Weekly reads 18/8/25

From structure to signals: new frontiers in cancer, spatial biology, and synthetic therapies

Aug 24, 2025

This week’s reading spanned some of the most exciting directions in cancer and computational biology, where mechanics, spatial context, and multi-omics reshape how we think about disease. From colorectal tumors that co-evolve with their architecture and nuclear shape, to a new method for disentangling cell-cell interactions in spatial data, to tools that uncover hidden transcription factor activity, the common theme is clear: biology is not just about the “what” (mutations, expression) but also the “where” and the “how.” Add to that discoveries on how viral infections awaken dormant cancer, bacteria and viruses teaming up in synthetic therapies, and axonal injury driving glioblastoma, and the picture becomes one of a deeply contextual, dynamic, and targetable disease process.

Preprints/articles that I managed to read this week

Tumor Architecture and Nuclear Shape Co-Evolve with Stem Cell States to Drive Colorectal Cancer Heterogeneity

Bertillot et al. bioRxiv (2025). https://doi.org/10.1101/2025.08.11.669722

The paper in one sentence

This study reveals how colorectal cancer (CRC) progression is driven by a co-evolutionary feedback loop between tumor architecture, nuclear morphology, and cellular plasticity—where fragmented tissue structures and rounded nuclei promote aggressive fetal-like stem cell states and EMT, independently predicting poor patient outcomes.

Summary

The paper integrates deep learning-based histopathology, spatial transcriptomics, and patient-derived organoids to dissect how CRC heterogeneity emerges from dynamic interactions between tissue structure and cell states.

Nuclear shape predicts survival: rounder nuclei correlate with fragmented tumor architectures and worse prognosis, especially in stage II/III patients.
Architecture dictates cell fate: fragmented tumor regions (linked to poor outcomes) enrich fetal-like stem cell (ANXA1+) and EMT programs, while glandular regions retain classic LGR5+ stem cells.
Co-evolutionary dynamics: mechanical confinement (e.g., compression, collagen stiffness) suppresses Wnt signaling, driving LGR5+ → ANXA1+ transitions; stromal signals further diversify these into spatially segregated EMT and stem states.

Personal highlights

Nuclear morphometry as a prognostic compass: round nuclei aren’t just a passive marker—they’re tightly coupled to tissue fragmentation and fetal-like stem cell states, offering a quantifiable window into tumor aggression.
Architecture as a fate sculptor: fragmented tumor structures aren’t mere consequences of invasion; they actively drive cellular plasticity by mechanically suppressing Wnt and inducing ANXA1+ regenerative programs, creating a permissive niche for metastasis.
EMT-stem state tango: EMT and fetal-like stem states initially co-emerge as a hybrid population but later segregate under stromal cues—revealing a hierarchical plasticity model where microenvironmental inputs dictate phenotypic bifurcation.
Beyond mutational dogma: tumor aggression isn’t just about genetics; mechanical confinement (e.g., compression, collagen stiffness) and soluble stromal signals (e.g., CAF-derived factors) are independent architects of cell state diversity.

Why should we care?

This work reframes CRC progression as a dialogue between structure and cell identity, where physical forces and spatial context dictate transcriptional programs. For clinicians, it offers a practical tool—quantifying tumor fragmentation from standard biopsies—to stratify high-risk patients who might benefit from aggressive therapy. For researchers, it uncovers a mechanical-epigenetic axis of plasticity, suggesting that targeting stiffness or ANXA1+ states could disrupt metastasis. By linking nuclear shape, tissue topology, and stemness, the study challenges reductionist views of cancer, arguing that where cells live is as critical as what mutations they carry. For patients, this could mean future therapies that disrupt the tumor’s physical niche alongside its genetic drivers.

QuadST: A Robust Method to Identify Cell-Cell Interaction-Changed Genes in Spatially Resolved Transcriptomics Data

Song et al. Genome Research (2025). doi: 10.1101/gr.279859.124

The paper in one sentence

QuadST is a novel statistical method that leverages quantile regression to robustly identify genes influenced by cell-cell interactions in spatially resolved transcriptomics (SRT) data, overcoming limitations of distance measurement errors and unmeasured confounders.

Summary

Song et al. introduce QuadST, a computational framework designed to detect interaction-changed genes (ICGs)—genes whose expression is modulated by proximity to neighboring cells—in single-cell SRT datasets. Unlike existing tools that rely on predefined cell-pair groupings or ligand-receptor pairs, QuadST treats cell-cell distance as a continuous outcome, models associations at multiple distance quantiles, and contrasts signals across quantile levels to identify ICGs with controlled false discovery rates (FDR). Applied to mouse cortex datasets (seqFISH+ and MERFISH), QuadST revealed synapse-related ICGs among excitatory neurons and demonstrated superior robustness to measurement errors and confounders compared to methods like Giotto and NCEM.

