Biology subjects

Lee, K.

Publications and source records attributed to Lee, K..

At least 55 records · Page 3Linked to original sources

bioRxiv2025

The ABCs of Cyanobacterial Glycogen: In Vitro Modelling of Glycogen Synthesis and Functional Divergence of Glycogen Synthases in Synechocystis sp. PCC 6803

Glycogen is the principal carbon reserve in Synechocystis sp. PCC 6803. We reconstituted its biosynthetic pathway in vitro--GlgC (Glucose-1-phosphate adenylyltransferase), two glycogen synthase isoenzymes (GlgA1, GlgA2) and the branching enzyme GlgB--to define how supply, polymerisation and branching set flux and product structure. GlgA2 shows higher specific activity and cooperates with GlgB-generated branched primers, whereas GlgA1 has higher substrate affinity and responds more to primer concentration. Product profiling links mechanism to architecture: GlgA1 produces more-branched glycogen, while GlgA2 yields longer, less-branched polymers, with GlgB biasing utilisation towards GlgA2. The complementary behaviours of GlgA1 and GlgA2 provide capacity for rapid accumulation versus steady-state maintenance and offer dynamic metabolic levers to tune glycogen content and architecture in cyanobacteria.

biochemistry↗

bioRxiv2025

Spared Nav1.8-Positive Nociceptors Drive Persistent Tactile Hypersensitivity After Sciatic Nerve Crush Injury in Mice

Peripheral nerve injury can lead to chronic mechanical hypersensitivity, yet the severity and persistence of pain are strongly influenced by the extent of axonal damage. Notably, partial sciatic nerve crush injury (PCI) produces persistent tactile hypersensitivity despite a less severe anatomical insult than full crush injury (FCI), yet the identity and post-injury state of the fibers that persist after PCI remain unclear. To define sensory neuron populations contributing to PCI-induced tactile hypersensitivity, we combined fiber-specific transgenic labeling (Thy1-YFP for A{beta} mechanoreceptors and Nav1.8-tdTomato for nociceptors) with pharmacological silencing using QX-314 coapplied with TRPV1 (capsaicin) and TLR5 (flagellin) agonists to selectively manipulate fiber subtypes. At day 7 after PCI, Nav1.8+ nociceptive terminals were still detectable in the hind paw. On day 30, acute silencing of TRPV1+ afferents transiently reduced mechanical hypersensitivity, indicating nociceptor activity in its maintenance. Whole-cell patch-clamp recordings of retrogradely labeled DRG neurons showed that remaining medium-diameter neurons exhibited reduced rheobase and increased action potential firings in response to step current injections. Besides, electrical stimulation of nociceptive fibers increased pERK expression in the spinal dorsal horn, indicating enhanced nociceptive signaling after PCI. Early ablation of TRPV1+ fibers with high-dose capsaicin during degeneration phase prevented the subsequent development of long-term tactile hypersensitivity. Collectively, our results suggest that spared nociceptors after PCI remain sensitized even during nerve repair, driving long-term tactile hypersensitivity. Targeting these spared nociceptive fibers after nerve injury may offer a potential strategy for preventing chronic pain associated with traumatic nerve injury. Graphical Summary O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=131 SRC="FIGDIR/small/687306v1_ufig1.gif" ALT="Figure 1"> View larger version (54K): org.highwire.dtl.DTLVardef@1331beorg.highwire.dtl.DTLVardef@26e75aorg.highwire.dtl.DTLVardef@1ecf05org.highwire.dtl.DTLVardef@a96e0b_HPS_FORMAT_FIGEXP M_FIG C_FIG

neuroscience↗

bioRxiv2025

Peripheral Nerve Transection Predominantly Drives Sympathetic Nerve Sprouting in Mouse Dorsal Root Ganglia

