Cerebral Cortex, Vol. 12, No. 7, 710-728,
July 2002
© 2002 Oxford University Press
Rapid Distributed Fronto-parieto-occipital Processing Stages During Working Memory in Humans
Institut National de la Santé et de la Recherche Médicale, CJF90-12, Neurologie, CHU Pontchaillou, F-35033 Rennes, France
E. Halgren, MGH NMR Center, 149 13th Street, Charlestown, MA 02129, USA. Email: halgren{at}nmr.mgh.harvard.edu.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Cortical potentials were recorded from implanted electrodes during a difficult working memory task requiring rapid storage, modification and retrieval of multiple memoranda. Synchronous event-related potentials were generated in distributed occipital, parietal, Rolandic and prefrontal sites beginning
130 ms after stimulus onset and continuing for >500 ms. Coherent phase-locked, event-related oscillations supported interaction between these dorsal stream structures throughout the task period. The Rolandic structures generated early as well as sustained potentials to sensory stimuli in the absence of movement. Activation peaks and phase lags between synaptic populations suggested that perceptual processing occurred exclusively in the visual association cortex from 90 to 130 ms, with its results projected to fronto-parietal areas for interpretation from 130 to 280 ms. The direction of interaction then appeared to reverse from 300 to 400 ms, consistent with mental arithmetic being performed by fronto-parietal areas operating upon a visual scratch pad in the dorsolateral occipital cortex. A second reversal, from 420 to 600 ms, may have represented an updating of memoranda stored in fronto-parietal sites. Lateralized perisylvian oscillations suggested an articulatory loop. Anterior cingulate activity was evoked by feedback signals indicating errors. These results indicate how a fronto-centro-parietal ‘central executive’ might interact with an occipital visual scratch pad, perisylvian articulatory loop and limbic monitor to implement the sequential stages of a complex mental operation. |
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Working memory maintains information for use in on-line cognition. As conceptualized by Baddeley, working memory is composed of a central executive controlling ‘slave systems’, primarily a ‘visuospatial scratch pad’ and an ‘articulatory loop’ (Baddeley, 1986
). Working memory thus combines a mental workspace with short-term buffers and executive processes for interpreting incoming stimuli, updating actively stored items accordingly, and maintaining the intermediate results of processing for further calculations. Working memory has become a key concept in many models of higher cognitive activities such as reading (Just et al., 1996). The amount of resources in working memory and the efficiency of their dynamic allocation are seen as the main factors underlying limitations in fundamental intellectual abilities (Kyllonen and Christal, 1990).
In some working memory tasks, a single stimulus is presented, then retained as presented for 2–30 s and finally retrieved. Such tasks probe what has been termed primary memory (James, 1890) and are amenable to neurophysiological analysis because the input, retention and retrieval stages are separated temporally. In contrast, other working memory tasks require a rapid interaction between perception, interpretation and memory. Previous studies have used positron emission tomography (PET) or functional magnetic resonance imaging (fMRI) for measuring haemodynamic or metabolic activation during working memory. Both varieties of working memory task activate an extended network of the prefrontal, parietal and medial frontal cortices (Swartz et al., 1995; Fiez et al., 1996; Smith et al., 1998; Ungerleider et al., 1998).
Haemodynamic and neuropsychological studies suggest that it may be possible to distinguish the contributions that these different areas make to working memory. Consistent with a long tradition in neuropsychology, the prefrontal cortex has been associated with executive function (Baddeley, 1986). In some haemodynamic studies, the more ventral prefrontal areas have been associated with the simple retention of material in primary memory, whereas the more dorsolateral areas have been engaged when the task requires a higher level of executive control (Petrides et al., 1993; D'Esposito et al., 1995; Cohen et al., 1997; Owen, 1997; Smith et al., 1998). Other studies have found the location of prefrontal activation to depend upon the nature of the material to be retained, with verbal/non-verbal material mapping to the dominant/non-dominant hemisphere and spatial/ non-spatial material mapping to the dorsal/ventral areas (Smith et al., 1998; Ungerleider et al., 1998). Similarly, lesion studies have suggested that the parietal cortex (in particular near the intraparietal sulcus) is specialized in spatial analysis (Critchley, 1953). Conversely, the left perisylvian activation is associated with the maintenance of information in the articulatory loop according to both neuropsychological (Shallice and Vallar, 1990) and haemodynamic evidence (Fiez et al., 1996). Finally, anterior cingulate activation has been associated with error monitoring and stimulus response mapping (Bush et al., 2000).
