Comparative analysis of amplified fragment length polymorphism and pulsed field gel electrophoresis in a hospital outbreak and subsequent endemicity of ampicillin-resistant Enterococcus faecium

Authors


*Corresponding author. Tel.: +47 55973075; Fax: +47 55975890, E-mail address: roland.jureen@med.uib.no

Abstract

Reliable molecular methods for determination of relatedness between bacterial isolates have become increasingly important to evaluate outbreaks and endemic situations with nosocomial pathogens. In the present study Simpson's index of diversity with calculated confidence intervals was used to compare amplified fragment length polymorphism (AFLP) and pulsed field gel electrophoresis (PFGE) analysis of a hospital outbreak of ampicillin-resistant Enterococcus faecium and subsequent endemicity. The outbreak, in a Norwegian tertiary hospital, of infections caused by these enterococci started in 1995 and increased in 1996 after which the situation turned endemic. The purpose of this study was to compare the two methods in this setting and to determine the length of time during an outbreak that these methods are sufficiently valid to be of value for hospital infection control efforts. One hundred and sixty clinical isolates from urine specimens collected during the period 1995–1999 were included. The findings indicate that PFGE and AFLP are equally discriminative and could in this setting be used for typing purposes over the whole 5-year period.

1Introduction

Enterococci are common causes of hospital-acquired (nosocomial) infections [1]. Urinary tract infections are most common although more serious infections also occur [2–4]. The emergence of high-level ampicillin resistance, high-level aminoglycoside resistance, and glycopeptide resistance has forced us to reconsider the importance of enterococci [5]. There have been several reports on the emergence of resistant enterococci in Scandinavia but so far glycopeptide resistance is uncommon [6–10]. Before 1995 acquired resistance in enterococci in Norway was rarely seen but in the spring of 1995 a tertiary hospital noticed an increasing number of ampicillin-resistant enterococcus isolates, almost exclusively Enterococcus faecium. The scenario of an outbreak at the hospital evolved [11]. The outbreak receded at the end of 1996 and entered an endemic phase, which still persists. When dealing with such an outbreak questions on typing techniques appear. Different DNA fragment techniques have frequently been used, but the interpretation of the data is often difficult due to experimental errors [12] and to spontaneous changes in the bacterial DNA with different modalities over time [13]. Criteria for interpreting pulsed field gel electrophoresis (PFGE) have earlier been given by Tenover et al. [14] and Struelens et al. [15].

Simpson's index of diversity was developed for the description of species diversity within an ecological habitat. Simpson's mathematical formulas enable us to calculate the probability that two unrelated strains sampled from the test population will be placed into different groups [16]. This index of diversity has been applied to compare typing methods in order to select the most discriminatory system [17]. Hunter and Gaston [18] suggest that discriminatory power can be defined mathematically as the probability that two strains chosen at random from a population of unrelated strains will be distinguished by that typing method.

In the setting of an ampicillin-resistant E. faecium outbreak with subsequent endemicity we compared amplified fragment length polymorphism (AFLP), a relatively new fingerprint-based typing technique successfully adopted for typing E. faecium[19,20], and PFGE, still regarded as the gold standard for molecular typing of E. faecium, with regard to their discriminative power using Simpson's index of diversity with the approximated 95% confidence interval as proposed by Grundmann et al. [21]. The aim was to compare the discriminatory power of PFGE and AFLP and to determine how well these techniques are able to identify the epidemic ampicillin-resistant clone during the course of a hospital outbreak and subsequent endemicity.

2Materials and methods

2.1Setting

Haukeland University Hospital is a fully specialized 1100-bed hospital serving a population of 1 million as a referral hospital and 300,000 as an acute-care hospital. The number of bacteriological specimens examined each year by the hospital laboratory exceeds 80,000, some 50,000 of which are from outpatients.

2.2Bacterial isolates

All ampicillin-resistant E. faecium (minimum inhibitory concentration (MIC) ≥32 mg l−1) from urine specimens from in-patients isolated at the hospital laboratory between 1995 and 1999 were collected and kept in the freezer. During this period a total of 167 isolates were reported. Three isolates were misclassified as Enterococcus faecalis and four were lost for further analysis. Thus 160 isolates were included in the study. The isolates were identified by standard biochemical tests [22] and species identification was verified by means of a polymerase chain reaction method described by Dutka-Malen et al. [23]. E. faecium ATCC 19434 and E. faecalis ATCC 29212 were used as quality control strains as well as an unrelated vancomycin-resistant E. faecium strain from The Netherlands.

