Genetic and Environmental Contribution to the Co-Occurrence of Endocrine-Metabolic Disorders and Depression: A Nationwide Swedish Study of Siblings

Depression and endocrine and metabolic disorders are major causes of disability and mortality, representing a growing public health concern at the global level (1–3). Several studies have established elevated co-occurrence of depression and endocrine-metabolic diseases, including thyroid gland disorders, type 1 diabetes (T1D), type 2 diabetes (T2D), obesity, and polycystic ovary syndrome (PCOS) (4–8). A bidirectional relationship has been reported between depression and obesity, as well as between depression and T2D, where one disorder increases the risk of subsequent development of the other and vice versa. However, other studies examining the bidirectional phenotypic causation hypothesis have reported mixed results (6, 9, 10).

Several biological, psychological, and behavioral factors have been proposed to influence the associations between these conditions, acting as mediating mechanisms in the causal relationships between depression and endocrine-metabolic diseases, and/or as common underlying factors affecting the risks of these conditions. Shared genetic influences could represent some of these common origins contributing to the development of such co-occurring disorders. This is known as genetic pleiotropy, when certain conditions or traits share genetic risk variants conferring a disposition for both disorders or traits in the same individual. Studies of families and twins have shown that genetic factors explain ∼40% of the phenotypic variation in liability for depression (11), ∼40% for obesity (12), 81% for T1D (13), 72% for T2D (14), between 38% and 71% for PCOS (15), and ∼60% for both Hashimoto’s thyroiditis and Graves’ disease (13). Growing evidence suggests genetic overlap between depression and some endocrine-metabolic disorders, in particular obesity (16), T2D (17), and PCOS (4). This finding has been supported by a number of studies using molecular genetic data (18–20), while other investigations failed to replicate some of these results (21, 22). The inconsistency in evidence is likely due to differences in disorder measurements, depression heterogeneity, and methodological approaches.

A greater understanding of the possible risk factors shared between these conditions is needed in order to elucidate potentially modifiable targets that could lead to tailored treatment strategies for individuals with co-occurring depression and different endocrine-metabolic disorders. Longitudinal population-based research using clinical diagnoses could help enlighten the etiological underpinnings of these conditions.

We investigated the familial coaggregation of clinically diagnosed depression and endocrine-metabolic disorders to test for the presence of shared familial liability, and we estimated the genetic and environmental contributions to the co-occurrence of these conditions using quantitative genetic modeling.

Methods

Study Population

We conducted an observational cohort study using nationwide Swedish registers linked by unique personal identity numbers (23). The cohort selection process is illustrated in Figure 1. All individuals born in Sweden between 1973 and 1996 were identified from the Medical Birth Register (24). Demographic information was obtained from the Total Population Register (25), Cause of Death Register (26), and Multi-Generation Register (27). We excluded children with severe congenital malformations, stillbirths, and neonatal deaths. We further excluded individuals who emigrated or died before age 10, had missing information regarding their biological parents, or were adopted. The final cohort consisted of 2,263,311 individuals who were followed up from birth until emigration, death, or December 31, 2013, whichever occurred first. Each individual was linked to their biological parents to create three cohorts of siblings: full siblings, maternal half siblings, and paternal half siblings (see Table S1 in the online supplement). Monozygotic twins were excluded from the sibling analyses because their genetic sharing is higher than that of other full siblings. This study necessitated no informed consent and was approved by the Regional Ethical Review Board in Stockholm (2013/862-31/5).

Assessment of Endocrine-Metabolic Disorders and Depression

We used information from the National Patient Register (28), which reached full nationwide coverage for hospitalizations in 1987 and 80% coverage for outpatient diagnoses in 2001, coded according to ICD-8, ICD-9, and ICD-10. We investigated endocrine-metabolic disorders diagnosed during the follow-up period, grouping these into autoimmune disorders (autoimmune hypothyroidism, Graves’ disease, and T1D) and non-autoimmune disorders (T2D, obesity, and PCOS), including the first-time diagnosis for each disease, irrespective of prior endocrine-metabolic disorders. Because PCOS affects only women, only females were included in the analyses investigating this disorder. Depression cases were defined as individuals who received a depression diagnosis during the follow-up. A complete list of all ICD codes included in this study is provided in Tables S2 and S3 in the online supplement.

