Weekly reads 24/11/25

Hidden circuits: how tumors, neurons, microbes & AI reveal new biology

Dec 01, 2025

This week’s reads focuses on uncovering hidden layers of communication and control inside tissues: from mitochondria behaving as an intercellular currency to neurons remotely suppressing anti-tumor immunity. Terasaki et al. reveal that cancer cells run a mitochondrial “economy,” importing, refurbishing, and redistributing organelles to build an immunosuppressive ecosystem. Liu et al. push live-cell genomics forward with CRISPR PRO-LiveFISH, enabling multiplexed imaging of non-repetitive loci and direct observation of enhancer–promoter dynamics. Morad et al. provide rigorous spatial evidence that intracellular bacterial elements are a real, immunologically active component of brain tumors. Yao et al. introduce spacer, an interpretable AI framework that decodes how cells recruit or repel each other in situ across tissues and diseases. Zhang et al. map a long-range neuroimmune circuit where tumor-evoked pain signals rewire lymph nodes into an immune-suppressed state. And Deshpande et al. develop MIMYR, a generative model that reconstructs missing regions in spatial transcriptomics with realistic cell types and gene expression.

Preprints/articles that I managed to read this week

Mitochondria Redistribution Organizes the Immunosuppressive Tumor Ecosystem

Terasaki et al. bioRxiv (2025). https://doi.org/10.1101/2025.11.11.687895

The paper in one sentence

Cancer cells act as metabolic hubs, importing host mitochondria, fusing them with their own network to boost their biosynthetic capacity, and then redistributing these hybrid organelles to neighboring immune cells to drive immunosuppression and CD8+ T cell exhaustion.

Summary

The tumor microenvironment is metabolically hostile, yet cancer and immunosuppressive cells mysteriously maintain robust mitochondrial function. This study reveals a sophisticated strategy tumors use to resolve this paradox: they actively orchestrate a mitochondrial redistribution program. Using advanced mitochondrial reporter systems in mouse models and human data, the authors show that cancer cells import mitochondria from host immune cells, fuse them with their own, and then relay these “refurbished” hybrid mitochondria back out to specific neighbors in the tumor microenvironment. This transfer has divergent effects: it reprograms neutrophils, macrophages, and CD4+ T cells into highly immunosuppressive states, while simultaneously exhausting anti-tumor CD8+ T cells. For the cancer cell itself, the act of fusing incoming mitochondria triggers a structural reorganization of the metabolic enzyme P5CS, enhancing its own biosynthetic output. Disrupting this mitochondrial redistribution collapsed the immunosuppressive network and impaired tumor growth, revealing it as a central organizing principle of the tumor ecosystem.

Personal highlights

Cancer cells as active conductors of a mitochondrial economy: the study reframes cancer cells not as passive resource hoarders, but as active organizers that import, refurbish, and redistribute mitochondria to shape their microenvironment, directly fueling immunosuppressive allies.
Fusion-driven metabolic optimization within cancer cells: prior to redistribution, cancer cells fuse exogenous mitochondria with their endogenous network, a process that triggers a conformational change in the P5CS enzyme into a filamentous state, dramatically boosting the cancer cell’s own biosynthetic and metabolic capacity.
Repurposing of “incompetent” host mitochondria: the fusion mechanism allows cancer cells to incorporate and functionally benefit even from mitochondria lacking a key complex I subunit (from Ndufs4KO mice), demonstrating an extraordinary ability to salvage and refurbish suboptimal organelles.
Divergent immune reprogramming via organelle transfer: the same redistributed mitochondria have opposing, context-specific effects on immune cells: potentiating immunosuppressive functions in Tregs, macrophages, and neutrophils while inducing exhaustion and metabolic paralysis in CD8+ T cells.
Tunneling nanotubes as the primary redistribution highway: the majority of mitochondrial redistribution from cancer cells to immune partners occurs via direct intercellular connections (tunneling nanotubes), with extracellular vesicles playing a minor role, highlighting a targetable physical mechanism.

