Unraveling phenotypic diversity in Cynodon spp. germplasm for forage accumulation and nutritive value in the transition zone

Abstract

Characterizing phenotypic variation in germplasm collections is crucial for plant breeding. Cynodon spp. accessions maintained at the USDA National Plant Germplasm System (NPGS), and the core forage collection were screened in a replicated trial under two nitrogen rates (0 and 150 kg ha⁻¹ per harvest) in Ardmore, OK. The goals for this study were to: (a) estimate genetic parameters for forage accumulation (FA) and nutritive value (NV), (b) estimate genotype × harvest (r_gh) and genotype × environment (r_ge) interaction for all traits, (c) quantify FA and NV for commercial cultivars and selected accessions. The experiment was setup as a row-column design with two replicates and augmented representation of controls: Tifton 85, Wrangler, Midland, and Cheyenne. The trial was harvested five times (twice in 2016 and three times in 2017) and data were analyzed using linear mixed models. Genetic parameters revealed the presence of significant phenotypic variation for FA and NV. Low genotype × harvest and genotype × environment interactions indicated genotypic stability and potential for selecting improved accessions for FA and NV. Several plant introductions (PIs) were identified for their improved FA and NV. The application of N increased FA and CP concentration in all harvests but did not significantly impact phenotypic variation in other NV traits. Phenotypic data will be deposited in the Germplasm Resource Information Network (GRIN)-Global database.

Abbreviations

ADF

acid detergent fiber

Ash

ash concentration

CP

crude protein

dNDF48

NDF digestion for 48 h

FA

forage accumulation

IVTDMD

in vitro true dry matter digestibility

Lignin

lignin concentration

NDF

neutral detergent fiber

NPGS

National Plant Germplasm System

NV

nutritive value

PCA

principal component analysis

PIs

plant introductions

TDN

total digestible nutrients

1 INTRODUCTION

The genus Cynodon L. C. Rich. is characterized by morphologically well-defined species that often cross and produce viable offspring under experimental conditions (de Wet & Harlan, 1970). The species C. dactylon (L.) Pers., commonly known as bermudagrass, is one of the most genetically variable and economically important species for warm-season forage and turfgrass breeding. The species C. nlemfuensis, known as stargrass, is widely used as forage due to its large stems, long stolons, and it spreads rapidly when planted vegetatively. Stargrass can readily cross with bermudagrass (de Wet & Harlan, 1970), and several interspecific hybrids between both species have been selected and released as commercial cultivars (Burton & Monson, 1972; Burton et al., 1967, 1984, 1993). All Cynodon species have a basic chromosome number of nine (x = 9) and ploidy levels range from diploid (2n = 2x = 18) to hexaploid (2n = 6x = 54; Anderson et al., 2009; Grossman et al., 2021; Taliaferro et al., 1997).

A Cynodon germplasm collection is maintained at the USDA-NPGS, and accessions can be requested online through the USDA Germplasm Resources Information Network (USDA-GRIN) (https://npgsweb.ars-grin.gov/gringlobal/search/). These plant introductions (PIs) were collected around the world and include various Cynodon species; however, passport information and taxonomy are incomplete or unknown for most PIs. Due to the complexity of multiple species, ploidy levels, and interspecific hybrids present in the Cynodon germplasm, the term bermudagrass will be used to refer to all accessions tested in this study. A bermudagrass forage core collection was assembled using phenotype data for 22 traits, including forage accumulation (FA), nutritive value (NV), and cold tolerance (Anderson, 2005). The core collection is composed of PIs from around the world (54% from Africa), as well as interspecific hybrids (26%) from the breeding program led by Dr. Glenn Burton at the USDA-ARS in Tifton, GA and commercial cultivars (12%; Anderson, 2005; Anderson et al., 2009; Supplemental Table S1). The core collection and several PIs obtained from USDA-GRIN were evaluated in Florida and showed significant phenotypic variation for NV traits, such as crude protein (CP), phosphorous concentration [P], in vitro digestible organic matter (IVDOM) and neutral detergent fiber (NDF; Souza et al., 2020), and for FA and bermudagrass stem maggot (BSM; Atherigona reversura Villeneuve) tolerance (Grossman et al., 2021). The core collection was developed with phenotypic data collected in the U.S. state of Georgia, and Souza et al. (2020) and Grossman et al. (2021) collected data in Florida. Given the increasing interest in growing bermudagrass in the transition zone in the United States (Baxter et al., 2022b), there is a need to quantify the adaptation and phenotypic variability of this germplasm in a broader range of environmental and management conditions within the transition zone.

Breeding methods applied in forage bermudagrass have relied on phenotypic selection to develop seeded varieties (Taliaferro, 2003) and vegetatively propagated polyploid interspecific hybrids (Burton & Monson, 1972; Burton et al., 1967, 1984, 1993). Two of the most planted forage bermudagrass cultivars are ‘Coastal’ and ‘Tifton 85’. Coastal is a hybrid between ‘Tift’ bermudagrass and an accession from South Africa, and its release in 1943 was the result of an extensive breeding program led by Dr. Glenn Burton, USDA-ARS, Georgia Coastal Plains Experiment Station at Tifton, GA (Burton, 1948). Tifton 85 is an interspecific hybrid between ‘Tifton 68’′, an hexaploid stargrass (2n = 6x = 54), and the plant introduction (PI) 290884, a tetraploid C. dactylon (Burton et al., 1993). Tifton 85 is considered one of the best bermudagrass cultivars released to date (Baseggio et al., 2015; Hanna & Anderson, 2008). Other cultivars are commercially available; however, most of them were selected and released decades ago, except for the ecotypes ‘Mislevy’ (Vendramini et al., 2021), and ‘Newell’ (Rios et al., 2021).

Core Ideas

• Cynodon germplasm collections preserve sources of phenotypic and genetic diversity that can be used for breeding.
• Phenotypic variation exists for FA and NV in Cynodon germplasm.
• Tifton 85 had the greatest FA and NV among commercial cultivars.
• Some PIs and breeding lines had greater FA and NV than cultivars.

