Bud set in poplar – genetic dissection of a complex trait in natural and hybrid populations


  • Antje Rohde,

    1. Department of Plant Systems Biology, Flanders Institute for Biotechnology (VIB), BE–9052 Gent, Belgium
    2. Department of Plant Biotechnology and Genetics, Ghent University, BE–9052 Gent, Belgium
    3. Institute for Agricultural and Fisheries Research, BE–9090 Melle, Belgium
    Search for more papers by this author
  • Véronique Storme,

    1. Department of Plant Systems Biology, Flanders Institute for Biotechnology (VIB), BE–9052 Gent, Belgium
    2. Department of Plant Biotechnology and Genetics, Ghent University, BE–9052 Gent, Belgium
    Search for more papers by this author
  • Véronique Jorge,

    1. Unité Amélioration, Génétique et Physiologie Forestières, Institut National de la Recherche Agronomique, F–45075 Orléans Cédex 2, France
    Search for more papers by this author
  • Muriel Gaudet,

    1. Dipartimento di Scienze dell’Ambiente Forestale e delle sue Risorse, Università degli Studi della Tuscia, I–01100 Viterbo, Italy
    Search for more papers by this author
  • Nicola Vitacolonna,

    1. Dipartimento di Matematica e Informatica, Università di Udine, I–33100 Udine, Italy
    Search for more papers by this author
  • Francesco Fabbrini,

    1. Dipartimento di Scienze dell’Ambiente Forestale e delle sue Risorse, Università degli Studi della Tuscia, I–01100 Viterbo, Italy
    Search for more papers by this author
  • Tom Ruttink,

    1. Department of Plant Systems Biology, Flanders Institute for Biotechnology (VIB), BE–9052 Gent, Belgium
    2. Department of Plant Biotechnology and Genetics, Ghent University, BE–9052 Gent, Belgium
    3. Institute for Agricultural and Fisheries Research, BE–9090 Melle, Belgium
    Search for more papers by this author
  • Giusi Zaina,

    1. Dipartimento di Scienze Agrarie e Ambientali, Università di Udine, I–33100 Udine, Italy
    Search for more papers by this author
  • Nicolas Marron,

    1. Research Group of Plant and Vegetation Ecology, Department of Biology, University of Antwerp, BE–2610 Wilrijk, Belgium
    Search for more papers by this author
  • Sophie Dillen,

    1. Research Group of Plant and Vegetation Ecology, Department of Biology, University of Antwerp, BE–2610 Wilrijk, Belgium
    Search for more papers by this author
  • Marijke Steenackers,

    1. Research Institute for Nature and Forest, BE–9500 Geraardsbergen, Belgium
    Search for more papers by this author
  • Maurizio Sabatti,

    1. Dipartimento di Scienze dell’Ambiente Forestale e delle sue Risorse, Università degli Studi della Tuscia, I–01100 Viterbo, Italy
    Search for more papers by this author
  • Michele Morgante,

    1. Dipartimento di Scienze Agrarie e Ambientali, Università di Udine, I–33100 Udine, Italy
    Search for more papers by this author
  • Wout Boerjan,

    1. Department of Plant Systems Biology, Flanders Institute for Biotechnology (VIB), BE–9052 Gent, Belgium
    2. Department of Plant Biotechnology and Genetics, Ghent University, BE–9052 Gent, Belgium
    Search for more papers by this author
  • Catherine Bastien

    1. Unité Amélioration, Génétique et Physiologie Forestières, Institut National de la Recherche Agronomique, F–45075 Orléans Cédex 2, France
    Search for more papers by this author

Author for correspondence:
Antje Rohde
Tel: +32 9 2722967
Email: antje.rohde@ilvo.vlaanderen.be


  • The seasonal timing of growth events is crucial to tree distribution and conservation. The seasonal growth cycle is strongly adapted to the local climate that is changing because of global warming. We studied bud set as one cornerstone of the seasonal growth cycle in an integrative approach.
  • Bud set was dissected at the phenotypic level into several components, and phenotypic components with most genetic variation were identified. While phenotypic variation resided in the timing of growth cessation, and even so more in the duration from growth cessation to bud set, the timing of growth cessation had a stronger genetic component in both natural and hybrid populations.
  • Quantitative trait loci (QTL) were identified for the most discriminative phenotypic bud-set components across four poplar pedigrees. The QTL from different pedigrees were recurrently detected in six regions of the poplar genome.
  • These regions of 1.83–4.25 Mbp in size, containing between 202 and 394 genes, form the basis for further molecular-genetic dissection of bud set.


The seasonal cycle of growth and dormancy is a distinct feature of perennial plants and represents one of the most basic adaptations of trees to their environment. Many events in the annual growth cycle, such as bud flush, growth, growth cessation, bud set, bud dormancy and release from dormancy through chilling, are regulated by aspects of local climate. The recurrent transitions of meristems between growth and dormancy are tightly linked to the yearly dates of bud flush and bud set. These two events delimit the growing season and are under strong selection to avoid cold injury by either early frost in autumn or late frost in spring. Bud flush and bud set show strong genetic differentiation along latitudinal and altitudinal clines, typically resulting in locally adapted ecotypes (Howe et al., 2003; Savolainen et al., 2007; Aitken et al., 2008). Along large geographical clines, bud set is often more differentiated among populations than bud flush. QST values, which estimate the proportion of the total trait variation between populations, are generally higher for bud set than for bud flush (Howe et al., 2003).

In the light of climate change, bud flush and bud set have gained considerable attention. Changes in phenology have been noted during the past decades and informed the discussion on whether trees will cope with the expected speed of climate change (Saxe et al., 2001; Aitken et al., 2008; Marris, 2009). Because phenology is so crucial to tree distribution and conservation, research efforts to understand its genetic basis are of paramount importance.

Bud set is the net result of growth cessation and bud formation. Many broad-leaved trees cease growth in response to increasing night length, sometimes in combination with or completely dependent on decreasing temperature (Nitsch, 1957; Heide, 1974; Junttila, 1982). In poplar, increasing night length is perceived and mediated through phytochrome and the CONSTANS (CO)/FT regulon (Howe et al., 1996; Olsen et al., 1997; Böhlenius et al., 2006). In addition to growth cessation, autumnal bud development is a composite of three processes, namely bud formation, simultaneous acclimatization to dehydration and cold, and acquisition of dormancy (Ruttink et al., 2007).

Once a critical night length – the night length triggering growth cessation – is achieved, it takes several weeks until the completion of bud set. During this time dynamic changes take place that are reflected at both the molecular and phenotypic levels (Ruttink et al., 2007). Traditional measurements of bud set have focused only on the timing of bud set, when all biological processes are completed (Frewen et al., 2000). To better understand the genetic components and the plasticity of bud set, it is important to also describe the initiation of bud set at cessation of growth and its dynamics until bud set. These aspects may contribute significantly to genetic variation at both among- and within-population levels.

