Spatiotemporal variation in abundance and genetic structure in an urban–rural landscape: Aedes albopictus (Skuse, 1894) (Diptera: Culicidae) in Wake County, NC

Abstract

Since its invasion of the United States in the 1980s, Aedes albopictus (Skuse, 1894) has become a major pest and a significant public health threat in the Southeastern United States. Despite its importance, we know little about its population genetics at fine spatial scales that correspond to the level of management units. To remedy this lack of information, we analyzed Ae. albopictus spatial variation in mosquito abundance and genetic structure in an urban–rural landscape over 2 years (2016 and 2018) in Wake County, NC, United States. We used a reduced representation sequencing method to generate between 1,100 and 30,000 single-nucleotide polymorphisms for population genetic analyses. We found spatial variation in both the abundance and genetic diversity, and significant differences in genetic divergence among sites that varied between the 2 years. The year-to-year variation in the population genetic patterns at the within-county scale suggests a dynamic system that requires extensive geographic, temporal, and genomic sampling to resolve.

Graphical abstract

Graphical Abstract.

Introduction

Aedes albopictus is a cosmopolitan mosquito that is invasive outside of its range in East Asia, spanning a large latitudinal range under current climate conditions, from the Equator to 50°. This mosquito prefers suburban environments, where it threatens human, domestic animal, and wildlife health as a vector of viral and filarial pathogens (Hawley et al. 1987, Barker et al. 2003, Gratz 2004, Spence Beaulieu et al. 2019). This is of concern as changes in global temperatures and land use may allow Ae. albopictus to both invade new temperate regions, especially in warmer urban landscapes, and increase the transmission of human and animal pathogens (Gratz 2004). Given its historic arrival in the Southeastern United States (in 1984, Hawley et al. 1987) and more recent expansion into the Northeastern United States (Hahn et al. 2017) and California (Metzger et al. 2017), we need a clear understanding of the behavior and movement of the species between urban and rural landscapes or within urban settings, regions with the greatest human population density.

Landscape, the spatial pattern of suitable habitat patches and unsuitable matrix, and human population density help determine the distribution, abundance, age structure, and genetic structure and diversity of many species. These demographic characteristics of species can be altered by land-use change, including urbanization, the conversion of land into the human-built environment characterized by increased impervious surface cover and greater human densities. Thus, urban landscapes are mosaics of habitat patches, such as parks, forests, and the built environment, with impacts on abiotic conditions and biotic interactions (Vitousek et al. 1997, Irwin and Bockstael 2007, Baudouin et al. 2018). From an ecological perspective, urbanization describes an increase in human population density and intensity of its influence on the environment (Herrero-Jáuregui et al. 2019). For native species, the degree of urbanization is negatively correlated with population abundance and genetic diversity and positively correlated with the degree of population genetic differentiation (Johnson and Munshi-South 2017, Reed et al. 2020). Increases in genetic differentiation or divergence among populations within an urban setting is caused by habitat fragmentation and a decreased habitat size that isolated populations. This increases the effects of genetic drift and lowers the homogenizing effects of gene flow between populations (reviewed by Reed et al. 2020; examples in urban settings: McKinney 2002, Vandergast et al. 2009, Johnson and Munshi-South 2017). Measures of lower genetic diversity and increased inbreeding can identify isolated habitat patches that are population sinks. In addition, the increased effects of genetic drift in smaller populations that cause lower genetic diversity can reduce the adaptive capacity of these populations (reviewed by Reed et al. 2020). This is context-dependent because species’ response to urbanization depends both on the ecology of the taxa, the urban area(s), and the surrounding landscape (McIntyre 2000, McDonnell and Hahs 2008). For human-dependent, invasive species, there may be an opposite pattern of greater abundance in urban centers and potential for more genetic connectivity and greater genetic diversity due to larger human population sizes (Reed et al. 2020). The patterns of abundance and population genetic structure of public health threats like Ae. albopictus are critical to document for urban and rural landscape networks because they can inform the potential for population movement, the spread of insecticide resistance or other adaptive alleles, and what areas might be most effective to target with chemical pesticides during an outbreak (eg, population sources not sinks) (McDonnell et al. 1997, Blair and Johnson 2008, Ariori et al. 2017).

