Single-cell RNA sequencing (scRNA-seq) profiles cellular heterogeneity but captures only static snapshots, limiting inference of gene expression dynamics. We developed PROFET (particle-based reconstruction of generative force-matched expression trajectories), a framework that reconstructs continuous, nonlinear single-cell trajectories from sparsely sampled scRNA-seq time series. PROFET combines a particle-based gradient-flow algorithm with simulation-free force matching to accurately infer cellular dynamics. Across mouse and human in vitro datasets and an in vivo axolotl regeneration dataset, PROFET achieved 2.6-12.5× lower prediction error than ten state-of-the-art trajectory inference methods. Applying PROFET to newly generated scRNA-seq data from a palbociclib-treated MCF7 cell line and three published breast cancer patient datasets, we reconstructed treatment-response trajectories and identified a resistant cell subpopulation exhibiting large phenotypic shifts and enrichment of the surface markers UNC5B, TLR3, PCDH19, PROCR, SLITRK6, and SEMA6B. PROFET provides a biologically grounded framework for reconstructing cell-state dynamics from static single-cell data across development, regeneration, and therapeutic response. A record of this paper's transparent peer review process is included in the supplemental information.
Even after folding, proteins sample unfolded intermediates at risk of irreversible alteration (e.g., via proteolysis, aggregation, or posttranslational modification). Thus, kinetic stability impacts protein lifetime and abundance. However, we have very few measurements of unfolding rates, largely due to technical challenges. To address this, we developed SPARKfold (simultaneous proteolysis assay revealing kinetics of folding), a microfluidic platform to express, purify, and measure unfolding rate constants at high throughput via native proteolysis. We applied SPARKfold to determine unfolding rate constants for 1,104 protein samples comprising 31 dihydrofolate reductase orthologs with up to 78 chamber replicates each, providing statistical power to resolve subtle effects. SPARKfold rate constants for 5 constructs agreed with traditional measurements across a 150-fold range and provided information about the folding transition state via φ analysis. In future work, SPARKfold can reveal mutations that drive misfolding and aggregation and enable the rational design of kinetically hyperstable variants for industrial use. A record of this paper’s transparent peer review process is included in the supplemental information.
Protein/RNA condensates participate in many cellular processes, but their functional roles often remain unclear. We recently found that condensates of the RNA-binding protein ATXN2 oscillate with the circadian cycle, recruit translation machinery and circadian mRNAs such as Per2, and promote translation. Using mathematical modeling, we show that dynamic condensate assembly enhances the robustness and phase coherence of oscillations in a transcription-translation feedback loop (TTFL). This effect cannot be reproduced by simply increasing repressive-protein translation through static condensates. Condensates that oscillate with the core TTFL at an appropriate phase shift can increase protein translation without depleting the corresponding mRNA. Two circuit designs can generate this timing: TTFL-regulated condensate transcription or a hybrid architecture in which an intrinsically oscillatory condensate loop is coupled to the core loop. Knockdown of Per2, Bmal1, and Clock showed residual 24-h ATXN2 condensate oscillations, supporting the hybrid model.A record of this paper’s transparent peer review process is included in the supplemental information.
Many molecular signals are known to be organized in subcellular regions termed microdomains. One example entails Rho GTPases, which control a range of cell behaviors through their functional patterning in spatiotemporal units. While fluorescent biosensors have allowed precise measurement of Rho GTPase activity in living cells, computational tools to delineate and track signaling microdomains in time-lapse image sequences of such biosensors do not exist. Here, we introduce a method that centers on the notion of activation time series coordination to identify signaling microdomains in space and time. After validating the algorithm with simulated microdomains, we show that our method identifies domains that support current hypotheses on the signaling architecture governing Rho GTPase organization and also respond to perturbation with optogenetics. A record of this paper’s transparent peer review process is included in the supplemental information.
