Flowering phenology in a Eucalyptus loxophleba seed orchard, heritability and genetic correlation with biomass production and cineole: breeding strategy implications

Reproductive synchronicity within a seed orchard facilitates gene exchange and reduces self-fertilisation. Here we assessed key flowering traits, biomass and foliar 1,8-cineole concentrations of Eucalyptus loxophleba (subsp. lissophloia and gratiae) in an open-pollinated seed orchard. Monthly flowering observations were made on 1142 trees from 60 families and nine provenances across 2 years. The percentage of trees flowering in both years was similar at 87%. There were differences between provenances and families within provenances for flowering traits, biomass and 1,8-cineole and interactions between provenances and year for flowering traits. Heritability of start and end flowering, and 1,8-cineole were high to moderate (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{h}^{2}$$\end{document}h^2 = 0.75–0.45) and duration of flowering, propensity to flower and biomass estimates were moderate to low (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{h}^{2}$$\end{document}h^2 = 0.31–0.10). Genetic and phenotypic correlations between flowering traits were high (rg = 0.96–0.63 and rp = 0.93–0.34) except between duration and end of flowering. The correlations were weaker between flowering traits and biomass or 1,8-cineole. ‘Dual flowering’, when trees underwent two reproductive cycles in a year, was responsible for out-of-phase flowering and those with low biomass and 1,8-cineole concentration should be removed from the breeding programme to hasten selection for desirable traits.

In southwest Western Australia, as elsewhere in Australia, the large scale conversion of native vegetation to agricultural land was followed by extensive salinity problems 1,2 . In the early 1990s, a research program was initiated to identify multi-purpose perennial crops that could mitigate the salinity problems while providing additional economic benefit when integrated with the annual crop based farming systems of the region 3 . The research identified mallee eucalypts as the preferred candidates due to their capacity to coppice after short-cycle harvest [4][5][6][7] . While the initial selection of mallee species focussed on foliar 1,8-cineole concentration (hereafter referred to as cineole), for large-scale bio-renewable feedstock [8][9][10][11] , subsequently, biomass yield emerged as an additional selection criterion as the opportunity for carbon sequestration and bioenergy became prospective during the late 1990s [12][13][14] .
Two species, Eucalyptus loxophleba subsp. Lissophloia L.A.S. Johnson & K.D. Hill, hereafter referred to as E liss , and E. loxophleba subsp. gratiae Brooker, hereafter referred to as E grat were selected for development given their high concentrations of foliar cineole, fast growth rates and prominence in native woodlands of the southwest of Western Australia 15,16 . Successful development of improved tree crops, with multiple desirable traits, presents a challenge with limited understanding of key genetic parameters, such as heritability, genetic and phenotypic correlations. Thus, in 1993 a breeding programme was initiated; and by 2002, 11 E. loxophleba trials had been established using progeny from a total of 78 parents selected in the wild for high cineole content 17,18 . The trials open 1 School of Molecular and Life Science, Curtin University, GPO Box U1987, Perth, WA 6845, Australia. 2 Department of Biodiversity, Conservation and Attractions, Kensington, WA 6151, Australia. * email: beren.spencer-@ postgrad.curtin.edu.au It is expected that knowledge of genetic control and heritability of the various traits considered here, and culling of those which flower outside of the peak flowering period from the seed orchard will speed up production of seeds with desired traits and support commercial viability of mallee as bioenergy tree corps.

Methods
Study site. The study site was located approximately 200 km south east of Perth near Toolibin Lake (32.88°S, 117.62°E). The average rainfall is approximately 400 mm and falls mainly between May and September.
