Abstracts

Recently, impressive advances in AI were achieved by training extremely large-scale models (GPT-3, CLIP, DALL-e, and many others), with hundreds of billions parameters on dozens of terabytes of data. Such models demonstrate remarkable improvements in few-shot generalization and transfer, both forward (few-shot generalization to novel tasks) and backward (reducing catastrophic forgetting in continual learning). It is remarkable that their test-loss performance typically follows simple power laws as the data, parameters and compute increase, although a more complex scaling, including emergent behaviors and sharp transitions, can be sometimes observed on certain “downstream” tasks. Overall, scaling AI models seems to bring us closer than ever to the “holy grail” of achieving general AI, but also opens a heated debate around the question: is scaling “all you need” to achieve AGI? We argue that, while scaling might be indeed a necessary condition for higher level of intelligence to emerge in complex networks, as it apparently did in nature, figuring out WHAT type of networks to scale, and HOW to scale them, remains crucial. Thus, inspirations from biology and neuroscience about the most effective scaling processes can be highly useful for achieving further breakthroughs in scaling artificial neural networks.

Tatyana Sharpee

Salk Institute

Reading out responses of large neural population with minimal information loss

Classic studies show that in many species – from leech and cricket to primate – responses of neural populations can be quite successfully read out using a measure neural population activity termed the population vector. However, despite its successes, detailed analyses have shown that the standard population vector discards substantial amounts of information contained in the responses of a neural population, and so is unlikely to accurately describe how signal communication between parts of the nervous system. I will describe recent theoretical results showing how to modify the population vector expression in order to read out neural responses without information loss. Compared to the standard population vector, the information-preserving read out includes just one additional weighting factor that describes the sharpness of neuronal nonlinearity and represents a measure of neuronal variability. In this way, more reliable neurons are weighted more than weakly tuned neurons. It is noteworthy that there is no simple expression for the information-preserving read-out when written in terms of parameters of neural tuning curves. Although noise correlation affect the amount of the information contained in the responses of the neural population, the same read-out expression continues to work when noise correlations increase or decrease in strength. These results demonstrate how to quantify information transmitted by neurons with irregular tuning curves. I will describe three approximations that make it possible to quantify information transmitted by large neural populations containing thousands of neurons.

UMass Faculty speakers

Gottfried Schlaug

Biomedical Engineering

Behavioral and Imaging Effects of Brain Stimulation

Non-invasive brain-stimulation (NIBS) techniques can modulate brain activity and change local cerebral excitatory/inhibitory relationships. In pre- versus post-stimulation comparisons one can see longer-lasting effects on behavior and cognitive operations allowing researchers to test causality of brain-behavior relationships. Concurrent NIBS and MRI allow us to examine changes in local activity, functional connectivity, and neurotransmitters of targeted brain regions and optimize the parameter space for therapeutic options. Current research on these issues and their applications in health and disease/disorder will be presented.

David Moorman

Psychological and Brain Sciences

Neural representations of value and action for learning and behavioral control

Learning and behavioral control, in both biological and artificial systems, is commonly driven by associations between either cues or actions, and the value of the associated or consequent outcomes. Despite years of research and many significant advances, we do not yet have a clear mechanistic description of the neural basis of each of these components, or how they integrate to regulate learning and behavior. In this talk I will present some of our work investigating how neural systems, particularly in the rodent frontal cortex, integrate information related to action and value in the service of learning and behavior. Our results demonstrate that outcome value representations in neural activity are dependent on the actions associated with the outcome. I will also describe some of our recent work in which we are further probing how outcome learning occurs under different action regimens, and what the neural basis of this learning might be. Finally, I will briefly discuss how, by probing the mechanisms of learning, action, and value, we believe our work has broad implications for both biological and computational frameworks.

Margaret Stratton

Biochemistry and Molecular Biology

The mechanism of CaMKII regulation: from fertilization to encoding long-term memory

Cell to cell communication is critical for function in all multicellular organisms. A key factor for intercellular communication is regulation by Ca2+ concentration. Ca2+/calmodulin dependent protein kinase II (CaMKII) is a Ca2+ sensitive enzyme that is encoded by four genes in mammals: alpha, beta, gamma, and delta. There is an incredible amount of diversity generated from the four vertebrate CaMKII genes. Alternative splicing produces up to 386 transcripts, which leads to the production of 386 proteins that are then differentially post-translationally modified, and mix to form hetero-oligomeric complexes, ultimately culminating in thousands of chemically distinct CaMKII proteoforms. We are specifically interested in the crucial roles CaMKII plays in long-term memory formation (neurons: alpha, beta), fertilization (oocytes: gamma), and cardiac physiology (cardiomyocytes: delta). Intriguingly, these cells all communicate using Ca2+ oscillations but on vastly different timescales (minutes to milliseconds). How does one enzyme accommodate this multifunctionality? We hypothesize that selective splicing and modification creates a unique set of CaMKII variants expressed in specific cell types, thereby leading to differential functional outputs. Fully elucidating these complex biological roles requires a deeper understanding of CaMKII variation at the sequence and protein level, structural and conformational ramifications of these variations, and how these variables affect CaMKII interactions within the cell. In this proposal, we seek to expand our understanding of CaMKII function inside cells using a combinatorial approach of sequencing, biochemistry, structural biology, and cellular assays. Completion of the proposed work will allow us to uncover the molecular basis for the many roles of CaMKII in neurons, cardiomyocytes, and oocytes – with far-reaching implications on therapeutic intervention for neurologic disease, cardiac dysfunction, and infertility.

Erik Learned-Miller

Computer Science

How can computer vision benefit from neuroscience and biology more generally?

There’s a longstanding tradition in computer vision of thinking we can find solutions that are

better than evolution has found, only to ultimately realize that the biological solution was superior.

In this talk, I’ll discuss a few recent trends in computer vision (and my own research) that

embody this idea of “returning to the biological solution”. In particular, I’ll discuss computer

vision with “event cameras” that work more like the retina and new techniques for “self-supervised learning” in which an agent (either an animal, or an artificial agent) can provide data

for its own training.

Hava Siegelmann

Computer Science

Intelligent Autonomy via Lifelong Learning and temporal aware AI

AI embedded in real systems, such as in satellites, robots and other autonomous devices, must make fast, safe decisions even when the environment changes, or under limitations on the available power; to do so, such systems must be adaptive in real time. To date, edge computing has no real adaptivity – rather the AI must be trained in advance, typically on a large dataset with much computational power needed; once fielded, the AI is frozen: It is unable to use its experience to operate if environment proves outside its training or to improve its expertise; and worse, since datasets cannot cover all possible real-world situations, systems with such frozen intelligent control are likely to fail.

Lifelong Learning is the cutting edge of artificial intelligence - encompassing computational methods that allow systems to learn in runtime and incorporate learning for application in new, unanticipated situations. Until recently, this sort of computation has been found exclusively in nature; thus, Lifelong Learning looks to nature for its underlying principles and mechanisms and then translates them to this new technology. Our presentation will introduce a number of state-of-the-art approaches to achieve AI adaptive learning from DARPA’s L2M program and subsequent developments.

