Representation of visual scenes by local neuronal populations in layer 2/3 of mouse visual cortex

Abstract

How are visual scenes encoded in local neural networks of visual cortex? In rodents, visual cortex lacks a columnar organization so that processing of diverse features from a spot in visual space could be performed locally by populations of neighboring neurons. To examine how complex visual scenes are represented by local microcircuits in mouse visual cortex we measured visually evoked responses of layer 2/3 neuronal populations using 3D two-photon calcium imaging. Both natural and artificial movie scenes (10 seconds duration) evoked distributed and sparsely organized responses in local populations of 70-150 neurons within the sampled volumes. About 50% of neurons showed calcium transients during visual scene presentation, of which about half displayed reliable temporal activation patterns. The majority of the reliably responding neurons were activated primarily by one of the four visual scenes applied. Consequently, single-neurons performed poorly in decoding, which visual scene had been presented. In contrast, high levels of decoding performance (>80%) were reached when considering population responses, requiring about 80 randomly picked cells or 20 reliable responders. Furthermore, reliable responding neurons tended to have neighbors sharing the same stimulus preference. Because of this local redundancy, it was beneficial for efficient scene decoding to read out activity from spatially distributed rather than locally clustered neurons. Our results suggest a population code in layer 2/3 of visual cortex, where the visual environment is dynamically represented in the activation of distinct functional sub-networks.

Keywords: 3D imaging; calcium imaging; natural movies; neocortex; two-photon microscopy.

Figure 1

3D calcium imaging of visual responses in layer 2/3 neuronal populations. (A) Stimulus set of visual scenes used in this study. (B) Top: Reference stack of a layer 2/3 cell population labeled with OGB 1 70–130 μm below pial surface (neurons green; astroglia counterstained with SR101, red). Bottom: 3D spiral scan trajectory used to collect data from layer 2/3 population. Neuron positions are indicated by green spheres. (C) Simultaneous 3D population imaging and single-cell juxtacellular recording. Top: 3D reconstruction of the imaged neurons with recorded neuron in red. Bottom: Juxtacellular recorded spikes binned to same sample rate as imaging data (10 Hz). (D) Example responses to Movie A and B with binned spikes (top) and simultaneously imaged fluorescence transients (middle; raw data in blue; filtered data in green). Dotted line indicates the 95th percentile of baseline. Bottom traces show estimated spike rates obtained by deconvolving calcium signal (blue) superimposed with the filtered actually recorded spike rates (black). (E) Average response to 10 consecutive Movie B presentations in a juxtacellularly recorded neuron and the surrounding population. Top: Mean traces for raw and deconvolved calcium signal, filtered spike rate, and peri-stimulus time histogram (PSTH) for the recorded neuron. Bottom: Intensity graph showing the average population response (recorded cell indicated by arrow).

Figure 2

Reliable and specific activation of 3D populations by visual scenes. Upper rows: Example 3D activation pattern for two time points (arrows) during the presentation of Movie A, Movie B, Grating, and Noise stimulus (black bars). Middle rows: Example responses of four neurons (locations indicated by inserted box plots) to repeated stimulation with different visual scenes. Average relative fluorescence changes (ΔF/F) from six trials are shown together with the individual trials (grays). Blue traces are the estimated underlying spike rates (f_s). Bottom row: Intensity graphs showing the average firing rate of the entire population (rows represent individual neurons; time runs on the horizontal axis). Start and end of visual stimulation are indicated by dotted lines. Black arrows indicate example neurons.

Figure 3

Response specificity of local population. (A) and (B) Method to calculate trial-to-trial correlation matrix. (A) Population trial-to-trial correlations were obtained by correlating all pairs of trials of entire population responses to visual scenes. (B) The correlation matrix was filled with pair-wise correlation coefficients of f_s responses for each pair of trials. The example shows six trials per presented visual scene. Trials are sorted by the presented visual scenes indicated on right and top. (C) Trial correlation matrix for entire network response of the population shown in Figure 2. Dashed white lines separate trials with different visual scenes indicated on the left and top. (D) Cumulative distribution of correlation coefficients from the population analysis for all experiments with trials with same visual stimulus (red) or with different visual stimuli (black). Trial-to-trial correlations were significantly higher for same stimulus trials compared to different stimulus trials.

