Natural Genetic Diversity of Nutritive Value Traits in the Genus Cynodon

: The Cynodon spp. collection maintained by United States Department of Agriculture National Plant Germplasm System (USDA-NPGS) has limited information on nutritive value (NV) traits. In this study, crude protein (CP), phosphorous concentration (P), in vitro digestible organic matter (IVDOM), and neutral detergent ﬁber (NDF) were determined to (i) estimate genetic parameters for NV, (ii) obtain genetic values for the whole population across two harvests, (iii) estimate genotype by harvest interaction (GHI) for NV traits, and (iv) select accessions exhibiting improved NV traits compared to ‘Tifton 85 (cid:48) . The experiment was setup as a row-column design with two replicates and augmented representation of controls: Tifton 85, ‘Jiggs’, and ‘Coastal’. The whole-population was harvested twice, and data were analyzed using linear mixed models with repeated measures. In addition, a selected population of 15 genotypes were evaluated across 11 harvests to determine the extent of GHI. Genetic parameters revealed the presence of signiﬁcant genetic variability, indicating potential improvements for NV through breeding. Speciﬁcally, P and IVDOM presented large variation, while NDF had lower diversity but some accessions exhibited lower NDF than Tifton 85. Low GHI, except for IVDOM, indicated genotypic stability and potential for selecting improved accessions under fewer harvests. Breeding line 240, PI-316510, and PI-3166536 presented superior NV than Tifton 85. 146 Cynodon clonal core the Cynodon in Gri ﬃ and 3 cultivars: Tifton and


Introduction
The investigation of natural genetic diversity present in germplasm banks is a key step to improve traits with narrow genetic variability. The use of germplasm banks is even more important for perennial forages, as several important releases worldwide originated from selections of plant introductions made in large collections [1]. Bermudagrass (Cynodon spp.) is the most widely used warm-season perennial forage for hay and pasture in the southeastern United States, covering more than 12 million hectares [2]. Its popularity among livestock and hay producers lays on high biomass production, nutritive value, animal performance, fast-curing for hay production, and drought tolerance [3][4][5][6][7]. The genus Cynodon is composed of genetically diverse species of variable ploidy level [8,9]. The most agronomically valuable species are Cynodon dactylon Pers. and Cynodon nlemfuensis Vanderyst. Both of The soil was a Chipley sand (thermic, coated Aquic Quartzipsamments) with a pH of 6.9 and characterized by high P2O5 content (164 kg ha −1 ), and low K2O (38.2 kg ha −1 ), S (5.6 kg ha −1 ), and Mg (44.8 kg ha −1 ) content. The experimental plot size was 1.8 m × 3.0 m. Planting was done on 2 July 2014 using a single plug 5 × 5 cm planted in the center of the plot, and it was allowed to grow up to cover the plot. The long period between planting and the beginning of evaluations was necessary to guarantee a suitable plot establishment. The plants were fertilized with 40 g plant −1 with a mix of nitrogen (N), phosphorous (P2O5) and potassium (K2O) 15-0-15 and micronutrients. An additional 75 kg ha −1 of nitrogen (N) and 45 kg ha −1 of K2O were applied two months after planting. In early spring 2015, 90 kg ha −1 of N and K2O were applied to promote spring regrowth. The field was re-fertilized with 90 kg ha −1 of N and 45 kg ha −1 of K2O after each harvest, except for the last harvests of the growing seasons.

