Duodenal and faecal microbiota of celiac children: molecular, phenotype and metabolome characterization

Molecular analysis of the bacterial community of duodenal biopsies and faecal samples

The dominant microbiota and specific subgroups (Bifidobacteria and Lactobacillus) from stool samples and from duodenal biopsies (mucus and mucosa associated bacteria) were analyzed by PCR (Polymerase chain reaction)-DGGE (denaturing gradient gel electrophoresis). Universal primers targeting V6-V8 regions of the 16S rRNA gene were used. Eubacterial profiles from PCR-DGGE analysis of duodenal biopsies of treated celiac disease (T-CD) children showed high richness with two to eight well resolved and strong bands (Figure 1A). Only the electrophoretic profile of 19 T-CD duodenal biopsy contained one band. Profiles of non-celiac children (HC) had only one to three strong bands. Banding patterns were processed using the Bionumerics software. Pearson correlation coefficients ranged from 4.6 to 99.5%. Except for two duodenal biopsies (33 and 34 HC) which showed high similarity to T-CD samples, all HC banding patterns were grouped together with 98.2% similarity coefficient. The major part of the T-CD samples were grouped together at 95% of the similarity. Overall, DGGE profiles of the PCR amplicons obtained with primers Lac1 and Lac2 had two strong, common and well-resolved bands, and a few bands with low intensity (Figure 1B). High similarity was found among samples belonging to T-CD and HC groups. Most of the T-CD and HC duodenal biopsies were grouped together at ca. 90% of similarity and all samples at 72.9%. Sequencing of the DGGE bands revealed the common presence of L. plantarum (band a). Although Lac1 and Lac2 primers were commonly used to detect Lactobacillus species [9, 24, 25], human DNA (band b) was also found. Finally, no PCR amplicons were found by using three different sets of primers targeting the Bifidobacteria group. This suggested that Bifidobacteria were probably absent from duodenal biopsies of both T-CD and HC.

Compared to duodenal biopsies, the PCR-DGGE profiles of faecal samples were more rich. Although fingerprints contained many well-resolved and strong bands, unresolved bands or very weak separate fragments were present in some regions of the gel. The PCR-DGGE profiles from universal primers (Table 1) targeting V6-V8 regions of the 16S rRNA gene were very rich in bands quite different for each of the 34 children (Figure 2A). Only some common bands were present. The uniqueness of the patterns was confirmed by cluster analysis. The values of Pearson similarity were always low. The mean similarity coefficient was 24.1%. No clustering differentiated T-CD and HC samples. Figure 2B shows the PCR-DGGE profiles from primers Lac1 and Lac2 specific for Lactobacillus group. Depending on the faecal sample, one to four strong and well-resolved amplicons were detected. Nevertheless, the values of Pearson similarity coefficient were low and all samples grouped together at ca. 4.2%. According to PCR-DGGE profiles of duodenal biopsies, the UPGMA clusterization grouped separately T-CD and HC samples with the only exceptions of sample 5 T-CD coupled to HC, and samples 22, 20 and 25 HC which showed high similarity to T-CD. Anyway significant differences were present within groups of T-CD or HC children.

Primer	Primer sequence (5'-3')	Amplicon size (bp)	Annealing temperature (°C)	DGGE gradient (%)	Target group	Reference
V6-V8: F968-GC V6-V8: R1401	GC clampa-AACGCGAAGAACCT CGGTGTGTACAAGACCC	489	55	45-55 (feces) 40-65 (biopsies)	Eubacteria	This study
g- Bifid F g-Bifid R-GC	CTCCTGGAAACGGGTGG GC clampa-GGTGTTCTTCCCGATATCTACA	596	65	45-60	Bifidobacterium	This study
Lac1 Lac2GC	AGCAGTAGGGAATCTTCCA GC clampa - ATTYCACCGCTACACATG	380	61	35-50 (feces) 35-70 (biopsies)	Lactobacillus groupb	[24]
Bif164-f Bif662-GC-r	GGGTGGTAATGCCGGATG GC clamp a- CCACCGTTACACCGGGAA	520	62	45-55	Bifidobacterium	[47]
Bif164-GC-f Bif662-r	GC clamp a - GGGTGGTAATGCCGGATG CCACCGTTACACCGGGAA	520	62	45-55	Bifidobacterium	[47]

^aGC clamp sequence: CGCCCGCCGCGCCCCGCGCCCGGCCCGCCGCCCCCGCCCC.

