Selection on a behaviour‐related gene during the first stages of the biological invasion pathway

Human‐induced biological invasions are common worldwide and often have negative impacts on wildlife and human societies. Several studies have shown evidence for selection on invaders after introduction to the new range. However, selective processes already acting prior to introduction have been largely neglected. Here, we tested whether such early selection acts on known behaviour‐related gene variants in the yellow‐crowned bishop (Euplectes afer), a pet‐traded African songbird. We tested for nonrandom allele frequency changes after trapping, acclimation and survival in captivity. We also compared the native source population with two independent invasive populations. Allele frequencies of two SNPs in the dopamine receptor D4 (DRD4) gene—known to be linked to behavioural activity in response to novelty in this species—significantly changed over all early invasion stages. They also differed between the African native population and the two invading European populations. The two‐locus genotype associated with reduced activity declined consistently, but strongest at the trapping stage. Overall genetic diversity did not substantially decrease, and there is little evidence for new alleles in the introduced populations, indicating that selection at the DRD4 gene predominantly worked on the standing genetic variation already present in the native population. Our study demonstrates selection on a behaviour‐related gene during the first stages of a biological invasion. Thus, pre‐establishment stages of a biological invasion do not only determine the number of propagules that are introduced (their quantity), but also their phenotypic and genetic characteristics (their quality).

considers a variety of aspects related to invasions, including factors associated with their success, as well as assessments and intense, controversial discussions of their impacts (Ricciardi et al., 2017). Previous studies have led to a better understanding of the ecology and evolution of invasive species, with knowledge that can be applied to their management. Although these studies have made progress in predicting which factors enhance invasion success and which species may successfully establish and spread in the new area, much of the variability in invasion potential remains unexplained (Hayes & Barry, 2008). This may partly be due to the focus on species characteristics, even when substantial variation in invasion potential can be found among populations and can be expected among individuals of the same species (Cardador, Carrete, Gallardo, & Tella, 2016;Edelaar et al., 2015;Ochocki & Miller 2017).
The invasion process is typically divided into distinct stages, namely uptake (entering transport, including deliberate trapping), transport (including captivity), introduction (including escape), establishment and spread (Blackburn et al., 2011). Recently, it has been hypothesized that phenotypes can be selectively "filtered" while passing through the early stages of the invasion process (Carrete et al., 2012;Chapple, Simmonds, & Wong, 2012). If so, the characteristics of the introduced individuals may be different from those of the native donor population, which could promote or decrease invasion potential and impacts. While some studies have paid attention to selection acting on establishing and spreading populations (i.e., the final invasion stages; Bock et al., 2015), selection during the preestablishment invasion stages has been neglected. This is surprising, because (i) pre-establishment selection might be severe, as suggested for example by the high mortality rates between catching and export for wild-caught birds in the pet trade (7%-62%; Thomsen, Edwards, & Mullikan, 1992), and (ii) pre-establishment selection is important, because any variation that is removed in an earlier stage will no longer be present and exposable to selection in later ones. A good understanding of the selective processes acting during the early stages of the invasion pathway hence may be a key issue to assess invasion potential and impact.
Nonetheless, we are not aware of any empirical study dealing with pre-establishment selection during the invasion process. Such selection seems highly plausible given that individuals with certain behavioural, physiological or morphological traits might be more likely to be caught, to survive transport and captivity or to escape or be released (Carrete et al., 2012;Chapple et al., 2012). For example, it has been shown that variation in risk-taking behaviour causes sampling bias in wild animals (Biro, 2013;Biro & Dingemanse, 2009;Stuber et al., 2013) and relates to the exploration of novel food sources (Sol, Griffin, Bartomeus, & Boyce, 2011). In addition, in many species -including invasive ones-other behavioural traits affecting invasion potential and impact, such as neophobia, aggression, sociability and dispersal, are often linked to risk-taking behaviour (Cote, Fogarty, Weinersmith, Brodin, & Sih, 2010;Duckworth & Badyaev, 2007;Reale, Reader, Sol, McDougall, & Dingemanse, 2007).
