Adaptive Evolution of Class 5 Fimbrial Genes in Enterotoxigenic Escherichia coli and Its Functional Consequences*

Background: Class 5 fimbriae of ETEC represent important colonization factors that mediate small intestinal adhesion. Results: ETEC recently acquired these fimbriae, where fimbrial genes accumulated mutations under positive selection. Conclusion: Mutations in both adhesive and non-adhesive subunits altered function, probably affecting bacteria-host interactions. Significance: Class 5 adhesin mutations suggest pathoadaptive evolution in ETEC, with insights in structural and functional properties of these virulence factors. Class 5 fimbriae of enterotoxigenic Escherichia coli (ETEC) comprise eight serologically discrete colonization factors that mediate small intestinal adhesion. Their differentiation has been attributed to the pressure imposed by host adaptive immunity. We sequenced the major pilin and minor adhesin subunit genes of a geographically diverse population of ETEC elaborating CFA/I (n = 31), CS17 (n = 20), and CS2 (n = 18) and elucidated the functional effect of microevolutionary processes. Between the fimbrial types, the pairwise nucleotide diversity for the pilin or adhesin genes ranged from 35–43%. Within each fimbrial type, there were 17 non-synonymous and 1 synonymous point mutations among all pilin or adhesin gene copies, implying that each fimbrial type was acquired by ETEC strains very recently, consistent with a recent origin of this E. coli pathotype. The 17 non-synonymous allelic differences occurred in the CFA/I pilin gene cfaB (two changes) and adhesin gene cfaE (three changes), and CS17 adhesin gene csbD (12 changes). All but one amino acid change in the adhesins clustered around the predicted ligand-binding pocket. Functionally, these changes conferred an increase in cell adhesion in a flow chamber assay. In contrast, the two mutations in the non-adhesive CfaB subunit localized to the intersubunit interface and significantly reduced fimbrial adhesion in this assay. In conclusion, naturally occurring mutations in the ETEC adhesive and non-adhesive subunits altered function, were acquired under positive selection, and are predicted to impact bacteria-host interactions.

CFA/I fimbriae have been subjected to detailed structurefunction studies. Crystal structures reveal that its CfaB pilin folds into a single domain and CfaE adhesin into two domains. Each domain folds into an Ig-like ␤-sandwich (11,13). In fimbrial assembly, subunits are interconnected by donor-strand exchange and complementation, a mechanism first described for E. coli type 1 fimbriae (13)(14)(15)). An unidentified sialylated glycoprotein on the small intestinal surface purportedly serves as the natural substrate for CfaE adherence (16 -18). In an adaptation of a classic erythrocyte adherence model for class 5 ETEC fimbriae (13,15,16), binding of CFA/I and more specifically CfaE was enhanced by shear stress, reminiscent of the catch bond properties of FimH, the mannose-specific E. coli type 1 fimbrial adhesin (19).
One approach to elucidating structure-function relationships of bacterial adhesins as it relates to pathogenesis is to examine the functional impact of naturally occurring mutations that are acquired under positive selection (20 -22). Such pathoadaptive mutations represent an alternative mechanism of virulence evolution, complementing horizontal gene transfer. Here, we investigated the natural population diversity of the adhesin and pilin subunits of representative fimbriae of subclasses 5a (CFA/I), 5b (CS17), and 5c (CS2). Our findings underscore common themes in microevolution of fimbrial adhesins from different pathotypes of E. coli.
Serotyping-All strains for which pre-existing E. coli serotype data were not available were serotyped using traditional methods.
Enterotoxin Genotyping-All ETEC strains included in this study have been identified previously by their expression of heat-labile enterotoxin (LT) and/or heat-stable enterotoxin I (STI). Further discrimination of STI genotype into estA1 (STIa or STp) and estA2 (STIb or STh) was performed by PCR analysis of plasmid preparations using the primer pairs and conditions described in supplemental Table S2.
