THEM6-mediated lipid remodelling sustains stress resistance in cancer

Despite the clinical benefit of androgen-deprivation therapy (ADT), the majority of patients with advanced prostate cancer (PCa) ultimately develop lethal castration-resistant prostate cancer (CRPC). In this study, we identified thioesterase superfamily member 6 (THEM6) as a marker of ADT resistance in PCa. In patients, THEM6 expression correlates with progressive disease and is associated with poor survival. THEM6 deletion reduces in vivo tumour growth and restores castration sensitivity in orthograft models of CRPC. Mechanistically, THEM6 is located at the endoplasmic reticulum (ER) membrane and controls lipid homeostasis by regulating intracellular levels of ether lipids. Consequently, THEM6 loss in CRPC cells significantly alters ER function, reducing de novo sterol biosynthesis and preventing lipid-mediated induction of ATF4. Finally, we show that THEM6 is required for the establishment of the MYC-induced stress response. Thus, similar to PCa, THEM6 loss significantly impairs tumorigenesis in the MYC-dependent subtype of triple negative breast cancer. Altogether, our results highlight THEM6 as a novel component of the treatment-induced stress response and a promising target for the treatment of CRPC and MYC-driven cancer.


INTRODUCTION
Androgen deprivation therapy (ADT) and androgen receptor (AR)-targeted therapies have significantly improved outcomes for patients suffering from advanced prostate cancer (PCa). However, treatment resistance ultimately leads to the development of lethal castrationresistant prostate cancer (CRPC), which remains a major therapeutic challenge.
Resistance to treatment is accompanied by a plethora of cellular and metabolic adaptations that allow cancer cells to cope with stress-inducing factors (1). Together with mitochondria, the endoplasmic reticulum (ER) plays a central role in the regulation of stresssignalling pathways. Indeed, the ER is critical for the establishment of a complex stress response, termed the unfolded protein response (UPR), that orchestrates the cellular adaptation to various perturbations such as impairment of protein or lipid homeostasis. Activation of the UPR relies on the coordinated action of three major branches, each of them characterized by the activity of a specific ER stress sensor: Inositol Requiring Enzyme 1 (IRE1α), PRKR-like Endoplasmic Reticulum Kinase (PERK) and Activating Transcription Factor 6 (ATF6). In cancer, correct induction of the UPR is required to support oncogenic transformation (2).
Persistent activation of ER stress responses can further promote tumour progression, metastasis dissemination and resistance to therapy (3). Therefore, clinical targeting of the UPR, alone or in combination with other treatment modalities, is regarded as a promising strategy for the treatment of aggressive cancer (4)(5)(6)(7). This strategy is particularly effective in the context of PCa, where inhibition of the IRE1a branch of the UPR significantly impairs tumour growth and activation of c-MYC signalling (8), an important feature of ADT resistance (9). Similarly, targeting of the PERK-eIF2a-ATF4 branch of the UPR effectively reduces tumour progression and metastasis dissemination in preclinical models of CRPC (10), while ATF4 signalling is essential for PCa growth and survival (11). 5 ER is also the primary site of lipid and cholesterol biosynthesis. Changes in lipid homeostasis, such as impaired membrane lipid saturation (12) or imbalance in the levels of phospho-and sphingolipid species (13,14), can also contribute to ER stress. Recently we and others demonstrated the importance of lipid remodelling in tumour resistance to antiandrogen therapy (15,16). Thus, targeting lipid-mediated ER stress could be considered as a potential therapeutic option for the treatment of CRPC.
Acyl-CoA thioesterases (ACOTs) are a class of enzymes that hydrolyse acyl-CoA molecules. In contrast with Type I enzymes, Type II ACOTs are related by structure rather than sequence. Type II ACOTs are characterised by the presence of an evolutionarily conserved domain, the "Hotdog" domain, which confers the thioesterase enzymatic activity (17). Type II ACOTs also include members of the Thioesterase Superfamily (THEM), which primarily function to deactivate fatty acyl-CoA thioesters and generate free fatty acids (FA) and CoA.
THEM proteins are involved in the regulation of intracellular FA trafficking and have been shown to influence both lipogenesis and beta-oxidation depending on the physiological context (18). Interestingly, Them1 and Them2 knockout mice exhibit severely attenuated responses to ER stress induction (19,20), thus supporting the role of these proteins in lipid-mediated activation of the UPR.
In this study, we identified Thioesterase Superfamily Member 6 (THEM6) as a clinically relevant protein associated with resistance to ADT. Mechanistically, we show that THEM6 maintains lipid homeostasis by controlling intracellular levels of ether lipids.
Consequently, THEM6 expression is critical for ER membrane protein trafficking, sterol biosynthesis and ATF4 activation under ADT conditions. Importantly, we further observe that THEM6 amplification is a frequent genomic alteration in cancer, and that THEM6 is functionally required to sustain MYC-induced UPR activation. Taken together, our results highlight the potential of THEM6 as a future therapeutic option for MYC-driven cancer. 6 7

