Major depressive disorder is one of the most common and burdening mental diseases worldwide,Reference Kessler, Berglund, Demler, Jin, Koretz and Merikangas1 and is associated with a mood-congruent bias in emotion processing.Reference Stuhrmann, Suslow and Dannlowski2,Reference Groenewold, Opmeer, de Jonge, Aleman and Costafreda3 The course of the illness shows a large interpersonal heterogeneity, varying from one single lifetime depressive episode to recurrent or ongoing episodes, with the risk for a future relapse being increased with each episode.Reference Kessler, Berglund, Demler, Jin, Koretz and Merikangas1,Reference Kanai, Takeuchi, Furukawa, Yoshimura, Imaizumi and Kitamura4 In fact, it is the course of illness over years that determines chronic disease burden and loss of functioning.Reference Ormel, Oldehinkel, Nolen and Vollebergh5 The neural correlates that underlie depressive relapses are of high psychosocial and economic interest, albeit poorly understood. Prospective neuroimaging studies could help to understand how disease progression affects the brain function during emotion processing, and identify patients at high risk of relapse who could benefit from applying intensified treatments and prevention strategies.
Neural correlates of negative emotion processing in depression
On the neural level, the mood-congruent bias in emotion processing in depression has been linked to a dysfunction of a fronto-limbic network, with a hyperactivity of (para-)limbic regions – such as the amygdala, hippocampus and insula – and a hypoactivity of prefrontal regions, including the dorsomedial and dorsolateral prefrontal cortex (DLPFC), in response to negative stimuli.Reference Stuhrmann, Suslow and Dannlowski2,Reference Groenewold, Opmeer, de Jonge, Aleman and Costafreda3 So far, longitudinal functional magnetic resonance imaging (fMRI) studies mainly focused on short-term effects of treatments for depression on brain function, mostly indicating a normalisation of depression-associated alterations in brain activity after successful treatment; for example, a reduction of pre-treatment amygdala and hippocampus hyperactivity during processing of sad or fearful faces was reported after treatment with antidepressantsReference Delaveau, Jabourian, Lemogne, Guionnet, Bergouignan and Fossati6 and electroconvulsive therapy.Reference Redlich, Bürger, Dohm, Grotegerd, Opel and Zaremba7 Regarding the insula, the state of research is less clear, with some studies indicating activity increases to negative stimuli after treatment,Reference Delaveau, Jabourian, Lemogne, Guionnet, Bergouignan and Fossati6 and other studies pointing to decreases.Reference Delaveau, Jabourian, Lemogne, Guionnet, Bergouignan and Fossati6,Reference Ma8
Long-term fMRI investigations
Only two fMRI studies examined changes in brain function in patients with depression during emotion processing covering more than 1 year.Reference Ai, Opmeer, Marsman, Veltman, van der Wee and Aleman9,Reference Opmeer, Kortekaas, van Tol, Renken, Demenescu and Woudstra10 They revealed that in remitted patients, insula activity in response to positive stimuli was reduced after 2 years,Reference Opmeer, Kortekaas, van Tol, Renken, Demenescu and Woudstra10 and symptomatic improvement was associated with increased amygdala and hippocampus activity in response to positive and negative words.Reference Ai, Opmeer, Marsman, Veltman, van der Wee and Aleman9 However, regarding negative stimuli, these results seem to stand in contrast to evidence of short-term treatment studies described above, which mainly showed a reduction in limbic hyperactivity.Reference Delaveau, Jabourian, Lemogne, Guionnet, Bergouignan and Fossati6,Reference Ma8 As Opmeer et alReference Opmeer, Kortekaas, van Tol, Renken, Demenescu and Woudstra10 found no activity changes for negative stimuli, there is a need to further investigate the long-term changes in brain function during negative stimuli processing.
Furthermore, the study of Ai et alReference Ai, Opmeer, Marsman, Veltman, van der Wee and Aleman9 suggested that brain functional changes were a result of remission status rather than of duration of depression during the study interval, whereas Opmeer et alReference Opmeer, Kortekaas, van Tol, Renken, Demenescu and Woudstra10 did not examine the influence of duration in depression or relapses during the study interval on their results. Therefore, in depression, the impact of course of illness and remission status on brain functional changes is still unclear and remains to be explored.
