The study was approved by the Institutional Review Board at the University of Nebraska (IRB Approval Number: 2009019551EP), and written informed consent was obtained from all subjects. The details for the study design were previously described [27]. Briefly, caramel chews were administered to 18 healthy human volunteers during a 16 week period. The first two weeks were established as the baseline period, followed by four sequential testing periods during which chews were administered for three weeks with GOS dosages at levels of 0.0 g, 2.5 g, 5.0 g, and 10.0 g GOS per day. A final two-week washout period was performed at the end of the fourth testing period. Fecal samples were obtained weekly, and DNA was isolated using a method that includes both an enzymatic and mechanic cell lysis [27].

Pyrosequencing of 16S rDNA tags was performed from fecal DNA as described previously [39]. Briefly, the V1-V3 region of the 16S rDNA gene was amplified by PCR from fecal DNA using primers adapted for the Roche-454 Titanium kit. A mixture (4∶1) of the primers B-8FM

(5′-CCTATCCCCTGTGTGCCTTGGCAGTCTCAGAGAGTTTGATCMTGGCTCAG-3′) and B-8FMBifido

(5′-CCTATCCCCTGTGTGCCTTGGCAGTCTCAGAGGGTTCGATTCTGGCTCAG-3′), were used as the forward primers. The primer A518R

(5′-CCATCTCATCCCTGCGTGTCTCCGACTCAGBBBBBBBBATTACCGCGGCTGCTGG-3′) containing an 8-base barcode sequence was used as the reverse primer. Sequences were then assigned to their respective samples via the barcode. The 8FMBifido was used in combination with primer 8FM, as 16S DNA sequences within the genus Bifidobacterium are not well amplified by the latter primer [40].

Equal amounts of the PCR products were combined and gel purified and then sequenced with the 454/Roche A sequencing primer kit using a Roche Genome Sequencer GS-FLX. Sequences were binned according to barcodes, using the Ribosomal Database Project (RDP) Pyrosequencing Pipeline (http://pyro.cme.msu.edu/) ‘Initial Process’ tool [41]. Default parameters were established to remove sequences containing any ambiguous nucleotides, except for the minimum sequences length, which was set to 300 bp. BioEdit Software was used to trim the quality approved sequences to 450 bp before submission to the sequence analyses (see below).

Sequences obtained by pyrosequencing were analyzed using taxonomy-dependent and taxonomy-independent approaches. First, the Classifier tool of the RDP was applied (with a minimum bootstrap value of 80%) to obtain a taxonomic assignment of all sequences. The Classifier approach allowed a fast determination of the proportions of bacterial groups at different taxonomic levels (phylum to genus). Alternatively, the sequences were assigned to Operational Taxonomic Units (OTUs). Accordingly, all sequences from each subject were individually aligned using the RDP Aligner web tool, and then clustered using the RDP Complete Linkage Clustering web tool (with a maximum distance cutoff of 97%; [41]). The OTU picking was performed on a per subject basis, as the entire data set from all of the subjects contained too many sequences for a quality alignment. OTUs that contained less than three sequences were excluded from the analyses. Using Statistical Analysis Software (SAS) to perform ANOVA, the OTUs that were significantly affected by the treatments in each subject were identified.

Representative sequences from each OTU whose abundance was significantly influenced by GOS were subjected to taxonomic classification using SeqMatch, an RDP web tool. From each statistically significant OTU identified, five random representative sequences were aligned to form consensus sequences using SeqMan Software (DNASTAR Lasergene). The consensus sequences were grouped and aligned according to phylum (Actinobacteria, Bacteroidetes, Firmicutes, Fusobacteria, Proteobacteria, and Verrucomicrobia), together with the most closely related type strains or entry in the NCBI database using Muscle 3.6 [42]. Phylogenetic trees were assembled by neighbor-joining with 1,000 bootstrap replicates with MEGA 4.0 Software [43]. Using visual analyses and a distance matrix, OTUs were assigned as sequence clusters with >97% identity, and consensus sequences were generated for each of the OTU sequence clusters, as described above.

