Research Radar — 2026-06-18

Summary: Reports a spatially resolved map of the human proteome across a variety of healthy tissues and cancers, representing one of the most comprehensive surveys of protein-level spatial organization in the human body. While spatial transcriptomics has produced detailed atlases of gene expression across tissues, proteins — the functional effectors of cellular phenotypes and the direct targets of most therapeutics — have lacked a comparable spatially resolved resource. This is a critical gap because mRNA and protein levels often correlate poorly, particularly for secreted proteins, cell-surface receptors, and proteins regulated by post- translational mechanisms. The authors use spatially resolved mass spectrometry-based proteomics to profile protein abundance and distribution across dozens of healthy human tissue types and matched cancer specimens. The resulting atlas reveals that protein spatial organization is substantially more structured than transcript-level organization — proteins form sharper spatial boundaries, exhibit more tissue-specific distributions, and show cancer-specific redistribution patterns that are not visible at the transcript level. The resource identifies proteins with highly restricted spatial expression patterns that may represent ideal therapeutic targets — highly expressed in specific cancer regions but absent from critical normal tissues. The atlas provides wide-ranging insights in developmental biology and oncology and is made available as an interactive resource for the research community.

Why it matters: This study fills a critical gap between the transcript-centric view of tissue biology that has dominated the genomics era and the protein-level reality that matters for drug development and disease mechanism. The finding that protein spatial organization is more structured than transcript organization has practical implications: therapeutic targets identified from transcript-level data may show broader tissue distribution at the protein level, creating toxicity risks that transcript-level analysis would miss. Conversely, proteins with highly restricted spatial expression — visible only at the protein level — represent opportunities for highly selective therapies. The cancer vs. normal tissue comparisons are particularly valuable for oncology drug development, where the therapeutic window is defined by differential expression between tumour and normal tissue. This resource is likely to become a standard reference for preclinical target validation, analogous to how GTEx and the Human Protein Atlas are used today but with the added dimension of spatial resolution.

Why for Yiru: The TME is fundamentally a protein-level ecosystem — checkpoint ligands (PD-L1), cytokines (IL-10, TGF-β), chemokines, and extracellular matrix proteins mediate the cell-cell interactions that determine immune infiltration, activation, and suppression. Yet most TME atlases to date are transcript-level, missing the protein-level spatial organization that actually determines immune cell behaviour. This spatial proteome atlas could be used to annotate TME regions in cancer specimens with their protein-level characteristics: which immune checkpoint proteins are actually present (not just transcribed), where in the TME they are located, and how protein-level immune signatures differ from transcript-level signatures. The finding that protein spatial boundaries are sharper than transcript boundaries is also relevant to TME analysis — it suggests that protein-level immune exclusion boundaries (e.g., the boundary between tumour nest and immune infiltrate) may be more sharply defined than transcript-level analysis would suggest. For computational TME research, this resource provides a reference for validating transcript-based TME classifications against protein-level ground truth.

Summary: Introduces CANVAS, a computational platform that decodes spatial tumour ecosystems from routine H&E histopathology to enable population-level prognostic modeling and immunotherapy response prediction. Routine H&E-stained histopathology slides are universally available for cancer patients — they are generated as part of standard clinical care and archived in pathology departments worldwide. However, the rich spatial information encoded in these images — cellular composition, tissue architecture, tumour-immune spatial relationships — has been underutilized because extracting it has required specialized staining (multiplex immunohistochemistry) or spatial omics technologies that are not routinely available. CANVAS addresses this by using deep learning to infer cellular architecture and spatial neighbourhood features directly from H&E images. The platform constructs a virtual spatial profile of each tumour that includes inferred cell-type composition, spatial neighbourhood characteristics, and tumour-immune interaction features. Applied to large retrospective cohorts across multiple cancer types, CANVAS-derived spatial features predict patient prognosis and immunotherapy response with accuracy comparable to or exceeding that of specialized spatial omics measurements. Critically, because H&E slides are universally available, CANVAS can be applied to population-scale retrospective cohorts — hundreds of thousands of patients — enabling epidemiological-scale spatial tumour ecology studies that would be impossible with specialized spatial technologies.

