Varied Pathways of Infant Gut-Associated Bifidobacterium to Assimilate Human Milk Oligosaccharides: Prevalence of the Gene Set and Its Correlation with Bifidobacteria-Rich Microbiota Formation

The infant’s gut microbiome is generally rich in the Bifidobacterium genus. The mother’s milk contains natural prebiotics, called human milk oligosaccharides (HMOs), as the third most abundant solid component after lactose and lipids, and of the different gut microbes, infant gut-associated bifidobacteria are the most efficient in assimilating HMOs. Indeed, the fecal concentration of HMOs was found to be negatively correlated with the fecal abundance of Bifidobacterium in infants. Given these results, two HMO molecules, 2′-fucosyllactose and lacto-N-neotetraose, have recently been industrialized to fortify formula milk. As of now, however, our knowledge about the HMO consumption pathways in infant gut-associated bifidobacteria is still incomplete. The recent studies indicate that HMO assimilation abilities significantly vary among different Bifidobacterium species and strains. Therefore, to truly maximize the effects of prebiotic and probiotic supplementation in commercialized formula, we need to understand HMO consumption behaviors of bifidobacteria in more detail. In this review, we summarized how different Bifidobacterium species/strains are equipped with varied gene sets required for HMO assimilation. We then examined the correlation between the abundance of the HMO-related genes and bifidobacteria-rich microbiota formation in the infant gut through data mining analysis of a deposited fecal microbiome shotgun sequencing dataset. Finally, we shortly described future perspectives on HMO-related studies.


Metagenomic Data Mining Analysis
The occurrence of HMO and GOS utilization genes (identity ≥70%, read coverage ≥60%) in infant stool DNA was examined by using the tblastn algorithm (BLAST+ v2. 6.0) against the fecal microbiome shotgun sequencing dataset (MG-RAST version 4.0.3 (accession number "qiime:621"); http://metagenomics.anl.gov/) [39] (Supplementary Figure S1B). The quality-filtered reads were obtained as described previously [40], and the resulting shotgun sequencing data of 117,492 ± 55,056 reads/sample (means ± standard deviations) for 34 breast-fed and 27 formula-fed infants (≤1 year old), who resided in USA (n = 37), Malawi (n = 14) and Venezuela (n = 10), were used as the local database. The amino acid sequences used for queries were the same as described above. LnbY was omitted from the analysis due to its short length. The relative abundance (%) of each homolog gene in the fecal DNAs was calculated by dividing the hit read counts with the above identity criteria by total read counts. When redundant mappings were detected, the reads with the highest tblastn scores were chosen to count. If multiple reads had equally high scores, the count value was divided by the number of reads with the same score (e.g., count value = 0.5 for mappings with two reads with equally high scores (1/2); count value = 0.25 for mappings with four reads with equally high scores (1/4)). The number of the reads with multiple equally high tblastn scores was six in total. The abundance (%) of the genus Bifidobacterium was obtained from MG-RAST under the same accession number as above.

Lacto-N-Biosidase
Lacto-N-biosidase (EC 3.2.1.140) hydrolyzes LNT to produce LNB and lactose (Lac) (Supplementary Figure S2). The enzyme from B. bifidum (LnbB) is classified as GH20 [43], while the enzyme from B. longum (LnbX) is categorized into GH136 [60]. GH136 enzymes require a designated chaperon (LnbY) for its proper folding. LnbB and LnbX are highly specific for LNT hydrolysis. However, LnbX additionally accepts LNFP I and α-2,3-sialyl LNT (LSTa), although the activity is considerably low [44]. The different activity was also mentioned for two enzymes using the sugar chains of gangliosides and globosides [61]. Disruption of lnbX gene caused severe growth retardation on LNT in B. longum [60]. Inactivation of lnbB in B. bifidum has not been attained due to the difficulty of gene manipulation in this species. It is interesting to note that B. bifidum and B. longum have evolved two completely different genes to catabolize the same substrate, LNT (the most abundant core structure of HMOs), with almost equal kinetic parameters, even though they belong to the same genus and share the same environmental niche.

