ENIGMA MDD: seven years of global neuroimaging studies of major depression through worldwide data sharing

Subcortical brain regions

The first ENIGMA MDD project focused on differences in subcortical volume between MDD patients (N = 1728) and healthy controls (N = 7199)³⁵. Consistent with prior studies and retrospective meta-analyses^8,43–45, in this IPD-based meta-analysis we found significantly lower hippocampal volumes in individuals with MDD compared with healthy controls. This effect was also consistently observed across individual cohorts, although the overall effect size was modest (Cohen’s d = −0.14). The hippocampal volume deficit was greater in MDD patients with recurrent episodes (N = 1119, Cohen’s d = −0.17), compared with healthy controls, whereas no hippocampal volume alterations were observed in first-episode patients (N = 583). Our findings may suggest that depression-related reductions in hippocampal volume are a result of longer illness duration or greater number of episodes, instead of a premorbid vulnerability factor. This is consistent with prior longitudinal studies showing greater hippocampal atrophy in individuals with persistent, recurring, or worsening of depressive symptoms over time^46–49. Nonetheless, it is unclear whether hippocampal atrophy associated with prolonged illness duration or recurrence represents a state marker instead of a permanent scar. Fortunately, hippocampal alterations may normalize as hippocampal enlargement has been observed following remission or treatment of MDD^48,50,51. We also found smaller hippocampal volumes in patients with an adolescent onset of MDD (⩽21 years; N = 541; Cohen’s d = −0.20), compared with controls, whereas no differences were observed in those with an adult onset of MDD (N = 997) (Fig. 3a). This is in line with previous studies showing smaller hippocampal volumes in adolescents and even children with depression^52–55, whereas other studies found lower hippocampal volumes only in adults with an age of onset >30⁵⁶ or no differences in hippocampal volume between adults with an adolescent versus adult age of onset of MDD⁵⁷. Because only about half (57%) of the adolescent-onset patients had a recurrent episode of MDD, adolescent disease onset may, in part, have an independent association with hippocampal volumes. Smaller hippocampal volumes may precede disease onset, especially in this early-onset group, perhaps as a result of factors commonly associated with early-onset MDD including childhood adversity^58,59 and genetic influences⁶⁰. Longitudinal studies designed to track hippocampal volume changes prior to disease onset and over the disease course are required to elucidate whether hippocampal abnormalities result from a prolonged duration of chronic stress associated with depressive episodes, represent a vulnerability factor for MDD, or both.

Interestingly, we did not detect significant differences for any of the other subcortical volumes, including the amygdala, nucleus accumbens, caudate, putamen, thalamus, and pallidum, or the lateral ventricles and intracranial volume (ICV). Previous reports have varied regarding volume abnormalities in subcortical regions other than the hippocampus, such as the amygdala^8,61,62. Nonetheless, associations with MDD may still be present for functionally distinct subregions within these broader subcortical regions. Patterns of depression-related alterations in subregions of subcortical surfaces have been difficult to identify as there are few identifiable surface landmarks, in contrast with the landmarks consistently found in cortical surfaces (e.g., deep sulcal patterns). In this context, subcortical shape analysis may be a more sensitive method to identify more localized effects in subdivisions in subcortical regions that were not captured by the volumetric analysis of subcortical regions in the first paper from the ENIGMA MDD consortium³⁵. To address this, we conducted an additional multi-site meta-analytic investigation to test whether MDD patients, and specific subgroups of MDD based on important clinical characteristics, differ from healthy controls in subcortical shape. Specifically, we applied meta-analytic models on effect sizes generated from 1781 patients with MDD and 2953 healthy controls across 10 study cohorts. Consistent with the findings from our first meta-analysis, we found that relative to healthy controls (N = 2879), patients with an adolescent onset of MDD (N = 476) had lower thickness (Cohen’s d = −0.17) and smaller surface area (Cohen’s d = −0.18) in the hippocampus, with most pronounced effects in the subiculum and cornu ammonis (CA) subfields two and three of the hippocampus³⁷ (Fig. 3a). Extending our prior findings, we also observed lower thickness (Cohen’s d = −0.16) and smaller surface area (Cohen’s d = −0.17) in the amygdala in adolescent-onset patients, specifically within the basolateral subdivision of the amygdala³⁷ (Fig. 3a). These subregions are rich in glucocorticoid receptors, emphasizing that disturbed glucocorticoid signaling during stress response promotes the development of MDD⁶³. Importantly, shape analyses of subcortical structures clarify results in the extant literature of smaller hippocampal volumes and ambiguous effects in the amygdala in patients with MDD; delineating nuanced effects in depression-related subregions of subcortical structures may help to identify more precise intervention targets or more sensitive biomarkers of treatment response. Noteworthy, additional ENIGMA MDD analyses regarding associations between MDD and FreeSurfer-derived hippocampal subfields are currently underway.

