Introduction
The Iowa Gambling Task (IGT) was developed as a psychometric probe for deficits in real-life decision-making manifested by neurologic patients with lesion of the ventromedial prefrontal cortex (VMPFC; Bechara, Damasio, Damasio, & Anderson, Reference Bechara, Damasio, Damasio and Anderson1994). The standardized computer-administered version of the IGT has undergone extensive demographically-based norming and is available as a clinical and research tool (Bechara, Reference Bechara2007). The task involves choosing cards from four decks (A–D) in which all decks yield monetary awards and penalties. The contingency schedule for awards and penalties is such that selections from decks C and D are advantageous after the first 20 trials, while those from decks A and B are disadvantageous after the first 20 trials. There are also parallel differences between decks A and B as well as C and D (Bechara, Tranel, & Damasio, Reference Bechara, Tranel and Damasio2000). For Deck A only the frequency of punishment increases every 10 trials while for Deck B only the magnitude of punishment increases every 10 trials. For Deck C only the frequency of reward increases every 10 trials while for Deck D only the magnitude of reward increases every 10 trials. Test-takers generally become aware of the risk parameters after approximately 20 (Maia & McClelland, Reference Maia and McClelland2004) to 40 (Brand, Recknor, Grabenhorst, & Bechara, Reference Brand, Recknor, Grabenhorst and Bechara2007) trials. The IGT is a measure of heterogeneous decision-making processes first under ambiguity and later under known risk (Brand et al., Reference Brand, Recknor, Grabenhorst and Bechara2007), and early trial selections reflect the construct of rational exploratory behavior (Dunn, Dalgleish, & Lawrence, Reference Dunn, Dalgleish and Lawrence2006).
Psychometric Concerns
Despite extensive norming and standardization (Bechara, Reference Bechara2007), there is no published data on the reliability of the IGT (Buelow & Suhr, Reference Buelow and Suhr2009). Inter-block consistency was shown to be low even when the analysis was restricted to the last three blocks, limiting convergent and criterion validity to modest values (Gansler, Jerram, Vannorsdall, & Schretlen, Reference Gansler, Jerram, Vannorsdall and Schretlen2011a). Metrics removing decks B and C improve construct validity (Gansler et al., Reference Gansler, Jerram, Vannorsdall and Schretlen2011a; Gansler, Jerram, Vannorsdall, & Schretlen, Reference Gansler, Jerram, Vanorsdall and Schretlen2011b). Given that early trials measure decision-making under ambiguity and later trials measure decision-making under known risk, late trial metrics are more representative of executive function (Brand et al., Reference Brand, Recknor, Grabenhorst and Bechara2007; Gansler et al., Reference Gansler, Jerram, Vanorsdall and Schretlen2011b) while metrics based on all trials or early trails represent novel problem-solving and attentional processes (Gansler et al., Reference Gansler, Jerram, Vanorsdall and Schretlen2011b). Despite concerns about the reliability and validity of the IGT, compelling reasons to examine its neuroanatomic correlates include its widespread use, the absence of large-scale brain-wise anatomic studies, the need to inform future efforts to modify stimulus presentation and administration format in the direction of greater reliability and validity, the opportunity to contribute to the theoretical work surrounding the somatic marker hypothesis, and, the complement that structural imaging of the healthy brain provides to the lesion and fMRI literature on the IGT.
Theory and Evidence for the Paradigm
The impetus for developing the IGT and the somatic marker hypothesis (SMH; Damasio, Reference Damasio1994) came from patients with VMPFC lesions who manifested psychosocial dysfunction secondary to poor decision-making but performed normally on neuropsychological testing (e.g., Eslinger & Damasio, Reference Eslinger and Damasio1985). These individuals had access to the rational processes underlying a typical cost-benefit analysis but lacked the influence of emotional input contributing to sound decision-making (Bechara et al., Reference Bechara, Damasio, Damasio and Anderson1994). Further investigation incorporating variations on the paradigm with VMPFC lesioned patients indicated they were insensitive to future negative or positive consequences, but not hyper-sensitive to reward (Bechara et al., Reference Bechara, Tranel and Damasio2000). The IGT is strongly associated with the somatic marker hypothesis (Damasio, Reference Damasio1994), which concerns the representation of somato-sensory states in relation to decision-making processes. The SMH posits that decision-making is influenced by somatic markers, crude biasing signals arising from the neural machinery in somato-sensory cortex, such as the insula and parietal sensory cortex (Bechara & Damasio, Reference Bechara and Damasio2005; Dunn et al., Reference Dunn, Dalgleish and Lawrence2006). The VMPFC is considered a critical region for decision-making as it integrates representation of bio-regulatory states and response options in an “as-if” loop (Bechara & Damasio, Reference Bechara and Damasio2005; Dunn et al., Reference Dunn, Dalgleish and Lawrence2006). Physiological (autonomic) under-arousal during IGT performance on the part of prefrontal lesioned patients relative to healthy controls lends credence to the SMH (Bechara, Tranel, Damasio, & Damasio, Reference Bechara, Tranel, Damasio and Damasio1996), as do several lesion studies. Bechara Damasio, Tranel, and Anderson (Reference Bechara, Damasio, Tranel and Anderson1998) have reported IGT impairment associated with orbital but not dorsal lesions. Subsequent work has shown the volume of the right middle orbital gyrus to be positively associated with decision-making performance in healthy controls but not individuals with schizophrenia (Nakumara et al., 2008).
