While most previous QTL mapping studies of diameter growth and wood density in Eucalyptus[12, 16–18, 40, 41] were based on F1 hybrid pedigrees, the present study focused on trait dissection in an F2 interspecific backcross pedigree. QTL mapping in the shared F1 hybrid parent and the E. grandis and E. urophylla BC parents allowed assessment of the architecture of interspecific as well as intraspecific genetic variation affecting trait variation. In this approach, fixed genetic differences between the parental species are likely to be in heterozygous state in the F1 hybrid and segregate in either or both backcross families depending on the degree of dominance. In addition, genetic factors that are heterozygous in the backcross parents (i.e. intraspecific variation) also segregate in the backcross progeny. If fixed genetic differences between the pure-species parents were purely due to additive genetic effects, the majority of QTLs in the F1 hybrid would segregate in both backcross families. However, the majority of QTLs in the two F1 hybrid maps were detected in only one of the two backcross families. Only one QTL for DBH on LG6 in the F1 hybrid parent was shared in both backcross families (Additional file 3: Figure S1). Failure to detect QTLs segregating from the F1 hybrid in both backcross families may be due to dominance effects (Additional file 8: Figure S4) playing a significant role in the expression of QTLs in alternative genetic backgrounds [42–44], or may be the result of epistatic interactions , or due to differences in QTL effects for the same alleles segregating in different genetic backgrounds (i.e. in the presence of a different set of segregating QTLs). For example, we identified a significant epistatic interaction between the wood density QTLs on LG8 and LG10, and between the wood density QTLs on LG2 and LG8 in the F1 hybrid map (E. urophylla BC family). This may explain why we detected the wood density QTLs on LG2 and LG8 in the E. urophylla BC family only (Additional file 3: Figure S1).
Overall, more QTLs were identified in the F1 hybrid parent (14) than in the backcross parents (3) for DBH (three compared to one in the BC parents) and density (eleven compared to one in the respective backcross parent, Tables 2 and 3). The majority of the positive QTL effects for DBH in the F1 hybrid were associated with the E. grandis allele and most of the positive QTL effects for density were associated with the E. urophylla allele. This is congruent with the expected interspecific and intraspecific genetic variation segregating in the backcross families (Table 1, Figure 1). The number of QTLs detected for DBH in Eucalyptus has generally been lower than that detected for wood density (Additional file 9: Table S4). The lower number of QTLs identified for DBH in this study (Tables 2 and 3) is consistent with published QTL reports, reflecting the lower heritability associated with growth traits compared to wood density in Eucalyptus and the limited statistical power to detect larger numbers of small effect QTLs. The well-described Beavis effect  certainly also applies in our study which means that some QTL effects listed in Tables 2 and 3 may be inflated and we fully expect that more QTLs of lower effect would be detected if our mapping populations were to be expanded.
DBH and density QTLs were detected in different regions of the genome (Additional file 3: Figure S1) suggesting that the two traits are affected by independent polymorphic loci in this pedigree. This is further supported by the low phenotypic correlation observed between DBH and wood density in the two backcross families (0.06 for E. grandis BC and −0.02 for E. urophylla BC). Freeman et al. also identified independent QTLs for DBH and wood density in an F2 outbred pedigree of E. globulus. However, other QTL studies in Eucalyptus have identified co-located QTLs affecting DBH and wood density in addition to independent QTLs [12, 40]. This could be explained by the occurrence of different polymorphisms affecting the two traits in each mapping pedigree resulting in different levels of correlation reported for DBH and density in previous studies [46–50]. The identification of independent QTLs for DBH and wood density in this study suggests that MAB could be used to improve growth and wood density simultaneously in this hybrid pedigree by selecting for combinations of QTL alleles with positive effects on DBH as well as wood density.
Comparative genetic mapping facilitates the identification of QTLs across different environments, ages and in different genetic backgrounds. Previous comparative genetic mapping studies in Eucalyptus suggested high levels of synteny and co-linearity among the genomes of eucalypt species [37, 38, 51, 52] enabling the comparative analysis of QTLs in different species [14, 16, 17, 40, 41, 51]. QTLs for DBH and wood density were detected on homologous linkage groups in the parental maps in this study (Additional file 3: Figure S1) and E. globulus and E. nitens linkage maps in previous studies (Additional file 10: Table S5 and Additional file 11: Table S6). QTLs identified for wood density on LG1, LG6 and LG10 of the F1 hybrid map may correspond to wood density QTLs previously identified on the same linkage groups in E. globulus, while QTLs identified for wood density on LG6 and LG8 of the F1 hybrid map may correspond to wood density QTLs identified in E. nitens. A QTL identified for wood density on LG9 of the F1 hybrid map (E. urophylla BC) may correspond to a wood density QTL previously identified in E. nitens. Similarly, QTLs identified for DBH on LG6 and LG10 (Table 2) of the same parental map may represent the same genomic regions as DBH QTLs reported in E. nitens and E. globulus, respectively. Common regions affecting trait variation across species should be the priority targets for the identification of candidate genes, the development of gene-based markers, association genetic studies and eventually MAB. We expect the resolution of comparative QTL analysis to drastically improve with the use of large numbers of trans-specific and trans-pedigree markers such as microsatellite, DArT and SNP markers linked to the E. grandis reference genome sequence.
