The origin, invasion history and resistance architecture of Anopheles stephensi in Africa

Abstract

The invasion of Africa by the Asian urban malaria vector, Anopheles stephensi, endangers 126 million people across a rapidly urbanising continent where malaria is primarily a rural disease. Control of An. stephensi requires greater understanding of its origin, invasion dynamics, and mechanisms of widespread resistance to vector control insecticides. We present a genomic surveillance study of 551 An. stephensi sampled across the invasive and native ranges in Africa and Asia. Our findings support a hypothesis that an initial invasion from Asia to Djibouti seeded separate incursions to Sudan, Ethiopia, and Yemen before spreading inland, aided by favourable temperature, vegetation cover, and human transit conditions. Insecticide resistance in invasive An. stephensi is conferred by detoxification genes introduced from Asia. These findings, and a companion genomic data catalogue, will form the foundation of an evidence base for surveillance and management strategies for An. stephensi.

Anopheles stephensi is a primary malaria vector native to South Asia and the Persian Gulf(1). Well adapted to urban settings, it was discovered in Djibouti in 2012, and has since been discovered in the wider Horn of Africa (HoA), Yemen, and Kenya, in an apparently ongoing range expansion. An. stephensi is a highly competent vector of Plasmodium falciparum and Plasmodium vivax malaria, and has been associated with substantial increases in malaria cases in Djibouti(2) and Ethiopia(3). The potential spread of An. stephensi to cities across Africa - to an estimated 126 million malaria-naïve people(4) - has raised international alarm(5–7). This has spurred urgent efforts to understand and contain its expansion, particularly in the HoA and Yemen - a region facing geopolitical instability, millions of displaced and vulnerable people, and high susceptibility to the effects of climate change(8). Identifying the origin of An. stephensi, and the landscape and environmental factors facilitating its spread would enable targeted vector control and surveillance at points of entry, and the development and refinement of models predicting both the spread of the mosquito and of potential implementation of biological and gene drive control techniques. Furthermore, identifying threats to insecticide-based control methods via genomic surveillance data will enable more focused and proactive resistance management strategies, supporting more optimal programmes to control and eradicate An. stephensi in its invasive range.

Previous genetic analyses of An. stephensi have found greater similarity between Asian and African mitochondrial barcodes(9, 10), and a stepping stone pattern of diversity and population in structure moving inward into Ethiopia from the coast(11). The relatively coarse resolution of mitochondrial barcodes compared to genomic data(12), a focus on specific geographic regions(11), often with small sample sizes(13), leaves many unanswered questions about the fine-scale structure and invasion history of An. stephensi in Africa. Analysing whole-genome-sequence (WGS) data of individuals sampled systematically from the native and invasive ranges provides clarity on the often complex dynamics of biological invasions, such as invasion timing, origin, and the number of introductions(14). Genomic surveillance studies are transforming the evidence base for the control of both invasive species(14) and malaria vectors, based on the analysis of vector population structures, the distribution of insecticide resistance mechanisms and gene drive targets(15–20). Here, we present a genomic surveillance study of 551 An. stephensi specimens sampled from 9 countries across the invasive and native ranges (Figure 1a, Table S1, Supporting Material). We present evidence of the likely origin and routes of the An. stephensi invasion, characterise the genomic basis of resistance, and provide a foundational genomic data resource for use in combating the challenge to malaria control presented by An. stephensi.

Figure 1. Global population structure suggests an Asian origin of An. stephensi.

(A) map showing sequenced individuals grouped into 12 analysis cohorts based on collection location, principal component analysis (PCA) (B) and neighbour-joining tree (Figure S2) clustering. Shaded region denotes the native range(1), with date of detection in invaded countries and regions (italics). PCA of global samples (B) was downsampled to include no more than 30 individuals per cohort, with shaded regions indicating super-population. (C) a PCA of samples from the HoA and Yemen only. In both PCAs, points are coloured by cohort, and were generated using 1e5 randomly selected SNPs from accessible regions of chromosome 2. (D) Heatmap of mean Hudson’s Fst calculated between cohorts, where 0 indicates no differentiation, and 1 indicates complete differentiation. Fst was generated using SNPs from accessible regions of chromosome 2. (E) A consensus unrooted admixture graph, produced by resampling 1e5 accessible SNPs 1000 times from chromosome 2. Shaded regions denote super-populations, with tips labelled and coloured by cohort. Node labels indicate bootstrap support, and scale bar indicates drift parameter.

