| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Department of Neurology and Neuroscience, Weill Medical College and Graduate School of Medical Sciences of Cornell University,New York, New York 10021
Submitted 7 January 2003; accepted in final form 10 July 2003
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIX AAPPENDIX BAPPENDIX CDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIX AAPPENDIX BAPPENDIX CDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
; Green 1957). These signals appear at the end of the saccade and are differentially modulated by the direction of the eye movement (Skrandies and Laschke 1997). Tasks requiring attention, such as reading, increase the amplitude of these signals and shorten their latency (Barlow 1971; Fourment et al. 1976).
In animals, saccades are correlated with transient local field potentials (LFPs) in the lateral geniculate nucleus (LGN) of the monkey and cat (Brooks and Gershon 1971; Feldman and Cohen 1968; Jeannerod and Sakai 1970), and in the temporal (Sobotka et al. 2002), parietal (Pesaran et al. 2002), and frontal lobes (Klostermann et al. 1994; Seidemann et al. 2002) of the cerebral cortex and mesolimbic structures (septum and hippocampal formation: Sobotka and Ringo 1997) of the monkey. As with the human scalp-EMPs, the LFP-EMPs recorded in mesolimbic structures of the alert monkey are differentially modulated by the direction of the saccade and by the visual environment (Sobotka and Ringo 1997). In a lighted environment, these LFP-EMPs are generally faster (shorter latency for time-to-peak) and more biphasic (larger peak-to-trough excursions) than in the dark.
A strong extraretinal contribution to the EMPs is emphasized by the following observations: 1) complete retinal loss through optic nerve section or photocoagulation does not entirely eliminate EMPs in the cat LGN (Brooks and Gershon 1971; Jeannerod and Saki 1970); 2) EMPs in the cat cortex and LGN appear both with caloric stimulation of the labyrinths under flaxedil paralysis sufficient to prevent saccades and with electrical stimulation of the pontine reticular formation with concomitant saccades when the paralytic is removed (Brooks and Gershon 1971); 3) EMPs are not produced in the monkey LGN by rapid mechanical displacement of the eye (Feldman and Cohen 1968).
Taken together, both the human and animal studies suggest that EMPs represent a signal in thalamic and cortical populations that is produced, in part, by the networks responsible for generating and controlling saccadic eye movements. The appearance of EMPs in the LGN, ventral visual pathway (including the temporal lobe), and mesolimbic structures—all regions of the brain not typically included in the oculomotor network (Wurtz 1996)—may therefore reflect the arrival of corollary motor activity (McCloskey 1981; Sommer and Wurtz 2002). Such corollary discharge signals may originate in the brain stem, superior colliculus (SC), lateral intraparietal area (LIP), and/or the frontal eye fields (FEF). The role of this corollary activity in the occipitotemporal (OT) pathway has not been fully resolved.
Here we examine EMPs in alert monkeys performing a pattern-recognition task. The timing and frequency content of EMPs, their regional differences in the OT pathway, their interactions with visual transients, and the influence of behavior (direction of saccadic eye movement) on their dynamics are quantified. Our goal is to determine, in more detail, the extent to which EMPs reflect corollary motor signals, to suggest how they are generated, and what their role may be in active vision.
The relationship between EMP signals recorded in the OT pathway and corollary oculomotor activity can be clarified by the temporal relationship between the time of saccade onset and the change in power in local population activity. For example, activity that precedes and is coincident with saccade onset may be associated with remapping of receptive fields (Duhamel et al. 1992; Umeno and Goldberg 1997; reviewed in Ross et al. 2001), "interrupts" from attentional circuit breakers (Corbetta and Shulman 2002) that lead to the generation of orienting saccades, and executive commands originating in the frontal eye fields that are associated with the selection of targets and implementation of scan paths (Bruce et al. 1985). Because of their timing, these EMPs cannot be generated by the visual reafference, the activity generated in the retina by the movement of the eye. On the other hand, activity that follows saccade onset and offset may represent either visual reafference or the exafference activity, that is, the activity produced by an interaction between the visual reafference and the corollary discharge (McCloskey 1981).
In addition to the timing of these signals with respect to saccade onset, other characteristics may also be useful for illuminating the origin of EMPs and their relationship to active vision. For example, regional differences in the timing and frequency content of EMPs would suggest that signals related to saccadic eye movements enter and influence the different stages of visual processing in the OT pathway through a variety of channels. Directional EMPs would indicate that the oculomotor network provides a significant component to the saccade-related activity in the OT pathway. Finally, interactions between visual transients and EMPs can be evaluated in terms of the gating of visually evoked activity by the saccade and its corollary activity.
We recorded LFPs using high-impedance microelectrodes at locations in the occipital and temporal lobes of monkeys during a task that required sustained attention and the scanning of a series of high-contrast dynamic images, conditions that maximize the amplitude and shorten the latency of EMPs (lambda waves). We examine the LFPs in a 500-ms interval centered on the onset of the horizontal component of saccades made during the trials. Here we seek to produce a time–frequency representation of the average LFP around the time of the saccade as one might do when processing acoustic signals that have a high signal-to-noise ratio. The EMP is not subtracted from each saccade-triggered LFP in the time-domain before spectral analysis. Instead, a spectrogram is computed for a number of different estimates of the EMP to establish confidence intervals for the EMP spectrogram. Time–frequency spectrograms are calculated to capture a portrait of the presaccadic, transsaccadic, and postsaccadic activation generated in the local cortical population. The spectrograms are evaluated to determine the influence of saccade direction, cortical location, and timing of visual transients on the time and frequency content of the LFPs during the perisaccadic interval.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIX AAPPENDIX BAPPENDIX CDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
LFPs were recorded from sites within the occipital and temporal lobes of 2 adult male macaque monkeys (Macaca mulatta), monkeys M and S, performing a pattern-recognition task. All animal care and both surgical and experimental procedures conformed to guidelines established by the National Institutes of Health and the Institutional Animal Care and Use Committee of the Weill Medical College of Cornell University.
