32Interplay of Tumor-Intrinsic and microenvironmental gene expression predicts immune checkpoint blockade response in metastatic renal cell carcinoma

Abstract Background We previously defined a Genomic Dedifferentiation Signature (GDS) derived from differential gene expression analysis of sarcomatoid renal cell carcinoma (sRCC) tumor clones compared to clear cell RCC (ccRCC; Salgia et al. KCRS. 2024). GDS expression is negatively prognostic across clinical RCC cohorts, yet carries predictive capacity for identifying patients likely to achieve benefit from immune checkpoint blockade (ICB). Furthermore, we have also demonstrated that sRCCs are enriched for expression of tertiary lymphoid structure (TLS) signatures. Crucially, TLS signatures have been associated with response to ICB in ccRCC (Jammihal et al. Nature Cancer. 2025). Therefore, we sought to investigate if an interplay between expression of the GDS—a tumor-intrinsic gene signature—and TLS signatures—characterizing the inflammatory immune microenvironment—could best identify ICB sensitivity in patients with metastatic RCC. Methods Transcript-per-million normalized gene expression counts matrices and clinical metadata were collected from three trials of ICB regimens in patients with metastatic RCC: IMmotion 151 (Motzer et al. Cancer Cell. 2020), Javelin101 (Motzer et al. Nature Medicine. 2020), and HCRN-GU16-260 (Hugaboom et al. Cancer Discovery. 2025). Gene expression scores for the GDS and the Imprint TLS signature (Meylan et al. Immunity. 2022) were assigned to cases using single sample gene set enrichment analysis. Correlation between GDS and TLS scores was performed via Spearman’s rank-order correlation. Cases were stratified into -high and -low signature expression (TLShigh-GDShigh, TLShigh-GDSlow, TLSlow-GDShigh, and TLSlow-GDSlow) based on median signature expression within trial cohorts. Progression-free survival (PFS) was compared between groups and trial arms within each cohort using Cox proportional hazards models, while objective response rates (ORR) were compared using Fisher’s exact tests. Results The correlation between GDS & Imprint TLS expression was moderate across IMmotion151 (n = 822; R = 0.44; p < 2.2x10-16), Javelin101 (n = 723; R = 0.43; p < 2.2x10-16) and HCRN-GU16-260 (n = 72; R = 0.53; p = 2.6x10-6). Within IMmotion151, patients classified as TLShigh-GDShigh experienced prolonged PFS with the ICB-containing atezolizumab/bevacizumab regimen (median PFS: 10.9 months) compared to sunitinib (median PFS: 6.9 months; Hazard Ratio [HR]=0.68; P = .009; Figure 1), whereas no significant differences between atezolizumab/bevacizumab and sunitinib PFS were observed in TLShigh-GDSlow (HR = 1.18; P = .5), TLSlow-GDShigh (HR = 0.71; P = .09), or TLSlow-GDSlow (HR = 0.93; P = .6) subgroups. ORR to atezolizumab/bevacizumab within the TLShigh-GDShigh population (47.5%) was also significantly higher compared to atezolizumab/bevacizumab ORR within the TLSlow-GDSlow population (30.6%; P = .008) and compared to ORR with sunitinib in the TLShigh-GDShigh population (31.0%; P = .009). Conversely, ORR with sunitinib (43.6%) was significantly greater than with atezolizumab/bevacizumab (30.6%; P = .004) in the GDSlow-TLSlow population. Consistent results were observed in Javelin101: the TLShigh-GDShigh population experienced prolonged PFS with the ICB-containing axitinib/avelumab regimen (median PFS: 12.5 months) compared to sunitinib (median PFS: 6.9 months; HR = 0.47; P < .001). Again, no significant differences between axitinib/avelumab and sunitinib PFS were observed in TLShigh-GDSlow (HR = 1.04; P = .9), TLSlow-GDShigh (HR = 0.81; P = .4), or TLSlow-GDSlow (HR = 0.81; P = .3) subgroups. In the HCRN-GU16-260 trial of front-line nivolumab in metastatic RCC, TLShigh-GDShigh status was associated with prolonged PFS, albeit non-significantly, compared to TLShigh-GDSlow (HR = 2.02; P = .10), TLSlow-GDShigh (HR = 1.99; P = .13) and TLSlow-GDSlow (HR = 1.72; P = .09). Additionally, ORR to nivolumab was significantly enriched in the TLShigh-GDShigh cohort (76.9%) compared to those classified as TLShigh-GDSlow (23.1%; P = .0009), TLSlow-GDShigh (25.0%; P = .01), and TLSlow-GDSlow (18.5%; P < .0001) within HCRN-GU16-260. Conclusions Across metastatic RCC patient cohorts, simultaneous enrichment of tumor-intrinsic and TLS gene expression best predict response to ICB versus anti-angiogenic therapy. This suggests that a pro-inflammatory milieu is not sufficient for effective immune sensitivity, but that a biologic interplay between tumor cells and their surrounding microenvironment is necessary for effective immune surveillance and ICB responsiveness in RCC.