Author response: Modelling the drivers of the spread of Plasmodium falciparum hrp2 gene deletions in sub-Saharan Africa

Abstract

Article Figures and data Abstract eLife digest Introduction Results Discussion Materials and methods Data availability References Decision letter Author response Article and author information Metrics Abstract Rapid diagnostic tests (RDTs) have transformed malaria diagnosis. The most prevalent P. falciparum RDTs detect histidine-rich protein 2 (PfHRP2). However, pfhrp2 gene deletions yielding false-negative RDTs, first reported in South America in 2010, have been confirmed in Africa and Asia. We developed a mathematical model to explore the potential for RDT-led diagnosis to drive selection of pfhrp2-deleted parasites. Low malaria prevalence and high frequencies of people seeking treatment resulted in the greatest selection pressure. Calibrating our model against confirmed pfhrp2-deletions in the Democratic Republic of Congo, we estimate a starting frequency of 6% pfhrp2-deletion prior to RDT introduction. Furthermore, the patterns observed necessitate a degree of selection driven by the introduction of PfHRP2-based RDT-guided treatment. Combining this with parasite prevalence and treatment coverage estimates, we map the model-predicted spread of pfhrp2-deletion, and identify the geographic regions in which surveillance for pfhrp2-deletion should be prioritised. https://doi.org/10.7554/eLife.25008.001 eLife digest Since the turn of the millennium, a large increase in international funding has helped to reduce the public health impact of malaria. The introduction of rapid diagnostic tests has played a central role in these efforts, particularly in remote areas that are heavily affected by the disease. These tests analyse human blood samples for specific proteins that are produced by malaria parasites. The most common rapid diagnostic tests for malaria detect a protein called HRP2, which is produced by the deadliest malaria parasite, Plasmodium falciparum. Recently, however, cases have emerged where the tests have failed to detect these malaria infections. The first occurred in South America, and were found to be because some malaria parasites no longer possessed the gene that produces HRP2. Since then, malaria parasites that lack this gene have been found in several locations in Africa. This raises the question of whether using the tests favours the survival and spread of parasites that cannot produce the HRP2 protein. Using mathematical modelling techniques, Watson et al. now present evidence that suggests that the use of HRP2-detecting rapid diagnostic tests over the past 10 years could have favoured the evolution of malaria parasites that lack this protein. Furthermore, the models suggest that the conditions that are most likely to cause such selection are places where malaria infections are not common but people seek treatment at high rates. Using this information, Watson et al. created a map of 160 locations in Africa most at risk of rapid diagnostic test-driven selection against the gene that produces HRP2. Public health authorities could use these maps to determine where they should more closely monitor malaria parasites to see if they lack this gene. Future genetic investigations will be required in the high-risk areas to confirm and refine the predictions. The development of rapid diagnostic tests that detect other malaria proteins will also be essential if malaria parasites that lack HRP2 continue to spread. https://doi.org/10.7554/eLife.25008.002 Introduction Efforts to control malaria globally have made substantial progress in the last 15 years (World Health Organization, 2015a). This progress reflects the impact made by reinvigorated political commitment that has yielded a twenty-fold increase in international funding for the control and elimination of malaria (World Health Organization, 2015a). The World Health Organisation (WHO) Global Technical Strategy for Malaria 2016–2030 sets ambitious goals to further reduce incidence and mortality rates by 90% by 2030 (World Health Organization, 2015b). Central to achieving these goals is the need to test, treat and track all malaria (World Health Organization, 2010). In sub-Saharan Africa (SSA), diagnostic testing of suspected malaria cases has risen from 36% to 60% between 2005 and 2014 (World Health Organization, 2015a). Microscopy was historically the most common method for diagnosis; however rapid diagnostic tests (RDTs) accounted for 