Sequence Analysis of Fimbrial Subunit Genes-DNA sequences of the major and minor subunit genes for CFA/I (cfaB, cfaE) were determined by direct sequence analysis of wild type plasmid preparations using the primers listed in supplemental Table S3. Sequence analysis of the same genes for CS17 (csbA, csbD) and CS2 fimbriae (cotA, cotD) was performed on amplicons generated by PCR using forward and reverse primers that annealed 5Ј and 3Ј to each open reading frame (supplemental Table S3). Using the same primers, and additional internal primers for the adhesin genes csbD and cotD, both strands of each amplicon were sequenced using an ABI Prism BigDye Terminator cycle sequencing kit. Products were purified and sequence was analyzed on an ABI Prism 3130xl genetic analyzer (Applied Biosystems, Foster City, CA). Sequences were assembled using Sequencher software (version 4.10) (Gene Codes Corp., Ann Arbor, MI).
Multilocus Sequence Typing-Multilocus sequence type (MLST) analysis was performed using four housekeeping genes, fumC and gyrB from the E. coli MLST database maintained at the University College Cork and uidA and fadD from the EcMLST database at Michigan State University (23,24). Internal fragments of these four genes were amplified and sequenced from all isolates using the primers and methods shown in supplemental Table S4.
Microevolutionary Analysis-Phylogenetic trees were generated as maximum likelihood phylograms using PAUP* (version 4.0b) (25) based on the general time reversible substitution model with codon position-specific estimated base frequencies. Sequence diversity was measured by the average pairwise diversity index () and the rates of nonsynonymous (dN) and synonymous (dS) mutations (26) using MEGA (version 4) (27). Analysis of statistical significance was performed using the z-test for and dN/dS values (28). The presence of structural hotspot mutations was determined using zonal phylogeny software (29).
Homology Modeling-Protein structure prediction software Phyre (32) was implemented to generate a three-dimensional atomic model of CsbD based on the available CfaE crystal structure (13). The resolved structures of CfaE and FimH and the predicted CsbD structure were viewed using PyMOL to calculate distances with reference to the C␣ atom of each mutation that increases mannose (FimH) or erythrocyte binding (CsbD). Using a cut-off of 4 Å, distances of each mutated position from the nearest amino acid in the binding pocket and in the interdomain interface were measured. Then each of the two distances was normalized by dividing by the distance between the corresponding binding pocket and interdomain interface residues, to eliminate any bias of length differences among CfaE, CsbD, and FimH proteins in the analysis.
Flow Chamber Experiments-Polystyrene tissue culture plates were coated with proteins in 0.02 M NaHCO 3 buffer for 1 h at 37°C. Protein concentrations for purified fimbriae were 100 g/ml (33). To avoid nonspecific binding of bovine erythrocytes to BSA, plates were quenched with 0.1 mg/ml of nonbinding FimH mutant type 1 fimbriae preparation in PBS. A parallel plate flow chamber (PPFC) (GlycoTech, Gaithersburg, MD) was assembled over the plates according to the manufacturer's instructions. The assembly was mounted on a Nikon TE200 inverted microscope with a 10ϫ phase-contrast objective. Experiments were recorded using a Roper Scientific high resolution CCD camera and MetaMorph or MetaView (Universal Imaging Corporation TM ) video acquisition software. All additional experiments were performed in 0.2% BSA-PBS buffer with 0.5% ␣-methyl D-mannopyranoside. Bovine erythrocytes were washed at least twice and were resuspended at 0.04% before use.
Accumulation Experiments-Erythrocyte suspensions were manually loaded into assembled PPFC and allowed to settle for 4 min. The recording was started, and buffer was driven into the chamber using a Harvard syringe pump at varied flow rates to produce the desired wall shear rates. Erythrocyte movement was recorded at two frames per second with 100-ms shutter time to distinguish cells that were moving at speeds slower than hydrodynamic velocity. The recording time varied depending on the period of delay for each shear rate. Recorded movies were analyzed in three steps. First, settled erythrocytes were counted before establishing flow. Second, after in-chamber flow was initiated, erythrocytes that were in focus (i.e. moving slower than hydrodynamic velocity) were counted in several sequential frames. Finally, these counts were used to calculate the average percent of erythrocytes that stayed near the surface (i.e. bound erythrocytes). Each experiment was repeated at least twice using different protein and erythrocyte batches.
Erythrocyte Velocities-Velocities of moving erythrocytes were calculated either from the accumulation movies as the distance covered by the erythrocyte within a given time period, or from the recording in stream video mode (1 frame per 37 ms) using the tracking object mode of Metamorph software. The velocity of every cell in the field of view was determined, and the average velocity with standard errors was calculated from the velocities of ϳ100 cells within the same video.

