Immune features are associated with response to neoadjuvant chemo-immunotherapy for muscle-invasive bladder cancer

Abstract

Neoadjuvant cisplatin-based chemotherapy is standard of care for muscle-invasive bladder cancer (MIBC). Immune checkpoint inhibition (ICI) alone, and ICI in combination with chemotherapy, have demonstrated promising pathologic response (<pT2) in the neoadjuvant setting. In LCCC1520 (NCT02690558), a phase 2 single-arm trial of neoadjuvant chemo-immunotherapy (gemcitabine and cisplatin plus pembrolizumab; NAC-ICI) for MIBC, 22/39 patients responded (pathologic downstaging as primary outcome), as previously described. Here, we report post-hoc correlative analyses. Treatment was associated with changes in tumor mutational profile, immune gene signatures, and RNA subtype switching. Clinical response was associated with an increase in plasma IL-9 from pre-treatment to initiation of cycle 2 of therapy. Tumors harbored diverse predicted antigen landscapes that change across treatment and are associated with APOBEC, tobacco, and other etiologies. Higher pre-treatment tumor PD-L1 and TIGIT RNA expression were associated with complete response. IL-8 signature and Stroma-rich subtype were associated with improved response to NAC-ICI versus neoadjuvant ICI (ABACUS trial, NCT02662309). Plasma IL-9 represents a potential predictive biomarker of NAC-ICI response, while tumor IL-8 signature and stroma-rich subtype represent potential predictive biomarkers of response benefit of NAC-ICI over neoadjuvant ICI. Future efforts must include additional independent biomarker discovery and validation, ultimately to improve the selection of patients for ICI-related treatments.

Fig. 1. LCCC1520 clinical data.

A Overview of LCCC1520 clinical trial timeline and data collection timepoints. B Clinical T stage and (C) baseline ECOG performance status by response (n = 39, Fisher’s exact tests, no FDR correction). D Response by occurrence of an adverse treatment-related event that required cessation of chemotherapy (n = 39), Fisher test, no FDR correction.

Fig. 2. MIBC tumors change with NAC-ICI.

A TMB in pre-treatment and post-treatment tumors from non-pCR patients (paired two-sided Wilcoxon test; n = 15). B Oncoplots of the 10 most common mutations in pre-and post-treatment tumors compared between timepoints (n = 51 samples). C Mutations in non-pCR tumors compared between pre-and post-treatment (n = 28 samples). D Mutations gained, lost, or retained from pre-to post-treatment in non-pCR tumors (n = 28 samples). E Pre-treatment molecular subtype by patient according to 5 subtyping schemes (n = 37). F Change in Consensus subtype among non-pCR patients from pre-treatment to post-treatment (n = 37). G Change in MDA subtype among non-pCR patients from pre-treatment to post-treatment (n = 37). H Changes in tumor RNA-based immune gene signatures from pre-to post-treatment among non-pCR patients (two-sided Wilcoxon test; n = 37). I Relative changes (differences divided by pre-treatment mean) in TCR and BCR diversity metrics among non-pCR patients (two-sided Wilcoxon test; n = 37). For all boxplots: centre = median; lower bound of box = 25^th percentile; upper limit of box = 75^th percentile; lower whisker = minimum value, 25^th percentile − 1.5*IQR; upper whisker = maximum value, 75^th percentile + 1.5*IQR.

Fig. 3. Immune populations and plasma analyte features in peripheral blood change with NAC-ICI.

A Changes in peripheral blood flow cytometry markers from pre-treatment to pre-cycle 2 (n = 28; two-sided paired Wilcoxon test). B Changes in plasma analyte concentrations from pre-treatment to post-treatment (n = 37; two-sided paired Wilcoxon test). C Changes in plasma analyte concentrations between pre-treatment and pre-cycle 2 (n = 32; two-sided paired Wilcoxon test). D Changes in plasma analyte concentrations between pre-treatment and pre-cycle 2 in responders (n = 11) versus non-responders (n = 21). E Changes in IL-9 concentrations from pre-treatment to pre-cycle 2 in responders (n = 11) and non-responders (n = 21, two-sided paired Wilcoxon tests, no FDR correction). F Receiver operator characteristic curve of change in IL-9 concentration from pre-treatment to pre-cycle 2 versus response (n = 32). G Logistic regression of change in IL-9 concentration versus response. Error bands span the 95% confidence interval (n = 32).

