Poster Abstracts

Networks of actin filaments play central roles in numerous cellular processes, including the intracellular transport and positioning of organelles. Some myosin motors facilitate the movement of cargoes along actin filaments. However, these motors also apply forces onto the filaments as they transport their cargo. This reciprocal interaction creates a feedback loop in which the motion of cargoes not only depends on actin organization, but can reshape the actin network that guides their movement. Here, we use coarse-grained computer simulations to study the interplay of motor-driven cargo transport and the organization of crosslinked and confined actin networks. We independently vary the numbers of crosslinkers, cargoes, and motors per cargo to reveal that that cargo-bound motors can drive actin bundling, compaction of the actin network, and clustering of cargoes. Strong network compaction requires sufficient crosslinking, indicating that cargo-generated forces must be transmitted through a connected actin network to produce large-scale remodeling. Increasing the number of cargoes strengthens network compaction and promotes clustering of cargoes in actin-rich regions. These results demonstrate that when cargo-bound motors engage a crosslinked actin network, they generate forces that can reorganize both the cytoskeleton and the cargo population itself. This work provides a quantitative framework for understanding how motor-driven transport, actin crosslinking, and cellular confinement together shape large-scale organization.

Local feature attribution is a standard tool for auditing and debugging deep learning predictions, but existing attribution methods are not designed for systems that chain pretrained, frozen, or API-only modules. Gradient-based methods such as Integrated Gradients require an end-to-end computational graph that may be unavailable. Perturbation-based methods such as KernelSHAP require a reference input or background distribution whose choice can substantially alter attributions and may not be defensible for composite pipelines. We present SensX, a local attribution method that treats the model as a black box and replaces arbitrary design choices with interpretable, application-grounded parameters. SensX adapts Morris-style coordinate walks from global sensitivity analysis to local attribution. It requires no access to model internals, training data, or arbitrary reference inputs. We validate SensX across four case studies, each targeting a distinct limitation of existing methods. On a synthetic benchmark where ground-truth relevant features vary per input, SensX reaches 95% top-k attribution accuracy versus 58% for the best KernelSHAP/Integrated Gradients variant. On a ViT with >150,000 pixel-channel features, SensX produces spatially coherent maps and exposes systematic intra-patch bias where KernelSHAP is infeasible and Integrated Gradients yields task-irrelevant attributions. On single-cell classifiers with unstructured gene-expression features, SensX attains the lowest top-k perturbation AUC. On a composite spatial transcriptomics system where neither method is applicable, SensX reveals reliance on preprocessing grid artifacts and a bias toward low-staining regions.

While work has been done to develop kinetic models of voltage-dependent ion channels which consult the gating physiology of the channel and whose parameters are inferred from current relaxation, the literature is ambiguous on how best to distinguish between these models. Here we propose to use hysteretic conductance resulting from oscillatory voltage protocols to assess the relative performance of these models in replicating nonequilibrium experimental data. We perform patch clamp experiments on transfected HEK293, tsA201 cells which over-expresses mutant Shaker Potassium ion channels, applying both activation and inactivation voltage protocols as well as a variety of oscillatory voltage traces and measuring the resulting currents. We optimize the parameter coefficients for six proposed Markov models from the literature using the experimental activation and inactivation data. We then compare each model’s ability to predict the hysteretic conductance from the experimental oscillatory voltage protocols.

Saccharomyces cerevisiae, or budding yeast, experiences heat shock (ambient temperature to 42–44°C) when ingested by birds, which serve as ecological dispersal vectors for yeast. After dispersal, yeast cells are expelled into new environments where they are likely to encounter starvation. Under starvation conditions, budding yeast can undergo sporulation, a meiotic process that increases cell survival in unfavorable conditions. We are investigating the predictive capacity of heat shock on cell-fate determining processes such as sporulation. To assess this relationship, we performed RNA sequencing on a sporulating diploid lab strain exposed to temperatures ranging from 35–46°C over 5 minutes to 2 hours. Dimensionality reduction analysis reveals distinct gene expression programs that segregate by temperature and time, indicating that yeast mount qualitatively different transcriptional responses across their thermal niche. Notably, after a 42–44°C heat shock for 60–120 minutes, we observe an upregulation of sporulation transcripts. A subset of these upregulated transcripts is under the control of Sum1, a middle-sporulation transcriptional repressor known for its role in regulating the meiotic recombination checkpoint. Sum1 is also thermally sensitive and has been shown to pellet at 42°C via biochemical sedimentation. These data support a model in which heat shock acts as a predictive signal that prepares cells for sporulation by engaging a Sum1-dependent transcriptional program. Our ongoing work, combining transcriptomics with a high-throughput sporulation assay, aims to define the regulation of this uncharacterized long-term heat shock response and its functional consequences for cell-fate determination.

Somitogenesis in vertebrates proceeds through sequential and periodic segmentation of the presomitic mesoderm (PSM) into somites. Both somite length and PSM length vary systematically with developmental stage. Mechanistic models of this process have grown increasingly detailed, incorporating gene regulatory networks, morphogen reaction-diffusion dynamics, and tissue mechanics. These approaches capture extensive molecular detail. They also require many parameters that are difficult to constrain from data, and they obscure which features of the system actually produce the observed morphology.

We present a simplified, phenomenological model that retains only the minimal ingredients. A segmentation clock sets the timing of boundary determination. A posterior elongation velocity captures tailbud growth. A gradient of FGF informs the determination of somite boundaries, followed by a delay before segmentation. With these ingredients, the model can match somite and PSM lengths measured experimentally. The same framework predicts some of the morphological consequences of perturbing the certain key features of somitogenesis.

The model reproduces several features of the data and of perturbation experiments quantitatively and clearly fails to capture others. We treat those failures as informative. Since the framework assumes little, the assumptions it does make are visible and individually testable. A minimal phenomenological model of this kind provides a transparent baseline against which the necessity of additional mechanisms can be judged.

