A consortia of clinical E. coli strains with distinct in-vitro adherent/invasive properties establish their own co-colonization niche and shape the intestinal microbiota in inflammation-susceptible mice

Background Inflammatory bowel disease (IBD) patients experience recurrent episodes of intestinal inflammation and often follow an unpredictable disease course. Mucosal colonization with adherent-invasive Escherichia coli (AIEC) are believed to perpetuate intestinal inflammation. However, it remains unclear if the 24-year-old AIEC in-vitro definition fully predicts mucosal colonization in-vivo. To fill this gap, we have developed a novel molecular barcoding approach to distinguish strain variants in the gut and have integrated this approach to explore mucosal colonization of distinct patient-derived E. coli isolates in gnotobiotic mouse models of colitis. Results Germ-free inflammation-susceptible interleukin-10-deficient (Il10−/−) and inflammation-resistant WT mice were colonized with a consortia of AIEC and non-AIEC strains, then given a murine fecal transplant to provide niche competition. E. coli strains isolated from human intestinal tissue were each marked with a unique molecular barcode that permits identification and quantification by barcode-targeted sequencing. 16S rRNA sequencing was used to evaluate the microbiome response to E. coli colonization. Our data reveal that specific AIEC and non-AIEC strains reproducibly colonize the intestinal mucosa of WT and Il10−/− mice. These E. coli expand in Il10−/− mice during inflammation and induce compositional dysbiosis to the microbiome in an inflammation-dependent manner. In turn, specific microbes co-evolve in inflamed mice, potentially diversifying E. coli colonization patterns. We observed no selectivity in E. coli colonization patterns in the fecal contents, indicating minimal selective pressure in this niche from host-microbe and interbacterial interactions. Because select AIEC and non-AIEC strains colonize the mucosa, this suggests the in vitro AIEC definition may not fully predict in vivo colonization potential. Further comparison of seven E. coli genomes pinpointed unique genomic features contained only in highly colonizing strains (two AIEC and two non-AIEC). Those colonization-associated features may convey metabolic advantages (e.g., iron acquisition and carbohydrate consumption) to promote efficient mucosal colonization. Conclusions Our findings establish the in-vivo mucosal colonizer, not necessarily AIEC, as a principal dysbiosis driver through crosstalk with host and associated microbes. Furthermore, we highlight the utility of high-throughput screens to decode the in-vivo colonization dynamics of patient-derived bacteria in murine models.


Introduction
In ammatory bowel disease (IBD) affects over one million people in the United States and costs billions of dollars in direct medical expenses annually [1]. IBD-linked chronic intestinal in ammation de nes both Crohn's disease (CD) and ulcerative colitis (UC), which currently have no medical cure. Additionally, IBD patients with longstanding in ammation experience a greater risk of developing in ammation-associated colorectal cancer (CRC) [2][3][4]. The resident microbes in the intestine are implicated in both IBD and CRC in humans and mouse models. This dysbiotic state of the gut microbial community includes an increased abundance of Escherichia coli (E. coli) and reduced microbial diversity [5][6][7][8][9][10].
Intestinal tissues of IBD and CRC patients often harbor high loads of mucosal adherent/invasive E. coli (AIEC), which are believed to perpetuate intestinal in ammation [10][11][12][13]. AIEC are distinguished by functional attributes tested through exhaustive in vitro co-culture assays, based on the ability to adhere to and invade epithelial cells and survive in macrophages [11]. However, it has been di cult to assess AIEC abundance across patient populations, because AIEC do not contain a universal genetic signature and cannot be screened molecularly from other gut-resident E. coli [14,15]. This limitation has prevented us from a comprehensive understanding of how the AIEC pathobiont takes control of microbiome community composition and interactions with the host, and what molecular features permit colonization of the in amed mucosa. This information is important for us to identify IBD patients colonized with highrisk E. coli strains, but might be hindered by inaccurate in-vitro classi cation. Instead, reclassi cation based on in-vivo phenotype (i.e. mucosal colonization) can potentially bridge this gap.
The mammalian intestine contains at least two distinct bacterial communities: the luminal and mucosally-adherent [16]. These two microbial communities differ in composition and function, and it is the mucosal community that readily interacts with intestinal epithelial and mucosal immune cells. It is well-documented in IBD patients and animal models that disturbance or dysbiosis of the mucosal community strongly impacts health and disease. In fact, a microbiome study on 447 CD patients and 221 unaffected controls revealed disease-associated dysbiosis of the mucosally-adherent, but not the fecal, microbiota in CD patients [17]. Notably, increased mucosally-adherent bacteria and bacterial translocation are observed in IBD patients and mouse models [10][11][12][13]18]. Several lines of evidence underscore the importance of mucosal interactions in microbial-driven disease. Both pks + E. coli and cagA + Helicobacter pylori must have physical contact with mammalian cells to exert their procarcinogenic activities [19,20]. Mucosal bio lms promote colorectal cancer in mouse models [21], and in humans are populated by potentially pro-carcinogenic bacteria including pks + E. coli, enterotoxigenic Bacteroides fragilis and Fusobacterium spp. [4]. We have recently described that siderophore production by pathobiont E. coli NC101 promotes tissue colonization and in ammation-associated brosis [22,23]. As the mucosal niche positions bacteria in an ideal location to exert pro-in ammatory and procarcinogenic activities, it is essential for us to predict which resident microbes are likely to colonize the mucosa of IBD patients.
The gut E. coli characteristics of AIEC (de ned in vitro) and mucosal colonization (observed in vivo) are two determinant factors in IBD. Although both phenotypic traits are prevalent in IBD, how AIEC/non-AIEC pathotypes vary in their ability to colonize mucosa remains to be fully understood. To clarify this, we sought to determine the extent to which the in-vitro AIEC de nition predicts mucosal colonization in-vivo, and whether that speci c colonizing consortia can contribute to a dysbiotic event in gut microbiota. We hypothesize that mucosally-adherent E. coli consortia, de ned as either AIEC and/or non-AIEC strains, utilize common functional capabilities and genomic features to colonize and play a central role in modulating the crosstalk between in amed host and mucosal microbiome during the progression to a dysbiotic state.
Here we describe a polymicrobial colonization strategy that uses a novel barcoding technology to easily distinguish genetically similar, but functionally distinct, E. coli strains in a complex community. We have incorporated this strategy with a collection of well-characterized clinical AIEC and non-AIEC strains isolated from human intestinal tissues and a well-established IBD mouse model to pro le microbiome community dynamics. Our ndings demonstrate that non-AIEC strains can colonize together with AIEC, but with variable patterns in different host contexts. Guided by mucosal colonization status, molecular features unique to in-vivo colonizers were identi ed and regarded as genomic advantages for colonization. We separated the luminal from the mucosal microbiota to highlight E. coli and microbial assembly that occurs in these distinct niches within the colon. High levels of mucosally co-colonized non-AIEC and AIEC strains promoted structural alternations and loss of diversity in the mucosal microbiome within the in amed intestines, which is normally associated with a dysbiotic state. Collectively, diversi ed E. coli colonization patterns mark the evolutionary response of mucosal colonizers to host selection and microbiome interactions by utilizing their conserved metabolic advantages.

