SOX10 Regulates Melanoma Immunogenicity through an IRF4–IRF1 Axis

Unresponsiveness to cancer immunotherapies, including immune checkpoint inhibitors (ICB), is mainly determined by the immunogenicity of the cancer cells as well as surrounding tumor microenvironment (TME; ref. 1). In general, an immunologically inflamed “hot” TME shows a signature of antitumor CD8⁺ T-cell responses and corresponds to better clinical responses to ICBs, whereas an immunologically “cold” TME corresponds to poor clinical responses to ICBs (2). In such immunologically “cold” TME, cancer cells often show low immunogenicity because of their low mutation burden (3), lacking expression of cancer-specific or cancer-associated antigens, or impairing antigen-presentation machineries (4–6); therefore, they cannot be recognized by antitumor CD8⁺ T cells.

IFNγ is a proinflammatory cytokine and mainly produced by cytotoxic CD8⁺ T cells, natural killer cells, and type I helper CD4⁺ T cells. Once it binds to a specific receptor, IFNγ activates the JAK–STAT signaling pathway and induces proinflammatory gene expression through the transcription of IFN regulatory factor 1 (IRF1). IRF1 is considered as a master regulator of cell immunogenicity because many of its target genes are critical for determining the immunogenicity of cells such as antigen-presenting molecules (MHC class I, TAP, β2M; ref. 7) and ligands of immune inhibitory receptors (PD-L1 and PD-L2; ref. 8). In particular, the association between PD-L1 expression on cancer cells and clinical responsiveness to ICBs has been reported in a variety of cancers including melanoma, which is one of the most immunogenic cancer types (9–12). Recently, JAK1/2 mutation was identified as both a primary and acquired mechanism of resistance to ICBs in cancers including melanoma (5, 6). By lacking responsiveness to IFNγ with loss-of-function mutations of JAK1/2, the immunogenicity of cancer cells is significantly impaired; therefore, having an understanding of other bypassing pathways that activate IRF1 in cancer cells is clinically important.

In this study, we demonstrate that SOX10 regulates the immunogenicity of melanoma through transcriptional control of the IRF4–IRF1 axis. Using both genetic and pharmacologic approaches, we also demonstrate that SOX10 represses IRF1 transcription through the direct induction of a negative IRF1 regulator, IRF4. Importantly, the SOX10–IRF4 axis independently regulates IRF1 expression from the JAK–STAT signaling pathway in melanoma, and the suppression of SOX10 is involved in the efficacy of combination therapy of anti-PD-1 antibody and histone deacetylase inhibitor (HDACi) against a clinically relevant murine melanoma model.

JAK inhibitor, baricitinib, was purchased from MedChem Express. HDACis used were vorinostat (suberoylanilide hydroxamic acid; Cayman Chemical), TMP269 (Cayman Chemical), ACY-1215 (BioVision), RGFP109 (MedChemExpress), CAY10683 (Cayman Chemical), and BRD73954 (Sigma-Aldrich).

The plasmids used were pLenti CMV-E2/Crimson and pLenti-CMV-SOX10, which subcloned E2-Crimson fluorescent gene (Takara Bio) human SOX10 cDNA into pLenti CMV Hygro DEST [a gift from Eric Campeau and Paul Kaufman (Addgene plasmid #17454); ref. 13]. pLNCX and pLNCX-SOX10 were described previously (14).

Human melanoma lines UACC62, UACC257, and Malme-3M were obtained from the NCI. 501mel was a gift from Ruth Halaban (Yale University Medical School, New Haven, CT). Other human melanoma cell lines were from ATCC. UACC257, A2058, SK-MEL-28, Malme-3M, UACC62, and SK-MEL-2 human melanoma cells were cultured in RPMI1640 medium (Nissui) containing 10% FBS and penicillin/streptomycin/l-glutamine. A375, MeWo, Hs 940.T, M14, and SK-MEL-5 human melanoma cells were cultured in DMEM (Nissui) containing 10% FBS and penicillin/streptomycin/l-glutamine. RPMI7951 human melanoma cells were cultured in Eagle Minimum Essential Medium (Nissui) containing 10% FBS and penicillin/streptomycin/l-glutamine. The D4M melanoma cell line was a gift from David Mullins (15). The D4M.3A.3 cell line, referred to as D4M in this study, was derived from single-cell cloning of D4M.3A, as described previously (16).