Quantification of each OTU in each sample was performed by BLASTn analysis with a local database including all the quality controlled sequences generated by pyrosequencing. A BLASTn algorithm was used with a 97% cutoff (min. length 300 bp) to quantify each OTU within each sample. The OTUs that were closely related to Bifidobacterium adolescentis were quantified by BLASTn using a cutoff of 98% (min. length 300 bp) as clearly differentiated clusters could be identified that showed overlap with 97%. The quantification of OTUs in all subjects was then verified to ensure that individual sequences were not being assigned to different OTUs. In three occasions, OTUs that were initially identified as distinct had very high sequence similarities, and were thus merged together as single OTUs.

Two different methods, the generation of rarefaction curves and Shannon's index, were applied to determine the diversity of the fecal microbiota using 16S rDNA sequence data. The DNA sequences within each sample were aligned and clustered using RDP web tools Aligner and Complete Linkage Clustering. Individual cluster files corresponding to each fecal sample were used to construct Rarefaction curves and determine the Shannon's Index.

To identify differences in the composition of the fecal microbiota induced through dietary treatments (0.0 g, 2.5 g, 5.0 g, and 10.0 g GOS) in all eighteen subjects, one-way ANOVA tests with repeated measures were performed. Samples obtained during the baseline and washout periods were not included within the statistical analysis. Post hoc pair-wise comparisons were done using Tukey's method. P-values of <0.05 were considered significant unless otherwise stated.

A total of 39 strains of bifiodbacteria were screened for their ability to use GOS as a growth substrate. Included were 19 lab strains (from ATCC, commercial sources, and the Department of Food Science Culture Collection) and 20 isolates obtained from subjects in the previous study [27]. Strains were grown anaerobically at 37°C in MRS broth containing 2% GOS (GTC Nutrition, Golden CO). Because the latter material contains 92% GOS, with the balance as lactose, glucose, and galactose, control cultures were prepared that contained an equivalent amount of these sugars (i.e., 0.16% final concentration). In addition, twenty-two anaerobic bacteria that were mainly of intestinal origin were also screened for their ability to use GOS as a growth substrate. All bacteria were obtained from the USDA ARS Culture Collection (Peoria, IL) and included strains of Bacteroides thetaiotamicron, Bacteroides distasonis, Bacteroides fragilis, Bacteroides uniformis, Bacteroides ovatus, Clostridium butyricum, Clostridium histolyticum, Clostridium bifermentans, Clostridium difficile, Clostridium innocuum, Clostridium paraputrificum, Clostridium perfringens, Clostridium ramosum, Clostridium rumen, Clostridium sporogenes, Enterococcus faecalis, Enterococcus faecium, Enterobacter aerogenes, and Streptococcus salivarius. Bacteria were initially propagated in Brain Heart Infusion (BHI) or Reinforced Clostridial Agar (RCA). To assess growth on GOS, cells were transferred (2%) into a basal medium [5 g/L Peptone No 3 (Becton, Dickinson, and Company), 5.0 g/L Casitone (Becton, Dickinson, and Company), 0.5 g/L L-Cysteine (Sigma), 40 mL Salt Solution, 10 mL Hemin (Sigma), 900 µL Vitamin K₃ (Sigma), and 1 g/L Yeast Extract (Becton, Dickinson, and Company)] containing 1% GOS (GTC Nutrition, Golden, CO). Control cultures containing 0.08% mono- and disaccharides were prepared as above.

All cultures were incubated at 37°C in an anaerobic chamber (Forma Scientific, Marietta, Ohio) containing an atmosphere of 85% nitrogen, 10% hydrogen, and 5% carbon dioxide and assessed for growth by optical density measurement at 600 nm in a Beckman Model 640 spectrophotometer. Each experiment was done in triplicate and the average optical densities were determined.

A total of 288 fecal samples were included in this study. Pyrosequencing resulted in a total of 2.3 million sequences. After quality control analysis (see Methods), an average of 8,200 sequences per sample was obtained. The mean sequence length was approximately 450 bp. An average of 2,022 OTUs was identified per subject. To assess the effect of GOS on the bacterial diversity in fecal samples, rarefaction curves for all eighteen subjects were generated (data not shown), and Shannon's diversity indices were calculated. This analysis revealed that consumption of GOS did not alter bacterial diversity of the fecal samples (p<0.0713).

