Lactate Metabolism Is Strongly Modulated by Fecal Inoculum, pH, and Retention Time in PolyFermS Continuous Colonic Fermentation Models Mimicking Young Infant Proximal Colon

The metabolism of lactate is important for infant gut health and may lead to acute lactate and/or H2 accumulation, pain, and crying as observed in colicky infants. Functional human studies often faced ethical challenges due to invasive medical procedures; thus, in this study, we implemented PolyFermS fermentation models to mimic the infant proximal colon, which were inoculated with immobilized fecal microbiota of two 2-month-old infants. We investigated the impact of pH, retention time, and accumulation of dl-lactate on microbiota composition and metabolic activity. We found that a drop in pH from 6.0 to 5.0 led to increased LPB and decreased LUB concomitantly with lactate accumulation. Increasing the RT resulted in complete lactate consumption associated with increased LUB. Our data highlight for the first time the impact of key abiotic factors on the metabolism of lactate, which is an important intermediate product for ecology and infant health.

inoculated with immobilized fecal microbiota to mimic 2-month-old formula-fed infant gut microbiota. Using this model with the PolyFermS platform (12), we investigated the effect of important parameters (pH and RT) for lactate metabolism on the gut microbiota composition and activity. Furthermore, because lactate is an important intermediate product associated with infant colic, we investigated accumulation of lactate by supplementing DL-lactate and two infant lactate-utilizing bacterial strains (Propionibacterium avidum and Eubacterium limosum), selected for little or no H 2 production, for their potential to colonize and metabolize residual lactate. A recent classification of the genus Propionibacterium allocated the cutaneous P. avidum to the new genus Cutibacterium (18).

RESULTS
Colonization of donor fecal microbiota in fermentation model. Two PolyFermS continuous fermentation models were used in this study to mimic the conditions in the proximal colon of a 2-month-old formula-fed infant. The fermentation setup consisted of a first inoculum reactor (IR) inoculated with 30% (vol/vol) gellan-xanthan gel beads that immobilized the fecal microbiota, which was connected to a control reactor (CR) and four treatment reactors (TRs). All TRs and the CR were operated in parallel, continuously inoculated with 5% fermentation effluent from the IR, and additionally fed with 95% fresh medium, as presented in Materials and Methods and illustrated in Fig. 1a. After an initial colonization and stabilization time of 11 days, the fermentations inoculated with fecal beads from donor 1 (fermentation F1) or donor 2 (fermentation F2) were divided into two experimental periods. Detailed experimental conditions for F1 and F2 with total times of 79 and 57 days, respectively, are depicted in Fig. 1b. During period 1, the effects of three levels of pH (5.0, 6.0, and 7.0) and two RTs (5 and 10 h) were studied. The effects of lactate supplementation (60 mM DL-lactate) and LUB strain addition on composition and activity of infant gut microbiota were investigated during period 2. Each period consisted of stabilization with CR control conditions for 11 to 23 days, followed by treatment application for 8 to 14 days.
Feces from two 2-month-old healthy infant donors with similar gestational ages, birth weights, ages, and feeding practices (see Table S1a in the supplemental material) were used for immobilization and model inoculation. The microbial composition of donor fecal samples was determined using qPCR targeting the 16S rRNA gene of specific characteristic bacterial groups for the infant gut microbiota ( Table 1). The two selected donors harbored similar levels of Firmicutes and Bacteroides. Compared to donor 2, the fecal sample from donor 1 harbored higher levels of Enterobacteriaceae (9.4 and 7.9 log gene copies g Ϫ1 , respectively), Bifidobacterium (9.2 and 7.6 log gene copies g Ϫ1 ), Veillonella (9.4 and 8.6 log gene copies g Ϫ1 ), and Lactobacillus (9.1 and 6.0 log gene copies g Ϫ1 ).
The IR and CR were operated under constant conditions at pH 6.0 with an RT of 5 h during the entire experiment and used for testing stability over 79 and 57 days in F1 and F2, respectively. After the first stabilization time that allowed gut microbiota to colonize the reactors and to reach steady conditions, the microbiota composition of the IR and CR detected by qPCR during both fermentations was very similar to the corresponding donor fecal sample for most of the targeted groups, including Firmicutes, Bacteroides, and Bifidobacterium (Table 1). Differences were observed for the levels of Enterobacteriaceae, which were 1.5 log lower in the IR (7.9 log gene copies ml Ϫ1 ) and 0.8 log lower in the CR (8.6 log gene copies ml Ϫ1 ) compared to the donor 1 fecal sample (9.4 log gene copies g Ϫ1 ). Also, 1.5-and 2.3-log-higher levels of Veillonella were detected in the IR (10.1 log gene copies ml Ϫ1 ) and CR (10.9 log gene copies ml Ϫ1 ) compared to the donor 2 fecal sample (8.6 log gene copies g Ϫ1 ), respectively.