Why should we care?

This work uncovers a previously hidden layer of ecosystem-level control in cancer, where the battle is not just over nutrients but over the very powerhouses of the cell. It explains a key paradox of tumor metabolism and provides a unified mechanism for how cancer cells can be both metabolically robust and potently immunosuppressive. For cancer biologists and immunologists, it reveals a new targetable axis: disrupting mitochondrial redistribution could simultaneously starve the tumor, disarm its immunosuppressive guards, and rescue exhausted killer T cells.

CRISPR PRO-LiveFISH: Live-cell imaging reveals chromatin dynamics and enhancer interactions at multiple non-repetitive loci

Liu et al. Nat Biotechnol (2025). https://doi.org/10.1038/s41587-025-02887-3

The paper in one sentence

Researchers developed a new live-cell imaging technique, CRISPR PRO-LiveFISH, which uses an expanded genetic alphabet to create highly sensitive probes, enabling them to watch up to six unique, non-repetitive genomic locations at once and reveal how chromatin moves and how genes interact with their enhancers in real time.

Summary

This study introduces CRISPR PRO-LiveFISH (Pooled gRNAs with Orthogonal bases LiveFISH), a major advancement in live-cell genomics. The method overcomes a key limitation of previous CRISPR-based imaging techniques, which struggled to visualize unique, non-repetitive genomic regions without using large numbers of probes or signal amplification, which can be inefficient or produce background noise.

The core innovation is the integration of orthogonal unnatural base pairs (UBPs) from expanded genetic alphabet technology. This allows for the rational design and site-specific fluorescent labeling of sgRNA probes via in vitro transcription, producing high-quality probes that are brighter and more specific. This high sensitivity means that as few as 10 sgRNAs are needed to image a non-repetitive locus, a significant improvement over earlier methods that required hundreds.

The authors demonstrate the power of PRO-LiveFISH by:

Revealing a correlation between epigenetic states and chromatin dynamics, showing that more active chromatin (marked by H3K27ac) is less mobile.
Showing that certain enhancer-promoter interactions (like PCDHa-HSS1) are persistent over time, maintaining close proximity despite general chromatin motion.
Providing direct, live-cell evidence that the protein BRD4 is crucial for maintaining the 3D interaction between the MYC oncogene and its super-enhancer, and that inhibiting BRD4 disrupts this loop and increases mobility.

Personal highlights

Expanded genetic alphabet for superior probe design: PRO-LiveFISH leverages unnatural base pairs (X-Y) to enable site-specific, rational labeling of sgRNAs at an exposed stem-loop, resulting in a more than three-fold increase in signal-to-noise ratio compared to randomly labeled sgRNAs or fluorescent tracrRNA.
Highly sensitive imaging of non-repetitive loci with minimal probes: the method achieves efficient labeling of unique genomic regions using pools of as few as 10 sgRNAs, a dramatic reduction from the hundreds required by previous state-of-the-art techniques like CAS-LiveFISH (288 sgRNAs) or Oligo-LiveFISH (340 sgRNAs).
Multiplexed live-cell tracking of chromatin dynamics: the platform enables simultaneous, multi-color imaging of up to six distinct genomic loci in living cells, allowing for direct observation of their individual movements and relative spatial relationships over time.
Direct evidence for persistent enhancer-promoter loops: contrary to some models suggesting highly transient interactions, PRO-LiveFISH imaging revealed that the PCDHa gene and its enhancer (HSS-1) maintain close spatial proximity (>90% of the time within <0.4 µm) despite dynamic chromatin motion, suggesting a relatively stable looping configuration.

Microbial signals in primary and metastatic brain tumors

Morad et al. Nat Med (2025). https://doi.org/10.1038/s41591-025-03957-4

The paper in one sentence

This multi-institutional study provides robust, spatially-resolved evidence that intracellular bacterial elements are a detectable component of the brain tumor microenvironment in both gliomas and brain metastases, correlating with distinct antimicrobial immune responses and distant oral/gut microbiota.

