Genomic DNAs were extracted from blood cells. Blood samples from Lanyu (TAPS), Spotty Lanyu, Binlang Lanyu, and Mitsai Lanyu (50% Lanyu [TAPS] and 50% Duroc) were obtained from the Taitung Animal Propagation Station. Animal handling protocols conformed to those approved by the National Taiwan University Animal Care and Use Committee. Blood samples of Lanyu (NTU) and that of the Lee-Sung pig (75% Lanyu [NTU] and 25% Landrace) were obtained from National Taiwan University teaching farm. The blood of KHAPS Black pig (KHAPS) was obtained from Kaohsiung Animal Propagation Station. The blood samples of Taoyuan, Taiwan Livestock Research Institute Black Pig No. 1 (TLRI Black Pig No. 1), Meishan, Landrace, Yorkshire, Duroc, and Berkshire were obtained from the Taiwan Livestock Research Institute (TLRI). Besides Lanyu (TAPS), most of our samples in each population have recorded pedigrees and unrelated pigs up to 3 generations. Lanyu (TAPS) pigs were bred through natural mating in semi-wild environments. Sample sizes obtained from each population were collected respective of population size and blood relations between individuals. Thirty-two individual Lee-Sung pigs, 39 Lanyu (TAPS), 5 Lanyu (NTU) and 17 Mitsai Lanyu were sampled from respective stock populations. Samples from 30 Landrace, 30 Yorkshire, 30 Duroc, 29 Berkshire, 29 Meishan, 30 Taoyuan, 18 Spotty Lanyu, 21 Binlang Lanyu, 21 TLRI Black Pig No.1 and 21 KHAPS black pigs were selected based on there being no blood relationships among candidates, according to pig-breeding pedigree registrations in Taiwan. The non-related individuals of each population were chosen based on their pedigrees (except for the Lanyu TAPS population); therefore, the number of individuals in each population used in this study varies. Extraction and purification procedures for genomic DNA from blood samples were conducted using slightly modified QIAamp DNA Blood Maxi Kits (Qiagen, Valencia, CA, USA). First, 500 μL QIAGEN protease, 10 mL blood and 12 mL of buffer AL were mixed thoroughly in a 50 mL centrifuge tube. The mixture was then incubated at 70°C for 10 min before 10 mL of absolute ethanol was added and mixed. The solution was then transferred onto QIAamp Maxi column and centrifuged at 1,850×g for 3 min. Buffer AW1 and Buffer AW2 were separately added to the QIAamp Maxi column and centrifuged at 1,850×g for 1 min and 15 min, respectively. To elute the DNA, 1 mL buffer AE was applied to the column and centrifuge at 1,850×g for 5 min. Table 2 describes the number of individuals in each population and the type of pig breeds.

A total of 19 microsatellite markers located on 15 chromosomes were chosen as suggested by the Domestic Animal Diversity Information System of the Food and Agriculture Organization of the United Nations (ISAG-FAO, 2004). The primers used for microsatellite loci amplification were end-labeled with FAM, TET or HEX fluorescent dye. The fragments of microsatellite DNA in the genome were amplified by PCR (PTC-200 Thermal Controller, M. J. Research, Waltham, MA, USA). Each 15 μL reaction volume contained 50 ng of genomic DNA, 1.5 μL of 10×PCR buffer 0.375 μL of 8 mM deoxynucleoside triphosphate, 4.5 pmol sense and antisense primers, and 0.6 units of DNA Taq polymerase (Amersham Biosciences, Arlington Heights, IL, USA). Thermal cycling conditions were as follows: an initial 95°C for 5 min, then 37 cycles at 95°C for 30 s, 48°C to 62°C for 30 s (depending on the locus), and 72°C for 45 s with a final extension at 72°C for 7 min. The fluorescent end-labeled PCR products were detected by capillary electrophoresis with a MegaBACE 1000 DNA sequencer (Amersham Biosciences, Arlington Heights, IL, USA) and a fluorescent-labeled marker ET-400 (Amersham Biosciences, Arlington Heights, IL, USA) was used as an internal size standard for length calibration. The allele lengths of loci were presented by Genetic-Profiler (version 2.2; Amersham Biosciences, Arlington Heights, IL).

