Subjects with type 2 diabetes (N = 18) and non-diabetic controls (N = 18) were all males at age 31 to 73 years and body mass indices (BMIs) ranging from 23 to 48 (Table 1). The diabetic group had elevated concentration of plasma glucose as determined by a fasting oral glucose tolerance test (OGTT). Subject C17, though having high plasma glucose, was referred to the non-diabetic group based on the measurements of baseline glucose and biochemical analysis of blood samples. The two groups were comparable with regard to their characteristics.

The total number of reads obtained for 20 subjects by V4 16S rRNA pyrosequencing was 1028955. After applying quality control and trimming we obtained 382229 high quality sequences from diabetic persons and 357782 sequences from healthy controls, accounting for 71.9% of the total reads (Table 2). The number of sequences varied between the subjects from 10521 to 66999 with a mean of 37000 (SD 16062). The average sequence length of trimmed sequences was 234 bp.

The mean bacterial diversity, as estimated by Chao1 indices from the equalized data sets (Table 2), was not different between the diabetic subjects and the controls, comprising 2287 (SD 587) and 2363 (SD 1113), respectively. However, the variability of Chao1 estimates between the diabetic subjects (Chao1 from 1364 to 2939) was lower compared to the controls (Chao1 from 589 to 4010). As shown by the rarefactions curves, bacterial diversity and richness in diabetic subjects with BMI more than 31 (Figure 1, D2, D5, D8 and D9) was somewhat higher than in lean diabetics, with the means Chao1 of 2785 (SD 436) and 1945 (SD 399), respectively. The same tendency was observed in the control group (Fig. 1, subjects C4, C6, C7 with BMI >31).

Sequences were distributed among 5 bacterial phyla including Firmicutes and Bacteroidetes, together harboring on average up to 90% of sequences, as well as phyla Proteobacteria, Actinobacteria and Verrumicrobia, each accounting for 1–4% of the sequences (Figure 2). The proportion of Firmicutes was significantly higher (P = 0.03) in the controls (mean 56.4%) compared to the diabetic group (mean 36.8%). Accordingly, phylum Bacteroidetes and Proteobacteria were somewhat but not significantly enriched in the diabetic group. Furthermore, ratios of Bacteroidetes to Firmicutes correlated positively and significantly with the values of plasma glucose determined by OGTT (R = 0.47, P = 0.04) and negatively, though not significantly, with BMI (R = −0.32, P = 0.17; Figure 3A, D). Relative abundances of Actinobacteria and Verrumicrobia were not significantly different between the groups (Figure 2).

Phylum Bacteroidetes was primarily presented by class Bacteroidetes that was on average slightly increased in diabetics (44%) compared to controls (33%; Figure 2). Most of the sequences from Firmicutes belonged to the class Clostridia, varying from 14 to 72% between the subjects. Other classes included Erysipelotrichi and Bacilli, accounting for less than 6% and 1% of the sequences, respectively. The proportion of Clostridia in diabetics was significantly lower than in controls (P = 0.03; means of 53% versus 34%, respectively) and showed a tendency to decrease with higher levels of plasma glucose (R = −0.42, P = 0.06; Figure 3B). The relative abundance of class Bacilli was increased in diabetics at close to significant levels (mean 0.19% versus 0.03% in controls, P = 0.06; Figure 2). Similarly, class Betaproteobacteria, belonging to Proteobacteria, was highly enriched in subjects with diabetes (mean 2.09% versus 0.81% in controls; P = 0.02) and positively correlated (R = 0.46, P = 0.04) with plasma glucose (Fig. 3C). The relative abundances of bacterial genera (see SI Figure S1) were not significantly different between the groups. However, the proportion of genus Roseburia negatively correlated with plasma glucose at levels close to significance (R = −0.52, P = 0.06; data not shown), whereas the opposite trend was observed for the genus Prevotella (R = 0.32, P = 0.15; data not shown).