Fermentation stability. To measure the metabolic and compositional stability of both fermentation models, we performed HPLC, qPCR, and MiSeq sequencing analyses of effluent samples of the IR and CR of F1 and F2. The metabolite ratios and concentrations for short-chain fatty acids (SCFAs; acetate, propionate, and butyrate), lactate, and formate measured with HPLC indicated overall stable microbial metabolic profiles in the IR and CR of both fermentations after an initial colonization and stabilization period of 17 days (Fig. S1). During F1, we observed an effect of time where the acetate concentration decreased (day 22, 128.8 mM; day 79, 94.7 mM) and the butyrate concentration increased (day 29, 8.1 mM; day 79, 21.6 mM), while the total C-mol concentration (mole of carbon per liter) calculated from addition of all metabolites remained stable (Fig. S1). These data suggest that the observed time drift in F1 is associated not with a loss of metabolic activity but instead with discrete equilibration of metabolism, with more acetate, as an intermediate metabolite, being converted into butyrate. Acetate was the main metabolite in effluents of both fermentations, followed by propionate and butyrate. While formate was not detected in the IR and CR of F1, it represented a significant fraction of approximately 20% (Ͼ20 mM) of the total metabolites of F2. The propionate concentration was lower, while the butyrate concentration was higher, in the IR and CR of F1 compared to F2. Furthermore, qPCR data showed stability of the bacterial groups of infant microbiota that were analyzed, including Firmicutes, Enterobacteriaceae, Bacteroides, Bifidobacterium, Veillonella, and Lactobacillus (Fig. S2). The model stability was confirmed by MiSeq data that showed an overall stable relative abundance of microbiota at the genus level in both reactors, with some fluctuations in the relative abundance of Ruminococcus, Veillonella, and Prevotella (Fig. S3). The fermentation setup consisted of an inoculum reactor (IR) containing 30% (vol/vol) fecal beads, connected to a control reactor (CR) and four treatment reactors (TRs) continuously fed with 5% fermentation effluent from the IR and 95% fresh medium. All reactors were constantly flushed with CO 2 to maintain anaerobiosis. Temperature was set at 37°C, stirring speed was set at 180 rpm, and pH was controlled automatically by the addition of 2.5 M NaOH. All reactors had a total working volume of 200 ml. (b) Setup of the experiment conditions at different periods. The fermentation of fecal samples from donor 1 (fermentation 1) and donor 2 (fermentation 2) was divided into 2 periods; each period consisted of stabilization and treatment and washout. RT, retention time; LUB, lactate-utilizing bacteria. No significant differences in composition and metabolic activity of the microbiota between the CR and TRs after the stabilization period were found in both fermentations (Fig. S4).
In conclusion, qPCR detected similar levels of predominant groups for donor samples and the IR and CR. Differences in bacterial levels detected for the 2 donor fecal samples were well reproduced in the IR and CR of F1 and F2 and are reflected in distinct metabolic profiles. After an initial stabilization time of 17 days, we also demonstrated high stability of composition and metabolic activity of the microbiota over 79 and 57 days of continuous operation in the IR and CR operated with constant conditions for both F1 and F2, respectively. Impact of pH. During period 1 stabilization, all reactors were set at pH 6.0 and an RT of 5 h. Combinations of different pHs and RTs were then assigned to TRs during the following treatment period while CR conditions were kept constant (Fig. 1b). The pHs 5.0 and 7.0 were chosen to mimic the colonic pH of breast-fed (fecal pH 5.1 to 5.4 in the first 6 weeks) and formula-fed (fecal pH 7.0 to 8.2 from the second to the fifth week; fecal pH 6.4 after the fifth week) infants, respectively (19). For the statistical analysis of qPCR and metabolite data pooled from the two fermentations F1 and F2 inoculated with different microbiota, we calculated differences (delta) between treatment and previous stabilization period for each reactor. We compared the delta values of each treatment reactor (TR1 to TR4) with that of the control reactor (CR) measured during the same periods, using the nonparametric Wilcoxon rank sum test with false-discovery rate correction. Reducing the pH from 6.0 to 5.0 led to a significant increase in lactate (P Ͻ 0.001), and decreases in propionate (P Ͻ 0.001), isobutyrate (P Ͻ 0.001), and butyrate (P Ͻ 0.001) production at pH 5.0 compared to pH 6.0 were shown (Fig. 2a). For both fermentations, significant lactate accumulation (from 0.6 Ϯ 0.1 to 54.9 Ϯ 3.9 mM in F1; from 0.0 to 47.7 Ϯ 8.0 mM in F2; P Ͻ 0.01) and significantly decreased propionate, butyrate, and isobutyrate (P Ͻ 0.01 for F1 and P Ͻ 0.05 for F2) production were measured at pH 5.0 compared to pH 6.0 ( Fig. S5a and b). Moreover, a pH of 5.0 resulted in decreased acetate in F1 (P Ͻ 0.001) or formate in F2 (P Ͻ 0.05) relative to pH 6.0. Significantly lower levels of Veillonella (P Ͻ 0.01) and Bacteroides (P Ͻ 0.001) and higher levels of Lactobacillus (P Ͻ 0.001) and Enterobacteriaceae (P Ͻ 0.001) were measured using qPCR for effluent samples at pH 5.0 compared to pH 6.0 when combining data from the two fermentations ( Fig. 3a) with fermentation (donor) effects (Table 1). Furthermore, lower relative abundances of Veillonella (F1, 1.5% versus 9.2%; F2, 1.2% versus 17.3%) and Prevotella (F1, 0.5% versus 5.7%; F2, 1.4% versus 5.4%) and higher relative abundances of Lactobacillus (F1, 22.2% versus 0.5%; F2, 2.8% versus 0.03%), Enterococcus (F1, 12.2% versus 4.4%; F2, 32.8% versus 1.0%), and Bifidobacterium (F1, 41.7% versus 30.5%; F2, 47.6% versus 3.5%) were recorded for both fermentations at pH 5.0 compared to pH 6.0 using MiSeq; however, no sequencing replicates prevent statistical analysis on MiSeq data (Fig. 4). During F1, low relative abundances of Ruminococcus (0.19% versus 19.3%) and Peptostreptococcaceae (0.19% versus 7.2%) and high relative abundances of Citrobacter (7.6% versus 2.8%) and Enterobacteriaceae (11.0% versus 6.6%) were measured at pH 5.0 compared to pH 6.0 (Fig. 4a). On the other hand, a strong decrease of the relative abundances of Collinsella (5.9% versus 29.9%) and Bacteroides (2.3% versus 21.4%) was observed in F2 at pH 5.0 compared to 6.0 (Fig. 4b).
