Erratum: Comparative transcriptomics reveals key differences in the response to milk oligosaccharides of infant gut-associated bifidobacteria

Breast milk enhances the predominance of Bifidobacterium species in the infant gut, probably due to its large concentration of human milk oligosaccharides (HMO). Here we screened infant-gut isolates of Bifidobacterium longum subsp. infantis and Bifidobacterium bifidum using individual HMO, and compared the global transcriptomes of representative isolates on major HMO by RNA-seq. While B. infantis displayed homogeneous HMO-utilization patterns, B. bifidum were more diverse and some strains did not use fucosyllactose (FL) or sialyllactose (SL). Transcriptomes of B. bifidum SC555 and B. infantis ATCC 15697 showed that utilization of pooled HMO is similar to neutral HMO, while transcriptomes for growth on FL were more similar to lactose than HMO in B. bifidum. Genes linked to HMO-utilization were upregulated by neutral HMO and SL, but not by FL in both species. In contrast, FL induced the expression of alternative gene clusters in B. infantis. Results also suggest that B. bifidum SC555 does not utilize fucose or sialic acid from HMO. Surprisingly, expression of orthologous genes differed between both bifidobacteria even when grown on identical substrates. This study highlights two major strategies found in Bifidobacterium species to process HMO, and presents detailed information on the close relationship between HMO and infant-gut bifidobacteria.

Interestingly, the energetic value derived from conventional digestion of these molecules is negligible for the nursing infant. Intestinal enzymes are incapable of breaking down the majority of complex linkages found in HMO 7 . While initially puzzling for scientists, several major functions have emerged and attributed to these molecules. For example, HMO can prevent pathogen binding to the intestinal epithelium given their structural similarity with glycoconjugates in the brush border 2 .
In addition, HMO are substrates for beneficial microbes in the developing infant colon 8 . Studies have consistently shown that the composition of the intestinal microbiome among breast-fed and formula-fed infants is different [9][10][11] . This is consistent with HMO enriching for the populations capable of efficiently utilizing these substrates.
In general, it is understood that species of the Bifidobacterium genus dominate the infant gut microbiota in the first year of life [12][13][14] . Interestingly, there are only a few bifidobacterial species that are consistently isolated from infant feces, including Bifidobacterium longum subsp. longum (B. longum), Bifidobacterium longum subsp. infantis (B. infantis), Bifidobacterium breve, and Bifidobacterium bifidum, among other lesser taxa 11,15-17 . It has been previously shown that certain infant-associated bifidobacteria can utilize HMO purified from breast milk as the sole carbon source 18 . This first suggested potential co-evolution between bifidobacteria and their mammalian hosts mediated by milk oligosaccharides. This consumption phenotype was observed for some B. infantis and B. bifidum strains 19,20 , and recently it has been shown that also several B. breve can target these molecules, including fucosylated HMO 21 .
Comparative genomics and functional studies have been very useful in providing a molecular framework underlying bifidobacterial HMO utilization, particularly for B. infantis ATCC 15697 22 and B. bifidum PRL2010 20 . B. infantis deploys a HMO utilization strategy involving translocation of intact oligosaccharides using ABC transporters 23 . Once internalized, milk oligosaccharides are degraded by several intracellular glycosyl hydrolases that target linkages inherent to HMO 24,25 . Released monosaccharides are fermented in the central fructose-6-phosphate phosphoketolase pathway also referred to as the "bifid shunt", resulting in the production of ATP and secretion of acetate and lactate as end products. In contrast, the HMO consumption strategy of B. bifidum differs dramatically from B. infantis 24 . In B. bifidum, HMO are initially processed extracellularly, primarily via secreted glycosyl hydrolases that degrade HMO into constituent mono-and disaccharides. B. bifidum then imports the resultant products to be catabolized intracellularly through the bifid shunt 26 .
In order to gain a fundamental understanding of how these two different bifidobacteria have adapted to milk oligosaccharide utilization and therefore grow in the infant intestinal environment, it is essential to address the diversity of phenotypes found in infant-gut associated strains and also to dissect the global transcriptional responses of these microorganisms to individual and chemically different oligosaccharides found in breast milk. In this study, we determined the ability of B. bifidum and B. infantis isolates to utilize pooled HMO and the synthetic HMO LNT, LNnT, 2FL, 3FL and 6 SL as well as mucin. In addition, we detailed genome-wide transcriptional responses of B. bifidum and B. infantis during growth on these HMO substrates as the sole carbon source.

