Introduction

Between 2015 and 2022, The Gambia achieved a more substantial reduction of Plasmodium falciparum malaria prevalence than most other countries in Africa [1]. Nevertheless, this reduction in malaria prevalence is not sufficient to meet the 75 % decrease milestone expected for 2025, implying that additional efforts are needed to achieve malaria elimination.

In regions characterised by seasonal transmission, malaria clinical cases peak towards the end of the wet season, a pattern mirrored in The Gambia’s transmission cycle. Its climate is characterised by a rainy season from late June to September, followed by a ∼8 month-long dry season, typically without rainfall. Malaria cases are almost exclusively reported from September to December (the high transmission season), while malaria is virtually absent for the remaining 8 months (the low transmission season) [2]. Malaria in The Gambia is highly unequal, affecting mostly individuals with low income from rural areas in the eastern part of the country [3–6]. Little is known about how the low transmission season affects the parasite population at genetic level. Understanding how the Plasmodium falciparum parasite population adapts to these seasonal variations is critical for informing targeted malaria elimination strategies.

Evidence suggests that imported cases play a minor role in maintaining malaria transmission, with the resurgence of malaria attributed to a persistent reservoir of asymptomatic chronic carriers that bridge two transmission seasons separate by several months of low to no transmission [7–10]. Previous studies conducted between 2012 and 2016 in The Gambia and neighbouring Senegal demonstrated that a substantial proportion of asymptomatic individuals remained infected from the end of the high transmission season to the end of the low transmission season, underscoring the ability of these infections to establish chronicity over extended periods [6,11]. Importantly, asymptomatic infections may carry gametocytes capable of transmitting to mosquitoes [12,13].

The complexity of infection (COI) is the number of distinct parasite genotypes within an infected individual. At the population level, the average COI correlates with transmission intensity, offering a practical approach to estimate prevalence [14–18]. Furthermore, genetic relatedness analyses between parasites can shed light on fine-scale spatio-temporal transmission dynamics. Genetic relatedness is assessed through Whole Genome Sequencing (WGS) or the cost-effective barcode genotyping, the latter able to identify a unique parasite strain with as few as 24 loci [19]. From pairwise distances between such barcodes, it has been shown that relatedness tends to decrease with time and distance of the sampled infections both at the national level in The Gambia and at the local level in neighbouring Senegal [18,20]. Although the parasite population genetic diversity has been well characterized from clinical infections isolated within a single high transmission season, the impact of the low transmission seasons on the long-term genetic diversity of the parasite populations is still poorly understood.

Identity-by-Descent (IBD) serves as a metric of genetic relatedness between parasites, incorporating recombination rate to provide insights into recent recombination events. While Fst-based studies are well adapted to characterize the haplotype diversity of populations of parasites across countries or over extended periods, IBD highlights recent recombination events between two parasites with varying levels of relatedness [21–23].

Extensive genomic epidemiology studies to date focused on clinical cases, yet such symptomatic cases only represent a minority of all P. falciparum infections [24,25]. The extent at which allelic diversity observed in clinical cases reflects the total parasite population remains unclear. Here, we genotyped P. falciparum isolates from a longitudinal study conducted over a 2.5-year period in four nearby villages in the Upper River Region of The Gambia [11,26]. Through identity by descent (IBD) analysis of parasite genetic barcodes and genomes collected during the low and high transmission seasons, we aimed to elucidate the parasite genetic diversity at the community level.

Material and methods

Study design and participants

Starting in December 2014, we recruited all residents from two villages (Madina Samako and Njayel, identified respectively with the letters ‘K’ and ‘J’), with two additional villages (Sendebu and Karandaba, identified respectively with the letters ‘P’ and ‘N’) recruited from July 2016, all four villages being in the Upper River Region in The Gambia within 5 km of each other. Active case detection was conducted a total of 11 times over the 2-year period, with each sampling session occurring within a ten-day window. Symptomatic cases that occurred between September to December 2016 were also sampled. More information about the recruited participants can be found in a previous study [26]. In December 2016, a cohort of 42 asymptomatic P. falciparum carriers was recruited and sampled monthly for 6 months until May 2017, as previously reported [11]. The study protocol was reviewed and approved by the Gambia Government/MRC Joint Ethics Committee (SCC 1476, SCC 1318, L2015.50) and by the London School of Hygiene & Tropical Medicine Ethics Committee (Ref. 10982).

