Genetic divergence among Psidium accessions based on biochemical and agronomic variables

Sixty-nine Psidium accessions collected in six Brazilian states were analyzed by two non-hierarchical clustering methods and principal components (PC), to provide orientation for breeding programs. The variables ascorbic acid, βarotene, lycopene, total phenols, total flavonoids, antioxidant activity , titrable acidity, soluble solids, total soluble sugars, moistur e content, lateral and transversal fr uit diameter , fruit pulp and seed weighs, and plant fr uit number and weight wer analyzed. Specific gr oups were obser ved for the araçazeir os accessions, by the T ocher and the k-means methods, as well as by the thr e -dimensional dispersion of the four PCs. The clustering separated accessions of araçazeiros from the guava. There was no specific grouping in terms of States of origin, indicating the absence of barriers in the guava propagation accessions. Analyses suggested the collection of a greater number of guava germplasm samples from a smaller number of regions and divergent accessions with high nutritional compound levels to develop new cultivars.


INTRODUCTION
In Brazil, in 2007 guava ranked 11 th in cultivated area and production value, 10 th in yield and 5 th in productivity among other important fruits such as orange, banana, grapes, papaya, apple, mango, tangerine, passion fruit, lemon, peach, persimmon, avocado, fig, pear, and quince.The main guava-producing states in 2007 were Pernambuco, São Paulo, Goiás, and Bahia, which together accounted for almost 75 % of the national production (http://www.sidra.ibge.gov.br).
Guava (Psidium guajava) belongs to the Myrtaceae family, with approximately 130 genera and 3000 species of trees and shrubs distributed mainly in the tropics and subtropics (http://delta-intkey.com).Guava is native to northern South America and widely distributed in the tropical regions of America (Risterucci et al. 2005).Other fruits of the genus Psidium, with about 150 species, are the Brazilian guava called araçazeiros, a term that refers to native Psidium species, with P. cattleyanum Sat., P. incanesces Martius, P. gradiflorum Martius, and P. arboretum Vell (Raseira and Raseira 1996).Although not as interesting as guava, Brazilian guava species have some important attributes such as earliness, tolerance to certain pests and diseases and exotic fruit flavor (Manica 2000).
Despite being regarded as one of the diversity centers of guava and araçazeiros (Soares-Silva and Proença 2008), Brazil has few research institutions with gene banks for this Psidium species which, in total, have compiled 310 accessions and 174 accessions of Brazilian guava.Important cultivars for the Brazilian guava agribusiness are Paluma, Rich, Sassaoka and Pedro Sato, which were developed 1 Embrapa Semiárido, C.P. 23,Petrolina,PE,Brazil. *E-mail: casantos@cpatsa.embrapa.br 2 Universidade Estadual Paulista (Unesp), Campus de Botucatu, 18.618-000, Botucatu, SP, Brazil from selections of producers or research institutions in open-pollinated seed orchards (Pommer et al., 2006).
The available collections of guava and araçazeiros in Brazil have been almost exclusively characterized for agronomic traits and only few studies focused on the chemical characterization, e.g., the paper of Lima et al. (2002).The importance of studies of functional compounds in guava can be exemplified by lycopene, the carotenoid responsible for the red color of the guava pulp.According to Yan et al. (2006), lycopene is considered an excellent antioxidant, with a high resistance potential to diseases such as prostate cancer and arteriosclerosis.According to these authors, the lycopene level found in the guava cultivar Paluma is the double found in tomatoes.
For Pereira and Nachtigal (2003) the aim of the improvement of guava characteristics would be an average fruit weight > 100 g, a red pulp and pulp yield > 70 %, soluble solids > 10 °Brix, sugar/acidity ratio > 11, vitamin C content of around 100 mg of ascorbic acid per 100 g of pulp and a minimum production of 30 t ha -1 .
The agronomic descriptions of guava collections in Brazil (Souza Junior et al. 2002, Gonzaga Neto 2002, Santos et al. 2008, Silva Junior et al. 2008) did not establish estimates of genetic distances between accessions, to support controlled crossings and the development of cultivars with improved traits, including higher levels of nutritional compounds.In the international literature divergence studies have been published based on morphological traits for collections of Mexican (Hernandez-Delgado et al. 2007) and Cuban accessions (Rodriguez et al. 2004).
The purpose of this study was to group accessions belonging to the Psidium gene bank of Embrapa Semiarid by multivariate techniques based on biochemical and agronomic variables, to provide orientation for genetic breeding programs of the genus Psidium.

