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
/
Systems approaches to understanding metabolic vulnerabilities in cancer cells
(USC Thesis Other)
Systems approaches to understanding metabolic vulnerabilities in cancer cells
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
SYSTEMS APPROACHES TO UNDERSTANDING METABOLIC
VULNERABILITIES IN CANCER CELLS
by
James H. Joly
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the Requirements for the Degree
DOCTOR OF PHILOSOPHY
(CHEMICAL ENGINEERING)
December 2020
Copyright 2020 James H. Joly
ii
TABLE OF CONTENTS
List of Figures .............................................................................................................................................. iv
Abstract…………………………………………………………………………………………………….v
Introduction ................................................................................................................................................... 1
Materials and Methods .................................................................................................................................. 5
Chapter 1: A synthetic lethal drug combination mimics glucose deprivation–induced cancer cell death in
the presence of glucose ............................................................................................................. 14
1.1 Metabolomics identifies accumulation of L-cystine and L-cysteine and depletion of glutathione
as markers of glucose deprivation sensitivity ............................................................................ 14
1.2 Glutamate-cysteine ligase (GCL) activity moderately regulates resistance to glucose deprivation.
................................................................................................................................................... 18
1.3 Glutamate-cysteine ligase (GCL) is activated by glucose withdrawal. ..................................... 18
1.4 NADPH is the limiting reducing agent in glucose-deprived cancer cells. ................................ 22
1.5 L-cystine import, but not L-cysteine import, induces ROS mediated cell death in glucose-
deprived cancer cells. ................................................................................................................ 26
1.6 L-cystine import from the glutamate/cystine antiporter xCT/SLC7A11 promotes cell death
following glucose starvation. ..................................................................................................... 30
1.7 Inhibition of glucose metabolism is synthetically lethal with inhibition of GCL. .................... 34
1.8 Discussion .................................................................................................................................. 43
Chapter 2: Differential Gene Set Enrichment Analysis: An enrichment-based method to infer tradeoffs from
-omics data. ............................................................................................................................... 45
2.1 DGSEA measures the relative difference between two gene sets ............................................. 47
2.2 Simulation Study ....................................................................................................................... 50
2.3 DGSEA accurately captures hypoxia-induced coordinate increases in glycolysis and decreases
in oxidative phosphorylation ..................................................................................................... 53
2.4 Benchmarking DGSEA against QuSAGE ................................................................................. 57
2.5 DGSEA is more predictive than GSEA of lactate secretion and glucose consumption in cancer
cell lines. .................................................................................................................................... 59
2.6 DGSEA is correlated with high concentrations of intracellular lactate and low concentrations of
intracellular AMP in adherent cancer cell lines. ........................................................................ 62
2.7 Untargeted DGSEA predicts differential metabolic pathway activity in senescent and
proliferating cells ....................................................................................................................... 65
2.8 Limitations of DGSEA .............................................................................................................. 66
2.9 Discussion .................................................................................................................................. 67
Chapter 3: The landscape of metabolic pathway dependencies in cancer cell lines ................................... 69
3.1 Introduction ............................................................................................................................... 69
3.2 Genetic Pathway Dependency Enrichment Analysis Identifies Metabolic Pathway Dependencies
by Integrating Metabolic Pathway Activity with CRISPR Screens .......................................... 71
3.3 Simulation Study ....................................................................................................................... 72
3.4 Metabolic Pathway Dependency is Highly Context Specific .................................................... 73
iii
3.5 Media Composition Influences Metabolic Pathway Dependency ............................................. 79
3.6 Metabolic pathway activity is correlated with anti-cancer drug sensitivity .............................. 81
3.7 Pharmacological Pathway Dependency Enrichment Analysis Reveals Common Metabolic
Pathway Vulnerabilities ............................................................................................................. 84
3.8 Integration of pharmacologic and genetic screens reveals common metabolic vulnerabilities . 90
3.9 Discussion .................................................................................................................................. 93
Bibliography ............................................................................................................................................... 96
iv
List of Figures
Figure 1. Accumulation of L-cystine and L-cysteine and depletion of reduced glutathione are
metabolic markers of sensitivity to glucose deprivation. .................................................. 17
Figure 2. Glutamate-cysteine ligase (GCL) activity moderately regulates resistance to glucose
deprivation. ....................................................................................................................... 19
Figure 3. NADPH is the limiting reducing agent upon glucose deprivation. ................................... 24
Figure 4. L-cystine import induces oxidative stress and ROS-mediated cell death in glucose deprived
cancer cells. ....................................................................................................................... 29
Figure 5. L-cystine import from the glutamate/cystine antiporter, system xc-, is required for glucose
deprivation-induced cell death. ......................................................................................... 32
Figure 6. Inhibition of glutathione synthesis is synthetically lethal with GLUT1 inhibition. .......... 37
Figure 7. Differential Gene Set Enrichment Analysis (DGSEA) quantifies the enrichment between
two gene sets relative to each other. ................................................................................. 50
Figure 8. DGSEA is more sensitive than GSEA for co-regulated pathways. .................................. 51
Figure 9. DGSEA accurately captures hypoxia-induced coordinate increases in glycolysis and
decreases in oxidative phosphorylation. ........................................................................... 55
Figure 10. DGSEA is a better predictor of cellular metabolism than GSEA. .................................... 60
Figure 11. DGSEA is a better predictor of intracellular lactate and AMP levels than GSEA for
adherent cell cultures. ....................................................................................................... 63
Figure 12. Untargeted DGSEA predicts differential metabolic activity in senescent and proliferating
cells. 79 metabolic pathway gene sets from the Kyoto Encyclopedia of Genes and Genomes
(KEGG) were queried with untargeted DGSEA to identify metabolic differences upon
ionizing radiation-induced senescence in IMR90 cells. .................................................... 66
Figure 13. Integration of gene expression and CRISPR gene dependencies to identify metabolic
pathway dependencies. ..................................................................................................... 73
Figure 14. Global analysis of metabolic dependency data reveals context-specific pathway
essentialities. ..................................................................................................................... 76
Figure 15. Media composition influences metabolic pathway dependency. ...................................... 81
Figure 16. Metabolic pathway activity is correlated with anti-cancer drug sensitivity. ..................... 83
Figure 17. Pharmacological PDEA Reveals Common Metabolic Pathway Vulnerabilities. ............. 86
Figure 18. Integration of Pharmacological and Genetic Screens Reveal Common Metabolic Pathway
Vulnerabilities. .................................................................................................................. 92
v
Abstract
Cancer cells exhibit an altered metabolic state compared to normal cells. In fact, some of the most
successful chemotherapeutics target aberrant tumor metabolism. Metabolic vulnerabilities in cancer arise
from dysregulated signaling and metabolic pathways, which can cause increased reliance on certain
substrates for survival. Here, we took systems biology approaches to identify metabolic vulnerabilities in
cancer cells. Using liquid chromatography-mass spectrometry (LC-MS) based metabolomics, we identified
a co-dependence on glucose metabolism and L-cystine metabolism. Building upon that, we defined a
synthetic lethal drug combination to mimic glucose deprivation-induced cell death in the presence of
glucose
1
. Next, we sought to quantify metabolic crosstalk by developing a bioinformatic method
Differential Gene Set Enrichment Analysis
2
that quantifies the relative enrichment of two metabolic
pathways. Lastly, we defined a computational framework to identify context-specific dependence on
metabolic pathways
3
. By integrating gene expression, CRISPR screens, and pharmacologic screen data, we
uncovered novel metabolic crosstalk and co-dependencies between metabolic pathway activity and
metabolic pathway essentiality. The results and frameworks here serve as a foundation for the systems
biology-driven discovery of metabolic vulnerabilities and therapeutic targeting of metabolism in cancer
cells.
1
Introduction
Proliferating cancer cells exhibit markedly different metabolic requirements compared to most
normal differentiated cells. For example, in order to support their high rates of proliferation, cancer cells
consume additional nutrients and utilize those nutrients to produce the macromolecules necessary to sustain
cellular function. Therefore, metabolic pathways must be altered in a way that balances biosynthetic
processes with sufficient ATP production and redox balance to support cell growth and survival. As all
cancer cells are dependent on this change in metabolism, metabolic reprogramming is considered a hallmark
of cancer
4
and represents an attractive therapeutic target
5–9
.
In addition to metabolic reprogramming, dysregulated cellular signaling is a hallmark of cancer
4
.
Aberrant signaling presents opportunities for molecularly targeted therapies, such as the use of EGFR
10
and
c-Met
11
inhibitors. Cancer cell signaling and metabolism have been shown to be highly interconnected, with
many protein kinases being linked to alterations in energy production and redox homeostasis. For example,
increased reliance on glucose has been linked to kinase signaling in the form of AKT and AMPK
12–14
,
mTORC1
15
, and tyrosine kinases
16
. The classical nutrient sensing-kinase mTORC2 has been linked to
regulation of glutamate and glutathione metabolism
17
. In addition to protein kinases, other studies have
linked Ca
2+
influxes through the L-type calcium channel Ca v1.3 and methylation of PP2AC to increased
reliance on glucose and reactive oxygen species generation
18
. These examples demonstrate the
interconnected nature of cancer cell signaling and metabolism. However, while there are many cancer
therapies targeting protein kinases, therapeutic strategies targeting the metabolic phenotype(s) of cancer
cells remain underdeveloped.
One consequence of altered signaling and metabolism is that some cancer cells become dependent
on certain nutrients for survival. For example, cancer cells often exhibit increased reliance on glucose. The
glycolytic phenotype of cancer cells was originally observed by Otto Warburg in 1927
19
. Warburg found
2
that cancer cells preferentially perform aerobic glycolysis instead of diverting carbon to the more
energetically favorable process of oxidative phosphorylation. The glycolytic phenotype has been observed
in cancer patients and is often utilized as a diagnostic in the form of
18
FDG-PET scanning
20
, but
therapeutically targeting the glycolytic phenotype of cancer cells remains a challenge because,
pragmatically speaking, tumors cannot be completely starved of glucose in vivo. While there have been
recent efforts to modulate diet to affect cancer patient response to treatment
21,22
, this strategy alone would
likely not eradicate a tumor. In addition, chemical inhibitors of glucose metabolism have achieved limited
success
9,23–25
, perhaps because other organs, such as the brain and heart, also exhibit high glucose uptake.
Another complication for therapeutically targeting the glycolytic phenotype of tumors is that cancer cells
exhibit a broad spectrum of dependence on glucose
16
. In addition, since tumors are often highly
heterogenous, individual inhibitors of glucose metabolism may result in a reduction, but not an eradication,
of a tumor. As glucose deprivation-sensitive tumor cells die, a treatment-resistant tumor is likely to arise,
resulting in relapse of an aggressive, difficult-to-treat cancer. Therefore, alternative treatment strategies are
necessary to enable therapeutic targeting of the glycolytic phenotype of cancer cells.
While the glycolytic phenotype of cancer cells has been extensively studied, metabolic pathways
(e.g., glycolysis) do not exist in isolation. Rather, they produce similar byproducts (e.g. ATP and NAD(P)H)
and can compensate for one another. As such, methods to study the relative activity of two metabolic
pathways are necessary. We developed a new algorithm, Differential Gene Set Enrichment Analysis
(DGSEA), to quantify the relative enrichment of two gene sets in genome-wide RNA expression data. Using
metabolic pathways as a test, we found that DGSEA more accurately captured cellular phenotypes than
analyzing individual metabolic pathways using Gene Set Enrichment Analysis (GSEA)
26
. For example, we
found that DGSEA accurately captures the hypoxia-induced coordinate upregulation of glycolysis and
downregulation of oxidative phosphorylation. We also found that DGSEA is more predictive than GSEA
of the metabolic state of cancer cell lines, including lactate secretion and intracellular concentrations of
lactate and AMP. Finally, we demonstrated that DGSEA can also be used for hypothesis generation about
3
differential metabolic pathway activity in senescent cells. To our knowledge, DGSEA represents the first
statistical framework for analyzing tradeoffs between metabolic pathways on a genome-wide scale.
Two promising trends in cancer therapy are the concepts of synthetic lethality and context-
dependent essentiality. In synthetic lethality, cell death occurs following the simultaneous inactivation of
multiple gene products but not the inactivation of either gene product alone
27
. Synthetic lethal treatment
strategies are thought to allow for larger therapeutic windows and less incidence of acquired resistance to
therapy. In addition to synthetic lethality, context-dependent essentiality occurs when the inactivation of
genes results in cell death or decreased cell growth only upon certain contexts. For example, when cells
highly express a single isoform of an enzyme but have little or no expression of other isoforms, inactivation
of the highly expressed isoform results in cell death, since no compensatory mechanism is present
28
. In an
effort to understand context-dependent essentialities, we developed a computational framework to analyze
metabolic pathway crosstalk and co-dependency in cancer cell lines
3
. To our knowledge, our approach,
termed pathway dependency enrichment analysis (PDEA) is the first method to interrogate cancer cell
dependencies on the pathway level. By integrating gene expression, gene dependency, and drug response
data, we were able to identify which contexts result in increased dependency on metabolic pathways. We
laid a foundation for using metabolic pathways as biomarkers and recapitulated known interactions between
metabolic pathway activity and drug response. Thus, PDEA serves as a novel tool to integrate
transcriptomic, CRISPR gene dependency, and drug response data to identify cancer cell vulnerabilities on
pathways.
We reasoned that a synthetic lethal therapeutic strategy using combinations of inhibitors to mimic
glucose deprivation-induced cell death might enable therapeutic targeting of the glycolytic phenotype of
cancer cells. We thus sought to identify the molecular mechanisms regulating cancer cell death following
glucose deprivation. Using mass spectrometry-based techniques to study the metabolic effects of glucose
deprivation, we identified a synthetic lethal interaction that induces cell death in glucose deprivation-
4
sensitive cancer cells in the presence of glucose. Furthermore, we have identified a biomarker (xCT /
SLC7A11) that is predictive of sensitivity to glucose deprivation and correlates with response to
combinations of inhibitors mimicking glucose deprivation-induced cell death. In other words, we have
identified a drug combination that synergistically induces cancer cell death in vitro and a corresponding
biomarker that is predictive of response to the drug combination
1
. This work will provide new methods to
therapeutically target glucose-addicted cancer cells and improve our fundamental understanding of cancer
cell metabolism.
5
Materials and Methods
Cell Culture
All cell lines were cultured in high-glucose DMEM (4.5 g/l glucose, 110mM pyruvate; Mediatech)
supplemented with 10% (v/v) FBS (Omega Scientific) plus 1% (v/v) anti-anti (Invitrogen). For glucose
starvation, cells were washed twice with PBS and then incubated in DMEM without glucose and pyruvate
(0 g/l glucose, 0 mM pyruvate; Invitrogen) supplemented with 10% dialyzed FBS (Omega Scientific) plus
1% anti-anti. GBM cell lines LN18, LN229, T98, and U87 and sarcoma cell lines HT161 and TC32 were a
gift from Thomas Graeber (University of California, Los Angeles). GBM cell lines A172 and U118MG
were gifts from David Changhan Lee (University of Southern California) and Matthew Lazzara (University
of Virginia), respectively.
Genetic modifications
Knockdown and overexpression of genes using CRISPR interference
29
and CRISPR activation
30
was done as previously described. Briefly, guide RNAs were expressed using pLenti-hygro-mTagBFP2
and paired with either lenti-EF1a-dCas9-KRAB-Puro or lenti-EF1a-dCas9-VPR-Puro. Wild-type GCLM
(pLX304), GCLC (pLX302), luciferase (pLX304), or RFP (pLX302) was expressed using lentiviral
vectors. Following infection, cells were selected using hygromycin (pLenti-hygro-mTagBFP2), puromycin
(lenti-EF1a-dCas9-VPR-Puro or lenti-EF1a-dCas9-KRAB-Puro or pLX302), or blasticidin (pLX304). For
CRISPRa experiments, guide RNAs were designed to target upstream of the transcriptional start site. For
CRISPRi experiments, guide RNAs were designed to target downstream of the transcriptional start site.
Cell Viability Analysis
Viability was measured by Trypan blue exclusion using a TC20 automated cell counter (BioRad)
using optimized, cell type-specific imaging parameters to distinguish live and dead cells.
6
Western blotting
Cells were lysed in modified RIPA buffer (50 mM Tris–HCl (pH 7.5), 150 mM NaCl, 10 mM β-
glycerophosphate, 1% NP-40, 0.25% sodium deoxycholate, 10 mM sodium pyrophosphate, 30 mM sodium
fluoride, 1 mM EDTA, 1 mM vanadate, 20 g/ml aprotinin, 20 g/ml leupeptin, and 1 mM
phenylmethylsulfonyl fluoride). Whole cell lysates were resolved by SDS–PAGE on 4–15% gradient gels
and blotted onto nitrocellulose membranes (Bio-Rad). Native lysates were collected in a native sample
buffer (50 mM tris, pH 8.0, 1% NP-40, 150 mM NaCl) containing protease and phosphatase inhibitors.
Native PAGE gels were run in Tris/Glycine/SDS at 4
o
C without boiling or addition of reducing agents.
Membranes were blocked over- night and then incubated sequentially with primary and IRDye-conjugated
secondary antibodies (Li-Cor). Blots were imaged using the Odyssey Infrared Imaging System (Li-Cor).
Liquid Chromatography-Mass Spectrometry Metabolomics
Cell lines were plated onto 6-well plates at a density of 1 x 10
5
cells/well. Cells were washed twice
with PBS before being treated with media. Metabolite extraction was performed 3 h after adding glucose
starvation unless otherwise specified. For extraction of intracellular metabolites, cells were washed on ice
with 1 mL ice-cold 150 mM ammonium acetate (NH 4AcO, pH 7.3). 1 mL of -80
o
C 80% MeOH was added
to the wells, samples were incubated at -80
o
C for 20 min, then cells were scraped off and supernatants were
transferred into microcentrifuge tubes. Samples were pelleted at 4C for 5 min at 15k rpm. Supernatants
were transferred into LoBind Eppendorf microcentrifuge tubes and the cell pellets were re-extracted with
200 uL ice-cold 80% MeOH, spun down and the supernatants were combined. Metabolites were dried at
room temperature under vacuum and re-suspended in water for injection. For extraction of extracellular
metabolites, 20 uL of cell free-blank and conditioned media samples were collected from wells. Metabolites
were extracted by adding 500 ul -80C 80% MeOH, dried at room temperature under vacuum and re-
suspended in water for injection.
7
NAD(P)H metabolites were extracted using a method described here
31
. Briefly, for extraction of
intracellular metabolites, a two solvent method is used. Solvent A is 40:40:20 acetonitrile:methanol:water
with 0.1 M formic acid, and solvent B is 15% NH 4HCO 3 in water (w:v), precooled on ice. Solvent A was
added to the cell pellet, vortexed for 10 s, and allowed to sit on ice for 3 min. For each 100 μl of solvent A,
8.7 μl of solvent B was then added and vortexed to neutralize the sample, and the mixture allowed to sit on
dry ice for 20 min. Volumes were calculated to produce a total volume of 50 μl solvent per 1 μl of cell
volume. Samples were then centrifuged at 16,000 g for 15 min at 4°C and supernatant taken for to be dried
at room temperature under vacuum and resuspended in water for LC-MS analysis.
Samples were randomized and analyzed on a Q-Exactive Plus hybrid quadrupole-Orbitrap mass
spectrometer coupled to an UltiMate 3000 UHPLC system (Thermo Scientific). The mass spectrometer was
run in polarity switching mode (+3.00 kV/-2.25 kV) with an m/z window ranging from 65 to 975. Mobile
phase A was 5 mM NH4AcO, pH 9.9, and mobile phase B was acetonitrile. Metabolites were separated on
a Luna 3 μm NH2 100 Å (150 × 2.0 mm) column (Phenomenex). The flowrate was 300 μl/min, and the
gradient was from 15% A to 95% A in 18 min, followed by an isocratic step for 9 min and re-equilibration
for 7 min. All samples were run in biological triplicate.
Metabolites were detected and quantified as area under the curve based on retention time and
accurate mass (≤ 5 ppm) using the TraceFinder 3.3 (Thermo Scientific) software. Raw data was corrected
for naturally occurring 13C abundance (49). Extracellular data was normalized to integrated cell number,
which was calculated based on cell counts at the start and end of the time course and an exponential growth
equation. Intracellular data was normalized to the cell number and cell volume at the time of extraction.
Pathway maps were made with Cytoscape software (50).
Flow cytometry
8
Cells were incubated with either 5 mM CM-H2DCFDA for 10min before analysis using a
MACSQuant® Analyzer 10 Flow Cytometer. Cells were gated using forward scatter and side scatter to
remove debris and dead cells. To quantify changes in DCF-DA signal, mean fluorescent intensity after
gating was used.
GCL activity assay
Enzymatic activity for GCL was assessed using a naphthalene dicarboxaldehyde (NDA)
derivatization method to form cyclized, fluorescent products
32
. Native lysates were collected and samples
were equally loaded to total protein content. Samples were split into 2 samples (A and B). A GCL reaction
cocktail (400 mM Tris, 40 mM ATP, 40 mM Glutamate, 40 mM Cysteine, 2 mM EDTA, 20 mM sodium
borate, 2 mM serine, 40 mM MgCl2) was added to sample A. Lysis buffer was added to sample B for
background measurements. Samples were incubated at 37C for 15 min, followed by quenching with 200
mM 5-sulfosalicyclic acid (SSA). Samples were placed on ice for 20 min then centrifuged at 2,000g for 10
min at 4C. The supernatant was collected and 20 µL was transferred to a 96 well plate designed for
fluorescent detection. A standard curve of γ-GCS was prepared (0 to 140 µM). 20 µL of each standard
solution was added to the 96 well plate. 180 µL of NDA derivatization buffer was added to each well,
followed by incubation for 30 min at RT in the dark. NDA-γ-GCS was measured by a plate reader (472 nm
excitation / 528 nm emission).