Aims of the study
The present study was designed to examine course of illness-dependent changes in negative emotion processing in a prospective approach. As the few longitudinal studies that examined brain functional changes over the long-term course of depression pointed to changes in the amygdala, hippocampus and insula as parts of brain circuits relevant for emotion processing and salience detection, a particular focus was placed on these regions. The objectives were as follows: (a) to investigate long-term changes in brain function during negative emotion processing and its dependence on the course of illness and remission status in patients with depression, in a prospective naturalistic design; (b) to subsequently control for the influence of medication dose and psychotherapy on brain function; and (c) to identify baseline differences in brain activity between patients with depression with different course of illness (full remission versus recurrent or ongoing depression).
Method
Participants and study design
In this case–control study, participants were investigated as part of an ongoing study of the Münster Neuroimaging Cohort from May 2010 to June 2015, and were reassessed from September 2012 to May 2018 (interscan interval: mean 2.25 years, s.d. 0.30 years, range 1.92–3.33 years). At baseline, all patients were experiencing a moderate-to-severe depressive episode and were receiving in-patient treatment (within the Department of Psychiatry at University Hospital of Münster or within the Psychiatric Hospital of the Landschaftsverband Westfalen-Lippe in Münster). Healthy controls were recruited by newspaper announcements and public notices. Participants received a financial compensation. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures were approved by the ethics committee of the University of Münster (2007-307-f-S). Written informed consent was obtained from all participants.
Exclusion criteria for all participants were any neurological abnormalities, chronic medical diseases, a history of alcohol or substance dependence, electroconvulsive therapy, benzodiazepine intake at study time, or any magnetic resonance imaging contraindications. For patients with depression, a further exclusion criterion was a diagnosis of bipolar or psychotic disorder, whereas healthy controls had to be free from any lifetime psychiatric disorder at baseline and follow-up (for a visualisation of exclusion process, see Supplementary Fig. 1 available at https://doi.org/10.1192/bjp.2021.223).
At baseline and follow-up, participants underwent fMRI during processing of negative emotional faces. Furthermore, the structured clinical interview for the DSM-IVReference Wittchen, Wunderlich, Gruschwitz and Zaudig11 and the Hamilton Rating Scale for Depression (HRSD)Reference Hamilton12 were conducted by trained clinical raters. Course of illness was assessed at baseline and follow-up. Type and frequency of psychotherapeutic interventions before baseline and during the study interval were documented. To quantify psychopharmacological treatment at baseline and follow-up, a composite medication load index was computed for each time point (see Supplementary Methods 1.1). For details on psychopharmacological medication intake, see Supplementary Table 1.
The final sample comprised 72 patients with depression and 42 healthy controls. Participants’ characteristics are summarised in Table 1. There was an overlap of 53 patients with depression and 31 healthy controls with a sample reported in one of our earlier studies investigating long-term morphological brain changes.Reference Zaremba, Dohm, Redlich, Grotegerd, Strojny and Meinert13 The patient group was further divided into two subgroups following a previously published protocol:Reference Zaremba, Dohm, Redlich, Grotegerd, Strojny and Meinert13,Reference Opel, Redlich, Dohm, Zaremba, Goltermann and Repple14 patients who were in full remission at follow-up and who did not experience any further depressive episode after baseline (no-relapse group, n = 25), and patients who suffered from ongoing depression since baseline or experienced at least one further depressive episode according to the DSM-IV15 after baseline (relapse group, n = 47).
