The genetics and genomics of the drought response in Populus


*(fax +46 90 7866676; e-mail


The genetic nature of tree adaptation to drought stress was examined by utilizing variation in the drought response of a full-sib second generation (F2) mapping population from a cross between Populus trichocarpa (93-968) and P. deltoides Bart (ILL-129) and known to be highly divergent for a vast range of phenotypic traits. We combined phenotyping, quantitative trait loci (QTL) analysis and microarray experiments to demonstrate that ‘genetical genomics’ can be used to provide information on adaptation at the species level. The grandparents and F2 population were subjected to soil drying, and contrasting responses to drought across genotypes, including leaf coloration, expansion and abscission, were observed, and QTL for these traits mapped. A subset of extreme genotypes exhibiting extreme sensitivity and insensitivity to drought on the basis of leaf abscission were defined, and microarray experiments conducted on these genotypes and the grandparent species. The extreme genotype groups induced a different set of genes: 215 and 125 genes differed in their expression response between groups in control and drought, respectively, suggesting species adaptation at the gene expression level. Co-location of differentially expressed genes with drought-specific and drought-responsive QTLs was examined, and these may represent candidate genes contributing to the variation in drought response.


Plants have evolved a host of response mechanisms to enable survival under stress conditions. Drought stress is the primary cause of crop loss worldwide (Boyer, 1982), reducing yields by as much as 50% (Bray et al., 2000), and, along with temperature, determines the global distribution of major vegetation biomes (Graham et al., 2002). According to the latest Intergovernmental Panel on Climate Change report (IPCC, 2001) drought frequencies are expected to increase globally over the next 50 years and there will be increasing shifts in patterns of rainfall.

Populus is now firmly established as a model organism (Bradshaw et al., 2000; Brunner et al., 2004; Taylor, 2002; Wullschleger et al., 2002), and has been added to the list of organisms for which the entire genome has been sequenced ( Populus as a model offers the opportunity to study stress response in a perennial, deciduous tree that is also being grown extensively as a commercial biomass crop for the production of carbon-neutral energy (Tuskan and Walsh, 2001), often in environments subjected to drought stress (Amlin and Rood, 2003) and where yield may be impaired (Lindroth and Båth, 1999; Pinon and Valadon, 1997). Previous work has shown that drought tolerance varies considerably between genotypes of Populus, both inter- and intra-specifically (Cochard et al., 1996; Gebre and Kuhns, 1993; Gebre et al., 1998; Harvey and Van Den Driessche, 1997; Marron et al., 2002; Robison and Raffa, 1998; Tschaplinski et al., 1998), suggesting that the genus provides a good model in which to investigate the genetic architecture and adaptive responses to this and other stresses. Despite extensive physiological and morphological descriptions of the Populus response to drought, to date little work has been undertaken to explain differences at the level of the gene and to examine the degree of similarity in stress response between this perennial, deciduous species and that of annual crop species or Arabidopsis.

At the molecular level, adaptation and speciation are the result of mutation in protein coding regions or in regulatory regions (Kirst et al., 2005), and, at least in the case of speciation, genome rearrangements. We were interested to see whether differences in drought response, which is a highly adaptive trait, could be used to gain insight into the genetic differences between closely related species and also to examine the degree of similarity or differences between the induced transcriptional events in response to drought stress between the species. As the poplar genome sequence and poplar microarrays are now available, it is possible to form a bridge between the approaches of quantitative trait loci (QTL) mapping, the ‘candidate gene’ approach, and transcriptomics (Borevitz et al., 2002; Weigel and Nordborg, 2005). Expressed sequence tags (ESTs) spotted on microarrays can be located on the physical genome sequence, and QTL mapping aims to identify regions involved in the control of a trait; it therefore follows that a list of all ESTs on a microarray and existing within a QTL region can be examined for differential expression. The expression levels of these genes can then be quantified within a mapping population and the data used to map QTL, with such QTL being termed expression QTL (eQTL; Kirst et al., 2005). When mapping eQTLs, there are two possible outcomes: either a cis- or a trans-regulatory element will be mapped. If the polymorphism lies in a cis-regulatory region, then the mapped QTL should co-locate to the gene of interest and is likely to represent a structural gene; if the polymorphism lies in a trans-regulatory region, the QTL will locate the transcription regulator(s). This approach also has potential for identifying transcription regulators acting on a regulon. In this case, the expression of a set of genes would co-locate to a single QTL, which would locate the transcription factor regulating the expression of that regulon. This approach was proposed by Jansen and Nap (2001), but, to date, has not been fully exploited. It was recently shown to be viable in a study of lignin synthesis in Eucalyptus (Kirst et al., 2004, Kirst et al., 2005). A number of studies have attempted to identify candidates that co-locate within QTL regions, such as for the cell-wall composition of Zea mays (Hazen et al., 2003), and for the co-location of transcriptional differences between drought response of Oryza (Hazen et al., 2004), and other work has examined the co-location of candidate genes with QTL or regions of introgression (Baxter et al., 2005; Silva et al., 2005). These studies offer useful insights into the control of a trait response, and prove useful in identifying cases where a full eQTL study may be of value, as well as identifying a subset of genes for targeted expression analysis within a mapping or natural population. The approach has also been applied to asthma susceptibility in mice (Karp et al., 2000), and for ovariole number in Drosophila (Wayne and McIntyre, 2002). Limitations to this approach exist – principally that it assumes that expression changes serve as the underlying causal mechanism of phenotypic divergence and that a gene is being actively transcribed at the time of sampling – however, it seems reasonable to assume that this will be true in at least some cases. These limitations aside, the combination of these approaches offers new opportunities to investigate causal mechanisms underlying adaptive traits and speciation.

In this study, we have used the grandparents and the F2 mapping pedigree initially described by Bradshaw and Stettler (1993) for which many QTL for traits of adaptive significance are already available. Variation in response to drought in this population was assessed, and this was complemented by analysis of the transcriptional response to drought of both the grandparent species and a subset of the F2 genotypes. We hypothesize that this pedigree will be a valuable tool in understanding tree adaptation to drought as the grandparent genotypes were selected from a relatively wet (P. trichocarpa) and dry (P. deltoides) regions of the USA.

Results and discussion

Contrasting physiological mechanisms of drought response in P. deltoides and P. trichocarpa

We tested the hypothesis that the grandparent species would exhibit contrasting drought responses by exposing them to an acute drought and performing detailed physiological and transcriptional assessments as the drought stress progressed. P. deltoides and P. trichocarpa manifested contrasting physiological responses, perhaps reflecting the adaptation of the species to the drought regimes of their natural ranges. In the leaves of P. deltoides, senescence was initiated in older leaves (Figure 1a), stomata closed rapidly in response to drought with a concomitant reduction in the rate of photosynthesis (Figure 1b), and leaf area was significantly reduced at all leaf plastochron indices (LPIs) tested (Figure 1c). In contrast, leaves of P. trichocarpa formed rapidly spreading necrotic lesions in semi-mature to mature leaves after exposure to drought stress for 14 days, followed by leaf shedding. Smaller reductions in stomatal conductance and photosynthetic activity and no reduction in leaf area were observed. It has been argued that leaf shedding may serve as an adaptive drought response, particularly in perennial species, designed to reduce load on the transpiration stream (Rood et al., 2000) and typically resulting from fatal xylem cavitation in the leaf mid-ribs (Hukin et al., 2005) as opposed to the formation of an abscission zone. P. deltoides is typically more prone to xylem cavitation, with the later onset of cavitation resulting from the ability of the species to rapidly close stomata (Rood et al., 2000). Cochard et al. (1996) have shown that the hybrid poplar P. koreana× P. trichocarpa cv. ‘Peace’ is unable to close stomata in response to drought or abscisic acid (ABA) application, and Hukin et al. (2005) have recently shown that this clone readily develops mid-rib cavitations in response to drought. The data presented in Figure 1(b) reveal a similar, limited stomatal response for P. trichocarpa in response to drought, and it is therefore likely that this limited ability to close stomata may lead to rapid xylem cavitation, ultimately resulting in leaf shedding. Whether this is an evolved, adaptive trait (hydraulic segmentation through cavitation; Hukin et al., 2005) or is merely an unavoidable consequence of the xylem structure of Populus is not clear. The results shown in Figure 1(b) for P. trichocarpa additionally indicate that the lack of stomatal response may be dependent on the developmental age of the leaf, which is consistent with the findings of Hukin et al. (2005) and Ridolfi et al. (1996).

Figure 1.

 Physiological response to drought stress in P. trichocarpa and P. deltoides.
(a) Progression of leaf senescence/damage in response to drought for P. trichocarpa and P. deltoides. The time taken to progress through the stages after the first appearance of any visible symptoms is indicated in parentheses for each species, and the degree of visible damage at various time points is shown. In both species, symptoms had appeared in semi-mature/mature leaves after 14 days of drought stress.
(b) Percentage effect on the rate of photosynthesis in response to drought for P. deltoides (closed grey bars) and P. trichocarpa (open spotted bars) and percentage effect on stomatal conductance for P. deltoides (closed black bars) and P. trichocarpa (open bars).
(c) Percentage effect on leaf area (%) in response to drought at contrasting LPIs for P. deltoides (closed black bars) and P. trichocarpa (open bars). Results of a two-way anova test are shown for each LPI: ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.01. In all cases, percentage effect was calculated as [(drought − control)/control] × 100.

