Summations
-
∙ Ten studies have examined the effects of CBT using brain imaging. They consistently demonstrate that it is associated with measurable changes in brain function.
-
∙ Changes in anterior cingulate cortex activity following CBT are most consistently reported. Given the role of this region in regulating cognitive control it is a predictable target for CBT effects.
-
∙ Though the model has appeal, the imaging data available to date do not fully support the proposal that CBT works by increasing an individual’s capacity for ‘top-down’ emotion regulation.
Considerations
-
∙ To date only a relatively small number of studies have examined the effects of CBT with brain imaging, and study heterogeneity hampers comparison of findings across studies.
-
∙ Ethical considerations prevent comparison of CBT-associated brain changes to those occurring in an untreated depressed group.
-
∙ The current study cannot disentangle effects specific to CBT to those consequent simply to being in a therapeutic relationship.
Introduction
Over the last few decades our understanding of psychiatric illness has been transformed by elucidation of its associated biological abnormalities. Depression, for example, is associated with particular allelic variants, demonstrable endocrine abnormalities and replicated structural and functional brain changes (Reference Willner, Scheel-Krüger and Belzung1). Further, the mechanism of antidepressant action has been examined with both structural and functional imaging (Reference Bellani, Dusi, Yeh, Soares and Brambilla2).
This ‘biological’ understanding of depression must however acknowledge the efficacy of psychological treatments, and raises the question of whether these treatment effects are also biologically identifiable. Of psychological treatments, cognitive behavioural therapy (CBT) has arguably the strongest evidence base. It was developed by Beck in the 1960s, building on his cognitive model of depression which proposed that the individual has a profoundly negative view of oneself, the world and the future, with biased information acquisition and processing crucial to depression onset and maintenance (Reference Beck3). These biases are demonstrable experimentally, and functional neuroimaging can associate them with changes in brain function. It has been shown, for example, that the depressed show amygdala hyperactivity when processing emotionally negative information (Reference Sheline, Barch, Donnelly, Ollinger, Snyder and Mintun4), this excessive activity persisting after the aversive stimulus is removed (Reference Schaefer, Jackson, Davidson, Aguirre, Kimberg and Thompson-Schill5). This explains why depressed individuals are more likely to attend to negative stimuli (Reference Peckham, McHugh and Otto6), and experience a stronger and longer lasting neural response. In another example, activity in the right ventrolateral prefrontal cortex (VLPFC), right dorsolateral pre-frontal cortex (DLPFC) and right superior parietal cortex is decreased compared with healthy controls when shifting attention from negative stimuli (Reference Fales, Barch and Rundle7,Reference Beevers, Clasen, Stice and Schnyer8). This may reflect dysfunction of these regions, potentially underpinning the reduced ability to shift attention from stimuli associated with negative affect (Reference Disner, Beevers, Haigh and Beck9).
Related to the above is a body of evidence emphasising the centrality of ‘cognitive control’ in brain function. This refers to those executive processes that allow information processing and behaviour to vary adaptively from moment to moment depending on current goals, rather than remaining rigid and inflexible. In a highly influential synthesis of these and other data related to prefrontal cortex (PFC) function, Miller and Cohen proposed that the PFC represents and maintains context for responding or goals, which in turn biases processing in posterior and premotor areas in order to support task appropriate responding (Reference Miller and Cohen10). The depression-associated cognitive biases outlined above can easily be conceptualised to reflect changes in these processes, their reduced efficiency also potentially underpinning the deficits in attention, concentration and executive function reproducibly reported in depression (Reference Rock, Roiser, Riedel and Blackwell11). Consequently, if CBT does operate through normalising deranged cognitive control processes, one may expect imaging studies to identify effects in the PFC and related subcortical circuits important in executive function.
Though psychological treatment effects may be assumed too subtle to detect with structural imaging, existing evidence suggests this is not so. CBT for chronic fatigue syndrome has been shown to increase DLPFC volume for example (Reference de Lange, Koers and Kalkman12), and skill acquisition in the healthy can be associated with both grey and white matter changes (Reference Draganski, Gaser, Busch, Schuierer, Bogdahn and May13–Reference Scholz, Klein, Behrens and Johansen-Berg15). It is in this context that we undertook a systematic review of imaging studies examining the effects on the brain of CBT for depression. Though previous reviews have examined functional brain changes associated with psychological treatment, they have tended to pool findings for different psychotherapies and/or psychiatric conditions (Reference Stein, Federspiel and Koenig14–Reference Liberati, Altman and Tetzlaff16).
