Effect of drug dose and timing of treatment on the emergence of drug resistance in vivo in a malaria model

Abstract Background and objectives There is a significant interest in identifying clinically effective drug treatment regimens that minimize the de novo evolution of antimicrobial resistance in pathogen populations. However, in vivo studies that vary treatment regimens and directly measure drug resistance evolution are rare. Here, we experimentally investigate the role of drug dose and treatment timing on resistance evolution in an animal model. Methodology In a series of experiments, we measured the emergence of atovaquone-resistant mutants of Plasmodium chabaudi in laboratory mice, as a function of dose or timing of treatment (day post-infection) with the antimalarial drug atovaquone. Results The likelihood of high-level resistance emergence increased with atovaquone dose. When varying the timing of treatment, treating either very early or late in infection reduced the risk of resistance. When we varied starting inoculum, resistance was more likely at intermediate inoculum sizes, which correlated with the largest population sizes at time of treatment. Conclusions and implications (i) Higher doses do not always minimize resistance emergence and can promote the emergence of high-level resistance. (ii) Altering treatment timing affects the risk of resistance emergence, likely due to the size of the population at the time of treatment, although we did not test the effect of immunity whose influence may have been important in the case of late treatment. (iii) Finding the ‘right’ dose and ‘right’ time to maximize clinical gains and limit resistance emergence can vary depending on biological context and was non-trivial even in our simplified experiments. Lay summary In a mouse model of malaria, higher drug doses led to increases in drug resistance. The timing of drug treatment also impacted resistance emergence, likely due to the size of the population at the time of treatment.

In almost all cases sequencing of the amplicon in both the forward and reverse direction resulted in identical sequences, however, a minority of samples differed with respect to either the majority genotype represented or in the presence of mixed infections detected. In all main text analyses, we refer to the majority genotype, obtained from Sanger sequencing in the forward direction. In our analyses here, we consider the presence of mixed infections, defined as: infections in which forward and reverse sequences differed and/or and minority peaks were detected in sequencing (at least 25% peak similarity, as detected by GeneiousÒ version 9.1.8 and confirmed via manual inspection).
Mixed infections were common within our relapsing populations. We re-classified sequences from infections into three categories: wildtype (no detection of minority peak alleles), mutant (no detection of minority peak alleles) and mixed (evidence of minority peak alleles); based on both the forward and reverse sequences. Data from our phenotypic measures of resistance suggested that resistance phenotype varied depending on genotype in experiments 1 and 2 (genotype: F9,55 = 6.6, p < 0.001, Supplementary Fig. S12b). We found that resistance phenotype was determined in part by the presence or absence of mixed infections. Wildtype only infections were found to have the lowest slopes of parasite growth in the presence of drug in naïve mice (0.04 +/-0.19, 95% confidence interval), followed by mixed infections (0.58 +/-0.09, 95% confidence interval) and finally mutant only infections (0.62 +/-0.10, 95% confidence interval).
The large spread in phenotypic variance for wildtype genotypes is likely because some of these relapsing populations contained sub-dominant clones. Given the low ability of Sanger sequencing to fully resolve mixed infections, it is likely that our estimations of mixed infections are highly conservative and that drug treatment resulted in relapse with highly diverse pathogen populations that differed in genotype and relative frequency in most cases. Selection here thus occurs as a soft selection sweep involving multiple and independent origins of the same or related alleles that confer drug resistance. Soft sweeps have been theorized to be common under scenarios of high mutation rates and/or high selection coefficients 1,2 and have even been predicted to occur given the complex biology of mitochondrially encoded atovaquone resistance 3 .

SUPPLEMENTARY TABLES
Supplementary Table S1. GLM models of relapse and resistance. Effects of atovaquone dose on relapse and resistance for experiment 1 and 2. Relapse is defined as sustained parasite growth following drug treatment. Resistance is defined as a relapse dominated by mutations in the Qo2 region of the cytb gene (see main text for further details). For experiments 3-5, we tested significance of experimental manipulation on relapse and resistance and then determined whether adding population size accounted for any additional explanatory power via likelihood testing as indicated.

SUPPLEMENTARY FIGURE CAPTIONS
Supplementary Fig S1. Five species alignment of the cytochrome b gene from malaria parasites. Sequences of the full-length cytochrome b gene are shown for a representative atovaquone sensitive (wildtype) strain of P. falciparum and P. vivax (human infecting species; PlasmoDB reference IDs: Pf_M7661101900.1, PVAD80_MIT0003.1, respectively) and P. yoelli, P. chabaudi and P. berghei (rodent infecting species; PlasmoDB reference IDs: PYYM_MIT00900.1, PCHAS_MIT01800.1, PBANKA_MIT01900.1, respectively). Sequences were aligned via Geneious version 9.1.8. Sequence homology is represented above the alignment with green bars, indicating large levels of conservation even among these distantly related species.
Blue arrows indicate positions of our forward and reverse primers used in sequence amplification and sequencing. Red bars indicate known and reported mutations associated with resistance to atovaquone in previously reported in vitro and in vivo studies. Black arrows indicate the quinone binding 2 (Qo2) region of the gene in which high level resistant mutations have been reported. The red box indicates where all the presently reported mutations were located. S2. Estimating population sizes at time of treatment for early treatment groups in experiment 3. We fit a 3-parameter logistic model to data from untreated mice during parasite growth following inoculation of an estimated 100 parasites during days five to 12 post infection. In blue is data from our untreated mice, while red points are estimates for parasite numbers on days when parasites were below the level of detection in our qPCR assay (dotted line), as predicted by model fit. S3. Growth rates of Qo2 mutant (red) and wildtype (blue) parasites in naïve drug-treated mice.