Eucalyptus applied genomics: from gene sequences to breeding tools

Authors

  • Dario Grattapaglia,

    1. Plant Genetics Laboratory, Embrapa – Genetic Resources and Biotechnology, Parque Estação Biológica, Brasília 70770-970 DF, Brazil;
    2. Graduate Program in Genomic Sciences and Biotechnology, Universidade Católica de Brasília – SGAN 916 módulo B, Brasília 70790-160 DF, Brazil;
    Search for more papers by this author
  • Matias Kirst

    1. School of Forest Resources and Conservation, Graduate Program in Plant Molecular and Cellular Biology, and University of Florida Genetics Institute, University of Florida, PO Box 110410, Gainesville, FL 32611, USA
    Search for more papers by this author

Author for correspondence:
Dario Grattapaglia Tel: +55 61 3448 4652 Fax: +55 61 3340 3624 Email: dario@cenargen.embrapa.br

Abstract

Contents

 Summary911
I.Introduction912
II.Eucalyptus biology and domestication913
III.Eucalyptus breeding and clonal forestry913
IV.Marker-assisted management of genetic variation in breeding populations914
V.Genetic mapping and quantitative trait locus (QTL) analysis915
VI.Gene discovery and genetical genomics916
VII.Association mapping918
VIII.Molecular breeding920
IX.From gene sequences to breeding tools922
X.Future developments and challenges924
 Acknowledgements926
 References926

Summary

Eucalyptus is the most widely planted hardwood crop in the tropical and subtropical world because of its superior growth, broad adaptability and multipurpose wood properties. Plantation forestry of Eucalyptus supplies high-quality woody biomass for several industrial applications while reducing the pressure on tropical forests and associated biodiversity. This review links current eucalypt breeding practices with existing and emerging genomic tools. A brief discussion provides a background to modern eucalypt breeding together with some current applications of molecular markers in support of operational breeding. Quantitative trait locus (QTL) mapping and genetical genomics are reviewed and an in-depth perspective is provided on the power of association genetics to dissect quantitative variation in this highly diverse organism. Finally, some challenges and opportunities to integrate genomic information into directional selective breeding are discussed in light of the upcoming draft of the Eucalyptus grandis genome. Given the extraordinary genetic variation that exists in the genus Eucalyptus, the ingenuity of most breeders, and the powerful genomic tools that have become available, the prospects of applied genomics in Eucalyptus forest production are encouraging.

I. Introduction

Intensive production forestry based on exotics began in the southern hemisphere c. 50 yr ago. Since then, the world forest industry has experienced a slow, steady, but now increasing shift of plantation forestry from the northern hemisphere to the tropics and subtropics. Eucalyptus species have been playing a significant role in this process (Kellison, 2001). High-productivity eucalypt forests supply high-quality raw material for pulp, paper, wood, and energy that would otherwise come from native tropical forests. The expansion of these ‘fiber farms’ may become limited by land needs for growth of food and other biofuel crops and, in some cases, by public pressure. Strategically, the increase in forest productivity and refinement of the quality of wood products through the use of genome-assisted breeding and transgenic technologies will become increasingly important to the forest industry.

While a number of genes affecting wood formation in forest trees have been intensively investigated and manipulated in recent years (reviewed by Bhalerao et al., 2003; Boerjan, 2005), significant applications of transgenic technology to eucalypt production forestry are still to come. Challenges are also faced in molecular breeding applications of genomics. Seventeen years have passed since the first experiments in genetic mapping and molecular breeding of forest trees were described (Neale & Williams, 1991; Grattapaglia et al., 1992). From the outset, many expectations of fast and accurate methods for early marker-based selection for growth and wood properties in trees were generated. Considerable progress has been made but significant challenges still exist for the implementation of high-impact applications.

Reverse and forward genomics approaches have been pursued in Eucalyptus research. Reverse genomics operates by generating altered phenotypes from the manipulation of a given gene through transgenic technology or induced mutations. The forward genomics approach, that is, analysis of existing phenotypic variation to identify causal genetic variants, is based on the wide natural intra- and interspecific diversity that exists in Eucalyptus. The technologies involved in this latter approach include genetic mapping, quantitative trait locus (QTL) discovery, association genetics, physical mapping and genome sequencing, some of which will be discussed in this review (Fig. 1).

Figure 1.

Schematic representation of the forward genomics approach. Genetic mapping and quantitative trait locus (QTL) discovery are carried out in segregating populations specifically generated to display large amounts of natural phenotypic variation for breeding relevant traits. Major-effect QTLs are detected with a resolution of ~10 centiMorgans, defining target genomic segments involved in the control of the measured traits. Narrower genomic windows of < 1 cM are defined that can be anchored to physical maps of large insert clones such as Bacterial Artificial Chromosomes (BACs) covering ~500–1000 kbp of DNA by high-resolution mapping. Several reverse genomics approaches (boxes) can be used to identify, test and validate specific genes that underlie the target QTL.

Recent reviews have described Eucalyptus genome research, including gene discovery, candidate gene mapping and functional genomics (Poke et al., 2005; Myburg et al., 2007). Here we discuss how the combination of the unique biology of the eucalypts and the current and emerging technologies, as well as the access to a draft of the Eucalyptus genome sequence, will allow a deeper understanding of complex quantitative traits and the translation of genomics into applied breeding tools. Three main points should be highlighted to summarize our main views.

  • • The wide phenotypic variation across Eucalyptus species has prompted remarkable gains through hybrid breeding. This same variation has been and will increasingly be a powerful resource for detecting major-effect QTLs in experimental hybrid populations as species-specific properties might be associated with high-frequency alleles fixed in contrasting species. The most significant gains from marker-assisted selection (MAS) may be achieved through the introgression of such valuable species-specific alleles into breeding populations.
  • • The broad genetic diversity of Eucalyptus natural populations is likely to contain an abundance of low-frequency alleles, some of which may be the next gems to be uncovered by a combined QTL/genetical genomics approach and used by breeders for MAS.
  • • The rapid improvements of genotyping and sequencing methods and major advances in bioinformatics will soon make it feasible to identify all the sequence variants that occur at moderate to high frequencies in any Eucalyptus breeding population by ultra-high-throughput shotgun sequencing of the genome of each individual tree. At this point the lack of well-replicated and accurately phenotyped experimental populations will be more limiting than genomic resources and technologies for an effective move from gene sequences to breeding tools.

II. Eucalyptus biology and domestication

The genus Eucalyptus includes over 700 species, some of which are the most widely planted hardwoods worldwide. They are long-lived, evergreen trees belonging to the angiosperm family Myrtaceae, which occurs predominantly in the southern hemisphere (Ladiges et al., 2003). Eucalypts are native to Australia and islands to its north, occurring from sea level to the alpine tree line, from high-rainfall to semi-arid zones, and from the tropics to latitude 43° south (Ladiges et al., 2003). Broadly speaking the eucalypts include species of the genera Eucalyptus L’Hérit., Corymbia Hill and Johnson, and Angophora Cav., although these last two (bloodwood taxa) warrant a phylogenetic separation from the Eucalyptus (nonbloodwood taxa) in the strict sense, based on molecular studies (e.g. Steane et al., 2002) The latest taxonomic revision (Brooker, 2000) of the eucalypts recognizes over 700 species that belong to 13 main evolutionary lineages, still considering the bloodwood eucalypts as subgenera of Eucalyptus. Most species belong to the subgenus Symphyomyrtus, and it is mainly species from three sections of this subgenus that are used in plantation forestry such as Eucalyptus grandis and Eucalyptus urophylla (section Transversaria), Eucalyptus globulus (section Maidenaria) and Eucalyptus camaldulensis (section Exsertaria).

