A comprehensive assessment of inbreeding and laboratory adaptation in Aedes aegypti mosquitoes

Abstract Modified Aedes aegypti mosquitoes reared in laboratories are being released around the world to control wild mosquito populations and the diseases they transmit. Several efforts have failed due to poor competitiveness of the released mosquitoes. We hypothesized that colonized mosquito populations could suffer from inbreeding depression and adapt to laboratory conditions, reducing their performance in the field. We established replicate populations of Ae. aegypti mosquitoes collected from Queensland, Australia, and maintained them in the laboratory for twelve generations at different census sizes. Mosquito colonies maintained at small census sizes (≤100 individuals) suffered from inbreeding depression due to low effective population sizes which were only 25% of the census size as estimated by SNP markers. Populations that underwent full‐sib mating for nine consecutive generations had greatly reduced performance across all traits measured. We compared the established laboratory populations with their ancestral population resurrected from quiescent eggs for evidence of laboratory adaptation. The overall performance of laboratory populations maintained at a large census size (400 individuals) increased, potentially reflecting adaptation to artificial rearing conditions. However, most individual traits were unaffected, and patterns of adaptation were not consistent across populations. Differences between replicate populations may indicate that founder effects and drift affect experimental outcomes. Though we find limited evidence of laboratory adaptation, mosquitoes maintained at low population sizes can clearly suffer fitness costs, compromising the success of “rear‐and‐release” strategies for arbovirus control.

The extent of laboratory adaptation can vary between insect orders (Hoffmann & Ross, 2018), and this could reflect differences in the range of conditions that can be tolerated relative to the conditions experienced in the laboratory (Ochieng'-Odero, 1994). Laboratory environments that are suboptimal will impose strong selective pressures on mosquito populations, leading to rapid adaptation (e.g., Watson et al., 2000). Colonized mosquito species can require a specific set of conditions such as swarm markers (Watson et al., 2000), artificial horizons (Marchand, 1985), dusk periods (Marchand, 1985), or exposure to stroboscopic light (Lardeux et al., 2007) to improve their reproductive success in the laboratory. Other species will not freely reproduce in the laboratory at all, requiring induced copulation over successive generations before free-mating colonies can be established (Bryan & Southgate, 1978;McDaniel & Horsfall, 1957). In contrast, Ae. aegypti collected from the field perform well in the laboratory without any of these specific requirements (e.g., Munstermann, 1997), and therefore, less drastic differences in traits would be expected between laboratory and field populations due to a lack of selective pressures.
Rear-and-release programs with modified Ae. aegypti mosquitoes are now underway in several countries, and many of these programs rely on the use of mosquitoes that have been inbred or maintained in the laboratory for extended periods. We colonized replicate Ae. aegypti populations collected from Queensland, Australia, to assess the effects of laboratory maintenance and inbreeding on life history traits in this species. We find that inbreeding is costly and is associated with a reduction in effective population size, but we find limited evidence of laboratory adaptation for most life history traits.
Modified mosquitoes reared for disease control programs should therefore be maintained at large population sizes and/or crossed with field populations prior to field release. Our research highlights potential issues with maintaining colonized insects that are destined for field release, and informs protocols for the maintenance of Ae. aegypti in the laboratory.