As seen from principal component analysis (PCA), a separation between the diabetic group and the control group at the level of bacterial phyla could be best observed from PC1 and PC2 (Figure 4A; 45% and 28% of explained variance, respectively). It was attributed to Proteobacteria and Actinobacteria in the second direction (PC2) in combination with Bacteroides versus Firmicutes and Verrumicrobia in the first direction (PC1). At the level of bacterial classes the variation was predominantly linked to a higher positive score for Betaproteobacteria, Bacteroidetes and Bacilli in the diabetic group, versus Clostridia and Erysipelotrichi in the control group (Figure 4B). The PCA plots of bacterial families generally indicated low levels of systematic variation where only 20–30 percent of the variation was explained by successive PCs, and no distinct clustering was observed (see SI Figure S2A). No association of the diabetic group with particular bacterial genera was apparent. Nevertheless, some of the diabetic persons (see SI Figure S2B, subjects D1, D3, D4, D6, D8, D10) could be discriminated from most of the controls by genus Prevotella versus the combination of genera Lachnospiraceae IS, Roseburia and Subdoligranulum (PC1 50%). PCA of multiple genera showed no clear grouping of the subjects with diabetes probably due to the high individual differences masking the systematic variation (data not shown).

The total bacterial counts were similar in the diabetic and the control group having a median values of 10.6 (quartile ranges (QR) 10.1–11.3) and 10.5 (QR 10.0–11.3) log₁₀ bacteria per g stool, respectively. The estimates of bacterial groups and genera by qPCR are presented in Figure 5. In diabetic group the median counts of Bacteroides-Prevotella group (10.2, QR 9.2–10.6), genus Prevotella (8.8, QR 7.6–11.5) and C. leptum subgroup (9.6, QR 9.0–9.7) were slightly but not significantly increased compared to the controls by approximately 0.7 log₁₀ bacteria. Genus Prevotella was assessed in 25 (14 diabetic subjects and 11 controls) out of 36 samples analysed, whilst for the remaining samples it was below the detection limit of the applied qPCR assay. The difference in median counts of C. coccoides-E. rectale, C. coccoides and Lactobacillus groups, and genus Roseburia between the diabetic persons and controls was less than 0.2 log₁₀ bacteria per g stool. The estimates of Prevotella and Bacteroides-Prevotella group were in some qPCR assays higher that the counts of total bacteria. The explanation might be that the primers, used for quantification of total bacteria, were not universal enough to amplify all bacterial populations, or alternatively that the group-specific primers amplified other targets as well. To overcome this limitation and to assess the differences between the groups of subjects we used ratios of bacterial counts. The ratios of Bacteroides-Prevotella to C. coccoides-E. rectale group correlated significantly and positively (R = 0.38; P = 0.03) with plasma glucose (Figure 6A) but not with BMI (R = 0.06, P = 0.71). Likewise, a positive correlation with plasma glucose was found for the ratios of Bacteroides-Prevotella to C. coccoides subgroup (R = 0.46, P<0.01; data nor shown) and for the proportion of the Lactobacillus group (R = 0.33, P = 0.05; Figure 6B). Higher ratios were in general related to a reduction in the C. coccoides-E. rectale and C. coccoides groups.

The study protocol was approved by the Ethical Committee of Copenhagen and Frederiksberg Municipalities (KF 01-320695) and performed according to the declaration of Helsinki. Both written and verbal consent was obtained from the subjects of the study. The study included 36 males, diagnosed with type 2 diabetes (N = 18) or as being non-diabetic (N = 18) by their general practitioner confirmed by OGTT. The OGTT included the measurements of plasma glucose (mmol/l) at baseline and two hours after administration of 75 g glucose diluted in 500 ml of water. The BMIs were calculated from the formula: weight (kg)/height (m)². Subjects' ages, BMIs (kg/m²) and 2-hour plasma glucose concentrations are presented in Table 1. Fecal samples were kept at 5°C immediately after defecation, brought to the laboratory within 24 hours and stored at −80°C before analysis.

