Nature Biomedical Engineering

Despite advances in immunotherapy, most solid tumors remain resistant to treatment. Immune cell engagers redirect cytotoxic lymphocytes against cancer, but limited tumor access, immunosuppressive microenvironments and systemic immune activation limit efficacy. Here we develop live immune modulating engagers (LIME), a modular platform where non-pathogenic, tumor-tropic Escherichia coli display tandem single-chain variable fragments targeting a tumor-associated antigen and an activating receptor on T or natural killer cells. LIME bridged effector and tumor cells, induced transcriptional programs of T cell activation, metabolism and proliferation, and enhanced cytotoxicity across cancer cell lines and patient-derived organoids. In mouse models, LIME safely accumulated in tumors, outperformed tarlatamab in small cell lung cancer, and induced durable immunity in lymphoma. RAS inhibition and PD-L1 blockade enhanced LIME activity in pancreatic cancer and induced humoral responses. Multi-lineage immune modulation remained tumor-confined, without organ toxicity. These findings establish LIME as a versatile living therapeutic platform for programmable, tumor-restricted immune orchestration.

Recent studies show that pallidal and subthalamic local field potentials (LFPs) encode locomotor state and can guide adaptive deep brain stimulation (DBS) for gait impairment in Parkinson's disease. Here, in one participant implanted with the Picostim DyNeuMo-2c, we demonstrate a simpler and more direct approach for inferring locomotor state using the device's onboard accelerometer. Triaxial acceleration was classified independently on each axis to select among preconfigured stimulation programs. Using a cranially mounted digital twin, we characterized inertial signatures across medication and activity states, developed a classifier that distinguished walking from rest while rejecting tremor, and verified the intended stimulation switches during walking. In an exploratory comparison, a gait-adaptive program improved objective gait measures relative to open-loop stimulation optimised for resting tremor. These findings provide a first-in-human demonstration of the feasibility of device-embedded inertial sensing for gait-responsive DBS. They establish a practical framework for further evaluation in larger cohorts.

Scalable human models that capture interactions between distinct tissues remain limited, constraining mechanistic insight and therapeutic prediction. Here, we established a scalable, automation-compatible human neuromuscular organoid (NMO) platform that enables integrated analysis of neuronal and muscle lineages in spinal muscular atrophy (SMA). Patient-derived NMOs reproducibly self-organise into spinal cord and skeletal muscle compartments and form functional neuromuscular circuits. SMA NMOs recapitulate early disease features, including reduced survival motor neuron (SMN) protein levels and impaired neuromuscular junction (NMJ) maturation. Single-nucleus RNA sequencing identifies lineage-specific transcriptional changes across neuronal and muscle compartments preceding functional deficits. Using this platform, we compared two clinically relevant SMN2 splicing modulators and observed distinct, cell-type-dependent responses. While both compounds increased SMN levels and NMJ number, only one enhanced myofiber growth and improved contractile function. These findings highlight muscle maturation, rather than NMJ number alone, as a key determinant of functional recovery and establish NMOs as a scalable system for studying cell-type-specific therapeutic responses.

Deep brain stimulation (DBS) is transforming from a static therapy toward adaptive systems that adjust stimulation based on neural biomarkers. However, the detection of reliable biomarkers that capture the multi-dimensional nature of complex symptoms is often challenging. Here we demonstrate volitional DBS (vDBS)--a paradigm in which patients use brain-computer interface (BCI) training to learn self-regulation of a neural signal that then controls closed-loop DBS. Two patients with Parkinson's disease implanted with sensing-enabled neurostimulators completed chronic, at-home BCI training by playing an airplane simulation game. Through training, they were able to effectively down-regulate their cortical beta signal (p's < 1e-10), represented as the real-time position of a plane in the BCI game. Following training, this cortical beta signal served as the input to a closed-loop DBS algorithm. By modulating their beta signal to cross personalized thresholds, patients voluntarily increased or decreased neurostimulation amplitude at will, in the absence of physical movement (p's < 1e-10). This proof-of-principle demonstration establishes that volitional control of intracranial neurostimulation is achievable without the need of an externalized manual controller. BCI-vDBS could potentially be used for a range of neuropsychiatric conditions and brain rehabilitation to support personalized control of neurostimulation.

