Functional impairments seen in pediatric BD include dysfunction in cognitive flexibility, defined as the ability to adapt one's thinking and behavior in response to changing environmental conditions, such as rewards (5, 6). The ability to alter one's response to changing reinforcement contingencies is one aspect of cognitive flexibility that can be measured in the laboratory using reversal learning tasks. In such tasks, two stimuli, A and B, are presented, and participants must use trial-and-error learning to determine that stimulus A but not B is rewarded. Then, without warning, the stimulus/response relationship is reversed, so that participants must learn that now B but not A is rewarded (7).

From an affective neuroscience perspective, many clinical symptoms of BD suggest impaired cognitive flexibility and adaptation to changing stimulus-response-reward contingencies, as occurs on reversal learning tasks. For example, mania can be viewed as a hyperhedonic state characterized by excessive involvement in pleasurable activities with high potential for painful consequences and abnormally inflated self-esteem or grandiosity. In contrast, depression may represent a hypohedonic state characterized by decreased interest in such activities, anhedonia, and feelings of worthlessness (8). Furthermore, reversal learning deficits may play a role in the irritability found during mania, depression, and euthymia in BD, since participants may become frustrated when they continue to anticipate but not receive rewards (9–12).

BD youths show significant impairment in reversal learning (13–17), a task that in healthy people engages neural regions previously implicated in BD. Specifically, successful reversal learning depends on engagement of the medial orbital frontal cortex (mOFC) for representation of reinforcement contingencies and prediction error signaling, allowing recoding of the reinforcement value associated with the response (18–20). Ventral striatal engagement further facilitates performance, allowing transformation of concrete stimulus exemplar information into motor responses (5, 21, 22). Additional areas of the frontal cortex that are involved in reversal learning include the dorsomedial frontal cortex (dmFC) and cingulate cortex. These regions enable mediation of response conflict between the executed incorrect response and the alternative response, as signaled by punishment (23, 24). In turn, the dmFC is hypothesized to recruit other regions implicated in top-down attentional control, including the lateral superior frontal cortex (lsFC) and parietal cortex (25, 26), as well as regions implicated in object/response selection, including the inferior frontal cortex (iFC) (27, 28), to orchestrate a change in response on the subsequent trial (5, 18, 24, 29).

While no prior imaging studies use reversal learning paradigms to study BD, prior work suggests that regions implicated in reversal learning may be dysfunctional in BD. For example, pediatric and adult BD structural magnetic resonance imaging (MRI) studies have shown volumetric alterations in components of the circuit engaged by reversal learning, including the dmFC/cingulate cortex, lsFC, iFC, striatum, and parietal cortex (30–41). Functional MRI (fMRI) studies of BD adults have demonstrated altered neural activation in these regions on tasks besides reversal learning paradigms. For example, BD adults have reduced OFC activity during an emotional go/no-go paradigm, reduced lsFC activity during affect discrimination and in response to emotional stimuli, and reduced iFC activity during decision making (42–45). Notably, a recent study reported reduced OFC, striatal, and anterior cingulate cortex (ACC) habituation-related activity in BD adults (46). Work with pediatric BD has also typically revealed reduced attention-related activity within iFC and lsFC in response to emotional provocation (47–49). Moreover, reduced striatal and iFC activity was seen in pediatric BD patients when failing to inhibit their responses on a motor inhibition paradigm (50, 51), whereas increased lsFC activity was necessary for successful performance on change trials on a motor control task for BD youths (52).

Based on these prior findings, we hypothesized that reversal deficits observed in youths with BD result from dysfunction either in regions putatively signaling reinforcement/prediction errors (i.e., mOFC) and/or in regions involved in related tasks, including processing response conflict, implementing alternative responses, and controlling attention (i.e., dmFC, lsFC, iFC). We tested this hypothesis in euthymic BD youth (n = 16) and typically developing controls (n = 16) using fMRI during performance of a probabilistic reversal paradigm. The paradigm we used was adapted from one on which we previously demonstrated behavioral impairments in pediatric populations, and whose mediating neural circuitry has been characterized in both adult and youth samples using fMRI (7, 15–18, 53). We examined euthymic patients with pediatric BD in this study to minimize the impact of mania or depression on any between-group differences.

