Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
DNA methylation and gene expression profiles in Vidaza treated cultured cancer cells
(USC Thesis Other)
DNA methylation and gene expression profiles in Vidaza treated cultured cancer cells
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
DNA Methylation and Gene Expression Profiles
in Vidaza Treated Cultured Cancer Cells
by
Mengyu Liu
A Thesis Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(BIOSTATISTICS)
Graduation Date: Aug.11
th
,2015
Copyright 2015 Mengyu Liu
ii
DEDICATION
I dedicate this document to my family.
iii
ACKNOWLEDGMENTS
I would like to express my sincere gratitude to my mentor, Dr. Kim Siegmund for her guidance
and support for my thesis. I also extend my gratitude to my thesis committee for their insight
and support. I thank Dr. Joe Hacia for providing the dataset, his guidance and biological
expertise, and Dr. Dan Stram for providing guidance and advice on presenting the results.
iv
TABLE OF CONTENTS
DEDICATION ii
ACKNOWLEDGMENTS iii
TABLE OF CONTENTS iv
LIST OF TABLES v
LIST OF FIGURES vi
ABSTRACT viii
CHAPTER 1: INTRODUCTION 1
CHAPTER 2: METHODS 3
2.1 Methylation Profile 3
2.2 Expression Profile 6
2.3 Linked DNA Methylation and Gene Expression 6
CHAPTER 3: RESULTS 9
3.1 DNA Methylation Profile 9
3.2 Expression Profile 21
3.3 Linked DNA Methylation and Gene Expression 32
CHAPTER 4: DISCUSSION AND CONCLUSIONS 40
REFERENCES 43
v
LIST OF TABLES
Table 1 Experimental design of gene expression analysis 3
Table 2 Cut-off of beta values for defining different DNA methylation patterns over time 4
Table 3 Counts of Probes by genomic context 11
Table 4 (a-c) Counts of methylation probes for varied patterns based on beta value cut-offs
given in Table 2. 12
Table 5 Counts and percentage of negatively significant demethylated genes and positively
significant increased expression genes of five main patterns for two probe genes 38
vi
LIST OF FIGURES
Figure 1 Distribution of distance to TSS in base pairs .................................................................... 7
Figure 2 Distribution of DNA methylation beta values for all samples .......................................... 9
Figure 3 Number of probes for genes in methylation data .......................................................... 10
Figure 4 (a,b) Venn diagram for significant methylation probes (by proportion and unscaled, BH-
adjusted p<0.05) ........................................................................................................................... 14
Figure 5(a,b) Venn diagram for negatively significant methylation probes (by proportion and . 14
Figure 6 (a-c) Histogram of log2 fold change for vidaza-treated samples compared to mannitol-
treated samples ............................................................................................................................ 15
Figure 7 Histogram of raw p-values for vidaza 5µM vs mannitol samples ................................... 17
Figure 8 (a,b,c) Volcano plots for 10µM, 5µM, and 1 µM vidaza-treated samples (raw p-values
from pairwise comparison without days). Horizontal lines represent p=0.05 cut-off, and vertical
lines represent positive and negative 2 fold change cut-off. ....................................................... 18
Figure 9 Venn diagram for probes showing differential trend over time compared to mannitol-
treated group (by proportion and unscaled, BH-adjusted p<0.05) ............................................. 19
Figure 10 Mean Beta values for the three most significant probes among different samples over
days (based on BH adjusted p-values from pairwise treatment comparison) ............................. 21
Figure 11 MDS plot for all samples ............................................................................................... 22
Figure 12 Heatmap of the top 500 most variable genes .............................................................. 23
Figure 13 (a,b) Venn diagram for differentially expressed probes (by proportion and unscaled,
BH-adjusted p<0.05). .................................................................................................................... 24
vii
Figure 14 (a,b) Venn diagram for probes showing increased expression after drug treatment (by
proportion and unscaled, BH-adjusted p<0.05). .......................................................................... 25
Figure 15 (a-c) Histogram of log2 fold change for vidaza-treated samples compared to mannitol-
treated samples ............................................................................................................................ 26
Figure 16 Histogram of raw p-values comparing expression in vidaza 5µM vs mannitol group . 27
Figure 17 (a-c) Volcano plots for 10µM, 5µM, and 1 µM vidaza-treated sample (raw p-values
from pairwise comparison). Horizontal lines represent p=0.05 cut-off and vertical lines
represent positive and negative 2 fold change cut-off................................................................. 28
Figure 18 (a-c) Mean expression values for the three most significant probes showing
differential expression comparing 5µM vidaza vs mannitol control. ........................................... 29
Figure 19 (a-c) Expression value for three probes selected from clusters in the volcano plots. . 30
Figure 20(a,b) Venn diagrams for significant expression probes showing differential treatment
effects over time in the trend model (by proportion and unscaled) ........................................... 31
Figure 21 (a,b) Venn diagram for methylation and expression (by proportion and unscaled) .... 33
Figure 22 (a,b) Starburst plot of the BH-adjusted p-values for gene-linked probes (in log10
scales) ............................................................................................................................................ 34
Figure 23 Pie graph for patterns of methylation genes that have two probes ............................ 35
Figure 24 Five main gene patterns for the two-probe genes ....................................................... 35
Figure 25 DNA methylation and gene expression values for three matched genes (left:
methylation, right: expression) ..................................................................................................... 37
viii
ABSTRACT
The methylation of DNA is an epigenetic modification that can play an important role in
the control of gene expression in mammalian cells. The genes involved include tumor
suppressor genes, genes that suppress metastasis and angiogenesis, and genes that repair DNA
suggesting that epigenetics plays an important role in tumorigenesis. I studied the genome-
wide DNA methylation, and gene expression profiles in HL60 cells following 5-azacytidine
(Azacitidine, Vidaza
TM
), using microarray technologies. Three doses of vidaza (1µM,5 µM,10
µM) and mannitol (control) were used and DNA methylation and gene expression were
measured in Day 1,Day 3 and Day 5. Quality control analysis was conducted, followed by two
statistical analyses to study the effects of drug treatment on the methylation profile and
expression profile. In the first analysis we compared the average DNA methylation (gene
expression) levels between treatment groups (‘pairwise treatment comparison’) and in the
second analysis we compared the treatment-specific change in average levels over time (trend
comparison). A decrease in DNA methylation was found in all three doses of vidaza-treated
samples compare to the mannitol treated samples and the largest number of differentially
methylated probes were found in lowest doses (1µM vidaza). Moreover, gene expression
values also changed, with a similar number of probes with increased or decreased expression
identified. Furthermore, for DNA methylation profiles, genes with multiple probes were studied
by genomic context. For the majority of genes that contain two probes, a large number of
genes with non-CpG islands promoters is significantly changed after drug treatment (BH-
adjusted p<0.05, fold-change (FC) >1 or <-1). However, only a small number of genes showed
ix
concomitant changes in gene expression. This study provides insights about changes in DNA
methylation induced by vidaza and potential changes of gene expression for those same genes.
1
CHAPTER 1: INTRODUCTION
Epigenetic research is increasingly recognizing the importance of DNA methylation in
cellular neoplastic development. DNA methylation occurs when a methyl group is added to
cytosine at palindromic CpG dinucleotides by the DNA methyltransferase enzymes [1].
