Genetic gain and selection efficiency of loblolly pine in three geographic regions

Genetic gain and selection efficiency of loblolly pine in three geographic regions

Xiang, Bin

ABSTRACT. Genetic parameters of annual height to 8 yr and 4 to 8 yr volume were examined for Northern, Coastal, and Piedmont populations of loblolly pine (Pinus taeda L.) from analyses of 23 disconnected half-diallel progeny tests in the southern United States. Genetic gains in year 8 volume predicted by various selection methods at age 6 revealed that selection on volume yielded more gain than selection on height. Among test regions, the Coastal population had the greatest correlated response, followed by the Piedmont population and Northern populations. Family plus within family selection was the most effective to achieve genetic gain for early selection on both height and volume. Additional gain (10-40%) can be achieved by capturing the nonadditive genetic component through mass production of full-sib crosses or vegetative propagation. Early selection efficiency was examined by using the ratio of either gain per year or present value between indirect selection and selection of year 8 volume. Optimal selection ages were determined for various selection methods. The analysis of selection efficiency showed earlier selection could be more efficient than selection on volume at age 8 or later. Family selection can be performed as early as age 3 for height and at age 4 for volume, which was the earliest measurement year for volume in this study. Family plus within family had its optimal selection age at age 3 or age 4. Based on the relationship of age 8 with rotation gain, the selection criteria and timing were also appropriate. FOR. SCI. 49(2):196-208.

Key Words: Diallel mating, genetic gain, selection methods, age-age genetic correlation, selection efficiency, Pinus taeda L.

IN A TREE BREEDING PROGRAM, the accurate and timely assessment of genetic parameters is critical for predicting future gains and developing successful breeding strategies (Zobel and Talbert 1984). Genetic parameter estimates of loblolly pine (Pinus taeda L.) have been reported from several recent studies (Svensson et al. 1999, McKeand et al. 1999, Li et al. 1996, Williams and Megraw 1994). Numerous age-age correlation studies have been conducted in the last two decades to explore the opportunity of achieving the greatest genetic gain per unit of time; results have been variable for early selections for different species (Wu et al. 2000, Woods et al. 1995, Danjon 1994, Magnussen 1993, Li et al. 1992, Hodge and White 1992, Bastien and Roman-Amat 1990, Cotterill and Dean 1988, Lambeth 1980). Early selection studies on loblolly pine (Li et al. 1992, Li et al. 1991, McKeand 1988, Foster 1986, Lambeth et al. 1983, Franklin 1979) indicated that selection on growth traits at early ages could be effective. These studies used large time gaps between two adjacent measurement years, thus conclusions were based on rather rough trends of genetic parameters. In addition, in most of these studies, the limited family sample size may have restricted the reliability and precision of genetic parameter estimates.

Balocchi et al. (1994) studied selection efficiency from the analysis of data from an unimproved population of loblolly pine. Time trends in genetic parameters for height indicated that, if a single measurement were used, measurement at age 6 and selection 1 year later would maximize the gains per year. However, genetic parameters and trends over time may be different with the improved loblolly pine population that was generated with different mating designs and field layouts.

The loblolly pine breeding program at North Carolina State University has been using a disconnected half-diallel mating to produce progenies for field tests that were established on very uniform land for the second-cycle breeding program (Li, et al. 1996). Some of these well-balanced tests have been measured annually at early ages to provide genetic parameter information and to assess selection efficiency over time for different geographic regions.

In this study, these well-balanced data sets from three major geographic regions were used to estimate genetic parameters for growth traits, such as height, dbh, and volume at early ages. With better estimates of parameters, time trends of selection efficiency were reevaluated for different selection methods in order to achieve maximum genetic gains per unit time. The impact of nonadditive component on genetic gain was considered, since nonadditive variance was found to be an important part of genetic variance at the early growth stage of loblolly pine in many earlier studies (Balocchi et al. 1993, McKeand and Bridgwater 1986, Foster et al. 1986). Such assessments are not only important to provide valuable information for deployment strategies, such as mass production of controlled crosses and vegetative propagation, but are also essential for developing an optimum breeding strategy for the next cycle of loblolly pine breeding.

Material and Methods

Mating Design and Field Design

A total of 275 parents or 690 full-sib families from 23 diallel tests were included in this study. These tests were grouped into the following three general geographic regions for the loblolly pine breeding program in the southeastern United States: (1) Northern: Virginia and northern North Carolina, (2) Coastal: Coastal Plains of South Carolina, Georgia, Florida, Southern Alabama, and Southern Mississippi, and (3) Piedmont: Piedmont of Georgia, South Carolina, and North Carolina.