Sampling and molecular detection of parasites

Details on the P. falciparum detection is provided in previous works [11,26]. Briefly, fingerprick blood samples were tested for P. falciparum by qPCR. From July 2016 onwards, individuals testing positive for P. falciparum were invited to provide an additional 5 to 8 mL venous blood sample that was leucodepleted with cellulose-based columns (MN2100ff) and frozen immediately. DNA was extracted with QIAgen Miniprep kit following manufacturer procedure.

Genotyping and genome sequencing

A total of 522 P. falciparum DNA positive samples, 307 from fingerprick and 215 from venous blood, were processed for genotyping (442 samples) and whole genome sequencing (331 samples of which 251 were genotyped) as part of the SpotMalaria consortium. Genotyping was performed by mass-spectrometry based platform from Agena MassArray system. The output consisted of 101 bi-allelic SNPs located on the 14 chromosomes and concatenated into a ‘molecular barcode’ [27], plus six markers of resistance to antimalarials (aat1 S528L, crt K76T, dhfr S108N, dhps A437G, kelch13 C580Y and mdr1 N86Y). Genetic barcode SNPs had been picked for their variable allele frequencies within the P. falciparum population [27]. Whole genome sequencing (Illumina) was performed after a Selective Whole Genome Amplification step [28]. Paired end DNA sequence-reads (150 bp) were aligned to 3D7 reference genome version 3. Variants were called by a script from the MalariaGen consortium using GATK HaplotypeCaller [29,30]. Two drug resistance markers, aat1 S528L and kelch13 C580Y were only available in sequencing data and absent from genotyping data. If distinct calls of drug resistance markers were obtained between genotyping and whole genome sequencing data (Figure S7), only the call of whole genome sequencing was considered for the rest of the analysis.

Parasite relatedness

To accurately assess the parasite genetic similarity between different sampled infections, we estimated pairwise mean posterior probabilities of Identity-By-Descent (IBD) between genomes or barcodes using hmmIBD, a hidden Markov model-based software relying on meiotic recombination events given a recombination rate of Plasmodium falciparum of 13.5 kb/cM [31,32]. The probability of two samples to be in IBD represents the expected shared fraction of their genomes. An IBD of more than 0.9 is considered identical, hence describing the same parasite genotype.

Multi-locus genotype barcode data analysis pipeline

To analyse whole genome data formatted in a VCF format, we developed the following pipeline to (Figure S1):

1. Filter out genomic loci for which the QUAL is inferior to 10000, with more than 2 alleles identified in the population or that are located outside of the core genome [32].

2. Remove genomes comprising less than 4000 SNPs covered by at least 5 reads.

3. Estimate the proportion of polyclonal samples using the F_ws metric and heterozygous loci.

4. Format SNPs into a binary matrix as required by hmmIBD (mixed and unknown positions set to 0). SNP calls were considered mixed if the within-sample Minor Allele Frequency (MAF) was greater than 0.2. The MAF of 0.2 was chosen according to the good agreement between molecular barcodes and genomic barcodes (high number of mixed locus matches and low number of mixed locus mismatches) (Figure S2).

5. Use the paired IBD values obtained from hmmIBD and build a network of genome relatedness. To limit the number of false positive matches between genomes containing mostly population-level major alleles (in the case of high polyclonality for example), IBD values were considered unknown between pairs of genomes having less than 100 informative loci, i.e., 100 pairs of loci available in both genomes with each time one of them being the population-level minor allele.