The seed orchard, containing both E liss and E grat was planted in 1999 and the population contains 60 families from nine provenances and three broad regions encompassing the full natural distribution of these taxa (Table 1, Fig. 1). Here a family is defined as the sexually produced open-pollinated progeny from a wild parent tree and a provenance refers to the progeny of wild trees from a known geographic range. The distribution of E liss is much wider than E grat and to reflect this, seven of the nine provenances were E liss (Fig. 1). The 60 families were planted in six-tree row plots in a randomised row-column design with 20 replicates. The orchard was thinned in 2005 from 7200 trees to 1142 based on breeding values for foliar cineole concentration and above ground biomass. After thinning, all 60 families and nine provenances were still represented in the seed orchard with the number of trees retained in each family ranging from 14 to 22. floral assessments. Floral assessments were completed approximately every 4 weeks from May 2012 to January 2013 and from February 2014 to January 2015. To carry out these assessments the canopy of each tree was scanned for presence of reproductive activity from the ground using 8 × 40 binoculars in teams of two to reduce assessor bias. In addition, several times a day, observer teams were calibrated with each other to further reduce bias. flowering criteria. Where a bud density score of one was observed, that reproductive flush was eliminated from the analysis. These flushes were sparse and scores fluctuated widely between assessments. This appeared to be due to loss of reproductive structures through natural abortion or predation processes. These flushes also tended to receive inflated flowering scores because flowers are more prominent than buds on a sparse crop. We defined flowering as a phase when at least two percent of the buds in a reproductive phase were in flower with fresh anthers present. A score of 1% was used to represent few extant flowers to assist the team in the following assessment to indicate that the tree had commenced flowering. Additionally, a tree was defined as flowering when more than 5% of its pre-flowering buds had progressed past anthesis between assessments. This was rarely seen during autumn and winter assessments, but was on occasions in spring and early summer, when flowering proceeded more rapidly.
Biomass assessment. Two biomass assessments were taken for individuals within this seed orchard.
The first assessment was taken pre-thinning in autumn 2004 by measuring the Crown Volume Index (CVI), as described in Spencer et al. 5 , of each of the 5372 healthy trees in the orchard. Briefly, CVI is the measurement of the height and two perpendicular crown widths expressed as m 3 .
The stem basal area of the remaining trees post-thinning (1163 trees) was assessed in February 2014. The stems of each tree were measured with a diameter tape approximately 10 cm above the ground. Loose and fibrous bark was removed and burls and buttressing associated with the lignotubers were avoided. A single diameter estimate was obtained by calculating the Equivalent Diameter near Root Collar (EDRC) as specified in Chojnacky and Milton 44 using the formula: Table 1. Descriptions of regions, provenances, and climate of locations from where the two subspecies of Eucalyptus loxophleba for the seed orchard planting were sourced. The number of families in each provenance and individual trees assessed for flowering traits are also described. Climatic data and elevation which were obtained from SILO dataset 43 from 1985 to 2015. www.nature.com/scientificreports/ Heat sum. Heat sum was calculated by averaging the maximum and minimum daily temperatures above a base temperature below which an organism will not develop 46 . Climatic data was obtained through SILO from interpolated dataset 43 . We used a base temperature of 5 °C which was determined for E. globulus 47 and used in eucalypt flowering studies by Jones et al. 26 . The annual heat sum for the flowering year was calculated from the first day of summer preceding the assessment year.
Statistical analysis. The floral traits (start, end, duration of flowering and the number of reproductive flushes), biomass and foliar cineole of each tree were analysed using a series of mixed linear models in SAS 9.4 48 : where y ijklmn and y ijklm are the floral, biomass and cineole traits, r i is the random replicate effect, c j (r i ) is the random effect column j nested within replicate i , b k (r i ) is the random effect row k nested within replicate i , p l is the provenance l , f m (p l ) is the family m within the provenance l , y n is the year n and e ijklmn and y ijklm are the residual errors of the respective models. Interactions were tested between year and provenance (p l .y n ), as well as between year and family nested within provenance (y n .f m (p l ); however, the proportion of dual flush trees, cineole and biomass analyses excluded the year factor y n from the model (Eq. 2b) because these data were not collected over multiple years. Given that the dataset was unbalanced due to mortality and thinning, replicate was treated as a random effect so that inter-block information could be recovered 49 . Biomass data was transformed using natural logarithms to conform to homogeneity assumptions. The proportion of trees that flowered and proportion of dual flush trees were tested with the same model following arcsine transformation. Tukey-Kramer tests were used to determine the differences in the least square means of key traits. Start of flowering week was standardised to compare the 2012 and 2014 assessments. At the first assessment in 2012, 9% of the trees were flowering, and the 2014 data were truncated to match the 9% flowering from the 2012 assessment.
