Cerebral Cortex, Vol. 10, No. 9, 829-839,
September 2000
© 2000 Oxford University Press
Neurophysiological Measures of Working Memory and Individual Differences in Cognitive Ability and Cognitive Style
San Francisco Brain Research Institute & SAM Technology, 425 Bush Street – Fifth Floor, San Francisco, CA 94108-3708, USA
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionConclusionsNotesReferences |
|---|
The capacity to deliberately control attention in order to hold and manipulate information in working memory is critical to higher cognitive functions. This suggests that between-subject differences in general cognitive ability might be related to observable differences in the activity of brain systems that support working memory and attention control. To test this notion, electroencephalograms were recorded from 80 healthy young adults during spatial working memory tasks. Measures of task-related neurophysiological and behavioral variables were derived from these data and compared to scores on a test battery commonly used to assess general cognitive ability (the WAIS-R). Subjects who scored high on the psychometric test also tended to respond faster in the experimental tasks without any loss of accuracy. The amplitude of the late positive component of the event-related potential was larger in high-ability subjects, and the frontal midline theta component of the EEG signal was also selectively enhanced in this group under conditions of sustained performance and high working memory load. These results suggest that subjects who scored high on the WAIS-R were better able to focus and sustain attention to task performance. Changes in the EEG alpha rhythm in response to manipulations of task practice and load were also examined and compared between frontal and parietal regions. The results indicated that high-ability subjects developed strategies that made relatively greater use of parietal regions, whereas low-ability subjects relied more exclusively on frontal regions. Other analyses indicated that hemispheric asymmetries in alpha band measures distinguish between individuals with relatively high verbal aptitude and those with relatively high nonverbal aptitude. In particular, subjects with a verbal cognitive style tended to make greater use of the left parietal region during task performance, and subjects with a nonverbal style tended to make greater use of the right parietal region. These results help clarify relationships between task-related brain activity and individual differences in cognitive ability and style.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionConclusionsNotesReferences |
|---|
Working memory (WM) refers to the limited capacity to hold and manipulate information in mind for several seconds in the context of cognitive activity. In a sense, WM is an outcome of the ability to control attention and sustain its focus on a particular active mental representation (or set of representations) in the face of distracting influences (Engle et al., 1999
). This ability plays an important role in comprehension, reasoning, planning and learning (Baddeley, 1992). Indeed, the use of active mental representations to guide performance appears critical to behavioral flexibility (Goldman-Rakic, 1987, 1988), and measures of this ability tend to be positively correlated with performance on psychometric tests of cognitive ability and other indices of scholastic aptitude (Carpenter et al., 1990; Kyllonen and Christal, 1990). Modulation of Brain Metabolism and Neuroelectrical Activity by Variations in Attention and WM Demands
Metabolic studies suggest that WM involves a functional network linking regions of prefrontal cortex with posterior association cortices (Petrides and Milner, 1982; Frisk and Milner, 1990; Owen et al., 1996). Activation of this network also occurs during reasoning and problem solving (Roland, 1984; Haier et al., 1988; Prabhakaran et al., 1997;) Summated dendritic potentials of cortical pyramidal neurons associated with such activation can be detected in measurements of neuroelectric activity recorded at the scalp. More specifically, a task-imposed change in WM requirements tends to produce characteristic changes in the amplitude of spectral components of the ongoing electro- encephalogram (EEG), and in components of the averaged event-related potential (ERP) elicited by a stimulus.
For example, in recent studies (Gevins et al., 1997, 1998), we have found that an EEG signal in the theta (4–7 Hz) range, largest over midline frontal regions of the scalp (‘frontal midline theta’ or fm
), was enhanced in tasks with greater WM demands. More generally, fm has been noted to be larger in amplitude in a variety of tasks that require sustained focused attention (Miyata et al., 1990; Yamamoto and Matsuoka, 1990; Gundel and Wilson, 1992; Gevins et al., 1997, 1998). Current evidence suggests that fm is generated in the anterior cingulate cortex (Gevins et al., 1997; Ishii et al., 1999), an important component of an anterior attentional network critical to the performance of complex cognitive tasks (Posner and Peterson, 1990; Posner and Rothbart, 1992).
Conversely, signals in the 8–12 Hz or alpha (
) EEG band tend to be attenuated by attention-demanding tasks (Berger, 1929; Glass, 1966; Galin et al., 1978; Gevins et al., 1979a, 1997, 1998; Gevins and Schaffer, 1980; Gundel and Wilson, 1992). The magnitude of activity during cognitive tasks is inversely proportional to the fraction of cortical neurons recruited into a transient functional network for purposes of task performance (Gevins and Schaffer, 1980; Pfurtscheller and Klimesch, 1992). Thus progressive increases in task-imposed WM load result in monotonic attenuation (Gevins et al., 1997, 1998). Both fm and tend to increase as subjects develop skill in task performance (Gevins et al., 1997; Smith et al., 1999).
Components of the ERP are also sensitive to the attentional and WM demands of tasks (Gevins et al., 1996; McEvoy et al., 1998). To take the most common example, stimuli in tasks that require active discrimination between classes of events typically evoke a large positive voltage deflection in the interval between about 300 and 500 ms following stimulus onset (Picton, 1992). The amplitude of this ‘late positive component’ (LPC or ‘P300’) is larger when attention is exclusively focused on analyzing the eliciting stimulus, and smaller when attention is consumed by some other mental activity. It is diminished when there is a concurrent requirement to hold information in WM (Gevins et al., 1996; McEvoy et al., 1998), and when multiple tasks compete for attention (Kramer et al., 1987). Its amplitude can be positively correlated with individual differences in WM capacity (Nittono et al., 1999).
