Genetic material originating from contrasting European chestnut ( Castanea sativa) populations of Greece, Italy and Spain was evaluated in a common garden test situated in Greece. The aim of the study is to device an appropriate selection strategy by identifying and conserving superior genotypes for current and future use in breeding programs. Breeding material consisted of 143 open-pollinated families growing in a common garden provenance-progeny experimental trial. Growth trait genetic parameters were estimated and response to selection was evaluated using family, within family and combined selection methods. Two models were employed for the estimation of family variance and genetic parameters. The CVA varied between 12.1% and 67% among traits and models, showing an increasing trend with age. Heritability estimates were high; however their variation with age was irregular. Selection of superior families regarding three years of measurement for height, diameter, volume index and number of leaves showed a potential gain of 12% - 25%, 12% - 28%, 33% - 73% and 21% - 49% over the mean of all plants. Genetic gain for volume index was the highest among the traits studied and the joint model used presented a more effective selection strategy. Results indicate that the experimental trial studied presented substantial genetic variation and sufficient genetic gain opportunities for quantitative traits of economic importance. These findings suggest that inferior trees can be rogued from the experimental trial and a seedling seed orchard of Castanea sativa in Greece can be established.
European chestnut (Castanea sativa Mill.), is one of the most important multipurpose broadleaf species in Western Europe and the Mediterranean. The species is widely cultivated for its valuable timber and edible starchy fruits, while it is also used for landscape architecture. In rural areas, chestnut cultivation constitutes an integral part of the economies (Aravanopoulos et al., 2001) . However, the species is considered threatened since it is under direct pressure from a number of environmental and anthropogenic risks. The combination of a higher demand for wood and chestnuts and increased threats, make the selection and conservation of superior genotypes highly impotent. Superior phenotypes can be selected as plus-trees in forest stands and their potential genetic superiority can be established in progeny trials (Zobel & Talbert, 1984) . Response to selection may be predicted on the basis of biostatistical models provided appropriate statistical and genetic parameters are determined from field experimentation (Falconer, 1989) . This approach permits the prediction of the effectiveness of a wide range of various methods of selection, based on different selection criteria. Of particular interest is the “combined family and progeny within family” selection. This method results in a higher genetic gain especially for low heritability traits (Falconer, 1989) which may not be amenable to QTL mapping and MAS selection. Growth traits are often evaluated in breeding programs, since they are important components of plant vigour and biomass production, directly related to wood economic value (Bradshaw & Stettler, 1995) .
In order to study growth traits in C. sativa, a provenance-progeny trial was established in Taxiarchis, Greece based on provenance material that originated from six populations of contrasting environmental conditions across Europe. This trial is part of a network of six similar trials that have been established in Greece, Italy and Spain and involve reciprocal transplantations of provenances. Considerable family variation was already detected at the plant juvenile stage and at the first years after plantation establishment (Pliura & Eriksson, 2002; Fernandez-Lopez et al., 2005; Tchatchoua & Aravanopoulos, 2010) . In addition genetic parameters of growth traits using different models were estimated in another study (Tchatchoua & Aravanopoulos, 2010) . Results indicated that genetic improvement is possible through selection of superior genotypes. The aim of this paper is to use estimates of genetic parameters in order to predict effective selection strategies among open-pollinated families.
Genetic material originated from six natural contrasting European chestnut (Castanea sativa) populations of Greece, Italy and Spain (
The following growth traits were measured four years after planting in the field: diameter (basal stem diameter taken with a calliper), height (shoot height measured from the ground to the top of the highest stem or branch
with a measuring rule), volume index (calculated as
number of leaves per individual). Variances of the independent variables [age, block, provenance, and family (provenance)] were tested for homogeneity by using Levene’s test (Sokal & Rohlf, 1981) . An analysis of variance (ANOVA) was carried out using two models presented below and Duncan’s multiple range tests were calculated for each measured traits. The following models were employed:
1) Model I (Joint Model)
This joint model was employed to test the effect of provenance × age and family × age interactions. The sources of variation taken into consideration were: age, block, provenance, family (provenance), provenance × age and family × age interactions. The model included age, block, provenance and provenance × age interaction as fixed effect and family (provenance) and family × age interaction as random effect.