Diverse Clonal Association of Class 5 Fimbrial Genes-MLST
analysis was performed on the 69 ETEC strains and also applied to 21 reference E. coli and Shigella strains using available genome sequences ( Fig. 1 and supplemental Table S5). These ETEC strains partitioned into 16 sequence types (assigned as ETEC1 to ETEC16), among which the pairwise nucleotide diversity averaged 0.5 Ϯ 0.07%. This compared with a 3-fold higher diversity among the reference strains (1.4 Ϯ 0.17%). Some clades were composed of distinct ETEC STs, whereas others contained a mixture of ETEC, other diarrheagenic E. coli pathotypes and non-pathogenic (fecal) isolates (Fig. 1). ETEC strains were confined to the previously described major phylogenetic groups A, B1, AϫB1. None fell within the clonally distinct B2 and D groups, where most extra-intestinal pathogenic E. coli are found.
All but one of the CS2-ETEC strains were clonally related and shared the O16 serogroup. The one unrelated CS2-ETEC, M145C2, had a different serogroup (O128ab). In contrast, ETEC strains carrying CFA/I or CS17 genes were clonally diverse. Some strains expressing the same fimbrial type belonged to separate clades, whereas some strains individually expressing CFA/I and CS17 belonged to closely related or even the same sequence type (ST) (Fig. 1), suggesting frequent horizontal movement of the different types of class 5 fimbrial genes between ETEC strains.
Fimbrial Genes under Positive Selection-The structural genes of the three fimbrial types were similar in length (Table  1), but their sequences were divergent, with nucleotide diversity ranging from 38 -43% (188 -213 pairwise nucleotide changes) and 35-39% (361-401 pairwise nucleotide changes) for the pilin and adhesin genes, respectively (Fig. 2). The average rates of non-synonymous (dN) and synonymous (dS) mutations between primary alleles of the pilin (dN ϭ 0.42 Ϯ 0.03, dS ϭ 1.87 Ϯ 0.23) and adhesin genes (dN ϭ 0.54 Ϯ 0.03, dS ϭ 0.85 Ϯ 0.03) were considerably higher than the corresponding values for the housekeeping genes (dN ϭ 0.001 Ϯ 0.0004, dS ϭ 0.03 Ϯ 0.0042). For the housekeeping genes, dN was significantly lower than dS (dN/dS ϭ 0.03, p Ͻ 0.0001), indicating strong purifying selection against structural changes. In contrast, for the class 5 fimbrial genes, the ratio of dN/dS for the pilin (dN/dS ϭ 0.22) and adhesin genes (dN/dS ϭ 0.64) was greater. In several regions across the pilin and adhesin genes, this ratio exceeded 1, likely due to positive selection for structural changes.
In contrast to the high diversity of pilin and adhesin genes between discrete fimbrial types, within a single fimbrial type there was nominal or no sequence variability (Fig. 2, A and B,  solid line branches). For CS2, all cotA pilin and cotD adhesin gene sequences were identical. For CFA/I, there were three separate alleles of cfaB (pilin) with an average pairwise change of 1.33 Ϯ 0.87 nucleotides (0.3% diversity). There were four alleles of cfaE (adhesin) with an average pairwise change of 1.50 Ϯ 0.88 nucleotides (1.4% diversity). For CS17, csbA (pilin) was invariant, whereas csbD (adhesin) divided into eight alleles with an average pairwise change of 4.64 Ϯ 1.35 nucleotides (0.4% diversity). Strikingly, within fimbrial types, all but one of a combined 18 single nucleotide polymorphisms were non-synonymous. For individual genes, the low number of alleles precluded analysis of differences in dN over dS. However, for concatenated loci dN was significantly above dS (p Ͻ 0.0001), the dN/dS ratio of 8.4 indicating strong positive selection for amino acid replacements. Additionally, multiple independent mutations conferring different amino acid changes were localized to codons 62 and 85 of csbD. Occurrence of these structural hotspot mutations provides further evidence of positive selection for structural changes (34,35).