Transporters
As described later, bifidobacteria have evolved two different strategies for assimilating HMOs. One is transporter-dependent (intracellular digestion strategy), and the other is extracellular glycosidase-dependent (extracellular digestion strategy) (Table 1, Figure 1, Supplementary Figure S4). The presence of a transporter-dependent HMO consumer was first revealed by the genomic analysis of B. infantis ATCC 15697 T by David A. Mills's group [30]. The strain possesses the so-called HMO cluster that is comprised of several intracellular HMO-related glycosidases as well as SBPs for ATP-binding cassette-type (ABC) transporters. Seven SBPs are in the cluster, while the genome encodes for an additional thirteen SBPs in different loci. Mills's group subsequently analyzed the specificity of the SBPs using glycan array [62], which enhanced our understanding of HMO catabolism in bifidobacteria with the intracellular digestion strategy. However, studies on HMO transporters are still behind compared to the aforementioned glycosidases involved in the decomposition of HMOs. Indeed, only three transporters for LNB, FL, and LNnT have been physiologically characterized.
Nutrients 2020, 12, x FOR PEER REVIEW 7 of 22 for an additional thirteen SBPs in different loci. Mills's group subsequently analyzed the specificity of the SBPs using glycan array [62], which enhanced our understanding of HMO catabolism in bifidobacteria with the intracellular digestion strategy. However, studies on HMO transporters are still behind compared to the aforementioned glycosidases involved in the decomposition of HMOs. Indeed, only three transporters for LNB, FL, and LNnT have been physiologically characterized.     Table 1. The prevalence (%) of the genes in each (sub)species was determined by dividing the number of the retrieved genes with the above identity criteria by the number of the genomes examined (values in parentheses). The results are shown as a heatmap.

GNB/LNB Transporter
The GNB/LNB transporter was first isolated and identified to be responsible for the uptake of LNB that is extracellularly liberated from LNT by lacto-N-biosidase (LnbB from B. bifidum and LnbX from B. longum) [43,44] and for the uptake of GNB that is liberated from mucin O-glycans by endoα-N-acetylgalactosaminidase [63]. Introduction of a plasmid carrying the transporter genes (gltABC) into B. longum increased the accumulation level of 14 C-labeled LNB inside the cells, compared to the control strain carrying an empty vector [31,64]. The SBP of the transporter (GltA, or GNB/LNB-BP) from B. longum specifically binds LNB and GNB with Kd values of 87 and 10 nM, respectively [46]. The recent study also revealed that GltA of B. breve (78% identity to GltA of B. longum) can bind 3'galactosyllactose (GOS) with Kd values of 2 × 10 4 nM. Note that eight amino acid residues involved in the binding of LNB are completely conserved between all GltA homologs mentioned in this study, except for the asparagine residue at the position of 206 of B. longum GltA [46]. 9 1 ) B . k a s h i w a n o h e n s e ( 2 ) B . in f a n t is  Table 1. The prevalence (%) of the genes in each (sub)species was determined by dividing the number of the retrieved genes with the above identity criteria by the number of the genomes examined (values in parentheses). The results are shown as a heatmap.

GNB/LNB Transporter
The GNB/LNB transporter was first isolated and identified to be responsible for the uptake of LNB that is extracellularly liberated from LNT by lacto-N-biosidase (LnbB from B. bifidum and LnbX from B. longum) [43,44] and for the uptake of GNB that is liberated from mucin O-glycans by endo-α-N-acetylgalactosaminidase [63]. Introduction of a plasmid carrying the transporter genes (gltABC) into B. longum increased the accumulation level of 14 C-labeled LNB inside the cells, compared to the control strain carrying an empty vector [31,64]. The SBP of the transporter (GltA, or GNB/LNB-BP) from B. longum specifically binds LNB and GNB with K d values of 87 and 10 nM, respectively [46]. The recent study also revealed that GltA of B. breve (78% identity to GltA of B. longum) can bind 3 -galactosyllactose (GOS) with K d values of 2 × 10 4 nM. Note that eight amino acid residues involved in the binding of LNB are completely conserved between all GltA homologs mentioned in this study, except for the asparagine residue at the position of 206 of B. longum GltA [46].

