Bacterial clonal diagnostics as a tool for evidence-based empiric antibiotic selection

Despite the known clonal distribution of antibiotic resistance in many bacteria, empiric (pre-culture) antibiotic selection still relies heavily on species-level cumulative antibiograms, resulting in overuse of broad-spectrum agents and excessive antibiotic/pathogen mismatch. Urinary tract infections (UTIs), which account for a large share of antibiotic use, are caused predominantly by Escherichia coli, a highly clonal pathogen. In an observational clinical cohort study of urgent care patients with suspected UTI, we assessed the potential for E. coli clonal-level antibiograms to improve empiric antibiotic selection. A novel PCR-based clonotyping assay was applied to fresh urine samples to rapidly detect E. coli and the urine strain's clonotype. Based on a database of clonotype-specific antibiograms, the acceptability of various antibiotics for empiric therapy was inferred using a 20%, 10%, and 30% allowed resistance threshold. The test's performance characteristics and possible effects on prescribing were assessed. The rapid test identified E. coli clonotypes directly in patients’ urine within 25–35 minutes, with high specificity and sensitivity compared to culture. Antibiotic selection based on a clonotype-specific antibiogram could reduce the relative likelihood of antibiotic/pathogen mismatch by ≥ 60%. Compared to observed prescribing patterns, clonal diagnostics-guided antibiotic selection could safely double the use of trimethoprim/sulfamethoxazole and minimize fluoroquinolone use. In summary, a rapid clonotyping test showed promise for improving empiric antibiotic prescribing for E. coli UTI, including reversing preferential use of fluoroquinolones over trimethoprim/sulfamethoxazole. The clonal diagnostics approach merges epidemiologic surveillance, antimicrobial stewardship, and molecular diagnostics to bring evidence-based medicine directly to the point of care.


Introduction
The rising prevalence of antimicrobial resistance among bacterial pathogens is one of today's greatest medical challenges [1][2][3][4]. Because conventional cultures can take days to determine a a1111111111 a1111111111 a1111111111 a1111111111 a1111111111

7-SNP test
The qPCR-based 7-SNP test was performed essentially as described previously [19], with minor modification (see below). The test determines the clonotype identity of the bacteria based on a combinatory number (barcode) derived from the presence/absence of seven single nucleotide polymorphism (SNP) positions within two genes, fumC and fimH. An additional test reaction determines presence of the uidA locus, which is specific to E. coli. The qPCR test was performed using the Rotor-Gene Q instrumentation platform (QIAGEN, Inc). The Rotor-Gene qPCR tube strips, functionalized with reaction mixes containing SNP-and uidA-specific primers, as described previously [19], were supplied by ID Genomics, Inc. (Seattle, WA) and stored at -20˚C until use. To shorten the test time, two minor modifications were made to the previously described protocol. First, for bacterial lysis and crude DNA extraction, the bacteria/ Chelex beads pellet was heated at 96˚C for 3 minutes instead of 5 minutes as described previously. Second, the bacterial DNA was added to the functionalized qPCR test tubes after the latter were pre-heated for 3 min as a 'jump start' step (that was done in parallel to the bacterial DNA prep) and not before the pre-heating step as described previously. Neither step affected the test quality.

Urine culture and isolate testing
Each urine specimen also underwent quantitative culture, species identification and susceptibility testing according to standard procedures. A clinically significant bacterial load was defined as ! 10 3 cfu/ml [26].
positive test as the predictor and a positive culture as the outcome. Correlation between time to the positive test and E. coli load as detected by culture was analyzed using linear regression, with bacterial load given as log 10 cfu/ml. Comparisons between the study and reference isolate sets for the prevalence of the major constituent clonotypes were done using a chi-square test of independence or Fisher's exact test (if required). Resistance prevalence for all E. coli isolates within the study and reference sets was compared individually for each antibiotic using a twosided Fisher's exact test. The difference in overall resistance prevalence between study and reference isolates was additionally evaluated using multivariable logistic regression, unadjusted or adjusted for the major clonotype CT561. Disparities between resistance of major clonotypes within study set and either of the two reference sets in clonotype/antibiotic profiles for major clonotypes were compared using McNemar test.

