Transcriptome analysis of bud burst in sessile oak (Quercus petraea)

Authors


Author for correspondence: Antoine Kremer Tel +33 557 122832 Fax: +33 557 122881 Email: kremer@pierroton.inra.fr

Summary

  • • Expression patterns of hundreds of transcripts in apical buds were monitored during bud flushing in sessile oak (Quercus petraea), in order to identify genes differentially expressed between the quiescent and active stage of bud development.
  • • Different transcriptomic techniques combining the construction of suppression subtractive hybridization (SSH) libraries and the monitoring of gene expression using macroarray and real-time reverse transcriptase polymerase chain reaction (RT-PCR) were performed to dissect bud burst, with a special emphasis on the onset of the process.
  • • We generated 801 expressed sequence tags (ESTs) derived from six developmental stages of bud burst. Macroarray experiment revealed a total of 233 unique transcripts exhibiting differential expression during the process, and a putative function was assigned to 65% of them. Cell rescue/defense-, metabolism-, protein synthesis-, cell cycle- and transcription-related transcripts were among the most regulated genes. Macroarray and real-time RT-PCR showed that several genes exhibited contrasted expressions between quiescent and swelling buds, such as a putative homologue of the transcription factor DAG2 (Dof Affecting Germination 2), previously reported to be involved in the control of seed germination in Arabidopsis thaliana.
  • • These differentially expressed genes constitute relevant candidates for signaling pathway of bud burst in trees.

Introduction

Vegetative bud phenology of long-lived species, such as forest trees, is crucial to consider as it may profoundly affect their productivity, adaptability and distribution (Chuine & Beaubien, 2001). Indeed, the length of the growing season, which largely determines the annual production, is defined by the time between spring bud burst and autumn bud set. Forests in temperate zones are well adapted to the seasonal cycle, having a dormancy period in response to winter conditions (low temperature and short photoperiod). Changes in timing of phenological events may cause frost or drought injuries or even a failure to produce mature fruits or seeds. As a result, any modification of the seasonal phases could alter productivity or affect species abundance and distribution (Chuine & Beaubien, 2001).

These concerns are particularly relevant in the context of global climate changes. As bud burst of temperate-zone forests is driven mainly by temperature (Chuine, 2000; Kramer et al., 2000), global warming is expected to modify the length of the growing season and distribution of forest tree species. Impacts of global climate change have been reported in plants and animals (Penuelas & Filella, 2001; Walther et al., 2002). At present time, the average annual growing season has already increased by 10.8 d since the 1960s (Menzel & Fabian, 1999) for trees and shrubs in Europe, according to records of the European International Phenological Garden. Consequences of these changes have been monitored or forecasted in forest trees (Cannell & Smith, 1986; Murray et al., 1989; Beuker, 1994; Kramer, 1995; Visser & Holleman, 2001; Howe et al., 2003). On one hand, for species with small chilling requirements, an increase in temperature results in an earlier bud burst, increasing the risk of frost damages. On the other hand, changes in the duration of growing period alter the competitive balance between species, through modifications of ecosystem interactions. Both for ecological and economical reasons, a better understanding of phenological switches, in particular bud burst, is needed.

Congruent geographic patterns of variation in bud burst were observed in forest trees, suggesting local adaptation of populations to the climatic conditions (Morgenstern, 1996). Ducousso et al. (1996) have studied the phenology of bud burst for 50 populations of sessile oak (Quercus petraea) planted in four provenance tests located in France. Both latitudinal and altitudinal trends were observed. On all sites, populations from northern latitudes flush later than populations from southern latitudes, and the ranking of populations remains remarkably stable across the four test sites. However, the interaction of environmental cues, such as temperature, photoperiodicity and winter chilling, with genetic components driving the timing of bud burst remains obscure. For a broad range of forest tree species, the timing of bud burst exhibits a continuous phenotypic variation typical of a quantitative trait. Overall, moderate to strong genetic control is associated with bud phenology related traits, as shown by the heritability values (h2) of bud burst in Populus (h2 = 0.48–0.80) (Howe et al., 2000), Salix (h2 = 0.43–0.72) (Tsarouhas et al., 2003) or Pseudotsuga menziesii (h2 = 0.44–0.95) (Li & Adams, 1993). Heritability in Quercus has been estimated to 0.87 by Jensen (1993) and from 0.15 to 0.51 by Scotti-Saintagne et al. (2004). These heritability values are evidence for a strong genetic control of timing of bud burst in forest trees. Despite these observations, there is still a lack of knowledge on the genes responsible for variation of this trait.