Cortical thickness and surface area

Following these studies of subcortical morphology, we examined cortical thickness and surface area in relation to MDD and clinical characteristics in a meta-analysis of data from 20 participating ENIGMA MDD cohorts³⁶. Because more research groups joined ENIGMA MDD after the publication of our subcortical volume meta-analysis study, we were able to conduct separate analyses in young people (⩽21 years) and adults (>21 years). Most published studies to date have focused on regional cortical volume, which is a function of cortical thickness and surface area. Advances in neuroimaging data processing have made it possible to separate cortical surface area and cortical thickness, which is important to do in the context of understanding brain correlates of MDD, given that these neural characteristics are genetically and phenotypically distinct^27,64,65. In adults, we observed subtle cortical thickness alterations in 13 of 68 cortical regions in patients with MDD (N = 1902) compared with healthy controls (N = 7658) (Cohen’s d’s between −0.10 and −0.14), including lower thickness of the bilateral medial orbitofrontal cortex (OFC), fusiform gyrus, insula, rostral anterior (ACC) and posterior cingulate cortex (PCC) and unilaterally in the left middle temporal gyrus, right inferior temporal gyrus and right caudal ACC (Fig. 3b). Our findings in adults with MDD were consistent with prior meta-analyses showing depression-related structural alterations in the medial PFC and ACC^8,66–69; however, they extended previous findings by demonstrating structural abnormalities in the temporal regions (middle and inferior temporal gyri and fusiform gyrus), posterior cingulate cortex and insula. These cortico-limbic thickness alterations may contribute to the broad spectrum of emotional, cognitive, and behavioral disturbances in MDD.

The largest effect size was observed in the medial OFC, which—in contrast to lower hippocampal volume—was already detectable in first-episode patients. In contrast to the hippocampal volume finding that were driven by adult patients with an adolescent onset of their first depressive episode, the lower cortical thickness findings were driven mostly by adult patients with an adult onset of MDD (N = 1214; Cohen’s d −0.11 to −0.18) relative to controls. No cortical surface area differences were found among the adult groups. We speculated that the more pronounced effects in adult patients with an adult age of onset may be driven in part by their older age compared with adult patients with an adolescent age of onset of their first depressive episode, which was confirmed by a post hoc moderator analysis with mean age of patients in each sample. This suggests that mental illness has a greater impact on cortical thickness in the context of aging. Indeed, cortical thickness has been shown to be a more sensitive indicator of aging than is surface area or volume^70–72. Because our meta-analytic approach did not allow us to pool all data across samples, we were not able to investigate age-by-diagnosis effects across the entire age range (individual samples had restricted age ranges). Therefore, future mega-analyses could further elucidate these dynamic relations with development and aging.

Surprisingly, in contrast to adults with MDD, adolescents with MDD showed no cortical thickness alterations, but rather, alterations in global cortical surface area (Fig. 3a); the effect sizes of these results were larger (Cohen’s d −0.31 to −0.41 in local regions) than the cortical thickness alterations observed in adults. Cortical surface area has been understudied in the context of MDD. Nonetheless, a recent longitudinal study showed that lower surface area was specifically observed in young people experiencing depressive symptoms in early adolescence but not in those developing depressive symptoms later in adolescence, and that lower surface area was already observable in young people with subclinical depressive symptoms, not all of whom will develop a full-threshold MDD diagnosis³⁶. Thus, cortical surface area reductions may represent an early developing subtype of depressive disorder, shaped by genetic factors or early life adversity (prenatal^73,74 or perinatal or during childhood^75–77), and potentially precede the onset of MDD. This hypothesis is consistent with the observation that, compared with cortical thickness, cortical surface area has a higher genetic heritability^27,64,78, has a genetic correlation with MDD and depressive symptoms (this genetic association is absent for cortical thickness²⁷), is determined earlier in development, and is less strongly affected by later environmental influences^71,79.

