Laminar microcircuitry of visual cortex producing attention-associated electric fields

Cognitive operations are widely studied by measuring electric fields through EEG and ECoG. However, despite their widespread use, the neural circuitry giving rise to these signals remains unknown because the functional architecture of cortical columns producing attention-associated electric fields has not been explored. Here, we detail the laminar cortical circuitry underlying an attention-associated electric field measured over posterior regions of the brain in humans and monkeys. First, we identified visual cortical area V4 as one plausible contributor to this attention-associated electric field through inverse modeling of cranial EEG in macaque monkeys performing a visual attention task. Next, we performed laminar neurophysiological recordings on the prelunate gyrus and identified the electric-field-producing dipoles as synaptic activity in distinct cortical layers of area V4. Specifically, activation in the extragranular layers of cortex resulted in the generation of the attention-associated dipole. Feature selectivity of a given cortical column determined the overall contribution to this electric field. Columns selective for the attended feature contributed more to the electric field than columns selective for a different feature. Last, the laminar profile of synaptic activity generated by V4 was sufficient to produce an attention-associated signal measurable outside of the column. These findings suggest that the top-down recipient cortical layers produce an attention-associated electric field that can be measured extracortically with the relative contribution of each column depending upon the underlying functional architecture.


37
Research into extracranial electric fields provides fundamental insights into the mechanisms of human perception, 38 cognition, and intention. For instance, event-related potential (ERP) components like the N2pc (Luck and Hillyard, 39 1994; Eimer, 1996; Woodman and Luck, 1999) and Pd (Hickey et al., 2009) reliably index selective attention in 40 humans and monkeys, alike. However, the interpretation of these extracranial measures of attention is severely limited 41 by uncertainty about the exact neural processes that generate these signals (Nunez and Srinivasan, 2006).

42
Understanding what brain processes an electric field indicates requires knowing how it is generated (e.g., Cohen, 43 2017).

44
One avenue to localize neural generators of electric fields is through inverse source localization (Michel et al.,     . EEG traces and inverse source localization for the N2pc index of attention in monkeys. (A) EEG was recorded from electrodes a rranged according to the 10-20 system in monkeys performing visual search for a colored oddball stimulus (see monitor diagrams showing two example search arrays in top panels). Blue and red shading indicates the relationship between visual hemifields and cerebral hemispheres to highlight the mapping between lateralized EEG signals and target location. (B) Trial-averaged EEG traces from monkey Z following presentation of search arrays with the target in either the right (blue) or left (red) visual hemifield. The voltage differential that characterizes the N2pc arises ~150 ms after array presentation for the target in the left vs. right visual fields (orange arrows). The N2pc was significant at posterior sites P5 and P6 (dependent samples t test between response polarizations averaged between 125-250 ms following array onset for contralateral and ipsilateral target presentations (t(35) = 2.42, p = 0.02)). (C) Inverse solution using sLORETA to determine current distribution consistent with voltage distribution during the N2pc (113-182 ms) when the target was in the left hemifield. Current density is displayed over the 3D boundary element model derived from a magnetic resonance scan of monkey Z. Data was clipped below the 85% maximum value for display purposes. Cyan disks indicate EEG electrode positions. Current density is concentrated beneath electrode P6 caudal to the lunate sulcus and in area V4 on the prelunate gyrus. Both results in B and C are reproduced for a second monkey in Figure S1.
Here we show that visual cortex generates dipoles through layer-specific transsynaptic currents that give rise to 68 electric fields that track the deployment of selective attention. ERPs as voltages aligned to array onset, averaged over all trials when the target was presented contra-(solid) or ipsilateral (dashed) to the electrode. Inset magnifies the N2pc window, highlighted in orange, defined as the difference in potentials 150-190 ms following array onset. (B) Cortical (laminar) current source density (CSD), aligned on array presentation when the target appeared in the population receptive field of the column. Dashed lines delineate estimated boundaries between supragranular (L2/3), granular (L4), and infragranular (L5/6) layers. CSD values were interpolated and smoothed along depth for display only. Current sinks are indicated by hotter hues and current sources by cooler hues, respectively. The earliest sink arises in putative L4, likely from rapid feedforward transmission. (C) CSD evoked by target outside the receptive field. (D) Subtraction of CSD responses shown in B and C. The only statistically significant differences (determined through a t test across time with p < 0.05, outlined by magenta line) were due to a current sink in L2/3 that arose gradually ~100 ms after array presentation. This relative sink was associated with a weak relative source in L5/6. (E) Mutual information between CSD and the extracranial signal for L2/3 (blue), L4 (purple), and L5/6 (green), aligned on array onset. Timepoints with significant mutual information were computed through Monte Carlo shuffle simulations (MCS). Epochs with significant mutual information persisting for at least 10 ms are indicated by horizontal bars.