All neural network AI systems are sensitive to changing environmental conditions as well as to attacks. A viable solution to alleviate these weaknesses may include joint human-AI collaboration; but to date, this approach has not been sufficiently well implemented - particularly in terms of avoiding work overload or boredom for the human partner. Another way to make adaptive systems robust is by making them aware of time and able to comprehend temporal patterns in the environment. We will describe our current research in temporal AI, while also considering power constraints.

Posters

1. UMass Amherst Genomics Core Facility: Capabilities and Resources

Ranjan, Ravi, Genomics Resource Laboratory, Institute for Applied Life Sciences (IALS)

The Genomics Facility provides a suite of services to address high-throughput next-generation sequencing (NGS), including solutions for sample processing such as nucleic-acid isolation, nucleic-acid quantitative and qualitative analysis, NGS library preparation, quantitative-PCR analysis, etc. The facility has recently acquired 10x Genomics Chromium Controller system, enabling single cell genomics research projects.

The Genomics Facility is a fee-for-service lab and provides a sample processing and library preparation such as whole genome sequencing, shotgun metagenomics, metatranscriptomics, targeted amplicon sequencing, RNA-Seq, ChIP-Seq, Exome Sequencing, Single Cell Genomics, etc., to address genomics project needs. The facility accepts samples and performs requested analysis. We offer training to users to conduct experimentation for use on a fee for service basis to both internal and external researchers, academic or industry.

The Genomics Facility is equipped with - Illumina NextSeq500 and MiSeq NGS systems, 10x Genomics Chromium Controller System, Fluidigm C1 Single-Cell Auto Prep System, Nexcelom Cellometer K2 Cell Counter, Diagenode Bioruptor Pico, Agilent 2100 Bioanalyzer, SageScience BluePippin, BioRad CFX96 Touch Real-Time PCR system, Qubit Fluorometer, Gel Electrophoresis, FastPrep 24-G Homogenizer, Savant Speed-Vac, and other ancillary lab equipment.

2. Loss of Neurodevelopmental Gene CASK Disrupts Neural Connectivity in Human Cortical Excitatory Neurons

McSweeney, D, Gabriel, R, Jin, K, Pang, ZP, Aronow, B, and Pak, C, Biochemistry and Molecular Biology

Loss-of-function (LOF) mutations in calcium/calmodulin-dependent serine protein kinase (CASK) cause severe developmental phenotypes, including microcephaly with pontine and cerebellar hypoplasia, X-linked intellectual disability, and autism. To unravel the pathogenesis of CASK-related disorders in human cells, we engineered CASK LOF knockout mutations in isogenic human embryonic stem cells using CRISPR/Cas9 and differentiated these cells into cortical excitatory neurons using Ngn2 overexpression. Using these cell types, we dissected CASK at two developmental time windows – at day 7 when active neurite outgrowth occurred and at day 28 during robust synaptic transmission. We performed bulk RNA sequencing at both time points to investigate changes in gene expression and characterized their neurite morphology. We also quantified synaptic morphology and assayed synaptic transmission using both multi-electrode arrays and whole-cell patch-clamp electrophysiology. While immature CASK KO neurons showed robust neuronal outgrowth, mature CASK KO neurons displayed severe defects in synaptic transmission and synchronized burst activity without compromising neuronal morphology and synapse numbers. In developing human cortical neurons, CASK functioned to promote both structural integrity and establishment of cortical excitatory neuronal networks. These results lay the foundation for future studies to identify suppressors of such phenotypes relevant to human patients.

3. No Evidence for a Visual Testing Effect for Novel, Unnameable Objects

McCarter, A., Huber, D., Cowell, R., Psychological and Brain Sciences

The well-documented verbal testing effect shows a memory benefit when verbal materials are recalled. The Representational-Hierarchical Account posits that similar cognitive processes operate on verbal and visual information, suggesting that visual stimuli should also benefit from recall practice. Here, we tested that claim. To minimize the impact of semantic information, we used novel, abstract images consisting of a shape and fill. Following initial study, half of the images were re-studied while the others were subject to recall practice. Practice involved presenting one feature (shape/fill) of a stimulus and asking the participant to recall and select the feature that goes with it. In a series of experiments, we repeatedly found no memory benefit for the items that were recall-practiced over the items that were re-studied. This suggests that the testing effect does not occur for purely visual material, and therefore may not be a universal process across content types.

4 The Contribution of 5-HT1A/2A Receptors and GABAergic Neurons of the Pedunculopontine Tegmental Area to Sensorimotor Gating

Correll E; Fenelon K, Neuroscience and Behavior / Biology

Sensorimotor gating is a fundamental process by which a sensory event can inhibit a motor output. When reduced, this pre-attentive process is associated with disturbances in cognition and attention. Sensorimotor gating is commonly measured using the translational prepulse inhibition (PPI) of the startle reflex behavioral assay, which is affected by many neuropsychiatric disorders including schizophrenia. While there are still gaps in our understanding of the underlying circuitry, PPI is known to be governed by startle-mediating neurons located in the caudal pontine reticular nucleus (PnC). These neurons send inputs to spinal motoneurons to initiate a motor response. The PnC receives inputs from many regions including the pedunculopontine tegmental region (PPTg) which contains three major cell types: cholinergic, glutamatergic, and GABAergic. Lesion studies indicate that the PPTg is critical for PPI, however, the contribution of PPTg cholinergic and glutamatergic cells have been ruled out. These findings suggest that GABAergic neurons, which constitute one third of PPTg neuronal population and express various serotonin (5-HT) receptors, play a crucial role in PPI. Notably, GABAergic and serotonergic neurotransmission are impaired in schizophrenia. Our study aims to determine how 5-HT1A/2A+ PPTg GABAergic neurons inhibit startle-mediating PnC neurons during PPI.

5. Endocannabinoid Receptor Knockout Protects Against Circadian Desynchronization-Induced Metabolic Disruption

Falcy, B; Pearson, G; Karatsoreos, I, Neuroscience and Behavior

Disruption of the circadian clock can lead to several changes in metabolic phenotypes, including weight gain, elevated plasma triglycerides, and increased hepatic lipid deposition. We have shown that environmental circadian desynchronization (ECD), by housing mice in 20h cycles (10h light – 10h dark), leads to weight gain, higher adiposity, and altered metabolic hormone levels in mice. Understanding the mechanisms underlying these effects is critical. While classical metabolic hormones certainly contribute to these phenotypes, it is important to expand our understanding of other potential systems regulating metabolism. Endocannabinoid (eCB) signaling can affect metabolism from the cellular to behavioral levels via effects both in the periphery (e.g., liver) as well as centrally in the brain. Interactions between eCBs and circadian rhythms have been documented in humans, where circadian misalignment can lead to increases in eCBs measured in the blood. To explore these phenomena on a more mechanistic level, we determined the effects of global cannabinoid type 1 receptor (CB1r) knockout on the metabolic and behavioral consequences of ECD. We undertook detailed behavioral and metabolic phenotyping in CB1r WT and KO littermates during baseline 24-hour and experimental 20-hour light-dark cycles. We found that in ECD, weight gain was significantly higher in WT male and female mice compared to KO littermates, demonstrating that CB1r KO mice are resistant to the metabolic effects of ECD. We also measured changes in metabolic hormones including insulin and leptin, further showing interactions between ECD and genotype. We also explored changes in several behaviors including locomotor activity, feeding, and drinking, as well as a detailed examination of changes in the respiratory exchange ratio. While there were genotype and sex differences in the organization of some of these behaviors, changes in overall activity levels do not explain the observed weight differences, as no effect of genotype was found on overall locomotor activity. Together, our data support a role for eCB signaling in ECD induced metabolic dysregulation.