There were 4 to 11 test series in each test region. In each test series, 2 tests were planted in each of 2 yr, resulting in a total of 4 tests. In each test series, 2 disconnected 6-parent half-diallels without selfs produced 30 full-sib crosses. Each test was planted in a randomized complete block design with 6 blocks each containing all 30 crosses in 6-tree row plots (Li et al. 1996). All four tests in each test series were measured annually for height through age 8. Dbh was measured annually from age 4 through age 8. Tree height was measured to the nearest centimeter, while dbh was measured to the nearest millimeter. Stem volume was calculated using the equation of Gebel and Warner (1966) in cubic decimeters.

Linear Model and Genetic Parameter Estimation

For each test series, the following linear model was used to estimate variance components.

Y^sub ijoklm^ = [mu] + T^sub i^ + B^sub j(i)^ + D^sub o^ + G^sub k(o)^ + G^sub l(n)^ + S^sub kl(o)^ + TG^sub ik(o)^ + TG^sub il(o)^ + TS^sub ikl(o)^ + P^sub ijokl^ + E^sub ijoklm^ (1)

where Y^sub ijoklm^ is the mth observation of the jth block within ith test (location) for the klth cross within oth diallel; [mu] is the overall mean; T^sub i^ is the ith fixed test environment effect, i = 1 to t; B^sub j(i)^ is the fixed effect of jth block within ith test, j = 1 to b; D^sub o^ is the oth fixed diallel effect, o = 1 to d; G^sub k(o)^ or G^sub l(o)^ is the random general combining ability (GCA) effect of the kth female or lth male parent within oth diallel, k, l = 1 to p, k

All random effects are assumed to be independent of each other. The model is a typical mixed model and was analyzed by using the new mixed analytical method (Xiang and Li 2001) with SAS (SAS Institute Inc. 1996). Using the new procedure, dummy variables were first created for GCA effects using SAS PROC IML, then SAS PROC MIXED was used to estimate variance components (SAS Institute Inc. 1996). REML is chosen as the model fitting method, as it was shown to be superior over the ANOVA based estimator (Huber 1993, Searle et al. 1992). This procedure not only can simultaneously perform variance component estimation and BLUP analysis of random genetic effects (e.g., GCA and SCA) like GAREML (Huber 1993), but also can provide convenience and flexibility in data analysis (Xiang and Li 2001).

The variance components of random effects (i.e., [sigma]^sup 2^^sub g^ [sigma]^sup 2^^sub s^ [sigma]^sup 2^^sub gt^ [sigma]^sup 2^^sub st^ [sigma]^sup 2^^sub p^ [sigma]^sup 2^^sub e^)were estimated for each test series (data not shown, see Xiang 2001). Additive and nonadditive genetic variances were derived from their causal relationships with variance of GCA and SCA effects, and various genetic parameters were further calculated by formulas presented in Appendix 1. These parameters include: additive and nonadditive genetic variance; narrow-sense and broad-sense individual-tree heritability; half-sib family-mean heritability; full-sib family-mean heritability (narrow sense); full-sib family-mean heritability (broad sense); narrow-sense within-full-sib family heritability; and broad-sense within-full-sib family heritability (Xiang 2001). To distinguish different selection methods that use the corresponding appropriate heritability in gain predictions, the notations used in this article are summarized in Table 1.

Using the same mixed model analytical procedure, the variance components were estimated on the sum of the values of two variables X and Y, and their genetic correlation was calculated using the following formula:

where X, Y are two traits of interest (e.g., height or volume at any age); [sigma]^sup 2^^sub gx^ or [sigma]^sup 2^^sub gy^ is the GCA variance of trait X or Y; [sigma]^sub gxgy^ is the GCA covariance (i.e., 1/4 of the additive genetic covariance between two traits); [sigma]^sup 2^^sub gxgy^ is the GCA variance (i.e., 1/4 of the additive genetic variance for the sum of values of two traits). The genetic correlation between height or volume at any earlier age and year 8 volume was calculated using the above method. Since the data were generally well balanced in terms of number of trees within each plot, and the estimation of [sigma]^sub e^^sup 2^ was not of particular interest, plot means were used in the analysis to obtain [sigma]^sup 2^^sub gx+gy^.

