Continent-wide population structure of Anopheles funestus.

To investigate the population structure of An. funestus throughout Africa, we removed the two microsatellites putatively under selection to analyse the remaining 9 neutral markers. Genetic divergence (F_st) among countries ranged from 2.9% for Ghana-Uganda to 10.8% for Uganda-Mozambique (S2 Table). Relative divergence among samples from different countries was in broad agreement with their geographical locations, with a major exception: Uganda clustered with Ghana and Cameroon (Fig 1C; F_st 2.9–4.1%) while Benin formed an outlier from the other samples (Fig 1C; F_st 4.9–9.6%). The southern African populations from Malawi and Mozambique were both highly divergent from the other samples (Fig 1C; F_st 6.5–10.8%) though, interestingly, also from each other (6.3%).

The impact of the loci putatively under selection on chromosome 2R was investigated by repeating the analyses on two datasets, one comprising 8 microsatellites on chromosome 2R and one comprising 8 microsatellites on the other chromosomes (S1 Fig). The most striking effect of analysing 2R alone is on Ghana, which becomes highly divergent from all other populations including Benin, the other West African population (S1A Fig; Fst 14.4–21.9%). When the 8 microsatellites on the other chromosomes are analysed, the pattern of differentiation is more aligned with geographical distance between all samples, with Ghana samples less differentiated from Benin (F_st = 4.3%) than with the 2R markers (F_st = 21.9%) (Fig 1C; S1B Fig). The major divergence between southern Africa and other populations is maintained (S1 Fig). This apparent discontinuity between southern and the rest of Africa is consistent with the contrasting resistance patterns and mechanisms seen among An. funestus populations [17] suggesting possible barriers to gene flow between regions. This is supported for example by the complete absence throughout southern Africa of the DDT 119F resistance allele of the glutathione-S transferase gene GSTe2, which is predominant in West/Central Africa [15, 19].

Bayesian analysis of population structure was also undertaken on the 9 putatively neutral loci. The most likely number of clusters estimated from the data was 2, and represents the major divergence between southern Africa and elsewhere (Fig 1D). When analyses were repeated using 8 2R microsatellites and 8 non-2R microsatellites, the most likely number of clusters were 3 and 2, respectively. Analysis of the cluster assignment probabilities (to 3 clusters) for all three datasets showed that Malawi and Mozambique were consistently assigned to their own ‘southern’ cluster, even with the 2R marker set (S3 Table). By contrast, the other populations were more variable, and affected by markers on 2R. For instance, using non-2R markers Ghana and Benin were assigned to the same cluster (44% and 41%, respectively), but using 2R markers they were clearly assigned to separate clusters (Ghana cluster 1: 76% and Benin cluster 2: 65%).

Overall, the analyses are consistent in indicating that gene flow is restricted between southern and the rest of Africa. They also suggest that selection on chromosome 2R may be independent among different geographical regions but appears to be common to both southern African An. funestus populations.

Patterns of polymorphism across the 120kb rp1 genomic region.

To assess the signature of a selective sweep associated with pyrethroid resistance, we carried out fine-scale polymorphism analysis across a 120kb genomic region spanning the rp1 QTL in two highly pyrethroid resistant populations (Malawi and Mozambique) and a moderately resistant population (Cameroon). This revealed markedly reduced polymorphism in the vicinity of CYP6P9a (Fig 2A; S2A and S2B Fig), a key pyrethroid resistance gene, in Malawi and Mozambique but not in Cameroon. The greatest loss of diversity was closest to CYP6P9a at BAC25 (-9kb) (Fig 2A; S4 Table), which also produced geographical clustering based on an ML-tree (Fig 2B). The extent of this low polymorphism in Malawi is characterised by the complete absence of polymorphic sites at +61kb from CYP6P9a (BAC95), only 1 polymorphic site at +86kb, and 2 polymorphic sites at +36kb from CYP6P9a.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Africa-wide analysis of genetic diversity across the rp1 pyrethroid resistance locus.