Perennial pastures are the main feed source for livestock, and they also provide various ecosystem services, including carbon sequestration and nutrient cycling (Gerber et al., 2013; O'Mara, 2012; Teague et al., 2019). Bermudagrass is well adapted to various climate and soil types, and provides high production potential, drought tolerance, and grazing tolerance (Redfearn et al., 2016; Redfearn & Wu, 2013). Cow–calf type enterprises in the transition zone in the United States use bermudagrass primarily as a grazed and preserved forage (Redfearn & Wu, 2013). Additionally, heat and drought tolerance in summer, and sufficient winter hardiness of bermudagrass resulted in improved adaptation of bermudagrass in Oklahoma (Martin, 2007). Livestock performance is dependent on diet, among other factors, and greater nutrient intake results in greater production (Oba & Allen, 1999). Nutritive value indicates the nutrient content in feed, and it is an important factor to livestock diets (Board on Agriculture and Natural Resources et al., 2001). Improving NV in forage species could be achieved by multiple approaches, such as breeding (Casler, 2001) and fertilizer applications (Schneider-Canny et al., 2019). Breeding and selection in bermudagrass resulted in improved varieties exhibiting greater FA and NV (Burton & Monson, 1972; Burton et al., 1967, 1984, 1993). Nitrogen (N) is a common limiting nutrient in forage production and environments with low or high N availability might affect potential performance of cultivars (Schneider-Canny et al., 2019). Nevertheless, nitrogen fertilization often results in greater FA and NV increases in bermudagrass (Hill et al., 2001; Schneider-Canny et al., 2019; Woli et al., 2019).

Cynodon spp. germplasm collection preserves unique sources of diversity that can be exploited in forage breeding. The estimation of genetic parameters for FA and NV, such as repeatability (R), genotype × environment interaction, and trait correlations can guide breeding approaches in bermudagrass. The main goal of this study was phenotype a Cynodon spp. collection for FA and NV in Ardmore, OK grown under two N rates, calculate genetic parameters, and select accessions that could be explored for public cultivar releases. The specific objectives were to: (a) estimate genetic parameters for FA and NV traits across five harvests and two N environments, (b) obtain predicted values for all traits and make guided selections for accessions exhibiting improved traits, and (c) quantify the response to N fertilization for FA and NV in this germplasm.

2 MATERIALS AND METHODS

2.1 Germplasm

A set of 290 accessions of Cynodon spp. with different ploidy levels (Grossman et al., 2021) and originating from 34 countries were evaluated in this study (Supplemental Table S1). The germplasm included 148 Cynodon clonal accessions from a bermudagrass forage collection maintained at Tifton, GA (William Anderson, personal communication, 2022), and 137 accessions from the USDA-NPGS Cynodon collection maintained at Griffin, GA. The cultivar Mislevy was also included in the study and planting material was obtained from Florida (Vendramini et al., 2021). Several other cultivars were included: African Star (23-20), African Star (24-17), African Star (339), African Star (340), African Star (8-13), Alicia, Burton 200 (Alicia × Tifton 78), Callie, Cheyenne, Coastal, Coastcross I, Coastcross II, Coastcross III, Coffee, D-8 (329), Florida 44, Florakirk, Jiggs, Midland, Russel, T292, Tifton, Tifton 44, Tifton 68, Tifton 78, Tifton 84, Tifton 85, Winter hardy (319), Winter hardy (320), and Wrangler. Three cultivars were considered as controls due to their popularity in the transition zone: Wrangler, Cheyenne, and Midland; while Tifton 85 was also included as control due to its performance in previous studies (de Grossman et al., 2021; Souza et al., 2020).

2.2 Location description

The experiment was established in July 2015 at the Noble Research Institute in Ardmore, OK (34° 11′ 33″ N and 97° 53′ 8″ W), at 266 masl. The soil at the location is a Wilson silt loam (Fine, smectitic, thermic Oxyaquic Vertic Haplustalfs). Historical weather data was obtained from the Oklahoma Mesonet weather data (https://www.mesonet.org/) and summarized in Figure 1.

FIGURE 1

Maximum (Max Temp.), minimum (Min Temp.) and mean (Mean Temp.) monthly temperature and rainfall during the experimental period (May 2016 to August 2017). The arrows show the sampling dates

2.3 Experimental design and data collection

The trial was established as a row-column design with two replicates, and a split-plot treatment arrangement with two N rates (whole plots) and genotypes (subplots). The N rates consisted of nonfertilized and N-fertilized treatments, and the N-fertilized plots received N after each harvest with split application of 150 kg N ha⁻¹ in June and August 2016, and again 150 kg N ha⁻¹ in April, June, and July 2017. Harvests were performed on 7 July and 9 Sept. 2016, and 22 May, 18 July, and 21 Aug. 2017. Forage accumulation was collected to a 10-cm stubble height from a 1.5-by-1.5-m² area in each plot, and the remaining plot area was mowed to the same stubble after data collection. The experimental plot size was 1.8 by 3.0 m. Fresh biomass per plot was weighed before subsamples (approximately 500 g) were taken, weighed, dried in a forced-air oven at 60 °C for 72 h, and weighed again to estimate FA in kg ha⁻¹. Samples were ground to 1 mm using a Wiley Mill (Model 4, Thomas Scientific), and stored at room temperature until further analysis. Samples were dried at 95% dry matter concentration, then were analyzed using near-infrared reflectance spectrophotometer (NIRS), with the FOSS NIRS DS2500 equipped with the NIRS Consortium Equations for grass hay (https://www.nirsconsortium.org). The NV traits included CP, calcium concentration [Ca], [P], potassium concentration [K], magnesium concentration [Mg], acid detergent fiber (ADF), NDF, total digestible nutrients (TDN), lignin concentration (Lignin), in vitro true dry matter digestibility (IVTDMD), ash concentration (Ash), and NDF digestion for 48 hours (dNDF48). All these traits were expressed as g kg⁻¹.

2.4 Statistical analyses

The statistical model used to fit the phenotypic data across repeated measurements (harvests) and the two levels of N (environment) can be described as:

$$$$\begin{equation*}{\bf{y}} = X{{\beta}} + {Z_1}gh + {Z_3}ge + {Z_4}ghe + {Z_5}r + {Z_6}c + {\bf{\varepsilon }}\end{equation*}$$$$

where y is the vector of phenotypic observations; β is the vector of fixed effects of the intercept, environments, harvests, replicates and the interaction between environments and harvests; g, gh, ge, and ghe are the random effects of the genotypes, genotypes × harvests interaction, genotypes × environments interaction, and the triple interaction between genotypes, harvests and environments; r and c are the random effects of rows and columns within replications; and finally ε is the vector of the residuals. The random effects can be described as:

