Neuroimaging And The Time Measurement System

We, like many others, are using functional magnetic resonance imaging as a tool to study human timing. Our survey of the literature shows that more than 25 imaging papers of interval timing have been produced so far, and like any other topic to which neuroimaging has been applied, we expect many more to follow in the next few years. Therefore, in this section we aim to discuss some of the conceptual limits to the imaging of human timing and explore ideas about what these studies may be expected to achieve. In the limit, any single functional imaging technique on its own (whether fMRI, PET, MEG, or EEG) is unlikely to be sufficient. As the bulk of this book has demonstrated, these imaging techniques must be complemented by patient studies, lesion experiments, drug interventions, and electrophysiological recording studies, spanning the range from system to cellular analyses. Nevertheless, neuroimaging by itself will prove an important tool (see Hinton, this volume; Pouthas, this volume).

21.3.1 The Functional Anatomy of Time Measurement

The first level of imaging analysis is to simply identify the areas involved in timing tasks (see Hinton, this volume; Meck, this volume; Meck and Benson, 2002; Sakata and Onoda, this volume). In many of the studies we have reviewed, this is achieved by using block analysis of timing tasks contrasted with nontiming control conditions. In these studies, activity is measured in blocks of 30 or more seconds at a time, so there is no real temporal resolution to the data. More specific localization of timing components can be achieved with event-related imaging techniques, but there are also clear limits on this technique, as we will describe below. The second level of attack is the use of imaging techniques to explore interactions between the timing subsystems or to approach the neural mechanism of their functions. At this level, we must consider whether the operation of each component in a time measurement system depends on neural mechanisms that we can actually detect. The most basic measures afforded by functional imaging studies are the changes in the activity of neural populations from one moment to another. In PET and fMRI, these are detected using the resulting changes in local blood flow or oxidation; thus, if a component of the time measurement system does not cause a significant change in metabolic cost, we may not detect its presence.

The most obvious example of this problem is the time-dependent process or "clock" central to the timing system, perhaps a "ticking" oscillator or similar circuit: if the clock is always ticking, but other components (e.g., the accumulator) only intermittently use its output, it may be very difficult to detect this process using neuroimaging. One solution may be to selectively speed or slow the clock, independent of all other neural processes (Meck, 1996), and detect the changing activity that correlates with these alterations. However, it is possible that a neural clock circuit could be accelerated or slowed without leading to gross change in metabolic load: if the duty cycle (active to inactive states) is kept constant, then the main metabolic costs (e.g., dendritic processing and some contribution to ionic pumping across the membrane after spike activity) could be nearly identical in a cell or a circuit oscillating slowly or rapidly. Because we do not yet know what form of clock ticking, if any, is used in the timing process, we cannot predict whether the changes in neural activity associated with changes in clock speed would be imageable.

MEG and EEG techniques complement PET and fMRI with regard to temporal precision, as they can detect neural activity in the millisecond range (for a discussion of how EEG and PET techniques can be used to inform each other, see Pouthas, this volume). Hence, for example, these techniques would be invaluable for detecting a rhythmically active clock, as they could differentiate between the signals of different clock rates. However, these techniques also have their limitations, as both depend on the synchronous activity of a group of aligned neurons (or rather, their dendritic processes) and are insensitive to currents in tissue that are oriented perpendicular or tangential to the scalp respectively. They are also insensitive to deep brain sources. One could certainly imagine time measurement processes that would be invisible to MEG or EEG.

21.3.2 Imaging the Timer Components

Bearing knowledge of the limiting characteristics of neuroimaging techniques in mind, let us think about the basic components of the scalar expectancy theory (or scalar timing) model (see Church, this volume; Gibbon et al., 1984) and ask how we can identify the mechanisms and the neural loci of each. The various components are the time-dependent process (the pacemaker), the local memory stores (the accumulator and the reference memory), and the comparator, as well as sensory input and modulatory output systems. Temporal information processing would also include the attentional system and the cognitive output structures or the motor systems using information from the timer.

For much of the imaging literature, the sensory input systems are treated as items of secondary interest. PET and fMRI depend on contrasting different behavioral states, and thus any process in common to the two states is not visualized. Hence, it is typical to attempt to balance the contribution of systems of secondary interest between the timing task and the control (baseline) task. Sensory inputs or motor outputs, if carefully balanced, do not confound the final results of the imaging study. However, this strategy has the implicit danger that it may obscure data suggesting that the timing functions actually depend on the specific sensory structure. Thus, if the time-dependent process is active from the start of the sensory stream, then it will be nearly impossible to distinguish between these two using functional neuroimaging. However, because interval timing is easily achieved across the gaps between delimiting stimuli, or by using stimuli in different modalities, this should not pose a real problem for the investigation of non-sensory-specific timing systems. The possibility that parts of the motor system may be obligatory components of some timing operations is less easy to dismiss.

