Biology subjects

Zhang, R.

Publications and source records attributed to Zhang, R..

At least 19 recordsLinked to original sources

bioRxiv2026

RnR-ExM: Robust Non-rigid Registration Challenge for Expansion Microscopy Volumes

Expansion Microscopy (ExM) enables nanoscale imaging of biological samples while preserving structural and molecular details. The Robust Non-rigid Registration Challenge for ExM (RnR-ExM) was launched to benchmark 3D registration algorithms on real-world ExM datasets. The challenge released 24 pairs of high-resolution 3D volumes from zebrafish, mouse, and C.~elegans, deliberately selected for their difficulty with standard alignment methods. To evaluate performance, we provided curated segmentation masks of nuclei and blood vessels as landmarks for half of the dataset. To encourage learning-based approaches, we released code for generating synthetic training data reflecting the deformation and appearance of the unlabeled pairs. By submission, 168 participants registered, with six teams submitting results by the deadline; the top three were invited to present at ISBI 2023. Accurate registration of ExM volumes is essential for downstream biological analyses such as multiplexed molecular mapping and cell-type-specific structural profiling. In this paper, we detail the methods and results of these top teams, highlighting current capabilities and future directions for 3D registration in ExM. All datasets and code are available at https://rnr-exm.grand-challenge.org/

neuroscience↗

bioRxiv2026

Reinforcement learning with evolutionary priors unlocks multi-site enzyme optimization from sparse experimental feedback

Industrial enzymes underpin sustainable manufacturing, yet natural proteins rarely combine the activity, robustness and application compatibility required for practical use. Engineering such properties requires identifying beneficial combinations of interacting mutations within vast sequence spaces that are too large to search exhaustively and too sparsely sampled to model reliably. Here we introduce AlphaVariant, a reinforcement-learning (RL) framework that couples a generative evolutionary prior with iteratively updated low-N fitness models to enable feedback-driven search of large combinatorial protein sequence spaces. Across six experimentally measured protein fitness landscapes, AlphaVariant ranked first on the three large, non-enumerable multi-site landscapes. In two successive rounds of savinase activity optimization, AlphaVariant-designed libraries yielded 75.4% and 83.8% activity-improved variants among experimentally characterized candidates even as the accessible mutation space expanded from 25 to 67 positions. The search identified beneficial mutation combinations involving up to 17 structurally dispersed sites and produced a variant with 7.1-fold higher activity than wild type after only two design-test cycles. We then extended AlphaVariant to jointly optimize activity, thermostability and a patent-informed constraint for intellectual-property freedom. This multi-objective search yielded variants that outperformed the commercial protease Uno 100 L in both biochemical assays and application-relevant detergent washing tests. These results demonstrate that generative RL can efficiently navigate vast protein fitness landscapes, progressively incorporate multiple engineering objectives and convert sparse experimental feedback into enzyme variants with industrially relevant performance, effectively resolving the limits of biocatalyst engineering.

bioinformatics↗

bioRxiv2026

Intelligent systems distribute memory across internal and embodied resources

Intelligent systems do not operate in isolation from the body, yet how bodily actions support internal computation remains unclear. Here we show that primates and artificial agents recruit eye movements to maintain beliefs about hidden goals when internal computation is constrained. In a naturalistic navigation task, both systems spontaneously tracked goals with their eyes after they disappeared from view. This strategy emerged in artificial agents only when internal memory was constrained, identifying resource constraints as a driver of memory externalization. Restricting gaze impaired performance in both systems, while natural behavioral variability and causal perturbations in agents established a direct influence of oculomotor state on internal beliefs about hidden goals. Across agents, varying the reliability of internal memory and gaze redistributed goal information between internal and embodied states. The same resource-allocation principle predicted the distribution of goal information in macaque parietal cortex, where greater reliance on gaze was accompanied by weaker gaze-independent neural representations. Together, these findings identify eye movements as embodied memory and reveal a convergent principle across intelligent systems: memory is allocated across internal and embodied resources according to their reliability.

