Cystathionine-β-synthase is essential for AKT-induced senescence and suppresses the development of gastric cancers with PI3K/AKT activation

Hyperactivation of oncogenic pathways downstream of RAS and PI3K/AKT in normal cells induces a senescence-like phenotype that acts as a tumor-suppressive mechanism that must be overcome during transformation. We previously demonstrated that AKT-induced senescence (AIS) is associated with profound transcriptional and metabolic changes. Here, we demonstrate that human fibroblasts undergoing AIS display increased Cystathionine-β-synthase (CBS) expression and consequent activation of the transsulfuration pathway controlling hydrogen sulfide (H2S) and glutathione (GSH) metabolism. Activated transsulfuration pathway during AIS maintenance enhances the antioxidant capacity, protecting senescent cells from ROS-induced cell death via GSH and H2S. Importantly, CBS depletion allows cells that have undergone AIS to escape senescence and re-enter the cell cycle, indicating the importance of CBS activity in maintaining AIS. Mechanistically, we show this restoration of proliferation is mediated through suppressing mitochondrial respiration and reactive oxygen species (ROS) production and increasing GSH metabolism. These findings implicate a potential tumor-suppressive role for CBS in cells with inappropriately activated PI3K/AKT signaling. Consistent with this concept, in human gastric cancer cells with activated PI3K/AKT signaling, we demonstrate that CBS expression is suppressed due to promoter hypermethylation. CBS loss cooperates with activated PI3K/AKT signaling in promoting anchorage-independent growth of gastric epithelial cells, while CBS restoration suppresses the growth of gastric tumors in vivo. Taken together, we find that CBS is a novel regulator of AIS and a potential tumor suppressor in PI3K/AKT-driven gastric cancers, providing a new exploitable metabolic vulnerability in these cancers.


56
Hyperactivation of oncogenic pathways such as RAS/ERK or PI3K/AKT can cause cellular 57 senescence in non-transformed cells, termed oncogene-induced senescence (Serrano et al., 58 1997;Zhu et al., 2020). In addition to the well-studied RAS-induced senescence (RIS), we and 59 others have demonstrated that hyperactivation of PI3K/AKT signaling pathway causes a 60 senescence-like phenotype, referred to as AKT-induced senescence (AIS) or PTEN loss- 2005). Therefore, understanding the molecular mechanisms that regulate AIS and how they are 71 subverted will provide opportunities to identify therapeutic strategies for suppressing 72 PI3K/AKT-driven cancer development. 73 We identified 98 key regulators in a whole-genome siRNA AIS escape screen and validated a 74 subset of these genes in the functional studies to confirm their role in AIS maintenance (Chan 75 et al., 2020). Intriguingly, 11 genes were associated with the regulation of metabolism, 76 suggesting that an altered metabolism could be integral for maintaining AIS. In particular, the 77 cystathionine-β-synthase (CBS) was ranked as one of the top metabolic gene candidates with 78 loss of expression leading to AIS escape (Chan et al., 2020), but how it does so is not known. 79 CBS is an enzyme involved in the transsulfuration metabolic pathway. CBS converts 80 homocysteine (Hcy), a key metabolite in the transmethylation pathway, to cystathionine which 81 is subsequently hydrolysed by cystathionine gamma-lyase (CTH) to form cysteine, the crucial 82 precursor for glutathione (GSH) production ( Fig. 1). CBS also catalyses the production of 83 hydrogen sulfide (H2S), a diffusible gaseous transmitter that modulates mitochondrial function 2012) and -suppressive roles (Kim et al., 2009). These studies underscore the context-100 dependent roles of CBS in cancer development. 101 In this study we explored the molecular mechanisms underpinning CBS's role in maintaining 102 AIS and how the loss of CBS promotes AIS escape.  (Carpten et al., 2007), also enhanced CBS protein expression (Fig.S1B). However, AKT 120 hyperactivation did not affect CBS mRNA expression (Fig S1C), suggesting a post-121 transcriptional regulatory mechanism underpinning increased CBS protein expression. 122 We hypothesised that the increased CBS expression in AKT hyperactivated cells was 123 associated with upregulation of transsulfuration pathway activity and cysteine metabolism. We 124 thus examined the expression levels of CTH, a key enzyme in the transsulfuration pathway and 125 xCT, the Xc-amino acid antiporter responsible for the up-take of cystine (an oxidised form of 126 cysteine) (Fig 1A). Both CTH and xCT were upregulated during AIS compared to proliferating 127 control cells, suggesting an elevated cysteine synthesis via the transsulfuration pathway and 128 cysteine uptake (Fig. 1B). In contrast, senescent cells expressing HRAS V12 cells exhibited a 129 moderate increase of CBS but no change in CTH expression level. An increase in xCT was 130 also observed during RIS, albeit to a lesser extent than during AIS, in line with the finding that 131 upregulation of xCT facilitates RAS-mediated transformation (Lim et al., 2019). 132 To assess transsulfuration pathway activity, we measured H2S production. A significant 133 increase in transsulfuration pathway activity was observed in BJ-TERT fibroblasts upon AKT 134 but not HRAS hyperactivation (Fig.1C), suggesting that activation of transsulfuration pathway 135 is a specific cellular response to constitutive activation of AKT. Cysteine starvation has been 136 reported to induce necrosis and ferroptosis in cancer cells (Chen et al., 2017). Since the 137 transsulfuration pathway mediates De Novo cysteine synthesis, an increase in transsulfuration 138 pathway activity may support survival of AKT expressing cells upon cysteine limitation.

