Seasonal migration is a dynamic natural phenomenon that allows organisms to exploit favorable habitats across the annual cycle. While the morphological, physiological, and behavioral changes associated with migratory behavior are well characterized, the genetic basis of migration and its link to endogenous biological timekeeping pathways is poorly understood. Historically, genome-wide research has focused on genes of large effect, whereas many genes of small effect may work together to regulate complex traits like migratory behavior. Here, we explicitly relax stringent outlier detection thresholds and, as a result, discover how multiple biological timekeeping genes are important to migratory timing in an iconic raptor species, the American Kestrel (Falco sparverius). To validate the role of candidate loci in migratory timing, we genotyped Kestrels captured across autumn migration and found significant associations between migratory timing and genetic variation in metabolic and light input pathway genes that modulate biological clocks (TOP1, PHLPP1, CPNE4, and PEAK1). Further, we demonstrate that migrating individuals originated from a single panmictic source population, suggesting the existence of distinct early and late migratory genotypes (i.e. chronotypes). Overall, our results provide empirical support for the existence of a within population-level polymorphism in genes underlying migratory timing in a diurnally migrating raptor.

We designed Fluidigm SNPtype assays and used them to screen additional breeding and migrating American Kestrels that were independent of the RAD-seq analyses above.  Specifically, we used the R package snps2assays (Anderson 2015) to evaluate the efficacy of designing assays for candidate loci. We considered the assays designable if GC content was less than 0.65, there were no insertions or deletions (indels) within 30bp of the target variant, and there were no additional variants within 20bp of the targeted variable site. We filtered out assays with primers that mapped to multiple locations in the genome (bwa mem: Li and Durbin 2009), resulting in assays for nine loci in nine candidate genes. We used the resulting Fluidigm assays to genotype the nine candidate migration genes in 738 breeding American Kestrels from 83 sites and 165 migrating American Kestrels from a single migration station in Boise, Idaho collected in a three-month time-series spanning fall migration over two years.

We then used a multi-gene and single gene framework to determine whether migratory timing was significantly associated with allele frequency shifts in the nine candidate migration genes. To determine how the nine candidate genes covary with each other, we conducted an ordinal principal component analysis (PCA) using the R software package gifi (Mair and De Leeuw 2019) . We used a linear regression to evaluate whether migration timing (day of year when a fall migrant was captured) was associated with genetic variation as measured by PC1 and PC2, and included a covariate of sex to account for the potential influence of differential migration between sexes on migration timing. To investigate single gene effects, we fit linear regression models of each allele frequency of the top 4 candidate genes, i.e. those that loaded strongly on PC1, TOP1, PEAK1, PHLPP1 and CPNE4, to migration timing as defined by the midpoint day of each week during the autumn migration period and using the lm model in the R software package stats v 3.6.2 (R Core Team 2019). The nonlinear decline in allele frequency over time prompted the fitting of a curved regression model, and we tested whether this linear regression polynomial model provided a better fit using a likelihood ratio test in the R package lmtest v 0.9-37 (Zeileis and Hothorn 2002).

To test whether seasonal allele frequency trends result from different populations migrating through the migration station at different times or distinct migratory chronotypes, we examined the association between PC1 and latitude as well as allele frequency in our 4 top ranked loci and latitude of kestrels breeding across the west. Further, we genotyped 151 of the 165 migrating birds from Boise, Idaho (all samples for which we had high quality DNA remaining) with population-specific SNP-type assays used in Ruegg et al. (2021), and assigned these birds to the breeding population of origin using rubias (Anderson and Moran 2018).

Anderson, E. C. snps2assays: Prepare SNP assay orders from ddRAD or RAD loci. (2015)

Anderson, E. C. & Moran, B. rubias: Bayesian Inference from the Conditional Genetic Stock Identification Model. (2018)

Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows–Wheeler transform. Bioinformatics 25, 1754–1760 (2009).

Mair, P. & De Leeuw, J. Gifi: Multivariate analysis with optimal scaling. (2019).

R Core Team. R: A language and environment for statistical computing. (R Foundation for Statistical   Computing, 2019).

Ruegg, K. C. et al. The American Kestrel genoscape (Falco sparverius): Implications for monitoring, management, and subspecies boundaries. Auk (2021)

Zeileis, A. & Hothorn, T. Diagnostic checking in regression relationships. R News 2(3), 7-10. R News 2, 7–10 (2002).