An Asian origin of African An. stephensi.

An. stephensi has undergone repeated global range expansions: migrating from Northern to Southern India in the mid-20th century(21), westwards across the Arabian Peninsula in the early 2000s, and from an unknown origin to Africa in 2012 (Figure 1a). Since first detection in the Port of Djibouti(2), An. stephensi has been detected across the HoA region (Figure 1a)(5, 22–26), with possible transient reported occurrences in Nigeria (2020)(5) and Ghana (2023)(27). WGS of 551 An. stephensi, including collections across the HoA and Yemen between 2021–2023, (see Supporting Material and Table S1 for more detail on collection methods, dates, and locations), indicate that An. stephensi is genetically structured into three population groups across its range: Iran and Saudi Arabia (KSA), Central Asia and India, and the HoA and Yemen (Figure 1b, Figure S1). Variation in genetic structure is dominated by the split between the KSA cohorts, and those from Central Asia and Africa. This signal of population structuring is consistent across the An. stephensi autosomal genome (Figure S3), with no sign of variation characteristic of large polymorphic inversions, in contrast to An. funestus and An. gambiae(15, 28). Afghanistan and Pakistan are the closest related population to those in the HoA and Yemen across both the autosomal chromosomes (Figure 1d), and the mitochondrion (Figure S4). Shared selective sweeps and profiles of copy number variation at metabolic insecticide resistance gene families further support a close relationship between Afghanistan and Pakistan and HoA and Yemeni An. stephensi (Figure 2a, b, Figure S5, Table S2). Without further sampling from Pakistan and Northern India, confident inference of the original invasive origin, single or multiple, is challenging, and it is likely that genetic drift brought on by the population bottleneck of the invasion (Figure 3) makes HoA and Yemeni An. stephensi look even more genetically distinct (Figure 1b). Nevertheless, the relative genetic distance between KSA and the HoA and Yemen (Figure 1b, d, e, Figure S1, S2, S3), clearly precludes an introduction from the Arabian Peninsula(2, 29)

Figure 2: Shared selective sweeps and copy number variation suggest a metabolic basis of resistance, shared between African and Asian An. stephensi.

A) genome-wide selection scans (GWSS) using Garud’s H12 statistic. Peaks indicate localised regions of reduced haplotype diversity, which can indicate the presence of a selective sweep. Y axis indicates H12, X axis indicates chromosomal position in 10s of MB. GWSS were performed on phased haplotypes in windows of 5kb. Plots are panelled columnwise by chromosome (label at bottom, see annotations at top for location of resistance genes), and row-wise by cohort (see label). The noisy signal in the HoA and Yemen is a result of numerous ROH (Figure 2b). Table S2 details the presence and sharing of selective sweeps. (B) heatmap of estimated copy number variation (CNVs) at selected metabolic resistance loci. Gene families were selected according to the presence of a selective sweep in (a), and prior involvement in resistance in other Anopheles taxa(17, 19, 58). CNVs were estimated using a HMM of deviations from modal coverage in read alignment files(19). Heatmap columns indicate individual samples, panelled vertically by cohort. Rows indicate genes found in swept regions (Table S2), labelled according to the corresponding ortholog in the An. gambiae PEST genome(59). Genes without an identifiable An. gambiae ortholog are labelled with a ‘U’. Cell darkness indicates increasing copy number (see key). Figure S5 shows further evidence of shared sweeps between populations.

Figure 3. An. stephensi has undergone recurrent bottlenecks and diversity loss in multiple invasions.

(A) shows the nucleotide diversity (π) and (B) Tajima’s D of An. stephensi calculated using accessible SNPs chromosome 2-wide, with error bars indicating 95% confidence intervals estimated by block-jackknife resampling. Point colour and X-axis label indicate cohort, with invasive populations denoted by asterisks. Shaded bars at top show super-population. (B) number (y axis) and fraction of an individual genome (x axis) in runs of homozygosity (ROH). ROH are informative for recent inbreeding, with greater nROH indicating small populations or admixture, and greater fROH indicating recent inbreeding(34). ROH were inferred from accessible SNPs genome-wide with a Hidden Markov Model(15, 60), and filtered to include ROH greater than 1e5 in length(15, 60). Points are coloured by cohort (key).