neuroscience↗

bioRxiv2026

Predicting cell-state-specific gene regulation from DNA across species

DNA encodes the cell's phenotype, but the same DNA sequence is interpreted differently in every cell type, developmental stage, and activity state. Existing deep-learning models solve only half of this puzzle: sequence-to-function models read DNA but treat cell state as a fixed label, while single-cell foundation models represent cell state but not sequence. We developed Bolero, a sequence-to-function model whose parameters condition on atlas-scale cell-state embeddings and additional biological variables to predict cell-state-specific chromatin accessibility and transcript abundance from DNA. Trained on 10.7 million cells from 33 datasets across six mammals, Bolero matched or exceeded specialised models and resolved regulatory differences between closely related cell states, including cortical regional and developmental gradients. Extrapolating cell-state embeddings extended Bolero's predictions to cell states that are absent from training. Cell-state conditioning facilitated tracing of the effects of non-coding variants to the cell types in which they act. Using an AP-1 activity score as a further continuous conditioning axis, Bolero predicted activity-regulated chromatin changes without requiring actual stimulation data. Across human cell types, we reconstructed the cooperative grammar of motif identity and spacing underlying predicted AP-1 responses. Bolero further predicted cell-type-specific chromatin accessibility and AP-1 responses in species excluded from training, extending comparative genomics from sequence conservation to predicted functional conservation. Across 84 mammals, we found that predicted AP-1 responses were less conserved than chromatin accessibility at cis-regulatory elements, driven by the human-specific sequence variants on AP-1 and cooperating motifs. By combining reference genomes with reference cell atlases, Bolero allowed genome sequences, either native or mutated, to be queried across cell states, perturbations, and different species for their chromatin and expression signatures.

genomics↗

bioRxiv2026

CucurbitAgent: An AI agent for exploring cucurbit multi-omics resources

Functional genomics in crops increasingly depends on integrating information from an extensive literature and heterogeneous multi-omics datasets. Large language models (LLMs) offer a powerful approach to synthesizing these resources and revealing hidden connections across datasets; however, they frequently hallucinate, producing plausible but unsupported claims, when responding to gene-specific queries. Here, we present CucurbitAgent, an AI research agent for functional genomics in the Cucurbitaceae, a major family of horticultural crops. CucurbitAgent integrates multi-omics resources from nine cultivated cucurbit species and more than 18,000 publications, encompassing approximately 256,000 genes, tissue-resolved expression profiles, over 16 million natural genetic variants, and roughly 242,000 predicted protein structures with more than 14 million predicted binding-site residue annotations. Responses link retrieved evidence to source records, including gene entries, expression measurements, structural predictions, and literature passages. On a curated benchmark of 50 tasks assessed by domain experts under blinded conditions, CucurbitAgent achieved an overall pairwise win rate of 96.0% against nine general-purpose LLMs, based on weighted scores for factual accuracy, completeness, relevance, and traceability. The platform is freely available at https://cucurbitagent.com and supports queries in 37 languages collectively spoken by an estimated 70% or more of the world's population. CucurbitAgent not only unifies evidence retrieval and interpretation but also reveals latent associations across datasets, helping researchers formulate testable hypotheses and substantially accelerating candidate-gene research. Its framework is readily transferable to other crops and, more broadly, provides a blueprint for building specialized agents in any domain.