Physiological drought traits segregate within the population

We examined segregation of drought response using a full-sib second generation population (referred to as F2) formed from a cross between the P. trichocarpa and P. deltoides species detailed above (Bradshaw and Stettler, 1993). In total, the population contains approximately 350 genotypes, and for this study we used 167 of those (with the selection of genotypes used being dictated by our ability to produce adequate cutting material). A genetic map for this population has been produced by the laboratory of G.A. Tuskan (Oak Ridge National Laboratory, Oak Ridge, TN, USA; personal communication), and was used for this study in conjunction with the out-breeding package of the QTL mapping software QTLExpress (Seaton et al., 2002, see Experimental procedures). We have previously shown that this population shows marked segregation for all phenotypic traits examined (Rae et al., 2004), and, due to the contrasting regimes of the grandparent species, we expected that drought response would also segregate.

Marked transgressive segregation (i.e. segregation both above and below parental values) in drought response was observed for all traits recorded in the F2 population. Total chlorophyll, relative leaf expansion rate, percentage of leaves with visible necrosis and abscission, leaf length extension rate, carotenoid content, the carotenoid:total chlorophyll ratio and absolute leaf area expansion were all significantly altered by drought (Figure 2a). The percentage of abscised leaves, leaf area expansion and chlorophyll and carotenoid content were the most highly affected traits in response to drought.

Figure 2.

  Physiological response of grandparents and F2 population to drought stress. (a) Physiological overview of the response to drought of the F2Populus mapping family for the Populus trichocarpa grandmother (open bars), P. deltoides grandfather (closed black bars), and the mean F2 trait value (hatched bars).
(b) Physiological overview of the F2 extreme genotype groups for abscission response to drought. Each extreme group is composed of the highest or lowest five genotypes for the trait percentage abscission in response to drought: (closed black bars) high-abscission genotypes, (open bars) low-abscission genotypes.
Values shown are percentage effect, calculated as [(control − drought)/control] × 100. The results of anova are shown for each trait: *P > 0.05, **P > 0.01, ***P > 0.01. anova codes are: T, treatment; G, genotype; T×G, treatment by genotype interaction; E, extreme group; G(E), genotype nested within extreme group.

QTL were mapped from the physiological data in control conditions, drought and for percentage effect response to drought (these being termed ‘response’ QTL), with percentage effect being the percentage difference between the control and drought conditions. Reymond et al. (2004) offer insightful interpretation on the mapping of QTL in differing conditions and in response to conditions. In total, 25 QTL were mapped in control conditions, 44 in drought and 30 for percentage effect. Six QTL were common to control and drought (constitutive QTL; see Reymond et al. 2004) and 54 were specific to the drought and/or response to drought. Information on all QTL, including confidence intervals, maternal and paternal effects, percentage variance explained, and statistical information, can be found in Table S1.

Co-location of drought-specific QTL for a number of traits was observed on linkage groups (LGs) IV, VI, VII, VIII, IX, XI, XVIII and XIX (a subset for abscission QTL is shown in Figure 3). These all represent regions of the genome that function in controlling drought response with effects on multiple drought-response/tolerance-related traits, and, in the context of this experimental approach, may represent likely locations of a regulon-controlling transcription factor. Trait-specific drought/response QTL were also identified for some traits, for example a QTL for height on LG V(I) that explains approximately 8% of the trait variance under both conditions. These QTL are more likely to represent structural genes affecting a single trait and the response of that trait to drought stress.

Figure 3.

 The co-location of abscission QTL with genes differentially expressed in response to drought in the grandparent species P. deltoides and P. trichocarpa (red), the extreme F2 genotype groups for the abscission response to drought (green), and both (purple). Each extreme group is composed of the highest or lowest five genotypes for the trait percentage abscission in response to drought. Where QTL for other traits co-locate to abscission QTL, these have been shown. Trait names are in capital letters, followed by a lower-case letter indicating treatment conditions. Traits are: A, abscission; N, necrosis; CHL-B, chlorophyll b content; CHL-A:B, chlorophyll a:b ratio; CAR, carotenoid content; H, height. Conditions are: d, drought; p, percentage effect calculated as [(control − drought)/control] × 100. Only those linkage groups (LGs) with abscission QTL identified are shown. Chromosomes are represented by the open bars with the chromosomes number given above. The values to the left of the bar represent calculated base-pair positions. Gene names and SSR identifiers are shown to the right of the bar. The position of genes and SSRs along the chromosomes and of SSRs and QTL along LGs were calculated on a ratio scale and then converted to base-pair positions. Only SSR markers are shown, but QTL were mapped using a genetic map containing additional (non-sequence-based) markers as described in Experimental procedures. Confidence intervals were defined using an F-2 drop-off.

Having established that there was a clearly identifiable genetic basis to the segregation of drought response within the population, we examined the transcriptional response to drought within the grandparents and the F2 genotypes. High-throughput mRNA quantifications are a powerful method by which the differentially expressed gene(s) mapping within a QTL region may be identified. Such an approach may prove to be a useful, and relatively simple, means of breaching the QTL–gene barrier (Hazen and Kay, 2003).

The Populus drought transcriptome

We carried out two microarray experiments to characterize the transcriptome of Populus in response to drought, as well as to test whether we could identify species-specific patterns of gene expression when comparing the grandparent species response to drought or between genotypes with contrasting drought responses within the F2Populus mapping family. Before beginning the microarray experiment, we first confirmed that no global species-specific effects due to sequence divergence and homology issues could be detected on the basis of self–self hybridizations (data not shown, available in upsc-base, Sjödin et al., 2006: However, we do not feel that sequence homology should be a serious concern considering the average length of ESTs spotted on the microarrays and how closely related the two grandparent species used are.

Drought-induced remodelling of the grandparent transcriptomes.  We exposed the grandparent species to an acute drought-stress treatment, sampled leaf material (l0, see Experimental procedures) from control and drought-stressed plants (after 14 days of drought), prepared RNA from the leaves and analysed the samples on the POP1 microarrays using the experimental design shown in Figure 4(a). Genes with a high probability of differential expression were identified using the ebayes (B-statistic) method available in the R package ‘LIMMA’ (Smyth, 2004). This method returns a ‘B’ (B-statistic) score value and an FDR-adjusted P value. B values represent the log-odds ratio (natural logarithm of the odds) of differential expression (where a score of 0 represents a 50–50, or 1-to-1, chance of a gene being identified as differentially expressed due to chance, and scores >0 represent greater than random chance of differential expression). The B value is automatically adjusted for multiple testing (Smyth, 2004). We have used a B value cut-off of 5 for selecting those genes classed as differentially expressed, with a value of 5 representing approximately 99% certainty of differential expression (exp[5] = 148.4, i.e. about 148 to 1; the probability is therefore 148/[1 +s148] = 0.99, or 99%). The results showed that, in both species, a highly similar set of changes in gene expression were induced in response to drought. As in other species, drought stress resulted in a profound remodelling of the transcriptome, with many genes undergoing more than an eightfold change in expression (Table S2). To gain a holistic view of changes in gene expression and to place them within a biological context, the functional role of genes that were commonly differentially expressed in both species was examined. Functional classification of genes in PopulusDB (Moreau et al., 2005; Sterky et al., 2004; is based on gene ontology (GO; assignment of mapped gene models from the poplar genome sequence. GO terms are of three types (‘molecular function’, ‘biological process’ and ‘cellular component’). Here we show only data on ‘biological process’ categories as we felt these provide the most insightful means of interpreting the changes in gene expression that occured. The dataset classified according to ‘molecular function’ or ‘cellular component’ is available on request. GO categories identified are those significantly over-represented by differentially expressed genes, as identified using the ‘GOstats’ package available via Bioconductor (Gentleman et al., 2004; and as implemented in upsc-base (details in Experimental procedures).

Figure 4.

 Experimental design of microarray experiments.
Arrows indicate dye orientations. Replicate arrays conducted with cDNA synthesized from the same RNA extraction but with dyes in the reverse orientation were performed for all comparisons made.
(a) Microarray experimental design to examine the transcriptional drought response of P. deltoides and P. trichocarpa to drought stress. The number of biological replicate arrays is indicated in parentheses.
(b) Microarray experimental design to examine and compare the transcriptional response of extreme F2 genotype groups for the abscission response in control and drought-stress conditions. Each circle represents a pool of cDNA, with pools representing the five highest or lowest abscission drought-response genotypes sampled in either control or drought conditions. The number of independent biological replicates constituting the pool represented by each circle is shown in parentheses.

The most prominent change in gene expression was the downregulation of many genes involved in photosynthesis, a commonly observed outcome of exposure to drought stress (Kreps et al., 2002, Flexas and Medrano, 2002, Seki et al., 2002). Figure S1 shows the GO category hierarchy for downregulated genes. A clear downregulation of the photosynthetic apparatus can be seen, with this decrease in photosynthesis feeding into the observed decrease in metabolism. Although the clear trend within the photosynthesis category was downregulation (the category mean was 1.5-fold down), a small number of photosynthetic genes were upregulated, including those encoding early light-inducible proteins (ELIPs, which are know to be stress-induced) (Bhalerao et al., 2003; Kreps et al., 2002), β-amylase, sucrose synthase, sucrose-phosphate synthase, catalase 2, pyruvate kinase and the photosystem II 10 kDa polypeptide PsbR. These upregulated genes were not induced to the same extent in the two species: those encoding β-amylase, sucrose synthase and ELIPs were more highly induced in P. deltoides, while those encoding sucrose-phosphate synthase, pyruvate kinase and PsbR were more highly induced in P. trichocarpa.