2) Model II
In model II, analyses were performed to determine the effect of age at measurement. The sources of variation were block, provenance, and family (provenance),
where,
Yijkl = phenotypic observation for the ijklth seedling
μ = the overall mean of the experiment
bi = the fixed effect of the ith block, i = 1,…, 10
pj = the fixed effect of the jth provenance, j = 1,…, 6
f(p)kj = random effect of the kth family within the jth provenance, k =1,…, 9
al = the fixed effect of the lth age, l = 1,…, 3
εijkl = the sampling error
Variance components were estimated by means of the Restricted Maximum Likelihood (REML)/VARCOMP of SPSS software. The derived variance components were used in the computation of additive coefficient of variation (CVA), heritability and response to selection as follows:
Additive coefficients of variation (CVA)
Heritability estimates (Shelbourne, 1992) :
1) family heritability:
n = mean number of individuals per family
a = ages
2) within family heritability:
Response to selection from 10% selection intensity (Zobel & Talbert, 1984) :
1) family selection which involves the choice of entire families on the basis of their phenotypic values:
where,
2) within family selection where individuals are chosen on the basis of their deviation from the family mean, and family values are not given weight when selections are made:
where,
3) combined selection which is a two stage process, involving selection on families followed by selection of individuals within families.
where,
The analysis of variance of all variables over years (
Source | df | Diameter | Height | Volume index | No. of leaves | ||||
---|---|---|---|---|---|---|---|---|---|
Mean square | p-value | Mean square | p-value | Mean square | p-value | Mean square | p-value | ||
Age | 2 | 113.8 | 0.00 | 74565.9 | 0.00 | 191011893.2 | 0.00 | 22205277.1 | 0.00 |
Prov. | 5 | 6.5 | 0.00 | 17789.1 | 0.00 | 15429312.6 | 0.00 | 203505.2 | 0.57 |
Fam. (Prov.) | 49 | 5.5 | 0.00 | 4268.2 | 0.00 | 10852582.5 | 0.00 | 742813.6 | 0.00 |
Age × Prov | 10 | 0.6 | 0.9 | 743.3 | 0.67 | 2136933.5 | 0.72 | 177477.9 | 0.75 |
Age × Fam | 98 | 0.6 | 1.0 | 240.5 | 1.00 | 1985625.1 | 0.99 | 204789.1 | 0.94 |
Error | 925 | 1.37 | 990.6 | 30006766.2 | 263307.1 | ||||
Total | 1142 |
Prov.: provenance, Fam. (Prov.): family within provenance, df: degrees of freedom.
Age | Source | df | Diameter | Height | Volume index | Number of leaves | ||||
---|---|---|---|---|---|---|---|---|---|---|
Mean square | p-value | Mean square | p-value | Mean Square | p-value | Mean Square | p-value | |||
Prov. | 1.75 | 0.050 | 3668.6 | 0.000 | 2117311.1 | 0.001 | 53644.2 | 0.136 | ||
Age 4 | Fam. (Prov.) | 49 | 0.95 | 0.16 | 727.40 | 0.30 | 467233.30 | 0.49 | 18152.60 | 0.99 |
Error | 317 | 0.78 | 656.03 | 470197.70 | 31702.90 | |||||
Total | 380 | |||||||||
Prov. | 2.13 | 0.238 | 5541.08 | 0.000 | 3346911.9 | 0.261 | 118174.0 | 0.45 | ||
Age 5 | Fam. (Prov.) | 49 | 1.90 | 0.17 | 1297.12 | 0.13 | 3246662.50 | 0.12 | 383276.60 | 0.05 |
Error | 317 | 1.60 | 1033.65 | 2564736.20 | 261299.20 | |||||
Total | 380 | |||||||||
Prov. | 5.33 | 0.044 | 11253.8 | 0.000 | 16285610.8 | 0.039 | 298293.6 | 0.715 | ||
Age 6 | Fam. (Prov.) | 49 | 3.57 | 0.02 | 2037.30 | 0.14 | 9817247.50 | 0.04 | 682002.60 | 0.08 |
Error | 317 | 2.30 | 1640.50 | 6848196.60 | 514378.40 | |||||
Total | 380 |
Prov.: provenance, Fam. (Prov.): family within provenance, df: degrees of freedom.
provenances. Regarding the analysis of variance for volume index, the provenance effect decreased at age five and increased at age six (p = 0.039). Provenance effects showed no significant differences at all ages for the number of leaves. The age × provenance was not significant for all the variables measured. The absence of significant interaction variation at the experimental test site for all traits indicated a high correspondence in the different variables with age. The separation of means obtained from the Duncan Multiple Range Test (DMRT), presented the following range: 3.7 - 4.0 for diameter, 95.5 - 119.2 for height, 1229.8 - 1725.0 for volume index and 510.3 - 596.7 for the number of leaves at age six.
