019_ Genetic association and protein-protein interaction networks between MC4R and obesity in bipolar affective disorder
Research Question and Aims
Bipolar disorder (BD) is a severe, devastating disorder characterized by mood swings from depression to mania or hypomania (Dilling H, 2010). Obesity frequently occurs in patients affected with BD independent of medication side effects (McIntyre et al., 2010a, McIntyre et al., 2010b, Goldstein et al., 2013). These observations strongly indicate that obesity and BD share common pathways. A more refined knowledge is of clinical relevance, given the fact that cardiovascular comorbidities are responsible for a nine-year shorter life span in BD (Kasper S, Kapfhammer HP, Bach M, Butterfield-Meissl C, Erfurth A, Haring C, Hausmann A, Hofmann P, 2013). Therefore, it is of high clinical relevance to understand shared biological pathways and molecular pathways associated with BD and obesity to find better therapeutic targets.
Results from the "BIPFAT" study in Graz (PI Eva Reininghaus) showed that obesity and obesity related mechanisms such as oxidative stress, Endoplasmatic Reticulum stress, chronic inflammation, imbalances in the kynurenine-system and altered circadian mechanisms are associated with Bipolar Disorder (Bengesser et al., 2018, Dalkner et al., 2018, Morkl et al., 2018, Queissner et al., 2018, Bengesser et al., 2016, Bengesser et al., 2016, Mangge et al., 2019, Bengesser et al., 2015, Lackner et al., 2015a, Lackner et al., 2015b).
A panel of risk genes predicting antipsychotics induced weight gain (AIWG) was implemented at the pharmacogenetics department, CAMH Toronto, which included GLP-1, Orexin A, Orexin B, OX1R, OX1R, NPY, NDUFS1, TSPO, HCRTR2, GCG, GABRA2, CNR1 and MC4R. More precisely, the Canadian collaboration partners established a model including the gene variants CNR1 (rs806378), NPY (16147), GCG (rs13429709), MC4R (rs489693), HCRTR2 (rs3134701 and rs4142972), TSPO (rs6971) and NDUFS1 (rs6435326) to predict antipsychotics induced weight, but was not yet analyzed in BD. Most promising is the melanocortin 4 receptor (MC4R) gene, which belongs to high risk-genes for obesity and antipsychotic induced weight gain (AIWG), but protective gene-variants also lead to decreased appetite and resilience. Melanocortins are expressed in the hypothalamus and play an important role in appetite regulation as stimulation of brain melanocortin results in a decrease in food intake and weight. Mutations and gene variants of MC4R have been implicated in severe forms of obesity (Yilmaz et al., 2015). Numerous studies suggested impact of genetic variants of MC4R on overweight, obesity, type 2 diabetes (T2D) and antipsychotic induced weight gain (AIWG). AIWG is a leading factor in patient non-compliance and has previously been shown to increase the risk of T2D, metabolic syndrome, and cardiovascular events (Lett et al., 2012). Numerous factors contribute to inter-individual differences in the risk for AIWG, including age, diet, smoking, eating habits, concurrent medications, and most important genetic factors, including variations/alterations in MC4R and other AIWG associated genes (Malhotra et al., 2012, Chowdhury et al., 2013).
As previous studies have additionally shown that obese individuals with BD suffer in many cases from more severe illness course than individuals with BD being normal-weight. Understanding of the MC4R pathways and recognition of possible genetic variations may lead to new therapeutic options to prevent weight gain in BD (Roubert et al., 2010, Clement et al., 2018) or to positively influence the course of BD itself. To date genetic analyses of MC4R in obesity of patients with BD and medication targeting MC4R in BD treatment are completely lacking and the current investigation will help closing this scientific gap. Furthermore, the network-analysis of MC4R and interacting genes (encoding for interacting proteins) is novel and will improve scientific knowledge about obesity associated mechanisms in BD!
