Solving analogies has been considered to be a paramount accomplishment in relational reasoning (Gentner, 1983; Hummel and Holyoak, 2003). Analogy performance requires the ability to understand relations among two or more elements in a source analog and to map that relation onto other relations present in a target analog. These capacities develop across childhood (Goswami and Brown, 1989; Gentner and Rattermann, 1991; Goswami et al., 1998; Rattermann and Gentner, 1998; Goswami, 2001) as the ability to understand similarity improves (Goswami and Brown, 1989; Goswami et al., 1998). Another key factor relevant to analogy performance is working memory (WM) function with increases in WM capacity leading to concurrent expansion of relational reasoning abilities (Waltz et al., 2000; Morrison et al., 2004). Achieving successful solutions to analogy problems places high demand upon WM and may also invoke inhibitory control processes under conditions when perceptual or feature similarities between individual objects exist (Gentner and Toupin, 1986). Such conditions have been demonstrated to increase processing time and error rates in analogy problems (Morrison et al., 2004; Viskontas et al., 2004).

A clear form of analogy problems in which relational and perceptual feature similarity compete is in picture analogies. Markman and Gentner (1993) tested participants on a set of scene analogy problems that included relational matches as well as perceptual distractors. Solutions to such problems were demonstrated to be dependent upon executive WM in a dual-task study (Waltz et al., 2000). A more extensive set of picture analogy problems was recently developed by Richland et al. (2006). These investigators varied relational complexity of the analogies finding that problems involving two-relations were more difficult than those that involved only one relation. Additionally, the presence of perceptual distractors was varied such that problems with distractors were solved at lower levels than those without distractors. Developmentally, children ranging in age from 3- to 7-years old were not as competent as 13- to 14-year-old adolescents. Thus, analogical reasoning abilities increase with age and depend upon WM abilities that enable relational processing and inhibitory control over competing perceptual alternative matches.

Theoretical models of analogical reasoning have implicated executive control functions as being central to the analogical mapping process. Learning and inference with schemas and analogies (LISA) is a model that has emphasized that the limits of working memory also limit the ability to map elements of analogous situations together (Hummel and Holyoak, 2003). Further, Hummel and Holyoak (1997) demonstrated that developmental limits on analogy can be simulated by reducing the working memory capacity of that model. Broader models of cognition, such as ACT-R have further linked analogical reasoning to working memory processes through emphasizing active memory retrieval (Salvucci and Anderson, 2001). These theoretical models provide motivation to test the degree to which executive functions are related to analogical reasoning in the current investigation.

Networks involving the prefrontal cortex (PFC), the temporal cortex, and the parietal lobes may be particularly relevant to analogical reasoning based on prior findings in populations with frontotemporal lobar degeneration (Morrison et al., 2004; Krawczyk et al., 2008). The results of TBI working memory studies suggest that older children and early adolescents may have deficits in n-back task performance (Levin et al., 2002, 2004) and that WM performance is related to injury severity and time post-injury with children sustaining mild to moderate TBI showing little change over time, whereas children with severe TBI perform worse overall, and show a decline in performance with time. Evidence from neuroimaging suggests that adolescents with moderate to severe TBI show over-activation of PFC and parietal regions related to maintenance of information in WM relative to typically developing (TD) youth, and that this may represent an inefficient over-recruitment of cortical activity in compensation for the brain injury (Newsome et al., 2008).

Inhibitory control deficits in working memory are likely to be particularly problematic for the reasoning abilities of individuals with TBI, as inhibitory deficits have been strongly implicated in failures of reasoning in adults (Viskontas et al., 2004; Richland et al., 2006). Dennis et al. (2001) demonstrated that children with TBI had difficulties screening out irrelevant digit information in a vigilance task. Other studies have reported inhibitory control deficits in TBI individuals in go-no-go tasks (Konrad et al., 2000a,b), stop-signal tasks (Schachar et al., 2004), and shifting tasks (Levin et al., 1993).

Adolescents with TBI frequently have disruptions of the frontal networks associated with executive functions relevant to reasoning (Levin and Hanten, 2005). This leads to deficits in working memory and inhibitory control. Such deficits may also lead to disabilities with abstract reasoning. MRI studies of children with TBI (Levin et al., 1997; Wilde et al., 2005) have supported this link revealing both focal lesions and diffuse injury, particularly to the prefrontal, frontal, and temporal regions relevant to cognitive control. Prefrontal regions are vulnerable due to their proximity to the sphenoidal ridges and bony protrusions on the base of the skull and gray matter in these regions has been found to be reduced even in the absence of focal lesions (Berryhill et al., 1995). MRI studies conducted several months to years post-injury have shown that loss of white matter can be extensive (Wilde et al., 2005; Ghosh et al., 2009; Bigler et al., 2010). Diffuse cortical thinning has been reported in children with moderate to severe TBI as compared to typically developing children (Merkley et al., 2008) or children with orthopedic injury (McCauley et al., 2010). Thus both gray and white matter disruptions likely contribute to the working memory deficits observed in TBI adolescents. While there is little current literature on relational reasoning after TBI, the presence of executive function deficits in TBI predict that they may be impaired on analogical reasoning. TBI has been demonstrated to decrease performance on the Raven's progressive matrices task (Hiscock et al., 2002), which is a relational reasoning task closely linked to analogical reasoning performance (Morsanyi and Holyoak, 2010; Geake and Hansen, 2010). Given the declines after TBI in both basic executive functioning (Levin and Hanten, 2005; Muscara et al., 2008) and in Raven's matrices (Hiscock et al., 2002), it is predicted that adolescents with TBI will have decreases in analogical reasoning performance. Additionally, the evidence of inhibitory control declines after TBI support the prediction that distractor items are likely to decrease performance on picture analogies (Leblanc et al., 2006; O'Keeffe et al., 2007).

