Phylogeography and invasion history of Aedes aegypti, the Dengue and Zika mosquito vector in Cape Verde islands (West Africa)

Abstract Aedes‐borne arboviruses have spread globally with outbreaks of vast impact on human populations and health systems. The West African archipelago of Cape Verde had its first outbreak of Dengue in 2009, at the time the largest recorded in Africa, and was one of the few African countries affected by the Zika virus epidemic. Aedes aegypti was the mosquito vector involved in both outbreaks. We performed a phylogeographic and population genetics study of A. aegypti in Cape Verde in order to infer the geographic origin and evolutionary history of this mosquito. These results are discussed with respect to the implications for vector control and prevention of future outbreaks. Mosquitoes captured before and after the Dengue outbreak on the islands of Santiago, Brava, and Fogo were analyzed with two mitochondrial genes COI and ND4, 14 microsatellite loci and five kdr mutations. Genetic variability was comparable to other African populations. Our results suggest that A. aegypti invaded Cape Verde at the beginning of the Holocene from West Africa. Given the historic importance of Cape Verde in the transatlantic trade of the 16th–17th centuries, a possible contribution to the genetic pool of the founding populations in the New World cannot be fully discarded. However, contemporary gene flow with the Americas is likely to be infrequent. No kdr mutations associated with pyrethroid resistance were detected. The implications for vector control and prevention of future outbreaks are discussed.