Other GO categories containing many downregulated genes were ‘biosynthesis’ and the child categories ‘macromolecule biosynthesis’ and ‘protein synthesis’, indicating that overall protein synthesis was decreased in the drought-stressed leaves. Extensive upregulation was seen in the categories ‘response to water’ and ‘response to water deprivation’, as well as the more general ‘response to stress’ category. Upregulation was also observed in the ‘protein ubiquitination’ category, suggesting that, as well as the downregulation of protein biosynthesis, there was also an increase in protein degradation/turnover. Categories involved in wax biosynthesis and cuticle biosynthesis were also upregulated, suggesting that biomechanical adaptation was occurring at the level of leaf structure. Alterations to the cuticle and wax layers are often observed in response to drought stress in order to minimize latent water loss through the epidermis (Aharoni et al., 2004).

It is of course also interesting to examine the gene-by-gene list of differentially expressed genes (Table S2). Here, upregulated genes of interest included an ABA-binding factor (AREB1), a negative modulator of ABA activity (CBL-9), a ubiquitin-conjugating enzyme, an early response to drought protein (ERD7), and ethylene-insensitive 3 (EIN3), which has recently been shown to be involved in the ABA-induced closure of stomata in response to drought (Tanaka et al., 2005). Many genes observed to be upregulated in response to drought by previous studies (Bray, 2004; Seki et al., 2002) were also present, including LEA genes (Lea 4, 5-D, 14-A), osmoprotectants, chaperones and other stress-related genes (HSPs, glutathione-S-transferase, LTCOR 11), transcription factors (bzips, zinc finger family and myb family), and alcohol dehydrogenase, among others. A number of candidate genes for leaf expansion control, including RGA2 (response to gibberellic acid 2) and GAI1 (gibberellic acid-insensitive 1), Phantastica, and a number of cell-cycle genes (cyclin A1, A2, mitotic check point protein, retinoblastoma), were downregulated during drought stress (see Fleming, 2005 for a recent review on leaf development).

The EST database PopulusDB makes possible rapid digital expression profiling of many Populus genes to gain information on their expression levels in different tissues and after different treatments (Sterky et al., 2004), and viewing gene expression responses in comparison to EST distribution among tissue libraries can prove highly informative. We investigated the library distributions of ESTs from different clusters, corresponding to genes, for the list of 100 genes with most significant differential expression in both species. When presented as a clustered correlation map (according to Ewing et al., 1999), it became evident that the most upregulated genes (Figure 5a) have a very different expression pattern than the most downregulated (Figure 5b). This can be expressed numerically as the fraction of upregulated/downregulated genes found in each library, a relative occurrence factor (ROF, Table 1) (described in Experimental procedures). If this ratio is 1, the genes found in a library have an equal chance of being in the up- and downregulated datasets, and ratios over 1 indicate that they are more likely to be found in the upregulated dataset. The upregulated regulon(s) are abundantly expressed especially in the dormant cambium (UA), but also in the dormant buds (Q) and senescing leaves (I). By contrast, the genes in these regulons are almost not expressed in cambial tissues (cambial zone, AB; tension wood, G), in apical meristems (T) and not even in young leaves (C). The genes in the downregulated regulon(s) are almost all expressed in young leaves (C) and represent principally photosynthetic genes. Notably, the regulons induced during the stress conditions employed here did not overlap with the regulons induced by cold stress (L), the only stress condition library present on the arrays. It has previously been reported that cold-stress response is less dependent on ABA-mediated changes in gene expression than other stress responses, and the observed lack of similarity to cold stress and similarity to dormancy may be a reflection of the crucial role of ABA in mediating plant responses to drought.

Figure 5.

 The number of ESTs for gene models identified as differentially expressed is shown according to cDNA tissue library in PopulusDB.
The various tissue libraries are represented by letter codes and form the columns of the figure. Each row represents a gene model. Cells are shaded according to the number of ESTs identified for that gene model in a particular tissue library.
(a) ESTs commonly upregulated in response to drought in P. deltoides and P. trichocarpa, (b) ESTs commonly downregulated in response to drought in P. deltoides and P. trichocarpa, (c) ESTs upregulated in response to drought in low-abscission extreme genotypes, (d) ESTs regulated in response to drought in low-abscission extreme genotypes, (e) ESTs that are differentially expressed between high- and low-abscission extreme genotypes in control and drought conditions.

Table 1.   Similarities between up- and downregulated genes during drought and the transcriptome of different Populus tissues/treatments
  1. ROF, relative occurrence factor (see Experimental procedures).

UADormant cambium9.14
QDormant buds3.05
ISenescing leaves2.81
XWood cell death1.49
UBActive cambium1.19
LCold-stressed leaves1.05
SImbibed seeds0.90
GTension wood0.88
FFlower buds0.65
VMale catkins0.65
KApical shoot0.57
MFemale catkins0.53
TApical meristem0.53
CYoung leaves0.33
ABCambial zone0.30

Taken together, the gross transcriptome responses upon drought stress in Populus seem to be biologically adequate and were largely similar to changes reported in earlier studies in other plants.

Species-specific transcriptional response to drought.  Rather than simply gaining an overview of the transcriptional response to drought of Populus in general, our aim was to test whether we could identify genes that may contribute to the genetic architectural differences accounting for the contrasting adaptive drought-response mechanisms of P. deltoides and P. trichocarpa. Such an analysis requires stringent statistical treatment, and to this end we undertook an anova analysis to determine genes differentially expressed between the two species based on ratio-of-expression values. A total of 569 genes that were differentially expressed between the species were identified, with 59 of these remaining significant after applying a Bonferroni correction (Table 2). anova was the most suitable statistical test in this case, where no direct connection between species existed on arrays and so ratio-of-expression comparisons were tested. This list identified genes in regulons that were differently regulated in the two species – regulons that may be important for the different drought responses in the two species. Among the genes more highly expressed during drought in P. trichocarpa were those encoding blight-associated protein P12, a putative disease resistance protein, a ripening-related protein, an expansin (with an apparent leaf-specific expression pattern, deduced from the library distribution of EST clones), a methyl transferase and an epimerase/dehydratase. Genes with an informative annotation that were more highly induced in P. deltoides included those encoding enzymes in starch metabolism, a granule-bound glycogen/starch synthase, a 1–4-α-glucan branching enzyme (starch branching enzyme) and proteases (ClpR1 and ubiquitin). In addition, a 1-aminocyclopropane-1-carboxylate oxidase (ACC oxidase) was induced in P. deltoides to a considerably greater extent than in P. trichocarpa. There are many ACC oxidase genes expressed in Populus, and data-mining of PopulusDB showed that these genes often had drastically different expression patterns; this particular ACC oxidase gene (PU10422, PU10214, Table 2), appears to be almost exclusively expressed in senescing leaves. Ethylene plays a key role in controlling the senescence and abscission of leaves, both in response to stress and in autumnal senescence (Andersson et al., 2004; Buchanan-Wollaston et al., 2005, Buchanan-Wollaston et al., 2003; Lim et al., 2003), and it is therefore interesting that in these two species ethylene was more highly induced in the species exhibiting senescence in response to drought.

Table 2.   ESTs that are differentially expressed in response to drought between P. deltoides and P. trichocarpa, as identified by one-way anova after Bonferroni correction
PUP. deltoidesP. trichocarpaDescription
  1. PU numbers represent gene identifiers in PopulusDB ( Values for P. trichocarpa and P. deltoides are M values (log2 R/G).

