Antimalarial Drug Resistance Profiling of Plasmodium falciparum Infections in Ghana Using Molecular Inversion Probes and Next-Generation Sequencing

A key drawback to monitoring the emergence and spread of antimalarial drug resistance in sub-Saharan Africa is early detection and containment. Next-generation sequencing methods offer the resolution, sensitivity, and scale required to fill this gap by surveilling for molecular markers of drug resistance. We performed targeted sequencing using molecular inversion probes to interrogate five Plasmodium falciparum genes (pfcrt, pfmdr1, pfdhps, pfdhfr, and pfk13) implicated in chloroquine, sulfadoxine-pyrimethamine (SP), and artemisinin resistance in two sites in Ghana.

ABSTRACT

A key drawback to monitoring the emergence and spread of antimalarial drug resistance in sub-Saharan Africa is early detection and containment. Next-generation sequencing methods offer the resolution, sensitivity, and scale required to fill this gap by surveilling for molecular markers of drug resistance. We performed targeted sequencing using molecular inversion probes to interrogate five Plasmodium falciparum genes (pfcrt, pfmdr1, pfdhps, pfdhfr, and pfk13) implicated in chloroquine, sulfadoxine-pyrimethamine (SP), and artemisinin resistance in two sites in Ghana. A total of 803 dried blood spots from children aged between 6 months and 14 years presenting with uncomplicated P. falciparum malaria at the Begoro District Hospital in Begoro and the Ewim Polyclinic in Cape Coast, Ghana, from 2014 to 2017 were prepared on filter paper. Thirteen years after the removal of drug pressure, chloroquine-sensitive parasite strains with pfcrt K76 have increased nearly to fixation in Begoro, in the forest area (prevalence = 95%), but at a lower rate in Cape Coast, in the coastal region (prevalence = 71%, Z = −3.5, P < 0.001). In addition, pfmdr1 184F-bearing parasites are under strong selection. The pfdhfr/pfdhps quadruple genotype (IRNGK), associated with SP resistance, is near saturation. Our study identified at a 2 to 10% prevalence pfdhps 581G, which is a sulfadoxine resistance marker that correlates with the failure of SP prophylaxis in pregnancy and which has not been observed in Ghana. The differences in the reexpansion of chloroquine-sensitive strains observed at the two study sites, the stronger SP resistance, and the high prevalence of pfmdr1 184F should be further monitored to inform malaria control strategies in Ghana.

INTRODUCTION

The scourge of Plasmodium falciparum malaria continues to have a debilitating effect on the world, and sub-Saharan Africa (SSA) still bears the main brunt (1). In 2018, an estimated 228 million malaria cases were reported worldwide, and these numbers have remained at similar levels since 2014, with Ghana contributing about 3% of the global burden (1). Following reports of the spread of resistance to chloroquine in sub-Saharan Africa and trends toward increasing rates of resistance in Ghana, chloroquine was replaced with artesunate-amodiaquine (ASAQ) in 2005 following the recommendations of WHO (2). Subsequent issues of adverse reactions to ASAQ led to the introduction of two more artemisinin combination therapies (ACTs), artemether lumefantrine (AL) and dihydroartemisinin piperaquine (DHAP), as alternative first-line drugs in 2009 (2). The 2014 Ghana Demographic and Health Survey (DHS) (3) and the 2016 Ghana Malaria Indicator Survey (MIS) (4) reported the use of ACTs by 78.2% and 53% of Ghanaian children under age 5 years, respectively. Sulfadoxine-pyrimethamine (SP) has been used for intermittent preventive treatment in pregnancy (IPTp) since 2004 (2), and sulfadoxine-pyrimethamine and amodiaquine (SP-AQ) have been used for targeted seasonal malaria chemoprevention (SMC) in the upper west and upper east regions of Ghana since 2015 (5).

Chemotherapy and chemoprevention have the tendency of engendering drug resistance and subsequently compromising the effectiveness of treatment and thwarting control efforts (6). Over the years, demonstrable resistance to antimalarials that have been used at appreciable levels in clinical settings has been shown (7). Resistance to artemisinin has developed and has risen to significant levels in Southeast Asia (SEA) (8). While artemisinin resistance has not yet been established in the Americas or Africa, it is thought to be only a matter of time until it establishes and spreads in Africa, as has happened with all previous antimalarials (9–11). This calls for preemptive surveillance for the development and spread of resistance in SSA.

Chloroquine pressure exerts a directional selection of pfcrt K76T, the main molecular marker for monitoring chloroquine resistance in field isolates (12). In Africa, the other amino acid substitutions that compensate for the fitness cost on pfcrt K76T form the mutant haplotype CVIET at positions 72 to 76 (13). pfmdr1 N86Y is the major modulator of pfcrt-mediated chloroquine resistance (14, 15). Furthermore, pfmdr1 haplotypes are differentially selected by other antimalarials. For instance, pfmdr1 YYY (86Y, Y184, and 1246Y) was selected for after exposure to ASAQ, while AL selects for the pfmdr1 NFD haplotype (N86, 184F, and D1246) (6, 16). In addition, variations in pfmdr1 copy number (CNVs) have been shown to modulate sensitivity to the partner drugs mefloquine, lumefantrine, and piperaquine (12, 17, 18). The accumulation of point mutations in the dihydrofolate reductase gene (dhfr) (S108N, N51I, and C59R) (19) and the dihydropteroate gene (dhps) (S436A, A437G, K540E, A581G, and S436F) (20) is associated with resistance to pyrimethamine and sulfadoxine, respectively. Resistance studies indicate that tolerance/resistance toward artemisinin may be conferred by a variety of mutations in the propeller domain of the P. falciparum kelch13 gene (pfK13) (21, 22). Recent reports have identified three key mutations in the propeller region of pfK13 (C580Y, R539T, and Y493H) in Southeast Asia (SEA), and these have been confirmed by in vitro studies to mediate a slow parasite clearance time (23). Other low-frequency pfK13 nonsynonymous polymorphisms have been identified in Africa (24–28), but they have not been linked to artemisinin resistance. However, the pfK13 C580Y mutant has independently emerged in Guyana (29), and pfK13 561H has also emerged independently in Rwanda (30).

