Excitatory and inhibitory intracortical circuits for orientation and direction selectivity

L. Federico Rossi, Kenneth D. Harris, Matteo CarandinibioRxiv 2019 · DOI

TL;DR

This study reveals how the specific spatial arrangement of excitatory and inhibitory inputs to neurons in the visual cortex dictates their selectivity for orientation and direction. By mapping these inputs, the researchers found that the visual cortex employs a "canonical" circuit motif, similar to earlier stages of visual processing, to achieve these fundamental computations.

What Problem Does This Paper Address?

Neurons in the primary visual cortex (V1), particularly those in layer 2/3 (L2/3), are remarkably adept at detecting specific orientations and directions of visual stimuli. This selectivity is crucial for understanding the visual world. However, how the complex network of excitatory and inhibitory connections within the cortex (intracortical circuits) contributes to this selectivity has been a long-standing question. While it's known that L2/3 neurons receive input from thousands of presynaptic neurons, the precise functional organization of these inputs – how they are distributed spatially and how their tuning properties relate to the postsynaptic neuron's preferences – remained largely unknown, especially for inhibitory inputs. This paper aims to dissect this intricate wiring to understand its role in shaping orientation and direction selectivity.

Methods

The researchers employed a sophisticated combination of techniques to achieve their goals:

  • Rabies Monosynaptic Tracing: This is a powerful technique for tracing direct synaptic connections. By injecting a modified rabies virus into a target neuron (the postsynaptic neuron), they could identify and label its direct presynaptic partners. This allowed them to map the "who's talking to whom" at a single-neuron level.
  • Two-Photon Microscopy: This allowed for high-resolution imaging of neuronal activity (calcium signals) in both the postsynaptic neuron and its presynaptic partners in vivo. They used GCaMP6, a genetically encoded calcium indicator, to visualize excitatory neuron activity.
  • Targeted Electroporation: To label specific neurons for tracing, they used electroporation to introduce genetic material (including a rabies virus receptor) into individual postsynaptic neurons.
  • Photoablation: To ensure they were only studying the inputs to a single, well-defined postsynaptic neuron, they used targeted laser pulses to eliminate any supernumerary electroporated neurons.
  • Retinotopic Mapping: They used sparse noise stimuli to map the visual receptive fields of neurons, allowing them to translate cortical locations into visual space. This was crucial for understanding how the spatial arrangement of inputs in the brain relates to the visual stimuli they respond to.
  • Computational Analysis: They developed algorithms to classify neurons as excitatory or inhibitory based on their GCaMP6 expression and to analyze the spatial distributions and visual tuning of these presynaptic populations.

Datasets

The study utilized data from experiments performed on mice. The specific datasets are not presented as distinct, publicly available datasets for training or evaluation in the traditional machine learning sense. Instead, the "dataset" consists of the meticulously collected and analyzed electrophysiological and anatomical data from the experimental procedures described above.

Key Insights and Ideas

  • Inverted Mexican Hat for Orientation: The spatial distribution of excitatory and inhibitory inputs around a postsynaptic neuron resembles an "inverted Mexican hat." Inhibitory neurons are clustered close to the postsynaptic neuron, while excitatory neurons are more spread out distally. This arrangement is hypothesized to enhance orientation selectivity by creating a center-surround antagonism.
  • Laminar Specialization: Excitatory inputs predominantly come from Layer 4 (L4), while inhibitory inputs are concentrated in Layer 2/3 (L2/3). This laminar segregation suggests distinct roles for different cortical layers in shaping neuronal responses.
  • Coaxial Alignment of Excitation: The spatial distribution of excitatory inputs in visual space is not random; it's elongated along the axis of the postsynaptic neuron's preferred orientation. This alignment directly contributes to the neuron's ability to detect specific orientations.
  • Offset Excitation and Inhibition for Direction: A key finding is the spatially offset arrangement of excitatory and inhibitory inputs relative to the postsynaptic neuron's preferred direction of motion. Excitatory inputs tend to be upstream, while inhibitory inputs are downstream. This offset, combined with the known delay in inhibitory neuron activation, provides a compelling circuit mechanism for direction selectivity.
  • Canonical Circuit Motifs: The identified circuit motifs for orientation and direction selectivity in L2/3 are remarkably similar to those found in earlier stages of the visual pathway (thalamocortical connections) and even in the retina. This suggests these are fundamental, "canonical" building blocks of visual processing that are reused across different brain regions.

