Human Milk Composition Is Associated with Maternal Body Mass Index in a Cross-Sectional, Untargeted Metabolomics Analysis of Human Milk from Guatemalan Mothers

Background Maternal overweight and obesity has been associated with poor lactation performance including delayed lactogenesis and reduced duration. However, the effect on human milk composition is less well understood. Objectives We evaluated the relationship of maternal BMI on the human milk metabolome among Guatemalan mothers. Methods We used data from 75 Guatemalan mothers who participated in the Household Air Pollution Intervention Network trial. Maternal BMI was measured between 9 and <20 weeks of gestation. Milk samples were collected at a single time point using aseptic collection from one breast at 6 mo postpartum and analyzed using high-resolution mass spectrometry. A cross-sectional untargeted high-resolution metabolomics analysis was performed by coupling hydrophilic interaction liquid chromatography (HILIC) and reverse phase C18 chromatography with mass spectrometry. Metabolic features associated with maternal BMI were determined by a metabolome-wide association study (MWAS), adjusting for baseline maternal age, education, and dietary diversity, and perturbations in metabolic pathways were identified by pathway enrichment analysis. Results The mean age of participants at baseline was 23.62 ± 3.81 y, and mean BMI was 24.27 ± 4.22 kg/m2. Of the total metabolic features detected by HILIC column (19,199 features) and by C18 column (11,594 features), BMI was associated with 1026 HILIC and 500 C18 features. Enriched pathways represented amino acid metabolism, galactose metabolism, and xenobiotic metabolic metabolism. However, no significant features were identified after adjusting for multiple comparisons using the Benjamini–Hochberg false discovery rate procedure (FDRBH < 0.2). Conclusions Findings from this untargeted MWAS indicate that maternal BMI is associated with metabolic perturbations of galactose metabolism, xenobiotic metabolism, and xenobiotic metabolism by cytochrome p450 and biosynthesis of amino acid pathways. Significant metabolic pathway alterations detected in human milk were associated with energy metabolism-related pathways including carbohydrate and amino acid metabolism. This trial was registered at clinicaltrials.gov as NCT02944682.


Introduction
Maternal obesity has been linked to the increased risk for adverse maternal and fetal outcomes, including gestational diabetes mellitus (GDM), hypertensive disorders of pregnancy, cesarean delivery, miscarriage, stillbirth, preterm birth, and fetal macrosomia [1,2].Maternal obesity is also one of the strongest predictors of childhood obesity, increasing risk by more than 2-fold, and accounts for 12%-15% of the population attributable risk of childhood obesity [3][4][5].Several studies have indicated that maternal overweight and obesity is associated with changes in infant feeding practices, including delayed lactogenesis and reduced duration of feeding [6,7].Human milk has a dynamic composition of nutrients and bioactive compounds, including enzymes, hormones, cytokines, and more, and varies in composition within a feeding, diurnally, over the entire course of lactation and between mothers and populations [8].Animal models demonstrate that maternal obesity and high-fat diets negatively impact milk fat production, by blunting the diurnal fluctuation in metabolism, suppressing induction of carbohydrate utilization and altering trafficking of dietary and de novo-derived calories to the mammary glands, impacting energy balance and fuel utilization [9].
However, gaps remain in our understanding of metabolic consequences between maternal overweight and obesity and milk composition.Epidemiological studies have produced contradictory results because it is challenging to disentangle the effects of certain exposures on human milk composition from other prenatal and maternal exposures, as well as other infant feeding practices [10,11].It has been hypothesized that milk from mothers with obesity may contain different amounts of obesogenic components, compared with protective components [12].A meta-analysis identified a 0.56 g/L increase in milk fat for every 1 kg/m 2 increase in maternal prepregnancy BMI, translating to an estimated 16.5% increase in milk fat for a mother with a BMI of 30 compared with a mother with a BMI 18.5 [13].
High-resolution metabolomics (HRM) is an analytic technique to identify a broad spectrum of endogenous metabolites, food and microbiome metabolites, and environmental chemical exposures [14,15].Untargeted HRM identifies thousands of metabolites across different metabolic pathways, can provide relative metabolic quantification, and enables investigations of association between external exposure and endogenous processes or adaptations, as well as hypothesis generation for future studies [16].For example, an untargeted analysis of human milk samples from 94 females at 6 mo postpartum identified enriched monosaccharides and sugar alcohols in milk from mothers with obesity, suggesting that maternal BMI may influence the maturation of transitional milk to mature milk [17].Traditional composition studies have linked maternal BMI to differences in lipid composition (a higher proportion of SFAs and lower n-3:n-6 PUFA ratio), increased milk content of insulin, leptin, C-reactive protein, IL-6, and TNF-α, lower adiponectin and differences in oligosaccharide composition [1,18,19].Investigations into the role of maternal BMI or obesity on associated changes in human milk composition, largely via targeted metabolome-associated changes, have returned mixed results and indicate the need for broader untargeted studies [20].
We conducted an untargeted HRM analysis of human milk samples from 75 mothers in Jalapa, Guatemala to expand our understanding of the metabolic consequences of maternal overweight and obesity on human milk composition.This crosssectional analysis was nested in a randomized control trial (RCT) that investigated the effect of a cleaner cookstove and fuel intervention on health outcomes in pregnant females and their infants.We hypothesized that mothers with overweight and/or obesity would have metabolic perturbations associated with inflammation and oxidative stress compared with normal weight mothers.