Experimental Design and Data Collection
The trial was established as a row-column design with two replicates and augmented representation of three controls, and the cultivars Tifton 85 and Jiggs were replicated thirteen times, and Coastal twelve times, in total. The plots were mowed to a stubble height of 10 cm in the beginning of each year (24 March 2015, and3 March 2016). Plots were harvested every five weeks from April to November in both years. Biomass was collected to a 5-cm stubble height from a 1.2 × 3.0 m area in each plot. The remaining areas non-harvested were mowed to the same stubble after data collection. The fresh samples were weighed, and sub-samples (approximately 450 g) were taken, dried in a forced-air oven at 55 °C for 72 h, and reweighed to estimate forage harvested (FH). The samples were The soil was a Chipley sand (thermic, coated Aquic Quartzipsamments) with a pH of 6.9 and characterized by high P 2 O 5 content (164 kg ha −1 ), and low K 2 O (38.2 kg ha −1 ), S (5.6 kg ha −1 ), and Mg (44.8 kg ha −1 ) content. The experimental plot size was 1.8 m × 3.0 m. Planting was done on 2 July 2014 using a single plug 5 × 5 cm planted in the center of the plot, and it was allowed to grow up to cover the plot. The long period between planting and the beginning of evaluations was necessary to guarantee a suitable plot establishment. The plants were fertilized with 40 g plant −1 with a mix of nitrogen (N), phosphorous (P 2 O 5 ) and potassium (K 2 O) 15-0-15 and micronutrients. An additional 75 kg ha −1 of nitrogen (N) and 45 kg ha −1 of K 2 O were applied two months after planting. In early spring 2015, 90 kg ha −1 of N and K 2 O were applied to promote spring regrowth. The field was re-fertilized with 90 kg ha −1 of N and 45 kg ha −1 of K 2 O after each harvest, except for the last harvests of the growing seasons.

Experimental Design and Data Collection
The trial was established as a row-column design with two replicates and augmented representation of three controls, and the cultivars Tifton 85 and Jiggs were replicated thirteen times, and Coastal twelve times, in total. The plots were mowed to a stubble height of 10 cm in the beginning of each year (24 March 2015, and3 March 2016). Plots were harvested every five weeks from April to November in both years. Biomass was collected to a 5-cm stubble height from a 1.2 × 3.0 m area in each plot. The remaining areas non-harvested were mowed to the same stubble after data collection. The fresh samples were weighed, and sub-samples (approximately 450 g) were taken, dried in a forced-air oven at 55 • C for 72 h, and reweighed to estimate forage harvested (FH). The samples were cleaned using sieves to avoid sand contamination and ground to 1 mm using a Wiley Mill (Model 4, Thomas Scientific, Swedesboro, NJ, USA) for NV analysis.
The analyses were performed using two subsets: (i) Whole-population: all accessions were harvested twice, first on June 1st and then in 11 August 2015; and (ii) Selected-population: a selected group of 15 accessions was sampled nine extra times between 1 June 2015 to 1 November 2016. The selected group resulted from a multi-location evaluation for biomass yield and other agronomic traits. The FH predictor values only were used to elucidate genetic correlations with NV traits. The NV traits included concentration of crude protein (CP), phosphorous (P), in vitro digestible organic matter (IVDOM) and neutral detergent fiber (NDF). These traits were determined by wet chemistry at the Forage Evaluation Support Laboratory, University of Florida (FESL). Crude protein was calculated as nitrogen multiplied by 6.25. Nitrogen and P samples were digested using a modification of the aluminum block digestion procedure [37]. Sample weight was 0.25 g, catalyst used was 1.5 g of 9:1 K 2 SO 4 :CuSO 4 , and digestion was conducted for at least 4 h at 375 • C using 6 mL of H 2 SO 4 and 2 mL H 2 O 2 . Nitrogen in the digestate was determined by semiautomated colorimetry [38]. In vitro digestible organic matter was performed by a modification of the two-stage technique [39]. Neutral detergent fiber was determined using the filter bag technique according to ANKOM 200 (ANKOM Technology, Macedon, NY, USA) procedure [40].