Total bacterial DNA was extracted from the fecal samples using the QIAamp DNA Stool Mini kit (QIAGEN, GmbH, Germany) according to the manufacturer's protocol for pathogen detection with slight modifications [23]. DNA concentration and quality in the extracts was determined by agarose gel electrophoresis (1% wt/vol agarose in Tris-acetate-EDTA (TAE) buffer) and with a NanoDrop 1000 spectrophotometer Thermo Scientific (Saveen Werner ApS, Denmark).

Tag-encoded amplicon pyrosequencing of fecal DNA was conducted for 10 persons with type 2 diabetes and 10 controls, matching in BMI and age. A selection criterion for diabetic persons was high severity of diabetes as evaluated by OGTT plasma glucose (Table 1). The primers used for pyrosequencing were a modified 530F-mod (GCCAGCMGCNGCGGTA; [24]) and 1061R (CRRCACGAGCTGACGAC; [25]) amplifying a 562 bp DNA fragment flanking the V4, V5 and V6 regions of the 16S rRNA gene ([26]; see SI Table S1). Modification of the primer 530F included two bases (TA) added to 3′-end in order to increase primer specificity. The sequence coverage of the forward and reverse primers was tested using the Probe Match feature at the Ribosomal Database Project release 10 (RDP 10; http://rdp.cme.msu.edu; [27]). The DNA concentration was measured with a NanoDrop spectrophotometer and adjusted to 5 ng/µl for all samples. PCR amplification (in a volume of 40 µl) was performed using 1x Phusion HF buffer, 2.5 mM magnesium chloride, 0.2 mM dNTP mixture, 0.8 U Phusion Hot Start DNA Polymerase (Finnzymes Oy, Espoo, Finland), 0.5 µM of each primer (TAG Copenhagen A/S, Denmark) and 1 µl diluted DNA sample. PCR was performed using the following cycle conditions: an initial denaturation at 98°C for 30 s, followed by 30 cycles of denaturation at 98°C for 5 s, annealing at 53°C for 20 s, elongation at 72°C for 20 s, and then a final elongation step at 72°C for 5 min. The PCR products were run on an agarose gel and purified using the QIAEX II Gel Extraction Kit (QIAGEN). Second round of PCR was performed as described above, except that the primers with adapters and tags were used, the number of cycles was reduced to 10 and the annealing temperature was increased to 56°C. Addition of adapter and tags specific to each sample was done using the custom primers with adapter A and 10 tags (see SI Table S1) required for pyrosequencing [24]. The concentration of the tagged PCR product was determined by qPCR. The standard DNA used in qPCR was prepared from Pseudomonas putida and quantified against a standard concentration of 200 base ssDNA oligo (TAG Copenhagen A/S, Denmark). The qPCR was performed in a 25 µl volume, using 1x Brilliant buffer (Stratagene, Cedar Creek, Texas, USA), 0.5 µM of each primer FLX (forward: GCCTCCCTCGCGCCATCAG and reverse: GCCTTGCCAGCCCGCTCAG) and 1 µl of diluted PCR products with tags and adapters. Amplification was conducted using a Mx-3000 thermocycler (Stratagene) at the following conditions: an initial denaturation at 95°C for 10 min, followed by 40 cycles of denaturation at 95°C for 30 s, annealing at 60°C for 60 s. Amplicons were then mixed in approximately equal concentration (5×10⁷ copies per µl) to ensure equal representation of each sample. A two-region 454 sequencing run was performed on a GS FLX Standard PicoTiterPlate (70X75) by using a GS FLX pyrosequencing system according to the manufacturer's instructions (Roche).

