Training recurrent neural networks with inhibitory neurons in multisensory integration tasks.
In recent years, Recurrent Neural Networks (RNNs) trained to perform cognitive tasks have become a useful tool in neuroscience. However, many RNN implementations are lacking in terms of biophysical realism. Most notably, typical RNNs don't distinguish between excitatory and inhibitory neurons, a realistic feature of neural circuits known as Dale's rule. This limits their applicability to understand perception and cognition. In this talk, we will present our approach for using RNNs, with explicit excitatory and inhibitory units, that are trained to perform several multisensory perception tasks. We will analyze the role of excitatory and inhibitory neurons in the underlying neural computations. Our models are able to reproduce the behavioral performance of rodents in several tasks that involve detection, discrimination and integration of multisensory information. In addition, our results suggest that not just excitatory, but also inhibitory cells, develop a marked specificity to sensory and choice modalities, in agreement with recent experimental data involving multisensory stimuli. These results highlight the importance of implementing biophysically realistic constraints in RNNs, in particular selectivity in excitatory and inhibitory units, to decipher the neural computations for cognitive functions.
Paradoxical response reversal of top-down modulation in cortical circuits with three interneuron types
Pyramidal cells and interneurons expressing parvalbumin (PV), somatostatin (SST), and vasoactive intestinal peptide (VIP) show cell-type-specific connectivity patterns leading to a canonical microcircuit across cortex. Experiments recording from this circuit often report counterintuitive and seemingly contradictory findings. For example, the response of SST cells in mouse V1 to top-down behavioral modulation can change its sign when the visual input changes, a phenomenon that we call response reversal. We developed a theoretical framework to explain these seemingly contradictory effects as emerging phenomena in circuits with two key features: interactions between multiple neural populations and a nonlinear neuronal input-output relationship. Furthermore, we built a cortical circuit model which reproduces counterintuitive dynamics observed in mouse V1. Our analytical calculations pinpoint connection properties critical to response reversal, and predict additional novel types of complex dynamics that could be tested in future experiments.
Cholinergic modulation of hierarchal inhibitory circuitry in the prefrtonal cortex controls resting state activity.
Local circuitry in the superficial layers includes a hierarchal sub-circuit of interneurons with inhibitory and disinhibitory, subtractive and divisive, influence on the principle cells. These neurons play a significant role in structuring on-going activity and are subject to specific modulation by acetylcholine through nicotinic acetylcholine receptors (nAChRs). We show how the various types of nAChRs, located on specific interneurons, impact the properties of ultra-slow transitions between high and low activity states (H-states and L-states, respectively), recorded in mice during quiet wakefulness. Recent data indicate that a genetic mutation of the α5 nAChR receptors located on vasoactive intestinal polypeptide (VIP) inhibitory neurons, appears to be responsible for hypofrontality observed in schizophrenia. Chronic nicotine application mice restores neural activity to control levels.
We will present a reduced circuit model of the hierarchically organized neural populations. Using this model we show that the change of activity patterns recorded in the genetically modified mice can be explained by a change of activity state stability, differentially modulated by cholinergic inputs to parvalbumin (PV), somatostatin (SOM) or VIP inhibitory populations. We demonstrate that desensitization and upregulation of β2 nAChRs located on SOM interneurons by chronic nicotine application could account for activity normalization recorded in α5 SNP mice. We will use the model to predict effects of nicotine withdrawal on the on going activity. Time permitting we will show how the local circuitry can also explain hyper-activity observed in the early stages of Alzheimer’s disease and how this circuitry interacts with brain oscillations.
Engagement of the thalamic reticular nucleus in decision confidence computations
Computational modeling of brain mechanisms of cognition has largely focused on the cortex, but recent experiments have shown that higher-order nuclei of the thalamus participate in major cognitive functions and are implicated in psychiatric disorders. In this computational study, we focus on the macaque pulvinar to show that corticopulvinar projections that engage the thalamic reticular nucleus (TRN) - a sheet of inhibitory neurons surrounding the thalamus - enable the pulvinar to estimate decision confidence. This computation depends on both excitatory and inhibitory plasticity of the corticopulvinar projections. Overall, our model suggests that the TRN is an important and understudied component of the distributed circuitry involved in decision making and possibly other cognitive computations.
Post-inhibitory rebound interacts with preventing or deleting mechanisms to generate theta spiking resonance in hippocampal CA1 pyramidal cells.
A crucial issue in the understanding of neuronal oscillations is to elucidate the microcircuits that are the substrate to these rhythms in the different brain areas. This raises the question of whether rhythmic activity results solely from the properties of the network connectivity (e.g., excitation and inhibition) and topology or it involves the interplay of the latter with the intrincic properties (e.g., ionic currents) of the participating neurons. In this project we address this issue theoretically in the context of the hippocampal area CA1 microcircuits, which include excitatory (PYR) and inhibitory (INT) cells. It has been observed that PYR exhibit a preferred subthreshold frequency response to oscillatory inputs at (4 - 10 Hz) frequencies (resonance) "in vitro". Contrary to expectation, these cells do not exhibit spiking resonance in response to "in vivo" direct oscillatory optogenetic activation, but, surprisingly, spiking resonance in PYR occurs when INT are activated. We combine dynamical systems tools, biophysical modeling and numerical simulations to understand the underlying mechanisms of these rather unexpected results. We show that the low-pass filter results form a combination of post-inhibitory rebound (the ability of a cell to spike in response to inhibition) and the intrinsic properties of PYR. The band-pass filter requires additional timing mechanisms that prevent the occurrence of spikes at low frequencies. We discuss various possible, conceptually different scenarios. These results and tools contribute to building a general theoretical and conceptual framework for the understanding of preferred frequency responses to oscillatory inputs in neuronal networks.
Interneuron circuits for guiding local plasticity based on top-down signals.
Animals learn better when it matters to them. For example, they learn to discriminate sensory stimuli when they receive a reward. As a result of learning, neural responses to sensory stimuli are adjusted even in the first processing stages of sensory areas (Goltstein et al. 2013, Khan et al. 2018, Poort et al. 2015). It is thought that behaviourally relevant contexts, such as rewards, trigger an internal top-down signal available to these early sensory circuits. This could be mediated by cholinergic inputs from the basal forebrain for example (Letzkus et al. 2011). One challenge remains: contextual signals are typically present for a short time only, but synaptic changes require time to be expressed. How can these time scales be bridged? In this talk, I will present recent work on how interneuron circuits can bridge the time scales by guiding synaptic plasticity of excitatory connections. Interneuron circuits recently emerged as key players during learning and memory. We hence investigated how temporary top-down modulation by rewards can interact with local excitatory and inhibitory plasticity to induce long-lasting changes in sensory circuitry. We propose that learning can happen in two stages: 1) Unspecific top-down signals rapidly induce an inhibitory connectivity structure between different interneuron types. 2) The inhibitory structure induces changes in sensory representations by guiding excitatory plasticity between pyramidal cells. Using a computational model of layer 2/3 primary visual cortex, I will demonstrate how inhibitory microcircuits could store information about the rewarded stimulus to guide long-term changes in excitatory connectivity in the absence of further reward. Our model makes specific testable predictions in terms of activity of different neuron types.