Natural microbial communities exhibit immense levels of diversity as well as significant functional redundancy. This complexity has been a major roadblock to understanding these ecosystems and their responses to perturbation. Nevertheless, several recent theoretical and empirical studies indicate that the predictive power of simple coarsened models may improve with community richness ("emergent predictability"). Importantly, common modeling frameworks like Lotka-Volterra and consumer resource models do not generally display this behavior. This makes understanding the origins and properties of this phenomenon an important question. This poster will investigate a simple extension of a consumer-resource model that exhibits two forms of emergent predictability: one results in an observable well predicted by the environment alone, while the other requires compositional information to predict the observable.

Cardiac arrhythmias arise from irregular and often chaotic electrical activity in the heart and remain a major cause of morbidity and mortality worldwide. Early prediction is essential for timely intervention but remains challenging due to the highly individualized, nonlinear, and nonstationary nature of electrocardiographic (ECG) signals. Most existing machine learning approaches treat arrhythmia detection as a static classification problem, limiting their ability to provide early warning or adapt to patient-specific cardiac dynamics.

We introduce CASCADE (Chaotic Attractor Sensitivity for Cardiac Anomaly Detection), an online and personalized framework for early arrhythmia prediction. CASCADE continuously forecasts short-term ECG dynamics and flags anomalies when observed signals deviate significantly from model predictions. The framework is built on Dynamical Systems Machine Learning (DynML), a novel reservoir computing paradigm that uses ensembles of continuous-time nonlinear dynamical systems to capture complex heartbeat dynamics, while requiring training only a simple linear readout. This design enables efficient online adaptation without retraining the underlying dynamical system. Rather than directly classifying heartbeats, arrhythmic events are detected as statistically significant deviations from predicted short-term dynamics, relative to subject-specific baseline behavior learned from normal rhythms.

We show that prediction performance is governed by reservoir dynamical complexity, quantified using topological entropy. Reservoirs tuned near optimal entropy detect subtle arrhythmic signatures earlier and more reliably. Evaluation on the MIT-BIH arrhythmia database demonstrates strong performance across diverse patients, highlighting CASCADE as a scalable, interpretable, and efficient framework for real-time cardiac monitoring and early-warning clinical support.

To survive environmental stress, cells rely on the protein homeostasis (proteostasis) network (PN), a coordinated system of chaperones, degradation pathways, and transcription factors. PN function and stress responsiveness decline markedly with age, contributing to the accumulation of aggregated proteins associated with the onset of age-related diseases. Ubiquitin signaling is central to the PN, with K48 and K63 linkages having established roles in proteasomal degradation and vesicular trafficking, respectively. However, the functional roles of other ubiquitin chains remain poorly understood. Specifically, K29-linked ubiquitin chains are evolutionarily conserved and strongly induced during cell stress, such as during the heat shock response, yet their cellular organization, substrates, and functions remain unknown.

To this end, I have utilized a previously published, high-affinity K29-specific synthetic antibody fragment (K29-sAB) that selectively binds and enriches K29-modified proteins. With this tool, I can differentiate K29-linked ubiquitin from other ubiquitin moieties and examine questions related to K29 linkage function during heat stress. In preliminary IP/MS in budding yeast, I have now identified strong heat shock-dependent interactions between K29 chains and quality control pathways, such as the plasma membrane protein quality control (PMQC) network, highlighting its potential as a stress-adaptive signaling platform for protein turnover and degradation. In parallel, I have utilized this K29-sAB to visualize subcellular localization of K29-linked ubiquitin in fixed yeast cells through immunofluorescence (IF) and in live cells with genetically encoded K29-sAB fused to fluorescent proteins under an inducible promoter system. The use of these cell biological assays further dissects the interactions observed from my IP/MS data set, allowing me to examine the adaptive signaling roles of K29 ubiquitin linkages during the heat shock response.

Every hour of inappropriate antibiotic therapy can worsen patient outcomes. We show ATR-FTIR spectroscopy with an antibiotic probe distinguishes resistant strains in just 20 minutes. By exposing seven S. aureus strains (three MSSA, four MRSA) to sub-MIC ampicillin and collecting spectra at 0, 20, 30, and 60 minutes, we found cell wall restructuring signatures. These signatures paired with linear methods achieved balanced accuracies of 0.91 at 20 minutes and 0.90 at 30 minutes under leave-one-strain-out cross-validation. Discriminative bands mapped to peptidoglycan and carbohydrate precursor regions, consistent with differential cell-wall stress between susceptible and resistant strains. We independently confirmed this by TEM and AFM at 20 minutes. These results establish a proof of concept for phenotypic MRSA/MSSA discrimination in under 30 minutes using a rapid, reagent-free, and interpretable approach.

Understanding complex microbial interactions remains a significant challenge because conventional batch cultures lack physical architecture to capture sequential cell-to-cell communication or simulate structured environments without cross-contamination. To overcome these spatiotemporal constraints, this study presents a large-scale microfluidic platform featuring an ultra-high-density 1024-chamber array with automated valve actuation. High-fidelity multiplexing capabilities enable independent fluidic addressing and localized environmental transitions across parallel double-chamber units.

To demonstrate platform utility, GFP-expressing carbenicillin-resistant (CarbR) and RFP-expressing carbenicillin-sensitive (CarbS) Escherichia coli strains were encapsulated in multiplexed compartments, utilizing isogenic mutants harboring either wild-type TEM_WT or catalytically inactive TEM_DEAD as controls. Automated protocols executed precise, sequential valve actuation, establishing a programmed delay of top-valve opening to regulate diffusion of donor-produced beta-lactamase into bottom chambers under lethal carbenicillin concentrations (4 ug/mL).