Our work is the first to begin to elucidate the neural underpinnings of cognitive flexibility using a reversal learning task in pediatric BD. Compared to controls, BD youths had greater activation during the reversal phase, particularly during reversal errors, in the dmFC, iFC, lsFC, and parietal cortex. It has been hypothesized that the dmFC is engaged during the response conflict following punished reversal errors, and that this region recruits regions implicated in top-down attentional control and object/response selection to orchestrate a change in response on the subsequent trial (18, 23). We found increased activity in BD versus controls during punished reversal errors in the dmFC, lateral regions of the superior and middle frontal cortex, and the parietal cortex. This finding suggests two possibilities. First, since task performance did not differ between groups, this increased activity may reflect inefficient recruitment of these regions to achieve satisfactory task performance. Second, this increased activity may reflect compensatory activity in response to dysfunction elsewhere.

The available literature provides more support for the first possibility. In previous work with euthymic adults with BD, there have been reports of enhanced activity within these regions during successful performance on other tasks that engage psychological processes involved in reversal learning. Thus, for example, euthymic BD adults show greater iFC activation than controls during an emotional go/no-go task when inhibiting emotional versus neutral stimuli (62). Euthymic BD adults also show greater neural activation than controls in the parietal cortex during a two-back working memory task (63). In addition, euthymic BD adults show greater PFC activation than controls during an affective face-matching task (64). Moreover, pediatric BD participants demonstrated increased lsFC activity when successfully performing change trials on a motor control task (52). In contrast, we did not find support for the second possibility (compensation) because we did not detect neural alterations in other areas, such as the mOFC, for which these fronto-parietal alterations would be compensating. However, caution is urged so as not to commit a type II error by overinterpreting a negative finding, and thus the current data cannot fully distinguish between these two possibilities.

In addition to showing increased activity versus controls in dmPFC and parietal regions during punished reversal errors, on these trials BD youths also showed less deactivation than controls in the posterior cingulate cortex and precuneus. Previous work is consistent with ours in finding that, during punished reversal errors, healthy individuals show decreased activation to punishment versus reward within these regions (7, 18). Decreased neural responding following punishment has been assumed to relate to recoding of the reinforcement value associated with the response (7, 18). Thus, our data suggest the possibility that a reduced ability to recode reinforcement values requires a compensatory increased recruitment of the dmFC, lsFC, and parietal cortex in order to achieve successful response change.

Importantly, we did not find between-group differences in the mOFC. It is possible that this represents a type II error, and that OFC dysfunction would be detected with either a larger sample of BD and controls, or a more homogeneous sample of BD youths. For example, gender effects have been demonstrated in BD youths in several brain regions, including the OFC (40, 65). While our study is underpowered to examine such possibilities, another possibility is that reversal learning impairments do not involve the OFC in pediatric BD. Support for the latter possibility comes from another study using this PRR paradigm, which found that both typically developing control children and those with ADHD had the expected decrease in BOLD signal in the OFC to reversal errors, but those with psychopathic tendencies did not (7). That study, by Finger et al. (7), involved three groups of 14 children, none of whose data are included in our present study. Additional support comes from a recent study of BD adults (66) that did not find differences in OFC volume or that of its subregions (including the mOFC) in BD adults compared to controls, although differences were found between depressed and euthymic BD adults. Thus, it is possible that there are diagnosis-specific effects on the brain / behavior interactions underlying reversal learning in child psychiatric disorders. Our future work is geared toward addressing both possibilities.

Although highly speculative, there are several potential clinical implications of our present study. Our data potentially suggest inefficient functioning of a series of regions implicated in attentional control and response selection. In turn, this suggests the possibility that BD youths may have difficulty maximally utilizing psychotherapies relying on such cognitive capacities. However, studies have demonstrated that when such therapies are modified to address BD-specific deficits in cognitive flexibility—e.g., when they include skills training to address impaired problem solving and affect regulation—they have great promise in the treatment of BD. Such therapies include cognitive behavioral therapy [studied in BD adults and children (67)], family-focused therapy [studied in BD children (68)], and interpersonal social rhythm therapy [studied in BD adults (69)]. In this context, it is also interesting to consider whether the brain / behavior interactions underlying cognitive flexibility and reversal learning impairments in pediatric BD may respond to training, such as the use of specialized computer games for `cognitive remediation' (70, 71).

