- Research Paper
- Open Access
Genetic associations with micronutrient levels identified in immune and gastrointestinal networks
© The Author(s) 2014
Received: 1 February 2014
Accepted: 12 May 2014
Published: 31 May 2014
The discovery of vitamins and clarification of their role in preventing frank essential nutrient deficiencies occurred in the early 1900s. Much vitamin research has understandably focused on public health and the effects of single nutrients to alleviate acute conditions. The physiological processes for maintaining health, however, are complex systems that depend upon interactions between multiple nutrients, environmental factors, and genetic makeup. To analyze the relationship between these factors and nutritional health, data were obtained from an observational, community-based participatory research program of children and teens (age 6–14) enrolled in a summer day camp in the Delta region of Arkansas. Assessments of erythrocyte S-adenosylmethionine (SAM) and S-adenosylhomocysteine (SAH), plasma homocysteine (Hcy) and 6 organic micronutrients (retinol, 25-hydroxy vitamin D3, pyridoxal, thiamin, riboflavin, and vitamin E), and 1,129 plasma proteins were performed at 3 time points in each of 2 years. Genetic makeup was analyzed with 1 M SNP genotyping arrays, and nutrient status was assessed with 24-h dietary intake questionnaires. A pattern of metabolites (met_PC1) that included the ratio of erythrocyte SAM/SAH, Hcy, and 5 vitamins were identified by principal component analysis. Met_PC1 levels were significantly associated with (1) single-nucleotide polymorphisms, (2) levels of plasma proteins, and (3) multilocus genotypes coding for gastrointestinal and immune functions, as identified in a global network of metabolic/protein–protein interactions. Subsequent mining of data from curated pathway, network, and genome-wide association studies identified genetic and functional relationships that may be explained by gene–nutrient interactions. The systems nutrition strategy described here has thus associated a multivariate metabolite pattern in blood with genes involved in immune and gastrointestinal functions.
Systems thinking and methodologies hold greater promise in understanding the complex phenotypes of chronic disease or response to nutrients in foods than the focus on individual genetic variants or the identification of independent environmental factors (Patel et al. 2010, 2012a, b) that influence biological processes. An increasing number of reports employ systems designs and analysis of high-dimensional data from studies of obesity, cardiovascular, nutrition, diabetes, drug, toxicology, immunology, gut microbiota, medicine, health care, and health disparities (Slikker et al. 2007; Auffray et al. 2009; Gardy et al. 2009; Kalupahana and Moustaid-moussa 2011; Kleemann et al. 2011; Roux 2011; Karlsson et al. 2011; Afacan et al. 2012; Meng et al. 2013).
“…Since the components of such a [sic] system are likely to be interrelated in complex ways, and since the synthesis of the parts of individual genes are presumably dependent on the functioning of other genes, it would appear that there must exist orders of directness of gene control ranging from simple one-to-one relations to relations of great complexity” (Beadle and Tatum 1941). (Emphasis added)
With the exception of several publications that include dietary intake variables as a part of omics-based systems (Morine et al. 2010, 2011, 2012) or genomic analysis (Nettleton et al. 2010), many systems studies have implicitly analyzed biological processes as closed systems since environmental variables were not included in the analysis. Biological processes occur in open systems (Von Bertalanffy 1950), and ex vivo factors, which include nutrients and other naturally occurring chemicals in food, can alter biochemical processes and signaling networks occurring within the organism (Kaput and Rodriguez 2004). Excluding external factors that influence internal biological processes generates an incomplete system at best, likely an inaccurate understanding of the interactions between environment and genetic makeup, and from a practical standpoint, misses an opportunity to identify modifiable factors that influence health.
This report details the design and conduct of a discovery-based pilot study that accounts for (1) the known genetic uniqueness of individual humans (Olson 2012), (2) the intra-individual variability in homeostatic measurements (Williams 1956; Illig et al. 2010; Suhre et al. 2011), and (3) the challenge of characterizing complex phenotypes resulting from small contributions of many genetic and environmental factors (Goldstein 2009). The participants in the Delta Vitamin Obesity intervention study were children and teens (age 6–14) enrolled in a summer day camp that was a component of a community-based participatory research (CBPR) program. CBPR is a form of translational research that engages the participant, members of the community, and scientists in research, education, and health-promoting activities for improving personal and public health (McCabe-Sellers et al. 2008). A detailed description of the intervention and results obtained by aggregating data from individuals for population-level analysis such as metabolite and protein variation in relation to BMI, sex, and age has been reported (Monteiro et al. 2014).
In this report, analysis of the data from the Delta Vitamin Obesity (Monteiro et al. 2014) is extended to further characterize metabolite–metabolite interactions with discovery-based methods that identify systems-wide relationships between metabolites, proteins, nutrient intakes, and genetic makeup. Principal component analysis (PCA) was used to analyze plasma homocysteine (Hcy); vitamins A, D, E; riboflavin; thiamine; pyridoxal; and erythrocyte S-adenosylmethionine and S-adenosylhomocysteine (SAH) metabolites. A quantitative variable (met_PC1) from the PCA was defined and used for discovering metabolite–protein correlations as well as thousands of genotypes associated with met_PC1 values. Two recent studies have used inferences based on heritability and Bayesian approaches to identify thousands of SNPs associated with height and weight (Hemani et al. 2013) and rheumatoid arthritis (Stahl et al. 2012) to demonstrate that complex phenotypes are the result of thousands of SNPs. Subsequent data mining methods associated genes and proteins identified in this report to biological functional classes including predominantly immune and gastrointestinal function. Finally, the challenges of conducting case–control studies in light of genetic and cultural differences within and between populations are discussed.
