Designing the best breeding strategy for Coffea canephora: Genetic evaluation of pure and hybrid individuals aiming to select for productivity and disease resistance traits

Breeding programs of the species Coffea canephora rely heavily on the significant genetic variability between and within its two varietal groups (conilon and robusta). The use of hybrid families and individuals has been less common. The objectives of this study were to evaluate parents and families from the populations of conilon, robusta, and its hybrids and to define the best breeding and selection strategies for productivity and disease resistance traits. As such, 71 conilon clones, 56 robusta clones, and 20 hybrid families were evaluated over several years for the following traits: vegetative vigor, incidence of rust and cercosporiosis, fruit ripening time, fruit size, plant height, canopy diameter, and yield per plant. Components of variance and genetic parameters were estimated via residual maximum likelihood (REML) and genotypic values were predicted via best linear unbiased prediction (BLUP). Genetic variability among parents (clones) and hybrid families was detected for most of the evaluated traits. The Mulamba-Rank index suggests potential gains up to 17% for the genotypic aggregate of traits in the hybrid population. An intrapopulation recurrent selection within the hybrid population would be the best breeding strategy because the genetic variability, narrow and broad senses heritabilities and selective accuracies for important traits were maximized in the crossed population. Besides, such strategy is simple, low cost and quicker than the concurrent reciprocal recurrent selection in the two parental populations, and this maximizes the genetic gain for unit of time.


Introduction
The Coffea genus has great economic and social importance, with Brazil being the world's largest producer and exporter [1]. Within  from the both populations were from 16% to 140% higher than the two commercial clonal varieties, and the most productive cross was more vigorous. By selecting 5% of the best plants, they showed high genetic gain expectations with a value higher than 60% for production; moderated with rates of 14% to 18% for young plant vegetative vigor and low for canopy diameter, being possible to predict 60% of the genetic gains in selected plant productivity compared to the most productive clone used as control. Taken together, improving interpopulation hybrids seems promising and studies on comparative hybrid populations strategies are needed and should be prioritized. The objectives of this study were to evaluate parents and families from the populations of conilon, robusta, and its hybrids and to define the best breeding and selection strategies for productivity and disease resistance traits.

Genetic materials and experimental design
The studied conilon coffee population consisted of 71 clones (S1 Table in S2 File) that were evaluated in two trials, each with 34 clones and three controls (UFV 3629-11, UFV 3628-2, and UFV 513). The robusta coffee population consisted of 56 clones (S2 Table in S2 File) evaluated in two trials, each with 27 clones and two controls (UFV 3366-134 and UFV 3366-139). Each experimental trial used a randomized block design with five replications and one plant per plot.
The hybrid coffee population consisted of 20 families obtained by controlled crossbreeding between five conilon coffee clones (male genitors) and five robusta coffee clones (female genitors) (S3 Table in S2 File). The hybrids were evaluated using a randomized block design with 1 to 35 (average number 11.90) repetitions and one plant per plot, obtained from seeds.
Planting was done at a spacing of 3.0 m between rows and 1.5 m between plants. The conilon and robusta clones were planted in July 2009 and the hybrids were planted in March 2011. Throughout the experimental period, the plants received the treatments necessary for cultivation based on Matiello et al. [11], but without pruning or the use of chemicals to control diseases.