Analysis of pooled data from the two fermentations showed a significant decrease in acetate (P Ͻ 0.001) and an increase in butyrate (P Ͻ 0.05) (Fig. 2b) and a significant increase in Enterobacteriaceae (P Ͻ 0.001), Firmicutes (P Ͻ 0.001), Veillonella (P Ͻ 0.001), Bacteroides (P Ͻ 0.05), and total bacteria (P Ͻ 0.01) at pH 7.0 compared to pH 6.0 (Fig. 3b). The impact of the high pH of 7.0 (TR3) on microbial composition and metabolic activity was fermentation (donor) dependent (Table 1; Fig. S5c and d). No significant effect of pH 7.0 on either microbial composition or metabolic activity was found in F2 compared to pH 6.0. In contrast, during F1, pH 7.0 significantly decreased acetate production (P Ͻ 0.01) and Firmicutes and Bifidobacterium levels (P Ͻ 0.05) and increased butyrate (P Ͻ 0.01) and formate (P Ͻ 0.05) accumulation compared to pH 6.0. Lower relative abundances of Bifidobacterium and Prevotella and higher relative abundances of Enterococcus were also observed in both fermentations at pH 7.0 compared to 6.0 ( Fig. 4a and b). Furthermore, the relative abundance of Anaerococcus increased and that of Veillonella decreased during F1, while Bacteroides and Streptococcus increased and Collinsella decreased during F2, at pH 7.0 compared to pH 6.0, although no statistical testing could be done. Impact of retention time. The effect of RT (5 and 10 h) on the gut microbiota composition and metabolic activity was tested at pH 6.0 in TR2 of both models during experimental period 1 (Fig. 1b). An RT of 10 h significantly increased butyrate production (P Ͻ 0.001) compared to 5 h with pooled data from both fermentations (Fig. 2c) and by 4-fold and 2.5-fold in F1 and F2, respectively ( Fig. S6a and b). A longer RT also led to significantly lower acetate in F1 (P Ͻ 0.05) and total metabolite (P Ͻ 0.01) levels compared to an RT of 5 h (Fig. S6a). A 10-h RT significantly increased total bacteria (P Ͻ 0.01), Firmicutes (P Ͻ 0.05), Enterobacteriaceae (P Ͻ 0.01), and Bacteroides (P Ͻ 0.001) when pooling data from the two fermentation (Fig. 3c). We also measured decreased Bifidobacterium (P Ͻ 0.05) and Veillonella (P Ͻ 0.05) in F1 compared to those with a 5-h RT (Table 1). In contrast, no impact of RT was found for the microbial composition of F2 using qPCR. However, decreased Bifidobacterium and increased Enterobacteriaceae abundances during F1 and at 10-h RT compared to 5-h RT were confirmed by MiSeq data ( Fig. 4a and b). Furthermore, lower Ruminococcus and higher Anaerococcus abundance during F1, lower Streptococcus abundance during F2, and higher Prevotella abundance during both fermentations were observed at the 10-h RT than at the 5-h RT ( Fig. 4a and b).
We also compared TR1 (pH 5.0; RT, 5 h) and TR4 (pH 5.0; RT, 10 h) during period 1, because similar conditions were used for all TRs during the stabilization period, resulting in similar microbiota composition and activities ( Fig. S4). At pH 5.0, a 10-h RT significantly decreased lactate accumulation compared to a 5-h RT (P Ͻ 0.001) in both fermentations ( Fig. S6c and d). MiSeq data showed a trend for higher relative abundance of lactate-producing Enterococcus in both fermentations at 10-h RT compared to 5-h RT. Moreover, a lower abundance of Lactobacillus and higher abundances of Bifidobacterium, Enterococcus, and Anaerococcus during F1 were observed at 10-h RT compared to 5-h RT at pH 5.0 (Fig. 4b). In contrast, a lower abundance of Bifidobacterium and a higher abundance of Collinsella and Veillonella were observed during F2 at 10-h RT compared to 5-h RT at pH 5.0 (Fig. 4b).