The evolutionary trajectory of SARS-CoV-2 is shaped by competing pressures for angiotensin-converting enzyme 2 (ACE2) binding, viability, and escape from neutralizing antibodies targeting its receptor-binding domain (RBD). Here, we present EscapeMap, a modular framework that enables the prediction and design of variants escaping antibodies. EscapeMap integrates deep mutational scanning data for ACE2 and 31 monoclonal antibodies with a generative sequence model trained on pre-pandemic Coronaviridae. To experimentally probe escape potential, we designed RBD variants under pressure from four clinically relevant antibodies (SA55, S2E12, S309, and VIR-7229). Among these designs, bearing up to 21 mutations from wild type, 50% expressed as stable proteins. Binding assays confirm that S309 and VIR-7229 retain recognition across diverse mutation combinations. EscapeMap accurately forecasts which antibodies are vulnerable to escape by our designed sequences. Finally, by identifying correlated escape routes, we predict and experimentally verify antibody combinations less prone to simultaneous escape, offering a quantitative basis for guiding therapeutic strategies.
In cell biology, optical techniques can measure cells’ internal states (biosensors) and stimulate cellular responses (optogenetics). Yet the design of all-optical experiments is often manual: a predetermined stimulus pattern is applied to cells, biosensors are measured over time, and data are processed offline. Here, we develop PyCLM, a Python-based suite enabling closed-loop measurement, image segmentation, and optogenetic control of thousands of cells per experiment. We showcase PyCLM on diverse applications, including performing feedback control on single cells and delivering developmental signaling patterns to Drosophila embryos. We compare single-cell versus tissue-scale optogenetic control of epithelial migration, revealing that fast and slow waves of receptor tyrosine kinase activity determine the direction of tissue movement, matching prior in vivo observations in zebrafish and mouse. PyCLM enables simple setup of dynamic experiments to probe cell and tissue properties and provides a first step toward real-time control of single-cell states at the tissue scale.
Exposed to diverse pathogenic and non-pathogenic insults, the airway epithelium must balance effective host defense while minimizing unnecessary inflammation and tissue damage. In studies of influenza A virus infection including spatial transcriptomics, Nguyen et al. illustrate how tissue-level organization of viral sensing may tune the intensity of innate immune responses.
Synthetic biology can program cellular behavior but remains underused in regenerative medicine. This commentary argues that integrating synthetic biology with tissue engineering should target vascularization and immune integration through compact, context-aware circuits. Embedded within engineered tissues, these circuits could enable adaptive grafts that sense stress and coordinate regenerative responses.
We introduce a systems-level approach to sensing and computing in which Escherichia coli acts as a living reservoir computer, performing complex information processing through its native growth responses without requiring genetic modification or specialized instrumentation. We validate this framework by accurately classifying early-stage COVID-19 plasma samples according to subsequent disease severity using only bacterial growth data, highlighting its prognostic potential without the need for infrastructure-dependent methods. By controlling nutrient media compositions, we also demonstrate that E. coli growth encodes nonlinear transformations that outperform linear regression, support vector machines, and multilayer perceptrons across diverse regression and classification tasks. More broadly, simulations across genome-scale metabolic models from multiple bacterial species support a link between phenotypic diversity and computational capacity. These findings position biological reservoir computing as a robust, scalable, and low-cost platform for intelligent biosensing, diagnostics, and hybrid bio-digital computation, while providing new mechanistic insights into the computational capabilities of living systems.
This study assessed three mammalian expression platforms—human embryonic kidney (HEK293), Chinese hamster ovary (CHO), and HepG2 cells—for their ability to produce recombinant human butyrylcholinesterase (rHuBChE), a promising bioscavenger for organophosphate (OP) poisoning treatment. Key challenges remain in producing rHuBChE with a glycosylation profile similar to that of the native human plasma enzyme butyrylcholinesterase (HuBChE), which is critical for its stability and in vivo half-life. HEK293 cells showed the most promising results, prompting extensive glycoengineering targeting over ten glycosylation-related genes to optimize the glycan profile. This involved knocking out genes associated with fucosylation, enhancing sialylation, and reducing glycan branching. Glycan analysis revealed a substantial reduction in aberrant glycan structures, improved sialylation, and a glycoprofile closely resembling that of plasma-derived HuBChE. These comprehensive modifications to glycosyltransferases and nucleotide sugar donor levels are expected to enhance pharmacokinetics and reduce immunogenicity, positioning rHuBChE as a viable candidate for further preclinical and clinical development against OP poisoning and other medical emergencies. A record of this paper’s transparent peer review process is included in the supplemental information.
