We genotyped the propeller domain of the pfk13 gene by Sanger sequencing 854 samples collected between March 2016 and September 2017 from different malaria-endemic regions of Guyana. Fourteen samples bearing the pfk13 C580Y variant were identified, representing a prevalence of 1.6% (CI₉₅0.8–2.5%). The prevalence of mutants was 8.8% (CI₉₅3.6–14.0%) in Region 1, 0.7% (CI₉₅0.0–2.0%) in Region 8 and 0.5% (CI₉₅0.0–1.1%) in Region 7 (Figure 1, Figure 1—source data 1). We also genotyped the sequence outside of the propeller domain for 283 samples. Among these, 57.2% (CI₉₅51.5–63.0%) were the 3D7 reference genotype, 42.0% (CI₉₅36.3–47.8%) carried the pfk13 K189T mutation and 0.7% (CI₉₅0.0–1.7%) had a mixed genotype (wild-type/K189T). All samples bearing the pfk13 C580Y variant also exhibited the pfk13 K189T mutation, which to date has not been associated with artemisinin resistance.

To understand the origin of the C580Y mutation and examine the genetic relatedness of the mutant strains, we performed whole-genome sequencing (WGS) on thirteen of the fourteen Guyana samples exhibiting the pfk13 C580Y mutation, as well as 40 comparator samples exhibiting wild-type pfk13 and collected at comparable locations and times. The variant profile observed in the sequenced samples identified that the pfk13 C580Y variant arose on a single Guyanese parasite genetic background, and was not imported from Southeast Asia (Figure 2a). Deeper examination of the genomic similarity was performed by estimating the proportion of the genome that was identical-by-descent between sample pairs. This analysis revealed an extremely high level of relatedness between sample pairs bearing pfk13 C580Y (identity by descent (IBD) >0.77) relative to pairwise comparisons containing at least one wild-type pfk13 sample (Figure 2b). This indicates that the parasite lineage on which C580Y arose in Guyana in 2010 probably engaged in limited sexual outcrossing with other parasite lineages in Guyana, despite having persisted for a sufficient duration of time to be observed in multiple regions of the country and to have risen to a non-negligible frequency in the Region 1 population.

In the 75 kb segments flanking the 2016–2017 C580Y-coding variant, WGS identified only two low-quality single nucleotide variants among the 13 mutant samples (Figure 2a). We also analyzed pfk13 C580Y parasites using eight microsatellite loci flanking the pfk13 gene. Despite missing data for some loci in some samples, two different haplotypes were identified, differing at one locus positioned at −6.36 kb (Table 1). The previously identified Guyana A mutant haplotype (Chenet et al., 2016) matched the most common haplotype in the 2016–2017 samples, which was markedly distinct from the one observed in pfk13 Cambodian mutants (Chenet et al., 2016). The second 2016 haplotype exhibiting the 280 allele at the locus −6.36 kb has not been previously identified. As replicate genotyping efforts reproduced the allelic variant at locus −6.36, it is likely that a de novo microsatellite mutation occurred at locus −6.36 following the origin of the C580Y mutation. The high similarity of the microsatellite haplotypes bearing C580Y between 2010 (Chenet et al., 2016) and 2016 suggests that the chromosome 13 resistance haplotype, and perhaps the full clonal lineage, has persisted over this timespan, as several of the allelic markers associated with the C580Y mutation are rare in both studies (e.g. allele 277 at marker −6.36: 11%; allele 206 at marker −0.15: 9%; allele 244 at marker 72.3: 6% Chenet et al., 2016). We were not able to procure remaining DNA from the samples collected in 2010 to determine whether the entire genomic background has been preserved intact as a clonal lineage between 2010 and 2016. However, the genotypes of the molecular markers for resistance (pfcrt, pfdhps, pfdhfr and pfmdr1) were also similar between 2010 and 2016 mutant samples (Supplementary file 1).