Materials and methods
Participants and CBPR methods
A description of the summer day camp in the Marvell, AR (USA) school district, 24-h dietary intakes, body weight and height, blood sampling and processing, and proteomic and genomic analysis are provided in (Monteiro et al. 2014). In brief, assessments were conducted before the beginning of the camp (baseline), at the end of 5 weeks of the camp (end of camp), and 1 month after camp ended (post-camp). Metabolite and dietary intake data were averaged across the three assessments for the analysis in this study. Thirty-six participants were recruited in year 1, and 19 completed all three assessments. In the second year, 72 participants enrolled and 42 completed three assessments. None of the children or adolescents (age 6–14) was taking prescribed medicines, nor did they have overt malnutrition, active infection, or known genetic disease that could alter metabolism. All participants were healthy African American children and adolescents. Results for the three assessments are reported. The biomedical research protocol was approved by the FDA’s Research Involving Human Subjects Committee (RIHSC) and the University of Arkansas for Medical Sciences (UAMS) Institutional Review Board (IRB).
Total Hcy was analyzed in plasma using a Hcy HPLC Kit (ALPCO Immunoassays, Salem, NH) and a UPLC Waters Acquity HSS T3 column (2.1 × 50 mm, 1.8 µm) coupled with an Acquity HSS T3 1.8 µm VanGuard pre-column at 40 °C.
Vitamins were determined using LC/MS/MS (NCTR-FDA-USA): 250 µL of plasma, in a 1.5-mL Eppendorf microcentrifuge tube, was spiked with stable isotope-labeled standards and mixed with 740 µL of MeOH. Samples were held at 4 °C for 30 min. About 500 µL of hexane was added, and samples were centrifuged at 13,000×g for 12 min (4 °C). The (top) hexane layer was transferred into a total recovery autosampler vial, and the sample was subsequently extracted with two additional 500 µL hexane portions, each time transferring the hexane layer into the autosampler vial. The combined hexane extracts were placed under a stream of nitrogen gas, dried, and reconstituted in 50 µL of 50:50 MeOH/ACN. Ten microlitre sample was injected on an Acquity UPLC equipped with a 2.1 mm × 50 mm (1.7 µm particle) BEH C18 column held at 35 °C. The mobile phase A was 90:10 water/ACN, and the mobile phase B was 50:50 MeOH/can with a flow rate of 0.5 mL/min. Metabolites were analyzed on a Xevo TQ operated in positive APCI ionization mode using the following parameters: source temperature was 145 °C, corona was 15 uA, probe temperature was 575 °C, and desolvation gas flow rate was 600 L/h. Multiple reaction monitoring (MRM) was optimized by direct infusion of standards. The transitions monitored for vitamin A were m/z 269 → 109 (cone E = 35 V, collision E = 15 V) and m/z 269 → 93 (cone E = 26 V, collision E = 14 V). The transition monitored for vitamin E was m/z 431 → 165 and for (d3) vitamin E was m/z 434 → 165 (cone E = 35 V, collision E = 15 V). The transition monitored for 25-hydroxy vitamin D3 was m/z 401 → 159 and m/z 407 → 159 for the (d6) 25 hydroxy vitamin D3 with cone and collision energies of 24 and 28 V, respectively.
About 250 µL of plasma was mixed with 1 mL of (4 °C) acetonitrile in a 1.5-mL Eppendorf microcentrifuge tube. The sample was vortexed briefly and then centrifuged at 13,000×g for 10 min at 4 °C. The supernatant was transferred to a total recovery autosampler vial, and the solvent was evaporated. Samples were reconstituted in 250 µL of water (Optima grade), and 10 µL of sample was injected onto an Acquity UPLC equipped with an HHS T3 2.1 × 100 mm, (1.8 µm particle) UPLC column. Mass spectrometric detection was performed on a Xevo TQ (Waters) operated in ESI positive mode using the following parameters: source temperature was 150 °C, capillary voltage (kV) was 2.2, desolvation temperature was 400 °C, and desolvation gas flow rate was 800 L/h. MRMs for target analytes were optimized by direct infusion of standards. The transition monitored for pyridoxal was m/z 168 → 94 (cone E = 16 V, collision E = 22 V), for pyridoxine m/z 170 → 134 (cone E = 22 V, collision E = 22 V), for thiamine m/z 265 → 122 (cone E = 20 V, collision E = 12 V), for riboflavin m/z 377 → 243 (cone E = 40 V, collision E = 22 V), and for folic acid m/z 442 → 295 (cone E = 22 V, collision E = 12 V).
Red blood cell S-adenosyl-l-methionine (SAM) and S-adenosyl-l-homocysteine (SAH)
Red blood cell samples stored at −70 °C were randomly assayed in batches of 20. About 600 µL of red blood cells was added to tubes containing 150 µL of ice-cold trichloroacetic acid (40 % w/v), plus 330 µL 0.1 M sodium acetate trihydrate, and then vortexed. Samples were incubated at 4 °C for 30 min, followed by centrifugation at 15,000 rpm for 15 min. About 150 µL of supernatant was filtered using a 0.22-µm filter, spun at 5,000 rpm for 5 min, and transferred to vials for chromatographic analysis of SAM. The remainder of the supernatant was transferred to a clean tube for ether extraction. Samples were extracted twice with 300 µL, and any remaining ether was evaporated under argon before filtration and transferred to UPLC vials for the analysis of SAH. Standards for SAM and SAH were obtained from Sigma (St. Louis, MO). Chromatographic separation was achieved on an Acquity HSS T3 column (2.1 × 50 mm, 1.8 µm) coupled with an Acquity HSS T3 1.8 µm VanGuard pre-column at 40 °C. The peaks were separated isocratically with an elution time of 5.0 min for SAM and 2.0 min for SAH at 97 % A (buffer) and 3 % B (methanol). The buffer composition for SAM was 50 mM potassium phosphate and 10 mM heptane sulfonic salt adjusted to pH 4.38 with phosphoric acid. The composition for the SAH mobile phase was 50 mM potassium phosphate. Column equilibration time required for SAM was 90 min, while equilibration time for SAH was just 30 min at flow rates of 0.575 mL/min. Buffers and solvents are filtered using 0.22-µm filters prior to use. Samples were held at 4 °C for the duration of the analysis. The injection volume for samples and standard was 10 µL. Detection was performed with a photodiode array detector set to monitor wavelengths 210–400. Standard was prepared in a range from 0.78 to 25.00 pmol/µL for SAH and from 0.32 to 10.40 µL for SAM. A standard curve was generated to allow for automated calculation of results using the Waters Empower software.
