Neuroimaging Studies of Human Anxiety Disorders

Lewis R. Baxter, Jr.

We may look forward to a day when paths of knowledge and, let us hope, of influence will be opened up, leading from organic biology and chemistry to the field of neurotic phenomena, . . .

—Sigmund Freud, 1926 (25)

Freud's remark, actually concluded: ". . . That day still seems a distant one." Were he alive today, however, Freud might not be so pessimistic about the distance in time to an understanding of brain mechanisms mediating these psychological disorders (26, 72). Present day instruments for studying the chemistry and physiology of the living human brain are helping to bring to psychiatry the scientific analysis about which Freud could only dream.

In the past 5 years, there has been significant progress in in vivo brain-imaging studies of some of the anxiety disorders that interested Freud, especially obsessive–compulsive disorder (OCD), sufficient to warrant this fresh optimism. The purpose of this chapter is to aquaint the reader with relevant human brain-imaging studies that shed light on how the central nervous system (CNS) may mediate the symptomatic expression of anxiety disorders.

The author does not believe that all the DSM-IV designated anxiety disorders are intrinsically any more closely related to each other than some of them are to other psychiatric problems, like depression or Gilles de la Tourette's syndrome, in the case of OCD (12, 16, 56), but this is the conventional lumping, and all these maladies do share anxiety (or dread) as a major symptom.

Of the official nonorganic anxiety disorders, social phobia and posttraumatic stress disorder have not been investigated sufficiently to merit review at this time. In this chapter, discussion of the other anxiety disorders are frequently interwoven, because this seems most efficient and the only large body of brain-imaging work to date is on OCD. In this vein, studies of induced anxiety states in normals are discussed, where appropriate, as they are often relevant. Review of studies in normals are limited, however, to those works judged directly relevant to understanding psychiatric disorders. Likewise, studies on the pharmacological effects of antianxiety agents in normals or animals, of which there are many, or anxiety states induced by structural brain lesions or drugs, are omitted. Brain-imaging studies of medication effects in the primary anxiety disorders themselves are reviewed when they illuminate possible brain mechanisms of symptom mediation.

The technical modalities covered are the present tomographic brain-imaging techniques—X-ray computed tomography (CT), magnetic resonance imaging (MRI) for structural anatomy (there are no published functional MRI studies of anxiety disorders to date, although several are in progress), single photon computed tomography (SPECT), and positron emission tomography (PET). The first two give information on brain structure, the latter two probe biochemical and physiological functions. The nontomographic brain-mapping techniques, such as various electroencephalographic (EEG) methods, are not surveyed. Although some relevant technical information on the functional imaging techniques will be given in passing in the context of critical review, readers are referred to other chapters of this volume, and other references (2, 3, 57, 58) for discussion of the methods for each form of brain imaging.

A search of the literature uncovered no structural brain-imaging studies of a series of patients with these disorders. Wu and colleagues (75) have studied 18 patients diagnosed with primary generalized anxiety disorder (GAD) with PET and the fluorodeoxyglucose (FDG) method to determine local cerebral metabolic rates for glucose (lCMR-Glc). Patients were studied during a passive viewing task off medication, and then in an active visual vigilence task after oral clorazepate (7.5 mg; n = 8) or placebo (n = 10); there were 15 normal controls. Significant findings included lower basal ganglia lCMR-Glc in patients compared to normal controls during passive visualization, but increased lCMR-Glc in the left inferior primary visual cortex, right posterior temporal cortex, and right precentral frontal gyrus. Active vigilance resulted in basal ganglia activation in GAD compared to controls, although the benzodiazepine decreased metabolic rates in the whole cortex, basal ganglia, and some limbic structures. Interpretation of this study is difficult because of the extreme number of post hoc analyses and summaries reporting of lCMR-Glc for various brain regions, but it is significant in noting that basal ganglia were significantly lower in anxious patients in the passive state—different than in FDG-PET studies of OCD (see below).

Two studies have looked at regional cerebral blood flow (CBF) in patients with small animal phobias. Mountz and colleagues (50) examined seven such phobics in a test–retest manner with PET for regional CBF. Studies were done as an intrasubject cross-over at baseline and when exposed to the animal. Eight normal controls studied under similar conditions. Although global absolute CBF was lower in the phobics when stimulated than at baseline and compared to results in the controls, when the method was corrected for hypocapnia, no global or regional differences were significant for either group. Wik and coworkers (73), on the other hand, studied six women with snake phobia in three states: (a) at baseline viewing a neutral stimulus, (b) viewing a video tape of snakes, and (c) viewing a tape of other adverse scenes not involving snakes. They observed significant increases in secondary visual cortex CBF, and reduced CBF in the hippocampus, orbitofrontal, prefrontal, temporal poles, and posterior cingulate cortex with snake exposure, compared to the neutral stimulation. The nonsnake adverse stimulus showed similar, but less intense patterns as those seen with snake viewing.

Two studies of normals are relevant to the phenomona of generic situational anxiety. In one (28), 43 normal volunteers were administered the State-Trait Anxiety Inventory before and after FDG-PET scanning with no specific stimuli. No relationship was observed between measured anxiety level and lCMR-Glc findings, and the authors concluded that the effects of simple situational anxiety at the levels experienced under these study conditions, if they exist, are obscured by the normal variance inherent in the FDG-PET method used. Reiman and collegues (63) studied anticipatory anxiety (threatened painful electrical shock) and measured CBF changes compared to baseline in normal volunteers (n = 8). They observed significant increases in CBF in the region of the lateral temporal poles, bilaterally, and attributed these to anxiety-related brain activation. Unfortunately, subsequent work by the same and other groups suggest that the temporal pole findings may have been generated by an artifact resulting from jaw clenching and consequent increased glucose metabolism in maseter and temporalis muscles. The data reduction methods employed, whereby brain regions are identified by idealized stereotactic coordinates, may have resulted in misidentification of muscle activity as being inside rather than outside the skull (22). Whether there are any specific patterns of functional brain activation in normals with situational anxiety is unclear at this time.