The aim of this study was to dissect bud set at the phenotypic level into its components, to identify those expressing a high level of genetic variation, and to detect genomic regions associated to this genetic variation by quantitative trait loci (QTL) mapping. First, we developed a new scoring system for bud set to measure all aspects of the phenotype, including the onset and the duration of autumnal bud development. Bud set was assessed in a wide collection of European riparian Black poplars (Populus nigra) and in four interspecific and intraspecific poplar pedigrees. The elaborate phenotypic data revealed that phenotypic variation resided in both the timing of cessation of growth and the duration from cessation of growth to bud set. Although the duration of bud formation varied at the phenotypic level, it contained, under the environmental conditions of this experiment, a smaller genetic component than the timing of growth cessation. Based on this information, quantitative trait loci for the most discriminative components of bud set, covering onset of growth cessation and duration of bud formation, were detected across four poplar pedigrees. The QTL from different pedigrees coincided in six regions of the poplar genome. Together, we integrate all information from phenotype, the contribution of its components to genetic variation, the genomic regions that control bud set variation and candidate genes. Because, this study is based on heterogeneous genetic material, the shared QTL regions provide a robust result to initiate the identification of the underlying genes.

Materials and Methods

Genetic material and field sites

A collection of European Populus nigra A total of 466 P. nigra genotypes were collected along riparian ecosystems in five European countries and planted in Geraardsbergen (Belgium) (Table 1; Supporting Information, Table S1). Bud set was scored for 437 genotypes with at least three ramets per genotype. The following populations had been previously sampled and characterized with neutral markers as true populations: Drôme1, Drôme6, TicinoSN, TicinoN, Kühkopf, Ebro1 and Ebro2. The overall genetic differentiation between populations (FST) is low to moderate with values between 0.03 and 0.28 (van Dam & Bordács, 2001; Imbert & Lefèvre, 2003; Smulders et al., 2008). The regional collections consisted of the individual trees from narrow regions (Durance, LoireE, LoireW, Ijssel/Rhine, and Waal/Maas; Storme et al., 2004). A few singular genotypes from France and the Netherlands were included (individual trees F and NL, respectively).

Table 1.   The collection of Populus nigra accessions from Europe and means for growth cessation and duration of bud formation. A total of 437 genotypes (2217 individuals), collected in riparian ecosystems in Europe were evaluated in a common garden in Belgium (50°N). Bud set was evaluated twice a week from day 234 to day 318 (24 times)
CountryNameTypeLocation (nearest town or landmark)FIS1Latitude at originEffective number of genotypesEffective number of individualsdate2.5_cnl (mean ± SE)subproc1_Δcnl (mean ± SE)subproc2_Δcnl (mean ± SE)
  1. 1FIS based on 105 amplified fragment length polymorphism (AFLP) markers (Smulders et al., 2008).

SpainEbro1PopulationNear Novillas0.022141°N1975915.72 ± 6.67138.08 ± 1.18159.78 ± 1.89
SpainEbro2PopulationNear Alfranca0.012741°N32136931.40 ± 6.24134.79 ± 1.09160.44 ± 1.88
FranceDuranceRegional collectionAlong the Durance 43°N1041751.99 ± 14.19146.47 ± 2.02154.95 ± 2.82
FranceDrôme1PopulationAlong the Drôme0.092744°N61315766.85 ± 4.54149.31 ± 1.15157.21 ± 1.01
FranceDrôme6PopulationAlong the Drôme0.132144°N63346803.61 ± 4.56142.16 ± 0.84158.98 ± 1.36
ItalyTicinoSNPopulationNear La Zelata 45°N42213817.01 ± 4.69135.76 ± 0.67146.28 ± 0.96
ItalyTicinoNPopulationInside Bosco Siro Negri 45°N60298822.37 ± 4.45137.73 ± 0.55152.48 ± 0.97
FranceIndividual trees FIndividual trees  44°–47°N529832.08 ± 16.84146.13 ± 0.76162.53 ± 2.55
FranceLoireERegional collectionEastern part Loire 46°–47°N22103621.57 ± 7.98145.63 ± 2.41156.22 ± 2.86
FranceLoireWRegional collectionWestern part Loire 47°N1989686.06 ± 13.89141.83 ± 2.01162.42 ± 1.79
GermanyKühkopfPopulationGinsheim/Kühkopf0.029649°N54285628.87 ± 3.03143.17 ± 0.88161.74 ± 1.09
The NetherlandsWaal/MaasRegional collection  51°N1272556.55 ± 2.86109.74 ± 1.28137.13 ± 1.17
The NetherlandsIndividual trees NLIndividual trees  50°–52°N738602.88 ± 9.54127.81 ± 2.59146.91 ± 2.61
The NetherlandsIjssel/RhineRegional collection  50°–52°N31177577.17 ± 4.95113.54 ± 1.08138.05 ± 0.85

Pedigrees  Four intraspecific and interspecific poplar F1 hybrid families were used (Table 2; Table S1). POP2 with 330 progeny was derived from an interspecific cross of Populus deltoides‘73028-62’ (IL, USA; 37°32′ N 89°49′ W) and Populus trichocarpa‘101-74’ (Nisqually River, WA, USA; 47° N 123° W) (INRA, Orleans, France) (Jorge et al., 2005). POP3a and POP3b were two hybrid F1 interspecific poplar maternal half-sib families (INBO, Geraardsbergen, Belgium) (Dillen et al., 2009). POP3a consisted of 180 progeny of an interspecific cross between P. deltoides‘S9-2’ and P. nigra‘Ghoy’ (Ghoy, Belgium; 50°43′ N 3°47′ E). P. deltoides‘S9-2’ was derived from a cross of P. deltoides‘V1’ (Ontario, Canada; 42°40′ N 80°10′ W) and P. deltoides‘V5’ (Bellevue, IA, USA; 42°15′ N 90°45′ W). POP3b consisted of 182 progeny of an interspecific cross involving the same female parent P. deltoides‘S9-2’ and P. trichocarpa‘V24’ (Camas, OR, USA; 45°30′ N 122°40′ W). POP5 consisted of 165 progeny of an intraspecific cross of P. nigra‘58-861’ (Val Cenischia, Italy; 45°09′ N 7°01′ E) with P. nigra‘Poli’ (Sinni river, Italy; 40°09′N 16°41′E) (DISAFRI, Viterbo, Italy) (Gaudet et al., 2008).

