The Effect of Human Milk Oligosaccharides and Bifidobacterium longum subspecies infantis Bi-26 on Simulated Infant Gut Microbiome and Metabolites

Human milk oligosaccharides (HMOs) shape the developing infant gut microbiota. In this study, a semi-continuous colon simulator was used to evaluate the effect of 2 HMOs—2′-fucosyllactose (2′-FL) and 3-fucosyllactose (3-FL)—on the composition of infant faecal microbiota and microbial metabolites. The simulations were performed with and without a probiotic Bifidobacterium longum subspecies infantis Bi-26 (Bi-26) and compared with a control that lacked an additional carbon source. The treatments with HMOs decreased α-diversity and increased Bifidobacterium species versus the control, but the Bifidobacterium species differed between simulations. The levels of acetic acid and the sum of all short-chain fatty acids (SCFAs) trended toward an increase with 2′-FL, as did lactic acid with 2′-FL and 3-FL, compared with control. A clear correlation was seen between the consumption of HMOs and the increase in SCFAs (−0.72) and SCFAs + lactic acid (−0.77), whereas the correlation between HMO consumption and higher total bifidobacterial numbers was moderate (−0.46). Bi-26 decreased propionic acid levels with 2′-FL. In conclusion, whereas infant faecal microbiota varied between infant donors, the addition of 2′-FL and 3-FL, alone or in combination, increased the relative abundance and numbers Bifidobacterium species in the semi-continuous colon simulation model, correlating with the production of microbial metabolites. These findings may suggest that HMOs and probiotics benefit the developing infant gut microbiota.


Introduction
The establishment of the microbiota in early life also shapes one's health later in life [1][2][3]. Several factors, such as delivery mode, diet, medication, and environment, influence the composition and numbers of the developing microbiota in infants [2,4]. As an abundant genus in infancy, Bifidobacterium is critical to the development of the infant gut microbiota, in contrast to the adult microbiota, which primarily harbours other bacteria from the phyla Firmicutes and Bacteroidetes [5]. The ability of certain Bifidobacterium species to utilise oligosaccharides from human milk increases their colonisation and abundance [5,6]. Bifidobacterium species also support host-microbiota homeostasis [1].
In addition to live commensal bacteria, human breast milk contains bioactive compounds, such as human milk oligosaccharides (HMOs), a group of structurally diverse carbohydrates [2,7]. HMOs are not hydrolysed by intestinal enzymes in the upper gastrointestinal tract [8] and promote the growth of certain bacteria in the colon, such as Figure 1. Schematic of the EnteroMIX ® colon simulator and overview of the study products. (a) A diagram of a single unit of the EnteroMIX ® colon simulator system. Vessel 1 (V1, proximal) to V4 (distal) represent various parts of the colon. Nitrogen was used to maintain anaerobiosis and as a carrier gas for ammonia and liquid transfers. The system was computer-controlled (figure modified from [37]). (b) An overview of the treatments in a single simulation. Three simulations were performed. Control = no FLs; 2′-FL = 2′-fucosyllactose; 3-FL = 3-fucosyllactose; Bi-26 = Bifidobacterium longum subspecies infantis Bi-26.

