Major Depressive Disorder (MDD) is a prevalent mental disorder with a recurrent or chronic course, often leading to significant suffering, disability, and increased mortality. Treatment resistance affects approximately one-third of patients, highlighting the need for a more comprehensive biological model. MDD exhibits remarkable heterogeneity in its clinical presentation and underlying neurobiology. Genome-wide association studies (GWAS) have identified divergent pathways and cellular processes, but these are often nonspecific. Genetic correlation studies reveal shared genetic contributions with somatic and mental disorders, reflecting the pleiotropy of genetic variants and biological processes. Twin studies suggest that only 40% of phenotype heritability is attributable to additive genetic effects, indicating the involvement of comorbid conditions and non-genetic factors. Including additional phenotypic information, such as multimorbidity patterns, can reduce genetic heterogeneity and improve the development of aetiological models. A recent study demonstrated strong multimorbidity patterns among numerous diseases, both phenotypically and genetically, revealing correlations between psychiatric and cardiovascular/respiratory disorders. This multimorbidity paradigm shift emphasizes the importance of considering these patterns when identifying depression subtypes and associated biological pathways. The TRAJECTOME project aimed to leverage information on trajectories of MDD-related multimorbidities from large population cohorts to identify biologically and clinically informative depression subtypes, their neurobiological and genetic backgrounds, and potential biomarkers for precision medicine. The central hypothesis was that age-dependent, strongly relevant MDD-related multimorbidities would enrich the genetic basis of MDD, leading to the identification of participant clusters with distinct genetic profiles contributing to the pathology.

Existing studies attempting to identify subgroups within chronic somatic diseases primarily used cross-sectional data and did not consider associations with psychiatric disorders. Recent advancements extended this approach to genetic data, explaining a significant proportion of multimorbidities based on shared genetic components and identifying central hub diseases. A Danish study identified time-dependent psychiatric multimorbidity clusters in schizophrenia patients, associated with heterogeneity in aetiological factors. However, these studies lacked a comprehensive approach integrating temporal multimorbidity trajectories with genetic and non-genetic risk factors to define MDD subtypes.

The TRAJECTOME project utilized data from 1,576,598 participants across seven European general-population cohorts. Dynamic Bayesian networks were employed to identify a minimal, non-redundant set of MDD-related multimorbidity trajectories. This involved three steps: (1) identifying a minimal set of multimorbidity trajectories relevant to MDD using the Markov boundary concept; (2) developing a data-driven method to measure patient similarity in this filtered set; and (3) performing privacy-preserving Bayesian federated clustering across cohorts. Five cohorts (N=1,189,509) were used for discovery, and two (N=387,089) for validation. The cohorts varied in age range, birth year, socioeconomic factors, and MDD prevalence rates. Eighty-six predetermined cross-cohort diseases strongly related to MDD were selected for analysis. Weighted direct MDD-related multimorbidity scores were computed for each participant, using these scores as input for cluster analysis. Seven clusters were identified, reflecting different temporal trajectories of MDD-related multimorbidity burden. The clinical characteristics of these clusters were investigated by examining temporal disease patterns, mean onset ages, and disease risk using Cox regression and Kaplan-Meier estimates. Genome-wide association studies (GWAS) were conducted in the UK Biobank cohort to explore the genetic contribution of each cluster. Polygenic risk scores (PRSs) were used to validate genetic findings in the Finnish cohorts. Non-genetic risk factors, including behavioral and physiological factors, were also examined within each cluster. Validation of the MDD-related multimorbidity profiles was performed at both genetic and non-genetic risk factor levels in additional cohorts, including one with limited disease information (SHIP). The biological meaningfulness of the profiles was evaluated by calculating PRSs and comparing non-genetic risk factor profiles with those of the UK Biobank cohort.

Seven distinct MDD-related multimorbidity clusters were identified, exhibiting different temporal trajectories of disease burden throughout the lifespan. Clusters 1-4 showed a low prevalence of MDD and related disorders, later onset ages, and a longer period of low disease burden. Clusters 5-7 displayed a high prevalence of MDD and related disorders. GWAS analyses revealed 6141 distinct genome-wide significant SNPs spanning 42 risk loci on 20 chromosomes in the UKB cohort. Clusters 1-4 showed a significant association with immune system-related genes, including major histocompatibility complex genes (HLA genes), receptors (interleukin- and Toll-like receptors), and cytokines. Clusters 5-7 exhibited distinct genetic patterns. Cluster 5 showed overlap with psoriasis, Cluster 6 with cardiovascular conditions, asthma, rheumatoid arthritis, and blood measures, and Cluster 7 demonstrated the strongest genetic contribution and negative correlation with other clusters. The non-genetic risk factor profiles of the clusters were consistent with clinical and genetic findings. Clusters 1-2 were associated with higher age and lower behavioral risk factors. Clusters 3-4 showed increased behavioral and physiological risk factors, while Clusters 5-6 displayed increased risk overall, and Cluster 7 had a more favorable profile. Validation in independent cohorts largely replicated the findings at genetic and non-genetic levels, supporting the robustness and generalizability of the results. The SHIP cohort, with limited disease information, showed an accuracy of 67.5% in replicating the clusters.

The TRAJECTOME project successfully identified seven MDD-related multimorbidity clusters with distinct clinical, genetic, and non-genetic risk factor profiles, highlighting the involvement of neuroinflammatory processes. Four clusters (1-4) exhibited low MDD risk, favorable genetic and non-genetic profiles, and late onset of age-related disorders. Clusters 5-6 displayed high MDD risk, unfavorable profiles, and earlier onset. Cluster 7 showed a strong contribution of inflammation-related genetic predispositions, with increased prevalence of MDD and respiratory disorders despite a generally low disease burden. The observed heterogeneity in MDD risk might explain previous contradictory findings regarding the relationship between MDD and inflammatory genes. The results corroborate existing inflammatory, metabolic, and related hypotheses regarding depression's pathophysiology. The study's findings demonstrate that integrating temporal multimorbidity trajectories with genetic and non-genetic factors provides valuable insights into MDD subtypes, informing personalized prevention and treatment strategies.

This study identified seven MDD-related multimorbidity clusters with unique genetic and non-genetic risk-factor profiles, emphasizing the role of neuroinflammatory processes. These findings provide a framework for subtyping depression patients and guiding personalized prevention, early intervention, and therapeutic approaches. This approach can be extended to other complex diseases with high comorbidity burdens.

The study's limitations include differences in healthcare systems across cohorts affecting disease rates and the inability to distinguish between chronic and acute diseases. The Bayesian network methodology is sensitive to unknown confounders and selection bias, although the large and comprehensive nature of the cohorts mitigates this risk. Further research is needed to explore the clinical implications of these findings and to develop targeted interventions.