Statistical Analyses
A dataset for the whole-population recorded traits of the two harvests were analyzed using linear mixed models with repeated measures implemented in ASReml-R (VSNI, Hemel Hempstead, Hertfordshire, UK) [41] using the R software [42]. The following model was fitted for single-trait and multi-harvest data: where µ is the overall population mean; 1 is a vector of ones; X 1 , X 2 , Z 1 , Z 2 , Z 3 , and Z 4 are design matrices; α is the fixed harvest effect; β is the fixed effect of blocks; g is the random effects of entries, with g~multivariate normal distribution (MVN) (0,σ 2 g I); gα is the random interaction effect between entry and harvest, with gα~MVN (0,σ 2 gα I); r β and c β are the random effects of row and column nested into block, with r β~M VN (0,σ 2 rβ I) and c β~M VN (0,σ 2 cβ I); and e is the random errors, with e~MVN (0,σ 2 e I). For each trait, the genotypic values for the entries were predicted and the variance components estimated: genotypic variance (s 2 g ); variance of the genotype-harvest interaction (s 2 gh ). The statistical significance of variance components were tested using the likelihood ratio test (LRT) with a Chi-square test with 1 degree of freedom [43]. Then, variance components estimates were used to calculate broad sense heritability (H 2 ) for each trait as and genotype by harvest correlation (r gh ) was calculated for each trait as where s 2 g is the estimated genotypic variance, s 2 p is the total phenotypic variance, and s 2 gh is the variance for the genotype-harvest interaction. The random error variance s 2 e for the multiharvest model was an average for both harvests (i.e., s 2 e = (s 2 e2 + s 2 e5 )/2).
The genetic coefficient of variation (CV g ) was assessed for each trait using the following equation: where s 2 g is the estimated genotypic variance, and X is the mean value of the trait [10]. Accuracy refers to the correlation between the parametric genetic value and predicted genetic value, and it considers the residual variation, the settled experimental design and the proportion between the genetic and residual variations associated with the trait under evaluation [44]. Accuracy (Acc) was estimated as an average based on the standard error of predicted genotypic value (PVse) of each genotype [44]. PVse is related to accuracy through the equation: where s 2 g is the genotypic variance. Reliability (Rel) was obtained as average of genotypes Acc elevated to square. Overall mean, maximum, and minimum of predicted genotypic value were also computed. Additionally, predicted genetic values obtained with single-trait models were used to rank populations for each trait, which are given in Supplementary Table S1. A Principal Component Analyses (PCA) was performed with the prcomp function in R using a correlation matrix of the genotypic values obtained with the multiharvest model and predicted values of FH. These values were also used to obtain genetic correlations by Pearson method using cor function in R.
For the subset of selected population composed by 15 genotypes evaluated across 11 harvests, a linear model was fit for analysis of variance (ANOVA) using the package agricolae [45], with genotype, harvest, genotype-harvest interaction, and replicate factors considered fixed effects. A Tukey-Honest Significant Test was used to separate treatment means at p = 0.05 using the package agricolae [45], and graphs were created with the package ggplot2 [46] in R. The coefficient of variation was calculated as standard deviation divided by mean multiplied by 100 and expressed as percentage. Additionally, a performance index was calculated by counting the number of times each genotype placed in the top of the ranking for each trait, and their mean response was statistically superior compared to any other genotype.

Whole-Population
The whole Cynodon population showed significant genetic variability for all NV traits. Genetic variances for each trait were higher than zero (p < 0.001) based on LRT (Table 1), while GHI was only significant for IVDOM (p < 0.001). Genotype by harvest correlations were high for all traits (Table 1), and the H 2 ranged from low (CP and IVDOM) to moderate (P and NDF). The genetic coefficient of variation was low and exhibited a range from 2.6 to 9.6% (Table 1). Accuracy were moderate for CP, and high for other traits, whereas reliability was low for CP and moderate for the other traits.
