National Science Review

Hepatocellular carcinoma (HCC) is particularly aggressive and difficult to treat. Due to the lack of early clinical diagnosis and the unsatisfactory clinical treatment effect, it is particularly important to identify novel markers that can predict tumor behavior in HCC. biogenesis of ribosomes BRX1 (BRIX1) is abundant in various tissues of the human body. However, the regulatory mechanisms and its role in various tissues are not fully understood. Here, we analyzed the expression pattern of BRIX1 in HCC from public gene expression databases and tissue samples from clinical HCC. We confirmed that BRIX1 was upregulated in both HCC cell lines and HCC paraffin section samples. BRIX1 depletion significantly dicreased the capacity of cells to grow and migrate in vitro, and knockdown BRIX1 suppressed tumor growth in xenograft tumor model. Mechanistically, BRIX1 depletion suppressed the MAPK/ERK pathway, as reflected by reduced phosphorylated ERK (p-ERK) levels. In summary, we provide a rational clue for the further investigation of BRIX1 as an invaluable biological marker for diagnosing and predicting prognosis of patients with HCC.

SynNotch-CAR circuits improve T cell specificity by coupling antigen recognition to inducible CAR expression. However, basal CAR expression without receptor activation, termed here as leakiness, can reduce the separation between killing of intended target cells and sparing of antigen-positive off-target cells, limiting target-cell discrimination. Here, we systematically quantified basal CAR expression for several synNotch-CAR designs and developed a coupled ordinary differential equation model to show that discrimination depends on basal output, CAR potency, and effector-to-target ratio. We introduced C-terminal tags such as fluorescent proteins, degron domains, endocytosis signals, and endoplasmic reticulum retention motifs as a strategy to reduce CAR leakiness. We found that fluorescent proteins and degron-containing tags reduced basal CAR surface expression while preserving antigen-induced CAR expression, improving discrimination of antigen-density sensing and combinatorial circuits in vitro. In xenograft models, fluorescent protein-tagged CARs improved discrimination by reducing activity against off-target cells while retaining activity against high-antigen tumors. Degron-containing constructs reduced basal CAR expression in vitro but showed suboptimal performance in vivo, revealing a trade-off between basal CAR suppression and induced CAR persistence. Together, these findings demonstrate that basal output expression is a key parameter for inducible genetic circuit designs and establish layered transcriptional and post-translational regulation as a strategy to improve the fidelity of inducible T cell circuits.

Dissecting the complexity of antibody responses in orthopoxvirus (OPXV) infected individuals is essential for elucidating protective mechanisms and identifying candidate protective immunogens. Here, we profiled the acute humoral response in 51 mpox cases, showing distinct IgG trajectories among multiple antigens alongside the rise of plasma neutralizing activities to plateau within 6 weeks after symptom onset. Utilizing a single-cell transcriptomic and BCR sequencing based antigen-agnostic mAb isolation workflow, we further generated monoclonal antibodies (mAbs) from 254 expanded peripheral B cell clones of 3 patients. We discerned 97 specific mAbs recognizing at least 12 different OPXV proteins via integrated screening approaches, which comprised neutralizing antibodies binding unconventional viral targets and antibodies exhibiting extraordinary in vitro and in vivo anti-OPXV effects. The number of OPXV-specific mAbs recovered per donor reflected the percentage of expanded clones among circulating B cells. More interestingly, we demonstrated that the inferred unmutated common ancestors (UCAs) of neutralizing antibody clones did not necessarily react with OPXV, implying that OPXV neutralizing antibodies might frequently originate from B cells previously activated by unknown antigens. Our work establishes an efficient workflow for antigen-agnostic isolation of pathogen specific mAbs and reveals previously unclarified features of antibody responses induced by acute MPXV infection.

Investigating the neural mechanisms of social group interactions and other naturalistic behaviors in small animals remains limited by current technology. Tethered neural recording systems are incompatible with many of these behaviors, while existing wireless devices for small animals are constrained by weight, bandwidth, recording duration, and the lack of closed-loop modulation capabilities. To overcome these limitations, we developed a Wireless, Interactive, Lightweight Datalogger (WILD) with integrated flexible neural probes, optogenetics, an inertial measurement unit, an ultrasonic microphone, and a head-mounted camera. This platform enables simultaneous, long-term recording of neural activity, locomotor variables, vocalizations, and eye movements from groups of freely moving mice in both laboratory and outdoor settings. Model-based real-time signal processing detects specific neural events and behavioral motifs to trigger closed-loop neural interventions. By combining multimodal recordings with advanced onboard signal-processing capabilities in a compact device, WILD enables the investigation of neural mechanisms underlying a broad range of natural behaviors in small animals.