The plasma proteome was quantified for 110 samples from 6 different time points (3 in year 1 and 3 in year 2) but data from 61 at time point 1 were used in these analyses due to missing samples. SomaLogic Inc. (Boulder, CO) performed all proteomic assessments and was blinded to the clinical characteristics of participants in this study. Samples were analyzed as previously described (Gold 1995; Brody and Gold 2000; Gold et al. 2010; Ostroff et al. 2010; Brody et al. 2012).
About 1 mL of whole blood sample from each participant was used for DNA extraction. The genomic DNA samples were extracted and purified using the QIAamp DNA Blood Mini Kit (QIAGEN, Valencia, CA), following the protocol provided by the manufacturer. The quality and quantity of each DNA samples were measured using a NanoDrop 8000 (Thermo Scientific, Wilmington, DE). The Infinium Whole Genome Genotyping technology with the HumanOmni1-Quad version 1.0 kits (Illumina, San Diego, CA) was used for genotyping analyses following the manufacturer’s protocol. The arrays were scanned on a high-resolution iScan (Illumina) and processed using the BeadStudio software version 3.1 (Illumina). The overall genotyping call rate on all samples was above 98 %. Data from 45 unique participants (15 participants attended both years) met these criteria.
Preprocessing of genotyping data
Raw SNP data were first preprocessed, removing SNPs with a GC score <0.7, and those that were not genotyped in all participants. SNPs with minor allele frequency <0.1 and those significantly diverging from Hardy–Weinberg equilibrium were also removed. The remaining SNPs were filtered to include only those present in the metabolic/protein–protein interaction (PPI) network used in the analysis, resulting in a final dataset of 125,959 SNPs.
A metabolic/PPI network was constructed based on the human interaction networks manually curated databases (Ma et al. 2007; Yu et al. 2012). The largest connected component of this network comprised 116,210 interactions between 13,705 genes, containing 125,959 SNPs present on the Illumina 1 M Quad Array. The network was partitioned into topological modules using the spinglass.community function in the igraph library in R (Csardi and Nepusz 2006) resulting in 58 topological modules (mean module size: 236 nodes; SD: 564 nodes).
SNP-, gene-, and network-level analyses
Significant correlations between genotype and met_PC1 levels were assessed in each SNP using generalized estimating equations (GEE), as implemented in the geepack library in R (Højsgaard et al. 2006). Met_PC1 was modeled as a function of genotype at each SNP locus, controlling for age, gender, average Healthy Eating Index, and sibling relationships among the participants (the latter being included as an independence correlation structure in the GEE models). Although some participants attended both years of the camp, only one genotype per participant was used in this analysis. Resulting p values were corrected for multiple testing using the procedure proposed by Benjamini and Hochberg (1995). Nominal p values were used as input for the VEGAS algorithm, which accounts for size, level of polymorphism, and linkage disequilibrium relationships within genes to determine genewise p values from SNP-level results (Liu et al. 2010). Genes reaching significance (q < 0.1) were used in hypergeometric tests (implemented using the HTSanalyzeR library in R) to determine significant enrichment of each of the 58 modules in the interaction network. Modules with q value <0.1 were considered as significantly enriched in genes related to micronutrients. In order to assess the biological processes that may be directly or indirectly implicated by genetic variation in our met_PC1 genes, the functional profile of each significant module was determined using the ClueGO (Bindea et al. 2009) plugin for Cytoscape. ClueGO functional profiles illustrated in Fig. 6 and Supplementary files include KEGG pathways that are significantly overrepresented among module nodes, using hypergeometric tests and correcting p values using the Benjamini and Hochberg method (see Bindea et al. 2009) for technical details on the generation of functional profile networks.
Significant genes were also analyzed in the context of the ArrayTrack QTL database (Harris et al. 2009; Xu et al. 2010) to determine significantly overrepresented QTL phenotypes. Gene sets were constructed by combining all genes within 1 Mbp of QTL mapping to each of the 36 phenotypes (containing at least one significant gene from our analysis) in the ArrayTrack database. Hypergeometric tests were then performed to identify which QTL phenotypes were significantly enriched in the significant genes from our analysis.
PCA of metabolite levels
Given the strong patterns of correlation among the plasma metabolites, PCA was used to identify latent metabolite variables. The first principal component (met_PC1) explained 41 % (Fig. 1b, c) of the variation in metabolite profile and stratified the participants primarily based on their levels of vitamin A, Hcy, SAM/SAH, thiamine, pyridoxal, and vitamin E. Vitamin D and riboflavin contributed to the second principal component and explained 5 % of the variation (Fig. 1b, c) in the dataset. To our knowledge, these nutrient–nutrient associations have not been previously reported and would not have been identified by standard single-variant analysis. Although met_PC1 is a continuous variable, the analysis and heat map indicate metabolic patterns that could be used to group individuals for different nutritional interventions.
Proteomic associations with metabolite patterns
Correlation analysis and hierarchical clustering produced two main branches differing in the percentage of plasma-soluble and membrane proteins versus cytosolic proteins (Fig. 2; see brackets at bottom). We previously observed two clusters of blood versus cytosolic proteins associated with erythrocyte SAM/SAH ratios (Monteiro et al. 2014). The cytosolic proteins in the blood were likely produced by apoptotic processes, although the current data cannot discriminate between normal and pathological cell death.
Analysis of genotype–metabolite correlations within a global protein interaction network
Nominal p values were used as input for the VEGAS algorithm, which accounts for size, level of polymorphism, and linkage disequilibrium relationships within genes to determine genewise p values from SNP-level results (Liu et al. 2010). The result was 1,875 statistically significant genes associated with the met_PC1 variable, which were unevenly distributed among 46 of the 58 modules (Supplement 3).
