Undergraduate Poster Symposium

Savoring is the mindful attention to and appreciation of positive experiences, serving as a key strategy in positive emotion regulation (PER). Past research has shown that difficulties with savoring are linked to symptoms of depression and anxiety. However, there is little known about the function of savoring in daily life and whether its benefits vary depending on the presence of psychopathology. The present study utilized a 21-day Ecological Momentary Assessment (EMA) to investigate the use of emotion regulation strategies in daily life as part of a larger study. Participants completed four daily surveys for 21 days, reporting on emotional context (e.g., situational positivity/negativity), momentary savoring use, affective outcomes, and symptoms of depression and anxiety. Multilevel modeling revealed that individuals with higher depression and anxiety engaged in less savoring overall, but among those with depressive symptoms, savoring was more strongly tied to momentary positive experiences. Importantly, savoring predicted more successful emotion regulation and less negative affect, although the benefits were weaker among individuals with elevated depression. The study’s findings suggest that while individuals with higher depressive symptoms may not frequently engage in savoring, it may be particularly beneficial for them when they do. This work supports strides towards targeted savoring-specific interventions that might be more beneficial for individuals with depression, therefore contributing to more effective clinical prevention and intervention efforts for people with depression.

Absence epilepsy is a form of generalized epilepsy characterized by 5-10 second episodes of brief staring spells and an inability to recall events that occurred during the seizure. Genetic Absence Epilepsy Rats of Strasbourg (GAERS) is an animal model with robust correspondence to human clinical absence epilepsy, and it can be used to understand the unknown mechanisms of sensorimotor disruption. Similarly to humans, GAERS seizure events are tied to the propagation of spike-wave discharges throughout the cortex and thalamus. Previous research in awake, freely behaving GAERS using in vivo electrophysiological recordings has shown that auditory evoked potentials (AEPs) in the primary auditory cortex are preserved during spike-wave discharges despite a failed motor response. Since primary auditory processing is intact, we identified the anterior insula as a potential locus of disruption during seizures because of its extensive connectivity with sensory and motor regions and its role as an associative hub. Using local field potential recording, we compared AEPs in the insula of GAERS at baseline and during seizures for a conditioned auditory lick-response task. We also investigated the effects of satiation on the AEPs to dissociate motor planning and motivation in the insula. We found stable oscillatory AEPs in the Wistar’s insula. In GAERS at baseline, this oscillation was unstable and significantly distorted. During GAERS seizure, this oscillation was further altered. In all conditions, the amplitude and power of insular AEPs were significantly reduced without changes in waveform. These findings suggest that seizure-induced AEP changes in the insula obstruct sensory integration and point to an important role of oscillatory activity in the insula for normal, conscious processing.

Understanding the neural mechanisms that drive social cooperation is crucial for advancing our knowledge of social behavior and developing treatments for conditions such as autism spectrum disorder and early-life stress. Animals, such as rats, can model these mechanisms, although their behavioral variability makes analysis inherently challenging. While reinforcement learning (RL) has emerged as a promising framework for modeling animal behavior, multi-agent RL algorithms have yet to be applied to the study of social cooperation. Such models could offer insight into the neural computations underlying these behaviors. To address this gap, we analyzed the behavior of pairs of rats performing a cooperative task, quantified using pose estimation methods, to identify strategies and behaviors predicting success. Specifically, we observed two primary strategies: (a) waiting for the partner and (b) synchronized movement. We also identified two social interactions that influenced cooperation: gazing and physical interaction. These behavioral metrics were compared across rat cohorts varying in training history, social familiarity, and environmental visibility. Notably, we found that gazing increases in pairs with lower social familiarity and decreases during successful trials, suggesting that gaze serves primarily as a mechanism for social recognition rather than real-time coordination. To better understand the underlying cognitive mechanisms, we are training a model using the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, which leverages centralized critics to guide decentralized decision-making. By aligning the model’s learned strategies with observed behavior and neural recordings during key cooperative moments, we aim to uncover interpretable links between computational principles and brain function. This approach will not only test the validity of RL-based models in social contexts but also provide a biologically grounded framework for understanding how cooperation is represented in the brain. Ultimately, our findings may offer novel insights into how disruptions in social learning circuits contribute to psychiatric conditions marked by social deficits.