Study population
Data were collected from participants who were enrolled in the HAPIN (Household Air Pollution Intervention Network) trial in Guatemala.Detailed information about the trial design and methods has been previously described [21].HAPIN was a multicenter, parallel-group, individually RCT to determine the effect of a liquefied petroleum gas (LPG) cookstove and free fuel intervention on infant health outcomes including low birth weight, severe pneumonia, and stunting in the first year of life [21].HAPIN was implemented between May 2018 and September 2021 in ten resource-poor, rural settings of Jalapa, Guatemala; Tamil Nadu, India; Puno, Peru; and Kayonza, Rwanda [21].The current study utilized data from HAPIN participants residing in rural communities in Guatemala's Jalapa municipality.
The study has been registered with ClinicalTrials.gov(NCT02944682).The study protocol was reviewed and approved by institutional review boards or ethics committees at Emory University (00089799), Universidad del Valle de Guatemala (146-08-2016/11-2016), and the Guatemalan Ministry of Health National Ethics Committee .

HAPIN eligibility
Pregnant females 18 to <35 y of age, with a confirmed single viable pregnancy between 9 and <20 weeks of gestation, were eligible to participate in the HAPIN trial if they lived in the study area, predominately used biomass stoves (for example, wood and animal dung or charcoal) and provided written informed consent [21].Pregnant females were excluded if they smoked cigarettes or other tobacco products, planned to move outside of the study area, or used or were likely to predominately use an LPG stove in the near future [21].

HAPIN randomization and trial group assignment
After eligibility and consent, enrolled females were randomly assigned in a 1:1 ratio to the intervention (LPG stove with continuous fuel supply for ~18 mo, education and behaviorbased messaging to promote safe, exclusive use of the LPG cookstove) or control arms throughout the ~15-mo recruitment period [21].Females in the control arm were expected to follow routine cooking practices but received compensation to minimize loss to follow-up and offset the economic advantage accorded to intervention households.Blinding at the participant and field staff level was not possible [21].

HAPIN data collection
Detailed information about data collection for HAPIN participants has been previously described [21].In brief, pregnant females were followed through pregnancy and the first year of their infant's life [21].This analysis used data collected at baseline, including maternal anthropometry, demographic characteristics, and data collected at 6 mo postpartum, including dietary diversity and milk samples.