Azacitidine (5-azacytidine, Vidaza
TM
) is a DNA methylation inhibitor with FDA approval
for treatment of all subtypes of myelodysplastic syndrome and has off-label efficacy in other
hematologic malignancies, including acute myeloid leukemia (AML) and chronic myeloid
leukemia (CML) [2].
It is commonly known that inactivation of certain tumor-suppressor genes occurs as a
consequence of hypermethylation within the promoter regions and numerous studies have
demonstrated a broad range of genes silenced by DNA methylation in different cancer types.
Apart from DNA methylation alterations in promoter regions and repetitive DNA sequences,
this phenomenon is associated also with regulation of expression of noncoding RNAs such as
microRNAs that may play role in tumor suppression [3]. As epigenetic alterations have gained
acceptance as legitimate players in neoplastic pathogenesis, there has been renewed interest in
therapies that can reverse deleterious hypermethylation.
Vidaza's anticancer effects are believed to be twofold. One way that it works is by
demethylation or interfering with the methylation of DNA. When incorporated into DNA,
azacitidine covalently binds DNA methyltransferase, resulting in inhibited methylation [3]. By
this process of demethylation, normal function to the tumor suppressor genes is restored, thus
restoring control over cell growth. Azacitidine also belongs to the category of chemotherapy
called antimetabolites. Antimetabolites are very similar to normal substances within the cell.
2
When the cells incorporate these substances into the cellular metabolism, they interact with a
number of targets within the cell to produce a direct cytotoxic effect that causes death of
rapidly dividing cancer cells.
It is of considerable interest to explore the inter-relationships between DNA methylation
and miRNA transcription profiles of vidaza-treated cancer cells. HL60 cell lines were available in
the lab and were used to do experiment. In this thesis, I analyzed DNA methylation data
(measured by Illumina Infinium HumanMethylation27 BeadChip) and RNA data (measured by
Affymetrix U133A 2.0 microarrays) and integrated the two profiles by Entrez gene ID.
Treatments were designed to result in a range of promoter demethylation ranging from 10 to
over 70 percent. Likely, the higher end will be accomplished when longer treatment times are
used and the cells have divided at least two times during the course of treatment.
3
CHAPTER 2: METHODS
All data were provided by Dr. Joseph Hacia. The experimental design is described in
Table 1[4] and all experiments were performed in his lab. Samples from HL60 cell line were
collected and assigned into one of three groups: control group, Mannitol-treated control group
and Vidaza-treated group. Three concentrations of Vizada (1 µM, 5 µM and 10 µM) were
administered and cells from all treatment groups were analyzed at three different time points
(Day1, Day3 and Day5). A Mannitol-treated control group was run separately as all cells treated
with Vidaza were also treated with Mannitol. DNA and RNA from the untreated and drug-
treated samples was isolated. All experiments were performed in triplicate to minimize
platform-specific noise. Three samples were run in duplicate on two chips, which resulted in 48
samples in total.
Table 1 Experimental design of gene expression analysis
Cell line
Origin
Experiments treatment concentration
time
points(days)
times of
experiments
HL 60 Leukemia
Control 1,3,5 9
Mannitol 1,3,5 9
Vidaza
1 µM,5 µM, 10
µM 1,3,5 27
total numbers of experiments:45
2.1 Methylation Profile
The DNA samples were prepared following the guidelines suggested by the manufacturer
(Illumina, Inc.), and then measured by Illumina Infinium HumanMethylation27 BeadChip, which
measures 27578 CpG sites. The Beta-value, which is the ratio of the methylated probe intensity
over the overall intensity (sum of methylated and unmethylated probe intensities) [5] was
4
obtained and used in the analyses. Density plots for all samples were generated to study the
consistency of the three replicates for each sample. Beta-values were converted to M-values
(the log2 ratio of the intensities of methylated probe versus unmethylated probe) for the
differential analysis of methylation levels [5]. Numbers of probes for methylation genes were
summarized. Probe locations were assigned into six categories (A=promoter CpG islands;
B=promoter non-CpG islands; C=exon plus intron transcribed CpG-island; D= exon plus intron
transcribed non CpG-island; E=intron non-transcribed CpG islands; F=intron non-transcribed
non-CpG islands) to study methylation and linked methylation and expression profile by
different locations.
Probes were classified based on their pattern of DNA methylation over the three time
points. Three trends of methylation (linear, immediate decrease and late decrease) were
identified, and within each main pattern, three sub-patterns (strict, intermediate and
permissive) were identified. The beta value cut-offs used for defining the different patterns are
listed below in Table 2.
Table 2 Cut-off of beta values for defining different DNA methylation patterns over time
Categories Day 1 Day 3 * Day 5 **
Linear
Strict >0.8 0.4 - 0.6 <0.2
Intermediate >0.75 0.4 - 0.6 <0.25
Permissive >0.7 0.4 - 0.6 <0.3
immediate
decrease
Strict >0.8 <0.2 <0.2
Intermediate >0.8 <0.25 <0.2
Permissive >0.8 <0.3 <0.2
late decrease
Strict >0.8 >0.8 <0.2
Intermediate >0.8 >0.75 <0.2
Permissive >0.8 >0.7 <0.2
* Out of probes meeting criteria for Day 1
** Out of probes meeting criteria for Day 1 and Day 3
5
Statistical models were built to study the effects of drug treatment. Vidaza-treated
samples and mannitol-treated samples were compared using linear regression models fit using
the LIMMA package [6] in the R programming language version 3.1.0. Pairwise comparisons
were conducted between the three doses of vidaza-treated samples and mannitol-treated
samples and moderated t-statistics and moderated F-statistic were computed by empirical
Bayes [6]. Raw p-values for each of three comparisons were obtained and volcano plots were
generated to provide visual identification of probes that display large-magnitude differences
with small p-values. Venn diagrams were generated to show the numbers of significant probes
for the three treatment group comparisons after p-values were adjusted by the Benjamini-
Hochberg approach (BH) to control false discovery rate [15]. Log2 fold change of methylation
values between vidaza-treated and mannitol-treated samples were studied to provide cut-offs
used in the analysis linking the methylation and expression data.
In a second analysis, we studied the change of DNA methylation over days in the vidaza-
treated samples compared to the mannitol-treated samples. For each probe, we fit a linear
regression model with methylation M value as the outcome and Day (1,3,5), treatment and
pairwise interactions between day and treatment as predictor variables. Parameter estimates,
moderated t-statistics, and p-values were obtained for each probe. BH-adjusted p-values for
the three interaction terms were collected and counts of significant p-values (BH-adjusted
p<0.05) were summarized in a Venn diagram. Lastly, results from the two statistical models
were compared.
6
2.2 Expression Profile
Gene expression data was measured by Affymetrix U133A 2.0 microarrays targeting a
total of 48783 assays. Control sample and drug-treated expression data were read into R
(version 3.1.0) using LIMMA. Quality control analyses were done to check for poor performance
of the arrays. Data processing was performed in LIMMA using background correction and
quantile normalization [7] including negative and positive control probes to get low bias, low
variance estimates of gene expression data. Similar to the analysis of methylation data, gene
expression data for the control samples, mannitol-treated control samples and vidaza-treated
samples were fitted using linear models and LIMMA. Pairwise comparisons were conducted
between the three different vidaza-treated groups and mannitol control. Raw p-values for each
of three comparisons were obtained and volcano plots were generated. Venn diagrams were
generated to show the number of significant probes for the three comparisons (BH-adjusted
p<0.05). Log2 fold change of mean expression between vidaza-treated probes and mannitol
probes generated and the distributions of fold change were studied to provide cut-offs used in
the analysis linking the methylation and expression data. In addition, a trend linear regression
model was generated to study the changes of expression over days among vidaza-treated
samples compared to mannitol-treated samples for each probe.
