Signatures of transmission in within-host M. tuberculosis variation

Abstract

Background: Because M. tuberculosis evolves slowly, transmission clusters often contain multiple individuals with identical consensus genomes, making it difficult to reconstruct transmission chains. Finding additional sources of shared M. tuberculosis variation could help overcome this problem. Previous studies have reported M. tuberculosis diversity within infected individuals; however, whether within-host variation improves transmission inferences remains unclear.

Methods: To evaluate the transmission information present in within-host M. tuberculosis variation, we re-analyzed publicly available sequence data from three household transmission studies, using household membership as a proxy for transmission linkage between donor-recipient pairs.

Findings: We found moderate levels of minority variation present in M. tuberculosis sequence data from cultured isolates that varied significantly across studies (mean: 6, 7, and 170 minority variants above a 1% minor allele frequency threshold, outside of PE/PPE genes). Isolates from household members shared more minority variants than did isolates from unlinked individuals in the three studies (mean 98 shared minority variants vs. 10; 0.8 vs. 0.2, and 0.7 vs. 0.2, respectively). Shared within-host variation was significantly associated with household membership (OR: 1.51 [1.30,1.71], for one standard deviation increase in shared minority variants). Models that included shared within-host variation improved the accuracy of predicting household membership in all three studies as compared to models without within-host variation (AUC: 0.95 versus 0.92, 0.99 versus 0.95, and 0.93 versus 0.91).

Interpretation: Within-host M. tuberculosis variation persists through culture and could enhance the resolution of transmission inferences. The substantial differences in minority variation recovered across studies highlights the need to optimize approaches to recover and incorporate within-host variation into automated phylogenetic and transmission inference.

Funding: NIAID: 5K01AI173385.

Keywords: M. tuberculosis; epidemiology; genomics; transmission.

Figure 1.. M. tuberculosis consensus genomes are closely related but are not always predictive of epidemiological linkage.

(a) Histograms indicate pairwise genetic distances between M. tuberculosis consensus genomes, with facets indicating study and pairwise comparison type. (b) Phylogeny of consensus sequences for each study, with branch tips colored to indicate samples from a single household or with known epidemiologic links.

Figure 2.. Limited M. tuberculosis within-host diversity is recovered with culture-based Illumina sequencing.

Ridgeline plot of the minor allele frequency distribution for five randomly selected samples from each study, indicated by ridge color. Panels indicate genomic region: outside PE/PPE genes and within PE/PPE genes.

Figure 3.. Pairwise shared variants above a 1% minor allele frequency.

Boxplots of the number of high-quality shared minority variants between sample pairs in three previously published M. tuberculosis transmission studies (columns) with jittered points indicating pairwise observations. Colors indicate comparison type: sample, within-host minority variants; household, minority variants shared between household pairs; unlinked, minority variants shared between individuals in different households. Boxes indicate group interquartile ranges and center lines indicate group medians.

Figure 4.. Shared minority variants contain information about household membership.

(a) Stacked barplot showing the proportion of sample pairs across different levels of shared minority variants ≥ 1% minor allele frequency threshold. Panels indicate study. (b) ROC curves showing sensitivity (true positive rate) as a function 1 – specificity (true negative rate) for predicting household membership in general linear models that include both shared iSNVs and consensus sequence-based clusters (Full model), consensus sequence-based cluster only (Consensus sequences), and Shared iSNVs only (Shared iSNVs). All models include study as a predictor.