Introgression Shapes Genetic Variation Associated With Humid Environments in Aedes aegypti Mosquitoes From Saudi Arabia

ABSTRACT

Introgression between genetically distinct populations generates variation that may promote adaptation to diverse environmental conditions. In the mosquito Aedes aegypti , the forest‐associated subspecies Aedes aegypti formosus (Aaf) underwent a domestication process approximately 5000 years ago, giving rise to the globally invasive, human‐associated arbovirus vector Aedes aegypti aegypti (Aaa). Despite extensive human‐mediated transport, the two subspecies remain largely distinct, with admixture documented mainly in coastal Africa and Saudi Arabia. We examined Saudi Arabian Ae. aegypti across climatically contrasting regions – the humid south and arid west‐ using a Saudi Arabian dataset (11,939 SNPs) to examine within‐country genetic and environmental variation and a global dataset (13,455 SNPs) to assess ancestry and admixture in a worldwide context. Genomic analyses revealed strong west–south differentiation, with Jazan Highland showing admixture between subspecies and the highest proportion of Aaf ancestry. Multiple pulses of gene flow into and within southern Saudi Arabia suggest Aaf populations act as a reservoir of genetic variation. Genome‐wide scans identified a candidate region associated with the humidity gradient, harbouring an elongase of very long‐chain fatty acids (AAEL024147), essential for cuticle production and minimizing water loss. Because the Saudi Arabian genetic background is predominantly the arid‐adapted subspecies Aaa, the presence of Aaf ancestry at this locus may reflect localized introgression with adaptive relevance under an increasingly humid environment. Our findings highlight how secondary contact between divergent lineages may contribute to local adaptation in Ae. aegypti, with implications for predicting responses to environmental change.

1. Introduction

Introgression is the transfer of genetic material between divergent populations or lineages via hybridization and repeated backcrossing. It commonly follows secondary contact and has been documented across diverse taxa, often producing heterogeneous genomic signatures shaped by demography, reproductive isolation, and selection (Hedrick 2013; Taylor and Larson 2019). Introgression can also provide raw genetic variation on which selection may act, particularly in rapidly changing or heterogeneous environments. Adaptive introgression occurs when the introduced variation increases fitness (Hedrick 2013) and has been associated with stress resistance in plants (Whitney et al. 2010) and locally adaptive traits in animals (Huerta‐Sánchez et al. 2014; The Heliconius Genome Consortium 2012). In mosquitoes, hybridization and introgression have been documented in Culex and Anopheles species (Fonseca et al. 2009; Norris et al. 2015). However, the evolutionary consequences of introgression in Aedes aegypti mosquitoes, particularly its potential contribution to local adaptation, remain less understood.

Aedes aegypti is the primary mosquito vector of arboviral diseases such as dengue and Zika worldwide. The species originated in the islands of the Southwest Indian Ocean and spread to Africa ~85,000 years ago (Soghigian et al. 2020). It exists as two subspecies with distinct ecology, behaviour, viral transmission capacities, and geographical distributions: the dark forest‐dwelling Aedes aegypti formosus (Aaf) from sub‐Saharan Africa and the paler, globally distributed domestic form Aedes aegypti aegypti (Aaa). The latter spread from West Africa via global trade routes to the rest of the tropics and subtropics (Abuelmaali et al. 2024; Gloria‐Soria et al. 2016; Powell et al. 2018). Unlike the sylvatic Aaf, the Aaa mosquito has evolved behaviours that promote human contact, such as entering homes, using man‐made containers as larval habitats, and exhibiting a strong preference for human scents and blood (Bennett et al. 2021; Brown et al. 2014; McBride et al. 2014; Rose et al. 2020). It has also been suggested that Ae. aegypti evolved enhanced drought tolerance in environments with limited water availability, together with an increased reliance on human water storage (Rose et al. 2020). This adaptation to human‐associated habitats has facilitated the global spread of Aaa (Bennett et al. 2021; Powell and Tabachnick 2013), with human movement and trade promoting introductions into new regions (Gloria‐Soria et al. 2014, 2022; Lima et al. 2016; Pless and Raman 2018; Seixas et al. 2019), and increasing opportunities for secondary contact among genetically differentiated populations. Until recently, evidence of admixture between Aaa and Aaf was geographically restricted to Africa, where introduced Aaa has come into secondary contact and hybridized with local Aaf populations, including in coastal regions of Senegal, Kenya, Sudan, and Burkina Faso (Crawford et al. 2025; Elnour et al. 2022; Gloria‐Soria et al. 2016; Mulwa et al. 2025). Notably, Aaf ancestry has now been detected in Saudi Arabia, providing evidence of secondary contact between these divergent lineages outside Africa (Mashlawi et al. 2024).

Aedes aegypti in Saudi Arabia is abundant year‐round, but the timing and intensity of peak activity vary regionally in association with rainfall and temperature (Alahmadi and Atkinson 2019). This spatial variation occurs across a strongly heterogeneous precipitation landscape, ranging from the rain‐fed western and southwestern highlands to the hyper‐arid interior plateau and the coastal plain along the Red Sea (Figure 1A). Aaa populations have been reported from arid regions of Saudi Arabia (Crawford et al. 2025; Gloria‐Soria et al. 2016; Mashlawi et al. 2024), whereas a strong signal of Aaf ancestry was described in the Jazan Highlands of southwestern Saudi Arabia (Mashlawi et al. 2024), a region characterized by relatively humid conditions and low levels of urbanization.

FIGURE 1.

Genetic differentiation between Aedes aegypti mosquitoes from two distinct regions in Saudi Arabia. (A) Schematic map of Saudi Arabia (white area) and its neighbouring countries (grey area), showing the collection sites of Ae. aegypti mosquitoes included in this study. The Saudi Arabian territory was coloured according to the levels of precipitation during the driest month (mm). (B) Principal component analysis (PCA), with values on each axis representing the percentage of total variation explained by each component; individuals are represented by dots and populations coded by colour. (C) Pairwise F _ST values among all Saudi Arabian samples. (D) Admixture analysis showing K = 2 and K = 3 genetic clusters, with K = 2 identified as the optimal number of clusters (see Figure S3A). The height of the colour bar indicates the proportion of individual ancestry assigned to each genetic group identified by ADMIXTURE, represented by different colours. Individuals belonging to multiple genetic clusters are shown as bars with varying colour proportions. Thick black lines within the plots delineate population boundaries. (E) Mean linkage disequilibrium (r ²) as a function of physical distance (kbp). The inset shows the LD₅₀ distance (kbp), calculated as the distance at which r ² decays to half of its maximum value (r ² _max), along with the corresponding 95% confidence intervals. Each population was highlighted using the same colours as in (A). (F) RDA analysis where the first two axes are shown, with points representing sampling locations coloured by precipitation levels in each collection site. Arrows indicate environmental variables and Moran's eigenvector maps (MEMs) included in the model, with arrow length and direction reflecting the strength and orientation of their association with the constrained genetic variation. Percentages on the axes indicate the proportion of variance explained by each constrained axis.

Recent shifts in rainfall patterns and increased climate variability in the Jazan region (Islam et al. 2025) are expected to impose changing ecological pressures on the local populations. Resident Ae. aegypti populations may respond to these conditions through selection on standing or newly arising genetic variation. The introduction of Aaf into Saudi Arabia, likely facilitated by human‐mediated movement, could bring new genetic variants into the region through introgression, as it creates opportunities for secondary contact, hybridization, and gene flow with local Aaa populations.

We hypothesize that Aaf‐derived introgression has facilitated local adaptation of Saudi Arabian Ae. aegypti to humid conditions. Here, we examine the genetic structure, gene flow, and genomic variation of Saudi Arabian Ae. aegypti populations and further investigate Aaf ancestry in southern populations to explore the evolutionary mechanisms underlying its persistence in the region.

2. Materials and Methods

2.1. Mosquito Collections

Aedes aegypti samples were collected from seven locations in Saudi Arabia, spanning years 2012–2022 (Figure 1A), with a minimum of 10 individuals sampled per site (Table 1), as described in (Gloria‐Soria et al. 2018; Mashlawi et al. 2024). This investigation spans multiple locations across Saudi Arabia, including four southern populations—Jazan Lowland, Jazan Highland, Sahil, and Najran—and three western populations—Jeddah, Makkah, and Madinah; as shown in Figure 1. Briefly, mosquito samples were collected from a combination of urban, semi‐rural, and rural areas, at different altitudes, including villages, with distances between sampling locations ranging from 22 to 58 km. Immature mosquito stages were collected with a larval dipper and reared at the Saudi Center for Disease Control and Prevention, Ministry of Health, Sabya. Mosquitoes were reared at 28°C ± 2°C and relative humidity of 75% ± 10%, as in Mashlawi et al. (2024). Species identification was performed using morphological characteristics and a mosquito taxonomic key (Rueda 2004). Mosquitoes from Jeddah and Makkah collected in 2012 were also included in this dataset (Soghigian et al. 2020).

TABLE 1.

Aedes aegypti larval collection sites.

