Distant metastases of melanoma exhibit varying extent of intrapatient proteogenomic heterogeneity

Abstract

Background: Metastatic melanoma is a highly aggressive disease with poor survival rates despite recent therapeutic advancements with immunotherapy. The proteomic landscape of advanced melanoma remains poorly understood, especially regarding proteomic heterogeneity across metastases within patients.

Methods: We collected 83 melanoma metastases from 19 different metastatic sites in 24 patients with advanced metastatic melanoma almost exclusively from the pre-immunotherapy era, using semi-rapid autopsies. The metastases were subjected to histopathological evaluation, RNA-sequencing and mass spectrometry-based proteomics for protein quantitation and non-reference peptide (NRP) sequence detection using a proteogenomic data integration approach.

Results: NRPs associated with mutations frequently occurred in proteins related to focal adhesion, vesicle-mediated transport, MAPK signalling and immune response pathways across the cohort. Intrapatient heterogeneity was negligible when considering morphology and driver gene mutation status but was substantial at the proteogenomic level. This heterogeneity was not driven by metastasis location, albeit liver metastases exhibited distinct proteogenomic patterns, including upregulation of metabolic pathways. Cluster analysis outlined four proteomic clusters (C1-4) of the metastases, characterised by the upregulation of cell cycle and RNA-splicing (C1), mitochondrial processes (C3), extracellular matrix (ECM) and immune pathways (C2) and ECM and vesicle-mediated transport pathways (C4). Around two-thirds of patients had metastases that had strongly distinct phenotypes. Patients in our cohort whose metastases were primarily assigned to clusters C1 and C3 exhibited shorter overall survival than patients whose metastases were categorised mainly into the C2 and C4 clusters.

Conclusion: Our unique multi-metastasis cohort captured the proteogenomic heterogeneity of immunotherapy-naïve melanoma distant metastases, establishing a foundation for future studies aimed at identifying novel therapeutic targets to complement current immunotherapies.

Key points: Comprehensive proteogenomic profiling of post-mortem melanoma metastases, collected primarily before the immunotherapy era. Description of 1177 protein sequence variants predicted by RNA-Seq and validated via mass spectrometry-based proteomics. Empirical evidence of prominent intrapatient heterogeneity, driven by heterogeneous protein expression related to cell cycle- and mitochondrial processes, immune system and extracellular matrix organization.

Keywords: RNA‐sequencing; distant metastasis; histopathology; mass spectrometry‐based proteomics; melanoma; post mortem; proteogenomics.

FIGURE 1

Overview of the metastatic melanoma post mortem cohort. (A) Overview of the patients' clinical data, including details on the number and localisation of their individual metastases, along with the specific driver mutations identified in each metastasis via targeted sequencing. The NRAS and cKIT mutation status is unknown for the metastasis labelled as BRAF WT. (B) Kaplan–Meier curves for patients with or without developing spleen metastases before their death. (C) Examples of histology images from distinct metastasis locations. The representative haematoxylin and eosin images of the melanoma metastases at selected regions were taken at 10× magnification with the QuPath v.4.0 software image viewer.

FIGURE 2

Multi‐omics analysis of the melanoma metastases. (A) Summary of the histological annotations (tumour, adjacent tissue, necrosis and lymphocyte percentage) for the individual metastasis samples, highlighting samples that were excluded from proteomic and/or RNA‐Seq analysis. (B) Pathways upregulated in samples with higher tumour percentages across both proteomic and RNA‐Seq data. The pre‐ranked GSEA results for the proteomic and RNA‐Seq data are displayed on the x‐ and y‐axis, respectively. Proteins and transcripts (both collapsed to gene IDs) were ranked according to the correlation between their abundance/expression levels and the tumour percentage of each sample. (C) Distribution of gene‐wise Spearman correlation coefficients (rho‐s) between transcript and protein abundance. Genes with a strong positive correlation are highlighted in red. (D) Summary of RNA‐Seq and proteomic data quality per metastasis sample. Metastases are grouped by patient.

