Phenotypic and Genotypic Predictors of Low Retinal Response to Diets High in Lutein

Abstract

Purpose: The ability to classify individuals as responders to dietary intake of lutein and zeaxanthin (LZ) may be useful in post hoc analyses, or design, of LZ supplementation trials. The purpose of this project is to build a model of phenotypic and genotypic variables that jointly predict retinal response to dietary LZ in the Carotenoids in Age-Related Eye Disease Study (CAREDS).

Methods: Macular pigment optical density (MPOD) was assessed using customized heterochromatic flicker photometry in 2001-2004. Diet (including LZ) was estimated from food frequency and supplement questionnaires. Visits also measured waist, queried for chronic diseases, medication use, sunlight exposure, smoking history and physical activity. WHI visits (1994-1998) collected blood for analysis of serum LZ, triglycerides, and genotyping of 439 variants from carotenoid status candidate genes. Among women with intake of LZ >1.7mg/day and complete predictor data, women with MPOD ranking ≥2 quintiles below their quintile for LZ intake were deemed low responders (n=222) whereas women with MPOD which equaled or exceeded their quintile ranking for LZ intake were responders (n=276). Partial least squares discriminant analysis was used to simultaneously test 439 genotypes and 24 phenotypic variables for classifying response, and to assign a variable importance in projection (VIP) score. Iterations were run including subsets of predictors based on VIP thresholds. The model which minimized the leave-one-out cross-validation (LOO CV) misclassification rate was selected.

Results: The model with the lowest LOO CV misclassification rate (33%) included 26 predictors with VIP scores ≥1.7 (range 1.7-3.1); 23 were genetic variants. Variables most strongly predicting low response included waist circumference (VIP=3.1), rs838879 in SCARB1 (VIP=2.8), and diabetes (VIP=2.3). Other predictors included variants from BCMO1, ABCA1, ELOVL2, and FADS1 and hypertension status; all of which have been previously related to MPOD. Newly identified for classifying retinal response were 10 variants within STARD3, which encodes the lutein-specific retinal binding protein and rs3211883 in CD36 and rs7664889 in SCARB2.

Conclusions: We present a model containing 23 carotenoid candidate genotypes and 3 phenotypes which best predicts retinal response to diets high in LZ in internal validation samples of the CAREDS. This model needs to be further validated in supplementation trials.