Mosquito subspecies	Sites	Country	ISO code	Continent	Region	Latitude	Longitude	Year	Unrelated individuals	References
Ae. aegypti formosus	Johannesburg	South Africa	ZA	Africa	East	−25.9904	27.9006	2015	7	Kotsakiozi, Evans, et al. (2018)
	Nairobi	Kenya	KE	Africa	East	−1.2897	36.8180	2012	8	Kotsakiozi, Evans, et al. (2018)
	Karary	Sudan	SD	Africa	East	16.2091	32.3841	2019	12	Elnour et al. (2022)
	Al Nasr	Sudan	SD	Africa	East	15.6556	32.5229	2019	9	Elnour et al. (2022)
	Lunyo	Uganda	UG	Africa	East	0.0721	32.4634	2013	5	Kotsakiozi, Evans, et al. (2018)
	Yaoundé	Cameroon	CM	Africa	West	3.8667	11.5167	2014	8	Kotsakiozi, Evans, et al. (2018)
	Benoue Park	Cameroon	CM	Africa	West	8.3711	13.8660	2014	9	Kotsakiozi, Evans, et al. (2018)
	Franceville	Gabón	GA	Africa	West	−1.6332	13.5830	2014	14	Kotsakiozi, Evans, et al. (2018)
	Lope Village	Gabón	GA	Africa	West	−0.3790	11.5274	2014	13	Kotsakiozi, Evans, et al. (2018)
	Sedhiou	Senegal	SN	Africa	West	12.7070	−15.5552	2012	5	Kotsakiozi, Evans, et al. (2018)
	Goudiry	Senegal	SN	Africa	West	14.1844	−12.7167	2012	9	Kotsakiozi, Evans, et al. (2018)
Ae. aegypti aegypti	Punjab	Pakistan	PK	Asia	West	33.9072	72.3111	2010	13	Kotsakiozi, Gloria‐Soria, et al. (2018)
	Cebu City	Philippines	PH	Asia	East	10.2833	123.9468	2013	5	Kotsakiozi, Evans, et al. (2018)
	Bangkok	Thailand	TH	Asia	East	10.3575	123.8518	2006	6	Kotsakiozi, Evans, et al. (2018)
	Ho Chi Minh	Vietnam	VN	Asia	East	10.8032	106.6947	2013	18	Kotsakiozi, Evans, et al. (2018)
	Tbilisi	Georgia	GE	Asia	West	41.9614	43.3624	2015	4	Kotsakiozi, Gloria‐Soria, et al. (2018)
	Hopa	Turkey	TR	Asia	West	39.5989	36.4483	2015	6	Kotsakiozi, Gloria‐Soria, et al. (2018)
Ae. aegypti aegypti	Jazan Lowland	Saudi Arabia	SA	Asia	East	17.2398	42.6024	This study	15	This study
	Jazan Highland	Saudi Arabia	SA	Asia	East	17.2446	43.0921	This study	14	This study
	Madinah	Saudi Arabia	SA	Asia	East	24.4565	39.5929	This study	13	This study
	Najran	Saudi Arabia	SA	Asia	East	17.4946	44.1378	This study	13	This study
	Sahil	Saudi Arabia	SA	Asia	East	18.0034	41.6760	This study	14	This study
	Jeddah	Saudi Arabia	SA	Asia	East	21.4941	39.2162	2012	10	Gloria‐Soria et al. (2018)
	Makkah	Saudi Arabia	SA	Asia	East	21.4060	39.8106	2012	12	Gloria‐Soria et al. (2018)

For global comparisons, previously genotyped Ae. aegypti mosquitoes representing the two subspecies were included in the analysis. These samples were collected worldwide from 2010 to 2020 (Elnour et al. 2022; Kotsakiozi, Evans, et al. 2018; Kotsakiozi, Gloria‐Soria, et al. 2018). The global dataset included 242 unrelated samples that we grouped by geographical region: Aaf from East Africa: Johannesburg (South Africa, ZA), Nairobi (Kenya, KE), Al Nasr and Karary (Sudan, SD), and Lunyo (Uganda, UG); Aaf from West Africa: Yaoundé and Benoue Park (Cameroon, CM), Franceville and Lope Village (Gabon, GA), Goudiry and Sédhiou (Senegal, SN); and Aaa from Asia: Punjab (Pakistan, PK), Cebu City (Philippines, PH), Bangkok (Thailand, TH), Ho Chi Minh (Vietnam, VN), Tbilisi (Georgia, GE) and Hopa (Turkey, TR) (see Table 1).

2.2. DNA Extraction and Genotyping

Mosquitoes were processed and genomic DNA extracted using the DNeasy Blood and Tissue Kit (QIAGEN Sciences, Germantown, MD, USA), following the manufacturer's instructions. Genotypes were inferred using the Axiom aegypti1 genotyping array (Life Technologies Corporation, CAT#550481) specifically designed for Ae. aegypti (Evans et al. 2015). This genotyping array includes 50,000 Single Nucleotide Polymorphisms (SNPs) and was designed using representative Ae. aegypti populations from around the world to capture the species' genetic diversity, which facilitates the identification of the two Ae. aegypti subspecies (Evans et al. 2015; Gloria‐Soria et al. 2018; Gómez‐Palacio et al. 2024). The genotyping was done at the Functional Genomics Core, University of North Carolina, Chapel Hill, NC, USA.

Data files were processed with the Axiom Analysis Suite (v5.4) software (Affymetrix, Santa Clara, CA) to retain high‐quality samples based on the following parameters: sample's dish quality control (DQC) ≥ 0.82, sample's QC call rate (QC) ≥ 66, average call rate for passing samples (ACRPS) ≥ 90. Genotypes were inferred using the same software on samples that passed all quality filters. We retained only loci that met Mendelian expectations (Evans et al. 2015) and whose probes were previously shown to map uniquely to the AaegL5 genome assembly (Gómez‐Palacio et al. 2024; Matthews et al. 2018). Using PLINK (v1.9) (Purcell et al. 2007), loci and individuals with high levels of missing data were excluded by removing loci with > 20% missing genotypes (‐‐geno 0.2) and individuals with > 5% missing data (‐‐mind 0.05), followed by the exclusion of loci with a minor allele frequency (MAF) < 0.01.

Inclusion of related individuals can bias estimates of population genetics and demographic parameters by overrepresenting family structure. Larvae collected from the same water source may include closely related individuals, as multiple offspring from the same female can occur within a single breeding container (Colton et al. 2003). Therefore, sibling mosquitoes were identified by estimating pairwise relatedness using the KING coefficient (Manichaikul et al. 2010) in VCFtools (v0.1.16; ‐‐relatedness2) (Danecek et al. 2011). Individuals in a pair with a KING coefficient ≥ 0.177 were considered first‐degree relatives, so one individual was removed from each pair prior to downstream analyses.

Finally, loci in linkage disequilibrium were pruned with PLINK using the indep‐pairwise filter with 50‐SNP windows, a 10‐SNP step size, and a squared correlation coefficient threshold of (r ²) = 0.3. After this filtering process two datasets were obtained: one comprising only Saudi Arabian mosquitoes (N = 88) and another including worldwide samples (N = 242). The datasets retained 11,939 and 13,455 high‐quality genomic sites, respectively. For some analyses, we used a reduced worldwide dataset that excluded populations from Senegal, Cameroon, and Gabon (N = 174) to focus the introgression analysis on Saudi Arabian populations within a global context, while minimizing potential confounding effects from previously documented admixture in these countries (Crawford et al. 2025; Gloria‐Soria et al. 2016; Kotsakiozi, Evans, et al. 2018; Kotsakiozi, Gloria‐Soria, et al. 2018; Rose et al. 2020), which was outside the scope of the present study. Details of the software and databases used for data processing and downstream analyses are provided in Tables S1 and S12.

2.3. Environmental Characterization and Variable Selection

High‐resolution bioclimatic variables (Bio1–Bio19) and elevation were obtained from the WorldClim database (Fick and Hijmans 2017). Elevation was included because sampling sites covered a wide altitudinal range, from 23 m above sea level in Jeddah to 1588 m in Najran (Figure S1D). To identify the predominant climatic gradients and the environmental variables contributing most strongly to variation across sampling locations in Saudi Arabia, we conducted a principal component analysis (PCA) based on these environmental variables in R using the base stats (v4.4.3) package (R Core Team 2022). PCA results were visualized using factoextra (v1.0.7) (Kassambara and Mundt 2026).

Multicollinearity was assessed using variance inflation factors (VIF) (Dormann et al. 2013) to further reduce redundancy among environmental variables. VIF measures the extent to which each predictor can be explained as a linear combination of the remaining variables. High VIF values indicate strong multivariate redundancy among predictors. Variables exceeding the commonly used VIF threshold of 10 were removed using the vifstep function implemented in the R package usdm (Naimi et al. 2014).

Environmental variables were selected for downstream analyses considering both their contribution to the major PCA axes of climatic variation and the reduction of multicollinearity based on VIF.

