Landscape and Climate Inﬂuence the Patterns of Genetic Diversity and Inbreeding in Cerrado Plant Species

: The anthropization of the landscape of the Cerrado biome that has occurred over the past few decades has fragmented its natural environments, impacting the connectivity of the plant populations and altering their gene ﬂow. Plant species may also reduce population size in response to sub-optimal climatic and environmental conditions, and observed distribution center–periphery edge genetic center we whether genetic and inbreeding coe ﬃ cients Cerrado plant species ﬀ ected by landscape features and in particular, if the distribution of the genetic diversity of these plants is consistent with the center–periphery model. To do this, we conducted a literature search for genetic studies of Cerrado plant populations using Scopus, Web of Science, and Scielo databases and the species found were used as a proxy to explore patterns throughout the biome. The data were analyzed using generalized linear mixed models (GLMM) and multiple matrix regressions (MMRRs) to evaluate the e ﬀ ects of landscape features and climatic variables on the observed (H O ) and expected heterozygosity (H E ), allelic richness ( AR ) and inbreeding ( Fis ) patterns of the local populations. The landscape was evaluated in terms of the percentage land cover of agriculture (AG), forestry (FO), remnant vegetation (RV), urban areas (UA), pasture (PA), and water (WA) within bu ﬀ ers of 1 km, 3 km, and 5 km around the study populations. We analyzed 121 populations of 31 plant species. The GLMMs showed that H O was a ﬀ ected by FO regardless of bu ﬀ er size, while H E was also a ﬀ ected by FO, but also by WA and UA. AR was a ﬀ ected by WA and UA in all three bu ﬀ er zones while the Fis was a ﬀ ected by FO and AU. The MMRRs showed that WA may a ﬀ ect H O , H E , and Fis within the 1 km bu ﬀ er, while FO a ﬀ ects H O and UA a ﬀ ects AR within the 5 km bu ﬀ er. In the case of the 1 km and 3 km bu ﬀ ers, the geographic distance between populations was identiﬁed as a factor determining the genetic diversity and inbreeding indices, populations. The GLMMs and MMRRs also showed that the mean annual temperature (MAT) and, to a lesser extent, isothermality (ISO) can explain the variation in genetic diversity observed in the Cerrado plant populations. We also found that the center–periphery model ﬁts the distribution pattern observed in most of the species evaluated, including Annona crassiﬂora, Annona coriacea , Copaifera langsdor ﬃ i , and Eugenia dysenterica . Our results indicate that changes in the climate and the landscape of Brazilian Cerrado must be considered carefully to guarantee minimizing the impacts of these processes on the genetic diversity of Cerrado plant species and ensuring the long-term conservation of these species in this biome.


Introduction
Little is known about the influence of shifts in landscape configuration on the levels of gene flow and genetic drift in plant populations [1]. The effects of habitat fragmentation are often confused with those of the matrix type and composition across a landscape [2], but previous studies suggest that open areas in the landscape may affect gene flow [3,4] and, in combination with limited dispersion capacity, can intensify distance isolation, accentuating inbreeding rates [5]. On the other hand, there is a complex relationship between habitat loss and genetic drift. The additional loss of habitat results in a threshold, after which the time needed for the allele fixation decreases rapidly [6].
Understanding the effects of landscape modifications is crucial, given that habitat loss and fragmentation are the principal drivers of the ongoing loss of biodiversity [7]. The genetic diversity of a population may decline through the loss of allelic richness, rare alleles, and heterozygosity, reaching an extreme through the fixation of alleles, indicating increased levels of inbreeding in plant populations [8][9][10], because when population fragmentation reduces effective size many ecological processes may be disrupted, such as pollination and seed dispersal, reducing contemporary gene flow and augmenting the risk of extinction due to multiple Allee effects (see [11]), characterized by correlating population size-or density-with fitness.