Phenotyping
Phenotyping was carried out at the time of fruit physiological maturity for seven consecutive years (2010 to 2016) in the conilon and robusta coffee populations and for five consecutive years (2012 to 2016) in the hybrid population. The following traits were assessed: vegetative vigor, incidence of rust and cercosporiosis, time of fruit ripening, fruit size, plant height, canopy diameter, and yield per plant.
The vegetative vigor was assessed using a scale from 1 to 10, where a score of 1 was assigned to highly depleted plants and 10 was assigned to highly vigorous plants. The incidence of rust (caused by the fungus Hemileia vastatrix Berk. & Br.) and cercosporiosis (caused by the fungus Cercospora coffeicola Berk. & Cooke) were evaluated on a scale of 1 to 5, where 1 was assigned to asymptomatic plants and 5 was assigned to plants highly susceptible to the pathogen. The fruit ripening time was classified as early, medium or late, with scores of 1 to 3, respectively. Fruit size was classified as small, medium or large, with scores of 1 to 3, respectively. Plant height (cm) was obtained by measuring from ground level to the apical point of the most developed orthotropic branch. Canopy diameter (cm) was measured as the canopy projection perpendicular to the planting row. Yield per plant was obtained by measuring the total volume of fruits, in liters.
There are several advantages in using the REML/BLUP procedure as it enables the incorporation of kinship information, comparisons of genotypes over time and space, correction of environmental effects, as well as the simultaneous estimate of variance components and prediction of genetic values. Moreover, it deals well with complex data structures and can be applied to unbalanced data and non-orthogonal designs [27,28]. Then, this statistical approach is especially suited to coffee breeding.
Variance components and genetic parameters were estimated via REML [29] and genotypic values were predicted via BLUP [30]. All statistical analyses were performed using the Selegen-REML/BLUP software [31].
Conilon and robusta coffee populations. To evaluate the conilon and robusta coffee clones, the following linear mixed model was used [32]: where, y is the vector of phenotypic data; u is the vector of year or crop effects (assumed as fixed), added to the general average; g is the vector of genotypic effects (assumed as random); b is the vector of block effects across experiments (coded sequentially and assumed as random); p is the vector of permanent environmental effects (assumed as random); and e is the vector of residuals (random). The capital letters (X, Z, W, and T) represent the incidence matrices of these effects.
The average and variance structures associated with this model are [33]: yju; V � NðXu; VÞ; gjs 2 g � Nð0; Is 2 g Þ; bjs 2 b � Nð0; Is 2 b Þ; pjs 2 p � Nð0; Is 2 p Þ; and ejs 2 e � Nð0; Is 2 e Þ; where, V is the phenotypic variance and equals ZZ 0 s 2 g þ WW 0 s 2 b þ TT0s 2 p þ Is 2 e ; I is an identity matrix; and s 2 g ; s 2 b ; s 2 p and s 2 e are the genotypic variances among progenies, blocks, permanent environment, and residuals, respectively. The covariances between these effects are equal to zero.
The mixed model equations associated with this model are [33]: is the coefficient of determination for the effects between blocks; and p 2 ¼ is the coefficient of determination for permanent environmental effects. The estimators of the variance components associated with this model using the expectation-maximization algorithm are [33]: s 2 e ¼ ½y 0 y Àû 0 X 0 y ÀĝZ 0 y ÀbW 0 y ÀpV 0 y�=½N À rðXÞ�; where, C is the generalized inverse of the matrix of coefficients of mixed model equations; tr is the trace matrix operator; r(X) is the rank of matrix X (number of linearly independent columns); N-r(X) is the degrees of freedom of the error; N, q, and r are the total number of data, clones, and blocks, respectively.
The coefficient of genetic variation (CV g ), coefficient of residual variation (CV e ), and coefficient of relative variation (CV r ) were calculated, respectively, with the following equations [33]: CV g ¼ ðs 2 g =mÞx100; CV e ¼ ðs 2 e =mÞx100; and CV r = CV g /CV e ; where μ is the overall average. The genotypic correlation (Pearson's correlation) among the evaluated traits in the conilon and robusta coffee populations were estimated according to Resende [32]. The classification rules of the parameter estimates (heritability, correlation, accuracy) were based on Resende [36], Resende and Alves [37] and Resende and Duarte [38].
Hybrid coffee population. To evaluate the hybrid families, the following linear mixed model was used [32]: where, y is the vector of phenotypic data; u is the vector of the effects of years or harvests (assumed as fixed) added to the overall average; c is the vector of the effects of the specific combining ability between the conilon and robusta genitors (assumed as random); f is the vector of the effects of the general combining ability of the robusta female parent (assumed as random); m is the vector of the general combining ability effects of the conilon male parent (assumed as random); s is the vector of the individual permanent environmental effects (assumed as random); b is the vector of block permanent environmental effects (assumed as random); and e is the vector of residuals (random). The capital letters (X, T, W, Z, Q and S) represent the incidence matrices of these effects.
The average and variance structures associated with this model are [33]: yju; V � NðXu; VÞ; mjs 2 m � Nð0; Is 2 m Þ; fjs 2 f � Nð0; Is 2 f Þ; cjs 2 c � Nð0; Is 2 c Þ; sjs 2 s � Nð0; Is 2 s Þ; bjs 2 b � Nð0; Is 2 b Þ; and ejs 2 e � Nð0; Is 2 e Þ; where, V is the phenotypic variance and equals ZZ 0 s 2 m þ WW 0 s 2 f þ TT 0 s 2 c þ QQ 0 s 2 s þ SS 0s 2 b þ Is 2 e ; I is an identity matrix; and s 2 m ; s 2 f ; s 2 c ; s 2 s ; s 2 b and s 2 e are the variances among the conilon and robusta genitors, the specific combining ability among the conilon and robusta genitors, the individual permanent environment, the block permanent environment, and the residuals, respectively.
The mixed model equations associated with this model are [33]:  ; The estimators of the variance components associated with this model using the expectation-maximization algorithm are [33]: tr ðC 66 Þ�=l; and s 2 e ¼ ½y 0 y Àû 0 X 0 y ÀmZ 0 y Àf W 0 y ÀĉT 0 y ÀŝQ 0 y ÀbS0y�=½N À rðXÞ�; where, C is the generalized inverse of the matrix of coefficients of mixed-model equations; tr is the trace matrix operator; r(X) is the rank of matrix X (number of linearly independent columns); N-r(X) is the degrees of freedom of the error; and N, q, n, m, and l are the total number of data, genitors in the conilon group, genitors in the robusta group, combinations between the conilon and robusta groups, and blocks, respectively.
The coefficient of genetic variation (CV g ), coefficient of residual variation (CV e ), coefficient of relative variation (CV r ) and genotypic correlation (Pearson's correlation) among the evaluated traits in the hybrid coffee population were calculated as described above. The parameter estimates (heritability, correlation, accuracy) were classified as described above.
Likelihood ratio test. The significance of the random effects of the statistical models was tested using the likelihood ratio test (LRT) according to the following expression [34]: where, LogL is the logarithm of the maximum (L) of the restricted likelihood function of the full model; and LogL R is the logarithm of the maximum (L R ) of the restricted likelihood function of the reduced model (without the effect being tested).
Genetic selection. The selection of superior conilon and robusta coffee clones and hybrid families was performed using the Mulamba-Rank index [32]. This index consists of classifying the genotypes in order of performance for each trait then calculating the average rank of each genotype for all categories. Based on the results of this index, the gains from selection were estimated, considering the predicted genotypic values of the selected genotypes.