Biosample collection and processing
Within HAPIN, we conducted a pilot substudy to differentiate the infant microbiome and maternal metabolome in rural Guatemala infants and their mothers relative to household air pollution exposure.HAPIN-Guatemala mothers who gave birth between February and August 2019 were recruited from the hospital postpartum ward following birth for the collection of nasal and oropharyngeal samples at birth and milk samples at 6 mo after birth using a convenience sampling approach.Mothers were excluded from participation if they delivered by cesarean section and if the mother reported that she did not intend to nurse her infant at the birth visit.Mothers were further excluded from human milk sampling at the 6-mo visit if they had mastitis, had used antibiotics in the past 48 h, or were no longer nursing.Mastitis was determined by self-report and visual observation by a female surveyor at the time of collection.
Milk samples were collected by a trained Guatemalan female surveyor in the privacy of the participant's home using a standard protocol for aseptic collection at the same time in the morning (AE2 h) for all participants.Nurses showed mothers how to clean the nipple and surrounding area before initiating pumping via breast pump or manual expression.Mothers were asked to fully empty one breast to ensure full expression of foremilk and hindmilk.Milk was expressed into a sterile collection bottle and the aliquot of expressed milk was poured back into 2 sterile tubes.Samples were stored in a À20 C frost-free freezer and then transported to the Universidad del Valle de Guatemala laboratory where they were stored in a À80 C freezer.Samples were shipped on dry ice to Emory's School of Nursing Research Lab and then transferred to the Clinical Biomarker Laboratory at Emory University for metabolomics profiling.Prediabetes or diabetes status was assessed among females via self-report in a survey questionnaire.

Maternal anthropometry
Baseline height and weight were measured at the first baseline visit when mothers were in their first or second trimester of pregnancy.Maternal height was measured using a Seca 213 stadiometer (seca GmbH) in centimeters to the nearest 0.1 cm.Maternal weight was measured using a Seca 876 scale (seca GmbH) and measured to the nearest 0.1 kg.Two measurements for height and weight were taken.If the first 2 measurements for height and weight had a difference of >1 cm (height) or 0.5 kg (weight), then a third measurement was taken, and the 2 closest measurements were used to calculate average height and weight.Maternal BMI was operationalized as a continuous variable because baseline maternal BMI measurements are not reflective of prepregnancy weight and BMI.Traditional BMI categories (underweight, normal, overweight, and obese) were not calculated given that BMI was based on weight measurements at the baseline assessment when participants were late in their first, or early in their second, trimester of pregnancy.

Covariates
Maternal age and education were assessed at baseline, whereas maternal dietary diversity was measured at 6 mo postpartum.Maternal education was categorized as follows: 1) no formal education or some primary school, 2) primary school or some secondary school, or 3) secondary, vocational, or some university.Maternal dietary diversity was assessed using the tool adapted from the FAO of the United States to calculate Minimum Diet Diversity for Women [22].Females were asked to recall the consumption of a prespecified list of food groups in the past 24 h at baseline and at 6 mo postpartum, which were then categorized into 10 categories including grains, white roots, tubers, and plantains; pulses; nuts and seeds; dairy; meat, poultry, and fish; eggs; dark green leafy vegetables; other vitamin A-rich fruits and vegetables; other vegetables; and other fruits.Individual consumption was summed into a score ranging from 0 to 10 based on yes/no responses to the prespecified food groups and each woman was classified as achieving dietary diversity using a cutoff value of !5 [23].

Metabolomics profiling
Human milk samples were analyzed in triplicates across 4 batches using liquid chromatography coupled with highresolution tandem mass spectroscopy (LC-MS) following an established protocol [28,29].Two chromatography columns were used to enhance the coverage of metabolic features detection-the hydrophilic interaction liquid chromatography (HILIC) with positive electrospray ionization (ESI) and reverse phase (C18) chromatography with negative ESI.A pooled milk sample that was referenced against the National Institute of Standards and Technology standard reference material was added to each batch.Quality control reference samples were included at the beginning and end of each analytic batch for normalization, control for background noise, batch evaluation, and post hoc quantification.
Data were extracted using an adaptive apLCMS R package (Tianwei Yu, version 6.6.9) which performs peak detection, noise removal, peak quantification, peak alignment, and recovery of weak signals [30].Another R package, xMSanalyzer (version 2.0.6.1) was used to optimize peak detection by merging results from different parameter settings, quality evaluation of samples and features, mass-to-charge ratio (m/z) calibration using internal standards and confirmed metabolites, and m/z calibration using internal standards and confirmed metabolites and batch-effect evaluation and correction [31].Detected signals are known as metabolic features and are uniquely defined by m/z, retention time, and ion intensity.