Individuals with lesions of the anterior orbitofrontal cortex (OFC), an area that overlaps the VMPFC, show selective impairment of IGT decision-making. Those with posterior OFC lesions have working memory deficits as well. Thus, the former area may be more specialized for the influence of emotional processes upon decision-making (Bechara, Reference Bechara2007). Patients with lesions of the right parietal lobe and adjacent insular and somatosensory cortices, but without homologous left-sided lesions, also display IGT decision-making difficulty (Tranel, Bechara, & Damasio, Reference Tranel, Bechara and Damasio2000). Lesions of the amygdala, an area with reciprocal connections to the VMPFC, produce the same pattern of deficit in advantageous IGT decision-making, as do VMPFC lesions (Bechara, Damasio, Damasio, & Lee, Reference Bechara, Damasio, Damasio and Lee1999), although psychophysiological data point to differences in emotional reactivity.
Critical Consideration of the IGT
The SMH and the IGT task as a operational measure of it, have come under criticism due to lack of empirical support. It has been reported that dorsomedial and dorsolateral, but not ventromedial, prefrontal lesions impair IGT performance (Manes et al., Reference Manes, Sahakian, Clark, Rogers, Antoun, Aitken and Robbins2002). While fMRI investigations support the involvement of neural networks proposed in the SMH in decision-making and IGT performance, particularly the OFC, they also indicate the involvement of broader prefrontal and other networks not explicitly part of the original SMH (Bolla, Eldreth, Matochik, & Cadet, Reference Bolla, Eldreth, Matochik and Cadet2005; Christakou, Brammer, Giampietro, & Rubia, Reference Christakou, Brammer, Giampietro and Rubia2009; Tucker et al., Reference Tucker, Potenza, Beauvais, Browndyke, Gottschalk and Kosten2004; Windmann et al., Reference Windmann, Kirsch, Mier, Stark, Walter, Gunturkun and Vaitl2006). The IGT has also been shown to be sensitive to diseases that do not necessarily involve the neural substrates of the SMH, such as fibromyalgia (Verdejo-Garcia, Lopez-Torrecillas, Calandre, Delgado-Rodriguez, & Bechara, Reference Verdejo-Garcia, Lopez-Torrecillas, Calandre, Delgado-Rodriguez and Bechara2009) and eating disorders (Brogan, Hevey, & Pignatti, Reference Brogan, Hevey and Pignatti2010). These findings suggest the IGT is sensitive but not specific to the SMH neural substrate, as Alvarez and Emory (Reference Alvarez and Emory2006) found executive functioning tasks are sensitive but not specific to frontal lesions. The IGT paradigm has also been criticized as an inadequate measure of how people behave based on future consequences as it may not operationalize “future blindness” (Colombetti, Reference Colombetti2008). The composite IGT score of total selections from advantageous decks (C and D) minus disadvantageous decks (A and B) (IGT Professional Manual, Bechara, Reference Bechara2007) could represent such a heterogeneous amalgam of diverse processes so as to create more confusion than clarity (Buelow & Suhr, Reference Buelow and Suhr2009). For example, an IGT fMRI study of nineteen healthy males revealed broad neural networks involved in IGT performance and decision-making with a proposed organization for representation of decision-making and outcome assessment in the context of expectancy (Christakou et al., Reference Christakou, Brammer, Giampietro and Rubia2009). These two processes combine in several ways to mediate shifting away from disadvantageous choices, potentiating adaptive decisions by identifying negative outcomes and potentiating adaptive decisions by detecting expected positive outcomes. Each of these combinations was associated with separate neural correlates: shifting was associated with ventromedial and dorsolateral prefrontal activation; potentiating adaptive decisions through negative outcome detection was associated with dorsomedial prefrontal and anterior cingulate activation; potentiating adaptive decisions through confirmatory, positive outcome detection was associated with widespread activation in frontal and temporal cortices, as well as other regions including the insula, caudate and thalamus (Christakou et al., Reference Christakou, Brammer, Giampietro and Rubia2009). Separating the valence of punishment and reward processes from the processing of differing contingencies delivered indicated bilateral caudolateral OFC (Brodmann area 47/inferior frontal gyrus) involvement in the processing of inconsistent outcome delivery (Windmann et al., Reference Windmann, Kirsch, Mier, Stark, Walter, Gunturkun and Vaitl2006).