Bacterial strains
We utilized seven clinical E. coli strains that were isolated from ileal tissue of Crohn's disease and non-Crohn's disease patients [13,24,25]. Strains were classi ed as AIEC or non-AIEC using standard in-vitro assays to evaluate adhesion/invasion to Caco2 colonic epithelial cells and uptake/survival in J774 macrophages [9,13]. These strains were referenced by their blinded laboratory designation: JA0018, JA0019, JA0022, JA0036, JA0044, JA0048, and JA0091 [25]. The identity of these strains is in Table 1.

Bacterial growth conditions
Bacteria were grown in Luria broth (LB) at 37°C and 250 rpm unless otherwise indicated. Barcoded strains were selectively plated on LB with kanamycin. Serial dilution plating to enumerate fecal and tissue-associated CFU were performed as described [23].

Insertion of molecular barcode
To track individual strains of E. coli from among our complex microbiota, each strain was marked with a kanamycin antibiotic resistance cassette and molecular barcode inserted into a neutral chromosomal region using site-speci c Tn7 transposon insertion [25][26][27]. Barcode designations are listed in Table 1 with a schematic in Fig S1. Animal care Germ-free mice were reared in the National Gnotobiotic Rodent Resource Center at UNC Chapel Hill. Once colonized, mice were maintained in sterile cages in an isolation cubicle under speci c pathogen free (SPF) housing conditions. All animal experiments and procedures were approved by UNC's Institutional Animal Care and Use Committee (IACUC).

Murine FMT preparation
Fecal microbial transplant 1 (FMT1) was prepared anaerobically from 7 C57BL/6 WT speci c pathogen free (SPF) mice that were Helicobacter spp. free. Brie y, colonic and cecal content were removed in an anaerobic chamber and resuspended in sterile, reduced PBS to make a slurry. The slurry was homogenized by vortexing and physical disruption with a lter pipette. Glycerol was added to 15% nal concentration before aliquoting and storing at -80 o C. Fecal microbial transplant 2 (FMT2) was prepared in the same manner from 7 germ-free C57BL/6 WT male mice that were colonized via oral gavage with 100 µl of FMT1.

Murine Model
In these studies, we used the established interleukin-10 de cient (Il10 −/− ) mouse model [3,28], described in detail in Fig. 1. Il10 −/− (in ammation-susceptible) and wild-type (in ammation resistant) 129S6/SvEv mice were reared germ-free to adulthood (8-10 weeks) in the cohorts listed in Fig. 1. Mice were colonized with an even mixture of a total of 10 8 CFU of the barcoded E. coli strains and maintained in an isolation cubicle in SPF housing. One week after colonization, mice were given 100 µl of thawed FMT for colonization competition with a normal, murine microbiota. Two weeks post-FMT, we gave kanamycin water (0.4 g/L) ad libetum for 2 weeks to suppress the microbiota and ensure that some of the barcoded E. coli strains could persist to the end of our model. We had noted in pilot studies that E. coli quickly became undetectable in some WT mice without the addition of kanamycin. In addition, germ-free mice housed in the same isolation cubicle on kanamycin water did not become colonized with any kanamycin resistant aerobes able to grow on LB (i.e. E. coli and similar Enterobacteriaceae). Throughout the experiment, E. coli loads per gram of stool were determined by serial dilution and quantitative culture on LB-kanamycin selection plates. Mice were harvested by CO 2 asphyxiation after a total of 10 weeks colonization post-FMT. Stool samples were removed from the lumen and after ushing contents with sterile PBS, 1 cm of colonic tissue (mucosal sample) was taken for sequencing analysis. Mice that received just FMT were treated the same as above without the one week of colonization with barcoded E. coli strains. Some Il10 −/− mice in internal cohort designation JA123 that received barcoded E. coli with FMT1 required premature sacri ce due to concerns for their care as they appeared hunched and in poor health, as can occur in this colitis model. Early sacri ces happened: mouse 17 -day 5 post FMT due to rectal prolapse, mice 12-14 -day 36 post FMT, mice 5-11 -day 55 post FMT. Histological analysis revealed no abnormalities in the colon. Pro-in ammatory cytokine expression analysis of proximal colon tissue revealed higher Il6 expression in mice sacri ced early ( Fig S2). We repeated this experiment with the internal cohort designation JA218 (7 E. coli + FMT2) and no mice required early sacri ce.

Colon Histology
At necropsy, the majority of the colon not stored for sequencing analysis was swiss rolled with distal colon in the center of the roll, xed in 10% neutral buffered formalin, sectioned and stained with H&E.
Colitis scores (0-4) of rolled colons were blindly assessed by an experienced investigator, as previously described [3,23,28] . DNA Extraction, Libraries Preparation (16S rRNA & E. coli Barcode) and Illumina sequencing DNA was isolated from murine fecal and mucosal samples using a phenol/chloroform extraction with physical disruption, followed by clean-up using the Qiagen DNeasy Blood and Tissue extraction kit as previously described [3,25,29,30]. Ampli cation of the variable V4 region of the 16S rRNA gene (515-806 bp) via a two-step PCR strategy to create libraries for Illumina MiSeq PE-250 sequencing was performed as previously described [31]. Ampli cation of the molecular barcode occurred via the same two-step PCR strategy, but using primers targeting the conserved area anking the unique barcode region in the rst round of ampli cation (F-5'-ATCTCCACTAGTTACCTACAACCTCAAGCT-3', R-5'-CTGCAAACTAGTTACCCATTCTAACCAAGC-3'), with a primer-binding (P2/P4) schematic in Fig S1. Both libraries were sequenced at the UNC High Throughput Sequencing Facility. Raw sequencing reads were processed and analyzed as described in a subsequent section.

