Distinct Neural Correlates of Washing, Checking, and Hoarding SymptomDimensions in Obsessive-compulsive Disorder

Context  Obsessive-compulsive disorder (OCD) is clinically heterogeneous, yet most previous functional neuroimaging studies grouped together patients with mixed symptoms, thus potentially reducing the power and obscuring the findings of such studies.

Objective  To investigate the neural correlates of washing, checking, and hoarding symptom dimensions in OCD.

Design  Symptom provocation paradigm, functional magnetic resonance imaging, block design, and nonparametric brain mapping analyses.

Setting  University hospital.

Participants  Sixteen patients with OCD (11 inpatients, 5 outpatients) with mixed symptoms and 17 healthy volunteers of both sexes.

Intervention  All subjects participated in 4 functional magnetic resonance imaging experiments. They were scanned while viewing alternating blocks of emotional (washing-related, checking-related, hoarding-related, or aversive, symptom-unrelated) and neutral pictures, and imagining scenarios related to the content of each picture type.

Main Outcome Measure  Blood oxygenation level–dependent response.

Results  Both patients and control subjects experienced increased subjective anxiety during symptom provocation (patients significantly more so) and activated neural regions previously linked to OCD. Analyses of covariance, controlling for depression, showed a distinct pattern of activation associated with each symptom dimension. Patients demonstrated significantly greater activation than controls in bilateral ventromedial prefrontal regions and right caudate nucleus (washing); putamen/globus pallidus, thalamus, and dorsal cortical areas (checking); left precentral gyrus and right orbitofrontal cortex (hoarding); and left occipitotemporal regions (aversive, symptom-unrelated). These results were further supported by correlation analyses within patients, which showed highly specific positive associations between subjective anxiety, questionnaire scores, and neural response in each experiment. There were no consistently significant differences between patients with (n = 9) and without (n = 7) comorbid diagnoses.

Conclusions  The findings suggest that different obsessive-compulsive symptom dimensions are mediated by relatively distinct components of frontostriatothalamic circuits implicated in cognitive and emotion processing. Obsessive-compulsive disorder may be best conceptualized as a spectrum of multiple, potentially overlapping syndromes rather than a unitary nosologic entity.

Obsessive-compulsive disorder (OCD) is clinically heterogeneous. Factor-analyticstudies have consistently identified at least 4 temporally stable symptomdimensions: contamination/washing, aggressive/checking, hoarding, and symmetry/ordering.^1-7 Thesesymptom dimensions have been related to different patterns of genetic transmission,^8-10 comorbidity,^1-3,11-13 andtreatment response.^3,5,14-17

Despite this heterogeneity, most previous neuroimaging studies of OCDhave grouped together patients with mixed symptoms,^18-23 thuspotentially reducing their power and obscuring their findings. Few studieshave examined the neural correlates of different symptom dimensions. Rauchet al²⁴ found that checking symptoms correlatedwith increased, and symmetry/ordering with reduced, regional cerebral bloodflow in the striatum, while washing symptoms correlated with increased regionalcerebral blood flow in bilateral anterior cingulate and left orbitofrontalcortex. Using functional magnetic resonance imaging (fMRI), Phillips et al²⁵ compared OCD patients with mainly washing or checkingsymptoms while they viewed generally aversive or washing-related pictures.When viewing washing-related pictures, only washers demonstrated activationsin regions implicated in emotion and disgust perception, ie, visual regionsand the anterior insula^26-28;checkers demonstrated activations in frontostriatal regions and the thalamus.In another fMRI study, patients with OCD with predominantly washing symptomsdemonstrated greater activation than controls in the right insula, ventrolateralprefrontal cortex, and parahippocampal gyrus when viewing disgust-inducingpictures.²⁹ Limitations of these studies includedthe artificial division between washers and checkers and the exclusive useof washing-related material, but taken together, they suggest that differentsymptoms may be mediated by distinct neural systems and that previous discrepantfindings may result from phenotypic variations in the studied samples.

