2024.02.18.580900v1.full-html.html

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

ABSTRACT

Large-scale metabolomic analyses of pan-cancer cell line panels have provided significant

insights into the relationships between metabolism and cancer cell biology. Here, we took a

pathway-centric approach by transforming targeted metabolomic data into ratios to study

associations between reactant and product metabolites in a panel of cancer and non-cancer cell

lines. We identified five clusters of cells from various tissue origins. Of these, cells in Cluster

4 had high ratios of TCA cycle metabolites relative to pyruvate, produced more lactate yet

consumed less glucose and glutamine, and greater OXPHOS activity compared to Cluster 3

cells with low TCA cycle metabolite ratios. This was due to more glutamine cataplerotic efflux

and not glycolysis in cells of Cluster 4.

In silico

analyses of loss-of-function and drug

sensitivity screens showed that Cluster 4 cells were more susceptible to gene deletion and drug

targeting of lactate and glutamine metabolism, and OXPHOS than cells in Cluster 3. Our results

highlight the potential of pathway-centric approaches to reveal new aspects of cellular

metabolism from metabolomic data.

Keywords:

metabolomics / metabolic pathways / cell lines / cancer / glutamine metabolism /

glucose metabolism

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

INTRODUCTION

Cancer cells are characterized by dynamic plasticity of nutrient utilization that supports tumor

growth and survival (Altea-Manzano

et al

, 2020; Fendt

et al

, 2020; Pavlova

et al

, 2022). Our

understanding of the role of metabolism in cancer cell biology has predominantly arisen from

studies taking cancer type-specific and/or metabolic pathway-focused approaches (for

example, Hensley

et al

, 2016, Kamphorst

et al

, 2015, and Wang

et al

, 2023).

More recently, several publications have reported the outcomes of high-throughput

metabolomics using pan-cancer cell line panels, such as the NCI-60 (60 cell lines, Jain

et al

2012; Ortmayr

et al

, 2019) and CCLE panels (928 cell lines, Li

et al

, 2019), CAMP (988 tissue

samples, Benedetti

et al

, 2023), and other panels containing 180 cell lines (Cherkaoui

et al

2022), and 173 cell lines (Shorthouse

et al

, 2022). These studies primarily aimed to identify

links between cancer cell metabolic phenotypes and transcriptional regulation (Benedetti

et al.

2023; Ortmayr

et al.

, 2019), or genetic alterations and dependencies (Li

et al.

, 2019; Mullen &

Singh, 2023) that were associated with drug-sensitivities (Shorthouse

et al.

, 2022). Notably,

Cherkaoui and colleagues (2022) took a top-down approach by clustering the metabolome

acquired by untargeted metabolomics across 49 KEGG metabolic pathways of 180 cancer cells.

Pathway activity was determined for each cell line, and they identified only two clusters that

were defined by either high carbohydrate metabolic activity or high aerobic mitochondrial

activity, that was associated with epithelial or mesenchymal status, respectively (Cherkaoui

al.

, 2022).

Here, we took a different approach to identify common signatures based upon high-flux

metabolic pathways only in a smaller pan-cancer panel of proliferating cells. Based on

metabolite levels, we initially identified four clusters of cells, but this approach failed to

provide insights into pathway differences. To overcome this issue, we introduce a conceptual

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

innovation to transform our metabolite data into pathway-centric ratios that resulted in the

formation of five clusters of cells that displayed different ratios of metabolites of glycolysis,

pentose phosphate pathway, pyruvate-TCA cycle, proline metabolism, serine metabolism,

glutamine metabolism, and methionine metabolism. Of these five clusters, we used a

combination of techniques to show that cells in Cluster 4 had higher ratios of TCA cycle

metabolites when normalized to pyruvate and produced more lactate, despite lower glucose

and glutamine consumption, and greater OXPHOS activity than Cluster 3 with low TCA cycle

metabolite ratios. These differences were, in part, explained by increased glutamine

cataplerotic efflux and glutaminolysis. These phenotypes were supported by

in silico

analyses

of pan-cancer loss-of-function and drug sensitivity screens to show that cells in Cluster 4 were

more susceptible to gene deletion and drug targeting of lactate and glutamine metabolism, and

OXPHOS compared to the cells in Cluster 3. These results highlight the benefit of converting

metabolite levels into pathway-based ratios as a starting point for gaining insights into cellular

metabolic activity.

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

RESULTS

Targeted metabolomic profiling of high flux pathways in cell lines from 11 tissue origins.

We quantified the metabolite levels of high flux pathways, including central carbon (glycolysis,

TCA cycle, pentose phosphate pathway) and amino acid metabolic pathways, in 57 adherent

cell lines (49 tumor- and 8 normal epithelial-derived) from 11 cancer types cultured in basal

media conditions (Appendix Table 1)

Samples were generated in triplicate across 6 batches,

including control cell lines for batch correction, ensuring consistency throughout the complete

data set (Fig 1A). K-means algorithms with Pearson’s correlation of the batch-corrected dataset

identified 4 distinct clusters of cells (Fig 1B), which were not a consequence of the culturing

conditions, tissue type (normal epithelial vs. tumor), cancer type, tissue origin, or the mutation

status of common oncogenic drivers that influence cell metabolism (Cairns

et al

, 2011; Jia

, 2008; Jones & Thompson, 2009; Oermann

et al

, 2012; Vousden & Ryan, 2009) (Fig 1C).

All clusters contained tumor and normal cell lines from different tissue origins (Fig 1B).

However, there were some instances where cell lines from the same tissue origin clustered

together, such as Cluster 3 that was enriched with prostate cancer cells and Cluster 4 with

endometrial cancer cells (Fig 1B), which has been observed in another pan-cancer metabolome

study (Shorthouse

et al.

, 2022). Despite the identification of heterogeneous clusters of cells

based upon the levels of metabolites of high flux pathways, there was no clear organization of

these metabolites into pathways (Fig 1B), likely because strong metabolite interactions are

often localized at the reaction level (Benedetti

et al.

, 2023), that could underpin a testable

hypothesis centered on differences and similarities of pathway activity.

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

Analyses of pathway-centric metabolite ratios uncover distinctive metabolic signatures.

Next, we took a physiological-based approach and evaluated the hypothesis that there were

differences in high-flux metabolic pathways in our panel of cancer and epithelial cells. To

achieve this, we transformed our metabolomic data by calculating the ratios between an

upstream precursor or reactant metabolite (applied as the denominator) and downstream

pathway product metabolites (numerator) for each central carbon and major amino acid

metabolism pathway, similar to Benedetti

et al.

(2023). This pathway-centric transformation

100

was based on the idea that metabolite conversion forms a cascade, and therefore, intrinsic

101

correlations likely exist between reactant and product that provide biologically meaningful

102

insights into pathway activity. For example, glucose is the precursor of glycolysis, and as such,

103

ratios of the abundance of glycolytic metabolites relative to glucose were calculated (Fig 2A).

104

K-means algorithms with Pearson’s correlation of the entire metabolite ratio dataset spanning

105

seven metabolic pathways of interest (glycolysis, pentose phosphate pathway, pyruvate-TCA

106

cycle, proline metabolism, serine metabolism, glutamine metabolism, and methionine

107

metabolism) identified 5 distinct clusters of cells (Appendix Fig S1A). These clusters differed

108

from what was identified using metabolite abundances alone (Fig 1) and were composed of

109

cells from different tissue origins, tissue types, cancer types, mutation status of common

110

oncogenic drivers, and culturing conditions (Appendix Fig S1B).

