Weekly reads 7/7/25

From sparse signals to spatial insights: reconstructing tumor evolution, resistance, and microenvironments

This week’s papers brought serious power to the challenge of seeing more with less—making sparse data richer, turning standard assays smarter, and uncovering hidden regulatory logic. We saw copy number variations inferred directly from spatial transcriptomics, cancer resistance mapped through gene co-context shifts, and histology repurposed as a high-resolution cell type predictor.

The spatial/transcriptomic/cellular landscapes continue to get more layered, more mechanistic—and more actionable.

Preprints/articles that i managed to read this week

In Situ Inference of Copy Number Variations in Image-Based Spatial Transcriptomics

Jensen et al. (2025). bioRxiv. doi: 10.1101/2025.07.02.662761

The paper in one sentence

A novel method enables spatially resolved copy number variation (CNV) inference from image-based spatial transcriptomics (iST) data, overcoming technical limitations to map tumor subclones and their microenvironment at single-cell resolution.

Summary

This study introduces a computational approach to infer CNVs directly from iST data, leveraging transcriptomic-neighborhood smoothing to enhance signal detection in sparse datasets. By systematically evaluating the impact of gene panel size and detection efficiency, the authors demonstrate the feasibility of CNV inference in iST platforms like CosMx and Xenium, enabling spatial mapping of tumor heterogeneity and microenvironment interactions. The method bridges a critical gap between genomic alterations and their spatial context, offering new avenues for studying cancer evolution in situ.

Personal highlights

• Spatially aware CNV inference: Adapts inferCNV to iST data by incorporating transcriptomic-neighborhood smoothing, averaging expression profiles of neighboring cells to reduce noise and improve CNV signal detection—critical for sparse spatial datasets.

• Technical benchmarking of iST constraints: Rigorously evaluates how gene panel size (500–20k genes) and detection efficiency (1–100% counts/cell) impact CNV prediction, revealing thresholds for reliable subclone identification (e.g., >2k genes, >1k counts/cell).

• Platform-agnostic applicability: Validated across two iST technologies (CosMx WTx and Xenium Prime), demonstrating robustness despite platform-specific differences in resolution and gene coverage.

Why should we care?

This work transforms iST from a purely descriptive tool into a platform for mechanistic insights into cancer evolution. By enabling CNV inference in situ, it bridges the gap between genomic alterations and their spatial context—answering not just "where" mutations occur, but "how" they shape tumor heterogeneity and microenvironment interactions. For computational biologists, the method offers a reproducible framework to adapt CNV tools for sparse spatial data; for cancer researchers, it provides a new lens to study clonal dynamics and therapy resistance.

Tracing the Shared Foundations of Gene Expression and Chromatin Structure

Liang et al. (2025). bioRxiv. doi: 10.1101/2025.03.31.646349

The paper in one sentence

A "bag-of-genes" framework integrates single-cell foundation models and chromatin architecture to reveal how topologically associating domains (TADs) systematically enhance transcriptional coordination across development, aging, and disease.

Summary

This study introduces two key innovations: (1) a species-wide TAD Map, which represents chromatin domains as gene sets (inspired by NLP’s bag-of-words) to enable cross-cell-type analysis without new Hi-C data, and (2) contextual transcriptional similarity (CTS), a gene-embedding metric derived from a 33M-cell foundation model (scGPT) that captures functional relationships beyond co-expression. By combining these tools, the authors demonstrate that TADs multiplicatively boost gene-gene regulatory coordination by ~20%—independent of genomic distance—suggesting synergy with transcriptional condensates. The framework reveals dynamic TAD-mediated organization in development (stronger in early stages), aging (declining plasticity), and cancer (chemotherapy-induced divergence between tumor/normal cells).

Personal highlights

• Bag-of-genes abstraction for chromatin structure: Represents TADs as simple gene sets, bypassing boundary ambiguity and enabling species-wide analysis. Validated against CTCF binding, eQTLs, and perturbation data, the TAD Map reveals conserved regulatory neighborhoods across 7 human and 4 mouse cell types.