139
Consistent with this hypothesis, inhibition of H2S production by aminoxyacetate (AOAA) 140 (Szabo, 2016) selectively impaired cell proliferation (Fig. 1D) and increased SA-βGal activity 141 (Fig. S1D) of BJ-TERT cells overexpressing myr-AKT1, suggesting that the transsulfuration 142 pathway has a protective effect on cells expressing hyperactivated AKT. Furthermore, cysteine 143 deprivation potently increased the expression levels of CBS and CTH in AIS cells (Fig.1B)  144 and did not affect the survival of AIS cells, consistent with increased cysteine synthesis due to 145 elevated CBS expression being critical for cell viability (Fig.1E). 146

Depletion of CBS promotes escape from AKT-induced senescence 147
To validate the functional role of CBS in AIS maintenance, we depleted CBS using two 148 independent shRNAs in BJ-TERT cells overexpressing myr-AKT1 ( Fig.2A). Loss of CBS in 149 AIS cells significantly decreased the proportion of cells with SA-βGal activity, increased EdU 150 incorporation and enhanced colony formation, demonstrating an essential role of CBS in AIS 151 maintenance ( Fig 2B). To further confirm the on-target specificity of the knockdown, we 152 generated BJ-TERT cells expressing a doxycycline-inducible CBS shRNA and an shRNA-153 resistant 4-OHT-inducible estrogen receptor (ER)-tagged-CBS fusion ( Fig.2C and S2A). Upon 154 expressing myr-AKT1, these cells underwent AIS, as indicated by a significant increase in SA-155 βGal-positive cells and decrease in EdU-positive cells and, consistent with the finding in the 156 siRNA AIS escape screen. CBS depletion induced by doxycycline decreased SA-βGal activity 157 and increased EdU incorporation in AIS cells (Fig.2D). Importantly, simultaneously depleting 158 CBS and expressing ER-CBS prevented senescence escape, confirming the on-target 159 specificity of the knockdown and the critical role of CBS in maintaining AIS (Fig.2D). 160 Modulation of CBS expression in proliferating (pBabe) control cells did not affect the 161 percentage of SA-βGal-and EdU-positive cells (Fig.2D), suggesting that CBS is a downstream 162 effector of AKT and its depletion alone is insufficient to induce senescence. Similar to the 163 findings in BJ-TERT cells, AKT1 hyperactivation also caused senescence in IMR-90 lung 164 fibroblasts ( Fig S2B). CBS knockdown in AIS cells significantly suppressed SA-βGal staining 165 and enhanced EdU incorporation but not in proliferating cells ( Fig S2B). Knockdown of CBS 166 in BJ-TERT cells with constitutive RAS activation did not affect colony formation, suggesting 167 CBS has a specific regulatory role during AIS but not RIS maintenance ( Fig.2E and S2C). absence of CBS ( Fig 2F). Consistent with our previous findings, p53 and its downstream target 176 p21 were upregulated during AIS, but depletion of CBS had no effect on these levels.