Table 2.   Interspecific and intraspecific pedigrees and progeny means for growth cessation and duration of bud formation. Pedigrees POP2, POP3a and POP3b were evaluated in France and POP5 was evaluated in Italy
PedigreeFemale parentMale parentLatitudinal origin (female parent)Latitudinal origin (male parent)Latitude of progeny selectionScoring period1Effective progeny size2date2.5_cnl (mean ± SE, n)subproc1_Δcnl (mean ± SE, n)subproc2_Δcnl (mean ± SE, n)
  1. 1Scoring period is given in days of the year. The number of measurements during this period of time is given within parentheses.

  2. 2Effective progeny size includes genotypes with no fewer than three individual trees with data.

  3. 3n corresponds to the number of genotypes with data for the trait. See also the Supporting Information, Table S1. For POP3a and POP3b, average performances related to date2.5_cnl and subproc1_Δcnl are biased at population level because information is available for a limited number of genotypes.

POP2P. deltoides‘73028-62’P. trichocarpa‘101-74’37°N47°N47°N240–279 (10)330649.73 ± 0.95 (n = 234)107.96 ± 1.18 (n = 234)143.42 ± 1.45 (n = 325)
POP3aP. deltoides‘S9-2’P. nigra‘Ghoy’42°N50°N50°N231–256 (8)179464.75 ± 2.07 (n = 64)68.00 ± 1.65 (n = 64)144.29 ± 2.00 (n = 138)
POP3bP. deltoides‘S9-2’P. trichocarpa‘V24’42°N45°N50°N244–262 (6)179592.74 ± 3.85 (n = 35)71.88 ± 2.27 (n = 35)106.72 ± 1.43 (n = 161)
POP5P. nigra‘58-861’P. nigra‘Poli’45°N40°N44°N252–286 (17)152863.25 ± 2.66 (n = 151)100.08 ± 1.44 (n = 151)124.05 ± 2.22 (n = 139)

Field sites  A completely randomized block design with six blocks and one ramet per genotype and block was used for the establishment of experimental plantations in Belgium (Geraardsbergen; 50°46′ N 3°52′ E; P. nigra collection), Central France (Ardon, Loire valley; 47°46′ N 1°52′ E; POP2, POP3a, and POP3b), and Northern Italy (Cavallermaggiore, Po valley; 44°21′ N 8°17′ E; POP5). Trees were planted at 0.75 m by 2 m distance. A double border row of poplars was planted around the experimental plot to reduce border effects. The field trials were established from hardwood cuttings in spring 2003 (France and Italy) and spring 2004 (Belgium). The sites were irrigated (only France and Italy), weed controlled and treated with insecticides and fungicides as necessary throughout the growing seasons. Trees were cut back in January 2005 and only the best-performing shoot was preserved after resprouting. Cumulative night length (cnl) was calculated from daylength data retrieved from the US Naval Observatory (http://aa.usno.navy.mil/) with the respective geographic coordinates of the field sites. Temperature sums after July 1 were calculated from the temperature records at the respective field sites. These considered either the daily mean temperature (cumulative mean temperature, cmt) or daily minimum temperature (cumulative daily minimum temperature when below 10°C, cmt10). When the daily minimum temperature was below 10°C, the minimum temperature was subtracted from 10°C to give lower temperature values a larger contribution to cmt10. July 1 was chosen to separate from the first growth flush that is typically accomplished before June 24. Both Δcnl and Δcmt10, used to describe the temperature sums for duration traits, were calculated by subtracting cnl and cmt10 at stage 1.5 from those at stage 2.5 (subproc1) or by subtracting cnl and cmt10 at stage 0.5 from those at stage 1.5 (subproc2).

Data collection, treatment and statistics

Bud set scoring and data analysis  Bud set was scored in autumn 2005, in the main apical bud of the 1-yr-old shoot on 2-yr-old (P. nigra collection) and 3-yr-old (pedigrees) roots, applying the scoring scheme (Fig. 1). Local polynomial regression was used to smooth the bud-set curve of each individual and to estimate dates of bud-set stages that had not been observed in the field because of the speed of bud set. Observed scores were fitted to a local polynomial regression of degree 2 with, as predictors, the different dates of observation in the given site (Cleveland et al., 1992). Date x was fitted using the 75% nearest data points and weighted least squares. High quality of fit was obtained for all individual plants with R2 ranging from 0.885 to 0.999. Pearson correlations between observed and predicted dates surpassed 0.986 for all bud-set stages. From the fitted curve, discrete values for the day of the year were retrieved for all scores from 3 to 0, respecting the range defined by first and last observation. This data were analysed with a two-step ANOVA in all genotypes that had at least three ramets. The first model considered Yij = μ + Bi + Gj + εij, where μ is the general mean, Bi is the effect of block i (fixed), and Gj is the effect of genotype j. Gj and εij are assumed to be normally distributed random variables with zero means and variance components, VG and VR, respectively. From the results, the individual estimates of each block were retrieved and used to correct Yij as follows: inline image. εij from the first ANOVA, plotted onto maps of the field sites, showed no additional unaccounted microspatial influences. inline image was taken to the second ANOVA. The second ANOVA for the P. nigra collection considered Yjkl = μ + Pk + Gj(k) + εjkl, where μ is the general mean, and Pk and Gj the effects of population k and genotype j nested under population k (both considered as random). The second ANOVA for the pedigrees considered Yjk = μ + Gj + εjk, where μ is the general mean, and Gj the effect of genotype j, considered as random.

Figure 1.

 Scoring scheme for bud set in poplar. Seven discrete stages were delineated to cover onset, dynamics and duration of bud set in different poplar species. Phenotypic aspects of the apical part, such as appearance of last emerged leaves, internode elongation, appearance of bud scales, color of the bud and presence of balsam, allowed the various stages to be distinguished. 1Balsam is indicative, primarily in P. nigra; 2Last leaf is not always reliable; sometime an atypical last small leaf sticks out from the bud (arrows) and dies after bud set. Not to be considered for scoring. L1, last emerged leaf; L2, one but last leaf; delt., deltoides; tricho., trichocarpa.

Principal component analysis  Multivariate analyses using principal component analysis (PCA) were performed separately for the P. nigra collection and each mapping pedigrees on individual tree basis. For POP3a and POP3b, because a high proportion of trees had already reached stages 2 or 1.5 at the beginning of the experiment (Fig. 2a), multivariate analysis could be performed either on a limited progeny sample (POP3a) or on a limited set of traits (POP3b). All initial variables were standardized and orthogonal factors (PC1 and PC2 axis) were successively constructed as linear combinations of initial variables to maximize percentage of phenotypic variance explained.

Figure 2.

