Plasma proteome correlates of lipid and lipoprotein: biomarkers of metabolic diversity and inflammation in children of rural Nepal[S]

Children in this proteomics study are a subset of a large child follow-up cohort in the southeastern plains district of Sarlahi, Nepal. Briefly, a community-based cluster-randomized controlled trial of antenatal micronutrient supplementation was conducted between 1999 and 2001 (). Among more than 4,000 liveborn infants during the trial, we revisited about 3,500 children when they were 6–8 years of age, from 2006 to 2008, to assess their health and nutritional status (). Field workers collected data on child demographic and anthropometric characteristics, education, dietary and morbidity history in the past week, and household socioeconomic status during home visits. Phlebotomists visited households of children early in the morning and collected venous blood samples in 10 ml sodium heparin vacutainers to prevent blood clots. Collected blood samples were transported on ice to the project's field laboratory in Sarlahi where plasma samples were separated by centrifugation (1,720 g for 15 min). Plasma samples were used for lipid assessments and remaining plasma was aliquoted into cryovials that were stored and shipped in liquid nitrogen tanks to the Center for Human Nutrition laboratory at Johns Hopkins University in Baltimore, MD where they were stored at −80°C until thawed for biochemical assessments and/or proteomics analysis (). Among 2,130 children who met inclusion criteria of having available plasma samples and complete epidemiological data from the original maternal micronutrient supplementation trial and follow-up study, samples from 1,000 children, balanced across the five original maternal intervention groups (n = 200 in each) were randomly selected for micronutrient and inflammation status assessments (). Of these, 50#x0025; of the samples (n = 100 from each maternal intervention group) were randomly selected for proteomics analysis (). The original antenatal micronutrient supplementation trial was registered at ClinicalTrials.gov as NCT00115271. Oral informed consent was obtained from the parents of eligible children during the child follow-up due to high illiteracy in the study population. Ethical approval for the original maternal trial and child follow-up study was obtained from the institutional review board at the Johns Hopkins Bloomberg School of Public Health, Baltimore, MD and the Nepal Health Research Council in Kathmandu, Nepal. All methods were carried out in accordance with the principles of the Declaration of Helsinki.

Assessments of lipid profiles and inflammation markers were previously described in detail (, ). Plasma concentrations of total cholesterol, HDL-C, and triglycerides were measured using a Cholestech LDX analyzer (Alere Inc.) by enzymatic colorimetric methods. This instrument has been certified for clinical application by the US Center for Disease Control's Lipid Standardization Program (Atlanta, GA). The instrument is also robust and thus has been used for point-of-care diagnostic testing of lipid levels in remote areas of rural Nepal. The detectable ranges for total cholesterol, HDL-C, and triglycerides were 100–500, 15–100, and 45–650 mg/dl, respectively. Estimates of plasma LDL-C concentrations were calculated using the Friedewald formula (). Quality control materials provided by the manufacturer were tested weekly and with the arrival of each new lot of cassette supplies for quality assurance. Plasma AGP was chosen as a biomarker of subclinical inflammation over C-reactive protein (CRP) because AGP is more likely to detect chronic inflammation. For example, in Nepal, AGP identified five times more children with systemic low-grade inflammation than CRP using conventional clinical thresholds (30#x0025; of children exhibited AGP >1.0 g/l and 6#x0025; of children showed CRP >5 mg/l) (). AGP was measured using a radial immunodiffusion assay (Kent Laboratories; 0.90 ± 0.1 g/l, CV = 10.0#x0025;).

The processes of immune-depletion of high abundant proteins and MS-based proteomics analysis have been previously described (). Plasma specimens were depleted of six high abundance proteins (albumin, transferrin, immunoglobulin G, immunoglobulin A, anti-trypsin, and haptoglobin) using a Human-6 Multiple Affinity Removal System LC column (Agilent Technologies). Seven depleted samples randomly chosen from the plasma samples of 500 participants and one masterpool sample, which served as an internal standard (), were digested at 37°C overnight with trypsin using a 1:10 enzyme to protein ratio. Samples were randomly labeled with eight different isobaric tags for relative and absolute quantification (iTRAQ) reagents, which have reporter ions to be detected for relative quantification, and were incubated at room temperature for 2 h. All labeled samples were combined and 90 μl of the combined peptide sample was dissolved in 4 ml of strong cation exchange loading buffer [25#x0025; v/v acetonitrile and 10 mK KH₂PO₄ (pH 2.8)]. The sample was fractionated into 24 fractions by strong cation exchange chromatography on an Agilent 1200 capillary HPLC system using a PolySulfoethyl A column. Peptides were loaded on to a reverse-phase nanobore column and eluted using a 2–50#x0025; acetonitrile and 0.1#x0025; formic acid gradient for 110 min at 300 nl/min. Eluting peptides were sprayed into an LTQ Orbitrap Velos mass spectrometer (Thermo Scientific) and interfaced with a NanoAcquity ultra-HPLC (Waters). Precursors and the fragment ions were analyzed at resolutions of 30,000 and 15,000, respectively. Spectra from full MS scans and fragmented MS/MS scans were extracted with and without deconvolution using Thermo Scientific Xtract software and searched against the RefSeq 40 protein database. Peptides were identified from Mascot (Matrix Science v2.3) searches through the Proteome Discoverer software (v1.3; Thermo Scientific) with a confidence threshold of 5#x0025; false discovery rate (FDR). A total of 4,705 nonredundant proteins were detected at least one time among 72 iTRAQ experiments conducted to assess plasma samples of the 500 study children (supplemental Table S1). We included 982 proteins quantified in >10#x0025; of all 500 plasma proteins of children (n > 50) in the proteomics analysis.