RECORDING LOCATIONS. Using sterile surgical technique, recording chambers (Crist Instrument) and a head holder were attached under isoflurane general anesthesia to the skulls of the monkeys at positions selected to provide access to regions of the extrastriate visual cortex in both the occipital and temporal lobes. For monkey M, one chamber was placed over the occipital lobe (P12, L20) and a second chamber was centered over the temporal lobe (A10, L21). Both of these chambers were placed over the right hemisphere. For monkey S, 3 chambers were used: occipital lobe chambers for the left and right hemispheres (P12, L20) and a right hemisphere temporal lobe chamber (A10, L21). The recording chamber placements were made after MR imaging (T1-weighted images made in a 1.5-T field, juvenile head coil) with the animal held in an MR-compatible stereotaxic frame that was then used in the subsequent surgeries. With monkey S a second set of images was taken during the course of the studies with tungsten microelectrodes left at productive recording sites. A custom-designed head coil was used for this second imaging session and the axial, sagittal, and coronal images were combined into a full 3-dimensional volume to better trace the path of the microelectrodes through the brain. For monkey M, some of the elements used for building the cranial implant were made of steel, so it was not possible to obtain an MRI after surgery. Instead, a fluoroscope was used to produce images of the microelectrodes and guide tubes in vivo.
Based on the imaging data, and with reference to a recently published Macaca fascicularis atlas (Martin and Bowden 2000), detailed microelectrode track reconstructions relating depth to receptive field size and eccentricity (Gattass et al. 1988), and from our own single unit recordings, we assigned our occipital lobe LFP recording sites as follows: V1 for recording depths in the range of 0–1.6 mm; V2–V3, 2–8 mm; ventral V1–V2, 8–14 mm. Ventral V4 was at the same depth as ventral (upper and parafoveal visual field) V1–V2 but accessed from more anterior positions in the chamber. The temporal lobe recordings were subdivided into 3 groups. The first, shallower group of visual units and LFP responses encountered in the temporal lobe tracks (at depths of 18–20 mm in monkey M, and starting at 16 mm in monkey S) were assigned to the inferior bank of the superior temporal sulcus (STS). The deeper recordings, encountered after passing through a region devoid of single units (white matter), were assigned to the inferotemporal cortex (area TE at depths beginning at 29 mm in monkey M). More medial penetrations at this depth entered the parahippocampal gyrus (PHG).
LOCAL FIELD POTENTIAL RECORDING. Extracellular recordings were made with epoxy-insulated tungsten microelectrodes (FHC, Bowdonham, ME), with a nominal impedance of 1–4 M
. All the data reported here were obtained by recording simultaneously with 2 microelectrodes. In one series of recording sessions, one electrode was placed in the occipital lobe with the other electrode placed at temporal lobe sites (monkeys M and S). In the other series of experiments, one electrode was placed in the left occipital lobe and the second microelectrode was situated in the right occipital lobe (monkey S). The microelectrodes were advanced with hydraulic microdrives (Narashige) through guide tubes inserted through the dura. For the temporal lobe recordings the guide tubes were lowered 13 mm into the brain to protect the microelectrode tips and to straighten the trajectory of the microelectrode travel. A grid system (Crist Instrument) was used in the chambers to organize guide tube placements.
The signals from each electrode were separated into spike and LFP channels (Tucker-Davis Technologies, Gainesville, FL). After appropriate band-pass filtering the LFPs were sampled at 200 Hz and the 1- to 10-kHz channels were sorted on-line for spike shapes with either matched template filters or neural network classifiers (Chandra and Optican 1997).
All the recordings were monopolar with the ground tied to either skull screws (monkey S) or a scalp EEG lead (monkey M). The same ground was used for both microelectrodes in the dual recordings. Monopolar LFP recordings (like those reported here) are not as localized as single-unit or bipolar LFP recordings, nor as delocalized as scalp EEGs and visually evoked potentials (VEPs). An important question is: how localized or delocalized are these LFP recordings? In a separate series of experiments we demonstrate that EMPs extracted from monopolar LFP recordings vary significantly across distances as little as 400 µm along vertically oriented electrode tracks. These data and the relevant methods of analysis are explored in SUPPLEMENT 3 (the Supplemental Material is available at the Journal of Neurophysiology web site).1
ASSESSMENT OF VISUAL ACTIVITY. All of the LFP recordings were obtained from positions in the recording track associated with single-unit and multiunit activity. Visually driven units were located by detecting the presence of modulated responses to a checkerboard pattern that could be varied in contrast, check size, check aspect ratio, and frequency of stimulation period (as appearance–disappearance or contrast–reversal). After isolation of single units, activating regions (Snodderly and Gur 1995) of some of the units were outlined by hand with a mouse controlling the positions of bars and squares, and the rate and type of contrast modulation (black–white, red–green). In many recording sessions, a standard checkerboard m-sequence (50 Hz), sized to cover one of the hand-mapped receptive fields, was run with the animal performing a concurrent fixation task.
