Weekly reads 8/12/25
Why mutations, cells, and signals behave differently in space
This week’s papers collectively emphasize that cancer and tissue biology cannot be understood by mutations or cell states alone, context, spatial organization, and non-canonical molecular layers fundamentally shape outcomes. Lourenço et al. show that even canonical colorectal cancer drivers like APC can be actively eliminated unless they arise in a genetically primed environment, reframing tumor initiation as a context-dependent evolutionary process. Migliozzi et al. reveal that spatial architecture itself restrains or unleashes cancer cell plasticity, with homotypic clustering stabilizing glioblastoma identities while dispersion fuels malignant adaptability and poor prognosis. Salas et al. challenge the assumption that unassigned RNA in spatial transcriptomics is mere noise, demonstrating that extrasomatic RNA carries structured biological information about cellular morphology, protrusions, and contacts. Finally, Zhang et al. introduce RamanOmics, a multimodal framework that unifies gene expression, spatial context, and label-free biochemical fingerprints to uncover conserved and tissue-specific programs of cellular senescence in aging and repair.
Preprints/articles that I managed to read this week
Decay of driver mutations shapes the landscape of intestinal transformation
Lourenço et al. Nature (2025). https://doi.org/10.1038/s41586-025-09762-w
The paper in one sentence
The order and context in which cancer-driving mutations occur, not just their presence, determine whether they persist or are eliminated, reshaping our understanding of how colorectal tumors begin and evolve.
Summary
This study challenges the traditional stepwise model of colorectal cancer (CRC) by showing that mutations in genes like APC and CTNNB1 can be negatively selected and lost unless they occur in a “primed” cellular environment, created by pre-existing mutations in genes such as KRAS, TP53, or FBXW7. Using mouse models and human CRC data, the authors demonstrate that priming changes the selection landscape, allowing otherwise deleterious mutations to take hold and drive tumor formation. The findings reveal that mutation order, tissue context, and negative selection against strong drivers play critical roles in shaping tumor evolution.
Personal highlights
Priming reshapes mutation selection: pre-existing driver mutations (e.g., KRAS, TP53) alter the intestinal microenvironment, allowing otherwise negatively selected APC and CTNNB1 mutations to persist and drive tumors.
Negative selection against strong drivers: in unprimed tissue, potent APC and CTNNB1 mutations are often eliminated, revealed by reversing the order of mutagenesis and priming in rescue experiments.
Mutation order dictates tumor evolution: the sequence of genetic events (e.g., KRAS before APC) determines whether clones survive, influencing tumor multiplicity, mutation spectrum, and clinical progression.
Human CRC mirrors priming effects: in human tumors, APC truncation patterns differ depending on KRAS status, suggesting that priming shapes mutation selection in patients, with implications for tumor subtyping and risk stratification.
Context-dependent transformation landscape: the study introduces a dynamic, context-sensitive model of tumor initiation where the fitness effect of a mutation depends on the genetic and epigenetic background of the tissue.
Why should we care?
This work suggests that cancer risk may depend not just on which mutations arise, but when and in what order they occur. In addition, it highlights the importance of studying mutation selection in realistic tissue environments, not just in isolated cells, while reminding us that cancer is as much about context as it is about mutations.
Restraint of cancer cell plasticity by spatial homotypic clustering
Migliozzi et al., Cancer Cell (2025). DOI: 10.1016/j.ccell.2025.08.009
The paper in one sentence
Using single-cell spatial transcriptomics, this study reveals that glioblastoma cells organized in homotypic clusters maintain stable identities, while dispersed cells exhibit greater plasticity and are linked to worse patient outcomes.