For siRNA knockdown experiments, siRNA for SOX10 (s13309 and s13310, Thermo Fisher Scientific), siRNA for IRF1 (s7503, Thermo Fisher Scientific), siRNA for IRF4 (s7511, Thermo Fisher Scientific), or negative control #1 (Thermo Fisher Scientific) was used for transfection at a final concentration of 12.5 nmol/L of melanoma cells by Lipofectamine RNAiMAX reagent (Thermo Fisher Scientific).

The stable A375 transformed cells containing pLNCX empty or pLNCX-SOX10 (full-length SOX10; ref. 14) were selected with G418 (500 μg/mL; Sigma-Aldrich), and referred to as A375 (−) and A375/SOX10. The stable A2058 transformed cells containing pLenti CMV-E2/Crimson or pLenti-CMV-SOX10 were selected with hygromycin (400 μg/mL; Sigma-Aldrich), and referred to as A2058 (−) and A2058 SOX10. The stable D4M/SOX10 cells were established similarly to A2058/SOX10 cells.

Gene profiles were analyzed by GenePattern (Broad Institute of MIT and Harvard, Cambridge, MA). Publically available datasets used in this article were GSE7127 containing 63 melanomas (17) and data on 62 skin cancers were obtained from the Broad Novartis Cancer Cell line Encyclopedia (http://www.broadinstitute.org/ccle/home; ref. 18). Pearson correlation values of other mRNA expressions with PD-L1 mRNA expression were calculated by Excel.

Total RNAs were prepared using RNeasy Plus Mini Kit (Thermo Fisher Scientific) from human melanoma cells. qRT-PCR was performed using One Step SYBR PrimeScript RT-PCR Kit II (Perfect Real-Time; Takara Bio). The primers used were: 5′-AGC TCA GCA AGA CGC TGG-3′ (sense) and 5′-CTT TCT TGT GCT GCA TAC GG-3′ (antisense) for SOX10 mRNA, 5′-CCA GCA CAC TGA GAA TCA ACA-3′ (sense) and 5′-ATT TGG AGG ATG TGC CAG AG-3′ (antisense) for PD-L1 mRNA, 5′-CTT CCA TGG GAT CTG GAA GA-3′ (sense) and 5′-GAC CCT GGC TAG AGA TGC AG-3′ (antisense) for IRF1 mRNA, and 5′-GCA CAG AGC CTC GCC TT-3′ (sense) and 5′-GTT GTC GAC GAC GAG CG-3′ (antisense) for β-actin mRNA. All reactions were run in triplicate, and mRNA levels were normalized to β-actin mRNA expression. To check correlations, expression values for each gene were normalized to a mean of zero and a SD of one.

Cells were reverse transfected with 12.5 nmol/L siRNA for 96 hours, as above, treated with each HDACi for 24 hours, or treated with IFNγ for 24 hours. Human or mouse CD274 was stained with PE-conjugated anti-human or mouse CD274 antibody (eBioscience). Human CD273 or HLA-A/B/C was stained with PE-conjugated anti-human CD273 antibody (eBioscience) or FITC-conjugated anti-human HLA-ABC (BD Pharmingen). For staining immune cells in melanoma tissues, FITC-anti-CD3ϵ (2C11), PE-anti-CD4, PerCP Cy5.5-anti-NK1.1 (PK136), PE Cy7-anti-4–1BB, and APC-anti-CD8 antibody were purchased from BioLegend or eBioscience. All flow cytometry data were analyzed using FlowJo software (Treestar Software).

Whole-cell extracts (10 μg/lane) or nuclear extracts were prepared using the method of Schreiber and colleagues (19) and then subjected to Western blotting analysis using anti-SOX10 antibody (Santa Cruz Biotechnology), anti-MITF antibody (C5), anti-IRF4 antibody (Santa Cruz Biotechnology), anti-IRF1 antibody (Cell Signaling Technology), anti-α-tubulin antibody (Sigma), or anti-histone H3 antibody (Abcam). The band intensities were measured by ImageJ and normalized to that of each control lane.