| Replicate population establishment
Aedes eggs collected from ovitraps near Townsville, Australia, in September 2015 (Ritchie, 2001) were hatched and reared in the laboratory (see Colony maintenance). Aedes aegypti larvae were separated from other species based on an identification key (Rueda, 2004).
A total of 327 Ae. aegypti adults (171 males and 156 females) were obtained and added to a single 19.7-L (27 cm 3 ) BugDorm-1 ® colony cage (MegaView Science Co., Ltd., Taichung City, Taiwan). Females were blood-fed, and all eggs laid were pooled and hatched in a single plastic tray (30 × 20 × 10 cm) containing 3 L of water. Larvae were selected at random and divided into groups to establish replicate populations ( Figure 1). Five populations were maintained at a census size of 400 adults (large populations), and five populations were maintained at a census size of 100 adults (small populations). Twenty adult females were also isolated for oviposition. The offspring from F I G U R E 1 Maintenance scheme for replicate Aedes aegypti laboratory populations. An ancestral population was established from eggs collected from Townsville, Australia, that all other populations were derived from. Replicate populations were maintained separately beginning from F 2 and were not interbred five isolated females were used to establish five additional populations maintained at a census size of 100 adults (isofemale lines), while the offspring from 10 females were maintained with a single male and female each (inbred lines). At least two mating pairs were established for each inbred line (Supporting Information Table S2), but only a single pair was used to found the next generation. Their offspring underwent full-sib mating each generation for nine generations, and then all progeny were interbred during F 12 to build up numbers for experiments. All replicate populations were maintained until F 13 when experimental comparisons were performed. All adults from the ancestral Townsville population (F 1 ) and the replicate populations at F 13 were stored in absolute ethanol at −20°C for pooled double-digest RADseq. Only two inbred lines had sufficient numbers for RADseq due to the loss of most inbred lines over the course of full-sib mating (Supporting Information Table S2).
Aedes aegypti eggs can withstand desiccation and remain viable for up to 1 year (Faull & Williams, 2015). We utilized this ability to perform direct comparisons between the ancestral population and the derived populations simultaneously. Eggs laid by F 2 and F 3 females were stored under humid conditions for 7-8 months at 26°C and then hatched at the same time as eggs laid by F 11 females from the other populations. A colony derived from larvae that hatched was maintained under standard conditions for one generation, and their progeny (F 4/5 ) were used for experiments alongside the populations at F 13 . We used a relaxed generation to avoid deleterious effects associated with extended quiescence (Perez & Noriega, 2012); however, we cannot rule out any indirect effects on fitness. Colonies derived from eggs collected from Cairns and Innisfail, Australia, were also used for experimental comparisons. These colonies were maintained as single caged populations with a census size of 400 individuals. Quiescent eggs from the Innisfail population were also used to generate a colony that had experienced fewer generations of maintenance under laboratory conditions. Eggs collected from Cairns at a later stage were used to establish a colony for comparisons with the Cairns colony at F 22 .

| Colony maintenance
All populations were maintained in a controlled temperature laboratory environment (26 ± 0.5°C and 50%-70% relative humidity, with a 12:12-hr light:dark photoperiod) following the protocol described by Ross, Axford, Richardson, Endersby-Harshman, and Hoffmann (2017). This protocol is designed to reduce selection against individuals that are slow or quick to develop, mature, mate, blood feed, oviposit, or hatch, and to minimize mortality at each life stage. To maintain each population, all eggs from the previous generation were pooled and a random subset of larvae was provided with food (TetraMin ® tropical fish food tablets, Tetra, Melle, Germany) ad libitum and reared to adulthood. For the large populations, 400 adults were selected at random and added to 19.7-L cages, while for the small populations and isofemale lines, 100 adults were added to 12-L (30 × 20 × 20 cm) cages. For the inbred lines, a single male and female were added to a 1.5-L (10 × 10 × 15 cm) cage. Except for the inbred lines, sex ratios were maintained naturally, and equal numbers of males and females were not counted. All cages were provided with a source of water and 10% sucrose. Approximately three days after the last adult had emerged, females were blood-fed on a single human volunteer. Two days after blood feeding, cups containing larval rearing water and lined with sandpaper strips were introduced into the cages. Eggs laid on the sandpaper strips were collected over the span of 1 week, and all eggs were hatched three days after females had ceased oviposition. We followed this procedure until the Townsville populations were at F 13 , with each generation taking 28 days to complete. Blood feeding of mosquitoes on human subjects was approved by the University of Melbourne's Human Ethics Committee (approval #: 0723847). All volunteers provided informed written consent.