A) Genetic diversity (pi) across the 120kb rp1 in Cameroon (orange), Malawi (blue), and Mozambique (green). B) Phylogeny of the ‘BAC25’ fragment (located 25kb along the BAC sequence and 9kb upstream of the CYP6P9a gene), which shows the most extreme difference in diversity between Cameroon (orange) and southern populations Malawi (blue) and Mozambique (green), which form a single clade. C) Phylogeny of the CYP6P9a gene sampled from throughout Africa, showing clear geographical divergence between southern Africa (Malawi and Mozambique; 100% bootstrap support), East Africa (Uganda; 87% bootstrap support) and West/Central Africa (Ghana, Benin, Cameroon; 78% bootstrap support). D) Haplotype network for Non-synonymous nucleotide variants in CYP6P9a. Light blue = haplotype shared between Malawi and Mozambique (MAL/MOZ); pink = haplotype shared between Benin, Cameroon and Ghana (BN/CAM/GH); teal = Benin (BN); red = Ghana (GH); orange = Cameroon (CAM); green = Mozambique (MOZ); blue = Malawi (MAL); purple = Uganda (UG). The size of the shapes is proportional to the frequency of the haplotype and numbers on each branch show the mutational steps separating haplotypes.

https://doi.org/10.1371/journal.pgen.1006539.g002

Analysis of the population from Cameroon shows that the relative diversity remains high across the rp1 genomic fragment although a slight reduction is observed at CYP6P9a compared to other loci. Individual analysis of these loci shows that at -34kb from CYP6P9a (BAC 0) high haplotype diversity is noted. No predominant haplotype is observed with the most common haplotype, “CAM1”, having a frequency of 13.1% (5/38) (S2E Fig). A high proportion of haplotypes were singletons (16 out of 23) even in Malawi (7) and Mozambique (4). Additionally, there was a high number of mutational steps observed between haplotypes from each (>12) or between (>20) country (S2E Fig). A drastic change is observed closer to the CYP6P9a at -9kb (BAC25) where a strong signature of positive directional selection is observed in both Malawi and Mozambique. This reduced haplotype diversity is shown by the presence of a highly predominant haplotype MAL/MOZ1 only found in Malawi and Mozambique with a frequency of 92.8% (S3F Fig).

Maximum likelihood phylogenetic trees and haplotype network of rp1.

Construction of Maximum likelihood (ML) phylogenetic trees of the five fragments further highlighted the reduced diversity close to CYP6P9a in the resistant populations of Malawi and Mozambique. The ML tree of the -34kb (BAC0) shows a high diversity of haplotypes for each of the three countries (S2C Fig), while the profile is very different at -9kb from CYP6P9a where all sequences from Mozambique and Malawi belong to a major haplotype while the Cameroon sample retained its high diversity (S2D Fig). Analysis of the TCS haplotype network confirms profiles obtained with the maximum likelihood phylogenetic trees for the 5 loci from -34kb to +86kb from CYP6P9a (S2E and S2F Fig). In general, Malawi and Mozambique formed their own clusters, with reduced genetic diversity shown by the fact that haplotypes from both countries had significantly fewer mutational steps between them (1–3 steps), while the more susceptible samples formed their own clusters with higher mutational steps (4–16 steps) for almost all of the BACs except for +86 (BAC 120).

Genetic diversity of the CYP6P9a pyrethroid resistance gene across Africa.

To further confirm whether the selective sweep was driven by pyrethroid resistance across Africa, the full 1,965 bp of the CYP6P9a resistance gene was sequenced for a total of 59 mosquitoes from six countries (Benin, Cameroon, Ghana, Malawi, Mozambique and Uganda). Analysis of the genetic diversity of CYP6P9a revealed signatures of strong directional selection in Malawi and Mozambique but not in other regions (S5 Table; S3A Fig). In Malawi, only 5 polymorphic sites are observed (S5 Table; S3A Fig) and an AA indel in the 5’ UTR was found mainly in Southern Africa and not in the other countries. Nucleotide diversity (π) was lowest in Malawi (π = 0.0008) and Mozambique (0.0011), compared to Uganda (0.0015), Ghana (0.0020), Benin (0.0022) and Cameroon (0.0029) (S5 Table).

Maximum likelihood phylogenetic tree and haplotype network for CYP6P9a.