To estimate heritability, the data were analysed with ASReml Version 4.1 50 using a linear mixed model (Eq. 2a). To estimate univariate heritability in flowering traits from 2012 and 2014, Eq. (2b) was used but family within the provenance was a random effect as required to calculate variance for heritability estimates 51 . The heritability analysis for number of flushes and propensity to flower in each year used a binomial model with a logit link. Genetic correlations were estimated from bivariate analysis of traits. Insignificant random effects were removed if they introduced instability to the model and log-likelihood was used to determine the model of best fit.
Eucalypts have a mixed breeding system in open-pollinated seed orchards and self-fertilisation is common 19 . Based on the outcome of the Sampson and Byrne 38 study of a third closely related subspecies in the loxophleba group, E. loxophleba subsp. supralaevis, it is assumed that E liss and E grat have a mixed mating system. Griffin and Cotterill 20 suggested using a coefficient of relationship of ρ = 1/1.25 to compensate for mixed mating and a selfing percentage of 30% for seed sourced from wild populations of E. regnans. This approach has been assessed and confirmed appropriate for correcting heritability estimates in an open-pollinated eucalypt progeny trial 52 and subsequently applied to a series of E liss progeny trials 18 .
Narrow-sense heritability was calculated from variance components of the individual traits using the formula: where ĥ 2 is the narrow sense heritability, σ 2 a is the additive genetic variance and σ 2 ƞ residual error component of variance.
Genetic and phenotypic correlations were calculated between all combinations of flowering traits (start, duration and end), dry biomass weight and foliar cineole concentration using the formula: where r is either r g , the genetic correlation, or r p , the phenotypic correlation. σ xy is the additive genetic covariance for the genetic correlations and σ 2 x and σ 2 y are the additive genetic variances of the two traits. For the phenotypic correlations, σ xy is the phenotypic covariance and σ 2 x and σ 2 y are the phenotypic variances for the traits.

Results
General flowering observations. Of the population of 1142 trees, 87% flowered in each year ( Table 2, Fig. 2a). The reproductive activity differed between the subspecies with E grat from the two south provenances having the lowest (79-82%) flowering in both years ( Table 2). E liss showed consistent reproductive activity: 91% and 93% from the eastern and western provenances, respectively. From the monthly observations, the population commenced flowering in late summer to early autumn, with peak flowering in spring (Fig. 2a). Flowering was about 2 weeks earlier in 2014 and this trend continued for most of the assessment period. The earlier flowering in 2014 corresponded with a higher heat sum (4515 °C) compared to 2012 (4332 °C).
Tracking of individual reproductive flushes in 2014 revealed that 16% of trees flowered on two separate occasions ( Table 2). Two provenances from the E liss east region had the highest proportion of dual flush trees followed by the two E grat provenances. The trees that had dual reproductive phases exhibited two different flowering  Table S1).
In assessing the duration of flowering, flowering phases that commenced after week 38 were omitted because a high proportion had not concluded by the time the study was terminated.

Effect of provenances on traits.
Flowering start, end, duration, proportion of flowering, proportion of dual flush, foliar cineole content and biomass differed, significantly (P < 0.0001) between provenance and family within provenance and between family within provenance for start of flowering (P < 0.001) ( Table 3). High levels of synchronicity in flowering events occurred within provenances, but there were substantial differences in the timing of flowering between provenances, as shown in (Fig. 3 and Table 4). Two of the E liss east provenances, Norseman and Coolgardie, had the highest proportion of dual flush trees followed by the two E grat provenances (Tables 2, 5). All of the E liss west provenances had a consistently low (1-8%) percentage of dual flush trees. The average foliar cineole concentrations for each provenance ranged from 1.6% in Southern Cross to 2.7% of green leaf weight in Coolgardie (Table 5). On average, the E liss east provenances tended to have higher cineole concentration, while the E liss west region contained the two worst performing provenances. Table 2. Proportion of individuals flowering (%) for trees derived from each provenance for each assessment year and the proportion of trees with dual reproductive flushes (%) in the 2014 assessments.  www.nature.com/scientificreports/ There was consistency in the biomass of different provenances between first and second measurement (Table 5).