The total number of alleles per microsatellite locus, observed and expected heterozygosities of alleles (Nei, 1973), polymorphic information content (PIC) of each locus (Botstein et al., 1980), and deviations from the Hardy-Weinberg equilibrium were calculated with CERVUS version 2.0 software (Marshall, 1998). The mean of the effective number of alleles (MEA) was estimated following the formula provided by Kimura and Crow (1964) and calculated by POPGENE (Yeh et al., 1999). The number of polymorphic loci (NPL) equates to the frequency of the most frequent allele at a locus being <0.95. GENEPOP 4.0 computer package was used to calculate F-statistics (F_IT, F_IS, and F_ST) to understand variation in heterozygosity of each locus and differentiations among populations (Rousset, 2008). To compare the number of alleles between different sample sizes, the allelic richness (AR) of each locus, which measures the number of alleles independent of the sample size, was calculated using the program FSTAT v. 2.9.3 software package (Goudet, 2001). The mean number of alleles per breed, mean observed and expected heterozygosities (Nei, 1973) per breed, and mean polymorphic information content (PIC) per breed (Botstein et al., 1980) were calculated with CERVUS version 3.0 software (Kalinowski et al., 2007). Pairwise genetic distances between sampling populations were calculated by POPULATIONS, software version 1.2.28 (Langella, 2002). A neighbor-joining (NJ) population tree, based on Cavalli-Sforza chord distances (Cavalli-Sforza and Edwards, 1967), was used to represent relationships among groups and a consensus tree was obtained by bootstrapping (1,000 replications) distance values over loci.

Genetic distances between 14 populations are displayed in the form of a geometric picture by non-metric multidimensional scaling (MDS) techniques using the program PRIMER (Carr, 1996). Standardized Residual Sum of Squares (STRESS) shows the difference between MDS and raw data. MDS and STRESS methods are described in Krzanowski (1987). STRESS values of <0.2 mean genetic distances given by MDS can be trusted.

Assignment and exclusion testing for individuals into their reference populations is calculated by GeneClass v.2.0 (Piry et al., 2004). In this study, a total of 14 reference populations were defined as within-breeds, crossbreeds (synthetic breeds), or inbred breeds. The principle of assignment and the exclusion method are based on the Bayesian method (Rannala and Mountain, 1997). This method was suggested as a superior method by Cornuet et al. (1999) and Koskinen (2003). The assignment test for each individual is excluded from the dataset when performing own assignment (Efron, 1983). The number of simulated individuals was set 10,000 for Monte Carlo simulations. The criterion for assigning individuals to populations is based on the frequency of simulated genotypes in the reference populations; the degree of confidence is ≤0.001. To ensure an effective success rate in the analysis of the 19 markers and the aforementioned variability criteria, the 19 markers are ranked and assignment testing performed according to the highest allelic number (K), H_E, F_ST and relative discriminatory power (D-score) estimated from the 14 populations (Supplementary Table S1, approach 1: See e-version for supplement.) and 6 Lanyu related populations (i.e., Lanyu [TAPS], Lanyu [NTU], Spotty Lanyu, Binlang Lanyu, Lee-Sung and Mitsai Lanyu; Supplementary Table S1, approach 2: See e-version for supplement). The D-score of each locus to assign individuals to source cultivars was calculated with the software WHICHLOCI (Banks et al., 2003). This program ranks microsatellite loci based on their relative allelic differentials as derived from different populations (Banks and Eichert, 2000). As a control, the markers were also numbered from 1 to 19 at random and processed. The performance of each assignment test was measured both by the small number of microsatellite loci used and the proportion of individuals correctly assigned to their original populations.

The total number of alleles found for the 19 microsatellite markers from the 14 populations, including the 7 breeds, 4 synthetic breeds and 3 inbred breeds (Lanyu, Spotty Lanyu and Binlang Lanyu) was 239. The number of alleles per marker ranged from 7 (SW951, SW911) to 29 (S0005), with a global mean for the 19 markers of 12.6. On the other hand, the PIC values per locus varied from 0.555 (S0215) to 0.879 (S0005), with the mean across the 19 markers being 0.771 (Table 1). The results indicate that these microsatellite markers are polymorphic for genetic structure analysis in this study (Anderson et al., 1993). The Wright’s F-statistic values for each marker are shown in Table 1. F_IS, the parameter of within-breed deficit in heterozygosity, resulted in a global mean of 0.110 for all loci, and ranged from −0.118 (SW24) to 0.477 (S0386). F_IT, the parameter of divergence between expected and observed heterozygosity for all individuals, had a global mean of 0.468 for all loci, and ranged for the different markers from 0.335 (SW24) to 0.686 (S0386). F_ST, the parameter for genetic differentiation among breeds, varied from 0.289 (S0005) to 0.502 (S0215), and the average F_ST of all loci was 0.402. These data indicate a high degree of differentiation among the populations and further analyses of genetic differentiation and diversity within and among the studied populations are required to evaluate how informative the rankings of microsatellite loci are for individual identification.