Growth of B. infantis and B. bifidum on pooled and individual HMO.
In a previous study, we created a collection of 461 Bifidobacterium isolates from 40 breast-fed infant fecal samples 21 . Among these, 297 were identified as B. longum (subsp. infantis or subsp. longum), and 22 as B. bifidum 21 . In order to discriminate these isolates at the strain level, Multilocus Sequencing Typing (MLST) analysis 27 was performed on purified genomic DNA of these isolates as well as 15 other strains obtained from culture collections (Table S1). Isolates from unrelated infants that shared identical MLST profiles were conservatively considered as separate isolates in this study. The MLST analysis identified 5 B. bifidum and 14 new B. infantis strains (Table S1). Together with the collection strains also analyzed, a consensus phylogenetic tree of the concatenated MLST data was created ( Figure S1). Additional information about the MLST analysis is shown in Tables S2-S5. Previous studies have characterized the HMO utilization phenotype using pooled HMO purified from breast milk. Here we evaluated the ability of B. infantis and B. bifidum strains to use individual milk oligosaccharides as a carbon source. We examined strains typed in this study, in addition to other B. infantis and B. bifidum collection strains 28 (Table S1). In addition to HMO purified from pooled breast milk samples, we used synthetically generated HMO species such as lacto-N-tetraose (LNT), lacto-N-neotetraose (LNnT), 2-fucosyllactose (2 FL), 3-fucosyllactose (3FL), and 6′ sialyllactose (6 SL). These individual HMO were selected to represent major structural classes such as neutral non-fucosylated HMO (LNT and LNnT), fucosylated HMO (2FL and 3FL), and sialylated HMO (6 SL). The structures of these HMO species are depicted in Fig. 1. In addition, all strains were tested for growth on porcine mucin, which is a glycosylated protein accesible to certain gut microbes 29 .
In general, all B. infantis strains grew to high OD 600 values on pooled HMO and every individual HMO tested (Table 1). One exception is the utilization pattern of 6 SL, where in a few cases moderate or poor growth was observed. In contrast, mucin was not a utilizable substrate for any B. infantis strain, results in agreement with previous observations 30 for the type strain ATCC 15697 (Table 1).
Interestingly, B. bifidum strains exhibited more phenotypic variability while subsisting on HMO (Table 1). Although most strains were capable of utilizing pooled HMO, four of the thirteen strains (JCM 1255, JCM 1209, JCM 7004 and S28a) showed little capacity to grow on pooled HMO, LNT, LNnT or even mucin (Table 1). JCM 1255 is the type strain and it is known to not represent the growth of B. bifidum on HMO 31 . Utilization of 2 FL, 3FL and 6 SL was also strain-dependent, with several strains unable to use 6 SL as the sole carbon source (Table 1).
We performed a principal component analysis (PCA) with the maximum OD 600 values of each strain growing on every substrate used in the study (Fig. 2). The descriptors HMO, LNT, LNnT, LAC, 2 FL, 3FL and 6 SL were mainly explained by the first axis of PCA analysis (Principal component 1) and are clearly associated with most of the bifidobacteria strains studied. The clustering of these variables indicates that B. infantis and B. bifidum strains grew on HMO similarly to LNT and LNnT. In addition, maximum OD 600 values on 2 FL, 3FL and 6 SL were more similar to each other ( Fig. 2A). The second axis of PCA (Principal component 2) was defined by mucin growth and was positively correlated with most of B. bifidum strains. In addition, B. infantis strains clustered more tightly together compared to B. bifidum strains, which spread in the plot and displayed more variable growth on HMO included in the analysis (Fig. 2B). Collectively, these results show that the HMO growth phenotype is mostly conserved among B. infantis isolates, and that B. bifidum displays a larger diversity of phenotypes in HMO consumption. Genome sequencing of B. bifidum SC555. In order to analyze the global transcriptome of representative strains of B. infantis and B. bifidum in response to major milk oligosaccharides, we first obtained the genome sequences of representative species. B. infantis ATCC 15697 was chosen due to the availability of a completely sequenced genome 22 and the conservation of genetic and phenotype observed among B. infantis strains here and previous studies 28 . We sequenced the genome of B. bifidum SC555, a strain isolated from breast-fed infant feces, as it exhibited the best growth on HMO fractions tested in this study. Genome sequencing of SC555 generated over 40 million reads generated on a Illumina HiSeq sequencer. Reads were assembled using the genome of B. bifidum PRL2010 as reference 20 , annotated and  (Table S6). The SC555 genome totals 2.24 Mbp, with a G + C content of 62.65%. These values are similar to those observed for other available B. bifidum genomes (Table S6). Analysis of orthologous genes shared in strain SC555 and B. bifidum PRL2010 indicates a high degree of similarity ( Figure S2). In addition, unique genes on each genome are mostly hypothetical without a predicted function, associated with prophages, or restriction-modification systems (data not shown). Finally, B. bifidum SC555, PRL2010 and JCM1255 have an identical number of transporters and glycosyl hydrolases previously characterized in the context of HMO utilization 20,26 (Table S7).