Quantitative trait loci (QTL) mapping experiments have been conducted in several tree species and showed that several genes are involved in the control of the timing of bud burst. In Populus species (Frewen et al., 2000), Castanea sativa (Casasoli et al. 2004), Pseudotsuga menziesii (Jermstad et al. 2003) and Salix (Tsarouhas et al., 2003) 6, 31, 11 and 6 QTL controlling bud flush have been detected, respectively. In Quercus, the genetic architecture of bud burst involved 8–15 QTL depending on the field test used (Scotti-Saintagne et al., 2004). With respect to the genes underlying these QTL, two candidate genes have been identified, whose map positions coincide with QTL for bud set and/or bud burst in Populus (Chen et al. 2000). However, the molecular mechanisms underlying bud set, bud dormancy and bud burst are still unclear. Only a few genes that may play a role in bud dormancy have been identified so far, and mutant analysis has provided support for the molecular basis of dormancy (Shimizu-Sato & Mori, 2001). Recently, chromatin remodeling has been suggested to be involved in bud burst by Horvath et al. (2003), following earlier observations made for vernalization responses during flowering (Bastow et al., 2004; Sung & Amasino, 2004).

The objective of the present study was to identify genes regulated during early bud burst in sessile oak, using a transcriptomic approach. We first constructed SSH (suppression subtractive hybridization) libraries (Diatchenko et al., 1996) at different stages of bud burst, then used high-density colony arrays (HDCA) and cDNA macroarrays to identify genes differentially expressed during bud burst, and validated our results with real-time reverse transcriptase polymerase chain reaction (RT-PCR).

Materials and Methods

Accession numbers

The sequences reported here have been deposited in the EMBL nucleotide database (Kulikova et al., 2004) (http://www.ebi.ac.uk/embl/index.html) under accession numbers CR627501 to CR628310.

Plant material and sample preparation

The plant material consisted of a collection of 1000 seedlings of sessile oak (Quercus petraea (Matt.) Liebl.) originating from two forest stands belonging to the same provenance region (north-east France). Seeds were sown in the spring of 2001 and seedlings were grown outdoors in the nursery of the INRA forest station (Cestas, France) under natural conditions. Terminal buds were collected along six different developmental stages (Fig. 1) during bud break on 1-yr-old seedlings. Harvests were made at seven different times from February 21 to April 24 2002. Each harvest time corresponded to a different developmental stage of the bud (Stage 0 to Stage V) excepted for stage 0 for which the first two samplings were bulked (Fig. 2), as buds were rather small for RNA extractions. These developmental stages are routinely used as standard scores for assessing bud burst in genetic experiments (Ducousso et al., 1996; Scotti-Saintagne et al., 2004). At each harvest time, four sets of 10 seedlings each were sampled (Fig. 2). These 40 seedlings had all reached the required stage of apical development. Exception to this was the first stage of development (Stage 0) for which buds were smaller and required three times more seedlings to be sampled. Apical buds were mixed within each set, and RNA extractions were carried out separately on each bulk of 10 (or 30) buds. Hence, four different extractions were made for each stage. Finally, two sets of RNA were bulked to account for extraction variations. As a result, each stage consisted in two statistical independent replicates of RNA extraction. As described in Fig. 2, one replicate was used for the construction of SSH libraries and for monitoring of gene expression (macroarray and real time RT-PCR (qPCR)). The second replicate was used for the monitoring of gene expression using real time RT-PCR only.