FL Transporter
The FL transporter of Bifidobacterium was first identified by Yakult Co. Ltd.'s group and Mills's group, independently [12,65]. They reported that the transporter has an ability to import 2 -FL, 3-FL, LDFT, and LNFP I. B. infantis ATCC 15697 T possesses two paralogous FL transporters with their SBPs sharing 60% identity. Biochemical and structural analyses, combined with genetic analysis, unequivocally revealed that the functions of the two transporters are overlapping but distinct [40]. One (designated FL transporter-1 [SBP name: FL1-BP]) imports 2 -FL and 3-FL only, while the other (FL transporter-2 (SBP name: FL2-BP)) is capable of internalizing LDFT and LNFP I in addition to 2 -FL and 3-FL. The wild type strain of B. infantis outcompeted the transporter knockout strain to grow on a purified HMO mixture [40]. While FL1-BP binds preferentially to 2 -FL (the binding to 3-FL was not observed by isothermal calorimetry analysis under the tested conditions), FL2-BP binds to 2 -FL and 3-FL with the K d values of 5.4 and 6.0 µM, respectively. This intriguing dual specificity has been rationalized by X-ray crystal structure analysis of the ligand-complexed structures [40].

LNnT Transporter
LNnT transporter has been genetically identified in B. breve UCC2003 by van Sinderen's group [48]. James et al. [48] conducted a series of targeted gene knockouts based on the results obtained in the transcriptome analysis of cells grown on HMO-related compounds including LNnT and then monitored the growth of the mutants in media supplemented with several oligosaccharides. Consequently, nahS (LNnT-BP) was found to be the SBP responsible for the uptake of LNnT.
HMO assimilation by B. pseudocatenulatum and B. kashiwanohense has also been recognized. Xiao et al. [67] demonstrated that 33 out of 61 strains of B. pseudocatenulatum are able to utilize LNB. Matsuki et al. [12] reported LNT consumption by B. pseudocatenulatum. They also mentioned that some strains of this species assimilate 2 -FL, 3-FL, and LDFT. B. kashiwanohense was shown to be a good consumer of 2 -FL and 3-FL, and the ability is well conserved in this species [70]. The ability of B. kashiwanohense to consume LDFT and LNFP I has been recently suggested and might be a common characteristic of this species [40]. Not LNnT, nor 3 -sialyllactose (3 -SL), nor 6 -sialyllactose (6 -SL) were consumed by this species (Supplementary Figure S2) [70]. Assimilation ability for the other HMOs has not been described to our knowledge.

Two Different HMO Utilization Strategies and Prevalence of the Gene Set Required for HMO Assimilation
Infant gut-associated bifidobacteria have evolved the two different strategies for utilizing HMOs (intracellular and extracellular digestion strategies). B. bifidum and some B. longum strains use extracellular glycosidase(s) to liberate mono and/or disaccharides from HMOs outside the cells (Table 1, numbers 1-8). The liberated HMO degradants are subsequently imported by the cells and further degraded in the cytoplasm (e.g., Table 1, numbers 9, 10, and [27][28][29][30]. In contrast, the other Bifidobacterium species/strains directly internalize intact HMOs by specific transporters (e.g., Table 1, numbers [11][12][13] and degrade them intracellularly (Table 1, numbers [14][15][16][17][18][19][20][21][22][23][24][25][26]. The HMO utilization strategies employed by infant gut-associated bifidobacteria are schematically drawn in Figure 1 and Supplementary Figure S4. We carried out BLAST analysis to uncover the prevalence of the gene set required for HMO assimilation among Bifidobacterium species (Table 1). The species that are known to proliferate in infant guts, adult guts, or mouths, and non-human animal intestines or dairy products were chosen for the analysis (Figure 2). The results clearly showed that the HMO-related genes are almost exclusively found in the genomes of infant gut-associated Bifidobacterium species, and hardly detected in the genomes of other bifidobacterial species (Figure 2). The prevalence of the genes in the respective species varied significantly.

Bifidobacterium Bifidum
B. bifidum (60 genomes examined) showed high conservation of all genes required for extracellular HMO degradation (note that LnbB and LnbXY catalyze the same reaction) (Figure 2). The results are consistent not only with the shared high HMO assimilation ability [32,58,66,67,69,70], but also with the shared cross-feeding ability of this species as described later (Section 7).