The three-dimensional organization of chromatin into topologically associating domains (TADs) may impact gene regulation by bringing distant genes into contact. However, studies of TADs’ function and their influence on transcription have been constrained by ambiguities in TAD boundary definitions and challenges in directly measuring their regulatory effects. We overcome these limitations by developing species-level consensus TAD maps for human and mouse by using a bag-of-genes approach that exposes an emergent regulatory structure. To quantify TAD-mediated relationships, we use a foundation model trained on 33 million transcriptomes to define a contextual similarity metric that captures higher-order relationships missed by co-expression. We find that TADs are regions of elevated co-regulation, with our framework yielding testable hypotheses about chromatin organization across cellular states. This TAD-linked enhancement is strongest during early development and declines with aging, while cancer cells show distinct TAD usage that shifts with chemotherapy. Together, these findings suggest that chromatin organization acts through probabilistic rather than deterministic mechanisms.
Gut Bacteroides are abundant and critical to human health, yet most are genetically cumbersome, non-model microbes. A widely applicable editing tool for Bacteroides is essential for gut microbiome manipulation. Here, we develop STIB (ShCAST-based transient insertion system for Bacteroides), an efficient genome-editing tool derived from CRISPR-associated transposases that enables rapid and site-specific insertions independent of homologous recombination. By fusing a nicking homing endonuclease to the transposase and an ATPase to Cas12k, we systematically optimize STIB to minimize plasmid cointegration and achieve >97% on-target insertion. STIB exhibits broad applicability across different genomic loci in diverse Bacteroides species, including non-model species. Finally, we apply STIB to achieve species- and site-specific editing of distinct Bacteroides species within a complex synthetic gut microbiota. Overall, STIB expands the toolbox for the functional investigation and engineering of the human microbiome. A record of this paper’s transparent peer review process is included in the supplemental information.
CRISPR-based high-throughput mutagenesis screens enable systematic mapping of mutations to phenotypes, yet deciphering mutation-phenotype links remains challenging. Here, we present ProTiler-Mut, a versatile computational framework that leverages tiling mutagenesis screens, which introduce variants across entire protein sequences, to analyze mutation effects at the levels of residues, substructures, and protein-protein interactions (PPIs). Applying ProTiler-Mut to multi-condition base-editing (BE) screens targeting DNA damage response proteins and T cell regulators, we define a separation-of-function (SoF) category beyond the conventional loss-of-function (LoF) and gain-of-function (GoF) classes, where SoF mutations show the strongest enrichment for ClinVar-annotated pathogenic variants. ProTiler-Mut also identifies candidate substructures that enable functional inference of unscreened pathogenic mutations and prioritizes candidate phenotype-associated PPIs potentially disrupted by functional variants. Using ProTiler-Mut, in cells with elevated programmed cell death 1 (PD-1) expression, we identify pathogenic GoF mutations that constitute a substructure that may disrupt mitogen-activated protein kinase (MAPK)1-RSK1 interactions and lead to MAPK activation. Finally, we show that ProTiler-Mut is applicable across different mutagenesis screening platforms. A record of this paper’s transparent peer review process is included in the supplemental information.
Macrophages can remember prior activation and subsequently augment their response to restimulation through trained immunity. However, it remains uncertain how trained immunity phenotypes manifest in individual cells. Here, we leverage highly quantitative single-molecule RNA imaging across 90,857 individual macrophages from 26 human donors to reveal inflammatory response dynamics in trained vs. untrained populations at single-cell resolution. Different inflammatory response genes showed distinct single-cell behavior in trained populations upon restimulation. Although training increased transcription of these response genes early after restimulation, untrained populations eventually "caught up" to the transcriptional output of trained populations, highlighting the importance of sampling timescale when interpreting transcriptional assays of training. Training did not significantly alter the relationship between the transcriptional activation of different genes within the same single cell, and any single cell appeared to be capable of training. Overall, these results revealed gene-specific single-cell transcriptional changes that generate population-wide training phenotypes in macrophages.
