Human milk oligosaccharide-sharing by a consortium of infant derived Bifidobacterium species

Bifidobacteria are associated with a host of health benefits and are typically dominant in the gut microbiota of healthy, breast-fed infants. A key adaptation, facilitating the establishment of these species, is their ability to consume particular sugars, known as human milk oligosaccharides (HMO), which are abundantly found in breastmilk. In the current study, we aimed to characterise the co-operative metabolism of four commercial infant-derived bifidobacteria (Bifidobacterium bifidum R0071, Bifidobacterium breve M-16V, Bifidobacterium infantis R0033, and Bifidobacterium infantis M-63) when grown on HMO. Three different HMO substrates (2′-fucosyllactose alone and oligosaccharides isolated from human milk representing non-secretor and secretor status) were employed. The four-strain combination resulted in increased bifidobacterial numbers (> 21%) in comparison to single strain cultivation. The relative abundance of B. breve increased by > 30% during co-cultivation with the other strains despite demonstrating limited ability to assimilate HMO in mono-culture. HPLC analysis revealed strain-level variations in HMO consumption. Metabolomics confirmed the production of formate, acetate, 1,2-propanediol, and lactate with an overall increase in such metabolites during co-cultivation. These results support the concept of positive co-operation between multiple bifidobacterial strains during HMO utilisation which may result in higher cell numbers and a potentially healthier balance of metabolites.

Establishment of the infant gut microflora is a highly complex process, with feeding regime profoundly impacting on the development of neonate's microbial gut ecology [1][2][3] . Human milk contains a high concentration of indigestible carbohydrates, known as human milk oligosaccharides (HMO), which can reach the large intestine in an intact form, where they can act as growth substrates for beneficial bacteria 4 . To date, it is estimated that over 200 different HMO structures exist, with overall levels in human milk varying from 5 to 23 g/L. All HMO consist of a lactose backbone which is decorated (see below) and/or elongated into a variety of oligomeric structures by the addition of β1-3-or β1-6-linked lacto-N-biose (Galβ1-3GlcNAc-, type 1 chain) or N-acetyllactosamine (Galβ1-4GlcNAc-, type 2 chain) 5,6 . HMO decoration is achieved by the addition of fucose residues (e.g., 2′-fucosyllactose [2′-FL] and 3-fucosyllactose ), or N-acetylneuraminic acid residues [NeuAc-also known as sialic acid] (e.g., 3′-sialyllactose [3′-SL] and 6′-sialyllactose [6′-SL]) 5,6 . Variability in the chemical and structural conformations of HMO is biologically relevant 7 .
These glycans serve as selective regulators of infant gut microbiota composition, and facilitate the development of a Bifidobacterium-dominant microbiota in breast-fed infants 8,9 . A high level of bifidobacteria in the gut microbiota of infants is believed to contribute to enhanced development of the immature gut immune system, and is associated with reduced levels of intestinal infections and diarrhea 10,11 . Due to their purported health-promoting effects, bifidobacteria have been incorporated into many functional foods and supplements as probiotic ingredients 12,13 . Many members of the genus Bifidobacterium are known to dedicate a relatively large proportion of their genome to carbohydrate utilization, reflecting their adaptation to metabolism of hostproduced and/or host-ingested/dietary carbohydrates. Some of these genomes, particularly infant bifidobacterial species, display a clear adaptation to the nursing period when the diet of their host may be solely composed of (HMO-rich) breast milk [14][15][16] . One of the reasons as to why certain bifidobacterial species are such abundant and prevalent members of the (early) gut microbiota is due to their remarkable capacity to metabolize and consume  17,18 . These studies have enabled detailed profiling of (bifido)bacterial consumption of individual HMO and demonstrate specific and preferential oligosaccharide consumption by particular bifidobacterial strains [19][20][21] . However, in order to fully appreciate the role that HMO play in the creation of a Bifidobacterium dominant microbiota, consideration must be given to the cross-feeding relationships that exist between members of the infant microbiota 22 . A key determinant of microbial dynamics in the gut microbiota is represented by competition and sharing of nutrients. In the current study, we performed co-cultivation experiments involving four bifidobacterial strains when grown on HMO substrates to elucidate if, and to what extent, trophic interactions prevail among bifidobacteria when present in a competitive environment. Parameters intrinsic to efficient HMO metabolism (ability to support microbial growth, utility as substrates for fermentation, and production of organic acid metabolites) were measured to determine the impact of this syntrophic/competitive relationship. Unravelling HMO assimilation phenotypes and understanding the oligosaccharide structural complexity required to enrich specific beneficial bacterial communities allows for the development of targeted and personalised HMO-containing probiotic products with well-defined microbial, metabolic, and infant health outcomes.