The transcriptomes of B. infantis and B. bifidum in response to HMO.
To profile the global transcriptional response of B. bifidum and B. infantis to major milk oligosaccharides, two sets of experiments were carried out. First, strains B. infantis ATCC 15697 and B. bifidum SC555 were grown on HMO purified from pooled breast milk 18 . Samples were extracted from duplicate cultures at early, mid and late exponential phase of growth (referred to as HMOearly, HMOmid and HMOlate). Second, these two strains were cultured on single purified milk oligosaccharides as the sole carbon source: LNT, LNnT, 2 FL, 3FL, and 6 SL. Sampling was performed at the mid-exponential phase. The two bifidobacterial strains were also grown on lactose as a reference sugar, and B. bifidum SC555 was also grown on porcine mucin for comparative purposes.
RNA extracted from the above-mentioned samples were depleted of ribosomal RNA and used to quantify absolute transcript concentrations using RNA sequencing (RNA-seq). It has been recently estimated that 5-10 million reads is an adequate number to interrogate the bacterial transcriptome with high resolution 32,33 . RNA-seq experiments generated over 365 million reads in total, with each sample averaging 10.4 million reads for B. infantis samples, and 8.8 million reads for B. bifidum. Over 70% of these reads aligned to coding genes; 15% aligned to intergenic regions (Table S8). Ribosomal RNA depletion led to total average representation of 16S and 23S rRNA transcripts of less than 1%. General features and statistics from the RNA-Seq experiments are summarized in Table S8.
We first compared the whole transcriptomes of both species on every glycan substrate and growth phase (Fig. 3A,B) using hierarchical clustering. In all RNA-seq experiments conducted on both bifidobacterial species, biological replicates were more similar to each other than to other samples as expected. For B. infantis, whole transcriptomes in response to 2 FL, 3FL, and 6 SL formed one primary cluster that was distinct from lactose ( Fig. 3A). In contrast, we found that global transcriptomes of growth on LNT and LNnT were much more similar to HMO (Fig. 3A). In addition, temporal responses to growth on HMO were also similar among each other, indicating that the initial response mounted to HMO is rather stable during exponential growth on HMO.
Unlike B. infantis, the global transcriptional responses of B. bifidum to 2 FL, 3FL, and 6 SL were very similar to lactose (Fig. 3B), forming a separate branch in the analysis. Transcriptomes of B. bifidum SC555 during growth on HMO at different time points were also closely similar to profiles on LNT, LNnT and also mucin. In summary, for both species global transcriptomes of growth of LNT and LNnT  (Tables 2 and 3). These are housekeeping genes and provide insights into bifidobacterial physiology. These genes include ribosomal proteins and elongation factors used in protein synthesis. Interestingly, the most transcribed locus across all samples was RNase P, which is a catalytic RNA that participates in post-transcriptional modification of tRNA molecules 34 . The gene is encoded by BBIF_01522 in B. bifidum and what was previously annotated to be an intergenic region in B. infantis 22 (coordinates 1,013,344-1014356). Several genes encoding subunits of the F 1 F 0 ATP synthase were also highly expressed (Tables 2 and 3). Other genes, such as a bacterial histone-like protein (Blon_0679 and BBIF_00999), chaperonines, and key enzymes of the bifid shunt such as fructose-6-phosphate phosphoketolase (Blon_1722 and BBIF_00286) and glyceraldehyde-3-phosphate dehydrogenase (Blon_0900 and BBIF_01046) were highly expressed as well.

