Effects of Sulfamethizole and Amdinocillin against Escherichia coli Strains (with Various Susceptibilities) in an Ascending Urinary Tract Infection Mouse Model

Abstract

Resistance to antibiotics used for the treatment of urinary tract infections (UTIs) is increasing worldwide. The impact of in vitro resistance on clinical outcome in UTIs requires further study, since most studies of both humans and animals have evaluated only the efficacy of antibiotics toward bacteria susceptible in vitro. We were interested in evaluating the relationship between the in vitro antibacterial effect and the in vivo efficacy after antibiotic treatment. We simulated a natural ascending UTI by use of the ascending UTI mouse model and used Escherichia coli strains with various susceptibilities to amdinocillin (mecillinam) and sulfamethizole. Mice were treated for 3 days with antibiotic doses approximating human urinary tract concentrations after a standard oral dose. For a susceptible strain (MIC, 0.5 μg/ml) and a resistant strain (MIC, 128 μg/ml), respectively, there were significant reductions in bacterial counts in the urine, bladder, and kidneys after treatment with amdinocillin, whereas for a strain for which the MIC was 16 μg/ml, there was a significant reduction in bacterial counts in the kidneys only (P < 0.05). Treatment with sulfamethizole resulted in a significant reduction in bacterial counts in all samples from a susceptible strain (MIC, 128 μg/ml) and a resistant strain (MIC, 512 μg/ml). Infection with a sulII gene-positive strain (MIC, >2,048 μg/ml) could not be treated with sulfamethizole, as no effect could be demonstrated in the urine, bladder, or kidneys. For amdinocillin, there was no clear-cut relationship between the in vitro susceptibility and the in vivo outcome, while for sulfamethizole, we found a relationship between the MIC for the strain and the effect in the urinary tract.

The resistance to antibiotics used for the treatment of urinary tract infections (UTIs) is increasing worldwide (28). The common opinion that antibiotics used against UTIs achieve such high and long-lasting concentrations in the urine that practically all bacteria can be eradicated might not be true. The impact of in vitro resistance on clinical outcome in UTIs requires further study, since most studies with both humans and animals have evaluated only the efficacy of antibiotics toward bacteria susceptible in vitro.

For β-lactam antibiotics, it has been determined that in vivo efficacy is correlated with the length of time for which the serum antibiotic concentration exceeds the MIC (9, 32). It has been shown that a similar connection exists in the urinary tract, where the length of time for which the urinary tract antibiotic concentration exceeds the MIC determines the efficacy of β-lactam antibiotics in the urine and kidneys (H. Hvidberg, S. N. Rasmussen, and N. Frimodt-Møller, submitted for publication).

In Denmark, the majority of uncomplicated UTIs are treated with sulfamethizole, although the resistance of Escherichia coli at present is 22% among this group of patients (18). It has therefore been widely debated whether the empirical treatment should be changed to amdinocillin, to which resistance is negligible (18). This is the background for the choice of evaluating these two antibiotics in the animal model.

To our knowledge, we are the first investigators to use the ascending UTI mouse model to evaluate the relationship between the in vitro antibacterial effect and the in vivo efficacy.

(This work was presented in part at the 11th European Congress of Clinical Microbiology and Infectious Diseases (Istanbul, Turkey, 2001) and at the 41st Interscience Conference on Antimicrobial Agents and Chemotherapy (Chicago, Ill., 2001).)

MATERIALS AND METHODS

Bacteria.

Twenty-one type 1-fimbriated E. coli strains were tested for their pathogenicity in the animal model as described below (five mice per strain). All strains were obtained from patients with UTIs or bacteremia originating from the urinary tract and expressed type 1 fimbriae, as determined by mannose-sensitive hemagglutination with 3% guinea pig erythrocyte and 2% d-mannose solutions (14).

MICs.

The MICs were determined by the agar dilution method as described by the NCCLS (20).

Inoculum.

The strains were cultured overnight on 5% blood agar plates (Statens Serum Institut). Morphologically similar colonies obtained from the cultures were suspended in 0.5 M phosphate-buffered saline (PBS; pH 7.4). The suspension was adjusted to a concentration of approximately 10⁹ to 10¹⁰ CFU/ml by using a colorimeter (Sherwood colorimeter 254). The actual concentration was confirmed by determination of viable counts as described below.