Much of Ae. albopictus’s invasion success can be attributed to its ecology. Aedes mosquitoes in the subgenus Stegomyia lay desiccant-resistant eggs in small, ephemeral pools of water. Containers used for oviposition can be natural (eg, tree holes, bamboo shoots, ivy) or anthropophilic (eg, planters, clogged gutters, bird baths, and human refuse). These artificial containers are commonly exploited by Ae. albopictus and have contributed to its spread along with international trade (Hawley et al. 1987). Aedes albopictus larvae may have a competitive advantage over other mosquito species, like Aedes aegypti, contributing to its invasion success through competitive displacement of resident species (Juliano 2010). The global distribution of Ae. albopictus is limited primarily by climate (Benedict et al. 2007, Cunze et al. 2016, Ding et al. 2018, Laporta et al. 2023). At finer spatial scales, Ae. albopictus presence, abundance, and growth depend on microclimate and landscape features, including urbanization, with evidence pointing toward an optimum of moderate urbanization akin to a suburban setting (Manica et al. 2016, Murdock et al. 2017, Reiskind et al. 2017, Spence Beaulieu et al. 2019, Hopperstad et al. 2021, Medley et al. 2015). Population genetic studies of Ae. albopictus in eastern North America have suggested significant population structure at coarse scales spanning hundreds to thousands of kilometers, but have not addressed finer scale questions at the level of 100 km or finer (Stone et al. 2020, Gloria-Soria et al. 2022). Given the mobility of flying insects and the transportation of mosquitoes by humans in cars, one would predict that there would be little to no genetic structure within a smaller geographic range, such as the level of a county.

To better understand how local Ae. albopictus abundances, stability, and connectedness vary across small spatial scales, we used both abundance and genetic data to describe population-level patterns within Wake County, North Carolina, United States. Aedes albopictus was first recorded in Wake County in 1993 and is the most frequently encountered mosquito in the county, making it a major pest (Kraemer et al. 2015, Kraemer et al. 2017, Reed et al. 2019, Reiskind et al. 2020). Previous studies did not examine distributions of Ae. albopictus at a within-county scale and did not investigate the population genetic patterns. To examine these patterns, we analyzed the population abundance and genetic characteristics of Ae. albopictus across different landscape “Zones” related to degree of impervious surface and human density. We used abundance data and genomic data from preserved Ae. albopictus individuals from a 2016 Wake County survey (Reed et al. 2019), reared from eggs collected in ovitraps, and from adult traps from more locations in 2018. We looked broadly at patterns within and between designated Zones of urbanization intensity. Integrating genomic methods with surveillance further clarifies the effects of urbanization on the population dynamics of invasive species at fine spatial scales and provides place-based information on Ae. albopictus population structure is critical for vector control.

Materials and Methods

Study Site

Wake County is 2,220 km², has a population of 1.1 million in 2019, and an average population density of 495.5 people/km² at the time of this study. Urban areas of Wake County are patchily distributed around the urban center, Raleigh, its largest city. To compare different urbanization categories, we grouped sites in the county into geographically delineated “Zones,” corresponding to differing levels of urbanization (human density and impervious surface) and by the major highways surrounding Raleigh (Fig. 1, Supplementary Table S1): (1) Inner Zone within the interstate I-440 and Interstate 40 beltway (urban), (2) Outer Zone between I-440, the I-540 beltway (suburban), and US route 1, and (3) Outside Zone as the remainder of Wake County (rural) (Fig. 1). These Zones are characterized in Supplementary Table S1.

Fig. 1.

Delineation of zones for Wake County, which act as proxies for the overall pattern of urbanization. In order from most to least urbanized: the Inner Zone is defined as inside the Interstate 440 beltline, the Outer Zone is between the Interstate 540 beltline and I-440, and the Outside Zone is the remainder of the county.

Sampling

We sampled over two different years in Wake County, 2016 and 2018. In 2016, we collected Ae. albopictus eggs from 15 sites weekly between 15 April and 26 October 2016 as part of a state-wide mosquito survey effort to characterize the distribution and abundance of Aedes species in North Carolina (Reed et al. 2019; Fig. 2A). The 15 locations were chosen for accessibility and habitat suitability for Aedes: five waste and recycling management centers, four gas stations, two residential backyards, and four miscellaneous buildings (school, government building, museum, retail) in each of our Zones (Inner: 4 sites, Outer: 4 sites, Outside: 7 sites). We collected eggs at each site using three ovitraps and hatched, reared, and identified the mosquitoes in the lab (for full methods, see Reed et al. 2019).

Fig. 2.

Wake County sampling locations. (A) sites and site names from 2016; all have genetic data for a portion of collected individuals (note we limit the number of individuals per ovitrap). There are four sites in the Inner Zone, five in the Outer Zone, and six in the Outside Zone. (B) Sampled sites from 2018; points overlaid with asterisks are sites that also have genetic data for mosquitoes. The number of sites in each Zone (and number of sites used in population genetic analyses in parentheses) are as follows. Inner Zone: six (four); Outer Zone: 18 (14); Outside Zone: 37 (24).