Personal highlights

Distance-as-outcome innovation: QuadST reverses traditional modeling by treating cell-cell distance as the dependent variable, minimizing bias from measurement errors inherent in SRT technologies (e.g., inaccurate cell boundary detection).
Quantile regression for localized signals: by modeling gene-distance associations at multiple quantile levels (e.g., 0.02–0.98), QuadST captures nuanced interaction effects that are strongest at short distances, bypassing the need for arbitrary "on/off" distance thresholds.
Robust FDR control via contrastive analysis: QuadST contrasts P-values from nearby vs. distant quantiles to bound the FDR empirically, ensuring reliability even with unmeasured confounders or gene-gene correlations—a critical advance over methods like NCEM, which inflate false positives.
Cell-type-agnostic design: unlike tools requiring predefined interacting pairs (e.g., Giotto), QuadST autonomously analyzes all cell-type combinations, revealing non-canonical ICGs beyond ligand-receptor pairs (e.g., synaptic genes like CpIx1 and Nsmf in excitatory neurons).
Scalability and interpretability: QuadST efficiently processes large SRT datasets (tested on 13,745 MERFISH cells) and outputs directional ICG associations, clarifying whether genes are upregulated or downregulated with proximity.

Epiregulon: Single-cell transcription factor activity inference to predict drug response and drivers of cell states

Włodarczyk et al. Nat Commun (2025). https://doi.org/10.1038/s41467-025-62252-5

The paper in one sentence

Epiregulon is a new computational method that builds gene regulatory networks from single-cell multi-omics data to accurately predict transcription factor activity, even when it's decoupled from gene expression, enabling better prediction of drug response and identification of key drivers of cell states.

Summary

This paper introduces Epiregulon, a tool designed to infer the activity of transcription factors (TFs) and co-regulators from paired single-cell RNA-seq and ATAC-seq data. Traditional methods often fail when a TF's activity (what it's doing) is disconnected from its mRNA expression level (how much of it exists), a common scenario in drug treatment, cancer mutations, and cellular reprogramming. Epiregulon overcomes this by leveraging the co-occurrence of TF expression and chromatin accessibility at its binding sites in each cell. Crucially, it uses empirical ChIP-seq data to define these binding sites, making it uniquely capable of modeling co-regulators that lack DNA-binding motifs. The authors rigorously benchmark Epiregulon, showing it outperforms existing methods in predicting target genes, accurately quantifying the effect of AR-targeting drugs in prostate cancer models (including those with neomorphic mutations), uncovering context-specific partners of a chromatin remodeler (SMARCA4), and identifying drivers of lineage reprogramming.

Personal highlights

Co-occurrence over correlation for decoupled activity: instead of relying on linear correlations, Epiregulon's default method uses a Wilcoxon test to compare target gene expression in cells where both the TF is expressed and its binding site is accessible versus all other cells. This makes it robust to scenarios where a drug inhibits a TF's function without immediately reducing its mRNA levels.
Motif-agnostic inference via empirical ChIP-seq integration: the method sidesteps a major limitation of other tools by directly integrating a massive, pre-compiled database of ChIP-seq binding sites for 1377 factors, enabling the inference of activity for transcriptional co-regulators like SMARCA4 that lack defined DNA-binding motifs.
Pan-cell-type ChIP-seq as a powerful proxy: The study demonstrates that even when cell-type-matched ChIP-seq data is unavailable, using a pan-cell-type compendium of binding sites yields far more accurate activity estimates than relying on motif annotations alone, a key practical advantage for most research settings.
Sensitive detection of exogenous drivers in reprogramming: In a reprogramming assay, Epiregulon was the only method that could accurately detect the activity of introduced factors like NKX2-1, whose mRNA was poorly captured by the scRNA-seq protocol, by instead reading out their functional impact on downstream target genes.

Why should we care?

Epiregulon moves beyond the "what" and "where" of gene expression to tackle the critical "why" and "so what" in single-cell biology. For cancer researchers and drug developers, it provides a powerful lens to predict whether a therapy is hitting its intended target,especially important for emerging modalities like PROTAC degraders, and to identify compensatory mechanisms and novel co-dependencies. For basic scientists studying development or differentiation, it offers a more accurate way to pinpoint the master regulators driving cell fate decisions, even when their mRNA levels are misleading. By bridging the gap between chromatin state, regulator binding, and transcriptional output, Epiregulon provides a more mechanistic and actionable view of cellular regulation, turning complex multi-omics data into testable hypotheses about disease drivers and therapeutic vulnerabilities.

