Molecular Characterization of Tumor-Microenvironment Interactions to Predict Drug Resistance in Gastric and GEJ Adenocarcinoma (GASTROIMMUNE)

This observational multicenter study will prospectively enroll approximately 250 patients with operable, locally advanced gastric or gastroesophageal junction adenocarcinoma who receive perioperative FLOT plus durvalumab according to current clinical practice. Patients typically present with locally advanced disease (T ≥2 and/or node-positive) as defined by the AJCC 8th edition, and are candidates for perioperative chemo-immunotherapy followed by surgery.

Peripheral blood samples for PBMC isolation and serum/plasma storage will be obtained at baseline before neoadjuvant treatment (N1), before surgery (N2), and approximately 2 months after completion of adjuvant therapy (N3). Tumor tissue will be collected through diagnostic endoscopic biopsies and surgical resections; fresh and FFPE specimens will be used for immunohistochemistry, pathomics and omics analyses, including organoid cultures.

The study will implement a multiparametric registry integrating clinical, imaging and omics data, ensuring harmonized data collection across the two IRCCS centers (IRCCS Saverio de Bellis and Humanitas Clinical and Research Center). Tumor-infiltrating lymphocytes and PBMCs will be profiled by spectral flow cytometry (up to 40 markers), single-cell RNA sequencing and TCR/BCR sequencing to define immune subsets, activation and exhaustion states, and clonal expansion patterns.

Whole-slide digital histology (HE-stained sections) will be analyzed to derive robust pathomic biomarkers related to vascular architecture and immune cell distribution, previously associated with response to anti-angiogenic therapies; these biomarkers will be validated in the perioperative gastric/GEJ cohort. Serum cytokine and chemokine panels will be assessed using Luminex xMAP technology, while CT imaging acquired for staging and follow-up will be annotated to calibrate mathematical models predicting treatment response.

Statistical analyses will treat this project as a pilot study, given the lack of previous evidence on predictors of complete response to perioperative therapy. Patients will be stratified by Tumor Regression Grade (TRG), with TRG 1-2 considered responders and TRG 3-4 non-responders, and effect-size measures, ensemble machine-learning algorithms and multilevel longitudinal models will be used to identify and evaluate angiogenic and immune markers associated with short-term outcomes.