Standard cut-offs for low or high plasma lipid levels for children were derived from the National Cholesterol Education Program Expert Panel on Cholesterol Levels in Children (). Estimation of relative abundance of proteins from reporter ion intensities across all obtained spectra have been previously described (). Linear mixed effects (LME) models were employed to estimate linear associations between each log2 transformed-lipid or lipoprotein concentration as dependent variables and relative abundance of individual plasma proteins as a fixed effect, and each iTRAQ experiment as a random effect. Lipid values below the lower limit of detection were excluded from these analyses. We also adjusted for fasting status of children in all statistical models, as more than one-third of the children in the study were not fasted at the time of blood draw (). We report summary statistics from the LME models, including number of child plasma samples in which a protein was detected, percent change (#x0025;) in lipid or lipoprotein concentration per 2-fold (100#x0025;) increase in relative abundance of a protein, P value calculated by using a two-sided test of a null hypothesis that there is no association between an individual protein and lipid or lipoprotein, q as FDR-adjusted P value to correct multiple comparisons (), and r as a correlation coefficient between measured plasma lipid or lipoprotein concentration and its respective best linear unbiased prediction from the LME models (). Proteins were considered significantly correlated with an outcome when passing a FDR threshold of 5#x0025; (q < 0.05). Because of their large number, we explored correlations between proteins associated with HDL-C concentration by calculating pairwise protein:protein correlation coefficients for proteins quantified in relative abundance within each iTRAQ experiment and averaged values across iTRAQ experiments.

Among proteins correlated with HDL-C, LDL-C, and triglyceride concentrations, we estimated and compared mean differences in relative protein abundance in plasma samples between children with and without inflammation (plasma AGP concentration >1.0 g/l or ≤1.0 g/l), adjusting for potential confounders, including child sex, age, ethnicity, fasting status, and stunting and underweight status (based on height-for-age and weight-for-age z-scores less than −2, respectively, derived from the World Health Organization growth reference for children) (). P values were adjusted to control FDR using the Benjamini-Hochberg method (). Gene symbols of protein GenInfo identifier (gi) numbers were derived from the Human Genome Organization (HUGO) gene annotation and used in the tables and figures (). General descriptions of proteins were extracted from the NCBI protein database, UniProt, and in-depth review of literature (, ). Data visualization was performed using the Cytoscape (). The data­sets of lipid profiles and relative abundance of reported proteins in this study are available in (supplemental Table S2. All analyses were performed using the R Environment for Statistical Computing (version 3.1.2; R Foundation for Statistical Computing, Vienna, Austria).

In this cohort of 500 Nepalese children, median (interquartile range) values of plasma concentrations for total cholesterol, HDL-C, and LDL-C (n = 324) were 111 (100, 127), 27 (21, 33), and 71 (62, 83) mg/dl, respectively, and 89 (66, 118) mg/dl for triglycerides. These medians approximate plasma values less than fifth percentiles for total cholesterol and HDL-C, less than tenth percentile for LDL-C, and greater than ninetieth percentile for triglycerides of distributions among child populations between the ages of 6 and 9 in the United States (, ). The prevalence of high total cholesterol (≥200 mg/dl) and high LDL-C (≥130 mg/dl) was 0.4#x0025; and 0.6#x0025;, while the prevalence of high triglycerides (≥100 mg/dl) and low HDL-C (<40 mg/dl) was 40#x0025; and 90#x0025;, respectively. About 30#x0025; of children had inflammation indicated by elevated plasma AGP (>1.0 g/l). Child characteristics including lipid and lipoprotein profiles are compared by inflammation status in Table 1. Children with inflammation had lower plasma HDL-C (25 vs. 28 mg/dl; P = 0.005) and higher plasma triglyceride (97 vs. 87 mg/dl; P = 0.032) concentrations than children without inflammation, while there were no significant differences in plasma concentrations of total cholesterol and LDL-C. Stunting (48#x0025; vs. 36#x0025;; P = 0.015), underweight (57#x0025; vs. 45#x0025;; P = 0.017), and reported episodes of fever (16.8#x0025; vs. 4.6#x0025;; P < 0.001) and diarrhea (7.4#x0025; vs. 1.4#x0025;; P = 0.001) in the past week were more common in children with inflammation than without inflammation. No differences were observed in child demographic and educational characteristics, economics of household, and dietary history between the two groups.

