Aberrant intestinal microbiota in individuals with prediabetes

We included 134 individuals with prediabetes and 134 individuals with normal glucose regulation. In accordance with the criteria from the WHO [10, 11], prediabetes was defined as fasting plasma glucose levels of 6.1–7.0 mmol/l or HbA_1c levels of 42–48 mmol/mol (6.0–6.5%). Normal glucose regulation was defined according to the more strict ADA criteria [12], i.e. fasting plasma glucose <5.6 mmol/l and HbA_1c <39 mmol/mol (5.7%). Individuals with known diabetes and individuals receiving any glucose-lowering drugs were ineligible for inclusion. Individuals with prediabetes were recruited from the Danish part of the Innovative Medicines Initiative – DIabetes REseraCh on patient straTification (IMI-DIRECT) (n = 63) and the Danish study of Functional Disorders (DanFunD) (n = 71) [13, 14]. At the initiation of the present study, 228 Danish individuals were eligible from IMI-DIRECT and 598 were eligible from DanFunD; all individuals fulfilling the abovementioned inclusion and exclusion criteria were included. Individuals with normal glucose regulation were recruited only from DanFunD and matched 1:1 to individuals with prediabetes with respect to sex and age (n = 134). Dietary habits could only be reliably compared between individuals with prediabetes and normal glucose regulation in a subset of the cohort (n = 142) where dietary information was obtained by identical questionnaires.

All participants gave written informed consent and both studies conformed to the principles of the Declaration of Helsinki and were approved by the Regional Scientific Ethical Committee, the Capital Region of Denmark (H-1-2011-166 and H-3-2012-015).

Biochemical analyses were performed on fasting blood samples. Plasma insulin and C-peptide were measured by chemiluminometric immunoassays (DiaSorin Liaison Analyzer, DiaSorin, Saluggia, Italy, at University of Eastern Finland, Kuopio, Finland). Plasma high-sensitivity C-reactive protein (hsCRP) was measured by an immunoturbidimetry assay (Roche/Hitachi analyser, Roche, Basel, Switzerland, at Vejle Hospital, Vejle, Denmark). HbA_1c, plasma glucose and triacylglycerol were measured at the Steno Diabetes Center, Gentofte, Denmark (IMI-DIRECT participants) or Rigshospitalet Glostrup, Copenhagen University Hospital, Denmark (DanFunD participants). HbA_1c was measured by high-performance liquid chromatography (TOSOH analyser, TOSOH, Tokyo, Japan) and plasma triacylglycerol by colorimetric slide analysis (Vitros analyser, Ortho Clinical Diagnostics, Raritan, NJ, USA). In IMI-DIRECT, plasma glucose was measured twice by a glucose hexokinase assay (Konelab Analyser, Thermo Fisher Scientific, Waltham, MA, USA) and the mean was calculated. In DanFunD, plasma glucose was measured by colorimetric slide analysis (Vitros analyser). Insulin resistance was estimated as HOMA-IR: (fasting plasma glucose [mmol/l] × fasting serum insulin [pmol/l])/135.

Body weight (kg) and height (cm) were measured in light indoor clothes and without shoes. Waist circumference was measured midway between the iliac crest and the lower costal margin.

Fresh stool samples were collected by the participants at home and were immediately frozen in their home freezer at about −20°C. Frozen samples were transported to the laboratory using insulating polystyrene foam containers or dry ice. The maximum transportation time in the insulating polystyrene foam containers was 4 h and at delivery study nurses ensured that all samples were frozen. After delivery, the samples were stored at −80°C until DNA extraction. Total faecal genomic DNA was extracted from 300 mg of faecal material using the NucleoSpin Soil kit (Macherey-Nagel, Düren, Germany) [15]. Briefly, the faecal material was suspended in SL2 buffer containing SX enhancer and cell disruption was carried out by bead beating at 30 Hz for 5 min using a TissueLyser instrument (Qiagen, Hilden, Germany).

Faecal microbiota composition was profiled by sequencing the V4 region of the 16S rRNA gene on an Illumina MiSeq instrument (llumina RTA v1.17.28; MCS v2.5) with 515F and 806R primers designed for dual indexing [16] and the V2 Illumina kit (2 × 250 bp paired-end reads). Details on amplification of 16S rRNA genes are provided in the electronic supplementary material (ESM) Methods.