$$$$\begin{equation*}\left[ \def\eqcellsep{&}\begin{array}{l} g\\ gh\\ ge\\ ghe\\ r\\ c\\ {{\rm{\varepsilon }}_1}\\ {{\rm{\varepsilon }}_2} \end{array} \right] = \left( {\left[ \def\eqcellsep{&}\begin{array}{l} 0\\ 0\\ 0\\ 0\\ 0\\ 0\\ 0\\ 0 \end{array} \right],\left[ { \def\eqcellsep{&}\begin{array}{@{}*{7}{c}@{}} {I{{\sigma}}_g^2}&{}&{}&{}&{}&{}&{}\\ {}&{I{{\sigma}}_{gh}^2}&{}&{}&{}&{}&{}\\ {}&{}&{I{{\sigma}}_{ge}^2}&{}&{}&{}&{}\\ {}&{}&{}&{I{{\sigma}}_{ghe}^2}&{}&{}&{}\\ {}&{}&{}&{}&{I{{\sigma}}_r^2}&{}&{}\\ {}&{}&{}&{}&{}&{I{{\sigma}}_c^2}&{}\\ {}&{}&{}&{}&{}&{}&{{R_1} \oplus {R_2}{{\sigma}}_{\rm{\varepsilon }}^2} \end{array} } \right]} \right)\end{equation*}$$$$

where $${{\sigma}}_g^2$$, $${{\sigma}}_{gh}^2$$, $${{\sigma}}_{ge}^2$$ and $${{\sigma}}_{ghe}^2$$ are the variance components associated with genotypes, genotype × harvest, genotype × environment, and genotype × harvest × environment effects, respectively; $${{\sigma}}_r^2$$and $${{\sigma}}_c^2$$ are the variance components associated with rows and columns within replications; $${{\sigma}}_{{{\rm{\varepsilon }}_1}}^2$$and $${{\sigma}}_{{{\rm{\varepsilon }}_2}}^2$$ are the variance components of whole plot and repeated measure errors effects, respectively; I is an identity matrix; R₁ and R₂ are the (co)variance matrix structures of the residuals associated with repeated measures considered as a heterogeneous autoregressive structure, described as:

$$$$\begin{equation*}{R_1} \oplus {R_2} = \left[ { \def\eqcellsep{&}\begin{array}{@{}*{10}{c}@{}} {{{\sigma}}_{{{\rm{\varepsilon }}_{11}}}^2}&{{{{\rho}}^1}}&{{{{\rho}}^2}}&{{{{\rho}}^3}}&{{{{\rho}}^4}}&{}&{}&{}&{}&{}\\ {{{{\rho}}^1}}&{{{\sigma}}_{{{\rm{\varepsilon }}_{12}}}^2}&{{{{\rho}}^1}}&{{{{\rho}}^2}}&{{{{\rho}}^3}}&{}&{}&{}&{}&{}\\ {{{{\rho}}^2}}&{{{{\rho}}^1}}&{{{\sigma}}_{{{\rm{\varepsilon }}_1}3}^2}&{{{{\rho}}^1}}&{{{{\rho}}^2}}&{}&{}&{}&{}&{}\\ {{{{\rho}}^3}}&{{{{\rho}}^2}}&{{{{\rho}}^1}}&{{{\sigma}}_{{{\rm{\varepsilon }}_{14}}}^2}&{{{{\rho}}^1}}&{}&{}&{}&{}&{}\\ {{{{\rho}}^4}}&{{{{\rho}}^3}}&{{{{\rho}}^2}}&{{{{\rho}}^1}}&{{{\sigma}}_{{{\rm{\varepsilon }}_{15}}}^2}&{}&{}&{}&{}&{}\\ {}&{}&{}&{}&{}&{{{\sigma}}_{{{\rm{\varepsilon }}_{21}}}^2}&{{{{\rho}}^1}}&{{{{\rho}}^2}}&{{{{\rho}}^3}}&{{{{\rho}}^4}}\\ {}&{}&{}&{}&{}&{{{{\rho}}^1}}&{{{\sigma}}_{{{\rm{\varepsilon }}_{22}}}^2}&{{{{\rho}}^1}}&{{{{\rho}}^2}}&{{{{\rho}}^3}}\\ {}&{}&{}&{}&{}&{{{{\rho}}^2}}&{{{{\rho}}^1}}&{{{\sigma}}_{{{\rm{\varepsilon }}_{23}}}^2}&{{{{\rho}}^1}}&{{{{\rho}}^2}}\\ {}&{}&{}&{}&{}&{{{{\rho}}^3}}&{{{{\rho}}^2}}&{{{{\rho}}^1}}&{{{\sigma}}_{{{\rm{\varepsilon }}_{24}}}^2}&{{{{\rho}}^1}}\\ {}&{}&{}&{}&{}&{{{{\rho}}^4}}&{{{{\rho}}^3}}&{{{{\rho}}^2}}&{{{{\rho}}^1}}&{{{\sigma}}_{{{\rm{\varepsilon }}_{25}}}^2} \end{array} } \right]\end{equation*}$$$$

where $${{\sigma}}_{{{\rm{\varepsilon }}_{eh}}}^2$$ is the variance component of the errors in harvest h and nitrogen rate e; ρ is the residual autocorrelation parameter between harvests.

The data were analyzed using linear mixed models with repeated measures implemented in ASReml-R (Butler et al., 2017) using the R software (R Development Core Team, 2020). A model was fitted for single-trait and multiharvest data, where harvest, blocks, and environment (N rate) were included as fixed effects, and genotype, row, and column, as well their interactions with other factors were considered as random effects. The statistical significance of variance components was tested using the likelihood ratio test with a Chi-square test with one degree of freedom (Satorra & Saris, 1985). Then, variance components estimates were used to calculate genetic repeatability (R) for each trait as:

$$$$\begin{equation*}R = \frac{{{{\sigma}}_g^2}}{{{{\sigma}}_p^2}} = \frac{{{{\sigma}}_g^2}}{{{{\sigma}}_g^2 + \frac{{{{\sigma}}_{gh}^2}}{h} + \frac{{{{\sigma}}_{ge}^2}}{e} + \frac{{{{\sigma}}_{ghe}^2}}{{he}} + \frac{{{{\sigma}}_{\rm{\varepsilon }}^2}}{b}}}\end{equation*}$$$$

and the genotype × harvest correlation (r_gh) was calculated for each trait as:

$$$$\begin{equation*}{r_{gh}} = \frac{{{{\sigma}}_g^2}}{{{{\sigma}}_g^2 + {{\sigma}}_{gh}^2}}\end{equation*}$$$$

and genotype × environment correlation (r_ge) was calculated for each trait as:

$$$$\begin{equation*}{r_{ge}} = \frac{{{{\sigma}}_g^2}}{{{{\sigma}}_g^2 + {{\sigma}}_{ge}^2}}\end{equation*}$$$$

where $${{\sigma}}_g^2$$ is the estimated genotypic variance, $${{\sigma}}_p^2$$ is the total phenotypic variance, $${{\sigma}}_{gh}^2$$ is the variance for the genotype × harvest interaction, $${{\sigma}}_{ge}^2$$is the variance for the genotype × environment interaction, and $${{\sigma}}_{ghe}^2$$ is the variance component for the genotype × harvest × environment interaction. The random error variance $${{\sigma}}_{\rm{\varepsilon }}^2$$ for the multiharvest–environment model was an average for all harvests–environment $${{\sigma}}_{\rm{\varepsilon }}^2 = \frac{{{{\sigma}}_{{{\rm{\varepsilon }}_{11}}}^2 + {{\sigma}}_{{{\rm{\varepsilon }}_{12}}}^2 + {{\sigma}}_{{{\rm{\varepsilon }}_{13}}}^2 + \cdots + {{\sigma}}_{{{\rm{\varepsilon }}_{he}}}^2}}{{he}}$$. h, e and b are the number of harvests, N rates and replicates, respectively.

The genetic and experimental coefficient of variation (CV_g and CV_e) was assessed for each trait using the following equation:

$$$$\begin{equation*}C{V_g} = \left( {\sqrt[2]{{{{\sigma}}_g^2}}\frac{1}{{\bar X}}} \right)100\end{equation*}$$$$

$$$$\begin{equation*}C{V_e} = \left( {\sqrt[2]{{{{\sigma}}_{\rm{\varepsilon }}^2}}\frac{1}{{\bar X}}} \right)100\end{equation*}$$$$

where $$\bar X$$ is the mean value of the trait.

Additionally, predicted genotypic values obtained with single-trait models were used to rank accessions for each trait, and are given in Supplemental Table S1. Principal component analyses (PCA) were performed with the prcomp function in R using a correlation matrix of the genotypic values obtained with the multiharvest model across both environments for FA and NV traits. The genetic correlation between traits were calculated based on variance of the sum between two traits, as described by the following equation:

$$$$\begin{equation*}{{\sigma}}_{\left( {g1 + g2} \right)}^2 = {{\sigma}}_{{\rm{g1}}}^2 + {{\sigma}}_{{\rm{g2}}}^2 + 2{\rm{CO}}{{\rm{V}}_{g1,g2}}\end{equation*}$$$$

$$$$\begin{equation*}{\rm{CO}}{{\rm{V}}_{g1,g2}} = \frac{{{{\sigma}}_{\left( {g1 + g2} \right)}^2 - {{\sigma}}_{\left( {g1 + g2} \right)}^2}}{2}\end{equation*}$$$$

$$$$\begin{equation*}{r_{g1,g2}} = \frac{{{\rm{CO}}{{\rm{V}}_{g1,g2}}}}{{\sqrt {{{\sigma}}_{g1}^2{{\sigma}}_{g2}^2} }}\end{equation*}$$$$

The graphs were created with the package ggplot2 (Wickham, 2016) in R 4.2.1 (R Core Team, 2022).

3 RESULTS

3.1 Genetic parameters for forage accumulation

Genetic variability for FA was significant in the bermudagrass germplasm tested in this study. Genetic variance, genotype × harvest interaction, and genotype × environment interaction were greater than zero (P < .001) based on likelihood ratio tests (Table 1). The R estimate for FA was the lowest among all traits, and genotype × harvest correlation (r_gh) and genotype × environment correlation (r_ge) estimates were high. The genetic coefficient of variation (CV_g), accuracy was high for FA (Table 1).

TABLE 1. Variance components for the genotypic ($${{\sigma}}_g^2$$), genotype × harvest ($${{\sigma}}_{gh}^2$$), genotype × environment (nitrogen rates; $${{\sigma}}_{ge}^2$$), genotype × environment × harvest ($${{\sigma}}_{ghe}^2$$), and the residual ($${{\sigma}}_{\rm{\varepsilon }}^2$$) components

Trait	$${{\sigma}}_g^2$$	$${{\sigma}}_{gh}^2$$	$${{\sigma}}_{ge}^2$$	$${{\sigma}}_{ghe}^2$$	$${{\sigma}}_{\rm{\varepsilon }}^2$$	R^a	rgh	rge	Mean	CVg	CVe
ADF	3.675***	0.753***	0.122**	0.000 ns^†	3.062	.8690	0.8299	0.9679	34.30	5.59	5.10
Ash	0.711***	0.125***	0.031 ns	0.000 ns	2.705	.7771	0.8505	0.9582	9.98	8.45	16.47
Ca	1.391***	0.334***	0.107*	0.000 ns	3.046	.8033	0.8064	0.9286	0.50	7.45	11.02
CP	2.299***	0.488***	0.059 ns	0.101*	3.055	.7767	0.8249	0.9750	11.79	12.86	14.82
FA	1.163***	0.443***	0.045 ns	0.005 ns	2.216	.7668	0.7242	0.9627	3581.60	30.11	41.55
dNDF48	3.674***	0.556***	0.228***	0.000 ns	2.272	.8937	0.8686	0.9416	29.36	6.53	5.13
IVTDMD	10.332***	1.643***	0.214 ns	0.000 ns	9.561	.8850	0.8628	0.9797	65.35	4.92	4.73
K	16.232***	5.829***	2.106***	0.000 ns	32.928	.7652	0.7358	0.8852	1.39	9.18	13.07
Lignin	0.241***	0.057***	0.013*	0.000 ns	0.531	.8070	0.8087	0.9488	6.56	7.48	11.10
Mg	1.967***	0.206***	0.043 ns	0.000 ns	1.805	.9069	0.9052	0.9786	0.39	11.25	10.77
NDF	10.186***	1.444***	0.309**	0.000 ns	6.202	.9030	0.8758	0.9706	65.42	4.88	3.80
P	0.186***	0.032***	0.009**	0.000 ns	0.315	.8472	0.8532	0.9538	0.21	6.52	8.47
TDN	4.345***	0.916***	0.144**	0.000 ns	3.553	.8674	0.8259	0.9679	60.94	3.42	3.09

• Note. Variances were used to calculate genetic parameters for the 290 bermudagrass accessions evaluated in Ardmore, OK, across five harvests and two environments.
• ^a R, repeatability; r_gh, genotype × environment correlation; r_ge, genotype × nitrogen rates; CV_g, coefficient of genetic variation; CV_e, coefficient of environmental variation; ADF, acid detergent fiber; Ash, ash concentration; Ca, calcium concentration; CP, crude protein; FA, forage accumulation; dNDF48, neutral detergent fiber digestion for 48 h; IVTDMD, in vitro true dry matter digestibility; K, potassium concentration; Lignin, lignin concentration; Mg, magnesium concentration; NDF, neutral detergent fiber; P, phosphorus concentration; TDN, total digestible nutrients.
• * Significant at the .05 probability level.
• ** Significant at the .01 probability level.
• *** Significant at the .001 probability level.
• ^† ns, not significant.

