The activation of inflammasomes is a crucial component of the early immune response to pathogens and initiates a form of inflammatory programmed cell death called pyroptosis. During infection, a cytosolic inflammasome-forming protein sensor detects a pathogen to assemble the inflammasome complex, which subsequently activates the protease Caspase-1 (CASP1). CASP1 processes Gasdermin D (GSDMD), inducing pyroptosis through pore formation in the plasma membrane, while also facilitating the release of proinflammatory cytokines, such as IL-1β and IL-18. Adenovirus (AdV) is a common pathogen that causes inflammatory symptoms by infecting multiple mucosal epithelial tissues in the respiratory tract and intestinal tract such as the nose, mouth, and eyes. We wanted to test whether AdV infection could activate one of the main inflammasome sensors in human conjunctival epithelial cells (hCjE cells), which is NLRP1. However, we found that upon AdV infection, NLRP1-mediated cytokine release is absent, suggesting that CASP1 signaling is suppressed. Interestingly, despite the loss of IL-1β and IL-18, pyroptosis remains unaffected. Recent studies indicate that in the absence of CASP1, inflammasomes can activate Caspase-8 (CASP8), leading to the cleavage of Caspase-3 (CASP3) and Gasdermin E (GSDME), resulting in an alternative, incomplete form of pyroptosis. Thus, I hypothesize that during AdV infection, host cells are still able to induce pyroptosis by activating the alternative CASP8-GSDME pathway. To test this hypothesis, we generated and validated genetic knockouts of CASP8, GSDMD, GSDME, and CASP3 in hCjE cells to assess their roles in pyroptosis during AdV infection. These findings will provide new insights into viral immune evasion strategies and inflammasome regulation in epithelial cells.

Inflammasomes are cytosolic innate immune complexes that initiate pyroptotic cell death and the release of inflammatory cytokines. Inflammasomes are a critical component of the host innate immune response to viral pathogens. The inflammasome-forming sensor NLRP1 functions in barrier defense against a diversity of viral and bacterial pathogens, necessitating multiple modes of pathogen recognition. For instance, NLRP1 directly senses viral infection by detecting viral protease activity. NLRP1 is also activated indirectly by the ribotoxic stress response caused by radiation or toxins. Moreover, NLRP1 has been proposed to directly bind dsRNA. However, it is now understood that dsRNA-induced NLRP1 activation also requires p38-mediated phosphorylation. Thus, it is unclear whether NLRP1 directly or indirectly senses dsRNA. To address how dsRNA activates NLRP1, we reconstituted the NLRP1 inflammasome in inflammasome-deficient 293T cells. We found that reconstitution of the minimal NLRP1 inflammasome responds to viral proteases and other activating stimuli but not to dsRNA. This suggests that NLRP1 is insufficient to respond to dsRNA and instead requires uncharacterized host cofactors. We then hypothesized the NLRP1 response to dsRNA is an indirect event that requires upstream sensing events by canonical dsRNA receptors, and we found that co-expression of RIG-I or MDA5 restores NLRP1 responsiveness to dsRNA in 293T cells. We further investigated this pathway in the context of pathogen infection. During viral replication, dsRNA is generated, and the host has evolved mechanisms to detect it. Since viral dsRNA sensing is detrimental to viral replication, viruses have evolved strategies to evade detection. Notably, influenza A virus (IAV) encodes NS1, a protein that limits dsRNA accumulation. To investigate how IAV potentially counteracts NLRP1 activation by dsRNA, we transfected NS1 into 293T cells reconstituted with the NLRP1 inflammasome system and observed that NS1 significantly attenuated dsRNA-induced NLRP1 activation.

Tetralogy of Fallot (TOF) is the most common cyanotic congenital heart defect and consists of four structural defects that prevent babies’ hearts from delivering oxygenated blood to their body. When life saving surgeries correct these defects, the resulting scar changes the way the heart conducts electrical impulses, causing abnormal heart beats later in life. These abnormal heart beats, arrhythmias, often present as sudden cardiac arrest. Due to the high risk of arrhythmias in patients with repaired TOF, it is clinically important to understand the exact mechanisms causing them. These mechanisms provide insight that is essential to developing personalized methods for preventing arrhythmias. In our lab, we use late gadolinium enhanced MRI scans from TOF patients to create personalized computational models of their heart and particular scar distribution. We then attempt to induce arrhythmias in our models, which are individualized and represent subcellular and cell-scale electrophysiological phenomena. These models are useful because they allow us to study arrhythmia mechanisms noninvasively. We expect that patients whose computational models are susceptible to arrhythmias will also be more likely to experience arrhythmias in real life. We also aim to use the mechanistic insights from these simulations to determine new ways of predicting arrhythmia risk in this vulnerable patient population. Our results should predict what subset of patients would benefit from invasive preventative procedures, and help give patients with TOF a better understanding of their personal risk with or without those procedures. We hope to use our methods and results to create a simple and accessible arrhythmia risk stratification tool.

