Restricted Maximum Likelihood/best Linear Unbiased Prediction Research Articles

Selection of Cordia trichotoma Vell. clones for adventitious rooting and determination of the plantlet quality produced by mini-cuttings

This study aimed to evaluate and define strategies for the selection of C. trichotoma clones with competence for adventitious rooting and production of high-quality plantlets by mini-cuttings. For adventitious rooting, mini-cuttings were treated with 3,000 mg L-1 of indole butyric acid and grown in 110 cm³ tubes, consisting of equal proportions of the commercial substrate and vermiculite. The percentage of rooting, number of roots, and mini-cuttings rooted per mini-stump were evaluated after 30 days of cultivation in a humid chamber. The rooted mini-cuttings were transferred to a shade house and the survival percentage, stem diameter, height, and number of leaves were evaluated at 120 days. Data were analyzed using the restricted maximum likelihood/best linear unbiased prediction (REML/BLUP) method and Pearson´s correlation analysis. The number of rooted mini-cuttings and stem diameter presented the highest repeatability and accuracy values for rooting and plantlet growth, respectively. These traits also presented significant positive correlation estimations with others, indicating the possibility of indirect selection gain. Cordia trichotoma clones can be selected for vegetative propagation, considering the number of rooted mini-cuttings and the stem diameter of the plantlets produced by mini-cutting. Early selection for adventitious rooting and plantlet quality can aid in genetic improvement programs for the development of new cultivars for vegetative propagation via the use of mini-cuttings.

Forage peanut genetic variability: Multi-trait selection for forage production and ornamental purposes

The selection of superior genotypes considering several traits simultaneously allows the release of more adapted, productive, and nutritive cultivars. To select forage peanut genotypes for use as animal feed and for ornamental purposes, 67 genotypes were evaluated. Twelve agronomic and nutritive value traits were evaluated in three temporally separated trials: a randomized complete block design. The mixed model method (Restricted Maximum Likelihood/Best Linear Unbiased Prediction (REML/BLUP)) was employed for estimating genetic parameters and predicting genotypic values, which were used in the selection indexes. The selection criteria considered vegetative and seed propagation traits corresponding to the use of forage peanut as animal feed and for ornamental purposes. Seed productivity was also evaluated and correlated with other traits. Genetic variability with high heritability in seed production was observed. Plant vigor, ground cover, and dry matter yield were negatively correlated with seed production. Among the genotypes evaluated, there is a possibility of selecting 14 and 16 genotypes corresponding to seed propagation for animal feed and ornamental purposes, respectively. For vegetative propagation, there are 16 and 14 promising genotypes for animal feed and ornamental purposes, respectively. Vegetatively propagated genotypes generally performed better in terms of agronomic traits related to feed production.

Genetic variability and <i>ex situ</i> conservation strategies for the neotropical tree <i>Parkia platycephala</i> Benth.

Parkia platycephala, known commonly as faveira, is a native species that occurs widely throughout Brazil and has high nutritional value and ecological potential. The objective of this study was to estimate the genetic variability in a P. platycephala provenance and progeny test. The test was established using a randomized block design, with 45 progenies from three provenances. The evaluated quantitative traits included height (HEI), diameter at ground level (DGL), and diameter below the first bifurcation (DFB). The data were submitted to Restricted Maximum Likelihood/Best Linear Unbiased Prediction (REML/BLUP) analysis, from which estimates of genetic parameters, genetic correlations, BLUPs, genetic gains, and effective population size were obtained. Estimates of individual narrow sense heritability (h2a) and within progeny heritability (h2w) ranged from low to moderate (0.01 a 0.39), with the highest values observed for DGL (0.32 and 0.39, respectively). The coefficient of individual variation (CVgi(%)) and between progeny genetic variation (CVgp(%)) indicate the existence of genetic variability between and within provenances. The FST vaues demonstrate low to moderate genetic differentiation among the three populations, and genetic correlations between traits were all positive and significant. The results obtained show that there is genetic variability between the studied provenances and progenies. This ensures not only the ex situ conservation of the species, but it can also be exploited in breeding programs.

