A MEG analysis of the P300 in visual discrimination tasks

Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

A MEG analysis of the P300 in visual discrimination tasks Axel Mecklinger a ,*, Burkhard Maess a, Bertram Opitz a, Erdmut Pfeifer a, Douglas Cheyne b, Harold Weinberg b a

Max-Planck-Institute of Cognitive Neuroscience, Inselstrasse 22–26, 04103 Leipzig, Germany b Brain Behaviour Laboratory, Simon Fraser University, Burnaby, B.C., Canada Accepted for publication: 28 January 1997

Abstract Based on recent research that indicated that P300 scalp topography varies as a function of task and/or information to be processed, this study examined scalp-recorded magnetic fields correlated with the P300 by means of whole-head magnetoencephalography. Subjects performed two discrimination tasks, in which targets, defined on either object or spatial characteristics of the same visual stimuli, had to be discriminated. Based on the across-subject root mean square (RMS) functions a sequence of 4 components could be identified in both tasks, N1m, P3m, and two later components, which, based on their estimated neuronal sources, were classified as representing motor processes during and following the manual responses to target stimuli. Reliable between-task differences in source localization were obtained for the P3m component, but not for the other components. Inferior-medial sources were found for the P3m evoked by both spatial and object targets, with these sources being located about 3.5 cm more anterior for object targets. These results suggest that different neuronal sources, possibly located in subcortical regions in the vicinity of the thalamus, contribute to the P3m evoked by target stimuli defined by either object or spatial stimulus characteristics.  1998 Elsevier Science Ireland Ltd. Keywords: P300; MEG; Visual discrimination; ‘What’ and ‘where’

1. Introduction Event-related potentials have been widely used to examine brain activity and cognitive functions in a wide variety of stimulus-response paradigms. Task stimuli evoke a consistent series of deflections (i.e. components) in the ERPs which can be distinguished based on their relative latency, polarity, scalp topography and correlations with physical or cognitive task manipulations. Sensory components are typically followed by an N1-P2 complex (Na¨a¨ta¨nen and Picton, 1986), a negative component with a minimum latency of 200 ms, the N200, which is followed by a positivity, the P300. The P300 is evoked by rare and task-relevant stimuli in the auditory, visual or somato-sensory modalities and occurs at a latency from 240 to 700 ms, depending on subject and task variables (Halgren et al., 1986). Although most task variables appear to produce a single P300 peak, more recent models suggests that the overall P300 is not a unitary * Corresponding author. Tel.: +49 341 9940114; fax: +49 341 9940113; e-mail: [email protected]

0168-5597/98/$19.00  1998 Elsevier Science Ireland Ltd. All rights reserved PII S(97 )0 0092-0

component but, instead, represents the summation of activity from widely distributed neuronal structures each related to the processing of a different type of information (Johnson, 1993). Evidence for the latter notion comes from studies showing that proportions of P300 amplitude associated with different experimental variables differ in scalp distribution and/or latency. The best established distinction between P300 components is the one between the novel P3 (P3a) and the target P3 (P3b). Novel, non-target stimuli evoke a large positive deflection in the ERP, with a frontocentral scalp distribution. This novelty P3 is assumed to be related to the momentary shift of attention towards an unexpected change in the environment (Courchesne, 1978; Friedman and Simpson, 1994) and can be contrasted with the P3b, a parietally maximal positivity evoked by attended target stimuli with a slightly longer latency than the P3a (Squires et al., 1975). The P3b component (in the following: P300) has been associated with processes of memory access and is assumed to be proportional to the amount of updating or reorganization of working memory contents as a function of incoming information (Donchin and Coles, 1988). Fur-

EEP 96601

46

A. Mecklinger et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

ther evidence for the view that different neuronal generators contribute to the novelty P3 and the P3b comes from ERP studies in patients with focal brain lesions. Unilateral prefrontal lesions do not affect the P3b component to target stimuli but lead to a specific reduction of the novelty P3 (Knight, 1984). Different scalp distributions were also reported for P300 components evoked by visual and auditory events (Johnson, 1989), suggesting that different neuronal generators contribute to the P300 in the two modalities. Recent studies using intracranial recordings also provided evidence for modalitydependent P300 generators (Halgren et al., 1995). The notion of information specific P300 generators was recently examined in a series of experiments in which different discriminations were required on visual stimuli. When object features had to be discriminated from a visual-spatial input for further processing, P300 scalp topography was significantly different from the P300 evoked in a condition in which spatial stimulus features from the same visual input had to be processed (Mecklinger and Mu¨ller, 1996; see also Mecklinger and Pfeifer, 1996). Several divergent methods have been employed to delineate the sources of the P300, like intracranial recordings in humans and animals (Paller et al., 1988), mapping of EEG or MEG scalp topographies (Okada et al., 1983; Mecklinger and Ullsperger, 1995), EEG recordings in brain lesioned subjects (Knight et al., 1989), or neuroimaging techniques like PET, SPECT and fMRI (cf. Ebmeier et al., 1995). In contrast to neuroimaging techniques like fMRI or PET whose temporal resolution for localization of functional relevant brain structures varies between 20 s and 30 min, MEG measures provide a temporal resolution in the millisecond domain allowing to examine the time course of activation of the neuronal structures, relevant for P300 generation. In contrast, the spatial accuracy of the MEG method in identifying neuronal sources is necessarily approximate because neuronal sources can only be estimated from the two-dimensional scalp topography of the magnetic data using inverse problem methods like dipole fitting algorithms (for an overview: Romani and Rossini, 1988). This approach has recently been used to estimate the generators of scalp-recorded magnetic fields correlated with the P300 mainly in the auditory modality (Gordon et al., 1987; Lewine et al., 1990; Rogers et al., 1991; Siedenberg et al., 1996). Gordon et al. (1987) tested two subjects in an auditory discrimination task and localized the magnetic P300 (in the following: P3m) in the superior temporal lobe. These results were extended by a study of Rogers et al. (1991). Using a time sequence approach, the authors showed that during the evolution of the auditory P300 the generating sources move from medial to lateral, i.e. from the thalamus nuclei to structures in the posterior superior temporal lobe. In the light of these latter findings the MEG methods seems to be a promising technique to estimate the locations and the time course of the neuronal structures engaged in the generation of the visual P300.