This programmed delay allowed beta-lactamase from CarbR donors to degrade antibiotic before CarbS recipients reached irreversible damage. The temporal causality shifted phenotypic growth patterns, proving that sequential valve opening creates a critical pre-incubation window required for community-wide enzymatic protection. High-throughput screening using TEM_DEAD mutants confirmed that recipient survival is strictly governed by donor catalytic activity, ruling out horizontal gene transfer or physical shielding. Replicated robustly across 1024 chambers, this platform provides a scalable framework for high-throughput screening of synthetic microbial consortia and complex community dynamics.

Collective invasion is a key mechanism by which tumors disseminate and metastasize, involving coordinated migration of heterogeneous cell populations. Experimental studies in spheroid-based assays have identified specialized leader and follower cells that work together during this process, but the biophysical rules governing their interaction remain unclear. We present a mechanistic, cell-based computational model using the Cellular Potts framework to investigate how heterotypic adhesion, leader motility, and follower proliferation jointly shape invasion. Leader–follower tumors were simulated across 13310 parameter sets, and invasion was quantified by invasive and infiltrative areas, finger-like protrusions, solitary defectors, and detached clusters. From these simulations, we identified four distinct invasion phenotypes: non-invasive, bulk collective, single-cell, and multimodal. Multimodal invasion—-the coexistence of cohesive strands, solitary cells, and small clusters-—emerged as the most prevalent phenotype, particularly under moderate adhesion, high motility, and intermediate proliferation. Proliferation primarily scaled tumor bulk and cluster composition rather than determining invasion mode. Mapping outcomes across the parameter space revealed sharp transitions between invasion modes, underscoring trade-offs between adhesion and motility in shaping invasion complexity. Our results show that hybrid invasion behaviors, previously considered rare, arise robustly from simple mechanical rules and are favored in a broad region of the parameter space. This framework reconciles binary models of invasion with experimental observations of heterogeneity, providing predictive insights into how modulating adhesion, motility, or proliferation can restrict metastatic spread.

To maintain and repair adult tissues, a balance must be maintained between stem cell proliferation and generation of more differentiated cells. When dividing, stem cells can either self-renew into stem cells, or their progeny can become progenitor cells that can then differentiate into more specialized cells. Feedback from the differentiated cell population onto regulation of division controls tissue growth and maintains tissue homeostasis. Here I consider how to differentiate between multiple cell lineage models with potential nonlinear feedback terms. I consider the influence of several commonly made assumptions in stem cell models including: (1) how division of more differentiated progeny is achieved , (2) whether or not stem cell death is included, (3) the impact of symmetric and asymmetric stem cell divisions. We find that including differentiated cell division in many published stem cell models can lead to the existence of a spurious steady state that may not be biologically realistic, and the parameter region for existence of nontrivial equilibrium, which corresponds to tissue at homeostasis, shrinks rapidly as more feedbacks are added to the model. We consider potential ways to modify a stem cell lineage model to get rid of this unrealistic steady state such as getting rid of differentiated cell division altogether or modifying how we model progenitor division.

Sequential segmentation of the body axis is conserved across diverse animal phyla. The timing of somite segmentation in vertebrates is thought to be controlled solely by the period of the segmentation clock but their periods mismatch. By combing real-time imaging of the segmentation clock with pharmacological and surgical treatments, we show that the segmentation timing is governed by two sequentially acting distinct timers: while the segmentation clock controls segmental commitment, mesenchymal to epithelial transition (MET) of committed cells into a somite is regulated by an hourglass-like timer, whose duration is proportional to somite size and regulated by cytoskeletal activities. By modulating the hourglass timer, we altered the number of predetermined compartments, changed somite lengths, and converted zebrafish segmentation from periodic to irregular, mimicking the formation of anterior somites in chicken. These results break the textbook dogma by showing somite lengths and segmentation timing are regulated by two distinct timers.

Biomolecular condensates are membraneless compartments within cells hypothesized to regulate processes such as transcription (1), splicing (2), and protein aggregation (3). Condensates commonly form complex heterogenous structures, including well-recognized multi-layered condensates such as the nucleolus or the paraspeckle (4). Lack of condensate formation or disruption to their morphology is linked to cellular diseases such as cancer and viral infections (5, 6). While the importance and organization of condensates are observed, there is limited understanding regarding the formation of internal structures and their influence on the condensate’s function. These are challenging to predict due to the variability in RNA and protein composition among condensates, making predicting structure solely from composition difficult.

To solve this, I will identify the specific interactions that drive the formation and organization of multi-layered condensates through available thermodynamic models and bioinformatics data. This work investigates interactions assembling three types of biomolecular condensates found in three different organisms: paraspeckles in mammals (7), embryo speckles in mice embryos (8), and omega speckles in Drosophila melanogaster (9). Each condensate has an observed architecture that is built around an essential long non-coding RNA (lncRNA). I have determined the specific sequences that drive RNA-RNA as well as RNA-protein interactions through bioinformatics tools like ViennaRNA, NUPack and protein motif profiles. I will then use this data as initial hypotheses to test in molecular dynamics simulations and investigate how interaction networks among biomolecules form condensate morphology. Condensate morphology will be characterized by looking at the number, size, shape, and distribution of species within each cluster, then comparing these parameters to the known profiles of each condensate.

This work will help design a workflow to help predict condensate internal structures to better understand condensate functions. Future applications include implementing design of condensate structures with constraint-based designs, such as when prompted with specific characteristics, this platform will design appropriate/accurate condensate properties that will achieve the desired architecture. This would enable scientists to investigate and design condensates for therapeutics applications ranging from cancer biology (5) to antiviral targets (6). Ultimately, this work will reveal the defining interactions that are necessary to understand how condensate architecture emerges.