Eucalypts were rapidly adopted for plantation forestry following their discovery by Europeans in the 18th century (Eldridge et al., 1993). They were introduced into India, France, Chile, Brazil, South Africa, and Portugal in the first quarter of the 1800s (Doughty, 2000) and quickly selected for plantations as their remarkable growth and adaptability were realized. Several seed collection expeditions followed the first introductions, resulting in germplasm being broadly distributed throughout the world. Today several countries maintain large and diverse germplasm collections of Eucalyptus that include a broad range of provenances for the most widely planted species.

Eucalyptus are predominantly outcrossers (Moran et al., 1989; Gaiotto et al., 1997) and insect pollinated. Protandry, late-acting self-incompatibility barriers and, possibly, cryptic self-incompatibility mechanisms (Horsley & Johnson, 2007) are responsible for the preferential outcrossing system and strong selection against inbreeding (Potts & Savva, 1988). Interspecific hybridization in natural populations has been observed (Pryor & Johnson, 1971; Griffin et al., 1988) and becomes relatively frequent in exotic conditions. A textbook case is the hybrid swarm derived from the Rio Claro arboretum established by Navarro de Andrade in Brazil c. 1904 with 144 Eucalyptus species (Campinhos & Ikemori, 1977; Brune & Zobel, 1981). Large commercial plantation forests were established with seeds from this arboretum in the 1960s. Most of them, however, had inferior quality because of the extensive segregation observed. However, some outstanding hybrid trees were selected that displayed superior growth, form, and disease resistance. Today, this property is successfully exploited by eucalypt breeders in tropical countries who take advantage of the naturally occurring genetic variation for growth, adaptability and wood properties among species to consolidate complementary traits in hybrid clones (de Assis, 2000).

III. Eucalyptus breeding and clonal forestry

The advent of industrially oriented eucalypt plantations in the 1960s and 1970s led to a formal approach to breeding in several countries (Eldridge et al., 1993). Breeding of eucalypts for industrial plantation forestry developed rapidly in countries such as Brazil, South Africa, Portugal, and Chile. Target traits include volume growth, wood density, and pulp yield. Traits related to biotic and abiotic stresses are usually secondary targets that become more important when they impact the main traits. Following the standard concepts in tree breeding, large genetic gains have been obtained in the early stages of eucalypt domestication through species and provenance selection followed by individual selection and establishment of seed orchards. However, a major breakthrough in eucalypt plantation technology occurred in the 1970s with the plantation of the first commercial stands of selected clones derived from hardwood cuttings (Martin & Quillet, 1974; Campinhos & Ikemori, 1977). Since then, vegetative propagation has enhanced the genetic gains of breeding programs by capturing both additive and nonadditive effects into elite clones (Fig. 2).

Figure 2.

Basic breeding scheme used in Eucalyptus breeding involving recurrent selection cycles from which elite trees are derived for the establishment of seed orchards and/or to be tested as potential commercial clones.

Clonal propagation and hybrid breeding have become a powerful combination of tools for the improvement of wood product quality. Segregation of high genetic diversity in hybrid populations, combined with intensive within-family selection and clonal propagation, resulted in exceptional genetic gains for growth and adaptability to tropical conditions, wood with higher pulp yield and vegetative propagation capability. Eucalypt hybrids deployed as clones currently make up a significant proportion of the existing commercial plantations and are recognized as some of the most advanced genetic materials in forestry. In the last 15 yr, with the increasing realization that the actual ‘pulp factory’ is the tree, the breeding focus of pulp and paper companies has shifted from volume growth to wood quality. The objective is to improve pulp yield per hectare by reducing wood specific consumption (WSC), that is, the amount of wood in cubic meters necessary to produce one ton of pulp. Clonal forestry of E. grandis hybrids in the 1980s reduced WSC by 20% (Ikemori et al., 1994). Second-generation clones derived from hybridization with E. globulus, the superior species for wood quality, has allowed a further reduction of 20% (de Assis et al., 2005). Trees that yield more cellulose create gains along the entire production chain by generating savings from tree harvesting and transportation to chipping and pulping, while mitigating the expansion of the commercial forest land base and reducing effluent waste. It is in the context of a specialized, industrially oriented breeding program that fully exploits the power of hybrid breeding and clonal forestry that we discuss the prospects of genomics and molecular breeding of Eucalyptus. From the genetics perspective, however, eucalypts are still in the early stages of domestication when compared with crop species. Most eucalypt breeding programs are only a few generations removed from the wild and this fact has important implications when applying genomics approaches.

IV. Marker-assisted management of genetic variation in breeding populations

The use of genome information for the practice of directional selection of superior genotypes still represents a challenge that depends on further and more refined experimental work. Nevertheless, molecular markers have been used to solve several questions related to the management of genetic variation, identity and relationship in breeding and production populations. The correct identification of clones is currently the most common application of molecular markers in Eucalyptus operational breeding and production forestry. Quality control of large-scale clonal plantation operations is crucial, especially in vertically integrated production systems where the pulp mill relies on the availability of wood from clones with specific wood properties. Plantation forest planning and propagation happen several years before wood consumption – therefore, mislabeling can seriously affect the whole production process. Correct clonal identity also has important implications in breeding procedures where mislabeled clones can significantly affect the expected gains from breeding. Several technologies have been used to resolve clonal identity in Eucalyptus such as random amplification of polymorphic DNA (RAPD) (Keil & Griffin, 1994), amplified fragment length polymorphism (AFLP) (Gaiotto et al., 1997) or DNA microarray technology (Lezar et al., 2004). However, co-dominant microsatellites have provided the most powerful method for the unique identification of Eucalyptus trees (Kirst et al., 2005a; Ottewell et al., 2005). Most of the microsatellites currently used are derived from dinucleotide repeats that provide powerful discrimination, but present challenges when multilocus profiles need to be compared between laboratories. To this end we have developed a novel set of microsatellites for Eucalyptus, based on tetra- and pentanucleotide repeats, the gold standard established by human forensics. Selected markers of this kind display heterozygosities above 0.45 while providing significantly more robust multilocus genotype profiles (C. Sansaloni and D. Grattapaglia, unpublished).

Molecular markers have been used to characterize the range of genetic variation in germplasm banks of Eucalyptus globulus, thereby assisting in the design of new seed collections (Nesbitt et al., 1995). Marker-based genetic distance data have been used to improve the structure of breeding populations (Marcucci-Poltri et al., 2003) and seed orchards (Zelener et al., 2005). Given the wide genetic diversity and multiple sources of germplasm for eucalypt breeding, choices have to be made as to which elite parents should be mated. Any means of predicting tree performance would be valuable for the breeder. Vaillancourt et al. (1995a) showed that the ability of RAPD-based genetic distance to predict heterosis was significant but accounted for less than 5% of the variation in specific combing ability (SCA) in E. globulus progenies. However, Baril et al. (1997) showed that a genetic distance based on RAPD markers with similar frequencies in two intercrossed Eucalyptus species successfully predicted the value of a hybrid cross with a global coefficient of determination of 81.6%. In a recent large-scale study, de Aguiar et al. (2007) showed that microsatellite-based genetic divergence was significantly correlated with SCA and hybrid family mean for volume growth. However, only 6.3% of the SCA was explained by the genetic distance, revealing that it is still of limited value for practical purposes.

Knowledge of outcrossing versus selfing rates in breeding populations is essential for maintaining adequate levels of genetic variability and continuous gains over generations. Eucalypts are preferentially outcrossed both in natural populations and in seed orchards (Moran et al., 1989; Gaiotto et al., 1997) although complex patterns of mating influenced by crop fecundity and orchard position of mother trees (Burczyk et al., 2002) and high rates of pollination from outside the seed orchard (Chaix et al., 2003) have been described. The ability to retrospectively select better combining parents using DNA paternity testing resulted in significant gains in volume growth in the forest planted with the improved seeds (Grattapaglia et al., 2004a).