The second part of the pipeline imputes missing SNPs in the molecular barcode from the WGS data to build a ‘consensus barcode’. Then, it estimates the duration of infection by the same genotype (Figure S1):

1. Build a ‘molecular barcode’ out of the initial 101 genotyped SNPs and a ‘genomic barcode’ out of the same 101 SNPs called from high-quality genomes.

2. Remove 12 loci that are absent from all high-quality genomes.

3. Estimate the within-sample Minor Allele Frequency (MAF) to use as a cutoff to call a genomic locus mixed when building barcodes out of genomic data. Genomic barcodes are built using different cutoffs of within-sample MAF and aligned against molecular barcodes from the same isolates. The cutoff of within-sample MAF of 0.2 showed a high number of mixed locus matches and low number of mixed locus mismatches, meaning that molecular barcodes and genomic barcodes were in good agreement (Figure S2). This cutoff was retained to build all genomic barcodes. Each pair of molecular and genomic barcodes obtained from the same isolates were aligned. For isolates sampled after May 2016, molecular barcodes are most of the time not matching genomic barcodes for 21 loci, suggesting that a change in the molecular genotyping protocol led to incorrect calling (Figure S3). As a result, these 21 loci were considered unknown for all the molecular barcodes obtained after May 2016.

4. Using the alignment between molecular and genomic barcodes from the same isolates, replace unknown and mismatched SNP of the molecular barcodes by the SNP of genomic barcodes. These improved molecular barcodes are referred to as ‘consensus barcodes’.

5. Remove consensus barcodes with fewer than 30 SNPs.

6. Estimate the proportion of polyclonal samples using heterozygous loci (described in following section).

7. Format consensus barcodes into a binary matrix (mixed and unknown loci set to 0) and run hmmIBD.

8. Use the paired IBD values obtained from hmmIBD and build a network of barcode relatedness. To limit the number of false positive matches between barcodes containing mostly population-level major alleles (in the case of high polyclonality for example), IBD values were considered unknown between pairs of barcodes having less than 10 informative loci, i.e., 10 pairs of loci available in both barcodes with each time one of them being the population-level minor allele.

Complexity of infections

The clonality of each isolate was estimated from whole genome sequenced samples by the F_ws metric based on allelic frequencies from genomic data [33]. Additionally, the complexity of infections was estimated by the proportion of heterozygous loci (polyclonal if the proportion is above 0.5 % of available sites) in both consensus barcodes and genomes using a within-sample MAF of 0.2.

Genetic relatedness between groups of multi-locus genotype barcodes

All ‘consensus barcodes’, hereafter referred to as ‘barcodes’, available after December 2016 were excluded from the genetic relatedness analysis as they were obtained exclusively from the 42 individuals recruited in the monthly dry season cohort [11]. Barcodes were grouped by their collection date (11 time points from December 2014 to December 2016), sampling location (households or villages) or collection date split by households. Within the same group, the genetic relatedness was estimated by the proportion of related barcodes (IBD ≥ 0.5) over all possible pairs of barcodes. When comparing different groups, the genetic relatedness was estimated by the proportion of related barcodes (IBD ≥ 0.5) over all pairs of barcodes of each group, excluding pairs involving barcodes from the same individual. We used a stringent criteria of filtering out pairs of households with less than 5 comparisons (1013/2829 removed pairs) and pairs of collection dates split by households with less than 5 comparisons (25/192 removed pairs). All pairs of villages (10 pairs) and all pairs of dates (66 pairs) contained at least 10 comparisons.

Results

Combined barcode and whole genome analysis pipeline

Overall, 5322 fingerprick and 253 venous blood samples were collected from 1516 individuals aged 3 to 85 years in four nearby villages of the Upper River Region of The Gambia between December 2014 and May 2017, as detailed in previous studies (Figure 1A) [11,26]. To characterise the P. falciparum genetic diversity and identify the impact of antimalarial drugs, we attempted parasite genotyping and whole genome sequencing of 522 isolates over 16 time points (Figure 1B). Whole parasite genomes were also successfully sequenced for 199 isolates. Through the concatenation of SNPs from barcode genotyping and whole genome sequencing, we obtained a high-quality ‘consensus’ barcode for 425 isolates comprising on average 59 SNPs (range 30 to 89 SNPs) (Figure 1). A detailed description of the pipeline is available (Figure S1).

