Estimation of genetic parameters for growth in a population of Eucalyptus grandis × Eucalyptus nitens hybrids in KwaZulu-Natal and Mpumalanga

Hybridising Eucalyptus grandis with Eucalyptus nitens produces progeny intended to combine the favourable traits of both parents. E. grandis exhibits good growth, stem form, wood properties and rooting ability for vegetative propagation. E. nitens displays superior cold tolerance, including superior frost and snow tolerance, and high wood density. Eucalypt hybrids developed specifically for improved cold tolerance tend to produce fewer viable hybrid progeny and are more recalcitrant rooters than their sub-tropical counterparts. To overcome these challenges, a novel testing strategy was developed and used to identify promising varieties for commercial deployment. As a result, a large population of E. grandis × E. nitens hybrids was developed through controlled pollinations, propagated via mini-cuttings and established in a total of 21 field trials, across the KwaZulu-Natal and Mpumalanga provinces of South Africa, to investigate growth and adaptability. Data of three growth traits, namely diameter at breast height (DBH), total tree height and volume were analysed to determine the genetic parameters of these traits. Combined-site broad sense heritability for DBH, height and volume were 0.39, 0.46 and 0.35 respectively, indicating a moderate level of genetic control. This, in conjunction with large genetic coefficients of variation (CVg = 20%, 16% and 41% for DBH, height and volume respectively) shows that substantial gains can be made through selection. Combined-site Type B genetic correlations for DBH, height and volume were 0.71, 0.68 and 0.65 respectively, indicating that there was notable genotype by environment interaction. Pairwise comparisons of rBg for volume showed large variation in the correlations of clonal rankings between trials, with rBg ranging from 0.0 to 0.90.


Introduction
Southern Forests is co-published by NISC (Pty) Ltd and Informa UK Limited (trading as Taylor & Francis Group) High-yielding plantation forestry is an important source of renewable resources (Jaradat 2010).A key component in increasing and sustaining the yield of forestry is tree breeding (Gladstone and Ledig 1990).Tree breeding requires the development, identification and deployment of germplasm with desirable traits.To determine whether genetic improvement is feasible in a breeding population, calculation of genetic parameters is required.
Quantitative genetic parameters are calculated from measurements of genetic material tested in field trials.Genetic parameters such as levels of genetic variation and heritability enable breeders to determine if the selection of superior genetics would lead to improved plantation yields.Type B genetic correlation describes levels of genotype by environment interaction.This informs breeders whether selections will perform well across different environments or whether specific selections need to be made for specific environments.Correlations between traits assist in determining selection criteria and their potential relative importance.Selection is possible at a parent, family or individual level, properties and rooting ability of E. grandis, a sub-tropical to warm temperate species, with the cold tolerance of E. nitens, a temperate species (Thompson 2013).
Breeding temperate eucalypt hybrid combinations offers numerous challenges when compared to their sub-tropical counterparts, namely poor flowering, small flowers, low pollen yields, low seed yields (Horsley 2009 et al.), poor coppicing ability and reduced rooting ability (Murugan 2007).Additionally, in South Africa, temperate hybrid clones need to be adaptable to a large and geographically diverse area, requiring clonal testing on numerous sites.
In order to overcome these challenges a novel testing strategy was developed.
The strategy aimed to circumvent the issues that contributed to low seed yield through by-passing the seedling progeny trial phase.This was achieved by using seedling ortets as hedge mother plants for clonal trials, which also negated the fact that temperate species often do not coppice well and their field coppice does not root well.
The challenge of limited ramet availability for diverse site testing was solved by establishing trials containing fewer ramets, in unreplicated trials on multiple sites, with site serving as replication.The drawback of this strategy is that microsite effects could therefore not be detected or adjusted for.In an attempt to mitigate the effects of microsite, each trial was randomised independently, sites were selected based on homogeneity and an emphasis was placed on the establishment of a large trial set, in the hope that sheer repetition would help dilute individual cases of microsite effect.Additionally, rolling controls were established between blocks running parallel to the trial block structure.These would serve as visual reference and could be measured to determine microsite effects at a block level.
We anticipate that this strategy will enable the determination of quantitative genetic parameters for a population of temperate eucalypt hybrid clones, albeit with lower precision than the traditional testing strategy, and enable the selection of superior clones for further testing in larger replicated trials.This article serves to detail this strategy and assess its effectiveness at deploying hard-toproduce novel hybrid clones over a diverse geographic area, its efficacy in screening these varieties for further testing and compare the genetic parameters calculated to those of a simulated traditionally screened population.

