Tomsk Review Says Integrative Tumor Microenvironment Models Outperform Single Biomarkers for Predicting Immunotherapy Response in Gastric Cancer

Which trials the authors lean on

Springer gates the full text. What follows comes from the abstract, the PubMed record, and the one component Springer does display in full, the reference list. The checkpoint-inhibitor trials cited are ATTRACTION-2, the Phase 3 nivolumab study in chemotherapy-refractory gastric and gastroesophageal junction cancer, and three pembrolizumab trials: KEYNOTE-059, KEYNOTE-061 and KEYNOTE-062. Neither CheckMate 649 nor KEYNOTE-859, the two first-line chemotherapy-combination trials that now anchor most guideline language, appears among the 84 entries. Also cited are the 2014 Cancer Genome Atlas molecular classification of gastric adenocarcinoma, Davis and Patel's 2019 analysis of PD-L1 as a predictive biomarker across all FDA checkpoint-inhibitor approvals, Shitara and colleagues' exploratory gene-signature analysis from KEYNOTE-061, and Lei and colleagues' 2021 analysis of PD-L1 and inflammatory gene expression with nivolumab with or without ipilimumab.

From PD-L1 and CD8 counts to spatial phenotypes

The abstract frames resistance to immunotherapy, both primary and acquired, as a product of tumor-immune-stromal interactions in what the authors call "a dynamic ecosystem," and describes immunotherapy response as "a systems-level phenomenon." Existing markers, they write, "have clinical value, yet their insufficiency reflects the multidimensional nature of tumor-immune crosstalk." The references show what fills out that claim: studies pairing PD-L1 expression with CD8+ T-cell infiltration, work on CXCL13+ CD8+ T cells and tertiary lymphoid structures, the WJOG10417GTR study of myeloid subsets that blunt anti-PD-1 activity, papers on tumor-associated macrophage CXCL8 and IL-10 production, cancer-associated fibroblast subsets marked by MFAP2 or SUSD2, tumor mutational burden estimated from targeted panels, Epstein-Barr virus-associated tumors, ARID1A deficiency, m6A RNA methylation and lactylation signatures, and a multiplex immunohistochemistry study that split gastric cancers into two cholesterol-metabolism patterns.

On the integrative side, the list includes Chen and colleagues' 2022 Nature Communications paper predicting immunotherapy response from multi-dimensional analysis of the immune microenvironment, the TIMES001 trial of a tumor microenvironment RNA test in advanced gastric cancer published in Med in 2024, a 2023 American Journal of Pathology study that predicted molecular features relevant to immunotherapy from hematoxylin and eosin whole-slide images with deep learning, and a 2024 pathomics-driven ensemble model in the Journal for ImmunoTherapy of Cancer.

The group's own 16-patient dataset

Among the references is the Tomsk group's own study, published in Cancers on July 21, 2025, with Tashireva as first author and Tsarenkova, Kalinchuk, Shmakova and Vtorushin as co-authors. That paper enrolled 16 patients with PD-L1-positive (CPS of 1 or higher) gastric adenocarcinoma treated with eight cycles of FLOT plus pembrolizumab, graded response by Mandard tumor regression grade, and profiled tumors before and after treatment with 10x Genomics Visium spatial transcriptomics and multiplex immunofluorescence. Responders expressed more IL1B, CXCL5, HMGB1 and IFNGR2; non-responders had more LGALS3, IDO1 and CD55, a higher density of FoxP3+ regulatory T cells (median 5.36% versus 2.41%, p = 0.0032) and more PD-1+ CD8+ T cells and PD-1+ FoxP3+ cells. PD-L1 CPS did not differ between the two groups. That result is the kind of single-marker shortfall the new review generalizes.

Limits of an abstract-only reading

Because the review's body is not open, it is not possible to report here how the authors weight the familiar clinical markers against one another, which assay platforms or cut-offs they favor, or what they ask pathology laboratories to put in a report. The abstract's stated conclusion is that predicting response "requires understanding the tumor-immune ecosystem as an integrated, dynamic system to improve patient stratification and outcomes," and that static, single-timepoint biomarker assessment is complicated by the microenvironment changing over the course of treatment.

Why This Matters to the APO|APE Reader

The trial set the Tomsk authors cite stops at KEYNOTE-062, which means the review is arguing for composite microenvironment scores without engaging the CPS-threshold debates that CheckMate 649 and KEYNOTE-859 produced in first-line practice. That gap, plus the 16-patient size of the group's own spatial dataset, puts the burden on the prospective validation the authors themselves say is missing. For now the gastric cancer immunotherapy selectors with regulatory standing remain PD-L1 CPS by immunohistochemistry and MSI or mismatch-repair status, and a laboratory adopting a multiplex or RNA-based microenvironment score would be doing so ahead of any endorsement in the literature this paper surveys.