For the pre-and post-thinning growth, both E grat provenances were ranked among the top performers along with E liss from Goongarrie. Coolgardie and Westonia were the poorest performers in both assessments. On average, E liss east and E liss west were similar in their biomass ranking.  www.nature.com/scientificreports/ Effect of families on traits. There was substantial variation in flowering times within provenances. There was, for instance, about 3 months difference between the commencements of flowering among families derived from parents with Lake Grace, Norseman or Southern Cross provenance (Fig. 4). There was a spread of eight weeks in end of flowering within the individuals from the Lake Grace provenance and nearly 6 weeks range for Goongarrie and Norseman. There was also substantial overlap of flowering between most families although there are a few early or late families which were reproductively isolated from other families in the seed orchard.     Supplementary Table S2. Table 6 shows substantial differences in the proportion of trees that flowered, proportion of trees with dual flushes, foliar cineole content and biomass from the families within provenances. The families within provenances that showed wide ranges in the proportion of trees that flowered were from Lake Grace (62-100%), Southern Cross (81-100%), and Dumbleyung (82-100%), while the rest had narrower ranges (90-100%). Similarly, there were families within provenances with considerable variation in dual flush proportions: in families from the Dumbleyung, Lake Grace, Coolgardie and Norseman provenances, the percentage of dual flush trees ranged from < 1 to 69%. For families from Goongarrie, there were no variation (both at 1%), while for the rest of the families within the remaining provenances, trees with dual flush were in the range of 0-7%.
There were two families (46 and 58) that averaged foliar cineole under 0.9% of green leaf weight whereas the top performing families had cineole concentration exceeding 3.0%. Several provenances have tight ranges (little variation among families) while other provenances show wide ranges between families. For example, families within the Narembeen provenance show little variation (2.5-2.7%), whereas those within Westonia are highly variable (0.9-2.6%). Likewise, there is more variation in biomass estimates across both measurements within the Lake Grace provenance compared to Westonia. There was nearly double the average back-transformed biomass from the high yielding families in the pre-thinned assessment (1, 6, 15 and 16 vs 59, 34, 22 and 46) and this difference was more pronounced after thinning (26, 3, 13 and 6 vs 46, 24, 22 and 34).
Heritability, genetic and phenotypic correlations. Narrow-sense heritability ranged from high for end of flowering ( ĥ 2 = 0.66-0.75), moderate for start flowering traits ( ĥ 2 = 0.45 ± 0.10) and foliar cineole content ( ĥ 2 = 0.53 ± 0.09) and low for duration of flowering and both biomass estimates ( ĥ 2 = 0.10-0.33) ( Table 7). Narrow-sense heritability for dual flush flowering was ( ĥ 2 = 0.61 ± 0.16), propensity to flower in 2012 ( ĥ 2 = 0.19 ± 0.10) and 2014 ( ĥ 2 = 0.24 ± 0.11). There were strong genetic correlations between the same flowering traits from different years (r g = 0.84-0.96) but the phenotypic correlations were lower (r p = 0.40-0.55). Genetic correlations were also high between most of the key flowering traits across years except for end of flowering and duration of flowering which generally had high standard errors. The genetic correlations between the start and end of flowering were all positive and above 0.62. The magnitudes of the genetic correlations between start and duration of flowering were also high although these correlations were negative. Phenotypic correlations were generally lower

Discussion
Understanding flowering phenology is critical in seed orchard design, as synchronicity of flowering amongst parents is key to maximising outcrossing, genetic quality of the seed, and its ability to deliver traits of commercial interest. The results from this study displayed large variation in flowering phenology within and between subspecies, provenances and families. With the exception of a few families, there was a high level of synchronicity of flowering within the seed orchard but substantial differences between provenances. However, caution should be used when directly comparing the provenance-level data as there are different numbers of families in each provenance; for example, Coolgardie, Goongarrie and Westonia were poorly represented with a small number of families.