Aims of the study
We sought to systematically review all studies which utilised functional or structural imaging to examine changes in brain function or structure associated with treatment of depression using CBT. Given what is known about brain structural and functional changes in depression, we expected treatment with CBT to be associated with changes in frontal and subcortical regions important for cognitive control, and the limbic system.
Methods
The systematic review was conducted according to PRISMA guidance (Reference Liberati, Altman and Tetzlaff16). The databases EMBASE (from 1980), PsycINFO (from 1980) and MEDLINE (from 1980) were searched for papers published up to November 2014 using the following criteria: (imaging OR magnetic resonance imaging (MRI) OR functional magnetic resonance imaging (fMRI) OR nuclear magnetic resonance imaging OR positron emission tomography (PET) OR single photon emission computer tomography (SPECT) OR spectroscopy OR diffusion tensor imaging OR diffusion weighted imaging AND cognitive behavioural therapy (CBT) AND depression) (limit to human). This was supplemented by examining citations of identified studies. Hits from the three search portals were collected in Endnote.
Ideally a study aiming to identify brain changes specific to CBT would randomise a cohort of depressed people to no treatment (control group), a non-CBT talking therapy (active control group) and CBT. The active control group would balance for therapeutic benefit resulting from a sympathetic relationship, enabling the specific effects of CBT (distinct from the non-specific therapeutic relationship effects) to be distinguished. This is important, as the non-specific benefits of being ‘in therapy’ contribute much efficacy of all psychological treatments (Reference Martin, Garske and Davis17). Studies of complex treatments with active controls are difficult to conduct however, and rarely undertaken. We therefore considered any study examining CBT effects by comparing brain structural or functional measures before and after treatment in a depressed cohort. As any changes observed could occur spontaneously, ideally changes in CBT-treated patients would be compared with untreated depressed patients. Given the ethics of depriving unwell people of treatment however, we imagined the comparator group would generally be antidepressant-treated patients. This obviously may obscure some effects, as both treatments could bring about similar changes.
Studies were also included if the comparator group were healthy receiving no intervention. These studies, unable to control for spontaneous recovery effects, could conversely overestimate CBT effects. Studies including non-responders to treatment in analyses were included, while acknowledging this may reduce sensitivity to identification of CBT effects bringing about remission. If separate analyses included and excluded non-responders, both will be discussed. Given the increased risk of confounding in non-randomised studies, greater weight must be given to those with randomised design. Comparison of imaging data across studies can be hampered by inconsistent labelling of brain regions. The use of the Brodmann area system can facilitate comparison, and for this reason when Brodmann areas were not given they were determined using Talairach Demon (if necessary after converting coordinates from MNI to Talairach space using MNI2Tal) (Reference Lancaster, Summerln, Rainey, Freitas and Fox18,Reference Brett19). This can facilitate, if feasible, meta-analytic synthesis of findings.
Results
Our search strategy, after omission of duplicates, yielded 199 papers. On review of abstracts 143 were clearly not relevant to this study, the remaining 56 being obtained in full text form. Of these nine met inclusion criteria (outlined in Table 1). Common reasons for exclusion included not focussing on a depressed group and lack of longitudinal data. One additional eligible paper was identified in the process of peer review.