PU132066.4755982.198361Inositol-3-phosphate synthase, putative/myo-inositol-1-phosphate synthase, putative/MI-1-P synthase, putative very strong similarity to SP|Q38862 Myo-inositol-1-phosphate synthase isozyme 2 (EC (MI-1-P synthase 2) (IPS 2) {Arabidopsis thaliana}; identical to SP|Q9LX12|Probable inositol-3-phosphate synthase isozyme 3 (EC (Myo- inositol-1-phosphate synthase 3) (MI-1-P synthase 3) (IPS 3) {Arabidopsis thaliana}; contains Pfam profile PF01658:Myo-inositol-1-phosphate synthase
PU072714.1001531.825729Calcium-transporting ATPase, plasma membrane-type, putative/Ca2+-ATPase, putative (ACA10) identical to SP|Q9SZR1 Potential calcium-transporting ATPase 10, plasma membrane-type (EC (Ca(2+)-ATPase isoform 10) {Arabidopsis thaliana}; similar to SP|Q9LF79 Calcium-transporting ATPase 8, plasma membrane-type (EC (Ca(2+)-ATPase isoform 8) {Arabidopsis thaliana}
PU103802.6016068.789288Metallothionein protein, putative (MT2A) identical to Swiss-Prot:P25860 metallothionein-like protein 2A (MT-2A) (MT-K) (MT-1G) [Arabidopsis thaliana]
PU045992.4181987.885675Hypothetical 7.2 kDa protein
PU001642.2560461.147243Vacuolar calcium-binding protein-related
PU091112.2088625.768294ATP synthase delta chain, chloroplast, putative/H(+)-transporting two-sector ATPase, delta (OSCP) subunit, putative similar to SP|P32980 ATP synthase delta chain, chloroplast precursor (EC {Nicotiana tabacum}; contains Pfam profile PF00213: ATP synthase F1, delta subunit
PU109342.1729326.569877ABC transporter family protein
PU105781.963257.085015ABC transporter family protein
PU050801.7969043.205011Kinesin light chain-related low similarity to kinesin light chain [Plectonema boryanum] GI:2645229; contains Pfam profile PF00515 TPR Domain
PU059491.7714643.079621Dihydroorotate dehydrogenase family protein/dihydroorotate oxidase family protein low similarity to SP|Q12882 Dihydropyrimidine dehydrogenase [NADP+] precursor (EC (DPD) (DHPDHase) (Dihydrouracil dehydrogenase) (Dihydrothymine dehydrogenase) {Homo sapiens}; contains Pfam profile PF01180: Dihydroorotate dehydrogenase
PU041041.769849.1423461,4-alpha-glucan branching enzyme/starch branching enzyme class II (SBE2-1) nearly identical to starch branching enzyme class II [Arabidopsis thaliana] GI:619939
PU095611.7117793.286455Oxidoreductase, 2OG-Fe(II) oxygenase family protein similar to IDS3 [Hordeum vulgare][GI:4514655], leucoanthocyanidin dioxygenase [SP|P51091][Malus domestica]; contains PF03171 2OG-Fe(II) oxygenase superfamily domain
PU009871.6600263.753532Expressed protein contains Pfam profile PF05129: Putative zinc binding domain (DUF701)
PU129881.5801982.957931Dihydroorotate dehydrogenase family protein/dihydroorotate oxidase family protein low similarity to SP|Q12882 Dihydropyrimidine dehydrogenase [NADP+] precursor (EC (DPD) (DHPDHase) (Dihydrouracil dehydrogenase) (Dihydrothymine dehydrogenase) {Homo sapiens}; contains Pfam profile PF01180: Dihydroorotate dehydrogenase
PU094121.5351740.327163RWP-RK domain-containing protein similar to nodule inception protein [Lotus japonicus] GI:6448579; contains Pfam profile: PF02042 RWP-RK domain
PU028451.4974412.74512ABC transporter family protein similar to ABC transporter homolog PnATH GI:7573600 from [Populus nigra]
PU047081.4381450.440084Invertase/pectin methylesterase inhibitor family protein/DC 1.2 homolog (FL5-2I22) similar to SP|Q42534 Pectinesterase 2 precursor (EC (Pectin methylesterase 2) (PE 2) {Arabidopsis thaliana}; contains Pfam profile PF04043: Plant invertase/pectin methylesterase inhibitor; FL5-2I22 mRNA for DC 1.2 homolog, partial cds GI:11127598
PU121031.4191922.464271ATP-dependent Clp protease proteolytic subunit (ClpR1) (nClpP5) identical to nClpP5 GB:BAA82069 GI:5360595 from [Arabidopsis thaliana]; identical to cDNA nClpP5 (nuclear encoded ClpP5) GI:5360594
PU070501.4118560.310094Expressed protein
PU100241.3501120.465823Leucine-rich repeat family protein contains leucine rich-repeat (LRR) domains Pfam:PF00560, INTERPRO:IPR001611; similar to Hcr2-0B [Lycopersicon esculentum] gi|3894387|gb|AAC78593
PU120911.3377085.679871Starch synthase, putative similar to starch synthase SP:Q42857 from [Ipomoea batatas]
PU044241.3362570.168624GCN5-related N-acetyltransferase (GNAT) family protein contains Pfam profile PF00583: acetyltransferase, GNAT family
PU135211.3350285.671338Starch synthase, putative similar to starch synthase SP:Q42857 from [Ipomoea batatas]
PU025601.275360.209652O-methyltransferase family 2 protein similar to caffeic acid O-methyltransferase, Pinus taeda, gb:U39301
PU132361.2564612.771864AP2 domain-containing transcription factor, putative contains similarity to AP2 domain transcription factor
PU125381.2239850.278769NAD-dependent epimerase/dehydratase family protein similar to sugar epimerase BlmG from Streptomyces verticillus GI:9937230; contains Pfam profile PF01370 NAD dependent epimerase/dehydratase family
PU047601.18442.537971Sucrose-phosphatase 1 (SPP1) identical to sucrose-phosphatase (SPP1) [Arabidopsis thaliana] GI:11127757
PU119921.1676032.389933Expressed protein contains Pfam profile PF05129: Putative zinc binding domain (DUF701)
PU048581.1629623.141314Pseudo-response regulator 7 (APRR7) identical to pseudo-response regulator 7 GI:10281004 from [Arabidopsis thaliana]
PU125641.1435052.291262NADP-dependent oxidoreductase, putative similar to probable NADP-dependent oxidoreductase (zeta-crystallin homolog) P1 [SP|Q39172][gi:886428] and P2 [SP|Q39173][gi:886430], Arabidopsis thaliana; similar to allyl alcohol dehydrogenase GI:9758497 from [Arabidopsis thaliana]
PU003511.1165863.863891Polyubiquitin (UBQ14) identical to GI:166795; similar to N. sylvestris hexameric polyubiquitin, GenBank accession number M74101
PU075501.1051910.245513Beta-galactosidase, putative/lactase, putative similar to beta-galactosidase precursor SP:P48980 from [Lycopersicon esculentum]
PU118001.0770153.639903Expressed protein contains Pfam profile: PF04601 protein of unknown function (DUF569
PU047541.0675330.394291Expressed protein contains Pfam profile PF04784: Protein of unknown function, DUF547; expression supported by MPSS
PU048821.0660442.698525Starch synthase, putative similar to starch synthase SP:Q42857 from [Ipomoea batatas]
PU104221.0600482.5893271-aminocyclopropane-1-carboxylate oxidase
PU102141.0317352.2336071-aminocyclopropane-1-carboxylate oxidase
PU059571.0259862.597359CBL-interacting protein kinase 25 (CIPK25) identical to CBL-interacting protein kinase 25 [Arabidopsis thaliana] gi|17646697|gb|AAL41008
PU128280.9538393.718451Expressed protein
PU115450.9444253.078609Expressed protein
PU012310.935810.331329Expansin, putative (EXP8) similar to expansin 2 GI:7025493 from [Zinnia elegans]; alpha-expansin gene family, PMID:11641069
PU092750.9242740.493374Monodehydroascorbate reductase, putative similar to cytosolic monodehydroascorbate reductase GB:BAA77214 [Oryza sativa]
PU037330.9225220.308485Expressed protein
PU105600.9180332.316057Expressed protein
PU072190.9179323.605562Auxin-responsive protein, putative similar to auxin-induced protein AIR12 (GI:11357190) [Arabidopsis thaliana]; similar to stromal cell-derived factor receptor 2 (GI:20381292) [Mus musculus]
PU099650.8596180.124811Expansin-related similar to blight-associated protein p12 precursor [Citrus jambhiri] gi|4102727|gb|AAD03398; similar to beta-expansin [Oryza sativa] gi|8118428|gb|AAF72986; expansin-related gene, PMID:11641069,
PU088460.8470.342712Expressed protein
PU081400.8356282.696852Auxin-responsive protein, putative similar to auxin-induced protein AIR12 (GI:11357190) [Arabidopsis thaliana]; similar to stromal cell derived factor receptor 2 (GI:20381292) [Mus musculus]
PU094100.8174620.122454Expansin-related similar to blight-associated protein p12 precursor [Citrus jambhiri] gi|4102727|gb|AAD03398; similar to beta-expansin [Oryza sativa] gi|8118428|gb|AAF72986; expansin-related gene, PMID:11641069,
PU036080.8109012.2006Phosphorylase family protein contains Pfam PF01048: Phosphorylase family
PU086290.8106843.166807Expressed protein
PU012160.8088220.224556Zinc finger (C3HC4-type RING finger) family protein contains Pfam profile: PF00097 zinc finger, C3HC4 type (RING finger)
PU126380.7976440.369748DEAD box RNA helicase, putative similar to RNA helicase [Rattus norvegicus] GI:897915; contains Pfam profiles PF00270: DEAD/DEAH box helicase, PF00271: Helicase conserved C-terminal domain
PU017310.7412060.276708Hypothetical protein
PU083820.7197380.208747Expansin, putative (EXP8) similar to expansin 2 GI:7025493 from [Zinnia elegans]; alpha-expansin gene family, PMID:11641069
PU031270.7088412.755436Glutaredoxin, putative similar to glutaredoxin [Ricinus communis] SWISS-PROT:P55143
PU036150.6582812.462611Expressed protein
PU057300.628132.633208Expressed protein
PU123820.4965740.9461783′(2′),5′-bisphosphate nucleotidase/inositol polyphosphate 1-phosphatase/FIERY1 protein (FRY1) (SAL1)

Although the commonality in the transcriptional responses of the two species to drought was large, the data presented here indicate that there were a number of genes whose expression differed in response to drought between the two Populus species, suggesting that at least a part of their contrasting physiological responses to drought may be genetically controlled through differences in gene regulation. To provide further insight into this, we performed a drought-stress experiment using the F2 offspring of the grandparent clones.

Transcriptional separation of drought response within the population

In order to assess the extent to which gene expression may account for the separation of F2 genotypes in their drought response, ideally the transcriptome of all individuals in the population should be analysed during drought stress, as was undertaken for a population subset by Kirst et al. (2005) to examine eQTL in wood-forming tissues. Rather than assaying our entire population, we attempted to construct an experimental design that allowed us to identify those genes exhibiting strong segregation associated with drought response within the population. Two extreme groups of genotypes (the five highest and lowest) were identified based on the trait ‘percentage effect response to drought of the percentage of abscised leaves’, and the labelled cDNA obtained from individual replicate leaves from genotypes within each group was pooled for microarray analysis (Figure 4b). Our aim was to investigate whether large-scale differences could be identified in the transcriptional response of the two extreme groups, and we felt that a pooled-loop design provided the most efficient strategy for testing this. Although we are aware of the limitation of such a pooling strategy (we have no indication of the contribution of individual replicates or genotypes within the pool), we do not feel that this represents a significant limitation to our approach: our aim is to identify those genes with the most highly segregating expression levels between the two sets of extreme genotypes. The pooling strategy will achieve this goal but may also identify a proportion of false-positives that achieved significance through the influence of an individual replicate or genotype on the mean expression represented by the pooled sample. Such a degree of false-positives is acceptable as they can readily be screened through subsequent RT-PCR confirmation before progressing to eQTL mapping within the entire population. We are currently undertaking this confirmation and subsequent eQTL mapping strategy. A number of papers have recently addressed the statistical issues of pooling and interpretation of results from pools of RNA (Kendziorski et al., 2003; Kendziorski et al., 2005; McShane et al., 2003; Peng et al., 2003; Shih et al., 2004; Zhang and Gant, 2005), as well as the benefits of loop designs in comparison to other commonly used microarray experimental designs (McShane et al., 2003; Vinciotti et al., 2005).