In 2018, we collected adult mosquitoes at 61 sites (Fig. 2B) using BG-sentinel traps baited with BG Lures, a proprietary chemical attractant targeted at anthropophilic Aedes species (Biogents GmbH, Regensburg, Germany). We collected mosquitoes from 7 June to 25 June 2018, sampling each location once a week, leaving the traps out to collect mosquitoes for 24 h. We then conducted an extra day of sampling for locations where a trap had failed (9 occurrences/181 trap × nights). Three of our 61 sites were identical to the sites in 2016. To determine the locations of the remaining sites, we used the r.random.cells function in GRASS GIS (GRASS Development Team 2018) to generate 100 random points across Wake County, each with a 1,000 m buffer within which no other points could fall. We chose the 1,000 m buffer size because previous studies found that Ae. albopictus rarely disperses further than this distance (Honório et al. 2003, Medeiros et al. 2017). For each of these 100 points, we generated a 100 m buffer within which we could place a trap. We then selected points that were deemed “accessible,” which we defined as areas within 1 Km of a public road and that were not entirely water, which eliminated 21 potential sampling locations. Of the 79 remaining, we randomly selected 60 using a random number generator in RStudio (RStudio Team 2018). We then exported these points and their buffers to Google Earth (Google Earth, Google 2008). In 2018, sites were categorized based on the site Zone (Inner, Outer, Outside), and given the differences in geographic area, we had 6, 19, and 40 sites per Zone, respectively.

We used different collection methods for the different years, and because there can be differences in genetic patterns between sampling types (Reed et al. 2023), we limited our comparisons within years for both abundance estimates and measures of genetic structure. For the 2016 data, to address the potential that egg sampling may be biased toward greater genetic differentiation than adult sampling, we only used one to three individuals reared from each ovitrap for a given week and evaluated multiple weeks per site.

Genomic Library Preparation

We built genomic libraries for sequencing following Burford Reiskind et al. (2016), which uses a double-digest restriction enzyme-associated DNA (ddRAD) approach with two restriction enzymes, SphI and MluCI. We included sample locations in the final analysis that had more than three individuals (in 2016, from different weeks to avoid siblings) with at least 8 ng DNA/µL after DNA extraction following the manufacturer’s protocol (Qiagen DNAeasy Tissue Kits). We built and sequenced 11 genomic libraries for a total of 15 sampling sites (100%) with 192 individuals in 2016, and 42 sampling sites (68.9%, with those not sampled primarily due to too few adult mosquitoes captured) with 336 individuals in 2018. We completed all sequencing at the NCSU Genomic Sciences Laboratory on the Illumina HiSeq 2500, and we conducted single-end sequencing of 100 bp fragments.

Bioinformatic Processing

The Illumina platform de-multiplexed to indices differentiating the two libraries in each lane, producing one FASTQ file per library. For each library, we checked the phred score to ensure high quality of sequence reads using FASTQC (Babraham Bioinformatics; http://www.bioinformatics.babraham.ac.uk/projects/fastqc). We then ran the process_radtags script in STACKS version 2.00 (Catchen et al. 2011) to filter low-quality reads (phred score <33), trim reads to 90 bp, and demultiplex barcodes to produce FASTQ files for each individual following Burford Reiskind et al. (2016).

For SNP (single-nucleotide polymorphism) detection, we conducted one STACKS denovo pipeline with all individuals (n = 528), which generated a unique catalog of SNPs. We ran all samples through this pipeline using the following parameters to generate SNPs: (1) minimum read depth (-m) of six, (2) maximum number of mismatches between reads within an individual (-M) of 3, and (3) maximum number of mismatches allowed between loci (SNPs) when combining them in a catalog of all individuals sampled (-n) of 2. We ran STACKS populations pipeline to remove SNPs that did not meet the following parameters: (1) present in a minimum of two sampling sites (-p = 2) and (2) present in 75% of individuals in a sample site containing that SNP (-r = 0.75). We ran the populations pipeline separately in two groups with the same filtering conditions to remove error: (1) Wake 2016 and (2) Wake 2018, which resulted in a different total number of SNPs per sample year.

Following the populations pipeline, we further filtered SNPs using PLINK v.1.19 (Purcell et al. 2007), removing variants with a minimum allele frequency (MAF) of less than 0.01, which removes monomorphic loci and a genotyping rate (GENO) of less than 0.5 to remove loci with a lot of missing data. We then removed individuals who had less than 25% of their remaining SNPs genotyped (MIND). Finally, we used the hw.test function in the R package pegas v 0.15 (Paradis 2010) to identify SNPs out of Hardy–Weinberg equilibrium and removed variants with a P value below the threshold value after applying a Bonferroni correction (P < 0.05/# SNPs). All filtering steps provide a robust set of SNPs that meet all the assumptions critical for population genetic data and remove noise generated by having loci with a lot of missing data for downstream analyses.

