Detection of low-density Plasmodium falciparum infections using amplicon deep sequencing

Abstract

Background: Deep sequencing of targeted genomic regions is becoming a common tool for understanding the dynamics and complexity of Plasmodium infections, but its lower limit of detection is currently unknown. Here, a new amplicon analysis tool, the Parallel Amplicon Sequencing Error Correction (PASEC) pipeline, is used to evaluate the performance of amplicon sequencing on low-density Plasmodium DNA samples. Illumina-based sequencing of two P. falciparum genomic regions (CSP and SERA2) was performed on two types of samples: in vitro DNA mixtures mimicking low-density infections (1-200 genomes/μl) and extracted blood spots from a combination of symptomatic and asymptomatic individuals (44-653,080 parasites/μl). Three additional analysis tools-DADA2, HaplotypR, and SeekDeep-were applied to both datasets and the precision and sensitivity of each tool were evaluated.. Results:Amplicon sequencing can contend with low-density samples, showing reasonable detection accuracy down to a concentration of 5 Plasmodium genomes/μl. Due to increased stochasticity and background noise, however, all four tools showed reduced sensitivity and precision on samples with very low parasitemia (<5 copies/μl) or low read count (<100 reads per amplicon). PASEC could distinguish major from minor haplotypes with an accuracy of 90% in samples with at least 30 Plasmodium genomes/μl, but only 61% at low Plasmodium concentrations (<5 genomes/μl) and 46% at very low read counts (<25 reads per amplicon). The four tools were additionally used on a panel of extracted parasitepositive blood spots from natural malaria infections. While all four identified concordant patterns of complexity of infection (COI) across four sub-Saharan African countries, the COI values obtained for individual samples differed in some cases. Conclusions:Amplicon deep sequencing can be used to determine the complexity and diversity of low-density Plasmodium infections. Despite differences in their approach, four state-of-the-art tools resolved known haplotype mixtures with similar sensitivity and precision. Researchers can therefore choose from multiple robust approaches for analyzing amplicon data, however, error filtration approaches should not be uniformly applied across samples of varying parasitemia. Samples with very low parasitemia and very low read count have higher false positive rates and call for read count thresholds that are higher than current recommendations.

An open dataset of Plasmodium falciparum genome variation in 7,000 worldwide samples

“…84 To be of practical value for national malaria control programmes, genetic data must address well-defined use cases and be readily accessible. 85 Amplicon sequencing technologies provide a powerful new tool for targeted genotyping that could feasibly be implemented locally in malaria-endemic countries 86,87 , but there remains a need for the international malaria control community to generate and share whole genome sequencing data, e.g. to monitor for newly emerging forms of drug resistance and to understand regional patterns of parasite migration.…”

Genotyping cognate Plasmodium falciparum in humans and mosquitoes to estimate onward transmission of asymptomatic infections

“…We performed haplotype inference on mapped reads using DADA2 (version 1.8) as implemented in R (version 3.6.1) 21,73 . These putative haplotypes were then further filtered in order to mitigate the risk of false discovery by removing haplotypes from a sample that met any of the following criteria: (i) supported by <250 reads within the sample, (ii) supported by <3% of the sample's total read depth, (iii) deviation from the expected nucleotide length of 300 for pfama1 or 288 for pfcsp, or (iv) a minority haplotype distinguished by a one single-nucleotide polymorphism difference from another haplotype within the sample that had a read depth >8 times the read depth of the minority haplotype 58 . Finally, we removed haplotypes from the overall population if it was defined by a single variant position that was only variable within that haplotype (see Supplementary Information Figs.…”

“…We only measured transmission directly within households and cannot capture events occurring in other settings; this limitation is mitigated by the known nocturnal feeding preference of local vectors. Finally, many infections in participants and mosquitoes had low parasite densities, which increases the risk of haplotype false discovery 58 . To mitigate this risk, we enforced stringent haplotype censoring based on read quality and haplotype abundance consistent with prior studies [58][59][60] .…”

Implementing parasite genotyping into national surveillance frameworks: feedback from control programmes and researchers in the Asia–Pacific region

“…Amplicon deep sequencing has also been used in P. vivax to compare infection dynamics between day 0 and recurrent infections [ 44 ]. Several analysis tools have been developed to support the analysis of infection complexity using deep sequencing data, including PASEC, DADA2, HaplotypR and SeekDeep [ 43 , 45 – 47 ]. Despite differences in their approach, these four state-of-the-art tools resolved known haplotype mixtures with similar sensitivity and precision [ 47 ].…”