Nature Biomedical Engineering

Body surface gastric mapping (BSGM) enables non-invasive measurement of gastric electrophysiology, but the signals are approximately 100 times weaker than cardiac potentials and overlap spectrally with motion artifacts, necessitating labor-intensive manual review that limits clinical scalability. We present an uncertainty-aware deep learning framework combining a signal reconstruction network with a parallel uncertainty estimation network to automate artifact correction in high-resolution BSGM. Models were trained on 2,398 multihour, 64-channel recordings from 27 international clinical sites using weak supervision, a physiology-aware loss function, and uncertainty-gated quality control. In an independent cohort of 127 patients, the system achieved relative reductions of 39% in signal reconstruction error, 9% in total data removed, and 23% in amplitude--movement correlation compared with the industry-standard Wiener filter. Improved signal fidelity altered automated clinical phenotyping in 7% of patients by recovering previously obscured gastric rhythms. Uncertainty-aware deep learning enables reliable automated artifact correction in body-surface gastric mapping, improving signal fidelity and enabling scalable clinical interpretation. The system is FDA-cleared (510(k) K252504) and deployed in clinical practice, demonstrating that data-driven artifact correction can meet regulatory requirements for medical devices and reduce dependence on specialist manual review.

In vivo CAR-T cell therapy eliminates manufacturing complexities associated with ex vivo autologous approaches, but safety concerns have limited adoption. We developed viroVbot, a next-generation in vivo CAR-T platform, by combining computational immunogenicity prediction (CIMMEXTM) with envelope engineering. Screening 22,562 glycoprotein sequences, we identified 641 vesiculovirus homologs, from which we selected Piry virus glycoprotein (PIRYV) as the optimal candidate. PIRYV exhibited lower MHC-epitope density, reduced human seroprevalence, with decreased T cell activation compared to VSV-G. To enhance targeting specificity, we engineered receptor-binding-deficient PIRYV (ePIRYVRBD) displaying CD3/CD7 nanobodies for T cell-selective transduction. To maximize safety, we engineered CAR-TRAP producer cells to eliminate unwanted B cell transduction and incorporated machine learning-optimized T cell-specific promoters that restrict CAR activation exclusively to lymphocytes. Additional modifications suppressed hepatocyte expression and prevented phagocytic uptake. In humanized xenograft models, viroVbot3 generated potent BCMA/CD19 specific CAR-T responses against multiple myeloma and Claudin18.2-targeting gastric cancer, demonstrating sequential redosing with alternative envelopes. Critically, viroVbot3 exhibited minimal off-target organ biodistribution with CAR expression restricted to T lymphocytes. These findings establish viroVbot as a low-immunogenic platform for scalable in vivo CAR-T manufacturing with capability for sequential redosing across hematologic and solid tumors.

Adaptive deep brain stimulation (aDBS) relies on physiological biomarkers to infer motor state and guide therapeutic stimulation in Parkinson's disease. However, neural biomarkers may themselves be altered by stimulation, potentially limiting their utility for closed-loop control. We address this limitation by testing whether DBS device-embedded accelerometers can accurately track Parkinsonian motor state across stimulation conditions. We analysed over 1,900 hours of chronic recordings of subthalamic nucleus (STN), sensorimotor cortical and device-embedded accelerometry signals acquired before and during continuous STN stimulation, alongside continuous wearable assessments of bradykinesia and dyskinesia. Across stimulation conditions, accelerometry-derived features robustly tracked motor symptom severity and outperformed neural features for symptom decoding. Mechanistically, total STN beta power - a widely used biomarker for aDBS - proved less informative because it conflates periodic and aperiodic neural processes with opposing relationships to motor state. Under active stimulation, periodic beta activity showed reduced coupling to symptom severity, whereas STN aperiodic activity, cortical periodic activity and cortico-subthalamic coherence remained comparatively stable. Together, these findings demonstrate that neural and behavioural biomarkers exhibit differential robustness during deep brain stimulation and identify device-embedded accelerometry as a robust behavioural biomarker of motor state, motivating its use in next-generation adaptive DBS systems.