Illumina reads were merged using PEAR [17] and quality filtered by removing all reads that had at least one base with a q-score <20. Quality filtered reads were analysed with the software package QIIME, version 1.8.0 (http://​qiime.​org) [18] as described in ESM Methods. Chimeric sequences, low abundant operational taxonomic units (OTUs; relative abundance <0.002%) and OTUs that could not be aligned with PyNAST [19] were excluded from all analyses. We obtained a mean ± SD of 48,169.6 ± 11,131.7 sequences/sample (range 26,968–87,208). In total, 12,909,447 sequences and 1609 OTUs were included in the analyses. To correct for differences in sequencing depth between samples, 26,968 sequences were randomly subsampled from each sample and included in the analyses for the estimation of α- and β-diversity. OTUs that showed differential abundance among individuals with prediabetes and normal glucose regulation were blasted against the NCBI 16S ribosomal RNA sequences (Bacteria and Archaea) database to obtain a more specific annotation.

From the study population of 268 individuals, we selected four individuals with the poorest (fasting plasma glucose >6.1 mmol/l and HbA_1c >42 mmol/mol [6.0%]) and four individuals with the best glucose regulation (lowest levels of fasting plasma glucose). Case–controls were matched with respect to sex. Frozen stools (250 mg) were obtained from each donor. Case stools and control stools were pooled separately and re-suspended in 4.5 ml PBS prior to each transfer. Mice (48 male wild-type C57BL/6 J, Taconic, Lille Skensved, Denmark; 10 weeks of age) were co-housed with three mice per cage and kept at 22°C under a 12 h light cycle and fed ad libitum with free access to water. After 2 weeks of acclimatisation on a standard chow diet (Altromin 1310, Altromin, Lage, Germany), mice were transferred to a Western diet (D12079 mod.* customised; sucrose was the main carbohydrate source with mixed protein and fat sources) and sham-gavaged with PBS twice a week. After 3 weeks on the Western diet, mice were divided into two equal recipient groups of 18 mice each (six cages per group) and one sham-gavaged control group of 12 mice (four cages per group). The groups were stratified based on magnetic resonance determined fat mass, weight and response to a glucose challenge prior to intervention. For the following 4 weeks, mice were gavaged three times per week with 200 μl of either faecal microbiota (control or case) or PBS (sham) while remaining on the Western diet. Two mice (one case and one control) died during the experiment and were excluded from all analyses.

An OGTT was performed after 4 weeks of gavaging. The mice were fasted for 5 h and gavaged with 3 g glucose/kg lean mass. Blood glucose was measured in tail vein blood at 0, 15, 30, 60, 90 and 120 min using standard Contour Next Test Strips (Bayer Contour, Leverkusen, Germany). Plasma insulin was measured before the glucose bolus and during the OGTT using an electrochemiluminescence assay (Meso Scale Diagnostics, Rockville, MD, USA). Colonisation of the recipient mice by the human gut microbiota was tested by extraction of total genomic DNA from faecal samples collected before and at the end of the 4 week gavage period, as well as caecal samples at the end of the experiment, followed by sequencing of the 16S rRNA gene as described above. Faecal and caecal samples were randomised before DNA extraction.

All animal experiments were conducted in accordance with national Danish guidelines (Amendment number 1306 of 23 November 2007) as approved by the Danish Animal Inspectorate, Ministry of Justice, permission number 2014-15-2934-01027. Mice were kept under specific-pathogen-free conditions and experimental protocols were validated by in-house standard operation procedures.

All animal procedures were approved by the Gothenburg Animal Ethics Committee (152-2015).