The many emotions we feel throughout our daily lives are constantly changing. Both unconscious and subconscious emotional experiences dictate the ways we feel and think. The study of affective state dynamics allows us to gain more insight into how our emotions change and helps us better understand how the constant patterns and fluctuations within our emotions define us. This research is critical as it will allow for greater insight into more affective mental health traits and symptom-related diagnoses. Affective dynamics can vary significantly across individuals when watching different movies, but there has been limited research on the correlation between affective dynamic variation and heterogeneity in mood disorders. In this study, we collected behavioral valence data during movie watching from 200 participants diagnosed with varying levels of depression to test whether different affective dynamics at the individual level relate to distinct symptoms of major depression. Participants were given standard questionnaires that quantified their trait levels of depression, general anxiety disorder, behavioral apathy, and anhedonia. After watching two 5-minute movies with emotional content in an fMRI scanner, participants were tasked with watching the same two movies again, this time while providing real-time ratings of their affective state. By isolating ratings across all participants based on the levels of their traits, we conducted both group-wide and individual participant analyses to determine if there was a statistical correlation between certain segments within each movie and higher symptom severity for each trait. Statistical analyses used to determine this correlation included temporal analysis across each movie, feature engineering, and data-driven event segmentation. Our findings suggest that within specific segments of each movie, variability in affective dynamic ratings across participants may be used to identify reliable patterns that link individual emotional experiences to distinct depressive symptoms.

Autism Spectrum Disorder (ASD) affects millions worldwide, yet its neural underpinnings remain poorly understood. Understanding the complex brain circuits of ASD requires powerful tools to analyze neural activity patterns. Machine learning, particularly deep learning architectures, offers unprecedented capabilities for understanding complex systems, such as the brain. In the present study, we employ a transformer-based machine learning approach to investigate how neural activity gives rise to autism-like behaviors using genetically modified mouse models. We analyze wide-field calcium imaging data from mice with mutations in ASD-associated genes, including GRIN2A, GRIN2B, and MECP2. We developed a Brain Vision Transformer (BrainViT) architecture that processes spatiotemporal brain activity. This can be used to discriminate between the neural dynamics of the genetic ASD mouse models and wild-type controls. Our preliminary data suggest that the transformer model differentiates between the neural activity patterns of mice with autism-associated mutations and wild-type mice, achieving classification accuracies substantially above chance. These findings indicate that mice with autism-associated mutations may exhibit shared neural activity patterns that differ from those of typical mice. These data suggest common circuit-level disruptions in ASD models despite different genetic alterations. This machine-learning approach opens new avenues for understanding the neural basis of autism.

Vision relies on the conversion of light into an electrical signal within rod and cone photoreceptors and the subsequent transmission of light-dependent responses at synapses between retinal neurons. Many retinal neurons, including photoreceptors, bipolar cells, and amacrine cells, signal via graded changes in membrane potential with high rates of continuous synaptic vesicle release. These vesicles fuse with the plasma membrane to release neurotransmitters and are subsequently recycled through endocytosis. The endocytic process requires the scission of the newly-formed vesicle, a step commonly mediated by GTPase proteins called Dynamins (Dnms). Dnm1 and Dnm3 are enriched in neurons, while Dnm2 is expressed ubiquitously and appears to mediate housekeeping functions. In rods, loss of Dnm1 and Dnm3 leads to impaired synaptic function, synapse degeneration, and cell death, suggesting that Dnm2 alone cannot support cellular function. Whether other retinal neurons depend similarly on neuronal dynamins remains unclear. Here, we investigate dynamin function in rod bipolar cells and VIP+ amacrine cells. Dynamins were selectively deleted using floxed alleles in rod bipolar cells of BAC-Pcp2-IRES-Cre mice and VIP+ amacrine cells of VIP-IRES-Cre mice. We analyzed the effect of dynamin deletion via immunohistochemistry, electroretinography (ERG), and electron microscopy. In rod bipolar cells, the combined loss of Dnm1 and Dnm3 had no apparent effect, suggesting that these cells are less dependent on neuronal dynamins. The additional loss of one Dnm2 allele led to progressive synaptic structural abnormalities and reduced ERG b-wave amplitudes by six months, indicating impaired synaptic function. Despite these changes, rod bipolar cell density remained intact. Ongoing studies of VIP+ amacrine cells investigate the structural and functional consequences of dynamin loss in these cells. Collectively, our results demonstrate diverse requirements for specific Dynamin isoforms among retinal cell types. These findings advance our understanding of how endocytic mechanisms support neurotransmission across diverse retinal circuits.