Quantification of Fucose, 2′-FL and 3-FL
Fucose, 2′-FL, and 3-FL were quantified from the simulation units to which HMOs were added as per Salli et al. [26] with modifications. Standard solutions of fucose (Sigma-Aldrich, St. Louis, MO, USA), 2′-FL, and 3-FL were prepared in water to concentrations of 80, 60, 40, 20, and 10 mg/l and stored at +4 °C. Sample solutions were centrifuged at 16,000× g for 5 min; then, 50 μL of the supernatant and 200 μL of ethanol were mixed in a microcentrifuge tube and incubated at +4 °C for 30 min. After centrifugation at 16,000× g for 5 min, 200 μL of the supernatant was evaporated to dryness at 30 °C under stream of nitrogen, and the solid residue was dissolved in 2000 μL of water and filtered. Separation and detection of the analytes were performed using high-performance anion-exchange chromatography as previously described [26]. Retention times of fucose, 2′-FL, and 3-FL were 6.3 min, 26 min, and 18 min, respectively. (a) A diagram of a single unit of the EnteroMIX ® colon simulator system. Vessel 1 (V1, proximal) to V4 (distal) represent various parts of the colon. Nitrogen was used to maintain anaerobiosis and as a carrier gas for ammonia and liquid transfers. The system was computer-controlled (figure modified from [37]). (b) An overview of the treatments in a single simulation. Three simulations were performed. Control = no FLs; 2 -FL = 2 -fucosyllactose; 3-FL = 3-fucosyllactose; Bi-26 = Bifidobacterium longum subspecies infantis Bi- 26. Before the simulation began, 10 mL of individual inoculum was pumped into the first vessel, mixed, transferred to the second vessel, mixed, transferred to the third vessel, mixed, transferred to the fourth vessel, and mixed; the effluent was discarded. In the units with B. infantis Bi-26 (ATCC SD6720), 1 mL of an overnight culture of B. infantis Bi-26 (10 9 cells/mL) was added to 9 mL of inoculum (~1 × 10 10 cells/mL); B. infantis Bi-26 thus constituted approximately 1% of the microbes inoculated in the simulator. Samples from the inocula with and without B. infantis Bi-26 were drawn to determine their microbial composition.
The test products (2 -FL, 3-FL, 2 -FL + 3-FL) or controls (without added carbohydrates) were fed to the simulator system at 3 h intervals during the simulation, over a total of 48 h.
Then, the samples were collected from the vessels, and the composition of the simulated microbiota and microbial metabolites was analysed.

Quantification of Fucose, 2 -FL and 3-FL
Fucose, 2 -FL, and 3-FL were quantified from the simulation units to which HMOs were added as per Salli et al. [26] with modifications. Standard solutions of fucose (Sigma-Aldrich, St. Louis, MO, USA), 2 -FL, and 3-FL were prepared in water to concentrations of 80, 60, 40, 20, and 10 mg/l and stored at +4 • C. Sample solutions were centrifuged at 16,000× g for 5 min; then, 50 µL of the supernatant and 200 µL of ethanol were mixed in a microcentrifuge tube and incubated at +4 • C for 30 min. After centrifugation at 16,000× g for 5 min, 200 µL of the supernatant was evaporated to dryness at 30 • C under stream of nitrogen, and the solid residue was dissolved in 2000 µL of water and filtered. Separation and detection of the analytes were performed using high-performance anion-exchange chromatography as previously described [26]. Retention times of fucose, 2 -FL, and 3-FL were 6.3 min, 26 min, and 18 min, respectively.

Total Bacterial Cell Counts
Total bacterial cell counts were determined by flow cytometry in samples that were fixed with 4% formaldehyde, as described [39].

Quantitative Polymerase Chain Reaction (qPCR)
Microbial DNA was extracted from the colon simulation samples, and total bifidobacterial numbers were analysed by real-time quantitative polymerase chain reaction (qPCR) as previously reported [26,36,40]. B. infantis was quantified by qPCR using Taq-Man and Applied Biosystems Real-Time PCR equipment and software (ABI 7500 FAST, Applied Biosystems, Foster City, CA, USA) with the Bi26_F (400 nM GTCACGATGTCTC-CTTTGATATCAGCATG) and Bi26_R primers (400 nM CCTTTTGCGTCTCCCCCG) and the Bi26_P probe (200 nM, TCATTCATTGTAGTGGCGATCACCGTTACC). The annealing step for B. infantis was 60 • C for 30 s. Standard curves, consisting of 10-fold dilutions of target species DNA, were used for the quantification.

Microbial Composition by Barcoded 16S rRNA Amplicon Sequencing
The V4 variable region of the 16S rRNA gene was PCR-amplified from donor inoculum samples and control and the treated samples after 48 h of simulation, as previously described [41]. The amplicon pool was sequenced on the Illumina MiSeq system with 2 × 250 bp reads (Roy J. Carver Biotechnology Center, University of Illinois Urbana-Champaign, Champaign, IL, USA) and analysed using the Quantitative Insights Into Microbial Ecology pipeline (QIIME2, v.2018.6) [41,42]. In brief, the sequences were demultiplexed, and DADA2 was used to denoise and dereplicate the sequences. Representative amplicon sequence variants (ASVs) were assigned taxonomy against the Greengenes database (v. 13.8) [43]. Taxa compositions were reported as relative abundance (% of total sequences).