This approach can be considered similar to that of performing a bulk segregant analysis (Borevitz et al., 2003), although with the obvious difference that the data obtained from microarrays is continuous rather than categorical. RNA was extracted from individual replicates of each extreme genotype, and a pool of genotypes within each extreme group was then created for transcriptional analysis. For all genotypes except one, there were a minimum of two biological replicates per genotype. As the POP2 array had subsequently been developed, containing almost twice as many elements as the POP1 array, we were able to assay differential expression of many more genes in this experiment.

Abscission was selected as it is a late-onset indicator of drought-adaptation response, and showed a significant treatment and genotype effect within the F2 genotypes, with drought-specific QTL being mapped that account for a total of 43% of trait variance. Figure 2(a) shows the percentage effect response to drought for the physiological traits recorded, and Figure 2(b) shows the separation of the extreme genotype groups for those traits. Although percentage abscission was the most significant separator of the extreme groups, relative leaf area expansion rate and the carotenoid:total chlorophyll content also showed significant differences.

Analysis of this dataset revealed a striking difference in the transcriptional response of the two extreme groups. Again, differentially expressed genes were identified using the ebayes method of the ‘LIMMA’ package in R. When comparing significant changes in response to drought for the high-abscission extreme with those of the low-abscission genotypes, only 65 genes were common to both lists (55 of these being upregulated and 10 downregulated). In the high-abscission group, there were 128 genes with a high probability of differential expression in response to drought, and 386 in the low-abscission group. This proved that our experimental design using a pooling strategy was capable of detecting segregating patterns of gene expression, allowing us to gain insights into the genomic structure of the transcriptional drought response and showing that a major factor distinguishing the two phenotypic extreme groups was the differential expression of genes between them.

The loop design that we used allowed data to be interrogated from a high- versus low-extreme group perspective in both control and drought conditions. In drought, there were 48 genes more highly expressed in the high-abscission extreme group and 77 genes with lower expression. In this comparison, a gene with a higher expression level in the high-abscission extreme group than in the low will have a positive log2 ratio value, and a gene with higher expression in the low-abscission extreme group will have a negative log2 ratio. In control conditions, 96 genes were more highly expressed and 119 showed lower expression. Such control–drought expression differences represent constitutive differences in gene expression across the population that could contribute to drought susceptibility and tolerance.

The functional role of genes based on GO categories (as described above) was examined. Table 3 gives GO categories that were significantly over-represented by differentially expressed genes in either the high-abscission or low-abscission extremes in control and drought conditions. These data allow two questions to be answered: (i) are there classes of genes that differ in their constitutive expression and could account for differences in drought response?, and (ii) are there drought-responsive gene classes that differ in their degree or direction of expression change that could account for the differential drought response? In control conditions, the high-abscission extreme had higher expression levels of genes involved in hormonal response and signalling, including ABA-mediated signalling and response, jasmonic acid- and ethylene-mediated signalling, and categories involved in biotic stress and pathogenesis responses. The expression differences for ABA signalling, biotic stress and pathogenesis responses remained significant in drought conditions. As for the grandparental gene expression responses, it is interesting that ethylene signalling was more highly expressed in the high-abscission extreme. It is also of note that a number of biotic stress and pathogenesis response categories contain genes that are significantly more highly expressed in high-abscission lines, as this may suggest that these genotypes initiate a response that has a greater overlap with the hypersensitive-type response of plants. Indeed, these genotypes regulate genes involved in reactive oxygen species (ROS) removal such as glutathione peroxidase and superoxide dismutase (SOD) to a lesser extent than low-abscission response genotypes. The control of ROS scavenging is critical in determining the observed response to a stress condition (Mittler et al., 2004; Rao and Davis, 1999).

Table 3.   Gene ontology biological process categories that were significantly over-represented by genes with expression differences between high- and low-abscission extreme genotypes in control and drought conditions
GO IDP valueDescription
(i) Categories that have higher expression in high-abscission genotypes under control conditions.
 GO:00096070.003893Response to biotic stimulus
 GO:00069520.006005Defence response
 GO:00097550.008658Hormone-mediated signalling
 GO:00097370.013095Response to abscisic acid stimulus
 GO:00071650.013749Signal transduction
 GO:00097250.014563Response to hormone stimulus
 GO:00098710.016874Jasmonic acid- and ethylene-dependent systemic resistance, ethylene-mediated signalling pathway
 GO:00461730.016874Polyol biosynthesis
 GO:00061140.016874Glycerol biosynthesis
 GO:00508320.025209Defence response to fungi
 GO:00158030.025209Branched-chain aliphatic amino acid transport
 GO:00098140.027542Defence response to pathogen, incompatible interaction
 GO:00071540.028728Cell communication
 GO:00098730.033476Ethylene-mediated signalling pathway
 GO:00066620.033476Glycerol ether metabolism
 GO:00066410.033476Triacylglycerol metabolism
 GO:00066380.033476Neutral lipid metabolism
 GO:00066390.033476Acylglycerol metabolism
 GO:00060710.033476Glycerol metabolism
 GO:00464860.033476Glycerolipid metabolism
 GO:00428290.034527Defence response to pathogen
 GO:00064570.036371Protein folding
 GO:00067290.041675Tetrahydrobiopterin biosynthesis
 GO:0098270.041675Cell-wall modification (sensu Magnoliophyta)
 GO:00098160.041675Defence response to pathogenic bacteria, incompatible interaction
 GO:00428300.041675Defence response to pathogenic bacteria
 GO:00461460.041675Tetrahydrobiopterin metabolism
 GO:00092690.049808Response to dessication
 GO:00097380.049808Abscisic acid-mediated signalling
 GO:00096200.049808Response to fungi
 GO:00427420.049808Defence response to bacteri
 GO:00461650.049808Alcohol biosynthesis
(ii) Categories that have higher expression in low-abscission genotypes under control conditions
 GO:00096992.34E-07Phenylpropanoid biosynthesis
 GO:00098128.59E-07Flavonoid metabolism
 GO:00098138.59E-07Flavonoid biosynthesis
 GO:00423981.98E-06Amino acid derivative biosynthesis
 GO:00096981.98E-06Phenylpropanoid metabolism
 GO:00197482.58E-06Secondary metabolism
 GO:00194381.83E-05Aromatic compound biosynthesis
 GO:00065752.13E-05Amino acid derivative metabolism
 GO:00067250.0005389Aromatic compound metabolism
 GO:00197250.00490113Cell homeostasis
 GO:00068730.00490113Cell ion homeostasis
 GO:00508010.00490113Ion homeostasis
 GO:00065190.006082497Amino acid and derivative metabolism
 GO:00300020.00635389Anion homeostasis
 GO:00097140.00635389Chalcone metabolism
 GO:00097150.00635389Chalcone biosynthesis
 GO:00097180.00635389Anthocyanin biosynthesis
 GO:00466880.00635389Response to copper ion
 GO:00303190.00635389Divalent and trivalent inorganic anion homeostasis
 GO:00421800.00635389Ketone metabolism
 GO:00421810.00635389Ketone biosynthesis
 GO:00160980.00635389Monoterpenoid metabolism
 GO:00160990.00635389Monoterpenoid biosynthesis
 GO:00462830.00635389Anthocyanin metabolism
 GO:00306430.00635389Phosphate ion homeostasis
 GO:00094430.01266937Pyridoxal 5′-phosphate salvage
 GO:00086140.01894661Pyridoxine metabolism
 GO:00098280.01894661Cell-wall loosening (sensu Magnoliophyta)
 GO:00098310.01894661Cell-wall modification during cell expansion (sensu Magnoliophyta)
 GO:00100510.01894661Vascular tissue pattern formation (sensu Tracheophyta)
 GO:00100350.01894661Response to inorganic substance
 GO:00100380.01894661Response to metal ion
 GO:00428160.01894661Vitamin B6 metabolism
 GO:00425470.02518675Cell-wall modification during cell expansion
 GO:00096640.028254Cell-wall organization and biogenesis (sensu Magnoliophyta)
 GO:00098270.03138992Cell-wall modification (sensu Magnoliophyta)
 GO:00469160.03138992Transition metal ion homeostasis
 GO:00452290.04339928External encapsulating structure organization and biogenesis
 GO:00070470.04339928Cell-wall organization and biogenesis
 GO:00073890.04368284Pattern specification
(iii) Categories that have higher expression in high-abscission genotypes under drought conditions
 GO:00071650.00092717Signal transduction
 GO:00071540.00212908Cell communication
 GO:00096070.00382956Response to biotic stimulus
 GO:00100750.00423592Regulation of meristem size
 GO:00100730.00845532Meristem maintenance
 GO:00064680.00898486Protein amino acid phosphorylation
 GO:00069520.01471841Defence response
 GO:00099340.01684487Regulation of meristem organization
 GO:00098160.02101517Defence response to pathogenic bacteria, incompatible interaction
 GO:00428300.02101517Defence response to pathogenic bacteria
 GO:00067930.02370078Phosphorus metabolism
 GO:00067960.02370078Phosphate metabolism
 GO:00097380.02516925Abscisic acid-mediated signalling
 GO:00400080.02516925Regulation of growth
 GO:00427420.02516925Defence response to bacteria
 GO:00098320.03342809Cell-wall biosynthesis (sensu Magnoliophyta)
 GO:00099330.03342809Meristem organization
 GO:00096180.03753498Response to pathogenic bacteria
 GO:00096170.04569853Response to bacteria
(iv) Categories that have higher expression in low-abscission genotypes under drought conditions
 GO:00197250.00148348Cell homeostasis
 GO:00068730.00148348Cell ion homeostasis
 GO:00508010.00148304Ion homeostasis
 GO:00300020.003529827Anion homeostasis
 GO:00197400.003529827Nitrogen utilization
 GO:00068080.003529827Regulation of nitrogen utilization
 GO:00303190.003529827Divalent and trivalent inorganic anion homeostasis
 GO:00306430.003529827Phosphate ion homeostasis
 GO:00099370.007048436Regulation of gibberellic acid-mediated signalling
 GO:00099380.007048436Negative regulation of gibberellic acid-mediated signalling
 GO:00097400.01055586Gibberellic acid-mediated signalling
 GO:00068780.01055586Copper ion homeostasis
 GO:00097390.014052129Response to gibberellic acid stimulus
 GO:00469160.017537275Transition metal ion homeostasis
 GO:00300050.02101133Divalent and trivalent inorganic cation homeostasis
 GO:00068750.03136726Metal ion homeostasis
 GO:00099680.03136726Negative regulation of signal transduction
 GO:00063340.03136726Nucleosome assembly
 GO:00068010.034797263Superoxide metabolism
 GO:00099660.034797263Regulation of signal transduction
 GO:00194300.034797263Removal of superoxide radicals
 GO:00063330.03821633Chromatin assembly or disassembly
 GO:00068690.045021783Lipid transport