These studies support hypotheses associating different cortical areas with different working memory components. However, the task/area double dissociations predicted by these hypotheses are often difficult to obtain. In particular, parietal and anterior cingulate activations in working memory tasks tend to be correlated with prefrontal activation across task conditions (Carpenter et al., 2000; Diwadkar et al., 2000) and a strict correlation of working memory deficits with prefrontal lesions has not been found (Frisk and Milner, 1990). Co-activation of a parieto-fronto-cingulate network has also been found in other tasks, in particular those involving attention (Corbetta, 1998) and task difficulty/self-monitoring/response choice (Cohen et al., 2000). Similarly, neuroanatomical and neuropsychological evidence supports an integrated parieto-fronto-cingulate circuit in executive functions (Goldman-Rakic, 1988; Mesulam, 1990). Finally, efforts to isolate cortical areas that are exclusively activated by the necessity of controlled allocation of attentional resources in dual task situations have yielded inconsistent results, suggesting that high-level executive functions may arise through the interaction of multiple structures (Adcock et al., 2000).
One interpretation of these data is that, although the parietal and prefrontal cortices make distinct contributions to complex processing (as revealed by lesions), they are so interconnected that their activation patterns are nearly indistinguishable. Alternatively, it is possible that the different structures not only have distinct functions but that they are activated sequentially and this cannot be seen with haemodynamic measures because of their poor temporal resolution (>1 s) (Buxton et al., 1998; Dale and Halgren, 2001). That is, although these areas are coactivated on a time-scale of 1 s and belong to the same general system, it is conceivable that their activation is modular and sequential rather than continuously interactive and overlapping.
The spatiotemporal dynamics of cortical activation during working memory also have implications for whether the prefrontal cortex acts in working memory as a repository of primary memory, as opposed to as the central executive. As primary memory, specific and sustained firing of prefrontal neurons would encode significant memoranda as well as the cognitive context. This sustained firing would function not only as a memory store but also as a source of task-specific instructions and temporal markers for the posterior association cortex, thus freeing it from immediate sensory input (Goldman-Rakic, 1988; Fuster, 1989; Halgren, 1994). Alternatively, as central executive, the prefrontal cortex would provide goal-directed sequencing and control of the posterior association cortex. Initially, this control directs attention to a sensory channel by activating the associated cortical area (Desimone, 1996; Hillyard and Anllo-Vento, 1998; Mangun et al., 1998; Knight et al., 1999). Subsequent executive control can be conceptualized as an extension of this process, with the prefrontal cortex successively activating cortical areas concerned with perceptual processing, interpretation of the stimulus with respect to task instructions, consequent mental operations on active memoranda, memory updating and response selection. These central executive functions would imply phasic activation of prefrontal neurons corresponding to the successive processing stages.
A related question is whether primary memory is held in the prefrontal cortex or the posterior sensory association cortex or both. If it is only in the prefrontal cortex, then sensory association areas may complete their work soon after stimulus presentation, pass the results to anterior areas and then no longer be active. Conversely, the model mentioned above (Baddeley, 1986) suggests that sensory association areas are involved in later stages simultaneously with higher cortical areas, in particular in the prefrontal cortex.
In summary, an adequate neural model of working memory should specify not only which structures are involved, but also the dynamics of their interaction. The most extensive source of data regarding these dynamics is unit-recording studies in macaques during working memory (Goldman-Rakic, 1988). Sustained stimulus-specific firing by prefrontal neurons supports the hypothesized prefrontal contribution to primary memory. However, phasic firing patterns are also observed in prefrontal neurons and, conversely, the same categories of task-related firing are observed in temporal and parietal sites, with the time-course of activation in these areas entirely overlapping with prefrontal activation (Fuster, 1989; Chafee and Goldman-Rakic, 1998). Furthermore, the parietal and prefrontal firing patterns are modified but fundamentally maintained during cooling of either location, suggesting that they arise out of the coordinated activity of an extended network (Fuster et al., 1985; Chafee and Goldman-Rakic, 2000). The lack of stronger effects is interpreted as being due to the high degree of redundancy in connections in the generating circuit (Goldman-Rakic, 1988). However, the similarity of activation in different areas that seems to reflect their seamless integration leaves no leverage for explicating their distinct roles (Chafee and Goldman-Rakic, 2000). Furthermore, it is uncertain how to generalize these results to humans in more complicated working memory tasks that engage the central executive heavily.