Eleven proteins were positively correlated with LDL-C (Table 2). apoB was most strongly correlated with plasma LDL-C (r = 0.69; q = 8.84 × 10⁻⁴⁷). LDL-C concentrations increased by 121#x0025; per 100#x0025; increase in relative abundance of apoB. The rest of the proteins were mostly positive associates (r = 0.18∼0.58; q = 3.74 × 10⁻² to 4.41 × 10⁻⁷), including kinesin family member 20B, platelet activating factor acetylhydrolase (also known as LDL-C-associated phospholipase 2), cholesteryl ester transfer protein (CETP), proteoglycan 4, and apoM, and four intracellular proteins observed in <100 children. Chondroadherin, which regulates chondrocyte growth, was the only protein negatively associated with LDL-C (r = −0.42; q = 2.77 × 10⁻²).

Three apoC proteins were most highly correlated with triglycerides (apoC-II/C-III/C-IV) (r = 0.51∼0.54; q = 8.24 × 10⁻¹³ to 7.22 × 10⁻¹⁷) (Table 3). Other positive correlates included cathelicidin antimicrobial peptide, proteoglycan 4, retinol-binding protein 4 (RBP4), and apoE. Five negatively associated proteins included extracellular matrix-related proteins, such as anthrax toxin receptor (ANTXR)2, neuropilin 1, and insulin-like growth factor binding protein 1, and lipid transfer proteins such as CETP and phospholipid transfer protein (PLTP).

Thirty-six proteins were positively correlated with HDL-C (Table 4). apoA-I (r = 0.79; q = 1.06 × 10⁻⁹¹) and apoA-II (r = 0.58; q = 4.16 × 10⁻²¹) were two of the three most strongly associated proteins followed by seven other apolipoproteins [apoA-IV/apoC-I/apoC-III/apoD/apoM/apoF/serum amyloid A4 (SAA4)] (r = 0.45∼0.56; q = 4.23 × 10⁻³ to 2.72 × 10⁻¹⁶). Proteins with well-defined roles in HDL metabolism and function included PLTP and enzymes, such as LCAT and paraoxonase (PON)1 and PON3. Interestingly, intracellular proteins, such as interferon-related developmental regulator 2 (IFRD2), eukaryotic translation initiation factor 2D (EIF2D), Kruppel-like factor 17 (KLF17), and TATA element modulatory factor 1 (TMF1), were also strongly correlated with HDL-C. Proteins mediating cell-cell and cell-extracellular matrix interaction, such ANTXR1 and ANTXR2, were also among positive correlates. Associated fat-soluble vitamin transporters included retinol binding protein 4, transthyretin, and afamin. Sixteen proteins (Table 5), moderately negatively correlated with HDL-C (r = −0.33 ∼−0.49; q = 2.05 × 10⁻² to 7.59 × 10⁻⁴), are primarily involved in the acute phase response (i.e., orosomucoid 1 or AGP and leucine-rich α-2-glycoprotein 1) and the complement system (i.e., complement component 9 and complement factor I). Protein-protein correlations as well as molecular network and functional clusters of the plasma HDL-C proteome are visualized in Fig. 1A and B. Prominent correlations among proteins positively associated with HDL-C (Fig. 1A) were found between apoA-I and apoA-II (r = 0.75, colored in orange) and apoA-I with other intracellular proteins, such as IFRD2, EIF2D, KLF17, and TMF1 (r = 0.53∼0.91, colored in blue). Among negative HDL-C correlates (Fig. 1B), proteins involved in the acute phase response proteolytic inhibition were strongly correlated with each other (r = 0.36∼0.90, colored in red). Correlates of HDL-C appear to be involved in a wide range of functions that extend beyond known functions of HDL, such as regulation of gene transcription and translation and extracellular matrix homeostasis.