Summary

This prospective study investigated the controversial presence of microorganisms within brain tumors. Using a rigorous, multi-method approach on 243 human tissue samples, the researchers found no cultivable bacteria but validated the presence of intracellular bacterial 16S rRNA and lipopolysaccharides (LPS) in a subset of gliomas and brain metastases. High-resolution spatial imaging revealed these bacterial signals inside tumor, immune, and stromal cells. Spatial analyses showed that tumor regions with high bacterial signal were enriched with specific antimicrobial and immunometabolic signatures, such as activation of the TLR9/NF-κB pathway and neutrophil recruitment. Furthermore, sequencing data indicated that the bacterial signals in tumors shared sequence overlap with the patients’ own oral and gut microbiomes, suggesting a potential connection to distant microbial communities.

Personal highlights

Orthogonal validation of low-biomass bacterial signals: combines stringent fluorescence in situ hybridization (FISH), lipopolysaccharide (LPS) immunohistochemistry, and high-resolution spatial molecular imaging (SMI) to visually confirm the presence of intracellular bacterial elements, moving beyond sequencing-based evidence alone.
Spatial correlation of bacterial signal with local immune response: leverages digital spatial profiling (DSP) to demonstrate that tumor regions with high intracellular 16S signal are specifically enriched in antimicrobial response pathways (e.g., HMGB1, TLR9, NF-κB) and show increased neutrophil presence, providing a functional link to the tumor microenvironment.
Intracellular localization across diverse cell types: uses 3D spatial reconstruction to confidently identify bacterial RNA within the cytoplasm of not just tumor cells, but also various immune (macrophages/microglia) and stromal cells, revealing a complex cellular distribution.
Distinction from a cultivable microbiota: Clearly demonstrates that despite the molecular evidence of bacterial components, standard and enriched culture methods failed to yield any bacterial growth, highlighting the non-cultivable, likely low-abundance and fragmented nature of these microbial elements.

Why should we care?

This work expends the brain tumor microenvironment by establishing that bacterial elements, even if not alive and replicating, are a tangible component. For neuro-oncologists and cancer biologists, it introduces a new layer of complexity: the immune system within brain tumors is partially shaped by and responds to these microbial signals. The connection to the oral and gut microbiome opens up possibilities for understanding disease progression and potentially for novel diagnostic or therapeutic strategies targeting these distant microbial communities.

Decrypting cellular engagement and recruitment from spatially resolved transcriptomics

Yao et al. bioRxiv (2025). https://doi.org/10.1101/2025.11.20.689581

The paper in one sentence

Researchers developed a fully interpretable AI tool called “spacer” that uses spatially resolved transcriptomics data to uncover the molecular rules governing how cells are recruited to and interact with each other within tissues, with key applications in cancer immunology and heart disease.

Summary

This study introduces “spacer” (spatial analysis of cellular engagement and recruitment), a novel multi-instance deep learning framework designed to decode the complex patterns of cellular localization from spatially resolved transcriptomics (SRT) data. Moving beyond simple cell mapping, spacer models how the collective gene expression of resident “recruiting” cells (e.g., tumor cells) in a local neighborhood influences the infiltration of “engaging” cells (e.g., immune cells). Applied to a large panel of 37 human and mouse SRT datasets spanning various cancers and myocarditis, spacer identified key genes and pathways that either attract or repel specific cell types. In tumors, it discovered that genes involved in antigen presentation (like HLA molecules) are potent T-cell attractors but are often downregulated by tumors, while mucin expression creates a physical barrier that repels T cells. The tool also revealed that in a mouse model of myocarditis, CD4+ T cells, though fewer in number, are more functionally active and responsive than their CD8+ counterparts. By integrating orthogonal data like immunopeptidomics and spatial TCR sequencing, the study validates spacer’s findings and establishes a new, interpretable paradigm for uncovering the in situ mechanisms of tissue organization and cellular crosstalk.