in infants born to mothers infected during pregnancy, reported in 31 countries (WHO, 2017). For these arboviruses, a proportion of cases is asymptomatic (Bhatt et al., 2013;Petersen et al., 2016), which together with poor disease surveillance and lack of point-of-care diagnostic tests may be potentially misdiagnosed and underreported (Wilder-Smith & Byass, 2016). This is particularly important in Africa where the majority of febrile illnesses are treated presumptively as malaria (Amarasinghe, Kuritsky, Letson, & Margolis, 2011;Stoler & Awandare, 2016). As a consequence, little is known about the epidemiology of these arboviruses in the African continent (Amarasinghe et al., 2011;CDC, 2015;Nutt & Adams, 2017;Sang & Dunster, 2001;Were, 2012). The control of Dengue and Zika is largely dependent on efficient and sustainable vector control measures.
Arboviruses like Dengue, Zika, and Yellow Fever have spread worldwide following the expansion of their main vector Aedes aegypti. This mosquito has originated in Africa from an ancestral sylvatic and more zoophilic form A. aegypti formosus, which expanded from tropical forests to urban areas giving rise to a domestic and anthropophilic form known as A. aegypti aegypti (Bennett et al., 2016;Brown et al., 2011;Moore et al., 2013;Powell & Tabachnick, 2013).
This form was the only that succeeded in invading the rest of the world, forming a monophyletic group (Bennett et al., 2016;Brown et al., 2014Brown et al., , 2011Gloria-Soria et al., 2016). Aedes aegypti arrived to the New World together with the first Europeans and Africans during the historical transatlantic shipping traffic between 1500s and 1700s, followed by the first reports of Yellow Fever and Dengue in the region (Powell & Tabachnick, 2013).
Genetic evidence has confirmed a single out of Africa colonization event for A. aegypti and its historic route of expansion throughout the world (Bennett et al., 2016;Brown et al., 2014Brown et al., , 2011Moore et al., 2013).
After the invasion of the Americas, a cline of reduced genetic variability suggests spreading westwards with subsequent founder events, from the American continent to Asia and Oceania (Brown et al., 2014;Gloria-Soria et al., 2016). A mitochondrial DNA study reported two ancestral clades from which A. aegypti populations outside Africa have arisen (Moore et al., 2013), one associated with West Africa and another with East Africa. Nevertheless, a scenario where populations outside Africa originated from a single sample with the two mtDNA lineages from ancestral Africa cannot be discarded (Gloria-Soria et al., 2016). Presently in Africa, both formosus and aegypti subspecies co-occur and interbreed. The only exception seems to be a rural/forest population from the Rabai District in Kenya, where the two subspecies remain genetically distinct (Brown et al., 2014(Brown et al., , 2011Gloria-Soria et al., 2016). Most likely due to increasing urbanization, populations of A. aegypti in Africa can now be found in many urban settings (Kamgang et al., 2013;Paupy et al., 2008) even if they fall genetically into the formosus group (Gloria-Soria et al., 2016). A study (Crawford et al., 2017) using exome sequences proposed West Africa, in particular Senegal, as the source of the America invasion by A. aegypti. Another recent study (Kotsakiozi et al., 2018), based on SNPs from 20 African populations, suggested Angola as the most likely origin.
The archipelago of Cape Verde is located at around 500 km West of Senegal (West Africa). It comprises nine inhabited islands ( Figure 1) and a current local population of about 500,000 inhabitants. Cape Verde islands had no human occupation when they were first reached by the Portuguese in the 1450s (Lobban, 2018). During most of the 16th and 17th centuries, the islands served as portof-call in a transatlantic commercial slave network (Russell-Wood, 1998), where most of the ships from West Africa to Santiago island would continue to the New World (Lobban, 2018;Russell-Wood, 1998).
Dengue was reported for the first time in Cape Verde at the end of September 2009, in an outbreak considered at the time the largest ever recorded in Africa, having affected all islands. Over 20,000 cases (about 5% of the country population), 174 hemorrhagic fever cases and four deaths were reported (WHO, 2009). Dengue type-3 virus (DENV-3) was confirmed both in Cape Verde and in the concomitant dengue epidemic in Senegal (Franco et al., 2010). In 2015, F I G U R E 1 Map with the location of the Cape Verde archipelago, naming the nine inhabited islands. The Aedes aegypti samples from the present study have been collected from the islands in bold (Santiago, Brava, and Fogo) DENV-2 and DENV-4 were found circulating in field mosquito samples collected in Santiago Island (Guedes et al., 2017). In October 2015, an unprecedented epidemic of Zika virus was reported in Cape Verde (WHO, 2015), with more than 7,500 suspected cases.
By August 2016, 18 microcephaly cases were reported in the islands of Fogo, Santiago, and Maio (Monteiro, 2016). This was the first time that a Zika strain associated with neurological damage in infants was detected in Africa (WHO, 2016). Given this fact, the timing of the epidemic and the high number of travelers visiting Cape Verde from the Americas, it was suggested that the outbreak was likely caused by the Asian genotype circulating there (Lourenço et al.., 2018).
Aedes aegypti is the only mosquito vector of these arboviruses so far detected in Cape Verde. This species was reported for the first time in S. Vicente Island and later on in all islands since 1964 (Ribeiro, Ramos, Capela, & Pires, 1980). Morphological identification of the subspecies was attempted with specimens from a single island (Santiago), and only A. a. formosus was detected (Vazeille et al., 2013). Low vector competence has been associated with African strains of A. aegypti, particularly A. a formosus, with a lower susceptibility to DENV-2 (Diallo et al., 2013;Sylla, Bosio, Urdaneta-Marquez, Ndiaye, & Black, 2009). However, A. aegypti from Santiago Island collected in 2010 showed a moderate ability to transmit the epidemic DENV-3, and high susceptibility to chikungunya and yellow fever viruses (Vazeille et al., 2013). A more recent study based on collections carried out in 2012 found that mosquitoes displayed higher vector competence for DENV-2 and DENV-3 when compared to DENV-1 and DENV-4 .
In Cape Verde, integrated vector control strategies have been directed to both malaria and dengue vectors, Anopheles arabiensis and A. aegypti, respectively, through source reduction, diesel use, biological control with fish (Gambusia sp.), and chemical control with the insecticides temephos and deltamethrin (Ministry of Health Cape Verde, WHO, & University of California, 2012;de Pina, 2013).
The efficacy of insecticide-based vector control has been threatened by the evolution of insecticide resistance worldwide (Smith, Kasai, & Scott, 2016). Insecticide susceptibility tests performed in A. aegypti from Santiago island during the dengue outbreak in 2009 revealed resistance to DDT but susceptibility to pyrethroids (Dia et al., 2012). Later on, A. aegypti mosquitoes collected in the same island in 2012 already exhibited resistance to pyrethroids (deltamethrin, cypermethrin) and to the organophosphate temephos (Rocha et al., 2015). A major resistance mechanism affects a gene in the insect's voltage-gated sodium channel. In this gene, several mutations, known as knockdown resistance (kdr) mutations, have been described and associated with DDT and pyrethroids resistance worldwide (Du, Nomura, Zhorov, & Dong, 2016). These mutations were not detected in A. aegypti collected in 2012 (Rocha et al., 2015).
Since Cape Verde is in a strategic route linking Africa, Europe, and the Americas, by sea or by air, thus having a high risk of introduction of new strains of arboviruses or new vectors, it is of great importance to study the local population structure of A. aegypti. Therefore, we have developed a comprehensive phylogenetic and population genetics study of A. aegypti from three islands of Cape Verde with samples collected before and after the dengue outbreak of 2009. We have measured levels of genetic variability and determine the population structure, inferred the evolutionary history and detect possible origins for this insular population, and evaluated the allelic composition in three kdr sites.  Table 1). The samples consisted of larvae or pupae collected from 12 breeding sites. These were reared to adult stage and then morphologically identified to species (Ribeiro & Ramos, 1995;Ribeiro et al., 1980). The detailed procedure of the entomological survey has been described elsewhere (Alves et al., 2010). Adult specimens were individually preserved in tubes filled with silica gel and kept at room temperature until DNA extraction. Genomic DNA was extracted using a phenol: chloroform protocol (Donnelly, Cuamba, Charlwood, Collins, & Townson, 1999).