Individual Differences in Cognitive Ability and Neurophysiological Correlates of WM Task Performance
In sum, brain activation associated with WM task performance is reflected in modulation of scalp-recorded neuroelectric activity. Given that behavioral performance in WM tasks is closely correlated with overall cognitive ability (Carpenter et al., 1990; Kyllonen and Christal, 1990), the current study sought to determine whether human neuroelectrical activity during WM tasks display similar patterns of individual differences. In particular, given the results reviewed above, it might be expected that the magnitude of WM task-related modulation of the fm and EEG signals, and the amplitude of the LPC of the ERP elicited in WM task, would be associated with individual differences in cognitive ability. To address this hypothesis, EEGs were recorded from 80 subjects while they performed a WM task used in several prior studies (Gevins et al., 1996, 1997, 1998; McEvoy et al., 1998; Smith et al., 1999). Measures of the fm and EEG signals and the LPC of the ERP were extracted from these data and compared to scores on the Wechsler Adult Intelligence Scale – Revised (WAIS-R) (Wechsler, 1981).
At the time these data were collected the WAIS-R instrument had served for many years as the de facto standard neuro- psychological instrument for assessing cognitive ability. The WAIS-R is composed of 11 tests. From the results of these tests three composite scores are derived. These include a total score intended to summarize general cognitive ability, a ‘verbal’ ability subscore and a ‘performance’ ability subscore (to distinguish this later score from behavioral measures obtained in the cognitive task employed in this study, we will henceforth refer to it as a ‘nonverbal’ ability score). The verbal and nonverbal subscores on the WAIS-R tend to be highly correlated. However, a variety of factors can contribute to differences between these scores, including cultural and educational differences. Such differences motivated two separate types of analyses to assess individual differences. In the first, we sought to evaluate the main hypothesis that individual differences in neuroelectric measures of WM would covary with general cognitive ability. To address this question subjects were eliminated from the sample if they displayed more than trivial differences in their verbal and nonverbal WAIS-R subscores. A second analysis specifically examined differences in neuroelectrical activity between the remaining subjects with a relatively high verbal subscore versus those with a relatively high nonverbal subscore.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionConclusionsNotesReferences |
|---|
Subjects
Data were collected from 80 clinically healthy, right-handed young adults. The participants (38 women and 42 men) had a mean age of 21.4 years (range 18–28 years). Their ethnic mix was statistically representative of the diversity of the San Francisco Metropolitan Region. All participation was fully informed and voluntary. Subjects received an honorarium at the end of the experiment, a portion of which depended on task performance accuracy.
In a separate test session the WAIS-R was administered to each participant. On the WAIS-R, the total score and verbal and nonverbal ability subscores are age-group normalized with respect to a population mean of 100 (SD = 15). For entry into the study we required that participants have a WAIS-R total score of 85 or greater, and at least a high school diploma or equivalent (across the group, subjects had a mean of 2.5 years of post-secondary education, range 0–7 years). These requirements eliminated potential participants with WAIS-R scores below the average range. This was done for several reasons. First, we sought to avoid including subjects who might have clinically abnormal and hence easily distinguishable EEG patterns (undiagnosed brain damage is one source of below-average IQ test scores). Second, we sought to eliminate potential participants who might display excessively poor performance on the experimental tasks, because it is difficult to readily distinguish performance deficits that are related to ability from those that arise from a lack of effort or compliance with task demands. Together, these precautions minimized the possibility that results might only reflect abnormalities in the EEG or task noncompliance. In the final sample of 80 subjects, total IQ scores ranged from 94 to 149 (mean = 121, SD = 13.5), verbal scores ranged from 93 to 150 (mean = 119, SD = 13.8) and nonverbal scores ranged from 93 to 140 (mean = 117, SD = 13.0).
Experimental Tasks and Procedures
Subjects performed versions of an ‘n-back’ style spatial WM task (Gevins et al., 1990, 1996, 1997, 1998; Gevins and Cutillo, 1993; McEvoy et al., 1998; Smith et al., 1999). In this task the spatial position of an uppercase letter stimulus (drawn from a set of 12) was compared to the position of a stimulus that had appeared previously. Stimuli were presented on a computer monitor located 60 cm in front of the subject. At the beginning of each trial a warning cue, a small ‘x’, appeared in the center of the screen for 200 ms. A letter stimulus occurred 1.3 s after the onset of the warning cue in one of 12 locations on an imaginary circular grid. Stimuli were presented for 200 ms once every 4.5 s. The identity of the letter and its spatial position varied randomly from trial to trial. A small fixation dot was continuously present at the center of the screen.
In a difficult, high-load version of the task, subjects compared the current stimulus with the stimulus presented two trials previously. That is, subjects were required to maintain two positions (and their sequential order) in WM for the duration of two trials (9 s), and to update that information on each trial. In an easy, low-load version of the task, subjects were required to match the position of the current stimulus with the first one that appeared in a block of trials. In both versions of the task, stimuli were presented in blocks of 23 trials, with 50% matches. The first three trials in each block were excluded from analyses. For each block of trials subjects were informed as to whether it would be a high-load or low-load version of the task. They were requested to respond as quickly and as accurately as possible, and to indicate match detection with a right index finger key press, and non-match with the right middle finger.
In the experimental session subjects were first prepared for recordings as described below. The tasks and protocol were described, and then a block of 3 min of EEG data was recorded as the subjects rested quietly with their eyes open while they stared at a blank wall. They then performed eight blocks of the low load task (184 trials), followed by eight blocks of the high-load task. Feedback on performance was provided after every block, and subjects were given a brief break (~2 min) between blocks. At the end of the total of 16 blocks of trials another 3 min block of eyes-open resting data was collected.
