Single-cell and spatial multi-omics expose why some tumors resist immunotherapy
Immune checkpoint blockade has changed cancer care, but resistance still limits how far the benefit can reach. The new review by Shi, Guo, Tang, Liu, Guo, Fang, Xu, and Zhuang argues that the familiar biomarkers used in practice — PD-L1, tumor mutational burden, and microsatellite instability — capture only a narrow slice of the biology that governs response. The real problem sits inside the tumor immune microenvironment, where immune cells, stromal elements, blood vessels, and metabolic pressures interact in space and over time.
A single biopsy can miss the architecture that determines whether an antitumor immune response can actually function. In many cancers, immune cells are present but ineffective, excluded from malignant nests, or trapped inside suppressive niches that blunt their activity before they reach tumor cells.
Exhausted T cells, suppressive myeloid cells, and stromal barriers act in separate but connected programs
Single-cell profiling has made the immune ecosystem look far more fragmented than older bulk assays suggested. The review describes diverse immune and stromal states, including progenitor and terminally exhausted T cells, suppressive myeloid programs, B-cell and tertiary lymphoid structure associated immune-reactive states, and cancer-associated fibroblast-driven exclusion phenotypes. Each state carries distinct functional consequences for antitumor immunity.
Progenitor exhausted T cells may still retain some capacity to expand after checkpoint blockade, while terminally exhausted cells are further along a dysfunctional trajectory and are less likely to recover full effector activity. Myeloid cells can shape a local environment that favors immune suppression rather than immune attack, and cancer-associated fibroblasts can form physical and biochemical barriers that keep T cells at the margins of the tumor. Those barriers are not decorative features of the tumor; they directly determine whether cytotoxic lymphocytes can find malignant cells and receive the support they need to kill them.
A tumor may contain immune cells and still behave as if it is immunologically empty.
Spatial biology explains why location changes function
Spatial transcriptomics, spatial proteomics, and imaging-based methods add the missing geographic layer. They show that immune cell states do not float freely through tumors; they assemble into distinct niches with characteristic biology. The review identifies immune-inflamed niches, T-cell-excluded niches, myeloid-suppressive niches, metabolic or hypoxic niches, and tertiary lymphoid structure associated niches.
This spatial organization matters because proximity controls communication. Antigen-presenting support, inhibitory ligand-receptor signaling, oxygen deprivation, nutrient competition, and stromal restraint all depend on where cells sit relative to one another. A T cell positioned near a suppressive myeloid cluster may behave very differently from the same cell placed in a tertiary lymphoid structure rich in immune activation. Hypoxia can further reprogram local immunity, while stromal and vascular features can prevent immune infiltration altogether. The review frames the tumor as a set of competing neighborhoods, each with its own immunologic rules.
Ligand-receptor networks link cellular states to treatment failure
The review places strong emphasis on ligand-receptor communication networks as a mechanistic bridge between cell identity and spatial dysfunction. These signaling circuits help explain how one cell population can instruct another to become more suppressive, more exhausted, or less mobile. In that sense, resistance is not only a property of mutated tumor cells. It is also an emergent behavior created by constant messaging among immune, stromal, endothelial, and malignant compartments.
This matters for immunotherapy because checkpoint blockade can only work if enough of the immune machinery remains available for reinvigoration. When suppressive myeloid signaling dominates, or when fibroblast-rich barriers fence off tumor nests, checkpoint inhibition may release a brake that is only attached to a small fraction of the immune response. When tertiary lymphoid structure-associated niches are present, the biology looks different: these zones can support antigen presentation and immune priming, offering a more favorable setting for response.
Translational biomarkers will need to move beyond single markers
The most practical message in the review is also the most demanding one for clinical implementation. Future biomarkers will need to move from static, single-marker assays toward longitudinal, spatially resolved, and interpretable multi-omics models. That means incorporating where cells are, what states they occupy, how those states change over time, and how neighboring cells shape one another.
A single PD-L1 result or one-time genomic readout cannot reliably capture that complexity. The review argues that patient stratification should increasingly rely on combined measurements that can identify immune-inflamed disease, immune-excluded disease, myeloid suppression, metabolic stress, and tertiary lymphoid structure-associated activity. In practice, such a framework could help separate patients most likely to benefit from checkpoint blockade alone from those who may need combination strategies that target stromal barriers, myeloid suppression, or hypoxic remodeling.
That shift is not just technical. It changes the biological question from “Does this tumor express a marker?” to “What ecosystem is this tumor building, and can treatment rewire it?”
What the field still needs before these tools reach routine care
Despite the promise of single-cell and spatial platforms, the review leaves clear clinical hurdles on the table. These methods are complex, expensive, and not yet standardized across institutions or tumor types. Their current value is strongest as a discovery engine and as a way to refine biomarker design, rather than as a universal bedside test.
Another challenge is interpretation. Multi-omics data can produce a dense map of cell states and signals, but the goal is not to collect more information for its own sake. The goal is to define a model that is robust enough for patient selection and simple enough to guide treatment decisions. Longitudinal sampling may be especially important, because resistance is dynamic and can evolve after therapy starts.
The review makes a pointed case that immunotherapy biomarkers must become more spatial, more temporal, and more mechanistically grounded if they are going to predict benefit with real clinical accuracy. Until then, the gulf between what a tumor contains and what a tumor does will remain too wide for single markers to bridge.