Resistance of melanoma to immune checkpoint inhibitors is overcome by targeting the sphingosine kinase-1

Immune checkpoint inhibitors (ICIs) have dramatically modified the prognosis of several advanced cancers, however many patients still do not respond to treatment. Optimal results might be obtained by targeting cancer cell metabolism to modulate the immunosuppressive tumor microenvironment. Here, we identify sphingosine kinase-1 (SK1) as a key regulator of anti-tumor immunity. Increased expression of SK1 in tumor cells is significantly associated with shorter survival in metastatic melanoma patients treated with anti-PD-1. Targeting SK1 markedly enhances the responses to ICI in murine models of melanoma, breast and colon cancer. Mechanistically, SK1 silencing decreases the expression of various immunosuppressive factors in the tumor microenvironment to limit regulatory T cell (Treg) infiltration. Accordingly, a SK1-dependent immunosuppressive signature is also observed in human melanoma biopsies. Altogether, this study identifies SK1 as a checkpoint lipid kinase that could be targeted to enhance immunotherapy.

T he recent success of immunotherapy in melanoma relates to the use of monoclonal antibodies directed against the immune checkpoints such as programmed cell death protein-1 (anti-PD-1; pembrolizumab, nivolumab) and cytotoxic T-lymphocyte-associated protein-4 (anti-CTLA-4; ipilimumab). These immune checkpoint inhibitors (ICI) have demonstrated significant clinical efficacy in metastatic melanoma by reversing effector T-cell dysfunction and exhaustion, thereby enhancing their anti-tumoral properties 1 . However, a proportion of 40-60% of patients do not achieve any significant therapeutic response and a substantial proportion of responder experience tumor relapse within 2 years [2][3][4][5] . Consequently, innovative strategies and synergistic combination therapies, which block more than one immunomodulatory pathway are still needed to eradicate advanced melanomas 6 . A promising strategy is to modulate cancer cell metabolism, for instance sphingolipid metabolism, by targeting key enzymes involved in the production of oncometabolites, such as sphingosine-1-phosphate (S1P) 7 . S1P is produced by sphingosine kinases (SK) that catalyze the phosphorylation of sphingosine to S1P. The SK type 1 isoform (SK1), encoded by the SPHK1 gene, which is overexpressed in numerous human tumors, including melanoma, leads to increased levels of S1P 8,9 . The SK1/S1P axis could modulate different hallmarks of cancer such as cell proliferation, cell death, metastasis and angiogenesis 10,11 . Moreover, S1P is a well-known regulator of lymphocyte trafficking and differentiation under different pathophysiological conditions 12,13 . However, the impact of increased SK1 expression in melanoma cells on the abundance, the functions and the phenotype of tumor-infiltrating lymphocytes (TILs) is unknown. TILs are a heterogeneous population for which frequency, localization, and subset ratio in solid tumors correlate with prognosis and immunotherapeutic responses 14,15 . CD8 + T cells play a central role in anti-tumor immunity whereas accumulation of Foxp3 + regulatory T cells (Treg) dampens effector function. Consequently, the CD8/Treg ratio in the tumor microenvironment (TME) constitutes a critical factor in immunotherapy 16,17 . How tumor cell metabolism, particularly sphingolipid metabolism, modulates this ratio needs further attention.
Here, we observe that high expression of SK1 in tumor cells is associated with shorter survival in melanoma patients treated with anti-PD-1. Interestingly, silencing of SK1 in preclinical models leads to attenuated tumor growth and Treg recruitment, and enhances the CD8/Treg ratio in tumors. Moreover, using epigenetic and pharmacological approaches to target SK1, we show that SK1 expression in melanoma impairs the responses to ICI. Our results demonstrate, that combining ICI and SK1 antagonism may represent the basis for innovative anti-melanoma therapies.

Results
SPHK1 expression inversely correlates with survival after ICI therapy. Analysis of two different cohorts from the Oncomine database indicated that SPHK1 (encoding SK1) transcript levels were higher in human primary melanomas as compared to nevi (Fig. 1a, left panel); SPHK1 expression was further increased in metastatic melanomas (Fig. 1a, right panel), suggesting that SPHK1 expression might be associated with melanoma progression.
