Weekly reads 17/11/25

New rules of life: gene editing, neuro-immune niches, AI models and anti-tumor synergy

Nov 23, 2025

This week’s reads focus on how biology can be rewritten: by editing genes, shaping immune and neuronal identities, disentangling cellular signals, and reprogramming aging itself. Laffin et al. deliver a landmark first-in-human CRISPR therapy for a common cardiometabolic disease, showing that a single ANGPTL3 edit can profoundly lower atherogenic lipids. Kolter et al. uncover a precise neuro–immune circuit where sensory neurons sculpt macrophage identity through contact-activated TGF-β, enabling nerve regeneration. In computational biology, Liu et al. introduce FADVI, an interpretable framework that cleanly separates technical noise from true biological signal across single-cell and spatial omics. Weng et al. reveal a matrix-imposed exhaustion pathway in cancer, where TSP-1/CD47 signaling forces T cells into dysfunction — a vulnerability with therapeutic potential. von Locquenghien et al. engineer MITEs, tumor-activated immunocytokines that reprogram myeloid cells while supercharging T and NK cells, achieving potent tumor control with minimal toxicity. Finally, Qiu et al. present ACE, an explainable deep learning model that isolates conserved aging signatures across species and identifies genes that genuinely modulate lifespan.

Preprints/articles that I managed to read this week

Phase 1 Trial of CRISPR-Cas9 Gene Editing Targeting ANGPTL3

Laffin et al. N Engl J Med (2025). https://doi.org/10.1056/NEJMoa2511778

The paper in one sentence

A first-in-human trial demonstrates that a single infusion of CTX310, an in vivo CRISPR-Cas9 therapy targeting the ANGPTL3 gene, is safe and leads to potent, dose-dependent reductions in key atherogenic lipids in patients with refractory dyslipidemia.

Summary

This phase 1 trial evaluated the safety and efficacy of CTX310, a CRISPR-Cas9-based therapy designed to disrupt the ANGPTL3 gene in the liver. ANGPTL3 is a protein that inhibits the clearance of LDL cholesterol and triglycerides from the blood; naturally occurring loss-of-function variants are linked to lifelong lower lipid levels and reduced heart disease risk. Fifteen patients with uncontrolled hypercholesterolemia or hypertriglyceridemia, despite being on maximally tolerated lipid-lowering drugs, received a single intravenous dose of CTX310. The therapy showed an acceptable safety profile with no dose-limiting toxicities, featuring mostly mild infusion-related reactions and one transient liver enzyme elevation. Efficacy was dose-dependent: the two highest doses (0.7 and 0.8 mg/kg) produced profound reductions in ANGPTL3 protein levels (≈-75%) and concomitant reductions in LDL cholesterol (≈-50%) and triglycerides (≈-55%) at 60 days, mimicking the protective genetic phenotype.

Personal highlights

First-in-human in vivo CRISPR-Cas9 for a common cardiometabolic target: This trial represents a pivotal step in applying single-administration gene editing to a widespread condition like dyslipidemia, moving beyond rare monogenic diseases to a target with broad population-level impact.
Potent, dose-dependent lipid lowering after a single infusion: the highest doses achieved a near-physiological knockout of ANGPTL3, resulting in substantial and simultaneous reductions of both LDL cholesterol and triglycerides, a dual effect that is difficult to achieve with most existing therapies.
Manageable safety profile with no dose-limiting toxicities: the primary safety signal was transient, manageable infusion-related reactions, with no serious adverse events attributed to the gene-editing agent, supporting the feasibility of further clinical development.

Why should we care?

For the millions of patients with stubbornly high cholesterol and triglycerides despite available drugs, in vivo gene editing offers the hope of a “functional cure” that could drastically reduce their lifelong pill burden and cardiovascular risk. For the medical community, it demonstrates the expanding reach of CRISPR beyond ultra-rare diseases into common conditions, forcing a conversation about the long-term safety, accessibility, and ethical implications of permanent genetic therapies. While longer follow-up is essential, this study is a landmark proof-of-concept that a single infusion can safely reprogram human liver genetics to confer a protective, cardiometabolic profile.

tSensory neurons shape local macrophage identity via TGF-β signaling

Kolter et al. Immunity (2025). https://doi.org/10.1016/j.immuni.2025.08.004

The paper in one sentence

Sensory neurons in the skin activate TGF-β upon physical contact with macrophages, imprinting them with a specialized identity that is essential for patrolling nerves and facilitating their regeneration after injury.