Direct electrophysiological recordings in humans during working memory would thus be useful in understanding the temporal dynamics of engagement by different cortical areas. Such recordings with scalp electrodes have also been interpreted as indicating continuing interactions between the anterior and posterior cortex (Gevins et al., 1996). However, the cortical generators of these signals cannot be inferred unambiguously (Hamalainen et al., 1993; Dale and Halgren, 2001). Intracranial electroencephalography (iEEG) recordings have high temporal resolution and under some circumstances can also localize generators with high spatial resolution (Halgren et al., 1980). Such recordings can only be obtained from electrodes implanted for clinical purposes and, thus, are limited by possible contamination with pathological activity, as well as clinical limitations in spatial sampling. Nonetheless, they offer a view of the spatiotemporal activation patterns that underlie cognitive events, which is complementary to non-invasive measures.
In the current study, we report the results of intracranial recordings during Mental Counters, a working memory task that requires the rapid perception and evaluation of six successive stimuli interleaved with the maintenance and updating of active memoranda (Larson and Saccuzzo, 1989). The memoranda were digits or figures in different variants that were intended to probe different material-specific ‘scratch pads’. Following the input stimuli, a probe stimulus was presented and then a feedback tone following the response.
The results are most consistent with a sequential but not modular model of cortical function in working memory. Although early processing was briefly confined to the visual association cortex, activation soon spread to multiple occipital, parietal and frontal sites, which all remained active for the entire epoch. In particular, the co-activation of visual association with the fronto-parietal cortices suggests that it contributes to working memory beyond perceptual analysis. The results support the presence of at least four processing stages in working memory, with each lasting 50–200 ms. Phase-locked oscillations from 4 to 12 Hz were prominent in multiple structures including the prefrontal cortex, as would be expected if it contributes to the central executive. Single-trial spectral analysis of the dominant oscillations suggested organized interactions between the areas, with rapidly shifting directions of information flow. Event-related oscillations ranged from 4 to 50 Hz. They were commonly embedded in slower oscillations synchronized with the individual stimuli (1 Hz), which in turn were embedded in sustained activation from the first stimulus to the probe (<0.1 Hz). The latter supports a primary memory contribution of the prefrontal cortex to working memory. Limbic areas, in particular the anterior cingulate gyrus, were preferentially activated by the probe and feedback stimuli. Overall, the results reveal multiple embedded oscillations that may reflect the temporal organization of processing and memory functions across multiple cortical areas during this difficult task.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Patients
Seventeen patients (10 females and seven males) suffering from intractable partial epilepsy consented to participate in this study while undergoing iEEG for possible neurosurgical treatment. All subjects except one were right-handed and that subject was left dominant for language according to the Wada test. Their mean age was 26.2 ± 7.7 years.
Behavioral Tasks
The subjects performed different variants of the Mental Counters paradigm of Larson, which were modified to be appropriate for neurophysiological recordings in patients (Larson and Saccuzzo, 1989). These tests all required the maintenance of different memoranda and their updating in response to rapidly presented stimuli. In the standard numeric paradigm with two memoranda presented in the visual modality (N2V), the location of the stimuli indicated which memorandum was to be operated on and which operation was to be performed on it (Fig. 1). Based on preliminary testing with a control group, NV2 was chosen as the basic task variant to be given to all patients. Of the 17 subjects, 16 performed the N2V variant. Some subjects also performed variants of the Mental Counters paradigm that differed from N2V in material (using figures or letters instead of numbers, i.e. F2V or L2V), modality (auditory rather than visual, i.e. N2A) or memory load (three rather than two memoranda, i.e. N3V). Note that, in all of the visual variants, identical stimuli were used and the tasks were different only in the nature of the memoranda and the operations to be performed upon them. Of the 17 subjects, 10 performed the F2V variant, four the L2V variant, three the N3V variant and only one the N2A variant. Each trial consisted of six stimuli, 36 trials were presented per block and one to four blocks of a variant were obtained in a particular subject. In addition to electrophysiological recordings, reaction times and response accuracy were obtained.
|
The study was described to the subject in the subject's hospital room, usually on the day prior to recordings and informed consent was requested. The subjects then participated in a training session of 45 min during which each variant was initially presented with inter-stimuls intervals (ISIs) of 2500 ms. This was gradually decreased to 1240 ms and finally 960 (nine subjects) or 840 ms (eight subjects) if the patient was capable. Approximately half of the patients receiving training were unable to perform the task at an acceptable speed, presumably due to psychomotor slowing induced by anticonvulsant medications. Physiological recordings were only obtained from the remaining 17 subjects.