Spatial Distribution of Naturally Occurring Amino Acid Mutations-Crystal structures show that the CfaB pilin folds into a single domain (Fig. 3A), whereas CfaE features two domains, a lectin or adhesin domain with a putative receptorbinding site and a pilin domain anchoring the tip adhesin to the fimbrial shaft (Fig. 3B) (11,13). The two observed amino acid changes in CfaB, A37V and T136S, were spatially localized to the interface between CfaB and the pilin domain of CfaE (Fig.  3A). The three residue changes noted in CfaE, T91I, A128V, and Q142R, were all localized to the putative lectin domain of the adhesin, clustered on the upper loops that are predicted to form the ligand-binding pocket (Fig. 3B).
In CsbD (CS17 fimbrial adhesin), the 12 non-synonymous nucleotide mutations resulted in a total of 10 amino acid changes (two codons were affected by two separate mutations each). Seven changes accumulated within a stretch of 24 amino acids (N62S/S62R, S74T, T84N, and I85V/I85L/I85R). Two other changes (A144H and Y145N) targeted adjacent positions, indicating significant clustering of mutations (p Ͻ 0.0001). One CsbD change (Y293H) occurred in the pilin domain close to the C terminus. Phyre predicted that the best structural homologue of CsbD adhesin is the CfaE adhesin (with 53% protein sequence identity, E-value 3.4e-38 and 100% estimated precision). Hence, we used the CfaE crystal structure to compute a model of CsbD (Fig. 3C). Projected onto this model, all but one (i.e. Y293H) of the CsbD structural changes mapped to the upper loops of the lectin domain, surrounding the putative ligand-binding pocket.
Thus, the majority of changes in both the CfaE and CsbD adhesins spatially clustered around their putative binding sites. This suggested the potential functional significance of these point amino acid mutations.
Fimbrial Microevolution and Functional Consequences-Sufficient polymorphisms were detected in cfaE and csbD (but not cfaB) to examine them for evidence of adaptive evolution. A primary (or progenitor) allele of each adhesin was identified, to which a majority of the adhesin sequences matched (Fig. 4). For CfaE, three variants were independently derived from the progenitor (Fig. 4A). The primary CsbD variant was represented by two alleles (differentiated by one synonymous mutation), from which six other structural variants evolved. Four variants evolved directly from the progenitor, whereas two others emerged from these descendants (Fig. 4B). All natural CfaE and CsbD variants and the single CotD type induced mannose-resistant hemagglutination of bovine erythrocytes (data not shown). In place of this qualitative assay, we assessed the functional impact of mutations by measuring erythrocyte binding to purified fimbriae in parallel plate flow chambers. In this assay, all four CfaE variants showed shear-dependent binding, with significantly greater binding as shear increased from 0.049 to 0.19 Pa (Fig. 5A). In contrast to the progenitor CfaE (strain WS1933D), which showed a 10-fold rise in binding in this shear range (Fig. 5A), the relative binding   (as shown). The black dotted line in CsbD between residues 84 and 85 was used to fill the short span where PHYRE was unable to predict the structure with reasonable precision and therefore showed as a gap in the model. increase was much less for the three variants, due to higher erythrocyte binding at low shear. The CfaE/A128V variant showed 3-fold higher binding, whereas the CfaE/Q142R and CfaE/T91I variants exhibited more than four times higher binding than the progenitor under low shear (0.049 Pa) but similar levels of binding at a shear of 0.19 Pa.
The six CsbD variants were tested in the same manner. The progenitor CsbD variant (represented by LSN02-013966/A) mediated distinct shear-dependent binding of erythrocytes very similar to the primary variant of CfaE (Fig. 5B). At low shear, the other CsbD variants mediated ϳ2to 4-fold higher binding than the primary variant, with the highest binding exhibited by the most evolved mutants (represented by DS37-4 and E20738/0), which mediated essentially shear-independent erythrocyte binding. Hence, acquisition of point mutations in CfaE and CsbD was associated with higher erythrocyte binding at low shear and increasingly shear-independent binding thereafter.
We also assessed the erythrocyte binding of the one CS2 fimbrial adhesin (CotD) variant. Although CotD exhibited shear-enhanced binding, it was less pronounced than that of the primary CfaE and CsbD variants. Moreover, CotD binding was relatively strong at low shear (Fig. 5C).