Abundance Analyses

We tested significant differences in Ae. albopictus abundance between Zones using a generalized linear mixed model (GLMM). As we were comparing count data and to account for natural aggregation in arthropod distributions, we used a negative binomial data distribution in JMP 18.0 (2024, JMP Statistical Discovery LLC, Cary, NC). For 2016 data, we compared eggs per ovitrap × week in each of our 15 sites, categorized into our three Zones (Inner, Outer, Outside). For 2018 data, we compared adult Ae. albopictus per daylight trap hour to correct for trapping effort in 62 sites and to reflect the diurnal activity of Ae. albopictus (Unlu et al. 2021, Wynne et al. 2024), categorized into our three Zones (Inner, Outer, Outside).

Genomic Analyses

To characterize and analyze the genomic patterns among sampling sites and Zones, we measured genetic diversity within and genetic differentiation among sampling sites. For genetic diversity, we measured both the degree of genetic variation and inbreeding within a site by calculating the expected heterozygosity (H_E) and the inbreeding coefficient (F_IS) corrected for small sample size for each location using the genetic_diversity function in the R package gstudio v1.5.2 (Dyer 2016). We evaluated significant differences in mean genetic diversity between Zones using GLMM, as described above. To test for genetic differentiation among sites and Zones, we estimated pairwise F_ST among sampling sites for each year using the R package hierfstat v0.5-11 (wcfst; Goudet et al. 2015). We tested significant differences in pairwise F_ST using 1,000 bootstraps and generated 95% confidence intervals (CIs; boot.ppfst). We determined significant differences when CIs did not include zero up to four significant figures.

We further evaluated geographic genetic structure using three different analyses. First, to understand if increases in geographic distance of sampling sites drive increases in genetic differentiation, we measured isolation by distance among all sample sites and among sites within Zones using a Mantel test (Sokal 1979). Second, we used a Bayesian method that focuses on genetic differences at the individual level to determine genetic differentiation among sampling sites using the program STRUCTURE v.2.3.4 (Pritchard et al. 2000, Hubisz et al. 2009). This individual-based Bayesian iterative algorithm assigns individuals to k clusters. We ran STRUCTURE using the admixture ancestry model with 10,000 burn-ins, 10,000 Markov Chain Monte Carlo (MCMC) replications, and cluster numbers (k) ranging from 1 to 10 with 10 iterations per k for 2016 and 2018 individuals. We used STRUCTUREharvester (Earl and von Holdt 2012) to determine the likelihood of a specific number of clusters (k) that best fit the data v using the Evanno method (Evanno et al. 2005). Third, we evaluated structure by conducting a discriminant analysis of principal components (DAPC), implemented in the R package adegenet v2.1.11 (Jombart 2008). DAPC can reveal more complex spatial genetic structure than clustering algorithms such as STRUCTURE or a principal component analysis (PCA) and does not make assumptions based on population genetic models (Plue et al. 2019). In DAPC, we chose the optimal number of principal components (PCs) based on the PC value with the lowest root mean squared error and the highest mean success rate after cross-validation with a training set size of 0.95, 1,000 replicates, and a maximum number of PCs equal to a third of the individuals included in the dataset to prevent overfitting the data (Jombart et al. 2008, Plue et al. 2019).

Beyond analyses that focused on genetic differentiation among sites, we further investigated genetic divergences within and between Zones in Wake County using DAPC. For within Zones, we used cross-validation in DAPC to decide the optimal number of PCs to retain. Next, we fit Zones separately with the same number of PCs and compared correct assignment rates. We interpreted higher rates of assignment as indicating greater genetic differentiation among sites within a given Zone. To account for different numbers of sites within Zones, we used a rarefaction method to generate a frequency distribution for Zones with more sites. To do this, we determined the Zone with the fewest sites and randomly selected an equal number of sites from the remaining Zones. We then found the correct assignment rate for that subset of populations. We repeated this process over 1,000 iterations to generate the frequency distribution for the Zone, and from which we calculated a mean and 95% confidence interval.

Results

Mosquito Abundance

Over 80% of the mosquitoes trapped were Ae. albopictus (Supplementary Table S2A). In 2016, the sites in the Outer Zone between beltlines I-440 and I-540 had the highest average abundance with 110.88 ± 31.48 (Standard Error of the Mean (SEM)) eggs per ovitrap per week, followed by the Outside Zone (97.61 ± 22.70 (SEM) eggs per trap per week) with the lowest in the Inner Zone (65.37 ± 31.62 (SEM) eggs per trap per week). These differences were not significant between Zones (F = 1.919, df = 2, P = 0.1929; Fig. 3A).

