Brain-state mediated modulation of inter-laminar dependencies in visual cortex

Abstract

Spatial attention is critical for recognizing behaviorally relevant objects in a cluttered environment. How the deployment of spatial attention aids the hierarchical computations of object recognition remains unclear. We investigated this in the laminar cortical network of visual area V4, an area strongly modulated by attention. We found that deployment of attention strengthened unique dependencies in neural activity across cortical layers. On the other hand, shared dependencies were reduced within the excitatory population of a layer. Surprisingly, attention strengthened unique dependencies within a laminar population. Crucially, these modulation patterns were also observed during successful behavioral outcomes that are thought to be mediated by internal brain state fluctuations. Successful behavioral outcomes were also associated with phases of reduced neural excitability, suggesting a mechanism for enhanced information transfer during optimal states. Our results suggest common computation goals of optimal sensory states that are attained by either task demands or internal fluctuations.

Fig. 1. Dependency decomposition in a multi-variate system using pairwise and network models.

a Simplified Partial information decomposition (PID) framework^, based definition of types of information that multiple sources can have about a target (see Methods and Supp. Fig S1). b A synthetic ensemble of eight neural variables with two kinds of dependencies – unique or shared – between seven source variables (black) and one target variable (cyan). All interactions are excitatory. Strength of dependencies is determined by model parameters P_unique and P_shared (see “Methods”). c1, c2 Information fraction (reduction in the proportion of total entropy) as a function of parameters (P_unique, P_shared) that control unique and shared information in the model. Information fraction estimation as a function of P_shared (c2) was performed using a subset of variables in the simulated network for computational efficiency, (see Methods). c3 Normalized total mutual information, measured by uncertainty coefficient, as a function of the sum of model parameters (P_unique,P_shared) that varied unique and shared components of mutual information in a monotonic way. d Coefficients of a pairwise model (univariate logistic regression (UR)) as a function of P_unique and P_shared. White arrow provides a visual guide for direction of highest change in coefficients. e Coefficients of a network model (LASSO multivariate regularized regression (RR)) as a function of P_unique and P_shared. White arrow provides a visual guide for direction of highest change in coefficients. f Application of pairwise and network-based statistical models for approximate information decomposition in an example multivariate system. It illustrates interpretation of the modulation of these dependencies using the PID framework. g Schema for utilizing pairwise and network methods for the estimation of total (brown) and unique (green) information modulation respectively, and to infer the modulation of shared information (purple) based on the PID framework. Shaded blocks indicate indeterminate modulation direction of shared information in the network.

Fig. 2. Multi-timescale & weighted Dynamic Bayesian Networks based estimation of unique dependencies in a sparsely sampled recurrent neuronal network.

a Bayesian Networks for graph representation of dependencies in a multivariate system. b Dynamic Bayesian Networks (DBN) for graph representation of dependencies in multivariate time series data. c Analysis flow for multi-timescale weighted DBN (MTwDBN) graphical model fitting. d Edge weight of MTwDBN graph as a function of connection weight in a 2-population simulated network using the pipeline in c. Error bars indicate 95% confidence interval (n = 100 weighted DAGs). e Spiking activity of 6 subpopulations in a simulated network with recurrent connectivity. Connectivity is visualized in the overlaid schematic. f Directed dependencies (edge in the graph) in the simulated network in e, estimated using MTwDBN fitting. g Summary graph of dependencies across all timescales from f. Solid and dashed lines indicate two different timescales. h F-score (harmonic mean of precision and recall of dependency structure) as a function of % of neural population observed. F-score was estimated for shuffle corrected weighted DAGs (MTwDBN, green), weighted DAGs with a fixed threshold (weightedFT, blue), unweighted DAGs (red), or LASSO regression, an example of regularized regression (RR) models (black). Each point represents the average of five separate runs, except 100% (single run). Error bars indicate standard deviation, some error bars are smaller than symbol size. i DBN decoder accuracy with different sizes of MTwDBN DAGs. Decoders were trained to predict population activities using a subsample of shuffle-corrected edges (see “Methods”). Graph edges for the decoder were sampled from the learned structure either in an unbiased fashion (black) or biased with the edge weights (green). Box indicates lower quartile, median, and upper quartile; whiskers indicate range of data points (n = 100 model seeds). Asterisk (*) indicates significant differences between unbiased and weight biased M-Scores (p < 0.001, two-tailed paired t-test, Bonferroni adjusted). j Schema for estimating modulation of unique dependencies in a network of neural populations, using MTwDBN.