Based on recent studies that indicate that topographically different P300 components are evoked as a function of the type of information to be processed in the visual modality, the major goal of the present study was to examine whether the topographies and the estimated neuronal sources of the P3m were different when evoked by different visual discrimination tasks. In order to control for influences of physical stimulus properties on the P3m component (cf. Roth et al., 1982) two discrimination tasks were developed which consist of the same set of visual stimuli.

2. Methods and materials 2.1. Subjects Eleven male subjects between the ages of 21 and 36 years (mean 29 years) with no history of neurological disorders and normal or corrected to normal vision participated in this study. 2.2. Stimuli All stimuli were presented on a VGA monitor under the control of a 486 computer. The stimuli consisted of 16 geometric objects (e.g. circle, square, arc, triangle etc.) and were approximately 4 cm in diameter. The objects were presented in one of 16 equally spaced squares of a virtual 4 × 4 spatial matrix with sidelines of 20 cm. The objects were presented in blue against a light gray background. Stimulus presentation and the collection of behavioral data were controlled by the ERTS software (Iwanek, 1994). 2.3. Procedure The subjects were seated comfortably at a distance of 3 m in front of the 17 inch monitor and held a small response box in their hand. Each subject performed 3 oddball-like tasks. In the object discrimination task a total of 600 objects were presented and the subjects’ task was to press the response buttons whenever a round object occurred (cf. Fig. 1). The stimuli were presented sequentially at random positions in the 4 × 4 matrix with the restriction that presentations in the left and right part of the matrix were equally likely. Stimulus duration was 200 ms and the ISI varied randomly from 1400 to 1800 ms in steps of 100 ms. A fixation cross was presented in the center of the matrix throughout the task. In the spatial discrimination task the same sequence of objects was presented and the subjects’ task was to respond whenever an object was presented at one of the 4 target positions illustrated in Fig. 1. In the conjunction task the subjects were required to respond on the occurrence of round objects presented at one of the 4 target locations. All 3 tasks were subdivided in two blocks of 300 stimuli and the subjects were allowed to take a short break between the two blocks. Subjects were instructed to fixate the cross

A. Mecklinger et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

47

Fig. 2. Across-subject RMS functions for targets (solid line) and standards (dashed line) in the −200 to 1000 ms time interval for the object (upper panel) and the spatial task (lower panel). The arrows indicate the N1m, P2m and the two peaks in the P300 time interval evoked by targets and the areas used for quantification of the N1m, P2m and P3m are shaded.

in the middle of the screen and to respond as quickly and accurately as possible by pressing both response buttons with the left and right thumbs simultaneously. In all 3 tasks, the target stimuli were presented with a probability of 25%. The order of the object and the spatial task was counterbalanced across subjects and the conjunction task was always performed last. Only the data of the object and the spatial discrimination tasks will be presented here. 2.4. Data acquisition

Fig. 1. Schematic illustration of the target and standard stimuli used in the object and spatial discrimination tasks. In both tasks, objects were presented sequentially at one of the positions of the 4 × 4 matrix. In the object task, subjects responded whenever one of the 4 round objects, marked in black, were presented (upper panel), whereas in the spatial task, subjects responded whenever an object occurred at one of the 4 target locations filled with black objects (lower panel). Note that a fixation cross in the center of the screen but not the matrix was visible during the experiment.

MEG recordings were carried out with a whole cortex 64channel MEG system (CTF Systems Inc., Canada). The sensor elements were 1st order axial gradiometers, with 5 cm baseline and 2 cm diameters. They were uniformly distributed over the surface of the cortex with an intersensor spacing of ca. 4.5 cm. Total system noise was below 10 fT rms/Hz. The helmet shape was developed based on an average male head shape. Various reference channels were used for noise cancellation, through the formation of 2nd and 3rd order synthetic gradiometers (see Cheyne et al. (1992) for a detailed description of the magnetometer and noise cancellation techniques). A sensor position indicator system