Fig. 3.

Average A. albopictus abundance in Wake County. (A) Average egg abundance per week in 2016, larger points indicate higher average egg abundance. Aedes albopictus eggs were found at all sampling locations, and average count ranged from 10.75 to −213.59 per ovitrap per week. There were no significant differences in mean abundance between Zones. Sites in the Inner Zone are shown in red, those in the Outer Zone in gray, and in the Outside Zone, blue. (B) Average adult abundance per light-hour in Wake County 2018, at 61 locations. Larger points indicate higher average adult abundance. Aedes albopictus adults were found at 59/61 sites, and average abundance ranged from 0.02 to 10.31 adults per light-hour. There were no significant differences in mean abundance between zones. Sites in the Inner Zone are shown in red, those in the Outer Zone in gray, and in the Outside Zone in blue.

In 2018, we trapped 2,553 mosquitoes, of which 2,086 were Ae. albopictus (81.71%; Supplementary Table S2B). Of the 61 sites we sampled, we found Ae. albopictus at all but two (Fig. 3B). Overall, the trap rate per daylight hour, defined as the hours between sunrise and sunset on the day(s) of collection, of Ae. albopictus across all sites was 0.9757 ± 0.2143 adults/daylight hour. The highest average abundance was in the Inner Zone, with 1.58 ± 0.79 (SEM) Ae. albopictus adults per hour of daylight. Unlike in 2016, the Outer Zone average was close to the Inner Zone, with 1.53 ± 0.56 (SEM) adults trapped per daylight hour. The lowest average abundance was in the Outside Zone with 0.612 ± 0.17 (SEM) Ae. albopictus adults per daylight hour across 44 sites. As in 2016, there were no significant differences in adult abundance among Zones (F = 2.113, df = 2, P = 0.1303).

Double-Digest RAD Sequence Libraries

In 2016, the STACKS de novo pipeline identified 4,003,920 SNPs across 192 individuals from 15 sites, and 166,852 were retained after the populations pipeline. During the filtering step, SNPs were further culled for MAF, genotyping rate, and Hardy-Weinberg Equilibrium (HWE) (Table 1). We removed 11 individuals for low genotyping rates, leaving a final dataset of 181 individuals and 28,347 SNPs after filtering (Table 1).

Table 1.

Summary of sequencing and bioinformatic processing and single-nucleotide polymorphism (SNP) filtering for 2 years of samples for Wake County, North Carolina: 2016 and 2018

Year	No. of sites	Range n per site		De novo pipeline (STACKS)	Populations pipeline (STACKS)	MAF (plink)	GENO (plink)	MIND (plink)	HWE—final dataset
2016	15	7–18	Individuals	192	192	192	192	181	181
			Variants	4003920	166853	151184	34695	34695	28347
2018	42	5–10	Individuals	336	289	289	289	276	276
			Variants	3096027	183753	169713	16066	16066	11305

The table includes the total number of sites, the range of individuals sequenced per site (n), total number of individuals and variants, and number of individuals and variants per each step of filtering from the de novo pipeline to the HWE filtering. N.B. In 2016, we limited sampling to 1–3 individuals for a given ovitrap × week to avoid sampling siblings.

In 2018, the STACKS de novo pipeline identified 3,096,027 SNPs, and 183,753 were retained in the populations pipeline. During the filtering step, SNPs were further culled for MAF, genotyping rate, and HWE (Table 1), leaving a final dataset of 276 individuals and 11,305 SNPs (Table 1).

Population Genetics: 2016

Genetic diversity measured by expected heterozygosity (H_E) of sites ranged from 0.1203 to 0.1389, and mean H_E did not differ between the Inner, Outer, and Outside Zones (P = 0.4360; Table 2). The inbreeding coefficient F_IS ranged from −0.0250 to 0.0953 (Table 2). Only one site, WA in the Inner Zone, had a negative F_IS value, which indicates that individuals sampled at this location had higher levels of heterozygosity than expected under Hardy–Weinberg equilibrium. Zone had a significant effect on F_IS (P = 0.0463; Table 2), with the Outside Zone having a significantly higher mean F_IS value (0.0826) than the Inner Zone (0.0401; Table 2), indicating that individuals sampled in this Zone had significantly higher levels of homozygosity than would be expected under Hardy–Weinberg equilibrium. This result is supportive of either small population dynamics or admixed populations that are not interbreeding.

Table 2.

Mean expected heterozygosity (H_E) and inbreeding coefficient (F_IS) within groups for each year pair in Wake County, Raleigh, NC

Year	Zone	HE	FIS
2016	Inner	0.1300	0.0401*
	Outer	0.1350	0.0717
	Outside	0.1320	0.0826*
2018	Inner	0.0984	0.1120
	Outer	0.1020	0.0790
	Outside	0.1000	0.0850

Asterix (*) indicates groups whose differences in genetic diversity were statistically significant.

