Single clinical isolates from acute uncomplicated urinary tract infections are representative of dominant in situ populations.

Urinary tract infections (UTIs) are one of the most commonly acquired bacterial infections in humans, and uropathogenic Escherichia coli strains are responsible for over 80% of all cases. The standard method for identification of uropathogens in clinical laboratories is cultivation, primarily using solid growth media under aerobic conditions, coupled with morphological and biochemical tests of typically a single isolate colony. However, these methods detect only culturable microorganisms, and characterization is phenotypic in nature. Here, we explored the genotypic identity of communities in acute uncomplicated UTIs from 50 individuals by using culture-independent amplicon pyrosequencing and whole-genome and metagenomic shotgun sequencing. Genus-level characterization of the UTI communities was achieved using the 16S rRNA gene (V8 region). Overall UTI community richness was very low in comparison to other human microbiomes. We strain-typed Escherichia-dominated UTIs using amplicon pyrosequencing of the fimbrial adhesin gene, fimH. There were nine highly abundant fimH types, and each UTI sample was dominated by a single type. Molecular analysis of the corresponding clinical isolates revealed that in the majority of cases the isolate was representative of the dominant taxon in the community at both the genus and the strain level. Shotgun sequencing was performed on a subset of eight E. coli urine UTI and isolate pairs. The majority of UTI microbial metagenomic sequences mapped to isolate genomes, confirming the results obtained using phylogenetic markers. We conclude that for the majority of acute uncomplicated E. coli-mediated UTIs, single cultured isolates are diagnostic of the infection. IMPORTANCE In clinical practice, the diagnosis and treatment of acute uncomplicated urinary tract infection (UTI) are based on analysis of a single bacterial isolate cultured from urine, and it is assumed that this isolate represents the dominant UTI pathogen. However, these methods detect only culturable bacteria, and the existence of multiple pathogens as well as strain diversity within a single infection is not examined. Here, we explored bacteria present in acute uncomplicated UTIs using culture-independent sequence-based methods. Escherichia coli was the most common organism identified, and analysis of E. coli dominant UTI samples and their paired clinical isolates revealed that in the majority of infections the cultured isolate was representative of the dominant taxon at both the genus and the strain level. Our data demonstrate that in most cases single cultured isolates are diagnostic of UTI and are consistent with the notion of bottlenecks that limit strain diversity during UTI pathogenesis.

INTRODUCTION

Urinary tract infections (UTIs) are one of the most common bacterial infectious diseases of humans, responsible for an estimated 9.6 million doctor visits in the United States each year (1). Most are acute uncomplicated infections that occur in healthy individuals with no history of urological disorders. Uropathogens may colonize the urethral opening and infect the lower urinary tract up to the bladder, resulting in urethritis and cystitis, respectively (2). Some UTIs may progress further, resulting in infection of the kidneys (pyelonephritis), and even spread into the bloodstream, leading to a systemic infection known as urosepsis (3).

Over 80% of UTIs have been attributed to Escherichia coli infections, while other, less common uropathogens include a variety of Gram-positive and Gram-negative bacteria such as Staphylococcus, Klebsiella, Serratia, Enterococcus, and Proteus species (4, 5). The standard method used for the identification of uropathogens in most clinical laboratories is microscopy followed by conventional microbiological culturing (6, 7). Culture-independent analysis using sequencing of the 16S rRNA gene has indicated that UTIs may be more polymicrobial than initially believed and has implicated organisms, such as Actinobaculum schaalii and Aerococcus urinae, which are fastidious and may be overlooked by standard techniques (8–10). Furthermore, 16S studies of culture-negative UTI samples and healthy urine samples have demonstrated that bacterial colonization can occur despite the inability to cultivate organisms (8, 11–15).

For culture-positive UTIs, molecular analysis based on sequencing has shown that cultivation generally is able to identify the most dominant organism in infected urine at the species level (8, 10, 16). Numerous studies have compared cultured isolates across patient cross sections, but little is known about the strain-level diversity of uropathogens within individuals. The fimH gene has been used extensively as a phylogenetic marker for the characterization of uropathogenic E. coli (UPEC) isolates at the strain level (17–21). The fimH gene encodes the tip-located type 1 fimbrial FimH adhesin that mediates binding to α-d-mannosylated glycoproteins such as uroplakins on human bladder epithelial cells (ECs) (22–24) and is an essential virulence factor required for adhesion, invasion, intracellular bacterial community formation, and colonization of the bladder (25, 26). The sequence of fimH is highly conserved in virtually all E. coli strains sequenced to date (including nonuropathogens), with minor sequence variations often corresponding to functional differences, most likely the result of adaptation to enhance pathogenesis (19, 27–34). Compared to other established E. coli genotyping methods such as multilocus sequence typing (MLST) (20), typing based on fimH variations is a faster and more practical approach (18). A two-locus-typing scheme combining fimH with fumC (fumarase C) has also been shown to provide strong clonal discrimination power for molecular epidemiological analyses (21).

