The sampling sites are located in the Fernão Dias Environmental Protection Area (EPA), situated in a mountain range in southern Minas Gerais State, Brazil and occupy an area of 180,073ha. Samples were taken in six small fragments (area ≤15ha), of which two are also corridor fragments, and in two large fragments located in a continuous forest area, totaling eight sampled sites. Each sampled site was considered as a subpopulation (Table 1).

The Atlantic Forest was continuous in the study area until 1910, when human activity increased its footprint in the landscape. Intense and inappropriate logging along the past 60 years has produced severe habitat fragmentation. The deforestation became more intense after 1950, due to timber exploitation. These fragments decreased in size until 1985, when timber exploitation and the expansion of agriculture became less intense in the area. Currently, several fragments are well conserved, but others are still subject to selective logging and cattle grazing (Aguari, 2001).

Leaves of C. canjerana adult plants were collected randomly in the eight sampled sites. On average, 28 individuals were sampled from each subpopulation, totaling 221 individuals. Genotyping was carried out for six microsatellite loci described by Pereira et al. (2011).

Allelic richness and unbiased expected heterozygosity (HE) (Nei, 1973) were estimated for each subpopulation. The parameters intrapopulation fixation index (f), overall inbreeding coefficient (F) and the measure of genetic structure among subpopulations (θ) were estimated according to Weir and Cockerham (1984). These parameters are analogous to Wright's F-statistics, FIS, FIT and FST, respectively. These estimates were also produced using FSTAT 2.9.3.2 (Goudet, 2002) software. Slatkin's RST genetic structure parameter was estimated using RST-calc (Goodman, 1997). Recent population bottleneck was evaluated for each subpopulation using Luikart and Cornuet (1998) empirical test available in Bottleneck software (Cornuet and Luikart, 1996). The presence of null allele frequencies was checked for each loci with 104 interactions in Gibbs sampling as implemented in INEst software (Chybicki and Burczyk, 2009).

To identify the number of clusters that best explained the genetic structure of the data set, an analysis was performed using Structure 2.3.3 (Falush et al., 2003), with default parameters. K values ranging from 1 to 10 were evaluated using 20 independent runs for each K and 106 MCMC steps, assuming a burn-in period of 50,000 steps. The suggested K values were evaluated using the ΔK statistic developed by Evanno et al. (2005). The level of genetic structure among the K clusters suggested by this analysis was also estimated.

The level of spatial genetic structure among the eight subpopulations was evaluated by the Mantel test with 104 permutations using pairwise RST values and the Euclidean geographical distances matrix. The geographical distances matrix was calculated using geographical coordinates data in UTM. Geographical distance matrix, Mantel test and dispersion plots were made in R software (R Core Team, 2013).

C. canjerana subpopulations in the Fernão Dias EPA have a high level of genetic diversity. The genetic diversity estimates are in agreement with those found by Nybom (2004) in different plant tree species and are greater than those found for endemic species (Kang et al., 2008; Furches et al., 2009). The genetic diversity levels detected in Group A subpopulations were lower than those found in Group B, suggesting an effect of altitude (Fig. 2a) and that subpopulations from Group A have already suffered the effects of forest fragmentation. The intrapopulation fixation index estimates were strongly correlated to subpopulation densities (Fig. 2b). Interestingly, significant estimates of f were found in subpopulations 7 and 8 from a continuous forest area. Since C. canjerana is a dioecious species, this is probably a result of biparental inbreeding as a consequence of the spatial genetic structure within each subpopulation.

The estimate of θ=0.131 indicates a moderate level of genetic differentiation among the subpopulations. However, due to the small-scale geographical distribution of these subpopulations, the detected genetic structure can be considered strong. The similarity between the θ and RST estimates (Table 2) suggests that genetic drift is the major force shaping the genetic structure and that the dispersive effect of mutation is still of minor importance. The recent fragmentation scenario can explain the high levels of genetic diversity retained by the fragments. Franceschinelli et al. (2007), working in the same area with young plants of Myrciaria floribunda, a common Brazilian Atlantic Forest tree, also found high levels of genetic diversity.

Subpopulations from Group B are located at higher altitudes than those from Group A. The genetic differentiation within Group A was low, most likely because these subpopulations are very close to each other, favoring gene flow among them. In contrast, the genetic structure found in Group B can be related to the geographical distance and variation in altitude among these subpopulations. Although the apparent rate of gene flow in Group B was low, the genetic diversity measure in this group was high, most likely because subpopulations 7 and 8 are located in higher and better-preserved forest fragments.

The genetic structure between the two groups was high (θ=0.139), suggesting a low rate of gene flow between the subpopulations belonging to different groups (Nm=1.584). Our results suggest that the gene flow among subpopulations is affected by distance (Fig. 2c) and the landscape of the study area. These two groups are separated by a deforested valley, which explains the higher apparent levels of gene flow within each group than between them. Our results also support the hypothesis of an additional genetic isolation between these two groups beyond that expected by isolation by distance. The presence of a deforested valley separating the two groups is probably limiting the agents of seed and pollen dispersal. Primarily birds, bats and small mammals perform seed dispersal in C. canjerana and there is evidence that pollen dispersal is performed exclusively by small moths (Pizo & Oliveira, 1998). The pattern of gene flow found may reflect the dispersal behavior of these animals.

Two environmental scenarios clearly describe these eight C. canjerana subpopulations. The first scenario includes subpopulations located at lower altitudes with little landscape variation. These subpopulations are in the most severely degraded forest fragments and show reduced levels of genetic diversity. For this reason, these subpopulations deserve substantial attention by conservation programs. The second scenario includes two subpopulations located in a preserved continuous area at higher altitudes. These subpopulations are located in areas with less logging activity and maintain high levels of plant density and genetic diversity. Gene flow is still important evolutionary force acting on these subpopulations, and their survival depends on the preservation of the Fernão Dias EPA. These scenarios suggest that the high altitude subpopulations must be considered priority areas for conservation policy since they still retain the higher levels of genetic diversity. Once these areas are protected, conservation efforts that improve the gene flow mediated by pollen and seed dispersal must be stimulated, mainly on direction of subpopulations in small fragments.