Monitoring the development and spread of drug resistance in the parasite population is of paramount importance for modeling and optimizing public health and clinical treatment (8, 31). Ghana currently relies on the three WHO-recommended approaches for monitoring drug efficacy, comprising (i) in vivo drug efficacy estimates based on parasite clearance, (ii) in vitro/ex vivo drug efficacy assays, and (iii) genotyping of molecular markers for drug resistance (32). Assessment for molecular markers of resistance provides an excellent complement to in vitro and in vivo drug efficacy testing, but it is more expensive, laborious, and time-consuming than in vitro and in vivo drug efficacy testing (32). However, earlier molecular surveillance methods, like PCR and restriction fragment length polymorphism (RFLP) analysis or Sanger sequencing (33), single-nucleotide primer extension (34), and real-time allele-specific PCR (35, 36), among others, may miss minor alleles, underestimating the prevalence (37), or may inaccurately determine haplotypes in mixed infections. This is important in Ghana, where a large number of infections still contain multiple parasite strains. Also, drug-resistant parasites may occur at a low frequency in individual patient samples before rising to appreciable frequencies under selection, when drugs are used more abundantly in countries where malaria is endemic.

Next-generation sequencing (NGS) is a more sensitive alternative and can accurately capture mixed infection haplotypes and detect minor allele frequencies (MAFs) as low as 1% (38, 39). Here, we leveraged the high throughput, low cost, and specificity of molecular inversion probes (MIPs) (40, 41), combined with deep sequencing (with the Illumina MiSeq platform), to capture known mutations in five drug resistance genes (pfcrt, pfmdr1, pfdhfr, pfdhps, and pfk13) and determined the temporal prevalence of antimalarial resistance markers in two clinical sites with different transmission intensities in Ghana.

By exploring the dynamics of resistance using targeted NGS (TNGS), we were able to detect the emergence of a previously unreported rare mutation in clinical isolates from the two sites in Ghana, which has implications for the early detection and containment of emerging drug-resistant parasites in the population.

RESULTS

Validation and quality control of MIPs.

Testing of the accuracy and the sensitivity of the designed MIP panel was done previously (40) by using a serially diluted mixture of Plasmodium haploid genome copies of each of four laboratory strains and 1 ng human DNA. This method captures the majority of the targets with as few as 29 parasites per microliter and corrects for errors introduced during PCR amplification using the unique molecular identifiers (UMIs) linked to each arm (40). All four laboratory strains (DNA controls) were successfully captured, amplified, and sequenced in all the assays performed.

Sequence data overview.

This study used MIPs designed to target known and putative drug resistance mutations and NGS to sequence 803 clinical P. falciparum isolates, collected over a period of 4 years (2013 to 2017), that were microscopically confirmed to be responsible for monoinfections. After setting a minimum depth per sample per MIP of 10 UMIs, to minimize the missing of mixed infections (complexity of infection [COI] ≥ 2), a total of 721 samples gave useable data. At this cutoff, the average MIP UMI depth for the drug resistance-associated polymorphisms was 410 and the maximum was 22,027 (see Table S2 in the supplemental material). The proportion of samples with each known single nucleotide polymorphism (SNP) with at least the minimum read depth of MIP UMIs ranged from 63% for pfdhfr II64L to 89% for pfcrt mutations at positions 72 to 76 (Table S2). The proportion of mixed infections demonstrating wild-type and mutant alleles ranged from 1% for pfmdr1 D1246Y to 21% for pfmdr1 Y184F and pfdhps S436A. The complexity of infection determined by sequencing pfama1 was 1.27 in Begoro, Ghana, and 1.32 in Cape Coast, Ghana (data not shown).

Assessment and temporal analysis of Plasmodium falciparum antimalarial drug resistance.

The prevalence of known resistance markers was estimated for key mutations in pfcrt, pfmdr1, pfdhfr, and pfdhps and putative markers in the pfk13 gene.

Selection of markers associated with chloroquine resistance.

The prevalences of the targeted gene mutations are listed in Table S3 and Fig. 1. In pfcrt, a steady decline in chloroquine resistance mutant allele pfcrt 76T was observed, and hence, the haplotype CIET also declined in both sites over the study period. In Begoro, virtually all parasite isolates had reverted to the wild type by 2017 (95% wild-type pfcrt K76 and haplotype CMNK), while 29% of the Cape Coast patients’ parasites still carried chloroquine resistance mutations in 2017. The prevalences of other mutations, such as pfcrt 220S, also declined in the two sites, with a greater decline being seen in Begoro than in Cape Coast. Mutations in pfmdr1, on the other hand, did not change significantly over time. At both sites, pfmdr1 86Y remained at a low prevalence, while pfmdr1 184F remained at a high prevalence. The frequencies of these markers were slightly lower than the expected prevalences, and the values and trends were similar at both sites (Fig. 2).

FIG 1.

Prevalence of point mutations associated with P. falciparum antimalarial drug resistance in Begoro and Cape Coast between 2014 and 2017. The prevalence of resistance-associated mutations in pfcrt, pfmdr1, pfdhps, and pfdhfr is shown. Prevalence (Prev) was calculated by dividing the number of samples containing at least one mutant allele by the total number of samples genotyped each year.

FIG 2.

Frequency of molecular markers associated with antimalarial drug resistance in Begoro and Cape Coast between 2014 and 2017. Allele frequencies were estimated for resistance-associated mutations in pfcrt, pfmdr1, pfdhps, and pfdhfr and were estimated as the mean for all mutation fractions per sample. The allele fraction was estimated for each sample, and the population frequency of each resistant mutant was estimated as the average allele fraction for that population.

The pfmdr1 double mutant haplotype 86Y 184F , which is associated with chloroquine treatment failure, was maintained at a low prevalence, ranging from 10% to 14% and 4% to 18% in Begoro and Cape Coast, respectively, during the study period (Table 1).

TABLE 1.

