INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

Molecular strain typing of Mycobacterium bovis, the causative organism of bovine tuberculosis, has become commonplace in recent years in an attempt to address the continued persistence of this pathogen in the cattle populations of many countries, in spite of intensive and costly eradication programs. Much attention has been focused on the application of techniques that distinguish between clinical isolates of the bacillus based on polymorphisms within the genomic DNA. Such methods may be used to distinguish between epidemiologically related and nonrelated herd breakdowns and so assist in tracing transmission of this disease and in the identification of likely routes of infection in particular herds. Techniques which have been applied to this sort of application include restriction fragment length polymorphism (RFLP) analysis (4, 7, 9, 19, 22, 23, 25), spoligotyping (1, 4, 7, 14, 20, 21), ampliprinting (12, 18), variable number tandem repeat analysis (11), and pulsed-field gel electrophoresis (10).

RFLP analysis is perhaps the most widely used method for the typing of M. bovis due to the availability of multiple DNA probes that are directed against specific polymorphic regions within the mycobacterial genome and which are capable of distinguishing between genotypic strains. This technique has been the method of choice, alongside spoligotyping, for strain typing of M. bovis isolates in Ireland (4, 5, 22, 23). The most commonly applied RFLP probes for M. bovis typing are IS6110 (4, 8, 16), PGRS (4, 7, 9, 19), and DR (1, 4, 7), each of which has associated advantages and disadvantages in terms of discriminatory power and ease of analysis. The spoligotyping technique has also been widely used for strain typing of M. bovis (1, 7).

Recently, a newly identified DNA probe for RFLP typing of M. bovis isolates, pUCD, was described (17). This probe was found to detect higher levels of M. bovis polymorphism in cultured isolates from Ireland, based on the results obtained for 60 isolates, than analysis with the mycobacterial RFLP probes IS6110, PGRS, and DR and the spoligotyping method. pUCD was also useful in subdividing the most commonly encountered Irish strain type (strain type A1 A1 A with IS6110, PGRS, and DR, respectively) which has been reported to occur in 20% of isolates typed (4). We report here that this relatively high level of strain discrimination was maintained when an expanded sample set of 299 Irish M. bovis isolates was examined with this probe.

The original pUCD clone described generated a certain amount of background banding which did not contribute to strain type differentiation and which complicated the banding pattern unnecessarily. In this study, the relevant polymorphic bands generated by the pUCD clone were reproduced by a mixed hybridization containing three different oligonucleotide probes. Each of these oligonucleotide probes is homologous to a different repeated sequence contained within the original pUCD clone and, in combination, account for the observed pUCD polymorphism.

We also report here a comparison of the performances of the commonly used mycobacterial RFLP probes IS6110, PGRS, and DR, the spoligotyping technique, and pUCD based on the results obtained with 299 Irish isolates. A statistical function for generating a single, objective value (probe power [PP]) which rates the relative usefulness of different methods for strain typing of isolates specifically for epidemiological purposes is described, and a comparison of the performances of RFLP analyses with probes IS6110, PGRS, DR, and pUCD and spoligotyping by this statistical approach is reported.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

RFLP analysis of M. bovis isolates. Isolates originating from different geographical regions of the country and different host species were available for typing as part of routine epidemiological investigations and were identified as M. bovis by standard biochemical tests (6). Isolates were subcultured on 50 ml of Middlebrook 7H9 medium with oleic acid, albumen, dextrose, and citric acid enrichment (Difco, Detroit, Mich.) for 4 weeks at 37°C and were harvested after inactivation by heating at 75°C for 1 h. Two hundred ninety-nine isolates, both epidemiologically related and nonrelated, were examined in this study. Isolates were collected from cattle (n = 156), badgers (n = 111), deer (n = 8), humans (n = 19), pigs (n = 4), and a fox (n = 1). Spoligotyping and RFLP analysis with IS6110, PGRS, DR, and pUCD as probes were carried out as described previously (4).