2.4. Population Structure

Both the global and the Saudi Arabia datasets were analysed using PCA with PLINK (v1.9). Population structure was assessed with ADMIXTURE (v1.3) (Alexander et al. 2009), testing K 1 to 10 genetic clusters, followed by visualization and interpretation in Clumpak (v1.1) (Alexander et al. 2009; Kopelman et al. 2015). The NetViewR (v2.1.0.9000) (Steinig et al. 2016) package was used to validate the similarity between samples and infer population structure. This software utilizes a genetic distance matrix, previously generated with PLINK (using the flags ‐‐distance square 1‐ibs flat‐missing), and applies a mutual k‐nearest neighbours' (kNN) threshold to establish connections between samples within the dataset, revealing genetic structure. The kNN parameter was optimized as the smallest value yielding a stable, fully connected network topology (kNN = 40 for the Saudi Arabian dataset; kNN = 80 for the worldwide dataset). Community structure was then inferred using the fast_greedy community detection algorithm implemented in the NetViewR package.

The R package StAMPP (v1.6.3) (Pembleton et al. 2013) was used to analyse Saudi Arabian mosquito populations, including estimation of global pairwise F _ST with 1000 permutations to assess statistical significance, Nei's genetic distance (Nei 1972), and analysis of molecular variance (AMOVA).

Linkage disequilibrium (LD) was assessed using VCFtools (v0.1.16) with the flags ‐‐geno‐r2 and ‐‐ld‐window‐bp 100,000. Squared correlation coefficients between SNPs (r ²) were plotted against physical distance (kbp), and LD decay was modelled using non‐linear regression in R with the package ggpmisc (v0.6.1) (Aphalo 2026). To estimate LD₅₀ (the distance in kbp at which r ² decays to half of its maximum value) and its standard deviations (SD), 100,000 r ² values were predicted from each fitted model, followed by bootstrapping (1000 replicates) through random sampling of 0.2% of the predicted values in each replicate.

2.5. Introgression Inference

TreeMix (v1.13) (Pickrell and Pritchard 2012) was used to estimate potential gene flow events among populations in the reduced‐worldwide dataset. To prepare the input file, the reduced world dataset was converted into TreeMix format using PLINK (flags: ‐‐freq ‐‐family), followed by processing with a Python2 script publicly available on GitHub (Nelson 2020). TreeMix was then run to estimate up to ten gene flow events (m). TreeMix generates a maximum likelihood (ML) phylogenetic tree based on per‐site allele frequencies of the input populations while simultaneously inferring migratory events. A consensus tree was constructed by running TreeMix with 100 bootstraps. These bootstrap trees were summarized using IQ‐TREE2 (v2.1.2) (Minh et al. 2020), retaining branches present in at least 50% of the sampled trees (majority‐rule consensus; flag ‐minsup 0.50). The Johannesburg population was added manually as an outgroup because it represents the sylvatic Aaf lineage from sub‐Saharan Africa and has been consistently shown to be basal to domestic Aaa populations in previous phylogenetic and population genomic studies (Crawford et al. 2025; Gloria‐Soria et al. 2016; Kotsakiozi, Evans, et al. 2018; Kotsakiozi, Gloria‐Soria, et al. 2018). The optimal number of migration events (m = 0–10) was selected using OptM (v0.1.9) R library (Fitak 2021) based on the Δm statistic. This measures the second‐order change in log‐likelihood across models with increasing migration events (m = 0–10), and the model associated with the highest Δm value was considered the best‐supported.

To statistically evaluate potential admixture events, an f ₃ test (Reich et al. 2009) was conducted using the same input file and TreeMix software. The f ₃ statistic analyses allele frequency correlations between three populations to determine if a candidate population is the result of admixture between two parental populations. Negative values of the f ₃ statistic indicate that allele frequencies in the candidate population are intermediate between the parental populations and provide support for admixture (Patterson et al. 2012). For this analysis, we used multiple iterations with a random set of 2000 SNPs to estimate standard errors. It is worth mentioning that this test does not consider multiple waves of admixture.

The origin of hybrid signals in Saudi Arabian populations was investigated using the DIYABC Random Forest approach (Collin et al. 2021), which consists of two programs used sequentially to infer population demographic history by testing alternative evolutionary scenarios. Two datasets were analysed consecutively to investigate the origin of admixture in southern Saudi Arabian populations. The first dataset included the Lunyo (UG), Madinah, and Jazan Highland populations and was used to evaluate the origin of the admixture signal detected in Jazan Highland. The second dataset expanded this framework by adding Jazan Lowland to investigate the admixture history of the neighbouring lowland population in the context of the Jazan Highland admixture. In both cases, Lunyo (UG) was selected as the best parental Aaf donor based on the f ₃ statistic (Tables S5–S7). The analysis consisted of two main steps. First, a training set was generated by simulating custom evolutionary scenarios for the datasets using the DIYABC Random Forest (v1.1.27) software. A total of 20,000 iterations per scenario were performed to ensure comprehensive coverage of the possible evolutionary pathways. Second, the results from the training step were analysed with the ABC ranger software (v1.16.44) (Collin et al. 2021), which employs a random forest approach to identify the most likely scenario. For this analysis, 100,000 trees were used to enhance the statistical power of the results. Parameter priors are listed in Table S14.

To infer the direction of gene flow, two independent analytical approaches were used: Migrate‐n (v5.0.6) (Beerli 2006; Beerli et al. 2022), and fastsimcoal2 (v2.8.0.0) (Excoffier et al. 2021). Using Migrate‐n, a Bayesian coalescent‐based approach was applied to infer relative effective population sizes, asymmetric gene flow, and population divergence times, all scaled by the mutation rate. To minimize computational time, three independent datasets, each comprising 500 randomly selected SNPs from the Lunyo (UG), Madinah and Jazan Highland populations, were analysed. The migrate‐n input files were generated with the vcfR R package (v1.15.0) (Knaus and Grünwald 2017). For the analysis, we assumed the JC69 mutation model (Jukes and Cantor 1969), with a transition/transversion ratio of 2.0, a constant mutation rate across loci, an exponential distribution of the split time, and different seeds for each dataset. No explicit mutation rate value was specified, as Migrate‐n estimates parameters scaled by the mutation rate and it is therefore implicit in the prior distributions of these scaled parameters (Table S15). Posterior distributions of the parameters were evaluated using Metropolis sampling (Hastings 1970; Metropolis et al. 1953) with a single long chain, a sampling increment of 500, 2000 iterations, and a burn‐in of 200, without replicates. Each run was performed using four Markov Chain Monte Carlo (MCMC) simulations at different algorithmic temperatures (1, 1.5, 3.0, and 100,000) to facilitate exploration of the parameter space. MCMC convergence was assessed by verifying that the effective sample size (ESS) of each modelled parameter exceeded 1000 and exhibited low autocorrelation. Log marginal likelihoods were estimated using Bézier‐approximated thermodynamic integration (Beerli and Palczewski 2010), with values closer to zero indicating better model fit.

Fastsimcoal2 (v2.8.0.0) estimates various demographic parameters using a coalescent simulation framework (Excoffier et al. 2021). This software infers the likelihood of the observed multidimensional Site Frequency Spectrum (multi‐SFS) based on the expected SFS, which is approximated through coalescent simulations under a given model. The folded observed multi‐SFS was derived from a VCF file containing 13,000 sites from the Lunyo (UG), Madinah, and Jazan Highland populations, using the easySFS.py script (Gutenkunst et al. 2009) with a projection ‐‐proj = 10,10. The mutation rate was also set at 4.85 × 10⁻⁹. Each model was analysed with over 200 runs, with 200,000 simulations (flag ‐n) and 100 conditional expectation–maximization (EMC) loops (flag ‐L), each run starting from a different initial point to ensure convergence. Analyses were based on the folded multi‐SFS (flag ‐m ‐u ‐s 0), with parameter estimation performed by maximum likelihood (flag ‐M). Likelihood calculations excluded rare SFS entries below a minimum count threshold (flag ‐C 10), and convergence was improved by resetting parameters after unsuccessful optimization cycles and by progressively narrowing the parameter search range (flags ‐y 4 and ‐z 0.5, respectively). Computational efficiency was ensured through multi‐threading and batch processing (flags ‐c 12 ‐B 12). Parameter priors are shown in Table S15. The Akaike Information Criterion (AIC) was calculated for each model to identify the best‐fitting one, with lower AIC values indicating better model support.

To map the introgressed genomic regions in the Jazan Highland population, we applied two complementary approaches. First, we used Dsuite (v0.5 r58) (Malinsky et al. 2021) to calculate the D‐statistic, which quantifies correlations in allele frequencies among populations (P) within a phylogenetic framework of the form (((P1, P2), P3), Outgroup). The outgroup is used to infer ancestral (A) and derived (B) alleles and to count ABBA and BABA site patterns. Under a model without gene flow, ABBA and BABA patterns are expected to occur at similar frequencies as a result of incomplete lineage sorting (D‐statistic = 0), whereas an excess of either pattern indicates introgression between P3 and one of the sister populations (D‐statistic ≠ 0). In addition, we used Dsuite to perform sliding‐window scans across the genome using non‐overlapping windows of 50 SNPs to estimate the f _dM statistic, which is designed to detect localized signals of introgression (Malinsky et al. 2021). Candidate introgressed regions were defined as genomic windows falling within the upper 5% of the f _dM distribution (f _dM threshold = 0.32).

