Differential profiles of cognitive and behavioral inflexibility in addictive disorders

Patients with addictive disorders (encompassing substance and behavioral addictions) often behave in ways that have been described as rigid and inflexible. This behavioral profile has been proposed to be explained by impairments in cognitive and behavioral flexibility that are shared by all addictive disorders. To evaluate this assertion, we reviewed the literature to determine similarities and differences in the performance of patients with either substance-related or behavioral addictions on well-established paradigms of cognitive and behavioral flexibility. Contrary to the widely-held view, we found that different addictive disorders have contrasting profiles of cognitive and behavioral inflexibility, suggesting that inflexible behavior — though conceptually similar in all addictive disorders — appears to be underpinned by different neuropsychological processes


Introduction
Pathological drug use and gambling are often considered exemplars of inflexible behavior because they persist, even when they perpetuate physical and mental harm.From a neuropsychological perspective, this behavioral profile is thought to reflect impairments in flexibility that are shared across all substance-related and behavioral addictions [1,2].However, this proposal has raised the question as to whether inflexibility indeed manifests in the same way in different types of addictive disorders [3].Here, we present a narrative review of the experimental literature to examine the similarities and differences of inflexibility in different addictive disorders on widely established neuropsychological paradigms.Despite the large body of animal studies investigating the relationships between different addictive drugs and inflexibility, the present review deliberately focuses on studies in humans.

Experimental paradigms to study inflexibility
The construct of flexibility can be differentiated into a cognitive and a behavioral component [4].Cognitive flexibility typically refers to the ability to set shift, that is, the ability to mentally switch between an established mindset to a different one following a rule change, whereas behavioral flexibility refers to the ability to adapt an established response set following a change in reinforcement contingencies.Evidently, the differences between these two types of flexibility are subtle because both are measured in an experimental setting through a behavioral response, which is why many studies use them interchangeably.Here, we will discuss them separately for the following two reasons: first, neuroscientific evidence suggests that these two constructs are dissociable as they are subserved by different neural substrates -the ventrolateral prefrontal cortex underpins cognitive flexibility, whereas the orbitofrontal cortex governs behavioral flexibility [5,6].Second, there are notable differences in the experimental paradigms used to measure them (which we will explain below).
Cognitive flexibility is typically assessed using setshifting tasks that measure one's ability to adapt to changing rules.Two widely used paradigms that measure cognitive flexibility are the Wisconsin Card Sorting Task (WCST) [7] and the Intra-/Extra-Dimensional setshifting task (IDED) [8].In the WCST (Figure 1a), participants are instructed to sort cards based on a hidden rule (e.g.sort by shape, color, or number).However, participants are not told what the rule is but would need to learn it from deterministic corrective feedback.After 10 consecutive correct responses, participants will have completed one set, and the sorting rule changes implicitly; again, participants are not informed about this rule change but should learn about it through deterministic corrective feedback and update their sorting strategy accordingly.Consequently, correct card sorting on the WCST relies on intact learning and memory, which enables participants to switch mentally from one rule to another.It is therefore important to differentiate different task measures: general learning performance is reflected by the number of completed sets and non-perseverative errors, whereas cognitive inflexibility is reflected by the number of perseverative errors -the number of wrong card sorts after rule change that are based on the old rule.An increase in perseverative errors (i.e.consecutive wrong responses based on the old rule) in the WCST indicates a reduction in cognitive flexibility, as these errors occur when people fail to cognitively switch to new abstract rules.
The IDED task (Figure 1b) is a computerized attentional set-shifting task adapted from the WCST.On each trial, participants are presented with two images (pink shapes).Like in the WCST, participants are instructed to select the correct option on each go, following a rule that can be learned through deterministic corrective feedback (e.g. a specific shape).Once participants meet the learning criterion (as measured by six consecutive correct responses), the rule changes, requiring participants to adjust their strategy by selecting the other shape.However, unlike the WCST, more dimensions are introduced in the IDED during the task (e.g.lines overlaid on pink shapes).Out of nine stages, two are critical: the intra-dimensional shift (ID-shift; when the rule switches from stimuli to another of the same dimension, that is, shape to shape [stage 6 in Figure 1b]) and the extra-dimensional shift (ED-shift; when the rule switches from one dimension to another, that is, from shapes to lines [stage 8 in Figure 1b]).Crucially, the ED stage measures how well participants disengage their attention from one dimension to another -the key stage that measures cognitive flexibility.Task performance is typically assessed using the number of errors at different stages; errors that participants make immediately after the ID-shift (or the total errors before the ED stage) are thought to reflect learning deficits, whereas errors made after the ED-shift are indicative of cognitive inflexibility.It is noteworthy that some authors reported the total adjusted errors on the IDED task as a measure of cognitive inflexibility; however, this measure is unspecific to the phases and, therefore, more likely to reflect general learning performance rather than cognitive inflexibility.
In contrast to cognitive flexibility, behavioral flexibility is typically measured using a probabilistic reversal learning (PRL) task (Figure 1c) [9,10], which tests the ability to adjust behavioral responses according to changes in reinforcement contingencies.Participants are typically presented with two options on each trial and would need to learn by trial and error to select the option that is most often rewarded.Different from the WCST and the IDED task, the feedback provided after choice selection is probabilistic, which means that participants sometimes receive misleading negative feedback; that is, in 20% of the trials, they are told that they made a wrong choice, although they selected the correct option.Once the learning criterion has been achieved, the reinforcement contingencies reverse such that the option that was previously wrong becomes the correct one.Depending on the PRL task version, the learning criterion is defined either by the number of consecutive correct responses or the completion of a number of trials.Once the rule reversal occurs, participants have to then adjust their responses to this new rule.As feedback delivered in this task is probabilistic, optimal task performance thus requires a healthy trade-off between ignoring spurious misleading feedback and adjusting their actions when the contingencies change.Behavioral inflexibility is measured by the number of perseverative errors on the PRL task, which occur when people do not behaviorally adjust to changed reinforcement contingencies.Other measures of interest include the probability of sticking to a previously rewarded response (i.e.win-stay) and the probability of switching after receiving negative feedback (i.e.lose-shift) to gauge participants' sensitivity to reinforcing feedback.