The large range of commencement and end of flowering times within provenances may in part be reflected by the percentage of dual-flush trees in each provenance. In most cases, dual flush trees started flowering earlier (i.e. the first flush) in the year than their single flush contemporaries, resulting in greater flowering duration. These were the most of the out-of-phase trees in the orchard. We estimate that only a small proportion (< 4%) of the total buds in the orchard flowered out of phase with the single flush trees. Early dual flush trees which   Table 6. Least square means for foliar cineole content (% green weight), biomass estimates of pre-thinned (aged 5) and post-thinned (aged 15) orchard, arcsine transformed proportion of flowered and dual flush flowing with the standard error in brackets (SE). DUM Dumbleyung, LGR Lake Grace, COO Coolgardie, GOG Goongarrie, NOR Norseman, NAR Narembeen, SNC Southern Cross, TYG Trayning, WES Westonia. Table 7. Heritability, genetic and phenotypic correlations for Eucalyptus loxophleba seed orchard. Narrowsense heritability (± standard errors) of each trait is on the diagonal (in bold), above the diagonal is genetic correlation (± standard errors) and below the diagonal is phenotypic correlations (± standard errors) in italics.  www.nature.com/scientificreports/ were active before the main flush of the orchard will have a greatly reduced opportunity to outcross with other trees in the seed orchard.
After the E liss eastern provenances the southern E grat provenances had the second highest proportion of dual flush trees. There is no evidence that the two regions exhibiting high rates of dual flush flowering is due to their relatedness. Investigation on the chloroplast DNA has revealed that E grat is more closely related to the neighbouring E liss western and E. loxophleba subsp. loxophleba than to E liss eastern provenances 53 . However, the high heritability of the dual flush trait ( ĥ 2 = 0.61 ± 0.15) suggests that selections could be made to reduce its prevalence in breeding populations. A large percentage of these dual flush trees were E grat , and based on this alone, it may be appropriate to separate E grat into a separate breeding population. The taxonomic split between E liss and E grat is contentious 54,55 , however, a breeding population comprised of E grat alone would allow more out-crossing of these dual flush trees especially at the start and end of the annual reproductive cycles. Considering E liss has been shown to have weak reproductive barriers [38][39][40][41][42] , it would allow the subspecies to be planted in their natural range thereby minimising genetic 'pollution' of other subspecies.
The dual flush E liss trees, most of which originate from eastern provenances, flowered out of phase with the single flush E liss . Many of these families also performed poorly in biomass assessments. For example, the Coolgardie provenance (families 22, 23 and 24) ranked last for biomass in the two measurements of this trial, flowered considerably earlier than any other family and nearly 30% of the trees flowered twice. Family 22 had the highest foliar cineole concentration, but ranked second and third last for biomass and was the earliest family to flower. Some families in the Norseman provenance, in contrast, included seven of the latest ten flowering families with a high proportion of dual flush trees and below average biomass. High cineole families may be kept in the seed orchard to maintain cineole levels, but seed should not be collected from these families for biomass plantings and they should be carefully considered before inclusion in future breeding programmes. Families with a high proportion of dual flush trees, low cineole and biomass rank, could be removed from the breeding populations. The other E liss eastern provenance, Goongarrie, consisted of two families, both with a low proportion of dual flush trees. Family 26 was the fourth latest to flower, but this family ranked highest in biomass in the second assessment and family 25 flowered during peak time and ranked well for both biomass and cineole. Thus, although E liss east provenances were on average poor performers in the seed orchard, some families from the eastern provenances may be candidates for next-generation seed or clonal orchards.