Conilon coffee population
For all evaluated traits, the genotypic effects were significant (Table 1), indicating the existence of genetic variability among clones. Ramalho et al. [35] also observed genetic variability in their analysis of conilon coffee clones. Except for the trait time of fruit ripening, the effects of block were significant (Table 1). Moreover, except for the traits incidence of cercosporiosis, time of fruit ripening, and fruit size, the effects of the individual permanent environment were significant ( Table 1).
Estimates of genetic parameters (heritability, repeatability, genetic correlation) are fundamental for genetic improvement, because they help to define the best breeding and selection strategies for improvement and predict gains from selection [36]. The traits vegetative vigor, rust and cercosporiosis incidence, plant height, canopy diameter, and yield per plant showed individual broad-sense heritability estimates of low magnitude (0.01�h 2 �0.15) ( Table 2). Meanwhile, fruit ripening time and fruit size showed individual broad-sense heritability estimates of moderate magnitude (0.15<h 2 <0.50) ( Table 2).
The traits rust incidence, cercosporiosis incidence, fruit ripening time, fruit size, and yield per plant presented low estimates for repeatability (ρ�0.30) ( Table 2). The traits vegetative vigor, plant height, and canopy diameter presented medium estimates for repeatability (0.30<ρ<0.60) ( Table 2). The relative coefficients of variation were less than unity for all evaluated traits, given that the largest proportion of phenotypic variation was due to uncontrolled environmental effects (Table 2) In the context of genetic evaluation, the most important statistical parameter for assessing the experimental quality and predicted genotypic values is the selective accuracy [37]. This Table 1