Ambiguity is a pervasive feature of language, so large language models (LLMs), similarly to humans, must be able to handle ambiguity to process language adequately. In this work, we investigate how well LLMs capture one specific type of ambiguity: the use-mention distinction, which differentiates using a word to refer to its definition (e.g., “cats have four legs”) from mentioning that word to refer to the word itself (e.g., “cats has four letters”). The universality of this distinction makes it well-suited for a systematic study of ambiguity, because all words can be used or mentioned, whereas most other forms of ambiguity apply to only a narrow category of words. To understand how LLMs process the use-mention distinction, we developed a dataset of 6000 sentence pairs, which tests whether LLMs recognize four different linguistic cues that disambiguate a use or mention case (for example, quotation marks around a word usually indicate that the word is being mentioned). We then tested Meta Llama 3.1 8B, a recent LLM, using this dataset, and it performed very well, achieving an accuracy of 82% across all cues. These results suggest that Llama captures the ambiguity of the use-mention distinction and can determine the most likely interpretation based on context and linguistic cues. This research opens the door for further experimentation to better understand how LLMs deal with ambiguity more broadly, and potentially to suggest hypotheses regarding the human mind and how it processes language.

Emotional arousal can play a critical role in memory, but most research has used static stimuli that fail to reflect how arousal naturally fluctuates during real-life experiences. We asked how arousal affects memory for the overall structure of an experience (event memory) versus specific contextual associations (associative memory). Participants listened to a naturalistic spoken story while undergoing fMRI and, 24 hours later, completed free recall and associative memory tests. A separate group of online participants provided continuous arousal ratings, which we linked to scene-level memory performance. We found that higher arousal during encoding was associated with worse event memory, while associative memory was unaffected. This suggests that in emotionally arousing situations, people may remember specific details but forget the broader context—potentially contributing to behaviors like risky alcohol use.

While multi-photon imaging provides unparalleled in vivo imaging capabilities, behavioral tasks that can be used in combination with multi-photon imaging are limited due to the head-fixed requirement. This limitation could be overcome by using a system such as MouseGoggles, a novel immersive virtual reality headset for mice. We are interested in using and developing head-fixed behavioral paradigms that will allow for circuit-specific investigations of innate and associative threat coding and are collaborating closely with developers of the MouseGoggles system to achieve this goal. A commonly used innate threat paradigm in mice is a visual looming stimulus, a rapidly growing dark disc that models an overhead predator. Previous work in our lab and others has shown a visual looming stimulus elicits innate defensive behaviors as well as physiological markers of fear and stress in mice. We thus began validating the visual quality of our MouseGoggles setup using a visual looming stimulus, which will allow us to compare neural and behavioral responses to previously published freely-moving and head-fixed data. We will test treadmill running speed as a measure of escape-driven behaviors and pupil diameter as an indicator of sympathetic nervous system activity, which is typically involved in the “fight or flight” response. Finally, we will be recording norepinephrine dynamics, which is a neuromodulator that is typically associated with arousal, fear, and stress, in the medial prefrontal cortex, a region of the brain associated with decision-making, fear learning, and anticipation. Future work will focus on developing MouseGoggles head-fixed behavioral paradigms of other innate and learned threats, which will be used to investigate threat responses on the single-neuron level utilizing two-photon imaging.