Numbers–Two Memoranda–Visual (N2V)
At the beginning of each trial the initial values of the memoranda (three zeros) were presented on the left, right and middle of a 16 in colour monitor. When the subject pressed a key, three zeros were replaced in the same positions by three points. Then at ISIs of 840 ms (240 ms stimulus exposure) a short line appeared. If the line was horizontal (42%) or vertical (42%) then the subject made a subtraction or addition to either the left or right memorandum, according to the location of the target. If the line was above a dot then the operation was addition of one and if the line was below a dot then the operation was subtraction. The new result was maintained in memory to be used as a basis for the next line. Diagonal lines (16%) indicated that no operation was to be performed (i.e. ‘catch’ trials). Six lines were presented in each trial. The location of the lines and their orientation varied randomly on successive presentations with the exception that the correct value of any given memorandum was maintained between –3 and +3. At the end of these six presentations a proposed solution consisting of three numbers was provided on the screen, one at each of the original dot locations. Fifty percent of the proposed solutions matched the correct result. The subject had to press the left key in <3 s if the proposed solution was correct and the right key if it was not. A feedback tone indicated whether the subject's answer was correct (high tone) or not (low tone). Throughout the entire trial the subject maintained fixation on the centre of the monitor, which was occupied by either a dot or a number. The response and feedback procedures as well as all time-intervals and stimulus-type proportions were identical across all variants.
Numbers–Three Memoranda–Visual (N3V)
N3V was identical to N2V, except that the lines could appear above or below the middle point on the monitor, i.e. three rather than two memoranda needed to be maintained in memory and incremented or decremented accordingly.
Numbers–Two Memoranda–Auditory (N2A)
N2A used sounds as the stimuli. As in N2V each trial started with the presentation of three zeros on the left, right and centre of the monitor. The dots then remained on the screen until they were replaced with the response probe. Throughout this period the subject maintained fixation on the centre dot. Adding one to the memorandum was signalled by pure high-pitched sounds, subtraction by low-pitched sounds and catch trials by complex sounds. In order to indicate that the operation was to be carried out on the right-side memorandum, the sound was presented in the right ear and vice versa. After six sounds a solution was proposed on the screen.
Letters–Two Memoranda–Visual (L2V)
L2V used letters as the memoranda. Each trial began with three Ls on the left and right and in the centre of the monitor. These were replaced by dots and then short lines started appearing above or below the left or right dot, exactly as in N2V. Horizontal or vertical lines above dots signalled going ‘up’ a letter (or backwards) in the alphabet (e.g. L to K) and targets below dots signalled going ‘down’ a letter (e.g. L to M). Again, diagonal lines signalled no operation. After six lines were processed a response probe consisting of three letters was presented.
Figures–Two Memoranda–Visual (F2V)
F2V used two figures (left or right) as the memoranda, each consisting of a horizontal line, a vertical line or both (i.e. a plus). Each trial began with three vertical lines on the left, right and centre of the monitor. These were replaced by dots, and then short lines started appearing above or below the left or right dot, exactly as in N2V. The appearance of a horizontal or vertical line above a dot signalled that the corresponding element should be added to the corresponding figure (Fig. 1
). Similarly, the appearance of a horizontal or vertical line below a dot signalled that the corresponding element should be removed. Diagonal lines signalled no operation. After six lines were processed a response probe consisting of three figures was presented.
Stereoencephalographic Recordings and Stimulation
Stereoencephalographic (SEEG) recordings were obtained from a total of five to eight multicontact probes surgically inserted into the depths of one or both cerebral hemispheres of each patient (Fig. 2). The anatomical location of these probes was decided entirely on clinical grounds for localizing the origins of the patients' epileptic seizures (Chauvel et al., 1996a). Typically, the probes remained implanted for 4–7 days in order to obtain recordings during seizure activity. The recordings in the present study were obtained during seizure-free periods. Each probe contained between five and 15 stainless steel contacts, thereby allowing for recordings from different medial to lateral sites aligned perpendicular to the sagittal plane (Fig. 2). The contacts were 2.0 mm in length and were separated by 1.5 mm. Recordings were obtained across all 17 subjects from 667 left hemisphere and 701 right hemisphere sites. The contacts were localized using human stereotaxic statistical data (Talairach et al., 1967; Talairach and Tournoux, 1988) and confirmed with stereoscopic stereotaxic angiography (Szikla et al., 1977) and, in most cases, stereotaxic MRI (Musolino et al., 1990). The probes are lettered (primed in the left hemisphere). The contacts are numbered according to their distance from the midline in millimetres. The contact locations are also indicated in Talairach's coordinates (x = distance to the right of the midline, y = distance forward from the anterior commissure or AC and z = height above the anterior commissure–posterior commissure or AC–PC line) (Talairach and Tournoux, 1988). The actual coordinates are listed first, then normalized in italics. Since the y and z coordinates are constant for all contacts of a given probe, they are omitted where they are clear from the context.