Chromatin immunoprecipitation (ChIP) assays were performed as described previously (20). Antibody used was anti-RNA polymerase II serine2 phosphorylation (Abcam) or anti-SOX10 antibody (Santa cruz Biotechnology). Primers used were: 5′-GAG GAA ACT GAG GTC CAA AGA A -3′ (sense) and 5′-GCC CAG ACT TCA GAG CTA ATC -3′ (antisense) for the PD-L1 intron 3 gene region, 5′-GCA AAG GCA TTC CAC TGT TC -3′ (sense) and 5′-GCA TCT TCT ACC TCC ATC CAT AC-3′ (antisense) for the PD-L1 exon 7 gene region, 5′-TGG AGG GAA TCG TGA CCT A -3′ (sense) and 5′-GCT AAG ACC AGG ACG CTA AC-3′ (antisense) for the IRF1 intron 1 gene region, 5′-AGG GCA GCT GAT CTC TTC A-3′ (sense) and 5′-GGC TAA ACC TGG CAC CAA A-3′ (antisense) for the IRF4 intron 4 gene region, and 5′-TGG GCT GTT TCT GGT AAT CA-3′ (sense) and 5′-CAC CTT GGA ATT TCC TGT GC -3′ (antisense) for the MIA gene region.

We classified patients with The Cancer Genome Atlas (TCGA) melanoma as SOX10^low and SOX10^high based on the bottom and top quintile of SOX10 mRNA expression, respectively. We used the “lifelines” package for Python to plot Kaplan–Meier curves and to compute the associated log-rank P value. To control for age, we also evaluated age-adjusted P values using the Cox model with two parameters: patient age and indicator variable specifying SOX10^high and SOX10^low groups. Since in the latest version of TCGA data, the expression value of SOX10 is erroneously set to zero for all TCGA samples across all cancers, our patient survival analysis used the legacy TCGA data downloaded from the GDC Legacy Archive on March 26, 2021.

C57BL/6 mice (6 weeks old) were purchased from Japan SLC Inc.. All experiments were approved and performed according to the guidelines of the Care and Use of Laboratory Animals of University of Toyama (Toyama, Japan). The D4M or D4M/SOX10 cells were inoculated subcutaneously (5 × 10⁵ cells/100 μL in PBS/mice) into the flanks of anesthetized mice. Mice in each group intra-peritoneally received vorinostat in 10% DMSO solution (25 mg/kg/day) or vehicle every day and anti-PD-1 antibody (10 mg/kg/day, RMP1–14, BioXcell) on days 7 and 9. The tumor volume was assessed every 2 days starting from day 3. The data are presented as the mean luminescence ± SEM.

Significance was calculated using Graphpad Prism software (GraphPad Software, Inc.). All statistical analyses were performed using the data from at least three independent experiments. More than three means were compared using two- or one-way ANOVA with Bonferroni correction, and two means were compared using the unpaired Student t test. P < 0.01 was considered significant.

To identify genes that intrinsically regulate PD-L1 expression in melanoma, we first examined the correlation of gene expression relative to PD-L1 in melanoma using publically available datasets [GSE7127 (17) and datasets from the Broad Novartis Cancer Cell line Encyclopedia (CCLE; ref. 18; Fig. 1A; Supplementary Fig. S1]. A gene with one of the strongest negative correlations with PD-L1 expression in melanoma is SOX10 (Pearson correlation values are −0.554, top of 54,676 probes in GSE7127, and −0.621, ninth of 18,988 genes in CCLE, respectively), and it was also confirmed in mRNA expression (Pearson correlation value is −0.349; Fig. 1B). To determine the negative correlation at protein levels of SOX10 and PD-L1, we used four melanoma cell lines that have a distinct SOX10 expression status (Fig. 1C). As shown in Fig. 1C, the cell surface expression of PD-L1 on melanoma cell lines with a high SOX10 status (UACC257 and A2058) was lower than that on melanoma cells with a low SOX10 status (RPMI7951 and Hs.940.T). The expression of SOX10 in UACC257 and A2058 cells should be functional because MITF, which is a downstream target of SOX10 (21–24), was coexpressed in those cells. To examine the clinical relevance of our findings in patients with melanoma, we reanalyzed the dataset of gene expression of patients with melanoma by classifying the SOX10^high and SOX10^low status. As shown in Fig. 1D, there was a strong correlation with poor survival in patients with the SOX10^high status compared with the patients with the SOX10^low status, therefore SOX10 expression could serve as a prognostic factor in patients with melanoma (age adjusted Cox regression P value = 0.041, log-rank P value = 0.023).