Comparative transcriptomics of B. infantis in response to different milk oligosaccharides.
Comparative genomics and functional studies have previously provided a framework of HMO utilization in B. infantis and B. bifidum 8,20,22,26,35 , which includes ABC transporters and uniporters, glycosyl hydrolases and feeder pathways. Lists including these genes and their annotations are presented in Tables S9 and S10. In addition to confirming expression of genes of interest during growth on HMO, RNA-seq was used to characterize change in expression in response to specific HMO structures and how these responses evolve during exponential growth.
A first subset of genes displayed consistent expression values across all conditions tested (Fig. 4). These genes encode important functions such as an ATPase associated to ABC transporters (Blon_2475), galactose metabolism genes and a β -galactosidase with lactase activity 36 , Blon_2334 ( Fig. 4 and Table S9).
Among genes associated with HMO consumption in B. infantis (Table S9), RNA-seq confirmed that several of them are significantly upregulated during growth on HMO (Fig. 4). This induction was steady across exponential growth time points (Fig. 4). These genes are located mostly in the HMO cluster I 22 (Blon_2335 to Blon_2346), and Blon_0881 and Blon_0882, encoding enzymes responsible for GlcNAc derivation to the bifid shunt. These results suggest that the HMO cluster I acts as an HMO-inducible unit, and that it is co-regulated in a similar fashion with this GlcNAc utilization pathway.
It is interesting to note that these HMO-utilization genes are also induced several fold when B. infantis grows on LNT, LNnT and 6 SL (Fig. 4). While this is expected for LNT, which is the most abundant HMO 5 , and for LNnT which is an isomer of LNT, 6 SL eliciting a similar response to HMO was not  predicted. LNT and LNnT are GlcNAc containing neutral HMO, while 6 SL contains a negatively charged sialic acid residue. In contrast, the expression of these genes is different when B. infantis consumes 2 FL or 3FL (Fig. 4). When growing on these sugars as the sole carbon source, the expression of the HMO cluster displayed only basal RPKM values, and other gene clusters appear to be used instead. Genes Blon_0341 to Blon_0344, Blon_2202 to Blon_2204 and Blon_2305 to Blon_2309, were induced by 2 FL, 3FL but no other substrates. These genes encode functions related to FL import through ABC transporters 23 , and putative enzymes participating in fucose metabolism 37 (Table S9). Thus, the transcriptional results suggest that fucosyllactose is imported and catabolized using proteins encoded by these alternative genes.

Differential transcription of key functions of the HMO utilization in B. bifidum. A model has
been proposed regarding how B. bifidum obtains energetic value from HMO and mucin glycoconjugates 20,32 . Important determinants are extracellular glycosyl hydrolases, that generate mono and disaccharides that are captured by membrane porters and one ABC transporter, specific for HMO-derived lacto-N-biose (LNB) and mucin-derived galacto-N-biose (GNB). This gene set is shown in Table S10.
Some of these genes displayed high expression at all time points during growth on HMO, regardless of the substrates used (Fig. 5). These include important determinants in B. bifidum HMO and mucin consumption, such as lacto-N-biosidase (BBIF_00533), lacto-N-biose phosphorylase (BBIF_01217), and an endo-α -N-acetylgalactosaminidase (BBIF_00503).
Several of these genes were largely induced during growth on HMO compared to lactose. In general, this response was stronger during early growth, with lower values to the end of exponential growth, including the LNB/GNB ABC transporter cassette and the LNB/GNB utilization pathway (Fig. 5, Table  S10). Interestingly, these genes were similarly induced during growth on LNT, LNnT and even mucin, likely reflecting the importance of LNB and GNB to B. bifidum 38 .