Sympathetic sprouting in dorsal root ganglia (DRG) is a feature of sympathetically maintained pain (SMP) following peripheral nerve injury, yet the factors determining its occurrence remain unclear. Here, we compare transection and crush injury models to determine if injury type or site influence sympathetic remodeling and pain. Using TH-IR immunostaining and Phox2b reporter mice to selectively label sympathetic fibers, we found that an L5 spinal nerve transection (SpNT) triggered robust sympathetic fiber sprouting and elevated norepinephrine (NE) levels in the DRG, correlating with a mechanical hypersensitivity reversed by chemical sympathectomy. In contrast, a partial sciatic nerve crush injury (PCI) produced long-lasting mechanical hypersensitivity without sympathetic sprouting or NE elevation and was unaffected by sympathectomy. Importantly, sympathetic sprouting was consistently more pronounced after transection injuries at both spinal and sciatic nerve sites, suggesting that injury type, rather than location, is a dominant factor shaping sympathetic remodeling. These findings establish nerve transection as a key driver of sympathetic sprouting and SMP, whereas crush-induced pain likely involves distinct non-sympathetic mechanisms. This distinction has important implications for pain subtype identification and treatment strategies.

neuroscience↗

bioRxiv2025

Stimulation with ECoG electrodes modulates cortical activity and sensory processing in the awake mouse brain

Electrical stimulation has been widely used to probe neural network properties and treat dysfunction. Electrocorticography (ECoG) electrodes, long used for activity monitoring, can also stimulate the brain in a minimally invasive and chronic manner. However, how cortical surface electrical stimulation impacts cortical network activity remains poorly understood. Using in vivo calcium imaging in the awake mouse brain with chronically implanted ECoG electrodes, we measured how electrical stimulation modulates the activity of visual cortical neurons, including during concurrent visual stimulation. We found that cortical surface electrical stimulation initially activates L2/3 neurons followed by prolonged inhibition lasting seconds after stimulation. Electrical stimulation suppresses the activity of neurons at their preferred grating orientation but enhances their responses to non-preferred visual stimuli, thereby reducing sensory feature selectivity. By measuring how electrical stimulation modulates the activity of inhibitory neuron subtypes including PV, SST, and NDNF interneurons, we propose a circuit model in which L1 NDNF interneurons are strongly activated by cortical electrical stimulation and, in turn, inhibit L2/3 excitatory neurons and PV interneurons through volume transmission of GABA.

neuroscience↗

bioRxiv2025

PIONEER: A Periplasmic Display Platform for Synthetic Biology-Based Screening of Genetically Encoded Protein Regulators

Periplasmic display in yeast is a promising but underdeveloped method for screening and studying genetically encoded biomolecules. We present PIONEER, a modular platform that localizes peptides, proteins, and nanobodies to the membrane-proximal periplasmic space of S. cerevisiae, enabling direct interrogation of membrane protein function. By optimizing secretion signals and display scaffolds, PIONEER ensures stable ligand retention and robust autocrine signaling. The ability to assess peptide and protein ligand activity through signaling, rather than binding alone, marks a key advance toward function-first screening. Applied to human G protein-coupled receptors (GPCRs), the system enables detection of surface expression, ligand activity profiling, and classification of nanobody regulators, including antagonists and conformational stabilizers. It also distinguishes intrabodies based on their functional effects and chaperone activity. Although demonstrated with GPCRs, PIONEER is adaptable to a wide range of membrane and soluble protein targets. As AI-driven design expands the space of candidate ligands and binders, this platform offers a scalable method for linking predicted sequences and their structures to biological function.