Analysis of Microbial Metabolites
The concentrations of short-chain fatty acids (SCFAs), lactic acid, and branched-chain fatty acids (BCFAs) in infant colon simulation samples were analysed by gas chromatography, as described by Ouwehand et al. [44].

Statistical Analysis
Alpha diversity comparisons were calculated in QIIME2 for the Phylogenetic Diversity (PD) Whole Tree metric [45] using an ASV table, rarefied at a sequence depth of 11,672. The main effect of treatment was analysed by Kruskal-Wallis test and corrected by the Benjamini-Hochberg false discovery rate (FDR); subsequent pairwise comparisons were conducted by Wilcoxon rank-sum test [46]. Beta diversity was calculated using weighted UniFrac values [47] and visualised by principal coordinates analysis (PCoA) in QIIME2.
Differentially abundant taxa (>0.1% abundance) were determined by Kruskal-Wallis test, and p-values ≤ 0.05 were reported after adjustment using the FDR.
The remaining analyses were performed in R [48].
To determine the effects of B. infantis Bi-26 and HMOs on total bifidobacterial numbers by qPCR, a linear model with Gaussian error was used. First, the response was log2-transformed to better fit the modelling requirements, and a second-order term was included to account for nonlinear curves. Through an automated model selection process, the parameters were narrowed down to vessel, vessel squared, treatment group, B. infantis Bi-26, and the slope that was associated with B. infantis Bi-26, all of which were included in the model. The effect of B. infantis Bi-26 and the treatments on B. infantis were also analysed in a somewhat similar manner using log2-transformed response and a fixed effect model with terms for donor, treatment, and interaction between donor and vessel (i.e., donor-wise slope).
Differences in metabolite concentration curves across all vessels were analysed using nonparametric and robust methods by Brunner et al. [49] implemented in the R package nparLD [50]. This method allows us to analyse the variably shaped curves in a consistent fashion with minimal distribution assumptions. The effects of the treatment versus control on metabolites in the pooled vessels were analysed by student's t-test. All p-values were corrected for FDR using the Benjamini-Hochberg method [46]. p-values of 0.05 or less were considered statistically significant.
The dependence between changes in 2 -FL, 3-FL and fucose, total bifidobacterial, and SCFA and BCFA counts was analysed by non-parametric Spearman correlation. The correlation estimates and their 95% confidence intervals were obtained using bias-corrected and accelerated bootstrap using 5000 samples. For illustration purposes, local polynomial regression fitting was used.

Donor Demographics
Of the three faecal sample donors, Donors I and II were exclusively breast-fed, and Donor III was given breast milk and formula. None of the donors had begun consuming solid foods. Donor II was delivered by caesarean section, and donors I and II were delivered vaginally; all three were given vitamin D supplements, and Donors I and II were administered probiotics. When donating the faecal samples, Donor I was 3 months, Donor II was 3,5-4 months, and Donor III was 1 month old.

