Optimization of Pulsed-Field Gel Electrophoresis for Legionella pneumophila Subtyping

Bacterial strains and culture conditions.Thirty-two strains of L. pneumophila, isolated in China and including at least four serogroups, were used in this study (Table 1). All isolates were obtained from water systems in China and were stored in our lab. They were revived from lyophilized isolates and 0.7% semisolid culture medium. The bacteria were streaked onto buffered charcoal yeast extract (BCYE) agar plates, and typical colonies were picked up, identified by serotyping, and inoculated onto BCYE agar plates. The bacteria were grown at 35°C in 2.5% CO₂ for 48 h for preparation of PFGE plugs. A panel of 26 strains of various origins was selected first from the test strains to optimize the EPs of SfiI digestion and to evaluate the discriminatory power of the enzyme. Three strains were used in the pilot test to choose the enzyme. Eleven strains associated with five independent water systems were included to evaluate the concordance between PFGE and four other molecular typing methods and the epidemiologic concordance of PFGE. Another 11 strains were selected randomly to test the reproducibility of the protocol.

Salmonella serotype Braenderup H9812 was used as a DNA size marker, as recommended by PulseNet (18).

PFGE protocol.The PFGE protocol used was based on the PulseNet 1-day standardized PFGE protocol for Vibrio cholerae (10, 31). A cell suspension in a polystyrene tube (Falcon; 12 by 75 mm) was adjusted to an optical density of 3.8 to 4.0, using bioMérieux Densimat. L. pneumophila slices were digested with 50 U of SfiI per slice (Roche Diagnostics, Mannheim, Germany) or with a corresponding amount of other enzymes (New England Biolabs) for 4 h at 50°C or 37°C. Electrophoresis was performed with a CHEF-DRIII system (Bio-Rad Laboratories, Hercules, CA). Images were captured on a Gel Doc 2000 system (Bio-Rad) and converted to TIFF files for computer analysis. Plugs of strain H9812 were prepared and digested along with the test strains. Slices of H9812 were digested with 40 U/slice XbaI (TaKaRa Bio, Dalian, China). All electrophoresis steps were run with a voltage gradient of 6 V/cm, an included angle of 120°, and a linear ramp.

This protocol was similar to the improved PFGE protocol for L. pneumophila developed by Chang et al. (8).

Computer analysis of PFGE patterns.The PFGE patterns were analyzed using the BioNumerics software package (version 4.0; Applied Maths, Inc.). Similarity analysis was performed by calculating Dice coefficients (S_D) (12), with customized tolerance for each EP. S_D was calculated as follows: S_D = [2(n_xy)]/(n_x + n_y), where n_xy is the number of bands common to isolates x and y, n_x is the total number of bands for isolate x, and n_y is the total number of bands for isolate y. The tolerance was determined according to the value when all the H9812 patterns obtained with the same EP were defined to be indistinguishable. Clustering was created using the unweighted-pair group method using average linkages (UPGMA). Fragments smaller than 20.5 kbp were not analyzed.

Optimization of electrophoretic parameters for SfiI digestion.Twenty-six isolates were analyzed with SfiI digestion, using four EPs, which were named EP-a, EP-b, EP-c, and EP-d (Table 2). EP-a was the parameter used most in the literature (8), while EP-b, -c, and -d were selected from a pilot study. The Simpson diversity index (D value) (17) and similarity coefficients (31) were used to compare the discriminatory powers under each parameter.

The D value was determined by the equation D = 1 − {∑[n_j(n_j − 1)]}/[N (N − 1)], where n_j is the number of strains belonging to the jth type and N is the number of strains in the population. The similarity coefficients of every two PFGE patterns were compared. Two-tailed probability was calculated using the Friedman test by SPSS 11.5 for multigroup comparisons. The Friedman test is a nonparametric test for analyzing randomized complete block designs, testing the null hypothesis that the treatments have identical effects. If significance was achieved among groups, the Friedman test was performed for two-group comparisons, with an adjusted significance level of 0.007. EPs with higher discriminatory power can distinguish patterns better, thus yielding smaller similarity coefficients. Accordingly, the EP with minimal similarity coefficients was optimal for distinguishing strains.