THEM6 is overexpressed in CRPC and correlates with poor clinical outcome
We previously performed an in-depth proteomic analysis comparing in vivo hormonenaïve (HN) and castration-resistant (CRPC) orthograft models of PCa (21). From this comparison, we identified THEM6/c8orf55 as significantly upregulated in CRPC tumours (22rv1 and LNCaP AI) in comparison to HN counterparts (CWR22res and LNCaP, respectively) ( Fig. 1a and Suppl. Fig. 1a). Increased THEM6 levels in CRPC were validated in vivo (Fig. 1b-c) and in vitro (Suppl. Fig. 1b). Of note, THEM6 expression was the lowest in normal prostate epithelial cells (RWPE-1) when compared with multiple PCa cell lines (Suppl. We next assessed the clinical relevance of THEM6 in CRPC. For this purpose, we examined THEM6/c8orf55 expression in three publicly available datasets of localised and metastatic PCa (Suppl. Fig. 1c-f). Analysis of the PRAD TCGA data revealed that THEM6 mRNA expression was significantly higher in tumour tissue than in normal prostate epithelium (Suppl. Fig. 1c). Similarly, THEM6 levels gradually increased from benign to localised and metastatic PCa lesions ((22), Suppl. Fig. 1d), and were also elevated in distant metastases ((23), Suppl. Fig. 1e). Additionally, high THEM6 expression was significantly associated with poor patient survival ((23), Suppl. Fig. 1f). To further demonstrate the importance of THEM6 in PCa, we performed IHC on a tissue microarray (TMA, n = 297) comprised of tumours obtained from treatment-naïve (Untreated), treatment-responsive (neoadjuvant hormonal therapy, NHT-treated), treatment-resistant (CRPC) and neuroendocrine (NEPC) PCa patients (Fig. 1d). CRPC and NEPC tumours exhibited significantly higher levels of the THEM6 protein than untreated tumours, with the highest score obtained for CRPC (Fig. 1e). High THEM6 8 levels significantly correlated with high Gleason score and high T-stage tumours ( Fig. 1f and Suppl. Fig. 1g), and were further associated with the presence of metastases and cancer recurrence ( Fig. 1g and Suppl. Fig. 1h). Finally, we validated THEM6 as a prognostic factor in PCa by staining a second TMA composed of tumour biopsies from treatment-naïve patients documented with a 20-years follow-up (n = 69). In this cohort, high THEM6 expression was strongly associated with poor patient survival (recurrence-free and overall survival, Fig. 1h-i).
Finally, in both patient cohorts, high THEM6 levels were significantly associated with high Ki67 expression ( Fig. 1j and Suppl. Fig. 1i), indicating that THEM6 was expressed at higher levels in highly proliferative tumours.

Loss of THEM6 affects CRPC tumour growth
To investigate the role of THEM6 in CRPC, we generated stable CRISPR-based THEM6 knockout cell lines (hereafter referred to as THEM6 KO) (Fig. 1k). On average, THEM6 KO resulted in a mild (~20-25%) decrease in CRPC cell proliferation (Fig. 1l). In vivo, loss of THEM6 significantly reduced tumour volume in a CRPC model of 22rv1-derived orthografts (assessed by ultrasonography, Fig. 1m). In this model, orchidectomy is performed at the time of the orthotopic cell transplantation to mimic the in vivo environment of ADT. To complement the use of 22rv1 orthografts to model CRPC, we further applied the CWR22res orthograft model to study the effects of acute ADT on pre-established tumours (orchidectomy 3 weeks post injection), thus better resembling treatment in clinical patients (24). Similar to 22rv1, THEM6 KO strongly impaired CRPC tumour growth in the CWR22res orthograft model (Fig. 1n). In addition to decreased tumour size, THEM6-deficient orthografts exhibited large necrotic areas and decreased cellularity, especially under ADT conditions (Suppl. Fig.   1j). Taken together, these data support a pro-tumorigenic role for THEM6 in PCa. 9

