Microarray Analysis and Barcoded Pyrosequencing Provide Consistent Microbial Profiles Depending on the Source of Human Intestinal Samples^{▿ †}

Sample collection.

Fecal samples (F1 to F3) used in this study were collected at home from three healthy individuals (2 female and 1 male; aged 30 to 32 years), frozen in dry ice immediately, and transported to the laboratory, where they were kept at −80°C until further analysis.

Ileostomy effluent samples (S1 and S2) were previously collected (5) at least 3 h apart in the morning and afternoon, respectively, from a healthy 74-year-old male ileostomist as part of a previous project, results of which are reported elsewhere (5). The volunteer collected the ileostomy effluent samples by emptying the ileostomy effluent in freezer baskets as soon as the bulk of ileostomy effluent was collected in a clean empty ileostomy bag. Samples were stored on dry ice at approximately −80°C and were processed within 3 days after collection.

An ileum lumen sample (S3) was obtained from a 24-year-old healthy female individual by using an orally introduced catheter, which passed to the ileum (120 cm distal to the pylorus) by peristalsis. Sampling was done under gastroenterologist supervision, following flushing of the ileum with 10 ml physiological salt solution through a port of the catheter, after which the sample was frozen and stored at −80°C.

Bacterial reference strains and culture conditions.

Bifidobacterium longum (DSM 20219) was grown in ST medium as described in reference 41 with a substitution of proteose peptone for 1 g/liter Casitone (Becton Dickinson, Breda, Netherlands) and meat extract for 3 g/liter beef extract (Sigma, St. Louis, MO). Escherichia coli MC1061 was cultivated in Luria-Bertani (LB) broth at 37°C in a shaking incubator (135 rpm; Heidolph Instruments GmbH & Co., Schwabach, Germany). Streptococcus thermophilus (CNRZ 1066) was grown in M17 broth (Becton Dickinson) supplemented with 0.5% (wt/vol) glucose (Sigma) at 37°C. Veillonella atypica (DSM 20739) was grown in Veillonella medium described in the DSMZ catalogue (medium 136) under an N₂ atmosphere.

DNA extraction.

Genomic DNA (gDNA) extractions from reference strains were performed using the FastDNA Spin kit for soil (MP Biomedicals, Solon, OH) with pelleted cells from 2 ml pure culture as starting material (data not shown).

Total DNA was extracted from 0.25 g fecal sample and 0.25 ml ileal content, using the repeated bead beating method described in reference 40, and from 0.2 g ileostomy effluent as previously described (50) by using the QIAamp DNA stool minikit (Qiagen GmbH, Hilden, Germany). A recent study by Salonen et al. (40) concluded that the difference in microbial compositions between DNA extraction methods is relatively small in relation to that between subjects.

DNA was quantified using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE) and adjusted to 10 to 20 ng/μl as template for subsequent 16S rRNA gene PCR amplification.

16S rRNA gene amplicon pyrosequencing.

Amplicons from the V1 to V6 region of 16S rRNA genes were generated by PCR using two degenerated (27F-DegL and 27F-DegS) and two nondegenerated (27F-Nondeg and 35F-Nondeg) primers in combination with a single reverse primer (1061R-Deg) (Table 1) for each fecal and small intestinal DNA extraction.

To facilitate pyrosequencing using titanium chemistry, each forward primer was appended with the titanium sequencing adaptor A and an “NNNN” barcode sequence (Table 1) at the 5′ end, where NNNN is a sequence of four nucleotides that was unique for each sample and did not start with G or contain a triplicate of identical bases. The reverse primer carried the titanium adaptor B at the 5′ end.

PCRs were performed using a GS0001 thermocycler (Gene Technologies, Braintree, United Kingdom) in a total volume of 50 μl containing 1× PCR buffer, 1 μl PCR-grade nucleotide mix, 2.4 units of Faststart Taq DNA polymerase (Roche Diagnostics GmbH, Mannheim, Germany), 200 nM forward and reverse primers (Biolegio BV, Nijmegen, Netherlands), and 0.2 to 0.4 ng/μl of template DNA. The amplification program consisted of an initial denaturation step at 95°C for 5 min; 35 cycles of denaturation at 95°C for 30 s, annealing at 56°C for 40 s, and elongation at 72°C for 70 s; and a final extension step at 72°C for 10 min. The size of the PCR products was confirmed by gel electrophoresis using 1 μl of the reaction mixture on a 1% (wt/vol) agarose gel containing 0.4 μg/ml ethidium bromide (Bio-Rad, Hercules, CA). Control PCRs were performed alongside each separate amplification without addition of template and consistently yielded no product. The optimal annealing temperature for primers (56°C) with attached adaptors and barcodes was determined by a 12-degree temperature gradient (49°C to 61°C) PCR using DNA from fecal sample F2 (data not shown).

