In the developing brain of a fruit fly (Drosophila melanogaster), neural stem cells, called neuroblasts, divide to produce new cells that will become neurons. These divisions follow strict biological rules, but because many factors influence how and when neuroblasts divide, predicting their behavior is challenging. While lab experiments provide crucial insights, they are often limited in how many conditions can be tested at once (genetic, physical, or otherwise). To address these limitations, we developed an agent-based computer model that simulates neuroblast divisions and their interactions with neighboring cells. Our model allows exploration of different conditions to predict how neuroblasts behave in complex environments. This work focuses on three key hypotheses about neuroblast behavior: (1) post stem cell division, the larger cells are more likely to remain as stem cells, (2) the cell positioned on top during division will keep its stem cell identity, and (3) clustering of differentiated neural cells on the membrane of a neuroblast suppresses their division. To investigate these hypotheses, we examine emergent behaviors in our model through size-based, location-based, and clustering-based differentiation rules. By adjusting parameters such as cell placement, division timing, and proximity to other neuroblasts, we analyze how these factors influence neuroblast fate. We validate model predictions against experimental data by comparing division patterns observed in simulations to those seen in Drosophila brains through live imaging. By combining computational modeling with experimental data, this work provides a framework for understanding the factors responsible for neural development. Our findings will refine existing models of neural stem cell behavior and help guide future experiments, making it easier to uncover the fundamental rules of brain development.

The retina is a layer of neurons on the back of the eye that sense light and relay visual information to the brain. Our goal is to understand the role of epigenetic repression in retinal cell development by focusing on the polycomb complex, a complex of many proteins that repress gene expression through deposition of the H3K27me3 mark on histones. The goal of my project is to learn how the polycomb complex influences retinal development by altering specific aspects of the complex’s activity and observing how these alterations influence cell fate, using two complementary model systems: fetal-derived retinospheres and stem cell-derived retinal organoids. To perturb different aspects of the polycomb complex, I have treated retinospheres with Gskj4, a UTX inhibitor, and BRM014, a BAF inhibitor. During development, UTX is responsible for removing H3K27me3 so genes that are silenced can be expressed. When I added Gsjk4 to 135-day old retinospheres, I observed that cell proliferation decreased, and more cells expressed the marker OTX2, indicating an upregulation of either bipolar or photoreceptor cell differentiation. These data indicate that H3K27me3 removal is critical for proper specification of retinal cell types. BRM014 inhibits BAF, an ATP-dependent chromatin remodeler that has been shown to be recruited by UTX to remove nucleosomes and initiate transcription. When I added BRM014 to day 135 retinospheres, I also observed an increase in the expression of OTX2, similarly indicating an upregulation of either bipolar or photoreceptor cell differentiation. From these experiments, we conclude that removal of H3K27me3 is necessary for proper retinal cell specification and development. A better understanding of epigenetic regulation during retinal development will allow us to develop therapies to regenerate damaged retina lost in blinding diseases and restore sight to patients.

Congenital hydrocephalus is a condition that is characterized by an accumulation of cerebrospinal fluid in the brain. This increases pressure in the brain, leading to neurological deficits. Surgery is the only current intervention however, even after surgery, patients experience lifelong complications. The underlying genetic mechanisms that cause human hydrocephalus are still unknown. To investigate the mechanisms behind hydrocephalus, the lab developed a mouse model by selectively ablating Notch signaling in specific brain regions. These mice developed obstructive hydrocephalus due to a loss of cell adhesion in the Sylvian aqueduct, a thin channel connecting the 3rd and 4th ventricles of the brain. Interestingly, brain regions exposed to high levels of Hedgehog signaling retained cell adhesion, indicating a possible protective role. Based on this, I hypothesized that Hedgehog signaling plays an unexpected role in supporting Notch-mediated cell adhesion. To test this hypothesis, I utilized small molecule Notch inhibitors to suppress Notch signaling activity in cortical spheroids. Cortical spheroids are precursors to cortical organoids derived from mouse embryonic stem cells. Preliminary data showed that various concentrations of different inhibitors were able to reduce cell adhesion and disrupt neural rosette formation in the spheroids. Neural rosettes are structures that recapitulate early cortical formation. After treatment with Hedgehog agonists, there was an increase in the number of neural rosettes in both untreated and inhibitor-treated spheroids, indicating that Hedgehog signaling can compensate for a loss of Notch signaling and preserve cell adhesion in the developing brain. These results seem to suggest that Hedgehog signaling can compensate for a loss of Notch through cell adhesion maintenance and prevent premature neural progenitor cell differentiation. The goal of the project is to establish cortical spheroids as a model system to screen potential genes associated with human hydrocephalus along with future drug therapies.
 

