The intricate topology of vascular networks and the complex functions of vessel-rich tissues are challenging to reconstruct in vitro. Here we report the development of: in vitro pathological models of erectile dysfunction and Peyronie's disease; a model of the penis that includes the glans and the corpus spongiosum with urethral structures; and an implantable model of the corpus cavernosum, whose complex vascular network is critical for erectile function, via the vein-occlusion effect. Specifically, we 3D printed a hydrogel-based corpus cavernosum incorporating a strain-limiting tunica albuginea that can be engorged with blood through vein occlusion. In corpus cavernosum defects in rabbits and pigs, implantation of the 3D-printed tissue seeded with endothelial cells restored normal erectile function on electrical stimulation of the cavernous nerves as well as spontaneous erectile function within a few weeks of implantation, which allowed the animals to mate and reproduce. Our findings support the further development of 3D-printed blood-vessel-rich functional organs for transplantation.

Foundation models are pretrained on massive datasets. However, collecting medical datasets is expensive and time-consuming, and raises privacy concerns. Here we show that synthetic data generated via conditioning with disease labels can be leveraged for building high-performing medical foundation models. We pretrained a retinal foundation model, first with approximately one million synthetic retinal images with physiological structures and feature distribution consistent with real counterparts, and then with only 16.7% of the 904,170 real-world colour fundus photography images required in a recently reported retinal foundation model (RETFound). The data-efficient model performed as well or better than RETFound across nine public datasets and four diagnostic tasks; and for diabetic-retinopathy grading, it used only 40% of the expert-annotated training data used by RETFound. We also support the generalizability of the data-efficient strategy by building a classifier for the detection of tuberculosis on chest X-ray images. The text-conditioned generation of synthetic data may enhance the performance and generalization of medical foundation models.

Reactive oxygen and nitrogen species (RONS) contribute to the pathogenesis of neurodegeneration, but the inability to detect RONS in vivo in the central nervous system has confounded the interpretation of results of clinical trials of antioxidants. Here we report the synthesis and characterization of a positron emission tomography (PET) probe, [F]fluoroedaravone ([F]FEDV), for the in vivo quantification of oxidative stress. Derived from the antioxidant edaravone, the probe can diffuse through the blood-brain barrier and is stable in human plasma. In mice, PET imaging with [F]FEDV allowed for the detection of RONS after intrastriatal injection of sodium nitroprusside, in the middle cerebral artery after stroke by photothrombosis, and in brains with tauopathy. When using dynamic PET imaging coupled with parametric mapping, the sensitivity of [F]FEDV-PET to RONS allowed for the detection of increased oxidative stress. [F]FEDV-PET could be used to quantify RONS longitudinally in vivo and to assess the results of clinical studies of antioxidants.

Most antibodies for treating COVID-19 rely on binding the receptor-binding domain (RBD) of SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2). However, Omicron and its sub-lineages, as well as other heavily mutated variants, have rendered many neutralizing antibodies ineffective. Here we show that antibodies with enhanced resistance to the evolution of SARS-CoV-2 can be identified via deep mutational learning. We constructed a library of full-length RBDs of Omicron BA.1 with high mutational distance and screened it for binding to the angiotensin-converting-enzyme-2 receptor and to neutralizing antibodies. After deep-sequencing the library, we used the data to train ensemble deep-learning models for the prediction of the binding and escape of a panel of eight therapeutic antibody candidates targeting a diverse range of RBD epitopes. By using in silico evolution to assess antibody breadth via the prediction of the binding and escape of the antibodies to millions of Omicron sequences, we found combinations of two antibodies with enhanced and complementary resistance to viral evolution. Deep learning may enable the development of therapeutic antibodies that remain effective against future SARS-CoV-2 variants.

