Progression Toward an Improved DNA Amplification-Based Typing Technique in the Study of Mycobacterium tuberculosis Epidemiology

Mycobacterial cultures.Mycobacterial cultures were identified as M. tuberculosis complex using a combination of growth, microscopic, and biochemical characteristics, as well as DNA hybridization tests (Accuprobe; GenProbe, San Diego, Calif.). Cultures were grown on solid Lowenstein-Jensen medium or in Middlebrook 7H9 medium supplemented with 10% (vol/vol) oleic acid-albumen-dextrose-catalase supplement. Samples were incubated for up to 8 weeks in the case of solid media and for 2 weeks in the case of liquid culture.

Mycobacterial cultures were analyzed in three panels, designated studies 1, 2, and 3.

Study 1.Study 1 consisted of 71 isolates drawn from 11 outbreaks investigated by the Health Protection Agency (Mycobacterium Reference Unit), which were coded and analyzed blind using VNTR. Data consisting of RFLP, spoligotyping, rapid epidemiological testing (RAPET) (44), and epidemiological information were combined and formed the basis for the decision analysis for clustering and comparison with the VNTR-determined clustering. MIRU clustering was performed blinded to this information.

Study 2.Study 2 consisted of 126 isolates, for which full epidemiological data were available, out of a total of 174 isolates from patients in one city in the middle of England (Leicester) during a 1-year period known to contain isolates from one major outbreak. The VNTR analysis was again blinded to the RFLP, spoligotyping, RAPET, and epidemiological data that were used to define “true” clustering.

Study 3.Study 3 consisted of 248 isolates randomly selected from a collection of all isolates from culture-positive patients seen in London, England, during the first two months of 1998. This collection was part of a larger study involving 2,500 strains. Very limited epidemiological information was available, and cluster comparison was performed against HCN-RFLP and LCN-RFLP/spoligotype-defined clusters. In this case, these two molecular methods (which indeed are our standard practices) were used in the definition of true clustering.

Extraction of DNA.The extraction and quantification of DNA was performed as described previously (44). Genomic DNA was diluted in distilled water in a ratio of 1:100 before being used for spoligotyping and MIRU/VNTR typing.

IS6110 RFLP and spoligotyping.IS6110 RFLP was performed as previously outlined by van Embden et al. (39). Images were recorded and analyzed using Bionumerics software (Applied Maths, St. Marten-Latem, Belgium). The analysis parameters were as follows: the initial tolerance level was set at 3%, with confirmation of clustering being undertaken by visual inspection. The resulting dendrogram was created using the unweighted-pair group method using average linkages (UPGMA) program. In the case of the third panel, where there was limited epidemiological data, an approach of 100% identity between grouped RFLP patterns was taken to define a cluster. In the case of the first two panels, where this extra information was available, patterns that differed by one band were considered linked if these data were supportive. This occurred once in panel 2, where one band was missing from the 14-band RFLP pattern seen in the other members of an epidemiologically well-defined cluster.

Spoligotyping was performed as previously stated by Kamerbeek et al. (21). Images were transferred and analyzed using Bionumerics software (Applied Maths, St. Martens-Latem, Belgium) using the DICE setting with the dendrogram produced using the UPGMA program.

VNTR typing.Twenty-five VNTR loci were analyzed in each of the M. tuberculosis cultures (designated 25VNTR). The 25 VNTR loci consisted of the 12 MIRU loci described by Supply et al. (37) and Kwara et al. (23). ETR-A, -B, and -C were described by Frothingham and Meeker-O'Connell (15), and an additional 10 loci were from those described by Le Fleche et al. (24); for a listing of the various loci that were used in this study, see Table 2. Primers and dyes used for analysis of the 12 MIRU loci were those used by Kwara et al. (23). In each case only the forward primer was labeled (Proligo France, Paris, France), while reverse primers were unlabeled (Invitrogen, Paisley, United Kingdom). Primers for ETR-A and -B were as described by Frothingham and Meeker-O'Connell (15), with ETR-A labeled with dye 2 and ETR-B labeled with dye 4. Primers for the remaining 11 loci were designed using PrimerSelect (DNAStar, Madison, Wis.) from sequences obtained on the Tuberculist website (http://genolist.pasteur.fr/TubercuList/ ) for M. tuberculosis H37Rv. These primers and the dyes used are described in Table 1.