V. Genetic mapping and QTL analysis

1. Genetic markers and maps

The first Eucalyptus linkage maps were built with combinations of several hundred RAPD and AFLP markers (Grattapaglia & Sederoff, 1994; Verhaegen & Plomion, 1996; Marques et al., 1998; Myburg et al., 2003) together with restriction fragment length polymorphism (RFLP), isozymes, expressed sequence tags (ESTs), genes and more recently some microsatellites (e.g. Byrne et al., 1995; Bundock et al., 2000; Gion et al., 2000; Brondani et al., 2002; Thamarus et al., 2002). Genetic mapping was carried out mostly using a pseudo-testcross (Grattapaglia & Sederoff, 1994) with results of limited value for inter-experimental sharing of linkage and QTL data. Until mid 2006, only 137 autosomal microsatellite markers had been published for species of Eucalyptus, including 67 from E. globulus, E. nitens, E. sieberi, and E. leucoxyon (Byrne et al., 1996; Steane et al., 2001; Glaubitz et al., 2001; http://www.ffp.csiro.au/tigr/molecular/eucmsps.html; Ottewell et al., 2005) and 70 from E. grandis and E. urophylla (Brondani et al., 1998; Brondani et al., 2002). Recently a microsatellite only consensus map covering at least 90% of the estimated recombining genome of Eucalyptus was reported. This map has 234 mapped loci on 11 linkage groups, an observed length of 1568 cM and a mean distance between markers of 8.4 cM (Brondani et al., 2006). Microsatellite transferability across intensively planted species of the subgenus Symphyomyrtus varies between 80 and 100% depending on the section to which the species belongs. Transferability is approx. 50–60% for species of different subgenera such as Idiogenes and Monocalyptus and 25% for the related genus Corymbia (Kirst et al., 1997). Microsatellite comparative maps have also shown that genome homology across species of the same subgenus is high, as is marker collinearity along linkage maps (Marques et al., 2002). The generalized use of an increasingly large set of interspecific transferable markers and consensus mapping information will allow faster and more detailed investigation of QTL synteny among species, validation of QTLs and expression QTLs across variable genetic backgrounds, and positioning of a growing number of candidate genes co-localized with QTLs, to be tested in association mapping experiments.

2. QTL analysis and validation

QTL mapping in Eucalyptus has invariably found major-effect QTLs for all traits considered, in spite of the generally limited experimental precision, the lack of pre-designed pedigrees that maximize phenotypic segregation, and the relatively small segregating populations. The success in detecting those major-effect QTLs can be explained by the undomesticated nature and wide genetic heterogeneity of eucalypts, and the fact that most QTL mapping experiments were carried out in interspecific populations with contrasting gene pools. QTLs for juvenile traits such as seedling height, leaf area, and seedling frost tolerance have been mapped (Vaillancourt et al., 1995b; Byrne et al., 1997a,b). QTLs that regulate traits related to vegetative propagation ability have also been detected (Grattapaglia et al., 1995; Marques et al., 1999), as well as a major QTL for early flowering (Missiaggia et al., 2005). In addition, QTLs for insect resistance, essential oil traits and terpenes were mapped (Shepherd et al., 1999; Henery et al., 2007). A major QTL for Puccinia psidii rust resistance was found and mapped in E. grandis (Junghans et al., 2003) and later positioned on a microsatellite consensus map (N. Bueno and D. Grattapaglia, unpublished). Two major-effect QTLs for resistance to another fungal epidemic caused by Mycosphaerella were recently mapped in E. globulus (Freeman et al., 2008), providing further evidence for oligogenic control of fungal disease resistance in Eucalyptus. Major QTLs were also found for harvest age traits such as volume growth, wood specific gravity, bark thickness and stem form (Grattapaglia et al., 1996; Verhaegen et al., 1997; Kirst et al., 2004, 2005b; Thamarus et al., 2004).

In spite of a relatively large number of QTLs mapped in Eucalyptus, the application of QTL information for directional selection is still an unfulfilled promise. This is largely a consequence of some instrinsic properties of the genus, including (a) its recent domestication and hence the high genetic heterogeneity and linkage equilibrium of breeding populations; (b) the biological barrier that has been encountered to the development of inbred lines to allow a more precise understanding of the architecture of quantitative traits, and (c) the absence of simply inherited traits that could be immediately and more easily targeted. Furthermore, the limited resolution and allelic coverage of mapping experiments have represented a major obstacle. QTL studies have typically identified broad genomic regions which are likely to comprise several hundred genes or cis regulatory elements and, therefore, represent only a small step closer to the identification of the causative polymorphism. Larger population sizes and marker densities could effectively improve the resolution by an order of magnitude, but even under the best scenario, intervals of less than a few centiMorgans would hardly be obtained. Fine-mapping to narrow the QTL interval is difficult to implement in eucalypts because of the impossibility of inbreeding and thus isogenizing lines differing exclusively in the target segment.

Another limitation faced by QTL studies in eucalypts is the restricted number of alleles that are evaluated per cross/study. Because only the genetic variation in the two parental lines can be evaluated in a single cross, alleles that are relevant for the trait variation may either not be sampled or not be variable in the population. Low-frequency, high-value alleles are highly likely not to be captured in these populations. Furthermore, the map positions of most QTLs discovered to date in Eucalyptus cannot be compared because they were mapped onto RAPD or AFLP maps, although this picture has recently started to change. Where transferable markers were also mapped, a preliminary comparison of QTL locations has been made. Marques et al. (2002) showed that putative QTLs for vegetative propagation traits were located on homeologous linkage groups in different Eucalyptus species. Thamarus et al. (2004) reported that QTLs for wood properties traits could be detected in two related full-sib families of E. globulus, and Marques et al. (2005) were able to verify high-effect QTLs for adventitious rooting also in related E. globulus pedigrees. Freeman et al. (2008) identified two major-effect QTLs for Mycosphaerella resistance on a microsatellite map – the two QTLs could be validated in a second F2 family and one QTL in a third F2 family. Recently, comparative QTL mapping across three unrelated pedigrees of E. grandis and E. urophylla, as well as a comparison of QTLs mapped in E. globulus, revealed a number of syntenic QTLs for cellulose yield, lignin content and different but correlated fiber traits (D. Grattapaglia, unpublished). These results demonstrate the feasibility of validating QTLs and identifying target genomic regions for subsequent investigations. However, the lack of a genome sequence for a Eucalyptus species complicates the identification of genes and/or noncoding regulatory variants underlying a QTL, although this difficulty may soon be mitigated by the sequencing of the E. grandis genome.

VI. Gene discovery and genetical genomics

1. EST sequencing

Since the first Eucalyptus gene sequence was deposited in Genbank 14 yr ago (Feuillet et al., 1993), the number of public sequences has increased very slowly compared with other crops or major tree species. The reason has been the industrially oriented, proprietary nature of the genomic research carried out in Eucalyptus. About 10 yr ago, tens of thousands of sequences were generated by Genesis in New Zealand and later incorporated by the US-based company Arborgen (M. Hinchee, pers. comm.). Dupont, in collaboration with the former Australian company Forbio, generated a database of approx. 14 000 ESTs from E. grandis c. 1997 (S. Tingey, Dupont Crop Genetics, pers. comm.) that were later annotated and used in expression QTL studies (Kirst et al., 2004). Other mostly private EST databases for Eucalyptus were built between 2002 and 2004 in the Genolyptus project (Grattapaglia, 2004; Grattapaglia et al., 2004b) and the ForEST project (Vicentini et al., 2005) in Brazil and in Japan (S. Sato, unpublished). The different species planted around the world and the hybrid breeding system adopted in Eucalyptus have driven a distinctive multi-species approach to EST sequencing efforts.