PCR products were purified with the ZR-96 DNA Clean and Concentrator kit (Zymo Research, Orange, CA) followed by DNA yield quantification using a NanoDrop ND-1000 spectrophotometer.

Purified PCR products were mixed in equimolar amounts with a final DNA concentration of 100 ng/μl. The pooled amplicons were pyrosequenced using an FLX genome sequencer in combination with titanium chemistry (GATC-Biotech, Konstanz, Germany). Sequencing occurred on a picotiter plate, of which a quarter space was available for samples included in this study.

HITChip analysis.

Microbial community profiling was also performed using the HITChip (35), which is a phylogenetic microarray, produced by Agilent Technologies (Palo Alto, CA) in an 8 by 15K format (each chip has 8 arrays, each with 15,000 probes), with over 4,800 tiling oligonucleotides _targeting the V1 or the V6 region of the 16S rRNA gene from 1,132 microbial phylotypes present in the human gastrointestinal tract (35). (See the supplemental material for the hybridization and analysis procedure.)

Pyrosequence analysis and comparison with HITChip analysis.

Pyrosequences were sorted per barcode. To the best of our knowledge, no recommendations for quality filtering of reads generated by pyrosequencing using titanium chemistry have been published to date, and therefore, we applied previously reported recommendations for quality filtering of pyrosequences generated by the GS 20 platform (24). This filtering was performed using an in-house perl script that passed sequences with exact matches to the forward primer, no ambiguous bases (N), and read lengths not longer or shorter than 1 standard deviation (SD) from the average sequence length (>87 to 157 and <314 to 359 nucleotides; see Table S2 in the supplemental material for the actual upper and lower read length limits per sample). Additionally, primer sequences were removed from the pyrosequencing reads, and the remaining sequences were analyzed. The numbers of operational taxonomic units (OTUs), rarefaction curves, and total species richness estimations (abundance-based coverage estimators [ACE] and Chao1) (22) for the quality filtered sequences per sample were calculated using ESPRIT (43) with default settings (without removing low-quality reads) at an 0.02 distance level.

Taxonomic classification of sequencing reads employed a locally installed version of the Ribosomal Database Project (RDP) classifier (48), which by default produces classifications into the new higher-order taxonomy as proposed in Bergey's Manual of Systematic Bacteriology (17). The corresponding assignments differ from those that are produced by HITChip analysis, which has a standard output of the relative contributions of 1,132 microbial phylotypes at level 1 (phylum-like with Firmicutes divided into classes or clusters), level 2 (genus-like), and/or level 3 (phylotypes based on >98% sequence identity) (35) to the overall microbial community per sample in the phylogeny as proposed by Collins et al. (11). The 16S rRNA gene sequences of these phylotypes are present in a nonredundant ARB (27) database that was used for HITChip probe design, the human unique OTU database (35, 36). The sequences with corresponding assignments present in this database were exported and used to train the RDP classifier. This yielded a classifier that (in combination with a trial multiclassifier provided by the RDP staff) could classify pyrosequencing reads originating from different samples on a large scale with the same assignments as those produced by HITChip analysis. The confidence threshold used for classification of the pyrosequences was kept at 80%. Moreover, the multiclassifier summarized the assignments per taxon, which facilitated calculation of relative contributions and subsequent construction of microbial profiles for comparison with those that were generated by HITChip analysis.

Hierarchical cluster analysis of the microbial profiles was done in R version 2.9.2 by computing a distance matrix that was based on Pearson product-moment correlation coefficients (r) between pairs of profiles with level 2 community data. Visualization of hierarchical clustering was done by using the distance matrix in the hclust function in R with Ward's minimum variance method as agglomeration method. The Shannon diversity index was calculated in R using the diversity function with the level 2 community data from each sample. Screening of sequences for exact matches with the HITChip probes or primers was performed using in-house perl scripts.

Quantification of bacterial community members by qPCR.

All qPCRs were performed in 96-well PCR plates (Bio-Rad) sealed with Microseal B film (Bio-Rad) using a MyIQ Icycler with MyIQ software version 1.0.410 (Bio-Rad). Each reaction was carried out in a total volume of 25 μl using IQ SYBR green Supermix (Bio-Rad) according to the manufacturer's instructions with 200 nM forward and reverse primer in combination with 5 μl template DNA.