Cancer progression can be slowed down or halted via the activation of either endogenous or engineered T cells and their infiltration of the tumour microenvironment. Here we describe a deep-learning model that uses large-scale spatial proteomic profiles of tumours to generate minimal tumour perturbations that boost T-cell infiltration. The model integrates a counterfactual optimization strategy for the generation of the perturbations with the prediction of T-cell infiltration as a self-supervised machine learning problem. We applied the model to 368 samples of metastatic melanoma and colorectal cancer assayed using 40-plex imaging mass cytometry, and discovered cohort-dependent combinatorial perturbations (CXCL9, CXCL10, CCL22 and CCL18 for melanoma, and CXCR4, PD-1, PD-L1 and CYR61 for colorectal cancer) that support T-cell infiltration across patient cohorts, as confirmed via in vitro experiments. Leveraging counterfactual-based predictions of spatial omics data may aid the design of cancer therapeutics.

In serology, each sample is typically tested individually, one antigen at a time. This is costly and time consuming. Serology techniques should ideally allow recurrent measurements in parallel in small sample volumes and be inexpensive and fast. Here we show that mass cytometry can be used to scale up multiplexed serology testing by leveraging polystyrene beads uniformly loaded with combinations of stable isotopes. We generated 18,480 unique isotopically barcoded beads to simultaneously detect, in a single tube with 924 serum samples, the levels of immunoglobulins G and M against 19 proteins from SARS-CoV-2 (a total of 36,960 tests in 400 nl of sample volume and 30 μl of reaction volume). As a rapid, high-throughput and cost-effective technique, serology by mass cytometry may contribute to the effective management of public health emergencies originating from infectious diseases.

Patient responses to immune checkpoint inhibitors can be influenced by the gastrointestinal microbiome. Mouse models can be used to study microbiome-host crosstalk, yet their utility is constrained by substantial anatomical, functional, immunological and microbial differences between mice and humans. Here we show that a gut-on-a-chip system mimicking the architecture and functionality of the human intestine by including faecal microbiome and peristaltic-like movements recapitulates microbiome-host interactions and predicts responses to immune checkpoint inhibitors in patients with melanoma. The system is composed of a vascular channel seeded with human microvascular endothelial cells and an intestinal channel with intestinal organoids derived from human induced pluripotent stem cells, with the two channels separated by a collagen matrix. By incorporating faecal samples from patients with melanoma into the intestinal channel and by performing multiomic analyses, we uncovered epithelium-specific biomarkers and microbial factors that correlate with clinical outcomes in patients with melanoma and that the microbiome of non-responders has a reduced ability to buffer cellular stress and self-renew. The gut-on-a-chip model may help identify prognostic biomarkers and therapeutic targets.

Targeting the delivery of vaccines to dendritic cells (DCs) is challenging. Here we show that, by mimicking the fast and strong antigen processing and presentation that occurs during the rejection of xenotransplanted tissue, xenogeneic cell membrane-derived vesicles exposing tissue-specific antibodies can be leveraged to deliver peptide antigens and mRNA-encoded antigens to DCs. In mice with murine melanoma and murine thymoma, xenogeneic vesicles encapsulating a tumour-derived antigenic peptide or coated on lipid nanoparticles encapsulating an mRNA coding for a tumour antigen elicited potent tumour-specific T-cell responses that inhibited tumour growth. Mice immunized with xenogeneic vesicle-coated lipid nanoparticles encapsulating an mRNA encoding for the spike protein of severe acute respiratory syndrome coronavirus 2 elicited titres of anti-spike receptor-binding domain immunoglobulin G and of neutralizing antibodies that were approximately 32-fold and 6-fold, respectively, those elicited by a commercialized mRNA-lipid nanoparticle vaccine. The advantages of mimicking the biological recognition between immunoglobulin G on xenogeneic vesicles and fragment crystallizable receptors on DCs may justify the assessment of the safety risks of using animal-derived biological products in humans.

The development and wider adoption of adoptive cell therapies is constrained by complex and costly manufacturing processes and by inconsistent efficacy across patients. Here we discuss how microfluidic and other fluidic devices can be implemented at each stage of cell manufacturing for adoptive cell therapies, from the harvesting and isolation of the cells to their editing, culturing and functional selection. We suggest that precise and controllable microfluidic systems can streamline the development of these therapies by offering scalability in cell production, bolstering the efficacy and predictability of the therapies and improving their cost-effectiveness and accessibility for broader populations of patients with cancer.