The aim of the current investigation is to investigate the influence of known risk factors for obesity (MC4R* i.e. MC4R and all directly interacting proteins) in general population and AIWG on weight gain and obesity during BD course. Furthermore, it would be great if we can check your dataset to see if these polymorphisms CNR1 (rs806378), NPY (16147), GCG (rs13429709), MC4R (rs489693), HCRTR2 (rs3134701 and rs4142972), TSPO (rs6971) and NDUFS1 (rs6435326) or their surrogates (SNPs with r2>0.8) are present in the genome-wide genotyping dataset of the renown PsyCourse study and if they are associated with obesity and disease course in BD. We hypothesize that AIWG associated genes, especially MC4R and its gene-network encoding for interacting proteins are associated with obesity in BD and disease course.
I. First, we hypothesize that AIWG associated gene variants (e.g. MC4R* gene variants and genes involved in its network) are associated with obesity and weight gain in BD.
II. Second, we hypothesize that obesity and MC4R* genotypes are associated with the course and severity of BD (e.g. GAF score, CGI, BD I vs. II, number of affective episodes).
III. Thirdly, we will analyze MC4R and its related gene-network with modern system-biology approaches to depict genes encoding for proteins interacting with the gene product of MC4R to decipher how MC4R confers to obesity related mechanisms in BD.
We will need the following phenotypic data for exploratory analysis, sample description and covariates in regression models for cases:
-Anthropometric measures and obesity associated comorbidities at baseline (V1) for all patients and for each visit (V2, V3, V4):
Body mass index
Elevated cholesterol or triglyceride levels
hypertension, diabetes, heart attacks, angina pectoris, stroke, asthma, kidney failure, COPD/chronic Bronchitis, traumatic brain injury, kidney failure, cancer, hyperthyroidism, hypothyroidism, infectious diseases, liver cirrhosis or inflammation.
-Diagnosis and course of disease: diagnosis according to DSM-IV, DSM-IV diagnosis categories, Bipolar-I Disorder Bipolar-II Disorder, number of manic episodes, number of MDD episodes, number of hypomanic episodes, number of suicide attempts, Clinical Global Impression (CGI), illness severity scale, Global Assessment of Functioning (GAF)
age, sex, Smoking: "How many cigarettes do you presently smoke on average?", "Did you ever smoke tobacco products?", alcohol: "Lifetime alcohol dependence?", markers for measuring indirectly loneliness and life events: marital status, relationship status "Do you currently have a partner?", children, siblings, living alone, life events precipitating illness episode between study visits, change in housing situation since last study visit?, education: high-school level education, professional education, currently paid employment, disability pension due to psychological/psychiatric illness, duration of illness, first-episode patient
- medication: psychiatric medication change during the past six months, current antipsychotics intake (yes/no), current antidepressants intake (yes/no), current antiepileptics intake (yes/no), current lithium intake (yes/no).
-family history: family member ever affected by psychiatric disorder
-comorbidities: hyperthyroidism, hypothyroidism, heart attacks, hypertension, cancer
- disturbed sleep from Inventory of depressive symptomatology (IDS-C30): items for disturbed sleep:
Item 1 Sleep onset insomnia, Item 2 Mid-nocturnal insomnia, Item 3 Early morning insomnia, Item 4 Hypersomnia,
-Young Mania Rating Scale
-Hamilton Depression Scale, BDI-II,
-Ethnicity country of birth,
-Not in PsyCourse databank yet, for covariates:
length of study attendance (How many visits finished?), lengths of treatment with antipsychotics,
BMI before medication start or BMI with 18 years, relatedness correction (if related individuals were not excluded in QC by Pi-Hat).
Genotypes of interest
We are asking for the total PsyCourse data-set ("best guess data") from quality controlled, imputed PsyCourse 3.0 data (n= 1223 cases: Bipolar Disorder, no controls and exclusion of related individuals).
Phenotypes of interest
Our primary phenotype of interest will be BMI at baseline and BMI change across visits. As the secondary phenotype we will investigate obesity status yes/no (BMI >=30). Lastly, we will investigate trajectories of BMI change across visits.