Network modules significantly enriched in met_PC1 genes
Adjusted p value
1.33 × 10−10
3.86 × 10−9
2.23 × 10−3
3.23 × 10−2
6.23 × 10−3
6.02 × 10−2
1.22 × 10−2
8.86 × 10−2
Network modules significantly enriched in neighborhood micronutrient genes
Adjusted p value
8.23 × 10−35
2.39 × 10−33
9.07 × 10−21
1.32 × 10−19
1.74 × 10−22
1.68 × 10−19
Functional and genetic analyses of statistically significant genes and modules
In order to assess the biological processes that may be directly or indirectly implicated by genetic variation in our met_PC1 genes, the functional profile of each significant module was determined using data mining tools including the ClueGO plugin in Cytoscape (Bindea et al. 2009), the KEGG pathway database (http://www.genome.jp/kegg/pathway), ArrayTrack QTL (Xu et al. 2010) database, and literature mining. All pathways described in the ClueGO analysis results were significantly overrepresented in the given module (adjusted p value <0.05).
Module 18: Functional annotation
Module 2: Functional annotation
Module 2 is functionally enriched in immune function pathways and processes influenced by or involved in infectious diseases (Supplement 4, Barograph). Over 70 % of the genes involved in complement/coagulation pathways are found in this pathway. In addition, disease pathways affected by inflammation such as Alzheimer’s, type 1 diabetes, and rheumatoid arthritis are also represented in this module. Proteomic analysis of blood proteins demonstrated the association between a combination of metabolites including micronutrients (met_PC1) and inflammatory processes (Fig. 2).
Module 52: Functional annotation
Module 52 is the largest in the network with over 2300 genes of which 422 had SNPs statistically associated with met_PC1. Genes and pathways involved in immune functions are enriched in module 52 (Supplement 4, BarGraph52) with cytokine signaling and other immune pathways overlapping with Module 2. The secretory and absorption pathways in Module 18 also have components in Module 52. About 75 % of the phosphatidylinositol and inositol phosphate pathways involved in proliferation, survival, migration, and differentiation in different cell types including the development and regulation of B-lymphocyte and T-lymphocyte functions (So and Fruman 2012) are found in Module 52. Functional analysis also highlights the known links between diabetes and immune function, since type 1 and type 2 diabetes genes and pathways and ~50 % of the insulin signaling pathway occur in Module 52.
Quantitative trait loci mapping and cofactor analyses
Lists of statistically significant genes alone, in modules, or mapped to QTLs identify potential candidate genes for a given phenotype. However, biological processes are necessarily controlled by gene–environment interactions. To associate the genes identified by data mining methods with nutrients, GeneCards and EBI’s cofactor database were searched for each of the genes mapping to QTLs for plasma levels of leptin, adiponectin, glucose, and for type 2 diabetes mellitus (T2DM) loci. Many of the statistically significant met_PC1 genes that mapped to these loci had a metal cofactor, and only a few required organic cofactors (not shown). For example, CD320, the transcobalamin receptor, mapped to the GLUCO3_H QTL (glucose level) on chromosome 19. LRP2, which is involved in vitamin uptake, mapped to a chromosomal region (GLUCO15_H on chromosome 2) associated with hyperglycemia. Several met_PC1 genes (CHKA which is involved in choline metabolism; NOX4, TM7SF2, ALDH3B1, NDUFS8 are associated with NADPH) mapped to serum adiponectin level QTLs. Two genes (SHMT1, cofactors pyridoxal phosphate and folate; ALDH3A2–NADPH) mapped to serum leptin QTLs on chromosome 17. DNMT3B mapped to a serum cholesterol QTL and to a T2DM susceptibility locus on chromosome 20.
Met_PC1 correlated proteins mapping to quantitative trait loci (definitions)
Health and disease processes result from a complex interaction between multiple genes and environmental factors. The systems nutrition analyses reported here used data from dietary intakes, plasma and erythrocyte metabolite levels, plasma proteins, and genetic makeup in a cohort of children/teens aged 6–14. Discussed below are (1) the main biological results, (2) strategy and methodological considerations, and finally, (3) implications for health and disease research.
Met_PC1 and SAM/SAH
Principal component analysis (PCA) identified a metabolite pattern, met_PC1, with positive and negative correlations between plasma micronutrients, plasma Hcy, and SAM/SAH ratio in erythrocytes. Plasma vitamin A and Hcy correlated positively with SAM/SAH, and vitamin E, thiamine, and pyridoxal correlated negatively in this population. While statistical associations do not prove causality, these correlations suggest that micronutrients and metabolites operate within a network that includes SAM/SAH metabolism. Altering the proportion of metabolites relative to each other may alter methylation potential and therefore epigenetic reactions. Others have shown that SAM/SAH correlated with differences in methylation at metastable epialleles based on season and food availability (Waterland et al. 2010). Changes in epigenetic programming at critical developmental windows such as in utero, early childhood, or during puberty have been associated with developmental plasticity, health, and susceptibility to chronic diseases in adults (Barker et al. 1993; Gluckman et al. 2009; Kussmann et al. 2010).
Met_PC1 and plasma proteins
Met_PC1 was also associated with levels of pro-inflammatory proteins. Individuals with high vitamin A, Hcy (but still below the clinical cutoff of 15 µmol/L), and SAM/SAH had lower levels of many of these inflammatory proteins. The correlation was modest for any single protein to met_PC1 value. However, certain proteins shared similar correlation coefficients and functional analysis based on gene ontologies, and some of these correlated proteins that participated in the same networks. Since met_PC1 is an empirically defined value specific to this study, the correlations among these plasma metabolites will necessarily require testing in other genetic makeups and environments.