Next, to understand the functional involvement of SOX10 in regulating PD-L1 expression in melanoma, we knocked down SOX10 in several human melanoma cell lines with high SOX10 expression (UACC257, A2058, and A375) or medium SOX10 expression (MeWo, SK-MEL-28, and Malme-3M). SOX10 knockdown using SOX10 siRNA in UACC257, A2058, MeWo, and A375 cells resulted in upregulation of cell surface expression of PD-L1 (Fig. 2A; Supplementary Fig. S2A). Although SOX10 knockdown in SK-MEL-28 and Malme-3M cells did not solely affect their PD-L1 expression (Fig. 2A; Supplementary Fig. S2A), IFNγ-induced PD-L1 expression in both cell lines was increased by SOX10 knockdown (Fig. 2B). These results suggest that SOX10 also regulates PD-L1 expression in SK-MEL-28 and Malme-3M cells in response to IFNγ. Conversely, SOX10 overexpression inhibited IFNγ-induced PD-L1 expression in A2058 cells (Fig. 2C). Importantly, the JAK-dependent IFNγ signaling pathway was not involved in the SOX10-dependent control of PD-L1 expression, considering that treatment with the JAK inhibitor baricitinib did not affect PD-L1 expression induced by siSOX10 treatment (Fig. 2D; Supplementary Fig. S2B and S2C), although baricitinib could inhibit IFNγ-induced PD-L1 expression. These results clearly indicate that SOX10 negatively regulates PD-L1 expression independently of the JAK–STAT IFNγ signaling pathway in melanoma.

As mRNA expression of PD-L1 was inversely correlated with that of SOX10 (Fig. 1A), we subsequently examined the role of SOX10 in transcriptional control of PD-L1 gene expression by the ChIP assay using antibody against polymerase II with Ser2 phosphorylation (Pol-II S2). As shown in Fig. 3A, increased Pol-II S2 occupancy was seen in intron 3 and exon 7 of the PD-L1 gene by SOX10 knockdown in UACC257 cells. In addition, we also observed increased Pol-II S2 occupancy in intron 1 of the IRF1 gene, which is known as a key regulator of PD-L1 expression (Fig. 3A; ref. 8). Consistently, mRNA expression of PD-L1 and IRF1 was increased by SOX10 knockdown in UACC257 cells (Fig. 3B), suggesting the function of SOX10 to control IRF1 transcription machinery. To determine the role of SOX10 as a negative regulator of IRF1, we analyzed the protein expression of IRF1 in different melanoma cell lines by knocking down SOX10. As shown in Fig. 3C, SOX10 knockdown significantly upregulated IRF1 expression in A2058, UACC257, SK-MEL-28, and Malme-3M cells, suggesting the function of SOX10 as a repressor of IRF1. We also observed that such IRF1 induction after SOX10 knockdown was independent on JAK activity (Supplementary Fig. S2B). In addition to PD-L1, both PD-L2 and HLA-A/B/C are known as IRF1-target genes (7, 8), were upregulated by SOX10 knockdown in UACC257 cells (Fig. 3D). Conversely, IRF1 knockdown significantly impaired the PD-L1 upregulation by SOX10 knockdown in UACC257 cells (Fig. 3E). Collectively, these results indicate that SOX10 regulates PD-L1 expression in melanoma cells through the repression of IRF1.

Considering that IRF4 is a major pigmentation-associated gene (25) and known to repress IRF1 (26), we subsequently investigated the role of IRF4 in the SOX10-dependent regulation of PD-L1 expression in melanoma cells. Similar to MITF, which is a direct target of SOX10, the expression of IRF4 in four different melanoma cell lines was significantly decreased by knocking down SOX10 (Fig. 4A). Such IRF4 downregulation by SOX10 knockdown was likely due to the reduction of IRF4 transcription, as seen in the decrease of Pol-II occupancy in the PD-L1 gene (Fig. 4B). To further determine whether SOX10 directly regulates IRF4 expression, binding of SOX10 to intron 4 of the IRF4 gene, which contains a pigmentation-associated enhancer region (25), was tested using ChIP. As shown in Fig. 4C, the IRF4 region was immunoprecipitated using anti-SOX10 antibody similar to the MIA region, which is a direct SOX10 target gene (27), suggesting the direct binding of SOX10 to the IRF4 enhancer region. Consistently, IRF4 knockdown significantly enhanced IFNγ-induced PD-L1 expression, whereas either IRF1 knockdown or IRF1/IRF4 double knockdown reduced such an effect of IFNγ (Fig. 4D). Collectively, these results suggest that SOX10 negatively regulates PD-L1 expression by directly binding to the IRF4 enhancer region to repress IRF1.