Here, we profiled the microbial communities associated with 50 uncomplicated UTIs using 16S rRNA and fimH amplicon pyrosequencing. These data were compared to the corresponding markers from the paired clinical strains obtained in the hospital pathology lab to determine how representative single clinical isolates are of underlying infections at both the genus (16S rRNA) and the strain (fimH) level. Eight paired urine UTI samples and E. coli isolates spanning the observed marker gene diversity were then shotgun sequenced to determine if fimH is sufficiently representative of strain-level diversity at the whole-genome level.

RESULTS

Patient demographics and clinical microbiology.

Urine samples were obtained from 50 individuals with acute uncomplicated UTIs (see Table S1 in the supplemental material). The majority of study subjects were female (76%), and patient ages ranged from <1 year to 94 years (Table S1; median, 56 years). E. coli was the most commonly isolated organism (70%) followed by Pseudomonas aeruginosa (10%; Table S1). E. coli infection was associated with age (P < 0.0001, Mann-Whitney U test) and gender (P = 0.03, Fisher’s exact test), occurring most often in younger females. Age was also associated with E. coli type, with phylogroup B2 more common in younger individuals versus D in older individuals (P = 0.01, Mann-Whitney U test). Males had higher white blood cell (WBC) counts on average (P = 0.04, Mann-Whitney U test) than did females, but no other correlations between clinical parameters and gender were observed.

Cultured isolates are representative of UTI microbial communities at the genus level.

For the majority of patients, the most abundant member of the microbial community as determined by 16S pyrosequencing was concordant with the cultured isolate (Fig. 1). Over 80,000 amplicon sequences were obtained for the 50 patients, with a median of 1,359 sequences per sample (normalized to 1,000 for analysis). 16S sequences were analyzed using QIIME and CD-HIT-OTU, and microbial community profiles were compared to cultured isolates. The taxonomic identity of microbial isolates was verified by 16S Sanger sequencing (see Table S1 in the supplemental material). In all cases, the genus corresponding to the isolate(s) was observed in the profiles; however, in eight samples (16%) the cultured genus was not the most dominant. Analysis of biological replicates from these samples indicated that this was not likely due to heterogeneity within urine samples (see Fig. S1). In three of the 50 cases (6%), the most abundant genus in the community profile was anaerobic (either Anaerococcus or Peptoniphilus), while only aerobic organisms were cultured. No epithelial cells were identified microscopically in these samples; however, one of the Anaerococcus-dominated urine samples was noted as containing debris, which could be indicative of contamination during sample collection and a source of additional microbial diversity. Both Anaerococcus and Peptoniphilus species have also been identified in the urinary microbiome of healthy individuals (35). For a further three individuals, Gram-negative Enterobacteriaceae were cultured, while Gram-positive organisms were found to be most dominant by 16S sequencing (50 to 75%). In the two remaining individuals, different genera of Enterobacteriaceae were represented by the cultured isolate and the most abundant constituent in the community profiles.

FIG 1 

Microbial relative abundance and biomass in UTI samples. The heat map of genus (right)- and phylum (left)-level taxonomy for UTI microbial community profiles is based on 16S pyrosequencing with each sample normalized to 1,000 sequence reads. Olive green boxes indicate males, and darker gray shading indicates older individuals. The genus of the cultured isolate for each UTI is indicated by a black border around the appropriate box. Sample numbers are given for those UTIs which also underwent fimH pyrosequencing, and two samples with noted higher taxonomic richness are marked with an asterisk and a plus sign. Relative microbial biomass was estimated by 16S quantitative PCR.

UTI microbial communities were significantly different between age groups (P = 0.013, generalized PERMANOVA) (see Fig. S2A in the supplemental material). Principal component analysis using the generalized Unifrac distance indicated that communities dominated by Enterobacteriaceae, and more specifically E. coli, were more common in younger individuals. Older individuals tended to have communities with higher abundances of Pseudomonas and members of the phylum Firmicutes. Gender was a borderline significant factor influencing community composition (P = 0.054) (see Fig. S2B). UTI samples from females were hallmarked by microbial communities with a high relative abundance of Enterobacteriaceae and/or the presence of lactic acid bacteria and other bacilli. In particular, the genera Streptococcus, Lactobacillus, and Staphylococcus were not detected in any males in the study, even at low relative abundance (Fig. 1). These results are highly concordant with those based on isolate data alone and reflect the high correspondence between cultured isolates and microbial communities as described above.