The level of genetic diversity within each breed was determined by the number of polymorphic loci, mean number of alleles, mean polymorphism index content, allelic richness, mean effective number of alleles, expected heterozygosity, observed heterozygosity, and the F_IS value (Table 2). The within-breed analyses indicated that the number of polymorphic loci (the frequency of the most frequent allele being <0.95 at one locus) ranged from 13 to 19, the lowest appeared in the Lanyu (NTU) population. Mean number of alleles implies the degree of allelic diversity within breeds; this was greatest in KHAPS (5.158) and least in Lanyu (NTU) (2.000). Mean polymorphism index content ranged from 0.246 (Lanyu [NTU]) to 0.572 (KHAPS), indicating the 19 microsatellite markers used were informative for genetic analyses ‘within’ populations. Allelic richness measures the number of alleles per locus in a standardized uniform sample size; this was greatest in KHAPS (2.920) and least in Lanyu (NTU) (1.820). Mean effective number of alleles ranged from 1.574 (Lanyu [NTU]) to 2.970 (KHAPS). These numbers were less than the mean number of observed alleles in each population, suggesting these markers were polymorphic for genetic identification within these breeds. Mean observed heterozygosity ranged from 0.213 (Lanyu [NTU]) to 0.601 (TLRI Black pig No. 1), whereas mean expected heterozygosity ranged from 0.292 (Lanyu [NTU]) to 0.624 (KHAPS). F_IS were higher in Spotty Lanyu (0.364), Lanyu (TAPS) (0.253), Lanyu (NTU) (0.269), Binlang Lanyu (0.133), and Berkshire (0.131) while a near balance between observed and expected heterozygosity was observed in Taoyuan (0.049), Yorkshire (−0.012), Duroc (0.073), and Landrace (0.079). These data are consistent with the breeding history of breed generations used in this study.

Cavalli-Sforza chord distances and pairwise F_ST estimates in proportion to each pair of the 14 populations are shown in Table 3. The genetic distances ranged from 0.238 (between Spotty Lanyu and Binlang Lanyu) to 0.814 (Taoyuan versus Berkshire). Low genetic distances were detected between Lanyu (TAPS) versus Spotty Lanyu (0.391), and Binlang Lanyu (0.453). The shortest distance (0.238) was identified between Spotty Lanyu and Binlang Lanyu. Short genetic distances were also detected among European pig breeds (range from 0.479 to 0.598). The pairwise F_ST coefficients ranged from 0.142 (between Spotty Lanyu and Binlang Lanyu) to 0.568 (between Lanyu [NTU] and Berkshire). High F_ST coefficients were detected between several populations, including Lanyu (TAPS) and its derived breeds: Spotty Lanyu (0.210), Binlang Lanyu (0.294) and Lanyu (NTU) (0.283) as well as Lee-Sung versus Lanyu (TAPS) (0.269), and Lanyu (NTU) (0.263). Mid differentiation was detected between Binlang Lanyu and Spotty Lanyu (0.142) as well as TLRI Black pig No. 1 versus Duroc (0.165). High F_ST coefficients (>0.25) were identified among commercial breeds: Landrance, Duroc, Yorkshire, Meishan and Taoyuan. Separation from Lanyu (NTU) was greatest with Berkshire (0.568) and Meishan (0.523). For Lee-Sung, it was with Berkshire (0.529). Cavalli-Sforza chord distances based on polymorphisms of the 19 microsatellite markers from the 14 populations were then used to construct a Neighbor-Joining phylogenetic tree. Three major clusters with high bootstrap values (>95%) were obtained in the phylogenetic tree (Figure 1). Cluster I included European pig breeds (Landrace, Yorkshire, Duroc, and Berkshire) and TLRI Black Pig No. 1 breed. Cluster II included Meishan, Taoyuan and KHAPS Black. Lanyu, Lanyu derived breeds (Spotty Lanyu and Binlang Lanyu), Lanyu derived synthetic breeds, Mitsai Lanyu, and Lee-Sung pigs were clustered in Cluster III with a 100% bootstrap value. These results indicate mid- to high-genetic differentiation among these pig breeds.

