Artificial intelligence (AI)-powered HER2 quantification continuous score (QCS) and tumor microenvironment (TME) analysis in HER2-amplified metastatic colorectal cancer (mCRC) treated with pertuzumab plus trastuzumab.

34 Background: Pertuzumab plus trastuzumab (per-/tra-) has shown clinical benefits in patients (pts) with HER2-amplified mCRC, but still there is an unmet need of biomarkers to optimize treatment decisions. In this study, we applied AI-powered whole-slide image (WSI) analyzers, to investigate the association of HER2 QCS and TME with clinical outcomes of per-/tra- in pts with HER2-amplified mCRC enrolled in TRIUMPH, a phase II study. Methods: TRIUMPH is a multicenter phase II study to evaluate the efficacy of per-/tra- in pts with mCRC with HER2 amplification confirmed by tumor tissue or circulating tumor DNA (ctDNA) analysis. HER2 immunohistochemistry (IHC) and H&E-stained WSIs from 30 pts enrolled in TRIUMPH were included in the analysis. AI-powered WSI analyzers, Lunit SCOPE HER2 and Lunit SCOPE IO (Lunit, Republic of Korea) detects tumor cells (TC) by HER2 staining intensity (negative, 1+, 2+, or 3+) in HER2-WSI, and detects various class of cells including tumor-infiltrating lymphocytes (TIL), macrophages and fibroblasts in H&E-WSI, respectively. Immune-excluded score (IES) was defined as the proportion of high stromal TIL but low intratumoral TIL area in all analyzable TME. Tumor response was measured by RECIST v1.1, and the primary endpoint was progression-free survival (PFS) assessed by the investigators. Results: All 30 tumor samples had proven HER2-amplification by either HER FISH or ctDNA analysis. The concordance rate between pathologists and AI to examine HER2 IHC was 86.7% (26/30), AI-powered HER2 QCS showed the proportion of HER2 3+ TC was widely distributed (median 73.9%; min-max 11.9%-98.9%) in the samples with HER2 3+ assessed by pathologists. Objective response rates (ORR) of per-/tra- in the whole set and a subgroup of HER2 IHC 3+ assessed by pathologists were 26.7% (8/30) and 34.8% (8/23), respectively. AI-powered HER2 QCS enabled enrichment of responders, as a subgroup of HER2 3+ QCS ≥ 50%, which is a higher cutoff than ASCO/CAP guideline (10%), had 42.1% (8/19) ORR, since all 8 responders harbored HER2 3+ QCS ≥ 50%. PFS and overall survival (OS) were significantly favorable in HER2 3+ QCS ≥ 50% group compared to < 50% group (median PFS [mPFS] 4.4 vs 1.4 m, hazard ratio [HR] 0.12 [95% CI, 0.04-0.38], p = 0.0000994; median OS [mOS] 16.5 vs 4.1 m, HR 0.13 [95% CI, 0.05-0.38], p = 0.000117, respectively). Interestingly, IES and the densities of macrophages and fibroblasts within cancer stroma were correlated with poor response to per-/tra-. Among HER2 3+ QCS ≥ 50% group, 5 pts with high IES (≥ 54%), macrophage density (≥ 26.3/mm 2 ), and fibroblast density (≥ 1790/mm 2 ) had ORR 0%, mPFS 1.3 m, and mOS 4.5 m. Conclusions: AI-powered HER2 QCS and TME analysis may provide additional information to precisely predict per-/tra- response in HER2-positive mCRC.