153
Second, we measured target information across the layers of V4 during the N2pc temporal window. This analysis 154 revealed enhanced information in L2/3 and L5/6 but not in L4 ( Figure S2). Third, we computed the mutual information 155 between the extracranial signal and CSD during the N2pc window, irrespective of target position. This analysis showed 156 a significant relationship between extracranial signal and the CSD in L2/3 and L5/6 but not in L4 ( Figure 2E, S2).

157
Fourth, we measured the transmitted information about target location from CSD to extracranial signal during the N2pc 158 window (Timme and Lapish, 2018). This analysis demonstrated significant information transmission to the extracranial 159 signal from L2/3 and L5/6, but not L4 ( Figure 2F, S2).

160
Averaged across sessions, we observed that the attention-associated electric field during the N2pc window 161 ( Figure 3A) was associated with a consistent CSD pattern ( Figure 3B). This relationship was observable in each . Average +2 SEM of information transmission during the N2pc window (right). Panel below shows that Information transmission from L2/3 and in L5/6 was significantly greater than that from L4 (t test p < 0.05). Timepoints with significant information transmission were assessed through Monte Carlo simulations during >75% of sessions. Epochs with significance persisting for at least 10 ms are indicated by horizontal bars, color coded for each laminar compartment (bottom). monkey ( Figure S3A-B). Presentation of the search array in any configuration elicited an early current sink in L4, followed by a prolonged sink in L2/3 that was associated with a briefer source in L5/6. significantly greater in L2/3 and L5/6 relative to L4 (L2/3-L4: t(29) = 2.15, p = 0.040; L5/6-L4: t(29) = 2.20, p = 0.036). The hue of each point across cortical depth signifies the value of a color selectivity index (CSI), derived from local gamma power. CSI values <0 i ndicate preference for green, and values >0, preference for red. CSI is smoothed across 2 adjacent channels for display only. Sessions are sorted from left to right based on a second index that estimates each column's combined selectivity, termed column color selectivity index (CCSI). A bar plot representing each session's CCSI is plotted below. Asterisks indicate columns that were significantly color-selective (Wilcoxon sign rank, p < 0.05). Asterisk color indicates which monkey the column was recorded from (monkey Ca: cyan; He: magenta). (C) Average ERPs for trials when a red (top) or green (bottom) target or distractor appeared in the RF based on the 17 significantly color selective columns. Conventions as in Figure 2. (D) Difference in CSD when the target appeared within the columnar population receptive field (RF), compared to out-of-RF trials when a red (top) or green (bottom) target or distractor appeared in the RF (n = 17). (E) Average ERP for trials when the preferred color (top) or non -preferred color (bottom) target or distractor appeared in the RF for the 17 color selective columns. Conventions as in Figure 2. (F) Difference in CSD when the target was within vs. out of the RF, for trials when the preferred color (top) or non -preferred color (bottom) target or distractor appeared in the RF. (G) Average across colorselective columns for subtraction of information transmission from laminar CSD to the extracranial signal about non-preferred color target position from information transmission about preferred color target position. Conventions as before. L2/3 and L5/6 but not L4 contribu te significantly to the extracranial signal. (H) Correlation plots between the CCSI for each session and the difference in information transmission between the red and green stimulus conditions for L2/3 (blue, top), L4 (purple, center), and L5/6 (green, bottom) . Spearman correlation reported in lower righthand corner of each plot. Data from all 30 sessions included. (I) Comparison of feature selective (solid line, n s = 17) and non-feature-selective (dashed line, nn = 13) columns for each laminar compartment (L2/3: blue, top; L4: purple, upper middle; L5/6: green, lower middle). Differences in time are shown at the bottom for each compartment at two alpha levels (filled: 0.05; unfilled: 0.1) as computed by a two-sample t test. Average information transmission during the time of N2pc indicated with bars at right with upper limit of 95% confidence intervals (left bars, selective columns; right bars, non-selective columns). Significance is indicated with a magenta bracket and p value from a two-sample t test shown to the right of ordinate.

170
Across sessions, the three other information theoretic analyses were consistent with the example session 171 ( Figure S2). Significant information transmission during the time of the N2pc was observed in each monkey ( Figure   172 S3C). Thus, the current dipole in V4 generated by the L2/3 CSD sink and the L5/6 CSD source contributes to the N2pc 173 measured in the overlying extracranial electric field.