Inflammatory bowel disease (IBD) is a heterogeneous chronic disease driven by dysregulated mucosal immunity and impaired epithelial barrier function. Although biologics have improved disease management, they are frequently associated with systemic immunosuppression and adverse effects, highlighting the need for localized therapeutic strategies that both control inflammation and promote tissue repair. Here, we developed a protein bispecific termed 7A2-IgG4-IL22, composed of a human IgG4-Fc domain displaying an antagonistic anti-human MAdCAM-1 single chain (sc)-Fv and a human interleukin (IL-)22. The anti-MAdCAM-1 scFv retained the functional activity of the parental monoclonal antibody, inhibiting T cell activation, expansion and differentiation from naive precursors. Blockade of the MAdCAM-1 signaling axis also reduced production of pro-inflammatory cytokines relevant to IBD pathogenesis, including IFN{gamma} and TNF. On the epithelial side, the IL-22 cargo induces robust signaling in epithelial cells, promoting the expression of IL-22 response genes associated with antimicrobial defense, mucosal homeostasis, as well as IL-10 and CXCL1 expression. This effect contributes to immune cell trafficking to the intestinal mucosa. Together, this bispecific provides a localized dual-mechanism strategy for restoring intestinal immune homeostasis.

How ecological architectures within the gut microbiome convert complex inputs into specific host physiological outcomes remains poorly understood. We used CDD-2101, a multi-component botanical drug operating under an FDA (U.S. Food and Drug Administration) Investigational New Drug program, as a defined ecological perturbation in functional constipation (FC). Integrating a randomized, double-blind, placebo-controlled clinical trial with genome-resolved metagenomics, targeted metabolomics, staged prediction modeling, and receptor-level validation, we show that clinical efficacy of CDD-2101 depends on remodeling a function-specific substructure of the stable Two Competing Guilds (TCG) architecture. We term this substructure the FC-TCG, demonstrate its role along the gut-motility axis, and confirm its effect in three independent gut hypomotility cohorts. The two guilds responded asymmetrically: the intervention selectively suppressed the C1B guild (the pathobiont guild) while largely sparing the C1A guild, the foundation guild that anchors the core gut community, restoring its ecological dominance, producing a coordinated metabolic shift that elevates lithocholic acid and propionic acid. Through gnotobiotic transplantation and receptor antagonism, we demonstrate that lithocholic acid and propionic acid restore gut motility via concurrent engagement of Takeda G protein-coupled receptor 5 (TGR5) and G-protein coupled receptor 43 (GPR43). These findings identify microbial guild architecture as a function-resolved signal-transducing layer that converts multi-component botanical intervention into multi-receptor-mediated gut motility restoration, reframing the gut microbiome from a compositional system into a structural transducer between complex environmental inputs and host physiology.

Only about 10 percent of drugs that clear animal testing succeed in humans, and central nervous system programs carry an even steeper translational gap. Human-relevant NAMs are gaining global regulatory and funding support, creating an urgent need for interpretable preclinical systems that can generate comparable, decision-ready evidence across assays, models, and species. Yet the multimodal treatment-response data produced by these systems are still evaluated assay by assay, with no unified metric showing whether a compound moves neural tissue toward a desirable or undesirable state. Here we present the Neural Impact Score (NIS), a framework that translates multimodal CNS drug-response data into a bidirectional score across four predefined biological categories: neurodevelopment, neuroinflammation, neurodegeneration, and longevity. A positive score indicates a desirable shift, whereas a negative score indicates the opposite, placing compounds on a single scale across assays, model systems, and species. To demonstrate NIS, we analyzed a vascularized human day-200 multi-region brain organoid (MRBO) composed of cortical, endothelial, and brainstem lineages and mimicking a mid-gestational cortical window (GW18-GW22). We tested five compounds with distinct mechanisms of action: a glucagon-like peptide-1 receptor agonist (GLP-1RA), a norepinephrine-dopamine reuptake inhibitor (NDRI), a selective serotonin reuptake inhibitor (SSRI), a sphingosine-1-phosphate receptor modulator, and an Akt activator. We profiled responses using single-nucleus RNA sequencing, bulk RNA sequencing, proteomics, and multi-electrode array electrophysiology. NIS integrated these readouts into category-specific and composite scores, separating beneficial from adverse effects for each compound and sorting compounds into interpretation tiers. Applying the same framework to independent human and rodent datasets without retraining, we recovered conserved human antidepressant responses despite near-chance gene-level agreement between human MRBO and rat brain, and identified an endothelial-dependent, human-specific GLP-1 response absent from the murine dorsal vagal complex. NIS therefore provides a human-relevant framework for drug evaluation and cross-species benchmarking, with a design extensible to other neural systems.