Genetic material
To identify high potential eucalypt clones for commercial deployment in the cool temperate region of South Africa's summer rainfall region, development and testing of E. grandis × E. nitens hybrids commenced in 2002 (Louw 2011).Early mating designs focused on crossing E. grandis females, backward selected for superior growth from third-generation progeny trials, with E. nitens males, from forward selected second-generation E. nitens orchards (Figure 1).Where possible, parents with high General Combining Ability (GCA) for growth and low levels of relatedness were selected from advanced orchards, but eventually, particularly in the case of E. nitens, availability of flowers became a significant criterion in parental selection.Initial mating designs were 5 × 5 full factorials, but over the period of a decade, mating designs varied.Crosses which did not yield seed were often repeated and many new parents were added, often in incomplete partially connected factorial designs.The emphasis of the crosses was to produce a sufficiently large population for clonal testing, as opposed to a carefully constructed, well-balanced mating design to give detailed family or parental breeding values.
From 2003 to 2019, a total of 12 519 flowers were pollinated, using around 55 E. nitens pollen parents and around 90 E. grandis females.In all, 245 different crosses were made with an average of 51.4 buds pollinated per cross.A total of 223 crosses were made only once, with 22 being repeated.
E. nitens pollen was collected annually from several orchards located in cold areas throughout KwaZulu-Natal and Mpumalanga (Figure 1), from May to July, dried and freezer stored until the following season.This pollen was used to pollinate selected females annually from January until March at Sappi's Shaw Research Centre E. grandis clonal breeding orchard, located at 29°28′44.56″S, 30°10′48.32″E at 1 103 m above sea level (Figure 1).
Initially various controlled pollination (CP) techniques were trialled with varying levels of success, but eventually the One-Stop Pollination technique (OSP) became the standard due to it yielding the highest levels of hybrid seedlings (Horsley et al. 2009).
At six months after sowing, each seedling was visually assessed for runtiness and assigned a score for hybridity based on diagnostic morphological features (Table 1), including stem characteristics, leaf arrangement, leaf shape, and size, colour, petiole and apical bud leaf pair angle (Aimers-Halliday et al. 1999;Naidu 2011).
Initial challenges with low hybrid seed yields from controlled crosses, coupled with poor rooting from field coppice, lead to the decision to modify the traditional testing strategy by skipping the seedling testing phase of progeny trials, in which an initial screening of the population and an understanding of family performance is usually achieved.Instead, hybrid progeny from controlled crosses were established as hedges in the nursery to produce cuttings for establishment in clonal trials (CTs).

Propagation
Healthy seedlings, with a hybrid score of three or lower, were then selected for ortet hedge establishment to produce ramets for clonal trials and assigned a unique clonal ID.These seedlings were planted into larger pots and housed in a single greenhouse tunnel where temperature, humidity and fertigation could be controlled.Once the ortet hedges were well-shaped and growing vigorously (roughly 18 months after sowing), cuttings were taken from August to February to supply ramets for clonal trial establishment.Cuttings were set in Unigro ® 60 ml, 128-cavity trays containing a mix of coir and perlite and placed in a rooting tunnel with underfloor heating and controlled misting.Rooting success, per clonal ID, was assessed after six weeks then the trays were moved from the rooting tunnel to a growing area to grow out and be gradually hardened off.Final survival assessments were then performed to enable the design of field trials.

