Selective abrogation of S6K2 identifies lipid homeostasis as a survival vulnerability in MAPK inhibitor-resistant NRAS-mutant melanoma

Abstract

Although oncogenic NRAS activates mitogen-activated protein kinase (MAPK) signaling, inhibition of the MAPK pathway is not therapeutically efficacious in NRAS-mutant (NRAS^MUT) tumors. Here we report that selectively silencing the ribosomal protein S6 kinase 2 (S6K2) while preserving the activity of S6K1 perturbs lipid metabolism, enhances fatty acid unsaturation, and triggers lethal lipid peroxidation in NRAS^MUT melanoma cells that are resistant to MAPK inhibition. S6K2 depletion induces endoplasmic reticulum stress, and peroxisome proliferator-activated receptor alpha (PPARα) activation, triggering cell death selectively in MAPK inhibitor-resistant melanoma. We found that combining PPARα agonists and polyunsaturated fatty acids phenocopied the effects of S6K2 abrogation, blocking tumor growth in both patient-derived xenografts and immunocompetent murine melanoma models. Collectively, our study establishes S6K2 and its effector subnetwork as promising targets for NRAS^MUT melanoma that are resistant to global MAPK pathway inhibitors.

One Sentence Summary:

S6 kinase 2 is a vulnerability in mitogen-activated protein kinase inhibitor-resistant NRAS-mutant melanoma.

INTRODUCTION

RAS-mutant tumors are highly aggressive and mostly refractory to currently available targeted therapies. In cutaneous melanoma, mutations in NRAS occur in almost 30% of all tumors (1). Oncogenic NRAS elicits persistent activation of the mitogen-activated protein kinase (MAPK) cascade, which plays a key role in melanomagenesis; therefore, inhibition of MAPK has been evaluated as a potential therapeutic approach for NRAS^MUT melanoma. However, inhibitors of the MAPK pathway (MAPKi) as single agents elicit response rates below 20% and do not prolong the survival in patients with NRAS^MUT melanoma compared to chemotherapy (2, 3). Further, suppression of MAPK signaling often leads to feedback or compensatory activation of the phosphoinositide 3-kinase (PI3K)/AKT pathway (4–7). Unfortunately, drug combinations that inhibit both the MAPK and PI3K pathways are poorly tolerated in patients, and clinically efficacious doses have not been achieved (8–12). Although simultaneous inhibition of the MAPK and PI3K pathways is generally toxic, there is no evidence that the proximal molecular causes of that toxicity are the same in normal and in cancer tissues. This raises the possibility that targeting a convergent subnetwork or node downstream from the PI3K and MAPK pathways could elicit lethality in melanoma cells while sparing normal tissues. Thus, mapping out the MAPK/PI3K downstream subnetwork could help identify much needed actionable drug targets for NRAS^MUT tumors.

Dual inhibition of MEK and PI3K in NRAS^MUT melanoma pre-clinical models leads to substantial perturbation of metabolic pathways (13). Oncogenic NRAS controls cell metabolism through the mechanistic target of rapamycin protein complex 1 (mTORC1), a critical RAS effector (14) that integrates upstream signals, including the MAPK and PI3K pathways (15, 16). Key effectors of mTORC1 are the 40S ribosomal kinases (S6K), S6K1 and S6K2 (17). The S6K pathway plays important roles in gene transcription, protein translation, cellular metabolism, and cell survival (16–18). Despite the high degree of structural homology between S6K1 and S6K2, their expression, activation and cellular localization are differentially regulated (17). Hyperactivated S6K2 signaling is characteristic of breast cancer subtypes (19), with S6K2 expression promoting breast cancer cell survival (20). Loss of S6K1 typically leads to compensatory upregulation of S6K2 and vice versa (21–24). Furthermore, S6K1 and S6K2 have overlapping, as well as distinct biological functions (16–18, 23–25). This suggests that the functional activity of these two kinases is tightly coordinated and imbalance between them could lead to differential outcomes.

Here, we identified a S6K2/peroxisome proliferator-activated receptor alpha (PPARα) subnetwork downstream of the MAPK and PI3K pathways. Perturbation of this subnetwork by uncoupling S6K1 and S6K2 triggered lipid metabolic imbalance and endoplasmic reticulum (ER) stress, leading to oxidative cell death selectively in NRAS^MUT melanomas that are resistant to MAPK inhibition. Further, we provide proof-of-concept that this vulnerability can be exploited to curb MAPKi-resistant tumors and to inform the design of future strategies to combat NRAS^MUT melanoma.

RESULTS

S6K2 is a vulnerability in NRAS^MUT melanoma resistant to MAPK inhibitors.

Since MAPK inhibition elicits heterogeneous effects in NRAS^MUT melanoma (26), we sought to identify potential vulnerabilities in MAPK inhibitor (MAPKi)-resistant tumor cells. We first analyzed the effect of MEK (trametinib and MEK162) or ERK (SCH-984 and BVD-523) inhibitors, collectively referred to as MAPKi, on viability of NRAS^MUT melanoma cells (fig. S1A to C, table S1). MAPKi induced variable effects on viability, proliferation, and cell death (fig. S1A to E), despite effective and persistent inhibition of the MAPK pathway (fig. S1F to G). We classified cells as MAPKi-sensitive (MAPKi-S) based on their response to MAPKi, which was defined by ≥75% suppression of 5-Bromo-2’-deoxyuridine (BrdU) incorporation and ≥35% cell death (fig. S1D and E). Conversely, MAPKi elicited modest or no suppression of BrdU (<75%; fig. S1D) and marginal induction of cell death (< 35%; fig. S1E) in the MAPKi-resistant (MAPKi-R) cells.

To further explore signaling perturbations in MAPKi-R versus MAPKi-S cells, we analyzed cells treated with either the MEK inhibitor trametinib or the ERK inhibitor SCH772984 (SCH984) by immunoblotting and reverse phase protein array (RPPA) (Fig. 1A, fig. S1F and G, data file S1). MAPKi similarly suppressed the MAPK pathway in both sensitive and resistant cells (Fig. 1A, fig S1F). MAPKi-treatment did not induce notable differential activation of PI3K/AKT (Fig. 1A, data file S1). In contrast, MAPKi treatment blocked phosphorylation of S6 kinase (S6K) and its substrate S6 in MAPKi-S, but not in MAPKi-R melanoma cells (Fig. 1A, fig. S1F and data file S1).

Fig. 1. S6K is persistently activated in MAPKi-resistant NRAS^MUT melanoma.

(A) NRAS^MUT melanoma cells were treated for 48 h with the MEKi trametinib (M; 10 nM) or ERKi SCH984 (E; 100 nM) and changes in protein expression or phosphorylation were determined by RPPA analysis (n = 3). Heatmap depicts log₂ fold changes in MAPK, PI3K, mTOR, and S6K pathways relative to vehicle controls. Differences in response between resistant versus sensitive cell lines were calculated with unpaired, two-sided Mann-Whitney tests, indicating differential regulation of S6K signaling between MAPKi-resistant and -sensitive cell lines. (B) MAPKi-resistant (M93–047, WM1366) or -sensitive (WM3451, WM4113) melanoma cells were transduced with doxycycline-inducible constitutively active HA-tagged S6K2^T388E or empty vector (EV), pre-treated with 0.25 μg/ml doxycycline for 48 h, and then treated with DMSO vehicle (Veh), ERKi SCH984 (1 μM) or PI3K/mTORi GSK2126458 (100 nM) for an additional 24 h. Percentage of pS6(S240/244)⁺ cells was determined by flow cytometry; bar graphs show mean ± SD (n = 3); p-values were calculated by unpaired, two-tailed Student’s t-tests, with p-values of <0.017 considered statistically significant after adjusting for multiple hypothesis testing with Bonferroni’s correction. (C) Cells treated as in (B) were probed by immunoblotting with the indicated antibodies. (D) Diagram illustrating that in MAPKi-resistant NRAS^MUT melanoma cells, S6K is primarily regulated by the PI3K/mTOR pathway. In MAPKi-sensitive cells, S6K relies on both the MAPK and PI3K pathways.

