Association of Brain Cortical Changes With Relapse in Patients With Major Depressive Disorder

Key Points

Question  Is relapse in major depressive disorder associated with morphologic brain changes?

Findings  In this longitudinal, case-control, magnetic resonance imaging study, patients with major depressive disorder showed different trajectories of cortical gray matter changes depending on whether they experienced a relapse during the follow-up interval. These changes were not significantly associated with psychiatric medication or with severity of depression at follow-up.

Meaning  Relapse in major depressive disorder is associated with morphologic changes in brain regions that are crucial for regulation of emotions and cognitive control.

Importance  More than half of all patients with major depressive disorder (MDD) experience a relapse within 2 years after recovery. It is unclear how relapse affects brain morphologic features during the course of MDD.

Objective  To use structural magnetic resonance imaging to identify morphologic brain changes associated with relapse in MDD.

Design, Setting, and Participants  In this longitudinal case-control study, patients with acute MDD at baseline and healthy controls were recruited from the University of Münster Department of Psychiatry from March 21, 2010, to November 14, 2014, and were reassessed from November 11, 2012, to October 28, 2016. Depending on patients’ course of illness during follow-up, they were subdivided into groups of patients with and without relapse. Whole-brain gray matter volume and cortical thickness of the anterior cingulate cortex, orbitofrontal cortex, middle frontal gyrus, and insula were assessed via 3-T magnetic resonance imaging at baseline and 2 years later.

Main Outcomes and Measures  Gray matter was analyzed via group (no relapse, relapse, and healthy controls) by time (baseline and follow-up) analysis of covariance, controlling for age and total intracranial volume. Confounding factors of medication and depression severity were assessed.

Results  This study included 37 patients with MDD and a relapse (19 women and 18 men; mean [SD] age, 37.0 [12.7] years), 23 patients with MDD and without relapse (13 women and 10 men; mean [SD] age, 32.5 [10.5] years), and 54 age- and sex-matched healthy controls (24 women and 30 men; mean [SD] age, 37.5 [8.7] years). A significant group-by-time interaction controlling for age and total intracranial volume revealed that patients with relapse showed a significant decline of insular volume (difference, −0.032; 95% CI, −0.063 to −0.002; P = .04) and dorsolateral prefrontal volume (difference, −0.079; 95% CI, −0.113 to −0.045; P < .001) from baseline to follow-up. In patients without relapse, gray matter volume in these regions did not change significantly (insula: difference, 0.027; 95% CI, −0.012 to 0.066; P = .17; and dorsolateral prefrontal volume: difference, 0.023; 95% CI, −0.020 to 0.066; P = .30). Volume changes were not correlated with psychiatric medication or with severity of depression at follow-up. Additional analysis of cortical thickness showed an increase in the anterior cingulate cortex (difference, 0.073 mm; 95% CI, 0.023-0.123 mm; P = .005) and orbitofrontal cortex (difference, 0.089 mm; 95% CI, 0.032-0.147 mm; P = .003) from baseline to follow-up in patients without relapse.

Conclusion and Relevance  A distinct association of relapse in MDD with brain morphologic features was revealed using a longitudinal design. Relapse is associated with brain structures that are crucial for regulation of emotions and thus needs to be prevented. This study might be a step to guide future prognosis and maintenance treatment in patients with recurrent MDD.

Less than half of all patients with major depressive disorder (MDD) remain free of symptoms for 2 years after recovery.¹ Rates of relapse increase markedly with subsequent episodes, which leads to a continuous high risk of chronicity and negative psychosocial consequences among these patients.^2,3 Structural neuroimaging techniques may provide insights into the underlying neural mechanisms of relapse. In the long term, understanding the neuronal correlates could support prognosis of relapse and thus improve maintenance treatment in patients with recurrent MDD.

Major depressive disorder is associated with morphologic changes in specific brain areas, most prominently the hippocampus,^4-7 insula,^8-11 and the prefrontal and orbitofrontal cortex (OFC).^6,11-14 Cross-sectional studies have repeatedly reported that these brain alterations are strongly influenced by the course of illness (eg, number of recurrent episodes,^8,15 duration of untreated depression,¹⁶ and early age of depression onset¹⁷). These findings could indicate that depression causes damage to the brain that evolves over time and is most obvious in individuals with progressive and recurrent MDD.¹⁸ However, this conclusion must be treated with caution, since only correlative statements can be drawn from cross-sectional studies.