Some traits exhibited large genotypic variability ( Figure 2). For instance, CP ranged from 116 g kg −1 to 157 g kg −1 (Figure 2A), where PI 2922601, PI 292508, PI 547109, and breeding line 319 exhibited CP higher than 150 g kg −1 . The highest predicted value for p was almost double its lowest concentration. However, only four accessions, including PI 316507, Breeding line 240, 'Florakirk', and PI 364485 showed p higher than 3.5 g kg −1 , while Tifton 85 and Jiggs presented 2.9 and 3.4 g kg −1 , respectively, and all above the population mean ( Figure 2B). Similarly, accessions exhibited wide IVDOM values from 363 to 563 g kg −1 ( Figure 2C). 'Tifton 84' presented the highest IVDOM followed by Florakirk (555 g kg −1 ), PI 316507 (549 g kg −1 ), PI 255450 (548 g kg −1 ), PI 204438 (546 g kg −1 ), and Tifton 85 (546 g kg −1 ). The NDF ranged from 651 g kg −1 to 767 g kg −1 ( Figure 2D), where only PI 297827, PI 287156, 319, and Florakirk showed values below 660 g kg −1 . Despite the fact that NDF presented the lowest genotypic range and lowest genetic coefficient of variation (low genetic variability), NDF had medium broad sense heritability and high genotype by harvest correlation. Comparing the population mean with Tifton 85, CP, P, and NDF values were similar, but Tifton 85 exhibited a much higher IVDOM compared to the population mean and most other accessions ( Figure 2). Table 1. Genetic parameters for the United States Department of Agriculture National Plant Germplasm System (USDA-NPGS) bermudagrass germplasm collection evaluated in Citra, FL. Genetic coefficient of variation (CV g ), broad sense heritability (H 2 ± SE), and genotype by harvest correlation (r gh ) for crude protein (CP), phosphorous (P), In Vitro Organic Matter Digestibility (IVDOM), and Nutrient Detergent Fiber (NDF) expressed at grams per kilogram. Likelihood Ratio Test of Genotype and Genotype by Harvest, Reliability Mean, and Accuracy Mean for nutritive value (NV) traits. concentration. However, only four accessions, including PI 316507, Breeding line 240, 'Florakirk', and PI 364485 showed p higher than 3.5 g kg −1 , while Tifton 85 and Jiggs presented 2.9 and 3.4 g kg −1 , respectively, and all above the population mean ( Figure 2B). Similarly, accessions exhibited wide IVDOM values from 363 to 563 g kg −1 ( Figure 2C). 'Tifton 84' presented the highest IVDOM followed by Florakirk (555 g kg −1 ), PI 316507 (549 g kg −1 ), PI 255450 (548 g kg −1 ), PI 204438 (546 g kg −1 ), and Tifton 85 (546 g kg −1 ). The NDF ranged from 651 g kg −1 to 767 g kg −1 (Figure 2D), where only PI 297827, PI 287156, 319, and Florakirk showed values below 660 g kg −1 . Despite the fact that NDF presented the lowest genotypic range and lowest genetic coefficient of variation (low genetic variability), NDF had medium broad sense heritability and high genotype by harvest correlation. Comparing the population mean with Tifton 85, CP, P, and NDF values were similar, but Tifton 85 exhibited a much higher IVDOM compared to the population mean and most other accessions ( Figure 2). The Pearson genetic correlations among the NV traits were all significant (p < 0.05), although FH was only correlated with IVDOM ( Table 2). The NDF presented moderate negative correlations with CP, P, and IVDOM (Table 2). In contrast, positive correlations were observed among CP, P, and IVDOM, as well as between IVDOM and FH, in a moderate to low magnitudes. The Pearson genetic correlations among the NV traits were all significant (p < 0.05), although FH was only correlated with IVDOM ( Table 2). The NDF presented moderate negative correlations with CP, P, and IVDOM (Table 2). In contrast, positive correlations were observed among CP, P, and IVDOM, as well as between IVDOM and FH, in a moderate to low magnitudes. The PCA using predictive values for all accessions explained a large part of the variation found for NV traits and FH in this collection ( Figure 3A). The first two PCAs accounted for 77.8% of the existing variation in the bermudagrass germplasm collection.  