Since S6K is also a bona fide effector of the PI3K/mTOR pathway (27), we further examined the impact of MAPK or PI3K/mTOR signaling on S6K. We ectopically expressed doxycycline-inducible constitutively active S6K (S6K1^T389E or S6K2^T388E) constructs in MAPKi-sensitive or resistant cells and quantified pS6^S240/244, a site exclusively phosphorylated by S6K (15) (Fig. 1B and C, fig. S1H and I). Treatment of MAPKi-S cells with MAPKi (SCH984) or PI3K/mTORi (GSK458) decreased phosphorylation of S6, whereas doxycycline-induced expression of constitutively active S6K1^T389E or S6K2^T388E rescued S6 phosphorylation in cells treated with SCH984 or GSK458 (Fig. 1B and C, fig. S1I). These data indicate that S6K mediates both MAPK and PI3K/mTOR signaling in MAPKi-S cells. In contrast, treatment of MAPKi-R cells with MAPKi had no effect on phosphorylation of S6 or mTOR, whereas treatment with PI3K/mTORi effectively suppressed pS6 and p-mTOR (Fig. 1B and C, fig. S1I). Ectopic expression of S6K1^T389E or S6K2^T388E rescued S6 phosphorylation in MAPKi-R cells treated with PI3K/mTORi (Fig. 1B and fig. S1I). Together, these data indicate that S6K activity is regulated by the PI3K/mTOR pathway in MAPKi-R cells. Moreover, these findings highlight that different upstream pathways differentially impinge on S6 kinase in MAPKi-R versus MAPKi-S cells (Fig. 1D).

We next examined the expression of S6K1/2 in patient samples by interrogating the cancer genome atlas (TCGA) skin cutaneous melanoma dataset. We noted that mRNA expression of the gene encoding S6K2 (RPS6KB2) was inversely correlated with expression of the gene encoding S6K1 (RPS6KB1) (n=443, R=-0.47, p val=9.23e-26; Fig. 2A). We also noted that high expression of RPS6KB2 (but not RPS6KB1) was associated with poor survival in patients with melanoma (Fig. 2B). Furthermore, analysis of RNA-sequencing (RNA-seq) data from NRAS^MUT melanoma patient-derived cells revealed that RPS6KB2 mRNA expression is significantly (p=0.0078) higher in MEKi-R tumor cells than in MEKi-S tumor cells (28), whereas RPS6KB1 mRNA expression was not significantly (p=0.3901) different in MEKi-S versus MEKi-R patient derived tumor cells (Fig. 2C). Additionally, analysis of basal expression of S6K2 and S6K1 protein in NRAS^MUT melanoma lines (6 MAPKi-R and 6 MAPKi-S), showed that MAPKi-R cells are characterized by higher S6K2 expression whereas S6K1 expression was heterogeneous between MAPKi-R and –S cells (Fig. 2D). These data support the notion that S6K2 could have an important role in melanoma and MAPKi resistance and thereby constitutes a vulnerability in this tumor type.

Fig. 2. S6K2 is a vulnerability in NRAS^MUT melanoma resistant to MAPK inhibitors.

(A) RPS6KB1 mRNA expression inversely correlates with RPS6KB2 (443 patients; TCGA SKCM dataset). (B) Overall survival curves are depicted for patients with high versus low RPS6KB2 or RPS6KB1 expression (TCGA SKCM; n = 470; first versus last quartile cut-off). (C) Expression of RPS6KB2 and RPS6KB1 in trametinib-sensitive (MEKi-S; n=12) or -resistant (MEKi-R; n=11) patient-derived NRAS^MUT melanoma cell lines as determined by RNA sequencing. RNA-Seq by Expectation Maximization (RSEM)-normalized counts were used (28). (D) Basal S6K1/2 protein expression was analyzed in a panel of MAPKi-R (n=6) and MAPKi-S (n=6) NRAS^MUT melanoma cell lines by immunoblotting (representative image, n=2). Box and whisker plots show mean densitometry values normalized to actin loading control (n=2). Data in (C and D) are presented as mean ± SD; data were analyzed by unpaired, two-tailed Student’s t-tests. (E to H) Three MAPKi-R (E and F) or MAPKi-S (G and H) cell lines were transduced with lentiviruses encoding S6K1 or S6K2 shRNA and analyzed for cell death 7 days after transduction by Annexin V/PI (E and G) or by immunoblotting (F and H) (n = 2; arrowhead indicates S6K2). (I) M93–047 cells were transduced with constitutively active S6K1^T389E, treated with DMSO or the pan-S6K inhibitor LY4702 (LY4702) for 48 hours and analyzed for cell death (left panel, n=3) or by immunoblotting (right panel, representative image, n=2). Data in E, G and I are shown as mean ± SD; n=3; unpaired two-tailed Student’s t-test.

To explore the potential role of S6K2 as a vulnerability in NRAS^MUT melanoma, we silenced S6K1 or S6K2 in 6 NRAS^MUT melanoma lines (3 MAPKi-S and 3 MAPKi-R) using multiple hairpins. Depletion of S6K2, but not S6K1, induced cell death in MAPKi-R cells (Fig. 2E and F). In contrast, neither depletion of S6K1 nor S6K2 triggered substantial cell death in MAPKi-S NRAS^MUT melanoma cells (Fig. 2G and H). These data indicate that S6K2 is a vulnerability in NRAS^MUT melanoma cells that are resistant to MAPKi. To mimic the effects of an S6K2-specific inhibitor, we ectopically expressed activated S6K1 in cells treated with the pan-S6K inhibitor LY2584702 (LY4702) (Fig. 2I). Activated S6K1 enhanced LY4702 cytotoxicity, rather than compensated for it, suggesting that sensitivity to the pan-S6K inhibitor is mediated mainly through S6K2 inhibition in this scenario. Collectively, these results indicate that selective inhibition of S6K2, in the context of active S6K1, induces death of MAPKi-R NRAS^MUT melanoma cells.

Depletion of S6K2 triggers lipid peroxidation and facilitates cell death.

To investigate the mechanisms by which depletion of S6K2 triggers cell death in MAPKi-R cells, we surveyed the proteome. Our goal was to identify proteins that were differentially affected by the depletion of S6K2 compared with S6K1 in MAPKi-R cells. Depletion of S6K2 was coupled to enhanced expression of proteins involved in lipid synthesis, fatty acid synthesis, uptake and activation, and phospholipid remodeling (Fig. 3A, fig. S2A, and data file S2). To further assess if lipid metabolism was perturbed by S6K2 depletion, we performed unbiased global lipidomic analysis in S6K2 or S6K1-depleted cells. Depletion of S6K2 led to an enrichment of specific lipid species containing polyunsaturated fatty acyl chains (PUFAs) such as phosphatidylcholine, phosphatidylethanolamine and phosphatidylglycerol (Fig. 3B, data file S3), consistent with disrupted lipid homeostasis. These results suggest that S6K2 blockade is coupled to an enhanced degree of fatty acid unsaturation in some lipid classes.

Fig. 3. Depletion of S6K2 triggers lipid peroxidation.