Inflexibility in addictive disorders
A summary of the key findings on the aforementioned tasks in patients with addictive disorders can be found in Tables 1-3.It is important to note that published studies vary considerably in terms of patients' diagnoses, drug use status (active drug use versus drug abstinence), how rigorously patients have been matched with control participants, sample sizes, and the analytical methods used.Overall, the available evidence suggests that addictive disorders are generally associated with inflexibility, but the manifestation of this inflexibility differs substantially across different types of addictive disorders.

Cocaine use disorder
Converging evidence suggests that cocaine use disorder is associated with behavioral, but not cognitive inflexibility.Several studies have shown that task performance on the WCST and IDED task in patients with cocaine use disorder were not measurably different compared with healthy control participants, irrespective of whether patients were drug abstinent or currently using cocaine [11][12][13][14][15][16][17][18][19].There have, however, been reports that recreational cocaine polydrug users (who do not meet the Diagnostic and Statistical Manual-intraveneous-TR criteria for cocaine dependence) make more WCST perseverative errors [20] and more ED-shift errors [21] compared with controls.Given the small sample size of recreational users and the level of polydrug use, it is difficult to ascertain whether this unusual profile really is associated with recreational cocaine use or with the concurrent use of other drugs.Whilst there is little evidence for cognitive inflexibility in patients with cocaine use disorder, there is ample evidence for behavioral inflexibility.Several studies have identified increased perseverative errors during the PRL task -a finding observed in both active [22,23] and abstinent cocaine users [24], suggesting that these deficits may persist during recovery.It is also of note that the longer patients have been using cocaine, the more perseverative errors they made during the PRL task [25,26], which may reflect a drug-induced deficit (i.e.cocaine use may impair the ability to flexibly adjust behavior).Although correlations do not allow inferences about causation, preclinical research has provided compelling evidence for orbitofrontal dysfunctions following chronic cocaine exposure [27], which has been shown to impair PRL task performance [28,29].More recent studies have attempted to use computational models to explain why perseverative errors occur in cocaine use disorder.These studies have identified evidence for an increased tendency to repeat prior responses irrespective of outcomes (also known as 'stickiness') in patients [30,31], possibly indicating a latent preference for automatic responding (in other words, 'stickiness' might be a precursor of habits).

