In this model population representing 91% of culture-positive TB cases collected in 2003 in Hamburg, the optimized MIRU-VNTR set showed a comparable to slightly better predictive value for the study of TB transmission than the current gold standard IS6110 RFLP, especially when combined with spoligotyping. The conditions in which this comparison was performed were stringent, since the isolate population displayed more than 90% of IS6110 high-copy-number fingerprints, for which IS6110 RFLP is especially discriminatory, and 37% of isolates from the single Haarlem lineage for which the original set of 12 MIRU-VNTR loci showed a relatively low discriminatory power.
Isolates clustered with identical IS6110
fingerprints are generally assumed to be clonal if they contain high copy numbers of IS6110
(≥6) and, depending on the context, considered to be prognostic of ongoing transmission events. In the present study, IS6110
RFLP typing resulted in 13 clusters, 5 of which had a confirmed epidemiological link. These five IS6110
clusters were perfectly matched by MIRU-VNTR typing over the 24 loci used. Moreover, six of the eight IS6110
clusters with no identified epidemiological link were also identically retrieved with MIRU-VNTR typing. In these cases, the perfect concordance between two highly discriminatory molecular methods does suggest that the patients involved were infected by the same strain, despite the absence of confirmation by contact tracing. This lack of confirmation can then be explained by the occurrence of transmission through casual contacts, barhopping, or illegal activities (which are difficult to detect by traditional investigation) (5
), transmission chains not covered by the study period, or reactivation of a previous infection with an endemic strain (3
These findings confirm previous reports on the high stability of MIRU-VNTR types (based on 12, 15, or 24 loci) among epidemiologically linked isolates and in IS6110
). They are in contrast to a single study, which reported changes in MIRU-VNTR patterns among 48% of isolates from IS6110
clusters (in addition to spoligotype changes among 17% of isolates from IS6110
). The significance of these latter results remains unclear in the absence of contact tracing data.
Actually, several other studies, integrating contact tracing data, showed that not all clusters based on IS6110
RFLP patterns, even when multibanded, necessarily reflect clonal transmission of TB (3
). Systematic assessment of the epidemiological significance of IS6110
-RFLP clusters revealed a proportion of 14% of false IS6110
clusters in a Dutch population-based context (29
). These clusters without any evidence of epidemiological link after thorough epidemiological investigation differed by two to six loci from the 15-locus discriminatory subset (24
). Here, in the Hamburg population, one cluster of two isolates with an identical IS6110
high-copy-number fingerprint and an identical spoligotype was split by as many as seven MIRU-VNTR loci, four of which are part of the 15-locus subset. The IS6110
RFLP, spoligotyping, and MIRU-VNTR results were repeated using the same original DNA samples, ruling out trivial laboratory errors. Both prospective and retrospective contact tracing provided no evidence of any epidemiological link among the patients involved (see Results). Therefore, the differences in four of the 15-locus subset observed here, reinforced by three additional differences among auxiliary loci with a lower evolutionary rate, leave little doubt that the patients were falsely clustered by IS6110
high-copy-number fingerprints. More trivially, a second IS6110
RFLP cluster, including three isolates with a single IS6110
band, was split by 6 to 12 differences in the 15- or the 24-locus set, as well as by spoligotyping. These two examples show how the interrogation of 15 or even 24 independent markers can increase the degree of confidence for reliably excluding recent transmission, especially if the results obtained with IS6110
RFLP typing are conflicting.
Even differences by a single MIRU-VNTR locus remain strongly predictive of the absence of a link, in contrast to results reported by Scott et al. (19
). In the present study, a total of 15 SLV events were detected within the 15-locus subset. None of the corresponding isolates distinguished by these SLVs was clustered by IS6110
RFLP, which corroborates the absence of a link. These results are fully consistent with our recent evaluation of MIRU-VNTR stability using large sets of epidemiologically linked or clonal isolates (24
). We estimated the risk of erroneously excluding from a cluster an isolate showing an SLV in the 15-locus subset to be only of 5 to 6%. This level is just below the statistically predicted threshold of detection in the present study (6.7%, corresponding to 1 epidemiologically erroneous SLV among a total of 15 SLV events).
