Identification of Oligosaccharides in Feces of Breast-fed Infants and Their Correlation with the Gut Microbial Community *

Glycans in breast milk are abundant and found as either free oligosaccharides or conjugated to proteins and lipids. Free human milk oligosaccharides (HMOs) function as prebiotics by stimulating the growth of beneficial bacteria while preventing the binding of harmful bacteria to intestinal epithelial cells. Bacteria have adapted to the glycan-rich environment of the gut by developing enzymes that catabolize glycans. The decrease in HMOs and the increase in glycan digestion products give indications of the active enzymes in the microbial population. In this study, we quantitated the disappearance of intact HMOs and characterized the glycan digestion products in the gut that are produced by the action of microbial enzymes on HMOs and glycoconjugates from breast milk. Oligosaccharides from fecal samples of exclusively breast-fed infants were extracted and profiled using nanoLC-MS. Intact HMOs were found in the fecal samples, additionally, other oligosaccharides were found corresponding to degraded HMOs and non-HMO based compounds. The latter compounds were fragments of N-glycans released through the cleavage of the linkage to the asparagine residue and through cleavage of the chitobiose core of the N-glycan. Marker gene sequencing of the fecal samples revealed bifidobacteria as the dominant inhabitants of the infant gastrointestinal tracts. A glycosidase from Bifidobacterium longum subsp. longum was then expressed to digest HMOs in vitro, which showed that the digested oligosaccharides in feces corresponded to the action of glycosidases on HMOs. Similar expression of endoglycosidases also showed that N-glycans were released by bacterial enzymes. Although bifidobacteria may dominate the gut, it is possible that specific minority species are also responsible for the major products observed in feces. Nonetheless, the enzymatic activity correlated well with the known glycosidases in the respective bacteria, suggesting a direct relationship between microbial abundances and catabolic activity.

The role of milk in developing the gastrointestinal microbiome appears to be critical in infant development. HMOs are not catabolized directly by the infant, but instead pass undigested through the upper digestive tract to the large intestine where they function as prebiotics for the distal gut microbiota (4, 14 -16). HMOs help promote a healthy infant gastrointestinal tract in a number of ways: as anti-adhesives preventing viruses from binding to epithelial cell surface glycans, as antimicrobials acting as decoy receptors to pathogenic bacteria, and as energy sources for beneficial bacteria which in turn produce short-chain fatty acids for the infant (4,6,17,18). Commensal bacteria such as Bifidobacterium and Bacteroides are equipped with a suite of enzymes capable of breaking down HMOs (19 -21). For this reason, bifidobacteria can make up to 90% of the microbial community in the gut of breast-fed infants (5,22). Select bifidobacterial species grow well on HMOs in vitro (21,23). Bifidobacterium longum subsp. infantis has been shown to be particularly good at utilizing HMOs, displaying preferential consumption of smaller and fucosylated HMOs (21,23,24). Different microbial species and strains in the infant gastrointestinal tract have their own mechanisms for HMO catabolism leading to potentially diverse digested products in the feces. These digestion products are generally unknown, but characterizing these glycan species may reveal information regarding the repertoire of specific bacterial enzymes at work and indicate the active members in the communities inhabiting the infant gastrointestinal tract.
The in vivo catabolism of HMOs can provide important phenotypic information regarding the state of the gut microbiota. Along with the degraded HMO, intact HMOs have also been discovered in the feces of breast-fed infants (15,16,25,26). Computational, genomic, and in vitro methods have been combined to study bacterial consumption of HMOs and other carbohydrates. Eilam et al. created a computational program using genomic data of microbial enzymes in order to predict degraded glycan products (27). Asakuma et al. grew several bifidobacteria strains on select HMO-enriched media and discovered various mechanisms for producing digestion products from the in vitro bacterial growth (28). Further studies found important distinctions for the feeding habits among specific bacterial species. B. longum subsp. infantis catabolizes the entire HMO internally, whereas B. bifidum secretes extracellular glycosidases for cleaving HMOs, leading to digested HMO products in the intestinal lumen (29,30). These studies show that degraded glycans are created through the activity of key glycosyl hydrolases and can link relevant glycosidases with specific bacteria, thus providing insight into possible gut microbiota degradation pathways witnessed in feces.