^bLactobacillus group comprises the genera Lactobacillus, Leuconostoc, Pediococcus and Weisella.

As shown by PCR-DGGE analysis, all faecal samples contained Bifidobacterium DNA (Figure 2C). The level of similarity among faecal samples varied from 16.8 to 100%. Identical profiles were found for some T-CD stool samples (numbers 1, 8 and 12). The UPGMA analysis grouped most of T-CD and HC profiles separately, with similarity Pearson coefficients ≥ 48%.

Enumeration of cultivable bacteria

Selective media were used to enumerate cultivable cells of the main microbial groups (Figure 3). No statistical difference (P = 0.161) was found between T-CD and HC for total microbes. The median values of presumptive lactobacilli and enterococci of T-CD was lower (P = 0.035) than those of HC. The number of presumptive Bifidobacteria significantly (P = 0.023) differed between T-CD (median value of 5.34 ± 0.020 log CFU/g) and HC (median value of 6.72 ± 0.023 log CFU/g). Compared to HC, significantly (P = 0.014) higher counts of presumptive Bacteroides, Porphyromonas and Prevotella, presumptive staphylococci/micrococci and Enterobacteria were found in faecal samples of T-CD. Presumptive Salmonella, Shighella and Klesbiella, and Clostridium did not significantly (P = 0.830) vary between groups. Total anaerobes were the highest (P = 0.018) in HC.

Identification and typing of lactic acid bacteria

To exclude clonal relatedness and to characterize lactobacilli and enterococci, three primers (M13, P4 and P7), with arbitrarily chosen sequences, were used for RAPD-PCR analysis. The number of strains identified for each faecal sample is shown in Table 2 and Figures 4, 5 and 6A-B. The percentage of strains identified as lactobacilli significantly (P = 0.011) differed between T-CD (ca. 26.5%) and HC (ca. 34.6%) groups.

Volatile organic compounds (VOC)

VOC (107 compounds) were identified from faecal and urine samples (Table 3 and Additional file 1, Table S1). VOC were grouped according to chemical classes: esters (14 compounds identified); sulfur compounds (3), ketones (21), hydrocarbons (15), aldehydes (16), alcohols (15), alkane (4), alkene (1), aromatic organic compounds (6), hetpane (1) and short chain fatty acids (SCFA) (11). During sampling, the level of VOC of each child did not differ (P > 0.05). On the contrary, high variability was found among children. Statistical differences (P < 0.05) were found between T-CD and HC children. As expected, faecal samples had higher level of VOC compared to urines. The median value of esters was higher than in HC children. Nevertheless, the levels of ethyl-acetate, octyl-acetate, propyl-butyrate, propyl-propanoate and butyl 2-methylbitanoate were higher than in faecal samples of T-CD. Among sulfur compounds, carbon disulfide was at higher level than in faecal samples of HC. Dimethyl trisulfide and dimethyl disulfide were at higher level than in the urine samples of HC. With a few exceptions, hydrocarbons were found at higher levels than in urine and, especially, faecal samples of HC. Faecal samples of HC contained higher median values of aldehydes compared to T-CD. The level of aldehydes did not differ (P > 0.05) between urine samples of T-CD and HC. Compared to faecal samples of HC, some alcohols (e.g., 1-octen-3-ol, ethanol and 1-propanol) were present at higher level in T-CD. Median values of alkane and alkene did not significantly (P > 0.05) differ between T-CD and HC. Overall, faecal samples of T-CD showed the lowest levels of aromatic organic compounds. The median value of total short chain fatty acids (SCFA) was significantly (P < 0.05) higher in faecal samples HC compared to T-CD. Major differences were found for isocaproic, butyric and propanoic acids (P < 0.038, 0.021, and 0.012, respectively). On the contrary, acetic acid was higher in T-CD compared to HC samples. The differences of the metabolomes between faecal or urine samples of T-CD and HC was highlighted through CAP analysis which considered only significantly different compounds (Figure 7A and 7B). Variables appearing with negative values represent bins whose values decreased in T-CD compared to HC samples. On the contrary, variables represented with bars pointing to the right indicate bins whose values were the highest in T-CD samples.