2.3 Linked DNA Methylation and Gene Expression
Expression and methylation probes were linked by Entrez gene ID. Any single gene can
have multiple probes that target different genomic positions. For the DNA methylation data,
the distribution of distance from the target CpG to the transcription start site (TSS) for all the
7
methyaltion probes were plotted and shown in Figure 1. It can be seen that the distance to TSS
for all probes is smaller than 1500 basepairs.
Figure 1 Distribution of distance to TSS in base pairs
Single-probe linked gene
To start, we studied a single probe from each gene. The probes were selected as follows.
For the expression data, the probe with the largest variance in log2 expression value was
chosen since we assume that those probes can show larger effects of treatment. For the
methylation data, the probe having the minimum distance to the transcription start site (TSS)
was chosen. A starburst plot of the BH-adjusted p-values from the pairwise treatment
comparisons of methylation data and expression data was plotted to show the relationship
between DNA methylation differences with gene expression differences [14]. All treatment
groups were picked to compare with mannitol groups and the direction of changes were
determined by the signs of t-statistics from the pairwise comparison model.
8
Multiple-probe methylation genes
Next, we identified all probes in the DNA methylation data from genes with expression
data and annotated them by genomic context. Counts of significant genes (adjusted p<0.05
based on vidaza 1µM vs mannitol samples and log FC>=2 or <=-2 from pairwise comparison
model) were summarized among those linked methylation genes and expression genes by the
main genes patterns for the majority types of the multi-probe genes found.
9
CHAPTER 3: RESULTS
3.1 DNA Methylation Profile
Density plots of methylation beta values for all the samples are displayed in Figure 2. It is
shown that mannitol-treated and untreated samples are very similar. Vidaza-treated samples
lost methylation over days compared to mannitol-treated samples. 1µM vidaza-treated samples
lost methylation more noticeably than the 5 µM vidaza-treated samples and 10 µM vidaza-
treated samples compared to mannitol-treated samples. High doses of drug can kill cancer cells
so that the 5 µM and 10 µM samples may not lose methylation due to death of cells.
Figure 2 Distribution of DNA methylation beta values for all samples
10
Upper left: density plot of beta value for untreated and mannitol-treated samples;
Upper right: density plot of beta value for 1µM vidaza-treated samples;
Lower left: density plot of beta value for 5µM vidaza-treated samples;
Lower right: density plot of beta value for 10µM vidaza-treated samples.
Numbers of probes for those methylation genes were displayed in Figure 3. Probe
locations were assigned into six categories and counts for probes by locations were summarized
in Table 3. Since the majority of multiple probe genes in Table 3 have one or two probes, gene
patterns for those probes were further summarized. Methylation probes (total 27578 probes)
are filtered by the two-fold change of 1µM vidaza vs. mannitol group and significance of
adjusted p-values (by BH-adjusted p<0.05). There are in total 12410 probes significantly
differentially methylated. Specifically, 1057 probes are composed of genes that have only one
probe and 9233 probes are composed by genes that have two probes.
Figure 3 Number of probes for genes in methylation data
2237
10411
93 122
0
2000
4000
6000
8000
10000
12000
1 2 3~5 >5
counts
number of probes
11
Table 3 Counts of Probes by genomic context
Genomic Context Number of probes Percent (%)
CpG island
Promoter[A] 4628 16.8
exon and intron transcribed (not
promoter)[C]
6825 24.7
intron non-transcribed[E] 84 0.31
sub total 11537 41.8
non-CpG island
Promoter[B] 9764 35.4
exon and intron transcribed (not
promoter) [D]
6119 22.2
intron non-transcribed[F] 158 0.57
sub total 16041 58.2
total 27578 1
Average DNA methylation levels over time were studied and Table 4 (a-c) reported for
three treatments (Mannitol, 1µM Vidaza, 5µM Vidaza, respectively) the number of probes out
of 27578 achieving different levels of DNA methylation. Cut-offs used to define different
patterns over time were given in Table 2. Table 4 showed that the largest number of probes
with beta value>0.8 on Day 1 were found in the mannitol treated samples and the smallest
number of probes were found in 1uM vidaza-treated sample. These indicate that
demethylation for vidaza 1uM treated samples occurred in less than one day. For the mannitol-
treated (control) samples, none of the probes that showed decreasing methylation across the 5
days. Only a single probe showed a linear decrease between days 1 and 5 in the 1µM Vidaza-
treated group.
12
Table 4 (a-c) Counts of methylation probes for varied patterns based on beta value cut-offs
given in Table 2.
Table 4a.
Methylation
trend in
mannitol
treated
samples patterns Day1 Day3* Day5**
Linear
Strict 5591 0 0
Intermediate 7474 0 0
Permissive 8704 0 0
Immediate
Decrease
Strict 5591 0 0
Intermediate 5591 0 0
Permissive 5591 0 0
Late
Decrease
Strict 5591 5379 0
Intermediate 5591 5561 0
Permissive 5591 5562 0
Table 4b.
Methylation
trend dose 1
(treated 1µM
Vidaza) patterns Day1 Day3* Day5 **
Linear
Strict 205 0 0
Intermediate 621 17 0
Permissive 1495 125 1
Immediate
Decrease
Strict 205 0 0
Intermediate 205 0 0
Permissive 205 0 0
Late Decrease
Strict 205 25 0
Intermediate 205 82 0
Permissive 205 131 0
13
Table 4c.
Methylation
trend dose 1
(treated 5µM
Vidaza) patterns Day1 Day3 * Day5 **
Linear
Strict 3846 1 0
Intermediate 5880 47 0
Permissive 7526 314 0
Immediate
Decrease
Strict 3846 0 0
Intermediate 3846 0 0
Permissive 3846 0 0
Late Decrease
Strict 3846 1308 0
Intermediate 3846 2367 0
Permissive 3846 3262 0
* Out of probes meeting criteria for Day 1
** Out of probes meeting criteria for Day 1 and Day 3
Pairwise comparisons of average DNA methylation were conducted for each dose of
vidaza-treated samples compared to mannitol treated samples. Overall differences between
three doses of vidaza treated samples and mannitol samples were tested in a simple regression
model without adjustment for day. Venn diagrams in Figure 4 (a,b) display the numbers of
probes with altered methylation in each comparison. More specifically, 4813 probes were
significantly changed for all three doses of vidaza-treated samples compared to mannitol
samples (adjusted p-values<0.05). 5404 probes were significantly changed for both 5µM vidaza-
treated samples, and 1µM vidaza-treated samples compared to mannitol samples (adjusted p-
values<0.05). 4678 probes were significantly changed only for 1µM vidaza-treated samples
compared to mannitol samples (adjusted p-values<0.05). Probes that changed negatively in
DNA methylation were summarized in Figure 5(a,b). Specifically, after drug treatment, DNA
methylation in 18489 probes is lower than in the mannitol-treated samples (adjusted p<0.05),
14
of which 15051 probes were found in 1µM vidaza-treated samples. These indicate that there is
a decrease in global DNA methylation in three doses of vidaza-treated samples compare to
mannitol treated samples with the greatest decrease (most hypomethylated probes) found in
the lowest dose (1µM vidaza).