The estimated genetic parameters were then averaged across all test series in a test region. Since these tests were well balanced and shared the same mating design with equal number of parents and full-sib families, all estimates of each parameter were assumed to have similar precision. Hence, no weighting function was needed to obtain average parameter estimate and simple unweighted averages were calculated.

Genetic Gain Prediction

Genetic gain in trait Y resulting from indirect selection on trait X can be calculated using the following:

CG^sub x[middot]y^ = i [middot] r^sub Gxy^h^sub x^h^sub y^[sigma]^sub y^ (3)

where j = selection age; i = selection intensity; r^sub Gxy^ = genetic correlation between selected trait X and trait Y (e.g., age 8 volume); h^sub x^= square root of heritability of selected trait X; h^sub y^ = square root of heritability of response trait Y.

Trait X can be height or volume at any earlier age. The response trait Y, i.e., selection goal, should be the tree volume at rotation age, typically about 25 yr for loblolly pine. Since year 8 was the latest year measurement for most test series in this study, it is used for the response trait, with the assumption that year 8 volume is a good indicator of rotation age volume, i.e., highly correlated with volume at the age 25 program (Li et al. 1996, Balocchi et al. 1994, McKeand 1988). We will discuss this issue in the “selection efficiency assessment” section below.

Correlated responses in age-8 volume were calculated for indirect selection on height and volume at age 6, the recommended selection age in the current loblolly pine breeding program (Li et al. 1996, McKeand 1988). Different selection strategies were considered: individual or mass selection; individual selection for clonal deployment; backward selection on half-sib family; forward selection on full-sib family-mean based on additive or total genetic component; full-sib family plus within family selection based on additive or total genetic component. Selection intensity i was set to 2.665 (select top 1%) for individual selection and within-family selection. For family selection, a selection intensity of 1.755 (select top 10%) was used.

Selection Efficiency

Different selection methods were evaluated by comparing indirect selection at earlier ages for each trait with selection on volume at age 8. For individual selection methods, mass selection was used for selection on year-8 volume, while for family selection methods, full-sib family selection based on additive genetic effect was used. Comparison was not made between family selection methods and individual selection methods since different selection intensities are required for these two methods. In the combined selection, intensity was set to be 2.665 for within-family selection and 1.755 for among family selection.

Selection efficiency was calculated as either gain per year (GPY) or present value (PV). In our primary analysis, GPY was calculated assuming t = 8 yr from measurement to seed collection (e.g., the time required to complete the breeding for the entire population). Other t values (5 and 12 yr) were also considered to study the sensitivity of the optimal selection ages. Genetic gain was divided by the time to complete a breeding cycle, i.e., selection age plus t yr. The selection efficiency (SE^sub GPY^) was defined as the ratio of gain per year between indirect selection and direction selection. Assuming the same selection intensity i is applied to both indirect selection and direct selection,

where j = selection age, j

Present value utilizes the discount equation (Balocchi et al. 1993, McKeand 1988) with interest rate I fixed at 8% in the primary analysis. Four percent and 12% were also included to investigate the effect of I on optimal selection ages. The only assumption was that s additional years (rotation age) beyond selection were needed to realize the genetic gain. Selection efficiency based on present value (SE^sub PV^) was defined as the ratio of present values (PV) between indirect selection and selection at year 8.

where symbols were defined as same as (4). Notice that s was canceled out in the final expression of the equation.

Selection Efficiency Assessment

It would be ideal to use volume at a year that is as close to the rotation age (25) as possible for the response trait Y [Equation (3)]; however, such data are not available for these tests. For most test series used in this study, measurement was limited to the first 8 yr, about one-third of the rotation age. Generalized prediction models (Lambeth 1980, McKeand 1988, Magnussen 1989) may be used to estimate age-age genetic correlations up to rotation age based on juvenile information. However, such extreme extrapolations would result in large errors associated with derived figures (Woods 1995) and additional assumptions about the heritabilities at the rotation age are also needed.

Based on previous studies on loblolly pine (Balocchi et al. 1993) and various juvenile-mature correlation studies (Lambeth 1980), optimal selection ages were around age 6 for loblolly pine and other commercial trees with 30 yr economic rotation, indicating that the true optimal selection ages could be well within the range of age (age 1 through 8) in the current study. Hence, if the defined selection efficiencies (SE^sub GPY^ and SE^sub PV^) provide a fair comparison among first 8 yr, optimal selection ages for various selection methods can be determined with the reference to genetic gains at the rotation age.

