Candidate gene polymorphisms are linked to dispersive and migratory behaviour: Searching for a mechanism behind the “paradox of the great speciators”

The “paradox of the great speciators” has puzzled evolutionary biologists for over half a century. A great speciator requires excellent dispersal propensity to explain its occurrence on multiple islands, but reduced dispersal ability to explain its high number of subspecies. A rapid reduction in dispersal ability is often invoked to solve this apparent paradox, but a proximate mechanism has not been identified yet. Here, we explored the role of six genes linked to migration and animal personality differences (CREB1, CLOCK, ADCYAP1, NPAS2, DRD4, and SERT) in 20 South Pacific populations of silvereye (Zosterops lateralis) that range from highly sedentary to partially migratory, to determine if genetic variation is associated with dispersal propensity and migration. We detected genetic associations in three of the six genes: (i) in a partial migrant population, migrant individuals had longer microsatellite alleles at the CLOCK gene compared to resident individuals from the same population; (ii) CREB1 displayed longer average microsatellite allele lengths in recently colonized island populations (<200 years), compared to evolutionarily older populations. Bayesian broken stick regression models supported a reduction in CREB1 length with time since colonization; and (iii) like CREB1, DRD4 showed differences in polymorphisms between recent and old colonizations but a larger sample is needed to confirm. ADCYAP1, SERT, and NPAS2 were variable but that variation was not associated with dispersal propensity. The association of genetic variants at three genes with migration and dispersal ability in silvereyes provides the impetus for further exploration of genetic mechanisms underlying dispersal shifts, and the prospect of resolving a long‐running evolutionary paradox through a genetic lens.

Great speciators are those species that inhabit multiple islands reached by overwater dispersal events, suggesting excellent colonization ability, yet display multiple morphological subspecies (Diamond et al., 1976;Manthey et al., 2020;Moyle et al., 2009;Pedersen et al., 2018).This concept is applied to several subspeciesrich birds from Melanesia and the South Pacific, including representatives within the white-eyes, kingfishers, and cicadabirds (Mayr & Diamond, 2001).A common explanation to reconcile broad island distributions with phenotypically diverse taxa is that dispersal traits are strongly selected against in each newly established island population (Diamond, 1981).Selection may reflect the individual energetic cost of this behaviour relative to the benefits of remaining (Bonte et al., 2012;McNab, 2002) or the idea that once most islands are filled, the chance of successful overwater dispersal is diminished (Diamond, 1970).However, it is still unclear how these relatively rapid, post-colonization changes in dispersal propensity could arise.
Variation in dispersal propensity is largely determined by differences in a combination of morphological, physiological and behavioural traits (Matthysen, 2012).In birds, the types of morphological changes that indicate a reduced dispersal ability include a more rounded wing (as indicated by a lower hand-wing index) (Kipp, 1959) and a shift to a more graviportal body plan (reduced flight muscles and longer legs) (Wright et al., 2016).However, among the great speciators, the shift from dispersive to sedentary forms may primarily involve behavioural rather than morphological or physiological changes, at least initially (Diamond, 1981).This is termed "behavioural flightlessness", a reluctance to disperse (especially across water), despite the maintenance of normal wings and flight.Hence, focusing solely on morphological proxies for dispersal ability that take time to change risks missing an essential part of the early evolutionary process.
A useful but relatively unexplored framework to understand behavioural flightlessness and subsequent divergence considers the links between dispersive behaviours and personality traits (Ingley & Johnson, 2014).Dispersal (single movement to a new area) is a complex behavioural syndrome that co-varies with other personality traits: highly dispersive taxa often show increased levels of boldness, aggression and exploration (Botero-Delgadillo et al., 2019;Duckworth & Badyaev, 2007;Duckworth & Kruuk, 2009).For example, highly aggressive Western bluebird (Sialia mexicana) individuals were found to be more dispersive than their less aggressive counterparts (Duckworth & Badyaev, 2007;Duckworth & Kruuk, 2009); and dispersive great tit (Parus major) individuals had a greater exploration rate than non-dispersive individuals (Korsten et al., 2013).
Migration (seasonal movement) is also associated with personality in animals, in particular in those species that display partial migration (Nilsson et al., 2014) Migratory individuals within a species tend to be bolder than non-migratory ones (Chapman et al., 2011).
