFs that are almost always translating Having characterized the translational status of uORFs across species in different physiological and pathological contexts we focused on uORFs that are 'almost always' under active translation . The definition of 'almost always' here is relative to the protein-coding regions. Because of the inherent heterogeneity in the Ribo-seq data itself and given that the datasets are chosen across multiple physiological and pathological contexts, even the high-confidence protein coding regions are not translating always. First, following [149]. we learned the species specific for all the datasets cutoff for determining if an ORF is actively-translating. We determined ribotricer generated phase score for each annotated CDS and designated the median of this score calculated across all CDS spanning all datasets as the species-specific cutoff (Table 7.2) . Most of the CDS regions are actively translating in do not exhibit translation in around 60 - 70% of the CDS regions (Figure 7.12). The proportion of samples in which all the pro tein coding regions are actively-translating, i.e. they meet the minimum phase score cutoff for the corresponding species follows a bimodal distribution (Figure 7.12 and 7.13). We use the proportion corresponding to these peaks to classify any ORF as 'always' translating and 'never' translating. In particular, we use a gaussian mixture model to resolve the mean and variance parameters of the two mixtures for each species. The higher mean gaussian (N(µ 1 , 0' 1 )) will be used to define the 'always' translating and the lower the mean gaussian (N(µ 0 , 0' 0 )) will be used to define the 'never' translating regions. We focus on the 100(1 - a)%, where O < a < 1 quantile of X 1 ~ N(µ 1 , 0' 1 ) and let z 1 be such that P(X ::; z 1 ) = 1 - a or equivalently, P(X ~ z 1 ) = a . This defines our universal set - this value of the proportion determines the universality of an ORF. We now call an uORF to be universal as long as the proportion 167 assembly mean median std . dev Human 0.355 0.348 0.299 Mouse 0.286 0.255 0.295 Rat 0.327 0.379 0.282 Zebrafish 0.323 0.272 0.335 Drosophila 0.190 0.117 0.255 C. elegans 0.364 0.336 0.270 5. cerevisiae 0.411 0.436 0.225 5. pombe 0.467 0.467 0.149 Table 7.2 Species-specific cutoffs for active translation. The mean median and standard deviation of ribotricer generated phase score as calculated using only protein coding regions for all datasets used in this study. Any ORF in a species is labelled as actively-translating only if its phase score exceeds the median value for the corresponding sites. of samples which meet the threshold is higher than z 1 . This is the most conservative way of defining them, given that this represents the extreme tails of the distribution of CDS which are believed to be actively translating with a higher probability than uORFs. Similarly, for defining the 'never' translating regions, we rely on the middle most values of the gaussian with lower mean X 0 ~ J..f (µ 0 , a- 0 ) ensuring we sample similar proportion . In particular, we look for a value of d, such that P(µ 0 - d ::; X 0 ::; µ 0 + d) = o:. We then sample this region from the uuORFs distribution to define 'never' translating regions. We use o: = 0.95 for this study. 7.2.5 Conservation of uuORFs across species To identify uuORFs that are conserved across any two species, we translated the uORFs to their corresponding amino acid sequences. Considering one species as the reference, we used BLAST [284] to search for conserved uuORFs. In particular, the references species' uuORFs are treated as the target database for blast and then the amino acid sequence of each species is used 168 2 ~ 0 II) C: 0 Cl) C 1 Human 0.5 2 1 0-.,...........,....~,.....,.....,....__,.--t 1 0 2 1 0.5 1 4 2 0.5 Yeast 2 1 0 1 2 1 Zebrafish 0 0.5 1 0-.,...........,....~,.....,.....,....,.......,.~ o-.,...........,....~,.....,.. ............... -1 o-+-.,......;;=.-~,....,... ....... ~~ 0--.,............,....-.-,....,...--,-.,......._ 0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1 Proporption above threshold Figure 7.12 Distribution of phase scores of CDS regions that are above the minimum score required for active translation. Kernel density estimates of the phase scores of CDS regions that are above the threshold phase score required for the ORF of the corresponding species to be labelled as active translation. 0.0 Human 0.2 0.4 0.6 0.8 Proportion of samples meeting threshold Figure 7.13 Determining thresholds for 'always active' translation. 169 § 0.5 :.:; ro > m 0.4 Cl) C 0 ~ 0.3 a::: 0 ~ 0.2 (I) 0) ro ::R 0.1 0 0 - .I� 200 400 600 Divergence time (million years) Figure 7.14 800 Percentage of conserved uuORFs. Human is used as the reference sequence. A uuORF is said to be conserved if there is a unique hit for the corresponding amino acid sequence in the database of human uuORFs. as the query sequence. We restrict our search to unique hits ( -max_ target_seqs 1 , -evalue ie-3) to define 'orthologous' uuORFs. Around 45% of uuORFs are conserved between human and mouse, 40% are conserved between human and rat, 30% is conserved between zebrafish and human and 13% is conserved between human and fruitfly (Figure 7.14). Next, we hypothesized that the upstream sequence context of these uuORFs might give them a preferential context for translation initiation. The Kozak consensus sequence is a nucleic acid sequence that is present upstream of the start codon [55]. It is considered as the optimum sequence required to be present upstream to initiate translation in vertebrates. In prokaryotes, a similar sequence is present around eight bases upstream referred to as the Shine-Dalgarno sequence [285]. We hypothesized that the uuORFs might have a similar sequence context. Chew et al. [286] indeed found evidence for a Kozak like sequence context responsible for increased effect on the translational efficiency of the main CDS. 170 Human G CCG CCA CCA TGG Kozak 0.12 0.08 0.04 A ~ c Gc:A c:;;; C -.. 0 = = == -==~ _ ¢ T ~ T 0.12 0.08 Rat M • ~ + + + Always translating Never translating 0.12 0.08 0.04 0.15 Always translating 0.1 Mouse Always translating G CCG CCA CCA TGG Kozak A G Zebrafish Never translating N ~ ~ ~ ~ ~ ~ ~ 0 +++++++++ Always translating ca; c -.. c:::: -~-- ~ --- T ~ ¢ G CCG CCA CCA TGG Kozak 0.15 Never translating 0.1 Figure 7.15 Upstream sequence context of CDS in vertebrates. 171 0.08 0.08 0.6 0.4 0.2 C.elegans Always translating Never translating Yeast Always translating G CCG CCA CCA TGG Kozak Never translating = ==-= = --=== ~ ~ 0 + + + Drosophila 0.24 Always translating o .,6 A 0.08 GA ==-=.;;;;..-=-= -=~~~ G CCG CCA CCA TGG Kozak 0.24 0.16 A Never translating 0.08 G A --- c::::cc_ ="= ==- ~ = T ~ = --::'=:".==.- ~ = - -== - --- S. pombe 0.75 ~ 0.5 0.25 A ....., ~ - ~ c::: G c ~ I =-=-~:!:=~~------ N M ~ ~ ~ ~ ~ ~ 0 ++++++++ . Always translating G CCG CCA CCA TGG Kozak 0.75 Never translating 0.5 0.25 -'E'-- ~- -=---- - -- Figure 7.16 Upstream sequence context of CDS in invertebrates. 172 Human 0.08 G G CCG CCA CCA TGG Kozak 0.08 0.04 Rat 0.12 0.08 0.04 0.12 0.08 0.04 Always translating Never translating Always translating Never translating "' .,., <O r,.. + + + + + + + Mouse G CCG CCA CCA TGG Kozak 0.09 T A 0.06 A GGA 0.03 -r - l T A CC ~ - -r __ G C T G o ~~:::~ c: ~ C T Zebrafish G CCG CCA CCA TGG Kozak 0.18 0 ·" T AA Always translating Never translating Always translating Never translating 0.06 A. T A A CGT C -r-..-. y :J:.-.-rG T ~~~£.~~~AcG ~ = -------- ----- --------= Figure 7.17 Sequence similarity of upstream sequence contexts of always and never translating uORFs in vertebrates. Clustering was performed using pearson correlation coefficient on the position frequency matrix [287]. 173 C. e/egans 0.24 Always translating 0.16 0.08 G CCG CCA CCA TGG Kozak : ! J - ...... ~ --~ ¢ G ~ C Never translating ='l.= ~ ~ = = = A -.-G <( => " Yeast 0.45 0.3 0.15 0.45 0.3 0.15 - N M • ~ ~ ~ = ~ 0 +++++++++ . k tways translating "T T AGA G T c:: A T"TT -.-A ~ ~ <.ii C:c:C: ~ ,A,. «= Drosophila 0.24 Always translating 0.24 G CCG CCA CCA TGG Kozak "" A Ar 0.08 T A TAAA c:X- l:;"X~ CG¥ ~ c:.. ~ 3:~ ~ ~ G c:c: c:.. 0.3 0.15 0.3 0.15 " + S. pombe Never translating N M • ~ ~ ~ = ~ 0 ++++++++ . T Always translating C A A Ty A -.- T c:..,--.-c A ~ --~--=~ ::;:5 ~ ~ ~ Never translating Figure 7.18 Sequence similarity of upstream sequence contexts of always and never translating uORFs in invertebrates. Clustering was performed using pearson correlation coefficient on the position frequency matrix [287]. 174 We determined the 13 nucleotides sequence present upstream of the uuORFs across all species. Our null set is formed by the upstream sequence of those uORFs that are never translating. The sequence context of the never translating uORFs is different from that of the never translating uORFs indicating that the different sequence context might be responsible for initiating transla- tion in the uuORFs (Figures 7.17 and 7.18). Furthermore, the sequence contexts in vertebrates is different from the Kozak sequence context (Figure 7.17) while the sequence context in in vertebrates is A and T rich, distinct from both the Kozak and vertebrates' uuORF sequence context. Next, we investigated if the sequence contexts are similar across species. We make use of the pearson correlation coefficient [287] to quantify the similarity between two sequence motifs. Given two column vectors X and Y representing the frequencies of base frequencies of bases A, C, T and G, Pearson correlation coefficient (PCC) is defined as: ~ (X - X )(Y - Y) PCC(X , Y) = 6 aE A a a , ✓~aEA(Xa - X)(Ya - Y) - 1 ~ X=-IAI ~Xa, aE A A high PCC value between two sequence motifs reflects their similarity. In order to assess the similarity of the uuORFs motifs we performed hierarchical clusterintg using PCC as the similarity metric. The uuORFs are similar across vertebrates and across invertebrates (Figure 7.19 and 175 Figure 7.19 If) If) a, ::::i ·u ro a, ..... a. If) If) status low translating mid translating always translating Sequence similarity of upstream sequence contexts of always and never translating CDS. Clustering was performed using pearson correlation coefficient on the position frequency matrix [287]. 7.20). The uuORFs also show higher similar similarity to the sequence motifs upstream of CDS that are always translating both in vertebrates and in invertebrates (Figure 7.19 and 7.20). 7.3 Discussion uORFs have been known to play role in regulating translation initiation of the coding domain sequence. Most of the studies thus far have focused on potential uORFs, based on their sequence 176 Figure 7.20 (/) (/) a, ::::i ·u <ii a, ..... a. (/) (/) correlation 1 0.5 0 -0.5 species Human Mouse Rat Zebrafish Drosophila C.elegans status low translating mid translating always translating Sequence similarity of upstream sequence contexts of always and never translating CDS. Clustering was performed using pearson correlation coefficient on the position frequency matrix [287]. 177 Drosophila Zebrafish C. elegans Mouse Human Rat Human Rat Mouse S. pombe Yeast Yeast Drosophila S. pombe C. elegans Zebrafish ro (]) E a. >. -~ ~I 0 ..!!1 ..c Cl) Cl) C ..c ~ ro a. ro 0) 0 '- (]) Cl) ..0 a, 0 (]) '- N 0 0 (]) C ro C ro (]) Cl) ro 0:::: ro 0:::: Cl) ::::i E E ::::i 0 0 ~ ::::i ::::i ~ I I Figure 7.21 (]) U) U) ..!!1 (]) ..0 ro ro ..0 E ~ ~ ..c E a. 0 0 0 a. Cl) a. 0 CJ) '- CJ) 0 Cl) ..c C Cl) ro ~ 0) ro (]) '- a, ..0 (]) 0 N animal vertebrate invertebrate orf_type uORF cos correlation 1 0.5 0 -0.5 -1 Sequence similarity of upstream sequence contexts of always and never translating CDS and uORFs. Clustering was performed using pearson correlation coefficient on the position frequency matrix [287]. The translating motifs are similar across CDS and uORFs in both vertebrates and invertebrates. 178 Species I cutoff I cutoff ( al-1 mean I sigma I mean I sigma ( al- (never) ways) (never) (never) (always) ways) Human 0.194- 0.698-1 0.214 0.159 0.604 0.073 0.234 Mouse 0.080- 0.680-1 0.092 0.098 0.550 0.101 0.104 Rat 0.111- 0.806-1 0.131 0.158 0.685 0.094 0.151 Zebrafish 0.178- 0.790-1 0.199 0.164 0.684 0.0826 0.219 Drosophila 0.009- 0.631-1 0.010 0.011 0.383 0.193 0.012 C. elegans 0.177- 0.769-1 0.195 0.138 0.637 0.103 0.212 Yeast 0.408- 0.564-1 0.441 0.264 0.501 0.05 0.474 5. pombe 0.205- 0.901-1 0.217 0.094 0.738 0.127 0.229 Table 7.3 Thresholds for defining universal uORFs The cutoffs are learned as the value corresponding to 95 percentile Species Status Motif Human Always translating CGGGGCCTGA ATG GTGGCGGCCG Human Never translating ATAATAATAA ATG GCGATGGCGG Mouse Always translating GGGCGGCGGG ATG GCGGGGGCCG Mouse Never translating ATAATAATAA ATG AATATTTTTA Rat Always translating GGCCGGCTGA ATG GCGGCCCCCC Rat Never translating ATGATGATAA ATG GGTGTGATCG Zebrafish Always translating GCTCCGGAGA ATG CGCTGTATTA Zebrafish Never translating ATAATAATAA ATG AATATTATTA Drosophila Always translating CTCACAATAA ATG TGTACAACTA Drosophila Never translating ATAATAATAA ATG AAAAAAAATA C. elegans Always translating TCAACGGACA ATG TGCCTGAAAA C. elegans Never translating ATAATAATAA ATG TTAATTTTTA Yeast Always translating TATAGTTTAA ATG GTTAAAAAGA Yeast Never translating AGAATAAGAA ATG GAATTTACCG S. pombe Always translating TTAATATTAA ATG TCGAAGATTT S. pombe Never translating ATAATAATAA ATG ATTATAATTA Table 7.4 Sequence context of uuORFs Start codon is shighlighed in bold. 179 Species Status Motif Human Always translating CGCCGCCACC ATG GCGGAGGTGG Human Never translating CGCCGCCACC ATG GCGCTCGTGG Mouse Always translating GGCGGCCAAC ATG GCGGCGGTGG Mouse Never translating AATTAAAAAC ATG GAGAACATGA Rat Always translating CGCGGCCACC ATG GCGGCGGCGG Rat Never translating CATCGACAAC ATG GAGATCATCA Zebrafish Always translating AGAAGTCAAC ATG GCGGAGGAGG Zebrafish Never translating TATAGTAAAG ATG GATATTATGA Drosophila Always translating AAACAACAAA ATG GCCAACAACG Drosophila Never translating AAAAATAAAA ATG ACTATCATTA C. elegans Always translating TATTGTAAAA ATG GCTGTCGTCG C. elegans Never translating TATTTTAAAA ATG ATTATTATTA Yeast Always translating AAAAAAAAAA ATG TCTAAAAAAA Yeast Never translating AAAAAAAAAA ATG AATAATAATA S. pombe Always translating ATTAATCAAA ATG GCTGATAATA S. pombe Never translating TTTTTTTAAA ATG GATAATAATA Table 7.5 Sequence context of CDS. Start codon is surrounded by underscores. context. We bridge this gap by providing a catalogue of uORFs with their translation status across different contexts using public Ribo-seq datasets across multiple species. We use this learned information to investigate the conservation of regulatory roles of uORFs across species which further led to us characterizing uuORFs - uORFs that are almost always translating across different physiological and pathological conditions. The upstream sequence context of uuORFs in vertebrates is different from the well char acterized Kozak sequence context [55] required for translation initiation in vertebrates. The Kozak sequence is given by GCCGCC(A/ G)CCAUGG and was initially assumed to be a univer- sal signal required for initiating translation. Later, however, it was found that the sequence context is species dependent. Amongst the characterized species, the sequence context is ACAACCAAAAUGGC for Drosophila , AAAAAAAAAAUGTC for Saccharomyces cerevisiae, and UAAAT(A/ C)AACAUG(A/ G)C for other invertebrates [54]. Our observations with respect to 180 the sequence context of CDS are consistent with previous findings (Figures 7.15 and 7.16 and Table 7.5). We find that the sequence contexts of uuORFs have a similar context as the CDS of the corresponding species and are conserved within vertebrates and invertebrates (Tables 7.4 and 7.5). Though the consensus sequence is different from the respective expected sequence, we believe the translation initiation at uORFs might have a different mechanism than the corre sponding CDS as has been observed previously [272]. The characterization of uuORFs and their associated upstream sequence context enables us to characterize the functional role of the short polypeptides. Our study provides a rich catalogue of these uuORFs which can be further used to investigate their functional implications by proteomics-based or other related approaches. 