6. Investigating the role of neurexin-1 alternative splicing in autism spectrum disorders

English J, Lim E, Pak C, MCB

Autism spectrum disorders (ASDs) are increasingly diagnosed in the United States, affecting 1 in 54 individuals. The costs associated with providing care for adult ASD patients is projected to increase from the current cost of approximately $185 billion to $1.36 trillion by the year 2040. Aside from the economic impact ASDs have, there are also emotional and financial impacts to families of ASD patients, such as higher education costs, medical bills, and reduced opportunities for employment for patient’s caretakers. Currently, there are no effective treatments for ASDs and the mechanism by which they develop are poorly understood. Studies indicate pathology stems largely from gene by environment interactions, with over 100 genes associated with higher risk for developing ASDs, most of which are involved in either gene regulation or neuronal communication. One such gene, NRXN1, encodes neurexin 1 (NRXN1). Heterozygous deletions of this gene have been found to occur at higher rates in ASD individuals versus the general population. NRXN1 is a pivotal synaptic organizing molecule which participates in synapse specification, maintenance, and function. The NRXN1 gene has six canonical splice sites and three promoters, which in combination with post-translational modifications can encode thousands of possible splice variants. The composition and structure of the resulting presynaptic NRXN1 isoforms dictate which postsynaptic ligands NRXN1 can interact with, conferring synaptic specificity. However, the splice variant repertoire and the resulting isoform representation of NRXN1 has not been thoroughly researched. I will use transcriptomic approaches utilizing long-read mRNA sequences to identify all splice variants in cultured ASD patient iPSC-derived induced neuronal cultures and identify candidate splice variants for further study. Quantitative proteomic approaches such as parallel reaction monitoring will be used to validate the transcriptomic findings as well as characterize isoform presence at the synapse in the same scheme. Finally, the electrophysiological properties, synaptic transmission, and neuronal network connectivity of these cultures will be characterized using patch-clamp and multielectrode array recording, respectively. Together, these studies will provide insight into the mechanism underlying synaptic dysfunction in ASDs.

7. Chemically-bonded hybrid liposomes for non-viral gene editing in mouse nervous system

Hong E, Biomedical Engineering

Recently, gene editing techniques have drawn much attention to cure neurodevelopmental disorders such as autism spectrum disorder (ASD), including deficient social behavior. Unfortunately, due to the structural complexity and the difficulty of crossing blood-brain barrier, gene editing in the nervous system relies on viral vectors, capable of high transduction efficiency despite insertional mutagenesis and oncogene activation. Here, to overcome these challenges, this study aimed to elucidate non-viral based gene editing to manipulate genes with hybrid liposomes and to investigate neural circuits for understanding neurobiological mechanisms for ASD in mouse brain. Hybrid liposomes have been reported as alternatives of viral vectors for gene delivery and pharmaceutical applications to enhance poor sensitivity and less endocytosis efficiency. These hybrid systems not only provide stable colloidal stability for longer blood circulation in vivo, but also have synergistic merits. In this work, chemically bonded hybrid liposomes will be developed, an approach that allows CRISPR/Cas9 systems delivery in mutant Shank3 mice as an autism-associated in vivo model. Therefore, this hybrid liposome system will enable to improve autism-like behavior, such as social skills deficits, and to support neurobiological mechanism evidence for ASD via neural circuit investigation.

8. Impact of traumatic brain injury on glial and neural function in the hippocampus Introduction:

Dougan CE, Roberts BL, Karatsoreos IN, Peyton SR, Psychological and Brain Sciences

Traumatic brain injury (TBI) is an established risk factor for developing neurodegenerative disease. However, there is a lack of TBI models that relate injury forces to both macroscale tissue damage and brain function at the cellular level. Needle-induced cavitation (NIC) is a technique that induces highly localized injury to ex vivo brain tissue by applying fluid pressure. We have previously observed that NIC causes tissue damage along the hippocampus, a brain region critical for learning and memory formation. Injury to this region causes cognitive pathologies in humans and rodent models. However, the impact of NIC at the cellular level is unknown. NIC related injury activates specialized glial cells called astrocytes. We hypothesize that NIC induced astrocyte activation will lead to secretion of the signaling and remodeling proteins: tenascin-c (TN-C), thrombospondin (TSP), and connective tissue growth factor (CTGF) into the extracellular matrix (ECM). Although astrocyte secreted ECM proteins are known contributors to synapse formation, how NIC impacts synaptic function is unknown. We hypothesize that NIC will disrupt synaptic function in the hippocampus in the weeks following injury. Here we propose to combine a range of engineering and neuroscience techniques to test these hypotheses. Understanding how TBI impacts short- and long-term astrocyte responses and synaptic function is essential in determining the underlying mechanisms that relate acute brain injury with neurodegenerative disease. This collaboration lays the groundwork for advanced approaches in understanding how TBI impacts neural function, and the development of treatments that promote TBI repair and prevent neurodegenerative disease.

9. High throughput localized passive diffusion device for patterning human forebrain organoids

Feiyu Yang; Narciso Pavon; ChangHui Pak；Yubing Sun., Mechanical Engineering

Human pluripotent stem cell (hPSC) -derived 3D brain organoids provide an unprecedented opportunity to recapitulate human brain development. Human brains develop into highly stereotypical subregions, which is partially determined by signaling molecule gradients. While protocols have been developed for deriving brain organoids with specific regional identities, it is still not possible to fully recapitulate the cytoarchitecture of developing brains with high reproducibility. Here we engineered a high-throughput system to mimic the Sonic Hedgehog (SHH) gradient presented in vivo for dorsal-ventral (D-V) patterning in human forebrain organoids. Our passive diffusion-based system is easy to operate and requires only routine media changes. We have successfully demonstrated that D-V patterning in forebrain organoids in the presence of a purmorphamine gradient in 20 days of culturing. Future steps include optimizing the control of diffusive-based gradient for potential commercialization of the device, monitoring neural activities in the brain organoids through Ca 2+ imaging, performing single-cell RNA sequencing to investigate the cellular composition in the patterned human brain organoid, and studying tangential interneuron migration from ventral to dorsal organoid regions. We anticipate improving the reliability of the brain organoid patterning through device optimization for applications such as drug screening and disease modeling.

10. Precise spatiotemporal identification of amygdala neurons active during sensorimotor gating

Wanyun Huang, Karine Fénelon, Biology

Prepulse inhibition (PPI) of the acoustic startle reflex refers to the inhibition of a startle response when a weak stimulus (“prepulse”) is presented prior to an alarming stimulus (“pulse”). PPI is a standard operational measure of sensorimotor gating. PPI is impaired in many neuropsychiatric disorders including schizophrenia, obsessive-compulsive disorder and Huntington’s disease. PPI impairments are associated with cognitive overload, attention deficits and amygdala dysfunction (Braff et al., 2001). Therapeutic advances are limited by the gap in our knowledge of the neuronal circuitry underlying PPI. In fact, the currently used dopaminergic-based antipsychotics show inconsistent effects on PPI in affected individuals (Frau et al., 2014).