The observation of the C580Y mutation persisting in only one clonal lineage suggests that genomic background may be an important determinant of the emergence and persistence of pfk13 propeller mutations in Guyana, as was previously observed in Southeast Asia (Miotto et al., 2013; Miotto et al., 2015; Cerqueira et al., 2017; Amato et al., 2018). We therefore compared the genetic background of 53 isolates from Guyana (40 pfk13 wild-type and 13 pfk13 C580Y) and the artemisinin-resistant background observed in Southeast Asian parasites, in order to explore whether partner mutations associated with pfk13-mutant parasites may be responsible for the preservation of this clonal lineage (Miotto et al., 2015). None of the previously described mutations were observed, including the pffd D193Y, pfcrt N326S, pfcrt I356T, pfarps V127M and pfmdr2 T484I variants. The pfcrt gene was of South American origin with a 7G8 haplotype comprising the mutations C72S, K76T, A220S, N326D and I356L, with no particular difference between wild-type and pfk13 C580Y parasites (Supplementary file 1). However, within this candidate gene set, we observed two fixed differences between wild-type and pfk13 C580Y parasites, both within the gene PF3D7_1252100 (RON3) (Supplementary file 2). Although overall genotyping rates were low in this gene (successful for 15 out of 53), all genotyped pfk13 C580Y mutants (n = 5) contained a V1661L-coding variant and all genotyped pfk13 wild-type parasites (n = 9) carried a synonymous non-reference variant at codon 1801. We also analyzed other molecular markers for resistance (pfcrt, pfdhps, pfdhfr and pfmdr1) (Supplementary file 1). All pfk13 C580Y variants were pfcrt SVMNT (72-76)/A220S/N326D/C350/I356L (i.e. the pfcrt 7G8 haplotype), pfdhfr double-mutant (N51I/S108N), pfdhps triple-mutant (A437G/K540E/A581G) and pfmdr1 triple-mutant (Y184F/N1042D/D1246Y). In summary, Guyanese parasites bearing pfk13 C580Y also exhibited variants in genes previously associated with separate antimalarial drug resistance phenotypes in Southeast Asia.

The highest prevalence of the Guyanese pfk13 C580Y variant was found in Region 1. This variant was first observed in April 2016 and achieved a maximum prevalence of 25.0% (CI₉₅3.8–46.2%) in June 2016 (Figure 3, Figure 3—source data 1). Thereafter, it was not identified between September 2016 and March 2017 despite a constant transmission level of malaria in this region during this period. To understand these fluctuations in prevalence, we studied the parasite population dynamics in the country. We compared parasites from Guyana to samples from Africa and Southeast Asia (Supplementary file 3). At synonymous sites within 4888 genes, pairwise nucleotide diversity (π_syn) in Guyana was 3.4 × 10⁻⁴, less than half the level observed in Southeast Asia (7.4 × 10⁻⁴) and nearly a third of that measured in Africa (1.0 × 10⁻³). In Guyana, 46 samples had sufficient sequencing coverage to assess their complexity of infection (COI), and only one (2.2%) showed evidence of multiple parasite lineages (COI > 1). In contrast, multiclonal infection estimates from Africa and Southeast Asia are routinely higher (Chang et al., 2017; Galinsky et al., 2015; Assefa et al., 2014; Zhu et al., 2019). Despite the low nucleotide diversity and small proportion of multiclonal infections, Guyana harbors a relatively high number of distinct parasite lineages and few parasite pairs with identical genomes. We found that only 0.54% of wild-type Guyanese parasite pairs showed evidence of belonging to the same clonal lineage (IBD >0.75).

To evaluate the impact of the pfk13 C580Y mutation on artemisinin susceptibility in parasites, we culture-adapted two P. falciparum isolates (O141-A and R086) from French Guiana (a neighboring country in the Guiana Shield) and genetically edited these lines using a previously described zinc-finger nuclease (ZFN) based approach (Straimer et al., 2015). We also evaluated the impact of the variant arginine to threonine mutation at codon 539 (R539T), associated with one of the highest in vitro artemisinin resistance levels in Asian parasites (Ariey et al., 2014; Straimer et al., 2015). These two culture-adapted lines both exhibit the pfcrt SVMNT haplotype that is associated with chloroquine resistance, but only R086 carries the pfcrt C350R mutation that was earlier found to restore chloroquine susceptibility to South American parasites despite the presence of the SVMNT haplotype (Pelleau et al., 2015). These parasite lines from French Guiana exhibit a high level of genomic similarity to our genome-sequenced Guyanese parasites (Figures 2c and d; Figure 2—figure supplement 1). These findings suggest that phenotypes observed in these lines from French Guiana are relevant for understanding the impact of the pfk13 C580Y variant in Guyanese parasites, none of which have yet been adapted to in vitro culture.

