Stylosanthes guianensis (Aubl.) Sw. (2n= 20) is the most widespread Stylosanthes species and exhibits great phenotypic variation (Williams et al. 1984; Vieira et al. 1993). This species is native to South and Central America, where it is widely distributed, although not in the equatorial zone (Williams et al. 1984). Stylosanthes guianensis is considered to be a promising forage crop in the Brazilian savannahs, and many accessions produce a large amount of dry matter and retain their leaves during the dry season (Andrade and Karia 2000; Maass and Sawkins 2004).

The taxonomic classification of S. guianensis is controversial, and different taxonomic groups have been proposed based on different morphological characters. Mannetje (1977, 1984) recognized seven different S. guianensis varieties: S. guianensis var. dissitiflora, S. guianensis var. gracilis, S. guianensis var. guianensis, S. guianensis var. intermedia, S. guianensis var. longiseta, S. guianensis var. marginata and S. guianensis var. robusta. Ferreira and Costa (1979) considered Mannetje's varieties to be different species. They proposed that S. guianensis was composed of the Mannetje S. guianensis var. guianensis but subdivided this species into four different botanical varieties. Analyses of random amplified polymorphic DNA (RAPD) markers (Kazan et al. 1993; Vieira et al. 1997), sequence-tagged site and internal transcribed spacer (ITS) sequences (Vander Stappen et al. 1998, 1999b) demonstrated a high genetic diversity among the varieties proposed by Mannetje (1984), supporting their separation into distinct species. More recently, Costa (2006) used ITS sequencing and classical taxonomy to review the genus Stylosanthes. The ITS data support the classification proposed by Ferreira and Costa (1979) and Brandão et al. (1985), according to whom there are four botanical varieties of S. guianensis: S. guianensis var. guianensis, S. guianensis var. canescens, S. guianensis var. microcephala and S. guianensis var. pauciflora.

Vieira et al. (1993) analysed the karyotypes of the botanical varieties described by Ferreira and Costa (1979) and Brandão et al. (1985), and four species considered to be S. guianensis varieties by Mannetje (1977, 1984). Their results demonstrate that the varieties proposed by Ferreira and Costa (1979) and Brandão et al. (1985) have similar karyotypes, suggesting that S. g. canescens, S. g. microcephala and S. g. pauciflora may all have evolved from S. g. guianensis. The four species considered to be S. guianensis varieties by Mannetje (1977, 1984) have distinct karyotypes, supporting the classification proposed by Ferreira and Costa (1979) and Brandão et al. (1985).

Microsatellites, or simple sequence repeats (SSRs), are useful markers for a variety of applications in plant genetics because they are codominant, multiallelic, easily detected by polymerase chain reaction, relatively abundant and provide good coverage of the genome (Powell et al. 1996a). Microsatellite markers are available for three species of Stylosanthes: S. guianensis (Vander Stappen et al. 1999a; Santos et al. 2009a), S. capitata (Santos et al. 2009b) and S. macrocephala (Santos et al. 2009c). Vander Stappen et al. (1999a) described 18 genomic microsatellites in S. guianensis and tested them in 65 genotypes of S. guianensis and in the related species S. gracilis, S. hippocampoides, S. grandifolia, S. acuminata and S. longiseta. All 18 microsatellites were polymorphic among S. guianensis and its related species, and 16 were polymorphic within S. guianensis. Karia (2008) tested the microsatellites developed by Vander Stappen et al. (1999a) to evaluate the relationships among 437 accessions of the S. guianensis Brazilian germplasm collection and found that only seven were polymorphic among the accessions. Because of the limited number of markers, many relationships were unclear, and few could be clarified with respect to the population structure and to the clustering of the botanical varieties based on the results of Karia (2008). Considering the small number of polymorphic microsatellites available for this species and the need to better evaluate the Brazilian germplasm collection, Santos et al. (2009a) developed 46 new genomic microsatellites, 20 of which were polymorphic when tested in 20 S. guianensis accessions.

Santos-Garcia et al. (2011) tested both sets of microsatellites (Vander Stappen et al. 1999a; Santos et al. 2009a) in 20 accessions that were used as maternal parents in a progeny array and found that only five were polymorphic and could be used to determine the mating system of S. guianensis. Based on the microsatellite data, these authors estimated the outcrossing rate in S. guianensis to be 26 %, indicating that the species presents a mixed mating system with predominance of autogamy.

Despite the importance of S. guianensis as forage in tropical areas and the fact that some Brazilian regions contain most of the diversity of this genus, few studies have been conducted to determine the genetic diversity in the Brazilian germplasm collection. The Brazilian S. guianensis collection was established in a series of collecting trips around Brazil and other South American countries, and it has not been subjected to any breeding selection. Genetic studies of this collection may provide valuable information for planning new collection trips as well as for the study of natural populations. Karia (2008) was the first to study this collection using molecular markers, but this analysis did not provide enough information to evaluate the genetic diversity of the collection or to clarify the botanical classifications and the relationships among the accessions.

