A comprehensive evaluation of gene expression-based signatures for detecting tertiary lymphoid structures (TLS) in metastatic renal cell carcinoma (mRCC).

591 Background: TLS are organized lymphoid aggregates linked to improved response to immunotherapy in multiple cancers, including mRCC. Traditionally, TLS assessment relies on pathological staining of tumor sections. In this study, we evaluated the performance of TLS gene-expression signatures as an alternative diagnostic tool. Methods: Primary tumors from treatment-naïve patients with mRCC in the HCRN GU16-260 trial were analyzed using multiparametric immunofluorescence (IF), staining for CD3, CD20 (TLS), CD21 (mature TLS), and CD8. The counts and densities of total and mature TLS were calculated. Bulk RNA sequencing was performed on the same specimens, and 4 gene signature scores (Meylan, Cabrita, Coppola, and Ng) were estimated using gene set variation analysis (GSVA). Spearman's correlations between signature scores and TLS IF parameters, as well as the area under the receiver operator characteristic (AUROC) curve, were computed. Results: Our cohort included 78 patients with matched IF and RNA-seq data. All 4 TLS signatures demonstrated strong predictive performance for detecting the presence (vs. absence) of TLS by IF in each sample (AUROC 0.8 to 0.85). Quantitatively, the 4 TLS signature scores, measured as continuous variables, correlated with IF-based counts and densities of total TLS, and to a lesser extent, mature TLS (Table). In terms of comparative performance, Meylan’s signature showed the highest correlation with IF TLS features, while Ng’s was the least correlated (Table). Furthermore, Meylan’s signature achieved the best predictive ability to stratify high vs. low TLS (AUROC 0.83) and was more specifically correlated with TLS compared to CD8 density (Table). Conclusions: Expression-based signatures exhibit robust predictive ability for TLS detection and quantification, similar to previously established CD8 signatures, and represent valid surrogate tools to complement pathological diagnosis. Spearman’s correlations between IF parameters and TLS signatures. Parameter Meylan Cabrita Coppola Ng Total TLS count ρ=0.65, P= 1.1x10 -10 ρ=0.58, P= 2.9x10 -8 ρ=0.56, P= 7.1x10 -8 ρ=0.42, P= 1.4x10 -4 Mature TLS count ρ=0.51, P= 2.1x10 -6 ρ=0.41, P= 1.8x10 -4 ρ=0.34, P= 2.3x10 -3 ρ=0.39, P= 5.0x10 -4 Total TLS density ρ=0.63, P= 6.1x10 -10 ρ=0.58, P= 3.4x10 -8 ρ=0.55, P= 2.2x10 -7 ρ=0.42, P= 1.4x10 -4 Mature TLS density ρ=0.49, P= 5.1x10 -6 ρ=0.39, P= 3.5x10 -3 ρ=0.32, P= 4.3x10 -3 ρ=0.38, P= 6.8x10 -4 CD8 density ρ=0.42, P= 3.1x10 -4 ρ=0.62, P= 2.1x10 -8 ρ=0.67, P <2.2x10 -16 ρ=0.35, P= 2.9x10 -3