Genetic stability observed in third-generation progeny trial of Acacia mangium: the importance of genotype by environment interaction assessment in advance generation breeding strategy

Abstract

The breeding program for Acacia mangium has entered advanced-generation breeding cycles through adopting a recurrent selection system and a sub-lining breeding population. Genetic variation changes along the successive generations could affect its genetic stability on wide ranges of sites. The aim of this study is to observe genetic stability in third-generation progeny trials of A. mangium established at three different sites in Indonesia. Analysis was conducted, including single-site and multi-sites analyses for height, diameter, and stem forking that were grouped into two sets of analysis based on the genetic background of the trial: SET01 for the single sub-line and SET02 for the composite sub-lines. Index selection for multiple-traits was then used to identify the family changing ranks for multiple-traits and genetic gain prediction. The results showed that the recurrent selection system adopted in the breeding strategy for single-site analysis could maintain sufficient genetic variance of A. mangium in the third-generation progeny trial. Family heritability was moderate to high for almost all traits. However, a strong genetic-environment interaction (G × E) exists in multi-sites analysis for the single sub-line population (SET01), indicating a less sufficient genetic variation and a low Type B genetic correlation in anticipating a wider range of environment. On the contrary, compositing selected family from several sub-lines (SET02) could diminish the strength of G × E and increase Type B correlation. Selection and genetic gain prediction could be more effective in multi-sites analysis for SET02, but it was less effective for SET01. The results imply that adopting a recurrent selection system in advanced-generation breeding of A. mangium should consider structuring the breeding population. It could be practiced by compositing selected superior families from several sub-lines into one breeding population to maintain high genetic stability, while increasing genetic diversity and productivity.

Keywords:

• Recurrent selection system
• sub-lining system
• tree breeding
• genetic gain

Introduction

Tree improvement for advanced-generation breeding is commonly practiced using recurrent selection within a breeding population over successive generations, in which the appropriate genetic selection and deployment of improved seed are fundamental outputs (Zobel and Talbert 1984; White et al. 2007). However, there is often a discrepancy between the magnitude of gain predicted from breeding populations and that are realized commercially at operational scale (Verryn et al. 2009), which may be linked to genetic stability as the effects of genetic-environment interactions (G × E). The G × E is affected by the genetic background, including its quality and variation, and by environmental factors, including the soils, climates, and silvicultural treatments. In an advanced-generation breeding program, the genetic quality could be accumulated, but the magnitude of genetic variation tends to decrease due to the genetic selection process. This conflict in the selection impact between increasing genetic gain and decreasing magnitude of genetic variation in the breeding population can be linked to G × E (Li et al. 2017). The importance of observing G × E has been reported by many studies, such as in crops (Ghaed-Rahimi et al. 2015; Hassani et al. 2018; Shahriari et al. 2018), and in tree species (Wang et al. 2016; Chmura et al. 2021).

In Indonesia, Acacia mangium Willd is one of the major species for plantations. Along with the increased productivity, A. mangium is currently in its third-generation of breeding, and cumulative genetic gain has been reported (Sunarti et al. 2012; Nirsatmanto and Sunarti 2019). A recurrent selection system and sub-lining breeding populations have been adopted in the breeding strategy (Burdon and Namkoong 1983). Following the genetic selection in breeding population, molecular studies have reported that selection in breeding programs has led to a little reduction in the genetic diversity of first-generation breeding of A. mangium (Widyatmoko et al. 2006; Yuskianti and Isoda 2012). However, further breeding cycles through successive advanced-generations can simultaneously lead to reduced genetic diversity (Wu et al. 2016). Consequently, G × E within the advanced-generations of A. mangium may occur and weaken genetic stability, which then stimulates different genetic expression of trees and a loss of genetic gain if there have been inappropriate genetic selection and seed deployment strategies. Therefore, an assessment of G × E from several advanced-generation breeding populations of A. mangium established at different sites is necessary to find out the strategy and to ensure that selection during the breeding process is not linked to the loss of potential genetic gain on an operational scale.

Information about how G × E affects the genetic stability of A. mangium is sparse and mostly based on first- and second-generation breeding populations (Nirsatmanto et al. 1996; Setyaji 2013). Previous studies reported that family-site interactions were small for almost all growth and form traits in the first-generation and tended to increase in the second-generation. There is no information from such G × E studies on the advanced-generation breeding of A. mangium concerning the fact that the number of families that are related by descent can increase by up to 75% over three successive generations of breeding (Nirsatmanto and Hashimoto 1994; Chigira and Leksono 2001; Sunarti et al. 2012).

Despite the high cumulative genetic gain that occurs with successive generation breeding programs, any G × E for key traits must be accounted for genetic selection and seed deployment strategies. In this study, genetic variation in third-generation progeny trials of A. mangium established at each of two pairs of contrasting sites in Indonesia is used to evaluate the effects of G × E on their genetic stability and genetic gain prediction. Although many tools have been developed and used to investigate the effect of G × E, such as AMMI and GGE biplot (Wang et al. 2016; Hassani et al. 2018; Shahriari et al. 2018; Chmura et al. 2021), the simple statistical analysis through ANOVA was made to observe the G × E for A. mangium in this study. The results will be discussed in the context of establishing the most appropriate strategy for an advanced-generation breeding program of A. mangium that can minimize any potential effects of G × E on genetic stability and optimize genetic gain prediction.