Despite intense research on the clinical correlates of morphologic brain alterations in patients with MDD, longitudinal studies in well-characterized samples are needed to gauge neurobiological changes due to relapse. So far, most longitudinal studies have investigated morphologic changes in dependence of remission status at follow-up assessment. Several studies found that patients whose MDD was not in remission at follow-up showed significantly more volume decline compared with patients whose MDD was in remission (eg, in the hippocampus, anterior cingulate cortex [ACC], and dorsolateral prefrontal cortex [DLPFC]).^19-22 Other studies showed normalized volume and thickness in the hippocampus, rostral middle frontal gyrus (MFG), and OFC in patients who achieved remission.^23,24 However, to evaluate the association between brain alterations and relapse, it is crucial to consider the course of illness between scans, which has been neglected by most studies. To our knowledge, only 1 study reported an association of gray matter reductions and the number of relapses in a 7-year follow-up,²⁵ but control for confounding variables (eg, medication) was lacking in the statistical analyses. Furthermore, most results are limited to geriatric populations, and younger samples are required for generalizability of these findings.^21-23,26-30 Most previous studies used restricted region-of-interest analyses rather than whole-brain approaches, which a priori limited the outcomes to a subset of regions of the brain.³¹

To enhance our understanding of structural brain alterations in patients with recurrent MDD, the aim of our study was to investigate morphologic changes in patients as a function of relapse during the follow-up interval using both whole-brain voxel-based morphometry (VBM) and region-of-interest analysis by FreeSurfer (https://surfer.nmr.mgh.harvard.edu). We hypothesized that patients with at least 1 relapse between scans show more pronounced reductions in gray matter than patients who achieved stable remission during the follow-up interval. Severity of depression at follow-up was controlled for to investigate whether potential brain changes merely reflect state-dependent differences owing to remission status, or rather result from relapse. Our longitudinal design also implemented correlational control for confounding effects of psychiatric medication.

The study comprised 64 patients with MDD and 59 matched healthy controls who were recruited from the University of Münster Department of Psychiatry from March 21, 2010, to November 14, 2014, and were reassessed from November 11, 2012, to October 28, 2016. See eAppendix 1 in the Supplement for further information on recruitment and exclusion criteria. All participants underwent structural magnetic resonance imaging at baseline and approximately 2 years later (mean [SD], 2.3 [0.3] years). Following data quality checks (eAppendix 1 in the Supplement), the final sample consisted of 60 patients and 54 healthy controls. Participants received financial compensation. The study was approved by the institutional review board of the medical faculty of the University of Münster and all participants provided written informed consent.

Participants received diagnoses according to the Structured Clinical Interview for DSM-IV³² at both time points. At baseline, medical records were consulted to validate the diagnosis and duration of the index episode of MDD. We used structured interviews to retrospectively assess the type and dose of psychopharmacologic treatment at baseline and follow-up (Table 1), as well as the days between scans without medication. We used this information to compute a composite medication index (eAppendix 1 in the Supplement). The Hamilton Depression Rating Scale³³ was applied to capture the severity of symptoms at both time points. Furthermore, we assessed familial risk for MDD and childhood maltreatment with the Childhood Trauma Questionnaire³⁴ as known risk factors for both MDD and morphologic brain changes.^35-37

All patients were experiencing a moderate or severe depressive episode at baseline, resulting in inpatient treatment. Depending on the course of illness between scans, we divided patients into 2 groups: patients with no subsequent episodes (no relapse; n = 23), and patients who experienced at least 1 additional episode (relapse; n = 37). We applied DSM-IV criteria using the Structured Clinical Interview for DSM-IV for assessing relapse during the follow-up interval and determining full remission (absence of symptoms for at least 2 months), which was required before diagnosing a new episode. Given the limited number of possible episodes in a 2-year interval, and the insufficient number of patients with more than 1 episode, we refrained from assessing the parametric effects of the number of relapses.