In contrast, the LNB/GNB cluster and pathway were not induced by 2 FL or 3FL, probably indicating that B. bifidum SC555 is mostly unresponsive to these substrates (Fig. 5). This is in agreement with the observation that two genes encoding α -fucosidases (BBIF_00008 and BBIF_01261), are actually induced by HMO, LNT, LNnT and mucin, but not by 2 FL or 3FL (which contain fucose). Collectively, these results suggest that while B. bifidum can use larger fucosylated oligosaccharides, it does not show a preference for 2 FL or 3FL and it only benefits from the lactose on these oligosaccharides.   54 . We next determined which of these predicted functions were differentially enriched within and between the two species (Tables S11-S15). The two species displayed different functional enrichments when examining genes up-regulated in the presence of HMO at the initial sampling point. During the HMOearly time point relative to lactose, the B. infantis transcriptome is enriched for the GO terms "carbohydrate transport" and "transporter activity" as well as COGs of both components of the ABC-type sugar transport system (Table  S11). This enrichment was also observed during growth on individual HMO, including 2 FL and 3FL (Table S11). In contrast, the B. bifidum transcriptome was not enriched for "carbohydrate transport" until HMOlate samples (Table S13). Instead, genes upregulated by B. bifidum at the HMOearly time point were enriched for translation (e.g. ribosome, rRNA binding, translation) as well as "cell adhesion" and "carbohydrate metabolic process" (Table S13). GO terms and COGs related to carbohydrate transport were also enriched in the B. infantis response to individual HMO substrates, but not in the B. bifidum response to these substrates (Tables S11 and S13). Interestingly, B. infantis appears to down-regulate nucleotide biosynthesis, at the mid and late phase of growth on HMO as well as on LNT relative to lactose. These observations support our hypothesis that B. infantis has innovated several parallel systems to transport intact milk oligosaccharides. This is in contrast with B. bifidum which deploys a far more limited oligosaccharide transport system with a substrate acquisition strategy focused on extracellular mono-and disaccharides generated from HMO.  Table S9.

Direct comparison of B. infantis and B. bifidum by orthologs.
To directly compare the transcriptional response between the two bifidobacterial species, we identified orthologs for which there is a single copy present in each genome (see Methods). Subsequently, we placed all gene expression values on this common coordinate system. Between the B. infantis and B. bifidum genomes, 1266 single copy orthologs were identified. Of these, between 772 and 1113 were differentially expressed in the two species in response to the same substrate (Table S15). In response to each substrate, roughly half of the differentially expressed genes are up-regulated in B. infantis, relative to B. bifidum, and the other half are up-regulated in B. bifidum, relative to B. infantis (Table S15). Importantly, each species had a distinct response to each substrate, even lactose, when comparing the orthologous genes.
Even in situations where the two species have the same genetic capacity (i.e. both have a single copy ortholog), the expression of that single ortholog in response to the same substrate varied by nearly 150-fold. For example, components of the phosphotransferase system exist in both species but are up-regulated in B. bifidum 139-to 148-fold compared to B. infantis in response to HMO at the early time point (Table S16). Other orthologs are likewise highly expressed in one species and essentially unexpressed in the other (Tables S16 and S17). The NLPA lipoprotein (Blon_1721) in B. infantis is 130-fold higher than its ortholog in bifidum (BBIF_01591). It is most highly expressed in response to lactose; thus, its function is expected to be independent of HMO substrates.
The B. infantis short-chain dehydrogenase/reductase (SDRs) Blon_2339 and Blon_2308, together, are expressed 111.9-fold higher than their orthologous gene in B. bifidum, BBIF_01220 (Table S17). These genes are probably involved in fucose metabolism 37 , and are up-regulated in response to HMO relative to lactose (p < 2.3e-31). Blon_2308 is uniquely significantly higher, by over 900-fold, in response to 2FL relative to lactose (p = 0). These data support the idea of fucose being actively metabolized by B. infantis, while B. bifidum seems to prefer other monosaccharides such as Gal, GlcNAc and Glc, cross-feeding fucose and sialic acid to other species.