A slight increase in the number of publicly available sequences started in 2003 with some small-scale cDNA microarray experiments carried out in Eucalyptus, which led to the publication of a few thousand sequences (Kirst et al., 2004; Paux et al., 2004; Foucart et al., 2006). During 2007 the French group led by J. Grima-Pettenati deposited a set of ~12 000 sequences, mostly derived from E. globulus, while a Chilean research group deposited another 8743 sequences. As of 31 December 2007 there were only 24 698 EST sequences available in Genbank out of the 43 664 deposited nucleotide sequences for Eucalyptus. It is expected, however, that with the breakthrough advancements of speed and dramatic cost reductions brought about by pyrosequencing technologies (Margulies et al., 2005) the size of public Eucalyptus EST databases will very soon become much larger and more diverse. Recently, the generation of more than 1 million E. grandis reads with average lengths of 100–200 bp from three runs of Genome Sequencer 20 and FLX Systems (454 Life Sciences Corporation) was reported, generating a preliminary assembly with approx. 29 000 contigs (E. Novaes et al., unpublished). As such large numbers of sequences become public, there will be a stimulus for the publication of existing private EST databases and a significant increase in available resources.

2. Genetical genomics

In an attempt to partially bypass some of the limitations to detecting genes underlying QTLs, Kirst et al. (2004) used a genetical genomics approach. This approach was originally proposed by Jansen & Nap (2001) and relies on analyzing gene expression measured by microarrays as a quantitative trait, using QTL analysis. By combining the QTL analysis of gene expression and trait in a segregating population, loci significantly associated with both phenotype and transcription of specific genes can be identified. Such ‘double-QTLs’ identify potential candidate genes for the regulation of the trait. This approach was applied in Eucalyptus, with the expression of ~2700 genes putatively involved in cell wall formation, lignin and cellulose metabolism, cell growth, and protein targeting being monitored. The key role of some lignin biosynthesis genes was confirmed, and some other unexpected new genes with major effect were also identified as highly correlated with volume growth. In a subsequent study, it was also shown that expression QTLs could explain up to 70% of the transcript level variation in one gene, and hotspots with co-localized expression QTLs were identified for over 800 genes, suggesting coordinated genetic regulation. However, the gene expression data also demonstrated a lack of conservation of the genetic architecture of transcript abundance regulation in different genetic backgrounds, indicating that many different loci could be involved in modulation of transcription of these genes in a complex and variable network of gene expression control (Kirst et al., 2005b). For MAS, such ‘master expression QTL’ may be very promising targets for tree breeding as they appear to be key regulators of a broad range of genes and, possibly, phenotypic traits. The identification of such large-effect genes by genetical genomics could also reveal important positional leads to nearby noncoding regulatory variants responsible for quantitative variation, although the challenges both in understanding and in assigning such variants to a measurable fraction of the phenotypic variation of a trait will be significant in Eucalyptus.

A limitation of the genetical genomics approach in Eucalyptus is the lack of a full genome sequence, or at least a genetic map including large numbers of mapped genes. This information is required to test whether the genetic control of gene expression is cis- or trans-acting. Therefore, unless the genes have been genetically mapped previously (Fig. 3), it is not possible to distinguish between these two modes (cis/trans) of regulation. Genetical genomics studies are also constrained by the same limitations as traditional QTL studies in terms of population and allelic diversity range surveyed. However, unlike traditional QTL studies, the genetical genomics strategy permits the rapid identification of candidate genes that underlie a QTL, assuming that transcription profiles are available for the segregating populations. Genetical genomics may also be a particularly valuable approach for detection of genes that are important for hybrid breeding, allowing identification of genes where alternative alleles are fixed in one or both hybrid species. For these loci, the same QTL is expected to segregate consistently in multiple, independent crosses involving the two parental species. These genes are primary targets for breeders attempting to create a favorable combination of alleles in tree improvement programs.

Figure 3.

Likelihood ratio (LR) profiles of the quantitative trait locus (QTL) analysis of the gene expression of S-adenosylmethionine synthase (SAM; black line) and adenosylhomocysteinase (SAH; gray line) measured in a segregating population of Eucalyptus globulus and a hybrid of Eucalyptus grandis ×E. globulus. The genetic map linkage groups (LG1–LG11) are displayed on the x-axis. The genetic map positions of the genes (black triangles) suggest that SAM is cis-regulated, while SAH is controlled in trans by three loci (gray lines; peaks of LR ~10) located in different genomic regions.

VII. Association mapping

Association genetics or linkage disequilibrium (LD) mapping differs from traditional QTL analysis primarily at the level of resolution and allele diversity captured. The approach relies on the genetic analysis of a population with unknown ancestry, but which at some point in time shared a common (unknown) ancestor (Cardon & Bell, 2001). Higher resolution is achieved because lineages derived from the common ancestor may extend for many generations, therefore reducing linkage disequilibrium to neighboring loci. By working with a large sample of unrelated individuals, the breadth of alleles and genes that contribute to the trait variation can also be captured. Fundamentally, the finding of a statistically significant correlation between alleles and phenotype identifies a locus that is a candidate for regulating the trait or a locus in LD with the causal polymorphism. In the few Eucalyptus genes analyzed to date, significant LD has been found for only a few hundred base pairs (Santos, 2005; Thumma et al., 2005; M. Kirst, unpublished; D. Faria and D. Grattapaglia, unpublished) (Fig. 4). If LD decays rapidly (i.e. within a few hundred bases), a significant association would suggests that one or more genes in the close vicinity of the polymorphic site, that is, within a few hundred base pairs, are implicated in the trait variation.

Figure 4.

Rapid decay of intragenic linkage disequilibrium estimated as r2 in the cinnamyl alcohol dehydrogenase (cad) gene in two different Eucalyptus species (D. Faria and D. Grattapaglia, unpublished).

The successful identification of single nucleotide polymorphisms (SNPs) and haplotypes associated with microfibril angle in E. globulus by Thumma et al. (2005) demonstrated the feasibility of association genetic studies in Eucalyptus. Reports in other forest tree species are becoming increasingly common, particularly in pines (Gonzalez-Martinez et al., 2006, 2007) and other conifers, where QTL/positional cloning approaches are prohibitive because of the large genome size. In Eucalyptus and other woody angiosperms with moderate genome size, association genetics represents one alternative to QTL mapping and positional cloning. However, it is questionable how effective association genetics will be for detecting allelic combinations that are explored in hybrid breeding, which is currently the basis of the most successful tree improvement programs in Eucalyptus and several woody angiosperms. While overdominance appears to be a significant driving force behind heterosis in these hardwoods, it is unclear if association genetics will be able to identify interspecific allelic combinations that produce superior phenotypes. Therefore, it will become important to evaluate the use of association genetics in populations that combine pure species and hybrid individuals. Unfortunately, interspecific hybridization is one of the driving forces in the creation of LD. As a result, significant associations detected in an F1 hybrid are meaningless as there will be complete LD among all the loci inherited from each parental species. These issues will probably be of limited importance in conifer association studies, as interspecific hybrid breeding is uncommon and heterosis is not as prevalent as in hardwoods.

The application of association genetics in Eucalyptus and trees in general implies the need to detect polymorphisms in candidate genes that can be tested for association and to assess the level of LD around these polymorphisms. However, focus on candidate genes may become unnecessary in the near future as new high-throughput genotyping technologies are developed for discovery of polymorphisms in large numbers of genes and their testing for associations at a limited cost (see Section X: Future developments and challenges). Also essential is the assessment of any genetic structure that may exist in the association population as this may lead to detection of spurious, false-positive associations (Knowler et al., 1988; Pritchard et al., 2000). In this section we discuss the potential for carrying out association genetic studies in Eucalyptus species, in natural and breeding populations, and the expected pitfalls and limitations of the application of this strategy.