SELECTION BETWEEN AND WITHIN SEGREGATING SOYBEAN POPULATIONS VIA SIMULATED INDIVIDUAL BLUP (BLUPIS)

The use of methods with high statistical accuracy in the selection of genotypes can provide substantial gains to breeding programs. This study evaluated the suitability of the best linear unbiased prediction at individual level (BLUPI) with simulated individual BLUP (BLUPIS) procedure to select individuals in segregating soybean populations. Fifteen segregating populations were evaluated in the field during the 2014/15 crop season. Data for grain yield (g/plant) and growth cycle in days (divided in early and late) were collected and analyzed via restricted maximum likelihood/best linear unbiased prediction (REML/BLUP). In general, 67–92% of the individuals were selected in the three best populations. The efficiency of BLUPIS relative to BLUPI was 4.66 and 1.46 for grain yield and growth cycle, respectively. There was strong correlation between BLUPIS and BLUPI for grain yield. The selection by BLUPIS was more efficient than BLUPI for all the traits studied and contributes to the selection of individuals in representative numbers of the best populations, thereby providing substantial gain for soybean breeding programs.

Genetic variability of traits of the laticiferous system and association with rubber yield in juvenile and adult rubber tree progenies

The objectives of this study were to investigate genetic variation and association of the traits of the laticiferous system with rubber yield in young and adult rubber tree progenies Hevea brasiliensis, and compare the traits of the laticiferous system at different heights of the adult trees. The genetic material is composed of progenies from open pollination from the rubber tree breeding program of IAC, evaluated in the experimental field of UNESP in Ilha Solteira, SP, Brazil, in the young and adult phases, and in the Polo Regional Alta Mogiana of the Agência Paulista de Tecnologia do Agronegócio in Colina, SP, Brazil in the young phase. A randomized block experimental design was used with 30 treatments (progenies), three replications, and 10 plants per plot. The following traits were evaluated: rubber yield (RY), virgin bark thickness (BT), number of rings of laticiferous vessels (NR), and mean distance between consecutive rings of the laticiferous vessels (MD). The individual statistical analyses and repeated measurements were performed through mixed linear models using the REML/BLUP (Restricted Maximum Likelihood/Best Linear Unbiased Prediction) procedure. Genetic and phenotypic correlations were also estimated. The accuracy estimates (Aˆc) for the traits in the two locations were higher (> 0.7), indicating good perspectives in a breeding program. The coefficient of genetic variation (CVg %) showed low variation between the sites: BT (4.5–7 %), NR (14.1–18.3 %), and MD (15–18.2 %). The NR expressed higher heritabilities and was therefore the best trait of the laticiferous system for predicting genetic values; it showed expressive genetic correlations with RY, and may be used in indirect selection. BT did not show progeny × height interaction, indicating that the classification of the progenies was the same, regardless of the tree height at collection. To select progenies in common in the young and adult ages, using analysis of measurements repeated in time, the most suitable height for collection of samples in the adult progenies for the traits of the laticiferous system was up to 80 cm. The traits of the laticiferous system collected at 30 cm in young progenies are associated with RY, both in the youth phase and in the adult phase. They can be used to assist in early or indirect selection of rubber tree progenies.