 Progression of bud development and climatic conditions in three field sites. The progression of bud development, expressed as mean stage of a population/pedigree at a given day of the year (DOY), and relevant climatic variables are given for the 90-d period between day 230 and day 320, including autumn equinox at day 267. (a) Bud development in four poplar F1 pedigrees. POP2, POP3a and POP3b were evaluated in France and POP5 was evaluated in Italy. The lines represent the period and frequency of measurement that are also detailed in Table 2. (b) Bud development in 14 populations and regional collections of Populus nigra, evaluated in Belgium. All populations were evaluated twice a week from day 234 to day 318 (24 times). Populations are color-coded per country; more detailed results are available in the Supporting Information, Table S1. (c) Daylength progression for the three field sites situated at 44° N, 47° N, and 50° N. (d) Cumulative night length (cnl), calculated from July 1 for the three field sites. (e) Cumulative sums of daily mean temperature (cmt) and daily minimum temperature when below 10°C (cmt10), both calculated from July 1, for the three field sites. Dotted lines correspond to cmt, full lines to cmt10. See the Materials and Methods section for the calculation of cmt10. (f) Schematic situating bud-set traits relative to time. The progression through bud-set stages is shown for two hypothetical genotypes. Cessation of growth and bud formation together lead to bud set. From the description of stages, cessation of growth, the duration from cessation of growth to visible budscales (subprocess 1, depicted by first arrow-head line), the duration from visible bud scales to closed apical bud (subprocess 2, depicted by second arrow-head line) and duration until bud set (depicted by dashed line) can be delineated.

Heritability, genetic variation, and population differentiation  For the P. nigra collection, heritability, genetic variation and population differentiation were calculated from restricted maximum likelihood (REML) estimates of variance components using the block-adjusted phenotypic values (inline image), considering genotype (nested under population) and population as random factors. Pk, Gj(k), and εjkl were assumed to be normally distributed random variables with zero means and variance components inline image, inline image and inline image, respectively. Using the estimates of variance components, individual broad-sense heritability (inline image) and genotypic heritability (inline image) were calculated for each trait using inline image/(VG + VR) and inline image/(VG + VR/ni), respectively, where VG is the genetic variance (inline image), VR (inline image) the residual variance, and ni the number of ramets/genotype for this trait. QST was calculated using inline image/(inline image), considering the true populations and regional collections. For the pedigrees, genetic and residual variance components were calculated by equating observed mean squares to expected mean squares and solving the resulting equations according to the Henderson III procedure (Henderson, 1953; Searle et al., 1992). For pedigrees, the coefficients of genetic and residual variation (CVG%; CVR%) were the genetic and residual standard deviation of a trait divided by the mean and expressed as a percentage.

QTL detection

Quantitative trait loci were determined for selected traits in four poplar breeding pedigrees with previously established genetic maps (framework markers with logarithm of odds (LOD) > 2) (Jorge et al., 2005; Gaudet et al., 2008; Dillen et al., 2009) with MultiQTL 2.4 (http://www.multiqtl.com/; MultiQTL Ltd, Institute of Evolution, Haifa University, Haifa, Israel). Single traits were analysed with multiple interval mapping. The entire genome was first scanned with the one-quantitative trait locus and the two-QTL models. Permutation tests (1000 runs), comparing hypotheses H1 (one quantitative trait locus in the chromosome) and H0 (no QTL in the chromosome), were run to obtain chromosome-wise statistical significance. In a second step, the genome was scanned for QTL assuming a two-QTL model. For chromosomes for which a single quantitative trait locus was already detected, permutation tests (1000 runs) were run to compare the hypotheses H2 (two linked QTL in the chromosome) vs H1. Subsequently, when P(H2vsH1) < 0.05, permutations were run to compare H2 vs H0. A two-linked QTL model was accepted only, when the two intervals were not adjacent and the one-quantitative trait locus model was significant. In a last step, multiple-interval mapping was carried out on all significant QTL (P < 0.05). Permutations were run per chromosome, using P < 0.05 as threshold per chromosome. Bootstrap analysis was done to estimate the 95% confidence interval. The option ‘marker restoration’ was used to reduce the effect of missing information. The Kosambi mapping function was chosen for recalculation of maps on genotypic data. The 2-LOD support intervals of QTL peaks were calculated from the exported LOD curve.

Integration of genetic maps with the poplar genome

Simple sequence repeat (SSR) markers that had been mapped in the various pedigrees were placed on the poplar genome with blast algorithms (http://genome.jgi-psf.org/Poptr1_1/Poptr1_1.home.html; genome assembly version 1.1). In addition, amplified fragment length polymorphism (AFLP) markers from the genetic maps of POP3a and POP3b were sequenced and compared with the poplar genome with blast. From these, 155 new AFLP markers, for which genetic and suggested genome positions were in agreement, were added to the analysis as genome-anchored markers. All genetic and QTL maps and genome sequence are visualized with the web-based cmap tool (http://www.gmod.org) that can be publicly accessed at http://services.appliedgenomics.org/poplar_budset.


Definition of traits to dissect bud set

To dissect bud set, we developed a new detailed scoring system with seven stages that cover the different aspects of the progression from a growing apex to a closed bud (Fig. 1). Stage 3 corresponds to full, active growth. Stage 2.5 marks growth cessation and, thus, is the closest approximation for sensing the critical night length. Stage 1.5 pinpoints the important transition from shoot to bud (Fig. 1). Stage 0.5 (green, closed bud) or 0 (red–brown mature bud) correspond to bud set as used in most traditional analyses that have assessed accomplished bud set (Fig. 1). The identification of landmark stages throughout bud development offered the possibility to dissect the phenotype. Hence, the timing of growth cessation and the duration of bud formation could be considered as separate determinants of bud set.

Bud development was assessed with the new scoring system (Fig. 1) in poplar germplasm of natural and hybrid origin. A total of 437 accessions of riparian P. nigra populations and regional collections represented a wide range of previously characterized, autochthonous provenances across Europe (Table 1; Storme et al., 2004; Smulders et al., 2008). Together with additional hybrid material from four different poplar breeding pedigrees (Table 2), a total of 6847 individual trees were assessed and their bud development was expressed as stage over chronological time (Fig. 2a,b). At the same day of the year (DOY), trees in the three different field sites located at 44° N, 47° N, and 50° N experience different night lengths and temperatures. The actual night lengths differ by c. 15 min, a month before and after the autumn equinox per 3° change in latitude (Fig. 2c). To allow for the direct comparison of data obtained in different field sites, data were subsequently expressed to light- and temperature-based scales. These include the cumulative night length (cnl), and two different temperature sums: cumulative daily mean temperature (cmt) and cumulative daily minimum temperature when below 10°C (cmt10) (see the Materials and Methods; Fig. 2d,e). In all sites, the cnl metric roughly parallels the DOY scale and brings photoperiod evolution for the three field sites closer to each other (Fig. 2d). This metric is not biologically relevant because photoperiod perception is governed by a coincidence of internal circadian clock rhythms and external light. It is used to provide a decimal scale for easier statistic data treatment. Conversely, cumulative temperature is used for its biological relevance for phenological events and growth rates. The differences in mean (cmt) and minimum temperature (cmt10) sums are characteristic for the cooler northern and warmer southern locations (Fig. 2e).