2.6. Genome Scans

The Saudi Arabian dataset was used for all genome‐wide analyses aimed at identifying candidate loci potentially associated with selection and environmental variation.

Non‐overlapping genomic scans with a 1 Mbp window size and step size were performed to calculate pairwise F _ST using VCFtools (Danecek et al. 2011). Candidate regions were identified using an outlier‐based approach, defined by a conservative empirical threshold calculated as the average plus two standard deviations of the F _ST distribution across four pairwise comparisons between Jazan Highland and Jeddah, Madinah, Makkah, and Najran.

PCAdapt (v4.3.5) (Luu et al. 2017) was used to identify SNPs potentially under local adaptation in Saudi Arabian mosquitoes. This analysis is based solely on allele frequency data, and it does not require predefined populations. We applied the Mahalanobis method (Luu et al. 2017) implemented in PCAdapt. This approach jointly evaluates multiple genetic statistics while accounting for their correlations, allowing the detection of loci that deviate from the genome‐wide background. To minimize false positives, we used the self‐calibrated p‐values provided by PCAdapt.

Finally, we conducted a genome‐environment association analysis in Saudi Arabian Ae. aegypti populations using LFMM2 (v1.1) (Caye et al. 2019). Environmental variables were selected based on their contribution to the major climatic gradients and were used in the association analyses after accounting for multicollinearity. LFMM2 was used to test associations between genetic variants and environmental variation while controlling for hidden population structure through latent factors, using the analytical algorithm to obtain self‐calibrated p‐values.

In both approaches, the number of genetic clusters was set to K = 2, based on ADMIXTURE results, and self‐calibrated p‐values generated by PCAdapt and LFMM2 were converted to q‐values, with significance assessed using a false discovery rate (FDR) threshold of < 0.01.

To disentangle isolation by distance (IBD) and isolation by environment (IBE), distance‐based redundancy analyses (dbRDA) were performed using individual‐based Euclidean genetic distance matrices as response variables, implemented in the R package vegan (v2.7‐5) (Oksanen et al. 2025). Geographic coordinates (latitude and longitude) were used as predictors to test for IBD, whereas environmental effects were assessed by including precipitation of the driest month as an explanatory variable while conditioning on geographic coordinates. Statistical significance was evaluated using 999 permutations. Redundancy analyses (RDA) were conducted to further quantify the relative contributions of spatial structure and environmental variation to genetic differentiation. Spatial structure was modelled using Moran's eigenvector maps (MEMs), computed from geographic coordinates of sampling locations using the adespatial package (v0.3‐2) (Dray et al. 2025). The spatial structure of precipitation was assessed using Pearson's correlation coefficients (r) between precipitation of the driest month and individual MEM variables, with |r| representing the absolute magnitude of the correlation. Genetic variation was summarized using principal coordinates analysis (PCoA) based on the same genetic distance matrix, and the resulting PCoA axes were used as response variables in subsequent analyses (ape v5.8‐1) (Paradis and Schliep 2019). RDA models included precipitation of the driest month and MEM variables as explanatory predictors. Variance partitioning results were interpreted descriptively to assess the relative contributions of spatial and environmental predictors.

2.7. Gene Description

Positive SNPs identified across all genomic scans for selection were extracted using BCFtools (v1.16) (Danecek et al. 2021) and annotated with SnpEff (v5.2) (Cingolani et al. 2012) using the AaegL5 reference genome (Matthews et al. 2018) (Aedes_aegypti_LVP_AGWG in SnpEff). SNP coordinates were verified with the updated SNP chip coordinates (Gómez‐Palacio et al. 2024). Gene information was gathered from multiple databases. Gene annotations, functional descriptions, and genomic coordinates were obtained from NCBI Gene (Murphy et al. 2006) and VectorBase (Alvarez‐Jarreta et al. 2024). Gene Ontology (GO) terms were retrieved from the Gene Ontology repository (Ashburner et al. 2000; The Gene Ontology Consortium 2023) to characterize the functional categories associated with candidate genes. Protein domains were inferred using InterPro (PANTHER) (Blum et al. 2025), and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway information (Kanehisa 2000), which consists of manually curated pathway maps representing molecular interaction and reaction networks, accessed through the R package KEGGREST (v1.46.0) (Tenenbaum and Maintainer 2021).

Genes identified by SnpEff were verified by comparison with the corresponding General Feature Format (GFF) file to confirm gene annotations and genomic coordinates. Positive SNP coordinates were formatted into a BED file and cross‐referenced with the Ae. aegypti GFF file using BEDTools (v2.30.0) (Quinlan and Hall 2010). Genes associated with intergenic SNPs were manually reviewed and extracted. Descriptions for each gene were manually retrieved from the GO file, ENTREZ, and KEGG databases. For representation, only genes with complete descriptions were included in the results.

3. Results

3.1. Climatic Variation Across Saudi Arabian Populations

To characterize environmental variation across Saudi Arabian samples, we performed a PCA on WorldClim bioclimatic variables (Bio1–Bio19) and elevation (Fick and Hijmans 2017). The environmental PCA revealed climatic differences among sampling locations (Figure S1A), with PC1 primarily capturing a gradient driven by temperature‐ and aridity‐related variables. Variables with the highest contribution on the first two PCs included annual mean temperature, mean temperature of the driest quarter, precipitation of the driest month and driest quarter, as well as elevation (Figure S1B). Precipitation of the driest month was selected as the representative environmental variable for downstream analyses, based on its contribution to the major climatic gradients and the reduction of multicollinearity based on VIF analysis (Figure S1C). Figure 1A depicts the precipitation of the driest month (mm) across Saudi Arabia, showing that southern populations (Jazan Highland, Jazan Lowland and Sahil) experienced higher precipitation levels (2.8, 3 and 9 mm, respectively), except for the southern population from Najran (1 mm), whereas all western populations were located in dry areas (0 mm). Following this pattern, populations receiving ≥ 2 mm of precipitation/month were considered humid, whereas those receiving < 2 mm were classified as arid.

3.2. Genetic Diversity and Structure of Saudi Arabian Populations

The Saudi Arabia SNP dataset was used to corroborate the patterns of genetic structure previously observed using microsatellite data (Mashlawi et al. 2024) and conduct a higher resolution analysis of its genetic variation. PCA on the Saudi Arabia dataset (Figure 1B) showed that the three western arid populations (Madinah, Jeddah, and Makkah) clustered tightly together. In contrast, southern populations (Jazan Lowland, Jazan Highland, Sahil, and Najran) were genetically distinct from the western group and more dispersed, indicating higher within‐population variation. Sahil occupied an intermediate position between western arid and southern humid populations, whereas Najran formed a tight, distinct cluster relative to the other southern populations. Analysis of molecular variance (AMOVA) indicated that population explained an important proportion of the observed genetic variation (12.53%), although most variation was explained at the individual level; see Table S2. Pairwise F _ST and Nei genetic distance analyses revealed pronounced genetic differentiation between the southern humid Jazan Highland and all western arid populations (F _ST  = 0.149, Nei distance = 0.093 on average), as well as between Jazan Highland and the southern arid Najran (F _ST  = 0.171, Nei distance = 0.100). In contrast, the lowest differentiation values were observed within the southern humid cluster, specifically between Jazan Highland and Jazan Lowland (F _ST  = 0.068, Nei distance = 0.049) and Sahil (F _ST  = 0.084, Nei distance = 0.058) (Figures 1C and S2A; Table S3).

Admixture analysis at K = 2, the optimal number of ancestral populations (Figure S3A), revealed admixture signals in the southern humid populations, with the second genetic ancestry component predominantly present in the Jazan Highland population, while western arid populations and southern arid Najran were assigned to the first genetic ancestry component and displayed no admixture. At K = 3, a third genetic ancestry component was assigned to Najran (Figure 1D). Consistent patterns were observed in network analyses performed with NetViewR (Figure S4A). Using kNN = 40, two major communities were identified, largely separating southern humid populations from all arid populations. Sahil individuals were distributed between both groups, consistent with the intermediate position in the PCA.

Beyond differences in population clustering, we also observed contrasts in genome‐wide patterns of linkage disequilibrium (LD) among population groups. Jazan Highland exhibited the lowest baseline LD and a rapid LD decay, as reflected by an LD₅₀ distance of 5.0 kbp (SD: 2.13 kbp), compared with 8–9 kbp in western arid populations (SD: 2.73–3.20 kbp) and with the southern arid Najran (10.0 kb, SD: 3.23 kbp) (Figure 1E; Table S4).