Impact of DL-lactate supplementation. Supplementation with 60 mM DL-lactate in nutritive medium to mimic the accumulation of lactate in the infant gut resulted in significant lactate accumulation (P Ͻ 0.001) as well as an increase in acetate (P Ͻ 0.001), propionate (P Ͻ 0.001), and total SCFA (P Ͻ 0.001) production compared to no supplementation, when combining data from the two fermentations ( Fig. 5a). Lactate accumulations in the effluent were similar, of 11.7 Ϯ 1.9 mM and 12.8 Ϯ 2.7 mM for F1 and F2, respectively. Significant fermentation (donor)-dependent increases in propionate and butyrate were detected ( Fig. S7a and b). L-Lactate determination by enzymatic assay revealed the presence of both D-and L-isomers of lactate in reactors supplemented with 60 mM DL-lactate in F1 (47.9 and 52.1%, respectively) and F2 (71.5 and 28.5%). No significant effect of lactate supplementation on microbial composition by qPCR was observed (Table 1 and Fig. 6a), except for a small but significant decrease of Eubacterium hallii (P Ͻ 0.05) with addition of lactate. In contrast, adding 60 mM DLlactate appeared to affect microbial relative abundances of some groups, with observed increased Peptostreptococcaceae (10.9% versus 7.1%) and decreased Citrobacter (0.6% versus 3.9%) and Enterobacteriaceae (4.2% versus 11.3%) abundances in F1, and increased Collinsella (43.4% versus 29.3%) and Veillonella (22.8% versus 14.0%) and decreased Bacteroides (5.7% versus 14.1%) and Enterococcus (1.5% versus 8.8%) abundances in F2 (Fig. 4).
Impact of addition of lactate-utilizing bacteria with DL-lactate supplementation. P. avidum or E. limosum was selected among infant LUB for its capacity to utilize lactate with no or little H 2 production, respectively. We tested the impact of daily spiking with each strain individually at a high cell concentration (10 8 CFU ml Ϫ1 ) in reactors supplemented with 60 mM DL-lactate to mimic lactate accumulation.
The addition of P. avidum at 10 8 CFU ml Ϫ1 together with 60 mM DL-lactate resulted in a significant increase of lactate (P Ͻ 0.001), acetate (P Ͻ 0.001), propionate (P Ͻ 0.001), and total metabolite (P Ͻ 0.001) production when combining data from the two fermentations (Fig. 5b). In F1, this treatment also led to increased butyrate concentration (P Ͻ 0.001), decreased Enterobacteriaceae (0.5 log copy number), and increased E. limosum (1.5 log copy number) ( Table 2 and Fig. S7e). Combining data from the two fermentations indicated significant increase in Veillonella (P Ͻ 0.01) and decrease in E. hallii (P Ͻ 0.05) levels after the addition of P. avidum and lactate compared to stabilization with this treatment (Fig. 6b). Trends toward higher Anaerococcus and Ruminococcus and lower Enterococcus abundances were also observed during F1 (Fig. 4).
To further demonstrate the impact of treatments on the infant PolyFermS microbiota of donors 1 and 2, we performed principal-coordinate analysis (PCoA) of weighted and unweighted UniFrac distance (Fig. S9). In fermentation 2, PCoA showed a clear separation of the treated microbiota from the untreated control, whereas this separation was less clear in fermentation 1.

DISCUSSION
PolyFermS closely mimics the young infant gut microbiota. The initial colonization of the gut is important for both short-and long-term health of infants (3). Infant gut microbiota studies using 16S rRNA-based analysis of fecal samples have provided crucial data on the composition and diversity of the gut microbiota and the effects of many factors, such as delivery mode (20) and diet (21). However, molecular methods can provide only limited insights into mechanisms and functions of bacterial species. Moreover, functional in vivo studies in humans often face social and ethical challenges due to invasive medical procedures (9,10). In this study, for the first time we reported gut fermentation models to mimic the proximal colon of a 2-month-old infant and investigated the impact of abiotic and biotic factors to modulate infant gut microbiota composition and metabolic activity. Large individual variations in gut microbiota composition and diversity in the first months of life have been well demonstrated in recent studies (2,4,21). The two infant donors used to inoculate the PolyFermS models harbored very different microbial compositions and in vitro metabolic profiles. The levels of Enterobacteriaceae, Bacteroides, Bifidobacterium, Lactobacillus, and total bacteria of the two infant donors were within the ranges reported in previous publications (17,(22)(23)(24). Distinct microbial compositions of fecal inoculum samples were reflected in different microbial compositions and metabolic activities of the microbiota during fermentations, such as high propionate-producing Veillonella levels together with high propionate production for donor 2 and in the IR and CR of F2, compared to donor 1 and F1.
