Written consent was given by each subject and ethical approval was granted by ACT Health (protocol #: ETH.6/03.276) and the Australian National University (protocol #: 2007/131[LESC-CMHS]).

Twenty subjects (10 males, 10 females), with an average age of 58 y (range 46–69 y), were evaluated. No subject consumed antibiotics in the six months prior the study. Fifteen subjects ingested 10 mg bisacodyl and 2 L polyethylene glycol (PEG) and underwent colonoscopy for a range of indications, five did not. For those who consumed the preparation, fecal samples were collected one month before (−1m), one week before (−1w), and where possible, within one week (+1w), one month (+1m), and three to six months (+3m) after the procedure. An average of 3.7 samples was taken from each subject that underwent the procedure. For those who did not consume the preparation, two samples were collected three to six months apart. These subjects and their first sample are refered to as A, B, C H and W, and their second sample denoted +3m. All samples were frozen immediately after collection, and stored at −80°C until required.

All samples from all subjects were analysed by polymerase chain reaction – denaturing gradient gel electrophoresis (PCR-DGGE); samples from 10 subjects who consumed the preparation, with various PCR-DGGE results, were subsequently analysed by high-throughput sequencing (HTS). For HTS, samples −1m and −1w, and the last available sample for each of these subjects (either sample +1w, +1m, or +3m) were selected. Both samples from the five subjects who did not consume the preparation were also subjected to HTS analysis.

As bacteria are not evenly distributed throughout stool samples [14], material was collected from eight different sites from each specimen, combined and used for the DNA extraction. Total DNA was extracted using the QIAamp DNA Stool Mini-kit according to the manufacturer’s instructions (Qiagen). Bacterial cell lysis was conducted at 95°C for 5 min, and proteinase K incubation at 70°C for 20 min. DNA was quantified using a Nanodrop® ND-1000 spectrophotometer (Analytical Technologies).

The microbiota of the different samples was compared using the universal bacterial PCR-DGGE protocol of Walter et al. [15], with a few modifications. DNA was amplified using primers targeting the V2–V3 region of the 16S ribosomal RNA gene of bacteria (positions 339–539 of the E. coli 16S rRNA gene), and were as follows: HDA-1-GC (CGC CCG GGG CGC GCC CCG GGC GGG GCG GGG GCA CGG GGG GAC TCC TAC GGG AGG CAG CAG T-3′) (GC clamp in bold), and HDA-2 (5′-GTA TTA CCG CGG CTG CTG GCA C-3′). The 50 µl PCR reactions contained 20 ng template DNA, 1.25 U of Taq (Amplitaq), 2.0 mM MgCl₂, 5 µl 10×buffer, 0.2 mM of each deoxynucleoside triphosphate (Promega), sterile water, and 10 pmol of each primer. Amplification conditions were as follows: 1 cycle at 94°C for 4 min; 30 cycles of 94°C, 30 s; 56°C, 30 s; 68°C, 1 min, and 1 cycle at 63°C for 7 min. The intensity of the PCR product on agarose gels was used as a guide to determine the volume loaded on the denaturing gradient gel to ensure equal amounts of PCR product were used. A species ladder, referred to as the marker, comprising 16S rRNA gene amplicons that migrated to various positions, was run in the two end lanes of each gel.

DGGE was performed to separate the HDA-PCR amplicons, using a DCode universal mutation detection system (BioRad). Polyacrylamide gels were prepared and run with 1×TAE buffer (Amresco). The denaturing gradient was created using two 6% acrylamide-bis, 37.5∶1 (BioRad) stock solutions. The polymerisation catalyst was 10% ammonium persulfate (95 µl), the adjunct catalyst TEMED (55 µl) (Amresco). The gels contained a 22 to 55% gradient of urea and formamide increasing in the direction of electrophoresis. The electrophoresis was conducted at a constant voltage of 130 V at 60°C for 4 h. Gels were stained with fresh ethidium bromide solution (5 µg/ml) for 20 min, destained in MilliQ water for 20 min, visualised using an ultraviolet transilluminator and photographed.

For each subject, the PCR products from all fecal samples were run on the same gel to eliminate gel-to-gel differences in DGGE profiles [16], [17]. The gels were loaded into the BioNumerics software package (v 5.10, Applied Maths), normalised against the marker, and the presence/absence of bands compared using a 3% tolerance limit and 2% optimization. Dendrogram and cluster analysis were performed using algorithms within BioNumerics. Percent similarity among different bacterial community profiles belonging to samples within a subject were scored by the Dice coefficient. The Unweighted Pair Group Method with arithmetic means (UPGMA) was used to obtain the dendrograms.