During embryonic development, inner organ tissues such as the gut break left-right symmetry, forming chiral organ shapes in 3D. While genetic patterning is known to influence chirality, we now know that cell intrinsic mechanisms are sufficient to drive reproducible chiral shape changes. This ‘bottom-up’ mechanism for symmetry breaking generates stereotyped shape changes through mechanical interactions between constituent cells, and perturbations of chiral cytoskeletal components can invert organ chirality. However, the mechanical forces and torques that arise in these tissues composed of chiral cells are unknown, as are the feedback mechanisms that enable stereotyped shape changes. We hypothesize that initial cell chirality establishes anisotropic junctional tensions, which are actively controlled and dynamically updated to transmit chirality from the cell scale to tissue scale in a manner robust to perturbations. Here, I will describe our methods to infer intercellular forces from 3D microscopy data by using the observed cellular geometries, and determine their update rules during development. I will also discuss our goals to learn minimal mechanical feedback rules at the cell scale that yield robust yet invertible chirality at the organ scale.

Living organisms, from bacteria to humans, are more likely to survive if their traits enhance fitness. In populations well adapted to their environmental niches, natural selection proceeds via rarely beneficial mutations. But when a catastrophe wipes out niche diversity, sudden adaptation often follows. Here, we present a data-validated theory of natural selection in the wake of catastrophe and unveil a simple law that emerges during recovery: the mean fitness relaxes inversely with time, with a prefactor proportional to the number of traits coupled to the post-catastrophe environment. We put our approach to test using experimental fitness landscapes measured following antibiotic administration to E. coli. The resulting mean trait adaptation is not described by gradient ascent on a fitness landscape, instead it follows an algorithm known as Levenberg-Marquardt optimization. Near fitness peaks, evolutionary trajectories are biased against greediness - from an optimization perspective, post-catastrophic selection is optimistic.

Patterns are ubiquitous and contain a wealth of embedded information within their structures. In the microbial world, macroscale patterns can be characterized by features like shape, color, texture, and size, which collectively make up colony morphology. Colonies are routinely grown in microbiology labs for research or clinical purposes, but their visual signatures remain qualitative and are often overlooked. Despite many works demonstrating that microbial characteristics and behaviors are reflected in spatiotemporal growth, there have been few advances robustly mapping these relationships. To bridge this gap, we developed an encoding-decoding scheme where biological variables are sufficiently uniquely encoded into a colony’s spatiotemporal signature for a given environmental condition such that machine learning (ML) can decode back the intrinsic information from morphology alone. In doing so, we designed an integrated workflow featuring high-throughput data generation with up to 96 colonies on one plate, ML model training on latent representations of colony images, and convenient sample imaging using iPhone. We demonstrate the feasibility of this approach by inferring community compositions from colonies of a 2-member fluorescent community, a 4-member dual fluorescence community, and a synthetically assembled community of 10 members isolated from hospital sink p-traps. We applied the trained 10-member model to measure 60 communities to demonstrate that multi-level spatial partitioning results in higher diversity. When comparing the time, cost, and ease of use between the proposed method and 16S rRNA sequencing, the advantages are clear especially given the number of communities. In our second application, we demonstrated how the proposed method can replace laborious CFU counting in measuring conjugation rate, the main mechanism responsible for antibiotic resistance gene proliferation. Lastly, we applied the framework to predict phenotypic susceptibilities to 22 antibiotics using a dataset of over 300 clinical isolates grown in just two conditions. This ability to map an isolate’s behavior in response to one drug and its predicted response to a different drug condition enables time- and resource-saving clinical phenotyping. We expect these results to enable high-throughput, cheaper, and more convenient measurements of microbial properties for both engineered and naturally occurring strains without any engineering or genetic information. POSTER #18: Decoupling Temporal Control of Genetic Patterning from Morphogenetic Execution in Somitogenesis Mayesha Sahir Mim, Ertuğrul Özbudak Northwestern University - Feinberg School of Medicine, Chicago, Illinois, USA Contact: mmim@northwestern.edu Somitogenesis requires the precise coordination of genetic patterning and tissue morphogenesis to generate segmented structures along the vertebrate body axis. The classical clock‑and‑wavefront paradigm and the modernized clock-dependent oscillatory gradient model provide a foundational explanation for how oscillatory gene expression and signaling gradients specify somite boundaries within the presomitic mesoderm (PSM). However, a fundamental gap remains in our understanding of how somite determination is translated into the physical formation of a somite boundary. Specifically, while the timing and position of segmentation are genetically encoded, the mechanisms that govern the execution phase, the regulated interval between determination and boundary formation, remain inadequately characterized. Here, we investigate this overlooked temporal window in vertebrate somitogenesis, focusing on identifying previously unrecognized regulatory inputs that operate downstream of genetic patterning. By quantitatively examining the timing relationships between determination and morphogenetic boundary formation under conditions that preserve upstream segmentation parameters, we aim to decouple pattern specification from boundary execution. This approach allows us to interrogate whether additional regulatory layers, potentially mechanical or biophysical in nature, modulate the conversion of genetic timing information into physical tissue architecture. Collectively, this work seeks to extend existing models of segmentation by highlighting an execution phase that is distinct from patterning and subject to independent regulation. By reframing the existing model to include downstream control of morphogenetic timing, this study provides a conceptual framework for understanding how temporal genetic information is robustly translated into segmented tissue form during vertebrate development.

Differentiation plays a central role in the development of multicellular organisms and microbial communities. While some microorganisms can differentiate into various cell types in response to environmental stimuli, these behaviors are species-specific and difficult to dissect and control. Synthetic microbial consortia are efficient platforms to study multicellular behaviors. However, despite wide studies on cell interactions and dynamics, control on programmable differentiation remains limited. Here we construct a synthetic bacterial differentiation system that produces permanently differentiated populations and regulates the ratio of the two cell types through intercellular signaling. We show that the system achieves a stable and tunable ratio in the long term in a continuous culturing platform. We further show that the feedback strength and growth rate ratio of the two cell types collectively determines the steady state ratio. Finally, we show in the heterogeneous environment the system is capable of spatial pattern formation with localized differentiated cell types. This study demonstrates how differentiation coupled with feedback control can be used to create complex synthetic multicellular systems.