Materials and methods

Progeny trials

Four third-generation progeny trials of A. mangium were established as part of a comprehensive breeding program for the species in Indonesia. The trials were derived from selected plus trees in second-generation, which were set up under a recurrent selection system and sub-lining breeding populations system (Burdon and Namkoong 1983). Following the selection in each generation of sub-lines, the propagation population is then set up by establishing a composite seedling orchard (Kurinobu 1993; Kurinobu et al. 1998). The orchard consists of a combined the best ten families selected from each of four sub-lines of progeny trial, and for the purpose of this study it is referred to as a composite progeny trial.

Four third-progeny trials used in this study were established in three sites: two trials in Wonogiri, Central Java; one trial in Banten, West Java; and one trial in Parung Panjang, West Java. For this study, the four progeny trials were then grouped into two sets of analyses according to the genetic background of the original family resources: SET01 consisted of two progeny trials from one sub-line, and SET02 consisted of the two composite progeny trials. A detailed description and experimental design of the trials are presented in Tables 1 and 2.

Table 1. Site descriptions of the third-generation progeny trials of Acacia mangium Willd established at three locations.

Descriptions	Sites
	Central Java	West Java 1	West Java 2
District	Wonogiri	Banten	Parung Panjang
Latitude (South)	07^o 32′	5°50′	06^o 30′
Longitude (East)	110^o 4′	105° 7′	106^o 20′
Elevation (m)	141 m	70	51
Climate type (Schmidt and Ferguson)	C	A	A
Rainfall (mm/year)	1850	2000–2500	2000–2500
Max.temperature (^oC)	33.23	27.6	31.5
Min.temperature (^oC)	21.21	25.8	23.8
Slope (%)	<10	<10	<5
Soil type	Vertisols	Podsolik	Podsolik haplik

Table 2. Experimental design of the third-generation progeny trials of Acacia mangium Willd established at three locations.

Descriptions	Progeny trials
	SET 01		SET 02
Sites	Banten, West Java	Wonogiri, Central Java	Parung Panjang, West Java	Wonogiri, Central Java
Site Area (ha)	2.6	1.25	3.2	1.92
Provenances^a	PNG	PNG	PNG, QLD	PNG, QLD
Planting Date	February, 2010	February, 2012	December, 2011	December, 2011
Site cultivation	manual	manual	manual	manual
Fertilizer	NPK, 2 times/year	NPK, 2 times/year	Urea, 2 times/year	NPK, 2 times/year
Previous vegetation	Acacia spp.	Dalbergia latifolia	Acacia spp.	Dalbergia latifolia
Design^b	RCBD	RCBD	RCBD	RCBD
Family	30	30	40	40
Replication	12	6	50	30
Trees/plot	4	4	1	1
Spacing (m)	4 × 2	4 × 2	4 × 4	3 × 3

Remarks: ^aPNG: Papua New Guinea; QLD: Queensland; ^bRCBD: randomized complete block design.

Measurement and data analysis

At two years of age, measurement was taken for three traits: height, diameter at breast height (dbh), and stem forking for all trials. Height and diameter were measured using metric scales, while the stem forking was observed as axis persistance. Stem forking was assessed using a scoring system at five classes: score 1 is forking in the first fifth of the total tree height; score 2 is stem forking in the second fifth of the total tree height; score 3 is stem forking in the third fifth of the total tree height; score 4 is stem forking in the fourth fifth of the total tree height; and score 5 is stem forking in the top fifth of the total tree height, or free-stem forking.

Analysis of variance and covariance for statistical analysis was made at single-site and multi-sites for each of SET 01 and SET 02 by using the following linear model (Matheson and Raymond 1984; Johnson 1992):

Single-site

(1) $${\mathrm{y}}_{\mathit{ijk}}=\mu +{\mathrm{B}}_i+{\mathrm{F}}_j+{\text{BF}}_{ij}+{\mathrm{e}}_{\mathit{ijk}}$$(1)

where, y_ijk, µ, B_i, F_j, BF_ij, e_ijk were the individual tree observation, the overall mean, the effect of the ith block, the effect of jth familiy, the effect of the ith block and the jth family interaction, and error, respectively.

Multi-sites

(2) $${\mathrm{y}}_{\mathit{ijkl}}=\mu +{\mathrm{S}}_i+\mathrm{B}{\left(\mathrm{S}\right)}_{ij}+{\mathrm{F}}_k+{\text{SF}}_{ik}+\mathrm{B}\left(\mathrm{S}\right){\mathrm{F}}_{\mathit{ijk}}+{\mathrm{e}}_{\mathit{ijkl}}$$(2)

where, y_ijk, µ, S_i, B(S)_ij, F_k, SF_ik, B(S)F_ijk, e_ijkl were the individual tree observation, the overall mean, the effect of the ith site, the effect of jth block within the ith site, the effect of kth family, the effect of the kth family and the ith site interaction, the effect of kth family and the jth block within the ith site interaction, and error, respectively. In multi-sites analysis, families are considered to be random and sites fixed. The variance component of each trait and the covariance component between two traits for family and family-site interaction were then calculated from the expected mean squares of the analysis of variance.

