072_ The influence of genetic liability for circadian rhythms alterations on bipolar disorder outcomes
Research Question and Aims
Patients with bipolar disorder (BD) display persistent disturbances of circadian rhythms (CR). Studies suggested that changes in sleep duration in euthymic BD were associated with relapse of mood episodes, increased risk of suicidal behavior, poor functioning, and worse cognitive outcomes[2,3,4]. Among biological markers of CR, modern GWAS have been used to derive PRSs for several sleep-related traits such as chronotype (eveningness–morningness)5, sleep duration6, insomnia7, or relative amplitude8 - a measure of CR disruptions, but their role in predicting BD outcomes has been overlooked.
We aim to: Examine the association between PRSs for circadian disruptions and cognitive, functioning, and clinical outcomes (including sleep patterns) in a sample of individuals with BD or Schizophrenia and healthy controls. Furthermore, we aim to integrate PRSs for these traits, and clinical variables into a machine-learning algorithm able to predict BD phenotypes (BD with poor cognitive and global functioning and more severe illness vs BD with good cognitive and global functioning and less severe illness) or BD diagnosis.
1. PMID: 34850507
2. PMID: 29776774
3. PMID: 30290235
4. PMID: 29286594
5. PMID: 30696823
6. PMID: 30531941
7. PMID: 30804565
8. PMID: 30120083
Individual genetic liability for specific circadian rhythms traits is different in BD compared with HC or other psychiatric diagnoses. The integration between genetic and clinical factors would be able to discriminate different phenotypes of BD.
1. Subjects with bipolar disorder (BD) will have higher polygenic risk scores (PRS) for circadian rhythms alterations compared to psychoses spectrum disorders and to HC
2. Subjects with bipolar disorder (BD) with higher PRS for circadian rhythms alterations will have worse cognitive outcomes, functioning, and clinical outcomes (i.e. suicide, more illness severity) compared with patients with lower PRSs for these traits.
3. Integrating genetic (PRS) with socio-demographical and clinical factors will significantly increase the accuracy of prediction of BD (in the total population) and BD outcomes (among individuals with BD).
For this purpose, we would use cross-sectional data derived from two different cohorts:
1. The Bipogent Cohort: composed of outpatients of the Bipolar and Depressive Disorders Unit of the Hospital Clinic of Barcelona, diagnosed with BDI or BDII, recruited from February 2017 to July 2019. In the cohort, we have a total of 531 individuals phenotyped and genotyped (251 patients and 280 matched healthy controls).
2. The PsyCourse Cohort: 1200 participants of the PsyCourse study with schizophrenia- or bipolar-spectrum diagnoses (we will use only data of genotyped patients or healthy controls)
Baseline evaluations from the Bipogent and PsyCouse cohorts will be considered:
- Psychiatric and medical history
- Medication data
- Family history of psychiatric disease
- Clinical evaluations: mood assessment, functioning, illness severity, suicide (attempt, behavior)
- Neurocognitive assessment: Processing Speed (TMT-A total time, key number of WAIS-III and DST); attention (digit forward); attention and working memory (digit backward of WAIS-III digit span); executive functions (TMT-B); verbal memory (CVLT/VLMT), IQ (WAIS-III, MWT-B).
Polygenic Risk Scores:
Polygenic risk scores (PRS) for major psychiatric disorders for sleep duration, Morningness-Eveningness chronotypes, and relative sleep amplitude will be constructed by using the results of meta-analysis of genome-wide association studies (GWAS) on the topic. PRS-CS tool will be used to infer posterior SNP effect sizes under continuous shrinkage priors and estimate the global shrinkage parameter (φ) using a fully Bayesian approach. PRS will be then calculated in PLINK 1.9 using dosage data independently in the two cohorts, and data will be meta-analyzed.
Partial correlation analyses adjusted by age and gender will be performed to assess the correlation between PRSs and cognitive outcomes, functioning, or other clinical outcomes, including lifetime number of suicide attempts, rapid cycling, predominant polarity, illness severity, substance use, seasonality, and sleep patterns (derived from the sub-items of the HDRS and IDS). Multiple linear regressions will be performed to study the association between the measured PRS and the described variables (cognitive and clinical). Separate multiple regression analyses were performed for each PRS, considered the main independent variable while controlling for sex, age, ancestry, and diagnostic category. For the development of machine learning algorithms, super vector machines will be used to predict diagnosis and BD outcomes. In order to compare the accuracy of linear and nonlinear modeling with the objective of identifying the presence of SNP x SNP interactions, the full datasets will be used to build the SVM models. The data will be randomly split into train/test subsets using 75%/25% proportions, then the SVM models will be built and tested on a large number (100 times) of such splits for each of the different kernels to obtain distributions of accuracy scores for each model. The metric used for all of the performance assessments was the Area Under the receiver operating characteristic (ROC) curve (AUC) metric, also known (ROC) score. Analyses will be conducted with Python using two main packages (scikit-learn and pandas) and R (for univariate or bivariate analyses).
Diagnosis (personal and familiar):
Genetic data (imputed).