Undergraduate Poster Symposium

Developmental differences in cortical neuronal activation following peripheral inflammation

Peripheral inflammation has been linked to increased long-term risk of psychiatric and neurological disorders, particularly when immune activation occurs during childhood. However, the mechanisms through which peripheral immune signals alter cortical function remain unclear. This project examines how peripheral inflammation engages specific neuronal populations in the cerebral cortex and how developmental timing shapes that response. Using mouse models of gut and allergic lung inflammation, immune challenges will be induced in adult and early postnatal animals. Brain tissue will be processed using c-Fos immunohistochemistry combined with markers for defined cortical neuron subtypes. Confocal imaging and quantitative analysis will be used to map inflammation-responsive neurons across cortical regions. By comparing activation patterns between adult and early postnatal exposure, this project will determine whether early-life immune challenges recruit distinct or more extensive neuronal populations. These findings will provide a cellular framework for understanding how immune activation alters cortical circuits in a developmentally dependent manner and may clarify why early-life inflammation is associated with lasting changes in brain function.

Laminar mechanisms of attention in the primate visual cortex

Internal states continuously fluctuate and strongly modulate perception and attention to the surrounding environment, making them a major source of variability in neural and behavioral responses. In humans, facial expressions have been studied extensively and provide a rich, noninvasive window into these internal states. However, despite the growing use of marmosets (Callithrix jacchus) as a model for systems and translational neuroscience, very little is known about their facial expressions. We hypothesize that marmosets exhibit measurable facial micro-expressions whose dynamics reflect the visual content of natural scenes. To test this hypothesis, we developed an automated facial tracking pipeline using DeepLabCut to quantify subtle facial movements while marmosets freely viewed Teletubbies video stimuli. Animals were presented with three viewing conditions: the original movie, a frame-shuffled version of the same movie, and a spontaneous viewing control condition. We validated our tracking framework by comparing model predictions against manually annotated landmarks using pixel error analysis, demonstrating high accuracy and reliability. We then applied Principal Component Analysis (PCA) to the extracted facial features to identify dominant patterns of facial dynamics and reveal discrete, partially non-overlapping clusters associated with the different viewing conditions. Furthermore, we examined the relationship between facial dynamics and established physiological indicators of internal state, including pupil diameter and blink rate. Together, this work establishes an automated framework for quantifying facial dynamics in marmosets and provides a foundation for investigating how internal states interact with natural vision, attention, and scene understanding.

Understanding cannabis use in Borderline Personality Disorder

Borderline personality disorder (BPD) is a mental illness characterized by unstable moods, turbulent relationships, and impulsive behaviors, including substance use (Gunderson et al., 2018). Specifically, cannabis use (CU) has been strongly associated with BPD, though its interactions with symptoms remain poorly characterized (Trull et al., 2018). We examined the relationship between CU and symptom endorsement in individuals with current BPD. Additionally, we compared reported reasons for CU between participants with and without BPD.

​ We drew phone screen data from two prior studies and one ongoing treatment study. Interested adults were assessed for current BPD symptoms, substance use, and other eligibility criteria. Adults who were fully assessed for CU and current BPD were included (n = 646). We ran chi-square tests to evaluate 1) differences in CU endorsement between participants with and without BPD and 2) CU prevalence across four symptoms we hypothesized to be linked to CU in BPD. Furthermore, we used Fisher’s exact test (adjusted for eight tests) to compare endorsement of reported reasons for CU between groups. ​

We found that participants with BPD had significantly higher odds of endorsing CU than participants without BPD (OR = 1.94, 95% CI [1.40, 2.70], p < 0.001). However, for the four symptoms tested, CU prevalence did not differ significantly between individuals with BPD who endorsed each symptom and those who did not (p > 0.05 for all tests). Likewise, per-reason differences between groups were non-significant (p > 0.05 for all tests). ​

Analyses were cross-sectional and were likely underpowered given cohort imbalances. Moreover, high CU prevalence, comorbidity, and unmeasured symptom burden may have obscured group differences. Finally, phone screens captured only binary CU, without frequency, recency, or severity. ​

More robust data are needed to characterize CU patterns and interactions in BPD and to guide informed conversations about cannabis between mental healthcare providers and patients.

Characterizing OCD subtypes using geometric and topological representations of fMRI data

Obsessive-Compulsive Disorder (OCD) is broadly characterized by uncontrollable thoughts and/or repetitive behaviors. While scientists recognize the heterogeneous nature of OCD and the wide range of ways it can manifest in patients, there are not yet clinically defined subtypes of OCD. Here, we apply geometric and topological machine learning to OCD data to explore the potential to characterize subtypes. We model the neural dynamics of 41 participants (27 OCD, 14 Healthy Control) using resting-state fMRI data. Using Neurospectrum, a framework for encoding spatial and temporal neural data, we model each patient’s brain activity as a dynamic graph signal. We then use an autoencoder to extract latent trajectories, which are used to create various geometric and topological representations of the neural activity. We find that Neurospectrum can extract latent trajectories from resting fMRI data and meaningful geometric/topological features. We show that these features can be clustered using unsupervised learning methods (such as K-Means or Louvain) to identify natural clusters in the representations.