Inclusion/exclusion criteria
The 10 included studies are summarised in Table 2. Eight had a comparator group; in five healthy controls, in the others antidepressant treated depressed patients. Randomisation to treatment was only potentially possible in the latter, and only Kennedy et al. did actually randomise treatment allocation (Reference Kennedy, Konarski and Segal20). The other two studies comparing changes to antidepressant-treated patients used comparison data from an earlier study (Reference Goldapple, Segal and Garson21,Reference Sanacora, Fenton and Fasula22). Treatment was generally at least 12 sessions of individually delivered CBT, though one study used group treatment (Reference Yoshimura, Okamoto and Onoda23), and one internet-based treatment (Reference Tiger, Rück and Forsberg24). The reported efficacy of CBT in reviewed studies was generally comparable to existing efficacy data (Reference Beltman, Voshaar and Speckens25). Studies generally defined response as a >50% reduction on a depression rating scale; this was achieved in more than half of patients in all studies reporting it. Seven studies analysed all CBT-treated depressed patients with a second scan together. Others analysed treatment responders and non-responders separately (Reference Kennedy, Konarski and Segal20,Reference Siegle, Thompson and Collier26,Reference Amsterdam, Newberg, Newman, Shults, Wintering and Soeller27).
Summary of study findings
ACC, anterior cingulate; ATL, anterior temporal lobe; CBT, cognitive behavioural therapy; DAS, modified Dysfunctional Attitudes Scale 48; DLPFC, dorsolateral PFC; ECT, electro-convulsive therapy; fMRI, functional magnetic resonance imaging; HAM-D, 17-item Hamilton Depression Rating Scale; HC, healthy control; HDRS-25, modified 25-item Hamilton Depression Rating Scale; MADRS, Montgomery-Åsberg Depression Rating Scale; MDD, Major depressive disorder; MDE, major depressive episode; MRS, magnetic resonance spectroscopy; MTL, medial temporal lobe; PET, positron emission tomography; [123I]-ADAM, 123I-labelled ((2-((dimethylamino)methyl) phenyl)thio)-5-iodophenylamine; ROI, region of interest; SERT, serotonin transporter; SGACC, subgenual ACC; SSRI, selective serotonin re-uptake inhibitor; VLPFC, ventrolateral PFC; VMPFC, ventromedial PFC; VPFC, ventral PFC.
* BA not given, BA identified via Talairach Demon (if necessary after converting given coordinates from MNI to Talairach using MNI2Tal).
No identified studies examined CBT effects with structural MRI. Five applied fMRI, two PET with fluorine-18 labelled deoxyglucose, one PET with a 5HT1A receptor ligand, one SPECT with a serotonin transporter ligand, and one magnetic resonance spectroscopy (MRS). The PET studies measuring deoxyglucose levels examined the brain in the resting sate, whereas the fMRI studies all examined brain activity during processing of emotionally laden words or images. Many studies undertook whole brain analyses supplemented with targeted analyses focussing on specific regions, generally determined a priori. These regions were most commonly the amgygdala, hippocampus, cingulate and regions of the frontal cortex.
The study with arguably the highest quality design, having an antidepressant comparator group and randomly allocating patients to treatment, was Kennedy et al.’s PET study (Reference Kennedy, Konarski and Segal20). Twelve of 17 patients randomised to CBT completed treatment (15 having 12 sessions or more), seven fulfilling criteria for treatment response. By contrast, 12 of the 14 patients randomised to Venlafaxine were followed up, nine responding to treatment. Changes unique to CBT were decreased resting-state metabolism in the thalamus and posterior cingulate and increased metabolism in the left inferior temporal cortex, subgenual cingulate/ventromedial prefrontal cortex (VMPFC, BA32) and right occipital-temporal cortex (BA19). As not seen in antidepressant responders, it seems reasonable to assume that these changes are associated with CBT treatment rather than treatment response per se. Of course, changes seen with both CBT and antidepressant treatment could still theoretically be attributable to CBT, they would just represent common effects. Common effects were decreased metabolism bilaterally in the orbitofrontal cortex (OFC, BA11, 47) and in the left dorsomedial prefrontal cortex (DMPFC, BA8), and increased metabolism in the right inferior occipital cortex.
The study most directly comparable to Kennedy et al. is the PET resting state study of Goldapple et al., undertaken by the same group (Reference Goldapple, Segal and Garson21). They compared brain metabolism before and after treatment with 15–20 sessions of CBT, with post hoc comparison to paroxetine treatment. This study also reported decreased metabolism in various frontal regions with CBT. In common with Kennedy et al., decreased metabolism in the OFC/VLPFC (BA11, 47) was seen in both treatment groups. They also reported that reduced metabolism in the posterior cingulate was unique to CBT treatment. Unlike Kennedy et al. however they did not report CBT-associated increased subgenual ACC/VMPFC (BA32) activity or thalamic effects. Whereas Kennedy et al. reported increased metabolism in the (left) inferior temporal cortex (BA 20), Goldapple et al. reported CBT-associated decreased metabolism in this region. Goldapple et al. also reported CBT-associated increased metabolism in the hippocampus, which was reduced with paroxetine.