Statistical analysis
Descriptive baseline analyses of the 75 study participants who provided milk samples were conducted using SAS/STAT (version 9.4).We investigated the association of continuous maternal BMI with milk metabolic feature intensities using multivariate linear regression.In the following model: Model 1: Y ij refers to the log 2 intensity of metabolic feature j for participant I, β 0 is the intercept, which refers to the predicted value of the log 2 intensity of metabolic feature when setting continuous variables to zero and categorical variables to their referent value.ε ij is the residual random normal error and BMI i refers to the continuous maternal BMI for participant i.
Covariates and potential confounding factors were identified a priori, from a review of existing literature and use of a directed acyclic graph (DAG) (Figure 1), which include maternal age, maternal education, and maternal dietary diversity in the adjusted model.
Metabolomics analyses were performed using R (version 4.2.1), and statistical significance was defined at α < 0.05.The untargeted MWAS examined the association between maternal BMI and global metabolic signals and pathways without a priori knowledge of chemical identities.Separate MWAS models were built for each metabolic feature from each dataset (HILIC/þESI and C18/-ESI) [32].The FDR was controlled using the Benjamini-Hochberg procedure to correct for multiple comparisons [33].However, a limited number of significant features were detected when using the FDR, due to the small sample size used in this pilot study, and thus, a less conservative corrected P value threshold (FDR < 0.2) was used to decrease the possibility of false negatives.FDR < 0.2 was selected as the threshold to be able to generate a novel hypothesis because 0.2 is consistent with what has been commonly used in the field of environmental metabolomics [34].

Pathway enrichment
Pathway analyses predict biological functions and molecular mechanisms associated with significant metabolic features.We used both Mummichog (version 1.0.9) and Metapone (version 3.18) for the pathway testing to infer and categorize functional biological activity directly from mass spectrometry output.Mummichog (version 1.0.9) is used to infer functional activity and metabolic pathways and networks without prior chemical identification [35].The algorithm uses m/z values to match possible metabolites to metabolic features and construct pathways based on tentative identification.Significant features (raw P < 0.05 by limma test) were further studied by pathway enrichment analyses using Mummichog (version 1.0.9).This approach protects against type 2 statistical error by including all features at P < 0.05 and protects against type 1 statistical error by permutation testing in pathway enrichment analysis.Metapone (version 3.18) is a newer method that can jointly analyze positive-and negative-ion mode data to generate a more integrated and comprehensive view of the metabolic perturbations and avoid double counting [36].

Metabolite identification
Significant features were matched to the METLIN Metabolite and Chemical Entity Database, ChemSpider, Human Metabolome Database, and Kyoto Encyclopedia of Genes and Genomes using a mass error threshold of 5 ppm [31].Chemical identities were confirmed by manual visual inspection of extracted ion chromatographs for all significant metabolites.To distinguish true peaks from noise signals and minimize false positives, peaks must have exhibited clear Gaussian peak shapes and a signal-to-noise ratio greater than 3:1.True metabolic features were then confirmed with level 1 confidence using Metabolomics Standards Initiative criteria of features whose m/z (AE10 ppm difference) and retention time (AE10 s) matched the authentic compounds [30].Level 1 evidence indicates that the proposed metabolites have been confirmed via appropriate measurement of an authentic reference standard with the same experimental conditions as the milk samples.Metabolic features not assigned with level 1 confidence were annotated by xMSannotator (version 2.0.6.1)[32].

Sample population
Nurses approached all females who delivered at the hospital between February and August 2019 for eligibility in the pilot substudy.Seventy-six mothers were recruited and consented to participate in the pilot from the intervention group and 72 mothers consented to participate in the pilot from the control group.
However, based on budgetary constraints, a sample size of 75 was predetermined for human milk sample collection and analysis.A total of 38 mothers from the HAPIN intervention and 37 mothers from the HAPIN control groups provided milk samples at the 6 mo postpartum household visit (Figure 2).