bioinformatics↗

bioRxiv2026

Neural Representations of Visual Anagrams in Macaque Inferotemporal Cortex

Visual recognition requires stable representations of object identity despite large changes in the retinal image. Yet this stability must be selective: some transformations change not only how an image looks, but also what object it appears to depict. Visual anagrams highlight this challenge: a single image can resemble two different objects at different orientations, such as an elephant and a rabbit. Here, we asked how macaque inferotemporal (IT) cortex represents such transformations. We recorded large neuronal populations while presenting visual anagrams and unambiguous exemplars of their two interpretations across rotations. Neurons preferring one exemplar tended to prefer anagram orientations supporting the corresponding interpretation. At each of the two interpretation-relevant canonical angles, population responses more closely resembled those to the matching exemplar than those to the nonmatching exemplar. This matching advantage persisted when both exemplars were presented at the alternative canonical angle. Decoders trained exclusively on exemplars across rotations also tracked the shifts in anagram representations. Visual models showed anagram-exemplar alignment as well, with stronger alignment in models with greater relative shape preference. These findings suggest that transformation tolerance in IT is sensitive to what an image appears to depict, while relative shape preference is associated with similar representational organization in visual models.

neuroscience↗

bioRxiv2026

Axonal dysfunction precedes axonal degeneration in a mouse model of progressive parkinsonism

Axons of substantia nigra dopaminergic neurons "die back" in Parkinson's disease (PD), exhibiting early degeneration relative to cell bodies, a key feature also observed in PD models. A critical step towards understanding the process of axonal neurodegeneration is to determine the physiological properties of dopaminergic axons that are altered during the early stages of cellular dysfunction. Here, we compared the electrophysiology and calcium dynamics of dopaminergic axons in the MitoPark model of PD with age-matched controls. Our experiments show a pronounced reduction in action potential (AP)-evoked calcium signals in MitoPark axons that emerges well before the onset of dopaminergic fiber loss and motor deficits. In addition, calcium imaging and direct axon recording revealed deficits in AP propagation accompanied by alterations in the axonal AP waveform. Therefore, our experimental findings identify the progression of axon-specific physiological deficits that occur long before neurodegeneration, more precisely defining the period that precedes axonal loss.

neuroscience↗

bioRxiv2026

Hyperexcitable primary somatosensory cortex in the MitoPark mouse model of Parkinson's disease

The majority of Parkinson's disease patients suffer from decreased tactile acuity or neuropathic pain, both of which contribute to low quality of life and likely exacerbate motor dysfunction. How Parkinson's disease pathology relates to altered somatosensation remains unclear. One potential contributor is the progressive degeneration of dopamine neurons. To assess the extent to which gradual dopamine neuron loss contributes to altered somatosensation, we recorded spontaneous and stimulus-evoked activity in the primary somatosensory cortex of the MitoPark mouse model of Parkinson's disease. MitoPark mice experience progressive dopamine neuron degeneration that is independent of -synuclein. We found that MitoPark layer V regular-spiking (largely pyramidal) neurons, but not fast-spiking (largely parvalbumin-expressing inhibitory) neurons, had elevated spontaneous firing rates relative to littermate controls at late stages of symptom progression. Regular-spiking neurons also became hyperresponsive to tactile stimuli -- more neurons were recruited by stimuli across a wide range of intensities, and these neurons responded with higher firing rates. Fast-spiking neurons, but not regular-spiking neurons, became desynchronized at late stages of symptom progression. These physiological changes coincided with pronounced degeneration of dopamine neurons and axons, which we measured with tyrosine hydroxylase immunostaining in the midbrain, striatum, and primary somatosensory cortex. Altogether, this characterization revealed cell-type-specific changes in the primary somatosensory cortex that are consistent with both decreased tactile acuity and hyperalgesia in Parkinson's disease patients. This suggests that the loss of dopamine neurons, potentially including those that innervate primary somatosensory cortex, contributes to somatosensory impairments associated with Parkinson's disease.

neuroscience↗

bioRxiv2026

Multi-omic spatial mapping of the human habenula defines the molecular and anatomical organization of medial and lateral subregions

