Genetic Parameters and Genotype–Environment Interactions in Paulownia Clonal Tests in Temperate and Subtropical Regions of China

Abstract

Clonal forestry has developed rapidly in recent years and already plays a significant role in commercial tree plantations worldwide. Clonal breeding requires accurate assessments of genetic parameters, together with measurements of clonal productivity, stability, and adaptably. However, relevant studies for clones of Paulownia spp. genotypes are rare. We therefore conducted clonal tests on twenty Paulownia clones established at three sites in the temperate and subtropical regions of China. Trees were planted in a randomized block design, with four replications in each site, twenty plots in each block, and six to eight individuals of the same clone in each plot. We measured the trunk diameter at breast height (DBH), total trunk height (H_t), and individual stand volume of 7-year-old trees to estimate genetic parameters and analyze genotype–environment interactions. A combined analysis of variance indicated that clonal, site, and clone–site interactions significantly affected the three growth traits. Clonal heritability and individual heritability were 0.35–0.84 and 0.07–0.30, respectively. The phenotypic and genetic correlation coefficients among the growth traits were 0.46–0.93 and 0.85–0.99, respectively. There were extremely significant positive linear relationships between the best linear unbiased predictors for DBH and the original DBH values (R² > 0.98). Clones 10, 2, 18, and 13 were selected for deployment based on a selection intensity of 1.4, GGE biplots, and the relative performance of harmonic means on genotypic values analysis. For these clones, the genetic gains in DBH, H_t, and volume were 18.05%, 21.46%, and 46.03%, respectively. These results provide useful information for the selection of Paulownia clones at the target sites and will provide a sound basis for improving Paulownia clonal breeding programs in the future.

1. Introduction

The genus Paulownia is native to China and is among the fastest-growing broad-leaved deciduous species in the world. Paulownia species are widely planted for timber production in subtropical and temperate regions [1]. With its light, silky texture, high dimensional stability, and good acoustic resonance, Paulownia wood is used in the manufacture of furniture, decorative materials, musical instruments, handicrafts, and other finished products [2,3]. Paulownia species have been introduced from China to many countries, including the United States, Japan, Poland, Turkey, Israel, India, Mexico, Canada, and Australia [4,5,6]. The original Paulownia species that were introduced worldwide are increasingly being replaced by superior clones such as Cotevisa 2 (Paulownia elongata S. Y. Hu × Paulownia fortunei (Seem.) Hemsl.) [7], Shantong 3 (Paulownia tomentosa (Thunb.) Steud. × Paulownia fortunei), and 9501 (natural hybrid of Paulownia fortunei), among others [5].

The clonal breeding of forest trees has developed rapidly in recent years because it is capable of producing more productive, better adapted, and more stable genotypes than natural varieties, as well as providing more promising management outcomes in commercial plantations [8,9]. The systematic clonal breeding of Paulownia in China began during the 1970s. More than 40 superior clones were obtained through the combined efforts of several generations of experts. The promotion and application of these clones have made great contributions to the Paulownia industry. However, some clones used in commercial plantations have been cultivated for multiple generations and long-term asexual propagation has produced serious problems, including species degradation. Moreover, the investigations of growth traits in superior clones have been restricted to one site per study [10,11].

Genotypes that are superior in one environment may not be correspondingly superior elsewhere [12,13]. Genotype–environment interactions (GEIs) reflect the relative stability and adaptability of different clones. When the growth traits of a given set of clones are evaluated at multiple sites, the clones are often divided into those adapted to specific sites and those adapted to a wider range of environmental conditions. Clones adapted to a narrow range of conditions are an obvious source of GEIs and differential performance across sites. Clones adapted to a wider range of sites may not produce clear GEIs. When the relative adaptation of certain clones to different environments is unclear, their use in large-scale afforestation risks poor growth and potentially irreparable losses [14]. To sustain a healthy and productive Paulownia industry, it is important to evaluate the performance of Paulownia clones in multi-environment trials.

Accurate estimation of GEIs is a serious challenge in multi-environment trials. Well-designed experiments are critical to the success of any clonal breeding program. These experiments should permit the accurate estimation of genetic parameters to produce more reliable predictions of clonal performance [15]. Interactions between clones and their environment should be fully considered so that clones can be deployed to sites where they perform optimally. Both the calculation of genetic gains and the optimal deployment of superior clones are crucial for designing optimum strategies for clonal forestry [16].

