Analysis of false-negative rapid diagnostic tests for symptomatic malaria in the Democratic Republic of the Congo

The majority of Plasmodium falciparum malaria diagnoses in Africa are made using rapid diagnostic tests (RDTs) that detect histidine-rich protein 2. Increasing reports of false-negative RDT results due to parasites with deletions of the pfhrp2 and/or pfhrp3 genes (pfhrp2/3) raise concern about existing malaria diagnostic strategies. We previously identified pfhrp2-negative parasites among asymptomatic children in the Democratic Republic of the Congo (DRC), but their impact on diagnosis of symptomatic malaria is unknown. We performed a cross-sectional study of false-negative RDTs in symptomatic subjects in 2017. Parasites were characterized by microscopy; RDT; pfhrp2/3 genotyping and species-specific PCR assays; a bead-based immunoassay for Plasmodium antigens; and/or whole-genome sequencing. Among 3627 symptomatic subjects, 427 (11.8%) had RDT-/microscopy + results. Parasites from eight (0.2%) samples were initially classified as putative pfhrp2/3 deletions by PCR, but antigen testing and whole-genome sequencing confirmed the presence of intact genes. 56.8% of subjects had PCR-confirmed malaria. Non-falciparum co-infection with P. falciparum was common (13.2%). Agreement between PCR and HRP2-based RDTs was satisfactory (Cohen’s kappa = 0.66) and superior to microscopy (0.33). Symptomatic malaria due to pfhrp2/3-deleted P. falciparum was not observed. Ongoing HRP2-based RDT use is appropriate for the detection of falciparum malaria in the DRC.

Study procedures. Informed consent/assent was obtained from all study subjects prior to enrollment.
Parental permission and informed consent was obtained from parents or legal guardians for all minors younger than 18 years of age, and assent was obtained from all children and adolescents 7-17 years of age. All subjects received malaria RDT testing and treatment according to DRC national guidelines. Subjects underwent a study questionnaire and finger-or heel-prick whole blood collection for diagnostic testing by RDT and microscopy and DBS collection. RDT testing was performed using the World Health Organization-(WHO-) prequalified, HRP2-based SD BIOLINE Malaria Ag P.f. (05FK50, Alere, Waltham, MA) according to manufacturer instructions. Thick-smear microscopy slides were read in the field, and thin smears fixed and transported to the National AIDS Control (PNLS) reference laboratory for confirmation and determination of parasite density. All thin smears were read by two microscopists, with discrepancies resolved by a third reader. Dried blood spot (DBS) samples (Whatmann 903 Protein Saver cards, GE Healthcare Life Sciences, Marlborough, MA) were allowed to air dry at ambient temperature in the field, and stored in individual ziplock bags with desiccant at − 20 °C prior to and after shipment to the University of North Carolina at Chapel Hill for further testing. This study was approved by the Ethical Committee of the Kinshasa School of Public Health (approval number ESP/ CE/07B/2017). Analysis of de-identified samples and data was determined to constitute non-human subjects research by the UNC Institutional Review Board (study number . The study was determined to be non-research by the Centers for Disease Control and Prevention Human Subjects office (0900f3eb81bec92c). Experiments were performed in accordance with relevant guidelines and regulations. Pfhrp2/3 genotyping by PCR. DNA was extracted from DBS samples using Chelex and saponin 18 . All microscopy-positive, RDT-negative samples, in addition to an equal number of microscopy-positive, RDT-positive controls from each province were subjected to quantitative PCR (qPCR) testing targeting the single-copy P. falciparum lactate dehydrogenase (pfldh) gene 19 . Pfhrp2 and pfhrp3 PCR genotyping was performed as previously described 13 , using conventional single-step pfhrp2/3 PCR assays and a qualitative real-time PCR assay targeting the single-copy P. falciparum beta-tubulin (PfBtubulin) gene (Supplementary File) [20][21][22][23] . Only samples with ≥ 40 parasites/µL by qPCR (≥ tenfold higher concentration than the pfhrp2 and pfhrp3 assays' limits of detection) were subjected to pfhrp2 and pfhrp3 PCR to reduce the risk of misclassification of deletions 13 . Microscopy-positive, RDT-positive controls with ≥ 40 parasites/µL by qPCR were randomly selected from the same facility for pfhrp2/3 genotyping. Samples were called pfhrp2/3-negative if they had ≥ 40 parasites/µL by pfldh qPCR, their pfhrp2 and/or pfhrp3 PCR assays were negative in duplicate, and they had successful amplification of PfBtubulin during a final confirmatory assay.