Overall diversity in UTI microbial communities was very low (see Fig. S3 in the supplemental material). Rarefaction analysis was used to compare operational taxonomic unit (OTU) richness between samples, and overall diversity was assessed using the Shannon index at a depth of 1,000 sequences (see Table S1). The average UTI diversity by the latter metric is 0.51, which is lower than those of human skin (1 to 2.5) (36) and the gut (~6) (37). The Shannon index was inversely correlated with the relative abundance of E. coli (Tau = −0.54, P < 0.0001), and these UTIs had lower overall richness in general. E. coli infections also had a borderline significant association with higher microbial loads as determined by quantitative PCR (qPCR) (P = 0.061, Mann-Whitney U test). No significant differences in community diversity or microbial biomass associated with age or between males and females were observed. The most diverse UTI community, with a Shannon index of 3.84, was observed in an 86-year-old male (Fig. 1; indicated by asterisk). Clinical notes on this individual’s sample indicated that a large amount of debris was observed in the urine. Increased diversity was also observed in a UTI community from a 71-year-old female, which was dominated by E. coli but also contained several members of the phyla Actinobacteria, Bacteroidetes, and Firmicutes (Fig. 1; indicated by plus sign). This sample had an epithelial cell count greater than 50, which is suggestive of contamination by normal microbiota of the anogenital region.

Isolates are congruent with the dominant strain in E. coli UTIs.

Escherichia was the most common genus identified by both culture and community profiling in urine UTI samples. We further characterized Escherichia-positive samples by examining the urine UTI samples and their respective matching clinical isolates (n = 27) using the fimH gene as a strain-level marker. A 523-bp region of the fimH gene was targeted for amplicon pyrosequencing using primers designed to amplify all fimH genes belonging to the genus Escherichia. These primers were used to profile Escherichia strain-level diversity in the UTI samples, and representative sequences for all unique OTUs (clusters with 100% identity) were used to generate a phylogenetic tree with manually identified subtrees (Fig. 2). Two additional urine UTI samples from which other Enterobacteriaceae (Enterobacter aerogenes and Proteus mirabilis) were cultured but which had >1% relative abundances of E. coli in their community profiles were also included in the analysis. fimH sequences were generated from E. coli isolates using Sanger sequencing. Isolates were also classified using the Clermont method, which classifies E. coli into phylogroups based on a multiplex PCR targeting four discriminating loci within E. coli genomes (38). Additionally, the expression of functional type I fimbriae by all clinical isolates was confirmed using a yeast agglutination assay (39)

FIG 2 

Heat map of strain-level taxonomy and fimH phylogeny for UTI microbial community profiles based on fimH pyrosequencing. The fimH type of the cultured isolate for each UTI is indicated by a black border around the appropriate box. Each node of the phylogenetic tree corresponds to a unique 100% OTU representative sequence, and gray shading indicates manually identified subtrees. Amino acid substitutions are indicated by circles for each corresponding nucleotide sequence. The phylogroup is given for samples that had E. coli as the cultured isolate, and darker gray squares indicate higher white blood cell counts. Sample numbers are presented below the bar for white blood cell count.

In all cases where E. coli was cultured, the dominant fimH type in the community matched the cultivated isolate (Fig. 2). A total of 33 distinct nucleotide sequences were observed in the UTI communities. Ten of these represented dominant fimH phylotypes across the 50 UTI samples examined (nucleotide dominant types; ND), all of which corresponded to their paired isolate sequences. The remaining 22 phylotypes were typically found in low abundance (nucleotide secondary types; NS) and were not represented by the cultured isolates. Three of the ND types were congruent with fimH lineages identified in ST131 subclones, including the globally distributed fluoroquinolone-resistant fimH30 subtype (40) (see Fig. S4 in the supplemental material). fimH diversity was extremely low overall, and in 20 of 29 cases (69%), only one fimH type could be detected in UTI samples (Fig. 2; see also Fig. S5). Biological replicates were highly concordant (Fig. S5), and rarefaction analysis showed no increase in fimH OTU richness even at a depth of 70,000 sequences (Fig. 2; see also Fig. S3C), indicating that these results were not due to sampling bias or insufficient sequencing. The diversity of fimH was notably higher in UTIs where E. coli was not the cultured isolate and for samples with epithelial cell counts greater than 50 (see Fig. S2B).

Translation of fimH nucleotide sequences resulted in 10 unique amino acid sequence types, five of which were the dominant type in at least one UTI (Fig. 2; see Table S2 in the supplemental material). Within the amino acid sequence types, five individual substitutions were observed (Fig. 2; see Table S2), all of which have been previously described (19, 27, 32). None of the amino acid types or individual substitutions were significantly associated with microbial community diversity (as measured by the Shannon index) or total microbial load. FimH allele types were labeled according to the presence of amino acid changes in dominant FimH types (AAD1 to -5) or low-abundance FimH subtypes (AAS6 to -10) (see Table S2). FimH amino acid type AAD2 was significantly more abundant in UTIs that had phylogroup D isolates (P = 0.0002), while AAD3 was associated with phylogroup B2 (P = 0.0009). AAD1 had a more widespread distribution and was the dominant FimH type in both of the UTIs not dominated by E. coli. AAD3 correlated significantly with an increased white blood cell (WBC) count (P = 0.005), while AAD1 was associated with a lower WBC count in general (P = 0.02). This reflected an underlying association between the amino acid substitutions N70S and S78N, both of which appear in AAD3 (see Table S2), and a higher WBC count (P = 0.0008 and P = 0.001, respectively). S78N was also correlated with patient age, occurring more often in younger individuals (P = 0.04).