So, in light of this evidence, we evaluated the effects of the landscape matrix on the genetic diversity and the levels of inbreeding in the plants of the Cerrado savannah biome. This biome was chosen because, although it is the largest savannah in the world, with an enormous complexity of habitats [12], its landscape has suffered profound transformations over the past few decades. Originally covering much of Brazil's central plateau [13], this savannah has gradually been replaced by farmland, pasture, and urban zones over the past 50 years [14]. Between 2003 and 2013, in fact, the proportion of the biome covered by agricultural land almost doubled, with 74% of this area being established on intact Cerrado vegetation [15]. By the end of the 20th century, the extensive impacts that threaten the survival of endemic species and the maintenance of ecosystem services, such as stabilizing the water regime and providing refuge for many species (see [16]), led to the inclusion of the biome in the world's biodiversity hotspots [17]. The fragmentation and loss of habitat increase spatial isolation, reducing the size of populations and interrupting their connectivity through limitations to dispersal [18], which ultimately leads to a reduction in gene flow and a subsequent decline in genetic diversity [19]. Previous studies have shown this relationship between fragmentation and increased levels of inbreeding in plant populations in the Cerrado, e.g., [20,21]. In addition to landscape processes, research has shown that plants may also be affected directly by climate change.
Climate influences genetic diversity by inducing changes in the distribution of species and acting as a selective factor, with adverse conditions acting as effective barriers to colonization and gene flow, e.g., [22,23]. Collevatti et al. [24] showed that climatic change may affect the distribution and genetic diversity of Caryocar brasiliensis in the Cerrado biome. However, phylogeographic patterns tend to be species-specific, rather than universal [25][26][27][28][29], which led us to decide to evaluate the effects of current climate conditions on the genetic diversity and inbreeding coefficients of the Cerrado populations of different plant species. Despite the taxon specificity of these patterns, we also evaluate the hypothesis that the patterns of distribution of the genetic diversity of these plants are consistent with the center-periphery model (see [30][31][32][33][34]).
Given this, we propose the hypothesis that the Cerrado plant species have higher genetic diversity and lower inbreeding towards the geographical center of the biome, based on the convention that population size decreases and spatial isolation increases from the center to the periphery of a species' range, often because of the decreasing quality of habitats toward the edge of the range [19,35]. In this scenario, demographic bottlenecks, varying selection pressures, and restricted gene flow will result in genetic impoverishment, pronounced genetic differentiation among peripheral populations, and substantial divergence from more central populations [34,[36][37][38].
Reliable data on the genetic diversity and population structure of species, and the factors that determine their variation within the species' ranges have proven to be crucial for planning conservation strategies [39]. The value of conservation and the allocation of resources for the preservation of marginal species are still subjects of ample debate [40]. However, it is clear that strong selective pressures caused by stressful conditions can support the acquisitions of new adaptations in marginal areas, conferring a distinct evolutionary potential on these populations [41]. Consequently, our main objective was to evaluate whether the distribution patterns of the genetic diversity of plants in the Cerrado are consistent with the center-periphery model, contributing to understand the population dynamics of these species. Alternatively, concerned with the high fragmentation rate of this biome, we seek to understand whether the genetic variation patterns of native plants can be affected by landscape and climate characteristics, within a conservationist perspective.

Literature Search of Genetic Studies on the Vascular Plants of the Cerrado
The data were obtained through an extensive survey of the scientific literature, and for each population (or species) identified, data were obtained on observed heterozygosity (H O ), expected heterozygosity (H E ), allelic richness (AR), and the inbreeding coefficient (Fis) [42,43], together with the geographical coordinates of the population, thus, the selection of working species depended only on the availability of genetic data. This literature search focused on Scopus, Web of Science, and Scielo databases, which are considered to be the most reliable and comprehensive sources of bibliographic information available online, with Scielo being the standard for Brazilian papers. The search was based on the keywords "SSR (Simple Sequence Repeats) or microsatellite" (according to the method proposed by Vitorino et al. [44]) and "Cerrado". We chose to use data from microsatellites (nSSRs) because they reflect recent gametic gene flow. Data obtained from the same species, but recovered from different papers were combined for analysis. Species of the same genus, for which we found few population studies, were also combined for analysis. For this combination, we make sure that the combined species have the same life history traits and are therefore subject to similar selective pressures.

Retrieving Data on Landscape Features
Landscape features were defined using a map of land use and cover of the Cerrado, based on a shapefile available at http://www.dpi.inpe.br/tccerrado/ in 2013 [45]. The map was scanned at a default scale of 1:2,500,000 in ArcGis 10.5 (ESRI), with six classes of land use and cover being recognized (Table 1). The coordinates obtained for each population (defined by the sampling area) were plotted on the map of the Cerrado, and landscape buffers were created with radii of 1 km, 3 km and 5 km around Diversity 2020, 12, 421 4 of 20 each population. These buffers were designed to cover the relatively short pollination distances known for some Cerrado species-see [46,47]-and the relatively long distances covered by seed dispersers in this biome [48]. They also provide insights into the possible effects of spatial scale on the observed genetic diversity. The extract by mask tool (in ArcGis) was used to collect the data separately from each buffer, which permitted the identification the relative quantities of the different classes of land use and cover. These quantities were transformed into percentages corresponding to each class. The correlation between classes of variables was assessed using Pearson's correlation. Variables were excluded from the analysis whenever the correlation was equal to or greater than 0.5 see [49].