2.3Susceptibility testing

Inocula were prepared from overnight cultures on Columbia II agar plates (Oxoid, Basingstoke, UK) supplemented with 5% sheep blood. The susceptibility of the isolates to ampicillin was examined by an agar diffusion method using paper disks and PDM II agar plates (AB Biodisk, Solna, Sweden), according to recommendations given by the Norwegian Working Group on Antibiotics [24]. The MIC breakpoint for resistance to ampicillin according to this recommendation is 32 mg l−1. Susceptibility testing to vancomycin and teicoplanin was performed with Etest (AB Biodisk). Resistance to glycopeptides were defined as MIC≥32 mg l−1 as recommended by NCCLS [25]. E. faecalis ATCC 29212 was used as quality control for all susceptibility testing.

2.4PFGE

PFGE was performed as described previously [26] and modified by Dahl [27]. In addition to the 160 isolates E. faecium ATCC 19434 was included. One isolate with the outbreak banding pattern (F1 E. faecium) and multiple size markers (Lambda Ladder PFG Marker, New England BioLabs, Beverly, MA, USA) were applied to all gels. The DNA banding patterns were analyzed with BioNumerics, version 2.5 (Applied Maths, Kortrijk, Belgium). The size markers were used to normalize the different gels and the F1 E. faecium isolate was used for setting parameters for optimization and band tolerance. The Dice coefficient of similarity was calculated, and the unweighted pair group method with arithmetic averages (UPGMA) was used for cluster analysis. The optimization was set at 1.3% and tolerance at 0.9%, so that all the banding patterns of the F1 isolate were identified with a similarity of 100%. For calculation of Simpson's index of diversity the interpretation was performed in two different ways:

  • 1Visual interpretation. The number of different banding patterns, with one or more band differences, was determined by visual inspection. For the PFGE analyzed visually we decided to have a stringent definition with no band differences between isolates in one PFGE type.
  • 2Computer-assisted interpretation. For the computerized interpretations of PFGE we chose a cut-off at 90% of the similarity values to indicate identical PFGE types. This is a stringent definition but gives room for minor technical errors and is also in line with earlier studies [28]. All isolates within a group were at least 90% similar to all other isolates in that group. The values of the similarity matrix generated by BioNumerics software were used for this.

2.5AFLP

DNA was isolated, as described previously [29], with the addition of a final ethanol precipitation step to further purify the DNA. AFLP was performed as described previously [20]. The amplification products were run on a DNA sequencer (ABI Prism 3700, PE Biosystems, Norwalk, CT, USA). For this 1 μl of reaction mixture was mixed with 7 μl distilled water and subsequently 1 μl was diluted with 9 μl formamide containing approx. 0.125 μl GeneScan-500 (GS-500 ROX) standard. The GeneScan collection software (PE Biosystems) was used to collect data during electrophoresis. After tracking and extraction of lanes, data were exported to the BioNumerics software for further analysis. Normalization was done by use of the reference positions of the internal DNA size marker GS-500 ROX. Fragments ranging in size from 50 to 500 nucleotides were used for comparison. The Pearson coefficient of similarity of AFLP curves was calculated. Cluster analysis was done by UPGMA. DNA extracted at different times from one isolate (F1 E. faecium) was analyzed on the electrophoresis runs and was used for optimization and tolerance settings. These parameters were set at optimization 0% and tolerance 0.1%, so that the curve patterns of this strain was identified with a similarity of 100%. For calculation of Simpson's index of diversity isolates with AFLP curves with a similarity coefficient of at least 90% in the similarity matrix generated by the BioNumerics software were considered to be of the same AFLP type as described earlier [30].

2.6Comparison of PFGE and AFLP

The concordance between PFGE typing and AFLP typing was determined using the BioNumerics software. Similarity values from the similarity matrices of the typing methods were compared and plotted in an x- and y-graph. Each dot in this graph represents corresponding similarity values for two strains by the typing methods given on the x- and y-axes, respectively. This graph thus gives an indication of the degree of concordance between the two techniques. Kendall's correlation coefficient [31,32] was also calculated. In addition we compared how many isolates that grouped together in one method were grouped in different groups with the other method. Groups containing two or more isolates were included. For this calculation, isolates with 90% or higher similarity as given by the dendrograms were grouped together for PFGE interpreted with computer assistance and AFLP.

2.7Simpson's index of diversity and confidence interval

The discriminatory power was defined mathematically as the probability that two strains chosen at random from the population would be distinguished by the typing method as previously described by Hunter and Gaston [18]. An approximate 95% confidence interval was calculated as proposed by Grundmann et al. [21]. A Simpson's index close to zero indicates that there is little diversity as shown by the typing method (index=0 indicates no diversity at all) whereas a Simpson's index approaching 1 indicates a high diversity as shown by the typing technique (index=1 indicates maximum diversity where no two isolates are similar).