DNA was amplified using primers targeting the V1–V3 region of the 16S ribosomal RNA gene of bacteria, equivalent to positions 27–518 of the E. coli 16S rRNA gene, using the following primers: 27F (5-CCATCTCATCCCTGCGTGTCTCCGACTCAGMIDGAGTTTGATCMTGGCTCAG-3) and 518R (5-CCTATCCCCTGTGTGCCTTGGCAGTCTCAGWTTACCGCGGC TGCTGC-3) where the sequences in bold represent the forward primer A and reverse primer B for the GS FLX TitaniumMV em PCR (LibL) v2 kit (Roche), MID represents the 8 bp multiplex identifier (MID) and the remainder represents the 16S rRNA gene forward and reverse primers amplifying the V1–V3 regions. The 48 MID sequences were kindly provided by Professor Andrew Benson [18]. The 50 µl PCR reactions contained 20 ng template DNA, 1 U high fidelity Platinum Taq DNA polymersase (Invitrogen), 2.0 mM MgSO₄, 5 µl 10×PCR buffer, 0.2 mM of each deoxynucleoside triphosphate (Promega), sterile water and 10 pmol of each primer. PCR reactions were performed under the following conditions: initial denaturation at 94°C for 3 min, then 30 cycles of: denaturation at 94°C for 15 s, annealing at 55°C for 30 s, and extension at 68°C for 60 s. The final extension was at 68°C for 10 min. The amplicons were extracted from 2% agarose gels, and purified using a Wizard SV Gel & PCR Clean-Up System (Promega) and the Agencourt AMPure XP system (Beckman Coulter). Concentrations were determined using an Agilent 2100 Bioanalyzer (Agilent Technologies) with DNA 1000 chips. Equal amounts of the PCR products were combined (total 500 ng per 48 samples) to complete each library. Emulsion-based PCR amplification and sequencing on the 454 Genome Sequencer FLX-Titanium system was performed at the Biological Research Facility, Canberra, Australia, according to the manufacturer’s instructions (454 Life Sciences, Branford, CT). Signal processing and base calling were performed using the GS FLX+ System’s software (Roche).

Raw sequences of the V1–V3 hypervariable regions of the 16S rRNA gene were curated and processed using the open-source program, Mothur v.1.23.1 [19]; sequences were trimmed of primer and barcode sequences (primer differences allowed, 2 bp, barcodes, 1 bp) and denoised using the shhh.flows command, a translation of Chris Quince’s PyroNoise algorithim [20]. Sequences were aligned using the SILVA database, and chimeras removed using the Uchime code [21]. The taxonomy of sequences was determined using the classify.seqs command with RDP (2010) training sets, 1000 iterations, and a cut-off of 80%. Distance matrices were generated using a cutoff of 0.30. Files containing operational taxonomic units (OTUs) found across all samples (shared files) were used to describe the dissimilarity (1-minus similarity) among all samples. Shared files, in which the proportion of each phyla/OTU in each sample was computed, were used as inputs when calculating distances between all samples. The resulting distance matrices were used when calculating the Jaccard (presence/absence) and Yue and Clayton theta (relative abundance) measures of dissimilarity for the dendrograms and non-metric dimensional scaling (NMDS) plots. The minimum dimensions used for the NMDS calulations was two, the minimum number of iterations was 10, the maximum 500. Two NMDS axes were plotted in Excel. The statistical program JMP (v.9) was used for matched-pairs t-tests of the mean distances between samples in the Jaccard and Yue and Clayton theta NMDS plots.

All sff files generated by the 454 Genome Sequencer, their associated metadata, primer and MID sequences, are included in dataset S1.

There was an average of 26 16S rRNA gene fragments/bands per DGGE profile. To define normal temporal variation, Dice similarity coefficients were calculated for the DGGE profiles of the two pre-colonoscopy samples from those subjects who underwent colonoscopy, and from both samples from subjects who did not. The average similarity was 93.6% and the range was 86–100% for these 20 subjects (Table 1).