71% of all diagnostic testing of suspected cases in 2014 (World Health Organization, 2015a). The most widely used RDTs target histidine-rich protein 2 (HRP2), which is expressed by the Plasmodium falciparum (Pf) specific gene pfhrp2, with over 85% of RDTs tested in the WHO Foundation for Innovative New Diagnostics (FIND) Malaria RDT Evaluation Programme targeting PfHRP2 (World Health Organization, 2012a). False-negative RDT results due to a partial or complete deletion of the pfhrp2 gene have been reported in areas of South America since 2010, resulting in the recommendation against the use of PfHRP2-based RDTs in these areas (Akinyi et al., 2013; Abdallah et al., 2015; Cheng et al., 2014). These pfhrp2-deleted mutants may still possess a functioning pfhrp3 gene; however, the cross reactivity between PfHRP2-based RDT antibodies and PfHRP3 epitopes is such that a positive result may only occur at very high parasitaemia (Baker et al., 2005). Confirmed pfhrp2-deleted mutants are rarer in Africa, with the first cases reported in Mali in 2012 (Koita et al., 2012). However, recently confirmed occurrences in Ghana, (Amoah et al., 2016) Zambia, (Laban et al., 2015) the Democratic Republic of Congo (DRC), (Parr et al., 2016) Rwanda (Kozycki et al., 2017) and Eritrea (Berhane et al., 2017) (Table 1) have prompted the WHO to host Technical Consultations on pfhrp2/3-deletions and to issue interim guidance for malaria control programs (World Health Organization, 2017; World Health Organization, 2016a; World Health Organization, 2016b). This raises the concern that pfhrp2-deleted mutants may be selected for by RDT-guided treatment decisions – which if confirmed would be one of the first example of selection of a pathogen through diagnostic testing. Table 1 Published studies reporting P. falciparum in Africa with deletions or no deletions of the pfhrp2 gene (Cheng et al., 2014). https://doi.org/10.7554/eLife.25008.003 OriginSource of samples*Initial evidenceGene deletion analysis by PCRAntigen analysisRefPrevalence (no. of samples, year of collection)CountryAreaMicroscopyQuality RDTSpecies PCRpfhrp2 (exon 1 and 2)No. single copy genesFlanking genesHRP ELISA2nd quality RDTMaliBamakoA/SDNDDD1NDNDND(Koita et al., 2012)2% (480, 1996)DRC,Gambia, Kenya, Mozambique, Rwanda, Tanzania, UgandaSDNDDExon 2 onlyNDNDDND(Ramutton et al., 2012)0% (77, 2–19 per country, 2005–2010)SenegalDakarSDNDDD1NDNDND(Wurtz et al., 2013)2.4% (136, 2009–2012)GhanaAccra and Cape CoastADDDExon 2 only2NDNDND(Amoah et al., 2016)29.5% (315, 2015)ZambiaChoma, South ZambiaA/SDDDD1NDNDND(Laban et al., 2015)20% (61, 2009–2012)†DRCCountry-wideADDDD3DNDND(Parr et al., 2016)6.4% (783, 2013–2014)RwandaBusogo, Musanze, KayonzaSDDDExon 2 only1NDNDD(Kozycki et al., 2017)23% (140, 2014–2015)EritreaAnseba, Debub, Gash-Barka, Northern Red-SeaSDDDND1NDNDD(Berhane et al., 2017)80% (51, 2015) *Source of samples: S = Symptomatic case, A = Asymptomatic case, U = not specified, D = done; ND = not done. † Authors suggested that failure to detect pfhrp gene in some samples was more likely to be the result of low parasite density rather than deletion Note: Quality RDT indicates RDTs that meet the WHO RDT recommended procurement criteria based on WHO Malaria RDT Product Testing. Here we use mathematical modelling to characterise the impact of introducing PfHRP2-based RDTs on the emergence and spread of pfhrp2-deleted parasites. We adapt a previously published transmission model (Griffin et al., 2016), incorporating the transmission of pfhrp2-deleted mutants and the contribution of PfHRP3 cross-reactivity to identify settings in which the selective pressure favouring pfhrp2-deleted strains is greatest. In addition, we conduct sensitivity analyses to characterise the influence of assumptions within our model concerning adherence to RDT-guided treatment decisions, the use of microscopy-based diagnostic testing, fitness costs associated with the mutant parasite and the impact of non-malarial fevers upon the selective advantage of pfhrp2 gene deletions. We continue by using a nationally representative cross-sectional study of pfrhp2-deletion in the DRC (Parr et al., 2016) to estimate the prevalence of pfhrp2-deleted mutants prior to RDT introduction. This, in turn, allows us to map predicted geographical regions across SSA where pfhrp2-deletion surveillance should be focused. These mapped predictions are explored across a range of estimates of the prevalence of pfhrp2-deleted mutants prior to RDT