Building on recent pilot work from our group,³⁰ weused a symptom provocation paradigm to examine, within thesame patients, the neural correlates of washing, checking, and hoardingsymptom dimensions of OCD. This dimensional approach is methodologically superiorto categorically dividing patients into mutually exclusive subgroups becausemonosymptomatic patients are infrequent and such a division is therefore artificial.On the basis of previous studies, we hypothesized that (1) anxiety would beprovoked in response to all types of emotional material both in patients andin controls but more so in patients^22,30,31;(2) symptom provocation would activate regions previously implicated in OCD,^18-23 bothin patients and in controls,³⁰ but more soin patients; and (3) distinct patterns of neural response would be associatedwith the provocation of each symptom type. Specifically, in patients comparedwith controls, the provocation of (a) washing-relatedanxiety would predominantly activate areas involved in emotion and disgustperception, ie, ventromedial prefrontal and paralimbic regions^24,25,29,30;(b) checking-related anxiety, regions involved inattentional and motor functions, ie, dorsolateral prefrontal cortex, thalamus,and striatal regions^24,25,30;and (c) hoarding-related anxiety, ventromedial prefrontaland paralimbic regions.³⁰ Finally, we expectedthat, in patients, the magnitude of activation in the predicted regions withineach experiment would be significantly correlated with the corresponding subjectiveanxiety and/or symptom dimension scores.

Seventeen patients with OCD (11 inpatients, 6 outpatients) who wereat various stages of treatment were recruited from 2 specialized cognitivebehavioral therapy clinics in London, England. This was a consecutive sample,but an effort was made to ensure that washing, checking, and hoarding symptomswere sufficiently represented. One outpatient reported having closed his eyesin the scanner and was excluded. Axis I and II diagnoses were made accordingto DSM-IV.^32,33 Patientswith comorbid diagnoses were not excluded provided that OCD was the main problemfor which treatment was sought. Exclusion criteria were brain injury, anyneurologic condition, psychosis, and substance abuse.

The patients' mean illness duration was 14.2 years (SD, 8.3 years; range,1.5-29 years). The OCD severity was moderate to severe (Yale-Brown Obsessive-CompulsiveScale total: mean, 24.7; SD, 7.8; obsessions: mean, 11.6; SD, 4.6; compulsions:mean, 13.1; SD, 3.6). Nine patients (56%) had 1 or more comorbid Axis I orAxis II disorders. Additional Axis I diagnoses were major depressive disorder(n = 6), social phobia (n = 3), specific phobia (n = 2), and panic disorder,agoraphobia without panic, posttraumatic stress disorder, generalized anxietydisorder, and body dysmorphic disorder (each, n = 1). Comorbid personalitydisorders were obsessive-compulsive (n = 7), avoidant (n = 6), depressive(n = 5), dependent (n = 3), paranoid (n = 2), borderline (n = 2), and narcissistic(n = 1). Most patients (n = 12; 75%) were taking medication at the time ofthe study: clomipramine hydrochloride (4 patients; 175 mg), fluoxetine hydrochloride(3 patients; 20 mg), paroxetine hydrochloride (3 patients; 40 mg), and venlafaxinehydrochloride (2 patients; 200 mg). Additional medications included buspironehydrochloride (3 patients; 12 mg), lithium carbonate (1 patient; 800 mg),valproate sodium (1 patient; 600 mg), and chlorpromazine hydrochloride (1patient; 25 mg).

Seventeen healthy volunteers of similar demographic characteristicswere recruited among ancillary staff at the Institute of Psychiatry. Theyreported no history of neurologic or psychiatric disorder and were unmedicated.Data from 10 of these control subjects were partially reported elsewhere.³⁰ The Ethics Committee of the Maudsley Hospital/Instituteof Psychiatry, London, approved the study protocol, and all subjects signedan informed consent form before their participation.

