The relationship between health-related quality of life and melancholic depressive symptoms is modified by brain insulin receptor gene network

To investigate whether expression-based polygenic risk scores for the insulin receptor gene network (ePRS-IRs) modifiy the association between type of depressive symptoms and health-related quality of life (HRQoL). This cross-sectional study includes 1558 individuals from the Helsinki Birth Cohort Study. Between 2001 and 2004, the Short Form-36 questionnaire was employed to assess mental and physical components of HRQoL and Beck Depression Inventory (BDI) to assess depressive symptoms. Depressive symptoms were categorized into minimal (BDI < 10), non-melancholic and melancholic types of depression. The ePRS-IRs were calculated for the hippocampal (hePRS-IR) and the mesocorticolimbic (mePRS-IR) regions of the brain. General linear regression models adjusted for age, sex, population stratification, lifestyle factors and body mass index were applied to analyze the data. Both types of depressive symptoms were associated with lower HRQoL (p < 0.0001). HePRS-IR modified the association between the types of depression and mental HRQoL (p for interaction = 0.005). Melancholic type of depressive symptoms was associated with higher mental HRQoL compared to the non-melancholic symptoms among individuals with low hePRS-IR (adjusted mean 4.1, 95% CI 0.7–7.4, p = 0.018). However, no such difference was evident in moderate or high hePRS-IR groups as higher hePRS-IR was associated with lower mental HRQoL (B = − 3.4, 95% CI − 5.6 to − 1.2) in individuals with melancholic type of depressive symptoms. No direct associations were detected between the ePRS-IRs and type of depressive symptoms or HRQoL. Variations in the glucose-insulin metabolism can lower HRQoL in individuals with melancholic depressive symptoms.


Scientific Reports
| (2021) 11:21588 | https://doi.org/10.1038/s41598-021-00631-w www.nature.com/scientificreports/ were higher than that of the non-melancholic and vice versa. In the case of equal mean of summary scores, subjects were categorized into melancholic subgroups. The Finnish version of the BDI has been reported as reliable and well validated 35,36 . Covariates. At the baseline clinical examination, subjects answered questions about their current health situation and lifestyle characteristics. Anthropometrics including weight and height were measured with stadiometers (KaWe) and medical scales (SECA alpha 770), respectively. BMI was calculated as weight in kilograms divided with height in meters squared. Systolic and diastolic blood pressure was measured by a mercury sphygmomanometer from the right arm while the subject was sitting and was recorded as the mean of two successive readings. Smoking was coded as never, former and current, and alcohol use was coded as never or having quit, less than once a week or weekly. Highest attained socioeconomic status, obtained from Statistics Finland, was coded as high official, low official, self-employed and manual workers based on the original classification system 37 . The cohort members' past 12-month leisure-time physical activity was assessed using a validated leisure-time physical activity (LTPA) questionnaire; the Kuopio Ischemic Heart Disease Risk Factor Study (KIHD) 38 . Leisure-time physical activity was measured in metabolic equivalents of task 39 , which were multiplied with time (hours) and frequency to calculate MET-hours, as previously suggested 40 . Cohort members were also asked using validated questionnaires about chronic diseases and conditions, including diabetes, cardiovascular conditions (congestive heart failure, arrhythmias, claudication, angina pectoris, previous heart attack and stroke), lung diseases (asthma, emphysema and chronic bronchitis), musculoskeletal disorders (rheumatoid arthritis, osteoporosis) and presence of cancer. The presence of comorbidities was coded as none, one or two or more. Blood samples were obtained to measure levels of total cholesterol, high-density lipoprotein ( Polygenic risk score (PRS). Genotyping and ePRS-IR calculation was performed as previously described 24 .
DNA was extracted from blood samples and genotyping was performed with the modified Illumina 610 k chip by the Wellcome Trust Sanger Institute, Cambridge, UK, according to standard protocols. Genomic coverage was extended by imputation using the 1000 Genomes Phase I integrated variant set (v3/April 2012; NCBI build 37/hg19) as the reference sample and IMPUTE2 software. Before imputing, quality control filters were applied. Specifically, SNP clustering probability for each genotype was set at > 95%, call rate at > 95% for individuals and markers (99% for markers with minor allele frequency (MAF) < 5%), MAF was set at > 1%, and the p value for the Hardy-Weinberg Equilibrium exact test p > 1 × 10 -6 . Moreover, heterozygosity, gender check and relatedness checks were performed and any discrepancies removed. The total number of SNPs in the imputed data was 39282668.
For the ePRS calculation, lists of genes co-expressed with the insulin receptor in the mesocorticolimbic system or hippocampus were created. In the original work 24 , a polygenic risk score for the insulin receptor (PRS-IR) was calculated using SNPs from these gene networks and the association betas from a discovery GWAS (ADHD in the mesocorticolimbic score 43 and Alzheimer in the hippocampal score 44 ). In the current study on HBCS, instead of weighing the SNPs by the association with ADHD or Alzheimer, these linkage disequilibria clumped SNPs were weighted with the betas from the Genotype-Tissue Expression (GTeX), a resource database and tissue bank for examining the relationship between genetic variation and gene expression in human tissues. The final lists of SNPs included 16,556 independent functional SNPs for mesocorticolimbic ePRS (from 263 genes) and 30,652 SNPs (from 498 genes) for the hippocampal ePRS. Final ePRSs were obtained by summation over all SNPs accounting for the sign of correlation coefficient between the genes and insulin receptor gene expression in the different regions. Thus, variations in the score used in the current study represent variations in the expression of the insulin receptor gene network in the specific brain region. In other words, a higher score indicates a higher expression of the insulin receptor gene network. The selection of the SNPs within a given clumping window was based on the lowest p-value. As a result, biologically informed mesocorticolimbic (mePRS-IR) and hippocampal (hePRS-IR) specific insulin receptor polygenic risk scores were calculated. For the analyses the PRS-IRs were standardized and reported both as a continuous and a categorical variable (0 = low = < − 0.5 SD, 1 = moderate = − 0.5 to 0.5 SD, and 2 = high = > 0.5 SD).