1. Nucleotide diversity and linkage disequilibrium

Initial estimates of nucleotide diversity in a few Eucalyptus species suggest that levels of polymorphism are adequate for association genetic studies. Nucleotide diversity is similar to or higher than that detected in maize (Zea mays) and Populus. In the analysis of partial sequences of S-adenosylmethionine synthase, ferulate 5-hydroxylase and cinnamyl-alcohol dehydrogenase in ~16 genotypes of E. globulus we detected values for nucleotide diversity (π) of < 1%, in coding regions and untranslated regions (UTRs) (M. Kirst, unpublished). A twofold higher nucleotide diversity was seen in a 440-bp sequence of the Caffeoyl CoA O-methyltransferase gene for E. grandis (π = 0.00356) when compared with E. globulus (π = 0.00168), possibly reflecting the wider geographical distribution of the former species (Santos, 2005). Thumma et al. (2005) described similar levels of diversity in cinnamyl-CoA reductase in E. globulus and also a rapid decay of LD, attaining an r2 of 0.2–0.3 in a few hundred to a thousand nucleotides. With such rapid LD decay, the resolution should be adequate for identification of specific genes as regulators of phenotypic traits in Eucalyptus association studies. Despite this expectation, it is now clear from work in the model plant Arabidopsis thaliana that LD decay and nucleotide diversity are highly variable in the genome, reflecting evolutionary and natural selection effects at specific loci (Clark et al., 2007; Kim et al., 2007).

The studies described above have been limited to a few Eucalyptus species and loci. It is unclear if such levels of polymorphism and LD will also be found in all commercially relevant species, including those that occur as small populations, or that may have originated from relatively recent admixture events. Small effective population size contributes to lower nucleotide diversity (as a result of genetic drift) and admixture leads to higher LD. Selective sweeps, particularly around loci that may be essential for adaptation of some Eucalyptus species to particular biotic and abiotic stresses, also lead to high levels of LD, at least in some areas of the genome. Although typically localized, these regions of high LD may be precisely the ones that are being targeted by breeders for gene identification through genetic association analysis (e.g. for freezing tolerance).

Another factor that creates LD is population admixture, but a similar effect occurs in the event of species hybridization followed by speciation. Interspecific hybridization leads to complete LD among alleles fixed in the two species, in the first generation. Although hybridization does not occur among species in different subgenera (Griffin et al., 1988), natural hybrids from species within a subgenus can be observed in nature and are relatively common (Griffin et al., 1988; Potts & Dungey, 2004). In the event of one Eucalyptus species having originated from a relatively recent hybridization event, one may expect significantly higher levels of LD among loci. The rate of LD decay will be dependent on the frequency of recombination and the number of generations, but is likely to remain high for many generations. These effects may also be detected in commercial Eucalyptus breeding programs which frequently rely on the transgressive segregation of hybrids.

It has been suggested that historical interspecific hybridizations between tree species could be useful for detecting genes related to adaptation using admixture mapping. This possibility was evaluated in Populus, primarily because of the availability of well-characterized hybrid zones, where Populus alba and Populus tremula have been hybridizing for many generations (Lexer et al., 2007). While it is a useful approach in those circumstances, we believe that admixture mapping will have minor utility in Eucalyptus genomics. The primary reason is that, although natural hybrids have been observed, it is unclear whether they are common and widespread enough to be useful for admixture mapping. Alternatively, artificially managed admixed populations could be created. However, even with advanced breeding techniques that shorten rotation cycles to only a few years, creating such a population would require several generations to achieve adequate LD for pinpointing genes that control quantitative traits. Furthermore, admixture mapping depends on the availability of genome-wide panels of markers informative for ancestry between species. Microsatellite allele frequency differentials and estimates of Fst amongst provenances within species (Steane et al., 2006) or even between tropical and temperate plantation species that belong to different taxonomic sections such as E. grandis and E. globulus (Kirst, 1999; D. Faria and D. Grattapaglia, unpublished) are typically modest. It may therefore be challenging to develop a robust and large set of ancestry-informative markers for Eucalyptus, especially between species or provenances that are closely related. Considering the rapid development in technology for DNA sequencing and SNP genotyping, we expect that within the time period needed to develop such AIM, whole-genome association analysis based on several hundred thousand markers will be far more efficient in defining genes that regulate complex traits.

2. Population structure – natural and breeding populations

Population genetics studies in Eucalyptus have demonstrated that most species have very limited population structure. Low Fst values (~0.10) have been reported with nuclear biparentally inherited markers in E. grandis, E. globulus and E. urophylla, which are species with wide geographic distributions and large population sizes (House & Bell, 1994; Kirst, 1999; Steane et al., 2006; D. Faria and D. Grattapaglia, unpublished). Even between two geographically separated populations of E. grandis located in Atherton (Queensland) and Coffs Harbor (New South Wales) the Fst was estimated at 0.06 based on 33 microsatellites. Similarly, a low Fst (0.05) was detected between two disjunct E. globulus provenances, one continental (Victoria) and one from Flinders Islands (D. Faria and D. Grattapaglia, unpublished). Limited structure among populations is expected in these species considering their outcrossing nature and wide pollen and seed dispersal. As a consequence, identification of spurious associations as a result of population structure, which has plagued studies in humans and some crop species (Knowler et al., 1988; Pritchard et al., 2000), has typically not been considered a major concern in Eucalyptus association studies. However, exceptions may exist and species and populations will have to be considered individually. For instance, population structure measured on the basis of the JLA region of the maternally inherited chloroplast DNA indicates much higher levels of population differentiation in E. urophylla (GST = 0.58) than in E. grandis (GST = 0.30) (Jones et al., 2006; Payn et al., 2007). Eucalyptus urophylla is widely distributed in disjunct populations that occupy the islands of Indonesia and Timor. Population isolation may contribute to inbreeding, and, in the long term, population differentiation (Hartl & Clark, 1997). Other unknown factors related to the history of the natural populations (e.g. genetic bottlenecks and migration) may also contribute to higher population differentiation and are often unknown.

Association populations created from natural stands may represent a good source of germplasm for association studies. However, most of these studies will be carried out in the context of existing breeding programs developed by companies and research institutions. Many of these breeding programs are derived from seed collected and distributed directly from Australia in the late 19th century, or from species and provenance tests. The population structure – if any – of the material utilized in these programs is in most cases unknown, but could be significant. A lack of source data also suggests that some inbreeding may be taking place within these programs without the breeders’ knowledge. For instance, a structure (Pritchard et al., 2000) analysis and assignment test of a group of elite clones used in operational plantation forestry in Brazil showed that most of them were assigned with high confidence to E. grandis and E. urophylla with approx. 50% probability for each parental species, corroborating their expected very recent hybrid origin (D. Faria and D. Grattapaglia, unpublished). Failure to recognize those differences in an association study will lead to identification of false-positive associations.

VIII. Molecular breeding

1. Perspectives of marker-assisted selection

Considering the rapid decay of LD reported for Eucalyptus, the marker-trait association detected in one segregating population cannot be assumed to hold in a different pedigree. As a result, MAS based on long-range marker-trait association detected by QTL analysis may only be useful for within-family selection. This is in principle a limitation, considering that most Eucalyptus breeding programs attempt to evaluate a broad diversity of families to capture valuable allelic combinations in the progeny. However, Eucalyptus breeding strategies vary broadly according to the target species or hybrid, the possibility of deploying clones, and the amount of resources available to the breeder, so that within-family early marker-assisted selection might in fact be a viable and valuable tool.