Spatially defined multicellular functional units in colorectal cancer revealed from single cell and spatial transcriptomics

Avraham-Davidi et al. bioRxiv (2025). https://doi.org/10.1101/2022.10.02.508492

The paper in one sentence

By integrating single-cell and high-resolution spatial transcriptomics in mouse and human colorectal cancer, this study identifies conserved, organized "cellular neighborhoods" with distinct functions that predict clinical outcomes.

Summary

This study provides a comprehensive spatial atlas of colorectal cancer (CRC) by combining single-cell RNA sequencing (scRNA-seq), high-resolution Slide-seq spatial transcriptomics, and multiplexed in situ RNA analysis. Using genetically engineered mouse models that mimic human CRC, the authors developed a computational framework (based on the TACCO method) to map not only cell types but also coordinated gene expression programs to their precise locations within tumors.

They discovered that tumors are not chaotic masses but are organized into spatially distinct "cellular neighborhoods" or functional units. Each neighborhood has a specific composition of epithelial, immune, and stromal cells governed by different biological programs (e.g., Wnt signaling, inflammation, angiogenesis). Crucially, they identified three key pro-tumorigenic neighborhoods: an inflammatory-angiogenic region, a stem-like/tertiary lymphoid structure region, and an epithelial-mesenchymal transition (EMT) region.

By comparing these findings to human CRC data, they showed that these organizational principles are conserved. Most importantly, the activity of these neighborhoods in human patients is predictive of clinical outcomes: the inflammatory region correlates with better survival, while the EMT region correlates with worse progression-free interval.

Personal highlights

Spatial mapping of multicellular functional units: the study moves beyond cataloging cell types to define organized, recurring "neighborhoods" within tumors where specific cell types and coordinated gene programs co-localize, forming functional ecological units.
Cross-species conservation of tumor organization: the spatial architecture and cellular co-variation patterns discovered in mouse models are recapitulated in human colorectal tumors, validating the mouse models and highlighting fundamental principles of tumor organization.
Linking spatial neighborhoods to clinical outcome: the activity of the defined malignant neighborhoods (inflammatory, stem-like, EMT) in human patient data directly correlates with progression-free survival and overall survival, transforming a spatial observation into a clinically relevant metric.
An integrative computational framework for spatial omics: the extended TACCO method provides a powerful tool to integrate scRNA-seq and spatial data, simultaneously annotating cell types, expression programs, and the spatial regions they define, overcoming the limitations of each individual technology.
Non-canonical functions of tumor cells: the work reveals that dysplastic stem cells exhibit a unique expression profile, secreting negative regulators of Wnt signaling and inflammatory molecules, suggesting a sophisticated mechanism for outcompeting normal neighbors and shaping the tumor microenvironment.

Why should we care?

This work changes how we view a tumor: not as a disorganized crowd of cells, but as a structured city with distinct functional districts. This spatial perspective is crucial because a cell's function is deeply influenced by its neighbors. By identifying these conserved neighborhoods and linking them to patient outcomes, this research provides a new lens for understanding cancer progression, resistance, and immune evasion.

For oncologists and cancer biologists, it offers a new set of spatial biomarkers that could improve patient stratification and reveal the functional context of therapeutic targets—knowing which neighborhood a drug is affecting is as important as knowing which cell. For computational biologists, it provides a robust framework and toolset (TACCO) for extracting biological meaning from the complex interplay of single-cell and spatial data. Ultimately, this spatial blueprint of cancer is a critical step toward developing more effective, spatially-informed therapies

Respiratory Viral Infections Awaken Dormant Breast Cancer Cells in the Lungs

Chia et al. (2025). Nature. DOI: 10.1038/s41586-025-09332-0

The paper in one sentence

Respiratory viral infections like influenza and SARS-CoV-2 disrupt the dormancy of metastatic breast cancer cells in the lungs by triggering IL-6-dependent inflammation and immune suppression, leading to aggressive cancer resurgence.

Summary

This study reveals a startling connection between respiratory viral infections (influenza and SARS-CoV-2) and the reactivation of dormant breast cancer cells (DCCs) in the lungs. Using mouse models, the researchers found that these infections induce a surge in interleukin-6 (IL-6), which awakens dormant DCCs, prompting them to proliferate and form metastatic lesions. The process involves two phases: an initial IL-6-driven expansion of cancer cells and a later phase where CD4+ T cells suppress immune surveillance, allowing the cancer to persist. Human epidemiological data from the UK Biobank and Flatiron Health further support these findings, showing that COVID-19 survivors with prior cancer diagnoses face a significantly higher risk of cancer-related mortality and lung metastasis.

