RESEARCH ARTICLE Sensory Processing The primate cortical LFP exhibits multiple spectral and temporal gradients and widespread task dependence during visual short-term memory Steven J. Hoffman,1,3 Nicholas M. Dotson,1,4 Vinicius Lima,2 and Charles M. Gray1 1Department of Cell Biology and Neuroscience, Montana State University, Bozeman, Montana, United States; 2Aix Marseille Universit�e, INSERM, Systems Neuroscience Institute, Marseille, France; 3Department of Biomedical Engineering, University of Minnesota, Minneapolis, Minnesota, United States; and 4Salk Institute for Biological Studies, La Jolla, California, United States Abstract Although cognitive functions are hypothesized to be mediated by synchronous neuronal interactions in multiple frequency bands among widely distributed cortical areas, we still lack a basic understanding of the distribution and task dependence of oscillatory activity across the cortical map. Here, we ask how the spectral and temporal properties of the local field potential (LFP) vary across the primate cerebral cortex, and how they are modulated during visual short-term memory. We measured the LFP from 55 cortical areas in two macaque monkeys while they performed a visual delayed match to sample task. Analysis of peak frequencies in the LFP power spectra reveals multiple discrete frequency bands between 3 and 80 Hz that differ between the two monkeys. The LFP power in each band, as well as the sample entropy, a measure of signal complexity, display distinct spatial gradients across the cortex, some of which correlate with reported spine counts in cortical pyramidal neurons. Cortical areas can be robustly decoded using a small number of spectral and temporal parameters, and significant task-dependent increases and decreases in spectral power occur in all cortical areas. These findings reveal pronounced, widespread, and spatially organized gradients in the spectral and temporal activity of cortical areas. Task-dependent changes in cortical activity are globally distributed, even for a simple cognitive task. NEW & NOTEWORTHY We recorded extracellular electrophysiological signals from roughly the breadth and depth of a cortical hemisphere in nonhuman primates (NHPs) performing a visual memory task. Analyses of the band-limited local field potential (LFP) power displayed widespread, frequency-dependent cortical gradients in spectral power. Using a machine learning classi- fier, these features allowed robust cortical area decoding. Further task dependence in LFP power were found to be widespread, indicating large-scale gradients of LFP activity, and task-related activity. cognition; electrophysiology; large-scale; LFP; short-term memory INTRODUCTION Extensive evidence, from studies in both humans and nonhuman primates, has established that cognitive tasks engage synchronized neuronal oscillations in multiple fre- quency bands among widely distributed cortical areas (1–13). A number of functional roles have been proposed for these cortical rhythms, ranging from perceptual binding (14, 15), working memory (7, 12, 16), and attention (17, 18) to inhibi- tory gating mechanisms (19, 20) and consciousness (21), of- ten with the implicit assumption that the cortical rhythms of interest are expressed throughout the cortex. However, despite decades of research on the function and mecha- nisms of cortical oscillatory activity (22–25), it is still unclear how the various cortical rhythms are distributed across the cortical mantle. Early electroencephalographic (EEG) recordings of human brain activity revealed salient oscillations in the a (�7–14 Hz) and b (�15–30 Hz) frequency bands (26–28). Spatial mapping of the surface cortical EEG revealed a wide distribution of a ac- tivity throughout parietal, temporal, and occipital regions, as well as portions of the anterior frontal lobe andmore localized b activity in sensorimotor and premotor cortices (29). Recent magnetoencephalography (MEG) and intracranial EEG studies Correspondence: S. J. Hoffman (hoff1695@umn.edu); C. M. Gray (cmgray.montana@gmail.com). Submitted 10 July 2023 / Revised 17 May 2024 / Accepted 5 June 2024 206 0022-3077/24 Copyright© 2024 the American Physiological Society. www.jn.org J Neurophysiol 132: 206–225, 2024. First published June 6, 2024; doi:10.1152/jn.00264.2023 Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. https://orcid.org/0000-0002-4431-9131 https://orcid.org/0000-0002-0885-2182 https://orcid.org/0000-0001-7115-9041 mailto:hoff1695@umn.edu mailto:cmgray.montana@gmail.com https://crossmark.crossref.org/dialog/?doi=10.1152/jn.00264.2023&domain=pdf&date_stamp=2024-6-6 http://www.jn.org https://doi.org/10.1152/jn.00264.2023 have revealed distinct spectral features among the array of cortical areas (30, 31) and spatial gradients of peak frequency within the a and b bands (32). However, both approaches have significant limitations. Intracranial EEG recordings are limited in scope and the spatial resolution of MEG methods is ham- pered by the superposition of signals frommultiple brain areas and the general dominance of the ongoing a rhythm (30). Therefore, a high-resolutionmap of cortical oscillatory activity has remained elusive. Elucidating the spatial and task-dependent organization of cortical oscillatory activity is an especially timely question, given recent discoveries of anatomical and functional gradients underlying the hierarchical organization of cortex (33–36). Striking correlations have been established between gradients of dendritic spine counts (37, 38), cell density (39, 40), neuro- transmitter receptor expression (41, 42), and the anatomical hi- erarchy defined by interareal connections (43–45). Measures of the intrinsic time scale of neural activity across cortical areas also display a hierarchical organization consistent with the underlying anatomy (46–52). However, these functional stud- ies do not include, or even discard, the oscillatory components of the signals under consideration. Consequently, a large gap exists in our understanding of how these anatomical and func- tional gradients are organized in relation to cortical oscillatory activity and its task dependence. To address these questions, we developed and utilized a novel microdrive system to measure intracortical neural ac- tivity from a total of 55 identified cortical areas in two maca- que monkeys that were engaged in a visual short-term memory task (50, 53, 54). We measured both the power spec- trum and the sample entropy (SampEn) (55, 56) of the local field potential (LFP) to identify the predominant frequency bands of oscillatory activity, the relative and absolute ampli- tude of the signals in each band, and the temporal complex- ity of activity in each of the measured areas. These metrics displayed unique spatial gradients across the cortical areas, some of which were correlated with excitatory synaptic spine counts obtained from previous anatomical studies (37, 38). Using the same metrics as features in a classifier, we found that cortical areas can be robustly decoded from each epoch of the task. All features and frequency bands contributed approximately equally to the classification performance. Task-dependent changes in spectral power occurred in all cortical areas and nearly in all frequency bands in both ani- mals, revealing that even a simple cognitive task engages widespread areas of the cortical mantle (57). METHODS Detailed descriptions of the experimental methods for be- havioral training, neural recording, and preprocessing of the data have been described in two previous reports (50, 54). We provide brief descriptions of these methods here. All procedures were performed in accordance with NIH guidelines and approved by the Institutional Animal Care and Use Committee of Montana State University. Behavioral Task Two female macaque monkeys were trained to perform an object-based delayed match to sample task [dMTS; MonkeyLogic software (58, 59)]. A schematic illustrating the time course of events in the task is shown in Fig. 1A. A trial began when the animal acquired and fixated on a small white fixation spot displayed on a gray background [presample pe- riod; fixation window ¼ 3 degrees of visual angle (dva)]. After 500 ms for monkey E, and 800 ms for monkey L the fixation dot was replaced with one of five randomly selected sample images for 500 ms (size: 2.4 � 2.4 dva). The five daily images used were drawn from a larger pool of 92 images. During the sample period, the monkey had to maintain its gaze within a 3� window encompassing the image. Next the sample image was extinguished and replaced with the fixation dot for a vari- able delay period (800–1,200 ms monkey E; 1,000–1,500 ms monkey L). At the end of the delay period, the fixation target was extinguished, and the matching image and a nonmatch- ing image (one of the four other images) appeared at 5� from the center of the screen. For monkey E, the match and non- match images were always placed randomly in opposite hemi- fields along the horizontal axis. Formonkey L, the images were randomly aligned either vertically, horizontally, or 45� diago- nally across from each other (see Fig. 1A). Finally, while the match images were visible, the monkey had to make a sacca- dic eye movement to the matching image and maintain fixa- tion for a brief period (200 ms for monkey E, and 500 ms for monkey L). Correct trials were rewarded with a drop of juice. Electrophysiological Recording Broadband recordings (0.1 Hz–9 kHz, sampled at 32 kHz) were made across the breadth of a cerebral hemisphere in the two monkeys as described in a study by Dotson et al. (54). Briefly, a hemisphere-wide, large-scale microdrive was implanted in two female macaque monkeys with the capa- bility of simultaneously recording from up to 256 independ- ently moveable microelectrodes (interelectrode spacing ¼ 2.5 mm) (Fig. 1B). Neural activity was sampled from an over- lapping set of 62 cortical areas over the course of 6 and 9 mo formonkeys E and L, respectively (Fig. 1C). The broadband sig- nal was bandpass filtered at 1–250 Hz (4th-order Butterworth) and resampled at 1 kHz to obtain the LFP. Anatomical designa- tions [using the nomenclature of Markov et al. (43)] were achieved through reconstruction of each electrode’s track through histological sections (54). Only recordings with neural unit activity that exceeded a minimum mean firing rate of 1 Hz were used for further analysis. Spike waveforms were extracted by detecting local minima in the highpass signal (500 Hz–9 kHz) that exceeded 5 SDs of the noise level (7, 8, 50, 60). To avoid the oversampling of activity, only LFP signals where the electrode position differed from the previous day of recording by>250 lm in depthwere used in this analysis. Areal Grouping Data from adjacent cortical areas with few recordings and similar functional properties were pooled to form small groups of areas. Recordings in visual areas V1 and V2 with short-latency visual responses in the unit activity (SLVR) were analyzed separately (i.e., V1-SLVR, V2-SLVR). The SLVR designation was assigned if the unit activity dis- played a significant change in firing rate within 50–100 ms following the sample presentation. After these pooling procedures, there were a total of 29 and 34 different corti- cal areas or small groups of areas in monkeys E and L, respectively (Table 1). PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org 207 Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org Identification of Spectral Bands For each recording, the power spectral density, averaged across all correct trials in a session, was calculated from 1 to 100 Hz [Field Trip toolbox (61)]. A multi-taper smoothing win- dow of 1 Hz was used. In monkey E, residual 60 Hz line noise was removed with a notch filter. Spectra were calculated over a time window of 2,100 ms encompassing the task period from presample until match onset. Due to the variable length of the delay period, trials that were less than 2,100 ms in length were zero padded to achieve this length. Longer trials were truncated to fit the 2,100 ms time window. The fre- quency resolution of the power spectra was 0.5 Hz. We used an empirical method to designate the frequency band limits for further analysis. We applied a peak detection algorithm (findpeaks.m in Matlab) to the normalized power spectra from 4–100 Hz, in both linear and semi-log coordi- nates, to identify narrow band oscillatory activity. Local Figure 1. A: time course of events in the delayed match-to-sample task. The upper frames illustrate the sequence of images viewed by the monkeys during the task. The dashed circles represent the windows for monitoring eye position. These were not visible to the animal. The sample and match stimuli are symbolized by the let- ters A–E. The white dashed circle in the right plot indicates a correct match. The lower plot illustrates the timeline of task events. The presample duration was 500 ms for monkey E and 800 ms for monkey L. The duration of the delay period was 800–1,200 ms for monkey E and 1,000– 1,500 ms for monkey L. Match stimuli were randomly presented on opposite sides of the horizontal for monkey E and on opposite sides of axes lying at 0, 45, and 90 degrees for monkey L. B: design drawings of the chamber and semi- chronic microdrive systems used for re- cording neuronal activity. The left and right designs were used on monkeys E and L, respectively. Adapted from Dotson et al. (54). C: flatmaps of the recording counts for each cortical area/group [fol- lowing the nomenclature of Markov et al. (43)] inmonkeys E (left) and L (right). Areas with less than three recordings are shaded gray and were not included in the analyses. Areal groupings and recording counts are listed in Table 1. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE 208 J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org maxima<4 Hz were excluded, due to artifactual peaks intro- duced by the multi-taper smoothing filter. To limit small, spurious peaks, a peak prominence criterion (as defined by the Matlab findpeaks.m function) was used. The prominence threshold was set at 1% of the maximum prominence found across all spectra. Values exceeding the prominence thresh- old were used to generate a histogram of peak frequencies across all recordings in each animal. The histograms were fit with a probability density function (PDF) using a mixture of Gaussians (Fitdist.m function in Matlab). Frequency band limits were designated as the local minima in the PDFs. This analysis yielded nearly identical band limits when applied to the linear and semi-log spectra, with deviations in fre- quency bands being less than 1.5 Hz. Results from the two approaches were combined by averaging the local minima designations from the two methods. Classifier Features We first computed the average power spectral density (1- Hz multi-taper smoothing window) across all trials for each of five 400-ms epochs of the task (see Fig. 2B) on each re- cording [Field Trip toolbox (61)]. Using the band limits defined for each monkey, we measured the spectral content (SC) and the peak amplitude (PA) of the spectrum in each fre- quency band and epoch of the task for the entire data set in each animal. We defined SC as the percentage of power within each frequency band relative to the entire spectrum (0–80 Hz) and PA as the largest value of the average spec- trum in each band. To quantify the complexity, or degree of randomness, of the LFP, we calculated the sample entropy [SampEn (m, r, N)] on each task epoch of each trial for each record- ing (55). We used the average value of SampEn across trials for each epoch as a feature. SampEn is an event counting statistic, derived from approximate entropy (ApEn) (62), which quantifies the persistence or regularity of similar patterns in a signal. It is defined as “the negative natural loga- rithm of the conditional probability that two sequences simi- lar for m points remain similar at the next point, where self- matches are not included in calculating the probability” (55, Table 1. Recording counts for all areas and areal groupings in monkeys E and L Monkey E Monkey L Areas Area/Group Flatmaps Decoding Flatmaps Decoding OPRO, 11,12,13,14,32 OrbPFC 42 42 24c, 24d a24 14 40 40 9/46v, 46v vPFC 14 20 20 9/46d, 9, 46d dPFC 40 40 32 32 8B 8B 10 60 60 8L 8L 9 55 55 8M 8M 6 54 54 8r 8r 25 25 44, 45 A, 45B a44/45 5 33 33 F1 F1 90 90 102 102 F2 F2 120 120 F3 F3 3 F5 F5 15 F6, F7 F6/F7 24 24 49 49 1 1 17 41 41 2 2 43 43 31 31 3 3 17 79 79 Insula Ins 16 MB MB 6 7op 7op 16 7m 7m 12 a5, MIP a5/MIP 15 74 74 PIP PIP 22 22 7a 7a 7 30 30 STPc, TPt STPc/TPt 12 15 AIP, VIP AIP/VIP 35 35 7b 7b 23 23 41 41 LIP LIP 37 37 DP DP 30 30 56 56 V6A V6A 44 44 35 35 MT, MST MT/MST 12 59 59 TEOm TEOm 13 V4, V4t V4/t 54 54 26 26 V3 V3 6 12 V2SLVR V2SLVR 6 23 23 V2 V2 51 51 117 117 V1SLVR V1SLVR 27 27 16 V1 V1 169 169 225 225 Recording count 788 526 1,644 1,563 Area/groups 29 11 34 28 Area and areal group names are given in the first and second columns, respectively. Columns labeled as “Flatmaps” show the record- ing counts reported in flatmaps and rank ordered plots of spectral content (SC), peak amplitude (PA), and sample entropy (SampEn). Columns labeled as “Decoding” show the area/groups that are included in the classification analysis. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org 209 Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE 210 J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org 56). It reduces the bias associated with ApEn and is largely in- dependent of the record length of the data: SampEn m; r;Nð Þ ¼ �ln Am rð Þ=Bm rð Þ� � where Bm(r) is the probability that two sequences are similar for m points (possibles) Am(r) is the probability that two sequences are similar for m þ 1 points (matches)m is the length of the sequences being comparedr is the threshold for accepted matches, expressed as a fraction of the standard deviation (r) of the time seriesN is the length of the time series being evaluated. We used values of m ¼ 2 and r ¼ 0.2r. N ¼ 400 data points, which is the length of each task epoch. Each data epoch of length N was z-score normalized before computing SampEn. Together, these analyses resulted in nine features in each of five epochs in monkey E (SC and PA in bands 1–4, and SampEn) and 11 features in each of five epochs in monkey L (SC and PA in bands 1–5, and SampEn). Machine Learning Classifier Each of the features were included in a cubic support vec- tor machine (SVM) classifier. Cortical areas with small num- bers of recordings were grouped with those in adjacent areas with similar functional properties (Table 1). To avoid bias (in both the training and validation phases) toward areas with substantially more recordings, an equal number of record- ings (n ¼ 20) from each area/group were randomly drawn from the larger pool of data. The classifier was trained and validated using k-fold cross validation (k ¼ 5, 4/5th training, 1/5th cross validation). The training-validation process was iteratively repeated 1,000 times, with each iteration using a different set of 20 randomly selected recordings for an area/ group drawn from the data pool (Table 1). The theoretical level of chance classification is 1/no. of areas (1/11 monkey E, 1/28 monkey L). To further assess classification accuracy and avoid biases arising from the use of the theoretical chance level (63), the mean and 99% confidence interval for chance classification was calculated from a permutation distribu- tion. This distribution was created by randomly shuffling areal/group labels before the classification process, then repeating over 1,000 iterations. RESULTS We recorded broadband neuronal activity from a total of 55 cortical areas in two rhesusmacaquemonkeys (33 areas in monkey E and 50 areas in monkey L) while they performed a feature-based delayed match-to-sample (dMTS) task and an interleaved visual fixation task (Fig. 1A). The dMTS task required the monkeys to remember a centrally presented sample image (1 of 5 possible images) for a minimum of 800 ms (800–1,200ms inmonkey E; 1,000–1,500ms inmonkey L), before making a choice between a matching and a nonmatch- ing image. Details of the recording methods, the behavioral task, reconstruction of the recording sites, and analysis of the task dependence of unit activity have been reported previ- ously (50, 54). Here, we focus on the spatial and spectro-tem- poral organization of the LFP and its task dependence. LFP signals were selected for analysis only if there was detectable unit activity recorded on the same electrode, and if the elec- trode position had changed by more than 250 lm from the previous session of recording, yielding a unique recording site. LFP signals that were noisy or had frequent artifacts were discarded, following the methods described by Salazar et al. (7). All analyses were performed on the correct trials of the dMTS task, obtained from 25 sessions inmonkey E and 61 ses- sions inmonkey L (minimum 500 correct trials and>75% cor- rect performance on each session). Simultaneous recordings were made from up to 21 and 37 different cortical areas in monkeys E and L, respectively. The cortical area of the record- ing locations and sample sizes for each animal are shown in Fig. 1C and Table 1. Data from sparsely sampled and adjacent cortical areas with similar functional properties were merged into small groups. This resulted in a total of 788 recordings from 33 areas merged into 29 area/groups in monkey E and a total of 1,644 recordings from 50 areas merged into 34 area/ groups in monkey L (Table 1). For simplicity, we use the term “areas” throughout the text when describing our findings for cortical areas and small groups of areas. Spatial Organization of the LFP At the outset of these experiments, it was apparent that the spectral and temporal properties of the LFP varied sys- tematically across the cortex in a task-dependent manner. Figure 2 shows a representative example of the raw data, and corresponding LFP power spectra, sampled from a subset of the channels on a single session in monkey L. Prefrontal (9/ 46d), frontal eye field (8M, 8B, 8r), and premotor (F7, F2) areas tended to display desynchronized fluctuations of low amplitude interspersed with brief intervals of periodic activ- ity in the range of 6–14 Hz. Primary motor and somatosen- sory areas, F1 and a3, exhibited pronounced oscillations in the 25–35 Hz range, that occurred with lower amplitude in premotor areas. Anterior (a2, a5, 7B) and posterior (7A, V6A) Figure 2. Spectral properties of local field potential (LFP) signals vary markedly across cortex. A: schematic of the recording sites in monkey L. The out- line of the left hemisphere and major cortical sulci, drawn from a photograph, are shown in red. Each circle shows the entry location of an electrode that recorded neural unit activity at some point during the 9 mo of the experiment. Filled circles indicate electrode locations that recorded neural activity in this session (n ¼ 94). Cyan-filled circles mark the electrode locations that correspond to the signals shown in B. B: broadband (0.1 Hz–9 kHz) raw data recorded on a single trial from 14 separate cortical areas. The areal name (left of each trace) follows the nomenclature of Markov et al. (43). The bottom two traces show the vertical and horizontal components of the eye position signal. The black vertical lines, from left to right, mark the onset and offset of the sample stimulus and the onset of the match stimulus, respectively. The colored horizontal bars at the top indicate the time and duration (400 ms) of the Presample (P), Sample (S), Delay1 (D1), Delay2 (D2), and DelayM (DM) epochs, respectively. A saccadic eye movement, occurring �200 ms follow- ing the onset of the match, indicates the monkey’s behavioral choice. The mean sample entropy (SampEn) across trials is plotted to the right of each data trace. C: LFP power spectra (0–80 Hz) for each area indicated in A and B, averaged across all correct trials for each of the five epochs (Presample: black; Sample: green; Delay1: blue; Delay2: red; Delaym: magenta). Clear differences in LFP power and its task dependence are apparent between differ- ent areas of cortex. Black arrows mark notable local peaks or shoulders in the power spectra. Peaks occurring below 4 Hz are due to the absence of a DC component in the filtered signals. AS, arcuate sulcus; CS, central sulcus; IPS, intraparietal sulcus; LS, lunate sulcus; PS, principal sulcus. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org 211 Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org parietal areas exhibited higher amplitude fluctuations with salient oscillations in the 6–14 Hz range. Primary visual cor- tex (V1) was dominated by high amplitude fluctuations at low frequencies. Each of these aspects were clearly apparent in the corresponding power spectra (Fig. 2C), which were computed from five separate 400-ms epochs of the task [pre- sample (P), sample (S), early delay (D1), late delay (D2), match locked delay (DM), Fig. 2B]. There were notable peaks in the spectra that often varied between task epochs. Some spectra showed a sharp singular peak centered around 10 Hz (Fig. 2C, areas 8B, 7B, a2, a5, 7A, V6A), whereas others dis- playedmultiple peaks around 10Hz and 30Hz (Fig. 2C, areas F2, F1, a3), and in some cases small peaks near 65 Hz (Fig. 2C, areas F1, a3). Smaller peaks and shoulders were also apparent near 20 Hz (Fig. 2C, areas 9/46d, F7, 8M, 8r). Other record- ings, most notably in early visual cortex, displayed a steep fall off in power, with very low amplitudes at higher frequen- cies (Fig. 2C, area V1). To determine the appropriate frequency bands for subse- quent analysis, we identified local peaks in the average power spectra, in both linear and semi-log coordinates, computed across all correct trials over the full trial length for all record- ings in both animals (Supplemental Figs. S1 and S2). This analysis yielded a histogram of peak frequencies, and their corresponding areas of origin, spanning the full data set in each monkey (Fig. 3). The histograms derived from the linear and semi-log spectra were nearly indistinguishable and had no effect on the selection of frequency bands (Supplemental Fig. S3). Frequencies below 4 Hz were excluded to avoid the detection of spurious peaks due to filtering. We fit each histo- gram with a probability density function (PDF) and found the local minima in the distributions. This resulted in four fre- quency bands in monkey E (0–8 Hz, 8–21 Hz, 21–32 Hz, 32–80 Hz) and five bands in monkey L (0–6 Hz, 6–14 Hz, 14–26 Hz, 26–42 Hz, 42–80 Hz). These bands did not change after removing the data from all areas that were not present in both animals (V1SL in monkey E and OrbPFC, 8r, Ins, AIP/VIP, LIP inmonkey L). These data revealed a notable difference in the frequency distribution of narrow-band oscillations between the two monkeys. Low-frequency spectral peaks (band 1) were sparsely distributed in bothmonkeys. Spectral peaks in band 2, centered at 14 Hz in monkey E and 10 Hz in monkey L, were widely distributed and present in nearly all areas of cor- tex. Peak frequencies in band 3 were centered at 25 Hz in monkey E and 21 Hz in monkey L and were frontally distrib- uted. There were more obvious differences between the ani- mals at higher frequencies. Spectral peaks in band 4 were very sparse in monkey E, whereas monkey L showed pro- nounced narrow-band components in two bands (band 4, band 5) that were concentrated in somatomotor regions and some areas of the prefrontal and parietal cortices. The spec- tral peaks in band 5 of monkey L were notable in that, with one exception, they always co-occurred with and had twice the frequency of the peaks in band 4. This suggested the presence of a higher frequency harmonic known to result from the nonsinusoidal nature of oscillations in LFP, EEG, and MEG signals (64–68). Detailed analysis of the band 5 component, in both the frequency and time domains (see Frequency Harmonic Analysis in Supplemental Materials), confirmed this conjecture (Supplemental Figs. S4, S5, S6, and S7). Some high-frequency components were also sparsely present in early visual cortex in both animals, con- sistent with the occurrence of induced c-band activity in response to the sample stimuli (69). Many areas displayed a unimodal distribution, whereas other areas showed bimodal Figure 3. Spectral peaks in the local field potential (LFP) occur in distinct frequency bands that vary between animals. Distributions of peak frequencies obtained from semi-log power spectra on all channels and sessions inmonkeys E (A) and L (B). The top plots in A and B show the cumulative histograms of all peak frequencies (4–80 Hz) that exceeded the peak prominence threshold (see Supplemental Figs. S1 and S2). The continuous red lines show the probability density function (PDF) computed with a mixture of Gaussians fit to each distribution. The local minima in each PDF define the boundaries between selected frequency bands for each animal (dashed vertical lines). Four bands (B1–B4) were chosen for monkey E and five bands (B1–B5) were chosen for monkey L. The bottom plots show the normalized counts of peak frequencies for each cortical area/group. This revealed a rough spatial or- ganization of peak frequencies across the sampled cortical areas. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE 212 J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org or even multimodal distributions of spectral peaks. This was most evident in monkey L. These results reveal distinct dif- ferences between animals in the frequency composition of narrow band oscillations in the cortical LFP. To determine if cortical areas can be distinguished by the spectral properties of their LFP, we extracted two values from the spectra computed on each epoch: the percentage of power in each frequency band relative to the entire spectrum (0–80Hz), which we refer to as the spectral content (SC), and the peak amplitude (PA) of the power in each band. The PA was distinct from SC because it reflected the absolute ampli- tude of the signals, which varied widely across cortical areas, and was not bounded within a range of 0–100% (Fig. 2C). To assess the spatial organization of spectral power across the cortex, we plotted the mean value of SC in the presample epoch (across sessions) on cortical flatmaps (adapted from Ref. 43) for each frequency band (Fig. 4). The corresponding rank-ordered box plot of SC (sorted on the median value) is displayed to the right of each flatmap to visualize the distri- bution of values across cortical areas. These maps revealed striking spatial gradients of SC that differ between frequency bands and display similarities as well as differences between the two animals. In both monkeys, there was a clear poste- rior to anterior gradient of SC in band 1, with the greatest concentration of power occurring in visual cortices. Band 2 displayed an anterior shift in the distribution with relative power radiating out from more central/parietal regions. In the higher frequency bands [3 and 4 in monkey E (Fig. 4A), 3–5 in monkey L (Fig. 4B)] there was an anterior shift in the distribution of SC with increasing frequency. A notable difference between animals was also present in the higher frequency bands. Monkey L displayed a distinct focal distri- bution of high amplitude (band 4) in somatomotor areas (3, F1, and F2) with a declining gradient into premotor, prefron- tal, and anterior parietal areas that was not present in mon- key E. Similar plots of PA are shown for bothmonkeys in Fig. 5. These data reveal pronounced amplitude differences across areas within each band as well as striking differences in the LFP amplitude between different frequency bands. The spatial gradients are also apparent, but more variable than those shown by the normalizedmeasure of SC in Fig. 4. The LFP also displayed marked variation in temporal structure across different areas of cortex (Fig. 2B). To quan- tify this, we computed the sample entropy (SampEn) of each LFP time series (55, 56) (see METHODS). SampEn is an event counting statistic, derived from approximate entropy (ApEn) (62), which quantifies the persistence or regularity of similar patterns in a signal. It reflects the degree of a signal’s ran- domness or complexity. A low value of SampEn indicates more self-similarity in a time series of LFP data. For each channel on each session, we calculated the average SampEn across trials on each epoch of the task. Examples of the mean SampEn values, averaged across epochs, are plotted to the right of the data traces in Fig. 2B. SampEn was lower in posterior occipital and parietal areas and increased in soma- tomotor and prefrontal areas. This effect was consistent with the data as a whole. Flatmaps and corresponding rank-or- dered box plots of SampEn (presample epoch) for the full data set are shown for both monkeys in Fig. 6. This revealed a clear spatial gradient of SampEn from occipital to prefron- tal regions of the cortex. The organization of the LFP gradients suggested they may be related to the well-documented variation of synaptic spine counts on the basal dendrites of layer 3 pyramidal neu- rons (37), which exhibit a striking correlation to anatomical hierarchy (36, 44, 45). To test this conjecture, we computed the correlation between the published values of mean spine counts (38, 70–75) and the median values of SC, SampEn, and PA in each band in eachmonkey for all the cortical areas in which these data were available (Table 2). The distribution of average dendritic spine counts from a closely overlapping set of 14 areas in monkey E and 18 areas in monkey L were correlated with the spectral content and peak amplitude in band 1 (SC1, PA1) and SampEn in bothmonkeys. Scatter plots and corresponding correlation coefficients (computed from all values in both monkeys) for these parameters are shown in Fig. 7. The low-frequency spectral components (SC1, PA1) were negatively correlated with mean spine count whereas SampEn displayed a positive correlation. Significant correla- tions were also present in higher frequency bands inmonkey L (SC2, SC3, SC5, and PA2) but not inmonkey E (Table 2). Decoding Cortical Areas Having identified a set of features that reveal spatial gra- dients of the spectral and temporal properties of the LFP, we sought to determine if these features could be used to clas- sify cortical areas. We used a support vector machine (SVM) classifier algorithm, using a K-fold cross validation scheme (K ¼ 5), and applied it separately to the full data set from each monkey, on each epoch of the task (see METHODS). Data were included in the analysis if each area contained a mini- mum of 20 recordings (resulting in 11 areas formonkey E and 28 areas for monkey L; Table 1). The analysis included nine features for monkey E (SC and PA in 4 frequency bands, and SampEn) and 11 features for monkey L (SC and PA in 5 fre- quency bands, and SampEn). The results of the classification analysis are shown in Fig. 8. The confusion matrices (Fig. 8, A and B), and corresponding distributions of validation accuracies (VA) (Fig. 8, C and D), are shown for the presam- ple epoch inmonkey E (Fig. 8, A and C) andmonkey L (Fig. 8, B and D). The median values ranged from 35% to 85% in monkey E (Fig. 8, A and C) and 20% to 75% in monkey L (Fig. 8, B and D) and all areas exceeded the 99% confidence limit, computed from the permutation test (red lines in Fig. 8, C and D, see METHODS). As suggested by the gradients of SC, PA, and SampEn, classification errors tended to lie near the diagonal, indicating similarity in the spectral and temporal properties of the LFP among nearby areas within a cortical region. To determine if VA was task-dependent, we compared the distributions in each area across the five epochs of the task. Figure 8, E and F shows the median VA in each area as a function of task epoch in monkeys E and L, respectively. These are the median values along the diagonals of the con- fusion matrices computed on each epoch. The correspond- ing