In the OCD group, severity and types of OCD symptoms were assessed withthe Yale-Brown Obsessive-Compulsive Scale and the Symptom Checklist.^34,35

In both groups, symptom dimension scores were obtained with the PaduaInventory–Revised (PI-R).³⁶ We were interestedin 2 of its subscales, "washing" (10 items; score range, 0-40) and "checking"(7 items; score range, 0-28), which are particularly reliable and valid.^36-41 Hoardingsymptoms were assessed with the Saving Inventory–Revised (SI-R).^42,43 The SI-R is a reliable and validinstrument consisting of 23 self-administered items requesting a responseon a 0 to 4 scale (score range, 0-92). Factor analysis identified 3 robustsubscales: clutter (9 items; score range, 0-36), difficulty discarding (7items; score range, 0-28), and acquisition (7 items; score range, 0-28).⁴³

Depression was assessed with the Beck Depression Inventory (BDI).⁴⁴ The state subscale of the State-Trait Anxiety Inventory⁴⁵ was administered immediately before the scan.

Fifty color pictures of scenes rated as aversive or disgusting by normalsubjects (eg, insects, mutilated bodies, decaying food) and 50 pictures ofneutral scenes (eg, furniture, nature scenes, household items) were selectedfrom a standard set of stimuli.⁴⁶ These stimuliwere carefully chosen to avoid resembling common triggers of OCD symptoms.In addition, pictures depicting contamination/washing, aggressive/checking,and hoarding material (50 of each type) were obtained with a standard digitalcamera. For each symptom type, 3 clinicians with experience in OCD had previouslylisted the most common items that were provocative of anxiety and the urgeto ritualize in patients with OCD. Examples of the pictures are public telephoneor toilet, money, syringe, and ashtray (washing); electric appliances, stove,open door, and purse (checking); and old newspapers or magazines, old clothesor toys, empty bottles or cans, and trash bins (hoarding).

A total of 250 scenes were selected after an independent group of 9normal volunteers (unrelated to the study) had rated an originally largerpool of pictures according to their level of visual complexity, anxiety, anddisgust on a 0 to 3 scale (0, none; 3, high). Pictures that were too simpleor too complex were excluded, and an effort was made to avoid using washing-relatedpictures that could be perceived as very aversive by normal individuals. Thefinal 250 stimuli were well matched regarding visual complexity and, as intended,the normally aversive or disgusting pictures induced more anxiety and disgustthan the other 3 types of pictures (data not shown).

All subjects participated in four 6-minute experiments in which theyviewed ten 20-second alternating blocks of emotional (washing, checking, orhoarding related or normally aversive) and neutral pictures. The order inwhich the 4 experiments were conducted was fully counterbalanced, as was theorder of the emotional and neutral conditions within each experiment. Moredetails can be found in Mataix-Cols et al.³⁰

Before the presentation of each set of pictures, subjects were playeda prerecorded voice file by means of high-fidelity pneumatic headphones, instructingthem to imagine being in a particular situation while looking at the scenesthey were about to see. Examples of these instructions are as follows: "Imaginethat you must come into contact with what's shown in the following pictureswithout washing yourself afterwards" (washing); "Imagine that you are notsure whether you switched off or locked the following objects and it is impossiblefor you to go back and check" (checking); "Imagine that the following objectsbelong to you and that you must throw them away forever" (hoarding); "Imaginethat you must touch or stand by the following objects" (aversive); "Imaginethat you are completely relaxed while looking at the following scenes" (neutral).

After each set of pictures, another prerecorded sound file of the question"How anxious do you feel?" was played and the subjects rated their subjectiveanxiety on a 0 (no anxiety) to 8 (extreme anxiety) scale.

Gradient-echo echoplanar images were acquired on a 1.5-T MRI system(GE Signa Neuro-optimized MR system; General Electric, Milwaukee, Wis) atthe Maudsley Hospital. One hundred T2*-weighted whole-brain volumes depictingblood oxygen level–dependent contrast⁴⁷ andconsisting of 16 sections oriented according to the bicomissural plane (thickness,7 mm; 0.7-mm gap) were acquired during 6 minutes for each of the 4 experiments(repetition time, 2.0 seconds; echo time, 40 milliseconds; field of view,24 cm; flip angle, 70; 64 × 64 matrix). This echoplanar image data setprovided almost complete brain coverage.