A correlation map of the whole transcriptomes, based on single copy orthologs, for both species on all tested substrates is shown in Fig. 6. Perhaps most striking, the B. bifidum response to lactose, 2 FL, and 3FL is more similar to B. infantis HMO-late than it is to its own response to HMO, LNT, LNnT, or mucin (Fig. 6). In other words, the B. bifidum response to lactose, 2 FL and 3FL correlates poorly with its own response to LNT, LNnT, mucin, or HMO at any point. Relative to lactose, there are only 230 and 138 genes differentially expressed by B. bifidum in response to 2 FL or 3FL, respectively. In contrast, there are over a 1000 genes differentially expressed by B. bifidum in response to the other substrates (Tables S18 and S19). This confirms that B. bifidum has a fundamentally different strategy for uptake and utilization of lactose and FL compared to the more complex oligosaccharides.

Discussion
In this study we compared the physiological and transcriptional responses to HMO of two representatives of important species of infant-gut associated bifidobacteria: B. infantis and B. bifidum. Importantly, we determined the transcriptional responses to pooled and single isolated HMO molecules, linking in detail biological function with substrate structural diversity.
While almost all B. infantis strains were efficient in utilizing a diversity of HMO as the sole carbon source, B. bifidum strains were more variable (Table 1). This is interesting since most of the strains studied were from infant origin, and it could be expected that they are all competitive in HMO consumption. While this was the case for B. infantis, several B. bifidum strains did not show a major preference for HMO, indicating that they may target other carbon sources. Growth analysis in Table 1 showed that medium and poor HMO-utilizing strains still display the ability to grow on LNT or mucin. It is very interesting to note that key HMO genes (lacto-N-biosidase, galactosidases, fucosidases and sialidases;  Table S7), are present in the same numbers in high growers (SC555) and poor growers (JCM 1255), indicating that the physiological differences among these strains are due to point mutations or differential regulation of these genes. In addition to the implications to host-microbial interactions, this wide array of responses to HMO has the potential to impact applications such as probiotic design, where strain-level differences in substrate utilization require additional consideration. Another recent study studied a similar number of B. breve strains, showing that certain of these isolates were much more competitive in different HMO consumption than previously reported for the type strain 21 .
In this study we used RNA-sequencing for determining the global transcriptome of B. infantis and B. bifidum during the in vitro consumption of specific, structurally different HMO, and during exponential growth. While our group and others have determined certain genes participating in HMO consumption in both species, in this study we were able to screen how individual HMO elicit differential transcriptomes in both bifidobacteria.
An important observation is that B. infantis and B. bifidum have a markedly different physiological response to FL. Genetic factors (secretor genes) such as the α 1,2 fucosyltransferase (FUT2) and α 1,3/4 fucosyltransferase (FUT3) genes can significantly alter HMO composition 39,40 , especially increasing the abundance of 2 FL and α 1-2 fucosylated HMO. The impact of milks higher or lower in 2 FL on bifidobacterial species is still not clear, although certain studies suggest a correlation between secretor status and the overabundance of certain bifidobacterial species 41 .
In B. infantis, we have observed that 2 FL or 3FL induces the expression of discrete, alternative operons, distinct from the HMO cluster I. They contain functions related to fucosylated HMO import and fucose utilization, and interestingly are not induced by pooled HMO or single neutral HMO. In contrast, global transcriptomes of LNT and LNnT in both species were highly similar to HMO, and responses to HMO were mostly steady during in vitro growth. Noteworthy, several HMO species such as LNT or LNnT enhanced the expression of several HMO consumption genes in a similar manner as pooled HMO. This was evident in the acidic 6 SL as well. It is possible that the transcriptomes in response to neutral HMO and 6 SL are similar considering that sialic acid and GlcNAc metabolic feeder pathways are convergent in B. infantis 31 . Neu5Ac is metabolized to GlcNAc-6-P and later to fructose-6-P in Escherichia coli and Bifidobacterium breve 42,43 , and the respective genes were upregulated during growth on these substrates (Fig. 4, Table S9).