Personal highlights

Interpretable deep learning for spatial biology: Spacer is built as a fully interpretable multiple-instance learning neural network that directly embeds biological principles—like the collective influence of a neighborhood of cells and the importance of spatial proximity—into its architecture, moving beyond “black box” predictions to deliver mechanistic insights.
Linking gene expression to immunogenic potential: the model successfully identifies that tumor-cell genes with the highest power to recruit T cells are not only involved in antigen presentation but also empirically encode a greater number of immunogenic peptides, validated by orthogonal immunopeptidomics data from patient tumors.
Uncovering cellular “repulsion” signals: Spacer delineates genes that actively inhibit cell recruitment, revealing that extracellular matrix components, particularly mucins secreted by tumor cells, create a barrier that repels T-cell infiltration, offering a new perspective on immune exclusion.
Revealing context-specific engagement rules: the framework distinguishes between the “outcome” of T-cell engagement (anti-tumor response in permissive regions) and the “cause” of their presence in non-permissive regions (high intrinsic migratory capacity or stemness), providing a nuanced view of tumor-immune interactions.
Cross-tissue and cross-species discovery power: application to myocarditis uncovered a surprisingly active role for CD4+ T cells over CD8+ T cells, driven by MHC class II-presented peptides from developmental genes, demonstrating spacer’s versatility in uncovering novel biology beyond oncology.

Why should we care?

Spacer fundamentally shifts how we can extract meaning from the rapidly growing field of spatial genomics. It answers the critical “why” and “how” behind the cellular structures we see in tissues, moving from descriptive maps to predictive, mechanistic models. For cancer researchers and immunologists, it provides a powerful, AI-driven lens to identify the key tumor-derived signals that dictate immune cell recruitment and engagement, with direct implications for predicting immunotherapy response and discovering new antigen targets.

Cancer Hijacks Pain-Sensing Nerves to Shut Down Anti-Tumor Immunity

Zhang et al. Cell (2025). https://doi.org/10.1016/j.cell.2025.09.029

The paper in one sentence

Cancer cells, under pressure from immune cells, activate pain-sensing nerves that remotely rewire lymph nodes into an immune-suppressed state, promoting tumor growth and reducing the efficacy of immunotherapy.

Summary

This study uncovers a sophisticated “inter-organ neuroimmune circuit” that cancers exploit to evade the body’s defenses. Researchers found that in head and neck cancers, tumor cells feeling pressure from macrophages (a type of immune cell) secrete a protein called SLIT2. This protein activates nearby pain-sensing (nociceptive) neurons, which in turn send signals to neurons innervating distant tumor-draining lymph nodes (TDLNs). These activated lymph node neurons release a neuropeptide called CGRP, which reprograms the TDLNs into an immune-suppressed state. Consequently, these suppressed lymph nodes produce less of a key immune-signaling molecule, CCL5, which leads to a pro-tumor polarization of macrophages back in the main tumor, fueling cancer growth and blunting the effect of immune checkpoint blockade therapy. Crucially, blocking this nerve-mediated communication restored anti-tumor immunity, alleviated cancer pain, and improved the response to immunotherapy in mouse models.

Personal highlights

An inter-organ escape route for cancer: the study maps a complete circuit where a local signal in the tumor (SLIT2 from cancer cells) is translated into a neuronal signal that travels to a distant organ (the lymph node), which then systemically suppresses immunity, allowing the tumor to thrive.
Cancer cell-intrinsic ATF4-SLIT2 axis as the ignition switch: under immune pressure, cancer cells activate the transcription factor ATF4, which directly drives the production of SLIT2, the key molecule that kicks off the entire process by activating the tumor-innervating pain neurons.
CGRP as the effector in lymph nodes: the work precisely identifies the neuropeptide CGRP, released by activated neurons within the lymph nodes, as the direct agent responsible for remodeling the lymph node into an immune-suppressed state, reducing critical anti-tumor T cells and dendritic cells.
Mechanistic link to TAM polarization via CCL5: yhe research connects the dots from neuronal activity in the lymph node to changes in the tumor itself, showing that suppressed lymph nodes produce less CCL5, which in turn promotes the polarization of tumor-associated macrophages (TAMs) into a pro-tumor, M2-like state.
Therapeutic potential of blocking neuro-immune crosstalk: Demonstrating that targeting this axis, either by denervating pain neurons, inhibiting CGRP signaling with an existing anti-migraine drug (rimegepant), or blocking the ATF4-SLIT2 axis, not only reduces tumor growth but also alleviates cancer-induced pain and synergizes with immunotherapy.