THEM6 regulates cellular lipid metabolism
As a member of the THEM superfamily, THEM6 exhibits an evolutionarily conserved "Hotdog" domain predicted to confer thioesterase activity (18). Rewiring of lipid metabolism is a common feature of ADT resistance (15). Therefore, we hypothesised that THEM6 could participate in the lipid rearrangement required for CRPC progression. To evaluate the impact of THEM6 loss on the lipid composition of CRPC cells, we compared the lipid profiles of control (CTL) and THEM6 KO cells using LC-MS lipidomics. Strikingly, loss of THEM6 in 22rv1 cells resulted in a profound remodelling of the cellular lipidome. THEM6 depletion was associated with a strong reduction in the intracellular levels of multiple triglyceride (TG) and ether lipid species (ether triglycerides (ether TG) and ether phospholipids (ether PC or ether PE)). In contrast, THEM6 KO cells displayed increased amounts of ceramides (Fig. 2a). In addition to specific lipid changes, THEM6 KO also significantly affected the total amount of TGs, ether TGs and ceramides in 22rv1 cells (Fig. 2b). Interestingly, the intracellular levels of several ether lipid species (but not TG) were also strongly reduced in LNCaP AI THEM6 KO cells when compared to their respective CTL ( Fig. 2c and Suppl. Fig. 2a), suggesting that THEM6 loss primarily affects ether lipid homeostasis. We further assessed the effect of THEM6 KO on lipid content in vivo by performing Raman spectroscopy on ADT-treated CWR22res orthografts. Raman spectroscopy allows assessment and quantitation of lipids (band at 2845 cm -1 ) and cholesterol (band at 2880 cm -1 ) content on paraffin-embedded tumour slides in a non-destructive manner (Fig. 2d). Results from the analysis provided evidence that THEM6-deficient tumours (Fig. 1n) display significantly less lipids (Fig. 2e) and cholesterol ( Fig. 2f) than CTL tumours, thus confirming a role for THEM6 in the maintenance of the tumour lipidome.
THEM6 is an ER membrane protein that is essential to maintain ER integrity 10 Lipids are essential components of biological membranes. Interestingly, THEM6 depletion in CRPC cells resulted in increased cell stiffness (Fig. 3a), a phenotype which is indicative of membrane and cytoskeleton reorganisation and often associated with impaired migratory abilities (25). To gain further insights into the role of THEM6 in cancer cells, we performed a proteomic comparison of 22rv1 cells proficient or depleted for THEM6 (Fig. 3b).
Proteomic analysis highlighted a large cluster of ER-related proteins that were significantly down-regulated in the absence of THEM6 (FC = 1.3, p < 0.05, Fig. 3c and Suppl. Table 1). In addition, the majority of these proteins are described as membrane proteins (Fig. 3c). ER is particularly sensitive to lipid perturbation (26). Electron microscopy confirmed the negative impact of THEM6 depletion on ER morphology. Indeed, THEM6 KO cells presented with abnormal ER, exhibiting rounded structures with highly dilated lumens (Fig. 3d). The presence of abnormally enlarged mitochondria and large multilamellar lysosomes was also frequently observed in THEM6 KO cells, potentially reflecting general membrane perturbations following THEM6 loss (Suppl. Fig. 3a). Interestingly, THEM6 strongly co-localised with the ER marker calreticulin (Fig. 3e, Suppl. Fig. 3b), but not with mitochondria (Suppl. Fig. 3c). This result suggests that THEM6 is predominantly associated with ER, contrasting with the mitochondrial localisation reported for other THEM superfamily members (THEM2/4/5) (27). Furthermore, topological analysis of the THEM6 protein sequence highlighted a 17-amino acids signal peptide that corresponds to a well-defined N-terminal transmembrane domain (Fig. 3f). Finally, western blot (WB) analysis after subcellular fractionation confirmed the presence of THEM6 in the insoluble organellular/membrane fraction of CRPC cells (Fig. 3g, Suppl. Fig. 3d).

THEM6 regulates membrane protein trafficking in the ER
Lipid metabolism and protein homeostasis are highly interconnected processes (28). In particular, ether lipids and plasmalogen derivatives play a key role in regulating membrane 11 protein trafficking and homeostasis (29). Due to its localisation at the ER membrane and its role in regulating lipid balance, THEM6 might therefore be important for the maintenance of protein homeostasis in the ER. In line with this idea, pull-down experiments followed by MS analysis identified 152 proteins that significantly interacted with THEM6 in THEM6overexpressing HEK-293 cells (FC = 10, p < 0.05, Suppl. Fig. 3e and Suppl. Table 2). Most of the THEM6 interactors were located at the ER membrane, or at the interface between ER and nucleus (Fig. 3h), and were mainly involved in protein transport (Fig. 3i). Among others, several exportins, importins, transportins, as well as components of the oligosaccharyltransferase (OST) complex were identified as strong THEM6-interacting partners (Suppl. Table 2). Interactions between endogenous THEM6 and membrane proteins from different subcellular compartments (CRM1 at the outer nuclear membrane, AMFR and SEC61b in the ER) were confirmed using immunoprecipitation experiments in CRPC cell lines ( Fig. 3j). Importantly, acute THEM6 silencing in PCa cell lines led to a consistent decrease in the expression of the ER membrane-associated lectin calnexin (CALX) but did not affect the levels of the soluble homolog calreticulin (CALR) (Fig. 3k), indicating that THEM6 loss might preferentially affect the trafficking of membrane proteins.