There were too few natural variants observed in CfaB to determine their evolutionary significance. That the two variants contained one or both of two point mutations localized to the intersubunit interface ( Fig. 3A; CfaB/A37V in 13 strains and CfaB/A37S,T136S in one strain) suggested the possibility that these changes affect fimbrial function. Purified fimbriae from strains expressing the common CfaB type (WS1933D) and the two variants each showed marginal erythrocyte binding at low shear, which rose with increasing shear stress (Fig. 5D). Under 0.19 Pa shear, optimal for erythrocyte binding to fimbriae with the common CfaB variant, there was no significant difference in binding between fimbriae with the common CfaB and CfaB/ A37V. However, erythrocyte binding to the fimbrial carpet with CfaB/A37V,S136T was significantly lower than that with the common or single-site CfaB variant. Because all three of these fimbriae contained the progenitor CfaE type, and similar levels of CfaE were detected in each fimbrial preparation (as mea-sured by anti-CfaE antibody detection, data not shown), this difference could not be attributed to CfaE.
Comparative Distribution of Functional Mutations in CfaE, CsbD, and FimH Adhesins-The findings described above recall the microevolution of the type 1 E. coli fimbrial adhesin FimH from relative shear-dependent to shear-independent mannose-specific binding (36). On this basis, we compared the spatial distribution of the lectin domain mutations responsible for this phenotypic shift in FimH to those in CfaE and CsbD using the available crystal structures (FimH, CfaE) or computed atomic model (CsbD) as scaffolds. The lectin domain mutations in CfaE and CsbD mapped near the putative binding pocket at their upper pole, with all mutations significantly closer to this pocket than to the interdomain interface (p Ͻ 0.05) (see supplemental Fig. S1). In contrast, the functionally analogous mutations in FimH were positioned closer to the interdomain interface than to the binding pocket (p Ͻ 0.01). Although there was no statistical difference between the distribution of mutations in CfaE and CsbD (p ϭ 0.45), the difference between CfaE/ CsbD and FimH was highly significant (p Ͻ 0.05). These differences suggest that the basic structural mechanism underlying shear-dependent adhesion may be distinct in CfaE/CsbD and FimH.

DISCUSSION
A critical step early in bacterial pathogenesis is adhesion to host cells, often promoted by fimbrial attachment to specific host-cell receptors. As a principal target of adaptive immune responses, the major fimbrial shaft subunits (pilins) typically exhibit structural diversification (37), whereas the adhesive subunits of specific fimbriae are much more conserved. In this study, however, we show that the diversity of pilin and adhesin genes between three class 5 fimbrial types was high, whereas there was very low diversity within any particular fimbrial type, and this diversity was associated with functional adaptation.
Most silent mutations are functionally neutral and accumulate randomly over time, thus serving as a molecular clock (38,39). By using one of the proposed clock models (39; expecting 1.187 ϫ 10 Ϫ6 synonymous mutations/site/year rate), it was previously estimated that the ETEC pathotype emerged Յ30,000 years ago (24). This matches closely with our estimation of when the ETEC pathotype arose, based on the overall silent mutation rate in the four MLST loci of the ETEC strains studied herein (dS ϭ 0.030 Ϯ 0.004, i.e. ϳ25,000 years ago). Based on the silent mutation rate between adhesin or pilin genes, different class 5 fimbriae split over 0.7-1.6 million years ago, well before emergence of the ETEC pathotype. Yet, based on the extremely limited number of synonymous changes within the class 5 fimbrial genes, the latter were acquired by ETEC no earlier than 700 years ago. Recent homogenization due to evolutionary bottlenecks is unlikely to explain this discrepancy because it suggests that all three fimbrial subclasses that are found in clonally diverse ETEC strains underwent independent bottlenecks. Another possibility is that the fimbrial genes were acquired only recently and then exhibited fast horizontal spread among the relatively old ETEC lineages, possibly spreading along or replacing other colonization factors. Indeed, class 5a, 5b, and 5c fimbriae were found in 41, 12, and 10% of colo- nization factor-positive ETEC isolates respectively from hospitalized patients in Bangladesh, whereas about one-third of ETEC expressed atypical, non-class 5 fimbriae (40). It is also possible that the age of ETEC is overestimated. Indeed, the age of pathotypes like ETEC, Shigella etc. might be better estimated based not on the backbone genes as done previously, but the virulence genes that define the pathotype but spread under selection across strains.
