As summarized in Jung & Haier's (J&H's) review, there have been numerous studies aimed at the quantification of g. One observation that arises from their review is that it has been operationally defined in ways that are at least partly isomorphic with “supervisory attentional system” concepts as articulated by Shallice (Reference Shallice1988). Therefore, g appears to be critical for such executive operations as supervision of diverse streams of information, goal management, cognitive control, and strategy shifting, independent of the format, or processing domain (e.g., verbal or spatial).
Attempts to elucidate the nature of g have taken two forms that may be characterized as either “top-down,” or “bottom-up.” In the top-down approach, complex tasks are used that require many cognitive operations to be implemented simultaneously. Conceptually, high-g executive processes are characterized as “domain-independent” and are distinguished from low-g “domain-dependent” processing of spatial, object, or verbal information. Empirically, high-g and low-g processes have often been distinguished by the use of subtraction techniques in neuroimaging data analysis. In one study, for example, Duncan et al. (Reference Duncan, Seitz, Kolodny, Bor, Herzog, Ahmed, Newell and Emslie2000) subtracted high-g (domain-independent+domain-specific) from low-g (domain-specific) versions of the tasks that spanned different processing domains (spatial, verbal, and perceptuo-motor). Two results from these studies are important for our present argument. First, both frontal and parietal activity was observed in the high-g/low-g subtractions. It is important to note, however, that parietal activity was observed only in the spatial tasks. Second, only prefrontal cortical (PFC) regions were consistently isolated when the subtraction analysis was applied. These results suggest a more intimate connection between frontal brain regions and g-related processes than between parietal regions. The apparent parietal fit noted by J&H may have been a consequence of the spatial nature of many high-g tasks.
In the bottom-up approach, simpler cognitive constructs have been utilized to isolate particular domain-independent g-related processes. In one of the earliest such approaches, subjects performed domain-specific processing tasks (either spatial rotation or semantic judgment tasks) in separate scans (D'Esposito et al. Reference D'Esposito, Detre, Alsop, Shin, Atlas and Grossman1995). Domain-independent processing was evoked by the requirement to perform both tasks together. This “dual-task” requirement elicited activity in dorsal PFC regions, implicating a supervisory role (cf. Shallice Reference Shallice1988) for these regions, a function also closely related to g. Prabhakaran et al. (Reference Prabhakaran, Narayanan, Zhao and Gabrieli2000) utilized a similar strategy to show that integration of information in working memory can also be mapped to specific PFC regions, whereas posterior brain regions support domain-specific processing. Subjects performed tasks in which they maintained integrated or discrete verbal and spatial items in working memory, while load and duration of processing were controlled. It may be that PFC regions mediate the integrated representation that underlies the cognitive flexibility needed for domain-independent processing. Posterior brain regions may mediate discrete representations needed for domain-specific processing.
Intelligence research is rooted in attempts to identify factors that determine reliable performance differences between individuals. Such information, not readily available from the group-level analyses of data, yields a more nuanced picture of the neural basis of g. Our approach to exploring this question has been to account for individual performance variability in neuroimaging data. This approach has yielded important clues about the neural underpinning of intelligence.
Consistent with the P-FIT model, our initial study of the Raven's Progressive Matrices (RPM; Prabhakaran et al. Reference Prabhakaran, Smith, Desmond, Glover and Gabrieli1997) uncovered a fronto-parietal network for analytic reasoning. Taking into account the relative difficulty of individual problems, as well as individual subjects' reaction time (RT) and accuracy, highlighted the importance of PFC regions (Christoff et al. Reference Christoff, Prabhakaran, Dorfman, Zhao, Kroger, Holyoak and Gabrieli2001). Importantly, a composite analysis of our reasoning studies (Prabhakaran et al. Reference Prabhakaran, Rypma and Gabrieli2001) suggested consistent frontal-lobe involvement across tasks but more variable involvement of posterior brain regions, based on the specific processing domain of the task. These results are consistent with the P-FIT model insofar as prefrontal function is concerned, but they are less consistent with the model regarding parietal function.
Do frontal and parietal regions together perform the functions that we associate with g as P-FIT suggests? Or does it depend on the nature of the task and the characteristics of the individual? Data we have collected recently suggest that g-related functions may be mediated primarily by the PFC. Posterior regions including the parietal cortex play the major role when executive or supervisory requirements are minimal.
One study we conducted suggests that the prefrontal cortex exerts supervisory control over posterior parietal regions during visual search (Rypma et al. Reference Rypma, Berger, Prabhakaran, Bly, Kimberg, Biswal and D'Esposito2006). Subjects were scanned while they matched a “digit-symbol key” and a single digit-symbol probe that appeared below the key. Subjects performed this task with uniformly high accuracy. Reaction times were short but more varied between individuals than accuracy was. We used Granger analysis to test the hypothesis that, among slower individuals, PFC systems guide posterior systems (Fig. 1).
Granger causality permits characterization of the strength and direction of influence between discrete brain regions. Influences were calculated separately for faster and slower subjects (grouped by median split). Influences were considered significant for ps<.05 and trends when .05<p<.10.
Consistent with the notion of PFC supervisory control, the results indicated there were more directed frontal-to-parietal influences in slower performers than in faster ones. Regression analyses indicated that, across individuals, PFC-parietal connectivity was associated with longer RT. This result supports the hypothesis that PFC exerts control over parietal activity in slower performers during visual search. For faster performers, posterior systems operate more automatically, independent of PFC control.
Another study (Prabhakaran et al. Reference Prabhakaran, Raman, Grunwald, Mahadevia, Hussain, Lu, Van Zijl and Hillis2007) also shows increased frontal activity among poor performers (Fig. 2a). In this study, normal subjects and chronic stroke patients were scanned while they either subvocally generated words or rested passively. Normal subjects recruited a left-lateralized fronto-temporal network compared to rest. Chronic left Middle Cerebeal Artery (MCA) stenotic patients showed additional right fronto-temporal activity. In normal subjects, parietal regions actually showed more activity during rest than during word generation, suggesting a limited role in mediation of g functions and possibly reflecting a “default mode network” role for this region (Fig. 2b; cf. Raichle et al. Reference Raichle, MacLeod, Snyder, Powers, Gusnard and Shulman2001; Vincent et al. Reference Vincent, Snyder, Fox, Shannon, Andrews, Raichle and Buckner2006).
(a) Blood oxygen level dependent (BOLD) effect for task versus rest at p=0.001 (uncorrected). Top panel: Composite results of fMRI of normal subjects showing a left-lateralized fronto-temporal network of brain activity with predominantly frontal involvement.Bottom panel: fMRI result of one of the chronic left MCA stenotic patients MZ showing less lateralized fronto-temporal activity with increased right fronto-temporal/predominantly-frontal region involvement. (b) BOLD effect for rest versus task at p=0.001 (uncorrected). Composite results of fMRI of normal subjects. Bilateral parietal regions were noted to show more activity in the rest than in the task state.
We think that such results suggest modification of P-FIT. First, the observations made in the target article are consistent with the notion that g is mediated through the PFC. Second, we suggest that the parietal lobe is not a principal player in posterior mediation of g functions, but rather one stand-in among many that are called upon to mediate domain-specific task demands. Thus, the term “P-FIT” may be better phrased as a “posterior-frontal integration theory.” We suggest that performance differences between high-g and low-g individuals may result from the extent to which domain-specific cognitive processes can be implemented relatively “automatically” (cf. Schneider & Shiffrin Reference Schneider and Shiffrin1977) by posterior brain regions, independent of control exerted upon them by the PFC.