Trial establishment
Since 2011, a total of 21 CTs were established over five years across various site types across the KwaZulu-Natal and Mpumalanga provinces of South Africa.Due to rooting constraints, the trials were unbalanced, and contained whichever clones were available for establishment at the time of planting (Table 2).Once a clone had been established on several sites, it was taken out of production to free up propagation capacity for new clones.As a result, the various trials contained numerous clones at different locations.However, large portions of the population were common across many of the sites and common controls were used to ensure adequate linkage.A total of 1 020 clones were established.Table 3 summarises the number of trials on which a clone was established.For example, 103 clones were only established on one site each and 13 clones (including controls) were established on all of the 21 sites.
The trials consisted of randomised three-ramet row plots, with perpendicular rolling controls of commercial clones to serve as easily visible references and to indicate any spatial biases, such as might occur down a slope due to different moisture levels, soil or frost gradients.Due to limited availability of cuttings, the trials per site contained no replications, but since the same treatments occurred on multiple sites, the sites themselves served as replications.Two rows of surround trees, consisting of pure species seedlings well suited to the site, were established to mitigate edge effects.Site altitudes ranged from 812 m a.s.l. up to 1 657 m a.s.l.Site index, an indication of site growth potential, ranged from 15.1 to 25.9 m of dominant total-tree height at five years of age.Between four and five sites were established annually throughout summer.Higher altitude sites, specifically selected for high snow and/or frost prevalence, were established first, in order to enable the cuttings to be well-hardened before winter.
Cuttings were established at a 2 m × 3 m espacement.
Planting was done by hand into pre-made pits and water or hydrogel was added as dictated by conditions.
Trial measurements iButton temperature data loggers were deployed at establishment to monitor winter temperatures.One-month survival assessments were performed to enable blanking within the recommended five-week period.Survival was assessed after one year and all trials were visited frequently to ensure adequate weeding was maintained.Height (in metres) and DBH (measured in centimetres at a height of 1.3 m) were measured for all trees, using a Vertex® hypsometer and DBH tape respectively.Any stem defects, diseases or abnormalities were also recorded.Trials were measured between two and six years of age (Table 2).Three trials, selected for convenience of location, were remeasured at an age of greater than six years to allow for the assessment of age:age correlation between growth traits and to investigate the effects of early selection.These trials were ECT0076T measured at 3.6 and 7.8 years, ECT0088T measured at 2.6 and 6.3 years and ECT0089T measured at 2.5 and 6.4 years of age.
Individual tree volumes were calculated using Demaerschalk's Function using confidential form factor values (Crous 2006): ( ) (1) Dominant heights were calculated through the use of diameter:height regressions used to predict the corresponding height for the quadratic mean diameter of the 20% thickest trees for each site (Bredenkamp 1993).Dominant heights were used to calculate site indices, to the base age of five years, using growth curve coefficients (Bredenkamp 2000a;2000b) and are listed in Table 2.

Statistical analysis
Height, DBH and individual tree volume data were standardised to a mean of 100 and used to conduct the analysis.The data were standardised to deal with scale differences for phenotypic and genetic variances between different site types and measurement ages (van den Berg et al. 2018).The statistical model used for the analysis was: where y ijk = the k th observation of the for the j th clone at the i th site; µ = overall mean; S i = fixed effect of the i th site; c j = random effect of the j th clone; S i *c j = random clone by site interaction; and e ijk = random within plot error term.All effects, except overall mean and site effect, were assumed to be random and independently distributed.Best Linear Unbiased Predictions (BLUP) is able to predict genetic values by simultaneously adjusting even unbalanced data to the specified fixed effects, such as site or trial effects (Fritsche-Neto et al. 2010).The lme4 package (Bates et al. 2015) in R (R Core Team 2017) was used to estimate variance components and to obtain BLUP of random genetic effects.
Phenotypic variance, heritability and genotypic coefficient of variance were calculated respectively as: Type B genetic correlation measures genotype by environment interaction (G × E) that is due to rank changes as the same trait is expressed differently across different environments.This is expressed as a ratio from zero to one.An r Bg = 1 indicates a perfect correlation between performance in different environments, and therefore no G × E (Burdon 1977).Type B genetic correlations (r Bg ) were estimated as: An understanding of genetic correlations between traits enables breeders to determine how selecting for one trait will impact on another.Genetic correlations between growth traits were determined using the formula: An understanding of growth trait correlations measured at different ages enables breeders to determine whether earlier measurements could be used for selection (Rweyongeza 2016).Age:age correlations were calculated for three trials, namely ECT0076T measured at 3.6 and 7.8 years, ECT0088T measured at 2.6 and 6.3 years and ECT0089T measured at 2.5 and 6.4 years of age using: Where t = first measurement; T = second measurement; and I = growth increment between first and second measurement.
In an attempt to compare this novel testing strategy (of eliminating the hybrid seedling progeny trial step and establishing unreplicated trials on numerous sites), with the traditional testing strategy, a height screening was performed by screening out the bottom 50% of all clones based on their BLUP heights across all 21 trials.A hybrid seedling progeny trial only has one seedling ortet representing each potential future clone in one trial, therefore clonal BLUP heights are not truly comparable to ortet seedling heights, and so the screening is regarded as a poor surrogate.Nevertheless, this was considered the most feasible simulation for comparative purposes.A total of 86 clones suffered 100% mortality at every trial at which they were established.In all, 934 clones had surviving ramets in at least one trial.The bottom 467 clones based on BLUP height were removed from the population to simulate a population selected from hybrid seedling progeny trials.Genetic parameters were calculated for this population in the same way as for the unscreened population.