The habenula (Hb) is a small epithalamic midline structure critical for reward processing and mood regulation that is implicated in many complex brain disorders, including depression and addiction. The molecular anatomy of the Hb has been well-characterized in model organisms, but little is known about its organization in the human brain, including the transcriptomic and epigenetic signatures of spatially-organized lateral (LHb) and medial (MHb) cell types. Here, we generated a high resolution spatiomolecular map of the human Hb using both multiomic single nucleus sequencing (snMultiome) and spatially-resolved transcriptomics. We measured gene expression and chromatin accessibility across 4 MHb and 5 LHb neuronal cell types as defined by transcriptomic conservation with rodents and zebrafish. We identified cell type-specific molecular specializations related to GABAergic and glutamatergic signaling as well as regulatory relationships between open chromatin regions, transcription factors, and target genes associated with depression and addiction. High resolution spatial transcriptomics identified the anatomical location of heterogeneous LHb subpopulations, including two putative inhibitory populations in the LHb. We further characterized cell-type relationships between Hb neurons and glial subtypes and predicted ligand-receptor interactions between LHb and MHb neurons. Finally, we delineated Hb cell types and spatial domains enriched in genes associated with genetic risk for psychiatric disorders, identifying the strongest enrichment in the MHb. Collectively, the generated data and analyses resolve the spatiomolecular organization of the human Hb and identify molecular-genetic associations with neuropsychiatric disease.

neuroscience↗

bioRxiv2026

Two functionally distinct amyloid-β intermediates link aggregation pathways to pathology in Alzheimer's disease

Neurotoxic intermediates formed during amyloid-{beta}(A{beta}) aggregation are thought to drive Alzheimer's disease, yet their rarity and transient nature have obscured their identity and formation pathways in the human brain. Here, we combine super-resolution microscopy and single-molecule functional imaging with kinetic modelling to classify A{beta} intermediates according to their membrane-disrupting activity and connect their damaging effects to the aggregation pathways that generate them. Applying this framework to recombinant A{beta}42, an Alzheimer's knock-in mouse model and human Alzheimer's brain tissue, we find two functionally distinct intermediate populations present across all systems. They form through separate aggregation routes and differ in their neurotoxicity, abundance and susceptibility to inhibition. Both damage neurons and glia, but only one converts into plaques. Suppressing these intermediates pharmacologically in mice reduces plaque burden, attenuates astrogliosis and improves cognition. Our findings connect model systems with human pathology and identify intermediate formation pathways as actionable therapeutic targets across protein-misfolding disorders.

biophysics↗

bioRxiv2026

When label-free morphology is sufficient for targeted cellular measurements

Label-free morphology captures cellular responses, but its ability to substitute for targeted cellular measurements depends on the intended use. We assess measurement sufficiency using nearly 10 million paired phase-reporter observations across 1,000 gene knockouts, 52 fluorescent reporters and 73 A549 screens. Ten predictors differed in recovery across held-out fields, genes and screens. Same-cell pairing controls identified reporter-state information beyond perturbation averages and measured covariates. Perturbation ranking often persisted despite reduced variation in predicted response magnitudes. Predictions of held-out knockout responses retained a median 79% of the strongest hits, whereas functional-term retention varied. Atlas-derived criteria identified 30 quantitative and 7 ranking proxies for a fixed ensemble, with reference reproducibility constraining interpretation. In independent primary-hepatocyte data, brightfield predictions prioritized 10 of the 11 compounds with the largest measured metabolic-activity losses among 217 held-out compounds. Measurement sufficiency therefore depends on target, predictor, context and scientific task, beyond predictive correlation alone.

bioinformatics↗

bioRxiv2026

A single-nucleus multi-omic atlas of gene regulation across 21 adult human tissues

