Dietary Intake Influences Metabolites in Healthy Infants: A Scoping Review

Metabolites are generated from exogenous sources such as diet. This scoping review will summarize nascent metabolite literature and discriminating metabolites for formula vs. human- milk-fed infants. Using the PICOS framework (P—Patient, Problem or Population; I—Intervention; C—Comparison; O—Outcome; S—Study Design) and PRISMA item-reporting protocols, infants less than 12 months old, full-term, and previously healthy were included. Protocol was registered with Open Science Framework (OSF). Publications from 1 January 2009–2019 were selected, for various biofluids, study designs, and techniques (such as high-performance liquid chromatography (HPLC)). From 711 articles, blinded screening of 214 articles using Abstrackr® software, resulted in 24 for final review. Strengthening the Reporting of Observational studies in Epidemiology (STROBE) guidelines were adopted, which included a 24-point checklist. Articles were stratified according to biofluid. Of articles reporting discriminating metabolites between formula- and human milk-fed infants, 62.5% (5/8) of plasma/serum/dried blood spot, 88% (7/8) of urine and 100% (6/6) of feces related articles reported such discriminating metabolites. Overall, no differences were found between analytical approach used (targeted (n = 9) vs. un-targeted (n = 10)). Current articles are limited by small sample sizes and differing methodological approaches. Of the metabolites reviewed herein, fecal metabolites provided the greatest distinction between diets, which may be indicative of usefulness for future diet metabolite-focused work.


Introduction
Infant health is strongly correlated to their diet of human milk or infant formula [1][2][3]. Metabolomic analysis of biofluids, such as feces, urine, and blood, is useful to determine the effects of diet on the infant's development and health [4]. Metabolites are small molecules that are created from diet, chemicals, drugs or tissues when biochemical reactions occur in the body, which influence molecular interactions between genes and proteins leading to a phenotype [5]. Some metabolites are created by microbial processes occurring in the gastrointestinal tract of the individual [6]. The analysis of metabolites may enable individualized medicine allowing for prescribed diets that lead to positive health outcomes [7,8], and metabolomic profiles have been shown to differ between infants fed human milk and those fed infant formula [9].

Search Strategy and Inclusion Criteria
Searches were performed in PubMed, Embase, and Web of Science between 9 May 2019 and 25 June 2019. The search strategy used in PubMed is reported in Supplementary Materials Table S3. Searches in Embase and Web of Science were similar and limited to articles published in English, between 1 January 2008 and 31 December 2018. Studies were included if they reported results from healthy, full-term infants less than 12 months of age and reported dietary intake (human milk, formula, mixed diet) as well as metabolomic data from infant biofluid (plasma/serum, stool, urine, dried blood spots (DBS), human milk). All studies, regardless of the technique used to analyze metabolites (such as high-performance liquid chromatography (HPLC), matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF), nuclear magnetic resonance (NMR)), were included.
Grey literature, opinion papers, letters, case studies, case control studies, and review articles were excluded. Studies focused primarily on older children, premature infants, mothers rather than infants, or non-healthy infants were excluded as well, and studies including both human and non-human infants in the same study were excluded. As cord blood would not reflect human milk or infant formula intake, studies with only cord blood metabolites were also excluded.