THEM6 loss affects de novo sterol and FA synthesis in CRPC cells
In addition to protein trafficking, ER is also the primary site for lipid and cholesterol synthesis. Therefore, we postulated that THEM6 depletion would impact on these metabolic processes. Supporting this idea, our proteomic analysis of THEM6-deficient cells highlighted a down-regulation of multiple proteins involved in sterol biosynthesis (Fig. 4a). Sterol biosynthesis is of particular interest in the context of CRPC, as cholesterol serves as a precursor for de novo androgen synthesis and sustains ADT resistance (24,30). We first validated the decreased expression of several enzymes involved in sterol biosynthesis in THEM6 KO cells 12 (Fig. 4b). Next, we tested the effect of THEM6 loss on de novo sterol biosynthesis by incubating the cells with [U13C]-glucose and [U13C]-glutamine and following 13 C incorporation into sterols using GC-MS. Surprisingly, we were not able to detect a significant proportion of labelled cholesterol in 22rv1 cells. Instead, these cells accumulated large amounts of de novo synthesised desmosterol, an immediate precursor of cholesterol ( Fig. 4c and Suppl.  Table 3). We then assessed the contribution of THEM6 to FA synthesis by determining the relative proportions of 13 C-labelled palmitic, oleic and stearic acids in presence and absence of THEM6 ( Fig. 4g and Suppl. Fig. 4d). Similar to sterols, THEM6 KO cells accumulated significantly less labelled FA when compared to CTL, indicating reduced FA synthesis in absence of THEM6. Altogether, these results suggest that THEM6 expression is critical for the regulation of de novo lipid synthesis.
THEM6 is required to trigger ATF4 induction in CRPC cells 13 Perturbation of ER homeostasis results in the activation of a tightly regulated stress response program, the unfolded protein response (UPR), in order to rapidly alleviate ER stress (31). Enrichment pathway analysis of the THEM6-deficient 22rv1 cells highlighted "Response to ER stress" as the main pathway regulated in absence of THEM6 (Fig. 5a), with 17 proteins referenced in this pathway significantly down-regulated in both KO clones (FC = 1.3, p < 0.05, Fig. 5b and Suppl. Table 1). Interestingly, BIP (HSPA5), the main regulator of the UPR, was identified among the proteins significantly down-regulated in 22rv1 THEM6 KO cells (Fig.   5b), and we confirmed this result by WB (Fig. 5c). Impaired UPR activation in THEM6deficient cells was further evidenced by decreased levels of the UPR effectors XBP1s (spliced isoform of XBP1), ATF4 and CHOP (DDIT3) (Fig. 5c). As a consequence, THEM6 KO significantly sensitised CRPC cells to prolonged ER stress, caused by chronic tunicamycin treatment (Fig. 5d).
Proteomic analysis of THEM6 KO in LNCaP AI cells also highlighted ER perturbation as the main consequence of THEM6 depletion (Suppl. Fig. 5a and Suppl. Table 4). Of note, THEM6 KO in this cell type led to an accumulation of several ER chaperones, including BIP, and did not consistently affect XBP1s levels (Suppl. Fig. 5b). In contrast with IRE1a-XBP1s signalling, proteomic analysis highlighted many ATF4 targets that were down-regulated in LNCaP AI THEM6 KO cells (Fig. 5e), suggesting that THEM6 might be particularly important for ATF4 activation in PCa. In line with this idea, we found that ATF4 levels were strongly reduced in THEM6-deficient LNCaP AI cells (Fig. 5f), and that both ATF4 and CHOP were also decreased following transient THEM6 silencing in multiple PCa cell lines (Fig. 5g). In addition, analysis of the PRAD TCGA data revealed that EIF4G1, a member of the EIF4F complex required for ER stress-induced translation of ATF4, and ASNS, a canonical ATF4 target, were significantly enriched in tumours with high THEM6 expression ( Fig. 5h and Suppl. 14 Table 3), thus underpinning the importance of THEM6 in the establishment of the ATF4mediated stress response.
Lipid-mediated stress leads to UPR activation (26). Because of the implication of THEM6 in maintaining lipid homeostasis ( Fig. 2 and Fig. 4), we wondered whether the inability of the THEM6-deficient cells to induce ATF4 was due to changes in their lipid composition. While CTL cells strongly induced ATF4 expression following palmitic acid (PA) treatment, ATF4 levels remained unaffected by PA in THEM6 KO cells (Fig. 5i). Importantly, treatment with the ether lipid precursor hexadecylglycerol (HG) induced ATF4 expression to a similar level in both THEM6 KO and CTL cells (Fig. 5j). Taken together, these results suggest that the stress-induced activation of ATF4 depends, at least in part, on THEM6mediated lipid remodelling in CRPC.