Several analytical techniques have been proposed for the accurate evaluation of GEIs. These include additive main effects–multiplication interaction (AMMI) analysis [17], factor analysis (FA) [18], genotypic main effect plus GEI (GGE) analysis [19,20,21], and best linear unbiased predictor (BLUP)-GGE combined analysis [22,23]. Among these methods, GGE analysis has attracted increasing attention from researchers because it evaluates both genetic gain and GEIs, both of which are relevant to the evaluation of clones. However, the use of GGE biplots in the selection and breeding of forest trees is still rare. One challenge to the use of such fixed-effects-only models is that the phenotypic data used often have heterogeneous error variances, and datasets are frequently imbalanced [22,24].

The calculation of BLUPs has been proposed as a breeding-value prediction method for unbalanced data. Such analyses have been used frequently in recent years [25,26]; they make use of linear mixed effects models of fixed and random effects to calculate unbiased predictions of genetic parameters, such as breeding values [27]. BLUPs can effectively process unbalanced data and eliminate the influence of non-genetic factors or heteroscedasticity among experimental sites. Studies have indicated that a GGE biplot analysis based on BLUP data is more reliable than the phenotypic mean for predicting performance in unbalanced datasets, as well as being more suitable for the evaluation of clonal stability and adaptability [23,28,29].

There have been few investigations of Paulownia genetic parameters in field experiments and no studies of GEIs. Therefore, we conducted a 7-year study of clonal tests at three sites spanning the temperate to subtropical regions of China between 2013 and 2019. We investigated variation in diameter at the breast height (DBH), trunk height (H_t), and individual stand volume of 20 Paulownia clones at each site. The objectives of this study were to estimate genetic parameters, including the coefficients of genotypic/experimental variation, heritabilities, and inter-trait correlations and to investigate GEIs and identify superior clones based on clone productivity, genotypic stability, and adaptability. The results of this study will offer insights into the selection of genotypes when establishing Paulownia plantations.

2. Materials and Methods

2.1. Plant Materials and Experimental Design

The twenty Paulownia clones used in this study were drawn from different hybrids, including nine interspecific hybrids of P. tomentosa × P. fortunei, one interspecific hybrid of P. fortunei × P. elongata, four natural hybrids of P. fortunei, one natural hybrid of P. catalpifolia, two plus-tree clones of P. fortunei, two plus-tree clones of P. elongate, and one plus-tree clone of P. fargesii. Table 1 shows detailed information about these clones.

Table 1. Clones included in the trial and their parentage.

Genotype Code	Clone Identity	Clone Source	Female Parent	Male Parent
1	B9501	Natural hybrid	Paulownia fortunei (Seem.) Hemsl.	**
2	MB9502	Artificial hybrid	Paulownia tomentosa (Thunb.) Steud.	P. fortunei
3	L13	Plus-tree	Paulownia elongata S. Y. Hu	--
4	MB15	Artificial hybrid	P. tomentosa	P. fortunei
5	MB2	Artificial hybrid	P. tomentosa	P. fortunei
6	MB42	Artificial hybrid	P. tomentosa	P. fortunei
7	BL45	Artificial hybrid	P. fortunei	P. elongata
8	L46	Plus-tree	P. elongata	--
9	MB47	Artificial hybrid	P. tomentosa	P. fortunei
10	B49	Natural hybrid	P. fortunei	**
11	B52	Natural hybrid	P. fortunei	**
12	B5	Natural hybrid	P. fortunei	**
13	C61	Natural hybrid	Paulownia fargesii Franch.	**
14	MB70	Artificial hybrid	P. tomentosa	P. fortunei
15	MBP1	Artificial hybrid	P. tomentosa	P. fortunei
16	BP2	Natural hybrid	P. fortunei	**
17	MBP3	Artificial hybrid	P. fortunei	P. tomentosa
18	MBP5	Artificial hybrid	P. fortunei	P. elongata
19	BP6	Natural hybrid	P. fortunei	**
20	YG1	Natural hybrid	Paulownia catalpifolia T. Gong ex D. Y. Hong	**

Note: -- means asexual propagation; ** means male parent unknown.

The three sites selected for this study were located in Xinxiang in the temperate region of Henan Province (E1), in the subtropics at Chibi Town, Hubei Province (E2), and Gongqingcheng Town, Jiangxi Province (E3) (Table 2). Site locations are shown in Figure 1. At each site, 1-year-old clonally propagated buried root seedlings with a root-collar diameter of 4.23 ± 0.5 cm were planted in the spring of 2013 in a randomized complete block design with 4 blocks, each of which contained 20 randomly distributed plots containing 6–8 seedlings of a given clone. Seedlings were planted at 4 m × 5 m spacing, and similar planting and maintenance procedures were used at each test site.