| Fitness comparisons between Townsville F 13 populations
We compared all Townsville populations at F 13 for their development time and survival to adulthood under two nutrition conditions, and the fecundity and egg hatch rate of females reared under high nutrition conditions. Not all inbred lines were included in the experiments as the majority were lost by F 13 (Supporting Information Table S2).
Two of the four remaining inbred lines were only tested under high nutrition conditions due to low numbers, and these lines later became extinct (Supporting Information Table S2). Cairns (F 2 and F 22 ), Innisfail (F 4 and F 10 ), and Townsville (F 4/5 ) populations were included in all experiments.
One hundred larvae from each population were reared in containers with 500 ml of water and provided with TetraMin ® ad libitum (high nutrition) or with 0.1 mg of TetraMin ® per larva every 2 days (low nutrition). Four replicate containers were reared for each population, except for two inbred lines where less than 400 larvae were obtained.
A random subset of females from each population that emerged from the high nutrition treatment were blood-fed and then isolated for oviposition. Eggs collected from each female were counted and hatched 3 days postcollection. Egg hatch rates were determined by calculating the proportion of eggs that had a detached cap.
Fitness data from the Townsville populations at F 13 were used to estimate the performance of each population relative to the Townsville F 4/5 ancestral population. We simplified an equation from Livdahl and Sugihara (1984) to calculate performance from fecundity, egg hatch, survival, and larval development time data.
F is the mean fecundity of each population multiplied by egg hatch proportion, S is the mean proportion of larvae surviving to adulthood, and D is the mean larval development time in days. The performance index of each population at F 13 was divided by the performance index of the ancestral population to determine their relative performance.

| Male mating competitiveness
We tested the male mating competitiveness of populations from Cairns that were at F 2 , F 7 , or F 27 in the laboratory, and an inbred line from Townsville (Inbred A) at F 18 . Males from all populations competed against males infected with the wAlbB strain of Wolbachia for access to F 2 females in a caged laboratory environment. Males infected with wAlbB induce complete sterility (eggs do not hatch) when crossed to uninfected females under standard laboratory conditions (Axford, Ross, Yeap, Callahan, & Hoffmann, 2016;Xi et al., 2005).
Thus, the competitive ability of each population relative to wAlbB-infected males can be estimated by scoring egg hatch rate from crosses between uninfected females and a mix of Wolbachia-infected and uninfected males (Chambers et al., 2011;Segoli, Hoffmann, Lloyd, Omodei, & Ritchie, 2014

| Pooled double-digest RADseq library preparation
We used pooled double-digest RADseq to determine the effective population size (N e ) of the 17 replicate populations from Townsville at F 13 relative to their ancestral population (F 1 ). These included the five large populations, five small populations, five isofemale lines, and two inbred lines. We prepared a library following methods described by Rašić, Filipović, Weeks, and Hoffmann (2014) and Schmidt, Filipović, Hoffmann, and Rašić (2018)

| Data processing and effective population size estimates
We checked the quality of the raw sequencing data with FastQC (minimum mapping quality below 20) were discarded, and alignments were converted to SAM format and sorted using SAMtools v1.4 (Li et al., 2009). Sorted files were then converted to mpileup format, with each file containing the ancestral population and one of the seventeen derived populations. These files were converted to sync format using the mpileup2sync.jar tool from Popoolation2 (Kofler, Pandey, & Schlotterer, 2011). We then estimated effective population size (N e ) using the Nest R package v1.1.9 with three different methods (Jonas, Taus, Kosiol, Schlotterer, & Futschik, 2016).

| Statistics on life history traits
All data were analyzed using SPSS Statistics version 24.0 for Windows (SPSS Inc, Chicago, IL, USA). Not all groups could be easily compared as variances across isofemale lines, and inbred lines were expected (and observed) to be much larger than the other populations. Data that were normally distributed were analyzed using general linear models (GLMs) and ANOVAs, and data that could not be normalized by transformations (log for development