Results
Bifidobacterium strains display divergent and strain-specific growth phenotypes, with enhanced growth performance during co-cultivation. Growth

profiles of Bifidobacterium bifidum
Rosell®-71 (R0071), Bifidobacterium infantis Rosell®-33 (R0033), Bifidobacterium breve M-16V, and Bifidobacterium infantis M-63 cultivated in mono-culture on HMO-supplemented media were evaluated and compared to the growth profiles of the strains when grown in a four strain co-culture on the same carbohydrate sources ( Fig. 1a, Fig. S1). The 2′-FL which had been eliminated during the size exclusion process was then resubstituted to give a powder representative of secretor HMO (designated S-HMO) ( Supplementary Fig. 1). In total, therefore, three HMO substrates were assessed in this study: 2′-FL (at 1.2 g/L), NS-HMO (at 3.8 g/L), and S-HMO (at 5 g/L) (compositions of the HMO powders are outlined in Supplementary Fig. S1). Most notably, significant growth differences were observed between the two B. infantis strains (Fig. 1a,b,c) despite these strains showing similar growth in MRS media with glucose and lactose as sole carbohydrate sources (data not shown). When grown on 2′-FL, strain M-63 demonstrated a significantly higher maximum OD 600 (p = 0.0004) and higher growth rate during the exponential phase (p = < 0.0001) of growth, when compared to strain R0033 (Fig. 1). Similar findings were noted during growth on NS-HMO (Fig. 1). Although R0033 and M-63 both grew vigorously on S-HMO (reaching final OD 600 values of 0.81 and 0.99, respectively), the observed growth rate of M-63 during the exponential phase was significantly higher than that of R0033 (p = 0.0023) (Fig. 1a). M-63, however, did display comparable growth profiles to R0071. Bifidobacterium bifidum R0071 reached a maximum OD 600 of 0.81 and 0.98 on NS-HMO and S-HMO, respectively ( Fig. 1).
Bifidobacterium breve M-16V did not exhibit any significant growth on 2′-FL ( Fig. 1) yet reached an OD 600 of 0.27 and 0.51 when grown on NS-HMO and S-HMO, respectively (Fig. 1b). Co-cultivation of this strain with R0071, R0033, and M-63 on 2′-FL (p < 0.0001), NS-HMO (p = 0.0033) or S-HMO (p = 0.0002) resulted in significant increases of M-16V cell numbers when compared to its growth in mono-culture (Fig. 1c). Overall, co-cultivation of all four Bifidobacterium strains together on S-HMO resulted in significantly higher cell numbers when compared to single strain cultivations (Fig. 1b). During growth on S-HMO co-cultivation resulted in a 27.7% increase in cell numbers when compared to the highest mono-culture fermentation of R0071 (p = 0.0096) (Fig. 1b). Quantitative PCR analysis targeting the 16S rRNA region allowed for quantification of cell numbers of each species within the co-culture, but did not distinguish between the two B. infantis strains. Nonetheless, qPCR analysis provided insights into the microbial dynamics in the co-culture system and revealed significant shifts in microbial distribution across the fermentation period. Despite significant increases in bacterial cell numbers, no change in the relative abundance of species within the co-culture occurred from 0 to 10 h. At the initiation of fermentation on S-HMO the co-culture contained 1.23 (± 0.41) × 10 6 bifidobacteria/mL B. bifidum, 1.38 (± 0.39) × 10 6 bifidobacteria/mL B. breve and 2.61 (± 0.46) × 10 6 bifidobacteria/mL B. infantis. No significant change had occurred in this population distribution by late exponential phase (10 h) with each species within the co-culture displaying similar growth rates (µ) and doubling times (T d ). During this period (0-10 h) B. bifidum, B. breve, and B. infantis grew at a rate of 0.32 ± 0.01, 0.3 ± 0.02, and 0.33 ± 0.04 h −1 respectively. However, between exponential (10 h) and stationary phase (24 h), significant shifts in microbial distribution were observed. While bacterial cell numbers of B. bifidum (p = 0.8476) and B. infantis (p = 0.9419) remained constant between 10 and 24 h fermentation, growth of B. breve (p = 0.0010) persisted. Therefore, after 24 h growth on S-HMO, the coculture contained 2.63 (± 0.95) × 10 7 bifidobacteria/mL B. bifidum, 7.57 (± 2.14) × 10 7 bifidobacteria/mL B. breve and 7.63 (± 1.36) × 10 7 bifidobacteria/ mL B. infantis. qPCR analysis showed an overrepresentation of B. breve at 24 h when compared to bifidobacterial numbers of B. bifidum and B. infantis. Analysis was calculated using technical triplicate data from biological triplicate experiments and data are means + /−SD. Univariate analysis of variance (ANOVA) and post-hoc Tukey tests were performed to determine the significant differences between the groups (# = p < 0.05, ## = p < 0.001 vs. co-culture) (* = p < 0.05, ** = p < 0.01, *** = p < 0.001, **** = p < 0.0001 between other strain combinations).  Fig. S3). However, significant growth of this strain in 3′-SL was observed when cultured in combination with B. bifidum R0071 ( Supplementary Fig. S3). The co-culture of R0071 and M-16V contained 3.11 (± 0.86) × 10 8 CFU/ mL after 24 h growth on 3′-SL (p < 0.0001 vs. either of the mono-culture situations) with B. breve M-16V accounting for 86.89 ± 4.84% of the total cell number proportion ( Supplementary Fig. S3).
