Inherited mitochondrial genetics as a predictor of immune checkpoint inhibition efficacy in melanoma

Abstract

Response to immune checkpoint inhibitors (ICIs) in metastatic melanoma (MM) varies among patients, and current baseline biomarkers predicting treatment outcomes are limited. As mitochondrial (MT) metabolism has emerged as an important regulator of host immune function, we explored the association of host MT genetics (MT haplogroups) with ICI efficacy in 1,225 ICI-treated patients with MM from the clinical trial CheckMate-067 and the International Germline Immuno-Oncology Melanoma Consortium. We discovered and validated significant associations of MT haplogroup T (HG-T) with resistance to anti-programmed cell death protein-1-based ICI (both single-agent and combination) and have shown that HG-T is independent from established tumor predictors. We also found that patients belonging to HG-T exhibit a unique nivolumab-resistant baseline peripheral CD8⁺ T cell repertoire compared to other MT haplogroups, providing, to our knowledge, the first link between MT inheritance, host immunity and ICI resistance. The study proposes a host blood-based biomarker with stand-alone clinical value predicting ICI efficacy and points to an ICI-resistance mechanism associated with MT metabolism, with clinical relevance in immuno-oncology.

Fig. 1. NIVO treatment efficacy by European haplogroup.

Dark bars represent participants with NIVO CB (CR, PR or dSD). Light bars represent participants with NIVO NCB (pSD or PoD). a, Phylogenetic tree depicting MT-HG lineages and CM-067 NIVO discovery cohort treatment efficacy (n = 115 patients) with CB/NCB frequencies (counts) for the major European haplogroups (H, U/K, T, J, V, I, X, W). Inset shows CB/NCB proportions within each major European haplogroup (H, J, T, U/K). b, CM-067 and IO-GEM NIVO treatment efficacy frequencies by European haplogroup (H, U, T, J, K and Other, including W, R, V, X, I). Left to right (from top left), CM-067 NIVO discovery cohort (n = 115), CM-067 NIVO validation cohort (n = 82), CM-067 pooled (n = 197) and IO-GEM validation cohort (n = 174). c, Forest plot from logistic regression analysis showing the ORs, log odds, CIs and two-tailed P values when comparing NIVO treatment outcomes for HG-T versus other MT-HGs. Black boxes represent the point estimate for the main effect size (log odds) in discovery and validation cohorts and the lines extending from them represent the 95% CIs. Diamonds represent the log odds point estimate for pooled analyses estimating the overall magnitude of effect. Point estimates to the left of the dashed line indicate increased odds of NIVO CB, while point estimates to the right indicate increased odds of no NIVO CB. CIs that do not cross the dashed line indicate a statistically significant point estimate (P < 0.05). Source data

Fig. 2. ICI treatment efficacy in IPI, NIVO or IPI-NIVO cohorts comparing HG-T and other (H, U, J, K, W, R, V, X and I) MT-HGs.

