I

excess of oxyhemoglobin 1

magnetic susceptibility decreases discrimination of intervals in the millisecond range (Harrington et al., 1998; Jueptner et al., 1995; Maquet et al., 1996). However, only a handful of studies have examined timing of intervals longer than a few seconds. In addition, all of the studies mentioned above either used short durations or applied neuroimaging methods that lack the combined spatial and temporal resolution of FMRI.

Below, I present three experiments to illustrate some possible approaches to analysis of FMRI data that also provide converging evidence for the importance of the right putamen in interval timing. In all three experiments, participants were instructed not to count (for a discussion of this issue, see Hinton and Rao, 2002).

17.4.1 Experiment 1: FMRI of a Basic Peak-Interval Procedure Task

The first FMRI study of interval timing (Hinton et al., 1996) used the peak-interval procedure (Rakitin et al., 1998). This experiment tested reproduction of a single duration of 11 sec. Five fixed-time training trials were presented to participants immediately before each 4-min imaging run. The runs were composed of eight repetitions of a basic trial structure in which 10-sec intertrial intervals (ITIs) alternated with 20-sec signal periods that were timed, as illustrated in Figure 17.2. Four 7-mm axial echo-planar images were collected every second. The bottom surface of this volume was aligned with the horizontal plane encompassing the anterior and posterior commissures. The six male participants made two motor responses on each trial with their right hand: one when they felt that the 11-sec criterion time was near and another when they estimated that the 11-sec criterion time had elapsed. Trial types were distributed according to a Latin-square design, with one dimension being auditory and visual stimulus modalities and the other dimension foreground and background conditions. The foreground condition refers to the standard trial type described above, whereas the background condition is the reverse: the signal to be timed began when the stimulus went off, and the ITI was defined by the presence of the stimulus. The purpose of this manipulation was to attempt to minimize the contribution to the brain maps of sensory components, which should roughly cancel

30-second trial

Signal on off motor responses _+ +

11-second criterion time r

10-second ITI 20-second signal

FIGURE 17.2 Illustration of the basic trial structure of Experiment 1 with 20-sec signal periods and 10-sec intertrial intervals. Participants were instructed to make two motor responses bracketing the criterion time of 11 sec.

out when the two trial types are averaged. The use of two different stimulus modalities to present the timing signal was intended to minimize modality-dependent processes, as the internal clock was assumed to be modality independent (however, see Penney, this volume).

The formal method to perform a group analysis of activated neural regions is to transform functional data from individual participants into a standard coordinate system, such as Talairach space (Talairach and Tournoux, 1988), and average the voxel-wise time series. Technical limitations prevented such an analysis for experiment 1. However, a simpler approach was used in an attempt to derive qualitatively similar data. Four participants completed the full study design, generating two runs of each of the four trial types. Activation maps were generated from these four participants by correlating their FMRI time course data with a square wave representing the period from 8 to 11 sec after signal onset. This particular period was chosen because the earliest responses typically occurred around 8 sec after signal onset; therefore, motor response activation should peak beyond 12 sec due to the hemodynamic delay. Individual correlation maps of activation during this period and structural brain images were aligned with each other. The correlation maps were spatially smoothed with a 3 x 3 center-weighted, spatial convolution and averaged across participants. The composite image was then thresholded at a correlation coefficient of 0.15, which is fairly conservative due to the smoothing and interparticipant averaging that preceded the thresholding step. One of the most reliably activated areas across all participants was the right putamen, as shown in Figure 17.3 (for additional details on the lateralization of interval timing, see Pouthas, this volume). This finding is consistent with drug and lesion data in rats and with this region's pivotal role in interval timing (e.g., Cohen et al., 2000; Gibbon et al., 1997; Matell et al., 2000; Meck, 1996).

As mentioned, averaging across participants' FMRI data is ordinarily performed by transforming each brain into some standardized space (e.g., Talairach and Tournoux, 1988). However, for purposes of visualizing activation in the basal ganglia and other centrally located structures, one may simply align participants' brains manually and then average across them, as was done here. While this is not the preferred method, it was necessary with the data from experiment 1 because software tools for proper averaging across participants were not widely available. The practical justification for such a procedure is that there is very little anatomical variability across brains in these central structures, so true interparticipant averaging would confer little advantage here.

Figure 17.4 illustrates this point with a montage of axial slices ranging from -70 to +75 mm in Talairach space with successive images separated by 5 mm. The montage is an average of the anatomical data of 14 male participants whose brains were aligned with respect to the anterior and posterior commissures and the longitudinal fissure. Note the fairly precise definition of the subcortical brain structures, such as the caudate, putamen, and thalamus. Many of the main white matter tracts also appear fairly distinct, primarily because of their size and more central location. However, there is substantial blurring apparent in the more distal cortical areas, where anatomical variability is greatest. Thus, aligning participants' brains with this

FIGURE 17.3 (See color insert following page 438.) Composite images from Experiment 1 showing 7-mm axial slices ranging from 0 to 21 mm above the AC-PC line with overlaid timing-related activation occurring during the 3 sec preceding the criterion time of 11 sec. Color scale indicates correlation coefficients between brain activation and a square wave covering the period described.