An advantage of MAB in trees is the early selection of seedlings, reducing the time and cost normally involved in growing trees to maturity in the field before being able to identify elite trees . Experiments in crop plants have indicated that major effect QTLs and candidate genes associated with these QTLs are more reliable for MAB . Most of the QTLs detected in previous studies in Eucalyptus have likely been from the top end of the distribution of segregating QTL effects, some of which could be considered major effect QTLs [12, 14–17, 40]. However, there is a bottleneck between mapped QTLs and gene discovery mainly due to the low resolution of QTL mapping in populations of only several hundred individuals. To extend the information of QTL mapping, genetical genomic approaches  have been used to identify positional candidate genes and regulatory networks underlying phenotypic variation in several plant species [13, 22–24, 27]. In this study, the majority (70.8%, Additional file 6: Figure S3) of eQTLs identified for 294 xylem expressed genes underlying a major wood density QTL on LG9 (F1 hybrid map, Figure 2, Table 3), did not co-locate with the physical positions of the genes (i.e. were trans-acting eQTLs). The trans-eQTLs detected for these genes most likely correspond to diverse regulatory factors controlling the expression of the genes located in the QTL interval on LG9 one (or more) of which could harbor the trait altering polymorphism underlying the wood density QTL. eQTLs co-localizing with the physical genome position of the gene (cis-eQTL; 29.2%) were identified for only 86 genes in the interval, which is in agreement with the lower proportion of cis-eQTL previously reported for Eucalyptus (22%) in an interspecific backcross population of E. grandis and E. globulus and more recently for Populus (23%) using whole-genome microarray analysis in an interspecific hybrid pedigree .
Schadt et al. reported that genes whose transcript levels are correlated with trait variation could be considered potential candidate genes for the trait. In the present study the transcript levels of ten genes located in the wood density QTL interval on LG9 were found to be positively correlated (R
2 > 0.4) with wood density variation (Table 4). Some of these genes encoding a nucleotide-regulated ion channel family protein (DND1), histidine kinase (HK), S-adenosyl-L-methionine-dependent methyltransferase (SAM), an auxin efflux carrier family protein and calcium-binding EF-hand family protein, have previously been reported to be involved in plant growth and development and cell wall biogenesis. The transcript abundance of three of these genes (DND1
HK1 and a gene encoding a calcium binding EF-hand family protein) was affected by trans-acting eQTLs on LG10 (Table 4). Importantly, the same genomic region on LG10 (51 cM to 74 cM) co-localized with a major wood density QTL on the same linkage group suggesting that this genomic region may contain trans-acting factors affecting wood density as well as the transcript abundance of the candidate genes underlying the wood density QTL on LG9 (Figure 2). DND1 has been shown to be involved in plant defense responses in Arabidopsis. HK was reported to act as a cytokinin receptor  involved in diverse plant growth and developmental processes [57, 58]. SAM is a key enzyme for the phenylpropanoid pathway, involved in the synthesis of lignin . Auxin, essential for plant growth and development (e.g. vascular tissue differentiation, apical development, cell elongation and tropical growth) is transported from cell to cell by auxin efflux carrier proteins [60–62]. Besseau et al. showed that a reduction in the level of hydroxycinnamoyl-CoA shikimate/quinate hydroxycinnamoyl transferase (HCT), a gene involved in lignin biosynthesis, was correlated with the inhibition of auxin transport in Arabidopsis, suggesting that auxin efflux carrier family proteins might be important for cell wall deposition and lignification. Similarly, plant cells contain large amounts of calcium in their cell walls and previous studies highlighted Ca2+ playing a role in secondary cell wall biosynthesis [64, 65].
The observation that the top most positively correlated genes (at the level of transcript abundance) in the wood density QTL interval on LG9 (Table 4) prominently shared trans-eQTLs on LG4 and LG10 (F1 hybrid map, E. urophylla BC family) suggested the presence of trans-acting factors that also underlie wood density QTLs at the same loci. This, together with the detection of a significant epistatic interaction between wood density QTLs on LG8 and LG10, led us to investigate the transcript abundance of genes in the QTL interval on LG8, with the hypothesis that a similar cis-trans relationship would exist between LG8 and LG10 as was observed for LG9 and LG10. We indeed found that the top most positively correlated genes in the wood density QTL interval on LG8 also shared trans-eQTLs on LG4 and LG10 (and LG6, Additional file 3: Table S3a). The top most negatively correlated genes in the wood density QTL intervals on LG 8 and 9 (Additional file 7: Table S3b) did not exhibit such a strong pattern of shared trans-acting eQTLs, but it is formally possible that any of the positively or negatively correlated genes in these two QTL intervals affect trait abundance via a cis-acting and/or trans-acting eQTLs. Together, these findings suggest that at least some of the wood density eQTLs detected in this study may represent segregating components of a transcriptional network (Figure 3). Furthermore, our results suggest that transacting genes (e.g. transcription factors) located in the QTL intervals on LG4 and LG10, together with putative target genes located in the QTL intervals on LG8, LG9 and other identified wood density QTLs should be prioritized for further investigation. Trans-acting factors for which the parental species are differentiated would be heterozygous in the F1 hybrid and could have large effect on gene expression and trait variation in backcross progeny. Cases where transcription factors as well as their target genes segregate may give rise to detectable epistatic interactions as was putatively observed for the wood density QTLs on LG10 (trans-acting) and LG8 (cis-acting).