Previous stimulation and electrophysiological studies showed that giant glutamatergic neurons located in the brainstem caudal pontine reticular nucleus (PnC) mediate startle response (Lengenhöhl and Friauf, 1992). The PnC neuronal population includes both giant glutamatergic neurons and glycinergic neurons (Koch and Friauf, 1995; Rampon et al., 1996; Zeilhofer et al., 2005) which receive various glutamatergic inputs. We recently showed that the central nucleus of the amygdala (CeA) contributes to PPI by sending glutamatergic inputs to PnC glycinergic neurons (Cano et al., 2021). But which CeA neurons are active during PPI remains unknown.

To answer this question, we used Cal-light, a calcium-dependent and light-sensitive method that enables the identification and manipulation of active neurons during a given behavior. By delivering Cal-light viral components to the CeA of wildtype mice, we were able to identify CeA neurons active during PPI, because upon calcium entry and in the presence of blue light, these neurons became green fluorescent. Since Cal-light targeted CeA neurons also expressed the inhibitory optogenetic tool Halorhodopsin, photo inhibiting these neurons with yellow light led to reduced PPI. Overall, our results confirm that CeA-PnC glutamatergic synapses contribute to PPI and Cal-light allowed us to identify the CeA neurons involved in PPI with high spatiotemporal resolution. Our findings provide critical insights towards identifying potential therapeutic targets for diseases associated with PPI deficits.

11. Characterizing a Circuit Linking Auditory Pallium and the Social Behavior Network

Spool J; Lally A; Chen P; Remage-Healey L, Psychological and Brain Sciences

To engage in healthy social interactions, the brain must coordinate processing of complex social sensory cues with appropriate social responses. While complex social signals (i.e., visual, auditory) are processed in the telencephalic pallium, nuclei controlling social behaviors, called the social behavior network (SBN; conserved across vertebrates), reside mainly in the diencephalon. In songbirds, for example, auditory pallium allows birds to learn dozens of individuals by their songs and calls, while the SBN are necessary for appropriate social responses to songs and calls. Tremendous progress has been made in studying pallial sensory circuits and the SBN largely in parallel, but apart from mammalian olfactory systems we are lacking basic knowledge of how SBN nuclei integrate sensory information from pallium. We asked whether pallial sensory processing is necessary for the SBN to respond to complex social sensory cues. We transiently inactivated auditory pallium of female Zebra finches with inhibitory neurotransmitter receptor agonists during song playback, and examined song-induced immediate early gene (egr-1) activation in SBN nuclei. Auditory pallial inactivation specifically impaired egr-1 responses to song in the lateral ventromedial nucleus of the hypothalamus (VMHl), providing the first evidence in vertebrates of a connection between auditory pallium and the SBN. Egr-1 expression in VMHl also correlated with feeding behavior, consistent with its dual roles in homeostatic regulation and social behavior. However, egr-1 expression did not correlate with feeding in control animals exposed to auditory playback, indicating that auditory input from pallium to VMHl may mediate a trade-off between social attention and feeding. Electrophysiological recordings from VMHl reveal single units with complex auditory response profiles to ecologically-relevant bird songs and calls. These data highlight a role for VMHl in the integration of complex social auditory stimuli with internal state to influence social decision-making. Support from NIH F32DC018508 and NIH R01NS082179.

12. Sex differences in adrenergic α1 regulation of reinforcement behavior

Rodberg E; Yu S; Vazey E, Biology - Neuroscience and Behavior

Humans and animals need to exhibit appropriate behavioral responses to environmental cues, both to obtain resources and avoid threats. Appropriate behavioral responses to stimuli are mediated in part by brainwide norepinephrine (NE) levels and locus coeruleus (LC) neural activity. LC-NE has been shown to have an inverted-U shaped effect on behavioral responses where task engagement decreases when NE levels are outside an optimal range, either too high or too low. In addition to the relationship between NE and behavior, previous research has identified sex differences in the size, structure, and stress responsivity of LC. Independently, sex-differences have been found in positive and negative reinforcement learning.

To probe the role of α1 adrenergic signaling on reinforcement behavior, we used an α1 agonist (cirazoline; low dose 0.1mg/kg, high dose 0.3mg/kg) and antagonist (prazosin; low dose 0.5mg/kg, high dose 1mg/kg) in rats (male and female) during an active avoidance and reward seeking task. During this task, rats learned to press a lever in response to a cue for a reward (reward seeking trials) and press an opposing lever after a different cue to avoid a foot shock (active avoidance trials).

We found that the ability to appropriately respond to cues, to either obtain a reward or avoid a negative outcome, is sensitive to α1 adrenergic signaling in a sex-specific manner. On active avoidance trials, high doses of both α1 agonists and antagonists decreased accuracy in males whereas females were only affected by α1 agonists. This decreased accuracy was driven by increased omitted responses in both males and females. Reward seeking trials were less sensitive to α1 manipulation and accuracy only decreased after high doses of cirazoline in males. Overall, the data shows that reinforcement behavior in males, particularly to cues predicting a potential threat, was sensitive to α1 manipulation and followed a Yerkes-Dodson relationship as predicted for LC-NE. Females were more resilient to α1 manipulations during this task which may be due to different behavioral strategies or NE receptor distribution. Results from this study indicate that cirazoline and prazosin, by increasing and decreasing α1 noradrenergic signaling respectively, can impair reinforcement behavior and the degree to which α1 modulation impacts behavior differs based on sex.

13. Robust Multimodal Deep Learning for the Early Forecasting of Alzhiemer’s Disease

Zhang, Sidong; Fellman, Evan; Liu, Genglin; Ko, James; Fiterau, Madalina, Manning College of Information & Computer Sciences

Alzheimer’s disease (AD) is the most common form of dementia, causing progressive cognitive impairment, disorientation, and memory loss. Since developed Azheimer’s symptoms often indicate untreatable damage to the brain, forecasting AD years ahead of onset is critical for attempts at early treatment, the selection of subjects for clinical trials, and to facilitate neurologists’ study of the disease. As deep learning models are capable of leveraging potential information from complex data, we introduce a sequential deep learning approach to the forecasting task on multimodal patient records including cognitive test scores, highly engineered volumetric features, and brain MRI scans, while still maintaining stable forecasting performance over forecasting windows of two years or more. The forecasting task on the patient records is based on selective cognitive test scores supported by latent features extracted from brain MRI scans where two novel methods are implemented. The first method is an algorithm to estimate the brain surface roughness of the MRI scan as an indicator of the cortical shrinkage. The second method is to encode MRI scans to lower dimension spaces via a CNN encoder trained by minimizing the mutual information between the encoded feature and the corresponding cognitive scores. Both methods leverage information from the MRI scan to enhance the forecasting task and achieve an F1 score of 0.8199 and 0.8202 respectively on Alzheimer’s forecast 24 months ahead of time.

14. Age-related Impairments in Memory Recall Depend on What You are Recalling

Gove J; Sanders M; Jiang A; Cowell, R, Psychological and Brain Sciences

Dominant theories of memory within cognitive neuroscience have held that medial temporal lobe (MTL) structures, such as hippocampus, are specialized for memory. Memory processes such as “recall” have been ascribed to the hippocampus. The ventral visual stream (VVS) is a brain pathway that feeds into the MTL. Importantly, there has been thought to exist a stark, anatomical division between a “perceptual system” that mediates vision in the VVS, and a “memory system” in the MTL. Contradicting this traditional account, recent research has demonstrated both a role for the hippocampus in visual perception and a role for brain structures outside of MTL (in the VVS) in memory processes such as recall.