Calculations of prescription mismatch rates
Prescription mismatch rates were calculated for a subset of cases where E. coli was detected both in the 7-SNP test and by culturing, and an antibiotic was prescribed at the index day of the specimen collection. Observed prescription mismatch rate was calculated for individual classes of antibiotics and overall as the percent of cases where E. coli isolates were resistant to the prescribed antibiotic according to the isolates' culture-based antibiogram. Expected mismatch rate was calculated for individual classes of prescribed antibiotic based on the rate of prescription of this antibiotic in the study patients and the overall rate of resistance to that antibiotic in the study E. coli isolates according to the culture-based antibiograms. Predicted mismatch rate was calculated for individual classes of prescribed antibiotic as follows: in each case the drug was considered as an acceptable choice ("allowed") for a strain of a particular clonotype if in the reference set the resistance prevalence in that E. coli clonotype was not above the threshold level of 20% [11]. Otherwise, use of that antibiotic would be rejected ("disallowed"). If an antibiotic was allowed for a particular isolate, but this isolate happened to be resistant to this drug according to its culture antibiogram, it was defined as a predicted mismatch. Mismatch rates were compared using Fisher's exact test.

7-SNP test determination of E. coli presence and load
Of 750 urinalysis-positive samples from urgent care patients with suspected UTI, 274 (36%) contained E. coli according to the 7-SNP test (Table 1, S3 Dataset). Quantitative culture detected E. coli at ! 10 2 cfu/ml (the detection limit) in 267 (97%) 7-SNP-positive and 39 (8%)  Table 1). In 7-SNP negative samples neither E. coli-specific uidA nor any SNP-specific primers gave a distinct positive signal, indicating the inability to detect E. coli in those samples is not likely to be due to the lack of 7-SNP markers in the strains. The 7-SNP test's failure to detect E. coli in 37 of 47 specimen with a bacterial load <10 3 cfu/ml likely indicates the limit of qPCR-based detection of E. coli under the current protocol. The 7-SNP test's failure to detect E. coli in 2 of 259 specimen with a bacterial load ! 10 3 cfu/ml could also indicate the presence of qPCR inhibitors in the urine sample.
Of 306 urine samples where E. coli was cultured, 45 (15%) had other microorganisms cultured as well. In 32 (71%) of the 45 polymicrobial cases, the E. coli load was <10 3 cfu/ml, and only 6 of yielded a positive 7-SNP test. In contrast, in the remaining 13 mixed cases, which had an E. coli load ! 10 3 cfu/ml, the 7-SNP test was uniformly positive. Among the 19 polymicrobial cases with a positive 7-SNP test, the majority (90%) contained enteroccoci or other Grampositive species as the non-E.coli bacteria, and E. coli was a minority species in only 3 cases, i.e., <1% of all 306 cases. In 7 of 19 polymicrobial cases E. coli was by far the most prevalent organism. For further analysis, presence of non-E coli species in polymicrobial specimens was disregarded.
On average, the 7-SNP test detected E. coli in urine specimens in 20 qPCR cycles, or 20.5 ± 3.4 min. Time to detection corresponded inversely with E. coli load (S1 Fig). Including the 8-min sample preparation time, the 7-SNP test could detect on average a clinically significant E. coli load in 28.0 ± 3.4 minutes.

Clonotypes in urine samples vs. cultured isolates
In urine samples that contained E. coli according to the 7-SNP test, clonotype identity could be defined in 22 PCR cycles (or 23.0 ± 3.2 minute), or 31.0 ± 3.2 minutes after sample availability. Among the 267 samples that contained E. coli by both the 7-SNP test and culture, in 260 (97%) the clonotype determined directly in urine corresponded with that determined by single-colony testing. Of the seven discrepant samples, four had < 10 3 cfu/ml and one had two clonotypes by single colony testing, one of which matched the 7-SNP test-determined clonotype. The 268 culture-confirmed E. coli study isolates represented 33 clonotypes (Fig 1), 31 of which were present in the local reference set (S1 Table), with the two novel clonotypes represented by a single isolate each.
Among the study isolates, 12 clonotypes accounted for ! 10 isolates each, and for 78% of the collection (Fig 1). The only significant clonotype-specific resistance prevalence differences between the study and reference isolates involved CZ in CT620 (0% vs. 20%, respectively) and FQs in CT571 (57% vs. 25%, respectively) (Fig 2). Likewise, at a 20% resistance threshold for allowed empiric use of an antibiotic, the local reference set disagreed with the study set for only 3 antibiotic-clonotype combinations, i.e., AMP in CT531 and CT721, and T/S in CT361 (Fig 2). The observed difference between the study and reference database with respect to the resistance profiles of a few clonotypes could be due at least in part to a possible differential prevalence of sub-CT clonal groups that are not recognized as separate clonotypes by the current resolution of the 7-SNP test.
Notably, the difference in overall FQ resistance prevalence between the reference and study isolates became non-significant (P = .51) after adjusting for the corresponding difference in prevalence of CT561 (7.4% vs. 16.3%, respectively: OR 2.42, CI 95% 1.66-3.59, P < .001), which corresponds with the pandemic multi-drug resistant clonal group ST131-H30. Thus, differential inter-clonal rather than intra-clonal variability in resistance prevalence is a critical determinant of overall resistance patterns.