Both quantitative and molecular genetic studies provide evidence of a genetic influence on migratory and dispersive behaviours (Liedvogel et al., 2011).These behaviours often have significant narrow sense heritability, often greater than 0.4 (Dochtermann et al., 2019).Additionally, at the molecular level, genetic variation for migration, dispersive behaviour and personality traits are well documented (Bubac et al., 2020), including the identification of a number of candidate genes underlying personality traits such as tendency for boldness or exploratory behaviour (Canestrelli et al., 2016;Fidler et al., 2007;Ruegg et al., 2014;Steinmeyer et al., 2009), and migratory behaviour (Bazzi et al., 2015;Johnsen et al., 2007;Peterson et al., 2013;see Le Clercq et al., 2023 for an extensive review).
Changes in genetic variation at these candidate genes may act as a genetic switch catalysing population-level shifts in dispersive behaviour.
The Zosteropidae family (white-eyes, yuhinas and allies) consists of 142 species (Clements et al., 2019) many of which are highly dispersive as evidenced by colonization of numerous oceanic islands throughout the Indian and Pacific oceans, along with the broad continental distributions of some species (Clegg et al., 2002;Mees, 1969).This family shows one of the highest per-lineage diversification rates for vertebrates (Moyle et al., 2009) and divergence can occur even across minor geographic barriers (e.g.water gaps of just 2 km) (Bertrand et al., 2014;Cowles & Uy, 2019;Glor, 2011;Manthey et al., 2020;Mees, 1969).A particularly interesting species within this family is the silvereye (Zosterops lateralis), considered a great speciator having multiple subspecies (17 morphological subspecies; Clements et al., 2019) with a wide natural distribution-including the Australian mainland and Tasmania, the North and South Islands of Aotearoa New Zealand (henceforth New Zealand), outlying oceanic islands of Australia and New Zealand, and the archipelagos of New Caledonia, Vanuatu, and Fiji (Figure 1a).Silvereyes also display a variety of gene flow potentials: the Tasmanian subspecies (Z.l. lateralis) is a partial migrant (Mees, 1969); those in the central Vanuatu archipelago display high levels of outgoing gene flow, while more peripheral populations have high levels of incoming gene flow (Clegg & Phillimore, 2010); and others, such as on Heron Island and Lord Howe Island in the southern Great Barrier Reef, are sedentary and genetically isolated (Sendell-Price et al., 2020).
Here, we assess variation in six personality-related candidate genes in the silvereye at three different levels: individuals, populations and subspecies.Our aim is to determine whether there are signatures consistent with a genetic switch that would explain rapid shifts in dispersal, leading to reduced gene flow and subsequent increased opportunity for divergence.We ask (i) if candidate gene variation follows patterns of neutral genomic divergence, that  S3).The sampling included putative Tasmanian winter migrant individuals captured in the Australian mainland states of Queensland and New South Wales.Individuals were assigned as Tasmanian migrants based on plumage differences: resident silvereyes (subspecies Z. l. cornwalli) exhibit bright yellow throat and grey flanks, while migrant individuals (subspecies Z. l. lateralis) exhibit white-to-pale-yellow throat and dark buff flanks (Figure S1; Higgins et al., 2006).Birds were caught in mist-nets or traps and blood samples were taken via venipuncture of the brachial wing vein and stored in 90% ethanol or lysis buffer (10 mM EDTA pH 8.0, 10 mM TRIS-HCl pH 8.0, 20 mM NaCl, 1% SDS, Seutin et al., 1991).

| Candidate microsatellite genotyping
We extracted DNA from 422 samples using a standard chelex protocol (Walsh et al., 1991) and amplified four microsatellite candi- and NPAS2 [n = 375]), (Table S3) with the primers from Steinmeyer et al. (2009) using a touchdown polymerase chain reaction (PCR) protocol and fluorescent labelling (VIC and FAM) using M13 tags (Schuelke, 2000).Fluorescent labelling with VIC and FAM was done in multiplex PCR reactions containing two loci each (VIC: CLOCK and ADCYAP1; FAM: CREB1 and NPAS2); final reaction volume of 3 μL contained 10 ng DNA, 1 μL Type-it Master Mix (Qiagen), 0.17 μM of either FAM or VIC, and 0.01 μM forward M13-labelled primer and 0.05 μM reverse primer for each locus.
Thermal cycling consisted of an initial denaturation step of 95°C for 15 min, eight cycles of 94°C for 30 s, an annealing temperature of 60°C, reduced by 1°C each cycle, for 90 s and a 72°C extension step for 60 s, then 25 cycles of 94°C for 30 s, 52°C for 90 s, and a 72°C for 60 s, and a final extension of 60°C for 30 min.VIC and FAM multiplexes were pooled for each sample and allele sizes for the candidate gene microsatellites were determined in relation to LIZ600 size standard on an ABI 3730xl DNA Analyser and scored using the Microsatellite Plugin within Geneious 2020.1 (http://www.geneious.com, Kearse et al., 2012).We tested for the deviations from Hardy-Weinberg Equilibrium, allelic richness and heterozygosity using the R package diverRsity (Keenan et al., 2013).

