The Schedule

To unravel the functional properties of the brain, we need to untangle how neurons interact with each other and coordinate in large-scale recurrent networks. One way to address this question is to measure the functional influence of individual neurons on each other by perturbing them in vivo. Applying such single-neuron perturbations in the mouse visual cortex has recently revealed feature-specific suppression between excitatory neurons despite highly specific excitatory connectivity, which was deemed to underlie feature-specific amplification. Here, we studied which connectivity profiles are consistent with these seemingly contradictory observations by modeling the effect of single-neuron perturbations in large-scale neuronal networks. Our numerical simulations and mathematical analysis revealed that contrary to the prima facie assumption, neither inhibition dominance nor broad inhibition alone was sufficient to explain the experimental findings; instead, strong and functionally specific excitatory–inhibitory connectivity was necessary, consistent with recent findings in the primary visual cortex of rodents. Such networks had a higher capacity to encode and decode natural images, which was accompanied by the emergence of response gain nonlinearities at the population level. Our study provides a general computational framework to investigate how single-neuron perturbations are linked to cortical connectivity and sensory coding and pave the road to map the perturbome of neuronal networks in future studies. 

Recurrent excitatory connectivity of prefrontal neurons has been hypothesized to be a mechanism for working memory maintenance. Bump-attractor models can replicate several properties of prefrontal networks. The bump-attractor architecture may be a general property of PFC networks, such that most neurons have the architecture (e.g., such an arrangement is observed in the ellipsoid body of the drosophila brain). Alternatively, PFC networks could contain mixed architectures, such that a subset of the neurons are part of a bump-attractor network, while the rest are connected with a different architecture, such as randomly connected. Using noise correlation analyses in biological and artificial neural networks we found that the more anterior regions of the lateral prefrontal cortex have a lower proportion of bump-attractor connectivity compared to more posterior regions. These differences could underlie the functional differences observed between these regions.   

Diversity is the norm in most biological systems, and the brain is no exception: significant heterogeneity is observed in fundamental properties of similarly classified neurons in the human brain. Whether this variability drives physiological activity or is merely an epiphenomenon of noisy and stochastic biological processes remains an open question, although important initial evidence for a functional role for this diversity is found in the fact that it is reduced in neuropathology, particularly in the seizure-generating brain regions of patients with epilepsy. Techniques from computational neuroscience are uniquely situated to identify the direct effects that various experimentally-observed heterogeneities (as well as disruptions in neuropathology) have on the dynamics of neuronal circuits as well as the mechanisms by which these dynamics arise. In this seminar, I will review results from my postdoctoral research that strengthen the connection between heterogeneous neuronal circuits and resilience against pathological brain activity, and preview how my independent laboratory at the University of Connecticut is further delineating how experimentally-relevant heterogeneities drive physiological brain activity of functional relevance. 

The brain interdependencies can be studied from either a structural or functional perspective. The former focuses typically on structural connectivity (SC), while the second considers statistical interactions (usually functional connectivity, FC). Notably, while SC is inherently pairwise because it describes white-matter fibers projecting from one region to another, FC is not limited to pairwise interdependencies. Despite this, FC analyses predominantly concentrate on pairwise statistics, usually neglecting the possibility of higher-order interactions. Moreover, the precise relationship between high-order and SC is largely unknown, partly due to the absence of mechanistic models that can efficiently map brain connectomics to functional connectivity.

To investigate these interlinked issues, we have built whole-brain computational models using anatomical and functional MRI data in two applications: healthy aging and transcranial ultrasound stimulation (TUS). We show that non-linear variations in the structural connectome can largely explain the differences in high-order functional interactions between age groups. Moreover, we showed the extent of perturbations in dynamical models to describe the high-order effects of TUS in two different brain targets. 