The overall composition of the gut microbiota in the 18 individuals included in this study is in general agreement with that of previous studies [13]. During the baseline period (no dietary modulation), the microbiota was dominated by two phyla, Firmicutes (64%) and Bacteroidetes (28%). Other phyla detected included Actinobacteria (3%), Verrucomicrobia (1%), and Proteobacteria (1%). Approximately 3% of the sequences remained unclassified. At the family level, the predominant groups were the Lachnospiraceae (31%), Ruminococcaceae (18%), Bacteroidaceae (12%), and Bifidobacteriaceae (2%). The most common genera included Bacteroides (12.2%), Fecalibacterium (7.7%), Blautia (7.4%), Ruminococcus (3.7%), Roseburia (2.2%), Bifidobacterium (1.5%), and Dorea (1.3%).

Sequence proportions determined by pyrosequencing were used to determine the effect of GOS on the composition of the gastrointestinal microbiota. The groups that were significantly affected are shown in Table 1, according to phylum, family, genus (by RDP Classifier), and OTUs. The control treatment (0.0 g GOS in confections) had no effect on the fecal microbiota, as the microbial populations during this period were not significantly different from those during the baseline and washout periods (although slight increases in the family Bacteroidaceae and the genus Bacteroides were detected). In addition, no significant changes in the fecal microbiota were detected for a dose of 2.5 g GOS. In contrast, consumption of 5.0 g GOS led to several significant changes. There were significant increases (p<0.05) in the family Bifidobacteriaceae and the genus Bifidobacterium, compared to the control dose. At the species level, the abundance of only one OTU, corresponding to the species, Fecalibacterium prausnitzii, increased significantly at this dose. In contrast, significant decreases in abundance were observed at both the family and genus levels for Bacteroidaceae (p<0.01) and Bacteroides (p<0.01), respectively, at the 5.0 g dose compared to the control.

At the 10.0 g GOS dose, additional differences in the proportions of several phyla (using taxonomy-based analysis) were observed (Table 1). There was a significant increase in Actinobacteria compared to the control (p<0.001), as well as compared to the 2.5 g dose (p<0.05). This change was associated with an increase both in the family Bifidobacteriaceae, the genus Bifidobacterium, and several OTUs related to Bifidobacterium species. Although there were not significant differences between the 5 gram and 10 gram dose in Bifidobacteriaceae, the genus Bifidobacterium, and Bifidobacterium species, the amount of bifidobacteria at 10 gram GOS was consistently higher than at 5 gram. In addition, bifidobacteria were significantly increased at 10 gram GOS when compared to the 2.5 gram dose (Table 1). Collectively, the abundances of bifidobacteria determined by pyrosequencing were highly correlated (r = 0.7629, p<0.0001) with the cell counts previously obtained by qRT-PCR [27] (Figure S1). This supports previous findings that show that our pyrosequencing approach allows a quantitative determination of bifidobacteria in human fecal samples.

There were few bacterial taxa other than bifidobacteria that were influenced by GOS, based on a taxonomy-based analysis (Table 1). Statistically significant decreases were observed only within the family Bacteroidaceae (p<0.05) and the genus Bacteroides (p<0.05) when compared to the control dose of GOS. In contrast, the OTU-based approach identified two additional taxa, Coprococcus comes and F. prausnitzii, whose abundances differed significantly at 5 and 10 g doses. However, no trend was apparent from these results (Table 1). Although few taxa were identified that significantly decreased with the administration of GOS when all 18 subjects were assessed collectively, our analysis nonetheless showed that different bacterial lineages were decreased in individual subjects. As shown in Figure S2, the changes were detected in a small number of subjects and occurred primarily within taxonomically diverse members within the phyla Firmicutes (Figure S2A) and Bacteroidetes (Figure S2B). Most of these taxa were reduced by GOS, but no consistent pattern was detected among subjects. Thus, it appears that although GOS induces a rather selective increase of different lineages of bifidobacteria, GOS does not result in a consistent increase of another bacterial group or a significant decrease of particular bacterial groups.