| DRD4 and SERT sequencing
We targeted one region of DRD4 (n = 34), encompassing intron 2 and exon 3, and part of intron 3 (1480 bp) following Fidler et al. (2007).For SERT (n = 100), we amplified part of the promoter region (508 bp) following Holtmann et al. (2016).Each 25 μL PCR reaction was comprised of approximately 10 ng DNA template, 1.0 unit BioTaq (Bioline USA Inc.), 1× BioTaq reaction buffer, and 0.5 μM each forward and reverse primer, 200 μM dNTPs, 1.5 mM MgCl 2 , and made up to volume with Milli Q water.The PCR reaction profiles for both DRD4 and SERT fragments consisted of an initial denaturation step of 94°C for 3 min, followed by 10 cycles of 94°C for 3 min, a touchdown step at 65°C for 30 s, and an extension step of 72°C for 2 min.This was followed by 25 cycles of 94°C for 30 s, 55°C annealing step for 30 s, an extension step of 72°C for 2 min, and a final extension step of 72°C for 10 min.We purified PCR products using Acroprep 96 filter plates (Pall Corporation) following the manufacturer's protocol, then Sanger sequenced with forward and reverse primers on an ABI 3730xl DNA Analyser (Genetics Analysis Service, Otago University, NZ).We aligned the sequences using MUSCLE (Edgar, 2004) in Geneious version 2020.1 and identified SNPs and insertions/deletions (indels).For DRD4, we aligned sequences to the great tit DRD4 gene sequence (GenBank accession no.: DQ006801.1;Fidler et al., 2007) and for SERT, we used the dunnock (Prunella modularis) sequence (Genbank accession no.: KT967954.1;Holtmann et al., 2016).We called SNPs with a minimum minor allele frequency (MAF) of 0.05.

| Gene flow estimates
As a proxy of population dispersiveness, we estimated outgoing gene flow rates from each population.We used whole genome sequences published by Estandía et al. (2023) representing 336 of the 422 individuals included in the present study to examine population structure patterns in NGSadmix (Skotte et al., 2013), a module implemented in ANGSD (Korneliussen et al., 2014).All populations are represented in the whole-genome dataset except Erromango.We obtained a BEAGLE file containing genotype likelihoods and created a subset of 10 000 unlinked SNPs picked at random after applying filtering for a MAF of 0.05.We ran NGSadmix with a range of genetic clusters (k), from 2 to 20.We selected the best k for our dataset based on the mean estimate likelihoods, which indicated k = 2 as the optimal number of clusters, corresponding to ANZO and SM groupings.We re-ran NGSAdmix in each of the main cluster to explore potential substructuring.We visualized clustering patterns using a custom R script.
We generated a covariance matrix in PCAngsd (Meisner & Albrechtsen, 2018) using the BEAGLE file.Because not all individuals were screened for all candidate genes (e.g.375 individuals for NPAS2 but 258 for CLOCK), and not all individuals in the candidate gene and whole-genome dataset coincide, we produced a population-level covariance matrix.From this, we calculated contemporary rates of gene flow among Z. lateralis populations for each of the two clusters (ANZO and SM) separately using BA3-SNP (Mussmann et al., 2019), an extension of BayesAss that allows SNPs as input.To ensure that our gene flow estimations were not a biased product of the SNP selection, we created five independent datasets of randomly selected 10 000 SNP as input.BayesAss employs a Bayesian approach with Markov chain Monte-Carlo (MCMC) sampling to estimate migration rates defined as the proportion of immigrants from the donnor population to the sink population.We adjusted delta values for migration rates (m), allele frequencies (a) and inbreeding coefficients (f) using Autotune (Mussmann et al., 2019) to ensure that parameter space sampling acceptance rate was between 20% and 60% (Wilson & Rannala, 2003).We ran the program for half a million iterations, discarding the first 10% as burn-in.We estimated the 95% credible sets by calculating the mean ± 1.96 × Standard deviation (SD).
To assess model convergence, we compared results from 10 replicate runs each with a different random starting seed.We considered runs to have converged on a similar solution if gene flow estimates were within 0.005 (0.5%) across runs.We averaged the parameter estimates of the runs for subsequent analyses.
We explored population structure within the partial migrant population by doing a genomic PCA with PCAngsd and decomposed the covariance matrix into eigenvalues.We plotted the two first PCs to visually explore any emerging structure.