At baseline, patient groups did not differ significantly in clinical and sociodemographic variables. However, patients with relapse had significantly higher Hamilton Depression Rating Scale scores at follow-up than did patients without relapse. All 3 groups did not differ significantly in age, sex, educational level, and time between scans (Table 2). For further information on sample matching, see eAppendix 1 in the Supplement.

The T1 images were acquired using a 3-T scanner. For a detailed description of data acquisition, see eAppendix 1 in the Supplement.

Gray matter information was analyzed using the default longitudinal preprocessing pipeline of the Computational Anatomy Toolbox (http://www.neuro.uni-jena.de/cat, v933). Processing steps included intraparticipant realignment of baseline and follow-up images to a mean image for each participant, to evaluate within-participant dependency of images over time. Following bias correction, spatial normalization parameters were estimated based on the mean image using high-dimensional Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) normalization, and applied to baseline and follow-up images. Images were segmented into gray matter, white matter, and cerebrospinal fluid. Gray matter images were smoothed with an 8-mm full-width at half maximum Gaussian kernel. For details on data quality checks, see eAppendix 1 in the Supplement.

Based on recent studies reporting cortical alterations in patients with MDD,^11,24 we selected the insula, medial OFC, rostral ACC, and rostral middle frontal gyrus as regions of interest for additional analysis of cortical thickness. The rostral middle frontal gyrus includes Brodmann area 46 and was therefore chosen as a representative of the DLPFC.³⁸ Segmentation of all regions was based on the Desikan-Killiany atlas.³⁹ To extract thickness estimates, images were automatically processed with the longitudinal stream⁴⁰ in FreeSurfer, version 5.3. See eAppendix 1 in the Supplement for further information on data quality checks.

Statistical analyses of VBM volumes were performed in SPM12, version 6685 (Wellcome Department of Cognitive Neurology; http://www.fil.ion.ucl.ac.uk/spm), using an absolute threshold masking of 0.2. Surface-based morphometry (SBM) measures of cortical thickness were analyzed in SPSS Statistics for Windows, version 24.0 (IBM Corp).

For both VBM and SBM analyses, a 3 × 2 analysis of covariance (ANCOVA) with group (no relapse, relapse, and healthy controls) as between-participant factor and time (baseline and follow-up) as within-participant factor was computed. Age and total intracranial volume (for VBM only) were included as covariates. We performed F tests for the main effects of group and the group-by-time interaction. Bonferroni-corrected pairwise comparisons were calculated to further investigate significant effects. For VBM volumes, pairwise comparisons were computed on extracted contrast values of the significant clusters in SPSS.

Anatomical labeling of significant VBM clusters was performed by the automated anatomical labeling atlas⁴¹ implemented in the Wake Forest University pickatlas.⁴² We used a rigorous statistical threshold of P < .05 (2-sided), familywise error corrected on the voxel level.⁴³ Exploratory results are also reported at a statistical threshold of P < .001, uncorrected, and a cluster threshold of k = 300 voxels.

For the patient subsamples, VBM gray matter volume differences were computed by subtracting contrast values extracted at baseline from those at follow-up. We performed correlation analyses to investigate potentially confounding effects of medication and symptom severity at follow-up on gray matter volumes and on gray matter volume differences.

The study included 60 participants with MDD: 23 without relapse (13 women and 10 men; mean [SD] age, 32.5 [10.5] years) and 37 with relapse during the follow-up interval (19 women and 18 men; mean [SD] age, 37.0 [12.7] years). A total of 54 age- and sex-matched healthy controls (24 women and 30 men; mean [SD] age, 37.5 [8.7] years) were also included.

The ANCOVA showed a significant interaction of group by time when controlling for age and total intracranial volume, which comprised clusters in the right insula and the right DLPFC (Table 3 and Figure). Exploratory analyses revealed an additional interaction in the ACC and in the middle and superior temporal gyrus (eTable 1 and eFigure in the Supplement).

