| Shared resistance mechanisms |
Particularly relevant for antibiotics of the same class (e.g. β-lactamases and some penicillin-binding protein mutations conferring resistance against multiple β-lactams), but also applicable for some drugs of different classes: there are examples of efflux pumps acting on multiple drugs in numerous species [4] and evidence for clinical relevance of efflux pumps in multidrug resistance [5]. However, MDR over-representation is also observed where efflux pumps are not thought to be a major mechanism of resistance (e.g. between β-lactams and other classes of antibiotics in S. pneumoniae [2]). |
| Linkage between resistance genes (when resistance is associated with particular alleles) |
In general terms, linkage (i.e. being inherited together) between alleles is not a mechanism that generates association between alleles. However, linkage slows down the rate at which recombination breaks down associations between alleles [6], it may therefore play a role in temporarily maintaining associations between resistances. In this context, it is helpful to distinguish between resistance mechanisms where a particular allele of a gene confers resistance (e.g. changes to the protein targeted by the antibiotics) and those where resistance is associated with the presence of a resistance gene (e.g. enzymes that break down the drug). When resistance is associated with specific alleles, there is no a priori reason to expect the resistance allele for one antibiotic to be linked to the resistance, rather than sensitivity, allele of another antibiotic. |
| Linkage between resistance genes (when resistance is associated with presence of gene) |
When resistance is associated with presence of a particular gene, absence of the gene from a particular MGE does not necessarily imply sensitivity to the antibiotic (the gene may be present on another element). As a consequence, spread of the MGE will spread resistance for resistance genes present on the element, but not spread sensitivity when resistance genes are absent. Resistances are therefore more likely to be inherited together than resistance and sensitivity. However, we would still expect recombination and mutation to eventually eliminate resistance determinants which do not confer a fitness advantage, even from mobile genetic elements. For example, in the PMEN1 pneumococcal lineage, there is evidence for loss of aminoglycoside resistance from the Tn916 transposon which encodes tetracycline, and sometimes macrolide, resistance [7]. However, the timescale at which this loss would occur is unclear and there are examples of (presumably) non-advantageous resistance determinants persisting for long time periods [8, 9]. |
| Correlated drug exposure of individual host |
Correlated drug exposure at the individual patient level can arise through use of combination therapy, sequential drug exposure (due to treatment failure with the first drug or prophylactic therapy involving antibiotic cycling), or antibiotic exposure among certain patients being particularly high due to co-morbidities. While combination therapy is rare (in the UK for example, monotherapy accounts for 98% of primary care prescriptions [10]), the other two mechanisms play a substantial role in shaping prescription patterns: prophylaxis and repeat prescriptions make up 31% of primary care prescriptions in England [11] and under 10% of patients account for 50% of adult antibiotic prescriptions [12]. It is unclear, however, whether correlation in antibiotic exposure at the individual level can drive selection for MDR: in absence of assortative mixing between patients, the distribution of antibiotic consumption within a population has little effect, although this result may be sensitive to assumptions about the ecology of the bacteria in question (S1 Text Section 5). |
| Resistance status/risk informs antibiotic choice |
Prescription practices may also contribute to MDR over-representation in another way: the resistance status of an infection or the presence of risk factors for resistance (e.g. travel to certain areas) affect which antibiotic is prescribed. Strains resistant to a particular antibiotic therefore have higher rates of exposure to other antibiotics. The extent to which this mechanism plays a role likely depends on the type of pathogen: for mostly asymptomatic pathogens, the majority of antibiotic exposure arises from prescriptions due to infections with some other pathogen [13], and the resistance status of this pathogen would therefore not affect the choice of antibiotic. |
| Cost epistasis (lower than expected fitness cost when multiple resistance determinants are present) |
There is evidence of cost epistasis between resistance determinants occurring in laboratory competition experiments for some antibiotics (e.g. between streptomycin and rifampicin resistance in Pseudomonas aeruginosa [14] and in E. coli [15]; streptomycin and nalidixic acid resistance in E. coli [16]; and rifampicin and ofloxacin resistance in Mycobacterium smegmatis [17]). Furthermore, for plasmid-associated resistance genes, cost epistasis could also arise if the presence of the plasmid in itself incurs a significant fitness cost (rather than the fitness cost depending on the specific resistance genes it carries). However, the extent to which epistasis plays a role in vivo remains unclear [18]. In particular, we would not, a priori, expect to observe cost epistasis between resistance to antibiotics operating through entirely different mechanisms (e.g. antibiotics targeting protein synthesis and antibiotics targeting the cell wall). |