In all, only three isolates among the 115 defined as unique by IS6110
RFLP were clustered by the 24 MIRU-VNTR loci (see cluster 9b in Fig. ). These three isolates shared common IS6110
banding patterns typical of the Haarlem lineage, demonstrating their close genetic relationships. However, these three isolates judged unique by IS6110
RFLP were also distinguished by spoligotyping, clearly indicating the absence of a link in agreement with contact tracing. Moreover, the use of the 15-locus subset resulted in the addition of a single cluster containing only two other isolates, thus only marginally affecting the resolution power obtained with the full set of 24 loci. Interestingly, these two isolates were also discriminated by spoligotyping (see arrows in Fig. S1a in the supplemental material). These observations support the design of the 15-locus subset, which concentrates most of the resolution obtained with the full 24 loci across different M. tuberculosis
lineages. Furthermore, these findings show that the combined use of spoligotyping can add marginal but still meaningful resolution to the optimized 15- or 24-MIRU-VNTR locus sets (24
), resulting in the discrimination of all of the isolates already predicted as unique by IS6110
RFLP in the present case. In comparison, 9 or 14 isolates among the 115 or the 120 defined as unique by IS6110
RFLP or by combined optimized MIRU-VNTR and spoligotyping, respectively, were clustered by the use of the 12 original MIRU-VNTR loci with spoligotyping. This false clustering mainly resulted from lower discrimination among the predominant Haarlem isolates and would have led to an overestimate of the TB transmission rate in Hamburg, albeit to a lower degree than that reported for the use of these 12 original loci with spoligotyping in Montreal (19
Thus, if the specificity and sensitivity for cluster analysis are calculated as described previously (19
) (i.e., a proportion of cases classified as unique by one method among cases classified as unique by another and a proportion of cases identically clustered between two methods, respectively), but by additionally taking into account the contact tracing data, MIRU-VNTR typing based on the 15 or the 24 loci showed an overall specificity of 95.7 and 97.4% compared to IS6110
RFLP. The specificity went up to 100% when either of the two sets was combined with spoligotyping. In addition, both MIRU-VNTR typing sets showed a sensitivity of 100% to detect the proven or possibly epidemiologically linked IS6110
clusters, excluding the most likely false IS6110
clusters 1 and 12 (see above). Conversely, IS6110
RFLP also showed a sensitivity of 100% but a specificity of only 95.8% (due to false IS6110
clusters 1 and 12) compared to the combination of MIRU-VNTR and spoligotyping.
The study period for the comparison of these molecular methods as tools to study ongoing TB transmission was limited to a single year. The analysis of the ongoing transmission itself requires longer periods (usually 2 or 3 years). Over longer periods, it has been noticed that the rate of molecular clustering often increases because, among other reasons, transmission chains are more efficiently covered (9
). However, because our results are based on an almost complete population representative of 1 year, we expect that this phenomenon will occur with both MIRU-VNTR and IS6110
RFLP fingerprinting, without much affecting the relative performances of the two methods.
In conclusion, our population-based investigation supports the wide applicability of a combination of the newly defined 15 loci MIRU-VNTR typing and spoligotyping as a real-time approach for population-based studies of TB transmission, at least in the numerous low-incidence settings with epidemiological characteristics similar to those of Hamburg. In areas with a higher incidence of TB, where the epidemiological interpretation of the results may be more difficult, this approach can be used at least as a reliable exclusion method (i.e., to identify most likely unrelated isolates by genotypic differences) to get an estimate of the maximal rates of ongoing TB transmission. The 15 or 24 MIRU-VNTR typing combined with spoligotyping represents the first PCR-based method with operating parameters (specificity and sensitivity) comparable to those of the gold-standard method IS6110 fingerprinting in a population-based investigation and thus can be used as a stand-alone approach to the study of TB transmission. Because of its speed and ease of data exchange and comparison, this method may actually become the new gold standard method.