Contribution of production and seed variables to the genetic divergence in passion fruit under different nutrient availabilities

The objective of this work was to evaluate the relative contribution of variables related to fruit production and to seed morphophysiological characteristics to the genetic divergence in passion fruit (Passiflora edulis) progenies, aiming at selecting progenies potentially responsive to fertilization. Ten progenies were evaluated under conditions of low (50% of the recommended dose of fertilizers) and high soil fertility (dose 50% higher than the recommended one), regarding variable sets related to production and to the morphological and physiological characteristics of seeds. The association between these sets was determined by the canonical correlation analysis and by the nearest-neighbor clustering method. The nutritional environments interfered in the relative contribution of the variables to the genetic divergence of the progenies. The accelerated aging test of seeds – from the set of seed physiological quality – did not contribute significantly to the selection of genotypes responsive to soil fertilization. The most responsive progenies to the increased availability of nutrients were grouped according to the production variables that were evaluated under high soil fertility. Irrespectively of the evaluated environment, the set of production variables is the one that contributes more expressively to the identification of the genetic divergence of passion fruit progenies.


Introduction
Passion fruit (Passiflora edulis Sims) has achieved an outstanding position in the national fruit market, due to the increased demand from the fresh and industrial markets, and to the interest of producers in a crop with a cycle shorter than that of other fruit trees.Brazil produces more than 800 thousand tons of passion fruit Pesq. agropec. bras., Brasília, v.52, n.8, p.607-614, ago. 2017 DOI: 10.1590/S0100-204X2017000800006 per year; however, its average productivity is only of 14 Mg ha -1 (IBGE, 2014).
Genetic diversity is fundamental for breeding programs.In the case of passion fruit, the development of cultivars with better agronomic traits is closely related to the genetic variability in segregating generations (Bruckner et al., 1995;Viana et al., 2007).The multivariate analysis is an efficient method to calculate the similarity between genotypes representative of the population.It uses specific variables of interest and allows the formation of groups with minimal variation between individuals and maximum variation between groups (Crossa & Franco, 2004).Cruz et al. (2012) point out the importance of using canonical correlations to evaluate the maximum correlation between two sets of variables.This analysis has been used in fruit trees to assess: the relationship between soil and plant, in peach trees (Terra et al., 2014); agronomic traits of passion fruit (Neves et al., 2013); physical and chemical properties of passion fruit and juice (Espitia-Camacho et al., 2008); soil biological and chemical attributes in the apple crop (Maluche-Baretta et al., 2006); and fruit quality and production traits of passion fruit (Viana et al., 2003).
The hierarchical clustering method can be used to unify genotypes by similarity measures, until a dendrogram is established (Cruz et al., 2012).Several studies have been developed to quantify and characterize Passiflora genetic diversity, mainly seeking to identify superior genotypes (Viana et al., 2006;Gonçalves et al., 2007;Araújo et al., 2008;Silva et al., 2009;Krause et al., 2012;Paiva et al., 2014).Plant morphological and production variables are commonly used for this purpose; however, the variables related to seed morphological and physiological quality (germination and vigor) are not commonly employed (Brum et al., 2011).
In addition, little is known about the impacts of soil fertility or nutrient availability on the evaluation of passion fruit genetic divergence.This may be a problem, since progenies usually present different responses to fertilization management, which affects fruit productivity, as well as seed size and weight (Borges et al., 2006;Nascimento et al., 2011;Carvalho & Nakagawa, 2012).
The objective of this work was to evaluate the relative contribution of variables related to fruit production and to seed morphophysiological characteristics to the genetic divergence in passion fruit progenies, aiming to select progenies potentially responsive to fertilization.