The difficulty of delivering genes to the kidney has limited the translation of genetic medicines, particularly for the more than 10% of the global population with chronic kidney disease. Here we show that new variants of adeno-associated viruses (AAVs) displaying robust and widespread transduction in the kidneys of mice, pigs and non-human-primates can be obtained by evolving capsid libraries via cross-species cycling in different kidney models. Specifically, the new variants, AAV.k13 and AAV.k20, were enriched from the libraries following sequential intravenous cycling through mouse and pig kidneys, ex vivo cycling in human organoid cultures, and ex vivo machine perfusion in isolated kidneys from rhesus macaques. The two variants transduced murine kidneys following intravenous administration, with selective tropism for proximal tubules, and led to markedly higher transgene expression than parental AAV9 vectors in proximal tubule epithelial cells within human organoid cultures and in autotransplanted pig kidneys. Following ureteral delivery, AAV.k20 efficiently transduced kidneys in pigs and macaques. The AAV.k13 and AAV.k20 variants are promising vectors for therapeutic gene-transfer applications in kidney diseases and transplantation.

Advancing the understanding of the various roles and components of the immune system requires sophisticated methods and technology for the detection of immune cells in their natural states. Recent advancements in the development of molecular probes for optical imaging have paved the way for non-invasive visualization and real-time monitoring of immune responses and functions. Here we discuss recent progress in the development of molecular probes for the selective imaging of specific immune cells. We emphasize the design principles of the probes and their comparative performance when using various optical modalities across disease contexts. We highlight molecular probes for imaging tumour-infiltrating immune cells, and their applications in drug screening and in the prediction of therapeutic outcomes of cancer immunotherapies. We also discuss the use of these probes in visualizing immune cells in atherosclerosis, lung inflammation, allograft rejection and other immune-related conditions, and the translational opportunities and challenges of using optical molecular probes for further understanding of the immune system and disease diagnosis and prognosis.

The efficacy of bacteriophages in treating bacterial infections largely depends on the phages' vitality, which is impaired when they are naturally released from their hosts, as well as by culture media, manufacturing processes and other insults. Here, by wrapping phage-invaded bacteria individually with a polymeric nanoscale coating to preserve the microenvironment on phage-induced bacterial lysis, we show that, compared with naturally released phages, which have severely degraded proteins in their tail, the vitality of phages isolated from polymer-coated bacteria is maintained. Such latent phages could also be better amplified, and they more efficiently bound and lysed bacteria when clearing bacterial biofilms. In mice with bacterially induced enteritis and associated arthritis, latent phages released from orally administered bacteria coated with a polymer that dissolves at neutral pH had higher bioavailability and led to substantially better therapeutic outcomes than the administration of uncoated phages.

Patients undergoing radiation therapy experience debilitating side effects because of toxicity arising from radiation-induced DNA strand breaks in normal peritumoural cells. Here, inspired by the ability of tardigrades to resist extreme radiation through the expression of a damage-suppressor protein that binds to DNA and reduces strand breaks, we show that the local and transient expression of the protein can reduce radiation-induced DNA damage in oral and rectal epithelial tissues (which are commonly affected during radiotherapy for head-and-neck and prostate cancers, respectively). We used ionizable lipid nanoparticles supplemented with biodegradable cationic polymers to enhance the transfection efficiency and delivery of messenger RNA encoding the damage-suppressor protein into buccal and rectal tissues. In mice with orthotopic oral cancer, messenger RNA-based radioprotection of normal tissue preserved the efficacy of radiation therapy. The strategy may be broadly applicable to the protection of healthy tissue from DNA-damaging agents.