Why it matters: CANVAS represents a practical bridge between the promise of spatial tumour analysis and the reality of clinical pathology workflow. Specialized spatial technologies (imaging mass cytometry, spatial transcriptomics) provide rich data but are expensive, low-throughput, and available at only a handful of centers — they will never be applied to every cancer patient. H&E histopathology, by contrast, is universal. CANVAS extracts spatial tumour ecosystem features from these universally available images, making spatial tumour analysis potentially accessible to every cancer patient worldwide. The ability to apply CANVAS to massive retrospective cohorts — the kind of dataset sizes needed to identify robust biomarkers of immunotherapy response across heterogeneous patient populations — is a significant advance. If validated prospectively, CANVAS could enable routine spatial tumour profiling as part of standard pathology workflow, without requiring additional tissue sections, staining, or instrumentation.

Why for Yiru: Spatial organization of the TME — the proximity of CD8+ T cells to tumour cells, the formation of immune excluded vs. inflamed vs. desert phenotypes, the spatial arrangement of immunosuppressive niches — is central to understanding immunotherapy response and resistance. CANVAS provides a way to study these spatial features at unprecedented scale using existing H&E data. For computational TME research, CANVAS could be applied to cohorts from clinical trials to identify spatial TME features that predict which patients benefit from specific immunotherapies. The virtual spatial profiling approach — inferring spatial features from routine images — is also methodologically interesting: it demonstrates that much of the spatial information captured by expensive specialized technologies is also encoded, in a more subtle form, in routine H&E images. This insight could inform the design of more efficient spatial TME studies that use H&E as a screening tool and reserve specialized spatial technologies for validation.

Summary: Reports that a single intracerebroventricular dose of CAR T cells in patients with recurrent glioblastoma induces dose-dependent remodeling of the endogenous immune compartment, and that the nature of this remodeling — rather than CAR T cell persistence alone — is associated with clinical outcomes. CAR T cell therapy has shown limited efficacy in solid tumours compared to haematological malignancies, and glioblastoma has been particularly challenging due to the blood-brain barrier, profound immunosuppression, and tumour heterogeneity. This study, based on a clinical trial of intracerebroventricular CAR T cell administration, performs deep immunophenotyping of the tumour microenvironment and cerebrospinal fluid before and after treatment. The key finding is that CAR T cell therapy does not operate in isolation — it triggers a cascade of endogenous immune responses that have divergent effects on outcome. Expansion of cytotoxic NK cells after CAR T infusion is associated with prolonged survival, suggesting that CAR T cells can "spark" an endogenous anti-tumour immune response. Conversely, expansion of regulatory T cells (Tregs) and the presence of baseline scavenger myeloid cells are associated with poor outcomes, suggesting pre-existing immunosuppressive niches that resist CAR T cell-mediated remodeling. The study emphasizes that the therapeutic effect of CAR T cells in solid tumours may depend less on the CAR T cells themselves and more on their ability to reprogramme the endogenous immune microenvironment.

Why it matters: This study provides a mechanistic explanation for why CAR T cell therapy has succeeded in haematological malignancies but struggled in solid tumours: in liquid tumours, CAR T cells can directly access and eliminate their targets, but in solid tumours, the endogenous immune microenvironment — which is often profoundly immunosuppressive — determines whether CAR T cells can function. The finding that CAR T cell efficacy depends on the pre-existing immune context (scavenger myeloid cells, Treg levels) and on the CAR T cells' ability to recruit endogenous effectors (NK cells) has immediate clinical implications. It suggests that patient stratification based on pre-treatment TME composition could identify those most likely to benefit, and that combination therapies — CAR T cells plus agents that deplete Tregs, reprogramme myeloid cells, or activate NK cells — may be necessary for efficacy in immunosuppressive tumours. More broadly, the study reframes CAR T cell therapy for solid tumours from a "living drug" model (where the CAR T cells are the primary effectors) to an "immune catalyst" model (where CAR T cells initiate a cascade that requires endogenous immune cooperation).

Why for Yiru: This study is a powerful illustration of why understanding the TME is essential for immunotherapy development. The divergent effects of NK cell expansion (good) vs. Treg expansion and scavenger myeloid cells (bad) after CAR T cell therapy are exactly the kinds of TME features that could be identified computationally from pre-treatment tumour samples to guide patient selection. For computational TME research, this suggests a specific analytical framework: characterize the pre-treatment TME immune composition (myeloid, Treg, NK cell signatures) in glioblastoma patients undergoing CAR T cell therapy, and test whether these features predict response. The "immune catalyst" model also has implications for how we computationally model immunotherapy — rather than modeling CAR T cells as autonomous killers, we should model them as initiators of an immune cascade whose outcome depends on the state of the endogenous immune network.