EYE MOVEMENT RECORDING AND DETECTION OF SACCADES. The monkeys' heads were fixed to the primate chair through the implanted head holder during all of the recording sessions. Eye position measurements were recorded using the horizontal and vertical analog outputs from the E5000 infrared video eye-tracking system (ASL, Bedford, MA). The animal's gaze position was calibrated each day before experiments began, and then whenever necessary to ensure the accuracy of the calibration. Horizontal and vertical signals were processed to determine the occurrence of a saccade, its amplitude, velocity, direction, and positions of fixation.
Saccade onset is detected when the eye speed surpasses a threshold equal to the smallest reliable difference (about 1.3°) between 2 consecutive positions divided by 10 ms (2 x temporal sampling resolution of the eye tracker signals), corresponding to 130°/s. Saccade end is detected when the velocity falls back below the threshold. Saccades that are preceded by signals with unusually low variance, such as a flat line attributed to the loss of signal (as can be produced by a blink), are not included in the analysis. The average position recorded during periods that have signal but no detectable velocity components are taken as the centers of positions of fixation. Saccades that are separated by a fixation period 
80 ms are fused through interpolation if they are both in the same cardinal direction (left, right, up, down) and discarded if they are not.
EXPERIMENTAL CONTROL AND BEHAVIORAL TASK. The monkeys performed a pattern-recognition task that required them to report the appearance of a target during the presentation of a sequence of similar images. The rate of image change (2, 3, or 4 Hz) was kept in a range that produced better than chance performance on the tasks. Shown in Fig. 1C is the target image with superimposed numerals indicating the temporal modulation sequence (Fig. 1A) assigned to each hexagonal region ("cube") in the image. The modulation moves the white diamond from the top (-1, Fig. 1B) to the bottom (+1, Fig. 1B) of the hexagonal region; for the "Y"-junction this produces what appears to be a cube viewed and lighted from above (-1) or a cube viewed and lighted from below (+1). Eight of the 10 regions are modulated, whereas the center 2 regions (9, 10) are static. The modulation sequence is a 6th-order m-sequence (63 elements in length) (Sutter 1992). The 8 modulated cube regions are assigned to different shifts of the m-sequence. In this study, the neural activity was not correlated against the m-sequence to produce estimates of the responses to the appearances of either of the elements (-1 or +1) in Fig. 1B at each of the eight positions (1–8) in Fig. 1C (this is reported in Purpura et al. 1999). Here the modulation sequence is used to produce a set of images whose subregions (cube elements) are nearly uncorrelated in space and in time when considered pairwise, and also have nearly zero higher-order correlations at each position for up to 6 transitions in the past. At least one cube element is modulated on each transition and, on average, 4 of the elements are flipped.
|
The use of an m-sequence to specify the transitions in the subregions and the relationship between the transitions in different subregions reduces the degree of correlation between subsequent samples of the stimulus. This will be true whether the visual system samples a sequence in the stimulus through extended periods of fixation or by selecting different regions of the stimulus by making attentional shifts and saccades across the display. Thus the saccade-related neural activity in the perisaccadic interval will be less influenced by correlations in the visual display. Presaccadic spatial adaptation has been shown to influence the responses to visual stimuli presented during a saccade (Judge et al. 1980) and after the saccade (Gawne and Woods 2003). Here we have balanced these potential effects by presenting the same cube configuration (-1 or +1, followed by -1 or +1, respectively; see Fig. 1B) as often as the opposite configuration (-1 or +1 followed by +1 or -1, respectively) in subsequent frames in many regions of the stimulus. Having only one instance of pattern appearance at the beginning of each trial, followed by a period of pattern modulation, also provides a more balanced sequence of visual inputs across the trials: the overall values of luminance in each of the hexagonal subregions are the same for both configurations -1 or +1 (Fig. 1B), and regions of contrast modulation are fragmented in time and space across the sequence of images and trials.
Subsequences of the m-sequence were presented on each trial. The sequences overlapped on subsequent trials. A new trial always began one slot behind where it ended at the termination of the last trial, before the last presentation of the target. In Fig. 1A staggered bands (labeled "Trial #1," "Trial #2," and "Trial #3") beneath the m-sequences indicate which configurations of the multi-input m-sequence were used in each of the 3 trials shown. A vertical bar within each band shows where the target replaced the next configuration in the sequence. The length of each trial was selected at random from a choice of 0 (a catch trial where only the target is shown), 2, 4, 5, or 6 s before appearance of the target. In the examples shown here, the first 3 trials would run for 5, 4, and 5 s. Recognition of the target is signaled by a bar release between 150 to 600 ms following the target. The target is followed by the next configuration in the multi-input m-sequence to provide a backward mask of the target before the screen is returned to a featureless mean gray background. The size of the liquid reward (juice or water) is scaled to the length of the trial.
The visual stimuli were presented on a Mitsubishi XC-2930C CRT (27° H x 21° V screen) placed 114 cm in front of the monkey. Images were displayed with a frame rate of 95.5 Hz. The mean luminance of the featureless gray background was 50 cd/m2 with the white, light gray, and dark gray regions of the hexagonal elements (Fig. 1B) at 100, 80, and 15 cd/m2, respectively. Each hexagonal region, or cube, was approximately 3.5° H x 4° V, with the overall configuration of 10 conjoined cubes subtending about 14° H x 10° V on the screen (Fig. 1C). The surrounding gray background was left on the screen between trials. Graphics were produced with a VSG 2/3 (Cambridge Research, Cambridge, UK) video board running on a Pentium PC. Real-time control of the graphics computer, monitoring and recording of gaze position, recording of neural activity, monitoring behavioral responses, and reward delivery were performed with a second Pentium PC running TEMPO (Reflective Computing, St. Louis, MO). Coordination of the 2 PCs was administered through a proprietary data transfer protocol (Reflective Computing).