Table 2. Geographical locations and environmental conditions at the three sites.

	E1	E2	E3
City state	Xinxiang, Henan	Chibi, Hubei	Jiujiang, Jiangxi
Latitude (N)	34°22′36″	31°12′49″	29°10′59″
Longitude (E)	113°48′12″	112°52′21″	115°48′33″
Climate zones	Temperate	Subtropical regions	Subtropical regions
Soil	Sandy	Red nitosol	Red nitosol
pH	7.3~8.1	5.2~5.7	5.3~5.9
Mean annual precipitation (mm)	571	1690	1366
Mean annual temperature (°C)	14.4	16.8	16.7
Frost-free period (days)	193~217	247–261	239–266
Altitude (m)	78	535	196

Figure 1. Geographical locations of the three test sites.

2.2. Measurements

The genus Paulownia displays a pseudo-dichotomous branching pattern, in which the total trunk is usually formed by the original trunk and an extended trunk [30].

The 7-year-old Paulownia had reached the early selection stage [14]. We, thus, measured the following variables at the end of 2019 during winter dormancy: original trunk height (H₀), trunk DBH (D_1.3), trunk diameter at 2.6 m above ground (D_2.6), the height of the extended trunk (H₁), the diameter of the extended trunk at 0.5 m above H₀ (De_0.5), and the diameter of the extended trunk at 1.5 m above H₀ (De_1.5). An analysis of covariance was used to adjust heights and diameters to compensate for the influence of initial seedling height and DBH on experimental results. Total trunk height (H_t) and stand volumes were calculated on the basis of the H₀ and H₁ segments, based on Equations (1)–(7), which adapt the methods of Wu et al. [31]:

$$H_t=H_0+H_1$$

(1)

$$V_t=V_0+V_1$$

(2)

$${\beta }_1=\frac{D_{2.6}-D_{1.3}}{1.3}$$

(3)

$${\beta }_2=\frac{D_{e0.5}-D_{e1.5}}{1}$$

(4)

$$V_0=\frac{\pi }{\mathrm{40,000}}\left\{2.6{\left(D_{1.3}\right)}^2+{\left[D_{2.6}-\frac{1}{2}{\beta }_1\left(H_0-2.6\right)\right]}^2\left(H_0-2.6\right)\right\}$$

(5)

$$V_1=\frac{\pi }{4000}{\left(D_{ec}\right)}^2{H}_1$$

(6)

$$D_{ec}=D_{e1.5}-{\beta }_2\left(\frac{H_1}{2}-1.5\right)$$

(7)

where H_t is the total trunk height, V_t is the total trunk volume, V₀ is the segment volume of the initial seedling trunk, V₁ is the segment volume of the extended trunk, D_ec is the diameter at the midpoint of the extended trunk, D_e0.5 is the diameter of the extended trunk at 0.5 m height, D_e1.5 is the diameter of the extended trunk at 1.5 m height, β₁ is the taper factor of the initial seedling trunk, and β₂ is the taper factor of the extended trunk.

2.3. Estimates of Genetic Parameters

The variance components of individual sites and joint analyses were estimated using the linear mixed effects model described in Equations (8) and (9) [32]:

$$y=\mu +Xb+Z_{1}g+Z_{2}gb+e$$

(8)

$$y=\mu +X_{1S}+X_2{b}_{\left(s\right)}+Z_{1}g+Z_{2}gs+Z_{3}gs{b}_{\left(s\right)}+e$$

(9)

where y is the observed value; μ is the global mean; s, b, and b_(s) are the fixed effects of site, block, and block within the site, respectively; g, gb, gs, and gsb_(s) are the random effects associated with genotypes, the interaction between genotype and block, the interaction between genotype and site, and the interaction between genotype and block within site; e is the residual error; and X and Z are the known correlation matrices relating observed values in y with their respective fixed and random effects.

