Cues in the weather prediction task.The probabilities of each possible combination of cues that appear togetherare provided in Table 2.
Percentage-correct scores over trialblocks. Shown here are the percentage-correct scores of the patients withTourette syndrome (TS) and the normal controls (NC) plotted against trialblocks in both the child (A) and the adult (B) populations.
Latency scores over trial blocks.Shown here are the latency scores of the patients with Tourette syndrome (TS)and the normal controls (NC) plotted against trial blocks in both the child(A) and the adult (B) populations.
Marsh R, Alexander GM, Packard MG, Zhu H, Wingard JC, Quackenbush G, Peterson BS. Habit Learning in Tourette SyndromeA Translational Neuroscience Approach to a Developmental Psychopathology. Arch Gen Psychiatry. 2004;61(12):1259-1268. doi:10.1001/archpsyc.61.12.1259
Copyright 2004 American Medical Association. All Rights Reserved.Applicable FARS/DFARS Restrictions Apply to Government Use.2004
The etiology of Tourette syndrome (TS) involves disturbances in the
structure and function of the basal ganglia. The basal ganglia mediate habit
To study habit learning in persons with TS.
Patients with TS were compared with normal controls in performance on
a probabilistic classification, or habit-learning task (weather prediction).
University research institute.
One hundred twenty-three children and adults, 56 with a diagnosis of
TS and 67 healthy control subjects.
Main Outcome Measures
Habit learning was assessed by the extent of improvement in accuracy
of predictions and reaction times over trial blocks during performance of
the weather prediction task. Declarative learning was assessed by performance
on 3 tasks that required intact declarative memory functioning.
Children with TS were impaired at habit learning relative to normal
controls (P = .01). This finding was replicated
in the independent sample of adults with TS (P = .01).
The rate of learning correlated inversely with the severity of tic symptoms
across both samples (r = −0.34; P = .01). Thus, impaired learning accompanied
more severe symptoms. Measures of declarative memory functioning, in contrast,
were normal in the TS groups.
Striatal learning systems are uniquely dysfunctional in both children
and adults with TS. The correlation of habit learning with symptom severity
suggests that the number and severity of tics are a function of the degree
to which the system for habit learning is dysfunctional. Thus, both the deficits
in habit learning and the tic symptoms of TS are likely to be consequences
of the previously reported anatomical and functional disturbances of the striatum
in children and adults who have TS. The existence of a well-developed animal
model for this learning system, which permits study of the neural and molecular
bases of habit learning, has important implications for the neurobiological
study of TS and for the development of new or improved therapeutics for this
Neuroimaging studies suggest that the pathophysiology of Tourette syndrome(TS) involves disturbances of the basal ganglia and related corticostriatal-thalamocorticalcircuitry.1 Findings of smaller caudate volumesin children and adults with TS suggest that the motor and cognitive functionsthe striatum subserves may be dysfunctional in persons who have this condition.2- 4 One important functionof the striatum, in particular the dorsal striatum, is the learning of skills,motor sequences, and habits, which is variously termed procedural, stimulus-response(S-R), or habit-based learning.5 The neostriatalhabit-learning system has been shown, in lesion studies of both animals andhumans, to be anatomically and functionally distinct from neural systems thatsubserve declarative learning.5- 9 Declarativelearning, which includes the learning and memory of conscious facts, experiences,and semantics, is supported primarily by medial temporal lobe structures.Habit learning, in contrast, involves the gradual and incremental learningof S-R associations, a function that is based primarily within the basal ganglia.
Probabilistic classification learning is a form of habit learning inhuman subjects that circumvents the use of declarative memory by probabilisticallyassociating cues with specific outcomes. One version of a probabilistic classificationlearning task is a weather prediction game that requires the gradual learningof S-R associations. Declarative memory of the previous trial is not as usefulin improving performance as is information gleaned across many trials. Subjectstry to predict rain or sunshine based on the presentation of a varying combinationof a set of cards on a computer screen.6,7 Eachcard is independently and probabilistically related to the outcomes (rainor shine), each of which occurs equally often. For example, one card predictssunshine 25% of the time and rain 75% of the time, whereas another card predictssunshine 57% of the time and rain 43% of the time. A response is consideredcorrect on a particular trial only if the selected outcome is more stronglyassociated with the cue combination that appears on that trial. Although subjectsreceive positive or negative feedback after each prediction, they can receivenegative feedback even when they think that they have predicted the weathercorrectly. The cue-outcome associations are not absolute because cue combinationspredict different outcomes in differing percentages. Thus, because of theprobabilistic nature of the task, subjects usually believe that they are simplyguessing at the outcome. Normal subjects do, however, exhibit learning onthis task, in that they gradually improve in their ability to predict thecorrect weather outcome, although it is outside of their conscious awareness.Patients with diseases affecting the striatum, such as Huntington diseaseand Parkinson disease, exhibit impaired learning on this task, although theyare able to answer explicit factual questions about the task.6,7 Thispattern of findings in humans is consistent with earlier studies in loweranimals indicating that the dorsal striatum subserves habit learning.5,10- 12 Conversely,patients with temporal lobe lesions that affect declarative memory systemsare impaired at answering explicit factual questions about the task,5- 7,13 whereastheir learning on the probabilistic features of the task is intact.
