SO-12 CLDN18.2 expression in resectable gastroesophageal cancers: Associated clinical and molecular signatures and impact of neoadjuvant chemotherapy on intratumor heterogeneity

Claudin 18.2 (CLDN18.2), a key component of tight junctions, is overexpressed in 18-49% of gastroesophageal adenocarcinoma (GEA), emerging as an attractive targetable biomarker. Zolbetuximab, an anti-CLDN18.2 monoclonal antibody, plus chemotherapy has outperformed chemotherapy alone in a global biomarker-informed phase 3 study in the front-line setting, and interest is growing in exploring CLDN18.2-targeted agents in the perioperative space. However, data is required in the resectable stages, including chemotherapy effects, relationship with genomic and immune factors, and measures of heterogeneity. We evaluated CLDN18.2 and PD-L1 in 57 patients with resectable GEA with paired clinical and sequencing data (WGS and/or RNA-seq), prospectively collected between 2011-2019 in the OCCAMS dataset. CLDN18.2 (ab222512, Abcam) and PD-L1 (E1L3N, Cell Signaling) were assessed by immunohistochemistry on adjacent slides from FFPE surgical samples. Positivity was defined as moderate-to-strong (2+/3+) staining in either ≥40% or ≥75% of tumour cells for CLDN18.2 and a combined positive score (CPS) either ≥1 or ≥5 for PD-L1. Thirteen patients (23%) had matched additional surgical blocks (n=9) and/or pre-treatment endoscopic biopsies (n=5) to study CLDN18.2 spatial and temporal concordance, respectively, defined as a staining score change from positive to negative or vice versa. Tumour mutational burden (TMB), sample ploidy, and metrics of gained and lost genomic material (copy number aberrations - CNAs, and weighted genomic integrity index - wGII, %) were estimated as indicators of genomic instability. The bioinformatic tool MCP-counter was used to quantify immune and stromal populations in the tumour microenvironment (TME) from transcriptomic data. Mann Whitney test or t-test was used to compare baseline characteristics by CLDN18.2 status, and log-rank test to compare survival curves. Statistical significance was set at p = 0.05 (two-sided). Fifty-seven patients with resectable (stage I (5%), II (42%), III (22%)) GEA (GEJC (74%), GC (21%)) undergoing surgery with curative intent with (77%) or without prior neoadjuvant chemotherapy (nCT) (23%) were evaluated. CLDN18.2 was overexpressed in either 17.5% (≥40%) or 3.5% (≥75%), and CLDN18.2 overexpression (≥40%) was not significantly associated with clinicopathological features (age, sex, disease stage, primary tumour location, grade, Barrett’s, all p>0.05), or OS (HR 2.77; 95% CI: 0.67-7.03, p= 0.09). CLDN18.2 copy numbers (n=57, p=0.24) and RNA levels (n=20, p=0.48) were not associated with staining intensity. Compared to CLDN18.2-ve, CLDN18.2+ve cases did not differ in PD-L1 expression (30/10% versus 34/12.8% by CPS≥1/5), median TMB (5.75 versus 6.23 mut/Mb), sample ploidy (2.66 versus 3.27), CNAs (448 segments versus 501), wGII (45.85% versus 53.2%), or TME composition (all p>0.05). However, treatment-naïve patients were significantly more CLDN18.2+ve (≥40%) (OR=0.20; 95% CI: 0.05-0.90, p=0.04). CLDN18.2 spatial concordance was 100% in treatment-naïve CLDN18.2+ve (n=2/2) and 100% in CLDN18.2-ve receiving nCT (n=3/3) but only 25% (n=1/4) in CLDN18.2+ve receiving nCT. CLDN18.2 temporal concordance was found in 100% of CLDN18.2-ve (n=3/3) but only 1 out of 2 CLDN18.2+ve, with a conversion from CLDN18.2+ve to CLDN18.2-ve after nCT. CLDN18.2+ve and CLDN18.2-ve tumours had similar genomic and immune signatures. However, nCT substantially contributes to CLDN18.2 spatial and temporal heterogeneity. Larger cohorts will need to confirm these findings, which could have implications for therapy.