Figure 4

Response specificity of individual neurons. (A) Single-cell trial-to-trial correlations were obtained by correlating trials of individual cell responses to visual scenes. The example shows two trials from the same cell in response to a visual stimulus (Grating). Method to calculate trial correlation matrix is shown in Figure 3. (B) Spike rate intensity graphs and single-neuron trial correlation matrices for the example neurons shown in Figure 2. Black pixels indicate trials without responses to presented visual scenes. Trial order in spike rate intensity graphs and correlation matrices are the same. Dashed white lines separate trials with different visual scenes indicated on the left and top. (C) Cumulative distribution of correlation coefficients from single-cell analysis from all cells in all experiments for trials with same visual stimulus (red) or with different visual stimuli (black). Trial-to-trial correlation coefficients are from single-cell responses. To compare with network responses see Figure 3.

Figure 5

Decoding of visual scenes from 3D population and single-cell responses. (A) Trial classification by nearest mean clustering of population or single-cell responses. Example shows dimension-reduced population response for better visibility (Dim 1–3). Each visual scene was presented six times and corresponding trials were correctly classified. The crosses indicate the average responses (see Methods). (B) Example of single-cell responses to different visual scenes. Correctly classified trials are indicated by red tick marks. Cells with >50% correctly classified trials for any presented visual scene were classified as “reliable responders” to this particular scene. Example cell was classified as reliable responder to all four presented visual scenes. (C) Percentage of correctly classified trials for all population and single-cell responses comprising either all cells or reliably responding neurons only. (D) Pie-chart grouping neurons according to their response properties, pooled over all experiments. Reliable responders are further subdivided into single-scene and multi-scene preferring neurons. Note that the majority of reliable responders show single-scene preference.

Figure 6

Dependence of decoding performance on network size and number of stimuli. (A–C) Decoding performance is measured as percentage of correctly classified trials as shown in Figure 5 and Methods. (A) The percentage of correctly classified trials increases with growing population size in an example experiment. Different colored lines show discrimination of 2, 3, or 4 different visual scenes. (B) Average decoding performance across all experiments depends on number of discriminated visual scenes but is similar for entire population (blue) and networks of reliable responders alone (orange). (C) Required network size for near-optimal decoding depends on the number of discriminated visual scenes. Note that assembling networks of reliable responders alone reduces the required network size. (D–E) Decoding performance measured as mutual information. (D) Mutual information increases with growing population size in same example experiment as in (A). (E) Average decoding performance as in (B) corrected for mutual information content. Note that maximum mutual information depends on the number of visual scenes to discriminate (1, 1.58, and 2 bits, respectively; dashed lines). (F) Required network size to obtain near-optimal information is independent of number of discriminated visual scenes.

Figure 7

Spatial organization of 3D population responses to different visual scenes. (A) Example of 3D distribution of neuronal stimulus preference. Note that few neurons reliably respond to more than one visual scene. (B) Occurrences of neurons with same or different stimulus preferences at different cell-to-cell distances compared to shuffled data sets. Neurons have significantly more neighbors with the same visual scene preference at distances of up to 40 μm than neighbors with different scene preferences. (C) Occurrences of functional clusters of nearest neighbors with same stimulus preference compared to shuffled data sets. (D) Decoding performance for different clusters of nearest neighbors compared to randomly picked groups of five cells. (E) Cumulative distributions of correlation coefficients between cells with different stimulus preferences and spatial locations. “NN 1-4” indicates correlations with the first four nearest neighbors, “NN ≥5” indicates correlations with neurons further away. (F) Average correlation coefficients between neurons are highest for neurons with preference for the same visual scene and located within local clusters.