Monosaccharide analysis using HPAEC-PAD revealed that large quantities of sialic acid are released during growth of B. bifidum strain R0071 on NS-HMO (118.8 mg/L) and S-HMO (134.6 mg/L) ( Supplementary Fig. S5). Significant quantities of sialic acid (24 mg/L in NS-HMO and 30 mg/L in S-HMO) were also detected in the M-63 cell-free supernatant after 10 h growth ( Supplementary Fig. S5). The levels of sialic acid produced following fermentation with B. bifidum R0071 were in stoichiometric accordance with levels of depleted sialylated HMO. In S-HMO, R0071 assimilated 0.46 ± 0.04 mM sialylated HMO, and released 0.44 ± 0.03 mM sialic acid (molar ratio of depleted sialylated HMO to sialic acid produced = 1.07 ± 0.13). This molar balance of depleted sialylated HMO and production of sialic acid was not observed during M-63 fermentation. In S-HMO, B. infantis M63 consumed 0.46 ± 0.04 mM sialylated HMO but produced only 0.1 ± 0.02 mM sialic acid (molar ratio of depleted sialylated HMO to sialic acid produced = 4.82 ± 1.03).
Glycoprofiling of B. bifidum R0071 and B. breve M-16V cultures with 3′-SL as sole carbohydrate source was performed to assess whether the sialic acid released by B. bifidum R0071 is utilised by other strains in a coculture system. In accordance with its growth on more complex HMO blends, M-16V was incapable of utilising 3′-SL ( Supplementary Fig. S3). Mono-culture growth of R0071 resulted in the consumption of 60.51% 3′-SL at 10 h and 100% 3′-SL at 24 h which corresponded to a release of 159.3 mg/mL (10 h) and 234.6 mg/mL (24 h) of sialic acid ( Supplementary Fig. S3). Co-cultivation of the two strains resulted in similar depletions of 3′-SL (69.91% at 10 h, 100% at 24 h) but significantly lower amounts of sialic acid (41.7 mg/mL at 10 h, 0 mg/mL at 24 h) ( Supplementary Fig. S3).  www.nature.com/scientificreports/ Across all strains and carbohydrate combinations assessed in this study, levels of lactose after 10 h growth were below the limit of detection. All residual lactose found in the HMO powders at initiation of fermentation (approximately 130 mg/L) and any lactose released from the breakdown of HMO into its constituent building blocks were consumed by the strain(s) by 10 h.

Production of metabolic end-products is dependent on strain combination and HMO substrate.
To assess the metabolic consequences of bifidobacterial growth on HMO, levels of the possible fermentative end-products acetate, lactate, formate, and 1,2-propanediol (1,2-PD) were evaluated. The absolute concentrations of these metabolites following 10 h fermentation are presented in Fig. 3 and Supplementary  Table S1. The central metabolic pathways employed in Bifidobacterium for the production of these organic acids is shown in Supplementary Fig. S6. Measurable levels of acetate, lactate, and formate were detected across all strain/HMO combinations, albeit with significant variations (Fig. 3). Consistent with increased bifidobacterial numbers detected during growth analysis, co-culture fermentation of the four bifidobacterial strains resulted in significantly higher total metabolites compared to mono-culture fermentations ( Fig. 3 and Supplementary  Table S1). In comparison to single-strain cultivation with M63, metabolite production increased by 45.4%, 13.6%, and 15.9% during co-cultivation on 2′-FL, NS-HMO, and S-HMO, respectively. On the whole, acetate, and lactate levels were lowest during metabolism of 2′-FL and highest in cultures grown in S-HMO. Production of acetate ranged from 2.36 mM during M-16V growth on 2′-FL to 25.6 mM during co-culture metabolism of S-HMO. Likewise, lowest concentrations of lactate were detected following M-16V growth on 2′-FL (0.21 mM) and highest during co-culture growth on S-HMO (17.29 mM). Notable variances were observed between metabolic profiles of the two B. infantis strains. B. infantis M-63 produced significantly higher amounts of acetate when compared with R0033 during metabolism of 2′-FL (7.76 mM vs. 3.94 mM, p = 0.0233), NS-HMO (19.87 mM vs. 3.08 mM, p < 0.0001), and S-HMO (23.11 mM vs. 11.55 mM, p < 0.000).
