|
nif:isString
|
-
The research protocol was approved by the University of Florida Institutional Review Board. Informed consent was obtained from all participants prior to data collection.
Data were collected in Guayama, Puerto Rico, a town of approximately 44,000 people on the southeastern coast of the island [69]. Sampling procedures have been described previously [20], [70]. Briefly, the survey was conducted in four residential areas selected ethnographically to represent contrasts in two key independent variables: socioeconomic status and color. Within each area, we drew a simple random sample of 25 households and then randomly selected one adult, aged 25–55 years, from each household (N = 100). Ninety-six participants donated buccal swabs and gave informed consent for DNA analyses. After removing samples and markers with >25% missing data, our dataset included 87 individuals. The combination of probability and non-probability sampling methods limits the generalizability of our results but improves the efficiency of our attempt to isolate contrasts related to socioeconomic status and color.
Samples were genotyped for 100 biallelelic autosomal SNP markers at Prevention Genetics (Marshfield, Wisconsin). These ancestry informative markers (AIMs) were selected for large frequency differences between West African, European, and Native American populations, derived from a GeneMapping 10K Affymetrix array. Eighty-seven individuals were typed for 78 AIMs. Eighty-four individuals were also assayed via pyrosequencing at the University of Florida Center for Pharmacogenomics for six hypertension candidate genes within three different genes of the adrenergic receptor family: Ser49Gly and Arg389Gly in β1AR, Gly16Arg, Gln27Glu, and Arg523Arg in β2AR, and Del322-325 in α2CAR.
Genetic Ancestry and Social Classification: Three independent methods were used to estimate individual genetic ancestry: two Bayesian approaches using a Markov chain Monte Carlo algorithm, implemented in Structure 2.2 [71] and ADMIXMAP [72], and a maximum likelihood method [73] implemented in software provided by Xianyun Mao. Genotypes and allele frequencies from unadmixed populations of West Africans, Europeans, and Native Americans (needed for Structure and ADMIXMAP/MLE, respectively) were provided by Mark Shriver. Ancestral proportions for K = 3 were selected for all three programs because of the putative ancestral contribution from three distinct populations in Puerto Rico [74]. The three methods did not produce significantly different ancestry estimates and yielded similar results in our analyses; we present maximum likelihood estimates. Our measure of color estimates how people are perceived by other Puerto Ricans in mundane social interaction. In an earlier ethnographic study in the same community where the survey was conducted, Gravlee [75] identified locally relevant categories of color. He asked ethnographic informants to identify the color of 72 standardized facial portraits that varied systematically in skin tone, hair texture, and facial features. He then used cultural consensus analysis [76] to test the assumption that respondents shared a coherent cultural model of color classification and to determine the culturally appropriate categorization of each portrait. In the current study, two observers determined which of the same 72 portraits most closely resembled each respondent, as described elsewhere [20]. We then used the consensus categorization of the matching portrait as an estimate of each respondent's ascribed color. This method produces an ethnographically grounded estimate of how survey respondents would be perceived by others in everyday social interaction.
Blood pressure was measured using an automatic oscillometric blood pressure monitor (Omron HEM-737AC; Omron Healthcare, Inc., Vernon Hills, IL) that has been validated for population-based studies [77], [78]. Three measurements were taken at standardized intervals at the beginning, middle, and end of the hour-long interview. Respondents had been seated for at least 10 minutes and had not ingested caffeine or tobacco for at least 30 minutes before each measurement. We took measurements with the left arm supported at heart level. Mean systolic (SBP) and diastolic blood pressure (DBP) from the three measurements were treated as dependent variables. Standard covariates included sex (0 = female, 1 = male), age (years), socioeconomic status (SES), body mass index (BMI, weight in kg/height in m2), and current use of antihypertensive medication (0 = no, 1 = yes). Weight (0.1 kg) was measured with a digital scale; height (0.1 cm) was measured with a portable stadiometer. SES was estimated as a combination of self-reported education (years) and household income (nine categories; total from all sources, before taxes, in the last 12 months). We tested multiple ways of modeling SES [79]. Here we used scores on the first principal component of education and household income (88% common variance explained); other ways of modeling SES did not alter substantive results.
Regression analyses were performed separately for SBP and DBP. Ascribed color was entered into models as two categorical variables using a reverse Helmert coding scheme. The first variable tested for differences between trigueño (literally, “wheat-colored”) and blanco (white); the second for differences between negro (black) and the mean of trigueño and blanco. This coding scheme reflects both the natural ordering of categories and the expectation from ethnography that the stigmatized category negro would differ from the mean of trigueño and blanco. Based on previous research [20], [31], [32], [40], we constructed cross-product terms to test for two-way interactions between (a) SES and genetic ancestry and (b) SES and color. Continuous predictors were mean-centered to reduce multicollinearity and to ease interpretation of interaction terms. We examined variance inflation factors for multicollinearity and case diagnostics for evidence of influential observations. In models including α2CAR genotype, three participants with missing genotype data for this polymorphism were excluded from the analysis. We replicated all other regression models without these three participants and observed no significant differences in any coefficients.
|
![[This page as RDF]](http://data.gesis.org/somesci/static/rdf-icon.gif){kind=icon}