Fig. 4. MIBC tumors exhibit varied predicted antigen landscapes.

A Counts of predicted antigens (binding affinity < 500 nM) in pre-treatment tumors (n = 25). B Number of predicted self-antigens shared between pre-treatment tumors. C Counts of predicted antigens lost, retained, or gained from pre-to post-treatment in non-pCR tumors (n = 9). D Counts of predicted antigens in pre-treatment tumors filtered based on Wells criteria (n = 25). E Binding affinity and stability of Wells criteria-filtered predicted antigens (n = 25). F TPM expression levels of Wells criteria-filtered predicted antigens: ERV (n = 1 antigen), Self-antigen (n = 17 antigens), and SNV (n = 9 antigens). G COSMIC signatures of mutations in pre-and post-treatment tumors (n = 27 patients). (H) Counts of mutations lost, gained, and retained from pre-to post-treatment in non-pCR tumors, labeled by COSMIC signature (n = 9 patients). For all boxplots: centre = median; lower bound of box = 25th percentile; upper limit of box = 75th percentile; lower whisker = minimum value, 25th percentile − 1.5*IQR; upper whisker = maximum value, 75th percentile + 1.5*IQR. Unless otherwise noted, all pairwise comparisons are two-sided Wilcoxon tests.

Fig. 5. Clinical and immunogenomic features are associated with MIBC patient outcomes.

A TMB levels from pre-treatment tumors are compared by patient response status (n = 33 patients: 13 CR, 7 PR, 13 NR; two-sided Wilcoxon tests). B TMB levels by an adverse treatment-related event that required cessation of chemotherapy (n = 32 patients: 12 CR, 7 PR, 13 NR; two-sided Wilcoxon test). C–E RNA expression of PD-L1 and TIGIT, and Ayers IFNG immune gene signature values, from pre-treatment tumors are compared by patient response status (n = 37 patients: 14 CR, 8 PR, 15 NR; two-sided Wilcoxon tests). Groups are compared by Wilcoxon p-value. F Consensus subtype of pre-treatment tumors by response class (Fisher’s exact test, no FDR correction; n = 37 patients). G Kaplan-Meier survival curves of female and male patients by response status. The difference in survival between female and male responders is compared by log-rank p-value. H Cross-validated elastic net coefficients of response, with only the feature coefficients shown that have 95% confidence intervals from cross validation that do not span zero (n = 33 patients). Points represent feature beta coefficients in each of 10 folds of cross validation. Bars represent mean beta coefficient value for each feature across 10-fold cross validation. Gray error bars represent 95% confidence intervals for each feature. I Receiver operating characteristic curve of model predictions on the training set. J Cross-validated elastic net coefficients of survival, with only the feature coefficients shown that have 95% confidence intervals from cross validation that do not span zero (n = 33 patients). Points represent feature beta coefficients in each of 10 folds of cross validation. Bars represent mean beta coefficient value for each feature across 10-fold cross validation. Black error bars represent 95% confidence intervals for each feature. For all boxplots: centre = median; lower bound of box = 25th percentile; upper limit of box = 75th percentile; lower whisker = minimum value, 25th percentile − 1.5*IQR; upper whisker = maximum value, 75th percentile + 1.5*IQR. Unless otherwise noted, all pairwise comparisons are two-sided Wilcoxon tests.

Fig. 6. NAC-ICI could improve outcomes compared to neoadjuvant ICI in specific subsets of patients.

A Comparison of association of pre-treatment immune gene signatures with response in ABACUS (ICI only, n = 84) versus LCCC1520 (Chemo-ICI, n = 37). Each pre-treatment immune gene signature’s association with response is plotted as a rectangle spanning the 83.4% confidence interval of the coefficient. IL-8 signature and stroma-rich Consensus subtype are highlighted as the only signatures with significantly different associations with response in the two data sets, with confidence intervals that do not cross the y = x line. B Response status by IL-8 signature in the two data sets. Loess curves are plotted with error bands spanning the 95% confidence interval. C, D Response status by data set in patients with high or low pre-treatment IL-8 signature values (Fisher’s exact test, no FDR correction). E, F Response status by data set in patients with Stroma-rich or other Consensus subtype (Fisher’s exact test, no FDR correction).