Reproducible Colonization of Germ-Free Mice With the Oligo-Mouse-Microbiota in Different Animal Facilities

We established a catalog of the mouse gut metagenome comprising ∼2.6 million nonredundant genes by sequencing DNA from fecal samples of 184 mice. To secure high microbiome diversity, we used mouse strains of diverse genetic backgrounds, from different providers, kept in different housing laboratories and fed either a low-fat or high-fat diet. Similar to the human gut microbiome, >99% of the cataloged genes are bacterial. We identified 541 metagenomic species and defined a core set of 26 metagenomic species found in 95% of the mice. The mouse gut microbiome is functionally similar to its human counterpart, with 95.2% of its Kyoto Encyclopedia of Genes and Genomes (KEGG) orthologous groups in common. However, only 4.0% of the mouse gut microbial genes were shared (95% identity, 90% coverage) with those of the human gut microbiome. This catalog provides a useful reference for future studies.

Background Murine models are a crucial component of gut microbiome research. Unfortunately, a multitude of genetic backgrounds and experimental setups, together with inter-individual variation, complicates cross-study comparisons and a global understanding of the mouse microbiota landscape. Here, we investigate the variability of the healthy mouse microbiota of five common lab mouse strains using 16S rDNA pyrosequencing. Results We find initial evidence for richness-driven, strain-independent murine enterotypes that show a striking resemblance to those in human, and which associate with calprotectin levels, a marker for intestinal inflammation. After enterotype stratification, we find that genetic, caging and inter-individual variation contribute on average 19%, 31.7% and 45.5%, respectively, to the variance in the murine gut microbiota composition. Genetic distance correlates positively to microbiota distance, so that genetically similar strains have more similar microbiota than genetically distant ones. Specific mouse strains are enriched for specific operational taxonomic units and taxonomic groups, while the 'cage effect' can occur across mouse strain boundaries and is mainly driven by Helicobacter infections. Conclusions The detection of enterotypes suggests a common ecological cause, possibly low-grade inflammation that might drive differences among gut microbiota composition in mammals. Furthermore, the observed environmental and genetic effects have important consequences for experimental design in mouse microbiome research.

Introduction The mammalian intestine hosts a microbial community of astonishing density and complexity. This intricate association presumably required significant coevolution of the host and its microbiota. Apparently, this coevolution has been guided by positive selection for factors that result in a state of both mutual tolerance and benefit. Microbial colonization of the intestine takes place right after birth and complexity steadily increases henceforward. The temporal and spatial assembly of the gut microbiota is apparently not guided by specific rules but eventually, after weaning, a stable microbial ecosystem is formed [1]. The adult human intestine hosts 1013 to 1014 bacteria belonging to at least 500 different species or strains [2]. Up to 9 different bacterial phyla are usually found; however, the Firmicutes and Bacteroidetes account for over 90% of all bacteria [3]. Despite its striking conservation on a higher phylogenetic level, the abundance of bacteria on species or strain level varies extensively between non-related individuals. Nevertheless, a core gut microbiome ( = sum of microbial genes) that is shared among different individuals ensures conservation of metabolic functions provided by the microbiota [4]. It is assumed that the microbial ecosystem, once it is formed, efficiently prevents invasion by foreign species. This has been extensively studied in the case of enteric pathogens and is known as ‘colonization resistance’ (CR) [5]. The gut microbiota protects its host against infection by life-threatening pathogens such as Vibrio cholerae, pathogenic Escherichia coli strains, Shigella spp., Clostridium difficile and Salmonella spp. [6],[7]. To date, the molecular bases of CR as well as the key bacterial species involved remain poorly defined. It is clear that if the gut microbiota is absent or disturbed (i.e. germfree status, antibiotic