One newer theory suggests that regions in the brain pathway from early visual cortex to hippocampus are not well explained using labels based on cognitive processes, such as “visual perception” or “recall”. Instead, brain regions are best understood in terms of the neural representations that they are equipped to process – a brain region will be engaged in a task if the task uses the “kinds of stimuli” that the region represents with its patterns of neural activation. Damage to the hippocampus does not result in a loss of “memory,” but in a loss of the ability to process complex associative representations, such as high-dimensional spatial scenes or episodic memories. This is because these are the kinds of things that the hippocampus represents.

My experiment is a memory study that uses images of complex scenes, which should engage the hippocampus, and individual objects, which should not (instead, these should engage brain regions earlier in the “VVS” pathway). Given that the hippocampus deteriorates with age, I tested a control group of younger adults, and an experimental group of older adults. The task for both scenes and objects was a recall task. If the key "process" that engages the hippocampus (and thus is impaired in old age) is "recall", then older adults should be equally impaired at recall in both conditions (objects, scenes). But if the key function of the hippocampus is to represent high-dimensional stimuli (such as scenes), then I predict an interaction between stimulus type and age group, such that older adults are more impaired at scene recall, relative to younger adults, than at object recall. I will present preliminary results.

15. Electrical Modulation on Genetically Identifiable Neurons

Chen W; Naroian J & Tajchman J; Siyuan Rao*, Biomedical Engineering

Abnormal neural electrical signaling can result in various neurological diseases. Existing techniques allow for modulating neural electrical signals via electrical or magnetic fields; however, due to nature of electricity and negligible magnetoreception of biological tissues, these techniques lack the control of genetically identifiable neural populations and permit limited spatial precision. Here, we propose a transcranial magnetic stimulation (TMS)-assisted neural modulation technique that utilizes a genetic toolkit to provide tunable thresholds for neural depolarization and grant the selectivity of TMS on specific neural populations. We plan to on-demand express a series of protein with different conductivity on neural cell membranes. After validating the protein expression by immunostaining, we confirmed effect on membrane property regulation with one of our candidate proteins, pilA, in HEK 293 cells with patch clamp recording. We plan to further validate our hypothesis of modulating neural excitation/inhibition on mouse hippocampal and dorsal root ganglion neurons via protein expression on neural cell membranes. To prepare in vivo study, we plan to co-microinject AAV9 viral vector carrying candidate proteins and genetically encoded calcium indicator, GCaMP6s in mice brain. We will use patch clamp recording and calcium imaging of infected neurons in acute brain slices with or without magnetic field to validate the specific activation/inhibition of neurons via magnetic field ex vivo. The proposed technique, as the first phase of my PhD project, will provide new insight on cellular mechanism of neurological diseases related to abnormal neural electrical signaling and will be further developed into a useful tool for neuroscience community.

16. In vivo characterization of social representations in medial amygdala aromatase neurons

Dickinson, S; Bergan, J, PBS

The medial amygdala (MeA) is a central node in a network of brain regions that process pheromones and produce social behaviors. A subset of aromatase expressing (arom+) neurons within the MeA facilitate sex specific social behaviors including sex recognition, mating behavior and aggression. Arom+ neurons integrate information from a wide array of subcortical neural circuits. The functional, temporal and behavioral dynamics of arom+ neurons are unknown. The following project fills this gap using fiber photometry to study arom+ neuronal activity while animals are awake and socially behaving. Preliminary data examines the effects of social partners and sex differences on arom+ neuron activity and social behavior. Method development and optimization is also detailed.

17. Time of Day Modulates Neuroimmune Activation of the Olfactory Bulb

Pearson G; Falcy B; Wang J; Akli S; Karatsoreos I, Neuroscience & Behavior

Background: The circadian clock regulates immune responses and primes cells and tissues to anticipate physiologically relevant environmental changes. Given its proximity to the nasal cavity, the olfactory bulb (OB) needs to generate robust neuroimmune responses to defend against neurotropic pathogens. Since the OB also contains a circadian clock that oscillates independently of the suprachiasmatic nucleus, we hypothesized that daily changes in OB neuroinflammatory state would differentially "prime” responses to an intranasal inflammatory challenge. Aim 1 probed the OB’s neuroinflammatory transcriptional profile throughout the day. Aim 2 was to determine how time of day of an intranasal inflammatory challenge influences the transcriptional and cellular responses within the OB. Methods: For Aim 1, OBs were isolated from male mice at 8 times of day (n = 3 mice/time). Transcriptional profiles of OBs were assessed using a NanoString Murine nCounter Neuroinflammation Panel. For Aim 2, we intranasally challenged mice at different times of day with Poly(I:C). We then measured transcriptional and cellular responses within the OB using the nCounter Neuroinflammation Panel and imaging flow cytometry, respectively. Results: We report that mRNA transcripts within the gene sets of microglia function, inflammatory signaling, and innate immune response were among the most differentially regulated by time of day. Further, we observed upregulated expression of mRNA transcripts involved in virus detection, type I interferon response, and chemotaxis at the beginning of the active phase. Conclusions: The circadian clock primes the OB’s response to intranasal inflammatory stimuli differently depending on time of day of exposure, potentially providing an important gating mechanism underlying differential susceptibility to neurotropic pathogens.

18. Neural ensemble coding and topography in a sensory circuit

Cini, F. A.; Remage-Healey, L. , PBS

Songbirds communicate with other members of their species using songs and calls. A forebrain secondary region, the caudomedial nidopallium (NCM), is essential to understanding and processing these auditory stimuli. One piece of evidence that NCM is involved in auditory learning is that NCM neurons adapt to repeated presentation of song stimuli. Furthermore, NCM has been mostly treated as a single region when looking at song learning and processing. However, there is evidence that different subregions of NCM have different molecular profiles, such as the expression aromatase, an enzyme that synthesizes neuromodulatory estradiol. Also, the expression of immediate early genes in NCM shows that subregions of NCM can specialize in the processing of different features of a song. Therefore, we hypothesize that: (1) stimulus adaptation occurs because NCM neurons form ensembles that encode a song and sparsify over time; and (2) subregions of NCM work together to represent all features of a song. Using a neural probe to record single-unit activity systematically across NCM and a combination of machine learning algorithms to identify neural ensembles, we aim to identify how song adaptation in NCM is accompanied by the formation of neuronal ensembles. Furthermore, with the probes, we will be able to find if neurons from different regions of NCM (dorsal, ventral, rostral, and caudal) contribute equally to ensemble formation.