Seed yields
Seed yields varied significantly over the various years and between the various crosses.On average, 0.8 seeds were yielded per bud pollinated, with only 53 crosses out of the 245 crosses yielding more than one seed per flower pollinated and 15 of the crosses yielding no seed at all.The resultant mating design yield was of such an unbalanced and disconnected nature that no meaningful indication of parental or family performance could be gained.

Growth results
Data on DBH, height and volume from all sites are summarised in Table 4. Combined site overall survival was low at 64.54%.The lowest survival was at ECT0070T, the trial that experienced the lowest recorded temperature of −10 °C.
Overall standard deviation exceeded the overall mean for volume, however, individual trial mean was only ever exceeded by individual trial standard deviation in the case of one trial, ECT0096T (Table 5).This showed the large impact of trial site (largely attributable to measurement age), on growth traits.

Clonal heritability estimates
Broad-sense heritabilities for DBH, height and volume were 0.39, 0.46 and 0.35 respectively (Table 6), indicating that these traits are under a moderate level of genetic control.
Combined-site Type B genetic correlations for DBH, height and volume were 0.71, 0.68 and 0.65 respectively, indicating that there was notable genotype by environment interaction.Pairwise comparisons of r Bg for volume showed large variation in the correlations of clonal rankings between trials, with r Bg ranging from 0.00 to 0.90 (Supplementary material, Table S1).
Standardised volume, as a function of DBH and height, had the highest variance in all forms.Its high genotypic variance in relation to its mean lead to standardised volume possessing the highest genotypic coefficient of variation.Height was the most heritable trait but had the lowest genotypic coefficient of variation.

Genetic correlations between growth traits
The correlations between all growth traits were very high (> 0.9), as is summarised in Table 7, with x as the first growth trait compared to y as the second growth trait.

Age:age correlations for three of the trials
Correlations between first and second measurements of growth traits were high (Table 8).Age:age volume correlations were the highest at 0.99, followed by DBH at 0.86 and were the lowest for height at 0.67.Heritabilities generally decreased with age possibly due to higher mortality at the second measurement.

Clonal BLUP values
BLUP values of random genetic effects were calculated using 'ranef' function in the linear mixed effects (lme4) package (Bates et al. 2015) in R (R Core Team 2017).Since the data were standardised to a mean of 100, clonal values were expressed as a percentage of gain above the population mean.The top 20 clonal values can be found in Table 9.
Volume BLUP estimates revealed that the most promising clone WGN480, with a volume BLUP estimate of 110.09% gain compared to the population mean, was substantially superior to the closest clonal hybrid control of WIGN1026, with a volume BLUP estimate of 51.97 gain compared to the population mean.Volume BLUP estimates of the parental species were 12.63 and 7.23 for E. grandis and E. nitens respectively.