In order to evaluate whether SPHK1 expression was related to the therapeutic outcome in advanced melanoma patients receiving anti-PD-1 therapy (Table 1), we analyzed SPHK1 messenger RNA (mRNA) expression in tumor biopsies by in situ hybridization using the RNAscope technology.
Patients were separated in two groups according to the proportion of tumor cells positive for SPHK1 mRNA (Low SPHK1: ≤ 50% tumor cells positive and High SPHK1: > 50% tumor cells positive) (Fig. 1b). Figure 1c shows representative SPHK1 staining for these two groups. Patients with low SPHK1 expression had significantly longer progression-free survival and overall survival than those with high SPHK1 expression (p = 0.0112 and p = 0.0445, respectively; log-rank test) (Fig. 1d, e), and patients with high SPHK1 expression mostly failed to respond to anti-PD-1 therapy. These findings support the hypothesis that SPHK1 expression represents a potential biomarker to predict tumor progression and resistance to anti-PD-1 in metastatic melanoma patients.
SK1 silencing enhances anti-tumor immune response. In order to assess the impact of SK1 expression on melanoma growth, we generated stable SK1 knockdown cells using Yumm 1.7 cells derived from spontaneous murine melanoma driven by Braf activation, as well as Pten and Cdkn2a inactivation 18,19 . This cell line has previously been shown to resist PD-1 blockade 20 . The puromycin-resistant cell lines shSK1(1), shSK1(2) and shSK1 (3), silenced for SK1, were obtained with three different shRNA sequences. shSK1 cells exhibited a markedly reduced enzymatic activity of SK1 ( Fig. 2a and Supplementary Fig. 2a). While SK1 silencing did not modify in vitro Yumm cell proliferation (Fig. 2b), we observed a significant and sustained reduction in tumor growth after intradermal injection in C57BL/6 wild-type (WT) mice ( Fig. 2c and Supplementary Fig. 2a). Interestingly, tumor regression was observed after day 12 for SK1-silenced melanoma cells, suggesting an increased anti-melanoma immune response. In addition, SK1 silencing did not alter Yumm tumor growth in immunodeficient NSG (Fig. 2d). In vivo depletion experiments demonstrated that CD4 and CD8 cells, but not NK cells, were important in the inhibition of tumor growth triggered by SK1 silencing (Supplementary Fig. 2b).
To investigate the impact of SK1 silencing in melanoma cells on the anti-tumor immune response, we analyzed the lymphocyte infiltration of tumors in mice injected either with shCtrl or shSK1 (1) Yumm cells at day 11. Interestingly, SK1 silencing increased the proportion of CD8 + T cells while decreasing that of Foxp3 + CD4 + T cells (Treg), leading to an increased CD8/Treg ratio in shSK1 tumors as compared to shCtrl tumors-injected mice (Fig. 2e, f). Analysis of TIL proliferation, evaluated by Ki67 expression, showed that SK1 silencing markedly increased the percentage of Ki67 + CD8 + T cells and decreased that of Ki67 + Treg (Fig. 3a, b). Similar results were observed in tumor draining lymph nodes (TDLN), but not in non-TDLN ( Supplementary  Fig. 3).
Tumor SK1 downregulation increased the proportion of IFN-γproducing CD8 + TILs (Fig. 3c) and CD107a + and granzyme + CD8 T cells ( Supplementary Fig. 1c), as well as the expression of PD-1, CTLA-4, and TIM-3 on CD8 + TILs (Fig. 3d, e) that likely reflect an accumulation of activated CD8 + T cells at day 11. Indeed, these three checkpoint molecules are upregulated upon Tcell activation during the anti-tumor response, and tumorinfiltrating CTLA-4 + PD-1 + CD8 + T cells have been shown to contain the majority of tumor-antigen-specific T cells 21,22 . This phenomenon was accompanied by a significant decrease of CTLA-4 expression in Treg (Fig. 3f, g). These results strongly suggest that SK1 downregulation attenuated Treg accumulation into tumors, leading to an enhanced CD8 + T-cell activation.