MATERIAL AND M ETHODS
Fruits of 59 guava (P.guajava) and 10 Brazilian guava or araçazeiros (Psidium spp.) accessions of the field germplasm bank of Psidium (Table 1) installed in the experimental field Bebedouro of Embrapa Semiarid were collected.The accessions were sampled at physiological maturity in six different Brazilian states, as described by Santos et al. (2008).Each accession was established in the field from four-month-old seed-grown seedlings, in January 2007.The data of six plants were evaluated, from two replications with three plants each.The accessions were managed according to the regional practices for commercial guava production, using a drip irrigation system in a plant spacing of 4.0 m x 4.0 m.To induce fruiting, the plants were pruned in August 2008 and harvest began in December 2009.
Biochemical variables: acidity, % citric acid was determined by titration method proposed by the Instituto Adolfo Lutz (1985); soluble solids (%), ºBrix, determined using a digital refractometer, according to standard techniques of the Instituto Adolfo Lutz (1985), total soluble sugars, % M, quantification by the anthrone method (McCready et al. 1950); moisture content, % H 2 O determined as weight of oven-dried (under forced air circulation at 65 °C to constant weight) fresh fruit.
Variables of agronomic characters: lateral fruit diameter, mm; transversal fruit diameter, mm; yield per plant, g; seed weight per fruit, g; number of fruits per plant and pulp weight per fruit, g, were obtained from samples taken for these variables every two or three days, during the production cycle.
The data were statistically analyzed by: 1. analysis of variance, 2. principal component analysis, after standardizing the data to exclude the effect of different measurement scales of variables, using SAS (SAS 1989), 3. cluster analysis by the modified Tocher method (Vasconcelos et al. 2007), based on Mahalanobis´ generalized distances, supported by software Genes (Cruz 2008), 4. cluster analysis with standardized data, according to the (SAS) FASTCLUS Procedure (1989), with the formation of six similar groups corresponding to the number of states sampled.The FASTCLUS Procedure uses a method called nearest centroid sorting, where the dissimilarity measure is the mean Euclidean distance, which is a k-means clustering method (SAS 1989).