Recordings
EEG was recorded continuously from 27 scalp locations (Fp1, Fp2, AF3, AF4, FT9, F7, F3, FZ, F4, F8, FT10, T7, C3, CZ, C4, T8, P9, P7, P3, PZ, P4, P8, P10, O1, OZ, O2, Iz) during task performance, using electrically linked mastoids as reference. Electrooculographic (EOG) activity was recorded from electrodes located in the center of the supraorbital ridge above each eye, referenced to an electrode at the outer canthus of each eye. Physiological signals were band-pass filtered at 0.01 to 100 Hz and sampled at 256 Hz. Automated artifact detection was followed by application of adaptive eye movement artifact removal filters (Du et al., 1994). The results of this process were then reviewed and edited by expert human judges.
Analyses
Fast Fourier transforms were calculated for all 2 s, contaminant-free EEG segments from each subject in each task condition and then averaged across segments to produce summary power spectra. Preliminary inspection of these data indicated that no relevant differences in the amplitude or frequency characteristics of the EEG existed between trials that required match versus non-match responses, so data were collapsed across this dimension. Rather than collapsing over all trials within a condition, data were then collapsed over the first four blocks (trials 1–80) and last four blocks (trials 81–160) in the low-load and high-load task conditions. This permitted subsequent examination of how EEG spectral features change both with practice or time-on-task ie. it allowed us to examine the degree to which the brain adapted to a new task demand (Gevins et al., 1997; Smith et al., 1999). The EEG data were also compared between the two WM loads. In addition, the EEG of each trial, aligned in time with the stimulus, was averaged over all correctly performed trials of each task variant and stimulus condition to produce average stimulus-locked ERPs to match and nonmatch trials of both the low-load and high-load tasks.
For statistical analyses, data from only a small subset of electrodes (those directly over lateral and midline prefrontal and parietal cortex) were used. Decisions concerning which electrodes to include in these analyses were based on results of prior studies. Power of the fm signal was measured at an anterior midline (Fz) electrode (Gevins et al., 1997). Some investigators have used characteristic frequency, scalp distribution and task correlates to distinguish between components of the rhythm. A lower frequency (8–10 Hz), or ‘slow’ component predominates over parietal and dorsolateral prefrontal association cortex and is highly sensitive to the cognitive load that a task imposes (Klimesch et al., 1993; Gevins et al., 1997). In contrast, a higher-frequency (10–12 Hz) component has a predominantly occipital or occipital-parietal distribution and is most sensitive to the visuospatial factors rather than the cognitive demands of a task. The focus in the current study was the lower-frequency (8–10 Hz) slow signal measured at left (F3) and right (F4) dorsolateral prefrontal, and left (P3) and right (P4) superior parietal electrodes. The fm and signals have been shown to have very high test–retest reliability when measured in the context of the WM tasks employed in this study (McEvoy et al., 2000). Amplitude and latency of the LPC of the ERP were measured at midline parietal electrode Pz (Gevins et al., 1996; McEvoy et al., 1998).
Analyses of individual differences had two complementary objectives. The first objective was to test the hypothesis that individual variations in EEG and performance measures in the WM task are related to general cognitive ability. The second objective was to compare subjects who had a relatively high verbal ability with subjects who had a relatively high nonverbal ability. To address these objectives, the total sample of 80 subjects was partitioned into subgroups based on their WAIS-R total, verbal ability and nonverbal ability subscores. In order to minimize variance in WAIS-R total scores that might be due to cultural, educational or other extraneous differences rather than general cognitive ability, analyses addressing the first objective were restricted to a subset of the total sample who had minimal differences between their verbal and nonverbal subscores. This was accomplished by first taking the difference between the verbal and nonverbal subscores for each subject, and dividing it by his or her total score. Participants in the top (relatively high verbal scores) and bottom (relatively high nonverbal scores) quintiles of the distribution of the resulting difference score had an average absolute difference between their verbal and nonverbal scores of 16.7 points (range 8–36). These extreme groups of participants were eliminated from initial analyses that focused on the relationship between task-related performance and EEG measures and cognitive ability. Across the original sample of participants, verbal and nonverbal subscores were correlated with an r = 0.59 (P < 0.01). After eliminating the participants with the largest discrepancies between sub-scores, the verbal and nonverbal subscores in the remainder of the sample (n = 48) were much more highly correlated (r = 0.91, P < 0.0001). That is, variation in the WAIS-R scores of this subset of participants was likely to be primarily due to differences in general cognitive ability rather than differences in domain specific cognitive style.
In order to identify task-related variables sensitive to individual differences in cognitive ability, the group of 48 participants with balanced verbal and nonverbal subscores was further subdivided into low, medium and high ability subgroups (n = 16 for each of three groups) based on total WAIS-R scores (see Table 1). In a secondary analysis (see Table 2) subjects in the two extreme verbal and nonverbal groups (n = 16 per group) were compared in order to isolate differences related to characteristic verbal or nonverbal cognitive style.
|
|
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionConclusionsNotesReferences |
|---|
Analyses of Task Effects
A summary of accuracy and reaction time (RT) measures for all experimental conditions appears in Table 3. A repeated- measures ANOVA using task load (high load versus low load) and practice (trials 1–80 versus trials 81–160) as factors indicated that accuracy (as measured by the ability to discriminate match trials from non-match trials, or d´ (Swets, 1964), was lower in the high load task [F(1,79) = 22.2, P < 0.001]. Accuracy increased with practice or time-on-task [F(1,79) = 85.1, P < 0.0001]. That is, accuracy was higher in the second set of 80 trials in each task than in the first set. RTs were log normalized prior to statistical analyses. A repeated-measures ANOVA using task load and practice as factors indicated RTs were longer in the high load version of the WM task [F(1,79) = 200.8, P < 0.0001], and decreased with practice [F(1,79) = 229.2, P < 0.0001]. There were no interactions between the load effects and practice effects.
|
Grand average EEG spectra (Fig. 1
) and stimulus-locked ERP measures displayed task-related modulation equivalent to that observed in past studies. For example, the task-related modulation of fm measured at electrode Fz was compared in a repeated-measures ANOVA using task load (high load versus low load) and practice (trials 1–80 versus trials 81–160) as factors. Data for each subject was converted to z-scores across these conditions prior to this analysis. The fm signal was larger in the high-load task than in the low-load task [F(1,79) = 57.7, P < 0.001]. It also increased marginally with practice [F(1,79) = 2.9, P < 0.10]. There was no interaction.
|
The 8–10 Hz
signal was also compared across task conditions and between electrode sites (F3, F4, P3, P4) including cortical hemisphere (left versus right) and lobe (frontal versus parietal) as factors in a repeated-measures ANOVA. This signal was larger in the low-load task [F(1,79) = 37.0, P < 0.001], and larger after practice [F(1,79) = 25.1, P < 0.001]. It was also larger over the parietal lobes than over the frontal lobes [F(1,79) = 11.2 P < 0.002], and larger over the right hemisphere than over the left [F(1,79) = 13.0, P < 0.002]. In order to examine interaction effects independently of differences in amplitude between subjects or between electrode sites, the data from each subject at each electrode site was converted to z-scores across the different test conditions. These normalized results displayed an interaction between the task load and practice factors [F(1,79) = 11.8, P < 0.002], with a larger load-related change in power after subjects had practiced the tasks. They also displayed an interaction effect between the task load and cortical lobe factors, [F(1,79) = 18.0, P < 0.001], with larger load-related changes in amplitude over the parietal lobe than over the frontal lobe. No other interaction effects reached significance. The latency and amplitude of the stimulus-locked LPC of the ERP measured at the parietal midline electrode (Pz) was compared between the two task loads and between trials on which the eliciting stimulus was either a match or a nonmatch with respect to the criterion stimulus. LPC latency was shorter on match (mean = 340 ms, SD = 6 ms) trials than on nonmatch trials (mean = 355 ms, SD = 6 ms). No other latency differences were significant. There was no overall effect of task load on LPC amplitude [F(1,79) = 1.6, NS]. However, there was a significant difference in LPC amplitude between match and nonmatch trials [F(1,79) = 13.9, P < 0.001], and an interaction between the load and match status factors [F(1,79) = 24.5, P < 0.001]. Analysis of simple effects indicated that this interaction arose from the fact that, in the low-load task, LPC was larger on match (mean = 11.2 µV, SD = 4.6) than on nonmatch trials (mean = 9.5 µV, SD = 4.3; P < 0.001). In contrast, in the high-load task, no amplitude difference was observed between match (mean = 10.0 µV, SD = 4.7) and nonmatch trials (mean = 9.9 µV, SD = 4.9).
Finally, subsidiary analyses were performed in which gender was included as a between-subjects factor. None of the task-related behavioral or EEG variables significantly differed between men and women. There was, however, a significant gender difference in EEG power under resting conditions. In particular, the signal recorded at parietal sites (P3, P4) was larger in power [F(1,78) = 7.6, P < 0.01] in women than men (mean = 31.9 versus 29.0 dB respectively). There were no significant gender differences in resting alpha asymmetry.
Relation of Performance and Task-related EEG to General Cognitive Ability
To initially examine how subjects differed with respect to their task-related performance and brain activity as a function of ability level, mixed-effects ANOVAs were used to compare the measures between the low, medium and high ability groups (n = 16 per group). These groups did not significantly differ in age (mean age = 22.5, 21.6, 20.7 respectively for the three groups) [F(2,45) = 1.3, NS], nor in number of years of post-secondary education (mean = 3.0, 2.6, 2.2 years respectively) [F < 1, NS]. They also did not significantly differ in gender composition (male/female frequency = 8/8, 7/9, 10/6 respectively) [
(1) < 1, NS].
All of the significant main effects of the task-related manipulations reported above for the entire sample of 80 subjects were also significant in the analysis of 48 subjects; thus, those effects will not again be described here. Rather, the focus in this section is on differences in ability groups. Table 1
shows WAIS-R scores and task-related accuracy and RT (collapsing across task conditions) for each of the three groups. Not surprisingly, the three groups differed significantly on WAIS-R total score [F(2,45) = 105.2, P < 0.0001], as well as on verbal [F(2,45) = 63.5, P < 0.0001] and nonverbal [F(2,45) = 117.3, P < 0.0001] subscores. Furthermore, in post hoc analyses all pairwise comparisons of these scores significantly differed between groups [P < 0.001 for all pairwise comparisons]. Since the ‘digit span’ subtest of the WAIS-R might be seen as providing a simple measure of WM capacity, scores on this subtest were also compared between the three groups. There was an overall effect of group on digit span score [F(2,45) = 13.1, P < 0.001], with lowest digit span in the low-ability group, and highest in the high-ability group (mean scores of 9.8, 12.6 and 13.0 respectively). Post hoc pairwise analyses indicated that while both the middle- and high-ability groups had a higher digit span than the low-ability group (P < 0.001 for each pairwise comparison), the two higher-ability groups did not differ on this measure.
When comparing task performance between groups, no significant main effect of ability group on accuracy was obtained [F(2,45) = 2.1, NS], nor did ability group interact with the task load or practice factors with respect to accuracy. This insensitivity probably reflects a ceiling effect on the accuracy measure, as most subjects made few errors. In contrast, RTs were fastest in the high-ability group, and slowest in the low-ability group [F(2,45) = 5.4, P < 0.01]. Again, there were no interactions involving the grouping factor. Post hoc pairwise analyses indicated that both the middle- and high-ability groups had faster RTs than did the low-ability group (P < 0.05 and P < 0.001 respectively for the two pairwise comparisons with the low- ability group). However, the middle- and high-ability groups did not significantly differ on this measure.
When considering the LPC of the ERP, there were no significant differences between the groups in LPC latency. However, there were significant group differences in LPC amplitude. In particular (Fig. 2), although the low-ability (mean amplitude 8.2 µV, SD 2.8 µV) and middle-ability (mean amplitude 8.0 µV, SD 2.8 µV) groups did not differ in amplitude (F < 1), the high-ability group displayed a much larger (mean amplitude 13.3 µV, SD 3.8 µV) LPC than did the other two groups [F(1,46) = 29.2, P < 0.0001]. This group difference was observed across conditions and did not interact with task load or stimulus match status factors.
|
When comparing EEG measures, the groups did not differ in overall amplitude of fm
. However, the three groups differed in the fashion by which task manipulations affected this variable (Fig. 3). In particular, when considering each group separately, the high-ability subjects displayed a significant interaction between the task load and practice factors [F(1,15) = 7.3, P < 0.02], that was not observed in the other two groups (both Fs < 1). This interaction in the high-ability group resulted from a large practice-related increase in fm power in the high load task that was not observed in the low load task. In other words, with practice on the high-load task, the high-ability group displayed a significant increase in power in the fm rhythm. This phenomenon was not observed in the low-load task condition or for either of the other two ability groups.
|
Similarly, although the three groups did not differ in the overall amplitude of the 8–10 Hz component of the
rhythm, they differed in the way in which it was modulated by variations in the task conditions. In particular, including the grouping factor in a mixed-effects, repeated-measures ANOVA that also included load, practice, cortical hemisphere (left versus right) and lobe (frontal versus parietal) factors, a significant three-way interaction was obtained between the group, practice and lobe factors [F(2,45) = 3.7, P < 0.04]. This interaction (Fig. 4) reflected the fact that subjects in the high-ability group displayed a relatively larger practice-related increase in power over the frontal region [F(1,15) = 6.8, P < 0.02], whereas subjects in the low-ability group displayed a trend towards a relatively larger, practice-related increase over the parietal region [F(1,15) = 3.2, NS]. In contrast, the middle-ability group displayed an equal practice-related increase in amplitude over both frontal and parietal regions (F < 1). Finally, although there was no overall significant interaction between the group, load and lobe factors (F < 1), the three groups differed somewhat in the manner in which task load affected activation of the frontal versus parietal lobes. In particular, for the high-ability subjects the load related attenuation of the rhythm over the parietal lobes was significantly larger in magnitude than the corresponding attenuation over the frontal lobes [F(1,15) = 9.6, P < 0.01]. This difference was not significant for the middle-ability [F(1,15) = 3.8, NS], or low-ability (F < 1) groups. In sum, these group differences in how the rhythm was modulated by task practice and load suggest that the high-ability subjects relied on strategies that made relatively high demands on parietal regions relative to frontal regions. This anterior to posterior difference in task-related brain activation was less apparent for the lower-ability subjects.
|
To illustrate the relationships between variables, Pearson product–moment correlation coefficients were computed between WAIS-R total scores, WM task accuracy and RT, and neurophysiological measures derived from the group differences in the analyses reported above (Table 4
). Consistent with the analyses described above, the average RT across task conditions was inversely correlated with WAIS-R score. Several of the neurophysiological variables were also correlated with WAIS-R score. For example, a variable coding the practice-related increase in fm during the high-load task was positively correlated with WAIS-R score. A variable coding the difference in magnitude between the practice-related increase of the rhythm over the frontal region and the analogous measure made over the parietal region was also positively correlated with WAIS-R score, as was a variable coding the amplitude of the LPC collapsed across task conditions. A variable coding the difference in magnitude between the load-related decrease of the rhythm over the frontal region and the analogous measure made over the parietal region was negatively correlated with WAIS-R total score. Neither task performance accuracy nor LPC latency was significantly correlated with WAIS-R score. RT was negatively correlated with task performance accuracy and with two neurophysiological variables: LPC amplitude and the difference in magnitude between the practice-related increase of the rhythm over the frontal region and the analogous measure made over the parietal region. None of the neurophysiological variables were significantly intercorrelated. Finally, in addition to being positively correlated with overall WAIS-R score, scores on the digit span subtest were negatively correlated with RT, and positively correlated with accuracy. Digit span was also positively correlated with the variable coding the practice-related increase in fm during the high-load task, the variable coding the difference in magnitude between the practice-related increase of the rhythm over the frontal region and the analogous measure made over the parietal region, and LPC amplitude.
|
Stepwise multiple regression was used to derive multivariate functions for predicting WAIS-R scores from combinations of task-related behavioral, EEG and ERP variables. These functions were derived from a set of measures that included first-order predictor variables indexing performance speed and accuracy, EEG
and fm power, and the amplitude of the LPC. They also included second-order (derived) predictor variables. The second- order variables indexed changes in the first-order measures between high-load and low-load task conditions, between practiced versus naïve conditions, or between left and right hemispheres or frontal and parietal lobes ie. the variables described in the correlation analysis presented above. All regression functions were limited to a maximum of eight predictor variables in order to maintain a conservative 6:1 ratio of observations (subjects) per variable. A stepwise analysis using only behavioral variables yielded an eight variable function with a multiple R = 0.56 [R2 = 0.32, F(8,39) = 2.27, P < 0.05; Fig. 5a]. An analogous analysis restricted to EEG and ERP variables yielded an eight variable function with a multiple R = 0.73 [R2 = 0.53, F(8,39) = 5.59, P < 0.001; Fig. 5b]. When both behavioral and neuroelectric indices were included in an analysis, a multiple R = 0.80 [R2 = 0.64, F(8,39) = 9.0, P < 0.0001; Fig. 5c] was obtained with an eight variable function. The variables included in this combined function included average RT across task conditions, LPC amplitude to match stimuli in the low-load task, three second-order fm EEG variables, and three second-order EEG variables. The second-order fm variables included a variable coding the practice-related increase in fm during the high-load task, a variable coding the load-related increase in fm across all trials, and a variable coding the load-related increase in fm across just the last 80 trails in each series. The second-order variables included a variable coding the difference in magnitude between the load-related decrease of over the frontal region and the analogous measure made over the parietal region, a variable coding the difference in magnitude between the load- related decrease of at electrode site P3 on trials 1–80 versus trials 81–160, and a variable coding the magnitude of resting asymmetry in amplitude between left and right parietal sites.
|
The final combined behavioral and neuroelectric regression function was submitted to a leave-out-one jackknife cross- validation analysis (Efron, 1982
) to test whether the findings would generalize to data not used for deriving the regression weights. In this analysis, 47 of the participants were used to derive the function weights, and the resulting equation was used to estimate the IQ test score of the remaining participant. This procedure was performed over 48 iterations so that each participant's score could be estimated as an independent test sample. This cross-validation analysis produced a correlation between the observed and estimated WAIS-R scores of R = 0.71 (R2 = 0.50, P < 0.001). Differences between Subjects with Verbal versus Nonverbal Cognitive Styles
A second set of analyses focused on the participants who were eliminated from the preceding analysis because they displayed relatively large differences between WAIS-R verbal and non- verbal subscores. That is, for the two groups of participants (n = 16 per group) whose test performance indicated that they had a disproportionately high verbal or nonverbal ‘cognitive style’ (Table 2), we sought to determine whether there were corresponding differences in their task-related behavior and/or brain activity. These groups did not significantly differ in age (mean age = 20.8 and 21.5 respectively for the two groups) (F < 1, NS), nor in number of years of post-secondary education (mean = 2.0 and 2.7 years respectively) [F(1,30) = 1.7, NS]. They also did not significantly differ in gender composition (male/female frequency = 7/9 and 10/6 respectively) [(1) < 1, NS]. Furthermore, the groups did not differ in WAIS-R total score, and although the Nonverbal group had slightly higher accuracy and slightly faster RTs than the Verbal group, these differences did not reach significance (both Fs < 1).
Although the high nonverbal group had a shorter peak latency of the LPC of the ERP than the high verbal group, this difference was not significant [F(1,30) = 1.6, NS]. However, average amplitude of the LPC was larger in the high nonverbal group (mean 11.5 µV, SD 3.8 µV versus mean 8.6 µV, SD 3.0 µV) than the high verbal group [F(1,30) = 5.9, P < 0.03]. The groups did not differ in the absolute magnitude of the EEG power measures. However, they differed with respect to hemispheric asymmetries of band signals (Fig. 6). More specifically, the high verbal ability group displayed a relatively large asymmetry in at parietal electrodes, with a smaller signal over the left hemisphere. In contrast, the high nonverbal ability group displayed a relatively small asymmetry in the opposite direction, with a smaller signal over the right parietal region [F(1,30) = 8.9, P < 0.01].
|
activity between the left and right frontal versus parietal lobes for high verbal and high nonverbal groups. The y-axis values reflect normalization of each participant's data by computing z-score transforms across task conditions and across the two contralateral homologous electrodes in the frontal (F3, F4) or parietal (P3, P4) regions. Negative scores indicate relatively greater In sum, the two extreme groups were equivalent in terms of their WAIS-R total scores. However, neurophysiological measures indicate that they had different patterns of neural activation during task performance. To determine whether these differences were sufficiently characteristic of the two groups to permit classification of individual participants into one group or the other, stepwise LDA was used to derive a classification function, and this classification function was again cross- validated in a leave-out-one jackknife analysis. The final four-variable discriminant function thus derived included a measure of task-related parietal
asymmetry across the task conditions (cf. Fig. 6), measures of LPC amplitude to non- matching trials in both high-load and low-load task conditions, and performance accuracy in the high-load task condition. The original equation successfully classified 28/32 participants (87.5%); 27/32 participants (84.4%) were successfully classified in the cross-validation analysis (binomial P < 0.0001 for both results). |
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionConclusionsNotesReferences |
|---|
The effects on behavioral and EEG measures produced by increasing WM load in the type of task utilized herein have been described in detail elsewhere. Briefly, across-subjects accuracy was lower, and RTs longer, in the high-load WM task relative to the low-load task. The apparent increase in task difficulty with higher WM load was associated with an increase in power in the fm
EEG signal, and a decrease in power in the band signal (Gevins et al., 1997, 1998). Practice or time-on-task also affected the amplitude of these features of the EEG (Gevins et al., 1997; Smith et al., 1999). The effect of WM load on the amplitude of the LPC of the stimulus-locked ERP differed depending on whether the current stimulus matched the criterion stimulus for the task (Gevins et al., 1996; McEvoy et al., 1998). In sum, these aspects of the current results served to replicate previous findings and extend them to a larger sample. The major purpose of the current study was to determine whether and how such indices of WM-related functional brain activation might covary with individual differences in cognitive ability. This was accomplished by comparing neurophysiological and behavioral data obtained during WM task performance to scores on the WAIS-R. Three types of findings were obtained. First, as might be expected, individual differences in general cognitive ability as measured by the WAIS-R were correlated with performance measures on the WM task. Second, neuro- physiological measures of WM function appear to be good predictors of general cognitive ability as measured by standard psychometric instruments. Third, such measures are useful for distinguishing individuals with relatively high verbal aptitude from those with a relatively high nonverbal aptitude, and they characterize those individuals in terms of relative utilization of left and right cerebral hemispheres. These findings are discussed below.
Neurophysiological Measures of WM and Individual Differences in Cognitive Ability
A large number of prior studies have examined relationships between EEG or ERP measures and individual differences in cognitive ability [for historical reviews see (Callaway, 1975; Gale and Edwards, 1986; Dreary and Caryl, 1997)]. Such studies have typically assumed that individuals with high general cognitive ability would have brains that were either faster at processing information, or endowed with a greater processing capacity, or more efficient and adaptable. However, in most such studies, brain function was measured either while participants sat passively or while they attended to stimuli in tasks that made only trivial demands on their cognitive abilities. Hence, the majority of past results are somewhat equivocal with respect to interpretation in terms of these constructs. In contrast, the current study involved testing subjects under more challenging task conditions. High-ability subjects performed the WM tasks faster without compromising accuracy. The exceptional task performance for this group is consistent with the notion that performance on both WM tasks and psychometric tests of cognitive ability reflect contributions from some common mental resource (Carpenter et al., 1990; Kyllonen and Christal, 1990; Engle et al., 1999).
In the current study, several task-related neurophysiological measures were correlated with WAIS-R total scores. For example, the amplitude of the LPC of the ERP was positively related to individual differences in cognitive ability. This is consistent with the observation by Nittono and colleagues (Nittono et al., 1999) of a larger amplitude LPC in a relatively complex discrimination task in subjects with a greater WM capacity as measured by a reading span task. No significant correlation between LPC latency and test scores was found in either the current study or the study by Nittono et al. In contrast, Polich and colleagues have reported a negative correlation between LPC peak latency and both the digit span measure of WM capacity (Polich et al., 1983) and overall academic performance (Polich and Martin, 1992), but no association between individual differences in ability and LPC amplitude. In these later studies a simple ‘auditory odd- ball’ task was employed that for most healthy average ability individuals requires minimal mental effort to achieve perfect performance.
These differences indicate that any claims about the relationship between ERP latency and amplitude measures and individual differences in cognitive ability must be qualified with respect to the eliciting task employed. Given this qualification, the positive correlation observed herein between LPC amplitude and WAIS-R scores (as well as scores on the digit span subtest of the WAIS-R) is consistent with the notion that differences in ability might reflect differences in how individuals allocate attention and WM resources to task performance. That is, the LPC is generally found to be larger when attention is exclusively focused on analyzing the eliciting stimulus, and smaller when attention is consumed by some other mental activity (Kramer et al., 1987; Sirevaag et al., 1989; Picton, 1992; Gevins et al., 1996). A positive correlation between cognitive ability and LPC amplitude might indicate that high-ability subjects had greater capacity to allocate to the task, or that they were better able to concentrate and inhibit thoughts that might have competed for their cognitive resources, or both.
With respect to task-related spectral features of the ongoing EEG, no significant relationships were identified between the absolute power of the attention-related fm and signals and WAIS-R test scores. Given that the absolute power of the EEG is affected by variables such as skull thickness and bone density (which presumably have little relationship to cognitive ability), these negative findings are of little interest. Of more note is that variations in WAIS-R scores were associated with between- subject differences in how the fm and signals were regionally modulated by variations in task demands.
For example, high-ability subjects displayed a proportionately larger practice-related increase in the fm EEG signal than did the low-ability subjects, an effect seen in the high-load task but not in the low-load task. The fm signal is largest under conditions that require sustained mental effort (Miyata et al., 1990; Yamamoto and Matsuoka, 1990; Gundel and Wilson, 1992; Gevins et al., 1997, 1998). It appears to be generated in the anterior cingulate region of the medial frontal cortex (Gevins et al., 1997; Ishii et al., 1999), an important node of the cortical network involved with attention control (Posner and Peterson, 1990; Posner and Rothbart, 1992). The observed difference in fm between task conditions and subjects thus suggests that members of the high-ability group were better able to exert additional effort to focus and sustain attention in response to challenging task demands than were members of the other groups. Given that most demanding problem-solving activities require a sustained effort to focus concentration, individual differences in the ability or willingness to persistently exert such effort may play a key role in explaining individual differences in task performance outcomes. These results thus provide further support for the general notion that frontal-lobe dependent executive functions involved with working memory and controlled attention are closely related to general cognitive ability (Engle et al., 1999).
High-ability subjects also displayed a proportionately larger practice-related increase in activity over dorsolateral frontal cortex, and a proportionately smaller practice-related increase in activity over dorsolateral parietal cortex, than did low-ability subjects. Such regional variations in practice-related changes in power might be expected given that is enhanced differentially over particular cortical regions in conjunction with task-specific forms of strategy development and skill acquisition (Smith et al., 1999). Increased activity over a region of cortex suggests a relative decline in the proportion of local cortical neurons that are activated by task performance (Gevins and Schaffer, 1980; Pfurtscheller and Klimesch, 1992). A practice- related increase in activity is thus consistent with the finding from neuroimaging studies that cortical regions that are less essential to task performance become progressively deactivated as strategies become refined and partially automated (Risberg et al., 1977; Haier et al., 1992; Jenkins et al., 1994; Raichle et al., 1994).
In the current context, the data suggest that when high-ability subjects were initially confronted with the novel spatial WM tasks, their dorsolateral frontal regions were activated. As they developed an effective strategy for task performance, they began to rely relatively less on frontal regions for task performance, and relatively more on parietal regions. In contrast, low-ability subjects tended to show the opposite pattern of practice-related regional changes. Such results suggest that the high-ability individuals learned to adopt task performance strategies that took advantage of distributed cortical processing resources. In contrast, the low-ability subjects tended to adopt strategies that relied more exclusively on limited capacity frontal lobes circuits for organizing information in WM. Metabolic studies have indicated that the parietal regions tend to play an important role in both the spatial allocation of attention and in mental transformations of spatial relations (Alivisatos and Petrides, 1997; Coull and Nobre, 1998; Zacks et al., 1999). The current results thus may indicate that the high-ability subjects quickly learned to exploit these capabilities of the posterior cortex in order to perform the tasks more efficiently. This interpretation is also consistent with the subsidiary finding that, as task load increased, high-ability subjects displayed a proportionately greater attenuation of power over parietal regions than did low-ability subjects. More generally, these results suggest that higher-ability subjects tend to rapidly identify strategies that make more optimal use of the wide array of the cortical resources that are available to them than do the strategies that tend to be adopted by lower-ability subjects.
Neurophysiological Differences between Individuals with Verbal versus Nonverbal Cognitive Styles
In the population as a whole, the verbal and nonverbal subscales of the WAIS-R tend to be highly correlated. However, a variety of factors can contribute to differences between these scores. For example, the WAIS-R tests that are most highly correlated with the verbal score (the vocabulary and information sub-tests) assess what has been referred to as ‘crystallized intelligence’ (Cattell, 1971) ie. the stable declarative knowledge that a person has consolidated through past experience. Performance on such tests tends to be affected by cultural differences (Dreshowitz and Frankel, 1975; Tsushima and Bratton, 1977). Performance on tests of crystallized knowledge also tends to be highly correlated with amount of education (Denny and Thiessen, 1983), and shows little decline with advancing age (Denny and Thiessen, 1983; Kaufman et al., 1991; Isingrini and Vazou, 1997). In contrast, tests that measure processing speed and capacity, attentional flexibility and concentration, and nonverbal reasoning ability tend to be less sensitive to educational differences, and are more sensitive to age-related cognitive decline (Denny and Thiessen, 1983; Stankov, 1988; Kaufman et al., 1991; Isingrini and Vazou, 1997). Brain damage can also produce large discrepancies between the two subscores (Lezak, 1983). For example, tests of crystallized knowledge have been reported to be fairly insensitive to frontal lobe injury when compared with tests of nonverbal ability (Duncan et al., 1995). Finally, some discrepancies between verbal and nonverbal abilities would be expected from normal variation between subjects in the anatomical and functional organization of their brains.
In the current study we found that the subgroup with relatively high verbal ability did not differ from the subgroup with relatively high nonverbal ability in terms of gender composition, demographics or overall WAIS-R score. Although those subjects with a nonverbal cognitive style were slightly faster and more accurate in the WM task performance, the two groups did not significantly differ on these indices. Furthermore, the absolute magnitude of task-related EEG power for the fm and signals did not discriminate between groups, suggesting that their gross EEG biophysical characteristics were quite similar.
Nonetheless, a few measures of brain function did differ enough to permit members of the two groups to be reliably discriminated based solely on behavioral and neurophysiological measures made during WM task performance. For example, the high nonverbal group displayed higher amplitude in the LPC of the ERP than the high verbal group did. As noted above, the LPC is typically larger when attention is exclusively focused on the eliciting stimulus rather than shared among competing activities; a larger LPC in the high nonverbal group suggests that those subjects were more effective at directing attention to stimulus processing.
This group difference in LPC amplitude was accompanied by a group difference in the asymmetry in the task-related rhythm measured over the parietal cortex of the left and right hemispheres. Many prior studies have noted between-task differences in patterns of left–right EEG asymmetry. Most such asymmetries appear to be due to the specific perceptuomotor requirements of the tasks being compared (Gevins et al., 1979b, 1980; Gevins and Schaffer, 1980). Other studies have shown differences in hemispheric asymmetry between groups of subjects who are performing the same tasks.
For example, using blood flow imaging methods Wendt and Risberg (Wendt and Risberg, 1994) found that subjects who displayed better performance on spatial tasks had greater activation of right versus left posterior brain regions than did subjects with relatively poor performance. Similarly, the current study demonstrated differences in EEG hemispheric asymmetry between groups of subjects who are performing the same tasks, and who differ only with respect to their psychometric aptitude for verbal versus nonverbal activities. In particular, the high verbal group displayed a smaller signal over the left parietal area, and the high nonverbal group displayed a smaller signal over the right parietal area. Given that the rhythm is attenuated by cortical activation, this pattern of results suggests that the high verbal group tended to engage the left hemisphere relatively more during task performance, whereas the high nonverbal group tended to engage the right hemisphere more. Together, these differences are consistent with prior reports that individuals with different ‘cognitive styles’ might vary with respect to their characteristic patterns of hemispheric engagement during task performance (Butler, 1988; O'Boyle et al., 1991; Banich et al., 1992; Wendt and Risberg, 1994).
|
Conclusions |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionConclusions |
|---|