Tuberculosis (TB) control is a fundamental aspect of the U.S. public health system, whereby cases with TB disease and their contacts at risk are systematically identified, investigated, and treated. TB patients who are proximal to one another are considered clustered when their Mycobacterium tuberculosis isolates have indistinguishable genotypes. TB cluster investigations play an important role in tracking transmission of specific Mycobacterium tuberculosis (Mtb) strains in a population [1]. Among the tools used by investigators, polymerase chain reaction (PCR)-based methods such as spoligotyping (spacer oligonucleotide typing) and MIRU-VNTR (mycobacterial interspersed repetitive units-variable number of tandem repeats) are useful for identification of clusters of Mtb strains (through molecular characterization “fingerprints” or genotypes) due to their relatively fast laboratory turnaround time, suitability for discriminatory analyses, and their reproducibility [2–5]. These Mtb characterization methodologies have become the routine genotyping methods used by the U.S. Centers for Disease Control and Prevention (CDC) since 2004 [6]. However, the limited discriminatory power of the original 12-locus MIRU-VNTR (MIRU12) sometimes identified clusters that included cases among patients who may not have been connected by recent transmission. Therefore, in 2009, CDC’s National TB Genotyping Service (NTGS) began using 24-locus MIRU-VNTR (MIRU24) to improve the discriminatory power of TB molecular surveillance in the United States (U.S.) [7–9].

To better understand the practical utility, programmatic application, and added value of increased discriminatory power associated with the additional 12 loci in genotyping, TB transmission must be verified by epidemiologic investigations [8, 10–14]. With this background, we investigated the concordance of MIRU24 among epidemiologic-linked patients in four U.S. public health jurisdictions. We also attempted to determine the programmatic benefits of combining MIRU24 and spoligotyping.

From 44 PCRType clusters randomly selected in our previous study [17], fifty-nine patients with incomplete MIRU24 results (including 13 patients for the entire clusters 3 and 31 from MD and GA respectively) were excluded from the analysis. In total, 42 clusters consisting of 342 patients having complete MIRU24 results were included in the current study (Appendix A). The number of patients in clusters from the GA, MD, MA, and HOU study sites were 73 (21%), 53 (16%), 71 (21%), and 145 (42%), respectively (Figure 1). Patients’ median age at TB diagnosis was 47 years (interquartile range: 21–57); 73% were males, 70% were U.S.-born and 55% were black (Table 1). Eighteen of the 42 clusters (43%) had only U.S.-born patients (data not shown). In an effort to determine the extent of cluster homogeneity by birth country, Appendix B shows the birth country details for patients in the 24 clusters having at least one patient with a birth country other than the U.S. Eight out of 42 clusters (18%) contained only foreign-born patients. Among the 342 study patients, 150 (43.9%) were found to have at least one epidemiologic link to another patient in the cluster to form 156 linked patient-pairs. There were 33 patients that had more than one epidemiologic link. The strength of the linkages between each of 156 linked patient-pairs was most often considered definite (97/156, 62%), and less often probable (22/156, 14%) or possible (37/156, 24%) (data not shown).

This study assessed the impact of using MIRU24 versus MIRU12 by systematically investigating clusters for epidemiologic relationships among patients (i.e., evidence of transmission among patients within a genotype cluster). MIRU24 analysis further differentiated 342 patients in 42 PCRType clusters into 46 GENType clusters comprised of 282 patients, plus 60 patients with unique GENTypes. By routinely using spoligotyping and MIRU24 (GENType) as the primary cluster identification method, follow-up cluster investigations through chart review, extensive interview and contact investigation of 60/342 (18%) patients in PCRType clusters could have been avoided and as a result, TB program’s resources for these interventions could have been saved. Additionally, using spoligotyping and MIRU24 (GENType) could provide a more accurate determination of the transmission risk among TB patients clustered by PCRType. This overall finding supports the practice of defining genotype clusters by GENType. While current practice has sometimes been to consider including patients with a SLV genotype that already has identified links (e.g., through a contact investigation) in a cluster, re-reading or repeating MIRU24 when the results have missing or ambiguous values should be considered in order to definitively rule out a patient from a genotypic cluster.

