Haim Sompolinsky - US grants

? DESCRIPTION (provided by applicant): We propose to combine whole brain 2-photon imaging of neural activity in behaving larval zebrafish with detailed anatomical and connectivity information extracted from the same animals. The final goal is to generate quantitative models of brain wide neural circuits that explain the dynamic processing of sensory information as well as the generation of motor output by these circuits. Anatomical data will be generated by two complementary technologies: 1) whole brain EM data sets will be prepared from the same fish that were used for calcium imaging. Respective data sets will be registered to each other, functionally relevant neuronal ensembles will then be identified in the EM stacks and connectivity will be analyzed in these sub-networks via sparse reconstruction. 2) EM based connectivity information will be supplemented by trans-synaptic viral tracing technology. These two technologies for identifying synaptic connections have complementary strengths and weaknesses and are thus ideally suited for combination with in-vivo 2-photon calcium imaging studies. The specific power of this approach is that all three techniques, whole brain calcium imaging, viral tracing and EM reconstruction, can be done in the same animal. Functional, anatomical and behavioral data can then be analyzed in the context of the specific stimuli and quantified behavioral output and subsequently synthesized into a theoretical framework. To that end we will start with quantitative models of simple reflex behaviors, like the optomotor and optokinetic reflex, where the transformation of sensory input to motor output is relatively straightforward and well defined. These elementary models will serve as a scaffold that can be refined and complemented by additional data from structure function studies from fish performing in more sophisticated behavioral assays that involve more complex stimuli, different modalities and plastic changes. As such the process of building such a virtual fish will be an iterative, open ended process that requires continuous and bidirectional exchange of information between the theoretical and experimental groups of the research team.

Project 3 - Modeling and theory-Abstract One of the greatest challenges for a theory of brain function is the fractured nature of our experimental knowledge about neuronal circuits. The zebrafish larva offers a unique opportunity to mitigate this barrier due to the ability to interrogate whole brain activity at single neuron resolution during ethologically relevant behaviors, the availability of exhaustive genetic tools and the ability to monitor the animal?s behavior over long times. To utilize these benefits the overall project will study a battery of sensorimotor functions in the fish larva, recording its behavioral responses, the brain structures involved, and the neuronal activity at single cell resolution. The primary goal of Project 3 is to integrate findings from the disparate experimental conditions into a coherent quantitative brain wide circuit model, guiding further validation, refinement and hypothesis testing experiments. We will establish a decade long experimental-theoretical-data-processing collaboration, yielding unprecedented advances in our understanding of the functioning of whole brains in animals with complex neuronal structure and function. We will adopt a multiscale approach, summarized as follows: ?Behavioral models - ?capable of an accurate probabilistic prediction of the animal?s actions given its environment and recent behavioral history. ?Conceptual circuit models ?- initial charts identifying the brain structures involved in a given sensorimotor task and their potential projections. ?Functional circuit models ?- systematic quantitative network models, whose functional units are local populations identified through dimensionality reduction of brain wide activity traces. Importantly, this brain wide model will continuously integrate additional structures and circuits as experiments on specific tasks progress. ?Neuronal circuit models ? data from perturbation and EM validation studies at the single neuron/synaptic level will be integrated into single neuron level models for specific local circuits. We will also incorporate data on action selection, multi-sensory integration, and decision making gathered in experiments of Aim 2 of the Overall Project. Model circuits will be integrated into the Multiscale Virtual Fish software platform to allow visualization and interrogation of experimental and modeling data in a common framework. Our multiscale approach will extend to the time domain. To model the brain state dependence of behavioral and neuronal functions (Overall Aim 3) we will build a hierarchical switchable model, with brain state slow dynamics at the top tier, which modulate the lower tier, ongoing fast behavioral and neuronal models (described above). We will seek to extract from our models of the fish brain and behavior general principles likely to generalize across species. This will be tested by using our modeling approach to related experiments in the rat and the fruit fly larva (Project 4), especially in the domain of brain state dynamics and the neuromodulatory control of behavior and sensorimotor processing.

Project Summary - A realistic multiscale circuit model of the larval zebrafish brain The working group of the BRAIN initiative (BRAIN 2025, a Scientific Vision) identified ?the analysis of circuits of interacting neurons as being particularly rich in opportunity, with potential for revolutionary advances?. They further pointed out that ?truly understanding a circuit requires identifying and characterizing the component cells, defining their synaptic connections with one another, observing their dynamic patterns of activity as their circuit functions in vivo during behavior, and perturbing these patterns to test their significance. It also requires an understanding of the algorithms that govern information processing within a circuit and between interacting circuits in the brain as a whole?. We propose to generate a realistic multiscale circuit model of the larval zebrafish brain ? the multiscale virtual fish (MVF), which is well aligned with the BRAIN initiative's guidelines. The model will be based on algorithms inferred from behavioral assays and it will span spatial ranges across three levels: from the nanoscale at the synaptic level, to the microscale describing local circuits, to the macroscale brain-wide activity patterns distributed across many regions. The model will be constrained and validated by optogenetic interrogation and sparse connectomics of identified circuit elements 1? ,2?. The ultimate purpose is to explain and simulate the quantitative and qualitative nature of behavioral outputs in response to sensory inputs across various timescales, and to explore how these findings might integrate with parallel work in two other important behavioral model systems, ? the ?Drosophila larva and the rat. Our prior U01 project achieved the first instantiation of this model, whereby we successfully dissected the optomotor response (OMR)1? ?, where a larval zebrafish will turn and swim to match the direction of a whole-field visual stimulus ?3?5.? We will build on this model by achieving three further aims: First, we will expand the OMR project with four additional ethologically relevant behaviors: phototaxis, rheotaxis, escape, and hunting. We will extract the precise algorithms underlying each behavior and develop a version of the circuit model to understand their neural implementation. Second, we will further refine the model to account for multimodal integration and decision making, events that naturally happen when conflicting stimuli driving different behaviors are presented simultaneously. For example, a fish might be driven to execute a left turn by whole field motion moving to the left (OMR), while simultaneously being induced to turn right by increased brightness on its right side (phototaxis). Third, we will examine how internal brain states, such as hunger or stress, influence and modulate the specific behaviors (Aim 1) or behavioral interactions (Aim 2). Implementation of neurochemical modulation into the framework of the MVF will be achieved through simulation of highly conserved neuromodulatory neurotransmitter systems such as serotonin, acetylcholine, epinephrine and dopamine. To uncover generalizable principles of circuit design and function, we will compare our findings with those from two other model systems, the fruit fly larva and the rat. This will serve to elucidate the rules, motifs and algorithms of neural circuit function that transcend the potential idiosyncrasies of any given model.