Analysis of the ML tree of CYP6P9a further provided evidences of selection acting on this gene in southern Africa in contrast to other regions. CYP6P9a from Malawi and Mozambique formed a defined clade from the other four populations, dominated by a single haplotype (MAL/MOZ23-H25), which is the allele recently shown to exhibit the highest catalytic efficiency in metabolising pyrethroids (S3A Fig) [20]. Elsewhere, Benin and Ghana did not form a common clade despite their geographical proximity. Benin formed a cluster with Cameroon while Ghana was closer to Uganda, but they did not form a single cluster (Fig 2C; S3B Fig). All haplotypes from Uganda formed a cluster intermediate between Southern, West and Central African countries, which correlates well with the genetic structure obtained using microsatellite loci and reflects the geographical position of Uganda in East Africa, between Southern and West/Central Africa.

Genetic differentiation of the CYP6P9a gene.

An index of genetic differentiation based on the nucleotide polymorphism (K_ST) showed that the two Southern African populations of Malawi and Mozambique are genetically close (K_ST = 0.045; P<0.05). However, these highly pyrethroid resistant populations show extensive differentiation from all other populations, with K_ST estimates ranging from 0.347 (Malawi vs. Benin; P<0.001) to 0.574 (Mozambique vs. Uganda; P<0.001). The other four populations exhibit lower levels of genetic differentiation between them than seen between the two southern African populations, although in accordance to geographic proximity, the Benin samples are closer to Cameroon compared to the rest (S6 Table; S3D Fig). However, despite been geographically close to Benin in West Africa, the Ghana sample is more differentiated to Benin (K_ST = 0.50) than the Cameroon populations (K_ST = 0.13) in line with microsatellite results.

Microsatellite genetic diversity pre- and post-intervention.

When the microsatellite diversity of pre- and post-intervention samples was compared, the only major variation occurs on the 2R chromosome, with higher genetic diversity in pre-intervention samples (H = 0.45) (Fig 3A). As already seen for the Africa-wide analysis, reduced diversity was detected on the 2R chromosomes around the rp1 notably for the FunR, AFUB6 and AFND6, markers located close to the pyrethroid resistance gene. Per-locus gene diversity was compared among pre- and post-intervention samples. In both Malawi and Mozambique, the FunR locus showed a striking reduction in gene diversity between collection time-points, from 0.45 in 2002 to 0–0.15 in 2009–2011. The fact that this rp1 region is the only area of the genome with a significant change of diversity level between pre and post-intervention samples further supports the recent occurrence of a selective sweep associated with implementation of vector control interventions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The selective sweep at rp1 in southern Africa coincides with the scale-up in pyrethroid use in malaria control.

A) Gene diversity at microsatellites on chromosome 2R in mosquitoes collected before widespread pyrethroid-based intervention in Malawi, 2002 (purple) and Mozambique, 2002 (red), and ‘post-intervention’ in Malawi, 2010 (green) and Mozambique, 2009 (yellow). B) Fine-scale nucleotide sequence analysis of the 120kb rp1 locus in Cameroon (orange) or pre-intervention samples (Malawi (CKW2002) = brown and Mozambique (MOZ2009) = pink) compared to Post intervention samples from Chikwawa (green), Salima (light blue), Nkhotakota (blue) Malawi, Chokwe (purple) Mozambique, and Kaoma (dark blue) Zambia. C) TCS haplotype network at ‘BAC25’. Haplotypes including sequences from more than one location are denoted by an H and include the frequency, haplotypes from one location Malawi (blue) and Mozambique (yellow) are only present in pre-intervention samples. D) Change in average pairwise diversity (Pi) in the CYP6P9a gene pre- vs. post-intervention in Malawi and Mozambique. E) Phylogeny of CYP6P9a sequences including both Malawi (Ckw) and Mozambique (Moz) shows that post-intervention samples (red) are almost completely homogenous while pre-intervention samples (black) are diverse.

https://doi.org/10.1371/journal.pgen.1006539.g003

Signature of selective sweep around 120kb genomic region of rp1 in southern Africa.