In this study, we extracted and characterized intact oligosaccharides and digestion products from the feces of breastfed infants to obtain the most comprehensive oligosaccharide analysis of feces. We compare the fecal oligosaccharides with matched maternal breast milk, and gained insight into the active glycosidases in the intestinal microbiome. The fecal glycans act as a phenotypic signature indicative of the enzymes that are active in the microbial communities in the infant gut. Genomic profiling of the types of microbial species present in infants' guts was used to infer those glycosidases. Intact and degraded oligosaccharides from milk glycoconjugates were identified and a fecal glycan library was created for the rapid and comprehensive analysis of postpartum alterations to the fecal glycome.

EXPERIMENTAL PROCEDURES
Breast Milk and Infant Fecal Samples-Milk samples were obtained from healthy women enrolled in the Foods for Health Institute Lactation Study at the University of California, Davis. Sample collection and subject details have been previously reported (31,32). The UC Davis Institutional Review Board approved all aspects of the study, and informed consent was obtained from all subjects. This trial was registered on clinicaltrials.gov (ClinicalTrials.gov Identifier: NCT01817127). For this study, a subset of one mother/infant pair and six infant fecal samples were analyzed. Briefly, milk samples were collected in the morning on day 6, 21/24, 71, and/or 120 postpartum. Subjects fully pumped one breast into a bottle, inverted six times, transferred 12 ml into a 15 ml polypropylene tube, and subsequently froze the sample in their kitchen freezers (Ϫ20°C). Samples were collected from the subjects' freezers by study personnel biweekly, transported to the lab on dry ice, and stored at Ϫ80°C until processing. Infant fecal samples were collected at 6/7, 21/24, 71, and/or 120 days of life. Parents transferred their infant fecal samples into sterile plastic tubes and were instructed to immediately store the samples at Ϫ20°C until transported by study personnel. Fecal samples were transported to the laboratory on ice packs and stored at Ϫ80°C before processing.
Oligosaccharide Extraction-Free oligosaccharides were extracted from breast milk and infant feces following previously reported methods, with slight modifications (8,26,33). Fecal samples were weighed, diluted with nanopure water to 1 mg/10 l, and homogenized in the shaker overnight at 4°C. The samples were then centrifuged at 4000 ϫ g for 30 min, and 100 l of supernatant was taken and placed on a 96-well plate. For the milk samples, 50 l were aliquoted on a 96-well plate and defatted via centrifugation. The skimmed milk and fecal supernatants then underwent similar methods. Proteins were removed by ethanol precipitation at Ϫ80°C for 1.5 h, and after 30 min of centrifugation the supernatant was evaporated to dryness. The resulting glycans were reduced to their alditol form with 1.0 M NaBH 4 at 65°C for 1.5 h. Desalting and purification was achieved using solid phase extraction on both C-8 and graphitized carbon cartridges (GCCs) (Glygen Corp, Columbia, MD) for the fecal samples but only GCCs for milk. After the glycans were loaded onto preconditioned C8 cartridges, the flow-through was collected and the glycans were loaded onto preconditioned GCCs. The purified glycans were eluted with 20% acetonitrile (ACN)/water (v/v) and 40% ACN/water (v/v) in 0.05% trifluoroacetic acid, eluent solvent then evaporated to dryness, and reconstituted and diluted in nanopure water before analysis.

Nano-High Performance Liquid Chromatography-Chip/Time-of-Flight Mass Spectrometry (nanoLC-MS)-Milk and fecal glycans were
analyzed on a nano-high performance liquid chromatography (HPLC)chip/time-of-flight (TOF) mass spectrometry system. The HPLC system used was an Agilent 1200 series unit with a microfluidic chip, which was coupled to an Agilent 6220 series TOF mass spectrometer via chip cube interface. The capillary pump on the chromatography unit loads the sample onto the 40 nL enrichment column of the chip at a flow rate of 4.0 l/min with a 1 l injection volume. The nano pump is used for analyte separation on the analytical column of the chip, which is 75 l x 43 mm and packed with porous graphitized carbon. Separation is accomplished using a binary gradient of aqueous solvent A (3% acetonitrile (ACN)/water (v/v) in 0.1% formic acid (FA)) and organic solvent B (90% ACN/water (v/v) in 0.1% FA) with a previously developed method for HMO separation (7,8). The sample is then introduced into the TOF mass spectrometer via electrospray ionization, which was tuned and calibrated using a dual nebulizer electrospray source with calibrant ions ranging from m/z 118.086 to 2721.895, and data was collected in the positive mode.