Population characteristics
Characteristics of the study population (Table 1) were similar to those of the larger HAPIN-Guatemala cohort (Supplemental Table 1) [37]

Metabolome changes associated with BMI
After processing and data quality checks, 19,199 metabolic features from the HILIC/þESI column and 11,594 metabolic features from the C18/ÀESI column were included in the final analyses, as can be seen in Figure 3.
Two MWAS models were analyzed using data from each of the chromatography columns.After adjusting for covariates, maternal age, maternal education, and maternal dietary diversity, 1000 m/z features were significantly associated with continuous maternal BMI in the HILIC ESI and 590 m/z features in the C18 ESI (P < 0.05), as can be seen in Figure 3.After adjusting for multiple comparison testing, no significant m/z features were identified for either column (FDR BH < 0.2).Below, we present the significant metabolic pathways and metabolites identified from the covariate-adjust raw model.

Pathways enrichment analysis
Metapone identified 8 metabolic pathways that were significantly associated with maternal BMI including galactose metabolism, biosynthesis of amino acids, xenobiotics metabolism and metabolism of xenobiotics by cytochrome p450, fructose and mannose metabolism, C5-branched dibasic acid metabolism, mineral absorption, and serotonergic synapse (Supplemental Figure 1).Significant pathways, number of significant metabolites, and putative identification of significant metabolites associated with maternal BMI are presented in Table 2.
Pathway enrichment analyses were analyzed separately in Mummichog (version 1.0.9) for significant metabolic features identified in the HILIC/þESI and C18/ÀESI columns.Sixteen metabolic pathways, from the HILIC/þESI column, and 12 metabolic pathways, from the C18/ÀESI column, were significantly associated with maternal BMI.The Mummichog output for all pathways associated with continuous maternal BMI can be found in Supplemental Table 2.There was no overlap in pathways identified between features detected in the HILIC and C18 modes.Pathways significantly associated with maternal BMI include vitamin groups [thiamin (vitamin B-1)], pyridoxine (vitamin B-6), biotin (vitamin H/B-7), ascorbate (vitamin C), fatty acid metabolism (fatty acid metabolism, fatty acid activation, de novo fatty acid biosynthesis), lipid metabolism (linoleate metabolism, glycosphingolipid and glycosphingolipid-globoseries metabolism), carbohydrate metabolism (hexose metabolism, galactose metabolism, starch and sucrose metabolism), nucleotide metabolism (aminosugar metabolism and purine metabolism), and amino acid metabolism (tyrosine metabolism, purine metabolism, and lysine metabolism).

Metabolite annotation and identity confirmation
Twenty-eight metabolites associated with BMI were confirmed and identified with level 1 evidence (Table 3).Specifically, 23 metabolites were confirmed identified from the HILIC/þESI column and 5 metabolites were confirmed identified from the C18/ÀESI column.Confirmed metabolites from both columns were related to C5-branched dibasic acid metabolism and were negatively associated with maternal BMI.Metabolites from the HILIC/þESI column were related to caffeine metabolism and tryptophan metabolism and were positively associated with maternal BMI.In addition, metabolites from both datasets were related to amino acid metabolism including branched-chain amino acid (BCAA) metabolism (valine, leucine, and isoleucine), tyrosine metabolism, cysteine and methionine, phenylalanine, and arginine metabolism.Corresponding parameter estimates from the MWAS models are also presented in Table 3, indicating if the metabolites are positively or negatively associated with maternal BMI.

Sensitivity analysis
Sensitivity analyses were conducted to reduce false discoveries and to confirm the consistency of the results in our analyses.Different P values, including both raw (P < 0.05; P < 0.005) and FDR-corrected (FDR < 0.2), were used as different cutoff points for the pathway enrichment analyses [35].All pathways from both the HILIC and C18 columns that were identified at P value of <0.05 were also identified at a more stringent P value of <0.005, indicating consistency in our results.