Prevalence of haplotypes associated with P. falciparum antimalarial drug resistance in the two study sites in Ghana

Gene (chra) codons and drug(s) to which resistance is associated	Mutation	Genotypeb	Base call	2014			2015			2016			2017
				Prevalence (%)c		P value	Prevalence (%)		P value	Prevalence (%)		P value	Prevalence (%)		P value
				Begoro	Cape Coast		Begoro	Cape Coast		Begoro	Cape Coast		Begoro	Cape Coast
pfcrt (chr7) codons 72/74/75/76, chloroquine, amodiaquine	Wild type	C-M-N-K	TATA	82 (69/74)	62 (48/78)	0.006	91 (84/92)	79 (79/100)	0.029	92 (76/84)	89 (64/72)	0.772	98(54/55)	89 (47/53)	0.058
	Triple	C-I-E-T	TGAC	18 (15/74)	39 (30/78)		9 (8/92)	21 (21/100)		8 (7/84)	11 (8/72)		2 (1/55)	11 (6/53)
pmdr1 (chr5) codons 86/184, chloroquine	Wild type	NY	AA	28 (6/57)	24 (15/63)	0.37	21 (12/58)	42 (28/66)	0.009	37 (17/46)	23 (14/62)	0.051	31 (11/37)	38 (17/45)	0.779
	Single	Y-Y	TA	2 (1/57)	0 (0/63)		0 (0/58)	2 (1/66)		0 (0/46)	2 (1/62)		0 (0/37)	0 (0/45)
	Single	N-F	AT	61 (35/57)	59 (37/63)		70 (41/58)	55 (36/66)		59 (27/46)	76 (47/62)		61 (22/37)	55 (25/45)
	Double	Y-F	TT	9 (5/57)	17 (11/63)		9 (5/58)	2 (1/66)		4 (2/46)	0 (0/62)		8 (3/37)	7 (3/45)
pmdr1 (chr5) codons 86/184/1246, lumefantrine, amodiaquine	Wild type	N-Y-D	AAG	30 (16/53)	22 (13/58)	0.302	21 (11/53)	43 (27/63)	0.024	35 (13/37)	23 (14/62)	0.082	34 (11/42)	34 (15/43)	0.759
	Single	Y-Y-D	TAG	2 (1/53)	0 (0/58)		0 (0/53)	2 (1/63)		0 (0/37)	0 (0/62)		0 (0/42)	0 (0/43)
	Single	N-F-D	ATG	58 (31/53)	57 (34/58)		70 (37/53)	52 (33/63)		59 (22/37)	76 (47/62)		53 (17/42)	58 (25/43)
	Double	Y-F-D	TTG	9 (5/53)	19 (11/58)		9 (8/53)	2 (1/63)		5 (2/37)	0 (0/62)		9 (3/42)	7 (3/43)
	Double	N-F-Y	ATT	0 (0/53)	0 (0/58)		2 (1/53)	2 (1/63)		0 (0/37)	0 (0/62)		3 (1/42)	0 (0/43)
	Double	Y-Y-Y	TAT	0 (0/53)	0 (0/58)		0 (0/53)	0 (0/63)		0 (0/37)	2 (1/62)		0 (0/42)	0 (0/43)
pfdhfr (chr4) codons 51/59/108, pyrimethamine	Wild type	N-C-S	ATG	0 (0/78)	0 (0/76)	0.234	6 (5/79)	2 (2/80)	0.61	5 (3/65)	0 (0/74)	0.338	0 (0/48)	5 (3/61)	0.221
	Double	N-R-N	ACA	19 (15/78)	11 (8/76)		11 (9/79)	13 (12/80)		6 (4/65)	5 (4/74)		15 (7/48)	8 (5/61)
	Double	I-C-N	TTA	1 (1/78)	4 (3/76)		1 (1/79)	2 (2/80)		2 (1/65)	1 (1/74)		0 (0/48)	3 (2/61)
	Triple	I-R-N	TCA	78 (62/78)	86 (65/76)		81 (64/79)	82 (74/80)		88 (57/65)	93 (69/74)		85 (41/48)	84 (51/61)
pfdhps (chr8) codons 437/540, sulfadoxine	Wild type	A-K	CA	0 (0/65)	4 (3/70)	0.122	6 (4/66)	5 (4/73)	1	0 (0/58)	1 (1/71)	0.501	3 (1/38)	0 (0/49)	0.686
	Single	G-K	GA	100 (65/65)	93 (65/70)		92 (61/66)	92 (67/73)		100 (58/58)	96 (68/71)		97 (37/38)	98 (48/49)
	Single	A-E	CG	0 (0/65)	0 (0/70)		0 (0/66)	0 (0/73)		0 (0/58)	0 (0/71)		0 (0/38)	0 (0/49)
	Single	G-E	GG	0 (0/65)	3 (2/70)		2 (1/66)	3 (2/73)		0 (0/58)	3 (2/71)		0 (0/38)	2 (1/49)
pfdhps (chr8) codons 436/437/540/581/613, sulfadoxine	Wild type	S-A-K-A-A	GCACG	0 (0/52)	0 (0/61)	0.044	0 (0/54)	4 (2/54)	0.58	0 (0/39)	0 (0/58)	0.416	0 (0/28)	0 (0/32	0.2199
	Single	S-G-K-A-A	GGACG	48 (25/52)	66 (37/61)		50 (27/54)	57 (31/54)		41 (16/39)	55 (32/58)		60 (17/28)	78 (25/32)
	Single	S-A-K-A-S	GCACT	0 (0/52)	2 (1/61)		0 (0/54)	1 (1/54)		0 (0/39)	0 (0/58)		0 (0/28)	0 (0/32)
	Single	A-A-K-A-A	TCACG	0 (0/52)	3 (2/61)		7 (4/54)	4 (2/54)		0 (0/39)	2 (1/58)		4 (1/28)	0 (0/32)
	Double	A-G-K-A-A	TGACG	42 (22/52)	29 (18/61)		31 (17/54)	22 (12/54)		46 (18/39)	29 (17/58)		21 (6/28)	16 (5/32)
	Double	S-G-E-A-A	GGGCG	0 (0/52)	3 (2/61)		2 (1/54)	4 (2/54)		0 (0/39)	3 (2/58)		0 (0/28)	3 (1/32)
	Triple	A-G-K-A-S	GGAGT	2 (1/52)	3 (2/61)		6 (3/54)	6 (3/54)		10 (4/39)	7 (4/58)		14 (4/28)	3 (1/32)
	Quadruple	A-G-K-G-S	GGACT	8 (4/52)	0 (0/61)		4 (2/54)	4 (2/54)		3 (1/39)	3 (2/58)		0 (0/28	0 (0/32)
pfdhfr/pfdhps (chr4/8) codons 51/59/108/437/540, sulfadoxine-pyrimethamine	Wild type	N-C-S-A-K	ATGGA	0 (0/57)	0 (0/58)	0.186	0 (0/60)	2 (1/60)	0.242	0 (0/46)	0 (0/64)	0.151	0 (0/30)	0 (0/44)	0.332
	Single	N-C-S-G-K	ATGCA	0 (0/57)	0 (0/58)		7 (4/60)	0 (0/60)		7 (3/46)	0 (0/64)		0 (0/30)	2 (1/44)
	Double	N-R-N-A-K	ACAGA	0 (0/57)	2 (1/58)		3 (2/60)	15 (9/60)		0 (0/46)	0 (0/64)		0 (0/30)	0 (0/44)
	Double	N-C-S-G-E	ATGCG	0 (0/57)	0 (0/58)		0 (0/60)	0 (0/60)		0 (0/46)	0 (0/64)		0 (0/30)	0 (0/44)
	Triple	N-R-N-G-K	ACACA	23 (13/57)	12 (7/58)		8 (5/60)	8 (5/60)		9 (4/46)	3 (2/64)		17 (5/30)	7 (3/44)
	Triple	I-C-N-G-K	TTACA	2 (1/57)	3 (2/58)		0 (0/60)	3 (2/60)		2 (1/46)	2 (1/64)		0 (0/30)	5 (2/44)
	Triple	I-R-N-A-K	TCAGA	0 (0/57)	3 (2/58)		3 (2/60)	3 (2/60)		0 (0/46)	2 (1/64)		3 (1/30)	0 (0/44)
	Quadruple	I-R-N-G-K	TCACA	75 (43/57)	76 (44/58)		77 (46/60)	72 (43/60)		83 (38/46)	91 (58/64)		80 (24/30)	84 (37/44)
	Quadruple	I-R-N-A-E	TCAGG	0 (0/57)	0 (0/58)		0 (0/60)	0 (0/60)		0 (0/46)	0 (0/64)		0 (0/30)	0 (0/44)
	Quadruple	N-R-N-G-E	ACACG	0 (0/57)	0 (0/58)		0 (0/60)	0 (0/60)		0 (0/46)	2 (1/64)		0 (0/30)	0 (0/44)
	Quintuple	I-R-N-G-E	TCACG	0 (0/57)	3 (2/58)		2 (1/60)	3 (2/60)		0 (0/46)	2 (1/64)		0 (0/30)	2 (1/44)