(A and B) NRAS^MUT melanoma cells (M93–047) were transduced with lentivirus encoding S6K1 or S6K2 shRNA, or empty vector (EV) and subjected to proteomic (n=3) (A) or lipidomic (n=3) (B) analysis 4 days after transduction. (A) Heatmap depicting abundance of selected proteins involved in lipid metabolic pathways relative to EV control; p-values denote two-tailed Student’s t-tests of shS6K2 versus shS6K1 adjusted for multiple hypothesis testing at a 10% FDR (Benjamini-Hochberg). (B) Box and whiskers plot depicting classes of lipids differentially affected by shS6K2 versus shS6K1. Median, upper quartile and lower quartiles are shown. Each dot represents an individual lipid species. Light gray dots represent lipids without any PUFA tails, dark gray indicates lipids with at least 1 PUFA tail. Details of lipids are presented in data file S3. (C to F) Three MAPKi-R and three MAPKi-S cell lines were transduced with lentiviruses encoding S6K2 or S6K1 shRNA, or EV. ROS (H2DCFDA; C and D) or lipid peroxidation (BODIPY C11; E and F) were assessed 5 days post transduction by flow cytometry. (C and E) Representative histograms of three independent experiments (n=3) showing mean fluorescence intensity (MFI) relative to EV control. (D and F) Box and whisker plots of pooled data for MAPKi-R (n=3) and MAPKi-S (n=3) cell lines transduced with two different shRNAs for S6K1 or S6K2. Median, upper quartile and lower quartile are shown. For (B, D, and F) data are from three independent biological replicates. Kruskal-Wallis ANOVA tests were used to compare groups and corrected for multiple comparisons with Dunn's test.

We also noted that depletion of S6K1 or S6K2 led to increased expression of proteins involved in the oxidative stress response, including reactive oxygen species (ROS) sensing/detoxifying enzymes such as gamma-glutamyltransferase 2 (GGT2), superoxide dismutase 2 (SOD2) and parkinson disease protein 7 (PARK7) (Fig. 3A, fig. S2A and B and data file S2). Whereas depletion of either S6K1 or S6K2 (fig. S2C) led to increased ROS (Fig. 3C and D), depletion of S6K2 induced significantly higher relative ROS compared with S6K1 depletion in both MAPKi-R cells (p=0.001) and MAPKi-S cells (p=0.003) (Fig. 3C and D). Although depletion of S6K2 increased ROS in MAPKi-S cells, these cells did not undergo cell death (Fig. 2G), implying that ROS alone are not sufficient to induce cell death or that MAPKi-S cells are able to counteract redox imbalance. The perturbation of redox and lipid homeostasis upon S6K2 knockdown raised the possibility that S6K2 depletion could be triggering lipid peroxidation (29). Indeed, depletion of S6K2 (but not S6K1) substantially enhanced lipid peroxidation and oxidative stress, as indicated by the lipid ROS sensor BODIPY 581/591 C11 (BODIPY C11) (Fig. 3E and F), generation of 4-hydroxynonenal (4-HNE) adducts, and increased nucleic acid oxidative damage (8-hydroxy-2'-deoxyguanosine (8-OHdG); fig. S2D to H) as well as DNA double strand break formation (gamma-phosphorylated histone H2AX (p-γH2AX); fig. S2I) in MAPKi-R cells. Whereas S6K2 depletion induced a similar increase in general ROS in both MAPKi-S and MAPKi-R cells, lipid ROS were preferentially induced in MAPKi-R cells (Fig. 3E and F). Together, these results support the premise that depletion of S6K2 enhances lipid metabolism and ROS, which jointly trigger lethal lipid peroxidation and oxidative cell death in MAPKi-R NRAS^MUT melanoma.

S6K2 depletion activates a terminal unfolded protein response (UPR) with concomitant upregulation of PPARα.

Excessive lipid peroxidation has been linked to different types of cell death including apoptosis and ferroptosis (29–35). To identify S6K2 effectors that could trigger cytotoxicity, we assessed global transcriptional changes induced by S6K2 depletion. Transcriptomic analysis indicated a general enrichment of UPR regulators in both S6K1- and S6K2-depleted cells, whereas X-Box Binding Protein 1 (XBP-1) and target genes of the lipid metabolism regulator Peroxisome Proliferator Activated Receptor (PPAR) (36, 37) were selectively increased in S6K2-depleted cells (Fig. 4A and B, table S2). Oxidative or metabolic stress can disrupt endoplasmic reticulum (ER) homeostasis, promoting the aberrant accumulation of misfolded proteins, inducing ER stress and further dysfunction of lipid metabolism (38). ER stress initiates the UPR through binding immunoglobulin protein (BiP)-mediated activation of three canonical pathways: Inositol-Requiring Enzyme 1 alpha (IRE1α)-XBP-1, Protein Kinase RNA-like Endoplasmic Reticulum Kinase (PERK)-C/EBP Homologous Protein (CHOP), and Activating Transcription Factor 6 (ATF6) (39), either restoring ER homeostasis under conditions of mild ER stress or triggering apoptosis when ER stress is severe and persistent. We observed increased expression of activated, spliced XBP-1 and its regulator phospho-IRE1α, in both protein and mRNA in S6K2-depleted cells (Fig. 4A, C and D). In contrast, S6K2 depletion had a negligible effect on PERK-CHOP signaling (Fig. 4C). Treatment of MAPKi-R NRAS^MUT melanoma cells with the IRE1α inhibitor Kira6 attenuated cell death induced by S6K2 depletion (fig. S3A) indicating that the cytotoxic effects of S6K2 depletion are partially mediated by a terminal UPR that activates apoptosis (40). Additionally, treatment with the pan-caspase inhibitor zVAD(OMe)-FMK (zVAD) also attenuated cell death (Fig. 4E). In contrast, treatment of MAPKi-R NRAS^MUT melanoma cells with the ferroptosis inhibitor, ferrostatin-1, did not attenuate cell death triggered by S6K2 depletion (fig. S3B). Upregulation of XBP-1 splicing and PPARα expression caused by S6K2 knockdown was suppressed by IRE1α inhibition with Kira6 in a dose-dependent manner, suggesting that PPARα could be a key effector downstream of ER stress (Fig. 4F, fig. S3B). Since XBP-1 directly binds to and activates PPARα and PPARγ (40, 41), we assessed the expression of the PPAR gene family following S6K2 knockdown (Fig. 4G, figS3C). PPARA expression was consistently elevated by S6K2 depletion in MAPKi-R cells, whereas PPARB/D expression did not substantially change and PPARG was upregulated in a single cell line. Further, S6K2 depletion led to upregulation of representative PPAR target genes involved in lipid metabolism (Fig. 4B and H) and RPS6KB1 (fig. S3C) in MAPKi-R cell lines. Altogether these data suggest that S6K2 depletion causes ER stress and that the IRE1α-XBP-1-PPARα signaling network contributes to cell death triggered by S6K2 depletion.

Fig. 4. S6K2 depletion activates a terminal UPR and PPARα.

(A and B) M93–047 cells were transduced with S6K1 shRNA, S6K2 shRNA or empty vector (EV) and analyzed by RNA sequencing (3 days after transduction). (A) Genes significantly (p<0.05) altered by S6K1/2 depletion were analyzed by IPA. Heatmap depicts z-scores for predicted activity of UPR regulators, PPAR genes, and PPAR modulators. Positive/red: activated; negative/blue: inhibited by S6K1 or S6K2 shRNA; ns: no significant enrichment. (B) Heatmap depicts mRNA expression of selected PPAR targets relative to the mean across samples; p values were calculated with Student’s t-tests. (C and D) ER stress markers were monitored following transduction with shS6K2 or EV by immunoblotting (C; M93–047, 3 days after transduction) or qRT-PCR (D; M93–047, S6K2 sh4 or sh6, 3 or 4 days after transduction, respectively; WM1366 4 days after transduction). (E) M93–047 cells transduced with S6K2 shRNA or EV were treated with the IRE1α inhibitor Kira6 or the pan-caspase inhibitor zVAD 1 day after transduction. Cell death was analyzed by Annexin V/PI staining 5 days after transduction (mean ± SD, n=4; unpaired two-sided t-test). (F) M93–047 cells transduced with EV or shS6K2 were treated with the IRE1α inhibitor Kira6. mRNA expression of spliced XBP1 and PPARA were quantified by qRT-PCR (3 days post-transduction for sh4 or 4 days-post transduction for sh6). (G and H) PPAR gene family (G) and PPAR targets (H) mRNA expression were determined by qRT-PCR 4 days-post transduction with S6K2 shRNA or EV. For (D, F, G, and H), data are shown as mean ± SD from three independent experiments and analyzed by unpaired two-sided t-tests.

S6K2 negatively regulates PPARα.