Forage accumulation showed large phenotypic variability among accessions (Figure 2a) where the mean predicted value was 3,236 kg ha⁻¹, and FA ranged from 411 to 6,485 g kg⁻¹ across all accessions. Tifton 85 had the highest FA among cultivars (red dashed line in Figure 2a), and some PIs exhibited greater FA than Tifton 85, such as PI 316507, Breeding Line 8, and Breeding Line 240 (Supplemental Table S1). Among cultivars, only Tifton 85 and Cheyenne had FA greater than the population mean (blue dashed line in Figure 2a).

FIGURE 2

Histogram for forage accumulation (FA; a), in vitro true dry matter digestibility (IVTDMD; b), crude Protein (CP; c), phosphorus concentration (P; d), neutral detergent fiber (NDF; e), NDF digestible in 48 h (dNDF48; f), lignin concentration (g), and total digestible nutrients (TDN; h) using predictive values estimated averaged across harvests and environments for the USDA-NPGS bermudagrass germplasm collection evaluated in Ardmore, OK. Nutritive value all expressed in g kg⁻¹ and FA in kg ha⁻¹. The dashed lines represent mean values for each trait for: the population mean across all accessions (blue), and the cultivars Tifton 85 (red), Cheyenne (green), Midland (purple) and Wrangler (orange)

A principal component analysis was performed using predicted values for FA obtained for each harvest. The first two principal components (PC) explained 80.5% of the genetic variation observed in this bermudagrass germplasm. The first PC accounted for 69.9% of the existing variation for FA and had an eigenvalue of 3.49 (Figure 3). Germplasm in the left side of the panel (negative values for PC1) showed low forage accumulation in Oklahoma across all harvests. Several cultivars, including Wrangler and Midland, exhibited low FA. Approximately half of the accessions grouped in the right side of the panel (positive values for PC1), indicating greater FA across harvests. For this group, there is a set of germplasm that exhibited greater FA in harvests one, two, and three (positive values for PC1, and negative values for PC2), including the cultivars Tifton, Tifton 44, Tifton 78, Florakirk, and Newell, and several PIs. Another cluster showed greater FA in harvests four and five (positive values for PC1 and PC2), including the cultivars Tifton 85, Callie, and Coastal and several PIs. The cultivar Tifton 85 and entry 323 (PI 316507) produced the highest FA across all harvests (Figure 3; Supplemental Table S1).

FIGURE 3

Principal component analysis (PCA) of forage accumulation across five harvests (1 to 5, eigen vectors) for the USDA-NPGS bermudagrass germplasm collection: Harvest 1 (7 July 2016), Harvest 2 (9 Sept. 2016), Harvest 3 (22 May 2017), Harvest 4 (18 July 2017), and Harvest 5 (21 Aug. 2017). Mean values for forage accumulation for each genotype across environments were used to construct the PCA

3.2 Genetic parameters for nutritive value

Large genetic variation was observed for all NV traits in this germplasm (Table 1). Genotypic and genotype × harvest interaction variances were greater than zero (P < .001) for all NV traits (Table 1). Genotype × environment interaction was significant for [P] (P < 0.001), [K], [Mg], and dNDF48 (P < .05). Repeatability ranged from 0.24 (Ca) to 0.59 (NDF), and ADF, NDF, TDN, and dNDF48 had estimates greater than 0.50. Genotype × harvest correlations (r_gh) ranged between 0.73 to 0.90, and genotype × environment correlations (r_ge) ranged between 0.96 to one. The CV_g was low for all NV traits and ranged from 3.4 (TDN) to 12.9% (CP). Accuracy and reliability were very high for all NV traits.

A PCA using predicted values was created to characterize the genetic variation observed and NV traits. First, a PCA using all NV traits had its first two principal components (PC1 and PC2) accounting for 74.9% of the existing variation (Figure 4). The PC1 presented an eigenvalue of 6.28 and explained 52.3% of the variance; NDF, ADF, and Lignin had negative vectors; and the rest of the traits were positive. The magnitudes ranged from 0.16 (Ash) to 0.38 (IVTDMD). Principal Component 2 exhibited an eigenvalue of 2.71 and explained 22.6% of the variation. Potassium concentration [K], ADF, TDN, and dNDF48 had a larger contribution in PC2 than other NV, and TDN presented a positive contribution in PC2 (0.42). The remaining principal components accounted for 25.1% and had eigenvalues lesser than 0.95.

FIGURE 4

Principal components analysis (PCA) estimated for the USDA-NPGS bermudagrass germplasm collection evaluated in Ardmore, OK. Twelve traits were determined using near infrared reflectance spectrometry: CP, crude protein concentration; IVTDMD, in vitro dry matter digestibility; TDN, total digestible nutrients; ADF, acid detergent fiber; NDF, neutral detergent fiber; dNDF48, NDF digestible in 48 h; Lignin, lignin concentration; Ca, calcium concentration; Mg, magnesium concentration; P, phosphorus concentration; K, potassium concentration; and Ash, ash concentration. Forage accumulation (FA) measured in kg ha⁻¹

Germplasm in the left side of the panel (negative values for PC1) showed greater lignin, ADF, and NDF. Several cultivars, including Russel, Callie, and Midland, were part of this group. Approximately half of the accessions grouped in the right side of the panel (positive values for PC1), indicating greater TDN, IVTDMD, CP, and mineral concentration (P, Mg, Ca). For this group, germplasm with known C. nlemfluensis background exhibited greater IVTDMD and mineral concentration. Accessions exhibiting greater NV included Tifton 85, Coastcross II, Florakirk, Newell, African Star, and Tifton 68, and several PIs and breeding lines (black entry numbers in graph; Figure 4; Supplemental Table S1). Most C. dactylon cultivars, including Jiggs, Cheyenne, Midland, Wrangler, Coastal, Florida 44, and Alicia grouped in the center of the PCA, indicating lesser variation for NV traits, and lesser NV than C. nlemfluensis cultivars.