| Larval development time
When the Townsville populations had reached F 13 , we performed fitness comparisons to test for inbreeding effects, laboratory adaptation, drift, and founder effects. We measured larval development time for all populations under high nutrition and low nutrition conditions ( Figure 2). In the analysis of large and small populations under high nutrition conditions, there was a significant effect of sex, population, and replicate cage on (log) development time, but no interaction between sex and population or replicate cage (Table 1).
Under low nutrition conditions, there was an effect of sex and replicate cage and an interaction between sex and population but no overall effect of population (Table 1)

| Survival to adulthood and sex ratio
We compared the proportion of larvae that survived to adulthood between populations in the larval development experiment

| Overall performance
We calculated an index of performance for each Townsville population at F 13 relative to the ancestral population (Townsville F 4/5 ) using the data for fecundity, egg hatch proportion, larval development time, and survival to adulthood (under high nutrition conditions) available for each population ( Figure 5). The large populations (census size 400) consistently performed better than the ancestral popula- The Cairns laboratory population (F 22 ) had increased performance over the field (F 2 ) population (relative performance index: 1.080), but the Innisfail laboratory population (F 10 ) had decreased performance over the field population (F 4 ) (relative performance index: 0.960).
Laboratory populations therefore did not always exhibit increased performance over the populations that were colonized more recently.

| Effective population size
We estimated the effective population size (N e ) of the replicate Townsville populations at F 13 relative to the ancestral population (F 1 ) using pooled RADseq and the Nest R package v1.1.9 (Jonas et al., 2016). The N e (JR) and N e (P) methods provided similar estimates of N e , but N e (W) provided estimates that were in many cases much larger than the census sizes. For estimates calculated using the N e (JR) and N e (P) methods, N e declined substantially with decreasing census size ( Table 2). Ratios of N e to census size calculated using the N e (P) method were low, though the small populations (census size 100, mean N e /N = 0.250) had higher ratios than large populations (census size 400, mean N e /N = 0.143). The index of performance for each population increased dramatically with increasing N e but levelled off at higher N e (Supporting Information Figure S1). These findings demonstrate a clear association between N e and fitness (Spearman's rank-order correlation: ρ = 0.973, p < 0.001, n = 17) but suggests that an N e greater than used in the large populations will lead to only small fitness improvements.

| Mating competitiveness
Males from the Cairns F 2 , F 7 , or F 27 and inbred (Inbred A F 18 ) populations competed for access to F 2 females against a standard competitor infected with Wolbachia (wAlbB strain) ( Figure 6). Hatch proportions did not differ significantly between the F 2 , F 7 , and F 27 populations (one-way ANOVA: F 2,12 = 0.829, p = 0.460), but were markedly reduced for inbred males relative to the other populations (one-way ANOVA: F 1,18 = 39.784, p < 0.001). These results indicate that male mating success in laboratory cages is not affected by longterm laboratory maintenance, but can be decreased by inbreeding.
The poor performance of inbred males was likely due to reduced mating success and not a paternal effect on female fertility, as crosses between inbred males and Cairns F 2 females produced eggs with high hatch proportions (Supporting Information Appendix S4).