| Candidate gene association analysis
For all candidate genes, we assessed variation among populations using the R package brms (Bürkner, 2017).Individual mean microsatellite length was set as the response variable and population was set as a categorical explanatory variable.In the case of SERT and DRD4, the response variable was coded as 0 or 1 representing whether the individual carried the most frequent nucleotide or the variant, and we used a Bernoulli family for binary outcomes.For the two microsatellite candidate genes that showed obvious within and between population variation in length (ADCYAP1 and CREB1) we tested whether population age and dispersal propensity could explain this variation by running Bayesian linear mixed models in brms.CREB1 showed a clear distinction in allele lengths between a grouping consisting of the Australian mainland, Tasmania and recently colonized islands versus old island populations.Because of this structure (that does not completely align with ANZO and SM neutral structure groupings), we also applied a broken stick regression model for CREB1 only to test whether including a single change point would improve predictive performance over a null model and a linear model using mcp (multiple point change) (Lindeløv, 2020).
NPAS2 and CLOCK displayed little among population variation in average allele length, hence we did not apply Bayesian linear mixed models to the whole population set.However, CLOCK showed variation within the partial migrant population of Tasmania, so we tested whether individuals that migrated to the mainland had different CLOCK lengths to those that over-wintered in Tasmania.The Tasmanian sample was restricted to 14 winter-caught birds (non-migrants) as the resident summer population includes a mix of migrants and non-migrants that cannot be phenotypically distinguished.The 'migrant' group consisted of those caught at Australian mainland sites in winter that were phenotypically identified as Tasmanian migrants (26 individuals).
For Bayesian linear mixed models (CREB1 and ADCYAP1) and broken-stick regression models (CREB1), the following populationlevel (fixed) parameters used were: 1. Dispersal Index (DI): the sum of each outgoing gene flow estimate (g) from island i into island j (where 95% credible interval did not overlap with zero), multiplied by the geographic distance between the islands (d) Equation (1).The latter helps to account for differences in geographic opportunity for dispersal; for example, a geographically isolated island population that has moderate outgoing gene flow to few far islands would score higher than a centrally located island population with moderate outgoing gene flow to many close islands.DI was scaled from 0 (nondispersive) to 1 (maximally dispersive).S2): population ages for New Zealand, Chatham Island and Norfolk Island are known from historical records (c.190 years; Clegg et al., 2002;Mees, 1969); for Heron Island, the population age was set as a maximum of 4000 years based on length of time the island has been vegetated (Clegg et al., 2008); and molecular estimates for the remaining ancient populations were taken from a time-calibrated gene tree (Black, 2010).

Age (Table
The null model represents the starting hypothesis that mean CREB1 or ADCYAP1 length differs between populations (which was set as group-level parameter or random effect) but does not change as a function of increasing dispersal propensity or time since colonization.We controlled for population structure using the PCAngsd population-level covariance matrix.As mcp does not permit incorporation of distance matrices, we used a categorical variable that classified each individual according to its membership in one of the two population genetic clusters (ANZO or SM) identified from the top-level NGSadmix analysis.
Our models used a Gaussian distribution and application of the default link function for all mcp (CREB1 only) and brms: We set weakly informative priors for the intercepts of the segments before and after the change point (mcp).The same prior was applied for both CREB1 length intercepts due to the lack of a priori information about a positive or negative correlation of mean CREB1 length with population age and dispersal index.Priors for both intercepts were centred at 550 bp with a SD of 20 (N(550, 20)), based on known mean and variance of CREB1 lengths in other passerines (Steinmeyer et al., 2009).For the brms linear mixed model, we set weakly informative priors with a normal distribution centred on 0 with a standard deviation of 20 for the DI and population age coefficients.This prior does not assume an increase or decrease in length, but incorporates the prior information that changes will not be greater than 20 bp given that CREB1 tend to vary less than 10 bp in other birds (Bazzi et al., 2017;Bourret & Garant, 2015).We performed prior predictive checks, where data was generated according to the specified prior predictive distributions in order to assess their suitability (Gabry et al., 2019).
Since dominance of the longer allele occurs for some candidate genes that show Poly-Q polymorphism (e.g.Saino et al., 2015), we analysed all the models with the longer allele as the response variable.S4).Other values of k also had high likelihoods (Table S4); k = 3 indicated sub-structuring within ANZO due to separation of Heron Island and Lord Howe Island from other populations (Figure S2; see also k = 2 in the ANZO analysis Table S4.1);k = 4 indicated sub-structuring within the SM cluster, primarily separating New Caledonia from Vanuatu populations, with the Vanuatu island of Tanna showing some affiliation with New Caledonia (Figure S2, see also k = 2 in the SM analysis Table S4.2).Increasing the value of k resulted in structuring based on single islands.These population genetic patterns were consistent with those that emerged from the covariance matrix (Table S5; Figure S3).