From a literature survey, group-specific primers were chosen (Table 1) that were deemed optimal in their phylogenetic specificity and coverage (based on results of the probe match tool offered in the Ribosomal Database Project [http://rdp.cme.msu.edu/] [10]) as well as the minimal tendency to form secondary structures, including hairpin loops, heterodimers, and homodimers (assessed using the IDTDNA Oligoanalyzer 3.1 [Integrated DNA Technologies]) that may interfere with PCR efficiency (25). The optimal annealing temperature for each primer pair was determined by an 8-degree temperature (53°C to 64°C) gradient PCR using gDNA from _target bacterial reference strains as template (data not shown).

The amplification program for most qPCR assays consisted of an initial denaturation step at 95°C for 5 min; 40 cycles of denaturation at 95°C for 15 s, annealing at the optimal temperature for 30 s (with data acquisition), and elongation at 72°C for 30 s; and a final extension step at 72°C for 10 min. The elongation time for the Streptococcus qPCR assay was set at 20 s, whereas the denaturation and the elongation times for the Bifidobacterium qPCR assay were set at 30 s and 40 s, respectively. This was done, for practical reasons, to reduce the time to complete the Streptococcus qPCR assay and to provide sufficient time for denaturation and elongation of the relatively large amplicon (550 bp) produced during the Bifidobacterium qPCR assay. Melting curve analysis was carried out by incrementally increasing the temperature from 55°C to 95°C at 30 s per 0.5°C with continuous fluorescence collection.

For each qPCR assay, a standard curve comprising 8 serial 10-fold dilutions of full-length 16S rRNA gene PCR products was generated from _target gDNA preparations of the respective reference strains. For the total bacterial qPCR assay, a standard curve was generated using E. coli MC1061 gDNA. The standard curves of each qPCR assay were used to determine the relative contributions of _target bacterial groups to the total bacterial community in sample DNA preparations.

16S rRNA gene library construction and analysis.

The 27F-Nondeg and Uni-1492-rev primers (Table 1) were used for PCR amplification of 16S rRNA gene sequences from the undiluted extracted DNA of samples S1 and S2. Each reaction was performed in quadruplicate in a total volume of 50 μl containing 1× PCR buffer (Promega), 200 nM (each) primer (Biolegio), 200 μM (each) deoxyribonucleotide triphosphate (Promega), 1.25 U GoTaq DNA polymerase (Promega) and 1 μl of the extracted DNA. The amplification program was performed on a T1 thermocycler (Biometra, Göttingen, Germany) and consisted of an initial denaturation step at 94°C for 2 min; 35 cycles of denaturation at 94°C for 30 s, annealing at 52°C for 40 s, and elongation at 72°C for 90 s; and a final extension step at 72°C for 5 min. PCR products were verified by gel electrophoresis using 5 μl of the reaction mixture on a 1.0% (wt/vol) agarose gel containing 0.4 μg/ml ethidium bromide (Bio-Rad). Quadruplicate PCR products from the same sample were pooled and subsequently purified with a High Pure cleanup microkit (Roche) using 10 μl elution buffer (Roche) for elution. The purified PCR products were diluted 10 times, of which 1 μl was used for ligation into the pGEM-T Easy vector (Promega) overnight at 4°C according to the manufacturer's instructions. XL1-Blue competent cells (75 μl; Stratagene, La Jolla, CA) were transformed with 2 μl ligation mixture according to the manufacturer's instructions and subsequently plated on LB agar containing ampicillin (100 μg/ml; Sigma), isopropyl-β-d-thiogalactopyranoside (IPTG; 0.16 mM; Carl Roth GmbH, Karlsruhe, Germany), and 5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside (X-Gal; 100 μg/ml; Invitrogen, Carlsbad, CA) for blue-white color screening. White colonies were randomly selected and separately cultured overnight at 37°C in LB medium containing ampicillin. Subsequently, 2 sets of 96 clones from each clone library were randomly selected and the cloned inserts were sequenced from both ends using the T7 and SP6 priming sites (GATC-Biotech, Konstanz, Germany). The obtained sequences per clone were assembled using Clone Manager 9 Professional Edition (Scientific & Educational Software, Cary, NC), yielding nearly full-length 16S rRNA sequences, which were analyzed with DNA Baser v2.71.0 (Heracle Software, Lilienthal, Germany) to trim vector sequences. Subsequently, sequences were tested for chimeras using Mallard (2) according to the instructions of the authors with the default settings to identify chimeric sequences. Putatively anomalous 16S rRNA sequences identified by Mallard were further analyzed according to the anomaly confirmation protocol suggested by the authors, and unambiguously anomalous sequences were excluded from further microbiome interpretations. Nonchimeric sequences were taxonomically classified using the in-house customized RDP classifier described above. Sequences for which no classification could be obtained above the 80% confidence threshold were classified using the locally installed version of the RDP classifier version 2.2 (48) with a default confidence threshold of 80%.