| Mitochondrial DNA sequencing and phylogenetic analysis
We have analyzed two mitochondrial genes: cytochrome oxidase subunit I gene (COI) and NADH dehydrogenase subunit 4 gene (ND4). The gene COI was amplified with primers published by (Paupy et al., 2012 described by Paduan and Ribolla (2008). PCR conditions for both genes were optimized by Seixas et al. (2014).
PCR products were purified and sequenced directly with the PCR primers (forward and reverse sequences obtained for each individual). Only sequences with no double peaks, suggestive of nuclear mtDNA segments (or NUMTs), were subsequently analyzed.
Bayesian Markov chain Monte Carlo inference, as implemented in BEAST 1.8.4 (Drummond, Suchard, Xie, & Rambaut, 2012), was used to reconstruct the phylogenies that best describe the evolutionary history of the data set, given the sequences and a priori information of the geographic traits. The substitution model HKY (Hasegawa, Kishino, & Yano, 1985) with gamma and invariant sites and three partitions into codon positions were used. Furthermore, a Bayesian skyline population growth model was used (Drummond, Rambaut, Shapiro, & Pybus, 2005) which allows for flexibility in demographic reconstruction and a strict molecular clock was assumed with a substitution rate of 0.023 for COI and 0.0125 for ND4, as previously described (Brower, 1994;Yu et al., 1999). The analyses were run in 3-5 separate independent runs in BEAST with 100,000,000 generations, sampled every 50,000 runs (ND4), 100,000,000 sampled every 10,000 runs (COI). To analyze convergence and stability, we used TRACER v1.6 software (Rambaut, Drummond, Xie, Abebe, & Suchard, 2017).
TREEANNOTATOR (Drummond et al., 2012) was used to estimate the final Maximum Clade Credibility Tree with a burn-in of 10%. Trees were visualized and edited with FIGTREE 1.4.3. (Rambaut, 2016).
In order to more accurately estimate the time to the most recent ancestor (TMRCA), the clade with Cape Verdean haplotypes obtained from the previous analysis was re-analyzed in BEAST.
Because evolutionary relationships among populations within species can be reticulate rather than bifurcating, haplotypes were connected on median-joining networks followed by maximum parsimony to eliminate unnecessary median vectors and links (Bandelt, Forster, & Röhl, 1999) with the software NETWORK 5 available at website fluxus-engineering.com. This program was also used to connect ND4 haplotypes from Cape Verde and Africa (54 sequences). In this case, we have applied a star contraction preprocedure.
In order to estimate levels of contemporary gene flow between Cape Verde and other African populations, we have also used the above-mentioned microsatellite database to detect recent migrants with assignment tests performed with GENECLASS v.2 (Piry et al., 2004). The Bayesian-likelihood criterion (Rannala & Mountain, 1997) and the as assignment criterion likelihood ratio (L): L_home/ L_max (Paetkau, Slade, Burden, & Estoup, 2004) was used with 1,000 simulated individuals.
Bonferroni corrections were used to adjust critical probability values for multiple tests (Rice, 1989).

| kdr mutations
We

| ND4 mitochondrial gene
The alignment of 360 bp of the ND4 gene sequenced in 42 Cape Verdean A. aegypti samples resulted in seven distinct haplotypes (Table S1) (Yu et al., 1999).
The median-joining network based on the ND4 gene revealed closely related haplotypes in a star-like topology ( Figure S1a). The central haplotype ND4-1 was the most abundant (57% , Table S1) and the only present in the three islands sampled, while the remain-  3,866-7,827), assuming a substitution rate of 0.023 (Brower, 1994).

| COI mitochondrial gene
The median-joining network based on the COI gene revealed also related haplotypes in a star-like topology ( Figure S1b). Haplotype

| Microsatellite genetic diversity
We genotyped 70 mosquitoes from Santiago Island, 47 individuals collected in 2007 and 23 in 2010. The 14 loci were polymorphic with allele richness ranging from 4 to 8 and private allele richness from 0 to 4. The expected heterozygosity varied between 0.494 and 0.783 (Table S3).
The two temporal samples showed low but significant (p < 0.0001) genetic differentiation with Rst = 0.027 and Fst = 0.043.
Estimates of current effective population size decreased from 55 in 2007 to 8 in 2010, with nonoverlapping 95% confidence intervals, indicative of significant differences in Ne between years (Table 3).
Tests performed with BOTTLENECK v.1.2.02 (Piry et al., 1999) revealed significant heterozygote excess in the sample of Cape Verde 2010 under the TPM (Table 3). from West Africa, 21 and 23 from East Africa), from the rest (orange cluster). In order to better detail this clustering result, we selected individuals from the blue cluster with a probability of assignment (Q)> 0.90 re-analyzed them in STRUCTURE v.2.3.4 (Pritchard et al., 2000) with the same previous conditions. The result is presented in Figure 4c, including 59 out of the original 71 Cape Verdean samples.