It has been argued that the lack of association of IGT and general intellect supports the SMH differentiation of rational processes and intelligence (Toplak, Sorge, Benoit, West, & Stanovich, Reference Toplak, Sorge, Benoit, West and Stanovich2010). However, some studies have shown that IGT performance is moderately associated with fluid intelligence (Suhr & Hammers, Reference Suhr and Hammers2010; Gansler et al., Reference Gansler, Jerram, Vanorsdall and Schretlen2011b). Nevertheless, the IGT has been shown to be impaired in several populations in which decision-making difficulties would be expected, such as the chemically dependent (Grant, Contoreggi, & London, Reference Grant, Contoreggi and London2000; Rotheram-Fuller, Shoptaw, Berman, & London, Reference Rotheram-Fuller, Shoptaw, Berman and London2004) and the thought disordered (Nakamura et al., Reference Nakamura, Nestor, Levit, Cohen, Kawashima, Shenton and McCarley2008). Function imaging approaches with chemically dependent individuals find evidence for involvement of orbital as well as dorsolateral prefrontal cortex in IGT performance (Bolla et al., Reference Bolla, Eldreth, Matochik and Cadet2005; Tucker et al., Reference Tucker, Potenza, Beauvais, Browndyke, Gottschalk and Kosten2004).
Hypotheses
The following hypotheses compare the neural correlates of conventional and unconventional IGT metrics. The conventional IGT score is “CD-AB 1-100,” and scores excluding decks B and C are “D-A 1-100,” with the trial numbers adjusted according to which blocks are incorporated into the metric (e.g., CD-AB 41-100). The examination of neural correlates is a continuation of a construct validity approach: (1) The alternative IGT metric (i.e., D-A 1-100) will display more robust neural correlates than the conventional IGT metric CD-AB 1-100 in regions consistent with the SMH. (2) Total selections from Decks A and D will display more robust neural correlates than total selections from Decks B and C. (3) Total selections from Decks B and D, involving shifting magnitude of punishment and reward will display more robust frontal neural correlates than selection from Decks A and C, involving changing frequency of punishment and reward. (4) Alternative IGT metrics that distinguish early from late trials (e.g., D-A 1-40 and D-A 41-100) will display differing neural correlates in areas relevant to developing response biasing and shifting: ventro-medial and dorsolateral PFC.
Method
Participants and Procedures
Participants were drawn from an initial community sample of adults recruited from the Baltimore, Maryland area to participate in the Aging, Brain Imaging, and Cognition (ABC) study (Testa, Winicki, Pearlson, Gordon, & Schretlen, Reference Testa, Winicki, Pearlson, Gordon and Schretlen2009). Participants were recruited via random digit dialing, written invitation to Medicare beneficiaries aged 65 and older, and telephone calls to listings selected in pseudo-random manner from residential directories. The ABC study was conducted in two phases: Participants (n = 215) who entered the study during phase 1 (1995–1998) were recruited from Baltimore. The IGT was not used at that time. However, 110 phase 1 participants returned during phase 2 (1999–2005), and they completed the IGT, which was added to the phase 2 study protocol. Another 86 participants were recruited during phase 2. Thus, 196 participants were administered the IGT. Each participant underwent a physical and neurological examination, psychiatric interview, laboratory blood tests, a 1.5 T structural brain MRI scan, and cognitive testing over 1–2 days which included a range of neuropsychological tests across domains administered according to manual instructions and standardized objective personality measures also administered according to manual instructions. Participants were classified as unhealthy if they had a neurological condition known to affect cognitive functioning (e.g., Parkinson's disease, multiple sclerosis), severe/life-threatening medical problems (e.g., congestive heart failure with poorly controlled hypertension and diabetes), or significant psychiatric illness (e.g., schizophrenia, bipolar disorder, major depression, substance dependence). To reduce error, unhealthy participants were not included in this report. The remaining 162 participants were classified as reasonably healthy. As part of the intake process, each participant's mental health status was classified [none (51%), minor (49%), moderate (0%), or severe (0%)] as was physical health status [no medical problems (29%), minor medical problems (58%), multiple or poorly controlled problems (12%), or life threatening (0%)]. IGT data for 23 of these participants were excluded or lost and we excluded one individual who did not complete the remainder of cognitive testing. Nine participants were excluded either due to invalid performance on the IGT (responding exclusively to decks C and D), or due to statistical outlier status. Finally, visual inspection revealed six participants with problematic structural scans arising either from motion artifact or central nervous lesion or anomaly, leaving data available for 124 healthy participants.