177
Furthermore, total Rb, a key regulator of the G1/S phase transition, and its inhibitory 178 phosphorylated form at serine 807/811 were markedly suppressed in AIS and partially rescued 179 upon CBS knockdown. Cyclin A, which mediates S to G2/M phase cell cycle progression, was 180 also upregulated upon depleting CBS in cells with hyperactivated AKT. These results 181 demonstrate that CBS depletion can restore the proliferation of cells that have undergone AIS 182 in a p53-independent manner. 183

Depletion of CBS in AIS cells increases GSH metabolism in cysteine-replete conditions. 184
CBS is the key enzyme regulating transsulfuration and transmethylation pathways. By analysis 185 of the data from the AIS-escape siRNA screen, we found that except CBS, siRNA knockdown 186 of other genes involved in transsulfuration and transmethylation pathway did not significantly 187 affect AIS cell numbers (robust Z score < 2, Table S4). Therefore, it is likely that AIS escape 188 in cysteine-replete conditions by loss of CBS is through a transsulfuration/transmethylation 189 pathway-independent mechanism. In addition, knockdown of CBS did not affect the expression 190 levels of CTH and xCT when BJ-TERT cells were grown in cysteine-replete culture medium 191 (Fig.S3A). supporting that AKT activation stimulates the transsulfuration pathway, consistent with the 200 increased H2S production capacity in AIS cells (Fig. 1C). Activation of the transsulfuration 201 pathway can also increase GSH synthesis. Indeed, AIS cells exhibited a much higher level of 202 the GSH synthesis precursor γ-Glu-Cys (Fig.3H) and the GSH degradation product Cys-Gly 203 (Fig.3I) than proliferating cells despite lack of change in GSH (Fig.3G), suggesting an elevated 204 GSH metabolism in AIS cells. 205 Loss of CBS can cause upstream substrate hcy accumulation. Indeed, compared to AIS cells 206 with intact CBS, an elevated hcy level in CBS-depleted cells (myrAKT1-siCBS) was detected 207 (Fig.3A). However, the metabolites in the transmethylation pathway including methionine 208 ( suggesting loss of CBS further stimulates GSH metabolism. In contrast, H2S production was 212 not affected by CBS knockdown (Fig.3J). Upregulation of GSH metabolism upon CBS 213 depletion implicated that senescence escape due to CBS knockdown is associated with 214 regulation of oxidative stress and antioxidation activity. 215 To explore whether other metabolic alterations are associated with CBS-mediated AIS 216 maintenance, we performed gas chromatography mass spectrometry (GC/MS)-based 217 untargeted metabolomics (Table S6 and Table S8), which is the key mechanism for transporting 223 NADH from the cytoplasm to the mitochondria in order to support oxidative phosphorylation. 224 Thus, alteration of mitochondrial function required to maintain AIS may be associated with 225 AIS escape caused by CBS depletion. 226 To further investigate the molecular mechanisms of how CBS maintains AIS, we characterised 227 the transcriptomic changes upon depleting CBS during AIS. RNA-seq was performed on BJ-228 TERT cells expressing myrAKT1 transduced with control or CBS shRNA. Differential gene 229 expression analysis of CBS-depleted AIS cells compared with control AIS cells revealed 404 230 genes were significantly up-regulated (adjusted p-value < 0.05, Log2FC > 1) and 181 genes 231 significantly down-regulated (adjusted p-value < 0.05, Log2FC < -1) ( Fig.S3C and Table S9  232 and S10). Gene set enrichment analysis (GSEA) using the hallmark gene sets in the Molecular 233 Signatures Database (MSigDB) identified that pathways involved in oxidative phosphorylation 234 and reactive oxygen species (ROS) were significantly downregulated in CBS-depleted AIS 235 cells compared to the control AIS cells ( Fig.3L and 3M, Table S11 and S12). Collectively, 236 these data implicate altered mitochondrial energy metabolism and ROS production contribute 237 to CBS-dependent AIS maintenance. 238

CBS deficiency alleviates oxidative stress in AIS cells 239