treatment, gut inflammation) the infection risk increases drastically [8],[9],[10],[11],[12]. CR might not only exclude pathogenic bacteria but also acts against harmless or even beneficial bacteria, such as probiotics. For example, the efficiency of probiotic therapy can differ greatly among individuals [13],[14],[15],[16]. To increase effectiveness of probiotic therapy, research aims at improving the half-life of probiotic strains in the gut [17]. In this study we set out to identify characteristics of the bacterial gut microbiota that are linked to infectivity of the human pathogen Salmonella enterica. Conventional mice (CON) harbouring a complex gut microbiota are highly resistant to oral Salmonella enterica infection and concomitant induction of gut inflammation [18]. We tested colonization resistance of mice harbouring different types of gut microbiota. On a quantitative level, we found that mice having a higher gut microbiota complexity exhibited increased protection against Salmonella-induced gut inflammation. In addition we found that the invasion-success of novel species into an established gut ecosystem (i.e. Salmonella enterica, Lactobacillus reuteri RR) may be predetermined by the abundance of species that are closely related to the invader. Materials and Methods Animals We generated LCM mice by colonizing germfree mice with the Altered Schaedler flora (ASF) according to the protocol published on the Taconic webpage. Mice were inoculated at eight weeks of age by intra-gastric and intra-rectal administration of 107–108 c.f.u. of ASF bacteria on consecutive days (www.taconic.com/library). LCM mice (C57Bl/6 background) were maintained under barrier conditions in individually ventilated cages with autoclaved chow and autoclaved, acidified water. No mice with complex gut microbiota were housed in the same room to prevent contamination with natural gut bacteria. CON C57Bl/6 mice were obtained from Janvier (France), Charles River Laboratories (Sulzfeld, Germany), from the Rodent Center HCI (RCHCI Zürich) and the Biologisches Zentrallabor (BZL; Univeristy Hospital Zurich). CON transgene negative B6.129P-CX3CR1tm1Litt /J mice (CX3CR1) [19] and CON Ly5.1 (B6.SJL-Ptprca Pepcb ) were bred at the RCHCI Zürich and CON heterozygous MyD88+/− mice (C57BL/6 background) [20] at RCC Füllinsdorf, respectively. All mice were bred and kept specified pathogen free in individually ventilated cages. This restricts microbial transfers between mice housed in the same room and animal facility. LCM mice, CON mice or streptomycin-pretreated CON mice (20 mg/animal 24h prior to Salmonella infection) were infected by gavage with 5×107 CFU S. Typhimurium SL1344 wildtype or avirulent (sseD::aphT [21]) strains or S. Enteritidis 125109 (streptomycin-resistant variant M1525 [22]). Live bacterial loads in mesenteric lymph nodes (MLN), spleen and cecal content were determined by plating on MacConkey-agar (Oxoid) with respective antibiotics [21]. Lactobacillus reuteri RR (8*106 cfu i.g.) was administered by gavage and cultured anaerobically on MRS media (Biolife; 100 µg/ml rifampicin). To enoumerate bacteria, cecal content was stained with Sytox-green and bacteria were counted in a Neubauer-chamber. Bacterial density is given as Sytox-green positive bacteria per gram cecal content. Ethics statement All animal experiments were approved (license 201/2004 and 201/2007 Kantonales Veterinäramt Zürich) and performed as legally required. Bacteria The streptomycin-resistant wild type strain S. Typhimurium (SL1344 wildtype [23]), the isogenic mutant S. Typhimurium avir (ΔinvG sseD::aphT; kan R [24]) and wild type S. Enteritidis (M1525 [22]) were grown in LB 0.3 M NaCl as described [24]. L. reuteri RR [12] was isolated from our mouse colony selected on MRS media (100 µg/ml rifampicin) (Biolife) and grown anaerobically. Histology HE-stained cecum cryosections were scored as described, evaluating submucosal edema, PMN infiltration, goblet cells and epithelial damage yielding a total severity score of 0-13 points [21]. 