THEM6 is required for MYC-dependent activation of ATF4
ADT is a major source of metabolic stress (32). In PCa, MYC is an important regulator of the UPR and promotes tumour survival in androgen-independent conditions (9,33).
Importantly, at gene level, we observed that THEM6 was located alongside MYC on the chromosome locus 8q24.3, a genomic region that is frequently dysregulated in cancer (34).
Consequently, we found that THEM6 was co-amplified with MYC in multiple prostate cancer datasets ( Fig. 6a-b). On average THEM6 and MYC genomic alterations occurred in 34% and 38% of PCa patients, and this incidence was even higher in metastatic cases (average genomic alteration frequency of 51% and 56% for THEM6 and MYC respectively, Fig. 6a). Accordingly, MYC expression was high in CRPC tumour orthografts compared to their HN counterparts ( Fig. 6c), and thus correlated with THEM6 expression (Fig. 1b). Furthermore, acute ADT was sufficient to increase both THEM6 and MYC levels in the CWR22Res orthograft model (Suppl. Fig. 6a). To evaluate the importance of THEM6 function in the context of high MYC 15 signalling, we used mouse embryonic fibroblasts (MEFs) as well as U2OS cells expressing a tamoxifen-inducible MYC-ER construct, hereafter referred to as MEFs-MYC ER and U2OS-MYC ER respectively. MYC activation in MEFs-MYC ER strongly increased THEM6 expression (Fig. 6d), and transient silencing of THEM6 significantly impaired the MYC-induced proliferation of these cells (Fig. 6e). Of note, prolonged MYC activation resulted in a strong accumulation of the THEM6 protein over time (Suppl. Fig. 6b) while its increased expression was transient at the mRNA level (~3-fold after 4h to 6h, Suppl. Fig. 6c), indicating that additional regulations might occur at the protein level. Activation of MYC in U2OS-MYC ER increased THEM6 protein expression and further led to a strong accumulation of ATF4 and CHOP (Fig. 6f). Strikingly, transient THEM6 silencing was sufficient to abolish the MYCinduced expression of these two factors, as well as strongly reducing CALX levels ( Fig. 6f). In addition, THEM6 depletion significantly altered the lipid profile of MYC-activated cells, resulting in decreased levels of many ether PCs and PEs (Fig. 6g). Altogether, these results suggest that THEM6 is essential for the establishment of a MYC-dependent stress response.

THEM6 is an actionable target in multiple tumour types
In addition to genomic amplification, THEM6 overexpression has been reported in multiple cancer types (35). Moreover, the ability of THEM6 to facilitate MYC-induced UPR suggests that the pro-tumoural effects of THEM6 might not be restricted to prostate cancer.
Meta-analysis of THEM6 expression across the different TCGA datasets revealed that THEM6 overexpression occurred frequently in hormone-dependent (Suppl. Fig. 6d) as well as liverand gastrointestinal cancers (Suppl. Fig. 6e). We further investigated THEM6 in breast tumours where hormone therapy is also routinely used as primary cancer treatment strategy. Using a breast cancer TMA (n = 551), we found that high THEM6 levels were associated with poorer clinical parameters, such as enhanced lymph node infiltration ( Fig. 6h), increased tumour 16 invasiveness ( Fig. 6i) and high levels of Ki67 (Fig. 6j). Similar to what we observed in PCa ( Fig. 5h), high THEM6 expression correlated with high ASNS expression in the BRCA TCGA dataset (Fig. 6k), and THEM6 silencing led to reduced levels of ATF4, CHOP and CALX in multiple breast cancer cell lines (Fig. 6l).
Breast cancer patients with high THEM6 expression further displayed a trend towards reduced survival (Suppl. Fig. 6f). Strikingly, this effect was emphasised in triple negative breast cancer (TNBC) (n = 123, Fig. 6m), an aggressive molecular subtype that heavily relies on MYC expression (36). Therefore, in a proof-of-concept experiment, we assessed the impact of THEM6 loss in TNBC cells. Consistent with a role for THEM6 in UPR regulation, THEM6 KO in MDA-MB-231 cells led to a strong decrease in ATF4 expression (Fig. 6n). Finally, similarly to the PCa, loss of THEM6 increased cell stiffness (Fig. 6o) and impaired in vivo tumour growth in the TNBC model (Fig. 6p, Suppl. Fig. 6g), thus supporting a general protumoural role for THEM6 in cancer.