Understanding cell state changes, such as cell cycle stages, is essential for understanding cellular behavior such as cell growth, differentiation, and infection dynamics. Current techniques engineering fluorescent cell lines with tags on the necessary proteins is time consuming; furthermore, fluorescent tags are phototoxic and limit the temporal resolution available experimentally. We seek to develop tools to identify cell states from phase contrast label free imaging. In this work, we leverage the supervised contrastive learning framework to train a neural network to identify G1 cells in induced pluripotent stem cell colonies from label-free phase contrast data. We train these models on aligned live cells aligned with corresponding immunofluorescent fix-and-stain data, bypassing the need for specialized cell lines, and demonstrating the widespread applicability of this technique for classification tasks beyond the cell cycle. The model correctly identifies G1 cells with an accuracy of 68%, recall of 74%m and precision of 47% showing the neural network learns the morphological characteristics of G1 stage cells, despite the overwhelming class imbalance of S/G2 cells.

Breaking left-right (LR) symmetry is fundamental to the positioning and function of visceral organs, including the looped structure of the digestive tract. Asymmetric looping allows for compact elongation of digestive organs inside of the confined body cavity. While amniotes are known to break LR symmetry through early genetic patterning of growth factors, no LR asymmetric gene expression patterns have been identified in the developing fly embryo to date. Instead, Drosophila LR symmetry-breaking is dependent on the unconventional class I myosins, Myo1D and Myo1C. The gut is the first to break LR symmetry, forming a coiled loop. Myo1C activity reliably inverts both the hindgut and midgut when driven in tissues expressing the transcription factor Brachyenteron (Byn), which includes the hindgut and the longitudinal visceral muscles surrounding the midgut. How, where, and when does the activity of Myo1C inform cell and tissue-scale behavior to determine the chirality of the midgut? Because Myo1C overexpression in the byn domain is sufficient to invert coiling, we investigated byn mutants, which lack the hindgut and longitudinal muscles. We found that their midguts nonetheless coil, but show randomized chirality; therefore, the byn domain controls the initial asymmetry, but is not required for coiling mechanics. In further experiments, we found that genetically ablating the longitudinal muscle is likewise sufficient to randomize chirality. An overexpression screen suggests Myo1C acts in the hindgut epithelium to control midgut chirality, and we isolate the developmental timing of Myo1C’s chirality-inverting activity using heat-shock induced Myo1C expression. Altogether, our data suggest a muscle-dependent, post-translational mechanism for controlling organ chirality across tissue layers.

Somites are embryonic precursors of the vertebral column whose formation depends on coordinated oscillatory gene expression within the presomitic mesoderm (PSM). Notch signaling plays a central role in maintaining local synchronization during this process, and disruption of its activity leads to segmentation defects, such as congenital scoliosis. Somites form with progressively decreasing size along the anteroposterior axis; however, how tissue-scale properties such as somite size and axial position influence the emergence of synchronization remain unclear.

To investigate how developmental timing and signaling dynamics shape somitogenesis, we examined zebrafish embryos exposed to sustained perturbation of Notch signaling across multiple developmental stages. Sustained inhibition produced stage-dependent defects in somite boundary formation. Strikingly, despite ongoing inhibition, a subset of embryos exhibited partial recovery of boundary formation at later stages. This unexpected recovery suggests that segmentation robustness cannot be explained solely by the instantaneous state of Notch signaling. Instead, embryos display heterogeneous and stage-dependent responses, revealing latent capacity for self-organization within the segmentation program.

Together, these findings suggest that somitogenesis reflects a dynamic interplay between signaling state, developmental timing, and tissue-level properties. Our results highlight the importance of system-level robustness in shaping segmentation outcomes and raise the possibility that tissue-level memory or hysteresis-like behavior contributes to recovery under sustained perturbation.

The Wnt/β-catenin pathway governs a broad range of biological processes, from development to tissue homeostasis and cancer. In the canonical view, Wnt activation stabilizes β-catenin, which accumulates in the nucleus and drives TCF/LEF-dependent transcription, implicitly assuming that signaling output is determined by nuclear β-catenin abundance. However, across many signaling pathways, including NF-κB, ERK, and p53, it is the temporal dynamics of a transcription factor's nuclear level, rather than its concentration, that determines downstream gene expression and cell fate. β-catenin dynamics have similarly been implicated in shaping Wnt responses, yet what physiologically shapes these nuclear dynamics, and in particular how the mechanical state of the cell contributes, remains largely unknown. Unlike most transcription factors, β-catenin shuttles between cell-cell junctions, the cytoplasm, and the nucleus, which may make it uniquely sensitive to mechanical regulation by the actin cytoskeleton and nuclear shape.

Here, we ask how the temporal dynamics of nuclear β-catenin are regulated and how these dynamics are converted into transcriptional output. Using human induced pluripotent stem cells (hiPSCs), we investigate mesoderm induction, a developmental process in which Wnt/β-catenin signaling plays a dominant role. We established a live-cell imaging system to simultaneously monitor β-catenin dynamics and a TCF/LEF (TopFlash) transcriptional reporter activity in single hiPSCs. We find that transcriptional output corresponds to the temporal dynamics of nuclear β-catenin rather than to its steady level: cells with comparable nuclear β-catenin can produce markedly different transcription when the time course of accumulation differs. Perturbation of the actin cytoskeleton changes β-catenin nuclear retention and reshapes its dynamics, with corresponding changes in transcription, identifying actin-dependent mechanics as a regulator of signaling dynamics. In addition, we show that cell division emerges as a further physiological factor that reshapes nuclear β-catenin dynamics and tunes transcription. We are extending our findings across multiple cell lineages to test their generality.