GROUPINGS OF SACCADES FOR ANALYSIS. Saccades are generated in a number of different contexts during the trials. In this study, we consider groupings based only on the presence of a significant horizontal component in the eye movement signals. We classify the saccades into different analysis groups based on several factors: 1) The direction of the saccade with respect to the laterality of the recording site. Saccades are thus either contraversive (directed away from the hemisphere of the brain containing the recording site) or ipsiversive (directed toward the recording site). 2) The direction of the saccade in body-centered coordinates; that is, leftward saccades are to the left side of the body. 3) The time of a stimulus sequence interchange (visual transient) with respect to the time of the saccade in the 500-ms analysis window centered on the saccade. Groups are formed for: i) saccades that are preceded (within 250 ms) by a visual transient; ii) saccades that are followed (within 250 ms) by a visual transient; iii) saccades produced between 2 visual transients occurring within the same 500-ms window (3- and 4-Hz stimulation). 4) Consecutive contraversive (or leftward) saccades are placed in 2 separate groups as are consecutive ipsiversive (rightward) saccades. This separation is made to reduce the confluence of the postsaccadic phase of one leftward saccade, for example, with the presaccadic phase of the next, or previous, leftward saccade. Consecutive saccades that have the same direction but that are separated by <250 ms are dropped from the analysis.
All of the analysis groups for the EMPs were established before any statistical testing. The goal here was to first organize the field potential data by behavioral criteria such as the direction of saccade and then establish whether intrinsic components of the EMP waveforms could be segregated based on statistical tests.
Figure 2 illustrates how a sequence of saccades made over 2 trials is divided into analysis groups L1, L2, R1, and R2. Successive saccades in the same direction are alternately assigned to groups 1 and 2 (L1 and L2 or R1 and R2) regardless of the presence of intervening saccades of the opposite direction. For example, beginning with the first leftward saccade in Trial 1, every other leftward saccade is assigned to group L1 even if a rightward saccade intervenes or a new trial commences. The other leftward saccades are assigned to group L2. The rightward saccades are similarly assigned to groups R1 and R2. In this example, by the end of Trial 2, there will have been 3 L1 saccades, 2 L2 saccades, and 3 each of R1 and R2.
|
Data analysis
SPECTROGRAMS. Among the features one would like to collect from saccade-triggered averages of LFPs are: 1) the time-to-peak power in the signals; 2) the shape of the waveforms [are they dominated by one sinusoidal component or do they exhibit frequency modulation (FM)?]; 3) the extent to which they are influenced by the timing of stimulus transitions in the perisaccadic interval, saccade direction, and recording location.
Dominating this type of analysis, as with the analysis of all evoked field-potential data, are questions concerning what components of the LFP activity are robust and how much confidence can one place on the estimates of such extracted features. LFP averages are often about one-fourth to one-tenth the amplitude of the range exhibited on individual trials. Thus the issue becomes, in the face of a relatively low signal-to-noise ratio, what is the most efficient way of analyzing such data. The answer is to make estimates in a restricted range in the spectrotemporal domain (Percival and Walden 1993) using an approach that allows multiple independent estimates to be made from the same data set. Multitaper spectral analysis (Thomson 1982) provides a means to produce such estimates; the availability of these multiple estimates improves the reliability of conclusions that can be made about features extracted from the physiological signals of interest. Multitaper spectral analysis also establishes a rigorous set of guidelines for obtaining an optimal tiling of the spatiotemporal domain that avoids violating theoretical limits imposed on time and frequency by the uncertainty relation (Jarvis and Mitra 2001; Mitra and Pesaran 1999; Percival and Walden 1993).
The difficulty with spectral estimation is that the statistical properties of samples of neural activity change with time, violating the stationarity requirements of all spectral analysis methods. It is sometimes possible to produce cyclo-stationarity by taking samples collected during repetitive stimulation or repeated behavioral actions. During the temporal intervals defined by these sampling procedures, however, the neural activity is often still nonstationary. To produce an estimate of how the spectrum of the LFP changes with time around the saccade we calculate Thomson's high-resolution multitaper spectrogram (Thomson 2002) using the saccade onset time to define the temporal interval for drawing samples from the trials. Spectrograms minimize the violation of the stationarity requirement by calculating spectral quantities over a smaller temporal window in the interval of interest, in our case the perisaccadic interval, and shifting the window over that interval. A brief description of Thomson's version of the spectrogram technique is provided in SUPPLEMENTS 1 and 2 (the Supplemental Material is available at the Journal of Neurophysiology web site).2
ESTIMATE OF THE STANDARD ERROR IN THE SPECTROGRAM. The goal here is to extract features from the spectrograms of the average LFPs to characterize how saccades modulate the neural activity in local cortical populations. Spectrograms are 2-dimensional (2-D). As such, a salient feature or region within a spectrogram can be defined by its area, the orientation and eccentricity of a best-fitting ellipse, and the 2-D coordinates of its centroid. Thus one problem with carrying out such a region-of-interest (ROI) analysis is determining what is a salient and robust feature in the spectrogram. Here, we establish a threshold level of power spectral density in each spectrogram (see APPENDIX A). This value is then compared point by point in the spectrogram with a 95% confidence interval for the mean computed at each point using all the samples in a subset (e.g., those associated with rightward saccades only) of the LFPs. If a point in the spectrogram has a confidence interval that does not bracket the threshold (noise) level for the spectrogram, then this point is included in a region of interest. If the threshold value does fall within the point's confidence interval, then that point in the spectrogram is set to zero power spectral density.