Figure 1:

Complexity of infection is stable across seasons

The complexity of infection (COI), defined as the number of unique genotypes/genomes within an infected individual, serves as an indicator of parasite strain diversity within the population. We estimated COI using F_ws values from genomes and the proportion of heterozygous loci from genomes and barcodes. Across these metrics, the proportions of polygenotype isolates were estimated as 40 % using F_ws (79/199), 38 % using heterozygous loci of genomes (76/199) and 28 % using heterozygous loci of barcodes (120/425). As expected, the proportion of heterozygous loci in barcodes or genomes showed a strong negative correlation with F_ws values, indicating its efficacy as a predictor of COI (Figure S4). Although the proportion of polygenotype isolates fluctuated between timepoints, no discernible trend was observed between low and high transmission season, suggesting a relatively stable complexity of infections in the population throughout the 2.5-year study duration (Figure S5). At the individual level, we previously showed that the COI was stable in polyclonal infections during the dry season (Collins22).

High P. falciparum genetic diversity at the community-level

To determine the relatedness between isolates and compare malaria parasites from distinct geographical locations and distant times, Identity By Descent (IBD) was calculated pairwise. The reliability of consensus barcodes in identifying genetically related isolates was confirmed through a strong linear correlation between barcode-IBD and genome-IBD, particularly when both were above 0.5 (R² = 0.77, p-value < 10^-15), with this threshold chosen to distinguish between related (IBD ≥ 0.5) and unrelated (IBD < 0.5) samples for all 425 consensus barcodes (Figure S6).

To characterise the parasite population genetic diversity, only one barcode per continuous infection per individual was retained among barcodes sampled between December 2014 and December 2016, totalling 284 remaining barcodes. Among 34,325 pairwise comparisons of barcodes, only 443 (1.3 %) demonstrated relatedness with an IBD above 0.5, indicating extensive recombinatorial genetic diversity within the parasite population (Figure 2A). Out of the 284 barcodes, 73 (26 %) were identical (IBD ≥ 0.9) to a barcode from another individual, while 213 (75 %) were related (IBD ≥ 0.5) to at least one other barcode (Figure 2B). Although clusters of genetically related barcodes may suggest a higher connection within villages (1.5 % of related barcodes) than between them (0.7 % of related barcodes), the average proportion of related isolates was not significantly different (Welch t-test value = 1.26, p-value = 0.29), indicating interconnectedness between villages.

Figure 2:

Pattern of relatedness between infections is shaped by seasonality

To explore the spatio-temporal relationship between barcodes from distinct individuals, we took advantage of our frequent samplings at the community-level to calculate the percentage of related isolates across six temporal groups, ranging from 0 to 2 months apart between sample collections to 16 to 24 months apart (Figure 3A). Remarkably, at the temporal level, the average proportion of related barcodes within barcodes sampled less than two months apart is 4.7 %. This value is ten-fold higher than the proportion of related barcodes sampled more than 12 months apart equal to 0.29 % (Welch t-test value = 4.67, p value < 10^-4). This indicates that recombination between P. falciparum isolates breaks down IBD with time.

Figure 3:

Similarly, the pairwise barcode IBD was computed for spatial groups: pairs of barcodes from the same household, different households within the same village, and different households across villages (Figure 3B). At the spatial level, barcodes sampled less than 2 months apart and from the same household were six times more related than those sampled from different villages (average proportions of 0.095 and 0.015, Welch t-test value = 3.24, p value = 0.01). However, when barcodes were sampled more than two months apart, the correlation between genetic relatedness and sampling location gradually disappeared. Altogether, the overall large parasite recombinatorial genetic diversity combined with the increased proportion of related isolates within the same household, indicate a scenario in which the same infectious mosquito infected two or more individuals living together, or a direct transmission chain between two household members.