Diverse human cell types establish specialized functions through lineage- and context-specific regulatory programs. Interpreting non-coding genetic risk requires integrated multi-omic reference maps that directly connect regulatory DNA to cellular expression across human tissues. Here we present a single-nucleus multi-omic atlas comprising 459,856 transcriptomic and chromatin accessibility profiles from 21 adult human tissues and four donors, including paired measurements from 160,688 nuclei. The atlas resolves nine cell lineages, 61 broad cell types and 313 subclusters, and identifies 1,085,062 candidate cis-regulatory elements (cCREs), including 161,270 novel elements absent from ENCODE. Regulatory activity was dominated by cell identity but refined by tissue context. Joint profiling enabled 871,177 cCRE-gene associations and revealed lineage-specific regulatory architectures. Cross-tissue accessibility further identified lineage-restricted and constitutively inaccessible chromatin domains, the latter showing preferential hypomethylation across human cancers. Furthermore, we leverage this dataset to train sequence-to-function models to predict chromatin-accessibility effects for 548,656 fine-mapped variants, identifying 18,133 high-effect variants, including 1,120 broadly active variants. Models trained for eight endothelial subtypes further resolve predicted variant effects across vascular beds. Together, this atlas provides a comprehensive cellular and computational framework for interpreting regulatory sequence, context-dependent gene control, and complex trait genetics across the human body.

genomics↗

bioRxiv2026

Three dimensional reconstruction of the human nucleus accumbens reveals topographic organization and molecular heterogeneity of D1-islands across the anterior posterior axis

Spatial transcriptomics has transformed molecular characterization of the human brain, but most studies profile individual two-dimensional tissue sections and therefore cannot capture molecular organization across 3D neuroanatomical axes. Here, we developed a spatial genomics framework to reconstruct the 3D cellular architecture of the human nucleus accumbens (NAc) by densely sampling and aligning serial sections across its anterior-posterior (AP) extent. Combining single-cell Xenium profiling with near-transcriptome-wide VisiumHD, we mapped cell type composition and topography across AP, mediolateral and dorsoventral axes. Neuronal populations varied in relative abundance across the AP axis while maintaining characteristic spatial topographies. These differences were particularly pronounced among D1-islands, a topographically organized structure within the NAc in which transcriptionally distinct subpopulations of DRD1-expressing medium spiny neurons (MSNs) aggregate into discrete cellular formations. D1-islands comprised two molecularly distinct spatial domains with conserved counterparts across species. Both domains varied in relative abundance across the AP axis, but demonstrated distinct 3D distribution; one was present throughout the NAc, whereas the other, corresponding to islands of Calleja (ICj), was more restricted to intermediate-to-posterior levels. Local spatial analyses revealed region- and AP-dependent cellular relationships, including segregation of the two island populations despite their anatomical proximity. Together, these findings establish a framework for three dimensional spatial genomics of human brain tissue and reveal molecular and cellular organization not captured by individual tissue sections.

neuroscience↗

bioRxiv2026

Structural basis for differential inhibition of THIK1 by chemically related inhibitors

TWIK-related halothane-inhibited potassium channel 1 (THIK1) is a two-pore domain potassium (K2P) channel highly expressed in microglia that contributes to microglial homeostasis and neuroinflammatory responses. Pharmacological inhibition of THIK1 suppresses tonic K+ currents and attenuates NLRP3-dependent IL-1{beta} release, highlighting THIK1 as a potential therapeutic target for neuroinflammatory disorders. However, the structural basis of THIK1 inhibition remains unclear. Here we report cryo-electron microscopy (cryo-EM) structures of full-length human THIK1 in complex with C101248, a commercially available tool compound, and CVN293, a clinical-stage inhibitor that has completed Phase I evaluation. The two inhibitors occupy a similar site within the inner vestibule formed by TM2 and TM4 beneath the selectivity filter. Consistent with their closely related chemical scaffolds and shared binding site, both inhibitors remodel the Y273 inner gate but differ in their effects on the C-terminal gate. Electrophysiological analyses further demonstrate that the C-terminal gate contributes differently to inhibition by the two compounds. Together, our findings show how chemically related inhibitors exert different effects on THIK1 despite engaging a common binding site.

biophysics↗

bioRxiv2026

NicheScribe: consistent and generalizable annotation of stromal cells in the murine bone marrow niche