Table 1 Sociodemographic and clinical characteristics of the sample
HRSD, Hamilton Rating Scale for Depression.
a. Comparing patients with relapse, no-relapse and healthy controls by using a one-way analysis of variance, except where noted.
b. Comparing patients with relapse and patients no-relapse by using the unpaired two-tailed t-test, except where noted.
c. P-values were obtained with the χ²-test.
d. Psychotherapeutic treatment during study interval was coded as yes with ≥ 12 sessions of psychotherapy.
fMRI methods
fMRI data acquisition and preprocessing followed previously published protocols.Reference Redlich, Grotegerd, Opel, Kaufmann, Zwitserlood and Kugel16,Reference Dannlowski, Stuhrmann, Beutelmann, Zwanzger, Lenzen and Grotegerd17 For details regarding the fMRI paradigm, data acquisition and preprocessing methods, see Supplementary Methods 1.2–1.4 and previous works.Reference Redlich, Grotegerd, Opel, Kaufmann, Zwitserlood and Kugel16–Reference Hariri, Mattay, Tessitore, Kolachana, Fera and Goldman18 Briefly, T2* functional data were acquired with a single-shot echoplanar sequence with a 3T scanner (Gyroscan Intera 3T, Philips Medical System, Best, The Netherlands). The fMRI paradigm was a modified version of the face-matching task described by Hariri et al.Reference Hariri, Mattay, Tessitore, Kolachana, Fera and Goldman18 It consisted of four blocks of an emotional face processing task and five blocks of a sensorimotor control task. Faces were derived from the Ekman and Friesen stimulus setReference Ekman and Friesen19 and expressed either anger or fear (Supplementary Fig. 2(a)), whereas the control task consisted of geometric shapes (circles or ellipses, see Supplementary Fig. 2(b)).
Onsets and durations of the two experimental conditions (faces and shapes) were modelled with a canonical hemodynamic response function in the context of the general linear model. The model was corrected for serial correlations and a high-pass filter of 128 s was applied to remove low-frequency noise. The six movement parameters from the realignment step were further entered as nuisance covariates. For each participant and measurement, an individual contrast image (faces > shapes) was generated in each fixed-effects first-level analysis, which was then used in the following second-level, random-effects group analyses. For quality control, effects of interest were visually inspected for each participant and time point.
Statistical analyses
Analyses of demographic and clinical data were performed with SPSS Statistics (version 25.0 for Windows; IBM Corporation). Second-level analyses of fMRI data were conducted with statistical parametric mapping software (SPM12, version 6685 for MATLAB on Ubuntu; Wellcome Department of Cognitive Neurology, London, UK; http://www.fil.ion.ucl.ac.uk/spm). Significance thresholds for multiple testing were obtained at the cluster-level by threshold-free cluster enhancement (TFCE), using the TFCE toolbox (version 223; Structural Brain Mapping Group, Jena, Germany; http://dbm.neuro.uni-jena.de/tfce) implemented in SPM12. We established a family-wise error (FWE)-corrected threshold of P < 0.05, obtained by 10 000 permutations per test. All second-level analyses were performed based on the faces > shapes contrast.
Region-of-interest analyses
First, all second-level analyses were performed in a region-of-interest (ROI) approach, based on masked voxel-wise analyses using small volume correction as implemented in the TFCE toolbox in SPM. The three a priori defined ROIs were the bilateral amygdala, bilateral hippocampus with parahippocampal gyrus and bilateral insula, each created separately by means of the Wake Forest University PickAtlasReference Maldjian, Laurienti, Kraft and Burdette20 tool (version 3.0; NeuroImaging Tools & Resources Collaboratory; https://www.nitrc.org/projects/wfu_pickatlas) implemented in SPM12 according to the automated anatomical labeling atlasReference Tzourio-Mazoyer, Landeau, Papathanassiou, Crivello, Etard and Delcroix21 definitions. Subregions were labelled by using cytoarchitectonic probability maps, implemented in the JuBrain Anatomy ToolboxReference Eickhoff, Stephan, Mohlberg, Grefkes, Fink and Amunts37 (version 2.2b; Forschungszentrum Jülich, Jülich, Germany; https://www.fz-juelich.de/inm/inm-7/DE/Resources/_doc/SPM%20Anatomy%20Toolbox_node.html) implemented in SPM12. A 3 × 2 analysis of covariance (ANCOVA) was computed with group (no-relapse, relapse, healthy controls) as between-participants factor, time (baseline, 2-year follow-up) as a within-participant factor, and age and gender as covariates. We investigated the relapse (no-relapse, relapse) × time (baseline, follow-up) interaction, the main effects of time and main effects of group at baseline and follow-up.
