Methods

Study design and sampling strategy. From October 1999 through January 2005, we conducted a series of cross-sectional studies at a California university health care service that included the following four 3.5-month sampling periods: period 1, 11 October 1999–31 January 2000; period 2, 11 October 2000–31 January 2001; period 3, 11 October 2003–31 January 31 2004; and period 4, 11 October 2004–31 January 2005. The collection of period 1 and 2 samples was nested within a separate 2-year study that examined changes in the prevalence of a major drug-resistant clonal group of uropathogenic E. coli in this student population [4, 7]. Period 3 and 4 samples were collected within a second 2-year prospective study initiated in October 2003 that examined the dietary habits of students with UTI [8]. Details of the 2 studies are provided in the respective publications [4, 7, 8].

During each period, urine specimens were obtained from patients consecutively presenting to the clinic with symptoms suggestive of UTI. A patient with E. coli UTI was defined as a patient who received a diagnosis of UTI (as stated on the laboratory referral documents) and had a urine culture yielding ⩾10² colonies per mL of urine with presumptively identified E. coli. If multiple urine specimens from the same patient were obtained, only the first specimen yielding an E. coli isolate (primary E. coli isolate) was included in the analysis. Study protocols were approved by the Committee for Protection of Human Subjects at the University of California at Berkeley.

Urine specimen collection and microbiological methods. All urine specimens obtained from clinic patients were picked up daily, preserved in 15% glycerol, and stored at −80°C until testing. Urine specimens were cultured by standard methods of the American Society for Microbiology [9]. Colonies isolated at concentrations of ⩾10² colonies per mL of urine with presumptively identified E. coli [7, 10] were selected for additional testing.

Antimicrobial susceptibility testing for trimethoprim-sulfamethoxazole (TMP-SMX), ciprofloxacin, and nitrofurantoin was performed during periods 1 and 2 by Etest strip (AB Bio-disk). During periods 3 and 4, as part of a new study design protocol [8], susceptibility testing for 29 antimicrobial agents representing 11 drug classes was performed by the broth microdilution method (Microscan Dade-Behring). All susceptibility testing was interpreted according to Clinical and Laboratory Standards Institute standards. Isolates exhibiting intermediate resistance were interpreted as resistant during analysis. An isolate was considered to be MDR if it was resistant to ⩾2 separate classes of antimicrobial agents.

Genotype analyses. All TMP-SMX-resistant E. coli isolates and either a randomly selected subset (period 1, 49 isolates [27%]; period 2, 104 isolates [62%]) or 100% (periods 3 and 4, 290 isolates) of TMP-SMX-susceptible isolates were genotyped by enterobacterial repetitive intergenic consensus 2 (ERIC2) PCR, as described elsewhere [4, 11]. Groups of ⩾2 isolates with ERIC2 electrophoretic banding patterns that were indistinguishable by visual inspection were designated to belong to ERIC2 clonal groups. Prototype uropathogenic strains CgA (ATCC BA-457) and CFT073 were included as reference strains for the ERIC PCR tests.

Statistical analysis. Comparisons of proportions were tested by Fisher's 2-tailed exact test. Cuzick's test for trend was performed to detect trends in resistance prevalence of the antimicrobial agents tested over the 4 periods of the study.

A test of negative binomial regression versus Poisson regression was used to examine the hypothesis that ERIC patterns displayed the same underlying prevalence of TMP-SMX resistance, pansusceptibility, or MDR. The basis of this test is that the negative binomial distribution can be thought of as an extension of the Poisson distribution that allows for variation in the underlying rates of antimicrobial resistance between the ERIC PCR patterns. A comparison of the relative fit of the Poisson and the negative binomial distributions via the log-likelihoods provides for a pseudolikelihood ratio statistic.

Temporal clustering of the major clonal groups, defined as the ERIC groups with ⩾20 isolates per group, identified over the 4 sampling periods was investigated with Pearson χ² analysis. Poisson and negative binomial regression tests were used to examine the hypothesis that the rate of occurrence of these ERIC clonal groups was constant throughout the study period. All statistical analyses were performed using Stata, version 9.0 (StataCorp).