The massive parallelization of neuronal intracellular recording, which enables the measurement of synaptic signals across a neuronal network, and thus the mapping and characterization of synaptic connections, is an open challenge, with the state of the art being limited to the mapping of about 300 synaptic connections. Here we report a 4,096 platinum/platinum-black microhole electrode array fabricated on a complementary metal-oxide semiconductor chip for parallel intracellular recording and thus for synaptic-connectivity mapping. The microhole-neuron interface, together with current-clamp electronics in the underlying semiconductor chip, allowed a 90% average intracellular coupling rate in rat neuronal cultures, generating network-wide intracellular-recording data with abundant synaptic signals. From these data, we extracted more than 70,000 plausible synaptic connections among more than 2,000 neurons and catalogued them into electrical synaptic connections and into inhibitory, weak/uneventful excitatory and strong/eventful excitatory chemical synaptic connections, with an estimated overall error rate of about 5%. This scale of synaptic-connectivity mapping and the ability to characterize synaptic connections is a step towards the functional connectivity mapping of large-scale neuronal networks.

Messenger RNA vaccines have shown strong prophylactic efficacy against viral infections. Here we show that antigen-encoding small circular RNAs (circRNAs) loaded in lipid nanoparticles elicit potent and durable T cell responses for robust tumour immunotherapy after subcutaneous injection in mice, particularly when combined with immune checkpoint inhibition. The small circRNA vaccines are highly stable and show low levels of activation of protein kinase R as well as low cytotoxicity, enabling long-lasting antigen translation (longer than 1 week in cells). Relative to large protein-encoding unmodified or modified mRNAs and circRNAs, small circRNA vaccines elicited up to 10-fold antigen-specific T cells in mice and accounted for 30-75% of the total peripheral CD8 T cells over 6 months. Small circRNA vaccines encoding tumour-associated antigens, neoantigens and oncoviral or viral antigens elicited substantial CD8 and CD4 T cell responses in young adult mice and in immunosenescent aged mice. Combined with immune checkpoint inhibition, monovalent and multivalent circRNA vaccines reduced tumour-induced immunosuppression and inhibited poorly immunogenic mouse tumours, including melanoma resistant to immune checkpoint blockade.

High-dimensional multiplexed imaging can reveal the spatial organization of tumour tissues at the molecular level. However, owing to the scale and information complexity of the imaging data, it is challenging to discover and thoroughly characterize the heterogeneity of tumour microenvironments. Here we show that self-supervised representation learning on data from imaging mass cytometry can be leveraged to distinguish morphological differences in tumour microenvironments and to precisely characterize distinct microenvironment signatures. We used self-supervised masked image modelling to train a vision transformer that directly takes high-dimensional multiplexed mass-cytometry images. In contrast with traditional spatial analyses relying on cellular segmentation, the vision transformer is segmentation-free, uses pixel-level information, and retains information on the local morphology and biomarker distribution. By applying the vision transformer to a lung-tumour dataset, we identified and validated a monocytic signature that is associated with poor prognosis.

Gene therapies and cell therapies require precise, reversible and patient-friendly control over the production of therapeutic proteins. Here we present a fully human nitric-oxide-responsive gene-regulation system for the on-demand and localized release of therapeutic proteins through clinically licensed nitroglycerin patches. Designed for simplicity and robust human compatibility, the system incorporates human mitochondrial aldehyde dehydrogenase for converting nitroglycerin into nitric oxide, which then activates soluble guanylate cyclase to produce cyclic guanosine monophosphate, followed by protein kinase G to amplify the signal and to trigger target gene expression. In a proof-of-concept study, human cells expressing the nitroglycerin-responsive system were encapsulated and implanted subcutaneously in obese mice with type 2 diabetes. Transdermal nitroglycerin patches applied over the implant enabled the controlled and reversible production of glucagon-like peptide-1 throughout the 35-day experimental period, effectively restoring blood glucose levels in these mice without affecting heart rate or blood pressure. The approach may facilitate the development of safe, convenient and responsive implantable devices for the sustained delivery of biopharmaceuticals for the management of chronic diseases.