distributions of VA are shown in Fig. 9. Task-dependent changes in VA were apparent in every area in both monkeys. In some areas VA increased during the delay period (e.g., DP and V2 in monkey E; 7b and V6A in monkey L), other areas displayed the opposite pattern (a2 in monkey E; a1 and a2 in monkey L), and many others displayed differences between PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org 213 Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org Figure 4. Spectral content displays anatomical gradients that differ across frequency bands in both monkeys (A: monkey E; B: monkey L). The left col- umns in A and B show cortical flatmaps of the mean spectral content in each area/group, across all sessions, during the presample epoch for each fre- quency band. Areal boundaries and nomenclature follow that of Markov et al. (43). Visual area V1 with short-latency visual responses in the unit activity (V1-SLVR) and visual area V2 with short-latency visual responses in the unit activity (V2-SLVR) refer to the subset of recordings in areas V1 and V2, respectively, that displayed short latency responses of spiking activity to one or more of the sample stimuli presented on each session [Dotson et al. (50)]. The right columns in A and B show the corresponding distributions of spectral content across all sessions, ranked by the median values. The circle within each box shows the median, the box displays the interquartile range, the whiskers show the 5th and 95th percentiles, and the open circles show outliers. Areal labels are displayed at the bottom and top of each plot. The data are color-coded by cortical region (prefrontal: red; premotor/motor: yel- low; somatosensory: green; temporal: cyan; parietal: blue; visual: magenta). PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE 214 J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org epochs with no discernible pattern across areas. In fact, ev- ery area examined in each animal showed a significant dif- ference in VA across epochs (Kruskal–Wallis test, P << 10�5, FDR corrected), suggesting widespread changes in the spec- tral and temporal properties of the LFP during the task that span all the cortical areas measured. We performed several additional analyses to determine which features in the classifier were responsible for success- ful classification (see Feature Importance in Supplemental Materials). We first assessed the change in VA that occurs when the values of each feature are separately randomized. This led to a �5% decrease in the mean VA across areas for Figure 5. Flatmaps and rank-ordered box plots of the peak amplitude (PA) obtained from the power spectra in each frequency band inmonkey E (A) and monkey L (B) during the presample epoch of the task. Plotting conventions are the same as Fig. 4. Because of outliers and nonlinearities in the rank order plots, the flatmaps are scaled as a percentage of a threshold value shown by the blue line in each rank ordered plot. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org 215 Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org each feature in both monkeys as compared with the base- line (Supplemental Fig. S8). We ran two additional analyses, after excluding SampEn as a feature. We assessed VA inde- pendently for SC and PA using the corresponding features in all frequency bands and we assessed the effect on VA of removing both features in each frequency band separately (Supplemental Tables S1 and S2). In both analyses, we found widespread, and often weak, effects that occurred in nearly all areas. We conclude that each feature makes a small contribution to validation accuracy and that no fre- quency band was substantially more informative than another. We also sought to determine which features of the spec- tra account for classification errors between areas (see Areal Misclassification in Supplemental Materials). We found an inverse relation between VA and the variance of SC and PA in low frequency bands (SC1, SC2, PA1, PA2 in monkey E; SC1, SC2, PA1 in monkey L) (Supplemental Table S3) and a greater incidence of classification errors among nearby areas (Supplemental Fig. S9). Thus, both high var- iance in spectral features as well as similarity in feature values between nearby areas reduced validation accuracy. The latter result is consistent with our finding of spatial gradients in the spectral and temporal features of the LFP. Task-Dependent Changes in Power Although the classification analysis reveals widespread task-dependent changes in the cortical LFP, it does not pro- vide a direct measure of the incidence, magnitude, or sign of those changes for each cortical area. We therefore performed a separate analysis of the change in spectral power across task epochs and frequency bands for each recording in the data set used for the classification analysis (11 areas in mon- key E, 28 areas in monkey L). To screen out signals with low amplitudes, the mean spectral content in each band and epoch had to exceed 5%. For each trial in a session, we calcu- lated the mean power within each band and each epoch. We compared the distribution of values across trials in the pre- sample epoch to the distributions in each of the other four epochs using the Wilcoxon signed-rank test (P < 0.01, FDR corrected). This was repeated for every recording in an area resulting in a distribution of significant changes in power (both increases and decreases) occurring in each epoch and frequency band relative to the presample period (Fig. 10). Example results are shown in Fig. 10A [area 8B (band 2) and area F2 (band 4) inmonkey L]. The mean of each distribution (black filled circle) and the incidence of significant differen- ces (black horizontal line) are color coded and replotted in the lower two plots (see color scale in Fig. 10, C and D). In these examples, there was a mixture of power increases and decreases in band 2 of area 8B that vary with task epoch, while the power in band 4 of area F2 shows a sustained and increasing suppression throughout the task epochs (note that the incidence of significant suppression in area F2 exceeds 90% in each epoch). The overall incidence of changes in power, across all areas, epochs, and frequency bands, is shown for both monkeys in Fig. 10B (left plot: monkey E; right plot: monkey L). Significant changes in LFP power, relative to the presample period, occurred in 60% and 50% of the epochs in monkeys E and L, respec- tively. Decreases in power occurred nearly twice as often as increases in both monkeys. To visualize the overall spatial pattern of these changes, we plotted themean change in power and the incidence of occur- rence in all epochs and frequency bands for all areas in Figure 6. Sample entropy shows clear anatomical gradients. Variation of sample entropy (SampEn) across the cortex for both monkeys during the presample epoch. Cortical flatmaps of median SampEn (left) and corre- sponding distributions of SampEn, ranked by the median, for monkeys E (A) and L (B). Plotting conventions are the same as in Fig. 3. Table 2. Correlation coefficients, and corresponding P values, between the median value of each spectral parameter in each frequency band and mean spine count of the basal dendrites of layer 3 pyramidal neurons in an overlapping set of 14 areas in monkey E and 18 areas in monkey L Monkey E SC1 SC2 SC3 SC4 SampEn PA1 PA2 PA3 PA4 Correlation �0.55 0.36 0.44 0.51 0.54 �0.59 �0.27 �0.1 �0.06 P value <0.05 <0.21 <0.12 <0.06 <0.05 <0.03 <0.35 <0.74 <0.85 Monkey L SC1 SC2 SC3 SC4 SC5 SampEn PA1 PA2 PA3 PA4 PA5 Correlation �0.64 0.6 0.73 0.21 0.51 0.68 �0.71 �0.62 �0.39 �0.04 �0.1 P value <0.005 <0.009 <0.0007 <0.4 <0.03 <0.002 <0.0009 <0.006 <0.12 <0.89 <0.69 Adapted from Elston and Rosa (70–72); Elston and Rockland (73); Elston et al. (38, 74, 75). Spectral parameters are derived from the presample epoch of the task. Numerical values 1–5 indicate the separate frequency bands for each monkey. Significant values are high- lighted in boldface. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE 216 J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org monkey E (Fig. 10, C and D) and monkey L (Fig. 10, E and F). Although the results differed between the two animals, several common findings were apparent in these plots. First, significant task-dependent increases and decreases in power occurred in at least one frequency band in every epoch of all areas in both monkeys, demonstrating wide- spread cortical involvement in the task. Second, the mag- nitude and incidence of the changes varied widely across areas and frequency bands. The spatial organization of the changes was most apparent in the incidence maps of Fig. 10, D and F. In band 1, prefrontal areas showed a task-de- pendent enhancement, particularly in monkey L, whereas suppression was common and robust in central, parietal, and occipital regions in both monkeys. In band 2, a com- plex combination of increases and decreases in power occurred across the cortical areas, making the spatial orga- nization difficult to discern. Notably area V6A displayed a ramp-like enhancement in band 2 in both monkeys, whereas area 8L showed a progressive suppression in mon- key L. In band 3, the most apparent pattern was a combina- tion of suppression in somatomotor, premotor, and prefrontal areas, and enhancement in parietal and extras- triate areas, particularly area V6A in monkey L. In band 4 ofmonkey L, which was unique to this animal (see Fig. 3), a pronounced and ramp-like suppression of both the inci- dence and magnitude occurred throughout a broad region of cortex that included anterior parietal, somatomotor, premotor, and prefrontal areas. At the higher frequencies (band 4 in monkey E, band 5 in monkey L), the pattern of task-dependent changes differed between the two ani- mals. In monkey E, there was a distinct enhancement in response to the sample stimuli in early visual cortex, most notably in foveal and perifoveal V1 (V1-SLVR). In monkey L, the changes in band 5 showed a close spatial correspon- dence to the changes in band 4, but the responses were a mixture of increases and decreases of activity. Together, these results demonstrate widespread, task-dependent changes in the cortical LFP that span all the areas we recorded from. Moreover, the spatial organization of these changes varies with frequency and task epoch. DISCUSSION We made simultaneous measurements of intracortical neural activity in two monkeys performing a visual short- term memory task (50, 54) to investigate the spatial organi- zation and task-dependence of the LFP across a significant fraction of the anatomically identified cortical areas in non- human primates (43). Analysis of the peak frequencies in the LFP power spectra revealed multiple narrow frequency bands in both animals. These distributions overlapped, but also differed in some key respects, demonstrating that the spectral composition of the cortical LFP can vary signifi- cantly between subjects and does not always fall neatly into classically defined canonical frequency bands. These results, along with differences in the recording locations and sample sizes, required us to confine the subsequent analyses to each monkey individually. We found that spatial maps and rank- ordered plots of two spectral parameters (SC, PA) revealed spatial gradients and large amplitude differences across the cortical map that differed markedly between frequency Figure 7. Scatter plots of the median values of spectral content in band 1 (SC1), sample entropy (SampEn) and peak amplitude in band 1 (PA1) vs. mean dendritic spine count on the basal dendrites of layer 3 pyramidal neurons for an overlapping set of 14 and 18 cortical areas in monkey E (blue) and monkey L (red). The correlation coefficients were calculated on the combined data from both monkeys. Spine counts along the x-axis are the same in all three plots and areal labels are shown above the x-axis in the top plot. Data for the areas labeled in black text were available for both monkeys while those labeled in red were