Spatial modeling of population membership in indigenous Eastern Adriatic sheep breeds using codominant marker genotypes

Number of biological populations and their spatial boundaries for 13 sheep breeds of the Eastern Adriatic coast was inferred using Geneland 4.0.8. package for R. Random unrelated genotypes of 22 to 51 sheep from each breed were analyzed, where 51 samples of Istrian sheep were collected in Croatia and Slovenia. Clustering was performed under spatial and non- spatial models for the 28 genotyped microsatellite markers, with and without the assumption of correlated allele frequencies. Stochastic inference was tested for 4 different groups of models (10 repetitions each, 106 iterations of the Markov chain Monte Carlo, 200 burn-in runs). Models with one to 18 clusters of genotypes were tested, accounting for putative null-alleles and treating the double missing genotype as genuine missing data. Explicit spatial coordinates of the samples were clumped, covering area of approximately 160, 000 km2 and were treated with the parameter of uncertainty of 10. Maximal number of nuclei in non-spatial models was set to the number of samples (317), and 3 times as much in spatial models. Non-spatial models showed poor chain mixing implying problems with convergence that can arise due to departure from modeling assumptions. Spatial model without assumed frequencies correlation recognized only 3 clusters. The expected low differentiation of the clusters due to larger variability within the breeds than among them, and spatial information found to be relevant in other sheep clustering studies, adduced toward a posteriori preference of the spatial model with correlated allele frequencies. A map of population membership was created for the best run of modal k = 14. Samples clustered according to the breed origin with lower resolution for the southern populations. The estimated spatial scale parameter (b = 63) and the low a parameter (a = 3.9) confirm the admixture structured in space and the value of explicit spatial data in indigenous sheep diversity assessment. Possible isolation by distance pattern in the sample may adduce toward recommendation of clustering using tessellation polygons.