111

To assist in interpreting the patterns in the data, the primary heatmap (Appendix Fig S1A) was

112

separated into individual panels with the cell clusters conserved (Fig 2). The subset of cells in

113

Cluster 3 had greater glycolysis (Fig 2A) and pentose phosphate pathway metabolite ratios

114

(PPP; Fig 2B) but lower TCA cycle (relative to pyruvate; Fig 2C) and proline metabolism ratios

115

(Fig 2D) compared to Cluster 2. We identified differences in serine metabolism between cells

116

in Clusters 4 and 5 (Fig 2E) and less striking differences in glutamine (Fig 2F) and methionine

117

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

Figure 2: Pathway-centric metabolite ratios identifies clusters of cells.

123

A-G.

Metabolite ratio heatmaps formed by normalization of (A) glycolysis pathway and (B) pentose

124

phosphate pathway metabolites to intracellular glucose, (C) TCA cycle pathway metabolites to

125

pyruvate, (D) proline metabolism metabolites to proline, (E) serine metabolism metabolites to

126

serine, (F) glutamine metabolism metabolites to glutamine, and (G) methionine metabolism to

127

methionine. Heatmaps are slices from Appendix Figure S1.

128

Differences in TCA cycle metabolite to pyruvate ratios are due to glutamine and not

129

glucose metabolism.

130

We next addressed a significant limitation of metabolomic data. To date, irrespective of how

131

the metabolomics data is analyzed, we often cannot infer function/flux. Here, we use our

132

pathway-centric analyses to validate that the clusters formed (from where??) from our

133

pathway-centric analyses exhibited functional differences. Since the TCA cycle is an essential

134

hub where various pathways converge and is critical for energy and biomass production and

135

cell viability (Spinelli & Haigis, 2018) (Fig 3A), we chose to contrast Cluster 3 and Cluster 4

136

from the identified five clusters formed in Figure 2, as they had distinctive TCA cycle

137

metabolite levels relative to pyruvate (Fig 3B). Firstly, Cluster 4 cells had greater TCA cycle

138

metabolite levels relative to pyruvate (Multiple unpaired t-tests, P-value<0.001), except for

139

oxoglutarate and NADPH (Fig 3C). Differences in succinate and malate levels relative to

140

pyruvate were also evident at the cell line level (Fig 3D).

141

As pyruvate was used as the denominator to calculate TCA cycle metabolite ratios, we extended

142

our analysis to lactate since pyruvate is converted to lactate by lactate dehydrogenase. Cells in

143

Cluster 4 had a greater lactate-to-pyruvate ratio, a measure of the equilibrium constant of lactate

144

dehydrogenase, than cells in Cluster 3 (Student t-test, P=0.001; Fig 3E). There was no

145

difference in the NADH/NAD+ ratio (Appendix Fig S2A), which are co-factors of lactate

146

dehydrogenase and can influence its activity (Luengo

et al

, 2021). Thus, the greater lactate to

147

pyruvate ratio in Cluster 4 cells, compared to Cluster 3 cells, was unlikely driven by excess

148

NADH relative to NAD+, but possibly due to carbon surplus in the TCA cycle.

149

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

150

Figure 3: Metabolite ratio-based signatures identifies differences in glutamine

151

metabolism between clusters.

152

Schematic of glucose and glutamine sources to the TCA cycle.

153

Scaled metabolite ratio heatmap of Clusters 4 and 3, derived from TCA cycle pathway metabolites

154

calculated using the precursor pyruvate as denominator.

155

Abundance ratios of TCA cycle metabolites to pyruvate (C4 n=10, C3 n=19, multiple unpaired t-

156

tests, *p<0.05, error min to max values).

157

Waterfall plots comparing the scaled metabolite abundance (log 10) of succinate and pyruvate (left),

158

and malate and pyruvate (right) for cells of Clusters 3 and 4.

159

Abundance ratio of lactate to pyruvate (error min to max values) (C4 n=10, C3 n=19).

160

Lactate production intensity and glucose and glutamine consumption (consumed % of starting

161

substrate concentrations) after 24 hours (C4 n=8, C3 n-6, error min to max values).

162

Oxygen consumption rates (OCR) at 48 hours, normalized to percent confluency (C4 n=6, C3 n-5,

163

3 biological replicates and 4 technical replicates per cell line, mean ± standard error of the mean).

164

Fractional contribution of [U-

C]-glucose to intracellular lactate (C4 n=3, C3 n=4, 3 biological

165

replicates and 3 technical replicates per cell line, mean ± standard error of the mean).

166

Fractional contribution of [U-

C]-glutamine to intracellular lactate (C4 n=3, C3 n=4, 3 biological

167

replicates and 3 technical replicates per cell line, mean ± standard error of the mean).

168

Fractional contribution of [U-

C]-glutamine to m+3 fructose 1,6-bisphosphate (FBP), and m+3

169

phosphoenolpyruvate (PEP) (C4 n=3, C3 n=4, 3 biological replicates and 3 technical replicates per

170

cell line, mean ± standard error of the mean).

171

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

Schematic of the C4 phenotype (increased glutamine to the TCA cycle, gluconeogenesis, and

172

aerobic glycolysis) compared to C3 (increased glucose and glutamine consumption and glucose

173

oxidation).

174

Data information: (C-G) *P<0.05 vs Cluster 4 by Unpaired student’s t-test.

175

Glycolysis and glutaminolysis both feed the TCA cycle and can produce lactate as a by-product

176

(Smith

et al

, 2016). Next, we sought to determine whether the increased lactate production

177

relative to pyruvate in the cells of Cluster 4 compared to Cluster 3 cells was due to greater

178

consumption of glucose and/or glutamine. As expected, cells in Cluster 4 produced more lactate

179

compared to those in Cluster 3 (Student t-test, P=0.006; Fig 3F), with some cell line-specific

180

differences observed (Appendix Fig S2B). Likewise, there were cell line-specific differences

181

in glucose and glutamine consumption over 24 hours (Appendix Fig S2B), but somewhat

182

surprisingly, the net consumption of glucose (Student t-test, P-value=0.02) and glutamine

183

(Student t-test, P=0.04) were lower in Cluster 4 cells (Fig 3F). In isolation, the higher lactate-

184

to-glucose yield seen in Cluster 4 could be interpreted as higher aerobic glycolysis, whereas in

185

Cluster 3 cells, proportionally more glucose was oxidized instead of being converted to lactate.

186

We postulated that Cluster 4 cells are more oxidatively competent and thus have surplus carbon

187

that spills into lactate. To test this, we quantified the oxygen consumption rate as a readout of

188

oxidative phosphorylation (OXPHOS) activity and, thus, TCA cycle fluxes. In line with our

189

prediction, cells in Cluster 4 possessed greater OXPHOS activity than Cluster 3 cells (Student

190

t-test, P=0.04; Fig 3G, Appendix Fig S3C). The more efficient respiration may correlate with

191

our observation of more abundant TCA cycle metabolites relative to pyruvate (Fig 3C) and

192

suggests that the increase in lactate production (Fig 3F) is a consequence of cells using pyruvate

193

as a redox sink to regenerate NAD

194

To further resolve the intersection of glucose and glutamine metabolism at pyruvate, which

195

underpins the formation of Clusters 3 and 4, we performed [U-

C]-glucose and [U-

C]-

196

glutamine tracing experiments in a subset of cells selected from Clusters 3 and 4. As expected,

197

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

lactate was produced mainly from glucose (78-92% enrichment). However, there were no

198

differences in lactate and pyruvate enrichment between Clusters 3 and 4 (Fig 3H, Appendix

199

Fig S2D), nor were there differences in the enrichment of TCA cycle metabolites from glucose

200

(Appendix Fig S2E). These data, therefore, eliminate the role of glucose metabolism in

201

explaining the higher TCA metabolite to pyruvate ratios in Cluster 4 cells compared to Cluster

