Artificial intelligence-based spatial analysis of tertiary lymphoid structures and clinical significance for endometrial cancer

Abstract

With the incorporation of immune checkpoint inhibitors into the treatment of endometrial cancer (EC), a deeper understanding of the tumor immune microenvironment is critical. Tertiary lymphoid structures (TLSs) are considered favorable prognostic factors for EC, but the significance of their spatial distribution remains unclear. B cell receptor repertoire analysis performed using six TLS samples located at various distances from the tumor showed that TLSs in distal areas had more shared B cell clones with tumor-infiltrating lymphocytes. To comprehensively investigate the distribution of TLSs, we developed an artificial intelligence model to detect TLSs and determine their spatial locations in whole-slide images. Our model effectively quantified TLSs, and TLSs were detected in 69% of the patients with EC. We identified them as proximal or distal to the tumor margin and demonstrated that patients with distal TLSs (dTLSs) had significantly prolonged overall survival and progression-free survival (PFS) across multiple cohorts [hazard ratio (HR), 0.56; 95% confidence interval (CI), 0.36-0.88; p = 0.01 for overall survival; HR, 0.58; 95% CI, 0.40-0.84; p = 0.004 for PFS]. When analyzed by molecular subtype, patients with dTLSs in the copy-number-high EC subtype had significantly longer PFS (HR, 0.51; 95% CI, 0.29-0.91; p = 0.02). Moreover, patients with dTLSs had a higher response rate to immune checkpoint inhibitors (87.5 vs. 41.7%) and a trend toward improved PFS. Our findings indicate that the functions and prognostic implications of TLSs may vary with their locations, and dTLSs may serve as prognostic factors and predictors of treatment efficacy. This may facilitate personalized therapy for patients with EC.