Nature Biomedical Engineering

Timely detection of Barretts esophagus, the pre-malignant condition of esophageal adenocarcinoma, can improve patient survival rates. The Cytosponge-TFF3 test, a non-endoscopic minimally invasive procedure, has been used for diagnosing intestinal metaplasia in Barretts. However, it depends on pathologists assessment of two slides stained with H&E and the immunohistochemical biomarker TFF3. This resource-intensive clinical workflow limits large-scale screening in the at-risk population. Deep learning can improve screening capacity by partly automating Barretts detection, allowing pathologists to prioritize higher risk cases. We propose a deep learning approach for detecting Barretts from routinely stained H&E slides using diagnostic labels, eliminating the need for expensive localized expert annotations. We train and independently validate our approach on two clinical trial datasets, totaling 1,866 patients. We achieve 91.4% and 87.3% AUROCs on discovery and external test datasets for the H&E model, comparable to the TFF3 model. Our proposed semi-automated clinical workflow can reduce pathologists workload to 48% without sacrificing diagnostic performance.

Metastatic pancreatic ductal adenocarcinoma (PDAC) remains deadly, with minimal improvement in prognosis over the past 20 years despite expansion of our chemotherapeutic arsenal. The complex tumor microenvironment (TME) of PDAC in the advanced stage, which often accompany clinical diagnosis, likely contributes to the limited efficacy of current standard of care chemotherapy. Informed by mechanistic preclinical studies, we evaluated the impact of inhibition of Hedgehog (Hh) signaling to prime PDAC TME and leverage anti-tumor efficacy of Gemcitabine plus nab-Paclitaxel (GnP) together with anti-CTLA-4 immune check point inhibitor (ICI, zalifrelimab). Hh inhibition using NLM-001 (an oral small molecule inhibitor of Smo) aimed to polarize the PDAC TME, including cancer associated fibroblasts (CAFs) and intratumoral immune profile, and to foster an immunosuppressive milieu that engages ICI. In this Phase 1b/2, open label, single arm study, patients with metastatic PDAC received standard GnP every 28-day cycles. In addition, NLM-001 was given at 800 mg daily on days -4 to -1 and days 10 to 13 of the GnP cycles 1 to 3, 6 to 8, 11 to 13 onwards (3 cycles on, followed by 2 rest cycles). Anti CTLA-4 inhibitor, zalifrelimab, was administered at 1mg/kg on day 15 of cycle 1 and every 6 weeks thereafter. The primary end point was to assess efficacy by objective response rate (ORR) as per RECIST v1.1. Treatment was overall well tolerated in the 28 patients enrolled. Most frequent grade 3-4 adverse events (AEs) were neutropenia (46.4%), asthenia (21.4%), and neurotoxicity (14.3%). No patient discontinued treatment due to toxicity. ORR was 50% [95% CI, 29.1-70.9] and disease control rate was 95.5% [95% CI, 86.8 - 100.0]. Median progression-free survival (PFS) was 7.3 months 95.5% [95% CI, 5.564 - 9.041] and median overall survival (OS) was 11.5 months [95% CI, 10.23 -12.73]; 1-year PFS was 18.2% [95% CI, 2.1 - 34.3] and 1-year OS was 50% [95% CI, 29.0 - 710]. Patients who achieved ctDNA clearance at cycle 4 had a significant better PFS (10.7 vs 6.0 months; p<0.0001). Paired biopsies immunolabeling and spatial transcriptomic analyses showed polarization of the TME, with increased CD4+ and CD8+ T cells infiltration, down trending Tregs, and decreased SMA/FAP ratio. Hedgehog inhibitor NLM-001 in combination with gemcitabine/nab-paclitaxel and zalifrelimab was safe and well tolerated and showed encouraging objective responses in the first line treatment of advanced PDAC. Clinical Trial RegistrationEudraCT: 2020-004932-52; NCT04827953.

There is a need for better therapies for acute radiation injury (ARI) of the human intestine as current treatments offer limited efficacy. As the ileum is most sensitive to radiation in patients receiving cancer radiation therapy, we created human Organ Chip microfluidic culture models lined by primary patient-derived ileal epithelial cells interfaced with intestinal microvascular endothelium and exposed them to clinically relevant doses of {gamma}-radiation. These Ileum Chips recapitulated key features of ARI, including cell loss, barrier dysfunction, and inflammation, as well as a therapeutic response to a probiotic formulation (VSL#3) that protects against radiation injury in patients. Use of an AI-enabled drug repurposing algorithm (NemoCAD) with transcriptomic data led to the identification of the antifungal agent miconazole as a potential radiation countermeasure drug, and its protective activity was confirmed on-chip. Combination of AI and human Organ Chip studies may offer a powerful way to repurpose drugs for novel disease applications. HighlightsO_LIPrimary human Organ Chips lined by patient-derived ileal epithelial cells interfaced with intestinal microvascular endothelium faithfully recapitulate acute radiation-induced intestinal injury C_LIO_LIUse of the Human Ileum Chip in combination with an AI-based drug repurposing platform led to the identification that the FDA approved antifungal drug miconazole has the potential to be rapidly repurposed as a therapeutic countermeasure against acute radiation injury in the human intestine. C_LI