introduction. Results Using our newly adapted model incorporating the transmission of pfhrp2-deleted mutants, we first explored the potential for RDT-guided treatment decisions to exert an evolutionary pressure on the prevalence of the mutant. Figure 1 shows the predicted proportion of strains that are pfhrp2-deleted within the population after 10 years. Figure 1 with 5 supplements see all Download asset Open asset Predicted increase in pfhrp2-deletion upon RDT introduction after 10 years. Graphs show the time course of pfhrp2-deletion emergence under (a) different transmission intensities (10%, 25% and 60% PfPR) and 8% starting frequency of pfhrp2-deletion prior to RDT introduction and under (b) different assumed starting frequencies of pfhrp2-deletion prior to RDT introduction (2%, 8% and 12% starting frequency) and 25% PfPR. Five years after RDT introduction, the proportion of strains that are pfhrp2-deleted (c), and the proportion of the population that are infected with only pfhrp2-deleted mutants (d) is recorded. The dark grey dots denote individual simulation runs with a LOESS regression fit shown in blue. Source data for Figure 1 is provided within Figure 1—source data 1. https://doi.org/10.7554/eLife.25008.004 Figure 1—source data 1 Effect of transmission intensity and pfhrp2-deletion starting upon pfhrp2-deletion emergence. https://doi.org/10.7554/eLife.25008.010 Download elife-25008-fig1-data1-v4.csv Within all settings that explored different transmission intensities and starting frequencies of pfhrp2-deletion, RDT introduction is predicted to increase the proportion of pfhrp2-deleted mutants. The strength of selection is predicted to be greatest at low PfPR (Figure 1a); however, a selective pressure is still predicted at both high PfPR and at low starting pfhrp2-deletion frequencies (Figure 1b). The variance in the selection pressure exerted by RDTs is also predicted to be greatest at low PfPR (Figure 1c). A more gradual but analogous trend is predicted in the proportion of the population that were only infected with pfhrp2-deleted mutants (Figure 1d). The prevalence of malaria within Figure 1a was also observed to increase after RDT introduction (Figure 1—figure supplement 1), with the greatest increase in lower transmission settings resulting from untreated infections due to false-negative RDT results. Within the sensitivity analyses, a selective pressure is observed to exist at comparative fitness costs of greater than 90% (see Figure 1—figure supplement 2), however below this the pfhrp2-deletion allele is quickly lost. Both the introduction of additional diagnosis with microscopy-based methods and non-adherence to RDT results decreased the selective pressure, slowing the rate of pfhrp2-deletion emergence (see Figure 1—figure supplement 3). The introduction of non-malarial fevers, however, increased the rate of pfhrp2-deletion emergence (see Figure 1—figure supplement 4), even at 25% below the mean estimated rate of non-malarial fever. When these opposing factors were combined, RDT introduction is still predicted to increase the proportion of pfhrp2-deleted mutants (Figure 1—figure supplement 5). The proportion of clinical cases seeking treatment (assumed here to be treated on the basis of an RDT result) is also predicted to exert a strong selection pressure for pfhrp2-deletion (Figure 2). A consistent relationship was seen across comparable PfPR ranges, with the lowest treatment seeking rates (fT = 0.2) yielding the slowest increase in the proportion of infections due to only pfhrp2-deleted mutants. Again, the lower PfPR categories show the greatest selection pressures for pfhrp2-deletion, with treatment seeking rates >30% and PfPR <25% leading to 20% of infections due to only pfhrp2-deleted mutants in fewer than five years (Figure 2a). Figure 2 with 1 supplement see all Download asset Open asset The predicted rate at which the population is only infected with pfhrp2-deleted mutants. The graphs show the time in years after RDT introduction at which 20% of the population are only infected with pfhrp2-deleted mutants up to a maximum follow-up time of 20 years post RDT introduction. PfHRP3 epitopes were assumed to cause a positive RDT result in (a) 0% or (b) 25% of individuals only infected with pfhrp2-deleted mutants. The plotted years represent the mean time grouped in each prevalence and treatment setting, with black dots representing where 20% was reached in less than five years. Each simulation had a starting