a, Meta-analysis proportions of CB and NCB comparing HG-T (n = 113) and other MT-HGs (n = 1,087) for CM-067 and IO-GEM IPI cohort (CTLA-4); CM-067 and IO-GEM NIVO cohort (PD-1); CM-067 and IO-GEM COMBO cohort. b–d, Forest plots from logistic regression analysis illustrating the improvement in ICI outcomes across treatments for HG-T and other MT-HGs for the patient population in a, showing the ORs, log odds, CIs and two-tailed P values. When a treatment is shown to be a significant improvement over another, it is not presented again in subsequent comparisons. Comparison groups are as follows: other MT-HGs treated with IPI (n = 431), NIVO (n = 307) or COMBO (n = 349); HG-T treated with IPI (n = 46), NIVO (n = 39) or COMBO (n = 28). b, Logistic regression comparing ICI treatment outcomes to outcomes for other MT-HGs treated with IPI (n = 431); point estimates to the left of the dashed line indicate an improvement in outcomes over those of other MT-HGs treated with IPI. For example, for other MT-HGs, treatment outcomes in NIVO were a significant improvement over treatment outcomes in IPI (row 1), while there was no statistical difference in treatment outcomes for patients belonging to HG-T who were treated with NIVO compared to other MT-HGs treated with IPI (row 4). c, Logistic regression comparing ICI treatment outcomes to outcomes for other MT-HGs treated with NIVO (n = 307). d, Logistic regression comparing ICI treatment outcomes to outcomes for other MT-HGs treated with COMBO (n = 349). Black boxes represent the point estimate for the main effect size (log odds) and the lines extending from them represent the 95% CIs. Point estimates to the left of the dashed line indicate increased odds of ICI CB compared to the reference (patients belonging to other European haplogroups who were treated with IPI, NIVO or COMBO), while point estimates to the right indicate increased odds of no ICI CB compared to the reference. CIs that do not cross the dashed line indicate a statistically significant point estimate (P < 0.05). CB (CR, PR, dSD); NCB (pSD, PoD). Source data

Fig. 3. Tumor markers and haplogroups in CM-067.

a, Correlation of tumor markers with clinical variables including log-transformed TMB, tumor PD-L1 status, composite ‘immune’ score (CD8⁺ T cell tumor infiltration > median, tumor IFNγ expression score > 0, or both), haplogroup (HG-T versus others) and sex (n = 255). b, Distribution of tumor markers stratified by haplogroup and ICI response for IFNγ score (n = 166), log-transformed TMB (n = 376), percentage tumor infiltration by CD8⁺ T cells (n = 182) and PD-L1 status (n = 476). Box plots depict lower and upper hinges (25th to 75th percentiles), median value (central line), whiskers extending from the hinges (largest and smallest values no more than 1.5 times the interquartile range) and outlying points outside this range (solid points beyond the whiskers). P values derived from Kruskal–Wallis tests. Source data

Fig. 4. scRNA-seq analysis for pretreatment circulating CD8⁺ T cells from CM-067 NIVO patients (n = 21) and CM-915 patients (n = 31).

a, UMAP dimensionality reduction plot (CM-067, n = 21 patients). b, Bubble plot showing per-cluster cell-type classification based on T cell marker gene expression (CM-067, n = 21 patients). c, Box plots showing single-cell CD8⁺ T cell-type proportions stratified by NIVO treatment response (CB, NCB) and haplogroup (HG^T; other haplogroups, HG^O; CM-067, n = 21 patients). P value from one-tailed Wilcoxon rank-sum test for the following comparisons: cluster 1, HG^T versus CB, P = 0.038; cluster 2, HG^T versus CB, P = 0.021, HG^T versus HG^O, P = 0.045; cluster 10, HG^O versus HG^T, P = 0.011, HG^O versus CB, P = 0.011. d, Scatterplot showing per-cluster mean expression of T cell exhaustion markers grouped by cell type. The connecting lines are used for the visualization of the changes in the gene expression of the selected markers per cluster. The clusters are ordered from least differentiated to exhausted phenotypes, based on the levels of differentiation markers as derived from b. e, Box plots showing single-cell CD8⁺ T cell-type proportions of baseline (postsurgical resection, pre-ICI) patient samples from CM-915 stratified by haplogroup (T haplogroup versus other haplogroups; CM-915, n = 31 patients). P value from one-tailed Wilcoxon rank-sum test, cluster 1, HG-T versus other MT-HGs, P = 0.020. Box plots depict lower and upper hinges (25th to 75th percentiles), median value (central line), whiskers extending from the hinges (largest and smallest values no more than 1.5 times the interquartile range) and outlying points outside this range (solid points beyond the whiskers). Source data

Fig. 5. Baseline differences in gene expression by haplogroup comparing bulk RNA-seq from peripheral blood CD8⁺ T cells in CM-067.

