Migratory phenotype characterisation
We classified each of the 70 species included in our study, according to their migratory phenotype. Our classification was based on a careful literature review of bird guides (Svensson et al. 1999), as well as BirdLife (http://birdlife.org) and Handbook of birds of the World (HBW) (http://www.hbw.com). We defined the following categories: clearly non-migratory (resident, sedentary) (0/R; n = 32), obligate migrant (2/M; n = 17). However, sometimes it is not easy to clearly define a species as either clearly non-migrant or obligate migrant, this is especially true when a migratory trait segregates within a population such as in partial migrants where only some individuals of the population migrate, consequently we added a third category (1; n = 21). This category includes partial migratory species (i.e. not all individuals of the population migrate), and species that exhibit other kind of migration-independent movement behaviour (e.g. dispersal, homing, foraging flights). For partial migrant species we used additional information of the individual used for generating the reference, such as date and geographical origin of sample collection, in order to clearly define migratory phenotype and grouped that individual/species accordingly whenever possible.
Genome sequences, extraction and alignment
We downloaded genome sequences and annotations for most of the species used in our study from the NCBI database (Supplementary Table S1). We further included genome sequences of five additional migratory species here: Siberian stonechat Saxicola maurus (Van Doren et al. 2017), Swainson’s thrush Catharus ustulatus (Delmore et al. 2015a), European blackcap Sylvia atricapilla (Delmore et al., in preparation), Willow warbler Phylloscopus trochilus (Lundberg et al. 2017, accepted), and Greenish warbler Phylloscopus trochiloides (Irwin et al. 2016).
Once we had sequences data and annotation for all of the species, we used the Bedtools (Quinlan and Hall 2010) getfasta module to extract genomic sequences for each of the 25 candidate genes for every species. Sequences for unpublished genomes or genomes without annotations (for details see Table S1) were generated using Blastn and chicken cDNAs from the Ensembl database as a reference. All genomic sequences were aligned with MAFFT (Katoh and Standley 2013) and manually edited in AliView. Coding (CDS) sequences were also obtained from a multiple alignment of the genomic sequences and Ensemble cDNA sequences (including untranslated regions, UTRs) for the flycatcher, chicken, and Zebra Finch. Only sequences covering 50% or more of the chicken genes were considered for further analysis.
Evolutionary trees were constructed for each candidate gene using whole genomic sequences and cDNA as reference, using a Neighbour Joining approach in MEGA v5.2 (Tamura et al. 2011). The reliability of the trees was evaluated performing a bootstrap analysis of 1000 replicates with the Kimura 2 Parameters model. To visualise the trees we used Figtree (http://tree.bio.ed.ac.uk/).
We compared the pattern of evolutionary divergence of these gene trees with three different hypothetical scenarios: divergence driven by phylogeny (‘phylogenetic topology’); divergence constrained by migration, i.e. different migratory phenotypes clustering in separate braches, while keeping the evolutionary relationship of the phylogenetic topology within each branch, (‘migratory phenotype topology’); and random divergence (‘random topology’). These comparisons were carried out for (a) the full dataset including all three migratory phenotypes; (b) a restricted dataset only contrasting exclusively obligatory migratory and completely non-migratory (resident) species, and (c) a clade-specific analysis exclusively focusing on the genus of Passeriformes, as this is the only monophyletic clade in our dataset with a sufficiently high number of species for both obligate migrants and non-migratory species, thus allowing for a more fine-tuned assessment on a narrower phylogenetic scale. The clade-specific subset allows us to test if the migratory phenotype might be controlled by a different clade-specific subset of genes. This comparative approach allows us to identify the presence or absence of general patterns, using genetic variation at candidate genes to distinguish between patterns related to phylogenetic relationships and migratory behaviour. The divergence driven by phylogeny (i.e. the gene trees matching the species tree, ‘phylogenetic topology’) was constructed using the total evidence nucleotide species tree (TENT) phylogeny, published by Jarvis et al. (2014). For species not included in the TENT phylogeny we used timetree (Hedges et al. 2015) divergence times to position these species in our phylogeny. The divergence constrained by migration scenario (‘migratory phenotype topology’) was constructed by clustering each phenotype (once exclusively focusing on migratory versus resident species for the restricted dataset; and also for the full dataset including other movement as a third phenotype category) in one separate branch while keeping the evolutionary relationships of the phylogenetic topology within each branch. Random divergence (‘random tree’) was generated shuffling branches randomly from the gene trees obtained, in order to avoid bias regarding the method of random trees generation by TOPD/fmts that only randomises taxa, but not branches for the statistical comparison. Restricting these analyses to exclusively Passerine species allowed us to analyse the effects of the evolutionary patterns of each candidate genes on a smaller scale. An example of the topologies is illustrated for the candidate gene PER3 in Fig. S1.