202

203

As such, we turned our attention to quantifying glutamine metabolism, as it is a major TCA

204

cycle carbon source (Spinelli & Haigis, 2018), using [U-

C]-glutamine tracing. Our first

205

observation was that more glutamine carbons were incorporated in lactate, via either

206

phosphoenolpyruvate carboxykinase or malic enzymes (Mansouri

et al

, 2017; Montal

et al

207

2015), in Cluster 4 cells than in Cluster 3 (Fig 3I). This increased efflux of glutamine from the

208

TCA cycle was also observed as greater enrichment of glutamine carbons into m+3 fructose

209

1,6-bisphosphate, and, to a lesser extent, phosphoenolpyruvate (Fig 3J), intermediates of

210

gluconeogenesis. Combined, these data show that cells in Cluster 4 possessed a more oxidative

211

phenotype that compensates for increased aerobic glycolysis with glutamine cataplerosis and

212

explains the increased lactate production and more abundant TCA cycle metabolites in cells of

213

Cluster 4, compared to Cluster 3 that had greater glucose and glutamine consumption and

214

glucose oxidation (Fig 3K). Furthermore, these new insights into glucose and glutamine

215

metabolism and the discovery that some cells produce more lactate despite lower glucose

216

consumption support the new insights into high flux metabolic pathways from our pathway-

217

centric ratio-based analyses.

218

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

Differences in TCA cycle, lactate, and glutamine metabolism between Clusters 3 and 4

219

correlate with sensitivity to loss-of function.

220

We further validated the outcomes of our pathway-centric metabolite ratio analysis of targeted

221

metabolomic data (Fig 2) and functional studies (Fig 3) using

in silico

assessment of publicly

222

available datasets. Specifically, we used the pan-cancer DEMETER (Tsherniak

et al

, 2017) and

223

Project Score (Behan

et al

, 2019) loss-of-function screens, and the drug sensitivity PRISM

224

(Corsello

et al

, 2020) and GDSC2 (Yang

et al

, 2012) datasets and selected for gene or drug

225

targets within all KEGG metabolic pathways (Fig 4A). We then cross-referenced these large-

226

scale cell line panels with cells that were members of Clusters 3 and 4 and focused our analyses

227

to test the hypothesis that cells in Cluster 4 were more susceptible to depletion of genes

228

associated with OXPHOS, glutamine and lactate metabolism compared to Cluster 3 (Fig 4B).

229

In line with our hypothesis, Cluster 4 cells had greater sensitivity to genetic knockout of

230

OXPHOS complexes I-V and glutaminase isoforms (Multiple unpaired t-tests, P<0.05; Fig

231

4C). We also observed a trend for greater sensitivity to deletion of succinate dehydrogenase

232

(SDHD; Student t-test, P=0.06) and lactate dehydrogenase (LDHD; Student t-test, P=0.08) in

233

cells in Cluster 4 (Appendix Fig S3A). We complemented our targeted assessment of these

234

datasets by determining the top 20 metabolic targets from all KEGG pathways that possessed

235

the greatest difference between Clusters 4 and 3. From this unbiased approach, we identified

236

members of OXPHOS, glutamine metabolism, TCA cycle, and pyruvate/lactate pathways in

237

this list (highlighted in yellow in Fig 4D, E). We also determined the top 20 list of the most

238

different targets between Cluster 4 and 3 when we narrowed the coverage to just central carbon

239

metabolism from all KEGG pathways (Appendix Fig S3B). Consistent with our other

240

observations, the list of most sensitive targets again was enriched with enzymes from OXPHOS

241

and TCA cycle pathways (Appendix Fig S3B). Finally, we developed a scoring system to

242

consolidate the top 20 most sensitive pathways in Cluster 4 from all databases, and again found

243

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

that the TCA cycle, OXPHOS, and glutamine metabolism were most vulnerable to genetic and

244

drug targeting compared to Cluster 3 (Appendix Fig S3C). Combined, we have demonstrated

245

that analyzing metabolomic data with a pathway-centric basis by using ratios identifies

246

distinctive metabolic profiles that are evident in functional measures and loss-of-function

247

screens.

248

249

Figure 4: Validating pathway-centric metabolite ratio clusters and substrate

250

dependencies using loss-of-function screens and drug sensitivity databases.

251

Schematic of

in silico

analysis of loss-of-function and drug sensitivity screens against KEGG

252

metabolic pathways and identified metabolic ratio phenotypes.

253

Schematic illustrating the C4 phenotype of greater utilization of glucose and glutamine identified

254

by cluster analyses using metabolite ratios. Genes named in red were hypothesized to have greater

255

sensitivity to gene knockouts or inhibition in C4 compared to C3.

256

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

DISCUSSION

267

Cell metabolism is dynamic and differs depending on context. From the early observations of

268

Warburg (Warburg, 1925; Warburg & Minami, 1923) and the Coris (Cori & Cori, 1925), it is

269

now well accepted that cancer cells metabolize glucose differently from non-tumor tissue.

270

These differences are not a consequence of rewiring or transformation of the cascades of

271

biochemical reactions that form glycolysis and PPP but essentially from increased uptake of

272

extracellular glucose driving increased flux and production of end products, including lactate

273

(DeBerardinis & Chandel, 2020). Alongside these well-documented changes in glucose

274

metabolism, there have been significant advances in the understanding of the interaction

275

between metabolic pathways (Sung

et al

, 2023), outside the well-established convergence of

276

glucose, glutamine and fatty acid metabolism in the TCA cycle. A major challenge remains

277

how to infer mechanistic differences in metabolism based on metabolomics data alone, since

278

the latter may not correlate with pathway activity nor flux. Conventional dimensional

279

reduction, clustering, and over-presentation methodologies rely on coordinated changes to infer

280

co-dependency or co-regulation (Amara

et al

, 2022; Huang & Wang, 2022), but the

281

effectiveness of subsequent interpretations may be hinge on how the literature has delineated

282

metabolite memberships (Mahajan

et al

, 2024), which are continuously honed over time.

283

Our first round of clustering, which was based on metabolite abundance of proliferating cells,

284

formed groups that contained both epithelial and tumor-derived, and dismissed culturing

285

conditions, tissue type and origin, and mutation status of oncogenic drivers as potential factors

286

influencing cell metabolism. However, we failed to derive any metabolic signatures or

287

hypotheses that could be evaluated with functional assessment. Consequently, we took a

288

physiologically-based approach and transformed our metabolomic data into pathway-centric

289

ratios and explored the relationships between reactant and product metabolites of central

290

carbon and amino acid pathways. The approach draws upon thermodynamic principles in terms

291

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

of reaction equilibrium and Gibbs energy (Park

et al

, 2016), and in practice, there is robust

292

evidence for coordinated changes among proximal and hub metabolites (Martínez-Reyes &

293

Chandel, 2020). Using metabolite ratios, we identified five clusters of cell lines, two of which

294

(Cluster 3 and 4) we surmised to differ in TCA cycle activity based on the levels of TCA cycle

295

metabolites relative to pyruvate. Indeed, the ensuing functional assay of glucose and glutamine

296

metabolism verified pathway activity differences between Cluster 3 and 4, with Cluster 4’s

297

elevated TCA cycle metabolites showing concordance with higher lactate production and the

298

shift from glycolysis to an increased glutamine oxidation and cataplerosis. This work highlights

299

the value of pathway-centric ratio-based data transformation in distinguishing metabolic

300

pathway activity in a pan-cancer cell line panel and, therefore, supports the concept of a

301

physiological-based approach to analyze metabolomic data.