Seasonal influenza A viruses cause significant global morbidity each year. Although vaccination remains the primary preventive strategy, effectiveness is often reduced by antigenic drift. This challenge is particularly pronounced for influenza A(H3N2), which has required eight vaccine updates over the past decade. Here, we present a computational framework to engineer broadly reactive influenza A(H3N2) vaccines, using protein language models to generate novel hemagglutinin (HA) sequences and a machine learning model to predict antigenic distance from circulating strains. In a proof-of-concept study, seven HA candidates designed using sequence data from 2013-2018 were evaluated in mice against contemporary and subsequently circulating viruses. Two candidates elicited protective levels of reactive antibodies, robust H3-specific antibody-secreting cell responses, and cross-neutralization against contemporary clades and drifted 2019-2020 strains. These findings demonstrate that an integrated generation-selection strategy can enhance vaccine coverage across current and future A(H3N2) seasons and may be applicable to other influenza subtypes.

Microglia are central mediators of Alzheimers disease (AD) pathogenesis, yet the mechanisms driving disease-associated microglial states and their therapeutic modulation remain poorly understood. Here, we integrated single-nucleus transcriptomic datasets across the AD spectrum and identified disease- and lipid-associated microglia (DLaM) as a major AD-enriched population linked to genetic risk, neuropathology and cognitive decline. To model this state experimentally, we screened AD-relevant perturbations in human induced pluripotent stem cell (hiPSC)-derived microglia and found that ferric ammonium citrate (FAC) reproducibly induced a DLaM-like state characterized by lipid accumulation, lysosomal dysfunction and impaired A{beta} phagocytosis. Using a transcriptomics-based state-reversion screen, we identified LY2090314 as a potent modulator that restored microglial function and induced a distinct lysosomal-metabolic state. These findings establish a framework for transcriptomic disease-state-guided therapeutic discovery in AD.

Deep learning-based protein structure prediction methods that leverage evolutionary information from multiple sequence alignments (MSAs), exemplified by AlphaFold2, have achieved remarkable accuracy. However, existing methods still struggle to predict challenging proteins, particularly those with novel folds or limited evolutionary information, and to recover alternative conformational states. Here we show that structure prediction models trained under different MSA-depth distributions corresponding to different levels of evolutionary information exhibit complementary generalization behaviors, and that a model trained on a mixture of these distributions can combine their complementary generalization strengths. Building on this insight, we developed ProtMonomer, a deep learning framework trained on MSA-depth distributions representing a broad range of evolutionary information levels to improve structure prediction. Across benchmarks comprising CASP15 targets, non-redundant experimentally determined structures, orphan proteins, and short peptides, ProtMonomer performed comparably to or better than leading methods, including AlphaFold2 and AlphaFold3, with particularly strong performance on challenging targets. For fold-switching proteins, ProtMonomer also recovered alternative conformational states more accurately than AlphaFold2 and AlphaFold3 across diverse homologous sequence sampling strategies. In addition to improving predictive accuracy, ProtMonomer substantially reduced inference cost through an efficient architecture, enabling high-throughput applications. Together, these findings provide insights into the generalization of evolution-informed structure prediction models and support ProtMonomer as an accurate and efficient framework for protein structure prediction.

Individuality is a defining feature of human biology. The functional network architecture of the human brain harbors person-specific qualities and forms individualized connectivity profiles that function as a neural fingerprint, both stable and unique across time. Here, using fMRI data from 431 Human Connectome Project participants, we examined whether neural individuality is more strongly exhibited in brain regions bearing signatures of recent human evolution. We calculated region-wise fingerprinting accuracy and associated it with four properties of evolutionary cortical organization: cortical expansion, myelin content estimate (T1w/T2w), human-specific gene-expression profiles, and functional homology to other primates. Across all four measures, neural individuality was strongest in cortical areas showing greater evolutionary novelty in humans, particularly frontoparietal control and default mode networks, and weaker in more conserved primary regions. Our findings connect evolutionary variation across species with stable functional variation among individuals.