The levels of predominant bacterial groups detected by qPCR in the IR and CR, with the exception of Veillonella, were similar to the corresponding donor fecal samples, suggesting that the gut microbiota from donor fecal samples were well conserved during sampling, immobilization, and cultivation under the conditions selected for the formula-fed young infant model. The preparation of the bead inoculum used only small amounts of fecal microbiota which could be obtained from ca. only 1 g of fecal material for production of approximately 200 ml of beads, and only 60 ml of beads was required for inoculation of the IR. This, with the reproduction of both the planktonic and sessile microbiota of the colon, is a unique feature of immobilization and using PolyFermS for modeling young, and possibly preterm, infant gut microbiota when only very limited volumes of feces are available. High and stable microbial concentrations, and stable relative abundances comparable to the fecal sample, were measured in the IR and CR throughout the 79-and 57-day fermentations. These data, combined with SCFA data, indicate long-term stability of fermentation models inoculated with infant fecal beads. PolyFermS models can be expanded to various configurations, allowing comparison of treatments and a control with the same microbiota (12,13,15,25). In this study, the PolyFermS model, which combines four treatments with a control reactor operated with constant conditions and inoculated with identical microbiota as produced in the IR, appears well suited for testing a range of abiotic and biotic factors of infant gut fermentation and requires only a minimal amount of fecal material for inoculation.
Low pH increased LPB and decreased LUB concomitantly with lactate accumulation. In vitro fermentations with fecal inocula from 6-month-old infants, children, and adults have demonstrated the impact of environmental conditions, such as pH and RT, on the gut microbiota composition and lactate metabolism (12,16,26,27). Little is known about the impact of such factors on the gut microbiota of younger infants, mainly because suitable gut fermentation models were lacking. Furthermore, lactate is one of the most important intermediate metabolites in the infant gut, and its accumulation can be detrimental for health (28). Using PolyFermS models, we investigated the impact of colonic pH and RT, which are known to vary widely in infants, and simulated lactate accumulation to determine the impact on 2-month-old infant gut microbiota and lactate metabolism.
The colonic pH can have a profound effect on the composition and metabolic activity of the human gut microbiota. A study investigating the effects of pH (5.2, 5.9, and 6.4) on lactate production and utilization in batch cultures inoculated with fecal slurries from four adult donors showed that pH 5.2 induced lactate accumulation due to reduced utilization (27). Using a single-stage continuous model inoculated with immobilized 6-month-old infant fecal microbiota, Cinquin et al. reported that the proportion of lactate significantly decreased when both the pH and RT were increased simultaneously, mimicking conditions from proximal to distal colon (16). Lactate utilization plays a central role in the metabolism of infant gut microbiota and could have a direct impact on infant health (4,5,29). To our knowledge, this is the first study investigating the impact of pH on infant gut microbiota composition and metabolic activity using in vitro colonic fermentation models. The selection of pH 5.0, 6.0, and 7.0 in this study was physiologically relevant, considering that infant stool pH varies from 4.8 to 7.0 in the first month of life (30). A recent study investigating the effect of Bifidobacterium infantis supplementation on fecal pH showed that the mean fecal pH of the probiotic group was 5.15, whereas the control group had a fecal pH of 5.97 (31).
One important finding in this study is the effect of low pH on fermentation under conditions mimicking the infant proximal colon. A low pH of 5.0 led to lactate accumulation and significantly decreased propionate and butyrate production, which agrees with data in adults (27). The decrease of propionate levels at pH 5.0 compared to pH 6.0 could be explained by a lower abundance of Veillonella bacteria, which are the main producers of propionate in the infant gut (4). Similarly, the decrease in butyrate production at pH 5.0 may be associated with lower abundance of butyrate-producing Anaerococcus. The accumulation of lactate at pH 5.0 agrees with the observed higher abundance of LPB (i.e., Lactobacillus, Enterococcus, and Bifidobacterium) and lower abundance of LUB (i.e., Veillonella). Consistent with previous studies (12,26,32), we also observed an inhibition of Bacteroides by acidic pH, as shown by both qPCR and MiSeq analyses.
Increasing RT resulted in complete lactate consumption at low pH, associated with increased LUB. Formula-fed infants showed a large variation in gastrointestinal transit time, with mean RTs of 13.7 h (range, 7.1 to 35.2 h) and 17.4 h (range, 5.4 to 36.5 h) at age 17 and 113 days, respectively (33), while the proximal colon transit time is estimated to be about one-third of the total transit time. In this study, we demonstrated that proximal colonic transit time is a strong determinant of the 2-month-old infant gut microbiota composition and metabolism in vitro. We showed that the effect of RT is pH and donor dependent. Increasing RT from 5 to 10 h at pH 5.0 attenuated the effect of low pH on the gut microbiota composition and metabolic activity and reduced lactate accumulation. This effect could be explained by the lower abundance of lactateproducing Lactobacillus and the higher abundance of lactate-utilizing Veillonella upon increased RT. We suggested that increased RT promotes the establishment of the trophic chain and the reutilization of lactate. In agreement, a recent study using an in vitro continuous fermentation system inoculated with adult fecal microbiota also reported that the abundance of Veillonellaceae (including genus Veillonella) increased with prolonged RT (34). Increasing RT from 5 to 10 h at pH 5.0 resulted in a small but significant increase of isobutyrate, suggesting an elevation of proteolytic activity possibly due to carbohydrate limitations (35).