Antibiotic prescribing and mismatches for study subjects
An antimicrobial prescription accompanied the index visit for 220 (82%) of 267 subjects with urinary E. coli according to both the 7-SNP test and culture. Most prescriptions (116, 53%) were for FQs, followed by T/S (63 29%), CZ (19, 9%), and NIT (100, 5%) ( Table 2). Other antibiotics included the AMP congener amoxicillin (3 prescriptions), CTR (2), 2 nd -generation cephalosporins (3), amoxicillin/clavulanate (2), clindamycin (1), and vancomycin (1). Optimizing empiric antibiotic selection Overall, for 38 of 220 subjects (17.3%) the infecting strain was resistant to the prescribed antibiotic, resulting in antibiotic/pathogen mismatch ( Table 2). The observed mismatch rate was not different (P = .39) from the mismatch rate expected from specific resistance prevalence in the study set (see Materials and Methods), suggesting that the prescribers had no special insight into the probability of resistance.

Antimicrobial therapy options allowed by the 7-SNP test
We next evaluated how effectively the 7-SNP test could guide empiric therapy if used in conjunction with local reference isolate clonal antibiograms and a 20% allowed resistance threshold for empiric use. (For the 2 novel and 4 unidentified CTs, population-wide E. coli resistance values were used.) Among the 220 E. coli -positive study subjects who received antibiotics, the 7-SNP test would allow use of FQs for 78.6%, T/S for 57.3%, CZ for 73.6%, and NIT for 100% (Table 2), with a resulting frequency of antibiotic/pathogen mismatch of only 4.6%, 7.9%, 4.3%, and 0.3%, respectively. Additionally, it would allow use of CTR for 100% (4.1% mismatch), FOS for 100% (1.9% mismatch), and AMP for 8.2% (33% mismatch).
In contrast, if prescribers were to prescribe drugs disallowed by the 7-SNP test, antibioticpathogen mismatch would be much more frequent, i.e., 77% for FQs (P < .001), 47% for T/ S   Fig 2. Comparison of antibiotic resistance prevalence among study vs. local reference Escherichia coli isolates for the 10 most frequent study clonotypes. a For each CT (clonotype) the most prevalent (! 90% of isolates) sequence type (ST), as determined by multilocus sequence typing, is shown, with its subclone determined by CH typing (H) in parentheses, where applicable. For CTs represented by multiple STs (each accounting for 90% of the constituent isolates) up to four major STs and the corresponding H subclones are listed. b CT571 comprises ST14 (H27), ST404 (H27) and ST1193 (H64), which all belong to clonal complex of ST14. c For the study set and local reference set ('Ref'), antibiotic resistance prevalence for all E. coli and for major clonotypes is given as percent of resistant isolates in relation to the total number of isolates, overall or within the clonotype; green color indicates resistance below conventional threshold ( 20%), red color-above it (> 20%); the resistance within study and reference sets was compared individually for each antibiotic using a two-sided Fisher's exact test; * P < .05. (P < .001), and 14% for CZ (P = .014) ( Table 2). Importantly, for the actual antibiotic/pathogen mismatches observed in the study, the 7-SNP test would have disallowed the prescribed drug in 15 (94%) of 16 instances involving FQs and 11 (79%) of 14 involving T/S.