Together, these results indicate that the temporal structure of nuclear signaling, rather than transcription-factor abundance, is a key determinant of Wnt-dependent transcription, and that actin-based mechanics and cell division actively shape this temporal structure. This reframes Wnt signaling as a mechanically tunable system in which transcriptional output is governed by nuclear β-catenin dynamics.POSTER #25: MOVIES: Microfluidics and Optogenetics for Vivo Imaging of Embryos at Single-cell resolution Yang Yang, Chandel Angad Singh, Benjamin Caputo, Arya Desai, Ertuğrul M. Özbudak Cell and Developmental Biology, Feinberg School of Medicine, Northwestern University Contact: yangyang2025@u.northwestern.edu Live imaging has transformed zebrafish research by enabling the visualization of cellular signaling and tissue morphogenesis in real time. However, the large yolk volume and continuous embryonic movements makes it challenging for long-term, high-resolution imaging and quantitative single-cell analysis in embryos. Here, we introduce a microinjection-based strategy that generates yolk-reduced “skinny” embryos while largely preserving normal embryonic morphology and developmental integrity. These embryos are ideally suited for our integrated platform enabling long-term live imaging, high-resolution multichannel microscopy, and single-cell tracking. The platform combines microfluidics with optogenetic manipulation to achieve precise spatiotemporal perturbation of embryonic signaling dynamics. The microfluidic system enables rapid and temporally controlled pharmacological interventions through sequential drug pulses, while an integrated temperature-control module permits acute temperature shifts within minutes. In parallel, the optogenetic component allows targeted spatial and temporal modulation of signaling molecules. Using this platform, we simultaneously monitored segmentation clock activity and ERK signaling dynamics in zebrafish embryos and investigated how the expression of relevant genes regulates somitogenesis. Our single-cell analysis further provides accurate spatial mapping of embryonic tissue cells. Importantly, application of this workflow demonstrates that yolk reduction does not significantly perturb zebrafish embryogenesis, as evidenced by unchanged somite numbers and preserved segmentation clock periodicity.

Biological oscillators, from cell cycles to circadian clocks, achieve precise timing despite molecular noise. Theory predicts that sustaining such precision in far-from-equilibrium dynamics demands increased energy dissipation, reflecting fundamental thermodynamic trade-offs across biological and physical information processing, including oscillators, sensors, proofreading, and computing. However, directly testing this energy-precision relationship remains challenging due to the difficulty of systematically modulating energy while measuring precision across large ensembles of microscopic, stochastic oscillators. Here we developed a high-throughput droplet-microfluidics platform to reconstitute mitotic oscillator from Xenopus laevis egg extracts in thousands of cell-sized compartments spanning a broad free-energy landscape. These sub-nanoliter artificial cells isolate the cyclin-dependent kinase 1 (Cdk1) phosphorylation–dephosphorylation network underlying precise early embryonic cleavage cycles and enable controlled, systematic tuning of ATP levels, while minimizing downstream energetic loads. We find that oscillation speed depends non-monotonically on ATP, peaking near physiological levels and declining towards both high and low extremes before terminating at bifurcation points. In contrast, temporal precision improves monotonically with ATP, reaching levels adequate to sustain synchronized embryonic cycles. These results provide the first direct experimental evidence that mitotic timing is shaped by free-energy budgets and reveal an optimal energetic regime balancing speed and precision. Our droplet platform establishes a programmable framework for probing thermodynamic limits in nonequilibrium biological dynamics.

Understanding how cells transition between states — through the cell cycle, differentiation, or phenotypic switching — requires methods that capture genuine temporal dynamics rather than static snapshots. Current approaches each fall short in complementary ways. Pseudotime methods order fixed cells but recover only rank, not rate, and presume monotonic progression along a single axis. RNA velocity yields a vector field, but operates at the transcriptomic level, where mRNA is a noisy proxy for the protein activities that actually drive cell behavior. Live-cell imaging captures real dynamics, but is fundamentally limited to the one or two fluorescent channels that can be imaged simultaneously. The high-dimensional, protein-level dynamics that govern cell behavior thus remain largely inaccessible.

I will present a framework that infers protein-level velocity fields by combining live-cell imaging with endpoint multiplex immunofluorescence staining of the same cells. The key insight is that the cell-state manifold is locally coherent: cells that are neighbors in a low-dimensional embedding share similar dynamics, which lets tangent-space information be transferred across modalities. Continuous trajectories reconstructed from live imaging in a morphology–reporter space provide local velocity directions; a projection then transfers these velocities onto the high-dimensional staining feature space, producing a velocity field over protein states from purely endpoint measurements.

Simulating trajectories in this inferred field recovers dynamics consistent with biological expectation, including a perpendicular axis along which proliferation and mesenchymal programs trade off — providing direct support for the Go-or-Grow hypothesis of mutually exclusive proliferation and migration. Because new proteins can be added simply by extending the staining panel, with no need for a matching live reporter, the approach offers a general and scalable route to dynamically resolved, high-dimensional protein-level analysis of cell-state transitions.POSTER #28: Koopman Analysis of Genomewide Models Reveals Collective Modes Swapnil Dutta1, Wenqing Yu2, Zhiqian Zheng1, Amitava Giri1, Jianhua Xing1 1. University of Pittsburgh, 2. Peking University Contact: zhz187@pitt.edu Living cells are high dimensional dynamical systems whose behavior emerges from interactions among many genes and regulatory processes. Recent single cell modeling methods can infer cellular vector fields from transcriptomic data, but the resulting models remain difficult to interpret. Direct simulation shows how trajectories evolve, but not which dynamical components control different stages of the process or how gene groups contribute to these changes. In cell cycle checkpoint dynamics, progression, slowing, and recovery are distributed across many regulatory genes rather than controlled by a single variable.

