Metabolomic data presents challenges for epidemiological meta-analysis: a case study of childhood body mass index from the ECHO consortium.

INTRODUCTION: Meta-analyses across diverse independent studies provide improved confidence in results. However, within the context of metabolomic epidemiology, meta-analysis investigations are complicated by differences in study design, data acquisition, and other factors that may impact reproducibility. OBJECTIVE: The objective of this study was to identify maternal blood metabolites during pregnancy (> 24 gestational weeks) related to offspring body mass index (BMI) at age two years through a meta-analysis framework. METHODS: We used adjusted linear regression summary statistics from three cohorts (total N = 1012 mother-child pairs) participating in the NIH Environmental influences on Child Health Outcomes (ECHO) Program. We applied a random-effects meta-analysis framework to regression results and adjusted by false discovery rate (FDR) using the Benjamini-Hochberg procedure. RESULTS: Only 20 metabolites were detected in all three cohorts, with an additional 127 metabolites detected in two of three cohorts. Of these 147, 6 maternal metabolites were nominally associated (P < 0.05) with offspring BMI z-scores at age 2 years in a meta-analytic framework including at least two studies: arabinose (Coefmeta = 0.40 [95% CI 0.10,0.70], Pmeta = 9.7 × 10-3), guanidinoacetate (Coefmeta = - 0.28 [- 0.54, - 0.02], Pmeta = 0.033), 3-ureidopropionate (Coefmeta = 0.22 [0.017,0.41], Pmeta = 0.033), 1-methylhistidine (Coefmeta = - 0.18 [- 0.33, - 0.04], Pmeta = 0.011), serine (Coefmeta = - 0.18 [- 0.36, - 0.01], Pmeta = 0.034), and lysine (Coefmeta = - 0.16 [- 0.32, - 0.01], Pmeta = 0.044). No associations were robust to multiple testing correction. CONCLUSIONS: Despite including three cohorts with large sample sizes (N > 100), we failed to identify significant metabolite associations after FDR correction. Our investigation demonstrates difficulties in applying epidemiological meta-analysis to clinical metabolomics, emphasizes challenges to reproducibility, and highlights the need for standardized best practices in metabolomic epidemiology.

Patterns of infant fecal metabolite concentrations and social behavioral development in toddlers.

BACKGROUND: Gut-derived metabolites, products of microbial and host co-metabolism, may inform mechanisms underlying children's neurodevelopment. We investigated whether infant fecal metabolites were related to toddler social behavior. METHODS: Stool samples collected from 6-week-olds (n = 86) and 1-year-olds (n = 209) in the New Hampshire Birth Cohort Study (NHBCS) were analyzed using nuclear magnetic resonance spectroscopy metabolomics. Autism-related behavior in 3-year-olds was assessed by caregivers using the Social Responsiveness Scale (SRS-2). To assess the association between metabolites and SRS-2 scores, we used a traditional single-metabolite approach, quantitative metabolite set enrichment (QEA), and self-organizing maps (SOMs). RESULTS: Using a single-metabolite approach and QEA, no individual fecal metabolite or metabolite set at either age was associated with SRS-2 scores. Using the SOM method, fecal metabolites of six-week-olds organized into four profiles, which were unrelated to SRS-2 scores. In 1-year-olds, one of twelve fecal metabolite profiles was associated with fewer autism-related behaviors, with SRS-2 scores 3.4 (95%CI: -7, 0.2) points lower than the referent group. This profile had higher concentrations of lactate and lower concentrations of short chain fatty acids than the reference. CONCLUSIONS: We uncovered metabolic profiles in infant stool associated with subsequent social behavior, highlighting one potential mechanism by which gut bacteria may influence neurobehavior. IMPACT: Differences in host and microbial metabolism may explain variability in neurobehavioral phenotypes, but prior studies do not have consistent results. We applied three statistical techniques to explore fecal metabolite differences related to social behavior, including self-organizing maps (SOMs), a novel machine learning algorithm. A 1-year-old fecal metabolite pattern characterized by high lactate and low short-chain fatty acid concentrations, identified using SOMs, was associated with social behavior less indicative of autism spectrum disorder. Our findings suggest that social behavior may be related to metabolite profiles and that future studies may uncover novel findings by applying the SOM algorithm.

Multi-omics analysis of glucose-mediated signaling by a moonlighting Gß protein Asc1/RACK1.