Evaluating the interaction between neuroinflammation and phosphorylated tau in rhesus macaque dlPFC across age span and human Alzheimer’s disease across disease continuum

Sporadic Alzheimer’s Disease (AD) is the most common neurodegenerative disease that triggers and furthers the onset of neuronal cell death, causing perpetual cognitive function and development decline in patients. However, the etiology is still largely unknown. Past research demonstrates that tau pathology within human brains appears a decade before the formation of Aβ plaques, and is a reliable discriminator of cognitive decline. Neuroinflammation is increasingly recognized as a central contributor to AD progression. Activated microglia and reactive astrocytes are closely associated with tau pathology in the human brain and are thought to influence tau phosphorylation, propagation, and neurotoxicity. However, it remains unclear whether inflammatory responses precede the accumulation of early soluble tau epitopes, arise as a reaction to fibrillated aggregates, or co-evolve with distinct tau species in a stage-dependent manner. Resolving these relationships is critical for identifying mechanistic links between tau dysregulation and immune activation, and for defining therapeutic windows targeting upstream pathogenic processes. In this study, we systematically examine the emergence of fibrillated (AT8) tau epitopes and their relationships to markers of neuroinflammation within vulnerable cortical circuits. I hypothesize that aging rhesus macaques and humans will demonstrate a concomitant emergence of reactive states of both microglia and astrocytes alongside the appearance of tau pathology. By integrating epitope-specific immunolabeling with cell-type–resolved analyses, this work seeks to define the early molecular sequence linking tau phosphorylation, aggregation, and inflammatory activation. Clarifying these interactions will inform biomarker interpretation and guide strategies aimed at intercepting AD at its earliest pathogenic stages.

Neural correlates and computational basis of corrective movements

Mammalian motor control exhibits a distinctive “submovement” structure, with discrete moments of replanning or correction. This submovement structure, often hypothesized to be an adaptation to motor noise (Harris & Wolpert, 1998), is impaired in stroke (Rohrer et al., 2004), motivating improved understanding of its neural basis. Although motor cortex projection populations are implicated in reach control, the specific computations they implement to facilitate submovements and corrective behavior remain unclear.

We extract rodent neural activity during a forelimb reaching task, recording from cortico-striatal (IT, n=7 mice) and cortico-spinal (PT, n=6 mice) motor cortex projection populations. We demonstrate ramping activity prior to corrections in both IT and PT, as well as dissociation in their rates of return to baseline post-correction. We further demonstrate that behavioral signatures of submovement structure persist even at high performance.

To analyze the computational basis of submovement structure, we implement an actor-critic model trained to perform simulated reaches. We demonstrate that signal dependent motor noise produces realistic submovement structure above even corresponding fixed-noise controls. As such, architectures such as (Takagi et al., 2025), in which submovement structure is hard coded, are unnecessary. Increases in signal-dependent noise, while associated with submovement structure, also lead to monotonic decreases in performance. We asked whether simulated rollouts with a learned reward model could recover some of these losses. Simulated rollouts accelerated learning at every noise level, and increased robustness to start location perturbation.

A visual circuit from the medial entorhinal cortex to the hippocampus

Sensory cues, such as visual information, are essential for navigation. This process relies on the hippocampal formation, which includes two interconnected regions: the medial entorhinal cortex (MEC) and the hippocampus. During navigation, grid cells in the MEC fire at multiple locations arranged in a regular triangular pattern, whereas hippocampal place cells fire only when the animal occupies in a specific location. Together, these cells form the brain’s internal spatial map. However, the cellular and circuit mechanisms by which visual information influences these internal cognitive maps remain largely unknown.

Previous work identified a population of visually responsive neurons in layer 3 of the dorsal MEC (dMEC), revealing a potential pathway through which sensory signals encountered during navigation may shape spatial representations and guide behavior. Because MEC layer 3 projects directly to the hippocampal area CA1, these neurons may provide visual information about the environment to hippocampal circuit as the animal navigates. However, the hippocampal targets of visually responsive MEC neurons remain unclear. To address this question, we used fosTRAP transgenic mice to permanently label layer 3 MEC visually responsive neurons and trace their axonal projections.

We found that MEC visually responsive layer 3 neurons project predominantly to stratum lacunosum-moleculare (s.l.m.) of hippocampal area CA1 where we also identified visually responsive neurons. Together, these findings identify the MEC-to-CA1 projection as an important pathway for conveying visual information to the hippocampus, where it may contribute to spatial representations, memory, and navigation.

