All experiments followed institutional and NIH guidelines for the care and use of laboratory animals. Two adult, male capuchin monkeys (Cebus apella
) weighing 3–4 kg served as subjects for this study. Henceforth, these animals are referred to as monkey D and S. Under sterile conditions, each animal was implanted with a scleral search coil for monitoring eye position (Judge et al. 1980
) and a cranial post for head fixation. After a period of visual fixation training, we performed a second surgical procedure, in which a plastic recording chamber was mounted over the visual cortex. Area 17 was accessed by stereotaxic coordinates (Gattass et al. 1987
). A micromanipulator with up to 8 independently movable tetrodes could be attached to the chamber. After a small craniotomy was performed, an incision was made in the dura mater and the guide tube array was positioned over the cortical surface. After the animals participated in the recordings, their head post, eye coils, and manipulator were removed, and they were donated to a local zoo. During the recording, the animals were seated in a chamber dimly lit at a low scotopic level (1–2 lx, LX-110 Lux Meter). They were presented with a collection of 13 pictures of different natural scenes (consisting of pictures of animals, faces and landscapes, 800 × 600 pixel resolution; taken from Corel photo library), which were displayed on a computer monitor (frame rate: 60 Hz) located 57 cm in front of the animals subtending 40 × 30° of visual angle. As a control, for every third stimulus presentation, a blank frame with black background was presented instead of a natural image. We refer to the trials with natural image stimuli as image condition trials and those with the blank frame as blank condition trials. In order to maintain the alertness of the animals, before every trial, we forced them to perform a fixation task, where a black frame with a single central fixation spot was presented and they had to fixate to it (1° window) for 1 s in order to be rewarded (referred to as fixation part). Then, the natural images or the blank frame were presented for 3 or 5 s for monkey D or S, respectively (free-viewing part). In the free-viewing part, the animals were allowed to freely explore the monitor screen with self-initiated eye movements (), while the experimental protocol required the animals to maintain their gaze within the limits of the monitor for the 3- or 5-s presentation period, to be rewarded with a drop of juice. After a free-viewing part, another fixation part began, followed by the next free-viewing part, and so forth. This process was repeated as long as the animals were motivated to continue the task. Only the data from the free-viewing part with successful gazes served for the following analyses.
Figure 1. Eye movements and V1 activity during free viewing of a natural image. (A) Trace of eye movements of monkey D on 1 of 13 presented images. Red dots indicate fixation positions and blue curves represent the traces of saccadic eye movements. Green dots indicate (more ...)
Recording of Eye Position and Extraction of Eye Movement Events
Vertical and horizontal eye positions were monitored ( top) with a search coil driver (DNI Instruments, Resolution: 1.2 minutes of arc) and then digitized at 2 kHz. To extract the different types of eye movements from the eye traces, we developed an automatic algorithm (coded in C) based on the following definitions of eye movement events. Saccades were defined as eye movements with an angular velocity higher than 100°/s lasting for at least 5 ms. In addition, saccades were required to exhibit a minimum acceleration of 170°/s2. Fixation periods were classified as such when they lasted at least 100 ms with the eye position maintained within 1° of the gaze location reached at the end of a saccade. Sustained movements with angular velocities ranging from 70 to 150°/s, and durations of at least 100 ms were classified as drifts, during which we did not analyze the data in the present study. Only the unambiguous fixation periods that were initiated and terminated by unambiguous saccades were considered for further analysis. We call each combination of a saccade and the immediately following fixation period as a “saccade–fixation (S–F) trial.” By this definition, an S–F trial begins with a saccade-onset (corresponding to the end of the preceding fixation), followed by a saccade-offset, which is equivalent to fixation-onset and ends with a fixation-offset. The total number of S–F trials was 2452 for monkey D and 2686 for monkey S.