In this vein, we have discussed above and other authors in this book have highlighted the fact that motor areas of the brain (cerebellum, basal ganglia, and premotor cortical areas) are strong candidates for involvement in interval timing tasks (see in this volume, Diedrichsen et al.; Hinton; MacDonald and Meck; Malapani and Rakitin; Matell et al.; Pang and McAuley). If these circuits are recruited only for some timing operations, such as those in which repetitive motor outputs are needed (e.g., rhythmic tapping), then separation of motor timing and motor execution becomes very difficult. The wealth of evidence suggesting that imagined movement or mental rehearsal does activate the motor system compounds this difficulty because implicit use of motor systems to measure time, even without active movement, could cause neural activity. In the limit, we should perhaps ask if the attempt to separate timing from movement is sensible, if indeed the movement, or its internal rehearsal or planning, is what is actually used as the timing signal.

To approach this problem, it would be useful to know whether different motor timing circuits were selectively recruited for specific timing tasks. It seems likely that the neural operations involved in selecting or recruiting pattern-generating circuits during the first epoch of repeated, subsecond interval measurements would be detected by current functional imaging techniques. We have evidence (Lewis et al., in preparation) that this is the case, as areas known to be involved in movement selection show activity at the onset of different rhythm epochs, but appear inactive during the immediately following rhythm production. We believe that timing circuits are therefore actively recruited, or adjusted to the target intervals, but then continue to cycle with little additional cost.

Let us think about what this observation might mean at a finer level. In an earlier model of the neural mechanism of timing (Miall, 1989), it was suggested that different neural oscillators could be selected and combined to provide an interval timing system. Only the weighted output of the multiple oscillators could be said to encode a specific interval: many oscillators were active in each interval, and the selectivity of the system was generated by synaptic weighting of a subset of these to some output unit. Hence, the activity of this output unit, excited at the critical moment by the synchronous activity of its oscillating inputs, would easily be detectable, but the ongoing activity of the population of oscillators would not (for an extension of this model, see Matell et al., this volume; Meck, this volume). Miall (1989) proposed that additional neural machinery might be used to synchronize the oscillators at the start of each timed interval, but beyond that, the system could freerun with no additional metabolic cost. Again, this suggests that the neural activation required to start, select, or synchronize the oscillator system might be imageable, but its ongoing activity would be hard to detect. If the oscillator population itself became active at the start of each interval, from an inactive state, this should also be detectable, but this scenario seems unlikely.

The accumulator as described by Gibbon et al. (1984) is probably the most easily detectable timing element, as by definition its activation changes throughout each interval and must be reset. A naïve viewpoint might therefore be that the bulk of the imaging data produced so far reflects the activation of this accumulator circuit. However, using carefully designed baseline conditions, it should be possible to dissociate the accumulation process from related events such as the comparison or decision processes. The reference memory store in which the previous intervals are recorded might also seem easy to image, as it would accumulate traces of the previous intervals, changing with experience of the target interval. It is striking to us that cortical prefrontal areas are prominently active in cognitively controlled timing tasks: these may be the systems in which a trace of activity is set up and changes throughout the timed interval. Overlap of the observed regions with areas known to be involved in working memory is also important.

In contrast, it is likely that EEG and MEG techniques would be poor for studying the accumulator activity. A basic model of the accumulator (Miall, 1993) and a recent, more elaborate model (Koulakov et al., 2002) suggest that it may be made up of a population of independently active cells, and thus would not have the synchronous behavior necessary to cause a large signal. Some MEG analysis techniques have made use of the switch from synchronous activity in the idle state to desynchronized activity in an active state (Singh et al., 2002), and this could prove useful.

Lastly, the comparator function would appear to be difficult to detect. As a singular event at the end of each interval, comparison would contribute rather little to the overall signal within a typical block design imaging study. Event-related imaging designs would allow temporal separation of different events within the timing task if their occurrence could be varied with respect to each other. In such studies, the blood oxygenation signals are correlated with specific event times, for example, with the onset and offset of each interval, as long as these events are themselves uncorrelated (Buckner et al., 1996). Unfortunately, the comparator process will almost always be time-locked to other events, such as the initiation of whatever action is required at the end of the trial. For example, if the subject was asked to respond at the end of the target interval, the comparison operation, the transfer of that interval to reference memory, the resetting of the accumulator, possibly the stopping of the pacemaker clock, and the initiation of the response would all be very close in time. Better temporal differentiation using MEG or EEG, where events that cannot be decoupled could be temporally ordered, might provide a solution to this problem if it were clear what the order of their occurrence must be (see Sakata and Onoda, this volume).

0 0

Post a comment