Non-metric MDS placed 14 points that represented the Cavalli-Sforza chord distances among the 14 populations in a two-dimensional ordination plot (Figure 2). The STRESS value in Figure 2 was exceptionally low (0.09). This indicates that the diagram gives an excellent representation of the data (Bond et al., 2002). The three clusters are clearly indicated within the ovals in Figure 2. European breeds, TLRI Black Pig No. 1 and KHAPS Black are clustered in one cluster. Asian pig breeds other than Lanyu and Lanyu derived pig breeds are distributed in 2 different clusters. The MDS data correlated with the Neighbor-Joining tree except for the KHAPS Black pig breed that was clustered in the Asian pigs’ clade in the NJ tree. These data showed closer genetic relationships among Lanyu and its derived breeds than other breeds showed among their relationships. Accordingly, Lanyu and its derived breeds provide an excellent model for testing the strategy of selecting microsatellite markers for improving assignment testing.

An effective assignment test can be defined as that test which requires the least number and best combination of microsatellite markers for assigning individuals to their original populations. The efficiency of assignment testing is affected by the level of H_E, allele number (K), F_ST or D-score values of microsatellite markers among used reference populations. This process was evaluated and analyzed for varying conditions based on the Bayesian method (Rannala and Mountain, 1997). The performance of each assignment test was measured by the proportion of individuals correctly assigned to their original populations (Bank et al., 2003).

When random ordering criteria were used to select markers for assignment testing, 99% assignment accuracy was achieved when 7 and 16 markers were chosen, respectively (Table 4). Six markers were enough for commercial pig breeds (Landrace, Yorkshire, Berkshire, Duroc) to obtain 100% assignment accuracy; 5 markers were adequate for TLRI Black pig No.1 and KHAPS Black pig to obtain 100% assignment accuracy; and 2 markers were enough for Taoyuan and Meishan to obtain 100% assignment accuracy (Supplementary Table S2, A: See e-version for supplement.). In addition, according to analysis results from the 14 populations (Supplementary Table S1, approach 1: See e-version for supplement.), marker subsets were selected by ranking all 19 microsatellite markers according to highest H_E, K, F_ST, and D-score values to conduct assignment tests. Nine markers with highest D-scores, 10 markers with highest H_E or K and 14 markers with highest F_ST were required to achieve 99% assignment accuracy. The commercial pig breeds needed less than 7 markers ranked according to highest H_E, K, F_ST, and D-score values from 14 populations to obtain 100% assignment accuracy (Supplementary Table S2 B, C, and D: See e-version for supplement.). After checking row data, most of incorrect individual assignments came from missed assignment among close lineage breeds (Lanyu, Binlang Lanyu and Spotty Lanyu, Table 5; Supplementary Table S2 and S3: See e-version for supplement). This result suggests very close lineage among reference populations may affect assignment accuracy.

To increase the performance of assignment testing, markers were selected by ranking highest H_E, allele number, F_ST and D-scores obtained from 6 Lanyu related populations (Lanyu [TAPS], Lanyu [NTU], Spotty Lanyu, Binlang Lanyu, Lee-Sung and Mitsai Lanyu) and conducting assignment testing among the 14 populations (Table 4, approach 2; Supplementary Table S1, approach 2: See e-version for supplement.). Figure 3 and Table 4 show that 99% accuracy in assignment was obtained using the following marker requirements: 6 markers with highest H_E, 7 markers with highest F_ST, 8 markers with highest K, and 10 markers with highest D-scores. If 9 markers with highest H_E or K were selected, we obtained 99.7% assignment-test accuracy. On the other hand, assignment accuracy for commercial breeds decreased when marker selection was from just 6 Lanyu related populations by ranking highest H_E, allele number, F_ST and D-scores (Supplementary Table S2 E, F, and G: See e-version for supplement.). One-hundred percent assignment accuracy among commercial breeds was achieved using: 9 markers with highest H_E, 7 markers with highest F_ST, and 9 markers with highest K (Supplementary Table S2 E, F and G: See e-version for supplement.). Overall assignment performance significantly improved after subset markers were selected based on highest ordering of H_E, allele numbers, F_ST and D-scores from close lineage Lanyu populations.

