Enhanced detection of genetic association of hypertensive heart disease by analysis of latent phenotypes

Abstract

Hypertension and hypertensive heart disease (HHD) are inter-related phenotypes frequently observed with other comorbidities such as diabetes, obesity, and dyslipidemia, which probably reflect the complex gene-gene and/or gene-environment interactions resulting in HHD. The complexity of HHD led us to examine intermediate phenotypes (e.g., echocardiographically-derived measures) for simpler clues to the genetic underpinnings of the disease. We applied the method of independent component analysis to a prospective study of the metabolic predictors of left ventricular hypertrophy and extracted latent traits of HHD from panels of multi-dimensional anthropomorphic, hemodynamic echocardiographic and metabolic data. Based on the latent trait values, classification of subjects into different risk groups for HHD captured meaningful subtypes of the disease as reflected in the distributions of primary clinical indicators. Furthermore, we detected genetic associations of the latent HHD traits with single nucleotide polymorphisms in three candidate genes in the peroxisome proliferator-activated receptors complex, for which no significant association was found with the original clinical indicators of HHD. Consensus analysis of the results from repeated independent component analysis runs showed satisfactory robustness and estimated about 3-4 separate unseen sources for the observed HHD-related outcomes.

The St. Louis African American health-heart study: methodology for the study of cardiovascular disease and depression in young-old African Americans

“…Third, by focusing the analyses on continuous cardiovascular phenotypes, including both raw (e.g., coronary artery calcium volume) and extracted latent traits, the power to detect significant associations will be greater than comparable analyses of dichotomized CAD traits (e.g., history of myocardial infarction). Fourth, the proposed genetic investigations are designed to take advantage of well-established genetic analytic tools supplemented by novel analytic methods developed and validated by the investigative team [103]. Fifth, multiple biomarkers and other clinical data permit the examination of physiological pathways by which depression affects CAD outcomes.…”

“…Version-controlled analytic datasets will be made and distributed to all approved study investigators for downstream analyses. Latent factor analysis will be performed on panels of observed cardiovascular findings to extract underlying patterns of data (i.e., “ traits ”) more proximally associated with CAD [103]. The latent factors will first be extracted from panels of variables representing cardiovascular attributes, and then again from panels combining both cardiovascular and depressive characteristics.…”

Association and interaction of PPAR-complex gene variants with latent traits of left ventricular diastolic function

“…The present study was performed using state-of-the-art analysis techniques in genetic epidemiology and represents a continuation of our study of the genetics of hypertensive heart disease through analysis of endophenotypes, which lay proximal along the pathway from observed clinical phenotype to genotype[8]. The latent factor analysis by ICA was used to address the multidimensional data in which non-Gaussian structure such as clustering and independence represent important components.…”

“…We used a freely available implementation called FastICA (version 1.9) available in R (v. 2.7.0) to analyze the residuals of a matrix of the echocardiography-derived endophenotypes by pre-specified numbers of latent components[8,18,19]. For each component, the output from ICA consists of the extracted independent component (IC) represented by column vectors of the matrix of loadings on the echocardiography-derived endophenotypes and the coefficients of the extracted IC for each subject.…”

“…We have previously shown that independent component analysis (ICA)-derived latent traits of HHD can be used to enhance the detection of genetic association[8]. Herein we apply ICA to derive a single latent LVDF trait from multiple echocardiography-derived indices (or "endophenotypes") of LVDF.…”