In each 20-second stimulus presentation block, subjects viewed either10 provocative or 10 neutral pictures. Each picture was presented for 1950milliseconds, with an interstimulus interval of 50 milliseconds. Ten whole-brainvolumes were acquired during each stimulus presentation block. Each stimulusblock was followed by (1) an 8-second period of complete silence during whichsubjects were asked to rate their level of anxiety and (2) an additional 8-secondperiod during which the subjects listened to a sound file containing instructionspertinent to the next stimulus block. Four "dummy volumes" were excited duringthis 8-second period by means of exactly the same radiofrequency envelopeand gradient section selection parameter, with the same repetition time of2 seconds to allow the magnetization to reach an equilibrium amplitude beforethe next period of data acquisition. The frequency-encoding gradient was turnedoff during this period to minimize acoustic noise and ensure that the instructionswere heard clearly by the subjects.⁴⁸ The 4dummy volumes were later discarded from the time series.

Individual brain activation maps were coregistered to a "whole head"gradient-recalled echo planar imaging scan of superior spatial resolutionacquired on all subjects. This structural scan had the following acquisitionparameters: echo time, 40 milliseconds; repetition time, 3000 milliseconds;field of view, 24 cm; image resolution, 128 × 128; number of sections,43; section thickness, 3.0 mm; intersection gap, 0.3 mm; number of signalaverages, 8.

Data were analyzed with software developed at the Institute of Psychiatry,using a nonparametric approach. Data were first realigned⁴⁹ tominimize motion-related artifacts and smoothed by means of a gaussian filter(full width at half maximum, 7.2 mm). Responses to the experimental paradigmswere then detected by time-series analysis using gamma variate functions (peakresponses weighted between 4 and 8 seconds) convolved with the experimentaldesign to model the blood oxygen level–dependent response. A goodness-of-fitstatistic and a measure of the mean power of neural response (the sum of squares[SSQ] ratio) was computed at each voxel. This was the ratio of the sum ofsquares of deviations from the mean intensity value due to the model (fittedtime series) divided by the sum of squares due to the residuals (originaltime series minus model time series). To sample the distribution of SSQ ratiounder the null hypothesis that observed values of SSQ ratio were not determinedby experimental design (with minimal assumptions), the time series at eachvoxel was permuted by a wavelet-based resampling method.^50,51 Thisprocess was repeated 10 times at each voxel to produce the distribution ofSSQ ratios under the null hypothesis. Voxels activated at any desired levelof type I error can then be determined by obtaining the appropriate criticalvalue of SSQ ratio from the null distribution. Individual brain activationmaps were produced for each subject for each experiment vs the neutral condition.

To extend inference to the group level, the observed and randomizedSSQ ratio maps were transformed into standard space⁵² bya 2-stage process⁵³ using spatial transformationscomputed for each subject's high-resolution structural scan. Once the statisticmaps were in standard space, a generic brain activation map was produced foreach experimental condition by testing the median observed SSQ ratio overall subjects at each voxel in standard space (median values were used to minimizeoutlier effects), against a critical value of the permutation distributionfor median SSQ ratio ascertained from the spatially transformed wavelet-permuteddata.⁵³ For greater sensitivity and to reducethe multiple comparison problem encountered in fMRI, hypothesis testing wascarried out at the cluster level using methods developed by Bullmore et al.⁵⁴ This method estimates the probability of occurrenceof clusters under the null hypothesis using the distribution of median SSQratios computed from spatially transformed data obtained from wavelet permutationof the time series at each voxel (see preceding section). Imagewise expectationof the number of false-positive clusters under the null hypothesis is setfor each analysis at less than 1.