IGT Procedure
The IGT was administered using a pre-publication copy of the standardized computer-based version obtained from the test's author. This paradigm is constructed to replicate real-life decision-making with each of the 100 card selections leading to financial gains or losses. The examinee starts with a (make believe) loan of $2000 and a bar indicator on the screen provides an ongoing dollar tally. Bowman and Turnbull (Reference Bowman and Turnbull2003) demonstrated undergraduates perform similarly under conditions of real versus the standard facsimile reinforcement condition. They are faced with four identical card decks labeled A, B, C, and D, and they are told only that some decks are better than others. Although each deck of cards yields rewards and punishments, selections from decks A and B are ultimately disadvantageous (leading to a net loss of $250 every 10 trials) and selections from decks C and D are ultimately advantageous (leading to a net gain of $250 every 10 trials). Examinees “win” $100 with a deck A or B selection and $50 with a deck C or D selection, but the losses from deck A or B range from $35 to $1250 while losses from deck C or D range from only $25 to $250. Decks A and C involve a higher 50% frequency of penalized selections.
All participants gave written informed consent, and the study was approved by the Johns Hopkins Medicine and Hartford Hospital Institutional Review Boards.
Image Acquisition and Processing Protocol
All images were acquired on the same 1.5 Tesla GE Signa scanner using identical acquisition parameters: 124 contiguous 1.5 mm SPGR images in coronal plane.
All image preprocessing was conducted using MATLAB 7.9 (The Mathworks Inc., 2009) and SPM8 (Wellcome Trust Center for Neuroimaging, 2009). MR images were first manually reoriented and realigned along the anterior commissure-posterior commissure axis. The “unified segmentation” method (Ashburner & Friston, Reference Ashburner and Friston2005) and the standard SPM gray matter template were used to segment tissue into gray matter, white matter, and cerebral spinal fluid. Gray matter maps were saved for analysis. The DARTEL algorithm, demonstrated to have advantages over previous methods (Bergouignan et al., Reference Bergouignan, Chupin, Czechowska, Kinkingnéhun, Lemogne, Le Bastard and Fossati2009), was then used to create study-specific gray matter templates to which the anatomical images were spatially normalized. Modulation was applied to preserve regional and global volume of gray matter. Since modulation produces partial spatial smoothing and the DARTEL algorithm improves spatial normalization, images were smoothed with a small 6 mm full-width at half-maximum (FWHM) Gaussian filter. Images were also normalized to MNI space for purpose of generalization beyond the sample template. Total intracranial volume (TICV) was calculated by summing the cubic centimeter values for gray matter, white matter, and cerebral spinal fluid obtained during segmentation.
Participants were on average around 55 years of age (M = 54.9; SD = 16.4), slightly more than half were female (52.4%), on average they had several years of education beyond the high school level (M = 14.0; SD = 3.0), and were 88% right-handed. On average, the group performed at typical levels on a screening for overt dementia (MMSE mean raw score = 27.4; SD 1.87), on a single word reading task (NART-IQ mean standard score = 104.66; SD 10.08), and in terms of general intellect (WAIS-R seven sub-test FSIQ mean standard score = 105.75; SD 15.66) for a non-clinical sample.
Data Analysis Plan and Pre-analysis
Voxel-based morphometry (VBM) was performed on the whole brain gray matter maps. Positive and negative regressions were performed using conventional and unconventional IGT metrics as the predictors. TICV served as a covariate in each of the regressions, to control for age (Tisserand et al., Reference Tisserand, Pruessner, Sanz Arigita, van Boxtel, Evans, Jolles and Uylings2002) and gender differences (Cowell et al., Reference Cowell, Sluming, Wilkinson, Cezayirli, Romanowski, Webb and Roberts2007). Voxel level results were thresholded at a p-level of .005, uncorrected for multiple comparisons. Cluster level analysis was performed on voxels that survived that thresholding. A cluster was defined with a minimum of 20 voxels. Cluster level results were corrected for multiple comparisons with family wise error (FWE) alpha set at .05. IGT metrics and TICV were normally distributed with the exception of the distribution of scores for D-A 1-40 which was peaked (see descriptive data in Table 1).