Animals.

Outbred female albino mice (Ssc-CF1 mice; weight, 30 ± 2g [mean and standard deviation]) were used. The mice were housed six to a cage, and they were allowed a minimum of 24 h of acclimatization prior to inoculation. During the studies, the animals were allowed free access to chow. Three days before inoculation and during the study period, drinking water was replaced with a 5% glucose solution, as it has been shown to greatly increase the numbers of bacteria in the urine and kidneys (17). The glucose was administered in autoclaved bottles and changed every day to avoid bacterial growth.

Mouse model of UTI.

The mice were anesthetized either by intraperitoneal administration of 0.1 ml of a 1:3 mixture of fentanyl citrate (0.315 mg/ml)-fluanisone (10 mg/ml) (Hypnorm) and diazepam (5 mg/ml) (Stesolid) or by intravenous administration of 0.08 ml of a 1:2:1 mixture of fentanyl citrate-fluanisone (Hypnorm), sterile water, and midazolam hydrochloride (5 mg/ml) (Dormicum). Anesthetized mice were inoculated transurethrally with the bacterial suspension by use of a plastic catheter. The catheter was made from approximately 2 cm of soft tubing (INTRAMEDIC polyethylene tubing; inner diameter, 0.28 mm; outer diameter, 0.61 mm; Becton Dickinson) placed on a hypodermic needle (Microlance 3; 30.5 gauge; Becton Dickinson). To ensure sterility, on the day before use, the catheter on the needle was placed for 20 min in 99.9% ethanol, which was then allowed to evaporate.

The catheter was gently inserted through the urethra until it reached the top of the bladder, and 0.05 ml of the bacterial suspension was injected slowly into the bladder to avoid vesicoureteral reflux (VUR). The catheter was removed immediately after this procedure, leaving the bladder with 5 × 10⁷ to 5 × 10⁸ CFU.

Although some investigators have shown that a 50-μl inoculum produces VUR in about 30% of kidneys (10, 16) and others have shown that 10- and 25-μl inocula should avoid VUR (11, 15), these experiments were performed with other strains and sizes of mice. In our experiments, an inoculum of 50 μl did not produce VUR in the mice used.

Samples for determination of bacterial counts.

Urine from each mouse was collected in Eppendorf tubes by gentle compression of the abdomen, and then the mouse was killed by cervical dislocation. The organs were removed aseptically; the bladder was cut away near the urethra, and the kidneys were removed by blunt dissection to avoid bleeding. The urine and organs were processed immediately after sampling. The bladder and each kidney separately were homogenized in 0.5 ml of PBS by using a sterile grinder (RW 16 Basic; IKA Labortechnik), and viable counts were determined by using lactose-bromthymol blue agar plates (Statens Serum Institut) by either the 20-μl spot method or the spread plate technique (Eddy Jet and Countermat Flash; IUL Instruments). Each kidney was individually processed, and the variability between kidneys from a given mouse was low: for 85% of the mice, the kidneys were either both infected or both not infected; for the remaining 15% of the mice, only one kidney was infected, with a right kidney/left kidney ratio of ≈1.

The CFU per milliliter of urine, per bladder, or per kidney were determined after 18 to 24 h of incubation at 35°C. We report the number of bacteria per organ and not per weight of the organ, since a pilot study showed that the results were nearly identical. The detection limits were set to 50 CFU/ml and 50 CFU/organ.

Antibiotics, dosages, and sampling.

The antibiotics used were amdinocillin (Selexid; Leo, Copenhagen, Denmark) dissolved in sterile water and sulfamethizole (Sigma, St. Louis, Mo.); the latter was dissolved in 1 M NaOH, hot PBS (pH 7.4) was added, and 1 M HCl was added until the solution reached pH 7.4. The amdinocillin solution was prepared immediately before use, whereas the sulfamethizole solution was stored at −20°C and kept at 4°C for 1 day before treatment (stabile). Infected mice were injected subcutaneously (s.c.) in the neck with 0.25 ml of antibiotic solution by use of hypodermic needles (22 gauge; 0.25 in.). Amdinocillin was given at 0.5 mg twice 6 h apart for 3 days, and sulfamethizole was given at 1.25 mg twice 6 h apart for 3 days, starting on the day after bacterial inoculation. These doses were intended to mimic the concentrations in human urine after the oral administration of 400 mg of pivmecillinam and 1 g of sulfamethizole, respectively (M. B. Kerrn, N. Frimodt-Møller, and F. Espersen, in press). Groups of three mice were killed 0, 10, 20, 30, 40, and 50 min after injection; for sulfamethizole, additional serum samples were obtained after 90 and 120 min. Samples of urine were collected at 1.5, 2, 4, and 6 h after injection for both antibiotics.