Whole-brain analyses
All second-level analyses for the 3 × 2 ANCOVA described for the ROI analyses were also performed at the whole-brain level.
Additional analyses
The Supplementary Methods provide a description of our additional analyses on (a) effects of medication dose and psychotherapy on changes in brain function (Supplementary Methods 1.5.1), (b) relapse prediction based on baseline brain functional data compared with clinical data (Supplementary Methods 1.5.2) and (c) effects of remission status on follow-up brain activity (Supplementary Methods 1.5.3).
Results
ROI analyses
Longitudinal relapse×time interaction
There was a significant relapse×time interaction (one-tailed) for the bilateral amygdala (Fig. 1(a) and (b)) and bilateral insula (Fig. 1(c) and (d)). The post hoc t-tests for changes in brain function from baseline to follow-up within groups revealed that the relapse×time interaction resulted from significant amygdala and insula activity increases within the no-relapse group, whereas within the relapse group, insula activity was reduced from baseline to follow-up. Healthy controls showed no significant changes over time (see also Supplementary Table 2).
Fig. 1 Longitudinal relapse×time interaction (one-tailed) of amygdala and insula ROI analyses. (a) Significant clusters (left: x = −26, y = 2, z = −26, t (220) = 3.32, k = 104, P TFCE-FWE = 0.011; right: x = 34, y = 0, z = −20, t (220) = 3.08, k = 81, P TFCE-FWE = 0.017) of the amygdala ROI analysis for the relapse×time interaction effect (one-tailed), driven by activity increases in the no-relapse group and activity decreases in the relapse group. The figure displays clusters significant at P TFCE-FWE < 0.05. Bar indicates t-values. (b) Plot depicts typical amygdala responses at baseline and follow-up for healthy controls, and patients with MDD with relapse and no-relapse during the study interval. fMRI contrast values were computed by extracting the first Eigenvariate of the significant left cluster (x = −26, y = 2, z = −26, t (220) = 3.32, k = 104, P TFCE-FWE = 0.011) resulting from the amygdala ROI analysis of the relapse×time interaction (one-tailed). Error bars indicate 1 s.e.m. (c) Significant clusters (left: x = −44, y = −2, z = −2, t (220) = 4.18, k = 1350, P TFCE-FWE = 0.001; right: x = 44, y = 2, z = −8, t (220) = 3.81, k = 1088, P TFCE-FWE = 0.004) of the insula ROI analysis for the relapse×time interaction effect (one-tailed), driven by activity increases in the no-relapse group and activity decreases in the relapse group. The figure displays clusters significant at P TFCE-FWE < 0.05. Bar indicates t-values. (d) Plot depicts typical insula responses at baseline and follow-up for healthy controls, and patients with MDD with relapse and no-relapse during the study interval. fMRI contrast values were computed by extracting the first Eigenvariate of the significant left cluster (x = −44, y = −2, z = −2, t (220) = 4.18, k = 1350, P TFCE-FWE = 0.001) resulting from the insula ROI analysis of the relapse×time interaction (one-tailed). Error bars indicate 1 s.e.m. fMRI, functional magnetic resonance imaging; FWE, family-wise error; MDD, major depressive disorder; ROI, region of interest; TFCE, threshold-free cluster enhancement.
For the hippocampus ROI, no significant relapse×time interaction effect emerged (P TFCE-FWE = 0.055). There was no significant main effect of time for any of the three ROIs (all P TFCE-FWE > 0.126).
Cross-sectional group differences at baseline
Supplementary Table 3 summarises the results of the between-group differences in baseline activity for all three ROIs. Briefly, a main effect of group at baseline emerged for the right amygdala, bilateral hippocampus and bilateral insula. Post hoc t-tests revealed that the no-relapse group showed hypoactivity within the left amygdala, bilateral hippocampus and bilateral insula, compared with healthy controls. Furthermore, patient groups differed in baseline activity, with higher baseline activity in the relapse group in the bilateral amygdala (Supplementary Fig. 3(a)), bilateral hippocampus (Supplementary Fig. 3(b)) and bilateral insula (Supplementary Fig. 3(c)), compared with the no-relapse group. Comparing the complete patient group with depression (no-relapse and relapse groups together) with healthy controls, no difference in brain function emerged in any of the three ROIs (all P TFCE-FWE > 0.150).