Fermentation of HMOs during Colon Simulation
The utilisation of 2 -FL and 3-FL by complex bacterial communities varied between simulations (Table 1). Table 1 also reports the fucose levels.
The addition of B. infantis Bi-26 resulted in more efficient 2 -FL utilisation in the simulations with Donor III. In simulations with Donors I and II, 2 -FL was utilised in the upstream vessels without B. infantis Bi-26. Conversely, the addition of B. infantis Bi-26 resulted in earlier and later 3-FL utilisation in the simulations with Donors II and I, respectively; this utilisation did not change considerably with Donor III. When 2 -FL and 3-FL were combined, their utilisation by the microbiota was similar in the respective simulations; with this combination, 2 -FL was used in downstream vessels only in simulation with Donor II. Overall, all three simulations consumed all 2 -FL prior to vessel 3; in the simulation with Donor III, this pattern was observed only with B. infantis Bi-26. In general, 3-FL was utilised before vessel 4, which occurred in simulation II only with B. infantis Bi-26.
After complete utilisation of HMOs, the amount of fucose, a downstream metabolite, was undetectable. However, when combining all simulations, the addition of B. infantis Bi-26 had a statistically significant difference on fucose levels in treatments with 2 -FL (p = 0.001), based on nonparametric analysis. The difference in simulations with 2 -FL was due to the addition of B. infantis Bi-26, which was different in three simulations (Table 1). By barcoded 16S rRNA amplicon sequencing, we noted clear differences in microbial populations between the three donor inocula ( Figure 2 and Supplementary Table S1). Actinobacteria and Bacteroidetes were the most abundant phyla in all three donors (Figure 2a), totalling 70% in relative abundance; Firmicutes constituted between 10% and 20%, versus Proteobacteria at under 6% relative abundance.  Table S1). Actinobacteria and Bacteroidetes were the most abundant phyla in all three donors ( Figure  2a), totalling 70% in relative abundance; Firmicutes constituted between 10% and 20%, versus Proteobacteria at under 6% relative abundance. At the genus/species level (Figure 2b), only the inoculum from Donor I contained Bifidobacterium adolescentis, Collinsella aerofaciens, and the genus Blautia, at relative abundances of ~27%, 24%, and 10%, respectively. The microbiota composition in the inoculum from Donor II differed from the other two donors, based on the presence of Veillonella parvula and the genus Enterococcus at relative abundance of ~0.8% for both. It also had the highest relative abundance of the genus Bacteroides (47%; species unidentified but determined not to be Bacteroides fragilis, Bacteroides ovatus, or Bacteroides uniformis).

Alpha and Beta Diversity of Simulated Microbiota
There was no difference in alpha (phylogenetic) diversity between donors (p > 0 ( Figure 3a). Combining all three simulations, alpha diversity was greater in the cont simulation compared with the treatments with HMOs (p < 0.05), and there was no diff ence between HMO treatment groups (p > 0.1) (Figure 3b). When comparing all stu groups, the control group with B. infantis Bi-26 was significantly more diverse than other groups (p < 0.05, Supplementary Figure S1); however, alpha diversity did not dif between samples with and without B. infantis Bi-26 (p > 0.1, Supplementary Figure S2) Figure 3c shows the beta diversity (weighted Unifrac metric) of the samples by tre ment, with or without B. infantis Bi-26, by principal coordinates analysis (PCoA). At the genus/species level (Figure 2b), only the inoculum from Donor I contained Bifidobacterium adolescentis, Collinsella aerofaciens, and the genus Blautia, at relative abundances of~27%, 24%, and 10%, respectively. The microbiota composition in the inoculum from Donor II differed from the other two donors, based on the presence of Veillonella parvula and the genus Enterococcus at relative abundance of~0.8% for both. It also had the highest relative abundance of the genus Bacteroides (47%; species unidentified but determined not to be Bacteroides fragilis, Bacteroides ovatus, or Bacteroides uniformis).