Statistical Analysis

Association and Familial Coaggregation Analyses

We used logistic regression and Cox proportional hazards regression models to evaluate the association between endocrine-metabolic disorders and depression, comparing unadjusted (logistic regression) and adjusted (Cox regression) estimates for follow-up time. For the logistic regression, we coded exposures (endocrine-metabolic disorders) and outcome (depression) as lifetime diagnoses (0 or 1 for absence or presence, respectively), and estimated odds ratios with 95% confidence intervals. In the Cox models, we computed hazard ratios and 95% confidence intervals, using attained age as the underlying timescale (i.e., we compared individuals of the same age) and time-varying exposure—where an individual was considered unexposed until the first observed diagnosis of the exposure condition, and exposed afterward. Since Cox regression takes into account time at diagnosis, we compared models in which the exposure was an endocrine-metabolic disease with models in which the exposure was depression, to explore whether the order of these diagnoses influences the associations.

First, we compared the risk of depression in individuals with and without endocrine-metabolic disorders in the full cohort. Then we evaluated familial coaggregation patterns of these disorders, analyzing full siblings, maternal half siblings, and paternal half siblings separately, comparing the risk of the outcome disorder in individuals with a sibling with and without the exposure disorder. If an association between two disorders is observed in the index person, and if the association exists but is lower in the full-sibling sample, this indicates that shared familial liability contributes to the phenotypic association between these conditions. Further comparison of different types of siblings can point to the source of familial liability (genetic or environmental). Full siblings share on average 50% of their segregating genes, whereas half siblings share 25%. Full and maternal half siblings tend to share environmental factors to a greater extent than paternal half siblings, as they had a similar prenatal environment and often reside predominantly with their mothers after parental separation (29). Thus, a greater association among full siblings compared with maternal half siblings suggests that genetic factors may be of importance, whereas a greater association in maternal compared with paternal half siblings indicates shared environmental contributions.

All possible combinations of sibling pairs were identified, with each individual contributing at least once as an index person and once as a sibling. The analyses were adjusted for sex and birth year to account for differences in disorder prevalence between sexes and over the study period. We used a cluster robust (sandwich) estimator for standard error calculation, clustered by family, to account for nonindependence of family data (30). A p value <0.05 was considered statistically significant. Additionally, to adjust for multiple testing, we report a Bonferroni-corrected significance level of p<0.005, accounting for the 11 main analyses. Data management was performed using SAS, version 9.4.6 (SAS Institute, Cary, N.C.), and analyses were performed using R, version 4.0.3 (R Foundation for Statistical Computing, Vienna).

Sensitivity Analyses

We conducted several sensitivity analyses. First, we restricted the study population to individuals born between 1987 and 1996 (rather than 1973–1996) to minimize loss of early diagnoses for the oldest individuals in the main cohort due to the fact that inpatient information from the National Patient Register reached nationwide coverage in 1987. Second, because females have a twofold increased risk of depression compared with males (31), we conducted analyses stratified by sex of the outcome person to investigate differences between sexes. Finally, we performed analyses stratified by parental educational attainment as an indicator of socioeconomic status, defined as low (primary school and lower and upper secondary school) or high (≥1 year of university), using information from the Longitudinal Integration Database for Health Insurance and Labor Market Studies (32).

Quantitative Genetic Analyses

Quantitative genetic modeling was used to estimate the contribution of genetic, shared, and nonshared environmental factors to the familial liability and to evaluate the genetic and environmental correlations across disorders (33). In these analyses, full and maternal half sibling pairs were included. Within each family, we selected only one pair of siblings (the oldest, for a longer follow-up time) to avoid dependency between sibling pairs from the same family and to increase the precision of the estimates, resulting in 691,499 full sibling pairs and 86,948 maternal half sibling pairs (for PCOS, only female individuals were included: 211,178 full sibling pairs and 29,418 maternal half sibling pairs). Given the low prevalence of Graves’ disease in the sibling cohorts, this disorder was not included in the quantitative genetic analyses. Because diagnoses were treated as binary variables, we used a liability threshold framework, where disease risks were assumed to represent an underlying normal distribution of liability and the risk was above an estimated threshold if the disease was observed and below the threshold if no disease was observed (33). Correlations between these assumed normally distributed liabilities are referred to as tetrachoric correlations. We specified additive genetic (A), shared environmental (C; nongenetic factors that make siblings in a pair similar), and nonshared environmental (E; unique environmental components making siblings dissimilar) factors as sources of variance and covariance between depression and endocrine-metabolic disorders. The following assumptions were made: A factors correlate at 0.50 for full siblings and 0.25 for half siblings, as they share 50% and 25% of their cosegregating genes, respectively; C factors correlate at 1 for both full and maternal half siblings; E factors correlate at 0 across all siblings. For each combination of depression and endocrine-metabolic disorders, we fitted bivariate models that included A, C, and E components (ACE model). Because the prevalence of disorders tends to differ in nuclear families and extended families, prevalence rates were allowed to vary across full and half siblings (34). Analyses were adjusted for sex and birth year of both the index person and the sibling (except for PCOS analysis, where no sex adjustment was made) and were performed in OpenMx (35) using weighted least squares. More details on the model are provided in the supplementary methods section in the online supplement.