To understand how tumor SK1 expression and Treg infiltration were related, the expression of immunosuppressive cytokines (TGF-β and IL-10) and chemokines (CCL17 and CCL22) that support Treg recruitment 23,24 , was evaluated in tumors from mice injected with shCtrl or shSK1(1) Yumm cells. Our data show that SK1 silencing significantly reduced the expression of Foxp3, TGF-β, IL-10, CCL17, and CCL22 (Fig. 3h) without affecting that of IDO1, which exerts immunosuppressive effects on T-cell proliferation. Similar results were obtained using shSK1 (2) melanoma cells ( Supplementary Fig. 4), confirming that SK1 acts as a key driver of Treg-associated cytokine and chemokine expression.
Downregulation of SK1 improves the efficacy of ICI. Given that SK1 silencing led to an increased expression of immune checkpoint molecules on CD8 + T cells and to a reduced accumulation of Treg, we hypothesized that SK1 downregulation may improve the efficacy of ICI (e.g., anti-CTLA-4 and anti-PD-1 therapy). As illustrated in Fig. 4, whereas anti-CTLA-4 or anti-PD-1 monotherapy had limited effects (18% of tumor rejection) on established shCtrl tumors, SK1 silencing dramatically enhanced the response to anti-CTLA-4 or anti-PD-1 treatment, leading to tumor rejection in 100% and 67% of mice, respectively  Fig. 4a), and significantly improved overall survival (Fig. 4b). Of particular interest was the observation that, mice with complete responses fully resisted tumor rechallenge performed almost 2 months after discontinuation of therapy, indicating that downregulation of SK1 in tumors combined with ICI induced efficient long-term memory immune responses and a durable cure of the animals (Fig. 4b). Strikingly, SK1 silencing was significantly associated with a decreased infiltration of Treg and an enhanced CD8/Treg ratio in melanoma tumors and under both control conditions and ICI therapy (Fig. 4c). As compared to ICI alone or SK1 knockdown alone, combined therapies led to a significant additive reduction of Treg content. We observed a strong decrease of CCL17 and CCL22 expression (Fig. 4d). Moreover, Treg depletion using DEREG (Foxp3-DTR-GFP) mice triggered total tumor rejection of either shCtrl or shSK1 Yumm tumors, demonstrating that Treg are a major immunosuppressive population ( Supplementary Fig. 5). Altogether, the SK1 silencing-dependent reduction of Treg accumulation contributes to slow down Yumm tumor growth and promote total tumor rejection upon ICI therapy.
To investigate the impact of SK1 silencing on CD8 T-cell function, CD8 + T cells were stimulated with PMA and ionomycin in vitro prior to flow cytometric analysis of IFN-γ and TNF production ( Supplementary Fig. 6b). We observed that the frequency of IFN-γ CD8 positive cells was increased after SK1 silencing with or without ICI therapy. Of note, the frequency of TNF + CD8 + TILs decreased after SK1 silencing under basal conditions but not upon anti-CTLA-4 or anti-PD-1 treatment. Moreover, the SK1 silencing and anti-PD-1 therapy were associated with an increased expression of CD226 and a reduced expression of TIGIT on CD8 + and CD4 + T cells (Supplementary Fig. 6c, d). Whereas CD226 and TIGIT share the same ligands (CD112 and CD155), they exhibit opposite biological functions: CD226 enhances T-cell activation and TIGIT behaves as a co-inhibitory receptor towards CD8 + TILs 25,26 . We also analyzed the expression of inhibitory ligands, PD-L1 and PD-L2 on tumor and myeloid cells. We found a significant decrease for PD-L1 expression (but not for PD-L2) on tumor cells in shSK1 tumors with and without ICI therapy ( Supplementary Fig. 7).