DISCUSSION
The remarkable reliance of PCa on AR signalling led to the generation of effective targeted therapies, such as enzalutamide, abiraterone acetate and other small molecule inhibitors of the AR pathway, which have significantly improved the clinical management of non-resectable prostate tumours. However, resistance to such therapies ultimately results in the development of lethal CRPC. A detailed characterisation of the molecular signalling pathways associated with treatment resistance will therefore identify novel actionable targets and foster innovative therapeutic options. For example, a better understanding of treatment-induced metabolic rewiring (15) or specific dependencies on stress response pathways (5) might uncover vulnerabilities with the potential to synergise with current treatments. Our lab recently engaged in a comparison of different in vivo models of ADT resistance at multi-omics levels (21). From this comprehensive dataset, we identified Thioesterase Superfamily Member 6 (THEM6) as a clinically relevant protein overexpressed in CRPC. In patients, THEM6 expression correlated with tumour aggressiveness and was highest in treatment-resistant tumours. Furthermore, high THEM6 expression was associated with poor clinical tumour parameters, such as enhanced tumour proliferation and metastatic dissemination, and correlated with shortened overall and disease-free patient survival. Importantly, the prognostic potential of THEM6 was not restricted to PCa but could also be applicable to other cancer types, such as TNBC, which at present have few treatment options. THEM6, formerly c8orf55, was originally discovered as a potential biomarker for colon cancer in a proteomic-based analysis of human samples. The authors further validated THEM6 overexpression in several tumour types, including breast and prostate (35). THEM6 has also been reported in additional proteomic studies, mainly comparing tumoural and respective non-malignant tissues from various origins (37,38). 18 Despite no characterised biological function, THEM6 has recently been classified in the THEM thioesterase superfamily due to the presence of a "HotDog" domain, an evolutionarily conserved domain with a predicted thioesterase activity (17). Members of the THEM superfamily (THEM1/ACOT11, THEM2/ACOT13, THEM4 and THEM5) display an acyl-CoA thioesterase activity, although presenting differences in specificity towards their fatty acyl-CoA substrates. Consequently, the THEM proteins have been described to play various roles in FA and lipid metabolism (18). In line with these studies, we found that THEM6 depletion significantly altered the lipid composition of cancer cells. In particular, loss of THEM6 resulted in consistent decreases in the levels of several ether lipid classes (ether TGs and ether PCs). Despite being relatively understudied in this context, ether lipids are frequently over-represented in tumours (39) and support a pro-oncogenic phenotype (40). Ether lipids are a subset of glycerolipids characterised by the presence of an ether-alkyl or a vinyl ether-alkyl (plasmalogens) bond at the glycerol sn-1 position. Their synthesis is initiated in peroxisomes and terminates in the ER, where they undergo active acyl chain remodelling (41). In addition to their roles in redox homeostasis (42,43) and cellular signalling, ether lipids are essential components of biological membranes. The importance of ether lipids in membrane protein trafficking, as well as their coordinate regulation with sphingolipids, has recently been described in mammalian cells (44). Interestingly, we identified THEM6 as an ER membrane protein whose absence primarily impacts ER function and morphology. Indeed, THEM6deficient cells displayed abnormal ER structure with the presence of highly dilated lumen, a phenotype that is also observed following dysregulation of ether lipid metabolism (45). The observation that both mitochondrial and lysosomal structures were also affected in THEM6 KO cells further suggests that THEM6 loss induces a global perturbation of membrane homeostasis. Moreover, THEM6 closely interacted with many membrane proteins involved in protein transport (exportins, importins, transportins, components of the ERAD machinery and 19 the OST complex), and THEM6 silencing led to a selective down-regulation of the membraneassociated chaperone calnexin, without affecting its soluble homolog calreticulin. CALX and CALR share substantial sequence identity, but the former is a type 1 transmembrane protein while the second mainly resides in the ER lumen (46). Of note, this decrease in CALX levels was much more pronounced upon transient siRNA transfection than following CRISPRmediated KO, suggesting that THEM6-deficient cells might have adapted to compensate for this effect. Taken together, these results suggest that THEM6 is involved in the regulation of membrane homeostasis and trafficking. An appealing hypothesis would be that THEM6 influences membrane properties by controlling the balance between ester-and ether-bound lipids within the ER, for example by regulating the pool of FA that must be incorporated into ether lipids. The fact that THEM6 is the only member of the THEM superfamily presenting with a transmembrane domain supports this hypothesis, but at the same time renders the purification and structural analysis of the THEM6 protein very challenging. Therefore, additional work remains to be done to fully characterise THEM6 enzymatic activity and uncover its biochemical substrates.
ER is essential for the maintenance of both lipid and protein homeostasis. Perturbation of ER homeostasis, induced by increased proteo-or lipotoxicity, activates a complex signalling network, called the UPR, which aims at rapidly reducing ER stress (31). Importantly, we demonstrate that THEM6 is required for a correct activation of the UPR. In particular, the inability of cancer cells to activate the ATF4/CHOP pathway was a constant feature of THEM6-deficient cancer cells, irrespective of their origin. Furthermore, THEM6-depleted cells were unable to respond to palmitate-induced ER stress, but managed to activate the UPR in response to treatment with hexadecylglycerol, a precursor of the ether lipid synthesis. This result not only demonstrates the implication of THEM6 in ether lipid metabolism, but also highlights its importance in the establishment of the lipid-mediated stress response. 20 Importantly, failure to activate the UPR in response to ER stress has also been reported in the context of THEM1 and THEM2 deficiencies. Indeed, genetic ablation of Them1 in mice attenuated diet-and chemical-induced ER stress responses, as well as UPR-mediated lipogenesis (19). Of note, the authors also speculated that Them1 could influence ER stress levels by modulating the phospholipid composition of the ER membrane. Similarly, an elegant study from the same group demonstrated that Them2-mediated trafficking of saturated fatty acid critically regulates ER membrane fluidity, and is therefore required for calcium-dependent induction of ER stress in non-physiological condition (20).
ER stress induction and subsequent activation of the UPR have been described to promote tumorigenesis and treatment resistance in cancer (47). In PCa, resistance to ADT involves the activation of a sustained stress response together with major metabolic reprogramming (15,48). An important contributor to these two processes is the MYC oncogene, which promotes androgen-independent growth and survival in PCa cells (9). MYC amplification is frequent in advanced PCa and correlates with enhanced tumour aggressiveness and increased resistance to treatment (49). Concomitant activation of MYC and the UPR is commonly observed in cancer, and both processes are tightly interconnected (33). As such, therapeutic targeting of the UPR has emerged as a promising strategy for the treatment of MYC-reliant tumours, including prostate (2,(6)(7)(8)10). Interestingly, THEM6 is co-amplified with MYC in cancer patient biopsies, while THEM6 expression is consistently increased upon MYC activation in different cell models. Furthermore, we showed that THEM6 is required for the correct activation of ATF4 following MYC induction. In addition to its role in cancer metabolism (50), ATF4 is essential for the regulation of MYC-dependent translation in cancer (51). Moreover, ATF4 has also been highlighted as an important regulator of PCa growth and survival (11). Along the same lines, Nguyen et al. demonstrated that the PERK-eiF2α pathway is selectively activated in metastatic PCa, and that targeting of this pathway significantly 21 impairs tumour growth and metastasis dissemination in preclinical models (10). Overall, targeting oncogene-induced cellular or metabolic dependencies represents a promising approach for the treatment of aggressive cancers (6,52). Therefore, through its role in lipid remodelling and its ability to regulate oncogene-induced UPR, THEM6 might provide an actionable target with the potential to be considered for the development of future anticancer therapies. 22