In the insula, pairwise comparisons on the extracted cluster values revealed a significant increase in volume between baseline and follow-up in healthy controls (difference, 0.060; 95% CI, 0.034-0.085; P < .001) and a significant decrease in volume between baseline and follow-up in patients with relapse (difference, −0.032; 95% CI, −0.063 to −0.002; P = .04), whereas there was no significant change in patients without relapse (difference, 0.027; 95% CI, −0.012 to 0.066; P = .17) (Table 4 and Figure). In the DLPFC, there was a significant decrease in volume between baseline and follow-up in patients with relapse (difference, −0.079; 95% CI, −0.113 to −0.045; P < .001) but no significant change in patients without relapse (difference, 0.023; 95% CI, −0.020 to 0.066; P = .30) or healthy controls (difference, 0.010; 95% CI, −0.018 to 0.038; P = .48) (Table 4 and Figure). The pattern of interaction effects remained the same if the ANCOVA was repeated considering only patient subgroups (eAppendix 2 in the Supplement). For the patient subgroups, neither symptom severity nor medication load index were significantly correlated with gray matter volume differences or gray matter volumes at follow-up (eTable 2 in the Supplement).

At baseline, patients with relapse between scans had significantly higher gray matter volume compared with patients without relapse (only in the DLPFC) and healthy controls (both in the DLPFC and insula). Details on the main effect of group and group differences at baseline and follow-up can be found in eAppendix 2 in the Supplement.

The ANCOVA showed a significant group-by-time interaction for cortical thickness of the left medial OFC and left rostral ACC (eTable 3 in the Supplement). For both regions, patients without relapse had a significant increase of cortical thickness (ACC: difference, 0.073 mm; 95% CI, 0.023-0.123 mm; P = .005; OFC: difference, 0.089 mm; 95% CI, 0.032-0.147 mm; P = .003), as did healthy controls (ACC: difference, 0.087 mm; 95% CI, 0.056-0.118 mm; P < .001; OFC: difference, 0.048 mm; 95% CI, 0.013-0.083 mm; P = .007), whereas there was no significant change in patients with relapse (ACC: difference, 0.003 mm; 95% CI, –0.035 to 0.041 mm; P = .86; OFC: difference, –0.006 mm; 95% CI, –0.047 to 0.036 mm; P = .79). Although the group-by-time interaction failed to reach significance for the right insula, pairwise comparisons revealed the same pattern of change. In the right rostral middle frontal gyrus, we observed a decrease of cortical thickness in patients with relapse whereas patients without relapse and controls did not show a significant change. Results from pairwise comparisons of longitudinal effects within groups are presented in Table 4.

Distinct associations of relapse in MDD with brain morphologic features were revealed using a longitudinal magnetic resonance imaging design and a combined VBM and SBM approach. Patients who experienced at least 1 relapse between scans showed a decline of insular volume and DLPFC volume and thickness from baseline to follow-up. In patients without relapse, gray matter volume in these regions was stable, whereas cortical thickness increased in the left medial OFC, left rostral ACC, and right insula. Volume changes from baseline to follow-up were not associated with psychiatric medication or with severity of depression at follow-up. Together, these results highlight the effects of relapse on morphologic brain alterations in patients with MDD.

The observed neuroplastic changes in the insula, ACC, and prefrontal cortex are in line with neurobiological models of MDD that assume a dysfunction of the frontolimbic brain circuitry.⁴⁴ Cross-sectional studies often report changes of insular volume and thickness in patients with MDD.^9-11,45,46 Our observed effect of relapse on insular morphologic features is supported by previous studies showing an association of insular volume change over time and patients’ course of illness, such as duration of illness^9,10,47 and number of depressive episodes.²⁵ Previous studies assume that regional volume changes are associated with changes in glial cell density, neuronal size, and dendritic branching,^48-50 which are known to influence the function of neural circuitries in these regions.⁵¹ Although, to our knowledge, the direct link between morphologic and functional alterations has not been investigated, we hypothesize that insular atrophy might contribute to a selective mood-congruent activation toward emotional stimuli⁵² and use of suppressive emotion regulation strategies.⁵³