^achr, chromosome number.

^bBoldface amino acids indicate the mutated amino acids.

^cThe data in parentheses represent the number of individuals with the indicated haplotype/total number of individuals tested.

The trends of prevalence over a period of 4 years are shown for each study site in Fig. 1 and Table S4. In addition, the population frequencies of these resistance markers were also examined (Fig. 2). Overall, there was a temporal decline in the prevalence of both pfcrt 76T (χ² = 16, P = 0.0001) and pfmdr1 86Y (χ² = 6.6, P = 0.0099) between 2014 and 2017, although there was a rise in the prevalence of pfcrt 76T in Begoro in 2016 and of pfmdr1 86Y at both sites in 2017. While the prevalence of pfmdr1 184F remained high at both sites, no trend was observed during the study period. There was no statistically significant trend in the other SNPs associated with P. falciparum drug resistance over the 4-year period of the study (Table S4).

Haplotypic analysis of alleles conferring resistance to sulfadoxine and pyrimethamine.

Table 1 and Fig. 1 describe the prevalence of mutant alleles and haplotypes associated with resistance to the antifolate drugs sulfadoxine and pyrimethamine in the two study sites in Ghana. Three of the pfdhfr mutations associated with pyrimethamine resistance (51I, 59R, and 108N) were observed to be nearing fixation in the study sites. The pfdhfr 164L mutation was not observed. The sulfadoxine resistance-conferring pfdhps mutation 437G was also near fixation. However, pfdhps 540E, which is also a key mutation associated with sulfadoxine resistance, remained at a low prevalence, ranging from 1% to 7% across both sites. Other pfdhps mutant alleles, namely, 436A, 581G, and 613S, were also present at frequencies that may influence the efficacy of sulfadoxine-pyrimethamine. The prevalence of the pfdhps mutant 436A ranged from 55% to 65% in Cape Coast and from 44% to 54% in Begoro. The 581G and 613S mutant alleles were observed in less than 11% and 20% of the parasite isolates from Cape Coast and Begoro, respectively. This study also showed the presence of triple mutant pfdhps alleles AGKAS (at 3 to 14% from 2014 to 2017) and, for the first time in these sites, the quadruple mutant alleles AGKGS (3%) in 2016 (Table 1). The prevalence of the combined pfdhfr/pfdhps quadruple mutation (IRNGK), conferring resistance to sulfadoxine-pyrimethamine, ranged from 75% to 83% in Begoro and 72% to 91% in the Cape Coast. The prevalence of the quintuple mutant, IRNGE, which is common in East Africa, was very low due to the low prevalence of 540E observed in both our sites (Table 1).

Comparison of drug resistance mutations with those from a previous study in Ghana.