Intrinsic neural timescale (INT) quantifies the persistence of spontaneous neural dynamics and offers a principled metric for characterizing brain-wide temporal organization. Although a hierarchy of INTs has been established during rest, how task engagement reconfigures this organization and how it is constrained by the structural connectome (SC) remain poorly understood. Here, we systematically mapped whole-brain INT using high-resolution fMRI data from the Human Connectome Project during rest and seven tasks spanning working memory, gambling, motor, language, social, relational, and emotion domains. Task engagement induced robust, regionally heterogeneous changes in INT while largely preserving the brain-wide temporal hierarchy across cognitive states. SC-INT coupling remained strong but consistently decreased during tasks, indicating that anatomical architecture continues to constrain INT, although its influence is attenuated under task demands. To investigate these findings mechanistically, we employed a multiscale, whole-brain neuronal-network model, which revealed that INT increase and peak within a broad critical-like regime. Strong SC-INT coupling, as observed empirically, emerged in the subcritical regime, weakened progressively with increasing network excitability, and reversed in the supercritical regime. These results demonstrate that task engagement reconfigures INTs while maintaining their hierarchical organization, suggesting that both resting and task states operate largely within a common subcritical dynamical regime.

To evaluate the cost-utility and 5-year budget impact of first-line olaparib plus abiraterone versus abiraterone alone for metastatic castration-resistant prostate cancer (mCRPC) in China after the eleventh round of volume-based procurement (VBP). The intention-to-treat (ITT) population was assigned primary decision-analytic weight; the prespecified BRCA1/2-mutated (BRCAm) subgroup was a supporting analysis.

Mutations in the E3 Ubiquitin (Ub) ligase RNF216 cause Gordon Holmes syndrome (GHS), a neurodegenerative disorder accompanied by neuroendocrine disruption. We developed an orthogonal ubiquitin transfer (OUT) platform to capture RNF216 substrates in neuronal cells and identified OTUD4, a deubiquitinating enzyme (DUB) mutated in GHS, and FMRP, a neuronal-enriched translational repressor. RNF216 predominantly synthesizes K6-linked Ub chains on OTUD4 to induce its degradation, forming donut-shaped structures in neurons. In return, OTUD4 removes the ubiquitination of RNF216 and FMRP. Analysis of RNF216 substrates revealed biological functions regulating protein synthesis, a shared function of the OTUD4-RNF216 substrate interaction network. Indeed, RNF216 expression increased protein synthesis rates in different cell types while Rnf216 deletion decreased dendritic development in neurons. Overall, our findings show that RNF216 and OTUD4 balance rates of protein synthesis and degradation and suggest GHS-related mutations in RNF216 or OTUD4 may offset this balance, triggering neurodegeneration.

Maintaining a prescribed composition in engineered microbial consortia is difficult because small fitness differences can drive competitive exclusion. We study a two-strain consortium in continuous culture and develop a feedback architecture that regulates composition by selectively slowing the fast strain as a function of the population ratio. At the population level, we derive an idealized ratio-feedback law with a tunable positive coexistence equilibrium. We then propose a biomolecular realization using orthogonal quorum sensing, an sRNA-based ratiometric controller, and a ppGpp-mediated growth actuator. Exploiting the separation between slow population growth and faster intracellular controller dynamics, we use singular perturbation theory to show that, for sufficiently fast controller dynamics, the full implementation model inherits the coexistence equilibrium and its local stability properties from the reduced model. Numerical simulations validate the reduction and show how weaker timescale separation or loss of the assumed molecular regime degrades performance.

Accurate, reproducible interpretation of kidney allograft biopsies is critical for diagnosis of graft injury to guide prognosis and management. The international Banff classification is a consensus diagnostic system based on semiquantitative histological lesion scoring on either extent or severity of kidney transplant biopsies. However, pathologist scoring is limited by substantial interobserver variability, constrained scalability, and the inherent nature of the scoring system itself. Here we present BanffNET, a weakly supervised, probabilistic deep learning framework that combines self-supervised feature extraction with a novel Bayesian multiple-instance learning framework to predict (continuously) the full spectrum of Banff lesion scores directly from whole-slide images (WSIs). Using lesion-specific aggregation functions tailored to localized (modeling lesion severity) and diffuse pathologies (modeling lesion extent), BanffNET generates interpretable, patch-level probability maps and calibrated slide-level scores. BanffNET's performance was assessed relative to consensus, biological correlates of rejection and clinical outcome, demonstrating superior consistency, transportability and generalization. Trained on 7,249 WSIs from three cohorts, BanffNET demonstrates consistent performance on 11,028 WSIs across five external test sets, performing on par or exceeding expert consensus across lesions. BanffNET scores align more closely than pathologist Banff scores with molecular profiles of rejection, offering a transparent, biologically grounded framework for computational pathology with relevance beyond transplantation.