Materials and Methods
Ten progenies of passion fruit -57x15, 144x130, 112x42, 117x19, 68x135, 81x117, 132x15, 144x42, 68x15, and 46x14 -, from the third cycle of the intrapopulation recurrent selection program of Universidade Estadual do Norte Fluminense Darcy Ribeiro (Uenf), were subjected to two levels of fertilization, according to the results of the soil analysis.The dose used for fertilization 1 was 50% lower than the recommended average for the crop (Carvalho et al., 2000), with 135 g urea per plant per year and 270 KCl per plant per year.In fertilization 2, the dose used was 50% higher than the recommended average for the crop, with 405 g urea per plant per year and 810 g KCl per plant per year.Fertilizations were split and applied monthly.
The experiment was implemented at the experimental unit of Uenf, in the municipality of Itaocara, located in the northwest of the state of Rio de Janeiro, Brazil, from September 2012 to May 2014.A completely randomized block design was used, in a split plot with ten progenies in the main plot and two levels of fertilization in the subplots, with two replicates of three plants each.
The production traits evaluated were: productivity (Mg ha -1 ); average fruit mass (g), for which ten fruits per treatment were weighed in a scale with two decimal places; pulp mass (g), as the average of ten fruits, also in a scale with two decimal places; and number of fruits per plant.
The seeds of these fruits were obtained by friction in a steel mesh sieve in running water until the aril was removed.Then, they were dried at room temperature, for 48 hours, to assess the following variables: length/width ratio of 15 seeds, measured with a digital caliper; average thickness of the 15 seeds (mm), also using a digital caliper; 1,000 seed weight (g), obtained by the average of eight replicates, in a scale with three decimal places (Brasil, 2009); germination test, in which the means of the treatments were obtained 28 days after the start of the test, from four replicates with 50 seeds, which were arranged between three germination paper sheets moistened with distilled water, at a ratio of twice the substrate mass, at alternating temperatures of 20-30°C, in a germination chamber (Brasil, 2009); first germination count, evaluated together with the germination test, for which normal seedlings were recorded at 14 days; accelerated aging, in which the seeds were placed on an aluminum screen, in a transparent plastic box, with 40 mL of water, where they remained for 48 hours, at 40ºC, in a germination chamber (Larré et al., 2007), before being subjected to the same methodology of the germination test, for which normal seedlings were recorded at 28 days (Brasil, 2009); first count of the accelerated aging test, for which the normal seedlings were counted at 14 days; seedling size, by using ten seedlings located at the top row of the paper roll at the end of the germination test (Negreiros et al., 2008); rootlet size, obtained from the measurement of the rootlets of ten seedlings located at the top row of the roll at the end of the germination test (Negreiros et al., 2008); and germination rate index, for which the number of normal seedlings was recorded every four days (Maguire, 1962).
The canonical correlation was used to estimate the maximum correlation of the linear combinations (Cruz et al., 2012) between production traits (set 1: productivity, pulp mass, fruit mass, and number of fruits per plant), seed morphological traits (set 2: seed length/width ratio, seed thickness, and 1,000 seed weight), and seed physiological traits (set 3: germination potential, first germination count, first count of the accelerated aging test, accelerated aging, seedling size, rootlet size, and germination rate index).The following models were used to obtain the linear combinations: which X i and Y i refer to one of the likely linear combinations between the traits from the sets 1, 2, and 3, in fertilizations 1 and 2.
The first canonical correlation maximizes the relationship between X i and Y i , and constitutes the first canonical pair, expressed by the equation: In this case, S 11 is the p x p covariance matrix, between the traits of set 1; S 22 is the q x q covariance matrix, between the traits of set 2; and S 12 is the p x q covariance matrix, between the traits of sets 1 and 2.
The other canonical correlations and canonical pairs were estimated using the eigenvalues and eigenvectors of the described expressions, corresponding to the p-or q-th estimated correlation.The significance of the hypothesis that all possible canonical correlations were null was assessed by the chi-square (X²) test: The analyses of canonical correlations were processed using the Genes software (Cruz, 2013).The evaluated variables were used to study the genetic divergence in different environments, based on Mahalanobis' generalized distance and on the unweighted pair-group method with arithmetic mean (UPGMA).The relative contribution of the traits to divergence was assessed by the Singh method, and the following parameters were evaluated: productivity, pulp mass, fruit mass, number of fruits per plant, seed length/width ratio, 1,000 seed weight, germination potential, first germination count, accelerated aging, seedling size, rootlet size, and germination rate index.The variables were calculated with non-standard means, and the grouping analyses among genotypes, at both fertilization levels, were performed by the Genes software (Cruz, 2013).