3Results and discussion

Visual analysis of the PFGE banding pattern of the 160 isolates from the whole period revealed 46 types comprising 13–17 differently sized DNA fragments between 50 kb and 1000 kb. The number of types for the whole period is given in Table 1. The distribution of the four most common types for the whole period is given in Fig. 1A. Type 1 was the most prevalent with 45 isolates followed by type 2 with 34 isolates. There were two band differences between these two banding patterns. Isolates belonging to type 1 were found during the whole period and the first isolates carry this type. Isolates belonging to type 2 appeared in 1996 and were found from then on. Fifty-one additional isolates had one to three band differences, 17 isolates had four to six band differences and 13 isolates had seven or more band differences compared to the outbreak strain.

Table 1.  Different types and number of included isolates per type for the whole period with the different techniques
  1. n=160.

PFGE visualPFGE computer-assistedAFLP
TypeNumber of isolatesTypeNumber of isolatesTypeNumber of isolates
145125130
234224225
317317317
410411414
555–69511
647768
7–1128677
12–4619–1058–96
  11410–115
  12–14312–144
  15–21215–172
  22–36118–251
Figure 1.

Numbers of isolates in the four most common fingerprint types with the different techniques for the individual years and for the whole period. A: Each type consists of isolates with indistinguishable banding patterns. B: Each type consists of isolates with ≥90% similarity of the Dice coefficient. C: Each type consists of isolates with ≥90% similarity of the Pearson coefficient.

Computer-assisted interpretation of the PFGE revealed in total 36 different types among the 160 isolates. The two most common types comprised 25 isolates and 24 isolates respectively. The number of types for the whole period is given in Table 1 and the distribution of the four most common types for the whole period is given in Fig. 1B.

AFLP analysis discerned 25 different types among the 160 isolates. The number of types found in the whole period of the study is given in Table 1 and the distribution of the four most common types for the whole period is given in Fig. 1C.

The concordance of similarity values generated by PFGE and AFLP analysis of the 160 isolates, E. faecium ATCC 19434 and an unrelated Dutch E. faecium strain is shown in Fig. 2. The degree of concordance between PFGE and AFLP as determined by Kendall's τ correlation coefficient was calculated to be 46%. Comparison of group differentiation by PFGE interpreted visually, with computer assistance, and AFLP is shown in Table 2. Comparing the degree of strain differentiation in Table 2 it is clear that there are differences between the typing techniques. For instance when comparing AFLP with PFGE analyzed visually and PFGE interpreted with computer assistance the percentages of isolates that are identical by AFLP but different by the other techniques are 59% and 41%, respectively. However, when we compare Simpson's index of diversity of the different typing techniques hardly any difference is seen. The calculated Simpson's indices with approximate 95% confidence interval for each year and for the whole period are shown in Fig. 3. One should be aware that the sizes of the groups and the number of isolates in the groups in addition to the number of groups affect these values. In Fig. 3 there is only a very small difference between the different typing schemes when comparing Simpson's indices for the whole period. Thus when including all factors mentioned above we are able to conclude that all typing schemes are practically equally discriminative.

Figure 2.

Concordance of stratification by PFGE analyzed with computer assistance and AFLP analysis for 160 clinical isolates and two control E. faecium strains. Each dot represents corresponding similarity values for two isolates obtained by the typing methods given on the x- and y-axes. The degree of concordance, as determined by Kendall's τ correlation coefficient, was 46%.

Table 2.  Comparison of strain differentiation by visually interpreted PFGE, computer-assisted interpreted PFGE and AFLP for the period 1995–1999
  1. aPercentages of isolates that were indistinguishable by one typing method but different by another method. For example 1% and 39% of isolates with identical PFGE types interpreted visually had a different PFGE type when interpreted with computer assistance and AFLP type, respectively. Groups containing two or more isolates were included.

Type% of isolates analyzed bya
 PFGE visuallyPFGE computer-assistedAFLP
PFGE visually03259
PFGE computer-assisted1041
AFLP39350
Figure 3.

Simpson's index of diversity for the individual years and for the whole period. Ninety-five percent confidence intervals according to Grundmann et al. are marked.

Isolates recovered from urinary tract infections were included since this is the most common nosocomial infection with enterococci. All the included isolates were resistant to ampicillin and susceptible to vancomycin. They were all typeable by both typing methods and the reproducibility of both techniques has been shown to be excellent [19,33,34]. To reduce differences in subjective evaluation of banding patterns in PFGE we wanted, in addition to visual analysis, to evaluate computerized analysis using 90% similarity of the Dice coefficient as a level of PFGE type identity. This degree of similarity also allows for minor technical errors that occur infrequently [12]. The AFLP could not be visually compared due to a large number of curve peaks, making it too complicated for a manual approach.