Whole-genome sequencing. All pfhrp2/3-negative samples identified during initial testing were further assessed using whole-genome sequencing. DNA from these samples was enriched for P. falciparum prior to library prep using selective whole-genome amplification (sWGA) as previously described 24 . In brief, two sWGA reactions were performed in parallel, one using a custom primer set designed in our lab (JP9) and another Evaluation for pfhrp2/3 deletions using whole-genome sequencing. Adapter sequences were trimmed from raw, paired sequence reads using trimmomatic and aligned to the P. falciparum 3D7 reference genome (PlasmoDB version 13.0) using bwa mem with default parameters 26,27 . Duplicates were marked and mate-pair information corrected using Picard Tool's MarkDuplicates and FixMateInformation functions, respectively 28 . Candidate indels were identified and realigned using GATK's RealignerTargetCreator and IndelAligner functions, respectively 29 . Genome coverage was calculated using bedtool's genomecov function and visualized using ggplot2 in R (R Core Team, Vienna, Austria) 30 Antigenemia assessment by Luminex. All DBS samples subjected to pfhrp2/3 genotyping by PCR were also assayed for the following Plasmodium antigens: Plasmodium genus-specific aldolase (pAldolase) and lactate dehydrogenase (pLDH), as well as P. falciparum HRP2 by a bead-based multiplex assay as previously described 35 .
Samples were assayed at 1:20 whole-blood concentration after elution from filter paper. Thresholds for antigen positivity for the three targets were determined by assaying 92 blood samples from US resident blood donors without history of international travel and determining the lognormal mean and standard deviation of assay signal from this sample set. The lognormal mean plus three standard deviations of this sample set was used as the antigen positivity threshold ( Supplementary Fig. 1).

Non-falciparum PCR assays.
We used R to randomly select 1000 samples for PCR-based species identification. DNA from these samples was first subjected to a pan-Plasmodium real-time PCR assay targeting the 18S rRNA gene in duplicate 36 . Any sample with at least one positive pan-Plasmodium replicate was subjected to a series of four 18S rRNA real-time PCR assays specific to P. falciparum, Plasmodium malariae, Plasmodium ovale, and Plasmodium vivax in duplicate [37][38][39] . Species calls were only made if at least two total replicates were positive. Samples with only a single positive pan-Plasmodium replicate but negative species-specific assays were called negative. Samples in which both pan-Plasmodium replicates were positive but species-specific assays negative were subjected to a PCR assay specific to the Plasmodium knowlesi Pkr140 gene 40  www.nature.com/scientificreports/ copy-positive malaria, and self-reported malaria diagnosis within the past 6 months were highest in Bas-Uele and lowest in Sud-Kivu.
Pfhrp2/3 deletion genotyping by PCR. We performed pfhrp2/3 genotyping using PCR on a subset of samples, including those collected from all 426 subjects with RDT-/microscopy + results and from 429 RDT + / microscopy + controls selected at random from the same province (Fig. 3). Among the RDT-samples, only 23 had parasite densities sufficient for pfhrp2/3 deletion genotyping by pfldh qPCR (≥ 40 parasites/µL) 13 . We further characterized these samples and 74 RDT-positive controls selected from the same facilities (n = 97 total) using a series of PCR assays for pfhrp2 and pfhrp3, and a final confirmatory PCR assay for PfBtubulin. Eight parasites were PCR-negative for pfhrp2 or pfhrp3 in duplicate despite having parasite densities well above the PCR assays' limits of detection and successful amplification of a second single-copy gene, consistent with pfhrp2/3  Whole-genome sequencing of candidate pfhrp2/3 deletions. However, whole-genome sequencing (WGS) confirmed that all eight putative pfhrp2/3-deleted samples had parasites with intact pfhrp2 and pfhrp3 genes (Fig. 4). All eight samples had at least 5 aligned reads across > 80% of the genome, with median aligned reads ranging from 66-254 reads/position ( Supplementary Fig. 2). Regions of reduced sequencing depth corresponded to differences in the number of histidine repeats compared to the 3D7 reference sequence and did not introduce frame-shift mutations. Mutations in PCR primer binding sites were not observed. F ws values suggested monoclonal infection in only 3 (37.5%) of the 8 samples subjected to whole-genome sequencing (Supplementary Table 3).