Background and Aims: Endoscopic artificial intelligence is commonly validated on selected single images, whereas gastric cancer interpretation requires integrating whole examinations. We developed GutCore and evaluated whether whole-case endoscopic images could be used for patient-level assessment of gastric cancer depth, biomarkers, and prognosis. Methods: GutCore was pretrained on 5.6 million de-identified endoscopic images from more than ten hospitals. We compared it with five general, medical, and endoscopy-specific foundation models using open image-level datasets and an internal tertiary-center cohort of 11,035 de-identified endoscopic examinations (2019-2023): 8,049 with early or advanced gastric cancer and 2,986 with benign gastritis or intestinal metaplasia. All examination images were aggregated for patient-level assessment of cancer status, invasion depth, molecular biomarkers, and overall survival. Results: Aggregating all stored images from each examination enabled patient-level gastric cancer assessment without selecting representative frames. GutCore achieved AUCs of 0.995 for cancer detection, 0.960 for muscularis propria invasion, and 0.804 for SM2-or-deeper invasion. Prediction of tissue-defined biomarker status was strongest for Epstein-Barr virus status and MLH1 loss (AUC, 0.831 and 0.854), with lower HER2 performance (AUC, 0.673). In the held-out advanced gastric cancer test set, GutCore-derived risk groups showed marked survival separation (log-rank P < .0001; high-risk vs low-risk hazard ratio, 13.18; 95% CI, 6.06-28.66), with stratification persisting within pathological stage II and III disease. External frame-level benchmarks showed strong performance for anatomical landmark recognition, disease grading, and segmentation. Conclusions: GutCore supported whole-case patient-level gastric cancer assessment using routinely stored endoscopic images. Further validation in independent clinical cohorts is needed to establish generalizability and clinical utility.

IgG-degrading enzymes have emerged as innovative therapeutic agents for treating conditions driven by pathogenic antibodies. Here, we used structure-guided rational design to engineer IdeSM33, a double mutant (K167R/D226E) of the IgG-specific bacterial protease IdeS from Streptococcus pyogenes, with improved catalytic efficiency. Biolayer interferometry revealed a fourfold increase in binding affinity relative to wild-type IdeS (IdeSWT). This enhancement is likely attributable to mutations that strengthen hydrogen bonding at the enzyme-IgG Fc interface. In vitro, IdeSM33 has higher performance than IdeSWT in cleaving serum IgG. In vivo studies in rabbits demonstrated that IdeSM33 effectively depleted circulating IgG and showed better performance at a dose of 0.005 mg/kg than the IdeSWT. Although doses greater than 0.2 mg/kg demonstrated higher plasma concentrations of IdeS and a larger AUC 0 to last, they did not show a significant enhancement in the pharmacodynamics of IgG degradation. Importantly, a single dose of IdeSM33 (0.2 mg/kg) potently degraded binding and neutralizing antibodies against AAV9 within 1-2 days and restored hepatic AAV9 transduction in pre-immunized animals. Together, these findings highlight IdeSM33 as a potent and safe engineered enzyme with therapeutic potential for autoimmune disorders, transplant rejection, and overcoming pre-existing humoral immunity in gene therapy.

Chimeric antigen receptor (CAR)-T cell therapies have transformed the treatment of various tumor types by redirecting and activating T cells against tumor cells. However, CAR-T cell manufacturing approaches remain challenging and limit their widespread use in clinical settings. In vivo CAR-T therapy bypasses ex vivo cell manufacturing and patient preconditioning limitations; however, it faces a significant safety concern as CAR proteins on viral packaging cells are incorporated into budding virions, leading to off-target transduction of tumor cells. Here, we address this risk by developing the CAR-Less ER-Anchor Vector (CLEAN-V) system. By exploiting endoplasmic reticulum (ER) retention, CLEAN-V prevents the CAR protein from trafficking to the cell surface during viral packaging, thereby blocking its incorporation into the viral envelope. CLEAN-V particles exhibit near-complete loss of CAR-mediated tumor cell transduction. Furthermore, CLEAN-V integrates seamlessly into existing third-generation LVV workflows in four- or five-plasmid formats and generates CAR-T cells with preserved phenotypic and functional integrity. These results establish CLEAN-V as a robust platform for developing safe, targeted lentiviral vectors for in vivo CAR-T therapy.