Other Bifidobacterium Species
The genomes of other Bifidobacterium species except for the above six (sub)species, which reside in adult guts or mouths, and non-human animal intestines or dairy products (176 genomes examined), were found to lack HMO degradation genes with a few exception ( Figure 2). Indeed, some of these species showed poor growth on HMOs and LNB [12,43,67,72]. Instead, they appear to have gene sets for assimilating plant-derived glycans contained in adult or animal diet [73].

Abundance of HMO-Related Genes Is Associated with Bifidobacteria-Rich Microbiota Formation
Recently, Vatanen et al. [13,74] revealed, by fecal microbiome shotgun sequencing of the two different cohorts, that the HMO cluster homologs (Blon_2331-2361 of B. infantis ATCC 15697 T ) are enriched during breast-feeding, compared to after weaning. The results strongly suggest that the competitive advantage is conferred by the presence of the HMO cluster for bifidobacteria to dominate in the infant gut ecosystem. However, the increased fitness of bifidobacteria in the breast-fed infant gut microbial ecosystem should also be attributed to genes encoded in other loci. Our group recently showed that the FL1-BP and FL2-BP genes (Blon_0343 and Blon_2202 of B. infantis ATCC 15697 T ) are enriched in the fecal DNA of breast-fed infants and the relative abundance is positively correlated with the abundance of genus Bifidobacterium in the samples [40]. Interestingly, the abundance of the genes was negatively correlated with the fecal concentrations of fucosylated HMOs (2 -FL, 3-FL, LDFT, and/or LNFP I, the substrates of FL1-BP and FL2-BP). Thus, one interpretation of the results is that the bifidobacteria-rich microbiota is shaped by HMO consumption by Bifidobacterium in the breast-fed infant guts. The lnbX gene of B. longum was also found to be enriched in the exclusively breast-fed infant group, as compared with mixed (formula and breast)-fed infant group [60]. Given these findings, we analyzed the abundance of all the heretofore identified HMO utilization genes (enlisted in Table 1 and Figure 2) in a metagenomic database.
The metagenomic dataset we analyzed (MG-RAST, accession number qiime:621) [39] includes 34 breast-fed and 27 formula-fed infants (≤1 year old) residing in the USA (n = 37), Malawi (n = 14), and Venezuela (n = 10). The data mining analysis revealed that the abundances of the extracellular glycosidase genes are indistinguishable in between the fecal DNAs obtained from formula-fed infants and breast-fed infants (Figure 3 and Supplementary Figure S5). By contrast, the abundances of the transporter and intracellular enzyme genes (for intracellular digestion strategy) were significantly higher in the breast-fed infant group than the formula-fed infant group, except for nahS (LNnT transporter), lacZ2, and lacZ6 (GH2 β-1,4-galactosidases) (Figure 3; Mann-Whitney U test). The observed significant differences in the gene abundance between the two groups are mostly attributable to Malawian and Venezuelan breast-fed infants, which was revealed when the metagenomic reads were sorted out by the three countries (Supplementary Figure S5; Dunn's test).
We then analyzed the association between the relative abundance of respective HMO utilization genes and the relative abundance of genus Bifidobacterium in the infant stools ( Figure 4A; Spearman's rank correlation coefficient). The abundance of Bifidobacterium in the Malawian and Venezuelan breast-fed groups was positively correlated to the abundance of most of the genes responsible for the intracellular digestion strategy with statistical significances, while no strong positive or negative correlation was observed between the abundances of Bifidobacterium and the genes responsible for the extracellular digestion ( Figure 4A). In contrast, in the breast-fed samples from the USA, the Bifidobacterium abundance was positively associated with the genes responsible for extracellular digestion ( Figure 4A). In this group, relatively weak correlation was detected between the Bifidobacterium abundance and the genes responsible for intracellular digestion, although a strong positive correlation was observed for the LNT β-1,3-galactosidase gene that is highly conserved in infant gut-associated Bifidobacterium species (Figure 2). It should be noted the abundance of Bifidobacterium was essentially similar between the three country groups [40]. These results imply that bifidobacteria residing in the Malawian and Venezuelan infant guts mainly use the intracellular digestion strategy for HMO assimilation, while those colonizing USA infant intestines mainly adopt the extracellular digestion strategy (described later).
Nutrients 2020, 12, x FOR PEER REVIEW 12 of 22 breast-fed groups was positively correlated to the abundance of most of the genes responsible for the intracellular digestion strategy with statistical significances, while no strong positive or negative correlation was observed between the abundances of Bifidobacterium and the genes responsible for the extracellular digestion ( Figure 4A). In contrast, in the breast-fed samples from the USA, the Bifidobacterium abundance was positively associated with the genes responsible for extracellular digestion ( Figure 4A). In this group, relatively weak correlation was detected between the Bifidobacterium abundance and the genes responsible for intracellular digestion, although a strong positive correlation was observed for the LNT β-1,3-galactosidase gene that is highly conserved in infant gut-associated Bifidobacterium species (Figure 2). It should be noted the abundance of Bifidobacterium was essentially similar between the three country groups [40]. These results imply that bifidobacteria residing in the Malawian and Venezuelan infant guts mainly use the intracellular digestion strategy for HMO assimilation, while those colonizing USA infant intestines mainly adopt the extracellular digestion strategy (described later).    Figure S6 for GOS transporter genes (B). *, **, ***, and **** denote p < 0.05, p < 0.01, p < 0.001, and p < 0.0001 respectively. BF: breast-fed; FF: formula-fed; GL-BP: galactosyllactose-binding protein; N.A.: correlation analysis was not applicable due to the absence of the reads attributable to FL1/2-BP or GlcNAc-ase (Blon_2355) genes in the tested subjects.  Figure S6 for GOS transporter genes (B). *, **, ***, and **** denote p < 0.05, p < 0.01, p < 0.001, and p < 0.0001 respectively. BF: breast-fed; FF: formula-fed; GL-BP: galactosyllactose-binding protein; N.A.: correlation analysis was not applicable due to the absence of the reads attributable to FL1/2-BP or GlcNAc-ase (Blon_2355) genes in the tested subjects.
In the formula-fed infant group (the USA only), the abundance of Bifidobacterium appeared to show strong positive correlations with those of GOS transporter genes with statistical significance [47,75] ( Figure 4B; Spearman's rank correlation coefficient). GOSs are world-wide prebiotics supplemented in infant formula and are known to increase the population of bifidobacteria in the guts [76,77]. It is interesting to note that the abundances of the GOS transporter genes were similar between the formula-fed and breast-fed groups except for GNB/LNB-BP (GltA) gene (Supplementary Figure  S6A,B); nonetheless, the slightly higher positive correlations were detected for 4 -galactosyllactose-SBP (4 GL-BP) and 6 -galactosyllactose-SBP (6 GL-BP) in infants who received formula milk than those receiving breast milk ( Figure 4B and Supplementary Figure S6C) (note that sample numbers are different between the groups (formula-fed: n = 27, breast-fed: n = 10-14)). The GNB/LNB-BP (GltA) gene homolog is highly conserved in the genomes of infant gut-associated bifidobacteria ( Figure 4B and Supplementary Figure S7).

