INTRODUCTION

Top Abstract Introduction Materials & Methods Results Discussion References

P fimbriae, the adhesins most clearly implicated in the pathogenesis of extraintestinal infection due to Escherichia coli (6, 15), mediate Gal(1-4)Gal-specific binding to host epithelial surfaces (3, 26, 28, 31-33) via the tip adhesin molecule PapG (9, 39). A possible explanation for the subtle differences in binding preferences among P-fimbriated E. coli strains that were noted following the discovery of P fimbriae (5, 7) came with the later discovery that PapG occurs in three molecular variants (38, 40, 42). The PapG variants, sometimes categorized in classes I, II, and III, bind preferentially to different Gal(1-4)Gal-containing compounds (25, 60, 61) and are encoded by distinct alleles of the adhesin gene papG (42) (Table 1).

Since receptor repertoire presumably determines a pathogen's host range (36, 60), efforts to define the three PapG variants' receptor specificities and clinical associations have been made (12, 13, 29, 30, 35, 36, 38, 50, 52, 57, 60, 61). Based largely on the study of single representatives of each papG class (often a recombinant strain), the concept that the three PapG variants can be differentiated phenotypically by their distinctive hemagglutination (HA) patterns with rabbit, sheep, and diverse human erythrocytes, which possess unique combinations of Gal(1-4)Gal-containing glycolipids, has emerged (11, 12, 25, 29, 35, 38, 60, 61) (Table 1). Rabbit cells are reportedly agglutinated only by the class I variant (60, 61), sheep cells are agglutinated only by the class II and III variants (12, 42, 60, 61; but see reference 35), and human O cells are agglutinated only by the class I and II variants (35, 36; but see references 12 and 29). HA patterns of different erythrocyte types have been used in epidemiological studies to classify wild-type E. coli strains according to their PapG repertoire (2, 58) and in laboratory studies to define the receptor specificities and host ranges of the PapG variants (11, 12, 25, 35, 36, 60).

However, by using slide HA assays with single representatives of each PapG class, we recently found that the HA patterns of the three PapG classes overlap considerably, even with supposedly class-specific erythrocyte types (24). We also found an unacceptably high degree of irreproducibility of inter- and intraobserver results with slide HA assays (24), which challenged the validity of conclusions regarding the PapG variants previously derived by such assays. Others have interpreted the unexpected finding of diverse (slide) agglutination phenotypes among wild-type E. coli strains containing only papG allele III (13) as evidence that phenotype is an unreliable indicator of papG status. However, whether this apparent phenotypic diversity within class III was due to true biological diversity or rather to assay irreproducibility was not determined.

These conflicting findings prompted us to reexamine two hypotheses, namely, (i) that the different papG alleles confer sufficiently distinctive HA patterns with human, rabbit, and sheep erythrocytes to allow phenotypic differentiation of strains with different papG allele configurations and (ii) that such HA patterns are uniform among wild-type strains of the same papG allele configuration. Our experience with slide HA assays (24) prompted us first to evaluate an alternative microtiter tray (MT) HA assay to determine whether it could assess E. coli HA phenotypes more reproducibly.

	MATERIALS AND METHODS

Top Abstract Introduction Materials & Methods Results Discussion References

Strains and papG PCR assay. Representatives of each known naturally occurring papG allele configuration, i.e., those of classes I plus III, III, II plus III, and II (13, 60), plus several papG-negative strains, were arbitrarily selected from available collections of papG genotyped clinical isolates of E. coli (14, 18, 21, 43, 55, 59) (Table 2). In addition, because of previously reported phenotypic discrepancies between recombinant and wild-type class III strains (12, 35), recombinant class III strain P678-54 (pJFK102) (30) was included. Strains were stored at 70°C until used. Each strain's papG allele configuration was determined by an allele-specific PCR assay as previously described (16). Primers for allele I were j96-193f (5'-TCGTGCTCAGGTCCGGAATTT-3') and j96-653r (5'-TGGCATCCCCCAACATTATCG-3') (461-bp product), primers for allele II were ia2-383f (5'-GGGATGAGCGGGCCTTTGAT-3') and ia2-572r (5'-CGGGCCCCCAAGTAACTCG-3') (190-bp product), and primers for allele III were prs-198f (5'-GGCCTGCAATGGATTTACCTGG-3') and prs-455r (5'-CCACCAAATGACCATGCCAGAC-3') (258-bp product) (16). papG genotypes were confirmed by at least one replicate PCR determination and by dot blot DNA probe hybridization with allele-specific probes, as previously described (22).