48

A. Mecklinger et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

A. Mecklinger et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

consisting of 3 coils positioned at the nasion and the two pre-auricular points was used to determine the spatial locations of the sensors relative to the head. The sensor array was centered as near as possible to the subject’s head. Stimulus related epochs of 1200 ms duration (including a 200 ms prestimulus baseline) were recorded with a sampling rate of 250 Hz and stored for further analysis on a computer hard disk. A notch filter was used to remove line frequencies and all signals were recorded from DC to 50 Hz. 2.5. Data analysis 2.5.1. Behavioral data Reaction time was defined as the interval between the appearance of the stimuli and the subjects’ keypress. All of the reaction time averages were composed of correct responses only. 2.5.2. MEG data For each task, 150 and 450 stimuli were available for averaging in the target and the standard conditions, respectively. Epochs containing muscle or eye blink artifacts (criterion ±3 pT) were rejected from the averaging procedure. The data from two subjects were excluded from all further analyses because of the large amount of blink and movement artifacts. For the target and standard conditions only epochs with correctly classified stimuli were entered in the subject averages. The averaging epochs extended from 200 ms prior to stimulus onset until 1000 ms thereafter. The prestimulus epoch served as a baseline, i.e. its mean field strength was subtracted from each data point in the waveform. To assess the effects of task and stimulus type on the MEG signals root mean square (RMS) amplitudes (i.e. the square root of the sum of the squared fT values over all sensor positions) were calculated for each subject and experimental condition. 2.5.3. Mapping Magnetic field maps were generated using a spherical spline interpolation algorithm (Perrin et al., 1989). For this analysis the 64 sensors were considered to be located on the best fitting sphere. The spline interpolation was performed on the sphere and 2D maps were constructed by computing the radial projection from the vertex, which respects the length of the meridian arcs. 2.5.4. Source analysis Source analysis based on a spatio-temporal dipole modeling approach in a spherical volume conductor (Scherg, 1990) was applied to the magnetic field distribution for

49

the object and the spatial target stimuli. The EMSE software package (Neuroscan Inc.) was used for source analyses. The spatio-temporal approach to source analysis considers MEG components as arising from multiple neuronal sources located in different brain areas whose activities overlap in time. It is further assumed that neuronal sources can be modeled by combinations of equivalent current dipoles (Scherg, 1990; Scherg and Ebersole, 1993). For the volume conductor model the center and the radius of the best fitting sphere were determined first by fitting a sphere to the sensor array and second by rescaling this sphere to the individual head size as revealed by the indicator coils. Positions and orientations of the radial 1st order gradiometers relative to the head coordinate system were taken into account for the source analysis. The synthetic gradiometers used for noise cancellation were not taken into account for source analysis, because their contribution on source localization can be assumed to be minimal (J. Vrba, personal communication, May 1996). Source localizations, orientations and strengths were estimated in a head-based coordinate system. The origin of this coordinate system was set at the mid-point of the medio-lateral axis (y-axis) which connected the two preauricular points. The posterior-anterior axis (x-axis) was oriented from the origin to the nasion and the inferior-superior axis (z-axis) was perpendicular to the x-y plane and pointed to the vertex. 2.5.5. Statistical analyses To quantify the effects of the experimental manipulations a global multivariate analysis of variance (MANOVA) was performed for mean amplitudes in the N1m, P2m and P3m time interval of the RMS functions. The factors were task (object vs. spatial task), stimulus type (targets vs. standards) and time interval (N1m, P2m, P3m). Based on visual inspection of the RMS functions the time intervals 140–180 ms, 250–290 ms, and 350–600 ms were chosen for quantification of the N1m, P2m and P3m components, respectively. In case of significant interactions involving the factor time interval, separate two-way repeated measure ANOVAs with the factors task and stimulus type were performed for each of the time intervals. A MANOVA design was also used to assess the effects of the experimental manipulations on the source localization results, i.e. the 3 spatial coordinates for each of the estimated dipoles. For this analysis the factors were task, dipole, hemisphere and coordinate. To correct for violations of sphericity all ANOVA effects with two or more degrees of freedom in the numerator were adjusted with the Greenhouse and Geisser procedure (Greenhouse and Geisser, 1959). Post hoc tests were performed by means of a modified Bonferroni procedure (Keppel, 1991).

Fig. 3. Sets of 64 averaged waveforms of 1200 ms length (including the 200 ms pre stimulus baseline) of an individual subject in the object (Fig. 3a) and the spatial task (Fig. 3b). The waveforms are superimposed for target and standard stimuli. The layout of the waveforms corresponds to the sensor positions on the head, with the upper and lower waveforms displaying the most anterior and posterior sensor positions.

50

A. Mecklinger et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

3. Results 3.1. Behavioral data Subjects responded faster and more accurately to the object targets than to the spatial targets. The means of subjects’ median reaction times for correct responses were 451 ms (SE 17.7) in the object task and 583 ms (SE 23.3) in the spatial task. Performance accuracy was 99.1% in the object task and 95.1% in the spatial task. As revealed by t tests for dependent variables, both differences were statistically reliable: reaction times, t(8) = 4.56, P , 0.002; accuracy: t(8) = 2.23, P , 0.05. 3.1.1. MEG data Fig. 2 shows the RMS functions averaged across subjects for the target and standard stimuli evoked in both tasks. As is apparent from the figure, targets evoke larger magnetic signals than standards. This difference started earlier (i.e. around 170 ms) in the object task than in the spatial task and for both tasks extended throughout the recording epoch. Both stimulus types evoke an N1m component with a peak latency of 168 ms. For standards the RMS function starts to decline after the N1m, whereas for target stimuli P2m components with peak latencies of 272 ms in the object task and 268 ms in the spatial task were obtained. Moreover, in the P3m time interval, i.e. from 350 to 600 ms, two peaks, i.e. peak I and peak II, could be identified for the targets clearly visible in the object task at 404 ms and 520 ms and less clearly discernible in the spatial task at 516 ms and 588 ms. These observations could be confirmed by statistical analyses. The global MANOVA with factors task (two levels), stimulus type (two levels) and time interval (3 levels) yielded main effects of stimulus type, F(1,8) = 103,32, P , 0.0001, and time interval, F(2,16) = 4.69, P , 0.03. Moreover, the interactions, task × stimulus type, F(1,8) = 31,72, P , 0.0008, stimulus type × time interval, F(2,16) = 74.92, P , 0.0001, and task × stimulus type × time interval, F(2,16) = 13.67, P , 0.0003, reached the significance level. Based on the latter interactions separate ANOVAs were performed for each of the time intervals. For the N1m time interval a significant interaction task × stimulus type was found, F(1,8) = 5.69, P , 0.04. Post hoc tests indicated that targets produced larger N1m components than standards in the object task, P , 0.01, but not in the spatial task, P = 0.66. For the P2m component, a significant main effect of stimulus type was obtained, F(1,8) = 10.11, P , 0.01. Moreover the interaction task × stimulus type, F(1,8) = 9.65, P , 0.01, reached the significance level, indicating larger target than standard P2m components in the object task, P , 0.008, but not in the spatial task, P = 0.59. Finally, for the P3m time interval, there was a significant effect of stimulus type, F(1,8) = 245,11, P , 0.001, and an interaction task × stimulus type, F(1,8) = 44.65, P , 0.0002. The latter