Summary

This study reveals a precise molecular dialogue between the nervous and immune systems in the skin. The authors show that after an injury, bone marrow-derived monocytes are recruited to the site of damaged, sprouting sensory nerves. Through direct physical contact with these axons, the monocytes undergo a profound transformation, becoming specialized nerve-associated macrophages (sNAMs). This transformation is driven by the local activation of TGF-β, a growth factor presented by the neurons and activated by the macrophages themselves via integrin proteins. This TGF-β signaling is both necessary and sufficient to give the macrophages their unique patrolling identity, characterized by high CX3CR1 and low CD206 expression. Crucially, when this neuro-immune crosstalk is disrupted, nerve regeneration is severely impaired, highlighting that these neuron-shaped macrophages are not just bystanders but active players in healing.

Personal highlights

Localized TGF-β activation defines a neuronal niche: the study demonstrates that TGF-β signaling is not a general cue for all skin macrophages but is spatially restricted to the immediate vicinity of sensory nerves. This precision is achieved through integrin-mediated activation of latent TGF-β, which requires direct physical contact between macrophages and neurons.
Macrophage plasticity is guided by the neuronal microenvironment: both prenatally seeded and bone marrow-derived monocytes converge on the same sNAM identity when exposed to the nerve niche. This shows that the local microenvironment, not just cellular origin, is the dominant instructor of macrophage identity and function.
A conserved bi-directional neuro-immune circuit: the interaction is evolutionarily conserved from mice to humans. Using iPSC-derived cells, the authors show that human sensory neurons also use TGF-β to upregulate CX3CR1 on human macrophages, which in turn support the growth and bundling of neuronal axons.
Dual signaling input from neurons: the sNAM phenotype is co-shaped by two key neuronal signals: TGF-β (which drives the core sNAM program and suppresses CD206) and the neuropeptide CGRP (which helps maintain CD206 expression), illustrating the complexity of niche-specific imprinting.
Functional necessity for tissue repair: disrupting TGF-β signaling specifically in sNAMs doesn’t just alter their identity; it functionally impairs the entire process of nerve regeneration and wound healing after injury, positioning these specialized macrophages as critical mediators of tissue repair.

Why should we care?

This work fundamentally advances our understanding of how our body’s two major surveillance systems, the nervous and immune systems, communicate to maintain health. It moves beyond simply observing that these cells interact and identifies the precise molecular language (TGF-β) they use to talk to each other. For anyone interested in healing, chronic wounds, or neuropathies, this research suggests that targeting this specific neuro-immune axis could be a powerful therapeutic strategy. It illustrates that healing is not just about stem cells or growth factors alone, but about coordinating the complex dialogue between different cell types in a damaged tissue.

FADVI: Disentangled Representation Learning for Robust Integration of Single-Cell and Spatial Omics Data

Liu et al. bioRxiv (2025). https://doi.org/10.1101/2025.11.03.683398

The paper in one sentence

FADVI is a new AI framework that cleanly separates technical artifacts from true biological signals in single-cell and spatial genomics data, leading to more robust integration and interpretable results than existing methods.

Summary

Integrating data from different single-cell and spatial omics experiments is a major challenge due to “batch effects”, technical variations that can obscure biological discoveries. While current methods focus on minimizing these effects, they often cannot disentangle them from the actual biological signals. FADVI tackles this by using a specialized variational autoencoder that explicitly partitions the learned data representation (latent space) into three distinct parts: one for batch-specific technical variation, one for label-related biological variation (like cell type), and a residual factor. It uses a combination of supervised learning, adversarial networks, and a cross-covariance penalty to ensure these representations are independent. Benchmarked across scRNA-seq, scATAC-seq, and spatial transcriptomics datasets, FADVI consistently outperformed other state-of-the-art methods in integration quality. A key advantage is its interpretability: using feature attribution, FADVI can pinpoint specific genes associated with cell identity or batch effects, providing deeper biological insights.