To examine whether the use of selection on 8 yr volume instead of the rotation volume in the definition of selection efficiency could affect the results of this study, the following derivation was used to assess the accuracy of using earlier age than rotation for response trait. The same approach can be applied to other similar studies (Woods 1995, Matheson et al. 1994) as well.

The core component in SE^sub c^ formulae (c = GPY or PV), regarding the genetic parameters, is

Suppose that the ideal way for evaluating early selection is that we instead use rotation age m for the response trait in SE, and then compare SE at any age j (j

is a function of regression coefficient b in (8) and age j. Since r^sub jmiddot;8^ usually increases in j, this correlation ratio is also an increasing function in j. In this study, age-age genetic correlations were used to fit the LAR model [Equation (8)]. The estimates of slope b and genetic correlation were then used to calculate correlation ratio [Equation (10)] in order to assess the selection efficiency criteria used in this study.


Genetic Gains

Genetic gain estimates, as deviations from the population mean, ranged from 1.7 to 16.1 dm^sup 3^ and varied for different regions, traits and different selection methods (Figure 1). For each selection method in each region, selection on volume at age 6 yielded greater gain in year-8 volume than selection on height. Among geographical test regions, the Coastal population had the greatest correlated response in year 8 volume to selection, followed by the Piedmont and Northern populations.

Regardless of traits and test regions, similar patterns were observed when comparing different selection methods within a test region for each trait (Figure 1). Gain from within family selection alone was less than individual selection whether clonal deployment was used or not. Among the three family selection method, full-sib selection based on total genetic component had the greatest genetic gain, followed by mid- parent full-sib selection (narrow sense) and half-sib family selection. Selection among full-sib families had about 40% genetic gain over selection on half-sib families.

Comparison between individual or within-family selection and family selection was misleading, because selection intensities were not the same in this study. However with such different selection intensities, which favor individual selection (1% for individual and within family selection and 10% for family selection), the magnitude of genetic gain from family selection was still equivalent to that of individual selection. Both were greater than within-family selection.

For all selection methods, additional gain can be achieved by capturing the nonadditive genetic component through mass-producing full-sib crosses (full-sib selection) and vegetative propagation (FS+WFS_G). For individual selections, mass selection for clonal deployment increased gain 12-22% over mass selection. Selection of best full-sib families, thus capturing SCA, improved 10%-33% over selection of middle parent values. For family plus within family selection, this additional gain could be more than 40% (for the Northern and Coastal populations).

Selection Efficiency

The selection efficiency of various selection methods were plotted against age for both height and volume selection across three test regions, selectively illustrated in Figures 2-5. These graphs indicated that compared with direct selection on year-8 volume, indirect selections at certain or most early ages were more efficient (i.e., SE^sub C^ > 1) for most selection methods, except for individual selection for height. The results of optimal ages were listed in Tables 2 and 3.


Results of selection efficiencies based on trait height for individual selection, family selection, and combined family plus within family selection were shown in Figures 2 and 3. Selection efficiency was calculated for selection based on both additive genetic variance and total genetic variance when appropriate.

Selection efficiency varied over time for different selection methods and test regions, with a wide range, from 0.7 to 2. For all selection methods on height, the basic shape of selection efficiency over time had the same pattern with a unique mode somewhere between two extreme selection ages. Different test regions revealed different magnitudes of selection efficiency due to different time trends of genetic parameters (Figure 2).

In the Northern region, optimal age based on SE^sub GPY^ ranged from yr 3 to 5 for different selection methods. If SE^sub PV^ was used as the criterion, the same optimal age (yr 3) was applicable to family selections, but the optimal age for SE^sub PV^ delayed about 1 yr for individual selection and combined selection.

The selection efficiency (both SE^sub GPY^ and SE^sub PV^) was higher in the Coastal and Piedmont regions than in the Northern region (Figure 2). In addition, the peak value of SE appeared earlier for all selection methods in the Piedmont region and for family selections in the Coastal region (Figure 3). As a result, the optimal age for selection on height could be very early. If gain per year was used as the criterion as in SE^sub GPY^, the optimal age was as early as age 2 or 3 for family selections in these two regions. For other selection methods, optimal age varied from year 3 to 5. If present value was instead used as selection criterion (SE^sub PV^), age 3 was the most efficient for all selection methods in the Piedmont region except half-sib family selection. Although in the Coastal region the optimal age for individual selection was delayed to age 5, age 3 was only slightly less efficient in SE^sub PV^. Overall, age 3 was also a good choice for any selection method in the Coastal and Piedmont regions. For both SE^sub GPY^ and SE^sub PV^, selection methods based on total genetic component had significant improvement in efficiency than those based solely on additive genetic component.