From 2003 to 2010, Duah and colleagues genotyped pfcrt, pfmdr1, phdhfr, and pfdhps for the prevalence of chloroquine-resistant parasite isolates in samples collected from the same geographical areas evaluated in the present study (84). Table 2 compares the pfcrt and pfmdr1 mutations detected between the two study periods, while Table 3 compares the phdhfr and pfdhps mutations. In Begoro, the prevalence of pfcrt 76T mutant alleles was reduced from 92% in 2005 to 57% in 2010 (6 years after the drug policy change in 2004) and continued to decline from 27% in 2014 to 5% in 2017, as described in this study. The rate of reexpansion of pfcrt K76 was, however, slower in the coastal town (Cape Coast) than in Begoro (Z = −3.5, P < 0.001). pfcrt 76T was still near saturation (95%) after 4 years (2008) of policy change (42), but its prevalence was reduced to 59.5% 8 years after the policy change (43). This study shows a prevalence of 29% in Cape Coast after 13 years since the policy change. pfmdr1 mutant alleles 86Y and 1246Y maintained a steady decline in prevalence over the first 8 years (42) and the last 4 years of change in drug policy. Interestingly, the prevalence of pfmdr1 184F has remained high over the years in both sites. The pfdhfr triple mutation almost reached fixation during both study periods, and so did the pfdhps 437G mutation. However, the prevalence of the combined pfdhfr/pfdhps quintuple mutation IRNGE has remained at less than 20% during all these 13 years of follow-up.

TABLE 2.

Comparison of prevalence of P. falciparum drug resistance alleles (pfcrt 76T and pfmdr1 alleles) between a previous study conducted at the current study sites and the current study

Site and yr	Prevalence (%)a						Reference or source
	pfcrt 76T	pfmdr1 86Y	pfmdr1 184F	pfmdr1 1034C	pfmdr1 1042D	pfmdr1 1246Y
Begoro
2005–2006	92 (36/39)	67 (26/39)	82 (32/39)	0 (0/39)	36 (14/39)	13 (5/39)	42
2007–2008	78 (39/50)	60 (30/50)	70 (35/50)	0 (0/50)	0 (0/50)	6 (3/50)	42
2010	57 (8/14)	27 (4/15)	80 (12/15)	0 (0/15)	73 (11/15)	7 (1/15)	42
2014	27 (25/94)	18 (14/78)	78 (62/80)	0 (0/82)	0 (0/82)	0 (0/73)	This study
2015	13 (12/96)	12 (9/77)	84 (66/79)	0 (0/80)	0 (0/80)	1 (1/74)	This study
2016	17 (15/91)	11 (8/79)	73 (50/69)	0 (0/71)	0 (0/71)	0 (0/56)	This study
2017	5 (3/57)	15 (7/46)	78 (40/51)	0 (0/51)	0 (0/51)	2 (1/43)	This study
Cape Coast
2005–2006	95 (37/39)	59 (23/39)	46 (18/39)	0 (0/39)	15 (6/39)	8 (3/39)	42
2007–2008	95 (40/42)	49 (21/43)	58 (25/43)	0 (0/43)	0 (0/43)	7 (3/43)	42
2014	47 (42/90)	18 (15/82)	82 (68/83)	0 (0/86)	0 (0/86)	2 (2/76)	This study
2015	38 (38/127)	18 (16/91)	67 (67/101)	0 (0/102)	0 (0/102)	3 (3/97)	This study
2016	29 (26/90)	5 (4/75)	80 (62/78)	0 (0/76)	0 (0/76)	1 (1/76)	This study
2017	29 (19/66)	7 (4/61)	71 (45/63)	0 (0/64)	0 (0/64)	0 (0/60)	This study

^aThe data in parentheses represent the number of individuals with the indicated haplotype/total number of individuals tested.

TABLE 3.

Comparison of prevalence of P. falciparum drug resistance alleles (pfdhps and pfdhfr alleles) between a previous study conducted at the current study sites and current study

Site	Yr	Prevalence (%)									Reference
		51I	59R	108N	IN	RN	IRN	437G	540E	GE
Begoro	2005–2006	45	90	83	5	43	33	88	3	3	84
	2007–2008	56	86	88	4	30	52	92	0	0	84
	2010	87	100	93	0	13	80	100	0	0	84
	2014	83	99	100	81	99	78	100	1	0	This study
	2015	83	93	94	82	93	81	95	3	2	This study
	2016	91	95	96	90	94	88	100	2	0	This study
	2017	89	100	100	86	100	85	98	2	0	This study
Cape Coast	2005–2006	55	75	83	8	23	43	73	0	0	84
	2007–2008	66	88	88	6	20	60	100	0	0	84
	2010	60	85	80	0	20	60	80	5	5	84
	2014	91	97	100	89	96	86	97	7	3	This study
	2015	87	96	98	85	96	82	95	5	3	This study
	2016	95	99	100	95	99	93	99	5	3	This study
	2017	88	93	96	87	92	84	100	6	2	This study

Markers associated with artemisinin and partner drug resistance.

Table 1 and Fig. 3 show the prevalence of mutations associated with artemisinin resistance and resistance to some partner drugs. None of the artemisinin resistance-associated pfk13 mutations observed in Southeast Asia (M476I, Y493H, R539T, I543T, and C580Y) were observed in the sample set. Our study identified the nonsynonymous pfk13 propeller mutations (C532S, S522C, P553T, and Q613H) that have previously been reported to be variants in Africa (28, 44–47) at low frequencies, as well as some new mutations (Table S5). There were no temporal trends. Also, the study identified the nonpropeller high-frequency pfk13 K189T marker at both study sites. In addition to the observed reduction in pfcrt CIET and pfmdr1 86Y prevalence over the 4-year period, there was also a slight reduction in pfmdr1 1246Y polymorphisms (from 3% to 0%; P = 0.5613), albeit it was not significant. However, the prevalence of pfmdr1 184F remained high (>60%). When pfmdr1 codons N86Y, Y184F, and D1246Y were analyzed together, the prevalence of the haplotype YYY was very low (at less than 2% at both sites), as shown in Table 1. Individuals bearing the NFD haplotype were present at a high frequency in both study sites (53% and 58% for Begoro and Cape Coast, respectively, in 2017), while less than 10% carried the double mutant haplotype YFD at both study sites in 2017. There were no statistically significant differences between the two study sites.