0–3 = no to minimal signs of inflammation which are not sign of a disease; this is frequently found in the cecum of conventional mice. 4–8 = moderate inflammation; 9–13 = profound inflammation. Statistical analysis Statistical analysis of Salmonella colonization titers was performed using the exact Mann-Whitney U Test (SPSS Version 14.0). P-values less than 0.05 (2-tailed) were considered statistically significant. Pearson- and Spearman correlation coefficients for bacterial colonization levels were calculated using Graphpad Prism (Version 5.01). Other statistical analyses (Pearson correlation, Kolmogorov-Smirnov test) were performed using the statistical language and environment R (http://www.r-project.org/). To systematically detect differentially abundant OTUs in all mice and for different clustering distances, we used the R software Metastats [25]. Bacterial DNA extraction and 16S rRNA gene specific PCR Total DNA was extracted from cecal contents using a QIAmp DNA stool mini kit (Qiagen). Bacterial lysis was enhanced using 0.1 mm glass beads in buffer ASF and a Tissuelyzer device (5 minutes, 30 Hz; Qiagen). V5-V6 regions of bacterial 16S rRNA were amplified using primers B-V5 (5′ GCCTTGCCAGCCCGCTCAG ATT AGA TAC CCY GGT AGT CC 3′) and A-V6-TAGC (5′ GCCTCCCTCGCGCCATCAG [TAGC] ACGAGCTGACGACARCCATG 3′). The brackets contain one of the 20 different 4-mer tag identifiers [TAGC, TCGA, TCGC, TAGA, TGCA, ATCG, AGCT, AGCG, ATCT, ACGT, GATC, GCTA, GCTC, GATA, GTCA, CAGT, CTGA, CAGA, CTGT, CGTA;]. Cycling condition were as follows: 95°C, 10 min; 22 cycles of (94°C, 30 s; 57°C, 30 s; 72°C, 30 s); 72°C, 8 min; 4°C, ∞; Reaction conditions (50 µl) were as follows: 50 ng template DNA; 50 mM KCl, 10 mM Tris-HCl pH 8.3, 1,5 mM Mg2+, 0,2 mM dNTPs; 40 pmol of each primer, 5U of Taq DNA polymerase (Mastertaq; Eppendorf). PCR products of different reactions were pooled, ethanol-precipitated and fragments of ∼300 bp were purified by gel electrophoresis, excised and recovered using a gel-extraction kit (Machery-Nagel). Amplicon sequencing of the PCR products was performed using a 454 FLX instrument (70×70 Picotitre plate) according to the protocol recommended by the supplier (www.454.com). PCR to detect ASF bacteria in the feces was done as described in [26]. E. coli differentiation Candidate E. coli strains yielding large, red colonies on MacConkey agar were typed using Enterotubes (BD Biosciences). Additionally, in some cases 16S rRNA gene sequencing was performed. The amplification was performed with extracted DNA using ”broad-range” bacterial primers fD1 and rP1 [27]. Reaction conditions were as follows: Deoxyribonucleoside triphosphates (0.25 mM), primers (1 pmol/µl each), 5UTaq-DNA polymerase (Mastertaq; Eppendorf), 50 ng of template DNA. The following cycling parameters were used: 5 min of initial denaturation at 94°C followed by 35 cycles of denaturation (1 min at 94°C), annealing (1 min at 43°C), and elongation (2 min at 72°C), with a final extension at 72°C for 7 min. Amplified PCR products were purified by gel electrophoresis and sequenced using rP1 as sequencing primer. Sequences were assigned to the RDP taxonomy using the RDP classifier (http://rdp.cme.msu.edu/; [28]). Quantification of Lactobacilli Fecal samples were re-suspended in PBS and plated in appropriate dilutions on MRS agar (DE MAN, ROGOSA und SHARPE; Biolife) that supports growth of Lactobacillus spp. as well as Leuconostoc spp. and Pediococcus spp. Plates were incubated for 24 h in an atmosphere of 7% H2, 10% CO2 and 83% N2 at 37°C in anaerobic jars. Reads sorting and quality filtering The amplicon library was sequenced according to the 454 Amplicon Sequencing protocols provided by the manufacturer (Roche 454) at the McMaster University Hamilton (Canada). The sequence determination was made using GS Run Processor in Roche 454 Genome Sequencer FLX Software Package 2.0.00.22. Performance of the sequencing run was gauged using known pieces of DNA introduced in the sequencing run as DNAControl Beads. On average, 94% of reads from DNA Control Beads matched the corresponding known sequences with at least 98% accuracy over the first 200 bases, which was above the typical threshold (80% matches of 98% accuracy over 200 bases). To estimate the reliability of sample separation using our primer-tagging approach, we assessed the number of reads observed to have an illegitimate 4-mer tag (i.e., different from our set of 20 tags). The sequencing plate (including other non-analyzed samples) produced a total of 264,503 reads from which 1,339 contained a wrong tag (0.506%). Given that 256 distinct 4-mer tags are possible and that we used only 20 of these, the majority of sequencing errors in this region are detectable. Correcting for the small fraction of undetectable errors (20/256) and division by four yields a sequencing error rate of 0.137% per