One of the great outstanding challenges in neuroscience is to understand how the human brain represents distinct objects. Studies have shown that a significant number of neurons in the visual cortex of non-human primates responded differentially to the sides of figures in 2D images even when the figures were outside of their classical receptive fields. This selectivity, termed border ownership, is believed to be responsible for the Gestalt idea of figure-ground organization, a process that allows specific regions of an image to be grouped together and given "thing-like" qualities. Many computational models have been developed to reproduce the experimental results of border ownership studies. At the same time, convolutional neural networks (CNNs), especially those specialized in image segmentation, are able to learn to solve the problem of figure-ground organization through supervised learning, all without the need for explicitly defined computational rules. We hypothesize that there is knowledge to be gained from CNNs, for they are excellent computational models for visual selectivities. Our novel 'artiphysiology' technique allows us to study the border-ownership phenomena in CNNs at a single-unit level in the same way an electrophysiologist studies the brain. The technique takes advantage of the accessibility and lack of noise of CNNs to enable high-throughput identification and analysis of circuit mechanisms. Starting from border ownership, the research aims to elucidate the mechanism by which figure-ground organization occurs at different layer depths and in different CNNs, using natural and artificial visual stimuli. The research has several applications, including improving CNN efficiency and interpretability as well as allowing for a better understanding of object recognition.

In the retina, loss of neurons results in blindness, because the mammalian retina cannot regenerate. Although mammals cannot regenerate neurons, species such as fish and amphibians can make fully functional neurons after injury and restore their vision. Muller Glia (MG) cells in fish and amphibians are the source of this regeneration. These cells respond to injury by dedifferentiating into progenitor cells that then become neurons, replacing the dead neurons and in turn restoring sight. At the Reh Lab, we have found a way to stimulate functional regeneration in mammals through these MG cells by expressing a gene called Ascl1. This technique has limitations however, as it only causes 25% of MG to undergo neurogenesis. One of the variables potentially limiting regeneration is inflammation, as inflammation has been known to be detrimental to neurogenesis. Monocytes invade the retina after injury and potentially cause inflammation that limits retinal regeneration. To determine monocyte impact on retinal regeneration we employed a transgenic technique to ablate monocytes. I then performed our retinal regeneration paradigm and determined whether regeneration is enhanced in the absence of monocyte invasion. Using immunohistochemistry and confocal microscopy I found more regenerated neurons in retinas that lacked monocytes. These data further confirm that inflammation limits the regeneration capacity of the retina, and provides future topics to improve neural repair through modifying the immune response.

Blindness caused by retinal degeneration is untreatable, and is a condition currently suffered by millions. The field of retinal reprogramming aims to establish a way to treat this type of vision loss, by finding a way for the eye to essentially produce new retinal neurons after they have been lost to degeneration. Often, this is done through directing cell fate by expressing or repressing various transcription factors such as Ascl1, a potent pro-neural transcription factor involved in retinal reprogramming. Ascl1 expression in mouse Muller glia, a support cell in the retina, can stimulate the regeneration of certain subtypes of retinal neurons, but the variety of retinal cells produced through this strategy is still limited. The class of transcription factors known as ‘Krüppel-like factors,’ or KLFs, regulates important cell processes, such as cell proliferation and development, with several KLFs functioning in the development of neurons. Due to the inhibitory function of KLFs during neuron development, we propose that the loss of inhibitory KLF genes in Muller glia may allow for activation of neurogenesis. Therefore, coupling the knockout of KLF genes with Ascl1 expression in Muller glia may be a key to enhancing reprogramming capabilities of Ascl1. For this project, I knocked out KLF genes in young mouse Muller glia in culture using the CRISPR/Cas9 system, and induced Ascl1 expression to stimulate reprogramming of the Muller glia. We used scRNA sequencing to determine if the knockout reprogrammed cells were molecularly similar to various retinal cell types that are typically lost with degenerative blindness. Preliminary sequencing results revealed that knockout of one candidate, KLF15, appears to promote neurogenesis. Findings from this experiment will allow us to increase our understanding of the role of KLF genes in retinal cell development as we work towards a future treatment for degenerative vision loss.