The motor and vocal tics in persons who have TS are typically brief,nonpurposeful or semipurposeful fragments of motor behaviors that are oftenresponses to stimuli or environmental cues either from within their bodiesor from the outside world.14 Sensitivity tothese cues is usually experienced as a compulsory urge that is only relievedby performing the tic.1,14,15 Theseurges and the patient’s preoccupation with them bear a phenomenologicalresemblance to the obsessional urges to act that typically precede compulsivebehaviors. In fact, patients with TS are often affected with comorbid obsessive-compulsivedisorder (OCD). Evidence from family-genetic and twin studies indicate thatthe disorders are genetically related,15,16 andneuroimaging studies suggest that the neural bases of TS and OCD are relatedas well.17 The phenomenological similaritiesbetween tics and compulsions and their common genetic and neural basis suggestthat they might lie on a continuum of semi-involuntary or habitual behaviors.On one end of this spectrum are those behaviors with a strong ideational componentbelonging to OCD, on the other end are those with little or no ideationalcomponent belonging to simple tics, and in between are the complex tics ofpatients with TS and comorbid OCD.15 Tics aretherefore similar in their appearance and subjective experience to habits.Parents of children with tics and OCD symptoms in fact often describe thebehaviors as habits or mannerisms. This phenomenological similarity of ticsand compulsions to habits, together with both the documented striatal abnormalitiesin TS and the role of the striatum in habit learning, has prompted othersand us to suggest that tics could represent habit learning gone awry.3,4,15,18,19 Infact, impairments in habit learning have recently been reported in a preliminarystudy of 10 children with TS.20 Thus, deficienthabit learning in persons with TS could contribute to their habit-like, stereotypedbehaviors.
We report herein a study in which we used the weather prediction taskto study habit learning in 55 children and 68 adults, both subjects with TSand healthy controls. We tested our hypothesis that children with TS woulddiffer from healthy control children in habit learning. We then sought toreplicate this finding in an independent sample of adults with TS. In exploratoryanalyses, we assessed whether measures of habit learning were associated withthe severity of tic symptoms across individuals with TS. Finally, we measuredperformance on 3 tasks that require intact functioning of declarative memorysystems to assess whether learning impairments are specific to the habit-learningsystem.
Subjects were recruited to participate in 1 or more studies of childhoodneuropsychiatric disorders. The TS sample was ascertained through the TicDisorder Clinic of the Yale Child Study Center, New Haven, Conn, and throughthe local chapter of the Tourette Syndrome Association. The unaffected controlchildren and adults were recruited from randomly selected names on a telemarketinglist of approximately 10 000 families in the local community. These familiesreceived introductory letters, which were then followed by screening and recruitmenttelephone calls. Approximately 10% of the families who were contacted ultimatelyparticipated. Control subjects were group matched with the patients by age,sex, and socioeconomic status. Those with a history of neurological illness,past seizures or history of head trauma with loss of consciousness, mentalretardation, pervasive developmental disorder, psychosis, or major depressionwere excluded. Written informed consent was obtained from adult subjects andthe parents of participating children, and assent was obtained from the children.Subjects were paid for their participation.