Cross-sectional group differences at follow-up
There was no main effect of group at follow-up for any of the three ROIs (all P TFCE-FWE > 0.187).
Whole-brain analyses
Longitudinal relapse×time interaction
Figure 2(a) and (b) and Supplementary Table 4 display the results of the relapse×time interaction (one-tailed) at the whole-brain level. Briefly, a significant relapse×time interaction emerged, driven by activity increases in the no-relapse group, whereas activity in the relapse group decreased from baseline to follow-up – mainly represented in the insula, temporal pole, temporal gyri, striatal and thalamic areas, as well as in the prefrontal cortex, including the anterior cingulate cortex (ACC), inferior, middle and superior frontal gyrus. There was no significant main effect of time at the whole-brain level (P TFCE-FWE = 0.148).
Fig. 2 Longitudinal relapse×time interaction (one-tailed) of whole-brain analysis. (a) Significant clusters of the whole-brain analysis for the relapse×time interaction effect (one-tailed), driven by activity increases in the no-relapse group and activity decreases in the relapse group. The figure displays clusters significant at P TFCE-FWE < 0.05. Bar indicates t-values. (b) Plot depicts typical responses at baseline and follow-up for healthy controls, and patients with MDD with relapse and no-relapse during the study interval. fMRI contrast values were computed by extracting the first Eigenvariate of the largest significant cluster (x = −44, y = 2, z = −10, t (220) = 4.32, k = 2190, P TFCE-FWE = 0.027) resulting from the whole-brain analysis of the relapse×time interaction (one-tailed). The cluster included parts of the following regions: insula, superior temporal gyrus, inferior frontal gyrus, temporal pole and rolandic operculum. Error bars indicate 1 s.e.m. fMRI, functional magnetic resonance imaging; FWE, family-wise error; MDD, major depressive disorder; TFCE, threshold-free cluster enhancement.
Cross-sectional group differences at baseline
Supplementary Table 5 presents the results of the between-group comparisons in baseline brain function at the whole-brain level. There was a significant main effect of group at baseline, resulting from significantly lower baseline activity in the no-relapse group compared with the relapse group (Supplementary Fig. 4) and healthy controls in regions involving prefrontal and temporal regions, basal ganglia, cerebellum and the precentral gyrus.
Cross-sectional group differences at follow-up
At follow-up, there was no significant main effect of group at the whole-brain level (P TFCE-FWE > 0.999).
Additional analyses
Effects of treatments on changes in brain function
Detailed results can be found in Supplementary Results 2.1 and Supplementary Table 6. Briefly, after controlling for medical and psychotherapeutic treatments, there was still a significant relapse×time interaction for the amygdala and insula ROIs.
Relapse prediction
The results of additional analyses investigating the predictive value of baseline brain functional data compared with clinical data are described in Supplementary Results 2.2 and Supplementary Table 7. The model including both clinical and baseline brain functional data revealed the best prediction regarding subsequent relapses compared with clinical or brain functional data alone. Significant predictors were cumulative duration of depression before baseline and baseline left insula activity.
Effects of remission status on follow-up brain activity
The division of the relapse group into two subgroups according to the remission status at follow-up resulted in n = 14 patients with relapse in full remission and n = 33 patients with relapse in current depression. The detailed results regarding the effects of remission status on follow-up brain activity can be found in Supplementary Results 2.3. Briefly summarised, there was a significant main effect of subgroup for the insula ROI (Fig. 3), resulting from lower activity in the relapse group with current depression compared with the healthy control group, as well as the no-relapse group and the relapse group in full remission.