Bioinformatic analysis
Data from the cancer dependency map (DepMap
33
) was downloaded from the depmap data portal
(https://depmap.org/portal/). Cell lines were grouped by their cancer subtype and dependencies scores of
genes from glycolysis were correlated against xCT expression. Significance was assessed by permutations
shuffling dependency scores and xCT expression. Single-cell RNA-sequencing data from adult and
pediatric glioblastoma tumors
34
was downloaded from the Broad institute single cell portal
9
(https://portals.broadinstitute.org/single_cell). Cell assignment labels were taken from the original
publication.
DGSEA Methods
Enrichment Score 𝑬𝑺
𝑺 & 𝑬𝑺
𝑨𝑩
Calculation
The enrichment score ES is calculated using the Gene Set Enrichment Analysis
26
algorithm:
1. Rank the order of N genes in expression dataset according to their correlation with phenotype (i.e.
ranking metric).
2. Evaluate the fraction of genes in set S weighted by their correlation with the phenotype and the
fraction of genes not in S.
3. The Enrichment Score 𝐸𝑆
𝑆 is the maximum deviation from zero of P hit – P miss. The score is
calculated by walking down the gene list L and increasing a running-sum statistic by P hit when
encountering a gene in S and decreasing it by P miss when encountering a gene not in S.
4. For calculating 𝐸𝑆
𝐴𝐵
comparing two gene sets, the maximum deviation from zero for gene set B is
subtracted from the maximum deviation from zero for gene set A.
Estimating Significance
We estimate significance of the ES by using an empirical permutation test procedure that preserves the
ranking metric. For permutations, we shuffle the gene labels and recompute the ES of the gene set for the
shuffled data, which generates a null distribution for ES(A-B, ). The p-value is then calculated relative to
this null distribution, using the positive or negative portion of the distribution corresponding to the sign of
the observed ES(A-B) (Fig. 1).
Adjustment for Multiple Hypothesis Testing
10
We adjust the estimated significance level for differences for ES(A-B) by generating a null distribution that
compares A & B separately to all non-A&B gene sets, hence referred to as ES(X-Y). We then normalize
each ES for each gene set to account for differences in gene set size, separately rescaling the positive and
negative scores by dividing by the mean of the ES(A-B, ) generating NES(A-B), NES(X-Y), NES(A-B,
), and NES(X-Y, ). For a given NES(A-B) 0, the FDR is calculated as the ratio of the percent for all
NES(X-Y, ) NES(A-B) divided by the percentage of observed NES(X-Y) NES(A-B), and similarly if
NES(A-B) ≤ 0. For the analyses done in this study, the non-A&B gene sets used were metabolic pathways
defined by the Kyoto Encyclopedia of Genes and Genomes.
NCI60 Consumption and Release Analysis
Consumption and secretion rates for glucose and lactate were used from Jain. Et al.
35
. The raw data was
averaged per cell line. Gene expression data was downloaded from Gmeiner et al.
36
, then filtered for protein
coding transcripts. Gene expression data was centered and scaled across all cell lines, generating rank lists
for each cell line. GSEA and DGSEA were run using the weighted Kolmogorov-Smirnov-like statistic to
generate normalized enrichment scores (NES) for glycolysis, OxPhos, and DGSEA. The Spearman
correlation coefficient was calculated comparing NES to lactate secretion or glucose consumption.
Cancer Cell Line Encyclopedia Analysis
Gene expression and metabolomics data were downloaded from the CCLE. Gene expression data was
centered and scaled across all cell lines, then GSEA and DGSEA were run to generate NES for glycolysis,
OxPhos, and DGSEA. The Spearman correlation coefficient was calculated comparing NES to all
metabolite abundances.
Hypoxia Analysis
11
Gene expression data was downloaded from Ye and Fertig et al.
37
for 31 breast cancer cell lines in 20%
oxygen or 1% oxygen. For the results in Fig. 3B-C, the data was median normalized and the log 2 fold-
change was calculated for each gene upon subjection to 1% oxygen to generate ranked gene lists. Then,
GSEA and DGSEA were run to generate NES for glycolysis, OxPhos, Hypoxia, and DGSEA. The
HALLMARK_HYPOXIA M5891 gene set was used from MSigDB as a benchmark. Heatmaps were
generated using Morpheus, https://software.broadinstitute.org/morpheus . Mountain plots were generated
using R (see github for source code). For the results in Fig. 3D, the gene expression data was scaled and
centered for all 62 samples and then single-sample GSEA and DGSEA were run for glycolysis, OxPhos,
Hypoxia, and DGSEA.
Pathway Dependency Enrichment Analysis Methods
Data Sources
Cancer cell line gene expression data was downloaded from the Cancer Cell Line Encyclopedia (CCLE)
version 19Q4. Gene dependency data was downloaded from the Cancer Dependency Map (DepMap),
Achilles gene effect version 19Q4 was used for this study. Drug response data was downloaded from the
PRISM Repurposing database, version 19Q4 with secondary screen with dose response curve parameters
was used. Metabolic pathway annotations were downloaded from the Kyoto Encyclopedia of Genes and
Genomes (KEGG).
Simulation Studies
Gene expression data was simulated for 300 cell lines using a normal distribution for each cell line (µ = 0,
σ = 1). Then, a synthetic gene set of 25 genes was perturbed using a normal distribution gradient, where
cell line 1 would receive a value of µ = -X, σ = 1 and cell line 300 would receive a normal distribution of
µ = +X, σ = 1, with cell lines 2-299 receiving sequential values from -X to X. Single-sample Gene Set
Enrichment Analysis (ssGSEA) was calculated for the synthetic gene set for all 300 cell lines. Next, gene
dependency data was simulated for the same 300 cell lines using the same method. For both gene expression
12
and gene dependency data, values for X were varied from 0 to 1. Next, Spearman correlation coefficients
between synthetic gene set activity (NES) and gene dependency were calculated for all 16,643 genes.
Finally, Gene Set Enrichment Analysis was run to calculate the simulated Genetic Pathway Dependency
Enrichment Analysis values. For the Pharmacological PDEA simulation study in Figure 17, a similar
approach was used with 200 cell lines and 1390 drugs to mimic what was used in the real Pharmacological
PDEA study.
Calculation of Metabolic Pathway Expression
1019 cancer cell lines from the CCLE were separated by their culture type (adherent or suspension) and
then culture medium (RPMI or DMEM), respectively. Cell lines with missing information for either culture
type or medium were omitted. Gene expression values were unit normalized across all cell lines of the same
culture and medium type (e.g. Adherent-RPMI). Single-sample Gene Set Enrichment Analysis (ssGSEA)
across all metabolic pathways in the KEGG database was run on the normalized gene expression values,
giving normalized enrichment scores (NES) representing relative metabolic pathway activity for 69
metabolic pathways for each cell line.
Genetic Pathway Dependency Enrichment Analysis
For each metabolic pathway, the NES was correlated with the -CERES score for all 16,643 genes. For
Adherent RPMI cells, this meant correlating 300 data points. Since NES distributions are bimodal,
spearman correlations were used. The resulting correlations were ranked and GSEA querying KEGG
metabolic pathways was run to calculate Genetic Pathway Dependency Enrichment Analysis (Genetic
PDEA). Resulting positive NES represent increased essentiality upon increased metabolic pathway activity,
whereas negative NES represent increased essentiality upon decreased metabolic pathway activity.
Drug Response Correlations
13
For each metabolic pathway, the NES was correlated with the -auc (area under the curve) for 1448 anti-
cancer drugs in the PRISM repurposing database. Drugs with less than 150 cell lines were removed, leaving
1390 drugs. Spearman correlation p-values were calculated and a Benjamini-Hochberg false discovery rate
correction was applied for each metabolic pathway.
Pharmacological Pathway Dependency Enrichment Analysis
Drugs were mapped to their metabolic pathway using the annotated target(s) and genes from KEGG
metabolic pathways. Since the PRISM database contains both activators and inhibitors, we annotated all
activators by mechanism of action and multiplied their correlation coefficients by -1. Therefore, a pathway
activator would be counted similarly to a pathway inhibitor. Pathways with 4 or more drugs were kept.
Then, GSEA was run on the rank lists of 1390 correlation coefficients.
Integration of Individual Drug Response and Gene Dependency
Drug-gene dependency pairs were mapped using the target annotations for each drug. Correlation
coefficients for each drug and gene dependency were summed for each metabolic pathway, generating
187,818 drug-gene-pathway combinations. An empirical permutation test was run sampling 1000
combinations of each drug-gene correlation coefficient. P-values were calculated by dividing the number
of permutations that out-performed the real summed correlation coefficients by the number of same-signed
permutations. P-values were then adjusted using a Benjamini-Hochberg correction for each metabolic
pathway.
Integration of Genetic PDEA and Pharmacological PDEA
Results from Genetic PDEA and Pharmacological PDEA were filtered for same signed NES and p-values
of less than 0.05 and FDR values of less than 0.25 (per the original GSEA algorithm). 3 pathway-drug-gene
dependencies were identified out of a possible 3220 combinations.
14
Chapter 1: A synthetic lethal drug combination mimics glucose deprivation –
induced cancer cell death in the presence of glucose
1.1 Metabolomics identifies accumulation of L-cystine and L-cysteine and depletion of
glutathione as markers of glucose deprivation sensitivity
To investigate the metabolic profile of glucose deprivation-induced cell death, we first tested the
response of a panel of six GBM and two sarcoma cell lines to glucose deprivation. Among the GBM cell
lines, we found that A172 exhibited strong resistance, LN229 and U118MG exhibited medium sensitivity,
and LN18, T98, and U87 exhibited high sensitivity to glucose deprivation (Fig. 1A)
16,38
. For the sarcoma
cell lines, HT161 and TC32 exhibited resistance and sensitivity, respectfully, to glucose deprivation
(Supporting Fig. 1A). We then used liquid chromatography-mass spectrometry (LC-MS) metabolomics to
measure changes in metabolite abundance and TCA cycle flux upon glucose deprivation
39–41
. Using two
sensitive cell lines (LN18, T98) and one medium resistant cell line (LN229), we quantified levels of 97
polar metabolites in cells cultured with or without glucose for 3 h (Supporting Table 1). To identify
metabolites correlated with glucose deprivation sensitivity, we ranked metabolite abundances by a ratio-of-
ratios comparing sensitive to resistant cells (Fig. 1B). This metric identified L-cystine (the oxidized dimer
of L-cysteine, rank #1), L-cysteine (#2), and reduced glutathione (GSH) (#97) as the most differentially-
changed metabolites between sensitive and resistant cells. In addition, the GSH precursor -
glutamylcysteine ( -GCS) was also differentially regulated between sensitive and resistant cells (rank 93 of
97).
Examination of these differentially-regulated metabolites revealed that the glucose deprivation-
sensitive T98 and LN18 cells exhibited 800- and 30-fold accumulation of L-cystine and L-cysteine,
respectively (Fig. 1C). In contrast, reduced GSH was completely depleted and -GCS levels dropped by
~80% following glucose deprivation in sensitive T98 and LN18 cells (Fig. 1C). In contrast, LN229 cells
exhibited little to no change in levels of these metabolites. Next, we used a targeted LC-MS metabolomics
15
assay to confirm these trends in one additional sensitive cell line (U87) and one additional resistant cell line
(A172). Indeed, we corroborated that sensitive U87 cells accumulated L-cystine and L-cysteine and
depleted GSH following 3 h of glucose deprivation, whereas the highly resistant cell line A172 showed no
significant changes in these metabolites (Supporting Fig. 1B). Together, these data demonstrate that glucose
deprivation sensitive cells exhibit a metabolic signature of L-cystine and L-cysteine accumulation and GSH
depletion upon glucose deprivation.
We next investigated changes in TCA cycle flux upon glucose deprivation by incubating cells with
L-[U-
13
C]-glutamine in the presence or absence of glucose. We found that >90% of glutamine was labeled
in 3 h in both resistant (LN229) and sensitive (LN18, T98) cell lines (Supporting Fig. 1C and Supporting
Table 2 and Supporting Table 3). Examination of the isotopomer distributions for TCA cycle metabolites
showed a significant decrease in the percentage of citrate M5 upon glucose deprivation, indicating
decreased reductive carboxylation flux from α-ketoglutarate to citrate in glucose-deprived cells
42,43
(Fig.
1D). In contrast, forward flux through the TCA cycle was increased upon glucose deprivation in sensitive
but not resistant cell lines, as indicated by increased labeling of malate M4 and fumarate M4 upon glucose
deprivation (Fig. 1D). Interestingly, both sensitive and resistant cell lines exhibited increased labeling of L-
aspartate M4 upon glucose starvation, further suggesting increased forward flux through the TCA cycle
(Supporting Fig. 1C). Malate and fumarate pool sizes were not consistently affected by glucose deprivation
in either direction, whereas citrate and aspartate levels were substantially decreased and increased,
respectively, in all three cell lines (Fig. 1E and Supporting Fig. 1D). Taken together, these data demonstrate
that cancer cells shift towards oxidative, rather than reductive, glutamine metabolism when deprived of
glucose.
16
17
Figure 1. Accumulation of L-cystine and L-cysteine and depletion of reduced glutathione are
metabolic markers of sensitivity to glucose deprivation.
A) A panel of GBM cell lines was subjected to glucose deprivation for 24 h, and viability was measured by
trypan blue exclusion. A172 exhibited strong resistance, LN229 and U118MG cells exhibited medium
resistance, and T98, LN18, and U87 cells exhibited high sensitivity to glucose deprivation. N.S. denotes p
> 0.05, * denotes p < 0.05, ** denotes p < 0.01, *** denotes p < 0.001 by Student’s t-test (n=3). B) Two
glucose deprivation-sensitive cell lines (T98, LN18) and one medium resistant cell line (LN229) were
deprived of glucose for 3 h in the presence of L-[U-
13
C]-Glutamine and then profiled by LC-MS
metabolomics. A total of 97 metabolites were identified and quantified. Metabolite pool sizes were ranked
using a log 2 ratio-of-ratios metric ([Average of T98 -/+ glucose and LN18 -/+ glucose] / [LN229 -/+
glucose]). L-cystine and L-cysteine, and reduced glutathione (GSH) were the most differentially regulated
metabolites (red bars). -glutamylcysteine ( -GCS) was also differentially regulated in glucose deprivation-
sensitive cells (rank 93 of 97). C) Bar plots from LC-MS metabolomics in B) showing that L-cystine and
L-cysteine were accumulated, and GSH and -GCS were depleted in glucose deprivation-sensitive GBM
cells following 3 h of glucose deprivation. * denotes p < 0.05, ** denotes p < 0.01, *** denotes p < 0.001
by Student’s t-test (n=3). D) Individual bar plots from LC-MS metabolomics showing that the GSH
synthesis precursors glycine and L-glutamate, as well as oxidized glutathione (GSSG) and ATP, were not
significantly regulated by 3 h of glucose deprivation in either resistant (LN229) or sensitive cells (T98,
LN18). E) Log 2 fold change in metabolite pool sizes upon 3 h of glucose deprivation overlaid on the GSH
synthesis pathway. Red and blue represent accumulation and depletion, respectively, as shown on the
indicated color scale. L-gln, L-glutamine; L-glu, L-glutamate; Gly, glycine. Error bars are standard
deviation of the mean.
Supporting Figure 1. Glucose deprivation sensitivity of sarcoma cell lines and total cell number following
glucose deprivation.
A) Two sarcoma cell lines were subjected to glucose deprivation, and viability was measured by trypan
blue exclusion. HT161 exhibit strong resistance (<15% cell death at 24 h post glucose deprivation), whereas
TC32 cells are highly sensitive to glucose deprivation (<25% viability at 12 h post glucose deprivation). *
denotes p < 0.05, *** denotes p < 0.001 by Student’s t-test. B) Total cell number is unchanged upon glucose
deprivation. Three glucose deprivation-resistant and three glucose deprivation-sensitive cell lines were
starved of glucose for 24 hours and viability was measured by trypan blue exclusion. Total cell numbers
were normalized to plus glucose control (0-hour time point).
18
1.2 Glutamate-cysteine ligase (GCL) activity moderately regulates resistance to glucose
deprivation.
The ligation of L-glutamate and L-cysteine is catalyzed by glutamate-cysteine ligase (GCL), which
consists of a catalytic subunit (GCLC) and a modifier subunit (GCLM). GCLM binding can increase the
activity of GCL up to 10 fold (27) by decreasing the Km for glutamate and increasing the Ki for GSH (28).
We thus first tested whether modulating GCL activity could affect sensitivity to glucose deprivation.
Indeed, the GCL inhibitor buthionine sulfoximine (BSO) sensitized both resistant LN229 and HT161 cells
to glucose deprivation (Fig. 2A and Supporting Fig. 4A). In addition, overexpression of GCLM, but not
GCLC, using CRISPRa (dCas9-VPR (29)) promoted resistance in glucose deprivation-sensitive T98 and
LN18 cells (Supporting Fig. 4B). Because the magnitude of GCLC and GCLM overexpression using
CRISPRa was modest, we validated our CRISPRa findings with lentiviral overexpression of GCL subunits
which we hypothesized would provide more substantial overexpression. However, despite increased
overexpression, exogenous GCLM but not GCLC conferred only a modest but significant increase in
resistance to glucose deprivation in T98 cells, similar to our CRISPRa results. (Fig. 2B). To test whether
GCLC had become limiting upon GCLM overexpression, we co-overexpressed GCLC and GCLM and did
not observe increased resistance to glucose deprivation relative to GCLM alone (Fig. 2B). Taken together,
this data suggests that GCL only modestly regulates sensitivity to glucose deprivation.
1.3 Glutamate-cysteine ligase (GCL) is activated by glucose withdrawal.
We next tested whether glucose deprivation regulates GCL activity. First, we examined whether
glucose deprivation affected the expression of GCLC and GCLM. Western blotting revealed that neither
GCLC nor GCLM expression was significantly changed following glucose deprivation (Supporting Fig.
4C). Interestingly, the sensitive cell lines T98 and U87 exhibited higher GCLC and GCLM expression than
the resistant cell line LN229. Next, using native PAGE gels to monitor the formation of the GCL
holoenzyme, we found that 60 min of glucose deprivation increased holoenzyme formation in T98 sensitive
cells to an extent similar to the positive control H2O2 (Fig. 2C) (30). The resistant cell line LN229 also
demonstrated formation of the GCL holoenzyme following glucose deprivation, although holoenzyme
19
formation was slower (180 min) and less expressed than in T98 sensitive cells. Finally, using an in vitro
assay (31), we directly tested GCL activity and found that glucose deprivation increased GCL activity in
glucose deprivation-sensitive T98 but not -resistant LN229 cells (Fig. 2D). Although our metabolomic data
had suggested GCL activity was inhibited in sensitive cells following glucose deprivation (Fig. 1E), these
data demonstrate that GCL activity is in fact increased following glucose deprivation in sensitive cells.
This, in turn, suggests the possibility that GSH depletion following glucose deprivation in sensitive cells
occurs not because of inhibited GSH synthesis, but rather because of extremely rapid GSH consumption.
Figure 2. Glutamate-cysteine ligase (GCL) activity moderately regulates resistance to glucose
deprivation.
A) Treatment with the GCL inhibitor buthionine sulfoximine (BSO, 500 µM) sensitized glucose
deprivation-resistant LN229 GBM cells to glucose deprivation. Cells were treated with either solvent
(water) or BSO in the presence or absence of glucose for 16 h, and viability was measured by trypan blue
exclusion. * denotes p < 0.05 by Student’s t-test (n=2). B) Overexpression of GCLM, but not GCLC,
confers resistance to glucose deprivation. Lentiviral vectors (pLX302 and pLX304) were used to
20
overexpress either subunit of GCL. Glucose deprivation-sensitive T98 cells were starved of glucose for 8
h and viability was measured by trypan blue exclusion. Overexpression of GCLM, but not GCLC, increased
survival. Combined overexpression of GCLC and GCLM did not confer additional resistance compared to
GCLM alone. ** denotes p < 0.01 by Student’s t-test (n=3). C) The GCL holoenzyme is formed rapidly
after glucose deprivation in both sensitive and resistant cells. The glucose deprivation-sensitive T98 and -
resistant LN229 cell lines were starved of glucose for the indicated time and then lysed in a non-reducing
lysis buffer. Treatment with 10 mM H 2O 2 for 10 min was included as a positive control. Cell lysates were
run in native PAGE conditions, and formation of the GCL holoenzyme was monitored by formation of the
GCLC:GCLM complex (101 kDa) using a GCLC antibody. D) GCL activity increases in response to
glucose deprivation in sensitive T98 but not resistant LN229 cells. T98 and LN229 cells were starved of
glucose for 3 h, lysates were collected, and GCL activity was measured using a fluorescence-based
microtiter plate assay (31). ** denotes p < 0.01 by Student’s t-test (n=2). Error bars are standard deviation
of the mean.
21
Supporting Figure 2. GCL activity moderately regulates sensitivity to glucose deprivation.
A) Log 2 fold change data from LN229, T98, and LN18 cells overlaid on glutathione metabolism pathway.
B) Treatment with the GCL inhibitor, buthionine sulfoximine (BSO, 500 µM), sensitized glucose
deprivation-resistant HT161 sarcoma cells to glucose deprivation. Cells were treated with either solvent
22
(water) or BSO in the presence of absence of glucose for 24 h, and viability was measured by trypan blue
exclusion. * denotes p < 0.05 by Student’s t-test (n=3).
C) Overexpression of GCLM but not GCLC increased resistance to glucose deprivation in sensitive LN18
and T98 cells. Cells were first infected with CRISPRa machinery (dCas9-VPR
49
) and then infected with a
second vector carrying the indicated guide RNAs. Cells were then cultured by 8 h in the presence or absence
of glucose, and viability was measured by trypan blue exclusion. * denotes p < 0.05, ** denotes p < 0.01
by Student’s t-test (n=3). Western blotting with antibodies against GCLC, GCLM, and the equal loading
control actin confirmed overexpression. NT, non-targeting control.
D) Expression of both GCLC and GCLM is stable upon glucose deprivation in both glucose deprivation-
sensitive and -resistant cell lines. Glucose deprivation sensitive (T98, LN18, TC32) and resistant (LN229,
HT161) cell lines were deprived of glucose for the indicated time points. Lysates were collected and western
blots were analyzed with antibodies against GCLC, GCLM, and the equal loading control actin showed
stable expression upon glucose deprivation in all cell lines.