Analysis of sequencing data was conducted using Pyrosequencing Pipeline tools at RDP 10 (http://pyro.cme.msu.edu/index.jsp). The RDP Classifier was used to assign 16S rRNA gene sequences to taxonomical hierarchy with a confidence threshold of 80%. Bacterial diversity was determined by sampling-based analysis of operational taxonomic units (OTUs) and showed by rarefaction curves. Comparison of bacterial richness across the samples was performed by Chao1 estimate at the distance of 3%, usually applied to characterize richness at species level [12], [28]. As the Chao1 estimates are dependent on the size of the sequence libraries, the sample sizes from the different subject we equalized by random subtraction.

Bacterial groups in fecal samples from 18 subjects with type 2 diabetes and 18 controls were quantified by qPCR using the 7500 Fast Real-time PCR System (Applied Biosystems, USA) and the primers shown in Table 3 (TAG Copenhagen A/S, Denmark). Genus-specific primers for amplification of Prevotella and Roseburia were designed using 16S rRNA gene sequences from the RDP 10. Sequences were aligned by the ClustalW software provided by the European Bioinformatics Institute (http://www.ebi.ac.uk/Tools/clustalw2/index.html). The target-specific sites were assessed by “Oligo” Primer Analysis Software version 6.71 (Molecular Biology Insights, Inc., USA). Specificity of the primers was evaluated in silico using the nucleotide BLAST, blastn algorithm (http://blast.ncbi.nlm.nih.gov/Blast.cgi). Prevotella-specific primers were targeting Prevotella falsenii, P. copri, P. nigrescens, P. intermedia, P. pallens, P. maculosa and P. oris. Primers targeting Roseburia were specific for species Roseburia faecis, R. hominis, R. intestnalis and R. cecicola.

The qPCR reaction mixture (20 µl) was composed of 0.3 µM of each universal primer or 0.5 µM of each specific primer, 1x Power SYBR Green PCR Master Mix (Applied Biosystems, Calif., USA), and 4 µl fecal DNA added in 10-fold serial dilutions starting from 100-fold dilution to diminish the effect of inhibitors. The amplification program consisted of one cycle of 95°C for 10 min, followed by 40 cycles of 95°C for 15 s, and 60°C for 1 min. Standard curves were constructed for each experiment using 10-fold serial dilutions of bacterial genomic DNA of known concentration. Genomic DNA from L. acidophilus NCFM (ATCC 700396) and from Bifidobacterium animalis subsp. lactis (B. lactis) Bi-07 (ATCC SD5220) was extracted with the use of GenElute Bacterial Genomic DNA Kit (Sigma-Aldrich, Germany) according to the manufacture's instructions. DNA from Roseburia intestinalis DSM14610, Prevotella copri DSM18205, Clostridium leptum DSM 753, Clostridium coccoides DSM935, Clostridium nexile DSM1787 and Bacteroides thetaiotaomicron DSM 2079 was purchased from DSM collection (Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH, Germany). Standard curves were created according to Applied Biosystems tutorials (http://www3.appliedbiosystems.com) and normalized to the copy number of the 16S rRNA gene for each species. For the species which copy number of 16S rRNA operon was not published, it was calculated by averaging the operon numbers of the closest bacterial taxa from the ribosomal RNA database rrnDB (http://ribosome.mmg.msu.edu/rrndb/index.php; [29]). Cell numbers of bacteria in fecal samples were calculated from the threshold cycle values (Ct) and expressed as quantity of bacteria per gram feces [30].

PCA plots were generated by Matlab 2008b using in-house algorithms (http://www.mathworks.com). Differences in bacterial populations between the diabetic and control groups were assessed using the two-sided Wilcoxon Rank Sum test (Statistics Online Computational Resource (SOCR), http://www.socr.ucla.edu/SOCR.html). Correlation between the variables was computed by Spearman Rank correlation provided by Free Statistics Software (version 1.1.23-r4, Office for Research Development and Education, http://www.wessa.net/). The qPCR results were graphically presented by Box and Whisker charts (Microsoft Office Excel 2007) and expressed as medians with quartile ranges (QR).