In this study, 15% of Mtb isolates (n=59) had mixed or missing information for one or more of the 12 additional MIRU24 loci. Fortunately, the proportion of patients genotyped by NTGS with missing MIRU24 loci has decreased substantially due to improvement in laboratory methods (< 5% in 2013) since the time of this study (CDC, unpublished data).

More than half of the PCRType clusters (22/42, 52.4%) had 100% of patients who were also GENType clustered. In other clusters, however, each patient had a unique MIRU24 pattern –demonstrating that patients in some PCRType-based clusters should not be considered clustered. In particular, the five East Asian (L2) PCR00002 clusters had significantly decreased odds of epidemiologic linkages using GENType (OR=0.05, 95% CI 0.001, 0.95) and were heterogeneous both by birth country and MIRU24; only 26% (10/38) of patients clustered by GENType, which was significantly less than patients in other clusters (272/304, 90%, p<0.001). These findings appear to be consistent with the insufficient discrimination power of MIRU24 in isolates from the Beijing family that were reported by other authors [22–23].

Our results suggest that defining TB clusters based upon PCRType overestimated clustering, especially for genotypes that are common in the United States. Clusters associated with certain PCRTypes (e.g., PCR00002, PCR00015, PCR00022 and PCR00041, which are each reported in >40 U.S. states and associated with more than 400 TB patients nationally during 2006–2010) were clearly discriminated by MIRU24 (Appendix A). On the other hand, our results also show that SLVs for certain MIRU-VNTR loci can result in underestimation of clustering by MIRU24. Our comparison of the MIRU24 patterns of epidemiologic-linked patients showed ~6% of linked pairs were SLVs (Appendix A). If an epidemiologic linkage between patients exists where the only MIRU24 difference is missing data at a particular locus for one patient and an identified value at the same locus for another patient, it remains equally likely that they are in the same chain of transmission even though their isolates will be assigned different GENTypes. Additionally, the decreased proportion of epidemiologic links in PCRType and GENType clusters containing all foreign-born patients was (62.5% and 40.0%, respectively) and the GENType discordance found in 20 of the 156 epidemiologic linked pairs, which also suggests that using PCRType might be overestimating epidemiologic links in comparison with using GENType.

A notable limitation to this study was exclusion of clinically defined, and therefore non-genotyped, TB patients. In addition, we cannot rule out the possibility of misclassification due to false epidemiologic linkage in patient-pairs. Bennett et al. reported a higher proportion of discordant genotypes among epidemiologic linkages with 29% of epidemiologic-linked patient-pairs having discordant IS6110-based restriction fragment length polymorphisms (RFLP) patterns and 31% of epidemiologic-linked patient-pairs consisting of household members who were discordant [11]. Although the PPV to identify epidemiologic-linked cases was higher among clustered patients with GENType (47.2%) than PCRType (43.9%), this relatively small difference between these two methods suggested the discrimination power of MIRU24 is still needed to be further investigated. With increasing applications of whole-genome sequencing (WGS) and phylogenetic analyses for use by TB control programs, additional molecular resolution can help identify patients who are part of a recent chain of transmission [9]. In the United States, it is likely that WGS will be integrated into molecular surveillance and become the gold standard for strain identification and characterization of genetic relatedness among isolates [24].

Because MIRU24 further differentiates most MIRU12 clusters into smaller clusters, one might logically expect that GENType clusters would be associated with higher proportions of epidemiologic links identified. In general, as the proportion of cases clustered by GENType increased, the proportion of patients that were epidemiologic-linked within a cluster also increased. GENType-defined clusters with a high proportion of epidemiologic-linked patients provide strong evidence for recent transmission; prioritization of these clusters would improve public health action. We found that certain characteristics were associated with having epidemiologic links among patients clustered by GENType. Specifically, GENType clusters with epidemiologic links had greater odds of having patients residing in the same zip code as a risk factor and less likely to have had >20% of patients with only extrapulmonary TB or be infected by East-Asian lineage strain. These factors may be useful to guide local programs in determining which GENType clusters warrant field investigations.