A comprehensive review led by Associate Researcher Linnan Zhu and Academician Zemin Zhang at Biomedical Pioneering Innovation Center (BIOPIC), Peking University, China, and Chongqing Medical University, China, was published in Volume 2, article number 27 of the journal Immunity & Inflammation on June 05, 2026. The article systematically synthesizes recent advances in single-cell and spatial transcriptomics of the TME, summarizes tumor-enriched cell subtypes and their interactive networks, and prospectively introduces the "virtual tumor" concept. The review provides an integrated roadmap from fundamental TME biology to next-generation immunotherapies and AI-driven precision immunotherapy.
The TME is a complex multicellular ecosystem composed of tumor cells, immune cells, stromal cells, blood vessels, and neurons. These components co-evolve with malignant cells, collectively influencing tumor initiation, progression, immune evasion, and therapeutic response. Recent rapid advances in single-cell sequencing and spatial omics open an era of "high-dimensional panoramic understanding" of cancer, leading to the identification of numerous tumor-enriched cell subsets closely related to prognosis and treatment response.
Among lymphocytes, CD8+ cytotoxic T lymphocytes (CTLs) are the core effectors of anti-tumor immunity, recognizing tumor antigens via MHC-I and killing tumor cells through perforin and granzymes. However, they often enter a dysfunctional, exhausted state in the TME. CXCL13+ T cells represent a population with both predictive and therapeutic significance. CXCL13+ pre-exhausted CD8+ T cells are enriched in multiple tumor types and correlate with favorable responses to immune checkpoint blockade (ICB). CXCL13+ Th1-like cells exhibit a unique "exhausted yet activated" state that recruits B cells to promote tertiary lymphoid structure formation. In contrast, TNFRSF9+ regulatory T cells possess stronger immunosuppressive activity, forming a major barrier. Beyond T cells, B and NK cells also play important roles. Tumor-associated B cells (FCRL4+) express high MHC-II and co-stimulatory molecules, correlating with better prognosis and ICB response. Tumor-associated NK cells (DNAJB1+) show a dysfunctional state with reduced killing capacity, associated with poor prognosis and PD-1 therapy resistance.
In the myeloid compartment, SPP1+ tumor-associated macrophages are pro-tumorigenic, promoting angiogenesis, hypoxia responses, and extracellular matrix remodeling, and are strongly linked to poor prognosis. The mutually exclusive CXCL9-SPP1 expression defines a key macrophage polarity axis with better clinical predictive value than the traditional M1/M2 classification. LAMP3+ dendritic cells (DCs) are mature migratory DCs; the cDC1-derived subset expresses CXCL9 and IL-15, positively associated with CD8+ T cell infiltration and ICB response. HLA-DR+ antigen-presenting neutrophils can induce antigen-specific T cell responses, and antigen-presenting mast cells in triple-negative breast cancer promote anti-PD-1 responses, demonstrating myeloid functional diversity.
Among stromal and neural cells, LRRC15+ cancer-associated fibroblasts represent a terminally differentiated fibroblast lineage associated with immune-excluded tumors, dependent on TGF-β signaling. CXCR4+ endothelial tip cells promote angiogenesis and poor prognosis, while tumor-associated high endothelial venules and ACKR1+ endothelial cells enhance immune cell infiltration. Notably, recent discoveries in cancer neuroscience show that TGFBI+ Schwann cells induced by TGF-β promote tumor cell migration and correlate with poor prognosis, adding a new dimension to understanding neuro–immune–tumor interactions.
"These cell subsets do not act in isolation but form complex multi-cellular networks through spatial organization and functional cooperation," the authors pointed out. Concepts such as "cell modules" and "immunity hubs" have shifted focus from individual cell types to spatially organized, functionally coordinated, dynamically co-evolving multi-cellular units. Tumor progression involves progressive dissolution of healthy multi-cellular networks and acquisition of aberrantly conserved tumor-associated modules. This insight provides a conceptual basis for understanding shared TME remodeling trajectories across cancers and developing broad-spectrum TME-targeted therapies.
Building on the AI virtual cell concept, the review prospectively introduces the "AI virtual tumor" model: integrating cellular composition, spatial tissue architecture, cell-cell communication, and perturbation response rules to extend single-cell behavior modeling to tumor-scale ecosystem dynamics. "This could provide new computational frameworks for patient stratification, combination drug design, and efficacy prediction," the authors highlighted.
For immunotherapy, the review summarizes three frontiers. First, in ICB, CXCL13+ T cells predict a favorable response, while CCR8+ Tregs, SPP1+ TAMs, and LRRC15+ CAFs associate with resistance. Dual LAG-3/PD-1 blockade (relatlimab plus nivolumab) shows promise. Second, for adoptive cell therapy, CAR-T cells have succeeded in hematologic malignancies, while CAR-M cells show potential in solid tumors due to infiltration advantages, entering phase I trials. Third, for personalized cancer vaccines, cDC1-targeted vaccines may overcome ICB resistance in pancreatic cancer, and mRNA neoantigen vaccines have shown safety and immunogenicity in high-risk renal cell carcinoma. Together, these advances outline a new path integrating TME insights, next-generation immunotherapies, and AI-driven strategies to advance mechanism-based precision oncology.