Whole-genome analysis confirms fimH typing of UPEC strains.

Eight E. coli-dominated urine UTI samples and their paired cultured isolates were selected for shotgun metagenomic and genomic sequencing, respectively. Genomic contigs were assembled de novo from isolate sequence data (see Table S3 in the supplemental material), and metagenomic sequences were mapped to their paired isolate genome to further evaluate the strain-level diversity of E. coli and dominance of the cultured isolate in the UTI (see Table S4). Metagenomic sequences were also mapped to the human genome to detect contaminating host genomic DNA. All UTI metagenomes contained sequences corresponding to human genomic DNA with relative abundances ranging from <1% to >80% (Fig. 3; see also Table S4). The two samples (UQU-8 and UQU-57) with extremely low relative abundances of human sequences (<1%) had white blood cell counts of ≤60, compared to >250 for all others, suggesting that WBCs may have been the major source of human genomic DNA. Sequences that did not map to isolate genomic contigs or the human genome were taxonomically classified using MEGAN as described in Materials and Methods.

FIG 3 

Taxonomic classification of metagenomic sequences. Bars show the proportion of metagenomic sequences which mapped to the human genome. Pie charts show the proportions of non-human sequences which mapped to E. coli isolate scaffolds and those which were assigned to non-isolate E. coli genomes and other microbial genomes by MEGAN.

For all UTI metagenomes, the majority of microbial sequences identified corresponded to E. coli and more specifically mapped to the corresponding cultured isolate (Fig. 3). Community profiles for UQU-8, UQU-24, UQU-41, UQU-90, and UQU-98 identified Escherichia species at >90% abundance (Fig. 1), and correspondingly, >95% of microbial sequences in metagenomes for these UTIs corresponded to E. coli. Furthermore, in these five UTIs and all other cases, 80% or more of the E. coli sequences were representative of the isolate genome. In addition to high relative abundances of E. coli, UQU-26, UQU-50, and UQU-57 had significant (>5% of total sequences) populations of other microbes, including Streptococcus spp. (UQU-26 and UQU-57), Staphylococcus spp. (UQU-57), and A. urinae and Actinomycetales spp. (UQU-50). These results are consistent with 16S profiles for these samples.

DISCUSSION

Common clinical practice for acute uncomplicated UTIs is to diagnose and treat the infection on the basis of characterization of a single cultured bacterial isolate, under the assumption that this isolate represents the dominant in situ population (6, 7). Recent evidence suggests that UTIs may be more polymicrobial than previously suspected and also that detectable bacterial populations are present even in healthy urine (11, 14, 15, 41). Here, we used culture-independent sequencing-based methods to investigate microbial communities in urine samples from individuals with uncomplicated UTIs at the genus and strain level to determine if single isolates are adequate to describe the most abundant populations in the disease state.

For the majority of UTIs (80%), cultured isolates were representative of dominant organisms at the genus level. Previous culture-independent analyses of UTIs in hospitalized patients, catheter-associated UTIs, and asymptomatic bacteriuria demonstrated similar concordance between cultivation and molecular analysis for culture-positive specimens (8, 10, 16, 41). Comparable levels of agreement have been reported for other high-biomass bacterial infections, such as those implicated in cystic fibrosis lung disease (42) and chronic wounds (43).

In three of 50 UTIs, anaerobic genera were more abundant than the isolate genus, while in a fourth, fastidious Aerococcus spp. were most dominant. The use of denaturing high-performance liquid chromatography coupled with 16S rRNA gene sequencing in infected urine samples first identified the widespread presence of fastidious bacteria, including obligate anaerobes that were previously overlooked by routine microbiological cultivation (8, 10). While some of these, such as A. urinae (44, 45), have been implicated directly in the etiopathogenesis of UTIs, the role of others, especially anaerobes and Lactobacillus spp., is largely unclear, as they have often been isolated from polymicrobial infections in conjunction with more traditional pathogens such as E. coli (8, 10, 46). Furthermore, community profiling using high-throughput amplicon pyrosequencing has subsequently demonstrated that these organisms are common constituents of both healthy and diseased urine microbiomes (11–15, 41), suggesting that they may not be causally linked to UTI.

The majority of UTIs were associated with the genus Escherichia and more specifically E. coli, which is consistent with previous literature (4, 5). Most of these were largely monomicrobial, with other genera present only in low or negligible (<1%) abundances. Strain typing of E. coli isolates from different UTI samples using fimH gene sequencing indicated a wide diversity of strains in the study population. All E. coli strains were also shown to produce functional type 1 fimbriae. In order to determine if the isolated strains were representative of dominant E. coli populations within individual urine UTI samples, we used amplicon pyrosequencing of the fimH locus, as, in general, the phylogenetic resolution of 16S amplicon pyrosequencing is limited to the genus level (47, 48). While fimH has been used in a number of studies to type UPEC isolates (18, 21, 27, 30, 32, 40), to our knowledge this is the first instance where this marker has been used to profile UTI E. coli strain diversity in situ. Similar assays have also been developed for the rapid analysis of total species and/or strain-level community diversity of methanotrophs (49) and ammonia-oxidizing archaea (50) in environmental samples.