Mutations in CACNA1A, which encodes the alpha-1 subunit of P/Q-type calcium channels, have historically been linked to disorders like episodic ataxia type 2 (EA2) and familial hemiplegic migraine (FHM1) through mechanisms that disrupt channel function, neuronal signaling, and motor coordination in both mouse models and human patients. Additionally, growing evidence shows that CACNA1A mutations also contribute to a range of cognitive and neurodevelopmental impairments, including autism spectrum disorder (ASD), intellectual disability (ID), and developmental delay. However, the mechanism by which impaired channel function results in cognitive symptoms remains poorly understood. Proper channel function depends not only on the alpha-1 subunit, which forms the pore, but also on auxiliary subunits such as alpha-2/delta and beta, which promote membrane trafficking, stabilize the channel complex, and fine-tune gating properties. Auxiliary subunits support alpha-1 function and localization, but the impact of ASD and ID-causing alpha-1 mutations on channel and subunit expression remains an open question. Our study focuses on the D1634N missense mutation in CACNA1A, a variant identified in patients presenting with both motor and cognitive impairments but whose cellular effects on channel assembly and subunit stability have not yet been characterized. To address this, we transfected cultured cells with either D1634N-mutant or wild-type alpha-1, along with alpha-2/delta-2 and beta-4, and used immunocytochemistry and confocal microscopy to assess changes in subunit localization. We hypothesize that the D1634N mutation leads to reduced alpha-1 surface expression, which in turn decreases the recruitment and membrane localization of the auxiliary subunits. These findings will provide insight into how disruptions in subunit interdependence contribute not only to calcium channel dysfunction in motor disorders like EA2 and FHM1 but also to the broader spectrum of cognitive and neurodevelopmental symptoms associated with CACNA1A mutations, helping to explain the heterogeneity of phenotypes observed in affected individuals.

Aedes aegypti (yellow fever) mosquitoes pose a major public health threat by transmitting several mosquito-borne diseases that disproportionately affect low-income, tropical countries. Studying the neurobiology behind their blood-feeding behavior can help with addressing this crisis. However, the lack of genetic tools for precise targeting of specific neuronal regions of the mosquito brain posed a significant limitation to behavioral research. We propose a deep learning model, trained on scATAC-seq data, to predict cell-type specific chromatin accessibility. As many predicted highly accessibility regions are strongly linked with enhancer activities, the model can be used to generate cell-type specific enhancer sequences that have high activity in a particular cell type but low in others. A proof of concept was conducted in the mushroom body (MB), a critical region in the insect central brain regulating many important behaviors. After training, our model ran on a held-out dataset and showed strong (PCC: 0.81) predictive power compared to ground-truth chromatin accessibility values within the MB. We have used our model to generate enhancers with greater predicted specificity and activity compared to endogenous sequences. These in silico-derived sequences will be validated in the future via transgenic reporter assays in vivo. Future steps also involve leveraging the scRNA-seq data to generate an attention model that can identify and integrate cell-type information into the model, enabling a more flexible and on-target experimental design.

Childhood is a time period characterized by unique optimism, as children overestimate both their abilities and the likelihood of good things happening in the world. While optimism serves important developmental functions for children’s learning (Bjorklund, 1997; Lu et al., 2023), it declines between the ages of 3 and 9 (Leonard & Somerville, 2025). Though previous work has explained this decline through naturally occurring neurocognitive development as children age (e.g., Habicht et al., 2022; Stipek et al., 1984), it has overlooked concurrent changes in children’s social environment. As children get older, they are granted more independence, receive less support, and begin receiving more realistic and performance-based feedback when they enter formal schooling. In other words, children are increasingly introduced to environments with stakes, where optimism can be costly, a shift that may cause them to recalibrate their optimistic expectations. Here, we test the novel hypothesis that environmental stakes can cause children to be less optimistic. Specifically, we task 4-to-8-year-old children with making probabilistic predictions about which of two outcomes will occur, where one outcome is more desirable than the other. To investigate the causal effect of environmental stakes on children’s optimism, children are either told they will lose their tokens if they guess wrong (High Stakes condition) or that they will keep their tokens regardless of accuracy (Control condition). Preliminary data show that children in the High Stakes condition are less optimistic than children in the Control condition, suggesting that children temper their optimism when it is costly. The present work challenges our understanding of children’s optimism as an individual-level trait that evolves only with children’s neurocognitive development. Instead, we suggest that children’s optimism is an adaptive mechanism that is responsive to and malleable by high-stakes environments. These findings may have practical implications for creating developmentally appropriate educational and social environments.