Screened population
Broad-sense heritabilities in the screened population for DBH, height and volume were 0.10, 0.07 and 0.12 respectively (Table 10), indicating low levels of genetic control.This is a very large reduction in values compared to the unscreened population.Combined-site Type B genetic correlations for the screened population for DBH, height and volume were 0.34, 0.21 and 0.36 respectively, indicating very high levels of genotype by environment interaction.

Discussion and conclusion
Since all growth data were standardised to a mean of 100, BLUP values can be interpreted as a percentage gain above the population mean, however, the clonal BLUP values derived in this study should only serve to rank and select top performing clones and should not be seen as an indication of plantation gains.For meaningful plantation gain predictions, we recommend testing top performing clones in larger plots in the next phase of clonal testing, referred to as Clone by Site Interaction (CSI) trials.These typically involve the establishment of large blocks (6 × 6 tree plots) of the top performing clones, where they can be compared to current hybrid clonal and pure species seedling benchmarks.The three-tree row plot trial structure used in this study is useful for screening large populations but it exacerbates competition effects and may overemphasise differences and select for 'fast-starters' whereas a larger block plot structure, which has the added advantage of accurately incorporating survival due to larger numbers of ramets, is more effective at simulating a monocrop environment (Stanger et al. 2012).These larger plots could also serve as an important source of coppice material for vegetative propagation when selected clones are commercialised.
Additionally, gain comparisons to the hybrid seedling population are not informative as far as expected plantation yields are concerned, since the hybrid population mean is poorer than that of the pure species alternatives.This is in common with other studies where the population mean of the hybrid is lower than the population mean of either parental species, even though outstanding individual hybrid progeny substantially outperformed outstanding individual progeny from either parental species (Assis 2000).The low mean of the hybrid population may be attributed to negative heterosis which may have been amplified by the fact that no screening took place due to the exclusion of the hybrid seedling progeny trial phase.
However, commercially available clonal hybrid controls and pure species seedling controls were included in this testing strategy to give early indications of the relative performance of new promising clones, many of which substantially A moderate to low r Bg (~0.7), indicates that G × E was present in the tested population across the various trials.This was further highlighted by the large variation in pairwise r Bg comparisons showing that clonal rankings across certain trials changed significantly.Further investigation to find the environmental source of this G × E is warranted.This is usually achieved in CSI trials, where a reduction in genotypic variance is achieved by testing only the top clones and therefore G × E becomes more apparent.
This study found that a simulated pre-screening of poor performing clones, as is operationally achieved through hybrid progeny trials in other breeding strategies, resulted in a reduction in genotypic variation.This in turn lead to an increase in the relative importance of G × E and a reduction in heritability.
Broad-sense heritabilities for DBH, height and volume (0.39, 0.46 and 0.35 respectively) show that these traits are under moderate levels of genetic control.These values are consistent with those reported in eucalypt progeny trials (0.13 < h 2 < 0.41) (Hamilton and Potts 2007;Volker et al. 2008;Vargas-Reeve et al. 2013;Chen et al. 2018;Sumardi et al. 2016).Although much of the literature reports narrow sense heritability, which will be lower than broad-sense heritability due to the exclusion of non-additive genetic effects, these values are within the broad-sense heritability range found in E. grandis (Snedden et al. 2010).Using pedigree to calculate non-additive genetic variance has several limitations: potential pedigree errors, confounding due to species, provenance or environmental effects and an inability to account for linkage disequilibrium, inbreeding and genetic similarity.Several studies have found that non-additive genetic variance contributes significantly more than additive genetic variance towards growth in eucalypt hybrids (Bouvet et al. 2009;Tan et al. 2018).Since the deployment strategy in this scenario captures both additive and non-additive effects, broad sense heritability is of greater interest.
Genetic correlations between growth traits are usually very high (Harrand et al. 2009;Kien et al. 2009;Sumardi et al. 2016).The same was found in this study with genetic correlations of greater than 0.9 between DBH, height and volume.This indicates that selecting for any of these traits would lead to gains in the rest.The fact that growth traits correlate so highly, suggests that only one of the traits needs to be selected for and begs the question of whether cost savings could be achieved through a less labour-intensive measurement trait, such as using DBH instead of height or volume (which requires both DBH and height measurement).A trade-off between cost saving and a minor reduction in gain due to selecting for a 'secondary' growth trait could be calculated by determining how many top selections would not be selected if volume was not used as the growth selection trait.
Measurement of both DBH and height for this population could contribute to the further development of growth and yield equations for this hybrid.
Since one of the constraints of this trial series was to test a large population at a low testing intensity using few ramets, the use of basal area or plot volume would be problematic.A few missing ramets would result in excessive variation in treatment survival which would unrealistically affect basal area or plot volume calculations.Using the mean tree value of a growth trait, as opposed to plot volume or plot basal area, excludes survival, which is an important aspect in operational yields.However, BLUP would draw the estimates of clones with poorer survival towards the mean.Basal area calculated from CSI trial measurements, containing more ramets of each clone, may provide a more accurate and cost-effective indication of clonal performance.
A multiple-step testing strategy is time consuming, which has a negative impact on selection intensity and the ability to rapidly respond to changing biotic and abiotic threats and market conditions.Strong age:age correlations for volume (r tT ~ 0.99) and DBH (r tT ~ 0.87) (Table 8) indicate that first measurements are acceptable for selecting top clones from CTs.
Although the main objective of this trial series was to select clones for further testing and commercial deployment, the potential of advancing the hybrid population to second generation (F2) may seem of significant interest, considering the gains of the top individuals over the population mean.However, according to the breeder's pedigree, many of the top individuals come from only a few families.Most of the top individuals come from only one family.This appears to indicate that specific combining ability (hence non-additive 0.86 0.67 0.99 genetic effects) may account for much of the reported gains.This would be in agreement with the results of several studies which reported that non-additive genetic variance contributes significantly more than additive genetic variance towards growth in eucalypts (Bouvet et al. 2009;Tan et al. 2018).Non-additive genetic effects are not randomly transferable to the next generation.This, in addition to the high levels of relatedness of top individuals, casts doubt on the benefits of advancing the hybrid population.Incorporating other species with additional desirable traits to form three-way hybrids may be of greater interest.This testing strategy was developed to enable the testing of a large population of difficult to produce hybrids across a wide variety of sites.It has shown that this strategy is useful for efficiently testing a large population over a broad and diverse geographic area, given the constraints of limited pollen availability, low levels of parental compatibility and poor rooting from coppice material.However, the drawback of this strategy is clearly that it facilitated the establishment of inferior genotypes and tested them with comparatively poor precision.It appears to mask the importance of G × E within the CT testing stage.Delineation of environmental zones for CT testing may be necessary to ensure that clones suited to specific environments are selected and brought forward to CSI trials.
heritability; and CV g = Coefficient of genetic variance.