Alternatively, the landscape matrix was characterized in the area where each population occurred, because we wanted to know if genetic data were obtained from populations mapped to areas of natural vegetation, pasture, forestry, amongst others.

Retrieving the Data on Landscape Features
A total of 19 bioclimatic variables were retrieved from the ecoClimate database for analysis in the present study [50]. Variables were chosen for a pre-industrial time-frame (1720-1800) to represent the current climatic conditions, obtained through the atmosphere-ocean general circulation model CCSM4 (CMIP5/PMIP3-http://cmip-pcmdi.llnl.gov/cmip5/ and https://pmip3.lsce.ipsl.fr/). These variables were plotted on a grid of the Neotropical region in cells of 0.5 • × 0.5 • (longitude vs. latitude), also obtained for each population of climatic data relating to its occurrence cell. Subsequently, the correlation between the variables was also evaluated by Pearson correlation (>0.5) to avoid colinearity. Therefore, only seven of the bioclimatic variables evaluated were maintained ( Table 2).

Spatial Patterns of Genetic Diversity and Structure
Spatially explicit analyses were used to detect central-peripheral patterns in the genetic diversity of the plants evaluated in the present study. This analysis included only species represented by more than three populations. The coordinates of each population were plotted in ArcGis, and the geographical distance (in km) of these populations to the center of the Cerrado was estimated based on a line projected in the editor tool, using the near command. We use a line across the biome to represent the center, considering the longitudinally observed heterogeneity, as it would not be represented by a single geographical point. We chose to evaluate populations in relation to the center of the biome because we were interested in testing whether the instabilities observed at the edge could affect the genetic diversity of populations, since the transition areas are identified as areas of tension see [51]. The relationship between the parameters of genetic diversity (H E , H O , AR, and Fis) and the distance to the centroid of the Cerrado was analyzed using quantile regression [52]. Quantile regression is a way to estimate the conditional quantiles of a response variable distribution in the linear model that provides a more complete view of possible causal relationships between variables in ecological processes.

Effects of Landscape and Climate on the Genetic Diversity of Cerrado Plants
The effects of landscape features and climatic variables on the H E , H O , AR, and Fis, were evaluated using generalized linear mixed models (GLMM). We used the type of matrix, the percentage of each landscape attribute (remaining vegetation, agriculture) within each buffer, and climatic parameters as the explanatory variables. As the landscape data were collected within a grid of 0.5 • × 0.5 • , the influence of these variables was evaluated directly through the occurrence of populations in the cells, and not in the buffers. The landscape and the identity of the species were included as random factors, and the explanatory variables as fixed factors, with the construction of the models used being represented by the general equation: where y is a N × 1 column vector, the outcome variable; X is a N × p matrix of the p predictor variables; β is a p × 1 column vector of the fixed-effects regression coefficients; Z is the N × q design matrix for the q random effects; u is a q × 1 vector of the random effects; and ε is a N × 1 column vector of the residuals.
We also constructed a model to determine whether random effects could also account for the genetic response variables. This model considered the random effects on the identity of the species and the type of landscape matrix only, discarding the effects of climatic variables and landscape attributes. Alternatively, we also consider a general model, which incorporates all the variables considered (full model). The analyses were run in the MCMCglmm package [53] implemented in R 3.6.1 [54], in a Bayesian framework with the Markov Chain Monte Carlo algorithm. We used a total of 80,000 iterations with 20,000 burn-in chains and a Gaussian distribution. We used the Akaike information criteria (AIC) to select the best model, that is, the model with the smallest AICc (the AIC corrected for sample size and the number of parameters), which was considered to be the most plausible for the explanation of the observed patterns [55]. The Delta AICci (∆AICci, where i represents each model) was calculated as the difference between the AICc for the ith model and the smallest AICc observed. We also determined Akaike's weight (wAICc), which represents relative contribution of the ith model to the explanation of the observed pattern, given a set of competing models. Models with ∆AICc < 2 were all considered equally plausible as explanations of the observed pattern [56].