Because SOX10 expression is reportedly suppressed by (HDACis (refs. 14, 28), we examined the effects of a clinically available pan-HDACi vorinostat (suberoylanilide hydroxamic acid) on the expression of PD-L1 in melanoma. As shown in Fig. 5A, cell-surface PD-L1 expression was induced by vorinostat treatment in all four human melanoma cell lines tested. Importantly, the vorinostat treatment reduced IRF4 expression and thereby induced IRF1 expression along with suppressing SOX10 expression in those melanoma cell lines. Similar with that after SOX10 knockdown, PD-L1 induction after vorinostat treatment could not be impaired by JAK inhibitor (Supplementary Fig. S3A). Furthermore, the overexpression of SOX10 in A375 cells significantly impaired the effect of HDACi to induce PD-L1 expression (Fig. 5B), suggesting that SOX10 can be a major target of HDACi to induce PD-L1 expression through the IRF4–IRF1 axis in melanoma.

To determine the specific subtype of HDACs involved in SOX10 suppression, we tested the effect of HDACi with different specificities on SOX10 expression. ACY-1215 (inhibiting HDAC1, 2, 3, 6, and 8) and RGFP109 (inhibiting HDAC1 and 2) significantly suppressed SOX10 expression in UACC257 cells, whereas TMP269 (inhibiting HDAC4, 5, 7, and 9), CAY10683 (inhibiting HDAC2), and BRD73954 (inhibiting HDAC6 and 9) did not show any effects on SOX10 expression (Supplementary Fig. S3B); therefore, HDAC1/3 can be potential subtypes of HDAC involved in SOX10 suppression. Importantly, treatment with ACY-1215 or RGFP109 significantly up-regulated PD-L1 expression of UACC257 cells (Supplementary Fig. S3C), supporting the involvement of HDAC1/3 in controlling PD-L1 expression through SOX10.

Both tumor immunogenicity and the presence of effector T cells are key for successful immunotherapy against cancer. In this regard, it has been reported that HDAC inhibition potentiates the efficacy of immune checkpoint blockade to enhance T-cell infiltration into the tumor site by upregulating chemokines (29), although HDACi showed limited clinical efficacy as a single agent in a phase I trial involving patients with melanoma (30). To test the involvement of SOX10 suppression in the efficacy of such combination therapy with HDACi and immune checkpoint blockade, we used the D4M melanoma model. D4M is a mouse melanoma cell line derived from BRAF^V600E/Pten^−/− mice (15) with a low mutation burden (16); therefore, it is known to show the lower level responsiveness to PD-1 blockade (16). As seen in human melanoma cell lines (Fig. 5), the expression of PD-L1 in D4M cells was increased on vorinostat treatment in vitro (Fig. 6A). To test the effect of combination therapy with HDACi and PD-1 blockade in vivo, we used the subcutaneous implantation model of D4M melanoma cells. Although neither HDACi (vorinostat) nor anti-PD-1 antibody treatment was solely effective, their combination significantly suppressed D4M melanoma growth (Fig. 6B) and showed a beneficial effect on the survival of tumor-bearing mice (Fig. 6C). Consistent with a previous report (30), vorinostat treatment increased the frequency of tumor-infiltrating CD3⁺ T cells compared with control mice (Fig. 6D); however, it was not effective on its own to control D4M tumor growth (Fig. 6B). The reduction of SOX10 mRNA (Fig. 6E) and, conversely, the upregulation of PD-L1 (Fig. 6F) were observed in D4M tumors treated with HDACi in vivo. Importantly, combination therapy with HDACi and anti-PD-1 antibody induced the activation of tumor-infiltrating CD8⁺ T cells, as seen in the expression of 4-1BB (Fig. 6G). Finally, to determine the involvement of SOX10 in the efficacy of combination therapy with HDACi and anti-PD-1 antibody, we used D4M cells overexpressing SOX10 (D4M/SOX10). As shown in Fig. 6H, the effects of combination therapy with HDACi and anti-PD-1 antibody to control D4M/SOX10 tumor growth were completely blocked, suggesting that SOX10 suppression is an underlying mechanism explaining the efficacy of the combination of HDACi with PD-1 blockade therapy.