The similarity between the pre- and last post-colonoscopy samples is also reported in Table 1. For the majority (11/14) of subjects, the similarity co-efficients between a pre-colonoscopy sample and the last available post-colonoscopy sample were within the range of normal temporal variation. Seven subjects had similarity coefficient values outside the estimated range of normal temporal variation for any given sample comparison (Table 1). Those subjects who did not revert back to the pre-colonoscopy state, when sample −1w was compared with the last sample from that patient, had sample comparisons outside the range of normal temporal variation for at least one comparison involving sample +1w (subjects 4, 9, and 13). These three subjects underwent colonoscopy for different indications and were on different or no medications. We therefore sought to determine whether or not there were any unique clinical characteristics that could account for the changes seen in these three subjects before and after colonoscopy (4, 9, 13). Subject 4 was unique relative to the other subjects in that the indication for colonoscopy was low iron, and he had been on iron supplements until 10 days prior to the procedure, which may account for the changes post-procedure. Subject 9 suffers from ulcerative colitis and was on a number of medications relative to two other subjects with UC (3 and 5) who were only on 5ASA and showed no changes pre- and post-colonoscopy. Subject 13 underwent a surveillance colonoscopy for a family history of colorectal cancer and was on no medications. We conclude that the colonoscopy preparation did not significantly change the gut microbiota for the majority of subjects as determined by PCR-DGGE.

A total of 565,174 raw sequences of the V1–V3 hypervariable regions of the 16S ribosomal RNA gene were processed using Mothur v.1.23.1 [19], yielding a total of 339,428 sequences (average of 251 base pairs). Each sample was covered by an average of 8,054 high quality reads (range = 554–16,395) (Table 2). The total number of unique sequences, determined after filtering the alignment, was 19,169. Discordance between observed and estimated richness, determined by Chao (range of 53–73%) and Ace (40–69%), is likely due to the presence of rare species, which occur naturally in large numbers in microbial communities [22], [23]. A summary of observed and estimated richness, diversity, and sample coverage is provided in Table 2.

The composition of the microbial communities in the subjects’ samples was compared based on the presence/absence (Jaccard coefficient, Figs. 1 and 2), and relative abundance (Yue and Clayton theta [θ_YC], Figs. 3 and 4) of operational taxonomic units (OTU). The results of each comparison are viewed as a dendrogram (Figs. 1 and 3) and a corresponding NMDS plot (Figs. 2 and 4). NMDS is a nonparametric ordination-based method for reducing community data complexity and identifying meaningful relationships among communities.

When comparing the microbial composition of OTUs (Fig. 1), samples from the same individual are more similar and cluster together. In some subjects the pre-colonoscopy samples had the most similar microbial composition, whereas in others a pre- and post-colonoscopy sample was most similar (Fig. 2). Subjects 7, 8, and 15 had at least one sample that was was quite dissimilar.

There were few changes in the relative abundance of OTUs in the majority of subjects (Fig. 3): all samples from 4/10 subjects that underwent colonoscopy (3, 7, 11, and 15) and 1/5 subjects who did not (C) clustered together, while samples from the other subjects were in related clusters in the dendrogram. They all plotted closely together in the NMDS plot (Fig. 4), indicating minor changes in the relative abundance of OTUs.

In four of ten subjects (3, 5, 8, 9), one pre- or post-colonoscopy sample was quite dissimilar from the other two samples indicating differences in the relative abundance of OTUs; this was investigated by assessing the distribution of the two major phyla, Firmicutes and Bacteroidetes. The relative abundance of these two phyla varied widely among subjects (Fig. 5). A shift towards an increase in Firmicutes was observed in four of the five subjects who did not consume the preparation and averaged 18% (Table 3). Large shifts (>32%) in the relative proportions of Firmicutes and Bacteroidetes (Fig. 5, Table 3) were observed in comparisons involving the dissimilar samples; samples that were dissimilar to the other sample(s) from the same subject, and correlate with the results in the dendrograms and NMDS plots (Figs. 1, 2, 3, 4). To gain insight into these shifts, the raw data for the 50 predominant OTUs were compared to determine whether or not one or two OTUs could explain changes in the relative abundance in the Yue and Clayton NMDS plot. We found there were no consistent findings: different OTUs contributed to changes in the relative abundance of the dominant phyla in different subjects. For example, the changes in subject 5 are due to a decrease in OTU 2, Bacteroides (phyla Bacteroidetes), and an increase in Rikenella and other genera of the phyla Firmicutes. In subject 9 the changes are due to an increase in OTU 9, Tannerella (phylum Firmicutes), and a decrease in genera belonging to the phylum Bacteroidetes including Paraprevotella. The shift in subject A can be attributed to a decrease in OTU 1, Bacteroides, and an increase in OTUs belonging to the phylum, Firmicutes.

Because samples from the same subject were expected to be more similar than samples from another subject, standard analyses could not be employed; the distances between samples in the two NMDS plots were compared using matched pairs t-tests (Fig. 6 and Table 4). There were no significant differences in the distance between samples obtained in the absence of colonoscopy preparation and the distance between samples obtained pre- and post-colonoscopy. We conclude that colonoscopy preparation with PEG does not affect the microbiota of subjects to a greater degree than the temporal variation seen in those who did not undergo the preparation.