Neuronal activity in the primary visual cortex during the free-viewing task was recorded with an array of 8 individually adjustable custom fabricated nichrome tetrodes (1–2 MΩ impedance). The electrodes were positioned in a circular array, with a center-to-center distance of ~400 μm. The signals were amplified (10 K), separated into multiunit activity (MUA; 0.5–5 kHz) and LFPs (1–200 Hz) by band-pass filtering and then stored in an electronic device at 25 kHz and 3 kHz sampling rates, respectively. Only one LFP signal was selected from 1 of the 4 channels of each tetrode. A notch filter was applied to the LFP signals in order to remove hum noise (at 50 Hz from the power line and at 60 Hz from the monitor refreshing rate). To observe a single trial LFP activity in a frequency-resolved manner, we applied wavelet transform to a LFP trace recorded during a presentation of one of the natural image stimuli (). We used a Morlet wavelet defined at frequency f
and time τ by
, according to Le Van Quyen et al. (2001)
. The parameter σ was set to 5/(6f
), so that the wavelet contains about 5 wave cycles. The MUA signals were fed through an off-line sorting program (Gray et al. 1995
) to reconstruct the spike trains of single units recorded simultaneously by a single tetrode. On successive penetrations (i.e., recording sessions) through the same guide tube, recordings were resumed always at least 200 μm deeper than during the previous recording session. This sampling procedure was continued until activity could no longer be measured, and then the guide tubes were repositioned. We identified 153 single units from 26 recording sessions for monkey D and 251 units from 51 sessions for monkey S. Some penetrations crossed V1 twice in the anterior part of the calcarine sulcus, which led to systematic changes in receptive field (RF) position. The location of the RF of a multiunit response at one individual tetrode was assessed by hand mapping using a mouse-driven white bar, while the animals fixated on a small fixation point in the middle of the screen. Because all tetrodes were close to each other, multiunit RFs largely overlapped.
Event-Triggered LFP Averages
To study LFP activities in response to eye movements, we calculated LFP averages either triggered on the onset of fixations or on the onset of saccades (). For this calculation, we first band-pass filtered the recorded LFP signals between 1 and 100 Hz, by eliminating the frequency components outside this range in the Fourier space before inversely transforming the residual components back to the time domain. For calculating the fixation-onset–triggered LFP average, we extracted from each S–F trial (12408 for Monkey D and 8585 for Monkey S) a 300-ms segment of the filtered LFP signal (–100 to 200 ms relative to fixation-onset) and averaged them. To access the variance of the LFPs across trials, we calculated the standard error at each time bin. These calculations were performed at the time resolution of the sampling frequency of 3 kHz. The saccade-onset–triggered average was calculated in the same manner except that saccade-onsets were used as the reference time point for extracting the LFP segments.
Figure 2. Spiking and LFP activities related to eye-event onsets during free viewing of natural images. Panels (A,C,E) and (B,D,F) show data of monkey D and S, respectively. (A,B) Mean firing rates triggered on fixation-onset (red) and saccade-onset (blue), estimated (more ...)
Event-Triggered Mean Firing Rate
To derive the firing rate responses in relation to the eye events, we computed the mean firing rates of all neurons across all S–F trials aligned either to saccade- or to fixation-onset (). Before averaging, the spike trains were smoothed by convolution with a Gaussian kernel of 4 ms standard deviation. Then segments of 300-ms duration (between –100 and 200 ms relative to the respective trigger event) were extracted and averaged to retrieve the smoothed population peri-stimulus time histogram (PSTH). As for the LFP average, the variability of the firing rate response is captured by the standard error calculated at each time bin on the same time resolution.
Phase Consistency of LFP Activity Across Fixations
The average LFP shows a clear sinusoidal waveform, indicating that the LFP modulation in response to the eye events is on a specific time scale. Deriving this time scale by direct application of a spectral analysis to the average LFP time series would lead to a very poor frequency resolution due to the limited duration (300 ms) of the time series. Therefore, we determined the time scale by employing a phase consistency analysis of the LFP responses across fixations in a frequency-resolved manner in the range between 1 and 100 Hz (). To estimate the phase of the LFP activity for each frequency, we first applied a band-pass filter of a bandwidth defined by ±0.2·fc (Hz), with fc, the center frequency of the filter, which is varied from 1 to 100 Hz in steps of 1 Hz. This definition of the bandwidth renders different frequency resolutions for different center frequencies, so that it allows a fine temporal resolution for a high center frequency and vice versa. (For example, the bandwidth for the center frequency of 10 Hz becomes 4 Hz [i.e., from 8 to 12 Hz], which is identical to the typical definition of the alpha frequency band.) We obtain the instantaneous phase of the filtered signal as the arc tangent of the ratio between the filtered signal and its Hilbert transform. The phase consistency of the LFP signals across trials was obtained by calculating the vector average of the phases at any instant in time relative to the trigger event (i.e., fixation-onset or saccade-onset). The lengths of the resulting average vectors represent the phase consistency values. This procedure provides us with a measure of the reoccurrence of a specific phase of the signal at the same time relative to trigger onset across trials. Irrespective of the specific trigger event, the maximum phase consistency value was found during the LFP response at frequencies of fc = 16 (image) and 7.5 (blank) (Hz) for monkey D and fc = 13 (image) and 5.5 (blank) (Hz) for monkey S (). The frequency extracted for each of the monkeys was considered in further analyses as their dominant LFP response frequency.