We also ran a multiple matrix regression with randomization (MMRR) analysis [57] to evaluate the effects of the geographic distance between pairs of populations and the variation in landscape (e.g., percentage of remnant of native vegetation and agricultural areas), as well as climate differences in the areas where the populations occur on their genetic diversity and inbreeding indices. This analysis was run in the popgenreport package [58].

Results
In the present study, literature data were extracted for 31 plant species, totaling 122 populations distributed throughout the Cerrado (Table 3 and Table S1

Spatial Patterns of Genetic Diversity and Structure
We found a clear center-periphery pattern in the spatial distribution of the genetic parameters in Anonna coriacea and A. crassiflora (Figure 1). The triangular pattern observed for the quantiles reveals that the genetic diversity (H O , H E , and AR) of these species is higher in the center of the biome. As expected, additionally, inbreeding rates (Fis) are higher at the edge of the biome, with reduced Diversity 2020, 12, 421 7 of 20 genetic diversity in these areas. To visualize the trends and the significance of each regression by species, see Table S2.

Spatial Patterns of Genetic Diversity and Structure
We found a clear center-periphery pattern in the spatial distribution of the genetic parameters in Anonna coriacea and A. crassiflora (Figure 1). The triangular pattern observed for the quantiles reveals that the genetic diversity (HO, HE, and AR) of these species is higher in the center of the biome. As expected, additionally, inbreeding rates (Fis) are higher at the edge of the biome, with reduced genetic diversity in these areas. To visualize the trends and the significance of each regression by species, see Table S2.  In Campomanesia adamantium, we observed higher genetic diversity values (H O , H E , and AR) in the populations located closer to the edge of the biome. Given this, we found lower levels of inbreeding in the edge regions ( Figure S1).
In C. langsdorffii, as in Annona, genetic diversity tended to decrease with the increasing distance of the populations from the centroid of the Cerrado and, as expected, higher inbreeding rates were observed in the more outlying populations ( Figure S2).
When we analyzed the data on Dimorphandra mollis, we found no clear evidence of a center-periphery pattern and there is a divergence in relation to the quantile 0.99 for H O ( Figure S3). For H E , the pattern was consistent with an increase in this index, in populations located closer to the Cerrado border. Thus, a pattern similar to H E was observed for AR and also for Fis.
In the case of the data available on Dipteryx alata, we observed a center-periphery pattern only in H E , that is, the expected heterozygosity showed a tendency to decrease with increasing distance of the populations from the biome's centroid, which was demonstrated by the 0.99 quantile. The quantiles indicated different patterns for the H O and AR data ( Figure S4), although no systematic variation was found in the Fis.
The populations of Eugenia dysenterica presented a clear center-periphery pattern in their genetic diversity, with the H O , H E , and AR values all decreasing with increasing distance of the population from the Cerrado centroid ( Figure S5). However, the inbreeding coefficient does not follow the expected pattern, with one quantile showing an increase in the Fis at the edge of the Cerrado, whereas the other reflected a decrease in this coefficient toward the edge of the biome Fis.
A relatively well-defined central-peripheral pattern was also found in the genetic diversity of the Hancornia populations, with the H O and AR values indicating significant structuring of the genetic diversity of these species. This is reflected in the higher inbreeding rates found in the more peripheral populations in the biome ( Figure S6).
We analyzed the genera Handroanthus and Tabebuia together, given that their species represent the Tabebuia Alliance clade of Grose and Olmstead [59]. The H O and H E values presented a clear center-periphery pattern, which was reinforced by the higher Fis values observed in the populations at the edge of the Cerrado biome ( Figure S7). We also observed a clear center-periphery pattern in the H E and AR data obtained for Manihot esculenta, although the highest Fis values are highest in the populations located in the center of the biome. While this contradicts the pattern observed in the H E and AR, it is consistent with the H O , which was higher toward the edge of the Cerrado ( Figure S8).
The center-periphery pattern was also clear in Oryza glumepatula, with the indices of genetic diversity (H O , H E and AR) decreasing with increasing distance from the center of the biome, while the Fis increased with the increasing distance of the populations, at least to the quantile 0.99 ( Figure S9).