|
In many cases, the contacts were stimulated electrically in order to delineate the origin and spread of seizure discharges better (Chauvel et al., 1996a
). Unless otherwise noted, these stimulations were bipolar between adjacent electrode contacts, with pulses of 1 ms delivered in a train of 5 s at 50 pulses/s at stimulus intensities below that necessary for evoking a local after-discharge (as monitored by simultaneous SEEG recordings). Those stimulations with effects that provided confirmation of functional localizations are presented below. The contacts were located in the following regions (the counts include contacts located in the underlying white matter): the occipital cortex (n = 62), precuneus and superior parietal lobule (n = 15), posterior cingulate gyrus (n = 60), supramarginal gyrus (n = 128), lingual and fusiform gyri at the occipitotemporal junction (n = 30), middle temporal gyrus at the occipitotemporal junction (n = 48), posterior hippocampal formation (n = 70), posterior middle temporal gyrus (n = 105), anterior hippocampal formation (n = 80), middle level of the middle temporal gyrus (n = 117), amygdala region (n = 75), anterior level of the middle temporal gyrus (n = 107), posterior superior temporal gyrus (n = 80), anterior superior temporal gyrus (n = 18), ventromedial prefrontal cortex (n = 71), ventrolateral prefrontal cortex (n = 157), dorsomedial prefrontal cortex (n = 15), ventrolateral prefrontal cortex (n = 24), paracentral gyrus (n = 19) and Rolandic cortex (n = 87) (including the precentral and postcentral gyri and sulci). Scalp recordings were usually obtained at Cz or Pz. The specifics of localization are presented in the context of descriptions of individual responses.
Recordings were digitized on-line with an accuracy of 12 bits. In the initial eight subjects, recordings were made as separate epochs during the six input (I1–I6), probe and feedback stimuli at 6 ms per sample with a bandpass of 0.1–35 Hz. The epochs began 40 ms before the stimulus and ended 794 or 914 ms after the stimulus (for the 840 or 960 ms stimulus-to-stimulus onset intervals, ISIs, respectively) for the I1–I6 and probe stimuli. The epochs began 190 ms before the stimulus and ended 644 or 764 ms after the stimulus for the feedback stimuli. Data were acquired continuously in the last 10 subjects during the entire task at 200 Hz and a bandpass of 0.5–50 Hz. This bandpass may have reduced sustained potentials related to primary memory. Eye movement artefacts (>75 µV) were monitored with an electrode placed at the outer canthus of the right eye. Unipolar recordings to a relatively inactive reference (the nose) were employed in order to understand the field distributions in terms of both gradients in amplitude as well as absolute levels. The potentials recorded at the nose relative to a distant (i.e. non-cephalic) reference were negligible in amplitude compared to the amplitude of intracranial potentials and, thus, it can be considered an inactive reference in the current study.
Analysis
The recordings were screened for artefacts (e.g. epileptiform activity and eye or body movements) and then averaged within each task variant in order to yield event-related potentials (ERPs) of input stimuli indicating addition, subtraction or no operation (‘catch’), of true and false probes, and of correct versus incorrect responses. In addition, ERPs were made of the entire trial in order to identify longer duration processes. These whole-trial ERPs in the initial eight subjects were made by concatenating the ERPs to all conditions of the I1–I6, probe and feedback epochs. Consequently, the ERPs began 40 ms before the I1 epoch and ended 644 or 764 ms after the feedback epoch, with small gaps between the I6 and probe epochs and between the probe and feedback epochs. Data were acquired continuously in the last 10 subjects, thereby permitting a continuous epoch from before the I1 epoch until after the feedback epoch.