DNA amplification setup and conditions of all analyses were as described by Kwara et al. (23), with the following amendments. It was found that the addition of 1% dimethyl sulfoxide (final volume) improved the efficiency of the amplification reaction. Fragment analysis of labeled PCR products was performed using a CEQ-8000 genetic analysis system (Beckman-Coulter, Fullerton, Calif.) run at 6.0 kV for 45 min. The 12 MIRU loci and ETR-A, -B, and -C loci were analyzed using a 60- to 640-bp ladder labeled with dye 1 (Beckman-Coulter, Fullerton, Calif.). The additional 10 loci were analyzed using both the 60- to 640-bp ladder and a 600- to 1,000-bp ladder again labeled with dye 1 (Bioventures, Murfreesboro, Tenn.).

The resulting data were analyzed on Bionumerics (Applied Maths, St. Marten-Latem, Belgium) software and clustered based on a comparison of character. The dendrogram was created using UPGMA. Samples that could not be analyzed using the automated system were run on 1.5% (wt/vol) agarose gels and sized against a 100-bp ladder (Promega, Madison, Wis.), and sizes were estimated.

Cluster analysis was performed on the three isolate panels using three VNTR loci sets. The first set consisted of the 12 MIRU described by Supply et al. (37), the second was the same set plus the remaining ETR loci, giving a 15MIRU-VNTR set, and the third set was comprised of 25 VNTR loci (designated 25VNTR) as described above. A summary of the composition of the VNTR panels referred to in this study is shown in Table 2. All VNTR analysis was performed blinded to all other epidemiological data.

Calculation of the discrimination index.The discrimination index was calculated as described by Hunter and Gaston (20). The closer the value is to 1, the more discriminatory the locus. The results of this testing can be seen in Table 3.

Study 1.Study 1 consisted of 71 isolates involved in 11 outbreak investigations. This included investigations containing both linked and unlinked isolates. Linkage was established using a combination of RFLP (for HCN) or RFLP and spoligotyping (for LCN) and RAPET with other epidemiological (contact tracing) data. The results are summarized in Table 4. The 12MIRU set clustered the isolates that had previously been shown to be linked but also clustered with two further clusters, each containing two RFLP-IS6110 high-copy-number isolates that had no epidemiological links. In both cases the similarity between the RFLP patterns was over 70% (data not shown). The 15MIRU-VNTR set gave the same clustering results as those seen in the original outbreak investigations. Interestingly, the 25VNTR set gave the same results, except for one cluster of three strains, which was divided by two loci, 531 and 2163a, into three unique types (data not shown). Where isolates were analyzed using VNTR in duplicate, identical profiles were seen.

Study 2.Study 2 was designed to determine the value of the VNTR technique within a population (n = 125) where the majority of isolates present were shown by RFLP, spoligotyping, and available contact tracing data to be unlinked. This population, however, was known to contain an established outbreak of 12 individuals. As was the case with study 1, the VNTR technique was performed blinded to this information. The results of the various typing strategies on the entire population are shown in Table 4. From this study, it can be seen that in comparison with the conventional strategy of HCN-RFLP and LCN-RFLP/spoligotyping (defined here as “the conventional method”), while 12MIRU and 15MIRU-VNTR both overestimate clustering, 25VNTR is actually more discriminatory than what is currently used. If spoligotyping data are used together with 12MIRU or 15MIRU-VNTR, clustering is reduced to near or below that seen using the conventional method. Adding spoligotyping to 25VNTR does not further reduce clustering seen with this loci set. In order to determine the effect of the proportion of HCN and LCN strains on clustering, the population was sorted into these categories and the typing strategies were reapplied (Table 4). Among the 98 HCN isolates, both 12MIRU and 15MIRU-VNTR overestimated clustering almost twofold in comparison to RFLP. 25VNTR showed the same degree of clustering as RFLP. When applied to the 27 LCN isolates, all the VNTR-based methods proved more discriminatory than the combination of RFLP and spoligotyping. 25VNTR showed there were no clustered isolates within this small population.

Study 3.Study 3 examined a sample of isolates taken from London, which has a different population structure than Leicester, having far more ethnic diversity. Although RFLP and spoligotyping data were available, the VNTR-based techniques were done blinded to this information to mimic how prospective VNTR typing would work. The clustering seen within this panel using the different typing techniques is shown in Table 4.

From this panel of 248 isolates, it was observed that in comparison with the conventional method of HCN-RFLP and LCN-RFLP/spoligotyping, none of the VNTR-based methods used alone performed as well (Table 4). While a threefold drop in clustering was seen between 12MIRU and 25VNTR, 25VNTR showed more clustering than the conventional methods. The addition of spoligotyping data to the VNTR panels dramatically improved discrimination in both the 12MIRU and 15MIRU-VNTR data sets by approximately half and reduced clustering within the 25VNTR set to a figure close to that of the conventional methods.