Genetic variation was significantly associated with both spatial and environmental factors. When spatial structure and climate were considered jointly, the model accounted for approximately 5.4% of the total genetic variation (F = 1.99, p‐value = 0.001, adjusted R ² = 0.054). dbRDA analyses further supported these patterns: geographic location alone explained roughly 5.1% of genetic differentiation (F = 2.31, p‐value = 0.001), consistent with isolation by distance, while precipitation of the driest month remained significantly associated with genetic differentiation after conditioning on geography (F = 1.33, p‐value = 0.006), explaining an additional 1.5% of the variation. Variance partitioning indicated that most of the explained genetic variation was attributable to spatial structure (2.55%), whereas precipitation of the driest month contributed to a smaller independent fraction (0.41%). Precipitation of the driest month also showed pronounced spatial structure, particularly at local scales (|r|, up to 0.75). Consistent with these results, redundancy analyses combining precipitation and spatial predictors showed that the explained genetic variation was primarily structured along spatial gradients, with precipitation contributing a modest additional signal (Figure 1F).

3.3. Genetic Ancestry of Saudi Arabian Aedes aegypti in a Global Context

To determine how Saudi Arabian populations relate genetically to Aaf and Aaa populations worldwide, we compared their genetic profiles with representative populations from the global dataset (Table 1, Figure 2). PCA analysis revealed that most southern Saudi Arabian mosquitoes were genetically close to African Ae. aegypti formosus (Aaf), with Jazan Highland being the closest one (Figure 2A). Two exceptions were the southern arid Najran population and a few samples from southern humid Sahil, which both formed a separate cluster with western arid mosquitoes and had greater similarity to Asian Ae. aegypti aegypti (Aaa) (Figure 2A).

FIGURE 2.

Evidence of hybrid Aedes aegypti in southern Saudi Arabia. (A) Principal component analysis (PCA) of Saudi Arabian and worldwide Ae. aegypti. The values on each axis indicate the percentage of total variation explained by each principal component; individuals are represented by dots and populations are indicated by colours. Triangles represent samples collected outside Saudi Arabia. (B) Admixture analysis at K = 2, 3 and 5; with K = 5 identified as the optimal number of clusters (see Figure S3B). The height of the colour bar represents the proportion of individual ancestry assigned to each of the genetic groups identified by ADMIXTURE, each shown in a different colour. Individuals belonging to multiple genetic clusters are depicted as bars with varying proportions of these colours. Thick black lines within the plots delineate population boundaries. Countries and regions are labelled above the bar plots: Senegal (SN), Cameroon (CM), Gabon (GA), Uganda (UG), Sudan (SD), South Africa (ZA), Kenya (KE), Pakistan (PK), Philippines (PH), Thailand (TH), Vietnam (VN), Turkey (TR), and Georgia (GE). BS correspond to Black Sea region.

Admixture analysis at K = 2 revealed that the Aaa signal was predominant in the Saudi Arabian samples. However, in the three southern humid populations of Jazan Highland, Jazan Lowland and Sahil, an Aaf signal was also detected, with the highest proportion of Aaf present in Jazan Highland (average Aaf ancestry: Jazan Highland = 0.57, Jazan Lowland = 0.20, Sahil = 0.15) (Figure 2B). In Sahil, this signal was present in most, but not all, individuals. At K = 3 Asia splits in two, with Saudi Arabian populations separating from the rest of Asia and Punjab (Pakistan) displaying both Asian ancestries (Figure 2B). At K = 5, the optimal number of groups for this dataset (Figure S3B), the Black Sea region (Turkey and Georgia) forms its own group and a new ancestry component is observed within Africa, clustering Johannesburg and Nairobi (ZA and KE, respectively) apart from Franceville and Lope Village (GA). The southern humid populations of Jazan Highland, Jazan Lowland, and Sahil show evidence of four genetic ancestries, pointing to a complex history of hybridization and/or older ancestral contributions.

To minimize the influence from the complex genetic background of populations from Senegal, Cameroon, and Gabon on clustering and gene flow inference, these populations were excluded from subsequent analyses, using a reduced worldwide dataset. A network constructed using kNN = 80 on the reduced worldwide dataset was consistent with patterns observed in the PCA and admixture analyses, where the three southern humid populations clustered closer to African Aaf, and Jazan Highland samples fell within the Aaf group (Figure S4B).

Together, these results confirm the Aaf/Aaa hybrid nature of Ae. aegypti populations from southern humid Saudi Arabia.

3.4. Introgression of Aedes aegypti formosus Into Southern Saudi Arabia Aedes aegypti aegypti

To investigate the origin of the hybrid ancestry in southern humid Saudi Arabia mosquitoes, we conducted an f ₃ test (Reich et al. 2009) to examine potential admixture events between Aaf and Aaa populations, using the reduced worldwide dataset.

Among all tested donor pair combinations, those that included Lunyo (UG) as the Aaf donor and Aaa donors from either Saudi Arabia or another Asian region (Thailand, Vietnam, Philippines, Pakistan) displayed the highest support as parental populations of Jazan Highland (Table S5). In contrast, only western arid (Makkah, Madinah, or Jeddah) and the southern arid Najran populations from Saudi Arabia were supported as Aaa donors for Jazan Lowland and Sahil populations (Figure 3A). Furthermore, Jazan Highland was strongly supported as the parental Aaf donor for Jazan Lowland and Sahil (Tables S6 and S7).

FIGURE 3.

Origin of southern Saudi Arabia Aedes aegypti explained by introgression events from Africa. (A) Results from f ₃ analysis using Lunyo (UG) as Aaf donor 1 and different Aaa donors 2 for Jazan Highland, Jazan Lowland, and Sahil. Grey dots correspond to non‐significant f ₃ values. Yellow dots indicate good support for the admixture between the two donors (Z‐score > −3.0). “Aaf‐Aaf” or “Aaa‐Aaa” indicates that two Aaf or two Aaa populations were used as donors, with the average of f ₃ values plotted. The grey vertical dashed line highlights the zero value for f ₃. Results for additional Aaf donors are summarized in Tables S5–S7 for Jazan Highland, Jazan Lowland, and Sahil, respectively. (B) Consensus ML‐phylogenetic tree based on allele frequencies, after 100 bootstrap samplings, with 3 events of gene flow. Coloured arrows represent migration estimates derived from TreeMix. Numbers represent the frequency at which each group appeared in the subsampled trees. (C) Demographic scenarios tested to explain the hybridization pattern of the Jazan Highlands. Model 1: A recent introgression of Aaf from Lunyo (UG) into Saudi Arabian populations. Model 2: An ancestral Aaf population that diverged from Lunyo (UG) present in the Jazan Highlands and recently invaded by local Aaa. The comparison was performed using two programs, migrate‐n and fastsimcoal2. Likelihoods generated for each model by each approach are shown below the corresponding model diagrams.

Saudi Arabian Ae. aegypti appeared polyphyletic in the maximum likelihood (ML) population phylogeny based on allele frequencies from the reduced worldwide dataset (Figure 3B), with the admixed Saudi populations distributed across different branches, while all western arid and the southern arid Najran populations clustered together and with the rest of the Asian samples. By incorporating migration events into the TreeMix phylogeny, we found evidence for two discrete pulses of gene flow between African mosquitoes and Jazan Highland and Sahil, and one extra pulse between Sahil and Jazan Lowland (Figure 3B). Three pulses were identified as the optimal number of pulses explaining the admixture pattern observed, according to the ∆m criteria as implemented by OptM (Figure S5) (Fitak 2021).

Next, we conducted DIYABC analyses to determine the demographic scenario that best explains the origin of the hybrid signal in the southern humid populations of Jazan Highland and Jazan Lowland (Figure S6). Three possible scenarios were tested for Jazan Highland, considering the two putative parental Ae. aegypti identified by the f ₃ test (Figure S6A). The best supported scenario was Scenario 2 (posterior probability = 0.713) and involved admixture between Aaf (represented by Lunyo (UG), as the best Aaf proxy according to f ₃ tests; Tables S5–S7) and Aaa from the western arid region of Saudi Arabia (such as Madinah populations). Next, to investigate the origin of Jazan Lowland, two sets of models were considered. The first set included one admixture scenario (Scenario 5) and four scenarios without admixture (Scenario 1 to 4) (Figure S6B). The second set included scenarios with a variety of admixture events (Figure S6C). In both cases, the best supported scenario involved an admixture event between the southern humid Jazan Highland and the western arid Madinah (Set 1: Scenario 5, posterior probability = 0.791; Set 2: Scenario 1, posterior probability = 0.851).

To explicitly test the direction of gene flow between African Aaf and a western arid Saudi Arabian Aaa, we performed two complementary demographic analyses using migrate‐n and fastsimcoal2, comparing two alternative scenarios. Model 1 represents a recent introgression of Aaf (Lunyo, UG) into a western arid Saudi Arabian Aaa population. Model 2 represents an ancestral divergence of Aaf leading to Jazan Highland, followed by a more recent introgression from western arid Saudi Arabian Aaa (Figure 3C). In migrate‐n, the estimated log marginal likelihoods across replicate runs were −8066.05, −8952.92 and −9256.12 for model 1, and −8077.31, −8992.074 and −9273.798 for model 2. Similarly, fastsimcoal2 yielded lower AIC values for model 1 (107,094.88) than for model 2 (109,288.41). Both methods consistently supported model 1, indicating that recent introgression from Aaf into Saudi Arabian Aaa lineage is the most plausible explanation for the Aaf genetic signal detected in the southern Jazan Highland population.