181 Chapter 8 Conclusions Translational control is an integral part of the chain of processes that are employed in the gene to regulate the expression level of protein . The development of assays such as Ribo-seq have provided us with a window to peek into the transcriptome of the cell which is actively engaged by the ribosome thus enabling us to decipher the various modes by which a gene is regulated at the translational level. Ribo-seq has been used to understand translational regulation in a myriad of physiological and pathological contexts across species. The availability of these datasets with different contexts has presented us with avenues of discovering new biological phenomena that are conserved across different species. The aim of my research was to understand the regulatory role of short open reading frames located in the 5' UTR called upstream open reading frames (uORFs) . uORFs have been known to play a repressive role in translation . The previous research in this field is based on the assumption that a mere presence of an ORF characterized by an in-frame start and stop-codon, can be called an uORF. Ribo-seq provides actual evidence if such a potential ORF is actually an uORF, since it allows to evaluate its translation status. The length distribution of uORFs is similar to that 182 of coding exons and being a region of low signal to noise ratio, detecting active translation in uORFs using Ribo-seq data has remained challenging. In Chapter 3, I developed a novel method , ribotricer, for identifying active translating in both short and long open reading frames using Ribo-seq data. Our method takes an approach of evaluating the inherent periodicity in the Ribo-seq data through the 'high-low-low' pattern exhibited by an actively-translating ribosome as it traverses the mRNA. This approach overcomes the two key problems inherently present in almost all Ribo-seq datasets: uneven coverage and sparsity. While uneven coverage arises because the ribosome moves non-uniformly over the codons and hence is more of biological nature in its origin , sparsity is more of a technical limitation in a lot of experiments. Our approach of transforming the counts information to asses the consistency of qualitative 'high-low-low' pattern gives the highest accuracy on multiple datasets across different species. In Chapter 4, I use our method , ribotricer, to learn the changes in the translational landscape of a fungal pathogen Candida albicans during its morphological transition from yeast-like to filamentous growth. Using deep sequenced Ribo-seq and RNA-seq samples, I first re-learned the transcriptome of C. albicans since the available annotation is incomplete. Through this process, I discovered unannotated exons that are also engaged actively by the ribosome implying that these could be genes that were unannotated. I also discover hundreds of genes whose translational efficiency gets upregulated during the morphological transition . These genes could serve as potentially good targets anti-fungal drugs. In Chapter 5, by an integrated analysis in two glioblastoma cell lines, I mapped the changes that occur in gene expression at three levels of transcription , alternative splicing , and translation in response to ionizing radiation by utilizing Ribo-seq and RNA-seq data. Glioblastoma is the 183 most common intracranial malignant brain tumor with an aggressive clinical course. High-dose radiation is the main component of glioblastoma therapy. By characterizing the alterations at all three levels, I identified new biological processes that lead to altered expression of various oncogenic factors. I also suggested new target options that can increase radiation sensitivity and prevent relapse. I reanalyzed public Ribo-seq datasets from multiple species to distinguish ORFs that are under active translation from those that are not, using our tool ribotricer. In Chapter 6, I introduce a database of the reanalyzed public Ribo-seq projects that provides users direct access to de-noised Ribo-seq datasets. Finally, in Chapter 7, I utilized the resource developed in Chapter 6 to study upstream open reading frame (uORF) mediated regulation across eight species. I first establish that uORFs play a repressive role on the translational efficiency of the downstream protein coding regions. I characterized uuORFs - universal uORFs that are 'almost always' translating within a species given the diverse physiological and pathological contexts of the public datasets that were used to characterize them . Further more, I characterize that the upstream sequence context of these uORFs is similar to the upstream sequence context of the coding domain sequences within the species. The uuORFs are conserved across species with total conservation inversely proportional to the divergence time. The sequence context is conserved within the vertebrates and invertebrates. The characterization of uuORFs is the first step towards deciphering their functional role. The short polypetides synthesized from these uuORFs could also be potential regulators of house keeping operations in the regulatory system. Thus, one focus of future studies should be on characterizing their functional role, given that they are almost always present. 184 Bibliography [1] Y. Zarai, M. Margaliot, and T. Tuller, " Maximizing protein translation rate in the ribosome flow model: the homogeneous case," arxiv:1407.0207vl , 2014. [2] F. Supek, M. Bosnjak, N. Skunca, and T. Smuc, " REVIGO summarizes and visualizes long lists of gene ontology terms," PLoS ONE, vol. 6, no. 7, p. e21800, 2011 . [3] D. Szklarczyk, A. Franceschini, S. Wyder, K. Forslund, D. Heller, J. Huerta-Cepas, M. Si monovic, A. Roth, A. Santos, K. P. Tsafou, et al., "STRING vlO: protein-protein interac tion networks, integrated over the tree of life, " Nucleic Acids Research, vol. 43, no. Dl, pp. D447-D452, 2014. [4] R. L. Bowman, Q. Wang, A. Carro, R. G. Verhaak, and M. Squatrito, " GlioVis data portal for visualization and analysis of brain tumor expression datasets," Neuro-Oncology, vol. 19, no. 1, pp. 139-141, 2016. [5] R. Dahm , "Discovering DNA: Friedrich Miescher and the early years of nucleic acid re search, " Human genetics, vol. 122, no. 6, pp. 565-581, 2008. [6] H. B. Vickery, "The origin of the word Protein ," The Yale journal of biology and medicine, vol. 22, no. 5, p. 387, 1950. 185 [7] C. Tanford and J. Reynolds, Nature's Robots: A History of Proteins . OUP Oxford , 2003. [8] J. Brock and S. Davidson, Human Nutrition and Dietetics. 1973. [9] F. H. Crick, "On protein synthesis, " in Symp Soc Exp Biol, vol. 12, p. 8, 1958. [10] M. B. Hoagland, M. L. Stephenson, J. F. Scott, L. I. Hecht, and P. C. Zamecnik, "A soluble ribonucleic acid intermediate in protein synthesis, " Journal of Biological Chemistry, vol. 231, no. 1, pp. 241-257, 1958. [11] R. Kohler, "The Eighth Day of Creation: Makers of the Revolution in Biology by Horace Freeland Judson, " 1997. [12] T. Nakatani, T. Chen, J. Johnson, J. J. Westendorf, and N. C. Partridge, ''The deletion of hdac4 in mouse osteoblasts influences both catabolic and anabolic effects in bone, " Journal of Bone and Mineral Research, vol. 33, pp. 1362-1375, apr 2018. [13] G. Meisenberg and W. H. Simmons, Principles of Medical Biochemistry. Elsevier Health Sciences, 2016. [14] J. R. Warner, "The economics of ribosome biosynthesis in yeast, " Trends in Biochemical Sciences, vol. 24, pp. 437-440, Nov. 1999. [15] J. W. Hershey, N. Sonenberg, and M. B. Mathews, "Principles of translational control : an overview," Cold Spring Harbor Perspectives In Biology, vol. 4, no. 12, p. a011528, 2012. [16] G. C. Scheper, M. S. van der Knaap, and C. G. Proud, "Translation matters: protein synthesis defects in inherited disease," Nature Reviews Genetics, vol. 8, pp. 711-723, jul 2007. 186 [17] F. Jacob and J. Monod, " Genetic regulatory mechanisms in the synthesis of proteins," p. 39, 1961. [18] A. Mortazavi, B. A. Williams, K. McCue, L. Schaeffer, and B. Wold , "Mapping and quan tifying mammalian transcriptomes by rna-seq ," Nature methods, vol. 5, no. 7, p. 621 , 2008. [19] C. Vogel and E. M . Marcotte, "Insights into the regulation of protein abundance from proteomic and transcriptomic analyses, " Nature Reviews Genetics, vol. 13, pp. 227-232, Mar. 2012. [20] Y. Liu , A. Beyer, and R. Aebersold , "On the Dependency of Cellular Protein Levels on mRNA Abundance, " Ce//, vol. 165, no. 3, pp. 535-550, 2016. [21] F. Edfors, F. Danielsson , B. M . Hallstrom , L. Kall , E. Lundberg, F. Ponten , B. Forsstrom , and M . Uhlen, "Gene-specific correlation of RNA and protein levels in human cells and tissues," Molecular systems biology, vol. 12, no. 10, 2016. [22] T. Maier, M . Guell , and L. Serrano, "Correlation of mRNA and protein in complex biological samples, " FEBS Letters, vol. 583, pp. 3966-3973, Oct. 2009. [23] S. C. Schuster, "Next-generation sequencing transforms today's biology," Nature Methods, vol. 5, no. 1, pp. 16-18, 2008. [24] N. T. lngolia, S. Ghaemmaghami, J. R. Newman, and J. S. Weissman, "Genome-wide analysis in vivo of translation with nucleotide resolution using ribosome profiling," science, vol. 324, no. 5924, pp. 218-223, 2009. 187 [25] C. Gerard and A. Goldbeter, "Dynamics of the mammalian cell cycle in physiological and pathological conditions," Wiley Interdisciplinary Reviews: Systems Biology and Medicine, vol. 8, pp. 140-156, nov 2015. [26] S. Lee, B. Liu , S. Lee, S.-X. Huang, B. Shen, and S.- B. Qian, "Global mapping of transla- tion initiation sites in mammalian cells at single-nucleotide resolution ," Proceedings of the National Academy of Sciences, vol. 109, pp. E2424-E2432, aug 2012. [27] C. Fritsch , A. Herrmann, M. Nothnagel, K. Szafranski , K. Huse, F. Schumann , S. Schreiber, M. Platzer, M. Krawczak, J. Hampe, and M. Brosch, "Genome-wide search for novel human uORFs and n-terminal protein extensions using ribosomal footprinting, " Genome Research, vol. 22, pp. 2208-2218, aug 2012. [28] R. Shalgi , J. A. Hurt, I. Krykbaeva, M. Taipale, S. Lindquist, and C. B. Burge, "Widespread regulation of translation by elongation pausing in heat shock," Molecular Cell, vol. 49, pp. 439-452, feb 2013. [29] G.-L. Chew, A. Pauli, J. L. Rinn, A. Regev, A. F. Schier, and E. Valen, " Ribosome profil ing reveals resemblance between long non-coding RNAs and 5' leaders of coding RNAs, " Development, vol. 140, pp. 2828-2834, may 2013. [30] M. Guttman , P. Russell, N. T. lngolia, J. S. Weissman, and E. S. Lander, " Ribosome pro filing provides evidence that large noncoding RN As do not encode proteins," Cell, vol. 154, pp. 240-251, jul 2013. 188 [31] N. T. lngolia, G. A. Brar, N. Stern-Ginossar, M. S. Harris, G. J. Talhouarne, S. E. Jackson, M. R. Wills, and J. S. Weissman, " Ribosome profiling reveals pervasive translation outside of annotated protein-coding genes," Cell Reports, vol. 8, pp. 1365-1379, sep 2014. [32] A. A. Bazzini, T. G. Johnstone, R. Christiano, S. D. Mackowiak, B. Obermayer, E. S. Flem ing, C. E. Vejnar, M. T. Lee, N. Rajewsky, T. C. Walther, and A. J. Giraldez, " Identification of small ORFs in vertebrates using ribosome footprinting and evolutionary conservation .," EMBO J. , vol. 33, pp. 981-93, May 2014. [33] A. P. Fields, E. H. Rodriguez, M. Jovanovic, N. Stern-Ginossar, B. J. Haas, P. Mertins, R. Raychowdhury, N. Hacohen, S. A. Carr, N. T. lngolia, A. Regev, and J. S. Weissman, "A regression-based analysis of ribosome-profiling data reveals a conserved complexity to mammalian translation, " Molecular Cell, vol. 60, pp. 816-827, dee 2015. [34] Z. Ji, R. Song, A. Regev, and K. Struhl, " Many lncRNAs, 5'UTRs, and pseudogenes are translated and some are likely to express functional proteins," eLife, vol. 4, dee 2015. [35] L. Calviello, N. Mukherjee, E. Wyler, H. Zauber, A. Hirsekorn, M. Selbach , M. Landthaler, B. Obermayer, and U. Ohler, " Detecting actively translated open reading frames in ribo some profiling data," Nature Methods, vol. 13, pp. 165-170, dee 2015. [36] B. Malone, I. Atanassov, F. Aeschimann , X. Li , H. GroBhans, and C. Dieterich , "Bayesian prediction of RNA translation from ribosome profiling," Nucleic Acids Research, vol. 45, no. 6, pp. 2960-2972, 2017. [37] C. Barbosa , I. Peixeiro, and L. Romao, "Gene expression regulation by upstream open reading frames and human disease, " PLoS Genetics, vol. 9, p. e1003529, Aug. 2013. 189 [38] T. Dobzhansky, "Nothing in Biology Makes Sense except in the Light of Evolution," The American Biology Teacher, vol. 35, pp. 125-129, Mar. 1973. [39] G.-W. Li, D. Burkhardt, C. Gross, and J. S. Weissman, "Quantifying absolute protein synthesis rates reveals principles underlying allocation of cellular resources," Ce//, vol. 157, no. 3, pp. 624-635, 2014. [40] H. Tani, R. Mizutani, K. A. Salam, K. Tano, K. ljiri, A. Wakamatsu , T. lsogai, Y. Suzuki , and N. Akimitsu , "Genome-wide determination of RNA stability reveals hundreds of short lived noncoding transcripts in mammals, " Genome Research, vol. 22, no. 5, pp. 947-956, 2012. [41] L. Y. Chan, C. F. Mugler, S. Heinrich, P. Vallotton, and K. Weis, " Non-invasive mea surement of mRNA decay reveals translation initiation as the major determinant of mRNA stability," eLife, vol. 7, p. e32536, 2018. [42] G. A. Brar, " Beyond the triplet code: context cues transform translation," Cell, vol. 167, no. 7, pp. 1681-1692, 2016. [43] A. S. Zhokhin and V. P. Gachok, "Transition to chaos in the kinetic model of cellulose hydrolysis under enzyme biosynthesis control," arxiv:1707.08914vl, 2017. [44] L. Ribas de Pouplana, M. A. Santos, J.-H . Zhu, P. J. Farabaugh, and B. Javid, " Protein mistranslation: Friend or foe?," Trends in Biochemical Sciences, vol. 39, pp. 355-362, Aug 2014. 190 [45] C. Zhang, S. Zhou , E. Groppelli, P. Pellegrino, I. Williams, P. Borrow, B. M. Chain , and C. Jolly, " Hybrid spreading mechanisms and t cell activation shape the dynamics of HIV-1 infection," PLOS Computational Biology, vol. 11, p. el 004179, apr 2015. [46] R. J. Jackson, C. U. Hellen, and T. V. Pestova, "The mechanism of eukaryotic translation initiation and principles of its regulation, " Nature Reviews Molecular Cell Biology, vol. 11, no. 2, pp. 113-127, 2010. [47] M. Kozak, "An analysis of vertebrate mRNA sequences: intimations of translational con trol ," The Journal of Cell Biology, vol. 115, pp. 887-903, nov 1991. [48] J. KIEFT, A. GRECH , P. ADAMS, and J. DOUDNA, "Mechanisms of internal ribosome entry in translation initiation, " Cold Spring Harbor Symposia on Quantitative Biology, vol. 66, pp. 277-284, jan 2001. [49] P. Shah, Y. Ding, M. Niemczyk, G. Kudla, and J. B. Plotkin , " Rate-limiting steps in yeast protein translation, " Cell, vol. 153, no. 7, pp. 1589-1601, 2013. [50] M. V. Rodnina, ''Translation in prokaryotes," Cold Spring Harbor Perspectives in Biology, vol. 10, no. 9, p. a032664, 2018. [51] P. M. Gentz, G. L. Blatch, and R. A. Dorrington, "Dimerization of the yeast eukaryotic translation initiation factor 5a requires hypusine and is RNA dependent, " FEBS Journal, vol. 276, pp. 695-706, dee 2008. [52] Y. Furuichi , "Discovery of m 7 G-cap in eukaryotic mRNAs," Proceedings of the Japan Academy, Series 8, vol. 91, no. 8, pp. 394-409, 2015. 00000. 191 [53] M. Kozak, "Point mutations define a sequence flanking the AUG initiator codon that modulates translation by eukaryotic ribosomes," Cell, vol. 44, pp. 283-292, jan 1986. [54] D. R. Cavener and S. C. Ray, "Eukaryotic start and stop translation sites," Nucleic Acids Research, vol. 19, no. 12, pp. 3185-3192, 1991. [55] M. Kozak, ''The scanning model for translation: an update., " The Journal of Cell Biology, vol. 108, pp. 229-241, Feb. 1989. [56] G. Hernandez, V. G. Osnaya, and X. Perez-Martfnez, " Conservation and Variability of the AUG Initiation Codon Context in Eukaryotes," Trends in Biochemical Sciences, vol. 44, pp. 1009-1021, Dec. 2019. [57] M. Jalsenius and K. Pedersen , "A systematic scan for 7-colourings of the grid," arxiv:0704.1625v3, 2007. [58] J. McManus, G. May, P. Spealman, and A. Shteyman, " Ribosome profiling reveals post transcriptional buffering of divergent gene expression in yeast, " arxiv:1312.1765vl, 2013. [59] S. L. Wolin and P. Walter, " Ribosome pausing and stacking during translation of a eukary otic mRNA.," The EMBO Journal, vol. 7, pp. 3559-3569, nov 1988. [60] C. J. Woolstenhulme, N. R. Guydosh, R. Green, and A. R. Buskirk, " High-precision analysis of translational pausing by ribosome profiling in bacteria lacking EFP," Ce// Reports, vol. 11, pp. 13-21, apr 2015. 192 [61] A. V. Pisarev, V. G. Kolupaeva , M . M. Yusupov, C. U. Hellen, and T. V. Pestova, " Ribo somal position and contacts of mRNA in eukaryotic translation initiation complexes, " The EMBO Journal, vol. 27, pp. 1609-1621, Jun 2008. 00171 . [62] Y. Arava, Y. Wang, J. D. Storey, C. L. Liu , P. 0 . Brown , and D. Herschlag, "Genome-wide analysis of mRNA translation profiles in Saccharomyces cerevisiae, " Proceedings of the National Academy of Sciences, vol. 100, pp. 3889-3894, Apr. 2003. 00657. [63] P. Sampath , D. K. Pritchard , L. Pabon , H. Reinecke, S. M. Schwartz, D. R. Morris, and C. E. Murry, "A Hierarchical Network Controls Protein Translation during Murine Embryonic Stem Cell Self- Renewal and Differentiation, " Cell Stem Cell, vol. 2, pp. 448-460, May 2008. 00216. [64] C. G. Artieri and H. B. Fraser, "Accounting for biases in riboprofiling data indicates a major role for praline in stalling translation, " Genome Research, vol. 24, pp. 2011-2021, oct 2014. [65] J. Li and Y. Zhang, ''Translation with frameshifting of ribosome along mrna transcript," arxiv:1502.02109vl, 2015. [66] J. A. Hussmann, S. Patchett, A. Johnson, S. Sawyer, and W . H. Press, "Understanding biases in ribosome profiling experiments reveals signatures of translation dynamics in yeast," PLOS Genetics, vol. 11, p. e1005732, dee 2015. [67] D. A. Santos, L. Shi, B. P. Tu, and J. S. Weissman, "Cycloheximide can distort mea su rem en ts of m RNA levels and translation efficiency," Nucleic Acids Research, vol. 4 7, pp. 4974-4985, mar 2019. 