Meta-analyses frequently report gray matter alterations in the DLPFC and OFC of patients with MDD.⁵⁴ Interestingly, reductions of left medial OFC thickness were most pronounced in patients with recurrent MDD.¹¹ Thus, relapse seems to affect gray matter decline, as already indicated by a study showing less progressed decrease in subgroups with remission of MDD after 3 years¹⁹ and findings on opposing trajectories of DLPFC thickness in patients with remission vs those without remission.²⁴ On the functional level, magnetic resonance imaging studies on the prefrontal cortex might point at a deficit in cognitive control and emotion regulation.^55-57

The rostral ACC has been reported to be decreased in individuals with MDD, especially with multiple depressive episodes.^11,12 Our results are supported by findings on less decrease of the left ACC in patients without remission compared with those with remission.¹⁹ Changes in ACC volume have been shown to be correlated with faster rates of symptom improvement,⁵⁸ which fits with our observation that an increase in ACC thickness might be associated with a beneficial course of illness. Taken together, our present findings on insular, ACC, and prefrontal atrophy in patients with recurrent MDD shed light on the development of a bottom-up processing bias of emotional stimuli (eg, in limbic brain areas) and a disruption of top-down executive functions (eg, in the DLPFC or OFC and ACC).

For both the DLPFC and the insula, we found only right-lateralized effects. These findings would be consistent with theories on a right-hemispheric specialization to aversive emotion processing and regulation.⁵⁹ However, complementary analyses with FreeSurfer showed left-lateralized effects in the medial OFC and rostral ACC. Thus, laterality effects could not be supported, possibly owing to differences in preprocessing and segmentation. Future studies need to directly investigate hemispheric lateralization.

Previous longitudinal studies on morphologic changes in MDD provided no information on the course of illness between scans^19-22 but mainly reported associations with remission status, including current depressive symptoms at follow-up. Therefore, these studies could not disentangle the effects of remission status from differing courses of depression. The morphologic alterations observed when contrasting patients with and patients without remission might thus be better explained by differences in depressive episodes between scans and less by state differences. Regional morphologic changes of the cortex as a putative cumulative effect of illness duration and relapse are supported by studies that elucidate progressive neural changes in association with recurrent episodes.^60-63 We can only partly replicate previous findings on a gray matter volume increase in patients with remission^23,24 because there was only a tendency for an increase of cortical thickness in patients without relapse. Future studies should clarify whether there is a normalization of volume in patients without relapse, or whether an increase could be better explained by confounding variables exerting neuroprotective effects (eg, medication or psychotherapy).

In our study, we observed larger baseline DLPFC volumes in patients with relapse than in patients without relapse, although patient groups did not differ in important confounding variables at baseline (eg, medication, depression severity, childhood maltreatment, and familial risk). Volume differences in the DLPFC could be explained by other variables that have not been assessed in this study. For example, environmental factors other than childhood maltreatment, such as urbanicity, could be associated with reduced volume and cortical thickness in the DLPFC.^64,65 Furthermore, unknown genetic factors contributing to both susceptibility for relapse and brain structure could be responsible for this finding. Alternatively, a larger DLPFC volume could be a correlate of increased disease vulnerability or could reflect a compensatory mechanism in individuals who are prone to relapse. Furthermore, our study might be underpowered to detect reliable cross-sectional associations owing to the small sample size. However, longitudinal studies require far fewer participants than do cross-sectional studies to detect small morphologic changes in the brain.⁶⁶

We found an increase of insular volume and thickness in controls. Research on normal brain changes across the lifetime in control populations describes a trend of decrease in volume after the age of 30 years.^67-70 However, an age-related decrease in volume has been derived mostly from cross-sectional studies, which could be confounded by cohort effects (eg, secular changes in nutrition, medical care, or lifestyle).⁷¹ Furthermore, high and constant reliability of neuroimaging measures was demonstrated, both in our study (eAppendix 1 in the Supplement) and in other samples.⁷² Finally, findings were independent of the applied method, and there were no changes to the scanner setting that could have confounded our findings.

Surface-based morphometry and VBM partly revealed different results for the direction of morphologic trajectories. Because VBM incorporates information from multiple morphologic characteristics, such as cortical thickness, folding, gyrification, and volume, this might explain the inconsistencies in the results. Previous studies that used a multimethods approach showed a regionally dependent contribution of different surface properties to the VBM signal.⁷³ Furthermore, comparability between methods is limited owing to differences in delineation of anatomical regions and use of different atlases for labeling. Despite these methodological differences, significant associations between VBM and SBM measures could be reported in previous studies,^74,75 which indicates concordance of methods.