Personal highlights

Explicit disentanglement of variation sources: FADVI’s core innovation is its VAE architecture, which partitions the latent space into dedicated, statistically independent subspaces for batch effects, biological labels, and residual variation, forcing a clean separation that other methods lack.
Adversarial training and cross-covariance for exclusivity: employs adversarial networks with gradient reversal to actively prevent information leakage between subspaces, supplemented by a cross-covariance penalty that further decorrelates them, ensuring the batch factor is free of biology and vice versa.
Consistent top-tier performance across diverse modalities: demonstrates good integration scores not only on large-scale scRNA-seq atlases but also on scATAC-seq and the challenging task of integrating high-resolution spatial transcriptomics data with scRNA-seq across multiple platforms.
Interpretable feature attribution for biological discovery: leverages SHAP-based analysis to identify genes most strongly associated with cell type identity and batch effects, moving beyond integration to offer testable hypotheses about key drivers of biological and technical variation.
Robustness to real-world label noise: shows stable performance even when a subset of cell type labels is intentionally incorrect, a critical feature for dealing with the imperfect annotations common in large-scale single-cell studies.

Why should we care?

FADVI shifts the goal of data integration from simply mixing datasets to intelligently pulling them apart. By explicitly disentangling technical noise from biological signal, it provides a more truthful and trustworthy representation of the underlying biology. For researchers building large-scale cell atlases, it offers a robust method to combine data from different labs and technologies without losing biological resolution. For biologists, its interpretability features can highlight novel gene-batch interactions or confirm known markers, turning a computational pipeline into a discovery engine. Ultimately, FADVI provides a powerful, principled framework that enhances the reliability and interpretability of integrative analysis.

Thrombospondin-1-CD47 signaling contributes to the development of T cell exhaustion in cancer

Weng, C. et al. Thrombospondin-1-CD47 signaling contributes to the development of T cell exhaustion in cancer. Nat Immunol (2025). https://doi.org/10.1038/s41590-025-02321-5

The paper in one sentence

This study identifies a novel pathway where the interaction between the extracellular matrix protein Thrombospondin-1 (TSP-1) and the receptor CD47 on T cells drives their functional exhaustion in tumors by activating calcium-dependent calcineurin-NFAT signaling.

Summary

T cell exhaustion is a major barrier to effective cancer immunotherapy, rendering T cells dysfunctional and unresponsive. This study uncovers a previously unknown driver of this process: the signaling axis between Thrombospondin-1 (TSP-1), a protein abundant in the tumor matrix, and CD47, a receptor best known as a “don’t eat me” signal on cancer cells. The researchers found that CD47 is highly upregulated on exhausted T cells in human and mouse tumors. They demonstrate that TSP-1 binding to CD47 on T cells triggers a calcium influx, which activates the calcineurin-NFAT signaling pathway. This, in turn, induces the expression of the exhaustion master regulator TOX and other inhibitory receptors (like PD-1, LAG-3), while impairing the production of effector cytokines. Crucially, genetically or pharmacologically disrupting the TSP-1-CD47 interaction prevented T cell exhaustion, enhanced T cell infiltration into tumors, and improved tumor control, especially when combined with anti-PD-1 therapy.

