Molecular determinants of response to PD-L1 blockade across tumor types

Abstract

Immune checkpoint inhibitors targeting the PD-1/PD-L1 axis lead to durable clinical responses in subsets of cancer patients across multiple indications, including non-small cell lung cancer (NSCLC), urothelial carcinoma (UC) and renal cell carcinoma (RCC). Herein, we complement PD-L1 immunohistochemistry (IHC) and tumor mutation burden (TMB) with RNA-seq in 366 patients to identify unifying and indication-specific molecular profiles that can predict response to checkpoint blockade across these tumor types. Multiple machine learning approaches failed to identify a baseline transcriptional signature highly predictive of response across these indications. Signatures described previously for immune checkpoint inhibitors also failed to validate. At the pathway level, significant heterogeneity is observed between indications, in particular within the PD-L1⁺ tumors. mUC and NSCLC are molecularly aligned, with cell cycle and DNA damage repair genes associated with response in PD-L1- tumors. At the gene level, the CDK4/6 inhibitor CDKN2A is identified as a significant transcriptional correlate of response, highlighting the association of non-immune pathways to the outcome of checkpoint blockade. This cross-indication analysis reveals molecular heterogeneity between mUC, NSCLC and RCC tumors, suggesting that indication-specific molecular approaches should be prioritized to formulate treatment strategies.

Fig. 1. PD-L1, TMB, and global RNA-seq profiles in mUC, NSCLC, and RCC cohorts.

a Bar chart representing the distribution of PD-L1 expression on TC and IC by indication and response group. PD-L1 distribution between response groups within and across indications was statistically tested using the two-sided Pearson’s chi-squared test. b Boxplot representing TMB by indication and response group. TMB differences within each indication were tested using the non-corrected two-sided Wilcoxon rank-sum test. The center of the boxplot represents the median. The lower and upper hinges correspond to the first and third quartiles. The upper whisker extends from the hinge to the largest value no further than 1.5 × IQR (interquartile range) from the hinge. The lower whisker extends from the hinge to the smallest value at most 1.5 × IQR of the hinge. c Venn diagram representing the overlap between responders, PD-L1⁺ patients and TMB^high patients. d Two-dimensional scatter plot representing sample distribution according to the first two components obtained from principal component analysis (PCA) of the complete RNA-seq-evaluable dataset (n = 366) based on the 16,581 genes used for analysis. Dots are colored by indication and ellipses capture all samples within one standard deviation of the mean per normal probability statistics (68%). e Same as d, colored by response group.

Fig. 2. Machine learning to identify a transcriptional signature of response to PD-L1 blockade.

a Flowchart depicting the approach to identify the signature. b Left panel, bar chart representing the signature score by indication and response group within the RNA-seq and TMB-evaluable population in the atezolizumab arms of the training set. n = 144 mUC, n = 50 NSCLC, and n = 52 RCC biologically independent samples were examined. Right panel, bar chart representing the signature score in the control arms of POPLAR (NSCLC, docetaxel arm) and IMmotion150 (RCC, sunitinib arm) clinical trials. n = 75 NSCLC and n = 85 RCC biologically independent samples were examined. The center of the boxplots represents the median. The lower and upper hinges correspond to the first and third quartiles. The upper whisker extends from the hinge to the largest value no further than 1.5 × IQR (interquartile range) from the hinge. The lower whisker extends from the hinge to the smallest value at most 1.5 × IQR of the hinge. P values were calculated using the non-corrected two-sided Wilcoxon rank-sum test. c ROC curves for the 58-gene signature, PD-L1, TMB, and external signatures in the RNA-seq/TMB-evaluable population. AUC values are displayed in parenthesis. d Same as c for the independent phase I PCD4989g test set.

Fig. 3. Transcriptional correlates of response to PD-L1 blockade across indications.

a In all tumors combined. Upper panel, Forest plot representing the pathway activity of WGCNA modules significantly enriched by Q-Gen analysis between responders and nonresponders across the combined cohorts or on an individual cohort basis. The pathway-level model contrasts responders and nonresponders, including indication as a covariate in the cross-indication analysis. The 20 modules detected as significant (p < 0.05) in any of the four models conducted are included in the display. Modules are ordered from left to right according to pathway activity across indications. Error bars represent the 95% confidence interval. Lower panel, heatmap representing per-indication pathway activity for modules selected from the combined cross-indication and per-indication analyses. Module enrichment significance is highlighted as a red dot (FDR-corrected p < 0.05) or a black dot (nominal p < 0.05). n = 208 mUC, n = 81 NSCLC, and n = 77 RCC biologically independent samples were examined. b Same as a for PD-L1⁺ (upper heatmap) and PD-L1⁻ (lower heatmap) tumors separately.

Fig. 4. CDK4/6 inhibition associates with increased response to PD-L1 blockade.

a Volcano plot representing the genes differentially expressed between responders and nonresponders. The gene-level linear model contrasts responders (CR/PR) and nonresponders (SD/PD), including indication and PD-L1 expression as covariates. Genes significantly upregulated or downregulated after Benjamini–Hochberg correction (p < 0.1) and absolute log2 fold change ≥  0.5 are colored in red and blue, respectively. b Horizontal bar chart representing the percent of patients exhibiting partial (CN1) or complete copy-number deletion (CN0) of the CDKN2A locus across the Foundation Medicine database (n = 97,811 after QC) for selected indications. The ratio and percentage of patients with CDKN2A loss within each ontology is represented on the right of each bar. c Response rate by CDKN2A deletion status. Bar charts represent the proportion of responders to nonresponders by CDKN2A deletion status (no DEL: copy-number ≥ 2; DEL: copy-number < 2). P values were calculated using the two-sided Pearson’s chi-squared test. d Overall survival (OS) for mUC and NSCLC cohorts and progression-free survival (PFS) for the RCC cohort, split by transcriptional expression of CDKN2A (top) or CDK6 (bottom). Transcription level is defined as high (≥median, red) or low (<median, blue) within each indication.