In the current study, we identified the regulatory role of the SOX10–IRF4–IRF1 axis in melanoma immunogenicity independent of JAK-dependent IFN signaling. Regarding the clinical relevance of our findings, we demonstrated that HDACi can improve the efficacy of anti-PD-1 therapy through a SOX10-dependent mechanism, possibly by enhancing the immunogenicity of melanoma cells.

SOX10 is known to be an indispensable gene for development of melanocytes and peripheral glial cells, derived from neural crest lineages; therefore, loss of function via SOX10 mutation causes Waardenburg syndrome type 4 in humans, which is an auditory-pigmentary syndrome associated with a megacolon (31). MITF is a direct SOX10 target molecule (21–24) and master regulator of melanocyte development, function, and survival, and is a genomically amplified melanoma-specific oncogene in some cases (32–34). In addition, some direct target proteins of SOX10 have been reported, such as myelin protein zero (35), myelin basic protein (36), connexin 32 (36), involved in the differentiation of peripheral glial cells. Contrary to its important role in developmental processes, the biological role of SOX10 in melanoma cells is less understood. In this regard, we showed that SOX10 regulates PD-L1 expression in melanoma through the IRF4–IRF1 axis along with other IRF1-target genes, such as PD-L2 and HLA-A/B/C independently of the JAK–STAT IFN signaling pathway (Fig. 3D). These results strongly support the suggestion that SOX10 functions as a cell-intrinsic regulator of immunogenicity through IRF1 regulation in melanoma cells.

Such SOX10-dependent suppression of melanoma immunogenicity is supported by the negative correlation between SOX10 mRNA and PD-L1 mRNA in mRNA expression data from melanoma cell lines (Fig. 1A) and also CCLE from patients with melanoma (Supplementary Fig. S1). In addition, the correlation values of IRF4 (−0.239 vs. PD-L1) and IRF1 (0.547 vs. PD-L1) in the GSE7127 melanoma dataset also support the importance of the IRF4–IRF1 axis. Contrary to melanoma, there was no significant correlation between SOX10 and PD-L1 expression in breast cancer, which expresses SOX10 (Pearson correlation value of −0.042, 9,996 of 18,988 genes in CCLE; ref. 37). Although there is a report that IRF1 inhibits antitumor immune responses through the up-regulation of PD-L1 in several different types of mouse tumor models (38), no relevance of SOX10 and IRF4 to control IRF1 expression is known elsewhere. Considering that SOX10 directly regulates IRF4 through binding of a pigmentation-associated enhancer region (Fig. 4C), SOX10 regulates IRF1 expression specifically in melanoma cells.

Regarding the clinical relevance of our findings, it is known that a tumor in an immunologically “cold” state corresponds to low PD-L1 expression and is correlated with a poor prognosis of patients with cancer including patients with melanoma receiving PD-1/PD-L1 blockade therapy. Considering the current results (Fig. 6), SOX10 suppression by HDACi treatment may change such an immunologically “cold” tumor state into a “hot” state by activating IRF1 as seen by increasing CD8⁺ T-cell infiltration. However, those tumor-infiltrating CD8⁺ T cells were not fully activated because SOX10 suppression by HDACi treatment also induced PD-L1 expression on melanoma cells through an IRF1-dependent mechanism. Such SOX10-dependent PD-L1 induction may correspond to the lack of beneficial effects in a clinical trial of HDACi in patients with melanoma (30). In this context, it is clinically important to classify the patients with SOX10^high melanoma, because SOX10 suppression in their patients can improve or enhance the efficacy of PD-1/PD-L1 blockade therapy. Moreover, the defect of IFNγ signaling in cancers bearing JAK1/2 mutation is also involved in primary and acquired resistance to anti-PD-1 therapy (5, 6). Importantly, we found that HDACi could also induce PD-L1 independent of the JAK–STAT pathway (Supplementary Fig. S3A); therefore, HDACi may show “warm-up” effects in immunologically “cold” tumors with JAK1/2 mutation through the SOX10–IRF4–IRF1 axis.