Even more striking differences were revealed when examining genes that were more highly expressed in the low-abscission extreme. In control conditions, many categories involved in secondary metabolite synthesis, including biosynthesis of phenylpropanoids, flavonoids, anthocyanins, chalcones and monoterpenoids showed significantly higher expression in the low-abscission extreme. There were also categories involved in cell-wall modification, organization and pattern definition. The apparent emphasis on homeostasis and cellular protection may indicate that resources are being utilized to maintain the cellular integrity and biochemical functionality of leaves. For example, many of these categories could result in an enhanced ability to tolerate increases in ROS and other drought-induced biochemical stresses.

In drought conditions, many categories involved in cellular homeostasis maintenance were represented, as were categories for gibberellic acid (GA) response and signalling, suggesting a functional role for this hormone in the drought response of trees that do not readily abscise leaves when exposed to drought stress, which is interesting in light of the recent findings of Achard et al. (2006), who showed that the quadruple DELLA mutant of Arabidopsis (GAI, RGA, RGL1, RGL2) exhibited reduced growth inhibition in response to salt stress compared to wild-type plants.

We then investigated the library distribution of the regulons that were differentially expressed, either in drought or in control conditions, in drought-sensitive versus drought-resistant clones in the population. As for the common responses to drought in the grandparents, we studied the digital expression profiles in PopulusDB for the gene lists identified above. Within the high-abscission extremes, there was no strong pattern of library distribution, perhaps as a result of the small number of differentially expressed genes (data not shown). More obvious patterns were seen for the low-abscission extremes, with the greatest number of upregulated genes being located within the dormant bud and dormant cambium tissue libraries (Figure 5c), perhaps reflecting the adaptive response of these genotypes. A combined list of genes differing in their expression between the high and low extremes in drought and control conditions is shown in Figure 5(e), where the greatest number of ESTs were found in the young leaf, apical and shoot meristem libraries as well as the senescing leaf, flower bud and petiole libraries.

Integrating transcriptional and QTL data

We wished to examine the degree to which differentially expressed genes co-locate to genomic regions identified by QTL analysis. If co-location occurs, this could be due to differences in cis-acting elements (promoter sequences) in the differentially expressed genes. This increases the probability that a gene is involved in the control of the drought response, although it does not prove a causal link as the expression may be regulated in trans by changes in, for example, a transcription factor regulating the drought response. Weigel and Nordborg (2005) discuss the evidence required for forming a causal link between gene and phenotype.

Both the grandparent and extreme-genotype datasets provide candidates for explaining segregation of drought response at the transcriptional level, although not all grandparental differences would be expected to segregate within the population. As discussed above, there were considerable differences between the two sets of candidate genes, with a greater number of co-locating genes being identified from the extreme-genotype comparison.

Figure 3 shows the location of QTL mapped for leaf abscission. Using the sequence of simple sequence repeat (SSR) marker primers as a link between the physical and genetic maps, the positional ratios of genes along a chromosome and of QTL along LGs were calculated as a means to examine co-location. Although only SSR markers are shown in Figure 3, all available markers were used for the QTL mapping results shown. This resulted in 13 genes that differed in their expression between extreme groups and that co-located to genomic regions identified by QTL mapping. A number of these genes have known functions that could impact upon the ability to tolerate drought stress or on the nature of the response initiated upon exposure to drought stress. The poplar genome contains an estimated 40 514 gene models. We have calculated that the QTL regions presented in Figure 3 (i.e. the entire genome regions within the confidence intervals for all QTL shown) contain 2167 genes, and we are examining the co-location of a total of 340 genes (data not shown). The POP2 arrays used represent 16 435 gene models. We have therefore calculated that a random list of 340 genes could be expected to contain 7.4 genes that co-locate to our QTL regions – if all genes were on the array, then the expectation would be (2167/40 514) × 340 = 18.1. However, POP2 arrays only represent 0.41 (i.e. 16 435/40 514) of the predicted genes: 18.1 × 0.41 = 7.4. There are some obvious assumptions in this calculation – principally that differentially expressed genes are not genetically linked and are evenly spread throughout the genome. Our results therefore suggest a greater number of co-locations than would be expected by chance.

We are now undertaking work to map the expression of these genes, and other genes identified from this study, within the F2 population, and believe that the approach undertaken here has provided a highly efficient means by which to identify the genes of greatest interest and potential for eQTL mapping.


We have shown that P. deltoides and P. trichocarpa have contrasting responses to drought, and have used a genetic and genomic approach to study the genetics of this difference. We have shown that the divergent drought response of two Populus species exhibits transgressive segregation within an F2 population, and that this results in the emergence of highly contrasting adaptive drought responses. We observed dramatic segregation within the population in the abscission response to drought, and comparison of the transcriptional response of a set of high- and low-abscission genotypes revealed a striking and perhaps surprising degree of separation. Importantly, we have also shown that an experimental design using few microarrays serves as an efficient means to identify genes with segregating patterns of gene expression within a population.

Although the common transcriptional responses to drought stress provide information on genes and regulons induced by drought stress (i.e. genes typically expressed by dormant tissues), this type of study does not necessarily provide good candidates for genes responsible for natural variation in this trait, as it is quite likely that the different alleles of the genes regulating the drought response may not have particularly contrasting transcript levels. Instead, we believe that the ‘genetical genomics’ approach may be needed to understand the drought response at the level of genetic variation. A number of genes with differential levels of expression response to drought between the two extreme groups co-located to genomic regions identified by QTL analysis. These genes may provide clues as to the genetic mechanisms through which species adaptation to drought has been achieved. A similar study in rice identified large-scale divergence of the transcriptional response for genotypes with divergent osmotic adjustment responses to drought stress (Hazen et al., 2004), and this may suggest that gene expression control plays a key role in the mechanisms of species divergence.

Experimental procedures

Plant material and growth conditions

Grandparental response to drought.  Thirty uniform cuttings each of P. deltoides (clone ILL-129) and P. trichocarpa (clone 93-968) were obtained from a field site in the UK (see Rae et al., 2004 for details) Following storage at 4°C, cuttings were soaked in cold water at room temperature for 24 h prior to planting, and were then planted into rigid plastic tubes (length 75 cm, diameter 13 cm) with one bud above the soil line. The tubes were filled with 10 kg John Innes No. 3 compost (JI3; John Innes Manufacturers Association, Harrogate, UK).

Cuttings were grown in controlled growth rooms at the University of Southampton. The day and night temperatures were 22 and 16°C, respectively, photosynthetically active radiation was 160 μmol m−2 sec−1, and day length was 16 h. The vapour pressure deficit (VPD) was uncontrolled but showed limited fluctuation around 50% relative humidity (RH). All cuttings were watered to field capacity for 49 days. The drought treatment was initiated 50 DAP (days after planting). Water was completely withheld from stressed trees (n = 5). Control plants (n = 10) were watered to field capacity. Trees were given a balanced NPK (N:P:K, 14:13:13) fertilizer (Osmocote, Scotts Europe BV, Heerlen, The Netherlands) throughout the experimental treatment.