Previous studies have suggested that most task-related EEG activity is not phase locked to the evoking stimulus and thus does not contribute to the ERP (Klopp et al., 1999; Tallon-Baudry and Bertrand, 1999). Furthermore, averaging obscures the temporal relations between EEGs in different structures. In order to obtain a more comprehensive estimate of task-related responses, the individual trial spectral power, phase locking and phase lag were calculated using a wavelet function for different frequencies and latencies with respect to the stimulus. The resulting time–frequency plots were averaged across trials and normalized with respect to the pre-stimulus baseline. For these measures, the single trial signal si(t) for each channel i was convoluted with complex Morlet's wavelets w(t,f0):
 |
i(t,f0)] information. The estimated spectral amplitude is the average of Ai(t,f0) across all trials. The phase-locking factor is defined (Lachaux et al., 1999) as
 |
 |
) used here had a Gaussian shape both in the time domain (SD 
t) and in the frequency domain (SD f) around its central frequency f0:
 |
f = 1/2
t. Wavelets are normalized so that their total energy is unity, the normalization factor A being equal to (t0.5)-0.5. Relatively constant temporal and spatial resolution across target frequencies were obtained by adjusting the wavelet widths (f0/f ratio) according to the target frequency. In the initial eight subjects, the wavelet widths increased linearly from 1.5 to 10 as the frequency increased from 4 to 40 Hz, resulting in a t value of 60–40 ms and a f value of 2.7–4 Hz. In the last 10 subjects, the wavelet widths increased from 2 to 13 as the frequency went from 4 to 50 Hz, resulting in a t value of 79–41 ms and a f value of 2.0–3.8 Hz. Tests with simulated data confirmed that the methods used here can show clear phase-locking and lag patterns, even at 4 Hz. The findings were confirmed with standard spectral power and coherence measures on successive overlapping windows as described previously (Klopp et al., 1999, 2000).
Like extracranial EEG and magnetoencephalography (MEG), SEEG has high temporal resolution (limited by the sampling interval) but, unlike them, SEEG also has fine anatomical resolution (limited by the spacing between recording sites). Although recordings are only obtained from epileptic patients, recording locations and epochs may be selected where the EEG appears to be normal. Indeed, when it is possible to compare the results from SEEG with those from non-invasive measures in normal subjects, very similar results are obtained. The main limitation of SEEG is incomplete sampling: in order to sample the entire brain at a spacing of 3.5 mm over 10 000 contacts would be needed, but only 100 contacts are sampled in each patient. Furthermore, in the tissue that is actively generating the EEG response the amplitude and polarity change rapidly from contact to contact. Thus, it is very difficult to average responses across subjects, even if they are nominally recorded from the same region.
These considerations lead to a data analysis strategy where every recording is examined for rapid changes in amplitude and/or polarity over short distances. Electrode contacts with proof of locally generated ERP components are then further intensively examined for the cognitive correlates of those components (by comparison across other tasks), for the precise location of the contact (anatomically by comparison with MRI and stereotactic arteriography and functionally by comparison with the effects of local electrical stimulation) and for the presence of activation in the spectral domain. When multiple ERP generators in different sites are simultaneously recorded, then evidence for the presence and direction of communication between these sites is sought using phase-locking and phase lag measures. Initial analysis thus treats the data as a series of single-case studies, thereby proving local generation of particular ERP components in particular locations in particular people. The consistency of the data is evaluated by identifying replications across subjects. The certain localization and high spatiotemporal resolution of SEEG can then be used to help in interpreting particular activations identified by fMRI, PET, MEG and/or EEG studies.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Behaviour
Behavioural performance for the two most common task variants is shown in Table 1. On average, the subjects responded correctly in 69% of the trials with a reaction time of 1064 ms to the probe. Performance was slightly better (t-test, P < 0.05) on the version with numerical memoranda (N2V) than that with figural memoranda (F2V), but the reaction times did not differ.
|
Overview
The task evoked a sustained multiphasic response that was distributed across widespread cortical regions. This is illustrated in a single subject with probes sampling the occipital, parietal, Rolandic and medial regions (Figs 3–5). Despite the lack of somatosensory stimuli or behavioural responses during the input stimuli, they evoked reliable oscillations from 0.05 to 35 Hz in the peri-Rolandic area (Figs 6–8). In contrast, limbic sites, in particular the anterior cingulate gyrus, tended to be activated by the feedback tones indicating incorrect responses (Figs 9 and 10). Single-trial, phase-locking measures (Figs 5 and 6) confirmed the similarity of the activity in widespread occipital, parietal, central, medial and prefrontal sites, and phase lag measures suggested shifting patterns of information flow across the multiple processing phases.
