Skip to main content

Reverse engineering sensory perception and decision making: bridging physiology, anatomy and behavior

Periodic Reporting for period 4 - EngineeringPercepts (Reverse engineering sensory perception and decision making: bridging physiology, anatomy and behavior)

Reporting period: 2020-02-01 to 2020-07-31

Unraveling the mechanistic principles that underlie perceptual decision making is extremely challenging, because even the simplest sensory-motor task activates hundred thousands of neurons distributed throughout the entire brain. Moreover, the data provided by the sensory systems, representing the state of the world, is noisy. Yet, the brain is able to transform the noisy sensory input across the hierarchy of cortical processing stages, triggering flexible and nuanced decisions – a hallmark of higher cognition. How the brain is able to accomplish this is still unknown. This is why understanding the neuronal mechanisms underlying decision making is one of the major challenges in systems neuroscience and a crucial step to derive general principles that will influence such varied fields as psychology, economics and political science. The most promising approach to answer this question is monitoring the sensory-evoked signal-flow throughout the brain, at subcellular resolution and millisecond precision. The ERC project “EngineeringPercepts” proposes reverse engineering the sensory-motor pathways involved in a well-controlled decision task, by reconstructing the 3D structure and synaptic wiring of the underlying neural networks, transforming the reconstructed circuits into a model that is populated with sensory-evoked responses, and simulating signal flow at cellular/network levels. By providing simulation predictions that can be tested in the living animal, this ‘bottom-up’ approach will function as a showcase of how to derive generalizable principles across sensory modalities and species.
Part A1: We had reported an article (Egger, Schmitt et al., PNAS 2015) in which we performed in vivo recordings from individual INs in Layer 1 (L1) of rat vS1. We had complemented the L1 INs with data from all layers of rat vS1. We have now analyzed the intrinsic electrophysiological properties, molecular identities, as well as the morphological dendrite/axon characteristics of more than 290 in vitro recorded INs. Based on this dataset we have developed classification routines for assigning IN cell types. Currently we are analyzing more than 40 in vivo recorded INs, complementing our in vitro datasets.
Part A2: We had established a novel approach that enables mapping of putative synaptic contacts between in vivo recorded neurons. We have combined this approach with the development of procedures to validate putative contact sites as synaptic connections via super-resolution light or electron microscopy. Currently we are analyzing data collected with this approach, quantifying the numbers and dendritic locations of synaptic contacts between L5tt PTs and different excitatory/inhibitory cell types in vS1, as well as with thalamocortical neurons. T
Part A3: We had established injections of recombination competent rabies virus into individual muscles. We showed that these injections allow trans-synaptic labeling of L5tt PTs in rat vS1, as well as mapping of their local and long-range input populations (Guest, Seetharama et al., Neuroscience 2018). We had developed classification routines for predicting long-range subcortical target structures of L5tt PTs in rat vS1 (Rojas-Piloni, Guest et al., Nat Comm 2017). We have focused on input maps to L5tt PTs from thalamocortical neurons by combining a novel virus-based optogenetic in vivo approach.
Part B1: We had achieved the goal of subproject B1. As reported in (Rojas-Piloni, Guest et al., Nat Comm 2017), we discovered that L5tt PTs form disjoint output channels to different subcortical target structures, and obtained structural and functional data to constrain simulations of whisker-evoked activity patterns.
Part B2: In collaboration with Prof. P. Strick (University of Pittsburgh, USA), we had accomplished the primary goals of this subproject. We had reported in (Guest, Seetharama et al., Neuroscience 2018) how to combine injections of recombination competent rabies virus into individual muscles with brain-wide quantifications of trans-synaptically connected networks. We used the rabies data to investigate how brain-wide circuits of motor control are organized. We used these data to develop a precise digital model of the vibrissal related part of rat primary motor cortex (vM1).
Part B3: We had recovered the dendrite and axon morphologies of more than 90 excitatory neurons that were recorded in vivo in vM1. We have now fully analyzed this unique dataset, providing unprecedented insight into the cell type-specific axonal innervation patterns and whisker-evoked responses across all layers of vM1.
Part C1: We have published a paper (Egger, Narayanan et al., Neuron 2020) in which we used structural and functional data – as acquired in Parts A1-B3 – to constrain simulations. We directly tested the in silico predictions empirically by combining in vivo electrophysiology with micro-pharmacology and optogenetic input mapping. Our in silico and in vivo data uncovered a novel mechanism by which neurons in the deep layers of the neocortex gate action potential responses of L5tt PTs during sensory stimulation. Our study is first to provide mechanistic insight into how thalamocortical and intracortical circuits interact – in conjunction with the complex non-linear dynamics of synapses and dendrites – to drive sensory-evoked responses in the major output cell type of the neocortex. We have recently expanded the multi-scale model and investigated how empirically measured thalamocortical input distributions enable top-down modulation of sensory-evoked responses in L5tt PTs.
Part C2: We had accomplished one of the primary goals of this subproject. In collaboration with the department of Prof. Sompolinsky at the Hebrew University in Jerusalem (Israel), we reported a scientific article (Landau, Egger et al., Neuron 2016) in which we showed how to combine anatomically and functionally constrained network models of point neurons with simulations of recurrent dynamics that mimic whisker touch. We have explored how to reduce full-compartmental models of L5tt PTs into point neuron models.
We provided proof of principle that the proposed reverse engineering strategy can yield unprecedented mechanistic insight into the complex interplay between synaptic, cellular and network properties during sensory information processing in cortex. Similar to approaches in other complex dynamical systems (e.g. climate research), we have developed a multi-scale model that can be regarded as the most comprehensive digital representation of the cortical circuitry to date. The model allows performing predictive simulations of neuronal processing – with subcellular and millisecond resolution – that mimic in vivo conditions, and that can directly be tested empirically. In future resreach, these models and simulations will aid to uncover how cortical computations can orchestrate sensory-guided behaviors (e.g. perceptual decision making). To illustrate that our reverse engineering approach can be regarded as a general strategy to investigate the interplay between biophysical, cellular and network properties, we have collaborated with the laboratory of Prof. Long at NYU School of Medicine in New York (USA). Prof. Long investigates mechanisms that underlie vocalization in the song bird. As illustrated by two scientific articles, we have successfully transferred our approaches from the rodent to the bird brain.