In all urine UTI samples, cultured isolates represented the dominant and, in most cases, the only detectable E. coli strain. Three of the dominant fimH types corresponded to subclones of the globally disseminated multidrug-resistant ST131 clone of E. coli (40). Two of these, fimH30 and fimH41, represent common and ubiquitous fimH lineages, present in E. coli from a range of hosts and geographical locations (40). For a few samples, additional strains were detected but at extremely low abundances. This lies in stark contrast to other infections such as Pseudomonas aeruginosa respiratory infections, which have been shown to be highly heterogeneous and poorly described by single cultured isolates (51–53). P. aeruginosa respiratory disease is most often chronic, occurring in patients with anatomical or physiological abnormalities that render them susceptible to repeated bouts of infection, while acute uncomplicated UTI is more likely to be due to an episode of bacterial invasion of a normally closed system. Other types of UTIs, such as catheter-associated infections, may be more heterogeneous, as they allow for sustained exposure to colonizing microbial populations from external sources (16).

E. coli has an extensive pangenome with a large reservoir of genes undergoing frequent lateral transfer (54–56), and thus, individual marker genes may not be indicative of strain-level diversity. Multilocus sequence typing (MLST) has commonly been used for classification (57); however, it has been shown to lack discriminatory power, grouping genetically, phenotypically, and ecologically distinct strains under the same sequence type (21, 40). Molecular typing with fimH has been proposed as a feasible alternative to MLST for large-scale epidemiological studies of E. coli (18, 21, 40). It has been shown that fimH can be used as a single marker or in conjunction with fumC to successfully discriminate ecologically and clinically relevant sublineages within MLST strain types (21, 40). Comparative analysis of the genomes of two ST131 isolates confirmed the validity of this approach, revealing marked divergence in a subset of shared core genes, including fimH, despite identical MLST loci in the two strains (58). Differences in fimH loci were driven by homologous recombination as well as the acquisition of point mutations, which further distinguished closely related strains with the same ancestral recombinant fimH alleles (58). Here, we performed shotgun sequencing on eight isolate-urine UTI pairs to corroborate the results of fimH amplicon pyrosequencing analysis and determine the validity of fimH as a strain-level marker. Overwhelmingly, the majority of metagenomic sequences of microbial origin from urine UTI samples mapped back to their corresponding isolate genomes, confirming that fimH is indicative of the overall genomic content of UPEC strains.

The ability of fimH to uniquely define UPEC strains may be linked to infection bottlenecks that have been defined in the mouse UTI model (59). In mice, FimH-mediated binding to mannosylated glycoproteins on bladder epithelial cells is critical for UPEC colonization and invasion of the uroepithelium and the initiation of cystitis (24, 60, 61). Enhanced mannose binding due to pathoadaptive mutations under high-shear-force conditions is also beneficial as it allows bacterial cells to resist clearance by micturition (30, 62, 63). Chen et al. demonstrated that fimH also exhibits a second class of mutations that are independent from those that enhance mannose binding and are important for the formation and proliferation of intracellular bacterial communities (IBCs) (27). IBCs in mice infected with a panel of differentially tagged UPEC strains were clonally derived from a single invasive bacterium, despite initial infection with a diverse population (64). The formation of IBCs represents a stringent bottleneck that influences the bacterial population in the lumen of the mouse bladder (64), and UPEC diversity has been shown to decrease as infections progress from the bladder to the kidneys and into the bloodstream (65). Thus, the proliferation of a single dominant fimH strain in acute uncomplicated human UTI may therefore be a consequence of multiple dynamics during infection. Minor subpopulations exhibiting other fimH types could represent either genetic drift in the founder population or the transient flux of newly introduced strains ascending from the intestinal tract or descending from the kidneys (65).

Our results indicate that identification of dominant uropathogens in UTIs via single culture representatives is diagnostic at the genus level in the majority of cases, and also at the strain level for E. coli infections. This implies that the common practice of prescribing antibiotic treatment based on resistance profiles of isolates is an effective treatment strategy for acute uncomplicated UTIs. Newer, more rapid, and cost-effective methods such as flow cytometry (66) and real-time PCR assays (67) that are targeted at identifying highly abundant organisms rather than total microbial communities will similarly be effective for the diagnosis and monitoring of UTIs. However, the question remains whether other microbial populations present in infected urine samples are causally linked to UTI, and further experiments exploring the mechanisms of microbial pathogenicity in the urinary tract are necessary. Additionally, while we have demonstrated the utility of fimH as a molecular marker for strain typing of E. coli and verified the lack of strain-level diversity in E. coli-dominated UTIs, it remains to be determined how representative single clinical isolates are for other common UTI pathogens and other types of infections such as complicated or catheter-associated UTIs.