Figure 4 (a,b) Venn diagram for significant methylation probes (by proportion and unscaled,
BH-adjusted p<0.05)
Figure 4a Figure 4b
Figure 4a. yellow, green, purple circles are for comparing 10 µM Vidaza to mannitol (7792 probes), 5µM
vidaza to mannitol (12527 probes), and 1µM vidaza to mannitol (15051 probes), respectively.
Figure 4b. Numbers of significant probes (BH-adjusted p<0.05) comparing Vidaza treated cells to
samples treated by mannitol.
Figure 5(a,b) Venn diagram for negatively significant methylation probes (by proportion and
unscaled. BH-adjusted p<0.05)
Figure 5a Figure 5b
Figure 5a. green, blue, purple circles are for comparing 10 µM Vidaza to mannitol (4233 probes), 5µM
vidaza to mannitol (9462 probes), and 1µM vidaza to mannitol (13813 probes), respectively.
15
Figure 5b. Numbers of negatively significant probes (BH-adjusted p<0.05) comparing Vidaza treated cells
to samples treated by mannitol.
Log2 fold change of mean M value between vidaza-treated probes and mannitol probes
were generated from the regression model and histograms were shown in Figure 6 (a-c). Two
modes of changes of mean M values can be seen from histograms for vidaza 1 µM and 5 µM
groups compared to mannitol-treated groups. More probes are demethylated than methylated
and a cut-off of -0.7 can be used to distinguish the two modes of plots.
Figure 6 (a-c) Histogram of log2 fold change for vidaza-treated samples compared to
mannitol-treated samples
Figure 6a Figure 6b
16
Figure 6c
Figure 6a. Histogram of log2 fold change for 10 µM vidaza-treated samples compared to mannitol-
treated sample; Figure 6b. Histogram of log2 fold change for 5µM vidaza-treated samples compared to
mannitol-treated sample; Figure 6c. Histogram of log2 fold change for 1µM vidaza-treated samples
compared to mannitol-treated sample.
The histogram of raw p-values for 5uM vidaza-treated groups compared to mannitol-
treated groups from the pairwise comparison model is showed in Figure 7 and a large number
of significant p-values were observed. We used the natural cubic spline approach proposed by
Storey and Tibshirani [8] to estimate the proportion of true null tests and found it to be 24%.
This suggests that 76% of features are truly differentially methylated. Proportions of truly
significant features for 1 uM vidaza-treated groups and 10uM vidaza-treated groups compared
to mannitol-treated groups were estimated to be 77% and 71%, respectively.
17
Figure 7 Histogram of raw p-values for vidaza 5µM vs mannitol samples
Volcano plots for three doses of vidaza-treated samples were generated to give an
overview of interesting probes. Interesting probes have significant p-values (adjusted by BH)
and large fold changes. A larger number of probes that have negative fold changes were found
for 5 µM vidaza-treated and 1 µM vidaza-treated samples showed in Figure 8 (a,b,c). For 10 µM
vidaza-treated samples, methylation of 11076 (40.2%) probes increased and 14412 (52.3%)
probes decreased. For 5 µM vidaza-treated samples, methylation of 10275 (37.3%) probes
increased and 15217 (55.2%) probes decreased. For 1 µM vidaza-treated samples, methylation
of 7822 (28.4%) probes increased and 17691 (64.1%) probes decreased.
18
Figure 8 (a,b,c) Volcano plots for 10µM, 5µM, and 1 µM vidaza-treated samples (raw p-values
from pairwise comparison without days). Horizontal lines represent p=0.05 cut-off, and
vertical lines represent positive and negative 2 fold change cut-off.
Figure 8a Figure 8b
Figure 8c
Figure 8a Volcano plots for 10µM vidaza-treated samples in methylation profile; Figure 8b Volcano plots
for 5µM vidaza-treated samples in methylation profile; Figure 8c Volcano plots for 1 µM vidaza-treated
samples in methylation profile.
We subsequently studied changes over days using a linear trend model. The interaction
term of Day and treatment represent the additional difference in mean methylation M-value of
vidaza-treated samples compared to mannitol-treated samples on Day1, Day3 and Day5. P-
values (adjusted by BH) for the interaction terms comparing the three doses of vidaza to
19
mannitol-treated samples were collected and counts of significant p-values (p<0.05) were
summarized in the Venn diagram shown in Figure 9 (a,b). Results from the pairwise comparison
model for treatment effects (unadjusted for days) and trend model that sought treatment-
related changes over time were different. Significant probes of three vidaza-treated samples in
the trend model are mostly overlapped, while most significantly changed probes were captured
in 1 µM treated samples in the pairwise comparison model. In terms of the fact that lots of
probes in 1uM vidaza-treated samples were demethylated within one day (presented in table
4b), current scale of days (1,3,5) could not capture changes the occurred within one day, so that
the trend model presents probes share the same type of changes. Unique changes, such as
probes in 1 µM vidaza-treated samples were presented in the pairwise comparison model that
didn’t consider days.
Figure 9 Venn diagram for probes showing differential trend over time compared to mannitol-
treated group (by proportion and unscaled, BH-adjusted p<0.05)
Figure 9a Figure 9b
Figure 9a. yellow, blue, purple circles are for interaction of 10 µM Vidaza and Day (9342 probes), 5µM
vidaza and Day (11000 probes), and 1µM vidaza and Day (11731 probes), respectively.
Figure 9b. Numbers of significant probes (BH-adjusted p<0.05) for interactions terms in the trend model
20
For those 8243 probes with a differential methylation trend over time for each vidaza
dose compared to the mannitol-treated group, 3529 (42.8%) probes were also found to have
differential methylation in the pairwise treatment comparison for each dose vs mannitol
control (unadjusted for days), and 3908 (46.4%) probes were found differentially methylated in
the vidaza 1µM and 5 µM categories. This indicates that for probes in 10 µM vidaza treated-
samples and over half of probes in 1µM and 5µM treated samples, changes over days were not
captured in pairwise comparison model. For the 1699 probes showing a differential trend over
time for both vidaza 1µM and vidaza 5µM groups, 688 (40.5%) probes were also found
differentially methylated in both the vidaza 1µM and vidaza 5µM pairwise treatment vs control
comparison, and 843 (49.6%) probes were found differentially methylated for only the vidaza
1µM category. This indicates that for some 1µM vidaza-treated probes, mean differences were
not significant but changes over days were significant and captured in pairwise comparison
model since many probes were demethylated within one day. For the 1595 only vidaza 1µM
significant probes in the trend model, 1193(74.8%) probes were also found significant in the
vidaza 1µM category from the pairwise comparison model, and 162(10.2%) probes were found
significant for both the vidaza 1µM and vidaza 5µM groups.
Mean beta values for the three most significant probes (BH adjusted p<0.05 based on
5µM vidaza vs mannitol groups) were plotted by treatment group over days and shown in
Figure 10 (a,b,c). Both demethylation and hypermethylation can happen for those probes after
treatment of vidaza. It seems that beta values for vidaza-treated samples on Day1 were similar
but separated on day 3 and day 5 for the two shown demethylated probes, which indicates that
effects of drug were increasing as time increased.
