Leveraging large-scale genetic and neuroimaging data for the study of major depressive disorder: towards mechanistic insights and personalised approaches

Abstract

Major Depressive Disorder (MDD) is a prevalent psychiatric disorder and a leading cause of global disease burden, with women being twice as likely than men to be affected by MDD in the lifetime. Complex in nature, the symptoms of MDD are wide-ranging and its aetiology is multifactorial, involving both genetic and environmental risk factors, as well as their interaction. To elucidate mechanisms underlying MDD, large-scale neuroimaging studies have contributed robust findings on regional brain morphometric alterations and provided insights into their genetic underpinnings. There are, however, two overarching issues that remain unaddressed – (1) the use network-based approaches as a better representation of how the brain is structured as a complex network in large-scale studies, and (2) the translation of population-level neuroimaging findings to the individual level, which set the foundations for this thesis.

Most MDD studies thus far have focused on individual brain regions, but the brain is inherently structured as a network with brain regions linked together by anatomical white matter tracts (i.e., a connectome). As a complex network, the brain exhibits non-trivial organisational properties, such as having a strong “rich club” core where key brain regions are strongly connected together to function as a central backbone of brain communication. MDD is thought to be associated with dysconnectivity in the brain, where network architecture, such as the rich club core, is thought to be disrupted. However, studies adopting connectomic approaches to study these questions have been few and limited by sample size, resulting in low reproducibility. Hence, this was addressed in the first study of this thesis, where I leveraged structural connectome data that were processed locally from two large adult cohorts (UK Biobank (N=5,104), Generation Scotland (N=725)) and utilised graph theory measures to compare the structural connectomes of MDD cases and healthy controls from the global network level down to the individual connections. Here, it was found that the rich club architecture remains robust in MDD. There were, however, subtle reductions in efficiency (i.e., a measure to reflect the efficiency of information transfer between brain regions) across the brain, which added up in the order of the connectome hierarchy to effect an overall reduction in global network efficiency in MDD. As such, brain structural connectomic differences in MDD do not consist of large effects confined to one or two distinct brain regions, but rather involves subtle effects that are distributed across the whole brain.

Having established that structural connectomic differences exist in MDD, the next step was to investigate how the connectome is affected by key MDD risk factors, such as MDD polygenic risk (MDD-PRS) and early life adversity such as childhood trauma (CT). A series of analyses were conducted using connectome data from 14,881 subjects (6,812 males and 8,069 females) from the UK Biobank. As previous studies have reported sex differences in MDD-PRS and CT associations with neuroimaging variables, the analyses here were conducted separately for males and females. Specifically, network-based statistics was first used to identify regions of the connectome (hereby called subnetworks) that were associated with MDD-PRS and CT (i.e., higher MDD-PRS or CT exposure is associated with lower network connectivity in the identified subnetwork). Next, the roles of the (1) PRS-associated subnetwork as mediator in the relation between MDD-PRS and MDD, and (2) CT-associated subnetwork as mediator in the relation between CT and MDD were examined. As the impact of MDD-PRS on MDD may differ in people with different degrees of CT exposure, further analysis using a moderated mediation model was also done to test if the direct and indirect effects of MDD-PRS on MDD were further moderated by CT. The results showed that the PRS- and CT-associated subnetworks were largely distinctive from each other within each sex, and were also specific to each sex. For the PRS-associated subnetwork, connectivity of the subnetwork was significantly lower in MDD cases than controls in females but not males. Likewise, a significant mediation effect was present in females but not males. Both direct and indirect effects in females were, however, not moderated by CT exposure. For the CT-associated subnetworks, connectivity of the subnetwork was significantly lower in MDD cases than controls in both sexes, but a significant mediation effect was only present in females. As such, the results in this chapter showed that MDD risk factors have differential associations with male and female structural connectomes, with connectome mediation effects only observed in females.

Importantly, while these population-level neuroimaging findings are useful in providing mechanistic insights, their use is limited if they cannot be translated to the individual level. In other words, as a first step towards clinical utility, can we use neuroimaging findings to assess an individual’s vulnerability to MDD? Given that brain morphometric findings by large-scale studies, such as ENIGMA, are well-established and have been shown to be reproducible, they can be used as a starting reference for the development of personalised brain-based risk scores. Thus, in the third study, I utilised ENIGMA summary statistics based on the adult cohort to generate brain-based risk scores (Regional Vulnerability Index abbreviated ‘RVI’) for five different brain morphometric measures for each subject in an adult subsample of Generation Scotland (N=702) and tested their associations with MDD. This was done concurrently with MDD-PRS, which can be considered a personalised genetic risk score, to use it as a benchmark to evaluate the validity of the brain-based MDD-RVIs in terms of effect sizes. The additive effect of both scores were also tested to see if they could account for more variation in disease risk when used in combination. It was observed that MDD-RVIs, namely those based on white matter microstructural measures (mean diffusivity and fractional anisotropy), had stronger associations with MDD compared to MDD-PRS. Further, both brain-based and genetic risk scores worked additively to increase variance explained in MDD. As an exploratory study to investigate the similarities of brain features of adolescent depression with adult MDD, the same analyses were repeated in a large adolescent cohort (N=2,081 to N=3,825). In adolescents, it was observed that MDD-PRS outperformed MDD-RVIs. These results re-emphasise the notion of ‘dysconnection’ in MDD and highlights the dynamic nature of the brain that allows it to capture signals of additional risk factors beyond genetic risk that influence disease course. With the increasing focus on connectome studies, future research can also consider extending this work to personalised connectome-based markers.

Summarising these works, these three studies helped to fill in knowledge gaps in the existing literature, specifically pertaining to the use of connectome-based approaches and the translation of findings to the individual level. The key takeaways include: (1) the rich club architecture remains stable in MDD but there are subtle but widespread differences across the connectome, (2) genetic and environmental factors can influence the risk of MDD through the connectome and sex differences in these associations are prominent, and (3) personalised brain-based markers have potential clinical utility in aiding the early diagnosis of MDD.