Conversely, variability in formate production between strains was less pronounced, ranging from 2.68 to 9.1 mM ( Fig. 3 and Supplementary Table S1). Overall, it is presumed that the concentrations of organic acids produced during bifidobacterial fermentation of carbohydrates are positively correlated with bacterial cell numbers. Pearson correlation and linear regression analysis was performed to assess the relationship between bacterial biomass and the subsequent yields of acetate, lactate, and formate (data not shown). Findings indicate that, across all three HMO substrates, acetate and lactate concentrations are positively correlated with bacterial cell numbers. However, no correlation between bacterial cell numbers and levels of organic acids were noted in the case of formic acid production.
Quantitative analysis of metabolite 1,2-PD using GC-MS revealed that production of 1,2-PD only occurred in the presence of the B. infantis strains and not during either B. bifidum or B. breve mono-culture fermentations,   Table 1). During growth on 2'-FL as sole carbon source, 1,2-PD production was only observed after fermentation with the four strain Bifidobacterium coculture, with 1.7 mM detected after 10 h growth (Fig. 3). Co-culture fermentation of NS-HMO and S-HMO also resulted in significant quantities of 1,2-PD being produced (2.4 mM and 3 mM, respectively) (Fig. 3). 1,2-PD was also detected in B. infantis R0033 culture supernatants on S-HMO (2.4 mM), and in M-63 culture supernatants when grown on NS-HMO (2.1 mM) and S-HMO (2.1 mM) (Fig. 3). Molar ratios of organic acids were determined to provide further insights into the metabolic trajectories during bifidobacterial HMO fermentation. These ratios are given in Table 1. For every 2 mol of hexose that enter into the Bifidobacteriaceae-specific bifid shunt metabolic pathway, 3 mol of acetic acid and 2 mol of lactic acid are produced (Supplementary Fig. S6). The theoretical ratio of 1.5 acetate: lactate was observed during B. bifidum R0071, B. infantis M-63, and co-culture fermentation across all HMO substrates (Table 1). In contrast, growth of B. infantis R0033 and B. breve M-16V on 2′-FL as sole carbon source significantly increased the acetate to lactate ratio to 7.4 ± 1.8 and 9.9 ± 3.4 (Table 1). Findings show that B. infantis R0033 and B. breve M-16V metabolism was characterized by an increase in formic acid production (relative to acetic and lactic acid production) ( Table 1). A significant increase in the ratios of formate: acetate and formate: lactate was observed for R0033 and M-16V when compared to all other strain combinations (Table 1).

Discussion
This study aimed to characterise the metabolic capability and co-operative metabolism of a community of four infant Bifidobacterium strains when grown on HMO. Since formula-fed infants lack exposure to these potentially health-promoting oligosaccharides and microbes, delivery of HMO and probiotic bifidobacteria in infant formula or supplements remains a priority 23 . The four strains included in this study were selected based on their proven efficacy in clinical studies [24][25][26] . For instance, formula-fed infants receiving a probiotic supplement containing B. bifidum R0071 and B. infantis R0033 in addition to a Lactobacillus strain, maintained higher faecal SIgA levels at the end of the four-week treatment period 24 . Bifidobacterium breve M-16V is also a commonly used probiotic strain in infants and has been shown to offer potential in protecting infants from developing the necrotising enterocolitis (NEC) and allergic diseases (reviewed by Wong et al. 2019 25 ). A mixture containing B. infantis M63 and two other probiotic strains (one being M-16V) was associated with better control of functional abdominal pain and improved quality of life when compared to the placebo in children with IBS 26 .
A key factor in determining microbial dynamics in the microbiota is the sharing and/or competition for nutrients 27 . In this context, microbe-microbe interactions can either positively or negatively influence the fitness of organisms present. The current study demonstrates that, during growth on HMO, cross-feeding interactions between multiple bifidobacterial strains increases overall cell numbers. This increase was hallmarked by notable shifts in microbial distribution from exponential to stationary phase. We found that under certain conditions (i.e., when fermentable carbohydrates were readily available) these Bifidobacterium strains stably coexist and no (sub)species appeared to have a considerable competitive advantage. However, over time, consumption of limiting nutrients and release of acidic metabolites appeared to shape the course of competition, thereby altering the relative abundance of species in the culture. No increase in bacterial cell numbers was observed for B. bifidum or B. infantis from exponential to stationary phases of growth during co-cultivation. Any increase in growth was due to continued proliferation of B. breve. This indicates that, despite demonstrating limited capacity to metabolize HMO in pure culture, the ability of M-16V to act as a 'scavenger species' and potential higher resistance to acidic conditions allows the strain to persist at the lowest resource level and become the competitive dominant in a shared environment. These findings appear to reflect the distribution of bifidobacterial species in the gut of breast-fed infants. Prototypical HMO-consumers B. bifidum and B. infantis, are often detected in low numbers in the feces of breast-fed infants, while B. longum and B. breve are regularly found as the dominant species in infant stools, even though they demonstrate minimal growth on HMO in vitro 28 .