The independent BayesAss runs conducted to quantify the degree of migration rates agreed on the patterns of gene flow (Table S6).Of 42 pairwise comparisons within the ANZO cluster, and 132 in the SM cluster, nine and five respectively had estimates for which the credibility intervals did not overlap with zero.Within showing a split between ANZO and the SM populations.The average CREB1 allele length per population indicates a reduction in length for Heron and Lord Howe Islands, and all populations in the SM cluster.CREB1 was not screened for Ambae and Pentecost samples.
ANZO, this was primarily seen in relatively high outgoing gene flow estimates from Tasmania, and within SM, moderate outgoing levels from central islands of Pentecost and Espiritu Santo.
We did a genomic PCA with migrant and resident individuals and found that the two first PCs did not separate discrete clusters containing migrants and residents (Figure S4), suggesting a lack of general genome-wide divergence.CLOCK was monotypic in the majority of populations (Figure 3b; Table S7.showed some variation across populations but similar mean values for populations in the ANZO cluster, with the exception of Lord Howe Island (Figure 3d; Table S7.3).However, most SM populations

|
were not variable at this locus, with the exception of peripherally located islands of Gaua, Efate and Tanna in Vanuatu.Variation between and within populations for ADCYAP1 was evident for all populations (Figure 3c; Table S7.4).
We identified 10 DRD4 SNPs, seven of which were nonsynonymous and three synonymous.Five of them showed significant differences across populations but only SNP at position 83 (a non-synonymous substitution) displayed consistent differences between ANZO and SM clusters (Table S7

| DISCUSS ION
Rapid changes in dispersal propensity could help to explain why taxa present on many islands show high levels of phenotypic diversitythe paradox of the great speciators.Quantifying variation in six candidate genes thought to underlie dispersal and migratory behaviour in populations of Australian mainland and island-colonizing silvereyes has revealed three key results that hint at a role for allelic changes as a mechanism underlying this paradox.First, at the population level, we found that CREB1 allele length variation was associated with time since colonization and dispersal propensity, with recently colonized populations and those with high dispersal tendencies carrying long alleles.Second, at the individual level, we found allele length variation in CLOCK associated with migratory status when comparing Tasmanian residents (shorter) and Tasmanian migrants (longer).Third, more limited data from a single DRD4 SNP suggested that recently colonized populations carried guanines, evolutionarily older island populations carried adenines, and Tasmanian migrants and non-migrants carried both nucleotides.Together, these genes are likely to provide useful signatures of behavioural shifts in dispersal propensity in other silvereye populations, and possibly closely related species, though whether any of these genes act as a genetic switch remains unclear.
In the silvereye populations studied here, CREB1 showed a disjunct pattern of allele sizes: longer in the mainland, the partial migrant Tasmanian population, and all of the recently colonized populations and shorter in island populations colonized thousands, or hundreds of thousands of years ago.This cannot be explained by population genetic groupings, as both the Heron Island population (up to 4000 years old) and Lord Howe Island population (over 100 thousand years old) (Clegg et al., 2002;Sendell-Price et al., 2020) had shorter CREB1 allele lengths, yet clearly fell in the ANZO population genetic cluster that includes all of the long-allele-length populations.The Heron Island samples were monomorphic for a unique (551 bp) allele, and we cannot rule out that it may be fixed entirely due to drift in this small population (of ~300 breeding birds, McCallum et al., 2000).However, all of the older colonizations converge in a reduction in length, suggesting that it is unlikely that just by neutral processes each of these independent colonizations show a change in length in the same direction.While most of the island populations were not monomorphic, they showed a decrease in variance and a shift in allele lengths.Populations did not have a fixed genotype in young (dispersive genotype) versus old colonizations (sedentary genotype).This is because the colonizing flocks were likely a mix of dominant dispersive leaders and submissive individuals as the silvereye is a social species (Kikkawa, 1987).Natural selection could have acted onto this phenotypic variation and associated genotypes.Another mechanism by which the same phenotype could have arisen is through new mutations in each of the colonizations.However, this scenario is unlikely because standing variation is a more relevant mechanism driving parallel evolution at shallow timescales (Hedrick, 2013).