Importantly and paradoxically, no differences in surface area were observed in adult patients with adolescent-onset MDD. This might be explained by (1) normalization of cortical surface area when transitioning into adulthood; (2) cortical surface area alterations being only present in a specific subgroup (subtype) of adult patients with adolescent-onset MDD, which we were unable to detect; or (3) those with cortical surface area alterations in early adolescence may be at higher risk for transitioning from MDD to other mental disorders over time. This latter possibility is consistent with reports of lower cortical surface area in adolescents and adults with psychosis or schizophrenia^80,81, and in individuals at high risk for and/or transitioning to psychosis^82,83. Longitudinal studies are required to test the hypothesis that cortical surface area alteration is a pre-existing risk factor for the development of MDD, and to investigate the subsequent clinical course of these depressed young people with global surface area reductions.

Brain asymmetry

Altered brain asymmetry may have a role in MDD, especially at a functional level⁸⁴, but only a few studies have investigated structural asymmetry^85,86. In our studies, the majority of cortical thickness or surface area alterations were observed across both hemispheres, with a few regions showing only unilateral differences³⁶. However, we did not explicitly test whether the effect sizes of our findings differed significantly by hemisphere. Therefore, in a separate study, we investigated structural asymmetries by investigating asymmetry indices ((left − right)/(left + right)) for local and global cortical and subcortical brain regions in individuals with MDD (N = 2256) compared with healthy controls (N = 3504)⁸⁷. The results showed no significant differences in brain structural asymmetry between individuals with MDD and controls for any of the structural brain measures, nor any associations with clinical characteristics. These findings suggest that altered brain structural asymmetry is of little relevance to the pathophysiology of MDD, although functional asymmetries may still play a role.

Brain aging in MDD

MDD is associated with an increased risk of aging-related medical illnesses such as cardiovascular disease and cancer^88,89. Although aging is associated with loss of gray matter, depression may accelerate age-related brain atrophy⁹⁰. Therefore, we examined deviations from normative brain aging in adults with MDD and associated clinical heterogeneity by pooling data from >6900 healthy controls and individuals with MDD from 19 different scanners participating in the ENIGMA MDD consortium⁹¹. Normative brain aging was estimated by predicting chronological age (18–75 years) from 7 subcortical volumes, 34 cortical thickness and 34 surface area, lateral ventricles and ICV measures using Ridge Regression, separately in 952 male and 1236 female controls. We showed that our brain age prediction model generalized to unseen hold-out samples (927 male controls and 986 males with MDD, and 1199 female controls and 1689 females with MDD; correlations r between predicted and actual age ranged from 0.77–0.85, mean absolute errors (MAE) ranged from 6.32 to 7.18 years), as well as to completely independent samples from different scanning sites (N = 1330 from 23 different scanners; r = 0.71 and MAE = 7.49 for male controls, r = 0.72 and MAE = 7.26 for female controls)⁹¹.

Brain-predicted age difference (brain-PAD) was computed from the difference between predicted “brain age” and chronological age⁹². We found that, at the group level, MDD patients had a + 1.08 years (Cohen’s d = 0.14, p < 0.0001) greater discrepancy between their predicted and actual age compared with control participants. In other words, individuals with MDD were estimated to be ~ 1 year older than expected based on the brain age model. The brain age model relied mostly on cortical thickness measures (compared with subcortical volumes, cortical surface area and ICV) in order to make good age predictions. Brain-PAD differences were observed in all subgroups of patients compared with controls, with no significant differences in brain-PAD between the patient groups.

As many of the MDD patients did not show advanced brain aging compared with controls, the clinical significance of the observed higher brain-PAD in MDD patients may be limited. However, there may still be a subgroup of MDD patients with more extreme patterns of brain aging, which would be important to identify as accelerated brain aging may be reversed with targeted treatment. For example, one study showed that brain-PAD was temporarily reduced by 1.1 years in healthy controls owing to the acute anti-inflammatory effects of ibuprofen⁹³. Inflammatory biomarkers are commonly dysregulated in MDD and negative relationships between levels of inflammatory cytokine (e.g., interleukin-6) and cortical thickness have been found in medication-free, first-episode MDD patients⁹⁴, suggesting that inflammation may be a common biological mechanism between MDD and brain aging. Notably, brain-PAD has been shown to be a general predictor of psychiatric and neurological disorders, with low clinical disease specificity⁹⁵.