Panic disorder patients (n = 30) were studied with X-ray CT by Lepola and collegues (39). Despite the fact that this patient group had a high baseline suspicion of organic lesions, only six had any identifiable structural abnormality, and these did not fit any recognizable patterns. On the other hand, Ontiveros and collaborators (54) when studying 30 lactate-sensitive panic disorder subjects with MRI did find 43% of the panic subjects with right temporal lobe dysmorphias, whereas only 10% of normals (n = 20) could be so classified. Furthermore, panic disorder patients with these temporal abnormalities were significantly younger and had more panic attacks than those with normal MRI scans. These results are in line with a single CT report of a patient developing panic disorder after right parahippocampal infarction (42) and with functional brain-imaging studies (below).

Reiman and collegues have published two PET studies of panic disorder subjects. In the first (61), 16 patients with lactate-induced panic attacks were compared with 11 controls in the resting state; the panic disorder patients had L < R asymmetry of blood flow, blood volume, and metabolic rates for oxygen in the parahippocampal gyrus. In a follow-up report of these subjects and others (24 patients with panic disorder, 18 normals), subjects were studied before and during lactate infusion to induce a panic attack (62). Patients who did not experience panic, as well as normal controls, showed no significant change in CBF on lactate infusion, but patients who panicked with this challenge showed increased CBF in temporal poles, insular cortex, claustrum, lateral putamen, and superior colliculus—all bilaterally—and near the left anterior cerebellar vermis. This study, however, is subject to the same questions concerning neuroanatomical localization as the study of induced anxiety in normals (22, 63) above.

Two studies using very different functional imaging methods do tend to support the general finding of parahippocampal abnormalities in panic disorder subjects at baseline. Nordhal and coworkers (53) studied 12 panic disorder patients and 30 normal controls with FDG-PET tracer uptake during an auditory discrimination task. This investigation also found a left–right (L < R) hippocampal region asymmetry compared to normal controls, this time in lCMR-Glc. They also reported decreased left inferior parietal and anterior cingulate lCMR-Glc and trend-level increases in medial orbital frontal cortex. Teresa et al. (71), used CT and 99mTc-hexamethyl-propyleneamine oxime (HMPAO) SPECT, an indirect index of blood flow, to study seven lactate-sensitive panic disorder patients compared to five controls. No CT abnormalities were noted in the patients, but they did report a decrease in hippocampal perfusion (relative to the rest of the brain) bilaterally and in the left occipital cortex in patients compared to control subjects.

Thus, there are several functional and even structural brain-imaging studies that suggest abnormality in brain regions around the hippocampus in panic disorder. This should be noted in comparison to results in OCD.

My collegues and I have published several reviews of brain-imaging studies of obsessive–compulsive disorder and related conditions (4, 5, 11, 12, 13, 16). Although many new and significant studies are reviewed here that have not been reviewed before, of necessity much will be but a repetition of those previous reviews.

Insel and associates (32) studied 18 patients with DSM-III OCD, using neuropsychological testing (Wechsler Adult Intelligence Scale and Halstead-Reitan Battery) and EEG recordings. As part of this work they also studied a subgroup of 10 patients with CT. Two subjects with EEG abnormalities, four with neuropsychological testing abnormalities, and four others chosen at random underwent CT scanning. Controls were age- and gender-matched subjects without evidence of psychiatric or known CNS illness, although seven had other known "peripheral illnesses." The findings in this study were unremarkable. There were no significant differences between OCD subjects and controls.

Another CT study, this time of childhood-onset OCD by Behar and colleagues (17), did find significant differences in the ventricle-to-brain ratios (VBRs) between OCD and normal subjects. The 17 OCD subjects could have had secondary depression, the severity of which was not reported. The 16 control subjects, had CT scans that were not "questionable clinically" (it is not clear whether they were truly "normal") and were not from individuals with altered consciousness, psychiatric symptoms, or hard neurological signs. All controls were suspected of having CNS pathology, however; this was the clinical reason for the CT study. Here, VBRs were judged significantly greater in OCD subjects than in controls. The ventricular measures did not show a significant correlation with demographic, disease severity, or prior treatment variables, but it was noted that those with compulsions without obsessions tended to have higher VBRs than those who had obsessions.

This second research group also reported results of an other CT study in late adolescents with childhood-onset OCD (40amined caudate nuclei volumes. Subjects were ten young males. Control subjects were ten males in good health. In this study, volumes of the caudate nuclei were found to be significantly smaller in OCD subjects than in controls. All other structures and ventricles were remarkably similar in volume across the two groups, and there were no significant left–right asymmetries.

To date only two studies of OCD patients studied with MRI have been published, although several are reported in progress. Garber and colleagues (27) studied 32 patients with OCD during a double-blind treatment study while on either clomipramine (n = 19) or placebo (n = 13). These patients were compared with normal controls. There were no distinct structural abnormalities noted in the OCD patients by inspection. Neuroanatomic structure volumes or other morphological measurements were not calculated. In both patients and controls, a variety of nonspecific T2 hyperintensity lesions were seen in the white matter. The authors also used the controversial approach of T1 mapping in comparing OCD subjects to control subjects. Subjects with OCD with a positive family history had more negative mean and right–left (R > L) hemisphere asymmetry T1 differences in the anterior cingulate than other patients or controls. Right–left T1 differences for the orbital cortex gave significant correlations with symptom severity in unmedicated patients and patients with a family history positive for OCD. Although the exact meaning of such findings is not clear, the authors did feel that the findings were consistent with frontal–limbic–striatal pathology.