At the broad scale, the most dispersive populations, and those evolutionarily recent colonization events, carry longer allele lengths, compared to more sedentary, and more ancient island populations.This is the opposite pattern to that found in common buzzard (Buteo buteo) where non-dispersive individuals carried longer CREB1 allele lengths (Chakarov et al., 2013).CREB1 allele lengths have also been shown to correlate with other life-history traits that also depend on circadian and circannual clocks e.g.incubation duration (Bourret & Garant, 2015) and male moult speed (Bazzi et al., 2017).The broad scale patterns, and suggested association with dispersal, were not always reflected within each of the longer-and shorter-allele length groups.For instance, within SM, Gaua and Espiritu Santo displayed the longest lengths, but according to our gene flow analyses, Espiritu Santo was a dispersive population within its grouping but Gaua was not; likewise for Norfolk Island in the ANZO cluster.
While the obvious length variation pattern seen at CREB1 appears to have biological relevance for dispersal propensity in our study system, a mechanistic resolution of the paradox of the great speciators also requires dispersal shifts to be rapid (Diamond et al., 1976).We were unable to pinpoint when the CREB1 allele size that CLOCK could potentially be a gene under divergence.Previous studies in migratory birds focusing on CLOCK variation show mixed results.Longer allele lengths have been found to be positively associated with migratory propensity (Peterson et al., 2013), and variation in the phenology of migratory species (Bazzi et al., 2015;Caprioli et al., 2012;Liedvogel et al., 2009), but in some cases allele lengths are negatively correlated with migration date (Ralston et al., 2019) or not correlated at all (Contina et al., 2018;Mueller et al., 2011;Parody-Merino et al., 2019).
CLOCK plays a key role in regulating the circadian oscillator gene complex (Panda et al., 2002), and is associated with variation in the phenology of photoperiodic traits (e.g.migratory behaviour) (Table S1).For example, in migratory equatorial African stonechats annual-cycle timing is associated with CLOCK length (Justen et al., 2022).Increasing photoperiod can initiate migration (Assadi & Fraser, 2021;Robart et al., 2018) triggered only in migratory caged birds but not in resident ones, supporting a genetic link with migratory behaviour (Chan, 1994).Consequently, photoperiod changes could be the migration trigger with the onset of shorter autumnal days.The Tasmanian population is one of the few silvereye migrant populations situated where changes in day length during the non-breeding, winter period is substantially greater than changes experienced further north and closer to the equator.Another location where these conditions are met is New Zealand.In fact, even though records are few, large flocks of silvereyes seem to have migrated from the South to the North Island (Dennison et al., 1987), providing further support that photoperiod changes which are mediated at least partially by genetics, can lead to migration.Some overwater dispersal events might be a direct consequence of off-course migration.For example, the recent sequential colonization sequence of the silvereye from Tasmania to New Zealand and outlying islands was likely initiated by an off-course flock of Tasmanian migrants (Mees, 1969).However, Tasmanian migrants sampled in this study carry long CLOCK allele variants that were not recovered in the recently colonized populations.This could because we are not sampling the full allelic diversity of CLOCK in recently colonized populations (New Zealand (n = 6), Chatham Island (n = 13), Norfolk Island (n = 2); Table S3), and/or underlying effects of a different set of genes that we have not included in this study on dispersal ability in these populations.The genetic basis of partial migration in Tasmanian silvereyes and whether associated variants are present in recently colonized populations should be further explored in future work using a hypothesis-free genome-wide approach.
DRD4 is one of the most well-studied candidate genes related to exploratory and risk-taking behaviour (Bubac et al., 2020)-both of which could feasibly have links to dispersal propensity.We found a single polymorphism within DRD4 (SNP83) that showed fixed differences between two young (more dispersive) and seven old (more sedentary) island populations.The Tasmanian population was polymorphic (G and A represented), the recently established populations of Chatham Island and South Island New Zealand were fixed for A, and Southern Melanesian populations fixed for G. SNP83 is located in intron 2 and corresponds to base pair 9423 on the DRD4 orthologue of the great tit (Fidler et al., 2007).It has not been noted previously as having any phenotypic associations.Previous work assessing DRD4 variation and personality in the great tit focused on associations between exploratory behaviour and variation at 'SNP830', revealing large effects in certain populations but not in others (Fidler et al., 2007;Korsten et al., 2013;Riyahi et al., 2017); however, SNP830 was not variable in our dataset.Our data add to the evidence that variability in the DRD4 gene plays a role in a suite of behavioural phenotypes; however, the extent of importance of SNP83 will require screening more individuals and populations in silvereyes and other species.
