Alzheimer’s disease (AD) is a highly disabling neurodegenerative disease characterised by memory loss and a gradual cognitive, functional, and behavioural decline (Reitz and Mayeux, 2014). Its prevalence increases rapidly with age, affecting 13% of the population at age 80, and 37% of the population at age 90 (Von Strauss et al., 1999). Individuals with AD often have comorbid major depressive disorder (MDD), present in 22–59% of cases (Zubenko et al., 2003; Starkstein et al., 2005), while MDD has an estimated lifetime prevalence of 11–15% in the general population (Bromet et al., 2011). MDD is a heterogeneous disorder; in addition to the core symptoms of low mood, anhedonia, and loss of energy, it comprises behavioural, physiological, and psychological signs and symptoms that include changes in appetite, sleeping, and psychomotor patterns, fatigue, lack of concentration, feelings of worthlessness or guilt, and suicidal ideation (American Psychiatric Association, 2013).

It has been long discussed whether a history of depressive symptoms is a risk factor for later development of AD, or rather an early prodromal manifestation of AD (Devanand et al., 1996; Chen et al., 1999). While bidirectional effects between the two disorders is likely, there is more evidence that midlife onset depressive symptoms and/or MDD are a risk factor for AD than vice versa (Jorm, 2001; Wilson et al., 2002; Green et al., 2003; Ownby et al., 2006; Barnes et al., 2012; Gracia-García et al., 2015). Furthermore, AD patients with depressive symptoms show accelerated cognitive decline and neurodegeneration, with significantly more plaques and tangles in the hippocampus than non-depressed individuals with AD (Rapp et al., 2006), while AD symptom count (Verkaik et al., 2007) or tau pathology (Kramberger et al., 2012) does not appear to contribute to the incidence or severity of depressive disorders.

Neuroimaging studies have provided scattered evidence that AD and depressive disorders share neurobiological pathways. Early stage AD is associated with atrophy of the hippocampus, para-hippocampal regions (Jack et al., 1999), and temporo-parietal cortex (Acharya et al., 2019), with atrophy becoming generalised in later stages of the disease, including cortical thinning in primary motor and sensory regions (Fox et al., 2001; Sabuncu et al., 2011). Similarly, MDD and recurrent major depression (MD) are related to smaller hippocampal volumes (Sheline et al., 1999; Bremner et al., 2000; Mervaala et al., 2000), amygdala, and parahippocampal areas (Andreescu et al., 2008) as well as lower cortical thickness in medial orbitofrontal cortex, fusiform gyrus, insula, rostral, and caudal anterior and posterior cingulate cortex, temporal lobe in MDD (Schmaal et al., 2020), many of these changes correlating positively with the duration of the disease (Andreescu et al., 2008). In AD patients with comorbid symptoms of depression, MRI studies have shown specifically thinner cortex in temporal and parietal areas when compared to non-depressed AD patients (Lebedeva et al., 2014). Conjunction analysis on the brain morphological changes that overlap between AD and late-life onset depression has shown that, in addition to the previously mentioned structures, both conditions are associated with hippocampal atrophy (Boccia et al., 2015). Yet, the risk of developing AD in MDD does not seem to be mediated by hippocampal or amygdala volumes (Geerlings et al., 2008).

Both AD and depressive disorders are heritable, with twin studies indicating 37% broad heritability for MDD (Sullivan et al., 2000) and 74% for AD (Gatz et al., 1997). Molecular genetics studies show that both disorders have complex genetic architectures. AD has recently been characterised as oligogenic, with estimates indicating the involvement of relatively few genetic variants, in addition to the well-known, strong APOE-e4 risk variant (Holland et al., 2020; Zhang et al., 2020). MDD on the other hand has been estimated to be the most polygenic of all major brain disorders (Holland et al., 2020), involving many genetic variants with small effects that explain a small amount of its heritability (Wray et al., 2018). Regardless, given the high comorbidity and indications of shared neurobiological pathways, substantial genetic overlap is to be expected, which may be leveraged to better understand these disorders (Smeland et al., 2020). Indeed, several candidate gene studies have identified shared genetic risk factors (Ye et al., 2016) that implicate hypothesised shared mechanisms, such as chronic neuroinflammatory changes in the brain (Leszek et al., 2018). While negligible genetic overlap between AD and MDD has been reported (Gibson et al., 2017; Lutz et al., 2020), substantial genetic overlap may remain hidden from measures of global genetic correlation due to mixed directions of effects. Here, we assess the genetic overlap between AD and depression across the genome through tools that capture the extent of overlap or specific loci, regardless of directions of effect. This was followed-up by analyses of the associations between shared loci and regional brain morphology in the UK Biobank (UKB) population study, providing valuable insights into their shared neurobiology.