By contrast, in the case of the genus Qualea, one quantile indicated a tendency for H O and AR to decrease with increasing distance from the centroid, although this is consistent with the inbreeding coefficient, which tended to be higher in populations over the edge ( Figure S10).
In the genus Solanum, we verified a clear center-periphery pattern only in the AR, which decreased with increasing distance from the center of the Cerrado ( Figure S11).
We observed a clear center-periphery pattern in the H E and AR of the Vellozia gigantea populations, with genetic diversity decreasing with increasing distance from the Cerrado centroid. However, the inbreeding coefficient also followed this pattern, that is, it decreased with increasing distance from the centroid, which is consistent only with the pattern observed in H O , which increased towards the periphery of the biome ( Figure S12).
When we analyzed the whole dataset, that is, all the plant species together (Figure 2

Effects of Landscape and Climate on Genetic Diversity
When we analyzed the landscape effects, the models were only significant when we excluded the model that incorporates only the species and landscape matrix. When this model was excluded, the percentage of forestry (FO) in the landscape had a significant effect on the HO patterns in all three buffers, with ∆AICc = 0.0 in all cases, and wAIC = 0.7720 at 1 km, wAIC = 0.6330 at 3 km, and wAIC = 0.6410 at 5 km (Table 4)

Effects of Landscape and Climate on Genetic Diversity
When we analyzed the landscape effects, the models were only significant when we excluded the model that incorporates only the species and landscape matrix. When this model was excluded, the percentage of forestry (FO) in the landscape had a significant effect on the H O patterns in all three buffers, with ∆AICc = 0.0 in all cases, and wAIC = 0.7720 at 1 km, wAIC = 0.6330 at 3 km, Diversity 2020, 12, 421 9 of 20 and wAIC = 0.6410 at 5 km (Table 4). By contrast, H E was related to the percentage of water (WA) in the 1 km buffer (∆AICc = 0.0; wAIC = 0.9120). The incorporation of WA in the model was also more likely to explain H E in the 3 km buffer (∆AICc = 0.0; wAIC = 0.3300), although in this case, other environmental variables were also capable of explaining the patterns of this index, including VR (∆AICc = 1.3; wAIC = 0.1700), FO (∆AICc = 1.5; wAIC = 0.1590), and AG (∆AICc = 1.6; wAIC = 0.1450). Similarly, for the 5 km buffer, WA provided the best explanation for the observed pattern (∆AICc = 0.0; wAIC = 0.3550), although it was matched by FO (∆AICc = 0.6; wAIC = 0.2680) and AG (∆AICc = 1.7; wAIC = 0.1500). In the analysis of the landscape effects on the allelic frequency (AR) in the Cerrado plant species, the model that includes the percentage of urban area (UA) best explains the pattern observed in all three buffers, with ∆AICc = 0.0 in all cases, and wAIC = 0.3820 at 1 km, wAIC = 0.3350 at 3 km, and wAIC = 0.3600 at 5 km (Table 5). However, the percentage of water (WA) also had a significant effect on the AR, with ∆AICc = 0.5 and wAIC = 0.3000 at 1 km, ∆AICc = 0.9 and wAIC = 0.2050 at 3 km, and ∆AICc = 0.5 and wAIC = 0.2800 at 5 km (Table 5). In the 1 km buffer, the third most important model was that of pasture (PA), with ∆AICc = 1.4 and wAIC = 0.1930, whereas at 3 km, it was farmland (AG), with ∆AICc = 1.1 and wAIC = 0.1920, and at 5 km, it was forestry (FO), with ∆AICc = 1.3 and wAIC = 0.1920. The inbreeding coefficient (Fis) was most affected by the percentage of FO in the landscape assessed in the 1 km buffer (∆AICc = 0.0 and wAIC = 0.6100), whereas the percentage of UA was the most important variable at 3 km (∆AICc = 0.0; wAIC = 0.6920) and 5 km (∆AICc = 0.0; wAIC = 0.5190). Table 5. Models used to test the hypothesis of the influence of landscape features, that is, the percentage of agriculture (AG), water (WA), urban areas (UA), remnant vegetation (RV), pasture (PA), and forestry (FO), on the allelic richness (AR) and inbreeding coefficient (Fis) recorded in 121 plant populations in the Cerrado biome of central Brazil. Analyses were run using 1 km, 3 km and 5 km buffers. The models with ∆AICc < 2.0 are highlighted in bold type. AICc = AIC corrected by sample size and number of parameters in the model; wAIC = Akaike weight. In general, FO, WA, and UA affected the genetic diversity of the plant populations evaluated in the present study ( Figure 3). When we related landscape parameters to the genetic diversity data, it was possible to determine that populations with low H O , H E and AR values and high inbreeding coefficients tend to be associated with highly modified landscapes. For example, the populations of O. glumaepatula, are surrounded by a landscape dominated by pasture, while those of M. esculenta occur in areas dominated by urbanization, farmland, and pasture, with few vegetation remnants. A similar pattern was observed in D. alata. In Annona and Handroanthus/Tabebuia, by contrast, high percentages of remnants of natural vegetation were associated with high H O , H E , and AR values.  