Each LFP sample referenced to a saccade contributes additively and independently to our estimate of the spectrogram of the saccade-triggered average LFP. As a consequence, the jackknife standard error calculation can be used to produce a confidence interval for the spectrogram of the mean LFP (Efron and Tribshirani 1998; Thomson and Chave 1991). This is shown in APPENDIX A.
ROI (REGION-OF-INTEREST) ANALYSIS. To extract statistically significant regions from the spectrograms we need to define the lower bound and the maximum value for the distribution of values in the spectrogram (Savg in APPENDIX A). The estimates of the lower bound and maximum are used as inputs to the MATLAB function (MATLAB, v6.1) ROICOLOR. This function determines the coordinates in a 2-D array where the power ranges between 2 levels. We have chosen a noise level (seeAPPENDIX A) and an estimate of maximum power (see APPENDIX B) in the spectrogram as the 2 levels. The extracted regions are then passed through a continuity filter (BWLABEL function, MATLAB) to determine which of the values in the ROI are connected to other values in the spectrogram by more than the corners of a time–frequency pixel. The assumption here is that continuity in the time–frequency plane reflects waveform integrity in the LFPs. Because the spectrograms have been tested against a threshold (noise level) before application of ROICOLOR and BWLABEL, the ROIs are maps of areas in the spectrograms that have significant power as determined by the jackknife procedure.
Using the MATLAB function REGIONPPROPS, we also extract the values of several parameters from each ROI. These parameters include the area, the coordinates of the centroid (centers of mass), and the orientation and eccentricity of a best-fitting ellipse. The areas of the ROIs are used to identify the largest ROI in a spectrogram. The centroid of an ROI is its weighted center. The weights are formed from the distributions of power in the frequency and time domains. This estimate uses all of the data describing a region, so focusing on the largest ROIs strengthens the estimate of its centroid.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIX AAPPENDIX BAPPENDIX CDISCLOSURESACKNOWLEDGMENTSREFERENCES |
|---|
), we will treat these different types of runs as only repeats of the same pattern-recognition task for the analyses presented here. Studying active vision
An illustration of the events constituting a typical trial is shown in Fig. 3. Once the monkey's gaze enters the screen and dwells there for 
250 ms, the first configuration slotted for the current trial appears (Fig. 3A, image frame 49). In this trial, 10 configurations are presented (interchange rate of 2 Hz) before the target is displayed (Fig. 3B, top row, image frame 127). As can be seen from the eye movement traces in Fig. 3, the monkey was free to shift its gaze within a region bounded by the screen during the course of the trial. Eight saccades with detectable rightward components and 10 saccades with leftward components can be extracted from the eye movement records for this trial. The eye movement traces and LFPs recorded simultaneously in the occipital and temporal lobes for this trial are shown in Fig. 4. A typical recording session provided behavioral and neural data from 250 similarly organized trials.
|
|
As seen in Fig. 4, LFPs recorded in both occipital (V2) and temporal lobe (TE) sites contain contributions from external visual transients, and an additional set of transients more closely correlated with the times of saccades (asterisks in V2 LFP). Our goal here is to characterize the spectrotemporal content of the neural activity in an interval surrounding the saccades so that the activation time of the EMP can be reliably determined and associated with a particular saccade and recording location. The emphasis here is on the average LFP activity triggered by the saccade (the EMP). The average LFPs obtained from all the trials in one recording session, using saccade onset time for the averaging reference, are shown in Fig. 5, A and B. Transient components in the V2 and TE LFPs are seen in the perisaccadic intervals centered on the saccades. The saccades with significant leftward (contraversive) and rightward (ipsiversive) horizontal components are associated with different average LFP waveforms. For the V2 and TE recordings, the rightward saccades produce larger transient average LFPs in the postsaccadic interval. We will see later that the bulk of the occipital lobe EMPs in our recordings show, on average, less directional modulation (see Figs. 10 and 11) than the example presented here in Figs. 5, 6, 7, 8. We highlight this less-typical example because it allows us to demonstrate how our methods of analysis can reveal differences in the timing and waveforms of EMPs associated with leftward and rightward saccades even when the LFPs are recorded from the same cortical location.
|
|
|
|
|
|
Characterizing the signals in detail during active vision where saccades are often made in rapid succession requires addressing a number of concerns. An averaging epoch of 1 s on either side of the saccade onset time can introduce multiple eye movement events into the pre- and postsaccadic phases of each estimate. As can be seen in Fig. 4, the rate of saccade generation is approximately 3/s. Restricting the temporal averaging window to 500 ms centered on the saccade time increases the independence of the samples. A 250-ms postsaccadic phase of neural activity is less likely to contain the presaccadic activity appearing 250 ms before the next eye movement. A 500-ms window also reduces the chance that the activity associated with undetected saccades (i.e., saccades that did not pass the automated velocity criterion comparison) will contribute to either the pre- or postsaccadic components of the LFPs. Closer inspection of Fig. 4 reveals that there was 1 leftward (at about 900 ms) and 2 rightward (at about 3,800 and about 4,900 ms) small-amplitude saccades that were missed by our conservative saccadic detection algorithm in this trial alone. These 3 small-amplitude saccades probably represent fixational eye movements. They are not produced at an abnormally high rate and, as a consequence, are preceded and followed by periods of 250 ms that do not contain large-amplitude saccades. Thus these small-amplitude saccades would not have contaminated the perisaccadic intervals of the saccades obtained from this trial. As discussed in METHODS Data analysis, an attempt was made a priori to maximize the independence of the LFP samples by forming separate analysis categories for saccades of different directions, for consecutive saccades of the same direction, and for saccades preceded or followed by stimulus transitions.