Cell culture
LNCaP, C4-2, CWR22res, VCaP, DU-145, PC-3 and PC-3met were cultured in RPMI USA). Protein depletion was confirmed using western blot. 23 Cell proliferation 1×10 6 cells were seeded in 6-well plates and allowed to attach overnight. The next day, cells were harvested with trypsin (T0) or allowed to grow for additional 72 hours. Cells were then counted using a CASY cell counter (Roche, Basel, Switzerland). Final cell number was normalised to the initial cell count obtained at T0. Data are expressed as relative percentage of CTL cells.

Measurement of cell stiffness using Atomic Force Microscopy
The mechanical properties of individual cells were measured using an Atomic Force Microscope Nanowizard II (JPK Instruments, Bruker, Berlin, Germany) with cell heater attachment mounted on an inverted optical microscope (Zeiss Observer Axio A.1, Zeiss, Cambridge, UK). Force indentation measurements were carried out as described previously (25,54). Briefly, the AFM colloidal probes were prepared by gluing a 5. Agarose beads were resuspended in a 2 M Urea and 100 mM ammonium bicarbonate buffer 25 and stored at -20°C. Biological triplicates were digested with Lys-C (Alpha Laboratories, Eastleigh, UK) and trypsin (Promega, Madison, WI, USA) on beads as previously described (56). Prior mass spectrometry analysis, digested peptides were desalted using StageTip (57).
Peptides from all experiments were separated by nanoscale C18 reverse-phase liquid chromatography using an EASY-nLC II 1200 (Thermo Fisher Scientific

Lipidomic analysis
Lipidomic analysis was performed according to (15). Single-phase lipid extraction was carried out using an extracting solution of methanol-butanol (1:1 ratio, BuMe), kept at

