G9a Inhibition Enhances Checkpoint Inhibitor Blockade Response in Melanoma

Immunotherapy targeting the immune checkpoint molecules cytotoxic T-lymphocyte–associated protein 4 (CTLA-4) and programmed cell death protein 1 (PD-1) has become highly effective in the treatment of metastatic melanoma (1). Despite this, mortality from late-stage metastatic melanoma remains very high, with patients having only a median survival of 6 to 10 months due to metastasis resulting from either intrinsic or acquired resistance. Importantly, clinical follow-up has shown that a significant proportion of patients with melanoma who had shown initial response to immunotherapies are developing resistance (2, 3). Therefore, new therapeutic strategies and combinations that circumvent these issues are needed.

Currently, immunotherapies are faced with the challenge of identifying predictive biomarkers to checkpoint inhibitor blockade. There are several different pathways believed to be responsible for poor response or acquired resistance to checkpoint inhibitor blockade (3, 4). Although increased expression of neoantigens expressed on melanoma cells as a result of high frequency and mutational load allow better response to anti-immunosuppressive strategies focusing on checkpoint blockade, a significant proportion of patients with melanoma fail to show durable clinical responses. This is partly due to exogenous and endogenous environmental factors. Epigenetic modification of histones has shown to be malignant drivers of several cancer types, including melanoma (5–9). As epigenetic changes are dynamic, it may be possible to reverse both the malignant process and/or therapy-resistant phenotype by targeting the epigenetic processes that cause malignancy and resistance to checkpoint inhibitor blockade. Recently, results from the ENCORE 601 phase Ib/II clinical trial (NCT02437136) demonstrated that a combination of a histone deacetylase (HDAC) inhibitor with the anti–PD-1 therapy showed clinical responses in anti–PD-1 resistant patients with melanoma. G9a, also known as euchromatic histone lysine N-methyltransferase 2 (EHMT2), is a histone methyltransferase that has a primary role in catalyzing mono- and dimethylation of the lysine residue 9 on histone H3 (H3K9me1 and H3K9me2) in euchromatic regions (10–13). These histone marks are generally associated with transcriptional repression (14). Elevated levels of G9a expression have been observed in many types of human cancers and implicated in mediating resistance mechanisms. G9a knockdown leads to inhibition of cancer cell proliferation and therefore has been suggested as a potential therapeutic target (15, 16). Studies have shown that G9a represses the expression of genes involved in the autophagic process (17–19). Autophagy is a lysosomal catabolic process that serves as an adaptive response to protect cells during stress and has been linked to several important processes such as cellular homeostasis, clearance of intracellular pathogens, and immunity. Because the epigenetic mechanisms are directly involved in determining which genes are activated or repressed, they play an important role in determining the best treatment strategy for each individual patient. This suggests that altering histone modifications may enhance checkpoint inhibitor treatment responses. We hypothesize that targeting G9a has the potential to sensitize patients with melanoma with minimal clinical response to checkpoint inhibition.

In this study, using three independent clinical models of metastatic melanoma, we demonstrate that LC3B (MAP1LC3B) is a predictive biomarker of clinical response to anti–PD-1 therapy. We evaluate the therapeutic efficacy of G9a inhibition and investigated the value of G9a as a sensitizer for anti–PD-1 therapy. We show that G9a is elevated in all genetic subtypes of cutaneous melanoma, and that inhibition of G9a induced significant cell death via an autophagy-related mechanism. Our results suggest G9a histone methyltransferase is an attractive therapeutic target to sensitize melanoma to immune checkpoint inhibitors.

UNC0642 and Bafilomycin A1 (Baf.A1) were purchased from Sigma Aldrich. Anti-mouse PD-1 (CD279) clone RMP1–14 and the rat IgG2a isotype control clone 2A3 were purchased from Bio X Cell. Human melanoma cell lines (Supplementary Table S1) were maintained as described previously (20–23) in RPMI1640 media (Gibco) supplemented with 10% FBS (Sigma Aldrich), in a humidified atmosphere of 5% CO₂ at 37°C. Mycoplasma testing was routinely performed to all cell lines prior to experimentation.