Spread from Djibouti to the HoA and Yemen

Within each population group, signals of genetic structure and diversity provide an insight into the population history of An. stephensi (Figure 1c, d, e). Samples from the HoA and Yemen form single groups for Sudan, Southeastern Ethiopia, and Yemen, and a diffuse group containing samples from Djibouti and two populations from West/Central Ethiopia (ETW and ETB) (Figure 1c, Figure S1). These cohorts are tightly clustered together on the global PCA (Figure 1b) and neighbour joining trees (Figure S2), show low levels of differentiation with respect to one another (Figure 1d), form a monophyletic group in an admixture graph (Figure 1e), and contain multiple mitochondrial lineages (Figure S4). Furthermore, they share selective sweep and CNV profiles at the same metabolic resistance gene families (Figure 2, Figure S5), and are derived, to varying extents, from the same ancestry components (Figure S6).

Biological invasions may be accompanied by genetic signals of population bottlenecks: reductions in genetic diversity and increased inbreeding(30). An. stephensi shows signs of extensive diversity loss in multiple populations across the sampled range (Figure 3a). The HoA and Yemen, Djibouti and Central Ethiopian cohorts are the most diverse and least inbred. Sudan, Yemen, and Southern Ethiopia are the least diverse, with many individuals from these populations showing signs of intense inbreeding (Figure 3b), suggesting that these cohorts represent more severely bottlenecked, or recently established, An. stephensi populations.

We infer high levels of migration between Djibouti and West/Central Ethiopia (ETW and ETB), Sudan and Yemen, within Sudan and Djibouti and West/Central Ethiopia (Figure 4). Barriers to migration are apparent between Ethiopia and Sudan, Djibouti and West/Central Ethiopia and Southeastern Ethiopia, and Saudi Arabia and the HoA and Yemen (Figure 4a). When considering the spatial signal, the closer relationships between Ethiopia, Sudan, and Yemen populations to Djibouti, than to each other (Figure 1c, d, e, Figure S1), we aver that the An. stephensi invasion of the HoA and Yemen consists of at least three spatially distinct incursions, derived from the initial Djibouti population. Gene flow within the Ethiopian incursion (consisting of samples from Ethiopia and Djibouti City) and Sudanese incursion are significantly higher than gene flow between incursions (Figure 4b, Table S3, Supporting Methods). For the Sudanese and Ethiopian incursions, the further a sample is located from the locations of first detection for these incursions: Port Sudan for Sudan, and Port of Djibouti, for Djibouti and Ethiopia), the less diverse, and more inbred it is likely to be (Table S3, Figure 4c, d).

Figure 4: An initial introduction to Djibouti seeded incursions of Yemen, Ethiopia and Sudan in a stepping-stone pattern.

(A) The grid shows inferred migration (log10(w) across the HoA, Yemen, and Arabian Peninsula. Grid is coloured from dark brown (low) to bright blue (high) between sequenced sampling locations (yellow points). Blue and grey points show where An. stephensi was sampled(61), and found, or not found respectively. The grid was inferred with fEEMS(62), using 1e5 randomly selected SNPs from across Chr2, and selected using leave-one-out cross-validation. Note that edge effects in the model lead to spurious inferences where no data exist (e.g. over Southeastern Ethiopia). Based on the hypothesis (see main text) that initial introduction in Djibouti spawned three geographically separate invasions: into Ethiopia, Sudan, and Yemen, (B) shows a pattern of isolation-by-distance in An. stephensi in Sudan and Ethiopia/Djibouti, with the y axis showing observed genetic distance (in linearised Fst), and the x axis geographic distance, between sampling locations within and between (point colour) invasion subpopulations. (C) and (D) show how nucleotide diversity (π) and inbreeding (base-pairs in ROH) decrease and increase, respectively, in sampling locations, as distance increases from the main port in Sudan (Port Sudan, “P.S.” on the map) and Ethiopia and Djibouti (Djibouti City, “D.C.” on the map). For B, C and D, solid lines and shading show predicted regression lines and 95% confidence intervals from Gaussian (A, D) and gamma (C) GLMs. Colours correspond to different levels of interaction variables (see keys). (E) show rasters of environmental predictors facilitating An. stephensi dispersal: mosquito development rate - MDR, enhanced vegetation index - EVI and transit friction, selected by a Bayesian Gaussian GLM (see main text and methods). Dark - white gradient indicates increasing terrain permissivity (see key).