Retinal diseases such as macular degeneration and glaucoma lead to various forms of blindness as neurons in the retina die. Unlike amphibians or fish, the neurons of the mammalian retina cannot regenerate on their own and any damage is permanent. Previous research has shown that we can recover some of the lost regenerative capacity in mammalian retinas by mimicking the regeneration process in other species. This is possible by overexpressing transcription factor Ascl1 in Müller glia (MG), which are the main support cells in the retina. However, expressing Ascl1 alone can only lead to the neurogenesis of one type of neuron in the retina. This limits the therapeutic applicability of this approach because in diseases like macular degeneration or glaucoma, specific neuronal cell types are lost such as photoreceptors and ganglion cells respectively. Therefore, to regenerate these neurons we need to identify the right cocktail of transcription factors for MG reprogramming. Recent single-cell analysis from our lab has identified candidate developmental factors that could push reprogrammed cells to photoreceptors or retinal ganglion cells. This project aims to investigate the role of these factors in influencing cell fate after MG cells have been pushed to a progenitor state with Ascl1. Using lentiviral vectors, I will induce the overexpression of these candidate genes in primary cultures of young adult mouse MG that have been engineered to express Ascl1. In order to identify the nature of resulting neurons from these cultures I will perform immunocytochemistry paired with confocal microscopy, and to better understand changes in functionality of these cells I aim to perform calcium imaging. Since the electrophysiological responses of glia and neurons are distinct, cultures with reprogrammed neurons would record differently. Overall, this analysis evaluates the influence of new transcription factors on mammalian retinal regeneration.

In fish and amphibians there is a specialized zone of retinal stem-cells at the edge of the retina, called the ciliary margin zone (CMZ) which replenishes the retina with new cells if it is damaged in adult animals. However, the presence of these stem-cells has not been observed specifically in the CMZ of the developing human. Here, I investigate the developing human retina to understand if it contains stem-cells that could be harnessed for repair. I first utilized CMZ stem-cell markers found in fish and amphibians to assess in the developing human retina, including BLBP, C-myc, cyclin D3, Six3, SMAD1/5, and Zic1. Following the stainings, I observed expression of C-myc and BLBP. To maintain their long-term proliferation, stem-cells will replicate slowly. Therefore I analyzed cell cycle kinetics in the CMZ. Primary cultures of fetal human retina, called retinospheres, were made by dissecting the fetal retina into small pieces containing a portion of the CMZ and growing them in culture with retinal differentiation media. EdU, a dye that is integrated into DNA only in replicating cells, was then added to the media for different incubation intervals with EdU being 30min, 1hr, 2hrs, 4hrs, 6hrs, 8.5hrs and 25hrs, then retinospheres were fixed in PFA. Retinospheres were IHC stained with antibodies and dyes: Pax6 (a stem-cell marker), EdU (marker of cell division), Ki67 (marker of replicating cells) and DAPI. The total Ki67 positive cells in the CMZ and of EdU and Ki67 positive cells were counted so that the S-phase of mitosis could be measured to discover how fast the cells in the CMZ were dividing. I observed that cells in the CMZ were replicating slower than those further away from the CMZ, consistent with the possibility that there is a population of stem-cells in the CMZ of the developing human retina.

The pairing of high-resolution volumetric imaging methods with cellular markers of neural activity holds network-level explanatory power over behavior. However, common statistical analysis approaches fall short in capturing the functional relationships between brain regions across multiple spatial dimensions. Here, we propose combining unsupervised machine learning clustering methods with network graph theory visualization to reveal intricacies from these data beyond conventional standards. We demonstrate the feasibility of this approach on a recent single-cell dataset describing the longitudinal changes of brain-wide activation during relapse to palatable food in mice. We applied a new analytical framework combining the functionality of two open-source programs: (i) Histo-Cytometric Multidimensional Analysis Pipeline (CytoMAP), packaged with unsupervised k-means clustering and t-distributed stochastic neighbor embedding (t-SNE), and (ii) Cytoscape, a network analysis program. Hierarchical radial network diagrams were applied to the dataset in which we visualized the anatomical organization of regions that underwent statistically significant changes in activation. Across abstinence duration, we found an initial suppression in activation followed by widespread increases in activation. This increase correlated with observed behavioral changes and appeared to be triggered by activation hotspots. We next interrogated the co-activational relationships amongst the >900 brain regions by running unsupervised t-SNE dimensionality reductions on the data from each experimental group. These results were validated using k-means clustering and Davies-Bouldin indexing. We observed an intuitive segregation of regions dependent on activation status. Cluster number decreased in a time-dependent manner, suggesting increases in modular processing are associated with increased activation due to abstinence length. This trend indicated that cluster membership of regions likely also changed in a time-dependent fashion, indicating a dynamic recruitment effect at a regional level underlies abstinence-related relapse vulnerability. By analyzing cellular whole-brain data in this novel manner, we gained new insight into a previously unexplored dimension of brain activation dynamics underlying complex behavior.