Family mean heritabilities (h²_f) for single-site were calculated using the following formula (Zobel and Talbert 1984): (3) $${\mathrm{h}}^2{}_f={\mathrm{\sigma }}^2{}_f/\left({\mathrm{\sigma }}^2{}_f+{\mathrm{\sigma }}^2{}_{fb}/\mathrm{b}+{\mathrm{\sigma }}^2{}_e/\text{nb}\right)$$(3) while those for multi-sites were (Johnson 1992): (4) $${\mathrm{h}}^2{}_f={\mathrm{\sigma }}^2{}_f/\left({\mathrm{\sigma }}^2{}_f+{\mathrm{\sigma }}^2{}_{fb/}\text{bs}+{\mathrm{\sigma }}^2{}_{fs}/\mathrm{s}+{\mathrm{\sigma }}^2{}_e/\text{nbs}\right)$$(4) where, σ²_f, σ²_fb, σ²_fs, σ²_e were component variance of family, family-block interaction, family-site interaction and error, respectively. The n, b, s, were number of trees per plot, block and sites, respectively.

Genetic correlation (r_g) between traits within site were calculated by using the formula (Falconer 1981), (5) $${\mathrm{r}}_g={\text{cov}}_{xy}/{\left({\mathrm{\sigma }}^2{{}_f}_x\times {\mathrm{\sigma }}^2{}_{fy}\right)}^{½}$$(5) where, cov_xy, σ²_fx, σ²_fy were covariance between the trait x and y, component variance of the trait x and y, respectively.

The type B genetic correlation (r_B), that is genetic correlation between the same traits at two different sites, were calculated by using the formula (Burdon 1977): (6) $${\mathrm{r}}_{\mathrm{B}}={\mathrm{\sigma }}^2{}_f/\left({\mathrm{\sigma }}^2{}_f+{\mathrm{\sigma }}^2{}_{fs}\right)$$(6) where the notations in the formula are the same as defined in Equation (4)(4) $${\mathrm{h}}^2{}_f={\mathrm{\sigma }}^2{}_f/\left({\mathrm{\sigma }}^2{}_f+{\mathrm{\sigma }}^2{}_{fb/}\text{bs}+{\mathrm{\sigma }}^2{}_{fs}/\mathrm{s}+{\mathrm{\sigma }}^2{}_e/\text{nbs}\right)$$(4) .

In this study, genotype stability was assessed from rank changes for each family across the sites, and it was derived from two basis methods, the mean of each family based on each trait (single-trait) and multi-traits selection index. For single trait, the family ranking was made on the basis of height, diameter and stem forking, independently. While for multi-traits, the ranking was made on the basis of the index value which was calculated using procedures given by Nirsatmanto et al. (1996): (7) $$\mathrm{b}={\mathrm{P}}_f{}^{-1}\times {\mathrm{G}}_f\times \mathrm{a}$$(7) where, b, P_f, G_f and a are a vector of coefficient of weight on each trait, the phenotypic variance – covariance matrix, the family variance – covariance matrix, and a vector of relative economic weights, respectively. The elements of P_f were variance and covariances of family means, the size of which were 3 × 3 for single-site test, and 6 × 6 for multi-sites test. In the case of the index for across site, covariances between traits at different sites in both P_f and G_f: element of diagonal blocks, were the same and they were derived from Type B correlation (White and Hodge 1989). Relative economic weights representing the relative economic importance of a trait: elements of a, were determined as inverses of phenotypic standard deviations of respective traits.

Genetic gain on each trait (Δg) was predicted by the following formula (Hazel 1943): (8) $$\Delta \mathrm{g}=i\times {\mathrm{G}}_f{}^{\prime }\times \mathrm{b}/({\mathrm{b}}^{\prime }\times {\mathrm{P}}_f\times \mathrm{b})$$(8) where, intensity of selection (i) was assumed to be 1.0 for convenience of computation which is equal of 38% best families. To predict correlated responses on other sites, G_f′ used in the selection index for single-site were 6 × 3 in size which were composed of family variance-covariance matrix at respective site (3 × 3) and the off-diagonal block of Type B covariance (3 × 3).

The degree of overall family changing ranks were then obtained by the method of Matheson and Raymond (1984), which could be summarized in formula as follows: (9) $$\Delta Ȳ=\frac{{\Delta y}_1+ {\Delta y}_2}{n}$$(9) where, $\Delta Ȳ, {\Delta y}_1,$ ${\Delta y}_2,$ and n are a mean rank deviation, rank deviation in the site y₁ from the combined sites, rank deviation in the site y₂ from the combined sites, and number of combined sites, respectively. Family having greater mean rank deviation indicates more interactive to different sites.

Results

Growth and stem forking

In general, growth performances varied among the sites. Tree growth, including height and diameter, in the SET01 analysis was better than that in the SET02 analysis (Table 3). Within the SET01, the growth of trials in Banten was better than that in Wonogiri, while within the SET02, the growth was similar between the sites of Parung Panjang and Wonogiri. In the case of stem forking scores, SET02 was better than SET01, with similar values between the two trials within each set analysis.

Table 3. Growth performance, component of variances (σ²), heritability (h²_f) and genetic correlation (r_g) for height (H), diameter at breast height (D) and stem forking (SF) within each site and across the two sites of third-generation progeny trial of Acacia mangium Willd for SET01 and SET02.

