Genetic cause and effect interrelationships for grain quality attributes of irrigated rice

Abstract The objective of this work was to determine the genetic correlations and the direct and indirect associations of agronomic traits and grain quality attributes with the percentage of whole grains in flood irrigated rice. The experiment was carried out in two environments, in a randomized complete block design with three replicates, using 23 irrigated rice genotypes. The evaluated traits were: percentage of whole grains, caryopsis length, caryopsis width, panicle length, panicle weight, 1,000 grain weight, days to flowering, percentage of chalky grains with white belly, total chalky area, total whiteness, vitreous whiteness, and defects in coloring. The percentage of grains with white belly and total chalky area were positively correlated, whereas the percentage of grains with white belly and vitreous whiteness were negatively correlated. The traits caryopsis width, percentage of chalky grains with white belly, panicle weight, and 1,000 grain weight showed indirect effects on whole-grain yield response according to total chalky area and total whiteness. Total chalky area and total whiteness are the factors that most negatively influence the percentage of whole grains according to the genotypic correlations and direct effects.


Introduction
In genetic breeding programs, the development of superior genotypes of rice (Oryza sativa L.), a staple food and the second most produced cereal worldwide (FAO, 2021), has been taking new directions.In addition to the selection of genotypes with genetic constitutions of high yield potential, the search for grains with quality attributes has become an important standard for the launch of new cultivars (Rangel et al., 2000).
In flood irrigated rice, grain quality attributes are considered quantitative traits, strongly influenced by the environment and its interaction with the genotype (Silva Junior et al., 2022).The early selection and setting of these attributes in segregating generations are difficult due to a low heritability and the use of destructive methods to measure some traits, such as the percentage of whole grains, which requires grain polishing that damages the embryo and makes the sample unusable as a seed (Facchinello et al., 2022).In these cases, Nardino et al. (2016) recommend using indirect selection, in which selection is applied to a secondary trait to improve a desired one through a correlated response between both of them.However, Falconer (1987) added that, for a more efficient result, the direct and indirect methods should be used concomitantly.
An alternative is the path analysis method of Wright (1921) that proposes unfolding the estimated linear correlation in direct and indirect effects of several traits in relation to a main one, which can facilitate the understanding of existing correlations.However, for this method, it is important to verify the degree of multicollinearity between explanatory variables (Cruz & Carneiro, 2006), since a high degree of multicollinearity can generate an overestimation of direct and indirect effects, without any biological meaning (Toebe & Cargnelutti Filho, 2013).If a high degree of multicollinearity is detected, the path analysis may be either carried out under multicollinearity (crest path analysis), with the addition of a k-value to the diagonal elements of the correlation matrix, or traditionally, with the elimination of highly correlated variables (Cruz & Carneiro, 2006).
The path analysis under multicollinearity has been used very effectively in the genetic phyto-improvement of crops such as peanut (Arachis hypogaea L.) (Luz et al., 2011), maize (Zea mays, L.) (Toebe & Cargnelutti Filho, 2013), and kale [Brassica oleracea subsp.acephala (DC.)Metzg] (Azevedo et al., 2016), showing the possibility of selecting main traits through indirect selection.In rice, this methodology can be used to clarify how agronomic traits and grain quality attributes influence the percentage of whole grains, which may contribute to indirect selection and the development of rice genotypes with more whole grains.
The objective of this work was to determine the genetic correlations and the direct and indirect associations of agronomic traits and grain quality attributes with the percentage of whole grains in flood irrigated rice.