synthetic biology↗

bioRxiv2025

Pigs lacking Natural Killer T cells have altered cellular responses to influenza

It is increasingly recognized that innate T cells such as natural killer T (NKT) cells, mucosal associated invariant T (MAIT) cells, and {gamma}{delta} T cells play an important role in shaping adaptive immune responses following influenza infection or vaccination. This is largely through the multiple cytokines these cells release upon activation, which have downstream effects on the scope and magnitude of virus-specific T and B cells, and antibodies which form. Here, we examined the contribution of NKT cells using pigs, which are considered a highly translational model of human influenza A infection. CD1D-expressing and CD1D-deficient pigs that respectively possess and lack NKT cells were infected with the swine influenza virus H3N2 A/Swine/Colorado/23619/1999 (CO99), with or without prior mucosal immunization with a recombinant H3N2 A/Swine/Texas/4199-2/1998 (TX98) modified live vaccine encoding a truncated NS1 protein (TX98 NS1{Delta}126). Vaccination reduced virus load and pulmonary pathology by similar amounts in both genotypes. However, NKT cells status had a significant impact on the underlying immune response: Unlike following vaccination, virus-specific T cell expansion after infection was greater in CD1D-deficient than -intact pigs, indicating that NKT cells play opposing roles in different phases of the immune response. NKT cell-deficient pigs also had reduced T cells cuffing around airways and higher numbers of expanded T and B cell clones according to single-cell {beta} T cell receptor (TCR) and B cell receptor (BCR) sequencing. Using our newly developed porcine-specific {gamma} and {delta} chain primers, we characterized the pulmonary {gamma}{delta} TCR repertoire. This revealed a high frequency of expanded clones, especially within the CD2- subset, and a tendency for larger expansion in NKT cell-expressing pigs. Overall, our results indicate a homeostatic role for NKT cells on several important features of the influenza immune response, including the dynamics of T cell expansion and contraction.

immunology↗

bioRxiv2025

Sleep deprivation constrains dynamic configurations of integrated and segregated brain states impacting cognitive performance

The breakdown of cognitive control following sleep deprivation is widely recognised, but the physiological mechanisms and brain signatures that produce this vulnerability have not been resolved. Effective cognition relies on large-scale brain networks flexibly reconfiguring between states of integration and segregation. Here we combined functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and electrocardiography (ECG) collected during cognitive tasks under rested wakefulness, after sleep deprivation, and following a recovery nap to test the hypothesis that sleep deprivation constrains this dynamical repertoire and disrupts its physiological regulation. Using time-resolved functional connectivity and graph theory, we show that sleep deprivation increases the distribution of connections across networks, while reducing the temporal variability of between-network connectivity. Furthermore, dynamic fluctuations between integrated and segregated modes of network topology were dampened, with brain regions spending more time in intermediate configurations and showing greater instability of mode transitions. These alterations were tightly linked to behavioural impairment: participants who exhibited greater contraction toward intermediate topologies also showed poorer task accuracy and slower responses. Under well-rested conditions, thalamic activity peaked prior to transitions into integrated states and was suppressed during transitions into segregated states, consistent with a coordinating role in cortical dynamics. Sleep deprivation weakened and delayed this thalamic coupling. Finally, global and regional fMRI fluctuations were elevated after sleep loss, becoming decoupled from cardiac physiology while more strongly coupled to EEG delta power, further linking reduced arousal to constrained network flexibility. Together, these findings show that sleep deprivation narrows the brains dynamical repertoire, due to disrupted thalamic regulation and changes to the physiological integration with cortical networks.

neuroscience↗

bioRxiv2025

Single-cell analysis of brain-derived Toxoplasma bradyzoites reveals a novel cell cycle regulated by AP2XI-6.

Toxoplasma gondii prevalence is due, in part, to its ability to persist in hosts while retaining the capacity to transmit and recrudesce. A process that is poorly understood. Through single-cell RNA profiling of in vivo-derived bradyzoites, we discovered that they are heterogeneous and not G1-arrested, as expected from in vitro studies. Instead, they progress through two cell cycles that branch from a single G1. While in G1b, in vivo-derived bradyzoites express cyst wall proteins predicted to replenish the wall that shields cysts. One cell cycle is common amongst tachyzoites and in vitro- and in vivo-derived bradyzoites; the other is unique to in vivo-derived bradyzoites (BCC). We demonstrated that AP2XI-6, expressed in G1 in vivo, is dispensable for tachyzoite growth and cyst formation in vitro but necessary for encystation in vivo. We propose that AP2XI-6 is a driver of replication through the BCC that is required for cyst maturation and persistence.