Neuropsychiatric diagnoses were established through clinical evaluationand administration of the Schedule for Tourette and Other Behavioral Syndromes.21 This diagnostic interview includes the Schedule forAffective Disorders and Schizophrenia for School-Age Children: Present andLifetime Version for diagnoses in children,22 theSchedule for Affective Disorders and Schizophrenia for diagnoses in adults,23 and more detailed sections on TS and OCD for bothage groups. Using all available clinical and investigational materials, 2child psychiatrists performed a best-estimate procedure to establish diagnoses.24 The Yale Global Tic Severity Scale,25 theYale-Brown Obsessive Compulsive Scale,26- 28 andthe DuPaul-Barkley attention-deficit/hyperactivity disorder (ADHD) ratingscale29 were used, respectively, to obtainratings of current and worst ever severity of tic, OCD, and ADHD symptoms.Intraclass correlation coefficients30 for clinicianswho administered the Yale Global Tic Severity Scale and the Yale-Brown ObsessiveCompulsive Scale were greater than 0.90 on videotaped training interviews.Estimates of full-scale IQs were made using the Wechsler Abbreviated Scaleof Intelligence.31 Socioeconomic status wasquantified using the Hollingshead Four-Factor Index of Social Status.32 Sample characteristics are given in Table 1.
This measure of probabilistic learning was administered on a laptopcomputer (Macintosh iBook). The task required subjects to learn which of 2outcomes, rain or sunshine, would occur on each trial based on 1, 2, 3, or4 different cues (Figure 1) that occurredon each presentation in 1 of 14 possible combinations (Table 2). The sequence of cue combinations appearing on each trialwas randomized for each participant, with the constraints that the same cuecombination could not appear twice in succession and that each outcome didnot occur more than 5 times in succession.
Each cue was independently associated to each outcome (rain or sunshine)with a fixed probability, and the 2 outcomes occurred equally often. Acrossall 14 cue combinations, each cue-outcome association occurred at a consistentfrequency; ie, cue 1 was associated 26.2% of the time with sun and 73.8% ofthe time with rain, cue 2 was associated 44.4% of the time with sun and 55.6%of the time with rain, cue 3 was associated 55.6% of the time with sun and44.4% of the time with rain, and cue 4 was associated 73.2% of the time withsun and 26.8% of the time with rain. These percentages were calculated from Table 2 by adding, for example, the number oftimes that cue 1 was associated with sun (column G) and dividing that numberby the total number of times that cue 1 appeared (11/42 = 26.2%).The probability structure of this task (determined by the cue-outcome associationstrengths and cue patterns) was more difficult than those used in other studiesto minimize the likelihood that subjects would gain conscious, declarativeknowledge of the S-R contingencies in this task.6,7,13,20
Subjects were asked to read the instructions on the computer screenand to look up when finished. These instructions explained that they wouldbe seeing 1 to 3 cues on each trial and their task would be to decide whetherthe cues predicted sunshine or rain. The experimenter informed the subjectsthat predicting the outcome would feel like guessing but that their performancewould gradually improve. On each trial, 1, 2, or 3 of the 4 cues appearedvertically on the computer screen (in 1 of the 14 possible combinations),and subjects predicted sunshine or rain by pressing either the G or the Hkey, respectively. To minimize confusion, the G key was covered with a stickerof a sun and the H key was covered with a sticker of a rain cloud. Feedbackwas provided immediately to signal a correct or incorrect response. For eachtrial, correct responses were followed by the appearance of a smiling facealong with the sound of a bell. Incorrect responses were followed by a frowningface and the sound of a groan. The task consisted of 90 trials with a short,1-minute break after the first 40 trials.
A response was considered correct on a particular trial if the outcomeselected was more strongly associated with the cue combination that appearedon that trial. Because of the probabilistic nature of the task, a cue combinationwas sometimes followed by the less strongly associated outcome. Thus, subjectscould have been scored as making a correct response (because they selectedthe most likely outcome) even though the feedback they received on that particulartrial suggested to them that their response was incorrect. In this way, thepercentage-correct score reflected how well subjects learned the cue-outcomeassociations. Because the 2 outcomes occurred equally often, chance performancewas 50% correct. The data were not analyzed for trials on which the 2 outcomeswere equally associated with the cue combination and on which there was thereforeno correct answer (combination 6) (Table 2).Percentage-correct and latency scores were analyzed by averaging across 10trials in each of 9 successive blocks.
Each subject was administered a 5-item, 4-alternative multiple-choicequestionnaire that explicitly inquired about the nature of the cues and feedback,the layout of the screen, and the testing procedure. Proportion-correct scoreswere derived for each subject.