Sites	Banten			Wonogiri			Across sites
Traits	H (m)	D (cm)	SF	H (m)	D (cm)	SF	H (m)	D (cm)	SF
SET01
Mean	10.4	9.5	3.8	7.1	6.5	3.5	9.2	8.5	3.7
σ^2f	0.068*	0.369**	0.050**	0.095*	0.256**	0.018	0.007	0.075	0.001
σ^2fs	–	–	–	–	–	–	0.087*	0.228**	0.046**
σ^2e	1.180	3.688	0.702	1.051	3.142	0.723	1.135	3.491	0.709
h^2f	0.41	0.62	0.67	0.37	0.48	0.27	0.06	0.23	0.03
rg^a	–	–	–	–	–	–	0.07	0.25	0.02
Sites	Parung Panjang			Wonogiri			Across sites
Traits	H (m)	D (cm)	SF	H (m)	D (cm)	SF	H (m)	D (cm)	SF
SET02
Mean	6.0	6.4	4.6	6.3	6.1	4.5	6.1	6.3	4.6
σ^2f	0.059**	0.088**	0.011**	0.074**	0.275**	0.007	0.051**	0.153**	0.010*
σ^2fs	–	–	–	–	–	–	0.017	0.030	0.004
σ^2e	1.671	3.814	0.423	1.625	3.617	1.113	1.654	3.741	0.680
h^2f	0.61	0.50	0.52	0.55	0.67	0.14	0.60	0.68	0.45
rg^a	–	–	–	–	–	–	0.75	0.83	0.73

Remarks: ^aType B genetic correlation; **significant at level confidence 1%; *significant at level confidence 5%. H, D, SF are height, diameter at breast height and stem forking, respectively. σ²_f, σ²_fs, σ²_e are components of variance for family, family-site interaction and error, respectively.

Analysis of variance

The single-site analysis of variance in SET01 and SET02 indicates that family differences of A. mangium were statistically significant for all measured traits in all third-generation progeny trials, except for stem forking in Wonogiri (Table 3). The level of significance of height growth in SET02 was higher (p < 0.01) than that in SET01 (p < 0.05), while for the diameter was the same (p < 0.01). It indicates that the expression of genetic variation for tree height due to the difference in genetic background is larger than the diameter.

The strength of family variation in the single-site analysis is not always expressed in the multi-sites analysis. The family differences in multi-sites were statistically significant for all measured traits in SET02 but not significant in SET01 (Table 3). Following such family variation, family-site interaction in SET02 is not statistically significant for all traits, while that in SET01 is significant. It indicates that the magnitude of family variation would affect its genetic stability and expression across the different sites due to G × E.

Genetic parameter

For SET01, family means heritability in single-site analysis for all traits ranged from 0.27 to 0.67, with the stem forking in Banten having the highest value, and the stem forking in Wonogiri having the lowest one. While for SET02, the heritability ranged from 0.14 to 0.67, with diameter as the highest value and stem forking as the lowest one, both from the trial in Wonogiri. The low heritability for stem forking was linked to the lower genetic variation of the trait, as indicated by the non-significant family for this trait in the two trials in Wonogiri. For multi-sites analysis, combining data from the two sites showed the increased heritability in SET02 as the effect of the increased family variance. On the contrary, combining data from multi-sites for SET01 diminished the family variance and thereafter provided a lower level of heritability for all traits.

The genetic correlation between height and diameter was higher than the phenotypic correlation in almost all the sites (Table 4). But the correlation of height and diameter with stem forking was low, and it even tended to be negative, particularly in the most productive site of Banten. The negative genetic correlation between height and diameter with stem forking indicates the difficulty of improvement between growth and stem forking simultaneously. In terms of stem forking scoring, most of the trees were in score 4 for SET01 at ranges of around 40–50% of the trees, corresponding to score 5 for SET02 at around 70% (Table 5).

Table 4. Genetic correlation (above diagonal) and phenotypic correlation (bellow diagonal) among the traits within site of third generation progeny trial of Acacia mangium Willd for SET01 and SET02.

	SET 01						SET 02
Sites	Banten			Wonogiri			Parung Panjang			Wonogiri
Traits	H	D	Stem forking	H	D	Stem forking	H	D	Stem forking	H	D	Stem forking
H	–	0.92	−0.21	–	0.96	0.07	–	0.82	0.30	–	0.95	0.06
D	0.79	–	−0.46	0.78	–	0.07	0.86	–	−0.04	0.85	–	−0.79
Stem forking	−0.09	−0.27	–	0.13	0.03	–	0.15	0.03	–	0.05	−0.06	–

Table 5. Distribution of stem forking scores incorporated to mean of H and mean of D in third generation progeny trial of Acacia mangium Willd for SET01 and SET02.

Sites	Banten					Wonogiri
Stem forking scores	1	2	3	4	5	1	2	3	4	5
SET01
N (%)^a	2.86	4.20	23.02	50.52*	19.39	2.92	9.09	35.51	41.34*	11.15
Mean of H (m)	10.2	10.6*	10.6*	10.4	10.0	6.0	6.8	7.1	7.3*	7.1
Mean of D (cm)	10.0	10.9*	10.3	9.5	8.3	5.5	6.3	6.6	6.7*	6.1
Sites	Parung Panjang					Wonogiri
Stem forking scores	1	2	3	4	5	1	2	3	4	5
SET02
N (%)^a	0.28	0.56	7.68	18.42	73.05*	4.34	4.15	8.20	8.58	74.71*
Mean of H (m)	4.4	3.6	5.3	6.1*	6.0	5.6	6.0	6.4	6.9*	6.3
Mean of D (cm)	2.8	3.9	5.9	6.9*	6.3	5.1	6.5	6.9	7.3*	5.9

Remarks: ^aNumber of trees for each stem forking scores in percentage (%), *the highest value for respective parameter.

