Supplemental Material

1H NMR spectroscopy

Spectra of biofluids and polar extracts were acquired on a Bruker Avance DRX 600MHz NMR Spectrometer (Bruker Biopsin, Rheinstetten, Germany) operating at 600.13MHz using a standard 1-dimensional (1D) pulse sequence (16) [recycle delay (RD)-90°-t1-90°-tm-90°-acquire free induction decay (FID)] with water suppression applied during RD of 2s and the mixing time (tm) of 100ms and a 90° pulse set approximately at 10μs. For each spectrum, a total of 256 scans were accumulated into 32K data points with a spectral width of 12019Hz. A range of 2D NMR spectra were performed on the same equipment for selective samples, including correlation spectroscopy (COSY) (2), total correlation spectroscopy (TOCSY) (10) and heteronuclear single quantum coherence (HSQC) NMR spectroscopy (3). The FIDs were multiplied by an exponential function corresponding to 0.3Hz line broadening. All spectra were manually phased, baseline corrected and calibrated to the chemical shift of TSP (δ 0.00). Metabolites were assigned using data from literature (4, 6, 9, 16, 18) and confirmed by 2D NMR experiments.

Polar phase extraction

Kidney cortex, proximal colon and plasma were homogenized in 1mL of methanol/water (1:1) for 5 min at T = 1/25 (name tissue lyser). The homogenate was transferred to glass tubes where 500µL of chloroform was added and vortexed for 1min. Polar and organic phases were separated by centrifugation at 4000 rpm at 4°C for 20 min. The polar phase (top layer) was collected and evaporated overnight under a fume cupboard. It was then freeze-dried and redissolved in 600µL of D2O containing 10% of water and 0.05% of sodium 3-(tri-methylsilyl)propionate-2,3-d4 (TSP) as an 1H NMR reference and transferred to 5mm NMR glass tubes for analysis.

High resolution magic angle spinning 1NMR spectroscopy of intact liver biopsies

NMR spectra of intact liver samples were acquired on a Bruker Avance DRX 600MHz Spectrometer (Bruker Biospin, Rheinstetten, Germany) operating at 600.13MHz by high resolution magic angle spinning technique and using a Carr-Purcell-Meiboom-Gill (CPMG) spin-echo pulse sequence with water presaturation (14). FIDs were collected into 32K data points using 128 scans with a spectral width of 12000Hz and were multiplied by an exponential function corresponding to 1Hz line broadening. All spectra were manually processed and calibrated to the chemical shift of glucose (5.223ppm).

Measurement of CYP-dependent testosterone hydroxylation activities

Liver homogenates were prepared in cold phosphate buffer (20% monobasic sodium phosphate and 80% dibasic sodium phosphate), pH 7.4, with a Potter-Elvehjem homogenizer equipped with a fitting Teflon pestle and centrifuged for 20min at 10000 g at 4°C. The microsomal fraction was obtained by centrifuging the supernatant fluid for 70min at 105000 g for at 4°C. Then, the pellet of microsomes was suspended into phosphate buffer + 20% glycerol (1mL/mg of liver tissue) and stored at -80°C until analysis.

Testosterone hydroxylation activities were measured as follows: 1mg of microsomal protein were pre-incubated with glucose-6-phosphate (5mM), glucose-6-phosphate dehydrogenase (1 unit), 4-MA (17-β-N,N-diethylcarbamoyl-4-methyl-4-aza-5-α-androstan-3-one) (2.5 µM), and 50 µM [4]-14C-testosterone (2 × 105 dpm) in 0.1 M HEPES buffer (pH 7.4, 1 ml final volume), containing 0.1 mM EDTA, for 5 min at 37°C in a shaking water bath. The reactions were initiated by the addition of 25 µl of 0.5 mM NADPH and maintained at 37°C for 30 min. The formation of testosterone metabolites was linear up to 45 min in mouse liver microsomes but incubation was stopped after 30 min to avoid any secondary reactions that could be significantly detectable at a longer incubation time. Incubation was stopped by direct extraction of metabolites with 4 ml methanol maintained at 0°C in ice. Proteins were pelleted by centrifugation at 7500 g at 4°C for 15min and the solvent was decanted into glass vials, which were then evaporated under vacuum using a Savant Speed Vac centrifuge and reconstituted in methanol (500µl) and stored at -20°C until HPLC analysis.