21
Figure 10 Mean Beta values for the three most significant probes among different samples
over days (based on BH adjusted p-values from pairwise treatment comparison)
Figure 10a Figure 10b
Figure 10c
Figure 10a. Mean beta value for gene cg24070292; Figure 10b. Mean beta value for gene cg24713204;
Figure 10c. Mean beta value for gene cg 26399113
Black lines : vidaza 1µM, red lines: vidaza 5µM, green lines: vidaza 10 µM, blue lines: mannitol for all
three figures
3.2 Expression Profile
A multidimensional scaling (MDS) plot was generated to provide a visual representation
of the pattern of proximities (i.e. similarities or distances) among all 48 vidaza-treated,
mannitol-treated and untreated samples. Figure 11 shows Mannitol-treated samples and
untreated samples are clustered and they are apart from vidaza-treated samples, which
indicates a substantial difference in expression between vidaza-treated samples and
22
untreated/mannitol-treated samples. A heatmap of the 500 most variant genes in Figure 12
shows more red and yellow bars (lower left) indicating higher expression in the vidaza-treated
samples. The figure shows that some clusters of vidaza-treated samples have higher gene
expression than the untreated and mannitol-treated samples.
Figure 11 MDS plot for all samples
23
Figure 12 Heatmap of the top 500 most variable genes
Pairwise comparisons were conducted for each dose of vidaza-treated samples relative
to mannitol-treated samples. First, we describe the overall differences between three doses of
vidaza-treated samples and mannitol samples without adjusting for time. The venn diagrams in
Figure 13 display the numbers of probes with altered expression in each comparison.
Expression values of 10 µM vidaza-treated samples showed more changes than the lower doses
of treatment samples, with 8747 (17.9%) probes changed. 7051 probes of 5 µM vidaza-treated
samples and 2283 probes of 1µM vidaza-treated samples showed differential expression
compared to in the mannitol samples. Moreover, 1513 probes showed significantly altered
expression for all three doses of vidaza-treated samples compared to mannitol samples
(adjusted p-values<0.05). An additional 4530 probes showed differential expression for both
5µM vidaza-treated samples, and 10µM vidaza-treated samples compared to mannitol samples
(adjusted p-values<0.05). Lastly, 2559 probes showed significant differences only for 10µM
24
vidaza-treated samples compared to mannitol samples (adjusted p-values<0.05). Probes that
changed positively in gene expression were summarized in Figure 14 (a,b). For 4918 probes,
gene expression increased after drug treatment compared to samples treated with mannitol. A
total of 4090 probes with increased expression were found in 10µM vidaza-treated samples.
These indicate that gene expression for drug-treated samples compared to mannitol treated
samples was altered.
Figure 13 (a,b) Venn diagram for differentially expressed probes (by proportion and unscaled,
BH-adjusted p<0.05).
Figure 13a Figure 13b
Figure 13a. yellow, green, purple circles are for comparing 10 µM Vidaza to mannitol (8747 probes),
5µM vidaza to mannitol (7051 probes), and 1µM vidaza to mannitol (2283 probes), respectively.
Figure 13b. Numbers of significant probes (BH-adjusted p<0.05) comparing Vidaza treated cells to
samples treated by mannitol.
25
Figure 14 (a,b) Venn diagram for probes showing increased expression after drug treatment
(by proportion and unscaled, BH-adjusted p<0.05).
Figure 14a Figure 14b
Figure 14a. yellow, green, purple circles are for comparing 10 µM Vidaza to mannitol (4090 probes),
5µM vidaza to mannitol (3413 probes), and 1µM vidaza to mannitol (1042 probes), respectively.
Figure 14b. Numbers of significant probes (BH-adjusted p<0.05) comparing Vidaza treated cells to
samples treated by mannitol.
Log2 fold change (FC) of expression values between vidaza-treated and mannitol-treated
samples were generated from the regression model and histograms plotted. Histograms of the
changes in mean expression values for 1 µM, 5 µM and 10 µM vidaza-treated samples
compared to mannitol-treated samples were shown in Figure 20 (a-c). The distributions of log2
FC changes were symmetric above and below 0 for all three vidaza-treated samples.
26
Figure 15 (a-c) Histogram of log2 fold change for vidaza-treated samples compared to
mannitol-treated samples
Figure 15a Figure 15b
Figure 15c
Figure 15a. Histogram of log2 fold change for 10 µM vidaza-treated samples compared to mannitol-
treated sample; Figure 15b. Histogram of log2 fold change for 5µM vidaza-treated samples compared to
mannitol-treated sample; Figure 15c. Histogram of log2 fold change for 1µM vidaza-treated samples
compared to mannitol-treated sample.
Figure 16 shows the histogram of raw p-values from differential expression analysis
comparing 5µM vidaza-treated samples to mannitol-treated samples. From the distribution of
p-values we estimated the percentage of true null tests to be 71%, and the percentage of true
differential tests to be 29%. The estimated percentage of truly differential features for 1 µM
27
vidaza-treated groups and 10µM vidaza-treated groups compared to mannitol-treated groups
were found to be 23% and 29%, respectively.
Figure 16 Histogram of raw p-values comparing expression in vidaza 5µM vs mannitol group
Volcano plots for three doses of vidaza-treated samples were generated to give an
overview of interesting probes (Figure 17 a-c). Probes with significant p-values (adjusted by BH)
and large fold changes were of greatest interest. Among all 48783 probes, for 10 µM vidaza-
treated samples, expression value of 22171 (45.4%) probes increased and 26612 (54.6%)
probes decreased. For 5 µM vidaza-treated samples, expression value of 21647 (44.4%) probes
increased and 27163 (55.7%) probes decreased. For 1 µM vidaza-treated samples, expression
value of 22186 (45.5%) probes increased and 26597 (54.5%) probes decreased.
28
Figure 17 (a-c) Volcano plots for 10µM, 5µM, and 1 µM vidaza-treated sample (raw p-values
from pairwise comparison). Horizontal lines represent p=0.05 cut-off and vertical lines
represent positive and negative 2 fold change cut-off.
Figure 17a Figure 17b
Figure 17c
Figure 17a 10µM vidaza-treated vs mannitol-treated samples; Figure17b. 5µM vidaza-treated vs
mannitol-treated samples; Figure 17c. 1 µM vidaza-treated vs mannitol-treated samples.
Mean expression values for the three most significant probes (BH-adjusted p for vidaza
5µM vs mannitol groups) were plotted by treatment group over days and shown in Figure 18 (a-
c). Vidaza-treated samples can have higher or lower expression than the mannitol-treated
samples. It seems that expression values for 1µM Vidaza-treated samples were separated from
5 µM vidaza-treated samples and they were getting closer with days. Thus a more complicated
29
pairwise comparison model with days was tested. In this model, each dose of drug-treated
samples on Day1, Day3 and Day5 were compared to the same day mannitol-treated samples.
Figure 18 (a-c) Mean expression values for the three most significant probes showing
differential expression comparing 5µM vidaza vs mannitol control.
