Transcriptional downregulation of MHC class I and melanoma de- differentiation in resistance to PD-1 inhibition

Abstract

Transcriptomic signatures designed to predict melanoma patient responses to PD-1 blockade have been reported but rarely validated. We now show that intra-patient heterogeneity of tumor responses to PD-1 inhibition limit the predictive performance of these signatures. We reasoned that resistance mechanisms will reflect the tumor microenvironment, and thus we examined PD-1 inhibitor resistance relative to T-cell activity in 94 melanoma tumors collected at baseline and at time of PD-1 inhibitor progression. Tumors were analyzed using RNA sequencing and flow cytometry, and validated functionally. These analyses confirm that major histocompatibility complex (MHC) class I downregulation is a hallmark of resistance to PD-1 inhibitors and is associated with the MITF^low/AXL^high de-differentiated phenotype and cancer-associated fibroblast signatures. We demonstrate that TGFß drives the treatment resistant phenotype (MITF^low/AXL^high) and contributes to MHC class I downregulation in melanoma. Combinations of anti-PD-1 with drugs that target the TGFß signaling pathway and/or which reverse melanoma de-differentiation may be effective future therapeutic strategies.

Fig. 1. Predictive performance of transcriptome signatures and PD-1 inhibitor response heterogeneity.

a Details of the 94 melanoma biopsies analyzed in this study, including 53 pre-treatment (PRE) and 41 on-treatment samples. Of the 53 pre-treatment tumor specimens analyzed, 31 were obtained from patients who subsequently underwent complete response (CR) or partial response (PR) by irRC criteria and 22 pre-treatment biopsies were obtained from patients who had stable disease (SD) or progressive disease (PD) by irRC. b Venn diagram showing the 68 patients included in this study and the distribution of pre-treatment, responding and progressing tumor specimens (n = 94). c Immune-predictive transcriptome scores derived for each PRE-treatment melanoma biopsy (n = 44) and patient response data were used to generate receiver operator characteristic (ROC) curves measuring the performance of each indicated signature in predicting PD-1 inhibitor responses in our patient cohort. The resulting AUCs and p values are tabulated. The signatures applied to our dataset were derived from the following references: IPRES signature, IMPRES signature, CD8A/CSF1R ratio, 18-immune gene set, TIDE, CYT score and CIBERSORT estimated relative proportion of CD8+ T cells (see Supplementary Data 6). d CT scans from patient 45. Tumor metastases pre-treatment and on PD-1 inhibitor therapy (week 12 and 24) measured by CT images are shown. Regions of interest in CT images are circled in red. Top images show new lesion at week 12 that continued growing in size at week 24. Middle images show core biopsied lesion that underwent partial response. Lower images show pre-existing lesion that initially responded at week 12 but progressed by week 24. e CT scans from patient 49. Regions of interest in CT images are circled in red, and show partial response of large, inflamed pre-treatment inguinal LN metastasis (upper images) and the appearance of a new, subcutaneous buttock metastasis on treatment (week 8; lower images). Despite excision of the new metastasis, there were multiple new metastases in bone and lymph node on second restaging. Scale bar is shown.

Fig. 2. Immune cytolytic activity in longitudinal melanoma biopsies.

a CYT scores for the 44 PRE, 6 RES and 29 PROG melanoma biopsies. Box plots show the median and interquartile ranges, and CYT scores were compared using Kruskal-Wallis with Dunn’s multiple comparison test. Dotted line aligns with the lowest RES tumor CYT score. n.s, not significant. b CYT score, clinicopathological and response details are shown for each of the 79 melanoma biopsies with RNA sequence data derived from 55 patients. The 6 responding on PD-1 inhibitor treatment biopsies are boxed. Where patients had multiple on treatment tumor biopsies these are shown as individual colored boxes within the ‘On PD-1’ and ‘Biopsy site’ rows.