A BLASTn analysis revealed that eight OTUs had statistically significant changes in abundance at the 10 g GOS dose, six of which were assigned to the genus Bifidobacterium. Three of the OTUs showed a high similarity (>97%) to described Bifidobacterium species, B. adolescentis, B. longum, and B. catenulatum (Table 1, Figure 1A). The other OTUs (Bifidobacterium spp I, II, and III) showed lower sequence similarities (91–96%) to known Bifidobacterium species, and the phylogenetic analysis shown in Figure 1A revealed that these OTUs belonged to lineages clearly distinct from known type strains. Interestingly, two of these OTUs (Bifidobacterium spp II and Bifidobacterium spp III), showed the numerically highest response to GOS (Table 1, Figure 1A).

Although the consumption of GOS at the higher doses resulted in compositional shifts within subjects on a collective basis (Figure 1B), closer examination of samples from individual subjects revealed that the effect of GOS on the intestinal composition of participants was subject to considerable variation among individuals (Figure 2). Indeed, the data showed that there were some individuals that were essentially unaffected by GOS consumption, whereas other experienced significant changes. The most substantial alteration was the increase in the Actinobacteria (at the phylum, family, genus, and species levels) which was observed in sixteen of the eighteen subjects after 5.0 g and seventeen of the subjects after 10.0 g of GOS. At the genus level, in particular, substantial increases were observed in the abundances of Bifidobacterium, which increased approximately ten-fold (from 1–4% up to 18–33%) in four subjects (subjects 2, 4, 11, and 17), and five-fold in seven additional subjects (subjects 1, 9, 10, 15, 18). Several culturable isolates (NEGOS 1–3) were obtained from these subjects and were found to associate within the distinct Bifidobacterium spp. II lineage (Figure 1A), indicating that this GOS responding linage contains bacteria that can be cultured. There was a very consistent reduction in the Bacteroidetes (at the family, genus, and species levels), which occurred within all of the subjects at some point after 5.0 g of GOS was consumed (Figure 2). At the genus level, there was a decrease in the abundance of Bacteroides in 17 subjects after the 5.0 g GOS dose (all except subject 4), with 14 of those subjects having a further decrease after consumption of 10.0 g of GOS.

Analyses of the community profiles provided insight into how GOS influenced the population dynamics over the entire 16 week study period. All of the changes induced by GOS were reversible within one week, and no differences (Student's t-test, p>0.05) could be detected in the proportions of the bacterial groups between the first washout sample and the baseline sample (Figure 3). The temporal patterns of the three main phyla (Actinobacteria, Bacteroidetes, and Firmicutes) and two of the selected genera (Bifidobacterium and Bacteroides) for five representative subjects showed that these groups were stable in their temporal response to GOS. For example, levels of Actinobacteria, Bacteroidetes, and Firmicutes were remarkably stable in fecal samples at the baseline and washout periods, and their populations returned to the baseline level within one to two weeks after GOS consumption was stopped. The same observations were also made at the genus level for Bifidobacterium and Bacteroides.

As shown above, GOS induces alterations to the human fecal microbiota that are remarkably specific for bifidobacteria. However, GOS utilization was observed to be a strain-specific phenotype, at least based on in vitro growth studies (Table S1). We also considered whether or not the ability to utilize GOS as a growth substrate was restricted to bifidobacteria and absent in other colonic bacteria. Therefore, we tested the ability of twenty-two strains of bacteria that are associated with the human intestinal tract to utilize GOS. This was performed by comparing growth in media containing GOS with growth in basal medium (i.e., without an additional source of carbohydrate). This experiment revealed that 6 of the 11 Clostridium strains could utilize GOS (Figure S3), as indicated by higher final cell densities compared to growth without carbohydrates. In addition, three of the six strains of Bacteroides were also significantly enriched when GOS was present. Significant growth on GOS was not observed, however, for strains of the genera Enterococcus, Enterobacter, or Streptococcus used in this study.