Statistical analyses were performed using R version 3.2.1 (www.​r-project.​org) and Stata, version 13.1 (StataCorp, College Station, TX, USA). Differences in baseline characteristics between individuals with prediabetes and those with normal glucose regulation were assessed by use of Wilcoxon rank ∑ tests for continuous variables and χ² tests for categorical variables. Differences in microbiota composition as assessed by β-diversity metrics were tested using multivariate non-parametric ANOVA [20] implemented in the Adonis function (999 permutations) in the R vegan package [21]. Differences in α-diversity were tested using a two-sample unpaired t test and associations between α-diversity and clinical biomarkers were tested using simple linear regression analyses. Only taxa present in at least 50% of the samples were included in the analysis of differential abundance. Differences in abundance of genera and OTUs were tested using a negative binomial wald test implemented in the R DESeq2 package [22, 23]. When examining differential abundance of genera and OTUs, p values were adjusted (p _adj ) for multiple testing by the Benjamini–Hochberg procedure. Correlation between taxa abundances and clinical biomarkers were tested only for differentially abundant taxa (p _adj  < 0.05) using Spearman’s ρ [24]. Differences in blood glucose and plasma insulin levels in mice were tested by two-way ANOVA, repeated measurements with Bonferroni post hoc test, and differences in fasting blood glucose were tested for by Wilcoxon rank ∑ tests.

Five bacterial genera and 36 OTUs differed significantly in abundance between individuals with prediabetes and those with normal glucose regulation. At the genus level, the abundance of Clostridium was decreased, whereas the abundances of Dorea, [Ruminococcus], Sutterella and Streptococcus were increased among individuals with prediabetes (p _adj  = 0.0497 to 1 × 10⁻⁴, Fig. 1 and ESM Table 3). At the OTU level, eight OTUs were increased and 28 were decreased among individuals with prediabetes (Fig. 1 and ESM Table 3). The two OTUs that differed the most between the two groups were classified as a member of the Clostridiales (OTU 146564) and A. muciniphila, which both displayed lower abundance among individuals with prediabetes (mean log₂ fold difference −1.74 (SEM 0.41), p _adj  = 2 × 10⁻³ and −1.65 (SEM 0.34), p _adj  = 4 × 10⁻⁴, respectively). For Faecalibacterium prausnitzii, a prominent butyrate producer, we observed increased abundance of one strain but decreased abundance of another strain in individuals with prediabetes (ESM Table 3).

Next, we examined the correlations between the differentially abundant taxa and clinical biomarkers. The abundance of Clostridium was negatively correlated with fasting plasma levels of glucose, insulin, C-peptide, triacylglycerol and hsCRP, as well as HOMA-IR, BMI and waist circumference (p ≤ 0.01, Fig. 2 and ESM Table 4). Concordant with an increased abundance among individuals with prediabetes, [Ruminococcus] was positively correlated with fasting plasma levels of glucose and C-peptide, as well as HOMA-IR, HbA_1c, BMI and waist circumference (p ≤ 0.03), whereas Dorea was positively correlated with fasting plasma glucose and C-peptide, and BMI and waist circumference (p ≤ 0.03). Notably, the OTUs displaying decreased abundance among individuals with prediabetes correlated most strongly with plasma triacylglycerol and hsCRP (Fig. 3 and ESM Table 4). Two OTUs (OTU 3856408 and OTU 193129) classified as Lachnospiraceae (96% identity to Lachnobacterium bovis strain LRC 5382), two OTUs (OTU 4364405 and OTU 819353) classified as Ruminococcaceae (92% identity to Pseudoflavonifractor capillosus strain ATCC 29799) and OTU 4465124 classified as Clostridium (ESM Table 3) correlated particularly strongly and negatively with fasting plasma levels of glucose, insulin, C-peptide, triacylglycerol and hsCRP, as well as HOMA-IR, BMI and waist circumference (Fig. 3 and ESM Table 4). L. bovis and P. capillosus belong to the Clostridium clusters XIVa and IV, respectively, which are known to contain multiple butyrate-producing bacteria. Two OTUs (OTU 198646 and OTU 307113) classified as Blautia sp. and mapping with 99% identity to Blautia wexlerae strain DSM 19850 (ESM Table 3) were positively correlated with fasting plasma levels of glucose and insulin, as well as HOMA-IR and waist circumference (Fig. 3 and ESM Table 4). OTU 188047, which mapped with 99% identity to Dorea longicatena strain 111–35, correlated positively with BMI and waist circumference. OTU 181167 classified as Dorea sp. and mapping with 99% identity to Coprococcus comes strain ATCC 27758 (ESM Table 3) was positively correlated with plasma glucose and waist circumference (Fig. 3).