PLOS ONE
Coffea canephora breeding parameter reflects the quantity and quality of information and procedures used in the prediction of genetic values and depends on the heritability and repeatability of the analyzed trait [38]. The average accuracies for the traits plant height, canopy diameter, and yield per plant showed moderate magnitude (0:50 � rĝ g < 0:70) ( Table 2). The average accuracies for the traits vegetative vigor, rust incidence, cercosporiosis incidence, fruit ripening time, and fruit size showed high magnitude (0:70 � rĝ g < 0:90) ( Table 2). Similar results were reported by Ramalho et al. [35] in their evaluation of conilon coffee clones.
Genotypic correlation measures the degree of linear association between two traits, resulting from linkage disequilibrium (transient cause) and/or pleiotropy (permanent cause) [36,39]. The pairs of traits vegetative vigor/canopy diameter and plant height/canopy diameter showed a high magnitude of correlation (0.67 to 1) (Table 3), while the other pairs of traits showed low (0 to 0.33) or medium magnitude (0.34 to 0.66) correlation (Table 3).
By utilizing the Mulamba-Rank index and considering a selection intensity equal to 21% (15 clones), the predicted gain with selection for the genotypic aggregate was equal to 55% (S4 Table in S2 File). This demonstrates the possibility of significant genetic gains with selection in the conilon coffee population. Carias et al. [40] compared the Additive, Multiplicative, and Mulamba-Rank indices in the selection of conilon coffee genotypes and concluded that the Mulamba-Rank index was the most effective. Ramalho et al. [35], evaluating conilon coffee clones for the trait production of processed coffee, and considering a selection intensity of 10%, estimated gains with selection equal to 58%.

Robusta coffee population
Except for the trait canopy diameter, the genotypic effects were significant (Table 4), indicating the existence of genetic variability among clones. Mistro et al. [20], evaluating conilon coffee progenies for yield per plant, also found genetic variability. Only the traits canopy diameter and fruit ripening time showed significant block effects (Table 4). In addition, the traits vegetative vigor, cercosporiosis incidence, plant height, and canopy diameter showed significant permanent environmental effects ( Table 4).
The traits vegetative vigor, rust incidence, cercosporiosis incidence, plant height, canopy diameter, and yield per plant showed low magnitude estimates of individual broad-sense heritability (0.01�h 2 �0.15) ( Table 5). For the traits fruit ripening time and fruit size, the individual broad-sense heritability estimates were of moderate magnitude (0.15<h 2 <0.50) ( Table 5). Mistro et al. [20], evaluating progenies of conilon coffee regarding yield per plant, obtained a moderate magnitude heritability estimate.

Table 5. Estimates of variance components and genetic and environmental parameters for the following traits: Vegetative vigor (VV), incidence of rust (IR), incidence of cercosporiosis (IC), plant height (PH), canopy diameter (CD), fruit ripening time (FT), fruit size (FS), and yield per plant (YP), evaluated in the robusta
The pairs of traits vegetative vigor/canopy diameter and plant height/canopy diameter showed high magnitude correlation (0.67 to 1) ( Table 6), while the other pairs of traits showed low (0 to 0.33) or medium magnitude (0.34 to 0.66) correlation (Table 6).
Using the Mulamba-Rank index and considering a selection intensity equivalent to 27% (15 clones), the predicted gain with selection for the genotypic aggregate was equal to 32% (S5 Table in S2 File), a fact that demonstrates the possibility of significant genetic advancement with selection in the robusta coffee population.