Analysis of covariance was carried out on the SSQ ratio maps in standardspace by first computing the difference in mean SSQ ratio between groups ateach voxel. The BDI scores were used as covariates in all analyses. Subsequentinference of the probability of this difference under the null hypothesiswas made by reference to the null distribution obtained by repeated randompermutation of group membership and recomputation of the mean difference inSSQ ratio. Cluster-level maps were then obtained as described by Bullmoreet al.⁵⁴ We set a voxelwise P value of .025 and a clusterwise P valueof .0001. This method ensured a total number of false positives close to zero.Correction for multiple comparisons was not required, as thresholds were seton an image-wide basis, not a voxelwise basis.

Correlation of fMRI blood oxygen level–dependent responses withbehavioral measures was determined by first computing the Pearson productmoment correlation coefficient at each voxel between the standardized powerof the fMRI response (SSQ ratio) and the behavioral variable for each subject.The null distribution of correlation coefficients was then computed by randomlypermuting group membership (see previous section) and recomputing the correlationcoefficient in an analogous fashion to that used for computation of groupdifferences. Cluster level maps of significant correlations were then computedas described by Bullmore et al.⁵⁴ The voxelwiseand clusterwise P values were set at .05 and .0001,respectively, ensuring less than 1 false positive. For each of the significantclusters identified by the above method, we next extracted the SSQ ratio ofeach participant and conducted a series of partial correlation analyses withthe relevant anxiety and questionnaire measures, controlling for BDI scores.

There were no statistically significant differences between patientsand controls on any demographic variable, but patients had more severe obsessive-compulsive(PI-R, SI-R) and depressive (BDI) symptoms (Table 1). Scores on the PI-R and SI-R suggested marginal levelsof obsessive-compulsive symptoms in the control group. Patients and controlsexperienced similar moderate state anxiety levels (State-Trait Anxiety Inventory)in anticipation of having a scan. All patients endorsed more than 1 symptomtype (Table 2).

Mixed-model analyses of variance with group (patient vs control) asbetween-groups factor and experimental condition (emotional vs neutral) aswithin-subjects factor showed significant main group and condition effectsin all 4 experiments, indicating that the paradigm was effective in provokinganxiety and that patients with OCD experienced higher anxiety levels thancontrols. The group × condition interaction effect was also significantfor the washing and hoarding experiments, suggesting that the difference betweenthe emotional and neutral conditions was greater in the OCD than in the controlgroup (Figure 1).

A series of multiple regression analyses in the patient group showedhighly specific associations between subjective anxiety scores and correspondingquestionnaire measures. Thus, washing-related anxiety correlated only withPI-R washing (nonsignificant trend: r = 0.45, P = .06), checking-related anxiety correlated only withPI-R checking (r = 0.77, P<.001),and hoarding-related anxiety correlated only with SI-R discarding (r = 0.80, P<.001). No significant correlationsemerged in the aversive control experiment.

In response to all types of anxiety, regions activated by both patientsand controls included bilateral visual areas, cerebellum, striatum (caudateand putamen), thalamus, motor and premotor cortices, limbic and paralimbicareas (ventrolateral prefrontal and orbitofrontal gyri, insula, temporal pole,amygdala, ventral/subgenual cingulate gyrus, and hippocampus), and dorsolateralprefrontal areas (medial and middle frontal, dorsal anterior cingulate, andinferior frontal gyri).

Results of the analyses of covariance, covarying for BDI scores, areshown in Table 3 and Figure 2.

Patients demonstrated greater activation than controls primarily inventromedial prefrontal regions: left medial frontal gyrus (Brodmann area[BA] 32/11), right anterior cingulate gyrus (BA32), bilateral orbitofrontalcortex (BA11), and right subgenual anterior cingulate gyrus (BA25) (extendingto the ventrolateral prefrontal cortex [BA47] and the amygdala). Further differenceswere observed in left middle temporal gyrus (BA37), right caudate nucleus,middle frontal gyrus (BA9/46), and left dorsal anterior cingulate gyrus (BA32).Controls demonstrated greater activation than patients within left ventrolateralprefrontal (BA47) and occipital (BA17/19) cortices.