Sequencing Analysis and statistical analysis
The 16S rRNA sequencing generated 52 million reads and E. coli barcode sequencing generated 20 million reads. The DADA2 and Quantitative Insights Into Microbial Ecology 2 (QIIME2) pipeline were used for processing 16S rRNA sequencing reads. The forward and reverse reads were denoised using DADA2 'denoise-paired' method, and the chimera were removed using DADA2 'consensus' method. Taxonomy was assigned using QIIME2 feature classi er 'classify-sklearn' based on SILVA databases (release 132). Molecular barcodes were assigned using BLAST with a threshold of 99% similarity. The taxonomic abundance tables were normalized to correct for different sequencing depth across samples using a previously reported method [32]. Bray-Curtis dissimilarity between samples were calculated using normalized abundance of genera and visualized using Principal Coordinate Analysis (PCoA). The differences in microbial composition between groups were analyzed with PERMANOVA test using R function 'adonis' with 999 permutations. Alpha diversity was calculated using Shannon diversity index and the differences in Shannon diversity were analyzed with Wilcoxon test, and visualized with box and whisker plot. Correlations between Escherichia and other genera were calculated with Spearman's correlation. The calculation was based on all the stool and mucosa samples in each of ve speci ed cohorts, and the abundance was normalized to the same depth across samples to limit the effect of sequencing bias. P-values were adjusted with the Benjamini-Hochberg method to correct for multiple hypotheses testing. Biological data (histology scores and cytokine expression) were assessed for signi cance between two groups using the nonparametric Mann-Whitney test. P-values of < 0.05 were considered signi cant.

Pangenomic analyses
To characterize the distribution of the gene content, comparative analysis of the genomes of of seven E. coli isolates was performed through the pangenomics pipeline in Anvi'o v7.1 [33]. The similar pipeline was also applied to characterize the genotypic diversity of gut Akkermansia [34]. Here, the sequenced genomes of E. coli isolates were described in our previous study [25] and retrieved from NCBI under the BioProject accession number PRJNA759208. An Anvi'o-compatible contig database was constructed from each of genome fasta les using the command anvi-gen-contigs-database. This command uses Prodigal [35] for gene prediction and then KEGG KOfam database [36] for function assignment with the script parameter anvi-run-kegg-kofams. To facilitate the downstream comparison on gene sequence, the annotated contig databases were subsequently combined and converted to a single genome database by running anvi-gen-genomes-storage. To compare the genes for presence/absence among genomes (known as "pangenomes"), the command anvi-pan-genome was run based on BLASTP comparisons between all pairs of encoded amino acid sequences from all seven E. coli strains, followed by construction of clusters of homologous genes (designated as "gene cluster") with an MCL in ation factor of 10 for very closely related genomes. The created pangenome was visualized in an interactive interface by running the script anvi-display-pan, where seven isolates are clustered based on gene frequencies and the phylogram is clustered based on presence/absence pattern of gene clusters. Based on the observed colonization patterns of E coli strains in GI tissue in the context of mice model: Il10 −/− +7E.coli + FMT1, each isolate was assigned to the colonized or disappeared group (see details in result) using the script anvi-import-misc-data. Group-speci c KEGG gene functions, represented by the ones present in subset or all members of a given group and absent in the other group, were obtained using the command anvi-compute-functional-enrichment-in-pan. Initial pangenome comparisons were made between two colonization groups on the basis of KEGG pro led functions. The KEGG orthology database could fail in homology assignments for certain gene families due to the lack of archived orthologous templates, while further searching against another database is possible to achieve successful function annotation. Thus, the gene cluster with unknown functions in KEGG were extracted using anvi-split and checked for annotation against NCBI Clusters of Orthologous Groups (COG) [37]. The COG-annotated gene clusters were then used as input to identify the group-speci c COGs. To better display the numbers and shared patterns of group-speci c gene functions (annotated by KEGG or COG), UpSet plots were created using the UpSetR package in R [38]. More broadly comparisons at high-level for group-speci c functions, represented by BRITE hierarchies [39] in KEGG and COG categories, were summarized in the supplemental material.