We next sought to define how S6K2 regulates PPARα. The transcriptional activity of PPARα is inhibited by the nuclear receptor corepressor 1 (NCoR1) (41), and S6K2 has been implicated in this regulation (42). Hence, we examined whether S6K2 could regulate this interaction in NRAS^MUT melanoma cells using proximity ligation assays (PLA) (43). PLA allows for the visualization and quantification of endogenous protein complexes by detecting interactions between proteins within close proximity (43). PLA supported the existence of a complex between S6K2, PPARα, and NCoR1 in MAPKi-R cells (Fig. 5A to C). S6K2 depletion diminished the interaction of PPARα with its co-repressor NCoR1 in MAPKi-R cells (Fig. 5D and E fig. S4A), indicating that S6K2 blockade is coupled to PPARα activation. In contrast, MAPKi-S cells displayed a weak and diffuse PPARα-NCOR1 interaction, which was not disrupted by S6K2 depletion (Fig. 5E and F, fig. S4B). These data support a model whereby S6K2 regulates the interaction between NCoR1 and PPARα; S6K2 depletion disrupts this complex leading to activation of the PPARα axis, thereby facilitating transcription of genes involved in lipid peroxidation and oxidative cell death in MAPKi-R NRAS^MUT melanoma.

Fig. 5. S6K2 negatively regulates PPARα.

(A and B) Interaction of S6K2 with PPARα (A) or NCoR1 (B) was determined by PLA. Representative images are shown; scale bar, 30 μM. (C) Bar graphs show quantification of PLA signals for A (top) and B (bottom); data are presented as mean ± SD (n denotes fields quantified), and analyzed by unpaired, two-sided t-tests. S6K2 Ab, PPARα Ab or NCoR1 Ab indicates single antibody controls. (D) MAPKi-R M93047 cells transduced with lentiviruses carrying S6K2 shRNA were analyzed by PLA 4 days after transduction. Representative images are shown; scale bar, 30 μM. (E) Bar graphs show quantification of PLA signals for MAPKi-R, M93–047 (top) and MAPKi-S, WM3451 (bottom); data are presented as mean ± SD (n denotes fields quantified) and analyzed by unpaired, two-sided t-tests. (F). MAPKi-S WM3451 cells transduced with lentivirus carrying S6K2 shRNA (F) were analyzed by PLA (4 days after transduction). Representative images are shown; scale bar, 30 μM.

PUFAs potentiate the oxidative cell death inducing effect of PPARα agonists.

Based on our data indicating that PPARα facilitates lethal lipid peroxidation, we next asked if PPARα agonists could induce anti-melanoma effects. We noted that PPARα protein expression was higher in MAPKi-R NRAS^MUT melanoma cells (Fig. 6A and fig. S5A) and that MAPKi-R cells were more sensitive to the PPARα agonist fenofibrate (FNB) than MAPKi-S/PPARα-low cells (Fig. 6B and fig. S5B). FNB is a weak PPARα agonist, (44) and has been shown to modify lipid and lipoprotein composition and metabolism by a variety of mechanisms (45). Treatment with FNB alone did not induce cell death in MAPKi-R cells (Fig. 6C). Since our lipidomic data revealed S6K2 depletion induces accumulation of PUFAs (Fig. 3B), we wondered whether addition of docosahexaenoic acid (DHA) as a source of PUFAs could potentiate the effect of FNB in inducing lethal lipid peroxidation and cell death. To this end, we treated MAPKi-R and MAPKi-S cells with FNB with or without DHA. Whereas neither FNB nor DHA as single agents induced substantial cell death, the combination of FNB + DHA triggered significant cell death in MAPKi-R cells (p=1.02E-5, 1.61E-5), but not in MAPKi-S NRAS^MUT melanoma cells (p=0.399, 0.590) (Fig. 6C). Similar to S6K2 depletion, the FNB/DHA combination induced ROS (fig. S5C). Consistent with the effects of S6K2 depletion, treatment with FNB+DHA triggered lipid peroxidation selectively in MAPKi-R cells (Fig. 6D and E). Further supporting the notion that the FNB/DHA combination phenocopies S6K2 depletion, the pan-caspase inhibitor, zVAD, and IRE1α inhibitor, Kira6, attenuated FNB/DHA-induced cell death in MAPKi-R cell lines (Fig. 6F, fig. S5D). Additionally, the FNB/DHA combination induced both spliced XBP-1 and PPARα (Fig. 6G and H, fig. S5E), and cotreatment with Kira6 suppressed the activation of XBP-1. Moreover, the FNB/DHA combination induced cell death in MAPKi-R 3D-melanoma spheroids (Fig. 6I). Taken together, these results indicate that MAPKi-R melanoma cells are sensitive to pharmacological activation of PPARα in combination with enhanced fatty acid desaturation, which concomitantly trigger lethal lipid peroxidation.

Fig. 6. PUFAs potentiate the oxidative cell death inducing effect of PPARα agonists.

(A) Plots depict relative expression of PPARα protein in MAPKi-R (red, n=8) and MAPKi-S (blue, n=7) cells determined by immunoblotting (see fig. S5A) and compared by unpaired, two-sided Student’s t-test. Data represent average PPARα expression from two independent experiments. (B) Plots depict sensitivity of MAPKi-R (red, n=8) and MAPKi-S (blue, n=7) cells to the PPARα agonist FNB, calculated from the area under the curve (AUC) of dose-response curves (fig. S5B) and compared using Studenťs t-test. (C) Cells were treated with 50 μM FNB, 7.5 μM DHA, as single agents or in combination for 72 h. Cell death was assessed by Annexin V/PI positivity (mean ± SD; n=3; unpaired two-tailed Students t-test). (D and E) Cells were treated as in (C) and lipid peroxidation was assessed by BODIPY C11. Shown are representative histograms (D) and data from three independent experiments comparing mean fluorescence intensity (MFI) of treated cells relative with vehicle control (E). (F) M93–047 cells were pre-treated with the indicated compounds [pan-caspase inhibitor (20 μM zVAD), UPR/IRE1α inhibitor 0.2 μM Kira6 for 48h)] and then treated with 50 μM FNB, 7.5 μM DHA for 24 h. Cell death was analyzed by AnnexinV/PI positivity. Graphs show mean ± SD (n=3), individual dots represent independent experiments; data were analyzed by unpaired two-sided t-test. (G and H) M93–047 cells were treated with 0.2 μM Kira6 for 48 h and then treated with 50 μM FNB + 7.5 μM DHA for 24 h. Spliced XBP-1 mRNA expression was quantified by qRT-PCR (G) or immunoblotting (H). Data in (G) are presented as mean ± SD (n=3) and were analyzed by unpaired two-sided t-test. (I) MAPKi-R M93–047 and WM1366 cells grown as collagen-embedded 3D spheroids were treated with vehicle or 50 μM FNB, 7.5 μM DHA for 5 days. Spheroids were stained with Calcein (AM) (green; live cells) and EtBr (red; dead cells) and imaged with a fluorescence microscope. Representative images of three replicates are shown; the scale bar represents 1000 μm.

MAPKi-resistant NRAS^MUT melanoma is sensitive to the combination of PPARα agonists and PUFAs.

We next evaluated the efficacy of combining PPARα agonists and DHA in vivo. To investigate this, we implanted MAPKi-R NRAS^Mut human melanoma cells into immunodeficient mice. Additionally, we implanted MAPKi-R syngeneic mouse tumors into immunocompetent mice, allowing us to test the combination treatment in the context of a fully functional immune system. When tumors reached approximately 100mm³, mice were treated with FNB and DHA as single agents or in combination. Combining FNB with DHA suppressed the growth of established NRAS^MUT MAPKi-R cell-derived xenografts (Fig. 7A) and MAPKi-R LSL-Nras^Q61R/Q61R (TpN61R/61R; WHN89) transplanted syngeneic tumors (Fig. 7B) (46, 47) with no appreciable toxicity (fig. S6A). Furthermore, we evaluated the FNB + DHA combination in NRAS^MUT PDX models. To do this, we first assessed the response of the PDXs to MEKi and classified them as MAPKi-S if treatment with MEKi led to complete tumor growth inhibition, or as MAPKi-R if MEKi treatment had minimal to partial effect (fig. S6B to D). FNB+DHA significantly increased the survival of mice bearing MAPKi-R PDXs (p=0.0262, 0.0210) (Fig. 7C and D, fig. S6B and C) but not in MAPKi-S PDX (Fig. 7E, fig. S6D). Importantly, FNB+DHA induced substantial lipid peroxidation and oxidative damage in MAPKi-R PDXs, but not in MAPKi-S cell-derived xenografts (Fig. 8A to D, fig. S7A and B).