Serotypes and genomic fingerprints. O:H and O:K:H serotypes were determined at the Escherichia coli Reference Center at Pennsylvania State University and the International Escherichia and Klebsiella Centre (WHO), Copenhagen, Denmark, respectively, as previously described (48). For strain IA2 the published O6:H serotype (4) was used.

Erythrocytes. Human A₁P₁ and OP₁ erythrocytes were from two of the investigators, and p erythrocytes were from the University of Minnesota Blood Bank rare donor collection. Sheep and rabbit erythrocytes were obtained as needed during the course of the study from commercial suppliers (Micropure Medical, Wayzata, Minn.; Remel, Lenexa, Kans.); no information was available regarding the donors. Erythrocytes were washed in phosphate-buffered saline, pH 7.4 (PBS), and stored in Alsever's solution (Gamma Biologicals, Houston, Tex.) at 4°C for up to 1 month. (No decrement in HA behavior was observed over this storage interval.)

MT HA assays. MT HA assays were done as previously described (26, 34, 47, 49), with modifications, in experiments designed primarily to assess reproducibility of assay results (experiment 1), to compare HA patterns of strains within a single papG category (experiment 2), and to compare HA patterns between papG categories (experiment 3). For all experiments, bacteria freshly harvested from lawns after overnight growth at 37°C on Luria-Bertani agar plates (41) were suspended in PBS plus 5% -methyl-D-mannose (PBS-man; Sigma, St. Louis, Mo.). (Concentrations of approximately 10¹¹ CFU/ml, as determined by quantitative culture, were routinely achieved by suspending the entire lawn from a single culture plate in 1 ml of PBS-man.) Serial 2× dilutions of these stock suspensions were prepared in PBS-man with 96-well V-bottom MTs (Sarstedt, Inc., Newton, N.C.) at 25 µl/well with 12 wells per dilution series. To each well was added 25 µl of erythrocyte suspension (5% in Alsever's solution; Gamma Biologicals). In all experiments, separate dilution series were prepared for each strain with each erythrocyte type. In addition, duplicate dilution series were prepared for each strain from the initial stock suspension (experiment 1.0) or a separately prepared stock suspension of the same strain (all experiments). Strains were sequenced randomly throughout. Erythrocytes were sequenced either randomly (experiment 1.1) or in a fixed order (i.e., A₁P₁, OP₁, rabbit, and sheep) (other experiments).

View larger version (40K): [in this window] [in a new window]   FIG. 1.   Interpretation of results of MT HA assays. Circles represent wells in a 96-well V-bottom MT. Internal shading indicates erythrocytes. Numbers to the right are HA titers. Rows: A, all negative (no HA); B, partial HA only, changing to negative at low titer; C, partial HA only, but continuing with partial HA off scale to the right; D, slow transition from full HA to partial HA (wells 4 to 7) to negative; E, more brisk transition; F, high-titer full HA, continuing with partial HA off scale to the right; G, very-high-titer full HA, never changing even to partial HA.

HA titer reproducibility. HA titer reproducibility was assessed by comparing HA endpoints from duplicate determinations done on the same day by the same observer or different observers, with LFP- and FFN-well endpoints being considered separately. Paired HA titers that differed by 1 twofold dilution, that were both <1 (for LFP wells), or that were both >12 (for FFN wells) were said to agree (A); those that differed by 2 twofold dilutions were said to disagree. Comparisons were said to be indeterminate when one titer was 1 or 12 and the paired titer was <1 or >12, respectively; conflicting when only one titration showed HA; and negative (N) when neither showed HA. Reproducibility rates were calculated as (A + N)/total.