Milk Oligosaccharide Composition May Dictate Transporter-Driven Bifidobacteria-Rich Microbiota Formation
HMO abundances and compositions vary among mothers, across geography, and between lactation stages within the same individual [23][24][25]. One of the most influential factors that give HMO structural variation between individuals is the maternal secretor status. In the mammary gland, fucosyltransferase-2 (FUT2) introduces α-1,2 linkage onto Gal residue (Fucα1-2Gal-O-R or H-antigen) at the non-reducing ends of milk glycans. Thus, HMOs synthesized by secretors (FUT2 +/+ or FUT2 +/− ) contain many H-antigen-carrying structures, such as 2 -FL, LDFT, LNFP I, and LNDFH I (≈50% of total HMOs by weight [25]), whereas those produced by non-secretors (FUT2 −/− ) do not. Considering that the conservation and prevalence of H-antigen-acting enzymes and transporters vary among infant gut-associated bifidobacterial species/strains (Table 1; Figures 1 and 2), it is not surprising that the secretor status affects microbiota formation in infant guts. Indeed, Lewis et al. [34] reported that Bifidobacterium strains that are able to utilize 2 -FL are enriched in the stools of the infants receiving breast milk of secretors, compared to those of non-secretors. Several previous studies showed that the bifidobacterial population increases more rapidly and abundantly in infants fed by secretor mothers than those fed by non-secretor mothers [34,78,79]. Two FL transporters we mentioned above can import both 2 -FL and 3-FL [40]; therefore, bifidobacteria carrying the transporters can adapt to FL from both secretors and non-secretors. Nevertheless, the above results suggested that the secretors' milk may be more advantageous for proliferating bifidobacteria. Detailed time-series analysis of the abundance of the gene set for HMO assimilation, as well as comparative studies of milk and fecal HMO concentration of mother-infant pairs during the lactation period, are necessary to understand the mechanistic differences in bifidobacteria-rich microbiota formation between infants receiving secretor's milk and non-secretor's milk. But, it is worth noting that FL2-BP (which binds to 2 FL and 3FL with equal affinity) is higher in abundance than FL1-BP (which preferentially binds to 2 FL over 3 FL), not only in breast-fed infant fecal metagenomes, but also in the genomes of infant gut-associated bifidobacteria (Figures 2 and 3) [40]. Bifidobacteria may have evolved and shared FL transporter-2 to flexibly adapt to different host genetic statuses. The expansion of FL2-BP's ability to additionally import LDFT and LNFP I may have occurred during the co-evolution of bifidobacteria with humans.
Given the importance of HMOs in the infant gut microbiota formation, industrial-scale production of HMOs has attracted significant attention in an effort to fortify infant formula [80]. Recently, 2 -FL and LNnT were approved for commercial supplementation. The safety, well-tolerance, and other benefits of the formula supplemented with 2 -FL (and LNnT) have been verified by several clinical trials [81,82]. These formulas may selectively promote the proliferation of bifidobacteria carrying either or both of FL and LNnT transporters. The results shown in Figure 4B might also provide evidence for the effectiveness of GOS supplementation to formula. The proposed transporter-driven bifidobacteria-rich microbiota formation is schematically presented in Supplementary Figure S8.