Discussion
In this exploratory study, we conducted HRM to characterize human milk composition and explore changes in the milk metabolome associated with maternal BMI, from among a subpopulation of Guatemalan mothers participating in the HAPIN trial.This untargeted HRM analysis identified metabolites and metabolic pathways involved in energy metabolism that were significantly associated with maternal BMI, including galactose, biosynthesis of amino acids, and xenobiotic metabolism, including cytochrome p450.This is one of the few untargeted HRM studies in the literature to identify that maternal weight, BMI, or overweight and obesity are significantly associated with the human milk metabolome [1,10,20].
A key finding is that metabolites in the galactose metabolism pathway were associated with maternal BMI.Galactose is a component of lactose and a product of hepatic galactose metabolism [38].In this analysis, galactitol, a reduction product formed from excess galactose, was positively associated with maternal BMI.Our metabolome-wide study showing alteration in galactose metabolites contributes additional evidence that maternal BMI is associated with changes in galactose  metabolism.A secondary analysis of breastfeeding participants with normal weight and obesity identified that metabolites involved in galactose metabolism were enriched (P < 0.001) over the first 6 mo among females with obesity and remained significant following FDR correction (P ¼ 0.01) [17].In addition, BMI classification was associated with metabolites involved in galactose metabolism (P ¼ 0.001) [17].
Metabolites related to the pathways for biosynthesis of amino acids, including methionine, phenylpyruvate, and citrulline, were negatively associated with maternal BMI.Each of these metabolites is involved in the biosynthesis of arginine, cysteine, methionine, and phenylalanine, meaning that these biosynthetic pathways may be altered or dysregulated among mothers with higher BMIs.A study to differentiate the human milk microbiota in colostrum of mothers with GDM or obesity similarly identified that amino acids biosynthesis pathways were overrepresented in mothers with GDM [39].In addition, a 2-wk reduced-fat and reduced-sugar dietary intervention during lactation altered amino acid biosynthesis pathways (P ¼ 0.005) in the stool microbiome among exclusively nursed infants [40].Additional maternal dietary interventions including high-fat compared with high-carbohydrate and high-glucose compared with high-galactose diets increased bacterial metabolic pathways involved in amino acid biosynthesis in human milk [40].Results from the literature, and other omics analyses, indicate the need for multiomics studies to integrate metabolomics and microbiome methods to assess the role of maternal BMI on amino acid-related metabolic pathways and long-term health implications for mothers and infants as a result of dysregulated amino acid metabolic pathways [41].Xenobiotics are chemicals found in but not produced by the organisms or the ecological system being studied.Some naturally occurring endobiotic compounds can also become xenobiotics when present in the environment at excessive concentrations [42].Cytochrome p450s are a superfamily of protein enzymes involved in the synthesis and metabolism of a range of internal and extracellular components.In this study, xenobiotics metabolism and xenobiotic metabolized by cytochrome p450 were associated with maternal BMI.Caffeine, one of the most widely used xenobiotic compounds, is metabolized by the cytochrome p450 oxidase enzyme system into paraxanthine, theobromine, and theophylline [43].All L-1 confirmed caffeine metabolites identified in this study, including theophylline, paraxanthine, theobromine, and dimethylxanthine, were positively associated with maternal BMI.
The literature on the role of xenobiotics in human milk is sparse-1 study suggests that variations in diet or dietary patterns may be associated with concentration of xenobiotic metabolites in human milk [1].Another study examining alterations in urine samples from preterm infants consuming donor human milk compared with their own mother's milk identified differences in 11 metabolites that were related to metabolism of xenobiotics by cytochrome p450 [44].Our results add to the literature by providing the first evidence of associations, to our knowledge, among maternal BMI, xenobiotic metabolism, and metabolism of xenobiotics by cytochrome p450.
The pathways described above-galactose metabolism, biosynthesis of amino acids, and xenobiotic metabolism-have been examined in studies of the infant fecal metabolome [45,46].Fecal metabolites belonging to galactose metabolism were associated with infant age among exclusively nursed infants, whereas biosynthesis of amino acids was enriched in formula-fed infants, compared with nursed infants [45].Xenobiotics were also identified as the highest milk-associated bacterial compound found in the neonatal fecal metabolome, indicating milk-fecal cross-talk [46].These results indicate that maternal BMI, maternal diet, and infant diet (namely human milk) may interact in shaping infant health, as reflected in the infant fecal metabolome.
This analysis additionally confirmed the identities of 28 chemical metabolites, as presented in Table 3, which were associated with maternal BMI.As indicated in Table 3, all metabolites, except for caffeine metabolites, were negatively associated with maternal BMI.Our results indicate associations between maternal BMI and milk metabolites predictive of insulin resistance and diabetes risk.We found that tyramine, a monoamine derivative of tyrosine and product of amino acid metabolism, was negatively associated with maternal BMI.Tyrosine is an aromatic amino acid and has been found to be increased in both human milk and plasma of females with obesity and negatively associated with metabolic health [10].In 1 study among mother-infant dyads, the concentration of tyrosine was 30% higher (P ¼ 0.016) in human milk from mothers with obesity, compared with human milk from normal weight mothers [47].In another study, maternal exposure to nonnutritive sweeteners (NNS) during pregnancy and lactation caused critical changes in amino acid metabolism in NNS pups [48].Amino acids were the most deregulated class of metabolites with tyramine being highly upregulated [48].Our results are consistent with the literature as aromatic amino acids have strong predictive value for insulin resistance and type 2 diabetes [49,50].In addition, tyrosine competes with branched-chain amino acids for L-type amino acid transporter responsible for cell and tissue entry and this could explain the higher tyrosine concentrations found in the plasma of individuals with obesity [51].