• Contextual similarity > co-expression: Leveraging scGPT embeddings, CTS detects functional gene relationships (e.g., olfactory receptors) invisible to traditional co-expression metrics—showing 20.5% higher coordination within TADs, even at long genomic distances.

• Condensate-TAD synergy hypothesis: Multiplicative CTS enhancement and orientation-dependent transcriptional read-through suggest TADs create phase-separated hubs where condensates amplify co-regulation—supported by Perturb-seq hits (e.g., TAF12, INTS complexes).

• Aging and cancer as TAD dysregulation: CTS declines systematically with age (even within cell types) but spikes in chemotherapy-resistant tumors, linking TAD plasticity to disease progression and treatment resistance.

• TAD signatures for single-cell robustness: A probabilistic model of TAD activation improves clustering accuracy (e.g., distinguishing luminal progenitors) and predicts drug response from transcriptional skew, outperforming gene-level metrics.

Why should we care?

This work redefines TADs as probabilistic amplifiers of gene regulation rather than static compartments, bridging the gap between chromatin structure and transcriptional logic. By integrating foundation models with genomics, it answers the "how" of spatial gene coordination—revealing that TADs and condensates likely collaborate to create regulatory hotspots. For biologists, the framework generates testable hypotheses about developmental plasticity and cancer resilience; for computationalists, it offers a noise-resistant lens (TAD signatures) to dissect single-cell data. As a tool, it democratizes 3D genome analysis by eliminating the need for cell-specific Hi-C, while its aging/cancer findings hint at chromatin organization as a tunable axis for therapeutic intervention

Decoding Spatial Transcriptomics: A Methodological Deep Dive into KRAS and p53 Dynamics in Pancreatic Cancer

Reyes et al. (2025). Oncogenic and tumor-suppressive forces converge on a progenitor-orchestrated niche to shape early tumorigenesis. bioRxiv. DOI: 10.1101/2025.06.10.656791

The paper in one sentence

This study reveals how a rare progenitor-like cell state in pancreatic cancer becomes the battleground for oncogenic KRAS signaling and tumor-suppressive pathways (p53, CDKN2A, SMAD4), orchestrating a malignant-permissive niche through bidirectional communication with the microenvironment.

Summary

Using spatially resolved single-cell and lineage-tracing approaches in mouse models of pancreatic ductal adenocarcinoma (PDAC), the authors identify a transient, progenitor-like cell population that emerges during early tumorigenesis. These cells uniquely co-activate oncogenic (KRAS-driven plasticity) and tumor-suppressive programs (p53, senescence), remodeling their microenvironment into an immune-privileged, fibrotic niche resembling advanced PDAC. The study shows that KRAS inhibition collapses this progenitor niche, while p53 loss accelerates its expansion—highlighting a critical window where intercepting progenitor cell dynamics could prevent malignant progression.

Personal highlights

• Progenitor cells as convergence points for cancer drivers and suppressors: a rare KRAS-mutant progenitor-like subpopulation simultaneously engages oncogenic (EMT, glycolysis) and tumor-suppressive (p53, senescence) programs, acting as a pivot between benign and malignant states.

• Spatial mapping of niche assembly: High-resolution spatial transcriptomics reveals how progenitor-like cells progressively dismantle epithelial architecture and recruit immunosuppressive macrophages/activated fibroblasts, mirroring invasive PDAC.

• Oncogene addiction meets tumor suppression: progenitor-like cells are exquisitely dependent on persistent KRAS signaling; acute KRAS inhibition collapses their niche, while p53 loss licenses their expansion and mesenchymal transition.

• Bidirectional niche signaling: Progenitor cells and their microenvironment engage reciprocal TGFβ, IL-18, and ECM-mediated communication circuits, creating a self-reinforcing "cancer-like" ecosystem.

• p53 as a plasticity gatekeeper: Beyond genomic stability, p53 restrains progenitor cell expansion and niche immune evasion, mimicking wound-resolution mechanisms.