Similar to height, selection efficiencies of volume based on individual, family, within-family, and combined family plus within-family selections were calculated based on both additive genetic variance and total genetic variance when appropriate (Figures 4 and 5). The range of SE varied from 0.7 to 2, with a majority of SEs being close to or exceeding one. Shape and magnitude of SE^sub GPY^ and SE^sub PV^ varied from one test region to another (Figure 4). Although age 4 was the first evaluation age for volume, the change was quite evident for some SE curves for such a short period. For some selection methods, although the highest selection efficiency was achieved at the first measurement year, i.e., age 4, the true optimal age could potentially be even earlier if earlier measurement data are available.

In the Northern region, the optimal ages for different selection methods ranged from yr 4 to 7 based on GPYselection efficiency (SE^sub GPY^). For individual selection and within-family selection, age 5 had the lower selection efficiency with increasing trends spreading towards both directions. Due to the irregular shape of the SE curve, the optimal ages for SE^sub PV^ lagged behind about 2 yr for individual selection and within-family selection based on total genetic variance. For selection methods based on additive variance, optimal age was age 4 or 5. However, even if age 5 or 6 was optimal, age 4 was only slightly less efficient and should still be a good age for selection. Both SE^sub GPY^ and SE^sub PV^ were larger than unity optimal ages, indicating that indirect selection was more efficient than direct selection.

Within the measurement duration, age 4 was the optimal age for all selection methods in the Coastal region regardless of selection criteria (Figure 5). After age 4, selection efficiency declined almost linearly. This pattern was mainly because the age-age correlation for volume was very high in this test region, and time discount function in the SE formula largely determined the shape of the curve. Based on time trend of selection efficiency, optimal age for selection on volume could potentially be at an earlier age than yr 4, but this was the earliest measurement year in this study. In addition, the peak value of SE was very high, indicating that early selection was about 40%-90% more efficient than selection at age 8.

Selection efficiency for the Piedmont region exhibited a different pattern (Figure 4), with optimal ages depending on selection methods. Like the Northern and Coastal regions, family selections had earlier optimal ages, at age 4 or 5. Individual and within-family selection revealed different patterns for selection based on additive genetic variance and total genetic variance. SE based on total genetic variance had a lower point at age 5, while SE based on additive genetic variance peaked at age 6.

Selection Criteria: GPY vs. PV

Both gain per year and present value have been used in selection efficiency studies. Both criteria consider time as a discount factor. In SE^sub GPY^, gain-per-year (GPY) is to maximize the genetic gain per unit of time in a breeding cycle (Li et al. 1996, Matheson et al. 1994). Present value (PV) in SE^sub PV^ measures investment return. It emphasizes the reduction of value in the course of time by introducing accumulated interest rate power function into the denominator (Balocchi et al. 1994, McKeand and Bridgwater 1986). As an example for illustration, the comparison of selection efficiency for both criteria for height and volume in the Northern region is shown in Figure 6.

For both selection criteria, one parameter needs to be subjectively determined. In case of GPY, additional time (t) for breeding cycle to complete may vary for various reasons such as breeding strategies and techniques utilized in a certain breeding program. With longer breeding cycles, optimal selection ages remained the same or were delayed 1 to 3 yr (Tables 2 and 3). For SE^sub PV^, additional time (s) did not affect selection efficiency since it was canceled out in the ratio formula for SE^sub PV^. But interest rate (I) can vary from time to time. With increased interest rates, optimal ages were pushed 1 or 2 yr earlier in most cases. Results of optimal ages are summarized in Table 2 and 3.

Selection Efficiency Assessment

Age-to-age correlations up to age 8 for trait height and volume were used to fit the LAR model. The coefficients were reported in Table 4. Age 1 was eliminated from regression analysis, since its correlation with other ages had a different trend, which was previously revealed by Lambeth (1980). For trait height, regression intercept a was very close to 1, while the slope ranged from 0.09 to 0.3 over three test regions, in agreement with previous studies (McKeand 1988, Lambeth 1980).

When the ratio of age correlation r^sub j[middot]25^/r^sub j[middot]8 was calculated from Equation (10) and plotted against age, the difference from age 8 was minimum for volume after age 4 for b within the range of 0.1[similar]0.3 (Figure 7A). The differences among ratios were also small beyond age 2 for height (Figure 7B), mainly because the age correlations between height and age 8 volume were less variant than those of the same growth trait.