pfhrp2-deletion frequency of 8% before RDT introduction. Source data for Figure 2 is provided within Figure 2—source data 1. https://doi.org/10.7554/eLife.25008.011 Figure 2—source data 1 Years after RDT introduction at which 20% of the population are only infected with pfhrp2-deleted parasites, with an assumed PfHRP3 epitope effect equal to 0% and 0.25%. https://doi.org/10.7554/eLife.25008.013 Download elife-25008-fig2-data1-v4.csv The selection pressure favouring pfhrp2-deletions is predicted to be weaker when PfHRP3 epitopes are assumed to cause positive RDT results (Figure 2b). In settings where PfHRP3 epitopes are assumed to cause a positive RDT result in 25% of cases (ε = 0.25), there are four fewer prevalence categories that reach 20% of infections due to only pfhrp2-deleted mutants in fewer than five years. A similar effect is observed in the mean final frequency of pfhrp2-deletion, with 64% frequency recorded after 20 years when no PfHRP3 epitope effect is assumed in comparison to 56% when ε is equal to 0.25 (Figure 2—figure supplement 1). To estimate the starting frequency of pfhrp2-deleted mutants, we used estimates of the proportion of pfhrp2-deleted mutants from a national study in DRC (Parr et al., 2016) to calibrate the model. The calibration incorporated both the PfPR levels and estimates of the treatment rates in the 26 Divisions Provinciales de la Santé (DPS) that would drive selection of the mutant. We estimate a starting frequency of pfhrp2-deleted P. falciparum of 6% in the DRC prior to any introduction of RDTs. The observed relationship between the proportion of infections due to pfhrp2-deleted mutants and PCR PfPR among children 6–59 months of age (Figure 3a) displays a similar trend to the simulations, however with a notably steeper increase at lower prevalence. Of note, the same relationship was not predicted in the absence of selection pressure due to RDT-based treatments (i.e. purely on the basis of the variation in monoclonal infections) (Figure 3b). Figure 3 with 1 supplement see all Download asset Open asset Simulated province level burden of pfhrp2-deleted mutants within the DRC, with an assumed probability of a clinical case seeking treatment, who is only infected with pfhrp2-deleted mutants, producing a positive RDT result (ε) equal to 0.25. In (a) the mean simulated proportion of children aged 6–59 months who are infected with only pfhrp2-deleted mutants is shown in red. Each region had an assumed starting frequency of 6% pfhrp2-deletion prior to RDT introduction in 2010 (2007 in North- and South-Kivu). The results in grey represent the recorded burden from the DHS survey (Figure 3—source data 1), with both datasets fitted with a LOESS regression. Error bars show the 95% confidence interval. In (b) the same simulation conditions were used as in (a) however it is assumed that no selection pressure is exerted by the introduction RDTs, i.e. ε = 1. Source data for Figure 3 is provided within Figure 3—source data 1. https://doi.org/10.7554/eLife.25008.014 Figure 3—source data 1 Estimates of the proportion of pfhrp2-deleted mutants from a national study in DRC. Sourced from Parr JB, Verity R, Doctor SM, Janko M, Carey-Ewend K, Turman BJ, Keeler C, Slater HC, Whitesell AN, Mwandagalirwa K, Ghani AC, Likwela JL, Tshefu AK, Emch M, Juliano JJ, Meshnick SR. 2016. Pfhrp2-deleted Plasmodium falciparum parasites in the Democratic Republic of Congo: A national cross-sectional survey. J Infect Dis: 1–34. doi: 10.1093/infdis/jiw538. Data is provided additionally in an importable format for plotting (Figure 3.csv). https://doi.org/10.7554/eLife.25008.016 Download elife-25008-fig3-data1-v4.csv Figure 3—source data 2 Simulated proportion of children aged 6–59 months who are only infected with pfhrp2-deleted parasites within the Democratic Republic of Congo, with an assumed PfHRP3 epitope effect equal to 0.25% and 1%, that is under no selection pressure. https://doi.org/10.7554/eLife.25008.017 Download elife-25008-fig3-data2-v4.xlsx Finally, using the baseline frequency estimate of 6% prior to RDT introduction, we explored 1000 different prevalence and treatment seeking rates spanning the range of estimates of the PfPR (Bhatt et al., 2015) and treatment levels across sub-Saharan Africa (SSA) in 2010 (Cohen et al., 2012) (Figure 4—figure supplement 2). The model output was aligned with these estimates by first administrative units (Figure 4—figure supplement 1), which enabled us to project the potential