Cross-Feeding of HMO Degradants within Bifidobacterium Community
The transporter-dependent intracellular digestion strategy enables the bifidobacteria to efficiently capture preferred carbon sources in the competitive ecosystem over other gut microbes. However, in addition to this "selfish" strategy, bifidobacteria are considered to have another strategy to dominate the ecosystem; i.e., cross-feeding between the species/strains. Tannock et al. [83] examined the relationship between the presence of B. bifidum and the dominance of genus Bifidobacterium in the infant guts. Interestingly, they found that, when B. bifidum occupies >10% of genus Bifidobacterium, the corresponding microbiota possesses a higher ratio of Bifidobacterium abundance than those containing B. bifidum of less than 10% population among the genus. The predominance of genus Bifidobacterium in the presence of adequate amount of B. bifidum was observed only for breast-fed infant stools but not for cow-based formula milk-fed infant feces. Before their report, we had noticed that B. bifidum leaves HMO degradants unconsumed in the spent media, even after prolonged incubation on HMOs [58]. These results suggested that HMO degradants produced by B. bifidum extracellularly can be shared within bifidobacterial community. SL-mediated cross-feeding between B. bifidum and B. breve was shown by Egan et al. [84] and Nishiyama et al. [85] in co-culture experiments. In these cases, Neu5Ac liberated from SL by extracellular sialidase SiaBb2 of B. bifidum was utilized by B. breve carrying the nan cluster that is responsible for converting Neu5Ac into pyruvic acid and GlcNAc 6-phosphate with one ATP consumption.
When stool suspensions obtained from infants, children, and adults were cultured in the media supplemented HMOs in the presence and absence of B. bifidum, a remarkable difference was observed: i.e., the population of Bifidobacterium species other than B. bifidum significantly increased when B. bifidum was added prior to cultivation, as compared with non-added control [66]. Importantly, the relative abundance (%) of other Bifidobacterium species in the total bacterial community was increased as well in the fecal cultures supplemented with B. bifidum. On the other hand, in the stool samples grown on Glc, neither the relative abundance nor population of other Bifidobacterium species was significantly changed by the addition of B. bifidum. In addition, the growth stimulatory effect of B. bifidum on other Bifidobacterium species observed for HMO-supplemented fecal cultures disappeared when fuconojirimycin, a potent fucosidase inhibitor (both for GH29 and GH95), was added to the media [57,66]. Bifidobacteria-rich gut microbiota observed for USA infants that are associated with the abundance of extracellular digestion enzymes (Figure 4), thus, appears to be established by cross-feeding by B. bifidum. These "altruistic" behaviors might be a common characteristic of B. bifidum.