We compared adolescents with TBI and control participants who were TD youth on analogical reasoning and executive function tasks. Participants completed a battery of scene analogies based on a set previously developed by Richland et al. (2006). We manipulated the number of relations relevant to solving the analogies (one-relation versus two-relation). We predicted that both groups would show differences in relation numbers, but that TBI participants will show greater deficits in response to the greater relational demand, as these participants are likely to have both WM maintenance and executive control deficits (Levin et al., 2002, 2004). We also tested for the effect of distractors upon reasoning performance in TBI participants. We predicted that TBI participants would exhibit deficits in analogical reasoning due to their tendency to have reduced cortical thickness related to executive functioning (Levin et al., 2008). This prediction follows from prior studies indicating that executive control deficits lead to failures of reasoning in young children (Richland et al., 2006) and older adults with dementia (Morrison et al., 2004; Krawczyk et al., 2008). Lastly, we modified the problem set to compare living and non-living analogy problems. We included an exploratory social component in the current study by varying the type of problems so that some analogies focus on relations among living elements, emphasizing intentions, and cooperation, while others focus on relational correspondences among inanimate objects, emphasizing spatial relationships. This was included to test whether TBI adolescents would show selective deficits in reasoning about relationships among living items based on prior evidence of their deficits in perceiving social situations (Janusz et al., 2002; Yeates et al., 2002; Hanten et al., 2008).

We administered two executive function tasks in order to assess the degree to which WM updating and inhibitory control abilities interact with reasoning performance. These tasks assess the ability to update WM and to allocate attention strategically toward high priority information while screening out lower priority information. These skills were predicted to be important for analogical reasoning, as prior evidence suggests that executive WM dual tasks disrupt the ability to produce solve picture analogies (Waltz et al., 2000). Similar deficits have been observed in picture analogy performance under high anxiety conditions that would likely reduce WM resources (Tohill and Holyoak, 2000). We predicted that executive control abilities within WM would be predictive of performance on the analogical reasoning task and may underlie the performance deficits predicted to be present in TBI adolescents particularly when problem complexity is higher and distraction must be screened out.

We also collected MRI data from the participants in order to assess cortical thickness related to the analogy task. While there is little literature describing the relationship between morphometry data and analogical reasoning, prior studies of analogy using functional MRI have demonstrated activation of frontal, parietal, and temporal cortex, often with a left hemisphere asymmetry (Wharton et al., 2000; Bunge et al., 2005, 2009; Luo et al., 2003; Green et al., 2010). Several studies have indicated that the frontal pole is particularly important for relational responding (Smith et al., 2007; Christoff et al., 2009; Wendelken et al., 2008). Studies of relational reasoning in patients with frontotemporal lobar degeneration have also demonstrated the necessity of intact PFC for matrix reasoning, transitive reasoning (Waltz et al., 2000), verbal analogy, and visual analogy performance (Morrison et al., 2004; Krawczyk et al., 2008). Notably, individuals with PFC damage caused by dementia demonstrated additional impairments when distraction needed to be screened out in these prior studies. Furthermore, temporal cortex was implicated as relevant in these studies based on its central role in semantic memory (Morrison et al., 2004; Krawczyk et al., 2008; Morrison and Soohyun, 2008). We predicted that healthy control participants would show relationships between cortical thickness and analogy performance in the temporal cortex and regions in the lateral PFC including the left frontal pole.