1.4 NADPH is the limiting reducing agent in glucose-deprived cancer cells.
In sensitive cells, it has been shown that glucose deprivation induces high levels of reactive oxygen
species (ROS), and that elevated ROS levels are responsible for glucose deprivation-induced cell
death
14,16,38
. We thus measured ROS levels at multiple time points following glucose deprivation. In
sensitive T98 cells, ROS levels were increased as early as 10 min after glucose deprivation and steadily
increased over two h (Fig. 3A and Supporting Fig. 3A). We next performed metabolomic profiling of
sensitive T98 cells at the same time points (Supporting Table 4). Interestingly, we found that L-cystine and
L-cysteine accumulated 8- and 6-fold, respectively, within only 10 min of glucose deprivation (Fig. 3B). In
contrast, levels of GSH did not decrease until after 60 min of glucose deprivation, and levels of GSSG
increased slightly over time after glucose deprivation (Fig. 3C). This result is consistent with our finding
that GCL is active upon glucose deprivation (Fig. 2). Notably, other metabolic destinations of L-cysteine
including cystathionine, taurine, and hypotaurine did not dramatically increase upon glucose deprivation
(Supporting Fig. 3B), suggesting that L-cysteine is not diverted to other metabolic pathways in glucose-
deprived cells. Taken together, these data demonstrate that intracellular accumulation of L-cystine, L-
cysteine, and ROS precedes GSH depletion.
NADPH is primarily generated by three main cellular metabolic sources: the oxidative pentose
phosphate pathway (oxPPP), malic enzyme, and serine-derived one carbon metabolism
50
. Since glucose
23
fuels both the oxPPP and the TCA cycle, we hypothesized that NADPH may be limiting when cells are
deprived of glucose. We thus profiled the metabolome at the same time points using an LC-MS
metabolomics method designed to detect NAD(P)H
31
. We found that levels of NADPH decreased
dramatically within 10 min of glucose deprivation and were undetectable after 60 min (Fig. 3C). Notably,
this short time scale is similar to the time required for L-cystine and L-cysteine accumulation following
glucose deprivation. In contrast, intracellular pools of GSH and NADH were not depleted until later time
points (90-120 min). These results suggest that NADPH, rather than NADH or GSH, is the limiting reducing
agent when cancer cells are deprived of glucose.
To better understand the global metabolic changes upon glucose deprivation, we calculated the
Pearson correlation coefficient between all pairs of metabolites. Hierarchical clustering of these Pearson
correlation coefficients revealed a cluster of metabolites from glycolysis, the TCA cycle, and redox
metabolism which are all rapidly depleted following glucose deprivation (Fig. 3D). A second large cluster
included 18 of 20 proteinogenic amino acids and these metabolites were largely unperturbed by glucose
deprivation. The only proteinogenic amino acids not found in this cluster were L-cysteine, which clustered
with L-cystine, and L-proline which clustered with glycolytic and redox metabolites. Examination of L-
proline levels revealed that intracellular concentrations of this amino acid steadily dropped after glucose
deprivation (Fig. 3E). This may occur because DMEM does not contain L-proline, forcing cells to rely on
de novo synthesis, but glucose-starved cells cannot provide the NADPH required for conversion of
pyrroline-5-carboxylate to L-proline. Together, these results demonstrate that NADPH is the limiting
reducing agent in glucose deprived cancer cells.
24
Figure 3. NADPH is the limiting reducing agent upon glucose deprivation.
A) Reactive oxygen species (ROS) accumulate upon glucose deprivation. ROS was measured by flow
cytometry using DCF-DA at the indicated time points. ROS accumulation steadily increased with time after
glucose deprivation. B) Glucose deprivation-sensitive T98 cells were deprived of glucose and metabolites
25
were extracted at the indicated time points. Accumulation of L-cystine and L-cysteine occurred within 10
min after glucose deprivation and steadily increased over the course of 2 h. C) NADPH is rapidly consumed
following glucose deprivation. An LC-MS metabolomics method designed to detect NAD(P)H
31
was used
at the same time points as 3B. Depletion of NADPH preceded depletion of NADH and GSH, with NADPH
dropping below the lower limit of detection within 60 min. D) The Pearson correlation coefficient was
calculated between each metabolite, and hierarchical clustering was performed on the Pearson correlation
coefficients. Clustering revealed that L-cystine, L-cysteine, and L-proline behave differently from all other
proteinogenic amino acids. Glycolysis metabolites and reducing agents clustered together, reflecting that
they both deplete upon glucose deprivation. 18 of 20 proteinogenic amino acids also clustered together,
reflecting that amino acid metabolism and glycolysis are uncoupled upon glucose deprivation.E) L-proline
is steadily depleted after glucose deprivation in sensitive T98 cells. The de novo synthesis of L-proline
requires NADPH to convert pyrroline-5-carboxylate to L-proline.
Supporting Figure 3. Temporal measurements of reactive oxygen species and metabolite abundances upon
glucose deprivation.
A) Reactive oxygen species (ROS) accumulate upon glucose deprivation. The oxidation dependent
fluorogen DCF-DA was used to measure ROS accumulation upon glucose deprivation. T98 cells were
either starved of glucose (blue) or supplemented with fresh DMEM with glucose (red) for the indicated
times. Upon incubation with DCF-DA, cells were analyzed by flow cytometry.
26
B) Metabolite abundances of cystathionine, taurine, and hypotaurine do not increase upon glucose
deprivation. Glucose deprivation-sensitive T98 cells were starved of glucose, and metabolite levels were
measured by LC-MS metabolomics at the indicated time points. While levels of L-cysteine dramatically
increased upon glucose deprivation (Fig. 3B), the alternative metabolic destinations of L-cysteine did not
show substantial increases.
1.5 L-cystine import, but not L-cysteine import, induces ROS mediated cell death in glucose-
deprived cancer cells.
The DMEM in which GBM cells are cultured contains 200 µM L-cystine but 0 µM L-cysteine.
Thus, the rapid (<10 min) accumulation of L-cystine and L-cysteine following glucose deprivation is likely
driven by L-cystine import and subsequent reduction to two molecules of L-cysteine. Notably, the reduction
of L-cystine to L-cysteine consumes one reducing equivalent of NADPH
51
. Because NADPH is depleted
on the same rapid time scale as L-cystine and L-cysteine accumulation (~10 min), we hypothesized that the
intracellular conversion of L-cystine to L-cysteine might drive glucose deprivation-induced NADPH
depletion, ROS generation, and subsequent cell death. Therefore, we first tested whether L-cystine
deprivation could rescue cells from glucose deprivation. Indeed, sensitive T98 cells exhibited very high
viability after 8 h of glucose deprivation in cystine-free media, and supplementation with L-cystine restored
glucose deprivation-induced cell death (Fig. 4A). Because GBM cells express L-cysteine transporters
52
including EAAT3, ASCT1, SNAT1, and SNAT2, we next tested whether supplementation with L-cysteine,
as opposed to L-cystine, could promote cell death in glucose starved cells. Although extracellular L-
cysteine was consumed by T98 cells (Supporting Fig. 4A), L-cysteine supplementation did not increase cell
death relative to cystine-free media. Notably, supplementation with L-cysteine did not confer greater
resistance to glucose deprivation relative to L-cystine deprivation (data not shown). This may be due to the
fact that L-cysteine spontaneously oxidized to L-cystine in cell culture media (not shown), suggesting that
L-cysteine supplementation can result in L-cystine import. Thus, we tested whether supplementation with
the GSH precursor downstream of L-cysteine, γ-GCS, could affect cell viability following glucose
deprivation. Indeed, supplementation with γ-GCS in L-cystine-free media conferred a modest but
27
significant resistance to glucose deprivation (Fig. 4B). These results demonstrate that L-cystine import, but
not L-cysteine import, contributes to glucose deprivation-induced cell death.
We next sought to understand the metabolic consequences of L-cystine and L-cysteine import upon
glucose deprivation. Using LC-MS metabolomics, we found that L-cystine starvation prevented the
accumulation of L-cystine and L-cysteine that normally occurs upon glucose deprivation in sensitive T98
cells (Fig. 4C). Supplementation of glucose-free media with L-cysteine instead of L-cystine led to increased
intracellular levels of both L-cysteine and L-cystine, which may reflect either intracellular or extracellular
oxidation of L-cysteine. We next asked whether the presence of L-cysteine or L-cystine during glucose
starvation induced oxidative stress. Plotting the ratios of GSH/GSSG and NADPH/NADP+ indicated that
sensitive T98 cells experienced oxidative stress (lower GSH/GSSG ratio and NADPH/NADP+ ratios) when
deprived of glucose in media containing L-cystine but not in glucose-free media without L-cystine or
glucose-free media supplemented with L-cysteine (Fig. 4C). Together, this indicates that the presence of L-
cystine but not L-cysteine during glucose deprivation promotes GSH depletion and redox imbalance.
We next tested whether L-cystine import during glucose deprivation induces ROS. In glucose
deprivation-sensitive T98 cells, 3 h of glucose deprivation in the presence of L-cystine induced a six-fold
induction in ROS (Fig. 4D). Meanwhile, cells deprived of glucose in L-cystine-free medium exhibited only
a three-fold induction of ROS, demonstrating that L-cystine import significantly contributes to ROS
accumulation. We next sought to uncouple glucose deprivation from L-cystine import by starving cells of
glucose and L-cystine for 3 h and then supplementing cells with either L-cystine or L-cysteine for an
additional 3 h. Consistent with our hypothesis that L-cystine import drives ROS accumulation in glucose-
starved cells, L-cystine supplementation increased ROS levels. Addition of L-cysteine, in contrast,
decreased ROS levels compared to an additional 3 h without L-cystine or glucose. Next, we used LC-MS
metabolomics to profile the metabolic changes that occurred after supplementation of L-cystine. Again,
sensitive T98 cells were starved of glucose and L-cystine for 3 h, and then L-cystine was spiked into the
28
cell culture media. Following L-cystine addition, levels of intracellular L-cystine and L-cysteine levels
accumulated rapidly, with levels increasing within just 10 min of L-cystine spike-in (Fig. 4E). In addition,
the ratios of GSH/GSSG and NADPH/NADP
+
were dramatically reduced within 10 min of L-cystine spike-
in. Taken together, these data demonstrate that import of L-cystine, but not L-cysteine, promotes ROS
29
accumulation, GSH depletion, and redox imbalance in sensitive cancer cells deprived of glucose.
Figure 4. L-cystine import induces oxidative stress and ROS-mediated cell death in glucose deprived
cancer cells.
A) L-cystine, but not L-cysteine, import results in glucose deprivation-induced cell death. Glucose
deprivation-sensitive T98 cells were cultured in L-cystine-free medium and subjected to glucose
30
deprivation for 8 h. L-cystine starvation rescued T98 cells from glucose deprivation-induced cell death.
Addition of L-cystine (200 µM), but not L-cysteine (200 µM), sensitized T98 cells to glucose deprivation.
B) -GCS contributes resistance to glucose and L-cystine deprivation. Glucose deprivation-sensitive T98
cells were starved of glucose and L-cystine for 24 h. Addition of 200 µM -GCS conferred a modest
resistance compared to DMSO control. ** denotes p-value < 0.01 by Student’s t-test (n = 3).
C) L-cystine induces oxidative stress upon glucose deprivation. T98 cells were deprived of glucose in L-
cystine-free medium for 3 h and metabolites were quantified using LC-MS metabolomics. Addition of L-
cystine (200 µM), but not L-cysteine (200 µM), induced GSH depletion and oxidative stress as measured
by the ratio of GSH / GSSG and NADPH / NADP+.
D) ROS accumulation following glucose deprivation is driven by L-cystine import. T98 sensitive cells were
deprived of glucose in the presence or absence of L-cystine for 3 h, and ROS levels were measured by flow
cytometry by DCF-DA staining. Cells that had been starved of glucose and L-cystine for 3 h were then re-
supplemented with L-cystine (200 µM), L-cysteine (200 µM), or neither for an additional 3 h. (Left) Mean
fluorescent intensity of DCF-DA signal. (Center) Histograms of DCF-DA for T98 cells cultured with
glucose, without glucose but with L-cystine, or without glucose or L-cystine for 3 h. (Right) Histograms of
DCF-DA intensity for T98 cells starved of glucose and L-cystine for 3 h and then re-supplemented with L-
cystine, L-cysteine, or neither for an additional 3 h.
E) L-cystine import induces redox imbalance upon glucose deprivation. Glucose deprivation-sensitive T98
cells were starved of glucose for 3 h in the presence and absence of L-cystine, at which point L-cystine (200
µM) was spiked-in for 10, 30, and 60 min. Addition of L-cystine induced a redox imbalance as indicated
by dramatic decreases in the ratios GSH / GSSG and NADPH / NADP+.
Supporting Figure 4. L-cysteine is consumed by T98
cells. L-cystine-free medium was supplemented with
either nothing (control), L-cysteine (200 µM), or L-
cystine (200 µM). T98 cells and control without cells
were incubated for 3 h in each medium. Extracellular
metabolites were extracted and measured using LC-MS
metabolomics. L-cysteine was consumed by glucose
deprived T98 cells. * denotes p < 0.05 by Student’s t-
test.
1.6 L-cystine import from the glutamate/cystine
antiporter xCT/SLC7A11 promotes cell death
following glucose starvation.
The primary source for intracellular L-cystine is the glutamate/cystine antiporter, system x c
-
.
System x c
-
comprises a heterodimer containing a non-specific heavy chain, SLC3A2, and a specific light
chain, xCT/SLC7A11. Recent reports have shown that system x c
-
activity promotes cancer cell dependency
on glucose
38,53,54
, although the mechanism by which system x c
-
promotes glucose deprivation-induced cell
31
death remains disputed. Therefore, we sought to test the role of system x c
-
in L-cystine accumulation and
cell death observed following glucose deprivation. First, by Western blotting, we found that expression of
xCT/SLC7A11 correlated with sensitivity to glucose deprivation in both glioblastoma and sarcoma cell
lines (Fig. 5A and Supporting Fig. 5A). Next, we tested whether pharmacological and genetic inhibition of
system x c
-
affected sensitivity to glucose deprivation. Indeed, pharmacological inhibition of xCT with either
sulfasalazine (SASP) or erastin prevented glucose deprivation-induced cell death (Fig. 5B-C and
Supporting Fig. 5B). In addition, knockdown of xCT with CRISPRi
29
(dCas9-KRAB) rescued LN18 cells
from glucose deprivation (Fig. 5D). These findings corroborate recent reports that system x c
-
activity and
expression is required for glucose deprivation-induced cell death
38,53,54
.
Because system x c
-
both imports L-cystine and exports L-glutamate, we next tested whether L-
glutamate export, in addition to L-cystine import, plays a role in glucose deprivation-induced cancer cell
death. We thus cultured cells in media with L-cystine but without L-glutamate, media lacking L-cystine
and L-glutamate, or media lacking L-cystine but supplemented with L-glutamate (i.e., mimicking system
x c
-
activity). In both T98 and LN18 sensitive GBM cells, L-cystine starvation rescued cells from glucose
deprivation-induced cell death, but L-glutamate supplementation did not restore cell death (Fig. 5E). These
results demonstrate that import of L-cystine from system x c
-
, but not export of L-glutamate, regulates
glucose deprivation-induced cell death.
To further understand the metabolic effect of system x c
-
activity upon glucose deprivation, we
performed LC-MS metabolomics on sensitive LN18 cells deprived of glucose in the presence or absence
of the xCT inhibitor SASP. As expected, SASP treatment prevented the accumulation of L-cystine and L-
cysteine following glucose deprivation (Fig. 5F). In the presence of SASP, the basal concentration of GSH
in cells cultured with glucose was reduced 67%, reflecting the fact that SASP-treated cells must rely on de
novo synthesis of L-cysteine (rather than L-cystine import) in order to synthesize GSH. Surprisingly, when
cells were starved of glucose in the presence of SASP, the levels of GSH increased rather than decreased
32
(Fig. 5F). Taken together with our metabolomic profiling of L-cystine starved cells (Fig. 4C), these results
demonstrate that system x c
-
activity promotes glucose deprivation-induced cell death through import of L-
cystine, depletion of GSH and consumption of NADPH by L-cystine reduction to L-cysteine.
Figure 5. L-cystine import from the glutamate/cystine antiporter, system xc-, is required for glucose
deprivation-induced cell death.
33
A) Expression of the specific light chain of system x c
-
, xCT/SLC7A11, correlates with sensitivity to glucose
deprivation in GBM cell lines. The resistant and sensitive GBM cell lines were starved of glucose for the
indicated times and then lysed. Expression of xCT/SLC7A11 was assessed by Western blotting. Actin was
used as an equal loading control. B-C) Pharmacological inhibition of xCT rescued sensitive cells from
glucose deprivation-induced cell death. The indicated sensitive GBM cells were treated with either
sulfasalazine (SASP, 500 µM) or erastin (10 µM) in the presence or absence of glucose for 8 h, and viability
was assessed by trypan blue exclusion. *** denotes p < 0.001 by Student’s t-test (n=3). D) Genetic
knockdown of xCT promoted resistance to glucose deprivation in sensitive LN18 cells. Cells were first
infected with CRISPRi machinery (dCas9-KRAB (69)) and then infected with a second vector carrying the
indicated guide RNA. Cells were then cultured 8 h in the presence or absence of glucose, and viability was
measured by trypan blue exclusion. * denotes p < 0.05 by Student’s t-test (n=3). Western blotting with
antibodies against xCT and the equal loading control actin confirmed knockdown. NT, non-targeting
control. E) L-cystine starvation rescues sensitive cells from glucose deprivation-induced death. The
sensitive GBM cell lines were cultured in L-cystine-free DMEM in the presence and absence of glucose
for 16 h. The media was supplemented with either L-cystine (200 µM) or L-glutamate (100 µM), and
viability was measured by trypan blue exclusion. *** denotes p < 0.001 by Student’s t-test (n=3). F)
Inhibition of the xCT cystine/glutamate antiporter prevents GSH depletion following glucose deprivation.
The sensitive GBM cell line LN18 was cultured with and without glucose in the presence or absence of the
xCT inhibitor sulfasalazine (SASP, 500 µM) for 3 h, and then intracellular metabolite concentrations were
measured using LC-MS metabolomics. xCT inhibition prevented the accumulation of L-cystine and L-
cysteine and depletion of GSH following glucose deprivation. ** denotes p < 0.01, *** denotes p < 0.001
by Student’s t-test (n=3). Error bars are standard deviation of the mean.
Supporting Figure 5. L-cystine import from the glutamate/cystine antiporter, system x c
-
, promotes glucose
deprivation-induced
cell death.
A) Expression of the specific light chain of
system x c
-
, xCT/SLC7A11, correlated with
sensitivity to glucose deprivation in
sarcoma cell lines. The resistant cell line
HT161 and the sensitive cell line TC32
were starved of glucose for the indicated
time and then lysed. Expression of
xCT/SLC7A11 was assessed by Western
blotting. Actin was used as an equal
loading control.
B) Pharmacological inhibition of xCT
rescued sensitive cells from glucose
deprivation-induced cell death. The
sensitive sarcoma cell line TC32 was
treated with either sulfasalazine (SASP,
500 µM) or erastin (10 µM) in the presence
or absence of glucose for 8 h, and viability
was assessed by trypan blue exclusion. ***
denotes p < 0.001 by Student’s t-test (n=3).
34
1.7 Inhibition of glucose metabolism is synthetically lethal with inhibition of GCL.
Building upon our observations that depletion of NADPH and GSH promotes glucose deprivation-
induced cell death, we sought to identify drug combinations that mimic this phenotype in the presence of
glucose. To identify candidate glycolytic nodes to target, we analyzed data from the Cancer Dependency
Map (DepMap
33
) to identify glycolytic genes whose CERES dependency score was correlated with high
xCT expression in glioblastoma (GBM) and Ewing’s sarcoma cell lines, since xCT expression correlates
with sensitivity to glucose deprivation in those cell types (Fig. 5A and Supporting Fig. 5A). Of 22 glycolytic
enzymes, we found that xCT expression was most strongly correlated with dependency on the glucose
transporter GLUT1 (SLC2A1) in GBM and also strongly correlated with dependency on GLUT1 in sarcoma
cell lines (i.e., high xCT expressing cells have large negative CERES scores, indicating greater dependency
on GLUT1) (Fig. 6A and Supporting Fig. 6A-C). We thus hypothesized that glucose deprivation sensitive
GBM and sarcoma cell lines would be sensitive to combined inhibition of GLUT1 and GSH synthesis. To
test this hypothesis, we treated T98 sensitive GBM cells with increasing concentrations of an inhibitor of
GLUT1 (STF-31
23
), an inhibitor of GCL (BSO), or the combination of both inhibitors. Treatment with
either drug alone showed little effect on viability (Fig. 6B), whereas co-inhibition of GLUT1 and GCL
synergistically induced cell death in T98 cells, as measured by calculating the Combination Index
56
(CI <
1 indicates synergy). Because STF-31 has been reported to inhibit both GLUT1
23
and NAMPT
57
, we
investigated whether NAMPT may play a role in combined STF-31 and BSO treatment. Using the DepMap
data, we asked whether NAMPT dependency correlated with xCT expression but found no correlation
between NAMPT dependency and xCT expression in either GBM or Ewing’s sarcoma cell lines
(Supporting Fig. 6D). Taken together, these results demonstrate that combined inhibition of GLUT1 and
GCL provoke synergistic cell death in T98 cells.
We next sought to test whether sensitivity to glucose deprivation correlated with response to
combined GLUT1 and GCL inhibition. Using a panel of GBM cell lines with either high (T98), medium
(LN229), or low (A172) sensitivity to glucose deprivation, we treated cells with STF-31, BSO, or the
35
combination of both drugs. Treatment with individual drugs did not cause significant cell death in any of
the three cell types. However, at 36 h, highly glucose deprivation sensitive T98 cells treated with the
combination of STF-31 and BSO exhibited a dramatic reduction in cell viability compared to either
treatment alone (Fig. 6C). Medium glucose deprivation sensitive LN229 cells also responded to the drug
combination but only after 60 h of treatment. Finally, low glucose deprivation sensitivity A172 cells showed
a modest response to combined STF-31 and BSO treatment at 72 h. Similar results were observed in
sarcoma cell lines with high and medium sensitivity to glucose deprivation (Supporting Fig. 6E). Together,
these results demonstrate that sensitivity to glucose deprivation correlates with sensitivity to a drug
combination blocking glucose transport (GLUT1) and GSH synthesis (GCL).