193 [68] J. Li and Y. Zhang, ''Translation with frameshifting of ribosome along mrna transcript," arxiv:1502.02109vl, 2015. [69] F. Mohammad, R. Green, and A. R. Buskirk, "A systematically-revised ribosome profiling method for bacteria reveals pauses at single-codon resolution," eLife, vol. 8, p. e42591 , 2019. [70] M . Chevallet, P. Lescuyer, H. Diemer, A. van Dorsselaer, E. Leize-Wagner, and T. Rabilloud , "Alterations of the mitochondrial proteome caused by the absence of mitochondrial DNA: A proteomic view, " ELECTROPHORESIS, vol. 27, pp. 1574-1583, apr 2006. [71] L. V. Bock, M . H. Kolar, and H. Grubmuller, "Molecular simulations of the ribosome and associated translation factors, " arxiv:1711.06067vl, 2017. [72] N. T. lngolia, L. F. Lareau, and J. S. Weissman, " Ribosome profiling of mouse embryonic stem cells reveals the complexity and dynamics of mammalian proteomes," Cell, vol. 147, no. 4, pp. 789-802, 2011 . [73] P. Zhang, D. He, Y. Xu, J. Hou, B.-F. Pan, Y. Wang, T. Liu, C. M . Davis, E. A. Ehli, L. Tan, et al., "Genome-wide identification and differential analysis of translational initiation," Nature communications, vol. 8, no. 1, pp. 1-14, 2017. [74] N. T. lngolia, L. F. Lareau, and J. S. Weissman, " Ribosome profiling of mouse embryonic stem cells reveals the complexity and dynamics of mammalian proteomes," Cell, vol. 147, pp. 789-802, nov 2011. 194 [75] V. Kasari , T. Margus, G. C. Atkinson , M . J. Johansson, and V. Hauryliuk, " Ribosome profiling analysis of eef3-depleted saccharomyces cerevisiae," Scientific reports, vol. 9, no. 1, pp. 1-10, 2019. [76] Y. Han, X. Gao, B. Liu, J. Wan, X. Zhang, and S.-B. Qian , "Ribosome profiling reveals sequence-independent post-initiation pausing as a signature of translation ," Cell research, vol. 24, no. 7, pp. 842-851 , 2014. [77] R. Jackson, L. Kroehling, A. Khitun, W . Bailis, A. Jarret, A. G. York, 0 . M. Khan, J. R. Brewer, M. H. Skadow, C. Duizer, C. C. D. Harman, L. Chang, P. Bielecki, A. G. Solis, H. R. Steach , S. Slavoff, and R. A. Flavell , "The translation of non-canonical open reading frames controls mucosa I immunity," Nature, vol. 564, pp. 434-438, dee 2018. [78] V. Olexiouk, W . V. Criekinge, and G. Menschaert, "An update on sORFs.org: a repository of small ORFs identified by ribosome profiling," Nucleic Acids Research, vol. 46, pp. D497- D502, nov 2017. [79] S. Bianchini, A. Lage-castellanos, and E. Altshuler, " Upstream contamination in water pouring," arxiv:1105.2585v1, 2011 . [80] Q. Li , J. K. Eng, and M. Stephens, "A likelihood-based scoring method for peptide identifi cation using mass spectrometry," The Annals of Applied Statistics, vol. 6, pp. 1775-1794, dee 2012. [81] J. Radianti , J. Dugdale, J. J. Gonzalez, and O.-C. Granmo, "Smartphone sensing platform for emergency management, " arxiv:1406.3848vl, 2014. 195 [82] C. Barbosa , I. Peixeiro, and L. Romao, "Gene expression regulation by upstream open reading frames and human disease," PLoS Genetics, vol. 9, p. e1003529, aug 2013. [83] J. Li and Y. Zhang, ''Translation with frameshifting of ribosome along mrna transcript, " arxiv:1502.02109v1 , 2015. [84] D. E. Andreev, P. B. O'Connor, C. Fahey, E. M . Kenny, I. M . Terenin, S. E. Dmitriev, P. Cormican, D. W . Morris, I. N. Shatsky, and P. V. Baranov, "Translation of 5' leaders is pervasive in genes resistant to elF2 repression ," Elife, vol. 4, p. e03971 , 2015. [85] A. A. Bazzini, T. G. Johnstone, R. Christiano, S. D. Mackowiak, B. Obermayer, E. S. Flem ing, C. E. Vejnar, M. T. Lee, N. Rajewsky, T. C. Walther, and A. J. Giraldez, "Identification of small ORFs in vertebrates using ribosome footprinting and evolutionary conservation," The EMBO Journal, vol. 33, pp. 981-993, apr 2014. [86] A. Raj, S. H. Wang, H. Shim , A. Harpak, Y. I. Li, B. Engelmann , M . Stephens, Y. Gilad, and J. K. Pritchard , "Thousands of novel translated open reading frames in humans inferred by ribosome footprint profiling, " eLife, vol. 5, may 2016. [87] Z. Xiao, R. Huang, X. Xing, Y. Chen, H. Deng, and X. Yang, " De novo annotation and characterization of the translatome with ribosome profiling data," Nucleic Acids Research, vol. 46, pp. e61-e61, mar 2018. [88] C. Cai, R. Chen, and M. ge Xie, " Individualized group learning, " arxiv:1906.05533v1, 2019. [89] S. Y. Chun, C. M. Rodriguez, P. K. Todd, and R. E. Mills, "SPECtre: a spectral coherence based classifier of actively translated transcripts from ribosome profiling sequence data," BMC Bioinformatics, vol. 17, nov 2016. 196 [90] C. A. Brackley, D. Broomhead , M . C. Romano, and M . Thiel , "A max-plus model of ribosome dynamics during mrna translation ," arxiv:1105.3580vl, 2011 . [91] J. A. Hussmann, S. Patchett, A. Johnson, S. Sawyer, and W . H. Press, " Understanding Biases in Ribosome Profiling Experiments Reveals Signatures of Translation Dynamics in Yeast, " PLoS Genetics, vol. 11, p. e1005732, Dec. 2015. 00050. [92] P. B. O'Connor, D. E. Andreev, and P. V. Baranov, " Comparative survey of the relative im pact of mRNA features on local ribosome profiling read density," Nature Communications, vol. 7, p. 12915, 2016. [93] S. Choudhary, " pysradb: A python package to query next-generation sequencing metadata and data from NCBI sequence read archive," mar 2019. [94] S. Ganguly, E. Mossel, and M . Z. Racz, "Sequence assembly from corrupted shotgun reads," arxiv:1601 .07086vl, 2016. [95] A. Dobin , C. A. Davis, F. Schlesinger, J. Drenkow, C. Zaleski , S. Jha, P. Batut, M. Chaisson , and T. R. Gingeras, "STAR: ultrafast universal RNA-seq aligner," Bioinformatics, vol. 29, pp. 15-21, oct 2012. [96] H. H. von Grunberg, M. Peifer, J. Timmer, and M. Kollmann , "Variations in substitution rate in human and mouse genomes," Physical Review Letters, vol. 93, nov 2004. [97] M. Ruffier, A. Kahari, M . Komorowska, S. Keenan, M . Laird, I. Longden, G. Proctor, S. Searle, D. Staines, K. Taylor, et al., " Ensembl core software resources: storage and programmatic access for DNA sequence and genome annotation," Database, vol. 2017, 2017. 197 [98] M. S. Skrzypek, J. Binkley, G. Binkley, S. R. Miyasato, M. Simison , and G. Sherlock, ''The Candida Genome Database (CGD): incorporation of Assembly 22, systematic identifiers and visualization of high throughput sequencing data," Nucleic Acids Research, p. gkw924, 2016. [99] J. R. V. Zandt, " Efficient cubature rules, " arxiv:1712.07309v3, 2017. [100] S. A. Eales, M. Baes, N. Bourne, M. Bremer, M. J. I. Brown, C. Clark, D. Clements, P. de Vis, S. Driver, L. Dunne, S. Dye, C. Furlanetto, B. Holwerda, R. J. lvison, L. S. Kelvin, M. Lara-Lopez, L. Leeuw, J. Loveday, S. Maddox, M. J. Michafowski, S. Phillipps, A. Robotham, D. Smith, M. Smith, E. Valiante, P. van der Werf, and A. Wright, ''The causes of the red sequence, the blue cloud , the green valley, and the green mountain," Monthly Notices of the Royal Astronomical Society, vol. 481 , pp. 1183-1194, aug 2018. [101] K. D. Pruitt, J. Harrow, R. A. Harte, C. Wallin , M. Diekhans, D. R. Maglott, S. Searle, C. M. Farrell , J. E. Loveland , B. J. Ruef, E. Hart, M.-M . Suner, M . J. Landrum , B. Aken , S. Ayling, R. Baertsch , J. Fernandez-Banet, J. L. Cherry, V. Curwen , M. DiCuccio, M. Kel lis, J. Lee, M. F. Lin, M. Schuster, A. Shkeda, C. Amid , G. Brown, 0 . Dukhanina, A. Frank ish, J. Hart, B. L. Maidak, J. Mudge, M. R. Murphy, T. Murphy, J. Rajan , B. Rajput, L. D. Riddick, C. Snow, C. Steward, D. Webb, J. A. Weber, L. Wilming, W . Wu, E. Bir ney, D. Haussler, T. Hubbard, J. Ostell, R. Durbin, and D. Lipman, "The consensus coding sequence (CCDS) project: Identifying a common protein-coding gene set for the human and mouse genomes, " Genome Research, vol. 19, pp. 1316-1323, jun 2009. 198 [102] D. Hinkley, "Bootstrap methods: Another look at the jackknife," in Springer Series in Statistics, pp. 178-206, Springer New York. [103] T. T. Ivancevic, M. J. Bottema, and L. C. Jain, "A theoretical model of chaotic attractor in tumor growth and metastasis," arxiv:0807.4272vl, 2008. [104] J. Noorbakhsh, A. H. Lang, and P. Mehta, "Intrinsic noise of microRNA-regulated genes and the ceRNA hypothesis, " PLoS ONE, vol. 8, p. e72676, aug 2013. [105] M. Mariotti, S. Shetty, L. Baird , S. Wu, G. Loughran, P. R. Copeland , J. F. Atkins, and M. T. Howard, "Multiple RNA structures affect translation initiation and UGA redefinition efficiency during synthesis of selenoprotein p, " Nucleic Acids Research, vol. 45, pp. 13004- 13015, oct 2017. [106] M. Castellana, S. H.-J . Li , and N. S. Wingreen , "Spatial organization of bacterial tran scription and translation, " Proceedings of the National Academy of Sciences, vol. 113, pp. 9286-9291 , aug 2016. [107] Y. Chua and C. Tan , " Neurogenesis and multiple plasticity mechanisms enhance associative memory retrieval in a spiking network model of the hippocampus, " arxiv:1704.07526vl, 2017. [108] S. Sengupta , X. Yang, and P. G. Higgs, "The mechanisms of codon reassignments in mitochondrial genetic codes, " arxiv:q-bio/ 0703066vl, 2007. [109] Y. T. Lin, P. G. Hufton, E. J. Lee, and D. A. Potoyan, "A stochastic and dynamical view of pluripotency in mouse embryonic stem cells," PLOS Computational Biology, vol. 14, p. e1006000, feb 2018. 199 [110] G. Xu , G. H. Greene, H. Yoo, L. Liu , J. Marques, J. Motley, and X. Dong, "Global trans lational reprogramming is a fundamental layer of immune regulation in plants.," Nature, vol. 545, pp. 487-490, 05 2017. [111] P. Juntawong, T. Girke, J. Bazin , and J. Bailey-Serres, "Translational dynamics revealed by genome-wide profiling of ribosome footprints in Arabidopsis., " Proc. Natl. Acad. Sci. U.S.A. , vol. 111, pp. E203-12, Jan 2014. [112] R. Lukoszek, P. Feist, and Z. lgnatova, "Insights into the adaptive response of Arabidopsis thaliana to prolonged thermal stress by ribosomal profiling and RNA-Seq .," BMC Plant Biol., vol. 16, p. 221 , 10 2016. [113] M.-J . Liu , S.-H . Wu , J.-F. Wu , W .-D. Lin, Y.-C. Wu , T.-Y. Tsai, H.-L. Tsai , and S. H. Wu , "Translational landscape of photomorphogenic Arabidopsis.," Plant Cell, vol. 25, pp. 3699-710, Oct 2013. [114] H. M . Blank, R. Perez, C. He, N. Maitra, R. Metz, J. Hill, Y. Lin , C. D. Johnson, V. A. Bankaitis, B. K. Kennedy, R. Aramayo, and M. Polymenis, "Translational control of li pogenic enzymes in the cell cycle of synchronous, growing yeast cells.," EMBO J. , vol. 36, pp. 487-502, 02 2017. [115] N. R. Guydosh and R. Green, " Dom34 rescues ribosomes in 3' untranslated regions.," Cell, vol. 156, pp. 950-62, Feb 2014. [116] C. G. Artieri and H. B. Fraser, " Evolution at two levels of gene expression in yeast.," Genome Res., vol. 24, pp. 411-21, Mar 2014. 200 [117] C. J. McManus, G. E. May, P. Spealman , and A. Shteyman , "Ribosome profiling reveals post-transcriptional buffering of divergent gene expression in yeast., " Genome Res., vol. 24, pp. 422-30, Mar 2014. [118] N. T. lngolia , S. Ghaemmaghami, J. R. S. Newman, and J. S. Weissman , "Genome-wide analysis in vivo of translation with nucleotide resolution using ribosome profiling. ," Science, vol. 324, pp. 218-23, Apr 2009. [119] D. D. Nedialkova and S. A. Leidel , "Optimization of Codon Translation Rates via tRNA Modifications Maintains Proteome Integrity., " Cell, vol. 161, pp. 1606-18, Jun 2015. [120] M. Stadler and A. Fire, " Conserved translatome remodeling in nematode species executing a shared developmental transition .," PLoS Genetics, vol. 9, no. 10, p. e1003739, 2013. [121] M . Stadler, K. Artiles, J. Pak, and A. Fire, "Contributions of mRNA abundance, ribo some loading, and post- or peri-translational effects to temporal repression of C. elegans heterochronic mi RNA targets ., " Genome Res., vol. 22, pp. 2418-26, Dec 2012. [122] M . Stadler and A. Fire, "Wobble base-pairing slows in vivo translation elongation in meta zoans.," RNA, vol.17, pp. 2063-73, Dec 2011. [123] X. Chen and D. Dickman , "Development of a tissue-specific ribosome profiling approach in Drosophila enables genome-wide evaluation of translational adaptations.," PLoS Genetics, vol. 13, p. e1007117, Dec 2017. [124] J. G. Dunn, C. K. Foo, N. G. Belletier, E. R. Gavis, and J. S. Weissman, " Ribosome profiling reveals pervasive and regulated stop codon readthrough in Drosophila melanogaster.," Elife, vol. 2, p. e01179, Dec 2013. 201 [125] S. W . Eichhorn , A. 0 . Subtelny, I. Kronja, J. C. Kwasnieski, T. L. Orr-Weaver, and D. P. Bartel , "mRNA poly(A)-tail changes specified by deadenylation broadly reshape translation in Drosophila oocytes and early embryos.," Elife, vol. 5, 07 2016. [126] J. L. Aspden, Y. C. Eyre-Walker, R. J. Phillips, U. Amin, M. A. S. Mumtaz, M. Brocard, and J.-P. Couso, "Extensive translation of small Open Reading Frames revealed by Poly Ribo-Seq .," Elife, vol. 3, p. e03528, Aug 2014. [127] D. E. Andreev, P. B. F. O'Connor, A. V. Zhdanov, R. I. Dmitriev, I. N. Shatsky, D. B. Papkovsky, and P. V. Baranov, "Oxygen and glucose deprivation induces widespread al terations in mRNA translation within 20 minutes.," Genome Biol., vol. 16, p. 90, May 2015. [128] A. Ori, B. H. Toyama, M. S. Harris, T. Bock, M. lskar, P. Bork, N. T. lngolia, M. W. Hetzer, and M. Beck, "Integrated Transcriptome and Proteome Analyses Reveal Organ Specific Proteome Deterioration in Old Rats.," Ce// Syst, vol. 1, pp. 224-37, Sep 2015. [129] S. Schafer, E. Adami , M . Heinig, K. E. C. Rodrigues, F. Kreuchwig, J. Silhavy, S. van Heesch, D. Simaite, N. Rajewsky, E. Cuppen, M. Pravenec, M. Vingron, S. A. Cook, and N. Hubner, "Translational regulation shapes the molecular landscape of complex disease phenotypes.," Nat Commun, vol. 6, p. 7200, May 2015. [130] A. A. Bazzini, M. T. Lee, and A. J. Giraldez, " Ribosome profiling shows that miR-430 reduces translation before causing mRNA decay in zebrafish .," Science, vol. 336, pp. 233- 7, Apr 2012. 202 [131] M. T. Lee, A. R. Bonneau , C. M . Takacs, A. A. Bazzini , K. R. DiVito, E. S. Fleming, and A. J. Giraldez, " Nanog, Pou5fl and SoxBl activate zygotic gene expression during the maternal-to-zygotic transition .," Nature, vol. 503, pp. 360-4, Nov 2013. [132] D. Muzzey, G. Sherlock, and J. S. Weissman , "Extensive and coordinated control of allele specific expression by both transcription and translation in Candida albicans., " Genome Res., vol. 24, pp. 963-73, Jun 2014. [133] N. R. Guydosh, P. Kimmig, P. Walter, and R. Green, "Regulated !rel-dependent mRNA decay requires no-go mRNA degradation to maintain endoplasmic reticulum homeostasis in S. pombe.," Elife, vol. 6, 09 2017. [134] S. H. Wang, C. J. Hsiao, Z. Khan , and J. K. Pritchard , "Post-translational buffering leads to convergent protein expression levels between primates.," Genome Biol., vol. 19, p. 83, 06 2018. [135] X. Robin , N. Turck, A. Hainard, N. Tiberti , F. Lisacek, J.-C. Sanchez, and M. Muller, " pROC: an open-source package for R and S+ to analyze and compare ROC curves, " BMC bioinformatics, vol. 12, no. 1, p. 77, 2011 . [136] A. M. Michel, S. J. Kiniry, P. B. F. O'Connor, J. P. Mullan, and P. V. Baranov, "GWIPS-viz: 2018 update," Nucleic acids research, vol. 46, no. D1, pp. D823-D830, 2017. [137] H. Wang, L. Yang, Y. Wang, L. Chen, H. Li, and Z. Xie, " Rpfdb v2. 0: an updated database for genome-wide information of translated mrna generated from ribosome profiling," Nucleic acids research, vol. 47, no. D1, pp. D230-D234, 2018. 203 [138] M. A. Pfaller and D. J. Diekema, " Epidemiology of Invasive Candidiasis: a Persistent Public Health Problem, " Clinical Microbiology Reviews, vol. 20, pp. 133-163, Jan. 2007. [139] M.A. Pfaller, D. R. Andes, D. J. Diekema, D. L. Horn, A. C. Reboli, C. Rotstein , B. Franks, and N. E. Azie, "Epidemiology and Outcomes of Invasive Candidiasis Due to Non-albicans Species of Candida in 2,496 Patients: Data from the Prospective Antifungal Therap y (PATH) Registry 2004-2008," PLoS ONE, vol. 9, p. el 01510, July 2014. [140] M. E. Falagas, N. Roussos, and K. Z. Vardakas, "Relative frequency of albicans and the various non-albicans Candida spp among candidemia isolates from inpatients in various parts of the world : a systematic review," International Journal of Infectious Diseases, vol. 14, no. 11, pp. e954-e966, 2010. [141] T. H. Koornwinder, "q-special functions, an overview, " arxiv:math/ 0511148vl, 2005. [142] A. Cufar, A. Mrhar, and M. Robnik-Sikonja, " Identifying roles of clinical pharmacy with survey evaluation ," arxiv:1406.4287vl, 2014. [143] R. Hirano, Y. Sakamoto, K. Kudo, and M. Ohnishi, " Retrospective analysis of mortality and Candida isolates of 75 patients with candidemia: a single hospital experience, " Infection and drug resistance, vol. 8, p. 199, 2015. [144] Z. Xiao, Q. Wang, F. Zhu, and Y. An, " Epidemiology, species distribution, antifungal susceptibility and mortality risk factors of candidemia among critically ill patients: a retro spective study from 2011 to 2017 in a teaching hospital in China," Antimicrobial Resistance & Infection Control, vol. 8, no. 1, p. 89, 2019. 204 [145] A. Boullis, F. Francis, and F. Verheggen , "Aphid-hoverfly interactions under elevated CO2 concentrations: oviposition and larval development, " Physiological Entomology, vol. 43, pp. 245-250, jun 2018. [146] D. Kadosh , "Control of Candida albicans morphology and pathogenicity by post transcriptional mechanisms, " Cellular and Molecular Life Sciences, vol. 73, pp. 4265-4278, Nov. 2016. [147] M. I. Love, W . Huber, and S. Anders, "Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2, " Genome Biology, vol. 15, no. 12, p. 550, 2014. [148] N. T. lngolia, S. Ghaemmaghami, J. R. S. Newman, and J. S. Weissman, "Genome-wide analysis in vivo of translation with nucleotide resolution using ribosome profiling," Science, vol. 324, pp. 218-223, apr 2009. [149] S. Choudhary, W. Li, and A. D. Smith , "Accurate detection of short and long active ORFs using Ribo-seq data," Bioinformatics, 2019. [150] W . Li , W . Wang, P. J. Uren, L. 0 . Penalva, and A. D. Smith , " Riborex: fast and flexible identification of differential translation from Ribo-seq data, " Bioinformatics, vol. 33, no. 11, pp. 1735-1737, 2017. [151] V. M . Bruno, Z. Wang, S. L. Marjani, G. M. Euskirchen, J. Martin, G. Sherlock, and M. Snyder, "Comprehensive annotation of the transcriptome of the human fungal pathogen candida albicans using RNA-seq ," Genome Research, vol. 20, pp. 1451-1458, sep 2010. [152] A. S. Hinrichs, D. Karolchik, R. Baertsch, G. P. Barber, G. Bejerano, H. Clawson , M. Diekhans, T. S. Furey, R. A. Harte, F. Hsu, et al., "The UCSC genome browser 205 database: update 2006," Nucleic acids research, vol. 34, no. suppl_l, pp. D590-D598, 2006. [153] M. Pertea, G. M. Pertea, C. M. Antonescu, T.