Retroviral constructs expressing short hairpin RNA to G9a (shG9a) or a nonsilencing control (shNS) were used as described previously (24). pBABE-puro mCherry-EGFP-LC3B was obtained from Addgene (RRID:Addgene_22418). The viral supernatants were prepared by cotransfecting the constructs with pUMVC3 (RRID:Addgene_41792) and pVSV-G (RRID:Addgene_138479) into HEK293T cells using Superfect transfection reagent (Qiagen). The supernatants were harvested and used to infect cells in medium, which contained 8 μg/mL polybrene. Cells were selected using 1 μg/mL puromycin.

For proliferation studies, cells (5 × 10³) were seeded in 96-well plates and allowed to adhere overnight, and incubated in fresh growth medium in the presence of either the G9a inhibitor (UNC0642) or the vehicle control (DMSO). Proliferation was evaluated via real-time imaging using IncuCyte Zoom (Essen BioScience) and/or by performing a 3 (4,5 dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay. Cell-cycle analysis was performed using propidium iodide (PI) staining. For PI staining, cells were harvested, washed once with PBS, and fixed in ice cold ethanol for 24 hours. Cells were then washed and resuspended in PBS containing 10 μg/mL PI (Sigma Aldrich) and 25 μg/mL RNase A. Samples were incubated for 1 hour in the dark at room temperature prior to analysis using the FACS Canto A (BD Biosciences). Cell cycle was evaluated for 10,000 single cells using ModFit (Verity).

Whole-cell lysates were prepared using RIPA lysis buffer containing 20 mmol/L Tris, pH 8.0, 150 mmol/L NaCl, 10% glycerol, 1% Nonidet P-40, and Complete EDTA-free protease inhibitor cocktail (Roche). Nuclear extracts were obtained using a high salt extraction buffer [20 mmol/L HEPES (pH 7.9), 0.32 mol/L NaCl, 1 mmol/L EDTA, and 1 mmol/L EGTA) supplemented with 1 mmol/L DTT and protease inhibitor cocktail. Immunoblotting was performed with primary antibodies (Supplementary Table S2) and detected with either HRP-conjugated anti-rabbit or anti-mouse antibodies (Cell Signaling Technology). Protein bands were visualized by applying ECL Prime reagent (GE Healthcare) using the ChemiDoc Touch Imaging System (Bio-Rad).

The melanoma cases in the human skin cutaneous melanoma (SKCM) TCGA dataset were allocated to one of four quartiles based on the expression of EHMT2 (G9a) and/or MAP1LC3B (LC3B) and the survival of these patients was compared. Survival curves were constructed using GraphPad Prism (GraphPad Prism, RRID:SCR_002798), and the log-rank (Mantel–Cox) test was used for statistical comparisons of survival curves. RNA sequencing (RNA-seq) data from 27 melanoma tumors that had been treated with anti–PD-1 therapy were allocated to one of two groups based on the expression of MAP1LC3B and the survival of these patients was compared. Further details of patient data from these two datasets can be found in Supplementary Table S3.

Stable cells expressing mCherry-EGFP-LC3B construct were treated with DMSO, UNC0642 at 5 μmol/L, or 20 nmol/L Baf.A1 for 24 hours. Images were taken with an EVOS FL auto fluorescent microscope (Invitrogen) and quantitated using ImageJ software (ImageJ, RRID:SCR_003070).

Tumor sections were fixed in 4% paraformaldehyde. The antibodies used were G9a (Cell Signaling Technology), LC3B (Santa Cruz Biotechnology), SOX10 (Santa Cruz Biotechnology), and PD-L1 (Abcam). The universal secondary protocol and the 3,3′-Diaminobenzidine (DAB; Biocare Medical) or ImmPact NovaRed Peroxidase (HRP) substrate (Vector Laboratories) were used to detect and amplify the signal. Aperio ImageScope software (ImageScope, RRID:SCR_014311) was used for imaging and quantitation of five nonoverlapping tumor regions and evaluating the number of positive pixels per unit area in each region.

Formalin-fixed, paraffin-embedded melanoma primary tumor biopsies were processed in the BondRX for OPAL staining (Perkin-Elmer) using the instrument protocol: ER2 for 20 minutes at 100°C with Epitope Retrieval Solution (pH6 EDTA-based retrieval solution) followed by probing with rat host CD8a (eBioscience) and visualized with a chicken anti-rat AF-647 secondary antibody. Coverslips were mounted on glass microscope slides with ProLong Clear Antifade reagent (Life Technologies).