Personal highlights

CD47 as a novel functional marker of T cell exhaustion: identifies CD47 upregulation not just as a correlate, but as a functional driver of the exhausted T cell state, linking it directly to high expression of TOX, PD-1, LAG-3, and impaired cytokine production in human and murine tumors.
Mechanistic link from extracellular matrix to nuclear programming: delineates a clear signaling cascade where the TSP-1/CD47 interaction triggers intracellular calcium influx, leading to calcineurin activation, NFAT nuclear translocation, and subsequent induction of the exhaustion transcriptome via TOX.
Genetic evidence for a T-cell intrinsic role of CD47: employs sophisticated adoptive co-transfer models of CD47-WT and CD47-heterozygous T cells into the same host, conclusively showing that reduced CD47 expression cell-intrinsically protects against exhaustion and enhances anti-tumor efficacy.
Therapeutic rescue with a selective pathway disruptor: uses the TAX2 peptide, a specific inhibitor of the TSP-1/CD47 interaction, to phenocopy the genetic findings, reducing exhaustion markers, improving T cell function, and enhancing tumor infiltration without the confounding effects of global CD47 blockade.
Synergy with existing immunotherapy: demonstrates that disrupting the TSP-1/CD47 axis can enhance the efficacy of anti-PD-1 checkpoint blockade, positioning it as a promising combination strategy to overcome resistance in immunologically “cold” tumors.

Why should we care?

This work shifts the paradigm of CD47 from being solely a phagocytosis checkpoint on cancer cells to a direct regulator of T cell fitness within the tumor microenvironment. By identifying the TSP-1/CD47 axis as a potent driver of exhaustion, it reveals a new therapeutic vulnerability that is complementary to existing PD-1 blockade. For oncologists, this opens a novel avenue for combination therapies that could reinvigorate T cells more effectively. For immunologists, it provides a mechanistic bridge between the immunosuppressive extracellular matrix and the intracellular signaling pathways that lock T cells into a dysfunctional state. Ultimately, targeting this pathway offers a promising strategy to prevent T cell exhaustion at its root, potentially leading to more durable responses for cancer patients

Macrophage-targeted immunocytokine leverages myeloid, T, and NK cell synergy for cancer immunotherapy

von Locquenghien et al., Cell (2025). https://doi.org/10.1016/j.cell.2025.10.030

The paper in one sentence

Researchers developed a novel “pro-drug” immunocytokine (MTE) that is selectively activated within tumors to simultaneously block immunosuppressive macrophages and potently stimulate anti-tumor T and NK cells, achieving potent cancer control with minimal toxicity in preclinical models.

Summary

This study introduces a new class of cancer immunotherapy called Myeloid-Targeted Immunocytokines and NK/T cell Enhancers (MITEs). MITEs are engineered to overcome two major hurdles in treating solid tumors: the immunosuppressive environment created by Tumor-Associated Macrophages (TAMs) and the severe toxicity of powerful immune-stimulating cytokines like IL-2.

The key innovation is a dual-targeting “pro-drug” molecule. It consists of an antibody that blocks TREM2, a key checkpoint on immunosuppressive TAMs, fused to a potent IL-2 variant (a “superkine”) that is masked by a blocking domain. This blocking domain is linked via a cleavable sequence that is specifically cut by the protease MMP14, which is highly enriched in TAMs. This design ensures the IL-2 is only activated within the tumor microenvironment, not in healthy tissues.

The results show that MITE-144, the lead candidate:

Reprograms the tumor immune landscape: It switches TAMs from a suppressive to a more inflammatory state and enhances antigen presentation by dendritic cells.
Activates cytotoxic lymphocytes: It robustly stimulates the proliferation, cytotoxicity, and stemness of CD8+ T cells and NK cells while limiting their exhaustion.
Is safe and effective: It achieves powerful tumor control in multiple mouse models without the lethal systemic toxicity seen with unmodified IL-2 fusions.
Synergizes with checkpoint inhibitors: Its efficacy is further enhanced when combined with anti-CTLA-4, which also helps deplete regulatory T cells.
Translates to human tumors: Using patient-derived tumor fragments, the study confirmed that MITE-144 selectively activates human T and NK cells without expanding immunosuppressive Tregs.