Neurodegenerative diseases are increasingly associated with vascular dysfunction beyond progressive neuronal degeneration, yet how vascular pathology contributes to disease progression remains poorly understood, largely due to the lack of a neurodegenerative disease model capable of capturing neuronal pathology alongside associated vascular dysfunction. Here, we developed a microengineered 3D vascularized brain tissue model that integrates neurospheroids with a self-assembled, perfusable vascular network to recapitulate key features of the neurovascular interface. Using this model, we investigated the vascular contribution to Parkinson's disease pathology by introducing -synuclein preformed fibrils into the engineered vasculature. Intravascular -syn fibril exposure induced endothelial barrier disruption, vascular leakage, inflammation, and vascular regression. Notably, this vascular insult was accompanied by intraneuronal -synuclein aggregation within neurospheroids, suggesting that vascular dysfunction may facilitate the exposure of neural tissue to pathogenic -synuclein. Our neurodegenerative disease modeling approach establishes a versatile and tractable platform for investigating vascular contributions to neurodegenerative disease progression.

Single-cell and spatial omics offer unprecedented opportunities to decipher the mechanisms of disease, however, this process requires teams of experts, iterative trial-and-error and reasoning across modalities. Here we present LungChat (https://chat.lungmap.net), a conversational system for integrated multi-omics analysis and biomedical discovery, deployed as a hierarchical multi-agent architecture in which a supervisor decomposes natural-language questions into parallel, tool-grounded tasks spanning single-cell and spatial analyses, literature and clinical-trial synthesis, and drug repurposing. To predict new therapeutics, LungChat implements Direction-Aware Repurposing and Targeting (DART) to distinguish perturbations that reverse disease transcriptional programs from those that reinforce them, at the cell-type level, for safety prediction. Controlled architecture ablations showed that hierarchical orchestration improved grounded abstention and token efficiency and preserved strong performance on complex multi-step tasks. In pulmonary disease case studies, LungChat independently prioritized saracatinib for IPF through drug-connectivity screening, followed by DART-based cell-type analysis; the same compound has been evaluated in the STOP-IPF clinical trial (NCT04598919). The system also recovered fluticasone propionate, an established COPD therapy, through a single orchestrated analysis. This tissue-agnostic system provides a blueprint for verifiable agentic AI systems that support reproducible scientific discovery.

Deep brain stimulation is an established treatment for Parkinson's disease but does not adapt to dynamic changes in brain state. Here, in four patients with sensing-enabled DBS systems, we evaluated a movement-responsive DBS (mDBS) paradigm that modulated subthalamic stimulation based on volitional motion decoded from cortical activity. During structured motor tasks, mDBS improved average forearm speed and mitigated the progressive bradykinetic slowing observed under constant-amplitude DBS (cDBS), accompanied by a cumulative increase in sensorimotor cortical beta activity and connectivity. In unconstrained, daily activities, mDBS lowered average bradykinesia severity and demonstrated progressive symptom reduction over hours of therapy, which gradually reversed upon switching to cDBS. These findings highlight the enhanced therapeutic benefit of mDBS and its potential to reinforce functional motor circuits in disorders of movement.

Antimicrobial peptides (AMPs) often act against multiple pathogen classes, making multi-label activity prediction a more realistic screening target than binary antimicrobial classification. The ESCAPE benchmark formalizes this setting, but leading approaches typically rely on multimodal, structure-conditioned deep models that are costly to train and tune. We show that a simple, sequence-only pipeline can match and surpass these methods by combining 330 interpretable sequence descriptors with TabPFN, a tabular foundation model that performs in-context prediction in a single forward pass without gradient-based training or hyperparameter search. On ESCAPE (82,359 peptides; five labels), a label-powerset TabPFN model achieves mAP-5=77.8%, improving on the previously best reported 72.1%. A probabilistic classifier chain is the first method to match or exceed the best published average precision on each of the five labels simultaneously. The gains persist under the prior state-of-the-art single-fold training protocol, indicating they are not a training-set-size artefact, and are largest for remote homologues (+11.2 points below 30% sequence identity). Ablations further show that predicted structure is unnecessary at inference and that performance is not driven by any single descriptor family: ten global physicochemical scalars recover 91% of full-feature performance. Finally, explicitly modelling label dependence yields targeted benefits for scarce activities and supports ranking which activity to assay next from partial positive evidence.