-C. Chang, J. T. Mendell, and S. L. Salzberg, "String Tie enables improved reconstruction of a transcriptome from RNA-seq reads," Na ture Biotechnology, vol. 33, no. 3, p. 290, 2015. [154] I. H. Madshus, T. I. T¢nnessen, S. Olsnes, and K. Sandvig, "Effect of potassium depletion of hep 2 cells on intracellular pH and on chloride uptake by anion anti port," Journal of Cellular Physiology, vol. 131, pp. 6-13, apr 1987. [155] D. K. Okamoto, "Competition among eggs shifts to cooperation along a sperm supply gradient in an external fertilizer," arxiv:1510.08813v3, 2015. [156] H. Teppola, S. Okujeni, M. L. Linne, and U. Egert, "Ampa, nmda and gabaa receptor mediated network burst dynamics in cortical cultures in vitro," arxiv:1802.00217v1, 2018. [157] J. C. Darnell, S. J. V. Driesche, C. Zhang, K. Y. S. Hung, A. Mele, C. E. Fraser, E. F. Stone, C. Chen, J. J. Fak, S. W . Chi, D. D. Licatalosi, J. D. Richter, and R. B. Darnell, "FMRP stalls ribosomal translocation on mRNAs linked to synaptic function and autism, " Ce//, vol. 146, pp. 247-261, jul 2011 . [158] W . J. Faller, T. J. Jackson, J. R. P. Knight, R. A. Ridgway, T. Jamieson, S. A. Karim, C. Jones, S. Radulescu, D. J. Huels, K. B. Myant, K. M. Dudek, H. A. Casey, A. Scopelliti, J. B. Cordero, M. Vidal, M. Pende, A. G. Ryazanov, N. Sonenberg, 0 . Meyuhas, M. N. Hall, M. Bushell, A. E. Willis, and 0 . J. Sansom, "m TORCl-mediated translational elongation limits intestinal tumour initiation and growth," Nature, vol. 517, pp. 497-500, nov 2014. 206 [159] S. Andrews et al., "Fastqc: a quality control tool for high throughput sequence data," 2010. [160] F. Krueger, ''TrimGalore! A wrapper around Cutadapt and FastQC to consistently apply adapter and quality trimming to FastQ files," 2012. [161] G. Yu, L.-G. Wang, Y. Han, and Q.-Y. He, "clusterprofiler: an r package for comparing biological themes among gene clusters, " Omics: a journal of integrative biology, vol. 16, no. 5, pp. 284-287, 2012. [162] S. Zhang, H. Hu, J. Zhou, X. He, T. Jiang, and J. Zeng, "Analysis of Ribosome Stalling and Translation Elongation Dynamics by Deep Learning, " Cell Systems, vol. 5, pp. 212-220.e6, Sep 2017. 00006. [163] A. Savitzky and M. J. E. Golay, "Smoothing and differentiation of data by simplified least squares procedures. ," Analytical Chemistry, vol. 36, pp. 1627-1639, jul 1964. [164] E. Garcfa-Portugues, R. M. Crujeiras, and W . Gonzalez-Manteiga, "Smoothing-based tests with directional random variables," arxiv:1804.00230v1 , 2018. [165] Y. Benjamini and Y. Hochberg, "Controlling the false discovery rate: a practical and powerful approach to multiple testing," Journal of the Royal Statistical Society: Series B (Methodological), vol. 57, no. 1, pp. 289-300, 1995. [166] R. Stupp, W. P. Mason, M. J. Van Den Bent, M. Weller, B. Fisher, M. J. Taphoorn, K. Belanger, A. A. Brandes, C. Marosi, U. Bogdahn, et al., " Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma," New England Journal of Medicine, vol. 352, no. 10, pp. 987-996, 2005. 207 [167] M. R. Gilbert, J. J. Dignam , T. S. Armstrong, J. S. Wefel , D. T. Blumenthal , M. A. Vogelbaum, H. Colman , A. Chakravarti, S. Pugh , M. Won, et al., "A Randomized Trial of Bevacizumab for Newly Diagnosed Glioblastoma," New England Journal of Medicine, vol. 370, no. 8, pp. 699-708, 2014. [168] M. E. Hegi, A.-C. Diserens, T. Gorlia, M.-F. Hamou, N. De Tribolet, M. Weller, J. M. Kros, J. A. Hainfellner, W . Mason, L. Mariani, et al., "MGMT gene silencing and benefit from temozolomide in glioblastoma, " New England Journal of Medicine, vol. 352, no. 10, pp. 997-1003, 2005. [169] J. N. Weinstein, E. A. Collisson, G. B. Mills, K. R. M. Shaw, B. A. Ozenberger, K. Ellrott, I. Shmulevich , C. Sander, J. M . Stuart, C. G. A. R. Network, et al., ''The Cancer Genome Atlas Pan-Cancer analysis project," Nature Genetics, vol. 45, no. 10, p. 1113, 2013. [170] Y. Lee, K. H. Kim, D. G. Kim, H. J. Cho, Y. Kim, J. Rheey, K. Shin, Y. J. Seo, Y.-S. Choi , J.- 1. Lee, et al., "FoxMl Promotes Sternness and Radio-Resistance of Glioblastoma by Regulating the Master Stem Cell Regulator Sox2," PLoS ONE, vol. 10, no. 10, p. e0137703, 2015. [171] U. B. Maachani, U. Shankavaram, T. Kramp, P. J. Tofilon, K. Camphausen, and A. T. Tandie, "FOXMl and STAT3 interaction confers radioresistance in glioblastoma cells," Oncotarget, vol. 7, no. 47, p. 77365, 2016. [172] L. Cheng, Q. Wu, Z. Huang, 0 . A. Guryanova, Q. Huang, W. Shou, J. N. Rich, and S. Bao, "LlCAM regulates DNA damage checkpoint response of glioblastoma stem cells through NBSl, " The EMBO Journal, vol. 30, no. 5, pp. 800-813, 2011 . 208 [173] X. Han, P. Ranganathan , C. Tzimas, K. L. Weaver, K. Jin, L. Astudillo, W . Zhou , X. Zhu , B. Li, D. J. Robbins, et al., " Notch Represses Transcription by PRC2 Recruitment to the Ternary Complex," Molecular Cancer Research, vol. 15, no. 9, pp. 1173-1183, 2017. [174] A. Balbous, U. Cortes, K. Guilloteau, P. Rivet, B. Pinel , M . Duchesne, J. Godet, 0 . Bois sonnade, M. Wager, R. J. Bensadoun , et al., "A radiosensitizing effect of rad51 inhibition in glioblastoma stem-like cells, " BMC Cancer, vol. 16, no. 1, p. 604, 2016. [175] S.-H . Kim , K. Joshi, R. Ezhilarasan , T. R. Myers, J. Siu, C. Gu, M. Nakano-Okuno, D. Taylor, M. Minata, E. P. Su Iman, et al., " EZH2 protects glioma stem cells from radiation induced cell death in a MELK/ FOXMl-dependent manner, " Stem Cell Reports, vol. 4, no. 2, pp. 226-238, 2015. [176] S. U. Ahmed, R. Carruthers, L. Gilmour, S. Yildirim, C. Watts, and A. J. Chalmers, "Selective inhibition of parallel dna damage response pathways optimizes radiosensitization of glioblastoma stem-like cells," Cancer Research, vol. 75, no. 20, pp. 4416-4428, 2015. [177] A. Karim , K. McCarthy, A. Jawahar, D. Smith , B. Willis, and A. Nanda, "Differential cyclooxygenase-2 enzyme expression in radiosensitive versus radioresistant glioblastoma multiforme cell lines," Anticancer Research, vol. 25, no. 1B, pp. 675-679, 2005. [178] S. H. K. Vellanki , A. Grabrucker, S. Liebau , C. Proepper, A. Eramo, V. Braun , T. Boeckers, K.-M . Debatin, and S. Fulda, "Small-molecule xiap inhibitors enhance 1 -irradiation-induced apoptosis in glioblastoma," Neoplasia, vol. 11, no. 8, pp. 743-W9, 2009. [179] H. Ma, L. Rao, H. Wang, Z. Mao, R. Lei , Z. Yang, H. Qing, and Y. Deng, ''Transcriptome analysis of glioma cells for the dynamic response to 1 -irradiation and dual regulation of 209 apoptosis genes: a new insight into radiotherapy for glioblastomas," Ce// Death & Disease, vol. 4, no. 10, p. e895, 2013. [180] P. Godoy, S. Mello, D. Magalhaes, F. Donaires, P. Nicolucci, E. A. Donadi, G. Passos, and E. T. Sakamoto-Hojo, "Ionizing radiation-induced gene expression changes in TP53 proficient and deficient glioblastoma cell lines, " Mutation Research/ Genetic Toxicology and Environmental Mutagenesis, vol. 756, no. 1-2, pp. 46-55, 2013. [181] K. P. Bhat, V. Balasubramaniyan , B. Vaillant, R. Ezhilarasan , K. Hummelink, F. Hollingsworth, K. Wani, L. Heathcock, J. D. James, L. D. Goodman, et al., " Mesenchy mal differentiation mediated by NF-i'i:B promotes radiation resistance in glioblastoma," Cancer Cell, vol. 24, no. 3, pp. 331-346, 2013. [182] K. A. Effenberger, V. K. Urabe, and M. S. Jurica, " Modulating splicing with small molecular inhibitors of the spliceosome," Wiley Interdisciplinary Reviews: RNA, vol. 8, no. 2, p. e1381 , 2017. [183] H. Dvinge, E. Kim , 0. Abdel-Wahab, and R. K. Bradley, " RNA splicing factors as onco proteins and tumour suppressors," Nature Reviews Cancer, vol. 16, no. 7, p. 413, 2016. [184] F. M. Meliso, C. G. Hubert, P. A. F. Galante, and L. 0 . Penalva , " Rna processing as an alternative route to attack glioblastoma," Human Genetics, vol. 136, no. 9, pp. 1129-1141, 2017. [185] C. G. Hubert, R. K. Bradley, Y. Ding, C. M. Toledo, J. Herman, K. Skutt-Kakaria, E. J. Girard, J. Davison, J. Berndt, P. Corrin, et al., " Genome-wide RNAi screens in human brain 210 tumor isolates reveal a novel viability requirement for PHF5A," Genes & Development, vol. 27, no. 9, pp. 1032-1045, 2013. [186] M. Grzmil and B. A. Hemmings, "Translation regulation as a therapeutic target in cancer," Cancer research, vol. 72, no. 16, pp. 3891-3900, 2012. [187] J. Harrow, A. Frankish, J. M. Gonzalez, E. Tapanari, M. Diekhans, F. Kokocinski, B. L. Aken , D. Barrell, A. Zadissa , S. Searle, et al., "GENCODE: the reference human genome annotation for The ENCODE Project, " Genome Research, vol. 22, no. 9, pp. 1760-1774, 2012. [188] S. Anders, P. T. Pyl, and W. Huber, "HTSeq - a Python framework to work with high throughput sequencing data, " Bioinformatics, vol. 31, no. 2, pp. 166-169, 2015. [189] P. Langfelder and S. Horvath, "WGCNA: an R package for weighted correlation network analysis, " BMC Bioinformatics, vol. 9, no. 1, p. 559, 2008. [190] P. Langfelder, R. Luo, M. C. Oldham , and S. Horvath, "Is my network module preserved and reproducible?," PLoS Computational Biology, vol. 7, no. 1, p. e1001057, 2011. [191] K. Strimmer, "fdrtool : a versatile R package for estimating local and tail area-based false discovery rates," Bioinformatics, vol. 24, no. 12, pp. 1461-1462, 2008. [192] S. Shen, J. W. Park, Z.-x. Lu , L. Lin, M. D. Henry, Y. N. Wu, Q. Zhou, and Y. Xing, "rMATS: robust and flexible detection of differential alternative splicing from replicate RNA Seq data," Proceedings of the National Academy of Sciences, vol. 111, no. 51 , pp. E5593- E5601, 2014. 211 [193] J. A. Martin and Z. Wang, "Next-generation transcriptome assembly," Nature Reviews Genetics, vol. 12, no. 10, p. 671, 2011. [194] A. Fabregat, K. Sidiropoulos, P. Garapati, M. Gillespie, K. Hausmann, R. Haw, B. Jassal, S. Jupe, F. Korninger, S. McKay, et al., "The reactome pathway knowledgebase, " Nucleic Acids Research, vol. 44, no. D1 , pp. D481-D487, 2015. [195] H. Mi, B. Lazareva-Ulitsky, R. Loo, A. Kejariwal, J. Vandergriff, S. Rabkin, N. Guo, A. Mu ruganujan , 0 . Doremieux, M. J. Campbell , et al., "The PANTHER database of protein fam ilies, subfamilies, functions and pathways," Nucleic Acids Research, vol. 33, no. suppl_l , pp. D284-D288, 2005. [196] Z. Tang, C. Li , B. Kang, G. Gao, C. Li, and Z. Zhang, "GEPIA: a web server for cancer and normal gene expression profiling and interactive analyses, " Nucleic Acids Research, vol. 45, no. Wl , pp. W98-W102, 2017. [197] S. Ning, J. Zhang, P. Wang, H. Zhi, J. Wang, Y. Liu, Y. Gao, M. Guo, M. Yue, L. Wang, et al., "Lnc2Cancer: a manually curated database of experimentally supported lncRNAs associated with various human cancers, " Nucleic Acids Research, vol. 44, no. D1, pp. D980- D985, 2015. [198] K. C. Cotto, A. H. Wagner, Y.-Y. Feng, S. Kiwala, A. C. Coffman, G. Spies, A. Wollam, N. C. Spies, 0. L. Griffith, and M. Griffith, " DGldb 3.0: a redesign and expansion of the drug-gene interaction database," Nucleic Acids Research, vol. 46, no. D1 , pp. D1068- D1073, 2017. 212 [199] M. V. Kuleshov, M. R. Jones, A. D. Rouillard , N. F. Fernandez, Q. Duan , Z. Wang, S. Koplev, S. L. Jenkins, K. M. Jagodnik, A. Lachmann, et al., " Enrichr: a comprehensive gene set enrichment analysis web server 2016 update," Nucleic Acids Research, vol. 44, no. Wl , pp. W90-W97, 2016. [200] D. Nandi, P. S. Cheema, N. Jaiswal, and A. Nag, "FoxMl: repurposing an oncogene as a biomarker," in Seminars in Cancer Biology, vol. 52, pp. 74-84, Elsevier, 2018. [201] H.-Z. Chen , S.-Y. Tsai, and G. Leone, " Emerging roles of E2Fs in cancer: an exit from cell cycle control, " Nature Reviews Cancer, vol. 9, no. 11, p. 785, 2009. [202] L. S. Payne and P. H. Huang, "The pathobiology of collagens in glioma," Molecular Cancer Research, vol. 11, no. 10, pp. 1129-1140, 2013. [203] S. Monferran, N. Skuli, C. Delmas, G. Favre, J. Bonnet, E. Cohen-Jonathan- Moya!, and C. Toulas, "avfJ3 and avfJ5 integrins control glioma cell response to ionising radiation through ilk and rhob," International Journal of Cancer, vol. 123, no. 2, pp. 357-364, 2008. [204] Z. Du , C. Cai, M. Sims, F. A. Boop, A. M. Davidoff, and L. M . Pfeffer, ''The effects of type I interferon on glioblastoma cancer stem cells," Biochemical and Biophysical Research Communications, vol. 491 , no. 2, pp. 343-348, 2017. [205] M. Budhwani, R. Mazzieri, and R. Dolcetti, "Plasticity of type I interferon-mediated re sponses in cancer therapy: from anti-tumor immunity to resistance," Frontiers in Oncology, vol. 8, p. 322, 2018. [206] Z. Peng, C. Liu , and M. Wu , "New insights into long noncoding rnas and their roles in glioma," Molecular Cancer, vol. 17, no. 1, p. 61, 2018. 213 [207] X. Wu , Y. Wang, T. Yu , E. Nie, Q. Hu, W . Wu , T. Zhi, K. Jiang, X. Wang, X. Lu , et al., " Blocking mir155hg/ mir-155 axis inhibits mesenchymal transition in glioma," Neuro oncology, vol. 19, no. 9, pp. 1195-1205, 2017. [208] C. Miao, K. Zhao, J. Zhu, C. Liang, A. Xu, Y. Hua, J. Zhang, S. Liu , Y. Tian, C. Zhang, et al., "Clinicopathological and prognostic role of long noncoding rna linc00152 in var ious human neoplasms: Evidence from meta-analysis," BioMed Research International, vol. 2017, pp. 1-11, 2017. [209] H. Zhang, Y. Cai , L. Zheng, Z. Zhang, X. Lin , and N. Jiang, "Long noncoding rna neatl regulate papillary thyroid cancer progression by modulating mir-129-5p/ klk7 expression," Journal of Cellular Physiology, vol. 233, no. 10, pp. 6638-6648, 2018. [210] W . Zhang, Y. Bi, J. Li, F. Peng, H. Li, C. Li, L. Wang, F. Ren, C. Xie, P. Wang, et al., "Long noncoding rna ftx is upregulated in gliomas and promotes proliferation and invasion of glioma cells by negatively regulating mir-342-3p," Laboratory Investigation, vol. 97, no. 4, p. 447, 2017. [211] H. Climente-Gonzalez, E. Porta-Pardo, A. Godzik, and E. Eyras, "The functional impact of alternative splicing in cancer, " Cell Reports, vol. 20, no. 9, pp. 2215-2226, 2017. [212] E. Macaeva, Y. Saeys, K. Tabury, A. Janssen, A. Michaux, M. A. Benotmane, W. H. De Vos, S. Baatout, and R. Quintens, " Radiation-induced alternative transcription and splicing events and their applicability to practical biodosimetry," Scientific Reports, vol. 6, p. 19251, 2016. 214 [213] N. H. Binh , K. Satoh , K. Kobayashi , M. Takamatsu , Y. Hatano, A. Hirata, H. Tomita , T. Kuno, and A. Hara, " Galectin-3 in preneoplastic lesions of glioma, " Journal of Neuro- Oncology, vol. 111, no. 2, pp. 123-132, 2013. [214] K. Ori no and K. Watanabe, "Molecular, physiological and clinical aspects of the iron storage protein ferritin, " The Veterinary Journal, vol. 178, no. 2, pp. 191-201, 2008. [215] M. Pang, X. Liu, B. Slagle-Webb, A. Madhankumar, and J. Connor, " Role of h-ferritin in radiosensitivity of human glioma cells," J Cancer Biol Treat, vol. 3, no. 006, pp. 1-10, 2016. [216] D.-H. Cho, Y.-M . Hong, H.-J. Lee, H.-N. Woo, J.-O. Pyo, T. W . Mak, and Y.-K. Jung, "Induced inhibition of ischemic/ hypoxic injury by a pip, a novel apaf-1-interacting protein ," Journal of Biological Chemistry, vol. 279, no. 38, pp. 39942-39950, 2004. [217] C. Simion, M. E. Cedano-Prieto, and C. Sweeney, "The lrig family: enigmatic regulators of growth factor receptor signaling," Endocrine-related cancer, vol. 21 , no. 6, pp. R431-R443, 2014. [218] Q. Xiao, Y. Tan, Y. Guo, H. Yang, F. Mao, R. Xie, B. Wang, T. Lei, and D. Guo, "Soluble lrig2 ectodomain is released from glioblastoma cells and promotes the proliferation and inhibits the apoptosis of glioblastoma cells in vitro and in vivo in a similar manner to the full-length lrig2, " PLoS ONE, vol. 9, no. 10, p. e111419, 2014. [219] N. Dalla Venezia, A. Vincent, V. Marcel, F. Catez, and J.-J. Diaz, " Emerging Role of Eukaryote Ribosomes in Translational Control," International Journal of Molecular Sciences, vol. 20, no. 5, p. 1226, 2019. 215 [220] L. N. Kent and G. Leone, "The broken cycle: E2f dysfunction in cancer," Nature Reviews Cancer, vol. 19, no. 6, pp. 326-338, 2019. [221] I. Thurlings and A. de Bruin, "E2f transcription factors control the roller coaster ride of cell cycle gene expression ," in Methods in Molecular Biology, pp. 71-88, Springer New York, 2016. [222] M. M. Alonso, J. Fueyo, J. W. Shay, K. D. Aldape, H. Jiang, O.-H . Lee, D. G. Johnson, J. Xu , Y. Kondo, T. Kanzawa, et al., "Expression of Transcription Factor E2Fl and Telom erase in Glioblastomas: Mechanistic Linkage and Prognostic Significance, " Journal of the National Cancer Institute, vol. 97, no. 21, pp. 1589-1600, 2005. [223] X. Li, H. Zhang, and X. Wu, "Long noncoding RNA DLX6-AS1 accelerates the glioma car cinogenesis by competing endogenous sponging miR-197-5p to relieve e2fl, " Gene, vol. 686, pp. 1-7, 2019. [224] L. Xia , D. Nie, G. Wang, C. Sun, and G. Chen , " FER1I4/ miR-372/ e2fl works as a ceRNA system to regulate the proliferation and cell cycle of glioma cells," Journal of Cellular and Molecular Medicine, vol. 23, no. 5, pp. 3224-3233, 2019. [225] B. Yang, Q. Meng, Y. Sun, L. Gao, and J. Yang, "Long non-coding RNA SNHG16 con tributes to glioma malignancy by competitively binding miR-20a-5p with E2Fl.," Journal of Biological Regulators & Homeostatic Agents, vol. 32, no. 2, pp. 251-261, 2018. [226] N. Wu, L. Xiao, X. Zhao, J. Zhao, J. Wang, F. Wang, S. Cao, and X. Lin, "miR-125b regulates the proliferation of glioblastoma stem cells by targeting e2f2," FEBS Letters, vol. 586, no. 21, pp. 3831-3839, 2012. 