19. Where do I remember this? Recognition memory for low-level visual stimuli.

de la Rosa Rivera, N.; Cowell, R.A., Neuroscience and Behavior

Most theories of memory assume that long-term, declarative memory relies primarily on the medial temporal lobes (MTL). In contrast, the Representational-Hierarchical (RH) Account – challenges this claim, replacing it with two assumptions: (1) the brain contains a hierarchy of stimulus representations, from simple features (lines, blobs) in visual cortex to high-dimensional associative representations in MTL; and (2) any region in the hierarchy can support memory for the information that it represents. The RH Account thus predicts that long-term, declarative memory for simple, purely visual stimuli should recruit regions outside of the MTL, specifically, in visual cortex. In this study, participants studied a set of simple visual stimuli for 10-20 days via visual search training. Stimuli were built from conjunctions of shape and fill pattern. At test in the MR scanner, subjects saw Studied items, Recombination items (novel recombinations of features of studied items), and Novel items. Participants responded “Old”, “Recombo” or “New” with a button press. Average memory performance (i.e., accuracy) showed that subjects are capable of distinguishing studied visual stimuli (i.e. Old) from novel stimuli (i.e., Recombo, and New). Multivariate classifier analysis (i.e., MVPA) of fMRI brain patterns was used to investigate which brain areas drive successful identification of studied stimuli. Specifically, we asked which regions hold neural representations that permit recognition memory for visual conjunctions and for visual features. Using signal detection theory, classifier prediction results were analyzed by looking at the discriminability of Old and Recombo items (a measure of conjunction memory), against the discriminability of Old and New items (a measure of feature memory). Results suggests that early visual regions (V1-V3) hold memory representations most apt for recognition of low-level features. Meanwhile, more anterior regions, such as lateral occipital (LO) cortex and temporooccipital (TO) cortex seem to hold representations of memory for low-level visual conjunctions, necessary for discriminating Old and Recombo stimuli. In conclusion, results imply that “sensory” brain regions, posterior to MTL, are supporting long-term memory behavior by holding neural representations of visual conjunctions and visual features needed for the memory task.

20. Structure and organization of the olfactory system in the mollusc, Berghia stephanieae

Tait, CC; H Sant; MD Ramirez; Y Meirovitch; Y Wu; JW Lichtman; PS Katz, Biology

The chemosensory system of gastropod molluscs has a structural organization that is distinct from standard models such as Drosophila or mouse. We are using a variety of approaches to study the neuronal organization of the chemosensory system of a nudibranch mollusc, including fluorescence and high resolution volume electron microscopy. These techniques enable us to trace peripheral sensory cells and neuronal pathways. Fluorescence microscopy reveals several populations of primary sensory neurons expressing different peptides and neurotransmitters, which line the sensory epithelium. Peptidergic interneurons are found deeper within the tissue. Efferent serotonergic axons project throughout the rhinophore and there is a plexus of peptidergic axons as well. Additionally, in collaboration with Jeff Lichtman’s lab at Harvard University, we are using volume electron microscopy (EM) and machine learning algorithms to map all synaptic connections between neurons of the rhinophore ganglion from serially sectioned tissue and digitally reconstructing neurons and synapses. When completed, this project will determine the full wiring diagram of an enigmatic brain structure, the rhinophore ganglion, in a previously unstudied species, an undertaking made possible by recent technological advances. Thus far, we found surprising complexity, with the rhinophore ganglion having 9000 neurons, almost twice the number of the animal’s brain, and 30,000 axons in the rhinophore connective having a range of ultrastructural features. Simultaneous staining for gene expression in the rhinophore ganglion revealed that it contains peptidergic, GABAergic, and cholinergic interneurons, each in different patterns. In addition, there is extensive connectivity between the rhinophores and other brain regions and sensory modalities. Using neuronal tracing and EM, we found primary sensory axon projections that bypass the rhinophore ganglion entirely to communicate with the brain. Furthermore, retrograde fills of neurons shows evidence for multisensory integration in the periphery, including a photoreceptor in the eye that sends an axon directly to the rhinophore. Thus, there is widespread multimodal connectivity between the rhinophore and the rest of the brain. Work is continuing to reconstruct the connectivity of the rhinophore ganglion. The results from this project suggest that the organization of this molluscan olfactory system is substantially different from that seen in arthropods and vertebrates, with widespread afferent and efferent connectivity rather than all chemosensory information being funneled to a central location.

21. Robust Control in Real Time Reinforcement Learning using Adaptive Response Times

Patel Devdhar; Walsh Francesca; Zhang Zhongyang; Rahman Tauhidur; Sejnowski Terrence; Siegelmann Hava, Computer Science

As deep reinforcement learning is increasingly applied to real-world control, we must account for an agent’s processing and actuation delays during tasks. To minimize the effect of such delays, control tasks are often implemented on energy-intensive hardware, which processes inputs at high speeds. However, nature has created biological neural reflex circuits to increase survivability while processing environmental information. Taking inspiration from biological reflexes, we demonstrate the effect of reaction speed on the learned policy’s robustness and present a biologically inspired network with an adaptive response capable of more robust performance.

We performed continuous control experiments for the inverted pendulum task. To test the policy’s robustness, we introduced random horizontal perturbation forces lasting 20ms in the environment. Our experiment response times ranged from 10ms to 160ms, and the maximum perturbation forces during training ranged from 0g to 20g. The networks were evaluated in an environment where an increasing perturbation force is introduced every five seconds.

In general, agents with faster response times perform better as they can respond to environmental changes faster, creating a trade-off between the agent’s energy and performance. However, we find that agents with 20ms response times survive longer in the test environment (169 ± 5s) than agents with 10ms response times (153 ± 9s). In theory, it is possible to replicate the best policy with 20ms using a response time of 10ms; however, the agents fail to learn such a policy which suggests that learning becomes harder at faster response times.

A better solution for control in the real world is an adaptive response time depending on the situation. Humans are capable of different reaction times depending on the situation. Our fastest reaction speeds result from spinal reflexes, which involve circuits of very few neurons. Here, we designed a biologically inspired reflex loop with just three neurons that modify the action when the pendulum’s angle is greater than a certain threshold. Thus, the reflex provides a faster but inaccurate response to certain emergency states. We found that adding a quick reflex to networks improves their performance significantly, allowing the agents to withstand perturbation forces of higher magnitude. For slower networks, the reflexes allow similar performance to twice the response speed. Thus, reflexes drastically reduce the amount of actions required (half) while maintaining the same performance.

Together, these two results provide evidence that the agent’s reaction speed changes the control task’s difficulty, and an agent with adaptive response times can leverage this to learn better control policies.

22. Validating Sleep Staging in 6 Consumer Devices

Hall, B.A.; Thakkar, M.; Caccavaro, J.; Kainec, K.A.; Spencer, R.M.C., Department of Psychological and Brain Sciences

In recent years, consumer-grade devices for self-monitoring sleep stages have rapidly increased in both popularity and availability. Despite their prevalence, the accuracy of these novel devices remains untested. Here we sought to validate the accuracy of sleep staging for six commercially available sleep tracking devices, compared to gold standard sleep staging with polysomnography (PSG). In 30 young adults, overnight sleep sessions were concurrently monitored using PSG, research-grade actigraphy, and 6 novel consumer-grade devices. For each sleep variable and device, agreement between the estimated variable and PSG was visualized using Bland-Altman plots. Statistical differences in agreement were calculated using Wilcoxon signed rank tests. Overall, most devices accurately estimated deep sleep, time awake, and total sleep time. However, most devices also underestimated light sleep, while overestimating REM sleep. These results contribute to growing evidence that wearable sleep tracking devices may provide easy and low cost alternatives to PSG for certain sleep variables.