Summary: Uses genome-wide CRISPR screens combined with clonal barcoding to identify regulators of the stem-like TCF7+ CD8 T cell state, revealing Trim28 as a key mediator of T cell differentiation in tumours. Stem-like TCF7+ CD8 T cells sustain anti-tumour responses and are the subpopulation that responds to immune checkpoint blockade — their abundance in tumours before treatment predicts immunotherapy response, and their maintenance during treatment is associated with durable tumour control. However, the transcriptional regulators that control the TCF7+ state, and the signals that drive these cells to differentiate into exhausted effector cells, have been incompletely defined. The authors address this using a powerful experimental design: genome-wide CRISPR screens in primary CD8 T cells, combined with random barcodes that link clonal relationships to guide identity and transcriptional states in single cells. This clonally-resolved approach enables inference of differentiation trajectories and differentiation rates while controlling for confounding clonal effects — a critical advance because T cell clones expand at different rates, and failure to account for clonal structure can produce spurious gene-phenotype associations. The screen identifies Trim28 as a key regulator: Trim28-deficient T cells in tumours are enriched in the TCF7+ stem-like state, depleted in cycling and terminal effector states, and show reduced differentiation rates. This suggests Trim28 promotes the transition from stem-like to effector states, and that inhibiting Trim28 may help maintain the stem-like pool that sustains anti-tumour immunity.

Why it matters: Maintaining the stem-like TCF7+ CD8 T cell pool during immunotherapy is one of the central challenges in cancer immunology — these cells are the source of the effector cells that kill tumours, and their depletion (through excessive differentiation into terminally exhausted cells) is a major mechanism of immunotherapy resistance. Identifying the transcriptional regulators that control the balance between self-renewal and differentiation in these cells opens therapeutic opportunities: drugging Trim28 or its downstream targets could help maintain the stem-like pool during checkpoint blockade, extending the duration and depth of anti-tumour responses. The clonally-resolved CRISPR screening approach is also methodologically important — it addresses a fundamental challenge in single-cell CRISPR screens (confounding by clonal expansion) that has limited the interpretability of previous studies. This approach could be applied to dissect regulators of other T cell states relevant to tumour immunity.

Why for Yiru: The TCF7+ stem-like CD8 T cell state is one of the most clinically important immune states in the TME — its abundance predicts immunotherapy response across multiple cancer types, and its biology is a focus of intense investigation. Trim28 as a regulator of the transition from stem-like to effector states adds a new molecular handle to this biology. Computationally, TME single-cell data could be analyzed to determine whether Trim28 expression or activity correlates with TCF7+ cell abundance, differentiation trajectories, or immunotherapy outcomes across patient cohorts. The clonally-resolved CRISPR screening framework is also relevant to computational TME methods development — integrating clonal information (from TCR sequencing or lineage barcoding) with single-cell transcriptomics is exactly the kind of analysis needed to understand T cell dynamics in the TME, and methods developed for these screens could be adapted for analysing clinical TME data.

Summary: Identifies citrate compartmentalization as a cellular mechanism by which CD8+ T cells couple cytokine production to glucose availability, revealing a direct metabolic control point for effector function. Cytokine production is a core function of effector T cells, yet the mechanisms that regulate cytokine output during an immune response remain incompletely understood — particularly how T cells adjust their cytokine production in response to the nutrient-depleted tumour microenvironment. Under glucose-replete conditions, the authors find that citrate is transported from the mitochondria to the cytosol by the citrate carrier SLC25A1, where it suppresses calcium-dependent transcription factor activity in effector T cells. This suppression limits cytokine production. When glucose availability is reduced — as occurs when T cells infiltrate glucose-poor tumour microenvironments — or when citrate transport across the mitochondrial membrane is blocked, calcium-dependent transcription is de-repressed, and cytokine production increases. The SLC25A1-citrate-calcium-cytokine axis thus functions as a nutrient sensor: when glucose is abundant, citrate export keeps cytokine production in check; when glucose is scarce, reduced citrate export unleashes cytokine production. This mechanism may represent an adaptive response that enables T cells to produce cytokines specifically in nutrient-poor environments like tumours.