CBS has been reported to localise to both the cytoplasm and mitochondria and regulate to the proliferating cells ( Fig.4A) and an increased abundance of proteins involved in the 245 mitochondrial electron transport chain (Fig.S4A). 246 To investigate the role of CBS-mediated mitochondrial alterations in AIS maintenance, we 247 examined the oxidative phosphorylation status in AIS cells transfected with control or CBS 248 siRNA by measuring the oxygen consumption rate (OCR) using a Seahorse extracellular flux 249 analyser. OCR was markedly elevated during AIS (Fig.4C) and knockdown of CBS 250 significantly suppressed basal OCR and ATP production (Fig.4D). These results suggest that 251 CBS is required for enhanced oxidative phosphorylation during AIS maintenance and CBS 252 depletion reduces mitochondrial bioenergetics in AIS. To test whether these effects were 253 specific for AIS, we also performed Seahorse analysis on cells during RIS. While basal OCR 254 was also increased during RIS, knockdown CBS in RIS cells, in contrast to those undergoing 255 AIS, had an opposite effect on oxidative phosphorylation by further upregulating basal OCR 256 and ATP production (Fig.4D). This distinct effect on mitochondrial activity is likely to 257 contribute to the specific regulatory role of CBS in AIS maintenance. 258 Mitochondria are the major intracellular organelles of ROS production. Elevated ROS results 259 in oxidative stress which may underlie AKT-induced senescence. To test this hypothesis, we 260 first treated AIS cells with an antioxidant Trolox, which resulted in increased proliferation ( these results strongly support the concept that increased oxidative phosphorylation and ROS 267 production sustain AIS status and are impaired by CBS loss. In parallel, AKT activation 268 increases transsulfuration pathway activity, which consequently stimulates GSH and H2S 269 production, thereby protecting AIS cells from ROS-induced cell death. Notably, the antioxidant 270 activity of this metabolic pathway is still retained and GSH metabolism is even further 271 enhanced upon CBS loss. Together the alleviated oxidative stress contributes to escape of AIS 272 cells from cell cycle arrest (Fig. 4H). 273

CBS expression is frequently suppressed in gastric cancer 274
Given that we showed CBS loss promotes escape from AIS, we hypothesised that loss of CBS 275 could cooperate with oncogenic activation of the PI3K/AKT/mTORC1 pathway to promote 276 tumourigenesis. Analysis of TCGA stomach adenocarcinoma data from 478 samples using 277 cBioPortal (http://www.cbioportal.org) identified CBS deep deletions and mutations in gastric 278 cancer (Fig.5A). Further analysis of the TCGA gastric cancer patient data revealed that 279 PI3K/AKT/mTORC1 signalling pathway alterations occur in 33% of gastric cancer with 280 PIK3CA (15%) and PTEN (9%) alterations being the most common genetic alterations 281 (Fig.S5A). CBS genetic alterations were found in 3% of gastric cancers including 4 cases with 282 CBS deep deletions co-occurring with hyperactive PI3K/AKT/mTORC1 signalling (Fig.S5A). 283 Furthermore, CBS mRNA level was significantly decreased in tumours (N=406) compared to 284 adjacent normal tissues (N=211) in gastric cancer patients (Fig.5B), consistent with the 285 hypothesis that CBS is a tumour suppressor in gastric cancers characterised by 286 PI3K/AKT/mTORC1 pathway alterations. 287 To evaluate alteration of CBS protein expression in human gastric cancer, we assessed CBS 288 protein levels in paired samples of gastric tumours and adjacent non-cancerous mucosa from 289 62 gastric cancer patients in a tissue microarray using immunofluorescent staining (Fig.5C). 290 This tissue microarray was assembled from paraffin-embedded tissue blocks collected from 291 compared to the adjacent normal gastric tissues. 296 To establish a cell-based system to probe the interaction between activated 297 PI3K/AKT/mTORC1 signalling and loss of CBS expression, we first examined CBS protein 298 expression in six gastric cancer cell lines compared with a SV40-transformed gastric epithelial 299 cell line GES-1, which was derived from foetal stomach mucosa and was non-tumourigenic in 300 nude mice (Ke et al., 1994). Compared to gastric epithelial cells, CBS expression was markedly 301 decreased in all gastric cancer cell lines while elevated AKT activity, as indicated by increase 302 of AKT phosphorylation, was observed in in AGS, Hs746T, KATO III gastric cancer cell lines 303 (Fig.5F). To further test if CBS loss cooperates with PI3K/AKT/mTORC1 hyperactivation in gastric 321 cancer oncogenesis, we transduced myrAKT1 and CBS/control shRNA into GES-1 cells and 322 tested their ability to form colonies in soft agar (Fig.6A). Consistent with our results in 323 fibroblasts, AKT hyperactivation increased transsulfuration pathway activity and GSH 324 production, which was not affected by CBS depletion (Fig. S6A and S6B). Hyperactivation of 325 AKT1 had no significant effect on cell proliferation in 2D culture and did not induce 326 senescence due to inactivation of p53 and pRb by SV40T in the GES-1 cells (Fig. S6C). 