Table 1 Descriptive data for TICV and IGT metrics for study participants (n = 124)
Results
Behavioral Results
Analysis of variance for repeated measures indicated that performance changed significantly across the five trials in the expected direction of increased advantageous decisions over trials (F 4,120 = 5.41; p < .001). Selection of advantageous decks increased in a linear manner across the first four blocks and then on the fifth block reverted to levels similar to the third block (mean of CD-AB/20 trial block: block 1 = −1.27, block 2 = 1.90, block 3 = 2.68, block 4 = 3.77, block 5 = 2.69). Thus, within-subjects contrasts indicated the repeated measures fit both a linear (F 1,123 = 12.11; p < .01) and quadratic equation shape (F 1, 123 = 7.92; p < .01). See Table 1 for average number of total selections for each deck.
Age was not associated with IGT performance (CD-AB 1-100: Pearson r = −.06, ns). Fluid intelligence as measured by the matrix reasoning subtest of the Wechsler Abbreviated Scale of Intelligence was positively associated with IGT performance (CD-AB 1-100: Pearson r = .26; p < .01). Gender was not significantly associated with IGT performance (CD-AB 1-100 male mean = 13.53; female mean = 6.37; F 1,122 = 2.25; p > .05), although on average males scored higher than females.
Hypothesis 1. Comparing Metrics based on Trials 1-100
CD-AB 1-100
No statistically significant foci was observed when CD-AB 1-100 was positively or negatively regressed on whole brain gray matter volume (WBGMV).
D-A 1-100
Seven statistically significant foci of interest were found when D-A 1-100 was positively regressed on WBGMV (see Table 2). Two foci were sub-cortical and located in left and right cerebellum. Two foci were located in the left and right parahippocampal gyri, with the left hemisphere foci extending into the uncus and constituting, by far, the largest foci involving D-A 1-100 (9145 voxels). The remaining three foci were prefrontal (two lateral and one medial/bilateral). The lateral foci included the left middle and right superior frontal gyri, and another focus included the left inferior and middle frontal gyri. The bilateral foci was located dorsomedially. No foci of significant gray matter tissue was observed when D-A 1-100 was negatively regressed on WBGMV.
Table 2 Positive associations between IGT performance and regional gray matter volume
Hypotheses Two and Three. Total Selections From Individual Decks and Contrast of Decks Involving Shifting Frequency versus Magnitude Patterns
Deck A total selections
No statistically significant foci of interest were found when A 1-100 was positively regressed on WBGMV. Four statistically significant foci of interest were found when A 1-100 was negatively regressed on WBGMV (see Table 3). These included the left and right cerebellum, the right cuneus, and a large cluster in the thalamus bilaterally and extending into the right brainstem red nucleus.
Table 3 Negative associations between IGT performance and regional gray matter volume
Deck B total selections
No statistically significant foci of interest were found when B 1-100 was positively or negatively regressed on WBGMV.
Deck C total selections
No statistically significant foci of interest were found when C 1-100 was positively regressed on WBGMV (see Table 3). Five statistically significant foci of interest were found when C 1-100 was negatively regressed on WBGMV. A very large cluster was found in the right inferior temporal gyrus. A reasonably large cluster emerged in the left middle temporal gyrus. A right inferior frontal gyrus cluster also emerged. Substantive neural correlates were also identified along the mid-line bilaterally involving the left posterior cingulate and cingulate gyrus and the right cingulate gyrus.
Deck D total selections
Eight statistically significant foci of interest were found when the total number of selections for deck D was positively regressed on WBGMV (see Table 2). A significant foci emerged in the left insula and extended to the thalamus. Neural correlates similar in extent and spatial distribution to those for D-A metrics, emerged in the left parahippocampal gyrus, the right uncus, left and right dorsolateral prefrontal cortex, and the right cerebellum. No statistically significant focus emerged when total number of selections for deck D was negatively regressed on WBGMV.
Hypothesis 4. Comparison of Early and Late Trial Metrics
Two statistically significant foci of interest were found when D-A 1-40 was positively regressed on whole brain gray matter volume (see Table 2). These included a focus in the left ventrolateral prefrontal cortex, and a slightly larger focus extending from the right parahippocampal gyrus into the right middle temporal gyrus.