216 [227] 0 . K. Okamoto, S. M. Oba-Shinjo, L. Lopes, and S. K. N. Marie, "Expression of HOXC9 and E2F2 are up-regulated in CD133+ cells isolated from human astrocytomas and associate with transformation of human astrocytes," Biochimica et Biophysica Acta (BBA)-Gene Structure and Expression, vol. 1769, no. 7-8, pp. 437-442, 2007. [228] H. Song, Y. Zhang, N. Liu, D. Zhang, C. Wan , S. Zhao, Y. Kong, and L. Yuan , " Let-7b inhibits the malignant behavior of glioma cells and glioma stem-like cells via down regulation of E2F2, " Journal of Physiology and Biochemistry, vol. 72, no. 4, pp. 733-744, 2016. [229] S. Qiu , D. Huang, D. Yin , F. Li , X. Li , H. fu Kung, and Y. Peng, "Suppression of tu morigenicity by MicroRNA-138 through inhibition of EZH2-CDK4/ 6-pRb-e2fl signal loop in glioblastoma multiforme," Biochimica et Biophysica Acta (BBA) - Molecular Basis of Disease, vol. 1832, no. 10, pp. 1697-1707, 2013. [230] Y. Zhang, D. Han, W. Wei, W. Cao, R. Zhang, Q. Dong, J. Zhang, Y. Wang, and N. Liu , "MiR-218 inhibited growth and metabolism of human glioblastoma cells by directly targeting e2f2," Cellular and Molecular Neurobiology, vol. 35, no. 8, pp. 1165-1173, 2015. [231] Y. Chen , F. Zhao, D. Cui , R. Jiang, J. Chen , Q. Huang, and J. Shi, "HOXD-AS1/ miR-130a sponge regulates glioma development by targeting e2f8, " International Journal of Cancer, vol. 142, no. 11, pp. 2313-2322, 2018. [232] V. Gouaze-Andersson, M.-J. Gherardi, A. Lemarie, J. Gilhodes, V. Lubrano, F. Arnauduc, E. C.-J. Moyal, and C. Toulas, "FGFRl/ FOXMl pathway: a key regulator of glioblastoma stem cells radioresistance and a prognosis biomarker," Oncotarget, vol. 9, no. 60, p. 31637, 2018. 217 [233] Q. Ma, Y. Liu , L. Shang, J. Yu , and Q. Qu , ''The FOXMl / BUBlb signaling pathway is essential for the tumorigenicity and radioresistance of glioblastoma," Oncology Reports, vol. 38, no. 6, pp. 3367-3375, 2017. [234] S. Zhang, B. S. Zhao, A. Zhou, K. Lin, S. Zheng, Z. Lu, Y. Chen , E. P. Sulman, K. Xie, 0 . Bogler, S. Majumder, C. He, and S. Huang, "m 6 a demethylase ALKBH5 maintains tumorigenicity of glioblastoma stem-like cells by sustaining FOXMl expression and cell proliferation program," Cancer Cell, vol. 31, no. 4, pp. 591-606.e6, 2017. [235] J.-J . Quan , J.-N . Song, and J.-Q. Qu , "PARP3 interacts with FoxMl to confer glioblastoma cell radioresistance," Tumor Biology, vol. 36, no. 11, pp. 8617-8624, 2015. [236] A. hua Gong, P. Wei , S. Zhang, J. Yao, Y. Yuan, A. dong Zhou , F. F. Lang, A. B. Heim berger, G. Rao, and S. Huang, "FoxMl drives a feed-forward STAT3-activation signaling loop that promotes the self-renewal and tumorigenicity of glioblastoma stem-like cells," Cancer Research, vol. 75, no. 11, pp. 2337-2348, 2015. [237] N. Zhang, X. Wu, L. Yang, F. Xiao, H. Zhang, A. Zhou , Z. Huang, and S. Huang, "FoxMl inhibition sensitizes resistant glioblastoma cells to temozolomide by downregulating the expression of DNA-repair gene rad51, " Clinical Cancer Research, vol. 18, no. 21, pp. 5961- 5971 , 2012. [238] M. Fischer, P. Grossmann, M. Padi, and J. A. DeCaprio, " Integration of TP53, DREAM, MMB-FOXMl and RB-e2f target gene analyses identifies cell cycle gene regulatory net works, " Nucleic Acids Research, vol. 44, no. 13, pp. 6070-6086, 2016. 218 [239] H. Bai, A. S. Harmanc1, E. Z. Erson-Omay, J. Li , S. Co?kun , M. Simon, B. Krischek, K. Ozduman , S. B. Omay, E. A. Sorensen, $. Turcan , M. Bak1rc1glu, G. Carrion-Grant, P. B. Murray, V. E. Clark, A. G. Ercan-Sencicek, J. Knight, L. Sencar, S. Altinok, L. D. Kaulen , B. Gulez, M. Timmer, J. Schramm, K. Mishra-Gorur, 0 . Henegariu, J. Moliterno, A. Louvi, T. A. Chan, S. L. Tannheimer, M. N. Pamir, A. 0. Vortmeyer, K. Bilguvar, K. Yasuno, and M. Gunel, "Integrated genomic characterization of IDHl-mutant glioma malignant progression, " Nature Genetics, vol. 48, no. 1, pp. 59-66, 2015. [240] A. C. Belkina and G. V. Denis, "BET domain co-regulators in obesity, inflammation and cancer," Nature Reviews Cancer, vol. 12, no. 7, pp. 465-477, 2012. [241] L. Xu , Y. Chen , A. Mayakonda, L. Koh , Y. K. Chong, D. L. Buckley, E. Sandanaraj, S. W . Lim, R. Y.-T. Lin, X.-Y. Ke, M.-L. Huang, J. Chen, W. Sun, L.-Z. Wang, B. C. Goh, H. Q. Dinh, D. Kappei, G. E. Winter, L.-W. Ding, B. T. Ang, B. P. Berman, J. E. Bradner, C. Tang, and H. P. Koeffler, "Targetable BET proteins- and e2fl-dependent transcriptional program maintains the malignancy of glioblastoma, " Proceedings of the National Academy of Sciences, vol. 115, no. 22, pp. E5086-E5095, 2018. [242] F. M. Meliso, C. G. Hubert, P. A. F. Galante, and L. 0 . Penalva, " RNA processing as an alternative route to attack glioblastoma," Human Genetics, vol. 136, no. 9, pp. 1129-1141, 2017. [243] B. R. Correa , P. R. de Araujo, M. Qiao, S. C. Burns, C. Chen , R. Schlegel, S. Agarwal , P. A. F. Galante, and L. 0. F. Penalva, " Functional genomics analyses of RNA-binding 219 proteins reveal the splicing regulator SNRPB as an oncogenic candidate in glioblastoma," Genome Biology, vol. 17, no. 1, p. 125, 2016. [244] Y. Liu, Y. Shen, T. Sun , and W. Yang, " Mechanisms regulating radiosensitivity of glioma stem cells," Neoplasma , vol. 64, no. 05, pp. 655-665, 2017. [245] J. Wang, T. P. Wakeman, J. D. Lathia, A. B. Hjelmeland, X.-F. Wang, R. R. White, J. N. Rich, and B. A. Sullenger, " Notch Promotes Radioresistance of Glioma Stem Cells, " Stem Cells, vol. 28, no. 1, pp. 17-28, 2010. [246] R. Hannen, M. Hauswald, and J. W . Bartsch, "A rationale for targeting extracellular regu lated kinases ER Kl and ERK2 in glioblastoma, " Journal of Neuropathology & Experimental Neurology, vol. 76, no. 10, pp. 838-847, 2017. [247] L. Xu , Y. Chen , M. Dutra-Clarke, A. Mayakonda, M. Hazawa, S. E. Savinoff, N. Doan , J. W. Said, W . H. Yong, A. Watkins, et al., " Bcl6 promotes glioma and serves as a therapeutic target," Proceedings of the National Academy of Sciences, vol. 114, no. 15, pp. 3981-3986, 2017. [248] E. Cho and Y. Yen , " Novel regulators and molecular mechanisms of p53R2 and its disease relevance," Biochimie, vol. 123, pp. 81-84, 2016. [249] L. Zhai, E. Ladomersky, A. Lenzen , B. Nguyen, R. Patel , K. L. Lauing, M. Wu, and D. A. Wainwright, "Idol in cancer: a gemini of immune checkpoints, " Cellular & Molecular Immunology, vol. 15, no. 5, p. 447, 2018. [250] P. Kesarwani, A. Prabhu, S. Kant, P. Kumar, S. F. Graham, K. L. Buelow, G. D. Wilson , C. R. Miller, and P. Chinnaiyan, "Tryptophan metabolism contributes to radiation-induced 220 immune checkpoint reactivation in glioblastoma," Clinical Cancer Research, vol. 24, no. 15, pp. 3632-3643, 2018. [251] R. 0. Hynes and A. Naba, "Overview of the matrisome-an inventory of extracellular matrix constituents and functions," Cold Spring Harbor Perspectives in Biology, vol. 4, no. 1, pp. a004903-a004903, 2011. [252] C. Heit, B. C. Jackson, M. McAndrews, M. W. Wright, D. C. Thompson, G. A. Silver man, D. W . Nebert, and V. Vasiliou, "Update of the human and mouse SERPIN gene superfamily, " Human Genomics, vol. 7, no. 1, p. 22, 2013. [253] S. Bonnal, L. Vigevani, and J. Valcarcel, "The spliceosome as a target of novel antitumour drugs," Nature Reviews Drug Discovery, vol. 11, no. 11, pp. 847-859, 2012. [254] I. Pal , M. Safari, M . Jovanovic, S. E. Bates, and C. Deng, "Targeting translation of mRNA as a therapeutic strategy in cancer, " Current Hematologic Malignancy Reports, pp. 1-9, 2019. [255] S. L. Salzberg, " Open questions: How many genes do we have?," vol. 16, no. 1. [256] B. L. Aken, S. Ayling, D. Barrell, L. Clarke, V. Curwen, S. Fairley, J. F. Banet, K. Billis, C. G. Giron, T. Hourlier, K. Howe, A. Kahari, F. Kokocinski, F. J. Martin, D. N. Murphy, R. Nag, M. Ruffier, M. Schuster, Y. A. Tang, J.-H . Vogel, S. White, A. Zadissa, P. Flicek, and S. M. J. Searle, "The ensembl gene annotation system, " Database, vol. 2016, p. baw093, 2016. [257] A. Bairoch , "The SWISS-PROT protein sequence database and its supplement TrEMBL in 2000, " Nucleic Acids Research, vol. 28, pp. 45-48, jan 2000. 221 [258] S. J. Kiniry, A. M. Michel, and P. V. Baranov, "The GWIPS-viz browser," Current Protocols in Bioinformatics, vol. 62, p. e50, may 2018. [259] 5.-Q. Xie, P. Nie, Y. Wang, H. Wang, H. Li, Z. Yang, Y. Liu, J. Ren, and Z. Xie, " RPFdb: a database for genome wide information of translated mRNA generated from ribosome profiling," Nucleic Acids Research, vol. 44, pp. D254-D258, oct 2015. [260] W .-5. Wu, Y.-X. Jiang, J.-W. Chang, Y.-H . Chu, Y.-H . Chiu, Y.-H . Tsao, T. E. M. Nordling, Y.-Y. Tseng, and J. T. Tseng, "HRPDviewer: Human ribosome profiling data viewer," Database, vol. 2018, Jan 2018. 00000. [261] M. Matsui, N. Yachie, Y. Okada, R. Saito, and M. Tomita, " Bioinformatic analysis of post-transcriptional regulation by uORF in human and mouse," FEBS Letters, vol. 581 , pp. 4184-4188, aug 2007. [262] S. E. Calvo, D. J. Pagliarini, and V. K. Mootha, " Upstream open reading frames cause widespread reduction of protein expression and are polymorphic among humans," Proceed ings of the National Academy of Sciences, vol. 106, pp. 7507-7512, apr 2009. [263] G. Menschaert, W. V. Criekinge, T. Notelaers, A. Koch, J. Crappe, K. Gevaert, and P. V. Damme, "Deep proteome coverage based on ribosome profiling aids mass spectrometry based protein and peptide discovery and provides evidence of alternative translation prod ucts and near-cognate translation initiation events," Molecular &. Cellular Proteomics, vol. 12, pp. 1780-1790, feb 2013. [264] M. Kozak, "Pushing the limits of the scanning mechanism for initiation of translation ," Gene, vol. 299, pp. 1-34, oct 2002. 222 [265] H. M . Hood, D. E. Neafsey, J. Galagan , and M. S. Sachs, "Evolutionary roles of upstream open reading frames in mediating gene regulation in fungi," Annual Review of Microbiology, vol. 63, pp. 385-409, oct 2009. [266] N. Sonenberg and A. G. Hinnebusch, " Regulation of translation initiation in eukaryotes: Mechanisms and biological targets," Cell, vol. 136, pp. 731-745, feb 2009. [267] M . Iacono, F. Mignone, and G. Pesole, "uAUG and uORFs in human and rodent 5'untranslated mRNAs," Gene, vol. 349, pp. 97-105, apr 2005. [268] A. Franks, E. Airoldi , and N. Slavov, "Post-transcriptional regulation across human tissues," PLOS Computational Biology, vol. 13, p. e1005535, may 2017. [269] P. D. Lu , H. P. Harding, and D. Ron , ''Translation reinitiation at alternative open reading frames regulates gene expression in an integrated stress response, " The Journal of Cell Biology, vol. 167, pp. 27-33, oct 2004. [270] M. J. Clemens, " Initiation Factor elF2a Phosphorylation in Stress Responses and Apopto sis," in Signaling Pathways for Translation , pp. 57-89, Springer Berlin Heidelberg, 2001. [271] J. Medenbach, M . Seiler, and M . W . Hentze, "Translational control via protein-regulated upstream open reading frames, " Cell, vol. 145, pp. 902-913, jun 2011. [272] H. A. Meijer and A. A. M . Thomas, "Control of eukaryotic protein synthesis by upstream open reading frames in the 5h-untranslated region of an mRNA, " p. 11, 2002. 223 [273] P. D. Lu , H. P. Harding, and D. Ron , ''Translation reinitiation at alternative open reading frames regulates gene expression in an integrated stress response, " The Journal of Cell Biology, vol. 167, pp. 27-33, oct 2004. [274] J. Medenbach, M . Seiler, and M . W . Hentze, "Translational control via protein-regulated upstream open reading frames, " Cell, vol. 145, pp. 902-913, jun 2011. [275] S. STONE, "PATHOLOGICAL SCIENCE," in Flavor Physics for the Millennium, WORLD SCIENTIFIC, sep 2001. [276] J. Schulz, N. Mah, M . Neuenschwander, T. Kischka, R. Ratei, P. M . Schlag, E. Castanos Velez, I. Fichtner, P.-U . Tunn, C. Denkert, et al., "Loss-of-function uorf mutations in human malignancies, " Scientific reports, vol. 8, no. 1, pp. 1-10, 2018. [277] T. Ouspenskaia , T. Law, K. R. Clauser, S. Klaeger, S. Sarkizova , F. Aguet, B. Li , E. Chris tian, B. A. Knisbacher, P. M . Le, et al., "Thousands of novel unannotated proteins expand the MHC I immunopeptidome in cancer," bioRxiv, 2020. [278] F.-B. Gao, J. D. Richter, and D. W . Cleveland , "Rethinking unconventional translation in neurodegeneration, " Cell, vol. 171, pp. 994-1000, Nov. 2017. [279] C. W . Dunn , X . Luo, and Z. Wu , "Phylogenetic analysis of gene expression ," Integrative and Comparative Biology, vol. 53, pp. 847-856, jun 2013. [280] S. H. Wang, C. J. Hsiao, Z. Khan, and J. K. Pritchard, " Post-translational buffering leads to convergent protein expression levels between primates," Genome Biology, vol. 19, p. 83, Dec 2018. 00000. 224 [281] Z. Khan , M. J. Ford , D. A. Cusanovich , A. Mitrano, J. K. Pritchard , and Y. Gilad, "Primate Transcript and Protein Expression Levels Evolve Under Compensatory Selection Pressures, " Science, vol. 342, pp. 1100-1104, Nov 2013. 00133. [282] P. Spealman , A. W . Naik, G. E. May, S. Kuersten , L. Freeberg, R. F. Murphy, and J. Mc Manus, " Conserved non-AUG uORFs revealed by a novel regression analysis of ribosome profiling data," Genome Research, vol. 28, pp. 214-222, Feb. 2018. 00005. [283] T. F. Martinez, Q. Chu , C. Donaldson , D. Tan , M. N. Shokhirev, and A. Saghatelian , "Ac curate annotation of human protein-coding small open reading frames, " Nature Chemical Biology, Dec. 2019. 00000. [284] S. F. Altschul , W . Gish, W . Miller, E. W . Myers, and D. J. Lipman , "Basic local alignment search tool," Journal of Molecular Biology, vol. 215, pp. 403-410, Oct. 1990. [285] J. Shine and L. Dalgarno, "Occurrence of heat-dissociable ribosomal RNA in insects: The presence of three polynucleotide chains in 26 s RNA from cultured aedes aegypti cells," Journal of Molecular Biology, vol. 75, pp. 57-72, Mar. 1973. [286] G.-L. Chew, A. Pauli, and A. F. Schier, " Conservation of uORF repressiveness and sequence features in mouse, human and zebrafish," Nature Communications, vol. 7, May 2016. [287] S. Gupta, J. A. Stamatoyannopoulos, T. L. Bailey, and W . Noble, "Quantifying similarity between motifs, " Genome Biology, vol. 8, no. 2, p. R24, 2007. [288] A. Cassese, M. Guindani, M. G. Tadesse, F. Falciani, and M. Vannucci, "A hierarchical bayesian model for inference of copy number variants and their association to gene expres- sion, " The Annals of Applied Statistics, vol. 8, pp. 148-175, mar 2014. 225 [289] C. Briceno, "Massive variability surveys from venezuela," arxiv:astro-ph/ 0304081v1 , 2003. [290] D. Koop, J. Freire, and C. T. Silva, " Enabling reproducible science with vistrails, " arxiv:1309.1784v2, 2013. [291] N. J. Kalton, 'The basic sequence problem ," arxiv:math/ 9408205v1 , 1994. [292] R. Luo, J. Xu, Y. Zhang, X. Ren, and X. Sun, " Pkuseg: A toolkit for multi-domain chinese word segmentation ," arxiv:1906.11455v2, 2019. [293] R. Cogranne, "Determining jpeg image standard quality factor from the quantization ta bles, " arxiv:1802.00992v1, 2018. [294] Y. Zhu , R. M. Stephens, P. S. Meltzer, and S. R. Davis, "SRAdb: query and use public next-generation sequencing data from within r," BMC Bioinformatics, vol. 14, no. 1, p. 19, 2013. [295] J. Zhu and S. Davis, " Bioconductor:sradb," Dec 2018. [296] J. Wilson, "Statistical computing with r: selecting the right tool for the job-r commander or something else?," Wiley Interdisciplinary Reviews: Computational Statistics, vol. 4, pp. 518-526, aug 2012. [297] S. Choudhary, "saketkc/ pysradb v0.9.0, " Feb 2019. [298] D. Faes, "Use of python programming language in astronomy and science," Journal of Computational Interdisciplinary Sciences, vol. 3, no. 3, 2012. 226 [299] B. Gruning, , R. Dale, A. Sjodin , B. A. Chapman , J. Rowe, C.H . Tomkins-Tinch , R. Valieris, and J. Koster, " Bioconda: sustainable and comprehensive software distribution for the life sciences," Nature Methods, vol. 15, pp. 475-476, jul 2018. [300] M.A. Breddels and J. Veljanoski, "Vaex: big data exploration in the era of gaia, " Astronomy & Astrophysics, vol. 618, p. A13, oct 2018. [301] C. da Costa-Luis, S. L., H. Mary, noamraph, M. Korobov, I. Ivanov, M. Bargull, James, G. Chen, M. D. Pagel, S. Malmgren, Socialery, J. McCracken, F. Dill, D. Panteleit, A. Roth berg, Y. Halchenko, T. Ostasevicius, S. Pokharel , ReadmeCritic, P. VandeHaar, K. che Wu , jcea, Hugo, F. Hurley, E. Betts, D. Bau, A. Persaud , Alexander, and A. Umer, "tqdm / tqdm: tqdm v4.20.0 stable," Apr 2018. [302] R. Teymourzadeh, S. A. Ahmed, K. W . Chan, and M. V. Hoong, "Smart GSM based home automation system, " in 2013 IEEE Conference on Systems, Process & Control (JCSPC), IEEE, dee 2013. [303] T. Kluyver, B. Ragan-Kelley, F. Perez, B. Granger, M. Bussonnier, J. Frederic, K. Kelley, J. Hamrick, J. Grout, S. Corlay, P. Ivanov, D. Avila, S. Abdalla, and C. Willing, "Jupyter notebooks - a publishing format for reproducible computational workflows, " in Position ing and Power in Academic Publishing: Players, Agents and Agendas (F. Loizides and B. Schmidt, eds.) , pp. 87 - 90, IOS Press, 2016. [304] D. R. Schrider and A. D. Kern, " Inferring selective constraint from population genomic data suggests recent regulatory turnover in the human brain," Genome Biology and Evolution, vol. 7, pp. 3511-3528, nov 2015. 