Chemical class	Treated celiac disease (T-CD)children				Non-celiac children (HC)
	Faeces		Urines		Faeces		Urines
	Median	Range	Median	Range	Median	Range	Median	Range
Esters	20.31b	0 - 846.97	0.47c	0 - 40.00	47.73a	1.83 - 496.83	0.99c	0 - 8.05
Sulfur compounds	214.83b	0 - 890.86	1.46c	0 - 25.44	387.07a	0 - 499.88	3.49c	0 - 63.67
Ketones	90.88b	0 - 2402.50	54.01c	0 - 295.03	112.83a	0 - 416.20	64.49c	0 - 458.78
Hydrocarbons	16.69b	0 - 1327.15	4.25c	0 - 67.07	119.13a	0.22 - 635.25	3.14c	0.15 - 62.56
Aldehydes	17.59c	0 - 512.28	64.31a	0.34 - 166.31	37.46b	2.08 - 365.25	73.37a	0.50 - 199.56
Alcohols	230.14a	0 - 2311.29	2.25c	0 - 17.5	122.56b	0 - 934.22	2.14c	0 - 34.96
Alkane	6.73a	0 - 653.61	0.3b	0.05 - 1.57	9.37a	0 - 432.74	0.43b	0 - 1.47
Alkene	0a	0 - 32.51	0a	0	0a	0 - 31.99	0a	0
Aromatic organic compounds	178.24b	0 - 143.67	2.10c	0.04 - 28.16	480.20a	233.74 - 993.94	2.78c	0 - 16.30
Heptane	23.01a	0 - 837.50	0c	0 - 1.37	26.37a	0 - 65.75	0.34b	0 - 2.37
Short chain fatty acids (SCFA)	21.64a	0 - 1438.28	3b	0.08 - 31.14	27.85a	0 - 1037.50	3.82b	1.44 - 24.87

Data are the means of three independent experiments (n = 3) for each children.

^a-cMeans within a row with different superscript letters are significantly different (P < 0.05).

¹H Nuclear Magnetic Resonance (NMR) metabolite profiling of faeces and urine samples

Subjects

Collection of duodenal biopsies, faecal and urine samples

Each child had fasted overnight, and biopsies, which were taken always from the second duodenum, faecal and urine were collected in the morning pre-prandial. Urine samples were collected after the second mittus. Each child provided a duodenal biopsy and three faecal and urine samples over the time. Duodenal biopsy specimens were obtained from the second duodenum by upper intestinal endoscopy, frozen immediately at -80°C and kept until further processing. After collection, faeces (ca. 15 g), contained in sterile plastic box, were immediately mixed (1:1 wt/wt) with the Amies Transport medium (Oxoid LTD, Basingstoke, Hampshire, England) under anaerobic conditions (AnaeroGen, Oxoid LTD). Samples were immediately subjected to analysis (plate counts) or frozen at -80°C (DNA extraction). The urine samples were collected into pre-labeled sterile collections cups. Three aliquots per patient were immediately frozen and stored at -80°C until use.