The four fMRI studies used different emotion-processing tasks, hampering comparison between them. Fu et al. utilised a task seeking to engage implicit processing of sad facial expressions (Reference Fu, Williams and Cleare28). They reported CBT treatment was associated with a significant decrease in right amygdala–hippocampal complex (AHC) activity during task. Conversely within task activity was increased following CBT treatment in regions including the ACC (BA24, BA32) and extending to the superior frontal gyrus (BA8), posterior cingulate (BA31), inferior parietal cortex (BA40), and precuneus (BA7). Though baseline right AHC activity was elevated in depressed compared to controls, treatment-associated change resulted in no significant difference between case and control within-task AHC activity at follow-up. The increase in ACC and posterior cingulate within-task activity following CBT treatment meant patient activity in these regions actually exceeded controls at the second scan.
The other three fMRI studies used explicit emotion processing tasks. Yoshimura et al. examined brain activity during processing of positive and negative emotional trait words (Reference Yoshimura, Okamoto and Onoda23). This was compared before and after treatment with 12 sessions of group CBT. Both change over time and comparison to controls at each time point was examined. Following treatment activation in the left ventral ACC (BA32), superior temporal cortex (BA39) and medial prefrontal cortex (MPFC, BA8) was increased when depressed patients considered if positive words described them, but decreased considering negative words.
In the uncontrolled study of Ritchey et al., participants were shown pictures designed to evoke positive, negative or neutral emotions and instructed to experience feelings or thoughts evoked and rate picture pleasantness (Reference Ritchey, Dolcos, Eddington, Strauman and Cabeza29). They undertook various contrasts and reported increased VMPFC activity post-CBT (BA11, so synonymous in this study with the OFC), when comparing all picture exposures vs. baseline activity (fixation cross). In the combined arousal contrast, they reported a larger difference in the positive and negative valence exposures versus neutral contrast post-treatment in the right amygdala, right caudate, and left hippocampus; that is a correction of the reduced increase in activity in this contrast in depressed (compared with controls) before CBT. The increased activity in the anterior temporal lobe to negative versus positive stimuli seen at baseline was reversed after treatment. After treatment activity in this region was (similarly to controls) greater on exposure to positive than negative stimuli.
Siegle et al. compared brain activity in 49 depressed patients and 35 healthy controls while they rated if positive, negative, or neutral words were personally relevant (Reference Siegle, Thompson and Collier26). The depressed group were scanned before and after at least seven sessions of individual CBT. The primary focus of this study was to examine if pre-treatment measures could predict response to CBT, but the impact of treatment on subgenual ACC (BA25) activity on exposure to negative words was specifically examined. Though they state there is no clear evidence CBT reduces this, they report that high post-treatment activity was more common in non-remitters and that three of five remitters with high pre-treatment subgenual ACC activity had decreased activity post-treatment.
The most recent study, by Sankar et al., compared brain activity of 16 depressed patients endorsing extreme responses to dysfunctional attitudes before and after 16 sessions of CBT to 16 healthy controls (Reference Sankar, Scott, Paszkiewicz, Giampietro, Steiner and Fu30). Within the patient group task-related activation in right posterior cingulate gyrus (BA 37) decreased following CBT (p<0.01). Activation in the left parahippocampal gyrus (BA 37) also decreased, but this reduction was less than that observed in the control group.