The different radioactive metabolites were separated on an analytical Uptisphere ODB column (150 × 4.6 mm i.d., 3 µm) from Interchim (Montluçon, France) coupled to an analytical Uptisphere ODB precolumn (30 × 4.6 mm i.d., 3 µm), using a Pye Unicam 4100 HPLC system connected to a PU4110 UV detector using a detection wavelength set at 254 nm. Radioactivity was online recorded on a Radiomatic model A-515 flow-through detector (PerkinElmer), equipped with a 0.5-ml flow cell. The HPLC eluate was blended 1:2.5 (v/v) with Ultima-Flo M (PerkinElmer) scintillant, and the radiochemical signal was stored and processed by the Radiomatic detector. The following ternary gradient was used: from 0 to 20 min, solvent A: 100%, then from 20 to 30 min, a linear gradient from 100% solvent A to 100% solvent B, which was maintained to 35 min, then from 35 to 40 min, a linear gradient from 100% solvent B to 50% solvent B and 50% solvent C, which was maintained to 45 min, then a linear gradient from solvent B/C (50/50) to 100% solvent C from 45 to 50 min, which was maintained to 60 min. The solvents A, B and C all comprised of 6% tetrahydrofuran, 0.2% acetic acid and 20% methanol and 74% water for solvent A, and 25% methanol and 69% water for solvent B, and 35% acetonitrile and 59% water for solvent C. Metabolites were evaporated under a nitrogen flux and reconstituted in the solvent A just before analysis and then manually injected on a 500-µl sample loop Rheodyne injector. A flow rate of 1.2 ml/min was used, and all HPLC analyses were performed at 35°C. The following reference compounds 7α-hydroxytestosterone, 15α-hydroxytestosterone, 6β-hydroxytestosterone, 16α-hydroxytestosterone, 11α-hydroxytestosterone, 16β-hydroxytestosterone, 2β-hydroxytestosterone, and androstenedione were detected at 14.20, 16.80, 20.30, 27.40, 33.10, 35.90, 42.30, and 50.80 min, respectively.

Quantitative RT-PCR of mRNA extracted from liver samples

Total RNA isolation was performed using TriPure Isolation Reagent (Roche Applied Science). Approximately 30mg of liver tissue were homogenized in 1mL of TriPure Isolation Reagent using an automatic homogenizer (RETSCH MM300) for 2min at a frequency of 29sec-1. Total RNA was then isolated according to the manufacturer protocol.

Once DNase treatment and purification of the total RNA were performed using the RNeasy MinElute Cleanup kit (Qiagen), the amount of RNA was estimated by the absorbance ratio at 260nm/280nm and the quality of the RNA samples was determined by electrophoresis through denaturing agarose gels.

RNA was reverse-transcribed to complementary DNA (cDNA) using a High Capacity cDNA Archive kit in a final volume of 25µl containing 12.5µL of 1x RT buffer and 12.5µL containing 2µg of total RNA. Samples were incubated at 37°C for 2 hours and stored at –20°C. Detailed sequences and references for primers previously tested and validated are shown in Supporting Table 1 in the Supporting data. Nucleotide primer pairs were bought from Operon (Operon Biotechnologies GmbH, Cologne, Germany) (HYPERLINK "http://www.operon.com" http://www.operon.com). Amplicons were all between 60 and 140 nucleotides long. All PCR reactions were performed using an ABI Prism 7900 sequence detection system (Applied Biosystems, Foster City, CA). PCR was performed using the Absolute Blue QPCR SYBR Green ROX Mix (ABgene, Surrey, UK). The thermal cycling conditions comprised an initial denaturation step at 95°C for 15min and 50 cycles at 95°C for 15s and 60°C for 1 min and a dissociation stage. Each sample was then normalized using the 2Ct method, on the basis of its 18S ribosomal RNA content.

Microbiota profiling

DGGE analysis

The method used was derived from the one published by Muyzer et al. (15). DNA was isolated from faeces of both groups (conventional and germ- free/ex-germ-free) at D0, D5 and D20 using the FastDNA® SPIN Kit and the FastPrep® Instrument (Qbiogene, Inc., CA).

The V3 variable region of the 16S rDNA gene of bacteria was amplified by PCR using a combination of commercially synthesized primers (Sigma Genosys, UK), P2 (5’-ATTACCGCGGCTGCTGG-3’) and P3 (5’-CGCCCGCCGCGCGCGGCGGGCGGGG CGGGGGCACGGGGGGCCTACGGGAGGCAGCAG-3’) that contains the GC-clamp (15). PCR amplification was performed with a MJ Research PTC-200 Peltier Thermal Cycler (GRI, Essex, UK) using a PCR mixture (50 μL) of 10 μL of 5 x MgCl2-free reaction buffer (Promega, Southampton, UK), 1 μL of deoxyribonucleoside triphosphates (dNTPs, 10 mM; Promega), 3 μL of MgCl2 (25mM, Promega), 2.5 μL of each primer (10 μM), 0.2 μL of Taq DNA polymerase (5U/μL; Promega) and 5 ng of template DNA. Touchdown PCR was performed with an initial denaturation step of 94°C for 5 min. Following the hot start which minimizes nonspecific annealing primers to nontarget DNA, 2 cycles at 94°C for 1min, 65°C for 1 min and 72°C for 1 min were run. Then, the annealing temperature was subsequently decreased by 1°C every second cycle until a touchdown temperature of 55°C, at which point 5 additional cycles were carried out (27 cycles in total). A final elongation step at 72°C for 7min was also performed (15). All amplification products were analysed by gel electrophoresis on 1.5 % (w/v) agarose gel containing ethidium bromide (0.4 mg/mL).