Herdabilidade estimada por diferentes métodos em quatro gerações de progênies de um cruzamento de guandu

Abstract The objective of this work was to compare different methods to estimate heritability in 30 pigeon pea families from the F3, F4, F5, and F6 generations, for nine variables. The experimental design was a randomized complete block with three replicates and 20 plants per plot. Broad-sense heritability was estimated by the analysis of variance ( A N O V A) [ h 2b-E(MS)], restricted maximum likelihood/best linear unbiased prediction (REML/BLUP) (h2b-REML), parent-offspring regression (h2PO), and standard deviation unit (h2UP). The h2b-E(MS) and h2b-REML estimates were similar for seven of the analyzed variables. For a higher genetic control and easier selection, values of h2b-E(MS) and h2b-REML >0.70 were estimated for two variables in four generations, two variables in three generations, three variables in two generations, and one variable in one generation. Values of h2UP and h2PO >0.70 were obtained for four and five variables, respectively. The estimates via regression or parent-offspring correlation showed some values outside the expected range of 0 to 1. The ANOVA [h2b-E(MS)] and REML/BLUP [h2b-REML] methods are the best to estimate pigeon pea heritability.

Use of an index based on best linear unbiased prediction value for the selection of superior papaya lines for multiple traits

Selection indexes represent the real efforts of a breeding program to obtain genetic gains for various significant traits simultaneously. As such, this study selected superior F5 lines by combined selection for fruit quality using an index based on predicted genotypic values via residual or restricted maximum likelihood/best linear unbiased prediction (REML/BLUP) methodology weighted by agronomic weights. To do so, 97 F5 papaya lines obtained by the single seed descent method, resulting from the biparental cross between the genotypes JS-12 and Sekati, were evaluated for the main traits related to fruit quality. Results of the analysis of deviance and the genetic parameters indicated that there was genetic variability, indicating possible success in the selection process. Based on the selection index, 29 lines were selected as superior for fruit quality with ‘Formosa’ pattern. The strategy of the selection index combining BLUPs of multiple traits of interest associated with agronomic weights allows maximizing genetic progress while discarding less-promising genotypes.

Theobroma grandiflorum breeding optimization based on repeatability, stability and adaptability information

The cultivation of Theobroma grandiflorum in the Brazilian Amazon is mainly conducted by family farmers who use a range of different management strategies. Thus, breeding programs of the species must address the challenge of developing cultivars that are adapted and stable in a variety of cultivation environments. In this context, this study aimed to estimate the optimum number of harvests for genetic selection of T. grandiflorum progenies and identify the most promising ones in terms of productivity, stability, and adaptability. The trials were implemented in three environments, using a randomized complete block design, with 25 full-sib progenies, five replications, and three plants per plot. The traits mean number of fruits/plant, mean fruit production/plant, and rate of infection with witches’ broom (Moniliophthora perniciosa) were evaluated over 11 harvests. The Restricted Maximum Likelihood/Best Linear Unbiased Prediction (REML/BLUP) procedure was used to estimate genetic parameters and predict genetic values, which were then applied to assess stability and adaptability. The results show that there is genetic variability among the studied T. grandiflorum progenies and that accurate genetic selection aiming at recombination and recommendation is effective after three and eleven harvests, respectively. Six progenies were selected that met the requirements for productivity, stability, and adaptability to different cultivation environments. These results can be used to optimize and advance T. grandiflorum breeding programs.

Genetic (co)variation and accuracy of selection for resistance to viral mosaic disease and production traits in an inter‐ecotypic switchgrass breeding population