For high-throughput microscopy, protein targets were localized by confocal laser scanning microscopy. Single 0.5-μm sections were obtained using an Olympus-ASI automated microscope with 20× lens running ASI software. The final image was obtained by employing a high-throughput automated stage with ASI spectral capture software. Digital images were analyzed using automated ASI software (Applied Spectral Imaging) to automatically determine the distribution with automatic thresholding and background correction to calculate positive cells for expression of markers and the percentage population of cells expression the analyzed proteins.

A tissue microarray (TMA) containing melanoma tumor biopsies from 49 patients with melanoma, collected prior to anti–PD-1 based immunotherapy (either pembrolizumab or nivolumab, with or without ipilimumab) and categorized as responders or nonresponders, as described previously (24), was stained with antibodies against G9a (Abcam) and LC3B (Cell Signaling Technology). All multiplex tyramide labeling was performed using the PerkinElmer Opal 7-Color Tyramide Kit (PerkinElmer) using the cyclic staining method. The TMA slide was scanned using the Vectra 3.0 spectral imaging system (PerkinElmer). Cell segmentation was also carried out using the inForm 4.2.1 image analysis software (inForm, RRID:SCR_019155) (PerkinElmer) and was further analyzed using the FCS express 6 software (De Novo software) to determine the number and intensity of expressing cells in each individual patient sample. ROC curves were constructed using MedCalc for all the endpoints and the cut-off criterion for sensitivity/specificity using the DeLong and colleagues' method (25).

Circulating tumor cells (CTCs) were isolated, as reported previously (26), from Metastatic Melanoma liquid Biopsies employing the RosetteSep Human CD45 Depletion Kit (Stemcell Technologies No. 15162) to remove CD45⁺ cells, using density gradient centrifugation with SepMate-50 (IVD) density gradient tubes (Stemcell Technologies) and Lymphoprep density gradient medium (Stemcell Technologies). To examine the dynamics of G9a and LC3B with chemoresistant stem-like marker for CTCs, ABCB5, CTCs were permeabilized by incubating with 1% Triton X-100 for 20 minutes and were probed with rabbit anti-LC3B, mouse anti-G9a, and goat anti-ABCB5 and visualized with a donkey anti-rabbit Alexa Fluor 488 (Life Technologies), anti-mouse 568 (Life Technologies), and anti-goat 633 (Life Technologies). Coverslips were mounted on glass microscope slides with ProLong Diamond Anti-fade Reagent (Life Technologies). Protein targets were localized by confocal laser scanning microscopy. Single 0.5-μm sections were obtained using a Leica DMI8 microscope using 100× oil immersion lens running LAX software. The final image was obtained by averaging four sequential images of the same section. Digital images were analyzed using ImageJ software to determine either the total nuclear fluorescent intensity, the total cytoplasmic fluorescent intensity, or total fluorescent intensity (TFI).

qRT-PCR and chromatin immunoprecipitation (ChIP) assays were conducted as described previously (27). Briefly, total RNA was isolated from either the tumor cells or from xenografts using TRIzol (Invitrogen) and reverse transcription was performed from 2 μg of total RNA using the Superscript III cDNA synthesis system (Invitrogen). The abundance of mRNA was detected by an ABI VIIA7 system with SYBR Green Master Mix (Life Technologies). Primer pairs were designed to amplify 90 to 150 bp mRNA-specific fragments and were confirmed as unique products by melting curve analysis. The quantity of mRNA was calculated using the ΔΔC_t method and normalized to HPRT. All reactions were performed in triplicates. See Supplementary Table S4 for primer sequences used.

Groups of eight SCID or C57BL/6J mice (RRID: IMSR_JAX:000664) per treatment group were used for xenograft studies to ensure adequate power to detect biological differences. D20 melanoma cells (2 × 10⁶) or B16F10 cells (1 × 10⁵) were mixed in a 1:1 ratio with growth factor–reduced Matrigel (Corning) and injected subcutaneously into the flank of 6- to 8-week-old mice in 100-μL volume (day 0) and treatments, either DMSO or UNC0642 at 5 mg/kg for the SCID mice or DMSO + IgG at 250 μg per mouse, UNC0642 at 5 mg/kg + IgG at 250 μg, anti–PD-1 at 250 μg + DMSO or UNC0642 at 5 mg/kg + anti–PD-1 at 250 μg for the C57BL/6J mice, were administered as indicated in the figure legends. Tumor volumes (width² × length/2) were measured using a digital caliper, and presented as mean ± SEM.

