GENETIC DIVERSITY OF BRAZILIAN PEPPER BASED ON QUALITATIVE REPRODUCTIVE TRAITS

In Brazil, pepper cultivation gives producers more marketing opportunities in various segments, either for fresh consumption or industry. Reproductive traits are efficient methods for comparing genetic divergence in pepper, once most differences among accessions could be verified at this stage. This study aimed to evaluate the genetic divergence in the reproductive phase of pepper accessions by multivariate analysis and test the efficiency of different methods. Sixty-five genotypes from the Federal University of UberlândiaCapsicum spp. Germplasm Bank were evaluated 145 days after sowing, and morphological characterization was performed at reproductive phase. Graphical representation of genetic distances was obtained by UPGMA. Tocher optimization method was also used to group accessions. Genotypes were arranged in seventeen different groups by Tocher method. UPGMA dendrogram collaborated with Tocher method, indicating the wide genetic variability of genotypes. Tocher and UPGMA methods were partially in agreement, allowing grouping 44 of the 65 analyzed accessionsequally. Multicategorical traits have the advantage of easy observation and require less time and labor, being ideal for use in gene bank and collections that do not have high human and financial resources. Moreover, these traits are not affected by the environment. Genetic divergence detected in this paper encourages other researchers to perform the characterization of pepper collections as completely as possible, because it becomes possible to generate more reliable information of the variability and genetic divergence among accessions. Our research discloses the reproductive biodiversity of pepper in “Alto Paranaíba” and “TriânguloMineiro” regions and the importance of maintaining these genotypes.


INTRODUCTION
Pepper belongs to the genus Capsicum and family Solanaceae. It is estimated that the world's Capsicum production is over 30 million tons in almost four million hectares. India is the major producer and consumer (Faostat, 2016). In Brazil, pepper cultivation gives producers more marketing opportunities in various segments, either for fresh consumption or industry (Signorini et al., 2013). The importance of pepper cultivation is directly associated with its consumption, being considered one of the most important species used by family farmers (Costa et al., 2015;Signorini et al., 2013). According to ABCSEM (2016), more than 3,322 kg of pungent and not pungent hot pepper seeds were sold in Brazil in 2011, with an estimate of 2,460 ha planted area.
However, unlike other countries with higher productions, pepper cultivation in Brazil is restricted to low diversity of varieties available to farmers, mostly occurring with informal marketing. In addition, with the growth of large urban centers,of the number of small farmershas reduced, resulting in the extinction of many Capsicum genotypes, narrowing the genetic base (Domenico et al., 2012). Several researches onCapsicum have evaluated aspects related to fruit productivity (Büttow et al., 2010;Domenico et al., 2012;Moura et al., 2010;Paulus et al., 2015), but there is little database with descriptions of reproductive features in a pepper germplasm. Reproductive traits are efficient methods for comparing genetic divergence in pepper, once most differences among accessions could be verified at this stage (Bianchi et al., 2016). Qualitative traits are little influenced by the environment, allowing an efficiency genetic distinction among accessions. Furthermore, classification is simpler and faster compared to quantitative methods, and several qualitative descriptors have already been proposed for the species (IPGRI, 1995).
Efficient description of accessions in a gene bank can define the future of the genetic diversity of a species (Costa et al., 2015). In Brazil, few studies had reported the genetic diversity of pepper, compromising the possibility of interesting crosses for plant breeding. Moreover, there is no consensus regarding the best multivariate analysis method able to characterize the reproductive phase of Capsicum spp. accessions. Therefore, this study aimed to evaluate the genetic divergence in the reproductive phase of pepper accessions by multivariate analysis and test the efficiency of different methods.

MATERIAL AND METHODS
Sixty-five genotypes from the Capsicum spp. Germplasm Bank were evaluated at the Federal University ofUberlândia-UFU, Campus of Monte Carmelo, MG, Brazil (Table 1). Currently, the bank has 250 accessions that have been acquired through donation, collection and purchase in farmer's markets. All analyses were conducted at the Laboratory of Seed and Genetic Resource Analysis, UFU (LAGEN). This genetic bank includes part of the pepper biodiversity of "Alto Paranaíba" and "TriânguloMineiro" regions of Minas Gerais that is home of small-scale pepper producers.