The nonchimeric 16S rRNA gene sequences from each clone library were aligned using the SILVA Webaligner (34) and subsequently imported into ARB. Each clone library was manually screened (using a neighbor joining distance matrix, employing no correction, generated in ARB) for sequences showing <98% identity to 16S rRNA sequences represented in the human unique OTU database that was used for HITChip probe design.

Nucleotide sequence accession numbers.

The cloned 16S rRNA gene sequences from ileostomy effluent were deposited in the GenBank database and are available under accession numbers HQ176022 to HQ176318.

Analysis of pyrosequencing reads from 16S rRNA gene amplicons.

Pyrosequencing of the 16S rRNA gene PCR amplicons from fecal samples (F1 to F3), ileostomy effluent (S1 and S2), and ileal lumen content (S3) yielded in total 190,652 sequences with 7,944 ± 2,201 sequences per sample. Quality filtering passed approximately 50% of the pyrosequencing reads, with an average length of 224 nucleotides (nt) (Table 2; see Table S1 in the supplemental material for a comprehensive overview of the characteristics of pyrosequencing reads before and after quality filtering). Detailed analysis revealed that the majority (74.80% ± 4.74%) of the sequences that failed to pass quality filtering were due to sizes that were outside the sequence length thresholds (see Table S2).

Microbial profiles, based on the level 1 (phylum-like [Fig. 2]) and level 2 (genus-like [see Fig. S1 in the supplemental material]) taxonomic assignments of the pyrosequences that passed quality filtering, were constructed for all samples. As anticipated, most pyrosequences were assigned to the Firmicutes (85.6%), Actinobacteria (7.6%), and Bacteroidetes (2.9%; predominantly encountered in fecal samples F1 and F2 as well as in the ileal lumen content sample S3). Notably, only 1.5% of all pyrosequences could not be classified using the confidence threshold of 80% and are represented as Unclassified_Human unique OTU (Fig. 2). Furthermore, microbial profiles obtained using all pyrosequences (without quality filtering) revealed essentially the same profiles as those obtained with the quality-filtered sequences, albeit with a raised abundance of the Unclassified_Human unique OTU (see Fig. S2), indicating that the quality filtering step does not drastically influence the reconstruction of the microbial community but predominantly eliminates noise.

Hierarchical clustering of the microbial profiles revealed separate grouping of fecal samples and ileum-lumen content from ileostomy effluent samples (see Fig. S3A in the supplemental material). The divergence between these clusters was most apparent for phylogenetic groups belonging to the Firmicutes, with fecal samples and ileum-lumen content being abundant in Clostridium clusters IV (12.3% ± 7.3%), XIVa (58.6% ± 8.6%), XVI (4.4% ± 3.8%), and XVIII (3.0% ± 1.8%), while Bacilli (9.2% ± 4.5%) and Clostridium clusters I (24.0% ± 11.5%), IX (13.3% ± 7.1%), XI (14.0% ± 14.1%), and XIVa (27.2% ± 13.4%) were predominant in ileostomy effluent. Moreover, species richness, as reflected by Chao1 and ACE supported by the rarefaction curves, as well as the Shannon diversity index, was higher in fecal samples and ileum-lumen content than in ileostomy effluent (Table 2; see also Fig. S4).

The effects of different forward primers on microbial profiling by barcoded pyrosequencing.