Fine-scale sequence analysis of polymorphism in the 120kb genomic region spanning the rp1 locus revealed a loss of genetic diversity in the post-intervention samples occurring most strongly around the CYP6P9a resistance gene (S3B Fig; S4 Table). In contrast, samples collected from both Cameroon and pre-intervention in southern Africa showed no reduction of diversity across the entire rp1 region (S4A and S4B Fig). Additional mosquitoes collected from across Malawi were sequenced and showed the selective sweep is consistent throughout southern Africa post-intervention (Fig 3B). A haplotype network of the closest fragment to CYP6P9a, BAC25 (-9kb), revealed the striking difference caused by control interventions. 22 unique haplotypes (with up to 23 mutational steps) were observed in pre-intervention samples, but only 2 in post-intervention samples (Fig 3C), which showed evidence of significant positive directional selection using both Tajima’s D (-2.10, -2.97) and Fu and Li’s D* (-2.94, -3.45) (S4 Table). The predominant post-intervention haplotype comprises 95% (68/72) of the sequences, highlighting the risk of a resistance allele becoming rapidly fixed over an 8 year period, from an original frequency of only 7.5%. Additionally, pre-intervention samples had a greater divergence and diversity based on a ML phylogenetic tree and higher numbers of singleton unique haplotypes, while post-intervention samples were nearly homogenous (S4C–S4F Fig).

Comparative analysis of CYP6P9a genetic diversity pre and post intervention.

Analysis of the genetic diversity of CYP6P9a in pre- and post-intervention samples revealed this important pyrethroid resistance gene is under positive selection. Post-intervention samples show a signature of positive directional selection: negative Tajima’s D (-2.43, -2.54) and Fu and Li’s D* (-2.39, -2.25) (S5 Table); reduced pairwise diversity (Fig 3D) and few polymorphic sites (17 combined) (S5A and S5B Fig). In contrast, pre-intervention samples have 68 combined polymorphic sites, higher diversity, and no signature of selection. The ML tree revealed that pre-intervention samples are highly diverse in contrast to post-intervention samples, which are highly homogenous (Fig 3E). The contrast was also evident from the haplotype network, with a predominance of singleton haplotypes pre-intervention (82.5%) compared to only 7.5% post-intervention and the presence of a predominant haplotype in post-intervention samples (67.5%) corresponding to the resistance haplotype previously shown to be driving resistance in southern Africa (S6A Fig) [8]. When an Africa-wide comparison is performed for CYP6P9a polymorphisms, both pre-intervention samples are divergent and diverse while post-intervention samples form a cluster with low divergence (S6B Fig). Further analysis of the polymorphisms of the CYP6P9a coding region detected key amino acid changes shown to provide the highest catalytic efficiency when metabolising pyrethroids (S5B Fig) [20].

(i) Reduced diversity of microsatellites flanking pyrethroid resistance genes on 2R.

This study identified a valley of reduced diversity at the AFUB6 and FunR loci on the 2R chromosome, a typical signature of selective sweep similar to that described in malaria parasites in Southeast Asia in response to pyrimethamine treatment [25]. The reduced heterozygosity of these microsatellite markers and their position around the rp1 pyrethroid resistance QTL point to a selective sweep associated with pyrethroid resistance. This is further supported by the lower diversity observed in the more resistant Southern populations of Malawi and Mozambique [4, 18, 26].

(ii) Reduced polymorphism of the 120kb rp1 genomic region.

Selection in the southern populations is confirmed by the continuous reduced genetic diversity for Malawi and Mozambique and also the reduced haplotype diversity in both countries. Selection is strongest around -9kb from CYP6P9a (BAC-25). This is due to its proximity to the two resistance genes, CYP6P9a and CYP6P9b. However, several other tests that can detect positive selection (MK, HKA, dN/dS and the Ka/Ks ratios) did not show any sign of positive selection. This can be explained by the fact that in a situation of near fixation of a selective sweep as seen in this study for the southern African populations, directional selection is rather indicated by reduced genetic variation [27]. Nevertheless, when estimating Ka/Ks ratios, a 2.64 and 2.5 fold reduction of the Ka/Ks estimates was observed in post intervention samples from Malawi and Mozambique respectively compared to the pre-intervention. The selection on the rp1 region appears to be more extensive in Malawi than in Mozambique in line with previous reports [8]. In Malawi, the region under selection spans from -9kb from CYP6P9a and beyond +86kb, which reasonably could be above a region of 100kb. In Mozambique the region under selection spans from -9kb to +61kb, which is around 70kb.