single nucleotide - at the position of the tag in the primer (this includes errors during primer synthesis as well as sequencing). Because most errors are actually visible as errors, the rate of unintentional ‘miscall’ of the sample is 0.043%. We applied quality control of 454 reads in order to avoid artificial inflation of ecosystem diversity estimates [29]. Reads containing one of the exact 4 nt tag sequences were filtered with respect to their length (200 nt ≤ length ≤300 nt). Quality filtering was then applied to include only sequences containing the consensus sequence (‘ACGAGCTGACGACA[AG]CCATG’) of the V6 reverse primer and displaying at maximum one ambiguous nt ‘N’. The latter criterion has been reported as a good indicator of sequence quality for a single read [30]. We identified 5,268 reads shorter than 200 nt, 228 reads longer than 300 nt and 2,169 reads containing more than one ‘N’. After filtering, 190,728 reads remained (initial total of 197,949 reads containing the exact primer sequence and tag) and were processed as described below. Definition of OTUs OTUs were defined using the complete filtered dataset, with the exception of exactly identical reads, which were made non-redundant to reduce computational complexity. Before OTU generation, we added reference sequences for subsequent taxonomic classification of OTUs; for this, we used a reference database of selected 16S rRNA gene sequences downloaded from the Greengenes database (http://greengenes.lbl.gov/Download/Sequence_Data/Greengenes_format/greengenes16SrRNAgenes.txt.gz, release 01-28-2009 [31]). In Greengenes, all entries are pre-annotated using several independent taxonomy inferences including the RDP taxonomy. Our reference database was built using full-length non-chimeric sequences with a minimum length of 1100 nt (in order to fully cover the V6 region of all entries). No archaeal sequences were included. The alignment of non-redundant reads from all mice with the reference database was performed using the secondary-structure aware Infernal aligner (http://infernal.janelia.org/, release 1.0, [32]) and based on the 16S rRNA bacterial covariance model of the RDP database (http://rdp.cme.msu.edu/; [28]). Before defining OTUs, we first removed reference sequences for which the alignment was not successful (Infernal bitscore 108 cfu/gram; p<0.05; Fig. 1C). Interestingly, LCM mice also displayed high pathogen titers in the cecum. Owing to this high-level colonization, wild type S. Typhimurium triggered a fulminant inflammatory response in the cecum and colon of both smCON and LCM mice, while no pathological changes could be observed in the CON mice not pretreated with antibiotics (Fig. 1D,E; Fig. S1). This demonstrates that, in contrast to normal complex type of gut microbiota, colonization of mice with a LCM gut microbiota does not confer CR against S. Typhimurium. 10.1371/journal.ppat.1000711.g001 Figure 1 LCM mice susceptible to S. Typhimurium induced colitis. Groups (n = 5) of CON, streptomycin-treated mice (20 mg 24 h before infection) and LCM mice were infected with 5×107 cfu S. Typhimurium wild type by gavage and sacrificed at day 3 postinfection. S. Typhimurium levels in the mLN (A), spleen (B) and cecal content (C). (D) Cecal pathology scored in HE-stained tissue sections (see M&M). (E) HE-stained sections of cecal tissue from indicated mice. Enlarged section (white box) is shown in the lower panel. Scale bar: 100 µm. To verify that mucosal inflammation induced by S. Typhimurium in infected LCM mice is induced by Salmonella-specific virulence factors, we infected LCM mice with an avirulent mutant lacking a functional TTSS-1 and 2 (S. Typhimuriumavir; 5×107 cfu). Despite colonizing the gut to high titers, S. Typhimuriumavir did not cause observable signs of intestinal pathology in LCM mice, demonstrating that gut inflammation in LCM mice was triggered by the same pathogenetic mechanisms as shown for smCON mice (Fig. S2). Colonization resistance is transferred by re-association with a conventional gut microbiota LCM mice, with a low complexity gut microbiota are susceptible to oral S. Typhimurium infection and develop severe acute colitis comparable to germfree or antibiotic-treated mice. Of note, microbiota in the cecum of LCM mice had a similar density as in CON mice (Fig. S3). These findings suggested that their gut microbiota lacks key bacterial species responsible for mediating CR. We reasoned that these protective bacteria would be transferable by co-housing LCM together with CON mice in the same cage. To test this hypothesis, we re-associated 2 groups of LCM mice (n = 2, 4) with one CON donor mouse each for 21 days. As controls, we used groups of non re-associated LCM and CON mice. We infected all animals with S. Typhimurium wild type (5×107 cfu by oral gavage) to measure the degree of CR. Compared to unmanipulated LCM, all re-associated LCM mice had significantly lower S. Typhimurium loads in their feces at 1 day p.i. (Fig. 2A). 