Met_PC1 and global protein topological analysis
To discover whether the met_PC1 variable was associated with genetic variation in molecular interaction networks or subsystems, a metabolic/protein–protein interaction network was constructed based on two manually curated interaction databases (Ma et al. 2007; Yu et al. 2012). The network was partitioned into topological modules, each of which was assessed for significant enrichment with met_PC1-correlated genes using a hypergeometric test. Three modules were identified, 2 of which contained substantial numbers of immune and metabolic function genes and the third included genes in a range of secretory and gastrointestinal functions. Although the met_PC1 genes were not directly functionally annotated to every one of the identified processes/pathways in these modules, they may either be directly contained in these processes/pathways, or indirectly connected via a small number of degrees of separation. Variation in plasma micronutrient levels and Hcy, and erythrocyte SAM and SAH, and specifically the ratios of these metabolites relative to each other, was associated with genetic variations in immune and gastrointestinal functions. Chronic disturbance in gastrointestinal function (such as that seen in inflammatory bowel disease, Crohn’s disease, and environmental enteropathy) may directly contribute to micronutrient deficiencies due to altered nutrient absorption (Valentini et al. 2008). Although the cohort in the present study did not present with diagnosed intestinal disorders, it may be the case that a range of SNPs contributes to subclinical variation in gastrointestinal function, which then relates to variation in micronutrient levels. Additional focused work on enterocytes and intestinal immune cells would be required to clarify the potential functional consequences of the SNPs identified in this study.
Metabolite principal component 1 (Met_PC1) was derived from the relationship of vitamins A, E, thiamine, pyridoxal, and the metabolites Hcy, SAM/SAH ratio. Correlations of met_PC1 to immunity were not unexpected since a rich literature exists for individual micronutrients and various aspects of immune function and regulation (Bhaskaram 2002; Maggini et al. 2007; Baeke et al. 2010; Ströhle et al. 2011; Ooi et al. 2012). The genetic analysis nevertheless revealed new insights into the many genes and their functions that may be associated with different plasma levels of metabolites and therefore with diet.
Strategy and methodological considerations
The use of CBPR with biomedical and network biology analyses. Community-based research engages participants in the research study and provides opportunities for health- and nutrition-related exchanges between community members and researchers. Community-based research is done in “real” time with lifestyle and other environmental conditions that are not under the control of the researcher. These factors likely introduce noise into the study and analyses, but the measured biological “signals” include the contribution from those unmeasured influences. Our goal was to measure as many physiological and environmental variables as possible to associate signals with phenotype. In addition, community-based results from such studies are likely to be translated more rapidly to individuals and populations (McCabe-Sellers et al. 2008).
Data from this study were previously analyzed at the group level (such as between SAM/SAH groups) and at the population level (Monteiro et al. 2014). We have also extensively analyzed dietary intake patterns, metabolomics, proteomic, and genomic data for individual participants in this study. For example, dietary intake variables were compared to metabolite patterns in each participant to determine whether common patterns could be identified at the individual level, and DNA ancestry was analyzed for each individual for the possibility of using genetic admixture mapping methods (Cheng et al. 2010) (data not shown). Methods which identify groups of individuals with related metabolic features but still allow for n-of-1 analysis may extend the recent personal omics analysis for molecular and medical phenotypes (Chen et al. 2012). Reporting data and results from studies with more than one individual, however, may require development of novel publication strategies.
Levels of metabolites in each participant were analyzed and shown in one figure as opposed to reporting results of the average metabolite level in separate graphs. Although such methods are common in transcriptomic and metabolomics literature, we identified patterns of metabolite levels that revealed unanticipated nutrient–nutrient statistical interactions. Standard PCA converted the graphic representation of metabolite levels to a statistic specific called metabolite principal component 1 (met_PC1). Met_PC1 represented 6 strongly and one weakly associated (vitamin D) measured metabolites and their interactions. Although the value of this statistic is specific for the study reported here, similar methods may allow for more comprehensive analyses of interacting metabolites.
Analyzing genetic differences based on met_PC1 in a metabolic/PPI network partitioned into topological modules allowed for the identification of physiological functions (immune, metabolic, and secretory) associated with gene–metabolite relationships identified in our statistical analysis. Rather than seeking a small number of SNPs with large effect on our phenotype, our network-based approach inherently highlights multivariate groups of functionally related SNPs/genes that are statistically associated with a phenotype. Defined interventions can be developed from our results and, equally importantly, tested by measuring parameters of immune and GI function that were identified in this study.
Implications for reproducibility
The results described in this manuscript and recent publications on intra-individual variability in physiological status found in environments that have large changes in nutrient availability (Dominguez-Salas et al. 2013) demonstrate the difficulty in replicating biomedical research, particularly for genetic and gene–environment interaction associations. Hierarchical clustering of proteins (Fig. 2) and SNPs (Fig. 5) correlated with met_PC1 helps visualize the proteomic and genotypic differences as combinations of SNPs or proteins rather than as single markers (even though these were derived from univariate analysis and corrected for multiple comparisons). No single protein or SNP is always correlated with met_PC1. This is what we would expect to observe, in part because of gene–gene and gene–environment interactions, epigenetic regulation, and other interactions. This perspective better fits the biological reality of multiple genes and their products contributing to a complex phenotype (in this case met_PC1). Our working hypothesis is that testing these associations in other populations and also in others experimental designs would be needed to identify common patterns of variants within these genotype data sets that might explain the percentage of genetic contribution to a given phenotype. Subsets of genes will contribute or not contribute to a complex phenotype (e.g., obesity and diabetes) based on interactions with diet or other environmental factors. In addition, the adaptations to diverse environments over human evolution may have selected different collections of genes for similar environments in different environments. The most notable, and still controversial example, is the different functional adaptations to Tibetan and Andean high altitudes (Beall 2007). Nevertheless, we predict that gene–environment interactions producing the same phenotype will have overlapping genes (much like Venn diagrams). Some pathways and therefore genes will be shared, and others may contribute less significantly in different genetic subpopulations. Discovering these similarities and differences may lead to an understanding of targeting diet and lifestyles to optimize health.
This research was funded by the Division of Personalized Nutrition and Medicine at the US FDA National Center for Toxicological Research (Jefferson, AR), the USDA Agricultural Research Service Delta Obesity Prevention Research Unit (Little Rock, AR), and the Nestlé Institute of Health Sciences (Lausanne, Switzerland). The research team gratefully acknowledges the contributions of the Phillips County community members in the Marvell (AR) School District for their participation in this research. This research was funded by the Division of Personalized Nutrition and Medicine at the US FDA National Center for Toxicological Research (Jefferson AR), the USDA Agricultural Research Service Delta Obesity Prevention Research Unit (Little Rock, AR), and the Nestlé Institute of Health Sciences (Lausanne, Switzerland). We thank Donna Mendrick, Li-Rong Yu, and Ritchie Feurs for critically reading this manuscript.