Commercial cultivars (red) and the selected accessions (blue) also exhibited broad variation ( Figure 3B). Two cultivars, T292 and Florakirk, exhibited contrasting performance for NV and FH ( Figure 3B). Besides, most commercial cultivars showed lower NDF and higher values for CP, P and IVDOM in general. Tifton 85 exhibited the highest yield and IVDOM across cultivars in those two harvests. Comparing accessions and cultivars, some selected accessions exhibited higher combining Commercial cultivars (red) and the selected accessions (blue) also exhibited broad variation ( Figure 3B). Two cultivars, T292 and Florakirk, exhibited contrasting performance for NV and FH ( Figure 3B). Besides, most commercial cultivars showed lower NDF and higher values for CP, P and IVDOM in general. Tifton 85 exhibited the highest yield and IVDOM across cultivars in those two harvests. Comparing accessions and cultivars, some selected accessions exhibited higher combining of FH, IVDOM, CP, and P than commercial varieties, except by Tifton 85 and Florakirk. Tifton 85 presented high FH, while Florakirk presented high P and IVDOM. The accessions PI 316510 (322) and PI 255450 (242) grouped near Tifton 85, exhibited similar FH but higher IVDOM. In addition, some accessions exhibited high CP and P, such as PI 316507 (323), Breeding line 240 (240), PI 316536 (325), and PI 364484 (117).

Selected-Population
Significant genotype and harvest main effects (p < 0.001) were observed for all traits, but two-way interactions were non-significant for all traits (Table 3). For CP and IVDOM, the mean square values for harvest were considerably larger than for genotype, suggesting a higher environmental influence for these traits without influencing the genotypic response. However, the mean square value of harvest for P was lower than for genotype, which means a greater influence of genotype on P. The CV was higher for CP and P than for IVDOM and NDF but lower than 30%, which is the maximum threshold for field experiments [47]. Additionally, NDF exhibited CV lower than 5%, which was similar to the CV g observed in the whole population.  The NV traits varied greatly across harvests ( Table 5). The NV fluctuated within the year, with higher CP and lower NDF in the April and September harvests. Higher IVDOM values were measured in the spring and beginning of summer for both years, coinciding with high volumes of rainfall and mild temperature (Figure 1). Lower values for CP, IVDOM, NDF, and P were observed in November, with scarce rainfall and low temperature inducing dormancy. The NV measured on the same month on both years were similar, and for some traits, there were no significant differences (Table 5). Although the fluctuations in temperature observed during both years were similar, the rainfall regime was different (Figure 1). In 2015, July and August had 250 mm per month, which was above average, whereas the rainfall the same period in 2016 was less than 133 mm per month. The rainfall in June was 290 mm in 2015, whereas it was 43 mm in 2016.

Discussion
The genetic parameters indicated that genetic gains can be expected in all NV traits in this bermudagrass germplasm, as most traits presented moderate H 2 . Broad sense heritability explains the magnitude of phenotypic variance due to the genetic variance [36]. Indeed, a higher genetic variance than GHI was observed for all traits, and the GHI was only significant for IVDOM. GHI effect was considerably lower than the main genotypic effect, similar to previous reports in maize (Zea mays L.) and timothy grass (Phleum pratense L.) [48,49], if we treat harvest as an environment. Thus, high stability of genotypes for these traits across harvests would be expected as supported by the high estimates of r GH . The use of r GH , although it does not elucidate the significance of variance sources, indicates the stability of genotypes among harvests in a given location, meaning that traits with high stability can be subjected to selection with fewer harvests. Given the high r GH for all traits, using estimates from fewer harvests can be a reliable NV measure to screen large nurseries in forage breeding programs. However, considering a selection for multiple locations, other environmental variables, such as soil type, crop management, and climate variability, might diverge and cause genotype by environment interaction (GEI). Therefore, we encourage other studies to elucidate GEI for NV traits. Despite that, using only two harvests for the whole population, it was possible to achieve prediction accuracies higher than 0.70 for IVDOM, NDF, and P, considered appropriate for forage breeding experiments [50]. In general, an increase in the number of harvests leads to an increase in precision; however, it adds costs and labor when dealing with large populations [51].