Figure 18a Figure 18b
Figure 18c
Figure 18a. Mean expression value for gene ILMN_1689702; Figure 18b. Mean expression value for gene
ILMN_2090782; Figure 18c. Mean expression value for gene ILMN_1679929
Black lines : vidaza 1µM, red lines: vidaza 5µM, green lines: vidaza 10 µM, blue lines: mannitol for all
three figures
All three volcano plots (Figure 17) showed interesting clusters composed of probes with
large fold change but not very small p-values. In order to explore patterns for those features,
expression values for three of those probes were plotted and shown in Figure 19 (a-c). It seems
that expression values for vidaza-treated samples increased with days compared to mannitol-
30
treated samples. Those odds points observed from volcano plots can be explained by the facts
that the variation in expression over days was not modeled in the pairwise treatment
comparison resulting in high variance estimates and decreasing the test statistic and increasing
the p-value for some features with otherwise large effect sizes.
Figure 19 (a-c) Expression value for three probes selected from clusters in the volcano plots.
Figure 19a Figure 19b
Figure 19c
Figure 19a. Expression value for gene ILMN_1708728; Figure 19b. Expression value for gene
ILMN_2399363; Figure 19c. Expression value for gene ILMN_1715638.
Black lines : vidaza 1µM, red lines: vidaza 5µM, green lines:vidaza 10 µM, blue lines: mannitol for all
three figures
31
Next, changes over days were studied in the trend model. The interaction term of Day
and treatment represents an increasing (or decreasing) difference in mean expression of vidaza
treated sample compared to mannitol-treated samples over time. Figure 20 a,b shows a Venn
diagram summarizing the number of statistically significant interaction terms for the three
vidaza dose comparisons (BH-adjusted p<0.05). We found 653 probes had significantly changed
expression levels for all three doses of vidaza-treated samples compared to mannitol samples,
1599 probes showed significantly changed expression uniquely for 10µM vidaza-treated
samples, and 1485 probes showed significant changes for both 10µM vidaza-treated and 5µM
vidaza-treated samples.
Figure 20(a,b) Venn diagrams for significant expression probes showing differential treatment
effects over time in the trend model (by proportion and unscaled)
Figure 20a Figure 20b
Figure 20a. Yellow, green, purple circles are for interaction of 10 µM Vidaza and Day (3814 probes), 5µM
vidaza and Day (3376 probes), and 1µM vidaza and Day (1538 probes), respectively.
Figure 20b. Numbers of significant probes (BH-adjusted p<0.05) for interactions terms in the trend
model
32
Results from the pairwise comparison of treatment groups and the trend model that
looked for varying treatment differences over time were compared. For those 653 probes
found to have differential treatment effects over time for all vidaza-treated groups compared
to mannitol control, 472 (72.3%) probes were also found differentially expressed in the vidaza
10µM and 5 µM pairwise comparisons. For the 1599 only vidaza 10µM significant probes in the
trend model, 222(13.9%) probes were also found significant in the vidaza 10µM significant
category, and 483 (30.2%) probes were found in both the vidaza 10µM and vidaza 5µM
significant category in the pairwise comparison model. This indicates that the largest numbers
of significant probes were found in the 10 µM vidaza samples in both pairwise comparison
model and trend model, however, larger numbers of significant probes were found in 5 µM
vidaza-treated samples in the trend model. These indicate that for some µM vidaza-treated
samples, mean differences were significant but changes over days did not capture the same
differences in the trend model.
3.3 Linked DNA Methylation and Gene Expression
Methylation and expression probes were linked by Entrez gene ID and 12863 genes
were found to appear on both the methylation and expression platforms. There are in total
48783 probes for the expression data and 27578 probes for the DNA methylation data and one
gene can have multiple probes. 16985 genes were represented on expression array, 14491
genes on the methylation array, and 12863 genes common to both the expression and
methylation platforms. The number of genes represented on the two platforms was displayed
in Figure 21 (a,b). The distribution of minimal distance of TSS for those 14491 genes were
similar as what showed in Figure 1 and 6753 (46.6%) of those genes are in CpG islands.
33
Figure 21 (a,b) Venn diagram for methylation and expression (by proportion and unscaled)
Figure 21a Figure 21b
Figure 21a. Blue, purple circles are for methylation (14491 probes) and expression (16985 probes)
profiles.
Figure 20b. Number of probes in expression, methylation and matched profiles.
A starburst plot of the BH-adjusted p-values from pairwise comparison of 1µM Vidaza
compared to mannitol-treated samples for those 12863 gene-linked methylation and
expression probes were plotted in Figure 22 (a,b). The y-axis shows the -log10 FDR-adjusted p-
value times the sign (positive/negative) of the fold change for the expression data and the x-
axis shows the same for DNA methylation. Genes that showed both decrease in DNA
methylation with increased gene expression would appear in the top left quadrant. There does
not appear to be a noticeable cluster of genes in this area of the plot.
4122
1628
12863
Expression Methylation
34
Figure 22 (a,b) Starburst plot of the BH-adjusted p-values for gene-linked probes (in log10
scales)
Figure 22a Figure 22b
Figure 22a. Density heatmap for BH-adjusted p-values from methylation and expression profiles (x axis
in log10 scale: transformed methylation p-values; y axis in log 10 scale: transformed expression p-values.
Red: high density; blue: low density.)
Figure 22b. Transformed BH-adjusted p-values of gene expression vs. methylation. Transformation = -
log10 BH-adjusted pvalue * sign (Fold-change)
For the DNA methylation data, genes with multiple probes were further studied by
genomic context. 24569 probes were found for the 12863 genes with expression data.
Specifically, 17.4% (2237 out of 12863) of genes have one probe, 80.9% (10411 out of 12863) of
genes have two probes and others (1.67%) have more than two probes. Among those 24569
probes, 4131 of them (16.8%) are CpG island promoter probes, 8656 of them (35.2%) of them
are non-CpG island promoter probes, 6126 of them (25.2%) are CpG island body probes and
5582 of them (22.7%) are non-CpG island body probes.
For the 10411 genes with two probes (80.9%), 16 categories described the possible
genomic context of the two probes. Figure 23 shows the distribution of genes in the 16
categories. Five main categories explained the majority of genomic contexts (95.1% of all two
probe genes). The patterns of these five probe contexts are shown in Figure 24. The most
35
frequent context targeted one probe in the promoter and the second in the gene body, neither
in CpG islands (34%) (pattern BD). The next most frequent categories either targeted one probe
in a non-CpG islands promoter and the second in a CpG islands gene body (25%) (pattern BC),
or one probe in a promoter and one in a gene body, both in CpG islands (24%) (pattern AC).
The last two main categories targeted one probe in a promoter CpG island and the second in a
non-CpG island gene body (7%) (pattern AD), or both probes in non-CpG island promoter region
(5%) (pattern BB).
