A total of 101 clinical E. coli isolates were collected consecutively from IUCs, UTIs, blood, sputum, wounds and ears of patients from August 2010 to February 2011 and were identified as ESBL producers in the Medical Laboratory Division of Chiayi Branch, Taichung Veterans General Hospital. After the further identification of E. coli by biochemical methods, confirmation of ESBL at National Chiayi University and the removal of genetically identical isolates from same patient, 87 ESBL-producing isolates were analyzed. All of the E. coli isolates were routinely incubated on EMB agar for further use. Furthermore, 18 isolates from 9 patients who were infected recurrently or at different sites within a short period were investigated. A recurrent E. coli infection was defined as a patient who was infected by a culture-proven E. coli more than 1 month after the completion of the initial antimicrobial therapy.

Antimicrobial susceptibility to ampicillin (AMP, 30 μg), cefotaxime (CTX, 30 μg), ceftazidime (CAZ, 30 μg), ceftriaxone (CRO, 30 μg), cefotaxine/clavulanic acid (CTX/CLA, 30/10 μg), ceftazidime/clavulanic acid (CAZ/CLA, 30/10 μg), ciprofloxacin (CIP, 5 μg), nalidixic acid (NAL, 30 μg), levofloxacin (LVX, 5 μg), moxifloxacin (MXF, 5 μg), piperacillin (PIP, 100 μg), piperacillin/tazobactam (TZP, 100/10 μg), cefepime (FEP, 30 μg), ertapenem (ETP, 10 μg), imipenem (IPM, 10 μg) and meropenem (MEM, 30 μg) was determined using the disc-diffusion method and CLSI standards [17] . The antimicrobial discs were purchased from BD (Becton Dickinson and Company, Sparks, MD, USA). Escherichia coli ATCC 25922 was used as the reference strain. ESBL isolates were determined by the diameter differences of the inhibition zones between cefotaxine and cefotaxine/clavulanic acid or between ceftazidime and ceftazidime/clavulanic acid. AmpC production was also examined.

The primers used to amplify genes for five β-lactamases and 12 virulence factors are listed in Table 1. Three multiplex PCR sets were developed by modifying previous primer sequences. Antimicrobial resistance genes were detected by Multiplex PCR I for bla_SHV, bla_CTX-M3-like, bla_CTX-M14-like, bla_TEM and bla_DHA and by single PCR for bla_CMY-2. Twelve virulence genes were amplified by two multiplex PCR sets: hlyA, usp, sat, fyuA, ironN, iutA and iucD for multiplex PCR II and fimH, cnf1, traT, kpsMTII and papGII for Multiplex PCR III. All PCR products were separated by electrophoresis on a 2% agarose gel at 50 V for 120 min. After ethidium bromide staining and UV illumination, the gel image was recorded and analyzed.

The procedure for PFGE was described previously [18] . Briefly, 10 U of XbaI were used for the restriction digestion. Isolates were defined as the subgenotype for those with ≤3 fragment differences and as the genotype for those with >3 fragment differences. Plasmid size and number were analyzed using the Kado-Liu method [19] , and Salmonella Choleraesuis OU7085 (6.6- and 50-kb plasmids) and S. Choleraesuis OU7526 (50- and 90-kb plasmids) as a control. Furthermore, the plasmid replicon groups FIA, FIB, FIC, HI1, HI2, I1-Igamma, L/M, N, P, W, T, A/C, K, B/O, X, Y, F, and FIIA were identified for each isolate by PCR amplification [20] . Full-length bla_CTX-M-3-like was sequenced and aligned to the NCBI database. MLST was performed using previously described methods [21] . After PCR amplification and sequencing, MLST ST types were determined at http://mlst.ucc.ie/.

Sodium azide-susceptible clinical recurrent E. coli isolates and sodium azide-resistant E. coli strain J53 were used as donors and recipient, respectively, for conjugation experiments. After the co-culture of recipient and donor strains on filter paper on MHA agar overnight, the mixture was plated onto MacConkey agar with the appropriate antibiotic and sodium azide. Plasmid profile of the transconjugants was analyzed using the Kado-Liu method. The replicons of the conjugation plasmid were examined as above [20] .

In general, all isolates were multidrug resistant to AMP, CRO and CTX and PIP with the least antimicrobial resistance for the urine group and the highest antimicrobial resistance for the IUC group (Table 2). The average resistance rate was 54% to TZP (range: 42.3% for the urine group to 66.7% for the wound group); 76% to CAZ (range: 65.4% in the urine group to 91.7% in the blood group); 60.9% to FEP (range: 57.7% for the urine group to 70% for the sputum group); less than 10% to carbapenems, including ETP, IPM and MEM, with the value observed highest in the urine group; near 90% to fluoroquinolones; over 95% to nalidixic acid; and near 90% to CIP, LVX, and MXF, particularly differing between the urine and IUC groups, with the lowest found in the blood and wound groups.

The bla genes differed in prevalence among the isolation source (Table 3). The most prevalent bla gene was bla_CTX-M (87.4%; 54.0% for the bla_CTX-M3-like gene and 33.3% for bla_CTX-M14-like), followed by bla_CMY-2 (80.5%),