White matter microstructure

We also examined white matter microstructure in 1305 MDD patients, and 1602 healthy controls from 20 samples participating in ENIGMA MDD worldwide, again using a meta-analytic approach³⁸. The ENIGMA protocol for diffusion tensor imaging (DTI)³³ calculates fractional anisotropy (FA) for 25 atlas-defined white matter tracts of interest. FA is a commonly used measure in DTI analysis, and higher values indicate directionally constrained diffusion of water molecules within the white matter, which is mostly interpreted as higher degree of myelin integrity. In addition to FA, the ENIGMA DTI protocols also yield the following diffusivity metrics: axial diffusivity (AD), which is thought to represent the number, caliber, and organization of axons, radial diffusivity (RD), which may be a measure of myelination, and mean diffusivity (MD), which is often considered a measure of membrane density⁹⁶. As maturation of white matter tracts continues through adolescence and young adulthood, adolescent (age ≤21 years) and adult (age >21 years) patients and controls were analyzed separately. The meta-analysis showed subtle but widespread changes in FA, with lower FA in adult MDD patients observed in 16 out of 25 ROIs (Cohen’s d between 0.12 and 0.26) (Fig. 3b). Adult MDD patients also showed higher RD in multiple tracts (Cohen’s d between 0.12 and 0.18), potentially reflecting changes in the morphology of glial cells or myelination^97,98. These alterations in FA and RD appeared to be global effects, as after correction for average FA and RD across the white matter skeleton respectively, these effects were no longer significant. Nevertheless, and in accordance with previous studies, the strongest regional changes in FA were observed in the genu and body of the corpus callosum and the corona radiata^99,100. The corpus callosum connects brain regions in both hemispheres, including regions involved in mood regulation such as the anterior cingulate cortices and orbitofrontal cortices, whereas the corona radiata is part of the limbic-thalamo-cortical circuitry and is also implicated in mood regulation¹⁰¹.

The effect sizes for case–control differences in adults were small, but very similar to the effect sizes reported in the meta-analysis of subcortical volume and cortical morphology^35,36. Also, in line with the previous cortical and subcortical meta-analysis findings, the widespread alterations in FA in adult patients were driven by MDD patients with recurrent episodes (N = 645), as there were no significant differences between first-episode patients (N = 169) and controls (N = 816). This again suggests that these alterations may reflect the cumulative effect of stress on brain morphology, rather than a vulnerability factor for MDD, although the reduced statistical power for the first-episode patients comparison may also explain this difference. In line with the cortical meta-analysis findings, but in contrast with findings on subcortical volume, we observed lower FA in patients with an adult age of onset (N = 399) compared with controls (N = 869) (Fig. 3b), but no differences when comparing patients with an adolescent age of onset (N = 251) compared with controls (N = 853). We hypothesized that MDD may interact with the normal aging process of white matter, which was in line with findings from our diagnosis-by-age interaction analysis, in which we observed that MDD was associated with an accelerated decline in overall FA with increasing age compared to controls.

We could not replicate the case–control differences in white matter microstructure in adults in a sample from the UK Biobank (N = 2096 patients and 3275 healthy controls), which may be related to lower severity of MDD symptoms in the UK Biobank sample or subtle differences in image processing. In addition, there were no significant differences in FA or diffusivity measures between adolescent patients and controls in the ENIGMA MDD sample after correction for multiple testing, with smaller effects in adolescents potentially related to lower disease duration and number of episodes in adolescent patients compared with adult patients. However, we cannot rule out that our sample of adolescent participants may still have been too small to detect subtle effects (N = 372 patients and 290 healthy controls).

Sex differences in depression-related structural brain alterations

Major depression is more than twice as prevalent and the disease burden of MDD is 50% higher in females than in males¹, which may suggest different etiological pathways to developing MDD in males and females. However, across our ENIGMA MDD studies examining subcortical, cortical, and white matter integrity differences in MDD^35,37,38,80, we found no diagnosis-by-sex interaction effects in adult MDD patients, indicating that structural brain alterations likely do not contribute to these sex differences in MDD. In addition, even though the model fits of the brain aging models improved when trained separately in males and females, the (subtle) advanced brain aging that we observed in adults with MDD was not different for male versus female patients⁹¹. Nonetheless, sex differences in structural brain alterations may be present during specific sensitive periods of brain development, such as adolescence or more specifically, during puberty¹⁰². We did indeed found higher RD only in adolescent males with MDD, but not females, compared with adolescent controls³⁸. However, we did not observe sex differences in the cortical alterations in the adolescent MDD group³⁶. This could perhaps be explained by the observation that sex differences in white matter volumes increase from birth, through adolescence, until males and females reach adulthood, whereas sex differences in gray matter remain relatively stable across development¹⁰³. Future research would also benefit from a separation between gender and sex analyses.