Result
Novel barcoding approach allows tracking in-vivo colonization of clinical E. coli strains in a murine model Cost-effective techniques that enable us to pro le bacterial consortia at the strain level in host-associated tissues are still lacking. To discriminate closely related bacterial strains from mucosa in a highthroughput manner, we developed an approach that uses a Tn7 transposon system to molecularly tag strains through site-speci cally inserting a unique genomic barcode and kanamycin resistance cassette into a neutral chromosomal region ( Fig S1). The incorporated tag facilitates quick identi cation/enumeration with PCR/qPCR using barcode-speci c primers, or quanti cation with high throughput sequencing using primers for conserved regions outside of the barcode ( Fig S1).
To validate Illumina sequencing can be applied to read and quantify the barcodes, murine AIEC strain NC101 was tagged with 10 different barcodes to obtain 10 uniquely barcoded clones. We pooled clones with various known proportions in-vitro, which were subsequently processed in the same manner as traditional 16S sequencing but treating the barcode as a "variable" region. Percent ratio of 10 barcode reads calculated from sequencing were nearly identical with the corresponding expected ratios in each of ve different mixtures ( Fig S3A). This outcome demonstrates that we have no technique barrier for barcode sequencing in-vitro mock communities. To further test if the barcoding approach impacts initial strain colonization, three germ-free mice were gavaged with evenly-pooled ve barcoded NC101 clones and their stools were harvested after 8 hours (n = 1) or 24 hours (n = 2) based on the estimated in-vivo transit time ( Fig S3B). Barcode-targeted qPCR of three stool samples showed even colonization levels of ve barcoded clones in stools ( Fig S3C); that observation was con rmed by Pielou's evenness index ( Fig   S3D). Consistent evenness, as harbored in inoculum, supports that there is no bottleneck for strains to achieve successful colonization at the initial stage. Once we extended the sampling time to one week after inoculating with the same ve NC101 clones, sequencing the stool samples from 13 mice showed minor shift on an expected even colonization ratio, likely due to in vivo selective pressure over time ( Fig  S3E).
To evaluate the effectiveness of this approach in a physiologically relevant system, we utilized seven clinical E. coli strains (Table 1) in an equal proportion to colonize 17 gnotobiotic mice. A murine fecal transplant was provided after 1 week to provide niche competition (depicted as JA123 in Fig. 1). The clinical E. coli strains were from a well-characterized culture collection isolated from intestinal mucosa of IBD and non-IBD patients [24]. The particular seven strains were chosen from among the B2 phylogroup that includes ybt + and pks + strains, which we have previously shown can impact IBD-associated brosis and CRC [3,23,40]. Stools were collected after 10 weeks to characterize the colonization capabilities of E. coli strains by measuring their presence frequency and relative abundance. Barcode-targeted PCR revealed that three strains of E. coli with barcode A1/C5/D5 can colonize consistently across most of mice ( Fig S4A), in line with their high abundance revealed through qPCR of the molecular barcode ( Fig  S4B). The sequencing obtained a similar result, with the same three strains as the dominant colonizers at the lumen niche ( Fig S4C). We veri ed that barcode insertion did not affect the growth rate by comparing to each E. coli parent strain, ensuring the insertion event had no impact on bacterial tness and proliferation ( Fig S4D). Overall, our high-throughput barcode sequencing permits the accurate assignment of the sequence origin from among related intestinal E. coli strains, and paves the way to evaluate their in-vivo colonization dynamics at the mucosal niche in a murine model. we developed an in-vivo model using in ammation-susceptible Il10 −/− and in ammation-resistant WT germ-free mice gavaged with seven barcoded E. coli isolates premixed at equal ratio. An identical fecal transplant (FMT1) was given to all recipient mice after 1 week to initiate niche competition. Mice were subsequently maintained on kanamycin water for two weeks to ensure measurable amounts of E. coli persisted in WT mice, which historically harbor low E. coli loads (JA123 and JA143 in Fig. 1). Fecal E. coli levels of both cohorts were monitored throughout the 10-week experimental period by serial plating on LB-kanamycin plates to select for transplanted strains. As expected, E. coli bloomed over time to signi cantly higher levels in Il10 −/− mice, which was approximately 2-log greater than in WT mice by the experiment's end ( Fig. 2A). Meanwhile, the V4 region of the 16S rRNA gene were ampli ed from fecal DNA samples at three time points and from the colon mucosa at harvest. Calculation of Escherichia-belonging sequences revealed similar colonization dynamics as serial plating, showing a signi cant increase of Escherichia load in stool and mucosa of Il10 −/− mice as compared to WT mice (Fig. 2B). This elevated E. coli load in an in ammation-susceptible host is consistent with observations on the gut microbiome of mouse models and CD patients [17].
We next sought to characterize the structure of colonic luminal and mucosal microbiomes of Il10 −/− and WT mice after 10-weeks of colonization. Bray-Curtis distances visualized in PCoA plot showed that fecal and mucosal communities in Il10 −/− mice signi cantly differed from those in WT mice ( Fig. 2C-D). Il10 −/− mice also harbored a signi cantly reduced microbial diversity compared to WT mice ( Fig. 2E-F). These results thus far support previous observations that host genotype in uences E. coli abundance, and also the structure and diversity of the fecal and mucosal microbiomes.
We next determined the composition of E. coli consortia in the same DNA samples from above two cohorts using barcode-targeted sequencing. After inoculation with the equally-pooled clinical E. coli consortia, Il10 −/− mice exhibited a loss of diversity relative to WT mice (Fig. 3A), indicating host genotype might also affect colonization patterns of E. coli. We validated that in response to E. coli and FMT colonization, the two cohorts (Il10 −/− and WT) exhibited distinct in ammation states. Histopathology revealed signi cant levels of intestinal in ammation in Il10 −/− mice but not in WT mice, as expected in this IBD model (Fig. 3B). Since we administered kanamycin from weeks 1-3 to improve the engraftment of the barcoded clinical strains, we evaluated the impact on the microbiome after 10 weeks. We compared two cohorts of Il10 −/− mice that were both gavaged the FMT1, with one cohort administered kanamycin and one not (JA134 and JA156 in Fig. 1). Shannon diversity of the stool and mucosal microbiome was not signi cantly different between the two cohorts ( Fig. S5A-B), and colonic in ammation remained the same (Fig. S5C), indicating kanamycin's impact on the microbiome and in ammation is negligible. In summary, the in ammatory micro-environment that develops in Il10 −/− mice promotes selective expansion of E. coli strains with high colonization potential and re-structuring of the intestinal microbiota.