Next, to test the role of system x c
-
in response to combined GLUT1 and GCL inhibitors, we
evaluated the response to the drug combination in xCT knockdown cells. In sensitive LN18 cells, we found
that xCT knockdown conferred resistance to combined STF-31 and BSO treatment compared to non-
targeting control (Fig. 6D). We next tested whether L-cystine import, and subsequent NADPH
consumption, contributed to cell death in combined STF-31 and BSO treatment. Culturing T98 cells with
L-cysteine rather than L-cystine again conferred resistance to the drug combination (Fig. 6E), suggesting
that L-cystine reduction contributes to cell death upon combined STF-31 and BSO treatment. These data
further support that L-cystine import from xCT is important for response to combined GLUT1 and GCL
inhibition.
To further evaluate whether combined STF-31 and BSO treatment mimics glucose deprivation-
induced cell death, we measured ROS accumulation in cells treated with the drug combination. As expected,
24 h of BSO treatment induced a 3-fold increase in ROS accumulation, reflecting the fact that GSH
synthesis is inhibited (Supporting Fig. 6F). In contrast, 24 h of STF-31 treatment alone did not induce ROS
accumulation. However, combined STF-31 and BSO treatment induced a 4-fold increase in ROS
36
accumulation. Taken together, these data suggest that combined inhibition of GLUT1 and GCL mimics
glucose deprivation-induced cell death.
We next investigated xCT expression within GBM tumors using single-cell RNA-sequencing data
from 28 adult and pediatric patients
34
. Malignant GBM cells exhibited higher xCT expression when
compared to macrophages, T cells, and oligodendrocytes (Fig. 6F). Examining xCT expression in individual
patients revealed that malignant GBM cells exhibited significantly higher expression of xCT than non-
malignant cells in 10 out of 20 patients. Together, this illustrates the potential of xCT as a therapeutic target
for a subset of GBM patients (Supporting Fig. 7).
Having validated that combined inhibition of GLUT1 and GCL synergistically induced cell death
in xCT-high GBM and Ewing’s sarcoma cell lines, we next sought to identify other cancer types which
might exhibit sensitivity to this drug combination. Mining the DepMap database, we found that melanoma
and lung adenocarcinoma cancer cell lines exhibited a strong negative correlation between GLUT1
dependency and xCT expression (Supporting Fig. 8A), suggesting that these cells lines would respond
synergistically to GLUT1 and GCL co-inhibition. In contrast, high xCT expressing cell lines derived from
breast, kidney, ovarian, and colorectal cancers did not exhibit increased dependence on GLUT1, suggesting
that high xCT expressing cells from these tumor types would not synergistically respond to combined GCL
and GLUT1 inhibition. To test this hypothesis, we treated the breast cancer cell lines M453 (xCT low) and
M436 (xCT high) with STF-31 and BSO. We found that neither breast cancer cell line exhibited significant
cell death in response to combined treatment even at 72 h (Supporting Fig. 8B). Taken together, these data
demonstrate the ability of the DepMap data to predict response to co-inhibition of GLUT1 and GCL.
37
Figure 6. Inhibition of glutathione synthesis is synthetically lethal with GLUT1 inhibition.
38
A) Glioblastoma cells rely on SLC2A1 (GLUT1) as xCT expression increases. Using data from the Cancer
Dependency Map (DepMap (42)), glycolytic genes were filtered for expression in GBM and then ranked
by their Pearson correlation coefficient between dependency (CERES score) and xCT expression. A
negative CERES score indicates increased dependency. SLC2A1 (GLUT1) and PGM1 were the top hits for
GBM. B) Co-inhibition of GLUT1 and GCL synergistically induces cell death in glucose deprivation-
sensitive cells. T98 cells were cultured in 5 mM glucose and treated with STF-31 and/or BSO at the
indicated doses. After 36 h of treatment, the combination of drugs synergistically induced cell death. C)
Sensitivity to glucose deprivation correlates with response to combined STF-31 and BSO treatment. Highly
glucose deprivation-sensitive T98, medium-resistant LN229, and highly-resistant A172 GBM cells were
cultured in 5 mM glucose and treated with STF-31 (12.5 µM), BSO (500 µM), or both. T98 cells died after
36 h of treatment, whereas LN229 cells died after 60 h of treatment, and A172 cells died after 72 h of
combined STF-31 and BSO treatment. D) Genetic knockdown of xCT confers resistance to STF-31 and
BSO treatment in xCT-high LN18 cells. Cells were first infected with CRISPRi machinery (dCas9-KRAB
(69)) and then infected with a second vector carrying the indicated guide RNA. Cells were cultured in 5
mM glucose and treated with STF-31 (12.5 µM), BSO (500 µM), or both. After 48 h, combined STF-31
and BSO treatment induced cell death in non-targeting control cells but not in xCT-knockdown cells. E) L-
cystine import contributes to combined STF-31 and BSO-induced cell death in T98 cells. Cells were
cultured in 5 mM glucose supplemented with either L-cystine (200 µM) or L-cysteine (200 µM). L-cysteine
treatment conferred resistance to combined STF-31 (12.5 µM) and BSO (500 µM) treatment. ** denotes p
< 0.01 by Student’s t-test (n=3). F) xCT expression is increased in a subset of GBM. Single-cell RNA-
sequencing data from 28 adult and pediatric patients (44) was analyzed for xCT expression (logTPM). xCT
expression was overlaid onto a tSNE plot to compare malignant GBM cells to macrophages, T cells, and
oligodendrocytes. Error bars are standard deviation of the mean.
39
40
Supporting Figure 6. Co-Inhibition of GLUT1 and GCL is synthetically lethal in glucose deprivation
sensitive cell lines.
A,B) Using data from the Cancer Dependency Map (DEPMAP
33
), glycolytic genes were filtered for
expression in GBM or Ewing’s Sarcoma and then ranked by their Pearson correlation coefficient between
dependency (CERES score) and xCT expression. A negative CERES score indicates increased dependency.
SLC2A1 (GLUT1) was the top hit for GBM and the sixth highest hit for Ewing’s sarcoma. Permutation p-
values were assessed by permuting expression and dependency labels.
C) Ewing’s sarcoma cells with increased xCT expression are more sensitive to SLC2A1 (GLUT1) depletion.
Data is from the Cancer Dependency Map (DEPMAP
33
) as in panel B.
D) NAMPT dependency does not correlate with xCT expression in glioblastoma and Ewing’s sarcoma cell
lines. Data is from the Cancer Dependency Map (DEPMAP
33
) as in panels A and B.
E) Glucose deprivation sensitivity correlated with response to combined STF-31 and BSO treatment. High
sensitivity TC32 and low sensitivity HT161 sarcoma cells were cultured in 5 mM glucose and treated with
STF-31 (12.5 µM), BSO (500 µM), or both drugs for the indicated amount of time. Glucose deprivation-
sensitive TC32 cells died after 48 h of treatment, whereas resistant HT161 cells did not die until after 60 h
of treatment.
F) Co-inhibition of GLUT1 and GCL induces ROS in T98 cells. Histograms showing the distribution of
T98 cells treated with STF-31 (12.5 µM), BSO (500 µM), or both for 24 h. Combined treatment induced a
4-fold increase in the mean fluorescent intensity, whereas BSO alone induced a 3-fold increase in and STF-
31 alone showed no increase compared to DMSO control.
41
42
Supporting Figure 7. xCT expression is upregulated in a subset of glioblastoma patients. Single-cell
RNA-sequencing data was analyzed from adult and pediatric GBM patients
34
for xCT
expression (logTPM). Data was filtered for tumors in which non-malignant cells were
measured. 10 out of 20 patients exhibited significant increased expression of xCT in
malignant cells compared to non-malignant cells as assessed using a Wilcoxon rank
sum test.
Supporting Figure 8. Combined GLUT1 and GCL inhibition does not synergistically induce cell death in
breast cancer cell lines.
A) Cancer cell lines derived from certain tissues exhibit increased dependence on GLUT1 as
xCT/SLC7A11 expression increases. Negative CERES scores (e.g. increased dependency) correlate with
xCT expression in glioblastoma, melanoma, and lung adenocarcinoma cell lines, but not in breast cancer,
pancreatic cancer, or colorectal cancer cell lines.
B) The breast cancer cell lines M453 (xCT high) and MB436 (xCT low) do not respond to STF-31, BSO,
or combined STF-31 and BSO treatment after 72 h in 5 mM glucose. Expression of xCT/SLC7A11 was
assessed by Western blotting. Actin was used as an equal loading control.
43
1.8 Discussion
The metabolic reprogramming that occurs during oncogenesis can result in increased dependency
on certain substrates for survival (e.g., glucose and glutamine)
58,59
. As such, therapeutic strategies that target
altered glucose metabolism in cancer cells have attracted considerable attention
5–7,60–62
. Here, we
demonstrate a synthetic lethal therapeutic strategy targeting the interconnected nature of glucose and redox
metabolism. Our results show that glucose deprivation-induced cell death is driven not by lack of glucose,
but rather by NADPH depletion following L-cystine import and reduction to L-cysteine. Consistent with
this mechanism, glucose deprivation-induced cell death requires activity of the L-glutamate/L-cystine
antiporter xCT. By mining the cancer dependency database DepMap, we found that GBM cells with high
xCT expression are sensitive to depletion of the glucose transporter GLUT1. Building on these
observations, we demonstrate that co-targeting GLUT1 and GSH metabolism induced a synthetic lethal
interaction in glucose deprivation-sensitive glioblastoma and sarcoma cell lines. Taken together, our results
demonstrate a synthetic lethal approach to exploit the glucose addiction phenotype in cancer cells.
Our results add to the growing literature surrounding synthetic lethal targeting of tumor
metabolism. For example, passenger genomic deletions of metabolic enzymes, including enolase 1
63
, malic
enzyme 2
28
, or methylthioadenosine phosphorylase (MTAP)
64–66
, can create synthetic lethal vulnerabilities
unique to cancer cells. Similarly, in B cell malignancies, transcriptional repression of the pentose phosphate
pathway creates a cancer cell-specific lack of antioxidant protection
67
, whereas in renal cell carcinomas,
VHL mutation creates a sensitivity to inhibition of the glucose transporter GLUT1
68
. In addition, our work
builds upon previous studies demonstrating that co-targeting glucose metabolism and redox metabolism
can induce cell death in some cellular contexts. For example, the glycolytic inhibitor 2-deoxyglucose (2-
DG) demonstrates enhanced cytotoxicity when paired with BSO in human breast cancer cells
69
or with
antimycin A in human colon cancer cells
24
. Interestingly, however, 2-DG supplementation can rescue
cancer cells in the absence of glucose
14,18
. This may be due to the fact that 2-DG can be metabolized in the
44
pentose phosphate pathway
70
(PPP), thus allowing 2-DG to serve as an alternative source of NADPH when
glucose is restricted. Our results provide a rationale for co-targeting GLUT1 and GSH metabolism in high
xCT expressing GBM and Ewing’s sarcomas.
Multiple recent studies have demonstrated that the L-glutamate/L-cystine antiporter xCT is
required for cell death upon glucose withdrawal
38,53,54
, although the role of xCT is debated. Because xCT
exports L-glutamate and imports L-cystine, studies have suggested that xCT promotes glucose addiction
by depleting L-glutamate and thereby impairing the TCA cycle
53,54
. However, in our GBM cell lines, we
find no evidence of TCA cycle impairment following glucose starvation. Notably, in glucose withdrawal-
sensitive cells, the forward flux through the TCA cycle was increased and levels of L-glutamate were
unchanged following glucose deprivation (Figs. 1D and 2A). In contrast, our data suggest that xCT
promotes L-cystine import and subsequent depletion of NADPH as L-cystine is reduced to L-cysteine,
consistent with the model proposed by Goji et al.
38
. Notably, the depletion of NADPH by L-cystine import
is consistent with our observation of increased flux through the TCA cycle following glucose deprivation,
as cells attempt to generate NADPH through malic enzyme. This finding is consistent with a recent report
that knockdown of malic enzyme can sensitize cancer cells to glucose deprivation
71
. Finally, although it has
been shown that glucose deprivation induces oxidative stress in sensitive cells
16,72–74
, the source of reactive
oxygen species (ROS) has remained unclear. Our results suggest that L-cystine reduction is a primary
source of redox imbalance, thereby allowing enabling accumulation of ROS from sources including
mitochondria and NADPH oxidase
75
.
Notably, targeting GSH synthesis has long been viewed as a potential cancer therapy
76,77
, and the
GCL inhibitor BSO has been used in combination with the alkylating agent melphalan in clinical trials with
mixed results (NCT00005835 and NCT00002730)
78,79
. Our findings suggest that pairing BSO with
inhibitors of GLUT1 may prove synthetically lethal in cancers with high xCT expression including
glioblastomas, sarcomas, melanomas, and lung adenocarcinomas. Because high xCT levels are associated
45
with poor prognosis in a number of cancer types, including GBM
80
and triple-negative breast cancer
81
, the
xCT transporter has been suggested as a therapeutic target for cancer
81–83
. Additionally, engineered enzymes
that degrade extracellular L-cystine and L-cysteine and thereby indirectly inhibit xCT, blocking
intracellular GSH production and upregulating ROS, have shown promise in preclinical models
84
. Our
findings, in contrast, suggest that not all tumors may benefit from xCT inhibition, particularly tumor cells
in nutrient-limited microenvironments. In fact, our results suggest that pairing xCT inhibitors with GLUT1
inhibitors would be counterproductive, leading to greater tumor cell survival. The L-cystine dependent
mechanism described here, in addition to findings that GLUT1 is commonly upregulated in GBM
85
,
suggests that a synthetic lethal approach targeting GSH synthesis and GLUT1 activity may result in
selective toxicity towards cancer cells.
In summary, the highly interconnected nature of redox homeostasis and glycolysis in cancer cells
provides an opportunity for synthetic lethal drug combinations targeting GSH synthesis and glucose
metabolism. Future studies will be necessary to evaluate the efficacy and toxicity of combined GLUT1 and
GCL inhibition in vivo and also to understand why some tumor types (e.g., GBM and Ewing’s sarcoma)
but not others (e.g., breast cancer) are susceptible to this drug combination. Taken together, these results
demonstrate the power of integrating metabolomic profiling with public databases of cancer dependencies
(e.g., DepMap) to identify synthetic lethal vulnerabilities of cancer cells.
Chapter 2: Differential Gene Set Enrichment Analysis: An enrichment-based
method to infer tradeoffs from -omics data.
With an abundance of quantitative data characterizing the transcriptome, proteome, and
metabolome, there is a persistent need for approaches to mine these rich data sets for biological insights.
Gene Set Enrichment Analysis (GSEA)
26
has proved to be one of the most popular and powerful tools for
analyzing transcriptomic data sets with over 18,000 citations. While primarily used to analyze gene
expression data, the algorithm has been applied to other data types including proteomics
86
, phospho-
46
proteomics (Kinase Substrate Enrichment Analysis, KSEA)
87,88
, and metabolomics (Metabolite Set
Enrichment Analysis, MSEA)
89,90
. Regardless of the data type, the key concept underlying GSEA is that
pre-defined sets of functionally related genes can display significant enrichment that would be missed when
examining individual genes. In addition, the GSEA approach has the advantage of not requiring a pre-
selected threshold of significance. By using the entire dataset as background, researchers can identify
pathways upregulated and downregulated in their phenotype(s) of interest. The approach is widely
successful and has inspired many extensions, improvements, and variations that analyze individual gene
sets
91–94
.
However, to our knowledge, no tools exist for analyzing the difference between two gene sets.
Biological control throughout signaling and metabolism presents many circuits, feedback and feedforward
loops, and pathway branches that have important cellular consequences, and statistical methods that
accurately capture these biological tradeoffs are needed. Here, we present a variant of Gene Set Enrichment
Analysis that compares two pathways against each other, Differential Pathway Enrichment Analysis
(DGSEA). Using the metabolic pathways glycolysis and oxidative phosphorylation as a test case, we
demonstrate that DGSEA using RNAseq data accurately reflects cellular phenotypes. Applying DGSEA to
RNAseq data from a panel of cell lines subjected to hypoxia, we find that DGSEA accurately identifies the
metabolic tradeoff between glycolysis and oxidative phosphorylation when there is a difference to be
observed. Using metabolomics data from the NCI60 and CCLE, we show that DGSEA more accurately
predicts lactate secretion and abundance than GSEA examining either pathway alone. Furthermore, we find
that DGSEA examination of glycolysis and oxidative phosphorylation using the RNAseq data from the
Broad Institute’s Cancer Cell Line Encyclopedia reveals novel insights into cancer cell dependence on
metabolic pathways and mitochondrial complexes. DGSEA will serve as a tool to examine the difference
between two gene sets and
47
2.1 DGSEA measures the relative difference between two gene sets
To measure the relative difference between two gene sets, we adapted GSEA to create Differential
Gene Set Enrichment Analysis (DGSEA). Given two gene sets of interest (e.g., Gene Sets A and B), the
goal of DGSEA is to determine whether the members of Gene Sets A and B are randomly distributed with
reference to each other. If Gene Sets A and B are upregulated and downregulated, respectively, we expect
that A and B will be at opposite sides of the rank list. Although we use the terminology “gene set”, the
DGSEA algorithm is agnostic to data type and can be used with transcriptomic, proteomic, or metabolomic
data alongside sets of functionally related genes, proteins, or metabolites. DGSEA first ranks the data by
any suitable metric including fold change, signed p-value, signal-to-noise ratio, or correlation with
phenotype (Fig. 7). Second, we calculate an enrichment score (ES) for each gene set (ES A and ES B) by
walking down the rank list and finding the maximum deviation from zero of a running-sum, weighted
Kolmogorov-Smirnov-like statistic. This is equivalent to the GSEA algorithm. Then, the difference between
ES A and ES B is calculated to measure the enrichment of the two gene sets relative to each other (ES AB =
ES A - ES B). We then estimate the significance level of ES A, ES B, and ES AB using an empirical permutation
test as in the original GSEA algorithm. Specifically, the rank list is permuted, permutation enrichment
scores are calculated for each gene set (pES A, pES B, and pES AB), and then the nominal p-value is calculated
relative to the same-signed portion of the null distribution. Next, the normalized enrichment score (NES) is
calculated by dividing positive and negative ES by the mean of positive or negative pES, respectively.
Finally, to estimate the false discovery rate (FDR), a null distribution of NES values is generated using a
list of background gene sets. Background gene sets can be chosen based on the biological meaning of the
tested gene sets, or they can be randomly generated. Using the background gene sets, the null distribution
is the union of NES values comparing Gene Set A versus all background pathways (NES AX) with the NES
values comparing all background gene sets versus Gene Set B (NES XB). The combination of these
distributions is termed NES XY. For a given NES AB, the FDR is then calculated as the ratio of the percentage
of the same-signed NES XY greater than or equal to NES AB divided by the percentage of NES XY with the
same sign as NES AB. The FDR estimates for NES A and NES B are generated using a similar approach based
48
on single background gene sets, equivalent to GSEA. The output of DGSEA is thus the relative enrichment
and statistical significance of Gene Set A versus B, as well as the individual enrichment and statistical
significance of Gene Sets A and B.
49
50
Figure 7. Differential Gene Set Enrichment Analysis (DGSEA) quantifies the enrichment between
two gene sets relative to each other.
First, a data set is first ranked by any suitable metric inducing fold change, signed p-value, signal-to-noise
ratio, or correlation with phenotype. Second, the enrichment score (ES) is calculated for each individual
gene set by walking down the rank list and finding the maximum deviation from zero of a running-sum,
weighted Kolmogorov-Smirnov-like statistic (ES A and ES B, equivalent to GSEA, left & middle). Then, the
difference between the enrichment scores of two gene sets is calculated (ES AB= ES A - ES B, right). Third,
the statistical significance of ES A, ES B, and ES AB is estimated by using an empirical permutation test that
preserves the structure of the original data. Specifically, a null distribution is generated by shuffling the
rank list and calculating the permutation ES values (e.g., pES A), and the nominal p-value is calculated
relative to the same-signed distribution. Fourth, the normalized enrichment score (NES) is then calculated
by dividing the observed ES by the mean of the same-signed portion of the permutation ES distribution
( pES
AB
̅ ̅ ̅ ̅ ̅ ̅ ̅ ̅
). Fifth, to control for false discovery rate (FDR), a null distribution of NES values is generated
using a list of background gene sets. For DSEA, a null distribution of NES values (NES XY) is the union of
the NES values comparing Gene Set A versus all background gene sets (NES AX) and the NES values
comparing all background gene sets against Gene Set B (NES XB) and is termed NES XY. For a given NES AB
greater than zero, the FDR is then calculated as the ratio of the percentage of all NES XY greater than or
equal to NES AB divided by the percentage of observed NES XY is positive and similarly if NES AB is negative.
2.2 Simulation Study
To evaluate the sensitivity of DGSEA to quantify differences between gene sets, we generated
simulated data from a standard normal distribution (15,678 genes total). We then added two “per-turbed”
Gene Sets A and B with normally distributed positive and negative random deviations, respectively (µ=+X
or -X for Gene Sets A and B, respectively, σ=1, n = 10, 25, 50, or 100 genes). For values of X ranging from
0 to 0.5, we then performed DGSEA (Gene Set A minus Gene Set B) and GSEA (Gene Set A or Gene Set
B) for 50 independent replicates. For all gene set sizes, DGSEA demonstrated significant mean enrichment
(FDR q-value < 0.05) at smaller values of X than did GSEA (Fig. 8). Like GSEA, the mean enrichment
necessary to detect a statistically significant enrichment using DGSEA decreased as gene set size increases.