The EP with high D values and minimal similarity coefficients was optimal for distinguishing strains and was considered the standard for evaluating the discriminatory power of another enzyme, selected later.

Enzyme selection.The preliminary enzymes tested were selected using DNASTAR 5.01 software (DNASTAR, Inc., WI), based on whole nucleotide sequences (GenBank accession no. NC002942 , NC006368, NC006369, and NC009494) published in GenBank. The primary enzymes were then selected. A pilot test using three strains was conducted for further evaluation, and the optimal enzyme was selected based on the distribution of the bands. This optimal enzyme was further evaluated for use in PFGE of L. pneumophila.

Evaluation of optimal enzyme, focusing on typeability, discriminatory power, reproducibility, epidemiologic concordance, and correction with other molecular typing methods.Discriminatory power was compared using the Simpson diversity index and similarity coefficients as described above. Typeability was calculated as the percentage of distinct bacterial strains which could be assigned a pattern (27). Reproducibility was evaluated by repeat analysis of 11 isolates. They were analyzed on three different CHEF-DRIII machines, and the patterns obtained on different runs were clustered. Epidemiologic concordance and correction with other molecular typing methods were evaluated by analyzing 11 strains associated with five independent water systems.

SBT, ribotyping, randomly amplified polymorphic DNA (RAPD) analysis, and multiple-locus variable-number-tandem-repeat analysis (MLVA).The SBT protocols used here were recommended by the European Working Group for Legionella Infections (version 4.2), using seven specific gene loci (flaA, pilE, asd, mip, mompS, proA, and neuA).

Ribotyping was performed as previously reported, using an automated RiboPrinter system (4) following the manufacturer's recommendations (Qualicon, Wilmington, DE). All consumables and equipment used in further analyses were supplied by the manufacturer. EcoRI (Qualicon) was used as the restriction enzyme.

RAPD analysis was performed with primer SK2 (5′-CGGCGGCGGCGG-3′). All PCR and electrophoresis conditions used were as previously reported (21).

For MLVA, nine loci (Lp1, Lp3, Lp13, Lp17, Lp19, Lp31, Lp33, Lp34, and Lp35) were selected from the MLVA genotyping databases for L. pneumophila (http://minisatellites.u-psud.fr/ ). The protocols used here are available online (http://mlva.u-psud.fr/Legionella2006/PROTOCOL.htm ).

Optimization of electrophoretic parameters of SfiI digestion.For every enzyme, an EP could be recommended by the CHEF Mapper equipment, based on the sizes of restriction fragments. The EP recommended for SfiI digestion of L. pneumophila was a switch time of 6.8 to 63.8 s for 20 h (EP-b). However, with this EP, the bands were not well distributed, and it was not sufficient to distinguish fragments of less than 200 kbp. We fine-tuned the EP to provide the best possible resolution, and two more EPs were obtained (EP-c and EP-d). These three EPs were selected for comparison with EP-a, which was the parameter used most in the literature (Table 2).

For the 26 L. pneumophila strains, the slight difference between the patterns given with the four EPs emerged because two strains (Hu3 and JX3) were defined as different with EP-b but indistinguishable with EP-a, -c, and -d (Fig. 1). The differences between patterns of strains Hu3 and JX3 were so small that they could not be distinguished under the tolerance values used by EP-a, -c, and -d, but they could be distinguished by visual observation using the BioNumerics software package. The four EPs gave the same D value, 100%, based on the types obtained by visual observation. At the 100% similarity breakpoint given by the software, the D value was 99.69% under EP-a, -c, and -d and 100% under EP-b.

• Open in new tab
• Download powerpoint

FIG. 1.

Clustering results of patterns obtained using four EPs with SfiI digestion. Charts are shown for 26 L. pneumophila strains.

To compare the similarity coefficients, multigroup comparisons were made using SPSS 11.5. Nonparametric tests were performed because one-sample Kolmogorov-Smirnov tests demonstrated that the data were not normally distributed. The Friedman test showed that for the 26 test isolates (n = 351; chi-square = 14.286; df = 3), there were significant differences among the four groups (asymptotic significance, 0.003). Two-group comparisons were performed using the Friedman test, with an adjusted alpha value of 0.007 (Table 3). For the 26 isolates, similarity coefficients generated with EP-c and EP-d were significantly smaller than those obtained with EP-b. However, there was no significant difference between similarity coefficients generated by EP-a and those obtained with EP-b, -c, and -d, but EP-a exhibited smaller calculated similarity coefficients. Thus, EP-a was declared the optimal EP.