Results

The demographic and clinical characteristics of the study population are summarized in Table 1. A total of 2,263,311 individuals (48.8% female) were followed up for a median of 27 years (IQR=22, 34). The prevalence of depression was 5.2% during the observed period (N=118,051; 63.0% female), and it was generally consistent across advancing birth years. The mean age at first recorded inpatient or outpatient diagnosis of depression was 23 years for both sexes, with the majority of individuals receiving their first diagnosis between ages 17 and 30, reflecting a lower risk of depression in childhood as well as limited follow-up for the youngest participants in the study cohort. Individuals with depression appear to have a higher prevalence of endocrine-metabolic disorders than individuals without depression (see Table S4 in the online supplement for sex-specific prevalence), particularly of autoimmune hypothyroidism and obesity. The prevalences of all conditions among sibling cohorts are reported in Tables S5–S7 in the online supplement.

Association and Familial Coaggregation Analyses

Associations between endocrine-metabolic disorders and depression, and their familial coaggregation, are shown in Figure 2. Compared to the general population, individuals diagnosed with endocrine-metabolic disorders had statistically significantly increased risk of depression (for exposure to any autoimmune disease, odds ratio=2.05, 95% CI=1.99, 2.12; for exposure to any non-autoimmune disease, odds ratio=2.33, 95% CI=2.28, 2.39). The strongest associations were observed for exposure to T2D (odds ratio=3.48, 95% CI=3.25, 3.72), obesity (odds ratio=2.44, 95% CI=2.37, 2.50), and autoimmune hypothyroidism (odds ratio=2.34, 95% CI=2.25, 2.44). The risk of depression was considerably lower but present in full siblings of individuals with endocrine-metabolic disorders compared with full siblings of individuals without endocrine-metabolic disorders (odds ratios ranged from 1.15 [95% CI=1.02, 1.29] for exposure to Graves’ disease to 1.53 [95% CI=1.37, 1.69] for exposure to T2D). Associations between depression and T2D, obesity, autoimmune hypothyroidism, and PCOS were higher in full siblings compared to maternal half siblings, suggesting genetic influences on each disorder. In contrast, no difference in depression risk was observed between full, maternal half, and paternal half siblings of individuals with T1D, potentially indicating that unique environmental factors may underlie the association between the two disorders. Finally, the magnitude of odds ratios observed between maternal and paternal half siblings was overall comparable, suggesting a weak role of shared environmental factors in explaining these associations.

Overall, estimates were consistent when using Cox models, regardless of whether depression was modeled as the outcome or the exposure disorder (see Figure S1 in the online supplement).

Estimates, p values, and Bonferroni-corrected p values for all analyses are listed in Tables S8–S10 in the online supplement.

Sensitivity Analyses

When restricting the study population to individuals born between 1987 and 1996 (rather than between 1973 and 1996), patterns of association were comparable to those observed when using the full cohort (see Figure S2 in the online supplement). Statistically significant sex differences were observed for exposure to autoimmune hypothyroidism (females: odds ratio=2.27, 95% CI=2.17, 2.37; males: odds ratio=3.45, 95% CI=3.07, 3.88) and obesity (females: odds ratio=2.32, 95% CI=2.25, 2.40; males: odds ratio=3.02, 95% CI=2.85, 3.20) (see Figure S3 in the online supplement). Finally, parental educational attainment did not appear to influence the associations between endocrine-metabolic disorders and depression (see Figure S4 in the online supplement).

Quantitative Genetic Analyses

Heritability was estimated to be 39% in depression; approximately 70% in obesity, T1D, and T2D; 57% in autoimmune hypothyroidism; and 28% in PCOS (see Table S11 in the online supplement).

In agreement with the observed pattern of odds ratios of depression in individuals with and without endocrine-metabolic disorders, stronger phenotypic correlations were observed between depression and T2D (odds ratio=0.22, 95% CI=0.20, 0.23), obesity (odds ratio=0.19, 95% CI=0.18, 0.20), and autoimmune hypothyroidism (odds ratio=0.17, 95% CI=0.16, 0.18), with genetic factors explaining 74%, 56%, and 34% of these correlations, respectively, and unique environmental factors explaining 33%, 32%, and 53%, respectively (Figure 3). In contrast, we found little evidence of genetic influences underlying the correlation between depression and T1D, with nonshared environmental factors explaining 84% of the phenotypic correlation (odds ratio=0.11, 95% CI=0.10, 0.13). Finally, our results pointed to a weak shared environmental contribution to the co-occurrence of depression and all endocrine-metabolic disorders. Estimates (with 95% confidence intervals) of the phenotypic correlations and the bivariate heritability explained by A, C, and E are listed in Tables S12 and S13 in the online supplement. It is important to note that the confidence intervals for some of these estimates were particularly wide due to model optimization issues.