DNA extraction from duodenal biopsies and faecal samples

Biopsies specimens, the average weight was ca. 3.5 mg (biopsies are not usually weighted, however all were taken by the same endoscopist using the same biopsy forceps), were homogenized using a sterile plastic pestle in 200 μl of 20 mM Tris-HCl, pH 8.0, 2 mM EDTA buffer. The homogenate was subjected to mechanical disruption in a FastPrep^® instrument (BIO 101) and total DNA was extracted with a FastDNA^® Pro Soil-Direct Kit (MP Biomedicals, CA., USA) according to the manufacturer's instructions. Three samples of faecal slurry of each child were mixed and used for DGGE analysis [43]. An aliquot of about 300 μl of each faecal slurry sample containing 150 μg of faeces was diluted in 1 ml of PBS-EDTA (phosphate buffer 0.01 M, pH 7.2, 0.01 M EDTA). After centrifugation (14,000 × g at 4°C for 5 min), the pellet was washed two times to decrease the content of PCR inhibitors. The resulting pellet was resuspended in 300 μl of PBS-EDTA and used for DNA extraction [44] with a FastPrep instrument as above. The final product was 100 μl of application-ready DNA both for stool and tissue samples [45]. Quality and concentration of DNA extracts were determined in 0.7% agarose-0.5X TBE gels stained with Gel Red ™ 10,000X (Biotium, Inc.) and by spectrophotometric measurements at 260, 280 and 230 nm using the NanoDrop^® ND-1000 Spectrophotometer (ThermoFisher Scientific Inc., MI., Italy).

Polymerase chain reaction (PCR) amplification and denaturing gel electrophoresis (DGGE) analysis

DNA isolated from duodenal biopsy and faecal samples was subsequently used as the template in PCR assays applying eubacterial universal and group-specific 16S rRNA gene primer sets. All primers used in this study are listed in Table 1. The forward or the reverse primer of each set was extended with a 40 mer GC-clamp at the 5' end to separate the corresponding PCR products in the gradient gel [46]. The specificity of each primer pair was experimentally tested by using DNA extracted from the following bacteria species: Bacteroides fragilis DSM 2151, Bifidobacterium bifidum DSM 20082, L. plantarum ATCC 14917, Weissella confusa DSM2196, P. pentosoceus DSM 20336, Leuconostoc lactis DSM 20202, E. durans DSM 20633, E. faecium DSM 2918, Clostridium coccoides DSM 935, Staphylococcus aureus DSM 20714, Enterobacter aerogenes DSM 30053, Escherichia coli DSM 30083 and Yersinia enterocolitica DSM 4780. Each primer set gave positive PCR results for the corresponding target bacteria and did not cross-react with any of the non target microorganisms. Each PCR mixture contained 80 - 100 ng and 40 ng of template DNA extracted from bioptic materials and faecal samples respectively, 50 pmol of each primer, 10 nmol of each 2'-deoxynucleoside 5'-triphosphate (dNTP), 3 U of Taq DNA polymerase (EuroTaq, EuroClone, Italy) and 2.5 mM MgCl₂ in a buffered final volume of 50 μl. The following PCR core program was used for the first three primer pairs listed in Table 1: initial denaturation at 95°C for 3 min; 30 cycles of denaturation at 95°C for 20 s, annealing at primer-specific temperature for 45 s and extension at 72°C for 1 min; and final extension at 72°C for 7 min. DNA extracted from duodenal biopsies was amplified by two additional set of primers targeting Bifidobacterium group and the PCR thermocycling program used for both Bif164-f/Bif662-GC-r and Bif164-GC-f/Bif662-r was: 94°C for 5 min; 35 cycles of 94°C for 30 s, 62°C for 20 s, and 68°C for 40 s; and 68°C for 7 min [47]. PCR amplification products were checked by electrophoresis in 1.5% agarose Gel Red 10,000X stained gels and stored at -20°C. Amplicons were separated by DGGE, using the Bio-Rad DCode™ Universal Mutation detection System (Bio-Rad Laboratories, Hercules, CA, USA). Different linear denaturing gradients of urea and formamide were applied depending on the amplified target sequence and type of samples (Table 1). The denaturing gradient conditions proposed by Vanhoutte et al. [43] were modified as described below. For eubacterial amplicons the denaturing gradient was 45-55% for faecal samples and 40-65% for duodenal biopsies, respectively; Lac1-Lac2GC PCR products relative to faecal and biopsies samples were separated in 35-50% and 35-70% denaturing gradient, respectively and, finally, g-BifidF/gBifidR-GC amplicons from faecal samples were resolved by 45-60% gradient. Gels were electrophoresed at 60°C at 75 V for 15 h. Sybr Green I stained gels were photographed and acquired by the Bio-Rad Gel Doc 2000 documentation system (Bio-Rad Laboratories). To compensate for internal distortions occurring during the electrophoresis, binding patterns were digitally aligned using the Bionumerics software version 4.5 (Applied Maths, Belgium) by comparison with an external reference pattern obtained by appropriately mixing DGGE marker II, III and V (Nippon gene, Tokyo), depending on the gradient used. This normalization enabled comparison among DGGE profiles from different gels, provided that these were run under comparable denaturing and electrophoretic conditions. Comparison and cluster of profiles were carried out using the unweigthed pair-group method with the arithmetic average (UPGMA) clustering algorithm based on the Pearson product-moment correlation coefficient (r) [25, 48] and resulted in a distance matrix. DGGE fragments from primers Lac1 and Lac2 were cut out using sterile scalpel. The DNA of each band was eluted in 100 μl of sterile water overnight at 4°C. Two μl of the eluted DNA were reamplified as described above. PCR products were separated by electrophoresis on 1.5% (wt/vol) agarose gel (Gibco BRL, France) stained with ethidium bromide (0.5 μg/ml). The amplicons were eluted from gel and purified by the GFXTM PCR DNA and Gel Band Purification Kit (GE Healthcare Life Sciences, Milan, Italy). DNA sequencing reactions were performed by MWG Biotech AG (Ebersberg, Germany). Sequences were compared to the GenBank database with the BLAST program.