available for monkey L only. Spine count data were taken from the reports of Elston and col- leagues [Elston and Rosa (70–72); Elston and Rockland (73); Elston et al. (38, 74, 75)]. For some area groups (i.e., MT/MST, V4/t, 5/MIP, STP/TPt) spine data was obtained from just one area (i.e., MT, V4, 5, STP). For other area groups [i.e., orbital frontal cortex (orbFC) and dorsal prefrontal cortex (dPFC)], spine data was obtained from a subset of those areas (i.e., 12, 13, 9, and 46). PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org 217 Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org bands. These gradients were similar in the two monkeys, apart from band 4 inmonkey L that was absent inmonkey E. Separate analysis of the temporal complexity of the LFP, using the measure of sample entropy (SampEn), revealed a similarly striking frontal to occipital gradient across the cort- ical map in both monkeys (Fig. 6). Together, these analyses demonstrate clear areal differences in the spectral and tem- poral properties of the LFP that exhibit multiple, distinct spatial gradients across the cortical map. These gradients are reminiscent of the earliest qualitative reports in humans and monkeys (28, 29, 77) and exhibit some similarities to a recent report in humans (31). We also compared the spectral and temporal metrics of the LFP to the well documented gradients of synaptic spine counts on the basal dendrites of layer 3 pyramidal neurons (38, 70–75), which are closely correlated with areal hierarchy defined by interareal connections (43–45). This analysis revealed significant correlations between the synaptic spine counts and the low-frequency spectral components (SC1, PA1) and SampEn (Fig. 7) in both monkeys. Significant correlations were also present for SC and PA in bands 2 and 3 in monkey L. These findings are particu- larly interesting, given the central role of temporal sum- mation of dendritic synaptic currents in the genesis of the cortical LFP (25, 78, 79). However, these results were limited to spine counts on the basal dendrites of layer 3 pyramidal neurons because they have been exceptionally well studied. Much remains to be learned. We do not understand how the patterns of synaptic current give rise to the different spectral components of the LFP, nor how Figure 8. Results of the decoding analysis. Confusion matrices (A and B) and distributions of validation accuracies (C and D) for the presample epoch in monkey E (A and C) andmonkey L (B and D). In each box plot the red line shows the median, the blue box displays the interquartile range, the whiskers show the 5th and 95th percentiles, and the red asterisks show outliers. The dashed blue line and solid red line in C and D show the mean and 99th per- centile computed from the surrogate distributions. The plots in E and F show the median validation accuracies as a function of task epoch [Presample (P), Sample (S), Delay1 (D1), Delay2 (D2), and DelayM (DM)] for each area inmonkey E andmonkey L, respectively. The upper and lower yellow horizontal lines in the color-scale bar show the 99th percentile computed from the surrogate distributions formonkeys E and L, respectively. VA, validation accura- cies. Cortical brain area abbreviations as per Ref. 43. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE 218 J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org variations in synaptic organization might contribute to the gradients we observe. These findings are a new piece in a larger puzzle requiring further research. Using the measures of spectral power and SampEn, we could reliably classify cortical areas, or small groups of areas, well above the 99% confidence limit derived from surrogate distributions that randomized areal assignment. The validation accuracy varied across epochs of the task in all areas. However, no single feature or frequency band incorporated in the analysis stood out as particularly in- formative, and validation accuracies exhibited a wide range in both monkeys. This suggested that variance in the feature distributions, or similarities in those distributions among nearby areas, could have led to degraded classification performance. We found evidence supporting both of these conjectures. Together, these analyses demonstrate widespread task-dependent changes in the LFP and show that the spectral features of the LFP display a considerable degree of overlap among adjacent areas, indicating that the LFP varies at a re- gional level incorporating multiple areas with related func- tional properties (30). The decoding analysis, however, provided little or no information regarding the specific changes in spectral power that occur in each area during the task. We there- fore calculated the magnitude and incidence of changes in power that occur in each area with respect to each fre- quency band and epoch of the task. This revealed the striking result that every area of the cortex we sampled in both monkeys displayed both significant increases and decreases in LFP power in multiple frequency bands and multiple epochs of the task. Thus, even a simple cognitive task, such as remembering a salient visual object for 1–2 s, evokes changes in power across widespread areas of the neocortex spanning, visual, parietal, somatomotor, Figure 9. Distributions of validation accu- racies (VA%) as a function of task epoch [Presample (P), Sample (S), Delay1 (D1), Delay2 (D2), and DelayM (DM)] for each area in monkey E (A) and monkey L (B). The bottom right plot in A and B shows the cumulative distributions of validation accuracies across all areas. In each box plot the red line shows the median, the blue box displays the interquartile range, the whiskers show the 5th and 95th per- centiles, and the red asterisks show out- liers. The dashed blue line and solid red line in each plot show the mean and 99th percentile computed from the surrogate distributions. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org 219 Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org premotor, and prefrontal areas. This result is consistent with our earlier findings of widespread, task-dependent changes in unit activity in the same data set (50). These findings are also consistent with a functional imaging study in humans demonstrating task-dependent changes in the blood oxygen level dependent (BOLD) signal in widespread regions of the brain during a simple visual attention task (57). Figure 10. Task-dependent changes in local field potential (LFP) power as a function of task epoch, frequency band, and cortical area inmonkeys E and L. A: example results for area 8B (band 2) and area F2 (band 4) inmonkey L. The top plots show the distributions of significant changes in power across recording sites for each task epoch (S, sample; D1, delay1; D2, delay2; DM, delaym) as a percentage change relative to the presample epoch. The mean of each distribution is shown by the black filled circles. The incidence of significant values in each epoch is shown by the black horizontal lines. Recording counts are shown in parentheses. The mean change and the incidence values are color coded and displayed in the lower pair of plots for the two areas [mean change (blue/red), incidence (orange/green), see color scales in C and D]. B: histograms of the change in mean power, relative to the presample epoch, for all task epochs, frequency bands and cortical areas in monkey E (left) and monkey L (right). The ratios in each plot show the inci- dence (%) of significant decreases and increases in power. C–F: summaries of the significant changes in mean LFP power (C and E) and the incidence of those changes (D and F) as a function of task epoch (S, D1, D2, and DM), frequency band and cortical area formonkey E (C and D) andmonkey L (E and F). The number of recordings in each area are given in Table 1. The black boxes in E and F indicate the data shown in the lower two pairs of plots in A. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE 220 J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org Methodological Considerations Our approach represents a significant advance over other methods such as electrocorticography (ECoG) that are lim- ited to surface measurements with lower spatial resolution (80–82). However, the study also had several methodological limitations. First, we were unable to record from most ven- tral areas of the cortex, particularly in the temporal lobe. Second, the cortical areas we recorded from, and the size of the sample measured from each area, differed substantially between monkeys and between areas within each monkey. Part of this was due to an improvement in the methods used in monkey L (54). This sampling problem could have biased our results. For example, we obtained a large sample of recordings in area F2 in monkey L, but none in monkey E (Table 1). We made efforts to mitigate this problem by ana- lyzing the data from each monkey separately and by sub- sampling the data from each area in each iteration of the decoding analysis. Third, although we were able to recover the areal location of each recording, we could not identify the cortical layer at each recording site. Given that the LFP is known to vary in amplitude and frequency across the corti- cal layers (9, 83, 84), the lack of layer information could have introduced significant variance or biased our results. Fourth, variations in arousal level are known to have pronounced influences on the spectral properties of the LFP (85), and these were clearly present in our data, particularly when the animals became drowsy or disinterested in the task (see Fig. 11). We attempted to reduce these effects by restricting our analysis to correct trials during periods of high performance on the task. Finally, the type of cognitive task, and the stim- uli used in the task, are likely to have a significant influence on the properties of the LFP that depend on cortical area. There were three other factors that could have affected our ability to discriminate differences in the LFP between cortical areas. First, because we used monopolar recording methods, volume conducted signals could have blurred the differences between areas. However, multiple lines of evidence indicate that the LFP is highly local (25, 79, 82, 84, 86, 87), suggesting that volume conduction has minimal effects on our data. Second, the reference channel in our recordings was tied to the large titanium chamber implanted on the animals (54). This provided a “quiet” reference potential by integrating sig- nals from widespread areas of the skull. A separate analysis of the spectral coherence between simultaneously recorded LFPs revealed many instances of nonsignificant coherence at all frequencies (data not shown), suggesting that the reference signal is indeed quiet. Third, the 400-ms duration of the epoch-based spectral analysis, the multi-taper filtering of the power spectra, and the AC-coupling of our recording system, limited our ability to evaluate signals less than 4Hz. Finally, a further analytical issue concerns the spectral and temporal metrics we chose for characterizing the LFP. The SC and PA metrics provide overlapping and somewhat redundant information. However, they also made separate and useful contributions to the decoding analysis. In nearly all cortical areas, one feature performed significantly better than the other (see Supplemental Tables S1 and S2). This can occur, for example, when there is a big difference between areas in absolute power (PA), whereas the differences in rela- tive power (SC) remain small. A related issue is present with SampEn, where the value of the metric likely reflects the rel- ative power of low and high frequencies. In spite of this apparent redundancy, SampEn was informative in classify- ing cortical areas (see Supplemental Fig. S8). This metric also revealed a clear occipito-frontal gradient in both animals de- spite significant differences between animals in the distribu- tion of peak frequencies shown in Fig. 3 and notable differences in the distribution of peak amplitudes shown in Fig. 5. Such a result would not be expected if this metric only reflected the relative magnitudes of low- and high-frequency power. Frequency Differences Outside of the low frequencies in band 1, we found several notable differences in the distribution of spectral peaks in the two monkeys. Signals with spectral peaks in band 2 were widely distributed across the cortical areas in both monkeys but the frequencies were centered at 14 Hz in monkey E and 10 Hz in monkey L. Signals with spectral peaks in band 3 were frontally distributed in both animals and centered at �25 Hz in monkey E and �20 Hz in monkey L. The narrow range of spectral peaks in bands 4 and 5 of monkey L appeared to be completely absent in monkey E. Part of these differences may stem from the sampling problem discussed earlier. But perhaps the simplest explanation is that the occurrence and frequency distribution of narrow band oscil- latory activity can differ widely between individuals and do not necessarily fall into the classically defined canonical fre- quency bands. This conclusion is supported by studies in Figure 11. Plots of a 3-s segment of broadband raw data recorded during a period of rest with the room lights turned off in monkey L. The data is shown for the same channels on the same session as the plots in Fig. 2. A rapid-onset sleep spindle occurs halfway into the segment with high am- plitude in prefrontal, premotor, and parietal areas. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org 221 Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. http://www.jn.org both monkeys and humans (13, 88–90), where the frequen- cies of narrow band oscillatory activity can differ signifi- cantly between individuals. Regarding the higher frequency bands (band 4 in monkey E and bands 4 and 5 of monkey L), several results from this study were particularly noteworthy. First, the spectral peaks in band 5 ofmonkey L (Figs. 2 and 3) co-occurred at twice the frequency of the high-amplitude narrow-band oscillations in band 4, suggesting a higher harmonic that is unlikely to reflect a distinct spectral component. Our supplemental analysis (Supplemental Figs. S4, S5, S6, and S7) confirmed this conjecture. Second, visually evoked c-band (30–80 Hz) oscillations, characteristic of primary and early extrastriate areas of the visual cortex (69, 80, 86, 91, 92), were notably sparse and low amplitude in our data. This type of activity is known to depend on the use of appropriate visual stimuli that are tailored to the receptive field properties of the corti- cal neurons in those areas. Although we did observe clear instances of visually evoked c-band oscillations in V1 and V2, particularly in monkey E, this was largely by chance, since we made no attempt to measure cellular receptive fields and adjust the sample stimuli to optimally activate the neurons at those recording sites. Third, we found limited evidence of elevated activity at the higher frequencies in either monkey during the delay period of the task (Fig. 10). In contrast, the oscillatory activity in this frequency range was nearly uni- versally suppressed during the memory delay. This finding differs from several studies reporting elevated c-band activ- ity during the delay period of short-term memory tasks (10, 93–97), and argues against a role of c-band activity as a gen- eral information carrier in the neocortex (98, 99). Implications for Cortical Processing Our findings also raise important questions regarding the concept of a hierarchy of intrinsic timescales across the cor- tex. The prevailing view, based on functional imaging and ECoG studies in humans (46, 47, 51) and single-unit studies in monkeys (49, 50, 52, 100), indicates that the intrinsic timescale, and temporal receptive window (47), of cortical areas gradually increase across the cortical hierarchy from sensory to association areas. Correspondence between the functional imaging and single-unit measures has also been recently implied in monkeys (101). Our findings, however, suggest an additional perspective. We findmultiple, overlap- ping gradients of LFP power across cortical areas that display striking differences with respect to frequency (Figs. 4 and 5). Occipital-to-prefrontal gradients at low frequencies (band 1) transition to prefrontal-to-occipital gradients at higher fre- quencies (bands 3–5), whereas intermediate frequencies (band 2) display a unique gradient radiating from parietal and sensory-motor areas. If we consider 1/frequency as a simple measure of time scale, these gradients differ from and may be superimposed upon the hierarchical gradients established in previous studies. These findings may also account for the diversity of intrinsic time scales described in recent single-unit studies (52, 100). Our finding of wide- spread task dependence of LFP power further suggests that these gradients dynamically changewith respect to cognitive task, attentional demands, and the types of sensory stimuli andmotor actions (51). Finally, our findings have important implications for the enduring view of cortical organization and function pro- posed by Mountcastle (102). In that classical monograph, Mountcastle proposed two fundamental tenets of cortical organization: The first of these, the unit module, synony- mous with the cortical column, was proposed as a canonical circuit in which “. . . the processing function of neocortical modules is qualitatively similar in all neocortical regions . . . without the appearance of qualitatively different modes of intrinsic organization.” The striking differences we and many others find in the spectral and temporal properties of the LFP across cortical areas do in fact reveal “qualitatively distinct modes of intrinsic dynamics.” When these results are combined with the well documented anatomical and functional gradients across the cortex (33–39, 42, 46–49, 103), a strictly canonical columnar model of neocortex (104) appears untenable. The second tenet was the concept of the distributed system, whereby “. . . complex function con- trolled or executed by the system is not localized to any one of its parts. The function is a property of the dynamic activ- ity within the system: it resides in the system as such.” Our finding of widespread task-dependent changes in spectral power that occur even for a simple cognitive task (57) rein- forces the view of dynamic distributed processing as an in- tegral feature of the central nervous system (105). DATA AVAILABILITY Data will be made available upon reasonable request. SUPPLEMENTAL DATA Supplementary Figs. S1–S9 and Supplemental Tables S1–S3: https://doi.org/10.6084/m9.figshare.25122929. ACKNOWLEDGMENTS We are grateful to Dr. Chris O’Rourke for her excellent veteri- nary care of the animals and assistance in this study. We thank Susan Krueger for her help in the behavioral training and daily care of the animals. We are grateful to Drs. Martin Vinck, Joachim Gross, and Pedro Maldonado for their helpful comments on earlier versions of the manuscript. We also thank Neuralynx (Neuralynx, Inc., Bozeman MT, USA) for providing the data acquisition system. Present addresses: S. J. Hoffman, Dept. of Biomedical Engineering, University of Minnesota, Minneapolis, MN 55455, United States; N. M. Dotson, Salk Institute for Biological Studies, La Jolla, CA 92037, United States. GRANTS This work was supported by grants from National Institute of Neurological Disorders and Stroke (NINDS) under Grants R01 NS059312 and U19 NS107609, National Institute of Mental Health (NIMH) under Grant R01 MH081162, the McKnight Foundation Memory and Cognitive Disorders Award, an EPSCoR RII Track-2 award from the National Science Foundation (to C.M.G. and S.J.H.), and a T32 Neuroimaging training fellowship from NIH under Grant 1T32EB031512-01 (to S.J.H.). DISCLOSURES C.M.G. is affiliated with Gray Matter Research. None of the other authors has any conflicts of interest, financial or otherwise, to disclose. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE 222 J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. https://doi.org/10.6084/m9.figshare.25122929 http://www.jn.org AUTHOR CONTRIBUTIONS S.J.H., N.M.D., and C.M.G. conceived and designed research; S.J.H., N.M.D., and C.M.G. performed experiments; S.J.H., N.M.D., V.L., and C.M.G. analyzed data; S.J.H., N.M.D., V.L., and C.M.G. interpreted results of experiments; S.J.H., N.M.D., V.L., and C.M.G. prepared figures; S.J.H., N.M.D., and C.M.G. drafted manuscript; S.J.H., N.M.D., and C.M.G. edited and revised manuscript; S.J.H., N.M.D., V.L., and C.M.G. approved final version of manuscript. REFERENCES 1. Bressler SL, Coppola R,Nakamura R. Episodic multiregional cortical coherence at multiple frequencies during visual task performance. Nature 366: 153–156, 1993. doi:10.1038/366153a0. 2. Tallon-Baudry C, Bertrand O, Fischer C. Oscillatory synchrony between human extrastriate areas during visual short-term memory maintenance. J Neurosci 21: RC177, 2001. doi:10.1523/JNEUROSCI.21- 20-j0008.2001. 3. Jensen O, Gelfand J, Kounios J, Lisman JE. Oscillations in the a band (9–12 Hz) increase with memory load during retention in a short-term memory task. Cerebral Cortex 12: 877–882, 2002. doi:10.1093/cercor/12.8.877. 4. Brovelli A, Ding M, Ledberg A, Chen Y, Nakamura R, Bressler SL. Beta oscillations in a large-scale sensorimotor cortical network: directional influences revealed by Granger causality. Proc Natl Acad Sci USA 101: 9849–9854, 2004. doi:10.1073/pnas.0308538101. 5. Siegel M,WardenMR,Miller EK. Phase-dependent neuronal coding of objects in short-term memory. Proc Natl Acad Sci USA 106: 21341–21346, 2009. doi:10.1073/pnas.0908193106. 6. Liebe S, Hoerzer GM, Logothetis NK, Rainer G. Theta coupling between V4 and prefrontal cortex predicts visual short-termmemory performance. Nat Neurosci 15: 456–462, 2012. doi:10.1038/nn.3038. 7. Salazar RF, Dotson NM, Bressler SL, Gray CM. Content specific fronto-parietal synchronization during visual working memory. Science 338: 1097–1100, 2012. doi:10.1126/science.1224000. 8. Dotson NM, Salazar RF, Gray CM. Fronto-parietal correlation dynam- ics reveal interplay between integration and segregation during visual working memory. J Neurosci 34: 13600–13613, 2014. doi:10.1523/ JNEUROSCI.1961-14.2014. 9. Bastos AM, Loonis R, Kornblith S, Lundqvist M, Miller EK. Laminar recordings in frontal cortex suggest distinct layers for maintenance and control of working memory. Proc Natl Acad Sci USA 115: 1117– 1122, 2018. doi:10.1073/pnas.1710323115. 10. Lundqvist M, Rose J, Herman P, Brincat SL, Buschman TJ, Miller EK. Gamma and beta bursts underlie working memory. Neuron 90: 152–164, 2016. doi:10.1016/j.neuron.2016.02.028. 11. Lobier M, Palva JM, Palva S. High-a band synchronization across frontal, parietal and visual cortex mediates behavioral and neuronal effects of visuspatial attention. Neuroimage 165: 222–237, 2018. doi:10.1016/j.neuroimage.2017.10.044. 12. Rezayat E, Dehaqani MA, Clark K, Bahmani Z, Moore T, Noudoost B. Frontotemporal coordination predicts working memory perform- ance and its local neural signatures. Nat Commun 12: 3811–3811, 2021. doi:10.1038/s41467-021-21151-1. 13. Vezoli J, Vinck M, Bosman CA, Bastos AM, Lewis CM, Kennedy H, Fries P. Brain rhythms define distinct interaction networks with dif- ferential dependence on anatomy. Neuron 109: 3862–3878.e5, 2021. doi:10.1016/j.neuron.2021.09.052. 14. SingerW,Gray CM. Visual feature integration and the temporal corre- lation hypothesis. Annu Rev Neurosci 18: 555–586, 1995. doi:10.1146/ annurev.ne.18.030195.003011. 15. Singer W. Neuronal synchrony: a versatile code review for the defi- nition of relations? Neuron 24: 49–65, 1999. doi:10.1016/s0896-6273 (00)80821-1. 16. Miller EK, Lundqvist M, Bastos AM. Working memory 2.0. Neuron 100: 463–475, 2018. doi:10.1016/j.neuron.2018.09.023. 17. Fries P. A mechanism for cognitive dynamics: neuronal communica- tion through neuronal coherence. Trends Cogn Sci 9: 474–480, 2005. doi:10.1016/j.tics.2005.08.011. 