microbiology↗

bioRxiv2025

Low-intensity transcranial focused ultrasound engages parvalbumin-positive GABAergic interneurons in a humanized mouse model of chronic pain: from electrophysiology to cellular investigation

BackgroundLow-intensity transcranial focused ultrasound (tFUS) offers high spatial specificity and deep brain penetration, showing great promise as a non-invasive stimulation technology for modulating brain activity and behavior. Recent studies show that specific tFUS parameters targeted to pain-processing brain circuits can significantly alter pain-related behaviors in rodent models and humans. However, a comprehensive understanding of how tFUS influences brain networks and cellular mechanisms is essential to optimize efficacy and facilitate safe translation to clinical pain therapies. ObjectiveWe aimed to evaluate the modulation of inhibitory neural circuits induced by tFUS of 40 Hz pulse repetition frequency (PRF) in a humanized mouse model of chronic pain, integrating local and network-level electrophysiological investigations, molecular analyses, and histological assessment to confirm safety. Methods and ResultsWe used a 128-element random array transducer for stimulation, along with a non-invasive and flexible 30-channel electroencephalography (EEG) to assess local evoked responses, topographical brain activity, and global brain dynamics including excitation and inhibition (E/I) balance. To further assess tFUS neuromodulation effects at the cellular level, we performed immunohistochemistry (IHC) analysis and found that tFUS significantly increased the activity of inhibitory neurons as indicated by elevated expression of Glutamate Decarboxylase 67 (GAD67) and Parvalbumin (PV). Finally, safety was evaluated in the same brain samples used for mechanistic analysis, with blinded histological assessment revealing no signs of tissue damage. ConclusionsThese findings provide new evidence that tFUS non-invasively engages PV GABAergic inhibitory circuits in a chronic pain mouse model, supporting its development as a robust neuromodulation strategy. TopicsChronic pain; Transcranial focused ultrasound; Non-invasive brain neuromodulation; GABAergic neural circuit modulation HighlightsO_LIMulti-modal assessment of low-intensity tFUS in a humanized chronic pain model. C_LIO_LI40 Hz tFUS enhances inhibition, mirroring optogenetic PV neuron activation. C_LIO_LIRepeated tFUS restored chronic pain-disrupted E/I balance. C_LIO_LIMulti-session tFUS upregulates GAD67 and PV interneuron expressions. C_LI

bioengineering↗

bioRxiv2025

Circadian phosphotimer generates time cues through PER dimerization-mediated trans-phosphorylation

The circadian clock generates autonomous molecular rhythms that regulate sleep cycles, chronotypes, daily physiology, and expression patterns of almost the entire genome. Temporal phosphorylation of PERIOD (PER) proteins creates a phosphotimer that is critical for clock function. Although previous studies show that PER stably binds the casein kinase CK1, cis-acting phosphorylation of PER by CK1 is not compatible with slow PER phosphorylation extended over 24 hrs. So how can PER phosphorylation be programmed in a slow and controlled manner? Here, we show that the timing cues are encoded by PER-PER dimerization-mediated trans-phosphorylation, which enables the necessary time delay in phosphorylation and slow phosphorylation. When PER dimerization is disrupted, PER phosphorylation and circadian rhythms are severely compromised. In mouse models with point mutations in the PER dimerization domain, circadian period is shortened to [~]20 hrs, and the phase of wake/sleep cycles is dramatically advanced, switching a nocturnal animal to a half diurnal animal.

neuroscience↗

bioRxiv2025

Specification of frequency criteria for secondary findings genes to improve variant classification concordance