In order to determine the effect of the proportion of HCN and LCN strains on clustering in this population, the isolates were sorted into these categories and the typing strategies were reapplied (Table 4). RFLP of the HCN isolates (Table 4) gave the lowest clustering values. The VNTR panels gave between three (25VNTR) and eight (12MIRU) times the clustering seen with RFLP used alone. When spoligotyping was added to VNTR, this clustering was reduced to at best more than twice the clustering seen with RFLP.

In the case of the LCN isolates (Table 4), while there was no difference in the clustering rates seen between the conventional method and 12MIRU typing, both 15MIRU-VNTR and 25VNTR reduced the clustering seen from 34.1% to 18.1%. The addition of spoligotyping data to 12MIRU reduced the clustering seen when this technique was applied uniquely. However, it was found that the addition of more loci in this case had no effect on the discrimination seen.

Determination of the individual discriminatory powers of the loci used. While this study has shown that the 25VNTR panel was the most discriminatory of the three used, it is likely that a number of loci provide more discrimination than others, which in turn could allow for the reduction in the size of the panel. To demonstrate this, the Hunter-Gaston index value (20), seen as a marker of discrimination, was calculated for each locus from the results of study 3 and were ranked in decreasing order. Using this ranking as a guide and starting with locus VNTR-3232, the loci with the highest Hunter-Gaston index values were added until the same pattern of clustering seen in study 3 with 25VNTR was seen.

Although a number of loci were highly discriminatory, when added to the panel they provided little further discrimination, while some with lower individual values did. Through the addition and subtraction of loci, a panel of 18 loci was found that gave the same discrimination as 25 loci for study 3. Using the same strategy for determining the best panel to be used alongside spoligotyping, this panel was further reduced from 18 to 14 loci with the removal of 5 loci and the addition of locus MIRU-40.

False clustering with VNTR and the similarity of RFLP patterns.The ideal level of discrimination required between strains of M. tuberculosis is dependent upon the use to which the data are being put. At one extreme would be a level of discrimination that would distinguish between M. tuberculosis and other species. At the other extreme would be a level of discrimination capable of distinguishing between subpopulations within a clone. A useful level of discrimination for epidemiological purposes is somewhere between the two, although a higher level is probably required for typing patient populations than that required for investigating suspected outbreaks. Currently the best evidence for a transmission event and hence an epidemiologically useful cluster would be two patients with matching IS6110 RFLP fingerprints as well as spoligotype and conventional epidemiological data. This would be the best reference for evaluating other methods for detecting clusters. When evaluating molecular analytical methods, if higher levels of discrimination than those seen with the IS6110 RFLP and spoligotype fingerprints are seen, the use of conventional epidemiological data becomes essential for defining “true” clusters.

The VNTR analysis of M. tuberculosis makes routine prospective typing of large populations a practical proposition. The highest degree of discrimination is required for the identification of clustered cases in a population that are not apparently linked by any other means. In this study, we define cases that are clustered by one method but differentiated by another as false clusters. If clustering data are used to inform contact tracing, the choice of VNTR methodology must be driven by the need to minimize false clustering and hence “false alarms.” This study accesses the value of different combinations of VNTR loci compared to the reference IS6110-RFLP methodology in a strain panel that is highly characterized at both a molecular and epidemiological level and a population panel of strains.

In a preliminary analysis, we defined clusters based on identical high-copy RFLP as true, while VNTR clusters formed with nonidentical RFLP patterns were false. Previous studies using the same criterion for the definition of a true cluster have shown that the 12MIRU panel indeed gives rise to a high proportion of false clustering events when used alone (22, 26, 37). The addition of the three remaining ETR loci reduces false clustering in a small sample of isolates with known epidemiological linkage, which in turn is analogous to our study panels 1 and 2 (19). In this current study, the rate of false clustering in all three investigations was seen to decrease with an increase in the size of the VNTR panel. However, the addition of the three VNTR loci in each investigation, while reducing the apparent false clustering rate from that seen for 12MIRU, did not eliminate this error completely, an observation also seen with the addition of the other 10 loci.