Materials and Methods
The experiment was carried out in the 2015/2016 harvest, in the two following locations in the state of Rio Grande do Sul, Southern Brazil: the experimental field of Estação de Terras Baixas of Embrapa Clima Temperado, in the municipality of Capão do Leão (31°46'3"S, 52°26'55"W); and a property in the municipality of Nova Esperança do Sul (29°23'40"S, 54°50'32"W).
To obtain the genotypes in the field, sowing was carried out at a density of 100 kg ha -1 , using a mechanical plot seeder, under a no-tillage system.Base fertilization consisted of 300 kg ha -1 NPK (formula 5-20-20) and 90 kg ha -1 nitrogen, in the form of urea, half applied at the V4 stage (collar formed on the fourth leaf of the main stem) and the other half at the R0 stage (panicle initiation).The permanent flooding irrigation system was adopted until the final stage of genotype maturation.
The experimental design was a randomized complete block with three replicates, in which the plots consisted of four rows of 5.0 m in length, spaced at 0.17 m.The useful area of the plot were the four central meters of the two inner rows, to exclude any border effect.
The evaluated agronomic traits were: percentage of whole grains; caryopsis length (mm); caryopsis width (mm); panicle length (cm); panicle weight (g); 1,000 grain weight (g); and days to flowering, considering the number of days from emergence to 50% exposed panicles.The following physical attributes of intrinsic grain quality were also analyzed: sum of the percentage of chalky grains with white belly (%), total chalky area (%), total whiteness, vitreous whiteness, and sum of the percentage of defects in coloring (stained, streaked, moldy, and yellow grains).These attributes were determined with the aid of the S21 rice statistical analyzer (S21 Solutions, Santa Cruz do Rio Pardo, SP, Brazil), based on the analysis of digital images of each sample.
The results were subjected to statistical analyzes using the GENES software (Cruz, 2013).Individual analysis of variance and the F-test were performed for each studied variable, in order to verify the existence of variability between genotypes and to check for homogeneity of variances.This way, the following estimates of genetic parameters were obtained: environmental, genotypic and phenotypic variances; coefficient of variation; and heritability.
For the path analysis, multicollinearity was detected in the matrices of genotypic and phenotypic correlations according to the criteria of Montgomery & Peck (1981) by determining the ratio (condition number) between the highest and lowest eigenvalues of the X'X matrix.The variance inflation factor obtained for each variable on the inverse diagonal of the correlation matrix (X'X) was also considered as an indicator of multicollinearity.
If severe multicollinearity was detected, to avoid the exclusion of variables, a procedure similar to the crest regression analysis was used, where a constant (k) is added to the diagonal of the X'X correlation matrix, in order to reduce the variance of the trail analysis least squares estimator (Cruz & Carneiro, 2006).The selected k-value (between 0 and 1) should be the best value that minimizes multicollinearity, considering the lowest constant for the stabilization of the trail coefficients, a variance inflation factor lower than 10, and a condition number lower than 100 (Hair et al., 2009).Cruz et al. (2014) highlighted that the higher the k-value used, the more biased the information obtained by the trail analysis will be.
A moderate to strong degree of multicollinearity was detected, with a condition number between 100 and 1,000 eigenvalues of the matrix, as well as values of the variance inflation factor higher than 10.The constant k = 0.0860 was used to minimize multicollinearity since it was the lowest constant that allowed of the stabilization of trail coefficients and showed a variance inflation factor lower than 10 and a condition number lower than 100.
Subsequently, phenotypic and genotypic correlations were obtained through the bootstrap test with 10,000 simulations, determining the existence of a significant linear correlation between the studied variables.The phenotypic correlations were analyzed by obtaining an n number of correlation estimates across the original permuted (unstructured) dataset.For the estimation of genotypic correlation, empirical distributions were constructed to verify significance by the Mantel test, which requires a higher number of cultivars (Oliveira et al., 2013), as used in the present study.After verifying the existence of a significant correlation between pairs of traits, the trail analysis was carried out to indicate the direct and indirect effects of the evaluated agronomic traits in relation to the percentage of whole grains.