MATERIALS AND METHODS

Ethical approval.

Ethical approval for this study was obtained from the Royal Brisbane and Women’s Hospital (RBWH) ethics committee (HREC/11/QRBW/107). The need for informed consent was waived by the institutional review board.

Study inclusion criteria.

In order to be included in the study, individuals were required to have no history of urological disorders as well as a urine sample clinically diagnosed with UTI based on cell counts and cultures. Epithelial cell (EC), white blood cell (WBC), and red blood cell (RBC) counts were performed using phase-contrast microscopy with Kova slides (Hycor, CA) at ×10 and ×40 magnifications. Urine cultures were performed on both MacConkey and blood agar plates, with incubation at 35°C for 16 to 18 h under aerobic conditions. A clinical diagnosis of UTI required a WBC count of >107 liter−1, an EC count of <107 liter−1, and bacterial cultures in excess of 106 CFU ⋅ liter−1. Urine samples with a WBC count of >107 liter−1 and an EC count of >107 liter−1 were also considered positive if culture counts were >107 CFU ⋅ liter−1.

Sample collection and initial processing.

Midstream urine samples were collected from patients presenting at the RBWH for microbiological analysis. All urine samples were deidentified and frozen at −20°C prior to DNA isolation for pyrosequencing. DNA was extracted from 1 ml of each urine sample using the Nucleospin tissue kit (Macherey-Nagel, Düren, Germany) according to the manufacturer’s protocol for hard-to-lyse organisms. Extractions were performed in duplicate on a subset of samples for biological replication. DNA was stored at −20°C prior to further processing. Cultured bacterial isolates were also obtained for each urine sample. Patient demographic data (gender and age) were obtained and matched to all deidentified samples.

Phenotypic and molecular analysis of clinical isolates.

Bacterial isolates were identified to the species level using phenotypic assays, including the Vitek2 microbial identification system (bioMérieux). A single colony representing the dominant colony type was selected from the original culture plate and used to inoculate 5 ml of sterile Luria-Bertani (LB) liquid medium. Cultures were incubated overnight at 37°C with shaking (150 rpm). Cells were centrifuged at 4,000 × g for 15 min, resuspended in 20% glycerol liquid LB medium, and stored in 1-ml aliquots at −80° C.

Isolates were identified at the molecular level by Sanger sequencing of the 16S rRNA gene using the primers 926F and 1392wR (see Table S5 in the supplemental material). A 50-µl PCR was run for each sample. Direct lysis multiplex PCR was carried out using a 25-µl reaction volume containing 2.5 µl 10× buffer, 10 mM (each) deoxynucleoside triphosphate (dNTP) mix, 2 mM MgCl2, 15 µg bovine serum albumin (BSA), 12.5 pmol of each primer, 1 unit Taq polymerase (Fisher Biotec, Australia), and 0.5 µl of the culture glycerol stock as the DNA template. PCR was performed using an Applied Biosystems Veriti thermal cycler under the following conditions: cell lysis at 96°C for 10 min; initial denaturation at 95°C for 3 min; 30 cycles of 95°C (30 s), 55°C (45 s), and 72°C (1 min 30 s); and a final extension at 72°C of 10 min. Briefly, 2 µl of exonuclease and Antarctic phosphatase master mix (Affymetrix, Santa Clara, CA) was added to 20 µl of PCR product. Samples were then incubated in an Applied Biosystems Veriti thermal cycler for 15 min at 37°C and 80°C, respectively, to completely inactivate the enzymes. Purified PCR products were submitted for Sanger sequencing using the 926F primer at Macrogen Inc. (Seoul, South Korea). 16S Sanger sequences were manually trimmed and compared to the Greengenes database for taxonomic assignment using BLAST (68, 69).

Phylogroups for E. coli isolates were determined by a modified direct lysis triplex PCR method using primers targeting two genes (chuA and yjaA) and one anonymous DNA fragment (TspE4.C2) (see Table S5 in the supplemental material) as previously described (38). To assay for the presence of functional type 1 fimbriae, E. coli isolates were grown in overnight 5-ml static cultures under aerobic conditions at 37°C. Subsequently, 10 µl of each overnight E. coli culture was mixed with 10 µl of 5% yeast solution, respectively, and agglutination was observed as previously described (39).

Primers targeting the fimH gene were designed based on regions of the gene conserved across different E. coli strains, including an outgroup species, Escherichia albertii, using ARB (70). Using the fimH primers 72F and 563R (see Table S5 in the supplemental material), direct lysis PCR was carried out in a 25-µl reaction volume as described for 16S sequencing. PCR products were purified using the modified Exo-Sap protocol and sequenced from both the forward and reverse primers at Macrogen Inc. (Seoul, South Korea). Each pair of fimH Sanger sequences (forward and reverse) was aligned and assembled, and consensus calls were performed automatically according to base quality, using the Geneious software package. Ambiguous base calls were checked manually using chromatograms. Sequences were trimmed to 320 bp for comparison with fimH amplicon pyrosequences as described below.