Detection of P and non-P adhesins. Prior to experiment 2, to exclude strains with non-P adhesins, microscope slide HA assays (24) were used to determine each strain's mannose-resistant HA (MRHA) titer for human A₁P₁ and p erythrocytes (Table 2). If HA was observed with either cell type, inhibition of HA by pigeon egg white (PEW) and 3% bovine serum albumin (Sigma) in PBS was assessed at a bacterial concentration four times that required for HA (19, 20). Strains that exhibited PEW-inhibitable MRHA of A₁P₁ erythrocytes but no MRHA of p erythrocytes were considered to express P adhesins; those that exhibited PEW-inhibitable MRHA of both A₁P₁ and p erythrocytes were considered to express NPMR (formerly X) adhesins (19, 20).

HA pattern matching. In experiment 3, our ability to predict a strain's overall HA pattern for human, rabbit, and sheep erythrocytes on the basis of its papG genotype was assessed. Two of the eight NPMR adhesin-negative strains from each papG category, selected to represent different HA patterns observed within the category, were designated positive controls. The remaining six strains served as test strains for that papG category. Two test strains from each papG category (eight strains total) and all eight control strains were assayed daily until all test strains had been assayed on two days each. Each day, duplicate stock suspensions of each test and control strain were titrated against (known) A₁P₁, OP₁, rabbit, and sheep erythrocytes. Observers blinded to test strain identity matched each test strain to one or more (known) controls by direct inspection (observer A) or by analysis of quantitative HA titers as recorded by observer A (observers A to C), based on similarity of overall HA patterns between the test strain and the controls. After a washout period, the three observers independently repeated the latter analysis. Duplicate assessments by each of the three observers were combined to give the day's total score for matching of each test strain with each control strain. Concordance was defined as matching of a test strain to a control of the same papG category. Reproducibility was evaluated by comparing (i) an individual observers' repeat assessments of the same strain (intraobserver) and (ii) results between observers (interobserver).

Statistical methods. Comparisons of proportions were tested by a ² test or Fisher's exact test.

	RESULTS

Top Abstract Introduction Materials & Methods Results Discussion References

Experiment 1: reproducibility of HA titers. Reproducibility of HA titers was assessed initially with two strains from each of the four papG categories plus two papG-negative strains (experiment 1.0), selected without regard for the presence or absence of NPMR adhesins. Reproducibility of same-day intraobserver HA titers for duplicate dilutions of the same bacterial suspension was 86.8% overall (analysis 1.0.1) (Table 3), and levels of reproducibility did not differ significantly between the two observers (87.9 versus 86.6%) or between the two endpoints used (86.8% for both LFP and FFN wells). Disagreement as to the presence of HA occurred in only 4 to 7% of comparisons (Table 3). For separately prepared stock suspensions of the same strain, agreement of intraobserver results was slightly worse, i.e., 81.8% overall (analysis 1.0.2) (Table 3) (P = 0.03 versus agreement of intraobserver results with the same suspension), but again levels of agreement were similar for the two observers and for the two endpoints used (not shown). Reproducibility of interobserver assessments of the same dilution series (analysis 1.0.3) (Table 3) approximated that of intraobserver assessments for duplicate dilution series from the same stock suspension (analysis 1.0.1) (Table 3).

Experiment 2: screening for NPMR adhesins. Having established the MT HA assay's reproducibility with wild-type P-fimbriated E. coli, we next used the assay to assess papG category-specific HA phenotypes attributable to PapG adhesins per se in the absence of NPMR adhesins. Strains representing the four papG categories were screened by slide HA assays for expression of NPMR adhesins (Table 2). NPMR adhesin-positive strains were excluded and replaced by alternative strains from the same papG category to give eight wild-type strains per category. Of the wild-type strains screened, 2 of 10 from category I plus III, 3 of 11 from category III, 2 of 10 from category II plus III, and 0 of 8 from category II expressed an NPMR adhesin. Of strains expressing a P but no NPMR adhesin, seven of the eight class II strains, but only one from another papG category (a category II plus III strain), exhibited a pseudo-NPMR HA pattern (P < 0.001) (Table 1). The only non-pseudo-NPMR class II strain was strain H16 (Table 2), which subsequently exhibited other phenotypic anomalies, as described below.