This measure consisted of 1 stimulus card and 2 response cards, eachdepicting a spatial array of common objects. The stimulus card depicted anarray of 27 common objects, including nature items (ie, cat, elephant, bird,and flower), household items (ie, telephone, iron, teapot, and plant), andwork-related items (ie, briefcase, hat, and cane). The first response cardmeasuring memory for object identities depicted an array of the original 27objects in their original locations and 20 added objects. The second responsecard measuring memory for object location depicted the original 27 objects,with an exchange between positions of 7 pairs of objects. For the purposesof this study, we focused on the first response card. Subjects were presentedwith the stimulus card for 1 minute and were given the explicit instructionto try to remember the objects on the page because they would be asked toidentify them later. After 1 minute, the stimulus card was removed from sightand subjects were presented with the first response card. The response cardwas removed after they marked the added objects. Proportion-correct scoreswere derived for each subject. Errors included total omissions (false negatives)and total commissions (false positives); thus, scores were defined as 1−(omissions + commissions)/totalnumber of objects.
The Wechsler Abbreviated Scale of Intelligence31 wasused to estimate full-scale IQ in the children and adults. It is a short andreliable measure of intelligence, consisting of 4 subtests: vocabulary, blockdesign, similarities, and matrix reasoning. It was considered an indirectmeasure of declarative memory functioning because each subtest requires anintact, flexibly accessible, and relational memory for the facts learned,and these are the defining characteristics of declarative memory functions.8,34 Administration of the 4 subtests yieldsverbal, performance, and full-scale IQ scores for each subject. The full-scalescores were used in our statistical analyses.
The following statistical procedures were performed in SAS version 8.0(SAS Institute Inc, Cary, NC). A priori hypotheses were tested using mixedmodels analysis (PROC MIXED) with repeated measures over blocks of trials.Percentage-correct scores were entered as the dependent variable in a linearmixed model. Analyses were performed first in children (age, <18 years)and then in adults (age, 18-59 years) in an attempt to replicate findingsfrom the children. For both models, diagnosis (TS or healthy control) wasa between-subjects factor. Covariates included age, sex, and current diagnosisof OCD or ADHD. We entered block as an ordinal variable from 1 to 9 to ensurethat performance was modeled as a linear trend across blocks. We also consideredfor inclusion in the model all 2- and 3-way interactions of diagnoses (TS,OCD, and ADHD), age, sex, and block. Terms that were not statistically significantwere eliminated via backward stepwise regression, with the constraint thatthe models had to be hierarchically well formulated at each step (ie, allpossible lower-order terms had to be included in the model, regardless oftheir statistical significance).35 Habit learningfor each subject in the weather prediction task was quantified by analyzingthe linear trend in performance across trial blocks. The hypothesized differencein habit learning between the patients with TS and the healthy control subjectswas tested by assessing statistical significance of the diagnosis-by-blockinteraction. Because we tested this term separately in children and adults,we protected against false positives associated with these 2 comparisons byconsidering P values <.025 to be statisticallysignificant in a priori testing.
Subjects who demonstrate learning on habit-learning tasks should becomeprogressively faster at responding (ie, they should have shorter latency scores)over blocks of trials. With latency defined as the response time for eachtrial, we compared the changes in latency scores for the TS and control groupsby assessing the significance of the diagnosis-by-block interaction.
The associations of learning with the severity of symptoms, and of learningwith IQ estimates, were assessed in correlation analyses. To evaluate theseassociations, we first quantified learning across the 9 blocks of trials bymodeling the percentage-correct scores across blocks as linear trends withineach subject. Learning for each subject was thus characterized by the valuesof the coefficients of these trends in a linear model (ie, larger coefficientsindicated better learning).
The effects of comorbid illnesses and medication use on the findingsof a priori hypothesis testing were assessed in 2 complementary ways. First,their effects were assessed as statistical covariates in our final model forhypothesis testing, both as main effects and as interactions with trial block.Second, the stability of findings in the absence of these effects was assessedby examining the stability of parameter estimates for the effect of TS onlearning in separate models that included either subjects with pure TS (ie,without OCD or ADHD) or subjects with TS who were not taking any medication.Medication effects were assessed separately for any medication use and foruse of traditional neuroleptics (haloperidol or pimozide), risperidone, α-agonists(clonidine or guanfacine), or selective serotonin reuptake inhibitors.
Linear regression was used to compare TS and control subject scoreson the weather task questionnaire. Independent t testscompared IQ scores and scores from the Silverman and Eals memory task acrossgroups.
