In the current analysis we investigated the interaction effect between SNP-based genetic variation and childhood trauma on seven MDD-related multimorbidity clusters. These clusters reflect the temporal courses of MDD-related multimorbidity burden throughout life and could initially be associated with a unique clinical, genetic and modifiable risk-factor profile (Juhasz et al., 2023). Here, we extend this direct genetic characterization by moderation effects investigating childhood trauma, one of the strongest risk factors for MDD and other psychiatric diseases in general. In seven G×E GWAS including 76,856 UK Biobank participants, we replicated the pattern of high- and low-multimorbidity clusters concerning childhood trauma burden which has already been found on the level of genetic and non-genetic factors (Juhasz et al., 2023). The strongest genetic findings in the G×E analyses could be observed for the high CTS burden Clusters 5 and 6 with more than 19 independent loci exceeding suggestive significance. These clusters were also associated with a high MDD-related multimorbidity burden. On the genome level, the correlation pattern showed a strong similarity between Clusters 1–4 which were all associated with a lower CTS burden. In contrast, Clusters 5–7 seemed to exhibit three individual but contrary genetic profiles that contribute to the high multimorbidity load in individuals with high CTS burden. Thus, childhood trauma might promote the development of certain diseases by altering biological pathways and metabolic processes that might be traced back to the identified genes (Table 1, Table 3). From a clinical point of view, especially the Clusters 5, 6 and 7 are of interest, as they give insights into the genetic risk-profiles for the development of certain diseases depending on their childhood trauma burden. As our clusters are based on depression-related multimorbidity trajectories, we selected the five diseases with the most increased and decreased prevalences within each cluster (Table 3) to draw biological connections towards the suggestive genes identified in our analysis.
Table 3
Biological connections toward the suggested genes, which are genome-wide significant in the GxE analysis, and the five diseases with the most increased and decreased prevalence within each cluster plus effect direction for depression.
Cluster
|
suggested genes in G×E results
|
top five positive and negative associated disorders*
|
link between genes and disorders in the literature
|
1
|
LDLRAD4, MYOF
|
negative: depression, schizophrenia, reaction to severe stress, tonsillitis, allergic rhinitis, dorsalgia
positive: -
|
LDLRAD4: schizophrenia (Kikuchi et al., 2003);
MYOF: none
|
2
|
LDLRAD4
|
negative: depression, schizophrenia, reaction to severe stress, tonsillitis, allergic rhinitis, migraine
positive: -
|
schizophrenia (Kikuchi et al., 2003)
|
3
|
-
|
negative: depression, tonsillitis, allergic rhinitis, migraine, asthma, pain (female genital organs)
positive: hypertension, cerebral infarction, cerebrovascular disease, acute kidney failure, chronic kidney disease
|
-
|
4
|
UGT2A3
|
negative: depression, tonsillitis, allergic rhinitis, migraine, asthma, pain (female genital organs)
positive: hypothyroidism, lipidemia, hypertension, benign prostatic hyperplasia
|
None
|
5
|
MMP16, VGLL2, NRG3, DENND3
|
negative: -
positive: depression, schizophrenia, allergic rhinitis, intervertebral disc disorder, dorsalgia, pain (female genital organs)
|
MMP16: pain (and pain associated disorders like migraine) (Wotton et al., 2022);
NRG3: depression (Paterson et al., 2017), schizophrenia (Avramopoulos, 2018; Li et al., 2020b);
VGLL2, DENND3: none
|
6
|
FOXA3, DAW1, B3GNT7, TREX2, SBK2, FEM1A, HAUS7
|
negative: allergic rhinitis, asthma, migraine
positive: depression, reaction to severe stress, somatoform disorders, nasopharyngitis, bronchitis, soft tissue disorders
|
FOXA3: asthma (Park et al., 2009);
DAW1: allergy (Waage et al., 2018);
B3GNT7, TREX2, SBK2, FEM1A, HAUS7: none
|
7
|
C6orf89, TAAR2, PNLDC1, IL17A
|
negative: alcohol related disorders, nicotine dependence, hypertension, acute kidney failure, chronic kidney disease
positive: depression, allergic rhinitis, asthma, tonsillitis, migraine, dermatitis
|
C6orf89: maybe involved in allergic rhinitis and asthma (Liu et al., 2016; Xu et al., 2017)(regulation of human airway epithelium);
TAAR2: involved in dopamine regulation (Efimova et al., 2022) which is associated with affective disorders (Grace, 2016);
IL17A: kidney disease and transplantation (Kim et al., 2012; Romanowski et al., 2015) and asthma (Holster et al., 2018); inconsistent results for allergic rhinitis (Wang et al., 2012; Fatahi et al., 2016) and atopic dermatitis (Narbutt et al., 2015; Klonowska et al., 2022);
PNLDC1: none
|
*based on weighted Cox proportional hazards regression model (Juhasz et al., 2023) |
The genetic risk profile for Cluster 5 includes the genes MMP16, NRG3, VGLL2, DENND3. Previous associations for MMP16 with nociception can be linked to intervertebral disc disorder, dorsalgia and pain (female genital organs) (Wotton et al., 2022), three high prevalence diseases in Cluster 5 (Juhasz et al., 2023). The already known psychiatric risk gene NRG3, in turn, might have a potential developmental role in schizophrenia (Avramopoulos, 2018; Li et al., 2020b), bipolar disorder and major depression (Paterson et al., 2017). Pain and psychiatric disorders were associated with childhood trauma by overlapping brain mechanism: It has been shown that regions of the brain involved in the pain matrix (such as the anterior cingulate cortex, the amygdala, or the hippocampus) are altered after experiences of childhood abuse and trauma (Teicher et al., 2003; Brown et al., 2018). However, neither VGLL2 nor DENND3 showed any associations with the top five diseases with increased prevalence for Cluster 5, while DENND3 is associated with the volumes of different brain regions (e.g. cortical surface area (Shadrin et al., 2021) cortical thickness (Shadrin et al., 2021; van der Meer et al., 2021), cerebellum (Zhao et al., 2019; Chambers et al., 2022)) and Alzheimer’s disease (Chung et al., 2022).
