Horizontal gene transfer rate is not the primary determinant of observed antibiotic resistance frequencies in Streptococcus pneumoniae

Abstract

The extent to which evolution is constrained by the rate at which horizontal gene transfer (HGT) allows DNA to move between genetic lineages is an open question, which we address in the context of antibiotic resistance in Streptococcus pneumoniae. We analyze microbiological, genomic, and epidemiological data from the largest-to-date sequenced pneumococcal carriage study in 955 infants from a refugee camp on the Thailand-Myanmar border. Using a unified framework, we simultaneously test prior hypotheses on rates of HGT and a key evolutionary covariate (duration of carriage) as determinants of resistance frequencies. We conclude that in this setting, there is little evidence of HGT playing a major role in determining resistance frequencies. Instead, observed resistance frequencies are best explained as the outcome of selection acting on a pool of variants, irrespective of the rate at which resistance determinants move between genetic lineages.

Fig. 1. Schematic of the two possible explanations for variation in levels of antibiotic resistance between bacterial lineages.

The first hypothesis (“everything is everywhere”; left) is that lineages with a long duration of carriage have high resistance frequencies because resistance is more beneficial to these lineages (because longer duration of carriage translates into a greater probability of antibiotic exposure per carriage episode). The second hypothesis (“genetics as limiting factor”; right) is that lineages with high HGT rate have high resistance frequencies because they acquire resistance determinants at a higher rate. Positive associations have been observed between all three variables (resistance, HGT rate, and duration of carriage). If the first hypothesis is correct, the association between resistance and HGT rate could be confounded by the causal path through the duration of carriage; if the second hypothesis is correct, the association between resistance and duration of carriage could be confounded by the causal path through HGT rate. In the case of a confounding effect, these associations will not be robust to adjusting for the confounded variable.

Fig. 2. Association with resistance multiplicity.

Kendall rank correlation (τ) between SC resistance multiplicity and duration of carriage (unadjusted and adjusted for HR or GM); SC resistance multiplicity and HR (unadjusted and adjusted for duration of carriage); and SC resistance multiplicity and GM (unadjusted and adjusted for duration of carriage).Error bars represent 95% CIs and were computed by bootstrapping (see the Supplementary Materials). SCs with fewer than five episodes of carriage were excluded, giving a sample size of 43 SCs.

Fig. 3. Association with individual resistances.

Kendall rank correlation (τ) between SC resistance to each antibiotic and duration of carriage (unadjusted and adjusted for HR or GM); SC resistance to each antibiotic and HR (unadjusted and adjusted for duration of carriage); and SC resistance to each antibiotic and GM (unadjusted and adjusted for duration of carriage). Opaque error bars represent 95% CIs, while transparent error bars represent 99% CIs; both were computed by bootstrapping (see the Supplementary Materials). SCs with fewer than five episodes of carriage were excluded, giving a sample size of 42 SCs.