Characterization of pUCD. It was reported previously (17) that pUCD bears a high degree of homology with a region of the M. tuberculosis genome (segment 85/162 of the M. tuberculosis H37Rv complete genome [3]; GenBank accession no. Z97193). Oligonucleotides based on repeat regions identified within this region were synthesized and used as probes on AluI restriction enzyme-digested M. bovis genomic DNA to determine whether these regions were responsible for generating some or all of the polymorphism observed with pUCD. Additional repeat sequence elements were identified by comparing fragments of the pUCD clone which had been subcloned and sequenced with the sequences in the M. bovis genome database maintained at the United Kingdom Sanger Centre (http://www.sanger.ac.uk/Projects/M_bovis/) by using the BLAST sequence search algorithm. All subcloned sequences were aligned within a single region within this database, identified as contig 456 (note that contig identifiers from this database are subject to change as the M. bovis sequencing project progresses). This sequence was subsequently searched for repetitive sequence elements by using Tandem Repeats Finder, version 2.02, software (2). Candidate repeat sequences were synthesized and assessed as RFLP probes to determine whether they contributed to any subset of the total number of bands observed with the full-length pUCD probe.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Performance comparison of RFLP probes. Two hundred ninety-nine M. bovis isolates were typed by RFLP analysis with IS6110, PGRS, DR, and pUCD and also by spoligotyping. IS6110 identified 24 strain types, PGRS identified 36, DR identified 25, and pUCD identified 43. Spoligotyping detected 19 strain types. The IS6110-PGRS-DR combination identified 55 strain types, and the pUCD-DR combination identified 71. This study contained 63 examples of the most commonly occurring Irish A1 A1 A strain type which were subdivided into 11 strain types by both pUCD and pUCD-DR. The percentages of isolates grouped into a single, most common strain type by each method were as follows: spoligotyping, 52%; RFLP analysis with IS6110, 47%; DR, 44%; PGRS, 33%; IS6110-PGRS-DR, 21%; pUCD, 22%; and pUCD-DR, 16%.

View larger version (11K): [in this window] [in a new window]   FIG. 1.   Equations used to derive values for D, E, and PP. See text for full explanation of each function.

Epidemiological concordance. Of the 299 isolates typed for this study, 118 isolates from 44 different herds had an epidemiological association with one or more other isolates from the same herd breakdown. The strain type results relating to these isolates are listed in Table 3. Generally, typing with pUCD followed the expected epidemiological pattern and did not unnecessarily subdivide isolates from within individual herds. In six instances, pUCD did discriminate between within-herd isolates which the other probes had not segregated (Table 3, isolate 20 from Cork, isolate 52 from Kilkenny, isolate 63 from Limerick, isolate 77 from Monaghan, isolate 87 from Sligo, and isolate 96 from Waterford). Conversely, there were also three occasions in which within-herd isolates were subdivided by either IS6110, PGRS, or DR but not by pUCD (Table 3, isolate 2 from Carlow, isolate 9 from Clare, and isolate 95 from Waterford). In two cases in which within-herd isolate strain types were completely different with IS6110, PGRS, and DR, the result was reiterated with pUCD (Table 3, isolate 28 from Cork and isolate 76 from Monaghan).

Identification of repeat elements responsible for pUCD polymorphism. Three distinct repeat elements, named pUCD1, pUCD2, and pUCD3, were found to be responsible for generating the observed pUCD polymorphism, with each element producing a subset of the total banding pattern originally observed. To demonstrate the nature of the polymorphism identified by each oligonucleotide, 20 M. bovis isolates were probed in the first instance with the original pUCD plasmid, then with each oligonucleotide in turn, and finally, with a combination of all three oligonucleotides. The results obtained are illustrated in Fig. 2. It can be seen that a hybridization with a mixture of the three oligonucleotides generated a banding pattern identical to that obtained with the original pUCD clone, with the exception that nonpolymorphic, background banding has not been reproduced. As the same blot was reprobed several times, faint bands were visible (Fig. 2B and D) as a result of carryover from the previous hybridization.

View larger version (45K): [in this window] [in a new window]   FIG. 2.   Re-creation of the banding pattern obtained with the full-length pUCD clone by a combination of three oligonucleotide probes. (A) Pattern obtained with pUCD with 20 M. bovis isolates. (B) Subset of bands obtained with oligonucleotide pUCD1 (ATC GGA CCG ACA CCA CCC AGC GCG TTC AGG CTC AAC GGA ATA CCA GGA ATA GTA ATA TC). (C) Bands obtained with pUCD2 (CCG CCC ACA TCA ATA CCC AAC GGG ATT GCC GGA AGT GAG TAG CCA TCC GGG AAC ACC GTA). (D) Bands obtained with pUCD3 (CAT CGG TTT GGG TGT GAG TAG TCC GGG GCG TTG GTG CGC CAA TCA ATG TTC CGC CTA TTA). (E) Result obtained by hybridization with a mixture of oligonucleotides pUCD1, pUCD2, and pUCD3. Molecular size markers are indicated to the right of the figure, adjacent to panels A, C, and E.