Alpha and Beta Diversity of Simulated Microbiota
There was no difference in alpha (phylogenetic) diversity between donors (p > 0.1) (Figure 3a). Combining all three simulations, alpha diversity was greater in the control simulation compared with the treatments with HMOs (p < 0.05), and there was no difference between HMO treatment groups (p > 0.1) (Figure 3b). When comparing all study groups, the control group with B. infantis Bi-26 was significantly more diverse than the other groups (p < 0.05, Supplementary Figure S1); however, alpha diversity did not differ between samples with and without B. infantis Bi-26 (p > 0.1, Supplementary Figure S2).
The effect of HMOs was evaluated, pooling all three simulations. All HMO treatments, when the vessels were combined, increased the Actinobacteria abundance compared with the control (p = 0.004), whereas Bacteroidetes and Proteobacteria relative abundance were higher in the latter (p = 0.006 and p = 0.031, respectively). At the species level, B. longum/breve was the only species that increased with HMOs (p = 0.053); conversely, Leucobacter spp., Enterobacteriaceae spp., Rummeliibacillus spp., and Enterococcus spp. were higher in the control (p = 0.034 for Leucobacter spp., p = 0.048 Enterobacteriaceae spp. and p = 0.053 for others).
Because the microbiota composition varied between simulations with different donors, we have presented the results separately for each simulation. Figure 4 shows the microbial composition from the individual simulations and various treatments at the phylum ( Figure 4a) and species levels (Figure 4b), with the vessels combined, and individual vessels at the species level (Figure 4c). At the species level, the most prominent change in the simulation with Donor I was the increase in relative abundance of B. longum/breve, B. catenulatum/gallicum, and Lacticaseibacillus casei/zeae on treatment with HMOs versus the control. In addition, in this simulation, the relative abundance of Veillonella dispar was higher when B. infantis Bi-26 was not added. In the simulation with Donor II with B.  Figure 3c shows the beta diversity (weighted Unifrac metric) of the samples by treatment, with or without B. infantis Bi-26, by principal coordinates analysis (PCoA).
The effect of HMOs was evaluated, pooling all three simulations. All HMO treatments, when the vessels were combined, increased the Actinobacteria abundance compared with the control (p = 0.004), whereas Bacteroidetes and Proteobacteria relative abundance were higher in the latter (p = 0.006 and p = 0.031, respectively). At the species level, B. longum/breve was the only species that increased with HMOs (p = 0.053); conversely, Leucobacter spp., Enterobacteriaceae spp., Rummeliibacillus spp., and Enterococcus spp. were higher in the control (p = 0.034 for Leucobacter spp., p = 0.048 Enterobacteriaceae spp. and p = 0.053 for others).
Because the microbiota composition varied between simulations with different donors, we have presented the results separately for each simulation. Figure 4 shows the microbial composition from the individual simulations and various treatments at the phylum (Figure 4a) and species levels (Figure 4b), with the vessels combined, and individual vessels at the species level (Figure 4c). At the species level, the most prominent change in the simulation with Donor I was the increase in relative abundance of B. longum/breve, B. catenulatum/gallicum, and Lacticaseibacillus casei/zeae on treatment with HMOs versus the control. In addition, in this simulation, the relative abundance of Veillonella dispar was higher when B. infantis Bi-26 was not added. In the simulation with Donor II with B. infantis Bi-26, the relative abundance of B. longum/breve rose with 2 -FL and 3-FL, and Bacteroides fragilis decreased compared with control. In the simulation with Donor II without B. infantis Bi-26 the relative abundance of B. longum/breve increased with 2 -FL and 2 -FL + 3-FL, while with 3-FL it did not. The relative abundance of B. fragilis increased in downstream vessels with 2 -FL + 3-FL. In the simulation with Donor III, relative abundance of B. catenulatum/gallicum increased between HMO treatments and control. In addition, there was increased relative abundance of L. gasserii/johnsonii with all HMO treatments. An Excel spreadsheet with all 16S rRNA amplicon sequencing results can be found in Supplementary

Total Bacterial, Total Bifidobacterial, and B. infantis Numbers of Simulated Microbiota
Based on the flow cytometry results, the total bacterial cell numbers increased from Vessels 1 to 4 in all units, in the control and HMO treatments, with and without B. infantis Bi-26 (Supplementary Table S3 Total bacterial cell numbers).
Total bifidobacterial and B. infantis numbers were analysed by real-time qPCR. Figure 5a shows the total Bifidobacterium numbers from simulation samples with HMOs, with and without B. infantis Bi-26. Vessel, vessel squared, treatment group, B. infantis Bi-26, and the slope that was associated with B. infantis Bi-26 were the main factors that contributed to the Bifidobacterium numbers and were included in the statistical model. In comparing the control against individual treatments, 2′-FL and 3-FL increased the overall bifidobacterial numbers by 5.3% and 4.6%, respectively. Figure 5b shows a similar analysis of the B. infantis qPCR results in which only the simulations to which B. infantis Bi-26 was added were included. In simulations in which B. infantis Bi-26 was excluded, B. infantis numbers were below the limit of detection. Donor, treatment, vessel, and the interaction of donor and vessel were the main factors that contributed to the B. infantis numbers. Furthermore, 2′-FL and 3-FL increased B. infantis numbers versus control by 6.3% and 14.4%, respectively. The results for the Donor I V2 control treatment are missing, due to low-quality DNA, and were not included in the analysis.