Figure 23 Pie graph for patterns of methylation genes that have two probes
AA
1%
AB
1%
AC
24%
AD
7%
AE
0%
AF
0%
BB
5%
BC
25%
BD
34%
BE
0%
BF
1%
CC
0%
CD
0%
DD
1%
EE
0%
FF
0%
gene patterns
Promoter Gene Body
TSS
BD (34%)
Promoter Gene Body
TSS
CGI
BC (25%)
Figure 24 Five main gene patterns for the two-probe genes
36
The 10411 genes measured by two probes can have different numbers of significant
probes, having either both probes significantly differentially methylated, only one probe, or
neither probe. Three genes that show a decrease of methylation and an increase in expression
over time are shown in Figure 25. Numbers of genes with different patterns were summarized
and matched to significant expression for genes with genomic contexts (patterns BD, BC, AC, AD
and BB in Figure 24). Significantly demethylated genes (decreased in methylation beta value)
matched to positively significantly changed expression genes (filtered by 1 µM vidaza vs.
mannitol group) were shown in Table 5. It is showed that the majority of the genes had
decreased methylation with increased expression values. Furthermore, table 5 showed that
about 44% of the genes with non-CpG island promoter and gene body probes (BD pattern) are
significantly decreased in methylation for both probes, and about 30% of them are significantly
decreased in methylation for one probe (adjusted p<0.05, FC<-1). For those genes with one
promoter non-CpG island and one CpG island gene body probe (BC pattern), about 50% are
significantly decreased in methylation for at least one probe. These indicate that the drug
Promoter Gene Body
TSS
CGI CGI
Promoter Gene Body
TSS
CGI
AC (24%)
AD (7%)
Promoter Gene
Body
TSS
BB (5%)
37
worked on these methylated non-CpG island genes and demethylated them. It seems that
promoter non-CpG island (pattern B) genes tended to be easily affected by durg (1µM vidaza)
since a large proportion of significant changes occurred. Among those two probe genes (5415)
that at least only one significantly demethylation summarized in Table 5, only 96 (1.8%) of them
had an increase of gene expression.
Figure 25 DNA methylation and gene expression values for three matched genes (left:
methylation, right: expression)
Left: methylation beta value for gene cg 24070292 on three days; Right: gene expression value for the
same gene on three days
Left: methylation beta value for gene cg 02082571 on three days; Right: gene expression value for the
same gene on three days
38
Left: methylation beta value for gene cg10530713 on three days; Right: gene expression value for the
same gene on three days
Table 5 Counts and percentage of negatively significant demethylated genes and positively
significant increased expression genes of five main patterns for two probe genes
Gene
pattern: BD
(N=3520)
Numbers of
genes
significantly
demethylated
Percent
(%)
Number of genes
significant with
increased
differential
expression
Expression
percent
(%)
both sig 1548 44.5 23 1.49
Only B sig 522 15.0 10 1.92
Only D sig 500 14.4 8 1.60
non sig 910 26.1 12 1.32
Subtotal * 3480 1 53 1.52
Gene
pattern: BC
(N=2627)
both sig 641 24.7 12 1.87
Only B sig 489 18.9 2 0.41
Only C sig 166 6.4 8 4.82
non sig 1296 50.0 9 0.69
Subtotal* 2592 1 31 1.20
Gene
pattern: AC
(N=2467)
both sig 623 25.7 16 2.57
Only sig A 130 5.36 2 1.54
Only Sig C 117 4.82 6 5.13
non sig 1556 64.1 8 0.51
Subtotal * 2426 1 32 1.32
39
Gene
pattern: AD
(N=736)
both sig 145 20.4 1 0.69
Only sig A 46 6.48 2 4.35
Only sig D 80 11.3 1 1.25
non sig 439 61.8 1 0.23
Subtotal * 710 1 5 0.70
Gene
pattern: BB
(N=551)
both sig 256 46.7 3 1.17
only one sig 152 27.7 2 1.32
non sig 140 25.5 3 2.14
Subtotal * 548 1 8 1.46
Total
number of
genes 9901 9901
* Numbers of demethylated genes in the category
A = CpG island (CGI) promoter; B = non-CGI promoter; C= CGI gene body; D= non-CGI gene body;
Three cases in each pattern: (1) both sig: two probes of the gene are all significant (BH-adjusted p<0.05
and FC>1 or <-1); (2) one sig: only one probe is significant; (3) non sig: none of the probe is significant
40
CHAPTER 4: DISCUSSION AND CONCLUSIONS
The inter-relationships between DNA methylation and gene expression profiles of HL-60
cells treated with vidaza were studied in this thesis. When studied separately, methylation m
values for three doses of vidaza-treated samples decreased significantly compared to mannitol-
treated samples in the pairwise comparison model that compare the mean methylation m
value of the three doses of vidaza-treated samples to mannitol-treated samples. The largest
number of differentially methylated probes were discovered in 1 µM vidaza-treated samples,
with demethylation occurring within the first 24 hours. Because of the large change before day
1, these features were discovered in the pairwise treatment comparison model (unadjusted for
days) instead of the model that looked for differential trends in DNA methylation over time.
Gene expression for three doses of vidaza treated samples also increased, and 10µM vidaza-
treated samples showed more increased expression than 1µM and 5 µM vidaza-treated
samples. The fact that the largest number of differentially expressed genes occurred for the
highest dose and not the dose with the most DNA methylation changes (lowest dose), suggests
a treatment toxicity that affects gene expression directly and not as a consequence of DNA
methylation. After integrating the DNA methylation and gene expression profiles, there is a
small number of significantly changed expression probes matched to methylation probes (BH-
adjusted p<0.05, FC>1 or <-1).
The drug, vidaza is a DNA methylation inhibitor and can affect methylation through cell
division. The large number of changes in the methylation profile is expected given the known
function of this drug. However, many factors may result in changes in gene expression, and
they may not be direct consequences of the methylation changes. A previous study indicated
41
that concordant demethylation and up-regulation of gene expression was limited to a set of
160 genes (about 3.4% of demethylated genes) [9]. In this study, 1.8% of demethylated two
probe genes were found relate to up-regulation of gene expression (Table 5). This discrepancy
reflects the fact that many methylation changes in the genome are not necessarily
accompanied by functional consequences on gene expression. Weber et al. demonstrated that
the impact of methylation on the transcriptional activity of promoters depends on the CpG
content in these DNA regions, such as promoters with low CpG contents are hypermethylated
and this doesn’t preclude gene expression [10].In other words, demethylation will not
necessary lead to the increase of gene expression. Data analysis in this study leads to similar
result, and indicates that most of the methylation changes do not lead to a significant change of
promoter activity.
Furthermore, terminal differentiation in the myeloid lineage results in apoptosis. One of
the major effects of hypomethylating agents is their cytotoxic action by induction of apoptosis,
similar to that induced by cytostatic drugs, independent from cellular differentiation [11,12]. In
this study, the induction of apoptosis following vidaza treatment happened and a possible link
between hypomethylation and apoptosis induction can be studied in the future.
It is found that for genes in 1 µM vidaza-treated samples, demethylation occurred in one
day and measurements on day 1,3 and 5 in the experiment cannot capture those fast changes
happened in the first day. Further study may use a smaller time scale, such as 8 hr to measure
methylation value for 1 µM vidaza-treated samples.
Illumina Infinium HumanMethylation27 BeadChip was used to measure methylation
value in this study. However, a more comprehensive array platform, Infinium
42
HumanMethylation450 (HM450), has now been introduced. The HM450 array targets 485,512
cytosine probes covering 99% of RefSeq genes. The probes interrogate 19,755 unique CpG
islands with additional coverage in shore regions and miRNA promoters as well as 3091 probes
at non-CpG sites [13]. Further study can use the HM450 array that contains both direct effect of
promoters and indirect effect of enhancers, in this case, more genes that have concordant
demethylation and increased gene expression may be identified.
In conclusion, this study shows that DNA methylation and expression changes induced by
hypomethylation treatment are complex. DNA demethylation can be affected by the treatment
directly but many methylation changes in the genome are not necessarily accompanied by
functional consequences on gene expression.