Overall, the bacterial community composition assessed by principal coordinates analysis (PCoA) based on unweighted UniFrac distances showed a weak association with prediabetes (p = 0.02, r² = 0.006) (ESM Fig. 1a). PCoA based on weighted UniFrac (p = 0.35, r² = 0.004) and Bray–Curtis distances (p = 0.19, r² = 0.004) did not show any clustering (ESM Fig. 1b,c). As unweighted UniFrac, in contrast to weighted UniFrac, does not take abundance into account, these results demonstrate that the weak association between prediabetes and gut microbiota is driven by the presence/absence status of low abundant taxa.

Bacterial richness estimated as the number of observed OTUs was 543 ± 92 and 562 ± 93 (mean ± SEM) among individuals with prediabetes and normal glucose regulation, respectively (p = 0.09, ESM Fig. 2a). Corresponding values for α-diversity estimated as phylogenetic diversity were 33 ± 0.49 and 34 ± 0.51 (mean ± SEM), respectively (p = 0.06, ESM Fig. 2b). In the total group of 268 individuals, α-diversity was negatively associated with clinical biomarkers related to type 2 diabetes: the higher the α-diversity, the lower the fasting plasma levels of glucose, insulin, C-peptide, triacylglycerol and hsCRP, as well as HOMA-IR, BMI and waist circumference (p = 0.049–0.001; Fig. 4 and ESM Table 5). Plasma triacylglycerol and hsCRP were the clinical biomarkers which were most strongly associated with α-diversity.

To assess whether the aberrant microbiota could directly modulate host glucose metabolism, we transferred faecal microbiota from human donors to conventional C57BL/6 J mice. Prior to transfer, we pooled faeces from four individuals with fasting plasma glucose >6.1 mmol/l and HbA_1c >42 mmol/mol (6.0%) and from four sex-matched individuals with normal levels of fasting plasma glucose. Mean ± SD fasting plasma glucose and HbA_1c were 6.4 ± 0.16 mmol/l and 44 ± 0.5 mmol/mol (6.2 ± 0.05%), respectively, among donors with prediabetes, and 4.5 ± 0.12 mmol/l and 33 ± 0.9 mmol/mol (5.2 ± 0.08%), respectively, among donors with normal glucose regulation. We observed that the prediabetic phenotype did not precipitate in recipient mice possibly due to unsuccessful colonisation of key microbial taxa in the conventional mice as a result of antagonism from the indigenous mouse microbiota (ESM Fig. 3).

In an attempt to avoid competition by the indigenous mouse gut microbiota, we colonised germfree Swiss Webster mice and used a ‘personalised’ approach consisting of the transfer of stool samples from selected individual donors rather than pooled samples. First, we transferred the faecal microbiota of a male donor with screen-detected, treatment-naive type 2 diabetes (two measurements of fasting plasma glucose >7.0 mmol/l) (case donor 1) and a BMI-matched male donor with normal glucose regulation (control donor 1). Faecal transfer did not result in impaired glucose regulation (ESM Fig. 4). Interestingly, mice transplanted with faeces from the donor with screen-detected type 2 diabetes had lower fasting blood glucose (ESM Fig. 4c, p = 0.049), possibly due to the lack of colonisation of key taxa, such as potentially protective strains belonging to the Clostridiales in the control recipients (ESM Table 6 and ESM Fig. 4l). We also performed a second transplantation experiment with faeces from another male donor with screen-detected, treatment-naive type 2 diabetes (120 min plasma glucose >11.1 mmol/l) (case donor 2) and a BMI-discrepant male donor with normal glucose regulation (control donor 2). This experiment showed no difference in blood glucose levels in the fasting state or during GTT between the two groups of mice (ESM Fig. 4d–e). Furthermore, consistent with these results, the microbiota analysis showed low similarity of overall microbiota composition between recipient mice and the control donor, as well as a lack of colonisation of key microbial taxa (ESM Fig. 4 and ESM Table 6).