327 However, it effectively promoted anchorage-independent growth, and this was further 328 enhanced by CBS depletion (Fig.6A and 6B). CBS depletion also significantly enhanced 329 anchorage-independent growth of GES-1 cells with clustered regularly interspaced short 330 palindromic repeats (CRISPR)/Cas9-mediated PTEN knockout ( Fig.6C and 6D) but not in 2D 331 cell culture (Fig. S6D). These results support that CBS loss can cooperate with PI3K/AKT 332 signalling to promote oncogenic transformation. 333 To further investigate the functional cooperation of CBS and PI3K/AKT signalling in gastric 334 cancer pathogenesis, we engineered AGS gastric cancer cells, which harbour CBS deficiency 335 and PIK3CA mutations E545A and E453K resulting in AKT activation (Fig.5F), to express a 336 doxycycline-inducible wild type CBS or an inactive CBS I278T mutant. This mutation is the most 337 frequently observed CBS mutation in cancer cells and exhibits only ~2.4% of the enzyme 338 activity of wild type CBS (Kruger & Cox, 1995). Treatment of AGS cells with doxycycline 339 restored CBS wt and CBS I278T protein expression to a level comparable to that of GES-1 gastric 340 epithelial cells (Fig.6E). Restoration of wild type CBS increased H2S production (Fig.6F) but 341 did not significantly enhance GSH abundance (Fig.S6E). Interestingly, restoration of CBS 342 expression did not affect AGS cell proliferation (Fig.S6F). 343 To evaluate the functional impact of CBS restoration in vivo, we transplanted the AGS cells 344 expressing doxycycline-inducible CBS into immunocompromised mice. Induction of CBS wt 345 significantly suppressed AGS tumour growth (Fig.6G). Induction of CBS was also associated 346 with a marked decrease in Ki67 expression and inhibitory RB phosphorylation without altering 347 p53 and p21 expression levels in the tumour tissues, suggesting that restoration of CBS 348 expression could suppress gastric tumour formation independent of p53 ( Fig.6H and 6I). AKT-hyperactivated cells supports a protective effect of H2S from oxidative stress-induced 390 cell death (Fig.1D). The protective role of the transsulfuration pathway in AIS is further 391 supported by the finding that myrAKT1-expressing cells are resistant to exogenous cysteine 392 deprivation (Fig. 1E). Thus, we propose that increased levels of GSH and H2S through the 393 transsulfuration pathway during AIS maintenance enhance the antioxidant capacity of AIS 394 cells, protecting senescent cells from ROS-induced cell death (Fig.4H). 395 Startlingly, we found AIS cells escape upon CBS depletion under cysteine-replete conditions. 396 This is mediated by decrease of ROS production through suppression of mitochondrial 397 oxidative phosphorylation and increase of GSH metabolism (Fig. 4H). Intriguingly, a recent 398 publication reported that overproduction of H2S by increased mitochondrial-localised CBS 399 expression results in suppression of mitochondrial oxidative phosphorylation and ATP 400 production in the fibroblasts from Down syndrome patients (Panagaki et al., 2019). This 401 discrepancy could be partially explained by the bell shaped or biphasic biological effect of H2S 402 as mentioned above. It is unclear how loss of CBS increased GSH metabolism. One explanation 403 is that loss of CBS reciprocally upregulates xCT activity and leads to a compensatory 404 upregulation of cystine uptake and GSH synthesis.  has been removed and EBFP2 was replaced with the puromycin resistance gene. The resulting 462 plasmid was designated pRT3-puro-CBS. HEK293T cells were used for virus production and 463 viral transductions were performed as previously described. prepared. The fresh complete medium was replaced at 24 hours post-transfection. 474

Colony formation assay 482