A reasonable premise is that MAS will only be justifiable when the program has already reached a relatively high degree of sophistication, fully exploiting all the accessible breeding and propagation tools. Breeding for hybrid performance combined with clonal propagation of selected individuals is currently the method of choice for extracting new elite clones (Rezende & de Resende, 2000; Potts, 2004; Potts & Dungey, 2004; Bison et al., 2007). Progeny trials, together with expanded single-family plots where larger numbers (> 1000) of full-sibs are deployed per family, are used for very intensive within-family selection. Vegetative propagules are rescued from selected trees and used for establishing clonal trials. This strategy involves a significant amount of time and effort being devoted to clonal testing before effective recommendations can be made regarding new, operational clones. At the same time this breeding scheme generates large amounts of LD, a favorable condition for MAS in forest trees (Strauss et al., 1992). Furthermore, numerous reports have indicated that dominance variation is a key component to explain the superiority of tropical hybrid Eucalyptus clones, especially for growth (Bouvet & Vigneron, 1996; de Assis, 2000; Rezende & de Resende, 2000). This and other nonadditive sources of genetic variation can be captured by vegetative propagation in Eucalyptus, a process that is not yet fully operational in conifers. Therefore, genomic segments containing favorable alleles or allele combinations segregating within families could be efficiently tagged with microsatellite markers in strong LD with the actual causative polymorphisms in a QTL mapping stage, and later used for within-family MAS selection for superior individuals (O'Malley et al., 1994). QTL linked markers would allow very early selection (seedling stage), reducing the time required to identify elite trees, especially for traits related to wood properties, and reducing the number of trees to be selected, propagated, and advanced to clonal trials (Grattapaglia, 2008; Fig. 5). Given their relatively short rotations and the ability to capture nonadditive genetic variation, eucalypts in fact may be the first forest tree species where this kind of MAS scheme could be applied.

Figure 5.

Strategy for early within-family marker-assisted selection (MAS) of superior (plus) trees in a Eucalyptus clonal selection program. Quantitative trait locus (QTL) mapping phase: QTL mapping is carried out and flanking markers in strong linkage disequilibrium with favorable alleles or allelic combinations at major-effect QTLs are identified within-family. MAS phase: QTL linked markers are genotyped in an expanded number of progeny individuals of the same family to increase within family selection intensity at very early age for late expressing traits such as wood properties (modified from fig. 10 in Grattapaglia, 2008; with kind permission from Springer Science).

Although the costs of genotyping have dropped significantly in recent years, DNA analysis is still costly. The most likely cost-beneficial application of MAS in Eucalyptus will be for traits that provide significant added value to the final product, such as branching habit (for solid wood) or wood chemical and physical traits, or for traits that facilitate clonal deployment, such as adventitious rooting or somatic embryogenesis response. Among all possible quality traits, the preference would be for those that display medium to high heritabilities but where phenotype assessment is difficult, expensive, or requires waiting until the tree reaches maturity. Wood quality traits typically require the tree to start accumulating mature wood and involve relatively lengthy procedures for phenotypic evaluation in the laboratory. These kinds of traits could be interesting targets for MAS in Eucalyptus, given that the costs of genotyping are sufficiently competitive, and precision is high when compared with direct phenotype measurements. It is important to point out, however, that with the recent developments of fast sampling and indirect wood chemistry measurements, the potential gain will be realized only on the basis of the time savings provided by very early selection.

2. Breeding using transgenic technology

Transgenic technology is undoubtedly a powerful complementary tool available to the molecular breeder. Considering that industrial Eucalyptus forests are almost exclusively clonal, transgenics are likely to have an increasing role not only in wood quality improvement but in resolving problems related to pest and pathogen susceptibility and/or abiotic stress tolerance (e.g. tolerance to frost and drought) that might limit the expansion or survival of existing plantations, as in the case of annual crops. The introduction of genes that confer traits that do not display variation within the Eucalyptus gene pool, or are impossible to produce by natural recombination processes, might radically modify the ways in which forests are planted or in which forest products are derived (Grattapaglia, 2008).

The development of efficient transgenic technologies for Eucalyptus would represent a key step for functional genomics studies. Gene tagging through insertional mutagenesis approaches is a particularly appealing method in heterozygous trees that cannot be easily selfed, as it would generate dominant phenotypes. In Eucalyptus, the use of such an approach as a practical tool for genome-wide gene characterization and breeding faces the same kinds of logistical and biological obstacles as found in Populus, but these might be overcome by large-scale collaborative efforts (Busov et al., 2005). To mitigate the biological and biochemical limitations to the study of wood formation in trees using mutant phenotypes, in vitro wood formation systems can be employed to introduce transgenes transiently or stably into growing wood-producing tissue. Such a system has been developed in Eucalyptus (Spokevicius et al., 2005) and recently used to show that b-tubulin determines cellulose microfibril orientation during xylogenesis in E. globulus (Spokevicius et al., 2007).

In spite of the recognized economic importance of Eucalyptus in world forestry, very little has been published on transgenic experiments in species of the genus; most of this work has been summarized by Poke et al. (2005). While Eucalyptus tissue can be transformed by Agrobacterium tumefasciens, major difficulties are faced in the regeneration step. Several reports have documented the production of transformed callus, tissue, and root organs; however, reports on transformed plants are scarce (Machado et al., 1997; MacRae & van Staden, 1999). A marked genotype effect has been observed on the efficiency of regeneration and, consequently, stable transformation. This fact has prompted several groups to first identify Eucalyptus‘lab rats,’ that is, easily regenerable genotypes, and only then to develop improved protocols to generate large numbers of independent transformation events (Tournier et al., 2003). These ‘lab rats’ are not yet available in the public domain, and this is also the case for most of the transformation protocols. However, recently, Chen et al. (2006) described a basic Agrobacterium-mediated genetic transformation protocol employing organogenesis for the production of transgenic plants using Eucalyptus camaldulensis. Efficient proprietary transformation protocols have also been developed in Japan, where an E. camaldulensis‘lab rat’ has been used (Kawazu et al., 2003), as well as in different laboratories in the USA, where E. grandis and E. urophylla have been used (M. Hinchee and V. Chiang, pers. comm.).

The current information on Eucalyptus transgenesis points to a very promising future as far as the technical possibility of generating stably transformed Eucalyptus plants is concerned. However, some strategic issues regarding the adoption of transgenic technology for wood quality manipulation in Eucalyptus should be taken into account, including (a) the magnitude of the gain and cost/benefit relationship obtained by manipulating lignification or cellulose genes when compared with the exploitation of the wide natural genetic variation in the genus; (b) the specific biosafety and intellectual property issues relevant to transgenic eucalypts and the time and investment necessary to resolve them in order to actually be able to plant transgenic trees on a large scale; (c) the speed at which breeding programs generate new and better clones for adaptive traits (growth, wood properties, pest resistance, clonability, etc.) compared with the time needed for regulatory approval of every new transgenic clone; (d) the lifespan of a patent in local regulation as compared with the time needed to effectively make returns on the patent from the planted forest before the patent goes into the public domain; (e) the market issues that the company has to consider in adopting transgenics, in relation to both public perception and forest certification processes. All these and other issues will have to be carefully considered, including the fact that, just as occurred in annual crops such as soybean (Glycine max), maize, and cotton (Gossypium spp.), the use of transgenics could produce a major technological divide and become a necessary condition for the continuing competitiveness of forest-based industry worldwide (Grattapaglia, 2008).