Accurate polyp segmentation from colonoscopy images is critical for colorectal cancer prevention, yet the generalization of deep learning models under domain shift remains insufficiently explored. We propose Boundary-Explicit Guided Attention U-Net (BEGA-UNet), a boundary-aware segmentation architecture that introduces explicit edge modeling as a structural inductive bias to enhance both segmentation accuracy and cross-domain robustness. The framework integrates three components: an Edge-Guided Module (EGM) with learnable Sobel-initialized operators to capture boundary cues, a Dual-Path Attention (DPA) module that processes channel and spatial attention in parallel, and a Multi-Scale Feature Aggregation (MSFA) module to encode contextual information across multiple receptive fields. Evaluated on the combined Kvasir-SEG and CVC-ClinicDB benchmarks, BEGA-UNet achieves 88.53% Dice and 82.51% IoU, outperforming representative convolutional and transformer-based baselines. More importantly, cross-dataset evaluation demonstrates strong robustness under domain shift, with BEGA-UNet retaining 83.2% of its in-distribution performance--substantially higher than U-Net (64.5%), Attention U-Net (47.5%), and TransUNet (53.1%). In a zero-shot setting on an entirely unseen dataset, the model further maintains 72.6% performance retention. Comprehensive ablation studies indicate that explicit boundary modeling plays a central role in improving generalization, while multi-scale context aggregation further stabilizes performance across domains. Feature distribution analyses support this observation by showing that edge-oriented representations exhibit markedly reduced cross-domain variability compared to appearance-driven features. Overall, BEGA-UNet provides an effective and interpretable solution for robust polyp segmentation, demonstrating that explicit boundary modeling serves as a critical inductive bias for ensuring reliability under clinical domain shifts.

Intracortical brain-computer interfaces (BCIs) can restore communication to people with vocal tract paralysis by decoding cortical activity during attempted speech into text. State-of-the-art systems pairing neural-to-phoneme decoders with phoneme-to-word language models have achieved word error rates (WERs) as low as 1%, but only after collecting thousands of sentences of training data. Shortening the data collection process would facilitate scaling this new technology by reducing the time from device implant to high-accuracy communication. Here we introduce a transformer-based decoder model trained jointly across six intracortical speech BCI participants. For every participant -- regardless of sex, disease etiology, or attempted speaking strategy -- a multi-user model decoded speech more accurately (over 50% lower relative WER on average) than models trained on individual users data. Notably, the multi-user model could be finetuned on fewer than 200 sentences from a held-out user to achieve a WER below 7%. These results reveal how to pool intracortical data across people to yield more accurate, generalizable, and rapidly-deployable decoding models.

1.Deep brain stimulation is a widely used therapy for Parkinsons disease (PD) but currently lacks dynamic responsiveness to changing clinical and neural states. Feedback control has the potential to improve therapeutic effectiveness, but optimal control strategy and additional benefits of "adaptive" neurostimulation are unclear. We implemented adaptive subthalamic nucleus stimulation, controlled by subthalamic or cortical signals, in three PD patients (five hemispheres) during normal daily life. We identified neurophysiological biomarkers of residual motor fluctuations using data-driven analyses of field potentials over a wide frequency range and varying stimulation amplitudes. Narrowband gamma oscillations (65-70 Hz) at either site emerged as the best control signal for sensing during stimulation. A blinded, randomized trial demonstrated improved motor symptoms and quality of life compared to clinically optimized standard stimulation. Our approach highlights the promise of personalized adaptive neurostimulation based on data-driven selection of control signals and may be applied to other neurological disorders.