3.5. Genomic Variants Associated With Local Selection and Environmental Factors

To identify signatures of selection in the admixed southern humid Saudi Arabian populations, we conducted three complementary genomic scans: a pairwise F _ST scan to detect differentiated genomic regions, analysis of local adaptation using PCAdapt (Luu et al. 2017), and environmental association tests with LFMM2 (Caye et al. 2019). We synthesized results from the three complementary scans by retaining SNPs that were identified as candidates in at least one approach (pairwise F _ST outliers, PCAdapt, or LFMM2). This integrative strategy resulted in 78 candidate variants associated with 120 genes (Figure 4A, Tables S8 and S10). Detailed results and further analyses are provided in the Supporting Information (Figures S7–S13). Neither identified variants nor genes were shared across all three selection scans (including genes with incomplete descriptions), although few genes were found by both PCAdapt and LFMM2 when elevation or mean temperature of the driest quarter were considered (Table 2, Figure 4A).

FIGURE 4.

Introgressed genomic regions potentially involved in environmental adaptation of Aedes aegypti in Saudi Arabia. (A) Venn diagrams showing the overlap of candidate variants (left), genes (middle), and KEGG pathways (right) identified using three complementary approaches: F _ST scans (red), PCAdapt (brown), and LFMM2 (blue). (B) Network representation linking candidate genes detected by F _ST , PCAdapt, and LFMM2 (coloured nodes) to associated KEGG metabolic pathways (grey nodes). Pathways highlighted in dark grey represent the most relevant functional categories identified in this study. The gene AAEL024147 is not assigned to any metabolic pathway in the KEGG database. Schematic representation of the ABBA‐BABA test topology used to evaluate introgression among populations, with Johannesburg (ZA) as the outgroup, Lunyo (UG) as the gene flow donor (P3), Jazan Highland as the hybrid population (P2), and Madinah as the non‐admixed population (P1). (D) Genome‐wide distribution of f _dM statistics across chromosomes 1–3. Highlighted regions correspond to the top 5% of f _dM values. (E) SNP genotypes across populations for candidate genes identified in this study. Rows correspond to SNPs, columns represent individuals grouped by population, and colours indicate genotype states (0/0, 0/1, 1/1, and NA = missing data). Asterisk (*) denote SNPs detected by the applied analyses.

TABLE 2.

Summary of relevant genes associated to metabolic pathways according to the KEGG database.

	Chromosome	SNP ID a	SNP annotation a	Gene ID	Gene description	Domain protein (PANTHER)
Associated with a metabolic pathway	NC_035107.1	AX‐93242415	Downstream_gene_variant/intergenic_region	AAEL024147	Elongation of very long chain fatty acids protein	Fatty acid acyl transferase‐related
	NC_035108.1	AX‐93238678	3′‐UTR variant	AAEL006987	NA	Phosphatidate phosphatase‐related
	NC_035108.1	AX‐93238682	Synonymous	AAEL006990	NA	Phosphatidate phosphatase‐related
	NC_035108.1	NA	NA	AAEL007018	UDP‐glucose 4‐epimerase	UDP‐glucose 4‐epimerase
	NC_035108.1	NA	NA	AAEL007023	Peroxisomal Multifunctional Enzyme Type 2	Dehydrogenases, short chain
	NC_035108.1	AX‐93261522	Intergenic	AAEL009951	Dimeric dihydrodiol dehydrogenase	Oxidoreductases 1
	NC_035109.1	NA	NA	AAEL009838	Glycogen debranching enzyme	Glycogen debranching enzyme
	NC_035109.1	AX‐93214097	Downstream gene variant	AAEL000006	Phosphoenolpyruvate carboxykinase	Phosphoenolpyruvate carboxykinase, GTP‐utilizing
Associated with Elevation (shared among PCAdapt and LFMM2)	NC_035107.1	AX‐93254850	Missense variant	AAEL021955	Syntaxin 16	Vesicle‐mediated transport
	NC_035107.1	AX‐93254850	3′‐UTR variant	AAEL022343	lncRNA	NA
	NC_035107.1	AX‐93254850	Downstream gene variant	AAEL025137	lncRNA	NA
	NC_035107.1	AX‐93254850	Downstream gene variant	AAEL021596	lncRNA	NA
	NC_035107.1	AX‐93224965	Downstream gene variant	AAEL003002	NA	NA
	NC_035107.1	AX‐93224965	Downstream gene variant	AAEL016274	tRNA	NA
	NC_035108.1	AX‐93245021	Intronic	AAEL019489	Apterous, transcription factor	Regulation of DNA‐templated transcription
	NC_035109.1	AX‐93241802	Intronic	AAEL007794	Defective proboscis extension response	NA
	NC_035109.1	AX‐93231610	Intronic	AAEL019832	Heavyweight, phosphotyrosine residue binding	NA

^a NA indicates that no SNPs associated with these genes are present on the SNP‐chip platform; however, the genes are located within candidate regions identified by pairwise F _ST scans.

Given the lack of convergence among the three methods, we used the KEGG pathway database to assess the potential functional roles of the candidate genes. Eight of the genes under putative selection identified by F _ST, PCAdapt, and LFMM2 were highly represented in metabolic pathways and were therefore retained for subsequent interpretation. These pathways involved lipid and sugar metabolism and oxidative phosphorylation (Tables 2, S9, S11, and S13; Figure 4B).

Most of the variants in genes related to lipid metabolism exhibited contrasting genotype distributions between the southern humid Jazan Highland and western arid Saudi samples (Figure 4E). Notably, the allele frequencies of AAEL024147, AAEL006987, AAEL006990, and AAEL007023 in Jazan Highland closely resemble those found in Lunyo (UG). Moreover, three of these genes (AAEL006987, AAEL006990, and AAEL007023) are located within the same 1 Mbp genomic region on Chr 2 (242–243 Mbp), suggesting physical linkage. A similar asymmetry in allele frequencies was observed for a small number of LFMM2 candidate genes, restricted to variants associated with precipitation of the driest month: AAEL006056 in Chr 3: 74–75 Mbp, respectively (Figure S14).

Visual inspection of genotype distributions for SNPs identified by PCAdapt revealed differences among populations, with genotypes carrying the alternative allele more frequent in the high‐altitude populations of Jazan Highland (mean altitude 554 m), Madinah (mean altitude 624 m), and Najran (mean altitude 1311 m) than in the arid populations (Figures S1D and S14). Two of these variants were located within AAEL009951 (dimeric dihydrodiol dehydrogenase, Chr 2: 379,546,778, SNP AX‐93261522), which is involved in xenobiotic degradation, and AAEL000006 (phosphoenolpyruvate carboxykinase: Chr 3: 102,809,060, SNP AX‐93214097), a key enzyme in gluconeogenesis (Figure 4B).

To evaluate whether the eight candidate genes under selection fell within regions introgressed from Aaf (Lunyo, UG) into the southern humid Jazan Highland population, we conducted a genome‐wide ABBA‐BABA test and window‐based f _dM analysis. The genome‐wide ABBA‐BABA test revealed an excess of allele sharing between Jazan Highland and Lunyo relative to western arid Madinah samples (D = 0.023, Z‐score = 4.20), consistent with African gene flow contributing to the Jazan Highland genome (Figure 4C), as described above. To localize this signal, we performed a window‐based f _dM scan. This analysis revealed a highly heterogeneous pattern, with only nine regions showing pronounced peaks of f _dM (Figure 4D), indicating that the African affinity of Jazan Highland is driven by discrete genomic segments rather than a uniform genome‐wide signal. Of the eight candidate genes described in Figure 4B, AAEL024147 on Chr 1 (282–283 Mbp) was located within an introgressed region (Figure 4D). Four additional candidate genes from the 120 set derived from the selection scans (AAEL022121, AAEL005490, AAEL005496, and AAEL019832) were also located within an introgressed region in the Jazan Highland population, but these genes have not been fully characterized in the databases available.

4. Discussion

Colonization of new ecological niches during range expansion or rapid environmental change can expose populations to novel selective pressures. Secondary contact between genetically differentiated populations can provide a source of genetic variation through introgression, some of which may facilitate adaptation to new or changing environmental conditions (Hedrick 2013; Huerta‐Sánchez et al. 2014; Taylor and Larson 2019; Whitney et al. 2010). The ecological contrast between the arid cities of western Saudi Arabia and the increasingly humid Jazan Highland region provides a natural setting to investigate genetic adaptation mediated by introgression. In this context, we examined the Aaa/Aaf admixture signal in southern Saudi Arabia and its potential role in environmental adaptation. Our analyses support introgression from an eastern African population, genetically similar to Lunyo (UG), into the southern Jazan Highland population. Genome‐wide scans identified loci that may contribute to local adaptation in a geographic region that differs markedly in precipitation and elevation, one of these loci located within introgressed regions, suggesting a potential for adaptive introgression. The functions of genes identified within candidate regions from the F _ST scan align with pressures typical of arid environments, where persistence depends on traits such as water‐efficient metabolism, enhanced water retention, and drought dormancy (Holmes and Benoit 2019). Such traits are known to promote ecological diversification and survival in challenging habitats (see review Chown et al. (2011)).