Why should we care?

This work extends our understanding of how cancer systemically evades attack, positioning the nervous system as a central conductor of immune suppression across different organs. For cancer biologists and immunologists, it reveals a novel, therapeutically targetable pathway where existing neurological drugs could be repurposed to enhance cancer treatment. For clinicians and patients, it offers a compelling dual-benefit strategy: a common migraine medication could potentially be used to both relieve the debilitating pain of cancer and improve the efficacy of immunotherapy, directly linking symptom management to tumor control. It also suggests that a patient’s pain level could be a biomarker for this immune-escape pathway, making a routine clinical assessment a potential predictor of treatment response.

MIMYR: Generative modeling of missing tissue in spatial transcriptomics

Deshpande et al. bioRxiv (2025). https://doi.org/10.1101/2025.11.24.690239

The paper in one sentence

MIMYR is a generative AI framework that reconstructs missing or unmeasured regions in spatial transcriptomics data by synthesizing realistic cell locations, types, and gene expression profiles.

Summary

Spatial transcriptomics allows scientists to see where genes are active within a tissue, but physical sectioning often leads to missing pieces, tears, lost regions, or slices allocated to other experiments, which disrupts the spatial context and limits analysis. MIMYR tackles this by breaking down the reconstruction problem into three coordinated tasks: first, a diffusion model predicts where cells should be located in the missing area; second, a classifier assigns cell types based on those locations; and third, a transformer model generates the full gene expression profile for each cell, conditioned on its position, type, and other metadata like disease state. The framework is highly generalizable, working across different gene panels, slicing orientations, and even disease contexts, as demonstrated on mouse brain atlases and an Alzheimer’s disease dataset. By realistically “filling in the blanks,” MIMYR enables more complete and powerful analyses of tissue architecture and function.

Personal highlights

Three-stage generative pipeline for holistic reconstruction: MIMYR decomposes the complex problem into three sequential, coupled steps: generating plausible cell locations via a plane-conditioned diffusion model, assigning cell identities with a spatial-coordinate MLP, and synthesizing full transcriptomes with a context-aware transformer.
Biologically grounded gene expression generation: The transformer orders genes based on a Gene Regulatory Network (GRN) structure before generating their expression, imposing a biologically meaningful sequence that reflects regulatory hierarchies rather than an arbitrary order.
Robust cross-dataset and cross-panel generalization: The model demonstrates strong zero-shot and few-shot transfer learning capabilities, effectively reconstructing tissue from datasets with different gene panels (e.g., MERSCOPE to MERFISH) and even adapting to new slicing orientations like sagittal sections with minimal fine-tuning.
Conditional generation for biological inference: By conditioning on metadata tokens, MIMYR can perform powerful in silico experiments, such as extending a limited gene panel to predict the spatial patterns of unmeasured genes or generating “what-if” scenarios like synthetic wild-type control tissue for a diseased sample.

Why should we care?

MIMYR directly addresses a pervasive but often overlooked practical problem in spatial biology: spatial incomplete data. Tissue damage and experimental designs inevitably lead to missing regions, creating gaps that obscure biological insights. This framework provides a principled, AI-driven solution to “complete the picture,” enabling researchers to recover a more faithful representation of tissue architecture and gene expression.

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

Ready for more?