Type B genetic correlations for all measured traits for the SET01 and SET02 analyses are presented in Table 3. Type B genetic correlations for SET01 were moderately low, at ≤0.25 for all traits, while those for SET02 were typically high, at >0.70. These correlations reflected the small amount of family variation as compared to the family-site interaction observed in multi-sites analysis in SET01, while the amount of family variation was larger than the family-site interaction in SET02 (Table 3). These results imply that the deviation of families changing rank in multi-sites might be larger for SET01 than for SET02.

Rank changes

The summary of family changing ranks derived from single-trait and multiple-traits of selection indices in single-site and multi-sites are presented in Table 6. The changing ranks were calculated as a rank deviation between single-site and multi-sites (Equation 9(9) $$\Delta Ȳ=\frac{{\Delta y}_1+ {\Delta y}_2}{n}$$(9) ). The larger range of ranks deviation indicates an increase of unstable families in multi-sites. The number of stable families in SET01 was smaller as compared to SET02. For ranks deviation less than 3, around 20–30% stable family for all traits were observed in SET01, while around 50% were in SET02, except for stem forking (around 30%). The difference in genetic background between SET01 and SET02 seems to affect the strength stability of the family across sites, regardless of the growth superiority. In SET01, number family was similar between single-trait and multiple-traits analysis for all ranges of ranks deviation. While in SET02, the similar number family was observed only in the lower ranges of ranks deviation (<6) (Table 6).

Table 6. Summary of family changing ranks based on range of ranks deviation for SET01 and SET02 analysis.

Range of ranks deviation	Number of family SET01				Range of ranks deviation	Number of family SET02
	Single-traits			Multiple– traits		Single-traits			Multiple– traits
	H	D	SF			H	D	SF
<3	10	7	7	6	<3	21	25	13	21
3.1–6	11	13	14	14	3.1–6	12	12	9	14
6.1–9	4	7	5	6	6.1–9	6	3	10	3
9.1–12	4	3	3	3	9.1–12	1	0	4	2
>12	1	0	1	1	>12	0	0	4	0

Genetic gain

Relative genetic gain predicted by selection index at single-site and multi-sites for SET01 and SET02 analyses are presented in Figure 1. The relative gains varied among the traits and sets of analyses. In SET01, selection indices of multi-sites for height and diameter provided a lower gain than those of indices at single-site. While for stem forking, the gain was slightly greater in multi-sites than in single-site. In SET02, the gains derived by selection index, multi-sites were greater than those of indices at single-site for all traits, which indicates the effectiveness of multi-site analysis. In addition, the correlated response of gains at another trial site based on indirect selection tends to be greater than the direct selection in single-site SET02, but it is lower for SET01. These results suggest that the strength of genetic stability affected the effectiveness of genetic gain prediction in multi-sites analysis. The genetic stability in SET01 was lower than that in SET02, as was partially confirmed by the difference in the significance level of family-site interaction between the two sets of analyses (Table 3).

Figure 1. Relative gains (%) in third-generation progeny trial of Acacia mangium Willd for SET01 (A) and SET02 (B). gains predicted by selection index on each site test (dark grey), and their correlated responses gains at another site test (light grey). White bars are the gains by selection index across the two sites test. The dbh is diameter at breast height.

Discussion

In this study, the genetic stability caused by G × E from advanced-generations of breeding observed in third-generation progeny trial of A. mangium could be discussed in two points of view. It involves: (1) the impact of the recurrent selection system as guidelines of selection practiced in advanced-generations of A. mangium, in relation to the effects of genetic variation; and (2) the impact of the genetic background of family tested in the breeding population of A. mangium, in relation to the effects of structuring the breeding population.

The genetic stability of A. mangium through three successive generations of breeding selection was observed based on the two-years age in four third-generation progeny trials established in three sites. Although it might still be in juvenile growth for a fast-growing species, the genetic stability revealed through G × E observation here would be useful for understanding the breeding strategy of A. mangium in advanced-generation. In terms of structuring the breeding population, the analysis of the study here was made into two sets of analyses: SET01 and SET02, in which the such two sets of analyses contain one common sub-line population (Oriomo provenance from Papua New Guinea). While in terms of site, both contain one common site trial establishment (Wonogiri in Central Java) (Tables 1 and 2).

Growth and stem forking

Growth and stem forking are the traits repeatedly used as criteria of selection along the first- and second generations of A. mangium breeding population in implementing recurrent selection system (Kurinobu et al. 1996; Nirsatmanto and Kurinobu 2002; Nirsatmanto et al. 2004; 2015). In addition, these referenced studies reported that superiority of growth and stem forking varied among the four sub-lines provenance in which the single sub-line of Oriomo, as used in SET01, is the best one for growth, and another sub-line of Claudie River, as a part of composite sub-lines trials in SET02, is best one for stem forking. These results indicate that the adopted breeding strategy through a recurrent selection system and a sub-lining breeding population system might imply a new insight in managing the advanced-generation breeding populations for A. mangium for simultaneous improving some traits of interest.

