Multivariate analysis of a missense variant inCREBRFreveals associations with measures of adiposity in people of Polynesian ancestries

ABSTRACT Background The Pacific-specific minor allele of rs373863828, a missense variant in CREB3 Regulatory Factor ( CREBRF ), is associated with several cardiometabolic phenotypes in Polynesian peoples, but the variant’s function remains poorly understood. To broaden our understanding of this variant, we used joint multivariate and network analyses to examine the relationships between rs373863828 and a panel of correlated anthropometric and lipid phenotypes. Methods We tested the association of rs373863828 with a panel of phenotypes (body mass index [BMI], weight, height, HDL cholesterol, triglycerides, and total cholesterol) under a multivariate Bayesian association model in a cohort from Samoa (N = 1 632), a Māori and Pacific Island (Polynesian) cohort from Aotearoa New Zealand (N = 1 419), and the combined cohort (N = 2 976). An expanded set of phenotypes (adding estimated fat and fat-free mass, abdominal circumference, hip circumference, and abdominal-hip ratio) was also tested in the Samoa cohort (N = 1 496). Bayesian networks were learned to further understand the structure of the relationships. Results In the Samoa cohort, significant associations (log 10 Bayes factor ≥5.0) were found between rs373863828 and the overall phenotype panel (7.97), weight (8.35) and BMI (6.39). In the Aotearoa New Zealand cohort, suggestive associations (log 10 Bayes factor between 1.5 and 5) were found between rs373863828 and the overall phenotype panel (3.64), weight (3.30), and BMI (1.79). In the combined cohort, concordant signals with stronger magnitudes were observed. In the expanded phenotype analyses among the Samoa cohort, significant associations were also observed between rs373863828 and fat mass (5.68), abdominal circumference (5.37), and hip circumference (5.15). Bayesian networks provided evidence for a direct association of rs373863828 with weight and indirect associations with height and BMI. Conclusions When correlation structures were considered, multivariate Bayesian analyses provided additional evidence of rs373863828’s pleiotropic effects and highlighted a strong direct effect only on weight.