Precise, non-invasive manipulation of individual living cells remains a central challenge in biomedical science, with far-reaching implications for single-cell analysis, tissue engineering, and the study of cell-cell interactions. Here, we report the first demonstration of single-cell control using bulk acoustic standing-wave acoustofluidics with closed-loop feedback. We introduce VeLO (Vector-based Local Optimization), a model-free, reinforcement learning-inspired algorithm that enables programmable two-dimensional manipulation of individual cells using a single piezoelectric transducer. Without prior calibration or physical modeling, VeLO learns system dynamics online from acoustically induced cell displacements and automatically adapts to nonlinear, time-varying conditions. We achieve robust control across multiple cell types (DU-145, Jurkat, K-562) and independent manipulation of multiple cells, including controlled cell-cell contact. By combining simplicity of hardware with autonomous, adaptive control, this approach establishes multimodal acoustofluidics as a versatile tool for label-free, high-precision single-cell manipulation.

Therapeutic genome editing requires delivery of editing molecules to defined cell types, but targeting specificity and efficiency are currently limited. We hypothesized that properties inherent to immune cells, including tissue infiltration and programmed cell recognition, could be harnessed to engineer a cell-based delivery system. We show here that T cells can both produce and transfer editing machinery to target cells. In response to a programmable ligand, engineered T-lymphoid cells can transfer enzymes using complex spatiotemporal logic and deliver cargo in a cell contact-dependent or -independent manner. We demonstrate feasibility of this approach in primary human T cells, establishing a customizable genetic circuit for macromolecular delivery controlled by intercellular interactions.

How drug treatments reshape immune and metabolic states within intact tumors remains difficult to study with existing methods. We introduce a spatial pharmacology platform that enables parallel analysis of multiple agents within a single tumor, linking local drug exposure to immune and metabolic remodeling. Using a microdevice for localized drug delivery, we created a large-scale paired CyCIF-MALDI dataset spanning 1.5 million cells across 27 MMTV-PyMT tumor sections and nine treatment programs, enabling integrated spatial pharmacology at unprecedented scale. Metabolic signatures robustly predict proteomic spatial neighborhoods establishing metabolism as a powerful predictor of tumor organization and immune phenotype. Within this framework, we identify a dominant metabolic axis defined by the myeloid polarization between CSF1R+ tumor-associated macrophages and MPO+ infiltrating myeloid cells localized near regions of drug-induced tumor cell death. Finally, we detect putative lipid-associated macrophage (LAM)-like populations within drug-resistant treatment regions.

Tissues are spatially organized systems in which cell states, functions and interactions vary across spatial coordinates, forming compartments or gradients shaped by local microenvironments. Understanding how molecular features and cell-cell interactions change across space and time is central to studying development, homeostasis and disease. Addressing these questions increasingly requires the integration of multi-modal spatial data, which provides complementary views of cellular and structural organization. However, existing computational approaches typically combine modalities by weighting them equally, overlooking domain-specific technical artifacts, differences in spatial resolution and non-overlapping feature spaces. In addition, methods for spatial cell-cell communication analysis are largely developed for single-modality settings and do not model how interactions vary across the tissue. To address these gaps, we introduce LYNX, a deep generative framework that learns a shared latent representation of spatial dynamics from joint-measured modalities in the 2D or 3D domain, to provide a unified coordinate system for modeling how cell-cell interactions, phenotypes, and molecular programs vary along continuous spatial gradients. LYNX identifies spatial programs difficult to resolve with existing approaches, including metabolically coupled porto-central interaction remodeling in liver, recovery of degraded proteomic signals along the cortico-medullary axis in thymus, and branching trajectories towards DCIS and invasive niches marked by distinct stromal activation-states and immune-tumor crosstalk in breast tumor microenvironment. We demonstrate that LYNX robustly infers spatially resolved gradients, maps functional compartments and cell-cell interactions along spatial axes and is compatible across diverse spatial profiling technologies, modalities, and resolution disparities. LYNX provides a foundational and scalable framework to advance our understanding of healthy tissue physiology and to decode temporal evolution of complex diseases.