Hybrid coffee population
The selection of hybrids resulting from crosses between individuals of different varietal groups is particularly relevant for the success of C. canephora breeding programs [19]. In this study, the effects of specific combining ability were significant for the traits vegetative vigor and plant height (Table 7), indicating the existence of heterosis for these traits. The effects of general combining ability of conilon and robusta parents were significant for the traits rust incidence and fruit ripening time, respectively (Table 7). Permanent environmental effects were significant for the traits vegetative vigor, rust incidence, plant height, canopy diameter, and fruit ripening time ( Table 7). The effects of block were significant only for yield per plant ( Table 7).

Genotypic correlation among the following traits: Vegetative vigor (VV), incidence of rust (IR), incidence of cercosporiosis (IC), plant height (PH), canopy diameter (CD), fruit ripening time (FT), fruit size (FS), and yield per plant (YP), evaluated in the robusta
The estimates of individual narrow-sense heritability were greater than 10% for plant height and fruit ripening time in the conilon population, and vegetative vigor and rust incidence in the robusta population ( Table 8). The individual heritability estimates for the interpopulation Table 8  dominance effects were greater than 10% only for the traits vegetative vigor, plant height, and canopy diameter (Table 8), indicating the occurrence of heterosis for these traits.

Genetic Parameters
The traits rust incidence, fruit ripening time, fruit size, and yield per plant presented low repeatability estimates (ρ�0.30) (Table 8). Meanwhile, the traits vegetative vigor, plant height, and canopy diameter presented medium repeatability estimates (0.30<ρ<0.60) ( Table 8). The relative coefficients of variation were less than unity for all evaluated traits, given that the largest proportion of phenotypic variation was due to uncontrolled environmental effects ( Table 8).
The pairs of traits vegetative vigor/plant height, vegetative vigor/canopy diameter, and plant height/canopy diameter showed high magnitude correlation (0.67 to 1) (Table 9), while the other pairs of traits showed low (0 to 0.33) or medium magnitude (0.34 to 0.66) correlation (Table 9).
Using the Mulamba-Rank index and considering a selection intensity equal to 50% (10 families), the predicted gains with selection for the genotypic aggregate was equal to 17% (S6 Table in S2 File), demonstrating the possibility of significant genetic gains with selection in the hybrid population.