Conflict of interest
J.K. is employed by the Nestle Institute of Health Sciences, a for-profit company. No other authors declared a conflict of interest.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.
- Afacan NJ, Fjell CD, Hancock REW (2012) A systems biology approach to nutritional immunology—focus on innate immunity. Mol Asp Med 33:14–25. doi:10.1016/j.mam.2011.10.013 View ArticleGoogle Scholar
- Auffray C, Chen Z, Hood L (2009) Systems medicine: the future of medical genomics and healthcare. Genome Med 1:2. doi:10.1186/gm2 PubMed CentralPubMedView ArticleGoogle Scholar
- Baeke F, Gysemans C, Korf H, Mathieu C (2010) Vitamin D insufficiency: implications for the immune system. Pediatr Nephrol 25:1597–1606. doi:10.1007/s00467-010-1452-y PubMedView ArticleGoogle Scholar
- Barker DJ, Gluckman PD, Godfrey KM et al (1993) Fetal nutrition and cardiovascular disease in adult life. Lancet 341:938–941. doi:10.1016/0140-6736(93)91224-A PubMedView ArticleGoogle Scholar
- Beadle G, Tatum E (1941) Genetic control of biochemical reactions in neurospora. Proc Natl Acad Sci USA 27:499–506PubMed CentralPubMedView ArticleGoogle Scholar
- Beall CM (2007) Two routes to functional adaptation: Tibetan and Andean high-altitude natives. Proc Natl Acad Sci USA 104(Suppl.):8655–8660. doi:10.1073/pnas.0701985104 PubMed CentralPubMedView ArticleGoogle Scholar
- Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B Stat Method 57:289–300Google Scholar
- Bhaskaram P (2002) Micronutrient malnutrition, infection, and immunity: an overview. Nutr Rev 60:S40–S45PubMedView ArticleGoogle Scholar
- Bindea G, Mlecnik B, Hackl H et al (2009) ClueGO: a cytoscape plug-into decipher functionally grouped gene ontology and pathway annotation networks. Bioinformatics 25:1091–1093. doi:10.1093/bioinformatics/btp101 PubMed CentralPubMedView ArticleGoogle Scholar
- Brody EN, Gold L (2000) Aptamers as therapeutic and diagnostic agents. J Biotechnol 74:5–13PubMedGoogle Scholar
- Brody E, Gold L, Mehan M et al (2012) Life’s simple measures: unlocking the proteome. J Mol Biol 422:595–606. doi:10.1016/j.jmb.2012.06.021 PubMedView ArticleGoogle Scholar
- Bustamante CD, Burchard EG, De la Vega FM (2011) Genomics for the world. Nature 475:163–165. doi:10.1038/475163a PubMed CentralPubMedView ArticleGoogle Scholar
- Chen R, Mias GI, Li-Pook-Than J et al (2012) Personal omics profiling reveals dynamic molecular and medical phenotypes. Cell 148:1293–1307. doi:10.1016/j.cell.2012.02.009 PubMed CentralPubMedView ArticleGoogle Scholar
- Cheng CY, Reich D, Coresh J et al (2010) Admixture mapping of obesity-related traits in African Americans: the Atherosclerosis Risk in Communities (ARIC) Study. Obes (Silver Spring) 18:563–572. doi:10.1038/oby.2009.282 View ArticleGoogle Scholar
- Csardi G, Nepusz T (2006) The igraph software package for complex network research. InterJournal Complex Syst 1695. http://cran.r-project.org/web/packages/igraph/citation.html
- Dominguez-Salas P, Moore SE, Cole D et al (2013) DNA methylation potential: dietary intake and blood concentrations of one-carbon metabolites and cofactors in rural African women. Am J Clin Nutr 97:1217–1227. doi:10.3945/ajcn.112.048462.The PubMed CentralPubMedView ArticleGoogle Scholar
- Fenech M, El-Sohemy A, Cahill L et al (2011) Nutrigenetics and nutrigenomics: viewpoints on the current status and applications in nutrition research and practice. J Nutrigenet Nutrigenomics 4:69–89. doi:10.1159/000327772 PubMed CentralPubMedView ArticleGoogle Scholar
- Gardy JL, Lynn DJ, Brinkman FSL, Hancock REW (2009) Enabling a systems biology approach to immunology: focus on innate immunity. Trends Immunol 30:249–262. doi:10.1016/j.it.2009.03.009 PubMedView ArticleGoogle Scholar
- Gluckman PD, Hanson MA, Buklijas T et al (2009) Epigenetic mechanisms that underpin metabolic and cardiovascular diseases. Nat Rev Endocrinol 5:401–408. doi:10.1038/nrendo.2009.102 PubMedView ArticleGoogle Scholar
- Gold L (1995) Oligonucleotides as research diagnostic, and therapeutic agents. J Biol Chem 270:13581–13584PubMedView ArticleGoogle Scholar
- Gold L, Ayers D, Bertino J et al (2010) Aptamer-based multiplexed proteomic technology for biomarker discovery. PLoS One 5:e15004. doi:10.1371/journal.pone.0015004 PubMed CentralPubMedView ArticleGoogle Scholar
- Gold L, Walker JJ, Wilcox SK, Williams S (2011) Advances in human proteomics at high scale with the SOMAscan proteomics platform. N Biotechnol. doi:10.1016/j.nbt.2011.11.016 PubMedGoogle Scholar
- Goldstein DB (2009) Common genetic variation and human traits. N Engl J Med 360:1696–1698. doi:10.1056/NEJMp0806284 PubMedView ArticleGoogle Scholar
- Hamza TH, Chen H, Hill-Burns EM et al (2011) Genome-wide gene–environment study identifies glutamate receptor gene GRIN2A as a Parkinson’s disease modifier gene via interaction with coffee. PLoS Genet 7:e1002237. doi:10.1371/journal.pgen.1002237 PubMed CentralPubMedView ArticleGoogle Scholar
- Harris SC, Fang H, Su Z et al (2009) FDA bioinformatics tool for public use—ArrayTrack™. Methods Mol Biol 563:379–398. doi:10.1007/978-1-60761-175-2_20