FIGURE 17.3 (See color insert following page 438.) Composite images from Experiment 1 showing 7-mm axial slices ranging from 0 to 21 mm above the AC-PC line with overlaid timing-related activation occurring during the 3 sec preceding the criterion time of 11 sec. Color scale indicates correlation coefficients between brain activation and a square wave covering the period described.

FIGURE 17.4 (See color insert.) Brains of 14 male participants were aligned along the AC-PC line and the longitudinal fissure and averaged to produce this axial montage of 5-mm slices ranging from z = -70 to +75 mm in Talairach coordinates.

method makes it possible to generate a reasonable average activation map, although it is more faithful toward the middle of the image. The likelihood of cortical activations aligning across participants is reduced because anatomical variability is greatest toward the edges of the brain.

Thus, this analysis approach produces a statistical threshold that varies spatially across the brain. It is approximately accurate for central structures that are well aligned, but it becomes increasingly more conservative than the nominal threshold for more distal structures, such as the cortex. Determining in a quantitative manner how the threshold varies with distance from medial structures is probably not possible, but use of this method means that cortical activations are less easily detected.

17.4.2 Experiment 2: Whole-Brain Imaging and Group Averaging

A later study similar in design to experiment 1 used whole-brain imaging and true group averaging across nine participants (unpublished data). The durations tested in this study were 7 and 17 sec. Unlike in experiment 1, participants made only a single motor response with their right index finger on each trial. Two fixed-time training trials were presented immediately before each block of ten testing trials. Eight blocks of each duration were presented in an alternating fashion, and the order of presentation was counterbalanced across participants. The signal duration of the testing trials for the 7-sec duration was 27 sec with a 3-sec ITI, and for the 17-sec duration it was 37 sec with a 3-sec ITI. Twelve 10-mm axial slices were collected every second.

All FMRI data were processed using AFNI (Cox, 1996). The collected volumes were motion corrected, transformed into standard stereotactic space (Talairach and Tournoux, 1988), and blurred with a Gaussian kernel of 4 mm full width, at half maximum to account for anatomical variability. A voxel-wise analysis of variance (ANOVA) was performed on the 7- and 17-sec data across participants for the period from 0 to 6 sec before the response time. The mean response times were 8.7 sec for the 7-sec task and 22.0 sec for the 17-sec task. The period before the response time was selected to minimize contamination by motor-related activation. The t-statistic map was thresholded at P < .0001. A cluster threshold of 100 |l and a nearest-neighbor threshold of 1 mm were applied to eliminate small, discontiguous activations. The result was rendered onto a three-dimensional brain with opacity set to 0.5. The data are displayed as a montage of 5-mm axial slices, as shown in Figure 17.5.

The montage shows several distinct activation foci within the caudate, putamen, and globus pallidus. Also apparent are midline activations of the supplementary motor area, anterior and posterior cingulate cortex, and precuneus, as well as bilateral activations of the insula. Right hemisphere foci include the middle frontal gyrus and inferior parietal lobe, which are areas that have been associated with attentional networks in general (Pardo et al., 1991; Posner and Peterson, 1990) and with attention to time in particular (Rao et al., 2001). Finally, there is some activation in left primary motor cortex, an unexpected finding.

FIGURE 17.5 (See color insert.) Axial montage of activation from Experiment 2) during the period from 0 to 6 sec before the response time when 9 participants timed a 7- or 17-sec duration (P < .0001).

17.4.3 Experiment 3: ROI Analysis and a Motor Control Task

Another FMRI study similar to experiment 2 used a region-of-interest (ROI) analysis and two different durations to examine whether activation in the putamen could be reliably separated from that due to the motor response (unpublished data). The trial structure was the same as that in experiment 2. Each participant alternately performed a temporal reproduction task (similar to the one in experiments 1 and 2) and a motor task in which the timing of the response was cued to occur at an actual response time the participant had generated in the previous timing run. The timing and motor tasks alternated eight times in a session. In the motor control task, the attentional and sensory components of the task were similar, but participants were not required to time. They were instructed to attend to the central fixation cross and to respond when they saw it flash briefly. In both tasks, participants responded by pressing a button with their right index finger. An 11-sec duration was tested with six participants, and a 17-sec duration was tested with one participant. Twelve 10-mm axial slices were collected every second, and the imaging volume included the hand area of the sensorimotor cortex and the basal ganglia. The imaging data were motion corrected, and ROIs were traced for each participant on each slice for the left and right primary motor cortex (M1), primary somatosensory cortex (S1), and putamen. The motor response for each trial was time-locked to the criterion time, and like trial types were averaged across participants to extract time courses of activation for each ROI. Because of the relatively small number of participants and in order to better visualize differences by task and hemisphere, each time course was temporally smoothed using a 3-point moving average. The results of this study are displayed in Figure 17.6.