Fig. 3 Bar graph depicting typical insula responses (faces > shapes) at follow-up for subgroups divided by remission status. Error bars represent 95% confidence intervals. fMRI contrast values were computed by extracting the first Eigenvariate of the significant cluster resulting from the insula ROI analysis investigating the main effect of subgroup (x = 40, y = 26, z = −6, F (3,110) = 7.00, k = 11, P TFCE-FWE = 0.043) at follow-up of the one-way ANOVA. fMRI, functional magnetic resonance imaging; FWE, family-wise error; MDD, major depressive disorder; ROI, region of interest; TFCE, threshold-free cluster enhancement. *P TFCE-FWE < 0.05, **P TFCE-FWE < 0.01.
Discussion
The present study examined brain function during negative emotion processing in the long-term course of depression in a naturalistic 2-year follow-up design. The main finding was that patients with diverging course of depression showed different changes in brain activity during negative emotion processing within the amygdala, insula and temporal and prefrontal cortex. Although the majority of patients were receiving treatment, medication dose and psychotherapy did not affect the pattern of results in the amygdala or insula. Baseline differences in amygdala, hippocampus and insula activity were detected between patient groups with different subsequent course of illness.
Changes in amygdala and insula activity
Both the amygdala and the insula are regions relevant for the detection and processing of emotionally salient stimuli.Reference Stuhrmann, Suslow and Dannlowski2,Reference Phan, Wager, Taylor and Liberzon22 The association of the insula with the course of illness is also highlighted by a previous studyReference Opel, Redlich, Dohm, Zaremba, Goltermann and Repple14 that showed reduced cortical surface area of the insula to be associated with future depressive relapse. In our study, the insula appears to be most sensitive for brain functional changes during emotion processing in the long-term course of illness. Increased insula activity in the no-relapse group seems to represent a normalisation of brain function, as activity was abnormally low at baseline during the acute episode and showed no difference at follow-up compared with healthy controls. Furthermore, healthy controls also showed a descriptive tendency for activity increases. Likewise, we found increased amygdala activity in the no-relapse group in response to negative stimuli after 2 years. Accordingly, Ai et alReference Ai, Opmeer, Marsman, Veltman, van der Wee and Aleman9 reported increased amygdala activity in association with symptom improvement during the encoding of positive and negative words. Combining these results, long-term remission may be associated with an increase in amygdala activity in response to emotional stimuli, independent of emotional valence. For the insula on the other hand, the changes in activity appear to be valence-specific, as we found increased activity in response to negative stimuli in patients in remission, whereas Opmeer et alReference Opmeer, Kortekaas, van Tol, Renken, Demenescu and Woudstra10 reported reduced activity in response to positive stimuli in remitters. It should be noted, however, that there are differences in types of task and control conditions across studies, limiting the comparability of the results and interpretations regarding emotional valence.
Changes in frontotemporal activity
Our whole-brain results further reveal changes within the temporal pole, orbitofrontal and middle frontal gyrus, and within the dorsal part of the ACC. The temporal pole is involved in face recognition and plays a key role in the integration of visual input with semantic information.Reference Olson, Plotzker and Ezzyat23 It has been reported to produce functional responses comparable to the amygdala, and is also activated during emotionally salient stimuli.Reference Olson, Plotzker and Ezzyat23 In contrast to the amygdala and insula, the temporal pole is also involved in more complex socioemotional functions:Reference Olson, Plotzker and Ezzyat23 temporal pole activity is evoked during theory-of-mind tasks and atrophies are associated with social withdrawal, abnormal social behaviour and lack of empathy.Reference Gorno-Tempini, Rankin, Woolley, Rosen, Phengrasamy and Miller24 Our results indicate that patients in full remission show increased temporal pole activity during negative emotional face processing, which may depict enhanced integration of emotional and social information. In contrast, patients with relapse or ongoing depression show decreased activity of the temporal pole, suggesting further difficulties in activating socioemotional functions. One potential mechanism might be depression-associated social withdrawal and social anhedonia leading to a decrease in social interactions and, therefore, to deficits in employing mentalising and theory-of-mind capacities. However, as theory-of-mind tasks were not used in our study, this interpretation remains to be verified by future studies using more complex socioemotional paradigms.