This transcriptional analysis was also useful in providing additional detail on both mechanistic models of HMO consumption deployed by infant-associated bifidobacteria. The similarity of transcriptomes in response to 2 FL, 3FL, 6 SL and lactose suggests that B. bifidum SC555 appears to prefer the utilization of lactose when growing on short HMO, probably releasing fucose and sialic acid to the environment. This in turn might allow cross-feeding to other gut species. This has been recently shown using mucin oligosaccharides and 3 SL 43,44 , where B. breve can feed on sialic acid released from B. bifidum activity. This is supported by the absence or lack of activation of fucose and sialic acid metabolic pathways in the genome or transcriptome of B. bifidum SC555 (Table S10). In contrast, B. infantis seems to activate feeder pathways for fucose and sialic acid ( Fig. 4 and Table S9) during most of the conditions tested, reflecting its ability to further process these monosaccharides.
Finally, global transcriptome analysis and ortholog comparison presented another set of fundamental differences between each species' physiology. While we expected to observe an induction of carbohydrate transport and metabolism genes in B. infantis, this upregulation was just modest in B. bifidum SC555 especially during growth on FL (Tables S16 and S17). We also noted that expression of several orthologous genes was also drastically different, which indicates that both species operate and respond differently to HMO. For example, while the HMO utilization strategy in B. bifidum relies in part on a PTS system, the corresponding homolog in B. infantis does not appear to participate in this process. Putative fucose metabolism genes were in contrast induced several fold only in B. infantis.
The differences presented therefore reflect two different lifestyles found in the infant intestinal microbiome that have apparently converged in specialization in utilization of HMO. It could be possible that both B. bifidum and B. infantis display biogeographical differences as well, considering that most B. bifidum strains are competitive in mucin degradation, associating this species with the mucus layer. In contrast, B. infantis could reside mostly in the lumen. Certain species in the gut microbiome, such as Clostridium and Bacteroides, can be preferentially located in the mucus layer or the intestinal lumen 45 .

Conclusions
Both B. infantis and B. bifidum are able to utilize HMO as a sole carbon source, but with divergent strategies. In the present study, we showed that B. infantis strains are highly homogeneous in their ability to use specific HMO substrates, while B. bifidum isolates were more diverse, with some strains unable to grow on fucosylated or sialylated HMO. Choosing representative strains that grow well on HMO, we confirm that B. infantis expresses an assortment of transporters to internalize HMO substrates while B. bifidum expresses enzymes to hydrolyze HMO extracellularly. The transcriptome of each isolate varies dramatically even in response to structurally similar substrates. Likewise, the two strains have different transcriptional responses to the same substrate, even when lactose is the sole carbon source.
Our findings highlight the genetic and functional diversity between strains and between different isolates of the same strain as well as the specific mechanisms used by these isolates to utilize HMO. Such detailed mechanistic knowledge of the relationships between substrates and specific bifidobacteria will inform effective prebiotic and probiotic strategies. Synbiotic supplements could be designed to pair specific prebiotics with probiotics that will consume them via known strategies to achieve predefined functional outcomes. (Table S1) were obtained from fecal samples from exclusively breast-fed term infants 21 , the Japanese Collection of Microorganisms (Riken Biosource Center Japan), the American Type Culture Collection (Manassas, VA), and the University of California Davis Viticulture and Enology Culture Collection (Davis, CA). For routine experiments, bifidobacteria were grown on de Mann-Rogose-Sharp (MRS) broth supplemented with 0.05% w/v L-cysteine (Sigma-Aldrich, St. Louis, MO), and incubated for 18 h at 37 °C in an anaerobic chamber (Coy Laboratory Products, Grass Lake, MI), in an atmosphere containing 5% carbon dioxide, 5% hydrogen, and 90% nitrogen. Prior to each assay all bacteria were subcultured twice.