Adeno-associated virus (AAV)-based gene therapy has made steady progress towards efficient delivery to numerous target cell populations, yet the virus's 5 kb packaging limit remains a challenge for effective and in some cases cell-selective cargo expression. Here, we introduce Expression-Linked Promoter Selection (ELiPS), a high-throughput platform for generating and functionally screening >106 engineered, short promoter variants using an AAV expression platform. ELiPS relies on a Golden Gate cloning method to build random oligomers of selected transcription factor binding sites (TFBSs) upstream of a minimal promoter, GFP, and a unique 3' barcode. As a proof of concept, to engineer short (~250 bp), synthetic, ubiquitous promoters, we applied ELiPS to build two libraries composed of TFBSs for ubiquitously expressed transcription factors (TFs) and screened them via AAV-mediated transduction in vitro. This strategy identified promoters with expression surpassing human cytomegalovirus (CMV) and CAG in vitro, and one variant was capable of driving therapeutic expression of B-domain-deleted Factor VIII (BDDFVIII) in vivo at levels comparable to a liver-specific promoter benchmark. ELiPS thus establishes a scalable framework for promoter discovery, enabling the design of compact, ubiquitous or cell-selective expression cassettes that enable further precision and efficacy in AAV-based gene therapies.

Label-free scattering imaging is widely used in pathology because it enables sensitive tissue assessment without exogenous contrast agents. Yet its limited optical penetration has prevented scattering-based methods from being applied to whole-organ pathology mapping. Here we present clearing-assisted scattering tomography (CAST), a high-throughput, label-free whole-brain mesoscope enabled by selective lipid clearance for scattering enhancement (SELiC). SELiC modulates endogenous refractive-index heterogeneity in cleared tissue, providing whole-brain optical penetration while retaining strong scattering contrast from amyloid plaques and white-matter fibre bundles. CAST enables volumetric imaging of intact mouse brains and brain-wide mapping of amyloid plaque pathology across anatomical regions. This platform establishes a scalable route for label-free, system-level analysis of amyloid pathology and tissue architecture in Alzheimers disease (AD) models.

Despite the significant technical advancement in spatial transcriptomics, its clinical usage is largely untapped. Here, we develop an integrated system, ENDO-Genome, for minimally invasive in-body transcript sampling to facilitate live spatial transcriptomic analysis of human internal organs. This is achieved by integrating a nanoarrayed biochip with existing endoscope to perform pressure-sensor-calibrated "Touch & Go" RNA extraction directly from human internal organs, including the highly vascularized liver or kidney, without the need for tissue biopsy, voiding any bleeding risks. By a demonstration using gastrointestinal endoscopy, multiplexed landscape of 55 mRNA transcripts was obtained from multiple locations of human intestinal tract via a 5-minute operation in routine examinations. Benefiting from a sequencing-free approach, each assay costs less than 10 US dollars. For the clinical study involving 15 Crohn' s disease (CD) patients, no complication case was reported out of 47 ENDO-Genome operations, showcasing the gentle deposition and excellent safety of the technique. The live spatial transcriptomics provides direct in vivo pictures of the heterogenous spatial transcriptional programs underlying CD pathological response at different intestinal locations, revealing distinct ileal phenotypes. This is manifested by unique microscale scattering of inflammation gene clusters, along with the discovery of a tissue-specific cooperative mechanisms between inflammation and RNA methylation regulations at single- or multi-cell scales.

Research on targeted genetic therapies for myocardial diseases, such as cardiomyopathies, currently focuses on (r)AAVs as the delivery method. Despite substantial efforts and advances in animal trials, predicting biodistribution and transduction efficacy in human tissue remains challenging due to interspecies differences in tissue tropism and the difficulty of accurately assessing alternative delivery routes and vector differences. The application in humans has also proven challenging, in part due to severe adverse events associated with systemic administration of (r)AAVs. This necessitates implementing alternative trial designs and stringent evaluation methods that minimize harm or risk to patients. Applying a predesigned vector carrying a gene-editing tool to a normothermic machine-perfused living beating heart in an ex vivo setting could overcome conventional obstacles and limitations. This can serve as a basis for safe and effective gene-therapy testing and assist in evaluating effects at the molecular level. Boxed-Breathing-Heart is a translational trial assessing the feasibility of ex vivo gene editing and gene translation in normothermic machine-perfused human hearts. Human hearts explanted from cardiomyopathy patients undergoing heart transplantation are donated for research and immediately placed in an Organ Care System, where they are surgically connected. The viability of the heart is maintained through normothermic perfusion of system solutions and donor blood. A predesigned AAV containing a CRISPR-Cas system is infused into the circulation and dispersed throughout the tissue via coronary perfusion. The changes at the cellular and molecular levels are assessed continuously via frequent sequential myocardial biopsies. Furthermore, after the pre-planned 72-hour perfusion, the heart is sectioned and analyzed using spatial and single-cell omics. The aim is to provide a proof-of-concept for genetic therapeutic options delivered to the human heart via AAV in an ex vivo perfusion setup. In summary, Boxed-Breathing-Heart provides an ex vivo translational platform for evaluating targeted cardiac gene therapies, enabling molecular analysis directly in human hearts and accelerating clinical translation without posing risks to patients.