In addition to tyrosine, 2-methylmaleate, a precursor to BCAA biosynthesis, is predictive of diabetes risk and associated with diabetes development [52].In our study, 2-methylmaleate was negatively associated with maternal BMI.A study examining free amino acid concentration in human milk between females with obesity and females with normal weight mothers identified that females with obesity produce human milk that differs from females with normal weight [53].In addition, free BCAAs were higher in human milk from females with obesity and maternal BMI was positively associated with isoleucine and leucine concentrations over the first 6 mo of infant life [45].BCAAs are associated with the development of type 2 diabetes mellitus (T2DM) and nonalcoholic fatty liver disease and increased BCAA concentration strongly predict T2DM [53].Research has suggested targeting BCAA catabolism as a mechanism to treat obesity-associated insulin resistance [54].
Our results also indicate significant associations between maternal BMI and metabolites that may be predictive of future GDM risk [55].Specifically, we observed associations between maternal BMI and LPC(14:0) and LPC(20:1), which are part of a class of lipids, lysophosphatidylcholines (LPCs), that activate multiple signaling pathways involved in oxidative stress and inflammatory response [55].LPC(14:0) is associated with obesity risk in the literature [56].A study to identify BMI-related lipids and the role of lipids linking BMI and GDM identified a relationship between BMI and LPC(14:0) [57].Furthermore, BMI-associated lipids, including LPC(14:0), were estimated to explain ~66.4% of the relationship between higher BMI and GDM, suggesting that BMI-related lipids may play a role in GDM pathogenesis [57].Metabolomics literature on LPCs also indicates associations between LPCs with infant weight gain, childhood overweight, and obesity [58].Future studies could examine the role of metabolomic LPCs and obesity risk in maternal-infant dyads.
Similarly, we observed a positive association between maternal BMI and a lipid biomarker, Cer(D18:1/16:0), which is one of the most well-studied and validated lipid biomarkers predictive of morbidity and mortality, including GDM [59].A nested case-control study from a prospective cohort of pregnant females identified that high Cer18:1 was associated with higher GDM risk [60].Another study demonstrated that Cer18:1 concentration was higher among females who eventually developed GDM, compared with females with normal glucose tolerance [61].Cer18:1 concentrations were elevated in pregnancy and changes were not independent of BMI [61].Because increased maternal overweight or maternal obesity is associated with increased GDM risk, it is conceivable that Cer18:1, enriched in females with GDM risk or developed GDM, would also be associated with maternal BMI.
The main strength of this analysis is the use of cutting-edge untargeted HRM to investigate biological mechanisms associated with differences in maternal BMI in a cohort of Guatemalan mothers.This analysis is unique in the use of human milk as a nonintrusive biological tool to examine changes in the milk metabolome from females living in a rural environment.Although smaller than epidemiological studies, the sample size for this analysis is comparable with other hypotheses generating untargeted HRM in the literature [62,63].In addition, multiple comparison testing, and sensitivity analyses were conducted to reduce risk of false positive discoveries.
There are a few limitations that must be considered in the interpretation of our findings.First, the single time point sampling and cross-sectional analysis neither allow us to consider the effect of the known variations in human milk composition, such as lactation phase and variation within and between feeds nor would we be able to make any causal inference.Second, this analysis was conducted with a small sample size.Although this is common in hypothesis-generating metabolomics studies, the analysis was not well-powered to detect a significant difference after adjusting for multiple comparisons.Third, maternal BMI was only measured in the first and second trimesters of pregnancy for all mothers and may not reflect prepregnancy BMI or postpregnancy weight gain.However, Retnakaran et al. [64] prospectively evaluated the relationship between pregravid weight and first trimester weight.The authors recognize that pregravid and first trimester weight measurements are unlikely to be identical due to modest physiological weight gain in the first trimester [64].However, there was good concordance between weight measurements suggesting that first trimester weight measurement can be a reasonable surrogate for pregravid weight, with the caveat that differences between these measurements are generally modest but can vary in magnitude [64].Although all mothers in this study sample were nursing at 6 mo postpartum, we did not collect additional information on infant feeding practices.This study population was conveniently sampled from the larger Guatemalan HAPIN cohort, so the results from this analysis may not be generalizable to the full cohort or to a broader population.There is a risk of selection bias as mothers were sampled in postpartum wards following the birth of the infant, but this study sample is demographically similar to the larger HAPIN-Guatemala cohort, as presented in Supplemental Table 1 [37].Although a DAG was used to identify and control for significant covariates, potentially important confounders related to maternal BMI (such as patterns of infant feeding) were not collected or included in this analysis.Despite the limitations, this study contributes to the growing literature on human milk composition using untargeted metabolomics. (