4 out of 6 animals were completely protected from Salmonella-colitis and did not show any signs of cecal pathology (Fig. 2E,F) while 2 out of 6 animals developed signs of inflammation (pathoscore 6 and 7) at day 3 p.i., which correlated with higher S. Typhimurium loads in the cecum of these mice (Fig. 2B). Systemic S. Typhimurium colonization appeared also slightly reduced in re-associated LCM mice (Fig. 2C,D). This revealed that CR is transferable and suggested that discrete bacterial species transferred during the 3 week re-association contributed to colonization resistance and protection from colitis. 10.1371/journal.ppat.1000711.g002 Figure 2 LCM gain CR by re-association with normal CON microbiota. Groups (n = 2,4) of LCM mice were re-associated with 1 CON donor each for 21 days in the same cage. Afterwards, non-reassociated LCM (control; n = 5), CON (control; n = 5) and re-associated LCM (n = 6) were infected with 5×107 cfu S. Typhimurium wild type by gavage for 3 days. S. Typhimurium levels in the feces at day 1 post infection (A), cecal content (B), mLN (C), spleen (D). (E) Cecal pathology scored in HE-stained tissue sections (see M&M). (F) HE-stained sections of cecal tissue from indicated mice. Enlarged section (white box) is shown in the lower panel. Scale bar: 100 µm. Arrows point at 2 mice that developed inflammation. Microbiota analysis by high throughput amplicon-pyrosequencing This offered the opportunity to correlate the changes in microbiota composition in the LCM mice with acquisition of colonization resistance. Protection from Salmonella diarrhea is conferred by bacteria entering the gut microbiota of LCM mice. To identify bacteria transferred during re-association, we analyzed gut microbiota composition by high-throughput sequencing of bacterial 16S rRNA genes. We analyzed the fecal microbiota because this non-invasive sampling method allows monitoring the microbiota of a given animal at various time points (i.e. before/after re-association or Salmonella infection). In contrast to other studies [2],[4],[34], we decided to sequence the 16S rRNA hypervariable regions V5 and V6 (length in E. coli: ∼280 bp). Several studies have shown that sequencing of different hypervariable regions or full-length 16S rRNA genes yields to comparable results [30]. Thus we reasoned that V5V6 sequencing would not lead to a major bias in microbiota composition and at the same time would allow us to fully use current pyrosequencing capacity (the average output length of the 454FLX instrument is 250 bp). After read-quality filtering, we obtained 190,728 reads with a length between 200–300 bps in total. Among those, 50,860 were non-redundant. The frequency of chimera, using a simple identification approach was 6.9% of the total reads (13,206) and 14.7% of non-redundant reads (7,499) (Fig. S4). This chimera-frequency is relatively high considering that we probably detected only a fraction of chimeric reads using our method (materials and methods). Sequence reads were aligned with all quality-filtered sequences of our reference database generated from the Greengenes database [31] and operational taxonomic units (OTUs) were defined by hierarchical clustering at various distances, from 0.01 to 0.2. Taxonomy assignment was inferred using annotation from the reference sequences, if possible, or by predictions generated by the RDP classifier from the RDP database [28]. Microbiota complexity differs between LCM, LCMCON21 and CON mice Comparing the average number of OTUs at various distances, clearly the CON donor mice display the highest level of complexity (Tables S1, S2 and S3). We found an average of 767±233 OTUs at a Clustering Distance (CD) of 0.03 and 499±139 OTUs at a CD of 0.05 (before chimera removal: 971±290 OTUs at a distance of 0.03 and 662±186 OTUs at a CD of 0.05). Complexity of the LCM gut microbiota was, as expected, relatively low. By