Exploring the neural mechanisms modulating complex social behavior requires a holistic understanding of both central and peripheral body states. In freely interacting mice, social behaviors are often registered by changes in autonomic nervous function, including altered blood pressure, heart and breathing rates, and core body temperature. Unfortunately, these physiological metrics are difficult to obtain during complex social behavior due to substantial hardware requirements, like collars and tethers, restricting full movement and interaction. In collaboration with an industry partner, we are developing a fully implantable, battery-free device for wireless data acquisition of physiological data, including heart and respiratory rate, temperature, and other behavioral information such as locomotion and orientation of mice using biomechano-acoustic (MA) methods. Here, we validate the use of MA devices in both anesthetized and freely moving mice. First, we tested MA devices during emergence from anesthesia and compared anesthetized recordings using MA devices to a widely used and commercially available rodent pulse oximetry device. Second, we obtained MA recordings in freely interacting mice during complex social behaviors. This technology represents a crucial advanced tool for experimental behavioral research that enables non-invasive operations in cages with simple or complex environments in an individual or groups of animals.

Recent advances in neural recording technology are allowing scientists to record data at unprecedented scale. For example, the International Brain Lab (IBL) is a consortium of 22 labs working together to produce mouse brain recordings using high-density Neuropixels probes. In total, the IBL stores data from over 1000 sessions which provide a picture of neural activity across the entire mouse brain. At this scale, traditional static images and visualizations fail to communicate the results of neural analyses for brain wide activity and time-dependent behavior. As a tool for researchers in the IBL and undergraduate courses, we are developing 3D visualization software, the Virtual Brain Lab (VBL), capable of rendering a complete view of neural activity in the mouse brain in a simulated interactive laboratory. In my role as a developer, I am building tools to (1) rescale trial-averages around events and replay neural responses with variable event timings and (2) replay a single trial recorded in the IBL. Our simulation is built using the software package ‘Unity,’ allowing easy construction of custom 3D environments and publishing to diverse devices from laptop browsers to tablets, to virtual reality headsets. Displaying event-averaged neuron activity and single-trial replays will help IBL researchers spot anomalous data and holistically view their recorded sessions. Additionally, researchers can use our software to generate high-quality figures and videos for papers, outreach, and presentations. Finally, a host of key neuroscience concepts, such as sensory and motor coding and correlated variability are only communicated to students via dry lectures or textbook figures. Expensive lab classes in which students perform neuroscience experiments to rediscover these concepts are inaccessible to schools with fewer resources. Using our framework for visualization, we can build and distribute simulated tutorials and labs for students at little to no cost – reducing barriers for neuroscience education.

Pain is one of the key contributing factors on why people seek medical attention and healthcare globally due to its negative effects on both patients and their families. Therefore, understanding the circuits that control pain and how noxious stimuli contribute to the assignment of negative valence is critical for discovering novel analgesics for pain treatment. Analgesic Screen 1 (AS1), a small molecule screened in our lab, has been shown to reverse the valence of painful noxious stimuli and renders them attractive. Using zebrafish as a model for the study, we performed several different behavioral assays that give them a choice between a neutral and noxious stimulus for various sensory modalities. The stimuli that we tested were temperature, chemical and light/dark preference. Interestingly, we found out that in all these assays, AS1 managed to elicit attraction to the noxious stimuli. We tested different receptor agonists and antagonists to pinpoint the exact underlying mechanisms and found out that AS1 affected the dopaminergic circuitry, specifically the D1 receptors. Therefore, we believe that AS1 is valuable in understanding the neural mechanisms that allocate negative valence to nociceptive stimuli and help us find novel treatments for pain.

Rapid and smooth emergence from the anesthetized state to the awake state is important for patient safety and perioperative efficiency, yet is currently a passive process and the underlying mechanism is not well understood. In mice, emergence from anesthesia is modeled by the return of righting reflex (RORR) signaled by righting from the supine to prone position as the mouse emerges to an awake state. Using this model, it is possible to investigate the neuropharmacological mechanisms of emergence. While commonly studied in concert with neuronal recordings and optogenetic manipulation, these approaches can be combined with high-throughput automated behavior analysis using deep and machine learning approaches. Here, my goal is to create an automated behavioral classification pipeline for annotating the RORR in combination with experimental manipulations and recordings. I aim to characterize the transition between unconscious and awake states to define a binary output. This is accomplished by using DeepLabCut pose-estimation software to track subject mouse body parts, followed by the generation of supervised behavioral classifiers for RORR-related behaviors using the SimBA (Simple Behavioral Analysis) machine learning pipeline. My ongoing directions focus on performing unsupervised classification with this model to cluster additional behaviors. This use of advanced behavioral analysis will enable a better understanding of behaviorally-relevant neural activity in emergence and help bridge the gap between preclinical animal models and clinical intervention.