Figure 4. Firing rates and LFP activities related to eye-event onsets in the blank condition. The figure is organized in the same way as , except for the gray curves in (A–D) that show the results for the image condition (fixation-onset–triggered (more ...)
Saccade Duration–Resolved Averages of LFP and Firing Rates
To elucidate the temporal relationship between the eye movement events and the neuronal activities, we studied how the response latency of LFP and the firing rate depend on the duration of saccades. Therefore, we sorted S–F trials by saccade duration and calculated the respective fixation-onset–triggered averages of the LFPs for saccades of similar durations. Before averaging, the LFPs were band-pass filtered with a center frequency set to the respective dominant frequency. The result was smoothed across saccade durations with a sliding window (10-ms width) starting at saccade duration of 5 ms with 2 ms increments until 95% of the data were covered. We also computed the saccade duration–resolved firing rates accordingly. The obtained LFP and firing rate matrices are displayed in pseudocolor plots () as a function of time (x-axis, representing the time relative to fixation-onset) and saccade duration (y-axis, representing saccade duration).
Figure 3. Saccade duration–resolved averages of LFP and firing rate during free viewing of natural images. Panels (A,C) and (B,D) show data of monkey D and monkey S, respectively. (A,B) Top: grand average LFP calculated from all S–F trials irrespective (more ...)
To examine whether or not the timing of the spikes is related to the LFP activity, we assessed the degree of temporal locking of the spikes to the phase of the LFP modulations. For this purpose, we measured the phase-locking value (PLV, ). Thus, first we estimated the instantaneous phase of the LFP signals by applying the Hilbert transform (as done in the phase consistency analysis described above). Before the application of the Hilbert transform, the LFP signals were filtered with the bandwidth introduced above and around the center frequency fc
set to the monkey's dominant frequency, which differed in the different behavioral conditions (image and blank). For calculating the PLV, we extracted the phase values at the times of the simultaneously recorded spikes
that is, the timing of the k
-th spike of cell i
within the fixation period of the j
-th S–F trial. The resulting phase values were denoted as
. Based on these, we calculated the PLV defined as
, where N
is the total number of spikes taken into account. For the results presented in this study, we related spikes and LFPs recorded from the same electrode, however, comparable results were obtained when only signals from different electrodes were related.
We applied the phase-locking analysis either within a time window at a fixed position or in a sliding window fashion () to yield the time dependence of the phase locking. In the former, the width of the window was set to one cycle of the dominant frequency to avoid a bias in the sampling of phase values and was centered at the peak of the fixation-onset–triggered average firing rate (, yellow colored area). In the sliding window analysis, the window width was set to 20 ms and slid from 0 to 200 ms starting at fixation-onset. To explore a possible dependence of the PLV on the order of spikes during the fixation, we calculated the PLVs separately for the set of the first (1ST) and the second (2ND) spikes occurring after each fixation-onset and also for the set of all (ALL) spikes occurring during fixation periods (). For cross-checking, we also repeated all the analyses based on an alternative phase estimation method (wavelet phase estimation, same method as described in Electrophysiological Recordings), which confirmed our results (not shown here).
Figure 8. Time-resolved analysis of phase locking. Panels (A,C) and (B,D) show data of monkey D and monkey S, respectively. (A,B) PLV of the first spikes as a function of time relative to fixation-onset. PLVs calculated from the real data (red trace) are plotted (more ...)
Figure 7. Phase locking of spikes of individual neurons to LFP modulations during fixation periods. Panels (A,C,D,E,F) and (B,G,H,I,J) show data of monkey D and S, respectively. (A,B) Spike histograms of different sets of spikes aligned to fixation-onset (bin width: (more ...)
Significance Test for Phase-Locking Value
We assessed the significance of PLV with 2 different surrogate methods: one is based on random shuffling of data and the other on random resampling of data.