Fig. 3. Laminar recordings in area V4.

a Experimental protocol: Paired Gabor stimuli with varying contrasts (see “Methods”); one stimulus was presented inside the receptive fields (RFs) of the recorded neurons and the other at an equally eccentric location across the vertical meridian. Attention was cued either to the neurons’ RFs (IN) or to the location in the contralateral visual hemifield (AWAY). The orientation of one of the two stimuli changed at a random time. The monkey was rewarded for detecting the change by making a saccade to the corresponding location. Task difficulty was controlled by the magnitude of orientation change. b Left, Recording approach: Laminar recordings in visual area V4. Middle, Stacked contour plot showing spatial receptive fields (RFs) along the laminar probe from an example session. Alignment of RFs indicates perpendicular penetration down a cortical column. Zero depth represents the center of the input layer as estimated with current source density (CSD) analysis. Right, CSD is displayed as a colored map. The x-axis represents time from stimulus onset; the y-axis represents cortical depth. The CSD map has been spatially smoothed for visualization. c An example trial showing single-unit activity across the cortical depth in the attend-in condition. The time axis is referenced to the appearance of the fixation spot. Spikes (vertical ticks) in each recording channel come from either single units (blue, orange) or multi-units (black). Spike waveforms for an example narrow-spiking (blue) and a broad-spiking (orange) single unit are shown. The bars at the bottom depict stimulus presentation epochs, with height indicating relative stimulus contrast. The brain schematic in (b) is adapted from Nandy, A.S., Nassi, J.J., Jadi, M.P., Reynolds, J.H. (2019) Optogenetically induced low-frequency correlations impair perception eLife 8:e35123. 10.7554/eLife.35123 and is under a CC BY license: https://creativecommons.org/licenses/by/4.0/.

Fig. 4. Modulation of dependencies in a V4 laminar network across attention conditions.

a Neural populations used for fitting laminar MTwDBN. Current source density analysis identified different layers (superficial, input, deep), and isolated single units were assigned to one of these layers (see Methods). b Top: Average MTwDBN-based modulation (green) of all unique dependencies between the laminar populations. Modulation of the same dependencies as estimated by logistic regression (brown). Error bars indicate 95% confidence intervals. Bottom: Visualization of modulation sign of unique dependencies at different lags. Combining the modulation sign of unique and total dependencies (see Top), PID framework-based estimated modulation sign of shared dependencies (using schema in Fig. 1g) is also shown for different lags. Thicker line along the time axis indicates the timescales of attentional modulation in prior studies^,. c Sign of average modulation of unique dependencies between layers (bi-directionally). I: input layer; S: superficial layer. d Sign of average modulation of shared dependencies within layers. e Sign of average modulation of unique dependencies within layers. f Summary of dependency modulation pattern. g Neural populations used for fitting laminar MTwDBN. Isolated single units were classified as broad- and narrow-spiking based on peak-to-trough duration in their average spike shape (see Methods). h Top: Average MTwDBN-based modulation (green) of all unique dependencies between the cell-type specific laminar populations. Modulation of the same dependencies as estimated by logistic regression (brown). Bottom: Visualization of modulation sign of unique dependencies and PID framework-based estimated modulation sign of shared dependencies is also shown for different lags. i Sign of average modulation of unique dependencies between layers. j Sign of average modulation of shared dependencies within layers for all, broad or narrow populations. M1, M2: subject-wise; broad, narrow: cell-class specific. k Sign of average modulation of unique dependencies within layers. l Summary of dependency modulation pattern. See Fig S3 for modulation indices in (c–e, i–k). Data points in (b, h) indicate mean, error bars indicate 95% confidence interval (n = 5000 bootstraps).

Fig. 5. Modulation of dependencies in a V4 laminar network across behavioral outcomes at perceptual threshold.

a Example session showing performance as a function of task difficulty. Gray box: threshold orientation change at which the animal was equally likely to correctly detect (hit) or fail to detect (miss) the change. Error bars indicate standard deviation (n = 20 jackknifes). b Laminar populations used for multi-lag analysis. c Modulation magnitude (top) and sign (bottom) of all unique (green) and total (brown) laminar dependencies in b across Hits and Misses at perceptual threshold. Estimated modulation sign of shared dependencies (bottom, see Fig. 1g). Data points indicate mean, error bars indicate 95% confidence interval (n = 5000 bootstraps). d Modulation sign of between layer (BL) and within layer (WL) dependencies. See Fig S4 for modulation indices. e Summary of laminar dependency modulation pattern. f Wideband (5–40 Hz) LFP signals (colored lines) overlaid on the raw LFP (0–200 Hz) signals (gray) in the input layer in a portion of an example session. The generalized phase (color-coded) depicts the dominant phase of the wideband LFP. Vertical ticks indicate single-unit spikes in the corresponding channel (see “Methods”). g Top: Target stimulus presentation probability as a function of the generalized phase of the LFP (adjusted for cortical delay), separated by HIT and MISS trials. Asterisk (*) indicates phases with significant differences (p < 0.05) between the two trial types (Ranked sum test, corrected for multiple comparisons). Bottom: Spike probability in the input layer as a function of generalized phase of the LFP, separately estimated for putative excitatory (broad) and putative inhibitory (narrow) units. For other layers, see Fig S5. Error bands indicate standard error of the mean.