Six statistically significant foci of interest were found when D-A 41-100 was positively regressed on whole brain gray matter volume (see Table 2). A large sub-cortical foci was found in the right cerebellum. Another large foci extending sub-cortically from the left cerebellum into the left middle and left superior temporal gyrus was observed. The right middle and superior temporal gyri were the site of the remaining non-frontal foci. Dorsolateral prefrontal foci including the middle and superior frontal gyri on the left, and inferior, middle and superior frontal gyri on the right were observed. The remaining frontal focus had a left dorsomedial location.
No foci of statistical significance were found when either D-A 1-40 or D-A 41-100 scores were negatively regressed on WBGMV.
Discussion
To our knowledge, this is the first large-scale and healthy control based report of the gray matter neural correlates of decision-making as measured by the IGT. Results indicate the psychometric approach taken to the 100-trial IGT performances deserves critical consideration when asking clinical or research questions about brain-behavior relationships. Broad ranging neural correlates of the IGT emerged, invoking a hierarchical information processing perspective (Luria, Reference Luria1980) to understanding the processes underlying deck selection and the metrics used to evaluate decision-making capacity.
Consistent with evidence of greater construct validity for alternative D-A versus conventional CD-AB metrics (Gansler et al., Reference Gansler, Jerram, Vannorsdall and Schretlen2011a, Reference Gansler, Jerram, Vanorsdall and Schretlen2011b), the D-A metric displayed robust neural correlates along which decision-making processes are known to be represented, if not necessarily those most closely associated with the SMH. The conventional metric displayed none. D-A metrics are more robustly associated with fluid problem-solving abilities, general neuropsychological abilities, and attention and executive function specifically (Gansler et al., Reference Gansler, Jerram, Vanorsdall and Schretlen2011b). This result confirmed hypothesis one, and demonstrated that measures with better psychometric properties are more likely to display neural correlates. Deck A selections appear to be associated with neural circuitry for directing attention, integrating attention and motor actions, and confirmation of positive expectancy as well, across the cerebellum bilaterally, in the right cuneus, and the thalamus bilaterally. This was the least selected deck, with selections relating primarily to early and lower level attentional aspects of information processing. Deck C selections relate to neural circuitry confirming positive outcome expectancies (inferior and middle temporal gyri), processing negative outcomes, directing attention to goal-directed actions, and error monitoring (cingulate gyri), and verbal processing of stimuli under conditions of inconsistent outcome delivery (inferior frontal gyri). Deck C neural correlates suggest middle stages of information processing in terms of manipulation of input, and to a lesser extent final output phases of behavior. Deck D selections were related to neural circuitry sensing negative expectancies (insula), confirming and encoding positive expectancies (thalamus, parahippocampal gyrus, uncus) and shift of behavior and decision-making (dorsolateral prefrontal cortex bilaterally). The Deck D neural correlates relate to late information processing, in terms of decisional or output phases of behavior, and alone were associated with cortical networks for executive and strategic functioning (Chen, Rosa-Neto, Germann, & Evans, 2008). The contrasting neural correlates of Deck C (involves increasing frequency of reward) and Deck D selections (involves increasing magnitude of reward) appears to reflect the core demand of the IGT to maximize dollar value thru strategic decision-making. Deck B selections may not have yielded neural correlates not because they are irrelevant to the task, but because they do not represent a clear stage of information processing.
Hypothesis two, that decks A and D would display more robust neural correlates than decks B and C was partly confirmed, in that no neural correlates for Deck B were found, helping to explain the advantage of the D-A metric. Neural correlates of Deck D were all positive while those for Deck A were all negative, complimenting one another well for the subtraction metric. Deck C neural correlates were robust, including regions involved in action control (cingulate gyrus) and processing of positive expectancies (left middle temporal gyrus). These correlates could represent fairly disparate information processing constructs which may not have worked together well. By contrast the lateral frontal and medial temporal correlates of Deck D selections, thought to be involved in shift of decision-making set and emotional aspects of memory, respectively, may have worked together better with Deck A correlates in the thalamus and cuneus thought to be involved in earlier attentional stages of information processing. Thus, the successful display of neural correlates by Deck D 1-100 and D-A 1-100 is a representation of fronto-subcortical networks across information processing stages involved in attention and strategic decision-making. This interpretation fits with recent concerns the CD-AB metric represents an amalgamation of disparate constructs (Buelow & Suhr, Reference Buelow and Suhr2009).