High colonization with E. coli signi cantly alters the microbiome in the in amed intestine The in amed Il10 −/− mice colonized by high loads of E. coli experienced structural alternations in the intestinal microbiota, leading us to examine whether the bacterial community directly responded to the E.
coli expansion. We utilized the Il10 −/− cohort described above that received a pool of barcoded E. coli consortia, FMT1 and kanamycin, and a parallel cohort treated identically but without inoculation of E. coli(JA123 and JA134 in Fig. 1). PCoA analysis based on 16S rRNA gene sequences revealed overlapping but signi cantly different clustering of both the stool and mucosal microbiomes between mice that received E. coli and FMT1 and those only received FMT1 (Fig. 4A-B). To con rm these results, we repeated the experiment with another two cohorts that were treated identically, except given a different fecal microbial transplant (FMT2) (JA218 and JA240 in Fig. 1). Stool and tissue samples were collected and sequenced in the same manner in an independent sequencing run. Consistent with results from FMT1, Il10 −/− mice colonized with E. coli and FMT2 versus FMT2 alone had signi cantly different stool and mucosal microbiomes (Fig. 4C-D). Correspondingly, signi cantly decreased microbial diversity was observed in mice that had added E. coli (Fig. 4E-F). Measurements of colonic tissue cytokine expression demonstrated that Il10 −/− mice colonized with either FMT1 or FMT1 + 7 E. coli exhibited high in ammatory cytokine expression (Fig S6). We also note that both cohorts of Il10 −/− mice colonized with either FMT1 + 7 E. coli or FMT2 + 7 E. coli developed high histologic in ammation scores at 2.38 ± 0.10 and 2.62 ± 0.19 respectively (displayed as mean ± SEM), characteristic of this colitis model. Together our results demonstrate that high loads of E. coli contribute to microbial community changes in the in amed intestine.
Beyond microbiome pro ling, we continued to determine the colonization patterns of the clinical E. coli consortia in the above two cohorts of Il10 −/− mice (JA123 and JA218 in Fig. 1). Barcode sequences were ampli ed from the same stool and mucosal DNA templates to quantify each individual strain. Four E. coli strains became the obviously outcompeting colonizers in each cohort after 10 weeks of colonization.
Speci cally, three strains (CU42ET-1_D5, CU568-3_C5 and CU39ES-1_A1) were maintained in both Il10 −/− cohorts with a different proportion, while HM670_C2 and CU37RT-2_D2 were unique to either of one cohort (Fig. 4G). To better understand how this differential composition developed from evenly-pooled E. coli consortia, we checked temporal dynamics of their surrounding microbes (represented by FMT1 or FMT2 in our mice model). The microbial compositions were moderately comparable in the initial inoculum of FMT1 vs. FMT2, based on similar taxonomic proportions and a potential positive correlation on lineage abundances (Fig. S7A-B). However, the two FMTs appear to have evolved differently over time in vivo, as they progressed into distinct constitutions at the nal colonization stage in mice (JA134 and JA240 in Fig. 1; Fig. 4H). These differences were probably driven by stochastic events such as housing/air or cage effect [41]. Overall, our results suggest that the initial FMT communities experienced divergent ecological succession in Il10 −/− mice, which likely in uenced the colonization dynamics and outcomes of the co-evolved E. coli consortia.
We questioned whether a high abundance of E. coli comprised of 7 strains would elicit similar effects on the microbiome as colonization with fewer distinct E. coli strains. To test this, we repeated our E. coli + FMT colonization of Il10 −/− mice using either 1, 3, or 7 E. coli strains followed by FMT2 (JA218, JA216 and JA226 in Fig. 1). The structure of both the stool and mucosal microbial communities were minimally affected by the initial diversity (1, 3, or 7 E. coli strains) of the colonizing E. coli consortia (Fig. 5A-B). This observation was also supported by pairwise comparisons (Fig. S8). Furthermore, E. coli strain diversity in the initial inoculum did not signi cantly impact microbiome diversity (Fig. 5C). As for the related E. coli colonization pattern, the predominantly colonizing isolate CU568-3_C5 dramatically decreased in mice as the numbers of inoculated strains increased (Fig. 5D), implying intense niche competition occurs among this E. coli consortia over 10 weeks of colonization in the in amed intestines.
Altogether, these results indicate that high colonization with E. coli, regardless of the numbers of distinct strains, exerts strong selective pressure in shaping the structure and biodiversity of the stool and mucosal microbiome in the in ammation-susceptible host. Additionally, microbial interactions among E. coli consortia and/or with other members of the microbiota shape the intestinal E. coli colonization patterns.
Intestinal bacteria associated with E. coli strains under speci c in-vivo context Given that E. coli community assembly was compositionally diverse under the various in-vivo conditions described above (Fig. 3A/4G/5D), we next aimed to identify which taxa bloomed and contracted in response, as these associated members are likely candidates that in uence E. coli colonization patterns.
In in amed Il10 −/− mice, E. coli abundance was negatively correlated with the abundance of 32 genuslevel bacterial lineages and positively correlated with another 2 lineages. However, no genera correlated with E. coli abundance in unin amed WT mice (Fig. 6A), suggesting in ammation is required for these dynamics. Furthermore, such contrary correlation patterns are consistent with the microbiome shift observed between host genotypes and in response to in ammation-induced E. coli expansion ( Fig. 2A-D). These correlations also reveal potential interacting partners that drive the observed E. coli colonization pattern in the in amed intestine (Fig. 3A). To determine if these positive and negative correlations between the E. coli consortia and in ammation-associated microbiome differed by the diversity of the initially colonizing E. coli consortia, we compared Il10 −/− mice colonized with either 1, 3, or 7 E. coli strains followed by FMT2. Here we observed a similar positive correlation between the abundance of E. coli and 7 other taxa among all three cohorts (Fig. 6B). In concordance with PCoA results (Fig. 5A-B), this suggests that the intestinal microbiome responds to mucosally-colonizing E. coli regardless of their strain diversity. Furthermore, with the similarity in microbiomes, these data strongly support that niche competition within E. coli consortia is a strong driver of strain-speci c colonization of the mucosa, as displayed in Fig. 5D. Another important nding in these data is that, except for sharing a signi cantly positive correlation of E. coli abundance with Enterococcus abundance, no correlating taxa were shared between Il10 −/− mice that received FMT1 or FMT2 (Il10 −/− +7E.coli + FMT1 vs. 7E.coli + FMT2 in Fig. 6A