227 [305] J. D. Blair, D. Hockemeyer, J. A. Doudna, H. S. Bateup, and S. N. Floor, "Widespread translational remodeling during human neuronal differentiation, " Ce// reports, vol. 21, no. 7, pp. 2005-2016, 2017. [306] J. Merkin, C. Russell, P. Chen, and C. B. Burge, " Evolutionary dynamics of gene and isoform regulation in mammalian tissues," Science, vol. 338, no. 6114, pp. 1593-1599, 2012. [307] M. D. Schultz, Y. He, J. W . Whitaker, M. Hariharan, E. A. Mukamel, D. Leung, N. Ra jagopal, J. R. Nery, M. A. Urich, H. Chen, et al., " Human body epigenome maps reveal noncanonical dna methylation variation," Nature, vol. 523, no. 7559, p. 212, 2015. 228 Appendix A pysradb: A Python package to query next-generation sequencing metadata and data from NCBI Sequence Read Archive A.1 Introduction Several projects have made efforts to analyze and publish summaries of DNA- [288] and RNA-seq [289, 290] datasets. Obtaining metadata and raw data from the NCBI Sequence Read Archive (SRA) [291] is often the first step towards re-analyzing public next-generation sequencing datasets in order to compare them to private data or test a novel hypothesis. The NCBI SRA toolkit [292] provides utility methods to download raw sequencing data, while the metadata can be obtained by querying the website or through the Entrez efetch command line utility [293]. Most workflows analyzing public data rely on first searching for relevant keywords in the metadata either through 229 the command line utility or the website, gathering relevant sample(s) of interest and then down loading these. A more streamlined workflow can enable the performance of all these steps at once. In order to make querying both metadata and data more precise and robust, the SRAdb [294] project provides a frequently updated SQLite database containing all the metadata parsed from SRA. SRAdb tracks the five main data objects in SRA's metadata: submission, study, sample, experiment and run . These are mapped to five different relational database tables that are made available in the SQLite file. The metadata semantics in the file remain as they are on SRA. The accompanying package, SRAdb [295]. made available in the R programming language [296]. provides a convenient framework to handle metadata queries and raw data downloads by utilizing the SQLite database. Though powerful , SRAdb requires the end user to be familiar with the R programming language and does not provide a command-line interface for querying or download ing operations. The pysradb package [297] builds upon the principles of SRAdb, providing a simple and user-friendly command-line interface for querying metadata and downloading datasets from SRA. It obviates the need for the user to be familiar with any programming language for querying and downloading datasets from SRA. Additionally, it provides utility functions that will further help a user perform more granular queries, which are often required when dealing with multiple datasets on a large scale. By enabling both metadata search and download operations at the command line, pysradb aims to bridge the gap in seamlessly retrieving public sequencing datasets and the associated metadata. 230 pysradb [297] is written in Python [298] and is currently developed on GitHub under the open-source BSD 3-Clause License. To simplify the installation procedure for the end-user, it is also available for download through PyPI and bioconda [299]. A.2 Methods A.2.1 Implementation pysradb [297] is implemented in Python and uses pandas [300] for data frame based opera tions. Since downloading datasets can often take a long time, pysradb displays progress for long haul tasks using tqdm [301]. The metadata information is read in the form of an SQLite [302] database, made available by SRAdb [294]. Each sub-command of pysradb contains a self-contained help string that describes its pur pose and usage example. The help text can be accessed by passing the '-help' flag. There is also additional documentation available for the sub-commands on the project's website. We also provide example Jupyter [303] notebooks that demonstrate the functionality of the Python API. pysradb's development primarily occurred on GitHub and the code is tested continuously using Travis Cl webhook. This monitors all incoming pull requests and commits to the master branch . The testing happens on Python version 3.5, 3.6, and 3.7 on an Ubuntu 16.04 LTS virtual machine, while testing webhooks on the bioconda channel provide additional testing on 231 Mac-based systems. Nevertheless, pysradb should run on most Unix derivatives. A.2.2 Operation pysradb [297] can be run on either Linux- or Mac-based operating systems. It supports Python 3.5, 3.6 and 3.7. Requiring just two additional dependencies, pysradb can be easily installed using either a pip- or conda- based package manager via the bioconda [299] channel. An earlier version of this article can be found on bioRxiv https: / / doi. org/10. 1101/578500 A.3 Use cases pysradb [297] provides a chain of sub-commands for retrieving metadata, converting one acces sion to other and downloading. Each sub-command is designed to perform a single operation by default, while additional operations can be performed by passing additional flags. In the following section we demonstrate some of the use cases of these sub-commands. pysradb uses SRAmetadb . sqli te, a SQLite file produced and made available by SRAdb [294] project. The file itself can be downloaded using pysradb as: 232 $ pysradb srametadb The SRAmetadb. sqli te file is required for all other operations supported by pysradb. This file is required for all the sub-commands to function . By default, pysradb assumes that the file is located in the current working directory. Alternatively, it can supplied using the '-db path / to/ SRAmetadb.sqlite' argument. The SRAmetadb. sqli te that is required for all under- lying operations of pysradb is available at: https : / /s3 . amazonaws . com/starbuck1/sradb/ SRAmetadb . sqli te . gz or alternatively at https: / / gbnci -abcc. ncif crf . gov /backup/SRAmetadb. sqli te. gz. The examples here were run using SRAmetadb. sqli te with schema version 1.0 and creation timestamp 2019-01-25 00:38:19. A.3.1 Search Consider a case where a user is looking for Ribo-seq [148] public datasets on SRA. These datasets will often have 'ribosome profiling' appearing in the abstract or sample description . We can search for such projects using the 'search ' sub-command: $ pysradb search '"ribosome profiling"' I head 233 study _accession experiment_accession sam ple_accession run_accession DRP003075 DRX019536 DRS026974 DRR021383 DRP003075 DRX019537 DRS026982 DRR021384 DRP003075 DRX019538 DRS026979 DRR021385 DRP003075 DRX019540 DRS026984 DRR021387 DRP003075 DRX019541 DRS026978 DRR021388 DRP003075 DRX019543 DRS026980 DRR021390 DRP003075 DRX019544 DRS026981 DRR021391 ERP013565 ERX1264364 ERS1016056 ERR1190989 The results here list all relevant 'ribosome profiling' projects. A.3.2 Getting metadata for a SRA project Each SRA project (accession prefix 'SRP') on SRA consists of single or multiple experiments (accession prefix 'SRX') which are sequenced as single or multiple runs (accession prefix 'SRR'). Each experiment is carried out on an individual biological sample ( accession prefix 'SRS'). pysradb metadata can be used to obtain all the experiment, sample, and run accessions asso ciated with a SRA project as: $ pysradb metadata SRP010679 I head 234 study _accession experiment_accession sam ple_accession run_accession SRP010679 SRX118285 SRS290854 SRR403882 SRP010679 SRX118286 SRS290855 SRR403883 SRP010679 SRX118287 SRS290856 SRR403884 SRP010679 SRX118288 SRS290857 SRR403885 SRP010679 SRX118289 SRS290858 SRR403886 SRP010679 SRX118290 SRS290859 SRR403887 SRP010679 SRX118291 SRS290860 SRR403888 SRP010679 SRX118292 SRS290861 SRR403889 SRP010679 SRX118293 SRS290862 SRR403890 SRP010679 SRX118294 SRS290863 SRR403891 SRP010679 SRX118295 SRS290864 SRR403892 SRP010679 SRX118296 SRS290865 SRR403893 However, this information by itself is often incomplete. We require detailed metadata asso ciated with each sample to perform any downstream analysis. For example, the assays used for different samples and the corresponding treatment conditions. This can be done by supplying the '-desc' flag: $ pysradb metadata SRP010679 --desc I head -5 235 study _accession experiment_accession sam ple_accession run_accession sample_attribute SRP010679 SRX118285 SRS290854 SRR403882 source_name: PC3 hu- man prostate cancer cells II cell line: PC3 11 sample type: polyA RNA 11 treatment: ve- hide SRP010679 SRX118286 SRS290855 SRR403883 source_name: PC3 hu- man prostate cancer cells 11 cell line: PC3 II sample type: ribosome protected RNA 11 treat- ment: vehicle SRP010679 SRX118287 SRS290856 SRR403884 source_name: PC3 hu- man prostate cancer cells II cell line: PC3 11 sample type: polyA RNA 11 treatment: ra- pamycin SRP010679 SRX118288 SRS290857 SRR403885 source_name: PC3 hu- man prostate cancer cells 11 cell line: PC3 II sample type: ribosome protected R~A 6 I I treat- ment: rapamycin This can be further expanded to reveal the data in 'sample_attribute' column into separate columns via '-expand ' flag. This is most useful for samples that have associated treatment or cell type metadata available. $ pysradb metadata SRP010679 --desc --expand ... [truncated] run_accession cell - line sample_type source_name treatment SRR403882 pc3 polya rna pc3 human prostate cancer cells vehicle SRR403883 pc3 ribosome protected rna pc3 human prostate cancer cells vehicle SRR403884 pc3 polya rna pc3 human prostate cancer cells rapamycin SRR403885 pc3 ribosome protected rna pc3 human prostate cancer cells rapamycin SRR403886 pc3 polya rna pc3 human prostate cancer cells pp242 SRR403887 pc3 ribosome protected rna pc3 human prostate cancer cells pp242 SRR403888 pc3 polya rna pc3 human prostate cancer cells vehicle SRR403889 pc3 ribosome protected rna pc3 human prostate cancer cells vehicle SRR403890 pc3 polya rna pc3 human prostate cancer cells rapamycin SRR403891 pc3 ribosome protected rna pc3 human prostate cancer cells rapamycin SRR403892 pc3 polya rna pc3 human prostate cancer cells pp242 SRR403893 pc3 ribosome protected rna pc3 human prostate cancer cells pp242 Any SRA project might consist of experiments involving multiple assay types. The assay associated with any project can be obtained by providing --assay flag: $ pysradb metadata SRP000941 --assay I tr -s ' ' I cut -f5 -d' ' I tail -n +2 I sort I uniq -c 237 999 Bisulfite-Seq 768 ChlP-Seq 121 OTHER 353 RNA-Seq 28 WGS A.3.3 Getting SRPs from GSE The Gene Expression Omnibus database (GEO) [304] is the NCBI data repository for functional genomics data . It accepts array and sequence-based data from gene profiling experiments. For sequence-based data, the corresponding raw files are deposited to the SRA. GEO assigns a dataset accession (accession prefix 'GSE') that is linked to the corresponding accession on the SRA ( accession prefix 'SRP') . It is often necessary to interpolate between the two accessions. gse-to-srp sub-command allows converting GSE to SRP: $ pysradb gse-to-srp GSE24355 GSE25842 study_alias study_accession GSE24355 SRP003870 GSE25842 SRP005378 It can be further expanded to obtain the corresponding experiment and run accessions: 238 $ pysradb gse-to-srp --detailed --expand GSE100007 I head study_alias study _accession experiment_accession sam ple_accession experiment_alias GSE100007 SRP109126 SRX2916198 SRS2282390 GSM2667747 GSE100007 SRP109126 SRX2916199 SRS2282391 GSM2667748 GSE100007 SRP109126 SRX2916200 SRS2282392 GSM2667749 GSE100007 SRP109126 SRX2916201 SRS2282393 GSM2667750 GSE100007 SRP109126 SRX2916202 SRS2282394 GSM2667751 GSE100007 SRP109126 SRX2916203 SRS2282395 GSM2667752 GSE100007 SRP109126 SRX2916204 SRS2282396 GSM2667753 GSE100007 SRP109126 SRX2916205 SRS2282397 GSM2667754 GSE100007 SRP109126 SRX2916206 SRS2282400 GSM2667755 A.3.4 Getting a list of GEO experiments for a GEO study Any GEO study (accession prefix 'GSE') will involve a collection of experiments (accession pre fix 'GSM'). We can obtain an entire list of experiments corresponding to the study using the gse-to-gsm sub-command from pysradb: $ pysradb gse-to-gsm GSE41637 I head 239 sample_alias GSM2667747 GSM2667748 GSM2667749 GSM2667750 GSM2667751 GSM2667752 GSM2667753 GSM2667754 GSM2667755 study_alias experiment_alias GSE41637 GSM1020640_1 GSE41637 GSM1020641_1 GSE41637 GSM1020642_1 GSE41637 GSM1020643_1 GSE41637 GSM1020644_1 GSE41637 GSM1020645_1 GSE41637 GSM1020646_1 GSE41637 GSM1020647 _1 GSE41637 GSM1020648_1 However, a list of GSM accessions is not useful if one is performing any downstream analysis, which essentially requires more detailed information about the metadata associated with each experiment. This relevant metadata associated with each sample can be obtained by providing gse-to-gsm additional flags: $ pysradb gse-to-gsm --desc GSE41637 I head 240 study_alias experiment_alias sample_attribute GSE41637 GSM1020640 - 1 source_name: mouse_brain 11 strain : DBA/ 2J 11 tissue: brain GSE41637 GSM1020641 - 1 source_name: mouse_colon 11 strain: DBA/ 2J 11 tissue: colon GSE41637 GSM1020642 - 1 source_name: mouse_heart 11 strain : DBA/ 2J 11 tissue: heart GSE41637 GSM1020643 - 1 source_name: mouse_kidney 11 strain : DBA/ 2J 11 tissue: kidney GSE41637 GSM1020644_1 source_name: mouse_liver 11 strain : DBA/ 2J 11 tissue: liver GSE41637 GSM1020645 - 1 source_name: mouse_lung 11 strain : DBA/ 2J 11 tissue: lung GSE41637 GSM1020646 - 1 source_name: mouse_skm 11 strain: DBA/ 2J 11 tissue: skeletal muscle GSE41637 GSM1020647 - 1 source_name: mouse_spleen 11 strain : DBA/ 2J 11 tissue: spleen GSE41637 GSM1020648 - 1 source_name: mouse_testes 11 strain : DBA/ 2J 11 tissue: testes The metadata information can then be parsed from the sample_attribute column . To obtain more structured metadata , we can use an additional flag '--expand' : $ pysradb gse-to-gsm --desc --expand GSE41637 I head study_alias experiment_alias source_name strain tissue GSE41637 GSM1020640_1 mouse_brain dba / 2j brain GSE41637 GSM1020641_1 mouse_colon dba/ 2j colon GSE41637 GSM1020642_1 mouse_heart dba / 2j heart GSE41637 GSM1020643_1 mouse_kidney dba / 2j kidney GSE41637 GSM1020644_1 mouse_liver dba/ 2j liver GSE41637 GSM1020645_1 mouse_lung dba / 2j lung GSE41637 GSM1020646_1 mouse_skm dba/ 2j skeletal muscle 241 A.3.5 Getting SRR from GSM gsm-to-srr allows conversion from GEO experiments (accession prefix 'GSM ') to SRA runs (accession prefix 'SRR'): $ pysradb gsm-to-srr GSM1020640 GSM1020646 experiment_alias run_accession GSM1020640_1 SRR594393 GSM1020646_1 SRR594399 A.3.6 Downloading SRA datasets pysradb enables seemless downloads from SRA. It organizes the downloaded data following the NCBI hiererachy: 'SRP => SRX => SRR' of storing data. Each 'SRP' (project) has multiple 'SRX' ( experiments) and each 'SRX' in turn has multiple 'SRR' (runs) . Multiple projects can be downloaded at once using the download sub-command: 242 $ pysradb download -p SRP000941 -p SRP010679 download also allows Unix pipes-based inputs. Consider our previous example of the project SRP000941 with different assays. However, we want to be able to download only 'RNA-seq ' samples. We can do this by subsetting the metadata output for only 'RNA-seq ' samples: $ pysradb metadata SRP000941 --assay I grep 'studylRNA-Seq' I pysradb download This will only download the 'RNA-seq' samples from the project. A.4 Summary pysradb [297] provides a command-line interface to query metadata and download sequencing datasets from the SRA. It enables seamless retrieval of metadata and conversion between different accessions. pysradb is written in Python 3 and is available on Linux and Mac OS. The source code is hosted on GitHub and licensed under BSD 3-clause license. It is available for installation through PyPI and bioconda. A.5 Data availability Underlying data 243 Dataset from DDBJ Sequence Read Archive, Accession number DRP003075: https: //identifiers . org/insdc.sra/DRP003075 Dataset from EMBL-EBI Sequence Read Archive, Accession number ERP013565: https: / / identifiers . org/insdc.sra/ERP013565 Dataset from Gene Expression Omnibus, Accession number GSE24355: https: //identifiers. org/geo/GSE24355 Dataset from Gene Expression Omnibus, Accession number GSE25842: https: //identifiers . org/geo/GSE25842 Dataset from Gene Expression Omnibus, Accession number GSE100007: https: //identifiers . org/geo/GSE100007 ~0~ Dataset from Gene Expression Omnibus, Accession number GSE41637: https: //identifiers. org/geo/GSE41637 [306] Dataset from NCBI Sequence Read Archive, Accession number SRP010679: https: //identifiers. org/insdc . sra/SRP010679 [103] Dataset from NCBI Sequence Read Archive, Accession number SRP000941: https: //identifiers . org/insdc. sra/SRP000941 [307] A.6 Software availability Software available from: https: //pypi . org/project/pysradb/. 244 Source code available from: https: / / gi thub. com/ saketkc/pysradb. Archived source code at time of publication: https: / /doi . org/10 . 5281/zenodo . 2579446 [297]. License: BSD 3-Clause 245