Furthermore, to confirm the beneficial effects of combining SK1 downregulation with ICI in a highly metastatic model with a different genetic background (e.g., BALB/c), we performed experiments with the murine triple-negative breast cancer (TNBC) cells expressing luciferase (4T1-Luc cells). 4T1-Luc cells knocked-down or not for SK1, and exhibiting similar in vitro cell proliferation rates (Fig. 5a, b), were injected in the mammary fat pad and lung metastases were assessed. Whereas lung metastases were significantly reduced with shSK1 4T1 cells as compared with shCtrl 4T1 cells, this phenomenon was further pronounced by combining SK1 silencing and anti-CTLA-4 ( Fig. 5c).
In addition, albeit less potently than SK1 silencing, pharmacological inhibition of SK1 by SKI-I 9 significantly improved the anti-CTLA-4 therapy on established Yumm melanoma, and led to a potent increase of tumor rejection and animal survival (Fig. 5d, e), as well as CD8/Treg ratio (Fig. 5f). Under these conditions, SKI-I treatment slightly enhanced the anti-PD-1 efficacy (Fig. 5g). Interestingly, the potentiation of anti-PD-1 efficacy was more pronounced in the MC38 colon cancer model (Fig. 5h).

SK1 silencing decreases Pges expression in melanoma tumors.
To further characterize the molecular mechanisms triggered in tumor cells upon SK1 inhibition, we used a hypothesis-free, large scale label-free proteomic approach to compare in a global way protein abundance in shCtrl, shSK1(1), and shSK1(2) Yumm cells ( Supplementary Fig. 8). Total lysates (three biological replicates for each cell line) were fractionated by one-dimensional (1D) sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and each fraction was analyzed by mass spectrometry on an Orbitrap instrument, leading to the identification and relative quantification of 6197 proteins. Statistical analysis was performed on abundance values measured in samples silenced for SK1 (shSK1(1) + shSK1(2)) vs. shCtrl samples, and 131 proteins were found to exhibit a significant differential expression between the two groups of samples at 1% global FDR. Among these proteins, we found that the prostaglandin E2 synthase (Pges) protein (encoded by Ptges) was dramatically downregulated upon SK1 silencing, and appeared as the strongest and most significant variation observed using our unbiased proteomic analysis (Fig. 6a). We further demonstrated by quantitative reverse transcription PCR (RT-qPCR) that this regulation takes place at the mRNA level, as SK1 silencing induced a decreased mRNA expression of Ptges (Fig. 6b). Pges is an enzyme involved in the biosynthesis of PGE2, a bioactive eicosanoid that plays a complex role in inflammation and cancer 27 . As expected, lipidomic profiling by mass spectrometry showed a strong reduction of both intracellular and extracellular PGE2 levels in Yumm cells after SK1 silencing (Fig. 6c). Of note, SK1 downregulation did not alter the levels of prostaglandin F2α (PGF2α) and prostaglandin D2 (PGD2) (Supplementary Fig. 9), two arachidonic acid metabolites, which, like PGE2, are derived from prostaglandin H2 (PGH2) 28 , supporting the specific impact of SK1 silencing on Pges. To exclude an off-target effect of shRNA, we performed a rescue experiment by transfecting SK1 silencing Yumm cells with a plasmid encoding SK1 (+mSK1). Under these conditions, the inhibitory effect on Ptges expression was totally abolished by the rescue of SK1 expression, as well as the in vivo tumor growth ( Supplementary Fig. 10).
These results indicate that Pges expression may contribute, at least in part, to overcome the resistance of melanoma to anti-CTLA-4 but not anti-PD-1 therapy. Moreover, Pges silencing did not reduce the expression of Foxp3, IL-10, TGF-β, CCL17 and CCL22 (Fig. 6g). Altogether, our data demonstrate non-fully redundant immunosuppressive functions of SK1 and Pges, and ARTICLE that SK1 silencing leads to a better therapeutic benefit in combination with immunotherapy compared to Pges silencing in melanoma.