3.3 Forage accumulation and nutritive value in selected bermudagrass

A PCA was created using FA and seven NV traits selected based on trait correlations shown in Figure 4, and their importance in livestock nutritional diet (Board on Agriculture and Natural Resources et al., 2001). Figure 5 was created using a selected group of accessions: commercial cultivars (red loadings), and PIs and breeding lines exhibiting improved FA and NV traits (black loadings). The PCA in Figure 5 explained 87.9% (PC1 and PC2) of the variation found in this germplasm (Figure 5). The PC1 explained 58.4% of the variance with an eigenvalue of 4.5, traits provided similar absolute contributions (CP, 0.42; P, 0.40, NDF, −0.38; TDN 0.37; Lignin, −0.35; IVTDMD, 0.44), except dNDF48 (0.16) and FA (−0,21. Lignin, FA, and NDF had negative vectors, while the rest of the traits were positive. Principal Component 2 exhibited an eigenvalue of 2.1 and explained 26.3% of the variation. Forage accumulation and dNDF48 had a larger contribution in PC2 (−0.50 and −0.62, respectively).

FIGURE 5

Principal Components Analysis (PCA) estimated for 290 bermudagrass accessions evaluated in Ardmore, OK, for forage accumulation (FA) and seven nutritive value traits estimated using near infrared reflectance spectrometry. CP, crude protein concentration; IVTDMD, in vitro dry matter digestibility; TDN, total digestible nutrients; ADF, acid detergent fiber; NDF, neutral detergent fiber; dNDF48, NDF digestible in 48 h; Lignin, lignin concentration

Commercial cultivars (red loadings) and selected accessions (black loadings) were color-coded to highlight the phenotypic variation in the improved bermudagrass germplasm (Figure 5). The commercial varieties had greater variability for FA and NV traits than the selected accessions. Two groups were observed among the commercial varieties. Russel, Callie (293), African Star (340), Mislevy, Jiggs, Midland, FL44, Cheyenne, Wrangler, and Burton 200 showed lesser FA and NV. The other cultivars and most of the selected accession high FA and NV. In general, these accessions presented high IVTDMD, CP, [P], and dNDF48. Russel and Callie (293) showed high Lignin and NDF, whereas PI 364484 (117), PI 255456 (270), and PI 292143 (102) presented high TDN, IVTDMD, CP, P, and low NDF (Figure 5).

The range of phenotypic variation for NV traits is presented in Figure 2B–H. The overall mean for in vitro true dry matter digestibility (IVTDMD) was 646 g kg⁻¹, ranging from 579 to 754 g kg⁻¹. Tifton 85 presented a high IVTDMD (697 g kg⁻¹), but 24 accessions presented improved IVTDMD (Figure 2b). The mean CP across all genotypes was 121 g kg⁻¹, and CP ranged from 98 to 180 g kg⁻¹. Tifton 85 presented the highest CP among the checks (137 g kg⁻¹), and 35 accessions had greater CP than Tifton 85 (PI 364485 had the highest CP; Figure 2c). Phosphorus concentration varied from 1.9 to 2.6 and it had mean of 2.1 g kg⁻¹, Tifton 85 exhibited 2.4 g kg⁻¹ of [P] (Figure 2d). Neutral detergent fiber ranged between 551 and 723 g kg⁻¹ with a mean of 662 g kg⁻¹, while Tifton 85 presented 652 g kg⁻¹, a slightly lesser value than the population mean (Figure 2e). The minimum and maximum estimates for dNDF48 were 228 and 343 g kg⁻¹, average was 288 g kg⁻¹ and Tifton 85 presented 343 g kg⁻¹ (Figure 2f). Lignin concentration varied from 50 to 76 g kg⁻¹ and the mean was 65 g kg⁻¹ and Tifton 85 had 60 g kg⁻¹. Total digestible nutrients varied between 564 and 683 g kg⁻¹, and Tifton 85 showed similar TDN to the overall population mean (609 g kg⁻¹).

3.4 Trait correlations

The Pearson genetic correlations among the NV traits were significant (P < .05) except between NDF and [K] (Figure 6). Positive correlations were estimated among mineral concentrations (Ca, P, K, Mg), CP, IVTDMD, and dNDF48, and among ADF, NDF, and Lignin. Total digestible nutrients showed a negative correlation with [K] and dNDF48, at a low magnitude, but a positive correlation with Ash, [Ca], [Mg], IVTDMD, CP, and [P], where correlation coefficients ranged from 0.12 to 0.71. In general, correlations between fiber content indicators and other traits were negative and ranged from −0.12 to −0.82. Coefficients of genetic correlations above .80 were observed, in a positive sense, for CP × [P] and CP × IVTDMD and in a negative sense for IVTDMD × NDF, TDN × NDF, and TDN × ADF.

FIGURE 6

Pearson genetic correlations among nutritive value traits estimated for the USDA-NPGS bermudagrass germplasm collection, evaluated in Ardmore, OK. CP, crude protein concentration; IVTDMD, in vitro dry matter digestibility; TDN, total digestible nutrients; ADF, acid detergent fiber; NDF, neutral detergent fiber; dNDF48, NDF digestible in 48 h; Lignin, lignin concentration; Ca, calcium concentration; Mg, magnesium concentration; P, phosphorus concentration; K, potassium concentration; and Ash, ash concentration. Forage accumulation (FA) measured in kg ha⁻¹

3.5 Response to nitrogen fertilization for forage accumulation and nutritive value

The two N rates were considered as different environments in this study. Significant harvest × environment interaction (P < .001) was observed for all traits. Despite the significant interaction, the harvest effect had greater influence than the environment for some NV traits, as there was greater variation among harvests, than between the two environments (Figure 7). The traits IVTDMD, NDF, and dNDF48 less variation for mean values across harvests and environments (Figure 7). Forage accumulation, [P], Lignin, and CP showed significant variation among harvests and the two environments. Harvest three (May 2017) had the lowest FA and highest CP and [P] mean values, and the opposite results were observed in Harvest 5 (August 2017) (Figure 7). Total digestible nutrient was similar across the first four harvests and both environments in all harvests, and TDN remarkably decreased on Harvest 5. Except for Harvest 3, the application of N increased FA, and CP concentration was greater for the environment with N in all harvests. Lignin concentration was lesser in the N-fertilized plots in the first four harvests.