Genotypic variance among clones (σ_g²), variance of genotype by site (σ_gs²), variance of genotype by block (σ_gb²), and the residual variance (σ_e²) were estimated from Equations (8) and (9). Genetic parameters were estimated using Equations (10)–(16):

$$\mathrm{Coefficient} \mathrm{of} \mathrm{genotypic} \mathrm{variation} ({CV}_g[\%]): C{V}_g\left(\%\right)=\frac{\sqrt{{\sigma }_g^2}}{\overline{x}}\times 100$$

(10)

$$\mathrm{Coefficient} \mathrm{of} \mathrm{experimental} \mathrm{variation} ({CV}_e[\%]): C{V}_e\left(\%\right)=\frac{\sqrt{{\sigma }_e^2}}{\overline{x}}\times 100$$

(11)

$$\mathrm{Clonal} \mathrm{heritability} \mathrm{for} \mathrm{a} \sin \mathrm{gle} \mathrm{site} ({h_f}^2): h_f^2=\frac{{\sigma }_g^2}{{\sigma }_g^2+\frac{{\sigma }_{gb}^2}{b}+\frac{{\sigma }_e^2}{nb}}$$

(12)

$$\mathrm{Individual} \mathrm{heritability} \mathrm{for} \mathrm{a} \sin \mathrm{gle} \mathrm{site} ({h_s}^2): h_s^2=\frac{2{\sigma }_g^2}{{\sigma }_g^2+{\sigma }_{gb}^2+{\sigma }_e^2}$$

(13)

$$\mathrm{Clonal} \mathrm{heritability} \mathrm{for} \mathrm{combined} \mathrm{sites} ({h_j}^2): h_j^2=\frac{{\sigma }_g^2}{{\sigma }_g^2+\frac{{\sigma }_{gs}^2}{s}+\frac{{\sigma }_{gb}^2}{bs}+\frac{{\sigma }_e^2}{nbs}}$$

(14)

$$\mathrm{Individual} \mathrm{heritability} \mathrm{for} \mathrm{combined} \mathrm{sites} ({h_i}^2): h_i^2=\frac{2{\sigma }_g^2}{{\sigma }_g^2+{\sigma }_{gs}^2+{\sigma }_{gb}^2+{\sigma }_e^2}$$

(15)

$$\mathrm{Genetic} \mathrm{accuracy} \mathrm{in} \mathrm{clone} \mathrm{selection} (r_{gg}): r_{gg}=\sqrt{h^2}$$

(16)

where $\overline{x}$ is the mean value of the measured trait, n is the reconciled mean number of individuals per clone in each block [33], b is the number of blocks, s is the number of sites, and h² is the clonal heritability.

Phenotypic and genotypic correlations between traits were estimated using Equations (17) and (18):

$$r_{p12}=\frac{Co{v}_{p12}}{\sqrt{{\sigma }_{p1}^2{\sigma }_{p2}^2}}$$

(17)

$$r_{g12}=\frac{Co{v}_{g12}}{\sqrt{{\sigma }_{g1}^2{\sigma }_{g2}^2}}$$

(18)

where Cov_p is the phenotypic covariance between traits x and y, Cov_g is the genetic covariance between traits x and y, σ_p1² and σ_p2² are the respective phenotypic variances of trait x and trait y, and σ_g1² and σ_g2² are the genotypic variances of trait x and trait y.

The analysis was performed using the ASReml-R package from International Limited (VSNi) in the R statistical software environment (version 4.1.0.90).

2.4. Genotype Main Effect Plus Genotype by Environment Interaction Effects

The BLUP values of the 20 genotypes were estimated using a linear-mixed-effects model (Equation (19)) using the lme4 package in R. These BLUPs were then used in a GGE biplot analysis of genotype main effect plus genotype by environment interaction effects [32,34]. The GGEBiplotGUI package was used to conduct the GGE biplot analysis [35]. Biplots of relationships among sites, including Which Won Where/What, discriminative vs. representative, and mean vs. stability were established using the following parameters: Scaling = 0, Centered = G + GE, and SVP = Symmetrical.

$$y_{ij}=\mu +S_i+S{G}_{ij}+e_{ij}$$

(19)

where y_ij are the individual observations of genotype j in environment i, u is the overall mean, S_i is the fixed effect of environment, SG_ij is the random interaction between the jth genotype and the ith environment, and e_ij is the residual error.