Comparisons of these three focal topologies were carried out in TOPD/fmtS (Puigbò et al. 2007) using three different approaches: nodal, splits and disagree from the program. In brief, the ‘nodal approach’ counts the number of nodes that separate two taxa in a given topology and calculates the root mean squared deviation (RSMD) between each pair of trees. For identical topologies RMSD results in a value of zero. To calculate the significance of the RMSD obtained, TOPD/fmts calculates the distance between two contrasted tree pairs and 100 random trees obtaining one standard deviation (SD) confidence interval (CI). Compared topologies are characterised as statistically similar, within noise or different, depending on their distance with respect to CI. Specifically distances below CI denote similar topologies; distances above CI indicate statistical difference (distances around CI are within noise). The ‘disagree method’ characterises how many taxa need to be removed from the compared topology in order to end up with the exact same topologies for both trees (assessed as count of taxa/total taxa). Consequently, a value of 0/total indicates identical topologies. The ‘splits method’ evaluates if there are common branches between both trees, the lower the distance the more branches the tree pair shares.
Accounting for the fact that different species might have found different ways to alter similar phenotypes in the same gene (i.e. different changes in sequence), we also analysed synonymous and non-synonymous mutations of all candidate genes across species. In order to pick up on putative selective pressures on candidate genes for migration, a gene-wide dN/dS analysis was carried out on the Datamonkey server (Pond and Frost 2005). We used three different datasets for each candidate gene: one including all the species, one restricted to migratory, and another restricted to non-migratory species.
Gene-wide dN/dS ratios (w) were estimated by maximum likelihood (ML) methods using a different model for each gene. Each model was obtained from the CMS module of the server. Neighbour Joining (NJ) phylogenies obtained for each candidate gene were used as input to assess likelihood of the tree comparing the neutral null model M1 (w < 1) and the model M2 that allows w > 1. Positive selection was assessed if the likelihood shows a p < 0.05.
To evaluate if lineages with migratory species show a signature of selection, a branch-specific analysis of dN/dS was also carried on Datamonkey with the Branch-Site REL program (Pond et al. 2011). The dataset for each candidate included migratory and non-migratory species. Branches under episodic diversifying selection were identified with a Holm–Bonferroni corrected p ≤ 0.05.
Structural features and predictors of migration
We used a linear regression analysis to test for a correlation between the genotype of migratory species at each focal candidate gene and both breeding latitude and migratory distance. Models for both predictors were run separately. The genotype used for each gene was the microsatellite length (as number of bases) of the 3′UTR of ADCYAP1 and CREB1, or poly-Q (as number of predicted glutamine amino acids) on exon 20 of CLOCK and NPAS. For the CLOCK gene we included two separate polymorphic regions with variable poly-Q repeats in our analysis (both variable regions are located in the same exon). The significance of the fit was assessed with a simple linear regression, using a significance threshold p ≤ 0.05.
Within and across population variability in candidate gene sequence
Our comparative analyses focus exclusively on the sequence of one reference genome; inter-individual variation is not taken into account, mostly due to the limitation of available data to examine this level of variation. In order to make an attempt to see if variance within on candidate gene could be higher/lower in a specific migratory phenotype, we focused on CLOCK gene polymorphism, the only candidate gene with a sufficiently high number of individual sequence data available for several species (n = 10), including both migratory (n = 8) and resident (n = 2) species. Here we compare datasets of individually genotyped migratory species: flycatcher Ficedula hypoleuca (Saino et al. 2015; n = 226), willow warbler Phylloscopus trochilus ssp (unpublished data, n = 384), chiffchaff Phylloscopus collybita ssp (unpublished data, n = 56), nightingale Luscinia megarhynchos (Saino et al. 2015; n = 151), tree pipit Anthus trivialis (Saino et al. 2015; n = 144), barn swallow Hirundo rustica (Dor et al. 2011; n = 830), whinchat Saxicola rubetra (Saino et al. 2015, n = 374); and two non-migratory species: blue tit Cyanistes caerulea (Liedvogel et al. 2009; n = 950), great tit Parus major (Liedvogel and Sheldon 2010; n = 804). We compared averages and variances among different pairs of groups or species, employing a Welch t test and F test, respectively. We assume as statistically similar, distributions with a p > 0.001.