Heterogeneity analysis provides evidence for a genetically homogeneous subtype of bipolar-disorder.

Bipolar Disorder (BD) is a complex disease. It is heterogeneous, both at the phenotypic and genetic level, although the extent and impact of this heterogeneity is not fully understood. One way to assess this heterogeneity is to look for patterns in the subphenotype data, identify a more phenotypically homogeneous set of subjects, and perform a genome-wide association-study (GWAS) and subsequent secondary analyses restricted to this homogeneous subset. Because of the variability in how phenotypic data was collected by the various BD studies over the years, homogenizing the phenotypic data is a challenging task, and so is replication. As members of the Psychiatric Genomics Consortium (PGC), we have access to the raw genotypes of 18,711 BD cases and 29,738 controls. This amount of data makes it possible for us to set aside the intricacies of phenotype and allow the genetic data itself to determine which subjects define a homogeneous genetic subgroup. In this paper, we leverage recent advances in heterogeneity analysis to look for distinct homogeneous genetic BD subgroups (or biclusters) that manifest the broad phenotype we think of as Bipolar Disorder. As our data was generated by 27 studies and genotyped on a variety of platforms (OMEX, Affymetrix, Illumina), we use a biclustering algorithm capable of covariate-correction. Covariate-correction is critical if we wish to distinguish disease-related signals from those which are a byproduct of ancestry, study or genotyping platform. We rely on the raw genotyped data and do not include any data generated through imputation. We first apply this covariate-corrected biclustering algorithm to a cohort of 2524 BD cases and 4106 controls from the Bipolar Disease Research Network (BDRN: OMEX). We find evidence of genetic heterogeneity delineating a statistically significant bicluster comprising a subset of BD cases which exhibits a disease-specific pattern of differential-expression across a subset of SNPs. This pattern replicates across the remaining data-sets collected by the PGC containing 5781/8289 (OMEX), 3581/7591 (Illumina), and 6825/9752(Affymetrix) cases/controls, respectively. This bicluster includes subjects diagnosed with bipolar type-I, as well as subjects diagnosed with bipolar type-II. However, the bicluster is enriched for bipolar type-I over type-II and may represent a collection of correlated genetic risk-factors. By investigating the bicluster-informed polygenic-risk-scoring (PRS), we find that the disease-specific pattern highlighted by the bicluster can be leveraged to eliminate noise from our GWAS analyses and improve not only risk prediction, particularly when using only a relatively small subset (e.g., ~ 1%) of the available SNPs, but also SNP replication. Though our primary focus is only the analysis of disease-related signal, we also identify replicable control-related heterogeneity. Covariate-corrected biclustering of raw genetic data appears to be a promising route for untangling heterogeneity and identifying replicable homogeneous genetic subtypes of complex disease. It may also prove useful in identifying protective effects within the control group. This approach circumvents some of the difficulties presented by subphenotype data collected by meta-analyses or 23 andMe, e.g., missingness, assessment variation, and reliance on self-report.

A time-series analysis of blood-based biomarkers within a 25-year longitudinal dolphin cohort.

Causal interactions and correlations between clinically-relevant biomarkers are important to understand, both for informing potential medical interventions as well as predicting the likely health trajectory of any individual as they age. These interactions and correlations can be hard to establish in humans, due to the difficulties of routine sampling and controlling for individual differences (e.g., diet, socio-economic status, medication). Because bottlenose dolphins are long-lived mammals that exhibit several age-related phenomena similar to humans, we analyzed data from a well controlled 25-year longitudinal cohort of 144 dolphins. The data from this study has been reported on earlier, and consists of 44 clinically relevant biomarkers. This time-series data exhibits three starkly different influences: (A) directed interactions between biomarkers, (B) sources of biological variation that can either correlate or decorrelate different biomarkers, and (C) random observation-noise which combines measurement error and very rapid fluctuations in the dolphin's biomarkers. Importantly, the sources of biological variation (type-B) are large in magnitude, often comparable to the observation errors (type-C) and larger than the effect of the directed interactions (type-A). Attempting to recover the type-A interactions without accounting for the type-B and type-C variation can result in an abundance of false-positives and false-negatives. Using a generalized regression which fits the longitudinal data with a linear model accounting for all three influences, we demonstrate that the dolphins exhibit many significant directed interactions (type-A), as well as strong correlated variation (type-B), between several pairs of biomarkers. Moreover, many of these interactions are associated with advanced age, suggesting that these interactions can be monitored and/or targeted to predict and potentially affect aging.

A loop-counting method for covariate-corrected low-rank biclustering of gene-expression and genome-wide association study data.

A common goal in data-analysis is to sift through a large data-matrix and detect any significant submatrices (i.e., biclusters) that have a low numerical rank. We present a simple algorithm for tackling this biclustering problem. Our algorithm accumulates information about 2-by-2 submatrices (i.e., 'loops') within the data-matrix, and focuses on rows and columns of the data-matrix that participate in an abundance of low-rank loops. We demonstrate, through analysis and numerical-experiments, that this loop-counting method performs well in a variety of scenarios, outperforming simple spectral methods in many situations of interest. Another important feature of our method is that it can easily be modified to account for aspects of experimental design which commonly arise in practice. For example, our algorithm can be modified to correct for controls, categorical- and continuous-covariates, as well as sparsity within the data. We demonstrate these practical features with two examples; the first drawn from gene-expression analysis and the second drawn from a much larger genome-wide-association-study (GWAS).

Changes in brain function during administration of venlafaxine or placebo to normal subjects.

Previous research has demonstrated neurophysiologic effects of antidepressants in depressed subjects. We evaluated neurophysiologic effects of venlafaxine in normal subjects. Healthy adults (n=32) received a 1-week placebo lead-in followed by 4 weeks randomized double-blind treatment with venlafaxine IR 150 mg. (n = 17) or placebo (n = 15). Brain function was examined using quantitative electroencephalographic (QEEG) power and theta cordance. Normal subjects receiving venlafaxine showed a decrease in theta-band cordance in the midline-and-right-frontal (MRF) region at 48 hours and at 1 week after randomization. Decreases in relative power also were seen in the MRF region; there were no significant changes in absolute power. These changes were significantly different from those in subjects receiving placebo. Changes in MRF cordance accurately identified treatment condition at 48 hours in 81.3% of subjects, and relative power from this region identified 60.7% of subjects. In conclusion, cordance may detect the pharmacological effects of antidepressant medication in normal subjects. Future studies should examine other classes of medication, as well as antidepressants with other mechanisms of action, to determine if cordance detects antidepressant medication effects in general in normal subjects.