Amsterdam et al. used [123I]-ADAM single photon emission computed tomography (SPECT) to compare serotonin transporter (SERT) binding in 20 depressed patients before and after CBT to that in 10 untreated healthy controls (Reference Amsterdam, Newberg, Newman, Shults, Wintering and Soeller27). Depressed subjects demonstrated low pretreatment mean SERT standardised uptake ratios, speculated to reflect low brain serotonin levels, which significantly increased over time in the midbrain (p=0.011), right medial temporal lobe (p=0.008), and left medial temporal lobe (p=0.000) regions. Tiger et al. demonstrated reduced binding potential of a 5-HT1B receptor selective radioligand in the dorsal brain stem after treatment with internet-based CBT (Reference Tiger, Rück and Forsberg24). Sanacora et al. undertook the only study examining the effects of CBT using MRS (Reference Sanacora, Fenton and Fasula22). They examined resting state occipital cortex gamma-aminobutyric acid (GABA) concentrations before and after 12 sessions of CBT. Changes in the eight patients completing treatment were compared to those in two previous studies examining the effects of selective serotonin re-uptake inhibitors (SSRIs) and electroconvulsive therapy (ECT). There was a significant decrease in HDRS scores in the CBT treated patients, but whereas an increase in occipital GABA concentrations was observed in the earlier two studies this was not seen with CBT. While the authors acknowledge that small sample size means the study is underpowered to definitively state that GABA content is unchanged after CBT, they suggest CBT may have different effects on GABA content than SSRIs or ECT.
Discussion
Overall, changes in ACC activity following CBT are most consistently reported. This region is often defined as BA 32 (dorsal anterior cingulate), but also includes BA24 (ventral ACC) and BA33 (pregenual area). In one of the highest quality studies, that of Kennedy et al., decreased resting state activity in BA32 after CBT was reported (Reference Kennedy, Konarski and Segal20). The other PET resting state study reported that activity in adjoining cingulate regions in BA24 were increased after CBT, though they describe this region as more dorsal midcingulate than ACC proper (Reference Goldapple, Segal and Garson21). Yoshimura et al. reported that following CBT activity in the ACC (BA32) was decreased when considering if negative words described them (but increased on considering positive words) (Reference Yoshimura, Okamoto and Onoda23), and Fu et al. that it was increased (in a region encompassing BA24 and BA32) during implicit processing of negative facial expressions (Reference Fu, Williams and Cleare28). Siegle et al.’s 2012 study focused on the adjoining subgenal ACC (BA25) (Reference Siegle, Thompson and Collier26). Though they acknowledge no clear evidence that subgenual ACC activity reduces with treatment during processing of negative words, high post-treatment activity was more common in non-remitters. Kennedy et al. and Goldapple et al. compared CBT-associated changes to those seen with antidepressants; in both studies ACC changes were unique to CBT (Reference Kennedy, Konarski and Segal20,Reference Goldapple, Segal and Garson21).
After the ACC, the regions most commonly reported to exhibit change in activity following CBT were the OFC/VLPFC (BA11, 47), posterior cingulate (BA 30, 31), and amygdala and/or hippocampus. This is summarised in Fig. 1 Kennedy et al. and Goldapple et al. both reported decreased resting state metabolism in BA11/47 (Reference Kennedy, Konarski and Segal20,Reference Goldapple, Segal and Garson21), also seen after antidepressant treatment. Ritchey et al. however reported a greater increase in activity in this region during the task of rating picture pleasantness versus baseline state following CBT treatment (Reference Ritchey, Dolcos, Eddington, Strauman and Cabeza29); this is not necessarily incompatible with the resting state findings, and could potentially be explained by reduced baseline activity in the region. Four studies also reported changes in posterior cingulate (BA31) activity following CBT. The two resting state studies both reported it reduced (and this was unique to CBT), though Kennedy et al. localised the activity change to BA29 (Reference Kennedy, Konarski and Segal20,Reference Goldapple, Segal and Garson21). Fu et al. reported decreased activity after CBT in BA23, BA30 and BA31 (all include the posterior cingulate) during implicit processing of sad facial expressions (Reference Fu, Williams and Cleare28). Sankar et al. reported decreased activation in right posterior cingulate gyrus (BA30) during extreme attributions from the DAS (Reference Sankar, Scott, Paszkiewicz, Giampietro, Steiner and Fu30). Changes in amygdala or hippocampal activity following CBT were seen in three studies (Reference Goldapple, Segal and Garson21,Reference Fu, Williams and Cleare28,Reference Ritchey, Dolcos, Eddington, Strauman and Cabeza29). Activity in the hippocampus was increased in Goldapple et al.’s resting state study, decreased in the amygdala (extending to hippocampus) during implicit processing of sad facial expression in Fu et al.’s fMRI study, and increased in the right amygdala and left hippocampus (in the combined contrast of positive and negative valence pictures vs. neutral) in Ritchey et al.’s study (Reference Goldapple, Segal and Garson21,Reference Fu, Williams and Cleare28,Reference Ritchey, Dolcos, Eddington, Strauman and Cabeza29). Though specifically examined, Kennedy et al. and Sankar et al. did not report any change in amygdala activity following CBT (Reference Kennedy, Konarski and Segal20,Reference Sankar, Scott, Paszkiewicz, Giampietro, Steiner and Fu30).