Previous studies have suggested that human transport links may facilitate An. stephensi dispersal in Ethiopia(11), and various bioclimatic variables have been linked to An. stephensi abundance(4, 31, 32). We investigated the extent to which eight environmental variables may impact An. stephensi gene flow between sampling locations (see Supporting Material, Table S4). We applied circuit theory(33) to estimate the mean value of selected predictors between sample locations, and fitted general linearised models to test for associations between these environmental covariates and matched linearised Fst. These models show that presence of increased enhanced vegetation index (EVI) (mean −0.027, 95% CI −0.036:−0.019) mosquito development rate (mdr), (mean −0.007, 90% CI: −0.017:0.001), and decreased landscape travel friction (mean 0.030 90% CI: 0.018:0.044) (Figure S8) is associated with An. stephensi genetic connectivity (reduced Fst). More variation could be attributed to these variables than isolation-by-distance alone (Figure 4e, Table S5, Supporting Material), suggesting these factors are important determinants of mosquito spread into the HoA.

Genetic diversity and structure in the native range

Genetic structure and diversity of An. stephensi reflect repeated biological invasions worldwide. Southern Indian An. stephensi are less diverse and more inbred than those in Afghanistan and Pakistan (Figure 3a, b), reflecting the encroachment of An. stephensi southward in the mid-20th century(21). Curiously, KSA An. stephensi are particularly genetically depauperate (Figure 3a). The invasive KSA-Riyadh cohort is less diverse still than the native range in KSA-Eastern (Figure 3a), but both KSA cohorts lack the high fraction of the genome in runs of homozygosity (fROH) that would indicate severe inbreeding(34) (Figure 3b); instead low diversity may be a result of older bottlenecks and population expansion, or relatively mild inbreeding over a long period of time. Despite the relative geographic proximity of Iran to Afghanistan and Pakistan, the two Iranian cohorts are more closely related to KSA than to Central Asia and India, or Africa. The Sistan and Baluchistan cohort (IRS), which is geographically and genetically closest to Afghanistan and Pakistan (APA) (Figure 1b, d, e, Figure S1, S2, S6), is the most diverse. The Bandar Abbas cohort (IRH) is much less diverse, more inbred (Figure 3a, b), and more closely related to the KSA cohorts (Figure 1b, 1d, 1e). CNV and selective sweep profiles in Iran contain gene families found both in Afghanistan and Pakistan, and Saudi Arabia (Figure 2a,b, Figure S7): with swept Cytochrome p450 subfamily 6 (Cyp6) in IRS shared between Afghanistan, Pakistan, and HoA, and sweeps at the (Glutathione-S-Transferase) Gste cluster and Rdl being shared between Iran and KSA (Figure S7). Low diversity in KSA An. stephensi may be a result of relative isolation of the population: the historic range covers a slender region of the east coast of KSA(1), where it was first described in 1957(35). In a surveillance program beginning in 1971, An. stephensi was not detected in Kuwait - between KSA and Iran - until 1985(36), raising the possibility that An. stephensi in KSA may have been a historical introduction from Iran across the Persian Gulf.

Evolution and mechanisms of insecticide resistance

Compounding the threat of urban malaria transmission in Africa is widespread resistance of An. stephensi to widely used classes of insecticides(37–40) - the mainstay of malaria vector control efforts. However, there is limited understanding of resistance mechanisms in the native range(37, 41–43), and especially the invasive range(13, 39, 44). We screened the genomic data generated in this study for resistance-determinants via a combination of: searching for target-site resistance mutations in known resistance genes (Figure S7), performing genome-wide selection scans (Figure 2a) and haplotype clustering (Figure S5) to identify regions of the An. stephensi genome under selection, putatively as a result of insecticide selection pressures. Strong selective sweeps at detoxification enzyme families (Figure 2a, S5, Table S2), a lack of selection at target-site resistance genes, and low frequencies of mutations therein (Figure 2a, S7, Table S2), combined with extensive copy number variation at metabolic gene families (Figure 2b), suggests that resistance in An. stephensi in the HoA and Yemen is predominantly mediated through metabolic mechanisms (Supporting text 1). Asynchronous selective sweeps at different metabolic resistance haplotypes of the same gene families: Gste and Cyp6, in KSA, Asia and Africa, suggest convergent evolution through copy number variation at different members of these families may drive resistance in different populations in the absence of gene flow between the Arabian Peninsula and Africa/Asia (Figure 2, Supporting Text 1, Table S2, Figure S5).