Averages formed with 500-ms time windows are shown in Fig. 6, A and B. Here, for the purpose of demonstration, the averages for ipsiversive (rightward) and contraversive (leftward) saccades are formed from all of the saccades with no segregation based on m-sequence transition times (stimulus transients) or saccade ordering. The LFP activity associated with each saccade is shown in Fig. 6, C and D.
In the saccade-by-saccade V2 and TE Left plots (Fig. 6, C and D), 31 of 119 leftward saccades are followed by an m-sequence transition and a rightward saccade, with the rightward saccades having an average latency of 200 ms with respect to the stimulus transients. Fifty-three of the remaining leftward saccades are not paired with any rightward saccades. Thirty-five additional leftward saccades are paired with rightward saccades but their temporal relationship with respect to the former is more variable.
Two prominent features appear in the TE Left plot (Fig. 6D, top panel). The blue swatch in the presaccadic phase appears to track the stimulus transitions. However, this feature leads the m-sequence transitions by 50 ms. Thus the most likely source for a segment of this blue region is the LFP activity associated with rightward saccades that were generated in response to some of the stimulus transients in the postsaccadic phase of the leftward saccades. The second prominent feature in the TE left plot (Fig. 6D, top panel) is the region of positive LFP voltage values (yellow-reds) that appears to straddle the saccade onset time (0 ms) over the sample set. Because the stimulus transitions move through this region rather than delimit its onset side, it is unlikely that the positive peak in the EMP (see Fig. 6B, left) is a visually evoked potential. There may be interactions between the stimuli and saccade-related activity but these are difficult to extract from the postsaccadic components of the averages of the signals that we focus on in this study (see following text, Interactions between visual transients and saccades).
The extent to which the averages calculated for one particular saccade grouping parameter are distorted by the activity of oppositely directed surrounding saccades is reduced by several factors. First, putting sequential saccades with the same horizontal component (leftward, for example) into 2 separate analysis groups disrupts the "wraparound" effect seen in the saccade-by-saccade plots in the TE Left plot (Fig. 6D). This procedure allows for time to develop between the postsaccadic phase of one leftward saccade and the presaccadic phase of the next leftward saccade to be included in the analysis group. As a consequence, the timing of an interaction between a stimulus transition and a saccade appearing late in the postsaccadic phase of an oppositely directed saccade will tend not to reinforce a similar interaction 2 saccades away. There will be even less reinforcement of these types of interactions when the saccade groups are formed across trials. Second, the emphasis here is on estimating the EMP, and the quality of this estimate formed by using the saccade onset time as the averaging trigger. The averaging procedure reinforces the signals timed to this trigger and leads to partial cancellation of LFP components produced by oppositely directed saccades appearing in the perisaccadic interval.
Spectrograms
Time–frequency spectrograms for the average LFPs presented in Fig. 6 are shown in Fig. 7. For these spectrograms, the inputs to the spectral analysis are the averages calculated from 119 leftward saccade-referenced LFPs (Fig. 7, A and C) and 109 rightward saccade-referenced LFPs (Fig. 7, B and D). Traditionally, calculation of a spectrogram involves the subtraction of the average signal from each signal sample before computing the Fourier transform. Here we produce a time–frequency representation of the average LFP, not of the residual variance remaining after subtraction of the average. The nonzero power spectral density values in the spectrograms were established by setting a threshold based on an estimate of the standard error of the mean spectrogram (see METHODS, Data analysis, and APPENDIX A).
The centroids of the largest ROIs in the spectrograms in Fig. 8 are indicated by the white stars; the white circles indicate the maxima. The orientation and eccentricity of the major axis of the best-fitting ellipse to an ROI summarize the temporal waveform of the transient response in the perisaccadic interval. For example, an orientation of 90° from the vertical and an eccentricity near 1 are associated with a temporal waveform that does not sweep through frequency bands over the course of the response. The temporal waveform thus resembles a mixture of a few damped sinusoids restricted to a narrow band of frequencies (Figs. 6B, leftward saccades, and 8C). An orientation of 0° or 180° and an eccentricity near 1, on the other hand, indicates that the temporal waveform is more short-lived but with a much richer frequency content as is often produced by a transient response. Finally, a more oblique orientation is produced by temporal waveforms that display FM. In Fig. 6A (rightward saccades), the average LFPs exhibit a period of rapid modulation, after the saccade, that is then followed by a slower component. Note that the spectrogram for this waveform shows a frequency sweep with the largest ROI starting at higher frequencies and progressing to lower frequencies (Figs. 7B, 8B). The major axes are shown for the largest, or solitary ROIs in each part of Fig. 8. The major axes provide a useful descriptor of the temporal waveforms of the LFPs produced in the perisaccadic interval.