We produced isogenic lines expressing the wild-type pfk13 allele (R086^ctrl, O141-A^ctrl) or the pfk13 C580Y mutation (R086^C580Y, O141-A^C580Y), or the pfk13 R539T allele in the case of R086 (R086^R539T). We measured the phenotypic impact of those mutations on in vitro artemisinin resistance in edited and parental lines using the Ring-stage Survival Assay (Witkowski et al., 2013) that begins with 0–3 hr post-invasion rings (RSA_0-3h). Introducing the pfk13 C580Y mutation caused a significant increase in the survival rate (twenty-three fold in R086: 0.6% ± 0.2 for R086 to 13.7% ± 1.5 for R086^C580Y, p=0.006; fifty-five fold in O141-A: 0.5% ± 0.1 for O141-A compared to 27.6% ± 7.0 for O141-A^C580Y, p=0.030, Figure 4, Figure 4—source data 1). R086^R539T showed a survival rate fifty-five times higher than the parental line (33.0% ± 5.0 compared to 0.6% ± 0.2, p=0.011). Control assays found no significant difference in the survival rate of the parental vs. isogenic control lines (0.6% ± 0.2 for R086, 0.6% ± 0.4 for R086^ctrlp=0.968, 0.5% ± 0.1 for O141-A and O141-A^ctrlp = 0.899) (Figure 4, Figure 4—source data 1).

To assess the in vitro fitness of the pfk13 mutants relative to the wild-type isogenic lines, we performed a competitive growth assay by co-culturing each pair of mutant and wild-type isogenic parasite lines and measuring the pfk13 allele frequencies over 60 days (~30 asexual generations). This was achieved using a highly sensitive and robust Taqman allelic discrimination real-time PCR (qPCR) assay (Figure 5—figure supplement 1, Figure 5—figure supplement 2, Figure 5—figure supplement 1—source data 1, Figure 5—figure supplement 2—source data 1) that was able to accurately quantify the different proportions of the mutant alleles in these samples. Results showed a modest growth deficit in pfk13 mutants compared to the wild-type parent for both the R086 and O141-A lines, suggesting that the pfk13 C580Y and R539T mutations negatively impact parasite growth. Over the 60 day period, we observed an 18% reduction in the frequency of the R539T mutant for the R086 line, and larger significant reductions of 24% and 40% in allele frequencies of the C580Y mutants of R086 and O141-A compared to their respective isogenic wild-type counterparts (Figure 5a, Figure 5—source data 1). This reflected an average reduced growth rate of −0.6%, −0.8% and −1.3% per 48 hr generation of the R086^R539T, R086^C580Y and O141-A^C580Y mutant lines, respectively, across 30 generations (Figure 5b, Figure 5—source data 1). In addition to observing that the C580Y mutation conferred a higher fitness cost compared to the R539T mutation, we also saw that the parasite genetic background contributed to the severity of the fitness deficit, as the growth defect in the C580Y mutation was substantially more pronounced in O141-A compared with R086 parasites (Figure 5, Figure 5—source data 1).

In collaboration with PAHO, 854 P. falciparum isolates were collected in Guyana from March 2016 to September 2017 in febrile individuals visiting the malaria clinic. Parasites associated with these samples came mainly from Region 1 (n = 114), 7 (n = 572) and 8 (n = 150) (Figure 1, Figure 1—source data 1). Those are the three main endemic regions for malaria in the country. Sample collection was done in patients who provided informed consent. As part of molecular surveillance established in the country, consent from the patients was not required and filter papers were collected at the diagnosis time in different clinics and hospital of the country.

DNA was extracted from dried blood spots using the QIAmp DNA mini kit according to the manufacturer’s protocol (Qiagen, Germany). The pfk13 gene of each sample was amplified by nested PCR using published primers (Ariey et al., 2014; Menard and Ariey, 2013). For the primary PCR, 1 µl of DNA was amplified in a Mastermix containing: 1X of HOT FIREPol Blend Master Mix Ready to Load (12.5 mM MgCl₂, Solis BioDyne), 3.5 mM MgCl₂, and 0.2 µM of each primer to a final volume of 25 µl. The amplification program was: 15 min at 95°C, then 35 cycles of 30 s at 95°C, 2 min at 60°C, 2 min at 72°C, and a final extension of 10 min at 72°C. For the nested PCR, 1 µl of primary PCR product was amplified under the same conditions with the following program: 15 min at 95°C, then 40 cycles of 30 s at 95°C, 1 min at 60°C, 1 min at 72°C, and a final extension of 10 min at 72°C. Nested PCR products were detected using 2% agarose gel electrophoresis and ethidium bromide staining. Double-strand sequencing was performed by Eurofins (France). Sequences were aligned with Geneious v8.1.7 using the 3D7 pfk13 sequence as a reference. Mutant isolates were analyzed twice to confirm results.