18. Fries P. Rhythms for cognition: communication through coherence. Neuron 88: 220–235, 2015. doi:10.1016/j.neuron.2015.09.034. 19. Jensen O, Mazaheri A. Shaping functional architecture by oscilla- tory a activity: gating by inhibition. Front Hum Neurosci 4: 186, 2010. doi:10.3389/fnhum.2010.00186. 20. Hagan MA, Pesaran B. Modulation of inhibitory communication coordinates looking and reaching. Nature 604: 708–713, 2022. doi:10.1038/s41586-022-04631-2. 21. Singer W. Consciousness and the binding problem. Ann N Y Acad Sci 929: 123–146, 2001. doi:10.1111/j.1749-6632.2001.tb05712.x. 22. Gray CM. Synchronous oscillations in neuronal systems: mecha- nisms and functions. J Comput Neurosci 1: 11–38, 1994. doi:10.1007/ BF00962716. 23. Buzsáki G, Draguhn A. Neuronal oscillations in cortical networks. Science 304: 1926–1929, 2004. doi:10.1126/science.1099745. 24. Wang XJ. Neurophysiological and computational principles of corti- cal rhythms in cognition. Physiol Rev 90: 1195–1268, 2010. doi:10.1152/physrev.00035.2008. 25. Pesaran B, Vinck M, Einevoll GT, Sirota A, Fries P, Siegel M, Truccolo W, Schroeder CE, Srinivasan R. Investigating large-scale brain dynamics using field potential recordings: analysis and inter- pretation. Nat Neurosci 21: 903–919, 2018. doi:10.1038/s41593-018- 0171-8. 26. Berger H. Über das elektroenkephalogramm des menschen. Archiv f Psychiatrie 87: 527–570, 1929. doi:10.1007/BF01797193. 27. Jasper HH, Andrews HL. Human brain rhythms. I. Recording techni- ques and preliminary results. J Gen Psychol 14: 98–126, 1936. doi:10.1080/00221309.1936.9713141. 28. Jasper HH, Andrews HL. Electro-encephalography. III. Normal differ- entiation of occipital and precentral regions in man. Arch NeurPsych 39: 96–115, 1938. doi:10.1001/archneurpsyc.1938.02270010106010. 29. Jasper HH, Penfield W. Electrocorticograms in man: effect of volun- tary movement upon the electrical activity of the precentral gyrus. Arch F Psychiatr U Z Neur 183: 163–174, 1949. doi:10.1007/ BF01062488. 30. Keitel A, Gross J. Individual human brain areas can be identified from their characteristic spectral activation fingerprints. PLoS Biol 14: e1002498, 2016. doi:10.1371/journal.pbio.1002498. 31. Frauscher B, von Ellenrieder N, Zelmann R, Dole�zalová I,Minotti L, Olivier A, Hall J, Hoffmann D, Nguyen DK, Kahane P, Dubeau F, Gotman J. Atlas of the normal intracranial electroencephalogram: neurophysiological awake activity in different cortical areas. Brain 141: 1130–1144, 2018. doi:10.1093/brain/awy035. 32. Mahjoory K, Schoffelen JM, Keitel A, Gross J. The frequency gradi- ent of human resting-state brain oscillations follows cortical hierar- chies. eLife 9: e53715, 2020. doi:10.7554/eLife.53715. 33. Goulas A, Zilles K, Hilgetag CC. Cortical gradients and laminar pro- jections in mammals. Trends Neurosci 41: 775–788, 2018. doi:10.1016/ j.tins.2018.06.003. 34. Huntenburg JM, Bazin PL, Margulies DS. Large-scale gradients in human cortical organization. Trends Cogn Sci 22: 21–31, 2018. doi:10.1016/j.tics.2017.11.002. 35. Hilgetag CC, Beul SF, van Albada SJ, Goulas A. An architectonic type principle integrates macrosopic corico-cortical connections with intrinsic cortical circuits of the primate brain. Netw Neurosci 3: 905–923, 2019. doi:10.1162/netn_a_00100. 36. Wang XJ. Macroscopic gradients of synaptic excitation and inhibi- tion in the neocortex. Nat Rev Neurosci 21: 169–178, 2020. doi:10.1038/s41583-020-0262-x. 37. Elston G. Specialization of the neocortical pyramidal cell during pri- mate evolution. In: Evolution of Nervous Systems, edited by Kaas JH, Preuss TM. Amsterdam: Elsevier, 2007, vol. 4, p. 191–242. 38. Elston GN, Benavides-Piccione R, Elston A, Manger PR, DeFelipe J. Pyramidal cells in prefrontal cortex of primates: marked differen- ces in neuronal structure among species. Front Neuroanat 5: 2, 2011. doi:10.3389/fnana.2011.00002. 39. Collins CE, Airey DC, Young NA, Leitch DB, Kaas JH. Neuron den- sities vary across and within cortical areas in primates. Proc Natl Acad Sci USA 107: 15927–15932, 2010. doi:10.1073/pnas.1010356107. 40. Beul SF, Barbas H, Hilgetag CC. A predictive structural model of the primate connectome. Sci Rep 7: 43176, 2017. doi:10.1038/srep43176. 41. Froudist-Walsh S, Bliss DP, Ding X, Rapan X, Niu M, Knoblauch K, Zilles K, Kennedy H, Palomero-Gallagher N, Wang XJ. A dopamine gradient controls access to distributed working memory in the large- scale monkey cortex. Neuron 109: 3500–3520.e13, 2021. doi:10.1016/ j.neuron.2021.08.024. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org 223 Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. https://doi.org/10.1038/366153a0 https://doi.org/10.1523/JNEUROSCI.21-20-j0008.2001 https://doi.org/10.1523/JNEUROSCI.21-20-j0008.2001 https://doi.org/10.1093/cercor/12.8.877 https://doi.org/10.1073/pnas.0308538101 https://doi.org/10.1073/pnas.0908193106 https://doi.org/10.1038/nn.3038 https://doi.org/10.1126/science.1224000 https://doi.org/10.1523/JNEUROSCI.1961-14.2014 https://doi.org/10.1523/JNEUROSCI.1961-14.2014 https://doi.org/10.1073/pnas.1710323115 https://doi.org/10.1016/j.neuron.2016.02.028 https://doi.org/10.1016/j.neuroimage.2017.10.044 https://doi.org/10.1038/s41467-021-21151-1 https://doi.org/10.1016/j.neuron.2021.09.052 https://doi.org/10.1146/annurev.ne.18.030195.003011 https://doi.org/10.1146/annurev.ne.18.030195.003011 https://doi.org/10.1016/s0896-6273(00)80821-1 https://doi.org/10.1016/s0896-6273(00)80821-1 https://doi.org/10.1016/j.neuron.2018.09.023 https://doi.org/10.1016/j.tics.2005.08.011 https://doi.org/10.1016/j.neuron.2015.09.034 https://doi.org/10.3389/fnhum.2010.00186 https://doi.org/10.1038/s41586-022-04631-2 https://doi.org/10.1111/j.1749-6632.2001.tb05712.x https://doi.org/10.1007/BF00962716 https://doi.org/10.1007/BF00962716 https://doi.org/10.1126/science.1099745 https://doi.org/10.1152/physrev.00035.2008 https://doi.org/10.1038/s41593-018-0171-8 https://doi.org/10.1038/s41593-018-0171-8 https://doi.org/10.1007/BF01797193 https://doi.org/10.1080/00221309.1936.9713141 https://doi.org/10.1001/archneurpsyc.1938.02270010106010 https://doi.org/10.1007/BF01062488 https://doi.org/10.1007/BF01062488 https://doi.org/10.1371/journal.pbio.1002498 https://doi.org/10.1093/brain/awy035 https://doi.org/10.7554/eLife.53715 https://doi.org/10.1016/j.tins.2018.06.003 https://doi.org/10.1016/j.tins.2018.06.003 https://doi.org/10.1016/j.tics.2017.11.002 https://doi.org/10.1162/netn_a_00100 https://doi.org/10.1038/s41583-020-0262-x https://doi.org/10.3389/fnana.2011.00002 https://doi.org/10.1073/pnas.1010356107 https://doi.org/10.1038/srep43176 https://doi.org/10.1016/j.neuron.2021.08.024 https://doi.org/10.1016/j.neuron.2021.08.024 http://www.jn.org 42. Froudist-Walsh S, Xu T, Niu M, Rapan L, Zhao L, Margulies DS, Zilles K, Wang XJ, Palomero-Gallagher N. Gradients of neurotrans- mitter receptor expression in the macaque cortex. Nat Neurosci 26: 1281–1294, 2023. doi:10.1038/s41593-023-01351-2. 43. Markov NT, Ercsey-Ravasz MM, Ribeiro Gomes AR, Lamy C, Magrou L, Vezoli J, Misery P, Falchier A, Quilodran R, Gariel MA, Sallet J, Gamanut R, Huissoud C, Clavagnier S, Giroud P, Sappey- Marinier D, Barone P, Dehay C, Toroczkai Z, Knoblauch K, Van Essen DC, Kennedy H. A weighted and directed interareal connec- tivity matrix for macaque cerebral cortex. Cereb Cortex 24: 17–36, 2014. doi:10.1093/cercor/bhs270. 44. Markov NT, Vezoli J, Chameau P, Falchier A, Quilodran R, Huissoud C, Lamy C, Misery P, Giroud P, Ullman S, Barone P, Dehay C, Knoblauch K, Kennedy H. Anatomy of hierarchy: feedfor- ward and feedback pathways in macaque visual cortex. J Comp Neurol 522: 225–259, 2014. doi:10.1002/cne.23458. 45. Chaudhuri R, Knoblauch K, Gariel MA, Kennedy H, Wang XJ. A large-scale circuit mechanism for hierarchical dynamical processing in the primate cortex. Neuron 88: 419–431, 2015. doi:10.1016/j. neuron.2015.09.008. 46. Hasson U, Yang E, Vallines I, Heeger DJ, Rubin N. A hierarchy of temporal receptive windows in human cortex. J Neurosci 28: 2539– 2550, 2008. doi:10.1523/JNEUROSCI.5487-07.2008. 47. Hasson U, Chen J, Honey CJ. Hierarchical process memory: mem- ory as an integral component of information processing. Trends Cogn Sci 19: 304–313, 2015. doi:10.1016/j.tics.2015.04.006. 48. Honey CJ, Thesen T, Donner TH, Silbert LJ, Carlson CE, Devinsky O, Doyle WK, Rubin N, Heeger DJ, Hasson U. Slow cortical dynam- ics and the accumulation of information over long timescales. Neuron 76: 423–434, 2012 [Erratum in Neuron 76: 668, 2012]. doi:10.1016/j.neuron.2012.08.011. 49. Murray JD, Bernacchia A, Freedman DJ, Romo R, Wallis JD, Cai X, Padoa-Schioppa C, Pasternak T, Seo H, Lee D, Wang X-J. A hierar- chy of intrinsic timescales across primate cortex. Nat Neurosci 17: 1661–1663, 2014. doi:10.1038/nn.3862. 50. Dotson NM, Hoffman SJ, Goodell B, Gray CM. Feature-based visual short-term memory is widely distributed and hierarchically organized. Neuron 99: 215–226.e4, 2018. doi:10.1016/j.neuron.2018.05.026. 51. Gao R, van den Brink RL, Pfeffer T, Voytek B. Neuronal timescales are functionally dynamic and shaped by cortical microarchitecture. eLife 9: e61277, 2020. doi:10.7554/eLife.61277. 52. Spitmaan M, Seo H, Lee D, Soltani A. Multiple timescales of neuro- nal dynamics and integration of task-relevant signals across cortex. Proc Natl Acad Sci USA 117: 22522–22531, 2020. doi:10.1073/ pnas.2005993117. 53. Dotson NM, Salazar RF, Goodell AB, Hoffman SJ, Gray CM. Methods, caveats, and the future of large-scale microelectrode recordings in the non-human primate. Front Syst Neurosci 9: 149, 2015. doi:10.3389/fnsys.2015.00149. 54. Dotson NM, Hoffman SJ, Goodell B, Gray CM. A large-scale semi- chronic microdrive recording system for non-human primates. Neuron 96: 769–782.e2, 2017. doi:10.1016/j.neuron.2017.09.050. 55. Richman JS, Moorman JR. Physiological time-series analysis using approximate entropy and sample entropy. Am J Physiol Heart Circ Physiol 278: H2039–H2049, 2000. doi:10.1152/ajpheart.2000.278.6. H2039. 56. Delgado-Bonal A,Marshak A. Approximate entropy and sample en- tropy: a comprehensive tutorial. Entropy 21: 541, 2019. doi:10.3390/ e21060541. 57. Gonzalez-Castillo J, Saad ZS, Handwerker DA, Inati SJ, Brenowitz N, Bandettini PA. Whole-bran, time-locked activation with simple tasks revealed using massive averaging and model-free analysis. Proc Natl Acad Sci USA 109: 5487–5492, 2012. doi:10.1073/ pnas.1121049109. 58. Asaad WF, Eskandar EN. A flexible software tool for temporally pre- cise behavioral control in matlab. J Neurosci Methods 174: 245–258, 2008. doi:10.1016/j.jneumeth.2008.07.014. 59. Asaad WF, Eskandar EN. Achieving behavioral control with millisec- ond resolution in a high-level programming environment. J Neurosci Methods 173: 235–240, 2008. doi:10.1016/j.jneumeth.2008.06.003. 60. Yen SC, Baker J, Gray CM. Heterogeneity in the responses of adja- cent neurons to natural stimuli in cat striate cortex. J Neurophysiol 97: 1326–1341, 2007. doi:10.1152/jn.00747.2006. 61. Oostenveld R, Fries P, Maris E, Schoffelen JM. FieldTrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput Intell Neurosci 2011: 156869, 2011. doi:10.1155/2011/156869. 62. Pincus SM. Approximate entropy as a measure of system complex- ity. Proc Natl Acad Sci USA 88: 2297–2301, 1991. doi:10.1073/ pnas.88.6.2297. 63. Combrisson E, Jerbi K. Exceeding chance level by chance: the ca- veat of theoretical chance levels in brain signal classification and sta- tistical assessment of decoding accuracy. J Neurosci Methods 250: 126–136, 2015. doi:10.1016/j.jneumeth.2015.01.010. 64. Aru J, Aru J, Priesemann V, Wibral M, Lana L, Pipa G, Singer W, Vicente R. Untangling cross-frequency coupling in neuroscience. Curr Opin Neurobiol 31: 51–61, 2015. doi:10.1016/j.conb.2014.08.002. 65. Lozano-Soldevilla D, Ter Huurne N, Oostenveld R. Neuronal oscil- lations with non-sinusoidal morphology produce spurious phase-to- amplitude coupling and directionality. Front Comput Neurosci 10: 87, 2016. doi:10.3389/fncom.2016.00087. 66. Gerber EM, Sadeh B, Ward A, Knight RT, Deouell LY. Non-sinusoi- dal activity can produce cross-frequency coupling in cortical signals in the absence of functional interactions between neural sources. PLoS One 11: e0167351, 2016. doi:10.1371/journal.pone.0167351. 