Heterotrimeric G proteins were originally discovered through efforts to understand the effects of hormones, such as glucagon and epinephrine, on glucose metabolism. On the other hand, many cellular metabolites, including glucose, serve as ligands for G protein-coupled receptors. Here we investigate the consequences of glucose-mediated receptor signaling, and in particular the role of a Gα subunit Gpa2 and a non-canonical Gß subunit, known as Asc1 in yeast and RACK1 in animals. Asc1/RACK1 is of particular interest because it has multiple, seemingly unrelated, functions in the cell. The existence of such "moonlighting" operations has complicated the determination of phenotype from genotype. Through a comparative analysis of individual gene deletion mutants, and by integrating transcriptomics and metabolomics measurements, we have determined the relative contributions of the Gα and Gß protein subunits to glucose-initiated processes in yeast. We determined that Gpa2 is primarily involved in regulating carbohydrate metabolism while Asc1 is primarily involved in amino acid metabolism. Both proteins are involved in regulating purine metabolism. Of the two subunits, Gpa2 regulates a greater number of gene transcripts and was particularly important in determining the amplitude of response to glucose addition. We conclude that the two G protein subunits regulate distinct but complementary processes downstream of the glucose-sensing receptor, as well as processes that lead ultimately to changes in cell growth and metabolism.

Information enhanced model selection for Gaussian graphical model with application to metabolomic data.

In light of the low signal-to-noise nature of many large biological data sets, we propose a novel method to learn the structure of association networks using Gaussian graphical models combined with prior knowledge. Our strategy includes two parts. In the first part, we propose a model selection criterion called structural Bayesian information criterion, in which the prior structure is modeled and incorporated into Bayesian information criterion. It is shown that the popular extended Bayesian information criterion is a special case of structural Bayesian information criterion. In the second part, we propose a two-step algorithm to construct the candidate model pool. The algorithm is data-driven and the prior structure is embedded into the candidate model automatically. Theoretical investigation shows that under some mild conditions structural Bayesian information criterion is a consistent model selection criterion for high-dimensional Gaussian graphical model. Simulation studies validate the superiority of the proposed algorithm over the existing ones and show the robustness to the model misspecification. Application to relative concentration data from infant feces collected from subjects enrolled in a large molecular epidemiological cohort study validates that metabolic pathway involvement is a statistically significant factor for the conditional dependence between metabolites. Furthermore, new relationships among metabolites are discovered which can not be identified by the conventional methods of pathway analysis. Some of them have been widely recognized in biological literature.

Associations of maternal plasma and umbilical cord plasma metabolomics profiles with birth anthropometric measures.

BACKGROUND: The metabolomics profiles of maternal plasma during pregnancy and cord plasma at birth might influence fetal growth and birth anthropometry. The objective was to examine how maternal plasma and umbilical cord plasma metabolites are associated with newborn anthropometric measures, a known predictor of future health outcomes. METHODS: Pregnant women between 24 and 28 weeks of gestation were recruited as part of a prospective cohort study. Blood samples from 413 women at enrollment and 787 infant cord blood samples were analyzed using the Biocrates AbsoluteIDQ® p180 kit. Multivariable linear regression models were used to examine associations of cord and maternal metabolites with infant anthropometry at birth. RESULTS: In cord blood samples from this rural cohort from New Hampshire of largely white residents, 13 metabolites showed negative associations, and 10 metabolites showed positive associations with birth weight Z-score. Acylcarnitine C5 showed negative association, and 4 lysophosphatidylcholines showed positive associations with birth length Z-score. Maternal blood metabolites did not significantly correlate with birth weight and length Z-scores. CONCLUSIONS: Consistent findings were observed for several acylcarnitines that play a role in utilization of energy sources, and a lysophosphatidylcholine that is part of oxidative stress and inflammatory response pathways in cord plasma samples. IMPACT: The metabolomics profiles of maternal plasma during pregnancy and cord plasma at birth may influence fetal growth and birth anthropometry. This study examines the independent effects of maternal gestational and infant cord blood metabolomes across different classes of metabolites on birth anthropometry. Acylcarnitine species were negatively associated and glycerophospholipids species were positively associated with weight and length Z-scores at birth in the cord plasma samples, but not in the maternal plasma samples. This study identifies lipid metabolites in infants that possibly may affect early growth.

Longitudinal associations of pre-pregnancy BMI and gestational weight gain with maternal urinary metabolites: an NYU CHES study.