Patients demonstrated greater activation than controls in a large bilateralcluster including various subthalamic and brainstem nuclei, in right putamen/globuspallidus, right thalamus, various dorsal cortical regions (right inferiorfrontal [BA44], right anterior cingulate [BA32], left medial/superior frontal[BA6], bilateral middle and medial frontal [BA8/9], left precentral [BA4]gyri), and visual regions (precuneus/superior parietal lobule [BA7], middleoccipital gyrus [BA19]). There were few differences in limbic/paralimbic regions:right hippocampus and bilateral subgenual anterior cingulate gyrus (BA25,extending to BA11). Controls demonstrated greater activation than patientsin bilateral visual regions (lingual and fusiform gyri) and left inferiorfrontal/precentral gyrus (BA44/6).

Patients demonstrated greater activation than controls in left precentral/superiorfrontal gyrus (BA4/6), left fusiform gyrus (BA37), and right orbitofrontalcortex (BA11). Controls demonstrated greater activation than patients in bilateralvisual areas (BA7/19).

Patients demonstrated greater activation than controls in left occipitotemporalregions (BA19/37). Controls demonstrated greater activation than patientsin bilateral visual areas (BA37/7), posterior cingulate gyrus (BA31), leftanterior insula (extending to the ventrolateral prefrontal cortex and superiortemporal gyrus), and left cerebellum.

The PI-R washing scores were positively correlated with activation inbilateral fusiform and lingual gyri and right superior temporal gyrus, ventrolateralprefrontal cortex, and anterior insula (Figure3).

Checking-related anxiety was positively correlated with activation inleft precentral/superior and inferior frontal gyri, bilateral globus pallidus/putamen,and left thalamus (Figure 4). Similarly,PI-R checking scores were positively correlated with activation in bilateralglobus pallidus/putamen (left: −14, −7, −2, correspondingto Talairach coordinates x, y, and z, respectively; voxels: 7; partial r = 0.70; right: 29, −4, 4; voxels: 12; partial r = 0.53) and left thalamus (−11, −4, 9; voxels:9; partial r = 0.60).

Hoarding-related anxiety was positively correlated with activation inleft precentral/superior frontal gyrus (Figure5).

Correlation analyses within the control group showed few significantassociations between subjective anxiety or clinical scales and brain activation.Positive correlations were found in right occipitotemporal regions (BA19/18/37)in all experiments. Negative correlations were found in left occipital cortex(BA31/19/18; washing and checking experiments), right precentral/inferiorfrontal gyrus (BA6/44; washing experiment), and right middle frontal gyrus(BA46/9; checking experiment).

The "pure" (n = 7) and comorbid (n = 9) OCD groups had comparable sociodemographicand clinical characteristics (Table 4).Their generic brain activation maps were also similar, and there were fewconsistent differences in brain activation between the 2 groups, mainly inoccipitoparietotemporal regions (Table 5).

To our knowledge, this was the first symptom-provocation study to examinethe neural correlates of different symptom dimensions of OCD in a representativesample of multisymptomatic patients using a dimensional approach. The mainfinding was that washing, checking, and hoarding symptom dimensions of OCDwere mediated by distinct but partially overlapping neural systems.

