Several of the methods discussed below rely on interpreting, comparing, or thresholding pairwise genetic distances between isolates to make inferences. It is generally understood that small pairwise distances between isolates implies recent transmission [16–18], but that this method is not entirely accurate due to within-host evolution and variable mutation rates [19, 20]. To identify recent transmission pairs using pairwise distances, investigators must choose a threshold to determine what pairs are considered closely related [16–18]. The threshold for “closely related” depends on the pathogen mutation rate and genome size, and the setting. Within a given time period, a higher mutation rate or a larger genome size will lead to higher pairwise distances. Additionally, the mutation rate of pathogens in outbreak settings is often higher than endemic settings [21]. Thus, for a given pathogen, closely related isolate pairs in an outbreak setting will likely have a higher pairwise genetic distance than closely related pairs in an endemic setting. For this reason, knowledge of the epidemiologic context of the isolates, and the species or sequence type itself, is essential for correctly interpreting pairwise genetic distance distributions. Here, we discuss two methods for identifying reasonable pairwise SNV distance thresholds. Similar methods can be used to identify thresholds for other pairwise genetic distance metrics, such as pairwise patristic distances.

One way to choose a pairwise SNV distance threshold is to use the genome length and mutation rate. For instance, in the context of the dataset we use here, K. pneumoniae ST258 isolates from an endemic setting, we could calculate a pairwise SNV distance threshold of 11 SNVs based on the KPNIH1 reference genome length of 5,394,056 base pairs and a mutation rate of 1.03e-6 mutations per base pair per year [22] (2 isolates * 5,394,056 bases * 1.03e-6 mutations per base per year per isolate). Mutation rates can be estimated for measurably evolving populations with a temporal signal [21–23]. However, mutation rate estimates are less reliable for lineages with high among-lineage rate variation or phylogenetic and temporal clustering [24], and are unreliable for datasets with no temporal signal [24]. Therefore, it is often difficult or impossible to reliably calculate the evolutionary rate of the pathogen for the particular dataset being used, and more general estimates of mutation rate may not be translatable.

If a mutation rate cannot be reliably calculated, or to perform a sensitivity analysis, one way to identify potential pairwise SNV distance thresholds is to use the assumption that pairs of individuals from the same facility will be enriched for recent transmission relative to pairs from different facilities. In particular, this method requires you to plot the fraction of isolate pairs from the same facility for increasing pairwise SNV distances. A decrease in the fraction of intra-facility isolate pairs at a given SNV distance suggests a potentially reasonable threshold. Performing this analysis on our data indicated that using SNV distance thresholds of ≤ 10 and ≤ 6 appear to be reasonable (Figure 2). We chose to use these two thresholds for our sensitivity analysis as they are more directly supported by our data compared to the more general mutation rate analysis using a mutation rate that was not calculated from the same dataset. However, in this case the results would likely be similar as the mutation rate analysis suggests using a threshold of 11 SNVs, which is similar to the 10 SNV threshold we use here.

When performing analyses where choosing a pairwise SNV distance threshold is necessary, we suggest that investigators evaluate the robustness of their results by comparing their findings for different pairwise SNV distance thresholds. Here, we compare the results using a threshold of 6 and 10, but a wider range of values can be used, particularly in instances with more uncertainty about what threshold should be used.