The biophysical cellular and circuit mechanisms underlying brain function are not yet well understood. Understanding brain function requires studying its components and interactions at different scales: molecular, cellular, circuit, system and behavior. Biophysically detailed modeling provides a tool to integrate, organize and interpret experimental data at multiple scales and translate isolated knowledge into an understanding of brain function. We developed  a software tool called NetPyNE (www.netpyne.org) to help build, optimize, simulate and analyze multiscale brain circuit models. We also developed detailed biophysical models of multiple thalamocortical circuits, including motor, somatosensory and auditory, each with approximately 15k detailed neurons and 30M synapses. The model neuronal densities, classes, morphology, biophysics, and connectivity were derived from experimental data. The models were validated against in vivo firing rate and local field potential data under different behaviors and experimental conditions. These detailed models generated predictions about the long-range and neuromodulatory inputs underlying behavioral changes, their effects across specific layers and cell types, and how multiscale interactions generate different oscillatory patterns. They also provided insights into the biophysical underpinnings of different brain diseases and disorders, including Parkinson's, dystonia, schizophrenia and epilepsy. This provides a quantitative theoretical framework for researchers to evaluate hypotheses, make predictions, guide experiment design, and develop new treatments for brain disorders. 

Feature selectivity, i.e. neurons' heightened responses to specific configurations of stimuli, constitutes a fundamental building block of cortical functions. Various mechanisms have been proposed to explain its origins, differing primarily in their assumptions about the connectivity between neurons. Some models ascribe selectivity to structured, tuning-dependent feedforward or recurrent connections, whereas others have demonstrated that selectivity can emerge within randomly connected networks when interactions are sufficiently strong. This diversity of plausible explanations poses a challenge in identifying the genuine mechanism for feature selectivity in cortex. We developed a novel approach that seeks to minimize preconceived assumptions about the underlying connectivity by utilizing connectomic data at synaptic resolution to construct network models. With this approach, we investigate the mechanisms governing selectivity to oriented visual stimuli in mouse visual cortex. We show that a connectome-constrained network model mimics neural responses seen in experiments and points at randomness in connectivity as the dominant factor shaping selectivity in the circuit. These findings provide novel insights on the mechanisms underlying feature selectivity in cortex and highlight the potential of connectome-based models for exploring the mechanistic underpinnings of brain functions.

Neurological pathologies like Alzheimer’s Disease or Multiple Sclerosis lead to neurodegenerative processes that degrade Structural Connectivity (SC) in the brain, affecting the strength and transmission speed of long-range fiber tracts. At advanced stages, these processes disrupt large-scale brain dynamics and associated Functional Connectivity (FC), impairing cognitive performance. We aim to explore whether modulating local regional dynamics can compensate for the effects of structural degradation, effectively restoring FC as if the original SC strengths and conduction speeds were intact. We began with simple toy models involving a few regions to understand how structural and dynamic changes jointly control FC. We then extend the analysis to large-scale, connectome-based whole-brain models to simulate more realistic scenarios. We found that suitable modifications in local dynamics (e.g., excitability and excitation/inhibition balance) can compensate for SC degradation. Notably, the required dynamical changes are widespread and aspecific (i.e. they do not need to be restricted to specific regions) so that they could be potentially implemented via neuromodulation or pharmacological therapy, globally shifting regional excitability, and/or excitation/inhibition balance. Computational modeling and theory thus suggest that, in the future therapeutic interventions could be designed to “repair brain dynamics” rather than structure to boost.

Brain rhythms in the gamma and beta range are quite noisy in spite of the presence of frequency components. They are characterized by epochs known as rhythm “bursts” during which the power in their respective frequency bands is moderate-to-high. Certain theories of inter-areal communication advocate for the importance of these bursting events. We show that gamma rhythms with proper statistics for these bursts, e.g. their mean duration, are quite robustly generated by rhythm models (such as PING, including Wilson-Cowan with noise) involving excitatory-inhibitory interactions. We further show how to estimate from monkey data the dynamical regime of the model by considering the statistics of the slow envelope of the rhythm. This leads to the conclusion that the likely “operating” regime is one in which the rhythm is induced by neuronal noise. Further, we find how the stability, noise and frequency parameters of the rhythm change as a function of the contrast of a visual grating. We finally show how the problem of burst statistics can be governed by a single meta-parameter for the whole neural model governing the rhythm. This allows one to theoretically predict these statistics from Fokker-Planck theory.