The analysis of eight microsatellite loci located upstream (−31.9,–6.36, −3.74,–0.15 kb) and downstream (3.4, 8.6, 15.1, 72.3 kb) of the pfk13 gene within isolates from Guyana was performed as previously described (Chenet et al., 2016; Cheeseman et al., 2012; Talundzic et al., 2015).

Parasites were cultured in human red blood cells in enriched RPMI medium containing 10% human serum and were propagated at 37°C in 10% O₂, 5% CO₂ and 85% N₂. The enriched medium is composed of RPMI-1640 (ref 4130, Sigma Aldrich) with HEPES [25 mM], L-glutamine [5 mM], glucose [22 mM], NaHCO₃5% [25 mM], gentamycin [20 mM], hypoxanthine [0.37 mM] and orotic acid [1.6 µM]. Two P. falciparum isolates collected in French Guiana in 2011 (O141-A) and 2014 (R086) were genetically modified using the ZFN method as previously described (Straimer et al., 2015). Briefly, donor plasmids (pZFN-K13-18/20-hDHFR-bsmut carrying the wild-type pfk13 allele and pZFN-K13-18/20-hDHFR-C580Y and pZFN-K13-18/20-hDHFR-R539T carrying the mutated pfk13 allele) were purified from XL10-Gold bacteria using the QIAGEN plasmid Maxi kit (ref12162) and resuspended in Cytomix. Parasites were electroporated with 50 µg of donor plasmid using the Biorad Gene-PulserII electroporator with settings of 0.31 kV and 950 µF (Fidock and Wellems, 1997). The day after electroporation and for 6 days, parasites were exposed to 2.5 nM WR99210 (a gift from Jacobus Pharmaceuticals, Princeton, NJ). Between 20–35 days after electroporation, parasites were detectable by microscopy. To check plasmid integration into parasites, DNA was extracted from bulk culture with the QIAamp DNA Mini Kit (Qiagen) and was PCR amplified using primers p16 (5’-GCTAATAAGTAATATCAATATAAGGG-3’) and p17 (5’-GGTATTAAATTTTTACCATTCCCATTAGTATTTTGTATAGG-3’). Sequencing was done by Eurofins with the p16 primer and sequences were analyzed using Geneious v8.1.7. Edited parasites were cloned by limiting dilution and selected after 3 weeks by lactate dehydrogenase assay using Malstat reagent (Goodyer and Taraschi, 1997; Adjalley et al., 2010). Selected clones were pfk13 genotyped.

Parental and pfk13-edited parasites were phenotyped using the Ring-stage Survival Assay (RSA_0-3h) method as previously described (Witkowski et al., 2013). RSA_0-3h were performed independently three times for each line. The pfk13 I543T-mutant IPC4912 strain (MRA1241) was used as an artemisinin-resistant control. RSA_0-3h was interpretable if the initial parasitemia was greater than 0.25% and if the growth rate was greater than two-fold per 48 hr. Statistical significance between survival rates of the different lines were calculated using Student’s t-test.

Fitness assays were performed by inoculating an equal number of wild-type and mutant isogenic ring-stage parasites in a 6 ml culture at an initial parasitemia of 1%. Each co-culture experiment was performed in duplicate on two independent occasions and monitored for 60 days. Saponin-lysed parasite pellets of each co-culture were harvested every two days during 60 days and genomic DNA was extracted using the QIAmp DNA mini kit. The percentage of wild-type or mutant allele in each co-culture was determined using genomic DNA in a TaqMan Allelic Discrimination Real-Time PCR Assays. Wilcoxon signed-rank test was applied to test for statistical significance in the observed percentage of the pfk13 mutant allele in each of the individual paired co-cultures against an expected no change in the percentage of the mutant.