a, Top five significantly enriched pathways from GO analysis on top DEGs from CM-067 NIVO NCB cohort (n = 62 patients) comparing HG-T (n = 9) versus other MT-HGs (n = 53). −log₁₀-transformed P values derived from two-tailed Fisher’s exact test. b,c, Top five significantly enriched pathways from GO analysis (Fig. 5b) and COMPARTMENTS database (Fig. 5c) on top DEGs from CM-067 NIVO and COMBO cohorts (n = 212 patients) comparing HG-T (n = 17) versus other MT-HGs (n = 195). −log₁₀-transformed P values derived from two-tailed Fisher’s exact test. d, Volcano plot from differential expression analysis from NIVO and COMBO CM-067 analysis (n = 212) showing upregulation of DEGs in the SOD pathway in HG-T. Genes with negative log₂ fold change (log₂FC) are downregulated in HG-T; genes with positive log₂FC are upregulated in HG-T. Red points are the significant DEGs in both NIVO NCB and NIVO and COMBO analyses that are present in the SOD pathway. Yellow and blue points are DEGs that are significantly upregulated and downregulated, respectively, in HG-T in both NIVO NCB and NIVO and COMBO analyses. Light gray points are significant DEGs in the NIVO and COMBO analysis that did not overlap with DEGs in the NIVO NCB comparison. Dark gray points are genes that were not significantly differentially expressed and did not overlap between the NIVO NCB comparison and NIVO and COMBO comparison. −log₁₀-transformed P values derived from negative binomial Wald test in DESeq2. Source data

Fig. 6. PFS probability by haplogroup in CM-067.

a, PFS (in years) comparing HG-T (n = 39) and other MT-HGs (n = 339) in CM-067 NIVO and COMBO cohorts (n = 378). P value from two-tailed Kaplan–Meier (log-rank) test. b, PFS (in years) stratified by treatment, comparing HG-T treated by NIVO (n = 24) and COMBO (n = 15) and other MT-HGs treated by NIVO (n = 173) and COMBO (n = 166). P value from two-tailed Kaplan–Meier (log-rank) test. Source data

Extended Data Fig. 1. Percentage of subjects in each European haplogroup.

Haplogroup distributions comparing subjects in BMS-067 Nivo Discovery Cohort (left; n=115) and subjects with European ancestry from the 1000 Genomes Project (right; n=414). Most frequent European haplogroups (H, U, T, J, and K) are plotted individually, with less common haplogroups (W, R, V, X, I) grouped together as “Other”.

Extended Data Fig. 2. Response to ICI by haplogroup in CM-067 and IO-GEM cohorts.

Proportion of CB and NCB ICI treatment responses in NIVO, COMBO (IPI-NIVO), and IPI cohorts comparing HG-T and other (H, U, J, K, W, R, V, X, and I) MT-HGs in: (a) CM-067 NIVO cohort (n=197), two-tailed Chi-square p-value = 5.75E-03; (b) IO-GEM NIVO cohort (n=174), two-tailed Chi-square p-value = 0.04; (c) CM-067 COMBO cohort (n=181), two-tailed Chi-square p-value = 3.55E-03; (d) IO-GEM COMBO cohort (n=196), two-tailed Chi-square p-value = 0.12; (e) CM-067 IPI cohort (n=98), two-tailed Chi-square p-value = 0.39; (f) IO-GEM IPI cohort (n=379), two-tailed Chi-square p-value = 0.03; Total CheckMate-067, n=476; Total IO-GEM, n=749.