Twelve youth who sustained a moderate-severe TBI (MSTBI) at least 1 year prior were recruited from a cohort of children with TBI, specifically from the Baylor College of Medicine and University of Texas at Dallas sites (7 males and 5 females; age at test: M = 16.51 years, SD = 2.14, range 12.79–19.12; Glasgow coma scale (GCS) score: M = 7.33, SD = 3.89, range 3–15; post-injury interval: M = 2.65 years, SD = 0.76, range 0.94–3.62; parental education: M = 12.83 years, SD = 3.32, range 6–16; ethnicity: 7 Caucasian, 1 African-American, 4 Hispanic). Severe TBI involved a non-penetrating TBI arising from closed head trauma (i.e., acceleration, deceleration, or impact by a blunt object) with lowest post-resuscitation GCS score of 3–8 indicative of coma that is not due to child abuse and is not complicated by hypoxia (PO2 < 96 mmHg or systolic blood pressure ≥ 2 SDs below mean for age) for 30 min or longer after resuscitation. In cases of delayed neurologic deterioration, the worst GCS score was be used. Following convention (Teasdale and Jennett, 1974), the best level of response recorded for each component of the GCS at a specific time was used to obtain the sum GCS score. Injury variables, including in-hospital CT scan and pupillary response were recorded (Saatman et al., 2008). Other inclusion criteria were age 12–18 years, right handedness (Oldfield, 1971) to mitigate heterogeneity in brain asymmetries, cognitive ability to assent or consent (depending on age). The classification of moderate TBI differed from severe TBI by a lowest post-resuscitation GCS score of 9–12, indicating impaired consciousness but not coma (Teasdale and Jennett, 1974) or a GCS score of 13–15 with an acute brain lesion on the CT scan within 24 h after injury. Patients with brain lesions despite mild impairment of consciousness were included because they frequently exhibit cognitive deficit at 12 months post-injury (Levin et al., 2008). Other injury variables were recorded as described for severe TBI. Table 1 shows lesion location, GCS, and type of accident for TBI participants, 10 of whom had clear lesions. Eleven TD children were recruited from our cohort of healthy subjects in the local community (6 males and 5 females; age: M = 16.37 years, SD = 1.89, range 13.94–19.24; parental education: M = 15 years, SD = 2.37, range 11–19; ethnicity: 6 Caucasian, 1 African-American, 4 Hispanic). There were no significant group differences in gender, Fisher's exact test p = 1; age, t(21) = 0.17, p = 0.8702; parental education, t(21) = −1.87, p = 0.0888, or ethnicity, Fisher's exact test, p = 1. Informed consent was obtained prior to the experiment for all participants. Experiments were approved by the Institutional Review Boards at Baylor College of Medicine and The University of Texas at Dallas.

T1-weighted 3D turbo field echo (TFE) axial acquisition series were performed on Philips Intera 3.0T whole body scanners (Philips, Cleveland, OH, USA). Parameters included 1.0-mm thick slices, 0-mm slice gaps, echo time (TE) = 4.6 ms/repetition time (TR) = 9.9 ms, field of view (FOV) = 240 mm, and a reconstructed isotropic voxel size of 1.0 mm. Additional parameters included turbo factor = 175 and flip angle = 8°.

Repeated measures ANOVAs were conducted on the four analogy conditions (group, presence of distractor, relational level, animacy). Age was also entered into the model in order to control for age effect and to see whether it modifies the group differences on the outcomes.

The SLT generated measures of overall word recall for the two non-strategic recall conditions (i.e., pre- and post-strategic memory trials). Word recall means were also generated for the strategic conditions (i.e., slow, medium, and fast). In addition, mean recall efficiency scores were also calculated for the three strategic conditions. This resulted in five recall scores and three efficiency scores. These eight scores were also tested for correlation with mean proportion correct for each of six problem categories from the picture analogy task (one-relation problems, two-relation problems, living problems, non-living problems, distractor, and non-distractor problems) using Spearman correlation coefficients separately for each participant group.

The KTT produced one percentage correct score for each of the performance conditions (i.e., two, three, or four category conditions). These scores were tested for correlation with the six picture analogy problem type means using Spearman correlation coefficients for each participant group separately.

Typically developing youth were significantly better at reasoning analogous roles among elements in scenes than were youth with TBI, F(1,17) = 10.55, p = 0.0047 (refer to Figure 2). Age had an effect that depended on group, F(1,17) = 5.58, p = 0.0303, such that younger adolescents with TBI performed more poorly than older adolescents, but this age difference was not apparent in TD youth.

Correlational relationships were measured by correlating cortical gray matter thickness with task performance using a general linear model. The strongest correlations were found for the TD participants for overall task performance. Figure 3 shows regions of the brain showing an inversely correlated relationship between overall analogy accuracy and cortical thickness. Notably, the anterior PFC (left BA10), bilateral anterior and posterior lateral PFC, bilateral superior and inferior temporal gyri, and the medial PFC showed these correlations in the TD group. There was less evidence of a clear correlational relationship in the TBI group and performance overall, but some inverse correlations were observed in the left medial orbitofrontal cortex and the left superior frontal gyrus. Similar relationships in similar areas were observed for accuracy on all two-relation analogies and gray matter thickness in the TD group (see Figure 4). Accuracy on all problems with distractors present showed an inverse correlational relationship with cortical thickness with foci in the left superior temporal gyrus, left middle temporal gyrus, right inferior frontal gyrus, left posterior cingulate for the TD group. Additional foci observed in the TBI group included the anterior left dorsal PFC and right orbitofrontal cortex (refer to Figure 4). Overall, the cortical thickness data showed the strongest correlational relationships between thinner cortex and superior performance overall and at the higher relation level in TD adolescents. This is broadly consistent with cortical maturity, or efficiency in anterior PFC and temporal lobe regions.