Abstract (if available)

Abstract Any biological process is a manifestation of the protein levels. Protein are synthesized inside the body through the process of translation. Translation is one of the most energy-intensive processes and hence is highly regulated to ensure the proteins get synthesized in the right quantity in the cell. The advent of next-generation sequencing has enabled transcriptome-wide monitoring of translation regulation. In particular, Ribo-seq has provided us with a high-throughput sequencing method to profile mRNA undergoing active translation at sub-codon resolution. However, the resulting dataset has inherent noise arising from the sequencing of non-ribosomal and non-active translation fragments. Through this work, I develop a method to identify regions under active translation using Ribo-seq data. The method enables more accurate identification of both short and long open reading frames. I develop a database of uniformly processed public Ribo-seq datasets which have been appropriately de-noised to recover only the actively translating fragments. I also demonstrate the utility of our method as a quality control metric to asses the quality of Ribo-seq data generated in a variety of treatment and biological contexts. ❧ I utilized our method to understand the changes involved in translational landscape in two important biological contexts. First, I present a study on translational landscape changes involved in the irradiation of glioblastoma cell lines thereby providing evidence of radio-resistance in some target genes that can be likely candidates for increasing irradiation efficacy. Second, I provide evidence of translational landscape changes involved in the morphological transition of Candida albicans, a fungal pathogen involved in inducing diseases as Candidasis in immunocompromised patients. These target genes can serve as potential targets for anti-fungal drugs. ❧ Finally, by applying our method to identify actively-translating regions on public Ribo-seq datasets, I discovered thousands of upstream open reading frames under active translation across various biological contexts in multiple species. These upstream open reading frames (uORFs) often act as a repressive element for the downstream protein coding regions. Using Ribo-seq datasets from multiple species across different conditions, I characterize the sequence properties of a subset of these uORFs which are almost always under active translation. Further, I analyze it in an evolutionary context providing evidence for the conservation of the upstream sequence context.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Computational methods for translation regulation analysis from Ribo-seq data