We will include demographic and clinical confounders that may affect BMI and changes of BMI either based on literature or found by exploratory data analysis (correlation, visualization) or forward-stepwise regression for covariate selection.
Extraction of MC4R interacting proteins
The Integrative Interaction Database (IID) is one of the largest collections of experimentally validated as well as predicted protein protein interactions (Kotlyar et al., 2019). It can be used to select gene products that have known interactions with the MC4R protein. We found 24 experimentally validated interaction partners of MC4R (NPY, GRK2, NPM1, POMC, PRKACA, MGRN1, NPC1, MAL, AGRP, TSPAN3, AIG1, PDIA6, ATP6V0E2, MRAP2, MC1R, TMEM94, ATRNL1, PLLP, ASIP, MRAP, YIPF3, CD81, TMEM19 and IL1RAP). However, we aim to extend this list by including high confidence predictions and orthologs.
For the preliminary phenotype, we run a linear regression to investigate associations between MC4R SNPs and primarily phenotype corrected for clinical and demographic confounders. For the secondary phenotype we will fit linear regression models corrected for covariates found before. For the trajectory, we will fit linear mixed models for repeated measurements corrected for length of the study-attendance, medication, and other significant confounders. If SNPs will be found nominally significant, we will investigate which genotype is driving the effect using ANOVA with Tukey's HSD correction or t-test and visualize results as boxplots.
Correction for multiple testing
As SNPs in each gene (e.g. MC4R) will be in high LD, we will use Nyhol's correction than incorporates genetic correlation instead of Bonferroni correction where possible. In all cases, analyses will be conducted in R using different packages like 'nlme' for linear mixed modes.
QC details (only if still necessary and not yet performed at the LMU Munich):
-Standard QC performed by Dr. Manfred Sagmeister in Graz (only if still necessary and only if we get just the raw data, iDat Files):
Quality Control (suggestion from Manfred Sagmeister- he will publish his pipeline before sharing code!):
a. Data QC (quality control), filtering and imputation (Main tools: PLINK 1.9 & 2.0, BCFtools, GEMMA, SNPflip)
b. Pre-QC on single cohorts
- Raw data check: gender, call rates, batch effects
- Marker strand orientation check and correction
- Check for heterozygosity and duplicates
- Remove markers and samples that failed initial QC
- Filtering: call rate, minor allele frequencies (maf) and Hardy Weinberg Equilibrium (HWE)
c. Principal component analysis (PCA) on single cohorts
- Prune cohorts and check relatedness with identity by decent (IBD)
- Create unrelated datasets and calculate principal components with EIGENSOFT smartPCA
- Removal of ethnic outliers
- Repeat filtering: call rate, maf, HWE
d. Merging of cases with controls
- Stepwise gross filtering, flagging of ambiguous markers
- Merged dataset with all subjects will be used for imputation
- Repeat PCA steps on merged dataset
e. Phasing and imputation of merged dataset with Beagle 5.0
- Reference panel: Haplotype reference consortium (HRC) subset with about 21.000 European subjects
f. Post imputation QC
- Filter low quality imputed markers with different quality thresholds (min. 0.9)
- Create dosage files with genotype probabilities
- Check for strand issues, flag ambiguous markers and remove duplicates
- Removal of ethnic outliers according to previous calculated PCA
- Stepwise filtering for call rate, maf and HWE (controls only)
Isolating genotype data with PLINK:
-Isolation of target genotypes (all SNPs from each target gene region) from the AIWG associated target gene (GLP-1, Orexin A, Orexin B, OX1R, OX1R, NPY, NDUFS1, TSPO, HCRTR2, GCG, GABRA2, CNR1 and MC4R) from the MC4R network (MC4R, NPY, GRK2, NPM1, POMC, PRKACA, MGRN1, NPC1, MAL, AGRP, TSPAN3, AIG1, PDIA6, ATP6V0E2, MRAP2, MC1R, TMEM94, ATRNL1, PLLP, ASIP, MRAP, YIPF3, CD81, TMEM19 and IL1RAP). Extraction of all SNPs from each gene region (reference gene: human, assembly 37p13) from the imputed PsyCourse data (best guess data) +/- 200 kb.