Nano-HPLC-Chip/Quadrupole-TOF MS-Structural elucidation of fecal glycans was performed on a nano-HPLC-chip/quadrupole-TOF (Q-TOF) mass spectrometer, Agilent LC 1200 series and Q-TOF 6520 series. The chromatographic separation method on the Q-TOF was the same as that for the nano-HPLC-chip/TOF. Tandem mass spectrometry was also performed in the positive mode with collision induced dissociation for fragmentation. The following voltages were optimized for oligosaccharide fragmentation: fragmentor 175 V, skimmer 60 V, and octopole 1 RF 750 V, and nitrogen drying gas was used at a flow rate of 5 L/min at 325°C. Auto MS/MS was used with 0.63 spectra/second for both MS 1 and MS/MS. Precursor ions were selected based on abundance, with doubly charged ions given first priority, singly charged ions second, triply charged third, then last, other multiply charged ions. All calibrant ions were excluded from precursor selection with a 4 m/z isolation window. Collision energy (CE) was based on the ion's mass-to-charge (m/z) ratio with higher energy for larger ions, according to the following equation where 1.3 is the slope and Ϫ3.5 is the y-intercept. The equation was empirically determined by the manufacturer (7). Oligosaccharide Data Analysis-Data was collected using Agilent MassHunter Work station Data Acquisition version B.02.01 on the nanoHPLC-chip/TOF and version B.05.00 on the nanoHPLC-chip/ Q-TOF and then analyzed with Agilent MassHunter Qualitative Analysis software versions B.03.01 and B.06.00. Glycans were identified using the Find Compounds by Molecular Feature function to within 20 ppm of theoretical masses. Compound abundances were extracted as volume in ion counts, which was directly correlated to absolute abundances of the compounds present in each sample. Known structures were identified by matching accurate mass and retention time to previously developed annotated HMO libraries which contains 102 composition entries (7,8). Peak alignment for correcting for retention time shifts was performed using in-house software (33). Tandem MS data were then annotated by hand and in-house software. The MS data have been deposited to the Proteome Xchange Consortium via the PRIDE partner repository with the data set identifier PXD004434 (34).
Cloning and Protein Expression of BLNG_00015 from B. longum subsp. longum SC596 -BLNG_00015 (␤-galactosidase) was cloned using the pEco-T7-cHis cloning kit (Gentarget Inc., San Diego, CA). The Qiagen DNeasy Blood & Tissue Kit was used to purify genomic DNA from B.longum subsp. longum SC596 culture following the manufacturer's guidelines for purification of Gram-positive bacterial DNA. BLNG_00015 was amplified by PCR. 225 l of PCR product was gel purified and cloned into a pEco-T7-cHis (Gentarget Inc) vector. Sequencing of the insert of the recombinant BL21 Star clone was carried out to confirm the sequence of the entire glycosyl hydrolase insert. Protein expression was carried out using 700 ml of LB with 100 g/ml carbenicillin. Cells were grown until an OD of 0.5 was reached and induced overnight using a final concentration of 0.5 mM isopropyl-␤-d-thiogalactopyranoside. The culture was centrifuged using an Eppendorf 5804 centrifuge (Hauppauge, NY) for 20 min at 4°C and frozen at Ϫ80°C. Bugbuster protein extraction reagent (EMD chemicals, Gibbstown, NJ) was used to resuspend culture. 60 l of DNase I (Roche, Basel, Switzerland) from a 10 U/l stock and 120 l of lysozyme from a 50 mg/ml stock were added per 20 ml of Bugbuster suspension. After a 5 min incubation, the Bugbuster suspension was then centrifuged for 30 min at 13,200 rpm at 4°C. For purification a 1 ml Bio-Scale Mini Profinity immobilized-metal affinity chromatography column attached to a EP-1 Econo Pump (Bio-Rad, Hercules, CA) was used. Following purification, imidazole was exchanged with PBS using 10 kDa cutoff Amicon Ultra-15 centrifugal unit (Millipore, Billerica, CA). Purity and size of the recombinant protein were confirmed using a 4 -15% SDS-PAGE gel.
In vitro Enzyme Digestion-Expressed enzymes were stored at Ϫ80°C until analysis. A pool of HMOs (1.5 l, 5 mg/ml) reduced to the alditol was combined with 3 l of enzyme BLNG_00015 and 5.5 l of 0.1 M ammonium acetate buffer at pH 6.0. The solution was incubated for 1 h at 55°C followed by C18 zip-tip (Agilent Technologies, Santa Clara, CA) clean up. The samples were then dried and reconstituted in 50 l nanopure water for analysis. This experiment was performed in triplicate. An undigested HMO pool (1.5 l in 50 l nanopure water) was run with the digested pool on the nano-HPLC-chip/TOF mass spectrometer.