Additionally, we tested replacing synthetic gene sets using the sizes of the real glycolysis and OxPhos gene
sets (19 and 85 genes, respectively) with simulated values, and we found that DGSEA captured differ-ences
in means of ≥ 0.45 (q < 0.01) (Supp. Fig. 1A). Finally, we tested DGSEA using negative control data with
both the simulated glycolysis and OxPhos gene sets perturbed in the same direction. Because both pathways
are randomly perturbed in the same direction, DGSEA should not measure enrichment of the two pathways
relative to each other. Indeed, DGSEA q-values were not significant for any value of the perturbation X
51
(Supp. Fig. 1B). Taken together, this data established the sensitivity and specificity of the DGSEA method.
Figure 8. DGSEA is more sensitive than GSEA for co-regulated pathways.
Gene expression data was simulated from a standard normal distribution (µ=0, σ=1, n=15678 genes). Then,
gene expression values for Gene Set A and Gene Set B were substituted from normal distributions with
means +X and -X, respectively (σ=1, n = 10, 25, 50, or 100 genes). For values of X ranging from 0 to 0.5,
we then performed DGSEA (Gene Set A minus Gene Set B, red boxplots) and GSEA (Gene Set A (green
boxplots) or Gene Set B (blue boxplots)) for 50 independent replicates and plotted the negative log10 FDR
q-value. For all gene set sizes, DGSEA demonstrated significant mean enrichment (FDR q-value < 0.05 or
negative log10 FDR q-value > 1.3) at smaller values of X than did GSEA. Like GSEA, the mean enrichment
necessary to detect a statistically significant enrichment using DGSEA decreased as gene set size increases.
52
Supporting Figure 9. Benchmarking sensitivity of DGSEA to detect differences between Glycolysis and
OxPhos.
A) Testing positive control synthetic data. To evaluate the sensitivity of DGSEA, we generated simulated
gene expression data from a standard normal distribution (15678 genes). Then, we replaced genes from the
glycolysis and OxPhos gene sets with values drawn from positively shifted and negatively shifted normal
distributions, respectively (μ = ±X, σ = 1, 15678 genes for glycolysis and OxPhos, respectively). For each
value of X ranging from 0 to 0.5, we performed DGSEA (Glycolysis minus OxPhos, red boxplots) and
GSEA (Glycolysis (green boxplots) or OxPhos (blue boxplots)) 50 times and then plotted the negative log 10
of the false-discovery rates q-values. DGSEA can reliably detect a change in means of ≥ 0.45. The dotted
line indicates q-value = 0.01. B) Testing negative control synthetic data. As a negative control, we replaced
genes from the glycolysis and OxPhos gene sets with values drawn from a positively shifted normal
distribution (μ = +X, σ = 1, 19 and 85 genes for glycolysis and OxPhos, respectively). Because both
53
pathways are randomly perturbed in the same direction, DGSEA should not measure enrichment of the two
pathways relative to each other. For each value of X ranging from 0 to 0.5, we performed DGSEA
(Glycolysis minus OxPhos, red boxplots) and GSEA (Glycolysis (green boxplots) or OxPhos (blue
boxplots)) 50 times and then plotted the negative log 10 of the false-discovery rate q-values. Indeed, DGSEA
did not detect significant enrichment of the two pathways relative to each other for any value of X. The
dotted line indicates q-value = 0.05.
2.3 DGSEA accurately captures hypoxia-induced coordinate increases in glycolysis and
decreases in oxidative phosphorylation
Hypoxia is associated with a metabolic shift away from oxidative phosphorylation and towards
glycolysis. In the absence of oxygen, hypoxia-inducible factor 1- (HIF1 ) transcriptionally activates
glucose catabolism through expression of glucose transporters, glycolytic enzymes, and lactate
dehydrogenase A
95
. Moreover, the HIF1 -target gene pyruvate dehydrogenase kinase 1 (PDK1) suppresses
metabolic flux from pyruvate to acetyl-CoA, diverting carbon away from the mitochondria and thereby
reducing oxidative phosphorylation
96
. We therefore first tested the ability of DGSEA to detect the hypoxic
shift from oxidative phosphorylation to glycolysis. We applied DGSEA to RNASeq data from 31 breast
cancer cell lines subjected to either 1% or 20% oxygen
37
either as paired cell lines (i.e., hypoxia / normoxia)
or as individual samples (i.e., single sample DGSEA) (Fig. 8A). Analysis with a consensus hypoxia gene
set confirmed that all 31 cell lines demonstrated enrichment of hypoxia-regulated genes upon exposure to
1% oxygen. To test the differential enrichment of glycolysis and oxidative phosphorylation, we chose gene
sets A) core glycolysis (hsaM00001, which includes conversion of glucose into pyruvate) and B) oxidative
phosphorylation (hsa00190), respectively, from KEGG
97
. For the paired cell line analysis, DGSEA was
significantly upregulated in 21 of 31 cell lines (p-value < 0.05, FDR < 0.25, Fig. 2B). For the individual
pathways, GSEA demonstrated significant upregulation of core glycolysis in all 31 cell lines but significant
downregulation of oxidative phosphorylation in only 21 of 31 cell lines. Surprisingly, upon exposure to 1%
oxygen, 10 cell lines exhibited significant upregulation of oxidative phosphorylation. One cell line,
MCF12A, had a similar induction of both core glycolysis and oxidative phosphorylation in 1% oxygen (Fig.
8C). Notably, the cell lines with upregulated oxidative phosphorylation in 1% oxygen were the same cell
lines identified by DGSEA as not having a significantly differential response between core glycolysis and
54
oxidative phosphorylation. Notably, substitution of the full glycolysis pathway (Glycolysis-
Gluconeogenesis, hsa00010) as Gene Set A or TCA cycle (hsa00020) as Gene Set B yielded very similar
results to usage of Core Glycolysis (hsaM00001) and Oxidative Phosphorylation (hsa00190) as Gene Sets
A and B, respectively (not shown).
We next tested whether similar trends were observable in single sample DGSEA (ssDGSEA) and
GSEA (ssGSEA). Again, analysis with a consensus hypoxia gene signature confirmed that all cell lines
responded to 1% oxygen. Using DGSEA to compare the relative enrichment of core glycolysis and
oxidative phosphorylation, we found that nearly every cell line increased its NES score in 1% oxygen (Fig.
8D). In addition, the number of cell lines with significant ssDGSEA scores at 1% oxygen was increased
relative to 20% oxygen (7 versus 1). Consistent with the paired cell line analysis, most cell lines increased
the ssGSEA NES of core glycolysis in 1% oxygen. For oxidative phosphorylation, however, most cell lines
had only slightly negative or negligible changes in the ssGSEA NES between 20% and 1% oxygen. Taken
together, these results demonstrate although not all cell lines subjected to hypoxia exhibit gene expression
signatures consistent with shifting from oxidative phosphorylation to glycolysis, DGSEA does accurately
identify cell lines which exhibit the hypoxia-induced coordinate upregulation and downregulation of
glycolysis and oxidative phosphorylation, respectively.
55
Figure 9. DGSEA accurately captures hypoxia-induced coordinate increases in glycolysis and
decreases in oxidative phosphorylation.
A) Schematic of data normalization methods used to generate gene ranking metrics for breast cancer cell
lines subjected to hypoxia (1% O2) or normoxia (20% O2). B) Core Glycolysis and OxPhos are
significantly changed relative to other pathways when subject to hypoxia. Untargeted DGSEA was
run using all metabolic pathways and gene set comparisons were ranked by their normalized
56
enrichment score. DGSEAs containing Core Glycolysis (red) and OxPhos (blue) are highlighted.
C) Representative mountain plots and table of values for Core Glycolysis – OxPhos. D) DGSEA
on paired cell line data identified cell lines with coordinately increased glycolysis and decreased
OxPhos. DGSEA (Core Glycolysis-OxPhos) and GSEA (Core Glycolysis & OxPhos) NES values
were calculated for each cell line. Black outline denotes p-value < 0.05 and FDR < 0.25. E)
Representative mountain plots and table of values for Core Glycolysis – OxPhos for MCF10A and
MCF12A cells.
Supporting Figure 10. Validation of hypoxia gene signature. A) RNAseq data from 31 breast cancer cell
lines subjected to either 20% or 1% oxygen was analyzed using GSEA using a hypoxia gene set
(HALLMARKS_HYPOXIA) from the Molecular Signature Database (MSigDB) of the Broad Institute.
Paired analysis of hypoxia over normoxia revealed that all breast cancer cell lines upregulated the hypoxia
57
gene signature when subjected to 1% oxygen (p-value < 0.01 and GDR q-value < 0.25). B) Single sample
GSEA (ssGSEA) using the hypoxia gene signature was performed on 62 breast cancer cell lines subjected
to either 20% or 1% oxygen. Upon 1% oxygen, most but not all cancer cell lines increased the Normalized
Enrichment Score (NES) of the hypoxia gene signature. Filled circles denote nominal p-value < 0.05 and
FDR < 0.25. Open circles denote not significant (N.S.).
Supporting Figure 11. Single sample DGSEA captures hypoxia-induced upregulation in glycolysis and
downregulation of OxPhos. Line plots for ssDGSEA (left) and ssGSEA (middle and right) for each cell line
show the change in NES upon switch from 20% to 1% oxygen. DGSEA and glycolysis were upregulated
in most cell lines, whereas OxPhos showed neutral changes in most cell lines. Filled circles denote nominal
p-value < 0.05 and FDR < 0.25. Not significant, N.S.
2.4 Benchmarking DGSEA against QuSAGE
Having found that DGSEA accurately captures the coordinate upregulation of glycolysis and
downregulation of OxPhos induced by hypoxia, we next sought to compare DGSEA to QuSAGE
98
, another
algorithm which can measure the enrichment of two gene sets relative to one another. We thus compared
our results from untargeted DGSEA across all KEGG metabolic pathways using RNAseq data from breast
cancer cell lines exposed to either hypoxia or normoxia. We first tested whether there was agreement
between the two methods for calculation of p-values (Supp. Figure 4A-B). We found that DGSEA had 121
p-values < 0.01 and QuSAGE had 283 p-values < 0.01, with 73 p-values overlapping between the two
methods (Supp. Figure 4C). A threshold of 0.01 was chosen because DGSEA p-values that are between
0.01 and 0.05 often have a false discovery rate of > 0.05. Notably, QuSAGE recommends using a
Benjamini-Hochburg correction to account for multiple hypothesis testing. However, the Benjamini-
Hochburg correction is not appropriate here because the 3081 combinations of pathways are not
58
independent of one another. Since we cannot properly assess the false discovery rate of QuSAGE for
untargeted analyses, we cannot determine how many of the 283 p-values < 0.01 are true positives. The
ability of DGSEA to control for false discovery rate renders it suitable for untargeted analyses.
Supporting Figure 12. Comparison of DGSEA and QuSAGE. To compare DGSEA and QuSAGE, we
took gene expression data from 31 breast cancer cell lines in 1% and 20% oxygen and calculated the average
fold change in hypoxia. We then ran DGSEA and QuSAGE across all metabolic pathways in the Kyoto
Encyclopedia of Genes and Genomes (N = 79 pathways, M = 3,081 combinations). For DGSEA, 10,000
permutations were used to assess significance. For QuSAGE, P-values were calculated using the
TwoCurvePval() function from the qusage R package. A) Scatter heat plot demonstrating the global
distribution of p-values for both DGSEA and QuSAGE. The dashed lines represent p = 0.05. B) Scatter plot
of all p-values that are below 0.05 for both QuSAGE and DGSEA. C) Venn diagram demonstrating the
overlap of p-values < 0.01 for DGSEA and QuSAGE. The significance level was set at p = 0.01 because
DGSEA p-values < 0.01 generally have an FDR < 0.05.
59
2.5 DGSEA is more predictive than GSEA of lactate secretion and glucose consumption in cancer
cell lines.
Lactate secretion is often used as a marker of the metabolic shift between glycolysis and oxidative
phosphorylation. Because DGSEA measures the relative difference between glycolysis and oxidative
phosphorylation, we therefore hypothesized that DGSEA would be more predictive of lactate secretion
rates than GSEA using either the glycolysis or oxidative phosphorylation gene sets alone. To test this
hypothesis, we analyzed published gene expression and metabolite consumption and secretion rates from
the NCI-60 panel of cancer cell lines
35
(Fig. 9A). Indeed, we found that ssDGSEA NESs were more
significantly correlated with lactate secretion rates than either core glycolysis or oxidative phosphorylation
ssGSEA NESs (Fig. 9B). Similar to our results with hypoxia, we found that ssDGSEA was a better predictor
of lactate secretion than ssGSEA for all combinations of similar gene sets (e.g., when Gene Set A is either
Core Glycolysis or Glycolysis-Gluconeogenesis and Gene Set B is either Oxidative Phosphorylation or
TCA Cycle) (Data not shown). Interestingly, ssDGSEA was also more significantly predictive of glucose
uptake across the 61 cell lines than ssGSEA using either glycolysis or oxidative phosphorylation. This may
reflect the highly glycolytic nature of these cancer cell lines in tissue culture. Together, these results reveal
that DGSEA was more predictive of lactate secretion and glucose consumption than GSEA across a panel
of cancer cell lines.
60
Figure 10. DGSEA is a better predictor of cellular metabolism than GSEA.
A) Schematic of process used to correlate pathway activity, as measured by GSEA or DGSEA, with
metabolic phenotypes. B-C) DGSEA more accurately predicts lactate secretion and glucose uptake
rates than GSEA. Gene expression data was centered and scaled across 59 of the NCI-60 cancer
cell lines and glycolysis, OxPhos, and DGSEA Normalized Enrichment Scores (NES) were
calculated for each cell line. Spearman rank correlation coefficients were calculated between each
NES and lactate secretion or glucose uptake data from Jain et al
35
. Scatter plots showing the
61
spearman correlation are shown (right).* indicates p < 0.05, ** indicates p < 0.01, *** indicates p
< 0.001.
Supporting Figure 13. Testing Gene sets for DGSEA comparing Glycolysis and Oxidative
Phosphorylation. DGSEA is a better predictor of lactate secretion and glucose uptake than GSEA regardless
of which glycolysis and oxidative phosphorylation gene sets are used. Gene expression data was centered
62
and scaled across 59 of the NCI-60 cancer cell lines and glycolysis, OxPhos, and DGSEA Normalized
Enrichment Scores (NES) were calculated for each cell line using the indicated gene sets for glycolysis (All
Glycolysis, Core Glycolysis) and oxidative phosphorylation (OxPhos, TCA Cycle). Spearman rank
correlation coefficients were calculated between each NES and lactate secretion or glucose uptake data
from Jain et al (Jain et al., 2012).
2.6 DGSEA is correlated with high concentrations of intracellular lactate and low concentrations
of intracellular AMP in adherent cancer cell lines.
Although intracellular metabolite concentrations do not reflect pathway flux, we next hypothesized
that comparing DGSEA and steady-state metabolite abundance would reveal trends consistent with
coordinate upregulation of glycolysis and downregulation of OxPhos. For this purpose, we used paired
RNAseq and metabolomics data from 897 cancer cell lines from the Cancer Cell Line Encyclopedia (Li et
al., 2019). Since culture type has been reported to be a major determinant of metabolism, we separately
analyzed cancer cell lines cultured in adherent and suspension cultures. Correlating DGSEA NESs for 836
adherent cell lines against 225 intracellular metabolite concentrations, we found that the metabolite most
correlated with DGSEA was 1-methylnicotinamide (MNA), which has no known role in regulation of
glycolysis or OxPhos (Fig. 5A and Supp. Table 5). However, the second most correlated metabolite with
DGSEA NES was lactate, suggesting that DGSEA accurately captured the tradeoff between glycolysis and
OxPhos. As with the lactate secretion data, we found that DGSEA NESs correlated better with intracellular
lactate levels than did GSEA NESs using either glycolysis or OxPhos alone (Fig. 5B and Supp. Fig. 6A).
Interestingly, we found that the metabolite most anticorrelated with DGSEA NESs was AMP, a classical
readout of cellular energetic state (Herzig and Shaw, 2018). DGSEA again was a better predictor of AMP
levels than GSEA with either glycolysis of OxPhos alone. Notably, these results with adherent cultures
were not recapitulated in suspension cultures, perhaps due to sample size limitations (Supp. Fig. 6B). Taken
together, these results indicate that DGSEA testing the relative enrichment between glycolysis and OxPhos
strongly correlated with steady state levels of metabolites that indicate the tradeoff between glycolysis and
OxPhos (i.e., lactate) and anticorrelated with metabolites indicative of a low energetic state (i.e., AMP).
63
Figure 11. DGSEA is a better predictor of intracellular lactate and AMP levels than GSEA for
adherent cell cultures.
A) Increased intracellular lactate and decreased AMP correlated with increased glycolysis and decreased
OxPhos. RNASeq data was centered and scaled across all adherent cell culture lines in the Cancer
Cell Line Encyclopedia and then the spearman correlation coefficient was calculated between
64
DGSEA NESs and metabolite abundances. Lactate was the second most correlated metabolite and
AMP was the least correlated metabolic with DGSEA. B) Barplots showing the comparison of
DGSEA and glycolysis and OxPhos GSEA. Scatter plots showing the correlation between DGSEA
and lactate or AMP are shown. *** indicates p < 0.001.
65
Supporting Figure 14. DGSEA analysis with steady state metabolite levels in adherent and suspension
cells. RNASeq data was centered and scaled for 836 adherent or 173 suspension cell culture lines in the
Cancer Cell Line Encyclopedia. Then the spearman correlation coefficient was calculated for between
DGSEA NESs and metabolite abundances. A) Lactate was the third most correlated metabolite with
glycolysis NES for adherent cancer cell lines. Waterfall plots showing the spearman correlation between
all metabolites and glycolysis or OxPhos NES are shown. B) Increased lactate and decreased AMP levels
did not correlate with DGSEA in cell lines cultured in suspension. Waterfall plots showing the spearman
correlation between all metabolites and glycolysis, OxPhos or DGSEA NES are shown. C) Individual bar
plots showing the correlation between glycolysis, OxPhos, and DGSEA NES and intracellular lactate and
AMP for cancer cell lines cultured in suspension are shown.
2.7 Untargeted DGSEA predicts differential metabolic pathway activity in senescent and
proliferating cells
To demonstrate the usage of DGSEA without an a priori hypothesis, we next analyzed RNAseq
data from IMR90 cells undergoing ionizing radiation-induced senescence
99
. Using 79 metabolic pathways
in from KEGG, we found that 19 of 3,081 pairwise combinations exhibited significant differential
enrichment between senescent and proliferating cells (p-value < 0.05, FDR q-value < 0.05) (Fig. 11). In
contrast, using GSEA, only two gene sets exhibited significant enrichment, namely glycolysis and nitrogen
metabolism. Of the 19 significant DGSEA pathway combinations, there were 14 unique gene sets with
nitrogen metabolism (9 of 19) and core glycolysis (4 of 19) overrepresented. Notably, six of the significant
DGSEA pathway combinations compared two pathways that were not individually significant using GSEA
(e.g., glycosaminoglycan biosynthesis keratan sulfate and heparan sulfate, Fig. 11A). These results suggest
that the coordinate up- and down-regulation of these metabolic pathways, rather than the up- or down-
regulation of the individual pathways, may be required for cellular senescence. Together, these results
demonstrate the ability of DGSEA to generate de novo hypotheses from transcriptomic data.
66
Figure 12. Untargeted DGSEA predicts differential metabolic activity in senescent and proliferating
cells. 79 metabolic pathway gene sets from the Kyoto Encyclopedia of Genes and Genomes
(KEGG) were queried with untargeted DGSEA to identify metabolic differences upon ionizing
radiation-induced senescence in IMR90 cells.
A-B) Representative mountain plots of DGSEA comparing KEGG pathways glycosaminoglycan
biosynthesis - keratan sulfate (hsa00533) to glycosaminoglycan biosynthesis - heparan sulfate
(hsa00534) and pyrimidine metabolism (hsa00240) minus nitrogen metabolism (hsa00910).
2.8 Limitations of DGSEA
Since DGSEA is an adaptation of the original GSEA algorithm, many of the limitations of GSEA
apply to DGSEA. Most notably, the method does not account for gene-gene correlations and can produce
high Type I error
100
. However, there has been some debate on whether gene-gene correlations can be
ignored due to the significant variance inflation they produce on enrichment scores
101
. One advantage of
GSEA is the intuitive “Enrichment Plot” that allow the user to manually examine enrichment patterns. If
the pattern of the enrichment does not appear biologically meaningful, the user can dismiss the result even
if the results are statistically significant. Similar to GSEA, we have included in our R package a function to
generate enrichment plots that the user can use to decide whether or not to discard DGSEA results in which
the enrichment plot does not appear biologically meaningful, even if the p-value is statistically significant.
As an example, we queried the senescence and proliferating RNAseq data with transcription factor target
gene sets from the Broad Institute’s Molecular Signatures Database (Supp. Fig. 7A). We found that the
67
transcription factors HSF1 minus HSF2 were statistically significant (FDR q-value = 1.96e-4), but upon
inspection of the enrichment plot we noticed the pattern for HSF1 was quite random (Supp. Fig. 7B).
Supporting Figure 15. Example of DGSEA Type I Error. In the above enrichment plot, the difference
between HSF1 and HSF2 transcription factor target gene sets was deemed significant by DGSEA because
the enrichment for HSF1 is negative (ES = -0.223) and HSF2 (ES = 0.282) is positive. However, the
distribution of HSF1 around 0 appears to be random. Thus, the user would be expected to deem this result
as insignificant even though p < 0.05 and q < 0.01.
2.9 Discussion
Traditional gene set enrichment analyses are limited to examining one set of genes at a time. Our
DGSEA method builds upon the original GSEA algorithm to measure the enrichment of two gene sets
relative to each other. Our work thus builds on statistical frameworks to identify differentially expressed
gene set pairs
98,102
. DGSEA can be run using traditional ranking metrics (e.g., fold-change between
perturbation and control) or using single-sample methods across many samples (i.e., ssDGSEA). In this
way, DGSEA provides similar usability to GSEA while serving as an extension to pathway analysis. Our
DGSEA software is freely available at https://jamesjoly.github.io/DGSEA/ and can be installed directly as
an R package.