Discussion
This study provides the first evidence that melanoma SK1 behaves as an immune escape lipid kinase, leading to an increased expression of immunosuppressive factors in the TME and impairing the response to anti-CTLA-4 or anti-PD-1 treatment. SK1 overexpression has been described in many different cancer types, including lung 30 , gastric [31][32][33] , breast cancer 34,35 , and glioblastoma 36 . High SK1 expression correlates with decreased survival of patients with various cancers in a meta-analysis of clinical studies 37 . However, its role in melanoma patient survival has not yet been described. Herein, we show that high SK1 expression in melanoma cells is associated with anti-PD-1 resistance in patients with advanced melanoma. Considering the small number of patients, the association between the expression of SK1 and the tumor stage, the mutation status of BRAF and NRAS, as well as immune responses could not be performed due to weak statistical power in this retrospective study. This will be performed on a larger cohort of advanced melanoma patients treated with anti-PD-1 in combination or not with anti-CTLA-4, from a prospective clinical trial (IMMUSPHINX: NCT03627026) we are currently conducting in our institute.
SK1 produces the phospholipid S1P, which exerts its functions through intracellular actions or ligation to five cell surface Gprotein-coupled receptors (named S1PR1-5) expressed both on cancer cells and cells of the TME 38,39 . The SK1/S1P/S1PR axis can regulate tumor cell behavior, as well as the TME composition [8][9][10][11] . For instance, it promotes tumor angiogenesis 40 . Thereby, neutralization of extracellular S1P with an anti-S1P antibody significantly inhibited bFGF-and VEGF-induced angiogenesis, as well as growth of different tumors including melanomas 41 . The present study demonstrates that downregulation of SK1 reduced melanoma growth in wild-type but not immunodeficient mice or in CD8 cell-depleted mice, highlighting the involvement of CD8 + T cells in this phenomenon. Although a potential role for SK1 silencing on tumor angiogenesis inhibition cannot be excluded, it is unlikely the main process delaying tumor growth in our model.
To the best of our knowledge, this study provides the first evidence that tumor SK1 plays a key role in the modulation of TIL composition, leading to potent Treg accumulation. Indeed, we show that SK1 targeting combined with ICI resulted in a high CD8/Treg ratio. These findings highlight the major role played by tumor SK1 in immune escape. Thus, inhibiting SK1 may represent an original strategy to enhance ICI response. As targeting SK1 specifically in melanoma cells is not currently achievable, it became important to document the putative benefit of systemic pharmacological approaches based on the combination of a SK1 inhibitor and ICI. As documented here, SK1 inhibition significantly sensitized melanoma to ICI. Although, a single melanoma cell line was tested here, two other cancer models, i.e, colon and breast cancers, were shown to respond to the combination of ICI and SK1 targeting.
In line with our findings, previous studies demonstrated that the S1P/S1PR1 axis modulates the anti-tumor immune response. First, S1PR1 signaling in T cells enhances the tumor infiltration of Treg in a STAT3-dependent manner, reducing CD8 + TIL and increasing breast cancer and melanoma growth in mice 42 . Second, the role of S1P signaling in metastatic colonization and immune response was also illustrated through SPNS2 targeting 43 . SPNS2 is a S1P transporter that regulates the S1P levels in blood and lymph 44 . The deletion of SPNS2 either globally or in a lymphatic endothelial-specific manner decreased the number of pulmonary metastatic melanoma foci 43 , a phenomenon associated with an increased infiltration of activated CD8 + T cells and NK cells in the lung. Of note, the role of S1P signaling on ICI efficacy has not yet been addressed.