Kellner et al. (37) studied with MRI 12 OCD subjects and 12 age- and sex-matched healthy controls. Patients were five females and seven males, average age 40. They had Yale-Brown OCD scale scores above 16. Scanning was for T1- and T2-weighted images. Measures were made of the caudate, cingulate gyrus thickness, intracaudate/frontal horn ratio, and the corpus callosum. No significant differences were found between the OCD patients and the normal control subjects.

Two HMPAO studies of pretreatment OCD have been published at the time of this review, although several others are known to be in progress. In the first study (41), ten nondepressed, medication-free OCD subjects were compared to eight Johns Hopkins employees or students, matched by age, race, and sex (three females in each group). Studies were done on the Toshiba GCA-90B single-rotating Anger camera (resolution, 16 mm; slice thickness, 11 mm). Patients had higher medial frontal perfusion than controls (109.7% ± 3.7% of cortical mean vs. 102.9% ± 3.6), but this gave a significant correlation with anxiety (r = -0.84, p = 0.002), not OCD severity (p > 0.50). Orbital metabolic rates were not significantly different between the two groups. Besides the difficulty with the significant finding correlating only with anxiety and not OCD symptoms, and there being no other pathological comparison group to control for anxiety, there are inherent technical problems with this study. The SPECT camera used is of low resolution. It is doubtful that the orbital region, which is not trivial to localize even with higher resolution PET technology, could be reliably identified. Furthermore partial volume effects (58), would likely lead to serious overestimation of activity in midline structures, given the present resolution limitations of this technology.

The much higher resolution (8 to 9.6 mm) Shimadzu Headtome, Model SET/031 SPECT system was used for both 133Xe- and HMPAO-measured blood flow by Rubin et al. (64). They studied 10 nondepressed, drug-free males with OCD and 10 matched controls. Although there were no significant differences between the two groups with 133Xe (although blood flow did give a significant correlation with OCD symptom severity), on HMPAO scanning the OCD patients had significantly increased uptake in the bilateral orbital frontal cortex, normalized to either cerebellum or whole brain, and in the left posterofrontal cortex and bilaterally in the high dorsal parietal cortex. They also had decreased HMPAO uptake in the caudate nuclei, bilaterally. Although the most likely reason for detecting differences with HMPAO and not 133Xe is the resolution difference between the two methods, the authors also thought an HMPAO-sensitive difference in blood–brain barrier permeability or a difference in HMPAO trapping through hydrophilic conversion between the experimental groups might be responsible for the effects observed (see below).

Our group at UCLA examined a sample of OCD patients (n = 14), compared to both normal controls (n = 14) and patients with unipolar-type major depression (n = 14) (7). All subjects were matched for age, and OCD and depressed subjects had similar Hamilton Depression Rating Scale and Brief Psychiatric Rating Scale tension and anxiety item scores. Nine of the OCD subjects were in a clear-cut secondary major depression at the time of study. Nine of the OCD subjects were drug-free for at least two weeks, but the other five were on a variety of antidepressants, benzodiazepines, and neuroleptics. The normal and depressed subjects were all drug-free. The normal control group had equal numbers of males and females, whereas the male–female ratio in the depressed group was 5:9, but was 9:5 for OCD.

Scanning by FDG-PET was done with the Neuro ECAT (in-plane resolution = 11 mm, axial = 12.5 mm) in the eyes and ears open state, without specific stimuli. Arteriolized venous blood was used for the calculation of absolute metabolic rates, and an idealized, calculated (rather than measured), attenuation correction was employed. Absolute glucose metabolic rates were determined for the whole cerebral hemispheres, hippocampal–parahippocampal complexes, anterior cingulate gyri, heads of the caudate nuclei, putamen nuclei, and thalamic nuclei, as well as various gyri in the prefrontal and temporal cortex. A one-way ANOVA was run among the subject groups for each structure, using the first seven subjects only. For those structures yielding significant results, a second, prospective analysis was run using the second group of seven. A separate analysis was also done excluding OCD subjects on drugs. Neuroanatomical regions of interest had to pass all tests to be considered significant. This approach was chosen to decrease the likelihood of type I (false-positive) statistical error, but greatly increased the likelihood of type II (false-negative) error.

Absolute glucose metabolic rates for the whole cerebral hemispheres, caudate nuclei, and orbital gyri were found to be significantly elevated in OCD, compared to the control groups, by these criteria. Furthermore, metabolic rates in the left orbital gyrus, divided by those of the ipsilateral hemisphere (normalized metabolic rates) were significantly higher than those found in normals on the left. There was only a trend for this to be the case in the right hemisphere. However, there was no statistically significant left–right asymmetry on this measure, and values on the right were higher in OCD than in normal controls. Normalized caudate metabolic rates in OCD were not different from normal control values.

The subject groups in this first FDG-PET study had several characteristics that were not optimal: (a) There was an unequal number of male and female subjects across the experimental groups (8); (b) most of the OCD subjects also had concomitant major depression; (c) some of the OCD subjects were on medications; and (d) handedness was not equal across the groups. Therefore, a second FDG-PET study was undertaken to compare a new group of drug-free, nondepressed, right-handed OCD patients to normal control subjects of similar age, scanned under the same conditions as before (9).