Comparison among genetic gains from various selection methods at current selection age, yr 6, revealed that selection on volume yielded more gain than selection on height, primarily because of the higher genetic correlation of early volume with age 8 volume and comparable heritability when compared with height (Xiang 2001 ). This confirms the earlier results that volume should be considered in early selection of loblolly pine in order to achieve the higher genetic gain (Li et al. 1996).

Among test regions, the Coastal population had the greatest correlated response in year-8 volume to selection, followed by Piedmont population and Northern population. Such differences in gain prediction are expected because of their different growth rates. The Atlantic Coastal source is noted for its faster growth than the other two sources. The Northern population is cold-hardy but grows relatively slowly. The Piedmont source is also relatively cold-hardy but is intermediate in growth rate. These regional differences should be considered regarding the optimal age and selection efficiency when selections are made in each geographic region.

Selection among full-sib families had about 40% more genetic gain over selection on half-sib families. Family plus within-family selection based on total genetic component can capture the most genetic gain. In the combined selection, family selection contributed more to the total genetic gain than within-family selection. These results are similar to those reported for height selection in an unimproved population of loblolly pine (Balocchi et al. 1993) and second-generation of loblolly pine breeding program (Li et al. 1999).

The nonadditive genetic component was found to be very important and even to exceed the additive component at some point in an early growth stage of loblolly pine for growth trait (Balocchi et al. 1993, McKeand and Bridgwater 1986, Foster and Bridgwater 1986). In this study of early diallel tests, dominance accounted for 20%-40% of the total genetic variance. Though the nonadditive component is not as high as that in the unimproved population, it is still important for tree breeders to consider and to capture this additional genetic gain through mass-controlled pollination or vegetative propagation techniques such as rooted cutting (Li et al. 1999). As shown in this study, for all selection methods, additional and significant genetic gain can be achieved by capturing partial nonadditive genetic component through mass production of the best full-sib families, and all of nonadditive component through clonal deployment, or through vegetative propagation of the best individual trees in the best full-sib family. In he example of selection at age 6, mass selection with clonal deployment had 22% more gain than simple mass selection. Selection of best full-sib families can improve gain up to 33% over selection of mid-parent values. Selection of the best individuals within the top full-sib family through vegetative propagation was the best selection strategy, achieving as much as 40% more gain than selection through the current breeding cycle (for Northern and Coastal populations).

Selection efficiency is a comprehensive statistic that combines the information of both genetic parameters and discount factor of time. These genetic parameters are genetic correlation r^sub j8^ and heritability of both indirect and direct selection ages (h^sup 2^, and h^sub 2^g). The time trend of any of these three genetic parameters will affect the result of selection efficiency. Since heritability is usually of the same magnitude, genetic correlation has more influence on selection efficiency.

If selection is based on tree height for the Northern population, the optimal selection age ranged from age 3 to age 5 based on SE^sub GPY^ criterion for all selection methods, and indirect selection earlier was more efficient than selection on volume at age 8. Similar results were observed for selection efficiencies based on SE^sub PV^ criterion. SE^sub PV^ also indicated that based on present value, indirect selection on height would be nearly as efficient as direct selection on volume.

The higher selection efficiency on height for both SE^sub GPY^ and SE^sub PV^ in the Coastal and Piedmont regions compared with that in the Northern region is mainly due to he higher genetic correlation between height and age 8 volume in these two regions (Xiang 2001). Early optimal ages for selection on height (age 3-5) imply that selection on height at very early growth stage has the potential to increase both gain per year and present value.

Selection efficiency is also sensitive to the remaining time to complete the breeding cycle t or interest rate I. A shortened breeding cycle with t = 5 pushed the optimal selection ages earlier (Table 2, 3). With the top grafting and other flower stimulating techniques, t can be reduced as short as 5 yr for the current breeding program. Thus, more aggressive selection strategies with earlier selection ages could be rewarded with even higher efficiencies and may become preferred in the future.

Results from the selection efficiency study of Balocchi et al. (1994) showed that optimal family selection age based on total genetic component was 2-3 yr earlier than family selection based solely on additive component. In contrast, we found that selecting an additional nonadditive component does not have a significant effect on optimal age for selection on trait height. There are very few differences in optimal ages between these two strategies for selection on height (Table 3). This is also true for volume selection in the Northern and Coastal regions. For selection on volume in the Piedmont region, however, capturing a nonadditive genetic component does have the advantage of earlier optimal selection ages for family and combined family plus within-family selection besides higher genetic gain.