Further analyses show that SK1 knockdown in melanoma tumors potently reduced the production of various immunosuppressive cytokines such as TGFβ, IL10, CCL17, and CCL22, which is in line with the strong decrease of Treg tumor infiltration. To get insights into the molecular mechanisms underlying, we performed a proteomic analysis of SK1 knockdown and control melanoma cells. Pges was found to be dramatically decreased upon SK1 silencing, leading to a strong reduction of PGE2 production. It was previously demonstrated that SK1 downregulation attenuates PGE2 production through inhibition of PGES, but not COX2, via the p38 MAPK pathway in human umbilical vein endothelial cells 45 . Whereas in our mouse Fig. 6 Reduced PGE2 production induced by shSK1 partially enhances ICI efficacy. a Volcano plot showing proteins differentially regulated between control shRNA (shCtrl) and SK1-targeted shRNA (shSK1 (1 and 2)) Yumm cells. Blue and red points represent under-and overexpressed proteins, respectively. b Sphk1 (left panel) aexpression in shCtrl (black column, n = 17 mice), shSK1(1) (white column, n = 17 mice) or shSK1(2) (gray column n = 12 mice) Yumm cells and Ptges (right panel) expression in shCtrl (black column, n = 8 mice), shSK1(1) (white column, n = 8 mice) or shSK1(2) (gray column, n = 5 mice) Yumm cells; n = 5-10 independent experiments. c Intracellular (left panel, n = 5) and extracellular (right panel, n = 4) PGE2 levels were normalized to those in shCtrl; n = 2-6 independent experiments. d Ptges (left panel, n = 5) and Sphk1 (right panel, n = 7) expression was analyzed in shCtrl (P) (dark gray columns) or shPGES (clear gray columns) Yumm cells; n = 5-6 independent experiments. e, f Mice were challenged with shCtrl(P) or shPGES cells on day 0, and then treated with isotype control antibody (iso; black lines), anti-CTLA-4 (red lines) or anti-PD-1 (blue lines). e Individual growth curves are depicted for each tumor (n = 10 mice). Insets: numbers indicate percentage of tumor-free mice at day 25. f Kaplan-Meier survival curves with log-rank test. shCtrl(P) and shPGES tumors were represented by black and white squares, respectively. g Levels of Foxp3, Tgfb1, Il10, Ccl17 and Ccl22 transcripts in shCtrl(P) (dark gray columns) or shPGES (clear gray columns) tumors (n = 5). Data, expressed as relative expression, are means ± SEM. The bottom-most and topmost horizontal lines, the lower and upper hinges, and the middle line of the boxplots indicate the minimum and maximum values, the 25th and 75th percentiles, and the median, respectively. Samples were compared using Kruskal-Wallis test with Dunn's correction (b and c) or Mann-Whitney test (d).
melanoma model Pges knockdown sensitized, albeit to a lesser extent than SK1 knockdown, to anti-CTLA-4 therapy, no effect towards anti-PD-1 treatment was observed. This suggests that inhibition of Pges expression, along SK1 knockdown in melanoma cells, is probably not the sole molecular mechanism responsible for the synergy observed between SK1 silencing and ICI. Our data seem at odds with a recent study, which shows that the inhibition of PGE2 production, by aspirin or celecoxib (two COX inhibitors), increased anti-PD-1 treatment efficacy in mouse melanoma and colon cancer 46 . One should note however, that the pharmacological inhibition of COX-driven synthesis of PGE2 is likely associated with the decreased production of PGH2 and additional eicosanoid metabolites such as PGI 2 , PGD 2 , PGF 2a , and TXA 2, which may also influence platelet aggregation, as well as immune responses [47][48][49] .
In conclusion, our current observations demonstrate that targeting SK1 is a promising option as inhibition of this specific lipid kinase affects (i) the primary tumor growth, (ii) the intratumoral CD8/Treg ratio (iii) the immunotherapy efficacy by immune checkpoint blockade such as anti-CTLA-4 or anti-PD-1. Future work is needed to explore the potential synergism of SK1 antagonism with other ICI or with cancer vaccine. Moreover, the potential roles of other immune cell subsets like NK, γδ T cells and especially various dendritic cells, as well as the role of SK1 in a non-T-cell-inflamed tumors have not yet been explored. These will require further investigation.
Taken together, our findings support the notion that SK1 is a checkpoint lipid kinase, which modulates immune escape mechanisms. Also, the relative abundance of SK1 within tumors could be highly predictive of the response to ICI therapy. Finally, it is tempting to speculate that clinical strategies based on combining SK1 inhibition and ICI may improve the response rate in patients affected with melanoma or other malignancies.