Here, we investigate pre-establishment selection in an invasive bird, the yellow-crowned bishop (Euplectes afer). This songbird naturally occurs across wide regions of sub-Saharan Africa, but has recently and independently established populations in the USA, Venezuela, Jamaica, Puerto Rico, Japan, Italy, Portugal and Spain after escape or release of captive birds (Bird Life International 2016; Lever, 2005). Nowadays, the wildlife pet trade is a major source of biological invasion among vertebrates, in particular birds (Abell an, Carrete, Anad on, Dyer et al., 2017;Su, Cassey, & Blackburn, 2016). Specifically, we studied pre-establishment selection on genes that are related to invasion-relevant behaviours such as novelty seeking, activity and harm avoidance. In birds, primary candidates are the dopamine receptor D4 gene (DRD4) and the serotonin transporter gene (SERT, SLC6A4;Fidler et al., 2007;Korsten et al., 2010;Mueller, Korsten, et al., 2013;Mueller, Partecke, Hatchwell, Gaston, & Evans, 2013). Indeed, we have previously identified two SNPs in the DRD4 gene (SNP449 and SNP698, hereafter called candidate SNPs) that had strong and replicated effects on activity after exposure to a novel object in individuals from two invasive populations of the yellow-crowned bishop (Mueller et al., 2014). Hence, we test for frequency changes of these two behaviour-related DRD4 variants during the invasion process, assuming that these behaviours affect the probability that an individual will be caught and survive in captivity. Heterozygosity at a microsatellite in the second candidate gene SERT correlated with flight-initiation distance in dunnocks (Prunella modularis; Holtmann et al., 2016). SERT heterozygosity was also higher in blackbirds (Turdus merula) from recently colonized urban populations compared to those from the original forest habitat (Mueller, Partecke, et al., 2013).
Any observed allelic shifts can either be signals of selection, or they can be due to neutral random processes, such as genetic drift (Bock et al., 2015). Genetic drift due to small founding population size has the potential to decrease standing genetic diversity in invading populations relative to native populations, but evidence for the importance of this effect in invasions is mixed (Dlugosch, Anderson, Braasch, Cang, & Gillette, 2015). Hence, we first test whether there is an overall loss of genetic diversity between the population of origin (Senegal, SEN) and two introduced populations of E. afer from Spain (SPA) and Portugal (POR). Second, to assess the hypothesis that selection already acts during the early invasion stages (Carrete et al., 2012), we test whether the DRD4 candidate SNPs significantly change their frequency along early stages of the invasion pathway (relative to other markers). To test for selection during uptake, we compare allele frequencies among individuals caught by the traditional trapping methods used by bird exporters (potentially selective given that trapping involves baiting with food and decoy birds, referred to as the TRAP sample) and individuals caught with presumably less-selective mistnets (SEN sample). To test for selection during initial acclimation to captivity, we compare allele frequencies among individuals that successfully acclimated to captivity (ACCL yes ) and those that died (ACCL no ). To test for further selection during longterm captivity in storage cages, we compare allele frequencies among individuals that survived captivity (SURV yes ) and those that did not (SURV no ). We test for absolute allele frequency changes because we have no clear expectation about the direction of change.
Third, to assess the possibility that early selection (if any) left a genetic signature that is still noticeable after introduction, establishment and spread, we test whether allele frequencies at the DRD4 candidate SNPs differ between the native (SEN) and the two introduced populations (SPA, POR) in a consistent manner, and if so, whether the change is in the same direction as the allele frequency changes observed during the first stages of the invasion pathway.