Generally, it is assumed that B. breve has limited ability to metabolise complex HMO structures, but can grow on mono-and di-saccharide HMO degradants 29 . More recent studies have identified B. breve strains capable of metabolising a greater repertoire of HMO structures, including LNT and LNnT 30 , and those decorated with fucose and sialic residues 21 . Thus, the ability of B. breve to assimilate particular HMO is strain-dependent. In mono-culture B. breve M-16V was shown to exhibit rather poor growth on all assessed HMO substrates. Glycoprofiling revealed that M-16V consumed minimal amounts of 3-FL, 2′-FL, 3′-SL, 6′-SL, and LNnT, corroborating findings from previous in vitro growth studies 31 . We demonstrate here that M-16V is an efficient consumer of  www.nature.com/scientificreports/ lacto-N-tetraose (LNT), the central moiety of Type I HMO. Similar selective consumption patterns have also been observed in vivo during probiotic administration of B. breve M-16V to premature breast-fed infants 19,32 . Growth and sugar analysis presented here revealed notable strain-level differences in the ability of the two B. infantis probiotic strains to utilize HMO. Bifidobacterium infantis R0033 exhibited minimal growth on HMO and only demonstrated significant consumption of the fucosyllactoses after 10h of growth. In contrast, the probiotic strain B. infantis M-63 showed high levels of growth on 2′-FL, NS-HMO, and S-HMO. HPLC analysis of the resultant supernatant showed that the majority of HMO structures are degraded after 10 h fermentation with M-63, with preferential consumption of LNT and LNnT. Sakanaka et al. (2020), via genome data mining, found conservation of HMO-related genes (including GHs) across B. infantis to be very high 33 . It is assumed, therefore, that all B. infantis strains behave in the same manner when utilizing HMO. However, a recent study has reported genetic differences in HMO utilization clusters among B. infantis strains which has significant implications on fitness and performance of B. infantis strains in culture and in the infant gut 34 . Comparative genome analysis of commercial B. infantis probiotics revealed two main variants; one which harboured the full repertoire of HMO utilization genes and another variant which lacks an ABC-transporter needed for binding of core HMO structures (such as LNT and LNnT) 34 . We hypothesise that similar genotypic variations exist between B. infantis strains M-63 and R0033 which may have implications for their fitness and survival in the infant gut.

2'-FL NS-HMO S-HMO 2'-FL NS-HMO S-HMO 2'-FL NS-HMO S-HMO
In contrast to the intracellular consumption strategies used by B. infantis, B. bifidum employs cell wallanchored secretory glycosidases which hydrolyse the HMO extra-cellularly releasing mono-and di-saccharides, which are then internalized for metabolism and growth 19,33 . Here, we report that B. bifidum R0071 grows well on HMO and readily assimilates most HMO structures including type I structures, type II structures and sialylated structures. R0071 demonstrated preference for consumption of 3′-SL over 6′-SL. Both these structures are trisaccharide HMO that have the same molecular composition and only differ in the attachment of sialic acid residues around the lactose core. These findings highlight that the specific structure of the HMO isomers is biologically relevant. Bifidobacterium bifidum R0071 also demonstrated substantial consumption of 2′-FL, with the majority of the substrate degraded by late exponential phase. Remarkably, dramatic increases in the levels of 3-FL were observed during R0071 fermentation of HMO. To the best of our knowledge, this is the first time that an increase in the level of 3-FL during bifidobacterial growth on HMO has been reported. We speculate that this increase in 3-FL can be accounted for by differences in the catalytic activity of 1,2-α-L-fucosidase (AfcA) and 1,3/4-α-L-fucosidase (AfcB) as previously described 35 . It is plausible to hypothesise that the rapid action of AfcA cleaves the α-1,2 linked fucose from DFL to release 3-FL into the supernatant, which subsequently remains in the culture due to the low catalytic activity of AfcB. To test this hypothesis, growth experiments on DFL alone were performed. Glycoprofiling of culture supernatants confirmed that 3-FL was released during growth of R0071. However, this increase in 3-FL levels was found to be transient, suggesting that the AfcB harboured by R0071 begins catalysing α1,3-fucosylated HMO once other preferred substrates have been assimilated. Interestingly, when R0071 was co-cultured on the same media with B. infantis R0033, notable increases in bacterial cell numbers were observed. This suggests that, despite its transient nature, 3-FL released by R0071, becomes available as a metabolic cross-feeding substrate for other gut commensals, particularly B. infantis strains which demonstrate adaptation for uptake and assimilation of the major fucosylated HMO 36 .