FIGURE 3

Insights into the proteogenome of melanoma metastases. (A) Summary of the non‐reference peptides (NRPs) on a circular plot. NRPs are grouped based on the number of patients in which they were detected. ‘AF high’ indicates AF >1% for an NRP. Infr, in‐frame; SJ, splice junction. (B) Enriched pathways for proteins with at least one identified NRP. (C) Enriched pathways (red nodes) for proteins with novel single amino acid mutations (grey nodes). (D) Proteins with significantly more novel single amino acid mutations detected than expected based on the reference population. The number of detected novel and frequent mutations are coloured separately.

FIGURE 4

Proteogenomic differences associated with metastasis location. (A) Number of significantly upregulated proteins in distinct metastasis locations. Upregulated proteins that are tissue‐specific or not tissue‐specific according to the Human Protein Atlas (HPA) are displayed in separate rows. Information on spleen tissue specificity was not available in HPA. (B) Genes of non‐reference peptides more frequently detected in a metastasis location compared to other locations. (C) Pathway‐level signatures of metastasis locations as detected by pre‐ranked gene set enrichment analysis (GSEA) based on the global proteomics data. White cells depict results where GSEA p > .05, blue or red‐coloured cells without a star indicate p < .05 (normalised enrichment score [NES] lower than 0 or higher than 0, respectively) and cells with a star indicate adj. p < .05.

FIGURE 5

Proteogenomic heterogeneity of melanoma metastases. (A) Consensus matrix with number of clusters (k) = 4, as generated by the consensus clustering algorithm (partition around medoids algorithm with Pearson distance). (B) Alluvial diagram showing cluster assignment transitions for the samples across different values of k. Each vertical block represents the samples’ cluster assignment at k = 2, 3, 4 and 5. Samples (horizontal lines) are coloured based on their cluster membership at k = 4, and flows between the vertical blocks illustrate how samples are reassigned as the number of clusters changes. (C) Heatmap of metastasis clusters displaying upregulated proteins (top heatmap) and non‐reference peptides (NRPs) more frequently detected (bottom heatmap) in each cluster. The metastases, proteins and NRPs are clustered based on Euclidean distance and complete linkage. Metastases are annotated with their histological features, cluster assignment and various clinical data, while protein and NRP clusters are annotated with enriched pathways.

FIGURE 6

Prognostic implications of proteomic heterogeneity. (A) Boxplots showing differences in single‐sample scores across the proteomic clusters. From left to right: Immune/Keratin/MITF‐low TCGA subtype scores, mesenchymal scores and tumour immune dysfunction and exclusion (TIDE) scores. Kruskal‐Wallis and pairwise Wilcoxon test p values are shown on the top. (B) Pathways significantly associated with overall survival (OS) in our cohort, and their relationship to the proteomic clusters and survival relevance in external cohorts., , , , Survival relevance for the pathways was assessed with Cox regression analyses followed by pre‐ranked gene set enrichment analysis (GSEA). The heatmaps show normalised enrichment scores from GSEA, with stars indicating significance. Normalised enrichment score (NES) >0 indicates that the pathway was upregulated in patients with shorter OS. The results on the external cohorts are grouped based on the applied therapy regimen, including anti (a)‐CTLA‐4, a‐PD‐1, combination therapy (combo). The TCGA cohort was not collected in the context of immunotherapy response. Missing NES values for a gene set indicate that <10 members of the gene set were quantified in the dataset. (C) Overview of cluster assignments, TIDE scores and patient‐level clinical data for each metastasis. Metastases are grouped by patient, and the dendrograms are shown to demonstrate the varying levels of intrapatient similarities based on their protein abundance profile. Both the proteins and metastases are clustered based on Euclidean distance and complete linkage. The protein abundance matrix is vertically compressed. Patients are ordered based on the number of clusters across which their metastases are distributed. (D) Kaplan–Meier curves for patients who have more than 50% of their metastases classified into C1 and C3 versus patients with a maximum of 50% of their metastases classified into C1 and C3.