On the origin and spread of An. stephensi in Africa

Our data suggest that An. stephensi in the HoA and Yemen may be derived from an initial bridgehead population in Djibouti. This inference is based on three findings: i) that the population now spreading in Africa is very genetically similar (Figure 1c, S1); the presence of Djibouti at the base of a monophyletic clade (Figure 1e) of all HoA and Yemen cohorts, and that all are more closely related to Djibouti and to each other, than to any external population (Figure 1b, c, d, e); ii) a stepping stone pattern of diversity loss in inland, more recently established, populations. (Figure 2, Figure 4c, d); and iii) uniform resistance profiles (Figure 2, Figure S5). A maritime origin hypothesis(45) is supported by the pattern of genetic structuring (Figure 4a), and loss of diversity away from Port Sudan and Djibouti (Figure 4b,c), as well as first discoveries of An. stephensi in coastal cities: Djibouti(2), Port Sudan(5), Aden City(24), Hodeidah(46) and Bosaso(47). Based on the relative similarity of eastern Iranian, Afghan, Pakistan and HoA/Yemen resistance haplotypes and CNV profiles (Figure 2b, Figure S5) a possible scenario is that African An. stephensi originated in southern Pakistan. Widespread resistance in the invasive and native range is likely largely mediated by extensive copy number variation at metabolic mechanisms, supported by gene flow and convergent evolution.

To date, hypotheses for the dynamics and structure of the An. stephensi invasion have been based largely on phylogenetic and haplotype-network based analyses of short cytochrome c-oxidase subunit I (COI) sequences. COI phylogenies, and interpretations of high diversity (relative numbers of COI haplotypes) are used to propose hypotheses of multiple introductions(48, 49), ongoing connectedness with the native range(9) and wind-borne migration of An. stephensi as a primary method of introduction and spread(29). Our analysis of the COI phylogeny failed to resolve any population genetic structure beyond the split between An. stephensi from the KSA and Iran, and Central Asia and Africa (Figure S4b). Moreover, our evidence of severe genetic bottlenecks in Africa and the Arabian Peninsula (Figure 2) contradicts inference of substantial genetic diversity based on the presence of multiple COI haplotypes in the invasive range(9). Network analysis of reduced-representation sequencing data has been used to propose the eastern Ethiopian city of Dire Dawa as a hub of An. stephensi dispersal(11). An alternative explanation is spatial autocorrelation in the genetic data due to the central location of Dire Dawa in Ethiopia(50). Whilst we did not have samples from Dire Dawa in our dataset, similarity of genetic diversity (Figure 3a) and admixture profiles (Figure S6) between Djibouti and West/Central Ethiopia samples, and generally high gene flow in the region (Figure 1d, 4a) make it difficult to pinpoint a specific hub location. More confident inferences of possible dispersal hubs will require sampling schemes explicitly designed to detect spatial genetic structure(51), or specific hypothesised dispersal mechanisms or corridors (e.g. in Aedes spp. mosquitoes(52–55)). Multi-model comparison and selection can then be used to evaluate the relative fits of models of landscape features or dispersal corridors(33, 56) against alternative models, including isolation by distance.

A foundational data resource for genomic surveillance

Genomic data are transforming our understanding of the ecology, evolution, and resistance architecture of malaria vectors(15, 16, 19, 28). These studies are facilitated by genomic data that are accessible, well-documented, and free to use. In an effort to assist the community with the design of resistance allele surveillance, development of phenotyping assays, and gene drive targets, as well as for comparison with new population genomic data, we have made available a mature catalog of SNP, haplotype, and CNV data in partnership with the MalariaGEN Vector Observatory(57). This body of data and analysis will underpin future efforts to predict the spread of An. stephensi, manage and monitor resistance, surveil entry points in targeted vector control efforts, and facilitate the control and elimination of invasive An. stephensi.

Data Availability

Raw read data have been deposited in the European Nucleotide Archive (ENA), accession number ERP160386. Per-chromosome Zarr files containing SNP and CNV data, annotations and accessibility masks will be available from the MalariaGEN Vector Observatory (https://malariagen.github.io/vector-data/) upon publication. Code required to reproduce the bioinformatic workflows and analysis in this study is available at: https://github.com/tristanpwdennis/AsGARD.