Selection of another enzyme.A theoretical enzyme selection using DNASTAR 5.01 software was based on four whole nucleotide sequences to include all satisfactory enzymes. Because the number of bands required for analysis should not be too large, many of the enzymes had too many bands. Therefore, beside SfiI, six enzymes (AscI, RsrII, SgrAI, SbfI, AsisI, and AvrII) were chosen as candidate enzymes for the pilot study. We obtained three images using these six enzymes and SfiI. The data showed that the image restriction cuts by AscI and SfiI were clear enough to meet our needs (Fig. 2).

• Open in new tab
• Download powerpoint

FIG. 2.

PFGE images of three L. pneumophila isolates restricted with AscI (lanes 2 to 4), RsrII (lanes 5 to 7), SgrAI (lanes 9 to 11), SbfI (lanes 13 to 15), SfiI (lanes 16 to 18), AsisI (lanes 20 to 22), and AvrII (lanes 24 to 26). The size standard (M) was loaded in lanes 1, 8, 12, 19, and 23.

Typeability, reproducibility, and discriminatory power.PFGE with AscI digestion had the ability to type all L. pneumophila strains and achieved the same level of typeability (100%) as that with SfiI digestion. Eleven randomly selected isolates were repeatedly analyzed three times with AscI digestion, and the patterns of the same isolate from different runs were defined to be indistinguishable, proving the good reproducibility of AscI digestion (see Fig. S1 in the supplemental material).

We analyzed the 26 strains with the EP of a switch time of 6.8 to 54.2 s for 19 h, which was recommended by the CHEF Mapper equipment and could provide reasonable patterns. PFGE with AscI digestion divided the 26 strains of L. pneumophila into 26 different pulsotypes and gave a D value of 100%, equivalent to the results of SfiI digestion (Fig. 3). Two-group comparisons using the Friedman test were also performed to compare the similarity coefficients generated by AscI and SfiI digestion. The results showed that for the 26 test isolates (n = 351; chi-square = 8.142), AscI digestion generated significantly smaller similarity coefficients than did SfiI digestion (asymptotic significance, 0.004). AscI PFGE had a higher discriminatory power than that of SfiI PFGE in this study.

• Open in new tab
• Download powerpoint

FIG. 3.

Clustering results of patterns obtained with AscI digestion. Charts are shown for 26 L. pneumophila strains.

Concordance with epidemiologic data and other molecular typing methods.We selected 11 isolates from five different water systems in different locations to evaluate the concordance of PFGE data with epidemiological data and data obtained by four other molecular typing methods. All of the isolates from the same water systems had identical SBT types, ribotypes, RAPD types, and MLVA types, except for the strains isolated from source 4 (Table 4). The two isolates from source 4 had different SBT types, with one base difference in the mip locus, and different MLVA types, with two loci showing differences. These two strains had the same PFGE type when digested with AscI and very similar patterns with SfiI digestion. The five isolates from sources 1 and 2 had the same SBT types, ribotypes, RAPD types, and MLVA types. However, PFGE with both AscI and SfiI digestion could distinguish all of the isolates from the different sources (Table 4).

The analysis of isolates from various sources demonstrated that PFGE with AscI digestion was able to group isolates from all different water systems into clusters representing different pulsotypes, except for one system. The three isolates from one water system showed similar patterns (one-fragment difference; S_D = 0.9474), with one isolate found to be different. This result of AscI digestion was nearly the same as that of SfiI digestion (Fig. 4). The difference was that the isolates within the other two water systems gave very similar, but not identical, patterns (one-fragment difference [S_D = 0.9600] and three-fragment difference [S_D = 0.8800]).

• Open in new tab
• Download powerpoint

FIG. 4.

Clustering results of patterns obtained with SfiI digestion and AscI digestion of 11 L. pneumophila strains from five different water systems. The strains in each frame were from the same water system.