Results

Study population and bacterial isolates. During the 4 sampling periods from 1999 through 2005, 1667 patients (age, 13–68 years) presented to the university health clinic (University of California, Berkeley) with clinical suspicion of UTI. Of these patients, 780 (47%) had primary E. coli isolates recovered at concentrations of ⩾10² colonies per mL of urine (table 1). E. coli accounted for 81% of the uropathogens isolated.

Antimicrobial resistance. Among the 780 E. coli isolates, the prevalence of resistance to TMP-SMX, ciprofloxacin, and nitrofurantoin was 18%, 2%, and 1%, respectively. No trends in the prevalence of resistance to TMP-SMX, ciprofloxacin, or nitrofurantoin were detected over the 4 periods (table 1).

Eleven (8%) of the 141 TMP-SMX-resistant isolates were also resistant to ciprofloxacin, and 2 (1.4%) were also resistant to nitrofurantoin. Isolates that were resistant to ciprofloxacin or nitrofurantoin were uncommon. Thirteen ciprofloxacin-resistant E. coli isolates were identified; 12 (92%) of these were MDR, 11 (85%) were resistant to TMP-SMX, and 1 was resistant to both TMP-SMX and nitrofurantoin. All 9 nitrofurantoin-resistant isolates were MDR, and those from period 3 and 4 were resistant to 5–8 classes of antimicrobial agents, including ampicillin and cephalothin.

During periods 3 and 4 (when 29 antimicrobial agents were tested), 169 (49%) of 346 isolates from these periods were susceptible to all 29 drugs tested (pansusceptible), 60 (17%) were resistant to a single agent, and 117 (34%) were resistant to ⩾2 classes of drugs (MDR). Fourteen (12%) of the 117 MDR isolates were resistant to ⩾6 of the 11 classes of drugs tested.

Among 346 E. coli isolates from periods 3 and 4, the prevalence of cephalothin and ampicillin resistance was 32% and 31%, respectively. The prevalence of resistance to ampicillin decreased from 35% in period 3 to 24% in period 4 (P<.05).

ERIC2 PCR genotyping results. Among 584 E. coli isolates tested by ERIC2 PCR, 35 distinct clonal groups, defined as those comprising ⩾2 isolates per group displaying visually indistinguishable electrophoretic banding patterns, were identified. The number of clonal groups identified and the proportion of isolates belonging to these groups increased with the increasing percentage of isolates typed by ERIC2 PCR during each period (table 2).

During period 1, genotyping of 47 TMP-SMX-resistant isolates (100%) and 49 randomly selected TMP-SMX-susceptible isolates (27%) identified 3 clonal groups. Three additional clonal groups were identified during period 2, when 38 TMP-SMX-resistant isolates (100%) and 104 (62%) of 168 TMP-SMX-susceptible isolates were typed [4, 7].

Concurrent genotyping of the 346 E. coli isolates during periods 3 and 4 revealed 118 unique ERIC2 patterns. Two hundred sixty-four isolates (75%) were identified as belonging to 33 distinct clonal groups. Of the 26 clonal groups first identified among period 3 isolates, 9 (35%) were no longer circulating during period 4. Of the 24 clonal groups infecting patients during period 4, three had not previously been identified.

The 4 major clonal groups, CgC (49 isolates), CgA (40), CgH (33), and Cg3 (20), accounted for 41% of all the E. coli isolates and 54% of the clonally grouped E. coli isolates during periods 3 and 4. CgC (72 isolates) and CgA (61) were present during all 4 sampling periods, CgH (50) was isolated during each of the last 3 periods, and Cg3 was recovered only during periods 3 and 4 (table 2).

ERIC2 clonal groups and antimicrobial resistance. The association of ERIC2 clonal groups with the prevalence of drug resistance was assessed. We found no statistically significant differences in the prevalence of TPM-SMX-resistant (P=.74), MDR (P=.36), or pansusceptible isolates (P=.54) between clonal and nonclonal group isolates during periods 3 and 4. However, antimicrobial drug susceptibility pattern was significantly associated with specific clonal groups, as assessed by the test of negative binomial versus Poisson regression (table 2).