Glucose metabolic rates in the cerebral hemispheres, heads of the caudate nuclei, and orbital gyri were examined, using an a priori design in an attempt to confirm or refute previous findings. The results obtained in this study were similar to those of the previous study. The OCD subjects had significantly higher glucose metabolic rates for the whole cerebral hemispheres, heads of the caudate nuclei, and orbital gyri than those found in normal control subjects. The orbital gyrus–hemisphere ratio (the normalized rate) was also elevated in OCD when compared to normal controls, but in this second study this finding was bilateral. In the previous study, where significant results had obtained only on the left, there was a higher percentage of males in the OCD group than in the normal control group (male–female OCD = 9:5; controls = 7:7). The data in this second study, when analyzed with a repeated measures ANOVA for variance and covariance, showed that there was a significant effect of sex by hemisphere on this measure (with males showing higher values on the left and females showing higher values on the right). The unbalanced sex ratios between the diagnostic groups that obtained in the first study probably accounted for the observation of statistical significance between OCD and normals for the left orbital gyrus–hemisphere ratio, whereas there was only a trend toward significance in the right hemisphere. We now believe that there are significant elevations in both left and right orbital gyrus– hemisphere ratios in OCD, compared to normals. (Why both OCD and normal control females should have a higher orbital gyrus–hemisphere ratio on the right than the left, whereas males of both groups show the opposite pattern, is a mystery at this time.)

At the same time that the second UCLA study was being undertaken, two separate groups at the National Institute of Mental Health were performing similar studies independently. Nordahl et al. (52) studied eight nondepressed OCD subjects. These subjects were compared with 30 normal volunteers with no personal or family history of psychiatric problems. Scanning by FDG-PET was done with tracer uptake taking place during an auditory continuous-performance task with eyes closed. Arterial blood was used, and attenuation was measured with a transmission scan. Neuroanatomical regions of interest were determined by two independent raters. Sixty regions of interest were examined, with those found significant in the previous UCLA studies examined a priori with one-tailed Student's t test. Calculations of global brain metabolic rates, which were also used to normalize rates, were done with gray matter structures only, rather than all supratentorial brain regions, as in the UCLA work. Absolute metabolic rates for whole-brain gray matter were calculated, but only normalized values for individual regions of interest were examined.

This group also found normalized regional brain metabolic rates higher in OCD than in normals in both orbital gyri and normalized OCD caudate rates similar to normals. This group did not find elevated absolute global brain glucose metabolic rates, however. Two-tailed Student's t tests, with p < 0.05 considered significant, were done for other regions of interest in an exploratory analysis, with no corrections for the number of tests done. Normalized values in the right parietal and left occipital-parietal regions were higher in normals than in OCD subjects by this criterion. This group interpreted their findings in the orbital cortex as consistent with those of the UCLA group.

The other NIMH group (69) examined nine men and nine women with childhood-onset OCD but no concurrent major depression or other anxiety disorders. All were drug-free for at least 2 weeks and physically healthy. These subjects were compared to age- and sex-matched, physically and mentally healthy control subjects. Scanning by FDG-PET was done in the eyes- and ears-closed condition, with arterial blood sampling and measured attenuation correction. The global metabolic rate was calculated based on the simple mean of cortical gray matter only. This measure was used also for normalizing other regions of interest.

The OCD patient group showed increased absolute glucose metabolism in left orbital and right sensorimotor regions and, bilaterally, in the anterior cingulate gyri and lateral prefrontal areas. Normalized values were significantly increased in right lateral prefrontal and left anterior cingulate regions only. These authors, however, did observe a significant correlation between absolute and normalized right orbital glucose metabolic activity and a measure of OCD severity. Furthermore, six of these patients, who failed to respond to a subsequent trial of clomipramine had significantly higher right anterior cingulate and right orbital metabolism than 11 patients who did respond. This research group interpreted their findings in light of other work on OCD to suggest a dysfunction of a frontal cortex–basal ganglia loop (67, 69). This team, along with other collaborators, subsequently reported a complicated correlational analysis of data from these patients (31), but the methods are hard to follow, and the conclusions to be drawn are unclear at this time.

Another FDG-PET study (43) found results at odds with those of the other PET studies reviewed above. Sixteen OCD patients were compared to eight normal controls in the resting state. Both absolute rates and rates normalized to whole brain were analyzed. All brain regions examined were reported to show lower absolute rates in the OCD subjects than in the controls. For normalized values, however, the lateral prefrontal cortex had significantly lower rates in the OCD subjects than in normal controls. They could not confirm orbital abnormalities. Given the similarity between these results and our data on depressed OCD subjects (see below), it is particularly important to note that the authors assure us that no subjects were depressed at the time of PET scanning and results of drug-free subjects were similar to those scanned while on medication. However, it is possible that a different threshold of depression severity for a diagnosis of major depression was applied in this study than in those done in the United States. In a PET study of depression by this same group, recovered patients had a mean score of 32 on the 26-item Hamilton Depression Rating Scale (44).

Thus, four separate FDG-PET studies have found evidence for inferior prefrontal cortex abnormalities in OCD, and one has not.

Yet another FDG-PET study provides information relevant to the phenomenon of secondary, major depression in OCD. In this study (10), drug-free, age- and sex-matched, right-handed patients with unipolar depression (n = 10), bipolar depression (n = 10), OCD with secondary major depression (n = 10), OCD without depression (n = 14), and normal controls (n = 12) were evaluated under the same conditions and with the same machinery as in the other UCLA studies. Six non-sex-matched manic subjects (four males) were also evaluated under conditions that were otherwise the same. Depressed patients in all three groups had similar levels of depression on the Hamilton Depression Rating Scale (HDRS) and OCD patients with and without depression were of similar severity of OCD symptoms, measured on the NIMH OCD scale.

