Exploration of biases that affect the interpretation of restriction fragment patterns produced by pulsed-field gel electrophoresis

Abstract

Pulsed-field gel electrophoresis (PFGE) has been used extensively in epidemiological investigations of bacteria, especially during food-borne outbreaks or nosocomial infections. The relationship between similarities in PFGE patterns and true genetic relatedness is poorly understood. In this study, computer-simulated populations of Escherichia coli isolates were created by mutating the sequence of E. coli K-12 strain MG1655. The simulated populations of isolates were then digested, again through simulation, with different restriction enzymes and were analyzed for their relatedness by different techniques. Errors associated with band determination and band matching were incorporated into the analyses, as both of these error types have been shown to affect PFGE interpretations. These errors increased the apparent similarities of the isolates. The use of multiple enzymes improved the fidelity between the results of PFGE analyses and the true sequence similarities. These findings, when they are combined with results from laboratory studies, emphasize the need for the inclusion of multiple enzymes and additional epidemiological data in order to make more accurate interpretations.

FIG. 1.

Relationships of the isolates in the two populations. The outbreak population (A) consisted of the reference isolate plus an additional 16 isolates (17 isolates in total), each of which differed from the reference isolate by a certain amount. The percentage represents the probability of mutation of each base position and, therefore, is approximately equal to the overall sequence difference between the reference isolate and the simulated isolate. The ecological population (B) consisted of the reference isolate, 6 isolates that were simulated from the reference isolate, 2 isolates that were simulated for each of the isolates simulated in the first step, and then an additional 2 isolates that were simulated from the isolates at the second step (43 isolates in total). The complete branching structure is shown only for isolate C1 but was identical for all isolates, isolates A through F.

FIG. 2.

Dendrogram depicting the relatedness of the outbreak population of isolates. Relationships are based on the sequence similarity of isolates, and the dendrogram was generated by UPGMA clustering. REF, reference isolate.

FIG. 3.

Dendrogram depicting the relatedness of the ecological population of isolates. Relationships are based on the sequence similarity of isolates, and the dendrogram was generated by UPGMA clustering. REF, reference isolate.

FIG. 4.

Dendrograms depicting the relatedness of the outbreak population of isolates and perfect matching. The four different analyses included digestion of the REST data set with XbaI (A); digestion of the REST data set with NotI (B); digestion of the REST data set with SfiI (C); and digestion of the REST data set with XbaI, NotI, and SfiI (D). Relationships are based on the matrices of band-sharing similarity coefficients among isolates, and the dendrogram was generated by UPGMA clustering. REF, reference isolate.

FIG. 5.

Dendrograms depicting the relatedness of the outbreak population of isolates and imperfect matching. The four different analyses included digestion of the IMP-REST data set with XbaI (A); digestion of the IMP-REST data set with NotI (B); digestion of the IMP-REST data set with SfiI (C); and digestion of the IMP-REST data set with XbaI, NotI, and SfiI (D). Relationships are based on the matrices of band-sharing similarity coefficients among isolates, and the dendrogram was generated by UPGMA clustering. REF, reference isolate.

FIG. 6.

Dendrograms depicting the relatedness of the ecological population of isolates and perfect matching. The four different analyses included digestion of the REST data set with XbaI (A); digestion of the REST data set with NotI (B); digestion of the REST data set with SfiI (C); and digestion of the REST data set with XbaI, NotI, and SfiI (D). Relationships are based on the matrices of band-sharing similarity coefficients among isolates, and the dendrogram was generated by UPGMA clustering. REF, reference isolate.

FIG. 7.

Dendrograms depicting the relatedness of the ecological population of isolates and imperfect matching. The four different analyses included digestion of the IMP-REST data set with XbaI (A); digestion of the IMP-REST data set with NotI (B); digestion of the IMP-REST data set with SfiI (C); and digestion of the IMP-REST data set with XbaI, NotI, and SfiI (D). Relationships are based on the matrices of band-sharing similarity coefficients among isolates, and the dendrogram was generated by UPGMA clustering. REF, reference isolate.

FIG. 8.

Mantel's randomization test correlation coefficients for the complete and restricted analyses of each enzyme combination with the outbreak population (A) and the ecological population (B). All correlation coefficients were statistically significant (P < 0.0005).