interaction indicates that target stimuli evoked larger P3m components in the object task than in the spatial task, P , 0.01, whereas the P3m components evoked by standard stimuli were larger in the spatial than in the object task, P , 0.006. For each subject only the N1m and the two peaks in the P300 time interval were consistently present in both tasks, whereas the P2m was discernible for the object and the spatial target stimuli only in 3 subjects, and was thus not considered for further analysis. Note that, although no clear peaks were discernible in the time interval after 600 ms, there still was pronounced activity in the RMS functions for target stimuli in this time period. Fig. 3 displays the MEG waveforms recorded at all 64 sensor positions in the object and the spatial task for one of the subjects investigated. The N1m component is clearly visible for targets and standards in both tasks with a pronounced peak at the right posterior sensor positions, e.g. R26 and R35, and a polarity-reversed smaller peak at left posterior sensors, e.g. L26 and L35. Around 400 ms, i.e. in the P300 time interval, a more complex picture emerged: for object targets there are maxima at right frontal sensors, e.g. R23 and R24 and with reversed polarity at the corresponding left frontal sensors (L23 and L24). A similar dipolar pattern was found at more posterior left and right hemisphere sensors (e.g. R33 and L33) in this time period. These two dipolar patterns reflect the two peaks (i.e. Peak I and Peak II) in the P300 time interval in the RMS function of this subject. For the spatial targets, similar anteriorly and posteriorly focused polarity reversals between left and right hemisphere sensors were obtained in the P3m time interval, again reflecting the two peaks in the RMS function. In this task, however, the peaks in the MEG waveforms were substantially smaller than those evoked by object targets. In both tasks, magnetic activity was obtained in the time interval after 600 ms. This activity was sharply focused to a sensor located above left central cortical areas, i.e. L15. It is further noteworthy that the maximum of this late component was at 612 ms for object targets and at 668 ms for spatial targets (as measured at the L15 sensor), and thus 56 ms earlier in the object task. The longer latency of this late component in the spatial as compared to the object task was confirmed by statistical analyses. The mean baselineto-peak-latency of the late component averaged across subjects was 666 ms (SE 21.1) for object targets and 717 ms (SE 25.6) for spatial targets, with this difference being statistically reliable, t(8) = 2.51, P , 0.04. The topographic distributions of the N1m, the two peaks in the P300 time interval and the late component evoked by targets and standards in the object and spatial task for this subject are displayed in Figs. 4 and 5. For the N1m and the P3m time intervals the topographic distributions are plotted for those time points where the RMS functions displayed maxima, i.e. at 168 ms, 396 ms and 444 ms in the object task, and at 168 ms, 420 ms and 460 ms in the spatial task.

A. Mecklinger et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

51

Fig. 4. Radial projected magnetic maps for an individual subject for the N1m, the two peaks in the P3m time interval and the late component evoked by targets (a) and standards (b) in the object task. The amplitude scales are ±150 fT. Dark areas represent ingoing magnetic fields and white areas represent outgoing magnetic fields.

The late component is plotted for those time points where the MEG waveforms at the L15 sensor reached their maximum amplitude, i.e. at 612 ms and 668 ms in the object and spatial task, respectively. For the N1m clear dipolar patterns were obtained for targets and standards, in both tasks with a polarity reversal over the posterior left and right cortex. While the dipolar patterns corresponding to the N1m were highly similar for the two tasks, between-task topography differences emerged in the P3m time interval. For Peak I evoked by targets in both tasks a right anterior ingoing maximum and a left anterior outgoing maximum were obtained (cf. Figs. 4 and 5a). These patterns may have emerged from two bilateral dipoles oriented perpendicular to each other with anterior bilateral extrema for which the posterior extrema spatially overlap and thus appear as closeto-zero magnetic field values. For Peak II in the P3m interval similar double dipole patterns were obtained; however, as compared to Peak I, for both dipoles the anterior extrema were less distant from each other, presumably reflecting more superior dipole locations. Note that in the object task, no dipolar patterns were obtained for the standards in the P3m time interval (cf. Fig. 4b), whereas in the spatial task the standards evoked dipolar activity in the P3b time interval, though less pronounced than for the targets. This latter result is consistent with the finding of more pro-

nounced RMS activity evoked by standard stimuli in the spatial as compared to the object task (cf. Fig. 2). Finally, the late components evoked by targets in both tasks yielded dipolar patterns in each hemisphere with the extrema located on the lateral to medial dimension, whereas no such patterns were obtained for the standard stimuli. 3.2. Source analysis of the magnetic components The estimated source localizations of the N1m, the components in the P3m interval, and the late component evoked by the target stimuli were determined by the following procedure. First, fitting intervals were defined which contained a component’s onset and maximum in the individual RMS function. In the case of the P3m interval the larger one of the two RMS peaks was selected, and in the case of the late component which was not always discernible as RMS peak, the fitting interval was selected around the field maximum at the sensor position at which it was largest, i.e. L15. Based on these criteria, for the individual subject presented in Figs. 3–5 the fitting intervals were 148–172 ms, 380–408 ms and 592–620 ms for the N1m, P3m and late component in the object task. The corresponding intervals for the spatial task were 148–172 ms, 404–432 ms and 648–676 ms. Second, for each fitting interval an initial model was defined, com-

52

A. Mecklinger et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

prising one dipole pair located 1 cm to the left and right of the x-z plane and oriented tangentially to the surface. For the P3m fitting interval the initial model comprised two dipole pairs which were fitted sequentially. These initial models were used to fit the N1m, P3m and the late component for object and spatial targets. Goodness-of-fit was evaluated based on the cross-correlation-coefficient between the modeled and the empirical waveforms in the respective fitting intervals. Fig. 6 shows the dipole modeling results obtained for the MEG signals of the subject presented in Figs. 3–5. The residual variances (1 − R2) of the dipole models for the 3 fitting intervals were 9.5%, 15% and 29% for object targets, and 13%, 19% and 18% for spatial targets. The N1m is reflected by two dipoles (i.e. 1 and 2) located close to each other, approximately in the posterior part of the parietal lobe. Dipoles 3 and 4 and 5 and 6 represent the magnetic activity in the P3m fitting interval. Dipoles 3 and 4 were located medially and inferiorly, presumably in medial subcortical structures. Interestingly, these two dipoles are located considerably more posteriorly for spatial targets than for object targets. Dipoles 5 and 6 are located in the left and right hemisphere anterior to the central fissure in both tasks, presumably reflecting activity of the motor cortex. Finally, dipoles 7 and 8 are located in superior lateral regions posterior to the central fissure, approximately in the somatosensory cortex. The estimated locations of the dipole pair 1 and 2, reflecting the magnetic activity in the N1m interval, as well as pairs 3 and 4 and 5 and 6 reflecting the activity in the P3m interval, for all 9 subjects are displayed in Fig. 7. This across-subject analysis was restricted to the first 3 dipole pairs since the late component was not clearly discernible in all subjects. As is apparent from the figure the across-task variabilities in the localization of dipole pairs 1 and 2 and 5 and 6 are remarkably low1. In contrast, the estimated inferior-medial dipoles 3 and 4 show between-task differences on the anterior-posterior dimension, being located more posteriorly for spatial targets than for object targets. These observations were confirmed by statistical analyses. For these analyses the x, y and z coordinate values from each subject were realigned to an average-size head model which was obtained by averaging the individual head coordinates separately for each axis and both polarities. A MANOVA with the factors task (two levels), dipole (3 levels), hemisphere (two levels) and coordinate (3 levels) revealed a main effect of task, F(1,8) = 21.28, P , 0.002, and the interactions dipole × task, F(2,16) = 13.40, P , 0.0001, and dipole × task × coordinate, F(4,32) = 6.24, P , 1 Note that for the individual subject presented in Figs. 3–6, dipoles 1 and 2 were located less posterior (mean x-coordinate −3.3 cm) than the across subject mean (mean x-coordinate −5.3 cm) and the N1m was also more prominent over the right hemisphere. However, the subject’s dipole locations for dipole pairs 3 and 4 and 5 and 6 and the topographic distributions of the corresponding MEG components were highly representative for the group of 9 subjects under investigation.

0.002. Univariate ANOVAs were conducted to further examine these interactions. Between-task differences in dipole coordinates were significant for dipole pair 3 and 4, P , 0.003, but not for dipole pair 5 and 6, and pair 1 and 2 in the N1m interval, both P . 0.12. Moreover, for the inferior medial dipoles 3 and 4 an interaction task × coordinate, F(2,16) = 23.98, P , 0.003, was obtained. Post hoc test revealed that dipoles 3 and 4 were located more anteriorly for object targets than for spatial targets, P , 0.001, whereas no effects were found for the y and z coordinates, P . 0.16.

4. Discussion In this study we examined magnetic fields evoked by standards and rare target stimuli in two different visual discrimination tasks by means of whole head magnetoencephalography. The same stimuli were used in both tasks, and subjects had to discriminate either object or spatial stimulus features. Targets evoked more pronounced magnetic responses than standards in both tasks. Differences in the magnetic fields between standards and targets started in the N1m time interval, i.e. 140–180 ms, in the easier to perform object discrimination task but not before 350 ms post-stimulus in the more difficult to perform spatial task. For targets in both tasks two peaks in the RMS functions could be identified in the P3m interval, i.e. from 350 to 600 ms. As is apparent from Figs. 4 and 5 the standards in the spatial but not the object task evoked dipolar activity in the P300 time interval. This observation was also confirmed by the more pronounced RMS activity for spatial as compared to object standards. Since epochs with false alarms were excluded from the standard waveforms this dipolar activity to spatial standards cannot be attributed to a portion of falsely classified standards evoking P3m components. The dipolar pattern evoked by spatial standards could be associated with the more demanding discrimination process in the spatial as compared to the object task, during which some of the standards, though correctly identified, evoked target-like P3m components. Since motor responses were required for targets but not for standards, it is conceivable that one of the two components in the P3m time interval reflects activity arising from the motor cortex. This assumption was confirmed by the source localization results. When the magnetic activity in the P300 time interval for each of the subjects was modeled as two pairs of equivalent current dipoles, the dipole pair reflecting the first peak in the RMS function converged to bilateral inferior locations, whereas the dipoles reflecting the second of the two RMS peaks always converged to superior bilateral locations anterior to the central fissure, i.e. possibly representing the motor cortex. Based on these results the first dipole pair in the P3m time interval can be assumed to reflect the P3m component, whereas the estimated source locations for the second dipole pair suggest

A. Mecklinger et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

53

Fig. 5. Radial projected magnetic maps for an individual subject for the N1m, the two peaks in the P3m time interval and the late component evoked by targets (a) and standards (b) in the spatial task. Amplitude scaling is as in Fig. 4.

that their magnetic fields reflect activity in the primary motor cortex during response execution. This movementrelated neuromagnetic activity resembles movement-locked changes in the ERP like the readiness potential, a negative shift preceding voluntary movements maximal at precentral electrode sites (cf. Kornhuber and Deecke, 1965; Kno¨sche et al., 1996). The bilateral symmetrical sources presumably emerge from the fact that subjects used the thumbs of both hands for responding. It should be noted that although the residual variances of the dipole models for some subjects and fitting intervals were quite high (e.g. 29% for the late component evoked by object targets, cf. Fig. 6), the topographic distribution of the residual waveforms did not reveal any dipolar patterns for any of the fitting intervals, nor did addition of a further dipole substantially affect the dipole models. We therefore assume that the residual variances reflect the background noise level rather than residual dipolar activity. The contribution of motor-related processes to the target waveforms was also confirmed by the presence of late (i.e. after 600 ms) dipolar activity. In this time interval magnetic fields, restricted to very few sensor positions and thus not discernible as peaks in the RMS functions, were obtained. For the individual subject presented in Figs. 3–6 the field maxima occurred 62 ms (object targets) and 79 ms (spatial targets) after the corresponding median reaction times, indicating that these magnetic fields reflect post-response pro-

cesses. This assumption was confirmed by dipole analyses revealing two lateral and symmetrical sources in the late time interval located superiorly and posteriorly to the central fissure, presumably in the somatosensory cortex. This result suggests the occurrence of reafferent mechanisms activating parts of the somatosensory cortex and is also consistent with a previous report of post-movement neuromagnetic fields generated in the somatosensory cortex which were observed 90–130 ms after the onset of EMG activity related to finger movements (Cheyne and Weinberg, 1989). While for the N1m and the two motor-related components no reliable between-task differences in dipole localization were obtained, pronounced differences in source localization were found for the dipole pair reflecting the P3m component in both tasks. The sources, modeled as a pair of bilateral equivalent current dipoles, for object targets were located about 3.5 cm (averaged across subjects) more anterior than for spatial targets in a x-y plane, possibly reflecting thalamic structures or adjacent subcortical regions. This pattern of results was found for 8 of the 9 subjects, whereas for one subject the x-coordinate locations of dipoles 3 and 4 were the same in both tasks. This double dipole solution was also supported by the isofield maps in the P300 time interval (cf. Figs. 4 and 5) that did not indicate the presence of more than two dipoles. Previous MEG studies, examining neuronal generators of

54

A. Mecklinger et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

tional point of view, it is assumed that the P300 component reflects processes of memory reorganization by which current working memory contents are modified as a function of the incoming information (Donchin and Coles, 1988; Fabiani and Donchin, 1995). The present finding of different P3m generators for object and spatial targets is consistent with the view that these generators represent forms of memory access, with the activation of a particular generator depending on the type of memory representation required for task performance (cf. Johnson, 1993). This interpretation, however, is challenged by the fact that, in the present experiment, the two tasks not only differed in the type of

Fig. 6. The localizations of the 4 pairs of equivalent current dipoles in an individual subject representing the N1m (dipoles 1 and 2), the first (dipoles 3 and 4) and second peak (dipoles 5 and 6) in the P3m time interval and the late component (dipoles 7 and 8) for object targets (upper panel) and spatial targets (lower panel).

the P3m (Okada et al., 1983; Lewine et al., 1990), as well as studies employing intracranial electrical recordings (Halgren et al., 1995) indicate that regions in the medial temporal lobes are engaged in P300 generation. It is important to keep in mind that the MEG method is not capable of detecting radially oriented neuronal sources (cf. Siedenberg et al., 1996) and that the present dipole modeling approach is based on a number of unrealistic assumptions (i.e. spherical head model, punctual dipole sources) and is also less accurate especially for sources close to the center of the sphere. Despite these objections, the present results do not provide support for the notion that the medial temporal lobes are the primary sources of P300 generation. The locations of dipoles 3 and 4 on the superior-inferior axis (cf. Figs. 6 and 7) rather suggest that subcortical structures in the vicinity of the thalamus contribute to the scalp-recorded neuromagnetic fields of the P300. The current results thus tentatively support the findings of Rogers et al. (1991), which indicated that the neuronal activity contributing to the P3m is initially generated in medial subcortical regions. Within this horizontal plane, the estimated locations of the neuronal generators of the P3m component evoked by object targets were found to be located more anteriorly. This pattern of results is consistent with the notion of multiple and distributed neuronal generators for the P300 component, each related to a particular type of information to be processed. From a func-

Fig. 7. The dipole locations obtained for all 9 subjects for the N1m (upper panel) and the two peaks in the P3m interval (middle and lower panel) for the object (filled symbols) and the spatial targets (open symbols). The dipole localizations of the individual subject presented in Figs. 3–6 are indicated with light gray (spatial task) and dark gray (object task) symbols. Coordinate scaling is in cm.