Primers (forward and reverse) and TaqMan fluorescence-labeled MGB probes (Eurofins, Germany) for real-time quantitative PCR (qPCR) were designed to specifically detect the pfk13 propeller mutations R539T and C580Y, or the ZFN silent binding mutations, in parasites. The sequences of the forward and reverse primers, and probes (FAM and HEX probe) are shown in Supplementary file 4. We first determined the efficiency and sensitivity of amplifying the pfk13 allele in real-time PCR assays using standard curves comprising 10-fold serially diluted DNA templates ranging from 10 ng to 0.001 ng. Robustness was demonstrated by high efficiency (88–95%) and R² values (0.98–1.00) (Figure 5—figure supplement 1a and b, Figure 5—figure supplement 1—source data 1). Next, we tested the quantitative accuracy in genotype calling by performing multiplex qPCR assays using a pre-defined set of mixtures containing plasmids expressing wild-type to mutant alleles in fixed ratios (0:100, 20:80, 40:60, 50:50, 60:40, 80:20, 100:0). The triplicate points clustered tightly, indicating high reproducibility in the data across the fitted curve (linear polynomial order = 2, R² = 0.92 to 0.94) (Figure 5—figure supplement 1c and d, Figure 5—figure supplement 1—source data 1). Hence, our assay was able to accurately quantify the different proportions of the mutant alleles in these pre-mixed samples. For a single set of samples, we ran the qPCR on two separate occasions to test the consistency of amplification between runs. The results showed excellent correlations (Figure 5—figure supplement 2a, Figure 5—figure supplement 2—source data 1), and thus we concluded that it was not necessary to perform qPCR in replicate runs for each sample. Nevertheless, we always included three replicate qPCR reactions per sample in every run. The data showed consistently high correlations between independent sampling experiments (Figure 5—figure supplement 2b,Figure 5—figure supplement 2—source data 1).

Purified DNA templates were amplified with a species-specific primer set and the corresponding probe. Briefly, the qPCR reactions for every sample were run in triplicates consisting of 1x QuantiFAST reaction mix containing ROX reference dye (Qiagen, Germany), 0.66 µM of forward and reverse primers, 0.16 µM each of the FAM-MGB and HEX-MGB TaqMan probes, and 10 ng of genomic DNA. Amplification and detection of fluorescence was carried out on the QuantStudio 3 (Applied Biosystems, USA) using the genotyping assay mode with the following cycling conditions: 30 s at 60°C, 5 min at 95°C to activate the enzyme, and 40 cycles of 30 s at 95°C, 30 s at 60°C, and 30 s at 60°C and post-read out at 60°C for 30 s. The final optimized assays always included a positive control of plasmid expressing the mutant or wild-type allele in every PCR reaction to ensure the reaction was successful and a no template control of water as a negative control.

The software analyzed the background fluorescence level and calculated background normalized media dye fluorescence (ΔRn) as a function of cycle number for the wild-type or mutant allele. To determine the wild-type or mutant allele frequency in each sample, we first confirmed the presence of the allele by only taking values where the C_t of sample was less than that of the no template control minus three cycles. Next, we subtracted the sample’s ΔRn from the background (control plasmid: absence of wild-type or mutant expressing allele) and normalized to 100% (control plasmid: 100% wild-type or mutant expressing allele) to obtain the percentage of mutant and wild-type allele. Then we calculated the average of (mutant and 100% - wild-type) to derive the percentage of mutant allele.

For the Guyana samples, we performed selective whole-genome amplification (SWGA) on DNA samples as previously described (Oyola et al., 2016) to enrich parasite DNA prior to sequencing on an Illumina HiSeqX instrument at the Broad Institute. We used the enriched DNA to construct Illumina sequencing libraries from the amplified material using the Nextera XT library kit (catalog no. FC-131–1002). We aligned reads to the P. falciparum v3 reference genome assembly using BWA-MEM (Li, 2013) and called SNPs and INDELs using the GATK HaplotypeCaller (Van der Auwera et al., 2013; McKenna et al., 2010; DePristo et al., 2011) according to the best practices for P. falciparum as determined by the Pf3K consortium (https://www.malariagen.net/projects/pf3k). Analyses were limited to the callable segments of the genome (Miles et al., 2016) and excluded sites where over 20% of samples were heterozygous. Additional BAM files for comparative analyses among populations were downloaded from the Pf3k project (release 5; www.malariagen.net/projects/pf3k). For each of four countries in Africa (Democratic Republic of Congo, Ghana, Guinea, and Malawi) and two countries in Southeast Asia (Cambodia and Thailand), 50 samples were chosen based on their high coverage (greatest number of sites with at least 10 × coverage).

We calculated pairwise identity by descent (IBD) using a hidden Markov model (hmmIBD) (Schaffner et al., 2018). Samples with a missing call rate >0.7 were excluded from the analysis. We conducted PCA analyses using R (R Development Core Team, 2016) after removing samples that showed high relatedness (IBD >0.5) to another sample with higher sequencing coverage. Pairwise nucleotide diversity was calculated using a custom Perl script (Source Code File 1, Source Code File 2, Source Code File 3). All genomic analyses were limited to samples containing only a single clonal lineage, as determined by THE REAL McCOIL (Chang et al., 2017) or as identified by the Pf3K consortium (Zhu et al., 2019).