Due to the unavailability of genetic information from rotation age, yr 8 (one-third of rotation age) was used for the response trait in indirect selections. Based on comparison with the rotation age gain formula via Equation (10), the selection efficiency criteria used in this study turned out to be appropriate for most early ages except for age 1 and 2. Thus all results based on present value criteria in Table 2 and 3 remain unaffected, and optimal ages of age 3 or later remain the same for gain per year criterion. However, caution must be taken for selection methods for which age 1 or 2 was the optimal age. In practice, first year should always be avoided for selection because of planting shock and other adaptive environmental effects in the field. For family and combined selection, age 2 may be attractive in terms of very strong genetic correlation with later ages in growth trait and high family-mean heritability. However, its advantage over age 3 would largely disappear if gain were corrected by the correlation ratio Equation (10), as a noticeable drop in the correlation ratio was observed when selection age was moved from age 3 to age 2 [Figure (7B)]. Hence, 1 yr delay should be a conservative but yet necessary tactic for optimal ages of 2 or earlier.

The most important implications from this study are: (1) volume would be more effective than height in early selection of loblolly pine in order to achieve maximum genetic gain; (2) selection at earlier ages (e.g., 3 or 4 for height) would achieve the maximum gain per time unit in the loblolly pine breeding program; (3) because of regional differences regarding the optimal age and selection efficiency level, separate considerations should be applied to each region; (4) reduction in breeding cycle (i.e., reducing t) could further reduce early selection age and increase selection efficiency in terms of gain per time unit.

Literature Cited

BALOCCHI, C.E., F.E. BRIDGWATER, AND R. BRYANT. 1994. Selection efficiency in a nonselected population of loblolly pine. For. Sci. 40:452-473.

BALOCCHI, C.E., F.E. BRIDGWATER, AND R. BRYANT. 1993. Age trends in genetic parameters for tree height in a nonselected population of loblolly pine. For. Sci. 39:231-251.

BASTIEN, J.C.H., AND B. ROMAN-AMAT. 1990. Predicting Douglas-fir (Pseudotsuga menziesii Mirb. Franco) volume at age 15 with early traits. Silvae Genet. 39:29-34.

COTTERILL, P.P., AND C. A. DEAN. 1988. Changes in the genetic control of growth of radiata pine to 16-years and efficiencies of early selection. Silvae Genet. 37:138-146.

DANJON, F. 1994. Heritabilities and genetic correlations for estimated growth curve parameters in maritime pine. Theor. Appl. Genet. 89:911-921.

FLANKLIN, E. 1979. Model relating levels of genetic variance to stand development of four North American conifers. Silvae Genet. 28(5-6):207-212.

FOSTER, G.S. 1986. Trends in genetic parameters with stand development and their influence on early selection for volume growth in loblolly pine. For. Sci. 32: 4.

GOEBEL, N.B., AND J.R. WANER. 1966. Volume tables for small diameter loblolly pine, shortleaf and Virginia pine in the upper South Carolina piedmont. For. Res. No. 7, Clemson University, Clemson, SC.

HODGE, G.R., AND T.L. WHITE. 1992. Genetic parameter estimates for growth traits at different ages in slash pine and some implications for breeding. Silv. Genet. 41(4-5):252-262.

HUBER, D.A. 1993. Optimal mating designs and optimal techniques for analysis of quantitative traits in forest genetics. Ph. D. Diss., Univ. of Florida, Gainesville. 151 p.

LAMBETH, C.C. 1980. Juvenile-mature correlations in Pinaceae and implications for early selection. For. Sci. 26:571-580.

LAMBETH, C.C., J.P. VAN BUIJTENEN, S.D. DUKE, AND R.B. McCULLOUCH. 1983. Early selection is effective in 20-year-old genetic tests of loblolly pine. Silv. Genet. 32:210-215.

LI, B., S.E. MCKEAND, AND H.L. ALLEN. 1991. Seedling shoot growth of loblolly pine families under two nitrogen levels as related to 12-year height. Can. J. For. Res. 21:842-847.

LI, B., S.E. MCKEAND, AND R.J. WEIR. 1996. Genetic parameter estimates and selection efficiency for the loblolly pine breeding in the south-eastern US. P. 164-68 in Tree improvement for sustainable tropical forestry. Proc. of QFRI-IUFRO Conf., Caloundra, Queensland, Australia.