• KRAS inhibition resets the niche: Targeting KRAS not only depletes progenitor cells but also reverses immune suppression and fibroblast activation—a dual therapeutic effect.

Why should we care?

This work shifts the paradigm of early cancer progression from a linear genetic model to a cellular ecosystem battle, where the fate of a single plastic cell state determines malignancy. By pinpointing the progenitor niche as the nexus of oncogenic signaling, immune evasion, and stromal reprogramming, it offers:

• New interception strategies: Targeting progenitor cell plasticity or niche communication (e.g., TGFβ, IL-18) could prevent malignant transformation before irreversible genomic instability occurs.

• Context for KRAS inhibitor effects: Explains why KRAS inhibitors remodel the tumor microenvironment—they collapse the progenitor hub that sustains it.

• A unifying principle for tumor suppression: p53’s role in restraining plasticity (not just DNA repair) may generalize across cancers.

• Tools for early detection: Progenitor-like states and their niche signatures could serve as biomarkers for high-risk premalignant lesions.

For researchers, this study provides a spatiotemporal playbook of early tumorigenesis; for clinicians, it highlights the potential of niche-directed therapies to intercept cancer at its origin.

Mast Cell-Derived Histamine Drives Cancer Invasion Along Nerves—H1 Antihistamines May Offer a Solution

Srivastava et al. (2025). Epithelial tumor cells utilize mast cell-derived histamine to regulate perineural invasion. bioRxiv. doi: 10.1101/2025.06.23.661147

The paper in one sentence

This study reveals that mast cells release histamine to help epithelial cancers invade nerves, and blocking the H1 histamine receptor with common antihistamines could suppress this deadly process and improve patient survival.

Summary

Perineural invasion (PNI)—when cancer cells spread along nerves—is linked to poor outcomes in cancers like squamous cell carcinoma (SCC), head and neck cancer, and cholangiocarcinoma. Using single-cell RNA sequencing, organoid models, and patient data, the authors discovered that tumor cells recruit mast cells via KITLG-KIT signaling, triggering histamine release. Histamine then binds to H1 receptors on cancer cells, activating P38 and MMP pathways to degrade the nerve sheath’s collagen, enabling invasion. Crucially, H1-antihistamines (e.g., cetirizine, desloratadine) blocked this process in lab models and were associated with better survival in patients with PNI-positive cancers.

Personal highlights

• Mast cells as unexpected accomplices: tumor cells co-opt mast cells via KITLG (stem cell factor) to release histamine, creating a permissive niche for nerve invasion—a previously overlooked axis in cancer progression.

• H1 receptor as the linchpin: HRH1 (histamine H1 receptor) is upregulated in PNI-positive tumors, and its activation drives MMP secretion, while H1-antihistamines (but not H2 blockers) reverse this effect.

• Collagen destruction via MMPs: histamine-H1 signaling activates P38, boosting MMP2/9 production to degrade type IV collagen in the nerve sheath—a "molecular crowbar" for cancer invasion.

• STAT1 as a transcriptional switch: in tumors without KITLG amplification, STAT1 drives KITLG expression, linking inflammation to mast cell recruitment.

• Clinical gold in old drugs: retrospective analysis of >15,000 patients revealed H1-antihistamine users had better survival and immunotherapy responses in PNI-prone cancers.

• Broad relevance: the KITLG-histamine-MMP axis is conserved across SCC, cholangiocarcinoma, and head/neck cancers, suggesting a unified therapeutic strategy.

Why should we care?

This work transforms our understanding of PNI from a passive "tracking" of nerves to an active hijacking of allergy pathways by cancers. For patients, it offers hope: cheap, well-tolerated antihistamines like cetirizine might curb nerve invasion and improve outcomes. For researchers, it uncovers mast cells as key players in the tumor microenvironment and provides a blueprint for studying PNI with organoids. Clinicians gain a rationale to prioritize H1-antihistamines over H2 blockers (e.g., famotidine) in PNI-positive cancers. Finally, it highlights how "big data" (EHR analysis) can validate mechanistic discoveries—bridging lab insights to real-world impact.