Photoacoustic (PA) and ultrasound (US) imaging provide complementary molecular, functional, and anatomical contrasts. Here, we present a panoramic PA-US imaging platform that integrates multispectral PA computed tomography (PACT) along with reflection-mode and transmission-mode US imaging through a single shared full-ring ultrasound array. We employ an ultrafast planewave transmission scheme in reflection-mode US for power Doppler (PWD) imaging and ultrasound localization microscopy (ULM). Additionally, we use the transmission-mode US to reconstruct a spatially resolved speed of sound (SoS) map that corrects both PA and US reconstruction. Such correction sharpens the resolution of PACT, suppresses the artifacts of PWD, and improves microbubble localization of ULM. Elevational scanning further enables whole-body volumetric imaging with co-registered PA and US contrasts. The integrated system maps photoswitchable DrBphP1-expressing tumors alongside their blood perfusion and oxygenation environment. Applying the platform to monitor unilateral renal ischemia-reperfusion injury, we report that microvascular perfusion and renal oxygenation recover at different rates. Collectively, we demonstrate that the integrated PA-US imaging platform provides a unified framework for multiparametric study of anatomy, perfusion, microvascular flow, oxygenation, and molecular activities.

Synthetic data generation is increasingly used to enable data sharing and secondary analysis while protecting participant privacy, particularly for longitudinal tabular health data, where repeated measures per subject create within-subject dependence that most synthetic data methods are not designed to preserve. Existing generative methods, particularly generative adversarial network (GAN)-based approaches, can model complex distributions, but their estimated dependence structures are often difficult to interpret and their performance may be unstable or prone to overfitting in modestly sized datasets. Here we show that eCDF-copula, a statistically rooted approach using the empirical cumulative distribution function (eCDF) and copula modeling, preserves within- and between-visit dependence structure. To handle pervasive missing data, we propose a two-stage strategy combining multiple imputation with copula-based synthesis, enabling a variance decomposition that quantifies replication variability across methods. We benchmarked the proposed approach against four established methods on two longitudinal clinical datasets spanning markedly different sample sizes (n = 120 vs. n = 3, 612). eCDF-copula achieved resemblance and utility exceeding those of state-of-the-art synthetic data methods, while maintaining comparable privacy.

Genomic recording enables transient biological signals to be indelibly captured through DNA alterations, creating a permanent record of cellular history retrievable by sequencing. However, current methods are limited by scarce writing space, typically targeting only one or a few amenable genomic sites and requiring large cell populations for signal reconstruction. Here, we establish Repeats for Genomic Recording (RGRs): sequences with up to 400 copies targetable by a single CRISPR guide RNA, readable with a common primer pair, and predicted to have minimal functional impact. We demonstrate that RGRs enable both signal deconvolution in single cells and high-resolution recording in cell populations. Individual RGR sites exhibit distinct response kinetics; thus, combining them improves recording resolution beyond what redundancy alone provides, analogous to diversity reception in wireless communication. We develop a computational pipeline for systematic RGR identification, revealing 15,000 to 25,000 candidates per species across human, mouse, and zebrafish, thereby markedly expanding recording capacity and enabling cell-type-specific applications. Finally, we validate RGRs in live mice by recording long-term immediate early gene activity across the brain following epilepsy induction. This work establishes genomic repeats as a high-capacity platform for single-cell molecular recording in vivo.

The efficacy of engineered T cell therapies in solid tumors remains limited by T cell dysfunction, driven by complex processes that cannot be easily manipulated via genetic knockouts or overexpression of individual genes. Protein design can create new biological functions that can rewire these consequential cell fate decisions. Here, we introduce OUTLAST Regulators, designed proteins that reprogram critical T cell signaling pathways to enhance functional persistence. These proteins are capable of regulating diverse groups of proteins such as the NR4A family of pro-exhaustion transcription factors, E3 ligases Cbl-b and c-Cbl, and SOCS family proteins. Our designs markedly improve CAR-T and TCR-T performance in vitro and in vivo in stringent solid tumor preclinical models. OUTLAST Regulators are implemented as compact genetic modules compatible with standard viral vectors and cell therapy manufacturing processes, creating a powerful platform for programming new functions into enhanced cell and gene therapies.