Quantification of HMO degradants also revealed that B. bifidum R0071 produced sialic acid in stoichiometric accordance with consumption of sialylated HMO. No increase in the levels of lactose was observed after 10 h growth, which demonstrates that B. bifidum R0071 hydrolyses HMO in order to access the disaccharide moieties (such as lactose and lacto-N-biose) which it then internalises and utilizes for growth. Findings from previous studies suggest that extracellular breakdown of HMO and subsequent liberation of monosaccharides by B. bifidum strains can facilitate cross-feeding within microbial communities [37][38][39] . In particular, studies have highlighted the existence of syntrophic relationships between B. bifidum and B. breve strains 38,40 . We hypothesise that release of HMO degradants by B. bifidum R0071 facilitated growth of B. breve M-16V during co-culture growth, with syntrophy between the strains prevailing in the presence of other infant-derived bifidobacteria. When M-16V was co-cultured with B. bifidum R0071 on 3′-SL (without the B. infantis strains present) its growth was considerably enhanced. These findings highlight that B. bifidum, which is generally known to be the minority species in the early gut microbiota, can have a major influence on the prevalence of other species within the Bifidobacterium genus. The significance of B. bifidum altruistic behaviour in shared environments has been highlighted previously, with studies showing that B. bifidum supplementation results in enhanced HMO consumption and increased dominance of bifidobacteria 35,41 .
Here we demonstrate that all four Bifidobacterium, individually, have the metabolic machinery required to capture and metabolize HMO into complex health-promoting metabolites, with enhanced production during co-fermentation. We found detectable levels of lactate, acetate, and formate in all cultures. This has potential implications for infant nutritional and health outcomes since production of these key metabolites is critical to the delivery of fundamental ecosystem services [42][43][44][45][46][47] . In general, higher bacterial biomass resulted in increased secretion of acetate and lactate and correlated with superior rates of HMO consumption. It has been proposed that a theoretical molar ratio of 1.5 acetate to 1 lactate is produced during efficient metabolism of HMO, although other studies have shown that this theoretical ratio is not always obtained 48 . The variation in the theoretical ratio is explained by the production of other sugar metabolites, such as formate and ethanol, which limit the production of lactate. This theoretical ratio was only achieved during B. infantis M-63 and co-culture metabolism of S-HMO. In contrast, inefficient hydrolysis of 2'-FL by B. infantis R0033 and B. breve M-16V shifted metabolism towards greater acetate production resulting in higher average ratios. Studies have demonstrated that increased levels of acetate and formate allow Bifidobacterium with atypical HMO consumption phenotypes to balance their reduction-oxidation levels and produce enough energy to continue growing 49 . Formate is produced when pyruvate is shunted away from lactate production, and towards the production of acetyl-CoA 49 . When more than 50% acetyl-CoA is converted to acetate by B. longum, the theoretical formate: acetate ratio is 0.4 50  www.nature.com/scientificreports/ was achieved throughout fermentation with the co-culture, irrespective of carbohydrate source. Here, B. infantis R0033 and B. breve M-16V surpassed that threshold, suggesting that there is a shift to produce more acetyl-CoA than lactate from pyruvate. It is hypothesised that this metabolic shift to favour formate secretion is to prioritize ATP production through a branch in the bifid shunt. We speculate that by shifting metabolic pathways towards formate production, M-16V and R0033 can generate NADH from NAD + without having to use NADH to reduce pyruvate to lactate, resulting in a net positive NADH. Unlike glucose, galactose, sialic acid, and N-acetylglucosamine derivatives, fucose is not predicted to enter the bifid shunt central metabolic pathway. Instead, it is hypothesized that fucose derivatives enter in a separate fermentative pathway, which results in the production of an equimolar ratio of pyruvate and intermediary metabolite 1,2-PD per mole of fucose consumed 51 . Here it was found that co-fermentation enhanced production of 1,2-PD. During growth on 2′-FL, the co-culture and B. infantis M-63 exhibited similar levels of growth but 1,2-PD was only detected in the co-culture system. Therefore, factors other than biomass accumulation may play a role in production of this compound. It is possible that fucose transport is repressed in the presence of high concentrations of efficient substrates. James et al. (2019) putatively linked a LacI-family regulator in Bifidobacterium kashowanohense (fumR) to fucose metabolism 52 . Subsequently, Dedon et al. (2020) determined that B. infantis ATCC 15,697 possesses a homolog of this fumR and hypothesised that this putative repressor may restrict fucose metabolism in the presence of other efficiently utilized sugars. Therefore, it is possible that increased competition and demand for lactose, glucose, and galactose during co-culture drives metabolism of fucose and subsequent production of 1,2-PD. Enhanced production of 1,2-PD may also occur as a consequence of the trophic chain from the release of fucose during extracellular metabolism of HMO by B. bifidum R0071 53 . Cross-feeding based on the metabolite 1,2-PD has been proposed to have an important role in the establishment of trophic interactions among gut symbionts 54,55 . Accordingly, it has been reported that the infant-associated Eubacterium hallii metabolizes 1,2-PD to produce short chain fatty acids butyrate, formate, and propionate 56,57 . The shift to prioritize 1,2-PD production may therefore facilitate trophic relationships within the infant gut microbiome.