FIGURE 1 .
FIGURE 1. Directed acyclic graph to identify confounders associated with maternal BMI and milk composition.Minimally sufficient adjustment sets for estimating the effect of maternal BMI (exposure) on milk composition (outcome) are highlighted in pink.Green arrows represent causal paths and pink arrows represent biasing paths.Yellow circles represent ancestors of exposure (maternal BMI), whereas blue circles represent ancestors of outcome (milk composition).Pink circles are ancestors of exposure and outcome.

FIGURE 2 .
FIGURE 2. Participant flow chart for identifying n ¼ 75 sample size for breast milk sample collection.LPG, liquefied petroleum gas.

FIGURE 3 .
FIGURE 3. Manhattan plot of the association between changes in metabolic features associated with continuous maternal BMI.The x-axis presents the retention time of the metabolic features, and the y-axis presents the negative natural log of the P value in exposure to metabolites association.Red dots represent metabolic features that are positively associated with maternal BMI and blue triangles represent metabolic features negatively associated with maternal BMI.The dashed red line is the threshold-log 10 (P < 0.05).HILIC, hydrophilic interaction liquid chromatography.
. The mean age of participants at baseline was 23.62 AE 3.81 y, whereas the mean gestational age of participants at enrollment was 14.64 wk AE 2.88 wk.The mean BMI was 24.27 AE 4.22 kg/m 2 .However, there was a wide range of BMI measurements in this sample with the median at 23.32 kg/m 2 (IQR: 21.71-26.08kg/m 2 ).In addition, no females had prediabetes or diabetes at baseline.Diet diversity was low at baseline (mean: 2.65, 95% CI: 2.45, 2.86) and remained low at 6 mo postpartum (mean: 2.81; 95% CI: 2.56, 3.06).Diet diversity was almost uniformly lacking in diversity (with 97% consuming fewer than 5 food groups per day).The majority reported consuming grains and legumes daily, with over half reporting weekly or monthly consumption of chicken or eggs and less than half reporting weekly or monthly consumption of dark greens and fruit.

TABLE 1
Sociodemographic characteristics of study population

TABLE 2
Significant pathways, number of significant metabolites, and putative identification of significant metabolites associated with maternal BMI