Bacteria and media. Bifidobacterium strains used in this study
Multilocus sequence typing (MLST) of strains. MLST analysis of B. infantis and B. bifidum strains targeted intragenic regions of seven housekeeping genes clpC, purF, gyrB, fusA, Iles, rplB, rpoB that were selected based on a previous study 27 . For PCR amplification, one μ l of extracted DNA was added to 50 μ l reaction mixture containing 50 pmol of primers, 200 μ M of each dNTP, 0.1 vol of 10X PCR buffer, 2.5 mM MgCl2, and 1 U AmpliTaq gold polymerase (Applied Biosystems). Cycling conditions were optimized for every primer set (Table S2) and consisted of an initial denaturation at 95 °C for 4 min, followed by 35 cycles of 95 °C for 30s, annealing at 60-67 °C for 30s, elongation at 72 °C for 60s, final extension at 72 °C for 7 min, and holding at 4 °C. The resulting amplicons were separated using a 1% agarose gel, followed by GelRed staining (Phenix Research Products, Candler, NC), and purification using a QIAquick PCR Purification Kit (Qiagen, Valencia, CA). Sequencing was performed on an ABI 3730 Capillary Electrophoresis Genetic Analyzer using BigDye Terminator chemistries at the University of California Davis DNA Sequencing Facility. Sequencing data for all loci was edited using BioEdit 7.0 and aligned using CLUSTAL W 46 . Phylogenetic analysis and concatenations of the sequenced loci were performed using the Molecular Evolutionary Genetic Analysis (MEGA) software version 5 (http://megasoftware. net). Descriptive evolutionary analysis including mol% G + C content, number of polymorphic sites, nucleotide diversity π /site, average number of nucleotide differences k were calculated using DnaSP version 5.10 (Table S3). Allelic sequences were assigned as described previously 47 (Tables S4 and S5). A minimum evolution tree of the concatenated loci was calculated using MEGA 5.0 (Fig. S1). Table S1 were tested for growth in the presence of seven different substrates: HMO 18 , LNT, lacto-N-neotetraose (LNnT), 2′ -fucosyllactose (2 FL), 3-fucosyllactose (3FL) (Glycom, Denmark), 6′ -sialyllactose (6 SL) (GenChem Inc. Korean), and hog mucin type II (Sigma). B. animalis subsp. lactis JCM 10602 was also included as negative control for growth experiments. Two μ l of each resulting overnight culture were used to inoculate 200 μ l of modified MRS medium (mMRS), devoid of glucose and supplemented with 2% (w/v) of each substrate, except for mucin at 1%, as the sole carbohydrate source, and another 2 μ l inoculated into mMRS without added sugar. Mucin was autoclaved at 121 °C for 10 minutes before addition to broth. The media was supplemented with 0.05% (w/v) L-cysteine, and in all the cases the cultures in the wells of the microtiter plates were covered with 30 μ l of sterile mineral oil to avoid evaporation. The incubations were carried out at 37 °C in an anaerobic chamber (Coy Laboratory Products, Grass Lake, MI). Cell growth was monitored in real time by assessing optical density (OD) at 600 nm using a BioTek PowerWave 340 plate reader (BioTek, Winoosky, VT) every 30 min preceded by 15 seconds shaking at variable speed. Two biological replicates and three technical replicates were performed for every studied strain. The OD obtained for each strain grown on the different substrates, was compared with the OD obtained in the absence of a sugar source. This difference in OD (Δ OD) was used as a parameter to evaluate each strain's ability for growing on the different substrates.

Bifidobacterial growth in vitro on HMO. The 35 bifidobacterial strains in
Statistical analysis of growth. Statistical analysis of the data was carried out using SPSS for Windows, 15.0 (SPSS Inc Chicago, IL, USA). The relationships among the maximum OD 600 values of the growth on the different substrates by the bifidobacteria strains were evaluated by Pearson correlation coefficients and principal components analysis (PCA).
Genome sequencing. B. bifidum SC555 is a strain that was previously isolated from a 3-6 month old infant. For genome sequencing, the strain was grown on MRS broth under the anaerobic conditions described above. After 18 h, 5 ml of this culture was centrifuged at 12000 × g for 2 min, and used to extract DNA using the MasterPure Gram-positive DNA Purification Kit (Epicentre), following manufacturer instructions. DNA was quantified using the dsDNA HS Assay in a Qubit 2.0 fluorometer (Life Technologies, Carlsbad CA). DNA was also checked for integrity in 1% agarose gels. Next, 2 μ g of the genomic DNA in 100 μ l of elution buffer EB were sonicated in a Bioruptor Standard (Diagenode, Denville NJ), for 10 min with cycles of 30s on and 30s off. Samples were stored at − 20 °C, and fragmentation sizes were evaluated in a 2100 Bioanalyzer using the Agilent DNA 1000 kit (Agilent Technologies, Santa Clara CA). Fragmented cDNA (15 μ l at 20 ng/μ l) was used to prepare a library for Illumina sequencing using