Amphetamine-type psychostimulant use disorder
Unlike patients with cocaine use disorder, there is substantial evidence suggesting that patients addicted to amphetamines or methamphetamine have deficits in cognitive flexibility.The majority of studies in patients with methamphetamine use disorder found increased perseverative responses on the WCST [16,[32][33][34], indicating difficulties with mentally switching between different abstract rules (see Refs. [17,35] for some exceptions).Interestingly, this deficit seems to recover   after prolonged abstinence; that is, the longer patients have been drug abstinent, the more likely they perform on par with controls [34,36,37].Besides perseverative errors, patients with methamphetamine use disorder also show an increase in nonperseverative errors on the WCST, which points toward difficulties in learning or maintaining learned responses [32,34].These learning difficulties seen on the WCST (where the feedback is deterministic) need to be differentiated from the PRL task (where the feedback is probabilistic).Several studies did not find PRL perseverative errors in patients with methamphetamine use disorder but instead observed that they made errors due to increased switching between two options [38,39].The increased switching between the options is a reflection of difficulties with contingency learning in general, but not of behavioral inflexibility.This is also in keeping with preclinical evidence showing that chronic amphetamine exposure disrupts synaptic plasticity (e.g. the formation of dendritic spines) that is important for contingency learning [40].Although it seems striking that cocaine and amphetamine-like use disorders differ in their behavioral profile (given that both are psychostimulants that increase extracellular dopamine), there are noteworthy differences in the drugs' relative affinities for monoamine transporters.For instance, cocaine binds to serotonin transporters with a 5-fold greater potency compared with dopamine transporters, whereas amphetamines have a 10-fold lower affinity for serotonin transporters compared with dopamine transporters [41].
As serotonin has been shown to support behavioral flexibility [42][43][44], we speculate that the differences in serotonin receptor affinities may in part explain why the chronic use of cocaine and amphetamine can lead to different profiles of flexibility -a hypothesis that needs to be tested in future studies.Taken together, the behavioral profile in amphetamine/methamphetamine use disorder appears to be associated with impaired contingency learning rather than impaired behavioral flexibility.

Opioid use disorder
There is little evidence to suggest that patients with opioid use disorder have impairments with respect to cognitive flexibility, possibly because opioids are thought to be less neurotoxic than psychostimulants [45].Several studies that used the WCST [46][47][48] or the IDED task [49][50][51] did not find any impairments in cognitive flexibility in patients with opioid use disorder.
There have, however, been reports that patients on methadone maintenance therapy make more perseverative errors on the WCST compared with healthy control participants [47,52], which may reflect the pharmacological effects of methadone.Whilst cognitive flexibility appears to be largely intact in opioid use disorder, the existing evidence for behavioral inflexibility is less clear.
One study in patients with opioid use disorder who were

Inflexibility in Addictive Disorders
actively using heroin did not find any evidence for increased perseverative errors in PRL [23], but another study in a group of patients who underwent recent opioid detoxification reported increased perseverative errors on the same task [53].It is likely that the observed performance differences between these two studies could be attributed to differences in drug status, given that prior evidence suggests that the first few weeks of opioid abstinence are associated with cognitive impairments [54].The same study [53] further reported that these recently detoxified in-patients also made increased ED-shift errors, which is again inconsistent with prior work [49][50][51], but it is consistent with the notion of impairments associated with opioid withdrawal.The authors argued that their findings point toward an endophenotype for opioid addiction, but further research is warranted to clarify whether the observed deficits are an antecedent to, or a consequence of, chronic opioid use.

Alcohol use disorder
Accumulating evidence suggests that patients with alcohol use disorder have problems with cognitive flexibility, as reflected by an increased rate of perseverative errors on the WCST [55][56][57][58][59][60][61] and ED-shift errors on the IDED task [62,63] -findings that are also supported by a recent meta-analysis [64] (though see Refs.[17,[65][66][67][68] for a few exceptions).Although cognitive inflexibility has mainly been reported in patients who had been in either recent or protracted alcohol abstinence, the impairment profile appears to be similar in nonabstinent drinkers [69], suggesting that cognitive inflexibility does not recover following alcohol abstinence.By contrast, behavioral flexibility does not seem to be impaired in patients with alcohol use disorder, as most studies did not find any evidence for increased perseverative errors on the PRL task in patients but instead found that they were slow to learn the reinforcement contingencies [62,70].These observations receive further support from a study that analyzed patients' reversal learning performance using a computational modeling approach, which revealed significant impairments in updating values of unchosen choices during learning [71].