Data were acquired from 55 children (32 subjects with TS and 23 normalcontrol subjects) and 68 adults (24 subjects with TS and 44 normal controlsubjects). Among the children, the TS and control groups were of comparableages (mean ± SD, 12.38 ± 2.74 vs 12.65 ± 3.16years; t66 = −0.34; P = .73) and had similar proportions of boysand girls (χ2 = 1.02; P = .31).In the adult sample, the groups were also similarly aged (mean ± SD,35.28 ± 11.29 vs 31.68 ± 12.10 years; t66 = −0.03, P = .97) and had a similar sex composition (χ2 = 0.26; P = .61).
Among the 32 child subjects with TS, current diagnoses included OCDin 7 (21.8%), ADHD in 7 (21.8%), both OCD and ADHD in 6 (18.7%), depressedmood in 3 (9.3%), and oppositional defiant disorder in 8 (25%). In the 24adults with TS, current diagnoses included OCD in 6 (25%), ADHD in 2 (8%),and both OCD and ADHD in 4 (16.6%). At the time of the study, 39 (70%) ofthe subjects with TS (24 children and 15 adults) were taking medications.These included stimulants (n = 1 [1.7%]), traditional neurolepticagents (haloperidol or pimozide; n = 3 [5.3%]), risperidone (n = 5[8.9%]), α-adrenergic agonists (clonidine or guanfacine; n = 10[17.8%]), and selective serotonin reuptake inhibitors (n = 20 [35.7%]).
Analysis of the linear trend only in children revealed a significantdiagnosis × block interaction (F1,430 = 6.87; P<.01; effect size = −2.62).A plot of performance across trial blocks indicated that this effect derivedfrom a relative impairment in learning in children with TS compared with controls(Figure 2A).
A significant diagnosis × block interaction (F1,540 = 5.58; P<.02; effect size = 2.36) was confirmed in the adultsample. Similar to findings in the children, this effect derived from impairedlearning in the TS group (Figure 2B).
Because technical complications interfered with the recording of responsetimes in 29 subjects, this analysis was performed for 94 of the 123 participatingsubjects. Modeling latency scores also revealed significant diag nosis × blockinteractions in both children (F1,320 = 2.16; P<.05) and adults (F8,352 = 2.03; P<.05). Plots of meanlatency scores against trial blocks indicated in each age group that controlsubjects responded progressively faster over subsequent blocks of trials thandid subjects with TS, further suggesting that children and adults with TSwere impaired in learning the task (Figure 3).
Pearson correlation coefficients were calculated to assess the associationof tic severity in the TS group, measured with the Yale Global Tic SeverityScale score for combined motor and vocal tics, with the coefficients usedto define learning. These correlations revealed a significant inverse association(r = –0.34; P = .01)of tic severity and learning, indicating that those who had more severe ticsymptoms had the greatest deficits in habit learning. In contrast, IQ scoreswere not significantly associated with learning scores among the TS group(r = 0.03; P = .89),suggesting that any individual and group differences in IQ were not confoundingresults of other analyses.
Attention-deficit/hyperactivity disorder and OCD as comorbid illnesseswere not significantly associated with learning either as main effects (P>.13 and P>.44, respectively)or as interactions (P>.47 and P>.90, respectively) in either the children or the adults with TS.These results did not change when the actual symptom severity scores fromthe Yale-Brown Obsessive Compulsive and the DuPaul-Barkley ADHD rating scaleswere used as covariates in the models (P>.31 and P>.12 for ADHD and OCD, respectively). To assess furtherthe effects of comorbid illnesses on our findings, we examined the stabilityof parameter estimates and P values for the effectsof TS in the models that included or excluded all subjects with comorbid OCDor ADHD. The values of these parameters did not change appreciably. In addition,correlation analyses revealed that OCD and ADHD symptom severity was not associatedwith habit learning in the patients diagnosed as having comorbid OCD (r = 0.08; P = .78)or ADHD (r = −0.27; P = .48). Together these analyses confirmed that impairedlearning in the TS group was not caused by the presence of comorbid OCD orADHD. In addition, current oppositional defiant disorder as a comorbid illnesswas not significantly associated with learning as a main effect (P>.66) or as an interaction (P>.93), indicatingthat its presence in children with TS was also not the cause of their impairedhabit learning.