Conclusion
Deciphering relationships between early colonizing bacterial species and specific HMO types allows for the development of health-promoting products that more closely resemble natural infant microbiota interactions. The findings presented here contribute to our knowledge of HMO-microbe interactions and demonstrate the potential for synbiotic combinations of pre-and pro-biotics. The work demonstrates that small differences in the molecular structure of HMO can have considerable impacts on their biological efficacy and on the catalytic ability of bifidobacterial strains. The variances in HMO metabolism observed in this study highlight the importance of probiotic strain specificity for efficient metabolism of HMO. Genomic and proteomic analyses are required to further understand the functional implications of genotypic variations in HMO-utilization clusters. Development of "HMObiotic" or HMO-containing probiotic products should account for these variations and combine HMO with their preferred pairing of strains. Overall, we conclude that genotype, but also phenotype, of strains should be considered when selecting a strain for probiotic use, as variations in HMO utilization may determine fitness and performance in the gut. We demonstrate the potential benefits for the use of multi-strain bifidobacterial probiotics since co-operation between strains may expand HMO acquisition capabilities and help shift bacterial communities closer towards those observed in breast-fed infants.

Carbohydrate substrates. Generation of human milk oligosaccharide enriched fraction.
Oligosaccharide fractions were isolated from human milk as previously described 59 . In short, pooled human milk samples (of unknown secretor status) were kindly donated by Irvinestown Human Milk Bank (Co. Fermanagh, Ireland) and stored at − 80 °C on arrival. Lipids and proteins were removed before applying samples to a BioGel P2 size exclusion column (92 × 5 cm; Bio-Rad Laboratories, Inc., Hercules, CA, USA) to separate lactose and HMO components. High pH anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD) was used, as described below, to quantify lactose in the eluted fractions while a Pierce colorimetric peptide assay (Thermo Fisher Scientific, Reinach, Switzerland) was used to measure levels of peptides. Peptide-free and lowtrace lactose (< 80 mg/L) fractions were pooled and freeze-dried and the resultant HMO-enriched fraction was stored at 4 °C prior to use in experiments.
Preparation of HMO blends. Three HMO powders were assessed in this study; a HMO-enriched fraction with and without 2′-FL (Friesland Campina, Amersfoort, Netherlands) was used to mimic non-secretor (NS-HMO) and secretor (S-HMO) human milk oligosaccharide fractions, respectively. 2′-FL alone was used as a control ( Table 2). Stock suspensions (20% wt/v) of 2′-FL, NS-HMO, and S-HMO were prepared in MQ-water and sterile filtered (0.2 µm, Chromafil, Macherey-Nagel, Düren, Germany) Oligosaccharide substrates were used in fermentation studies at concentrations outlined in Pure culture and co-culture fermentations. Growth medium. Modifications were made to Man-Rogosa-Sharpe (MRS) 61 medium to generate a basal growth medium which contained no carbohydrate sources as previously described 62 . Media was sterilized by autoclaving at 121 °C for 15 min. For use in fermentation studies, cfMRS medium was supplemented with sterile oligosaccharide solutions at concentrations outlined in Table 2. Growth analysis. Optical density readings. Variability in bacterial biomass was assessed throughout fermentation by measuring optical density at 600 nm (OD 600 nm ). Growth analysis was performed on cultures at 0 h, 6 h (early exponential), 10 h (late exponential), 20 h (early stationary), and 24 h (late stationary). At each time point, 200 µl of culture was transferred to the wells of 96 well microtiter plates in an anaerobic chamber and the OD values determined using a Synergy 2 plate reader (BioTek Instruments, Inc., Vermont, US). Turbidity of cultures was assumed to correlate with numbers of bacterial cells within solution.

Preparation of inoculum and initiation
Quantitative PCR. Possible enhancement or reduction of bifidobacterial growth as a consequence of co-cultivation with other strains was evaluated through absolute quantification of bacterial numbers by quantitative PCR (qPCR). One ml samples were harvested from cultures at 0 h, 10 h, and 24 h. Bacterial cells were pelleted at 12,000×g for 2 min prior to the addition of the lysis buffer. Total bacterial DNA was isolated using GenElute Bacterial Genomic DNA Kit (Merck) according to manufacturer's instructions. The resultant DNA was quantified by a Nanodrop 2000 Spectrophotometer. Total bifidobacterial numbers were quantified on a 480 Lightcycler platform (Roche Applied Science, Penzberg, Germany) using the following programme: 95 °C for 5 min followed by 40 cycles of 95 °C for 10 s, 58 °C for 20 s and 72 °C for 30 s. The fluorescent product was detected at the last step of each cycle. Following amplification, melting temperature analysis of PCR products was performed to determine the specificity of the PCR. The melting curves were obtained by slow heating at 0. cfu/mL = no. of colonies × dilution factor volume of culture plated The copy-number of a given strain in co-culture was evaluated by comparing the cycle threshold (Ct) values with those from a standard curve of known copy number. Standard curves (10 9 -10 2 copy number/µL) were established by performing 1 in 10 serial dilutions of 16S rDNA of the specified bifidobacterial strain (in the case of B. infantis strains, M-63 was used as the reference rDNA). The cell numbers of strains within co-culture (denoted by bifidobacteria/mL) were then deduced by comparing these copy number values with copy numbers of standards with known cell numbers (as determined by viable count assessment). Cell numbers were calculated according to the formula 63 : where C/μl = Copy number/μl, TV = template volume, TCN = total copy number of the standard used, T cfu/ ml = total cell number of standard used, and bifidobacteria/mL(S) = cell number of test sample. Samples were run in triplicate, while standards and negative controls (where template DNA was replaced with PCR grade water) were run in duplicate.