Seventy-eight (55%) of 141 TMP-SMX-resistant E. coli isolates belonged to 11 clonal groups. During periods 3 and 4, these 11 clonal groups accounted for 41 (73%) of 56 TMP-SMX-resistant UTIs and only 159 (46%) of all 346 UTIs (P<.001). Four of these clonal groups (10 isolates) were entirely composed of TMP-SMX-resistant, MDR isolates and contributed 10 (18%) of 56 TMP-SMX-resistant UTIs (P<.001) and 10 (8.6%) of 117 MDR UTIs (P<.001), compared with 10 (2.9%) of all 346 UTIs, during periods 3 and 4.

The association of the major clonal groups (CgA, CgC, CgH, and Cg3) with antimicrobial resistance was examined further. These major clonal groups accounted for 203 (35%) of all of the 584 genotyped isolates in this study. A single clonal group, CgA, was responsible for 40 (12%) of all 346 UTIs during periods 3 and 4. However, during these periods, this clonal group accounted for 3 (60%) of 5 ciprofloxacin-resistant UTIs (P<.05), 19 (34%) of 56 TMP-SMX-resistant UTIs (P⩽.001), 22 (20%) of 108 ampicillin-resistant UTIs (P⩽.001), and 24 (21%) of 117 MDR UTIs (P<.001) (table 2).

None of the 61 CgC isolates found over the course of the study were resistant to ciprofloxacin or nitrofurantoin. Only 5 (8%; all isolated during period 3) were resistant to TMP-SMX. Although accounting for only 31 (13%) of the 230 UTIs during period 3, CgC was responsible for 17 (21%) of 80 ampicillin-resistant isolates (P<.05), 17 (20%) of 84 MDR isolates (P<.05), 5 (14%) of 37 TMP-SMX-resistant isolates (P=1), and 11 (10%) of 108 pansusceptible isolates (P=.18) during period 3. However, during period 4, CgC accounted for 18 (16%) of the 116 UTIs and 13 (21%) of 61 pansusceptible infections (P=.08) but only 1 (3.6%) of 28 ampicillin-resistant UTIs (P=.07), 1 (3%) of 33 MDR UTIs (P<.05), and none of the 19 TMP-SMX-resistant UTIs (P<.05).

During periods 3 and 4, CgH was responsible for 33 (9.5%) of all 346 UTIs, 3 (5.4%) of 56 TMP-SMX-resistant UTIs (P=.32), and 2 (50%) of 4 nitrofurantoin-resistant UTIs (P<.05). Cg3 infected 20 women during periods 3 and 4, accounting for 20 (5.8%) of 346 UTIs and 18 (11%) of 169 pansusceptible infections (P<.001). Eighteen (90%) of the Cg3 isolates were pansusceptible, and 2 (10%) showed intermediate susceptibility to cephalothin.

Temporal clustering of clonal groups. Temporal clustering, defined as the isolation of the same clonal group strain from ⩾2 women on the same day, was observed during all periods of our study. There were 33 instances (1.7% of all clinic visits) in which ⩾2 unrelated patients infected with the same ERIC2 clonal group presented to the clinic on the same day. Six clonal groups, including CgA and CgC, were responsible for these clusters (table 3). Although no significant temporal clustering was detected by χ² or negative binomial regression analysis, notable clusters of pansusceptible Cg3 isolates and TMP-SMX-resistant, MDR CgC isolates were observed during period 3.

Discussion

Large surveillance networks [2, 3] continue to report both increasing trends and marked geographic variation in the prevalence of antimicrobial resistance of uropathogenic E. coli strains; such data are often used to guide empirical treatment choices [12, 13]. However, reliance on UTI management strategies based on a limited number of routine urine cultures of specimens from patients with uncomplicated CA UTI may result in antimicrobial susceptibility data that are unrepresentative of these patients. To assess whether such biases exist in the estimation of prevalence of drug-resistant CA UTI, we conducted a population-based study in a single community over 4 different periods spanning 6 years. In each period, we assessed drug susceptibility of all consecutively collected E. coli isolates from women with CA UTI.