| D ISCUSS I ON
We performed a comprehensive assessment of inbreeding and laboratory adaptation in Ae. aegypti mosquitoes to inform rear-and-release programs for arbovirus control. Our study is the first to investigate the effects of inbreeding on Ae. aegypti fitness directly by comparing outbred and inbred lines derived from the same population, and the first that links fitness costs to reductions in effective population size as assessed through genomic markers. We look for evidence of adaptation by comparing laboratory populations to their direct ancestor concurrently and use replicate populations to separate fitness changes due to adaptation from drift and founder effects, two approaches which have not been previously applied in mosquitoes.
We find evidence of laboratory adaptation in colonized Ae. aegypti populations, but changes in trait means were small in magnitude and directions were often inconsistent between populations.
All replicate laboratory populations from Townsville developed faster and were smaller than mosquitoes from the ancestral population. These changes could be a response to selection for abbreviated development in the laboratory, despite efforts to avoid selection in our laboratory rearing protocol (Ross et al., 2017). Shorter developmental periods are often observed in laboratory-adapted insects (Allgood & Yee, 2014), particularly under mass-rearing conditions that favor the rapid production of insects (Economopoulos, 1992;Miyatake, 1993). In contrast, development times can increase in However, rearing mosquitoes on a larger scale may introduce additional selective pressures that affect field performance resulting from crowded rearing conditions (Zhang et al., 2018).
When fitness traits were combined into an overall index of performance, we found that laboratory populations maintained at a large census size usually had greater performance than field populations. This finding is consistent with other insects, where fitness under laboratory conditions tends to improve with laboratory maintenance (Hoffmann & Ross, 2018), though a recent review and set of experiments in Drosophila found a lack of clear directional trends across multiple species (Maclean, Kristensen, Sorensen, & Overgaard, 2018). The rate of adaptation in our laboratory colonies of Ae. aegypti was slower than other mosquito species and insects in general (Hoffmann & Ross, 2018). Aedes aegypti collected from the field performed well from the first generation in the laboratory, potentially because this species is already somewhat adapted to living in artificial environments (Cheong, 1967). Rates of adaptation are likely to be higher for species such as Aedes notoscriptus where the laboratory environment is suboptimal and only a small proportion of individuals can reproduce in the initial generations (Watson et al., 2000). Populations tested at F 2 did not tend to differ from laboratory populations in terms TA B L E 2 Effective population sizes (N e ) of Aedes aegypti F 13 laboratory populations maintained at different census sizes, calculated using three temporal methods A limitation of our experiments is that we assessed the effects of inbreeding and laboratory adaptation under laboratory conditions.
High fitness under these conditions does not necessarily indicate high fitness in the field (Kristensen, Loeschcke, & Hoffmann, 2007;Thomson & Hoffmann, 2002); therefore, the apparent lack of laboratory adaptation observed here might not translate to the field where conditions are more complex. Several factors may also confound the results of our experiments. Our main comparisons were between populations at F 4/5 and F 13 ; if substantial laboratory adaptation occurs then we would be unable to detect it with these comparisons. Our population comparisons could also be confounded by selection on the ancestral population due to eggs experiencing quiescence (Townsville and Innisfail populations) or differences present in populations collected from the same location but at different times (Cairns populations). Other factors such as gut microbiota could also confound our comparisons between laboratory and field populations because the microbiome can greatly influence mosquito life history traits (Coon, Brown, & Strand, 2016;Coon, Vogel, Brown, & Strand, 2014). Gut microbiota are much less diverse in colonized mosquitoes (Mwadondo, Ghilamicael, Alakonya, & Kasili, 2017) and tend to be similar in laboratory populations regardless of geographic origin (Dickson et al., 2017). This could be an issue when comparing field and laboratory populations.
Few studies on laboratory adaptation in insects attempt to separate the effects of laboratory adaptation from drift or founder effects (Hoffmann et al., 2001; is one exception) which are likely to be substantial when establishing small laboratory colonies. We used replicate populations to avoid this issue; consistent divergence in colonized populations from the ancestral population indicates adaptation, variation between replicate populations immediately after establishment indicates founder effects, and divergence between replicate populations at the time indicates drift. We found that replicate populations at the same census size differ significantly from each other for several fitness traits, both at F 5 and at F 13 , particularly for populations maintained at low census sizes. Fitness differences between replicate populations were not always consistent between F 5 and F 13 , suggesting that both founder effects and drift occur. These findings are of concern for laboratory studies that compare traits between populations maintained separately. Researchers should consider using replicate populations when conducting experiments or outcross populations frequently to maintain similar genetic backgrounds (Yeap et al., 2011).