68
To test the accuracy of DGSEA, we first used hypoxia as an example of a metabolic shift between
glycolysis and OxPhos. We found that DGSEA accurately captured the metabolic tradeoff between
upregulated glycolysis and downregulated OxPhos (Fig. 3). Notably, individual cell line analysis by
DGSEA identified a metabolic switch in only 21 of 31 cell lines, a finding confirmed by the observation
that the 10 other cell lines increased OxPhos in response to hypoxia. These surprising findings may be
explained by the fact that some cell lines require concentrations of oxygen lower than 1% to suppress
OxPhos
103
. Regardless, in cell lines with the classic hypoxia-induced metabolic shift, DGSEA correctly
identified coordinate increases in glycolysis and decreases in OxPhos.
Having established the accuracy of DGSEA, we proceeded to analyze how DGSEA using the
glycolysis and OxPhos pathways correlated with traditional metrics of cellular metabolism, namely lactate
secretion and glucose consumption (Fig. 4). Our finding that DGSEA more accurately predicted lactate
secretion rates than either GSEA with glycolysis or OxPhos alone confirmed that DGSEA accurately
captured the tradeoff between upregulated glycolysis and downregulated OxPhos. Furthermore, we found
that DGSEA NESs of adherent cancer cell lines were more correlated with intracellular lactate than either
GSEA with glycolysis or OxPhos (Fig. 5). Although steady-state levels of lactate do not necessarily reflect
the relative activity of glycolysis and OxPhos, they do suggest that DGSEA reflects the balance between
conversion of pyruvate to lactate and acetyl-CoA for the TCA cycle. In addition, we found that DGSEA
NESs were more significantly anticorrelated with intracellular levels of AMP than GSEA with glycolysis
or OxPhos alone. Notably, AMP regulates both glycolysis and OxPhos through AMP-activated kinase
(AMPK)-mediated activation of glycolytic enzymes and mitochondrial biogenesis
104
. Since the analyzed
metabolomic data did not include ATP levels, we cannot calculate the AMP:ATP ratio to infer the activity
of AMPK in these cell lines. However, taken together these results demonstrate that DGSEA analysis is
highly informative for metabolic pathway activity and intracellular energetic state.
69
While useful as a targeted tool, we wanted to explore how DGSEA could be used when the user
does not have a pre-defined hypothesis. We thus sought to use untargeted DGSEA to search for differential
enrichment of metabolic pathway gene sets from KEGG in senescent and proliferating cells. We found 19
out of 3,081 pairs of metabolic pathway gene sets that exhibited differential activity in senescent versus
proliferating cells. Of these 19 pairs of gene sets, 6 were uniquely significant in DGSEA but not individual
GSEAs. One of these results was differential enrichment between the biosynthesis of the glycans keratan
sulfate and heparan sulfate (Fig. 6). This result is particularly interesting since these molecules play critical
roles in the extracellular matrix
105,106
and changes to cell morphology are a hallmark of senescence
107
.
These results demonstrate that DGSEA can be used to detect differential enrichment in gene set activity
when there is not a pre-defined hypothesis.
In summary, DGSEA is a novel framework for analyzing the tradeoffs between two gene sets or
pathways. As such, we believe that the DGSEA will serve as a tool for analysis of a wide array of biological
contexts. Furthermore, since GSEA has been demonstrated to work on other -omic layers, we anticipate
that DGSEA will accurately capture trade-offs in phospho-proteomic and metabolomic data. As such,
DGSEA will serve as a useful tool to accurately quantify how tradeoffs between gene sets or pathways
regulate biological control.
Chapter 3: The landscape of metabolic pathway dependencies in cancer cell
lines
3.1 Introduction
The reprogramming of cellular metabolism was one of the earliest discovered hallmarks of cancer
19
.
Cancer cells rewire their metabolism to satisfy the bioenergetic, biosynthetic, and redox demands of tumors,
and these metabolic adaptations can create cancer-specific vulnerabilities that can be targeted
70
therapeutically
108
. Much research has focused on how individual mutations or DNA copy number
alterations reprogram tumor metabolism and create therapeutic opportunities
109–111
. However, given that
metabolic pathways consist of multiple enzymes which collectively regulate metabolic flux, studying the
effects of individual genes may not reflect cancer cell metabolic vulnerabilities at the pathway level. As a
result, our understanding of cancer cell dependency on metabolic pathways remains incomplete.
Recent developments in large scale CRISPR-based genetic
33,112
and pharmacologic screens
113
have
proved powerful tools for the identification of genes essential for cancer cell survival
114
, molecular markers
of drug sensitivity
112,115
, and novel candidate drug targets
116,117
. Furthermore, parallel integration of both
pharmacologic and gene loss-of-function data has been used to identify drug mechanism(s) of action
118–121
.
While these databases have served as a rich resource to explore individual gene vulnerabilities and drug
sensitivities, there exists a need to probe these datasets on the pathway level.
Here, we aimed to identify cancer cell dependencies on metabolic pathways rather than individual
metabolic genes. To do so, we used gene expression data from the Cancer Cell Line Encyclopedia (CCLE)
to infer metabolic pathway activity and then integrated these pathway activities with data from genetic and
pharmacologic screens across hundreds of cell lines. We show that this approach provides a comprehensive
characterization of cancer cell dependence on metabolic pathways. Furthermore, this approach confirmed
known metabolic vulnerabilities, gave insight into confounding factors such as media composition,
identified robust associations between drug response and metabolic pathway activity, and independently
found metabolic pathway essentialities in both genetic and pharmacological screens. Collectively, we
present an approach to integrate gene expression, gene dependency, and drug response data on the pathway
level to identify cancer cell dependencies on metabolic pathways.
71
3.2 Genetic Pathway Dependency Enrichment Analysis Identifies Metabolic Pathway
Dependencies by Integrating Metabolic Pathway Activity with CRISPR Screens
To identify metabolic pathway dependencies, we analyzed gene expression data and CRISPR‐Cas9
loss‐of‐function screens from 689 cancer cell lines overlapping between the Cancer Cell Line Encyclopedia
(CCLE)
116
and the Cancer Dependency Map
33
. First, we used gene expression data to infer metabolic
pathway activity. Because metabolism is influenced by culture type
122
and culture medium
114
, we first
divided cancer cell lines by culture type (e.g. adherent v. suspension culture) and media (e.g. RPMI v.
DMEM) (Fig. 1A). Cell lines without annotations for either of these features were removed, leaving 300
adherent cell lines cultured in RPMI, 153 adherent cell lines cultured in DMEM, 66 suspension cell lines
cultured in RPMI, and 2 suspension cell lines cultured in DMEM. Since the number of suspension cell lines
was small, we focused our analysis on adherent cell lines. To infer metabolic pathway activity within each
medium type, we ran single-sample gene set enrichment analysis (ssGSEA)
92
using RNAseq data from each
cell line querying 69 metabolic pathways from the Kyoto Encyclopedia of Genes and Genomes (KEGG)
123
.
The resulting normalized enrichment scores (NESs) represent the metabolic pathway activity relative to all
other cell lines within the respective cell culture medium. Next, we correlated the cell line-specific NESs
for each metabolic pathway with cell fitness effects from CRISPR‐Cas9 loss‐of‐function screens (16,643
gene knockouts). Here, each correlation coefficient represents the association between metabolic pathway
activity and gene essentiality, with positive values representing increased gene dependency in cell lines
with increased metabolic pathway activity. Conversely, a negative correlation indicates increased gene
dependency in cell lines with decreased metabolic pathway activity. Finally, to measure the essentiality of
the entire metabolic pathway, as opposed to individual genes, we then ranked the resulting 16,643
correlation coefficients and analyzed the rank list using GSEA again querying the KEGG metabolic
pathways. Here, positive NES values represent increased pathway dependency upon increased pathway
activity, whereas negative NES values represent increased pathway dependency upon decreased pathway
activity. Because this approach integrates the essentiality of all genes across a metabolic pathway into a
72
single metric of pathway dependency, we termed this approach genetic pathway dependency enrichment
analysis (Genetic PDEA).
3.3 Simulation Study
To analyze the sensitivity of our Genetic PDEA approach, we analyzed simulated gene expression
and gene dependency data using the pipeline outlined in Fig. 13A. Gene expression data (16,643 genes)
was simulated for 300 cell lines using a normal distribution for each cell line (µ = 0, σ = 1). Then, a synthetic
gene set of 25 genes was perturbed using a normal distribution gradient. In cell line 1, the 25 genes were
replaced with values from a normal distribution with µ = -X, σ = 1, and in cell line 300, the 25 genes were
replaced with values from a normal distribution with µ = +X, σ = 1. For cell lines 2-299, the 25 genes were
replaced with values from normal distributions with µ sequentially increasing from -X to X. Single-sample
Gene Set Enrichment Analysis (ssGSEA) NESs were then calculated for the synthetic gene set for all 300
cell lines. Next, gene dependency data was simulated for the same 300 cell lines using the same normal
distribution gradient method. For both gene expression and gene dependency data, values for the
perturbation X were varied from 0 to 1. Then, Spearman correlation coefficients between synthetic gene set
activity (NES) and gene dependency were calculated for all 16,643 genes. Finally, Gene Set Enrichment
Analysis was run to calculate the simulated Genetic PDEA values as outlined in Fig. 13A. We found that
both the correlation coefficients and Genetic PDEA results were more strongly influenced by expression
gradients added than dependency gradients added (Fig. 13 B,C). This suggests that our Genetic PDEA
approach to identify metabolic pathway vulnerabilities in cancer cells is more sensitive to changes in
metabolic pathway activity than changes in gene dependency. However, when the perturbation X was large
for the dependency gradient and small for the expression gradient (or vice versa), significant Genetic PDEA
values were still obtained.
73
Figure 13. Integration of gene expression and CRISPR gene dependencies to identify metabolic
pathway dependencies.
A) Schematic outlining the approach for Genetic Pathway Dependency Enrichment Analysis (Genetic
PDEA). Cancer cell lines from the CCLE were first stratified by culture type (adherent, suspension) and
culture medium (RPMI, DMEM), and then their metabolic pathway activity was inferred using single-
sample GSEA (ssGSEA). The resulting pathway activities were integrated with gene dependency and drug
response data to assess association with metabolic pathway activity. B-C) Simulated data (see methods)
was used to assess the sensitivity of the Genetic PDEA approach. The heatmaps represent the percentage
of significant results at each gradient added. Values added to the expression data resulted in slightly stronger
correlation coefficients and Genetic PDEA results compared to values added to dependency data.
3.4 Metabolic Pathway Dependency is Highly Context Specific
Having validated the sensitivity of our Genetic PDEA approach, we next investigated cancer cell
dependency across all KEGG metabolic pathways by clustering Genetic PDEA NES values across all
74
pathway activities (columns) and pathway dependencies (rows) (Fig. 14A & Supp. Fig 15A). Pathways
with similar functions clustered together within metabolic pathway activity more than metabolic pathway
dependency. For example, glycan biosynthesis pathways cluster together on the pathway activity level but
not on the pathway dependency level. We also found that no single metabolic pathway had uniquely positive
or negative NES values, suggesting that cancer cells are not universally dependent on any metabolic
pathway. Interestingly, the strongest metabolic dependency of adherent RMPI cancer cell lines was Folate
Biosynthesis (hsa00790) when One Carbon Pool by Folate (hsa00670) pathway activity was high (Fig.
14B). This enrichment was driven by the genes QDPR, ALPI, ALPP, and GCH1. Notably, Folate
Biosynthesis is directly upstream of One Carbon Pool by Folate. Furthermore, one of the strongest Genetic
PDEA results in both the adherent DMEM and adherent RPMI analyses recapitulated a link between one
carbon metabolism and the TCA cycle
124
(Fig. 14C & Supp. Fig. 15B). Additionally, some metabolic
pathways exhibited context-specific dependencies. For example, the dependency on the TCA Cycle
(hsa00020) was increased in cell lines with decreased Glycolysis-Gluconeogenesis (hsa00010) activity,
whereas dependency on the TCA cycle was increased in cell lines with increased Pentose Phosphate
Pathway (hsa00030) activity (Supp. Fig. 16). This suggests that the diversion of glucose from glycolysis
to the pentose phosphate pathway may confer increased dependency on the TCA Cycle. We next asked
whether there existed a general relationship between a metabolic pathway’s activity and its essentiality. Of
the 69 metabolic pathways queried, 36 had a negative NES from Genetic PDEA and 33 had a positive NES
for both adherent RPMI and adherent DMEM cell lines (Supp. Fig. 17A). Notably, the 36 negative NES
and 33 positive NES were not from the same metabolic pathways for the two analyses (Supp. Fig. 17B).
These results suggest that there is no generic rule regarding a metabolic pathway’s activity and its
essentiality. Instead, these results indicate that metabolic pathway dependency is highly context specific.
75
76
Figure 14. Global analysis of metabolic dependency data reveals context-specific pathway
essentialities.
A) Metabolic pathway activity was inferred using single-sample GSEA for 300 adherent cell lines cultured
in RPMI and correlated to gene dependency data from The Cancer Dependency Map (DepMap). Correlation
coefficients were then ranked and Genetic Pathway Dependency Enrichment Analysis (Genetic PDEA) was
run using the KEGG metabolic pathways (see Fig. 1). Hierarchical clustering was performed on the Genetic
PDEA normalized enrichment scores (NES). Results for pathways with FDR < 0.25 are plotted. Dots are
colored according to their NES and sized according to the -log10 of the false discovery rate (FDR). Results
shown in B and C are highlighted with a black outline. B) Cancer cell dependency on Folate Biosynthesis
(hsa00790) was increased when One Carbon Pool by Folate (hsa00670) pathway activity was high. The
scatter plots of pathway activity NES and gene dependency (-CERES) for leading-edge genes QDPR and
ALPI are shown. C) Dependency on One Carbon Pool by Folate metabolism (hsa00670) is increased when
TCA cycle (hsa00020) activity is increased. The scatter plots of pathway activity NES and gene dependency
(-CERES) for leading-edge genes MTR and MTHFD1 are shown.
77
78
Supporting Figure 16. Genetic PDEA results for Adherent DMEM cell lines. Metabolic pathway activity
was inferred using single-sample GSEA for 153 adherent cell lines cultured in DMEM and correlated to
gene dependency data from The Cancer Dependency Map (DepMap). Correlation coefficients were then
ranked and GSEA was run querying the KEGG metabolic pathways (see Figure 13). A) Hierarchical
clustering was performed on the Genetic PDEA normalized enrichment scores (NES). Results for pathways
with FDR < 0.25 are plotted. Dots are colored according to their NES and sized according to the -log 10 of
the false discovery rate (FDR). Numerical values for each pathway can be found in Supp. Table #. B) TCA
cycle activity is positively associated with increased one carbon pool by folate metabolism dependency.
Supporting Figure 17. TCA Cycle dependency changes upon glycolysis versus pentose phosphate
pathway activity. The dependency on TCA Cycle (hsa00020) is context specific. When Glycolysis-
Gluconeogenesis (hsa00010) pathway activity was low, the dependence on TCA Cycle was increased (top).
When Pentose Phosphate Pathway activity (hsa00030), the dependence on the TCA Cycle was increased
(bottom). The scatter plots of pathway activity NES and gene dependency (-CERES) for leading-edge genes
ACO2 and SDHD with Glycolysis-Gluconeogenesis and Pentose Phosphate Pathway, respectively, are
shown.
79
Supporting Figure 18. Pathway activity does not correlate with pathway dependency. Metabolic pathway
expression was correlated with gene dependency data and GSEA was run on the resulting correlation
coefficients (see Figure 13). Then, results were filtered for results that had the same pathway expression
and pathway dependency queried (e.g. Glycolysis dependency GSEA was queried against all correlations
with Glycolysis expression). A) The resulting normalized enrichment scores (NES) are presented as a
density plot for both Adherent RPMI and Adherent DMEM analyses. The distribution of NES is centered
around 0. B) The NES for each self-dependency is plotted for Adherent RPMI and Adherent DMEM.
3.5 Media Composition Influences Metabolic Pathway Dependency
We next investigated whether cancer cell line metabolic pathway dependencies were influenced by
media composition. To assess the essentiality of a pathway, we weighted each NES from Genetic PDEA
by its -log 10 FDR and then took the mean of all weighted NES. We found striking differences between
DMEM and RPMI metabolic pathway essentialities that can be partly explained by media composition,
consistent with the finding that the essentiality of individual metabolic genes is influenced by culture
medium
114
(Fig. 15). For example, cancer cells cultured in RPMI exhibited a strongly positive average
80
weighted NES for Folate Biosynthesis (hsa00790) whereas cancer cells cultured in DMEM did not. DMEM
contains four times the concentration of folate (4 mg/L) compared to RPMI (1 mg/L), suggesting that cancer
cells grown in DMEM need to synthesize less folate, thereby reducing their dependency on Folate
Biosynthesis. Similarly, cancer cells grown in DMEM were more dependent on Oxidative Phosphorylation
(hsa00190) than cancer cells grown in RPMI. One function of oxidative phosphorylation is to enable
aspartate synthesis to accept electrons from the electron transport chain
125,126
. Since RPMI and DMEM
contain 150 µM and 0 µM aspartate, respectively, the increased dependency on oxidative phosphorylation
in DMEM may reflect an increased need for aspartate synthesis. Taken together, these results suggest that
media composition influences cancer cell line metabolic pathway dependency and that future studies of
metabolic vulnerabilities should take media composition into consideration.
81
Figure 15. Media composition influences metabolic pathway dependency.
For adherent cancer cell lines cultured in RPMI (Fig. 14) and DMEM (Supp. Fig. 15A), the metabolic
pathway dependency NESs from Genetic PDEA analysis were weighted by -log 10 FDR. The weighted NESs
were then averaged across all 69 KEGG metabolic pathways. Pathways are ranked by the difference
between DMEM and RPMI. Relative media composition between RPMI and DMEM are on the right. The
concentration of folate in RPMI and DMEM is 4 mg/L and 1 mg/L, respectively. The dependency on Folate
Biosynthesis was much higher in DMEM than in RPMI because these cells must synthesize more folate.
Conversely, the dependency on oxidative phosphorylation is much higher in DMEM. This may be due to
differences in aspartate levels (RPMI 150 µM, DMEM 0 µM). The indicated pathways are highlighted in
bold.
3.6 Metabolic pathway activity is correlated with anti-cancer drug sensitivity
We next sought to integrate metabolic pathway activity with large scale pharmacologic screens.
We used the PRISM drug repurposing database
113
, which contains 1,448 compounds screened against 499
82
cell lines at 8 different doses. Once again, cell lines were separately processed by culture type and culture
medium with a focus on adherent cell lines. Compounds measured in less than 150 adherent RPMI cell
lines were removed from the analysis, leaving 1,390 compounds. We then correlated drug response area-
under-the-curve (AUC) with metabolic pathway activity after multiplying the response of drugs classified
as positive regulators (e.g., agonists) by -1 for directional consistency (Fig. 16A). First, we asked whether
there were any drugs with significant directional agreement across the RPMI and DMEM analyses. From a
possible 101,360 possible drug:metabolic pathway combinations, 66 combinations passed FDR-corrected
significance thresholds (q < 0.05) and were of the same sign in RPMI and DMEM. Notably, zero results
that passed FDR correction were of different sign. Many of the common associations in RPMI and DMEM
were tyrosine kinase inhibitors (TKIs), which have been extensively linked to metabolism
127
. Interestingly,
we found a strong association between decreased Core Glycolysis (hsa_M00001) pathway activity and
increased sensitivity to AZD8931, an inhibitor of EGFR and ERBB2 (HER2) (Fig. 16B). We also found a
strong association between increased α-linoleic acid metabolism (hsa00592) and sensitivity to afatinib,
another EGFR inhibitor (Fig. 16C). Of the non-TKI results, we found a link between decreased
phenylalanine metabolism (hsa00360) activity and increased sensitivity to atorvastatin, an HMGR inhibitor
(Fig. 16D). HMGR is the rate limiting enzyme in the cholesterol biosynthetic pathway
128
and multiple
reports have suggested that elevated levels of phenylalanine inhibit cholesterol biosynthesis
129–131
. Increased
response to atorvastatin when phenylalanine metabolism activity is low suggests that decreased
phenylalanine metabolism and HMGR inhibitors may be redundant. Lastly, we found a known link between
decreased mucin type O-glycan biosynthesis pathway (hsa00512) activity and increased sensitivity to the
HSP90 inhibitor NMS-E973
132
(Fig. 16E). Taken together, these results indicate that metabolic pathway
activity can be used as robust biomarkers of drug sensitivity.
83
Figure 16. Metabolic pathway activity is correlated with anti-cancer drug sensitivity.
A) Schematic representing the strategy used to integrate metabolic pathway activity with drug response
screens. Similar to Genetic PDEA (Fig. 1), the drug response (area-under-the-curve, AUC) for individual
cancer cell lines was correlated to metabolic pathway activity as measured by ssGSEA. Drugs classified as
activators (e.g., agonists) were multiplied by -1 for directional consistency. Cancer cell lines were
separately processed by culture type and culture medium with a focus on adherent cell lines. All correlation
p-values were FDR corrected using a Benjamini-Hochberg correction. Here, individual drug-pathway
correlations are shown. Drugs were also mapped to metabolic pathways using their annotated gene targets
and then drug set enrichment analysis (Pharmacologic PDEA) was run on the resulting drug sets. Those
results are presented in Fig. 5. B-E) Scatter plots of significant drug:metabolic pathway combinations (FDR
< 0.05) in both DMEM and RPMI mediums. Correlation coefficients and false discovery rates are shown
for each correlation. The annotated gene target of each drug is listed below the drug name.