Determination of antibiotic concentrations.

Drug concentrations in urine and serum were determined for amdinocillin by the bioassay method and for sulfamethizole spectrophotometrically. Blood samples were collected by ocular cutdown after the mice were anaesthetized with CO₂. The blood was centrifuged at 3,000 × g for 7 min, and the supernatant (serum) was transferred to an Eppendorf tube.

The bacterium used in the bioassay was amdinocillin-sensitive E. coli strain Leo HA2 grown on Mueller-Hinton BBL-II agar. Each sample was measured in duplicate and matched to a standard curve of five twofold dilutions. For serum, the dilutions started at 50 μg/ml, and for urine, they started at 2,500 μg/ml; the dilutions were prepared in mouse serum and PBS, respectively. The coefficients of variation were 7.9 to 9.2% in PBS and 1.6 to 3.1% in serum. The detection limit was determined to be 0.7 μg/ml in several pilot studies.

The spectrophotometric assay (Ultrospec 2000; Pharmacia Biotech) for sulfamethizole was performed as described by Bratton and Marshall (5) with standard curves of four fivefold dilutions starting at 500 μg/ml for both serum and urine (R², 0.998 to 1). About 10% of the sulfamethizole dose is excreted in acetylated form (3) and therefore is inactive, but the method of Bratton and Marshall does not measure this fraction. Some urine samples had to be diluted, as they attained concentrations higher than 500 μg/ml. The standards were diluted in mouse serum and mouse urine, respectively. The detection limit was 10 μg/ml.

Statistical methods.

Linear regression analysis of the concentration curves was performed with Microsoft Excel 2000. The area under the curve (AUC) from 0 to 360 min (noncompartmental) was determined by use of GraphPad Prism 3.02 (GraphPad Software 2000). A comparison of the numbers of CFU found in the specimens was performed by the Mann-Whitney U test (GraphPad Prism 3.02), and the P value was considered significant at <0.05. The number of animals needed in each group was calculated by using GraphPad StatMate 1.0 for comparing two proportions with the chi-square test (one tailed). The type 1 error was set to 5%, and the type 2 error was set to 20%. If the expected success rate in control groups was set to a maximum of 50%, then 12 mice in each group could reveal approximately a 45% difference between the treated group and the control group.

RESULTS

Selection of E. coli strains.

A range of E. coli strains with different susceptibilities to amdinocillin and sulfamethizole were tested in the model. On the day after inoculation, five mice infected with each strain were examined for the presence of bacteria in the urine, bladder, and kidneys. The results are shown in Table 1. On the basis of infection capability and MICs, the following strains were chosen for amdinocillin and sulfamethizole treatment: for amdinocillin—strain A, 21623884-114 (UTI; MIC, 0.5 μg/ml); B, 21773360-98 (UTI; MIC, 16 μg/ml); and C, Eco 518 (UTI; MIC, 128 μg/ml); and for sulfamethizole—strain D, KMA-26883 (bacteremia; MIC, 128 μg/ml); E, UVI-203 (UTI; MIC, 512 μg/ml); and F, KMA-28523 (sulII gene positive; bacteremia; MIC, >2,048 μg/ml).

TABLE 1.

Pathogenicity of 21 type 1-fimbriated E. coli strains in mouse urinary tract 1 day after inoculation^a