302

Many tumors are reliant on glutamine as a critical carbon and nitrogen source, with glutaminase

303

inhibition shown to be an effective therapy in several cancer types, such as triple-negative

304

breast cancer (Gross

et al

, 2014), non-small cell lung cancer (van den Heuvel

et al

, 2012), and

305

head and neck cancer (Wicker

et al

, 2021). Our findings showed that Cluster 4 cells had higher

306

OXPHOS levels yet consumed less glucose and glutamine; these cells appear more carbon

307

efficient. The relatively more plentiful TCA cycle metabolites may sustain higher TCA cycle

308

fluxes, which are tightly coupled to OXPHOS, and buffer critical anabolic and signaling

309

functions (Martínez-Reyes & Chandel, 2020). Glutamine’s proximity means oxidizing

310

glutamine repletes TCA cycle metabolites more directly than glucose (Quek

et al

, 2022), but

311

the displacement of glucose oxidation further entrenched the aerobic glycolysis phenotype as

312

seen in Cluster 4 cells. Additionally, glutamine cataplerosis and the contribution of PEP

313

carboxykinase to gluconeogenesis, converting oxaloacetate to PEP, augment the supply of

314

biomass precursors, which have been documented in several cancer types, including liver (Liu

315

et al

, 2018) and lung (Vincent

et al

, 2015). For example, glutamine-derived lactate production

316

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

via glutaminolysis in glioblastoma cells helps produce NADPH and support fatty acid synthesis

317

(DeBerardinis

et al

, 2007). Overall, we speculate that the increased glutamine utilization

318

among Cluster 4 cells may confer greater fitness to support proliferation and survival.

319

Excitingly, our assertion of elevated OXPHOS and enhanced glutamine utilization in Cluster 4

320

matched data mining results derived from drug sensitivity (Corsello

et al.

, 2020; Yang

et al.

321

2012) and loss-of-function screens (Behan

et al.

, 2019; Tsherniak

et al.

, 2017). Among the top

322

20 loss-of-function screens, Cluster 4 cells were most sensitive to targeting the TCA cycle,

323

OXPHOS, glycolysis, and glutamate metabolism pathways, compared to Cluster 3 cells.

324

Namely, we found Cluster 4 cells significantly more sensitive to glutaminase gene knockout

325

and inhibitors. These findings highlight the potential of a physiological pathway-centric

326

approach to translating metabolite signatures into effective strategies for identifying druggable

327

vulnerabilities in glucose and glutamine metabolism.

328

A key limitation of our approach is coverage. Our targeted LC-MS method covers central

329

carbon metabolism, and thus we have focused on the TCA cycle as it is a convergent point for

330

glucose and glutamine utilization; however, whether these relationships are consistent for other

331

TCA cycle substrates, such as branched-chain amino acids and fatty acids (Neinast

et al

, 2019;

332

Schoors

et al

, 2015) remains to be determined. Another major lesson from our approach was

333

that the most insightful outcome of our clustering analysis came from using a hub metabolite

334

(e.g., pyruvate) as the denominator rather than a starting substrate (i.e., glucose, glutamine,

335

serine). Since we only included one hub metabolite in our pathway cluster analysis, it is

336

conceivable that investigating other metabolic hubs not covered in our targeted approach, such

337

as NAD

(Benedetti

et al.

, 2023), could identify other distinctive signatures and targetable

338

vulnerabilities. Perhaps the abundance of hub metabolites, where multiple pathways converge,

339

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

MATERIALS AND METHODS

351

Reagents and Tools Table

352

Reagent/Resource Reference or Source

Identifier or Catalog Number

Experimental

Models

MCF10A

ATCC

ATCC CRL-10317 RRID: CVCL_0598

MDA-MB-231

ATCC

ATCC HTB-26 RRID: CVCL_0062

BT474

ATCC

ATCC HTB-20 RRID: CVCL_0179

MDA-MB-468

ATCC

ATCC HTB-132

MCF7

ATCC

ATCC HTB-22

BT20

ATCC

ATCC HTB-19

MDA-MB-134

ATCC

ATCC HTB-23

PNT1

ECACC

ECACC 95012614 RRID: CVCL_2163

PNT2

Prof. Lisa Butler, SAHMRI

(Nassar

et al

, 2020)

C4-2B

ATCC

ATCC CRL-3315

PC-3

ATCC

ATCC CRL-1435

22Rv1

ATCC

ATCC CRL-2505 RRID: CVCL_1045

LNCaP

ATCC

ATCC CRL-1740 RRID: CVCL_1379

DU145

A/Prof. Luke Selth, Flinders University

ATCC HTB-81

CWR-AD1

A/Prof. Luke Selth, Flinders University

(Li

et al

, 2013; Nyquist

et al

, 2013)

CWR-D567

A/Prof. Luke Selth, Flinders University

(Nyquist

et al.

, 2013)

VCAP

Prof. Lisa Butler, SAHMRI

(Nassar

et al.

, 2020)

V16D

Prof. Lisa Butler, SAHMRI

(Nassar

et al.

, 2020)

MR49F

Prof. Lisa Butler, SAHMRI

(Bishop

et al

, 2017)

MR42D

Prof. Lisa Butler, SAHMRI

(Bishop

et al.

, 2017)

AML12

Prof. Rob Parton, University of

Queensland

(Nagarajan

et al

, 2019)

PH5CH8

A/Prof. Susan McLennan, University of

Sydney

(Nagarajan

et al.

, 2019)

HEPG2

A/Prof. Susan McLennan, University of

Sydney

(Nagarajan

et al.

, 2019)

HUH7

Prof. Mark Gorrell, University of Sydney

RRID: CVCL_0336

SKHEP1

Prof. Mark Gorrell, University of Sydney

RRID: CVCL_0525

HEPA1-6

Prof. Mark Gorrell, University of Sydney

RRID: CVCL_0327

HUE-T

Dr Frances Byrne, UNSW

(Byrne

et al

, 2014)

MAD11

Dr Frances Byrne, UNSW

(Byrne

et al.

, 2014)

Ishikawa

Prof. Kyle Hoehn, UNSW

RRID: CVCL_2529

MFE296

Prof. Kyle Hoehn, UNSW

RRID: CVCL_1406

MFE319

Prof. Kyle Hoehn, UNSW

RRID: CVCL_2112

AN3CA

Prof. Kyle Hoehn, UNSW

RRID: CVCL_0028

RL952

Prof. Kyle Hoehn, UNSW

RRID: CVCL_0505

KLE

Prof. Kyle Hoehn, UNSW

RRID: CVCL_1329

HEC1A

Prof. Kyle Hoehn, UNSW

RRID: CVCL_0293

U251

Prof. Lenka Munoz, University of

Sydney

RRID: CVCL_0021

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

https://www.graphpad.com/

bioRxiv preprint

U87

Prof. Lenka Munoz, University of

Sydney

RRID: CVCL_3429

HPDE

A/Prof Thomas Grewal, University of

Sydney

RRID: CVCL_4376

PANC1

A/Prof Thomas Grewal, University of

Sydney

RRID: CVCL_0480

MIAPACA-2

A/Prof Thomas Grewal, University of

Sydney

RRID: CVCL_0428

BXPC3

A/Prof Thomas Grewal, University of

Sydney

RRID: CVCL_0186

ASPC1

A/Prof Thomas Grewal, University of

Sydney

RRID: CVCL_0152

A2780

A/Prof. David Croucher, Garvan Institute

RRID: CVCL_0134

OVCAR3

A/Prof. David Croucher, Garvan Institute

RRID: CVCL_0465

SKOV3

A/Prof. David Croucher, Garvan Institute

RRID: CVCL_0532

A375

Dr. Lorey Smith, Peter MacCallum

Cancer Centre

ATCC CRL-1619 RRID: CVCL_0132

HT144

Dr. Lorey Smith, Peter MacCallum

Cancer Centre

ATCC HTB-63 RRID: CVCL_0318

WM266.4

Dr. Lorey Smith, Peter MacCallum

Cancer Centre

ATCC CRL-1676 RRID: CVCL_2765

DO4-M1

Dr. Lorey Smith, Peter MacCallum

Cancer Centre

(Parmenter

et al

, 2014)