Clinical treatment planning of near-infrared (NIR) brain stimulation requires patient-specific light dosimetry to optimize fluence delivery to cortical targets. The gold-standard Monte Carlo (MC) photon transport forward solver is accurate but computationally expensive and non-differentiable for personalized inverse design across subjects. Here, we present a foundation-model (FM)-encoded, differentiable implicit-neural surrogate for the MC solver. A pretrained 3D MRI/CT foundation model, VISTA3D, is domain-adapted to head phantoms with known optical properties to encode the subject anatomy. Next, an implicit neural representation is used to predict light fluence at arbitrary continuous coordinates. This formulation enables off-grid queries and gradients with respect to illumination parameters. Trained with a physics-informed, decade-stratified loss, the surrogate attains R2 {approx} 0.90 on held-out subjects. Ablation results show that the FM benefit is contingent on domain adaptation. Benchmarked against standard learned surrogates, our model is the most accurate in the high-dose region and best on dose-fidelity metrics ({gamma}-index, treated-volume DICE). Finally, gradient-based optimization through the surrogate recovers MC-consistent illumination configurations 50-240 x faster.

Embryonic development is orchestrated by complex gene regulatory networks, and learning regulatory dynamics from developmental data could allow us to understand, predict, and ultimately engineer cell fates. Here we introduce Navigo (https://github.com/aristoteleo/Navigo-release), a biologically grounded generative modeling framework that learns a developmental vector field by integrating flow matching at the population level with RNA kinetics modeling at the molecular level. Navigo accurately maps developmental trajectories across lineages on a mouse embryogenesis scRNA-seq atlas spanning 43 time points and comprising 12.4 million cells. Applied to cardiac development, Navigo enables disease modeling by mechanistically resolving regulatory networks that distinguish congenital heart disease subtypes. Navigo also predicts perturbation effects in a zero-shot manner, as validated on independent in vivo data from six knockout genotypes without perturbation-specific training, uncovering lineage-specific gene-compensation mechanisms. Moreover, Navigo guides rational cell-fate engineering, exemplified by fibroblast reprogramming analyses, including identifying pro-fibrotic barriers to cardiac fates and evaluating hundreds of pairwise transcription factor combinations for neuronal fate, each consisting of one bHLH factor and one POU factor. Overall, Navigo provides a generalizable AI platform for perturbation-effect prediction, disease modeling, and rational cell-fate engineering, advancing toward AI-based virtual embryos for developmental biology and regenerative medicine.

Brain organoids offer an invaluable model system for studying human brain development and disease. However, the establishment of high-fidelity brain organoids with multiple cell lineages including vasculature and immune cells remains a huge challenge. Here, we present a new strategy to generate human cerebral organoids with vasculature and microglia-like cells using genetic code expansion technology (GCE-T) via site-specific protein engineering. The strategy integrates orthogonal genetic translation machinery in hPSCs via PiggyBac transposon system, enabling temporally control of ETV2 expression and endothelial differentiation in hPSC-derived cerebral organoids. The vascularized human cerebral organoids (vhCOs) exhibit coordinated development of multiple cell lineages and blood-brain barrier (BBB) features. Moreover, vhCOs form perfusable vascular network after transplanted in the immune-deficient mice. Single-nucleus RNA sequencing reveals enhanced neurovascular interactions, multi-brain-regional identities, diverse neuronal subtypes and specialized endothelial subclusters in vhCOs, closely resembling human fetal brain. Strikingly, we identify enriched microglia-like cells comprising three distinct subtypes in vhCOs, which contribute to microglia-vascular interactions and synergistically modulate vascular development. Upon Zika virus (ZIKV) infection, vhCOs show neurovascular dysfunction and impaired microglia development, offering new insights into viral-induced neurodevelopmental disorders. This study offers a unique platform for producing more valuable brain organoids with vasculature and immune components, opening a new avenue to advance organoid research and applications.