Habit learning differed significantly in those patients with TS whowere taking medications compared with those who were not taking medications(main effect for medication, F1,48 = 8.49; P<.001). This effect was caused by the difference inperformance between the nonmedicated patients and those taking neuroleptics(t33 = 3.24; P = .002). No significant associations with habit learningwere noted with the use of α-adrenergic agonists or selective serotoninreuptake inhibitors. Comparing the performance of the nonmedicated patientswith that of the healthy controls still produced a significant diagnosis × blockinteraction (F1,91 = 2.60; P = .008), indicating that medication effectswere not responsible for findings of a priori hypothesis testing. Comparingthe medicated patients with the healthy controls did not yield a significantdiagnosis × block interaction (P = .61),indicating that performance of the medicated patients was similar to thatof the healthy controls. Furthermore, partial correlation analyses in whichmedication was a covariate still revealed a significant inverse associationof tic severity scores with learning coefficients (β = −.39; P = .003), indicating that patients with moresevere symptoms had the greatest deficits in habit learning, regardless ofmedication use.
A linear regression model comparing TS patients with controls on theweather task questionnaire revealed a trend toward poorer performance in theTS group (mean ± SD, 70.0% ± 2.6% vs 77.8% ± 4.4%; F1,82 = 3.31; P = .08). However, this trend derived from the effects ofcomorbid OCD or ADHD in that the trend was absent when individuals with thesecomorbidities were excluded from the analyses (t83 = 1.03; P = .31). Independent-samples t tests comparing the TS and normal control subjects revealedno significant differences between groups on the Silverman and Eals LocationMemory Task in either the children (t85 = –0.65; P = .51) or the adults (t77 = –0.98; P = .33).Similarly, t tests revealed no significant differencesacross groups on IQ (children: t56 = −0.63; P = .53; adults: t52 = −1.61; P = .11),which remained true when the patients with the comorbid illnesses of OCD andADHD were eliminated from the analyses (children: t45 = –0.03; P = .98;adults: t32 = –1.63; P = .11). Taken together, these analyses suggestthat adults and children with TS in general perform normally on declarativememory tasks.
Both children and adults who have TS were impaired in a probabilisticclassification learning task that has previously been shown to depend on thefunctional integrity of the neostriatal system for habit learning.6,7 Compared with healthy control subjects,patients with TS did not improve in task performance, measured either by improvedprediction accuracy or by improved reaction times over trial blocks. Learningwas significantly and inversely associated with the severity of tic symptoms,indicating that subjects with TS who had more severe symptoms were proportionatelymore impaired in habit learning. Although the association of diagnosis withimpaired habit learning does not prove that impairments in striatally basedhabit learning cause tics, the association of symptom severity with learningscores does provide strong circumstantial evidence that impaired habit learningis centrally involved in the pathophysiology of TS. Patients with TS werenot impaired on measures requiring intact functioning of declarative memory,a system that requires the structural and functional integrity of the hippocampusand other medial temporal lobe structures. These findings suggest that deficitsin memory functioning in patients with TS are relatively specific to the striatallearning system.
Previous habit-learning studies have employed transfer tests, such asquestionnaires, as indexes of declarative memory functioning. Results of thosestudies have indicated that patients with Parkinson disease, Huntington disease,and TS lack awareness of the algorithm learned in the weather prediction task,but their explicit memory of the testing situation remains intact.6,7,13 Few studies, however,have systematically and comprehensively assessed declarative memory functionsin individuals with TS. One large study of adolescents with TS did reportnormal performance in declarative memory functions assessed with the AdultMemory and Information Processing Battery of story recall, the Rey AuditoryVerbal Learning Test, and the Visual Reproduction Test of the Wechsler MemoryScale.36 More extensive and detailed assessmentsof declarative memory functions are needed to determine whether the learningdeficits observed among patients with TS are indeed specific to striatallybased systems for habit learning.
Our findings replicate and expand on a previous preliminary report ofimpaired habit learning in 10 children with TS.20 Ourswas a much larger sample than that of the previous study, and it includedadults as well as children. We found that habit learning was deficient insubjects with TS from both age groups, suggesting that impairments in habitlearning are not likely to reflect simply the presence of an immature cognitiveprocessing skill that improves later in life. In addition, we used a moredifficult probability structure in the task than has been used in previousstudies to minimize the likelihood that declarative memory functions couldbe used to improve performance in later trials.6,7,13,20
Our behavioral findings provide a useful conceptual framework for understandingthe anatomical and functional abnormalities of the basal ganglia in TS thathave been reported previously.2- 4,18 Deficitsin habit learning, for example, seem likely to be a functional consequenceof the reduced caudate nucleus volumes previously reported in children andadults with TS.4,37 They are alsoconsistent with findings from a functional imaging study that demonstrateddisturbances in subcortical activity during the voluntary control of tics,disturbances that were directly proportional to the severity of tic symptomsmeasured clinically outside of the scanner.38 Basedon regional patterns of activation and the known connectivity within corticostriatal-thalamocorticalcircuits, this functional imaging study concluded that the disturbances incontrolling tics likely originated in or around the caudate nucleus.