Total Bacterial, Total Bifidobacterial, and B. infantis Numbers of Simulated Microbiota
Based on the flow cytometry results, the total bacterial cell numbers increased from Vessels 1 to 4 in all units, in the control and HMO treatments, with and without B. infantis Bi-26 (Supplementary Table S3 Total bacterial cell numbers).
Total bifidobacterial and B. infantis numbers were analysed by real-time qPCR. Figure 5a shows the total Bifidobacterium numbers from simulation samples with HMOs, with and without B. infantis Bi-26. Vessel, vessel squared, treatment group, B. infantis Bi-26, and the slope that was associated with B. infantis Bi-26 were the main factors that contributed to the Bifidobacterium numbers and were included in the statistical model. In comparing the control against individual treatments, 2 -FL and 3-FL increased the overall bifidobacterial numbers by 5.3% and 4.6%, respectively. Figure 5b shows a similar analysis of the B. infantis qPCR results in which only the simulations to which B. infantis Bi-26 was added were included. In simulations in which B. infantis Bi-26 was excluded, B. infantis numbers were below the limit of detection. Donor, treatment, vessel, and the interaction of donor and vessel were the main factors that contributed to the B. infantis numbers. Furthermore, 2 -FL and 3-FL increased B. infantis numbers versus control by 6.3% and 14.4%, respectively.

Microbial Metabolites
SCFAs (acetic acid, butyric acid, propionic acid), lactic acid, and BCFAs (2-methy butyric acid, isobutyric acid and isovaleric acid) were measured to examine the effects HMOs and B. infantis Bi-26 on bacterial metabolism. The effect for each metabolite w first analysed by subtracting the control values from the corresponding treatment valu and pooling the vessels, such that the samples with and without B. infantis Bi-26 we considered separate. The 2′-FL + 3-FL combination was excluded because no B. infantis B

Microbial Metabolites
SCFAs (acetic acid, butyric acid, propionic acid), lactic acid, and BCFAs (2-methylbutyric acid, isobutyric acid and isovaleric acid) were measured to examine the effects of HMOs and B. infantis Bi-26 on bacterial metabolism. The effect for each metabolite was first analysed by subtracting the control values from the corresponding treatment values and pooling the vessels, such that the samples with and without B. infantis Bi-26 were considered separate. The 2 -FL + 3-FL combination was excluded because no B. infantis Bi-26 samples for this treatment were available. Next, the data were analysed to determine the overall effect of B. infantis Bi-26, examining the vessel level data using nonparametric methods per Brunner et al. [49].

Effects of HMOs and B. infantis Bi-26 on Microbial Metabolite Production
The results showed that 2 -FL effected the biggest changes in general metabolite levels versus the control, but as there were only small number of simulations, these changes did not reach statistical significance (p-values above 0.05). However, we observed a trend for an increase in acetic acid (p = 0.064) and SCFA sum (p = 0.064) with 2 -FL and in lactic acid with both 2 -FL and 3-FL (p = 0.116 and p = 0.116, respectively) (Figure 6a).

Effects of HMOs and B. infantis Bi-26 on Microbial Metabolite Production
The results showed that 2′-FL effected the biggest changes in general metabolite levels versus the control, but as there were only small number of simulations, these changes did not reach statistical significance (p-values above 0.05). However, we observed a trend for an increase in acetic acid (p = 0.064) and SCFA sum (p = 0.064) with 2′-FL and in lactic acid with both 2′-FL and 3-FL (p = 0.116 and p = 0.116, respectively) (Figure 6a).
The addition of B. infantis Bi-26 had a statistically significant impact on propionic acid in the 2′-FL treatment (p = 0.013) (Figure 6a). The changes in the other metabolites were not statistically significant for the other treatments.

2′-FL and 3-FL Utilisation Correlates with SCFA and Total Bifidobacterial Counts
We then examined how strongly the change in 2′-FL and 3-FL related to the changes in fucose, total bifidobacterial, SCFA, and BCFA levels, recorded from each transition between each consecutive vessel and each simulation.  The addition of B. infantis Bi-26 had a statistically significant impact on propionic acid in the 2 -FL treatment (p = 0.013) (Figure 6a). The changes in the other metabolites were not statistically significant for the other treatments.

2 -FL and 3-FL Utilisation Correlates with SCFA and Total Bifidobacterial Counts
We then examined how strongly the change in 2 -FL and 3-FL related to the changes in fucose, total bifidobacterial, SCFA, and BCFA levels, recorded from each transition between each consecutive vessel and each simulation.