Figure 1.

Stages of bud development in Quercus petraea. Stage 0, bud is quiescent and protected by scales; Stage I, swelling bud; Stage II, opening of the bud has occurred; Stage III, leaves have grown; Stage IV, one leaf at least is completely out of the bud; Stage V, internodes have started growing.

Figure 2.

Flow chart describing the plant material, the RNA extractions and their use for analysis. HDCA, high-density colony arrays; qPCR, real time reverse transcriptase polymerase chain reaction; SSH, suppression subtractive hybridization.

At the first and second sampling dates, buds were considered to be in a quiescent stage, as confirmed by microscopic analysis (Fig. 3). Stage I corresponded to the first visible swelling of the bud, and was interpreted as the beginning of bud burst. At stage II, the buds started to grow (e.g. increase in length and diameter); leaves became clearly visible at stage III. Obviously, leaf and stem development occurred at stages IV and V, which corresponded to the transition between bud burst and the start of vegetative stem growth.

Figure 3.

Apical meristem at different bud burst stages. (a) Apical meristem at stage 0 (bar scale: 50 µm). No mitotic activity can be observed. (b) Apical meristem at stage 0 (bar scale: 10 µm). Typical meristematic cells (arrows): microvacuolization, nucleoles, high density cytoplasm, important nucleus volume. (c) Organization of apical meristem at stage II (bar scale: 50 µm). Leaves primordia are visible on each sides of the meristem. Mitotic activity is found as shown in (d) (arrows) (bar scale: 10 µm). Vacuoles are more visible than in stage 0.

For light microscopy (Fig. 3), samples were fixed for 4 h at 4°C in glutaraldehyde 2.5% (v : v) in phosphate buffer (pH = 7.2, 10 mm) and treated overnight at 0°C with osmium tetroxide 1% (w : v) in phosphate buffer (pH = 7.2, 10 mm). After three rinses in distilled water, buds were then fixed in uranyl acetate 2% (w : v) in water for 1 h in a darkroom at 4°C. Objects were then dehydrated progressively at room temperature with a crescent series of ethanol and embedded in Araldite (Epoxy resin). Sections (1 µm thick) were stained with toluidine blue and observed with a Leica DMLB microscope using AnalySIS software (Soft Imaging System, Münster, Germany).

Isolation of poly(A)+ RNA

Total RNA was extracted from buds, previously peeled off, following Chang et al. (1993) with modifications described by Dubos & Plomion (2003). Four independent extractions were done for each harvest and bulked among two sets of each stage (Fig. 2). RNA degradation and quality were evaluated on 1.2% agarose gel and spectrophotometrically between 180 nm and 340 nm. Before any other step, samples were treated with DNaseI (10 units per 30 µg total RNA) for 30 min at 37°C and then purified using RNeasy Plant Mini kit columns (Qiagen, Hilden, Germany). mRNA isolation and cDNA synthesis were carried out with the SMART PCR cDNA Synthesis kit (BD Biosciences Clontech, Palo Alto, CA, USA), according to the manufacturer's instructions.

Suppression subtractive hybridization

We performed SSH using the PCR-Select cDNA Subtraction kit (BD Biosciences Clontech). This method, originally described by Diatchenko et al. (1996), is based on selective amplification of differentially expressed sequences. The first library (forward), called Q (quiescent), was obtained with cDNAs isolated from buds at stage 0 for use as the tester and a pool of cDNAs from the five other stages was used as the driver. A second library (Early, E) was enriched for transcripts from stages I and II (tester), and a third library (Late, L) was enriched for transcripts from stages III, IV and V; stage 0 was used as the driver in both cases (reverse libraries). Subtracted PCR products were inserted into the pCR 4-TOPO vector and cloned into TOP10 Escherichia coli cells with the TOPO TA cloning kit (Invitrogen, Carlsbad, CA, USA). Each cloning product was spread on LB plates containing 50 µg ml−1 ampicillin and 80 µg ml−1 X-gal. About 600 randomly chosen positive clones per library were picked using the BioPick robot (BioRobotics, Woburn, MA, USA) for culture in 384-well plates.

