Real-world evaluation of CXCL9/10 with PD-L1 and TMB as predictors of pembrolizumab benefit in triple-negative breast cancer (TNBC).

1125 Background: Pembrolizumab is approved for TNBC; however, current biomarkers PD-L1 and tumor mutational burden (TMB) have limited predictive accuracy and show weak correlation with clinical benefit. We evaluated whether chemokines (CXCL9 and CXCL10) that recruit cytotoxic T cells into the tumor microenvironment (TME) may improve prediction of pembrolizumab benefit beyond PD-L1 and TMB. Methods: TNBC tumors (n=3662) were profiled by next-generation sequencing (592 NextSeq; WES/WTS NovaSeq; Caris Life Sciences, Phoenix, AZ). Chemokine expression was classified as high vs low based on 50 th percentile. Immune cells were estimated using WTS deconvolution (Quantiseq). PD-L1 was assessed by immunohistochemistry (22C3 ≥10%), and TMB by nonsynonymous mutations/Mb (high ≥10). Real-world median overall survival (mOS) was derived from insurance claims and calculated from pembrolizumab initiation to last contact. Associations were tested using chi-square and Mann-Whitney U with multiple comparison adjustment ( q<0.05 ). Results: CXCL9/ 10- high expression showed increased fraction of CD8⁺ T, dendritic cells and higher IFN-γ signatures ( CXCL9-high vs low : 1.3% vs 0%, 3.1% vs 2.8%, –0.09 vs –0.47; CXCL10 -high vs low: 1.1% vs 0%, 3.1% vs 2.7%, –0.1 vs –0.46), respectively all q<0.05 . Among patients treated with pembrolizumab (n=750), CXCL9 / 10- high expression was associated with improved mOS ( CXCL9-high : 24.5 months (m) vs 15.8 m; CXCL10-high : 23.4 m vs 17 m; all p<0.05 ) compared to CXCL9/10 -low; PD-L1+ tumors had improved mOS (24.3 m vs 18.6 m, p=0.007 ) compared with PD-L1-negative; and TMB-low tumors had numerically improved mOS (21.7 m vs 17.5, p=0.08 ) compared with TMB-high. Combining CXCL9 / 10 with PD-L1 or TMB identified subgroups of tumors with PD-L1-negative and TMB-high expression that showed better OS with pembrolizumab (Table). In multivariable cox analysis, adjusting for TMB and PD-L1, CXCL9 HR 0.67 (95% CI 0.54-0.83) and CXCL10 HR 0.79 (95% CI 0.64-0.97) remained independent predictors of pembrolizumab benefit, all p<0.05 . Conclusions: In this large real-world study, PD-L1 but not TMB predicted pembrolizumab benefit in TNBC. CXCL9 / 10-high define an immune-active TME and are independent predictors of pembrolizumab response. When integrated with PD-L1 and TMB, these chemokines help identify additional patients with TNBC who may derive benefit from pembrolizumab, supporting their inclusion as biomarkers in prospective studies. PD-L1 Chemokine CXCL9 mOS (in months) (95% CI) p CXCL10 mOS (in months) (95% CI) p + High 25.7 (22.1 - 32.1) <0.01 25.7 (22.2 - 34.1) <0.01 – High 26.9 (17.8 - 39.2) 21.7 (17.5 - 28.2) + Low 18.5 (11.9 - NR) 20.7 (13.3 - 30.1) – Low 15.8 (12.2 - 19.2) 16.7 (13.1 - 21.5) TMB Low High 25.0 (22.1 - 30.1) <0.01 23.7 (20.9 - 28.2) <0.01 High High 20.6 (17.5 -37.1) 19.5 (16.3 - NR) Low <jats:td c