(iii) Directional selection acting on CYP6P9a gene.

The strong directional selection observed on CYP6P9a is only seen in southern Africa and correlates with the extensive pyrethroid resistance observed in this region with a high over-expression of this gene [4, 8, 18, 26]. The reduced CYP6P9a haplotype diversity with limited mutational steps between haplotypes in southern Africa is indicative of selection while in contrast the west and central African countries maintained high diversity. This lack of selection in other African regions could suggest that rp1 may not be the main resistance region in other populations of An. funestus such as Uganda and Benin where pyrethroid resistance has been reported but with a less extreme over-expression of CYP6P9a [11, 13, 19].

The predominant haplotype in Malawi and Mozambique is a CYP6P9a allele recently shown to have greater efficiency in metabolising pyrethroids compared to susceptible alleles [8, 20]. Vector control interventions such as LLINs and IRS are largely implemented in these countries [4, 28]. This signature of directional selection on CYP6P9a is similar to the selection observed in the glutathione S-transferase GSTe2 gene, another detoxification gene conferring resistance to DDT in An. funestus populations West/Central Africa [19]. Resistant samples from Benin exhibited a nearly fixed GSTe2 haplotype in contrast to DDT susceptible populations where high genetic diversity is observed [19]. Evidence of a selective sweep around insecticide resistance genes has been reported in other species, such as CYP6G1 in D. melanogaster [29] for which a single haplotype, containing a partial Accord transposable element in the 5’ UTR, confers DDT resistance [30].

Patterns of genetic differentiation based on CYP6P9a aligned with F_ST estimates obtained with microsatellites, where Malawi and Mozambique were significantly genetically differentiated from other African populations. The overall pattern of differentiation obtained here with CYP6P9a is also similar to that obtained with GSTe2 gene [19] further supporting the presence of barriers to gene flow between southern Africa and other regions. Furthermore, the predominant CYP6P9a haplotype in southern Africa was completely absent from the other countries in line with the low level of gene flow observed between southern and West/Central Africa. This contrasting distribution of the CYP6P9a resistance haplotype is similar to the distribution of resistance mutations previously reported in An. funestus such as the L119F GSTe2 mutation, which is completely absent in Southern Africa [19]. However, one cannot exclude that the CYP6P9a resistant haplotype will spread to other regions with time as recently observed for the A296S RDL mutation in An. funestus. Indeed the 296S resistant allele initially located only in West to East Africa and completely absent from southern Africa in 2010 [16] was recently detected in southern Africa, although at low frequency [15]. Similar geographical spread of resistance alleles over time has been observed for other resistance mutations such as knockdown resistance (kdr) mutations in mosquito species such as An. gambiae [31] or Ae. aegypti [32, 33].

(iv) Temporal analysis of southern African populations revealed that the scale-up of insecticide-based interventions is driving selection for insecticide resistance.

In theory, causes other than insecticide use in public health could be selecting for insecticide resistance (e.g. agricultural pesticides or general pollution). However, while these may play a role, the scale-up of insecticide-based interventions across southern Africa was the only major influence that changed in such a short period of time (7–8 years). Based on the genetic diversity, haplotype networks, and the predicted maximum likelihood phylogenetic tree and the whole genome sequencing approach, the selective sweep observed around the rp1 in general and particularly on CYP6P9a only happened after implementation of control interventions. Furthermore, the selection to near fixation of an allele highly efficient to metabolize pyrethroids [20] strengthens the hypothesis that this selective sweep was driven by these pyrethroid-based interventions. Such selection of a favourable allele is similar to previous examples of positive selection observed for target site mutations (ex. kdr) conferring resistance in many mosquito vectors [34–38] and selection linked to the Ace-1 gene in An. gambiae [39]. In contrast to these examples, this is the first evidence of a signature of pyrethroid-based vector control such as LLINs on metabolic resistance mechanisms. The near fixation of a resistance haplotype over an 8 year period highlights the need to implement suitable resistance management early enough to prevent control failure. Selection due to pyrethroid interventions is especially alarming because currently pyrethroids are the only insecticide approved for use in bed nets.