Phylogenetic analysis
A 633-bp sequence was obtained from the Cyt b gene and a 478-bp sequence was obtained from the CR, which were integrated into one 1111-bp sequence. After excluding some samples of poor quality, 78 individual sequences from 2015 and 173 individual sequences from 2016 were obtained. For all mtDNA sequences, which combined the Cyt b and CR genes, coalescent sequences were aligned by the algorithm Clustal W in Mega v7 (Kumar et al. 2016). Then, DnaSP v5 and FaBox websites were used to generate haplotypes from the original individual sequences (Villesen 2007, Librado and Rozas 2009). In addition, 94 reported haplotypes were also downloaded from the National Center for Biotechnology Information (NCBI)for further alignment. The haplotypes were derived from 370 dunlin individuals breeding in eastern Asia and Alaska, including six subspecies, C . a. actites , C . a. kistchinski , C . a. sakhalina , C. a. hudsonia ,C. a. pacifica , and C. a. arcticola (GenBank accessions for CR: KP205084–KP205177; GenBank accessions for Cytb : KP205178–KP205271) (Miller et al. 2014).
To infer to the phylogenetic relationships between dunlins from the breeding areas and those sampled at Jiangsu, all sequences were analyzed with Bayes inference (BI) and maximum-likelihood (ML) algorithms. The program PhyML v3.1 was applied to infer phylogenetic relationships among haplotypes under the ML criterion, with 1000 bootstrap replicates used to evaluate clade support (Guindon et al. 2010). The best-fit nucleotide substitution model for ML analysis was distinguished using Jmodeltest v2.1.7 based on the Akaike information criterion (AIC) (Darriba et al. 2012). MrBayes v3.2.2 was used to infer phylogenetic relationships by a multigene Bayesian algorithm based on the model, as suggested by Mrmodeltest v3.7 software (Ronquist et al. 2012). Four concurrent Markov Chain Monte Carlo chains (one cold chain and three hot chains) were run simultaneously for 5.0 × 107 generations. Trees were sampled every 1000 generations, and the burn-in of the initial 25% of samples was discarded. Tracer v1.5 was used to test and check the reliability of the results of Bayesian analysis.