Statistical analysis.
The data are reported in means (standard deviation or 95% confidence intervals), medians (interquartile range) or in counts (percentage). The baseline characteristics were analysed with analysis of variance for continuous variables and chi-square test for categorical variables. General linear models (GLM) allowed us to assess the association between components of HRQoL, types of depression, mePRS-IR, hePRS-IR. GLM were also applied to investigate the depression type by PRS-IRs interaction effect on HRQoL. The regression models were adjusted for age, sex, smoking, alcohol usage, socioeconomic status (SES), presence of chronic diseases, BMI, LTPA and population stratification 45,46 , as well as for significant interactions between covariates and depression types on HRQoL when applicable. In the case of not meeting the assumptions (e.g. non-normal distribution) bootstrap style analyses (10,000 replications) were applied when analysing the baseline characteristics as well as when applying general linear models. Multinomial logistic regression with bootstrap style analyses (1500 repetitions) was applied to investigate the association between the depression subgroups. Significance level of 0.005 was applied when testing associations between the hePRS-IRs and subgroups of HRQoL. In all

Results
Our study comprised 1558 subjects, of which 876 (56.2%) were women. The mean age of the participants at the time of the examination was 61.5 years (SD 2.9 for all). Table 1 shows the characteristics of the cohort members according to type of depressive symptoms. Individuals with symptoms of non-melancholic depression suffered more frequently from multiple chronic disease and had higher BMI and blood pressure than individuals with normal BDI score or melancholic type of depressive symptoms. Both non-melancholic and melancholic depression were associated with lower HRQoL (p < 0.001) ( Table 2). A nominally significant association was found between the general health and hePRS-IR (p = 0.04). However, after correcting for multiple testing this association was no longer (Bonferroni corrected p value = 0.40) and www.nature.com/scientificreports/ no association was found between the hePRS-IR or mePRS-IR and the eight items of HRQoL, MCS, and PCS ( Table 3) or type of depression (Table 4, 5). The hePRS-IR was associated with 2 h glucose and insulin concentrations in the OGTT (p values = 0.02 and 0.002, respectively; Table 6), but not with HOMA. Figures 1 and 2 demonstrate the significant interaction between type of depressive symptoms and hePRS-IR on MCS when analysing hePRS-IR as categorical variable (p for interaction = 0.005), respectively as a continuous variable (p for interaction = 0.008). Regardless of the he-PRS-IR status, both non-melancholic and melancholic were associated with lower MCS compared to the BDI < 10 group (Fig. 1). However, in individuals with low hePRS-IR, melancholic depression was associated with higher MCS (4.1 95% CI 0.7-7.4, p = 0.018) compared to non-melancholic group. No such difference was detected among individuals with moderate or high hePRS-IR groups (p ≥ 0.18) (Fig. 1) as MCS decreased linearly in the melancholic group when applying either categorical (p < 0.001, Fig. 1) or continuous hePRS-IR (p = 0.003, Fig. 2).
Further adjustment for glucose regulation status did not substantially alter the found association or interactions (p for interaction = 0.009 for continuous hePRS-IR and p for interaction = 0.006 for categorical hePRS-IR).
Our analyses detected no significant association between type of depression by ePRS-IR interaction on PCS (p > 0.05) nor between type of depression and continuous or categorical mePRS-IR on MCS or PCS (p > 0.05).