Fecal DNA Extraction-DNA was extracted from 150 mg of stool sample using the ZR Fecal DNA MiniPrep kit (ZYMO, Irvine, CA) in accordance with the manufacturer's instructions, which included a bead-beating step using a FastPrep-24 Instrument (MP Biomedicals, Santa Ana, CA) for 2 min at 25°C at a speed of 6.5 m/s. Bifidobacterium-specific Terminal Restriction Fragment Length Polymorphism (Bif-TRFLP)-The method of Lewis et al. was used to perform the Bifidobacterium-specific terminal restriction fragment length polymorphism assay (35). Briefly, DNA from feces was amplified in triplicate by PCR using primers NBIF389 (5Ј-[HEX]-GCCTTCGGGTTGTAAAC) and NBIF1018 REV (GACCATGCACCAC-CTGTG). DNA was purified using the Qiagen QIAquick PCR purification kit and then cut with restriction enzymes AluI and HaeIII. The resulting fragments were analyzed on an ABI 3100 genetic analyzer, and sizes were compared against the published database for species identification.
Bifidobacterium longum/infantis Ratio (BLIR)-A PCR-based assay, BLIR, was used to determine which subspecies of B. longum were present in each sample and to gain an estimate of their relative abundance to each other (36). Briefly, PCR was performed using three primers (FWD_BL_BI (5Ј-[HEX]-AAAACGTCCATCCATCACA), REV_BL (5Ј-ACGACCAGGTTCCACTTGAT), and REV_BI (5Ј-CGCCT-CAGTTCTTTAATGT)) targeting a conserved portion of the genome (between Blon_0424 and Blon_0425) shared by both subspecies, but generating different amplicon lengths for each. Amplicons were analyzed by capillary electrophoresis on an ABI 3100 genetic analyzer (Applied Biosystems, Carlsbad, CA) and interpreted with PeakScanner 2.0 software (Applied Biosystems) to estimate relative abundances of each subspecies.
Marker Gene Sequencing-DNA samples were prepared for 16S rRNA marker gene sequencing as previously described (31,37) with the following modifications. Universal barcoded primers with Illumina sequencing adapters (Illumina, San Diego, CA) (adapters are italicized and the barcode is highlighted in bold) V4F (5Ј-AATGATACGGCGA-CCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCT NNNNNNNNGTGTGCCAGCMGCCGCGGTAA-3Ј) and V4Rev (5Ј-CAAGCAGAAGACGGCATACGAGATCGGTCTCGGCATTCCTGCT-GAACCGCTCTTCCGATCTCCGGACTACHVGGGTWTCTAAT-3Ј) were used to PCR amplify the V4 region of the 16S rRNA gene in each sample (37). A pooled library was sequenced at the University of California DNA Technologies Core Facility on an Illumina MiSeq platform (Illumina) (150 bp single read).
Sequence Analysis-The QIIME software package (version 1.7.0) was used to analyze the results of the Illumina sequencing run. Illumina V4 16S rRNA gene sequences (Illumina) were demultiplexed and quality filtered using the QIIME 1.7.0 software package with default settings unless otherwise specified (38). Reads were truncated after a maximum number of three consecutive low quality scores. The minimum number of consecutive high-quality base calls to include a read (per single end read) as a fraction of the input read length was 0.75. The minimum acceptable Phred quality score was set at 20. Similar sequences were clustered into operational taxonomic units (OTUs) using open reference OTU picking with UCLUST software (39). Taxonomy was assigned to each OTU with the Ribosomal Database Project (RDP) classifier (40) and the RDP taxonomic nomenclature (41). OTU representatives were aligned against the Greengenes core set (42) with PyNAST software (43).

Glycomic Profile of Feces from Breast-fed Infants-The
considerations of potential sources of oligosaccharides in feces include HMOs, digested HMOs, and the digestion of other milk glycoconjugates including glycoproteins and glycolipids ( Fig. 1). Fig. 2A shows the extracted compound chromatogram (ECC) of intact HMOs from a breast milk sample from one mother 24 days postpartum, and the ECC of a fecal sample from her offspring 24 days postnatal (mother/infant pair 1028) (Fig. 2B). The direct comparison of the nanoLC-MS glycan profile of the milk to that of infant feces yielded considerable overlap because of undigested HMO in feces. Nonetheless, even a casual inspection showed significant differences (Fig. 2). There were peaks in feces that appeared and peaks that disappeared compared with the milk sample. For example, peaks from 25-26 min found in the milk sample were diminished in the fecal sample, whereas a peak at 8.5 min appeared in the fecal chromatogram that was absent in the milk (Fig. 2).