At pH 6.0, a 10-h RT led to a lower abundance of Bifidobacterium and a higher abundance of Enterobacteriaceae relative to a 5-h RT. This observation agrees with previous studies that showed that Bifidobacterium spp. were less abundant in feces from functional constipated adult patients (36) and that Enterobacteriaceae levels were higher and Bifidobacterium levels were lower in constipated-irritable bowel syndrome (IBS) adults (C-IBS) compared to healthy adults (37). Increasing RT at pH 6.0 favored butyrate production in both fermentations concomitantly with a decrease of the intermediate products acetate (F1) and formate (F2). This observation could be explained by the slow kinetics and low levels of butyrate producers in the infant microbiota, which cannot efficiently reuse intermediate products such as lactate, succinate, and acetate when the RT is short. Our data provide initial mechanistic insights into the possible impact of transit time on infant gut microbiota composition and activity.
Supplementation with lactate and LUB reduced Enterobacteriaceae and increased SCFAs. Because most primary colonizers in the infant gut are LPB, lactate must be efficiently reused to prevent negative consequences of lactate accumulation. However, excess H 2 production from lactate utilization (e.g., by Veillonella) may also lead to flatulence and is a possible factor in infantile colic (38). Indeed, we recently reported higher lactate-utilizing, H 2 -producing bacteria in colicky infants (5). On the other hand, LUB that produce only minimal or no H 2 (e.g., E. limosum and P. avidum) were shown to compete with high H 2 -producing LUB (e.g., Veillonella) in pure and mixed cultures using anaerobic techniques (5).
In this study, a large amount of lactate (ca. 80% of 60 mM added DL-lactate) was reused, confirming the efficient utilization of lactate by LUB. Furthermore, adding 60 mM DL-lactate to mimic lactate accumulation increased butyrate and propionate formation. Interestingly, the impact of lactate was detected only on a functional but not on a taxonomic level, suggesting that lactate increased the activity of LUB by providing more energetic substrate but not by stimulating growth to detectable levels. Infant LPB, including Lactobacillus, produce both D-and L-lactate. The two isomers of lactate were detected at comparable levels after the addition of 60 mM DL-lactate, suggesting that the 2-month-old infant LUB community was able to utilize both D-and L-forms. Our data suggest that LUB of infant colonic microbiota have a high capacity to metabolize lactate, possibly as a natural protective mechanism in infant microbiota preventing lactate accumulation and detrimental health effects such as acidosis.
The E. limosum and P. avidum strains tested in this study were isolated from healthy infant feces and characterized for their ability to metabolize different substrates (5). While E. limosum utilizes lactate to produce butyrate, P. avidum produces propionate, acetate, and CO 2 . Lyophilized E. limosum fed to mice significantly attenuated colitis and increased cecal butyrate levels compared to the control group (5,39). In our study, E. limosum, combined with the supplementation with 60 mM DL-lactate, led to a lower relative abundance of the Enterobacteriaceae family. The treatment also promoted acetate and butyrate production in F1 and propionate in F2, consistently with the butyrogenic and propionigenic profiles of donors 1 and 2, respectively. The increase of propionate might be attributed to the addition of lactate, which further stimulates the lactate-utilizing propionate-producing bacteria. The increase of butyrate and propionate may be of clinical significance for the infant gut, because of their well-established beneficial impacts on host health. Butyrate is the main energy source for enterocytes and regulates the epithelial barrier and immunity functions of the epithelial cells (40,41). Furthermore, butyrate has been implicated in protection against colitis and colorectal cancer (42). On the other hand, propionate has been shown to stimulate an anti-inflammatory response (43).
Propionibacterium, recently reclassified in two different genera, Propionibacterium and Cutibacterium according to dairy and skin origin, respectively, is one of the dominant organisms of the skin microbiota (44). Recent studies have reported its natural occurrence in breast milk (45,46), as well as in neonatal feces (47,48). The addition of P. avidum with 60 mM DL-lactate increased concentrations of both lactate and the main SCFAs, decreased Enterobacteriaceae, and increased butyrate-producing E. limosum by 1.5 log. The increase of butyrate could be explained by the increase of E. limosum. Furthermore, P. avidum produces acetate, which could be used by butyrate producers. Moreover, in comparison with the theoretical washout curves of P. avidum spiked at 1 ϫ 10 8 and 5 ϫ 10 8 CFU ml Ϫ1 , calculated for a 5-h RT in a homogenous continuous stirred-tank reactor (see Fig. S8b in the supplemental material), our data demonstrated the ability of P. avidum to colonize the reactors 4 days after spiking.