View larger version (15K): [in this window] [in a new window]   FIG. 3.   Physical map of the pUCD region, derived from the M. bovis genome database, showing the number and relative positions of tandemly repeated sequences and recognition sites for the restriction endonuclease AluI.

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

We report here on the reduction of the pUCD probe to the level of oligonucleotides capable of detecting the same bands as those produced by the original clone. The use of oligonucleotides limits the final result solely to those polymorphic bands of interest, eliminating that proportion of bands which do not contribute to strain differentiation. A significant advantage of this approach is that it allows the DR oligonucleotide probe to potentially be incorporated directly into the same hybridization. This would allow the discrimination obtainable from a pUCD-DR probe combination to be achieved from a single round of RFLP analysis, as the restriction enzyme used for the preparation of samples for analysis with pUCD, AluI, is the same as that used for analysis with DR (17). It must be noted that, in this study, pUCD and DR were not included together in such a mixed hybridization and that the results reported here for a pUCD-DR combination were obtained by using these two probes independently and subsequently combining the results.

The fact that a single band is produced per isolate by each of the oligonucleotides described would suggest that these sequences are localized within a single region of the M. bovis genome; however, the single-locus nature of pUCD remains to be confirmed pending complete sequencing of the entire clone. The absence of the pUCD3 repeat element from the M. tuberculosis H37Rv genome might suggest that this repeat may be used to distinguish between these two organisms. Preliminary examination of M. tuberculosis isolates has demonstrated that the pUCD3 oligonucleotide does not produce the homologous band (data not shown). However, during the course of this study M. bovis isolates which also lacked a pUCD3 band were encountered, which would circumvent the use of pUCD3 as such a diagnostic marker. The reduction of pUCD to the level of oligonucleotides also facilitates automated computer band analysis, as no more than three bands are produced per isolate. Computer-assisted analysis has proved to be problematic in the past with the multilocus PGRS probe due to the complexity of the resulting fingerprint pattern and the difficulty associated with resolving a large number of closely spaced bands accurately (8, 15).

In assessing the value of an individual typing approach, the three main characteristics that must be considered are typeability, reproducibility, and discriminatory power (13). As a typing method, RFLP analysis is highly reproducible (15) and most isolates are readily typeable due to the availability of multiple DNA probes. The discriminatory power of RFLP analysis is directly determined by the particular probe(s) used and reflects the ability of individual probes to detect interstrain polymorphism within the organism's genomic DNA. Although typing with pUCD detected a greater number of individual M. bovis strain types than typing with IS6110, PGRS, or DR or spoligotyping, it is unrealistic to state that this makes pUCD a better probe for epidemiological purposes. A hypothetical probe with an ability to distinguish between every isolate would be of little practical value, as it would not allow meaningful epidemiological conclusions to be drawn in terms of disease transmission. A truly useful probe would differentiate well between unrelated disease outbreaks, while it would correctly identify instances of direct disease transmission and also facilitate easy downstream analysis. Simply comparing probes on the basis of the total number of strain types identified does not accurately reflect their practical value, and a more explicit means of comparison is required.

A statistical function which may be used to rank RFLP probes in order of epidemiological efficacy is described here. This function assesses both the discriminatory power of the probe and the level of within-herd variation generated by each probe to give a single value for PP. A proportion of herd breakdowns will generally exhibit within-herd variation (typically, 10% in Ireland [4]), which likely reflects more than one clonal origin of infection in a herd. This was taken to be a negative factor in the estimation of a probe's utility, as it cannot be distanced from any error introduced either by the collection process or by the typing procedure itself. Samples collected at abattoirs for culture and strain typing are invariably harvested under difficult conditions, and the possibility of error attributable to mislabeling of specimens or other sample contamination must be taken into account in the final analysis when multiple strain types are seen to arise from single herds.

Although the true level of probe error is necessarily exaggerated, the result represents a "worst-case scenario" in which all perceived errors, true or otherwise, are weighted negatively against the value obtained for a probe's discriminatory power. In terms of simply gauging a probe's ability to distinguish between herd breakdowns in an epidemiological study, this approach will overestimate the true level of probe error, as a proportion of herds in which more than one strain type is revealed will genuinely represent multiple, unrelated sources of infection. However, this will be the case for each probe under investigation and does not bias one probe over another.