Genetic parameters, such as H 2 , r GH , and CV g , are variable among species, germplasm, traits, and experiments [52]. Medium to high broad-sense heritability for IVDOM have been reported for other forage grasses, 0.56 to 0.93 [53][54][55]. Previous efforts with bermudagrass showed broad-sense heritability ranging between 0.27 and 0.78 across years and experiments [11], denoting that H 2 is inherent to the experimental conditions, as well as in where moderate to high estimates can be achieved for those species. In our study, H 2 estimates were within the range of previous reports. The magnitude of H 2 for P in the bermudagrass collection was similar to the broad-sense heritability of 0.46 reported for reed canarygrass (Phalaris arundinacea L.) [56] but lower than the 0.72 estimate reported in tall fescue (Schedonorus arundinacus (Schreb.) Dumort.) [57].
In turn, NDF had lower H 2 compared to previous studies, and lower variation (CV g ) in the population. Broad-sense heritability in NDF for populations of tall fescue [54] and koronivia grass (Urhochloa humidicola (Rendle), Morrone and Zuloaga [58]) were much greater (0.85 and 0.58, respectively) compared to the bermudagrass germplasm. Other reports also showed high narrow-sense heritability for NDF in tall fescue (0.6; Reference [59]) and signal grass (Urochloa decumbens (Stapf) R. Webster) (0.74; Reference [51]). Moreover, r GH across multiple harvests was also lower for the referred studies (0.36 in koronivia grass [58]; 0.82 in signal grass [51]). Thus, the efforts to genetically improve NDF from this germplasm would be similar to the above-mentioned species but fewer evaluations are needed since it has higher stability.
Crude protein exhibited lower H 2 than the estimates for the other NV traits, but similar H 2 estimates were found in other species. Low broad-sense heritability was reported for CP in meadow fescue (0.21) [53], as well as low narrow-sense heritability (0.07) in koronivia grass [58], 0.14 in congo grass [60], and 0.18 in tall fescue [59]. The estimates of H 2 , also, can be influenced by environment. In tall fescue, narrow heritability ranged from 0.18 to 0.54, with the winter harvest exhibiting a higher estimate than that for the fall harvest [59]. Thus, the choice of the harvest to characterize CP is important because managing data collection can minimize undesirable effects attributed to the large environmental variance compared to measured genetic variance in this study. Although CP had low H 2 that will potentially result in lower genetic gains for this trait, any gain in protein content in grasses can result in significant reductions of the use of external protein sources. The high stability of this trait over the two annual harvests (high r GH ), and non-significant GHI indicated that most of the variation in CP in this population was indeed due to genetics.
The genetic coefficient of variation complements the broad sense heritability estimates, and it allows the comparison of genetic variability among populations and traits [10]. The CV g for IVDOM was 8.1%, which is higher than values reported for meadow fescue (4%) [53], congo grass (2.2%) [61] and Urochloa spp. (5.8%) [55]. Digestibility has been a major target in previous bermudagrass breeding efforts (e.g., Coastcross I; Reference [11]. In an effort to improve digestibility in bermudagrass, approximately 500 accessions showed great variability for in vitro dry matter digestibility (IVDMD) and CV g ranged from 6.1 to 28.9%, and the two-year average for IVDMD ranged from 400 to 690 g kg −1 [10]. This large genetic variation can be exploited in breeding programs. In our study, the PI 255450 (242), which is a parental line for Kenya 61 [11], exhibited the highest IVDOM. Although it could be used in crosses aiming to combine higher digestibility with other traits, the ploidy level of this accession can be a limitation because the progeny from Coastal and Kenya 61 were all male sterile and not able to produce seeds [11]. Other Pis, such as Tifton 84, Florakirk, and PI 316507, also showed high IVDOM.