Fourth, we test whether heterozygosity at the SERT candidate locus changes along the filter steps and whether it is higher in the introduced populations (SPA, POR) than in the native one (SEN). To these ends, we genotyped 335 individuals for nine random microsatellites, the SERT candidate microsatellite, and 31 DRD4 SNPs including the two candidate SNPs previously found to associate with activity in the two invasive populations. According to the Senegalese bird export company and the CITES trade data (Sanz-Aguilar, Carrete, Edelaar, Potti, & Tella, 2015), this is the same area where this species has been caught for export to Spain and Portugal. All individuals were marked (to avoid resampling of the same individual), blood-sampled (Fig. S1c) from the brachial vein (10-30 ll) and released in situ. Mistnetting is a sampling method that is presumably the least biased with respect to behavioural traits. There are few studies on sampling bias using mistnetting, but Simons, Winney, Nakagawa, Burke, and Schroeder (2015) did not detect any bias in mistnet-caught birds for their fully monitored island population of house sparrows (Passer domesticus). We therefore considered our sample of mistnetted Senegalese birds as the reference for the native population.

| Sampling of individuals for the bird trade and follow-up during the first invasion stages
We studied potential selection during three stages of the original invasion pathway via the international exotic bird trade. This involved sampling of birds caught by the Senegalese bird trappers and monitoring the fate of these individuals between trapping and international export, usually 1-3 months later. In stage 1, we accompanied professional local bird trappers working for the Senegalese company that historically exported E. afer to Europe and currently to other continents. Between 6 and 13 September 2014, they caught individuals using a traditional clap net baited with seeds and stuffed decoys to attract birds (Fig. S1d,e) in the same area as described for the reference sample (SEN) above. We took blood samples from all these individuals and marked them with uniquely numbered plastic rings. We genotyped a random subset (approximately one-third) of all captured/blood-sampled birds. A first invasion filter of selective uptake can be assessed by comparing these genotyped, traditionally caught birds (TRAP, N = 144) with those caught using mistnets ("trapping" or TRAP-SEN comparison).
In stage 2, we monitored the early survival of these trapped individuals. All individuals were kept at high densities for one week in traditional storage cages (Fig. S1f,g) close to the trapping sites and were then transported 350 km in the same cages (Fig. S1h) to the installations of the bird-trading company in Dakar (about 7-hr driving on the roof of a bus). Therefore, a second invasion filter where selection could take place was a 14-to 18-day period during which individuals either acclimated successfully to entry in captivity and transport (ACCL yes , N = 99) or died (ACCL no , N = 44; one individual was excluded because it lost its ring). Such mortality soon after capture has been documented before (Thomsen et al., 1992) and might select for certain behavioural types. We thus compared the genotypes of the surviving and nonsurviving birds ("acclimation" or ACCL yes À ACCL no comparison).
In the last stage prior to export, the remaining birds were communally kept in storage cages (Fig. S1f) for 3 months. Thus, a third invasion filter during which selection was evaluated was this longerterm survival in captivity. Because of its long duration and as most birds had died at the end of this period, we split this period in early mortality/survival (survival in the first 30 days, SURV1) and late mortality/survival (survival in the next 60 days, SURV2). We assessed selection by comparing the genotypes of individuals that survived with those that died (SURV1 yes , N = 54 vs. SURV1 no , N = 45; SUR-V2 yes , N = 11 vs. SURV2 no , N = 43). Given that the conditions during these two periods were largely the same and to increase statistical power, we then averaged the allele frequency shifts and changes in genetic diversity during these two periods to represent a single invasion filter of long-term survival in captivity. Information about allele names, whether the SNP is synonymous or nonsynonymous, or in an intron or exon (coding status), and major allele frequencies are given in Table S1. Among the 31 SNPs, twelve showed a minor allele frequency > 5% in one of the samples (SEN, TRAP, SPA, POR). Estimated allelic correlations between DRD4 SNPs are generally weak with most r 2 values below 0.5; the average r 2 between the candidates SNP449 and SNP698 was 0.14 (Mueller et al., 2014).

