Background
In recent years, experts have introduced several new perspectives on the management of drug resistance in malaria and other infectious diseases. These include criticisms of the aggressive use of therapeutic agents [
1‐
3], the broader encouragement of more responsible use of antimicrobials [
4‐
7] and the exploration of drug cycling strategies [
8‐
13]. Drug stewardship programmes have been successful in several settings, and declines in drug resistance have been observed following changes in antibiotic use [
14,
15]. There are other settings, however, where more careful use of antibiotics was not so effective, microbial populations remaining highly resistant even after removal of drug [
16‐
19], an outcome with serious health and financial consequences. Pathogens might remain resistant to antimicrobials even after their removal for several reasons, among them compensatory mutations at other loci that counteract any fitness cost of drug resistance [
18,
20,
21]. While compensatory mutations at other loci underlie many long-term fixation patterns in clinical infections, it is not fully understood why compensatory mutation is necessary, rather than the evolutionary undoing of mutations that ‘fixed’ in the process of forward resistance evolution.
The lack of a coherent understanding of reverse evolution is partly due to conceptual ambiguity: the term ‘reverse evolution’ is misleading, as it implies directionality in a process (Darwinian evolution) that is near-sighted and agnostic with regard to goal. This has spawned similarly dubious concepts, such as Dollo’s Law, asserting that evolution is intrinsically irreversible [
22] because it would require two independent, low-probability events, occurring along the same pathway, but in opposite order [
23]. Consequently, few studies have examined the molecular pathways through which reverse evolution across an antimicrobial resistance adaptive landscape is likely to occur. One such study of cefotaxime/pipericillin resistance in
Escherichia coli highlighted that epistasis may wire ‘hidden randomness’ into adaptive landscapes that prevents reverse evolution [
24]. A landmark study of reverse evolution in the vertebrate glucocorticoid receptor identified a combination of five mutations, labelled an ‘epistatic ratchet’, that precludes evolution towards the ancestral state [
25]. Studies of this sort are even less frequent as they pertain to the problem of malaria drug resistance, which remains the cause of a global pandemic complicated by widespread resistance [
26].
Approaches utilizing all possible combinations of a suite of mutations associated with resistance can help to resolve the likelihood of adaptive evolution occurring through certain pathways [
27‐
31]. This study uses empirical data from a combinatorial analysis of
Plasmodium vivax dihydrofolate reducatase (DHFR) mutants, evolutionary theory, and individual-based simulations to uncover factors that affect the likelihood of reverse evolution across pyrimethamine (PYR) concentrations. In doing so, it proposes a method for determining whether reverse evolution will occur across an adaptive landscape. By measuring the fitness effects of individual mutations, the study uncovers the existence of a mutational pivot with potentiated genotype-by-environment (G × E) effects that may direct evolution towards or constrain evolution from areas of the landscape with high resistance or fitness. In addition, these mutations attract interactions with other mutation sites, creating an epistatic ratchet, limiting reverse evolution across a landscape. Lastly, the study discusses the implications of these findings for evolutionary theory, molecular epidemiology and in two clinically relevant contexts: (1) the use of existing drugs for resistance management in malaria, and, (2) the rational design of drugs that might target certain amino acid residues of a resistance determinant.
Conclusions
While irreversibility across an adaptive landscape for antimicrobial resistance has been observed in many pathogen types, this question has been relatively unexplored in malarial parasites and in particular, as it pertains to a mechanism underlying this constraint. In the case of malaria, several past studies from the field, involving both chloroquine and pyrimethamine, support the assertion that reverse evolution is improbable: in one instance, a population of
P. falciparum resistant to chloroquine reverted to wild type only after replacement with a migrant population composed of ancestral susceptible genotypes (rather than through
de novo mutation and selection) [
55]. In another, a population of
P. falciparum resistant to pyrimethamine compensated through copy number variation in GTP cyclohydrolase in lieu of reversing the mutations already fixed in DHFR [
56].