Phosphorous had the largest CV g among the NV traits in this population. Breeding line 240 and PI 316510 (322) had an average 30% more P concentration than Tifton 85, as well as desired levels of CP, IVDOM, and NDF. This genetic variability can be explored to generate information about P uptake and its relationship with forage yield, as previous research reported for N [18]. Accessions with high uptake P and N that combine reasonable agronomic performance can be used as phytoremediation agents for dairy farms [27,62]. On the other hand, genetic variability for NDF was the lowest among evaluated traits, concurrent with other reports from the literature, like 1.4 and 2.9% in tall fescue [59], 3.8 to 4.2% in leaves of Brachiaria spp. [55] 4.32, and 4.43% in leaves and stems of Arachis spp. [59]. These aspects confirm that difficulty in improving NDF by conventional breeding in forage species.
Entries with high NV traits were already included in the bermudagrass core collection except for four accessions (PI 292508, PI 547109, PI 297827, and PI 287156-01), as selection was based on FH across several environments [15]. Breeding line 240, PI 316510 (322) and PI 3166536 (325) presented lower NDF and greater P concentration than Tifton 85, and similar IVDOM and CP. Breeding for improved CP has received little attention in the past as nitrogen fertilizer has been an affordable and effective way to increase increasing CP concentration, IVDOM, and productivity in grasses [17,18,23]. Few studies looked at nitrogen use efficiency in bermudagrass [18], which found high variability for the trait, and a negative correlation to CP concentration especially at low N fertilization rates.
In general for bermudagrass, CP has negative correlation with biomass and NDF (−0.32 and −0.36, Reference [29]), but positive correlation with dry matter digestibility (0.34; [11]). Similar results were observed on a review of over one hundred forage species, where digestibility had a positive correlation with CP (0.62) and negative with NDF (−0.68) [63]. Our results showed similar trends, although FH was only significantly correlated to IVDOM. The moderate to low genetic correlations found in the current germplasm suggest genes related to the expression of these traits have higher independence or belong in different metabolic routes. This can be due to the genetic diversity present in this germplasm, particularly related to having different species and ploidy levels in the core collection [15,16]. Thus, breeding multiple nutritive value traits in Cynodon ssp. should be achievable, and there is still potential for NV to be improved with forage yield.
Selection on a single trait provides higher genetic progress but can be detrimental if some unwanted correlation exists with the trait under improvement. On the other hand, genetic correlations can be used to identify potential indirect selections for simultaneous breeding [36]. Although there was no clear cluster of genotypes in the PCA, it was possible to identify genotypes that presented superior NV traits and FH. For instance, except for Tifton 85 and Florakirk, other accessions in the selected population presented higher P, IVDOM, and FH than most commercial cultivars. In general, commercial varieties and the selected population showed variability and different patterns of NV performance. Accessions such the breeding line 240, PI 255450, and PI 316510 exhibited high FH and NV. Thus, they can be selected as parental lines in future crosses and public cultivar releases.

Conclusions
The results presented in this study complement previous findings and provide useful information for the entire forage bermudagrass collection, aiming at developing cultivars with improved NV. The high significance of the genotypic factor evaluated during eleven harvests showed differences among the accessions in the selected-population for all NV traits. These accessions have a good combination of NV and biomass production, and some of them had improved NV compared to Tifton 85. The lack of significant two-way interactions between genotype and harvest for NV traits confirmed the high stability of genotypes across harvests. Thus, selection for NV can be assessed with fewer harvests, resulting in savings for time, labor, and resources. Genetic parameters revealed that P has higher potential to be explored as breeding target, along with CP, while narrow variance for NDF, and the availability of varieties with high IVDOM, would require more effort than conventional breeding for improving those traits in bermudagrass. Breeding line 240, PI 316510 (322), and PI 3166536 (325) presented superior nutritive value than Tifton 85 and will be considered for public cultivar releases in the United States.