Table 1.
Processes which may contribute to MDR over-representation.
Fig 1.
Host population and strain structure in maintaining coexistence of antibiotic sensitivity and resistance.
Illustration of how host population (panel A) and strain (panel B) structure maintain coexistence by introducing heterogeneity in the fitness effect of resistance and thus creating niches for sensitivity and resistance within the population. Each of the SIS model diagrams represents the resistance dynamics described by Eq (1). A: The resistance dynamics of assortatively mixing host groups can be modelled as independent SIS models by assuming no transmission between groups. Heterogeneity in the fitness effect of resistance arises from between host group differences in antibiotic consumption rate or clearance rate. B: The resistance dynamics of pathogen strains maintained by balancing selection can be modelled as independent SIS models by assuming no recombination. Heterogeneity in the fitness effect of resistance arises from between strain differences in mean duration of carriage (i.e reciprocal of clearance rate).
Fig 2.
Example of a set of resistance profiles from a system with five strata and four antibiotics.
Each row in the table corresponds to the resistance profile of one isolate—i.e. there are three isolates from each strata (equal sampling/size of strata is not necessary). Competitive exclusion within a stratum means all isolates from one stratum have the same resistance profile. The strata have been arranged from top to bottom in order of decreasing resistance proneness (). The antibiotics have been arranged left to right in order of increasing resistance threshold (
), or, equivalently, decreasing resistance frequency. Resistance to a particular antibiotic outcompetes sensitivity in a stratum when the resistance proneness of the stratum is greater than the resistance threshold of the antibiotic. Resistance proneness being independent of antibiotic and resistance threshold being independent of stratum leads to nested resistance profiles (i.e. rarer resistances only observed in the presence of more common ones) and complete linkage disequilibrium between resistances. See Fig A in S1 Text for an example of a set of non-nested resistance profiles.
Fig 3.
Strain frequencies and mean linkage disequilibrium (LD) between resistances in three models with three antibiotics (A, B and C) consumed at different rates.
Left: a model with host population structure (five assortatively mixing host groups) with increasing levels of intergroup transmission. The rate of intergroup transmission on the x-axis (parameter m in the model represented by Eq (10), see Methods) reflects the proportion of transmission events that occur between, instead of within, host group. Middle: a model with strain structure (five strains differing in duration of carriage) with increasing rates of recombination at the duration of carriage locus. Recombination rate on the x-axis (parameter r in the model represented by Eq (11), see Methods) reflects the probability of co-infection, the probability of recombination occurring during co-infection and the probability of the recombinant strain being transmitted. Right: the same model with strain structure (five strains differing in duration of carriage) with increasing rates of recombination at the resistance loci.
Fig 4.
Linkage disequilibrium (measured as D′) between resistances in a two drug model with ten assortatively mixing host groups, as a function of the Spearman correlation between antibiotic consumption rates for the two antibiotics across the ten host groups.
For each drug, five of these host groups consume antibiotics at a high rate (τhigh = 0.075 which selects for resistance when μ = 1) and five consume antibiotics at a low rate (τlow = 0.025 which selects for sensitivity when μ = 1). Consumption rates vary from perfectly anticorrelated (all host groups consuming the first antibiotic at high rate consume the second antibiotic at low rate) to perfectly correlated (all host groups consuming the first antibiotic at a high rate also consume the second antibiotic at a high rate). Left-hand panel: all host groups have the same clearance rate (μ = 1). Middle panel: small variation in clearance rate between host groups (0.5 ≤ μ ≤ 1.5). Each red marker corresponds to one possible configuration of antibiotic consumption and clearance rates (see Methods for details), the black markers represent the average of all possible configurations with the same correlation between antibiotic consumption rates (horizontal jitter is for visualisation purposes only). Right-hand panel: similar to middle panel but with larger variation in clearance rate between host groups (0.25 ≤ μ ≤ 2). Other parameters are β = 2, cβ = 0.95 for both antibiotics, cμ = 1 for both antibiotics.
Fig 5.
Antibiograms (resistance profiles) in the six bacterial datasets.
Dark shading indicates resistance, light shading indicates sensitivity. Antibiograms with a nested structure are coloured red. Each row represents the antibiogram of one isolate (sorted by nestedness and resistance multiplicity). Columns represent antibiotics, ordered by frequency of resistance. A: aztreonam, B: tobramycin, C: cefepime, D: clindamycin, E: erythromycin, F: cefoxitin, G: gentamicin, H: chloramphenicol, I: imipenem, J: piperacillin, K: amikacin, L: ciprofloxacin, M: ampicillin, N: nitrofurantoin, O: oxacillin, P: penicillin, R: rifampin, S: trimethoprim-sulfamethoxazole, T: tetracycline, U: ampicillin-sulbactam, V: levofloxacin, X: ceftriaxone, Z: ceftazidime.
Table 2.
Mean pairwise LD between antibiotic pairs () and proportion of resistance profiles that are nested for six bacterial datasets.