In conclusion, we successfully implemented for the first time stable continuous colonic fermentation models to mimic the proximal colon of very young infants using immobilized fecal microbiota. Using the PolyFermS model platform, we observed a strong impact of pH and RT on the composition and metabolic activity of the gut microbiota involved in lactate metabolism, which is important for ecology and infant health. Using two different donors with different microbiota reflects the in vivo situation, where interindividual variability is inevitable and unavoidable and further strengthens the impacts detected in both fermentations.
E. limosum (strain 4119; Laboratory of Food Biotechnology, ETH Zurich), previously isolated from feces of a healthy infant (5), was activated from stabbed agar Hungate stocks (Ϫ20°C). The strain was subcultured daily at 3% (vol/vol) in YCFA medium supplemented with 60 mM DL-lactate (Sigma-Aldrich, Buchs, Switzerland) at 37°C under strict anaerobiosis using Hungate tubes flushed with CO 2 (42,49). Twenty Hungate tubes containing 10 ml of overnight E. limosum cultures were prepared for inoculation of 10 8 CFU ml Ϫ1 , by centrifugation at 2,000 rpm for 20 min and resuspension in 8 ml of prereduced peptone water (10 g liter Ϫ1 peptone, 5 g liter Ϫ1 sodium chloride) before being used to inoculate the reactors. The purity of P. avidum and E. limosum cultures was checked via Gram staining.
Fecal inoculum and immobilization. Two continuous colonic fermentation experiments were performed independently. Fresh fecal samples were obtained from healthy 2-month-old infants born without congenital disease. Because the composition of human milk is very complex and hence difficult to mimic in vitro, both infants selected for this study had been fed exclusively with infant formula (see Table S1 in the supplemental material). Exclusion criteria were variables known to affect the balance of the infant gut microbiota, including preterm birth, antibiotic usage, and gastrointestinal and immunological disorders during the neonatal period. The study was exempted by the Ethics Committee of ETH Zurich because the fecal sample collection was noninvasive and not in terms of intervention. Informed written consent was obtained from the mothers on behalf of the infants.
Experiment setup and fermentation procedures. The fermentation medium was based on the composition designed previously to mimic the chyme entering the colon of 6-month-old infants (17,50). The medium contained the following (g liter Ϫ1 ): lactose (6. The PolyFermS continuous fermentation model used in this study was designed to mimic conditions in the proximal colon of a 2-month-old formula-fed infant. The fermentation setup consisted of a first reactor with a working volume of 200 ml inoculated with 60 ml (30%, vol/vol) fecal beads from the respective donor (IR), which was connected to a control reactor (CR) and four test reactors (TRs) (Fig. 1). All TRs and the CR (200-ml working volume) were continuously inoculated with 5% (vol/vol) fermentation effluent from the IR and fed with 95% fresh medium. To maintain anaerobiosis, all reactor headspaces were constantly flushed with CO 2 . Temperature was set at 37°C, stirring speed was set at 180 rpm, and pH was maintained automatically at 6.0 by adding 2.5 M NaOH.
Initial batch fermentations were carried out at a temperature of 37°C and a pH of 6.0 with stirring (180 rpm) to colonize beads in the IR. During colonization (days 1 and 2), fermentation effluent was replaced by fresh medium every 12 h (17). Afterward, the IR was switched to continuous mode at a flow rate of 40 ml h Ϫ1 , corresponding to a mean RT of 5 h. This flow rate simulated the transit time in the infant proximal colon, which is estimated to be a total transit time of 17.4 h in formula-fed infants aged 113 days (33). After an initial IR stabilization of 5 or 7 days for F1 and F2, respectively, the CR and TRs were connected and operated in continuous mode with the same proximal colon conditions as the IR.
The IR and CR were operated with constant conditions of pH 6.0 and 5-h RT throughout the fermentation time, which was 79 and 57 days for F1 and F2, respectively. Detailed experimental conditions for the two PolyFermS fermentations are depicted in Fig. 1b. After initial stabilization times of 9 and 11 days in F1 and F2, respectively, the fermentations were divided into two periods. During period 1, the effects of pH and RT were studied, while the effects of lactate and LUB on composition and activity of infant gut microbiota were investigated during period 2. Each period consisted of stabilization at pH 6 and a 5-h RT, which was followed by treatment. During treatment 1, combinations of pH (5 or 7) and RT (5 h or 10 h) were assigned to TRs. The pHs (5.0 and 7.0) were chosen to simulate the colonic pH of breast-fed (fecal pH of 5.1 to 5.4 in the first 6 weeks) and formula-fed (fecal pH of 7.0 to 8.2 from the second to the fifth week; fecal pH of 6.4 after the fifth week) infants, respectively (19). During treatment 2, DL-lactate was added in all TRs to achieve a concentration of 60 mM, with or without daily addition of E. limosum (10 8 CFU ml Ϫ1 ) and P. avidum (1 ϫ 10 8 or 5 ϫ 10 8 CFU ml Ϫ1 ).
Sampling of effluents from all reactors was performed daily. The sample supernatant (10,000 rpm for 10 min) was used for HPLC analysis, while the pellet was stored at Ϫ80°C for DNA extraction. HPLC and qPCR were performed on samples collected during the last 3 days of each stabilization and treatment. MiSeq sequencing was performed on pooled samples collected during the last 2 days of the periods. Plate counts of P. avidum were performed in triplicate on samples collected during the last 3 days of stabilization, P. avidum treatment, and posttreatment periods (F2).