Fig. 7. MAPKi-resistant NRAS^MUT melanomas are sensitive to combination treatment with PPARα agonists and PUFAs.

(A) Mice bearing MAPKi-R M93–047-derived subcutaneous tumors were treated with FNB (200 mg/kg) or DHA (300 mg/kg) as single agents or in combination. Treatment was initiated once tumors reached approximately 100mm³, which was approximately 12 days after tumor cell inoculation. Tumor volume for individual mice (gray lines) or the mean (black lines, n=8–10) is shown. (B) C57BL/6j mice bearing MAPKi-R LSL-Nras^Q61R/Q61R (TpN^61R/61R; WHN89) syngeneic tumors were treated with vehicle, FNB (200 mg/kg, n=9), DHA (300 mg/kg, n=8) or FNB + DHA combination when tumors reached approximately 100 mm³ (n=8/group, approximately 19 days after inoculation). For (A and B), p-values and FDR adjusted p-values were estimated from a linear mixed-effect model with all follow-up subjects and time points. (C to E) Mice bearing PDXs were fed control chow or DHA (0.15% w/w) plus FNB (0.15% w/w) laced chow when tumors reached approximately 100 to 150 mm³. WM4023 (control n=5, DHA+FNB n=4; C) and WM4299–1 (control n=5, DHA+FNB n=5; D) were MAPKi-R. WM4319 (control n=4, DHA+FNB n=4; E) was MAPKi-S. Mice were followed until tumors reached a pre-defined volume (1200 mm³). P-values were calculated using log-rank (Mantel-Cox) test. n.s., not significant.

Fig. 8. Combined PPARα agonist and PUFA treatment induce lipid peroxidation and oxidative damage in melanoma models.

(A to D) Tumors derived from mice bearing MAPKi-R WM4023 (PDX), or MAPKi-S WM3623 (cell-derived xenograft) treated with FNB (200 mg/kg) or DHA (300 mg/kg) as single agents or in combination for 7 days were analyzed by immunohistochemistry for 4-HNE (A and B) or 8-OHdG (C and D). Representative images are shown in (A and C); scale bar, 25 mm. Tumors were scored using QuPath software and quantification is presented in (B and D). Bar graphs show H-score from representative fields (mean ± SD; WM4023 n=2, WM3623 n=3).

These results provide proof-of-principle that combining PPARα agonists with PUFAs facilitates lipid peroxidation, elicits anti-tumor activity, and restrains MAPKi-R NRAS^MUT melanoma. By selectively blocking S6K2 or harnessing the S6K2 subnetwork, we have identified a strategy that could be potentially exploited as an anti-tumor approach in NRAS^MUT melanoma.

DISCUSSION

Despite crucial advances in treating melanoma, effective therapies for NRAS^MUT tumors are sorely needed. One approach to inhibiting NRAS that has been investigated is targeting its proximal downstream effectors, mainly the MAPK and PI3K pathways. Although inhibiting either pathway alone is barely effective in NRAS^MUT tumors, inhibiting both pathways leads to unacceptable toxicities in patients. We hypothesized that identifying and inhibiting convergent subnetworks, within the MAPK and PI3K super-networks, could spare the toxicities while triggering death of cancer cells. We therefore performed a deep analysis of the signaling networks in MAPKi-resistant NRAS^MUT melanoma. By comparing the effects of MAPK inhibition in MAPKi-S versus MAPKi-R NRAS^MUT melanoma, we mapped a critical subnetwork downstream from the convergence of PI3K and MAPK. We identified a S6K1/2-PPARα subnetwork that is vital for cancer lipid metabolism. We uncovered that S6K1/2 is regulated by both the MAPK and PI3K pathways in MAPKi-S cells; in contrast, S6K1/2 is regulated by PI3K in MAPKi-R NRAS^MUT melanoma. Further, we found that uncoupling S6K1 and S6K2 by selective S6K2 depletion triggered a lipid metabolic imbalance featuring ER stress and lipid peroxidation, which led to cell death preferentially in MAPKi-R NRAS^MUT melanomas (fig. S7C).

Both the IRE1α-XBP-1 and PERK-CHOP pathways are implicated in initiating ER stress-induced apoptosis(48). Upon S6K2 knockdown, ER stress predominantly activated the IRE1α-XBP-1 pathway. Moreover, cell death was suppressed by the IRE1α inhibitor Kira6, indicating that IRE1α-XBP-1 signaling plays a key role in mediating cell death elicited by S6K2 suppression. Our findings underscore that MAPKi-resistant and -sensitive cells have different signaling dependencies both upstream and downstream of mTORC1 and illustrate specialized functions of S6K isoforms, highlighting the biological importance of isoform differentiation. Moreover, the identified subnetwork provides a strategy to pharmacologically induce lethal lipid peroxidation in vivo, which has been limited by the lack of compounds with sufficient bioavailability(33, 49–51). Of note, S6K2 knockout mice are viable and develop normally(21), indicating that S6K2 is not an essential gene. Collectively, these data suggest that selective targeting of S6K2 or its effectors represents an opportunity for therapeutic intervention in S6K2-dependent tumors such as MAPKi-R NRAS^MUT melanoma.

Our study establishes that selectively depleting S6K2 induces ER stress and lethal lipid peroxidation in MAPKi-R NRAS^MUT melanoma. This suggests that selective inhibition of S6K2 could curb melanoma through a mechanism unachievable with PI3K or mTOR inhibitors, which inhibit both S6K1 and S6K2. However, selective S6K2 inhibition has not been explored in pre-clinical melanoma models and only one S6K2-selective chemical probe has been reported (52). We further identified actionable targets whose modulation could be a surrogate for S6K2 inhibition and provide proof-of-concept for this strategy. The combination of FNB (PPARα agonist) plus DHA (PUFA) recapitulated the activation of ER stress and XBP-1-PPARα signaling observed with genetic depletion of S6K2, promoting lipid peroxidation and cell death. Importantly, both FNB and DHA are compounds that are used in humans with manageable toxicity profiles and, therefore, could be incorporated into new treatments for melanoma. Of note, MAPKi-R and MAPKi-S NRAS^MUT melanomas exhibit different degrees of PPAR expression, which correlate with sensitivity to MAPKi and PPARα agonists. This provides a context in which our proposed approach would be most effective. In addition to the effects of FNB on tumor lipid metabolism and homeostasis, PPARα agonists can also modulate inflammation and alter the immune cell profile within the tumor microenvironment (53, 54). For example, PPAR signaling is implicated in promoting interferon-γ production by natural killer cells (54, 55). Furthermore, FNB improves the efficacy of CD8⁺ T cell therapy for PDX melanoma mouse models (56). Similarly, DHA has been shown to exert antitumor activity by enhancing natural killer cell effector functions in B16F10 melanoma models (57). In addition, DHA exerts anti-melanoma effects by inducing apoptosis, inhibiting tumor growth, and reducing metastatic potential (58–60). The effect of DHA on Microphthalmia-associated transcription factor (MITF), a lineage-specific transcription factor which has been linked to drug resistance in BRAF-mutant melanoma, is intriguing. MITF functions as a rheostat governing a phenotypic switch, with low expression promoting a slow-cycling, drug-resistant state and high expression characteristic of a more proliferative, differentiated phenotype (61). DHA can upregulate MITF expression, whereas FNB downregulates MITF (62, 63). This raises the possibility that combining FNB with DHA could stabilize MITF, potentially mitigating its impact on drug resistance.