In this present study, growth (height and diameter) in the respective of two progeny trials within each set analysis varied in SET01, while it was similar in SET02. While among all of four trials, progeny trial established in Banten (SET01) showed the best growth. Such varied growth traits might be due to the differences in genetic background, climate, and soils that influence the growth and development of trees (Clutter et al. 1983). Appropriate genetic background and site conditions are expected to favor the growth of A. mangium. These factors may have contributed to a better growth of height and diameter in Banten (Table 3) than the other sites. Progeny trial in Banten is a single sub-line that was established using genetic material originated from provenance of Oriomo, Papua New Guinea. As a common sub-line available in the four trials studied here, the superiority of Oriomo provenance also confirmed in the SET02 in which >50% of the families from this sub-line was identified as the top-ten ranking for growth trait (data not shown). Other studies were also reported that Oriomo provenance is the most productive provenance of A. mangium for plantations in Indonesia (Kari et al. 1996; Nirsatmanto et al. 2004; 2014).

In terms of site conditions, the site in Banten was by far the most productive, and it has higher annual rainfall and better soil types (>2500 mm/year; podzolic soils) than the other two sites (1800 mm/year; podzolic haplic and vertisol soils). In addition, soil N and P contents were low in Parung Panjang and Wonogiri (unpublished data). Height and diameter growth at both sites may be declining due to the low nutrient supply. This is because A. mangium grows very quickly on high fertility sites (Panitz and Yaacob 1992).

In the case of stem forking, the trials in SET02 showed better average scores than that in SET01. This discrepancy in stem forking scores between SET01 and SET02 might be due to the different intensity of selection and the impact of recurrent selection system practiced for this trait over the past two generations breeding cycle of the parental population. SET01 consisted of the best-thirty families selected from one single sub-line of second-generation progeny trial, while SET02 consisted of the best-ten families selected from each of the four sub-lines of second-generation progeny trial, one of which is a common sub-line provenance used in SET01. It means that the intensity of selection of family in SET02 is three times higher than that practiced in SET01. As a result, the proportion of trees in highest scores of 5 tends to be higher in SET02 (>70%) than in SET01 (20%) (Table 5) which then affected the different average values of stem scores. It was also confirmed that the trend of frequency distribution of trees across the five scores were slightly different between SET01 and SET02, particularly on the peak of distribution. For SET01, the peak of distribution in the two trials was found in score 4, while for SET02 it was found in score 5 (Table 5). In addition, such different trends might be related to the effect of decreased genetic variance in third-generation breeding, particularly in poor site condition, such as in Wonogiri (Table 3). Therefore, the discrepancy in growth and stem forking between the trials and the sites might be due to the different site conditions, such as soil and climate, and the different genetic background used in the trials, such as provenance (Tables 1 and 2). Wu et al. (2016) stated that genetic variance tends to decline with the advancement of generations of selection. Other studies in pines species reported that stem forking effects might be mostly due to environmental control, such as site fertility, and the inconsistency of the phenotypic measurement on these characters (Zobel and Jett 1995; Jin et al. 2014). Thus, at age 2 years, differences in growth and stem forking performance among all sites may just be developing, and this may also depend on genetic stability across different sites for the third-generation breeding population of A. mangium.

Genetic × environment interaction

The magnitude of family-site interaction variance for the third-generation progeny trial of A. mangium in SET01 was substantially higher than the size of the family variance component, in which the ratio was more than 50%. It indicates the presence of G × E and its serious effect on genetic gain prediction from selection and testing (Shelbourne 1972). It suggests that under a single sub-line population, the relative performance of genotype derived from advanced-generation of A. mangium tend to vary across the different site environments, whereby certain genotypes may perform well in one site but poorly in another. Only 20–30% of the families in SET01 showed relatively stable ranks across the two sites, as it was indicated by the range of ranks deviation <3 (Table 6). However, the genetic background of A. mangium used in the progeny trial might become another factor to limit the presence of G × E. It can be observed in SET02 as a series of trials established using composited genetic materials from four sub-lines population. A non-significance of family-site interactions was observed for SET02, suggesting that all relative performances of measured traits in two sites are more genetically determined and less influenced by the difference in environmental factors. This means that most ranks of genotypes tested (>50%) tend to perform similarly across the different sites (Table 6). The high magnitude of genetic variation seems to be the key parameter to limit G × E effects in advanced-generation breeding population of A. mangium through decreasing the ratio of genetic-environment variance to genetic variance (Shelbourne 1972).

Corresponding to the genetic parameters, unlike the family heritability of the traits in single-site analysis, which was moderate to high, the multi-sites analysis in SET01 showed low heritability (h_f ≤0.23), even near zero for height and stem forking (Table 3). The type-B genetic correlation was also low (r_g≤0.25). On the contrary, the analysis in SET02 showed relatively similar family heritability between single-site and multi-sites analysis, even the stem forking increased in the multi-sites analysis. The type-B genetic correlation in SET02 was also higher (r_g>0.73) than that in SET01. The high heritability combined with the high type-B correlation could provide an effective multi-sites analysis for selection, estimating genetic gain, and determining the deployment strategy of the improved stocks (Li et al. 2017). The discrepancy in family heritability estimated between SET01 and SET02 indicates the differences in magnitude of genetic variation, in which the tested number of families in SET02 was around 30% higher than SET01 (Table 2). In addition, whereas the original provenance of families in SET01 was limited to Papua New Guinea, the provenances in SET02 were likewise wider, coming from both Papua New Guinea and Queensland. Intra- and inter-specific variation of species and high heritability estimates could be a good basis for genotype selection and the improvement of target traits in breeding programs (Hassani et al. 2018; Shahriari et al. 2018).