Conclusions
In the past 15 years, a series of the genes that are involved in HMO degradation and transport have been identified and characterized from bifidobacteria (Table 1), which led to the findings that these genes are almost exclusively present in the genomes of infant gut-associated bifidobacteria ( Figure 2). However, there have been no reports that systematically analyzed the conservation of genes among Bifidobacterium spp. In this review, we analyzed bifidobacterial genomes to understand the prevalence of the HMO-related gene set at the species and strain levels. We also conducted in silico data mining analysis using an infant fecal metagenomic dataset to examine how the abundances of the HMO-related genes are linked with bifidobacteria-rich microbiota formation. The results obtained were then discussed based on the HMO assimilation mechanism that each Bifidobacterium has evolved (Figure 1 and Supplementary Figure S4). Although our knowledge of HMO transporters is still incomplete, the presented data could dictate the importance that HMO assimilation pathways play in the formation of the infant gut microbiota.
Recent reports strongly suggest that in vivo consumption of HMOs is associated with bifidobacteria-rich microbiota formation in infant guts [12,[34][35][36][37]. Frese et al. [86] reported that continuous administration of B. infantis EVC001 to the breast-fed infants from day 7 to day 28 led to the decrease of fecal HMO concentrations in infant stools, which coincided with increased fecal abundance (>10 10 cells/g feces) of the strain. A similar observation was also reported by the same research group using B. breve M-16V [87]. Interestingly, in the former case, the fecal concentrations of most HMOs, including fucosylated ones, were decreased, whereas in the latter case, fecal consumption of HMOs was essentially limited to undecorated (non-fucosylated) HMOs. The two studies were conducted in different cohorts, and thus it is not appropriately comparable; however, it seems that the observed difference in the fecal HMO consumption behavior between the two cohorts may reflect the difference in the possession of HMO-related gene set of between B. infantis and B. breve (Figures 1 and 2).
Apparently, the formation of the infant gut microbiome is affected by the level of HMO consumption by bifidobacteria in infant guts. However, we must understand the extent to which endogenous (or exogenously supplemented) bifidobacterial species/strains that possess variable gene sets are responsible for HMO assimilation in order to gain mechanistic insight into bifidobacterial-rich microbiota formation in the guts. Such understanding helps us not only to understand the results obtained in clinical studies but also to design future clinical interventions that elucidate the effects of supplementation of HMOs and probiotic Bifidobacterium strains in formula.
Besides infant gut-associated bifidobacteria, it has been reported that a few other gut microbes belonging to different taxa have HMO consumption abilities. Several Bacteroides species were shown to assimilate 2 -FL, 3-FL, LDFT, 3 -SL, and 6 -SL [88]. Lactobacillus casei utilizes HMO degradants LNTri II and LNB [89,90]. Interestingly, Lactobacillus has evolved a specific phosphotransferase system to import LNB [90]. HMO utilization capability of these gut microbes is generally not very high, but it is worth exploring how minority species during the breast-feeding stage persist until the introduction of solid food (weaning), which potentially influences maturation of the microbiota in adulthood.

Supplementary Materials:
The following are available online at http://www.mdpi.com/2072-6643/12/1/71/s1. Figure S1: Graphical abstract of the in silico analysis procedures adopted in this study. Figure S2: Representative twelve HMO molecules described in this study, and the HMO degradants produced in the assimilation pathways. Figure S3: Unique specificity observed for α-l-fucosidases can be explained by the structural similarities between the different substrates. Figure S4: HMO utilization pathways of B. pseudocatenulatum and B. kashiwanohense. Figure  S5: Metagenomic data mining analysis of HMO-related genes of bifidobacteria. Figure S6: Metagenomic data mining analysis of GOS transporter homolog genes in bifidobacteria. Figure S7: Prevalence of GOS transporter genes in the genomes of bifidobacteria. Figure S8: Bifidobacteria-rich microbiota formation processes in infant guts can differ between feeding strategies and be partially dependent on specific transporters. Table S1: The accession numbers of Bifidobacterium strains used for the in silico analysis.