PurposeThe return of secondary findings (SF) is well-established in clinical testing and is increasingly employed in clinical research testing. Variant classification can be challenging due to the lack of monogenic disease entity (MDE, defined as a gene-phenotype pair) specific variant classification recommendations. A key criterion for variant classification is disease allele frequency thresholds (DAFTs), which are not established for all MDEs recommended for the return of SF. MethodsWe calculated DAFTs for SF MDEs considering prevalence, gene contribution, and penetrance. ACMG criterion BS1 was set at the calculated DAFT value, with BA1 set at ten times the calculated DAFT. For genes associated with multiple SF MDEs without clear genotype-phenotype correlation, DAFT values were combined. GnomAD Grpmax filtering allele frequencies (FAF) for pathogenic/likely pathogenic classified variants were compared to calculated thresholds. ResultsWe determined BS1 and BA1 values for 58 SF MDEs (47 genes). No pathogenic/likely pathogenic variant Grpmax FAF was greater than the relevant MDE-specific BA1. ConclusionSetting BA1 and BS1 thresholds should improve variant classification consistency and reduce misclassifications. For SF MDEs without current ClinGen Variant Curation Expert Panel (VCEP) specifications, these frequency specifications can be used until full criteria are available.

genetics↗

bioRxiv2025

Cereblon-related mild intellectual disability disrupts response inhibition and uniformity of group--individual strategies

Temporal processing, including duration, is essential for survival and communication across species. Intellectual disability (ID), which has diverse causes, including Cereblon (CRBN), impairs duration discrimination. CRBN-related ID, link to abnormal cognitive behaviors, may disrupt both perception and behavior during duration discrimination. However, cross-species behavioral strategies, their variation with ID, and associated behavioral indices remain unclear. Here, humans and wild-type (WT) mice with typical intelligence, and CRBN knockout (KO) mice with ID, performed an auditory duration discrimination task with Long (10 s) and Short (2 s) cues. All groups distinguished stimulus durations, but latency-based strategies diverged. KO mice showed impulsivity and divergent responses at both individual and group levels, whereas WT mice and humans consistently delayed their responses by [~]2 s to the Short cue length, reflecting inhibition and convergent responses. In typical intelligence models, latencies for both stimuli clustered between 2-5 s, while in ID model depended on stimulus duration. Duration perception is a conserved cross-species capacity, while task-specific cognitive strategies are intrinsically preserved though variation emerges with ID. We suggest that latency indices dissociated inhibition from impulsivity, and convergence from divergence. We further suggest that CRBN-related ID preserves perceptual understanding but disrupts the shared behavioral language of typical intelligence.

neuroscience↗

bioRxiv2025

Systematic Growth Factor Profiling Platform for 3D Tumor Models Reveals Estradiol-Responsive Cellular Mechanisms of Immunotherapy Resistance

Current organoid culture systems face critical limitations: standardized growth factor formulations fail to capture patient-specific signaling requirements, and single-cell-type approaches overlook tumor-stromal interactions essential for understanding immunotherapy resistance. To address these challenges, we developed an innovative biofabrication platform that systematically integrates patient-derived three-dimensional (3D) cultures with comprehensive growth factor profiling across 128 combinations. Through systematic screening of 23 ovarian cancer patient samples and single-cell RNA sequencing, we identified two estradiol-responsive cellular populations that coordinate immunosuppression: a malignant cell fraction (MAL.PDCD5) and a cancer-associated fibroblast (CAF) fraction (FB.TNFSF10). MAL.PDCD5 cells suppress immune infiltration by downregulating antigen presentation, while FB.TNFSF10 cells promote immunosuppression through enhanced TGF-{beta} signaling. Spatial transcriptomic analysis revealed striking mutual exclusivity between FB.TNFSF10 cells and T/NK cells, providing evidence of active immune cell exclusion. Most significantly, FB.TNFSF10 abundance emerged as a robust predictor of immune checkpoint inhibitor therapy resistance across multiple cancer cohorts, independent of conventional biomarkers. This biofabrication platform provides a scalable framework for drug screening and biomarker discovery, with immediate applications in precision medicine for patient stratification and combination therapy development.