- Recoding in additive model with PLINK
- Repeating the same extraction step for the other target genotypes (R loops or Python loops).
- Reading resulting file into R Studio
- Merging with PSYCOURSE phenotype data
- Analyses with R studio (see also section association analyses):
I) Association between target genes (e.g. MC4R) and obesity in BD:
a. Cross sectional analysis of the relationship between target genotypes (0, 1, 2) and BMI at baseline (V1): linear regression model in R including the dependent variable BMI, the independent variable MC4R genotype and covariates.
b. logistic regression: logistic regression model with the independent variable: normal weight vs. overweight, dependent variable BMI, correction for confounding factors.
c. BMI changes over time-points (baseline and end of the PSYCOURSE study): linear mixed models for repeated measurements corrected for tudy-attendance length, medication, and other significant confounders.
II.) Association between target genotypes and course of disease
linear regression model in R including dependent variable representing disease course (e.g. numbers of affective episodes or GAF or CGI), the independent variable target genotype (0,1,2) and significant covariates.
III.) Systems Biology analyses
The most interesting SNPs will be selected and used for in silico functional analyses to understand the biological context of our findings. We will use a set of established systems biology approaches, to annotate these genetic features with biomedical ontologies such as Gene Ontology and other contextual information. This allows us to identify molecular mechanisms and pathways that are relevant for the genetic characteristics of obesity in bipolar disorder.
A. Genotype data
We apply for the total genotype data (total sample, best guess data) from the complete quality controlled, imputed PsyCourse 3.0 data (n= 1223 cases). Nevertheless, for diverse tasks (e.g. to exclude related patients or to analyse relatedness for correction) our QC specialist would need a pruned dataset (~20k ; independent, not in LD SNPs from the whole genome). If he had to include related individuals at all, he would exclude PI_HAT >0.185 in Plink IBD check and would do an additional check with GEMMA - both on a pruned whole dataset. But, if related individuals are excluded in the sample, this is not necessary. B. Phenotype data for sample description, exploratory data-analysis, covariates and testing association between obesity and MC4R: Anthropometric measures at baseline (V1) for all patients and for each visit (V2, V3, V4):
Body mass index
Elevated cholesterol or triglyceride levels
Comorbidities (cardiovascular or putatively affecting weight):
Traumatic brain injury
Liver cirrhosis or inflammation
Diagnosis and course of disease:
Diagnosis according to DSM-IV, DSM-IV diagnosis categories, Bipolar-I Disorder Bipolar-II Disorder
Number of manic episodes
Number of MDD episodes
Number of hypomanic episodes
Number of suicide attempts
Clinical Global Impression (CGI) - illness severity scale
Global Assessment of Functioning (GAF)
For putative covariates and sample description:
markers for measuring indirectly loneliness and life events: marital status
"Do you currently have a partner?"
life events precipitating illness episode between study visits
Change in housing situation since last study visit?;
high-school level education
currently paid employment
disability pension due to psychological/psychiatric illness
duration of illness
psychiatric medication change during the past six months
family member ever affected by psychiatric disorder
Inventory of depressive symptomatology (IDS-C30):
items for disturbed sleep:
Item 1 Sleep onset insomnia
Item 2 Mid-nocturnal insomnia
Item 3 Early morning insomnia
Item 4 Hypersomnia
Young Mania Rating Scale
Hamilton Depression Scale
country of birth
Needed, but not in PsyCourse databank yet:
length of study attendance
lengths of treatment with antipsychotics or medication in general
BMI before medication start or BMI with 18 years
relatedness correction (if related individuals were not excluded in QC by Pi-Hat value!)
antidepressants (yes, no)
antipsychotics (yes, no)
antiepileptics (yes, no)
and lithium (yes, no)
Further necessary variables for filtering, which may be necessary: recruitment center, date of interview, case or control status.