As predicted a priori, ispilateral-hemisphere-normalized glucose metabolic rates for the left dorsal anterolateral prefrontal cortex were similar in the unipolar and bipolar subjects and significantly lower than values obtained for both normals and OCD subjects without depression. Bipolar depressed patients were significantly lower than manics on this measure. Likewise, OCD with major depression showed significantly decreased glucose metabolic values in this brain region, compared to OCD without depression. Furthermore, there was a significant negative correlation (r = -0.5; p = 0.0002) between this brain metabolism measure and scores on the HDRS.

With antidepressant medication, normalized left dorsal anterolateral prefrontal cortex activity increased significantly, and the percentage change in HDRS scores gave a significant (r = -0.5; p = 0.045) negative correlation with percentage change on the metabolic variable. It was concluded that, despite being different on other measures of cerebral glucose metabolic rates, OCD with major depression, like the other two primary major depressions, is distinguished from the nondepressed (OCD) state by a decrease in normalized left dorsal anterolateral prefrontal glucose metabolic rates.

Zohar et al. (76) studied ten subjects using SPECT and 133Xe, a low resolution but direct method for determining CBF. All the subjects had washing compulsions and reports of focal stimuli for their obsessions concerning contamination. The design was complicated. There were six women and four men, and all were drug-free for at least 3 weeks. Subjects underwent two single-blind "placebo Xe" test runs in an attempt to diminish test-situation anxiety. All studies were done with eyes closed for 16 min, with 30-min intervals in between. The subjects underwent, in sequence, Xe–flow studies (a) in a relaxation state, (b) during imaginal flooding, and (c) during in vivo exposure to stimuli that, in the past, had tended to induce contamination obsessions and washing rituals. All three stimulus state studies were done during a 4-hr time period on the same day. The relaxation stimulus was done with an audiotape of a relaxation scene (the same for all subjects), and the flooding was done with an audiotape of a situation specific to the subject's individual contamination fears. Exposure in vivo was done with the flooding tape, accompanied by placing the contaminating object in contact with the dorsum of the right hand. In an attempt to gain information as to whether results were biased by the fixed sequence of presentation, three subjects returned, 5 to 10 days later, for testing in which exposure was given before flooding.

Both symptom-rating-scale scores and physiological measures gave highest values during in vivo exposure and were lowest during relaxation. In contrast, CBF was significantly increased in the imaginal flooding over the relaxation state, but only in the temporal cortex, and was significantly decreased in virtually all cortical regions (specifically temporal, parietal, posterior, and in the cortex as a whole) during exposure in vivo. In all three conditions, flow in the left hemisphere was greater than on the right, and the left prefrontal cortex appeared slightly more reactive than the right in the directions cited above. There did not appear to be an effect of order of presentation in the three subjects in whom this was evaluated.

The authors found the results of this study puzzling. They had predicted that the peripheral physiological measures of anxiety would increase going from relaxation to imaginal flooding to in vivo stimulation, which they did. However, they predicted the same direction of effect on CBF as well. Instead, blood flow decreased in in vivo stimulation.

A number of speculations were offered that might have accounted for the results. Differences in the sensory modalities of the stimuli (i.e., auditory versus touch) could have been important. Imaginal flooding might require more active cortical involvement than the passive touch situation, but this explanation was not favored, due to patient reports of the degree of conscious activity in the various situations. Another possible explanation was that, in the highly anxiety provoking in vivo exposure situation, blood was being shunted away from the cortical areas of the brain to other areas, such as the caudate nucleus or orbital gyri (see PET studies, below), which cannot be visualized with the 133Xe-flow method.

One group has used 11C-glucose and PET to measure cerebral glucose metabolism in five patients with OCD refactory to conventional treatment who were studied 10 days before and 1 year after surgical capsulotomy (47, 48, 49). Drug status is not clear. These patients' scans were compared to scans of ten healthy males. The PET scanning and blood sampling were done over a 15-min period and the model for calculating metabolic rates used estimates of 11C—CO2 loss from brain based on other experiments in anesthetized monkeys where the arteriovenous difference across the brain is measured after an iv bolus and the Fick Principle applied. We are told that orbital and caudate metabolic rates in patients were significantly higher before than after surgery, but it is not clear whether these are absolute or normalized metabolic rates. We are told further that although the orbital region was not determined in normals, that Brodmann area 11 was higher in the five normal controls where it could be located than was the orbital cortex in OCD patient pretreatment. The investigators also correlated measured lCMR-Glc changes with various personality measures on the Karolinska Scales of Personality. They observed significant positive correlations of the left caudate nucleus lCMR-Glc with decreases in somatic anxiety, psychic anxiety, psychasthenia, and guilt, and in the right orbital gyri with decreases in psychic anxiety, psychasthenia, suspicion, and guilt. Although this report finds similar pre- to posttreatment changes in orbital and caudate metabolic rates as reported with other methods (below), besides the multiple questions about patient status and the small subject numbers, there are the problems with the 11C-glucose method itself. If the orbital region of the brain is in fact overstimulated in OCD, as supported by most FDG studies and theories advanced about brain function in OCD (see below), then the 11C-glucose method would underestimate the glucose metabolic rate in OCD patients versus normals.

The Johns Hopkins group examined six of the OCD subjects of their HMPAO-SPECT study after 3 to 4 months of 80 to 100 mg/day of fluoxetine (29). Obsessive– compulsive symptoms showed significant decreases with treatment, as did ratios of medial-frontal cortex to mean cortex. No Pearson correlations with clinical symptoms were significant, but this method of correlation is decidedly not valid with subject numbers this low. The orbital region of the brain did not change pre- to posttreatment. It should be noted that, although the exact localization of the medial-frontal region is not clear in these reports, it is said to include the anterior cingulate cortex (29, 41).