To confirm the identity of the intact HMOs, a previously optimized method was used to identify structures in human milk through the use of accurate masses, tandem MS, reproducible retention times, and an annotated structure library (7,8). For example, the peak at 10.3 min ( Fig. 2A) with m/z 491.18 corresponds to 2Ј-fucosyllactose (2ЈFL -structure inset). The same peak was found in feces at 10 Table I, with a complete list of intact HMOs found in milk and feces in supplemental Table S1. In general, we found that many of the species transit through the gut, albeit in depleted amounts at least early in the lactation period (day 24).
The HMO profile of feces displays preferential consumption of select glycan species. For example, fewer sialylated compounds were identified in feces that eluted from 25 to 26 min, compared with the milk (Fig. 2). There are 30% fewer sialylated HMO structures in feces compared with milk (supplemental Table S1). However, the largest drop in the number of HMO compounds was for the undecorated structures (containing neither Fuc nor Neu5Ac), which decreased by 60% in feces (supplemental Table S1).
Determination of Unidentified HMO and Non-HMO Species-The non-HMO compounds present in the chromatogram were further probed to determine if they were indeed oligosaccharides. To determine these new structures, their compositions were first determined by accurate masses and confirmed by tandem MS. For example, a compound in Fig.  2B with retention time 13.6 min was found to have a neutral molecular mass of 547.21 Da, corresponding to a composition of 2Hex:1HexNAc. The tandem MS spectrum confirmed the composition (supplemental Fig. S1). Two peaks with m/z 204.09 and 366. 16 were observed corresponding to HexNAc and 1Hex:1HexNAc, respectively. Also observed was a neutral loss of 162.06, because of the loss of one Hex, and 182.07, a Hex reduced to the alditol. Reduction of the glycans is incorporated in the extraction method to eliminate peak splitting during chromatographic separation because of the presence of anomers and to identify the reducing end saccharide in the tandem MS.
Another example is illustrated with the compound found at 17.2 min (Fig. 2B). The MS yields a neutral molecular mass of 693.26 Da, which corresponds to the composition 2Hex:1HexNAc:1Fuc. The tandem MS spectrum is consistent with an oligosaccharide but was not consistent with an HMO as there were no fragments indicative of the lactose (Gal(␤1-4)Glc) core, but a neutral loss of 182.08 (a reduced Hex) was present (supplemental Fig. S2). The results of these analyses are summarized in Table II with a list of the compound compositions and their respective retention times. Also listed are the potential sources of the oligosaccharides, which are either HMOs or glycoproteins as discussed below.
Oligosaccharides are not all derived from HMOs. There are structures that are consistent with N-glycans based on their fragmentation patterns. Fig. 3A shows the tandem MS spectrum of a doubly charged glycan with m/z 631.74, which corresponds to the composition 3Hex:3HexNAc:1Fuc. The loss of 111.53 correlates to a doubly charged reduced Hex-NAc, and the singly charged reduced HexNAc can be observed at m/z 224.10, both indicative of a reduced GlcNAc from an N-glycan core (Fig. 3A). The fragmentation pattern further suggests a fucosylated species with losses corresponding to Hex and Fuc (Fig. 3A).
Tandem MS was used to differentiate between oligosaccharides derived from either HMOs or N-glycans. One example is a compound with composition 5Hex:3HexNAc at m/z 1440.52, corresponding either to an HMO or a hybrid-type N-glycan. The neutral loss of 344.11 in the tandem MS spec-trum indicates a reduced lactose core of an HMO (Fig. 3B). The repeated neutral loss of 365.14 was also suggestive of the repeating lactosamine (Gal(␤1-3/4)GlcNAc) units that make up the backbone of HMOs. The lack of m/z 224.11, which would have indicated a reduced GlcNAc from the chitobiose N-glycan core, was another indication that this oligosaccharide was an HMO and not an N-glycan. Fragmentation of an antenna or the trimannosyl portion of the N-glycan core would lead to subsequent neutral losses of 162.05, but neither were present.
A more difficult example is the compound with composition 3Hex:2HexNAc (m/z 913. 35), which appears to be an N-glycan core. The fragmentation showed neutral losses of 344.12, which corresponds to a reduced lactose core of an HMO (Fig.  3C). Also observed were neutral losses of 182.05, the mass of a reduced Hex, which corresponds to the reduced Glc of the lactose core (Fig. 3C). Neutral losses of 223.10 or a peak at m/z 224.11 would have been indicative of a reduced HexNAc and evidence of the N-glycan core, however these were not observed. Because these diagnostic markers were not found in the MS spectra, composition 3Hex:2HexNAc was assigned as a degraded HMO (Table II).