Inbreeding of the Lanyu and its derived breeds, Spotty Lanyu and Binlang Lanyu resulted in low allelic richness (2.097 and 1.980) and mean effective allele numbers (1.955 and 1.738) (Table 2). Lanyu (NTU) possesses low numbers for polymorphic loci (13), allelic richness (1.82), and mean effective allele numbers (1.574). This might be due to the initial size of the breeding population being small and founder effect (Wu et al., 2007). Fixation indexes (F_IS) were significantly derived from 0 in Lanyu (NTU) and Lanyu-derived populations, suggesting a loss of heterozygosity in these populations. Berkshire breeds (not a popular commercial breed in Taiwan) also possess low allelic richness and mean effective allele numbers. This may also be due to high intensive inbreeding and founder effect (Wu et al., 2007). Crossbred synthetic breeds, including Lee-Sung, Mitsai Lanyu, TLRI Black Pig No.1 and KHAPS Black pig, show reasonably high numbers for polymorphic loci, allelic richness and effective allele numbers even though all of them are established from small populations experiencing founder effect. This suggests that heterosis within these two very small unrelated populations have overcome founder effect in the inbreeding index. A similar result was obtained when the hermaphroditic freshwater snail lymnaea stagnalis was used as a model to test heterosis and inbreeding depression in bottlenecked populations (Coutellec and Caquet, 2011). Both the TLRI Black Pig No. 1 (75% Duroc and 25% Taoyuan) and KHAPS Black pig (50% Meishan and 50% Duroc) commercial proposed breeds possess higher figures for mean allele numbers (4.105 and 5.158, respectively), allelic richness (2.698 and 2.920, respectively) and effective allele numbers (2.547 and 2.970, respectively) than the Lee-Sung and Mitsai Lanyu populations. These differences may result from variations in population size and inbreeding practices after crossing of original parents. Information richness from the 19 microsatellite markers was enough to allow analysis of Taoyuan derived and Duroc derived synthetic breeds. Hale et al. (2012) randomly sub-sampling 5 to 100 individuals from 4 empirical microsatellite genotype datasets (Formica lugubris, Sciurus vulgaris, Thalassarche melanophris, and Himantopus novaezelandia) showed 25 to 35 individuals per population are enough to accurately estimate allele frequencies in genetic studies of populations. In this present study, the numbers used from each population varied; therefore, to compensate for this drawback, Nei (1978) and Kalinowski (2005) suggested using a large number of markers when estimating the average heterozygosity of a population.

An index of genetic distance and differentiation among populations is given in Table 3. The genetic distances and differentiations range from 0.238 to 0.814 and 0.142 to 0.568, respectively. Lanyu and its derived populations showed little genetic distance but mid (0.142 between Binlang Lanyu and Spotty Lanyu) to high differentiation coefficients (0.475 between Lanyu [NTU] versus Mitsai Lanyu). In addition, there was lower observed heterozygosity (0.213 to 0.397) within Lanyu inbreeding populations (Table 2). Interesting, Lanyu (NTU) was co-ancestor to Lanyu (TAPS), Spotty Lanyu and Binling Lanyu but with high F_ST (from 0.283 to 0.416) values. These results indicate speciation among Lanyu and its derived breeds caused by intense inbreeding and founder effect. High genetic differentiation occurs when a highly intense inbreeding program is performed even among two pig populations with high levels of co-ancestry. High genetic distance and mid to high differentiation coefficients also exist between Duroc and Duroc synthetic breeds (Mitsai Lanyu and TLRI Black pig No. 1), ranging from 0.66 and 0.446 to 0.165 and 0.355, respectively (Table 3). In addition, Neighbor-Joining tree analysis formed 3 sub-clusters (Figure 1): the Lanyu, its derived breeds, and derived synthetic breeds. These sub-clusters formed a unique clade with a 100% bootstrap value. The European breeds and TLRI Black Pig No. 1, Asian pig breeds and KHAPS Black pig formed the other two clusters. The results from MDS plot analysis show similar results to Neighbor-Joining tree analysis except for the KHAPS Black pig breed. This might be due to close lineage and small genetic distances between KHAPS Black pig versus Duroc (0.512) and Meishan (0.544). The data indicate mid to high genetic divergence among these groups. Therefore, Lanyu populations provided both close genetic breeds and highly differentiated related breeds. This ideal animal model made it possible to examine and conduct different strategies for assignment testing. It allowed the development of an efficient method for identifying individuals based on our reference data.