strain-specific PCR [26], we only detected 4 members of the ASF (ASF361, ASF457, ASF500 and ASF519; Fig. S5). However, 29±10 OTUs at a 0.03 CD, and 17±5 OTUs at a 0.05 CD were detected (before chimera removal: 38±10 OTUs at a CD of 0.03 and 23±5 OTUs at a CD of 0.05). This was expected considering the way the LCM mice were generated. LCM status was created by inoculating germfree mice with bacteria of the ASF. Afterwards, LCM mice were kept in individually ventilated cages (IVCs). During this phase, a limited number of additional species might have been acquired. This might explain why our sequence analysis detected more than 8 different phylotypes in unmanipulated LCM mice. Alternatively, the relatively high number of phylotypes could be explained by PCR artifacts or most likely by the intrinsic error rate of pyrosequencing that can lead to a severe over-estimation of microbial diversity using the 16S rRNA marker gene [29]. In LCMCON21 mice, we observed a significant increase in gut microbiota complexity compared to LCM mice. At a 0.03 CD, 295±34 OTUs and at a CD of 0.05, 188±23 OTUs were detected (before chimera removal: 409±60 OTUs at a CD of 0.03 and 279±45 OTUs at a CD of 0.05). However, complexity in LCMCON21 mice remains significantly lower than that in CON mice. We assessed the richness (actual diversity) of our samples by calculating the Shannon index (H) and species evenness (E) as well as the Chao1 diversity estimate (Tables S1, S2 and S3). These calculations revealed that the community was clearly under-sampled; for small CDs (0.01 to 0.05), the Chao1 estimator was, for each mouse, higher than the total number of OTUs. Although under-sampling is limiting our view on the true microbial diversity, it is legitimate to use diversity measures for relative comparisons among samples. Within this context, it is interesting to ask whether, after re-association, LCM mice display similar or different species evenness E compared to the CON mice. Here, species evenness can be regarded as the equilibrium between community members; the less variation is observed between species, the higher is the E value (in other words, evenness is greatest when species are equally abundant). The E-value is defined as the ratio of the theoretically maximal Shannon-index (if all observed phylotypes were equally abundant) divided by the actual Shannon-index. For a 0.05 CD, CON mice displayed an average E-value of 0.76 compared to an average of 0.70 for the LCMCON21 mice (compared to E = 0.15 for LCM mice). Thus, there is no major difference between CON mice and re-associated LCM mice with respect to evenness. Hence the 21 days of co-housing were sufficient in order to adopt a relatively complex and ‘in equilibrium’ microbial gut community. To compare species richness between the 3 different groups, rarefaction curves were created for different CDs (Fig. 3; Fig. S6). For a CD of 0.01, slopes for CON and re-associated CON mice are rather steep, revealing again a considerable under-sampling in our experiment. However, slopes for 0.05 (for re-associated LCM) and 0.1 CD (for CON) seem to reach saturation, suggesting that for this level of analysis, the sampling was sufficiently complete. Therefore we decided to perform OTU analyses using a CD higher or equal to 0.05. This CD is in accordance with a recent report advising a stringent quality-based filtering of 16S- 454 reads and the use of a clustering threshold no greater than 97% [29]. Given the clear under-sampling and the sequencing strategy applied here, a species-level analysis is not conclusive and we decided to focus our further analysis at a higher taxonomic level (from the Family up to the Phylum). 10.1371/journal.ppat.1000711.g003 Figure 3 Collectors' curves of LCM, LCMCON21 and CON mice reveal different complexity. Collectors' curves were created for CD = 0.05 for each mouse from the total number of filtered sequences (A) or from chimera-removed sequences (B). CON mice (green), LCM mice (red) and LCMCON21 mice (blue). Analysis of differences between CON and LCMCON21 gut microbiota We next analyzed qualitative changes in microbiota composition during re-association. In particular, we focused at identifying which OTUs were transferred from the CON donor mice to the LCM recipients within 21 days. Those bacteria may contribute to protection against S. Typhimurium colonization. In order to predict taxonomy for each OTU, we used either the reference sequence