Orbitofrontal, medial prefrontal areas, as well as the insula, are structurally connected with the temporal pole by the uncinate fasciculus, explaining the similar activation patterns. Remission-associated activity increases in the orbitofrontal, prefrontal and dorsal cingulate cortex further suggest enhanced emotion regulation in the no-relapse group at follow-up, as stated by the emotion regulation model by Phillips et al.Reference Phillips, Drevets, Rauch and Lane25 This finding is supported by a previous study that showed an association between symptom improvement after 6 months and an increase in prefrontal activity during regulation of negative affect.Reference Heller, Johnstone, Peterson, Kolden, Kalin and Davidson26 Further, a meta-analysis found the dorsal ACC and anterior insula to be the only regions involved in both explicit and implicit emotion regulation strategies.Reference Picó-Pérez, Alemany-Navarro, Dunsmoor, Radua, Albajes-Eizagirre and Vervliet27 The ACC, inferior frontal gyrus extending to the insula, as well as the thalamus, are furthermore activated in applying the optimal policy in approach–avoidance conflicts, whereas activity of the amygdala and DLPFC is associated with a simpler heuristic strategy that considers only the probability of threat.Reference Korn and Bach28 Thus, the regions in which we found changes over time as a function of the course of depression also play an important role in the resolution of approach–avoidance conflicts in addition to emotion regulation.
Since structural neuroimaging studies found changes that are dependent on the course of illness in grey matter volume and cortical thickness in the insula, ACC, DLPFC and orbitofrontal cortex,Reference Zaremba, Dohm, Redlich, Grotegerd, Strojny and Meinert13 the findings suggest that in the long term, depressive relapses are associated with reductions not only in grey matter volume, but also in brain activity during negative emotion processing.
Effects of medication dose and psychotherapy on changes in brain activity
Corresponding with previous long-term fMRI research,Reference Ai, Opmeer, Marsman, Veltman, van der Wee and Aleman9 treatment with psychotherapy during the study interval did not influence the pattern of results. Medication load diminished the relapse-associated changes in brain function at the whole-brain level. Therefore, activity changes could partly be influenced by medication dose effects, which would be in line with short-term treatment studies revealing brain functional changes during negative emotion processing after antidepressant pharmacotherapy.Reference Delaveau, Jabourian, Lemogne, Guionnet, Bergouignan and Fossati6,Reference Ma8 Nevertheless, within the amygdala and insula, neither psychotherapy nor medication influenced the pattern of results. This suggests that the course of illness rather than specific effects of treatments may underlie long-term changes in amygdala and insula activity.
Cross-sectional differences in brain activity
In contrast to previous studies,Reference Stuhrmann, Suslow and Dannlowski2,Reference Groenewold, Opmeer, de Jonge, Aleman and Costafreda3 there was no difference in baseline amygdala, hippocampus and insula activity between the complete patient group including all patients with depression (both no-relapse and relapse) and healthy controls. Dividing the patient group into subgroups according to the course of illness revealed that patients without future relapse showed abnormally low amygdala, hippocampus and insula activity compared with healthy controls and the relapse group. This heterogeneity in baseline brain function within patients with depression may provide a rationale for the missing difference between the complete patient group and healthy controls at baseline.
Our analyses revealed that the inclusion of baseline brain functional data substantially improved the prediction of relapse based on clinical data alone. These results underline the potential of fMRI as an additional predictive marker to identify patients at high risk of relapse.
The finding of higher baseline brain activity in patients with an unfavourable subsequent course of illness is supported by short-term treatment studies that also suggest higher baseline activity of the insulaReference Fu, Steiner and Costafreda29 and amygdalaReference Williams, Korgaonkar, Song, Paton, Eagles and Goldstein-Piekarski30 is associated with subsequent non-response to pharmacotherapy. In contrast, high psychological resilience has been found to be linked with lower amygdala activity during rest,Reference Leaver, Yang, Siddarth, Vlasova, Krause and St. Cyr31 and insula activity has been shown to differentiate between high-resilient and low-resilient individuals.Reference Waugh, Wager, Fredrickson, Noll and Taylor32 The lower limbic activity detected in the no-relapse group could therefore be a function of higher resilience to subsequent depressive relapses. However, as the majority of patients included in our study was already under psychopharmacological treatment at baseline, lower baseline activity in the no-relapse group could also be a result of antidepressants applied before baseline. For example, studies showed that with antidepressant treatment, reductions in amygdala responses to negative stimuli occurred early, and even before the onset of mood changes,Reference Godlewska, Norbury, Selvaraj, Cowen and Harmer33 and that greater early reductions in amygdala and insula activity after antidepressant treatment were predictive for later response.Reference Godlewska, Browning, Norbury, Cowen and Harmer34
Notably, there was no difference in baseline amygdala, hippocampus and insula activity between the relapse group and healthy controls. Again, psychopharmacological effects could have masked potential differences in brain function, reducing potential limbic hyperactivity in this patient group. Thus, antidepressants could potentially have diminished limbic hyperactivity in both groups of patients with depression (no-relapse and relapse), with only the no-relapse group showing neural correlates of early response in terms of limbic hypo-activity.