Statistical differences in test and control samples were determined by Student t test and Mann–Whitney two-tailed test, and ANOVA was used for group comparisons. Numerical data were expressed as means ± SEM of independent determinations. Mantel–Cox and χ² tests were also used in the evaluation of the TCGA datasets and melanoma patient TMA analyses. A P value <0.05 was considered statistically significant. Statistical analyses were performed using GraphPad Prism.

All animal experiments were approved by the QIMR Berghofer Medical Research Institute Animal Ethics Committee. The human TMA was constructed from tumor biopsies samples obtained from patients and the Melanoma Biospecimen Tissue Bank, including patients from Royal Prince Alfred Hospital, Westmead Hospital and Melanoma Institute Australia, with ethical approval from the Sydney Local Health District Human Research Ethics Committee (Protocol No. X15–0454 and HREC/11/RPAH/444). All experimental procedures relating to isolating human CTCs from patients were performed in accordance with the guidelines and regulations approved by the ACT Health Research Ethics and Governance Office (Ethics ID ETH.5.16.073). Written informed consent was received from all patients.

As it has been shown previously that G9a (EHMT2) inhibition leads to changes in autophagy (17), and the observation that melanoma cell lines that had a lower basal amount of LC3B protein corresponded to greater sensitivity to G9a inhibitor (Supplementary Fig. S1A–S1C), we investigated the value of LC3B protein and transcript (MAP1LC3B) levels as a predictive biomarker of response to immunotherapy in cutaneous melanoma. We performed an in silico analysis of gene expression data (RNA-seq; ref. 28) where RNA was isolated from tumors from patients with metastatic melanoma prior to anti–PD-1 therapy (pembrolizumab; Fig. 1A). Patients were ranked from high to low based on the expression of MAP1LC3B transcript levels and were divided into tertiles. We compared each tertile in terms of patient survival and found that only the lowest tertile was associated with the worst overall survival (OS). Hence, the comparison was made between the lowest tertile (low, n = 9) against the rest (high, n = 18). A Kaplan–Meier survival curve using the expression of MAP1LC3B showed that patients with higher expression of MAP1LC3B was associated with significantly longer survival (MAP1LC3B high, red) compared with patients with lower expression (MAP1LC3B low, black; Fig. 1B). Patients that responded to the anti–PD-1 treatment had higher expression levels of MAP1LC3B, but lower levels of EHMT2 expression (Fig. 1C), suggesting that tumor expression of EHMT2 and MAP1LC3B can be used to predict response to anti–PD-1 therapy.

Next, we investigated the mRNA expression of both and EHMT2 and MAP1LC3B in the SKCM TCGA melanoma RNA-seq dataset (29) and found that the expression of MAP1LC3B was also inversely correlated to EHMT2 expression (Fig. 1D and E). Using a modified cohort of TCGA patient samples that only contained those samples that represented the first regional metastases, patients were grouped based on quartiles of expression and showed that EHMT2-high patients associated with the worst progression-free survival (Fig. 1F), whereas the expression levels of MAP1LC3B showed no correlation with survival (Fig. 1G). The expression of EHMT2 mRNA did not associate with NRAS or NF1 mutations but was higher in BRAF wild-type versus BRAF-mutant cases, whereas the expression of MAP1LC3B mRNA did not associate with BRAF, NRAS, or NF1 mutation status (Supplementary Fig. S2). Grouping the patients according to the expression of EHMT2 and MAP1LC3B (Supplementary Fig. S3A), we found that patients with high EHMT2 expression were associated with worse relapse-free survival (RFS) compared with all other groups (Supplementary Fig. S3B). Patients with high EHMT2 and low MAP1LC3B expression were associated with the worst OS (Supplementary Fig. S3C). It is noteworthy that the four groups based on the expression patterns of EHMT2 and MAP1LC3B did not differ significantly in their driver mutational subtype status (Supplementary Fig. S3D). Upon performing multivariate survival analysis using EHMT2 or MAP1LC3B mRNA expression alone or combined in comparison with other parameters including disease stage, gender, and mutational status, we found that the expression of EHMT2 alone was able to stratify patients into different prognostic groups in both OS and RFS, while the expression of MAP1LC3B alone did not (Supplementary Table S5). Only 4% (19/473) of patients in the TCGA dataset were treated with checkpoint inhibitor therapy and the substantial level of nonmetastatic tumors may explain the differences observed in the different datasets, suggesting that EHMT2 level is important in overall tumor development, regardless of stage, while MAP1LC3B levels might be more important in treatment response.