Alternative antimicrobials allowed by the 7-SNP test
As noted above, the 7-SNP test would have allowed NIT for all 220 E. coli -positive, antibiotictreated patients and at least one of the other three most frequently used antibiotics (FQs, T/S, and CZ) in 173 cases. In these 173 cases, all three of these non-NIT agents would have been allowed for 103 patients (46.8%); FQs and CZ, but not T/S, for 45 (20.5%), and FQs and T/S, but not CZ, for 23 (10.5%) (Table 3). Thus, in addition to NIT, both T/S and FQs would have been allowed in 126 of 173 (72.8%) cases, and both CZ and FQs in 148 of 173 (86.8%) cases, whereas FQs alone (but not T/S or CZ) would have been allowed in just 2 (0.9%) cases.  (4); "CZ" cephalexin (19); "AMP" amoxicillin (3); "CTR" cefixime (2); "other" cefuroxime (3), amoxicillin/clavulanate (2), clindamycin (1), and vancomycin (1) One can estimate that if FQs, T/S, and CZ were used in accordance with the 7-SNP test, but in the observed proportions (Table 2), the cumulative antibiotic/pathogen mismatch rate would be 8.0%, as compared with the observed mismatch rate of 19.1% (Table 4). If instead T/ S were to be used preferentially when allowed by the 7-SNP test (consistent with IDSA guidelines), followed by FQs or CZ as the 2 nd or 3 rd choice only if T/S was rejected, the combined mismatch rate would be 7.5%-9.1%. Thus, use of the 7-SNP test to guide empiric antibiotic selection could allow significantly increased T/S use and decreased FQ use, while significantly reducing the risk of antibiotic/pathogen mismatch.

Effect of using different resistance thresholds
We next estimated how the use of different arbitrary resistance thresholds (10% or 30%, instead of 20%) would affect allowed antibiotic use, with a focus on T/S. With a 10% threshold, T/S would be allowed in only 37% of cases (vs. 57% with the 20% threshold; P < .001), but with only 3.7% resulting mismatches (vs. 7.9% with the 20% threshold, P = .22; S2A Fig). With a 30% threshold, T/S would be allowed in 71% of cases (vs. 57%, P = .004), with 15% resulting mismatches (vs. 7.9%, P = .074), which is still lower than the IDSA-suggested 20% resistance prevalence threshold for cystitis, and the observed 25% mismatch rate. Notably, with a 30% resistance threshold, T/S would be allowed as an alternative to FQs in fully 82% of cases (vs. only 73% with the 20% threshold, P = .055).

Effect of using multi-national vs. local reference databases
We estimated next the effect of using clonotype-specific antibiograms from a multi-national (non-Seattle) reference set, rather than the local reference set, to determine clonotype-guided use of antibiotics. Clonotype/antibiotic resistance prevalence profiles for the top 10 study clonotypes were compared head-to-head between the study set and both reference sets, which were compiled, respectively from the local and multi-national isolate collections (S2 Table).
As noted above for the local reference database, with the multi-national reference database the major clonotypes usually did not differ significantly from the study clonotypes for the prevalence of resistance to the six antibiotics tested-AMP, CZ, CTR, FQ, T/S and NIT (resistance Table 4

Frequency of use of the 1 st , 2 nd and 3 rd -choice antibiotics (prescription rate, %) a
Total antibiotic-pathogen mismatch rate b P value Observed' indicates the prescription rate for these three classes of antibiotics that was observed in the study patients (see Table 2).

FQ a T/S a CZ a Predicted Expected
https://doi.org/10.1371/journal.pone.0174132.t004 Optimizing empiric antibiotic selection to fosfomycin was not determined in the multi-national dataset). However, significant resistance prevalence disparities vis-a-vis the study isolates were somewhat more frequent with the multi-national reference isolates (i.e., 7/60 clonotype/antibiotic combinations) than with the local reference isolates (2/60 total clonotype/antibiotic combinations) (P = .059 in McNemar's test). The most pronounced difference between the multi-national and local reference databases for non-concordance with the study isolates was when a 10% resistance threshold was used to classify clonotypes as susceptible vs. resistant to a given antibiotic. With this threshold, whereas the local reference clonotypes were non-concordant with the study clonotypes in 5 clonotype/antibiotic combinations, the multi-national reference clonotypes were non-concordant in 19 combinations (P < .001). In contrast, with a 20% resistance threshold, the local and multi-national reference clonotypes did not differ significantly for non-concordance with the study clonotypes (3 vs. 7 clonotype/antibiotic combinations, respectively, P = .16). With a 30% threshold even less non-concordance was observed (9 vs. 12 combinations, respectively, P = .26).
Although the multi-national reference database was not quite as accurate as the local reference database for predicting resistance among the study isolates, with the IDSA-recommended 20% resistance threshold its use would reduce only slightly the frequency of allowed T/S use relative to use of the local reference-51% vs. 57%, respectively (P = .11) (S2 Fig). Also, at this threshold, use of the multi-national vs. local reference would raise insignificantly the associated antibiotic/pathogen mismatch rate in T/S-8.9% vs. 7.9% (P = .48). At the alternative resistance thresholds (10%, 30%), however, T/S use would be allowed for fewer cases based on the multi-national reference database than the local reference database-22% vs. 37%, respectively, for the 10% threshold (P < .001), and 57% vs. 71%, respectively, for the 30% threshold (P = .002) (S2 Fig). Still, the resulting mismatch rate would not differ significantly-7.1% vs. 3.7%, respectively, for the 10% threshold (P = .40) and 10.4% vs. 14.8%, respectively, for the 30% threshold (P = .14).
Thus, a reference clonotype database that does not include local data, despite not being as concordant with the study isolates as the local clonotype database, could still be used for clonotype-based diagnostics and yield a significant reduction in mismatch rates with empirical antibiotic use.