Considering the classifications proposed by Ferreira and Costa (1979), Brandão et al. (1985) and Costa (2006), we have selected a representative sample of the S. guianensis Brazilian germplasm collection consisting of 150 accessions that were either classified as one of the four botanical varieties or remained unclassified. We are dealing with botanical varieties of S. guianensis in this study, and their classification is still controversial. Currently, microsatellite markers have been the most widely used for inferring population structure that ultimately defines variation at the infraspecific level.

The observed and expected heterozygosity values (H_O and H_E) ranged from 0.00 to 0.58 (mean of 0.18) and from 0.04 to 0.83 (mean of 0.55), respectively. The markers with a higher observed heterozygosity were SG01D7 and SG01B9. In the case of bulk samples, a heterozygous pattern may be observed as a result of the presence of heterozygous plants in the DNA pool or variation within the accession (Zhang et al. 2008).

Roger's genetic distances were estimated for each pair of accessions and varied from 0.00 (among several accessions) to 0.94 (between accessions 19 and 140), with an average of 0.66.

A model-based analysis performed by STRUCTURE was used to investigate the possible population structure within the collection sample (150 accessions). The number of clusters was set to nine, following the statistics presented in Evanno et al. (2005). In the ancestry plot (Fig. 1B), each accession is represented by a horizontal bar, and the length of each segment of the bar is proportional to the accession's estimated ancestry fraction from each of the nine groups. The same ancestry plot is shown in more detail in Supplementary Figure S1 [see Additional Information], which includes the sample codes for the accessions. The majority of the model-based groups were in agreement with the currently proposed S. guianensis taxonomic classification (Ferreira and Costa 1979; Brandão et al. 1985). Some of the groups could also be correlated with the geographic origins of the accessions.

The number of accessions belonging to each variety that were assigned to one of the nine groups with >80 % probability is shown in Table 3. Of the 40 accessions belonging to S. g. guianensis, 11 were assigned to Group C, 5 to Group H and 14 to Group I. These groups are formed mostly by accessions belonging to this variety, which showed the greatest genetic diversity among all Stylosanthes varieties. The remaining 10 accessions from S. g. guianensis were distributed along with S. g. pauciflora accessions. Three of the accessions were assigned to Group B, two were assigned to Group G, and five were not assigned to any group with >80 % probability. Of the 28 S. g. pauciflora accessions, 13 were assigned to Group A (formed mostly by this variety), seven were assigned with S. g. guianensis in Groups B and G, two were in distinct Groups (E and I), and six were not assigned to any group with >80 % probability.

The 15 S. g. microcephala were mostly clustered in Groups D and E, both with six accessions. Group D contained six accessions belonging to S. g. microcephala and two non-classified accessions, and Group E contained six accessions belonging to S. g. microcephala, one accession from S. g. pauciflora and three non-classified accessions. Of the remaining three S. g. microcephala accessions, two were assigned to Group I and one to Group A.

The S. g. canescens accessions were clustered in a distinct Group (F) composed exclusively of this variety. Of the 20 S. g. canescens accessions studied, 14 were assigned to Group F.

In the dendrogram constructed using DARwin and the UPGMA method (Fig. 1A), the clusters, with few exceptions, were consistent with the groups generated using the Bayesian approach in STRUCTURE. The same dendrogram labelled with the sample code for the accessions is provided in Supplementary Figure S2 [see Additional Information].

The mean number of alleles per locus observed in this study (4.7) was higher than that reported by Vander Stappen et al. (1999a) for 65 S. guianensis genotypes and some related species but lower than that observed by Karia (2008) among 437 S. guianensis accessions; those authors reported mean numbers of alleles per locus of 3.7 and 6.43, respectively. These differences can be attributed to the number of accessions studied by each of the authors.

Considering the observed and expected heterozygosities, our data revealed a deficit in heterozygosity that was consistent with the S. guianensis outcrossing rate (26 %) that was estimated based on microsatellite data (Santos-Garcia et al. 2011). The predominance of autogamy reduced the number of heterozygous samples, but some heterozygosity was maintained as a result of the 26 % outcrossing.

The genetic distances found among the studied accessions were higher than those previously reported for the species. The highest genetic distance (0.94) was observed between accessions 19 and 140. Accession 19 belongs to S. g. pauciflora and was collected in Venezuela; accession 140 belongs to S. g. guianensis and was collected in the Brazilian state of Minas Gerais. Accession 140 was one of the most divergent, and its mean genetic distance (0.75) is greater than the overall mean genetic distance (0.66). Kazan et al. (1993) studied 31 S. guianensis accessions and observed a mean genetic distance of 0.26 using RAPD. Faleiro et al. (2003) also used RAPD to study 35 accessions and observed genetic distances varying from 0.04 to 0.54. Microsatellite markers are codominant and multiallelic with a high degree of polymorphism, making them more useful than dominant markers for revealing diversity (Laborda et al. 2005). Powell et al. (1996b) compared the expected heterozygosities and the estimated genetic similarities based on different molecular marker types for the evaluation of a set of soybean accessions. Their results show that microsatellites have higher expected heterozygosity (0.61) and lower estimated genetic similarities (0.45) among accessions relative to RAPD markers (0.31 and 0.72, respectively).