RESULTS AND DISCUSSION
In the analysis of variance a significant effect (P < 0.01) was observed for each of the 16 variables by test F, indicating the existence of phenotypic variability in the 69 Psidium accessions studied.In the analysis of variance performed only with the guava accessions the effect was also significant (P < 0.01) by the F test for all variables, except moisture content (p < 0.05), indicating the existence of phenotypic variability among the 59 Psidium guava accessions under study.
The 10 araçazeiros accessions were grouped in a specific cluster when analyzed by the k-means (Table 2), while in the Tocher analysis, these same 10 accessions were grouped into two clusters, one with seven and the other with three accessions (Table 3).Results of this study indicate that the Brazilian guava showed biochemical and agronomic traits that distinguish them from guava accessions, which may be the result of the intense artificial selection in guava for the purpose of agronomic cultivation and fruit quality-related traits.Analyzing the araçazeiros and guava accessions from different Brazilian states for morphological traits, Santos et al. (2008) concluded that the fruit traits e.g., pulp color and fruit size, were the most influenced by the artificial selection in guava.
As discussed by Raseira and Raseira (1996), the term araçazeiro refers to several Psidium species with wide distribution in Brazil (Soares-Silva and Proença 2008).In the analysis by the k-means clustering method, all araçazeiros were grouped into a single cluster (Table 2), suggesting that the differences among araçazeiros ecotypes were smaller than of araçazeiros compared to guava accessions.The formation of two araçazeiros groups in the Tocher cluster analysis (Table 3) suggests that there were probably different species or even variability within a single species, considering that the total number of Brazilian guava species was not determined in this study.In view of the presence of different araçazeiros species in Brazil, this analysis indicates that the biological significance of the Tocher analysis was greater than of the k-means.In both analyses, results indicated that the differences in agronomic and biochemical variables were sufficient to form specific groups of araçazeiros and guava accessions.
The groups of Psidium accessions formed by the two clustering methods showed different results, with no coincidence: the 13 accessions clustered in group 4 by the k-means method were distributed in groups 4 and 1 by the Tocher method (Tables 2 and 3).Many clustering methods are biased in relation to group size, shape and dispersion; the k-means method, for example, tends to find groups with an approximate number of elements (SAS 1989), as observed in this study.The Tocher method clustered a first group with 34 accessions, in other words, 50 % of the accessions formed a single group and group 10 contained only one plant, indicating limitations to identify differences between accessions.As concluded in the review of Fonseca et al (2004), the application of the discriminant function technique can minimize the probability of genotype misclassification, in the case of a preexisting classification study.
There were no specific clusters for sampling states of P. guajava accessions by the two grouping methods (Tables 2 and 3).Not even by the k-means method, where the number of six clusters had been defined in reference to the number of sampled Brazilian states.The absence of relationships between clusters and germplasm sampling  (Visser andReheul 2001, Jagadeesh et al. 2007) or for species exposed to longterm selection, such as Solanum tuberosum (Rio et al. 2004).It should be noted that the guava genotypes were subject to selection for multiple traits and may have been vegetatively dispersed from one location to another, for various reasons, resulting in the formation of groups independent from the collection site.Although Brazil is considered an area of guava diversity (Risterucci et al. 2005), no naturally occurring guava plants were found among the accessions of the collection studied; guava was always associated with human presence, growing in backyards, on road sides or beside old houses, among other sites.
The accumulated variation of the three principal components was 69 %, indicating a restricted suitability for a three-dimensional dispersion model of Psidium accessions (Table 4).When the four principal components are considered, accumulated variation was 76 %, indicating good suitability, without major distortion of accessions in three-dimensional diagrams, as discussed by Cruz and Regazzi (1994).The 10 araçazeiros accessions formed a specific group in the three-dimensional diagrams (Figures 1 A and B), suggesting a smaller differentiation among araçazeiros than among guava accessions, probably due to the process of artificial selection with the latter, as observed in the analysis of non-hierarchical grouping methods (Tables 2 and 3).
In the principal component analysis, the variables with highest weight of eigenvalues which explained most of the variation among Psidium accessions were transversal fruit diameter, total antioxidant activity, number of fruits per plant and total phenols.The variables with highest weight of eigenvalues, which explain least of the variation among Psidium accessions, were pulp weight per fruit and total flavonoids, and are therefore dispensable (Table 4).Cruz and Regazzi (1994) reported that traits that are relatively invariant among accessions and/or redundant due to their high correlation with other characters are dispensable or less important in divergence studies.This is the first study to apply multivariate techniques to analyze genetic diversity in guava and araçazeiros accessions in Brazil.Hernandez-Delgado et al. (2007) analyzed 52 Psidium accessions in Mexico with 50 qualitative and quantitative characters and reported that the analysis of 14 principal components explained less than 30 % of the total variation in the characterization of the accessions, which is well below the value observed in this study.Sanabria et al. (2005) reported that the first three principal components explained 72 % of the total   variation in 53 Colombian accessions analyzed with 17 qualitative characters and 9 quantitative traits.In the same paper, the authors reported the formation of three groups, with 24, 12 and 17 accessions per group, and reported no correlation between sampling locations and grouping of accessions.
The multivariate analysis techniques used to separate 69 Psidium accessions were generally effective for the discrimination of araçazeiros from guava accessions, indicating that the selections for guava, the most important species of the genus, were efficient to induce changes in the biochemical and agronomic traits evaluated (Tables 2  and 3; Figures 1A and 1B).The groups formed by the two methods of non-hierarchical clustering as well as the dispersion of the four principal components indicated no relationship with the sampling location of both araçazeiros as well as guava accessions.For Psidium germplasm collection activities, the analysis of biochemical and agronomic variables suggests restricted collections of seed germplasm, in a smaller number of regions, with a larger number of samples per region.
The information compiled in this study of genetic divergence along with the levels of antioxidant compounds found in the tested accessions, can be used as a reference in guava breeding programs for cultivars with high levels of antioxidant activity.This property has been reported as important for human health because antioxidant compounds can reduce the damaging effects of free radicals in cells (Leong and Shui 2002) and consequently the predisposition to mutagenic diseases.
For the controlled crosses accessions should be selected that performed consistently different by the two clustering methods, as well as in the dispersion of the first four principal components, associated with higher values for the traits of agronomic and biochemical interest.For example, accessions G03MA, G10MA and G01MA, with highest values for antioxidant activity (AOX), may be suitable for crossings with accessions G73RO, G20MA and G32PE, with highest lycopene contents.

CAF
Santos et al.

Figure 1 .
Figure 1.Dispersion diagram based on the principal components prin1, prin2 and prin3 (A) and prin1, prin2 and prin4 (B) of 69 accessions of the Psidium, analyzed for 16 biochemical and agronomic traits.

Table 1 .
Identification of 69 accessions of guava (G) and Brazilian guava or araçazeiro (A) of the Psidium genebank CAF Santos et al.

Table 2 .
Grouping of accessions of the Psidium gene bank of Embrapa Semiarid, analyzed for 16 biochemical and agronomic traits, distributed in six groups by the k-means clustering method based on the Euclidean mean distance

Table 3 .
Grouping of accessions of the Psidium gene bank of Embrapa Semiarid, based on 16 biochemical and agronomic traits, distributed in 10 groups by the modified clustering method of Tocher based on Mahalanobis' distance Genetic divergence among Psidium accessions based on... regions has been reported for traits influenced by environmental conditions, such as phenotypic, for Stipa lagascae and Artocarpus heterophyllus

Table 4 .
Eigenvalues, accumulated percentage and variables associated to the eigenvectors with highest absolute value for 16 principal components in 69 accessions of the Psidium, analyzed for 16 biochemical and agronomic traits