Even though NPAS2, ACDYAP1 and SERT each showed some variability between populations this variability is not related to dispersal ability.Given the limitations of studying a handful of candidate genes for explaining complex behavioural phenotypes, failure to detect associations is not entirely unexpected.In birds, different associations (negative, positive or no correlation) in different populations, species and candidate genes are often reported (Table S1).Furthermore, few studies have considered candidate gene-environment interactions (but see Bourret & Garant, 2015;Liedvogel & Sheldon, 2010;Liedvogel et al., 2009) or methylation patterns that have been found to explain diverse complex behaviours in birds (Saino et al., 2019).
The lack of replicability even within the same species could also be a product of sampling in different locations with a low number of individuals leading to a lack of statistical power, choosing different variables and proxies to measure migration and dispersiveness, using different methodologies to analyse data, publication bias or a combination of all of these (Yang et al., 2022).Thus, the lack of association in NPAS2, ADCYAP1 and SERT does not rule out their potential role in dispersal behaviour as other factors might be masking their effects.

| Maintenance of high dispersal propensity in a continental island population
Despite being an old insular form (>200 K split from Australian mainland subspecies Black, 2010), the Tasmanian silvereye (Z.l. lateralis) has maintained high dispersal propensity: it is a partial winter migrant to mainland Australia, it was the original source population for the historical sequential colonization of New Zealand and outlying islands (Clegg et al., 2002;Mees, 1969), and as shown here, displays high levels of gene flow with Australian mainland subspecies.It also maintains the longer CREB1 average allele lengths, and putative Tasmanian migrants caught on the mainland have shown unusually long CLOCK mean allele lengths despite genome-wide lack of divergence.
The maintenance of high dispersal potential, and its partial migrant status, are most likely explained by its geography and history of connectivity with the mainland.Tasmania has repeatedly been connected to the Australian mainland during glacial periods and it is currently separated by a very shallow sea (Blom & Alsop, 1988).Over 50 islands of varying sizes can be found between the Australian mainland and Tasmania, which can act as migration stopovers, and facilitate connectivity (Belbin et al., 2021).

| The future of dispersal genomics
A candidate gene approach to understanding the paradox of the great speciators relies on knowledge of those genes in multiple systems.As discussed earlier, the candidate gene approach has many limitations and often shows conflicting results.Alternative hypothesis-free approaches, like genome-wide association studies (GWAS), partially overcome the obstacles imposed by the incomplete understanding of the mechanisms underlying complex behaviours.GWAS requires phenotypes, in this case dispersal propensity, to be characterized at the individual level.Tasmanian migrants were relatively easy to identify because of their different morphological differences with the mainland subspecies.However, assigning a dispersal score to individuals where morphology cannot be used to distinguish between dispersers and non-dispersers becomes challenging.Dispersal phenotypes could be assessed via use of individual tracking devices though the sample sizes for these types of studies are often smaller than required for GWAS in particular.
Even though our results suggest that migrants have genetic differences from non-migrant individuals, and that more dispersive populations have different CREB1 profiles to non-dispersive ones, a more thorough sampling involving a higher number of loci is necessary to explore whether standing genetic variation within a population or de novo mutations can provide the raw material for natural selection to act upon shifting a population to complete sedentariness.

| CON CLUS ION
In this study we assessed variation in six personality-related candidate genes in silvereye populations to examine whether signatures consistent with a genetic switch can explain rapid shifts in dispersal, leading to reduced gene flow and ultimately divergence in this great speciator.We find strong support for the idea that more dispersive populations carry longer CREB1 alleles, but length decreases with time and limited isolation, suggesting that selection could be acting against dispersal ability following island colonization.At the individual-level, partial migrants showed longer CLOCK alleles than non-partial migrant individuals in a population.Our results suggest that the paradox of the great speciators can be partially understood from a genetic perspective.
reflect drift and gene flow processes only; (ii) if candidate gene variation is associated with dispersal propensity among a range of dispersive and non-dispersive populations of silvereye, considering time since colonization; and (iii) if candidate gene variation is correlated with individual migratory status in the partially migratory Tasmanian population.F I G U R E 1 (a) Distribution range of the silvereye highlighted in blue.Red dots represent single islands or archipelagos where multiple islands have been colonized from the Australian mainland or Tasmania.Approximate colonization times are shown.French Polynesia is shown as inset; Locations of silvereye populations sampled for candidate gene variation: (b) ANZO cluster (c) SM populations.2 | ME THODS 2.1 | Sampling and DNA extraction Silvereye blood samples were collected between 1996 and 2016 from 20 sites in eastern Australia and New Zealand, including outlying islands, henceforth ANZO (Figure 1b), and Southern Melanesia (New Caledonia and Vanuatu) referred to henceforth as SM (Figure 1c; Table (i) Differences in length among populations for each candidate microsatellites (ii) Bayesian linear mixed models: 1. migration: CLOCK only 2. population age and DI: CREB1 and ADCYAP1 only.