IX. From gene sequences to breeding tools

The successful application of gene sequence information to Eucalyptus breeding has not yet been demonstrated, but this may change in the near future. The expected completion of the genome sequence of E. grandis by 2010 and the use of complementary or novel quantitative genomic approaches (e.g. genetical genomics and association genetics) may rapidly lead to such applications, much as has been the case in A. thaliana and some agricultural crops. The first example of the identification of SNPs associated with wood microfibril angle was reported in E. globulus (Thumma et al., 2005) and SNPs have also been identified for several wood property and physiological traits in pines (Gonzalez-Martinez et al., 2006, 2007) Not surprisingly, the identification of significant associations has generally occurred for traits with high heritability. In none of these cases, however, has the biological cause of the SNP–trait association been demonstrated, or the gene function verified by alternative approaches.

The strategy to be adopted for successful identification and introgression of valuable alleles discovered through genomic analysis into Eucalyptus breeding programs will be dependent primarily on (1) the successful identification of polymorphisms associated with traits of interest, (2) the frequency of superior alleles in the base breeding population, and (3) their phenotypic effect. We have discussed current and new approaches to the detection of genetic variants of interest (see Sections VI and VII on QTL and association mapping). Here we explore likely scenarios regarding the other two factors – allele frequency and phenotypic effect – considering the life-history characteristics of Eucalyptus. In the last sections we discuss how current technology and future developments may change the outcome of genomic research targeted to assist Eucalyptus breeding in the next few decades.

1. Superior alleles with high frequency in the breeding population

In plants, association genetic studies have typically only targeted the analysis of SNP loci where the frequency of the minor allele is relatively high (> 10%). Lower frequencies, much like small population sizes, severely hamper the statistical power of detection of association, particularly for traits with low heritability (Long & Langley, 1999). SNPs identified in association with quantitative phenotypes will probably be of small effect, explaining less than 5% of the phenotypic variance, much as has been observed previously in loblolly pine (Pinus taeda) (Gonzalez-Martinez et al., 2007) and recently by Thumma et al. (2005) in E. globulus. Therefore, it is unlikely that traditional breeders will, at least in the short term, make selection decisions based on such limited gain. Because of its recent domestication coupled to the capacity for clonal selection, every breeding cycle in Eucalyptus still produces gains that are almost one order of magnitude higher for growth traits (Snedden & Verryn, 2004). As a result, at least initially, marker-assisted breeding will only serve to complement traditional breeding. We anticipate that this scenario will remain unchanged until ‘whole-genome’ association approaches – where almost all polymorphisms with an effect on a phenotype can be detected – are developed.

2. Low-frequency alleles of large effect

It has been generally considered that alleles at low frequencies (< 5%) in the breeding population will be of limited use for MAS. Such alleles may not significantly shift population means and therefore do not have great relevance at the population level – very few individuals carry them, particularly in a homozygous state. However, increasing evidence from human association genetic studies indicates that rare alleles can have a tremendous impact in an individual's phenotype (Romeo et al., 2007; Topol & Frazer, 2007). This effect will probably be accentuated in Eucalyptus. Eucalyptus species are generally outcrossers, and have large effective population sizes, a broad geographic distribution and wide pollen and seed distributions. Therefore, rare mutants are frequently maintained at low frequencies in the population for a large number of generations, particularly if they are selectively neutral. The broad genetic and phenotypic diversity of Eucalyptus natural populations is likely to contain an abundance of unique haplotypes and low-frequency alleles; some of these alleles may be the gems that breeders are seeking for MAS.

A recent survey of a relatively small number of Eucalyptus haplotypes identified almost 300 large-effect SNPs – that is, polymorphisms that lead to the introduction or removal of a STOP codon – in transcribed sequences (E. Novaes, unpublished). Therefore, the largest breeding gains from MAS may be achieved not through the continuous incorporation of frequent alleles of small effect, but rather through the identification of rare, high-value alleles. The potential impact of such alleles in tree breeding programs is exemplified by a rare loblolly pine mutant, the cinnamyl alcohol dehydrogenase cad-null allele, which has been associated in some studies with the superior character of that genotype in terms of growth and wood properties (Mackay et al., 1997; Ralph et al., 1997; Wu et al., 1999). In species with the life history of Eucalyptus, where the probability of maintaining rare alleles in the population is high, it is possible that the majority of the superior phenotype detected in some individuals may be attributable to such uncommon allelic variants. Obviously, identifying those alleles and increasing their frequency to usable levels in a breeding population will be challenging. Furthermore, in a Eucalyptus hybrid breeding/clonal deployment situation, where the magnitude of nonadditive variation becomes important, the issue will be not only the detection of rare alleles of large effects but also the detection of rare allelic combinations both within and between loci. Those alleles and allele combinations are unlikely to be uncovered through association genetics because very few individuals will carry them, but they may be detected by an integrated sequential QTL analysis and genetical genomics approach. Nonetheless, capturing all the rare alleles with important additive and dominance effects in a population would require a QTL/genetical genomic analysis of a large number of crosses in a diallel mating design (which is too costly with the current technology). An alternative would be to create crosses that involve a subset of the breeding population that encompasses the majority of the genetic variation in the species, and carry out analysis of only those populations. This approach is currently being pursued in maize by a Nested Association Mapping (NAM) strategy where, in parallel to the development and analysis of an association population, 26 recombinant inbred line populations were generated. Each population was derived from the cross of a ‘reference’ common parent to each one of 26 diverse founder lines that best represents the genetic variability of the association population (Yu et al., 2008). One of the discoveries expected from this series of QTL populations is the detection of alleles that may be unique to one of a few maize lines, which segregate in the progeny and have an impact on the phenotypic variation. Although not all important rare alleles with significant additive effects will be detected using this approach, it will permit evaluation of how frequently these alleles are missed in an association population. Maize has many similarities with Eucalyptus and other tree species in terms of high nucleotide diversity and low LD, largely because of rare alleles.

3. Incorporation of species-specific high-value alleles

Different Eucalyptus species are recognized for their superior characteristics in terms of growth, wood quality and resistance to biotic and abiotic stress. For instance, E. nitens is highly tolerant to frost damage; E. grandis has superior adaptability and growth properties, and E. globulus has better wood properties. These species-specific properties are likely to be associated with alleles that are fixed or at very high frequencies in each species and may provide a selective advantage in the natural range of the species for individuals carrying them. In the short term it is possible that the most significant gains from MAS may be achieved through the introgression of valuable species-specific alleles into the breeding populations. As described previously (QTL analysis and genetical genomics), those alleles of large effect may be relatively easy to detect in the analysis of hybrid segregating populations. However, association genetic approaches may be hampered by spurious associations resulting from population structure. Although genetic structure is relatively modest among Eucalyptus species, there will be very significant genetic differentiation among species for those genes that have become fixed in any one species. Any allele fixed in one species may be identified as significantly associated with traits that are superior in that species.

The incorporation of species-specific superior alleles in any breeding program is also likely to be problematic, at least through traditional breeding. In the first generation following hybridization one should expect complete LD among alleles that are fixed in the two species and contribute to the hybrid genome. LD decays as a function of the recombination rate, effective population size and number of generations. Recombination rates are not constant in the genome (Kim et al., 2007), and some evidence suggests that lower rates occur in hybrids because of recombination suppression in parts of the genome that have diverged (Bradshaw & Stettler, 1994; Bradshaw et al., 1994). Therefore, transgenic approaches may be the best approach for rapidly and efficiently incorporating these alleles into the breeding program.