Enumeration of cultivable bacteria

Diluted faecal samples (20 g) were mixed with 80 ml sterilized peptone water and homogenized. Counts of viable bacterial cell were carried out as described by Macfarlane et al. [45, 49] The following selective media were used: MRS agar (lactobacilli); Beerens agar (bifidobacteria); Baird-Parker (staphylococci and micrococci); Blood Azide agar (enterococci); Wilkins-Chalgren agar (total anaerobes); Wilkins-Chalgren agar plus GN selective supplements (Bacteroides, Porphyromonas and Prevotella); Reinforced Clostridial Medium supplemented with 8 mg/l novobiocin, 8 mg/l colistin (Clostridium), MacConkey agar No2 (enterobacteria); and nutrient agar (total anaerobes) [50].

Lactic acid bacteria isolation

Fifteen to twenty colonies of presumptive lactic acid bacteria were isolated from the highest plate dilutions of MRS and Blood Azide agar media. Gram-positive, catalase-negative, non-motile rods and cocci isolates were cultivated in MRS or Blood Azide broth (Oxoid Ltd) at 30, 37 or 42°C for 24 h, and re-streaked into the same agar media. All isolates considered for further analyses showed the capacity of acidifying the liquid culture medium. All cultures were stored at -80°C in 10% (vol/vol) glycerol.

DNA extraction and molecular identification by 16S rRNA, pheS and recA genes sequencing