Non-falciparum malaria. Non-falciparum malaria is expected to cause HRP2-RDT-negative/microscopypositive results and was common in our study cohort. Among 1000 randomly selected samples that underwent species identification using a series of real-time PCR assays (Fig. 3), malaria was confirmed by PCR in 56.8% of samples, and non-falciparum co-infection with P. falciparum was common (13.2%, n = 75) (Table 3). However, only 1.9% (n = 11) of symptomatic cases were due to non-falciparum infections alone. P. ovale was observed in 11.2% (n = 64) of Plasmodium-PCR-positive symptomatic cases. Among the four (0.8%) symptomatic cases involving P. vivax, half involved P. falciparum and all were low density (< 5 parasites/µL by semi-quantitative 18S rRNA PCR). The majority of symptomatic P. malariae infections (86.9%, n = 20/23) occurred as part of mixed infections with P. falciparum (Supplementary Fig. 4). We were unable to determine the species in 19 samples that were positive by the pan-species 18S PCR assay in duplicate; all had negative P. knowlesi PCR results.   www.nature.com/scientificreports/ RDT performance. Assessment of RDT performance versus PCR suggested that false-negative RDT results in our cohort were commonly caused by RDT failure or operator error rather than parasite factors. Among the random subset of 1000 samples that underwent 18S rRNA testing for all species, 134 (24.9%) of 538 P. falciparum 18S rRNA real-time PCR-positive samples were RDT-negative. RDT performance varied by province, with a larger proportion of RDT-/PCR + results in provinces with higher P. falciparum prevalence by 18S rRNA PCR: Bas-Uele (19%), followed by Sud-Kivu (17%), and finally Kinshasa (5%) (Supplementary Table 4). Only a small proportion of samples (3.6%, n = 36) were RDT + /PCR−, a finding not unexpected and suggestive of persistent PfHRP2 antigenemia after recent clearance of parasitemia 42 . When compared to PCR, RDTs were 75% sensitive and 92% specific, with good agreement (Cohen's kappa = 0.66). Microscopy was 53% sensitive and 81% specific, with fair agreement with PCR (Cohen's kappa = 0.33). Parasite densities as determined by microscopy and pfldh qPCR had moderate correlation (Spearman correlation coefficient = 0.63, p < 0.001, Supplementary Fig. 5).

Discussion
We did not observe symptomatic malaria due to pfhrp2-or pfhrp3-deleted P. falciparum in this large, crosssectional survey across three geographically disparate DRC provinces. The majority of RDT-negative/microscopypositive results occurred in the setting of low or absent parasitemia. This finding implicates parasite densities below the RDT's limit of detection and false-positive microscopy results as the primary causes of RDT-microscopy discordance in the present study. Further assessment of RDT performance using microscopy, genus-and species-specific real-time PCR assays, and Luminex-based antigenemia assessment confirmed that RDT failure and/or user error also caused false-negative RDTs in the present study. However, the overall performance of HRP2-based RDTs was superior to microscopy and in good agreement with PCR. These findings support continued use of HRP2-based RDTs in the DRC. They also contrast with the results of our prior study of asymptomatic children enrolled in the 2013-2014 DHS. There are several possible explanations for these differences. The present study enrolled symptomatic subjects in order to directly inform policy decisions about malaria case management. This study design could have inhibited our ability to identify pfhrp2/3deleted parasites. We and others have proposed the hypothesis that parasites with deletions of the pfhrp2 and/or pfhrp3 genes and their flanking regions may be less fit 6,10,43 , and less likely to cause symptomatic disease. Direct assessment of this hypothesis has not yet been performed in vivo or in vitro, to our knowledge, and is limited by the challenges of confirming deletions in low parasite density infections. However, genetic cross experiments of the 3D7 (wild-type), DD2 (pfhrp2-deleted), and HB3 (pfhrp3-deleted) lab strains did not provide definitive evidence of a fitness cost associated with deletion of either gene 44,45 . In addition, reports from Eritrea confirm that pfhrp2/3-deleted parasites can cause symptomatic and sometimes severe disease 8 .
Exhaustive analysis of putative pfhrp2/3-deleted parasites was needed to discern the status of both genes. The use of rigorous parasite density thresholds well above the downstream pfhrp2/3 PCR assays' limit of detection 13 , confirmation of successful amplification of multiple single-copy genes, and adherence to commonly accepted criteria 5 reduced the risk of inappropriate pfhrp2/3 deletion calls. Only five of 426 (1.2%) RDT-negative/microscopy-positive samples were identified as putative pfhrp2/3 deletions during initial testing. However, we subsequently confirmed HRP2 antigenemia and intact pfhrp2 and pfhrp3 genes in all eight pfhrp2/3-PCR-negative samples using highly sensitive antigen detection methods and WGS, respectively. Pfhrp2/3 sequence variation and resulting changes in the structure of the HRP2 and HRP3 proteins could yield a negative RDT but positive Luminex result, or vice versa, due to differences in the anti-HRP2 monoclonal antibodies employed by both assays. However, previous evaluation of diverse P. falciparum strains failed to identify an association between pfhrp2/3 sequence variation and RDT sensitivity 46 . Taken together, Luminex and WGS results confirmed that the putative pfhrp2/3-deleted parasites were misclassified during initial PCR testing.
