The rationale and design of the OmniHeart trial as well as the main blood pressure and lipid results (21
) were published previously. Briefly, the OmniHeart trial is an investigator-initiated study sponsored by the National Heart, Lung, and Blood Institute that used a randomized, 3-period crossover design to compare the effects of macronutrients on blood pressure and plasma lipids. Study diets were modeled on the successful DASH diet and emphasized either carbohydrate (Carb diet), unsaturated fats (Unsat diet), or nonmeat protein (Prot diet). The DASH diet was chosen because it is effective at reducing diastolic and systolic blood pressures. DASH was found to lower LDL cholesterol, but it also lowered HDL cholesterol and increased TGs, as many carbohydrate-rich diets were shown to do. In this study, we examine the effect of replacing some of the carbohydrate with unsaturated fat or protein. A detailed description of these 3 diets was published previously (24
) and is summarized in .
Macronutrient composition of usual and assigned diets1
Trial participants lived in the greater Boston, MA, and greater Baltimore, MD, areas and were adult men and women aged ≥30 y with systolic blood pressure 120−159 mm Hg, diastolic blood pressure < 100 mm Hg, fasting LDL cholesterol < 220 mg/dL, and TGs < 750 mg/dL. Other exclusion criteria were described previously (21
). By design, we aimed to recruit 50% African Americans and 50% women. Eligibility was determined during screening visits, at which time a blood sample was taken for baseline measurement. The blood samples were immediately centrifuged to collect plasma, which was then divided into aliquots and stored at −80 °C pending analysis. Eligible participants then completed a 6-d run-in period during which time they were given the meals that were to be provided during the 3 diet periods. Subjects who failed to adhere to the protocol during the run-in period were excluded.
Each participant was randomly assigned 1 of 6 sequences of the 3 diets (Carb, Unsat, Prot). Each diet consisted of commonly available foods. The initial calorie content was determined for each participant, based on body size, sex, and physical activity level. Body weight was monitored daily, and calorie content of the diets was adjusted to maintain initial body weight. All meals, snacks, and beverages, except for discretionary calorie-free beverages, were provided to the participants. In addition, participants were requested to maintain their usual intake of alcoholic beverages, not to exceed 2 drinks/d. Participants were instructed to eat only the food provided and to maintain their usual levels of physical activity. Adherence was monitored through daily diet diaries and at their weekday visits to the study center. At the end of the fourth and sixth weeks of each diet period, blood was drawn, centrifuged, divided into aliquots, and stored at −80 °C pending laboratory analysis. Participants then ate their usual free-living diet for a minimum of 2 wk before beginning the next diet period. Controlled feeding took place from April 2003 to June 2005.
All 162 participants who successfully completed ≥2 of the 3 diet periods were included in the laboratory analysis. Each participant provided 4 samples for analysis, one at baseline and one after each of the three 6-wk diet periods. Blood samples were collected in tubes containing EDTA. The study was conducted from April 2003 to June 2005. Laboratory analysis began March 2004 and was completed August 2005. Samples were submitted to the laboratory in batches, and the batches were analyzed in the order received. Therefore, the longest possible period of storage was 12 mo, with most samples stored ≤10 mo. The 4 samples were analyzed in the same batch in random order to reduce analytic variation. Analysis batches consisted of 5 or 9 participants, depending on the week, and batches were completed within 5 d. All laboratory staff members were blinded to the diet sequences of the participants.
Samples were removed from cryogenic storage and thawed in the dark at room temperature for 30 min. Samples were filtered, and 700 μL filtered plasma was loaded into 20 mL Econo-Pac columns (Bio-Rad Laboratories, Hercules, CA) packed with anti–apo C-III resin (polyclonal goat anti–human apo C-III antibody bound to Sepharose 4B Resin; Academy Biomedical Company Inc, Houston, TX). Samples and resin were incubated for 16 h at 4 °C with mixing. The unbound fraction was eluted from the column by gravity followed by washes with phosphate-buffered saline. The bound fraction was then eluted from the columns with 3 mol/L sodium thiocyanate in phosphate-buffered saline and was immediately desalted with the use of PD-10 columns (GE Healthcare, Little Chalfont, United Kingdom).
The immunoaffinity columns consisted of 2.5 mL anti–apo C-III resin prepared with the use of polyclonal goat anti–human apo C-III antibody bound to Sepharose 4B Resin at a minimum concentration of 5 mg antibody/mL resin. The highest concentration of plasma apo C-III found in this study was ≈61 mg/dL (0.61 mg/mL). At a load volume of 0.7 mL, this is a maximum load of 0.4 mg, which is below the minimum theoretical capacity of 0.6 mg apo C-III based on column specifications. All columns were tested to ensure efficiency of >95% before the start of laboratory analysis and midway through the analysis period by application of a quality control plasma sample to each column and measurement of apo C-III concentration of both the retained and unretained fractions. In addition, a separate quality control sample was included in each sample batch that was randomly assigned to a different column each week. No column failures were found during this study.