The orofacial sensorimotor cortex plays an important role in controlling tongue and jaw movements, such as speaking and eating. Being able to reliably perform these movements has critical implications for people suffering from neurological diseases such as stroke and Alzheimer’s disease, which are known to affect orofacial functions. However, the features of the complex lingual function that drive motor cortical activity are still poorly understood. Here, we investigate how information in the orofacial primary motor cortex (MIo) varies based on factors such as availability of tactile sensation, axis of motion, and specific regions of the tongue. To answer this question, we tracked marker-based movements of the tongue and jaw while recording neural activity from implanted microelectrode arrays in the MIo of two rhesus macaques (Macaca mulatta) engaged in feeding. Decoding accuracies of models based on (i) axis of motion, i.e., antero-posterior (x-axis), supero-inferior (y-axis), medio-lateral (z-axis), (ii) tongue marker region (superficial vs. deep, anterior vs. intermediate vs. posterior), and (iii) local anesthesia applied to sensory branches of the trigeminal nerve, were then compared to evaluate the ability to predict marker position. Generally, decoding performance was best using the y-axis and worst with the z-axis. Additionally, model performance was best in the x-axis of posterior tongue markers. Lastly, we found significant differences in model performance between control and nerve block conditions across all motion axes, with the x-axis showing the largest decrease in performance post-nerve block. These findings indicate that information carried by MIo neurons differ as a function of the tongue's motion axis, region, available tactile information, and varying combinations of these factors. These have important implications for the development of evaluation tools, rehabilitation strategies, and neural prostheses to restore orolingual function in particular and limb sensorimotor function in general.

Multitasking, such as walking and talking, is common for humans and other animals, yet we are limited in how many behaviors we can perform simultaneously. The neural circuit mechanisms that limit multitasking are not well understood. Uncovering these mechanisms will help us understand how brains combine some, but not all, behaviors during normal function, but also in the context of aging and neurological disorders such as Parkinson’s disease, where multitasking gets compromised. The fruit fly Drosophila melanogaster walks and “sings” by vibrating a wing during courtship, in a natural example of multitasking. These stereotyped behaviors are controlled by a relatively simple brain, which can be experimentally driven via artificial stimulation of key neurons, making the fly an amenable model to study multitasking. I therefore developed a platform to record and manipulate the interaction between locomotion and “singing”. I will activate sing-inducing neurons during two contexts, when flies are stationary (single-tasking) vs. moving (multitasking). I hypothesize that singing characteristics will change depending on context. For example, multitasking may decrease the likelihood of singing because the fly’s nervous system is “busy” controlling locomotion. Alternatively, locomotor context may make it easier to drive wing vibrations because of the higher activity levels in the circuits involved. My results will therefore help uncover how neural circuit interactions shape an animal’s ability to multitask.

Artificial neural networks (ANNs) are now able to learn from data to recognize patterns, often equaling or exceeding human performance. If we can understand the learned internal representations of such networks, we stand to gain insights into the brain. By taking a visual neurophysiologist’s approach to studying internal representations in deep convolutional neural networks (CNNs) trained to solve challenging visual recognition tasks, I aim to further our understanding of the primate visual system. To do this, I have taken sets of visual stimuli used by neurophysiologists to study the encoding of shape, texture and color in the mid-level visual cortex of macaque monkeys and presented them to CNNs (e.g., ResNet and AlexNet).  I found that activations of units within CNNs show a higher level of invariance to changes in surface properties of simple shapes (e.g., shape selectivity remains consistent when filled shapes are replaced by their outlines) than do cortical neurons. I also found a correlation between this invariance and the sensitivity of units to shape vs. texture stimuli that holds up in several CNNs.  Specifically, units with lower invariance to surface properties tend to respond with a wider dynamic range to textures than to shapes.  If this holds in other classes of visual ANNs, it could establish a general principle for mid-level visual encoding in which the surface properties (texture and color) of an object are represented somewhat distinctly from the position and shape of the boundary of the object.  This is consistent with the observation in the primate visual cortex that some neurons specialize in texture encoding, while others specialize in shape encoding. Ultimately, a better understanding of how information is encoded and processed in the cerebral cortex can allow us to build devices that interface better with the brain and to someday address brain disorders.