Activation in Ml for the 11-sec duration is shown in Figure 17.6a. The left M1, contralateral to the effector, was activated for both the timing and motor tasks to a similar extent and concurrently beginning just after the 11-sec criterion time. Figure 17.6b shows a similar pattern for the left primary sensory cortex. A smaller ipsilateral activation is apparent in both M1 and S1. In contrast, the pattern of activation in the putamen is distinctly different. Figure 17.6c shows that activation for the timing task begins in both the right and left putamen some seconds before the 11-sec criterion is reached, although the activation is much stronger in the right putamen, which is ipsilateral to the effector and therefore should not be engaged by motoric aspects of the task. For the motor task, the timing of activation in the putamen looks similar to the time courses in M1 and S1. However, whereas the response in M1 and S1 is highly lateralized to the contralateral left hemisphere, activation in the putamen is less lateralized and, if anything, is stronger in the ipsilateral right putamen. Another point to note is that the timing of the response in the putamen to the motor task looks as though it may be slightly delayed relative to those in S1 and M1, similar to their ipsilateral activations. This suggests the possibility of some kind of delayed feedback effect from the sensorimotor system to the putamen. The response in the putamen for the 17-sec criterion time is shown in Figure 17.6d. With this longer duration, it is clear that the putamen activation begins substantially before the motor response occurs for the timing task, and again the activation is stronger in the right than in the left putamen. For the motor task, the activation looks more like the hemodynamic response in the left M1 and S1.

Overall, what is striking about these data is how the time courses in M1 and S1 are more differentiated by hemisphere, whereas the putamen activation is more differentiated by the task and the timing of the activation. One hypothesis is that the putamen, and especially the right putamen, signals other areas of the brain in circumstances in which a precisely timed motor response is required to estimate when a particular short interval has elapsed. Whatever the exact role played by the putamen in interval timing, its time course of activation when the participant is timing a signal is clearly separable from and anticipates those of the motor and sensory cortices.

17.4.4 Timing May Be Obligatory

Indirect evidence for the involvement of the right putamen in interval timing comes from a study of motor sequencing (Harrington et al., 1999). In this experiment, participants were cued to make a series of specific finger movements within a 12-sec activation period. These motor periods were followed by a 12-sec period of rest. These two states alternated repeatedly in blocks. When the authors contrasted the activation periods with the rest periods, they observed a consistent pattern of BOLD signal in striate and sensorimotor cortices and the cerebellum. However, somewhat

FIGURE 17.6 (See color insert.) Time courses of averaged activation from Experiment 3 extracted from region-of-interest analyses of (a) the primary motor cortex (M1), (b) the primary sensory cortex (S1), and (c) the putamen for an 11-sec signal. Panel (d) shows data from the putamen for a 17-sec signal. Dashed and solid lines respectively indicate activation measured from right and left hemisphere structures. Lighter and darker colors respectively indicate time courses of activation for the motor and timing tasks.

FIGURE 17.6 (See color insert.) Time courses of averaged activation from Experiment 3 extracted from region-of-interest analyses of (a) the primary motor cortex (M1), (b) the primary sensory cortex (S1), and (c) the putamen for an 11-sec signal. Panel (d) shows data from the putamen for a 17-sec signal. Dashed and solid lines respectively indicate activation measured from right and left hemisphere structures. Lighter and darker colors respectively indicate time courses of activation for the motor and timing tasks.

FIGURE 17.7 (See color insert.) Area in the right putamen of decreased MR signal intensity (t < -1.55) comparing a repetitive motor sequence task with rest (shown in coronal, axial, and sagittal views from left to right). This was the only area of decreased signal intensity reported in the experiment of Harrington et al. (1999). (Figure created with kind permission from D.L. Harrington and S.M. Rao.)

FIGURE 17.7 (See color insert.) Area in the right putamen of decreased MR signal intensity (t < -1.55) comparing a repetitive motor sequence task with rest (shown in coronal, axial, and sagittal views from left to right). This was the only area of decreased signal intensity reported in the experiment of Harrington et al. (1999). (Figure created with kind permission from D.L. Harrington and S.M. Rao.)

perplexingly, there was one brain region that was negatively activated, as shown in Figure 17.7: "Activation was greater during rest than repetitive sequencing in the right putamen, an unexpected result" (Harrington et al., 1999). The authors had no explanation for this observation, but when their experiment is viewed from the perspective of interval timing, their finding makes perfect sense. Participants were performing the complex motor movements during the activation period, as they had been instructed. During that period, their attention was fully engaged by the task. However, during the rest periods, there may not have been much else to occupy participants' minds other than predicting the onset of the next activation period. Thus, the individuals in this experiment may have inadvertently engaged in a timing task, albeit without any explicit instruction to do so. These data are consistent with a perceptual role of the putamen for precisely timing short intervals, as described for the timing task in experiment 3. The surprising finding of this motor sequencing study illustrates how pervasive interval timing behavior can be whenever there is temporal regularity in an organism's environment.

0 0

Post a comment