The range values of family means for the variables measured were 2.2 - 5.4 cm for diameter, 56.5 - 170.0 cm for height, 264.6 - 3084.0 cm3 for volume index and 75.0 - 1075.0 for the number of leaves, at all ages. Variation among families was significant for all traits regarding Model I, i.e. the joint Model (
The additive coefficients of variation (CVA) values for all traits are presented in
Variables | σ2f(p) | CVA % | h2 family | h2 within family | Rf | Rw | Rc | ||||
---|---|---|---|---|---|---|---|---|---|---|---|
Value | % | Value | % | Value | % | Value | % | ||||
Diameter | 0.19 | 8.1 | 23 | 0.76 | 0.61 | 0.55 | 14.8 | 0.56 | 14 | 1.1 | 31 |
Height | 153.4 | 8.5 | 24 | 0.78 | 0.66 | 15.6 | 15.1 | 16.17 | 15.1 | 31.8 | 35 |
Volume index | 359445.8 | 7.5 | 73 | 0.72 | 0.49 | 723.9 | 44.8 | 673.32 | 41.8 | 1397.2 | 87 |
Number of leaves | 20350.9 | 5.2 | 51 | 0.62 | 0.31 | 159.5 | 28.3 | 126.69 | 22.6 | 286.2 | 51 |
σ2f(p): variance component of family (provenance), CVA: coefficient of variation, h2: heritability, Rf: response to family selection, Rw: response to within family selection, Rc: response to combined selection.
Age | Variables | Diameter | Height | Volume index | Number of leaves | Age | Variables | Diameter | Height | Volume Index | Number of leaves | |||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Age 4 | σ2f(p) Value | 0.027 | 6.32 | 45.4 | 0.00 | Age 5 | σ2f(p) Value | 0.056 | 36.5 | 105411.7 | 16679.6 | |||
% | 2.8 | 0.68 | 0.01 | 0.00 | % | 2.8 | 2.5 | 3.3 | 5.3 | |||||
CVA % | 10.3 | 5.7 | 1.5 | 0.0 | CVA % | 12.5 | 12.1 | 41.6 | 36.2 | |||||
h2 family | 0.19 | 0.06 | 0.001 | 0.00 | h2 family | 0.20 | 0.20 | 0.22 | 0.31 | |||||
h2 within family | 0.14 | 0.04 | 0.001 | 0.00 | h2 within family | 0.14 | 0.14 | 0.17 | 0.28 | |||||
Rf | Value | 0.10 | 0.86 | 0.22 | 0.00 | Rf | Value | 0.13 | 3.87 | 214.9 | 102.3 | |||
% | 3.1 | 0.98 | 0.02 | 0.0 | % | 3.5 | 3.9 | 13.7 | 14.3 | |||||
Rw | Value | 0.17 | 1.39 | 0.37 | 0.0 | Rw | Value | 0.24 | 6.08 | 367.1 | 179.0 | |||
% | 5.2 | 1.6 | 0.04 | 0.0 | % | 6.2 | 6.06 | 23.6 | 25.0 | |||||
Rc | Value | 0.27 | 2.58 | 0.59 | 0.00 | Rc | Value | 0.37 | 9.95 | 582.0 | 281.4 | |||
% | 8.3 | 1.69 | 0.07 | 0.00 | % | 9.7 | 9.96 | 37.4 | 39.4 | |||||
Age 6 | σ2f(p) Value | 0.18 | 47.9 | 424233.8 | 20568.5 | |||||||||
% | 6.2 | 2.2 | 5.1 | 3.5 | ||||||||||
CVA % | 19.3 | 11.0 | 52.6 | 39.1 | ||||||||||
h2 family | 0.35 | 0.17 | 0.30 | 0.22 | ||||||||||
h2 within family | 0.31 | 0.21 | 0.25 | 0.16 | ||||||||||
Rf | Value | 0.35 | 4.08 | 506.1 | 96 | |||||||||
% | 8.1 | 3.4 | 20.4 | 13.1 | ||||||||||
Rw | Value | 0.64 | 6.6 | 883.9 | 154.9 | |||||||||
% | 14 | 5.6 | 35.7 | 21.1 | ||||||||||
Rc | Value | 1.0 | 10.6 | 1389.9 | 250.9 | |||||||||
% | 22.4 | 9 | 56.1 | 38.2 | ||||||||||
σ2f(p): variance component of family (provenance), CVA: coefficient of variation, h2: heritability, Rf: response to family selection, Rw: response to within family selection, Rc: response to combined selection.