2.5. Productivity, Stability, and Adaptability

We used the following parameters to analyze productivity, adaptability, and stability: the harmonic mean of genotypic values (MHVG), the relative performance of genotypic values (PRVG), and the relative performance of harmonic means on genotypic values (MHPRVG). These parameters were calculated with Equations (20)–(22) [36,37].

$$MHVG=\frac{e}{{{\displaystyle \sum }}_{j=1}^e\frac{1}{V{G}_{ij}}}$$

(20)

$$PRVG=\frac{1}{e}\left({{\displaystyle \sum }}_{j=1}^e\frac{V{G}_{ij}}{{\mu }_j}\right)$$

(21)

$$MHPRVG=\frac{e}{{{\displaystyle \sum }}_{j=1}^e\frac{1}{PRV{G}_j}}$$

(22)

where e is the number of sites where genotype i was evaluated, VG_ij is the genotypic value of genotype i in site j, and μ_j is the j-site average.

2.6. Expected Gain

We assumed that the selection proportions could be as high as 20% (4 clones out of 20) and that the selection intensity would be 1.4 [38]. The estimates of genetic gain (ΔG) and actual gain (G) were calculated using Equations (23) and (24) [32]:

$$\Delta G=\frac{i{h}^2{\delta }_p}{\overline{x}}$$

(23)

$$G=\frac{\left(x_i-\overline{x}\right)}{\overline{x}}$$

(24)

where i is the intensity of selection, h² is the clonal heritability, δ_p is the standard deviation, and x_i is the average trait value of the selected clones.

3. Results

3.1. Variance Analysis and Growth Performance

The respective means of DBH, H_t, and individual volume varied from 17.70 cm, 5.76 m, and 0.1319 m³ at the least productive site (E1) to 18.87 cm, 7.73 m, and 0.1465 m³ at the most productive site (E2). The means and variabilities of DBH, H_t, and individual volume at each site are shown in Figure 2. Analyses of variance were performed on clones within each site for DBH, H_t, and individual volume (Table 3). There were highly significant differences between clones for all three growth traits in each site (p < 0.001). An exception was H_t at E2, which was still significant at p < 0.05. This result indicates that artificial selection based on genetic evaluation has the potential to improve the growth traits of Paulownia spp. at the clonal level.

Figure 2. Box plots of DBH, H_t, and volume for clones within each site. Data points are extreme values outside the range of 1.5 times the box length.

Table 3. Analysis of variance of the growth traits of the clones within each site.

Site	Source	df	Mean Square
			DBH	Ht	Volume
E1	Clone	19	34.90 ***	6.52 ***	0.0117 ***
	Block	3	5.77	5.80 ***	0.0019 *
	Clone × Block	57	12.44 ***	4.35 ***	0.0042 ***
	Error	367	3.95	0.99	0.0013
E2	Clone	19	21.39 ***	1.41 *	0.0082 ***
	Block	3	91.13 ***	34.52 ***	0.0332 ***
	Clone × Block	57	20.81 ***	2.15 ***	0.0077 ***
	Error	365	7.53	0.76	0.0024
E3	Clone	19	29.18 ***	18.99 ***	0.0194 ***
	Block	3	139.15 ***	23.56 ***	0.0605 ***
	Clone × Block	57	17.07 ***	7.07 ***	0.0078 ***
	Error	395	4.20	2.68	0.0022

*** Significance at 1‰ of probability of error; * Significance at 5% of probability of error.

The joint ANOVA returned significant main effect for clones, sites, and blocks within sites for all three growth traits (p < 0.01). Significant clone–site interactions on DBH and individual volume (p < 0.05) and on H_t (p < 0.01) were observed (Table 4).

Table 4. Multisite ANOVA for the growth traits of the clones.

Effect	df	Mean Square
		DBH	Ht	Volume
Site	2	172.44 ***	441.0 ***	0.0339 ***
Clone	19	76.35 ***	16.8 ***	0.0336 ***
Block (site)	9	71.94 ***	19.4 ***	0.0271 ***
Clone × Site	38	7.26 *	5.3 ***	0.0032 *
Error	1293	5.65	1.8	0.0021

*** Significance at 1‰ of probability of error; * Significance at 5% of probability of error.

3.2. Estimation of Genetic Parameters

The coefficients of genetic variation (CV_g) and the coefficients of environmental variation (CV_e) were higher for individual volume than for the other two traits at all sites. This trend was observable in single-site analyses and in the joint analysis of all three sites (Table 5). The values of CV_g ranged from 4.44% to 11.34%, and those for CV_e ranged from 11.81% to 30.30%. The clonal heritabilities for single sites (h_f) ranged from 0.35 to 0.64, indicating that the three traits tested were under moderate genetic control, while individual heritabilities for single sites (h_s) ranged from 0.07 to 0.30. The clonal heritabilities in the joint analysis (h_j) ranged from 0.64 to 0.84, while the individual heritabilities returned by the joint analysis (h_i) were lower, ranging from 0.14 to 0.23. The genetic accuracy of growth traits for single sites was 0.55–0.80, while the genetic accuracy returned by the joint analysis was 0.80–0.92.