There was no signal of isolation by distance (Mantel test: observation = −0.3934; P = 0.977) across all sites. However, in the within Zone analysis, we did find evidence of isolation by distance in the Outside Zone (Mantel test: observation = 0.3833; P = 0.01). We found evidence of significant genetic differentiation in 2016 for 95 of the 105 comparisons (90%) to four significant figures (Table 3). Two sites, VDR and WA in the Inner Zone, showed greater genetic differentiation compared to other locations than other sites in the dataset. Both sites were either in a residential/commercial or solely residential area near downtown Raleigh. To a lesser degree, the pairwise F_ST results showed genetic differentiation at sites OOR, WNS, and KDB in the Outer Zone. All three sites were also either within or near residential areas with higher human density. Overall, we found a similar degree of genetic divergence among sample sites within Zones. Our analysis of genetic structure using STRUCTURE supported three clusters (k) using STRUCTUREharvester and the Evanno method (k = 3; Supplementary Fig. S1A), and these clusters were associated with specific sites, not specific Zones.

Table 3.

Aedes albopictus pairwise F_ST values among sites in (A) Wake 2016 and (B) Wake 2018 in the R package hierfstat

Background color of site names shows the Zone that the site belongs to: red = Inner, gray = Outer, and blue = Outside. Bolded numbers are comparisons that were identified as significantly differentiated following a bootstrap analysis in hierfstat (bootstraps = 1,000), generating 95% CIs. We restricted our significance test using CIs that did not include zero to three significant figures. Increases in shading indicated increases in pairwise F_ST values. For the 2018 dataset, the last row indicates the number of significant pairwise comparisons for each site. Two sites are identical between 2016 and 2018, WNS–SWN and VDR–SVD. These data are generated using 28,347 SNPs for 181 individuals from 2016 and 11,305 SNPs for 276 individuals from the 2018 datasets.

Our DAPC analysis showed sample sites primarily formed one cluster, with three sites, WNS and KBD (Outer Zone) and WA (Inner Zone), isolated from the remaining individuals (50 PCs and 0.087 assignment rate; Fig. 4A). The results of cross-validation comparisons for number of PCs retained and assignment rate varied across Zones for DAPC, with greater assignment in the Inner and Outer Zones when we corrected for number of sites and number of retained PCs (Supplementary Fig. S2). This indicates greater genetic divergences among sites within these two Zones compared to the Outside Zone. We found a similar trend with the DAPC analysis within each Zone, with less clustering of sites in the Inner Zone than the other two zones (Supplementary Fig. S2). However, there were sites that did not group within the other two Zones, such as KDB in the Outer Zone and DPD and SRC in the Outside Zone.

Fig. 4.

DAPC scatterplots for all sample sites. (A) 15 sites sampled in Wake County 2016, with the first two discriminant functions (DAs) on the x and y axes. Cross-validation retained 50 principal components, which produced a correct assignment rate of 0.807. (B) for the 42 sites sampled in Wake County 2018, with the first two discriminant functions (DAs) on the x and y axes. Cross-validation retained 60 principal components, which produced a correct assignment rate of 0.62.

Population Genetics: 2018

For measures of genetic diversity, there was minimal variation in H_E, which ranged from 0.0937 to 0.1058, and we did not find significant differences in mean H_E between Zones (P = 0.1019) (Table 2). F_IS was more variable and ranged between 0.0201 and 0.1562. Unlike what we found in 2016, there were no significant differences in F_IS among Zones (P = 0.3040; Table 2). Therefore, we found the sampling sites in general had similar genetic diversity with individuals that mostly conformed to Hardy–Weinberg equilibrium expectations.

In 2018, we did not find evidence of isolation by distance among sites or within any of the three Zones. However, we found significant genetic differentiation among sampling sites in the pairwise F_ST analysis (Table 3B). Of the 861 pairwise comparisons, there were 355 significant pairwise F_ST values (41.2%). The five sites with the greatest number of significant pairwise F_ST were S24 and SWN in the Outer Zone, and SO7, S26, and S41 in the Outside Zone, with between 12 and 31 significant differences (Table 3B). In general, all five sites showed significant pairwise differences to sites within and across Zones. We found greater genetic divergence among sites within the Outer Zone, with the Outside Zones as a close second compared to the Inner Zone. STRUCTURE harvester identified an optimal k = 4, though most individuals had much of their ancestry assigned to the same cluster (Supplementary Fig. S1B).