Quantitative real-time PCR.

Quantitative real-time PCR was performed on extracted urine DNAs using the 16S primers 803Fa and 1392wR (see Table S5 in the supplemental material) and SYBR-Green MasterMix (Life Technologies, Carlsbad, CA, USA) on the ABI 7900 platform to assess total microbial load in the UTI samples. Real time-PCR (qPCR) was performed in triplicate at two dilutions using the following cycling conditions: 10 min at 95°C and 40 cycles of 15 s at 95°C followed by 1 min at 60°C. A melt curve was produced by running a cycle of 2 min at 95°C and a last cycle of 15 s at 60°C. 16S copy number was determined based on a standard curve constructed from a dilution series of E. coli strain DH10B genomic DNA.

16S community profiling.

Fusion primers containing 454 adaptor sequences and oligonucleotide bar codes were used to amplify the V5, V6, V7, and V8 regions of the 16S rRNA gene. The forward primer was a mixture of four variants (803Fa, 803Fb, 803Fc, and 803Fd) in a ratio of 2:1:1:1 used in conjunction with the reverse primer 1392wR (see Table S5 in the supplemental material) (71). PCR was performed as described for isolate 16S Sanger sequencing using 50-µl reaction mixtures, with a unique bar-coded reverse primer used for each sample. Amplicons were purified with the Agencourt AmPure XP system (Beckman Coulter, Danvers, MA) according to the manufacturer’s instructions. The concentration of purified amplicons was quantified on a Qubit fluorometer (Invitrogen Corp., Carlsbad, CA) with the Qubit double-stranded DNA (dsDNA) high-sensitivity (HS) assay kit. Subsequently, equimolar amounts of bar-coded amplicons were pooled and sequenced on the 454 Genome Sequencer (GS)-FLX Titanium platform at the Australian Centre for Ecogenomics.

Amplicon sequences were checked for chimeras using UCHIME version 4.1 (72) and then quality filtered, trimmed to 300 bp, and assigned to their respective samples using QIIME (73). Sequence clustering at 97% identity was performed using CD-HIT-OTU-454, which detects and adjusts homopolymer errors in amplicon sequences (74). Representative centroid sequences for each cluster were compared to the Greengenes database (February 2011 release) for taxonomy assignment using BLAST (68, 69). Rarefaction analysis was performed using QIIME (73). Sequence libraries were normalized to 1,000 sequences per sample using a repeated subsampling procedure as described in reference 75 prior to further analysis. Genus-level taxonomy for the normalized OTU table was visualized using the heatmap.2 function in the R package gplots (76). The Shannon index was calculated using QIIME, and beta-diversity was assessed using the generalized Unifrac metric (77).

fimH community profiling.

Partial fimH gene sequences from urine samples were amplified using the primers fimH72F and fimH563R (see Table S5 in the supplemental material) modified to include 454 adaptor sequences and a sample-specific oligonucleotide bar code in the forward primer. PCRs and cycling conditions were identical to those used to amplify fimH from clinical isolates. Bar-coded amplicon preparation and sequencing were performed as described for 16S amplicons.

fimH sequences were quality filtered, trimmed to 320 bp (to maximize phylogenetic resolution), and separated by sample using QIIME. Clustering was performed at 100% similarity using CD-HIT-OUT-454, and singleton clusters were discarded. OTU representative sequences were manually checked for uncorrected homopolymer errors in Geneious using the conserved length of fimH. Briefly, reference sequences were aligned with the fimH sequence from E. coli K-12 (GenBank accession number NC_000913.2), and sequences not conforming to the conserved length due to deletions or insertions in homopolymer regions were manually corrected. Following correction, representative sequences were reclustered at 100% similarity using CD-HIT-EST (74). Chimeric sequences that did not align with the reference were also removed. Rarefaction curves were generated using QIIME (70). Sequence libraries were normalized to 1,000 sequences per sample prior to the creation of heat maps and calculation of the Shannon diversity index using QIIME (73). Full-length reference and partial pyrosequence amplicon fimH nucleotide sequences were aligned in ARB with ClustalW using default settings, and the alignment was masked using a 50% consensus filter (70). The fimH ARB database is available upon request. An evolutionary distance tree was constructed in ARB using the Olsen correction and neighbor-joining method. The naming system of the phylogenetic tree was adapted from reference 27, and subtrees were manually identified.

Nucleotide substitutions were described using fimH from E. coli K-12 as the reference sequence. OTU representative sequences were compared to isolate fimH sequences using BLAST and pairwise alignments in Geneious. Representative sequences were translated using TranSeq (78) and aligned in Geneious to the translated E. coli K-12 sequence and to fimH ST131 clone reference sequences (40) using ARB (70). Amino acid substitutions were named according to the convention in reference 21.