Simpson's index of diversity showed no major differences between the different typing schemes. This index was originally used to describe the diversity of unrelated species. This is important to keep in mind when this index is used to evaluate typing techniques. The low indices of diversity for the methods in 1995 do thus not necessarily indicate poorly differentiating techniques, but more likely indicate that there was little polymorphism as a consequence of a newly established clone of bacteria with few other competing clones in the sampled patient population. As Grundmann et al. [21] point out, the index of diversity will increase with increasing sample size, for highly diverse populations or typing techniques with extreme ability to discriminate genotypes. It is thus also possible that the low sample size for 1995 (n=16) could explain the low index of diversity in that year. The indices of diversity increased from 1995 to 1998 for all methods and for the whole 5-year period they are about 90% as indicated in Fig. 3, which may suggest genetic unrelatedness of the isolates. However, to highlight differences between typing techniques when evaluating which is the most discriminative it is important to have stringent criteria for type assignment when calculating this index. As Table 1 shows, with all methods most isolates belong to only a few types. Therefore a high index of diversity for the typing methods in this study does not contradict the notion that most isolates seem genetically related.

The nature and frequency of occurrence of genetic events that influence banding patterns is not known. It is therefore difficult to determine with certainty which of the banding patterns is a product of such genetic events and which is due to new bacterial strains infiltrating the study population. This type of information would be important since it would affect the way we interpret the findings. In a situation were only one clone circulates without influx of new strains, a large number of different genotypes resulting in a high index of diversity would indicate a large number of spontaneous mutations or horizontal gene transfer in the study population. In reverse, a low index of diversity would indicate a low rate of spontaneous mutations or horizontal gene transfer. Serial culturing of isolates followed by genotyping could give an indication of the genomic stability of these ampicillin-resistant E. faecium.

The influence of transferred antibiotic resistance determinants on the outcome of PFGE, AFLP and random amplification of polymorphic DNA typing of E. faecium has recently been reported by Werner et al. [35]. They found that almost half of their transconjugants had one to three band differences compared to the recipient pattern in PFGE and almost half had identical patterns, whereas the result of AFLP depended on how the data were interpreted. Since most isolates in our study seem related (one to three band differences in PFGE) additional information by performing plasmid fingerprinting could have resolved the question whether these band differences could be caused by transferred plasmids.

Morrison et al. [13] demonstrated that isolates belonging to the same vancomycin-resistant E. faecium strain by PFGE (which gave 19–23 bands, run under different conditions, as opposed to 13–17 in this study) could differ by up to seven bands, indicating a high degree of polymorphism. D'Agata et al. [36] compared PFGE and AFLP and reported that PFGE is more discriminatory for vancomycin-resistant E. faecium, however only 22 isolates were included. Antonishyn et al. [19] also compared PFGE and AFLP in a study that included 30 vancomycin-resistant E. faecium isolates and showed similar discriminatory power of the two techniques.

The PFGE as well as the AFLP outbreak pattern was found in isolates throughout the study (1995–1999) while a large number of isolates with a PFGE pattern differing in only two bands emerged in 1996 as indicated in Fig. 1. Overall the banding patterns are very similar with only 17 isolates that had four to six PFGE band differences compared to the outbreak strain and 13 isolates that had seven or more band differences. The E. faecium ATCC 19434 strain and the unrelated Dutch E. faecium strain differed by more than seven bands from the outbreak strain. This shows that both PFGE and AFLP are equally well able to identify the outbreak clone throughout the study period and confirms that most isolates are epidemiologically linked with no or only little influx of unrelated E. faecium isolates. Hence, minor differences observed in PFGE and AFLP patterns are most likely due to genetic alterations (mutations or recombinations or both) of the outbreak clone.

In conclusion, we have demonstrated that PFGE and AFLP typing can be used for identifying circulating clones during a 5-year period, in both an epidemic and an endemic setting. Computer-assisted interpretation of PFGE, using 90% or more similarity of the Dice coefficient as a criterion for similarity, was equally discriminative as visually interpreted PFGE during the individual years and for the whole period. This indicates that the time-consuming visual interpretation of PFGE in certain settings can be replaced by the faster computer-assisted interpretation. AFLP, first described by Vos and collaborators [34], was equally discriminative as PFGE. However, AFLP is much less labor-intensive than PFGE which makes it possible to test large numbers of isolates with an acceptable workload. Furthermore, AFLP typing, in contrast to PFGE, permits the study of genetic relationships among dissimilar, non-epidemiologically related strains [20].

Acknowledgements

This work was funded by the University of Bergen, Bergen, Norway.

Ancillary