Achieving durable CAR T cell responses against primary brain tumors and metastases requires strategies that enable intracranial control of therapy to overcome the barriers of solid tumor treatment without compromising safety. Here, we show that closed-loop sonothermogenetics enables remote regulation of CAR T cell therapeutic activity through the intact skull. Using MR-guided focused ultrasound with closed-loop temperature feedback, we modulate CAR T cells engineered with a genetically encoded thermal bioswitch to achieve metronomic activation in the brain without lasting adverse effects on healthy brain tissue. In murine models of brain cancer, metronomic production of NKG2D T cell engagers by intratumoral CAR T cells overcomes antigen heterogeneity in breast cancer brain metastasis and myeloid-derived immunosuppression in glioblastoma to drive antitumor responses. Our findings support the use of closed-loop sonothermogenetics for spatial and temporal control of CAR T cell therapies targeting solid brain tumors.

Cell-based therapies are a major emerging category of medicine. The ability of engineered cells to traffic to and function at specific anatomical locations is a major aspect of their performance. However, there is a lack of non-invasive, non-ionizing, cost-accessible methods to track these therapies inside the body and ensure proper function. Here, we establish a platform for in vivo imaging of primary cell therapies using ultrasound - a ubiquitously accessible technology for high-resolution non-invasive imaging. We introduce and optimize a lentiviral delivery system to express acoustic reporter genes based on gas vesicles in primary mammalian cells such as T cells, showing that this results in robust ultrasound contrast. Additionally, we develop genetic circuits making it possible to monitor T cell activation via activity-dependent promoters. We apply this technology to primary human T cells, using it to non-invasively track their accumulation and proliferation as a targeted therapy in a mouse tumor xenograft model and compare it to invasive, terminal measures such as immunohistology. By making it possible to visualize cell-based therapies and their function inside opaque living organs with unprecedented resolution and accessibility, this technology has the potential to significantly accelerate their development and effective use.

Recent advancements in flexible bioelectronics have enabled continuous, long-term stable interrogation and intervention of biological systems. However, effectively utilizing the interrogated data to modulate biological systems to achieve specific biomedical and biological goals remains a challenge. In this study, we introduce an AI-driven bioelectronics system that integrates tissue-like, flexible bioelectronics with cyber learning algorithms to create a long-term, real-time bidirectional bioelectronic interface with optimized adaptive intelligent modulation (BIO-AIM). When integrated with biological systems as an AI-cyborg system, BIO-AIM continuously adapts and optimizes stimulation parameters based on stable cell state mapping, allowing for real-time, closed-loop feedback through tissue-embedded flexible electrode arrays. Applied to human pluripotent stem cell-derived cardiac organoids, BIO-AIM identifies optimized stimulation conditions that accelerate functional maturation. The effectiveness of this approach is validated through enhanced extracellular spike waveforms, increased conduction velocity, and improved sarcomere organization, outperforming both fixed and no stimulation conditions.

Ultrasound imaging and acoustic reporter genes provide unique capabilities for in vivo biological imaging by leveraging ultrasounds ability to visualize opaque tissues with high spatiotemporal resolution. But until now, the expression of acoustic reporter genes - based on gas vesicle (GV) proteins - has been limited to ex vivo-modified cells due to the complexity of the GV gene cluster, precluding valuable in situ applications. Here, we develop a system capable of introducing GV genes directly into native tissues via stoichiometric multi-AAV delivery. We validate this system in the mouse brain, demonstrating well-tolerated in situ gene expression and repeated ultrasound imaging over more than a month in the same animal. Furthermore, by placing GV genes under the control of immediate early gene promoters, we demonstrate the ability to track in vivo gene expression changes arising from elevated neural activity during epileptic seizures. This work connects ultrasound to in situ transcriptional dynamics happening inside the opaque tissues of living creatures.

We developed an integrated platform combining high-throughput automated biofabrication, systematic patient-derived tissue experiments, and specialized artificial intelligence to enable patient-specific computational "digital twins" for thrombosis prediction. Our automated manufacturing platform fabricates 80 fully assembled, patient-specific vessel-on-chips within 10 hours from clinical imaging--a [~]100-fold improvement over manual methods--achieving sub-micron precision through novel two-stage pneumatic motion control and integrated optical feedback. Using these chips, we systematically captured thrombosis across 491 high-fidelity videos spanning 6 patient-derived vascular geometries, 5 distinct anatomical injury sites, and 14 anticoagulant/antiplatelet interventions, establishing a "physical twin" experimental corpus. We trained CLoT (Cascade Learner of Thrombosis), a conditional video diffusion model efficiently adapted via lightweight Low-Rank Adaptation (LoRA) to generate realistic thrombosis videos conditioned on patient-specific geometry, injury location, and drug treatment. Rigorous benchmarking against state-of-the-art commercial models (Sora, Wan, Kling, Seedance, Hailuo, Hunyuan) reveals CLoT achieves 7.38-fold superior temporal biological consistency and 5.3-fold higher spatial morphological fidelity. Prospective validation on unseen patients demonstrates >90% temporal accuracy. This integrated paradigm--combining automated fabrication with domain-specialized generative AI--establishes proof-of-concept for personalized medicine enabled by digital twins trained on human-derived vascular anatomy, enabling pre-treatment antithrombotic evaluation while providing a replicable template for translating tissue engineering into clinical practice.