84
3.7 Pharmacological Pathway Dependency Enrichment Analysis Reveals Common Metabolic
Pathway Vulnerabilities
Having identified strong associations between metabolic pathway activity and individual drugs, we
next asked whether there were commonalities in the response of cancer cell lines to families of metabolic
pathway inhibitors. Drugs were mapped to metabolic pathways using their annotated target(s) and grouped
according to the KEGG metabolic pathways database. Drugs were also classified as positive or negative
regulators based on their annotated mechanism of action. For example, drugs labeled as “agonists” or
“activators” were classified as positive regulators whereas “blockers” and “antagonists” were classified as
negative regulators. To enable a consistent pathway analysis, correlation coefficients for positive regulators
were multiplied by -1. Pathways with less than 4 drugs were omitted from the analysis, leaving 46 sets of
drugs targeting metabolic pathways. Pharmacological Pathway Dependency Enrichment Analysis
(Pharmacological PDEA) was run on the resulting correlation coefficients from Fig. 4A for each metabolic
pathway (Fig. 17A). To test the sensitivity of this approach, we again performed a simulation study with
1,390 drugs instead of 16,643 genes. Similar to Genetic PDEA, we found that expression gradients resulted
in stronger results than dependency gradients for both individual drug correlation coefficients and
Pharmacological PDEA (Supp. Fig. 17).
We then clustered the Pharmacological PDEA NESs and found that metabolic pathways clustered
closely in terms of their activity and less so for inhibition, similar to our results with Genetic PDEA (Fig.
14). For example, the activities of Core Glycolysis (hsa_M00001), Fructose and Mannose Metabolism
(hsa00051), Starch and Sucrose Metabolism (hsa00500), and Pentose Phosphate Pathway (hsa00030)
clustered together. Among the strongest Pharmacological PDEA results, we found that sensitivity to
inhibitors of Terpenoid Backbone Biosynthesis (hsa00900) cancer cells was increased in cancer cells with
high Alanine, Aspartate, and Glutamate Metabolism (hsa00250) (Fig 17B). We also found an interesting
link between decreased pentose phosphate pathway (PPP) pathway activity and increased sensitivity to
85
folate biosynthesis inhibitors (Fig. 17C). Because folate biosynthesis inhibitors prevent the generation of
NADPH via one-carbon metabolism, these inhibitors may be more damaging to cellular redox balance
when PPP expression is low. In fact, most strong results for inhibitors of folate biosynthesis occur when
metabolic pathway activity is low (Fig. 17D). Conversely, inhibitors of Ascorbate and Aldarate metabolism
(hsa00053) are more effective when metabolic pathway activity is high (Fig. 17E). This may be because
ascorbate (also known as Vitamin C) is an effective antioxidant used to detoxify reactive oxygen species
(ROS). ROS are a byproduct of many metabolic reactions such as oxidative phosphorylation and
methionine metabolism, which indirectly produces ROS by supporting polyamine synthesis
133
. Indeed, the
sensitivity to inhibitors of Ascorbate and Aldarate metabolism is stronger when expression of these ROS
producing pathways is high (Fig. 17F) suggesting that ascorbate’s role as an antioxidant is crucial in this
context. Taken together, these results reveal a set of contexts in which inhibition of metabolic pathways
results in decreased cell fitness.
86
Figure 17. Pharmacological PDEA Reveals Common Metabolic Pathway Vulnerabilities.
A) Pharmacological PDEA (see Fig. 4A) was performed on 1,390 anti-cancer drugs from the PRISM
database. Drugs were mapped to metabolic pathways by their annotated target(s). Hierarchical clustering
was performed on NES and results with FDR < 0.25 are plotted. Dots are colored according to the NES and
87
sized according to the -log 10 FDR. Dots with black outline correspond to results shown in panel B. B-C)
Representative mountain plots and the drug(s) driving enrichment are shown. D-E) Inhibitors of Folate
Biosynthesis (hsa00790) are more effective when metabolic pathway expression is low, whereas inhibitors
of Ascorbate and Aldarate Metabolism (hsa00053) are more effective when metabolic pathway expression
is high. F) Representative mountain plots and the drug(s) driving enrichment of metabolic pathway
activities that strongly correlate with response to inhibitors of Ascorbate and Aldarate metabolism are
shown.
88
89
Supporting Figure 19. Pharmacological Pathway Dependency Enrichment Analysis Simulation Study.
Simulated data (see methods) was used to assess the sensitivity of the Pharmacological PDEA approach.
Values added to the expression data resulted in slightly stronger correlation coefficients and
Pharmacological PDEA results compared to values added to dependency data.
Supporting Figure 20. Pharmacological PDEA Reveals Metabolic Pathway Vulnerabilities in Adherent
DMEM cell lines. A) Pharmacological PDEA (see Fig. 4A) was performed on 1,390 anti-cancer drugs from
the PRISM database for 97 adherent cell lines grown in DMEM. Drugs were mapped to metabolic pathways
by their annotated target(s). Hierarchical clustering was performed on NES and results with FDR < 0.25
are plotted. Dots are colored according to the NES and sized according to the -log10 FDR.
90
3.8 Integration of pharmacologic and genetic screens reveals common metabolic vulnerabilities
Next, we sought to integrate results from genetic and pharmacological screen data to identify
metabolic pathway dependencies found independently in both analyses. First, we integrated individual gene
correlations with their corresponding drug correlations by first annotating each drug with its gene target(s).
We then summed the gene dependency correlation coefficient and the drug correlation coefficient for each
drug target and assessed significance by a permutation test and FDR correction (Fig. 18A). Out of 187,818
gene+drug:metabolic pathway combinations, we found 176 results that passed an FDR-corrected
significance threshold of 0.01. Interestingly, all significant results targeted known cancer driver genes such
as EGFR, HER2, PIK3CA, and BRAF (Fig. 18B-E). These results included a known interaction between
HER2 inhibitors and retinol metabolism, whereby increased retinol metabolism enhances sensitivity to
HER2 inhibition
134
(Fig. 18C). Additionally, some results identified well known molecular interactions,
such as BRAF and PIK3CA driving sugar metabolism
135,136
(Fig. 18D,E). These results demonstrate robust
associations between metabolic pathway activity, gene dependency, and drug response.
Lastly, we sought to integrate the results from Genetic PDEA and Pharmacological PDEA to
identify metabolic pathway vulnerabilities that were consistent between gene dependency and drug
response data (Fig. 18F). By applying p-value and q-value filters to each analysis, we found 3 common
vulnerabilities. First, we found that when tyrosine metabolism is high, there is an increased vulnerability to
inhibition and knockout of terpenoid backbone biosynthesis genes (Fig. 18G). Interestingly, the drugs
driving the enrichment (fluvastatin and pitavastatin) target the top gene dependency within the pathway,
HMGCR. We also found a common vulnerability between inactivation of the folate biosynthesis pathway
and decreased aminoacyl tRNA biosynthesis (Fig. 18H). Here, the top hits in Genetic PDEA and
Pharmacological PDEA did not converge on a single protein product. Nevertheless, these results indicate
that inactivation of the folate biosynthesis pathway is more effective at slowing cancer cell growth when
aminoacyl tRNA biosynthesis pathway activity is low. Lastly, we found a strong association between
inhibitors and gene knockouts of terpenoid backbone biosynthesis when pathway activity for biosynthesis
91
of heparan sulfate is low (Fig. 18I). Once again, the statins driving the Pharmacological PDEA enrichment
do not align with the top gene dependencies (DHDDS, HMGCS1). Taken together, these results demonstrate
common metabolic pathway vulnerabilities by integrating gene dependency, drug response, and gene
expression data.
92
Figure 18. Integration of Pharmacological and Genetic Screens Reveal Common Metabolic Pathway
Vulnerabilities.
93
A) Schematic outlining approach to identify drugs targets and genetic dependencies that are commonly
increased or decreased with metabolic pathway activity. Significance was assessed by permutation testing
combined with Benjamini-Hochberg FDR correction. 176 significant associations of 187,818
gene+drug:metabolic pathway combinations passed the FDR threshold of 0.01. B-E) Scatter plots of four
drug response and CRISPR gene dependencies associated with metabolic pathway activity. The gene target
of each drug is listed below the drug name. F) Schematic outlining a filtering approach used to identify
common pathway-level vulnerabilities in Genetic PDEA and Pharmacologic PDEA. G-I) Mountain plots
and leading edge drugs and genes from the three common pathway vulnerabilities are shown.
3.9 Discussion
Traditional gene dependency analyses have been limited to associations between individual genes.
Here, using metabolic pathways as an example, we have demonstrated an approach for identifying cancer
cell dependencies on the pathway level. Illustrating the utility of our approach, we recapitulated known
interactions between metabolic pathway activity, drug response, and individual gene dependency. These
results build on a strong foundation of research identifying metabolic vulnerabilities in cancer cells
108,114
.
By extending this knowledge to include metabolic pathways we uncovered novel metabolic crosstalk,
identified robust associations between drug response and metabolic pathway expression, and have
discovered associations between metabolic pathway expression and essentiality.
A common, but not unexpected, theme from our analyses is the importance of the pathways Folate
Biosynthesis (hsa00790) and One Carbon Pool by Folate (hsa00670). In fact, two of the most widely used
chemotherapeutics, methotrexate and 5-fluorouracil, target folate biosynthesis. Folate metabolism supports
two key metabolic phenotypes commonly found in cancer cells by producing one carbon units for
nucleotide synthesis and maintaining redox balance through production of NADPH
137
. In our analysis, we
found that inhibitors of folate biosynthesis are highly effective when activity of other metabolic pathways
is low (Fig. 17D). This may be because the drugs classified as folate biosynthesis inhibitors are anti-
metabolites that cannot be metabolized by enzymes like thymidylate synthase or dihydrofolate reductase.
When adjacent metabolic pathway expression is low, compensatory mechanisms cannot pick up the slack
from decreased folate biosynthesis, causing a crisis in both nucleotide synthesis and redox homeostasis.
94
Our results suggest that analyzing metabolic pathway activity may provide biomarkers that would enable
advances in patient selection for antifolate chemotherapy.
The approach outlined here also sets the stage for the use of metabolic pathways to guide patient
selection to therapy. Even for some common cancer targets, there still exists a need to identify additional
features that inform patient selection. Here, we identified metabolic pathways that strongly correlate with
both CRISPR knockout and pharmacological inhibition of EGFR, HER2, BRAF, and PIK3CA. The use of
pathways, rather than individual genes, may serve as biomarkers that are more predictive of response to
therapy.
Our analysis identified media composition as a major confounding factor when analyzing cancer
cell metabolic pathway dependencies. This finding is consistent with recent studies and demonstrates the
effect to which metabolism and metabolic vulnerabilities are shaped by the tumor microenvironment
114,138
.
These results highlight the importance of formulating cell culture mediums that better recapitulate the tumor
microenvironment
139
. Furthermore, the data used in this study comes from adherent cell lines cultured on
tissue culture plastic. This removes environmental stresses such as concentration gradients and physical
stimuli that cells experience in real tumors. Recent efforts have demonstrated that CRISPR-Cas9 screens
can be performed in 3D organoids
138
. As this technology becomes more widely used, computational
approaches such as ours can be applied to identify differences between 2D and 3D culture. Accordingly,
culture conditions that better reflect the physiological conditions of tumors will enhance the therapeutic
relevance of our approach.
While our study identified robust associations between drug response and metabolic pathway
expression, these analyses (Figs. 17 and 18) rely on the annotated targets of these drugs. Off-target toxicity
is a major concern when using small molecule inhibitors. In fact, some recent studies have found that off-
target toxicity drives the anti-tumor effect of these compounds
119
. We cannot exclude the possibility that
95
off-target effects of these compounds could cause the associations identified here. Furthermore, some
compounds in this study are quite promiscuous and have multiple annotated protein targets. This
promiscuity confounds the Pharmacologic PDEA analysis (Fig. 17) since some drugs were mapped to
multiple metabolic pathways. As such, the utility of the Pharmacologic PDEA approach lies in the
aggregation of multiple drugs to arrive at a significant conclusion, rather than treating each individual drug
as significant. We believe that integrating the Pharmacologic PDEA results with the Genetic PDEA results
(Fig. 18) provides more confidence in the association between metabolic pathway activity and pathway
dependency.
Another potential weakness of our study is that we are “inferring” metabolic pathway activity from
gene expression data. Gene expression often does not correlate with protein expression
140
. Furthermore,
metabolic enzyme activity can be regulated by post translational modifications. By using gene expression
data, we have not accounted for these factors, and as such this may not reflect activity at the metabolic flux
level. More comprehensive proteomic profiling of cancer cell lines is necessary to extend this analysis to
the proteome.
This study extends previous efforts, and utilizes new pharmacologic screen data, to identify
metabolic vulnerabilities in cancer cells to the metabolic pathway level. We anticipate this approach to be
useful beyond analysis of metabolic pathways and could be extended to other biologically relevant
pathways. Furthermore, the utility of approaches like this is likely to increase as CRISPR-Cas9 and
pharmacologic screen data expands to include more cancer cell lines. The therapeutic relevance of this
approach will also continue to expand as large scale pharmacologic and genetic screens become more
ubiquitous in 3D organoids. In conclusion, this study uncovers cancer cell dependencies on metabolic
pathways and serves as a framework for integrating gene expression, cell fitness, and drug response data
on the pathway level.
96
Bibliography
1. Joly, J. H., Delfarah, A., Phung, P. S., Parrish, S. & Graham, N. A. A synthetic lethal drug combination
mimics glucose deprivation–induced cancer cell death in the presence of glucose. J. Biol. Chem. 295,
1350–1365 (2020).
2. Joly, J. H., Lowry, W. E. & Graham, N. A. Differential Gene Set Enrichment Analysis: A statistical
approach to quantify the relative enrichment of two gene sets. Bioinformatics
doi:10.1093/bioinformatics/btaa658.
3. Joly, J. H., Chew, B. T. L. & Graham, N. A. The landscape of metabolic pathway dependencies in
cancer cell lines. bioRxiv 2020.10.05.327429 (2020) doi:10.1101/2020.10.05.327429.
4. Hanahan, D. & Weinberg, R. A. Hallmarks of Cancer: The Next Generation. Cell 144, 646–674 (2011).
5. Vander Heiden, M. G. Targeting cancer metabolism: a therapeutic window opens. Nat. Rev. Drug
Discov. 10, 671–684 (2011).
6. Hamanaka, R. B. & Chandel, N. S. Targeting glucose metabolism for cancer therapy. J. Exp. Med. 209,
211–215 (2012).
7. Galluzzi, L., Kepp, O., Vander Heiden, M. G. & Kroemer, G. Metabolic targets for cancer therapy.
Nat. Rev. Drug Discov. 12, 829–846 (2013).
8. Janku, F., Yap, T. A. & Meric-Bernstam, F. Targeting the PI3K pathway in cancer: are we making
headway? Nat. Rev. Clin. Oncol. 15, 273–291 (2018).
9. Ganapathy-Kanniappan, S., Kunjithapatham, R. & Geschwind, J.-F. Glyceraldehyde-3-Phosphate
Dehydrogenase: A Promising Target for Molecular Therapy in Hepatocellular Carcinoma. Oncotarget
3, 940–953 (2012).
10. Sullivan, I. & Planchard, D. Next-Generation EGFR Tyrosine Kinase Inhibitors for Treating EGFR-
Mutant Lung Cancer beyond First Line. Front. Med. 3, 76 (2016).
97
11. Abdelaziz, A. & Vaishampayan, U. Cabozantinib for the treatment of kidney cancer. Expert Rev.
Anticancer Ther. 17, 577–584 (2017).
12. Elstrom, R. L. et al. Akt stimulates aerobic glycolysis in cancer cells. Cancer Res. 64, 3892–3899
(2004).
13. Priebe, A. et al. Glucose deprivation activates AMPK and induces cell death through modulation of
Akt in ovarian cancer cells. Gynecol. Oncol. 122, 389–395 (2011).
14. Jeon, S.-M., Chandel, N. S. & Hay, N. AMPK regulates NADPH homeostasis to promote tumour cell
survival during energy stress. Nature 485, 661–665 (2012).
15. Choo, A. Y. et al. Glucose addiction of TSC-null cells is caused by failed mTORC1-dependent
balancing of metabolic demand with supply. Mol. Cell 38, 487–499 (2010).
16. Graham, N. A. et al. Glucose deprivation activates a metabolic and signaling amplification loop leading
to cell death. Mol. Syst. Biol. 8, 589 (2012).
17. Gu, Y. et al. mTORC2 Regulates Amino Acid Metabolism in Cancer by Phosphorylation of the
Cystine-Glutamate Antiporter xCT. Mol. Cell 67, 128-138.e7 (2017).
18. Lee, H. Y. et al. Ca2+-dependent demethylation of phosphatase PP2Ac promotes glucose deprivation–
induced cell death independently of inhibiting glycolysis. Sci Signal 11, eaam7893 (2018).
19. Warburg, O. On the Origin of Cancer Cells. Science 123, 309–314 (1956).
20. Czernin, J. & Phelps, and M. E. Positron Emission Tomography Scanning: Current and Future
Applications. Annu. Rev. Med. 53, 89–112 (2002).
21. Hopkins, B. D. et al. Publisher Correction: Suppression of insulin feedback enhances the efficacy of
PI3K inhibitors. Nature 563, E24 (2018).
22. Gao, X. et al. Dietary methionine influences therapy in mouse cancer models and alters human
metabolism. Nature 1 (2019) doi:10.1038/s41586-019-1437-3.
23. Chan, D. A. et al. Targeting GLUT1 and the Warburg Effect in Renal Cell Carcinoma by Chemical
Synthetic Lethality. Sci. Transl. Med. 3, 94ra70 (2011).
98
24. Fath, M. A. et al. Mitochondrial electron transport chain blockers enhance 2-deoxy-D-glucose induced
oxidative stress and cell killing in human colon carcinoma cells. Cancer Biol. Ther. 8, 1228–1236
(2009).
25. Jae, H. J. et al. The Antitumor Effect and Hepatotoxicity of a Hexokinase II Inhibitor 3-Bromopyruvate:
In Vivo Investigation of Intraarterial Administration in a Rabbit VX2 Hepatoma Model. Korean J.
Radiol. 10, 596–603 (2009).
26. Subramanian, A. et al. Gene set enrichment analysis: A knowledge-based approach for interpreting
genome-wide expression profiles. Proc. Natl. Acad. Sci. 102, 15545–15550 (2005).
27. O’Neil, N. J., Bailey, M. L. & Hieter, P. Synthetic lethality and cancer. Nat. Rev. Genet. 18, 613–623
(2017).
28. Dey, P. et al. Genomic deletion of malic enzyme 2 confers collateral lethality in pancreatic cancer.
Nature 542, 119–123 (2017).
29. Qi, L. S. et al. Repurposing CRISPR as an RNA-Guided Platform for Sequence-Specific Control of
Gene Expression. Cell 152, 1173–1183 (2013).
30. Chavez, A. et al. Highly efficient Cas9-mediated transcriptional programming. Nat. Methods 12, 326–
328 (2015).
31. Lu, W., Wang, L., Chen, L., Hui, S. & Rabinowitz, J. D. Extraction and Quantitation of Nicotinamide
Adenine Dinucleotide Redox Cofactors. Antioxid. Redox Signal. 28, 167–179 (2018).
32. White, C. C., Viernes, H., Krejsa, C. M., Botta, D. & Kavanagh, T. J. Fluorescence-based microtiter
plate assay for glutamate-cysteine ligase activity. Anal. Biochem. 318, 175–180 (2003).
33. Tsherniak, A. et al. Defining a Cancer Dependency Map. Cell 170, 564-576.e16 (2017).
34. Neftel, C. et al. An Integrative Model of Cellular States, Plasticity, and Genetics for Glioblastoma. Cell
(2019) doi:10.1016/j.cell.2019.06.024.
35. Jain, M. et al. Metabolite profiling identifies a key role for glycine in rapid cancer cell proliferation.
Science 336, 1040–1044 (2012).
99
36. Gmeiner, W. H., Reinhold, W. C. & Pommier, Y. Genome-wide mRNA and microRNA profiling of
the NCI 60 cell-line screen and comparison of FdUMP[10] with fluorouracil, floxuridine, and
topoisomerase 1 poisons. Mol. Cancer Ther. 9, 3105–3114 (2010).
37. Ye, I. C. et al. Molecular Portrait of Hypoxia in Breast Cancer: A Prognostic Signature and Novel HIF-
Regulated Genes. Mol. Cancer Res. MCR 16, 1889–1901 (2018).
38. Goji, T., Takahara, K., Negishi, M. & Katoh, H. Cystine uptake through the cystine/glutamate
antiporter xCT triggers glioblastoma cell death under glucose deprivation. J. Biol. Chem. 292, 19721–
19732 (2017).
39. Graham, N. A. et al. Recurrent patterns of DNA copy number alterations in tumors reflect metabolic
selection pressures. Mol. Syst. Biol. 13, 914 (2017).
40. Delfarah, A. et al. Inhibition of nucleotide synthesis mediates replicative senescence of human
mammary epithelial cells. bioRxiv 423665 (2018) doi:10.1101/423665.
41. Yuan, M., Breitkopf, S. B., Yang, X. & Asara, J. M. A positive/negative ion-switching, targeted mass
spectrometry-based metabolomics platform for bodily fluids, cells, and fresh and fixed tissue. Nat.
Protoc. 7, 872–881 (2012).
42. Metallo, C. M. et al. Reductive glutamine metabolism by IDH1 mediates lipogenesis under hypoxia.
Nature 481, 380–384 (2012).
43. Mullen, A. R. et al. Reductive carboxylation supports growth in tumour cells with defective
mitochondria. Nature 481, 385–388 (2012).
44. Winterbourn, C. C. & Hampton, M. B. Thiol chemistry and specificity in redox signaling. Free Radic.
Biol. Med. 45, 549–561 (2008).
45. Meister, A. & Anderson, M. E. Glutathione. Annu. Rev. Biochem. 52, 711–760 (1983).
46. Franklin, C. C. et al. Structure, function, and post-translational regulation of the catalytic and modifier
subunits of glutamate cysteine ligase. Mol. Aspects Med. 30, 86–98 (2009).
100
47. Huang, C. S., Chang, L. S., Anderson, M. E. & Meister, A. Catalytic and regulatory properties of the
heavy subunit of rat kidney gamma-glutamylcysteine synthetase. J. Biol. Chem. 268, 19675–19680
(1993).
48. Krejsa, C. M. et al. Rapid Activation of Glutamate Cysteine Ligase following Oxidative Stress. J. Biol.
Chem. 285, 16116–16124 (2010).
49. Ho, S.-M. et al. Evaluating Synthetic Activation and Repression of Neuropsychiatric-Related Genes in
hiPSC-Derived NPCs, Neurons, and Astrocytes. Stem Cell Rep. 9, 615–628 (2017).