In the washing experiment, patients showed greater activations thancontrols predominantly in bilateral ventromedial prefrontal regions (anteriorcingulate and orbitofrontal gyri). Additional regions included the left middletemporal gyrus, right subgenual anterior cingulate gyrus (extending to theventrolateral prefrontal cortex and amygdala), left middle frontal gyrus,right caudate nucleus, and left dorsal anterior cingulate gyrus. Correlationanalyses showed significant positive correlations between scores on the PI-Rwashing subscale (but not subjective anxiety scores) and activations in bilateralvisual regions, right temporal pole, ventrolateral prefrontal cortex, andanterior insula. These results are consistent with previous OCD symptom provocationstudies that mainly recruited washers.^{20-25,29,30,55} Thesefindings also parallel those of symptom provocation studies in specific phobias,^56,57 which share elements of fear anddisgust with contamination/washing symptoms.⁵⁸ Takentogether, these findings suggest that washing-related anxiety is associatedwith regions involved in the processing of emotions,^59,60 specificallydisgust.^26-28

In the checking experiment, patients showed greater activation thancontrols predominantly in regions important for motor and attentional functions:a large bilateral cluster in the subthalamic region including various brainstemnuclei, right putamen/globus pallidus, right thalamus, and various dorsolateralcortical regions (inferior frontal, dorsal anterior cingulate, medial/superiorfrontal, middle/medial frontal, and precentral gyri). There were fewer statisticallysignificant differences in emotion-processing regions (right hippocampus anda small bilateral cluster in the subgenual anterior cingulate gyrus, extendingto the orbitofrontal cortex). Correlation analyses showed positive correlationsbetween subjective anxiety and activation in left precentral/superior frontal(BA6), left inferior frontal gyrus (BA44), bilateral putamen/globus pallidus,and left thalamus during this experiment. Correlations with PI-R checkingscores gave similar findings: positive correlations in bilateral globus pallidus/putamenand left thalamus. Thus, both state and trait checking-related anxiety correlatedwith activation in similar regions.

These findings are consistent with those of Rauch et al,²⁴ whofound positive correlations between scores on a checking scale and regionalcerebral blood flow in bilateral striatum, and Phillips et al,²⁵ whofound that only checkers activated dorsal prefrontal (anterior cingulate andinferior frontal gyrus) and visual regions, thalamus, and caudate nucleus.Hypermetabolism in the putamen has been inconsistently reported in OCD.^61,62 It is possible that an excess ofpatients with checking symptoms were recruited for these studies but few describedtheir samples in detail. A majority of patients (7 of 11) in the Perani etal⁶¹ positron emission tomographic study werelabeled as "checkers." These authors found increased metabolic rates in thecingulate gyrus, thalamus, and putamen/globus pallidus but no orbitofrontalor caudate involvement, findings similar to ours. We suggest that the provocationof checking-related anxiety (or the suppression of checking rituals) is associatedwith dysfunction in a circuit that is important for attentional and motorfunctions as well as the inhibition of unwanted impulses^59,63-65 ratherthan emotion processing per se.

In the hoarding experiment, patients showed increased activation inleft precentral/superior frontal (BA4/6), fusiform (BA37), and right orbitofrontal(BA11) gyri, compared with controls. Significant correlations with hoarding-relatedanxiety were found in left precentral/superior frontal gyrus (BA4/6). An associationbetween hoarding-related anxiety and activation in motor cortex was unpredicted,and its significance is uncertain. Increased activation in right orbitofrontalcortex is congruent with the intense emotional reactions these patients experiencewhen they are asked to discard their possessions.⁶⁶ Activityin this region has been shown to be negatively correlated with response topharmacotherapy^55,67-69 andpositively correlated with response to cognitive behavioral therapy.⁶⁸ The relationship between our finding of increasedactivation in this region during the provocation of hoarding symptoms andthe well-documented lack of treatment response of these patients^3,5,14,15,17 requiresinvestigation.

The inclusion of an aversive, symptom-unrelated experiment allowed usto explore the neural correlates of general emotional reactivity independentof the content of the patients' symptoms. Although patients experienced moresubjective anxiety than controls, they showed greater activation only in occipitotemporalregions, suggesting that the findings of the foregoing experiments were mostlysymptom specific.