Antibiotic	Strain	Origin	MIC (μg/ml)	Median no. of CFU per:
				ml of urine	Bladder	Kidney
Amdinocillin	UVI-63	Urine	0.25	1.3 × 104	1.8 × 104	2.6 × 102
	21623884-114	Urine	0.5	1.5 × 106	1.7 × 105	5.0 × 104
	UVI-105	Urine	1	3.5 × 105	5.0 × 103	5.0 × 101
	UVI-139	Urine	2	6.4 × 102	6.0 × 102	0
	21657801-43	Urine	4	1.5 × 104	1.3 × 103	1.1 × 102
	UVI-227	Urine	8	9.8 × 101	0	0
	21773360-98	Urine	16	5.8 × 107	2.5 × 106	1.6 × 104
	UVI-86	Urine	32	8.8 × 106	4.0 × 106	3.0 × 104
	77504954-48	Urine	64	6.3 × 106	2.4 × 106	6.2 × 103
	Eco 518	Urine	128	2.2 × 108	2.2 × 107	3.7 × 103
	GAIL I01137	Urine	>256	1.5 × 104	5.1 × 103	0
Sulfamethizole	KMA-4845	Bloodb	64	7.2 × 105	7.0 × 101	2.9 × 102
	KMA-21476	Bloodb	64	1.2 × 107	3.5 × 105	2.2 × 102
	KMA-26883	Bloodb	128	8.3 × 106	8.8 × 105	1.4 × 103
	21804403-132	Urine	512	5.0 × 102	0	0
	UVI-203	Urine	512	6.8 × 107	4.3 × 106	2.4 × 104
	UVI-202	Urine	1,024c	3.8 × 102	0	0
	21398355-97	Urine	1,024	0	0	0
	UVI-120	Urine	>2,048	1.7 × 105	6.3 × 104	2.6 × 102
	21426064-1	Urine	>2,048c	1.2 × 106	3.6 × 105	3.0 × 103
	KMA-28523	Urine	>2,048d	2.3 × 106	5.0 × 106	2.9 × 104

^aFive mice were tested per strain.

^bIsolated from a patient with bacteremia caused by a UTI.

^csulI and sulII gene positive.

^dsulII gene positive.

Concentrations of amdinocillin and sulfamethizole.

The curves for log concentration versus time are shown in Fig. 1. Immediately after s.c. administration of 0.5 mg of amdinocillin, the peak concentration in serum was measured at 18 μg/ml. In urine, the peak concentration of 3,415 μg/ml was obtained after 20 min. The amdinocillin concentration in serum declined to the detection limit of 0.7 μg/ml after 50 min (half-life [t_1/2], 8 min), whereas the concentration in urine was detectable after 360 min (t_1/2, 33 min). For sulfamethizole, the s.c. dose of 1.25 mg resulted in a peak concentration in serum of 91 μg/ml and a peak concentration in urine after 20 min of 3,276 μg/ml. Elimination of the sulfamethizole dose was observed after 90 min in serum (t_1/2, 26 min) and after 240 min in urine (t_1/2, 29 min).

FIG. 1.

Mean concentrations of amdinocillin (A) and sulfamethizole (B) in serum and urine. Error bars indicate standard deviations.

Table 2 shows the AUC related to the MIC (AUC/MIC) and the time above the MIC (T > MIC) in serum and urine for selected strains after one dose of amdinocillin or sulfamethizole. The T > MIC was calculated from the estimated regression curves used to determine the t_1/2 values.

TABLE 2.

AUC/MIC and T > MIC in serum and urine after one dose of drug

Drug	Strain (MIC, μg/ml)	AUC/MIC (min)		T > MIC (min)
		Serum	Urine	Serum	Urine
Amdinocillin	A (0.5)	736	234,700	46	400
	B (16)	1	7,074	12	235
	C (128)	0	783	0	136
Sulfamethizole	D (128)	0	741	0	87
	E (512)	0	121	0	55
	F (>2,048)	0	0	0	0

Effects in urine, bladder, and kidneys after 3 days of treatment with amdinocillin or sulfamethizole.

The antibiotic treatment was initiated on the day after inoculation. Results from two or three experiments with the same strain were grouped, since there was no significant difference between the observations. The bacterial counts in the urine (CFU per milliliter) and in the bladder and kidneys (CFU per organ) after 3 days of treatment with either amdinocillin or sulfamethizole are shown in Fig. 2.

FIG. 2.

Bacterial counts. (A) Bacterial counts from mice treated for 3 days with amdinocillin. The small horizontal line represents the median bacterial count. The large horizontal line indicates the detection limit. An asterisk indicates that there is a significant difference between the control group (closed squares) and the treated group (open circles). (B) Bacterial counts from mice treated for 3 days with sulfamethizole. Symbols are as defined for panel A. ns, not significant.