175
Feature selectivity determines columns' relative roles for attention-associated electric field generation 176 Given the selectivity of V4 neurons for color ( Figure 4A) (Roe et al. 2012) and the homogenous columnar 177 representation of V4 color selectivity (Zeki, 1973(Zeki, , 1980

188
We found that more than half of V4 columns (monkey Ca: 12/21, 57.1%; He: 5/9, 55.6%) show color selectivity 189 defined in this way. We computed the information transmission of target position for each of these color tuned

215
In a similar vein, we tested whether feature 216 selective columns, on average, transmitted more 217 information than their non-feature-selective counterparts.

218
We found that feature selective columns, along all laminar 219 compartments, transmitted significantly more information 220 (Figure 4I

226
Translaminar currents in V4 recapitulate the N2pc 227 CSD is computed by differentiating between local field 228 potentials to eliminate volume conducted signals that do 229 Figure 5. Comparing an estimated field potential generated from the CSD across the cortical columns to the actually observed extracranial event-related potential. Black lines indicate the empirically measured event-related potential (ERPobs, top), averaged across sessions. The pink line indicates the estimated event-related potential calculated from the synaptic currents across V4 columns, averaged across sessions (ERPcal, center). Synaptic currents at each electrode are measured and divided by the Euclidean distance of the electrode from the extracranial surface (see Methods; Nicholson and Llinas 1971; Kajikawa and Schroeder 2011). ERP for target present in the RF vs. target opposite the RF is shown as solid and dashed lines, respectively (example array for each condition shown at top right). Clouds around ERPcal lines indicate 95% confidence intervals across sessions for each condition. Note that despite differences in overall waveshape (which are likely due to the fact that V4 is not the only contributor to the attention-independent, visually evoked ERP), the timing of differences within signal types can be compared. The congruence in polarization of the difference in potentials is of similar note. not arise from local circuit activity. Using the inverse procedure (i.e., summing the CSD), it is possible to estimate the local field potential without contamination by volume conducted activity (Nicholson and Llinas, 1971; Kajikawa and period of the N2pc ( Figure 5). In other words, the summed potential generated by currents along V4 columns 235 differentiates between attention conditions simultaneous with the extracranially measured attention-associated signal.

236
Note that the shape of the of the empirically observed extracranial ERP (ERPobs) differs from the estimated extracranial

237
ERPcal. This is expected in part because the ERPobs reflects several more variables such as volume conducted 238 contributions of nearby columns as well as the filtering and attenuating effects of the tissue and cranium above the 239 gray matter (Nunez and Srinivasan, 2006). Given these expected differences, it is remarkable how well the difference

255
Unexpectedly, we discovered that the contribution of a cortical column to the overlying electric field depended on 256 whether the visual feature in the RF matched the selectivity of the columnan important consideration in the 257 mechanism producing EEG potentials that may not be observable through the macroscopic EEG signal alone.

259
Columnar mosaic underlying EEG interactions producing these attention-associated electric fields, the approach also affords the opportunity to better  (Zeki, 1973(Zeki, , 1980Tootell et al., 2004; 270 Conway and Tsao, 2009; Kotake et al., 2009). We demonstrate that color-feature selectivity was consistent along 271 cortical depth and discovered that the contribution of a column to the global electric field was greater when the feature 272 in the RF was the preferred feature of the column. Specifically, columns that preferred green (or red) contributed more 273 to the electric field when the item in the RF was green (or red) rather than red (or green).

274
The implications of this unexpected finding are illustrated in Figure 8A   descend to extragranular compartments. We computed CSD and identified the granular input sink session-wise.

396
Inverse modeling of 10-20 EEG recordings was performed in CURRY 8 (Compumedics Neuroscan). 3D head 397 reconstruction was created for each monkey (P and Z) using the boundary element method (Hämäläinen and Sarvas, 398 1989). This method takes into account individual monkey's surface morphologies to create models of cortex surface,

399
inner and outer skull, and skin boundaries. This model was used in conjunction with EEG to compute a voltage 400 distribution over the cortical surface. We calculated the current density with sLORETA, which calculates a minimum 401 norm least squares that divides current by the size of its associated error bar, yielding F scores of activation.

454
Above information theory analyses were performed using the Neuroscience Information Theory Toolbox 455 (Timme and Lapish, 2018). Pairwise mutual information and information transmission were computed at each 456 timepoint across trials for each session using default parameters. Five uniform count bins were used for data binning.

692
Organization of traces reflects electrode positions. Scale is consistent across traces and is indicated by OL. N2pc was found to be significant through 693 an ANOVA measured as the interaction between posterior electrode sites, the target position in the array, and the set size si tes (sites OR and OL,