A549

Dr. Lorey Smith, Peter MacCallum

Cancer Centre

(Chen

et al

, 2019)

NCI-H226

Dr. Lorey Smith, Peter MacCallum

Cancer Centre

RRID: CVCL_1544

Detroit562

A/Prof Thomas Grewal, University of

Sydney

RRID: CVCL_1171

SCC4

A/Prof Thomas Grewal, University of

Sydney

RRID: CVCL_1684

HT29

A/Prof. Kellie Charles, University of

Sydney

RRID: CVCL_A8EZ

HCT116

A/Prof. Kellie Charles, University of

Sydney

RRID: CVCL_0291

SW-48

A/Prof. Kellie Charles, University of

Sydney

RRID: CVCL_1724

SW-1417

A/Prof. Kellie Charles, University of

Sydney

RRID: CVCL_1717

Chemicals,

Enzymes, and

other reagents

Enzalutamide

MDV3100

Selleckchem

Cat#S1250

D-Glucose

Sigma-Aldrich

Cat#273053

L-Glutamine

Gibco

Cat#25030081

Fatty acid free BSA

Bovogen

Cat#BSAS0.10

D-Glucose (U-13C6,

99%)

Cambridge Isotope Laboratories, Inc.

Cat#CLM-1396

L-Glutamine (U-

13C5, 99%)

Cambridge Isotope Laboratories, Inc.

Cat#CLM-1822-H

Software

GraphPad Prism V9

GraphPad Software

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

https://au.mathworks.com/products/matlab.html

bioRxiv preprint

MATLAB

Mathworks

MSConvert

N/A

(Chambers

et al

, 2012)

MSDial

Riken

http://prime.psc.riken.jp/compms/msdial/main.html

Biorender

https://biorender.com/

Other

Resipher

Lucid Scientific

https://lucidsci.com/

Incucyte-SX5

Sartorius

https://www.sartorius.com/

Cell lines and culture conditions

353

A total of 57 adherent cell lines (8 normal and 49 tumor) spanning 11 cancer types were used

354

in this study. Cell lines obtained from vendors or that were kindly provided by academic labs,

355

overlapped with major cell line resources such as the Cancer Cell Line Encyclopedia (CCLE)

356

and National Cancer Institute 60 panel (NCI60) (Appendix table 1). Mutational status of

357

common oncogenes was characterized by cross-referencing the Cellosaurus database,

358

COSMIC database, and Depmap portal. Culturing media were supplemented with 10% fetal

359

bovine serum (Cytiva Hyclone) and 1% penicillin/streptomycin (Gibco) unless stated

360

otherwise (Appendix table 1). Cells were incubated at 37°C and 5% CO

. The full panel of

361

cells were generated across 6 batches over a period of 3 years due to COVID restrictions and

362

included overlapping cell lines to correct for batch effects.

363

Metabolomics experiments

364

Cells were seeded in triplicate in 6-well plates at a density of 5x10

cells/well in 2 mL of media.

365

After 24 hours, the media was removed, and cells were washed once with 2 mL ice-cold 0.9%

366

w/v NaCl. Cells were then scraped with 300 mL of extraction buffer, EB, (1:1 LC/MS

367

methanol:water (Optima) + 0.1x internal standards comprised of non-endogenous polar

368

metabolites (2-morpholinoethanesulfonic acid, D-camphor-10-sulfonic acid, and deuterated

369

thymidine) and transferred to a 1.5 mL microcentrifuge tube. A further 300 mL of EB was

370

added to the well and combined in the tube. 600 mL Chloroform (Honeywell) was added before

371

vortexing and incubating on ice for 10 minutes. Tubes were vortexed briefly and centrifuged

372

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

at 15,000 x

for 10 minutes at 4°C. The aqueous layer was collected and dried without heat,

373

using a Savant SpeedVac (Thermo Fisher). Dried samples were resuspended in 40 mL Amide

374

buffer A (20 mM ammonium acetate, 20 mM ammonium hydroxide, 95:5 HPLC H

375

Acetonitrile (v/v)) and vortexed and centrifuged at 15,000 x

for 5 minutes at 4°C. 20 mL of

376

supernatant was transferred to HPLC vials containing 20 mL acetonitrile for LC-MS analysis

377

of amino acids and glutamine metabolites. The remaining 20 mL of resuspended sample was

378

transferred to HPLC vials containing 20 mL LC-MS H

O for LCMS analysis of glycolytic,

379

pentose phosphate pathway, and TCA cycle metabolites. Amino acids and glutamine

380

metabolites were measured using the Vanquish-TSQ Altis (Thermo) LC-MS/MS system.

381

Analyte separation was achieved using a Poroshell 120 HILIC-Z Column (2.1x150 mm, 2.7

382

mm) (Agilent) at ambient temperature. The pair of buffers used were Amide buffer A and 100%

383

acetonitrile (Buffer B), flowed at 200 mL/min; injection volume of 5 mL. Glycolytic, PPP and

384

TCA cycle metabolites were measured using 1260 Infinity (Agilent)-QTRAP6500+ (AB

385

Sciex) LC-MS/MS system. Analyte separation was achieved using a Synergi 2.5 mm Hydro-

386

RP 100A LC Column (100x2 mm) at ambient temperature. The pair of buffers used were 95:5

387

(v/v) water:acetonitrile containing 10 mM tributylamine and 15 mM acetic acid (Buffer A) and

388

100% acetonitrile (Buffer B), flowed at 200 mL/min; injection volume of 5 mL. Raw data from

389

both LC-MS/MS systems were extracted using MSConvert (Chambers

et al.

, 2012) and in-

390

house MATLAB scripts. Concentrations of metabolites were calculated against a standard

391

curve of polar and amino acid metabolite standards similarly extracted as above. Log10

392

normalization was performed on the metabolite concentration data.

393

Metabolomics batch correction

394

Cell line samples were quantified over 6 batches during the experiment including overlapping

395

cell lines across each batch. To eliminate potential batch effects, we applied the normalization

396

method: Removing Unwanted Variation-III (Molania

et al

, 2019). We set k, the number of

397

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

unwanted factors, to 9. The cell lines that were measured across multiple runs were used as

398

replicates for the batch correction algorithm and to assess the quality of the batch correction

399

output.

400

Metabolomics clustering analysis

401

Clustering was performed on the batch-corrected matrix using K-means algorithms with “1 -

402

Pearson correlation” as the distance matrix (Gu

et al

, 2016) implemented in the

403

ComplexHeatmap package (k = 5). We visualized the results as a heatmap using the

404

ComplexHeatmap package. The clustering procedure was repeated 1000 times and the

405

consensus was taken to ensure a more robust clustering result. The clustering strategy was

406

applied to both the original batch-corrected matrix and the ratio-transformed matrix. We

407

visually assess whether the obtained clustering was not influenced by potential confounding

408

factors, such as culturing conditions and common oncogenic gene mutations and tissue type.

409

This is achieved by adding a color-coded legend on culturing conditions and mutational

410

information to the clustering heatmap and demonstrate that no patterns exist.