Personal highlights

Viral infections break cancer dormancy: influenza and SARS-CoV-2 infections disrupt the quiescence of dormant breast cancer cells in the lungs, triggering rapid proliferation and metastatic growth.
IL-6 as the key awakening signal: the cytokine IL-6, produced during infection, is essential for reawakening dormant cancer cells, shifting them from a mesenchymal-like state to a proliferative hybrid phenotype.
Immune suppression sustains metastasis: CD4+ T cells recruited post-infection create a niche that inhibits CD8+ T cell activity, allowing awakened cancer cells to evade immune destruction.
Human data mirrors mouse findings: analyses of cancer survivors show that SARS-CoV-2 infection increases the risk of cancer-related death and lung metastasis, peaking in the months following infection.
Dual-phase mechanism: the study uncovers a two-step process—first, IL-6-driven awakening, followed by CD4+ T cell-mediated immune suppression—that sustains metastatic progression.
Implications for cancer survivors: the findings highlight a previously overlooked risk for cancer survivors, suggesting that respiratory infections could trigger deadly cancer recurrence.
Potential therapeutic targets: blocking IL-6 or modulating CD4+ T cell activity could offer strategies to prevent infection-induced cancer resurgence.

Why should we care?

This research uncovers a hidden danger for cancer survivors: common respiratory infections like the flu or COVID-19 can awaken dormant cancer cells, turning a manageable condition into a life-threatening resurgence. For patients, it emphasizes the importance of infection prevention (e.g., vaccinations, masking during outbreaks) and monitoring for metastatic progression post-infection. For clinicians, it suggests that IL-6 inhibitors (already used for severe COVID-19) might also protect against infection-triggered cancer relapse. Beyond breast cancer, the findings could apply to other cancers with dormant metastases, reshaping how we view the interplay between infections and cancer progression. For researchers, the study opens new avenues to explore how inflammation and immunity influence cancer dormancy, a critical frontier in metastasis biology.

Engineered bacteria launch and control an oncolytic virus

Singer et al. Nat. Biomed. Eng (2025). https://doi.org/10.1038/s41551-025-01476-8

The paper in one sentence

Researchers engineered Salmonella bacteria to act as a living "capsid" that delivers, launches, and controls an oncolytic virus inside tumors, overcoming major hurdles in cancer virotherapy.

Summary

This study introduces a novel cooperative microbial therapy platform called CAPPSID (Coordinated Activity of Prokaryote and Picornavirus for Safe Intracellular Delivery). The system uses engineered Salmonella typhimurium bacteria, which naturally target tumors, to deliver the RNA genome of an oncolytic virus (Senecavirus A) directly into cancer cells. Once inside the tumor cell, the bacteria sense their intracellular environment and activate a genetic circuit that causes them to lyse, releasing the viral genome. This "bacterial capsid" approach shields the virus from pre-existing neutralizing antibodies in the bloodstream, allowing it to reach the tumor even in immune hosts. The launched virus then replicates, lyses the initial cancer cell, and spreads to neighboring cells, effectively destroying the tumor. Furthermore, the team engineered a second layer of control by modifying the virus to require a bacterially delivered protease (TEV protease) for its maturation. This ensures the virus can only complete its life cycle and spread in the immediate vicinity of the bacteria, adding a crucial safety switch to prevent uncontrolled systemic viral infection.

Personal highlights

Bacteria as a dynamic, synthetic capsid: instead of a traditional protein shell, the team uses entire engineered Salmonella bacteria to protect and deliver the viral RNA genome. This "living capsid" leverages the bacteria's natural ability to target and invade tumors, providing a sophisticated delivery vehicle that is also tunable and responsive to its environment.
Cloaking from the immune system: by packaging the viral instructions (RNA) inside bacteria, the system effectively hides the therapy from circulating antiviral antibodies. This allows for successful systemic (IV) delivery and tumor targeting even in vaccinated, immunocompetent hosts—a significant breakthrough for overcoming a major clinical limitation of oncolytic viruses.
Programmed intracellular launch sequence: the bacteria are engineered with a precise genetic circuit that only activates inside a host cell. It uses natural Salmonella promoters to first produce the viral RNA and then express lytic proteins to rupture both the bacterial cell and its surrounding vacuole, ensuring efficient release of the viral payload into the host cell's cytoplasm where it can replicate.
Trans-regulatory control of viral spread: the study goes beyond simple delivery by engineering a fundamental dependency between the virus and the bacteria. By replacing a native viral protease cleavage site with one for an orthogonal bacterial protease (TEVp), they created a virus that can only produce infectious particles and spread if the bacterium is present to provide the essential maturation enzyme.
Mutation-resistant safety engineering: the team anticipated and engineered a solution to a key challenge: the high mutation rate of RNA viruses. When the engineered virus evolved a single-point mutation to escape its dependency on TEVp, they redesigned the cleavage site to require two simultaneous mutations for escape, dramatically increasing the genetic barrier and safety profile of the controlled virus.