Table 5. Estimates of genetic parameters for DBH, H_t, and the individual volume of the clones at the three sites.

Site	Traits	CVg(%)	CVe(%)	hf2	hs2	hj2	hi2	rgg
E1	DBH (cm)	5.95	11.23	0.64	0.30	-	-	0.80
	Height (m)	5.04	17.30	0.30	0.10	-	-	0.55
	Volume (m3)	13.78	27.64	0.64	0.30	-	-	0.80
E2	DBH (cm)	3.69	11.88	0.41	0.13	-	-	0.64
	Height (m)	4.75	23.73	0.37	0.07	-	-	0.61
	Volume (m3)	8.98	27.07	0.38	0.13			0.62
E3	DBH (cm)	3.33	11.01	0.35	0.11	-	-	0.59
	Height (m)	9.95	25.11	0.60	0.22	-	-	0.78
	Volume (m3)	13.92	32.09	0.55	0.23	-	-	0.74
Joint analysis	DBH (cm)	4.44	11.81	-	-	0.84	0.23	0.92
	Height (m)	5.54	18.71	-	-	0.64	0.14	0.80
	Volume (m3)	11.34	30.30	-	-	0.81	0.22	0.90

CV_g(%): genotypic variation coefficient; CV_e(%): environment variation coefficient; h_f²: clonal heritability for single site; h_s²: individual heritability for single site; h_j²: clonal heritability for joint analyses; h_i²: individual heritability for joint analyses; and r_gg: genetic accuracy.

There were significant, positive genetic and phenotypic correlations with all three growth traits (Table 6). These correlations allow for indirect clonal selection on these variables, especially for individual volume. Pearson correlation coefficients between DBH and individual volume for genotype and phenotype were all > 0.93. DBH is easy to measure and is less affected by measurement errors than H_t or stem volume; thus, DBH can be used for direct clonal selection. We therefore used DBH as our reference trait in subsequent analyses for this paper.

Table 6. Correlation coefficients among the traits of the Paulownia clones.

Traits	DBH	Ht	Volume
DBH		0.4606 **	0.9329 **
Ht	0.8455 **		0.6247 **
Volume	0.9923 **	0.8984 **

Phenotypic correlation coefficients in the upper triangle, and genetic correlation coefficients in the below triangle. ** represents significant correlations at 0.01 level.

3.3. Relationships between BLUPs for DBH and Original DBH Values

Figure 3 shows the strong linear relationships between BLUPs for DBH and the original DBH values at each site (R² ≥ 0.98). As a result, BLUPs were used in subsequent GGE biplot analyses and optimal genotype screening, which not only had high reliability but also resulted in greater genetic gain [22].

Figure 3. Linear regressions between the original DBH values and the BLUPs of clones at the three sites.

3.4. GGE Biplot Analyses

The GGE biplots illustrated inter-site relationships in the multi-environment trial (Figure 4A). The biplot of DBH captured 94.59% of the total variation in G + G × E, indicating that the relationships were accurately displayed. The first principal component, which was related to the effects of G, explained 89.22% of the total variation. The second principal component, which represented the G × E interaction, explained 6.37% of the variation. The vector angles between the pairs of sites were all acute, indicating that the site values were positively correlated (Figure 4A). The correlation coefficient for sites E2 and E3 was close to 1.

Figure 4. The GGE biplot based on the BLUP values of DBH at three sites. (A) Site relationships with the environmental view of the GGE biplot. The genetic correlation coefficient between sites is estimated based on the cosine value of the angle between two site vectors. When the angle is less than 90°, the smaller the angle, the greater the correlation coefficient. (B) The Which Won Where/What view of the GGE biplot. Optimal genotypes into an irregular polygon were connected and perpendicular lines through each edge of the polygon to the origin were drawn. Tested sites were grouped. Red solid lines intersecting at the sides of the polygon represent mega-environments. (C) The discrimination and representative view of the GGE biplot. The blue solid line with an arrow passing through the origin represents the average environment axis. Dashed blue vectors represent the discrimination of the site. The longer the vector, the greater the discriminatory power. The angle between the dashed vector and the average environmental axis illustrates the degree to which a given site represents the average environment. (D) The mean and stability view of the GGE biplot. The vertical black dashed line segment represents the productivity and stability of each clone. Shorter dashed lines represent better adapted clones. The green solid line perpendicular to the average environment axis represents the mean population DBH. The farther away to the left of the average environment axis, the greater the values of DBH.