Experiment 2: HA phenotypes according to papG category. Within each papG category, both consistency and variability of phenotype were encountered. The most common HA pattern in each papG category is shown in Table 4 as pattern 1. Although the class I plus III strains typically gave only partial HA with any erythrocyte type, two strains (pattern 2) gave full HA of A₁P₁ and OP₁ erythrocytes. The class III strains varied mainly with respect to their complete (pattern 1) or partial (patterns 2 and 3) HA of sheep erythrocytes and their partial (patterns 1 and 2) or absent (pattern 3) HA of human erythrocytes. Of note, recombinant class III strain P678-54 (pJFK102) was a marked outlier among the class III strains, reproducibly exhibiting much-higher-titer HA for sheep erythrocytes (relative to that of any other erythrocyte type) than the wild-type class III strains, i.e., by 6 twofold dilutions, versus a median of 1.5 twofold dilutions (range, 1 to >7) (P < 0.001). The class II plus III strains usually gave full HA of rabbit cells (pattern 1), but some gave only partial HA, with either lower (pattern 2) or similar (pattern 3) titers relative to those of other cells. The class II strains varied mostly with respect to intensity of HA of rabbit versus sheep cells, i.e., higher (pattern 1), the same (pattern 2), or lower (pattern 3) intensity.

Experiment 3.0: comparison of HA phenotypes between papG categories. We next sought to determine whether, despite the phenotypic diversity within each papG category, the different papG categories could still be distinguished phenotypically from one another when they were compared on the same day. This was done by matching (unknown) test strains with positive controls based on overall similarity of HA patterns with the four erythrocyte types (as detailed in Materials and Methods).

Experiment 3.0: diversity of HA phenotype within papG categories. HA patterns differed significantly not only between papG categories but also within each papG category. Among the class I plus III test strains, although strain 518 matched exclusively (12 of 12 comparisons) the class I plus III controls, on the same days strain 553 matched other controls more frequently (5 of 12 comparisons [P = 0.04]), and although strain CP9 matched predominantly the class II control strain H16 (8 of 12 comparisons), paired strain BOS038 exhibited no such matches (P = 0.001). Among the class III only test strains, strain 1044 matched moderately well the class III controls (5 of 12 comparisons), but paired strain BOS0025 matched the class III controls exclusively (12 of 12 comparisons [P = 0.03]). Class III test strains BOS117 and CL09 exhibited similar levels of divergence when they were tested in parallel (they matched class III controls in 4 of 12 and 11 of 12 comparisons, respectively [P = 0.009]). Among the class II plus III test strains, matches with the class II plus III controls were significantly more frequent for strain CA032 (12 of 12 comparisons) than for paired strain BOS086 (1 of 12 comparisons [P < 0.001]) and for strain BOS119 (12 of 12 comparisons) than for paired strain BOS098 (3 of 12 comparisons [P < 0.001]). Among the class II test strains, a match to either of the class II controls was more common with strain BOS013 (11 of 12 comparisons) than with paired strain H26 (7 of 12 comparisons [P = 0.03]), and a match specifically to control strain IA2 was significantly more common with strain CL24 (9 of 12 comparisons) than with paired strain CL01B (4 of 12 comparisons [P = 0.005]).

Experiment 3.0: reproducibility of HA pattern assessments. The reproducibility of the phenotype matching system for overall HA patterns with human, rabbit, and sheep erythrocytes (Tables 6 and 7) approximated that of the underlying HA titer assay itself (Table 3). Discrepancies in experiment 3.0 with respect to overall HA pattern matches occurred in 18% of pairwise intraobserver comparisons and were similarly frequent for the three observers (who had considerable, moderate, and little experience with the assay system, respectively) (Table 6). Most intraobserver and interobserver discrepancies were conservative, i.e., they stayed within the same papG family (class I plus III and class III only, family A, versus class II plus III and class II only, family B) (Tables 6 and 7). Interobserver discrepancies (Table 7) occurred significantly more often with family A strains (23%) than with family B strains (10%).