- Hemani G, Yang J, Vinkhuyzen A et al (2013) Inference of the genetic architecture underlying BMI and height with the use of 20,240 sibling pairs. Am J Hum Genet 93:865–875. doi:10.1016/j.ajhg.2013.10.005 PubMed CentralPubMedView ArticleGoogle Scholar
- Højsgaard S, Halekoh U, Yan J (2006) The R package geepack for generalized estimating equations. J Stat Softw 15:1–11Google Scholar
- Illig T, Gieger C, Zhai G et al (2010) A genome-wide perspective of genetic variation in human metabolism. Nat Genet 42:137–141. doi:10.1038/ng.507 PubMed CentralPubMedView ArticleGoogle Scholar
- Kalupahana NS, Moustaid-moussa N (2011) Overview of symposium “systems genetics in nutrition and obesity research” 1(2):3–5. doi:10.3945/jn.110.130104.512
- Kaput J, Rodriguez RL (2004) Nutritional genomics: the next frontier in the postgenomic era. Physiol Genomics 16:166–177PubMedView ArticleGoogle Scholar
- Kaput J, Swartz D, Paisley E et al (1994) Diet–disease interactions at the molecular level: an experimental paradigm. J Nutr 124:1296S–1305SPubMedGoogle Scholar
- Kaput J, Klein KG, Reyes EJ et al (2004) Identification of genes contributing to the obese yellow Avy phenotype: caloric restriction, genotype, diet × genotype interactions. Physiol Genomics 18:316–324PubMedView ArticleGoogle Scholar
- Karlsson FH, Nookaew I, Petranovic D, Nielsen J (2011) Prospects for systems biology and modeling of the gut microbiome. Trends Biotechnol 29:251–258. doi:10.1016/j.tibtech.2011.01.009 PubMedView ArticleGoogle Scholar
- Kleemann R, Bureeva S, Perlina A et al (2011) A systems biology strategy for predicting similarities and differences of drug effects: evidence for drug-specific modulation of inflammation in atherosclerosis. BMC Syst Biol 5:125. doi:10.1186/1752-0509-5-125 PubMed CentralPubMedView ArticleGoogle Scholar
- Kraemer S, Vaught JD, Bock C et al (2011) From SOMAmer-based biomarker discovery to diagnostic and clinical applications: a SOMAmer-based, streamlined multiplex proteomic assay. PLoS One 6:e26332. doi:10.1371/journal.pone.0026332 PubMed CentralPubMedView ArticleGoogle Scholar
- Kussmann M, Krause L, Siffert W (2010) Nutrigenomics: where are we with genetic and epigenetic markers for disposition and susceptibility? Nutr Rev 68(Suppl. 1):S38–S47. doi:10.1111/j.1753-4887.2010.00326.x PubMedView ArticleGoogle Scholar
- Lee Y-C, Lai C-Q, Ordovas JM, Parnell LD (2011) A database of gene–environment interactions pertaining to blood lipid traits, cardiovascular disease and type 2 diabetes. J Data Min Genomics Proteomics 2:1–8. doi:10.4172/2153-0602.1000106 Google Scholar
- Liu JZ, McRae AF, Nyholt DR et al (2010) A versatile gene-based test for genome-wide association studies. Am J Hum Genet 87:139–145. doi:10.1016/j.ajhg.2010.06.009 PubMed CentralPubMedView ArticleGoogle Scholar
- Ma H, Sorokin A, Mazein A et al (2007) The Edinburgh human metabolic network reconstruction and its functional analysis. Mol Syst Biol 3:135. doi:10.1038/msb4100177 PubMed CentralPubMedView ArticleGoogle Scholar
- Maggini S, Wintergerst ES, Beveridge S, Hornig DH (2007) Selected vitamins and trace elements support immune function by strengthening epithelial barriers and cellular and humoral immune responses. Br J Nutr 98(Suppl. 1):S29–S35. doi:10.1017/S0007114507832971 PubMedGoogle Scholar
- McCabe-Sellers B, Lovera D, Nuss H et al (2008) Personalizing nutrigenomics research through community based participatory research and omics technologies. OMICS 12:263–272. doi:10.1089/omi.2008.0041 PubMedView ArticleGoogle Scholar
- Meng Q, Mäkinen V-P, Luk H, Yang X (2013) Systems biology approaches and applications in obesity, diabetes, and cardiovascular diseases. Curr Cardiovasc Risk Rep 7:73–83. doi:10.1007/s12170-012-0280-y PubMed CentralPubMedView ArticleGoogle Scholar
- Monteiro J, Wise C, Morine M et al (2014) Methylation potential associated with diet, genotype, protein, and metabolite levels in the delta obesity vitamin study. Genes Nutr 9(3):403–418. doi:10.1007/s12263-014-0403-9
- Morine MJ, McMonagle J, Toomey S et al (2010) Bi-directional gene set enrichment and canonical correlation analysis identify key diet-sensitive pathways and biomarkers of metabolic syndrome. BMC Bioinform 11:499. doi:10.1186/1471-2105-11-499 View ArticleGoogle Scholar
- Morine MJ, Tierney AC, van Ommen B et al (2011) Transcriptomic coordination in the human metabolic network reveals links between n-3 fat intake, adipose tissue gene expression and metabolic health. PLoS Comput Biol 7:e1002223. doi:10.1371/journal.pcbi.1002223 PubMed CentralPubMedView ArticleGoogle Scholar
- Morine MJ, Toomey S, McGillicuddy FC et al (2012) Network analysis of adipose tissue gene expression highlights altered metabolic and regulatory transcriptomic activity in high-fat-diet-fed IL-1RI knockout mice. J Nutr Biochem. doi:10.1016/j.jnutbio.2012.04.012 Google Scholar