Although we clearly demonstrated that HDACi can increase immunogenicity in melanoma cells through the SOX10–IRF4–IRF1 axis, the exact subtype of HDAC that specifically controls SOX10 expression remains unknown. The chemical inhibition of HDAC1 and HDAC3 by ACY-1215 and RGFP109 antagonized SOX10 expression in melanoma (Supplementary Fig. S3B), suggesting the involvement of those two HDACs. Using genetic inhibition by siRNA, the triple knockdown of HDAC1, 2, and 3, instead of the double knockdown of HDAC1 and 3, strongly suppressed SOX10 expression in melanoma (Supplementary Fig. S3D). Considering that HDAC1 and 2 are functionally redundant HDACs (39, 40) that colocalize in super enhancer regions together with HDAC3 (40), and SOX10 expression is predominantly regulated by superenhancers in melanoma (41, 42), we speculate that HDAC1, 2, and 3 predominantly control SOX10 expression among other subtypes of HDACs. Although there was a discrepancy in the response between SOX10 knockdown and HDACi treatment to induce PD-L1 expression in SK-MEL-28 and Malme-3M cells (Fig. 2A), such a discrepancy can be explained by the different efficiency of siRNA knockdown and pharmacologic inhibition of SOX10. Alternatively, HDACi may also induce IRF1 expression in those melanoma cells through mechanism other than the SOX10–IRF4 axis, considering that there is a report that HDACi treatment induces PD-L1 and IRF1 expression in myeloid cells (43).

In summary, we identified SOX10 as a regulator of the immunogenicity of melanoma through an IRF4-IRF1–dependent transcriptional mechanism. Regarding the clinical relevance of our findings, treatment with HDACi, such as clinically available vorinostat, reinvigorates the immunogenicity of melanoma cells and enhances the efficacy of anti-PD-1 therapy through a SOX10-dependent mechanism. Thus, we strongly consider the SOX10–IRF4–IRF1 axis to be a promising target to immunologically warm up patients with melanoma showing a “cold” state in their TME.

F.S. Hodi reports personal fees from Bristol-Myers Squibb, Merck, EMD Serono, Novartis, Surface, Compass, Apricity, Aduro, Sanofi, Pionyr, 7 Hills Pharma, Bicara, Checkpoint, Genentech, BioEntre, Gossamer, Iovance, Trillium, Catalym, Immunocore, and Amgen outside the submitted work; in addition, F. Hodi has a patent for Methods for Treating MICA-Related Disorders (#20100111973) pending, licensed, and with royalties paid, a patent for Tumor antigens and uses thereof (#7250291) issued, a patent for Angiopoiten-2 Biomarkers Predictive of Anti-immune checkpoint response (#20170248603) pending, a patent for Compositions and Methods for Identification, Assessment, Prevention, and Treatment of Melanoma using PD-L1 Isoforms (#20160340407) pending, a patent for Therapeutic peptides issued, a patent for METHODS OF USING PEMBROLIZUMAB AND TREBANANIB pending, a patent for Vaccine compositions and methods for restoring NKG2D pathway function against cancers pending, licensed, and with royalties paid, a patent for Antibodies that bind to MHC class I polypeptide-related sequence A pending, licensed, and with royalties paid, and a patent for ANTI-GALECTIN ANTIBODY BIOMARKERS PREDICTIVE OF ANTI-IMMUNE CHECKPOINT AND ANTI-ANGIOGENESIS RESPONSES pending. D.E. Fisher has a financial interest in Soltego, a company developing salt-inducible kinase inhibitors for topical skin-darkening treatments that might be used for a broad set of human applications. The interests of D.E. Fisher were reviewed and are managed by Massachusetts General Hospital and Partners HealthCare in accordance with their conflict of interest policies. No disclosures were reported by the other authors.

S. Yokoyama: Conceptualization, data curation, formal analysis, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing. A. Takahashi: Data curation, formal analysis, investigation. R. Kikuchi: Data curation, formal analysis, investigation. S. Nishibu: Data curation, formal analysis, investigation. J.A. Lo: Data curation, formal analysis, investigation. M. Hejna: Data curation, formal analysis, investigation. W.M. Moon: Data curation, formal analysis, investigation. S. Kato: Data curation, formal analysis, investigation. Y. Zhou: Data curation, formal analysis, investigation. F. Hodi: Supervision, investigation. J.S. Song: Data curation, formal analysis, investigation, writing–original draft, writing–review and editing. H. Sakurai: Supervision, writing–review and editing. D.E. Fisher: Supervision, writing–original draft, writing–review and editing. Y. Hayakawa: Conceptualization, data curation, formal analysis, supervision, writing–original draft, project administration, writing–review and editing.

The authors thank members of the Hayakawa laboratory and Sakurai laboratory for discussions and suggestions.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.