We performed IHC staining using specific antibodies to G9a and LC3B on a TMA, containing 40 melanoma samples, collected from patients prior to receiving anti–PD-1 treatment alone or in combination with anti–CTLA-4 treatment to assess G9a and LC3B as predictors of clinical response (Fig. 2A). This cohort was divided into “responder” (R; n = 28), including those that completely or partially responded to the treatment, or “nonresponder” (NR; n = 12) groups. In the responding group, the pretreatment level of LC3B was higher compared with nonresponding group (Fig. 2B). The number and mean intensity of the LC3B-expressing cells were significantly higher in the responding group compared with the nonresponding group (Fig. 2C). On the basis of the cutoffs from ROC curves (Supplementary Fig. S4A and S4B), we found that a high percentage of LC3B⁺ cells associated significantly with better survival due to better response and lower frequency of disease progression while LC3B staining intensity showed similar trends (Fig. 2D and E). On the basis of the pathology reports, the tissue was considered as PD-L1 positive if 1% or greater of PD-L1 staining was observed. Univariate and multivariate analyses were carried out for LC3B, PD-L1, and all other variables available in our cohort (Supplementary Table S6). For multivariate analysis, neither LC3B nor PD-L1 remained significant for survival, suggesting that they do not appear to be independent prognostic factors, meaning that there is an interaction. For response, both LC3B and PD-L1 are significant. For progression, PD-L1 remains significant, while LC3B was borderline significant, and for resistance, only PD-L1 is significant. This suggests that both PD-L1 expression and % LC3B⁺ cells have prognostic value and the combination of both increased the prognostic power (Supplementary Fig. S5). The most benefit from immunotherapy was when both PD-L1 and % LC3B⁺ cells were both positive and the least benefit was observed when both were negative. When either LC3B or PD-L1 are positive, there is benefit, but not to the extent when both are positive. Of note, PD-L1–negative but LC3B-positive patients would benefit from immunotherapy so LC3B can be used to include more patients for treatment who would have missed out if only judging by PD-L1 positivity alone.

We isolated CTCs using a less invasive liquid biopsy from patients with stage IV metastatic melanoma and examined the level of G9a and LC3B using a diagnostically relevant IHC staining method (Fig. 2F). CTC collections in all cases were from patients who had already commenced anti–CTLA-4 (two cycles of monotherapy) followed by anti–PD-1 therapy (one cycle) for at least 3 months. The cohort of patients with melanoma included three groups characterized as (i) complete response (CR); (ii) partial response (PR); and (iii) no response (NR) as per RECIST 1.1 (n = 12 patient samples with four patients/group). Immunofluorescence analysis of G9a and LC3B protein expression on these CTCs showed a higher ratio of LC3B to G9a was strongly associated with CR to checkpoint inhibitor therapy (Fig. 2G). Conversely, the lowest ratio of LC3B to G9a significantly associated with NR and the intermediate ratio was associated with PR to checkpoint inhibitor therapy (Fig. 2H). Together, our analyses suggest that G9a and LC3B protein and transcript levels could be used as potential predictive and response markers to checkpoint inhibitor therapy for patients with melanoma.