Discussion

In this nationwide register-based study of 2,263,311 individuals, we explored the associations between endocrine-metabolic disorders and depression using two complementary designs: familial coaggregation analysis and quantitative genetic modeling. To our knowledge, this is the largest sibling study investigating these associations and the first study to quantify and decompose the correlations between depression and T1D into genetic and environmental etiologies.

In line with previous research, we showed that individuals with endocrine-metabolic disorders were at increased risk of depression compared with the general population, especially for exposure to T2D, obesity, and hyperthyroidism (2.3–3.5 times higher odds), with males displaying elevated relative risks for the latter two compared to females’ relative risks. Full siblings of individuals with endocrine-metabolic disorders were also at higher risk of depression, suggesting that shared familial liability contributes to the co-occurrence of these conditions.

For the non-autoimmune endocrine-metabolic disorders (T2D, obesity, and PCOS), the shared familial liability was primarily due to genetics, with very small contributions from shared environmental factors, as suggested by the familial coaggregation analysis (i.e., the strongest associations with depression were found between full siblings, followed by maternal and paternal half siblings), and these findings were further supported by quantitative genetic modeling. Our results are consistent with previous studies suggesting shared genetic origins between depression and these conditions (4, 16, 17), which is also reflected in the shared mechanisms linking these disorders (36). For example, a number of common biological mechanisms have been proposed to explain the bidirectional depression-obesity link, including hyperactivation of the hypothalamic-pituitary-adrenal axis, immuno-inflammatory activation, and neuroendocrine system dysregulation (6). These biological pathways may act as common underlying mechanisms influencing the liability to both depression and obesity, and/or as mediating factors in the causal link between the two conditions. Similar shared biological origins have been suggested to explain the bidirectional relationship between depression and T2D (10). Intuitively, associations between obesity, T2D, and PCOS have also been established, all displaying common metabolic abnormalities such as insulin resistance (37, 38). Overall, our findings suggest that similar etiologies for depression and comorbid obesity, T2D, and PCOS exist, and that family history may help identify risk factors for these co-occurring disorders. Furthermore, characterization and modification of upstream biomarkers of both depression and these non-autoimmune conditions could potentially prevent development of such chronic and debilitating illnesses. It is important to note that ∼30% of the phenotypic correlation between depression and non-autoimmune endocrine-metabolic disorders was attributable to environmental influences not shared between siblings. Behavioral and psychosocial factors such as sedentary lifestyle, poor nutrition, and poor sleep hygiene may represent some of these risk factors, which are accessible and modifiable, and which should remain key targets for prevention strategies and treatments of both depression and these endocrine-metabolic disorders. Some antidepressants have also been shown to influence the risk of obesity and T2D, although inconsistency of the evidence highlights the need for further investigation of the causal role of antidepressants in the development of these conditions (10, 39).

In contrast, the phenotypic correlations between depression and autoimmune endocrine-metabolic disorders (autoimmune hypothyroidism and T1D), while not high (0.17 and 0.11, respectively), were largely explained by nonshared environmental factors. This was especially relevant in the association between T1D and depression, for which no evidence of shared genetic influences emerged. Our results are in line with a previous study, which found a weak genetic contribution to the co-occurrence of depression and autoimmune disorders using cross-trait polygenic risk score analyses and linkage disequilibrium score regression (40).

This may reflect the existence of a direct link between these disorders via environmental mechanisms and/or causal environmental factors influencing the risk of both conditions. In accordance with emerging evidence, our study showed that although both T1D and T2D result in persistent hyperglycemia, their association with depression appeared to be driven by different underlying mechanisms. T1D usually has an earlier onset than T2D, with the potential to disrupt developmental processes during childhood and adolescence. Psychological mechanisms (e.g., parental stress and the burden of a lifelong disorder requiring a complex management regimen) as well as biological mechanisms (e.g., inflammation and cerebral damage) have been suggested to mediate the link between T1D and depression (7).