Some factors such as assignment principles, degree of genetic differentiation among populations, sample size per population, number of markers used, and level of variability in markers affect the accuracy and efficiency of assignment testing (Cornuet et al., 1999; Bjørnstad and Røed, 2002). The rate at which accuracy of allele frequency-estimates increases should level out as sample size increases with the cost of genotyping more individuals increasing congruently (Hale et al., 2012). In addition, loci with high expected heterozygosity, K, and F_ST values across reference populations provides the best means of selecting loci to improve assignment testing efficiency (Rosenberg et al., 2001; Tadano et al., 2008). The Bayesian statistical approach proved more efficient than the genetic distance approach in calculating assignment of individuals to populations (Rannala and Mountain, 1997; Cornuet et al., 1999; Primmer et al., 2000). Even with as few as 10 microsatellite loci with heterozygosity ≈0.6, assignment accuracy performed by the Bayesian method can reach 100% based on 30 to 35 individuals from each of 10 populations when F_ST is near 0.1 (Cornuet et al., 1999). In this study, high heterozygosity (mean = 0.798) and a high level of genetic differentiation (mean F_ST = 0.403) among 14 pig populations increased the success rate of the Bayesian statistical approach in assigning individuals to populations (Supplementary Table S1, approach 1: See e-version for supplement.). Through applying the direct approach to the selection criteria needed for microsatellite markers, we could assign more than 99% of 352 animals to their correct reference populations. To make these assignments more efficient, we required 14 markers for ranking highest F_ST, 10 markers for highest H_E, 10 markers for highest K, and 9 markers for highest D-scores to obtain 99.1% correct assignment. This result is better than the assignment accuracy of previous studies (91.3% in Kim et al., 2005; 92.14% in Yang et al., 2003). Individual assignment based on markers of highest F_ST from 14 populations did not give ideal results.

Assignment error rates increase depending upon genetic similarities between populations and co-ancestry between individuals (Guinand et al., 2004). A perfect marker for understanding error-rates is mono-morphism within given breeds but polymorphism across breeds (Reed, 1973). Among the 352 individuals tested, one Spotty Lanyu was assigned to Lanyu in the TAPS population, and one Binlang Lanyu was assigned to the Spotty Lanyu population. The commercial breeds with high genetic differentiation obtained better assignment. These results reveal the closeness of these inbred groups and their preservation of original genetic characteristics (Table 4). This result led us to hypothesize that loci with high polymorphism, heterozygosity, F_ST and D-scores among Lanyu, its derived breeds, and synthetic breeds may provide a better ability to discriminate between individuals. High mean heterozygosity (0.631) and F_ST (0.305) were calculated for Lanyu and its derived populations, suggesting that genetic variation existed in related populations (Supplementary Table S1, approach 2: See e-version for supplement.). We, therefore, selected these microsatellite markers to correct individual assignments. By using markers within the Lanyu and Lanyu derived breeds and ranking these for highest H_E, highest K, high F_ST and high D-score. We found 6 markers with high H_E (S0218, SW951, S0355, SW122, SW857, S0266), 7 markers for high F_ST and 8 markers for high K (S0005, S0227, SW857, S0266, S0218, S0355, SW122, SW951) were required to obtain 99% assignment accuracy (Table 4 and Figure 3). High D-score microsatellite markers within Lanyu breeds did not increase individual-discrimination efficiency. Interestingly, microsatellite markers (SW857, S0355, SW951, and S0227) are used in the paternity control system for DNA tracking of pigs under selection (MAPS) by Spain and France (Boitard et al., 2010), indicating that these markers are informative for both Asian and European type pig breeds.