23. Engineered Morphogen Gradient Microsystem for Patterning Forebrain Organoids

Pavon, N; Yang, F; Pak, C; Sun., Y, BMB

The capabilities of human pluripotent stem cells (hPSCs) have been utilized to develop various in vitro models for studying human neurodevelopment. Recently, 3D stem cell aggregates have been utilized to produce representative fetal brain tissue called brain organoids. Notably, they can produce fast and robust data with accurate representation of human tissue and a large yield of material to work with. Further, the benefit of working with human tissue means that results are more easily translatable for biomedical applications as compared to animal models. However, current organoid technologies are plagued with challenges revolving around issues with reproducibility and controlled patterning. This is due to the large number of varying protocols which can rely on either the spontaneous differentiation of stem cells or make use of morphogens for guided differentiation to acquire regional specificity. While these region-specific organoids can then be combined into assembloids that better resemble the interactions of neural regions, it introduces the variable of artificial assembly through tissue contact. In this work, we report a new device design which leverages localized passive diffusion to produce a stable chemical gradient for the patterning of forebrain organoids. We successfully patterned forebrain organoids with a clear Dorsal-Ventral (D-V) axis using a sonic hedgehog (Shh) signal gradient generated by our novel device. These D-V patterned organoid offer a potential model which can better recapitulate early telencephalic development.

24. Cortical metrics of related memory network restructuring due to sleep dependent memory consolidation

Heimer-Bumstead, A; Kainec, K; Spencer, RM, Psychological and Brain Sciences

The sleeping brain is uniquely optimized to consolidate memory. As we sleep memories are successively replayed and overlapping themes are selectively strengthened leading to long term general memory networks. However, some of the mechanisms behind how memories are selected to be transferred from short-term storage to long-term memory are still unknown. The proposed study aims to investigate the role of spreading activation, a known process of waking cognition, in sleep dependent memory consolidation. Using targeted memory reactivation (reactivation of specific memory traces by playing sound cues previously associated with them), we will investigate consolidation of sound-associated memory traces in addition to strong and weak semantic associates. Testing recall of items associated with sound cued words can elucidate the effects of spreading activation in sleep dependent memory consolidation. In addition, we will investigate how semantic memory networks may restructure during sleep. By comparing changes in subjective semantic relatedness between words over sleep to quantitative measures of network interrelatedness, how our memory networks may change due to sleep-dependent memory consolidation can be investigated. Our results could provide a mechanism by which new memories are selected for consolidation and provide new information on how sleep changes our existing memory traces. This can fill a significant theoretical gap in our knowledge of sleep dependent memory consolidation within associated memory networks

25. All-in-one Miniaturized and Multifunctional Hydrogel Neural Probe

Sizhe, Biomedical Engineering

The nervous system replies to complex electrical, chemical, and mechanical signaling. Neural interfaces such as deep brain optogenetics devices, and spinal cored injury functional recovery probes have been applied across different regions in the nervous system to modulate and record these signals. But the limitation of integrating these multiple modalities into one single biocompatible probe with long-term stability has been hindered since miniaturization and minimal tissue damages cannot be achieved by current fabrication methods. This proposed work presents a one-step molding and extrusion method by leveraging the polymer crosslinking chemistry to fabricate poly(vinyl) alcohol (PVA) fiber probes with controllable shrinking behaviors. The preliminary works have demonstrated that by tuning the crosslinking network and crystallization with two crosslinkers, the PVA fiber probe can achieve up to 70% of shrinkage in diameter with high light transmission (96.40%) and low light loss (1.82±0.72dB/cm). Furthermore, a 100 µm microfluidic channel has been integrated on the side of a 600 µm hydrogel fiber via the similar molding and extrusion process. Therefore, this proposed work will utilize the shrinking behavior to incorporate electrodes and microfluidic channels with controllable diameters and aim to validate multiple functionalities of hydrogel probe in social circuit-related brain regions, including the ventral tegmental area (VTA), nucleus accumbens (Nac) and amygdala with photometric recording and chemical deliveries during mice social behaviors tests.

26. High dose tDCS modulates concentration of neurometabolites in targeted brain regions

Shinde A; Nagarajan R; Schlaug G, BIomedical Engineering

Transcranial direct current stimulation (tDCS) is a non-invasive brain-stimulation technique that can modulate brain activity and change the local cerebral excitability/inhibitory balance. Proton MRS is a non-invasive quantitative technique that can measure metabolites' concentrations such as excitatory and inhibitory neurotransmitters within a region of interest in the brain. Combing non-invasive brain-stimulation with Spectroscopic imaging might increase our understanding of how non-invasive brain-stimulation modulates bain function testing different polarities and current strengths.

In this study, we investigated the effects of tDCS (both anode and cathode) applied over the several regions on metabolites’ levels in single voxel MRS with constant high-dose dose levels. Five MRS scans with 128 average MEGA PRESS sequence were recorded using the Siemens scanner at Human Magnetic Resonance Center, Institute of Applied Life Sciences. tDCS was applied concurrently during the 3rd MRS scan. MRS scans were processed using Gannett toolbox to quantify the GABA and Glx(Glutamate + Glutamine) neurometabolite concentrations. The change in GABA from before to after the tDCS stimulation showed a polarity effect. Our study provides insights into how high-dose tDCS leads to polarity dependent modulation of cerebral metabolites and might provide a paradigm allowing us to test the effects of various stimulation parameters as well as effects of combinations of non-invasive brain-stimulation and gaba-ergic medications.

27. Detecting effects of brain-stimulation using a machine learning approach

Mohapatra S, Shinde A, Schlaug G, Computer Science

Non-invasive electrical stimulation can modulate not only local intrinsic brain activity, but also activity in remote, yet connected brain regions. Similarly, multiple electrodes can be applied to modulate activity in entire networks affecting the feedback and integrating of sensorimotor information. Stimulation of cortical networks with a backbone of a white matter fibertract such as the arcuate fasciculus provides an ideal scenario to examine the effects of brain stimulation on entire networks. We tested two different stimulation montages to study modulation of an entire network, namely single electrode (SE) with one electrode placed over nodal brain regions and multi electrode network stimulation (ME-NETS) with placing multiple electrodes simultaneously over nodal cortical regions.

Concurrent tDCS-fMRI data was used to calculate dynamic functional connectivity (DFC) matrices between various regions within and across networks. Machine learning methods were implemented to investigate if different stimulation montages modulate one network (e.g., AF network) differently from another network. Further, the brain regions that strongly contribute to the classification are also identified. Our investigations allowed us to detect montage dependent changes in brain activity immediately after the start of the stimulation providing a quick and easy measure of brain engagement through external brain-modulation.