The impact of seasonality on the parasite recombinatorial genetic diversity was assessed by comparing the proportion of related barcodes (IBD ≥ 0.5) between groups of collection dates within or between transmission seasons. Barcodes sampled during the high transmission seasons from 2014, 2015 and 2016, as well as the low transmission seasons of 2015 and 2016, were grouped into intraseasonal pairs if they were collected during the same season or interseasonal pairs otherwise. This resulted in 5 groups for intraseasonal pairs and ten groups for interseasonal pairs, with the most distant groups (high 2014 and high 2016) being 4 seasons apart. Similarly to Figure 3, pairs of collection dates close in time exhibited greater similarity than more distant collection dates (Figure 4A). This suggests a continuous recombination process among all parasites, rather than the transmission of one or more specific strains. To determine the specific time of the year the average genetic relatedness declines, the proportion of related barcodes was compared between pairs of sample collections from the same season or one season apart (Figure 4). Parasites belonging to the ‘within high’, ‘within low’ and ‘transition high to low’ groups displayed similar genetic relatedness with average proportions of related barcodes of 0.026, 0.026 and 0.025, respectively. However, the ‘transition low to high’ group exhibited a 4-fold lower genetic relatedness, with an average proportion of related barcodes of 0.006, indicating that most of the parasite differentiation occurs during the transition from the low transmission season to the subsequent high transmission season. This corresponds to the increase in transmission rate at the onset of the high transmission season, with parasite genetic diversity being reshuffled after sexual reproduction in the mosquito. This increase in genetic diversity is not observed in the ‘transition high to low’ group, demonstrating a reduced transmission in the low transmission season.

Figure 4:

Independence of seasonality and drug resistance markers prevalence

In The Gambia, drug resistant allele frequencies increased dramatically from the late 1980’s until the early 2000’s, then plateaued until 2008 [34]. The rise in clinical malaria cases during the high transmission season implies usage of anti-malaria drugs, particularly among children aged between 0 and 5 years receiving the monthly Seasonal Malaria Chemoprevention, from September to November. We investigated whether the prevalence of drug resistance alleles is influenced by malaria seasonality due to differential selective pressures applied between high (more pressure) and low (less pressure) transmission seasons. To determine the prevalence of resistant haplotypes over time, six drug resistance-related haplotypes were obtained by molecular genotyping and called from whole genome sequenced in genes aat1, crt, dhfr, dhps, kelch13 and mdr1, leading to a merged total of 438 isolates with drug resistance-related haplotypes. Overall, 89 % of haplotypes called from both molecular genotyping and whole genome sequencing were identical (Figure S7).

As expected, the haplotype kelch13 C580Y, related to artemisinin resistance, was absent. Overall, the proportion of isolates with resistant alleles was stable over time for each haplotype with 0.92 (95 % Wilson’s Confidence Interval: 0.87-0.95) for aat1 S258L, 0.64 (95 % CI: 0.59-0.70) for crt K76T (chloroquine resistance), 0.93 (95 % CI: 0.90-0.95) for dhfr S108N (pyrimethamine resistance), 0.46 (95 % CI: 0.41-0.51) for dhps A437G (sulfadoxine resistance) and 0.12 (95 % CI: 0.09-0.17) for mdr1 N86Y (multi-drug resistance) (Figure S8). In our dataset, no evidence of seasonality-related drug resistance fitness costs impacting population haplotype diversity was observed. Our data from 2015 to 2017 is very similar to the allele frequency levels from 2008, indicating a potential plateau in the cost-benefit of drug resistance alleles [34].