Fig. 3. HLA-A transcript downregulation associated with markers of de-differentiation.

a Scatter plot showing the Pearson correlation coefficient of CYT score with the expression of MHC class I genes, HLA-A, -B, -C and B2M in responding (RES; n = 6), and pre-treatment (PRE; n = 44) and progressing biopsies (PROG; n = 29). b Plots showing expression of HLA-A and SNAI1 in the CYT score-matched tumors (n = 38) with high or low HLA-A transcript expression. FDR-adjusted p-values (q) calculated using limma test. The HLA-A low melanoma tumor derived from patient 53 was found to express a STAT1^S316L mutation (highlighted). c Heat map showing differentially expressed genes (FDR adjusted p-value < 0.001 are shown) between CYT score-matched tumors (n = 38) with low or high HLA-A transcript expression. CYT score is also shown and HLA-A, SNAI1 and NGFR genes are highlighted. Best irRC response is also shown. d Correlation matrix of SNAI1 gene expression with ssGSEA scores derived from the Hallmark gene set collection and stromal cell-specific transcriptome signatures^, in 79 melanoma biopsies. The Spearman rank correlation coefficients are shown within the matrix, and the false discovery adjusted p-value was <0.01 for all signatures shown (see Supplementary Data 7). e Subset of top scoring genesets (GSEA PreRanked; Hallmark gene set collection and stromal cell-specific transcriptome signatures^,) upregulated in the tumors with low HLA-A transcript expression compared to HLA-A high expressing tumors.

Fig. 4. Immune checkpoint resistance in T-cell inflamed melanoma.

a Cell surface expression of HLA-ABC (relative to HLA-ABC in tumor-infiltrating lymphocytes) in melanoma cells from fresh dissociates of tumors derived pre-treatment (PRE) and progressing (PROG) on PD-1 inhibition. Solid line represents median and dotted line set at Y = 1. Patient-matched PRE and PROG tumors are connected with a dotted line. b Representative histograms showing levels of HLA-ABC expression in melanoma and tumor-infiltrating immune cells in PRE and PROG tumor dissociates. Red histograms show HLA-ABC expression in melanoma cells and blue histograms in tumor infiltrating tumors (TILs). Each tumor driver oncogene is also indicated. c Representative histograms showing cell surface expression levels of B2M and HLA-ABC in WMD-084 melanoma cells, 72 h post transduction with B2M-specific shRNA molecules (left panel). IFNγ production 72 h after co-culture of B2M silenced WMD-084 melanoma cells with the patient-matched tumor-infiltrating lymphocytes expanded from the same tumor biopsy. IFNγ was measured by ELISA (right panel). Data are means ± s.d. and individual data points represent the average of technical triplicates. Data were compared using one-way ANOVA with the Geisser-Greenhouse correction, *p < 0.05. d IFNγ production 72 h after co-culture of WMD-084 melanoma cells with the patient-matched TILs expanded from the same tumor biopsy. Lymphocytes were pre-treated for 1 h with 10 µg/ml HLA-ABC blocking or an isotype-matched antibody prior to co-culturing. Isotype antibody-treated control cells were compared to HLA-ABC blocking antibody-treated data using a paired t-test, *p < 0.05.

Fig. 5. TGFß promotes HLA-ABC downregulation at baseline and in response to IFNγ.

a IFNγ-mediated induction (IFNγ-treated/vehicle-treated control) of cell surface HLA-ABC in MITF^high/AXL^low or /MITF^low/AXL^high short-term PD-1 PROG melanoma cell lines. Each dot represents one cell line and HLA-ABC induction was measured by flow cytometry 24 h after treating cultures with vehicle control or 1000 U/ml IFNγ. Box plots show the median and interquartile ranges, and data were compared using Mann-Whitney test. b Cell surface expression (median fluorescence intensity; MFI) of HLA-ABC in WMD-084, SCC14-0257 and SMU17-0132 melanoma cells treated with vehicle (Control), 1000 U/ml IFNγ and/or 10 ng/ml TGFß for 72 h. Data (mean ± s.d.) were compared using one-way ANOVA with the Geisser-Greenhouse correction. c Expression of de-differentiation markers AXL, N-cadherin and SNAIL in WMD-084, SCC14-0257 and SMU17-0132 melanoma cells treated with vehicle (Control), 1000 U/ml IFNγ- and/or 10 ng/ml TGFß for 72 h. d IFNγ production after co-culture of TGFß pre-treated (10 ng/ml for 72 h) melanoma cells with the patient-matched tumor-infiltrating lymphocytes expanded from the same tumor biopsy. IFNγ was measured by flow cytometry. Data (mean ± s.d.) show relative IFNγ expression in T cells (TGFß pre-treated/BSA pre-treated) after background subtraction (spontaneous IFNγ production on immune cell-only cultures). Paired BSA-treated vs TGFß-treated data were compared using paired t-test, **p < 0.01, *p < 0.05.