Benkelfat et al. (18) restudied eight of their initial FDG-PET study OCD subjects after an average of 16 weeks of drug treatment with clomipramine. For the group as a whole, whether responders to treatment or not, metabolic rates normalized to whole brain decreased in some regions of the medial and right orbital cortex. Left caudate rates also decreased significantly, but right anterior putamen values increased. When poor responders were compared to those showing good clinical response, good responders showed a significantly greater decrease in left caudate values than poor responders (10.9% ± 5%, mean ± sd, vs 2.1% ± 3.6%). There was a similar directional change between these groups for the right caudate (7.4% ± 18.6% vs -0.07% ± 8.7%), but there was not a significant difference due to the higher variance in the data than on the left.

Swedo et al. (67) restudied 13 of their OCD patients, eight on clomipramine, two on fluoxetine, and three off drug, after at least 1 year of treatment (mean 20 months). Seven patients were treatment responders and six had not changed significantly. Normalized glucose metabolic rates were examined in orbital (lateral), prefrontal, and cingulate cortical regions and in the caudate nuclei. In responders, the orbital cortex decreased 9.0% ± 9.7%, whereas in nonresponders only 0.4% ± 6.1% (p = 0.04). (It should be noted, however, that orbital metabolic rates showed highly significant, positive correlations with levels of clomipramine and its metabolite desmethylclomipramine.) No other brain regions tested showed significant changes. Interestingly, both normalized orbital and caudate metabolic rates were higher in nonresponders before treatment than in responders.

Baxter et al. (15) have studied patients pre- and posttreatment either with fluoxetine alone (n = 9) or behavior treatment using exposure and response-prevention without medication (n = 9). Both treatment groups were well matched on OCD severity, anxiety, age, and sex. All were drug-free on initial scan. They were rescanned after 10 weeks of treatment. In both groups, normalized right caudate metabolic rates decreased significantly in treatment responders (-5.2% ± 2.3% for drug treatment and -8.0% ± 4.8% for behavior treatment), not in nonresponders (0.3% ± 1.0% and 2.6 ± 3.2%, respectively) and the differences between responders and nonresponders were significant for both treatment groups. They were also significant compared to normalized right caudate changes in a small group of normals (0.4% ± 2.0%) rescanned after a similar interval. Percentage change in normalized right caudate rates gave a significant correlation with percentage change on the Yale-Brown Obsessive–Compulsive Disorder Scale for drug treatment, although there was a trend in behavior-treated subjects. The left caudate decreased in responders to each treatment, but not significantly. The right anterior cingulate (cf., SPECT study, above) and left thalamus decreased with fluoxetine, but not with behavior modification.

We also found in this study (15) and by subsequent analysis (5) that, before treatment, eventual positive treatment responders had significant, positive, pathological correlations between glucose metabolic rates in the orbital cortex and the caudate nucleus, bilaterally, and between the orbital cortex and the thalamus, bilaterally. After effective treatment, however, these correlations were abolished, and the change in correlation strength pre-and posttreatment was significant. Although numbers were too small to establish significance, it did not appear that treatment nonresponders had such positive correlations among these structures before treatment (Baxter et al., unpublished data). Both normal controls and patients with unipolar depression do not show these significant correlations of glucose metabolic rates among the orbit, caudate, and thalamus (11). See Table 1.

Rauch and colleagues at Harvard (60) have used PET measures of CBF before and while acutely exposing subjects (n = 8) with OCD to stimuli that acutely increase their OCD symptoms (see Fig. 1). They demonstrated significant and striking increases in CBF in the right head of the caudate nucleus and orbital gyri, and lesser increases in thalamus, anterior cingulate, and dorsolateral cortex. Caudate laterality comes up again, this time consistent with our report (15), rather than with Benkelfat et al. (18). Because regional brain blood flow and glucose metabolism are usually tightly correlated (58), many of these PET findings by Rausch and collaborators on augmentation of OCD symptoms can be viewed as the reciprocal of changes we and others have noted with FDG-PET after effective treatment of OCD, and quite compatible with most FDG studies of OCD at rest. This is a remarkable validation that these brain regions are involved in the mediation of OCD symptoms. (Other regions, for example, the cingulate, may be for nonspecific distress; the same Harvard group has now studied six humans with small animal phobia in a similar manner, and found a different activation pattern than in OCD, namely cingulate and medial temporal blood flow increases (Jenike and Rausch, personal communication).

Rubin and colleagues have recently repeated the SPECT procedure on eight of the subjects of their initial 133Xe- and HMPAO-SPECT study (above) while still on clomipramine, after a mean treatment interval of 7 months (range 2 to 10 months) (Rubin, personal communication, and ref. 65). On clomipramine the previous increases in HMPAO uptake in the orbital, posterofrontal, and high dorsal parietal cortex showed significant reductions compared to the range of normal controls. Significantly reduced caudate HMPAO uptake in OCD patients compared to controls before treatment did not change with clomipramine.

We must note the conflicts in the functional brain-imaging reports of treatment effects on the brains of OCD patients. These could relate to specific differences in the treatment chosen for each study (e.g., clomipramine vs. fluoxetine), different stimulus conditions at the time of scanning, the scanning technique used, or other factors.