Mazanec et al. 18 reports on the variation of biomass from different provenances from three large E liss progeny trials across southern Australia. Each trial consisted of nine provenances with at least 13 families. It was found that the progeny from Norseman ranked last for biomass at two of the three trials and the progeny from Cardunia Rocks, the other eastern provenance, performed average or below at the three trials. These trials were measured at age three and gives further support for the poor biomass performance from the E liss eastern range. However, short-duration studies can at times fail to indicate longer-term performance and further measurements of these trial should reveal more about the top biomass producing E liss provenances.
This study shows that, with the exception of a few families, there is a considerable proportion of trees flowering in the seed orchard throughout most months the year. However, the amount of reproductive activity is uneven; the rate of anthesis is much more pronounced in late winter and spring when it is common for 30-50% of buds to have commenced flowering in a 1-month period. This contrasts with other studies of southern Eucalyptus species. For example, warmer temperatures are known to trigger key developmental stages in E. nitens including bud initiation and growth 56 and heat-sum has been found to be the main driver of anthesis time in E. globulus and E. nitens 26,57 . With only 2 years of data, the effect that heat sum had on the timing of anthesis for E. loxophleba is impossible to determine. The year of greater heat did correspond with earlier flowering and shorter flowering duration, both expected outcomes in the heat sum model. However, E liss and E grat, in contrast to E. nitens and E. globulus, are adapted to arid conditions. Moisture availability and rainfall events have been identified as important factors affecting anthesis for plants in arid environments [58][59][60][61][62] and that is a likely cause of slow reproductive development of E liss and E grat during late summer and autumn. Furthermore, Friedel et al. 59 demonstrated that soil moisture was a predictor of flowering for 46 arid-zone species although flowering events lagged between 1 and 9 months after rainfall. This is consistent with the flowering of E liss and E grat in this study where there were no significant rainfall events in 2012 until May and in 2014 until April. E liss and E grat are well adapted to drought and appear to exhibit drought-induced dormancy of the crown.
The plastic nature of the species recorded in the arid zone of Australia by Friedel et al. 59 , Davies 61 and others suggest that eastern provenances of E liss may be more strongly adapted to rainfall-induced growth and reproduction than western provenances. The multiple reproductive events observed in this trial may be a function of that adaptation. It is possible that trees from the low rainfall eastern provenances were responding to the higher rainfall experienced near Lake Toolibin. However, the two most southerly provenances (E grat ) also had a high percentage (20%) of dual flushes trees and are the two closest provenances to Lake Toolibin. E grat was less reproductively active, suggesting that these provenances did not benefit from the slightly higher rainfall of the study site when compared to their natural range. In contrast, western and eastern E liss provenances displayed high levels of reproductive activity at a site with higher rainfall and lower evaporation compared to their natural distribution. There was however, no trend in northern provenances to flower earlier than the more southerly E grat provenances, a trend recorded for E. marginata 63 and for Corymbia citriodora subsp. variegata and C. maculata 27 . However, the opposite trend has been observed in E. globulus where Victorian provenances flowered later than the more southerly eastern-coast Tasmanian provenances 26 .
Biomass generally has a lower heritability than either flowering traits or cineole production and this was observed in this trial. The pre-thinned biomass heritability was low ( ĥ 2 = 0.10 ± 0.03) but increased after thinning to 0.29 ± 0.08 suggesting that estimates of additive genetic variance for this trait were biased upwards as a result. The difference between the two estimates suggests that some degree of bias was introduced as a result of selective  64,65 , and E. cladocalyx ĥ 2 = 0.14 66 and ĥ 2 = 0.30 67 . Heritability of flowering traits in this study were under moderate to strong genetic control. These results are consistent with results from other studies in eucalypts. Jones et al. 26 reported high broad-sense heritability for peak anthesis time ( Ĥ 2 = 0.78) in an E. globulus clonal seed orchard but found weak heritability for duration of flowering ( Ĥ 2 = 0.09) with the low heritability attributed to the correlation between duration of flowering and flower abundance 68 . For the same species, Gore and Potts 25 found narrow-sense heritability over 0.64 for start, peak and end of flowering after a single year of assessment. Flowering intensity of E. cladocalyx has been recorded as ĥ 2 = 0.48 69 and ĥ 2 = 0.52 66 . The number of reproductive flushes was under strong genetic control ( ĥ 2 = 0.61 ± 0.16) but the propensity to flower in 2012 and 2014 were quite low ĥ 2 = 0.19 ± 0.10 and ĥ 2 = 0.24 ± 0.11 respectively, although this is much higher than the ĥ 2 = 0.06 ± 0.05 reported for E. globulus 26 .
