A deep sequencing tool for partitioning clearance rates following antimalarial treatment in polyclonal infections

Abstract

Background and objectives: Current tools struggle to detect drug-resistant malaria parasites when infections contain multiple parasite clones, which is the norm in high transmission settings in Africa. Our aim was to develop and apply an approach for detecting resistance that overcomes the challenges of polyclonal infections without requiring a genetic marker for resistance.

Methodology: Clinical samples from patients treated with artemisinin combination therapy were collected from Tanzania and Cambodia. By deeply sequencing a hypervariable locus, we quantified the relative abundance of parasite subpopulations (defined by haplotypes of that locus) within infections and revealed evolutionary dynamics during treatment. Slow clearance is a phenotypic, clinical marker of artemisinin resistance; we analyzed variation in clearance rates within infections by fitting parasite clearance curves to subpopulation data.

Results: In Tanzania, we found substantial variation in clearance rates within individual patients. Some parasite subpopulations cleared as slowly as resistant parasites observed in Cambodia. We evaluated possible explanations for these data, including resistance to drugs. Assuming slow clearance was a stable phenotype of subpopulations, simulations predicted that modest increases in their frequency could substantially increase time to cure.

Conclusions and implications: By characterizing parasite subpopulations within patients, our method can detect rare, slow clearing parasites in vivo whose phenotypic effects would otherwise be masked. Since our approach can be applied to polyclonal infections even when the genetics underlying resistance are unknown, it could aid in monitoring the emergence of artemisinin resistance. Our application to Tanzanian samples uncovers rare subpopulations with worrying phenotypes for closer examination.

Keywords: amplicon sequencing; artemisinin; drug resistance; ecology; malaria; within-host selection.

© The Author(s) 2016. Published by Oxford University Press on behalf of the Foundation for Evolution, Medicine, and Public Health.

Figures

Figure 1.

Clearance curves and determination of ‘clones’ by deep sequencing. Predicted relative abundances (A) and clearance curves (B) for hypothetical infections composed of sensitive (black, 3-h half-life) and resistant (gray, 6.5-h half life) parasite clones after drug treatment. The initial frequency of the resistant clone is 25% in the top row and 1% in the bottom row. In (B), the dotted line shows the standard parasite clearance curve, as fitted to the total parasite density. The half-life estimates from these curves, for the whole infection, are ∼5.5 h (top) and ∼4 h (bottom). When the resistant clone is rare it exerts little effect on the overall estimate of parasite half-life

Figure 2.

Schematic of library preparation for highly multiplexed amplicon deep sequencing. (A) Samples are amplified using primers targeting conserved sequences (black fragments) surrounding variable regions (colored fragments). The forward primer contains a barcode (MID) specific to the sample and replicate. (B) After amplification, each PCR reaction contains amplicons representing the haplotypes within that sample labeled with the specific MID. (C) Multiple samples are then mixed forming a final library containing multiple MIDs (white, blue, purple and brown) for preparation for sequencing. This mixture is library prepped with a specific index placed on the library during the process. Multiple libraries with different indexes can then be pooled and sequenced at the same time allowing for highly multiplexed sequencing

Figure 3.

Deep sequencing of control mixtures. The detected haplotype frequencies of the dilution series are shown, with parasitemias ranging from three genomic equivalents per microliter (GE/µl) to 1536 GE/µl. Each point represents the mean (dot) and 1 SD (error bars) of triplicate experiments. Each experiment involves two PCR replicates used to call haplotypes with a fixed minimum cutoff frequency of 0.5% using SeekDeep. Overall, there is little variation in the frequency estimates within a concentration and between concentrations. Beginning at 24 GE/µl, false-positive haplotypes begin to be detected. Plotted here is the sum total frequency of false haplotypes (red line). Individual samples could contain between 1 and 4 false haplotypes, which individually never exceeded 6% within a single experiment

Figure 4.

Estimating subpopulation clearance curves from infections in Tanzania. (A) Relative abundance of different parasite subpopulations (defined by haplotypes) within patients. For each patient (row), there are two bar graphs representing technical replicates, labeled 1 and 2. Within each patient, individual subpopulations have specific colors and are ranked by initial frequency. (The same color may thus represent different haplotypes in different patients.) (B) Densities of individual subpopulations (points) are calculated as the total density by quantitative PCR (crosses) multiplied by their relative abundance. Clearance curves are plotted for the total parasite density (dashed lines) and individual subpopulations (colored lines). Statistics reported represent model comparisons; P values < 0.05 indicate that a more complicated model that includes different slopes for each subpopulation within a patient explains significantly more variation than a model with a single slope. Similar plots for 12 additional patients for whom parasites could not be detected at 72 h by PCR can be found in the Supplementary Figs S2 and S3

Figure 5.

Estimating subpopulation clearance curves from infections in Cambodia. (A) Relative abundance of different parasite subpopulations (defined by haplotypes) within patients. (B) Densities of individual subpopulations (points) are calculated as the total density by PCR (crosses) multiplied by their relative abundance. Clearance curves are plotted for the total parasite density (dashed lines) and individual subpopulations (colored lines)

Figure 6.

Genetic relationships between parasite haplotypes in one Tanzanian patient. Pseudo minimum spanning tree figure showing the number of nucleotide differences (red dots) between haplotypes (circles) with the minimum number of mismatches and their ties shown to include all haplotypes in the graph for patient T40. Haplotypes are colored as in Fig. 3 (second row from the bottom) and the size of the circle corresponds to the relative abundance of that haplotype at (A) 0 h, (B) 24 h, (C) 48 h and (D) 72 h after the start of drug treatment. Across time points, the total area of the circles is constant

Figure 7.

Predicted effects on clearance time (by PCR) of increasing the initial frequency of the slowest clearing parasite subpopulation in individual infections. (A) The predicted effects on five of the patients from Tanzania in which significant variation in clearance rates was observed. One patient is plotted on its own (B) for visibility. (Note that the simulations assume that all relevant rates remain constant, i.e. there are no changes in pharmacokinetics/pharmacodynamics nor other within-host processes, over the relevant timescale. This is likely to be violated over the span of 100 days but does not affect our inference of having detected some very slow clearing subpopulations.)

Figure 8.

Comparison of clearance slopes of parasites isolated from patients in Tanzania (patient IDs beginning with T) and Cambodia (patient IDs beginning with C). For patients from Tanzania, in which significant variation in clearance rates was observed, the predicted clearance slopes ±1 standard error of individual subpopulations are given in different colors. There was no significant variation in infections from Cambodia. The dashed lines (shaded region) represent the mean (±1 standard error of the mean) clearance slopes of patients from each region, as estimated by fitting linear models to the decline in loge total parasite density