Two male rhesus monkeys (Macaca mulatta, 5 and 6 kg BW) were used in this study. The animals were on a regulated water schedule. Animal care conformed to the principles outlined in the Guide for Care and Use of Laboratory Animals (National Institutes for Health publication no. 85-23, revised 1985). The animal protocols were approved by the Institutional Review Board.

We used a task (Figure 1) in which two monkeys categorized a sample stimulus to either one of two half-spaces (upper and lower) in a task box, as described in detail in Fortes et al. (2004). Since the task box changed location from trial to trial, the task involved categorization of a relative spatial cue. There were three periods of interest in the task (Figure 1). During the “sample period,” the monkey received all the information needed to categorize the stimulus bar; during the following “pre-response period” the monkey prepared the response which was elicited during one of the ensuing two “response periods.” From an information processing point of view, the sample period is the most demanding and crucial one, for it is during that period that the process of spatial categorization is taking place. Cognitive load decreases after that period and becomes smallest during the response time. To test the hypothesis that fine-grain (1 ms temporal resolution) synchronous and lagged neuronal interactions are involved in information processing, we recorded LFP activity from two cortical sites simultaneously in the posterior parietal cortex, based on the known involvement of area 7a in spatial cognitive processing.

Neuronal activity was recorded extracellularly using two independently movable 16-microelectrode matrix systems (Eckhorn system, Thomas Recording, GMbH, Giessen, Germany, Figure 2B). The recording sites included symmetric areas 7a of the posterior parietal cortex, bilaterally (see Figure 2A). There were six recording sites in any given trial, of which three were in the left and three in the right hemisphere. The raw analog electrical activity was digitized at 40 kHz and stored in a personal computer via two PCI-DAS64/M2/16 high speed analog boards (Measurement Computing, Middleboro, MA). LFPs were extracted from three microelectrodes per matrix using a LFP-filter amplifier (Thomas Recording GmbH, Giessen, Germany) with a low-pass filter at 120 Hz and a high-pass filter at 0.1 Hz. All channels were recorded irrespective of neural activity during the task. The eye position was sampled at 200 Hz using an infrared video eye tracking system (ISCAN Inc., Burlington, MA). Two male rhesus macaque monkeys (weighing 8 and 7 kg) were used. They were prepared for recording using standard aseptic surgical techniques under Isoflurane (1–2%) gas anesthesia. In each animal, recording chambers (7 mm internal diameter) were implanted bilaterally, above a craniotomy overlying area 7a in each of the two hemispheres. Five titanium posts were attached to the skull with titanium screws, and a halo was attached to provide an anchor point to stabilize the head during neural recordings. Analgesia was provided for a period of several days following surgery (Buprenex, 0.05 mg/kg BID, i.m.).

The raw digitized extracellular record (collected at 40 kHz) was resampled (decimated) to 1 kHz, by picking every 40th sample of the original digitized data. The resulting time series were then prewhitened by taking the residuals after applying an [25,1,1] AutoRegressive Integrative Moving Average (ARIMA) model (Box and Jenkins, 1970; Priestley, 1981) (see Figure 3 for details). This model was arrived at after extensive model identification and yielded residuals (innovations) that were practically stationary with respect to the mean and variance, and flat autocorrelations (see Results below). The crosscorrelation function (±25 ms maximum lag) (Figure 4) was computed for all pairs of recorded LFP time series and for each one of the seven periods of the task (Figure 1). The following three measures were extracted from each crosscorrelogram: (a) The crosscorrelation at zero lag (CC⁰), indicating synchronicity, (b) the crosscorrelation with the maximum absolute value (CC^max), and (c) the lag at which CC^max occurred. For the purposes of this analysis, the sign of the lag was ignored, hence its absolute value was used (range: 0–25 ms); hereafter, we use “lag” to mean “absolute value of lag.” The signs of CC⁰ and CC^max were retained and their statistical significance calculated using the assumption that the series are white noise. The crosscorrelations were z-transformed to normalize their distribution (Fisher, 1958):

This is equivalent to

Finally, standard statistical methods were performed to further analyze the data (Snedecor and Cochran, 1989), as mentioned below, using the IBM SPSS Statistics package, version 22.

With respect to zero-lag interactions, there are two plausible sources of synchronized inputs to a cortical area, namely recurrent collaterals of pyramidal cell axons (Stefanis and Jasper, 1964a,b; Brooks and Asanuma, 1965) and thalamocortical inputs (Jones, 2001; Bruno and Sakmann, 2006). The former, and thalamic inputs directed specifically to a given area arising from parvalbumin-immunoreactive neurons (Jones, 2001), could account for local (within-area) input synchronization, whereas widely and multifocally distributed thalamic inputs, arising from calbindin-immunoreactive neurons (Jones, 2001), could subserve long-distance (between-areas) input synchronization. The results of the present study demonstrated that that such inputs can be finely synchronized (at 1 ms temporal resolution). This synchronization was mostly positive and occurred more frequently within area 7a of the same hemisphere, whereas it was mostly negative and occurred less frequently across symmetric 7a sites in the two hemispheres. Nevertheless, in spite of these differences, the strength of LFP synchronization was modulated in a similar fashion by the cognitive load of the task. These findings reinforce the role of area 7a in spatial cognitive processing, as shown in different studies (Merchant et al., 2001, 2003, 2004a,b, 2005), and extend the neuronal mechanisms involved from single cell activity (Chafee et al., 2005; Crowe et al., 2005, 2013; Chafee and Crowe, 2012) to encompass LFP synchronicity.

Our results demonstrated the orderly modulation of lagged neural interactions by cognitive load. Simply, interactions became stronger and more synchronous (at shorter lag) during the sample period, when the bar to be categorized was presented. This effect was observed both for interactions within as well as across hemispheres, even though the magnitude of correlations and lags were much smaller in the latter than in the former case. Altogether, our findings establish the strength and lag of neural interactions as meaningful variables under cognitive control.

Finally, our results provide a novel source of correspondence between neurophysiological and neuroanatomical measurement, namely on callosal conduction delays. Specifically, detailed measurements of white matter variables and fiber characteristics have furnished estimates of callosal conduction delays in monkeys and humans (Caminiti et al., 2013). For the monkey, evidence based on fiber histology provided an estimated mean callosal conduction delay between posterior parietal cortices of 6.42 ± 3.11 ms (mean ± SD) (Caminiti et al., 2013, Table 1) which is statistically indistinguishable from our estimate of 5.65 ± 7.34 ms (Figure 8). To our knowledge, this is the first time that crosscorrelation of LFP time series has been used to yield such an accurate estimate of callosal conduction delays in such a close agreement to estimates derived from purely anatomical measurements. We believe that this was made possible by the proper prewhitening procedure which converted the LFP time series to white noise, thus allowing for a correct estimation of neuronal interactions.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.