Traits were derived from the stages to describe the onset and dynamics of the process (Fig. 2f). Traits include date-of-onset of a defined stage (as described by the scoring scheme; Fig. 1) and the duration of two biologically coherent processes. Subprocess 1 (subproc1), corresponding to the duration from onset of stage 2.5 to onset of stage 1.5, essentially describes the changes in the growth-ceasing apical shoot. Subprocess 2 (subproc2), covering the time from stage 1.5 to stage 0.5, refers to changes of the bud after the first visible signs of bud initiation. Depending on the scale relative to which traits are expressed, date-of-onset traits are extended by _cnl (cumulative night length), _cmt10 (cumulative daily minimum temperature when below10°C) and duration traits by _Δcnl and _Δcmt10.

Phenotypic variation in bud-set traits

Typically, cessation of growth is initiated above a critical night length in poplar (Vegis, 1964; Howe et al., 1996). Accessions in the P. nigra collection, if ordered according to their latitude of origin, showed a roughly clinal response for growth cessation (date2.5_cnl; Fig. 3a). Based on the average date2.5_cnl for populations or regional collections, the critical night lengths were inferred to vary between 13:32 h (Waal/Maas, earliest) and 12:26 h (Ebro2, latest) within the collection, when assessed at 50°N (Figs 2b, 3a; Table 1).

Figure 3.

 Phenotypic variation in the cessation of growth and duration of bud formation. The date of cessation of growth or bud initiation, respectively, and duration of bud formation are given for 14 populations/regional collections of Populus nigra and the four pedigrees, based on block-adjusted genotypic means. Box plots represent the range of genotypic means between the 25% and 75% percentile by the box, the median by the middle horizontal line, and minimum and maximum observations by the whiskers. In each panel, populations or regional collections are numbered and ordered from southern to northern origin (see Table 1 for full information). (a) Population means for cessation of growth (date2.5_cnl). Additional informative scales on day of the year and night-length are given, but are not metric. (b) Population means for the duration of bud formation (subproc1_Δcmt10 and subproc 2_Δcmt10). (c) Pedigree means for bud initiation (date1.5_cnl). (d) Pedigree means for the duration of bud formation (subproc 2_Δcnl). The grand mean of the P. nigra collection is included for reference in (c) and (d).

In the poplar pedigrees, cessation of growth occurred first in the interspecific F1 families POP3a, then in POP3b, POP2 and finally in the intraspecific P. nigra family POP5 (Table 2). In the three pedigrees grown together in France, the female P. deltoides parent (and in POP3b also the male parent) originated south of the site of the progeny selection, thus carrying a critical-night-length character that delayed bud set relative to trees originating at the site. The order of pedigrees with respect to growth cessation reflected the extent of latitudinal shift, with the latest-setting POP2 shifted farthest north (10°-Δlatitude; Table 2).

In the P. nigra collection, the differences in the duration of bud formation were minor. Part of the variation might have gone undetected because of an exceptionally warm autumn in 2005. As little as 1.5 d constitute the difference for the duration of either subprocess 1 or subprocess 2 at the level of population means (data not shown). Expressing the duration to minimum temperature (cmt10) revealed, however, that early-setting populations (Dutch accessions) experienced less cold during bud formation than late-setting populations (Ebro1 and Ebro2) within a comparable period of time (Fig. 3b). Thus, the actual temperature sum for bud formation differed to a great extent. The effect of temperature was more pronounced for subprocess 1, when leaf primordia are formed and differentiated, than for subprocess 2, comprising the maturation of the bud (Fig. 3b). In the pedigrees, the phenotypic variation in the duration of bud formation is comparable to the variation in growth cessation and bud initiation (Table 2; Fig. 3c,d). Thus, duration of bud formation is yet another important factor determining the completion of bud set.

Genetic variation in bud-set traits

All traits showed relatively high broad-sense heritabilities, both at the individual as well as at the genotypic level (Fig. 4). In the pedigrees, genotypic heritabilities for onset-of-stage traits (0.5–0.9) were higher than those of the two-duration traits (0.25–0.5; Fig. 4). This difference in heritability will provide a higher precision for QTL detection for onset-of-stage traits. In the P. nigra collection representing a large range of geographical origins, genotypic heritabilities were even higher: > 0.9 for the onset-of-stage traits and c. 0.7 for the two-duration traits (Fig. 4). These values are comparable with, or in case of the P. nigra collection surpass the broad-sense heritabilities of 0.51 at individual level and 0.81 at genotype level found earlier for bud set in an F2 poplar pedigree (Howe et al., 2000).

Figure 4.

 Genetic variation of bud-set traits. Onset-of-stage and duration traits were defined from original data (Fig. 2f). The bars represent phenotypic variance components expressed as % of the trait mean: total phenotypic variance (open bars) and total genetic variance (tinted bars). For Populus nigra, the genetic variance is further partitioned to population (dark tinted bars) and individual within population (lighter tinted bars) components. Broad-sense heritability on individual (inline image) and genotype (inline image) basis and population differentiation (QST) are given. Genetic variance and heritability were not calculated (NA, not available), when genetic variance estimation was biased owing a to limited number of genotypes with data.

Bud set typically is a highly differentiated trait in forest trees. In the P. nigra collection, the effects of genotype and population were highly significant (P < 0.0001) for all traits, as estimated with ANOVA. For the various onset-of-stage traits, between 76% and 86% of the phenotypic variance was partitioned to genetic variance (VG) in the P. nigra collection (Fig. 4). The VG of duration traits reached only 27–36% of the phenotypic variance, underlining the higher influence of environmental conditions on duration traits (Fig. 4). Irrespective of the trait, most VG, on average 75%, was partitioned to the factor population (Fig. 4). Consequently, population differentiation (QST of c. 0.6) was quite high; only subprocess1_cnl showed less differentiation (Fig. 4). In a European aspen (Populus tremula) collection covering a steeper cline from 56° N to 66° N, QST for bud set was within the same range (c. 0.75; Hall et al., 2007).

Among all traits of the P. nigra collection, subprocess2_Δcnl, describing the maturation of the bud, appeared strongly influenced by environmental conditions (Fig. 4). Only the late-setting populations from the Ebro displayed significant genetic variation (Fig. S1). For cessation of growth (date2.5_cnl), the two populations from the Drôme displayed most phenotypic variation, but the two populations from the Ticino together with one population from the Ebro had a relatively higher genetic variation (Fig. S1). These examples of within-population genetic variation illustrate that natural selection can act at relatively small geographical scales.