Total DNA of presumptive lactic acid bacteria isolates was extracted from 2 ml samples of overnight cultures grown at 37°C in MRS or Blood Azide broth. Total DNAs were obtained as described by De Los Reyes-Gavilàn et al. [51]. The concentration and purity of DNA was assessed by a NanoDrop^® ND-1000 Spectrophotometer (Thermo Fisher Scientific Inc.). A primer pair (Invitrogen Life Technologies, Milan, Italy), LpigF/LpigR (5'-TACGGGAGGCAGCAGTAG-3' and 5'-CATGGTGTGACGGGCGGT-3') [52], corresponding to the position 369-386, and 1424-1441, respectively, of the 16S rRNA gene sequence of L. mucosae, (accession number AF126738) was used to amplify the 16S rRNA gene fragment of presumptive lactic acid bacteria. Fifty microliters of each PCR mixture contained 200 μM of each dNTP, 1 μM of both forward and reverse primer, 2 mM MgCl₂, 2 U of Taq DNA polymerase (Invitrogen Life Technologies) in the supplied buffer, and approximately 50 ng of DNA. PCR amplification was carried out using the GeneAmp PCR System 9700 thermal cycler (Applied Biosystems, USA). PCR products were separated by electrophoresis on 1.5% (wt/vol) agarose gel (Gibco BRL, France) stained with ethidium bromide (0.5 mg/ml). The amplicons were eluted from gel and purified by the GFX™ PCR DNA and Gel Band Purification Kit (GE Healthcare Life Sciences, Milan, Italy). DNA sequencing reactions were carried out by MWG Biotech AG (Ebersberg, Germany) using both, forward and reverse, primers. Taxonomic identification of strains was performed by comparing the sequences of each isolate with those reported in the Basic BLAST database http://www.ncbi.nlm.nih.gov. Primers casei/para were used to discriminate between the species L. casei, L. paracasei and L. rhamnosus [53]. Primers pheS-21-F/pheS-23-R were used to identify Enterococcus species [54]. Primers designed on recA gene were also used to discriminate between the species L. plantarum, L. pentosus and L. paraplantarum. Part of the recA gene was amplified using the degenerate primer pair (MWG Biotech AG, Ebersberg, Germany) recALb1F 5'-CRRTBATGCGBATGGGYG-3'/recALb1R 5'-CGRCCYTGWCCAATSCGRTC-3' derived from the homologous regions of the recA gene sequences of L. plantarum (accession no. AJ621668). PCR reactions and separation, and purification and sequencing of amplicons were carried out as described for 16S rRNA gene.

Genotypic characterization by Randomly Amplified Polymorphic DNA-Polymerase Chain Reaction (RAPD-PCR) analysis

Genomic DNA from each isolates was extracted as described above. Three oligonucleotides, P4 5'-CCGCAGCGTT-3', P7 5'-AGCAGCGTGG-3' and M13 5'-GAGGGTGGCGGTTCT-3' [55, 56], with arbitrarily chosen sequences, were used for isolates biotyping. Reaction mixture and PCR conditions for primers P4 and P7, and primer M13 were according to De Angelis et al. [55, 56]. PCR products (15 μl) were separated by electrophoresis at 100 V for 200 min on 1.5% (wt/vol) agarose gel and DNA was detected by UV transillumination after staining with ethidium bromide (0.5 μg/ml). Molecular sizes of the amplified DNA fragments were estimated by comparison with 1-kb DNA molecular size markers (Invitrogen Life Technologies). RAPD-PCR profiles were acquired by Gel Doc EQ System (Bio-Rad Laboratories) and compared using Fingerprinting II Informatix™ Software (Bio-Rad). The similarity of the electrophoretic profiles was evaluated by determining the Dice coefficients of similarity and using the UPGMA method.

Gas-chromatography mass spectrometry/solid-phase microextraction (GC-MS/SPME) analysis

After preconditioning according to the manufacturer's instructions, the carboxen-polydimethylsiloxane coated fiber (85 μm) and the manual SPME holder (Supelco Inc., Bellefonte, PA, USA) were used. Before head space sampling, the fiber was exposed to GC inlet for 5 min for thermal desorption at 250°C. Three grams of faecal sample were placed into 10 ml glass vials and added of 10 μl of 4-methyl-2-pentanol (final concentration of 4 mg/l), as the internal standard. Samples were then equilibrated for 10 min at 45°C. SPME fiber was exposed to each sample for 40 min. Both phases of equilibration and absorption were carried out under stirring condition. The fiber was then inserted into the injection port of the GC for 5 min of sample desorption. GC-MS analyses were carried out on an Agilent 7890A gas-chromatograph (Agilent Technologies, Palo Alto, CA, USA) coupled to an Agilent 5975C mass selective detector operating in electron impact mode (ionization voltage 70 eV). A Supelcowax 10 capillary column (60 m length, 0.32 mm ID) was used (Supelco, Bellefonte, PA, USA). The temperature program was: 50°C for 1 min, 4.5°C/min to 65°C and 10°C/min to 230°C, which was held for 25 min. Injector, interface and ion source temperatures were 250, 250 and 230°C, respectively. The mass-to-charge ratio interval was 30-350 a.m.u. at 2.9 scans per second. Injections were carried out in splitless mode and helium (1 ml/min) was used as the carrier gas. Sodium 3-(trimethylsilyl)propionate-2,2,3,3-d4 (TSP) was used as the internal standard. Identification of molecules was carried out based on comparison of their retention times with those of pure compounds (Sigma-Aldrich, Milan, Italy). Identification was confirmed by searching mass spectra in the available databases (NIST version 2005 and Wiley Vers. 1996) and literature [57]. Quantitative data of the identified compounds were obtained by interpolation of the relative areas versus the internal standard area [33].