Natural variation in population response to drought.  Hardwood cuttings of 167 genotypes of family 331 (referred to as POP1 in the POPYOMICS project, were obtained from a field site in the UK (see Rae et al., 2004 for details). Cuttings were stored at 4°C until used, then soaked in water at room temperature for 24 h before planting into 25 l plastic pots (one cutting per pot) filled with lightly compacted JI3 compost. Cuttings were planted with one bud above the soil line. Plant material was grown in a greenhouse facility near to the University of Southampton. Cuttings were pre-assigned to random positions within the experiment, and each pot was labelled with genotype, position and treatment. Each genotype was replicated three times in each condition, with each replicate being assigned to a random position in both control and drought treatments (total number of plants was 1002). Randomization was achieved by sequentially numbering pots within the greenhouse and then randomly assigning a pot number to a combination of genotype–treatment–replicate. Cuttings were planted on 5 June 2004. During establishment, all cuttings were watered daily by overhead sprinklers. Once established, daily watering continued by hosepipe. The drought treatment was initiated at 131 DAP. Water was withheld from drought trees for a period of 7 days. Drought trees were then given 0.5 l of water and soil drying continued. Percentage soil moisture (%SM) was recorded 17 days after drought (DAD) using a Delta-T ML2x ThetaProbe connected to an HH2 moisture meter (Delta-T Devices, Cambridge, UK). The average %SMs were 27% and 15% for control and drought treatments, respectively. An anova test for treatment and genotype was used to confirm that no genotype or treatment × genotype effects were present for %SM, and additionally no location (within greenhouse) effects were identified. Control trees were watered to field capacity throughout. Trees were sampled 17 DAD. The site was not affected by shading, so within-site variation in light, temperature, airflow and humidity was minimal. The average day temperature was 26°C and the average night temperature was 22°C. Day length and VPD were uncontrolled but the average recorded RH was 63%.

Physiological measurements for grandparental response to drought

Leaf area expansion.  The leaf plastochron index (LPI) was calculated as described by Erickson and Michelini (1957), and was shown to be applicable to poplar by Larson and Isebrands (1971). Detailed discussion can be found in Taylor et al. (2003). The time per LPI was calculated by examining the growth of individual leaves. The leaf area for each LPI was recorded at 61 DAP. A digital image of each leaf and a scale bar was obtained using a Nikon CoolPix 5000 camera (Nikon, Surrey, UK), and leaf area was calculated from the image using MetaMorph version 5 (Universal Imaging Corporation, Philadelphia, PA, USA).

Photosynthesis and transpiration.  Physiological gas exchange data for stomatal conductance and rate of photosynthesis were collected using a LiCor ‘6400 IRGA’ (Infra Red Gas Analyzer; LiCor, Lincoln, Nebraska, USA). Measurements were taken on leaves LPI 0, LPI 5 and LPI 9 1 day and 9 days after the initiation of drought (52 and 60 DAP). IRGA chamber parameters were: photosynthetically active radiation 300 μmol m−2 sec−1, CO2 400 μmol mol−1 and a constant fixed flow of 400 μmol sec−1. Measurements were made on attached leaves at mid-day.

Natural variation in population response to drought

The plastochron index (and consequently LPI) have been shown to be difficult to use as markers of developmental stage with this population (unpublished data), given the very large differences in leaf production rate and the contrasting durations of leaf expansion. Leaf l0 was defined as the first fully unfurled leaf. At 9 DAD , the petiole of leaf l0 was tagged with coloured cotton thread and a digital image of the leaf was acquired (Canon EOS 300D, Canon UK Ltd, Reigate, UK). Therefore, at 17 DAD, leaf l6 was flash-frozen in liquid nitrogen for subsequent analysis of transcript levels. Sampling was started at 11 am and was completed for all replicates by 12 am. Control and treated (drought-stressed) replicates for each genotype were sampled simultaneously to negate any time sampling effects. A 12 mm leaf disc from leaf l4 was sampled into DMF (dimethylformamide) for subsequent analysis of chlorophyll content and carotenoid content. At 18 DAD, all plants were scored for leaf number, the number of abscised leaves, the number of leaves with necrotic lesions, and the number of chlorotic leaves. Tree height was also scored to the nearest cm. A further image of the same leaf was then acquired at 13 DAD. Leaf area, leaf length (along the mid-rib) and leaf width (at widest point at right angle to leaf mid-rid) for both days were then measured from the digital images using ‘ImageJ’ ( Expansion and extension rates and the length:width ratio were then calculated.

Physiological data analysis

All data analysis and manipulation were performed using the statistical language R ( with the ‘nlme’ and ‘nortest’ packages. Data were first filtered to remove all genotypes that had fewer than two replicates present in both conditions. For the traits abscission, necrosis and chlorosis, the percentage of leaves that had abscised etc was calculated as (trait value/number of leaves) × 100. The percentage effect of drought was calculated for each trait using the formula: [(control−drought)/control] × 100. Normality was then tested for all traits, and data transformed using Box–Cox normalization where required. Homogeneity of variance was also tested using a Bartlett test. For normally distributed data, a two-way anova test was conducted with genotype set as a random factor. The normality of the residuals was additionally tested using an Anderson–Darling test (all traits returned a non-significant result). Non-normal data were tested using a Kruskall–Wallis test. Genotype means were calculated for all traits and exported for QTL analysis. Chlorophyll a, b, total chlorophyll and carotenoid content were calculated. Chlorophyll content was measured by measuring the absorption of DMF extracts (diluted 1:4) at 647, 664 and 480 nm using a spectrophotometer (Hitachi U20000; Hitachi Ltd., Tokyo, Japan) as described by Wellburn (1994) using the calculations stated for the 1–4 nm range. Subsequently the ratios chlorophyll a:b and carotenoid:total chlorophyll were calculated.

A set of ‘extreme’ sensitive and insensitive genotypes were defined on the basis of percentage response to drought for the trait percentage of abscised leaves. This trait was selected as a late-onset indicator of drought tolerance. The five highest and lowest ranked genotypes were selected for subsequent transcriptome analysis. These genotypes can be considered as transgressive segregants from the F2 population used.

QTL mapping

QTL were mapped using the freely available web-based program QTLExpress (Seaton et al., 2002). The out-breeding module of the program was used. Permutation testing implemented in QTLExpress was used to establish the critical F value for declaring a QTL present (1000 permutation; see Churchill & Deorge, 1994). QTL confidence intervals (CIs) were calculated using an F-2 drop-off (the cM distance taken for the F value to drop by two from the highest F value location). The genetic linkage map used was produced by G.A. Tuskan et al. (personal communication), and consists of 91 SSR markers genotyped on 350 of the full-sib progeny, and 92 fully informative amplified fragment length polymorphisms (AFLP) genotyped on 165 genotypes of the progeny. The resulting genetic map consists of 22 linkage groups. These have been aligned to the physical sequence of Populus and are numbered accordingly. Where more than one LG has been assigned to a chromosome, they are numbered with the LG number and a letter, with the letter order indicating the order of LGs along the chromosome. All genotypes were full-sib progeny (referred to here as the F2 generation) of family 331, an intra-Americana cross between Populus trichocarpa (93-968) and P. deltoides Bart (ILL-129) described by Bradshaw and Stettler (1993). The population is known to be highly divergent for a vast range of phenotypic traits (see, for example, Rae et al., 2004). SSR primer sequences ( were located on the genome sequence to align the genetic and physical maps and to provide correct orientation of linkage groups (i.e. 3′–5′).

Microarray experimental design

To assess the grandparental response to drought, at mid-day at 65 DAP (14 days of drought treatment), leaf LPI 0 from control and drought trees was flash-frozen in liquid nitrogen for subsequent RNA extraction (Doyle and Doyle, 1987: modified by Chang et al., 1993) and microarray analysis. Three of the biological replicates were randomly selected for RNA extraction. RNA was prepared as described above and in Chang et al. (1993) with the following modifications. No spermadine was used in the extraction buffer, 2.67%β-mercaptoethanol was used, and an additional chloroform extraction was performed after the LiCl precipitation. RNA concentrations were determined spectrophotometrically (GeneQuant, Amersham-Pharmacia Biotech, Uppsala, Sweden), and RNA quality was assessed by running a 1% w/v agarose gel. Further details of the array procedure can be found below in the ‘Common microarray method’ section.

Before conducting control versus drought arrays, a number of self–self hybridizations were performed in order to test for evidence of differential hybridization of the two species to the microarrays used. The microarrays are constructed from tissues of a number of species, but principally Populus tremula × tremuloides‘T89’ and Populus tremula ( These arrays showed no evidence for hybridization bias when tested for species effect using anova (the arrays are available in upsc-base). Such an anova design can, however, only test for global-scale differences in expression in the equivalent of a binary present/absent test (all genes are expected to have a ratio-of-expression of 1 on self–self arrays). Initial array results were also confirmed using RT-PCR (Figure S2), and these confirmed the array results obtained. Considering the close relationship between the two species and the average length of the ESTs spotted on the arrays, we do not feel that homology issues will significantly affect genes identified as differentially expressed. However, it should be expected that some genes will have diverged more than others, and all candidates should be sequenced in both grandparent species to identify possible polymorphisms before being used for further downstream application.

In total, 12 successful microarrays were hybridized, representing three independent biological replicates from each species and each treatment. Direct comparisons were made between treatment and control within each species (Figure 4a). Species comparisons were made indirectly by comparing across arrays. A technical replicate of each slide was made with dyes in the reverse orientation, as discussed by Cui and Churchill (2003). The microarrays used have been described by Andersson et al. (2004), and are referred to as POP1 arrays on the Umeå Plant Science Centre (UPSC) PopulusDB website ( The POP1 array is a cDNA microarray containing 13 490 elements spotted in duplicate. These elements were selected from 36 354 ESTs obtained from seven cDNA libraries. All EST sequences were annotated and functionally classified, and full information is available from the PopulusDB website ( Data analysis methods are described below in the ‘Common microarray method’ section.

