Global and local selection acting on the pathogen Stenotrophomonas maltophilia in the human lung

Abstract

Bacterial populations diversify during infection into distinct subpopulations that coexist within the human body. Yet, it is unknown to what extent subpopulations adapt to location-specific selective pressures as they migrate and evolve across space. Here we identify bacterial genes under local and global selection by testing for spatial co-occurrence of adaptive mutations. We sequence 552 genomes of the pathogen Stenotrophomonas maltophilia across 23 sites of the lungs from a patient with cystic fibrosis. We show that although genetically close isolates colocalize in space, distant lineages with distinct phenotypes separated by adaptive mutations spread throughout the lung, suggesting global selective pressures. Yet, for one gene (a distant homologue of the merC gene implicated in metal resistance), mutations arising independently in two lineages colocalize in space, providing evidence for location-specific selection. Our work presents a general framework for understanding how selection acts upon a pathogen that colonizes and evolves across the complex environment of the human body.

Figure 1. Sampling Stenotrophomonas maltophilia populations across explanted CF lungs.

From each lung, we took 4 parasagittal cross-sections along the medial to lateral axis. Each sample was homogenized and plated on MacConkey agar; 24 isolates were randomly selected from 23 tissue samples that exhibited growth. All isolates were whole-genome sequenced and phenotyped. Sample site locations are approximate. Lung figure adapted from original by Patrick J. Lynch and C. Carl Jaffe (goo.gl/iC8AjM), CC-BY-2.5.

Figure 2. Positive selection drove differentiation of S. maltophilia into distinct coexisting lineages within the patient.

(a) Parsimony tree of the population constructed from 334 polymorphic mutations. Most recent common ancestor (MRCA) on the left. The mutations of eight genes with recurrent mutations are indicated on the tree, each gene with its own symbol. Lineages A and S (small colony variant) form the majority of the population. Minor lineages B and C are also indicated. (b) Resistance profile of all isolates against three antibiotics used in treatment, in twofold drug concentrations. Each row is an isolate aligned to its position on the phylogeny. NA, not available; 0, isolate grew only on no drug plates. The differences in resistance between lineages A and S were highly significant: ceftazidime P=4 × 10⁻³⁹, ciprofloxacin P=3 × 10⁻⁴³, tobramycin P=2 × 10⁻³⁴, Kolmogorov–Smirnov test. (c) Lineage-separating mutations are under positive selection (green; dN/dS=1.9, P=0.027, 95% confidence interval (CI) 1.06–3.70), whereas within-lineage mutations are neutral (white; dN/dS=0.93). Error bars indicate 95% CIs.

Figure 3. Mapping the spatial distribution of lineages and estimating the dispersion rate of a mutation.

(a) The approximate location of each tissue sample is labelled on the lung. Each radial line is coloured by the lineage membership of an isolate with the exception of those that were undetermined because of inadequate sequencing coverage (grey). The sputum population is shown at top. Lung figure adapted from original by Patrick J. Lynch and C. Carl Jaffe (goo.gl/iC8AjM), CC-BY-2.5. (b) The observed ratios between lineage A and S across sites are significantly different from the expectation in a well-mixed environment (P<10⁻³, χ² test). (c) We calculated the likelihood η(d) that pairs of isolates separated by d or less SNPs are in the same site (orange line); grey line indicates null model. Shaded error bars indicate 1 s.d. with respect to null model. We inferred a dispersion time of d∼3.

Figure 4. Location-specific selection of merC homologue mutants.

(a) Independently occurring mutants of the distant merC homologue in both lineages A and S colocalize to the same sites (P=0.005, Fisher's exact test). (b) MerC is an inner membrane protein with four transmembrane domains. The three mutational events in the S. maltophilia distant merC homologue were mapped to a protein model (generated via EVfold3132). Mutation in lineage S (P35L) is highlighted in teal, whereas mutations in lineage A (C24Y, L16P) are highlighted in pink. The cysteine proposed to form a disulfide bond with C24 is highlighted in orange (C20); C24Y would disrupt this disulfide bond. All three mutations are found on the same side of the first transmembrane domain.