Constructing a Plasma Nutriproteome for Population Assessment: Analytical Considerations

OBJECTIVES: Micronutrient status is rarely assessed in low-income settings. Proteomics may offer a proxy by measuring plasma proteins correlated with nutrients on a single platform. However, the proteome is huge, diverse and measured in different ways. We describe an analytic framework and decision process to explore nutrient: protein (N:P) associations for micronutrient status assessment. METHODS: In plasma of 435 1(st) trimester women in rural Bangladesh, we compared relative protein abundance, revealed by a multiplexed slow off-rate modified aptamer assay (SomaLogic, Inc), to biochemical concentrations of vitamins A, D, E, B(9), B(12), Zn, Se, Cu, I, & Fe, 5 carotenoids, cholesterol and AGP. After log(2)-transforming protein abundance per convention, N:P relationships were summarized by simple linear regression. We assessed reliability by Pearson correlation (r(p)) and coefficients of variation (CV) in 20 blind duplicates. To define each plasma nutriproteome [proteins correlating at a false discovery rate < 0.05], in all samples we explored a) normalizing protein abundance to the median of our sample vs not, b) assessing correlations by r(p) vs Spearman rank (r(s)) estimators, and c) log(2)-transforming (log(2)) nutrients vs not. We compared differences in the number of proteins and N:P correlative strength (either more negative/positive by r(p) or r(s)) in each proteome when nutrient concentrations were left untransformed vs log(2)-transformed. RESULTS: In duplicates, log(2)-transformed proteins that were normalized, vs not, to the sample-specific median generated higher median r(p) (0.92 vs. 0.87) and r(s) (0.87 vs. 0.85) and lower CV (4.8% vs. 11.7%). The median (IQR) size of the nutriproteomes was 147 (41–340) proteins by r(p) and 87 (29–639) by r(s). For >50% of proteins in 13 nutriproteomes, r(p) was stronger than r(s) (in either +/− direction), favoring use of r(p). Log(2)-transforming folate (B(9)), Zn & cholesterol increased proteome size by 39, 223 & 875 proteins and strengthened r(p) for >50% of proteins than untransformed nutrients. Other proteomes remained larger when nutrient concentrations remained untransformed. CONCLUSIONS: Comparing plasma protein: nutrient associations via methods of normalization, transformation, and correlation offers a framework to guide plasma nutriproteome definition. FUNDING SOURCES: Johns Hopkins University.