To determine the effects of different primers on microbial profiling, 16S rRNA gene PCR amplicons were generated for each intestinal sample using four different forward primers (27F-DegL, 27F-DegS, 27F-Nondeg, and 35F-Nondeg). Microbial profiles constructed on the basis of pyrosequences per sample were highly correlated for the different primers (average r of 0.88 ± 0.14 at level 2 community data [data not shown]). Furthermore, hierarchical clustering (see Fig. S3A in the supplemental material) of the microbial profiles revealed distinct clusters of microbial profiles for each of the samples using the four forward primers, except for sample S3 as discussed below, indicating that the effect of primers on microbial profiling is smaller than the sample-specific effect and supporting a high level of technical reproducibility of the pyrosequencing method. Analogously, comparison of rarefaction curves (see Fig. S4) and species richness estimators Chao1 and ACE, as well as Shannon diversity indices (Table 2), did not reveal a particular primer giving consistently the highest or lowest value for any of these ecological metrics. Nonetheless, qualitative comparison demonstrated that the microbial profiles deduced from pyrosequencing using amplicons generated with the 27F-DegL, 27F-DegS, and 35F-Nondeg primers were notably more abundant in Actinobacteria than were those using the 27F-Nondeg primer (Fig. 2), confirming previous reports on the underestimation of the Actinobacteria using the 27F-Nondeg primer (21).

Because the PCR annealing temperatures used for amplicon generation differed between pyrosequencing and HITChip analysis (56°C versus 52°C, respectively), both annealing temperatures were employed for HITChip analysis of sample F3 to investigate the effects of different annealing temperatures on microbial profiling. Results demonstrated highly similar microbial profiles (r ≥ 0.99; data not shown), and therefore, comparison of microbiota profiling by pyrosequencing and that by HITChip analysis was not biased by different PCR annealing temperatures.

Comparison of barcoded pyrosequencing with HITChip analysis.

The concordance between microbial profiling by pyrosequencing and that by phylogenetic microarray analysis was evaluated. Although the principles for classification and abundance estimations of microbial community members differ between these technologies, hierarchical clusterings of the microbial profiles from the two methods matched (see Fig. S3 in the supplemental material).

Microbial profiles as a result of HITChip analysis were compared to those from pyrosequencing using the 27F-Nondeg primer, as this forward primer is used for amplicon generation in the HITChip analytic procedure. The resulting comparison of the community data at level 1 (phylum-like) showed a high correlation for the fecal samples (F1 to F3; r = 0.99 to 1.00) and ileum-lumen content (S3; r = 0.99), while the correlation was lower for ileostomy effluent samples (S1 and S2; r = 0.53 to 0.62) (Fig. 2). Correlations for the community data at level 2 (genus-like) were significantly lower but remained highest for the fecal samples (r = 0.63 to 0.78) and ileum-lumen content (r = 0.71) and lowest for the ileostomy effluent samples (r = 0.31 to 0.49) (see Fig. S1 in the supplemental material). This difference between ileostomy effluent and other intestinal samples is also demonstrated in the community data scatter plots at levels 1 and 2 (see Fig. S5). Remarkably, numerous phylogenetic groups at level 2 showed significant abundances with HITChip analysis but were absent from the pyrosequence data set (see Fig. S5 inset), suggesting that HITChip analysis enables detection of low-abundance bacterial groups by its dynamic range being broader than that of pyrosequencing.

To verify if suboptimal HITChip probe matches could potentially explain higher abundances per cluster with grouping of fecal samples and ileal lumen content (cluster I) apart from ileostomy effluent samples (cluster II) of some phylogenetic _targets in the pyrosequence data relative to HITChip analysis (see Fig. S6 in the supplemental material), pyrosequences were screened for exact matches with the HITChip probes designed for detection of these phylogenetic groups (see Fig. S7). In general, the two clusters showed approximately the same fractions (∼90%) of pyrosequences that had a perfect match with at least one of the HITChip probes. For cluster I, Eubacterium rectale et rel. and Ruminococcus obeum et rel. contained the most pyrosequences that lack a HITChip probe perfect match, while Veillonella showed the highest number of sequences (19.3%) without a perfect match for the HITChip probes for cluster II.

Phylogenetic groups for which the abundance estimates were higher in the HITChip analyses than in pyrosequencing included Streptococcus spp. (Bacilli), for which a deviation in relative abundance estimations between the two profiling technologies was as high as ∼7% (see Fig. S6 in the supplemental material). The abundances of this phylogenetic group and others were further investigated by means of qPCR as well as 16S rRNA gene cloning and sequencing (see below).

Quantification of bacterial groups by qPCR.

To evaluate the performances of pyrosequencing and HITChip analysis in estimating relative abundances of microbial community members, the results of the two techniques were compared with those obtained by means of group-specific qPCR, focusing on Bifidobacterium, Veillonella, Streptococcus, and E. coli (Fig. 3).