When we evaluated the effects of the climatic variables associated with the landscape matrix, the mean annual temperature and the isothermality were the principal factors determining the observed genetic patterns. The mean annual temperature MAT has a significant effect on both the H O (∆AICc = 0.0; wAIC = 0.9949) and the H E (∆AICc = 0.0; wAIC = 0.7116), whereas the AR was affected by the combination of all the climate variables analyzed, with ∆AICc = 0.0 and wAIC = 0.7116 ( Table 6). The patterns of inbreeding can be accounted for by ISO (∆AICc = 0.0; wAIC = 0.3530), MAT (∆AICc = 0.8; wAIC = 0.2380), and also by the model that incorporates only the species and landscape matrix (∆AICc = 1.3; wAIC = 0.1870). Table 6. Models used to test the hypothesis of the influence climatic parameters, that is, the mean annual temperature (MAT), isothermality (ISO), annual temperature range (ATR), mean temperature of driest quarter (TDQ), annual precipitation (AP), precipitation seasonality (PS), and precipitation of the warmest quarter (PWQ), on the observed (H O ) and expected heterozygosity (H E ), the allelic richness (AR), and the inbreeding coefficient (Fis) recorded in 121 plant populations in the Cerrado biome of central Brazil. Analyses were run using 1 km, 3 km and 5 km buffers. The models with ∆AICc < 2.0 are highlighted in bold type. AICc = AIC corrected by sample size and number of parameters in the model; wAIC = Akaike weight. Legend: K = number of parameters that consider the β of the explanatory variables and the distribution parameters of the residuals.
When we analyzed the influence of the landscape and climatic characters, together with the geographical distance between populations, on the genetic diversity variables, we found that the percentage of WA was the most important variable, in the 1 km buffer, related to the differences in genetic diversity observed among the populations (Table 7). In the 3 km buffer, the geographic distance affected the H O and Fis, although the H E and AR were not influenced by any of the factors analyzed.
In the 5 km buffer, geographic distance affected H O and Fis, although the H O was also affected by the percentage of FO, whereas AR was affected by the UA on the landscape. In the case of the climatic variables, the results were similar to those observed in GLMM, that is, the differences observed in the MAT and isothermality (ISO) in the areas occupied by the populations contribute to the variation observed in the genetic diversity. Table 7. Models tested by the multiple matrix regression with randomization (MMRR), based on the hypothesis that the geographic distance (Geo distance) between populations, landscape parameters (agriculture, AG; water, WA; urban areas, UA; remnant vegetation, RV; pasture, PA, and forestry, FO), and the climatic variables, that is, mean annual temperature (MAT), isothermality (ISO), annual temperature range (ATR), mean temperature of driest quarter (TDQ), annual precipitation (AP), precipitation seasonality (PS), and precipitation of the warmest quarter (PWQ), account for the pattern of variation observed in the genetic diversity (H O , H E ), allelic richness (AR), and the inbreeding coefficient (Fis) in 121 plant populations in the Cerrado biome. The analyses were run using buffers of 1 km, 3 km, and 5 km.

Discussion
The central-peripheral model of genetic diversity was in fact observed in many of the plant species evaluated in the present study, which tended to have higher heterozygosity and allelic richness in the more centrally-located populations, which also tended to have lower levels of inbreeding. This pattern likely reflects a decline in the adaptability of the plants to the abiotic conditions found toward the external limits of the biome. The marginal areas of a species' range tend to be a zone of transition, where selection pressures are generally more intense [31,60,61]. In this case, species with less phenotypic plasticity will become more stressed physiologically, e.g., [62], which would result in smaller populations, with reduced genetic diversity driven by low gene flow, genetic drift, inbreeding, and directional selection, leading to a marked genetic structure, e.g., [63][64][65][66]. Marginal areas have a number of challenges, including unfavorable abiotic conditions and competition with other species from other biomes [67], and the reduced fitness of the plant may lead to larger fluctuations in population size, reducing effective population size in comparison with larger or more stable populations [68][69][70][71].