We next sought to determine whether inhibition of G9a would lead to elevation of LC3B. The levels of LC3B I and II were examined in melanoma cell lines (D05 and C008) treated with the G9a inhibitor UNC0642 at 5 μmol/L for 24 hours (Fig. 3A). LC3B II protein levels significantly increased in both cell lines following G9a inhibitor treatment as well as following G9a knockdown in D05 cell line (Fig. 3B). Because the accumulation of LC3B II can result from either by the induction of autophagy or inhibition of autophagic flux, we used a pBABE puro mCherry-EGFP-LC3B reporter to examine the impact of G9a inhibitor on the autophagic flux. Although there was an increase in punctate pattern of EGFP-LC3B after treatment with UNC0642 (Fig. 3C; Supplementary Fig. S6), the level of mCherry-LC3B did not decrease with UNC0642 treatment, suggesting a block in autophagic flux, similar to that of cells treated with an autophagy inhibitor, Baf.A1. This was also consistent with the increase in p62 levels that was observed after G9a inhibitor treatment (Fig. 3A). qRT-PCR analysis revealed that the transcript level of MAP1LC3B was induced significantly in both D05 and C008 cell lines with approximately 2-fold induction compared with vehicle control (Fig. 3D). To ascertain whether the inhibition of G9a directly impacts MAP1LC3B expression via methylation of H3K9, ChIP was performed on the MAP1LC3B promoter. We found that there was a greater amount of H3K9me2 at the MAP1LC3B promoter of the two sensitive cell lines (D05 and C008; Fig. 3E). Consistent with the increase in gene expression observed in Fig. 3D, in D05 cells, there was a greater than 4-fold reduction in the repressive mark H3K9me2 concomitant with an increase in RNA Pol II recruitment at the MAP1LC3B promoter following UNC0642 treatment (Fig. 3F). No change in G9a, H3K9me2 and the recruitment of RNA Pol II to the irrelevant region (5 kb upstream of the MAP1LC3B promoter) was observed, suggesting a promoter-specific regulation of MAP1LC3B by G9a. Together, these results demonstrate that G9a inhibition elicits accumulation of LC3B II in melanoma cell lines by blocking autophagic flux as well as inducing MAP1LC3B gene expression via reduction of H3K9 methylation in the promoter region.

Using the TCGA/GTEx databases (30), it is evident that G9a is upregulated in melanoma patient samples compared with normal skin controls (Fig. 4A). We therefore sought to determine whether G9a inhibition impacts survival of melanoma cells in vitro. Cell survival in vitro was significantly attenuated by UNC0642 treatment in the melanoma cell lines, while survival of the normal melanocyte cells was unaffected (Supplementary Fig. S1B). We found that the D05 and C008 cell lines were particularly sensitive to G9a inhibition and were selected to perform further analyses. Real-time cellular imaging analysis using the IncuCyte ZOOM showed that G9a inhibition resulted in a significant reduction in proliferation in the D05 and C008 lines (Fig. 4B–D). We also performed a G9a knockdown (shG9a) analysis in the D05 cell line and observed a significant reduction in cell viability compared with that of control (shNS) cells (Fig. 4E and F). G9a inhibition led to a reduction in cells in both G1 and G₂–M phases but a more striking impact was observed in inducing cell death as indicated by greater than 4-fold increase in cells in the pre-G₁ phase (Fig. 4G). Together, these data suggest that the inhibition of G9a directly impacts melanoma cell survival.

We performed an in vivo experiment using the D20 cell line as the D05 cell line was not tumorigenic. UNC0642 administration led to almost 2-fold reduction in tumor growth and tumor weight (Fig. 4H and I), compared with vehicle-treated control. Quantitative PCR analysis showed a significant increase in expression of the MAP1LC3B gene in tumor RNA isolated from mice treated with UNC0642 compared with the vehicle-treated group (Fig. 4J). Expression of MAP1LC3B and CD274 (PD-L1) in the D20 cell line and xenografts were significantly increased with UNC0642 treatment (Supplementary Fig. S7). IHC analysis of LC3B protein in the excised tumors revealed a higher level of LC3B in the tumors from mice treated with UNC0642 compared with that of mice treated with vehicle (Fig. 4K).