Discussion
The early development of the infant gut microbiota is crucial for host-microbe interactions and in immediate and later human health [5,6]. Bifidobacterium is the dominant genus in the microbiota of healthy infants [5]. The ability of certain species of Bifidobacterium to utilise various HMO structures gives them a competitive advantage; conversely, the HMO composition can influence the microbiota composition [51]. Thus, it is important to better understand the interplay between bifidobacteria and HMOs. In this study, three in vitro colon simulations were performed with inocula from faecal samples from infant donors aged 1 to 4 months to determine the effects of 2′-FL, 3-FL, and their combination on the composition of the simulated infant microbiota and on the production of microbial metabolites. In addition, the probiotic B. infantis Bi-26 was examined in the presence and absence of 2′-FL and 3-FL.

Discussion
The early development of the infant gut microbiota is crucial for host-microbe interactions and in immediate and later human health [5,6]. Bifidobacterium is the dominant genus in the microbiota of healthy infants [5]. The ability of certain species of Bifidobacterium to utilise various HMO structures gives them a competitive advantage; conversely, the HMO composition can influence the microbiota composition [51]. Thus, it is important to better understand the interplay between bifidobacteria and HMOs. In this study, three in vitro colon simulations were performed with inocula from faecal samples from infant donors aged 1 to 4 months to determine the effects of 2 -FL, 3-FL, and their combination on the composition of the simulated infant microbiota and on the production of microbial metabolites. In addition, the probiotic B. infantis Bi-26 was examined in the presence and absence of 2 -FL and 3-FL.
Because HMOs resist digestion and generally reach the colon intact [57], we used the EnteroMIX ® colon simulation model, which simulates human colonic fermentation, proceeding from the proximal to distal colon [26,34,37]. Our earlier study described its use in modelling infant colonic fermentation, finding that the simulations that were performed could be grouped by efficiency of 2 -FL utilisation indicating differences in the microbiota composition and HMO utilization capabilities [26]. Compared with our earlier work, which examined only 2 -FL, all three simulations in the current study were considered rapidly fermenting HMOs (2 -FL, 3-FL, or both) [26]. This was determined by complete or almost complete utilization of 2 -FL, 3-FL, or both within vessels 1 and 2 of the simulator, mimicking proximal (ascending and transverse) colon [34,36]. This heterogeneity in infant microbiota composition has also been observed in other fermentation studies with several infant donors [25][26][27][28]31,32]. To reduce the variability across donors, we included only donors who had not started solid foods and were, at least in part, breast-fed.
In this study, all three infants who donated faecal samples for the inoculum had relatively high Actinobacteria (and/or Bifidobacterium) levels. On average, at the phylum level, the microbiota composition of the three inocula in our study were comparable with those of healthy, term, breast-fed 1-4-month-old infants from Ireland and the US [58,59], although a Spanish study has reported bacterial inocula composition from six infants with much higher relative abundances of Proteobacteria [25]. However, in a more detailed analysis, large differences in the microbiota composition of the inocula were observed between the three donors. The genera Blautia and C. aerofaciens were detected in the inoculum from Donor I but absent from Donors II and III. In addition, there were more diverse Lactobacillus species in the inoculum from Donor III versus the other two donors. Further, their abundance of Bifidobacterium also differed. Whereas B. longum/breve was present in all three donors, Donor I was the only microbiota that contained B. adolescentis, and only Donors I and III harboured B. catenulatum/gallicum. Similarly, Lawson et al. noted differences in bifidobacterial community and function between three healthy full-term infants (aged 3-5 months) [60].
The use of an in vitro system to study the effects of B. infantis Bi-26 and HMOs on the microbiota allows vessels that represent various section of the colon to be sampled, which would otherwise be invasive if taken from a live subject. Although the microbiota composition evolved during simulations, the differences in donor microbiota compositions remained both in control simulations and with HMO and B. infantis Bi-26 treatments. The disparities in HMO consumption and fucose levels between simulations could have resulted from differences in the faecal microbiota compositions between donor infants. Samples of the simulated microbiota with Donor II contained no B. catenulatum/gallicum. B adolescentis, C. aerofaciens, and the genera Lachnospira, Blautia, Dorea, and Sutterella were present only in simulations with Donor I, whereas B. animalis and V. parvula were primarily present in the simulation with Donor II. The simulation with Donor III had relative abundance of L. gasserii/johnsonii of 9% to 63% between vessels and treatments, and it was the only simulation to contain L. salivarius and L. vaginalis. Few articles have reported the microbiota composition of individual donors.