DGGE was performed with BDH DGGE V20-HCDC Unit (Merck Eurolab Ltd., Dorset, UK). Amplicons were separated on 8% (w/v) polyacrylamide gels (acrylamide-bisacrylamide 37.5:1, BioRad, UK). A linear denaturing gradient of 30 % to 70 % (100 % denaturant was defined as 7 M urea and 40 % (v/v) deionised formamide) was used. Electrophoresis was performed at a constant voltage of 200 V at 60°C for 16 hours.

16S rDNA pyrosequencing

DNA was extracted from fecal samples using the GNOME kit (BIO 101, La Jolla, CA, USA) as previously described (8). The influence of the priming sequences on the microbiota profile was studied and two sets of primers were selected, pV12-B and pV4. Primers were designed as previously proposed (11) and are listed in Supporting Table 4. For each set of primers, two 50 µl PCRs were prepared, containing 1X Expand Long Template buffer 1, 50 μM of each dNTP (Roche Applied Science, Basel, Switzerland), 20 pmole of each primers (Microsynth, Balgach, Switzerland) and 2.5 U Expand Long Template Enzyme mix (Roche Applied Science, Basel, Switzerland). To each reaction a minimum of 2 ng of DNA template was added. PCR amplifications were performed in GeneAmp PCR System 9700 (Applied Biosystems Inc, Foster City, CA, USA). The PCR parameters were 94°C for 5 min, 25 cycles of 94°C for 30 sec, annealing temperature (see Supporting Table 4) for 30 sec and 72°C for 30 sec, followed by 72°C for 7 min. After pooling the two PCRs, 10 l of PCR product were visualized on agarose gel (1.2% in TBE buffer) stained with SYBR Safe (Invitrogen, Eugene, Oregon, USA). Then, PCR products were sent to Beckman Coulter Genomics (Grenoble, France) where equal amount of each were pooled and sequenced by the GS FLX System (Roche). Low quality reads were identified using criteria adapted from Huse et al. 2007: A) Undefined sequence key; these reads could not be unambiguously assigned to any of the PCR samples. B) Lack of recognizable 5’ primer sequence. C) More than one error in 5’ primer sequence. D) Average quality score below 25. E) Presence of ambiguous characters (‘N’). F) Sequence reads not matching the expected length were excluded. First, primer sequences were removed using the vectorstrip software (-besthits 1 –mismatch 30 –besthits parameters). All reads with less than 100 bp were excluded. Subsequently, the median length and the Median Absolute Deviation (MAD) of the remaining reads were calculated separately for each of the 8 different primer pairs. All reads shorter than the median – 5 * MAD were excluded. G) Reads without a BLAST (1) hit to the ARB (13) 16S DNA database (Evalue cut-off: 10-10) were discarded, since these likely do not represent regions of 16S DNA genes. All high-quality reads were classified into Bergey’s taxonomy using the RDP-Classifier (80% confidence cut-off). Sequences that could not be assigned to any known taxonomic group at a certain rank (phylum, class, order, family, or genus) were classified as “Unknown” at that rank. Conversely, the “Unknown” category was defined for each rank. Reads were grouped into Operational Taxonomic Groups (OTUs) based on their best BLAST hit to full-length reference 16S DNAs. First, reads were compared to all type strain 16S DNAs from the RDP database using BLAST. Second, all reads with a perfect match (sequence-identity of 100%) to the same reference 16S DNA were grouped into one OTU. In an analogous manner, all remaining reads were iteratively grouped into OTUs using sequence-identity cut-offs of 98%, 95%, 90%, 80% and 70%. Each generated OTU was labelled by the type-strain of the respective best-matching reference 16S DNA and the sequence-identity cut-off employed.

Reconstruction of phylogenetic trees

8,000 sequences from the variable region pV1-V12 were randomly drawn from each of the following data sets: 16S DNA reads of A) Conv-R mice (D5 and D20) B) reconventionalized mice D5 C) reconventionalized mice D20. The obtained 24,000 reads were compared to all type-strain 16S DNA sequences from the RDP database (5). All best matching reference 16S DNAs were downloaded in aligned format. The alignment of these reference sequences was represented by a profile Hidden Markov Model (pHMM) using hmmbuild from the HMMER package (http://hmmer.wustl.edu/.) (with the –s option). The resulting pHMM was employed to generate a high-quality, multiple-alignment of all 24,000 randomly drawn reads using hmmalign (with –withali option). Hyper-variable regions that do not provide a phylogenetic signal were removed from the resulting alignment using gblocks (17) (with -t=d -b1=12000 -b2=12000 -obesitb3=10 -b4=2 -b5=a options). A simple PERL script was employed to compute the pair-wise distance of the 24,000 randomly drawn reads. The distance was defined as 1 – pair-wise sequence-identity; gaps were not taken into account. A phylogenetic tree was reconstructed from the resulting distance matrix using the neighbour software from the PHYLIP package (7) and colorized in ARB (12). Three different trees were derived by masking different parts of the original tree, i.e. by setting the color of some leaves to the background color (white).