As in previous symptom provocation studies,^22,30,31 thepresentation of OCD symptom–like material was associated with significantincreases in subjective anxiety not only in patients but also in controls.Consistent with a few previous reports,^25,30,31 controlsactivated brain regions similar to those activated by patients. These resultsare not surprising, as these areas have been repeatedly associated with theinduction of various emotional states in normal subjects^70-75 andpatients with other anxiety disorders.^76-78 Greateractivation in these regions among patients paralleled their higher anxiety,reflecting the greater salience of the provoked stimuli in the patient group.

Controls showed greater activation than patients in bilateral visualareas in all experiments. Furthermore, significant correlations with subjectiveanxiety and symptom rating scores were observed primarily in occipital regions.Increased activation within visual cortex was repeatedly demonstrated in responseto emotive compared with neutral visual stimuli in healthy individuals.^30,79,80 It is plausible thatcontrols directed their attentional resources to the processing of the pictures'visual details rather than their emotional salience.^25,81

Controls showed greater activation than patients in left inferior prefrontalregions during the washing (BA47) and checking (BA44/6) experiments. Similarregions have been associated with suppression of negative emotions⁸² and might reflect more successful regulation of anxietyin controls. During the aversive experiment, controls also showed greateractivation than patients in the left insula (extending to the ventrolateralprefrontal and superior temporal cortices), which are emotion and disgustperception areas. This might reflect a bias toward highly aversive (but notsymptom-related) material in controls and the opposite pattern in patients.

This study did have certain limitations. We did not exclude patientswith comorbidity. Comorbid depression has been found to affect resting-stateregional glucose metabolism in positron emission tomographic studies.^83-85 However, comorbidityhad little impact on our results: (1) it was constant across the 4 experiments;(2) patients with (n = 9) and without (n = 7) comorbidity had similar sociodemographicand clinical characteristics and showed no consistent differences in brainactivity; (3) BDI scores were used as covariates in all analyses; and (4)there were few consistent differences between patients and controls in theaversive control experiment.

Since most patients (n = 12 [75%]) were taking medications, we couldnot compare medicated and unmedicated patients. However, (1) medications wereconstant across the 4 experiments; (2) several studies have demonstrated thatdrug treatment has a normalizing effect on pretreatment functional abnormalities^85,86; medication would therefore haveattenuated rather than inflated our results; (3) symptom provocation studieswith²² or without²³ medicatedpatients reported similar results; and (4) there were no consistent differencesbetween patients and controls in the aversive control experiment.

The sample was relatively small, but this is the largest fMRI studyin OCD to date. The reported effects were strong and consistent across variousmethods of analysis. It is possible that our hoarding experiment was underpowered,since only half of our sample had current hoarding symptoms; further researchon the hoarding dimension is warranted. The neural correlates of the symmetry/orderingdimension remain to be investigated.

The relative inconsistency of findings from previous functional neuroimagingstudies of OCD may have resulted from phenotypic variations among subjectgroups. Replication of our findings would suggest that discrete neural systemsmight mediate the expression of different symptoms. Because of the neuroanatomicproximity within the frontostriatothalamic loops,⁵⁹ itis not surprising that the different symptom dimensions often coexist in anygiven patient. Obsessive-compulsive disorder could be better understood asa spectrum of multiple potentially overlapping syndromes that are likely tobe continuous with "normal" worries and extend beyond the traditional nosologicboundaries of OCD. Each symptom dimension might reflect the dysregulationof highly conserved complex and partially overlapping neural systems thatserve to detect, appraise, and respond to potential threats.⁸⁷

Corresponding author: David Mataix-Cols, PhD, Departments of PsychologicalMedicine and Psychology, Institute of Psychiatry, 5th Floor, Thomas Guy House,Guy's Hospital, London SE1 9RT, England (e-mail: d.mataix@iop.kcl.ac.uk).

Submitted for publication June 13, 2003; final revision received January8, 2004; accepted January 29, 2004.

This study was supported by project grant 064846 from the Wellcome Trustto Drs Phillips, Mataix-Cols, and Speckens.

This study was presented in part at the 58th Annual Meeting of the Societyof Biological Psychiatry; May 15, 2003; San Francisco, Calif.