The family heritability was high for all the traits regarding Model I, in particular it was 0.76 for diameter, 0.78 for height, 0.72 for volume index and 0.62 for the number of leaves (
Estimated genetic gain at 10% selection intensity among families ranged from 14% - 31% for diameter, 15.1% - 35% for height, 41.8% - 87% for volume index and 22.6% - 51% for the number of leaves among the different selection methods in the joint Model. Among the traits studied, the genetic gain for volume index was the highest (
Concerning the predicted genetic gain (
To aid genetic improvement it is important that superior genotypes are selected, fingerprinted and conserved for future use. Family variation in the provenance-progeny trial of Taxiarchis, Greece indicated that potential genetic improvement is possible through artificial selection. This study evaluated growth trait genetic parameters as an aid for the selection of an optimal breeding strategy for chestnut. The analysis of variance of all variables showed that the factor of provenance to be highly significant. In other provenance-progeny tests of broad leaved species, significant differences attributed to provenance (population) are generally found. For instance, significant differences between Q. robur provenances were found in growth characteristics (Shutz & Badoux, 1979; Kleinschmit & Svolba, 1979; Kleinschmit, 1993; Shutyyaeu, 1999; Jensen, 2000) .
Variation among families was significant for all traits in Model I, while the analysis of variance per year revealed that family effects were significant at age six, increasing with age. Other pertinent studies in Castanea sativa also recorded significant differences among families (Pliura & Erikson, 2002, Fernandez-Lopez et al., 2005) . Strong significant family effects were also encountered for height growth at the juvenile stage of another Fagaceae species, Quercus robur, in a study of open-pollination families in Lithuania (Baliuckas & Pliura, 2003) . Similar family variation was also observed among Betula pendula progenies (Armstrong, 1999) and among B. pendula seed sources from Northern Britain for height growth (Worrell et al., 2000) . Furthermore, congruent results were reported in Fraxinus excelsior where the differences in mean height at age five were found to be statistically significant (Cundall et al., 2003) . Similar results were reported for Prunus avium mean tree height variation among progenies as well (Owe, 2001) . Besides angiosperms, significant family variation for growth was extensively reported in conifers, for example in Picea abies (Skroppa, 1982) , and Pinus contorta spp. latifolia (Xie & Ying, 1996) .
The age × provenance and age × family interactions were not significant for all the variables measured. The absence of significant interaction variation at the experimental test site for all traits indicated a high correspondence in the different variables with age. This is the first time that such analysis has included age as source of variation in Castanea sativa. Similarly, family × year interactions were not significant for all traits measured in a complete diallel performed by crossing Norway spruce grafts of high altitude parents at a low elevation site (Skroppa, 1994) . On the contrary Douglas-fir provenance test showed a significant provenance × age interaction in early selection parameters (Zas et al., 2004) . In this Model I, the data were combined over the three years of measurement to test the effect of the interaction terms among sources of variation. Age × family interactions was statistically not significant indicating high stability across years for all the variables. Based on these results, early selection among families from the test site may be effective. Making selection at early ages is a common practice with the aim of shortening breeding cycles in advanced generation tree improvement programs. Besides accelerating breeding progress, selection at early ages may also offer other advantages such as smaller genetic tests, easier measurement, greater adaptability to changing demands, and quicker delivery of genetic gain to the production population (Lambeth, 1980) .
