In November 2002, while visiting New York, New York, USA, 2 persons from Santa Fe County, New Mexico, USA, became ill with fever and unilateral inguinal adenopathy; clinicians subsequently identified the illness as bubonic plague. Investigation by the New Mexico Department of Health and the Centers for Disease Control and Prevention indicated the patients were infected in New Mexico because Y. pestis–positive fleas were collected near the patients’ home (10). However, because plague is endemic to the region, and flea samples from which isolates were obtained were collected at the home and along a local trail on which the patients hiked, either location could be the source. To identify the most likely fine-scale source of their infections, we examined specific mutations separating the human isolate from closely related environmental isolates.

Figure 1

Figure 1. Distribution of rodent trapping stations along a hiking trail in Santa Fe County, New Mexico, USA. Each red circle indicates a single trapping site that had 3 traps. Trap stations (not...

We examined 5 Y. pestis isolates (Table 1) to develop a fine-scale spatial analysis of the infection. The reference isolate was obtained from 1 patient, 3 isolates were obtained from fleas collected in the patients’ yard (9) (2 were collected before their illness as part of a long-term investigation), and 1 isolate was obtained from the trail flea samples a short time later as part of the same long-term study (Figure 1). Other isolates were collected and examined but were excluded from this fine-scale analysis because they were more distinct from the human isolate, differing at >4 VNTR loci. DNA extracts were prepared from each isolate (11,12) and analyzed using a 43-loci MLVA system as previously described (1,8).

Figure 2

Figure 2. Alternate infection source hypotheses for the plague cases in the persons who visited New York, New York, USA. Closed circles indicate genotypes; black, red, and blue circles indicate genotypes A, B,...

We observed 3 MLVA genotypes (A–C) among the 5 samples (Table 1, Figure 2). The human isolate was assigned genotype A. Genotype B, observed in 3 isolates obtained from the yard, differed from the reference by single-repeat mutations at 2 VNTR loci (M25 and M34; Figure 2, panel A). Genotype C, observed in 1 isolate from a flea obtained along the trail, also differed from the reference isolate at loci M25 and M34. However, the mutation at M25 was a double-repeat mutation that could be explained 2 ways: as a single 2-repeat mutational event (Figure 2, panel B) or as 2 sequential single-repeat mutations at the same locus (Figure 2, panel C). Although all 43 VNTR loci are useful for identifying the coarse geographic origin of an unknown isolate by using a database approach (9), our analysis examined only polymorphic loci because monomorphic loci provided no additional information. The molecular epidemiologic goal was to identify the environmental isolate most closely related to the human isolate and thus the most likely fine-scale geographic source of the infection.

To this end, we examined the relative probability of each mutation (Table 2) using published mutation rate data (1,8). We used mutation rate estimates for specific mutational events to judge relative probabilities of different scenarios. This approach assumes 1) there is an intrinsic mutation rate at each loci for each event, 2) we have accurately estimated these rates (1,8), and 3) we can use intrinsic rates to judge the relative likelihood of >2 hypotheses. We multiplied individual probabilities of mutations within a scenario to calculate the overall relative probability (ORP) that an environmental isolate was related to the infection source (Table 2; Figure 2). To select the most likely source, we compared the ORP of each scenario with the others in a pairwise fashion (odds ratios, Table 2). In practice, only the most likely source needs to be compared with all other sources.

The patients most likely were infected from a source in their yard. Genotype B was observed in isolates from the yard, and this scenario had the highest ORP (7.9 × 10^–9; hypothesis B→A; Table 2). The first scenario for genotype C (C1→A; Table 2) is second most likely (ORP 1.0 × 10^–9). The odds ratio shows the most likely scenario (B→A) is just 7.9× more likely than this scenario (C1→A). These 2 near matches illustrate the power of this approach: one is the most likely source, but the other is statistically possible because this odds ratio difference would not be significant at α<0.05 (odds ratio >20). However, the ORP (1.0 × 10⁴) for the second scenario for genotype C (C2→A; Table 2) would be statistically significant, enabling it to be rejected.

When a high-resolution typing approach based on loci with fast mutation rates, such as MLVA, is used, near matches should be the rule rather than the exception. After transmission, the pathogen will continue to propagate in environmental sources and in the patient, leading to additional mutations before investigators obtain isolates. Mutations may also occur during routine laboratory procedures (e.g., culturing) before genotypic comparisons. Thus, perfect matches are rarely observed during phylogenetic analysis. Rather, the common ancestor (i.e., genotype of the source strain at time of infection) of the human isolate and each potential source isolate will need to be hypothesized. MLVA and probabilistic modeling provide a rigorous means to identify the most likely fine-scale environmental source. The same principles can be applied to other subtyping approaches used in investigations, including those with slower evolution patterns such as PFGE. In these cases, matches and near matches also should be judged by their relative evolutionary rates. Applying evolutionary probabilistic modeling to subtyping will generate stronger conclusions by evaluating the relative strengths of alternative hypotheses regardless of the subtyping approach.