There was mixed support for hypothesis three, that decks B and D involving shifting reward/punishment magnitude would display more robust frontal correlates than decks A and C, involving shifting reward/punishment frequency. The neural correlates for Decks D and A were consistent with the hypothesis, as were the majority of the correlates for Deck C with the exception of the right inferior frontal gyrus. The absence of correlates for Deck B did not fit the hypothesis. Thus the D-A metric might yield more robust neural correlates because it removes metrics combining frequency and magnitude evaluation (C+D or A+B).
The fourth hypothesis proposing that D-A 1-40 would correlate with brain areas representing positive and negative outcome processing, while D-A 41-100 would correlate with areas representing response biasing and shifting, received some support. Neural correlates of D-A 1-40 were observed in right lateral temporal cortex, which may represent the processing of positive outcomes (Christakou et al., Reference Christakou, Brammer, Giampietro and Rubia2009). This same focus also included the right parahippocampal gyrus, and although not hypothesized, could relate to reports that participants encode appropriate deck-based selections within the first 20 to 40 trials (Brand et al., Reference Brand, Recknor, Grabenhorst and Bechara2007; Maia & McClelland, Reference Maia and McClelland2004). Contrary to our hypothesis, medial regions thought to represent the processing of negative outcomes in fMRI studies did not emerge, although the cingulate gyrus correlates are close to those regions. This could be because fMRI paradigms create very specific item-based contrasts, whereas in this study broad measures of IGT performance were examined. Contrary to expectation, a fairly large cluster in the left ventrolateral prefrontal cortex was positively associated with D-A 1-40, a region representing processing of outcomes under changing circumstances (Christakou et al., Reference Christakou, Brammer, Giampietro and Rubia2009; Windmann et al., Reference Windmann, Kirsch, Mier, Stark, Walter, Gunturkun and Vaitl2006), that is also part of cortical modular architecture for auditory and language processing (Chen et al., Reference Chen, He, Rosa-Neto, Germann and Evans2008). Although later IGT blocks are more strongly associated with executive function (Brand et al., Reference Brand, Recknor, Grabenhorst and Bechara2007; Gansler et al., Reference Gansler, Jerram, Vanorsdall and Schretlen2011b) and more sensitive to frontal lobe pathology (Torralva, Roca, Gleichgerrcht, Bekinschtein, & Manes, Reference Torralva, Roca, Gleichgerrcht, Bekinschtein and Manes2009), the early trials demonstrate some of the same neural correlates expected to underlie later performance.
Neural correlates of D-A 41-100 were generally similar to those of D-A 1-100 (see Figure 1). It was notable that more widespread neural correlates in the right DLPFC emerged for trials 41-100 compared to 1-100. Taken together with the left VLPFC correlate of D-A 1-40, those results could indicate a greater role for verbal mediation early in the task and for response inhibition later in the task. The bilateral nature of the prefrontal correlates indicates the mix of positive and negative rewards on the IGT precludes the predominance of a positive or negative valence invoking hemispheric dominance (Windmann et al., Reference Windmann, Kirsch, Mier, Stark, Walter, Gunturkun and Vaitl2006). The broader prefrontal nature of the neural correlates is consistent with recent calls to revise the SMH and expand the neural substrate to regions of the PFC beyond the OFC (Christakou et al., Reference Christakou, Brammer, Giampietro and Rubia2009; Dunn et al., Reference Dunn, Dalgleish and Lawrence2006).
Neural correlates of D-A 1-100 did not emerge in the OFC, insular cortex, or parietal cortex as posited by the SMH, but did so for other regions implicated in the decision-making literature. One recent volumetry study found an association between right OFC volume and CD-AB 1-100 in healthy adults (Nakumara et al., 2008). IGT activation studies also demonstrate the role of anterior and midline structures (Christakou et al., Reference Christakou, Brammer, Giampietro and Rubia2009; Tucker et al., Reference Tucker, Potenza, Beauvais, Browndyke, Gottschalk and Kosten2004) consistent with the neural representation of emotion (Hornak et al., Reference Hornak, Bramham, Rolls, Morris, O'Doherty, Bullock and Polkey2003), if not the strict neural substrate of the SMH. Nevertheless, using structural equation modeling (Gansler et al., Reference Gansler, Jerram, Vanorsdall and Schretlen2011b), IGT metrics were shown to be associated with fluid intelligence, general neuropsychological ability, and a composite attention variable, while CD-AB 41-100 and D-A 41-100 related to an executive function composite. IGT metrics, especially D-A metrics, appear to have a considerable cognitive loading based on the convergence of psychometric and neural localization evidence, and may not represent an exclusive realm of decision-making and emotion apart from cognition (Toplak et al., Reference Toplak, Sorge, Benoit, West and Stanovich2010). When total selections for decks A or D were regressed on whole brain gray matter, a neural substrate emerged in the left insula (Deck D) and thalamus (Decks A and D). The insular correlate is consistent with the emotion neural substrate in the SMH (Dunn et al., Reference Dunn, Dalgleish and Lawrence2006). Furthermore, it was selections from Deck C, more than other deck selections, that were associated with mid-line structures involved in emotion regulation. The paradigm may be seen as drawing on neural substrates for both cognitive and emotional aspects of decision-making, with less amalgamated metrics teasing out the constituent elements.