P. falciparum chronic infections with persisting genotypes

To distinguish between ‘true’ chronic infections with the same parasite genotype from reinfections, we measured the minimal duration of infection using IBD values between barcodes obtained from the same individual sampled at different time points, which were not utilized for the spatio-temporal analysis of genetic relatedness. The beginning and end of an infection by the same parasite were attributed to the two most extreme time points separating two identical barcodes (IBD ≥ 0.9). Overall, 34 individuals (24 males and 13 females) were chronically infected with the same dominant P. falciparum genotype ranging from one month to one and a half years (Figure 6). Gender and age did not significantly influence the duration of infection (Figure S9). Three individuals (25, 26 and 28) were infected with two distinct parasite strains, as shown by monthly barcodes alternating between the two strains (e.g. in individual 25, barcodes from Jan and Mar, and from Feb and Apr, are identical). Interestingly, in each case, the two strains were genetically related (IBD ≈ 0.5), likely indicating co-infection of sibling strains from the same brood. The change in proportion between the two strains over time could indicate intra-host competition.

Figure 6:

Discussion

In anticipation of the pre-elimination phase of malaria, our study aimed to develop robust methods for estimating the complexity of infections (COI) and constructing genetic relatedness networks of P. falciparum parasites. Leveraging both barcodes and genomes offers complementary advantages: barcode genotyping provides cost-effective access to genetic information from numerous blood isolates, while whole genomes yields quantitative data on thousands of single nucleotide polymorphisms (SNPs), facilitating comprehensive analyses of parasite diversity, with the added potential to extend analyses to newly discovered haplotypes unidentified at the time of sequencing. Combining both data types validates base calling from barcode genotyping and guides method applicability across the limited positions available in barcodes.

The stable COI over time reflects a constant transmission intensity [16–18,35]. Such stable transmission in this area of The Gambia, reported already elsewhere, contrasts with the overall decrease of malaria transmission across the country, highlighting the heterogeneity malaria within The Gambia [6,26]. Our findings underscore the importance of understanding local malaria dynamics amidst broader regional trends.

With Identity by Descent (IBD) we evaluated parasite relatedness between genomes and between barcodes. Although using more SNPs (e.g. more than 200 SNPs) is ideal to reach an accurate relatedness, in silico data suggests that the accuracy is only marginally improved with 96 SNPs or more [36]. Furthermore, we showed that for IBD values above 0.5, the correlation between genome- and barcode-IBD was very strong. Finally, the qualitative aspect of our approach, with isolates either related (IBD ≥ 0.5) or unrelated (IBD < 0.5) enables to compensate for the lower accuracy of barcode-IBD.

Overall, 26 % of unique parasite strains were found in two or more individuals and around 28 % of isolates were polyclonal according to the proportion of heterozygous loci of barcodes. This seemingly contrasts with findings from Thiès, in neighbouring Senegal, where parasite haplotype diversity is low, the proportion of shared genetic types between isolates is 35 %, with less than 10% of polyclonal infections, suggesting self-mating transmission and low outcrossing levels [9,10,37]. However, the key difference of our approach was the active detection of asymptomatic infections, as opposed to the large number of studies sequencing parasites collected from clinical cases. Our results argue for active case detection as a necessary step to comprehensively characterise the parasite genetic diversity.

One limitation of genetic epidemiology studies such as this one, is the necessity to exclude IBD values obtained between isolates with high level of polyclonality and thus not enough informative loci. One important remaining challenge is to incorporate highly polyclonal infections in the network of genetic relatedness of parasites, while the proportion of complex infections tends to increase with malaria transmission intensity [38]. Achieving this goal is difficult as it requires whole genome sequencing followed by a deconvolution tool such as DEploid, with the caveat that the different strains are not too related or too rare [39,40].

A follow-up study carried out in rural villages of eastern Gambia between 2012 and 2016 reported that roughly half of asymptomatic infected individuals at the end of the wet season were still infected at the end of the dry season [6]. We previously reported a similar rate [11]. We showed that parasites collected during the dry season share significantly more genetic similarity with those from the previous wet season than those from the next wet season. Compared to parasites sampled 2 months apart, parasites sampled more than one year apart had very low level of similarity, implying that the whole population had been replaced in just one year. A study conducted in Colombia (with a 10-fold lower malaria prevalence than in The Gambia) showed that the same level of decreasing genetic similarity was reached in about 9 years [35,41].