Discussion
In this prospective observational cohort study we assessed a prototypic clonal diagnostics method for guiding selection of empirical therapy for E. coli UTI. Our findings support three main conclusions. First, when the test is performed at the point of care (here, a busy metropolitan urgent care clinic), it can detect E. coli and determine clonal identity with high specificity and sensitivity within 25-35 min of urine specimen availability, which is an acceptable timeframe for empiric antibiotic prescription. Second, if empiric prescribing were guided by clonal antibiograms from a pre-existing reference database, the frequency of antibiotic/pathogen mismatch could be reduced considerably. Third, use of clonal diagnostics could promote antimicrobial stewardship by encouraging empiric use of preferred antibiotics (e.g., T/S) over less preferred antibiotics (e.g., FQs).
Clonotype-level antibiograms provide much more accurate guidance for empirical treatment than do species-level antibiograms, the current standard [11,19,20]. In our population, adherence to IDSA-recommended resistance thresholds would preclude T/S for empiric therapy of uncomplicated cystitis due to the 25% resistance prevalence (cf. the suggested 20% threshold). In contrast, according to clonal antibiograms, T/S conceivably could be used empirically for twice as many patients as actually observed (i.e., from 29% to 57%), yet with an almost 67% relative reduction of the frequency of antibiotic/pathogen mismatch (from 22% to 8%). Similarly, given the~20% overall prevalence of FQ resistance, empiric FQ therapy was marginally acceptable even for uncomplicated cystitis, and was definitely unacceptable for pyelonephritis. In contrast, if guided by clonal diagnostics, empiric FQ therapy could be used for almost 80% of isolates, with only a 4.6% mismatch, which would qualify well for empiric treatment even of pyelonephritis.
As had been reported previously, the reagent cost of the clonotyping test is already quite low (< $4), and the hands-on time could be reduced significantly with future test optimization. We expect that the main cost-benefit advantage of using clonal diagnostics would be in reducing the prescription of ineffective antibiotics, while constraining overuse of broad-spectrum and/or last-lane antimicrobials. Indeed, while the commonly used antibiotics are relatively cheap (but vary significantly in frequency of use in relation to provider), prescription of an antibiotic regimen to which the E. coli isolate is resistant is strongly associated with clinical persistence and treatment failure and, thus, need for new prescriptions, extended patient discomfort, repeated outpatient visits or prolonged hospital stay [27][28][29]. However, quantifying the cost benefits of the improved accuracy of antibiotic prescription allowed by the 7-SNP test would require a separate study.
Although patient-specific factors (e.g., drug allergy history, renal dysfunction, pregnancy, and drug-drug interactions) often constrain prescribers' antibiotic choices, clonal diagnostics could reduce empiric FQ overuse by allowing safe substitution of T/S in nearly 3 of 4 cases when both drugs are allowed. Additionally, with clonal diagnostics guidance, expanded use of 1 st generation cephalosporins could further diminish empiric FQ use, which could be especially useful in children and pregnant women, in whom FQs (and sometimes T/S) may be undesirable.
Rapid molecular tools have been explored for antimicrobial resistance prediction by targeting genetic resistance markers [30,31]. Unfortunately, resistance to a given drug often depends on the presence and proper expression of a wide range of genes and variants. In E. coli alone, at least 14 different genes can confer T/S resistance, and a dozen different mutations in three chromosomal genes as well as from any of multiple plasmid-borne genes can confer FQ resistance [32][33][34][35]. Thus, accurate prediction of resistance/susceptibility to a broad panel of antibiotics currently remains infeasible for a test based only on resistance gene markers.