67. Cole SR, van der Meij R, Peterson EJ, de Hemptinne C, Starrr PA, Voytek BJ. Nonsinusoidal b oscillations reflect cortical pathophysiol- ogy in Parkinson’s disease. J Neurosci 37: 4830–4840, 2017. doi:10.1523/JNEUROSCI.2208-16.2017. 68. Schaworonkow N, Nikulin VV. Spatial neuronal synchronization and the waveform of oscillations: implications for EEG and MEG. PLoS Comput Biol 15: e1007055, 2019. doi:10.1371/journal.pcbi.1007055. 69. Friedman-Hill SR, Maldonado PE, Gray CM. Dynamics of striate cortical activity in the alert macaque. I. Incidence and stimulus-de- pendence of gamma-band neuronal oscillations. Cereb Cortex 10: 1105–1116, 2000. doi:10.1093/cercor/10.11.1105. 70. Elston GN, Rosa MG. The occipitoparietal pathway of the macaque monkey: camparison of pyramidal cell morphology in layer III of func- tionally related cortical visual areas. Cereb Cortex 7: 432–452, 1997. doi:10.1093/cercor/7.5.432. 71. Elston GN, Rosa MG. Morphological variation of layer III pyramidal neurones in the occipitotemporal pathway of the macaque monkey visual cortex. Cereb Cortex 8: 278–294, 1998. doi:10.1093/cercor/ 8.3.278. 72. Elston GN, Rosa MG. Complex dendritic fields of pyramidal cells in the frontal eye field of the macaque monkey: comparison with parie- tal areas 7a and LIP. Neuroreports 9: 127–131, 1998b. doi:10.1097/ 00001756-199801050-00025. 73. Elston GN, Rockland KS. The pyramidal cell of the sensorimotor cor- tex of the macaque monkey: phenotypic variation. Cereb Cortex 12: 1071–1078, 2002. doi:10.1093/cercor/12.10.1071. 74. Elston GN, Tweedale R, RosaMP.Cortical integration in the visual sys- tem of the macaque monkey: largescale morphological differences of pyramidal neurones in the occipital, parietal and temporal lobes. Proc Biol Sci 266: 1367–1374, 1999. doi:10.1098/rspb.1999.0789. 75. Elston GN, Benavides-Piccione R, DeFelipe J. A study of pyramidal cell structure in the cingulate cortex of the macaque monkey with comparative notes on inferotemporal and primary visual cortex. Cereb Cortex 15: 64–73, 2005. doi:10.1093/cercor/bhh109. 76. Dreyfus G, Guyon I. Assessment methods. In: Feature Extraction: Foundations and Applications, edited by Guyon I, Gunn S, Nikravesh M, Zadeh L. Cham, Switzerland: Springer, 2006, p. 65–88. 77. Garvin JS, Amador LV. Electrocorticograms of the cytoarchitectural areas of Macaca Mulatta. J Neurophysiol 12: 425–433, 1949. doi:10.1152/jn.1949.12.6.425. 78. Lind�en H, Tetzlaff T, Potjans TC, Pettersen KH, Gr€un S, Diesmann M, Einevoll GT. Modeling the spatial reach of the LFP. Neuron 72: 859–872, 2011. doi:10.1016/j.neuron.2011.11.006. 79. Einevoll GT, Kayser C, Logothetis NK, Panzeri S. Modeling and analysis of local field potentials for studying the function of cortical circuits. Nat Rev Neurosci 14: 770–785, 2013. doi:10.1038/nrn3599. 80. Bosman CA, Schoffelen JM, Brunet N, Oostenveld R, Bastos AM, Womelsdorf T, Rubehn B, Stieglitz T, De Weerd P, Fries P. Attentional stimulus selection through selective synchronization between monkey visual areas. Neuron 75: 875–888, 2012. doi:10.1016/j.neuron.2012.06.037. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE 224 J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. https://doi.org/10.1038/s41593-023-01351-2 https://doi.org/10.1093/cercor/bhs270 https://doi.org/10.1002/cne.23458 https://doi.org/10.1016/j.neuron.2015.09.008 https://doi.org/10.1016/j.neuron.2015.09.008 https://doi.org/10.1523/JNEUROSCI.5487-07.2008 https://doi.org/10.1016/j.tics.2015.04.006 https://doi.org/10.1016/j.neuron.2012.08.011 https://doi.org/10.1038/nn.3862 https://doi.org/10.1016/j.neuron.2018.05.026 https://doi.org/10.7554/eLife.61277 https://doi.org/10.1073/pnas.2005993117 https://doi.org/10.1073/pnas.2005993117 https://doi.org/10.3389/fnsys.2015.00149 https://doi.org/10.1016/j.neuron.2017.09.050 https://doi.org/10.1152/ajpheart.2000.278.6.H2039 https://doi.org/10.1152/ajpheart.2000.278.6.H2039 https://doi.org/10.3390/e21060541 https://doi.org/10.3390/e21060541 https://doi.org/10.1073/pnas.1121049109 https://doi.org/10.1073/pnas.1121049109 https://doi.org/10.1016/j.jneumeth.2008.07.014 https://doi.org/10.1016/j.jneumeth.2008.06.003 https://doi.org/10.1152/jn.00747.2006 https://doi.org/10.1155/2011/156869 https://doi.org/10.1073/pnas.88.6.2297 https://doi.org/10.1073/pnas.88.6.2297 https://doi.org/10.1016/j.jneumeth.2015.01.010 https://doi.org/10.1016/j.conb.2014.08.002 https://doi.org/10.3389/fncom.2016.00087 https://doi.org/10.1371/journal.pone.0167351 https://doi.org/10.1523/JNEUROSCI.2208-16.2017 https://doi.org/10.1371/journal.pcbi.1007055 https://doi.org/10.1093/cercor/10.11.1105 https://doi.org/10.1093/cercor/7.5.432 https://doi.org/10.1093/cercor/8.3.278 https://doi.org/10.1093/cercor/8.3.278 https://doi.org/10.1097/00001756-199801050-00025 https://doi.org/10.1097/00001756-199801050-00025 https://doi.org/10.1093/cercor/12.10.1071 https://doi.org/10.1098/rspb.1999.0789 https://doi.org/10.1093/cercor/bhh109 https://doi.org/10.1152/jn.1949.12.6.425 https://doi.org/10.1016/j.neuron.2011.11.006 https://doi.org/10.1038/nrn3599 https://doi.org/10.1016/j.neuron.2012.06.037 http://www.jn.org 81. Bastos AM, Vezoli J, Bosman CA, Schoffelen JM, Oostenveld R, Dowdall JR,DeWeerd P, Kennedy H, Fries P.Visual areas exert feed- forward and feedback influences through distinct frequency channels. Neuron 85: 390–401, 2015. doi:10.1016/j.neuron.2014.12.018. 82. Dubey A, Ray S. Cortical electrocorticogram (ECoG) is a local signal. J Neurosci 39: 4299–4311, 2019. doi:10.1523/JNEUROSCI.2917- 18.2019. 83. Murthy VN, Fetz EE. Oscillatory activity in sensorimotor cortex of awake monkeys: synchronization of local field potentials and rela- tion to behavior. J Neurophysiol 76: 3949–3967, 1996. doi:10.1152/ jn.1996.76.6.3949. 84. Spaak E, BonnefondM,Maier A, Leopold DA, Jensen O. Layer-spe- cific entrainment of gamma-band neural activity by the a rhythm in monkey visual cortex. Curr Biol 22: 2313–2318, 2012. doi:10.1016/j. cub.2012.10.020. 85. Magoun HW. Brain mechanisms for wakefulness. Br J Anaesth 33: 183–193, 1961. doi:10.1093/bja/33.4.183. 86. Gray CM, Singer W. Stimulus-specific neuronal oscillations in orien- tation columns of cat visual cortex. Proc Natl Acad Sci USA 86: 1698–1702, 1989. doi:10.1073/pnas.86.5.1698. 87. Katzner S, Nauhaus I, Benucci A, Bonin V, Ringach DL, Carandini M. Local origin of field potentials in visual cortex. Neuron 61: 35–41, 2009. doi:10.1016/j.neuron.2008.11.016. 88. Kilavik BE, Ponce-Alvarez A, Trachel R, Confais J, Takerkart S, Riehle A. Context-related frequency modulations of macaque motor cortical LFP b oscillations. Cereb Cortex 22: 2148–2159, 2012. doi:10.1093/ cercor/bhr299. 89. van Pelt S, Boomsma DI, Fries P. Magnetoencephalography in twins reveals a strong genetic determination of the peak frequency of visually induced c-band synchronization. J Neurosci 32: 3388– 3392, 2012. doi:10.1523/JNEUROSCI.5592-11.2012. 90. Confais J, Malfait N, Brochier T, Riehle A, Kilavik BE. Is there an intrinsic relationship between LFP b oscillation amplitude and firing rate of individual neurons in macaque motor cortex? Cereb Cortex Commun 1: tgaa017, 2020. doi:10.1093/texcom/tgaa017. 91. Fries P, Reynolds JH, Rorie AE, Desimone R. Modulation of oscilla- tory neuronal synchronization by selective visual attention. Science 291: 1560–1563, 2001. doi:10.1126/science.1055465. 92. Bartoli E, Bosking W, Chen Y, Li Y, Sheth SA, Beauchamp MS, Yoshor D, Foster BL. Functionally distinct c range activity revealed by stimulus tuning in human visual cortex. Curr Biol 29: 3345–3358. e7, 2019. doi:10.1016/j.cub.2019.08.004. 93. Pesaran B, Pezaris JS, Sahani M, Mitra PP, Andersen RA. Temporal structure in neuronal activity during working memory in macaque pari- etal cortex.Nat Neurosci 5: 805–811, 2002. doi:10.1038/nn890. 94. Howard MW, Rizzuto DS, Caplan JB, Madsen JR, Lisman J, Aschenbrenner-Scheibe R, Schulze-Bonhage A, Kahana MJ. Gamma oscillations correlate with working memory load in humans. Cereb Cortex 13: 1369–1374, 2003. doi:10.1093/cercor/bhg084. 95. Meltzer JA, Zaveri HP, Goncharova II, Distasio MM, Papademetris X, Spencer SS, Spencer DD, Constable RT. Effects of working mem- ory load on oscillatory power in human intracranial EEG. Cereb Cortex 18: 1843–1855, 2008. doi:10.1093/cercor/bhm213. 96. Roux F, Wibral M, Mohr HM, Singer W, Uhlhaas PJ. Gamma-band activity in human prefrontal cortex codes for the number of relevant items maintained in working memory. J Neurosci 32: 12411–12420, 2012. doi:10.1523/JNEUROSCI.0421-12.2012. 97. Honkanen R, Rouhinen S, Wang SH, Palva JM, Palva S. Gamma oscillations underlie the maintenance of feature-specific information and the contents of visual working memory. Cereb Cortex 25: 3788–3801, 2015. doi:10.1093/cercor/bhu263. 98. Fries P. Neuronal c-band synchronization as a fundamental process in cortical computation. Annu Rev Neurosci 32: 209–224, 2009. doi:10.1146/annurev.neuro.051508.135603. 99. Bas�ar E. A review of c oscillations in healthy subjects and in cogni- tive impairment. Int J Psychophysiol 90: 99–117, 2013. doi:10.1016/j. ijpsycho.2013.07.005. 100. Cavanagh SE, Hunt LT, Kennerley SW. A diversity of intrinsic time- scales underlie neural computations. Front Neural Circuits 14: 615626, 2020. doi:10.3389/fncir.2020.615626. 101. Manea A, Zilverstand A, Ugurbil K, Heilbronner S, Zimmermann J. Intrinsic timescales as an organizational principle of neural process- ing across the whole rhesus macaque brain. eLife 11: e75540, 2022. doi:10.7554/eLife.75540. 102. Mountcastle VB. An organizing principle for cerebral function: the unit module and the distributed system. In: The Neurosciences: Fourth Study Program, edited by Schmitt FO, Worden FG. Cambridge: MIT Press, 1979, p. 21–42. 103. Li S, Wang XJ. Hierarchical timescales in the neocortex: mathemati- cal mechanism and biological insights. Proc Natl Acad Sci USA 119: e2110274119, 2022. doi:10.1073/pnas.2110274119. 104. Hawkins J. A Thousand Brains: A New Theory of Intelligence. New York: Basic Books, 2021. 105. Tognoli E, Kelso JA. The metastable brain. Neuron 81: 35–48, 2014. doi:10.1016/j.neuron.2013.12.022. PRIMATE LFP SPECTRAL TEMPORAL GRADIENTS AND TASK DEPENDENCE J Neurophysiol � doi:10.1152/jn.00264.2023 � www.jn.org 225 Downloaded from journals.physiology.org/journal/jn at Montana State University Library (153.090.170.056) on August 14, 2024. https://doi.org/10.1016/j.neuron.2014.12.018 https://doi.org/10.1523/JNEUROSCI.2917-18.2019 https://doi.org/10.1523/JNEUROSCI.2917-18.2019 https://doi.org/10.1152/jn.1996.76.6.3949 https://doi.org/10.1152/jn.1996.76.6.3949 https://doi.org/10.1016/j.cub.2012.10.020 https://doi.org/10.1016/j.cub.2012.10.020 https://doi.org/10.1093/bja/33.4.183 https://doi.org/10.1073/pnas.86.5.1698 https://doi.org/10.1016/j.neuron.2008.11.016 https://doi.org/10.1093/cercor/bhr299 https://doi.org/10.1093/cercor/bhr299 https://doi.org/10.1523/JNEUROSCI.5592-11.2012 https://doi.org/10.1093/texcom/tgaa017 https://doi.org/10.1126/science.1055465 https://doi.org/10.1016/j.cub.2019.08.004 https://doi.org/10.1038/nn890 https://doi.org/10.1093/cercor/bhg084 https://doi.org/10.1093/cercor/bhm213 https://doi.org/10.1523/JNEUROSCI.0421-12.2012 https://doi.org/10.1093/cercor/bhu263 https://doi.org/10.1146/annurev.neuro.051508.135603 https://doi.org/10.1016/j.ijpsycho.2013.07.005 https://doi.org/10.1016/j.ijpsycho.2013.07.005 https://doi.org/10.3389/fncir.2020.615626 https://doi.org/10.7554/eLife.75540 https://doi.org/10.1073/pnas.2110274119 https://doi.org/10.1016/j.neuron.2013.12.022 http://www.jn.org bkmk_bookmark_1 bkmk_bookmark_2 bkmk_bookmark_3 bkmk_bookmark_4 bkmk_bookmark_5 bkmk_bookmark_6 bkmk_bookmark_7 bkmk_bookmark_8 bkmk_bookmark_9 bkmk_bookmark_10 bkmk_bookmark_11 bkmk_bookmark_12 bkmk_bookmark_13 bkmk_bookmark_14 bkmk_bookmark_15 bkmk_bookmark_16 bkmk_bookmark_17 bkmk_bookmark_18 bkmk_bookmark_19 bkmk_bookmark_AC bkmk_bookmark_20 bkmk_bookmark_21 bkmk_bookmark_22 bkmk_bookmark_23