Hematopoietic stem and progenitor cells (HSPC) depend on interactions with nearby stromal cells to support their homeostatic functions and mount responses to perturbations. The murine bone marrow (BM) niche can be profiled using single-cell RNA sequencing (scRNA-seq), but there is considerable inconsistency in cell type identification, and uncertainty over how populations can be isolated prospectively by flow cytometry. Here, we produced new scRNA-seq reference datasets using purified populations of murine BM stromal cells, which we used to create a bespoke classifier for stromal cells. Our NicheScribe approach is consistent, accurate, and generalizable to query datasets irrespective of methods used for cell isolation. Moreover, consistent annotation permits comparative analyses of the same stromal cells across different bone types and perturbations like inflammation, neoplasia, and aging. Our work provides new tools for consistent cell type annotation in the murine BM niche and new insights for combined molecular and functional analyses.

cell biology↗

bioRxiv2026

Single residues within the YTHDF intrinsically disordered region control m6A-dependent mRNA decay

Intrinsically disordered regions (IDRs) mediate protein interactions, condensate partitioning, and regulatory control in RNA-binding proteins, yet the residue-level logic underlying IDR function has remained difficult to define using conventional fragment- and truncation-based approaches. Here, we apply cytosine and adenine base-editor screens to map residue-level regulation within the intrinsically disordered region of YTHDF2, a cytoplasmic reader that couples m6A to mRNA decay. By tiling base edits across YTHDF2 in a YTHDF1/YTHDF3-null background, we identify IDR residues that regulate YTHDF2-dependent cellular fitness. Interestingly, we identified residues that limit YTHDF2 decay activity. Their mutation disrupts endogenous protein-protein interactions and generates hyperactive YTHDF2 variants that promote m6A-dependent mRNA decay and alter recruitment to cytoplasmic RNA granules, without affecting intrinsic m6A binding or protein stability. These findings define the residue-level regulatory logic of the YTHDF2 IDR and establish base-editor screening as a strategy to uncover functional mechanisms encoded within IDR regions.

systems biology↗

bioRxiv2026

Loss of TDP-43 function drives cryptic circular RNAs in neurodegenerative diseases

TAR DNA-binding protein 43 (TDP-43) is a key pathological hallmark of amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) and a critical regulator of RNA splicing. While loss of TDP-43 induces aberrant splicing in linear transcripts, its impact on circular RNA (circRNA) biogenesis remains unexplored. Here, we show that TDP-43 depletion in human neurons induces widespread circRNA changes, especially upregulation of a distinct class of cryptic circRNAs that arise specifically upon loss of TDP-43. Some of these cryptic circRNAs incorporate cryptic exons derived from intronic sequences. Notably, these cryptic circRNAs exhibit greater stability than their corresponding linear RNA isoforms and accumulate progressively in neurons. Moreover, cryptic circRNAs are elevated in postmortem brain tissues from ALS, FTD and Alzheimer's disease (AD) patients. These findings reveal a previously unrecognized role for TDP-43 in repressing cryptic circRNA formation and establish these circRNAs as stable molecular signatures of TDP-43 dysfunction.

neuroscience↗

bioRxiv2026

Reference-guided pseudotime inference across species and biological contexts

Cells collected at the same chronological age can vary substantially in biological age due to the heterogeneity in the timing of differentiation, speed of maturation, and degeneration. However, existing pseudotime inference methods either disregard chronological time information, or rely on accurate time-series labels within similar species or biological conditions of interest. As a result, both types of strategies often fail to faithfully order cells from biological contexts without reliable time labels, along the desired axis of interest such as human embryonic development or disease progression. Here, we propose Cavebear, a machine learning framework that enables pseudotime inference in a query species or condition guided by scRNA-seq time-series profiles from a reference species or condition. Cavebear achieves more accurate developmental pseudotime inference than existing methods and provides in vivo temporal mapping for in vitro experiments. Furthermore, we illustrate the potential of Cavebear to study cellular-level disease progression in human patients using mouse cancer development models as references. By transferring temporal information across species and conditions, Cavebear enables systematic investigation of biological variation in contexts where such annotations were previously unattainable.

bioinformatics↗