Results and Discussion
In the combination of variables set 1/set 2, the three canonical pairs were not significant for the fertilization levels 1 and 2 (Table 1).In the combination of variables set 1/set 3, the first and second canonical pairs were significant for fertilization 1, and only the first canonical pair was significant for fertilization 2. In the combination of variables set 2/set 3, the first and second canonical pairs were significant for the two levels of fertilization.The coefficients of the pairs were estimated in order to verify the relationship between the sets.When the tests detected significance, the first canonical pair was used for data interpretation, since it maximizes the relationship between X 1 and Y 1 (Cruz et al., 2012).
In the combination of seed morphological and physiological (germination and vigor) traits (set 2/set 3), the variable seed thickness showed greater effect on the following variables (Table 2): germination; first count of the accelerated aging test (FCAA); rootlet size and the germination rate index (GRI) in fertilization 1, with lower nutrient availability.In fertilization 2, the variable 1,000 seed weight had effect on the variables germination, first germination count, and seedling size.Regardless of the level of fertilization, accelerated aging was the only unaffected variable in set 3. Brum et al. (2011) verified significant canonical correlations between seed morphological and seedling variables of castorbean (Ricinus communis L.); however, the authors did not study these correlations in response to fertilization.
In the set 2/set 3 combination, the highest correlation coefficient in fertilization 1 was obtained for rootlet size.According to Tozzi & Takaki (2011), as germination occurs, the protein bodies identified in the cotyledons are degraded and almost completely consumed up to the moment when radicle protrusion occurs.Therefore, the faster the radicle protrusion process, the more vigorous the seedling is, because it shows greater capacity to transform the reserve supply into meristematic tissues, which may explain the great effect of the variable rootlet size on the coefficient of correlation.
In fertilization 2, the highest positive correlation was observed for the variables germination, first germination count, and seedling size; while, in fertilization 1, only the vigor variables FCAA, rootlet size, and GRI presented higher correlation values, with small contribution of the variable germination (Table 2).
In the combination of the production and of the seed germination and vigor traits (set 1/set 3), the variable number of fruits presented the greatest contribution to the effect of the variables seedling size and GRI, in fertilization 1 (Table 2).In fertilization 2, the variables productivity and fruit mass presented great contribution to the vigor variables fist germination count, rootlet size, and GRI.Balkaya et al. (2011) found that the variable total number of fruits (set 1) and the variables fruit diameter and length (set 2) contribute the most to the explanatory capacity of canonical variables; therefore, they should be used in the selection of more productive pumpkin (Cucurbita maxima Duch.) genotypes.In the present study, the variable number of fruits also provided the greatest contribution to crop production in fertilization 1.
In the set 1/set 3 combination, the variable productivity was negatively correlated with the variables of set 3, which presented greater effect on fertilization 1 (seedling size and GRI); in fertilization 2, productivity was the most influential variable of set 1, while the GRI was the only variable with great effect on set 3, for both levels of fertilization (Table 2).This confirms that the GRI is a good vigor test to identify superior lots or genotypes (Nakagawa, 1999).In the same combination, the variable germination correlated negatively to the variables of set 1; and, in the set 2/set 3 combination, it correlated positively with the variables of set 2, regardless of the level of fertilization.
Table 1.Degrees of freedom and chi-square values, obtained by the canonical correlation analysis among sets of production variables (set 1) and of seed morphological (set 2) and physiological traits related to germination and vigor (set 3), for passion fruit genotypes (Passiflora edulis), in treatments with two levels of fertilization (1) .The stress resistance trait accelerated aging did not correlate positively -or did not exert great effect -on the set 2/set 3 and set 1/set 3 combinations.Therefore, its use is not a good alternative for the selection of superior genotypes of passion fruit.
Of the variables of set 1 with relative contributions for the genetic divergence of the passion fruit progenies, pulp mass and fruit mass were more expressive in fertilization 1; and fruit mass, pulp mass, and productivity, in fertilization 2. For the variables of set 3, a lower contribution was observed, and, for those of set 2, the contribution was non-expressive (Table 3).
Table 2. Coefficients of canonical correlation among sets of production variables (set 1) and of seed morphological (set 2) and physiological traits related to germination and vigor (set 3), for passion fruit genotypes (Passiflora edulis), in treatments with two levels of fertilization (1) .1.0018 1.0022 (1) Fertilization 1, 50% of the average dose recommended for the crop; and fertilization 2, dose 50% higher than the average recommended for the crop.

Figure 1.
Dendrogram of genetic dissimilarities among ten progenies of passion fruit (Passiflora edulis), according to 12 variables evaluated at the levels of fertilization 1 (A, 50% of the average dose recommended for the crop) and 2 (B, dose 50% higher than the average recommended for the crop).
Fertilization 1, 50% of the average dose recommended for the crop; and fertilization 2, dose 50% higher than the average recommended for the crop.*Significant at 5% probability.ns Nonsignificant.CP, canonical pair.