PDF

Identification and analysis of shared epigenetic changes in extraembryonic development and tumorigenesis

PDF

Improved methods for the quantification of transcription factor binding using SELEX-seq

PDF

Genome-wide studies of protein–DNA binding: beyond sequence towards biophysical and physicochemical models

PDF

Efficient short-read sequencing on long-read sequencers

PDF

DNA shape at transcription factor binding sites: from purifying selection to a new alphabet

PDF

Genome-wide studies reveal the isoform selection of genes

PDF

Quantitative modeling of in vivo transcription factor–DNA binding and beyond

PDF

Probabilistic modeling and data integration to examine RNA-protein interactions

PDF

Detection, classification and functional annotation of mouse L1 retrotransposon promoters

PDF

Fast search and clustering for large-scale next generation sequencing data

PDF

Efficient algorithms to map whole genome bisulfite sequencing reads

PDF

Application of machine learning methods in genomic data analysis

PDF

Simulating the helicase motor of SV40 large tumor antigen

PDF

Model selection methods for genome wide association studies and statistical analysis of RNA seq data

PDF

Whole genome bisulfite sequencing: analytical methods and biological insights

PDF

Genome-wide studies reveal the function and evolution of DNA shape

PDF

Identifying allele-specific DNA methylation in mammalian genomes

PDF

A generalization of the accumulation curve in species sampling and its applications to high‐throughput sequencing

PDF

Data-driven approaches to studying protein-DNA interactions from a structural point of view

Asset Metadata

Creator Choudhary, Saket Kumar (author)

Core Title Comparative genomics of translational regulation

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Computational Biology and Bioinformatics

Publication Date 07/27/2020

Defense Date 07/24/2020

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag OAI-PMH Harvest,Ribo-seq,ribosome,ribosome profiling,translation,translation regulation

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Smith, Andrew D. (committee chair), Calabrese, Peter (committee member), Chaisson, Mark (committee member), Lu, Zhipeng (committee member), Rohs, Remo (committee member)

Creator Email saketkc@gmail.com,skchoudh@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-340422

Unique identifier UC11663555

Identifier etd-ChoudharyS-8780.pdf (filename),usctheses-c89-340422 (legacy record id)

Legacy Identifier etd-ChoudharyS-8780.pdf

Dmrecord 340422

Document Type Dissertation

Rights Choudhary, Saket Kumar

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA