Personality Predicts Brain Responses to Cognitive Demands

Introduction

Eysenck's model of personality (Eysenck, 1967) proposes that introverts have lower response thresholds and are consequently cortically more aroused than extraverts. It postulates further an inverted U-shaped relationship between cognitive performance and “level of arousal,” jointly determined by environmental arousal potential (defined in terms of a range of environmental manipulations and task parameters) and subject arousability reflected in questionnaire-measured extraversion (E) scores. These postulates jointly predict that, at low environmental arousal potential, introverts (low E scorers) should outperform extraverts (high E scorers); as environmental arousal increases, performance should improve and extraverts should catch up with introverts, and, at high levels of environmental arousal, performance of introverts should decline and extraverts should outperform introverts until the level of arousal is high enough to evoke transmarginal inhibition (TMI) (Gray, 1964; Eysenck, 1994). With evocation of TMI, introverts may experience lower increments in arousal than extraverts. These predictions are widely supported by data from a range of cognitive paradigms (Eysenck, 1981). At the neural level, Eysenck's model postulates that the level of arousal (jointly determined by environmental arousal and subject arousability) is mediated by activity in a “cortical arousal system” modulated by reticulothalamic- cortical pathways (Eysenck, 1967, 1981).

We examined the association between E, as assessed with the Eysenck Personality Questionnaire-Revised (EPQ-R) (Eysenck and Eysenck, 1991), and brain activity during an “n-back” task with four cognitive loads and a rest condition. The n-back task activates a network of frontal [particularly the dorsolateral prefrontal cortex (DLPFC) and anterior cingulate (AC)] and parietal regions (Smith and Jonides, 1997). The DLPFC is considered specialized for noting task-relevant contents of memory (MacDonald et al., 2000), anterior cingulate for on-line monitoring, error detection, and response execution (Botvinick et al., 2001), and the parietal cortex for short-term storage (Gathercole, 1994). Of particular relevance is that DLPFC and AC may modulate cortical arousal via their connections to the midline thalamic (Barbas et al., 1991) and brainstem monoamine nuclei (Crino et al., 1993). The prefrontal cortex may be particularly sensitive to E-related influences (Lieberman, 2000), because the pontine reticular nucleus projects to this region. A positive association of E with central executive efficiency (prefrontal cortex), but not storage (parietal cortex), supports this suggestion (Lieberman and Rosenthal, 2001). Finally, the AC and DLPFC are activated by a range of cognitive tasks pointing to their prominent role in cognitive processing (Duncan and Owen, 2000), presumably impacting the same cortical arousal system theorized by Eysenck (1967, 1981) to be associated with E.

In the present experiment, we observed no differences in behavioral performance as a function of E. Performance equivalence between introverts and extraverts implies that the latter increase subject arousal to a greater extent than the former from conditions of low to high environmental arousal. This, in theory, would require extraverts to mobilize more cortical resources than introverts (who should have a relatively higher level of resting cortical arousal). Assuming that activity in DLPFC and AC reflects subjects' performance-related cognitive arousal level, we predicted that, the greater the E score, the greater the increase from resting levels.

Materials and Methods

Results

Intercorrelations of E, N, and P suggested a modest yet nonsignificant relationship of E to P (r = 0.41; p = 0.21) and N (r = 0.32; p = 0.34) but no relationship between P and N (r = 0.04; p = 0.90). Mean ± SD response accuracy across all subjects was as follows: 0-back, 94.42 ± 9.86%; 1-back, 88.96 ± 13.53%; 2-back, 84.76 ± 19.56%; and 3-back, 45.30 ± 15.91%. Mean ± SD response latency across all subjects was as follows (in msec): 0-back, 190.05 ± 74.68; 1-back, 92.84 ± 95.03; 2-back, 113.35 ± 142.20; and 3-back, 147.37 ± 166.72. The level of performance shown by our subjects was very comparable at the 2-back with that reported in a recent study of normal healthy subjects, using a task similar to ours (Glabus et al., 2003) (accuracy, 83 ± 14.30%; 3-back condition not used) and at the 3-back condition, on average, was above chance level (which was equal to 25%). E, N, or P scores were not significantly associated with response accuracy or latency at any load (p > 0.20 for all tests), except for a marginally significant negative association between N and response accuracy at the 1-back level (r = -0.595; p = 0.054).

Cognitive activations: effects of E

None of the analyses revealed associations between E and load-related activations in regions other than right DLPFC and AC (see below) or between N or P scores and load-related activation in any region.

Increasing E predicted increasing activity from rest to the 3-back condition in right DLPFC [(44, 34, 26) (x, y, z); F_(1,9) = 10.26, p = 0.01; adjusted R² = 0.48; β = 0.73, p = 0.01 (Fig. 1a); β = 0.63, p = 0.05 after controlling for P and N; β = 0.64, p = 0.04 after controlling for response accuracy and latency at the 3-back] and AC [(-6, 20, 38); F_(1,9) = 8.02, p = 0.02; adjusted R² = 0.41, β = 0.69, p < 0.02 (Fig. 1b); β = 0.85, p = 0.02 after controlling for P and N; β = 0.77, p = 0.03 after controlling for response accuracy and latency at the 3-back]. For AC, but not DLPFC, E also predicted increased activity from rest to 1-back (F_(1,9) = 7.82, p = 0.02; adjusted R² = 0.41, β = 0.68, p = 0.02; β = 0.55, p = 0.10 after controlling for P and N; β = 0.60, p = 0.06 after controlling for response accuracy and latency at the 1-back) and to 2-back (F_(1,9) = 6.06, p = 0.04; adjusted R² = 0.40, β = 0.63, p = 0.04; β = 0.57, p = 0.11 after controlling for P and N; β = 0.62, p = 0.05 after controlling for response accuracy and latency at the 2-back). These results demonstrate that, in both AC and DLPFC, E strongly determined the change from rest to the most cognitively demanding 3-back condition but did not significantly relate to the degree of change from rest to the least demanding 0-back condition (Fig. 1a,b). These data can be taken to suggest a dose-response relationship between task-induced fMRI activity and E, although the difference between the slopes for the relationship of E to change from rest to 3-back and for the change from rest to 0-, 1-, or 2-back conditions failed to attain a conventional level of significance when tested formally using Fisher's exact z tests. In the ANOVA analysis, right DLPFC and AC were similarly modulated by E (load, F_(3,18) = 25.92, p = 0.001; E, F_(1,6) = 8.34, p = 0.03; E × linear component of load, F_(1,6) = 5.70, p = 0.05; no main or interactive effects involving region): high E subjects showed a greater fMRI signal change from rest to 3-back through 0-, 1-, and 2-back conditions (Fig. 1c,d) than low E subjects, who did not show an increase in brain activity during the low cognitive load, 0-back condition relative to rest, implying relatively higher brain activity during the rest condition itself and thus an overall lower increase with cognitive load conditions. The differential increase in response in AC and right DLPFC, but not in the parietal lobe, as a function of E scores in three representative subjects is illustrated in Figure 1e.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Increase in activity from rest to 0-, 1-, 2-, and 3-back in right DLPFC (a) and AC (b) as a function of individual E score. Memory load-related fMRI signal changes (increases from rest) in right DLPFC (c) and AC (d) in groups with high (≥17) and low (≤9) E scores. e, Activation, thresholded at p < 0.001 (uncorrected), in three individual subjects with E scores of 4 (left), 15 (middle), and 21 (right), superimposed on their own individual structural images. I, Introversion. Color bar shows SPM-derived T values. Task load-related activity in the DLPFC and AC, but not in superior parietal cortex, varied systematically with E. Axial view with associated z coordinates is shown; left hemisphere is shown on the left.

Resting fMRI signal intensity: effects of E, N, and P

Localized personality effects were evident in the globally normalized resting signal. The areas identified consisted of a region including both Wernicke's and Broca's areas [centered at (-40, -40, 10); Brodmann areas 22, 44, and 45] and the cuneus [(2, -70, 6); Brodmann area 17], both negatively related to E (Fig. 2a-c). At a lower threshold (p < 0.005; uncorrected), we found additional negative relationships: of E to signals in the left [(-8, -30, 10)] and right [(8, -34, 6)] thalamus (Fig. 2d,e); of P to signals in the left [(-18, -4, -4)] and right [(12, -10, -14)] globus pallidus, extending to the putamen (Fig. 3a-c); and of N to signals in the left [(-32, 58, 2)] orbitofrontal cortex (β = 0.91).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Localized personality correlates in the globally normalized resting signal intensity related to E. a, Regression map in SPM-derived T values thresholded at p < 0.005 (uncorrected) and superimposed on the average structural image. Color bar shows the strength of the correlation with E. Axial view with associated z coordinates is shown; left hemisphere is shown on the left. Localized resting fMRI signal in relation to intersubject local mean intensity (y-axis) in Wernicke's area (b), cuneus (c), and left (d) and right (e) thalamus as a function of E (x-axis). Graphs show SPSS-derived r values and probabilities.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Localized personality correlates in the globally normalized resting signal intensity related to P. a, Regression map in SPM-derived T scores thresholded at p < 0.005 (uncorrected) and superimposed on the average structural image. Color bar shows the strength of the correlation with P. Axial view with associated z coordinates is shown; left hemisphere is shown on the left. Localized resting fMRI signal in relation to intersubject local mean intensity (y-axis) in left (b) and right (c) global pallidus as a function of P (x-axis). Graphs show SPSS-derived r values and probabilities.

Discussion

We demonstrated powerful effects of questionnaire-measured E on brain activity in a parametric n-back task. Eysenck's model (1981), coupled with the assumption (Duncan and Owen, 2000) that AC and DLPFC form part of a general cognitive arousal system, can predict the major relationship that we observed between task-related activity and personality: the incremental activity in these regions with increasing cognitive load (from 0- to 3-back) was itself an increasing function of E. The n-back task involves sustained attention, encoding of information into working memory, active maintenance of stimulus representations, response selection, and updating of sequential order information. Correlations between task performance and BOLD activity, considered on their own, are therefore susceptible to interpretation in terms of any of these functions. However, the relationship reported here between DLPFC-AC activity and questionnaire-measured E scores [themselves related to a diversity of other behavioral and physiological measures (Eysenck, 1981)] suggests that only a very broad functional role, such as the influence of cortical arousal in effectiveness of cognitive processing, can account for these correlations. Also, consistent with Eysenck's model, which treats N and P as independent of E, these dimensions did not influence task-related activity.

There have been three previous studies of E or related traits and brain responses during cognitive activation paradigms. The first study (Haier et al., 1987) noted a positive relationship between EPQ-R E scores in 18 patients with generalized anxiety disorder and nine healthy subjects during a continuous performance test and regional glucose uptake in the putamen, the caudate nucleus, the cingulate gyrus, and the hippocampal and parahippocampal gyrus. The effects of this study are broadly consistent with our data in showing increased cingulate activity in extraverts relative to introverts. The second study (Stenberg et al., 1993) reported marked asymmetry in data collapsed across several brain regions in introverts, but not in extraverts, during one (mental rotation) of the two cognitive activation tasks. It is not possible to relate the results of this study to ours because anatomical details were not provided. In the third study, the only one that used fMRI (Gray and Braver, 2002), the Behavioral Activation System (BAS) scores (Carver and White, 1994) correlated negatively with an increase in AC activity from rest to the 3-back memory task (0-, 1-, or 2-back conditions not used). BAS scores correlated positively with E (Elliot and Thrash, 2002); therefore, these results seemed to run counter to ours. However, there were substantial methodological differences between the two studies. Of particular importance is the fact that the Gray and Braver (2002) study took place over a much longer time period. Extraverts get bored easily and long for social interactions, and the effects of E vary as a function of time on task (Eysenck, 1982). It is possible that the relationship of E to activity in the AC will be positive for short-duration experiments (current results) (Haier et al., 1987) and negative for longer-duration experiments. The fact that Gray and Braver (2002) manipulated mood may also be relevant (Canli et al., 2001). As stated in Introduction, environmental variables are expected to influence observed relationships between personality and brain activity and should be taken into account. The direction of personality-AC activity association is likely to be situation specific rather than static.

We also observed localized effects relating to E, P, and N in the globally normalized resting signal. Each dimension showed strong negative relationships with activity in distinct brain regions: E, with Wernicke's and Broca's areas, thalamus, and cuneus; P, with the globus pallidus and putamen; and N, with the left orbitofrontal cortex. Although nonphysiological factors limit the inferences that can be drawn from resting fMRI data, the negative relationship between E and activity in Wernicke's and Broca's areas found here confirms previous findings (Johnson et al., 1999) with positron emission tomography (PET) and may reflect the tendency of introverts to engage in internal self-talk. Similarly, two previous studies using the ¹³³Xenon inhalation technique have also reported negative associations between global (Mathew et al., 1984) or temporal lobe resting activity (Stenberg et al., 1990, 1993) with E but not N. Stenberg et al. (1993) have shown, using PET, that the anxiety-proneness aspect of introversion determines the inverse relationship between E and temporal lobe activity, as can be expected (Gray, 1987). One PET study (Ebmeier et al., 1994), however, observed a positive correlation between E and cingulate activity measured in 51 elderly subjects, of whom 33 were scanned at rest, 12 with a metronome beating in the background, and 6 during a word repetition task; the authors interpreted these effects as reflecting extraverts' superiority on attentional tasks (in line with our observations of E influences in task-related activity). The relationship between P and activity in the basal ganglia that we found using our fMRI method is also broadly consistent with the negative relationship between P and dopamine D₂ receptor binding in the basal ganglia demonstrated in single-photon (Gray et al., 1994) and positron (Farde et al., 1997) emission tomography studies. The negative relationship between N and resting activity in the left orbitofrontal cortex has not been described previously but perhaps reflects the fact that activity in this region increases with positive emotion (Canli et al., 1998).

Deserving some comment is the observation that, although the average E score of our sample was representative of men in this age range, both N and P scores were slightly lower. This occurred most likely because we did not include regular cigarette smokers or subjects with past or current drug abuse (an item measuring EPQ-R P). Cigarette smoking is associated with high scores on both N and P (Eysenck and Eysenck, 1991), as well as with altered brain structure and functions (Dager and Friedman, 2000). Thus, our results can be taken as demonstrating a true effect of E, free from any confounding effects of high N or high P scores, under the conditions of the present experiment. Our results are, however, pertinent only to the sociability component of E within the context of Eysenck's theory (1967, 1981), because the items measuring impulsivity subcomponent of the E scale in the Eysenck Personality Inventory (Eysenck and Eysenck, 1964) were not included in later measures of E [i.e., the EPQ (Eysenck and Eysenck, 1975) or EPQ-R (Eysenck and Eysenck, 1991)]. Although Eysenck's theory does not make a theoretical distinction between the power of sociability and impulsivity components of E in mediating cortical arousal/arousability, there has been an impressive body of research showing that the impulsivity, but not sociability, component of E putatively influences arousal-mediated performance (Revelle et al., 1980; Smith et al., 1981). There is, however, an equally convincing body of research demonstrating that sociability, rather than impulsivity, underlies arousal-related influences (Gupta, 1990; Matthew et al., 1990a,b; Wilson, 1990; Corr et al., 1995; Corr and Kumari, 1997). We had chosen the EPQ-R for studying personality-brain relationships within the context of Eysenck's personality theory (1967, 1981) because it continued to associate cortical arousal/arousability with sociability (i.e., EPQ-R E) (Eysenck and Eysenck, 1985). EPQ-E scores also correspond well to extraversion scale scores on the NEO Personality Inventory-Revised (Costa and McCrae, 1992), which is another commonly used measure of extraversion (Zuckerman et al., 1993). Nonetheless, the contribution of sociability and impulsivity components of E in arousal-mediated effects warrants additional research to ascertain whether they affect the same or separate cortical arousal systems. It would be valuable to replicate our observations in relation to E, N, and P and extend them to examine the influence of the two subcomponents of E with a larger sample. A final methodological issue relates to the nature of our cognitive task. Instructions to subjects in our study required them to recall the location of the stimuli (numerals) presented previously (1-, 2-, or 3-back) on the screen. Although they seem to have followed the instruction to perform with spatial cues judged by the debriefing data as reported previously (Kumari et al., 2003), the effects we observed might still in some way be related to the strategy used by high and low E subjects, because the task did not constrain strategy and could be performed with either spatial or verbal cues.

In the present and previous (Gray et al., 1994; Farde et al., 1997; Canli et al., 2002; Gray and Braver, 2002) functional neuroimaging studies, the influence of personality has been strong despite small sample sizes and the absence of special subject selection. It is likely, therefore, that this influence is present in virtually all such studies. It is already mandatory to remove from fMRI analyses motion-induced variance, which at 30-50% is less than the variance attributable to personality observed here. Measurement of personality by simple questionnaire should similarly become routine in functional neuroimaging, with likely benefits in reducing apparent “error” variance within studies and increasing comparability between them.

To conclude, we report that personality predicts brain responses during cognitive tasks, supporting the model conceived by Eysenck (1967, 1981). Our findings are of interest to the neuroimaging community but perhaps are more important in confirming a biological basis of individual differences.