From the quantitative genetic analyses, negative genetic and environmental contributions were observed for some phenotypes, for which there may be a number of possible reasons. For example, it is possible for a parameter whose true value is zero to be negative due to sampling error, which is particularly likely in smaller samples (41). Another possibility is that such negative estimates actually reflect the underlying genetic or environmental mechanisms under investigation (42, 43), for example, genes that have opposite effects on two phenotypes, or full-sibling cross-disorder correlations that are larger than the phenotypic correlations. Replication studies with larger sample sizes are needed.

Parental education level did not appear to influence the association and familial coaggregation of depression and endocrine-metabolic disorders; moreover, these findings were consistent in the quantitative genetic analyses, where we observed only a weak contribution of environmental factors shared between siblings (e.g., parental academic performance) to the co-presentation of these disorders.

Findings from the familial coaggregation analyses were consistent when comparing logistic and Cox regression models, indicating that there was no major difference when modeling the explored diagnoses as lifetime conditions or when considering time to diagnosis. Furthermore, similar results were obtained when depression was modeled as the outcome condition or as the exposure condition, suggesting a possible bidirectionality between the investigated disorders, but also encouraging caution when using diagnosis ordering to infer causality, since diagnostic records do not provide definitive information about when health problems occurred.

This study enhances our understanding of co-occurring depression and endocrine-metabolic disorders, and it underscores the importance of screening for comorbid symptoms in individuals with these conditions, which may lead to risk identification and early detection of concomitant conditions, as well as promotion of appropriate prevention and intervention strategies. Thus, collaboration between psychiatrists and endocrinologists should be prioritized in an effort to comprehensively manage both psychiatric and somatic conditions. Moreover, we present evidence of stronger shared genetic influences between depression and T2D, obesity, and PCOS compared with autoimmune disorders, suggesting different etiologies for these comorbid conditions and the need for tailored treatment strategies for individuals with depression and different comorbid endocrine-metabolic conditions.

This study had several limitations. Depression and endocrine-metabolic disorders treated within primary care could not be captured, possibly limiting the generalizability of the findings to the more severe clinical cases, as well as to treatment-seeking individuals. Yet, diagnoses given to children and adolescents are well captured in the National Patient Register, as it is standard for these patients to be referred to specialist health care for these conditions.

Exposure to T1D and T2D could have been misclassified, since ICD-8 and ICD-9 did not distinguish between the two types. Nevertheless, we were able to minimize this issue by restricting T1D measures to individuals with a diagnosis of diabetes before age 19, given that the majority (98%–99%) of Swedish children diagnosed with diabetes at 0–18 years of age have been shown to have T1D (44). Since the maximum age at the end of follow-up was 40 years, we could investigate only early-onset forms of T2D, autoimmune hypothyroidism, and Graves’ disease, which could represent more heritable forms of these disorders and therefore may explain why we found stronger genetic contributions to these disorders compared to a previous study using linkage disequilibrium score regression (45). Similarly, we defined obesity using clinical diagnoses rather than BMI measures, potentially capturing a more serious form of the disorder, which could explain the increased heritability estimated in this study compared with previous research (45, 46). Moreover, depression heterogeneity could have hampered the results, since the explored associations are likely to vary in different subgroups of patients. Further research is needed to investigate the role of age at onset of depression, symptom profiles, severity, depression history, and other psychiatric comorbidities.

Given the low power in our quantitative genetic analyses, caution in interpretation of the genetic and environmental correlations is warranted, and replication studies with increased sample sizes are encouraged. Furthermore, one key assumption of quantitative genetic analyses is that there are no correlations or interactions between genes and environment (47). As a consequence, estimates may be biased if this assumption does not hold. Future research assessing gene-environment interplay will be important to gain further insights into the etiology of depression and endocrine-metabolic disorders.

When investigating the possible role of socioeconomic status, we performed our analyses across different parental education levels; given that academic performance has been shown to be moderately heritable and correlated with offspring education (48), there is a risk that adjusting for parental educational attainment could lead to biased estimates (e.g., by conditioning on a collider). Finally, although we found no evidence of a major effect of socioeconomic status in the explored associations, it is important to note that parental educational attainment may inadequately capture the complexity of socioeconomic status.

In summary, we presented convergent evidence from two methods, supporting knowledge from previous research on the comorbidity between depression and various endocrine-metabolic disorders and providing new insights into the etiological sources of their co-occurrence. We reported considerable shared genetic influences between depression and non-autoimmune endocrine-metabolic disorders, and limited common genetic factors between depression and autoimmune disorders, particularly T1D. Our results encourage clinical vigilance for co-presentation of endocrine-metabolic disorders and depression, providing a useful foundation for future research aimed at characterizing and targeting the underlying biological mechanisms and modifiable risk factors.