29. Characterization and manipulation of sleep onset via wearable sensor suite

Cournoyer H; Kainec K; Liang X; Gummeson J; Busa M; Lacreuse A; Padilla S, Psychological and Brain Sciences

Many menopausal women report difficulty falling asleep during the menopause transition, which is associated with worse health outcomes. The current study uses a non-human primate model of advanced reproductive age to precisely measure sleep onset and develop intervention approaches to decrease the latency to sleep. Our first effort in this initiative is to minimize the need for invasive technology to study sleep through the development of a wearable Smart Jacket fitted with a sensor suite that can be used to characterize sleep onset. The external sensors will be validated using the “gold standard” signals of indwelling EEG and EMG of existing telemetry-instrumented animals. External wearable devices that track sleep already exist, but they fail to precisely detect the transition into stage one of sleep. We have incorporated temperature changes into the metric of sleep onset to improve the resolution of this time window. Sleep onset is characterized by a decrease in core body temperature facilitated by an increase in sympathetic vasomotor blood flow and increase in skin temperature. We have developed a method for sleep scoring in the common marmoset that encompasses all three stages of NREM sleep along with core-body temperature, allowing us to better pinpoint the transition to sleep. Current work is focused on the implementation of the Smart Jacket, and once this is established, we will work on intervention approaches to decrease the latency to sleep.

30. CRY 'Havoc!' and Let Slip the Phase of Clocks

Bittman E; Bahiru MS; Sisson C, Biology

Jet lag has adverse effects on health, as circadian oscillators in various organs re-entrain at different rates and the normal phase relationships between physiological processes are disrupted. We know little about what determines the rate of re-entrainment after shifts of the light:dark (LD) cycle. The phase lability of the endogenous clock may have a critical role. Circadian period may also contribute, as it regulates entrained phase angle and hence the phase illuminated when the LD cycle is shifted. We compared re-entrainment latency in hamsters and mice that have a fast circadian clock, utilizing mutant hamsters and mice in which the transcriptional-translational feedback loops (TTFLs) that govern circadian rhythms were altered. We used duper hamsters, which lack CRYPTOCHROME 1 due to a stop codon in exon 4, and generated tau mutant hamsters, in which gain of function of casein kinase 1e destabilizes PERIOD 2. We first established that among free running animals in constant darkness, only the dupers have a high amplitude phase response, i.e., they can shift up to 12h when given a 15' light pulse. Tau heterogygotes and Cry1-ko mice have low amplitude acute phase responses.

We next examined latency to re-entrain (jet lag) upon shifts of a full photoperiod. Hamsters were kept in 14L:10D and subjected to 8h phase advances or delays. Mice were housed in 12L:12D and subjected to 6h phase advances.Duper and tau heterozygote hamsters and Cry1-/- mice all re-entrained more rapidly than wild types after an 8h phase advance of the LD cycle, despite the difference in acute phase responses. Upon 8h delays of the LD cycle, tau heterozygotes advance by 16h, i.e., they entrain antidromically. This occurs only occasionally in dupers, and never in wild types.

These findings indicate that mutations affecting the negative limb of the TTFL (duper and tau) can accelerate phase shifts, but dissociate PRC amplitude from the rate of re-entrainment. When the LD cycle is delayed by 8h, light falls later in the subjective night of animals whose activity onset begins before lights out. 4h delays of the LD cycle elicit antidromic delays in only 30% of tau heterozygotes. As period determines entrained phase angle, it may determine the latency to re-entrain and determine whether orthodromic or antidromic entrainment occurs. Phase shifts occur more rapidly in animals whose endogenous period is shorter than that of the LD cycle, reducing jet lag and its deleterious effects on health.

31. The Artificial Retina - Cellular Neural Networks

Ravichandran V, Flannery B, Maurer T, Xia Q, ECE

Inspired in part by the biological retina, the Cellular Neural Network (CeNN) allows for massively parallel, analog computing schemes based on identical repeating processing elements (cells). These cells are able to communicate via weighted synapses to their eight direct neighbors during network inferencing, where the reconfigurable nature of the synapses allows for a single SoC to handle a multitude of applications with simulations suggesting processing speeds approaching 100,000 frames per second on certain tasks. The parallel nature of the hardware enables high-speed visual data processing, making this network ideal for a variety of image processing applications, including edge detection and object motion tracking.

32. Deletion of CRY1 Alters Neurogenesis in Adult Female Syrian Hamsters

Bahiru Michael; Bittman Eric, Biology

Rotating shift work and jet lag disrupt circadian rhythms and aggravate disease. Cognitive deficits may result from disruption of adult neurogenesis. Cell cycle entry and exit in hippocampal neural stem cells (NSC) is regulated by the circadian clock: proliferation increases in Per2-/- and Bmal1-/- mice. Internal misalignment, a state in which abnormal phase relationships prevail between and within organs, is proposed to account for adverse effects of circadian disruption. This hypothesis has been difficult to test because phase shifts of the entraining cycle lead to transient desynchrony. Thus, it remains possible that phase shifts, regardless of internal desynchrony, account for adverse effects of circadian disruption and alter rates of neurogenesis or the fate of newborn cells.

We examined cell birth and NSC differentiation in duper hamsters. Locomotor activity rhythms of these Cry1-null mutants re-entrain 5-fold faster than wild types after a phase shift of 8 hours. Mutant and wild type adult female hamsters were subjected to alternating 8h advances and delays at eight 16-day intervals. BrdU, a cell birth marker, was given midway through the experiment. Brains were stained for NeuN to assess differentiation of newborn cells. Repeated phase shifts increased the number of newborn neurons, and the duper mutation increased the number of BrdU+/NeuN cells and eliminated this effect of phase shifts. Immunocytochemical staining for proliferating cell nuclear antigen (PCNA) indicated no overall effect of genotype or repeated shifts on cell division rates at the time of sacrifice (P=0.09). Cell differentiation, assessed by doublecortin-ir, was higher in duper hamsters (P<0.02) but was not significantly altered by repeated phase shifts. Our results indicate that the deletion of Cry1 upregulates cell differentiation, but phase shifts may determine survival over weeks after cell birth.

33. The impact of Tau phosphorylation and lysine modification on Tau-LRP1 complex and Tau propagation

Wang C; Sebastian, N; , Biochemistry and Molecular Biology

The spread and aggregation of the protein tau has been linked to a variety of neurodegenerative diseases including Alzheimer’s disease (AD). A growing body of evidence suggests that aggregated tau spreads throughout the brain via cell-to-cell transmission in a “prion-like” manner (Iba et al., 2013, Clavaguera et al.,2009, Guo et al., 2016), “seeding” new tau aggregates and causing neurotoxicity as it goes.

High resolution quantitative mass spectrometry has provided exciting new data on the sequential deposition of tau PTMs during disease progression (Wesseling et al., 2020). Phosphorylation of tau has been historically viewed as an important trigger for tau aggregation (Arendt et al., 2016), and biomarkers for phosphorylated tau (p-tau) in cerebrospinal fluid (CSF) correlate with disease prediction (Barthelemy et al., 2020) suggesting p-tau as a critical culprit of tau spread.

In preliminary work, we have shown that lysine modifications can influence tau endocytosis (Figure 12), but it is still unclear if this phenotype is dependent on LRP1. LRP1 binding proteins have previously been reported to use lysine motifs to interface with LRP1 (Migliorini et al., 2020, Migliorini et al., 2003, van den Biggelaar et al., 2015). We have mutated several patches of lysine residues (K to A) on tau’s MTBR and have found that there are indeed potential “hotspots” that are important for tau endocytosis. Intriguingly, one of these patches (Mut5: K311, K317) is highly modified (both acetylated and ubiquitinated) in AD brains (Wesseling et al., 2020).. In this study, we will examine how phosphorylation and lysine phosphorylation influence interaction with LRP1 and the ability for tau to aggregate in vitro and in cells.