Experiment 3.0: day-to-day variability of HA phenotypes. Several test strains exhibited day-to-day shifts in HA pattern during experiment 3.0, such as class III only strain 1044's matching predominantly (5 of 6 comparisons) the class I plus III controls on day 1 (versus 1 of 6 comparisons on day 2 [P = 0.08), in contrast to its matching predominantly (5 of 6 comparisons) the class III controls on day 2 (versus 0 of 6 comparisons on day 1 [P = 0.015]). Similar shifts were seen among the control strains, as tested in duplicate on six successive days in experiment 3.0 (Table 8). The class I plus III controls failed to exhibit the more intense HA of rabbit erythrocytes seen in experiment 2 (Table 3). Strain J96 did not agglutinate OP₁ cells at all on day 5 (Table 8), and BF1023 gave full HA of A₁P₁ and OP₁ cells on days 1 and 2 but not thereafter, despite equally-high-titer HA (Table 8). The class III strains gave intermittent HA of A₁P₁, OP₁, and rabbit cells and variable intensity HA of sheep cells (Table 8). The class II plus III strains varied with respect to their levels of HA of rabbit cells (full, days 1, 2, and 6; partial, days 3 to 5). Of the class II strains, IA2 gave inconsistent HA of rabbit cells and H16 never did exhibit its HA pattern from experiment 2 (Table 3), instead usually appearing more like one of the family A control strains (Table 8). (Of note, although some of these day-to-day shifts occurred simultaneously in both control strains within a papG category, others were specific to individual strains.)

Experiment 3.1: phenotype matching with randomly selected test strains. To more realistically simulate the evaluation of unselected wild-type strains, on two different days eight test strains were selected randomly from the pool (without regard to papG category) and matched phenotypically to control strains with respect to their overall HA patterns for human, rabbit, and sheep erythrocytes (as in experiment 3.0) by observers blinded to test strain identity. Reproducibility of global HA pattern matching was better in experiment 3.1 than in experiment 3.0, for both intraobserver discrepancies (4% [2 of 48 comparisons] versus 18% [25 of 138 comparisons] [P = 0.017]) and interobserver discrepancies (6% [6 of 96 comparisons] versus 16% [45 of 276 comparisons] [P = 0.015]). All intraobserver or interobserver discrepancies involved family A test strains, and five of eight strains were conservative with respect to papG family (not shown).

	DISCUSSION

Top Abstract Introduction Materials & Methods Results Discussion References

We rigorously evaluated an MT HA assay and found it to exhibit a high degree of reproducibility of same-day intra- and interobserver results despite blinding of observers to strain and erythrocyte type. We then used the assay to assess the HA phenotypes of wild-type clinical E. coli isolates representing all four naturally occurring combinations of the three papG alleles, plus a recombinant class III strain. We found that despite statistically significant overall differences in the HA pattern of the four papG categories, there was reproducible overlap of phenotypes between class I plus III and class III strains and between class II plus III and class II strains, phenotypic diversity among strains within each papG category that were tested on the same day, and day-to-day phenotypic variability for individual strains.

We found that the MT HA assay exhibits significantly better reproducibility of same-day results than we obtained by conventional slide HA assays (24), which are the more commonly used phenotyping methods for extraintestinal pathogenic E. coli. When HA titer reproducibility results from the present study are calculated as in our previous study of slide HA assays (A/[total  N]) (24), the MT HA assay's overall reproducibility of same-day intraobserver results for different suspensions was 83% (1,169 of 1,408 comparisons; experiments 1 to 3), versus the slide assay's 39% (50 of 127 comparisons; experiments R2 to R4 [24] [P < 0.001); its reproducibility of same-day interobserver results was 85.8% (398 of 464 comparisons; experiment 1), versus the slide assay's 73% (87 of 120 comparisons; experiments B1 and B2 [24] [P = 0.001). The MT HA assay's reproducibility was high both for individual HA titers (experiments 1 to 3) and for assessment of overall HA patterns with A₁P₁, OP₁, rabbit, and sheep erythrocytes combined (experiment 3).