cancer biology↗

bioRxiv2025

Generalized brain-state modeling with KenazLBM

The large-scale functional state of a human brain remains difficult to characterize, much less predict. Regardless, techniques have been engineered to electrically neuromodulate the brain to treat a subset of neurologic and psychiatric disorders with moderate efficacy. Accurate characterization of a brains instantaneous functional state has stymied the development of more effective neuromodulation paradigms. Advanced computational methods are required to address this gap and enable large-scale neuroscience. Here we define the concept of generalized brain-state modeling across humans as Large Brain-State Modeling (LBM) and present KenazLBM as the worlds first example. KenazLBM can instantaneously characterize the functional state of a persons brain with raw iEEG data, and predict future brain-states. KenazLBM was trained on over 17.9 billion unique multichannel tokens from people undergoing intracranial electroencephalography (iEEG) recordings, and has learned to interrelate brain-states between people into a common interpretable topology. Most importantly, the model generalizes to unseen subject data with significant recording channel heterogeneity from the training set. We offer KenazLBM as a first generalized brain-state model to serve as a new paradigm of basic neuroscience inquiry and potential translation into neuromodulation therapeutics.

neuroscience↗

bioRxiv2025

Insights from U.S. beekeeper triage surveys following unusually high honey bee colony losses 2024-2025

In January of 2025, U.S. commercial beekeepers reported unusually high honey bee colony losses as they prepared colonies for almond pollination. Two industry groups launched nationwide surveys to document colony losses between June 2024 and March 2025 across all scales of beekeeping. This study analyzes these survey data to assess colony losses, estimate financial impacts, and identify correlations with beekeeper management practices and geographical locations. Unlike past surveys, commercial beekeepers experienced more severe losses than smaller-scale beekeepers during this period. Respondents, managing over half of U.S. colonies, most frequently cited Varroa mites as the cause for their losses. Varroa mites were followed by pesticides and pathogens in the case of commercial beekeepers and by queen failure and weather in the case of smaller-scale beekeepers. Although Varroa was the most frequently cited cause, losses did not significantly differ between users and non-users of amitraz, suggesting that rising amitraz resistance alone does not explain observed trends. Differences in protein and carbohydrate feeding frequencies also played a role in net losses. While colony loss rates and financial concern varied widely among respondents, commercial beekeepers understandably showed higher sensitivity to financial impacts, with concerns increasing linearly with loss severity. This study highlights the value of beekeeper surveys which, alongside direct analyses of bee samples and longitudinal studies, help identify effective management strategies and environmental risks. Such insights are crucial for addressing the leading causes of colony losses on a national scale, and ultimately aid in safeguarding honey bee health, pollination services, and agricultural production. HighlightsO_LIUnprecedented honey bee colony losses C_LIO_LIIndications of disease stress C_LIO_LIHigh economic pain for commercial beekeepers and growers C_LI

ecology↗

bioRxiv2025

Structures of dynamic interactors at native proteasomes by PhIX-MS and cryoelectron microscopy

Proteasome function depends on a network of transient interactions that remain structurally and functionally unresolved. We developed PhIX-MS (Photo-induced In situ Crosslinking-Mass Spectrometry), a structural proteomics workflow that stabilizes transient interactions in cells by UV-activated crosslinking to capture topological information. Applying PhIX-MS with cryo-electron microscopy (cryo-EM), we mapped redox sensor TXNL1 at the proteasome regulatory particle (RP), placing its PITH domain above deubiquitinase RPN11 and resolving its dynamic thioredoxin domain near RPN2/PSMD1 and RPN13/ADRM1, ideally located to reduce substrates prior to proteolysis. We also resolved chaperone PSMD5 bound to RP without the proteolytic core particle (CP) where its C-terminus inserts into the ATPase pore blocking CP binding. PhIX-MS and AlphaFold modeling tether ubiquitin ligase UBE3C/Hul5 along the RP placing its catalytic site above the RPN11 active site, enabling their coupled activities. Our integrative approach enables the localization of native, low-affinity protein interactions and is broadly applicable to dynamic macromolecular assemblies.