In the four pedigrees, the contribution of genetic variance to phenotypic variance was generally lower than that of residual variance. However, the onset-of-stage traits showed a relatively higher genetic component in the total variance (Fig. 4). Some pedigrees presented specific genetic variation, for example POP3a for the last stages of bud maturation and bud set (date0.5_cnl and date0_cnl; Fig. 4). Most variation was found for the date of growth cessation in the P. nigra collection, whereas the four poplar pedigrees presented similar total amounts of genetic variance for onset-of-stage and duration traits.

Definition of the most descriptive traits based on phenotypic variation

The onset-of-stage and duration traits together describe a developmental sequence and therefore tend to be correlated. In particular, the onset-of-stage traits correlated highly with each other (Fig. 5a,b). Duration and onset-of-stage traits were, on average, less (P. nigra) or even negatively (POP3a, POP5; Table S2) correlated at the trait level. Hence, in the pedigrees, the later cessation of growth occurred, the shorter the duration of bud formation was. While all stages are important for a profound physiological dissection of bud development, many highly correlated traits will provide redundant information when genetic trait architecture is investigated. Therefore, we used an unsupervised method to define those traits that are most discriminative in terms of phenotypic variation.

Figure 5.

 Correlations between phenotypic traits of bud set and effect on trait-pair quantitative trait loci (QTL) colocalization. (a) Trait correlations in the Populus nigra collection, based on individual values. All correlations are significant at P < 0.0001. (b) Average trait correlation across the four pedigrees above the diagonal, and the total number of colocalization events of traits below the diagonal. Trait correlations above 0.8 and total number of colocalization above 7 are in bold; negative trait correlations are in red. Because of the limited data, QTL for subproc1_Δcnl were excluded from the colocalization analysis. Trait correlations are detailed for each pedigree in Table S2. (c) Colocalization events in function of trait correlation within the individual pedigrees.

A principal component analysis using the six onset-of-stage and two duration traits allowed identifying those traits that differentiated most in the populations (Fig. 6). The phenotypic variation in P. nigra, the most comprehensive dataset, was partitioned into a major contribution from onset-of-stage traits (principal component 1, PC 1; 81.2%) and a minor contribution from the duration traits (PC 2, 18.7%) (Fig. 6). Similar to P. nigra, the phenotypic variation in the pedigrees is 50–75% partitioned to PC 1 and 20–35% partitioned to PC 2 (Fig. 6). In POP2 and POP3b, onset-of-stage traits (primarily date1.5_cnl and date0.5_cnl), determine the phenotypic spread along the axis of PC1 (Fig. 6). In POP5, date1.5_cnl has a major effect and subproc2_Δcnl a minor opposing effect on the phenotypic spread along the axis of PC1. subproc2_Δcnl contributes most to PC2 in POP2, POP3a and POP3b, while subproc1_Δcnl generates a distinct pattern along PC2 in POP5. Thus, without a priori assumptions, the onset of bud initiation (date1.5_cnl) and the duration from bud initiation to completed bud set (subproc2_Δcnl) were identified as major contributors to phenotypic variation. Based on this analysis, the following traits were selected for further analysis of genetic aspects. date0.5_cnl and subproc2_Δcnl cover most of the phenotypic variation. In addition, date1.5_cnl, subproc1_Δcnl were included for their specific importance in POP3a and POP5, respectively. However, we need to bear in mind that both date1.5_cnl and date0.5_cnl have a clearly higher genetic variation within the total variation (Fig. 4).

Figure 6.

 Principal component analysis (PCA) of bud-set traits. The PCA was applied to delineate bud-set traits that covered most phenotypic variation. The individual block-adjusted phenotypic values of six onset-of-stage and two duration traits were included, except for POP3b (date2.5, date2, and subproc1 excluded). For each pedigree and the Populus nigra collection, the portion of variance association to the principal components (PC) is given. Contributions of all original traits to PC1 and PC2 are listed for P. nigra. For the pedigrees, only traits with high contribution are shown.

QTL for bud set in four poplar breeding pedigrees

Across the four pedigrees, 105 QTL were detected for all six onset-of-stage and two duration traits. All QTL were projected onto the respective genetic maps (Table S3) and can be publically viewed at http://services.appliedgenomics.org/poplar_budset. Not unexpectedly, QTL for various traits tended to stack because of the correlated nature of phenotypic data. The higher the traits were correlated, the more the corresponding QTL colocated (Fig. 5c). Most colocation events, relative to the total number of QTL, were detected in POP2 and POP5 (Fig. 5c). Approximately 20% of the colocation events involved a colocation of duration and onset-of-stage traits. Most often, colocation of such negatively correlated traits was noticed in POP5 (Fig. 5c). If based on the same gene(s), the respective locus might be under evolutionary constraint owing to the opposite effects on the onset and duration of bud set.

For the purpose of understanding genetic trait architecture, only 53 QTL of the four most descriptive traits chosen from PCA are discussed further (Table 3). Relatively more QTL were found in the two pedigrees involving inter-American crosses (POP2 and POP3b) than in the Euramerican (POP3a) and the intraspecific crosses (POP5) (Table 3). The strong segregation for bud-set traits might be based on the large difference in latitudinal origin of the parents for POP2 and POP3b (Table 2). A single QTL explains on average 8.6% of the phenotypic variation. The average sum of explained phenotypic variation per trait in a pedigree was 38% (Table 3). Consistent with the phenotypic and genetic results, QTL are revealed for both the onset-of-stage and the duration traits (Table 3). In most cases, a trait was influenced by a small number of loci. Given the progeny size, it is likely many QTL of smaller effect that contribute to phenotypic variation were not even detected. Together, these results underscore the quantitative nature of bud set.

Table 3.   Quantitative trait loci (QTL) for the four most descriptive bud-set traits
TraitNumber of QTLTotal PVE
  1. Where no number appears, too few and/or skewed data were available. In POP5, phenotypic data of date0.5_cnl were subjected to QTL analysis but did not detect any QTL. A complete list of QTL and their characteristics is available in Table S3.

date0.5_cnl527No QTL140.3730.1680.474

Six robust QTL regions detected across various pedigrees

Tentatively similar QTL regions were observed in a number of chromosomes across two or more parental maps (Table S3). The investigation of the corresponding genetic and physical sequence regions appeared to be hampered, however, by problems in species with less-developed genetic resources, such as low marker density, low marker sharing among genetic maps and the lack of marker sequence information to establish a link to the genome sequence. In consequence, a conservative approach using sequence-anchored flanking markers was adopted to determine whether QTL occurring in two or more parental maps identified the same region.