We next examined whether inhibiting G9a enhances the efficacy of immunotherapy using the anti–PD-1 resistant B16F10 mouse melanoma model. We confirmed that UNC0642 treatment increases protein and mRNA levels of Map1lc3b as well as Cd274 transcript levels (Fig. 5A). UNC0642 administration led to a robust reduction in tumor growth and tumor weight (Fig. 5B and C), compared with vehicle or anti–PD-1 treated groups. UNC0642 combined with anti–PD-1 not only further reduced tumor volume compared with UNC0642 treatment alone, but also led to tumor regression. qRT-PCR analysis revealed that a significant increase in expression of the Map1lc3b, Cd274, and Irf1 genes following UNC0642 treatment compared with the vehicle-treated group (Fig. 5D). IHC analysis of LC3B and PD-L1 protein levels in the excised tumors revealed elevated level of LC3B and PD-L1 in the tumors from mice that were treated with UNC0642, or the combination (Fig. 5E). Another BRAF-mutant anti–PD-1 resistant melanoma model (SM1WT1) showed similar results to the treatments (Supplementary Fig. S8A and S8B), and it was found that this cell line had similar basal level of LC3B protein as the B16F10 cell line (Supplementary Fig. S8C).

IHC staining of the B16F10 tumors revealed a significant increase in CD8⁺ T cells in tumors derived from mice treated with both G9a inhibitor and anti–PD-1 (Fig. 5F). CD8⁺ T-cell tumor infiltration is a favorable prognostic factor for a broad spectrum of human cancers, either through direct cytolytic action on tumor cells or by releasing IFNγ and TNFα. To determine whether the expression of MAP1LC3B is correlated with the expression of T-cell signature genes, we examined the TCGA dataset and found that MAP1LC3B expression is inversely correlated with the expression of EHMT2, and positively correlated with T-cell signature genes (Fig. 5G). G9a inhibition has been shown to have antitumor activities by the reexpression of endogenous retroviruses in preclinical cancer models and thereby induce viral mimicry and IFN signaling to increase tumor immunogenicity and recognition (31). Consistent with other studies showing that the induction of endogenous retroviruses acting synergistically with checkpoint inhibitor therapies (32, 33) and IFN signaling increasing the expression of PD-L1 in melanoma (34, 35), G9a inhibition alone significantly increased Ifng expression and a further increase was observed when combined with anti–PD-1 (Fig. 5H). Interestingly, upon analyzing RNA-based melanoma phenotypes, lower expression of EHMT2 was enriched in the immune phenotype compared with the other two phenotypes (Supplementary Fig. S9). To ascertain whether G9a inhibition can induce IFN signaling on tumor cells in the absence of immune cells, we performed a gene set enrichment analysis on genes significantly upregulated by G9a inhibitor treatment on D05 melanoma cells in vitro, we found that the top two pathways enriched were TNFα signaling via NFKB and IFNγ response (Supplementary Table S7; Supplementary Fig. S10). All 11 genes in common between these two pathways (CCL2, PTGS2, TNFAIP3, NFKBIA, IFIH1, SOD2, IRF1, ICAM1, DDX58, BTG58, and NAMPT) were robustly induced upon G9a inhibition (Fig. 5I and J). Consistently, qRT-PCR analysis of six of these genes showed a significant increase upon UNC0642 treatment compared with the vehicle control (Fig. 5K). These results demonstrate that G9a inhibition enhances antitumor activity of checkpoint inhibitor blockade by modulating autophagy and IFN signaling.

We and others have previously shown that inhibiting G9a using small-molecule inhibitors such as UNC0642 or BIX01294 has efficacy in inducing cell death and blocking metastatic potential in other cancer types (15, 24, 36). In this study, we analyzed the impact of G9a inhibition on melanoma cell lines, the physiologic consequences on melanoma cells on inhibiting G9a, the efficacy of the treatment in vivo, and the importance of G9a in melanoma in terms of its association with patient survival using different melanoma gene expression datasets. It was shown that the inhibition of G9a was effective in inducing cell death in diverse melanoma cell types, and this treatment was effective in reducing tumor growth in vivo. G9a inhibition in melanoma cells modulated autophagic processes, suggesting that the mode of cell death is mediated by the autophagic pathway. Although the exact role of autophagy in human carcinogenesis is still debatable, autophagy has been shown to have a tumor-suppressive role by removing damaged organelles and misfolded proteins that could lead to a build-up of oxidative stress, activation of the DNA damage response, and increased genomic instability, which have all been shown to be involved in cancer initiation and progression (37–39).