For Cluster 6, our G×E analysis identified FOXA3, DAW1, B3GNT7, TREX2, SBK2, FEM1A, HAUS7 as suggestive genes. FOXA3 might be involved in the regulation of allergic airway diseases and asthma (Park et al., 2009). As asthma is a potential risk factor for migraine and vice versa, also possible connections between FOXA3 and migraine are conceivable (Wang et al., 2020). Asthma as well as migraine are usually triggered by stress. DAW1 is associated with allergy (Waage et al., 2018), which aligns with allergic rhinitis being one of the high prevalence diseases in Cluster 6. For the remaining genes no associations were reported. However, B3GNT7 may play a role in the formation of neurophils and perineuronal nets in the adult brain (Takeda-Uchimura et al., 2022).
The genetic risk profile for Cluster 7 includes the genes C6orf89, TAAR2, PNLDC1 and IL17A. The gene C6orf89 encodes the bombesin receptor-activated protein (BRAP) which might be involved in the stress response of lung epithelia (Liu et al., 2016; Xu et al., 2017) and can be linked to allergic rhinitis and asthma. In mice, the BRAP homologous protein may have a protective effect on the behavioral response to stress via regulating dendritic spine formation and synaptic plasticity in the hippocampus (Yao et al., 2023). As a consequence, it was concluded that chronic stress might cause damage to hippocampus function (Yao et al., 2023). TAAR2, in turn, expresses a protein which is involved in dopamine regulation and adult neurogenesis (Efimova et al., 2022). As dysregulations in the dopamine system are related to affective disorders like MDD (Grace, 2016), TAAR2 represents a promising target for treating neuropsychiatric disorders (Efimova et al., 2022). IL17A is, among others, associated with chronic kidney disease (Kim et al., 2012) and kidney transplantation (Romanowski et al., 2015) as well as asthma (Holster et al., 2018). Regarding allergic rhinitis (Wang et al., 2012; Fatahi et al., 2016) and atopic dermatitis (Narbutt et al., 2015; Klonowska et al., 2022) studies revealed inconsistent results. Especially, asthma and allergic rhinitis appear to have different genetic risk profiles for IL17A (Resende et al., 2017). A link towards the role of PNLDC1 instead could not be found.
With respect to the candidate genes, we did not confirm their impact in G×E analyses on our MDD-related multimorbidity clusters as a whole. However, some genes suggest a biological connection towards specific individual clusters, underscoring biological heterogeneity stemming from distinct temporal patterns of MDD-related multimorbidity.
From the 18 available candidate SNPs, 5 (located in ABCB1, NTRK2, TNF, IL6, TPH1) showed a nominally significant interaction in at least one cluster (Supplementary Table S4). From these 5 SNPs, no one had a significant interaction in more than three clusters. Hence, our observation that they are not significant in all clusters is in line with Border et al. (Border et al., 2019) and Li et al. (Li et al., 2020a) as they were also not able to confirm the impact of these SNP – Childhood Trauma interactions on MDD in general. However, the effect direction of the candidate SNP for NTRK2 in Cluster 4 was in line with previous findings (Juhasz et al., 2011; van der Auwera et al., 2018; Li et al., 2020a).
IL6 delivered the strongest genetic signal with significant results in both candidate SNP and gene-wise MAGMA analyses in several clusters (Cluster 4 and Cluster 6 on SNP-level, Cluster 3, 4, 6 and Cluster 7 on gene-level). The GWAS catalog lists an association between IL6 and asthma which is among the top five diseases in Clusters 3, 4, 6, 7. Conversely, asthma is not among the top five diseases in Cluster 1, 2, and 5, where the interaction effect with IL6 is not significant. The strong genetic signal of IL6 might be due to the close interrelation between inflammation, stress and depression (Ting et al., 2020). There, it was shown that psychosocial stress acts as a trigger for depression development by initiating changes in HPA-axis and immune/inflammatory system (Ting et al., 2020).