taxonomy information present within an OTU-cluster, if any, or the reads taxonomy predicted by the RDP classifier. To test if the taxonomy assignment via reference sequences provided a more resolved taxonomy, we compared taxonomy resolution obtained via reference sequences and via RDP-classifier annotated reads for OTUs which contained both reads and reference sequences (Fig. S7). For different CD and different taxon levels, the reference taxonomy always provided better taxonomic resolution from the phylum level (taxon_1) down to the genus level (taxon_5). Euclidean distances between relative abundance profiles were computed for each mouse and every time-point sampled. Hierarchical clustering (average method) of all mice for taxon_2 (class) taxon_3 (order) and taxon_4 (family) were visualized on distinct heatmaps (Fig. 4; Fig. S8A,B). All CON mice (day 0 and day 21) clustered together as well as all the LCM mice before re-association. Additionally, we included two unmanipulated CON mice (donor 9855 and 9856) that were only sampled at one time-point to provide more samples of independent CON mice from the same mouse colony (n = 4 in total). 10.1371/journal.ppat.1000711.g004 Figure 4 Heatmap showing OTU's distribution in different groups. Fecal microbiota of unmanipulated LCM mice was analyzed at day 0 (n = 8). 6 of these LCM mice (LCM_1 to LCM_6; blue) were conventionalized in two groups with 2 different CON-donors (CON_1 and CON_2; green) and fecal microbiota analyzed at day 21 (LCM_x_d21; grey). OTUs (CD = 0.05) were sorted according to taxon_4 (Family level; x-axis) and average clustering was performed on Euclidean distances calculated between abundance profiles for each mouse and every time-point sampled. Red color indicates high abundance (Log2), yellow color low abundance. CON_9855_d0 and CON_9856_d0 and LCM_9865_d0 and LCM_9866_d0 are 2 additional CON or LCM mice, respectively sampled only at day 0. All samples of LCM mice from day 0 (before re-association) were highly similar and clustered together. The highest identity (determined by BLAST, all against all) between the V5V6 regions of the 8 different ASF members is of 93% (data not shown); therefore it is theoretically possible, for a small clustering distance, to detect each ASF species by our sequencing and taxonomy inference approach. Seven OTUs were systematically detected in the LCM mice, all assigned to the Firmicutes and Bacteroidetes phyla. Thus, we assume that the most abundant species in the feces of LCM-mice are ASF500 (Firmicutes; Clostridia; Clostridiales; Lachnospiraceae; unclassified_Lachnospiraceae) and ASF519 (Bacteroidetes; Bacteroidetes; Bacteroidales; Porphyromonadaceae; Parabacteroides;). Abundance of ASF strains in different mice can be influenced by various factors [26],[40]. Hence, the sampling depth could explain the non-detection of the other ASF members, which were most probably less abundant. Gamma-Proteobacteria as indicators of susceptibility and resistance to Salmonella-infection The qualitative microbiota analysis revealed that within the 21 days of re-association, bacteria from all detected phyla in the CON donor mice were transferred (Fig. 4; Fig. S8A,B). However, the gut microbiota of LCMCON21 was significantly less complex than that of CON mice, suggesting that the microbiota might also differ on a qualitative basis. This might be causally linked to the increased susceptibility to Salmonella infection. Thus, we compared the microbiota of CON and LCMCON21 with respect to lack or enrichment of specific clusters of bacteria (i.e. on order or family level). We analyzed which OTUs were significantly over- or underrepresented comparing LCMCON21 and CON mice. Interestingly, among others, OTUs assigned to the family of the Enterobacteriaceae were enriched in LCMCON21 mice, as compared to CON mice (Fig. 4; Dataset S1). Since Salmonella Typhimurium is also a member of the Enterobacteriaceae, the enrichment of such close relatives in LCMCON21 mice might be an indicator of favorable growth conditions for this type of bacteria. This finding prompted us to investigate, whether there is a positive correlation between the abundance of Enterobacteriaceae (i.e. E. coli) and the susceptibility to Salmonella infection. We have previously observed that C57Bl/6 mice obtained from different sources (commercial breeders, other laboratories) exhibit differential degrees of CR against Salmonella. To analyze