Regions most commonly identified as exhibiting change in activity following treatment of depression with cognitive behavioural therapy.
Two fMRI studies reported post-CBT increases in superior frontal gyrus/DMPFC (BA8) activity when processing emotional stimuli, though Yoshimura et al. found it increased on judging whether positive emotional trait words applied to them and Fu et al. increased on processing sad facial expressions (Reference Yoshimura, Okamoto and Onoda23,Reference Fu, Williams and Cleare28). Activity in this region was decreased in Kennedy et al.’s (Reference Kennedy, Konarski and Segal20) resting state study (Reference Yoshimura, Okamoto and Onoda23,Reference Fu, Williams and Cleare28). Kennedy et al. and Goldapple et al. reported changes in opposite directions in the inferior temporal cortex (BA20) (Reference Kennedy, Konarski and Segal20), while Yoshimura et al., Fu et al. and Ritchey et al. reported changes in other temporal cortex regions (Reference Yoshimura, Okamoto and Onoda23,Reference Fu, Williams and Cleare28,Reference Ritchey, Dolcos, Eddington, Strauman and Cabeza29).
Limitations and themes arising
The limitations of this review arise predominantly from the relatively small numbers of studies identified as eligible for inclusion and the striking heterogeneity of these studies, a fact which precluded any meaningful meta-analytic synthesis of findings. Studies differ in sample characteristics (e.g. proportion experiencing first depressive episode vs. recurrent illness), amount of therapy, regions chosen for reduced threshold analyses, scanner resolution and nature/existence of comparator group. Possibly even more fundamental differences are neuroimaging technique employed, and whether resting state or task-dependent brain activity is examined. Considering imaging technique, PET with fluorine-18 labelled deoxyglucose measures glucose metabolism whereas fMRI measures deoxyhaemoglobin concentration. Given that these are different physiological parameters, and the modalities differ in temporal (PET acquires a single scan over 60–90s, fMRI over 1–2s), and spatial (2 mm with fMRI, but generally less with PET) resolution (Reference Turner and Friston31), one would expect findings from the two modalities could differ considerably. Whether resting state or task-related activity is examined (and in the latter the nature of the task used), could also lead to apparently divergent findings. Expanding on the latter point, it has been established that individuals with major depression show amygdala hyperactivity when processing emotionally negative information (see Introduction). If CBT addresses this information processing bias, then post-CBT amygdala activity may be expected to be decreased on exposure to sad faces, but unchanged (or even increased) with happy faces. Other stimuli characteristics, for example whether pictorial or linguistic, could also influence brain region activated and/or magnitude of associated brain activity. It is particularly striking that no studies used structural imaging approaches. Given that CBT-associated structural brain changes have been identified in treatment of other conditions, this is a clear gap in the research data.
Given the level of study heterogeneity, the degree of consistency in brain regions identified as exhibiting activity change following CBT is notable. First, ACC activity does seem decreased following CBT, both in the resting state and when processing negative words. It seems however increased when processing positive words and (possibly more unexpectedly) negative facial expressions. Posterior cingulate activity was reduced in both resting state studies, as well as when processing sad facial expressions and endorsing extreme responses to dysfunctional attitudes. Resting state activity in OFC/VLPFC also seems decreased following CBT, though the task-associated increase in activity observed rating picture pleasantness may be greater following CBT. There seem different CBT-associated changes in the amygdala and hippocampus. Though activity in the former may be decreased following CBT (both resting state and when processing sad facial expressions), resting state hippocampal activity is reported increased. The attenuation of left parahippocampal activity observed on follow-up scanning of healthy controls endorsing extreme dysfunctional attitudes is reduced in depressed patients treated with CBT.