Specifically, although FDG and 11C-glucose PET both measure a form of cerebral glucose metabolic rates, they are not at all equivalent. Unfortunately, this is not always recognized, leading to needless confusion. For this reason a few lines are necessary about each technique. Fluorodeoxyglucose is phosphorylated and transported into cells proportionally to glucose transport and equated to glucose via the lumped constant in the modified Sokoloff equation (58), but once in the cell FDG undergoes no further metabolism and is trapped. Thus, not only is FDG-PET scanning done after uptake is complete, but it also reflects total glucose uptake, whether that glucose is being used for aerobic or anaerobic processes.

11C-glucose, on the other hand, undergoes metabolism rapidly and the 11C label is lost. How rapidly depends on the position of the 11C in the glucose molecule. Under complete aerobic conditions, the carbon skeletons of glucose are largely retained in the amino acid pools of the Krebs cycle during the time period of dynamic PET scanning used for this tracer. Glucose metabolism in nonstimulated (baseline condition) brain regions is thought to be mainly aerobic. However, even in the baseline condition, although glucose labeled with 11C in the 1- or 6-position gives metabolic rates comparable to deoxyglucose in rats sacrificed quickly (6 min), labeling in the other four possible positions of the glucose molecule gives severe underestimations of glucose metabolic rates. This problem is worse in brain regions undergoing physiological stimulation where the underestimation even with 11C-glucose labeled at the 6-position can be as high as 10% in rats just 6 min after administration of the tracer. The error is even higher with 11C-glucose labeled at other positions.

Although the exact reason for this underestimation of brain glucose metabolic rates in stimulated regions with glucose versus deoxyglucose is still somewhat controversial, it is probably accounted for by the fact that, under physiological stimulation, a significant amount of cerebral glucose is metabolized anaerobically. Here, the tracer is quickly converted into lactate, which is quickly carried off and not available for counting in PET. (Remember that deoxyglucose does not undergo metabolism after initial phosphorylation, and is trapped in the cell.) This results in a serious underestimation of total glucose metabolism with 11C-glucose, which only accurately reflects oxidative metabolism. Fluorodeoxyglucose measures total glucose metabolism in stimulated or unstimulated brain regions. Readers interested in more detail are referred elsewhere (38, 58).

The studies by Zohar et al. (76), Rubin et al. (64), and Hoehn-Saric et al. (29) used SPECT techniques, which is itself of special concern. Using 133Xe is a very low-resolution technique that only measures summed activity across the brain surface and provides no information on deep structures. On the other hand, HMPAO SPECT with a high-resolution camera, as that used by Rubin et al. (64) does provide data on deep structures. How HMPAO is used to derive an index of blood flow must be kept in mind, however.

99mTc-hexamethyl-propyleneamine oxidase is a highly lipophilic compound that crosses the blood–brain barrier quickly, but once in cells is metabolically converted to a hydrophilic form that is trapped in the brain at concentrations roughly proportional to the blood flow through the region. Any HMPAO not so metabolized is in equilibrium across the blood–brain barrier, and is therefore subject to redistribution kinetics, that is, it can be carried off by the blood to other lipid sinks in the body. Thus, although HMPAO uptake is usually interpreted as a valid method of measuring the blood flow of one brain structure relative to that of another, there may be circumstances when the rate of the metabolic conversion from the lipophilic to the hydrophilic form in the brain may be a more important confound than usually admitted in the SPECT literature. Although the most likely reason for Rubin and coworkers detecting differences in brain function between OCD patients and normals with HMPAO and not with 133Xe is the resolution difference between the two methods, the authors also thought an HMPAO-sensitive difference between patients and controls in blood–brain barrier permeability, or even a difference in the rate of HMPAO trapping through hydrophilic conversion, might be responsible. Such possibilities must also be borne in mind when considering these studies compared to those using PET glucose and blood flow techniques (the later is a direct measure of blood flow). Rubin and coworker's finding of continued HMPAO abnormality in the head of the caudate nucleus after effective treatment might point to an important abnormality related to an OCD trait dysfunction that is partially compensated by other elements in the system after effective treatment, as measured with FDG-PET. In fact, it fits well with the model proposed by Swedo et al. (67) for a possible cause of one subtype of OCD in which autoantibodies selectively bind in the striatum, where they would be expected to disrupt the blood–brain barrier coating the microvasculature.

The reports of both Benkelfat et al. (18) and of Baxter et al. (15) found decreases in normalized caudate metabolic rates bilaterally; they differed only on which side made statistical significance. Rausch et al. (60) found that stimulating OCD thoughts increased blood flow in the right striatum. Although there is no proven explanation, we note that Rauch et al. (60) stimulated patients on the left side of the body; we performed blood sampling (often an intense focus of patient attention) on the left, whereas Benkelfat et al. sampled blood on the right. In all these studies, many of the subjects had doubting and checking; a lateralized focus of concern may have led to lateralized brain findings, clearly in need of control in further experiments.

Besides laterality, there are differences in the brain region of interest that are reported to change in the three PET reports of treatment effects in the brains of OCD patients. Our group has noted (15) that the report of both Benkelfat et al. and Baxter et al. found decreases in normalized caudate metabolic rates bilaterally but differed on which side made statistical significance. Swedo et al. (67) found only changes in orbital glucose metabolic rates after treatment, whereas Baxter et al. reported only changes in the striatum, although correlations among orbital cortex and other brain regions were broken up after effective treatment. Time-interval differences between pre- and posttreatment scans among these three studies may have been a critical factor. Our group defined the orbital cortex as involving all orbital gyri, whereas Swedo et al. looked at subregions of the orbital cortex. In addition, other PET work done at UCLA (44) has shown that as a person learns to execute a task more efficiently, the brain region mediating the behavior shows a decrease in both the extent and degree of activation on PET scanning, compared to when the task was new. With time the caudate might become more efficient in limiting OCD symptoms, and its change in a critical function might no longer be detectable with present FDG-PET methods (12).