AbstractObtaining good accuracy and reliability of estimated breeding values is essential to increase the efficiency of a plant breeding program. Genetic variation was assessed for categorical (Virc) and binary (Virb) mosaic (caused by Panicum mosaic virus), dry matter (DMY) and predicted ethanol (Etoh) yields, and lignin content (Klason or KL, and acid‐detergent or ADL) in a Summer–Kanlow switchgrass (Panicum virgatum L.) population. Breeding values were predicted with the restricted maximum likelihood–best linear unbiased prediction (REML‐BLUP) approaches using a multivariate phenotypic (PBLUP) and animal (ABLUP) models, integrating a three‐generation pedigree (1,622 half‐sibs) in ABLUP and not in PBLUP. Models were compared in their precision (accuracy and reliability) in assessing genetic parameters and estimating breeding values. The models were similar in most aspects, allocating the highest heritability () values to DMY (.38 ± .035 vs. .41 ± .035), Etoh (.46 ± .031 vs. .42 ± .033), and Virc (.43 ± .046 vs. .37 ± .047) and the lowest (.17 ± .032 to .30 ± .044) to KL, ADL, and Virb. Genetic correlations were always larger than residual and phenotypic correlations. Intermediate or strong additive genetic control suggest that selecting for high‐biomass genotypes will slightly increase lignin content and simultaneously impart mosaic tolerance. Mitigating an increase in lignin content will require including Etoh in a selection index based on its much stronger negative correlation (rG = −.63) with lignin. In this population, accuracy values ranged from .06 to .94 (PBLUP) and from .26 to .92 (ABLUP) and corresponding reliability ranged from .004 to .89 and from .07 to .87. However, ABLUP improved average reliability of DMY and Etoh by 11% and of other traits by 4–5% over the PBLUP model. The ABLUP was a better model over PBLUP, which is a valid analysis in the absence of a pedigree.

Use of the REML/BLUP methodology for the selection of sweet orange genotypes

Abstract: The objective of this work was to select superior sweet orange (Citrus sinensis) genotypes with higher yield potential based on data from eight harvests, using the residual or restricted maximum likelihood/best linear unbiased prediction (REML/BLUP) methodology. The experiment was carried out from 2002 to 2008 and in 2010 in the municipality of Rio Branco, in the state of Acre, Brazil. Analyzes of deviance were performed to test the significance of the components of variance according to the random effects of the used model, and parameters were estimated from individual genotypic and phenotypic variances. A selection intensity of 20% was adopted regarding genotypic selection, i.e., only the best 11 of the 55 genotypes tested were selected. The estimates of the genetic parameters show the existence of genetic variability and the selection potential of the studied sweet orange genotypes. The genotypic correlation between harvests is of low magnitude, except for the variable average fruit mass, and, as a reflex, there is a change in the ordering of the genotypes. Genotypes 5, 48, 19, 14, and 47 stand out as being the most productive, and, therefore, are the most suitable for selection purposes. Genotypes 14 and 47 show superior performance for the character set evaluated.

Genetic progress in upland rice breeding program for grain yield and plant height

ABSTRACT In the context of plant breeding programs, it is necessary to evaluate the efficiency of genotype selection over successive years. However, evaluating the genotype selection efficiency is not an easy task, since there is not just a single way to precede it. Besides that, the programs need to be dynamic; that is, they should be able to track the introduction and discard of genotypes each year. As a result, the available data is quite unbalanced, leading to difficulties in certain analyses. Thus, the present study aims to propose some approaches to verify the genetic progress in the preliminary trial of the Federal University of Lavras (UFLA) upland rice breeding program. We utilized mixed models for grain yield and plant height. Trials were performed with a total of 120 genotypes in seven environments, defined by the interaction between locations and years. Due to the imbalance in the available data, the mixed model approach, i.e., Restricted Maximum Likelihood/Best Linear Unbiased Prediction (REML/BLUP), was adopted for the joint analysis. Besides the genetic and phenotypic parameters, the expected gains were also obtained with the selection, genetic progress, renewal rate (RR%), and dynamism of preliminary trials. The efficiency of the selection of superior genotypes per year was verified, with genetic progress favorable for reducing the medium-sized plants associated with high yield.