Cells were seeded in 6-well culture plates. Media was refreshed every 2 days. At the end of 483 experiments, cells were fixed with 100% (v/v) methanol for 30 mins at RT and then stained 484 with 0.1% w/v crystal violet for 30 mins at RT. After intensively washing with H2O and drying 485 plates, images were obtained using ChemiDoc TM Imaging system. Total colony area, expressed 486 as percentage of cell coverage per well, was determined using the ImageJ plugin Colony Area. 487 Anchorage-independent soft agar assay 488 The anchorage-independent soft agar assay was performed as described by Borowicz  were washed with cool PBS and then harvested in cell lysis buffer (50mM Tris-HCl pH8, 514 150mM NaCl, 1% v/v IGEPAL® CA-630(SigmaAldrich-I3021), 1% v/v Triton-X100). 515 Samples were kept on ice for 1 hour and then centrifuged at 20,000g for 10mins at 4°C before 516 protein quantification by DC protein assay. 400 ug proteins were mixed with 100μl AzMC 517 reaction master mix and incubated at 37°C in dark for 2 hours. Samples were read at 340nm 518 excitation and 445nm emission wavelength using Cytation TM 3 Cell imaging multi-mode reader 519 (BioTek). 520

Gas chromatography-mass spectrometry (GC/MS) Metabolomics 521
After saline wash, cells were quenched by pouring liquid nitrogen into 6-well plates and then 522 harvested with ice cold methanol:chloroform:scyllo-inositol (MeOH:CHCl3 9:1 v/v) 523 containing 3μM scyllo-inositol as internal standard. The extracts were vortexed for 10 seconds 524 and incubated on ice for 15 minutes. By centrifugation at 4°C for 3mins at 16,100g, the 525 supernatant was collected and snap-frozen in liquid nitrogen and stored at -80°C. 526 The samples were evaporated to dryness by vacuum centrifugation. Prior to GC/MS analysis, 527 samples were derivatised with 25μl 3% (w/v) methoxyamine in pyridine (Sigma, 528 #226904/270970) for 60 min at 37°C with mixing at 750 rpm, followed by trimethylsilylation 529 with 25μl BSTFA + 1 % TMCS (Thermo, #38831) for 60 min at 37°C with mixing at 750 rpm. 530 The derivatized sample (1μl) was analyzed using Shimadzu GC/MS-TQ8040 system, running 531 the Shimadzu Smart Metabolites NRM database, comprising approximately 475 metabolite 532 targets. Statistical analyses were performed using Student's t test following log transformation 533 and median normalization. Metabolites were considered to be significant if their adjusted p-534 values after Benjamini-Hochberg correction were less than 0.05. Further data analysis and 535 enrichment analysis were performed through MetaboAnalyst 4.0. 536 Liquid chromatography-mass spectrometry (LC/MS) metabolomics 537 Thiol derivatization using N-Ethylmaleimide (NEM) and LC/MS based detection were 538 described previously and optimized (Ortmayr et al., 2015). Cells were seeded four days before 539 harvest. Cells were washed with ice cold Saline and then collected with ice cold extraction 540 buffer containing 50 mM NEM (Sigma-Aldrich-E3876) in 80% v/v methanol and 20% v/v 541 10mM ammonium formate (Sigma-Aldrich-516961) at pH 7. Final concentration of 2 µM L-542 Valine-13 C5, 15 N (Sigma-Aldrich-600148), D-Sorbitol-13 C6 (Sigma-Aldrich-605514) and L-543 Leucine-13 C6 (Sigma-Aldrich-605239) were added in the extraction buffer as internal 544 standards. Samples were mixed at 4°C on a vortex mixer for at 1 hour at 1,000 rpm before 545 centrifugation at 20,000 g for 10 mins at 4°C. Supernatants were stored on ice and processed 546 as per described previously (Ortmayr et al., 2015). Briefly, metabolites were separated on an 547 Dionex Ultimate 3000RS HPLC system (Thermo Scientific, Waltham, Massachusetts, USA) 548 using an InfinityLab Poroshell 120 HILIC-Z (100 x 2.1 mm, 1.9 µm) HPLC column (Agilent 549 Technologies, Santa Clara, California, USA) maintained at 25°C with buffer A (20 mM 550 (NH4)2CO3, pH 9.0; Sigma-Aldrich) and solvent B (100% acetonitrile) at a flow rate of 300 551 μl/minute. Chromatographic gradient started at 90% B, decreased gradually over 10 min to 552 65% B, then to 20% B at 11.5 min, stayed at 20% B until 13 min, returned to 90% B at 14 min 553 and equilibrated at 90% B until 20 min. Metabolites were detected by mass spectrometry on 554 an Q-Exactive Orbitrap mass spectrometer (Thermo Scientific) using a heated electrospray 555 ionisation source (HESI). LC-MS data was collected using full scan acquisition in positive and 556 negative ion mode utilizing polarity switching. Data processing was performed using 557 Tracefinder application for quantitative analysis (Thermo Scientific). NEM derivatised thiols 558 were detected in negative ion mode. Metabolite identification was based on accurate mass, 559 retention time and authentic reference standards. Peak intensities were normalised against 560 internal standards and cell numbers, and the statistical analyses were carried out using one-way 561 ANOVA. 562 563 564 RNA-sequencing and analysis 565 RNA sequencing and analysis were described previously (Chan et al., 2020