Overall, the presence of G × E highlights the importance of considering both genotype and environment when selecting and deploying A. mangium for improving productivity. The results of G × E analysis in this study reveal that the magnitude of genetic variation is one of the important parameters, which could not only affect the genetic parameters but also the genetic stability of families tested in the advanced-generation breeding population of A. mangium. Selection in breeding populations is commonly practiced producing high-quality genetically improved trees, in which the higher intensity of selection could provide better superior parent trees. In consequence, such selection stimulates a decrease in genetic variation in breeding populations (Zobel and Talbert 1984; White et al. 2007).

In this study, recurrent selection was practiced in the breeding strategy of A. mangium, which successively entered the third-generation of breeding through the progeny trial establishment. The SET01 analysis represents the appearance of high G × E in the third-generation progeny trial of A. mangium. Although the variation among families were still significantly different in each single-site analysis, combining the data from two progeny trials for multi-sites analyses showed a non-significant difference for all the traits (Table 3). In addition, following the rule of variance component ratio of family-site interaction to family (Shelbourne 1972), the ratio value is more than 50%, indicating a serious effect on the genetic stability of families across the two sites due to the presence of G × E. Imposing a recurrent selection system to generate successive advanced-generations up to third-generation breeding in A. mangium tends to diminish its genetic relative performance stability across varying environments.

In comparison, studies on the family-environment interaction of A. mangium in first-generation (Nirsatmanto et al. 1996) and second-generation (Setyaji 2013) progeny trials as parental of family resource used in the third-generation here showed that the family stability observed in multi-sites analysis on first-generation was better than that on second-generation. Family-site interaction in the second generation was statistically significant, but the strength was less than the result in the present study. Although the result from three sets of generational analysis here was conducted at different sites, it indicates that the G × E is more evident in the advanced-generation of A. mangium. Therefore, consideration in the deployment strategy of the improved seed produced is necessary for A. mangium when the breeding strategy is processed under a recurrent selection system to determine whether it will be regionalized in a specific adaptive area or in a wide range of site environments.

As mentioned in the preceding paragraph, the genetic background between SET01 and SET02 is different. In terms of the structure of the breeding population, the SET01 consists of families from single sub-line provenance in Papua New Guinea (Table 2). While the SET02 was a composite of four sub-lines from Papua New Guinea and Queensland. Therefore, the larger variation in SET02 is not only due to a higher number of families tested (40 families), but it also consists of a wide range of provenance origins. Families in SET01 were derived from one of the best superiors of growth of sub-line provenance-origin either in the first generation or the second-generation of breeding. While the SET02 was composed of the best ten families from each of four sub-lines of different provenance origin as a part of establishing a composite seedling orchard, or as an orchard consisting of seedlings from multiple genetically diverse selected parent trees origin. This study reveals that the genetic variation in such an orchard could be increased while maintaining better overall performance in terms of growth and stem forking at the two different trial sites. In this study, the larger genetic variation tends to increase the resilience of families in the composite orchard to environmental stressors such as drought and low rainfall.

Genetic gain

Genetic gain prediction is a representative measure for decisions on assessing the impact of genetic stability due to G × E on selection and deployment strategy (Li et al. 2017). The multiple-traits index selection used in this analysis confirms that the genetic gain from family selection, whether on a single-site or multi-sites basis, was affected by the magnitude of genetic variation. Single-site analysis in both analyses (SET01 and SET02) showed that recurrent selection practiced along the third-generation breeding pathway still retained sufficient genetic variation in all progeny trials, as indicated by a significant difference among families and the high estimated heritability for almost all traits (Table 3). However, the presence of high-level G × E could diminish the strength of family variance as observed in multi-site SET01 analyses. On the other hand, a low level of G × E could become a measure of a large family variation, as observed in multi-sites SET02 analyses (Shelbourne 1972).

Selection based on multi-sites in SET01 analysis through combining data from the two third-generation progeny trials of A. mangium showed that a high level of G × E, as indicated by a significant difference in family-site interaction, resulted in smaller genetic gain predictions for height and diameter as compared to direct selection based on each single-site analysis (Figure 1). Moreover, the response to indirect selection between the two sites was also substantially smaller than the direct selection at each site. It means that family selection at one site based on the selection criteria of another site or the combination of two sites would be less effective. On the contrary, the SET02, which showed a low level of G × E, showed that both multi-sites analysis and indirect selection resulted in higher genetic gains than those of direct selection in the single-site analysis. The low level of G × E indicates the effectiveness of family selection based on combining data from two sites for improving height and diameter in the third generation of A. mangium, particularly for the trial in Wonogiri in SET02. This means that selecting families based on performance across multiple environments can increase the chances of selecting genotypes that perform well across different locations (Li et al. 2017).