Key Results

  • Orientation Selectivity: Excitatory presynaptic neurons were predominantly tuned to the postsynaptic neuron's preferred orientation. This co-tuning was observed in both L2/3 and L4.
  • Spatial Distribution: Inhibitory inputs formed a more concentric distribution around the postsynaptic neuron, while excitatory inputs were more elongated and aligned with the postsynaptic neuron's preferred orientation.
  • Direction Selectivity Mechanism: A significant spatial offset was found between excitatory (upstream) and inhibitory (downstream) inputs relative to the postsynaptic neuron's preferred direction. This offset, when combined with the temporal dynamics of inhibition, predicted direction selectivity.
  • Correlation with Direction Selectivity Index (DSI): The predicted direction selectivity based on the spatial offsets of excitatory and inhibitory inputs correlated well with the measured DSI of the postsynaptic neurons (r = 0.67, p < 0.05).

Limitations and Future Work

  • Mouse Model: While powerful, the study was conducted in mice. Future work could explore whether these findings generalize to other species, particularly primates, which have more complex visual systems.
  • Specific Neuron Types: The study broadly classified neurons as excitatory or inhibitory. Further research could delve into the specific subtypes of inhibitory neurons and their distinct contributions.
  • Dynamic Nature of Circuits: The study provides a snapshot of circuit organization. Investigating how these circuits change with experience or during different behavioral states could offer deeper insights.
  • Generalizability: While the paper suggests canonical circuits, further investigation is needed to confirm their presence and specific implementation across a wider range of brain areas and functions. The authors suggest that these circuit motifs might be fundamental building blocks for sensory processing across the brain.

Summary updated

Abstract

The computations performed by a neuron arise from the functional properties of the circuits providing its synaptic inputs. A prime example of these computations is the selectivity of primary visual cortex (V1) for orientation and motion direction. V1 neurons in layer 2/3 (L2/3) receive input mostly from intracortical circuits 1 , which involve excitation 2-9 and inhibition 10-12 . To understand how an L2/3 neuron achieves its selectivity, therefore, one must characterize the functional organization of both its excitatory and inhibitory presynaptic ensembles. Here we establish this organization, and show how it predicts orientation selectivity and reveals a new cortical circuit for direction selectivity. We identified the presynaptic partners of pyramidal neurons in mouse V1 through rabies monosynaptic tracing 1,13 , and imaged the functional properties of the postsynaptic neuron and of its presynaptic ensemble. Excitatory presynaptic neurons were predominantly tuned to the postsynaptic neuron’s preferred orientation. Excitation and inhibition described an inverted Mexican hat, with inhibitory presynaptic neurons densest near the postsynaptic neuron and excitatory ones distributed more distally. Excitation and inhibition also differed in laminar origin: inhibitory presynaptic neurons concentrated in L2/3 while excitatory ones dominated in L4. The distribution of excitatory neurons in visual space was coaxial with the postsynaptic neuron’s preferred orientation and lay upstream of the neuron’s preferred direction. Inhibitory presynaptic neurons, instead, clustered more symmetrically around the postsynaptic neuron and favoured locations downstream of its preferred direction. These results demonstrate that L2/3 neurons obtain orientation selectivity from co-tuned neurons in L4 and beyond, and enhance it by contrasting an elongated excitatory input with a concentric inhibitory input. Moreover, L2/3 neurons can obtain direction selectivity through visually offset 14 excitation and inhibition. These circuit motifs resemble those seen in the thalamocortical pathway 15-20 and in direction selective cells in the retina 21,22 , suggesting that they are canonical across brain regions.
BibTeX
@misc{rossi2019excitatory,
  author = {L. Federico Rossi and Kenneth D. Harris and Matteo Carandini},
  title = {{Excitatory and inhibitory intracortical circuits for orientation and direction selectivity}},
  year = {2019},
  doi = {10.1101/556795},
  howpublished = {bioRxiv},
}
Figure