We demonstrate that inbreeding is extremely costly to Ae. aegypti fitness. Most inbred lines were lost across the experiment, and the remaining lines performed substantially worse than outbred populations. Relatively few studies have specifically addressed the effects of inbreeding on mosquito fitness. Powell and Evans (2016) observed that inbreeding Ae. aegypti through full-sib mating reduces heterozygosity by much less than expected based on theory, and deleterious recessive alleles must therefore be common. Koenraadt, Kormaksson, and Harrington (2010) reported fitness costs of inbred Ae. aegypti larvae relative to a wild population, and inbreeding through full-sib mating reduces fitness in other Aedes species (Armbruster, Hutchinson, & Linvell, 2000;O'Donnell & Armbruster, 2010). We demonstrate that mosquito populations inbred intentionally, for instance, to generate homozygous transgenic strains (Catteruccia, Godfray, & Crisanti, 2003;Phuc et al., 2007), will likely suffer from severe inbreeding depression. However, it may be possible to retain partial fitness if there is also selection for certain life history traits during inbreeding (Shetty et al., 2016). We show that laboratory populations maintained at low census sizes (N = 100) also experience inbreeding depression, and the loss of fitness correlates strongly with decreased effective population size. Thus, laboratories should ensure that population sizes in colonized mosquitoes are sufficiently high to maintain their fitness. Our laboratory populations for these experiments were each established from only a few hundred individuals, and we would recommend that larger numbers be used to avoid bottlenecks.
Our laboratory populations at F 13 had a substantially lower N e than field populations from Townsville (Endersby et al., 2011) and other locations around the world (Saarman et al., 2017). However, ratios of N e to census size (N e /N) were similar to ratios reported in nature (Saarman et al., 2017). N e /N ratios were larger in the small laboratory populations (N = 100) than in the large ones (N = 400), consistent with a study of Drosophila populations maintained at different census sizes (Schou, Loeschcke, Bechsgaard, Schlotterer, & Kristensen, 2017). Low N e /N ratios indicate that reproductive success varies greatly between individuals (Hedrick, 2005;Nunney, 1995), and this appears to be the case for large colonized populations of Ae. aegypti. Unequal contributions to the next generation occur because we sample only a few hundred individuals randomly from a pool of thousands of larvae, and we do not equalize offspring from each female to establish the next generation (Ross et al., 2017).
We demonstrate that the consequences of laboratory maintenance in Ae. aegypti can be minimized by maintaining large population sizes, but there are several other ways to maintain the fitness of colonized mosquito populations. The simplest approach is to cross laboratory colonies to an outbred population (Yeap et al., 2011).
Gene flow into inbred populations commonly leads to a fitness improvement (Frankham, 2015), and we also show that the fitness of inbred Ae. aegypti can be improved through a single generation of outcrossing. Increased performance of hybrids has been demonstrated in Anopheles mosquitoes (Baeshen et al., 2014;Ekechukwu et al., 2015;Menge et al., 2005) and the Queensland fruit fly (Gilchrist & Meats, 2014), with fitness improvements in the F 1 . Crosses between different laboratory lines can also be used to determine whether changes in fitness in laboratory populations are due to inbreeding or adaptation (Baeshen et al., 2014). Rates of laboratory adaptation can be slowed by using more natural rearing environments. Knop et al. (1987) compared two methods of rearing Culex tarsalis and found that colonies maintained in larger cages at a variable temperature and more complex environmental conditions had a slower rate of laboratory adaptation. Ng'habi et al. (2015) found that rearing Anopheles arabiensis under semi-field conditions preserved their similarity to the wild population and reduced the extent of inbreeding. Quality control methods such as screening mosquitoes for their flight capacity can also be used to increase fitness before their deployment for disease control programs (Balestrino, Puggioli, Carrieri, Bouyer, & Bellini, 2017).
In summary, we provide evidence for inbreeding depression effects and a small effective population size relative to census size in laboratory mosquito populations, along with some limited laboratory adaptation particularly in large populations. Our results have implications for the maintenance of insects in the laboratory, particularly for those destined for open field releases. While we find that life history traits of Ae. aegypti do not change consistently with laboratory maintenance, traits where selective pressures are absent in the laboratory, such as flight ability, feeding behavior, and thermal tolerance, might still be compromised.

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

DATA ACCE SS I B I LIT Y
Data for this study are available at the Dryad Digital Repository: https://doi.org/10.5061/dryad.84q8c68.