Figure 1: Map showing locations of parental Eucalyptus grandis and Eucalyptus nitens orchardsTable 1 :
Diagnostic morphological features used to assign hybridity scores for putative Eucalyptus grandis × Eucalyptus nitens hybrid seedlings

Table 2 :
Site and trial information of 21 Eucalyptus clonal trials in Mpumalanga and KwaZulu-Natal

Table 3 :
Trial representation of clones

Table 4 :
Descriptive statistics for combined site analysis (dbh, height and volume) for 21 Eucalyptus trials established in Mpumalanga and KwaZulu-Natal

Table 5 :
Descriptive statistics per site (DBH, height and volume) for 21 Eucalyptus trials established in Mpumalanga and KwaZulu-Natal

Table 6 :
Genetic parameters of standardised growth traits from 21 Eucalyptus trials established in Mpumalanga and KwaZulu-Natal

Table 7 :
Genetic correlations between dbh, height and volume for 21 Eucalyptus trials established in Mpumalanga and KwaZulu-Natal

Table 8 :
Age:age correlations for standardised growth traits from three Eucalyptus trials established in KwaZulu-Natal

Table 9 :
Treatment gains for the top 20 clones expressed as a percentage over the population mean

Table 10 :
Genetic parameters of standardised growth traits from the top 50% of the population screened for height from 21 Eucalyptus trials established in Mpumalanga and KwaZulu-Natal