What do these findings tell us about the impact of CBT on brain function?
The regions exhibiting activity changes following CBT are all parts of the limbic system and functionally related neocortical structures. This is unsurprising given the centrality of these regions to emotional experience and memory, abnormalities of which underpin depression. The cingulate, regions of frontal cortex, and the amygdala/hippocampus are crucial parts of a fronto-limbic network central to these functions, and numerous studies have identified functional (and some structural) abnormalities of these regions in depressed people (Reference Sacher, Neumann, Fünfstück, Soliman, Villringer and Schroeter32).
In studies including a healthy control group there is a tendency for CBT-associated changes to ‘normalise’ brain function, that is for functional imaging findings to be more like those in healthy individuals. This is reported in both resting state PET (Reference Yoshimura, Okamoto and Onoda23) and task-dependent fMRI (Reference Fu, Williams and Cleare28) studies. Intriguingly however, some studies examining how antidepressants work suggested this is not actually as simple as ‘normalising’ the depressed brain, with compensatory changes also important (Reference Willner, Scheel-Krüger and Belzung1). In short, rather than restoring brain function to that seen in controls, treatment brings about changes in brain function which compensate for the abnormalities giving rise to depression. In-keeping with this, the rapid (and potentially transient) effects of ECT (Reference Shapira and Lerer33) argue against it working through (presumably protracted) processes of neurogenesis and receptor synthesis. In studies explicitly comparing the effects of CBT and pharmacotherapy, though some common effects are seen (such as decreased ventral prefrontal cortex metabolism), other changes are frankly divergent. Opposite effects with each treatment are reported in the DLPFC and inferior parietal cortex by Goldapple et al. (Reference Goldapple, Segal and Garson21), and posterior cingulate and left inferior temporal cortex by Kennedy et al. (Reference Kennedy, Konarski and Segal20). It is thus conceivable that the relative degree to which each modality brings about remission through ‘normalisation’ of brain functional abnormalities versus ‘compensatory’ processes differs.
So what is CBT actually doing?
Previous writers have incorporated imaging findings into a parsimonious account of how CBT and pharmacotherapy work, and how their effects differ (Reference Shapira and Lerer33). As discussed above, functional imaging studies suggest CBT brings about changes in fronto-limbic systems, potentially normalising abnormal activity. Writers have commented on how functional imaging studies of depression particularly implicate change in regions regarded as crucial components of a well-established model of information processing in the forebrain (Reference Willner, Scheel-Krüger and Belzung1). This model recognises two distinct but interacting systems: a ventral ‘affective’ circuit involving the amygdala, anterior hippocampus, ventral striatum, insular cortex, ventral (subgenual) part of the ACC and ventral and orbital PFC; and a dorsal ‘cognitive’ circuit, involving the hippocampus, dorsal (pregenual) part of the ACC and dorsolateral PFC ((Reference Willner, Scheel-Krüger and Belzung1,Reference Mayberg, Liotti and Brannan34). It is proposed the ventral system is important for identification of the emotional significance of a stimulus and production of affective states and autonomic regulation related to emotionally significant situations, while the dorsal system is important for executive function, including selective attention, planning, and effortful regulation of affective states; essentially the ‘cognitive control’ functions described by Miller and Cohen (Reference Miller and Cohen10). DeRubeis et al. incorporated an understanding of such distinct (but intertwined) circuits into a proposed hypothesis of how CBT works. They emphasised differences from antidepressant medication effects by suggesting the former works in a ‘top-down’ manner while the latter works in a ‘bottom-up’ way (Reference Shapira and Lerer33). In essence, that CBT might allow a resetting of tonic prefrontal activity to yield greater capacity for ‘top-down’ emotion regulation when it is needed (such as when skills taught in CBT are engaged); conversely antidepressants might increase subcortical cingulate metabolism tonically, creating a ‘bottom-up’ effect whereby relevant limbic regions are inhibited during medication administration. This is of course not incompatible with a model of both modalities having ‘normalising’ and ‘compensatory’ effects. To speculate, CBT could normalise or lead to compensatory changes in the dorsal network which subsequently impact on subcortical regions, whereas pharmacotherapy could conversely promote normalising or compensatory subcortical changes which over time normalise cortical abnormalities.