For strain A (MIC, 0.5 μg/ml) treated with amdinocillin, there was a significant reduction in bacterial counts in the urine, bladder, and kidneys. Only in the kidneys was there a significant effect for strain B (MIC, 16 μg/ml). Although there was a reduction in CFU in the urine and bladder, it was not significant. The counts of resistant strain C (MIC, 128 μg/ml) were significantly reduced in the urine, bladder, and kidneys after treatment.

Treatment with sulfamethizole for strain D (MIC, 128 μg/ml) and strain E (MIC, 512 μg/ml) resulted in a significant reduction in bacterial counts in all samples from treated mice compared to control mice. The effects in the urine and bladder were more pronounced for strain D than for strain E. sulII gene-positive strain F (MIC, >2,048 μg/ml) could not be treated effectively with sulfamethizole; no effects could be demonstrated in the urine, bladder, or kidneys.

DISCUSSION

To our knowledge, we are the first investigators to use the ascending UTI mouse model for the evaluation of urinary tract-specific breakpoints. We were interested in simulating a naturally ascending UTI by using E. coli strains with type 1 fimbriae, which have an affinity for the bladder (25). The duration of the antibiotic treatment was chosen to resemble a standard treatment for an uncomplicated human UTI.

The variability in this animal model is considerable, and we had to use more mice than we would have for other models. The model is rather complex, and many factors can vary: (i) the time from inoculation until the mouse urinates, (ii) establishment of the infection in each mouse, (iii) expression of type 1 fimbriae when the infection is established (29), and (iv) little knowledge of how type 1 fimbriae are affected by antibiotics (30).

Only limited data on the urine pharmacokinetics of amdinocillin and sulfamethizole in humans are available. After oral treatment with 400 mg of pivmecillinam, 6-h samples of urine reached a mean concentration of 187 μg/ml (range, 37 to 787 μg/ml) (34). A study conducted with a similar dose resulted in an average urine concentration of 296 μg/ml (range, 84 to 1,324 μg/ml) 0 to 3 h after ingestion of pivmecillinam (Kerrn et al., in press). As can be seen from Fig. 1 and Table 2, the maximum concentration in urine is about 10 times higher and the T > MIC in urine is about 4 times shorter in mice than in humans. To solve this problem, we would have had to give more frequent doses, as the metabolism in mice is faster than that in humans.

It was reported that after ingestion of 1.0 g of sulfamethizole, the concentration in pooled urine samples after 0 to 3 h averaged 934 μg/ml (Kerrn et al., in press); the t_1/2 of sulfamethizole in urine is about 1.6 h (33). Which pharmacokinetic parameter(s) correlates with the efficacy of sulfonamides is unknown. We chose to treat the mice with a sulfamethizole dose that resulted in a maximum concentration in urine corresponding to human data, but since we did not alter the excretion of the antibiotic (e.g., probenecid, uranyl nitrate) or use repeated injections (refracted doses), the T > MIC was not as extended as would be expected in humans.

Thus, the dosing of both antibiotics used here was an approximation of the actual concentration profile in humans. Both antibiotics were evaluated in vivo against a susceptible strain, a strain for which the MIC is close to the resistance breakpoint, and a resistant strain (according to current NCCLS breakpoints) (21).

Basic studies have shown that the antimicrobial effect in vitro is not exclusively a predictor for the effect in vivo (26), and approximately 40% of new compounds lack this correlation (6). Although we studied old, well-known antibiotics, the pharmacodynamics of these have been poorly described.

Several other studies with animal infection models showed that for β-lactam antibiotics, if the time for which levels in serum exceeded the MIC was greater than 40%, the survival or bacteriologic cure rate was about 85 to 100% (7). The significant effects in the urine and bladder of 3 days of amdinocillin treatment on strain A (T > MIC, 56%) and the lack of an effect on strain B (T > MIC, 16%) indicated that our assumption about the relevance of the time for which the concentration in urine exceeds the MIC is relevant. The problem is in explaining the significant effects on strain C (T > MIC, 9%). It is well known that the effect of the inoculum size on the MIC is tremendous, since a 100-fold increase in the inoculum has been shown to increase the MIC by a factor of approximately 65 (22). The significance of this in vitro phenomenon was tested in vivo in an intraperitoneal infection mouse model by Isenberg et al., who found that about 44 to 60% of the infecting strains of Enterobacteriaceae that they tested and that were resistant to amdinocillin in vitro were susceptible to amdinocillin in vivo (13). The amdinocillin MIC for E. coli strain C (128 μg/ml, tested by agar dilution) did show marked inoculum dependence in vitro, as an E test with 10⁶ and 10⁸ CFU/ml resulted in MICs of 8 and >256 μg/ml, respectively.