|
) than do later peaks at 
). The internal contacts of probe O (left column) are in the right superior parietal lobule (sPl) at its junction with the superior occipital gyrus and the external contacts are in the inferior parietal lobule (iPl) at its junction with the middle occipital gyrus. Multiple large peaks are seen to polarity invert between contacts O25, O29 and O32, for example at 
). These contacts lie in the fundus of the intraparietal sulcus (iPs) at its limit with the superior occipital sulcus. The rapid oscillations in contacts D35–42 and O25–29 are superimposed on a slower oscillation with a period approximately equal to the ISI (
, m). Localization of the responsive contacts in visual motion areas was confirmed by the subjective illusions of head movement evoked by electrical stimulation between contacts D38 and D42 and between contacts O22 and O25 (1 s and 2.5 mA). In the lower traces, potentials in the same latency range are shown from probes passing through the parietal, cingulate and Rolandic cortices. The deepest contacts (P4–P11) in the cingulate sulcus (Cis) record a small but polarity-inverting N200 (
). The probe then passes though the white matter of the superior parietal lobule (contact P28) where it records slowly changing N200–P275–N360–P440 components (
). Polarity inversions of what appear to be the same components (
|
|
|
|
|
|
|
Visual Association Cortices
Multiple inverting evoked potentials were recorded in the dorsal visual pathway, with peaks as early as 100 ms and continuing until 700 ms, i.e. virtually the end of the recording epoch. Local potential inversions and gradients demonstrated that these potentials were locally generated. The precise anatomical location of the most responsive cortical recording sites indicated that they were in the visual association cortex of the dorsal stream, an inference confirmed in some cases by the effects of local electrical stimulation.
Probes sampling two such sites are shown at the top of Figure 3. Multiple large peaks separated by 100 ms inverted repeatedly between successive contacts of probes D and O. The different components inverted in different locations, indicating that the earlier components were generated in more medial cortex. These rapid oscillations were superimposed on a slower oscillation with a period approximately equal to the ISI, i.e. 840 ms. The slower oscillations also showed multiple inversions, large amplitudes and steep voltage gradients, and thus were locally generated (Figs 3 and 8). Leads passing through more anterior association cortex also recorded similar responses. Such leads could be located anterior to area MT+ (probe O, Fig. 9) or just below the posterior superior temporal sulcus (contacts C'56 and D61, Fig. 7).
Local electrical stimulation of probe D in patient 2 evoked illusory head movements. The location of probe D (visualized directly on his MRI, Fig. 5) corresponded to that of human visual motion area MT+. Earlier studies have found specific responses in this area to coherent as compared to incoherent movement in humans (Watson et al., 1993; Tootell et al., 1995; Ulbert et al., 2001) and macaques (Lappe and Duffy, 1999; Andersen et al., 2000), leading to theories that assign a role to area MT+ (and in particular to MSTd) in visual flow phenomena, inferring the direction and speed of head motion from coherent visual motion. The current finding that stimulation in this area resulted in a sensation of head motion is consistent with these theories. This contrasts with the experience of moving visual stimuli evoked by stimulation of another visual motion area in the medial parieto-occipital sulcus, which appears to correspond to SPO or V7 (Richer et al., 1991). Illusory head movements were also evoked by local electrical stimulation of probe O, which appeared to be located in an area that has also been associated with visual motion processing with fMRI, and termed LIP (Brandt et al., 1999). The neural correlates of activation in this structure and its primate homologue are poorly understood.
Occipital responses typically did not change with different stimulus materials (numbers, letters or figures; Fig. 3) or different processing requirements (adding, subtracting or no operation; Fig. 4). However, responses to auditory stimuli in the homologous task were absent (Fig. 4). The responses to the probe stimuli were similar to those evoked by the I1–I6 stimuli (Fig. 7). There were exceptions, for example the most external contacts in probe O of Figure 9, which showed an enhanced late response to the catch stimuli, peaking at 500 ms. This response was seen in the same leads to the feedback tones (in particular to those that indicated an incorrect response) and to the probe stimuli (in particular those that indicated a mismatch to the numbers held in working memory). In some cases, habituation was observed across successive input stimuli in a trial (e.g. contact O29, Fig. 8).