Remarkably, there is a clear tropism for non-AIEC strains to colonize the Il10 −/− mice mucosa: non-AIEC strain CU39ES-1_A1 colonized to high levels in the presence of competing community FMT1, and non-AIEC strain CU37RT-2_D2 colonized to high levels in the presence of competing community FMT2 (Fig. 4G). Thus, in vitro de ned AIEC status cannot explain these mucosal colonization patterns. The unin amed environment of WT mice appeared more hospitable to E. coli colonization, as all seven strains persisted in the mucosal microbiome to some extent (Fig. 3A). Importantly, while the diversity of E. coli strains was higher in the unin amed WT mucosa, E. coli were signi cantly less abundant (2-log fold) than in the in amed Il10 −/− intestines. Together these data suggest that colonization dynamics of individual AIEC and non-AIEC strains are more likely modulated by in ammation states and the composition of the competing microbiome, rather than in vitro-de ned AIEC status.
Common genomic features of clinical E. coli strains that colonize the in amed mucosa E. coli colonization is strongly in uenced by intrinsic metabolic potential that may be conserved across highly colonizing strains. To identify potential genomic features that drive colonization of the in amed intestines, we grouped clinical E. coli strains under two terms: "colonized" and "disappeared." Three E. coli strains (CU39ES-1_A1, CU568-3_C5 and CU42ET-1_D5) had colonized to high levels. Despite lower abundance, strain HM670_C2 widely persisted in colonic tissues and was also included in the "colonized" group ( Fig. 3A and S10A-B). Thus CU39ES-1_A1, CU568-3_C5, CU42ET-1_D5 and HM670_C2 comprise the "colonized" group, while the remaining strains were assigned the "disappeared" group. The CD-associated AIEC strain LF82 was among the "disappeared" group and served as an internal control. It is well known that LF82 poorly colonizes the mucosa of mice because they do not express CEACAM6, the binding partner for E. coli type I pili adhesin FimH [42], thus we would not expect mucosal colonization, as observed ( Fig. 4G/5D).
To identify genes that may promote colonization of the in amed intestines, we performed a pangenome analysis on the seven E. coli isolates. We identi ed a total of 6,482 gene clusters, with 3,585 core gene clusters shared among all genomes and 1,469 gene clusters found only in single genomes (Fig. 7A).
These results indicate that E. coli can adapt to diverse host niches with a large and exible gene pool. We were particularly interested in speci c accessory genes present in most or all members of the "colonized" group that were absent in the "disappeared" counterparts. These genes represent candidate colonization factors to be further explored in future studies. To identify them, we compared the distribution of KEGGannotated gene functions between two de ned groups. This approach highlighted 31 gene functions speci c to "colonized" and 4 gene functions speci c to "disappeared" strains (red and blue bars in Fig. 7B). We next queried common gene functions among in vitro de ned AIEC, regardless of their colonization capacity, and identi ed 8 gene functions (arrows in Fig. 7B). We identi ed only 8 unique gene functions shared among in vitro-de ned "AIEC," underpinning the lack of genomic biomarkers in AIEC. To overcome the annotation limit of single database, the 3,068 gene clusters with unknown KEGG functions were further explored with COG database to facilitate comprehensive comparisons.
Accordingly, we analyzed the pangenome of the same strains but narrowed it down to this unde ned gene pool, from which 1,546 gene clusters were successfully assigned with COG functional categories (Fig. 7C). Consistently, "colonizers" contained much higher numbers of group-speci c COG functions than AIEC strains (red bars vs. arrows in Fig. 7D).
The group-speci c gene functions annotated by either KEGG or COG databases represent potential colonization factors (Fig. 7E and Table S1). Three "colonized" strains (HM670_C2, CU39ES-1_A1, CU42ET-1_D5) putatively encode a distinct subclass of ATP-binding cassette (ABC) transporters that were absent in "disappeared population". The transporter components, including a cytoplasmic membrane-localized ATPase LapB and a membrane fusion protein LapC, are known to participate in the secretion of large adhesin to cell surface to facilitate bio lm development [43]. Another unique gene product in this "colonized" group, DNA helicase IV (HelD) can lead to the formation of a bio lm with lamentous morphology in E. coli [44]. The same three strains appear capable of utilizing energy-rich pectin from diet, due to possession of a series of pectinolytic enzymes (PelB and PelX) that sequentially cleave polygalacturonate to form oligogalacturonides [45]. Oligogalacturonides can be transported in through the TogT transporter [46] and metabolized by oligogalacturonides lyase Ogl [47], all present among three "colonized" genomes. Enzymes necessary for catabolic conversions of sucrose (ScrY/A/R and SacA), were identi ed as an exclusive mark of "colonized" CU42ET-1_D5 and CU568-3_C5 [48]. "Colonized" strains CU42ET-1_D5 and HM670_C2 harbored the pgtABCP operon encoding a phosphoglycerate transporter system, that mediates the transport of glycolysis intermediates across the cytoplasmic membrane [49,50]. These versatile mechanisms for carbohydrate metabolism may provide members of the "colonized" consortia access to a broad range of small carbon molecules for growth and persistence in the intestine.
Iron acquisition genes permitting additional strategies for iron acquisition were a feature of the dominant "colonizers" CU42ET-1_D5 and CU39ES-1_A1. Both genomes harbored a ferric citrate ABC transport system consisting of fecABDELR genes, that assists to uptake ferric dicitrate complex through the cytoplasmic membrane [51][52][53]. They also harbored genes for salmochelin conversion, a glycosylation of the ubiquitous siderophore enterobactin that is catalyzed by glucosyltransferase IroB. This modi cation allows bacteria to circumvent capture by the host immunity protein lipocalin-2 [54] and while characteristic of Salmonella species, has also been found in some intestinal E. coli strains [54][55][56]. Cytoplasmic and periplasmic esterases, encoded by iroD and iroE, allow the following cleavage of ironbound salmochelins, that is required for iron release and salmochelins reutilization [56]. This suggests that they might use multiple ways to compete with host and microbes for scavenging iron and promote growth during infection.
Unexpectedly, a key feature of the "disappeared" group is the Type VI secretion system (T6SS) that is dedicated to the delivery of toxins into neighboring bacteria and host cell [57]. We identi ed two T6SS proteins, ImpL and VasI, assigned through KEGG and then supplemented these ndings with eight COGpredicted T6SS components, demonstrating the advantage of using independent annotation systems to assign previously unknown KEGG functions. Another unexpected contributor to differences in colonization phenotypes may be the catalytic activity of propanediol dehydratase PduC/D/E to generate propionate [58], which is present only in two "disappeared" strains (indicated by triangle in Fig. 7B). The presence of such well-conserved genomic features among colonization groups may represent inherent molecular mechanisms permitting or inhibiting colonization of the mucosa. However, future in vivo colonization studies employing many more E. coli strains coupled with direct investigation of each molecular mechanism (i.e. knockout strains) will be necessary to truly de ne conserved colonization factors.