Most populations formed a single cluster in DAPC, with some differentiation from the major cluster at sites S24, S26, and to a lesser degree at sites S41, S58, S27, and S01 (40 PCs retained; Fig. 4B). The two sites that were the most differentiated, S24 and S26, were located in the Outer and Outside Zone, respectively. In comparison to the pairwise F_ST, these results suggest six of the 42 populations were genetically differentiated (F_ST  Table 3B; DAPC Fig. 4B).

When divided among Inner, Outer, and Outside Zones (Supplementary Fig. S2 Wake County 2018), after cross-validation, the results of the number of PCs retained and assignment were similar across all Zones (Supplementary Fig. S2), indicating a similar degree of genetic divergence among sites within Zones (Supplementary Fig. S2). For the DAPC analysis within Zones, the same locations that were outside of the main cluster in Fig. 4 were also within Zones, with the addition of S35 in the Outside Zone (Supplementary Fig. S2).

Discussion

Within the small geographic extent of Wake County, NC, we found a mosaic of spatial genetic variation, suggesting local, idiosyncratic factors at each site contribute to the population genetic structure of Ae. albopictus. We found genetic structure among sites in different Zones representing different levels of impervious surface and human density, such that the significant genetic differentiation was not unique to different degrees of urbanization as measured in our study. In 2016, we saw evidence of higher inbreeding levels and isolation by distance in the Outside Zone, indicative of small population size dynamics in this Zone. This is consistent with, in this year and Zone, sites acting as population sinks with the patchiness of habitat suitability limiting genetic connectivity. While we did not find isolation by distance and inbreeding in the Outside Zone in 2018, demonstrating the temporal fluctuation in these dynamics, we did see greater genetic differentiation among sites in this Zone. Overall, we found a greater number of sites genetically differentiated from other sites in 2016 than in 2018, which indicates there are other abiotic and biotic factors that drive these patterns year to year. Indeed, the patterns of abundance data were also variable across site and year, and these dynamics suggest that these year-to-year and among-site variations may be influenced by unmeasured abiotic factors and small population sizes.

Trends in abundance in Wake County within each year between 2016 eggs and 2018 adults were somewhat contradictory. Specifically, the Inner Zone had the lowest egg abundance relative to other sites in 2016 and the highest adult abundance in 2018, though differences were not statistically significant in either year. The scope of this study does not permit us to identify whether this switch was due to changes in mosquito population dynamics, land-use changes, or sampling methods and effort between years (both trapping approach and site selection). Targeting the egg-stage is most effective when there are fewer competing larval habitats, while targeting host-seeking adults is most effective when there are fewer hosts (Silver 2007). For our study, these two situations may differ by level of urbanization in different ways, affecting ovitrapping and adult trapping differently. Site selection may also have affected our measurements of abundance between years. In 2016, sites were specifically chosen based on where mosquitoes were likely to be found, while in 2018, sample sites were chosen to capture a wider area, with most sites chosen randomly. This expansion of sampling allowed us to capture a broader picture of the population dynamics of Ae. albopictus. However, many of the sites in the 2018 Outside Zone were in landscapes less hospitable to high densities of Ae. albopictus (Barker et al. 2003, Reiskind et al. 2017, Reed et al. 2019). For example, some sites were located by agricultural fields, in a forest interior, and in high grass, not areas previously associated with high Ae. albopictus abundance (Barker et al. 2003, Reiskind et al. 2017). Others have reported that highly urbanized areas are also marginal for Ae. albopictus relative to more suburban sites (Manica et al. 2016). While site selection may explain some of the differences in trends we observed in the abundance data, we still obtained adequate sampling for the population genomic analysis.

The population genetic analyses suggest variable genetic structuring between years and indication of the genetic signatures of small population dynamics. While we did not find significant differences in the expected heterozygosity, an indication of lowered genetic diversity, we did see a trend of increasing genetic diversity comparing the Inner Zone to the other Zones in both years. While this trend would need to be confirmed with further sampling and sequencing, it suggests that there are greater small population dynamics in urban, relative to suburban or more rural, settings in Wake County. The urban center of Wake County in Raleigh, while on average has a higher impervious surface, still has suitable suburban habitat intermixed, and if small-population dynamics are important, they may not operate at all sites. The significant degree of among-site genetic differentiation, found in both years to varying degrees, and evidence of inbreeding in the Outside Zone suggests that small population dynamics are not unique to the urban setting but also found in locations with marginal habitat or lower human densities.