There are several caveats and limitations of our study. First, while we previously found impaired performance on reversal learning paradigms in children with BD (15–17, 53), we observed no significant group differences in behavior here. This contrast with previous behavioral studies likely reflects task differences between the current fMRI study and previous behavioral work. Importantly, the failure to find between-group behavioral differences eliminates the possibility that our fMRI results are epiphenomena of, or are confounded by, group differences in task performance. Along these lines, it is common for fMRI studies in psychiatric patients to find between-group differences in neural activity without between-group behavioral differences; an example is the recent study in children with psychopathy versus controls that used the same reversal learning paradigm as here (7). Indeed, some have argued that fMRI is more sensitive than behavior in detecting important between-group differences (72). On the other hand, others suggest that the absence of between-group behavioral differences complicates attempts to link the observed differences in brain activity to symptoms (73). Although from this perspective, the lack of between-group behavioral differences might call into question whether reversal tasks engage neural circuitry relevant to the pathophysiology of BD, ample evidence, including our own work in pediatric BD and that of others in BD adults, documents that reversal learning and cognitive flexibility deficits are present in BD (74–76).

Other potential limitations of our study include potential heterogeneity in the BD sample, including that due to psychiatric comorbidities, psychotropic medications, BD subtypes, and possible subsyndromal mood symptoms, as well as the wide age range of our participants. All of our BD youth had one or more co-occurring psychiatric disorders, most commonly ADHD or anxiety disorders. This is consistent with prior research showing that both pediatric and adult BD are characterized by high rates of comorbidity (77–80). However, our exploratory secondary analyses showed that our primary findings between patients and controls remained significant even when restricting the BD sample to those without ADHD (n = 6) or to those without an anxiety disorder (n = 10), suggesting that our results are not confounded by comorbid ADHD or anxiety. A study of adolescents with ADHD found no atypical neural response associated with acquisition correct versus reversal incorrect responses when performing this same reversal learning task (7). Hence, our results here add to an emerging body of literature suggesting that ADHD presenting in the context of BD may be a phenocopy of ADHD presenting alone (14, 81). Nonetheless, further work is necessary to directly evaluate the specificity of brain / behavior interactions underlying reversal learning in pediatric BD. For example, such work might compare directly youth with BD to those with either primary ADHD or primary anxiety disorders, as well as comparing BD youth with versus without comorbid ADHD and BD youth with versus without comorbid anxiety (82, 83). Similarly, our post-hoc analyses excluding the 1 / 16 BD participants with type II BD (rather than type I), confirmed our primary result. However, further work to explore how BD subtypes differ in the brain / behavior interactions underlying reversal learning is warranted. In addition, while all of our participants were euthymic by the definition of having CDRS scores<40 and YMRS scores <12, we cannot rule out the possibility that subsyndromal mood symptoms were nonetheless present and impacted on our findings (84, 85). However, the lack of significant correlations between neural activation during reversal and YMRS or CDRS scores suggests that subsyndromal mood effects are less likely. Moreover, since the K-SADS was administered at enrollment and not at scan day, we cannot say for certain that BD participants did or did not meet K-SADS mania or depressive episode criteria on the scan day; however, it is highly unlikely given their mood (YMRS, CDRS) scores.

With respect to medications, most of our BD participants (13/16) were taking a combination of psychotropic medications, with no single agent predominating. To begin addressing the potential confound of medications on our results, we conducted exploratory analyses of those BD youth taking either atypical antipsychotic or anticonvulsant medications compared to those not, since 50% of the BD participants were taking either medication. Such analyses failed to show any difference in group-by-phase neural activation among BD youth based on either medication category. Recent evidence suggests that medication may bias toward type II, rather than type I, error (86). Also, given the variety of our patients' medication regimens, it is unlikely that medications would account for the consistent and relatively specific findings that we observed. Nevertheless, additional studies that pair treatment and neuroimaging in their design are needed to examine whether pharmacotherapy can reverse the cognitive flexibility deficits present in BD and, if so, the neural mechanisms that might account for this effect.