BACKGROUND/OBJECTIVES: Excessive gestational weight gain (GWG) and pre-pregnancy obesity affect a significant portion of the US pregnant population and are linked with negative maternal and child health outcomes. The objective of this study was to explore associations of pre-pregnancy body mass index (pBMI) and GWG with longitudinally measured maternal urinary metabolites throughout pregnancy. SUBJECTS/METHODS: Among 652 participants in the New York University Children's Health and Environment Study, a longitudinal pregnancy cohort, targeted metabolomics were measured in serially collected urine samples throughout pregnancy. Metabolites were measured at median 10 (T1), 21 (T2), and 29 (T3) weeks gestation using the Biocrates AbsoluteIDQ® p180 Urine Extension kit. Acylcarnitine, amino acid, biogenic amine, phosphatidylcholine, lysophosphatidylcholine, sphingolipid, and sugar levels were quantified. Pregnant people 18 years or older, without type 1 or 2 diabetes and with singleton live births and valid pBMI and metabolomics data were included. GWG and pBMI were calculated using weight and height data obtained from electronic health records. Linear mixed effects models with interactions with time were fit to determine the gestational age-specific associations of categorical pBMI and continuous interval-specific GWG with urinary metabolites. All analyses were corrected for false discovery rate. RESULTS: Participants with obesity had lower long-chain acylcarnitine levels throughout pregnancy and lower phosphatidylcholine and glucogenic amino acids and higher phenylethylamine concentrations in T2 and T3 compared with participants with normal/underweight pBMI. GWG was associated with taurine in T2 and T3 and C5 acylcarnitine species, C5:1, C5-DC, and C5-M-DC, in T2. CONCLUSIONS: pBMI and GWG were associated with the metabolic environment of pregnant individuals, particularly in relation to mid-pregnancy. These results highlight the importance of both preconception and prenatal maternal health.

LC-MS-based metabolomics analysis to identify meprin-ß-associated changes in kidney tissue from mice with STZ-induced type 1 diabetes and diabetic kidney injury.

Meprin metalloproteases have been implicated in the pathophysiology of diabetic kidney disease (DKD). Single-nucleotide polymorphisms in the meprin-ß gene have been associated with DKD in Pima Indians, a Native American ethnic group with an extremely high prevalence of DKD. In African American men with diabetes, urinary meprin excretion positively correlated with the severity of kidney injury. In mice, meprin activity decreased at the onset of diabetic kidney injury. Several studies have identified meprin targets in the kidney. However, it is not known how proteolytic processing of the targets by meprins impacts the metabolite milieu in kidneys. In the present study, global metabolomics analysis identified differentiating metabolites in kidney tissues from wild-type and meprin-ß knockout mice with streptozotocin (STZ)-induced type 1 diabetes. Kidney tissues were harvested at 8 wk post-STZ and analyzed by hydrophilic interaction liquid chromatography ultra-performance liquid chromatography-quadrupole time-of-flight mass spectrometry. Principal component analysis identified >200 peaks associated with diabetes. Meprin expression-associated metabolites with strong variable importance of projection scores were indoxyl sulfate, N-γ-l-glutamyl-l-aspartic acid, N-methyl-4-pyridone-3-carboxamide, inosine, and cis-5-decenedioic acid. N-methyl-4-pyridone-3-carboxamide has been previously implicated in kidney injury, and its isomers, 4-PY and 2-PY, are markers of peroxisome proliferation and inflammation that correlate with creatinine clearance and glucose tolerance. Meprin deficiency-associated differentiating metabolites with high variable importance of projection scores were cortisol, hydroxymethoxyphenylcarboxylic acid-O-sulfate, and isovaleryalanine. The data suggest that meprin-ß activity enhances diabetic kidney injury in part by altering the metabolite balance in kidneys, favoring high levels of uremic toxins such as indoxyl sulfate and N-methyl-pyridone-carboxamide.

Broad spectrum metabolomics for detection of abnormal metabolic pathways in a mouse model for retinitis pigmentosa.

Retinitis pigmentosa (RP) is a degenerative disease of the retina that affects approximately 1 million people worldwide. There are multiple genetic causes of this disease, for which, at present, there are no effective therapeutic strategies. In the present report, we utilized broad spectrum metabolomics to identify perturbations in the metabolism of the rd10 mouse, a genetic model for RP that contains a mutation in Pde6ß. These data provide novel insights into mechanisms that are potentially critical for retinal degeneration. C57BL/6J and rd10 mice were raised in cyclic light followed by either light or dark adaptation at postnatal day (P) 18, an early stage in the degeneration process. Mice raised entirely in the dark until P18 were also evaluated. After euthanasia, retinas were removed and extracted for analysis by ultra-performance liquid chromatography-time of flight-mass spectrometry (UPLC-QTOF-MS). Compared to wild type mice, rd10 mice raised in cyclic light or in complete darkness demonstrate significant alterations in retinal pyrimidine and purine nucleotide metabolism, potentially disrupting deoxynucleotide pools necessary for mitochondrial DNA replication. Other metabolites that demonstrate significant increases are the Coenzyme A intermediate, 4'-phosphopantothenate, and acylcarnitines. The changes in these metabolites, identified for the first time in a model of RP, are highly likely to disrupt normal energy metabolism. High levels of nitrosoproline were also detected in rd10 retinas relative to those from wild type mice. These results suggest that nitrosative stress may be involved in retinal degeneration in this mouse model.