Here we employed state-of-the-art biostatistical tools to improve our knowledge of the genetic underpinnings of the relation between AD and depression. In line with previous reports we have identified large differences in the genetic architecture of these disorders. We add new knowledge by revealing the presence of genetic overlap between them. We further illustrated how conjunctional analysis may be used to discover specific shared genetic loci, and substantially expanded on previous efforts by mapping the effects onto the brain in order to identify neurobiological mechanisms that contribute to the relation between these disorders.

We found that many of the causal variants for AD are overlapping with depression. This partly contradicts the previously reported negligible genetic correlation between AD and MD (Gibson et al., 2017), as well as the overall low genetic correlation reported between neurologic and psychiatric disorders (Anttila et al., 2018). However, whereas genetic correlations rely on globally consistent directions of effects between the two traits under investigation, bivariate Gaussian mixture modelling estimates the number of causal variants that have an effect on both, regardless of directions of effects. High levels of mixed directions of effects is likely to be commonplace for complex traits such as brain disorders. This can be seen in, for instance, another psychiatric disorder like schizophrenia, which has been estimated to share virtually all causal variants with educational attainment, despite a near-zero genetic correlation (Frei et al., 2019).

The strong heterogeneity of depressive disorders is likely to contribute to the mixed directions of effects between the two traits, which appears in contradiction to the high levels of reported comorbidity with AD. The heterogeneity of the depression phenotype is evident from its wide range of signs and symptoms and the likely existence of several depression subtypes. This may explain how the extent of genetic overlap can be large for the less polygenic AD, yet very small for depression, which fits with numerous reports that depressive disorders are an important predictor of AD pathology while the opposite is less true (Rapp et al., 2006; Verkaik et al., 2007; Kramberger et al., 2012). We speculate that some depressive disorder subtypes will be shown as genetically more concordant with AD than others, in line with indications that depressive disorders subtypes have significantly different genetic architectures (Jang et al., 2004; Milaneschi et al., 2016). The wide range of reported levels of comorbidity with AD across studies (Mega et al., 1999; Zubenko et al., 2003; Starkstein et al., 2005; Verkaik et al., 2007; Gracia-García et al., 2015) may also be due to this heterogeneity, as they differ in defining and subtyping of depression. A direct investigation of the relation of AD comorbidity and depressive disorders subtypes, coupled to neurobiological data, would be valuable. We postulate that studies using more narrow depressive disorder subtypes would find lower polygenicity and more concordant directions of effects with AD for specific subtypes.

Our use of cFDR to identify a specific locus shared by AD and depression is an example of how we may use genetics to improve our understanding of the neurobiology underlying the relation between these two disorders. The SNP rs5011436 is located in an intron of the gene TMEM106B, which encodes the transmembrane protein 106B. TMEM106B was the first genetic risk factor to be identified for fronto-temporal lobar degeneration (FTLD; Van Deerlin et al., 2010). Since then, it has also been reported in GWAS of both AD (Jun et al., 2016) and MD (Howard et al., 2019). The protein TMEM106b is thought to regulate lysosomal function, with a role in the clearance of TPD-43 (Nicholson and Rademakers, 2016). Both lysosomal function and specifically TDP-43 are highly related with the pathogenesis of AD (Nixon et al., 1992; Wilson et al., 2011) and MD (Modrego and Ferrández, 2004). TMEM106B expression has been shown to be downregulated in brains of individuals with AD (Satoh et al., 2014), while it has been found to be upregulated in individuals with MD (Dall’Aglio et al., 2020).