Discussion
A high throughput in-vivo system to track colonization dynamics of strain-level variants E. coli represent a diverse group of bacteria with pathogenic, commensal, and probiotic activities.
Techniques to distinguishing such closely-related but functionally-divergent strains from within a complex host-associated environment are still lacking. The common strategy of sequencing variable regions of the 16S gene is insu cient to resolve genetically similar strains [59,60], while metagenomics with higher resolution largely fails at the tissue niche from contaminating host material [61]. A compromised method sequencing on the entire rRNA operon achieves desirable classi cation accuracy, whereas much higher error rates of long-read sequencing should not make this as a general metataxonomic analysis procedure [25]. Due to technique limitations, our eld has been unable to e ciently and effectively link bacterial identity to their in-vivo functional and pathogenic potentials at the strain level. We have developed a high-throughput sequencing approach to quantify individual E. coli strains by incorporating a unique barcode sequence into a neutral genomic region, and demonstrated this molecular barcode permits accurate identi cation and discrimination of each clone from in vitro mock mixtures and in vivo IBD mouse models. Similar methods have been used to track the translocation of pathogenic bacteria (uropathogenic E. coli and Yersinia pestis) during infection in mouse models, however, barcode detection and quanti cation was limited by ine cient techniques like qPCR or Southern dot blot [26,27]. Our results demonstrate that barcode-targeted high-throughput sequencing can effectively resolve seven clinical E. coli strains, with results in concordance with barcode-targeted PCR and qPCR. This approach, which can be applied to consortia of tens or hundreds of strains, will permit a rapid, reliable, and thorough means to assess the colonization capacity of individual E. coli strains in order to re-classify and better de ne high-risk IBD-associated E. coli. In combination with in-vivo uorescent imaging, our method would facilitate connecting spatial dynamics to functional activities from both qualitative and quantitative perspectives [26, 62]. One obvious limitation of our genetic engineering approach is common in microbiology: the candidate strains must be culturable with known genomes in order to properly engineer the molecular barcode or uorescence markers.
Human intestinal tissue-derived E. coli can colonize and induce in ammation in the Il10 −/− mouse model For human-derived E. coli strains to stably colonize mice, they must be capable of adapting to different epithelial surface structures in mice. Previous studies have revealed certain human fecal-derived lineages have a poor transfer rate into GF mice [63]. Therefore, we tested the colonization capacity of seven E. coli strains isolated from human intestinal biopsies -thus all strains had the ability to colonize human intestinal mucosa. All strains were retained in the luminal compartment (feces) of at least some animals after 10 weeks (Fig S4A), demonstrating stable colonization. We observed distinct colonization patterns for mucosally-adherent E. coli communities across disease states (unin amed WT and in ammationsusceptible Il10 −/− mice) and in the presence of different competing microbial communities (FMT1 and FMT2). This illustrates human-derived E. coli strains can colonize GF mice, although to varying extents based upon both inherent ability and microenvironment. Inoculating human-derived strains into murine model in the setting of a complex microbial community, rather than in mono-association with murine E. coli strain or fecal slurry that used in previous studies [3,28,40,64], elevates the physiological relevance of our model. Accordingly, the expansion of these E. coli colonizing strains is linked to a signi cant in ammatory response and microbiome shift in Il10 −/− mice, in line with previous results [3,17,[65][66][67][68]. These coincident consequences demonstrate that colonized human E. coli strains can retain the same functional characteristic and utility in murine intestinal environment.
Crosstalk between the host, E.coli, and microbiome drives intestinal dysbiosis in the in amed gut Many reports have described the co-occurrence of intestinal in ammation, E. coli expansion, and reduced microbiome diversity in the intestine of Il10 −/− mouse model [67], yet it has been di cult to break apart their cause-and-effect relationships. Our data and the previous research support an ongoing dialogue between host and microbes, where in ammation fuels expansion of E. coli and microbiome changes, which in turn exacerbate in ammation [64,67]. Importantly, we illustrate here that this communication is occurring at the mucosal barrier, a key site of host:microbe interactions [69]. However, it is still unclear whether microbiome changes are directly driven by host in ammation or in ammation-induced E. coli expansion. We have thus compared microbiomes of two cohorts colonized with or without intestinal E. coli. PCoA analyses illustrated that high colonizing loads of E. coli, but not numbers of distinct introduced strains, shapes the structure of microbiome in an in amed environment, in concert with the observation that many microbes co-shifted with E. coli. Furthermore, in ammation is required for such causal relationships between E. coli and microbiome, as no E. coli correlating genera were identi ed in the unin amed intestine.
More importantly, we further revealed the identities of speci c components underlying E. coli-mediated microbiome change, since compositional disruption can be linked to the intestinal dysfunction. For instance, in response to the expansion of E. coli, a subset protective lineage of Clostridiales (including Ruminococcaceae and Lachnospiraceae) were depleted, which has also been observed in IBD patients [13,68]. As these lineages are involved in fermentation of complex dietary polysaccharides including ber [70], we expect their disappearance will reduce the availability of short-chain fatty acids (SCFAs) necessary for maintenance of gut health [71,72]. Unexpectedly, correlation revealed that E. coli promoted the expansion of probiotic Lactobacillus. It is possible that certain Lactobacillus might bene t from E. coli-derived metabolites [73]. While the above correlations were not shared with another cohort using FMT2 as an inoculum, Enterococcus always co-expanded with E. coli. In concordance with their synergistic correlation pattern, dual-association of Il10 −/− mice with E. faecalis and E. coli has been shown to trigger more aggressive intestinal in ammation than the degree induced by mono-association with each strain alone [28,74]. In turn, in ammation can cause a perturbation in gut anaerobiosis, which facilitates expansion of these two facultative anaerobic lineages [75]. Taken together, these results support E. coli expansion and intestinal in ammation both contribute to a dysbiotic mucosal microbiome and may drive compositional and even functional changes that enhance disease activity [76].
The in-vitro de nition of AIEC may not predict mucosal colonization in vivo Our ability to distinguish strain-level variations from among the complexity of the mucosally-adherent microbiota allowed us to assess AIEC status in relation to colonization potential. Speci c four strains persisting in WT and two strains persisting in Il10 −/− mice supports that the mucosal colonizers are not necessarily AIEC strains. There are likely non-AIEC strains that are more widely distributed than others in healthy and IBD-associated individuals [77,78], and possibly as prevalent as CU39ES-1_A1 shown here. At a population level, although total numbers of non-AIEC strains dropped, CU39ES-1_A1 and CU37RT-2_D2 became remarkably more abundant as in ammation progressed in two respective Il10 −/− cohorts. In line with our observation in mouse models, certain non-AIEC strains can also persist at high levels in the mucosa of IBD patients [10,13]. Indeed, our culture collection contains both AIEC and non-AIEC strains that were isolated from intestinal tissue of IBD patients. This observed dynamic and previous evidence imply that the expansion of non-AIEC colonizers can potentially contribute to in ammation, where non-AIEC cooperate with AIEC in Il10 −/− mice. Conversely, the "pathobiont" rather than "pathogenic" nature of these strains is evident as AIEC and non-AIEC strains failed to induce in ammation in non-susceptible WT mice [79]. The ability of non-AIEC to impair intestinal homeostasis could possibly be linked to putative virulence gene elements that are comparable to AIEC [25,80,81]. In this regard, mucosal colonization characteristics de ned in vivo may better predict pro-in ammatory potential than the current in vitro AIEC de nition. Taken together, our data suggest that the AIEC pathotype de nition cannot truly predict colonization phenotype, while high levels of non-AIEC strains can co-colonize with AIEC and contribute to dysbiosis and in ammation. Our work can be used as a guide to infer candidate colonization factors based on in vivo phenotype rather than AIEC classi cation.
E. coli colonization patterns may be determined by external environmental factors and intrinsic genomic potential In amed Il10 −/− mice harbored higher loads of colonizing E. coli, but with fewer strains retained than WT mice, indicating that in ammation may convey selective pressure for speci c strains to colonize and expand. This is well linked to the previous observation that most AIEC-positive IBD patients carry two or less AIEC strains [82]. We found that two similar inoculum communities (FMT1 vs. FMT2) exhibited distinctive compositions at the nal stage of colonization in the in amed intestine. These FMTassociated microbiomes may have a bifurcating evolutionary trajectory that diversi es colonization dynamics of the co-evolved E. coli consortia. This is supported by different correlation patterns between microbiome and E. coli in the two related Il10 −/− cohorts colonized with either FMT1 or FMT2.
Comparative genomics have been used to identify factors conferring metabolic tness of AIEC strains; several responsible features were recently identi ed [83,84]. Here we identify genetic determinants that may be important for mucosal colonization by E. coli in the in amed intestine, relative to their colonization capacity and AIEC status. Although conducted on only seven strains, our pangenomic analysis revealed that the genomes of "colonizing" E. coli are enriched in genes for carbohydrate catabolism and iron acquisition relative to "disappearing" members. Western diet low in ber confers an expansion of AIEC in the gut [85], and AIEC can reprogram its metabolic preference from carbohydrate to amino acids [86], thus "colonizing" E. coli strains may obtain a growth advantage over their commensal competitors by utilizing varied carbon metabolites through catabolism. Genes encoding iron acquisition via siderophores are overrepresented in AIEC vs. non-AIEC [83, 87], and iron is an important cofactor for metabolic enzymes that promote colonization of the gut. We observed dedicated iron-transporters were harbored by E. coli strains in the "colonized", which may enhance their ability to outcompete neighboring microbes through iron acquisition. Remarkably, these potential colonization-promoting features were nearly all conserved between AIEC and non-AIEC strains, suggesting that a molecular de nition of mucosal colonizers (rather than AIEC) should be further investigated. These studies will require mutating candidate colonization genes across multiple strains to determine their impact on colonization and in ammation.
Contrary to what we expected, we also observed that "colonizing" E. coli strains do not harbor T6SS and propanediol dehydratase that were harbored exclusively by the "disappearing" strains. This would suggest that T6SS and propanediol dehydratase are dispensable for E. coli to colonize the in amed mucosa, but this conclusion differs from many published observations on E. coli persistence in the in amed gut [83, 84, 87-89] and therefore warrants further study with a larger E. coli consortia. In fact, the distribution of T6SS is not restricted to speci c pathotypes, and in other studies has been found in the genomes of multiple AIEC and non-AIEC strains [81, 84, 88]. Thus the impact of E. coli T6SS in colonization of the in amed gut remains to be determined. Also in contrast to previous reports linking the ability to metabolize propanediol with virulence and persistence of AIEC in the gut [83, 87, 89], such crucial features were only speci c to one "disappeared" strain. This discrepancy can potentially be explained by the defect in propanediol transformation being compensated with metabolism of other carbon sources. Also, it is important to point out that due to a lack of host cell adhesion molecule CEACAM6 on mouse epithelial cells [42], LF82_B6 failed to colonize the mucosa of in amed Il10 −/− cohort, thus genotypic features of this strain may not be associated with the "disappeared" phenotype in a CEACAM6 + human host. Taken together, these initial pangenomic studies suggest that metabolic advantages harbored by in vivo colonizers do not necessarily correlate with those identi ed in the AIEC pathotype. We conclude that the E. coli intestinal colonization pattern is contingent on the combined effects of external co-evolved interactome and intrinsic metabolic exibility.