LI, B., S.E. MCKEAND, AND R.J. WEIR. 1999. Tree improvement and sustainable forestry-impact of two cycles of loblolly pine breeding in the U.S.A. For. Genet. 6:229-234.

LI, B., C.G. WILLIAMS, W.C. CARLSON, C.A. HARRINGTON, AND C.C. LAMBETH. 1992. Gain efficiency in short-term testing: Experimental results. Can. J. For. Res. 22:290-297.

LYNCH, M., AND B. WALSH. 1998. Genetics and analysis of quantitative traits. Sinauer Associates, Inc. Sunderland, MA. 980 p.

MCKEAND, S.E. 1988. Optimum age for family selection for growth in genetic tests of loblolly pine. For. Sci. 34:400-411.

MCKEAND, S.E., AND F.E. BRIDGWATER. 1986. When to establish advanced generation seed orchards. Silv. Genet. 35:245-247.

MCKEAND, S.E., AND J.C. SVENSSON. 1997. Sustainable management of genetic resources. J. For. 95(3):4-9.

MCKEAND, S.E., B. LI, AND H.V. AMERSON. 1999. Genetic variation in fusiform rust resistance in loblolly pine across a wide geographic range. Silv. Genet. 48(5):255-260.

MAGNUSSEN, S. 1989. Minimum age-age correlations in early selections. For. Sci. 43:928-938.

MAGNUSSEN, S. 1993. Growth differentiation in white spruce crop tree progenies. Silv. Genet. 42:258-266.

MATHESON, A.C., D.J. SPENCER, AND D. MAGNUSSEN. 1994. Optimum age for selection in Pinus radiata using basal area under bark for age:age correlations. Silv. Genet. 43:352-357.

SAS INSTITUTE INC. 1996. SAS/STAT Software: Changes and enhancements (through release 6.11). Cary, NC. 1094 p.

SEARLE, S.R., G. CASEELA, AND C.E. MCCULLOCH. 1992. Variance components. Wiley, New York. P. 254-255.

SVENSSON, J.C., S.E. MCKEAND, H.L. ALLEN, AND R.G. CAMPBELL. 1999. Genetic variation in height and volume of loblolly pine open pollinated families during canopy closure. Silv. Genet. 48:204-208.

WILLIAMS, G.G., AND R.A. MEGRAW. 1994. Juvenile-mature relationships for wood density in Pinus taeda. Can. J. For. Res. 24:714-722.

WOODS, J.H., D. KOLOTELO, AND A.D. YANCHUK. 1995. Early selection of coastal douglas-fir in a farm-field test environment. Silv. Genet. 44(4): 178-186.

WU, H.X., F.C. YEH, R.P. PHARIS, D. DHIR, AND B.P. DANCIK. 2000. Study of early selection in tree breeding: 3. A case study using early information to enhance selection efficiency in late trait in lodgepole pine (Pinus conforta spp. Latifolia). Silv. Genet. 49(3): 152-158.

XIANG, B. 2001. Genetic analysis of diallel tests of loblolly pine (Pinus taeda L.). Ph.D. Diss., North Carolina State University, Raleigh, NC. 143 p.

XIANG, B., AND B. LI. 2001. A new mixed analytical method for genetic analysis of diallel data. Can. J. For. Res. 31:2252-2259.

ZOBEL, B.J., AND J.T. TALBERT. Applied forest tree improvement. Wiley, New York. 1984. P. 232-234.

Bin Xiang; Bailian Li; Steve McKeand

The authors are former Graduate Research Assistant, Associate Professor, and Professor, respectively, Department of Forestry, Box 8002, North Carolina State University, Raleigh, NC 27695-8002. Bailian Li may be contacted at (919)515-6845; Fax: (919)515-3169; E-mail:

Acknowledgments: This research was supported by members of the North Carolina State University-Industry Cooperative Tree Improvement Program and the Department of Forestry at the North Carolina State University. The authors thank J.B. Jett and Robert Weir, and members of the Tree Improvement Program, for their inputs in design, establishment, and measurement of these progeny tests. Appreciation to Gary Hodge, Eugene Eisen, and Roger Berger for their inputs and helpful discussions at various stages of this work.

Manuscript received October 12, 2001, accepted April 22, 2002.

Copyright Society of American Foresters Apr 2003

Provided by ProQuest Information and Learning Company. All rights Reserved