X. Future developments and challenges

One of the major goals of genomics is to describe all levels of genetic information – from DNA sequence, to mRNA, to protein – and develop predictive models that define phenotype based on genetic, epigenetic, developmental and environmental properties. Therefore, a complete description of these networks would also imply full knowledge of the contribution of epigenetic and environmental factors to the phenotype. We are still a long way from achieving this goal for any phenotype in any organism, but technological developments, particularly in genome sequencing and transcription analysis, have advanced tremendously since the turn of the century. Comprehensive databases of gene expression have already been created for Populus, based on whole-genome microarrays and ESTs (Sjödin et al., 2006). This and other plant transcriptome databases (http://www.arabidopsis.org/portals/expression) allow any researcher to obtain gene expression information for almost any gene in a broad spectrum of tissues, developmental stages, growth conditions and genotypes. Similar advances in DNA sequencing technology may translate into the description of the genome sequence for any individual with a moderate genome size, such as Eucalyptus, during the next decade.

The US National Institute of Health request for proposals in revolutionary genome sequencing technologies (The $1000 Genome) was launched in 2004. The ‘goal of this initiative is to reduce costs by at least four orders of magnitude, so that a mammalian-sized genome could be sequenced for approximately $1000’ within the next 10 yr. Since then, novel sequencing methods have led to a reduction in cost of at least one order of magnitude (Margulies et al., 2005). Already, sequencing of an average Eucalyptus genome (~600 Mbp, 12X coverage) could be achieved using 454 Life Sciences (GS FLX) for less than $1 million. This cost is expected to be rapidly reduced further through use of the established technology. In the short term, approaches based on single-base addition methods (SBA) promise to deliver over several Gbp of sequence at a cost of a few thousand dollars (http://marketing.appliedbiosystems.com/mk/get/SOLID_KNOWLEDGE_LANDING). Other methods still in development may be able to deliver complete genome sequences of relatively large genomes within the next decade. If the $1000 human genome becomes feasible it can be expected that sequencing any Eucalyptus genome will be achievable at some fraction of that cost, creating the ideal scenario for the effective integration of molecular technologies into classical breeding programs.

As a first evaluation of these new sequencing technologies, our laboratory has recently completed the sequencing of 1 million short expressed sequence tags derived from a pool of tissues collected from 21 partially unrelated genotypes of E. grandis (E. Novaes, unpublished). The sequencing approach was based on massively parallel pyrosequencing, which can produce several hundred thousand short sequencing reads (~100–250 nucleotides) in a single assay. The purpose of our specific study was the discovery of genes and simultaneous identification of SNPs based on sequence alignments among unrelated haplotypes (Fig. 6). One limitation of this study was that LD and nucleotide diversity could not be estimated because the identity of each EST could not be related back to each E. grandis genotype. However, one may envision a scenario where cDNA material from individuals is sequenced either in separate reactions or utilizing some barcoding system that identifies the origin of each sequence (Parameswaran et al., 2007), permitting the assessment of individual haplotypes.

Figure 6.

Single nucleotide polymorphism (SNP) discovery by massively parallel sequencing of normalized cDNA from unrelated Eucalyptus trees. (a) SNPs identified in the Eucalyptus grandis homologs of the lignin gene caffeoyl-CoA 3-O-methyltransferase. The reference nucleotide at an SNP is defined by the most frequently observed allele. In this analysis, at least two reads containing the reference and alternative SNP nucleotide, in both forward and reverse directions, were required to declare the presence of an SNP. (b) Alignment of 15 sequences (eight forward and seven reverse sequences) that contain the alternative alleles for the SNP located in position 605 (box). The gray shading indicates other variable positions that have not been declared polymorphic because of the limited number of sequences to support them.

SNP discovery and validation in Eucalyptus is currently limited to identifying SNPs in silico in EST assemblies and resequencing in relevant germplasm. However, standard approaches used to produce automated, unsupervised alignments (e.g. Phrap; Lee & Vega, 2004) and to detect polymorphisms (e.g. PolyPhred and PolyBayes; Nickerson et al., 1997; Marth et al., 1999) are very limited when applied to highly diverse species such as Eucalyptus. To address the limitation of these software, new bioinformatics tools that are designed specifically for aligning and identifying SNPs in genetically diverse species are being developed (PineSAP; J. L. Wegrzyn, pers. comm.). Also, although SNPs are highly abundant in Eucalyptus, the lack of inbred lines or haploid tissue has rendered the detection of polymorphisms by resequencing almost intractable because of the very high frequency of indels which complicate direct sequencing from PCR without cloning. With the trend towards genome-wide association analyses, there is an immediate need for robust, inexpensive, high-density and high-throughput markers. An approach that would allow parallel genotyping of the Eucalyptus gene space for sequence polymorphisms is highly desirable. Given the very high levels of sequence polymorphism in the Eucalyptus genome, Diversity Array Technology (DArT) (Jaccoud et al., 2001; Wenzl et al., 2006) could provide several thousand markers enriched for low-copy regions to allow a first move toward selection on a genome-wide scale (Meuwissen et al., 2001) in specifically tailored small MAS Eucalyptus breeding populations. Unpublished data from an interspecific (E. grandis×E. globulus) expression profiling experiment carried out in the Genolyptus project (G. Pasquali, unpublished) on a 385 000 50mer Nimblegen oligoarray platform, representing 21 403 Eucalyptus spp. unigenes, was explored for putative single feature polymorphisms (SFPs) (Borevitz et al., 2003). By analyzing the contrasts of least square means for probe vs. species interaction we found 32 077 probes in 10 781 genes that differentially hybridized to E. globulus and E. grandis, and are thus candidate segregating markers in crosses between these two species. Based on these exploratory data we recently validated the inheritance and segregation of SFPs using a pseudo-test-cross configuration in two segregating families and a microarray with 15 180 short probes covering 1518 candidate genes involved in wood formation (10 probes/gene). A total of 1137 probes representing 658 genes were deemed polymorphic between the parents and segregated in the progeny sets, representing an efficiency of 43% for genetically mapping genes (D. Grattapaglia, unpublished). The expansion of this experiment will allow linkage mapping of several thousand genes and provide opportunities to verify co-localization with QTLs, thus supplying positional candidates to be tested in association genetics experiments.

Within the next decade we can expect the improvement of current SNP discovery, genotyping and sequencing methods, to the point that it will be feasible to identify all the sequence variants that occur at moderate to high frequencies in any Eucalyptus breeding population. The strategy will probably rely on ultra-high-throughput shotgun sequencing of each individual tree genome as has been proposed for human genomes (Mardis, 2008). With current technologies, assembling would still require a high-quality reference sequence and deep enough coverage so that the two homologous chromosomes of a heterozygous tree can be tentatively discriminated and separately assembled. Currently in progress, the complete sequence of the first Eucalyptus genome (E. grandis, clone BRASUZ1) will represent a milestone achievement and serve well as a first reference sequence for future genomic undertakings. At that time the main limitation to moving from genotypes to phenotypes will probably be the availability of appropriate material, sufficiently replicated across field sites and precisely phenotyped for traits of interest. Therefore, for short-rotation woody crops such as Eucalyptus, it is essential that development of these populations be initiated within the next few years. Research institutions and industry with such materials readily available will potentially achieve the greatest gains when the technology becomes available. Given the extraordinary genetic variation that exists in the genus Eucalyptus, the ingenuity of most breeders and the powerful genomic tools that have become available, the outlook for applied genomics in Eucalyptus forest production is encouraging.

Acknowledgements

DG acknowledges support from the Brazilian Ministry of Science and Technology, the participating forestry companies in the Genolyptus project and the Brazilian National Research Council (CNPq) as well as the students, research collaborators and breeders for continued discussions and scientific input. Special thanks are due to Danielle Faria for providing access to yet unpublished work. MK acknowledges support from the US Department of Energy and National Science Foundation and the Consortium for Plant Biotechnology Research, as well as Evandro Novaes and Derek Drost for assistance with the preparation of the data presented here.

Ancillary