The bound and unbound fractions were ultracentrifuged to separate particles by density. VLDL was isolated by overlaying 700 μL of sample with 300 μL of potassium bromide [with density (d) = 1.006 g/mL] aqueous solution (Sigma-Aldrich, St. Louis, MO) and spinning for 16 h at 15 °C and 25 000 rpm in the outer-most row of a Beckman 25-Ti rotor with a Beckman L8−70 M ultracentrifuge (Beckman Coulter, Inc, Fullerton, CA). The top 200 ± 10 μL from each tube was collected by careful aspiration and stored at 4 °C briefly, pending same-day analysis of lipids and apolipoproteins while the next ultracentrifugation step for LDL was prepared. LDL was isolated by overlaying the plasma remaining after VLDL aspiration with 34% potassium bromide solution to produce a final density of 1.063 g/mL and spinning for 24 h under the same conditions as for VLDL isolation. The top 300 ± 10 μL from each was collected by aspiration. Three density fractions of plasma were thus isolated: <1.006 g/mL (VLDL), 1.006 g/mL to <1.063 g/mL (LDL), and > 1.063 g/mL (very dense LDL, HDL, plasma proteins). The products of the immunoaffinity chromatography followed by density fractionation by ultracentrifugation were VLDL without apo C-III, VLDL with apo C-III, LDL without apo C-III, LDL with apo C-III, d > 1.063 g/mL lipoproteins without apo C-III, and d > 1.063 g/mL lipoproteins with apo C-III.
Determination of lipids and apolipoproteins
Sandwich enzyme-linked immunoabsorbent assay (ELISA) procedures with the use of affinity-purified antibodies (Academy Biomedical Company Inc) were performed to determine the concentrations of apo B, apo C-III, and apolipoprotein E (apo E) in whole plasma and the lipoprotein fractions. TGs and cholesterol were determined enzymatically (Thermo Scientific, Waltham, MA). Liquid transfer for 96-well plate loading and ELISA dilutions were handled robotically with a Multiprobe II (Perkin-Elmer, Waltham, MA) to minimize pipetting error. Both ELISA and lipid plates were read with a BioTek ELx808iu 96-well plate reader controlled by KCJUNIOR software (BioTek, Winooski, VT). All assays were completed in triplicate, and any sample with an intraassay CV > 15% was repeated. Final data were exported to Microsoft EXCEL (Microsoft, Reman, WA) for analysis and database management.
The study protocol was approved by the Institutional Review Boards at all affiliated institutions (Johns Hopkins University, Brigham & Women's Hospital, and the Harvard School of Public Health).
The primary outcome was apo B concentration in whole plasma and in the lipoprotein subfractions. Cholesterol, TG, apo C-III, and apo E concentrations were examined as secondary outcomes. The main comparisons in this study were made among the Carb, Unsat, and Prot diets, in particular the Prot-Carb diets and the Unsat-Carb diets; the Unsat-Prot diet was of secondary interest and was mathematically contained in the other 2 differences. Paired t tests of the between-diet differences were used to assess the differential effects of the diets. Secondarily, we examined the change from baseline elicited by each diet with paired t tests of the difference between baseline and postdiet samples. Analyses were performed with the use of SAS version 9.1 (SAS Institute Inc, Cary, NC). The t tests were all 2-sided, and statistical significance was defined as P ≤ 0.05. We did not adjust P values for multiple comparisons because there is no consensus for adjustment when many of the outcome variables (lipoprotein subtypes) examined are highly interrelated with the others. Our analysis produced patterns of significance that corroborate one another, which reduces the risk of reporting significance of false positives because of the sheer number of tests performed. Furthermore, the “proc mixed” procedure, which performs a global test reducing the risk of type I error, was performed to confirm the findings of the paired t tests. For a study such as this with many subsamples per participant, there are randomly occurring missing values, stemming from limitations in laboratory technology, eg, detection limits, lack of sufficient sample, or out-of-range value. Imputation of these missing data points must be handled differently for each different reason for the missingness. Choosing a single imputation strategy for all missing data introduces noise. Therefore, for this analysis no missing data were imputed. Instead, the t tests for each outcome are performed with the use of subsets of the data that exclude any subjects' missing data for that particular outcome. Although this reduces power compared with an analysis of all participants with imputation, it avoids regression to the null because of background noise.