Breeding strategies for productivity and disease resistance traits
In C. canephora, especially in Brazil, there is a lack of comparative studies involving alternative breeding strategies stemming from intrapopulation and interpopulation improvement programs in this species. The objectives of this study were to evaluate parents and families from the populations of conilon, robusta, and its hybrids and to define the best breeding and selection strategies for productivity and disease resistance traits.
C. canephora breeding germplasm is based on two contrasting groups native from Africa: the Guinean group (from Occidental Africa) and Congolese group (from Central Africa). In terms of breeding strategies of these species, three main ones can be used: intrapopulation recurrent selection within groups (RS); interpopulation (reciprocal) recurrent selection between groups (RRS); intrapopulation recurrent selection within the hybrid (crossed) population (RSH). RRS and RSH are suitable to improve heterosis. Studies performed in Ivory Coast showed that hybrids between these groups express heterosis [10,42]. In Brazil, the main representant of the Congolese group is known as conilon and the main representant of the Guinean group is known as robusta. Hybridization between parents of these groups has been performed in Brazil [19], as a mean to get complementary traits and also to exploit heterosis. Then, RRS   or RSH strategies can be used. Currently, the RS with cloning of the best clones is the approach more used. The best breeding strategy can be proposed by looking at the results from the three populations. The main results that can help the breeder are: high or favorable general mean of the population; good (higher than 2/3) selective accuracy in the main traits (this reflects also good genetic control and high genetic variability); favorable (and higher than 1/2) correlation between main traits; significant specific combining ability and high level of dominance, which demonstrate the presence of heterosis.
Significant specific combining ability and high level of dominance were found in the hybrid population for the traits vegetative vigor and plant height (Tables 7 and 8), indicating the existence of heterosis for these traits. Then in such aspect they are suitable for RRS or RSH strategies. Looking at the relevance of these two traits for the breeding we should analyze their correlation with the two principal traits (breeding objectives): coffee yield and rust resistance.
For conilon it can be seen no meaningful correlation between the traits (vegetative vigor and plant height) subjected to RRS or RSH strategies, with yield per plant and incidence of rust (all correlation below 0.50). The own vegetative vigor and plant height, had correlation of 0.58. These results reveal that the RRS or RSH strategies for vegetative vigor and plant height will not jeopardize the improvement of yield per plant and incidence of rust. However, they will not help either. Yield per plant and incidence of rust shows correlation of -0.20, then allowing the simultaneous improvement of both.
For robusta it can also be seen no meaningful correlation between the traits (vegetative vigor and plant height) subjected to RRS or RSH strategies, with yield per plant and incidence of rust (all correlation below 0.50). These results reveal that the RRS or RSH strategies for vegetative vigor and plant height will not jeopardize the improvement of yield per plant and incidence of rust (correlation of 0.24 between them), whereas they will not help either. For hybrid, the same conclusions can be drawn.
Concerning the selective accuracy for the traits, for conilon, the only below the 0.67 threshold were plant height and canopy diameter. This reveals a very favorable situation for breeding. Concerning the general mean of the population, the values are 8.11 and 2.06 for the breeding objectives yield per plant and incidence of rust. For robusta, the selective accuracy of the traits showed better results for disease resistance and fruit traits. Considering both populations together, only plant height and canopy diameter are at unfavorable situation. Then, it seems, in this aspect, that only vegetative vigor is suitable for SRR. For the heterotic traits vegetative vigor and plant height the general means are 5.79 and 146.64, respectively, for conilon; 5.75 and 158.46, respectively, for robusta; and 6.26 and 158.24, respectively, for the hybrid. All values are similar for the three populations, except by the plant height in robusta, which is higher.
For robusta, the general mean of the population showed the values are 7.07 and 1.25 for the breeding objectives yield per plant and incidence of rust. Then, the conilon showed slightly better for yield per plant and the robusta showed slightly better for incidence of rust. In this way, the selection and breeding in the hybrid population can be recommended as ideal, given the complementary involving yield per plant and incidence of rust. Results for the hybrid population confirm this, as the general mean of the population gave the values are 8.24 and 1.45 for the breeding objectives yield per plant and incidence of rust. We can conclude that the conilon and the hybrid are slightly better than robusta for yield per plant. After the choice to conduct the improvement of the Coffee canephora in the hybrid population, a decision should be made between RRS or RSH. As only vegetative vigor is suitable for RRS to increase heterosis and it is not the main trait, we choose the best breeding strategy as RSH in the composed new hybrid population originated from crossing conilon x robusta. This choice allows and enables the simultaneously improvement of all traits of interest. Such strategy is simple, low cost and quicker than the concurrent reciprocal recurrent selection in the two parental populations, and that maximizes the genetic gain for unit of time.
To exploit and improve heterosis in C. canephora, in Ivory Coast (Africa), Leroy et al. [42] have chosen the RRS, i.e, decided to invest in hybrids. Expressive genetic gain has been reported [10,43]. Our strategy has also strong features, such as relying on hybrids and being more cost effective and quick.
Concerning the selective accuracy for the traits, for the hybrids the values were below the 0.67 threshold. However, genomic selection can circumvent this by increasing accuracy and speeding the breeding cycle [22,23,25].
In conclusion, although the two parental populations themselves have shown low or absent genetic variability, the hybrid population exhibited higher variability and significant specific combining ability for some traits. Then, a sound breeding program can be run for this genetic material. And the best choice of breeding strategy would be the RSH because it is genetically efficient as it keeps and increases the heterosis in the population, is simple, low cost and quicker than RRS, and this maximizes the genetic gain for unit of time.