An integrative analysis of circulating and tumor microenvironment (TME) determinants of patient response in the Checkmate 9ER (CM 9ER) trial of nivolumab and cabozantinib (NIVO+CABO) in advanced renal cell carcinoma (aRCC).

Abstract

4511 Background: The CM 9ER trial demonstrated increased objective response rate (ORR), progression-free and overall survival in patients with aRCC treated with NIVO+CABO compared to sunitinib (SUN). For vascular modulating therapies (e.g. CABO and SUN) and immunotherapies (like NIVO), the state of the TME and activity of stromal cells can modulate tumor response to therapy. Methods: We investigated how the TME and circulating factors were associated with response to NIVO+CABO using pre-treatment tumor PD-L1 staining, human interpretable features (HIF) derived from H&E tissue sections, circulating immune cell populations quantified by flow cytometry, and circulating extracellular matrix (ECM) markers quantified by competitive ELISAs from 150 patients (23% of ITT) enrolled in CM 9ER. We employed principal component analysis, varimax rotation, and Feature Set Enrichment Analysis (FSEA) to identify a subset of biological measurements capturing 85% of the data variability. We constructed a logistic regression model to associate the most variable features with patient response (ORR per BICR) to NIVO+CABO or SUN therapy. Results: An unbiased clustering and feature extraction approach was used to identify measurements contributing to multi-modal variability in the TME across 150 patients (~4000 biological measurements reduced to 16 highly informative measurements): PD-L1 staining, 4 ECM markers, 4 PBMC markers, and 7 H&E HIF features. A final binary logistic regression model was built employing lasso regularization to associate these 16 features to short-term response to NIVO+CABO or SUN while minimizing spurious associations. Based on these regression models, the ECM marker VICM, a citrullinated fragment of vimentin released by matrix metalloproteases, and which measures macrophage activity and immune status, was prognostic across both therapies. Logistic regression models that integrated highly informatic features had AUCs of 0.76 for NIVO+CABO model and 0.72 for the SUN model. Within the NIVO+CABO arm, this integrative model uncovered high VICM (and therefore anti-tumor macrophage polarization), high PD-L1 staining, high plasma cell numbers and high cancer cell numbers at the epithelial stromal interface, low levels of circulating fragment of C-terminal type VIa3 collagen (Pro-C6), and low percentages of circulating regulatory (Foxp3+ CD4+) T cells as determinants of therapeutic response. Conclusions: Taken together, these findings indicate that the state of the tumor microenvironment and circulating factors together have an effect on patient responsiveness to NIVO+CABO in aRCC and provides a framework for integrative analysis for biomarker discovery.