On the basis of the results obtained with 151 Irish bovine M. bovis isolates obtained from 87 different properties, PP values were calculated for RFLP analysis with IS6110, PGRS, DR, IS6110-PGRS-DR in combination, pUCD, and pUCD-DR in combination and, finally, for spoligotyping. The PP values obtained may range from 0 to 1, where a value of 1 would represent an idealized epidemiological tool with the ability to distinguish between all unrelated herd breakdowns without detecting any degree of within-herd strain variation. In practice this degree of differentiation is unobtainable as a percentage of herds will demonstrate multiple strain types for valid reasons. Conversely, a value of 0 would indicate that a particular probe had classed strains from all herds into a single group. Another possibility for error, in which instances of true polyclonal coinfection of single herds are not detected by any of the methods used and all isolates examined are identified as the same strain, would not be recognized by this approach. By its very nature, however, this occurrence cannot realistically be accounted for.

The results obtained indicate that RFLP analysis with the pUCD-DR probe combination was the best typing approach (PP = 0.51) for this particular sample set, followed by RFLP analysis with pUCD (PP = 0.39), IS6110-PGRS-DR (PP = 0.37), IS6110 (PP = 0.28), PGRS (PP = 0.28), and DR (PP = 0.11) and, finally, spoligotyping (PP = 0.10) (Table 2). Although an IS6110-PGRS-DR probe combination detected more individual strain types (n = 55) than pUCD alone (n = 43) and pUCD detected more instances of within-herd variation, the PP value obtained for pUCD was higher (PP = 0.39 for pUCD as opposed to 0.37 for IS6110-PGRS-DR). This reflects the fact that pUCD generated a more balanced array of strain types, identifying fewer cases of isolated, singlet strain types and a correspondingly higher proportion of grouped isolates, while it also subdivided down some of the common strain types identified with the IS6110-PGRS-DR probe combination.

The key characteristics of the different probes and probe combinations are illustrated in Fig. 4, in which, for each method, the cumulative number of individuals allocated to particular strain types is plotted against the total number of strain types identified. On the graph in Fig. 4, the starting values for each probe and probe combination reflect the bias in strain distribution, in which a higher starting value corresponds to a higher proportion of individuals grouped into the most common strain types. The relative positioning of the profiles produced for each probe on the vertical axis reflects the degree to which individuals are allocated to unique stain types. The extent of the profile for each probe on the horizontal axis reflects the total number of strain types identified. This serves to illustrate the factors that give rise to one probe having a higher PP value than another. In the two extremes, it can be seen that spoligotyping produces the highest profile on the graph, grouping the majority of individuals into a small number of strain types followed by a rapid drop in the number of additional strain types identified. The pUCD-DR probe combination produced the lowest profile, allotting fewer individuals into each group and demonstrating a more steady accumulation of additional strain types. Each of the other probes discussed falls between these two extremes. In a comparison of the IS6110-PGRS-DR probe combination and pUCD alone, it can be seen from the graph in Fig. 4 that, overall, pUCD groups fewer individuals into common strain types than the three-probe combination and also demonstrates a more gradual accumulation of strain types. It is this property that gives pUCD a higher PP value than that for IS6110-PGRS-DR, even though RFLP analysis with IS6110-PGRS-DR detected a slightly larger number of individual strain types.

View larger version (24K): [in this window] [in a new window]   FIG. 4.   Cumulative number of strains identified by RFLP analysis with probes IS6110, PGRS, DR, pUCD, IS6110-PGRS-DR in combination, and pUCD-DR in combination and the spoligotyping method based on the results obtained with 151 bovine M. bovis isolates.

Strain typing by RFLP analysis is a highly reproducible and discriminatory technique for distinguishing between M. bovis isolates, although it is time-consuming when multiple DNA probes are required for maximum strain type resolution. The requirement for more than one round of hybridization and analysis in combining the results obtained with the IS6110, PGRS, and DR probes can make this technique cumbersome. The potential to obtain high levels of strain type discrimination through the combined use of pUCD-based oligonucleotide probes and the DR oligonucleotide probe on a single membrane may go some way toward shortening this procedure.

We describe here a statistical function which can be used to assess the relative performances of different RFLP probes in epidemiological studies. For the set of isolates examined in this study, it was determined by this statistic that pUCD alone outperformed an IS6110-PGRS-DR probe combination and that a pUCD-DR probe combination was more powerful still.