43
REFERENCES
[1] Ilan Bernstein , Hyang-Min Byun , Ann Mohrbacher , Dan Douer , Gerry Gorospe , Jeffrey
Hergesheimer , Susan Groshen , Casey O'Connell , Allen S. Yang (2010). A phase I biological
study of azacitidine (Vidaza
TM
) to determine the optimal dose to inhibit DNA methylation
Epigenetics Vol. 5, Iss. 8,
[2] Kaminskas, E., Farrell, A. T., Wang, Y., Sridhara, R., & Pazdur, R. (2005). FDA drug approval
summary: Azacitidine (5-azacytidine, vidaza) for injectable suspension. The Oncologist, 10(3),
176-182
[3] Kulis, M., & Esteller, M. (2010). DNA methylation and cancer. Advances in Genetics, 70, 27-
56.
[4] Joseph G.Hacia (2010) DNA Methylation and Gene Expression Profiles in Vidaza Treated
Cultured Cancer Cells, Progress Report
[5] Pan Du, Xiao Zhang, Chiang-Ching Huang, Nadereh Jafari, Warren A Kibbe, Lifang Hou, Simon
M Lin (2010) Comparison of Beta-value and M-value methods for quantifying methylation levels
by microarray analysis. BMC Bioinformatics.11:587
[6] Gordon K Smyth (2004) Linear models and empirical Bayes methods for assessing
differential expression in microarray experiments. Stat Appl Genet Mol Biol 3: Article 3.
[7] Jeremy D. Silver, Matthew E. Ritchie, and Gordon K. Smyth (2009) Microarray background
correction: maximum likelihood estimation for the normal–exponential convolution.
Biostatistics. 10(2): 352–363.
[8] John D. Storey and Robert Tibshirani (2003) Statistical significance for genomewide studies.
PNAS. 100 (16) 9440-9445.
44
[9] Emiliano Fabiani, Giuseppe Leone, et al (2010). Analysis of genome-wide methylation and
gene expression induced by 5-aza-20 -deoxycytidine identifies BCL2L10 as a frequent
methylation target in acute myeloid leukemia. Leukemia & Lymphoma. 12(51):2275-2284
[10] Weber M, Hellmann I, Stadler MB, et al (2007). Distribution, silencing potential and
evolutionary impact of promoter DNA methylation in the human genome. Nat Genet. 39:457-
466
[11] Khan R, Schmidt-Mende J, Karimi M, et al (2008). Hypomethylation and apoptosis in 5-
azacytidine-treated myeloid cells. Exp Hematol. 36:149–157
[12] Flotho C, Claus R, Batz C, et al (2009). The DNA methyltransferase inhibitors azacitidine,
decitabine and zebularine exert differential effects on cancer gene expression in acute myeloid
leukemia cells. Leukemia. 23:1019–1028
[13] Morris, Tiffany J., and Stephan Beck (2015). Analysis Pipelines and Packages for Infinium
HumanMethylation450 BeadChip (450k) Data. Methods (San Diego, Calif.). 72: 3–8
[14] Siegmund KD (2011). Statistical approaches for the analysis of DNA methylation microarray
data.Hum Genet. 129(6):585-595
[15] Yoav Benjaminia, Dan Draib, Greg Elmerc,et al (2001). Controlling the false discovery rate
in behavior genetics research. Behavioural Brain Research. 125(1-2):279-284.
Abstract (if available)
Abstract
The methylation of DNA is an epigenetic modification that can play an important role in the control of gene expression in mammalian cells. The genes involved include tumor suppressor genes, genes that suppress metastasis and angiogenesis, and genes that repair DNA suggesting that epigenetics plays an important role in tumorigenesis. I studied the genome‐wide DNA methylation, and gene expression profiles in HL60 cells following 5-azacytidine (Azacitidine, Vidaza™), using microarray technologies. Three doses of Vidaza (1µM, 5µM, 10µM) and mannitol (control) were used and DNA methylation and gene expression were measured in Day 1, Day 3 and Day 5. Quality control analysis was conducted, followed by two statistical analyses to study the effects of drug treatment on the methylation profile and expression profile. In the first analysis we compared the average DNA methylation (gene expression) levels between treatment groups (‘pairwise treatment comparison’) and in the second analysis we compared the treatment‐specific change in average levels over time (trend comparison). A decrease in DNA methylation was found in all three doses of Vidaza‐treated samples compare to the mannitol treated samples and the largest number of differentially methylated probes were found in lowest doses (1µM Vidaza). Moreover, gene expression values also changed, with a similar number of probes with increased or decreased expression identified. Furthermore, for DNA methylation profiles, genes with multiple probes were studied by genomic context. For the majority of genes that contain two probes, a large number of genes with non‐CpG islands promoters is significantly changed after drug treatment (BH‐adjusted p<0.05, fold‐change (FC) >1 or <-1). However, only a small number of genes showed concomitant changes in gene expression. This study provides insights about changes in DNA methylation induced by Vidaza and potential changes of gene expression for those same genes.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Understanding DNA methylation and nucleosome organization in cancer cells using single molecule sequencing
PDF
Finding signals in Infinium DNA methylation data
PDF
DNA methylation inhibitors and epigenetic regulation of microRNA expression
PDF
The relationship between DNA methylation and transcription factor binding in colon cancer cells
PDF
DNA methylation changes in the development of lung adenocarcinoma
PDF
Identification and characterization of cancer-associated enhancers
PDF
Differential methylation analysis of colon tissues
PDF
DNA methylation groups determined by GATA5 gene methylation level are correlated with tumor subtype, sex, smoking status, and body mass index in esophageal and gastric adenocarcinoma
PDF
Gene expression and angiogenesis pathway across DNA methylation subtypes in colon adenocarcinoma
PDF
DNA methylation of NOS genes and carotid intima-media thickness in children
PDF
Identification of DNA methylation markers in diffuse large B-cell lymphoma
PDF
An analysis of conservation of methylation
PDF
Functional DNA methylation changes in normal and cancer cells
PDF
Statistical analysis of high-throughput genomic data
PDF
Using average pairwise distance in a correlation analysis
PDF
Comparative analysis of DNA methylation in mammals
PDF
Preprocessing and analysis of DNA methylation microarrays
PDF
Using epigenetic toggle switches to repress tumor-promoting gene expression
PDF
DNA methylation as a biomarker in human reproductive health and disease
PDF
DNA methylation markers for blood-based detection of small cell lung cancer in mouse models
Asset Metadata
Creator
Liu, Mengyu
(author)
Core Title
DNA methylation and gene expression profiles in Vidaza treated cultured cancer cells
School
Keck School of Medicine
Degree
Master of Science
Degree Program
Biostatistics
Publication Date
07/23/2015
Defense Date
07/22/2015
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
DNA methylation,gene expression,linked profiles,OAI-PMH Harvest
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Siegmund, Kimberly (
committee chair
), Hacia, Joseph G. (
committee member
), Stram, Daniel O. (
committee member
)
Creator Email
liucathy09@gmail.com,liumengy@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-604037
Unique identifier
UC11300236
Identifier
etd-LiuMengyu-3679.pdf (filename),usctheses-c3-604037 (legacy record id)
Legacy Identifier
etd-LiuMengyu-3679.pdf
Dmrecord
604037
Document Type
Thesis
Format
application/pdf (imt)
Rights
Liu, Mengyu
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
Tags
DNA methylation
gene expression
linked profiles