FIG 3.

MIP coverage for molecular markers associated with antimalarial drug resistance. Box plot distribution values are shown with bars, which represent the median; the lower hinge represents the 25th percentile, the upper hinge represents the 75th percentile, the lower whisker represents the smallest value no greater than 1.5× the interquartile range (IQR) from the lower hinge, the upper whisker represents the largest value no greater than 1.5× IQR from the upper hinge, and dots represent outliers.

DISCUSSION

The extensive use of ACTs as first-line treatment and SP for IPTp exerts strong selective pressure on parasite populations. Monitoring the resistance profiles of the antimalarial drugs can help contain any emerging resistant strains and to model their spread. Methods currently available in SSA are unable to detect low-frequency resistance markers, and this affects early detection and containment and ultimately leads to the spread of resistance, especially in high-transmission areas. Leveraging the throughput and sensitivity that MIP targeting and deep sequencing have over existing molecular surveillance platforms, we accurately quantified drug resistance markers and determined their temporal profiles in two sentinel sites in Ghana. Our method was able to detect a minor allele frequency of a mutation in individual samples of as low as 0.5% at the two sites, whereas a minor allele frequency of 15 to 20% was previously detectable using Sanger sequencing (48). In effect, samples with a mutant frequency in mixed infections of less than 15% are likely to have been missed by Sanger sequencing (see Fig. S1 in the supplemental material). In addition, the UMI linked to each arm of the target allowed for the correction of errors introduced during PCR amplification. The use of MIPs was also cost-effective, with this work costing approximately $10 per sample.

Our multiyear sampling allowed us to observe temporal trends. A continued decline in mutations associated with chloroquine resistance in Begoro and Cape Coast was observed since the last reports from studies conducted in 2012 (84) and 2014 (43). Although other 4-aminoquinolines, such as amodiaquine, are ACT partner drugs for the treatment of uncomplicated malaria in Ghana, there was no surge in the prevalence of the modulating pfmdr1 mutant haplotype pfmdr1 YYY (86Y, Y184, and 1246Y, which is linked to amodiaquine resistance) in circulating Ghanaian P. falciparum isolates. The rate of resurgence of chloroquine-sensitive parasite strains after the withdrawal of chloroquine drug pressure was accelerated in Begoro (which has a high to moderate transmission intensity) compared to that in Cape Coast (which has a moderate to low transmission intensity). We also observed that some mutations associated with resistance to SP have remained very high in prevalence and near fixation and found other emerging SP resistance mutations. The prevalence of parasite isolates carrying mutations linked with resistance to ACT partner drugs, such as lumefantrine, was high. However, the Southeast Asian pfK13 mutations associated with artemisinin resistance were not found in the two study sites.

The steady decline in the prevalence of chloroquine resistance mediator pfcrt 76T over the 4-year study period parallels the findings of a study conducted in Malawi, Cameroon, Ethiopia, and Mozambique, where the removal of chloroquine pressure resulted in the reexpansion of drug-sensitive P. falciparum strains (49–53). Early work in Ghana (42) showed a decline in pfcrt 76T prevalence from 92% in 2005 and 2006 to 57% in 2010 for Begoro. Another study conducted on Ghanaian samples collected in 2011 (43) observed levels on par with the 58.5% in Ghana in 2010. This was considered a slow decline after the drug policy change compared to the decline seen in other countries. For Begoro, the findings of this work, extending to 13 years after chloroquine removal, are consistent with those of other studies conducted in some African countries (>90% recovery in Tanzania after 10 years, 100% recovery in Malawi after 13 years, and 80% recovery in Mozambique after 5 years), while the findings for Cape Coast fell short, with about 75% recovery after 13 years (43, 51, 54–57, 84). On the contrary, data from Cape Coast agree with the results obtained from studies conducted in Benin (58) and Nigeria (59), where the decline occurred at a much lower rate. Reversion to sensitive strains is dependent on multiple factors, including, but not limited to, the time that has elapsed since the withdrawal of drug, transmission intensity, illegal drug use, access to health care or drugs, the presence of SMC, cross-resistance shared between chloroquine and amodiaquine, parasite genetics, drug bioavailability, intrinsic human genetic factors, and immunity (59). There are unpublished reports on the illegal use of chloroquine in Cape Coast after its withdrawal, and these possible differences in continued use may underlie the differences in reversion to the wild type (43). Our study, however, was not designed to track drug usage or other possible confounders to the recovery of chloroquine sensitivity. In terms of the main chloroquine resistance gene pfcrt, it is quite evident that there was a difference in the selection of chloroquine-sensitive strains. It is likely that because pfcrt 76T has a distinct causal effect on chloroquine resistance, subtle differences in the evolution of drug resistance, as observed in this study, would be detectable. On the contrary, it is likely that the modulatory effects of mutations in pfmdr1 on the levels of chloroquine resistance were not remarkable enough for detecting subtle differences in the two study sites, if any. pfmdr1, however, is worthy of further study. We observed the pleiotropic effect of pfmdr1 mutations in the study sites. We observed significant directional selection for pfmdr1 184F and pfmdr1 NFD that persisted throughout the 4 years of this study, suggestive of AL and DHAP pressures, and this was the case in a previous Ghanaian study as well (42). The caveat to this observation, however, is that our study did not examine copy number variations in the pfmdr1 gene. Although amodiaquine is an ACT partner drug, our study did not observe an increase in the pfmdr1 Y184 or YYY haplotype at either study site. This suggests that the ACTs in use are highly efficacious with 4-aminoquinolines (such as amodiaquine) as partner drugs and select for wild-type alleles. The results compare favorably with previous findings that demonstrated a strong selection of pfmdr1 184F, which persisted for a prolonged duration (2 months) after treatment with AL but not after treatment with ASAQ in Uganda (60) and in a similar study conducted in Cameroon (61). The trend suggests that the usage of different or alternating ACTs may delay the development of resistance by exerting inverse selection pressure on the parasites (62).