Out of the 53 QTL for the four most descriptive traits, 36 are linked to the physical genome sequence with at least two sequence-anchored microsatellite or AFLP markers flanking the LOD peak of the QTL (http://services.appliedgenomics.org/poplar_budset). Of these, 21 QTL from different parents identified six recurrent QTL regions in the poplar genome that are important for bud set (Fig. 7). These six robust QTL regions ranged from 1.83 to 4.25 Mbp in size and contained from 202 to 394 annotated genes (Figs 7, 8; Table S4).

Figure 7.

 Robust quantitative trait loci (QTL) regions for bud set. The QTL regions were linked to the physical genome sequence by sequence-anchored markers flanking the LOD peak. Twenty-one QTL fall into recurrently detected regions across pedigrees and are depicted here by their physical regions above the respective genome linkage group (in Mbp). Only the flanking markers considered are shown; additional markers can be viewed through the cmap application at http://services.appliedgenomics.org/poplar_budset. Genomic regions (highlighted in gray) were delineated to the smallest overlapping regions. The features of the 21 QTL (LOD, PVE, LOD peak, genetic and physical size) are given. One quantitative trait locus on chromosome 6, detected in Populus trichocarpa of POP3b, turned out to have no overlap with the quantitative trait locus detected in Populus deltoides of POP3b (region therefore not highlighted). A size reference in Mbp is given in the lower right corner. M1, marker start; M2, marker stop; LOD, logarithm of the odds; PVE, phenotypic variance explained; LG, genome linkage groups; P. delt., respective P. deltoides parent; P. trich., respective Populus trichocarpa parent; P. nigra P, Populus nigra‘Poli’; P. nigra 58, P. nigra‘58-861’.

Figure 8.

 Six new genomic regions, previously identified quantitative trait loci (QTL) and candidate genes for bud set across the poplar genome. The 19 genome linkage groups are drawn to scale, with their total size in Mbp given below. Expressional candidate genes, as identified by Ruttink et al. (2007), are drawn at their respective positions with green dots. The total number of expressional candidate genes per linkage group is given within parentheses below each linkage group. The six robustly detected QTL regions from Fig. 7 are drawn to scale and position with green rectangles. The identities of the genes contained in each of the six QTL regions are given in Table S4. Functional candidate genes are drawn at their respective genome position, following previous studies (Frewen et al., 2000; Rohde et al., 2002; Böhlenius et al., 2006; Ingvarsson et al., 2008; Ruonala et al., 2008; Jiménez et al., 2009 and Li et al., 2009; Ibáñez et al., 2010). Three QTL identified in family 822 (Frewen et al., 2000) are drawn with open rectangles (not to scale) at their tentative location, by giving the genetic distance in cM from the respective anchoring marker with an arrow (on LG3, the simple sequence repeat (SSR) name of the anchoring marker is suppressed). TFL1/CENL1, TOC1 and CO are located on unassembled genome scaffolds.

QTL studies cannot reveal the specific genes underlying the traits. Still, for the sake of developing a strategy toward the cloning of genes carrying the causative polymorphisms, it is interesting to integrate the data with various genes known to be associated with bud set either through function or expression (Frewen et al., 2000; Böhlenius et al., 2006; Ruttink et al., 2007; Ingvarsson et al., 2008; Ruonala et al., 2008; Ibáñez et al., 2010). Of the 945 genes with differential expression during bud development in poplar (Ruttink et al., 2007), 757 genes were located on assembled genome linkage groups (Fig. 8). These expressional candidate genes were evenly distributed across the genome. Fifty-two expressional candidate genes (6.9%) resided within the six highlighted QTL regions for bud set. The proportion of these expressional candidate genes amounted, on average, to 2.8% of all genes contained in the respective regions (Fig. 8).

Next, we asked whether functional candidate genes were colocalizing with the six genomic regions. CO (gw., involved in the transduction of daylength signals for growth cessation (Böhlenius et al., 2006), was an obvious candidate but resided on an unassembled scaffold of the poplar genome. CO was mapped to chromosome 17 of P. nigra (POP3a) through two single nucleotide polymorphisms (SNPs). Because a very strong QTL was located close to this position in P. deltoides (POP3a; Table S3), additional single-marker t-tests (Kruskal–Wallis) between phenotypic data and markers adjacent to the CO-SNPs (and similarly contained in the P. deltoides map) were carried out. No significant correlations of marker polymorphisms with any bud-set trait were revealed, rejecting the possibility that a segregation of CO alleles would have caused the QTL effect in the POP3a pedigree.

The FT gene, similarly involved in the transduction of daylength signals (Böhlenius et al., 2006), colocated with a robust QTL region on chromosome 8, thereby constituting a good candidate to be tested for segregation in the pedigrees (Fig. 8). Also, GIGANTEA (GI), adjusting the expression of the photoperiod-regulated CO gene to the circadian clock, colocates to a robust region on chromosome 5 (Fig. 8). PHYTOCHROME B2, of which two SNPs explained part of the phenotypic variation in timing of bud set in European aspen (Ingvarsson et al., 2008), did not coincide with any of the robust QTL regions (Fig. 8). None of the poplar homologs of dormancy-associated MADS-box genes, that were functionally characterized in peach (Jiménez et al., 2009; Li et al., 2009), colocalized with a robust QTL region (Fig. 8).

The QTL for bud set had been previously identified on linkage groups 3 (F), 6 (PY), and 10 (J) (the linkage groups within brackets correspond to Frewen et al., 2000; Fig. 8). The QTL on chromosome 6 most probably corresponds to a robust region identified again in two different parental maps (Fig. 8).

In conclusion, the pedigrees displayed segregating genetic variation at various loci, of which six recurrently detected QTL regions define new, highly interesting and reasonably small areas (Fig. 8). These regions contain yet to be identified new causative genes and/or might in part be based on segregation in known functional genes, such as FT or GI, colocating with them (Fig. 8).


Dissecting the bud set phenotype

For an understanding of bud set and its rational exploration in forestry, it is of primary interest to disentangle the conundrum of various overlapping processes that co-occur in time and/or space. The in-depth analysis of bud set, as applied in the current study, offers the possibility of dissecting the phenotype into stages and to estimate their relative contribution to the accomplishment of bud set (Fig. 1). Most importantly, the time of cessation of growth and the duration of bud development can be considered as separate and crucial determinants of bud set, which is a major improvement compared with the commonly used date of completed bud set.

Bud set proved to be determined by the timing of the cessation of growth and the duration of bud formation. Phenotypic variation was split, using PCA, into the onset of cessation of growth and duration of bud formation in the natural and hybrid populations (Fig. 6). However, the timing of the cessation of growth had a relatively higher contribution of genetic variance to total phenotypic variance than the duration of bud formation (Fig. 4).

