Discussion
We have identified a sequential series of fitness-augmenting mutations in An. gambiae culminating in a triple-mutant haplotype with a large effect on pyrethroid resistance and which is spreading rapidly across East and Central Africa. The mutation that is probably the oldest in this series, the insertion of ZZB-TE, was first detected in 2004 in the malaria-endemic area around Lake Victoria, with theCyp6p4-236M SNP evident in 2005 samples and the third, a duplication in Cyp6aa1 detected in 2008. In samples collected only five years later, the triple mutant was detected hundreds of kilometres away in the DRC. These patterns suggest both a large fitness advantage arising from the triple mutant, and a frightening speed at which resistance-conferring mutations are able to spread within and across populations.
Second generation nets treated with the synergist PBO were shown to be much more effective than conventional nets in killing mosquitoes in populations where the triple mutant haplotype is present (Figure 5). Therefore the use of PBO bednets should be prioritised in the regions where this mutation is present. A strong corollary of this finding comes from the cluster randomised control trial conducted in Uganda, where the mutation is at high frequency, which demonstrated that malaria parasite prevalence in children <10 years old (12% vs 14%; Prevalence ratio = 0·84, 95% CI 0·72–0·98; P=0·029) and mean number of mosquitoes (Density ratio=0·25, 95% CI 0·18–0·3; P <0·0001) per house were significantly lower in villages that had received PBO LLINs relative to standard LLINs [4]. There is some evidence that the haplotype may be less strongly associated with resistance to permethrin than deltamethrin (and perhaps also alphacypermethrin), although both pyrethroids were metabolised by Cyp6aa1 and Cyp6p4 .
Our results highlight the importance of gene duplications for the evolution of insecticide resistance. In An. gambiae , duplications have recently been shown to be concentrated in regions associated with metabolic resistance, and over 40 such duplications have been described across the genome [20]. Thirteen different duplications have so far been described that encompass Cyp6aa1 (Figure 1), both in West and East Africa and in the two sister-species An. gambiae ss and An. coluzzii [20]. It seems likely that these other Cyp6aa1duplications are also associated with pyrethroid resistance. For example, in An. coluzzii sampled from a highly insecticide-resistant population from Cote d’Ivoire, greater than 95% of individuals had Cyp6aap duplications (Supplementary materials Appendix 5). Five different duplications were observed in the collections from 2012 and 2017 and, whilst we detected no association with pyrethroid resistance, two duplications (Cyp6aap-Dup7 andCyp6aap-Dup14 ) showed significant increases in frequency over time. Moreover the total number of CNVs per sample (measured as presence of each of the 5 duplications, summed for each sample) increased significantly from an average of 1.59/ individual in 2012 to 2.08 in 2017 (Mann Whitney U test; P<0.0001); the mean number of duplications >2 indicates that there are multiple CNVs on the same haplotype. Duplications of the Cyp6aa1 orthologue have been found in another important malaria vector, An. funestus , from west and central Africa [21]. The Cyp6aa1 orthologue inAn. funestus, which shares an 87% identity with An. gambiae, was also observed to metabolize permethrin and deltamethrin and, when expressed in transformed Drosophila , was associated with significant increases in resistance to both permethrin and deltamethrin relative to control mosquitoes [21]. This evolution of multiple Cyp6aa1 duplications suggests this is an important Africa wide resistance mechanism
There are now several cases of insecticide resistance evolution where an initial mutation in a genomic region is followed by the spread of additional mutations on the resistant haplotype background [14, 38-42]. It is not yet clear whether the sequence of mutations that we have identified rely on each-other for effect, and thus could only have spread sequentially, or whether each additional mutation coincidentally appeared on the background of an already common mutant haplotype. In the case of Cyp6p4-236M and Cyp6aap-Dup1 it seems that the latter is more likely, although we cannot exclude the possibility that the duplication affects the regulation of Cyp6p4 . In contrast, the putative enhancer inserted with the ZZB transposon may affect either or both of Cyp6p4 and Cyp6aa1 , and may thus interact with the other two mutations in ways that have yet to be determined. Transposable elements can sometimes affect gene expression of neighbouring genes [38] and are abundant in mosquito genomes [43, 44]. Interestingly, in the common house mosquito, Culex pipiens , TEs have been found in the flanking regions of theEster locus, a genomic region in which many independent gene duplications have arisen and spread worldwide in response to selection from organophosphates [45]. Clearly, both TEs and gene duplications are an understudied, yet common source of variation that may have important implications for vector control efforts. The appearance and rapid spread of the three mutations described here is broadly coincident with the scale up of LLIN coverage in DRC, Kenya and Uganda. The haplotype is a strongly predictive marker of high-level resistance to pyrethroids, is easily screened with a single diagnostic assay and we suggest should be used for both insecticide resistance monitoring strategies and for informing LLIN selection.[6, 29, 46].