Spinal cord injury (SCI) affects millions worldwide, with over half of all cases resulting in tetraplegia, where a complete injury can cause profound motor and sensory loss in all four limbs1. Here, we demonstrate an artificial double neural bypass (DNB) that integrates a bidirectional intracortical brain-computer interface with targeted spinal and brain stimulation to promote restoration of upper limb function in severe, complete paralysis. This hybrid assistive-therapeutic approach restores both hand movement and tactile sensation simultaneously via cortical mediation while promoting significant persistent sensorimotor improvements. The DNB uses a stable nested neural decoding architecture with deep reinforcement learning for fine grasping, along with patterned brain microstimulation ( cortical mirroring) and spinal cord stimulation to promote real-time and long-term functional recovery. Using the DNB, our participant with chronic C4 sensory/C5 motor complete tetraplegia regained the ability to self-feed, grasp delicate objects, and experienced persistent recovery of arm flexion and wrist tactile sensation. These findings represent a major advance in restoring meaningful function after severe, complete SCI, demonstrating that bidirectional neuroprostheses combined with targeted brain and spinal neuromodulation can drive durable sensorimotor recovery and improve independence and quality of life.

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.

Early cancer detection has the potential to reduce cancer mortality, yet endogenous tumor-shed biomarkers lack sensitivity for early-stage disease. We report OncoSCOUT, a cancer detection strategy using T cells engineered with protease-activatable receptors (PARs) that conditionally recognize tumor cells and release a synthetic biomarker for detection in urine. These PARs comprise masked synthetic Notch receptors in which antigen binding is blocked by a peptide mimotope tethered via a protease-cleavable linker. We demonstrate that requiring both extracellular protease activity and tumor antigen recognition improves spatial specificity and minimizes off-tumor activation of PAR T cells in vivo. To identify tumor-selective PARs, we adoptively transferred a HER2-targeted PAR library displaying [~]160,000 unique 4-mer amino acid linkers and discovered multiple variants significantly enriched in a HER2-positive cancer xenograft model. Using a single customized PAR, we show that OncoSCOUT can detect total tumor burdens as small as 10-30 mm3 with significantly improved sensitivity than the protein biomarker CA 15-3 or a 20-plex circulating tumor DNA (ctDNA) assay.

Monitoring neural population activity at single-cell resolution is essential for driving fundamental research and clinical innovations. However, translating advanced recording techniques from animal models to humans remains a significant challenge. Flexible neural electrodes have recently emerged as powerful tools for large-scale single-unit recordings due to their superior biocompatibility and high recording density. Here, we demonstrate reliable, high-density single-unit recordings during intraoperative procedures in human patients using ultra-Flexible Implantable Neural Electrode (uFINE) arrays. The uFINE array exhibited sufficient mechanical robustness to maintain structural integrity throughout surgical operations. We successfully recorded 616 single units from 10 patients, with up to 135 single units simultaneously recorded. The flexibility of uFINE array minimized signal disturbances from brain pulsations, enabling stable and continuous single-unit detection. Stimulus and response tuning were observed at the level of individual neurons in awake patients. This uFINE-based recording approach offers unique opportunities to investigate human-specific cognitive functions and develop next-generation brain-machine interfaces.

Human intestinal organoids (HIOs) derived from human pluripotent stem cells co-differentiate both epithelial and mesenchymal lineages in vitro but lack important cell types such as neurons, endothelial cells, and smooth muscle. Here, we report an in vitro method to derive HIOs with epithelium, mesenchyme, enteric neuroglial populations, endothelial cells, and organized smooth muscle in a single differentiation, without the need for co-culture. When transplanted into a murine host, these populations expand and organize to support organoid maturation and function. Functional experiments demonstrate enteric nervous system function, with HIOs undergoing peristaltic-like contractions, suggesting the development of a functional neuromuscular unit. HIOs also form functional vasculature, demonstrated in vitro using microfluidic devices to introduce vascular-like flow, and in vivo following transplantation, where HIO endothelial cells anastomose with host vasculature. Collectively, we report an in vitro model of the human gut that simultaneously co-differentiates epithelial, stromal, endothelial, neural, and organized muscle populations.