4.1. Genetic Differentiation and Hybridization Patterns

We find significant genetic differentiation among Saudi Arabian populations and evidence of introgression between the two subspecies of Ae. aegypti in southern humid locations, consistent with patterns previously identified using microsatellite markers (Mashlawi et al. 2024). Using genome‐wide SNP data, this study further explored the nature of the admixture signature and established the directionality of the gene flow.

Jazan Highland, Jazan Lowland, and Sahil populations exhibit varying degrees of Aaf ancestry, with evidence of a complex history of admixture and divergence suggested by the ancestry results at higher clustering levels. Our results provide evidence of directional gene flow from African Aaf populations into established western arid Aaa populations in Saudi Arabia, in line with earlier findings that report secondary contact zones between Aaf and Aaa along the coast of Africa (Crawford et al. 2025; Elnour et al. 2022; Gloria‐Soria et al. 2016; Mashlawi et al. 2024; Rose et al. 2020). Both the phylogenetic analysis and gene flow direction models support two distinct sequential gene flow events into southern humid Saudi Arabia: one from East African populations into Jazan Highland and subsequently another from Jazan Highland into Jazan Lowland (Figures 3 and S7), suggesting that within Saudi Arabia, Jazan Highland may act as a reservoir of genetic variation. We note that the varying levels of Aaf ancestry among individuals in Sahil (ranging from 0.018 to 0.230; Figure 2B) limited our ability to fully resolve the underlying evolutionary patterns; therefore, Sahil was excluded from our models. This variability in African ancestry within Sahil (Figure 3B) may be caused by the presence of genetic substructure within this population, likely a consequence of the large geographic region it covers, but the origin of this signature remains to be elucidated.

Genome‐wide patterns of linkage disequilibrium (LD) varied among Saudi Arabia populations and provide insights into their demographic history. Consistent with Aaa evolutionary history of bottlenecks leading to its arrival to Asia from Africa (Crawford et al. 2025; Gloria‐Soria et al. 2016), high baseline LD and slow LD decay were observed in the arid populations of Saudi Arabia (Jeddah, Madinah, Makkah, Najran). The representative parental Aaf population of Lunyo (UG) was characterized by high LD and rapid LD decay, as expected from a large and diverse ancestral population (Table S4). Meanwhile, a range of lower baseline LD levels and rapid LD decay was observed across the humid Saudi Arabian populations of Jazan Highland, Jazan Lowland and Sahil (Figure 1E), in line with secondary contact and admixture between subspecies breaking down parental haplotypes.

While Mashlawi et al. (2024) hypothesized that road connectivity and human activities facilitate passive dispersal and admixture, our multivariate analyses (RDA and dbRDA) revealed a significant spatial signal consistent with isolation‐by‐distance. This pattern indicates that genetic differentiation and admixture are primarily shaped by localized secondary contact among neighbouring populations rather than widespread mixing across the landscape. Such spatially structured gene flow is consistent with short‐range dispersal. Together, these results suggest that restricted dispersal, habitat heterogeneity, and region‐specific contact zones play a central role in structuring genetic variation across the study area.

4.2. Genomic Regions Under Selection and Role of Environmental Factors

Throughout its evolutionary history, Ae. aegypti adapted to survive in areas where water availability heavily depends on human activity, such as the arid regions in Africa (Rose et al. 2023). Approximately 5000 years ago, these changes facilitated Ae. aegypti invasion of human habitats from the ancestral forest and increased its preference for human hosts, behaviours that correlate with dry season intensity and have an underlying genetic component (Rose et al. 2020). Levels of precipitation during the driest month differ significantly between western and southern Saudi Arabia regions, especially across collection sites (on average, 4.86 mm and ~0.0 mm for southern humid and western arid populations, respectively; 1.0 mm in southern arid Najran, Figure 1A). Ae. aegypti inhabiting regions where water is limited, such as the western arid cities of Jeddah, Madinah, and Makkah, would experience strong selective pressure on genes acting to prevent water loss. These genes are therefore expected to be genetically differentiated from those in populations inhabiting more humid areas, such as Jazan Highland.

Our genomic scans identified several candidate genes under selection among Ae. aegypti in arid and humid regions. Notably, most of these genes were linked to different aspects of lipid metabolic pathways, including genes encoding phospholipid phosphatases (AAEL006990 and AAEL006987) and a peroxisomal multifunctional enzyme (AAEL007023). This pattern is consistent with previous evidence showing that lipid degradation is essential to produce protective molecules and the energy required for desiccation survival in Ae. aegypti eggs (Prasad et al. 2023). We also identified an elongase of very long‐chain fatty acids (AAEL024147), a gene involved in cuticular hydrocarbon biosynthesis, which has been associated with adaptation in Ae. aegypti populations from Sri Lanka (Ramasamy et al. 2021) and other genes associated with fatty acid metabolism (Balcazar et al. 2026). The repeated implication of cuticle‐associated processes and lipid metabolism points to these pathways as a recurring axis of environmental adaptation, likely related to the regulation of water balance under both saline and arid stress. Elongases of very long‐chain fatty acids have been associated with elevated cuticular hydrocarbon production in desert‐adapted Drosophila species (Wang et al. 2023), limiting water loss, a process conserved across insect taxa (Arcaz et al. 2016).

Several candidate genes show similar allele frequencies in the humid southern Jazan Highland and Aaf from Lunyo (UG); however, only AAEL024147 lies within an inferred introgressed region from Aaf. While the broader genomic background of Jazan Highland populations appears associated with adaptation to arid environments, the introgressed region containing AAEL024147 may reflect selection to more humid conditions, consistent with the climate of Lunyo (Figure S15; 69 mm of precipitation during the driest month). Together with the low levels of linkage disequilibrium observed in this region, these patterns suggest that introgression from Aaf may have introduced genetic variation that facilitated adaptation to the more humid, forested, and rural habitats of southern Jazan Highland.

These observations indicate that lipid metabolism broadly, and AAEL024147 specifically, may be under selection across humidity gradients. However, the functional roles of these candidate genes in desiccation tolerance, arid‐region colonization, and adaptation to humid conditions remain to be experimentally validated.

Additionally, we identified genes associated with elevation that were also detected in the selection scan by PCAdapt, and are related to glucose metabolism, oxidative processes, and other cellular functions. Most of these genes could not be placed within a molecular pathway (Table 2), but their allele frequencies differ between populations at higher (Jazan Highland, Madinah, Najran) and lower elevation (Jazan Lowland, Jeddah, Sahil and Makkah). Possible selecting pressures acting on these genes may include UV radiation exposure, oxygen levels, or human activity (Figures 4C and S1D). Interestingly, two genes identified by PCAdapt—AAEL000006 and AAEL009951—showed differences in allele frequencies between populations at different elevations and are known for their increased activity following a blood meal (Hou et al. 2015; Sanders et al. 2003). These associations may reflect a physiological response linked to metabolic demand or detoxification processes that vary with environmental conditions, highlighting the need for functional studies to assess their relevance in elevational adaptation.

4.3. Implications for Evolutionary Dynamics

Overall, these findings indicate that the genetic structure of Ae. aegypti populations in Saudi Arabia reflects the combined effects of environmental pressures—particularly climatic gradients in precipitation and elevation‐, gene flow from African Aaf populations, and population movements within Saudi Arabia. While environmental selection appears to be the dominant signal, introgression may act as a complementary process by providing genetic variation upon which selection can operate, particularly in populations occupying more humid regions.

Although introgression has been widely implicated in local adaptation across diverse taxa (Fonseca et al. 2009; Huerta‐Sánchez et al. 2014; The Heliconius Genome Consortium 2012; Whitney et al. 2010), its role in Ae. aegypti remains relatively understudied. Studies in Aaf populations have identified introgressed regions containing odorant‐binding protein (OBP) genes that may be linked to behaviourally relevant traits, although direct functional evidence remains limited (Redmond et al. 2020).

By identifying genomic regions associated with environmental variation, our findings emphasize the role of environmental selection in shaping the ecological responses of Ae. aegypti populations. Introgression may serve as a supplementary source of genetic variation, with potential implications for the species' continued geographic expansion.

5. Conclusion and Future Directions

Our study demonstrates significant genetic differentiation between western arid and southern humid Ae. aegypti populations in Saudi Arabia. Jazan Highland populations exhibit a significant high proportion of Aaf ancestry derived from secondary contact between the subspecies and subsequent introgression, serving as a potential reservoir of genetic variation. Although introgression from African populations may have facilitated the spread of Ae. aegypti into humid highland regions of Saudi Arabia, future studies should focus on elucidating the functional roles of candidate genes and on assessing how genetic variation and introgression shape physiological and behavioural traits, such as desiccation tolerance, feeding, and breeding preferences. These insights will be crucial for understanding geographic variation in vector competence, disease transmission dynamics, and insecticide resistance, ultimately improving the effectiveness of vector control strategies.