In contrast, patterns of mean inbreeding coefficients (F_IS) were more variable. The inbreeding coefficient, despite its name, is not a direct measure of relatedness between individuals. Instead, it indicates the degree to which observed levels of heterozygosity within a population deviate from the expected levels of heterozygosity given the population’s allele frequencies at the genetic markers under study. If we consider two populations with the same level of expected heterozygosity, the population with the higher F_IS value will have greater between-individual differences because within each individual more loci are homozygous, while the population with lower F_IS will have more heterozygous loci within the individual, but lower differences between individuals. Therefore, the relatively low levels of expected heterozygosity paired with low F_IS in urbanized Zones indicate that there were fewer biallelic SNPs in those populations, but that individuals were more likely to be heterozygous at those loci. This could indicate frequent mating events between unrelated individuals, perhaps from migration events or the admixture of multiple bottlenecks. In contrast, the higher F_IS found in the Outside Zone in 2016 would indicate that individuals here were more homozygous than expected of a large interbreeding population, and this Zone may support smaller, isolated populations that may not interbreed. Overall, the patterns in the inbreeding coefficient support the importance of local, small-scale population dynamics within each Zone.

Examination of the geography of genetic divergence and genetic structure showed that the Inner Zone was more genetically differentiated among sites, followed by the Outer and Outside Zones in 2016, indicating little gene flow between the urban-most locations. For instance, in 2016, the site WA formed a separate genetic cluster in both DAPC scatterplots, despite being located less than 1,000 m from another sampled site, NRC. However, this pattern was not as strong in 2018 as it was in 2016, with all three Zones showing similar degree of genetic differentiation. This suggests that in some years, there is minimal gene flow between more urban locations but does not preclude in both years gene flow in or out of rural areas to more urban populations.

Our results suggest that the spatial heterogeneity and sources and sinks across Wake County may vary year by year. This could be explained by differences in degree of connectivity, with some years supporting more widespread gene flow and lower genetic differentiation driven by factors we did not measure. While it was outside the scope of this study, a landscape genetic analysis would aid in better understanding source and sink dynamics and gene flow among these three Zones and how this relates to landscape features (Reed et al. 2020). Our study supports mosquito control by addressing small spatial scale dynamics, such as within residential zones, to better understand the degree of genetic differentiation among sites, indicative of low gene flow and population isolation, and identify locations that may be a year-to-year source of individuals for other locations in the urban and rural landscape (reviewed by Schmidt et al. 2021).

The patterns we observed point to several areas for future research. First, we saw more genetic differentiation and variation in abundance when we analyzed reared adults rather than trapped adults, and while we limited the number of individuals from a given ovitrap × week to avoid relatedness, we do not know whether the reared adults would have emerged naturally in the wild. Consideration of sampling methods that target trapped adult mosquitoes, while also evaluating suitable habitat, would be critical for any landscape genetic study that seeks to estimate the degree of realized gene flow associated with landscape features. The differing patterns between the 2 years in abundance measures and genetic differentiation we found in this study may in part be due to including sample sites in the Outside Zone in areas that are marginal and do not support large mosquito populations.

While the genomic approach we used is powerful and revealed genetic structure at a finer spatial scale than previous studies at coarser spatial scales, we did not identify consistent year-to-year genetic breaks. Considering the increase in genetic control methods of mosquito management, we argue that measuring genetic structure, gene flow, and directional migration between populations, especially between and within urban and rural locations, is critical for individual mosquito management units (Sinkins and Gould 2006). Combining this with an understanding of the role of the intervening landscape matrix in determining genetic diversity, connectivity, and source-sink dynamics would be important. In addition, careful consideration of species ecology, introduction history, and location-specific features of the urban–rural network, as well as a landscape genetic approach are necessary to predict population responses to future land-use change due to human development. Understanding gene flow can provide a basis for predicting the spatiotemporal dynamics of not only how individuals move in a population, but also the potential for pathogen transmission and the change in frequency of insecticide resistance.

Author Contributions

Emily M.X. Reed (Conceptualization [lead], Data curation [lead], Formal analysis [lead], Methodology [lead], Writing—original draft [lead]), Michael H. Reiskind (Conceptualization [supporting], Methodology [supporting], Writing—review & editing [supporting]), and Martha O. Burford Reiskind (Conceptualization [supporting], Formal analysis [equal], Methodology [supporting], Resources [lead], Supervision [lead], Writing—review & editing [equal])

Supplementary Material

Supplementary material is available at Journal of Medical Entomology online.

Funding

This study was supported by the USGS Southeast Climate Adaptation Science Center graduate fellowship awarded to E.M.X.R. and was funded by the Wynne Innovation Grant from the CAL Dean’s Enrichment Grant Program at NCSU awarded to M.O.B.R.

Conflicts of Interest

None declared.

Data Availability

The genomic datasets that support this manuscript will be available in M.O.B.R.’s Dryad account specific to this publication (https://doi.org/10.5061/dryad.7d7wm388p).