Sampling and analysis. (i) DNA extraction. Total genomic DNA was extracted from 200 mg fresh infant feces and the pellet from 2 ml of fermentation effluent samples using the FastDNA Spin kit for soil (MP Biomedicals, Illkirch, France) according to the manufacturer's instructions. DNA concentration and quality were assessed by absorbance measurements at 260 nm on a NanoDrop ND-1000 spectrophotometer (Witec AG, Littau, Switzerland), and samples were stored at Ϫ20°C before qPCR and MiSeq sequencing analyses.
(ii) qPCR analysis. qPCR was performed using an ABI Prism 7500 PCR sequence detection system (Applied Biosystems, Zug, Switzerland). Specific primers targeting predominant bacterial groups or species in the infant gut were used at a final concentration of 0.2 M (see Table S1 in the supplemental material). Amplification conditions were described previously (4).
(iii) MiSeq sequencing analysis. V3-V4 amplicons were prepared using specific forward primer F340 (5=-CCTACGGRAGGCAGCAG-3=) and reverse primer R805 (5=-GGACTACHVGGGTWTCTAAT-3=). Illumina MiSeq sequencing analyses of fecal and effluent samples were carried out at Genotoul (Toulouse, France). Thermocycling was performed with an initial step at 94°C for 60 s, followed by 30 cycles of denaturation at 94°C for 60 s, annealing at 65°C for 60 s, and elongation at 72°C for 60 s, with a final elongation of 10 min at 72°C. The raw data set containing paired-end reads with corresponding quality scores was merged and trimmed using settings as previously mentioned (51). Quantitative Insight Into Microbial Ecology (QIIME) open source software (1.7.0 and 1.8.0) was used for subsequent analysis steps. Purging the data set from chimeric reads and constructing de novo operational taxonomic units (OTU) were conducted using the UPARSE pipeline. The HIT 16S rRNA gene collection was used as a reference database.
Enumeration of P. avidum. Due to the lack of specific primers for Propionibacterium amplification by qPCR, P. avidum was enumerated in duplicate by plating 100 l of effluent sample, which had been serially diluted 10-fold, on 1.5% sodium lactate agar supplemented with metronidazole (4 mg liter Ϫ1 ) and kanamycin (10 mg liter Ϫ1 ) (both from Sigma-Aldrich, Buchs, Switzerland) (52). Antibiotics were used to obtain a higher degree of selectivity for Propionibacterium spp., as metronidazole is active against other anaerobic microorganisms (53), such as Veillonella species (54), and kanamycin inhibits most Gramnegative (such as Escherichia coli) and some Gram-positive bacteria (55,56). A combination of kanamycin and metronidazole allows differentiation of P. avidum, which forms smooth, cream-to orange-colored convex and circular colonies of various sizes (57). Plates were incubated for 5 days in anaerobic jars at 37°C, and cell counts were reported as log CFU ml Ϫ1 effluent.
Metabolite analysis. The concentrations of SCFAs (acetate, propionate, butyrate, valerate, isobutyrate, and isovalerate), formate, and DL-lactate in effluent samples from all reactors were determined by HPLC analysis. Supernatants from effluent samples were passed through 0.45-m nylon HPLC filters (Infochroma AG, Zug, Switzerland) before injection. HPLC analysis (Thermo Fisher Scientific Inc. Accela, Wohlen, Switzerland) was performed as described previously (4). Data were expressed as mmol liter Ϫ1 effluent (mM). L-Lactate concentration was measured by an enzymatic kit according to the manufacturer's instructions (Megazyme, Bray, Co. Wicklow, Ireland). D-Lactate concentration was determined by subtracting L-lactate concentration from total DL-lactate concentration.
Statistical analysis. Statistical analysis was done using IBM SPSS Statistics 20.0 (IBM Inc., Chicago, IL, USA). qPCR (log 10 -transformed) and HPLC data were expressed as the mean results Ϯ SD for the last 3 days of each fermentation period and compared pairwise between stabilization and treatment within each TR, using repeated-measures ANOVA. Comparisons between reactors within each fermentation period were performed using ANOVA after testing for normal distribution using the Shapiro-Wilk test.
We combined SCFA concentrations and bacterial population levels from the two fermentations for statistical analysis as follows. Differences (delta) between treatment and stabilization period within each reactor were calculated for each combination of 3 measurement days, resulting in 9 delta values per fermentation. Delta values between treatment (TR1 to TR4) and control (CR) reactors were compared using the Wilcoxon rank sum test with false-discovery rate correction. Pairwise comparisons of SCFA concentrations and bacterial population levels between each treatment reactor (TR1 to TR4) and control reactor (CR) during stabilization periods were carried out using the Wilcoxon rank sum test with false-discovery rate correction. For all tests, P values Ͻ 0.05 were considered significant.
Data availability. The sequence data reported in this paper have been deposited in the European Nucleotide Archive database (accession no. PRJEB32244).