Extended Data Fig. 3. Association between HG-T and tumor markers in CM-067.

a. Results of correlation analysis comparing HG-T to demographic and tumor characteristics (n=255): age, sex, log-transformed tumor mutation burden (TMB), tumor PD-L1 status, and tumor “immune” score (a composite variable for either positive IFN-γ gene expression, increased CD8⁺ tumor infiltration, or both). b-d. Distribution of tumor markers by haplogroup: b. IFN-γ score (n=166); c. log-transformed tumor mutation burden (TMB) (n=376); d. % Tumor Infiltration by CD8⁺ T cells (n=182). Boxplots depict lower and upper hinges (25th to 75th percentiles), median value (central line), whiskers extending from the hinges (largest and smallest values no more than 1.5 times the interquartile range), and outlying points outside this range (solid points beyond the whiskers). P-values derived from two-tailed Kruskal-Wallis tests.

Extended Data Fig. 4. Cluster annotation and cell-type determination from CM-067 and CM-915 scRNA-seq.

a. Top 8 significantly differentially expressed genes per cluster from single-cell (sc)RNA-seq of peripheral blood CD8⁺ T cells isolated from NIVO-treated CM-067 patients (n=21). b. Per-cluster cell-type classification based on differential expression of marker genes from scRNA-seq of peripheral blood CD8⁺ T cells isolated from NIVO-treated CM-067 patients (n=21). c. Per-cluster gene expression of T cell exhaustion markers from scRNA-seq of peripheral blood CD8⁺ T cells isolated from NIVO-treated CM-067 patients (n=21). d. Per-cluster gene expression of T cell exhaustion markers in CD8⁺ T cells from scRNA-seq of pre-treatment PBMCs from CM-915 patients (n=31). As this was an independent analysis, clusters are labeled by the cluster identified in CM-915 (top cluster number) and, based on their expression profiles, the corresponding cluster first identified in CM-067 (bottom cluster number).

Extended Data Fig. 5. Differentially expressed genes from CM-067 bulk RNA-seq analysis of circulating CD8⁺ T cells comparing HG-T and other MT-HGs.

Heatmaps showing upregulated (red) and downregulated (green) genes for HG-T (pink) or other MT-HGs (blue) among a. NIVO NCB patients (n=62) and b. cohort of NIVO and COMBO CM-067 patients with bulk RNA-sequencing regardless of outcome (n=212).

Extended Data Fig. 6. Progression-free and overall survival stratified by MT-HG in CM-067 NIVO cohort (n=197) and COMBO cohort (n=181), and IMCG prognostic cohort.

a-b. Progression-free survival (PFS) stratified by MT-HG in CM-067 a. NIVO and b. COMBO cohorts. c-d. Overall survival (OS) stratified by MT-HG in CM-067 c. NIVO and d. COMBO cohorts. e-f. OS stratified by MT-HG (HG-T vs Other MT-HGs) and response in CM-067 e. NIVO and f. COMBO cohorts. Light bands represent 95% confidence intervals. g-h. OS by MT-HG in IMCG prognostic cohort for g. early-stage (Stage I-II) disease (n=915) and h. late-stage (Stage III) disease (n=109). All p-values from two-tailed Kaplan-Meier (log-rank) test.

Extended Data Fig. 7. Summary of circulating pre-treatment CD8⁺ T cell phenotypes defined using single-cell (sc)RNA-seq on patients from BMS-067 NIVO (n=21).

1. Clinical Benefit (CB) “response” phenotype: predominantly T cells able to be reinvigorated by anti-PD-1. 2. No Clinical Benefit in haplogroups other than HG-T (NCB-O): predominantly dysfunctional or terminally exhausted effector cells unable to be reinvigorated by anti-PD-1. 3. No Clinical Benefit in HG-T (NCB-T): few cytotoxic or exhausted effector cells, anti-PD-1 reinvigoration is insufficient. Created with BioRender.com.

Extended Data Fig. 8. Study population flow diagram.

Flow diagram illustrating the patient population, sample collection, multi-omic material extraction, sequencing/genotyping, and haplogroup determination for the analytic cohorts. Created with BioRender.com.

Extended Data Fig. 9. Genomics data analysis workflow.

Workflow for computationally determining mitochondrial haplogroups.