Reaction norms‐based approach applied to optimizing recommendations of cotton genotypes

AbstractReaction norms fitted through random regression models (RRM) have been widely used in animal and plant breeding for analyses of genotype × environment (G × E) interaction. However, in annual crops, they remain unexplored. Thus, this study aimed to evaluate the applicability and efficiency of RRM fitted through Legendre polynomials as a tool to recommend cotton (Gossypium hirsutum L.) genotypes. To this end, a data set with 12 genotypes of cotton evaluated in 10 environments for fiber length (FL) and fiber fineness was used. The restricted maximum likelihood/best linear unbiased prediction (REML/BLUP) procedure was used to estimate the variance components and to predict the genetic values. Results showed that there was genetic variability among cotton genotypes and that the reaction norms over the environmental gradient illustrated the G × E interaction. Very high selective accuracies () were found for both traits in all environments, which indicates high reliability in the genotype's recommendation. The areas under the reaction norms were calculated for the recommendation of genotypes for unfavorable, favorable, and overall environments. Regarding genotypes recommendation, areas under reaction norms allow recommending genotypes for unfavorable and favorable environments, as well as for overall recommendation, for both traits. This study is the first considering reaction norms fitted through RRM for the recommendation of cotton genotypes and demonstrated the potential of this technique in cotton breeding, besides its great potential to deal with G × E interactions.

Association between physiological and agronomic traits and selection of tropical wheat

The aims of this study are to verify the existence of canonical correlations between physiological and agronomic traits of 40 lines of tropical wheat using the REML/BLUP (restricted maximum likelihood/best linear unbiased prediction) method and select genotypes with the best performance using four selection indexes. For such, 40 tropical wheat genotypes were evaluated in a field experiment in Vicosa, MG, Brazil. The physiological traits (group I) of these genotypes were measured using an infrared gas analyzer, group II was formed by agronomic traits. The data were submitted to REML/BLUP and the predicted genotypic values (BLUP) were obtained to estimate genotypic correlation coefficients, canonical correlation coefficients between groups I and II, and to perform the selection of superior genotypes using the method of ranks summation index, multiplicative index, genotype-ideotype distance index and Z index. The maximum likelihood test revealed a significant effect of genotypes for all traits evaluated. Net photosynthetic rate it showed a positive correlation with stomatal conductance to H2O (0.32) and transpiration rate (0.79). Hectoliter weight has a significant association with physiological traits, positive for intercellular CO2 concentration (0.56) and negative for leaf temperature (− 0.56) and transpiration rate (− 0.42). The correlation between groups was 0.78. The intercellular CO2 concentration was directly related to disease note and hectoliter weight. The lines VI 14050, VI 14197 and VI 14950 coincide in the three selection indexes, with a potential for registration as a new cultivar.

Genetic control of wood quality of Myracrodruon urundeuva populations under anthropogenic disturbance

Abstract Forest fragmentation in the Cerrado biome exposes the tree species Myracrodruon urundeuva to the risk of extinction. Thus, the objective of the study was to evaluate silvicultural characteristics related to wood quality, based on genetic parameters estimated in two M. urundeuva progenies from natural populations, assessed in Bauru-SP (PBAU) and Selviria-MS (PSEL), 31 years after planting. The genetic parameters were estimated according to the REML/BLUP (Restricted maximum likelihood/Best Linear Unbiased Prediction) method using a mixed linear model. The progeny tests detected genetic variation. In PSEL, more positive genetic correlations and a greater number of clusters were observed. Based on the heartwood/sapwood ratio, a higher genetic gain was predicted for PBAU, and for the basic wood density, the genetic gain of the populations was approximately the same. Consequently, the areas of the two progeny tests may be used as future seed orchards to ensure wood quality and long-term ex situ conservation.

Adaptability and stability of Coffea canephora genotypes cultivated at high altitude and subjected to low temperature during the winter

Climate change prospections indicate large fluctuations in terrestrial temperature, what may change the zoning of coffee cultivation. This is particularly true for Coffea canephora (conilon or robusta coffee) to which some suitable areas may become inappropriate. Thus, conilon cultivation at high altitudes may be an option to establish C. canephora, mitigating the possible effects of productive areas loss. In this context, the aim of this work was to evaluate the productivity, adaptability and stability of 28 C. canephora genotypes cultivated at 850 m above sea level, under climatic conditions where the minimum temperature is the limiting factor for the species. A random block design with four replicates x five plants per replicate was used. The genotype x environment interaction was investigated using stability and adaptability analyzes based on the mixed model of restricted maximum likelihood / best linear unbiased prediction (REML/BLUP). The genotypes presented heritability of high magnitude, indicating the precision of selection. Six genotypes (NV2, NV8, P1, 3V, Verdim TA and A1) were highlighted for productivity, adaptability and stability at high-altitude and low temperature.