The α-diversity was similar in all simulations, and all HMO treatments decreased the α-diversity compared with the control simulations. In all simulations, the addition of 2 -FL, 3-FL, and 2 -FL + 3-FL resulted in increase in the phylum Actinobacteria-specifically, B. longum/breve and B. catenulatum/gallicum, which was expected, because these are species utilising HMOs [10,22,24,51]. Similarly, 2 -FL and 3-FL increased the total bifidobacterial and B. infantis numbers (in the vessels where it was added) versus the control simulations as detected by qPCR. Thus, the addition of HMOs increased those species that utilise HMOs, albeit differentially between individual simulations.
The changes in the composition of infant gut microbiota also influence the production of microbial metabolites, such as SCFAs and BCFAs [61]. Tsukuda et al. followed 12 Japanese infants from birth to 24 months and found variability in the microbiota composition that was mirrored in their metabolite production [62]. The levels of SCFAs and BCFAs in the current study were comparable with our previous work [26], and with the results of a 24 h batch model [53]. A trend for an increase in acetic acid and SCFA sum was found with 2 -FL and in lactic acid with both 2 -FL and 3-FL. However, when examining how the HMO utilization relates to the production of fucose, SCFA, and BCFA, we noted a clear relationship between 2 -FL and 3-FL utilisation by the microbiota and higher SCFA and SCFA + lactate levels. This might indicate that HMO utilisation by the microbiota could translate into greater increased metabolite production. Similarly, 2 -FL upregulated acetate and lactate levels, corresponding to a rise in the relative abundance of Bifidobacterium and bifidobacterial numbers. Bifidobacterium species are known acetate and lactate producers [5,56,62]. Simulation studies enable samples to be drawn from various sections of the artificial colon, but they often do not consider the absorption of metabolites. The comparison of simulation studies with faecal SCFA levels is also complicated, because in faecal samples, SCFA levels comprise the sum of the production, absorption, and consumption by the surrounding microbiota and transit time [61].
The strengths of this study include its comparison of HMOs, individually and in combination, with and without HMO-utilising B. infantis Bi-26 bacterium and infant inocula in parallel in an in vitro colon model. Consequently, we could distinguish the effects of HMOs on individual microbiotas and analyse individual variations between the three donors. Conversely, this setup was a limitation, allowing us to draw conclusions solely from these three individual simulations. Because all three donors had relatively high Bifidobacterium levels from the outset of the simulation, the addition of B. infantis Bi-26 induced only minor changes. To have better detected the possible benefits of B. infantis Bi-26 on infant microbiota composition and metabolites, we would have had to pre-screen donors with lower Bifidobacterium levels or added more B. infantis Bi-26. Moreover, the microbiota composition in in vitro colon models is based on faecal inoculum, not actual colonic microbiota. To obtain sufficient faecal material for the seven parallel simulations, we also had to combine samples from a single donor within a maximum of 2 weeks. Due to the paucity of in vivo data in infants, discrepancies between other models, and interindividual variation, we did not alter pH, duration of the semi-continuous fermentation, or nutrient composition. In future studies, changes to these parameters, the addition of a mucosal component, and absorption of the nutrients could improve the in vitro modelling of fermentation by infant microbiota.

Conclusions
In conclusion, this in vitro model is a suitable alternative that can be used to increase our understanding of the effects of 2 -FL and 3-FL on microbial composition and metabolism. In our study, these HMOs increased the relative abundance and numbers of Bifidobacterium species that can ferment them. In particular, 2 -FL and 3-FL influence the intestinal microbiota composition in an individual manner. HMO utilisation correlated with greater SCFA and lactate production. At the species level, we noted differences in microbiota composition between donors. This study highlights the importance of better understanding the changes in the composition of individual infant microbiota and the effects of HMOs and probiotics on them.