STRUCTURE analysis was performed to investigate the population structure in the germplasm collection to compare the genetic population structure with the described botanical varieties. The nine groups generated based on microsatellite analysis were mostly consistent with the botanical classification. The S. g. microcephala and S. g. canescens varieties were well differentiated and formed individual groups. Although most of the accessions belonging to S. g. guianensis and S. g. pauciflora formed distinct groups, some of them were mixed together in other groups. These mixed groups generally contained small numbers of individuals, many of which were not classified into varieties. Stylosanthes g. pauciflora was recognized as a new botanical variety in 1985 (Brandão et al. 1985); the three other varieties were recognized in 1979 (Ferreira and Costa 1979). This may have affected the classification process, and some of the accessions belonging to S. g. pauciflora may have been incorrectly classified as S. g. guianensis.

In previous studies, most S. g. microcephala accessions did not group together (Faleiro et al. 2003; Karia 2008). In the present work, these samples clustered into two diverse groups. Of the 15 S. g. microcephala accessions, five from the state of Minas Gerais were assigned to Group D, and six accessions (two from the state of Goiás and four from the state of Tocantins) were assigned to Group E. In both groups, the majority of the accessions were S. g. microcephala. The soil in Goiás and Tocantins is considered less fertile than the soil in Minas Gerais, and this difference in soil fertility may be the main driver of the observed genetic differentiation (Karia 2008). Our data reinforce the idea that the presence of S. g. microcephala in Minas Gerais is associated with more fertile soils, as proposed by Costa (2006). However, further studies are necessary to address this question and should include the collection and analysis of more plants from those regions. If genetic differentiation based on soil conditions is confirmed by further studies, this information could be useful for the development of new commercial varieties that are adapted to specific soil conditions.

The genetic data obtained in this study are also consistent with previously published karyological findings. All varieties have similar total chromosome lengths except for S. g. microcephala, which has visibly smaller chromosomes and the most asymmetrical karyotype (Fig. 2) (Vieira et al. 1993). This variety predominantly formed Groups D and E (Fig. 1B). Stylosanthes g. canescens and S. g. microcephala are quite close in the dendrogram (Fig. 1A); in those varieties, chromosome 10 is submetacentric (2.58 and 2.55 as arm ratios, respectively) (Table 4), which is distinct from chromosome 10 in the other varieties. However, S. g. canescens has a uniquely submetacentric chromosome 8 (1.76 arm ratio). As revealed from the UPGMA-based dendrogram and the model implemented in the STRUCTURE software, S. g. guianensis and S. g. pauciflora share alleles at many of the microsatellite loci under investigation. The karyotypes of these two varieties are also very similar, except for chromosome 10, which is significantly smaller in S. g. guianensis than in S. g. pauciflora (7.08 and 8.56 relative lengths, respectively). As described above, this taxon was separated from S. g. guianensis.

The G_ST observed in the present study was higher than that observed with allozymes in plants that exhibit a mixed mating system (Hamrick and Godt 1996), probably because autogamy is predominant in S. guianensis (26 % of outcrossing, as described above). In general, G_ST values in autogamous or predominantly autogamous species are higher than those in allogamous species (Hamrick and Godt 1996; Maki et al. 2003). High variation values (30 %) were observed between groups of Mexican and South American accessions of Stylosanthes humilis using amplified fragment length polymorphism (AFLP) (Vander Stappen et al. 2000). The variation was even higher in AFLP studies of S. humilis (59 %) and S. viscosa (66 %) (Sawkins et al. 2001).

Stylosanthes g. microcephala and S. g. guianensis were distributed in more than one group of the STRUCTURE analysis. The genetic differentiation (G_ST) between the groups formed by S. g. microcephala was low (18 %) and, as discussed above, may be related to the soil conditions. However, the genetic differentiation among the three clusters formed mostly by the S. g. guianensis variety was 49 %, which is similar to the overall G_ST (46 %). Because most of the accessions from this variety that were included in this study were collected in the same Brazilian state, no conclusion can be drawn about the correlation between these data and the geographic origin of the accessions. More botanical and genetic studies of this variety should be conducted to determine whether it should be subdivided into multiple varieties based on the existing variation.

File 1. Fig. 1: The bar plot obtained from the model-based ancestry analysis of the S. guianensis accessions implemented in the STRUCTURE software. The numbers indicate the sample code given in Table 1.