It has been argued that a degree of variability in RFLP patterns can be tolerated when defining clusters (38). Minor changes may occur as the pathogen is transmitted from host to host over an extended period of time (2, 27, 34, 43). In study 2, one isolate which was clustered by all the VNTR panels but not by RFLP (based on our criterion of identical fingerprints) was indeed incorporated into the cluster on account of strong conventional epidemiology evidence (data not shown). If we consider RFLP clustering being based on ±1 band difference, as in the example above clustered by RFLP, overall only a very small percentage of false clustering would be seen in the study 3 population with the largest 25VNTR panel (2% of the overall total). It has been postulated that the molecular clock for RFLP is faster than that of VNTR (25), and therefore this may account for an overclustering of samples using the VNTR methods.

However, while some factors (such as preferential location sites) are known to affect IS6110 transposition (42), there may be other factors which promote more drastic rearrangements. While it might be correct to assume that high-copy patterns missing one or two bands may be related, this may also be true in cases where there is less similarity.

A general problem with molecular epidemiology of M. tuberculosis is that some strains are more common in a given geographical area or ethnic group from a given area (28, 40). Two indistinguishable strains imply transmission, but in these circumstances this may not necessarily have been recent and may reflect past transmission events in the individual's country of origin. Therefore, if we see two strains from different geographical areas in the United Kingdom that cluster by RFLP from patients of the same ethnic group, this may have occurred through the acquisition of a strain that is endemic overseas rather than a transmission event in the United Kingdom.

The use of other techniques in parallel with VNTR to improve discrimination.From our previous work performed on a subset of isolates from study 3, it was found that a combination of 12MIRU and spoligotyping was more discriminatory than RFLP alone (our unpublished data). It was initially thought that this may have been due to the population structure of this subset, which was mostly made up of low-copy isolates, where it is known that RFLP is less effective (6). However, when spoligotyping data was applied to 15MIRU-VNTR data from studies 2 and 3, it substantially reduced the number of clusters formed and had this same effect when combined with 25VNTR data for study 3. This substantiates the observation of Cowan and colleagues (9), who saw a reduction in clustering rates when combining 12MIRU with spoligotyping. Indeed, this group advocates, and has now introduced, a two-step procedure where a combination of 12MIRU and spoligotyping is used for first-pass typing and RFLP is used to test the clusters formed.

Population structure and the use of VNTR-based typing.Approximately 20% of the reference cultures received at the Health Protection Agency Mycobacterium Reference Unit are of low-copy RFLP patterns (four bands or less). Currently, spoligotyping is the secondary typing system used when testing these strains. However, in this work, we have shown that MIRU-VNTR and 25VNTR have far higher levels of discrimination than both LCN-RFLP and spoligotyping combined (see Table 4). This further validates the use of these VNTR panels in the typing of this subset of isolates. Splitting the populations into HCN and LCN isolates gives an indication of how VNTR typing techniques would perform in situations where a population has a higher proportion of HCN or LCN isolates than in London. Although the numbers of LCN isolates used in this study are small, the VNTR techniques alone or with spoligotyping are radically better than our current methods. Furthermore, we have observed that increasing the proportion of LCN isolates within a population improves VNTR-based techniques versus conventional methods (data not shown). In populations with a significantly higher proportion of LCN isolates, such as the Indian subcontinent, there would be more justification for using VNTR than in locales where HCN strains predominate.

The use of other loci to improve the discrimination.As already shown, the addition of more loci to the panel of 12MIRU elements improved discrimination of this technique. Using the minisatellite database, a further 10 regions have been shown to vary in size between the sequenced H37Rv (8) and CDC1551 (14) M. tuberculosis strains (data not shown). In addition, there are other loci that have been recommended as being highly discriminatory that may improve further on the panels we have utilized in this study (P. Supply, personal communication).

Further, a number of tandem-repeat regions are located within genes defined as PE-poly-GC-rich sequence-like (8). It has been suggested that these genes allow for antigenic variation within tuberculosis infections (3), and these may offer additional variable regions for analysis. However, their small size (9 bp long instead of the 50- to 70-bp elements that have been analyzed here) will offer technical challenges in terms of data interpretation.

The aim of this study was to compare the VNTR typing methods with those currently available to the molecular epidemiologist. From our work, we have shown that while the discriminatory power of VNTR can be improved with a selection of loci, at this time this value is still less than that of the conventional HCN-RFLP and LCN-RFLP/spoligotype techniques.

For prospective typing, the most discriminating combination of VNTR loci should be employed. The data in this study suggest that the 18-loci VNTR set alone or a 14-loci VNTR set, if used in tandem with spoligotyping, would be highly discriminating. Investigation of alternative or additional VNTR loci may improve discrimination further. Application of these tools would minimize the secondary RFLP typing required, which has been shown here to give the maximum discrimination between isolates containing IS6110 in HCN isolates.