Within each fimbrial type, replacement mutations clearly predominated over silent changes, suggesting the action of positive selection. Remarkably, in the CsbD protein, a stretch of 24 amino acids covering only 7% of the length of the protein accu-mulated 70% of the total mutations, of which most occurred at hotspot positions. All changes in the hotspot positions led to different amino acid changes, thereby ruling out any possibility that recombination contributed to these changes. Such strong footprints of recent positive selection for structural changes in the CsbD adhesin could be attributed to adaptive functional changes rather than to immune or phage escape where the swapping of large epitopes is more likely (41).
It was shown previously that adhesive interactions of fimbriated bacteria become stronger under increasing shear conditions (42,43), including those of the type 1 fimbrial adhesin FimH and the CFA/I fimbrial adhesin CfaE (19). Here, we show 19 Pa for each of them. Accumulation of erythrocytes at each shear value was recorded, and average velocity of bound erythrocytes was determined. The mutation information shown for each structural variant studied in PPFC was shown with respect to the "Primary" variant. The underlined mutations designate the mutations specific to the particular variants with respect to the corresponding ancestral variants (as shown in Fig. 4 for adhesin subunits). The strains subject to PPFC experiments are shown in parentheses as representatives of specific variants. RBC, red blood cells (or erythrocytes). that this mechanism of adhesion also is demonstrable for two other class 5 fimbrial adhesins. For FimH and CfaE, an allosteric catch bond mechanism has been invoked, where the binding domain toggles between low and high affinity conformations depending on the shear level. In FimH, naturally occurring mutations with shear-independent effect primarily occur near the interface of pilin and lectin domains, away from the mannose-binding pocket, highlighting that separation of the domains under shear results in switching of the mannose-binding pocket from low to high affinity conformation (33). Strikingly, for both CfaE and CsbD, mutations that converted binding from shear dependence to shear independence were localized primarily near the putative ligand-binding pocket rather than the interdomain interface. The markedly different distribution pattern of the binding enhancing mutations suggests that the shear-dependent catch bond binding properties of class 5 fimbrial adhesins could have a different structural basis than the one observed for FimH.
It has been shown that in FimH, the shear-dependent adhesin variants provide advantage over the shear-independent mutants in binding under high concentration of soluble inhibitors (44). Although for FimH, the shear-independent mutants are selected positively in strains causing urinary tract infections (especially upper tract infections in the kidneys, where low shear forces prevail), it is as yet unclear why such variants would be selected in ETEC and how this could affect the molecular pathogenesis of diarrhea. Possibly, the shear-dependent adhesin variants provide an advantage for binding to the intestinal epithelial surface, where the normal luminal milieu is highly mucoid. In contrast, the relatively shear-independent variants could provide advantage when watery diarrhea develops, whereby the intestinal fluid becomes more aqueous. Elucidation of the adaptive significance of different functional variants of the ETEC adhesins will require further experimental as well as epidemiological studies.
CfaB is the major, non-adhesive subunit comprising the CFA/I fimbrial shaft, arranged into a helical filament with 1000 or more subunit per single fimbria, rotating with respect to one another and the adhesin subunit CfaE (11,13). Both of the CfaB mutations observed here were positioned in the interface region of CfaB. These could potentially affect intersubunit interactions, changing the mechanical properties of the fimbrial shaft, which may uncoil and recoil under mechanical force (11). Therefore, natural mutations might play an important role in defining native helical filaments or uncoiled thin fibrillar structures. One of the mutations, T136S, resulted in decreased binding. These results highlight previous studies suggesting the importance of non-adhesive fimbrial shaft structure in modulating adhesive properties of the fimbrial tip adhesins (31,(45)(46)(47)(48). It remains to be determined, however, how exactly the point mutation(s) in the non-adhesive subunit could influence adhesin function.
Based on analytical and experimental work on adhesin variability, FimH in particular, a concept of pathoadaptive evolution of uropathogenic E. coli has been envisaged as a process complementary to horizontal transfer of virulence factors (20,36,49). The fact that ETEC class 5 adhesin mutations display patterns similar to FimH from a population perspective leads us to speculate that ETEC strains undergo pathoadaptive evolution as well. The present study thus builds on a common theme in the microevolution of bacterial pathogens and the virulence dynamics of specific pathotypes. Also, analysis of positive selection footprints in fimbrial adhesins provides insights in both the structural and functional properties of these virulence factors.