Previous studies from our group and others have linked enhanced or persistent mTORC1/S6K1 signaling with acquired resistance to MAPKi in BRAF-mutant melanoma (7, 15, 64) and to MEKi and cyclin-dependent kinase 4 (CDK4) inhibition (CDK4i) in NRAS^MUT melanoma (55, 65). Inhibition of the mTORC1/S6K1 axis restrains MAPK-dependent melanomas and tumors with acquired resistance to MAPKi or CDK4i. In contrast to strategies relying on mTOR1/S6K1 inhibition, we exploited active lipid metabolism to induce cell death (67) by disrupting the S6K1/S6K2/PPARα subnetwork through selective S6K2 blockade. We noted that, whereas S6K2 blockade led to enhanced expression of proteins involved in lipid synthesis and accumulation of PUFAS, inhibition of S6K1 minimally perturbed lipid homeostasis. These data support the notion that selective S6K2 inhibition is required to induce lethal peroxidation in NRAS^MUT melanoma. Inducing severe ER stress and lethal lipid peroxidation provides a therapeutic opportunity to offset drug resistance. Previous studies have shown that MAPKi-R melanomas display elevated oxidative stress and are vulnerable to compounds that enhance ER stress (68, 69). Moreover, activation of the UPR can promote anti-tumor immunity in a context-dependent manner (70, 71). Likewise, tumors characterized by distinctive metabolic states such as those displaying enhanced PUFA biosynthesis are innately vulnerable to lipid peroxidation (72, 73).

Our study has some limitations. First, whereas S6K2 expression is higher in MAPKi-R than in MAPKi-S tumors, the upstream regulators of S6K2 in NRAS^Mut melanoma remain unidentified. Additionally, while the FNB/DHA combination mimics the effects of S6K2 depletion, it may also induce pleiotropic effects, potentially leading to other physiological changes. Unfortunately, selective inhibitors of S6K2 are not currently available. The development of such inhibitors will allow for more targeted investigations to better understand the role of S6K2 in NRAS^Mut melanoma and treatment responses.

In sum, our studies establish S6K2 and its effectors as a vulnerability in NRAS^MUT melanomas that are resistant to MAPKi. Our results underscore the importance of oncogenic NRAS-induced metabolic dependency and the role of S6K2 in supporting this addiction. We propose that harnessing this S6K2-dependent addiction could be exploited as a strategy to combat MAPKi-resistant NRAS^MUT melanoma, a highly aggressive tumor type with limited treatment options. Future studies using S6K2 inhibitors will be crucial to fully evaluate the therapeutic potential of targeting S6K2 in melanoma, particularly in the context of drug resistance. Furthermore, investigating the immunomodulatory effects of FNB/DHA or S6K2 inhibition could provide valuable insights into their potential to overcome resistance and enhance anti-tumor immune responses, paving the way for more effective melanoma treatments.

MATERIALS AND METHODS

Study design

The objective of this study was to identify therapeutic vulnerabilities in MAPK inhibitor-resistant NRAS^MUT melanoma by using patient-derived cell lines and in vivo models. We combined high-throughput, genetic, and biochemical approaches to compare MAPKi-R and MAPKi-S melanoma models. Studies were conducted using techniques such as conventional immunoblotting, flow cytometry, RNA-seq, RPPA, global proteomics, and lipidomics in melanoma cells and mouse models to analyze treatment effects. All animal studies were approved by the Institutional Animal Care and Use Committee (IACUC) at the Wistar Institute and performed in accordance with institutional guidelines. Sample size for in vivo experiments was determined using power analyses based on pilot studies or previous studies and was sufficient to detect statistically significant differences between treatment groups. Mice were randomly assigned to different treatment groups without blinding. All data were included in the final analysis. In vitro experiments were performed in biological triplicates, except where otherwise specified. The number of biological replicates, statistical methods applied, and p-values are provided in the figure legends. Investigators were not blinded to the experimental conditions. Compounds used in this study were sourced from suppliers specified in table S3.

Melanoma cell and 3D spheroid culture

Melanoma cell lines and 293T cells were grown in RPMI-1640 supplemented with 5% or 10% fetal bovine serum (FBS), respectively (Tissue Culture Biologicals (#101; Lot#170708, 105094, 101258, 101929) or Clontech (#631107; Lot#A301117007)). Melanoma spheroids were generated and stained for live/dead cells as previously described (47) with the following modifications: spheroids were embedded into a collagen mixture (10% RPMI-1640, 1.5 mg/ml collagen, 10% FBS, and 7.5% NaHCO₃) and treated with dimethyl sulfoxide (DMSO) or 50 μM FNB + 7.5 μM DHA.

Animal Studies

Mice were housed in groups of five/cage in an AAALAC certified facility. NOD/LtScidIL2Rg-null (NSG) mice (4 to 8-week-old, approximately 25–30 g) were procured from The Wistar Institute. C57BL/6j mice (4-week-old, approximately 25 g) were obtained from Charles River (Strain code 027). For in vivo xenograft studies, 1 x 10⁶ cells were resuspended in 1:1 RPMI-1640/Matrigel (Matrigel Matrix, Corning #354230) and implanted subcutaneously into the flanks of male NSG mice. Patient-derived xenografts were minced, mixed with Matrigel and implanted into male NSG mice. For the syngeneic mouse model, Nras^mut tumors (WHN89) were derived from Tyr-CRE-ER^T2 p16^L/L LSL-Nras^Q61R/Q61R (TpN^61R/61R) mouse model (46). TpN^61R/61R tumors were minced and implanted into male C57BL/6j mice. Tumor volume was measured using digital calipers and calculated by length x width²/2. Mice in all treatment groups were euthanized when the vehicle-treated group reached the humane endpoint (tumor volume 1500 to 2000 mm³).

To determine MEKi sensitivity in melanoma PDXs, tumors were minced and subcutaneously injected into 8-week-old male NSG mice. When tumors reached 100 to 150 mm³, mice were randomly assigned into vehicle control or MEKi group. The MEKi PD0325901 was prepared in 0.2% Tween 80 + 0.5% hydroxypropyl-methyl cellulose + 3% DMSO in ddH2O and administered by oral gavage (1.5 or 5 mg/kg/day).

For experiments testing FNB and DHA treatment, tumor cells were subcutaneously injected into NSG mice. Syngeneic TpN^61R/61R tumor cells (WHN89) were subcutaneously injected into C57Bl/6j male mice. When tumors reached approximately 100 to 200 mm³, mice were randomly assigned into experimental groups and treated with vehicle, FNB, DHA, or FNB and DHA. FNB was prepared in 4% DMSO + 40% polyethylene glycol (PEG) 300 + 2% Tween 80 in ddH2O and administered by oral gavage (200 mg/kg/day). DHA was administered by oral gavage (300 mg/kg/day) in soybean oil. Mice bearing PDXs were given control chow or DHA (0.15% w/w) plus FNB (0.15% w/w) laced chow (BioServ).

Immunoblotting Cells were lysed with RIPA buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 1 mM EDTA, 2.5 mM sodium pyrophosphate, 0.05% or 0.1% SDS) supplemented with sodium vanadate (0.2 mM) and protease inhibitor cocktail (Sigma-Aldrich #11697498001). Immunoblotting was performed as previously described; primary and secondary antibody concentrations are listed in table S4 (47).

Flow cytometry

Cells were fixed in 90% ice cold EtOH in phosphate buffered saline (PBS). Cells were washed twice with PBS followed by 1 h incubation at room temperature with incubation buffer [1% bovine serum albumin (BSA) in PBS] containing primary antibodies (table S4). Samples were washed once (1 x PBS), incubated for 30 min at room temperature with incubation buffer containing secondary antibodies diluted to 1:500 (table S4), washed (1 x PBS) and resuspended in incubation buffer for analysis by flow cytometry (BD LSRII 14-color flow cytometer; Alexa Fluor 488 and allophycocyanin (APC) filter sets). Data analyses were performed using FlowJo software to calculate pS6 positivity. For 4-HNE staining, cells were fixed with 4% paraformaldehyde for 15 min, washed with PBS and stored at 4°C. Before immunostaining, cells were permeabilized with 0.5% Triton X-100/PBS for 10 min.