Blocking Osteoprotegerin Reprograms Cancer-Associated Fibroblasts to Boost Immune Infiltration

Wang et al. (2025). Blocking osteoprotegerin reprograms cancer-associated fibroblasts to promote immune infiltration into the tumor microenvironment. bioRxiv. doi: 10.1101/2025.07.04.663190

The paper in one sentence

This study reveals that osteoprotegerin (OPG), secreted by immunosuppressive cancer-associated fibroblasts (CAFs), blocks T-cell function—and targeting OPG reprograms the tumor microenvironment to unleash anti-tumor immunity in "cold" cancers.

Summary

In stroma-rich cancers like pancreatic and breast cancer, a subset of inflammatory CAFs (iCAFs) secrete OPG, a decoy receptor that neutralizes TRAIL and RANKL—key molecules for T-cell cytotoxicity. Using single-cell RNA sequencing and murine models, the authors show that OPG directly impairs CD8+ T-cell killing. Blocking OPG (genetically or with antibodies) reshaped the tumor microenvironment: immunosuppressive iCAFs decreased, while interferon-responsive CAFs increased, leading to robust T-cell infiltration, enhanced cytokine release, and tumor regression. Crucially, OPG blockade did not disrupt bone homeostasis, highlighting its potential as a stromal-specific immune checkpoint target.

Personal highlights

• OPG as a stromal immune checkpoint: iCAFs exploit OPG to paralyze T cells by intercepting TRAIL/RANKL signals—a novel mechanism of fibroblast-mediated immune evasion.

• Reprogramming the "cold" TME: OPG blockade transforms the tumor stroma, replacing immunosuppressive iCAFs with interferon-licensed CAFs that recruit and activate T cells.

• Dual-action T-cell revival: anti-OPG therapy restores both CD8+ cytotoxicity (via TRAIL unblocking) and CD4+ effector function (via RANKL signaling), doubling the immune attack.

• Beyond bone biology: unlike systemic RANKL inhibitors (e.g., denosumab), OPG blockade spared bone remodeling in mice, suggesting a cancer-specific therapeutic window.

• Conserved across cancers: OPG+ iCAFs were identified in human esophageal, breast, and pancreatic tumors, hinting at broad applicability.

• Stromal-immune crosstalk decoded: single-cell analysis revealed IFN-responsive CAFs as key orchestrators of post-OPG blockade immune infiltration.

Why should we care?

This work uncovers OPG as a linchpin of stromal immunosuppression—a long-overlooked barrier to immunotherapy success in pancreatic, breast, and other stroma-dominated cancers. For patients, it offers hope: targeting OPG could "heat up" resistant tumors without the bone toxicity of global RANKL inhibition. For researchers, it redefines CAFs as dynamic immune modulators, not just ECM architects. Clinicians gain a rationale to explore OPG blockade in "cold" tumors, while drug developers may repurpose existing OPG inhibitors (e.g., for osteoporosis) for oncology

Gene Context Drift: A New Lens on Cancer Treatment Resistance

Jassim et al., Cancer Cell (2025). https://doi.org/10.1016/j.ccell.2025.06.005

The paper in one sentence

RECODR, a computational pipeline leveraging gene co-expression context drift, identifies hidden drivers of cancer treatment resistance and predicts effective combination therapies.

Summary

The study introduces RECODR (Resistance through Context Drift), a graph-embedding tool that analyzes changes in gene co-expression networks during cancer treatment. By focusing on shifts in gene interactions—rather than just expression levels—RECODR uncovered resistance mechanisms in aggressive brain tumors (choroid plexus carcinoma, medulloblastoma) and triple-negative breast cancer. The approach successfully predicted drug targets (e.g., ATM, PARP1) and designed combination therapies that mitigated resistance in preclinical models, offering a roadmap for clinical translation.