¹H Nuclear Magnetic Resonance (NMR) spectroscopy analysis

To study the water soluble fraction of the faeces by means of ¹H NMR spectroscopy, 40 mg of thawed faecal or urine mass were thoroughly homogenized by vortex-mixing with 400 μl of cold deuterium oxide (D₂O) at pH 7.4 ± 0.02, containing 1 mM TSP as the internal standard. Mixtures were centrifuged at 14,000 rpm for 5 min and the supernatant was collected. To ensure the complete recovery of the water soluble species and highly reproducible spectra, the extraction procedure was repeated two times [31]. ¹H NMR spectra were acquired on the collected supernatants, with no further treatments, at 300 K on a Mercury-plus NMR spectrometer from Varian, operating at a proton frequency of 400 MHz. Residual water signal was suppressed by means of presaturation. ¹H NMR spectra were processed by means of VNMRJ 6.1 software from Varian. To minimize the signals overlap in crowded regions, all free induction decays (FID) were multiplied by an exponential function equivalent to a -0.5 line-broadening factor and by a gaussian function with a factor of 1. After manual adjustments of phase and baseline, the spectra were scaled to the same total area, in order to compare the results from samples of different weight and water and fiber content. The spectra were referenced to the TSP peak, then digitized over the range of 0.5 - 10 ppm. By means of R scripts developed in-house the residual water signal region, 4.5 - 5.5 ppm, was excluded from the following computations [58]. To compensate for chemical-shift perturbations, the remaining original data points were reduced to 218 by integrating the spectra over 'bins', spectral areas with a uniform size of 0.036 ppm. A 34 × 218 bins table was thus obtained for statistical analysis. As some parts of the spectra are very crowded, some bins may contain peaks pertaining to different molecules. In order to consider this potential source of error the bins containing peaks ascribed to the same molecules were not summed up [33].

Statistical analysis

All data coming from culture-dependent analysis and metabolomic analysis were obtained at least in triplicates. The analysis of variance (ANOVA) on culture-dependent analysis, GC-MS/SPME and ¹H-NMR analysis, was carried out on transformed data followed by separation of means with Tukey's HSD, using a statistical software Statistica for Windows (Statistica 6.0 per Windows 1998, (StatSoft, Vigonza, Italia). Letters indicate significant different groups (P < 0.05) by Tukey's test. Canonical discriminant Analysis of Principal coordinates (CAP) analysis was carried out for GC-MS/SPME data [33]. This was preferred to the more common Canonical Discriminant Analysis (CDA), because it does not assume any specific distribution of the data, thus giving more robust results in the case of reduced number of samples. The CAP constrained ordination procedure that was carried out is summarized as follows: (i) data were reduced by performing a Principal Coordinate analysis (PCO) of the parameters, using the dissimilarity measure calculated on euclidean distances; (ii) an appropriate number of PCO was chosen non-arbitrarily, which maximizes the number of observations correctly classified; (iii) the power of classification was tested through a leave-one-out procedure; and (iv), finally, a traditional canonical analysis on the first PCO was carried out. The total variance obtained in PCO used to perform CAP was 70 and 73% for faeces and urine data, respectively. The hypothesis of no significant difference in the multivariate location within groups was tested using the trace statistic based on 9999 permutations [33]. The permutation test performed correctly assigns ca. 90% of the samples.