We performed population analysis of strains A, B, and C and found that all contained a subpopulation that was inhibited at a concentration much higher than the MIC. The two strains for which MICs were 16 and 128 μg/ml showed the same inhibition patterns in the population analysis (data not shown). In vitro, we measured the MIC on a static inoculum, while in vivo, the inoculum fluctuates over time during treatment. In addition, given the availability of an intact immune system, a strain that contains a minor subpopulation of resistant bacteria may be effectively treated as the inoculum decreases over time. On the other hand, amdinocillin is difficult to test in vitro, so that the high amdinocillin MIC that we measured in vitro may be doubtful and may not be a good predictor for the effect that we measured in vivo.

Despite the relatively short exposure to an active sulfamethizole concentration, CFU reduction was observed in all samples for strain D and strain E (although less pronounced). There was no effect on resistant strain F.

Analysis of the distribution of amdinocillin in human kidneys revealed a 1.4-fold higher concentration in renal tissue than in serum (24), and it was shown that the concentration of sulfamethoxazole measured in the kidneys of a rat was lower than the concentration measured in serum (31). Still, it is difficult to explain the significant effects observed in the kidneys for strains A, B, C, D, and E. The concentrations of the antibiotics in serum do not exceed the MICs for strains C, D, and E; thus, perhaps the CFU reduction reflects the localization of the bacteria in the kidneys. The antibiotic treatment was started in the early phase of the infection; in this phase, most bacteria could be located in the renal pelvis or distal tubules, where the antibiotic concentration potentially corresponds to the higher concentration in urine. Investigators have found that bacteria in chronic pyelonephritis are limited to the medulla of the kidneys, and it seems virtually certain that the high concentration of antibiotic present in the collecting ducts is transferred to the interstitial tissue water of the medulla (27). The localization of bacteria in the kidneys could be further studied by investigation of, e.g., radioactively labeled bacteria or in situ hybridization. Also, the animals were treated just 1 day after inoculation, whereas in a study with cefuroxime and gentamicin, treatment was initiated 3 days after inoculation; in that study it was concluded that concentrations in the kidneys resembling those in serum were necessary for effects in the kidneys (12).

For sulfamethizole, we found a relationship between the MIC for the strain and the effect in the urinary tract, as measured by CFU reduction. This observation was also made in other animal models and with other antibiotics. In the murine thigh infection model (Enterobacteriaceae), the in vivo activities of tobramycin, pefloxacin, ceftazidime, and imipenem correlated with the in vitro susceptibility test results (8). The same model was used for the evaluation of the breakpoints for amoxicillin and amoxacillin-clavulanate against Streptococcus pneumoniae (2) and cefprozil against S. pneumoniae (23), where it was concluded that the extent of organism killing was dependent on the MICs. Clinical data have shown that standard treatment with sulfamethizole of uncomplicated UTIs caused by resistant strains results in cure rates of 58 to 74% (1, 4, 19). Whether the reported strains were actually resistant and to what degree they were resistant are in question, since none of the studies reported such data or stated the MIC breakpoint used.

The MIC of sulfamethizole at which 50% of the native sul gene-negative E. coli population is inhibited is 128 μg/ml (range, 8 to 512 μg/ml) (18). On the basis of the finding that a sul gene-positive strain is untreatable, the current MIC breakpoints for sulfonamides (susceptible, ≤256 μg/ml; resistant, ≥512 μg/ml) seem reasonable (21). For amdinocillin, there is no clear-cut relationship between the in vitro results and the in vivo outcomes; therefore, it is impossible to evaluate the NCCLS breakpoint of ≥32 μg/ml for resistance (21). Testing of further strains is required.

In conclusion, the ascending UTI mouse model can be used to study antibiotic efficacy. The urinary tract concentration of sulfamethizole is not high enough to treat an infection with a resistant sul gene-positive E. coli strain, and so the current MIC breakpoint is acceptable. Amdinocillin is a difficult antibiotic to study, since the in vivo outcome obtained with the presumed in vitro-resistant strain is not what would be expected.