Personal highlights

1. Gene co-context over expression levels: RECODR shifts the paradigm from differential gene expression to gene context drift—tracking how genes reorganize their functional partnerships during treatment, revealing hidden resistance drivers like ATM and PARP1.

2. "Recalled" embryonic programs fuel resistance: tumors reawaken fetal gene networks (e.g., cell cycle/DNA repair pathways) to evade therapy, a vulnerability RECODR pinpoints for targeted intervention.

3. Immune mimicry as a resistance shield: combination therapy-resistant tumors expanded myeloid-like gene modules—unmasked by RECODR—leading to dasatinib’s success in blocking this adaptive niche.

4. Benchmark-beating precision: outperformed WGCNA, diffcoexp, and scDRUG by prioritizing targets (e.g., PARP1) invisible to conventional methods due to its context-aware design.

5. Clinical bridge: validated in human medulloblastoma and breast cancer, RECODR identified high-risk patients and candidate drugs (e.g., ribociclib, regorafenib) based on pre-treatment drift patterns.

Why should we care?

RECODR tackles the black box of cancer treatment failure by decoding why therapies stop working—not just where genes are active. For oncologists, it offers a tool to preempt resistance by designing context-informed combination therapies. For patients, it signals hope for hard-to-treat cancers (e.g., TP53-mutant CPC) by repurposing existing drugs like dasatinib. For researchers, it provides a framework to dissect transcriptional plasticity in real-time, transforming single-cell data into actionable insights. Beyond cancer, its NLP-inspired approach could reshape how we study dynamic gene networks in other complex diseases.

STHELAR: Bridging Spatial Transcriptomics and Histology for Pan-Cancer Cell Annotation

Giraud-Sauveur et al. bioRxiv (2025). doi: 10.1101/2025.07.11.664123

The paper in one sentence

STHELAR integrates spatial transcriptomics (ST) with H&E histology images to create a multi-tissue dataset of 11 million cells, enabling AI models to predict cell types from routine pathology slides.

Summary

The study introduces STHELAR, a pipeline combining 10x Genomics Xenium-based ST data with paired H&E whole-slide images across 16 human tissues (22 cancerous, 9 healthy). Using Tangram and Leiden clustering, cells are annotated into 10 standardized categories (e.g., epithelial, immune subsets). The dataset includes 587,555 histology patches with segmentation masks, validated by pathologists and used to fine-tune CellViT, a vision transformer model, demonstrating that ST-derived annotations can translate to H&E-based predictions.

Personal highlights

• Multi-modal cell atlas: STHELAR links subcellular-resolution ST (Xenium) with H&E histology for 11 million cells across 31 slides, creating the largest spatially resolved pan-cancer cell-type resource to date.

• Annotation robustness: combines Tangram alignment with slide-specific Leiden clustering and differential expression, achieving 80–83% concordance between nuclear and whole-cell RNA-based labels.

• Pathology-ready AI training: delivers 500K+ curated H&E patches with segmentation/classification masks, optimized for fine-tuning CellViT to predict cell types directly from histology (PQ scores up to 0.54 for epithelial cells).

• Practical validation: shows that ST-derived labels can train models to classify morphologically similar immune subsets (e.g., T vs. B cells) in H&E, despite their visual ambiguity.

Why should we care?

STHELAR tackles a major bottleneck in digital pathology: the inability to extract rich cell-type data from ubiquitous H&E stains without costly molecular assays. By "distilling" spatial transcriptomics into histology-compatible annotations, it empowers researchers to infer tumor microenvironment composition from routine slides—democratizing precision oncology. For computational biologists, it’s a gold-standard benchmark; for clinicians, a step toward AI-powered pathology that deciphers tissue context beyond human vision.

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

• Identification of proliferating neural progenitors in the adult human hippocampus

• Perturb-Multimodal: A platform for pooled genetic screens with imaging and sequencing in intact mammalian tissue

• In vivo mapping of mutagenesis sensitivity of human enhancers

Thanks for reading.

Cheers,

Seb.