Dissecting the baseline tumor microenvironment in mNSCLC by transcriptional profiling identifies tumor microenvironmental conditions predictive for immunotherapy response

Background: The MYSTIC trial (NCT02453282), a phase 3 study of durvalumab ± tremelimumab vs. standard of care (SOC) chemotherapy in 1L metastatic NSCLC, did not meet its primary endpoint of overall survival (OS) in the intent-to-treat (ITT) population. This analysis aimed to identify whether baseline tumor-immune microenvironment (TME) features were predictive for either favorable or poor response to durvalumab (D) ± tremelimumab (T) treatment in this patient population. Methods: RNA sequencing was performed on baseline tumor samples from 185 patients in the MYSTIC biomarker-evaluable population (BEP). Non-negative matrix factorization (NMF) was used to resolve distinct clusters based on the expression of the top 10% most variable protein-coding genes. The resulting clusters were analyzed for differences in clinical outcomes, genomic alterations, immune cell composition, pathway activity, and T-cell and B-cell repertoire. Results: NMF identified four distinct clusters within the MYSTIC BEP population of which two clusters were notably characterized by tumor immune grouping. Cluster 1 (20% of patients) exhibited hallmarks of a pro-inflammatory TME, with high T-cell signatures, macrophage infiltration, and expression of checkpoint inhibitors. Patients in this cluster treated with D+T experienced a prolonged survival (mOS 19.78 months) compared with patients on this treatment in any of the other three clusters (HR 0.39, 95%CI 0.15-1). Cluster 1 was also enriched for patients with PD-L1≥50%. Cluster 3 (20% of patients) displayed features associated with immune suppression, including high granulocyte and Treg infiltration, low macrophage signature, and enrichment of TGFbeta and stemness pathways. These patients experienced worsened survival outcomes when treated with D+T (mOS 2.17 months), compared to other clusters (HR 3.13, 95%CI 1.48-6.6). Most notably, patients in cluster 3 had significantly worsened survival on D+T compared to patients in cluster 1 (HR 6.15, 95%CI 2.02-18.8). The remaining two clusters could be characterized by histology, with Cluster 2 (35% of patients) characterized by non-squamous histology and enrichment for STK11 and MYC alterations, and Cluster 4 (25% of patients) characterized by squamous histology, aggressive proliferation signatures, and enrichment for TP53, MLL2, PTEN, and PIK3CA alterations. Neither of these two clusters drove notable impact to D+T treatment. Conclusions: NMF clustering of baseline MYSTIC tumor samples revealed distinct TME profiles linked to divergent immunotherapy responses. A pro‑inflammatory cluster predicted benefit with D+T, while a suppressive cluster aligned with poor outcomes. Findings suggest response depends not just on overall immune infiltration but on the specific immune cell composition within the TME.