A. Mecklinger et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

information to be processed but also in task difficulty. Thus, the different source localizations found for the P300 on the anterior-posterior dimension could also result from the fact that the easily discriminable object targets evoked larger amounts of P3a components. However, based on the high consistency in the temporal order of dipole activity across tasks (cf. Fig. 6) and based on the fact that both tasks were completely matched for stimulus and response characteristics, it appears rather unlikely that the two tasks evoke qualitatively different P300 components (i.e. P3a and P3b). Based on these arguments, the results are more consistent with the notion that both types of target stimuli evoke P300 components differing in latency. The finding that different neuronal structures at inferior medial locations are engaged in tasks in which either object or spatial features of the same stimulus had to be discriminated is also consistent with the notion that ‘what’ and ‘where’ information is processed and stored by different neuroanatomical systems in the primate brain (cf. Ungerleider and Mishkin, 1982; Smith et al., 1995; Mecklinger and Pfeifer, 1996). The finding of different neuronal structures being active during P300 generation suggests that, besides perceptual processes, also higher level processing aspects (memory access and representation) for ‘what’ and ‘where’ information are mediated by different neuronal structures. Taken together, the present results suggest that the examination of scalp-recorded magnetic fields by means of whole head magnetoencephalography can shed some light on the temporal characteristics and the spatial locations of the neuronal structures underlying the generation of the P300 component in visual discrimination tasks. In showing a spatial dissociation for the neuronal sources of the P3m but not for the other components evoked by the two types of target stimuli the present results provide converging evidence for the notion of task- and information-specific generators of the P300 component. The present results can provide a guide for elucidating the neuronal structures involved in P300 generation in more detail by means of neuroimaging techniques.

Acknowledgements This experiment was conducted while B. Maess and E. Pfeifer worked as visiting scientists in the Brain Behaviour Laboratory at the Simon Frazer University in Burnaby, Canada, We wish to thank M. Burbank (CTF Systems Inc.) for the invitation and for financial support.

References Cheyne, D. and Weinberg, H. Neuromagnetic fields accompanying unilateral finger movements: pre-movement and movement-evoked fields. Exp. Brain Res., 1989, 78: 604–612. Cheyne, D., Vrba J., Crisp, D., Betts, K., Burbank, M., Cheung, T., Fife, A., Haid, G., Kubik, P., Lee, S., McCubbin, J., McKay, J., McKenzie,

55

D., Spear, P., Taylor, B., Tillotson, M., Weinberg, H., Basar, E. and Tsutada, T. Use of an unshielded, 64-channel whole-cortex MEG system in the study of normal and pathological brain function. In: Proc. Satellite Symp. on Neuroscience and Technology, 1992, pp. 46–50. Courchesne, E. Changes in P3 waves with event repetition: long-term effects on scalp distribution and amplitude. Electroenceph. clin. Neurophysiol., 1978, 45: 468–482. Donchin, E. and Coles, M.G.H. On the conceptual foundations of cognitive psychophysiology: a reply to comments. Behav. Brain Sci., 1988, 11: 406–417. Ebmeier, K.P., Steele, J.D., MacKenzie, D.M., O’Carrol, R.E., Kydd, R.R., Glabus, M.F., Blackwood, D.H.R., Rugg, M.D. and Goodwin, G.M. Cognitive brain potentials and regional cerebral blood flow equivalents during two- and three-sound auditory ‘oddball tasks’. Electroenceph. clin. Neurophysiol., 1995, 95: 434–443. Fabiani, M. and Donchin, E. Encoding processes and memory organization: a model of the von Restorff effect. J. Exp. Psychol. Learn. Mem. Cognit., 1995, 21(1): 1–17. Friedman, D. and Simpson, G.V. ERP amplitude and scalp distribution to target and novel events: effects of temporal order in young, middle-aged and older adults. Cognit. Brain Res., 1994, 2: 49–63. Gordon, E., Sloggett, G., Harvey, I., Kraiuhin, C., Pennie, C., Yiannikas, C. and Meares, R. Magnetoencephalography: locating the source of P300 via magnetic field recording. Clin. Exp. Neurol., 1987, 23: 101– 110. Greenhouse, S.W. and Geisser, S. On methods in the analysis of profile data. Psychometrika, 1959, 24: 95–112. Halgren, E., Stapleton, J.M., Smith, M.S. and Altafullah, I. Generators of the human scalp P3(s). In: R.Q. Cracco and R.H. Bodis-Wollner (Eds.), Evoked Potentials. Liss, New York, 1986, pp. 269–284. Halgren, E., Baudena, P., Clarke, J.M., Heit, G., Lie´geois, C., Chauvel, P. and Musolino, A. Intracerebral potentials to rare target and distractor auditory and visual stimuli I: superior temporal plane and parietal lobe. Electroenceph. clin. Neurophysiol.,1995, 94: 191–220. Iwanek, R. Review of the Experimental RunTime System. Psychology Software News, Vol. 5(2). Centre for Psychology, University of York, 1994, pp. 65–67. Johnson, R.J. Auditory and visual P300s in temporal lobectomy patients: evidence for modality-dependent generators. Psychophysiology, 1989, 26: 633–650. Johnson, R.J. On the neuronal generators of the P300 component of the event-related potential. Psychophysiology, 1993, 30: 90–97. Keppel, G. Design and Analysis (3rd edn.). Prentice Hall, Englewood Cliffs, NJ, 1991. Knight, R.T. Decreased response to novel stimuli after prefrontal lesions in man. Electroenceph. clin. Neurophysiol., 1984, 59: 9–20. Knight, R.T., Scabini, D., Woods, D.L. and Clayworth, C.C. Contributions of temporal-parietal junction to the human auditory P3. Brain Res., 1989, 502: 109–116. Kno¨sche, T., Praamstra, P., Stegeman, D. and Peters, M. Linear estimation discriminates midline sources and a motor cortex contribution to the readiness potential. Electroenceph. clin. Neurophysiol., 1996, 99: 183–190. Kornhuber, H.H. and Deecke, L. Hirnpotential :a¨nderungern bei Willku¨rbewegungen und passiven Bewegungen des Menschen. Bereitschaftspotential und reafferente Potentiale. Pflu¨gers Arch., 1965, 284: 1– 17. Lewine, J.D., Roeder, S.B., Oakey, M.T., Arthur, D.L., Aine, C.J., George, J.S. and Flynn, E.R. Localization of the generators of the magnetic P3. In: S. Williamson, M. Hoke, G. Stroink and M. Kotani (Eds.), Advances in Biomagnetism. Plenum Press, New York, 1990, pp. 229–238. Mecklinger, A. and Mu¨ller, N. Dissociations in the processing of ‘what’ and ‘where’ information in working memory: an event-related potential analysis. J. Cognit. Neurosci., 1996, 8: 453–473. Mecklinger, A. and Pfeifer, E. Event-related potentials reveal topographical and temporal distinct neuronal activation patterns for spatial and object working memory. Cognit. Brain Res., 1996, 4: 211–224.