Summary
This work applies high-resolution spatial transcriptomics and proteomics to glioblastoma, uncovering that tumor cells organize into either clustered or dispersed spatial patterns. Clustered cells, particularly glycolytic-plurimetabolic (GPM) and proliferative-progenitor (PPR) subtypes, maintain stable phenotypes through homotypic adhesion mechanisms mediated by CD44 and NOTCH signaling, respectively. In contrast, dispersed cells lose their original identity, exhibit increased plasticity, and acquire microenvironment-driven traits. Critically, enrichment of dispersed GPM cells is uniquely associated with shorter survival and higher relapse risk, highlighting the clinical relevance of spatial architecture in cancer progression.
Personal highlights
Single-cell spatial mapping of GBM subtypes and microenvironment: uses CosMx spatial transcriptomics at single-cell resolution across 16 glioblastoma samples and 2.8 million cells to classify malignant cells into four functional states and map their spatial organization relative to immune and neuroglial cells.
Homotypic clustering locks cell identity, dispersion fuels plasticity: demonstrates that GBM cells in homotypic clusters maintain transcriptional stability and phenotype, while dispersed cells lose subtype-specific programs, acquire alternative states, and exhibit increased phenotypic randomness.
Mechanistic validation of adhesion-dependent clustering: identifies and experimentally validates that CD44-mediated collagen/fibronectin signaling maintains GPM clustering, while NOTCH ligands sustain PPR clustering, revealing adhesion pathways as key inhibitors of plasticity.
Spatial patterns dictate clinical outcome: shows that the proportion of dispersed GPM cells—but not other dispersed subtypes—uniquely predicts worse overall survival and progression-free survival, linking spatial architecture directly to patient prognosis.
Generalizability across cancer types: extends findings to circulating tumor cells in breast cancer, where clustered circulating cells show higher transcriptional homogeneity than single cells, suggesting homotypic clustering as a conserved mechanism to restrain plasticity in cancer.
Exploration of RNA outside segmented cells in spatial transcriptomics reveals extrasomatic RNA organization
Salas et al., bioRxiv 2025. doi:10.64898/2025.12.07.692889
The paper in one sentence
This study systematically analyzes unassigned RNA in image-based spatial transcriptomics data, revealing that a large fraction represents biologically meaningful extrasomatic transcripts rather than technical noise.
Summary
The authors investigate RNA molecules detected in spatial transcriptomics data that are not assigned to segmented cells, so-called unassigned RNA (uRNA). Across 14 public datasets, they show that a significant portion of uRNA cannot be explained by technical artifacts like noise or diffusion. Instead, uRNA is enriched near cells with complex morphologies (e.g., neurons, glia) and reflects RNA localized in cellular protrusions and extrasomatic compartments. The work introduces troutpy, a Python package for uRNA analysis, and demonstrates how uRNA can reveal subcellular RNA localization, cellular architecture, and cell–cell contacts missed by standard segmentation.
Personal highlights
Systematic dissection of uRNA origins: combines segmentation-based and segmentation-free strategies to quantify contributions from missegmentation, technical noise, and diffusion, showing that over half of uRNA arises from segmentation errors, while ~30% has non-technical biological origins.
Cross-platform consistency of biological uRNA: demonstrates that non-technical uRNA patterns are reproducible across different imaging platforms (e.g., Xenium and CosMx), with genes linked to synaptic plasticity and cellular projections consistently enriched in the uRNA pool.
Spatially structured extrasomatic RNA organization: applies DRVI and NMF to uRNA bins, revealing spatially coherent gene programs that align with anatomical structures (e.g., corpus callosum, hippocampal layers) and reflect subcellular compartmentalization beyond segmented cell bodies.
Inference of protrusion-associated RNA and cell–cell contacts: introduces probabilistic source–target scoring to map RNA to cellular protrusions and uses uRNA displacement to infer directional cell–cell interactions, uncovering contacts (e.g., astrocyte-mediated) invisible to centroid-based approaches.
troutpy: an open-source toolbox for uRNA exploration: provides a flexible Python package for quantifying, characterizing, and interpreting unassigned RNA in spatial transcriptomics, enabling systematic study of extrasomatic RNA across tissues and technologies.