As shown in Figure 4B, the four clones (16, 17, 11, and 10) that were farthest from the origin were connected, and the polygon was divided into four sectors. The 20 clones tested were distributed within these sectors. The three sites occupied a single sector and were therefore grouped together. The clones located close to the top corner of the polygon were the highest-yielding clones in each environmental group [37]; thus, the optimal clone in this group (or among the three sites) was clone 10. As shown in Figure 4C, E1, E2, and E3 were all suitable for the tested clones. E3 is representative of the average environment at all sites, while E1 had the greatest discriminatory power. Clones 10, 2, 18, and 13 were optimal performers with both high productivity and stability according to Figure 4D.

3.5. Stability, Adaptability, and Productivity

Although the rank order of test genotypes varied among sites, the two top-ranked clones were the same at all three sites (see Table 7). Furthermore, clonal rankings based on the values of MHVG, PRVG, and MHPRGV were identical across these variables, and the two top-ranked clones were once again 10 and 2. Clones 10 and 2 therefore displayed stable performance, adaptability, and high yields at all three sites. Clones 18 and 13 were also of note for their ability to be cultivated in a wide range of environments, based on the criteria of 20% selection proportion and a selection intensity of 1.4.

Table 7. Ranking of the 20 clones for each site and selection of the optimal genotypes based on DBH; genotypic stability (MHVG); genotypic adaptability (PRVG); and productivity, adaptability, and stability (MHPRVG).

Rank	For Single Site			For Stability and Adaptability
	E1	E2	E3	MHVG	PRVG	MHPRVG
1	10 (a)	10 (a)	2 (a)	10	10	10
2	2 (ab)	2 (ab)	10 (ab)	2	2	2
3	17 (abc)	18 (bc)	18 (abc)	18	18	18
4	18 (abcd)	19 (bcd)	13 (bcd)	13	13	13
5	13 (abcde)	17 (bcd)	3 (bcde)	3	3	3
6	3 (abcde)	3 (bcd)	1 (bcde)	17	17	17
7	14 (bcde)	14 (bcd)	15 (cdef)	14	14	14
8	7 (cde)	13 (bcd)	20 (cdef)	20	20	20
9	20 (cde)	8 (bcd)	7 (cdef)	19	19	19
10	1 (cde)	15 (bcd)	19 (cdef)	15	15	15
11	19 (cde)	12 (bcd)	14 (cdef)	7	7	7
12	15 (cde)	20 (bcd)	12 (cdef)	1	1	1
13	9 (cdef)	7 (cd)	6 (cdef)	9	9	9
14	8 (defg)	6 (cd)	9 (def)	8	8	8
15	4 (efg)	4 (cd)	8 (def)	12	12	12
16	6 (efg)	9 (cd)	5 (def)	6	6	6
17	12 (efg)	5 (cd)	11 (def)	4	4	4
18	11 (fgh)	1 (d)	16 (def)	5	5	5
19	5 (gh)	16 (d)	17 (ef)	11	11	11
20	16 (h)	11 (e)	4 (f)	16	16	16

Note: Clones with different letters in parentheses in the same column are significantly different at the 0.05 level.

3.6. Actual Gain and Genetic Gain

The actual gain (G) was greater than the genetic gain (ΔG) for all growth traits in the four top-ranked clones (Figure 5). Volume displayed the highest G and ΔG among the growth variables, with values of 23.74% and 14.46%, respectively.

Figure 5. Actual and genetic gains of growth traits under a selection intensity of 1.4.

4. Discussion

4.1. Variance among Sites and Clones

Site characteristics affect the performance of different genotypes through the combined effects of the soil and local climate [39]. Analysis of variance is an effective technique to evaluate the magnitude of differences among clones, and variation among populations is the basis for clonal breeding [40]. The coefficient of variation reflects the variability in each population and determines the selection space that is used to decide which clones to choose for breeding purposes [41]. In this study, we found significant differences in sources of variance (clone, site, and clone–site; p < 0.05) for all three growth traits measured in 7-year-old Paulownia clones. These results indicate that the measured variables are feasible criteria for clone selection.