Our study has several strengths and methodological weaknesses. To our knowledge, this is the first longitudinal study that directly addressed the association of relapse with morphologic brain alterations in middle-aged patients with MDD. Owing to adaptive neural processes in MDD, it seems of great importance to assess the course of depression for a given time interval, rather than the remission status at a single time point.

Furthermore, we used a multimethodological approach with SBM and VBM. Both methods are optimized for longitudinal designs, instead of nonblinded manual segmentation. Manual methods are subject to errors in both accuracy and reproducibility (eg, through subjectivity in structure interpretation).⁷⁶

We implemented control for important confounding variables (eg, age, depressive episodes, medication, severity of depression, childhood maltreatment, and familial risk). However, there might be other confounding variables (eg, psychotherapy, social support, urbanicity, and duration of depressive episodes) that should be addressed by future studies. Finally, our study is limited by the retrospective assessment of the course of illness during the follow-up interval based on patients’ self-disclosure, because self-reported clinical measures often have low accuracy owing to recollection biases in MDD.⁷⁷

Our study revealed distinct effects of relapse in patients with MDD using a longitudinal design and a multimethodological approach. Observed cortical changes from baseline to follow-up affected brain structures that are crucial for regulation of emotions and thus need to be prevented. Our results illustrate the negative association of relapse with morphologic brain alterations and might be a step to guide future prognosis and maintenance treatment in patients with recurrent MDD.

Accepted for Publication: January 18, 2018.

Corresponding Author: Udo Dannlowski, MD, PhD, Department of Psychiatry, University of Münster, Albert-Schweitzer-Campus 1, A9, 48149 Münster, Germany (dannlow@uni-muenster.de).

Published Online: March 28, 2018. doi:10.1001/jamapsychiatry.2018.0123

Author Contributions: Mr Zaremba and Ms Dohm contributed equally to the present work and should therefore both be regarded as co-first authors. Mr Zaremba and Dr Dannlowski had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of data analysis.

Study concept and design: Zaremba, Dohm, Redlich, Repple, Kugel, Dannlowski.

Acquisition, analysis, or interpretation of data: Zaremba, Dohm, Redlich, Grotegerd, Strojny, Meinert, Bürger, Enneking, Förster, Opel, Baune, Zwitserlood, Heindel, Arolt, Kugel, Dannlowski.

Drafting of the manuscript: Zaremba, Dohm, Baune, Heindel, Dannlowski.

Critical revision of the manuscript for important intellectual content: Redlich, Grotegerd, Strojny, Meinert, Bürger, Enneking, Förster, Repple, Opel, Baune, Zwitserlood, Heindel, Arolt, Kugel, Dannlowski.

Statistical analysis: Zaremba, Dohm, Redlich, Bürger, Förster, Dannlowski.

Obtained funding: Dannlowski.

Administrative, technical, or material support: Zaremba, Dohm, Redlich, Grotegerd, Strojny, Meinert, Enneking, Förster, Repple, Opel, Heindel, Arolt, Kugel, Dannlowski.

Study supervision: Redlich, Grotegerd, Repple, Baune, Zwitserlood, Kugel, Dannlowski.

Conflict of Interest Disclosures: Dr Arolt reported serving as a member of the advisory board of, or has given presentations on behalf of, AstraZeneca, Janssen-Organon, Lilly, Lundbeck, Servier, Pfizer, Otsuka, and Trommsdorff. Dr Kugel reported receiving consultation fees from MR:comp GmbH, Testing Services for MR Safety. No other disclosures were reported.

Funding/Support: This work was funded by grants FOR2107 DA1151/5-1, DA1151/5-2, and SFB-TRR58, Project C09, from the German Research Foundation (Dr Dannlowski); grant Dan3/012/17 from the Interdisciplinary Center for Clinical Research of the medical faculty of the University of Münster (Dr Dannlowski); and a research scholarship from the Deanery of the medical faculty of the University of Münster (Dr Opel).

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.