Article Screening and Data Abstraction
Authors followed the PRISMA guidelines for item reporting (Figure 1) [30]. All abstracts were screened for inclusion/exclusion by two of the authors using Abstrackr ® [31]. After the first round, articles selected for inclusion were then re-screened by the same authors in their full-text form.
Disagreements were discussed until a consensus was reached. The initial search yielded 214 relevant results after de-duplication. Of these, 180 were excluded through abstract screening, and a further 10 were excluded during full-text screening (analyzed bacteria in feces rather than metabolites (n = 4); analyzed <3 metabolites (n = 5); dietary exposure unclear (n = 1)). Accordingly, a total of 24 articles were analyzed fully . This scoping review focused on metabolomic analyses as opposed to individual and single biomarkers, therefore studies with fewer than three metabolites analyzed were excluded from further review.
Nutrients 2020, 12, x FOR PEER REVIEW  3 of 19 were excluded during full-text screening (analyzed bacteria in feces rather than metabolites (n = 4); analyzed <3 metabolites (n = 5); dietary exposure unclear (n = 1)). Accordingly, a total of 24 articles were analyzed fully . This scoping review focused on metabolomic analyses as opposed to individual and single biomarkers, therefore studies with fewer than three metabolites analyzed were excluded from further review. Abstracted data as summarized in the tables included: the PMID, first author, year of publication, diets, ages, sample sizes (where available initial cohort samples size was listed as well as the sample size for metabolite analysis), assay, targeted vs. un-targeted, whether results were discriminatory, and if so at what age and for which metabolite.

Quality of Reporting Assessment
Given that one of the weaknesses of scoping reviews is the lack of a quality of reporting assessment [56], selected papers were evaluated using the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [57]. These guidelines provide a basis Abstracted data as summarized in the tables included: the PMID, first author, year of publication, diets, ages, sample sizes (where available initial cohort samples size was listed as well as the sample size for metabolite analysis), assay, targeted vs. un-targeted, whether results were discriminatory, and if so at what age and for which metabolite.

Quality of Reporting Assessment
Given that one of the weaknesses of scoping reviews is the lack of a quality of reporting assessment [56], selected papers were evaluated using the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [57]. These guidelines provide a basis to determine quality of reporting of the studies, providing a minimal list of information to be included for a study to be interpretable and replicable. It is a checklist of items to determine if all the relevant metadata and study design information is reported within the manuscript, as previously described [58]. STROBE guidelines are gaining momentum with journals [59], and have previously been adopted to determine changes in the quality of reporting of the literature, using a pre-/post-test study design [60]. STROBE is designed for observational studies of which 18 (75%) of 24 papers met this criterion (in the case of experimental trials, these were reviewed regardless, and received 1 point for study design in the total checklist). Two reviewers scored each manuscript following STROBE checklist electronically (M.L.L.L., W.K.), and discrepancies were analyzed by a third reviewer (S.S.C). The remaining discrepancies were eliminated through discussions between two of the three reviewers both electronically and verbally (M.L.L.L., S.S.C.), where discrepancies were largely in study design, and sources of bias. Reviewers discussed each case of the remaining discrepancies and came to full agreement. A point was given for each positively evaluated checklist item. From this, the scoring was deemed neutral (16)(17)(18)(19) and positive (20)(21)(22)(23)(24), no studies were excluded based on their STROBE result.

STROBE
All initial discrepancies were resolved. STROBE results from two independent reviewers revealed total points of neutral (16)(17)(18)(19) and positive (20-24) scored manuscripts. All manuscripts met 12 of the 24 STROBE criteria. For seven of the STROBE criteria, fewer than half of the 24 papers reviewed herein met the criteria. These missing criteria included: addressing potential sources of bias, explaining how study size was arrived at, explaining how missing data were addressed, describing any sensitivity analyses, giving reasons for non-participation/sample exclusion at each stage, indicating the number of participants with missing data for each variable of interest, and translating estimates of relative risk into absolute risk for a meaningful time period. Additionally, fewer than 20 studies but more than 12 studies noted the following: how quantitative variables were handled in the analyses, how loss of samples were accounted for, numbers of individuals at each stage of the study, demographic/clinical or social characteristics of study participants, unadjusted estimates of risk, limitations of the study (including sources of potential bias or imprecision), as well as the source of funding and role of the funders in the study. Notably, few articles described how they quantified several of the variables included in their analyses. This directly impacted how the authors described their outcome data. Lastly, the point of generalizability was not always implicitly stated in the text but was assumed to be present when the text presented the greater clinical implications of the work. The weakest sections were: reporting of study size, sample size calculation, and explanation of missing data.