The MT HA assay's documented reproducibility indicates that a valid method for assessing the HA phenotypes of the P-adhesin variants exists, quite the opposite of our pessimistic conclusion regarding slide HA assays (24). Whereas others have studied uropathogenic E. coli adhesins using MT HA assays (20, 45, 55, 56), the present study to our knowledge represents the first formal, quantitative assessment of the performance characteristics of MT HA assays in this application. It supersedes prior work in its use of blinding (56), randomization, replicate determinations, interobserver and intraobserver comparisons, and statistical testing; its detailed assessment of phenotype stability over time; and its use of a diverse sample of wild-type clinical isolates from each papG category, with careful exclusion of strains coexpressing NPMR adhesin. In addition, our pattern-matching system (experiment 3) is unique in its use of same-day positive controls for each papG category, which compensates for day-to-day phenotypic shifts due to erythrocyte variability (24), and in its use of two endpoints per titration, which enhances discriminating power.

Our findings with respect to the PapG variants' phenotypes challenge several prevailing assumptions, the first being that the three papG alleles confer sufficiently distinctive receptor specificities that they can be distinguished phenotypically based on differential HA patterns of human, rabbit, and sheep erythrocytes (11, 25, 60). We found that in the combinations in which the three papG alleles occur in nature, they do not confer sufficiently distinct HA phenotypes for accurate stratification of strains according to papG allele genotype. Although differing significantly from one another in the aggregate, the four papG categories overlapped substantially with respect to types of erythrocytes agglutinated and relative HA titers for each erythrocyte type. Control strains from different papG categories were often phenotypically indistinguishable (Table 8).

This phenotypic overlap between strains possessing different papG variants fits with our previous slide HA assay-based findings for the individual papG alleles (24). In the present study we observed pyenotypic merging between papG categories I plus III and III only and between categories II plus III and II only but little overlap between these two larger groups. These two aggregate groups, i.e., classes I plus III and III and classes II plus III and II, might be considered phenotypic families, internally similar but usually dissimilar to one another. A possible explanation for phenotypic convergence between papG categories I plus III and III or between II plus III and II is phase variation (37, 45, 62), with decreased or absent expression by dual-allele strains of one allele.

A second common assumption challenged by our findings is that papG genotype equates with phenotype, i.e., that strains of the same genotype have a uniform and stable MR adherence phenotype (11, 25, 60). Corollaries are that phenotype assessments not only are definitive for a given strain but also can be generalized to others of the same papG genotype (12, 60, 61) and that phenotype testing may even be dispensable (8, 52). We observed substantial diversity of phenotype within each papG category, even for the pure genotypes (classes II only and III only), where differential expression of papG alleles of alternate classes should not cause shifts in HA patterns. In addition, there was substantial day-to-day variability of phenotype for individual strains, including the class II only and class III only strains. That these findings were not artifactual is suggested by their reproducibility in replicate, blind determinations by the same observer (experiments 2 and 3) and different observers (experiment 3).

This reproducible phenotypic diversity among strains with the same papG allele(s) (and no NPMR adhesin), even when they are tested on the same day and with the same reagents, is consistent with the diversity of agglutination patterns reported among class III only strains (13) and extends this phenomenon to include the other papG categories. Such diversity may arise from subtle variations in receptor specificity within each papG category due to sequence differences within papG (42). There may also be strain-strain differences in P-fimbrial expression not involving PapG, unrecognized secondary adhesins, or differences in nonadhesin background properties.

Some strains clearly changed their phenotype considerably from day to day, even when other strains tested in parallel with the same reagents did not (e.g., Table 8), which argues against instability of erythrocytes as an explanation. For dual-allele strains (classes I plus III and II plus III), such shifts may occur by alternating expression of alleles of different papG classes (37, 45, 46, 62). A similar process may also occur in single-allele strains (class III only or II only), if multiple copies of pap are present (1, 10, 51), that contain subtly different versions of papG, albeit of the same class.