Most infections were sub-microscopic, which is typically observed when prevalence falls below 20 % [26,42,43]. These low-density infections that may persist for months are typically asymptomatic unless the host immune system is compromised [44]. In this study, the infection with the longest duration was in an asymptomatic individual aged between 5 and 9 years in which it persisted as the same parasite strain for one and a half year. School-age children contribute the most the malaria reservoir through their higher carriage of asymptomatic infections and their ability to effectively infect mosquitoes [6,12,45,46].

In the late 20th century, when the malaria burden in The Gambia was much higher, clinical cases from the same household were more likely to be caused by the same mosquito bite [47]. More recently, Ngwa et al. showed that parasite genetic similarity during the 2013 transmission season was inversely correlated with each spatial and temporal distances in The Gambia [20]. We also found that parasites sampled in neighbouring households are more genetically similar but only when sampled less than three months apart. Similarly, two other studies observed parasite strains with varying levels of spatio-temporal propagation in Thiès, Senegal, suggesting that some parasites are more actively transmitted than others [9,18]. These results add evidence that anti-malarial strategies should target all members of a household with an infected individual. Such “reactive” strategies include treatment of a malaria case and all its household members (without testing) or testing all household members and treating if necessary. The impact of reactive strategies on malaria transmission is at best very limited [48–51]. Based on the sheer size of the P. falciparum asymptomatic reservoir, with parasitaemias typically below microscopy or RDT detection level [26], it is not surprising that targeting clinical cases and their immediate families is not sufficient to break the transmission. For a malaria elimination campaign, based on our insights into parasite genetic diversity and transmission dynamics in The Gambia at the community level, we argue for mass detection with a highly sensitive method such as qPCR, followed by treatment of P. falciparum positive cases and all their household members.

Data availability

The barcode data analysis pipeline can be found at https://github.com/marcguery/gambiarcodes.

Barcode and genome metadata are in ‘Barcode_SuppTables.xlsx’.

• ST1: Gender, age and households of participants.

• ST2: Sampling, falciparum-malaria test results and treatment.

• ST3: Locations of genotyped SNPs and their rank in barcodes.

• ST4: Statistics about barcode genotyping.

• ST5: Statistics about genome sequencing.

• ST6: Complexity of infection metrics from barcodes.

• ST7: Complexity of infection metrics from genomes.

• ST8: Drug resistance markers genotyped from barcodes.

• ST9: Drug resistance markers genotyped from genomes.

• ST10: Duration of continuous infection estimated from barcode relatedness.

This publication uses data from the MalariaGEN SpotMalaria project as described in ‘Jacob CG et al.; Genetic surveillance in the Greater Mekong Subregion and South Asia to support malaria control and elimination; eLife 2021;10:e62997 DOI: 10.7554/eLife.62997’ [27]. The project is coordinated by the MalariaGEN Resource Centre with funding from Wellcome (206194, 090770).

Acknowledgements

The authors would like to thank the staff of MalariaGEN, Wellcome Sanger Institute Sample Management, Genotyping, Sequencing and Informatics teams for their contribution. We thank Michael Fontaine, Franck Prugnolle and Virginie Rougeron for insightful comments on data analysis.

Additional information

Funding

This work was funded by grants from the Netherlands Organization for Scientific Research (Vidi fellowship NWO 016.158.306) and the Bill & Melinda Gates Foundation (INDIE OPP1173572), the joint MRC/LSHTM fellowship, CNRS Transversales, the French National Research Agency (ANR 18-CE15-0009-01), the Fondation pour la Recherche Médicale (EQU202303016290).

Author contributions

Conceptualization by AC, TB, UDA and DC; field work and sample processing by SC, SD, FJ. Data analysis by MAG. Writing of original draft by MAG and AC. Reviewing and editing of manuscript by all authors. Funding acquisition by AC and TB.