Discussion
Our findings indicate that higher hippocampal expression of the insulin receptor gene network is associated with lower mental HRQoL specifically in individuals with symptoms of melancholic depression, suggesting that individual differences in the function of the hippocampal insulin receptor gene network can possibly play a role in the relationship between type of depressive symptoms and HRQoL. No evidence of such modifying effect was found for the physical component of HRQoL nor by the mePRS-IR. This lowering effect decreased as the gene expression score decreased, and HRQoL did not distinguishably differ between melancholic and non-melancholic depressive type in individuals with low ePRS-IR. Lower ePRS-IR in itself showed no association with decreased HRQoL nor with any of the two types of depressive symptoms. The hePRS-IR displayed a significant association with 2 h glucose and insulin concentrations in the OGTT. None of the ePRS-IRs were associated with HOMA. Thus, we showed that the ePRS-IRs measure what they were intended to when they were created, i.e. variations in the glucose insulin metabolism. As such, the application of the ePRSs allowed us to investigate the role of variations in insulin metabolism in specific tissues in individuals regardless of insulin resistance status.
Insulin is considered to be a cognitive modulator 47 and disturbances in glucose and insulin metabolism are linked to several psychiatric conditions, including depression 16 . As established earlier, the ePRS-IRs measure Table 4. The biologically informed hippocampal (hePRS-IR) and mesocorticolimbic (mePRS-IR) polygenic risk score for insulin receptor according to depressive type. Data are shown as predicted mean probability (standard error). Model adjusted for age, sex and population stratification, smoking, alcohol usage, socioeconomic class, body mass index, leisure-time physical activity and comorbidities. BDI Beck Depression Inventory, hePRS-IR hippocampal biologically informed polygenic risk score for insulin resistance, mePRS-IR mesocorticolimbic biologically informed polygenic risk score for insulin resistance, SD standard deviation. P values were based on the bootstrap style method of 1500 repetitions.   Table 5. The association between the biologically informed polygenic risk score for insulin receptor (ePRS-IR) and non-melancholic and melancholic depression with ePRS-IR as continuous variable. Model 1 = adjusted for age, sex and population stratification. Model 2 = adjusted for Model 1 + smoking, alcohol usage, socioeconomic class, body mass index, leisure-time physical activity and comorbidities. hePRS-IR hippocampal biologically informed polygenic risk score for insulin receptor, mePRS-IR mesocorticolimbic biologically informed polygenic risk score for insulin receptor, B regression coefficient. P values were based on the bootstrap style method of 1500 repetitions.  www.nature.com/scientificreports/ variation in insulin glucose metabolism and not metabolic states such as insulin resistance. This could explain the lack of direct association between the ePRS-IRs and depressive symptoms. Both types of depression were associated with lowered HRQoL regardless of genetic markers for different expression of the insulin receptor gene network. However, individuals with melancholic, but not non-melancholic, depressive symptoms showed a clear descending trend in HRQoL with the hePRS-IR. Previous studies have shown an association between non-melancholic depression and peripheral insulin resistance and metabolic syndrome 10,11 , while brain insulin sensitivity has been shown to correspond with favourable body fat distribution and weight loss 48 . The lack of association between non-melancholic depressive symptoms and the ePRS-IRs is compliant with the ePRS-IRs not being associated with HOMA. Individuals with melancholic depression have been found to suffer less frequently from metabolic diseases associated with insulin resistance compared to non-melancholic depression 8 , which could explain the role of high expression of the insulin receptor gene network on HRQoL in our study. HRQoL is lower among individuals with peripheral insulin resistance 49 . According to our results, HRQoL did not display direct association with either of the ePRS-IRs, which suggests that the insulin receptor gene network does not affect HRQoL the same way insulin resistance does. HRQoL has repeatedly been demonstrated to be lowered in depressed individuals 50 , as was the case in our study. Although the modification of the hePRS-IR was associated with further decreased HRQoL in the melancholic depressive subtype, both subtypes have significantly lower HRQoL compared to the individuals with BDI score < 10.
The modification by a high genetic risk score in depressed individuals was detected only by the hePRS-IR whereas the mePRS-IR did not affect HRQoL. Moreover, only the hePRS-IR was associated with worse glucose metabolism. Studies have suggested that changes in the hippocampal region of the brain play part in the pathophysiology of depression 13,14 . The mesocorticolimbic region of the brain has previously been considered to be associated primarily with other neuropsychiatric disorders such as ADHD and addiction 51 . In light of these studies, it can be argued that any modification by high expression of the insulin receptor gene network in depressed individuals would be observed in those with high hePRS-IR, as our results suggest.
The strengths of our study involve the use of the well-characterized HBCS and the use of both clinical and registered-based data in producing our results, as well as the validated SF-36 for assessing HRQoL and the standardized BDI questionnaire for assessing depression. Moreover, the biologically informed ePRS-IR employ whole gene networks instead of single genes, SNPs or discovery GWAS based PRSs to predict risk. However, our cohort is from a homogenous, restricted area in Finland and the findings from this study might therefore have to be cautiously implemented in other populations. Concerning the application of our findings, ePRSs are best utilized in combination with environmental and clinical risk factors when interpreting an individual's lifetime risk of disease.
In conclusion, a high expression of the insulin receptor gene network is associated with lower mental HRQoL in individuals with symptoms of melancholic depression, suggesting that differences in the function of the insulin receptor share an association with the relationship between type of depressive symptoms and HRQoL.