“Nobody else like me”: Consequences of under-representation on competence beliefs

Imagine a 10-year-old girl successfully tries out for a competitive chess club only to find out that she is the only girl in the club. On the one hand, she might disregard the gender imbalance and focus on the validation of her abilities. At the same time, this gender disparity may cause her to conclude that girls must not be as good at chess. Here, we explore whether statistical data about the composition of groups causes 8- to 10-year-old children to internalize gender stereotypes. These issues of group composition are critical to understand given that they plausibly relate to stereotype acquisition as well as gender and racial gaps in the workforce (Arnold et al., 2023; Cimpian et al., 2020; Galitis, 2002; Garibaldi, 2014; Martinot et al., 2025). In this pre-registered study, a total of 200 8-10-year-old children participated in an asynchronous online experiment on ChildrenHelpingScience.com (Scott & Schulz, 2017). Children reported their age and gender, and then learned that they were going to try out for an elite club (the “Daxing Club”). To try out for the Daxing Club, children had to complete four rounds of a novel visual search task (i.e., find a green alien named Dax in a busy picture). To control for performance and elicit uncertainty about ability, the task was rigged such that Dax was only present in three out of four trials. After trying out, all children learned that they were accepted into the club and were shown a picture of existing club members. In the under-representation condition, children saw that they were the only ones of their gender. In the equal-representation condition, children saw that the club had five boys and five girls. We then assessed children’s views of their own and other children’s competence. Children learned group stereotypes from statistical information, inferring that the opposite gender was more competent in the under-representation condition but not in the equal-representation condition (B = 0.52, p < .001). Moving beyond prior work on third-party inferences about group membership (Kumar et al., 2023; Vélez & Gweon, 2020), we also found that children themselves felt more competent than same-gender peers when they got into a group where they were a minority (B = 0.53, p = .005). Children felt equally good about their competence compared to other-gender children across conditions. Finally, we found effects of participant gender: Across conditions and measures, boys rated their own competence higher than girls did. These findings suggest that under-representation carries immediate consequences for children’s reasoning about their own ability as well as group-level stereotypes. Combining first- and third-person inference revealed novel insights: When children were under-represented in a group, they thought the other gender was better at the task, but did not apply this stereotype to themselves. Taken together, these results show that children learn from under-representation in ways that shape their beliefs about themselves as well as others.

Identifying the neural signatures of conscious tactile perception

Processing sensory inputs is a fundamental aspect of consciousness, but differentiating between behavior-driven activity and perception poses a significant challenge in consciousness research. Previous research has demonstrated activation of cortical and subcortical networks during visual perception in the absence of overt responses. However, neuroimaging of tactile perception has not been as extensive as that on other sensory modalities. To address this gap, we developed an MRI-compatible tactile report/no-report paradigm to investigate the neural signatures of tactile perception. Across two visits, each with six runs, participants (N = 14) received bilateral pneumatic tactile stimulation at an individually calibrated perceptual threshold. Participants reported whether they perceived stimuli delivered to one hand, while stimuli delivered to the opposite hand were presented under no-report conditions. Pupil diameter, microsaccades, and blink rate were recorded using an EyeLink® 1000 Plus eye tracker at a 1000 Hz sampling rate. Function MRI data were acquired on a Siemens Prisma 3T scanner with a repetition time of 1s, and 2-mm isotropic resolution. Preprocessing included gray-matter masking, high-pass filtering, motion regression, and DVARS/framewise displacement censoring. Stimulus-locked BOLD responses from perceived and non-perceived threshold trials were analyzed in MNI space across S1, S2, the task-positive network, and the default-mode network. We found significant eye metrics to effectively predict perceived and non-perceived tactile stimuli, supporting the use of eye metrics in a tactile no-report paradigm. We found BOLD signal differences between perceived and non-perceived stimuli, suggesting that distinct neural networks contribute to conscious tactile perception. Differences between report and no-report conditions revealed a shift in activity from subcortical and arousal-related regions to frontal regions in the report condition, suggesting the report-related processing that no-report paradigms aim to eliminate. Overall, we developed a functional tactile no-report paradigm that has the potential to study conscious tactile perception with preliminary results showing promise.

Logical quantification in preverbal infants

Logical quantifiers (e.g., some, all) are found in all known human languages. Strikingly, quantifiers appear early in newly emergent languages such as Nicaraguan Sign Language (NSL), with quantifiers produced even by the first generation of NSL signers despite the absence of systematic language input. This raises the possibility that the concepts underlying quantifiers are available before the corresponding words are acquired. ​

Here, we test whether 10-month-old infants represent quantifiers. Across three looking-time experiments, infants viewed animations of causal launching events, in which colorful shapes bumped into and launched varying proportions of gray squares. In Experiment 1, infants were habituated to events in which not all (e.g., 2 out of 3) of the squares were launched, and looked longer at all events (e.g., 3 out of 3) than at other not all events (e.g., 1 out of 3). This suggests that infants can discriminate events that are universally quantified (i.e. all) from those that are not (i.e., not all). Experiment 2 controlled for the number of launching and non-launching events at test and replicated the finding that infants distinguished all from not all. Experiment 3 (ongoing) habituates infants to all events to test whether they distinguish not all events at test. ​

These findings suggest that pre-linguistic infants may represent the concepts of all and not all before acquiring the corresponding words, raising the possibility that the concepts underlying quantifiers are not only linguistic universals but also cognitive universals.