For|
both the brms and mcp analyses, we used MCMC with four chains of 4000 iterations each, including a warm-up of 400 iterations.We evaluated convergence via visual inspection of the MCMC trace plots, checking that the ESS > 200 and the R values for each parameter (R = 1 at convergence).To evaluate model performance, we compared our fitted models with an null model using leave-oneout cross-validation (LOO), a robust, fully Bayesian model selection approach (Vehtari et al., 2017).Population genetic structure and gene flow NGSadmix analysis of WGS data supported two main genetic clusters (k = 2): Cluster 1 comprised Australia, New Zealand and outlying island populations (ANZO), and Cluster 2 comprised Vanuatu and New Caledonia populations in Southern Melanesia (SM) (Figure 2; Figure S2; Table 2).Compared to Tasmanian resident silvereyes, Tasmanian migrants showed longer allele lengths.Migrant individuals had long CLOCK variants (allele lengths of 289 and 291) not observed in any winter-caught Tasmanian birds (i.e.residents) (Figure 4a), or any other population except for a Heron Island individual.NPAS2 Mean CREB1 variation was better explained by a single change point model than the null model or a linear regression model F I G U R E 3 Individual mean allele lengths (shadowed dots) with mean and standard error for each population (large dots) for the four microsatellite candidate genes.Blue and grey dots correspond to two population genetic clusters identified from admixture analysis.(a) CREB1 is longer in the source populations (Australia and Tasmania), and recently colonized populations (New Zealand, Chatham Islands and Norfolk Island) compared to older populations (Heron Island, Lord Howe Island, and SM populations).(b) Migrant individuals highlighted in orange have longer allele lengths at CLOCK.(c) ADCYAP1 shows extensive variation, including shorter lengths in Lord Howe Island and Ambrym and longer lengths in New Zealand, Ouvea and Gaua.(d) NPAS2 shows little variation with southern Vanuatu populations (Erromango, Efate and Tanna) showing an increased mean allele length.(Figure S4; Tables S9.1-S9.2).Time since colonization and DI failed to explain the variability in ADCYAP1 (Tables S10.1-S10.2).All models performed with the longer allele yield the same results.The single change point model indicated that mean CREB1 length decreased six base pairs with increasing population age, however the timing of this change had high uncertainty (Intercept 1 (before change point) = 552.462,Intercept 2 (after change point) = 557.494;Table S10.1; Figure 5a).The posterior probability density of change point ranged between 200 and 4000 years ago when we see the step reduction in allele length in the Heron Island population (maximum age 4000 years, Clegg et al., 2008), a reduction that is observed in other older populations (Lord Howe Island and Southern Melanesian populations).The exception is Tasmania, an evolutionarily old population with long average CREB1 length.CREB1 decreased five base pairs with decreasing DI (Intercept 1 = 551.22,Intercept 2 = 556.68;Table S10.2; Figure 5b).The posterior probability density of change point for DI ranged from Chatham Island to Espiritu Santo, an island in Vanuatu that shows low levels of outgoing gene flow into nearby islands.
Individual CLOCK lengths for Tasmanian residents and Tasmanian migrants (Tasmanian silvereyes caught in winter on the Australian mainland).Tasmanian migrants show longer allele lengths.(b) Posterior estimates of the group means obtained from the brms model.Migrant individuals have a 1 bp increase when compared with residents.
reduction occurs because of a lack of island populations of intermediate ages.At best we can say that for silvereyes, a shift in dispersal propensity is relatively rapid, taking more than 63-95 generations (South Island New Zealand, 190 years since colonization, generation time of 2-3 years) and <1-1000 generations (Heron Island, 3000-4000 years old) (Clegg et al., 2008).Variation at a second gene, CLOCK, was associated with migratory status of individuals in a partial migrant population, with migrant individuals having longer CLOCK allele variants, including long variants absent in non-migrant individuals from Tasmania, Australian mainland and all island populations.Migrant and sedentary individuals did not show genome-wide population structure, suggesting

Variation in candidate microsatellites, DRD4 and SERT
and Norfolk Island (FigureS4; TableS7.1).Heron Island, Lord Howe Island and all Southern Melanesian populations displayed shorter allele lengths on average.
.5).SM individuals exclusively carried adenines, resulting in production of lysines, while those from New Zealand and Chatham Island only carried guanines producing arginines.Tasmanian residents and Tasmanian migrants had both nucleotides represented.We found one SNP and an INDEL in SERT but none were significantly different in frequency among populations.