411

The original batch-corrected matrix was subset to targeted metabolites from several key

412

metabolic pathways including the TCA cycle, pentose phosphate pathway, glycolysis, amino

413

acid metabolism, and glutamine metabolism, followed by clustering analysis. For the ratio-

414

transformed matrix, batch-corrected abundance ratios were calculated between a precursor

415

metabolite for each pathway (used as the denominator) and the remaining pathway metabolites.

416

The corresponding precursor (denominator) metabolites were pyruvate for the TCA cycle,

417

glucose for PPP and glycolysis, and glutamine for the glutamine metabolism pathway. For

418

amino acid pathways, serine, proline, and methionine were used as the precursor metabolites.

419

The clustering structure determined from the processes above were also used to visualize the

420

oncogene mutation status, tissue type, cancer type, tissue origin, and culturing media type of

421

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

the cell lines using a heatmap and inspect the relationship of these variables with the determined

422

clusters.

423

U-

C stable isotope tracing

424

Cells were seeded in triplicate in 6-well plates at a density of 5x10

cells/well in 2 mL of media.

425

After 24 hours wells were washed with warm PBS and media replaced with 600 mL of DMEM

426

no glucose, no glutamine media supplemented with 2% (wt./vol) FA-free BSA, and 5 mM

427

glucose and 1 mM glutamine, replaced with their respective U-

C forms. Cells were incubated

428

in U-

C containing medium for 6 hours. Samples were extracted and measured using the

429

Vanquish-TSQ Altis and Agilent-QTRAP6500+.

430

Extracellular substrate experiments

431

Cells were seeded in triplicate in 6-well plates at a density of 5x10

cells/well in 2 mL of media.

432

After 24 hours wells were washed with warm PBS and media replaced with 1 mL of DMEM

433

no glucose, no glutamine media supplemented with 5 mM glucose, 1 mM glutamine and 150

434

mM palmitate. 100 mL of extracellular media were collected from wells at 3, 6, 12, 24 hour

435

timepoints. Media samples were centrifuged at 16,000 x

for 5 minutes at 4°C and the

436

supernatant collected for subsequent extraction and LC-MS analysis.

437

To extract media samples for LC-MS, 20 mL of supernatant media was first diluted with 80

438

mL water and vortexed, and then 10mL of the diluted media was transferred to 90 mL of

439

extraction buffer containing 1:1 (v/v) acetonitrile and methanol + 1x internal standards (non-

440

endogenous standards) at -30°C. The mixture was centrifuged at 12,000 x

for 5 minutes at

441

C and transferred into HPLC vials for LC-MS analysis measured using the Vanquish-TSQ

442

Altis.

443

Oxygen consumption rate

444

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

Cells were seeded in 96-well plates (Nunc) in 100 mL basal medium. Oxygen consumption

445

rates were continuously measured using Resipher (Lucid Scientific) at 37°C, 10% CO

as per

446

manufacturer instructions over 48 hours, starting 24 hours post-seeding. Media was replenished

447

at 24 hours. Cell-free wells contained 200 mL of PBS to avoid evaporation. To account for

448

differences in cell growth over 48 hours, parallel plates were similarly cultured, and media

449

replenished at 24 hours for cell confluency measured using IncuCyte-SX5 (Sartorius).

450

Analysis of drug and CRISPR databases

451

Drug sensitivity area under the curve (AUC) data were downloaded from the PRISM (Corsello

452

et al.

, 2020) and GDSC2 (Release 8.4, July 2022) (Yang

et al.

, 2012) databases. Loss-of-

453

function fitness score data were downloaded from the DEMETER (Tsherniak

et al.

, 2017) and

454

Project Score (July 2021) (Behan

et al.

, 2019) databases. First, cell lines were filtered by

455

overlapping cell lines within our panel, and then inhibitor or loss-of-function gene targets were

456

filtered by KEGG metabolic pathway genes. We then performed a differential expression

457

analysis on the drug response of the cell lines belonging to clusters of interest, specifically the

458

response of Cluster 3 cell lines versus Cluster 4 cell lines. The top 20 drugs or gene targets with

459

the greatest differential response between Clusters 3 and 4 was identified.

460

Statistical analysis for

in vitro

and database analysis

461

For all

C-tracing, extracellular, and oxygen consumption experiments, at least 3 technical

462

replicates and 3 independent biological replicates were used for each sample group. Descriptive

463

data summary in Figures 3 and 4 were presented as mean

standard error of the mean (SEM),

464

mean

standard deviation (SD), or mean

min to max values, as indicated in each of the figure

465

legends. We determine statistically significant differences between cell clusters 3 and 4 in

C-

466

tracing, extracellular, and oxygen consumption experiments and loss-of-function analysis by

467

performing unpaired Student’s t-test. We assessed the differences between clusters 3 and 4 of

468

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

481

482

Appendix Figure S2: Substrate preferences for cell line clusters identified by TCA cycle

483

ratios.

484

Schematic of the lactate dehydrogenase catalyzed reaction. NAD+ to NADH ratio in cells of

485

Clusters 4 and 3 (Unpaired student t-test, ns p>0.05, error min to max values).

486

Lactate production and glucose and glutamine consumption over 24 hours (C4 n=8, C3 n-6, 3

487

biological replicates and 3 technical replicates per cell line, mean

standard error of the mean).

488

Oxygen consumption rates (OCR) measured over 48 hours, normalized to % confluency. Media

489

changed at 24 hours (left). OCR measurement for cell lines at 48 hours (right) (C4 n=6, C3 n-5, 3

490

biological replicates and 4 technical replicates per cell line).

491

Fractional contribution of [U-

C]-glucose to intracellular pyruvate (C4 n=3, C3 n=4, 3 biological

492

replicates and 3 technical replicates per cell line, mean

standard error of the mean).

493

Fractional contribution of [U-

C]-glucose to TCA metabolites citrate, oxoglutarate, succinate and

494

malate (C4 n=3, C3 n=4, 3 biological replicates and 3 technical replicates per cell line, mean

495

standard error of the mean).

496

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

497

Appendix Figure S3: Loss of function and drug database screen validation of pathway-

498

centric metabolite ratio clusters.

499

Genes related to the glucose and glutamine utilization phenotype of C4 with trends of greater

500

sensitivity to gene knockouts. (Unpaired student t-test, ns p>0.05, mean

standard error of the

501

mean, n=2-6).

502

Top 20 gene knockouts and associated KEGG pathways with the greatest fitness scores between

503

C4 and C3 in DEMETER and Project Score databases from glycolysis, TCA cycle, pyruvate

504

metabolism, glycolysis/gluconeogenesis, and alanine, aspartate, and glutamate metabolism

505

pathways.

506

Metabolic pathways associated with the top 20 drug and gene knockout sensitives for Cluster 4

507

for each database. Drugs and pathways matching C4 vulnerability model (yellow) are highlighted.

508

509

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

Acknowledgements

510

N.T.S. is supported by the Australian Rotary Health/Rotary Club of Blacktown City ‘Mel Grey’

511

PhD scholarship. A.J.H. is supported by a Robinson Fellowship. This work was supported by

512

funding from The University of Sydney. We thank the Sydney Mass Spectrometry facility for

513

access to LC-MS instruments; and all collaborators who kindly donated cell lines.

514

Author contributions

515

Nancy T Santiappillai:

Project administration, formal analysis, investigation, validation,

516

visualization, writing – original draft, writing -review & editing.

Yue Cao:

Formal analysis,

517

methodology, software, validation, visualization, writing -review & editing.

Mariam F

518

Hakeem-Sanni:

Investigation.

Jean Yang:

Methodology, supervision, writing -review &

519

editing.

Lake-Ee Quek:

Conceptualization, methodology, project administration, software,

520

supervision, writing – review & editing.

Andrew J Hoy:

Conceptualization, funding

521

acquisition, project administration, resources, supervision, writing – review & editing.