FIGURE 7

Nutritive value traits of bermudagrass collection evaluated in Ardmore, OK, across five harvests (1 to 5 in x-axis): 7 July, 9 Sept. 2016, and 22 May, 18 July, and 21 Aug. 2017; and two environments (see legend): non nitrogen (N) fertilization (red bars) and 150 kg N ha⁻¹ per harvest (blue bars). TDN, total digestible nutrients; IVTDMD, in vitro dry matter digestibility; CP, crude protein concentration; P, phosphorus concentration; NDF, neutral detergent fiber; dNDF48, NDF digestible in 48 h; L, lignin concentration; and forage accumulation (FA) measured in kg ha⁻¹. All nutritive value traits expressed as g kg⁻¹

4 DISCUSSION

Subtropical and transition regions worldwide offer unique agroecosystems to grow forage crops with high forage accumulation and nutritional value for both grazing and hay managements. Bermudagrass is the most planted warm-season perennial forage species grown on 12 million ha in the southeastern United States (Taliaferro et al., 2004). Recent changes in temperature and precipitation patterns have contributed to an expansion of warm season species towards the transition zone, especially bermudagrass (Baxter et al., 2022a, 2022b). Cynodon dactylon germplasm normally exhibit more cold tolerance and lesser NV than C. nlemfuënsis (Taliaferro et al., 2004); therefore, there is a need to evaluate forage germplasm with the goal of selecting accessions with improved adaptation to the transition zone, and with high FA and NV. The 290 bermudagrass accessions were phenotypically characterized for FA and NV traits under two N rates in Ardmore, Oklahoma. Phenotypic data collected across five harvests in 2 yr, which amounts to 5,800 phenotypic records for each trait, were used to estimate variance components and calculate genetic parameters, and predicted values for each trait are presented in Supplemental Table S1.

4.1 Phenotypic diversity for forage accumulation

The PIs, breeding lines, and cultivars evaluated in this study comprise of various Cynodon species and represent germplasm collected from around the world (Anderson, 2005; Anderson et al., 2009). Significant phenotypic variation was observed for FA in the 290 accessions included in this study. Despite the significant genotype × harvest interaction and genotype × N environment interaction (Table 1), the genotype × harvest correlation (r_gh) and genotype × N environment correlations (r_ge) were high, indicating stability in the phenotypic performance for FA across harvests and N rates. Forage accumulation had the lowest R estimate (Table 1), but the trait showed large phenotypic variability among accessions (Figure 2a), indicating the presence of significant phenotypic variation for this trait. Midland, Wrangler, and Cheyenne are cultivars commonly used in the transition zone for their cold tolerance (Baxter et al., 2022b), but they produced lesser FA than Tifton 85 and most PIs (Figure 2a). Tifton 85, PI 316507, Breeding Line 8, and Breeding Line 240 produced the highest FA across the five harvests and represent valuable germplasm for improving FA in bermudagrass (Supplemental Table S1). The same germplasm was evaluated in Florida, and Tifton 85, Breeding Lines 8 and 240, PI 316507, and other PIs showed high FA and NV traits (Grossman et al., 2021). Breeding line 240 and PI 316507 are tetraploid (2n = 4x = 36; Grossman et al., 2021), and they can be used as parents in crosses to create new breeding lines; while Tifton 85 and Breeding Line 8 are pentaploid (2n = 5x = 45) (Grossman et al., 2021) and sterile. Fine textured varieties (thin leaves and stems), which correspond to tetraploid C. dactylon cultivars (Baxter et al., 2015; Grossman et al., 2021), produced lesser FA than coarse texture (thicker leaves and stems) varieties, which correspond to C. nlemfluensis cultivars or interspecific hybrids (Baxter et al., 2015; Grossman et al., 2021). These results provide insights into the potential area of adaptation for this germplasm in the southeastern United States. The germplasm evaluated in the current study were exposed to two winters with an average minimum temperature of 0.56 °C for the coldest month in Ardmore, OK; however, further studies are needed to quantify the persistence of these lines under cold temperatures in long-term studies, as C. nlemfuënsis germplasm normally exhibits less cold tolerance than C. dactylon germplasm (Baxter et al., 2022b; Taliaferro et al., 2004).

4.2 Phenotypic diversity for nutritive value

The forage core collection was developed using phenotypic data collected in the U.S. state of Georgia for 22 traits including NV, and phenotypic variability is present in the collection for those traits (Anderson, 2005). Significant phenotypic variation was observed for all NV traits in the 290 accessions included in this study. Some NV traits exhibited significant genotype × harvest interaction and genotype × N environment interaction (Table 1), but the genotype × harvest correlation (r_gh) and genotype × N environment correlations (r_ge) were high for all traits, indicating stability in the phenotypic performance for NV. The R estimates were medium to high across traits (Table 1), and large phenotypic variability was present among accessions for all traits (Figure 2b-h). Midland, Wrangler, and Cheyenne had low NV traits (Figures 2b–h, Figure 4). Cynodon nlemfluensis germplasm often exhibits greater NV than C. dactylon (Taliaferro et al., 2004). In this study, germplasm with known C. nlemfluensis (coarse texture) background exhibited greater NV than C. dactylon germplasm (fine texture; Supplemental Table S1). Tifton 85, Coastcross II, Florakirk, Newell, African Star, and Tifton 68, and several PIs and breeding lines had the highest NV (Figure 4; Supplemental Table S1). Newell, Breeding line 240, PI 316507, and other tetraploid PIs also exhibited high NV traits in Florida (de Souza et al., 2020), and represent valuable genetic resources to improve these traits in future breeding efforts.

4.3 Correlation between forage accumulation and nutritive value traits

The goal of forage production is to obtain high accumulation levels of good quality forage. The presence of a positive genetic correlation between two traits exhibiting high heritability suggests that genetic improvement for both traits could be achieved by conducting phenotypic selection only in one trait. The high positive and significant correlations among mineral concentrations (Ca, P, K, Mg), CP, IVTDMD, and dNDF48, and among ADF, NDF, and Lignin, indicate that indirect selection can be achieved for them. Positive phenotypic correlations were estimated for FA and fiber traits (ADF, NDF, dNDF48), and negative estimates between FA and IVTDMD and CP, indicating that simultaneous selection for multiple traits will lead to lower genetic gains for each trait individually. Despite these correlations between FA and NV traits, germplasm with improved FA and NV were selected (Breeding Line 240, PI 316507, and others) and they will be further studied to confirm their merit as parental lines or for cultivar releases.

