The Critical Importance of Sampling Fraction to Inferences of Mycobacterium tuberculosis Transmission

To the Editor—In the study by Manson et al [1], whole-genome sequencing was performed on 223 Mycobacterium tuberculosis isolates from the Tiruvallur and Madurai districts of India. They subsequently examined local strain diversity and mutations associated with phenotypic drug resistance in these regions. In this important study, the authors show that lists of published resistance mutations (including [2–4]) have lower positive predictive values for phenotypic resistance in this context; lineages 1 and 3 predominate in India [1], but these published mutations were largely identified using strains from different M. tuberculosis lineages. This highlights a key obstacle to the implementation of genomics for resistance prediction in India and potentially other endemic regions with diverse lineages of M. tuberculosis. By extension, this study emphasizes the critical need to collect strains and categorize the mutations circulating in these regions to better inform such predictions.

Although we reiterate the importance of this work, we have some considerations that warrant further attention. First, we note that the authors apply a threshold of 10 single nucledotide polymorphisms (SNPs) distance for “recent transmission,” stating this threshold was derived in previous publications. Given this, they then conclude that transmission “was occurring among patients from the same and not different regions.” We would argue that this inference cannot be made from the data available in this study. The sampling fraction, which corresponds to the proportion of total cases included in the study, is an essential (yet often overlooked) consideration in genomic epidemiology. Previous studies have shown that, as sampling fraction decreases, clustering is underestimated [5, 6]. With a low sampling fraction, numerous potential transmission events may be missed due to failure to observe source or secondary cases. When making inferences about transmission in genomic epidemiology (or deciding which inferences should be made), this is therefore a critical consideration [7]. In the Manson et al study, the authors included samples from 196 unique patients from 2 districts of India that were collected over a 6-year period. Because India accounts for >2 million cases of tuberculosis per year [8], the Manson et al study clearly includes only a small proportion of the total cases that would have been diagnosed in this time. We therefore argue that transmission between districts cannot and should not be excluded. To do so not only sends a potentially erroneous message to regional public health units but also risks promoting a “silo effect,” wherein public health officials within regions overlook risk factors for transmission beyond their administrative borders, which may ultimately prove detrimental to tuberculosis control in India and elsewhere.

We would also caution about the general application of SNP thresholds derived from external studies. Although such thresholds are useful from a public health perspective, it is important to note that their sensitivity and specificity for transmission often depends on local strain diversity (eg, [9]) and may not be readily transferrable across settings. We agree that ≤10 SNPs distance does suggest a close genetic relationship; however, it is important to keep in mind that direct person-to-person transmission cannot be ruled in absent more detailed epidemiologic and contextual data.

Notes

Financial support. R. S. L. is supported by a Fellowship from the Canadian Institutes of Health Research (Funding Reference Number 152448). B. P. H. holds a Practitioner Fellowship from the National Health and Medical Research Council, Australia (GNT1105905).

Potential conflicts of interest. Both authors: No reported conflicts of interest. Both authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.