Random Shuffling Surrogate
This surrogate method was employed both in the fixed time window and in the moving time window analyses of PLV. An inhomogeneous distribution of spikes (i.e., nonstationary firing rates) within the PLV analysis window may cause spuriously large PLVs due to a bias in the considered phase values. Similarly, if the width of the analysis window is shorter than one cycle of the dominant LFP frequency, only a restricted set of phase values is contained and thus may bias the resulting PLV. In order to avoid wrongly assigned significance to PLV estimates, we quantified the significance of the empirical PLV by comparison to a distribution of PLVs derived from surrogate data sets, which were generated by shuffling the spike trains of the individual neurons across different S–F trials (). Thereby the spikes were related to LFPs of nonsimultaneous, randomly selected S–F trials. Each randomization generated one surrogate data set, which resulted in one surrogate PLV. The PLVs estimated from 10000 such surrogate data sets were used to construct a distribution of the surrogate PLVs to derive the P
value Ψ of the PLV obtained from the original data. We quantified the significance by the surprise measure (SM)
defined as (Palm et al. 1988
; ). The trial shuffling procedure destroyed any possible correlation between the spike trains and the LFPs but preserved the potential sampling biases of spikes and phases within the analysis window. Thus, the obtained surrogate PLVs reflect the degree of phase locking resulting only from these biases. This approach yields a more conservative estimate of the PLV of the original data as compared with surrogates that do not preserve the biases inherent in the data, e.g., by spike time randomization.
Figure 6. Schematic illustration of phase-locking analysis and generation of surrogate data. (A) Each row sketches data from one S–F trial around fixation-onset (dashed vertical line): LFP (black curve) and a simultaneously recorded spike train (vertical (more ...) Random Resampling Surrogate
This surrogate method was employed only in the PLV analysis with the fixed time window. In order to directly assess whether 1ST and 2ND spikes are significantly more strongly locked to the background LFP than arbitrarily selected spikes, we compared the SMs for 1ST and 2ND spikes with the SMs for arbitrarily chosen subsets of ALL spikes. For this purpose, we randomly picked from ALL spikes within the analysis time window the same number of spikes as 1ST or 2ND spikes and computed the SM of PLV for this subset of ALL spikes. We repeated this 1000 times and estimated the median and the 95 percentile of the SMs for the randomly resampled surrogate data sets. The SM for 1ST or 2ND spikes was considered to be significantly higher if it exceeded the 95 percentile of the corresponding surrogate.
Effect of LFP Amplitude on Phase Locking
To study the relationship between the amplitude of the LFP responses and the phase locking of spikes, S–F trials were separated into 2 groups (hi-peak and lo-peak group) according to the height of the first positive peak of the filtered LFP signal (filter details, see above) after fixation-onset. For deriving potential differences in the locking degree of the respective groups, we calculated separately for the 2 groups the time-dependent PLV () and the LFP averages ().
Figure 9. Influence of LFP amplitude on spike time precision. (A) Proposed model to account for spike timing precision based on the LFP modulation. Modulations of the field potentials are assumed to correlate with changes in the effective firing threshold of neurons (more ...)
Unitary Events Analysis
To examine the relation of the phase relation of spikes to the LFP and the occurrence of excess spike synchrony between neurons, we used the unitary events (UEs) analysis method for the detection of significant spike synchrony (Grün et al. 2002a
; Grün 2009
). The method enables to retrieve the time-resolved occurrence of excess spike synchrony and to relate its time course to the LFP modulations in response to eye events (). To evaluate UEs, we used the same approach as applied in our previous study (Maldonado et al. 2008
), where it is outlined in detail. In brief: For each pair of simultaneously recorded neurons, we extracted the empirical number of coincidences (tolerated temporal jitter: 5 ms) from all trials within a given time window. To evaluate the significance of the detected number of coincidences, its count was compared with the number of coincidences expected on the basis of the firing rates of the neurons within the same time window. This expected number was derived as the sum of the trial-by-trial products of the firing probabilities of the 2 neurons, multiplied by the number of time steps within the window derived counts. The firing probabilities were estimated from the corresponding trial-by-trial spike counts within the window, normalized to the number of bins. The significance of the empirical count is derived as the P
value estimated from a Poisson distribution with the mean set to the derived expected number. For a P
value smaller or equal to a predefined significance level (here set to 0.05), the window is considered to contain UEs, that is, significant excess spike synchrony. The application of that procedure in a sliding window fashion (window size: 50 ms, increment: 1 ms) permits to extract the time-dependent UE rate for each neuron pair. The UE-PSTH in , bottom, represents the UE rate averaged over all neuron pairs.
Figure 10. Temporal relationship between LFP, firing rate, and UE rate. Red solid and blue dashed curves in the top panel represent saccade-onset–triggered average LFP and its 1ST temporal derivative (dashed blue), respectively. The pink vertical line indicates (more ...)