522

Disclosure and competing interests statement

523

The authors declare that they have no conflict of interest.

524

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

REFERENCES

525

Altea-Manzano P, Cuadros AM, Broadfield LA, Fendt SM (2020) Nutrient metabolism and

526

cancer in the in vivo context: a metabolic game of give and take.

EMBO Rep

21: e50635

527

Amara A, Frainay C, Jourdan F, Naake T, Neumann S, Novoa-del-Toro EM, Salek RM, Salzer

528

L, Scharfenberg S, Witting M (2022) Networks and Graphs Discovery in Metabolomics Data

529

Analysis and Interpretation.

Frontiers in Molecular Biosciences

530

Behan FM, Iorio F, Picco G, Gonçalves E, Beaver CM, Migliardi G, Santos R, Rao Y, Sassi F,

531

Pinnelli M

et al

(2019) Prioritization of cancer therapeutic targets using CRISPR–Cas9 screens.

532

Nature

568: 511-516

533

Benedetti E, Liu EM, Tang C, Kuo F, Buyukozkan M, Park T, Park J, Correa F, Hakimi AA,

534

Intlekofer AM

et al

(2023) A multimodal atlas of tumour metabolism reveals the architecture

535

of gene–metabolite covariation.

Nat Metab

5: 1029-1044

536

Bishop JL, Thaper D, Vahid S, Davies A, Ketola K, Kuruma H, Jama R, Nip KM, Angeles A,

537

Johnson F

et al

(2017) The Master Neural Transcription Factor BRN2 Is an Androgen

538

Receptor–Suppressed Driver of Neuroendocrine Differentiation in Prostate Cancer.

Cancer

539

Discov

7: 54-71

540

Byrne FL, Poon IKH, Modesitt SC, Tomsig JL, Chow JDY, Healy ME, Baker WD, Atkins KA,

541

Lancaster JM, Marchion DC

et al

(2014) Metabolic Vulnerabilities in Endometrial Cancer.

542

Cancer Res

74: 5832-5845

543

Cairns RA, Harris IS, Mak TW (2011) Regulation of cancer cell metabolism.

Nat Rev Cancer

544

11: 85-95

545

Chambers MC, Maclean B, Burke R, Amodei D, Ruderman DL, Neumann S, Gatto L, Fischer

546

B, Pratt B, Egertson J

et al

(2012) A cross-platform toolkit for mass spectrometry and

547

proteomics.

Nat Biotechnol

30: 918-920

548

Chen P-H, Cai L, Huffman K, Yang C, Kim J, Faubert B, Boroughs L, Ko B, Sudderth J,

549

McMillan EA

et al

(2019) Metabolic Diversity in Human Non-Small Cell Lung Cancer Cells.

550

Mol Cell

76: 838-851.e835

551

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

Cherkaoui S, Durot S, Bradley J, Critchlow S, Dubuis S, Masiero MM, Wegmann R, Snijder

552

B, Othman A, Bendtsen C

et al

(2022) A functional analysis of 180 cancer cell lines reveals

553

conserved intrinsic metabolic programs.

Mol Syst Biol

554

Cori CF, Cori GT (1925) the carbohydrate metabolism of tumors: ii. Changes in the sugar,

555

lactic acid, and co2-combining power of blood passing through a tumor.

J Biol Chem

65: 397-

556

405

557

Corsello SM, Nagari RT, Spangler RD, Rossen J, Kocak M, Bryan JG, Humeidi R, Peck D,

558

Wu X, Tang AA

et al

(2020) Discovering the anti-cancer potential of non-oncology drugs by

559

systematic viability profiling.

Nat Cancer

1: 235-248

560

DeBerardinis RJ, Chandel NS (2020) We need to talk about the Warburg effect.

Nat Metab

561

127-129

562

DeBerardinis RJ, Mancuso A, Daikhin E, Nissim I, Yudkoff M, Wehrli S, Thompson CB (2007)

563

Beyond aerobic glycolysis: transformed cells can engage in glutamine metabolism that exceeds

564

the requirement for protein and nucleotide synthesis.

Proc Natl Acad Sci U S A

104: 19345-

565

19350

566

Fendt SM, Frezza C, Erez A (2020) Targeting Metabolic Plasticity and Flexibility Dynamics

567

for Cancer Therapy.

Cancer Discov

10: 1797-1807

568

Gross MI, Demo SD, Dennison JB, Chen L, Chernov-Rogan T, Goyal B, Janes JR, Laidig GJ,

569

Lewis ER, Li J

et al

(2014) Antitumor Activity of the Glutaminase Inhibitor CB-839 in Triple-

570

Negative Breast Cancer.

Mol Cancer Ther

13: 890-901

571

Gu Z, Eils R, Schlesner M (2016) Complex heatmaps reveal patterns and correlations in

572

multidimensional genomic data.

Bioinformatics

32: 2847-2849

573

Hensley CT, Faubert B, Yuan Q, Lev-Cohain N, Jin E, Kim J, Jiang L, Ko B, Skelton R, Loudat

574

et al

(2016) Metabolic Heterogeneity in Human Lung Tumors.

Cell

164: 681-694

575

Huang Z, Wang C (2022) A Review on Differential Abundance Analysis Methods for Mass

576

Spectrometry-Based Metabolomic Data.

Metabolites

577

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

Jain M, Nilsson R, Sharma S, Madhusudhan N, Kitami T, Souza AL, Kafri R, Kirschner MW,

578

Clish CB, Mootha VK (2012) Metabolite profiling identifies a key role for glycine in rapid

579

cancer cell proliferation.

Science

336: 1040-1044

580

Jia S, Liu Z, Zhang S, Liu P, Zhang L, Lee SH, Zhang J, Signoretti S, Loda M, Roberts TM

581

(2008) Essential roles of PI(3)K–p110

in cell growth, metabolism and tumorigenesis.

582

Nature

454: 776-779

583

Jones RG, Thompson CB (2009) Tumor suppressors and cell metabolism: a recipe for cancer

584

growth.

Genes Dev

23: 537-548

585

Kamphorst JJ, Nofal M, Commisso C, Hackett SR, Lu W, Grabocka E, Vander Heiden MG,

586

Miller G, Drebin JA, Bar-Sagi D

et al

(2015) Human pancreatic cancer tumors are nutrient

587

poor and tumor cells actively scavenge extracellular protein.

Cancer Res

75: 544-553

588

Li H, Ning S, Ghandi M, Kryukov GV, Gopal S, Deik A, Souza A, Pierce K, Keskula P,

589

Hernandez D

et al

(2019) The landscape of cancer cell line metabolism.

Nat Med

25: 850-860

590

Li Y, Chan SC, Brand LJ, Hwang TH, Silverstein KA, Dehm SM (2013) Androgen receptor

591

splice variants mediate enzalutamide resistance in castration-resistant prostate cancer cell lines.

592

Cancer Res

73: 483-489

593

Liu M-X, Jin L, Sun S-J, Liu P, Feng X, Cheng Z-L, Liu W-R, Guan K-L, Shi Y-H, Yuan H-X

594

et al

(2018) Metabolic reprogramming by PCK1 promotes TCA cataplerosis, oxidative stress

595

and apoptosis in liver cancer cells and suppresses hepatocellular carcinoma.

Oncogene

37:

596

1637-1653

597

Luengo A, Li Z, Gui DY, Sullivan LB, Zagorulya M, Do BT, Ferreira R, Naamati A, Ali A,

598

Lewis CA

et al

(2021) Increased demand for NAD+ relative to ATP drives aerobic glycolysis.

599

Mol Cell

81: 691-707.e696

600

Mahajan P, Fiehn O, Barupal D (2024) IDSL.GOA: gene ontology analysis for interpreting

601

metabolomic datasets.