Parietal Lobe
A series of potentials from 130 to 500 ms were recorded in multiple parietal sites, including the intraparietal sulcus, inferior and superior parietal lobules, posterior cingulate and supramarginal and postcentral regions (Figs 3–9). Typically, parietal waveforms exhibited negative–positive–negative–positive peaks at 200–280–350–430 ms. Frequent polarity inversions, most prominently in lateral inferior parietal sites, indicated local generation, for example components in Figure 3 inverted at approximately the same latency in the most anterior part of the intraparietal sulcus (a-iPs) and the supramarginal gyrus (sMg) (see also probe P, Fig. 9). Similar potentials were also recorded in several locations near the cingulate sulcus (Cis; probes P, S and G of Fig. 3) and in the postcentral sulcus (posCs; probe S of Fig. 6), where they again polarity inverted over short distances. Although relatively low amplitude, the double inversion of peaks at 210–265 ms between successive contacts of the probe confirmed local generation. Stimulation of the probe evoked somatosensory responses that were typical of non-primary cortex (Penfield and Jasper, 1954).
Later potentials were also found to invert in the postcentral sulcus, for example at 450 ms (probe P', Fig. 6). When viewed over a longer time-scale, this late potential was seen to form a slow oscillation with approximately the same periodicity as the ISI (Fig. 7). Superimposed on these slow oscillations are the rapid oscillations reflected in the early peaks. Even longer duration modulatory responses (6 s) were rare in the parietal lobe recordings, but an example of a potential slowly developing over the course of I1–I6 stimuli was seen at the anterior limit of the intraparietal sulcus (Fig. 8).
Responses in the supramarginal gyrus from 200 to 500 ms commonly habituated over the course of I1–I6 stimuli. The parietal responses usually did not significantly differ to I1–I6 stimuli signalling addition versus subtraction versus no action, nor in relation to the material of the stimuli. However, there were rare exceptions. For example, the longest latency component to no-operation trials was somewhat larger than to the subtraction or addition trials in the supramarginal contacts shown in Figures 4 and 9. Similarly, the response to numbers was clearly larger than that to figures in the left postcentral sulcus (P'52) site shown in Figure 6 (N2I versus F2I). Oscillatory potentials were also observed to auditory stimuli when the homologous task was administered (Fig. 4). These components appeared to be similar to those evoked by the visual stimuli, except for an 90 ms earlier latency in the auditory modality. This is consistent with other SEEG studies finding both visual and auditory responses in this area (Halgren et al., 1995a).
Parietal sites also generally responded to the probe and feedback stimuli. Often it appeared that fewer oscillations might be present to these stimuli, in particular to the feedback stimuli. However, this was difficult to confirm because of signal:noise considerations: parietal responses are often small and require substantial averaging, but six times as many trials were available by combining across input stimuli I1–I6 as compared to the probe or feedback stimuli. For example, in the left inferior post-central sulcus (probe P') (Fig. 6), focal and/or polarity-inverting components from 200 to 500 ms were evoked by the probe stimuli. These resembled those evoked by the input stimuli, but had smaller amplitudes. The feedback stimuli also evoked similar waveforms, except they were smaller and of earlier latency, corresponding to the shorter latencies often observed in the auditory modality. Similarly, the supramarginal site shown in Figure 9 showed an enhanced late potential to the wrong feedback, catch input trials and probe stimuli.
Frontal Lobe
The parietal pattern of rapid oscillations (12 Hz) from 130 to 600 ms sometimes superimposed on slower waves with the period of an entire stimulus or trial was also observed in frontal recordings. For example, the latencies, polarities and waveforms of the components recorded by probe G in the precentral area in Figure 3 are essentially identical to those recorded by probe S in the supramarginal gyrus of the same patient. Direct visualization of the probe's path with MRI (Fig. 8), as well as the effects of direct electrical stimulation, confirmed that this probe was located in the primary motor cortex. Although no inversions were seen in this probe, other nearby probes did record them. Two such probes (K and M) in the precentral sulcus are shown in Figure 6. Note that, as in the postcentral cortex, these responses were small (20–30 µV), but the multiple inversions between contacts separated by 1.5 mm provide clear evidence of local generation. Stimulation-induced movements indicated that the superior probe is located in the premotor cortex (Chauvel et al., 1996b). Habituation of frontal responses was common across stimuli I1–I6 (Fig. 8).
Cycling activation with periodicity of the input stimuli (1.2 Hz) was also recorded in and near Broca's area (see, for example, contacts I'42 and R'60 in Fig. 7). Note that the onset of each cycle of this activation appeared to precede the stimulus onset. The slower oscillations in the frontal cortex were focal, but frank polarity inversions were not observed.
Long duration responses (6 s) were observed in several frontal contacts. A contingent negative variation (CNV) that was negative within the precentral gyrus is shown to polarity invert at its surface in Figure 8. Electrical stimulation of these contacts identified the generating cortex as pri