In terms of stem forking, the result seems to be unclear. The levels of G × E do not change the benefits of multi-sites analysis for genetic gain prediction for stem forking in spite of the fact that there was significant family-sites interaction. High percentage number of families with a little fluctuation in the ranking (<6) for stem forking was observed that is similar to height and diameter at around 70% (Table 6). As a results, the genetic gain from multi-sites analysis for this trait was greater (>17%) than the single-site analysis, whether under a high or low level of G × E. This result might be also supported by the nature characteristic procedure for calculating the coefficient weight of the trait in the multiple-trait index selection (Nirsatmanto et al. 1996). This procedure stimulated to obtain more weight for stem forking over the other traits in multi-sites analysis. Overall, these findings suggest that structuring of the breeding population in a composite seedling orchard can be valuable tools for increasing genetic variation to diminish G × E, which would provide further better response in selection and genetic gain prediction of important traits in A. mangium.

Implication

Tree breeding is practiced as a long-term strategy over several successive generation breeding cycles to ensure a simultaneous increase in productivity (Zobel and Talbert 1984; White et al. 2007). Regarding this purpose, a recurrent selection system is often adopted for forest tree breeding through repeating the genetic selection process over generations while maintaining sufficient genetic variation in the target breeding generation (Burdon and Namkoong 1983). Whatever the system, genetic selection could improve the productivity of trees, but on the other hand, it tends to gradually diminish the magnitude of genetic variation, whereby as more advanced-breeding generations are achieved, more variation simultaneously decreases. This study revealed that the magnitude of genetic variation in the advanced-generation breeding population of A. mangium affects the relative performance of genotypes at differing sites, which in turn would affect heritability, genetic gain, and deployment strategy unless breeding populations are appropriately structured to deal with G × E.

The SET01 analysis indicates the presence of a high level of G × E due to low genetic variation in the third-generation progeny trial of A. mangium that was established under a recurrent selection system. In this trial, structuring multiple populations through a sub-lining system in the breeding population resulted in a high improvement over family selection based on a single-site. But a such selection would not be followed by the response of improvement in another corresponding site. Thus, the deployment of genetic materials on specific and known sites having similar environmental conditions would be beneficial for a sub-lining breeding population of this kind.

Maximize potential genetic gains for varying environments, and selecting the right individuals best adapted to differing environments is likely to be another essential approach, particularly to support an efficient genetic selection and deployment (Li et al. 2017). Following the sub-lining system strategy in structuring the breeding population, other benefits could be realized through establishing the composite seeding orchard (Kurinobu 1993; Kurinobu et al. 1998). In this type of orchard, higher intensities of selection (IS) could be applied within each sub-line, whereby all the selected genotypes from all the sub-lines are then mixed to establish a composite seedling orchard, such as the trial used in SET02. The genetic variation in multi-sites analyses could be increased to diminish the strength of G × E, particularly in terms of moving into advance generations breeding.

Overall, the results in this study imply that the effect of G × E could be quantified in terms of family ranking change and genetic gain prediction. In this case, the breeding program for A. mangium for advanced-generations could be structured to achieve high genetic gain from selection and wide deployment for a variety of sites. Regionalization of breeding strategies could be applied through the sub-lining population and recurrent selection system. However, anticipating wide ranges of deployment in the next generation of breeding cycles, the composite seed orchard would be one of the most valuable tools for the breeding strategy for A. mangium because it allows for increasing genetic variation. In turn, it can lead to better overall performance of genotypes for growth and form traits in differing sites to make an efficient genetic selection and deployment of genetic material strategy. In addition, observing the effect of age maturity on G × E is necessary to strengthen the comprehensive results of the study in the advanced generation breeding program of A. mangium.

Conclusion

Advanced-generation breeding cycles for A. mangium have been proven to increase tree productivity. A recurrent selection system practiced in the breeding population of A. mangium for single-site analysis could maintain sufficient genetic variation in the three successive generations of breeding. However, in multi-sites analysis, the presence of G × E affected the effectiveness of further selection and deployment. The G × E had affected the reduction of genetic gain in multi-sites analyses in the third-generation progeny trial of A. mangium, which indicates the limitations of the improved material deployment for wide ranges of site environments. Breeding populations management through a sub-lining system for A. mangium (SET01) showed less genetic stability across multiple sites in more advanced-generations of breeding, such as the third-generation progeny trial studied here. However, compositing the best genetic materials from several sub-lines into one breeding population (SET02) could be used to increase genetic variation in advanced-generation breeding of A. mangium, which in turn would minimize the presence of G × E and increase the genetic stability of multiple sites.

Author contributions

Arif Nirsatmanto (A.N.), Sri Sunarti (S.S.), Asri Insiana Putri (A.I.P.), Liliek Haryjanto (L.H.), Noor Khomsah Kartikawati (N.K.K.), Toni Herawan (T.H.), Fajar Lestari (F.L.), Sugeng Pudjiono (S.P.) and Anto Rimbawanto (A.R.) contributed in conceptualization, methodology, software, validation, formal analysis, investigation, resources, data curation, writing—original draft preparation, writing-review and editing, visualization, supervision, and project administration. All authors had an equal role as main contributors in discussing the conceptual ideas and the outline, providing critical feedback for each section, and writing the manuscript. All authors have read and agreed to the published version of the manuscript

Acknowledgements

We would like to express our deepest gratitude to The Centre for Forestry Instrument Standard Assessment, The Ministry of Environment and Forestry for providing the plots as the valuable resources in support of our research. Our thanks also go to the entire team involved in this research for their hard work and dedication in measurement and data entry. Thanks are due to Dr. Christopher Beadle, who kindly reviewed earlier drafts of the manuscript and suggested improvements.

Disclosure statement

No potential conflict of interest was reported by the author(s).