Large-scale clinical genome sequencing yields vast numbers of variants of unknown significance (VUSs). The high frequency of VUSs and the paucity of platforms to characterize their functional impact pose significant challenges for clinical decision making. Here, we present an integrated end-to-end platform, REVi-SCOPE (Rapid evaluation of variants in single cells by optogenetics and prime editing), for characterization of the impact of VUSs on cardiac physiology. Our strategy consists of (1) introduction of variants directly into wild-type (WT) human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) via prime editing; (2) optogenetic assessment of calcium and membrane voltage dynamics in single hiPSC-CMs within the pool of edited and unedited cells; and (3) in situ single-cell genotyping of the phenotyped hiPSC-CMs with single-allele resolution. By optimizing and integrating each of these steps, we created a platform that enables VUS characterization in 10 days. We validated the REVi-SCOPEs capabilities by analyzing the properties of established arrhythmogenic variants. We then used REVi-SCOPE to reveal the functional impact of a VUS, TRPM4A320V, identified in a child with a conduction block. Together, our results show that REVi-SCOPE enables functional characterization of VUSs linked to cardiac arrhythmias with unprecedented throughput.

Personalized antisense oligonucleotides (ASOs) have achieved positive results in the treatment of rare genetic disease. As clinical sequencing technologies continue to advance, the ability to identify rare disease patients harboring pathogenic genetic variants amenable to this therapeutic strategy will likely improve. Here, we describe a scalable platform for generating patient-derived cellular models and demonstrate that these personalized models can be used for preclinical evaluation of patient-specific ASOs. We establish robust protocols for delivery of ASOs to patient-derived organoid models and confirm reversal of disease-associated phenotypes in cardiac organoids derived from a Duchenne muscular dystrophy (DMD) patient harboring a structural deletion in the dystrophin gene amenable to treatment with existing ASO therapeutics. Furthermore, we design novel patient-specific ASOs for two additional DMD patients (siblings) harboring a deep intronic variant in the dystrophin gene that gives rise to a novel splice acceptor site, incorporation of a cryptic exon, and premature transcript termination. We show that treatment of patient-derived cardiac organoids with patient-specific ASOs results in restoration of DMD expression and reversal of disease-associated phenotypes. The approach outlined here provides the foundation for an expedited path towards the design and preclinical evaluation of personalized ASO therapeutics for a broad range of rare diseases.

Bioengineered probiotics enable new opportunities to improve colorectal cancer (CRC) screening, prevention and treatment strategies. Here, we demonstrate the phenomenon of selective, long-term colonization of colorectal adenomas after oral delivery of probiotic E. coli Nissle 1917 (EcN) to a genetically-engineered murine model of CRC predisposition. We show that, after oral administration, adenomas can be monitored over time by recovering EcN from stool. We also demonstrate specific colonization of EcN to solitary neoplastic lesions in an orthotopic murine model of CRC. We then exploit this neoplasia-homing property of EcN to develop early CRC intervention strategies. To detect lesions, we engineer EcN to produce a small molecule, salicylate, and demonstrate that oral delivery of this strain results in significantly increased levels of salicylate in the urine of adenoma-bearing mice, in comparison to healthy controls. We also assess EcN engineered to locally release immunotherapeutics at the neoplastic site. Oral delivery to mice bearing adenomas, reduced adenoma burden by [~]50%, with notable differences in the spatial distribution of T cell populations within diseased and healthy intestinal tissue, suggesting local induction of robust anti-tumor immunity. Together, these results support the use of EcN as an orally-delivered platform to detect disease and treat CRC through its production of screening and therapeutic molecules.

Functional ultrasound imaging (fUSI) is a promising neuroimaging method that infers neural activity by detecting cerebral blood volume changes. It offers high sensitivity and spatial resolution relative to fMRI and is an epidural alternative to electrophysiology for medical and neuroscience applications, including brain-computer interfaces. However, current fUSI methods require hundreds of compounded images and ultrasound pulse emissions, leading to high computational costs, memory demands, and potential probe heating. We propose VARiable Sampling fUSI (VARS-fUSI), the first deep learning fUSI method to allow for different sampling durations and rates during training and inference by using neural operators. VARS-fUSI reconstructs high-quality fUSI images using 10 - 15% of the time or sampling rate needed per image while preserving decodable behavior-correlated signals. Additionally, VARS-fUSI offers efficient finetuning for generalization to new animals and humans. Demonstrated across mouse, monkey, and human data, VARS-fUSI achieves state-of-the-art performance, enhancing imaging efficiency by significantly reducing storage and processing needs.