50. Fan, J. et al. Quantitative flux analysis reveals folate-dependent NADPH production. Nature 510, 298–
302 (2014).
51. Pader, I. et al. Thioredoxin-related protein of 14 kDa is an efficient L-cystine reductase and S-
denitrosylase. Proc. Natl. Acad. Sci. U. S. A. 111, 6964–6969 (2014).
52. Ottestad-Hansen, S. et al. The cystine-glutamate exchanger (xCT, Slc7a11) is expressed in significant
concentrations in a subpopulation of astrocytes in the mouse brain. Glia 66, 951–970 (2018).
53. Shin, C.-S. et al. The glutamate/cystine xCT antiporter antagonizes glutamine metabolism and reduces
nutrient flexibility. Nat. Commun. 8, ncomms15074 (2017).
54. Koppula, P., Zhang, Y., Shi, J., Li, W. & Gan, B. The glutamate/cystine antiporter SLC7A11/xCT
enhances cancer cell dependency on glucose by exporting glutamate. J. Biol. Chem. jbc.M117.798405
(2017) doi:10.1074/jbc.M117.798405.
55. Ho, S.-M. et al. Evaluating Synthetic Activation and Repression of Neuropsychiatric-Related Genes in
hiPSC-Derived NPCs, Neurons, and Astrocytes. Stem Cell Rep. 9, 615–628 (2017).
56. Chou, T. C. & Talalay, P. Quantitative analysis of dose-effect relationships: the combined effects of
multiple drugs or enzyme inhibitors. Adv. Enzyme Regul. 22, 27–55 (1984).
57. Adams, D. J. et al. NAMPT Is the Cellular Target of STF-31-Like Small-Molecule Probes. ACS Chem.
Biol. 9, 2247–2254 (2014).
58. Heiden, M. G. V., Cantley, L. C. & Thompson, C. B. Understanding the Warburg Effect: The Metabolic
Requirements of Cell Proliferation. Science 324, 1029–1033 (2009).
101
59. DeBerardinis, R. J., Lum, J. J., Hatzivassiliou, G. & Thompson, C. B. The Biology of Cancer:
Metabolic Reprogramming Fuels Cell Growth and Proliferation. Cell Metab. 7, 11–20 (2008).
60. Cairns, R. A. & Mak, T. W. The current state of cancer metabolism. Nat. Rev. Cancer 16, 613–614
(2016).
61. Martinez-Outschoorn, U. E., Peiris-Pagés, M., Pestell, R. G., Sotgia, F. & Lisanti, M. P. Cancer
metabolism: a therapeutic perspective. Nat. Rev. Clin. Oncol. 14, 11–31 (2017).
62. Schulze, A. & Harris, A. L. How cancer metabolism is tuned for proliferation and vulnerable to
disruption. Nature 491, 364–373 (2012).
63. Leonard, P. G. et al. SF2312 is a natural phosphonate inhibitor of Enolase. Nat. Chem. Biol. 12, 1053–
1058 (2016).
64. Kryukov, G. V. et al. MTAP deletion confers enhanced dependency on the PRMT5 arginine
methyltransferase in cancer cells. Science 351, 1214–1218 (2016).
65. Mavrakis, K. J. et al. Disordered methionine metabolism in MTAP/CDKN2A-deleted cancers leads to
dependence on PRMT5. Science 351, 1208–1213 (2016).
66. Marjon, K. et al. MTAP Deletions in Cancer Create Vulnerability to Targeting of the
MAT2A/PRMT5/RIOK1 Axis. Cell Rep. 15, 574–587 (2016).
67. Xiao, G. et al. B-Cell-Specific Diversion of Glucose Carbon Utilization Reveals a Unique Vulnerability
in B Cell Malignancies. Cell (2018) doi:10.1016/j.cell.2018.02.048.
68. Chan, D. A. et al. Targeting GLUT1 and the Warburg effect in renal cell carcinoma by chemical
synthetic lethality. Sci. Transl. Med. 3, 94ra70 (2011).
69. Andringa, K. K. et al. Inhibition of glutamate cysteine ligase activity sensitizes human breast cancer
cells to the toxicity of 2-deoxy-D-glucose. Cancer Res. 66, 1605–1610 (2006).
70. Miwa, H. et al. Leukemia cells demonstrate a different metabolic perturbation provoked by 2-
deoxyglucose. Oncol. Rep. 29, 2053–2057 (2013).
71. Murai, S. et al. Inhibition of malic enzyme 1 disrupts cellular metabolism and leads to vulnerability in
cancer cells in glucose-restricted conditions. Oncogenesis 6, e329 (2017).
102
72. Lee, Y. J. et al. Glucose deprivation-induced cytotoxicity and alterations in mitogen-activated protein
kinase activation are mediated by oxidative stress in multidrug-resistant human breast carcinoma cells.
J. Biol. Chem. 273, 5294–5299 (1998).
73. Ahmad, I. M. et al. Mitochondrial O2*- and H2O2 mediate glucose deprivation-induced stress in
human cancer cells. J. Biol. Chem. 280, 4254–4263 (2005).
74. Aykin-Burns, N., Ahmad, I. M., Zhu, Y., Oberley, L. W. & Spitz, D. R. Increased levels of superoxide
and H2O2 mediate the differential susceptibility of cancer cells versus normal cells to glucose
deprivation. Biochem. J. 418, 29–37 (2009).
75. Graham, N. A. et al. Glucose deprivation activates a metabolic and signaling amplification loop leading
to cell death. Mol. Syst. Biol. 8, 589 (2012).
76. Arrick, B. A., Nathan, C. F., Griffith, O. W. & Cohn, Z. A. Glutathione depletion sensitizes tumor cells
to oxidative cytolysis. J. Biol. Chem. 257, 1231–1237 (1982).
77. HIRONO, I. Mechanism of natural and acquired resistance to methyl-bis-(beta-chlorethyl)-amine N-
oxide in ascites tumors. Gann Gan 52, 39–48 (1961).
78. Bailey, H. H. l-S,R-buthionine sulfoximine: historical development and clinical issues. Chem. Biol.
Interact. 111 –112, 239–254 (1998).
79. Villablanca, J. G. et al. A Phase I New Approaches to Neuroblastoma Therapy Study of Buthionine
Sulfoximine and Melphalan With Autologous Stem Cells for Recurrent/Refractory High-Risk
Neuroblastoma. Pediatr. Blood Cancer 63, 1349–1356 (2016).
80. Robert, S. M. et al. SLC7A11 expression is associated with seizures and predicts poor survival in
patients with malignant glioma. Sci. Transl. Med. 7, 289ra86 (2015).
81. Timmerman, L. A. et al. Glutamine Sensitivity Analysis Identifies the xCT Antiporter as a Common
Triple-Negative Breast Tumor Therapeutic Target. Cancer Cell 24, 450–465 (2013).
82. Yoshikawa, M. et al. xCT inhibition depletes CD44v-expressing tumor cells that are resistant to EGFR-
targeted therapy in head and neck squamous cell carcinoma. Cancer Res. 73, 1855–1866 (2013).
103
83. Arensman, M. D. et al. Cystine–glutamate antiporter xCT deficiency suppresses tumor growth while
preserving antitumor immunity. Proc. Natl. Acad. Sci. 116, 9533–9542 (2019).
84. Cramer, S. L. et al. Systemic depletion of L-cyst(e)ine with cyst(e)inase increases reactive oxygen
species and suppresses tumor growth. Nat. Med. 23, 120–127 (2017).
85. Labak, C. M. et al. Glucose transport: meeting the metabolic demands of cancer, and applications in
glioblastoma treatment. Am. J. Cancer Res. 6, 1599–1608 (2016).
86. Mertins, P. et al. Proteogenomics connects somatic mutations to signalling in breast cancer. Nature
534, 55–62 (2016).
87. Drake, J. M. et al. Oncogene-specific activation of tyrosine kinase networks during prostate cancer
progression. Proc. Natl. Acad. Sci. 109, 1643–1648 (2012).
88. Drake, J. M. et al. Phosphoproteome Integration Reveals Patient-Specific Networks in Prostate Cancer.
Cell 166, 1041–1054 (2016).
89. Xia, J. & Wishart, D. S. MSEA: a web-based tool to identify biologically meaningful patterns in
quantitative metabolomic data. Nucleic Acids Res. 38, W71-77 (2010).
90. Delfarah, A. et al. Inhibition of nucleotide synthesis mediates replicative senescence of human
mammary epithelial cells. bioRxiv 423665 (2018) doi:10.1101/423665.
91. Hänzelmann, S., Castelo, R. & Guinney, J. GSVA: gene set variation analysis for microarray and RNA-
seq data. BMC Bioinformatics 14, 7 (2013).
92. Barbie, D. A. et al. Systematic RNA interference reveals that oncogenic KRAS-driven cancers require
TBK1. Nature 462, 108–112 (2009).
93. Lavallée-Adam, M., Rauniyar, N., McClatchy, D. B. & Yates, J. R. PSEA-Quant: A Protein Set
Enrichment Analysis on Label-Free and Label-Based Protein Quantification Data. J. Proteome Res.
13, 5496–5509 (2014).
94. Yoon, S., Kim, S.-Y. & Nam, D. Improving Gene-Set Enrichment Analysis of RNA-Seq Data with
Small Replicates. PLOS ONE 11, e0165919 (2016).
104
95. Majmundar, A. J., Wong, W. J. & Simon, M. C. Hypoxia-Inducible Factors and the Response to
Hypoxic Stress. Mol. Cell 40, 294–309 (2010).
96. Kim, J., Tchernyshyov, I., Semenza, G. L. & Dang, C. V. HIF-1-mediated expression of pyruvate
dehydrogenase kinase: a metabolic switch required for cellular adaptation to hypoxia. Cell Metab. 3,
177–185 (2006).
97. Kanehisa, M. et al. Data, information, knowledge and principle: back to metabolism in KEGG. Nucleic
Acids Res. 42, D199–D205 (2014).
98. Yaari, G., Bolen, C. R., Thakar, J. & Kleinstein, S. H. Quantitative set analysis for gene expression: a
method to quantify gene set differential expression including gene-gene correlations. Nucleic Acids
Res. 41, e170–e170 (2013).
99. Baar, M. P. et al. Targeted Apoptosis of Senescent Cells Restores Tissue Homeostasis in Response to
Chemotoxicity and Aging. Cell 169, 132-147.e16 (2017).
100. Goeman, J. J. & Bühlmann, P. Analyzing gene expression data in terms of gene sets:
methodological issues. Bioinformatics 23, 980–987 (2007).
101. Tamayo, P., Steinhardt, G., Liberzon, A. & Mesirov, J. P. The Limitations of Simple Gene Set
Enrichment Analysis Assuming Gene Independence. Stat. Methods Med. Res. 25, 472–487 (2016).
102. Cho, S. B., Kim, J. & Kim, J. H. Identifying set-wise differential co-expression in gene expression
microarray data. BMC Bioinformatics 10, 109 (2009).
103. Chandel, N. S., Budinger, G. R. S., Choe, S. H. & Schumacker, P. T. Cellular Respiration during
Hypoxia ROLE OF CYTOCHROME OXIDASE AS THE OXYGEN SENSOR IN HEPATOCYTES.
J. Biol. Chem. 272, 18808–18816 (1997).
104. Herzig, S. & Shaw, R. J. AMPK: guardian of metabolism and mitochondrial homeostasis. Nat. Rev.
Mol. Cell Biol. 19, 121–135 (2018).
105. Buczek-Thomas, J. A., Rich, C. B. & Nugent, M. A. Hypoxia Induced Heparan Sulfate Primes the
Extracellular Matrix for Endothelial Cell Recruitment by Facilitating VEGF-Fibronectin Interactions.
Int. J. Mol. Sci. 20, (2019).
105
106. Caterson, B. & Melrose, J. Keratan sulfate, a complex glycosaminoglycan with unique functional
capability. Glycobiology 28, 182–206 (2018).
107. Herranz, N. & Gil, J. Mechanisms and functions of cellular senescence. J. Clin. Invest. 128, 1238–
1246 (2018).
108. Luengo, A., Gui, D. Y. & Vander Heiden, M. G. Targeting Metabolism for Cancer Therapy. Cell
Chem. Biol. 24, 1161–1180 (2017).
109. Graham, N. A. et al. Recurrent patterns of DNA copy number alterations in tumors reflect
metabolic selection pressures. Mol. Syst. Biol. 13, 914 (2017).
110. Maddocks, O. D. K. et al. Serine starvation induces stress and p53-dependent metabolic
remodelling in cancer cells. Nature 493, 542–546 (2013).
111. Muller, F. L. et al. Passenger deletions generate therapeutic vulnerabilities in cancer. Nature 488,
337–342 (2012).
112. Meyers, R. M. et al. Computational correction of copy number effect improves specificity of
CRISPR–Cas9 essentiality screens in cancer cells. Nat. Genet. 49, 1779–1784 (2017).
113. Corsello, S. M. et al. Discovering the anticancer potential of non-oncology drugs by systematic
viability profiling. Nat. Cancer 1–14 (2020) doi:10.1038/s43018-019-0018-6.
114. Lagziel, S., Lee, W. D. & Shlomi, T. Inferring cancer dependencies on metabolic genes from large-
scale genetic screens. BMC Biol. 17, 37 (2019).
115. Gonçalves, E. et al. Drug mechanism-of-action discovery through the integration of
pharmacological and CRISPR screens. Mol. Syst. Biol. 16, e9405 (2020).
116. Barretina, J. et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer
drug sensitivity. Nature 483, 603–607 (2012).
117. Garnett, M. J. et al. Systematic identification of genomic markers of drug sensitivity in cancer cells.
Nature 483, 570–575 (2012).
118. Deans, R. M. et al. Parallel shRNA and CRISPR-Cas9 screens enable antiviral drug target
identification. Nat. Chem. Biol. 12, 361–366 (2016).
106
119. Lin, A. et al. Off-target toxicity is a common mechanism of action of cancer drugs undergoing
clinical trials. Sci. Transl. Med. 11, (2019).
120. Subramanian, A. et al. A Next Generation Connectivity Map: L1000 Platform and the First
1,000,000 Profiles. Cell 171, 1437-1452.e17 (2017).
121. Wang, T. et al. Gene Essentiality Profiling Reveals Gene Networks and Synthetic Lethal
Interactions with Oncogenic Ras. Cell 168, 890-903.e15 (2017).
122. Li, H. et al. The landscape of cancer cell line metabolism. Nat. Med. 25, 850–860 (2019).
123. Kanehisa, M., Goto, S., Furumichi, M., Tanabe, M. & Hirakawa, M. KEGG for representation and
analysis of molecular networks involving diseases and drugs. Nucleic Acids Res. 38, D355-360 (2010).
124. May, J. L. et al. IDH3α regulates one-carbon metabolism in glioblastoma. Sci. Adv. 5, eaat0456
(2019).
125. Birsoy, K. et al. An Essential Role of the Mitochondrial Electron Transport Chain in Cell
Proliferation Is to Enable Aspartate Synthesis. Cell 162, 540–551 (2015).
126. Sullivan, L. B. et al. Supporting Aspartate Biosynthesis Is an Essential Function of Respiration in
Proliferating Cells. Cell 162, 552–563 (2015).
127. Poliaková, M., Aebersold, D. M., Zimmer, Y. & Medová, M. The relevance of tyrosine kinase
inhibitors for global metabolic pathways in cancer. Mol. Cancer 17, (2018).
128. Burg, J. S. & Espenshade, P. J. Regulation of HMG-CoA reductase in mammals and yeast. Prog.
Lipid Res. 50, 403–410 (2011).
129. Bhat, C. S. & Ramasarma, T. Effect of Phenyl and Phenolic Acids on Mevalonate-5-Phosphate
Kinase and Mevalonate-5-Pyrophosphate Decarboxylase of the Rat Brain. J. Neurochem. 32, 1531–
1537 (1979).
130. Castillo, M., Martínez-Cayuela, M., Zafra, M. F. & García-Peregrín, E. Effect of phenylalanine
derivatives on the main regulatory enzymes of hepatic cholesterogenesis. Mol. Cell. Biochem. (1991)
doi:10.1007/BF00230371.
107
131. Shefer, S. et al. Is there a relationship between 3-hydroxy-3-methylglutaryl coenzyme a reductase
activity and forebrain pathology in the PKU mouse? J. Neurosci. Res. 61, 549–563 (2000).
132. Zhang, F., Snead, C. M. & Catravas, J. D. Hsp90 regulates O-linked β-N-acetylglucosamine
transferase: a novel mechanism of modulation of protein O-linked β-N-acetylglucosamine modification
in endothelial cells. Am. J. Physiol. - Cell Physiol. 302, C1786–C1796 (2012).
133. Zhang, T. et al. Polyamine pathway activity promotes cysteine essentiality in cancer cells. Nat.
Metab. 1–15 (2020) doi:10.1038/s42255-020-0253-2.
134. Koay, D. C., Zerillo, C., Narayan, M., Harris, L. N. & DiGiovanna, M. P. Anti-tumor effects of
retinoids combined with trastuzumab or tamoxifen in breast cancer cells: induction of apoptosis by
retinoid/trastuzumab combinations. Breast Cancer Res. 12, R62 (2010).
135. Fruman, D. A. et al. The PI3K Pathway in Human Disease. Cell 170, 605–635 (2017).
136. Parmenter, T. J. et al. Response of BRAF-Mutant Melanoma to BRAF Inhibition Is Mediated by a
Network of Transcriptional Regulators of Glycolysis. Cancer Discov. 4, 423–433 (2014).
137. Locasale, J. W. Serine, glycine and one-carbon units: cancer metabolism in full circle. Nat. Rev.
Cancer 13, 572–583 (2013).
138. Han, K. et al. CRISPR screens in cancer spheroids identify 3D growth-specific vulnerabilities.
Nature 1–6 (2020) doi:10.1038/s41586-020-2099-x.
139. Cantor, J. R. et al. Physiologic Medium Rewires Cellular Metabolism and Reveals Uric Acid as an
Endogenous Inhibitor of UMP Synthase. Cell 169, 258-272.e17 (2017).
140. Liu, Y., Beyer, A. & Aebersold, R. On the Dependency of Cellular Protein Levels on mRNA
Abundance. Cell 165, 535–550 (2016).
Abstract (if available)
Abstract
Cancer cells exhibit an altered metabolic state compared to normal cells. In fact, some of the most successful chemotherapeutics target aberrant tumor metabolism. Metabolic vulnerabilities in cancer arise from dysregulated signaling and metabolic pathways, which can cause increased reliance on certain substrates for survival. Here, we took systems biology approaches to identify metabolic vulnerabilities in cancer cells. Using liquid chromatography-mass spectrometry (LC-MS) based metabolomics, we identified a co-dependence on glucose metabolism and L-cystine metabolism. Building upon that, we defined a synthetic lethal drug combination to mimic glucose deprivation-induced cell death in the presence of glucose. Next, we sought to quantify metabolic crosstalk by developing a bioinformatic method Differential Gene Set Enrichment Analysis that quantifies the relative enrichment of two metabolic pathways. Lastly, we defined a computational framework to identify context-specific dependence on metabolic pathways. By integrating gene expression, CRISPR screens, and pharmacologic screen data, we uncovered novel metabolic crosstalk and co-dependencies between metabolic pathway activity and metabolic pathway essentiality. The results and frameworks here serve as a foundation for the systems biology-driven discovery of metabolic vulnerabilities and therapeutic targeting of metabolism in cancer cells.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Metabolomic and proteomic approaches to understanding senescence and aging in mammary epithelial cells
PDF
System biology approaches to cancer and diabetes metabolism
PDF
Understanding the impact of cell-to-cell variability on intracellular signaling in CAR cells through mathematical modeling
PDF
Genome-scale modeling of macrophage activity in the colorectal cancer microenvironment
PDF
Data-driven learning for dynamical systems in biology
PDF
Investigating hematopoietic stem cells via multiscale modeling, single-cell genomics, and gene regulatory network inference
PDF
Engineering therapeutics for the improved antitumor efficacy of chimeric antigen receptor T cell therapy
PDF
Genome-scale screening in mammalian cells using CRISPR-Cas9 system
PDF
Development of bioorthogonal probes to reveal proteome modifications
PDF
Non-invasive live-cell imaging for monitoring and evaluating pancreatic islet and beta cell metabolism
PDF
Mechanistic model of chimeric antigen receptor T cell activation
PDF
Dissecting the heterogeneity of mouse hematopoietic stem cells in vivo
PDF
Deciphering protein-nucleic acid interactions with artificial intelligence
PDF
Dissecting metabolic changes in muscle stem cells during activation
PDF
Understanding anti-angiogenic signaling and treatment for cancer through mechanistic modeling
PDF
Associations between perfluoroalkyl substances exposure and metabolic pathways in youth
PDF
Insights from a computational approach to natural killer cell activation
PDF
Using chemistry to reveal the consequences of post translational modifications in cancer
PDF
Identification and analysis of shared epigenetic changes in extraembryonic development and tumorigenesis
PDF
Using chemical biology approaches to investigate the consequences of protein concentration and activity in cancer cells
Asset Metadata
Creator
Joly, James Henry
(author)
Core Title
Systems approaches to understanding metabolic vulnerabilities in cancer cells
School
Viterbi School of Engineering
Degree
Doctor of Philosophy
Degree Program
Chemical Engineering
Publication Date
11/21/2020
Defense Date
10/08/2020
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
bioinformatics,cancer,Computational Biology,mass spectrometry,metabolism,OAI-PMH Harvest,systems biology
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Graham, Nicholas (
committee chair
), MacLean, Adam (
committee member
), Pratt, Matthew (
committee member
), Roberts, Richard (
committee member
)
Creator Email
jamesjoly2@gmail.com,joly@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-398107
Unique identifier
UC11666145
Identifier
etd-JolyJamesH-9137.pdf (filename),usctheses-c89-398107 (legacy record id)
Legacy Identifier
etd-JolyJamesH-9137.pdf
Dmrecord
398107
Document Type
Dissertation
Rights
Joly, James Henry
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
bioinformatics
mass spectrometry
metabolism
systems biology