The observed phenotypic heterogeneity within papG categories suggests that in epidemiological studies phenotypes may need to be directly assessed, since phenotype presumably is more closely related to pathogenesis than is genotype but cannot be predicted reliably based on genotype. Additionally, conclusions regarding papG-specific adherence phenotypes derived from the study of single representatives of a particular papG category should be validated in a larger population. (Of note, since we tested only eight representatives of each papG category, broad generalization from our data may be premature and additional strains should be studied.) Finally, the day-to-day shifts in relative intensity of HA of human, rabbit, and sheep erythrocytes seen with individual strains argue that one day's HA testing is insufficient to define a given strain's P-adhesin-based phenotypic repertoire.

A third prevalent assumption regarding the papG variants which is challenged by our findings is that phenotypes associated with papG alleles are best studied for each allele in isolation, preferably with the use of recombinant strains (12, 36, 60, 61). We found that each of the two naturally occurring multiple allele papG categories (I plus III and II plus III) exhibits a characteristic range of HA phenotypes that in the aggregate differ significantly from the phenotypes of the single-allele papG categories (III only and II only). Thus, since both class I plus III and class II plus III strains are significant pathogens (13, 21, 23), the phenotypes associated with these combined papG alleles merit study.

In addition, a recombinant strain expressing J96's class III papG allele exhibited a reproducibly different HA phenotype than did wild-type class III strains, which calls into question the generalizability of previous in vitro observations regarding the class III papG variant derived through study of this or similar recombinant strains (12, 29, 30, 60, 61). Simple quantitative differences in the level of P-adhesin expression between recombinant and wild-type strains does not seem to explain our findings, since the recombinant class III only strain exhibited a selective HA titer increase for sheep erythrocytes, rather than the across-the-board increases for all cell types that would be expected from enhanced expression alone. This strain's outlier phenotype may be due instead to actual receptor specificity differences or to other factors such as host strain background. Whatever the explanation, our findings suggest that the class III recombinant strain may not be a reliable surrogate for wild-type class III strains for in vitro assays or in vivo models (38). Alternative approaches, such as knockout mutants of wild-type strains (44, 53), might allow more valid assessments of the receptor specificities and adherence phenotypes of allele III.

(Of note, no class I or II recombinant strain was included in present study because no phenotypic discrepancies between wild-type and recombinant strains for these adhesins have been reported in the literature. There is no known wild-type class I only strain against which a class I recombinant strain can be compared, and in our previous study of slide HA assays, a class II recombinant strain and its wild-type parent exhibited similar HA patterns with diverse erythrocytes [24].)

The similar behaviors of human A₁P₁ and OP₁ erythrocytes indicate that these cells give redundant information in the MT HA assay: either alone would suffice. The absence of the ONAP phenotype (as previously described for some class III strains) (35, 36, 57) fits with the concept that globo-A is not a necessary receptor on human erythrocytes for the class III adhesin and that alternative broadly prevalent receptors (such as Luke antigen, or stage-specific embryonic antigen 4) are sufficient (29). Our findings also reconfirm that the class III adhesin does not agglutinate only sheep or canine erythrocytes (24, 35). These observations are consistent with reports that the class III papG variant predominates among cystitis isolates from children (13) and women (22) and can be as prevalent as the class II variant among certain bacteremia isolates (17). In contrast, they conflict with the concept that class III strains narrowly target dogs (60) or humans of the A (36) or A₁ secretor (12) phenotypes. Further study is needed of the clinical implications of class III strains' diverse and variable adherence capabilities (14).

In summary, we found that individual HA titers for human, rabbit, and sheep erythrocytes, as well as overall HA patterns for these erythrocyte types combined, can be reproducibly assessed for P-fimbriated E. coli by an MT HA assay. Wild-type clinical isolates exhibit reproducible diversity of HA phenotypes both between and within the four naturally occurring papG categories, day-to-day phenotypic variability for individual strains, and phenotypic overlap between class I plus III and class III only strains and between class II plus III and class II only strains. The biological basis for and epidemiological implications of these findings remain to be determined.