The associations of learning with both diagnosis and symptom severitypersisted even when excluding medicated subjects from the analyses, suggestingthat medication use did not contribute to our main findings. The use of neurolepticagents, however, was associated with better habit learning in the subjectswith TS who were taking them. This finding suggests that tic medications mayimprove not only tic symptoms but also habit learning. Study of habit learningbefore and after initiation of medication use is warranted in future clinicaltrials to help clarify whether neuroleptic use can improve habit learningin persons with TS. We found no evidence that the presence of comorbid OCDor ADHD affected our findings for habit learning, assessed either throughstatistical covariation for these diagnoses or through the analysis of subgroupsof patients with TS who did not have these comorbid illnesses.
The mediation of habit learning by the basal ganglia and the independenceof habit learning from declarative learning and memory functions based withinmedial temporal lobe structures have been demonstrated in both animal5,10- 12 andhuman studies.6,7,13 Animalstudies have shown, for example, that electrolytic or neurochemical lesionsof the dorsal striatum impair performance of habit-learning tasks but notdeclarative memory tasks, whereas lesions of the hippocampal system impairperformance of declarative memory tasks but not habit-learning tasks.11,39 Patients with temporal lobe amnesiawhose performance is impaired in declarative memory tasks tend to learn normallyin probabilistic classification tasks,6,7,13 whereaspatients with Huntington and Parkinson diseases who perform normally on testsof declarative memory are impaired at probabilistic classification learningtasks.5,6 Consistent with theimplication from these behavioral findings that habit learning is based withinthe dorsal striatum, one human functional imaging study has demonstrated increasedneuronal activity in the striatum and reduced activity in the hippocampusduring habit learning in the weather prediction task.40 Inaddition, other human imaging studies of similar tasks of skill learning havealso shown activation of the caudate nucleus, confirming the mediation ofthese tasks by the striatum.41,42
Changes in activity of dopaminergic neurons within the striatum areknown to contribute to learning in response to reward. Striatal dopamine neuronsare sensitive to prediction errors (the discrepancies between outcomes andtheir predictions). Their firing increases in the presence of unpredictedrewards and decreases in the absence of predicted rewards. These neurons donot fire when rewards occur as predicted, consistent with learning theoriesin which learning is induced when rewards or reinforcers occur more oftenthan predicted and in which learning is extinguished in the absence of reward.43,44 Dopamine systems therefore contributeto learning most powerfully when reward contingencies change, ie, when rewardsare unpredictable. The weather task requires subjects to make predictionsbased on cues presented in the current trial and on the feedback, or reward,received after previous trials. However, rewarding feedback is predicatedon probabilistic associations with the stimulus cue and not always consistentwith the correct prediction or feedback anticipated with the current stimulus.Thus, the rewards in this task are unpredictable, and this unpredictabilityshould provide an important role for dopaminergic firing within the striatumduring learning of this task.
Electrophysiological recordings in animals have also elaborated thisunderstanding of striatally based learning. They have shown, for example,that habit learning of complex action sequences is associated with gradualchanges in the task-related firing of neural ensembles within the striatum.45 The new firing patterns stabilize during subsequentperformance and consolidation of the learned action sequence. The striatumthus develops a neuronal representation of the action pattern or sequenceof movements. Firing is most frequent at the beginning and the end of theaction sequence. This change in firing pattern is thought to indicate thateach of the behavioral fragments composing the entire action sequence arechunked together within the striatum into a single, coherently executed behavior.Once a chunked action sequence is activated, it tends to execute smoothlyand in entirety. Intact dopaminergic innervation seems to be important forthese chunking functions of the striatum.46 Thelong-term acquisition of memories for habit-learning tasks, similar to theacquisition of all long-term memories, presumably involves the alterationof cellular architecture within the striatum. This may include the modificationof dendritic spines, dendritic arborization, or synaptic remodeling, processesthat require modification of gene expression and protein production withinthe cell.47,48 Whether these presumedultrastructural correlates of learning are visible macroscopically in thestriatum is unknown.