Growth rate (µ) of each subspecies within cultures was calculated using the following equation: where X 2 and X 1 are the cell densities at t 2 and t 1 . Mean doubling time (T d ) was calculated as: Chemical analysis of culture supernatants. For chromatographic analysis of cultures, supernatants were extracted at late exponential phase (10 h) and filtered using 0.2 µm nylon syringe filters (Chromafil, Macherey-Nagel, Düren, Germany) to remove bacterial cell debris.
Quantification of HMO utilization. The concentrations of oligosaccharide in culture after 10 h microbial fermentation were determined by HPAEC-PAD using a Dionex ICS-3000 Series system (Dionex Corporation, Sunnyvale, CA, USA). Samples were diluted 100-fold with MQ-Water, and 20 µl injected onto a CarboPac PA100 column (250 × 4 mm) (Thermo Fisher Scientific) equipped with a guard column. 100 mM NaOH (A) and 100 mM NaOH plus 500 mM Na-acetate (B) were used as eluents at a flow rate of 1.0 mL/min. The column temperature and detector were maintained at 32 °C. HMO were separated for 22 min using an eluent gradient as follows: 95% A and 5% B for 3 min, 88% A and 12% B for 10 min, 84% A and 16% B for 4 min and hold time of 3 min, followed by 82% A and 18% B for 2 min. At 22.1 min the gradient immediately changes to 50% A and 50% B for 8 min to regenerate the column and then back to 95% A and 5% B for a further 15 min to re-equilibrate the column. The eluent was monitored by pulsed amperometric detection (PAD). The eluted sugars were identified from their elution times relative to those of external standards: Lactose, 3-FL, 2′-FL, LNnT, LNnH, LNT, LNH, sialic acid, sialyllacto-N-tetraose a (LSTa), LSTc, 6′-SL, 3′-SL, disialyllacto-N-tetraose (DSLNT). Utilization of HMO was calculated as % consumption versus the uninoculated control.
Quantification of organic acid metabolites. To examine organic acid production, supernatants were extracted after 10 h fermentation and filtered as above. Levels of organic acid metabolites were then quantified by HPLC using a Waters Alliance Separations module e2695 coupled to a Waters 2414 refractive index (RI) detector (Waters, Milford MA, USA). Samples or standards at a volume of 20 µl were injected on to an Aminex HPX 87C fixed ion resin column (300 × 7.8 mm) operated at 60 °C. The samples were eluted with H 2 SO 4 (0.005 N) at a flow rate of 0.6 mL/ min. Sample detection was performed by comparing retention times of standards. Analytical grade acetic acid, lactic acid, and formic acid supplied by Merck were used as standards.
Quantification of 1,2-propanediol. To assess whether co-culture fermentation of HMO increased 1,2-PD production when compared to single culture fermentation, samples were sent to MS-Omics ApS, Frederiksberg, Denmark for analysis using GC-MS 64 . A mixed pooled sample (QC sample) was created by taking an aliquot from each sample. This sample was analyzed with regular intervals throughout the sequence. For 1,2-PD analysis, samples were acidified using hydrochloric acid, and deuterium-labeled internal standards were added. Analysis was performed using a high-polarity column (ZebronTM ZB-FFAP, GC Cap. Column 30 m × 0.25 mm × 0.25 μm) installed in a GC (7890B, Agilent) coupled with a quadrupole detector (59977B, Agilent). Raw data was converted to netCDF format using Chemstation (Agilent), before the data was imported and processed in Matlab R2014b (Mathworks, Inc.) using the PARADISe software.

Statistical analysis
Statistical analysis was calculated using technical triplicate data from biological triplicate experiments. Graphs were created using in PRISM v8 (GraphPad, La Jolla, CA). Univariate analysis of variance (ANOVA) and post-hoc Tukey tests were performed to determine the significant differences between the groups for growth, oligosaccharide consumption and metabolite production. Statistical significance was accepted as p ≤ 0.05.