| Genotyping
We genotyped a microsatellite that is either in exon 1 or in the promoter of the SERT homologue (exact location unknown in this species) using the primers Sert_Ex1_F2 ATCTCCACACATTYCCCAGA and Sert_Ex1_R2 AGGAACCCTAAATCTGCCCTAC (see Mueller, Partecke, et al., 2013).
To assess population structure, genetic diversity and genetic  To assess population structure, we applied exact tests for allelic differentiation using Genepop (Rousset, 2008). We visualized population structure with a discriminant analysis of principal components (DAPC) using the R package adegenet (Jombart, 2008). DAPC first reduces allelic variance of all loci across all individuals (a total of 195 alleles) to the main principal components (we used 50 components explaining 88% of the total variance) and then uses these principal components in discriminant functions to maximize between-group variance while minimizing within-group variance (Jombart, Devillard, & Balloux, 2010). We explored potential genetic substructuring within the populations using the program STRUCTURE with default settings of the underlying model, that is, allowing for admixtured individuals and correlated allele frequencies between genetic clusters (Pritchard, Stephens, & Donnelly, 2000). The web tool STRUCTURE HARVESTER was used to combine the STRUCTURE output of 10 independent runs (Earl & von Holdt, 2012). We also tested for inflated genetic relatedness within the samples SEN, TRAP, SPA and POR by calculating all pairwise maximum-likelihood estimates of relatedness (Milligan, 2003) using the R package related (Pew, Muir, Wang, & Frasier, 2015). We compared the mean and distribution of all these values with the correspondent means and distributions of 1,000 random samples of simulated unrelated individuals while maintaining observed allele frequencies and sample sizes.

| Data analyses
We also tested whether mean relatedness among the surviving individuals of ACCL yes , SURV1 yes and SURV2 yes is higher in comparison with the traditionally caught birds (TRAP). Here, pairwise relatedness was calculated using the allele frequencies of the TRAP sample as reference. The first test assesses the potential confounding influence of relatedness structure for all samples, whereas the second test evaluates whether surviving individuals tended to be more related. Both effects could lead to nonrandom changes in allele frequencies across all loci.
We calculated allele frequencies and genetic diversity (expected heterozygosity) for each population and filter group using the R packages hierfstat and adegenet (Goudet & Jombart, 2015;Jombart, 2008). Individuals were randomly permuted between groups to obtain a null distribution for testing differences in heterozygosity.
For each invasive-native population comparison and for each filter stage, we calculated changes (delta values) in major allele frequencies such that a positive value indicates an increase and a negative value a decrease along the introduction process: SPA À SEN (Spain minus Senegal); POR À SEN (Portugal minus Senegal); TRAP À SEN (traditionally trapped minus mistnetted); ACCL yes À ACCL no (surviving acclimation minus nonsurvivors); SURV yes À SURV no (surviving captivity minus nonsurvivors). Similar delta values were calculated for genetic diversity changes.
For each of the three filter stages (trapping, initial acclimation and longer-term survival) and across all three stages combined, and for each marker, we used a permutation procedure to estimate the likelihood of the observed (or more extreme) absolute allele frequency changes (irrespective of increase or decrease). The group affiliation of each individual (e.g., TRAP or SEN when assessing the trapping filter) was randomly permuted against the genotypes within each comparison and new delta values were computed; this was repeated 10,000 times. This procedure simulates the random assortment of individuals into the contrasting groups of a specific filter stage (traditionally trapped versus mistnetted or survivors versus nonsurvivors). Similar permutation tests were performed for genetic diversity changes across all filter stages and for compar- in the previous association study (Mueller et al., 2014)  Also, the distributions of the observed relatedness values were similar to those of the simulated relatedness values (Fig. S4). Mean relatedness in the surviving filter groups ACCLyes (0.054), SURV1yes (0.056) and SURV2yes (0.076) did not increase more than expected under random subsampling (all p > .1). In addition, there were no obvious clusters of genetically related individuals within populations (discriminant analysis, Fig. S2). We thus conclude that it is unlikely that our tests for nonrandom allele frequency shifts among the filter and invasive-native comparisons are confounded by population or relatedness substructuring.