GC-MS-based determination of 13 C-sterols and fatty acids
Fatty acids (FA) and sterol measurements were performed as previously described in (15) and (65)   Fatty Acid Methyl Esters (FAMEs) and sterols were analysed using an Agilent 7890B GC system coupled to a 7000 Triple Quadrupole GC-MS system, with a Phenomenex ZB-1701 column (30 mm × 0.25 mm × 0.25 μm). For FAMEs, an initial temperature of 45 o C was set to increase at 9 o C/min, held for 5 min, then 240 o C min -1 , held for 11.5 min, before reaching a final temperature of 280 o C min -1 , held for 2 min. For cholesterol, the initial temperature was set at 200°C and increased at 20°C/minute up to 280°C, and held for 9 minutes. The instrument was operated in pulsed splitless mode in the electron impact mode, 50eV, and mass ions were integrated for quantification using known standards to generate a standard curve. Palmitic, stearic and oleic acid peak areas were extracted using mass-to-charge ratios (m/z) 270, 298 and 296 respectively. Cholesterol peak areas were extracted from m/z 458. Mass Hunter B.06.00 software (Agilent) was used to quantify isotopomer peak areas before natural abundance isotope correction was performed using an in-house algorithm. 30 Cholesterol was analyzed using an Agilent 7890B GC system coupled to an Agilent 7000 Triple Quadrupole GC-MS system, which was operating in a single quadrupole mode, with a Phenomenex ZB-1701 column (30 mm × 0.25 mm × 0.25 μm). An initial temperature of 200°C was set to increase at 20°C/minute up to 280°C, and held for 9 minutes. The instrument was operated in splitless mode in the electron impact mode, 70eV, for quantification and 50eV for labeling experiments. Cholesterol was quantified and isotope labeling pattern analyzed using Mass Hunter B.06.00 software (Agilent). Cholesterol and lathosterol internal standard peak areas were extracted from mass-to-charge ratio (m/z) 458 for both. Cholesterol was normalized to the internal standard, and a standard curve was used to quantify mg cholesterol per sample.

Bioinformatics analysis
Gene expression data were downloaded from TCGA and the GEO website. TCGA RNASeqV2 data were shift log transformed; GSE35988 and GSE21034 data were log transformed, using mean of probes per gene. Expression values were grouped according to sample type and group distributions were plotted using the matlab routine boxplot with the bar indicating the median, the box spanning from the 25th to the 75th percentiles, and whiskers spanning 2.7\sigma. Outliers beyond that span are indicated in red. Significant difference between the groups was measured using overall ANOVA.
For survival analysis (GSE21034), gene expression data from human samples (excluding cell lines) was normalised as before, mean-centered, and clustered into three groups using kmeans. Kaplan-Meier survival curves for the groups were plotted using the matlab routine kmplot.

Raman Spectroscopy analysis
Raman spectra were acquired on a Renishaw inVia Raman microscope equipped with a 532 nm Nd:YAG laser giving a maximum power of 500 mW, 1800 l mm -1 grating, and a Nikon NIR Apo 60×/1.0 N.A. water dipping objective. Prior to Raman measurements, tissue sections were dewaxed in xylene (2 × 15 min), 100% ethanol (5 min), 95% EtOH (5 min) and 90% EtOH (5 min). A water dipping objective was used to map the tissue sections with a step 32 size of 100 µm in x and y, with 1 s acquisition time, 100% laser power and a spectral center of 3000 cm −1 .
For initial pre-processing steps Renishaw Wire 4.1 software was used. The inbuilt software functions were used to remove cosmic rays followed by baseline subtraction. Baseline was subtracted using the baseline subtraction intelligent fitting function (with an 11 th order polynomial fitting and noise tolerance set to 1.50, applied to the whole spectral dataset). Further data analysis steps were then performed using custom MATLAB® scripts. Outlier spectra which gave high intensity due to saturation or fluorescence were removed using a threshold function. Spectra were cut to the region of interest between 2800 cm −1 and 3020 cm −1 . Spectra from tissue regions were then extracted (based on total spectral intensity) and all extracted spectra were min-max scaled for comparison between conditions. For each map the resultant spectral data set went through further quality control steps which involved firstly, excluding spectra with a total spectral intensity less than 60000, and then removing spectra out with one standard deviation of the mean. Spectra were then scaled to the peak at 2933 cm −1 , and spectral data sets for all CTL and THEM6 KO samples were combined. The average for each condition was plotted for comparison. The ratio of the intensities of the peaks at 2845 cm −1 and 2935 cm −1 as well as the ratio of the intensities of the peaks at 2880 cm −1 and 2935 cm −1 were determined for each spectral data point. GraphPad Prism 8.4.2 was used to produce graphs and perform statistical analysis of the data. Pictures were taken on a Nikon A1R confocal microscope (Nikon Instruments Europe B.V.,

Statistical analysis
Statistical analyses were performed using GraphPad PRISM software v8.4.2 (GraphPad Software Inc, San Diego, CA, USA).      The following databases were used in this study: