Innate immune checkpoint inhibitor resistance is associated with melanoma sub-types exhibiting invasive and de-differentiated gene expression signatures

Abstract

Melanoma is a highly aggressive skin cancer, which, although highly immunogenic, frequently escapes the body's immune defences. Immune checkpoint inhibitors (ICI), such as anti-PD1, anti-PDL1, and anti-CTLA4 antibodies lead to reactivation of immune pathways, promoting rejection of melanoma. However, the benefits of ICI therapy remain limited to a relatively small proportion of patients who do not exhibit ICI resistance. Moreover, the precise mechanisms underlying innate and acquired ICI resistance remain unclear. Here, we have investigated differences in melanoma tissues in responder and non-responder patients to anti-PD1 therapy in terms of tumour and immune cell gene-associated signatures. We performed multi-omics investigations on melanoma tumour tissues, which were collected from patients before starting treatment with anti-PD1 immune checkpoint inhibitors. Patients were subsequently categorized into responders and non-responders to anti-PD1 therapy based on RECIST criteria. Multi-omics analyses included RNA-Seq and NanoString analysis. From RNA-Seq data we carried out HLA phenotyping as well as gene enrichment analysis, pathway enrichment analysis and immune cell deconvolution studies. Consistent with previous studies, our data showed that responders to anti-PD1 therapy had higher immune scores (median immune score for responders = 0.1335, median immune score for non-responders = 0.05426, p-value = 0.01, Mann-Whitney U two-tailed exact test) compared to the non-responders. Responder melanomas were more highly enriched with a combination of CD8+ T cells, dendritic cells (p-value = 0.03) and an M1 subtype of macrophages (p-value = 0.001). In addition, melanomas from responder patients exhibited a more differentiated gene expression pattern, with high proliferative- and low invasive-associated gene expression signatures, whereas tumours from non-responders exhibited high invasive- and frequently neural crest-like cell type gene expression signatures. Our findings suggest that non-responder melanomas to anti-PD1 therapy exhibit a de-differentiated gene expression signature, associated with poorer immune cell infiltration, which establishes a gene expression pattern characteristic of innate resistance to anti-PD1 therapy. Improved understanding of tumour-intrinsic gene expression patterns associated with response to anti-PD1 therapy will help to identify predictive biomarkers of ICI response and may help to identify new targets for anticancer treatment, especially with a capacity to function as adjuvants to improve ICI outcomes.

Keywords: cancer-associated fibroblast; de-differentiation; gene expression signatures; immunotherapy; melanoma; neoantigen; neural crest-like; tumour mutation burden.

Figure 1

Tumour mutation analysis between responding and non-responding melanomas. Driver mutations in pre-treatment melanomas of responding versus non-responding patients prior to anti-PD-1 therapy. The heatmap represents only 31 samples because these samples passed an in-built QC (quality control) test with ≥80% depth of read uniformity in sequencing using the Oncomine Tumour Mutation Load – w3.0 – DNA – Single Sample in the Ion Reporter cloud server.

Figure 2

Correlation of mutation burden with neoantigens and immune score analysis. (A) Pearson correlation between tumour mutation burden and expressed neoantigens in melanomas. (B) Pearson correlation between tumour mutation burden and immune score in melanomas. (C) Beta-2-microglobulin (B2M) expression level in responder and non-responder tumours in log-normalised expression value (p = 0.06, not significant, Mann-Whitney U). Here, because RNA-Seq was performed on twenty samples, it was only possible to match twenty samples from TMB and RNA-Seq data (responder = 10, non-responder = 10) were used to perform Pearson correlation analysis. On the horizontal axis in C), R, responder; N, non-responder.

Figure 3

Deconvolution study for malignant cells, cancer associated fibroblasts and immune score between responding (n = 10) and non-responding (n = 10) melanomas. Significance tests were performed using Mann–Whitney U test. On the horizontal axis, R, responder; N, non-responder; CAF, cancer associated fibroblast; and MEScore, micro-environment score.

Figure 4

Deconvolution study to estimate different immune cell abundance in responding (n = 10) and non-responding (n = 10) melanoma tissues. Significance tests were performed using Mann–Whitney U test. Here, R, responder; N, non-responder, DC, dendritic cells; NK cell, Natural killer cell; Th1, T helper cell 1; Th2, T helper cell 2; Tregs, regulatory T cells.

Figure 5

Transcriptomic states of responding (n = 10) versus non-responding (n = 10) melanomas. The non-responder group of melanomas exhibited undifferentiated and neural crest-like gene expression signatures. In contrast, the responder group of melanomas mainly exhibited melanocytic and transitory gene expression signatures. Values are in z-score. Here, Pt, patient; R, responder; N, non-responder.

Figure 6

Gene set enrichment analysis (GSEA). The upregulated GSEA groups in the responder group of patients were enriched for genes involved in the immune response pathway and transcription, and the down-regulated genes were enriched for genes in differentiation, viral mimicry, and metabolic pathways. Number of responders = 10 and non-responders = 10. In total, 34 gene sets were identified in GSEA using CAMERA test. The p-value is <0.05.

Figure 7

NanoString analysis of pathway-associated gene expression differences between responder (n = 4) and non-responder (n = 4) melanomas to anti-PD1 therapy. Values are in z-score. Here, Pt, patient; R, responder; N, non-responder.

Figure 8

Enrichment score analysis. Barcode plots show that genes involved in the epithelial mesenchymal transition and viral mimicry detection were upregulated in non-responders whereas differentiation-associated genes and proliferative signature genes were downregulated. Genes expressed in the profile datasets were ranked by log2 fold changes (high-expression/low-expression) in the responders (left side, n = 10) and non-responders (right side, n = 10). Overall shifts in the genes (represented by vertical bars) towards the left or right give an indication of the overall gene level for specific pathways rather than at the individual gene level. Above each barcode plot enrichment scores are shown. Positive and negative values for enrichment scores mean positive and negative enrichment, respectively. Here, R, responder; N, non-responder.

Figure 9

Hypothetical presentation of responding tumour microenvironment versus non-responding tumour microenvironment. The upper panel is representing melanoma de-differentiation trajectory and the lower panel is focusing on the interplay between melanoma cell types and immune cell composition. This image was generated using BioRender.com.