Why should we care?

CAPPSID represents a paradigm shift in microbial cancer therapy by moving from single-agent approaches to designing cooperative, multi-kingdom consortia. For oncologists and patients, it offers a promising strategy to make potent oncolytic virus therapies work even in patients whose immune systems would normally neutralize them. For bioengineers and synthetic biologists, it is a masterclass in programming complex interactions between vastly different biological systems (bacteria and viruses) to achieve a unified therapeutic goal with built-in safety controls. It demonstrates that the future of advanced therapeutics may not lie in a single magic bullet, but in rationally designed teams of engineered organisms working together inside the body.

Fig. 1 from Singer et al. Overview of CAPSID system

Axonal injury is a targetable driver of glioblastoma progression

Clements et al. Nature (2025). https://doi.org/10.1038/s41586-025-09411-2

The paper in one sentence

This study identifies that early glioblastoma cells cause physical damage to axons in the brain's white matter, triggering a self-destruction process called Wallerian degeneration, which in turn fuels tumour inflammation, progression, and neurological decline—a vicious cycle that can be broken by inhibiting the key enzyme SARM1.

Summary

Glioblastoma (GBM) is the most aggressive and incurable primary brain cancer. While advanced tumours are well-studied, their earlier, more treatable stages are not. This research reveals that early GBM cells preferentially grow in the brain's white matter, where they physically compress and injure axons. This injury activates Wallerian degeneration (WD), an active program of axonal death mediated by the SARM1 protein. This process doesn't just destroy neurons; it creates a pro-inflammatory environment that accelerates tumour growth and progression. Crucially, the study shows that genetically inactivating Sarm1 preserves axons, suppresses this tumour-promoting neuroinflammation, and results in less aggressive, more diffuse tumours. This significantly extends survival and, importantly, preserves motor function in mouse models. The findings position SARM1 inhibition as a promising therapeutic strategy to intercept GBM progression by targeting the injury microenvironment.

Personal highlights

Early tumour cells home to and injure white matter axons: using mouse models and spatial transcriptomics, the authors show that the earliest GBM cells have a strong tropism for white matter tracts, where they cause direct physical compression and injury to axons, long before widespread tumour formation.
Wallerian degeneration is the mechanistic link between injury and progression: the research definitively shows that the axonal damage isn't passive; it triggers the active, SARM1-mediated program of Wallerian degeneration. This process is the key driver that transforms a localized injury into a tumour-promoting inflammatory response.
Experimental axon transection accelerates tumour growth via SARM1: surgically cutting axons in tumour-bearing mice accelerates tumour proliferation and neuroinflammation in wild-type mice, but not in Sarm1 knockouts, proving the causative role of this pathway.
Sarm1 deletion locks tumours in a less aggressive state: inhibiting this injury pathway doesn't just slow growth; it fundamentally changes the tumour's identity. Sarm1-deficient tumours are more diffuse, less vascularized, display a more neurodevelopmental-like cell state, and are less immunosuppressive compared to the aggressive, mesenchymal-like tumours that form in a WT background.
Dual benefit: suppressing progression and preserving function: Targeting SARM1 offers a two-for-one therapeutic benefit: it suppresses tumour advancement and protects neurological function by preventing axonal loss, directly addressing the debilitating cognitive and motor symptoms that plague GBM patients.

Why should we care?

This work reframes our understanding of how brain tumours progress by highlighting axonal injury as a central, targetable driver of the disease. For patients, it opens a promising new therapeutic avenue: drugs that inhibit SARM1 (already in development for neurodegenerative diseases) could potentially be used to "intercept" glioblastoma, locking it in a less aggressive, more manageable state and preserving quality of life by protecting brain function. For researchers and clinicians, it provides a compelling mechanistic link between the tumour's physical presence and the hostile microenvironment that makes it so resilient, suggesting that protecting the brain's architecture could be as important as directly attacking the cancer cells themselves

Other papers that peeked my interest and were added to the purgatory of my “to read” pile

Thanks for reading.

Cheers,

Seb.

Sebcentrism

Discussion about this post