Conclusion
Mucosal colonization is essential for in ammation, a characteristic generally attributed to the in vitro de ned AIEC pathovar. Yet it has been unclear whether in vitro AIEC de nition can well explain colonization phenotype in vivo. Indeed, the lack of a genetic de nition for AIEC could be due to a classi cation problem, where perhaps the in vitro de nition must be re ned by in vivo colonization studies. Our current study has developed and evaluated an innovative high-throughput in vivo approach that can overcome current technological limitations of distinguishing strain-level variations from among the complexity of the mucosally-adherent microbiota, permitting us to probe the AIEC de nition. By using the mouse as a model for human intestinal E. coli colonization, we demonstrated the in vitro AIEC de nition may underestimate the mucosal colonizing dynamic of non-AIEC populations, while colonization phenotype is potentially driven by inherent metabolic plasticity under speci c host contexts. Accordingly, our study highlights the centrality of E. coli mucosa colonizers, consisting of AIEC and non-AIEC, in dysbiosis-associated crosstalk between host and microbiome. Moreover, comparisons of genomes based upon colonization phenotype provide new insight into the genetic determinants of E. coli colonization of the mucosa, which we expect will improve the molecular identi cation of IBD patients harboring high-risk E. coli, who may be at risk for a complicated disease course.  Figure 1 Work ow for novel high-throughput in vivo approach to identify highly colonizing E. coli strains and structural shifts in the mucosally-adherent intestinal microbiome. Germ free mice (in ammationsusceptible Il10 -/and in ammation-resistant WT) were gavaged with various pools of barcoded clinical E. coli strains followed by fecal microbial transplants (FMT) from pooled WT C57Bl/6 mice. We reference each experimental group by their cohort ID and treatment designation (i.e. WT+7E.coli+FMT1). Tissue and stool were harvested from each cohort. DNA was extracted from harvested samples and used as template to PCR amplify the barcode region identifying each E. coli strain and V4 region of 16S rRNA gene for microbiome analysis. Illumina sequencing of barcoded regions revealed highly-colonizing E. coli strains at the mucosally-adherent niche for further pangenomic analysis to pinpoint candidate colonization factors.  In amed (Il10 -/-) vs. un-in amed (WT) states promote divergent patterns of E. coli colonization in the stool and mucosa. Il10 -/and WT mice were the same as Figure. 2. A. Column graph showing relative abundance of 7 E. coli strains in inoculum, distal colon mucosa and stool collected at the harvest, based on analyzing the barcode region. B. In ammation of the distal colon (DC) and proximal colon (PC) in Il10 -/mice was scored histologically (0-4) as compared to WT mice. Line is at mean, and p-values determined by Mann-Whitney test compared to WT (**** p < 0.001). Representative H&E histology at 40X of the colon. Scale bar = 100 µm.

Figure 4
Colonization by the E. coli consortia alters the structure of the stool and mucosal microbiome of the in amed colon. A and B. PCoA based on Bray-Curtis distance for the stool (A) and mucosal (B) microbiome in Il10 -/mice that received FMT1 with (red) or without (blue) E. coli. C and D. PCoA based on The mucosal and stool microbiota respond similarly to inoculation with different numbers (1, 3 or 7) of E. coli strains, whereas the abundance of unique E. coli strains differs by the number of originally inoculated strains. A-C. PCoA based on Bray-Curtis distance for the mucosal (A) and stool (B)

microbiome and
Shannon diversity (C) in Il10 -/mice that received 7, 3, or 1 E. coli isolate(s) followed by FMT2. (A-B) Each symbol represents an individual mouse, and circles represent 95% con dence limits. Box and whisker plots show the median, rst/third quartiles, and min/max index values. There were no signi cant differences between cohorts. D. Column graphs show relative abundance of barcoded E. coli strains colonizing the distal colon mucosa and stool in Il10 -/mice that received 3 or 7 E. coli isolates followed by FMT2. Figure 6