There are compositions whose origins could not be differentiated between HMOs and N-glycans, even when their frag-  mentation patterns are known. This is because of their low abundances, which made tandem MS difficult to obtain. The masses of these overlapping compositions were added to the Fecal Glycan Library, with both oligosaccharide sources listed (Table II). Selection and Expression of Microbial Enzymes-Previous analysis on the feces of these infants showed that the infants were dominated by types of bifidobacteria that are capable of catabolyzing HMOs (supplemental Fig. S3) (24,31). For this reason, glycosidases from bifidobacterial strains were expressed to determine whether the enzymes can act on HMOs and produce the structures present in the infant feces. For this work, we selected an exogalactosidase and endoglycosidase from bifidobacteria to determine whether these acted on milk glycoconjugates. An example is illustrated in Fig. 4 with a ␤-galactosidase from B. longum subsp. longum SC596, BLNG_00015. For these experiments, pooled HMOs were treated with the enzyme (a galactosidase, pH 6.0 at 55°C for 1 h). The abundances of HMOs in the digested pool were compared with those in an undigested HMO pool. The chromatograms in Fig. 4A show digestion of select structures with m/z 611.24 (4Hex:2Hex-NAc:1Fuc). Two isomers -monofucosyllacto-N-hexaose I (MFLNH I, retention time 14.5 min, inset structure 4) and isomer 3 fucosyl-paralacto-N-hexaose (IFLNH III, retention time 15.2 min, inset structure 5) -were noticeably lower in abundances in the digested HMO pool compared with the undigested HMO pool (Fig. 4A). Isomer 1 fucosyl-paralacto-N-hexaose (IFLNH I, retention time 18.3 min, inset structure 6), however, has no terminal Gal residue and therefore the exogalactosidase was unable to digest this structure as shown by the overlapping chromatograms (Fig. 4A). The exogalactosidase was also able to degrade select undecorated intact HMOs with m/z 1075.41 (4Hex:2HexNAc) (Fig.  4B). Lacto-N-hexaose (LNH, retention time 15.6 min, inset structure 7) and lacto-N-neohexaose (LNnH, retention time 16.2 min, inset structure 8) decreased in abundance, but para-lacto-N-hexaose (p-LNH, retention time 19.5 min, inset structure 9) went undigested (Fig. 4B).
After analyzing intact HMO digestion, we turned our attention to the digestion products that were formed by the exogalactosidase activity. If LNH and LNnH (4Hex:2HexNAc) were to lose a terminal Gal residue the digestion products would have the composition 3Hex:2HexNAc. This composition was seen in high abundance in the digested sample (Fig.  4C). Further digestion would lead to the loss of another terminal Gal and composition 2Hex:2HexNAc (Fig. 4D). The abundance in the undigested pool was at baseline for these  digestion products, revealing that they are not found as intact HMO structures in the pooled breast milk but were only observed after performing enzymatic digestion reactions. These new digested HMO products were added to the Fecal Glycan Library (supplemental Table S2).
An endoglycosidase from B. longum subsp. infantis was previously expressed to determine its activity. Instead of degrading HMOs, this enzyme (EndoBI-1) was active on Nglycans, demonstrating preferential cleavage between the two GlcNAcs of the chitobiose core (44). These types of degraded N-glycans were observed in feces. For example, the m/z 1034.37 peak detected in feces corresponds to an oligosaccharide with composition 5Hex:1HexNAc (fragmentation shown in Fig. 5). This composition is not consistent with an intact or digested HMO for several reasons. Large HMO structures have repeating lactosamine units, and this compound only has one HexNAc. The tandem MS yields a peak at m/z 224.11, consistent with a reduced HexNAc, which is a rare occurrence in HMOs (Fig. 5). All the neutral mass losses that led up to that peak were 162, indicative of Hex losses, most likely mannose residues (Fig. 5). Because of the reduced GlcNAc and the five successive Hex, we deduced that this oligosaccharide is indeed a high mannose type N-glycan cleaved by an endoglycosidase. Both intact and digested N-glycans, their masses, and corresponding compositions were also added to the Fecal Glycan Library (supplemental Table S2).