The estimated community proportion of Streptococcus spp. was consistently highest with HITChip analysis relative to qPCR assays and pyrosequencing, while Bifidobacterium abundance levels were expectedly low when determined with the HITChip as a result of using the 27F-Nondeg primer for the initial sample preparation. Surprisingly, Bifidobacterium abundances assessed with the qPCR assay were relatively low as well, whereas estimations by pyrosequencing using the 27F-DegL, 27F-DegS, and 35F-Nondeg primers showed relative abundances varying from 1.10% to 6.98% for samples F1, F2, F3, and S3 and from 4.04% to 16.37% for samples S1 and S2. For Veillonella and E. coli, relative abundances were highest by pyrosequencing analysis, followed by intermediate values obtained from qPCR and lowest values assessed by HITChip analysis. Sample S3 was the only sample for which an E. coli abundance above 1.5% was detected, and it showed relative contributions as high as 31.8% and 17.4% in microbial profiles from pyrosequencing using the 27F-DegL and 27F-DegS primers, respectively. Although the proportion of E. coli bacteria as assessed by the HITChip was considerably lower, with an abundance of 0.44%, the E. coli-specific qPCR assay revealed a contribution of 5% and confirms the abundant presence of E. coli in sample S3. Additionally, this observation suggests that microbial profiling using primers 27F-DegL and 27F-DegS results in a more accurate representation of the microbial composition in intestinal samples that contain higher proportions of E. coli.

Screening ileostomy effluent for novel bacterial phylotypes.

Based on the prominent deviations between pyrosequencing and HITChip analysis for ileostomy effluent, these small intestinal samples are a potential source for a range of bacteria with 16S rRNA sequences that are absent from the human unique OTU database to date. To support this notion, 16S rRNA clone libraries for the two ileostomy effluent samples were constructed and analyzed (Fig. 4).

A total number of 139 and 158 cloned nonanomalous 16S rRNA gene sequences were obtained for the S1 and S2 16S rRNA clone libraries, respectively. Out of the total of 297 cloned sequences, 6 could not be classified at level 2 assignments above the 80% confidence threshold using the RDP classifier trained with the human unique OTU database. Consequently, these sequences were grouped to Unclassified_ with the specific level 1 assignment (Fig. 4). To determine if the resolution of the identifications could be improved, sequences were reclassified using the standard RDP classifier (see Materials and Methods), which is based on a set of 16S rRNA gene sequences more exhaustive than that of the human unique OTU database. This showed that both Unclassified_Bacilli sequences were assigned to the order Lactobacillales, though deviating in their genus-level classifications as Granulicatella for one and Streptococcus for the other. The sequence classified as the latter also showed <98% identity to 16S rRNA gene sequences represented in the human unique OTU database. Sequences of the Unclassified_Clostridium cluster XIVa group could be classified no further than the family Lachnospiraceae, whereas the Unclassified_Proteobacteria were assigned to the genus Variovorax belonging to Betaproteobacteria.

Further analysis of the libraries revealed that 20 sequences, predominantly belonging to Veillonella, showed <98% identity to 16S rRNA gene sequences represented in the human unique OTU database and were therefore considered to be phylotypes not previously reported as being associated with the human intestine (Fig. 4). The finding of a relatively large proportion of these phylotypes among the Veillonella 16S rRNA gene sequences supports the suggestion that the HITChip probes display relatively poor sequence matches with these sequences (see above). Detailed analysis indeed showed that 5 out of 7 HITChip probes specific for Veillonella had more than 2 mismatches with sequences classified as Veillonella, whereas the remaining two probes at most had 1 mismatch (see Fig. S8 in the supplemental material).

The same screening strategy was applied to determine if the cloned sequences classified as Veillonella had exact matches with the forward and reverse primers used for the Veillonella qPCR assay. Out of the 72 cloned sequences classified as Veillonella, only one sequence did not show a perfect match with the reverse primer. In contrast, 16 sequences had a single mismatch with the forward primer (see Fig. S8 in the supplemental material). Interestingly, 13 of these sequences were also identified as novel intestinal phylotypes (see above). This shows that the forward primer for the Veillonella qPCR assay is not in agreement with the 16S rRNA gene sequences of Veillonella and could also explain the lower Veillonella abundances as determined by qPCR in comparison with those estimated by pyrosequencing (Fig. 3). Taken together, these results suggest that the human small intestine is inhabited by novel Veillonella phylotypes that previously have not been reported to inhabit this niche.