Methods
In silico analysis with Oncomine database. SPHK1 expression in human nevi or melanomas was analyzed using the Oncomine tool (www.oncomine.org) in 2 different cohorts 50,51 .
4T1 Metastasis analysis. At day 31, mice with 4T1-Luc shCtrl or shSK1 tumors and treated or not with anti-CTLA-4 were sacrificed, lungs were collected and homogenized using Precellys. RT-qPCR was performed using primers for transcripts encoding murine Luciferase and Cyclophilin-A (Qiagen). Evaluation of the metastatic load was performed by quantification of luciferase mRNAs in lungs by RT-qPCR and normalized to Cyclophilin-A. Luciferase_F CTCACTGAGACTAC ATCAGC; Luciferase_R TCCAGATCCACAACCTTCGC.
MC38 tumor growth. Mice were inoculated with 300,000 MC38 cells at day 0, and then treated or not with vehicle (Ctrl), SKI-I (50 mg/kg by i.p. injection on days 5, 7, 10, 13, and 15) and anti-PD-1 (100 µg per mouse at day 7 and 10). Cells were analyzed with a BD LSR Fortessa X-20 (BD Biosciences), followed by Flow-Jo software analysis. Gating strategies followed for flow cytometry analysis are shown in Supplementary Fig. 1.
Quantification of eicosanoids. Eicosanoid concentrations in conditioned media or Yumm cell lysates were determined by ultra-performance liquid chromatography, using an Agilent 1290 system coupled to a G6460 triple quadripole mass spectrometer (Agilent Technologies).
Analysis of SPHK1 expression in human melanoma. SPHK1 expression was analyzed using the TCGA melanoma cohort. TCGA genomic and clinical data were downloaded from the UCSC Xena platform (https://xenabrowser.net/). The analysis population consisted of 342 patients with distant metastasis for whom RNAseq and clinical data overlap. No formal sample size calculation was performed concerning TCGA analysis. Gene expression was measured experimentally using the Illumina HiSeq 2000 RNA Sequencing platform and log2(x + 1) transformed. The strength of the relationship between SPHK1 transcript levels with those of transcripts encoding immunosuppressive factors was assessed using Spearman's rank correlation coefficient.
Statistical analyses. Results are expressed as means ± SEM, and group comparisons were performed with Mann-Withney, Kruskal-Wallis test with Dunn's correction or two-way Anova repeated measures with Bonferoni correction for multiple comparisons. The strength of relationship between continuous biomarkers was assessed using Spearman's rank correlation coefficient. The major clinicopathological characteristics and available treatment information of the cohorts are presented in Table 1 and summarized by frequency and percentage. All survival times were calculated from the initiation of immunotherapy and estimated by the Kaplan-Meier method with 95% confidence intervals (CI), using the following first event definitions: progression or death from any cause for progression-free survival (PFS) and death from any cause for overall survival (OS). Patients still alive were censored at their date of last follow-up. Comparison between groups (low expression vs. high expression) was performed using log-rank test. All tests were two-sided and differences were considered statistically significant when p < 0.05 (*p < 0.05; **p < 0.01; ***p < 0.001, ****p < 0.0001). Data were analyzed using GraphPad Prism (GraphPad Software, San Diego, CA) and STATA v13 (Stata Corporation, College Station, TX, USA).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD016535 [http://www.ebi.ac.uk/pride/archive/projects/PXD016535]. The oncomine data referenced during the study in Fig. 1 are available in a public repository from the oncomine website (www.oncomine.org). Talantov data can be accessed here. Xu data can be accessed here. The TCGA source data underlying Fig. 7 are available in UCSC Xena platform: https://tcga.xenahubs.net/download/TCGA.SKCM.sampleMap/HiSeqV2.gz. All the other data supporting the findings of this study are available within the article and its supplementary information files and from the corresponding author upon reasonable request. A reporting summary for this article is available as a Supplementary Information file.