Sci Rep

14: 1299

602

Mansouri S, Shahriari A, Kalantar H, Moini Zanjani T, Haghi Karamallah M (2017) Role of

603

malate dehydrogenase in facilitating lactate dehydrogenase to support the glycolysis pathway

604

in tumors.

Biomed Rep

6: 463-467

605

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

Martínez-Reyes I, Chandel NS (2020) Mitochondrial TCA cycle metabolites control

606

physiology and disease.

Nat Commun

607

Molania R, Gagnon-Bartsch JA, Dobrovic A, Speed TP (2019) A new normalization for

608

Nanostring nCounter gene expression data.

Nucleic Acids Res

47: 6073-6083

609

Montal ED, Dewi R, Bhalla K, Ou L, Hwang BJ, Ropell AE, Gordon C, Liu WJ, DeBerardinis

610

RJ, Sudderth J

et al

(2015) PEPCK Coordinates the Regulation of Central Carbon Metabolism

611

to Promote Cancer Cell Growth.

Mol Cell

60: 571-583

612

Mullen NJ, Singh PK (2023) Nucleotide metabolism: a pan-cancer metabolic dependency.

Nat

613

Rev Cancer

23: 275-294

614

Nagarajan SR, Paul-Heng M, Krycer JR, Fazakerley DJ, Sharland AF, Hoy AJ (2019) Lipid

615

and glucose metabolism in hepatocyte cell lines and primary mouse hepatocytes: a

616

comprehensive resource for in vitro studies of hepatic metabolism.

Am J Physiol Endocrinol

617

Metab

316: E578-E589

618

Nassar ZD, Mah CY, Dehairs J, Burvenich IJ, Irani S, Centenera MM, Helm M, Shrestha RK,

619

Moldovan M, Don AS

et al

(2020) Human DECR1 is an androgen-repressed survival factor

620

that regulates PUFA oxidation to protect prostate tumor cells from ferroptosis.

Elife

621

Neinast MD, Jang C, Hui S, Murashige DS, Chu Q, Morscher RJ, Li X, Zhan L, White E,

622

Anthony TG

et al

(2019) Quantitative Analysis of the Whole-Body Metabolic Fate of

623

Branched-Chain Amino Acids.

Cell Metab

29: 417-429.e414

624

Nyquist MD, Li Y, Hwang TH, Manlove LS, Vessella RL, Silverstein KAT, Voytas DF, Dehm

625

SM (2013) TALEN-engineered AR gene rearrangements reveal endocrine uncoupling of

626

androgen receptor in prostate cancer.

Proc Natl Acad Sci U S A

110: 17492-17497

627

Oermann EK, Wu J, Guan KL, Xiong Y (2012) Alterations of metabolic genes and metabolites

628

in cancer.

Semin Cell Dev Biol

23: 370-380

629

Ortmayr K, Dubuis S, Zampieri M (2019) Metabolic profiling of cancer cells reveals genome-

630

wide crosstalk between transcriptional regulators and metabolism.

Nat Commun

631

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

Park JO, Rubin SA, Xu YF, Amador-Noguez D, Fan J, Shlomi T, Rabinowitz JD (2016)

632

Metabolite concentrations, fluxes and free energies imply efficient enzyme usage.

Nat Chem

633

Biol

12: 482-489

634

Parmenter TJ, Kleinschmidt M, Kinross KM, Bond ST, Li J, Kaadige MR, Rao A, Sheppard

635

KE, Hugo W, Pupo GM

et al

(2014) Response of BRAF-mutant melanoma to BRAF inhibition

636

is mediated by a network of transcriptional regulators of glycolysis.

Cancer Discov

4: 423-433

637

Pavlova NN, Zhu J, Thompson CB (2022) The hallmarks of cancer metabolism: Still emerging.

638

Cell Metab

34: 355-377

639

Quek L-E, van Geldermalsen M, Guan YF, Wahi K, Mayoh C, Balaban S, Pang A, Wang Q,

640

Cowley MJ, Brown KK

et al

(2022) Glutamine addiction promotes glucose oxidation in triple-

641

negative breast cancer.

Oncogene

41: 4066-4078

642

Schoors S, Bruning U, Missiaen R, Queiroz KC, Borgers G, Elia I, Zecchin A, Cantelmo AR,

643

Christen S, Goveia J

et al

(2015) Fatty acid carbon is essential for dNTP synthesis in

644

endothelial cells.

Nature

520: 192-197

645

Shorthouse D, Bradley J, Critchlow SE, Bendtsen C, Hall BA (2022) Heterogeneity of the

646

cancer cell line metabolic landscape.

Mol Syst Biol

647

Smith B, Schafer XL, Ambeskovic A, Spencer CM, Land H, Munger J (2016) Addiction to

648

Coupling of the Warburg Effect with Glutamine Catabolism in Cancer Cells.

Cell Rep

17: 821-

649

836

650

Spinelli JB, Haigis MC (2018) The multifaceted contributions of mitochondria to cellular

651

metabolism.

Nat Cell Biol

20: 745-754

652

Sung Y, Yu YC, Han JM (2023) Nutrient sensors and their crosstalk.

Exp Mol Med

55: 1076-

653

1089

654

Tsherniak A, Vazquez F, Montgomery PG, Weir BA, Kryukov G, Cowley GS, Gill S,

655

Harrington WF, Pantel S, Krill-Burger JM

et al

(2017) Defining a Cancer Dependency Map.

656

Cell

170: 564-576.e516

657

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.

bioRxiv preprint

van den Heuvel AP, Jing J, Wooster RF, Bachman KE (2012) Analysis of glutamine

658

dependency in non-small cell lung cancer: GLS1 splice variant GAC is essential for cancer cell

659

growth.

Cancer Biol Ther

13: 1185-1194

660

Vincent Emma E, Sergushichev A, Griss T, Gingras M-C, Samborska B, Ntimbane T, Coelho

661

Paula P, Blagih J, Raissi Thomas C, Choinière L

et al

(2015) Mitochondrial

662

Phosphoenolpyruvate Carboxykinase Regulates Metabolic Adaptation and Enables Glucose-

663

Independent Tumor Growth.

Mol Cell

60: 195-207

664

Vousden KH, Ryan KM (2009) p53 and metabolism.

Nat Rev Cancer

9: 691-700

665

Wang P, Ma J, Li W, Wang Q, Xiao Y, Jiang Y, Gu X, Wu Y, Dong S, Guo H

et al

(2023)

666

Profiling the metabolome of uterine fluid for early detection of ovarian cancer.

Cell Rep Med

667

101061

668

Warburg O (1925) über den Stoffwechsel der Carcinomzelle.

Klin Wochenschr

4: 534-536

669

Warburg O, Minami S (1923) Versuche an Überlebendem Carcinom-gewebe.

Klin Wochenschr

670

2: 776-777

671

Wicker CA, Hunt BG, Krishnan S, Aziz K, Parajuli S, Palackdharry S, Elaban WR, Wise-

672

Draper TM, Mills GB, Waltz SE

et al

(2021) Glutaminase inhibition with telaglenastat (CB-

673

839) improves treatment response in combination with ionizing radiation in head and neck

674

squamous cell carcinoma models.

Cancer Lett

502: 180-188

675

Yang W, Soares J, Greninger P, Edelman EJ, Lightfoot H, Forbes S, Bindal N, Beare D, Smith

676

JA, Thompson IR

et al

(2012) Genomics of Drug Sensitivity in Cancer (GDSC): a resource for

677

therapeutic biomarker discovery in cancer cells.

Nucleic Acids Res

41: D955-D961

678

679

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 21, 2024.