Metabolites
Summary tables were generated for the following biofluids: fecal samples (Table 1), plasma, serum and DBS (included together in Table 2), and urine (Table 4). In cases where manuscripts included both healthy, term-infants, as well as unhealthy and/or pre-term infants, only the healthy infants were discussed, and specific results listed. In manuscripts where more than one biofluid was described, the article is cited in multiple tables. A column was included to describe whether discriminatory markers between diets were found in the outcomes of the studies. This is listed as a separate column in the tables, as well as detailed in the following paragraphs for each biofluid. The use of targeted (n = 9) vs. un-targeted (n = 10) analytical approach did not reveal differences. Similarly, the techniques used to qualify metabolites was extremely heterogenous, whereby there was little repetition in any of those reporting for discriminatory markers: fecal (n = 6 total: GC/MS n = 1, LC/MS/MS n = 1, H 1 -NMR n = 2, MALDI-TOF-MS n = 1, HPLC n = 1, UHPLC-MS n = 1); plasma/serum/DBS (n = 5: LC/MS n = 2, H 1 -NMR n = 1, HPLC n = 1, HRMS n = 1); urine (n = 7: CE-TOF/MS n = 2, GC/MS n = 1, LC/MS n = 1, H 1 -NMR n = 1, HPLC n = 1, ELISA n = 1). From this we may state that H 1 -NMR was more frequently used for fecal metabolite detections, LC-MS for plasma/serum/DBS, and CE-TOF/MS for urine with twice as many times reported. Given the small sample size we are unable to speculate as to which technique is best for each biofluid.
Fecal metabolites were discussed in the following reports [35,37,39,40,44,45] and are summarized in Table 1. Methods for analyzing human fecal metabolites used in these studies employed analytical techniques that were previously validated, and these methods are reviewed in [61]. From review of these studies, age of the infant emerged as an important covariate to measure because metabolomic analyses of stool from infants of various ages are distinct [35,44]. One year of age is a distinct cutoff, where each individual generates a unique repertoire of fecal metabolites, whereas 3-6-month-old infants have similar fecal metabolites if fed the same diet [44]. None of the studies reviewed associated a specific dose of exposure to human milk or formula with discriminatory metabolites. Rather, fecal metabolites, in the reviewed studies, discriminated between any or no exposure to human milk. Discriminatory metabolites included human milk oligosaccharides themselves, which had traversed the intestinal tract intact, and short-chain fatty acids (SCFAs) synthesized by intestinal bacteria. In all studies which included both formula-and human milk-fed infants (n = 6 out of 6 such articles), fecal metabolites were discriminatory for dietary exposure.
The discriminatory ability of the articles was qualified as either yes (n = 5), no (n = 3), and largely based on the detection of lipids and amino acids. In summary, Acharjee et al. found that an exclusively human milk diet resulted in the presence of distinct lipids species (PC (35:2), SM (36:2), SM (39:1) [32]. Kirchberg determined that a total of 29 plasma metabolites differed between dietary groups, of which BCAA's were the most discriminant [43]. Slupsky et al. noted increased circulating plasma amino acids, creatinine and urea compared with human milk-fed infants within 2 h postprandial [48]. A similar observation was made with Socha et al. where essential serum amino acids, IGF-1, and urea increased significantly in both the LP and HP groups compared to human milk [51]. Finally, Uhl et al., control formula showed 40% (AA) and 51% (DHA) plasma levels as compared to human milk-fed infants [52].   Table 2. Plasma and serum were the biofluids for the following reports [41,43,48,[51][52][53]. Two articles discussed metabolites as revealed from dried-blood spot samples [32,46].     [34,38,40,42,44,[49][50][51]55,62] and the outcomes and trends are presented in Table 4. Scalabre et al. was finally excluded from further review due to a lack of explicit detailing in methods on dietary intake [47]. Table 3. Manuscripts reporting urine metabolites from human milk-and/or formula-fed infants (n = 10).  Yes, urinary C-peptide: creatinine ratio higher in HP, as compared to LP and human milk-fed. Yes, Choline metabolites (choline base solution, n,n-dimethylglycine, sarcosine, and betaine) and l(-)-threonine and l-carnosine excretion at 1 mos were statistically significantly higher in human milk-fed infants; 1(-)-threonine and 1-carnosine. Lactic acid was lower in human milk-fed infants at both 1 and 6 mos.