Observation of Infant Fecal Glycome and Microbiome Profile Changes Postnatal-The compilation of fecal glycans (the Fecal Glycan Library) with their retention times and accurate masses were used to determine the fecal glycome profiles of infants over lactation. Fecal samples for three of the infants from the Davis Lactation Study were analyzed at several postnatal time points. For all three infants, there was a drastic decrease in the abundance of intact HMO from the first week of life (days 6 and 7) to week 17 (day 120) (Fig. 6). The absolute abundance dropped 75.3% on average, also observed was an opposite trend for digested HMO over lactation, which increased 124.0% at week 17 (Fig. 6). The bacterial profiling revealed that bifidobacteria, known to consume HMOs comprised on average 68.0% of all bacteria in the infant gut at week 17 (Supplemental Fig. 3) (20,21). For two of the infants, (1004 and 1054) the abundance of digested N-glycans also increased from week 1 to week 17 (Fig. 6). B. longum subsp. longum comprised 9.8% of all bacteria in the infant 1054 fecal microbiota at week 17 (supplemental Table S3). Schell et al. showed that B. longum contains a genomic cluster with an endoglycosidase for cleaving GlcNAcs which explains our observance of digested N-glycans (45). B. longum subsp. infantis, known to catabolize the entire HMO internally, was found to comprise 6.7% and 4.2% of all bacteria in infant 1040 at weeks 3 and 17, respectively (supplemental Fig. S3) (29,30). This correlated with lower amounts of digested HMOs at those time points (Fig. 6). This trend is consistent with the gut microbiota becoming established, niches being filled, and more of the oligosaccharides being degraded later in lactation. DISCUSSION Human breast milk contains both free and bound oligosaccharides. As the infant consumes breast milk, the contents pass through the digestive tract through the intestine, where they encounter the gut microbiota. The bacteria are equipped with glycosidases that cleave HMOs and other glycoconjugates, potentially leaving digestion products and free oligosaccharides to pass through the gut and exit as feces. Glycoconjugates such as N-glycans may be degraded while still linked to proteins, but they can also be enzymatically released from glycoproteins to form new free oligosaccharides. After their release, free N-glycans can be further degraded by exoglycosidases to form additional digestion products. Our oligosaccharide enrichment method may also extract O-glycans. In vitro studies have shown bacterial growth on, binding to, and degradation of mucus O-glycans, yet no glycosidases have been characterized that can release large O-glycans (larger than disaccharides) from glycoproteins (46 -49). Furthermore, O-glycans are in much lower abundances in milk compared with HMOs and N-glycans and are therefore not expected to be present in appreciable abundances (50,51). Free oligosaccharides may also be released from glycolipids (52), yet much like O-glycans, glycolipids are far less abundant compared with HMOs and N-glycosylated proteins and are not likely to be observed. In order to identify structures that are digestion products of HMOs, enzymatic in vitro experiments were performed. One involved the use of an exogalactosidase (BLNG_00015) from B. longum subsp. longum SC596, as this species was found in high abundance in the infant fecal samples. Digestion of intact HMOs showed that the enzyme was capable of degrading structures with a terminal galactose residue, but more interesting were the structures not degraded. The lack of digestion for p-LNH is explained by the preference for ␤(1-4)Gal linkages of this specific enzyme. Both LNH and LNnH have at least one terminal ␤(1-4)Gal so those isomers were digested, but because p-LNH only has a terminal ␤(1-3)Gal linkage, the hy-drolase was unable to cleave the linkage leaving the structure intact. This specificity was also observed in the fucosylated structures MFLNH I and IFLNH III, which both contain a terminal ␤(1-4)Gal residue so the exogalactosidase was capable of degrading those structures. Having a Fuc present on the same GlcNAc as a ␤(1-4)Gal appeared to hinder the enzyme from hydrolyzing that linkage. This type of analysis reveals the specificity of microbial enzymes. Observance of certain isomers, along with the absence of others, can indicate the preferences of those enzymes present. These experiments also confirmed that degraded oligosaccharides obtained in infant feces can be produced under in vitro conditions with the appropriate enzymes.