To assess the extreme genotype response to drought, leaf l6 from each replicate of the extreme genotypes was used for RNA extraction. RNA was extracted from leaf material as detailed in the grandparent experiment above. However, RNA concentration was measured using a Nanodrop spectrophotometer (Nanodrop Technologies, Wilmington, DE, USA), and quality was checked using a bioanalyser (Agilent 2100 Bioanalyzer, Agilent Technologies, Waldbronn, Germany). A pooled analysis strategy was then used for microarray investigation. Four RNA pools were prepared: high-abscission, control (Hc, n = 11); high-abscission, drought (Hd, n = 12); low-abscission, control (Lc, n = 9); low-abscission, drought (Ld, n = 11). RNA was extracted from individual replicates of each extreme genotype, and a pool of genotypes within each extreme group was then created for transcriptional analysis. For all genotypes except one, there were a minimum of two biological replicates per genotype. The number of biological replicates of each of the five genotypes within each pool reflects the number of surviving plants. Therefore, the number of replicates within each pool is a reflection of this (the maximum number of replicates in a pool would be 15 if all replicates of all genotypes had survived). The amount of RNA used to construct the pools from each genotype was normalized to ensure that each genotype had equal representation within the pool. A loop design was then used to make all possible comparisons (Figure 4b). For each point of the loop, a technical replicate was performed with one replicate in the dye orientation Cy5/Cy3 and the other in the opposite dye orientation (Cy3/Cy5) to allow for dye effects to be tested for. The entire loop was then technically replicated, giving a total of 24 microarrays. The POP2 array was used for this experiment and contains 24 735 probes representing >100 000 ESTs from 18 tissues (described by Sterky et al., 2004). Full details are available from the PopulusDB website. Data analysis methods are described below in the ‘Common microarray method’ section.

Common microarray method

cDNA synthesis.  Aliquots (50 μg) of total RNA suspended in 9 μl DEPC-H2O were denatured with 1 μl oligo(dT)-anchor (Cybergene AB, Huddinge, Sweden) at 70°C for 5 min and then cooled on ice. mRNA was then reverse-transcribed with 6 μl 5x RT buffer, 0.6 μl 50x dNTP mix (25 mm dATP, dCTP, dGTP, 20 mm aa-dUTP, 5 mm dTTP), 3 μl 10 mm DTT, 1 μl RNase inhibitor (30 U, Invitrogen, Stockholm, Sweden), 1.5 μl Superscript II (300 U, Invitrogen) and 7.9 μl DEPC-H2O by incubating at 42°C for 3 h. The reaction was stopped with 10 μl 0.5 m EDTA, and RNA was degraded by adding 10 μl 1 m NaOH and incubating at 65°C for 15 min. The reaction was then neutralized by addition of 50 μl 1 m HEPES (pH 7.0).

cDNA was purified using Qiaquick columns (Qiagen, Stockholm, Sweden) according to the manufacturer's instructions with the exception that the wash buffer was replaced by a phosphate wash buffer (pH 8.0, 5 mm KPO4; cDNA was eluted twice with 30 μl ddH2O with a 1 min incubation for each elution. Samples were dried in a Speedvac (Savant DNA SpeedVac, Thermo Electron Corporation, Waltham, MA, USA) at 40°C for 60 min.

Indirect Cy3/5 dye coupling.  Dyes were coupled by first re-suspending the Cy3/Cy5 (Amersham Biosciences, Uppsala, Sweden) in 120 μl 0.1 m NaHCO3 (pH 9.0); 15 μl of Cy3/Cy5 was then added to the dry cDNA. Dyes were coupled in the dark for 2.5–3 h at room temperature. Cy3-labelled cDNA was purified using a Qiaquick column as described above, with the exception of an extra washing step and that labelled cDNA was eluted with 41 μl phosphate elution buffer (pH 8.5, 4 mm KPO4; that was incubated on the membrane for 1 min. The Cy5-labelled target was then eluted into the same Eppendorf tube in the same manner.

Hybridization.  Hybridization was performed in an automated slide processor (Lucidea ASP hybridisation station, Amersham-Pharmacia Biotech). Pre-hybridization buffer was 50% formamide, 5x SSC, 2.5x Denhart's solution. The hybridization solution contained the labelled cDNA, 25% formamide, 5x SSC, 0.22% SDS, 1 μl tRNA and 0.42 μg oligo-dA (80-mer). Wash buffer 1 was 0.8x SSC, 0.03% SDS. Wash buffer 2 was 0.2x SSC. Wash buffer 3 was 0.05x SSC, 2 mm KPO4. Isopropanol (100%) was used to clean slides after washing.

Scanning and image analysis.  Arrays were scanned at 5 μm resolution, using a Scanarray 4000 microarray analysis system scanner (Perkin-Elmer, Boston, MA, USA). Scanner settings were PMT (Photo Multiplier Tube) 80% for Cy5 and 85% for Cy3, and laser power of 90–99% depending on signal strength for the acute experiment. For the extreme experiment, arrays were scanned at four settings of increasing laser power and PMT (laser power – 60, 80, 100, 100%; PMT – 70, 70, 70, 80%) at 10 μm resolution, using a ScanarrayLite microarray analysis system scanner (Perkin-Elmer). Regression analysis was then applied to the scans using a upsc-base plug-in. This produces a unified data file from all four scans with the effect of increasing the dynamic range of intensities for which spot intensity data can be extracted (Dudley et al., 2002). Spot data were extracted using GenePix versions 4.1 and 5.0, respectively (Axon Instruments Inc, Union City, CA, USA) for the grandparent and population experiments. Settings for the spot diameter re-size feature were set to <75% and >150%, and CPI (composite pixel intensity) was set to 300.

Data analysis

The data output from GenePix was imported into upsc-base, and is publicly available ( (seeSjödin et al., 2006, for details of upsc-base). All data were examined for quality control purposes using plug-ins integral to the upsc-base analysis pipeline (see Sjödin et al., 2006). Background subtraction was achieved by removing the local median background intensity from the spot foreground median intensity. A print-tip LOWESS normalization was then applied, and spots flagged as bad had a negative weighting factor of 0.1 applied to them. Data were subsequently filtered based on A value [(log2 R + s log2 G)/2] to remove spots with low intensities in both channels. The log2 A value threshold was set to 8.0, corresponding to a raw intensity of 256. B statistics implemented in the LIMMA package for R (Smyth, 2004; and made available as a upsc-base plug-in were then used to select genes with a high probability of differential expression in response to drought for the grandparent experiment and for all comparisons between the population extreme groups. A B value of zero indicates a 50:50 probability of differential expression, and, as the B value increases, so does the probability that the gene is differentially expressed. In this way, the B value can be considered as a confidence measure of differential expression. We used an arbitrary cut-off of five to yield lists of genes with a high probability of differential expression. FDR-adjusted P values are also given.

Grandparent species differences were tested by exporting normalized data from upsc-base and importing them into GeneSpring version 7.2 (Agilent Technologies, Redwood City, CA, USA) where a one-way anova was performed for species effect, both with and without Bonferroni correction. These gene lists were then re-imported into upsc-base for visualization and functional analysis.

GO categories significantly over-represented by differentially expressed genes were identified using a modified version of the GoHyperG function from the GOstat package available via Bioconductor (Gentleman et al., 2004; and as implemented in upsc-base, with GO-directed acyclic graphs (DAGs) produced using the GOGraph function of the same package. The GoHyperG function uses an equivalent form of Fisher's exact test to identify over-representation at certain nodes in the DAG produced.

In silico comparisons with PopulusDB

To obtain digital expression profiles of genes with B values >5 for the grandparent and extreme arrays, the library distribution of all clusters corresponding to represented uni-genes was exported from PopulusDB using a upsc-base plug-in. For each of the 100 genes, clone frequency in a particular library was determined, and the resulting tables (one for upregulated and one for downregulated genes) were analysed according to the method described by Ewing et al. (1999) and clustered correlation maps were generated. To calculate ROF (relative occurrence factors), the similarities between transcriptomes and occurrence in a particular library (yes or no) were scored for each of the genes. The relative frequency of clusters of the upregulated list found in each of the libraries was calculated, for example 18 of the 74 (24%) most upregulated clusters/genes were found in the young leaf (C) library, and 19 of the 26 most downregulated (73%). ROF was calculated as the ratio of these relative frequencies, i.e. 24/73.

All microarray data collected from the two experiments are available to download from the upsc-base website (, as are all gene lists discussed. The grandparent and population experiments are numbers 0009 and 0036, respectively. Upon request, all novel materials described in this publication will be made available in a timely manner for non-commercial research purposes.


This research was supported by the European Commission through the Directorate General Research within the Fifth Framework for Research – Quality of Life and Management of the Living Resources Programme, contract number QLK5-CT-2002-00953 (POPYOMICS), coordinated by the University of Southampton. We thank Caroline Dixon for her assistance during all stages of planting and sampling in both experiments, and Toby Bradshaw for the original supply of plant material used in this study. We also thank Anne Rae, Penny Tricker, Sarah Elhag, Sera Bowden, Mike Cotton, Charlotte Freer-Smith, Dan Stapleton, Matthieu Pinel, Harriet Trewin, Nicole Harris, Laura Graham and Matthew Tallis for their help planting and sampling the population experiment, and Anne Rae for her advice on the use of QTL mapping approaches. We are grateful to G.A. Tuskan and co-workers for supply of the improved molecular genetic map of POP1, made freely available to us. N.R.S. and J.T. thank Natural Environment Research Council (NERC) for the award of research studentships and G.T. acknowledges support for research on Populus from the Department for Environment, Food and Rural Affairs (DEFRA) to her laboratory (grant numbers NFO410 and NFO424).