Before broader clinical implementation, the studied clonotyping test would benefit from technological improvements, including a) fewer hands-on steps and automation of the readout; b) increased sensitivity to detect low concentrations of E. coli that still could be indicative of UTI; c) finer clonotyping resolution to identify distinct sub-CT clonal groups, and, potentially, d) use of cheaper and point-of-care-friendly non-PCR DNA amplification/detection platforms. Nonetheless, the current findings provide solid proof-of-principle that a clonal diagnostic approach can optimize selection of empiric therapy for E. coli UTI. Due to their temporal and geographic stability, clonal antibiograms also supersede species-level antibiograms in being usable in different locales and over several years [22][23][24]. In the future, clonal reference databases could also include data on a broad range of antibiotics for different clonotypes as well as their minimal inhibitory concentrations, which also could differ by clonotype [36,37].
Diagnostic reliability of clonal reference database is one of the most critical conditions to use it as a successful tool for guiding empiric treatment based on the clonal identify of the infecting agent. Indeed, compilation of such database went through multiple refining cycles [19,20] to balance reference strains based on, among other parameters, patient demographics and clinical presentation. Furthermore, continued sustainability and improvement of the quality of clonal diagnostics will require the database to incorporate self-learning features to fine tune to both local and global dynamics of the antimicrobial resistance.
Our study has limitations. First, it was observational, leaving in question whether clinicians would act on guidance from the 7-SNP test, and the test's actual clinical impact. (Notably, the clinicians did not adhere to IDSA-recommended empirical regimens.) Second, it focused on E. coli-containing urine samples, whereas other organisms cause a variable proportion of UTIs, depending on the context [5,6]. Third, it used antibiotic/pathogen mismatch as a surrogate for ineffective therapy, which, although documented previously for T/S and FQs [38,39], was not directly studied here. It also has notable strengths, including its prospective cohort design, urgent care clinic setting, attention to point-of-care feasibility, local and multi-national reference databases, ample-sized study population, and assessment of test performance with respect to turn-around time and potential impact on prescribing.
As a key strategy for confronting the threat of antimicrobial resistance, the WHO 2014 report lists the need to "improve antimicrobial use supervision and support of clinical practices, especially diagnostic and treatment strategies" [40]. This study provides proof-of-principle that clonal diagnostics is a promising approach for optimizing empirical antibiotic therapy, promoting antimicrobial stewardship, and tracking the dynamics of emerging antibiotic-resistant pathogens.
Supporting information S1  As the validation set we used the 220 subjects whose urine was positive for E. coli both by culture and 7-SNP test, and who had an antibiotic prescribed at the index visit or the next day. As the reference set we used either the local reference set (A) or a multinational reference set (B). For 10%, 20% and 30% resistance prevalence thresholds we calculated the percent of allowed T/S use (solid grey line) and the rate of mismatches in allowed cases (solid red line) and in disallowed cases (dotted red line). (TIFF) S1 Dataset. Escherichia coli isolates from the local reference collection. Following variables are listed: ID (isolate's name in UW collection), clonotype (septatype), resistance to seven antibiotics (AMP, ampicillin, CZ, cefazolin, CTR, ceftriaxone, T/S, trimethoprim/sulfamethoxazole, CIP, ciprofloxacin, NIT, nitrofurantoin, FOS, fosfomycin). Resistance is given in '1' for resistant, '0' for susceptible, '.' for missing data. (XLSX) S2 Dataset. Escherichia coli isolates from the multi-national (non-Seattle) reference collection. Following variables are listed: ID (isolate's name in UW collection), clonotype (septatype), resistance to six antibiotics (AMP, ampicillin, CZ, cefazolin, CTR, ceftriaxone, T/ S, trimethoprim/sulfamethoxazole, CIP, ciprofloxacin, NIT, nitrofurantoin). Resistance is given in '1' for resistant, '0' for susceptible, '.' for missing data. (XLSX) S3 Dataset. Samples from prospective clonal diagnostic study. Following variables are listed: study ID (name of the de-identified sample); bacterial load determined by 7-SNP test and bacterial load determined by culture (log10 cfu/ml); clonotype (septatype) determined in 7-SNP test directly from urine sample and clonotype from cultured bacteria; and resistance to seven antibiotics (AMP, ampicillin, CZ, cefazolin, CTR, ceftriaxone, T/S, trimethoprim/sulfamethoxazole, CIP, ciprofloxacin, NIT, nitrofurantoin, FOS, fosfomycin). Resistance is given in '1' for resistant, '0' for susceptible, '.' for missing data. (XLSX)