cell biology↗

bioRxiv2025

The Impact of Delayed Evacuation on the Quality of Human Fetal Tissue

BackgroundDigoxin and other agents are frequently administered to arrest fetal circulation before the evacuation of human fetal tissue (HFT) in pregnancy terminations to address concerns about fetal viability. However, the impact of delayed evacuation on HFT quality remains unknown. Analyzing HFT is critical for diagnosing pregnancies affected by fetal abnormalities and for driving progress in biomedical research. ObjectiveThis study aims to assess the effects of delayed evacuation following fetal circulation cessation on the quality of HFT for both diagnostic and research purposes. Study DesignHFT samples were collected from second-trimester dilation and evacuation (D&E) procedures, with and without agent injection approximately 24 hours prior, as per standard care protocols. We assessed multiple parameters relevant to diagnostics and research, including: 1) cell morphology, 2) cell proliferation, 3) apoptosis, 4) cell viability in culture, and 5) nucleic acid quality. To simulate in utero conditions and determine the timeline for tissue degradation, we incubated brain tissue obtained from D&E without induced demise at 37{degrees}C for up to 18 hours. ResultsWe analyzed 18 HFT samples from D&E procedures performed 19-25 hours after digoxin or potassium chloride (KCl) injection and compared them to 26 HFT samples from immediate evacuations without prior injection. Induced fetal demise resulted in 1) disrupted cell morphology, 2) decreased cell proliferation, 3) increased apoptosis, 4) reduced cell viability in culture, and 5) lower RNA quality. Despite these findings, all samples yielded DNA of sufficient quality for polymerase chain reaction (PCR). ConclusionD&E procedures performed after fetal demise induced by digoxin or KCl lead to decreased HFT quality, limiting its diagnostic and research potential beyond gross tissue evaluation and DNA extraction. Limiting the time between fetal circulation arrest and evacuation may improve HFT quality for clinical and research applications.

pathology↗

bioRxiv2025

Prediction of cellular morphology change under perturbations with transcriptome-guided diffusion model

Investigating the cell morphology change after perturbations with high-throughput image-based profiling is of growing interest, considering its wide applications in phenotypic drug discovery, including MOA (Mechanism Of Action) prediction, compound bioactivity prediction, and drug repurposing. However, the vast space of chemical and genetic perturbations makes it infeasible to fully explore all the potential perturbations with image-profiling technologies. Consequently, developing a powerful in-silico method to simulate high-fidelity cell morphological response under perturbations can reduce the experiment costs and accelerate drug discovery. Motivated by this, we proposed MorphDiff, a transcriptome-guided latent diffusion model for accurately predicting the cell morphology response to perturbations. We applied MorphDiff to two large-scale datasets, including one drug perturbation and one genetic perturbation cell morphology dataset covering thousands of diverse perturbations. Extensive benchmarking and comparison with baseline methods show the remarkable accuracy and fidelity of MorphDiff in predicting cell morphological changes under unseen perturbations. Furthermore, we explored the utilities of MorphDiff in identifying and retrieving the MOAs of drugs, which is a crucial application in phenotypic drug discovery. With the designed pipeline for MOA retrieval, we demonstrated MorphDiffs capability to boost the retrieval of the drugs MOAs (Mechanism Of Actions) by generating realistic cell morphology profiles. The average MOA retrieval accuracy of MorphDiff-generated morphology is comparable with that of the ground truth cell morphology, and consistently outperforms the baseline method and gene expressionbased retrieval by 29.1% and 9.7% respectively. We also validated that complementary information provided by cell morphology generated by MorphDiff can help discover drugs with dissimilar structures but the same MOAs. In summary, with its strong capabilities in generating high-fidelity cell morphology on unseen perturbations, we envision MorphDiff as a powerful tool in phenotypic drug discovery by accelerating the phenotypic screening of vast perturbation space and improving MOA identification.

cell biology↗