The accumulation of mutations in pfdhps and pfdhfr, the targets of sulfadoxine and pyrimethamine, are associated with sulfadoxine-pyrimethamine (SP) treatment failure (63, 64). Similar to the findings of the study conducted in 2012 at the same study sites reported by Duah and colleagues (42), the prevalences of the pfdhfr triple mutation IRN and the pfdhfr/pfdhps quadruple mutation IRGNK have remained high over the years and have almost approached fixation. This suggests that the use of SP for IPTp over the past 13 years has selected for this less-SP-resistant mutant rather than the highly resistant quintuple mutant. This result is consistent with the results of other studies conducted in West Africa (58, 65) and other African countries (66–70). Sulfadoxine-pyrimethamine is still used for IPTp because the prevalence of the 540E mutation, the surrogate marker for the quintuple mutant IRNGE, required for the high rate of the failure of treatment with SP in South and East Africa (50, 64, 71), is still very low in Ghana (6%) and in the subregion (58, 72). The pfdhps triple mutation AGKAS has previously been reported in Ghana, but at a lower frequency (0.8 to 2.6%) (73); however, the quadruple mutation AGKGS is being reported for the first time in our study sites. In addition, the pfdhps 581G mutation has not been previously reported in these parts of Ghana (74), even though one study reported it in one sample in the northern part of Ghana (75). Evidence indicates that the number of mutations is positively correlated with the levels of both in vivo and in vitro SP resistance (73, 76, 77). It is likely that our more sensitive MIP capture and deep sequencing detected these new mutations in the Ghanaian parasite isolates that were previously undetected.

Ghana is on course with the reexpansion of chloroquine-sensitive parasites after 13 years of removal of the chloroquine pressure. However, the seemingly slow pace of reexpansion observed in the Ewim Polyclinic catchment area and in other parts of the Cape Coast metropolis of Ghana (43) should be further explored across the country to see how the dynamics can be targeted for malaria control in Ghana. The high prevalence of pfmdr1 184F and pfmdr1 NFD should be further explored in in vitro and in vivo studies with ACT partner drugs to see the correlation with partner drug resistance in Ghana, if any. The use of SP in the IPTp policies should be encouraged and continually monitored, especially as new mutations emerge. Because of their high sensitivity and low cost, the use of MIP capture and deep sequencing should be further explored to promote the preemptive control and containment of emerging resistant parasite strains in Ghana.

MATERIALS AND METHODS

Study area and sites.

The study was conducted in two of the sentinel sites established by the National Malaria Control Programme (NMCP) to monitor antimalarial drug efficacy in Ghana: the Begoro District Hospital in Begoro and the Ewim Polyclinic in the Cape Coast metropolis (Fig. 4). These health care facilities diagnose the majority of malaria cases in their catchment areas. The Begoro District Hospital is the only state health facility in the district and is therefore responsible for the management of most malaria cases. The Ewim Polyclinic is an initial site of presentation for malarial diagnosis and treatment in the Cape Coast municipality, while the regional hospital caters mainly to referrals.

FIG 4.

Map of Ghana showing the two sites selected for this study. Begoro, which is in the Fanteakwa District, is considered a forest zone where malaria is hyperendemic and which has a high transmission intensity. Cape Coast, in the Cape Coast district, is considered a coastal savanna zone where malaria is also hyperendemic, but it has a low to moderate transmission intensity. Both axes indicate kilometers.

Cape Coast is in the coastal ecological zone of Ghana and is situated 165 km west of Accra (the capital of Ghana) on the Gulf of Guinea (05°06′N, 01°15′W). Cape Coast has a population of 169,894 (78) and has annual rainfall of 750 to 1,000 mm, with double maximum rainfall occurring in June and October. The temperature in the district ranges from 21°C to 36°C throughout the year, and the mean monthly relative humidity varies between 85% and 99%. Malaria transmission in the metropolis is low to moderate and perennial, with the most intense transmission occurring in June.

Begoro is in the forest ecological zone of Ghana and is the capital of the Fanteakwa District in the eastern part of Ghana. It is situated at 6°23′N, 0°23′W. Begoro has a population of 86,154 people (78). The district has annual rainfall of 1,500 to 2,000 mm, with double maximum rainfall occurring in June and October each year. The temperature ranges from 26.0ºC to 31.5ºC, and the mean relative humidity is 74.4%. Malaria transmission is high and perennial, with the most intense transmission occurring in June.

Study design and participants.

This study primarily aimed to compare the two study sites for the proportion of clinical subjects with blood samples containing parasites with a mutant allele associated with drug resistance in the targeted genes. For an individual mutation, the study had an 80% power to detect baseline differences between the two study sites of as low as 20%, using a two-sided, 0.05 alpha significance level.

Study samples were drawn from a one-arm prospective study of the efficacy of ASAQ in Ghana conducted between 2014 and 2017 (79). Briefly, the participants enrolled were children aged 6 months to 14 years presenting with uncomplicated P. falciparum malaria at the Begoro and Cape Coast sites. The participants were identified on the basis of routine clinical laboratory procedures. Parasitological assessment was done by microscopy, for which 400 μl of finger-prick blood was aseptically drawn for thick and thin smears and hemoglobin measurement. Approximately 50 μl was then spotted onto 110-mm Whatman 3MM filter paper, dried, and stored individually in plastic bags with silica gel desiccant.

The study protocol was reviewed and approved by the Institutional Review Board (IRB) of the Noguchi Memorial Institute for Medical Research, University of Ghana (IRB approval no. 056/12-13). Written informed consent was obtained from the parents or guardians of all children before enrollment. Additionally, children aged 12 years to 14 years gave their assent before enrollment. The objectives, methods, potential risks, and their right to withdraw from the study without any penalty were explained to the parents or guardians and children.

DNA extraction.