| Changes in genetic diversity during different invasion stages
Overall, genetic diversity did not decrease during the first stages of the invasion pathway (from SEN to TRAP, TRAP to ACCL yes , to SUR-V1 yes and to SURV2 yes; note that survival was assessed at two stages, whereby statistics were averaged because the final surviving group was small; see Section 2). Expected heterozygosity estimates  Figure 1 shows the allele frequency changes of the major alleles of all loci for each invasion stage, that is, during trapping (TRAP vs SEN), acclimation to captivity (ACCL yes vs ACCL no ) and survival in captivity (SURV yes vs SURV no ). A few loci showed significant changes in single contrasts, but there were only two loci (DRD4 SNP449 and SNP698) showing repeated allele frequency shifts in the top 10% along two or all three filter stages. The significance of the absolute frequency shifts (irrespective of direction) was evaluated for all major alleles using a permutation procedure (see Section 2) that controls for sample size and major allele frequency. Each of the two DRD4 candidate SNPs changed its frequency across the three filter comparisons more than expected by chance (SNP449: p = .0018, SNP698: p = .0018; Figure 2). The likelihood that the observed extreme frequency shifts occurred in both candidate SNPs together by chance was very low (p = 1.6 9 10 À5 ). Both candidate SNPs were also among the four table-wide significant markers after adopting a Bonferroni correction for the number of genomic regions tested or for the effective number of independent polymorphic markers tested (Figure 2). The other table-wide significant loci were another DRD4 marker from the genic region associated with activity (SNP458; Mueller et al., 2014) and a random microsatellite marker (WBSW7). However, these other two markers showed an extreme allele frequency shift in only one filter comparison. In separate analyses of each filter stage, SNP449 and/or SNP698 were always among the loci with the strongest frequency shifts, although not always significant due to lower power (Fig. S7). Both in females and in males SNP449 or SNP698 was among the loci with strongest frequency shifts, indicating that both sexes contribute to the overall effects ( Fig. S8). In single filter comparisons, the frequency changes of each SNP mostly follow the same direction in females and males, but the effect strengths might differ among the sexes. This needs further evaluation given the small sample sizes for each sex. SNP449 and SNP698 still belonged to the top 21% of polymorphic markers with the most extreme frequency shifts (Figure 4). The likelihood that the observed allele frequency changes in SNP449 and SNP698 between the native and the invasive samples were due to chance was p = .021 and p = .012, respectively (permutation test).

| Allele frequency shifts during different invasion stages
The likelihood of obtaining the extreme allele shifts in both candidate SNPs together by chance was very low (p = .0009). As expected given the smaller sample sizes, the sex-specific analyses mostly show nonsignificant allele frequency changes at the two candidate SNPs (Fig. S9). Figures 2 and 4  F I G U R E 1 Frequency changes of the major alleles in all polymorphic loci (subset of 31 DRD4 SNPs, and one SERT and nine random microsatellites) for the comparisons of (a) the trapping filter (TRAP À SEN), (b) the acclimation filter (ACCL yes À ACCL no ) and (c) the survival filter (SURV yes À SURV no ). For the latter, we used averages over two comparisons (SURV1 and SURV2) evaluated at two time points, because the final surviving group was small (see Section 2 for details). The 5% and 95% percentiles are indicated as dotted lines. All markers with changes more extreme than these percentiles are labelled, and the two candidate loci DRD4 SNP449 and SNP698 are marked in red

| Changes in DRD4 SNP genotype combinations during different invasion stages
We now consider five categories of SNP449-SNP698 genotype combinations that likely differ in additive expression of activity (high, medium high, intermediate, medium low, low; see Section 2). The most significant absolute change along the three filter contrasts (trapping, acclimation, long-term survival) was in the low-activity genotype combination (permutation test: p = .019). The frequency of the low-activity genotype decreased strongly in the first invasion stage (TRAP-SEN), with smaller changes in the following invasion stages ( Figure 5).