It was not surprising to find intact N-glycans in the infant feces at all sampled time points, as these molecules have been described in breast milk. The observation of degraded N-glycans in vivo in infant feces is, however, a novel finding. Degradation of N-glycans has been predicted before by enzymatic studies of individual microbes. For example, Bacteroides is known to inhabit breast-fed infant digestive tracts, was found in the feces of the infants in this study, and is known to have an enzyme capable of cleaving intact N-glycans from glycoproteins (19,31,53). Another enzyme, Peptide:N-glycosidase F (PNGase F), is known to cleave N-glycans at the glycosylamine linkage, but this enzyme was discovered in the pathogenic bacterium Flavobacterium meningosepticum (54,55). Experiments were not performed in FIG. 5. Oligosaccharide extracted in infant feces was determined to be a degraded N-glycan based on tandem mass spectra. Monosaccharides represented as mannose (•) and N-acetylglucosamine (GlcNAc) (f). Precursor ion with m/z 1034.37 shows neutral subsequent losses of hexose residues consistent with a high mannose type N-glycan. Only one GlcNAc present in the core is indicative of endoglycosidase activity on the glycoprotein this structure was derived from. this study using this specific enzyme because of its widely established activity on glycoproteins. Although F. meningosepticum was not specifically detected in the feces of the infants, it is a potential glycoprotein degrader and contributor to intact N-glycans in feces. Other N-glycans observed were hydrolyzed at the chitobiose core. We previously discovered an endoglycosidase, EndoBI-1, in B. longum subsp. infantis that was able to release all types of N-glycans from lactoferrin and immunoglobulins A and G by cleaving the oligosaccharide between the two core GlcNAcs, which is consistent with the product found in feces and may explain other similar digested N-glycan products (44). B. longum was found in all infants (B. longum subsp. infantis and B. longum subsp. longum), and one strain of B. longum was also found to contain genes for an endoglycosidase (45). Endoglycosidases have also been found in other bifidobacteria species, along with Streptomyces plicatus, and pathogenic Enterococcus faecalis, Streptococcus pyogenes, and Streptococcus pneumoniae (44, 56 -59). These studies employed genetic analysis and in vitro experiments to determine the enzyme's activity, but did not test their in vivo implications.
Fecal glycan profile changes were observed for infants postnatally. There was an initial increase in total oligosaccharide abundance for infants 1004 and 1054, but not for infant 1040. This initial increase with a later decrease has been observed previously (60), and has been attributed to the initial lack of saccharolytic bacteria. The HMOs accumulate until the establishment of HMO-consuming microbiota in the infant's gut. This process of maturation occurs at different rates for infants, however, the general trend is that the HMOs decrease. There is a drastic decrease in intact HMOs from week 1 to week 17 postpartum for all three infants, as well as an increase in digested HMOs. It has been shown that the abundance of HMOs in breast milk decrease over the course of lactation, which could explain the diminished abundance of observed intact HMOs in infant feces (61). However, infants consume more volume and feed more frequently as they age before solid foods are introduced. Although the mother may be producing milk with a lower concentration of oligosaccharides, the infant could still be consuming the same amount of HMOs because of the increased milk intake. We also observed an increase in bifidobacterial abundance in the infants' gastrointestinal tracts, and bifidobacteria are known to consume HMOs (20,21). This suggests that the decrease in intact HMOs and increase in digested HMOs is because of bacterial metabolism. An interesting correlation was observed, however, in infant 1040, whose feces contained decreased amounts of digested HMOs during weeks 3 and 17, but also a presence of B. longum subsp. infantis at those same time points. B. longum subsp. infantis is unique among saccharolytic bacteria in that it imports the entire HMO structure for complete internal catabolism. This process could explain the low amounts of HMO digestion products in feces during later weeks (29,30). Even though the microbiota community becomes dominated by bifidobacteria over the postnatal period leading to fewer intact HMOs and more digested HMOs in infant feces, it appears that specific strains can have an impact as to the types of fecal oligosaccharides observed. Analysis of fecal oligosaccharides has the potential of becoming a noninvasive method of assaying biomarkers likely relevant to the health of an infant (62).
Analyzing which milk glycans remain intact in the feces of an infant can help provide insight into the types of bacteria present in the infant gut microbiome and the functions they provide. We observed HMO and bacterial interaction by directly comparing a mother's HMO composition to the oligosaccharide profile of her infant's feces. Many of the HMOs passed through intact in the feces, but in significantly depleted amounts. The two most abundant structures in breast milk, LNT and LNnT, were completely lacking from the feces, showing preferential consumption of those structures. Some HMOs were found in larger abundance in feces compared with milk, would could be attributed to large HMOs being digested to form smaller HMOs, adding to their abundance. These types of observations can allude to the interactions between microbiota and oligosaccharides. Large-scale functional characterization of the different types of bacteria in the infant's gastrointestinal tract is typically performed via metagenomic, proteomic, or metabolomic experiments. Here, we propose that fecal glycans are suggestive of the bacterial enzymes present in an infant's gut. This information allows greater understanding of how bacteria interact with HMOs and other glycoconjugates in the gut. Interestingly, a small number of select enzymes can account for the vast majority of the products, although the glycosidases may come from different strains and still represent different catabolic activities. Although additional work is still needed to describe all the relevant microbial enzymes and their specificities, we have shown that a glycan-centered analysis can provide information about the active glycosidases operating on human milk.