Molecular staging of Alzheimer's disease (AD) increasingly defines transition boundaries along single-cell pseudo-progression trajectories, yet whether such boundaries reproduce across brain regions, cohorts and molecular modalities is rarely tested. We present a permutation-controlled audit that combines nine boundary-detection algorithms with a fixed marker panel and four orthogonal reproducibility axes-algorithmic consensus, region, cohort and modality. On synthetic data with planted ground-truth boundaries the audit reaches 100% sensitivity and 94% specificity, rejecting four distinct artefact classes each by a different axis. Applied to the Seattle Alzheimer's Disease Brain Cell Atlas middle temporal gyrus, it localizes a transition that is robust across algorithms and recovered in most cell types but does not generalize: its leading marker is attenuated or absent in prefrontal cortex, entorhinal cortex and cerebrospinal fluid, and an apparent cross-region conservation of glial metabolic genes proves to be a global-expression offset rather than a shared program. The same audit nonetheless certifies an externally validated marker (astrocytic PTGDS) as reproducible across regions and modalities, showing that it separates generalizable anchors from dataset-specific ones rather than rejecting all signals. We provide this four-axis audit as a transferable, code-available standard to apply before a trajectory boundary is read as a biological stage, in AD and other progressive proteinopathies.

Klebsiella pneumoniae (Kp) is a WHO high-priority pathogen for vaccine development, yet previous efforts failed largely because key protective immune mechanisms remain unclear. Here we show that protective immunity conferred by mucosal mRNA vaccines (but not parenteral) require neither serum IgG nor airway secretory IgA, but instead depends on a previously unrecognized lung-resident CD8IL-17 T-cells (Tc17) that rapidly recruits neutrophils/macrophages to eliminate bacteria. To therapeutically harness this paradigm, we developed INSPIRE, a machine learning-engineered exosome platform incorporating donor-screened, miRNA-bioactive backbones (miR-21-mediated airway barrier penetration and miR-155-associated dendritic-cell activation through SOCS1/Inpp5d axis) and computationally designed peptides that boosts 11.6-fold mRNA encapsulation and 3-fold dendritic-cell cross-presentation. Intranasal INSPIRE-mRNA vaccination confers near-complete protection against clinically relevant Kp strains while intramuscular counterparts fail (below ~30% survival). Leveraging pIgR-/- and IL-17-/- mice coupled with T-cell depletions, we demonstrate the protection is Tc17-dependent. This work overturns the antibody-centric dogma and redefines a non-canonical Tc17-correlate for extracellular bacterial pneumonia.

Autosomal dominant retinitis pigmentosa (adRP) caused by RHO mutations is a leading form of inherited retinal degeneration. Extensive allelic heterogeneity of RHO pathogenic variants limits the translational applicability of mutation-specific gene therapies. To address this, we developed SNARE (SNP-guided Silencing of Aberrant RHO Expression), a mutation-independent, allele-specific antisense oligonucleotide (ASO) strategy. SNARE selectively suppresses mutant RHO transcripts by targeting the common, benign c.-26A/G single-nucleotide polymorphism (SNP) as an allelic discriminator. Candidate gapmer ASOs were screened in engineered reporter lines and validated in patient-derived retinal organoids, identifying RHOligo-A as the lead c.-26A-targeting candidate. In vitro, RHOligo-A achieved robust, preferential knockdown of the target allele, improving RHO localization in retinal organoids, and demonstrated a favorable safety profile with minimal transcriptomic off-target effects and no detectable immunostimulatory activity. Subsequent validation in a novel, humanized RHOP347L/WT mouse model, achieved sustained c.-26A-linked allele-selective suppression, retinal structure preservation, and significantly restored visual function, upon a single intravitreal administration. These findings establish RHOligo-A and SNARE as a scalable, mutation-independent therapeutic platform with strong translational potential and substantial clinical reach for RHO-associated adRP.

Deep learning has accelerated drug discovery, yet most existing models are trained using in vitro affinity datasets and consequently remain disconnected from the cellular context in which functional ligand-protein interactions occur. This limitation hinders the ability to reflect the complexity of native interactomes and characterize biological responses to molecular perturbation. Here we introduce C-PLANK (Chemi-Proteome Language Attention NetworK), a deep learning framework trained on fragment-protein interactions profiled directly in living cells using fully functionalized fragment (FFF) chemoproteomics. C-PLANK combines physicochemical embeddings with a bilinear attention network (BAN) to model both global cellular context and local residue-atom interactions, generating interpretable interaction fingerprints. Particularly, C-PLANK incorporates Cellular Interaction State Index (CISI), a systems-level evidential metric that contextualizes the biological plausibility of each predicted interaction against the global cellular interaction landscape. Across 431 ligand interactomes curated from eight independent chemoproteomic studies, C-PLANK consistently outperformed current state-of-the-art interaction prediction frameworks under both random and cold-protein evaluation settings. The inferred interaction fingerprints aligned with orthogonal evidence from structure-based pocket predictions, co-crystal structures, and cellular binding-site annotations. C-PLANK further generalized to unseen ligands. In a cellular target-focused discovery campaign, C-PLANK identified a previously unrecognized ligand that was subsequently advanced into an active chemical probe acting as a SIRT3 agonist in cellular assays. By learning directly from cellular chemoproteomics, C-PLANK moves beyond isolated interaction prediction toward cellular interaction-state modelling, establishing a computational foundation for future digital-twin frameworks in drug discovery.