We not only demonstrate that G9a inhibition reduces cell proliferation in vitro and tumor growth in vivo by regulating autophagy but also the clinical utility of G9a and LC3B as patient stratification markers. Although EHMT2 levels alone can stratify patients with melanoma into distinct prognostic groups, when combined with its target gene, MAP1LC3B, the stratification power is significantly improved. The fact that MAP1LC3B gene expression in patients with metastatic melanoma prior to anti–PD-1 therapy predicted survival not only emphasizes the potential value of MAP1LC3B gene expression as a predictive marker of immunotherapy response, but also suggests that there may be therapeutic potential in modulating MAP1LC3B expression by targeting G9a (Fig. 6A). Recently, induction of the autophagic pathway has been shown to be beneficial for anti–CTLA-4 blockade in patients with melanoma (40), whereas inhibition of autophagy, or the blockage of autophagic flux, has been shown to induce PD-L1 expression (41). This study is limited; despite the interesting association LC3B has on anti–PD-1 response in the tested patient cohorts, the relatively small number of patients in each group means that these findings are still exploratory and would need to be further validated using a larger cohort. These findings suggest that a G9a inhibitor has the potential to be used as a novel agent to treat melanoma, potentially as an adjuvant to anti–PD-1 or anti–CTLA-4 therapies, either potentiating the efficacy or extending the proportion of patients with melanoma responding to checkpoint inhibitor treatment (Fig. 6B).

F. Al-Ejeh reports a patent for WO2020047604A1 pending. I.P. Silva reports personal fees from BMS, MSD, and Roche outside the submitted work. G.V. Long reports personal fees from Highlight Therapeutics S.L., Merck Sharpe & Dohme, NovartisPharma AG, Pierre Fabre, QBiotics Group Limited, Regeneron Pharmaceuticals, SkylineDX B.V., Specialised Therapeutics Australia Pty Ltd, and OncoSec outside the submitted work. R.A. Scolyer reports grants from National Health and Medical Research Council of Australia, National Health and Medical Research Council of Australia, personal fees from QBiotics, Novartis, Merck Sharp & Dohme, Bristol-Myers Squibb, Amgen, GlaxoSmithKline, NeraCare, and Myriad Genetics outside the submitted work. N.K. Hayward reports grants from NHMRC during the conduct of the study. J.S. Lee reports grants from Queensland Emory Development, Australian Skin and Skin Cancer Centre, and other from QIMR Berghofer Medical Research Institute during the conduct of the study, as well as a patent for WO2020047604A1 pending. No disclosures were reported by the other authors.

G.M. Kelly: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. F. Al-Ejeh: Conceptualization, data curation, formal analysis, investigation, writing–original draft, writing–review and editing. R. McCuaig: Data curation, formal analysis, investigation, visualization, writing–original draft, writing–review and editing. F. Casciello: Data curation, formal analysis, investigation, writing–original draft, writing–review and editing. N.A. Kamal: Formal analysis. B. Ferguson: Formal analysis, visualization. A.L. Pritchard: Resources, data curation, writing–original draft, writing–review and editing. S. Ali: Resources, formal analysis, methodology. I.P. Silva: Resources, methodology. J.S. Wilmott: Resources, data curation, writing–original draft, project administration, writing–review and editing. G.V. Long: Resources, data curation, supervision, methodology, writing–review and editing. R.A. Scolyer: Resources, data curation, supervision, writing–review and editing. S. Rao: Resources, data curation, supervision, investigation, visualization, writing–original draft, writing–review and editing. N.K. Hayward: Resources, data curation, supervision, writing–review and editing. F. Gannon: Resources, supervision, funding acquisition, writing–review and editing. J.S. Lee: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

We thank Drs. Glen Boyle (QIMR Berghofer) for providing normal melanocytes, Mark Smyth for the B16F10 and SM1WT1 cell lines, and Tam Hong Nguyen and QIMR Berghofer histology services for their contribution. This work was supported by a QIMR International Research Fellowship (J.S. Lee), an Australian Research Council Future Fellowship (F. Al-Ejeh; ARC-FT130101417), an Australian Skin and Skin Cancer Centre Enabling Grant (J.S. Lee), and a National Health and Medical Research Council of Australia (NHMRC) Fellowship (N.K. Hayward).

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.