Although the study focused on
P. vivax DHFR, it provides a conceptual basis for irreversibility in other resistance proteins. The findings from the field, in combination with these results, imply that modern whole-genome sequencing efforts will reveal a more complex genomic signature of resistance and reversal in settings where antimicrobial use has waxed and waned. This has implications for the practice of molecular epidemiology: while sequencing selected SNPs in resistance determinants might be sufficient for identifying resistance alleles in settings where populations of pathogen are ‘forward’ evolving resistance, the genomic picture is likely more complicated in reverse. The results, corroborated by findings from the field, suggest that re-evolution of the susceptible phenotype (without the growth defects of the more resistant phenotypes) is likely to occur either through the introduction of susceptible migrant genotypes from elsewhere or compensatory mutations at sites other than the ones originally arising during resistance evolution [
57,
58].
This study dissected barriers to reverse evolution from the most resistant genotype (1111) toward the most susceptible (0000) across an adaptive landscape for drug resistance mediated by DHFR in P. vivax. Among four amino acid replacements resulting in pyrimethamine resistance, a single site, S117N (**1*) had a strong effect on the fitness of alleles in the landscape across a breadth of drug concentrations. At high drug concentrations, double and triple mutants containing the S117N mutation, and in particular those in combination with the second site mutation S58R (0110 and 1110, for example) have a reproductive advantage across most drug environments. More specifically, while these higher order alleles have lower fitness than the 0000 ancestral allele in the absence of drug, they have substantially higher fitness than the single mutant neighbours that separate the higher-order mutants carrying the S117N (**1*) mutation (1000, 0100, 0010, 0001), which explains the low likelihood of reverse evolution across this drug resistance adaptive landscape.
Simulations of evolution across the landscape demonstrate the consequences of genotype-by-environment interactions involving S117N: whereas past studies have shown that forward evolution from the 0000 ancestor to an absolute fitness peak occurs readily at drug concentrations greater than 0 [
32], evolution starting from the population fixed for the 1111 quadruple mutant becomes trapped at the 1110 triple mutant local fitness peak, even in the drugless environment (Fig.
4). Even more, when the landscape starts with a population distributed equally between the double mutants (the centre of the landscape; 0011, 0101, 1001, 1010, 0110, 1100), the evolutionary dynamics are still driven by the S117N site (**1*), usually resulting in fixation of the 1110 triple mutant (Fig.
5; Table
1). In this sense, the S117N mutation serves as a pivot point for mutation: its arrival provides a bridge to high fitness areas of the landscape that are trapped onto local peaks through their interaction with other mutations, unable to move to other areas of the landscape.
These findings support the existence of epistatic ratchets that inhibit reverse evolution towards ancestral states, such as that observed in the evolution of the vertebrate glucocorticoid receptor [
25]. While the pivotal S117N mutation creates a ratchet through epistatic interactions, its average effect alone (across all genetic backgrounds and across environments) is larger than that of the other sites (1***, * 1**, ***1), indicating that its fitness effects are not limited to singular genetic backgrounds or certain environments. In this sense, S117N opens evolutionary ‘forks in the road’ towards higher mutation regions of the landscape (double mutants, triple mutants and the quadruple mutant), serving as the starting material for the epistatic ratchets that ultimately prevent reverse evolution.
Other than the implications for molecular epidemiology discussed above, these findings are most relevant to debates surrounding best practices for antimicrobial resistance management. The notion that ceasing use of antimicrobials is a viable strategy for decreasing resistance is based, in part, on the assumption that reverse evolution can occur across a landscape because of the fitness cost of resistance. The results suggest that such strategies may not be generally valid, and they should be tailored to the nature of the actual adaptive landscape and its G × E interactions affecting the accessible trajectories towards resistance and susceptibility. Please note that these comments apply to stepwise reverse evolution in a situation where a derived resistance allele (1111 in this manuscript) is fixed. Alternatively, if a population retains the ancestral allele at low frequency, it can increase in frequency in the absence of drug. This latter scenario is not stepwise evolution, however, but canonical selection on standing genetic variation, a different population genetic context than the one simulated in this study.
Lastly, and most provocatively, the identification of the S117N pivot mutation suggests a possible strategy for identifying targets for chemotherapeutic intervention: if a single mutation is a pivot point to large fitness effects (as found in this study), it might be an ideal drug target. Compounds that perturb the interaction between these pivot residues and others might have a destabilizing effect on protein structure or function, and diminish the evolutionary potential of alleles carrying the resistance determinant. Because this strategy would target the ability of a protein to reach high fitness areas of adaptive landscapes, it would constitute a strategy directed against the evolvability of the pathogen, an unexplored avenue for the treatment of microbial pathogens.