The Spanish and Portuguese populations also showed a reduced frequency of the low-activity genotype in comparison with the Senegalese sample ( Figure 5). Indeed, for the two invasive-native population comparisons combined (SPA-SEN and POR-SEN), the frequency of the genotype combination with low activity showed the largest difference (p = .041). The medium-high-and medium-low-activity genotype combinations also significantly changed frequency along the invasion stages (p = .036 and p = .030, respectively), but their frequency did not differ between the native and invasive populations (p = .19 and p = .29, respectively). The high and intermediate activity genotype combinations did not show consistent changes, neither for the invasion stages (p = .15 and p = .08, respectively), nor for the native-invasive population comparisons (p = .10 and p = .23).

| DISCUSSION
We analysed allelic changes in behaviour-related genes as well as presumably neutral microsatellite loci during the earliest stages of a F I G U R E 3 Frequency changes of the major alleles in all polymorphic loci (subset of 31 DRD4 SNPs, and one SERT and nine random microsatellites) for the comparison between the Senegalese native population and (a) the Spanish invasive population (SPA À SEN) and (b) the Portuguese invasive population (POR À SEN). The 5% and 95% percentiles are indicated as dotted lines. All markers more extreme than these percentiles are labelled, and the two candidate loci DRD4 SNP449 and SNP698 are marked in red human-induced biological invasion (i.e., uptake and captivity before introduction) by a well-known biological invader (a pet-traded wild bird, see Abell an, Tella, Carrete, Cardador, & Anad on, 2017 for its invasion process in Spain and Portugal). Among all markers, the two candidate SNPs in the DRD4 gene were the only variants that showed consistently large, significant changes in allele frequency along two or more comparisons of selective filters (Figures 1 and 2).
Remarkably, these exact same two SNPs explained on average between 11% and 15% of the variation in activity and neophobic behaviour in two replicate invasive populations of this species   Mueller et al., 2014) for the Senegal reference population (SEN), the different filtered groups (trapped, acclimation survivors and captivity survivors) and the two invasive populations (Spain and Portugal). The frequency in SURV2 yes group is not plotted due to its small sample size (Mueller et al., 2014). Specifically, SNP449 which appears to be conserved among bird species, has a high functional potential (Mueller et al., 2014). This suggests that selection on behaviour acts already during the initial invasion stages, as proposed by Chapple et al. (2012) and Carrete et al. (2012). As far as we know, this is the first empirical test of pre-establishment selection. Whether pre-establishment selection is common in biological invasions remains to be seen, but this seems likely (Carrete et al., 2012;Chapple et al., 2012). In this system, there is also evidence for sex-or size-biased trapping (A. Baños-Villalba and P. Edelaar, unpublished data). In particular when mortality is high, as in our study (92%), there is potential for strong selection. The observation of significant allele frequency differences at the same two SNPs when comparing two invasive populations with the native population of origin (Figures 3 and 4) suggests that the effects of such pre-establishment selection might be long-lasting. Such selection could therefore potentially affect the probability of successful establishment (e.g., through the degree of behavioural adaptation to novel conditions), the further development of the invasive population (e.g., activity levels may play an important role in range expansion) and its impacts on other species. Hence, our results highlight the importance of studying selective processes during the first stages of a biological invasion, because these stages may not only determine the number of propagules that are introduced (quantity) but also their phenotypic and genetic characteristics (quality).
In the first invasion stage (the "uptake" stage), we observed a downward shift in the frequency of the combined DRD4 genotype associated with low activity in response to novel objects ( Figure 5). adaptive shifts has been suggested for great tits Parus major (Mueller, Korsten, et al., 2013) and humans (Ding et al., 2002;Wang et al., 2004).