ROI centroids and spectrogram maxima in the time–frequency plane
The numerous estimates of EMP characteristics obtained from multiple recording locations, and from saccades produced in a variety of different contexts during active vision, can be condensed into a smaller set of numbers that describe these waveforms. Figure 9 illustrates that the EMPs can populate different portions of the time–frequency plane. Here the centroids and maxima split largely over recording site. Figure 9A maps the time–frequency positions of the centroids of the largest ROIs extracted from the spectrograms obtained from the left and right occipital lobes during one recording session. In Fig. 9B the maxima of the spectrograms are mapped. For both Fig. 9A and Fig. 9B, the black disks are values extracted from left occipital cortex spectrograms, whereas the white disks are obtained from the right occipital cortex. Each disk is from a spectrogram calculated from a unique set of LFPs defined by one of the analysis groups (see METHODS, Grouping of saccades for analysis, and Fig. 2). For the sake of clarity, the disks are not labeled with analysis group designators. In Fig. 9, C and D similar maps are shown for the data obtained from a right occipital lobe (V2 location in black) and right temporal lobe (TE location in white) pair recording. Examples of LFPs, EMPs, spectrograms, and ROIs drawn and composed from this data set appear in Figs. 4, 5, 6, 7, 8. Figure 9C is a map of centroids of the largest ROIs in the spectrograms and Fig. 9D maps the spectrogram maxima.
|
Other spatial relationships exist in these data and those extracted from the other recordings. An analysis of clustering in the time–frequency plane appears below in ROI centroids and maxima in the time–frequency plane: measures of clustering. We will show (see Fig. 9) that if the 2 channels in the data sets are combined, distinct clusters of centroids and maxima in the time–frequency plane can be found that are formed around the direction of the saccade. The more dramatic separation of centroids (Fig. 9C, and to a lesser extent Fig. 9A) and maxima (Fig. 9B) based on recording location, apparent from visual inspection of the positions of the black and white disks, is also reinforced by the quantitative analysis of the spatial relationships in the time–frequency plane. We will also show (Figs. 10 and 11) that if we base our analyses on the time or frequency positions of the centroids and maxima alone, a separation based on saccade direction will not be apparent, on average, for the occipital lobe EMPs, but will be more evident for those recorded in the temporal lobe.
Response timing and frequency content of EMPs
In Figs. 10 and 11 and in Table 1 we show that occipital and temporal lobe EMPs can be differentiated based on their timing and frequency content.
|
Figure 10 summarizes the positions of the centroids of the largest ROIs in the spectrograms, in time, with respect to saccade onset, and in frequency, for all the analysis groups, runs, and recording sessions obtained for this study. Saccade directions are referenced to whether they are contraversive or ipsiversive to the recording site. Reorganizing the saccade directions by body-centered coordinates (leftward or rightward) made no significant changes to the distributions for the occipital lobe. For the temporal lobe results, no recordings were made in the left hemisphere; consequently, contraversive is synonymous with leftward and all ipsiversives are rightward saccades. The total number of centroid estimates used to produce the distributions is given in the title of each relative-frequency time histogram.
For both contraversive and ipsiversive saccades, signal power of the EMPs is centered in the postsaccadic phase of the perisaccadic interval. The range between 75 and 150 ms after saccade onset contains about 67% of the centroids associated with contraversive saccades in the occipital lobe (Fig. 10A). That same range includes about 61% of ipsiversive occipital lobe EMP centroids (Fig. 10B). The mode of both distributions is 100 ms with similar means, 73 ms for contraversive, and 78 ms for ipsiversive saccades.
For temporal lobe EMPs, the distributions of the centroid latencies (with respect to saccade onset) of the largest ROIs are more dissimilar when comparing across direction of saccade. In Fig. 10C, only 30% of the centroids are clustered between 75 and 150 ms. This percentage inflates to 41% if the interval expands to 175 ms. On the other hand, for ipsiversive saccades in the temporal lobe, 53% of the centroids lie in the interval including 75 and 150 ms, and 65% are in the 75- to 175-ms postsaccadic interval. Correspondingly, the modes and means of the 2 distributions are dissimilar: modes of 0 and 150 ms for contraversive and ipsiversive saccades, respectively; means of 41 and 85 ms for contraversive and ipsiversive saccades, respectively.
The centroid distributions in time and frequency for the occipital lobe EMPs are significantly different from those compiled for the temporal lobe. Visual inspection of the temporal frequency histograms in Fig. 10 (bottom boxes in A–D) suggests that the distributions are skewed toward higher frequencies in the occipital lobe than they are in the temporal lobe. This result agrees with the picture that emerges from Figs. 7 and 8 that shows elongation of spectrogram ROIs into higher frequencies for occipital lobe EMPs. This elongation raises the centroids of the occipital lobe ROIs in the frequency dimension above those found in many temporal lobe spectrograms. Note that even though ipsiversive saccades produce EMPs in the temporal lobe that are more localized in time (Fig. 10D), their frequency content is about the same as that for the more temporally delocalized contraversive temporal lobe EMPs.
To more objectively evaluate the uniqueness of the centroid distributions presented in Fig. 10, the 2-sample Kolmogorov–Smirnov (KS2) test was performed (KSTEST2 in MATLAB v6.1) on each pair of distributions in the figure. The time and frequency coordinates of the centroids were evaluated separately. The hypothesis that any 2 of the centroid ensembles, such as the time values for contraversive occipital lobe centroids and their ipsiversive counterparts, were drawn from the same distribution was rejected if the P value obtained from the test was <0.025 (2-tailed, 95% confidence level). The results of these tests are summarized in Fig. 11.