The evidence for involvement of TMEM106B in both disorders is further substantiated by our neuroimaging analyses, indicating effects on several brain regions that have been tied to both AD and depression. In particular the corpus callosum was implicated by our analyses, in line with previous neuroimaging findings on TMEM106B (Adams et al., 2014), with higher volume of several callosal subregions for carriers of the rs5011436 c-allele, the allele that we found to convey risk for depression and to be protective for AD. MDD is associated with abnormal cerebral lateralisation, and individuals with familial MDD have been found to have significantly larger callosal volume than individuals with non-familial forms of MDD (Lacerda et al., 2005), while AD has been repeatedly linked to degeneration of the corpus callosum (Di Paola et al., 2010). Thus, specific, genetically mediated, forms of depression have been found to have opposing directions of effects on the corpus callosum than other forms of depression and AD.

Our estimates of heritability, polygenicity, and discoverability highlight the complexity of the genetic architecture of both disorders. While twin studies have indicated high broad heritability of AD (Gatz et al., 1997) and MD (Sullivan et al., 2000), we replicated previous findings of low SNP-based heritability, as captured by GWAS data (Lambert et al., 2013; Wray et al., 2018; Jansen et al., 2019). This possibly implicates an important role for rare variants, as well as a high degree of genetic and environmental interaction effects. Contrasting the two disorders, it is clear that AD polygenicity is relatively low, with a recent study even qualifying late-onset AD as oligogenic (Zhang et al., 2020). Depressive symptoms on the other hand are highly polygenic, partly reflecting the substantial clinical heterogeneity which may capture a broad range of conditions, each likely with partly distinct genetic determinants (Jang et al., 2004). Regardless of these differences in genetic architectures, our analyses made clear that, with current approaches, GWAS sample sizes will need to reach millions of individuals to uncover a substantial fraction of the common genetic variance influencing both disorders.

Our findings once again reiterate the complexity of the genetic architectures of brain disorders, highlighting the limitations of the GWAS approach. Our power analyses suggest that, despite tremendous efforts from worldwide consortia to bring together large samples, we are only at the very beginning of uncovering the genetic determinants of AD and depression through the standard GWAS approach. Clearly, more powerful biostatistical tools are needed, ones that better match this complexity and that leverage genetic signal shared across traits of interest (Frei et al., 2019; Smeland et al., 2020; van der Meer et al., 2020) in order to lower the required sample sizes and provide more meaningful metrics. While approaches like Gaussian mixture modelling are a step in the right direction, the current implementation does still suffer from oversimplified assumptions about the nature of the genetic architecture of brain disorders; AD is enriched for rare variants, while the MiXeR analysis focuses on common variants only. Further, low polygenicity implies a handful of large genetic effects – there is a bigger chance that the distribution of those effect sizes won’t follow a Gaussian distribution, violating model assumptions. We are developing extensions of this method that will handle such characteristics.

To conclude, in this study we provided further insights into the genetic relationship between AD and depression, providing evidence of significant genetic overlap, and neuropathological effects reflected in brain morphological changes, warranting further genetic research. However, it seems that the complex relation between AD and depression will require future research to employ larger sample sizes, cleaner phenotype definitions and further improvements of biostatistical tools. It will also be important to study interaction effects between genetic variants and between genetic and environmental factors, as well as the dynamic interplay between relevant factors over the lifespan. These all will influence the underlying biological mechanisms that account for the complex relationship between these disorders. Ultimately this knowledge may provide a path toward more effective treatments, thereby reducing the enormous burden that AD and depression place on patients and their care-givers.

Publicly available datasets were analysed in this study. This data can be found on the UK biobank datasets under accession number 27412. This data can be accessed following a formal application at https://www.ukbiobank.ac.uk/enable-your-research.

The studies involving human participants were reviewed and approved by UK North West Multi-centre Research Ethics Committee. The patients/participants provided their written informed consent to participate in this study.

JM-S and DM conceived the study. DM and TK pre-processed the data. JM-S and DM performed all analyses, with conceptual input from MS and DL. All authors contributed to interpretation of results. JM-S and DM drafted the manuscript and all authors contributed to and approved the final manuscript.

Dr. Andreassen has received speaker’s honorarium from Lundbeck, and is a consultant to HealthLytix. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The handling editor is currently organizing a Research Topic with two of the authors OA and OS.