Genomic DNA was extracted from the dried blood spots (DBSs), using a QIAamp DNA minikit (Qiagen) according to the manufacturer’s instructions. In all, DNA was extracted from 6 (3-mm diameter for each punch) punches, placed into 200 μl of eluent. The quality and quantity of the DNA were estimated using a Qubit (version 2.0) fluorometer, and quantification was based on a Quant-iT PicoGreen double-stranded DNA (dsDNA) assay (catalog number P11496; Invitrogen). A total of 803 DNA samples were extracted and quantified for MIP capture. In addition, genomic DNA was extracted from four laboratory strains, strains 3D7 (wild type), HB3, 7G8, and DD2, and used as wild-type and mutant allele controls in all assays.

Quality control.

Previously validated MIP assays were used (40). Briefly, serial dilutions of a control mixture of DNA isolated from the laboratory strains 3D7, HB3, 7G8, and DD2, mixed at relative frequencies of 67%, 14%, 13%, and 6%, respectively, were used to assess the sensitivity and the accuracy of the MIP captures. In all, 15 serial dilutions of a mixture of four laboratory strains containing from 7,469 down to 0.5 of P. falciparum haploid genome copies and 1 ng human DNA (∼1,650 haploid genome copies) per microliter were used. Quantification was based on the results of the Quant-iT PicoGreen dsDNA assay (catalog number P11496; Invitrogen). These DNA mixtures were combined with 1 ng/μl human DNA to better mimic the DNA isolated from DBSs and whole blood.

Molecular inversion probe design, capture, amplification, library preparation, and sequencing.

MIP design, capture, amplification, and sequencing were done as previously described (40) (Fig. 5). Briefly, the DNA samples were hybridized with probes designed to capture the region of interest in the drug resistance genes, including pfcrt, pfmdr1, pfdhps, pfdhfr, and pfk13 (see Table S1 in the supplemental material). A mixture of the genomic DNA, the MIPs, a thermostable ligase, polymerase, and unlabeled nucleotides was preheated and brought to the annealing temperature. Upon hybridization of the MIP arms, the target region (the template sequence between the arms) was filled in by the DNA polymerase, after which DNA ligase sealed the remaining gap, with the ligation arm forming a covalently closed circular molecule.

FIG 5.

Step-by-step diagram showing the procedure for targeted sequencing using MIP capture and enrichment by PCR amplification. In a single tube, multiplexed reaction MIPs (a) are first hybridized to the genomic region of interest (b), and the gap is filled by polymerization and ligation. Linear (nonreacted) probes are then removed by exonuclease I/III, and the circular MIP capture containing the structure of a MIP, ligation arm (Lig), extension arm (ext) (c), and target are amplified (d). gDNA, genomic DNA.

Next, all noncircular DNAs (unbound probes, original template DNA) were removed by exonuclease treatment, leaving the circular MIPs. To create the sequencing library for a sample, primers that anneal by binding to the shared backbone were designed and used with unique nucleotide barcodes that were specific for each sample. The amplified products were viewed on agarose gels. Equal volumes of PCR products from each reaction were pooled, cleaned, and concentrated using AMPure XP beads (catalog number A63881; Beckman Coulter) at a 0.8× bead/DNA ratio using the manufacturer’s protocol to remove unwanted adapter dimers of ∼200 bp. If the adapter dimers remained after bead cleanup, the eluted DNA was electrophoresed on a 1.5% agarose gel and the relevant bands of between 500 and 1,000 bp were extracted from the gel using a Monarch DNA extraction kit (catalog number T1020S; New England Biolabs). The purified amplicons were then sent to the sequencing core of the University of Massachusetts Medical School (UMMS) for sequencing on a MiSeq platform for 250-bp paired-end sequencing.

Molecular inversion probe processing, variant calling, and annotation.

The sequences generated were processed using MIPWrangler software (N. J. Hathaway, unpublished data) and other software described elsewhere (40). Briefly, after demultiplexing the sequences by their dual sample barcodes, the paired end reads were stitched together using the FLASH program. The raw sequences were filtered by discarding sequences if the fractions of base quality scores above 30 were less than 70% (Q30 < 70%). Another demultiplexing step was carried out to generate a file for each target per sample using the extension and ligation arm sequences (Table S1). To remove a significant proportion of PCR errors that occurred in late cycles, including polymerase stutter and subsequent sequencing errors, target sequences were corrected by clustering the sequences using their unique molecular identifiers (UMIs) per sample. A consensus sequence for each specific UMI was then created, and the clustering algorithm from MIPWrangler used this sequence to subsequently cluster the corrected UMI sequences for accurate detection of single base differences and indels at levels of 1% or less (40, 80). For this analysis, we set a minimum threshold for an MIP-captured sequence at ≥2 UMIs, and a ≥0.5% relative abundance was set for each haplotype to minimize false calls. In addition, a coverage filter of ≥10 UMIs per variant locus per sample was required to avoid undercalling mixed infections with abundant minor strains. The captured sequences for each MIP and sample were then extracted from the MIPWrangler output, and pairwise alignment was conducted using LastZ software to determine the difference between the observed sequence and the reference sequence for each probe (81). Finally, single nucleotide variants and indels from the LastZ output were annotated using Annovar software (40).

Data analysis. (i) Estimating prevalence.

Definitions of the outcomes used in this study are as previously described (82). The prevalence of drug resistance alleles (Prev), defined as the proportion of infected children carrying at least one resistant parasite clone, was calculated as Prev = P(mix) + P(res), where P(mix) is the proportion of infections with mixed (both wild-type and resistant) parasite clones and P(res) is the proportion of infections with only resistant parasite clones. P(mix) was calculated as P(mix) = 1 − [P(res) + P(wt)], where P(wt) is the proportion of infections with all wild-type strains.

(ii) Estimating frequency.

The allele fraction was estimated for each sample, and the population frequency of each resistant mutant was estimated as the average allele fraction for that population.

(iii) Statistical analysis.

All statistical analyses were performed using the statistical software R (version 3.5). A Bonferroni-adjusted significance level of a P value of 0.05/25 was used to account for testing of multiple mutations. The prevalence and frequency of drug resistance alleles (Prev) were estimated and compared between study sites using the chi-square and Fisher’s exact tests. The trends of prevalence over the 4 years were tested using the Cochrane-Armitage test for trends. Statistical tests for monotonic trend between study sites were performed using the regional Kendall test (83).