Table 1 presents some of the values useful for characterizing the distributions shown in Fig. 10 (see Table 1 caption for details), and Fig. 11 presents the results of the KS2 tests. The "Tdist" values summarize how the distributions are balanced about saccade onset time (0 ms). "Tdist" is defined as Tdist = [n(pst) - n(pre)]/[n(pst) + n(pre)], where n(pst) and n(pre) are the number of spectrograms that had centroids in the presaccadic and postsaccadic phase, respectively, of the perisaccadic interval. Thus if the number is 1, all of the centroids in the distribution are piled up at 250 ms in the postsaccadic phase; for -1 the concentration is at -250 ms in the presaccadic phase; and for small fractions, the centroids are then more evenly dispersed around saccade onset time. In the columns arranged below the header "Maxima" in Table 1 are values extracted from the distributions of the maxima of the spectrograms. These distributions are not shown.
The conclusions to be drawn from Table 1 and Fig. 11 are that: 1) the direction of the saccade does not have as much influence on the activation time of the EMPs (as measured by the position of the centroids in time) in the occipital lobe recording sites as it does in the temporal lobe sites; 2) most of the activation elicited by the saccade in the local neural population is postsaccadic in the occipital lobe, but can be either presaccadic or postsaccadic in the temporal lobe if the saccade is contraversive, and is mostly postsaccadic for ipsiversive saccades; 3) the EMP waveforms are centered at 100 ms after saccade onset in the occipital lobe for both directions of saccade; 4) temporal lobe EMPs are centered at 150 ms for ipsiversive saccades and at saccade onset (0 ms) for contraversive saccades; 5) there are more EMPs with higher temporal frequency content in the occipital lobe than are found in the temporal lobe.
Similar conclusions can be drawn from an examination of the spectrogram maxima. Overall, the pattern of distributions that are distinguishable from one another is the same as that found with the centroids. However, there is a general shift of the distributions of the maxima to longer latencies and lower frequencies than what is seen with the centroids (Table 1). Thus the trend is that the peak power of the EMPs comes later in the postsaccadic interval and is associated with a slower component in the waveform. A notable exception is evident for the OC spectrograms (Table 1). The maximum power in these EMPs appears 50 ms after saccade onset but the LFP activity in the EMP is centered at 100 ms after saccade onset.
Interactions between visual transients and saccades
As illustrated in Figs. 3 and 4, saccades occur during the ongoing presentation of a sequence of images. Thus the perisaccadic interval used to analyze the EMPs frequently contains a stimulus transition that appears either before or after the saccade. For 2 of the modulation rates (3 and 4 Hz), stimulus transitions can occur in both the presaccadic and postsaccadic phases of the 500-ms perisaccadic interval. Table 2 lists the modes and means extracted from distributions of ROI centroids and spectrogram maxima for EMPs with a visual transient in the presaccadic phase of the perisaccadic interval ("O1" and "T1"), in the postsaccadic phase ("O2" and "T2"), or in both presaccadic and postsaccadic phases ("O1&2" and "T1&2"). Tdist values for the centroid and maxima time distributions are also listed.
|
As determined by KS2 tests (test results are not plotted), the distributions of the times of the occipital and temporal lobe ROI centroids are not differentiated by the appearance of a single stimulus transient in either the presaccadic or postsaccadic interval alone. The distributions of spectrogram maxima follow a pattern similar to that of the centroids: the maximum power in the occipital lobe EMPs is shifted to the postsaccadic phase of the perisaccadic interval regardless of whether a visual transient appears before or after the saccade. This suggests that the visual transient does not interact very strongly with the saccade in the generation of the EMP. As seen in Fig. 6C (bottom panel, rightward saccades), the stimulus transitions move through the saccade-triggered LFPs with little impact on the time course of the voltage fluctuations in the postsaccadic phase.
When 2 stimulus transitions appear within the perisaccadic interval, however, the activation times of the occipital lobe EMPs are distributed more evenly between the pre- and postsaccadic intervals (in Table 2, Tdist = 0.59 for "O1&2" compared with values of 0.79 and 0.69 for "O1" and "O2," respectively). The KS2 tests indicate that the "O1&2" distributions are only significantly different from the "O1" distributions of centroid times (P < 0.01) and frequencies (P < 0.0001). When 2 visual transients appear in a 500-ms perisaccadic interval (3- or 4-Hz m-sequence modulation rates), their appearance within the interval is restricted to a smaller range of times than is possible with the single transients (see SUPPLEMENT 2, Fig. S1). The more regular timing of the double transients may thus allow for reinforcement of the visually evoked activity in the presaccadic phase that then shifts the distribution of the activation times, and the frequency content of the EMPs toward the presaccadic phase and to higher frequencies, respectively. The visual transients appearing in the postsaccadic phase are masked by the postsaccadic concentration of power of the occipital lobe EMPs (see Fig. 10). As a consequence, the "O2" and "O1&2" distributions are not significantly different. In the temporal lobe, the distributions of the centroids and maxima are not significantly influenced by the timing of visual transients: activation times of the EMPs remain more evenly distributed about saccade onset time than those recorded in the occipital lobe EMPs regardless of the timing of the visual transients (Table 2).
Analysis of the orientation and eccentricity of the spectrogram ROIs
In Figs. 12, 13, 14 and in Table 3 we show that there are significantly more occipital lobe EMPs that exhibit FM than are seen in the temporal lobe.
|