The experiment was conducted at the Horticultural Experiment Station, (18º42'43"S, 47º29'55"W and 873 ma.s.l., humid temperate climate with hot summers and dry winters). Sowing was performed on September of 2013, in 128-cell polystyrene trays filled with commercial coconut-fiber substrate. After sowing, trays were kept in gableroofed greenhouse covered by transparent 150-micron polyethylene film activated against ultraviolet rays with 30% shade. Seedlings were transplanted 54 days after sowing, with twenty seedlings of each accession planted in four rows, spaced one meter between lines and 0.7 meter between plants, with total experimental area of 1,316 m 2 . Four replicates of each accession were planted in a randomized block design (RBD).
Graphical representation of genetic distances was obtained by Unweighted Pair-Group Method using Arithmetic Average (UPGMA). The NbClust package from the R software was used to determine the best number of clusters, based on 24 indexes that establish the compactness and separation criteria among clusters. Tocher optimization method was also established to group accessions. Cophenetic correlation coefficient (CCC) was performed for UPGMA and Tocher methods to identify the clustering quality of both. Mean Decrease Impurity (MDI) was measured to evaluate the importance of variables and Cramer's V measured the correlation among traits with Chi-Square test (x 2 ) to validate correlations, with 0.05 significance. Data were analyzed with R software, using clusters packages to estimate dissimilarity matrix, random Forest to measure MDI (Liaw and Wiener, 2002); NbClust for the number of clusters (Charrad et al., 2014); stats for UPGMA, its CCC and Pearson's Chi-Square; biotools for Tocher and its CCC (Silva and Dias, 2013). Graphic artwork was performed with the SigmaPlot® software (version 11.0).

RESULTS AND DISCUSSION
From the 16 evaluated traits, only seed color did not change among genotypes, being all categorized with yellow seeds. This variable also did not differ among 40 pepper accessions from Amazonas-Brazil, studied by Costa et al. (2015). The other 15 traits were capable of detecting genetic divergence among accessions. Vasconcelos et al. (2012) also obtained efficient distinction among 22 Capsicum chinense accessions, with ten qualitative descriptors used in our work.
Gower's algorithm was capable of detecting divergence among genotypes. This methodology was widely used in dissimilarity papers with qualitative and quantitative traits (Silva et al., 2016), including pepper (Moura et al., 2010;Costa et al., 2015). However, Knezovic et al. (2005) warned that this approach is not suitable, because the distance between two accessions will immediately be 0 or 1; but also, questionable, because variables will be unequally weighted, where in a multi-level situation, more weight will be given to variables with more levels, while on a single level incompatibilities will outweigh correspondences.
Genotypes were arranged in 17 different groups by the Tocher optimization method ( Table 2), showing that these descriptors were efficient to detect genetic divergence in Capsicum accessions, justified by the large variability at reproductive stage for that genus. Six clusters had only one genotype. Other five clusters had two genotypes. These clusters could be used to search genotypes with different traits to improve genetic diversity among crosses. On the other hand, if the aim of crossingsis to keep high fruit quality (for example, dark red fruit and small seeds), Group 1 and 2 detected large number of genotypes that would be recommended for that case. UPGMA dendrogram collaborated with the Tocher method, indicating the wide genetic variability of genotypes (Figure 1). Thirteen of the 24 indexes from NbClust package indicated ideal formation of 15 clusters for the data, which helped for the cut of 37.3% of similarity on the graphic, generating 14 different groups. UPGMA also isolated UFU-35, UFU-69, UFU-32, UFU-04 and UFU-75. UFU-48 and UFU-10 genotypes once they were paired with other genotypes at Tocher's optimization. UFU-37, isolated at Tocher's, was grouped with UFU-39, UFU-54 and UFU-68, showing to be genotypes with close phenotyping in relation to UFU-37. Larger clusters had 20 and 13 genotypes, respectively, showing that UPGMA created larger groups than Tocher. Tocher and UPGMA grouping methods were partially in agreement, allowing grouping 44 of the 65 analyzed accessionsequally. The same results were obtained by Bento et al. (2007), which grouped 19 of the 28 analyzed chili pepperaccessions equally. Tocher and UPGMA methods were also partially coincident in the separation of 137 pepper genotypes from SouthwesternGoiás (Alvares et al., 2012).