increase of the mutant strain and its impact on RDT-guided treatment (Video 1). Our results suggest that 160 of 850 first-administrative regions may have over 20% of all infections due to only pfhrp2-deleted mutants by 2016 (Figure 4c). These areas, which we term of ‘high HRP2 concern’, are largely located in areas where PfPR2-10 in 2010 was less than 25% (Figure 4a). A number of other regions, classified as ‘moderate HRP2 concern’ have high treatment rates, and hence potential selective pressure, despite having comparatively higher transmission (Figure 4b). Our results also illustrate that regions with low transmission may have low HRP2 concern if the frequency of people seeking treatment is very low. Figure 4 with 5 supplements see all Download asset Open asset Predicted areas of HRP2 concern in comparison to recorded prevalence and treatment seeking rate, with an assumed probability of a clinical case seeking treatment, who is only infected with pfhrp2-deleted mutants, producing a positive RDT result (ε) equal to 0.25. The graphs show (a) the recorded malaria prevalence in children aged 2–10 by microscopy in 2010, (b) the frequency of people seeking treatment in 2010 and (c) the predicted concern for the impact of pfhrp2-deleted mutants. In (c), high, moderate and slight risk represent >20% infection due to only pfhrp2-deleted mutants by 2016, 2022 and 2030 respectively, and marginal risk represents <20% by 2030. In 2010, each region was assumed to have a starting frequency of 6% pfhrp2-deletion. Source data for Figure 4 is provided within Figure 4—source data 1. https://doi.org/10.7554/eLife.25008.018 Figure 4—source data 1 Recorded malaria prevalence in children aged 2–10 by microscopy in 2010 (sourced from the Malaria Atlas mapping project [see Metadata - Datasets]), the frequency of people seeking treatment in 2010 (sourced from Cohen et al., 2012 [see Metadata – Datasets]) and the simulated predicted concern for the impact of pfhrp2-deleted mutants, with an assumed PfHRP3 epitope effect equal to 0.25%. High, moderate and slight risk represent >20% infection due to only pfhrp2-deleted mutants by 2016, 2022 and 2030 respectively, and marginal risk represents <20% by 2030. https://doi.org/10.7554/eLife.25008.024 Download elife-25008-fig4-data1-v4.csv Video 1 Download asset This video cannot be played in place because your browser does support HTML5 video. You may still download the video for offline viewing. Download as MPEG-4 Download as WebM Download as Ogg The projected increase in individuals who are only infected with pfhrp2-deleted parasites, from 2010 to 2030, with an assumed starting frequency of 6% pfhrp2-deletion, and an assumed PfHRP3 epitope effect equal to 0.25%. The video relates directly to Figure 4. https://doi.org/10.7554/eLife.25008.025 Discussion Our results demonstrate that the key drivers of pfhrp2-deletion selection are low malaria transmission and a high frequency of people seeking treatment and being correctly treated on the basis of diagnosis with a PfHRP2-based RDT. Based on Africa-wide estimates of parasite prevalence and treatment-seeking behaviour at the time of RDT-introduction, we identified 160 first-administrative units which we classify as ‘high HRP2 concern’. These are areas where the pfhrp2-deleted strain is expected to increase in frequency over a relatively short timescale, and hence where further surveillance efforts should be concentrated. Our results are based on calibration to a large representative survey of malaria across DRC. Due to its size and location in the centre of SSA, the DRC is arguably one of the most representative countries for endemic malaria in Africa. That the model was able to predict the observed relationship in the DRC, despite variability at a province level, provides support for the hypothesis that the variability in pfhrp2-deletion frequency with transmission is driven by selection. However, in contrast to other reported surveys, the samples in this survey were primarily drawn from asymptomatic infections, and hence may not be representative of other reports of pfhrp2-deletion in symptomatic cases with higher parasite density. However, it is interesting to note that our results show broad agreement with published data sets from Zambia (Laban et al., 2015) and Ghana (Amoah et al., 2016) (Table 1). In particular, our predictions confirm that the HRP2 concern would be greater in Ghana than in Southern Zambia. However, one study in Senegal found a lower