Results and Discussion
A significant difference as to genotype effect was observed by the F-test among 11 of the analyzed variables (Table 1).Regarding environmental variance, only caryopsis width and percentage of defects in coloring showed significant differences, whereas, for the genotype x environment interaction effect, most variables differed significantly.The coefficients of variation of the studied traits ranged from 1.47% for caryopsis width to 19.01% for chalky area, which means there was a good experimental precision.However, the sum of the percentage of chalky grains with white belly (111.68%) and the percentage of defects in coloring (51.64%) showed high values, an effect mainly due to the nature of these variables.Moreover, there was a high heritability from 66.62 and 99.63% for the evaluated traits, which shows an adequate efficiency in the selection process.
A total of 32 pairs of significant genotypic correlations between the studied traits were obtained for the 23 evaluated cultivars (Table 2).The highest positive magnitude of 0.967 was observed between the sum of the percentage of chalky grains with white belly and total chalky area, which was already expected since both are a result of the variation of the same defect caused by the air spaces generated by disturbances at the time of the accumulation of starch and protein molecules in the grains (Shen, 2000).The highest negative magnitude was found between the sum of the percentage of chalky grains with white belly and vitreous whiteness, which were inversely proportional.
Regarding whole grains, six significant genetic correlations were observed, with negative effects for caryopsis width, sum of the percentage of chalky grains with white belly, chalky area, total whiteness, panicle weight, and 1,000 grain weight, with values ranging from -0.373 to -0.756 for caryopsis width and total whiteness, respectively.The relationship between the sum of the percentage of chalky grains with white belly and chalky area confirm the finding of Liu et al. ( 2009) that chalky grains, due to air spaces, break easily during processing.Zhou et al. (2009) added that these characteristics are among the greatest challenges in rice cultivation, as they directly affect rice quality and price, together with the percentage of whole grains.Therefore, simultaneous selection considering these two traits, resulting from this high genetic correlation, could generate considerable genetic gains for grain quality attributes.
For panicle weight and 1,000 grain weight, the negative correlations in relation to whole grains (Table 2) is attributed to the japonica subspecies of the cultivars, whose grains are rounder and have a different composition and inferior grain quality than those of the indica subspecies, as shown by Table 1.Summary of the analysis of variance and estimates of the genetic and phenotypic parameters of the agronomic traits evaluated in 23 flood irrigated rice (Oryza sativa) genotypes in two locations in the state of Rio Grande do Sul, Brazil (1) .the significant negative correlations between these traits and vitreous whiteness (-0.726 and -0.802, respectively).The intrinsic characteristics of grain type and plant architecture, for example, led BRS AG to be the first rice cultivar intended for uses other than human consumption, such as alcohol production or animal feed, explaining its high genetic distance in relation to other commonly used cultivars, with a practically double average 1,000 grain weight of 52 g (Streck et al., 2017).
Panicle length did not present significant correlations with whole grain yield (Table 2), contrasting with the finding of Jongkaewwattana & Geng (2002), who concluded that this character could be a useful selection index for the improvement of whole grain yield.These different results can be explained by the small amplitude variation for this character in the estimates of genetic correlations in the present study, since only cultivar BRS Querência showed higher panicle length values.
Days to flowering was the only variable that did not present a significant correlation with the explanatory variable percentage of whole grains, nor relationships with the other analyzed variables (Table 2).Similar results were reported in other studies with rice, where no correlation or few low-magnitude correlations were found between days to flowering and several other traits (Nandan et al., 2010;Kishore et al., 2015).Therefore, for an efficient response of indirect selection in breeding, it is important to identify which characters with a high correlation with the main trait present the greatest direct effect in a direction favorable to selection (Cruz & Carneiro, 2006).
In this sense, total whiteness and chalky area presented high negative magnitudes regarding the direct effect on whole grain yield, explaining much of the correlation between these traits (Table 3).In addition, indirect effects via chalky area and total whiteness were detected for caryopsis width, sum of the percentage of chalky grains with white belly, panicle weight, and 1,000 grain weight, with negative correlations in relation to the response variable whole grain yield.
Chalky area, sum of the percentage of chalky grains with white belly, and total whiteness were the factors that showed the greatest direct and indirect losses in terms of percentage of whole grains (Table 3).Several studies have sought to minimize these effects, both through genetic improvement and through different management methods, in order to avoid possible problems related to the occurrence of chalky grains (Marchezan et al., 1992;Ishimaru et al., 2009;Gu et al., 2015;Londero et al., 2015).
For variables with a high correlation, Cruz & Carneiro (2006) concluded that the best strategy to be used is simultaneous selection, with emphasis on characters whose indirect effects are significant, as was the case for the sum of the percentage of chalky grains with white belly and 1,000 grain weight in the present study, with a genotypic correlation equal to -0.563 and -0.516, respectively, but a low direct effect on the response variable.
Indirect selection through the correlations shown in the present work, mainly of chalky area and total whiteness, can be effective to increase the percentage of whole grains.

Conclusions
1. Total chalky area, together with total whiteness, is the factor that most negatively influences the percentage of rice (Oryza sativa) whole grains according to the genotypic correlations and direct effects.
2. The variables caryopsis width, percentage of chalky grains with white belly, panicle weight, and 1,000 grain weight have indirect effects on wholegrain yield response according to total chalky area and total whiteness.