Our finding that habit learning is impaired in children with TS andthe replication of these findings in adults with TS suggest that the chunkingtogether of action sequences is dysfunctional in persons of all ages who haveTS. It has long been noted that the sudden, repetitive, and jerking movementsthat constitute tics appear to be fragments of normal motor and vocal behavioralrepertoires.1,17,49 Thepresence of impaired habit learning in persons with TS suggests that thesebehavioral fragments are not concatenated together properly but instead occurin isolation and independently of normal S-R contingencies.50 Trait-likeabnormalities previously documented in the structure and function of the striatumin persons with TS2- 4 mayimpair the concatenation or chunking of these behavioral fragments into coherentaction sequences that are executed smoothly as habits. The disturbances inhabit learning may not produce the cellular changes within the striatum thatsupport long-term learning; deficient habit learning might thereby producethe macroscopic hypoplasia of the caudate nucleus that has been observed invivo in the brains of children and adults with TS.4
In addition, large, controlled imaging studies of individuals with TSsuggest that additional disturbances in frontostriatal projections may releasefrom regulatory control this trait-like predisposition for behavioral fragmentationthat is based within the striatum.4,51,52 Thus,the tics of TS seem to be the product of core disturbances in the structureand function of the striatum that predispose an individual to impairmentsin habit learning and to the expression of fragmented motor and vocal behaviors.These predispositions to tic behaviors may then be released from regulatoryinfluences of the prefrontal cortex.52,53
Medication use, particularly the long-term use of dopamine receptorantagonists,1,51 was associatedwith better performance on the habit-learning task. In apparent contrast,acute posttraining peripheral54- 57 andintradorsal striatal11,58 administrationof dopamine agonists has been shown to enhance habit-memory formation in rats.The effects of long-term administration of dopaminergic agents, however, oftendiffer from the effects of acute administration.59 Long-termuse of dopamine receptor blockers, for example, decreases the overall firingrate of dopamine neurons60 while increasingburst firing in response to prediction errors. Decreased background firingand increased burst firing associated with long-term dopamine blockade maytogether increase the signal-to-noise ratio of the information carried tothe dorsal striatum by bursting dopaminergic neurons during reward-based learning.Thus, we speculate that the long-term use of dopamine antagonists may havecontributed to relatively better performance in subjects with TS by enhancingthe salience of dopamine bursting as these subjects learned the task. Thisinterpretation is perhaps consistent with previously reported clinical andexperimental evidence for disturbances of dopaminergic transmission in thestriatum of individuals with TS,1,52,61- 66 andit further suggests that the impaired habit learning detected in our patientswith TS may have its basis in dysfunction of the nigrostriatal dopamine system.Consistent with this suggestion, 6-hydroxydopamine lesions of the nigrostriatalpathway impair learning in lower animals.67,68 Alternatively,the observed medication effect could have arisen from normal dopaminergicneurons interacting with other abnormal striatal tissues.
The findings from this study have important implications for our understandingof the behavioral basis and pathophysiology of TS in children and adults.They also help us to understand better the role of the basal ganglia and frontostriatalcircuits in habit learning. This line of work is the first to implicate directlyin TS a deficit in a specific and discrete cognitive-behavioral system, theneostriatum-based habit-learning system. Furthermore, animal models permitstudy of the neural and molecular bases of habit learning. By extension, thesemodels should permit study of the neural and molecular basis of the tic-likebehavioral fragments that fail to enter the unified action sequences of S-Ror habit learning. The potential availability of both animal models and humanparadigms for studying habit learning therefore offers the exciting promisenot only of improving our knowledge of the neurobiological origins of TS butalso of developing novel therapeutics through bona fide translational researchprograms and methods that are not available to human clinical studies alone.Future research should evaluate the effects of medication and behavioral interventions,such as dopamine blockers and habit-reversal therapy,69 onhabit learning in randomized clinical trials.
Correspondence: Rachel Marsh, PhD, ColumbiaUniversity and New York State Psychiatric Institute, 1051 Riverside Dr, Unit74, New York, NY 10032 (firstname.lastname@example.org).
Accepted for Publication: June 11, 2004.
Funding/Support: This study was supported inpart by grants MH01232, MH59139, and MH068318 from the National Instituteof Mental Health, Bethesda, Md; the Suzanne Crosby Murphy Endowment at ColumbiaUniversity College of Physicians and Surgeons, New York, NY; and the ThomasD. Klingenstein & Nancy D. Perlman Family Fund, New York, NY.
Acknowledgment: We are grateful to James Leckman,MD, Robert King, MD, Larry Scahill, PhD, and Diane Findley, PhD, as well asthe Tourette Syndrome Association, for their help with patient recruitment.