Recent data from our laboratory may help to resolve these apparent conflicts. We have studied seven OCD subjects with FDG-PET pre- and posttreatment with paroxetine hydrochloride. New image-subtraction techniques have been employed in data analysis that allow unbiased assessment of small brain regions that change from one scanning session to another. With this treatment and these techniques we observed significant (>10%) reductions of glucose metabolic rates in the medial orbital gyri, bilaterally, when there is treatment response. There are also glucose metabolic decreases localized to the anterior medial inferior region of the head of the caudate nucleus and nucleus accumbens [precisely that striatal region to which the medial orbital gyri project (13)] with effective treatment (Colgan et al. unpublished data) (see Fig. 2).

We believe we missed detecting a significant change in orbital gyri glucose metabolic rates with effective OCD treatment in previous studies because we lumped the whole orbital gyri, rather than evaluating the medial gyrus separately. Similarly, in studies by other groups, changes in the head of the caudate nucleus may have been missed because only that subregion of the striatum that receives limbic cortex projections is affected. These specific orbital and striatal subregions are the ones that theory would predict are the points of OCD symptom mediation (13).

Many of the brain-imaging studies reviewed here provide evidence for symptom-related functional activity in the orbital prefrontal cortex and caudate nucleus in OCD. This pattern of orbital-striatal dysfunction is different from that reported in panic (above), depression (4), and schizophrenia (3).

Based on these functional-imaging results reviewed above, the results of neurosurgeries for OCD that interrupt tracts coursing among these brain elements (48), and studies in animals (36, 76), we have proposed (15) a model for the mediation of symptoms in treatment-responsive OCD (see Fig. 3). These ideas grew from theories first put forward by others (20, 33, 35, 49, 60; see ref. 15 for review) of a dysfunctional "[cortico/limbic]-basal ganglia-thalamic circuit" in OCD.

In the present formulation of this theory, inadequate sensory information gating (66), sieving, or, to use Freud's term, "repressive" functions in the basal ganglia might allow cortical (in this case orbital) inputs to capture and drive a self-sustaining loop (see Fig. 3). In turn, this loop drives behavioral routines (adaptive in other circumstances) that are difficult to interrupt, even when maladaptive. It should be considered that sources of such context-specific behavioral interference (referant to survival needs) may consist not only of irrelevant sensory stimuli (external interference or distractors) but may also consist of well-established but currently irrelevant sensory or motor representations (internal interference) that assert themselves regardless of new, significant environmental inputs that, if they could get through, would alter behavior.

In this regard, we have argued (see ref. 15 for further discussion and literature review) that one function of the orbital cortex, in conjunction with the neostriatum and thalamus as an interactive system, is to help the animal rivet critical, prepackaged response behavior (macros) to specific stimuli that are judged significant, to the exclusion of less important signals clamoring for other actions. Such a system may have evolved to allow significant threats to capture and direct attention for needed action to the exclusion of other distractors and to rivit behavior to such concerns until the danger is judged passed. To function optimally, however, such a system would have to be adjusted so that goal-directed behavior would continue to implementation, despite insignificant external distractors that may intervene to redirect behavior, yet could still be suppressed if more significant new information intervenes that requires a change in behavior. Thus, we have postulated that if orbital cortex function is set too high in relationship to the rest of this critical circuitry (i.e., in relationship to the strength of the gate closure in the basal ganglia), such internal inputs might capture behavior to the exclusion of significant new information (resulting in preoccupation), whereas if the orbital region is hypofunctional, less important external inputs may prevail, resulting in distractability.

With successful OCD treatment, however, we believe that the gating functions of the caudate are adequately restored, and this driving circuit broken up.

What is not at all clear at this time is what neurotransmitter system(s) might mediate the proposed critical changes in caudate function that might facilitate treatment response. Studies by PET and SPECT with more specific chemical tracers, especially probes for specific neurotransmitter systems, are clearly needed. Since such radioligand development is difficult and expensive, precluding a shot gun approach, the selection of likely candidates might best be persued with autoradiographic studies of many possibilities in animal models.

Circuits involving limbic- and prefrontal–striatal– thalamic systems have been postulated to mediate the symptomatic expression of several psychiatric disorders (21), including schizophrenia (70) and depression (6, 22, 70), as well as OCD and Gilles de la Tourette's syndrome (16). Indeed, the sensory gating functions of the striatum (70) seem to fit the metaphor of successful vs. unsuccessful repression, classically invoked by Freud in many mood and anxiety disorders (72). We propose that specific, different patterns of abnormal neural function in subcircuits of these general cortical- and limbic–basal ganglia–thalamic systems (1, 12, 13, 16, 51) may mediate the symptomatic expression of many of the neurotic phenomana of whose neurological basis Freud could only dream.

This work was supported, in part, by a Research Scientist Development Award (KO1-MH00752), and ROI MH37916 from the National Institute of Mental Health, an Established Investigator Award from the National Alliance for Research on Schizophrenia and Affective Disorders, contract AM03-76 SF00012 from the Department of Energy, the Jennifer Jones Simon Foundation, Los Angeles, and the Judson Braun Chair in Psychiatry at UCLA.

I am indebted to many collaborators, but especially would like to acknowledge here the late Daniel X. Freedman, M.D., a psychiatrist who had a life-long interest in both the workings of the brain and the observations of the classical psychoanalysis, and without whose encouragement and support this work would not have progressed.

published 2000