whether CR is linked to different E. coli titres, we defined fecal E. coli levels of mice from five different breedings (C57Bl6 background from our animal facility and others) before infecting them with S. Enteritidis wild type by oral gavage (5×107 cfu; no antibiotic-treatment). We used S. Enteritidis because pilot experiments in our laboratory had shown that this serovar generally leads to a higher disease incidence (colitis at day 4 after oral infection) in non-antibiotic-treated mice, than S. Typhimurium. E. coli is readily differentiated from other Enterobacteriaceae by colony color and morphology on MacConkey agar (see Materials and Methods for typing details). One day after infection, we determined fecal S. Enteritidis titers by plating. The mice were sacrificed at day 4 postinfection and we analyzed S. Enteritidis titers at systemic sites, in the intestine as well as cecal pathology (Fig. 5A; Fig. S9). Indeed, we observed a positive linear correlation between fecal E. coli levels before infection, S. Enteritidis colonization efficiency (r2 = 0.434: Spearman p = 0.0015). If S. Enteritidis titres were above 1.5×105 cfu/g feces at day 1 p.i., mice developed colitis at day 4 p.i. This suggests that E. coli titres may predict whether mice are susceptible to Salmonella induced gut inflammation. 10.1371/journal.ppat.1000711.g005 Figure 5 Infection experiments in conventional mice reveal correlation of bacterial infectivity with the prevalence of related species. (A) Groups normal unmanipulated CON mice (6-12 weeks; symbols indicate different sources) were infected with 5×107 cfu S. Enteritidis wild type by gavage. Fecal E. coli titres before infection were determined (x-axis; Log10 cfu/g). 1 day post infection, S. Enteritidis titres in the feces were determined (y-axis; Log10 cfu/g). Spearman and linear correlation were calculated (p = 0.0015; p<0.0001). The degree of gut inflammation was determined in the infected mice. Half-filled symbols indicate mice with inflammation score ≥4. (B) Groups normal unmanipulated CON mice (6-12 weeks; symbols indicate different sources) were infected with 5×107 Lactobacillus reuteri RR (rifampicin-resistant) by gavage. Fecal levels of Lactobacilli were determined on MRS agar and plotted against fecal Lactobacillus reuteri RR titers at day 1 (left) and 5 (right) postinfection. Higher levels of Lactobacilli predict higher intestinal colonization with a commensal L. reuteri RR after oral inoculation We observed that higher E. coli levels positively correlate with increased Salmonella infectivity. This might be due to the close relatedness of these two species as they might have similar environmental requirements. Thus, we hypothesized that the same principle might apply for other intestinal bacteria. We tested this hypothesis using Lactobacillus reuteri RR, a rifampicin-resistant isolate from our mouse colony that can be specifically detected by culture [12]. We determined whether higher titres of intestinal Lactobacilli would correlate with increased gut colonization by Lactobacillus reuteri RR upon oral gavage. Lactobacilli are Gram-positive, of low G+C content, non-spore-forming, aerotolerant anaerobes and can be differentiated on selective media (i.e. MRS-agar). We determined fecal levels of Lactobacilli of mice from different sources and subsequently infected them with Lactobacillus reuteri RR (107 cfu by oral gavage). 1 and 5 days post infection we determined Lactobacillus reuteri RR titres in the feces. Indeed, we found significantly enhanced colonization of Lactobacillus reuteri RR in mice with higher titres of Lactobacilli (Fig. 5B). This suggests that, like in the case of E. coli and Salmonella, higher levels of Lactobacilli correlate with increased colonization efficiency by a commensal Lactobacillus strain. Closely related phylotypes generally display significantly correlated abundances in the intestine In order to investigate whether our observations with Enterobacteriaceae and Lactobacillaceae correspond to a more universal phenomenon that applies to closely related bacterial groups in general, we performed a systematic abundance correlation analysis between OTUs detected in 9 distinct CON mice (Fig. 6; Fig. S10). We limited our analysis to OTUs detected in at least 6 mice in order to lower the under-sampling bias in our 454 sequence data. Upon examination of OTUs defined at various CDs, we found that closely related phylotypes (i.e. 0