Estimation of genetic parameters and prediction of genotypic values in common beans using mixed models#

In the international scenario of agriculture, Brazil stands out as the main producer and consumer of common bean (Phaseolus vulgaris L.) The increase in the productive potential of the crop is mainly due to breeding programs. The objective of this study was to estimate genetic parameters, predict genotypic values ​​with REML/BLUP (Restricted Maximum Likelihood/Best Linear Unbiased Prediction) and, based on these values,​​ study the variability in common bean cultivars with carioca and black grain. Twenty three agromorphological descriptors were evaluated, among them grain yield. Deviance analysis detected significant differences between the cultivars in both groups. Selective accuracy (Ac) was considered high for most of the traits. Broad-sense heritability ( ) ranged from 0.05 to 0.72, but it was low for the trait yield (YLD). In the carioca grain group, the ​​ values for the traits related to plant morphology were higher than in the black group. Nevertheless, the ​​ values in the black group were higher in relation to the pod and seed traits. The correlations for YLD were moderate but different in the two commercial groups studied. In the black group, variables related to the seed morphology were correlated with grain yield, and in the carioca group, traits related to seed quantity. Based on the groupings, variability among the cultivars was observed. Three distinct clusters were formed for the carioca group and four for the black group. Based on the predicted genetic values, genetic variability and the most adapted and stable cultivars were detected among the cultivars in the studied environments.
 
 #This article has been posted in Preprints with a doi 10.20944/preprints201803.0215.v1 (https://www.preprints.org/manuscript/201803.0215/v1) (Creative Commons CC BY 4.0 license)
 *Corresponding author, Email: leandrosag@uel.br

Estimates of repeatability for selection of genotypes of seedless table grapes for Brazilian semiarid regions

Grapevine (Vitis vinifera L.) produces one of the most consumed fruits in the world. Nonetheless, there is a lack of table grape cultivars that are adapted to Brazilian semiarid conditions. Therefore, there is a need for new table grape cultivars for formation of commercial vineyards in these areas. Thus, the objective of this study was to select promising genotypes for cultivation under semiarid conditions using the methodology of mixed models: restricted maximum likelihood/best linear unbiased prediction (REML/BLUP) models. Eighty-one individual plants (genotypes) were evaluated, in three crop seasons, during the years 2016 and 2017, for six descriptors. The estimates of the repeatability coefficient were high for most of the key traits during table grape breeding, except for total soluble solids (°Bx), 0.28, and the ratio of total soluble solids to titratable acidity (0.30). Predicted selection gains for the 30 superior genotypes provided maximum average values of 46.84 t ha―1 per crop (the yield), 150.64 bunches per plant (number of bunches), and 745.21 g (bunch weight). The selective accuracy predicted for the six evaluated traits revealed a good degree of confidence in the inferences and high precision and selection gain. Using the minimum selection indexes proposed in this study for three crop seasons, genotypes CPATSA 01.02, 02.15, 05.02, 06.123, 15.05, 15.06, 19.06, 19.10, 21.42, and 68.06 were found to be the most promising. In this group, genotypes CPATSA 15.05 and 15.06 stood out because they are completely devoid of seeds. Therefore, these genotypes will be chosen for advanced selection to prove their potential for cultivation in Brazilian semiarid regions.