Taken together, these results suggest heterogeneity in the neural correlates of depression that may be associated with later disease progression, and therefore emphasise the importance of longitudinal phenotyping.
Exploratory analyses revealed that at follow-up, only patients with current depression had abnormal insula activity compared with healthy controls and patients in full remission. Instead, patients in full remission showed no difference in limbic activity compared with healthy controls, independent of the occurrence of relapses during the study interval. This finding demonstrates the importance of taking the mood state into account when investigating the course of illness, and is in line with cross-sectional studies showing mood-congruent brain functional alterations of the limbic system only in patients with current depression.Reference Stuhrmann, Suslow and Dannlowski2,Reference Arnone, McKie, Elliott, Thomas, Downey and Juhasz35 Nevertheless, the small sample size per subgroup in our exploratory investigations should be considered and, subsequently, these results should be interpreted carefully.
Strengths and limitations
The major strengths of our study are the prospective naturalistic design providing a high external validity, the long study interval covering 2 years and a careful investigation of the course of illness. Nevertheless, some limitations should be mentioned. First, although we were able to examine dose-dependent effects of pharmacotherapy, the small number of drug-free patients did not allow us to examine the effect of medication versus no medication. Second, given the nature of our paradigm, it cannot be clearly stated whether the findings are a result of processing of the emotional valence of the faces or face processing in general. However, the paradigm applied in this study is well-established and frequently used to investigate brain functional correlates of emotion processing.Reference Redlich, Grotegerd, Opel, Kaufmann, Zwitserlood and Kugel16–Reference Hariri, Mattay, Tessitore, Kolachana, Fera and Goldman18 Third, the reliability of paradigm-based fMRI is limited.Reference Elliott, Knodt, Ireland, Morris, Poulton and Ramrakha36 Consequently, there is a high need for further longitudinal studies covering more participants, to improve the reliability of the results.
Implications
In summary, our study reveals long-term activity changes in emotion processing areas that are dependent on the course of illness, and point to a normalisation of brain function in remission. Our findings contribute to a more profound understanding of the brain functional correlates of conscious negative emotion processing in the context of depressive relapses, and thereby shed some light on the long-term neural trajectories of major depression. Common treatment approaches may be effective in reducing the negative bias in emotion processing.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1192/bjp.2021.223.
Data availability
The data that support the findings of this study are available from the corresponding author, R.R., upon reasonable request.
Author contributions
V.E., M.K., K.D., R.R. and U.D. contributed to the conception and design of the study. All authors contributed to the organisation of the conduct of the study (including acquisition of data). V.E., M.K., L.M.F., R.R. and U.D. analysed and interpreted the data. V.E., M.K., R.R. and U.D. drafted the work and all other authors revised it critically for important intellectual content. All authors have read and approved the final version of the manuscript.
Funding
The study was supported by grants from the Innovative Medizinische Forschung (grant RE111722 to R.R., grant I-KO121806 to K.D.), the German Research Foundation (grant RE4458/1-1 to R.R., grants FOR2107 DA1151/5-1 and DA1151/5-2 to U.D.) and the Interdisciplinary Center for Clinical Research of the Medical Faculty of Münster (grant Dan3/012/17 to U.D.).
Declaration of interest
None.
Open access
eLetters
No eLetters have been published for this article.