The CVA for all growth traits followed an increasing trend with age in the C. sativa trial studied. The CVA for height was comparable to that of Pliura & Eriksson, (2002) . However, it was higher compared to the CVA estimates reported by Lauteri et al., (2004) for C. sativa and higher than height CVA estimates reported for other broadleaves species (Baliuckas et al., 1999; 2000) . Family heritability can be high among years and models, while heritability values changed in an irregular way from age four to age six. Family heritability is usually high because it is based on averages estimated with a sample of many progenies. The effects of environmental factors within a test are thus averaged out for the family mean (Zobel & Talbert, 1984) . Heritability estimates were calculated assuming the open-pollinated progenies studied were half-sibs. Any remote possibilities of consanguineous mating, or full-sib mating, might influence heritability estimates. Lauteri et al., (2004) also studied C. sativa heritability on different individuals of the same open pollinated families (i.e. progeny individuals originating from the same mother tree), however using fewer families than this study. They reported higher heritability estimates than the present study. They noted however that only the family and family × treatment components were used in their analysis of variance and the few factors among which total variance was distributed in conjunction to juvenile stage of the plant material (six months old) and the quite uniform environmental conditions (Phytotron experiment) may have led to the higher heritability estimates. Pliura & Eriksson, (2002) also carried out studies on the same open pollinated families. They also estimated higher heritability values than the present study regarding carbon isotope discrimination and height. They attributed their high heritability estimates to the reason explained above which likely resulted in a low environmental variance. Discrepancies in heritability estimates are not unusual and could be attributed to the experimental design, size and composition of the population under study, as well as to the method of data collection (Falconer & Mackay, 1996) . Heritability values that are derived from a single test site as in this report tend to be overestimated (Illingworth, 1978; Ying et al., 1986) . Nevertheless, the estimated heritability values obtained in this study do indicate that the growth traits evaluated are at least moderately heritable. Therefore, they provide evidence that artificial selection may be effective for the improvement of these traits. Heritability estimates showed changes over time which can be attributed to the changes in the external environment with age and to changes in genetic control through gene × environment interaction (Namkoong et al., 1972; Namkoong & Conkle, 1976) . However, no particular trend could be determined in this present study, potentially due to the relatively short measurement period. Elevated heritability values for diameter and height with age were found in Douglas-fir, but at a more advance age of measurement than in this present study (Namkoong et al., 1972; King et al., 1988; Magnussen & Yanchuk, 1994; St. Clair, 1994; Johnson et al., 1997) .
The estimated genetic gain from a 10% selection intensity among families ranged from a low of 12% (diameter, height) to 73% (volume index) among the different selection methods with combined selection having the highest gain. Among the traits studied, the genetic gain for volume index was the highest. Similar results were noticed in the analysis of Tchatchoua & Aravanopoulos, (2010) using models with interaction terms. Hence, family variance has specifically influenced the analysis of genetic gain. Genetic gain estimates for height are comparable to those of Savill et al., (1999) , who estimated a genetic gain of 16% from selecting the best 15% of families obtained at six years from four ash (Fraxinus excelsior) progeny tests. Rink, (1984) predicted genetic gain for height to be up to 25% using family selection strategies in black walnut (Juglans nigra). Higher genetic gain results were obtained in open-pollinated sycamore (Platanus occidentalis) families measured at age seven years (Nebgen & Lowe, 1985) . The estimated high values were perceived to be inflated because members of the open-pollinated families were probably more closely related than half-sibs. The genetic gain for the number of leaves at age four was recorded as zero, since this variable presented zero heritability. Therefore no gain can be accumulated concerning this variable at this age. At age five, combined selection yielded a higher value than family selection. Among all the traits, the highest genetic gain was recorded for volume index and the joint model used resulted in a more efficient selection strategy. Gain can be increased if superior genotypes are vegetatively propagated (Randall & Cooper, 1973; Beineke, 1983) , a possibility that exists for elite mother trees of Greek origin regarding this experimental plantation (Tchatchoua et al., 2014) . This result indicated that the provenance-progeny experimental trial of Taxiarchis has ample amounts of genetic variation in the quantitative traits studied and sufficient gains can be foreseen.
The following conclusions can be drawn from this study: 1) Castanea sativa growth traits at the juvenile stage were under strong genetic control; 2) sufficient heritability was recorded to permit selection; 3) combined selection had the highest genetic gain and the joint Model used resulted in a more efficient selection strategy; 4) notable genetic gains can be realized after rouging inferior trees from the experimental plantation and converting the trial to a seedling seed orchard.
A seedling seed orchard could cover immediate needs for improved seeds for wood production. The economic and temporal gains of this option outweigh the alternative of creating a seed orchard as an entirely new effort.
This work is part of the Ph.D. thesis of the first author, funded by the Hellenic State Scholarship Foundation (IKY), Greece. Assistance in the field measurements by the Aristotle University Forest Fund is gratefully acknowledged. Special thanks are extended to the Taxiarchis Head Forester Mr. G. Panourgias for providing logistical support.