Repeatability coefficients and genetic gains in table grape progenies for the Brazilian semi-arid region

Table grape stands out among the main fruit-bearing plants of irrigated agriculture in the Sao Francisco Valley region of Brazil. This study estimated repeatability and heritability coefficients and predicted genetic gains in order to select superior genotypes in grape progenies from controlled hybridizations using the Restricted Maximum Likelihood/Best Linear Unbiased Prediction (REML/BLUP) methodology. Individual plants were evaluated for the variables of production (kg per plant), number of bunches, bunch weight (g), berry diameter (mm), and soluble solids content (°Brix). We evaluated 194 hybrids from 30 crosses between Vitis vinifera and interspecific hybrids in Juazeiro, Bahia State, Brazil, over four growing seasons. Repeatability coefficients, ranging from 0.164 for soluble solids content to 0.72 for berry diameter, were estimated with accuracy values higher than 80 % for all variables, except for soluble solids content (66 %). The 30 best individuals classified for each variable exhibited genetic gains and their new estimated mean values were higher than the overall mean of the population in all variables. Regarding production and number of bunches per plant as the main variables, 15 genotypes were selected simultaneously for both variables. Among which, hybrids CPATSA 15.05, 15.06, 15.06T, and 23.103 stand out because they have trace seeds and should be selected for the following steps of genetic breeding to develop new table grape cultivars for the Brazilian semiarid region.

Selection of Jatropha curcas families based on temporal stability and adaptability of genetic values

Despite its numerous important traits for biodiesel production, little research exists on the Jatropha curcas crop, mainly with respect to repeated measures over time (crop years). The present study was thus developed to evaluate repeated measures over time in Jatropha curcas and examine the applicability and efficiency of the harmonic mean of the relative performance of genetic values (HMRPGV) method in the selection of families for yield, stability, and adaptability, simultaneously. The work involved data from the evaluation of 730 individuals of 73 half-sib families, in a complete randomized block design, during six crop years, for the trait grain yield. Restricted maximum likelihood/best linear unbiased prediction (REML/BLUP) procedure was performed to estimate the variance components and predict the genetic values. Ten crop years must be evaluated to obtain 80% of the maximum coefficient of determination and 90% accuracy in terms of genetic gain from selection. The efficiency of performing 10 measurements compared with the situation in which only one measurement is used was 67%. The HMRPGV method has great potential in the selection of Jatropha curcas families for yield, stability, and adaptability, simultaneously.

GENETIC DIVERGENCE AMONG COWPEA GENOTYPES BASED ON CHARACTERS ASSOCIATED WITH THE GREEN POD AND GRAIN MARKET

ABSTRACT The objective of this study was to evaluate genetic divergence among cowpea genotypes and to select parents for crosses aimed at the fresh pod and grain market. Two experiments were carried out during 2014, corresponding to two sowing times, in the municipality of Mossoró, State of Rio Grande do Norte. Twenty-three cowpea genotypes were evaluated in a randomized complete block design with four replicates. Fifteen descriptors were used to quantify divergence, using the Mahalanobis distance as a measure of dissimilarity, obtained from the genotypic mean predicted by the Restricted Maximum Likelihood/Best Linear Unbiased Prediction (REML/BLUP) method. The Unweighted Pair Group Method with Arithmetic Mean (UPGMA) hierarchical method was used to group genotypes and the Singh criterion to quantify the contribution of traits to genetic divergence. The genotype × environment interaction (G × E) influenced divergence, both in the contribution of traits and in the grouping of genotypes. In the experiment 1, the genotypes were distributed among four groups. In the experiment 2, less discrimination occurred and the genotypes were represented by only two groups. When joint analysis of two evaluations was carried out based on two sowing times, genotypes were represented by six groups. The G × E interaction influences the contribution of traits and the grouping of cowpea genotypes in the study of divergence. The genetic divergence of the set of cowpea genotypes evaluated is mainly due to green grain and pod yield. BRS Aracê and BRS Xiquexique cultivars are the most divergent among the genotypes studied, representing 75% of the recommended crosses.