The distribution of both Anonna crassiflora and A. coriacea is associated with typical Cerrado environments, that is, nutrient-poor latosols [72]. Hancornia speciosa occurs in the cerrado and cerradão physiognomies, and is thus also associated with poor soils [73], and a similar distribution pattern was observed in E. dysenterica, which inhabits poor, sandy, and acidic soils, predominantly in the Cerrado region and on coastal plains [74]. These preferences, combined with the modifications provoked by the changes in climate occurring during the Pleistocene, have influenced the distribution patterns of these species in the Cerrado biome, in particular, the greater concentration of genetic diversity in the centroid of the biome. Correa Ribeiro et al. [75] concluded that the area to the north of the center of the Cerrado, in the central Goiás highlands, acted as a refuge for the A. coriacea populations during the Pleistocene. Collevatti et al. [76] also showed that a large area of central Brazil, which coincides approximately with the central Cerrado, served as a historical refuge for the H. speciosa populations during long-term climate change.
The patterns observed in the heterozygosity of the Cerrado plant populations were influenced by the matrix in which the populations were located, with the greatest effects being observed in the landscapes dominated by FO, WA, and UA. However, the effect appears to be positive when the natural vegetation is replaced by areas of preservation and restoration of native forests, rather than farmland or pasture. This is consistent with the fact that woody plants, especially fruiting trees, will attract pollinators and dispersers. Pollinators, in particular, will transit between cultivated areas and natural environments [77], and the natural landscape provides an important refuge for a diversity of pollinators which provide pollination services in forestry plantations [78]. Dispersing animals may be attracted to fragments of vegetation close to forestry plantations where they can obtain food. There is also considerable evidence that the crop matrix can provide habitats that support many animal species [79]. These movements of pollinators and dispersers between different habitats within the matrix may thus contribute to the genetic diversity of the plant populations sampled near these areas.
In buffers where the percentage of WA was greater, H E tended to decrease, which is consistent with barrier isolation models. Aquatic environments can create discrete barriers, such as waterfalls and reservoirs, which reduce gene flow between the populations located at their margins, e.g., [80]. However, the river's slope and the physical-chemical dissimilarities of water can also act as barriers to gene flow and cause detectable differentiation in the genetic constitution of different populations [81,82].
Urban areas also appear to affect the genetic diversity of the plant species evaluated. Some populations sampled in buffers with a high percentage of the urban development presented higher levels of diversity, such as C. brasilienses and T. aurea. Urban areas surrounded by fragments of vegetation can function as barriers to gene flow [83], although they may also facilitate dispersion among populations, leading to greater genetic diversity overall, and reduced differentiation between populations of species that are attracted to urban areas [84]. Many native pollinators, such as bats and bees, may benefit from urban forest resources [85,86], but they usually return to the fragments of native vegetation associated with these areas, thus increasing the flow of pollen between fragments, or even between the urban zone and the fragments. The C. brasilienses populations found in urban areas may be pollinated by the glossophagine bats, Glossophaga soricina and Anoura geoffroyi [87], while T. aurea is pollinated by bees of the genus Centris [88]. These populations can serve as a source of alleles for populations located in fragments of natural vegetation in the area surrounding the urban zone.
Geographical distance affected the variation in the H O and Fis values recorded in the present study, indicating that distance is an important landscape component in the determination of the genetic diversity and structure of plant populations in the Cerrado. These findings are consistent with the previous studies that have verified the occurrence of isolation by the distance between populations of a number of Cerrado plant species, e.g., [89][90][91][92].