Among the 24 SNPs detected in the same-sized native and/or invasive samples (SEN, SPA, POR), only one was unique to the invasive samples, whereas eight appeared only in the native sample and may have been lost in the invasive populations (Table S1). It has been shown that founder events more often lead to loss of rare alleles than to a decrease in heterozygosity (Greenbaum, Templeton, Zarmi, & Bar-David, 2014). This indicates that selection on the remaining standing allelic variation seems important here, which can lead to rapid adaptive shifts (Bock et al. 2015). New mutations, however, appear to play a minor role in the genetic changes of the DRD4 system of E. afer during invasion. Mueller et al. (2014) speculated that the observed strong association between the two DRD4 SNPs and activity-related behaviour in the introduced populations might be partly rooted in the invasion history of these populations. It can be argued that the power to detect genotypephenotype associations may increase as a result of allele frequency changes (a rare variant with a strong effect might become more common; e.g., Zoledziewska et al., 2015), because of changes in the genomic background (e.g., a general diversity loss may "free" additive genetic variation at epistatically interacting loci, i.e., release cryptic genetic variation; Dlugosch et al., 2015) or because of changes in the ecological environment during invasion (Dlugosch et al., 2015). We can exclude the first reason, because the two candidate SNPs already had high minor allele frequencies in the native population. However, our results indicate that a few neighbouring SNPs in the exonic DRD4 region were lost or changed frequency during the invasion process. This leaves potential for changes in the neighbouring interactive genetic environment (epistasis). Furthermore, genetic variants at other, more distant, loci-in particular rare large-effect alleles-could have changed their frequency and thus their interactive influence on the DRD4 variants (Dlugosch et al., 2015). Only large-scale genomewide genotypephenotype association studies in the native range of Euplectes afer would provide the necessary information.
Overall genetic diversity as measured by heterozygosity did not decrease significantly between the native and invasive populations, further supporting that the reported allele frequency changes in the common DRD4 SNPs are not a mere consequence of genetic drift.
Genetic diversity at SERT was only slightly, but significantly higher in the two invasive than in the native population (Figs S5 and S6). This is similar to findings from blackbird populations which invaded urban areas (Mueller, Partecke, et al., 2013). Although the higher diversity of SERT in E. afer was not exceptional in comparison with the other tested loci and needs to be verified in future studies, its direction is opposite to that expected by drift. Thus, the invasive populations might have experienced selective bias for rare variants with deviating serotonergic signalling characteristics, similar to urban blackbirds (Mueller, Partecke, et al., 2013). If so, selection would presumably take place during the later stages of the invasion pathway, because we did not obtain statistical support for selection on SERT variants during the first stages (Fig. S5a-c). Selection during later stages of the invasion might act via risk-taking behaviour: in dunnocks (Prunella modularis), heterozygous females had shorter flight-initiation distances than homozygous females (Holtmann et al., 2016). Interestingly, heterozygosity of the SERT microsatellite homologue was also higher in an invasive dunnock population (in New Zealand) than in the native British one, while all other tested markers showed the opposite pattern (Holtmann et al., 2016). This suggests a similar selection regime to the one in Euplectes afer.
In summary, this study provides the first empirical evidence for the operation of selection during the earliest, pre-establishment stages of biological invasions, in this case selection on genetic variation in behaviour. Some of these early selective changes appear maintained in two successful invasive populations, and the reduction in low-activity genotypes could conceivably have influenced invasion success and impact in the habitats where the birds were introduced (Carrete et al., 2012). Selection could also be important in unintentional introductions where nonrandom uptake and survival during transport (e.g., in ships, containers) also represent the first steps of the invasion process (Blackburn et al., 2011;Chapple et al., 2012).
Further exploration of this hypothesis is therefore necessary to better understand and effectively manage biological invasions and to gain insight into the evolution of behaviour and other traits in introduced populations.

DATA ACCESSIBILI TY
The data that support the findings of this study are available in the supplementary materials.