Foliar cineole concentration was observed to be under strong genetic control ( ĥ 2 = 0.53 ± 0.09). Mazanec et al. 18 found similar narrow-sense heritability at an E liss progeny trial in Brookton, Western Australia, of over 1700 trees from 126 families ( ĥ 2 = 0.53 ± 0.07) and from an E grat progeny trial of 90 families of ( ĥ 2 = 0.50 ± 0.08) 70 . Similar heritability was found in E. camaldulensis 29 . Heritability as high as ĥ 2 = 0.83 was found in a small E. kochii seed orchard 71 . In the current study, the parent trees had been tested for foliar cineole concentration prior to selection with seed only sourced from elite individuals (foliar cineole concentration > 2.5%) so selection was biased to high cineole individuals. This is because at that time, in the genetic selection of this species, cineole was considered likely to be the major product for commercial planting of these species 8,11 . The two more recent E liss and E grat progeny trials mentioned above were not subject to pre-selection for foliar cineole but the heritability results were similar suggesting that pre-selection for high cineole did not influence estimates of additive variance in this trial.
The strong genetic and phenotypic correlations between duration and start of flowering have been found in other species, for instance, Lythrum salicaria (r p = − 0.92) 72 . The correlations for E liss are much higher when comparing within years than across years and may be attributable to the annual variation in dual flush flowering which exhibit longer duration of flowering. It is unknown if the trees that flowered twice in 2014 also did so in 2012 and although the heritability for this trait is high, another assessment following individual reproductive flushes would be useful. Furthermore, assessment of the performance of progeny from the first-and second-flush of the dual flush trees and single flush trees could indicate the degree of inbreeding or selfing.
Genetic correlations suggest that selection for high biomass and cineole may result in selection of early flowering trees. For example, if selection is biased toward biomass then the low genetic correlation for that trait with flowering time will allow greater freedom in selection of trees with later flowering times. The orchard was initially conservatively thinned, in most cases an individual from each family was retained from each replicate. Poor biomass producing E liss east families with high levels of dual flush flowering could be eliminated from the orchard which would increase flowering synchronicity without negatively impact biomass production.

conclusion
This study has shown that timing of anthesis is strongly influenced by genetic factors. Genetically, there is a large amount of variation with broad-scale differences among provenances and families. Most flowering traits, along with cineole, were moderately to strongly heritable whereas biomass heritability was low. At the start and end of the annual flowering cycle, dual flush trees were reproductively isolated and because the trait is strongly heritable, this could result in greater flowering asymmetry in progeny collected from the dual flush trees. Dual flush flowering was evident in the E grat provenances. For this reason and because of the potential spread of genetic material to native stands through pollen dispersal, E grat should be treated as a separate breeding population for use within its natural range. The E liss east provenances included families with the highest proportion of dual flush trees and the poorest biomass yield. These families should be eliminated from the general breeding programme. Due to the moderate to negligible genetic correlations between flowering traits, biomass and foliar cineole concentrations, selections should be based on biomass and cineole. Families with a high proportion of dual flush trees and otherwise desirable characteristics should be further studied to determine if this was a one-off event or if it is a recurring phenomenon. Opportunistic use of available soil moisture may be responsible for dual flush flowering in eastern provenances, however, the 2-week divergence in flowering time between assessment years may be driven by heat-sum.