Recurrent somatic mutations as predictors of immunotherapy response

Abstract

Immune checkpoint blockade (ICB) has transformed the treatment of metastatic cancer but is hindered by variable response rates. A key unmet need is the identification of biomarkers that predict treatment response. To address this, we analyzed six whole exome sequencing cohorts with matched disease outcomes to identify genes and pathways predictive of ICB response. To increase detection power, we focus on genes and pathways that are significantly mutated following correction for epigenetic, replication timing, and sequence-based covariates. Using this technique, we identify several genes (BCLAF1, KRAS, BRAF, and TP53) and pathways (MAPK signaling, p53 associated, and immunomodulatory) as predictors of ICB response and develop the Cancer Immunotherapy Response CLassifiEr (CIRCLE). Compared to tumor mutational burden alone, CIRCLE led to superior prediction of ICB response with a 10.5% increase in sensitivity and a 11% increase in specificity. We envision that CIRCLE and more broadly the analysis of recurrently mutated cancer genes will pave the way for better prognostic tools for cancer immunotherapy.

Fig. 1. An aggregated cohort of immune checkpoint blockade (ICB) patients replicates known correlations between tumor mutational burden and age with treatment response.

a Overview of the two-stage approach for immunotherapy response prediction. We pooled 6 cohorts of immune checkpoint blockade (ICB) recipients with matched whole-exome sequencing (WES) and Response Evaluation Criteria in Solid Tumors (RECIST) classification. We identified genes and pathways under positive selection and tested the nominated genes and pathways for their ability to predict ICB response. The significant predictors were used to develop and test an ICB response prediction algorithm. b Number of patients from the aggregated set of 6 cohorts in each RECIST response group. Patients with stable disease were excluded from analyses and the RECIST classifications of complete response and partial response were both considered responders. c Proportion of tumor types amongst ICB responders and non-responders. d Enrichment (effect size, Hedge’s g) for different types of mutations in responders (n = 94) and non-responders (n = 178) to ICB therapy. Error bars represent the 95% confidence interval and significance was determined using a two-sided Welch’s t test with Bonferroni correction. Tumor Mutational Burden (TMB) is the union of High and Moderate mutations. e TMB for responders (n = 94) and non-responders (n = 178) to ICB therapy by tumor type. Statistical significance was tested using two-tailed Welch’s t tests of log₂ TMB. f Patient ages for different RECIST response groups (complete response n = 14, partial response n = 80, progressive disease n = 178). Statistical significance was tested using a two-tailed Welch’s t test. In e and f, the boxplot center line denotes median, with box limits being the 25th and 75th percentile. Boxplot whiskers indicate 1.5 times the interquartile range, while outliers above/below the whiskers are represented individually as points.

Fig. 2. A two-stage approach identifies BCLAF1 somatic genotype  as a predictor of ICB response.

a Quantile-quantile plot of fishHook p-values to assess significance of gene mutational burden after removing confounders. The p-values were obtained by comparing observed mutational rate to the right tail (one-sided) of the expected mutational rates derived from a gamma-Poisson model of genome-wide mutational density and the covariates replication timing, epigenetic state, and sequence context. In the first stage of CIRCLE, six significant genes were identified below a false-discovery threshold (FDR < 0.1). b Odds ratios (ORs) of response to ICB therapy in patients with a high or moderate impact mutation in the indicated gene as compared to patients that do not have a high or moderate mutation in the given gene (n = 272 patients). Error bars indicate the 95% confidence interval of the odds ratio. ORs greater than one indicate enrichment in responders and ORs less than one indicate enrichment in non-responders. Statistical significance was tested using a two-sided Wald’s test of coefficients with multiple-hypothesis correction (FDR < 0.2).

Fig. 3. BCLAF1 mutations identify a subset of non-responders with high tumor mutational burden (TMB).

a Age, TMB and tumor type for responders and non-responders with (red) and without (gray) BCLAF1 mutations. b TMB of patients with (n = 33) and without (n = 239) mutations in BCLAF1. Significance was calculated using a two-sided Welch’s t test and error bars indicate 95% confidence intervals. c Age of patients with (n = 33) and without (n = 239) mutations in BCLAF1. Significance was calculated using a two-sided Welch’s t test and error bars indicate 95% confidence intervals. d Protein location of mutations in BCLAF1 in responders (top) and non-responders (bottom). Mutations are color-coded by mutation type. Horizontal black lines indicate mutational clusters. The red horizontal line indicates a mutational cluster not present in responders. e Prevalence of BCLAF1 mutations in melanoma, bladder, and NSCLC cancer by ICB response status. f Distribution of BCLAF1 mutations by tumor type.

Fig. 4. Somatic mutations in genes encoding DNA damage, immune-associated, and mitogen-activated protein kinase (MAPK) pathways correlate with ICB response.

a Volcano plot of fishHook-nominated pathways with log₂ odds ratio for ICB response (x-axis) and significance of association with ICB response (y-axis). Statistical significance was tested using a two-sided Wald’s test of coefficients with multiple-hypothesis correction (FDR < 0.2). b fishHook-nominated pathways that overlap top-ranked genes from a genome-wide CRISPR screen for immunotherapy resistance (FDR-corrected one-sided hypergeometric test). c–d Volcano plots of odds ratio (responders/non-responders) and nominal p-values for genes in two of the fishHook-nominated pathways: Scavenging by Class A Receptors (c) and MAP2K and MAPK Activation (d). Red outlines indicate genes that were also found in the CRISPR screen. Indicated p-values are from the fishHook model using the observed mutation rate for each gene.

Fig. 5. The cancer immunotherapy response CLassifiEr (CIRCLE) predicts ICB response and patient survival.

a Averaged areas under the receiver-operator curve (AUCs) from 100 Monte Carlo cross validation iterations of the CIRCLE classifier and the FoundationOne CDx tumor mutational burden (FO-TMB) classifier. Error shading indicates the standard deviation of AUCs calculated from the 100 cross validation iterations. b Absolute values and percent change in the true positive rate (sensitivity), true negative rate (specificity), false positive rate and false negative rate of the CIRCLE classifier and FO-TMB classifier. c–d Waterfall plots of per patient scores from the CIRCLE (c) and FO-TMB (d) classifiers. Each patient is represented as a vertical bar and the adjusted score is equal to the score of the indicated classifier minus the optimal cutoff derived from the respective receiver operating characteristic curves. e Kaplan–Meier plot of overall survival for patients classified as CIRCLE responders versus CIRCLE non-responders. Shaded areas indicate the 95% confidence interval. Statistical significance was calculated using a two-sided Cox proportional hazards test with tumor type as a covariate.