LC-MS Targeted
Yes, Urinary t-PGDM at 1 and 6 mos was significantly lower in breastfed infants than FF. In spite of the lack of a comparative group, other observations in this sub-cohort of studies looking at plasma, serum and DBS, from exclusively human milk-fed included those describing serum phospholipids, acylcarnitines, amino acids, and LPC 14:0 [41]. This finding was similar to that of Neto et al. which found long-chain acylcarnitines, palmitoylcarnitine, stearoylcarnitine, and oleolycarnitine increased by 27%, 12% and 109% in the first week of life, this article however, did not evaluate any additional time points [46]. Additional markers for the exclusively human milk-fed included higher levels of serum cholesterol, TG, ALT, AST, GGT, T-bil and D-bil levels were significantly higher in the human milk-fed group, at both 1 and 2 months [53].
Dietary-specific urinary markers were found in seven studies [38,44,[49][50][51]55,62]. Cesare Marincola et al. found age-dependent differences for choline, betaine, myoinositol, taurine, and citrate for three types of nutrition, and no differences between the two formula types [62]. In Dessi et al. it was determined at three days that infants fed formula milk had higher levels of glucose, galactose, glycine and myo-inositol; aconitic, aminomalonic, adipic acids were elevated in human milk-fed infants [38]. Martin et al. was listed in both the table for fecal metabolites and urine, which found that at 3-, 6-, and 12-month, formula-fed differed from breast-fed infants, based on both lipid profiles and energy metabolism (carnitines, ketone bodies, and Krebs cycle); this study was powered with a large sample size (n = 236) and many urine samples (n = 587) [44]. Socha et al. looked at infants (n = 636) at six months, where urinary C-peptide: creatinine ratio higher in HP, as compared to LP compared to human milk-fed [51]. Shoji et al. (2017) found that choline metabolites (choline base solution, N, N-dimethylglycine, sarcosine, and betaine) and l(-)-threonine and l-carnosine excretion at 1 month were statistically significantly higher in human milk-fed infants; 1(-)-threonine and 1-carnosine. Lactic acid was lower in human milk-fed infants at both 1 and 6 months [55]. Another study by the same group reported urinary t-PGDM at 1 and 6 mos was significantly lower in breastfed infants than FF [49]. A final study by the same group found that choline metabolites (choline, N, N-dimethylglycine, sarcosine, and betaine) differed between human milk-and term formula-fed [50].
One manuscript did not provide discriminatory results. Anderson et al. looked at exclusive human vs. exclusive formula milk in 175 infants and DCA (adipic, suberic, sebacic acids) excretion amounts did not differ between groups [34]. Two additional manuscripts did not have a comparative group [40,42].