Author Contributions

Abadi M. Mashlawi: managed and performed field collections, performed research, wrote original draft, reviewed and edited manuscript draft. Dario Balcazar: performed research, analysis, visualization, wrote original draft, reviewed and edited manuscript draft. Sara A. Abuelmaali: data discussion, reviewed and edited manuscript draft. Mohammed Fahad Albeshr: data analysis, reviewed and edited manuscript draft. Mohamed F. Sallam: collection of environmental data, reviewed and edited manuscript draft. Hesham M. Al‐Mekhlafi: study design, data analysis, reviewed and edited manuscript draft. Catherine Walton: project manager, analysis discussion, reviewed and edited manuscript draft. Andrea Gloria‐Soria: research design, genetic data collection and curation, analysis supervision, reviewed and edited manuscript draft.

Funding

This work was supported by National Institutes of Health (Grant RO1 AI155562) and Jazan University (Grant RG24‐S054).

Disclosure

Benefit‐sharing statement: All the research complies with applicable laws on sampling from natural populations (including the ARRIVE guidelines). No vertebrates were used in this study.

Sample metadata: Metadata associated with the samples included in this study can be found in Table 1.

Ethics Statement

Our work follows all ethical and research integrity standards.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Bioclimatic and elevation differences for Saudi Arabian samples. (A) PCA of environmental conditions based on bioclimatic variables. Points represent sampling locations. (B) Loadings of bioclimatic variables on the first two principal components. Arrows indicate variable loadings, while colours represent the relative contribution of each variable to the PCA axes. (C) Variance inflation factor (VIF) values for bioclimatic variables retained after stepwise filtering (threshold = 10). Bioclimatic variables correspond to WorldClim definitions: bio1, annual mean temperature; bio2, mean diurnal range; bio3, isothermality; bio4, temperature seasonality; bio5, maximum temperature of the warmest month; bio6, minimum temperature of the coldest month; bio7, annual temperature range; bio8, mean temperature of the wettest quarter; bio9, mean temperature of the driest quarter; bio10, mean temperature of the warmest quarter; bio11, mean temperature of the coldest quarter; bio12, annual precipitation; bio13, precipitation of the wettest month; bio14, precipitation of the driest month; bio15, precipitation seasonality (coefficient of variation); bio16, precipitation of the wettest quarter; bio17, precipitation of the driest quarter; bio18, precipitation of the warmest quarter; bio19, precipitation of the coldest quarter. (D) Map of elevation in Saudi Arabia (meters above the sea). Red colours indicate high elevation above the sea, while blue colours represent lower levels.

Figure S2: Nei distance matrix comparing all Saudi Arabian populations. Heatmap illustrating the pairwise Nei distance values between all the Saudi Arabian populations included in the study. The Nei distance is represented with colour gradients, where darker red shades indicate greater genetic distance and darker blue shades suggest closer genetic relationships.

Figure S3: Cross‐validation error along different K values for admixture tests. Analysis of the results of cross‐validation error, used to select the optimal K value for admixture analysis. (A) shows the analysis performed with Saudi Arabian populations only, while (B) presents the results for the world context dataset. The lowest point in the curve indicates the most suitable K value.

Figure S4: Network visualization showing the connection of Saudi Arabian samples in two different datasets. (A) Illustration of the connections of Saudi Arabian mosquitoes at a k‐nearest neighbour (kNN) threshold: kNN = 40, while (B) displays the grouping analysis of Saudi Arabian samples alongside mosquito populations from different regions of the world, using a kNN = 80. Light grey areas indicate the communities detected by the “fast greedy” algorithm.

Figure S5: Three migration edges represent the optimal number of migration events in hybrid Saudi Arabian populations. Line plot showing the optimal number of migration edges, as determined by OptM based on TreeMix output from 10 independent runs, each testing 1 to 10 migration events and repeated 10 times.

Figure S6: Evidence of two different admixture events into Saudi Arabian mosquitoes. (A) DIYABC analysis under different scenarios for the Lunyo, Jazan Highland, and Madinah dataset. (B and C) Same populations as before but including Jazan. In (B), scenarios without admixture events and with just one admixture event were tested, while in (C) different scenarios with varying numbers of admixture events were included. The diagrams only represent the analysed scenarios, and the length of the branches and the timing of bifurcations do not have any specific meaning.

Figure S7: Genomic regions differentiated between Jazan Highland from non‐hybrid Saudi Arabian populations. (A) Results from the non‐overlapping F _ST genomic scan using a window size 1 Mb, comparing Jazan Highland with Jeddah, Madinah, Makkah, and Najran, respectively. Grey dots correspond to genomic windows with non‐significant F _ST values (the dashed black line denotes the F _ST threshold: 0.279); yellow, green, and blue dots represent significant F _ST values present in one, two, or three pairwise comparisons, respectively, while red dots indicate regions that were significant in all comparisons. (B) Sankey diagram showing the relationship between genes, GO terms, and KEGG pathways.

Figure S8: Analysis of SNPs and genes identified with F _ST genomic scanning. (A) After F _ST genomic scanning (window size 1 Mb), SNP annotations were analysed to determine the most represented types of variation. (B–D) correspond to the analysis of different types of GO terms according to their ontology (biological process, cellular component, and molecular function, respectively). (E) Analysis of the pathways in which the genes are involved.

Figure S9: Significance of SNPs through PCAdapt. PCAdapt p‐values were transformed into q‐values to control for false discovery rate (FDR < 0.01). Brown dots represent SNPs with q‐values higher than the threshold (−log₁₀ (q) = 2). Labels in bold italics correspond to SNPs also identified by LFMM2 using Elevation or Mean Temperature of Driest Quarter as the environmental predictor. In purple are SNPs associated with genes involved in a metabolic pathway.

Figure S10: Annotation of candidate SNPs and associated genes identified by PCAdapt. (A) Proportion of SNPs in each annotation category. (B) Counts of associated genes annotated with Gene Ontology (GO) terms, grouped into biological process, cellular component, and molecular function. (C) Counts of associated genes assigned to KEGG pathways.

Figure S11: Annotation of candidate SNPs and associated genes identified by LFMM2. (A) Proportion of SNPs in each annotation category. (B) Counts of associated genes annotated with Gene Ontology (GO) terms, grouped into biological process, cellular component, and molecular function. (C) Counts of associated genes assigned to KEGG pathways.

Figure S12: SNPs detected with PCAdapt or LFMM2. Sankey diagrams showing the relationship between genes, GO terms, and KEGG pathways corresponding to SNPs highlighted with (A) PCAdapt or (B) LFMM2.

Figure S13: SNP‐environment associations identified by LFMM2. An LFMM2 scan was performed to identify SNPs associated with elevation, precipitation of the driest month, precipitation of the driest quarter, mean temperature of the driest quarter, temperature annual range, and temperature seasonality. Each row shows an environmental predictor and each column a chromosome. The p‐values were transformed into q‐values to control for false discovery rate (FDR < 0.01). Blue dots represent significant SNPs with q‐values higher than the threshold (−log₁₀ (q = 2). The threshold is indicated by the dashed lines. Labels in bold italics correspond to SNPs also identified by PCAdapt. In purple is the SNP associated with a gene involved in a metabolic pathway.

Figure S14: Genetic profiles of SNPs associated with multiple environmental variables. SNP genotypes mapped to different genes. Positive SNPs (*) were identified using LFMM2 when different environmental variables were tested as environmental parameters.

Figure S15: Map of precipitation levels during the driest month in Uganda. Data was retrieved from the BioClimate database (BIO_14) and plotted using R. Red colours indicate high precipitation levels during this season, while blue colours represent lower levels. Precipitation levels equal to or above 12 mm were manually saturated with red to facilitate comparison with the precipitation levels in Saudi Arabia.

Table S1: Software used for this work.

Table S2: AMOVA test for Saudi Arabian Ae. aegypti mosquitoes.

Table S3: F _ST and Nei distance values. Values under the diagonal correspond to F _ST values, while Nei distance are above the diagonal.

Table S4: Linkage disequilibrium measures for Saudi Arabian populations.

Table S5: f ₃ values for Jazan Highland being the hybrid population.

Table S6: f ₃ values for Jazan Lowland being the hybrid population.

Table S7: f ₃ values for Sahil being the hybrid population.

Table S8: Descriptions of the positives SNPs detected with F _ST genomic scan.

Table S9: Gen description associated to those F _ST positive SNPs.

Table S10: Descriptions of the positives SNPs detected with PCAdatp or LFMM2 scans.

Table S11: Gen description associated with those PCAdapt positive SNPs.

Table S12: Environmental variables used for our LFMM2 scans.

Table S13: Gen description associated with those LFMM2 positive SNPs.

Table S14: Parameter prior distributions for DIYABC analysis.

Table S15: Parameter priors and values for n‐migrate (top table) and fastsimcoal2 (bottom table) analysis.

Data Availability Statement

The analysis software and databases used for this work are listed in Table S1. The scripts for this study can be found at: https://github.com/dbalcazar‐notes/Saudi_Arabian_aedes. Individual genotype data are available at Dryad: 10.5061/dryad.fj6q57469.