Immunohistochemistry (IHC)

Tumors were fixed with 10% buffered formalin, embedded in paraffin and cut into 4 mm consecutive sections. Deparaffinized sections were steamed in citrate buffer (pH 6) for 20 min or incubated in pressure cooker with DAKO EDTA (pH9, Agilent Dako #S2367) for 20 min to retrieve antigens. Samples were treated with 3% H₂O₂ for 10 min, then 0.3% Triton X-100/PBS for 15 min at room temperature. Samples were blocked with 5% horse serum in PBS for 1h at room temperature, incubated overnight at 4°C with primary antibodies (table S4) in 4% BSA/PBS, and then incubated with secondary antibodies for 30 min at room temperature. All wash steps were performed with 0.5% PBST. Antigens were detected using diaminobenzidine (DAB). Slides were counterstained with hematoxylin. Immunohistochemistry images were acquired using a Nikon 80i upright microscope.

Proximity ligation assay

Cells were grown on chamber slides (Lab-Tek II #154534), fixed with 4% paraformaldehyde for 10 min and washed three times with PBS. Proximity ligation assay was performed to assess protein-protein interactions using Duolink In Situ Red Starter Kit Mouse/Rabbit (Sigma-Aldrich #DUO92101) according to the manufacturer’s protocol, using primary antibodies listed in table S4. Slides were mounted with media containing DAPI and images were captured and analyzed on a Leica TCS SP5 II scanning spectral confocal microscope.

Plasmids and lentivirus

Packaging plasmids pS-PAX2 and pMD2.G (Didier Trono; Addgene plasmid #12260, #12259) were used to transfect 293T cells and produce lentiviral particles. Melanoma cells were transduced with lentivirus and selected with the appropriate antibiotics as previously described (74). HA-tagged S6K1^T389EΔCT in pSLIK inducible vector was a gift from Kevin James (G418 selection, 2mg/mL; Addgene #58516). Protein expression was induced with doxycycline (0.25 mg/mL). FLAG-tagged S6K1^T389S in pLJM vector was a gift from David Sabatini (puromycin selection, 2 mg/mL; Addgene #48800). HA-tagged S6K2^T388E (Addgene#17731) was subcloned into pLUT-poly inducible vector (puromycin selection, 2 mg/mL; gift from Meenhard Herlyn, The Wistar Institute). The following short hairpin RNAs were used for S6K1 and S6K2 in pLKO.1 vector: S6K1 shRNA (sh1: TRCN0000003158, sh4: TRNC0000003161, sh5: TRNC0000003162). S6K2 shRNA (sh1: TRNC0000000729, sh4: TRNC0000010540, sh5: TRCN0000010541, sh6: TRCN0000199513, sh7: TRCN 0000432273).

Quantitative reverse transcription polymerase chain reaction (qRT-PCR)

Cell pellets were homogenized using QIAshredder (QIagen #79656) and RNA was extracted with PureLink RNA mini kit (Thermo Fisher Scientific, #12183018A). DNA was removed by on-column PureLink DNase treatment (Thermo Fisher Scientific, #12185010). First-strand cDNA was synthesized using Maxima cDNA kit (Thermo Scientific #K1642). qRT-PCR was performed using SYBR green PCR Master Mix (Applied Biosystems #4385612). Tumor samples for qRT-PCR were preserved in RNAlater solution (Invitrogen #AM7020). Primer sequences are specified in table S5.

Cell viability assays

For the MTT assays, cells were seeded in 96-well plates (2500 cells per well) and treated with drugs by adding 2X compounds (prepared in fresh growth medium). After 72 h, cell viability was determined by a standard MTT assay. Absorbance was determined at 490 nm using a BioTek plate reader. For the CellTiter-Glo assay, cells were seeded in 384-well plates (500 cells per well) and treated with drugs using the Janus MDT Nanohead. Viability was measured after 72 h using CellTiter-Glo (Promega). Data were normalized where 100% viability equals luminescence of DMSO-treated cells, and 0% viability equals luminescence of 1 mM bortezomib-treated cells.

Cell proliferation assay

BrdU incorporation was performed per the manufacturer’s instructions (BrdU Cell Proliferation Assay Kit; Cell Signaling Technology #6813). Briefly, cells were seeded in 96-well plates (2500 cells per well) one day before drug treatment. Cells were treated with drugs for 72 h, followed by BrdU labelling for 16 h and 30 min incubation in fixing/denaturing solution at room temperature. Cells were then incubated with primary antibody (table S4) for 1 h at room temperature, washed three times followed by incubation with a horseradish peroxidase (HRP)-conjugated secondary antibody solution at room temperature for 30 min. Colorimetric reactions for BrdU incorporation were started by incubation with 3,3,5,5-Tetramethylbenzidine (TMB) substrate (10 min at room temperature) and terminated by adding STOP solution; signal was determined by absorbance at 450 nm.

Cell death assay

Floating and attached melanoma cells were harvested, washed once with PBS, and stained with PSVue 643 (5 min at room temperature) and propidium iodide (PI, 10 min at room temperature) or Annexin V-fluorescein isothiocyanate (FITC), and PI (10 min at room temperature). Annexin V staining was performed as described previously (47).

Measurement of ROS and lipid peroxidation

Melanoma cells were trypsinized, washed with PBS, and incubated with 10 mM H2DCFDA (2',7'-dichlorodihydrofluorescein diacetate) in PBS (25 min at 37°C). Cells were then washed once with PBS, incubated in phenol red-free growth medium (10 min at 37°C) and analyzed by flow cytometry. For lipid peroxidation, cells were harvested and incubated with 5 mM BODIPY C11 in growth medium for 30 min at 37°C. Cells were washed 5 times with PBS and analyzed by flow cytometry (BD LSRII 14-color flow cytometer). Data analyses were performed using FlowJo software.

Proteomics sample preparation and liquid chromatography-mass spectrometry/mass spectrometry (LC-MS/MS)

M93–047 cells were transduced with vector control, S6K1 shRNA (sh1/4 or sh5) or S6K2 shRNA (sh1 or sh4), lysed four days post-transduction with SDS buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% SDS and 1 mM EDTA) supplemented with protease inhibitors. In-gel trypsin digestion was performed using 15 mg of lysate from triplicates as described previously (67). LC-MS/MS was performed using a Waters nanoACQUITY UPLC in-line with a Thermo Q Exactive Plus mass spectrometer. For each sample, 1 mg of tryptic digest was loaded onto a 180 mm x 2 cm nanoACQUITY UPLC Symmetry C18 trap column with 5 mm particle size (Waters #186006527) followed by analytical separation on a 1.7 mm x 2.5 cm ACQUITY UPLC Peptide BEH C18 column with 1.7 μm particle size (Waters #186003546). Samples were analyzed in data-dependent mode using a 245 min LC gradient with water (solvent A) and acetonitrile (solvent B) containing 0.1% formic acid: 5–30% B over 225 min, 30–80% B over 5 min, 80% B hold for 10 min and return to initial conditions for 5 min. Full MS spectra were recorded at a resolution of 70,000 using a 400–2000 m/z scan range. Data-dependent MS/MS was performed on the top 20 most abundant ions selected with an isolation width of 1.5 m/z and recorded at a resolution of 17,500.