Personal highlights

Spatially-informed design of a protease-activated immunocytokine: the MITE platform leverages the TAM-specific expression of MMP14 to create a tumor-restricted “on/off” switch for IL-2 activity, ingeniously confining potent cytokine effects to the tumor while avoiding systemic toxicity.
Dual targeting of myeloid and lymphoid immune compartments: MITEs represent a trans-acting immunomodulator that concurrently disrupts the immunosuppressive TREM2 pathway in macrophages and delivers a potent IL-2 signal to nearby T and NK cells, creating a synergistic anti-tumor immune response.
Comprehensive multi-omic deconvolution of mechanism of action: using single-cell and spatial transcriptomics across murine and human systems, the study meticulously demonstrates how MITE-144 remodels the entire tumor microenvironment, from reprogramming myeloid cell states to enhancing cytotoxic and proliferative programs in lymphocytes.
Validation in a clinically relevant human ex vivo platform: the use of patient-derived tumor fragments (PDTFs) provides strong translational evidence, showing that the immunostimulatory effects of MITEs are conserved in human tumors, activating a coordinated cytotoxic program in patient T and NK cells.
Superior efficacy and safety profile over existing modalities: MITE-144 outperforms monotherapies targeting TREM2, PD-1, or CTLA-4, and its masked design solves the historical problem of IL-2-related severe toxicity, paving the way for a safer, more effective cytokine-based therapy.

Why should we care?

MITEs tackle a central problem in oncology: how to effectively attack tumors that are “cold” or resistant because they are protected by a shield of immunosuppressive cells. By simultaneously dismantling this shield (via TREM2 blockade on TAMs) and weaponizing the immune system’s best attackers (via localized IL-2 to T/NK cells), this strategy represents a powerful new multi-pronged assault on cancer. For oncologists and patients, it offers a promising path to potent, durable responses with a dramatically improved safety window compared to previous cytokine therapies.

ACE: An Explainable AI Framework for Identifying Universal Aging Signatures in Cell Embeddings

Qiu et al. bioRxiv (2025). https://doi.org/10.1101/2025.11.07.687286

The paper in one sentence

ACE is an explainable deep learning model that disentangles aging-related gene expression changes from other biological variations in single-cell data, revealing conserved aging genes and pathways across mice, flies, and humans, which were experimentally validated to affect lifespan.

Summary

Aging is a complex process influenced by many factors, making it difficult to isolate its specific molecular signatures from single-cell data, where signals are often dominated by cell type, tissue, and sex. To address this, researchers developed ACE (Aging Cell Embeddings), a framework based on a variational autoencoder that learns two separate sets of latent variables: one for aging-related variation and another for all other “background” biological factors. Using explainable AI techniques, ACE identifies key genes driving aging both globally (across tissues and cell types) and locally (in specific contexts). Applied to large-scale datasets from mice, flies, and humans, ACE accurately predicts biological age and uncovers evolutionarily conserved aging pathways related to proteostasis and immune function. Crucially, experimental validation in C. elegans confirmed that knocking down ACE-prioritized genes, such as Uba52, significantly impacts lifespan, demonstrating the model’s power to reveal biologically meaningful and universal aging mechanisms.

Personal highlights

Disentangling aging from dominant biological variation: ACE employs a dual-encoder architecture, explicitly separating aging-related gene expression signatures from other strong sources of variation like cell type and tissue, which typically obscure subtle aging signals in single-cell data.
Identification of globally conserved aging mechanisms: the model uncovers aging signatures shared across diverse tissues and cell types, highlighting key roles for pathways in proteostasis, ribosomal function, and synaptic activity in both mouse and fly models.
Accurate biological age clocks at single-cell resolution: ACE’s aging embeddings enable highly accurate prediction of chronological age from individual cell profiles, and these predictions generalize robustly to entirely held-out individuals and even unseen cell types.
Cross-species alignment reveals universal aging genes: by integrating data from humans, mice, and flies and aligning their aging trajectories, ACE identifies conserved aging genes like Uba52, whose importance was unknown in the context of aging across species.

Why should we care?

ACE moves beyond simply cataloging which genes change with age to pinpoint which changes are fundamental drivers of the aging process itself, separating these true signals from the biological “noise” of cell identity. By providing a universal, interpretable framework for aging biology, it opens the door to identifying conserved therapeutic targets that could slow aging across multiple tissues.

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?