| Population structure
The best K was again K = 2, but this time, the Senegalese population of Goudiry (8 in Figure 4) grouped with the Kenyan Rabai-outdoor collection (21 in Figure 4), leaving the rest to another cluster, where Cape Verde was incorporated.
GENECLASS v.2 (Piry et al., 2004) assigned only one potential contemporary migrant from Goudiry, Senegal (population 8 in Figure 4) to Cape Verde 2010, with a L = 4.090. No migrants from Cape Verde to continental Africa were detected.

| kdr mutations
The analysis of the kdr gene from 74 individuals revealed that the A. aegypti population from Cape Verde was monomorphic with 100% of wild-type alleles at the three loci analyzed (i.e., 1011I, 1016V, and 1534F).

Phylogeographic and population genetic analysis of A. aegypti from
Cape Verde Islands revealed the main elements of the invasion history and recent demographic events of this insular mosquito population.
Our results suggest an ancient West African origin and no evidence of recent founder events. We report comparable levels of genetic diver-

| West African origin
Clustering analysis of microsatellite data indicates that the Cape Verdean A. aegypti population belongs to the subspecies formosus.
This result agrees with previous analyses based on morphological data (Vazeille et al., 2013). Microsatellite data also suggest a West African origin, possibly having A. aegypti from Senegal as the source population. The levels of expected heterozygosity are within the average of African populations (Gloria-Soria et al., 2016) and similar to the only African insular sample from Guinea-Bissau (sample 10-Bijagos, GW in Figure 4). All other island populations analyzed in that study showed lower levels of genetic diversity than Cape Verde.

| Invasion history and expansion time
Our analysis suggests that the TMRCA of Cape Verdean and West African mtDNA lineages is in the late Pleistocene (ND4Median = 443,000 years, COI Median = 126,000 years, 95% HPD = 61,000-686,000).
In the three Cape Verdean islands surveyed, we found several unique haplotypes very closely related in a star-like network in both ND4 and COI ( Figure S1). Given the highest haplotype diversity in Santiago Island and ubiquity of central haplotypes in the trees (ND4-1 and COI-15 in Figure S1), it is likely that the colonization of the more western and smaller islands of Fogo and Brava was made from the island of Santiago (Figure 1). However, a more detailed history of these founder events is Focusing on Santiago Island from 2007, besides the already mentioned star-like tree of very close haplotypes, we found a concordant unimodal mismatch distribution of mutations and also a significantly negative Fu's FS neutrality test (Table 2). Together, these are genetic signatures of population expansion (Fu, 1997;Harpending, 1994 (Heatwole & Shine, 1976) or bird (Tandon & Ray, 2000) hosts already present on the island. In Cape Verde, there is evidence for reptile presence before 6.2 mya in the late Miocene (Carranza, Arnold, Mateo, & López-Jurado, 2001) and bird species establishment during the Pleistocene (2.6-0.01 mya) (Bourne, 1951).
Cape Verde played an important role in the historical slave trade network (Lobban, 2018). Between the 16th and 17th century, Santiago island was an obligatory stop for ships going from the coast of Angola, São Tomé, Ghana, and Guinea-Bissau to Brazil and the West Indies (Lobban, 2018;Russell-Wood, 1998  with the first reports of resistance to these insecticides (Rocha et al., 2015). This reinforces the hypothesis of insecticide-based reduction of Ne of the mosquito population, and the increase of selection pressure for insecticide resistance during the 2009 dengue epidemic.

| Epidemic preparedness and implications for vector control
The role of airports and airlines in the spread of vector-borne diseases has been helpful in predicting the risks of vector-borne disease importation and establishment Tatem et al., 2012). Cape Verde is in a strategic route linking Africa and South America. The major Dengue outbreak in Cape Verde was caused by a DENV3 virus that originated from Senegal (Franco et al., 2010)  Zika outbreaks (WHO, 2009(WHO, , 2015. In this context, the adoption of alternative noninsecticidal vector control strategies should be considered a priority for the Cape Verde health authorities.

CO N FLI C T O F I NTE R E S T
None declared.

DATA ACCE SS I B I LIT Y
Data for this study are available at GenBank (MK359818-MK359844) and VectorBase (VBP0000346) .