56

A. Mecklinger et al. / Electroencephalography and clinical Neurophysiology 108 (1998) 45–56

Mecklinger, A. and Ullsperger, P. The P300 to novel and target events: a spatio-temporal dipole model analysis. NeuroReport, 1995, 7(1): 241– 245. Na¨a¨ta¨nen, R. and Picton, T.W. The N1 wave of the human electric and magnetic response to sound. Psychophysiology, 1986, 24: 375–425. Okada, Y.C., Kaufman, L. and Williamson, S.J. The hippocampal formation as a source of the slow endogenous potentials. Electroenceph. clin. Neurophysiol., 1983, 55: 417–426. Paller, K.A., Zola-Morgan, S., Squire, L.R. and Hillyard, S.A. P3-like brain waves in normal monkeys and in monkeys with medial temporal lesions. Behav. Neurosci., 1988, 102: 714–725. Perrin, P., Pernier, J., Bertrand, O. and Echallier, J.F. Spherical splines for scalp potential and current density mapping. Electroenceph. clin. Neurophysiol., 1989, 72: 184–187. Rogers, R.L., Baumann, S.B., Papanicolaou, A.C., Bourbon, Th.W., Alagarsamy, S. and Eisenberg, H.M. Localization of the P3 sources using magnetoencephalography and magnetic resonance imaging. Electroenceph. clin. Neurophysiol., 1991, 79: 308–321. Romani, G.L. and Rossini, P. Neuromagnetic functional localization: principles, state of the art and perspectives. Brain Topogr., 1988, 1: 5–21. Roth, W.T., Blowers, G.H., Doyle, C. and Kopell, B.S. Auditory stimulus

intensity effects on components of the late positive complex. Electroenceph. clin. Neurophysiol., 1982, 54: 132–146. Scherg, M. Fundamentals of dipole source potential analysis. In: F. Grandori, M. Hocke and G.L. Romani (Eds.), Auditory Evoked Magnetic Fields and Electrical Potentials. Advances in Audiology, Vol. 6. Karger, Basel, 1990, pp. 40–69. Scherg, M. and Ebersole, J.S. Models of brain sources. Brain Topogr., 1993, 5: 419–423. Siedenberg, R., Goodin, D.S., Aminoff, M.J., Rowley, H.A. and Roberts, T.P.L. Comparison of late components in simultaneously recoded eventrelated electrical potentials and event-related magnetic fields. Electroenceph. clin. Neurophysiol., 1996, 99: 191–197. Smith, E.E., Jonides, J., Koeppe, R.A., Awh, E., Schumacher, E.H. and Minoshima, S. Spatial versus object working memory: PET investigations. J. Cognit. Neurosci., 1995, 7: 337–356. Squires, N.K., Squires, K.C. and Hillyard, S.A. Two varieties of longlatency positive waves evoked by unpredictable auditory stimuli in man. Electroenceph. clin. Neurophysiol., 1975, 83: 387–401. Ungerleider, L.G. and Mishkin, M. Analysis of visual behavior. In: D.J. Ingle, M.A. Goodale and R.J.W. Mansfield (Eds.), Two Cortical Visual Systems. MIT Press, Cambridge, MA, 1982, pp. 549–586.