Contrary to expectation, we found no evidence of increasing or decreasing prevalence of drug resistance in our community, except for 1 instance of a decrease in ampicillin resistance between periods 3 and 4. Notably, 75% of this decrease in the prevalence of ampicillin resistance could be attributed to a single E. coli clonal group (CgC).

Our results are consistent with those from a study performed with a similar sampling strategy at the Stoneybrook University health service (Stoneybrook, NY). After comparing results from a 7-month study period in 2003 with those from a similar period in 1993, Ansbach et al. [14] found no significant increase in the prevalence of drug resistance. Interestingly, the 14% prevalence of TMP-SMX resistance among E. coli isolates recorded by Ansbach et al. [14] was observed in a community where TMP-SMX remained the most commonly prescribed empirical therapy for UTI, although in our community, with an 18% prevalence of TMP-SMX resistance, the health service had switched (in early 1999) from prescribing TMP-SMX to treating UTI with nitrofurantoin or ciprofloxacin. The prevalence of nitrofurantoin and ciprofloxacin resistance remained similar in both communities over the different study periods.

Our genotyping results support the growing body of evidence that most drug-resistant CA UTIs are associated with a limited number of strains of E. coli that belong to distinct phylogenic groups [15, 16] and are sometimes associated with community outbreaks [4–6]. Our study documents that the majority (75%) of all CA E. coli UTIs in our community were associated with ERIC2 clonal group membership. Earlier studies based on the typing of selectively or randomly sampled collections of isolates did not reveal this level of clonality [7, 17, 18].

Although clonal group E. coli isolates were no more likely to be antibiotic resistant than nonclonal group isolates, antibiotic resistance was concentrated within a small number of specific clonal groups. Furthermore, the 6-year comparison in the same community provided us with an opportunity to determine if pansusceptible clonal group strains became resistant over this period. Interestingly, there was little evidence that the acquisition of resistance by these initially pansusceptible strains contributed substantially to the prevalence of drug-resistant UTIs during any of the sampling periods.

During the first sampling period, investigators identified a previously unrecognized MDR genetically related group of E. coli, CgA. This single group was responsible for 11% of all E. coli isolates and 49% of TMP-SMX-resistant E. coli isolates from patients with CA UTI at the university health service during period 1 [4]. Subsequent studies have revealed that CgA is responsible for cystitis, pyelonephritis, and septicemia in the United States and Europe [17, 19–21]. Many CgA isolates exhibit similar MDR antimicrobial susceptibility patterns, PFGE profiles, and multilocus sequence type membership, and many carry a class 1 integron with a single arrangement of class 1 drug resistance gene cassettes (dfrA17-aadA5) [22, 23]. The isolation of E. coli strains indistinguishable by ERIC2 PCR that belonged to CgA from animals and retail poultry meat products [24, 25] suggests that contaminated food products could be a source of human drug-resistant CA UTI. Over the 6 years of our study, CgA accounted for 12% of all typed isolates and 30% of isolates resistant to TMP-SMX, ciprofloxacin, or nitro-furantoin (P<.00).

Our study demonstrates that the prevalence of drug-resistant E. coli UTI at any 1 time is greatly affected by the prevalence of a small number of circulating clonal groups of uropathogenic E. coli that are sampled during the study period. The probability that different women with no obvious common exposure would be infected with such drug-resistant clonal groups is highly unlikely. The resistant clonal E. coli groups that we detected are more likely to have already been resistant when introduced into this university community. These observations suggest that fluctuations in the prevalence of drug-resistant UTI in a community cannot be solely explained by local drug prescribing practices, regardless of what these prescribing practices may be. If the prevalence of resistant UTI in this community was a result of the human antibiotic prescribing or use practices, the selective pressures of the drugs should have yielded many more genetically distinct drug-resistant E. coli isolates. Thus, the usual recommendation to restrict human antibiotic use at the community level may not have the expected impact on diseases such as drug-resistant UTI. Strategies developed to maintain the usefulness of CA UTI empirical treatment options may need to include interventions that target sources of drug-resistant E. coli.