AI-powered characterization of the tumor microenvironment landscape in HER2-overexpressing non–small cell lung cancer.

Abstract

8545 Background: Antibody-drug conjugate (ADC) targeting HER2 in NSCLC is an opportunity for biomarker-guided therapy independent of oncogenic driver status. As such, HER2 overexpressing (OE) tumors remain a clinically relevant yet under-characterized subgroup of NSCLC patients. Given the role of the tumor microenvironment (TME) in modulating response to ADCs, we applied AI-powered spatial analysis to comprehensively characterize the immunological landscape of NSCLC stratified by HER2 expression. Methods: We performed a retrospective analysis of a proprietary dataset comprising 2,054 H&E whole-slide images (WSIs) of NSCLC cases collected globally and enriched for HER2 OE tumors. HER2 OE was defined as IHC 3+ by pathologist reading based on ASCO/CAP guidelines and compared against IHC 0-2+. An AI-powered H&E analyzer (Lunit SCOPE IO) was used to quantify tumor-infiltrating lymphocytes (TILs) in the epithelial and stromal compartments. Immune phenotypes were classified as inflamed, excluded, or desert. For a subset of 1,099 cases with paired HER2 IHC WSIs, the proportion of tumor cells with 3+ intensity was quantified using an IHC analyzer (Lunit SCOPE HER2). Findings were further validated using 318 H&E WSIs and differential gene expression analysis of the TCGA Lung Adenocarcinoma (LUAD) cohort, stratified by ERBB2 protein expression (RPPA). Results: Of the 2,054 NSCLC cases, the primary analysis focused on adenocarcinoma cases (n = 1,641), which included 229 (14.0%) HER2 OE tumors. These tumors exhibited a significantly lower proportion of inflamed immune phenotype compared to the non-OE group (10.2% vs 22.1%, P < 0.0001). Intratumoral TIL and stromal TIL densities were reduced by 26.1% (P < 0.0001) and 14.7% (P < 0.001), respectively. Notably, even among HER2 OE tumors, a subset with high 3+ cell proportion (≥50% 3+ tumor cells) demonstrated a dose-dependent trend with an even lower inflamed proportion (3.7% vs 14.0%, P = 0.004) and reduced intratumoral TIL density (-36.7%, P < 0.001). The non-adenocarcinoma cohort (n = 413), where HER2 OE was present in 57 (13.8%) cases, showed a trend towards reduced intratumoral TIL density (-20.6%, P = 0.080). Analysis of H&E WSIs in TCGA LUAD validated that ERBB2-high tumors had significantly lower intratumoral TIL density (-16.1%, P = 0.027). Furthermore, gene expression analysis revealed that these tumors had downregulation of immune effector (GZMB, PRF1) and exhaustion (LAG3) markers, while maintaining an epithelial phenotype (OCLN, CDH1) with decreased mesenchymal markers (CDH2). Conclusions: NSCLC with HER2 OE is characterized by a distinct, immune-cold phenotype that is more pronounced with higher HER2 expression. These findings highlight a unique TME landscape that warrants further investigation to understand its potential impact on therapeutic responses.