prevalence of pfhrp2-deletion than we predict (Wurtz et al., A key in the potential spread of pfhrp2-deletion due to selective pressure is the of use and adherence RDT results and the availability of treatment. the one if adherence to RDT results is with who show clinical of malaria in the absence of a positive or additional microscopy-based is used (Figure 1—figure supplement if treatment is not due to or if treatment is not due to or the spread of these deletions will be than the other in areas in which case or in which treatment is for fevers (Figure 1—figure supplement 4), RDT-based treatment may also asymptomatic infections and hence increase the rate of spread of the However, when these with potential fitness costs associated with pfhrp2-deletion, were we still observed an increase in pfhrp2-deletion (Figure 1—figure supplement which a similar rate of increase to that predicted by our model. data on RDT and as as on non-malarial fevers and the fitness of pfhrp2-deletion, however, could to refine mapping of areas of HRP2 A relates to the prevalence of the prior to RDT introduction. is variability in the estimates that have been both before and after RDT introduction, and it is that the of could at a range of However, this variation is the lack of a in reports based on clinical cases and the relatively our results should be not as predictions of the levels of the gene but rather of geographical areas in which surveillance should be focused. these results should not be as predictions of the impact on malaria prevalence as a result of increased gene deletions (Figure 1—figure supplement 1), but of the potential impact of false-negative results upon malaria prevalence and the of diagnostic methods (Figure 1—figure supplement 3). the same further data in the months and years be incorporated to and refine our with any modelling there are a number of we not or any fitness associated with pfhrp2-deletion. a transmission level, locations are likely to have a lower frequency of pfhrp2-deletion in comparison to regions with could however cause substantial which may result in that could resulting in a decreased of or an if it occurred et al., In incorporating a fitness the selection pressure was found to be weaker (Figure 1—figure supplement 2). The fitness despite being is likely as our fitness would cause the strain to be at less than 90% comparative In addition, concerning the role of PfHRP2 a more role in than previously PfHRP2 have been shown to be et al., with more on the recently protein et al., Furthermore, in South America the first cases of pfhrp2-deleted P. falciparum were confirmed prior to the introduction of RDTs et al., 2010). This suggests that these mutants may possess high fitness such that the frequency of pfhrp2-deletion is in the absence of a selective advantage exerted through the use of RDTs. our results on assumptions made the contribution of PfHRP3 epitope cross-reactivity and the potential for RDT results. We found that increased cross-reactivity with PfHRP3 epitopes selection for pfhrp2-deletion and was due to confirmed of PfHRP2-based RDTs PfHRP3 epitopes at high parasitaemia (Baker et al., 2010). In with no epitope the model the in DRC (Figure supplement 1) and a higher estimate of HRP2 concern (Figure 4—figure supplement 1), the same patterns are identified (Figure 4—figure supplement 2). Furthermore, RDT results would the strength of the selection pressure, with infections being However, rates observed within of WHO RDT testing were found to be with the rate on both samples and samples other equal to and the rate on samples factors equal to (World Health Organization, in the absence of introduction we assumed introduction of RDTs in all countries from 2010 in with the WHO recommendation of testing in 2010 (World Health Organization, 2010). The from region to region is more however 2010 is a estimate the reported years at which RDTs were at the level in SSA by the WHO (Figure 4—figure supplement (World Health Organization, However, the of testing microscopy RDT is likely to have decreased over this and hence our estimate of RDT use our model is from is likely The sensitivity of the output to this is in the data from the DRC, in which higher levels of pfhrp2-deletion are observed in an in which RDT introduction likely occurred than in the the starting frequency of pfhrp2-deletion strains from the DRC across the of SSA is a however, in the absence of similar we it provides a first To the of this we also the