Meprin ß metalloproteases associated with differential metabolite profiles in the plasma and urine of mice with type 1 diabetes and diabetic nephropathy.

BACKGROUND: Meprin metalloproteases are abundantly expressed in the brush border membranes of kidney proximal tubules and small intestines. Meprins are also expressed in podocytes and leukocytes (monocytes and macrophages). Meprins are implicated in the pathophysiology of diabetic nephropathy (DN) but underlying mechanisms are not fully understood. Single nucleotide polymophisms (SNPs) in the meprin ß gene were associated with DKD in human subjects. Furthermore, meprin α and ß double deficiency resulted in more severe kidney injury and higher mortality rates in mice with Streptozotocin (STZ)-induced type 1 diabetes. Identification of meprin substrates has provided insights on how meprins could modulate kidney injury. Meprin targets in the kidney include extracellular matrix (ECM) proteins, modulators of inflammation, and proteins involved in the protein kinase A (PKA) and PKC signaling pathways. The current study used a global metabolomics approach to determine how meprin ß expression impacts the metabolite milieu in diabetes and DKD. METHODS: Low dose STZ was used to induce type 1 diabetes in 8-week old wild-type (WT) and meprin ß knockout (ßKO) mice. Blood and urine samples were obtained at 4 and 8 weeks post-STZ injection. Assays for albumin, creatinine, neutrophil gelatinase-associated lipocalin (NGAL), kidney injury molecule - 1 (KIM-1), and cystatin C were used for biochemical assessment of kidney injury. Data for biomarkers of kidney injury utilized two-way ANOVA. Metabolomics data analysis utilized UPLC-QTOF MS and multivariate statistics. RESULTS: The number of metabolites with diabetes-associated changes in levels were significantly higher in the WT mice when compared to meprin ßKO counterparts. Annotated meprin ß expression-associated metabolites with strong variable importance in projection (VIP) scores play roles in lipid metabolism (LysoPC(16:1(9Z)), taurocholic acid), amino acid metabolism (indoxyl sulfate, hippuric acid), and neurotransmitter/stress hormone synthesis (cortisol, 3-methoxy-4-hydroxyphenylethylene glycolsulfate, homovanillic acid sulfate). Metabolites that associated with meprin ß deficiency include; 3,5-dihydroxy-3',4'-dimethoxy-6,7-methylenedioxyflavone 3-glucuronide, pantothenic acid, and indoxyl glucuronide (all decreased in plasma). CONCLUSION: Taken together, the annotated metabolites suggest that meprin ß impacts complications of diabetes such as DKD by altering distinct metabolite profiles.

Neonatal Metabolomic Profiles Related to Prenatal Arsenic Exposure.

Prenatal inorganic arsenic (iAs) exposure is associated with health effects evident at birth and later in life. An understanding of the relationship between prenatal iAs exposure and alterations in the neonatal metabolome could reveal critical molecular modifications, potentially underpinning disease etiologies. In this study, nuclear magnetic resonance (NMR) spectroscopy-based metabolomic analysis was used to identify metabolites in neonate cord serum associated with prenatal iAs exposure in participants from the Biomarkers of Exposure to ARsenic (BEAR) pregnancy cohort, in Gómez Palacio, Mexico. Through multivariable linear regression, ten cord serum metabolites were identified as significantly associated with total urinary iAs and/or iAs metabolites, measured as %iAs, %monomethylated arsenicals (MMAs), and %dimethylated arsenicals (DMAs). A total of 17 metabolites were identified as significantly associated with total iAs and/or iAs metabolites in cord serum. These metabolites are indicative of changes in important biochemical pathways such as vitamin metabolism, the citric acid (TCA) cycle, and amino acid metabolism. These data highlight that maternal biotransformation of iAs and neonatal levels of iAs and its metabolites are associated with differences in neonate cord metabolomic profiles. The results demonstrate the potential utility of metabolites as biomarkers/indicators of in utero environmental exposure.