Here, we use extended dynamic mode decomposition as an approximation of the Koopman operator to analyze cell cycle dynamics reconstructed from single cell RNA-seq data. In this view, a nonlinear cellular trajectory is represented as a sum of modes, each with its own timescale and gene-level contribution. This allows us to ask whether high dimensional learned vector fields can be described by a smaller number of interpretable dynamical components. The analysis can be viewed as a nonlinear extension of normal mode analysis for potential-gradient governed systems.

We apply this idea to cell cycle models inferred by a new version of Dynamo, focusing on A549 and RPE1 cells and 399 cell cycle-related genes. A reduced mitotic control model is used as a reference for interpreting how Koopman spectra and modes reflect known cell cycle dynamics. Preliminary analysis of the A549 model suggests that many modes are fast decaying, while a smaller subset captures slow relaxation of cell states toward G1/S and G2/M arresting states. These modes differ in timescale, reconstruction contribution, and gene-level organization. Cell cycle marker genes from the same phase tend to show similar phase and amplitude patterns within important modes, and initial enrichment analysis links different mode groups to DNA replication, DNA damage repair, and stress-related signaling.

In the literature Koopman analysis and related methods are widely used for reconstructing nonlinear dynamics. Here, with a known high dimensional nonlinear cellular dynamics model, we use Koopman spectrum analysis to identify collective modes describing cellular relaxation dynamics and cell phenotype stability.

Thalidomide’s teratogenic effect that emerged in the late 1950s taught us the importance of understanding the mechanisms of limb development. While we have made huge strides towards understanding aspects of this process, such as the FGF signaling network involved in limb bud initiation and outgrowth, there are still critical unknowns left to resolve. One unknown is the role of basic Helix-Loop-Helix (bHLH) transcription factors (TF) in early limb development. Beyond research acknowledging their presence, little has been done to investigate these TFs. Here, we investigate one such bHLH TF her7 and its role in the Notch signaling network governing chondrogenic development prior to structures forming in the zebrafish pectoral fin. Through live-imaging, immunohistochemistry, fast time-lapse videos, in situ hybridization, and drug treatments, we have seen that her7 plays a critical role in the proper development of strong muscles by keeping the main chondrogenic gene, sox9, in check and in its own compartment. Understanding the cross-talk between myogenic and chondrogenic regions of the fin is critical and required to prevent another tragedy like Thalidomide, and we have discovered part of a complex signaling network that appears to govern this communication.

Somitogenesis, the process that establishes the segmented vertebrate body plan, proceeds through two modular steps: molecular determination of somite boundaries followed by cellular execution. Although oscillatory signaling encodes positional information at the determination front (DF), a definitive marker of this molecular boundary has remained elusive. Here, we identify nuclear alignment as a robust, quantitative readout of the DF in zebrafish embryos. Using high-resolution confocal and multiphoton live imaging, we developed computational pipelines to localize aligned nuclei with single-cell precision across the presomitic mesoderm. In wild-type embryos, a defined band of aligned nuclei consistently anticipates the position of future somite boundaries. By contrast, genetic (her1⁻/⁻; her7⁻/⁻) and pharmacological perturbations of FGF/Notch signaling or the cytoskeleton disrupt nuclear alignment with defective boundary formation. Quantification of cell dynamics reveals that regions of nuclear alignment correspond to local decreases in cell displacement and neighborhood exchange rate, consistent with localized tissue compartmentalization at predetermined boundaries. Together, these findings elevate nuclear alignment from a descriptive morphological feature to a structural decoder that links segmentation-clock output to the mechanics of boundary execution, providing the long-sought physical link between molecular determination and cellular execution of somite boundary formation.

Blood glucose homeostasis relies on the well-coordinated rhythmic activity of millions of islets throughout the pancreas. While glucose-driven rhythmicity in individual islets has been well characterized, the mechanisms underlying inter-islet coordination in vivo remain poorly understood. Using simultaneous multi-islet Ca²⁺ imaging and continuous glucose monitoring in live mice, we report that robust global synchronization of islets with a 15-minute period—a phenomenon that persists following sympathetic neuron or adrenal gland ablation, pointing to pancreas-liver crosstalk. global synchronization arises from GLP1r activation, either through agonisnt or hypoglycemia. Dynamically, transitional phase analyses showed inhibited activation of leading islets, and mathematical modeling revealed that this global synchronization stems from an oscillatory-to-excitable state transition. Collectively, these findings elucidate the in vivo mechanism underlying islet synchronization and its essential contribution to precise blood glucose homeostasis.

How cells generate discrete signaling outputs from continuous biochemical reactions is a fundamental question in cell biology. Here we show that SOS processivity, the ability of a single membrane-bound SOS molecule to catalyze the activation of many Ras molecules before dissociating, can produce discrete Ras activation states in confined membrane corrals through a general mechanism that does not require bistability or cooperative feedback. Building on membrane reconstitution experiments and computational modeling by Lee et al. [Sci. Adv. 10, eadi0707 (2024)] demonstrating that processivity underlies bimodal Ras activation, we develop a mathematical framework that explains how quantization arises and predicts when it will occur. Using a rule-based model of Ras activation, we first demonstrate that distinct SOS occupancy states can generate a characteristic RasGTP level, producing a multimodal distribution of Ras activity concealed by apparent bimodality. To identify the essential requirements underlying this behavior, we develop a minimal activator–target framework. Four dimensionless conditions jointly determine whether discrete states are observable and precisely where each peak appears: low activator number, resolvable peak spacing, timescale separation between enzyme and target dynamics, and unsaturated catalysis. When these conditions are satisfied, the target distribution takes the form of a Poisson-binomial mixture with analytically predicted peak positions. This framework is not specific to Ras/SOS and describes any system in which a small number of processive enzymes act on a confined target pool. Output patterns consistent with the framework's predictions have been described previously in other contexts, including single-enzyme partitioning experiments by Rotman [PNAS 47, 1981 (1961)] and stochastic gene expression studies by Cai et al. [Nature 440, 358 (2006)], suggesting that such discrete distributions may be more widespread than recognized. Allosteric positive feedback present in the Ras/SOS system shapes the relative weight of quantization levels by biasing occupancy toward higher states without shifting peak positions, while eliminating processivity abolishes quantization and collapses the distribution to mean-field behavior. The four conditions of the framework serve as a practical diagnostic for predicting whether discrete signaling levels will be observable in any given system. These results establish processivity-driven quantization as a general mechanism for discrete signaling in localized enzymatic systems, with potentially broad implications for understanding digital responses across cell biology.