Why should we care?
This work challenges the current view that unassigned RNA is mere technical noise and reframes it as a window into subcellular biology and tissue architecture. For researchers studying complex cell types, uRNA offers clues about RNA trafficking, localized translation, and cellular connectivity that are missed by cell-centric segmentation. For method developers, it highlights the limits of current segmentation algorithms and calls for integrative models that account for extrasomatic RNA.
RamanOmics Decodes Spatial Vibrational-Molecular Architecture and Rewiring in Aging and Repair
Zhang et al. bioRxiv (2025). https://doi.org/10.64898/2025.12.04.692337
The paper in one sentence
Researchers developed a multimodal platform called RamanOmics that integrates single-nucleus RNA sequencing, spatial transcriptomics, and label-free Raman imaging to create a unified biochemical and molecular map of cellular senescence in aging and repair, revealing tissue-specific programs and a conserved fatty-acid signature.
Summary
This study introduces RamanOmics, a pioneering framework that fuses three powerful technologies: single-nucleus RNA sequencing (snRNA-seq) for molecular profiling, spatial transcriptomics (STARmap-ISS) for cellular location, and hyperspectral Raman imaging for label-free biochemical fingerprinting. By applying this integrated approach to aging mouse lung and skin, the team created the first spatially resolved maps that directly link a cell’s gene expression to its underlying biochemical state (e.g., lipids, proteins, metabolites). They uncovered that senescent cells (marked by p21) are not just transcriptionally distinct but also have unique vibrational “signatures,” including a conserved increase in branched-chain fatty acids (peaks at ~1134 cm⁻¹). The study further shows how these signatures shift from a reparative state in young tissues to a dysfunctional, pro-fibrotic state in old tissues and are reactivated during wound healing. Finally, they used machine learning to create a “multimodal barcode” that combines both gene and biochemical features, significantly improving the accuracy of identifying senescent cells directly in intact tissue.
Personal highlights
Multimodal integration of vibrational chemistry with genomics: ramanOmics directly fuses label-free Raman imaging, which captures the vibrational “fingerprint” of biomolecules like lipids and proteins, with single-cell and spatial transcriptomics, bridging the long-standing gap between biochemical composition and gene expression programs in intact tissues.
Discovery of a conserved biochemical signature of senescence: across different tissues (lung and skin), *p21+* senescent cells consistently show elevated Raman peaks (~1134-1135 cm⁻¹) corresponding to branched-chain fatty acids (BCFAs), revealing a previously invisible, core biochemical hallmark of senescence beyond traditional genetic markers.
Tissue-specific senescent programs revealed: the study elegantly shows that while the BCFA signature is shared, the functional state of senescent cells is tissue-defined: lung senescence is dominated by ECM remodeling and TGF-β signaling (e.g., Serpine1), while skin senescence is characterized by keratinization and barrier maintenance (e.g., Krt10, Lor).
In situ validation of signatures in wound repair: the newly discovered molecular (Dmkn, Sbsn, Sfn) and biochemical (BCFA) signatures were functionally validated in a wound-healing model, confirming they are dynamically induced during tissue regeneration, moving beyond steady-state aging to a relevant repair context.
Machine learning-derived “multimodal barcode” for precise identification: by integrating top Raman spectral peaks with differentially expressed genes, the team created a quantitative barcode that significantly outperforms transcriptomic data alone in classifying senescent cells, paving the way for non-destructive, in situ diagnostics.
Other papers that peeked my interest and were added to the purgatory of my “to read” pile
Mapping spatial gradients in spatial transcriptomics data with score matching
Inferring cell differentiation maps from lineage tracing data
RamanOmics Decodes Spatial Vibrational-Molecular Architecture and Rewiring in Aging and Repair
Single-cell resolution spatial analysis of antigen-presenting cancer-associated fibroblast niches
Thanks for reading.
Cheers,
Seb.