The mean annual temperature (MAT) was the principal climatic variable influencing the H O , H E , and Fis, although isothermality (ISO) also affected genetic diversity. Bonte et al. [93] showed that temperature can determine the limits of the geographical ranges of a species, principally by affecting the ability of the plants and juveniles to disperse and become recruited successfully. The gametocytes are especially sensitive to fluctuations in temperature, both during their development, before pollination, and in the post-pollination stage [94]. To avoid negative impacts on their reproduction and physiology [95], plants tend to disperse to regions that correspond to their optimal temperature range [96]. The more central regions of the biomes in which these species occur may encompass this optimal temperature range and other abiotic conditions appropriate for the species, possibly reflecting a relationship between optimal suitability and the typical climatic conditions of the biome. Our model thus supports the hypothesis that current and future climate change will impact the genetic diversity of many plant populations [97][98][99], given that some species, especially those that are already threatened or restricted to habitats created by climate change, are vulnerable to the loss of specific alleles through inbreeding [100]. Combined with climate change, landscape modifications need to be considered carefully in order to guarantee the survival and conservation of the genetic diversity of plant species in Cerrado ecosystems, since they represent the main mechanism for maintaining the food webs.
This study includes patterns for a few species in the Cerrado biome, or even for a few populations of these species, which may be improved in the future, if genetic data for other populations or other plants are available in the literature. Likewise, updated information on land use and land cover in this biome will allow us in the future to establish levels of change at different points in the landscape and assess whether there is a real loss of contemporary gene flow in the most modified areas over time.

Conclusions
The distribution of the genetic diversity of many Cerrado plant species follows a typical centerperiphery pattern, with greater heterozygosity and lower levels of inbreeding in the populations located more toward the center of the biome. These plants are affected directly by the landscape matrix in which they are located, and in particular by the percentage of forestry cover, water, and urban areas. As modifications to the matrix may affect gene flow, they will ultimately determine the levels of genetic heterozygosity and inbreeding found in a population. The mean annual temperature also influenced genetic diversity in the Cerrado plant species, indicating the influence of the optimal temperature range for the occurrence and genetic diversity of the species. Overall, the results of the study indicate the need for the systematic management of the effects of both changes in climate and shifts in landscape characteristics, in order to minimize the impacts on the genetic diversity of Cerrado plant species.
Supplementary Materials: The following are available online at http://www.mdpi.com/1424-2818/12/11/421/s1, Table S1: Species or subspecies, number of microsatellite locus and reference article used to obtain the genetic parameters: Observed Heterozygosity (H O ), Expected Heterozygosity (H E ), Allelic Richness (AR), and inbreeding coefficients (Fis) for Cerrado plant populations. Figure S1: Distribution of genetic diversity indices and the inbreeding coefficient for the Campomanesia adamantium populations sampled in the Cerrado biome. Figure S2: Distribution of the genetic diversity indices and inbreeding coefficients of the Copaifera langsdorffii populations sampled in the Cerrado biome. Figure S3: Distribution of the genetic diversity indices and inbreeding coefficients for the Dimorphandra mollis populations sampled in the Cerrado biome. Figure S4: Distribution of genetic diversity indices and inbreeding coefficients of the populations of Dipteryx alata sampled in the Cerrado biome. Figure S5: Distribution of the genetic diversity and inbreeding coefficients of the Eugenia dysenterica populations sampled in the Cerrado biome. Figure S6: Distribution of the genetic diversity indices and the inbreeding coefficients of the populations of Hancornia speciosa cuyabensis, Hancornia speciosa gardinerii, Hancornia speciosa pubescens, and Hancornia speciosa speciosa sampled in the Cerrado biome. Figure S7: Distribution of the genetic diversity indices and inbreeding coefficients of the populations of Handroanthus chrysotrichus, Handroanthus serratifolius, Handroanthus impetiginosus, Tabebuia aurea, and Tabebuia roseoalba sampled in the Cerrado biome. Figure S8: Distribution of the genetic diversity indices and inbreeding coefficients of the of Manihot esculenta populations sampled in the Cerrado biome. Figure S9: Distribution of the genetic diversity indices and inbreeding coefficients of the Oryza glumaepatula populations sampled in the Cerrado biome. Figure S10: Distribution of the genetic diversity indices and inbreeding coefficients off the populations of Qualea grandflora, Qualea multiflora, and Qualea parviflora sampled in the Cerrado biome. Figure S11: Distribution of the genetic diversity indices and inbreeding coefficients for the populations of Solanum crinitum and Solanum lycocarpum sampled in the Cerrado biome. Figure S12: Distribution of the genetic diversity indices and inbreeding coefficients recorded in the populations of Vellozia gigantea sampled in the Cerrado biome. Table S2: Values referring to parameter β (b 1 ) and significance (P)