The neural correlates observed here do not overlap substantially with the IGT lesion studies (Bechara, Reference Bechara2007; IGT Professional manual), with the exception of the positive association between the number of total deck D selections and left insula gray matter volume. However, these findings are more convergent with a report that dorsolateral and dorsomedial lesions, but not orbitofrontal lesions produce IGT impairment (Manes et al., Reference Manes, Sahakian, Clark, Rogers, Antoun, Aitken and Robbins2002). The results of lesion and gray matter volume estimation studies do not necessarily converge. The current results also converge better with activation studies, that VBM has more in common with methodologically (Bolla et al., Reference Bolla, Eldreth, Matochik and Cadet2005; Christakou et al., Reference Christakou, Brammer, Giampietro and Rubia2009; Tucker et al., Reference Tucker, Potenza, Beauvais, Browndyke, Gottschalk and Kosten2004). While lesion studies can clearly inform on the neural substrate of IGT performance and decision-making, the results are limited to the selected regions. Brain-wise VBM provides a “bird's eye” view of the substrate. Perhaps the disparate findings can be viewed with an information processing perspective (Barbas & Zikopoulos, Reference Barbas and Zikopoulos2006). The amygdala, insular, parietal cortex, and posterior OFC may serve as “up-stream” regions for emotional decision-making, while the lateral prefrontal cortex represents the downstream end-point (Barbas & Zikopoulos, Reference Barbas and Zikopoulos2006; Gansler et al., Reference Gansler, Lee, Emerton, D'Amato, Bhadelia, Jerram and Fulwiler2011). Participants in neuroimaging studies have functional upstream components of the neural circuit used to complete the task, which may contrast with participants in lesion studies. The bias of neuroimaging studies to find results in downstream components is likely to be a consequence of this correlation of observed task performance in a relatively intact brain, as they are the last regions to influence behavior before it is affected. This does not indicate that upstream elements have been taken out of the circuit; it simply suggests that more sophisticated neuroimaging experimental methods (i.e., functional connectivity) may be necessary to adequately correlate activation in those upstream regions with observed behavior in a manner that more closely reflects the results of lesion studies.
In sum, the IGT paradigm displays robust neural correlates supporting its use in clinical or research measurement of decision-making processes. However, D-A IGT metrics appear to be much more robustly represented across the decision-making neural pathways than the conventional CD-AB metric. The D-A 1-100 metric appears to represent overall decision-making capacity well, and would represent a more sound omnibus metric for clinical or research use than CD-AB. Previous work shows the D-A 41-100 metric best captures the executive functioning aspects of decision-making, and, consistent with that, it does correlate with dorsolateral prefrontal cortex, while D-A 1-40 correlates with ventrolateral prefrontal cortex. The CD-AB metric appears to be based on too broad an amalgamation of processes to yield clear representation in brain, whereas the D-A metric may succeed because its discrete lateral frontal and sub-cortical correlates represent complimentary aspects of the information processing sequence. Furthermore, the decks containing frequency of reward clues (C) and magnitude of reward clues (D) have distinct neural correlates reflecting a decision-making hierarchy. The reward frequency clues appear to be represented limbically, whereas the reward magnitude clues are represented fronto-limbically. As real and facsimile reinforcers produce the same performance patterns, the neural correlates observed here likely subserve those involved in “real-world” decision-making.
Limitations of the current study include a study sample based on healthy controls, which might preclude generalization of the findings to clinical populations. IGT performance was recorded by blocks of trials rather than individual items, precluding a more detailed study of the impact of response contingency “history” on performance. Furthermore, the neural substrate in this study is operationalized by gray matter volume estimates and conclusions could differ from studies in which the neural substrate is operationalized by hemodynamic response function.Fig. 1
Fig. 1 Positive regression of D-A 41-100, clusters displayed restricted to those greater than 1000 voxels in size.
Acknowledgments
The authors of this study have no financial conflicts of interest based in this project. This work was supported by the National Institute of Mental Health (MH60504: Aging, Brain Imaging and Cognition study).