Understanding how changes in gene function affect disease risk is central to drug development. Genetic variants are natural perturbations of gene activity and provide an opportunity to systematically model dose-response relationships between genes and phenotypes. Here, we present Variant-Informed Dose-Response Analysis (VIDRA), a computational framework that integrates trait-associated variants spanning a spectrum of allele frequencies and functional consequences within a hierarchical Bayesian regression model to systematically infer genetic dose-response relationships. Using 1,607,115 phenotypically associated germline variants available through Open Targets data (including common trait, rare disease, and gene burden associations), we systematically modelled how genetically driven alterations in gene function relate to disease risk, generating 148,350 dose-response-like gene-phenotype relationships across 7,681 phenotypes. Incorporating rare variant information alongside common variants resulted in a 7.23-fold increase in unique phenotypes, and a 51% increase in gene-disease pair associations. We calibrated our model against known drug targets to derive VIDRA Therapeutic Potential Score to rank genes based on their likelihood of succeeding as therapies, and identified 1,860 significant genes, 87% of which are currently therapeutically not targeted. VIDRA framework captures the direction and magnitude of gene-phenotype dependencies, enabling insights beyond target identification. Sixty-two percent of high-scoring targets are predicted to benefit from agonistic modulation, highlighting untapped potential for therapeutic activation. Furthermore, VIDRA framework extends to modelling relationships between genes, intermediate phenotypes, and diseases, enabling identification of biomarker-disease correlations. Applied to blood-cell and cardiovascular traits, VIDRA recovered known genetic links, suggesting a capacity to identify novel biomarker-disease relationships. Lastly, we observed a positive correlation by comparing VIDRA slopes with drug dose-response data from clinical trials, supporting the use of genetic data as a proxy for pharmacological titration. Together, VIDRA framework provides a generalizable, interpretable approach to inform multiple stages of drug development, from target prioritization and therapeutic direction of modulation prediction to biomarker identification, dose guidance, and safety risk assessment.

Immunoglobulins (e.g., IgGs) are critical effectors of the adaptive immune system that when overexpressed or dysregulated can result in autoimmune diseases. Thus, depletion of IgGs can be a promising therapeutic avenue. Here we developed genetically-encoded lysosome targeting chimeras (GELYTACs) that target circulating IgGs for clearance and degradation. The GELYTACs comprised two protein modules derived from insulin-like growth factor 2 (IGF2) and an IgG-binding nanobody, respectively, and mediated clearance of plasma IgG via the lysosomal trafficking receptor IGF2R. To achieve long-lasting IgG depletion, we encoded GELYTACs in an AAV gene therapy vector and established continuous expression in mice. We also developed conditional GELYACs that are activatable with disease-specific proteases or small molecule drugs. This work establishes GELYTACs as a possible therapeutic modality that is deliverable using genetic medicine approaches.

Large-scale single-cell perturbation sequencing provides an unprecedented opportunity to construct virtual cells for the in silico simulation of cellular responses and the inverse design of optimal interventions. However, most perturbation-response models treat cellular responses primarily as mass-preserving shifts in transcriptomic state, whereas single-cell perturbation measurements are inherently unbalanced: the recovered endpoint population is shaped by technical sampling as well as biological perturbation-induced proliferation, apoptosis and selection. Here we introduce U-Pert, an unbalanced generative framework that learns condition- and context-dependent perturbation dynamics from unpaired single-cell snapshots. U-Pert jointly models transcriptomic state transitions and cell-number dynamics, enabling scalable and robust forward prediction of unseen perturbations and contexts, as well as inverse design to screen for desired genetic or pharmacological interventions that achieve user-defined transcriptomic or population-level outcomes. Across controlled simulations, genetic perturbation benchmarks, sciPlex3 drug responses and PBMC cytokine perturbations, U-Pert predicts unseen responses, captures both molecular and abundance changes, and performs inverse design for target gene-expression programs and cell-type compositions. These results show that cell abundance is an integral component of the perturbation phenotype, providing a mass-aware framework for virtual-cell modeling and perturbation cell fate design.