This model for understanding the mechanism of action of CBT in depression has much appeal. It is parsimonious, but also fits with the idea that the benefits experienced with CBT treatment derive from cognitive restructuring and improved ability to rationally appraise automatic thoughts; essentially bolstered cognitive control of emotionally salient automatic thoughts. The reality is however that the reviewed data provide at best only partial support for this model. In the case of the DLPFC for example, a brain region believed to play a crucial role in cognitive control, changes in activity post-CBT were only seen in the study of Goldapple et al. (Reference Goldapple, Segal and Garson21). By contrast changes in DLPFC have been reported following successful treatment with antidepressants, ECT and even placebo (Reference Willner, Scheel-Krüger and Belzung1). CBT-associated changes were also reported in several of the reviewed studies in the OFC and amygdala, these of course being parts of the proposed ‘affective’ circuit, function of which is purportedly influenced by medication rather than CBT. Decreased resting state activity in the dorsal ACC (cognitive circuit) and increased resting state activity in the ventral ACC (affective circuit) following CBT (Reference Kennedy, Konarski and Segal20,Reference Goldapple, Segal and Garson21) can be explained by CBT resetting tonic prefrontal activity (i.e. reduced resting state activity) resulting in greater capacity for ‘top-down’ emotion regulation when needed. If they are ‘opposing’ networks however, why task-related activity in both the dorsal and ventral ACC would be increased during the processing of negative facial expressions is much harder to explain within this model (Reference Amsterdam, Newberg, Newman, Shults, Wintering and Soeller27).
It may be that some of the apparent inconsistencies discussed can be explained by factors such as heterogeneity in study design, inconsistency in region definition and the limited spatial resolution inherent in current functional imaging methodologies. Even accounting for these factors however, the reality is that the existing imaging data are not fully supportive of the existing models purporting to explain how CBT brings about its effects. Clarification of this will clearly require further study, with inclusion of appropriate comparator groups essential if effects specific to treatment with CBT are to be identified. Design of such studies will be challenging, separating the effects specific to CBT from those of spontaneous recovery or the non-specific effects of being in a therapeutic relationship being hampered by the ethical implications of delaying treatment in a suffering and potentially high-risk group. The question of whether CBT and antidepressants bring about recovery through distinguishable mechanisms is intriguing and potentially informative. As suggested above however, it is also conceivable that even though these treatment modalities do have different neurobiological mechanisms of action early in treatment, as recovery progresses these differences diminish. Consequently, if the effects of the two treatments modalities are to be distinguished, this may necessitate repeat scans during treatment, following remission, and beyond. As it is thought that patients treated with CBT are less vulnerable to relapse than those who were treated with antidepressants (Reference Hollon, DeRubeis and Shelton35), it would be particularly interesting to examine if differences in brain function are detectable post-remission. Could differences specific to CBT treatment be particularly crucial to protection from relapse?
In conclusion, though limited in number and variable in methodology, the reviewed studies do suggest that CBT is associated with functional brain changes. Whether it results in structural brain changes, which has been demonstrated when used to treat conditions such as chronic fatigue syndrome, remains unexplored. The data summarised demonstrate CBT is having ‘biological’ effects. Further work, applying the expanding variety of imaging methodologies of increasing sophistication and resolution, may enable elucidation of the effects of CBT in ever more localised brain regions. This could expand our understanding of how all treatment modalities actually work, and potentially elucidate if specific psychological treatments do indeed have unique effects. Further randomised studies certainly seem essential in disentangling if antidepressant and psychological treatments have distinct modes of action, and characterising what these specific effects are.
Acknowledgement
Authors’ Contributions: Killian Welch devised the study, updated the systematic review, and revised the manuscript. George Franklin undertook the original systematic review and produced an early draft of the manuscript. Alan Carson revised the manuscript, including input to data presentation.
Financial Support
No funding was provided for this study
Conflicts of Interest
There are no conflicts of interest.