Treatment-related cardiomyopathy is a significant cardiotoxic complication for cancer patients treated with chemotherapy or radiotherapy and a leading cause of premature morbidity in childhood cancer survivors. Predicting a patient’s cardiomyopathy risk could help clinicians intervene early but is not possible with standard echocardiogram analysis methods. Preliminary research at the CardSS lab demonstrated that a deep convolutional neural network has modest success at predicting a pediatric patient’s risk for developing CM but is significantly limited by insufficient pre-diagnosis data for training, impairing its ability to learn generalizable disease progression patterns. This research aims to develop a generative AI model that generates synthetic echocardiogram data for training to improve the prediction model’s ability to learn distinctive patterns representing cardiomyopathy risk. By training on a longitudinal dataset containing echocardiograms from several cardiomyopathy stages before diagnosis, we aim to produce synthetic echocardiograms conditioned on specific classes: 0-1 years before diagnosis, 1-3 years before diagnosis, cardiomyopathy present, and control. Thus far, I have preprocessed echocardiogram data and implemented three experimental diffusion model architectures to investigate how the addition of a cross-attention layer to the encoder, bottleneck, and decoder regions of the model affects its ability to produce echocardiograms of different classes. I also implemented an analysis pipeline that calculates the Fréchet Video Distance (FVD), Structural Similarity Index Measure (SSIM), and Peak Signal-to-Noise Ratio (PSNR) between two sets of echocardiograms, which provide measures of image/video similarity. Using this pipeline, I am evaluating two key standards for synthetic data—intraclass fidelity and interclass separability—to quantify each model’s ability to generate data that is (1) representative of its class and (2) distinct from data produced for another class, and how these metrics change as training progresses. Preliminary data has shown that these models are producing synthetic echocardiograms that closely resemble real echocardiograms, but inconsistently.

Vibrio parahaemolyticus is a gram-negative marine bacterium that causes acute gastroenteritis in humans generally following the consumption of raw or undercooked shellfish. Mice are highly resistant to many human gut pathogens, including Vibrio, Salmonella, and Shigella spp., which has hindered our understanding of bacterial pathogenesis, immunity, and the development of therapeutics. Inflammasomes are cytosolic innate immune complexes that assemble in response to pathogen infection or harmful stimuli. Once the inflammasome is assembled, inflammatory caspases like caspase 1 are activated, driving a lytic cell death termed pyroptosis and the maturation and release of pro-inflammatory cytokines (i.e., IL-1β, IL-18). Inflammasomes have recently emerged as a necessary mediator of mouse resistance to Shigella and Salmonella, suggesting that inflammasomes may also be the cause of mouse resistance to V. parahaemolyticus. Consistent with that possibility, our preliminary data suggest that inflammasomes prevent intestinal inflammation in mice infected with V. parahaemolyticus, although the mechanism of protection is unknown. To identify the inflammasome(s) responsible for mouse resistance, I reconstitute specific murine inflammasomes in HEK293T cells, which lack most components of the inflammasome pathway. Then, I assess their activation in response to V. parahaemolyticus infection. Our previous findings demonstrated that V. parahaemolyticus robustly activates the mouse NAIP-NLRC4 inflammasome. However, we unexpectedly observed that V. parahaemolyticus infection also induces inflammasome activation in HEK293T cells even in the absence of NAIP-NLRC4 inflammasome reconstitution. This suggests the presence of an inflammasome-sensor in 293T cells that is responsive to V. parahaemolyticus infection. I am currently using inflammasome inhibitors and gene knockouts to identify this unknown inflammasome, which will ultimately aid in our understanding of host factors that mediate host defense against V. parahaemolyticus.

Rapid-growth wildfires disproportionately contribute to loss of life and destruction of property. Further improving our understanding of longer-term signals of impending fire-associated weather is crucial if we are to mitigate future destruction. Recent work compared local conditions, including surface wind and 100-hour dead fuel moisture (FM100) to fire growth (Murphy and Mass 2025). We investigate the evolution of larger scale weather patterns prior to rapid wildfire growth. Using two individual-fire-growth datasets, Fire Events Data Suite (FEDS) and Fire Events Delineation (FIRED), we separate fires by season, growth rate, and region. We conduct analyses of several meteorological variables for periods preceding maximum growth in rapid-growth wildfires. Using the European Centre for Medium-Range Weather Forecasts Reanalysis v5 (ERA5) dataset, we compare weather patterns at different heights in the atmosphere prior to maximum growth for fires of different growth rates and in different seasons, to identify any signals comporting to eventual fire extremity. We also consider how the patterns affect FM100 and near fire winds and the impacts of region of wildfire within California.

Hydrogels with tunable stiffnesses are a versatile method to study the interactions of human cells in vitro. These systems recreate human extracellular matrix (ECM) and capture the stiffness changes associated with a variety of biological processes and diseases, like cancer and cirrhosis. Photoresponsive chemistries allow light to be used to modulate the stiffness in these materials with high resolution. However, when creating more complex patterned gels with photomasks, bulk property analysis cannot capture the variation. To circumvent this and measure the stiffness of these complex gels, I performed rheology and fluorescence recovery after photobleaching (FRAP) to establish a correlation between diffusivity and stiffness in flood-illuminated gels. By finding and using the correlation, I am able to calculate the stiffness of the more complex patterned gels based off of their FRAP-derived diffusivity measurements. This method allows for better fine tuning of gels for use as a platform to study human cell growth through a range of stiffening events in multiple different parts of the body.