Cannabis use disorder
Compared with other addictions, relatively little work has been conducted on cannabis use disorder.However, of the limited studies available, there seems to be a trend toward cognitive inflexibility in active cannabis users, as reflected by increased perseveration errors on the WCST [69,[72][73][74][75].This impairment profile has, however, not been observed in cannabis users following prolonged drug abstinence [76], suggesting that cognitive inflexibility might reflect a cannabis-induced effect.With respect to behavioral flexibility, there is currently no evidence suggesting impairments in cannabis use disorder [77], but polydrug recreational users who were administered cannabis acutely have shown impairments in contingency learning irrespective of valence [78].Again, this finding may point toward impairments caused by the acute effects of cannabis in reducing general motivation [79], possibly by attenuating the activity of the mesolimbic system [80].

Gambling disorder
The evidence for cognitive inflexibility in patients with gambling disorder is mixed, with some studies reporting increased perseverative errors on the WCST [81,82], whereas others did not find any case-control differences [68,[83][84][85][86][87][88][89].However, one large cross-sectional study (N = 552) reported increased ED-shift errors as a function of increased gambling frequency [90], suggesting that individuals who gambled more are less cognitively flexible.A similar picture emerges with respect to behavioral flexibility; studies with small sample sizes did not identify increased perseveration errors during PRL in patients [24,70,91,92], but one study with a larger sample size (N = 148) did [93].It is thus conceivable that the null evidence from most studies was related to insufficient statistical power, so further investigations are warranted.

Concluding remarks
Addiction has repeatedly been conceptualized in dimensional terms, such that disparate addictive disorders share common abnormalities in certain aspects of neuropsychological functioning [3].Consequently, all addictive disorders and some psychiatric disorders [94] that share conceptually similar behavioral phenotypes, such as inflexible behavior, are assumed to also share similar problems with flexibility.However, the current literature does not fully support this widely held assertion but instead suggests that the profiles of cognitive and behavioral inflexibility diverge across different addictive disorders.These differences allude to the possibility that inflexible drug use (or gambling) -though appears similar on the surface -may be underpinned by different neuropsychological processes.The exact mechanisms that underpin diverging profiles of cognitive and behavioral inflexibility are a subject of ongoing research, but recent advances in computational modeling approaches may provide the means to effectively disambiguate the cognitive processes that contribute toward inflexibility in addictive disorders [95,96].
Although there is a large body of literature on flexibility in addictive disorders, there remain several avenues for future research.First, most existing tasks probe cognitive and behavioral flexibility in a non-drug-related context.Considering that drug users behave differently in drug-related contexts [97][98][99][100], future studies should consider investigating the effects of drug-related context in modulating flexibility.Second, it is tempting to speculate whether inflexibility is a cause or consequence of pathological drug use (or gambling).However, the paucity of longitudinal and endophenotype research in this area precludes any conclusive inferences on whether cognitive and behavioral inflexibility are risk factors or consequences of addictive disorders.Nevertheless, the present review, as well as others [95], reinforces the point that addictive disorders are multifaceted and are unlikely to be explained by one single theoretical framework.

Figure 1 Current
Figure 1

Table 1 (
Where possible, effect sizes (unstandardized Cohen's d) are reported in brackets in the group differences columns.IQ, intelligence quotient; DSM, Diagnostic and Statistical Manual; N/A, not applicable.a Diagnosis as reported in the publication, unless specified otherwise.b Case-control matching includes matching at the time of recruitment and statistically post hoc.

Table 2
a Diagnosis as reported in the publication, unless specified otherwise.b Case-control matching includes matching at the time of recruitment and statistically post hoc.

Table 3
a Diagnosis as reported in the publication, unless specified otherwise.b Case-control matching includes matching at the time of recruitment and statistically post hoc.