Illumina genomic and metagenomic sequencing.

Indexed whole-genome sequencing libraries were prepared with the Nextera DNA sample preparation kit (Illumina, San Diego, CA). Libraries were pooled such that each isolate genome was approximately 1/80 of the pool and the metagenomes were approximately 9/80 of the pool. This pool of libraries was submitted to the Queensland Centre for Medical Genomics (University of Queensland, QLD, Australia), and a single lane of 2 × 100-bp paired-end data was generated on an Illumina HiSeq2000 sequencer (see Table S1 in the supplemental material).

Read pairs were checked for overlap using SeqPrep (https://github.com/jstjohn/SeqPrep) and clipped to remove low-quality bases using Nesoni (http://bioinformatics.net.au/software.nesoni.shtml). De novo assemblies were generated from the isolate reads for each sample using Velvet version 1.2.10 with the help of VelvetOptimiser to determine optimum assembly parameters (79). Isolate genomes were checked for completeness using Phylosift (https://github.com/gjospin/PhyloSift) and Metachecka (https://github.com/Ecogenomics/PhylogeneticM) to identify sequences corresponding to 111 single-copy genes.

For each of the metagenome samples, human contamination was removed by mapping reads to the human genome (hg19) using BWA on default settings (80). Reads that did not map to the human genome were recovered and mapped to their respective de novo isolate contigs using BWA (default settings). Reads that remained unmapped against the isolate were compared to the NCBI database of complete microbial genomes using BLAST. Microbial taxonomy from best BLAST hits was summarized using MEGAN (81).

Statistical analysis.

All statistical analyses were conducted in R (82). Two-way correlations between clinical, microbiological, and molecular covariates were determined using Fisher’s exact test (two discrete covariates), the Mann-Whitney U test (one discrete and one continuous covariate), or Kendall’s Tau (two continuous covariates). The Mann-Whitney U test was also used to compare average Unifrac distances between microbial community biological replicates. Associations between covariates and total microbial communities were determined using generalized PERMANOVA (83) with the unweighted, weighted, and generalized Unifrac distance matrices as inputs (77). Principal component analysis based on generalized Unifrac distances was performed using the function prcomp and visualized with the s.class function in the ADE4 package (84).

Nucleotide sequence accession number.

All sequence data obtained from this study have been deposited in GenBank under BioProject PRJNA174753.

Analysis of biological replicates for a subset of UTI samples (n = 14). Genus-level community profiles (A) were not significantly different between pairs of replicates (P > 0.05 in all cases; Kolmogorov-Smirnov test). Comparisons were made at a normalized depth of 1,000 sequences per sample. Comparison of weighted Unifrac distances (B) indicated that distances were significantly greater between different samples than between replicates of the same sample (P = 0.017; Mann-Whitney U test). Download

Figure S1, PNG file, 0.1 MB

Principal component analysis plots based on generalized Unifrac distances for UTI samples by age (A) and gender (B). Significance testing for differences between indicated groups was conducted using generalized PERMANOVA. Note the predominance of Escherichia in younger individuals. Download

Figure S2, TIF file, 0.3 MB

Rarefaction curves for 16S 97% OTUs (A) and fimH 100% OTUs (B), highlighting the extremely low microbial diversity associated with UTIs in relation to other human-associated microbiomes. Rarefaction curves for four samples (samples 23, 67, 50, and 78) where fimH amplicons were sequenced to 10,000-plus sequences are shown in panel C. Download

Figure S3, PNG file, 0.5 MB

Heat map of fimH sequence types in biological replicates for 5 UTIs with hierarchical clustering based on Euclidean distance. Download

Figure S4, PNG file, 0.1 MB

Nucleotide alignment of fimH amplicon 100% OTU representative sequences. ND represents dominant fimH types, while NS represents secondary types. Positions are numbered according to positions on the E. coli K-12 reference genome. Download

Figure S5, PNG file, 0.4 MB

Patient demographics, clinical microbiology, E. coli types, counts of Pyrotag reads for both 16S and fimH amplicons, Shannon indices for each sample, and amount of raw data generated by high-throughput sequencing.

Table S1, XLSX file, 0 MB.

Amino acid mutations in OTU sequences and amino acid sequence types. D in the type name indicates a dominant type corresponding to a sequenced isolate, while S indicates low-abundance subtypes found only in the FimH community profiles. Positions are based on the E. coli K-12 reference sequence following cleavage of the signal sequence.

Table S2, DOC file, 0.1 MB.

Primer sets used for 16S and fimH amplification.

Table S3, XLSX file, 0.1 MB.

Mapping statistics for metagenomics reads (human sequences removed) and isolate reads versus de novo-assembled isolate contigs.

Table S4, XLSX file, 0.1 MB.

Primer sets used for 16S and fimH amplification. Primers used for pyrosequencing assays have 454 sequencing adaptors, and reverse primers contain oligonucleotide bar codes unique to each sample.

Table S5, DOCX file, 0.1 MB.