4.4 Bermudagrass cultivars and accessions with high forage accumulation and nutritive value

Commercial cultivars showed large phenotypic variation for FA and NV traits (Figure 5). In addition, 14 accessions were selected based on FA and NV results from this study, as well as for results performed in Florida by de Souza et al. (2020) and Grossman et al. (2021). The commercial varieties had greater variability for FA and NV traits than the selected accessions. Most of the fine texture varieties such as Russel, Callie (293), African Star (340), Mislevy, Jiggs, Midland, FL44, Cheyenne, Wrangler, and Burton 200 showed lesser FA and NV. Coarse texture cultivars and most of the selected accessions had high FA and NV (Figure 5), indicating the improved traits in the selected accessions. Tifton 85, Breeding Lines 8 and 240, PI 316507 (Entry 323), and Tifton had the best combination of high FA and NV (Supplemental Table S1; Figure 5). Four fine texture accessions (Entries 286, 282, 281, and 276 in Supplemental Table S1; Figure 5) showed greater forage accumulation than most fine texture cultivars. These four entries are C. dactylon types and tetraploids (Grossman et al., 2021), and represent valuable resources to improve FA in bermudagrass. PI 316507 and the cultivar Newell are coarse texture types and tetraploids, which can also be exploited for improving FA and NV in bermudagrass. Tifton 85 produced the highest FA and had the best NV among cultivars; however, its utilization for further crop improvement using traditional selective breeding is prohibited given its odd ploidy level (2n = 5x = 45) that results in sterility.

4.5 Phenotypic response to multiple harvests and nitrogen rates

Given the need for reliable phenotypic records to estimate genetic parameters and make informed decisions to select germplasm as parental lines in breeding programs, a combination of multilocation, multiyear, multienvironment, and multiharvest trials are conducted with germplasm collections (Anderson, 2005; Anderson et al., 2009; de Grossman et al., 2021; Souza et al., 2020; Rios et al., 2019;). However, screening large germplasm pools for quantitative traits that are costly and laborious, such as FA and NV, requires major efforts. In this study, 290 accessions were tested in two N environments replicated two times (290 × 2 × 2 = 1,160 experimental units), and plots were manually harvested five times in 2 yr (1,160 × 5 = 5,800 phenotypic records for FA and each NV trait). Genetic parameters estimated using linear mixed models revealed the presence of significant phenotypic variation for all traits with medium to high R estimates (Table 1), which was expected given the previous work performed with this collection (Anderson, 2005; Anderson et al., 2009; de Grossman et al., 2021; Souza et al., 2020). While the genotype × harvest and genotype × N environment interactions were statistically significant for most traits in this study, the high estimates for genotype × harvest and genotype × N environment correlations indicated that genotype rankings (predicted performance for each trait, i.e., best linear unbiased predictor) for each harvest and for each N environment were consistent. In other words, best cultivars and/or accessions were consistently ranked high, and worst cultivars and/or accessions were consistently ranked in the bottom (despite minor changes in ranking positions). These results showed that screening for highly diverse and large germplasm pools, such as this collection (Anderson, 2005; Anderson et al., 2009), could provide meaningful results even if performed with fewer harvests and/or environments. After making selections for target traits in large populations, multilocation, multiyear, multienvironment, and/or multiharvest trials can be performed with fewer cultivars and/or accessions to reduce the efforts and focus on germplasm with true merit for the traits of interest.

Multiple approaches can be implemented in forage species to improve NV. Efforts in forage breeding have resulted in improvements in NV (Burton et al., 1967, 1984, 1993; Casler, 2001), as well as forage management practices through fertilizer applications (Schneider-Canny et al., 2019), and NV varies throughout the year (de Souza et al., 2020). Nitrogen is a limiting nutrient in forage production and environments with low or high N availability might affect potential performance of cultivars (Schneider-Canny et al., 2019). Despite the statistically significant genotype × N environment interaction, the application of N did not result in significant phenotypic variation between both environments (high correlation between genotypes and N environment), meaning that the cultivars and/or accessions producing high FA and NV were the same in both environments. Similarly, there was variation across harvests for FA and NV, but the best cultivars and some selected accessions were consistently ranked high across the five harvests. The magnitude of the genotypic variance was much greater than the genotype × harvest and the genotype × N environment variances, resulting in consistent genotypic performance.

5 CONCLUSION

A Cynodon spp. collection composed of 290 from the USDA-NPGS and the core forage collection were screened in a replicated trial under two N rates (0 and 150 kg ha⁻¹ per harvest) in Ardmore, OK. Significant phenotypic variation for FA and NV is present in this germplasm, and PIs were selected for further evaluation as cultivar candidates. Low genotype × harvest and genotype × environment interactions indicated genotypic stability and potential for selecting improved accessions for FA and NV. The application of N increased FA and CP concentration in all harvests but did not significantly impact phenotypic variation in other NV traits. All phenotypic records will be deposited in the Germplasm Resource Information Network (GRIN)-Global database for the benefit of the forage bermudagrass community. The results obtained in this study will aid plant breeders and geneticists in the development of new bermudagrass cultivars and breeding populations.

ACKNOWLEDGMENTS

This research was partially funded by the USDA National Institute of Food and Agriculture Project 2022-67013-36252, and National Institute of Food and Agriculture, Grant/Award Number: Hatch Project 1018058.

AUTHOR CONTRIBUTIONS

Cleber H. L. de Souza: Data curation; Formal analysis; Investigation; Visualization; Writing – original draft; Writing – review & editing. Claudio C. Fernandes Filho: Data curation; Formal analysis; Writing – review & editing. Raquel S. Canny: Data curation; Investigation; Methodology; Writing – review & editing. Normal Sharma: Data curation; Investigation; Writing – review & editing. Malay Saha: Conceptualization; Data curation; Funding acquisition; Investigation; Methodology; Project administration; Resources; Supervision; Writing – review & editing. Marcelo O. Wallau: Formal analysis; Resources; Supervision; Writing – review & editing. Lisa Baxter: Formal analysis; Resources; Writing – review & editing. William Anderson: Conceptualization; Project administration; Writing – review & editing. Karen Harris-Shultz: Formal analysis; Resources; Writing – review & editing. Esteban F. Rios: Conceptualization; Formal analysis; Funding acquisition; Project administration; Resources; Supervision; Visualization; Writing – original draft; Writing – review & editing.

CONFLICT OF INTEREST

The authors report no conflicts of interest.