Discussion
The current review of the literature describing metabolites in biofluids of healthy infants fed human milk, infant formula or a mixed diet demonstrates wide variability in metabolites but provides evidence that metabolites in fecal samples may be best suited to differentiating human milk-fed from formula-fed infants. Furthermore, it highlights the variety of methods used to profile these metabolites and the failure of much of the literature to report necessary details about study design, participants, and generalizability of reported results. It was beyond the scope of this review to determine if compositional differences within human milks or within formulas can be detected in biofluids from infants consuming those foods. However, one of the manuscripts that was reviewed compared formulas of differing composition (high-and low-protein formulations). That article did not identify differences in infant metabolites by formula composition. Another of the manuscripts (Dotz 2015 [40]) did identify fecal metabolite differences related to differences in the human milk oligosaccharide composition of the milk.
Based on the STROBE results and to report on the primary objective of the work, we can deduce that the overall quality of the articles is fair to positive. None of the articles were excluded based on their STROBE results and this internal metric should be considered for future reviews, as it is not customary for scoping reviews to screen articles using this metric.
This literature review also demonstrates that metabolites are measured using a variety of methods, including targeted (by MS only) or untargeted (by NMR or MS) approaches. Targeted approaches are those which set out to measure specific metabolites whereas untargeted approaches measure all metabolites [63]. Typically, targeted approaches allow for quantitative assessment of metabolite abundance whereas untargeted can only compare relative amounts within the study population. Untargeted approaches are often used because of their ability for discovery of new metabolites. Variability in methodologies was not differentiated, given the limited number of manuscripts using each technique that were included in the final analysis.
Blood, plasma, and DBS are attractive sources of metabolic differences. It is known that blood reveals metabolic changes reflective of organ function and is commonly used in clinical biochemistry, therefore there is ongoing interest to develop blood markers of interest for this patient population. The use of blood for patient analytics remains at the forefront of current bedside medicine, playing a major role in decision making, deductive reasoning for physicians and how we derive a diagnosis. It has evolved in the field as a common biofluid of interest [5].
There also is a desire in the pediatric community to determine novel ways to run studies using non-invasive practices, of which urine and fecal matter, both constituted as waste-products, have a certain appeal. However, urine has specific challenges [64][65][66]. For instance, urine must be considered over a 24-h period of time, as opposed to a single spot urine, for some analytes [67,68]; however, this can be impractical in infant populations [68]. Fecal samples may be an attractive specimen for collection, but this sample type requires that the patient be consuming a diet enterally. In the case of hospitalized patients, they often are not consuming an enteral diet and may be nil per os (NPO). In such cases, little fecal matter is produced, making it difficult to use this biofluid for metabolite analyses. Nonetheless, 100% (6/6) of articles describing fecal metabolites reported discriminatory metabolites between formula-and human milk-fed infants. These discriminatory metabolites included the human milk oligosaccharides themselves as well as short-chain fatty acids. Only 62.5% (5/8) of plasma/serum/dried blood spot and 88% (7/8) of urine reports had such discriminating metabolites. Thus, when the goal is to discriminate between human milk-and formula-fed infants, feces may be the preferred biospecimen.

Limitations
This review excludes articles published after 31 December 2018. Findings cannot be used to recommend policy/practice. In many cases biomarkers still need to be validated in a separate cohort. Given the limited number of manuscripts, we were not able to further stratify based on technique, which may introduce additional heterogeneity. Studies may not have been adequately powered. We did not differentiate here between human milk-fed to infants at the breast or human milk given to infants via bottle feeding.

Conclusions
There are many articles that characterize dietary metabolites in infants, as described here (n = 24). Most are of high quality, but some do not clearly describe the populations from which samples were collected nor the methods used to handle missing data/samples. These articles reported differentiating between human milk-and formula-fed infants based on metabolites in their urine, plasma, serum, blood or feces. Although the studies identified several metabolites that differ between human milkand formula-fed infants, only some metabolites were reported in more than one study. These are summarized here. Human milk oligosaccharides and short-chain fatty acids were discriminatory in fecal samples. In serum/plasma, amino acids and urea were discriminatory. For urine, myoinositol was discriminatory. It is beyond the scope of this review to state whether sufficient evidence exists; however, if dietary markers are of interest, then the studies listed in this review may serve as reference body of literature. Further work is required. In healthy infants, fecal samples provide the best discriminatory prospect for dietary metabolite differentiation.
Supplementary Materials: The following are available online at http://www.mdpi.com/2072-6643/12/7/2073/s1, Table S1: Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) Checklist. Table S2: Participants: interventions, comparisons, outcomes and study design (PICOS) used to systematically review relationship between infant diet and the range of metabolites present in infant biofluids. Table S3