These findings emphasize the challenges of confirming pfhrp2/3 gene deletions and support the argument that a portion of pfhrp2/3 deletion calls in our original study of asymptomatic children in the DRC resulted from experimental artifact 10,11 . Even complex laboratory workflows conducted in accordance with commonly used deletion classification criteria are not always sufficient to eliminate the risk of misclassification of pfhrp2/3 deletions. The use of advanced serological and next-generation sequencing methods improved the quality of our pfhrp2/3 deletion assessment, allowed for a more robust evaluation of RDT performance, and enabled visualization of the genetic structure of the pfhrp2 and pfhrp3 genes and their flanking regions. While these methodologies are not widely available in resource-limited settings, they are now accessible through a network of laboratories that collaborate with the World Health Organization to support pfhrp2/3 deletion surveillance 47 and in select locales in sub-Saharan Africa with advanced laboratory capacity.
Symptomatic malaria due to non-falciparum species was common but usually occurred as part of mixed infections with P. falciparum. Although non-falciparum species are not detected by widely deployed HRP2based RDTs, co-infection with P. falciparum is expected to trigger a positive RDT result and prompt treatment with artemisinin-combination therapy according to current DRC guidelines. Therefore, complications due to untreated symptomatic, non-falciparum malaria are likely uncommon, although the risk of relapse by P. vivax or P. ovale without proper diagnosis and terminal prophylaxis remains. Our findings are generally in-line with prior reports of non-falciparum infection among asymptomatic subjects in the DRC [48][49][50] .
Strengths of this study include its geographically diverse sampling locations, robust pipeline of conventional and advanced laboratory methodologies, and relevance to malaria case management. Indeed, these findings directly informed the DRC national malaria control program's decision to continue the use of HRP2-based RDTs, despite evidence of pfhrp2-negative parasites from our initial study of asymptomatic subjects. Our experience in the DRC confirms the importance of basing policy decisions on careful studies of the target population-individuals presenting to health facilities with symptomatic malaria-rather than convenience sampling determined by sample availability or access to study sites. www.nature.com/scientificreports/ Limitations include our inability to discriminate pfhrp2/3-deleted from pfhrp2/3-intact strains in individuals infected by multiple P. falciparum strains. Neither the conventional methods nor the advanced Luminex-based HRP2 antigenemia assessment and WGS methods employed here are well-suited to identify gene deletions in mixed infections. Indeed, assessment of whole-genome sequencing data confirmed mono-infection in only three of eight samples tested. We therefore cannot exclude the possibility that the five remaining samples included mixed infection that involved pfhrp2/3-deleted strains. Recently developed multiplexed qPCR methods 51 and amplicon-based deep sequencing approaches 52 have potential to elucidate pfhrp2/3-deleted minor variants in future large-scale surveys. Second, we restricted our pfhrp2/3 deletion analysis to samples with ≥ 40 parasites/ µL. This requirement was necessary to reduce the risk of misclassification due to DNA concentrations below the pfhrp2/3 PCR assays' limits of detection 13 , but it prevents us from commenting on the prevalence of deletions in lower density infections. Third, enrollment occurred primarily in the rainy season, with only limited dry season enrollment near the end of the study in Bas-Uele. A modeling study predicts that pfhrp2/3-deleted parasite prevalence may be underestimated during the rainy season, when individuals are more likely to be infected by multiple parasite strains 53 . Fourth, we could not evaluate associations between specific symptoms or temperature and RDT performance, as this data was only available for a subset of subjects. Finally, this study was restricted to three provinces. These provinces spanned a range of malaria prevalences, but they do not capture the full diversity of the DRC, which is Africa's second largest country by land mass and neighbors nine other countries.
In conclusion, ongoing HRP2-based RDT use is appropriate in the DRC. False-negative RDT results due to pfhrp2/3 deletions were not observed among symptomatic subjects. Most false-negative results in the DRC are likely due to low parasite densities, RDT failure, or operator error. Only a minority of non-falciparum malaria cases would be missed by an HRP2-based RDT testing strategy; co-infection with P. falciparum and non-falciparum species was common. Careful laboratory workflows are required during pfhrp2/3 gene deletion analyses. Advanced serological and next-generation sequencing approaches can be used to improve the rigor and reproducibility of pfhrp2/3 deletion surveillance efforts and to inform malaria diagnostic testing policy.

Data availability
Genomic sequencing data is available through the Sequence Read Archive (BioSample accession numbers: SAMN16711875-82). Datasets generated during the current study are not publicly available because they contain protected health information but are available from the authors upon reasonable request and with permission of SANRU.