In this study, sites E2 and E3 had warmer climates, higher rainfall, and longer frost-free periods than E1. The clones at site E1 were smaller for all three growth traits than those at E2 and E3, in accordance with the results of Houspanossian et al. [42] and Goldblum et al. [43]. Environmental differences among sites underpinned the differential performance of the clones in our study, indicating that the conditions at sites E2 and E3 were more suitable for the growth of the tested clones.

4.2. Genetic Parameters

The coefficient of environmental variation (CV_e) for DBH and H_t obtained from the joint analysis ranged from 11.81% to 30.30%, demonstrating reasonable experimental precision. These results indicate that the model captured the majority of the variation in the study and that the estimates can be trusted [8]. These values are similar to those found by Zhao et al. [10]. However, the genotypic variation coefficient in this study seems to have been low compared to that reported by Shi et al. [44]. Restricted genetic variability among our 20 clones may have arisen because they were selected from 45 germplasm resources, which were collected and preserved at an early stage based on superior growth traits.

As one of the most important genetic parameters, heritability indicates significant genetic control and can reflect the reliability of selected genotypes [45]. A review of heritabilities in growth trait tests showed that the average heritabilities of DBH and height in forestry species were low, with values of 0.16 and 0.21, respectively [46]. We obtained comparable results for individual heritabilities at single sites (h_s) (0.07–0.30). Higher values of heritability were observed for clonal heritability at single sites (h_f) (0.30–0.64), indicating that superior clones could be identified from the 20 test clones. Some studies of natural populations reported higher heritabilities [47,48], suggesting that increasing the diversity of the breeding population may increase heritability and, consequently, increase the performance reliability of selected genotypes.

4.3. GGE Biplot and Selection of Optimal Clones

Because it is one of the major traits reflecting site quality, tree height has been selected as a representative variable for evaluating the adaptability and stability of forest growth [49]. DBH has also been selected as a representative trait under the same stand densities at the experimental sites when the production of large-diameter timber is a breeding objective [28,37]. In the present study, we used similar planting densities with 4 m × 5 m spacing between individual trees. Furthermore, there were significant positive genetic and phenotypic correlations between DBH and H_t and DBH and volume. Therefore, DBH was used as the target trait in GGE biplot analysis to select clones for fast growth, high yield, and stability. To reduce the influence of heterogeneous environmental errors or unbalanced data, and to improve selection accuracy [50,51], BLUPs were used to conduct GGE biplot analysis in this study. The BLUP values of DBH for the three sites showed good linear relationships with the original DBH values, making it possible to use BLUPs to replace original values in the GGE biplot analysis.

The biplot of DBH captured nearly 95% of the total variation in G + G × E, indicating that our results are reliable. The Which Won Where/What biplot grouped the three sites together in one sector, and clone 10 was located at the vertex of the sector, indicating that clone 10 had the largest DBH at the three sites. Clones located in the other biplot sectors were considered inadvisable for deployment at our experiment sites. However, some of these clones, especially 16, 11, and 17, which occupied the vertices of their respective sectors, might be well adapted to environments not represented in this study [21].

The discrimination vs. representativeness biplot showed positive correlations among the three sites, especially between E2 and E3. E1 had the greatest discriminatory power and representativeness, followed by E3 and E2. The mean vs. stability biplot is used to exhibit the productivity and stability of test clones in a visual comparison of their performance in different environments. An optimal clone should display both high yield and stability within the tested environments [52]. The DBH of clones 10, 2, 18, and 13 were the top four clones on the average environment axis, consistent with the ranking based on MHVG, PRVG, and MHPRVG. These four clones can therefore be selected as optimal genotypes with high productivity and stability. It is noteworthy that, although clone 17 showed mediocre performance in our analysis, it also had a long vector directed toward E1, implying that it had high adaptability in this experiment.

5. Conclusions

The 20 clones used in this study included five Paulownia species obtained through different breeding methods, which established a genetically diverse test population. Genotype, environment, and GEI significantly affected the values of key growth traits among these clones. DBH was positively correlated with H_t and volume, and the BLUPs of DBH for the three sites had strong linear relationships with the original DBH values. Clones 10, 2, 18, and 13 were the best-performing, most stable clones at all three sites, as shown in GGE biplots and MHVG, PRVG, and MHPRVG analyses. Our study points to a robust breeding strategy for the genus Paulownia and should play a significant role in increasing productivity of Paulownia and enhancing the incomes of plantation workers.