Also, results for TPH1, which catalyzes the first and rate limiting step in the biosynthesis of serotonin, revealed significant interactions on SNP (Cluster 1, Cluster 2 and Cluster 6) as well as on gene level (Cluster 6) in clusters that tend to have a low as well as high CTS burden. As TPH1 is associated with a broad range of psychiatric conditions (Shnayder et al., 2022) this might explain the link toward these clusters. Different effect directions for TPH1 on SNP level are reflected by the decreased prevalence rates of psychiatric conditions in the low-risk Clusters 1 and 2 in contrast to the increased prevalence rates for the high-risk Cluster 6.
On a gene-based level several candidates might be of special interest as they either survive multiple testing within a cluster or show a strong association pattern towards several clusters.
The gene DRD2 (which encodes a D2 subtype of the dopamine receptor) showed a significant interaction effect in several clusters (Clusters 1, 2, 5, and 7; Fig. 3, Supplementary Table S9) and was the only gene that could be confirmed on the gene-level by Border et al. (2019). DRD2 is associated with schizophrenia (Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014; Shioda, 2017), which is among the top five associated disorders in all of these clusters. Interestingly, similar to Border et al. (2019), we found no significant interaction effect of the candidate SNP (rs1800497). This SNP was initially assigned to the DRD2 gene, but was later found to be located in the ANKK1 gene (https://www.ncbi.nlm.nih.gov/gene/1813). Instead, we found a significant interaction for rs4936274 in Clusters 1–5, and 7 (Supplementary Table S6) which is not in LD with the historical candidate SNP. Hence, we propose that rs4936274 might be an alternative candidate SNP for the DRD2 x CTQ interaction on depression, as we observe a clear reversal of the effect direction when comparing clusters with high vs low MDD burden.
Furthermore, the DBH and MTHFR genes revealed significant interaction results with CTS in the gene-based analyses in Cluster 5 (Supplementary Table S9), although the proposed candidate SNP was not available in our analyses. In addition, MTHFR had a nominally significant interaction in Clusters 3 and 4 (Supplementary Table S9) and four independent MTHFR SNPs (r2 < 0.8; Supplementary Fig. S7) showed a significant interaction in five clusters which might also be due to the broad association of MTHFR with neurological and psychiatric disorders (Liew and Gupta, 2015; Zhang et al., 2022).
The oxytocin receptor gene (OXTR) was significant in Clusters 1, 2 and 6 (Supplementary Table S9). Our dataset does not contain the candidate SNP, but in other studies that SNP had no significant interaction effect (Tollenaar et al., 2017; van der Auwera et al., 2018). Instead, we found two independent SNPs (rs60345038, rs62243375) that showed a nominally significant interaction in Cluster 1, 2 and 6. One of them (rs60345038) also had a significant association with social cognitive performance in individuals with schizophrenia (Davis et al., 2014) and could also be a novel risk variant that is possibly linked to and associated with familial type 2 diabetes (Amin et al., 2023).
Our study has several limitations: The CTS, being a retrospective self-reported measurement, is likely to be influenced by recall bias (Baldwin et al., 2019). Further, the cluster probabilities are based on MDD-related multimorbidities, which makes it difficult to compare our results with previous G×E findings for depression, although we present results for the first analysis on G×E interaction for temporal MDD-multimorbidity clusters. In addition, the cluster assignment strongly depends on the reliability of the data from the healthcare system where missassignments can lead to wrong results for the G×E analysis. Due to strong correlations among our parameters (MDD, diseases, environmental factors and CTS), interpreting the correlations between GWASes in terms of causes and mechanisms may prove challenging. Results could be biased by the direct GWAS results for the clusters.
To conclude, our results underscore that some of the former candidate SNPs exert their effects on MDD-related multimorbidity patterns depending on the level of childhood trauma. Such multimorbidity patterns may explain previously inconclusive results on G×E analyses. This genetically based susceptibility for early trauma effects may root in differences in brain phenotypes. Each SNP can have its distributed impacts across numerous brain regions (van der Meer et al., 2020), and these brain-wide differences may establish inter-individual differences in sensitivity to environmental (e.g., traumatic) factors, and thus in multimorbidity patterns, as suggested by our present results. Furthermore, our findings indicate that the moderation of SNP effects by CTS may exert a more prominent influence on the high multimorbidity clusters compared to the low multimorbidity clusters. However, future studies are to reveal the exact etiopathological mechanisms from G×E SNPs through brain phenotypes towards multimorbidity patterns. Regarding the role of candidate SNPs, we conclude that rs4936274 is likely a better candidate SNP for the DRD2 × CTQ interaction on depression than the former candidate SNP and should be tested in further analysis. Investigating MDD-related multimorbidity patterns may be a promising approach in G×E analyses.