Why it matters: Immunometabolism — the intersection of cellular metabolism and immune function — is one of the most dynamic areas in immunology, and this study adds a novel mechanism: citrate as a signaling molecule that links nutrient availability to effector function. The finding that citrate compartmentalization controls cytokine production has therapeutic implications: modulating SLC25A1 activity or citrate levels could be used to tune T cell cytokine output — enhancing it in the context of cancer immunotherapy (where stronger T cell effector function is desired) or suppressing it in autoimmune disease. The glucose-sensing aspect of this mechanism is also important for understanding T cell function in the TME, where glucose competition between tumour cells and T cells is a well-documented mechanism of immune suppression. This study suggests that glucose deprivation may actually enhance certain aspects of T cell function (cytokine production) through the citrate pathway, even as it impairs others (proliferation).

Why for Yiru: Metabolic competition in the TME is a central determinant of anti-tumour immunity — tumour cells consume glucose and glutamine at high rates, starving infiltrating T cells of the nutrients they need for effector function. The citrate compartmentalization mechanism adds a new dimension: glucose deprivation may not simply impair T cell function uniformly but may trigger specific metabolic signaling responses that alter the quality of the T cell response. For computational TME research, this suggests that metabolic gene expression signatures — including SLC25A1, citrate pathway enzymes, and calcium signaling components — could be analyzed in TME single-cell data to predict T cell functional states and cytokine production capacity. The finding also connects to TME metabolomics, where direct measurement of citrate and other TCA cycle intermediates in tumour interstitial fluid could provide a metabolic readout of the T cell functional environment.

Summary: Systematically dissects the molecular determinants of cancer cell killing by cytotoxic lymphocytes, revealing that the Fas-FADD-caspase-8 death receptor pathway is a critical cancer cell-intrinsic determinant of killing efficacy — independent of the perforin-granzyme pathway. Cytotoxic lymphocytes (CD8+ T cells and NK cells) kill target cells through two major pathways: the death receptor pathway (FasL-Fas, TRAIL-DR5) and the perforin-granzyme pathway. These pathways are generally thought to converge on executioner caspases (caspase-3/7) to drive apoptosis. Using a reductionist approach, the authors systematically disrupted key cell death mediators in cancer cells and tested their sensitivity to cytotoxic lymphocyte-mediated killing. Strikingly, loss of executioner caspases (caspase-3/7) conferred only limited resistance to killing, suggesting that cytotoxic lymphocytes can kill cancer cells through executioner caspase-independent mechanisms. In contrast, loss of the Fas-FADD-caspase-8 axis — the upstream death receptor signaling module — conferred substantial resistance, revealing that this pathway is a critical determinant of killing that operates at least partly independently of the canonical executioner caspase cascade. The study identifies cancer cell-intrinsic expression of Fas pathway components as a predictor of sensitivity to cytotoxic lymphocyte killing and suggests that tumours may evade immune killing by downregulating this pathway.

Why it matters: Understanding the molecular determinants of whether a cancer cell lives or dies when engaged by a cytotoxic T cell is fundamental to cancer immunotherapy. The finding that the Fas-FADD-caspase-8 axis, rather than executioner caspases, is the critical determinant of killing efficacy challenges the canonical model of cytotoxic lymphocyte-mediated apoptosis and suggests that cancer cells can evade killing by mechanisms upstream of the executioner caspases. This has immediate translational implications: Fas pathway component expression could serve as a biomarker for immunotherapy response, and therapeutic strategies that upregulate Fas or sensitize the Fas pathway in tumour cells could enhance the efficacy of checkpoint blockade, CAR T cell therapy, and other T cell-based immunotherapies. The identification of executioner caspase-independent killing mechanisms also raises the possibility that tumours believed to be apoptosis-resistant (due to BCL-2 overexpression or caspase downregulation) may actually be sensitive to T cell killing through this alternative pathway.

Why for Yiru: Immune evasion is a hallmark of cancer, and understanding the molecular mechanisms by which tumour cells resist T cell killing is central to TME research. This study identifies Fas-FADD-caspase-8 as a cancer cell-intrinsic determinant of killing sensitivity — meaning that whether a tumour cell dies when contacted by a CD8+ T cell depends on the tumour cell's own expression of these pathway components. For computational TME analysis, this suggests that Fas, FADD, and caspase-8 expression levels in tumour cells — measurable in single-cell RNA-seq data — could be used to predict T cell killing efficacy and immunotherapy response. The finding that caspase-3/7 loss confers limited resistance also suggests that common apoptosis resistance mechanisms in cancer (BCL-2 overexpression, caspase downregulation) may not protect against T cell killing as effectively as assumed. This has implications for how we interpret apoptosis gene signatures in TME data: canonical apoptosis resistance markers may not be good predictors of immune evasion.