Cophenetic correlation coefficient (CCC) can be used to compare clustering results of the same data set, using different distance measures or clustering algorithms. In general, CCC is a measure of how accurately a cluster method preserves the pairwise distances among individuals (Kumar & Toshniwal, 2016). Both clustering methods showed similar CCC (0.72 for Tocher and 0.65 for UPGMA) significant at 1% probability. Furthermore, Jaccard bootstrap method showed that clusters formed by UPGMA were consistent. Both clustering methods should be used in combination, because one statistical approach complements the other.
Traits with less impact on Mean Decrease Impurity (MDI), and respectively with less impact to detect genetic divergence among accessions were endof flowering appendage, calyx margin and calyx pigmentation (Figure 2). Endof flowering appendagewas absent for all formed clusters. Calyx pigmentation was present only in two clusters, for Tocher and UPGMA method. Calyx margin had small variation among clusters, being intermediate or dentate (Tables 3 and 4).    , Rio Largo, v. 17, n. 3, p. 1-11, 2019 Cramer's V correlation among the 105 intercoefficients presented 22 significant correlations by Chi-Square test at 5% significance (Table 5). Traits with more correlations were Fruit shape (seven); Fruit shape at pedicel attachment (five); Fruit surface, Flower position, Corolla spot color, Calyx margin, Fruit shape at the endof flowering, Fruit cross-sectional corrugation (four). Fruit color at mature stage and endof flowering appendage had no correlation with any other trait. On the other hand, fruit shape at the endof flowering and fruit color at mature stage had great contribution to distinguish genotypes, presenting higher MDI values. Large diversity between clusters can be observed for fruit color at mature stage varying between lemon-yellow, dark red, and purple. All fruit shape classes at the endof flowering were observed in clusters (pointed, blunt, sunken, sunken and pointed and others), demonstrating high variation of this genetic trait among accessions. Fruit shape was also detected as one of the most effective descriptors for Capsicum accessions by Costa et al. (2015).
MDI exhibit desirable properties for assessing the relevance of a variable: it is equal to zero only if the variable is irrelevant and depends only on relevant variables (Liaw and Wiener, 2002). MDI also showed that all 15 traits contributed to analyze genetic divergence and should be used in further similar studies. If the number of traits under analysisshould be reduced, Calyx margin and Calyx Pigmentation should be discarded due to their lower reduction on mean impurity and also because Calyx margin was correlated with FP, FNA, CSC and Fsh; and Calyx Pigmentation with CSC. Fruit blossom end appendage also showed lower score on MDI; however, this trait was not correlated with any other variable.
Capsicum spp. Germplasm Bank from UFU has high variability and qualitative reproductive traits were capable of distinguishing genotypes in several clusters. These results are in accordance with other published papers that accessed genetic divergence in peppers of the genus Capsicum (Baba et al., 2015;Maciel et al., 2016;Sudré et al., 2010). The same accessions were used to detect divergence based on the physiochemical characteristics of fruits (Maciel et al., 2016). Unlike this study, genotypes were clustered in only four groups, evidencing that reproductive traits were more variant in the germplasm. Our study also allowed isolating more genotypes. For plant breeders, isolated genotypes represent a possibility of new crossings that could explore maximum heterosis.
Multicategorical traits have the advantage of easy observation and require less time and labor, being ideal for use in gene bank and collections with short human and financial resources. Moreover, these traits are not affected by the environment. Genetic divergence detected in this paper encourages other researchers to perform the characterization of pepper collections as completely as possible, because it becomes possible to generate more reliable information of the variability and genetic divergence among accessions. Our research discloses the reproductive biodiversity of peppers in 'Alto Paranaíba' and 'TriânguloMineiro' regions and the importance of maintening these genotypes. For users of other pepper collections, there is possibility of using this genetic bank, ensuring new materials for plant breeders.