- Nettleton JA, McKeown NM, Kanoni S et al (2010) Interactions of dietary whole-grain intake with fasting glucose—and insulin-related genetic loci in individuals of European descent. Diabetes Care 33:2684–2691. doi:10.2337/dc10-1150 PubMed CentralPubMedView ArticleGoogle Scholar
- Olson MV (2012) Human genetic individuality. Annu Rev Genomics Hum Genet 13:1–27. doi:10.1146/annurev-genom-090711-163825 PubMedView ArticleGoogle Scholar
- Ooi JH, Chen J, Cantorna MT (2012) Vitamin D regulation of immune function in the gut: why do T cells have vitamin D receptors? Mol Asp Med 33:77–82. doi:10.1016/j.mam.2011.10.014 View ArticleGoogle Scholar
- Ordovás JM, Robertson R, Cléirigh EN (2011) Gene–gene and gene–environment interactions defining lipid-related traits. Curr Opin Lipidol 22:129–136. doi:10.1097/MOL.0b013e32834477a9 PubMedView ArticleGoogle Scholar
- Ostroff R, Foreman T, Keeney TR et al (2010) The stability of the circulating human proteome to variations in sample collection and handling procedures measured with an aptamer-based proteomics array. J Proteomics 73:649–666. doi:10.1016/j.jprot.2009.09.004 PubMedView ArticleGoogle Scholar
- Patel CJ, Bhattacharya J, Butte AJ (2010) An Environment-Wide Association Study (EWAS) on type 2 diabetes mellitus. PLoS One 5:e10746. doi:10.1371/journal.pone.0010746 PubMed CentralPubMedView ArticleGoogle Scholar
- Patel CJ, Chen R, Butte AJ (2012a) Data-driven integration of epidemiological and toxicological data to select candidate interacting genes and environmental factors in association with disease. Bioinformatics 28:i121–i126. doi:10.1093/bioinformatics/bts229 PubMed CentralPubMedView ArticleGoogle Scholar
- Patel CJ, Cullen MR, Ioannidis JPA, Butte AJ (2012b) Systematic evaluation of environmental factors: persistent pollutants and nutrients correlated with serum lipid levels. Int J Epidemiol 41:828–843. doi:10.1093/ije/dys003 PubMed CentralPubMedView ArticleGoogle Scholar
- Ramos E, Rotimi C (2009) The A’s, G’s, C’s, and T’s of health disparities. BMC Med Genomics 2:29. doi:10.1186/1755-8794-2-29 PubMed CentralPubMedView ArticleGoogle Scholar
- Reichardt J, Bornholdt S (2006) Statistical mechanics of community detection. Phys Rev E Stat Nonlin Soft Matter Phys 74:016110PubMedView ArticleGoogle Scholar
- Roux AVD (2011) Complex systems thinking and current impasses in health disparities research. Am J Public Health 101:1627–1634. doi:10.2105/AJPH.2011.300149.Complex View ArticleGoogle Scholar
- Schulz LO, Bennett PH, Ravussin E et al (2006) Effects of traditional and western environments on prevalence of type 2 diabetes in Pima Indians in Mexico and the U.S. Diabetes Care 29:1866–1871. doi:10.2337/dc06-0138 PubMedView ArticleGoogle Scholar
- Slikker W Jr, Paule MG, Wright LK et al (2007) Systems biology approaches for toxicology. J Appl Toxicol 27:201–217. doi:10.1002/jat.1207 PubMedView ArticleGoogle Scholar
- So L, Fruman DA (2012) PI3K signaling in B and T lymphocytes: new developments and therapeutic advances. Biochem J 442:465–481. doi:10.1042/BJ20112092.PI3K PubMed CentralPubMedView ArticleGoogle Scholar
- Stahl EA, Wegmann D, Trynka G et al (2012) Bayesian inference analyses of the polygenic architecture of rheumatoid arthritis. Nat Genet 44:483–489. doi:10.1038/ng.2232 PubMedView ArticleGoogle Scholar
- Ströhle A, Wolters M, Hahn A (2011) Micronutrients at the interface between inflammation and infection–ascorbic acid and calciferol. Part 2: calciferol and the significance of nutrient supplements. Inflamm Allergy Drug Targets 10:64–74PubMedView ArticleGoogle Scholar
- Suhre K, Shin SY, Petersen AK et al (2011) Human metabolic individuality in biomedical and pharmaceutical research. Nature 477:54–60. doi:10.1038/nature10354 PubMedView ArticleGoogle Scholar
- Tanaka T, Ngwa JS, van Rooij FJA et al (2013) Genome-wide meta-analysis of observational studies shows common genetic variants associated with macronutrient intake. Am J Clin Nutr 97:1395–1402. doi:10.3945/ajcn.112.052183 PubMed CentralPubMedView ArticleGoogle Scholar
- Valentini L, Schaper L, Buning C et al (2008) Malnutrition and impaired muscle strength in patients with Crohn’s disease and ulcerative colitis in remission. Nutrition 24:694–702. doi:10.1016/j.nut.2008.03.018 PubMedView ArticleGoogle Scholar
- Von Bertalanffy L (1950) The theory of open systems in physics and biology. Science 111(2872):23–29View ArticleGoogle Scholar
- Waterland RA, Kellermayer R, Laritsky E et al (2010) Season of conception in rural Gambia affects DNA methylation at putative human metastable epialleles. PLoS Genet 6:e1001252. doi:10.1371/journal.pgen.1001252 PubMed CentralPubMedView ArticleGoogle Scholar
- Williams RP (1956) Biochemical individuality: the basis for the genetotrophic concept. Keats, New CanaanGoogle Scholar
- Xu J, Wise C, Varma V et al (2010) Two new ArrayTrack libraries for personalized biomedical research. BMC Bioinform 11(Suppl. 6):S6. doi:10.1186/1471-2105-11-S6-S6 View ArticleGoogle Scholar
- Yu X, Wallqvist A, Reifman J (2012) Inferring high-confidence human protein–protein interactions. BMC Bioinform 13:79. doi:10.1186/1471-2105-13-79 View ArticleGoogle Scholar