Night length is the most stable predictor of coming winter hazards in many environments. Accordingly, the latitude of origin is often strongly correlated with the timing of bud set (Pauley & Perry, 1954). Here, cessation of growth triggered by an increasing night-length was the most decisive factor for bud set (Figs 3, 4, 6). The timing of the cessation of growth differed by approx. 27 d, equaling a difference of 2 h night-length, in all P. nigra accessions covering origins over 11° latitude (Fig. 3a; Table 1). Similarly, in the four poplar pedigrees, the timing of the cessation of growth contributed most to the timing of bud set (Fig. 6).

The duration of bud formation varied by only a few days in the P. nigra collection. Northern ecotypes set bud slightly quicker than southern (Table 1; Fig. 3b), as was also observed in birch (Betula alba) under controlled growth conditions (Junttila et al., 2003). The duration of bud formation, because of its lower dependence on genetic factors (Fig. 4), probably allows for the accommodation of environmental influences and might be particularly important in the adaptation to short-term fluctuations or year-to-year variation in temperature, or even in the adaptation to different environments. Together, the timing of the cessation of growth as well as the duration of bud formation determine, albeit not to the same extent genetically controlled, the timing of bud set.

The unknown contribution of phenotypic plasticity to variation in bud set

For a rational selection of ecotypes in breeding programs as well as for the deployment of seeds or other propagules in forest plantation programs, phenology is of primary importance. Phenology largely determines to which degree trees will synchronize growth to the local climate. Quite rapid changes in climate might uncouple populations from the local environment to which they had strongly adapted. Adequate genetic adaptation through migration and evolution in situ are considered rather unlikely scenarios for forest trees (Aitken et al., 2008). Another short-term option for adaptation, namely phenotypic plasticity, remained largely uncharacterized for phenology traits.

Genetic determinants of phenotypic plasticity, the degree to which the expression of characters is changed by different environments, are not well understood (Via & Lande, 1985). Variation of phenotypic response at population level ideally draws from large within-population variation and results in low clinal and/or ecotypic differentiation. Phenology traits, however, usually exhibit less within-population phenotypic variations than many growth-related traits (Howe et al., 2003; Aitken et al., 2008). The within-population genetic variation, an indicator for the potential adaptive response to natural selection, remained below 25% of the total genetic variation in the P. nigra populations (Fig. 4). A small, but significant within-population genetic variation for critical night length at bud set was also noted in a collection of European aspen (Ingvarsson et al., 2006). Because variation in phenotypic plasticity can contribute to short-term adaptation, replicated clonal experiments in several field sites need to be carried out to study the genetic variation for phenotypic plasticity.

The genetic architecture of bud set with small-effect genes implies a strong genetic redundancy (Table 3). The fairly steep phenotypic clines of adaptive traits, such as bud set, may not correlate with allelic clines in single genes. Many genes of small effect determine and perpetuate the cline. Selection acts to reinforce phenotypic (and QTL) effects through generating covariance between individual alleles of several loci (Le Corre & Kremer, 2003). Many loci (genes or QTL) stay as undifferentiated as neutral markers; a few loci, at best, might show signs of differentiation, which do not need to be the same across populations or pedigrees. This genetic architecture argues for sufficient loci that alone and through epistatic interactions can contribute to phenotypic plasticity, despite the limited within-population variation.

Other, yet little investigated, mechanisms might also enable a quick adaptation to the local climate. In Norway spruce, the prevailing temperature during embryo development determines in each seed, probably epigenetically, the timing of bud set of the future plant (Kvaalen & Johnsen, 2008). Such a mechanism, if widespread among forest trees, would not only enable a quick (within one generation) adaptation to changes in climate variables, but also generate variability within a population, depending on warm or cold temperatures during seed set from year to year (Rohde & Junttila, 2008).

In addition to phenotypic plasticity, bud set is clearly influenced by a number of phenotype-confounding factors, such as growth rate, branching system and drought stress or pathogen infection at the end of the growing season. Foliar rust infection by Melampsora larici-populina can significantly advance or delay bud set (C. Bastien, unpublished). These current biotic or other abiotic (drought) constraints on phenology might become more important cofactors for bud set when trees need to adapt to changing climatic variables.

Toward the causative genes for genetic variation in bud set

Quantitative trait loci mapping remains challenging in forest trees, primarily because of the limited manageable progeny sizes that impede the resolution of genetic maps. However, additional robustness can still be achieved through searching QTL in different genetic backgrounds, implying a multiplicity of epistatic effects. Here, four different poplar pedigrees were investigated for bud-set QTL (Table 3) and a number of QTL regions coincided across different parents (Fig. 7). The number of coinciding genomic regions is most probably underestimated: only 36 of 53 considered QTL could be investigated for physical overlap because of a lack of genome-anchored markers across the parental maps. Nevertheless, the six genomic regions with coinciding QTL from several parents are considered as more robust than those found in one parent only (Fig. 8). Still, it remains to be seen whether the same causative genes underlie the QTL in the different parents.

In addition to the ability to detect meaningful quantitative variation, the delineation of narrow QTL regions for the identification of the genes carrying the causative polymorphisms remains extremely challenging in forest trees. Through the comparative approach, we have established reasonably sized regions in which to start looking for the causative genes (Fig. 8). The best-supported genomic region, detected in three different parents on chromosome 8, still contains 394 genes (region 8b; Figs 7, 8). Genes within this or the other region(s) can be scanned for allelic variation to start pinpointing the gene(s) that underlie(s) the QTL and to identify the relevant alleles for adaptation. To estimate their putative significance for adaptation under the climate change scenarios, genes with demonstrated function at bud set will need to be characterized not only for their functional polymorphisms, but also for plasticity of their expression in ecologically relevant, natural environments.

Together, the data from bud-set-related gene expression studies (Ruttink et al., 2007) and the QTL presented here indicate that probably many genes are involved in the regulation of bud set and, thereby, in the adaptation to local climate in general. The need to consider many different genes underscores the difficulties in predicting complex interactions of trees with their environment. However, the diversity in the genes with a functional role during bud set will determine how rapidly populations will respond to selective environmental pressures. Combining genomics, functional and developmental approaches with ecological mechanisms will advance our understanding of phenotypic adaptation in the context of global warming.


The authors thank Stephane Rombauts for kindly providing the genome locations of SSR and AFLP markers, Kurt Schamp, Bart Ivens, Patrick Poursat, Jean Gauvin, Luca Riciotti, Michele Baldasso, Aurélien Marron and Els Van Goethem for assistance in data collection, and Martine De Cock for help in preparing the manuscript. This work was supported by the European Union (QLK5-CT-2002-00953 POPYOMICS), the Research Foundation-Flanders (predoctoral and postdoctoral fellowships to SD and AR, respectively), and the Institute for the Promotion and Innovation by Science and Technology in Flanders (postdoctoral fellowship to TR).