Personalized forecasting in oncology requires models that are both mechanistically interpretable and effective with limited longitudinal data. A major clinical challenge is predicting whether an individual patient will respond to chemotherapy, radiotherapy, immunotherapy, or combination therapy before treatment begins. We present a framework that combines mechanistic tumor growth models with machine learning to predict treatment response in both murine and human cancer datasets.

The model library includes one-dimensional mechanistic formulations describing exponential growth, logistic growth, weak and strong Allee effects, and therapy-induced tumor death. These models were fitted to longitudinal tumor measurements from murine chemotherapy and radiotherapy studies, together with human clinical datasets including CAR-T therapy, platinum-based salvage chemotherapy for large B-cell lymphoma, and five chemo/immunotherapy clinical trials in non-small cell lung cancer (NSCLC) and bladder cancer.

Model performance was evaluated using both goodness-of-fit metrics and the ability to predict treatment outcome (durable control versus relapse/progression). Model-derived parameters were subsequently used as candidate biomarkers for machine-learning classifiers, including logistic regression, random forest, gradient boosting, linear discriminant analysis, and support vector machines, with Elastic Net feature selection to reduce overfitting. The framework achieved excellent predictive performance in murine datasets, with balanced accuracy and AUC values approaching 1.0, while demonstrating encouraging predictive performance across all clinical datasets.

These results demonstrate that combining low-dimensional mechanistic models with machine learning enables accurate, interpretable, and data-efficient prediction of treatment response. The proposed framework provides biologically meaningful biomarkers that may support personalized treatment planning and adaptive oncology strategies.

During early embryogenesis in many species, cell cycle periods remain nearly constant until the Mid-Blastula Transition (MBT), after which they elongate due to the emergence of gap phases and the activation of checkpoint pathways. Both the DNA-to-cytoplasm ratio and the nuclear-to-cytoplasmic volume ratio have been proposed as regulators of this transition in vivo and in vitro. However, the relative contributions of these mechanisms, and whether they act independently or through interconnected pathways, remain unresolved. Here, we provide evidence supporting contributions from both mechanisms by independently controlling cytoplasmic volume and initial DNA concentration in Xenopus laevis egg extract in cell-free microfluidic droplets. Droplets exhibited self-sustained Cdk1 oscillations, nuclear growth, and DNA replication without division.

We identified two distinct components of period elongation: a gradual elongation correlated with nuclear volume, and an abrupt elongation associated with maternal resource depletion. Cell cycle periods increased progressively with nuclear size. Consistent with our previous findings of elevated nuclear Cdk1 activity during interphase, we propose that nuclear accumulation of cyclin B1-Cdk1 complex enables it to reach the mitotic entry threshold, thereby linking nuclear size to period duration. Droplets with similar nuclear sizes but larger cytoplasmic volumes shortened periods, likely by increasing the pool of import factors that facilitate cyclin B1-Cdk1 nuclear translocation. Perturbations inhibiting nuclear translocation elongated cycle periods, while reducing nuclear size shortened them.

Varying cytoplasmic volume and initial DNA amount revealed that maternal resources limit DNA replication capacity and nuclear growth, correlating with a switch in Cdk1 activation dynamics. After this switch, Cdk1 activity patterns resembled the somatic cell G2-to-M transition, suggesting checkpoint emergence. Chk1 inhibition shortened periods throughout cycles, indicating that Chk1 is active even before the abrupt transition, possibly due to incomplete DNA replication. Notably, the nuclear size-period length relationship persisted despite Chk1 inhibition, indicating that this gradual elongation mode operates independently of Chk1 signaling.

Together, our results suggest that before MBT, cell cycle duration is primarily governed by nuclear size, with cytoplasmic volume compensating to maintain rapid synchronous early divisions. The latter reflects a Cdk1 dynamic shift driven by maternal resource depletion, corresponding to MBT timing when scaled to embryo volume.

Detection of micro- and nanoplastics (MNPs) in human tissues has raised growing concern about their biological effects on tissue and cell function. While previous studies have examined MNP-cell interaction, most focused on limited cell and plastic types. Here, we present a comprehensive, quantitative investigation into how different types of nanoplastics (NPs) associate with and affect diverse cell types under physiologically relevant conditions. Using microfluidic-calibrated fluorescence microscopy, we quantify NP accumulation in cells in vitro and match cellular NP concentrations to levels reported in human tissues. While cell-associated NPs could be gradually released in vitro, they persist in vivo for over one month without detectable reduction in a mouse model. Across epithelial, endothelial, fibroblast, and immune cells, we found that NP exposure at these levels most strongly affects immune cell proliferation, with sensitivity varying by cell type. NP exposure also reduces motility in T cells and fibroblasts, with more complex effects observed in macrophages. Mechanistically, NP-cell association and trans-epithelial transport involved not only classical endocytic regulators but also pathways related to ion and water transport. Notably, NP association and release were highly sensitive to the extracellular fluid environment within the physiological range. By testing inhibitors of these pathways, we identified molecules that reduce NP-cell association and promote release. We further compared common NPs found in human samples and widely used in research: polystyrene (PS), polyethylene (PE), and polypropylene (PP). Although these NPs similarly impaired proliferation and motility, they showed markedly different cellular association and release dynamics. These findings reveal the impact of NPs on tissue cell functions and uncover novel regulatory pathways, establishing a quantitative framework for studying NP-cell interactions in biologically relevant conditions.