Constructing interpretable disease models from longitudinal omics data is a central challenge in precision medicine. The goal is a low-dimensional representation in which a patient's position encodes their molecular state and clinical severity, and along which disease progression can be read directly. Existing dimensionality reduction methods (e.g., UMAP, Variational Autoencoders) fall short of this goal: they optimize a single generic objective and are blind to clinical labels and to the temporal ordering of measurements. Consequently, trajectory inference is typically applied after the fact to an embedding that was never optimized to reveal progression, decoupling the representation from disease dynamics. Manifold learning offers a natural route to such representations, and we build on Locally Linear Embedding (LLE) to preserve the local geometry of the omics data (i.e., keeping molecularly similar patients close together in the low-dimensional space). Geometry alone, however, yields a space that is faithful to molecular similarity yet uninformative about clinical severity and progression. We therefore recast the problem as multi-task learning: MT-LLE jointly optimizes five objectives: geometric reconstruction, supervised organization by clinical stage, embedding and phenotype forecasting, and clustering. Because naively combining such heterogeneous objectives induces gradient conflicts that distort the molecular geometry, an embedded reinforcement learning agent dynamically schedules their weights during training, establishing global geometry before refining clinical boundaries. Across two independent Chronic Obstructive Pulmonary Disease (COPD) cohorts (SPIROMICS and COPDGene), MT-LLE deliberately relaxes exact geometric reconstruction, by a modest margin, in exchange for substantial gains in clinical structure. On held-out patients, a linear model reads disease severity (GOLD stage, 0--4) from the MT-LLE embedding 35--40\% more accurately than from standard dimensionality reduction (0.53 vs.\ 0.38 F1-Macro). The gap is starker for progression: forecasting a patient's next-visit severity from their trajectory reaches 0.38 F1-Macro, while unsupervised baselines sit near zero (0.06--0.09 F1-Macro), a temporal signal those methods fail to capture. To test whether the reinforcement learning agent earns its place, we compared it against a fixed schedule that imposes the same ordering of objectives but cannot adapt during training; the learned agent outperforms it by 13--20\% across clinical metrics, showing the gains come from adapting the weights to how training unfolds, not merely from ordering the objectives correctly, and at no cost to geometric fidelity. Beyond these quantitative gains, the manifold supports complementary analyses that surface structure invisible to standard staging: static phenotyping isolates subjects with active molecular pathology despite preserved lung function; trajectory inference maps two mechanistically distinct progression axes (inflammatory fibrosis and pan-immune activation); and kinematic analysis of each patient's speed and acceleration identifies subjects whose molecular trajectories accelerate ahead of detectable spirometric decline.

Accurate long-term forecasting of cognitive trajectories across the Alzheimer's disease continuum is essential for early intervention, personalized prognosis, patient stratification, and clinical trial enrichment. Despite the promising predictive performance of recent longitudinal forecasting methods, they remain largely data-driven, struggle with irregularly sampled, incomplete longitudinal data and often neglect established disease biology, leading to biologically implausible trajectories. To address this, we propose a biologically constrained continuous-time framework for long-horizon cognition forecasting from limited baseline observations. The proposed method models the complete amyloid-tau-vascular-neurodegeneration-cognition (ATVNC) cascade using hierarchical Neural ODEs with biologically motivated monotonicity constraints. Each pathological stream is governed by a dedicated Neural ODE initialized from irregular longitudinal observations using a GRU-D encoder, capturing intrinsic disease evolution while being modulated by directed upstream pathological influences. A bounded cognition readout ensures physiologically valid cognitive score (MoCA) predictions, while teacher-student knowledge distillation improves learning from sparse longitudinal supervision. Evaluated on the ADNI dataset, the proposed framework achieves a long-horizon extrapolation MAE of 2.06 on 188 held-out participants while eliminating biologically implausible trajectory violations. It further demonstrates robust zero-shot cross-cohort generalization on OASIS-3 (MAE 2.68 on 300 participants), with fine-tuning improving MAE to 1.90. The model also supports prognostic enrichment for Alzheimer's clinical trials, achieving up to 2.70x enrichment over the cohort base rate. These results demonstrate that embedding biological disease mechanisms within continuous-time deep learning improves the accuracy, biological plausibility, and clinical utility of long-horizon cognitive forecasting. The code is publicly available at: https://github.com/PonDeepika/BEACON.

