Pre-existing partner-drug resistance to artemisinin combination therapies facilitates the emergence and spread of artemisinin resistance: a consensus modelling study

Abstract

Background: Artemisinin-resistant genotypes of Plasmodium falciparum have now emerged a minimum of six times on three continents despite recommendations that all artemisinins be deployed as artemisinin combination therapies (ACTs). Widespread resistance to the non-artemisinin partner drugs in ACTs has the potential to limit the clinical and resistance benefits provided by combination therapy. We aimed to model and evaluate the long-term effects of high levels of partner-drug resistance on the early emergence of artemisinin-resistant genotypes.

Methods: Using a consensus modelling approach, we used three individual-based mathematical models of Plasmodium falciparum transmission to evaluate the effects of pre-existing partner-drug resistance and ACT deployment on the evolution of artemisinin resistance. Each model simulates 100 000 individuals in a particular transmission setting (malaria prevalence of 1%, 5%, 10%, or 20%) with a daily time step that updates individuals' infection status, treatment status, immunity, genotype-specific parasite densities, and clinical state. We modelled varying access to antimalarial drugs if febrile (coverage of 20%, 40%, or 60%) with one primary ACT used as first-line therapy: dihydroartemisinin-piperaquine (DHA-PPQ), artesunate-amodiaquine (ASAQ), or artemether-lumefantrine (AL). The primary outcome was time until 0·25 580Y allele frequency for artemisinin resistance (the establishment time).

Findings: Higher frequencies of pre-existing partner-drug resistant genotypes lead to earlier establishment of artemisinin resistance. Across all models, a 10-fold increase in the frequency of partner-drug resistance genotypes on average corresponded to loss of artemisinin efficacy 2-12 years earlier. Most reductions in time to artemisinin resistance establishment were observed after an increase in frequency of the partner-drug resistance genotype from 0·0 to 0·10.

Interpretation: Partner-drug resistance in ACTs facilitates the early emergence of artemisinin resistance and is a major public health concern. Higher-grade partner-drug resistance has the largest effect, with piperaquine resistance accelerating the early emergence of artemisinin-resistant alleles the most. Continued investment in molecular surveillance of partner-drug resistant genotypes to guide choice of first-line ACT is paramount.

Funding: Schmidt Science Fellowship in partnership with the Rhodes Trust; Bill & Melinda Gates Foundation; Wellcome Trust.

Figure 1

Artemisinin selection with respect to starting partner-drug resistance frequency In these scenarios, DHA-PPQ is used as the first-line therapy, with 40% population-level drug coverage and 5% malaria prevalence. Results are shown for five different starting piperaquine resistance frequencies (0·00 [green]–0·50 [purple]). The top row shows the median time to three resistance milestones for the five partner-drug resistance scenarios. The middle row shows the fixation pattern of artemisinin-resistant genotypes, characterised by the 580Y allele, with the median shown with solid lines and IQR shown with shaded bands. The bottom row shows the early patterns of emergence for five median simulations, where the median simulation is defined as the one whose time to 0·10 580Y allele frequency is the median time. DHA-PPQ=dihydroartemisinin–piperaquine. MORU=Mahidol-Oxford Research Unit. PSU=Pennsylvania State University.

Figure 2

Number of years until 580Y allele frequency reaches 0·25 in regions with DHA-PPQ deployed as first-line therapy, starting from 0·00 580Y frequency Results for different coverage levels (three columns) and different prevalence levels (four rows) are shown. As the initial genotype frequency of piperaquine resistance increases, the time to 580Y establishment gets shorter. The box (median and IQR) and whisker (95% quantile range from 100 simulations) plots presented are censored box plots, with simulations only being run for 40 years. Values greater than 40 years contribute to the median, IQRs, and 95% ranges if calculable. DHA-PPQ=dihydroartemisinin–piperaquine. MORU=Mahidol-Oxford Research Unit. PSU=Pennsylvania State University.

Figure 3

Median time (T_Y,0·25) for artemisinin-resistant genotypes to reach a frequency of 0·25, under different pre-existing allele frequencies for partner-drug resistance Simulations were evaluated over a 40-year period with median times taking longer than 40 years indicated in grey (40+ years). Times are shown for each model at four different malaria prevalences (1%, 5%, 10%, and 20%) and three different treatment coverages (20%, 40%, and 60%). In all settings and across all models, a decrease in time to 0·25 artemisinin resistance frequency was observed with increasing initial partner-drug resistance. AL=artemether–lumefantrine. ASAQ=artesunate–amodiaquine. DHA-PPQ=dihydroartemisinin–piperaquine. ICL=Imperial College London. MORU=Mahidol-Oxford Research Unit. PSU=Pennsylvania State University.

Figure 4

Comparison of fitness landscapes of DHA-PPQ and AL Network diagrams connect genotypes that are one mutation apart and arrange all genotypes from the wild type (top) to multi-allelic resistant types (bottom); each horizontal level corresponds to one additional genetic mechanism (mutation or copy number variation) of resistance. The genotype is labelled for each node (eg, the wild type at the top, KNY1C1, contains pfcrt K76, pfmdr1 N86 and Y184, one copy of pfmdr1, pfkelch C580, and one copy of the pfpm2 and pfpm3 genes). The 28-day probability of treatment failure for each parasite genotype is indicated by the discrete colour of the nodes in the network for DHA-PPQ (A) and AL (B), which is detailed in the colour bar. Yellow shades show increased resistance (increased probability of treatment failure). Genotypes with the 580Y artemisinin-resistant allele are shown with circles, and genotypes with the wild-type C580 allele are shown with squares. The network highlights the comparatively more complex fitness landscape associated with AL resistance (16 different treatment failure phenotypes) than with DHA-PPQ resistance (4 different phenotypes). AL=artemether–lumefantrine. DHA-PPQ=dihydroartemisinin–piperaquine.