Paternally-inherited epigenetic information can influence phenotype in offspring (1). Here, we identify a critical mechanistic contribution of KDM6A (UTX), an X-linked histone modifier and tumor suppressor, in regulating transmissible epigenetic information in mammalian sperm. Paternal loss of KDM6A increases cancer risk in genetically wild type offspring, but how Kdm6a knockout sperm transmit this effect at the molecular level is unknown (2). We find that KDM6A functions in spermatogenesis to promote methylation of histone H3 lysine 4 (H3K4) via selective interaction with the COMPASS complex methyltransferase KMT2C (MLL3). KMT2C and KDM6A are coordinately recruited to promoters of active genes in spermatogenic cells, contrasting with recruitment to intergenic enhancers in other cell types (3, 4). Loss of KDM6A disrupts H3K4 methylation at promoters of tumor suppressor genes in spermatogonia, and some of these defects persist in epididymal sperm and correspond to impaired expression in preimplantation embryos. These genes are also misregulated in normal and malignant hematopoietic tissue of genetically wild type offspring, indicating that impaired H3K4 methylation in KDM6A-deficient male germ cells may preferentially alter regulation of tumor suppressor gene networks in development across generations.

Standard methods for ancestral sequence reconstruction (ASR) rely on substitution models for the residues in a biological sequence and assume independent evolution across these sites, ignoring the epistatic interactions that shape molecular evolution. In contrast, deep learning models like variational autoencoders (VAEs) can learn low-dimensional representations ("embeddings") of sequences in a protein family that may implicitly handle these dependencies, raising the possibility of performing more accurate ASR by interpolating between extant sequence embeddings within the VAE's latent space. In this study, we test this hypothesis by developing and evaluating a VAE-based ASR pipeline. Benchmarking this approach against established likelihood-based and parsimony methods using various simulations of protein evolution, including scenarios with and without epistasis, we find that the VAE-based approach is consistently and significantly outperformed by standard methods, even in epistatic regimes where it was hypothesized to have an advantage. We further show that this failure is not due to a lack of phylogenetic structure in the latent space, which does contain evolutionary signal. Rather, the primary limitation is the information loss inherent to the autoencoding process: the VAE's decoder cannot generate sequences with sufficient fidelity for the precise demands of ASR.

The gastric network, comprised of brain regions whose activity synchronizes with the stomach's slow-wave rhythm, offers a unique window into the brain-body interaction involved in interoceptive processing. While previous work has established the existence of this network, its intrinsic organization and temporal unfolding remain poorly understood. Here, we reanalyzed resting-state fMRI-electrogastrogram data from 43 healthy adults of both sexes to characterize the time-averaged architecture and time-varying reconfiguration of the gastric network. We identified regions exhibiting phase-locked synchronization with the stomach slow electrical rhythm (0.05 Hz) and characterized cortical parcels comprising this network. Time-averaged graph-theoretical analysis revealed a fixed unimodal organization of functional communities, with primary visual, default mode network (DMN) and dorsal attention regions emerging as the principal time-averaged hubs. Next, we applied edge-centric functional connectivity (eFC) to capture the network state during transient high-amplitude "bursts". Time-varying community detection revealed communities whose compositions formed integrative combinations of DMN, visual, attentional and control elements. Edge-derived hubs shifted away from primary visual dominancy in the time-averaged analysis, and were instead directed by DMN regions, suggesting that moments of heightened connectivity in the network are coordinated by multisensory integration rather than passive sensory processing. These findings demonstrate that the gastric network is not merely a time-averaged, sensory-bound system, but rather a flexible and dynamically reconfiguring interoceptive network whose organization is selectively coordinated by transient cofluctuation events. This work provides a comprehensive network analysis of gastric-brain coupling and reveals a temporally structured mode of interoceptive integration that may support adaptive physiological and cognitive regulation.