Spatial transcriptomics now profiles patient cohorts at single-cell resolution, enabling analysis of disease-associated cell organization in situ. However, discovering such spatial biomarkers remains challenging because relevant structures occur at unknown scales and cell-or niche-level annotations are rarely available. We present spHOT, a deep learning framework that localizes phenotype-associated spatial biomarkers from sample-level labels. spHOT combines spatial foundation model embeddings, a hierarchical domain tree for multi-resolution tissue representation, and a teacher-student multiple instance learning architecture that converts sample labels into cell-level biomarker scores. In controlled simulations and real-tissue benchmarks, spHOT outperformed existing spatial and single-cell methods in localizing ground-truth biomarkers. Across fibrotic, metabolic, and autoimmune disease datasets, spHOT recovered disease-relevant niches and tissue states reported by supervised analyses in the original studies. Cross-disease application of spHOT transferred biomarkers across chronic lung diseases without retraining. spHOT enables scalable, annotation-efficient spatial biomarker discovery in cohort-scale spatial transcriptomics.

Transcriptome-wide association studies (TWAS) can identify genes where genetically predicted gene expression is associated with disease risk, but translating those signals into therapeutic opportunities remains time-consuming, manual, and difficult to reproduce. We developed TRACE (TWAS-driven Repurposing through AI-assisted Curation of Evidence), a gene- and phenotype-agnostic computational pipeline that accepts a TWAS gene and effect-size direction, normalizes the gene symbol, retrieves FDA-approved drug-gene candidates from four online resources, collects related peer-reviewed literature from PubMed, and uses a fine-tuned biomedical language model to classify whether the literature supports a direct drug-gene relationship, the mechanism of action, and the direction of effect. The pipeline then compares the drug-derived direction with the direction implied by the TWAS effect estimate to rank candidate therapeutic pairs and flag potential drug safety concerns. The local classifier, built on BiomedBERT, was trained using pipeline-derived labels, BioCreative VI ChemProt gold-standard chemical-protein relation examples, and author-reviewed active-learning cases, reaching a held-out macro F1 of 0.809 across three simultaneous classification tasks. We validated the pipeline against a manually curated endometriosis gold standard of 43 drug-gene pairs spanning six TWAS-identified genes, developed through S-PrediXcan analysis of endometriosis GWAS summary statistics, manual querying of four drug-gene interaction databases for each gene, literature review of drug-gene mechanistic evidence, and Mendelian randomization validation of candidate pairs. External validation used two independently published genetically informed drug-repurposing studies in metabolic dysfunction-associated steatotic liver disease (MASLD) and type 2 diabetes (T2D). The pipeline recovered 90.7% of endometriosis pairs, 88.2% of MASLD pairs, and 92.9% of T2D pairs that were present in at least one queried database. Applied to 99 endometriosis-associated TWAS genes, the pipeline identified 1,089 FDA-approved drug-gene pairs, 32 candidate therapeutic pairs, and 77 potential safety concerns, including independent recovery of leuprolide acetate, an established endometriosis therapy. This framework provides a scalable, literature-grounded bridge from TWAS discovery to prioritized therapeutic hypotheses, while preserving uncertainty through manual-review flags and requiring downstream Mendelian randomization, electronic health record-based validation, and experimental follow-up before clinical interpretation.

Activation of the STING pathway can induce potent antitumor immunity, but effective delivery of STING agonists to the tumor while limiting systemic exposure remains challenging. We previously developed MUSIC, an ultrasound-guided platform that uses microbubbles (MBs) to deliver the STING agonist 2'3'-cGAMP and locally activate antitumor immunity. However, the vascular confinement of MBs and the need for intratumoral administration limit the potential for systemic tumor targeting. To overcome these limitations, we developed SONATA (Systemic Oncotherapy using Nanobubbles for Acoustically-guided Tumor Activation), which employs nanobubbles (NBs) that are approximately 10-fold smaller than conventional MBs, enabling systemic administration and tumor extravasation. Following NB accumulation within tumors, ultrasound exposure triggers localized cGAMP release, facilitating delivery to targeted CD11b+ antigen-presenting cells (APCs) and STING activation with spatial and temporal control. NBs are composed of the same components as MBs, including phospholipid shells and a perfluorobutane core and are functionalized with anti-CD11b antibodies to target CD11b+ APCs and spermine-modified dextran to stably load cGAMP through nanocomplex formation. Upon ultrasound activation, SONATA induced phosphorylation of STING, TBK1, and IRF3 and increased IFN-{beta} production in bone marrow-derived macrophages. In an orthotopic breast cancer model, intravenously administered SONATA combined with tumor-localized ultrasound significantly inhibited tumor growth compared with controls. Furthermore, SONATA synergized with immune checkpoint blockade prolonged the median survival of tumor-bearing mice. Collectively, these findings establish SONATA as a systemically administered immunotherapy platform that enables ultrasound-guided, spatially controlled STING activation.