Proteomics LC-MS/MS data analysis

RAW files were processed using MaxQuant 1.5.1.2 in a single run (68). Database searches were performed against the UniProt human sequence database (June 29, 2017; 159,819 sequences) and an in-house database of common laboratory contaminants (July 28, 2014; 3,671 sequences) with Trypsin/P specificity, a maximum of 2 missed cleavages and a minimum peptide length of 7 residues. Precursor and fragment mass tolerances were set to 4.5 ppm and 20 ppm, respectively. Protein N-terminal acetylation and methionine oxidation were set as variable modifications. Cysteine carbamidomethylation was set as a fixed modification. A maximum of 5 modifications were allowed per peptide. Peptide and protein false discovery rates were both set to 1%. Match between runs was enabled with a 0.7 min match time window and 20 min alignment time window. Protein quantification was based on unique peptides only. Label-free protein quantitation (LFQ) to normalize between samples was performed with a minimum peptide ratio of 1. Statistical analysis was performed using Perseus 1.5.0.31 (69). Protein groups identified by a single unique peptide or corresponding to contaminants were removed. LFQ intensities were log2 transformed to reduce outlier effects, and missing values were imputed from a downshifted normal distribution. Statistical significance between conditions after pooling data from the two different shRNAs for the same S6K was defined as Student’s t-test p-value less than 0.05 and absolute fold-change greater than 2 based on log2-transformed LFQ intensities. Where indicated, correction for multiple testing was performed by permutation-based false discovery rate (FDR). Canonical pathway analysis was performed using QIAGEN’s Ingenuity Pathway Analysis software (IPA) with all identified proteins as the reference dataset.

Lipidomics sample preparation and LC-MS/MS

M93–047 cells were transduced with vector control, S6K1 shRNA or S6K2 shRNA. Four days post transduction, lipids were extracted using a modified Folch method (63, 70). Cells were washed twice with PBS, quenched with ice-cold MeOH and transferred to a glass tube. Splash (Avanti Polar Lipids #330707) internal standard was added to each sample. Ice-cold 0.88% NaCl in water and CHCl₃ were added to a final ratio of 2:1:1 CHCl₃:MeOH:0.88% NaCl. Extracts were then sonicated and centrifuged. The lower organic phase was transferred to a new glass tube. The upper aqueous phase was re-extracted with synthetic organic phase. The organic phases were pooled and dried under nitrogen, before reconstitution in 1:8:1 CHCl₃:MeOH:H₂O. Global lipid extracts from biological triplicates were analyzed by reversed-phase LC-MS/MS on a Thermo Q Exactive HF-X in both positive and negative polarities. For each polarity, 1% (2 mL) of the resuspended lipidome was loaded onto a 150 mm x 2.1 cm Accucore C30 column with a 2.6 mm particle size (Thermo Fisher Scientific #27826–152130). The lipidome was analyzed using data-dependent acquisition on a 40-minute gradient with 1:1 acetonitrile:water with 5 mM ammonium formate and 0.1% formic acid (solvent A) and 88:10:2 isopropanol:acetonitrile:water with 5 mM ammonium formate and 0.1% formic acid (solvent B). The gradient is as follows: 0–60% B over 10 minutes, 60–85% B over 10 minutes, 85–100% over 10 minutes and holding for 5 minutes and re-equilibrating to initial conditions over 5 minutes. Full MS spectra were recorded at a resolution of 120,000 using a 300–1,200 m/z scan range in positive mode and 250–1,200 m/z scan range in negative mode. Data-dependent MS/MS scans were recorded at a resolution of 15,000 for the top 20 most abundant ions isolated with a 0.4 m/z isolation width. Stepped collision energy (NCE) of 20/30 was used in positive mode and 20/30/40 was used in negative mode.

Lipidomics data analysis

Global lipids quantitation was performed using Lipid Search 4.1 (Thermo Fisher Scientific). Raw files were processed in a single batch. Lipids were identified based on mass, retention time and predicted fragmentation patterns. All identified lipids were aligned and quantified with a 5 ppm mass tolerance and a retention time window of 0.2 minutes. Low scoring lipid identifications and unlikely adducts were filtered out based on standards. To quantify lipid distribution, the abundance (peak area) of each lipid species in a sample was normalized to total lipid peak area in that sample. Statistical significance was defined as a Student’s t-test p-value less than 0.05.

RNA-seq

Cells were lysed in Tri-Reagent (Sigma-Aldrich, #T9424) and RNA extracted using the Direct-zol MiniPrep Kit (Zymo Research, #R2050) with on-column DNAse I treatment. Total RNA (200ng) was used to prepare 3’ QuantSeq mRNASeq libraries (Lexogen, #KO152x96). Fourteen PCR cycles were used for library amplification. Library qualities were checked using the TapeStation High Sensitivity D5000 ScreenTape (Agilent Technologies, #5067–5592) and quantified by quantitative polymerase chain reaction (Roche #KK4835). Sequencing was performed on the NextSeq 500 (Illumina) using a 75 cycle High Output sequencing kit. A final concentration of 1.9pM was loaded onto the flowcell.

RNA-seq data was aligned using bowtie2 (75) against hg19 version of the human genome and RSEM v1.2.12 software (76) was used to estimate raw read counts and reads per kilobase of transcript per million mapped reads (RPKM) using Ensemble transcriptome. DESeq2 (77) was used to estimate significance of differential expression between each shRNA and controls. Genes with expression changes passing a false discovery rate (FDR) threshold of 5% were considered significantly differentially expressed. Only genes that overlapped between the two shRNAs for S6K1 (sh1/4 and sh5) or S6K2 (sh1 and sh4) were analyzed. Heatmaps were normalized by DESeq2 log2 scaled counts relative to average across all samples. Gene set enrichment analysis was done using QIAGEN’s Ingenuity Pathway Analysis software and significance of enrichment for upstream regulators was defined at p-value<0.05.

Reverse phase protein array

Cells were lysed with 1% Triton X-100, 50 mM HEPES, pH 7.4, 150 mM NaCl, 1.5 mM MgCl₂, 1 mM EGTA, 10% glycerol, protease and phosphatase cocktail (Sigma-Aldrich #11873580001 and 04906837001). Lysates were denatured in sample buffer (10% glycerol, 2% SDS, 0.06 M Tris-HCl, pH 6.8, 1/40th volume of 2-ME) for 5 min at 95°C. RPPA was performed by the MD Anderson Center RPPA core facility as previously described (78).

Analysis of TCGA data

To determine the relationship between S6K2 and S6K1 expression in human melanoma patient tumors, log2-transformed mRNA expression values for the genes encoding RPS6KB1 and RPS6KB2 were extracted from the The Cancer Genome Atlas Program (TCGA) skin cutaneous melanoma (SKCM) dataset (n=443) through cBioPortal (accessed 5^th Feb 2022). Kaplan–Meier survival curves for patients with melanoma (n=470) were obtained from the TCGA SKCM dataset using R2: Genomics Analysis and Visualization Platform (http://r2.amc.nl) to stratify by upper versus lower quartiles (n=115 each). P values correspond to log-rank tests comparing the two Kaplan–Meier curves.

Statistical analysis

Individual-level data are presented in data file S4. Data were analyzed and plotted using GraphPad Prism or Microsoft Excel. Data are presented as mean ± SD or median with interquartile range as indicated in each figure legend. The number of independent biological replicates (n) is indicated in each figure legend. Statistical differences between two experimental groups were determined using unpaired, two-tailed Studenťs t-tests, unless otherwise specified in the figure legends. Linear mixed effects models were applied to compare tumor growth rates between treatment groups from in vivo tumor growth experiments. For multiple comparisons, an FDR-adjusted p-value <0.05 was considered significant.

DATA AND MATERIALS AVAILABILITY:

All data associated with this study are present in the paper or the Supplementary Materials. Cell lines and PDXs from the Wistar Melanoma collection can be obtained from Rockland or by UBMTA.

All correspondence and requests for materials or reagents can be addressed to JV. Proteomics dataset generated from M93–047 cell line has been deposited to the MassIVE repository (accession # MSV000083480). RNA-seq dataset generated from M93–047 cell line has been deposited into the Gene Expression Omnibus (GEO) repository (accession #GSE127916).

Supplementary Material

Fig. S1 to S7

Table S1 to S5

MDAR Reproducibility Checklist

Data file S1 to S4