Differential screening using colony array

The 600 selected clones, from each subtracted cDNA library, were arrayed onto two nylon membranes, using the MicroGrid II robot (BioRobotics). Each clone was spotted four times on each membrane. Bacterial colonies were grown onto filters overnight. Bacterial cells were then lysed and DNA was crosslinked to the filters with standard protocols. cDNAs from each pool (Q, E and L) were cleaned using chloroform–isoamyl alcohol (24 : 1) method. Labeled targets were synthesized using Prime-a-Gene Labeling System (Promega, Madison, WI, USA) with 30 µCi of [α-33P]dATP and 30 µCi of [α-33P]dCTP. Purification was carried out with QIAquick Nucleotide Removal Kit (Qiagen). Filters were preincubated for 5–6 h at 65°C under shaking in hybridization solution (5× standard saline citrate (SSC), 5× Denhardt's, 0.5% sodium dodecyl sulfate (SDS), 1 mg ml−1 shared salmon sperm DNA). Filters were incubated in the same solution with radiolabelled targets at 65°C overnight. Hybridized filters were washed for 5 min in 2× SSC, 0.5% SDS solution at room temperature and for 15 min in 2× SSC, 0.5% SDS at 65°C twice. Filters were exposed to Imaging Screen K film (Bio-Rad, Hercules, CA, USA) for 3 d. At the end of the exposure, autoradiographies of the filters were digitized using PhosphorImager system (Amersham Biosciences, Little Chalfont, UK). Detection and quantification of signal intensities were analysed with ArrayVision software (Imaging Research Inc., St Catharines, Canada). Local mean background was subtracted from each spot's intensity to calculate net signals. Interfilter normalization was performed using signal from an internal control. Three hundred and eighty-eight replicates of Desmine DNA were spotted at specific positions on the filters. Each spot intensity signal value was normalized by dividing it by the ratio between Desmine's mean intensity on the filter and global Desmine's mean intensity throughout all filters.

Data analyses were carried out with SAS (Statistical Analysis System version 6.12; SAS Institute, Cary, NC, USA). Analysis of variance (anova) was performed for target, filter and interaction effect for the intensity values of each clone, using PROC GLM with the following model:

Yijk = µ + Fi + Tj + (FxT)ij + eijk(Eqn 1)

(Yijk is the intensity value of the kth replicate of the clone; µ is mean overall replicates; Fi is the filter effect (i = 1–2); Tj is the target effect (j = 1–3); (FxT)ij is the interaction effect between filter i and target j; eijk is the residual part of observation not accounted for by the above effects (k = 1–8)). The mean values of the targets were compared using the MEANS/SNK option in PROC GLM in order to rank intensity levels of each clone. For further steps, we selected clones that were differentially expressed (P < 0.05) and whose target means were statistically different, by at least twofold, between at least any two conditions (Q, E and L), according to the Student–Newman–Keuls (SNK) method.

Sequencing and bioinformatics

Differentially expressed ESTs were selected and DNA inserts from the clones were first amplified, using M13 forward (5′-GTAAAACGACGGCCAG-3′) and reverse (5′-CAGGAAACAGCTATGAC-3′) primers. Then, PCR products were purified with QIAquick 96 PCR purification kit (Qiagen) and, sequenced with a Megabace 1000 automated DNA sequencer (Amersham Biosciences Inc.) using the DYEnamic ET Dye Terminator kit (Amersham Biosciences Inc.). Sequences were analysed using the Stack–Pack line (http://www.egenetics.com/stackpack.html) in order to create a nonredundant dataset indexed by gene. A putative function was assigned to each gene using the blast procedure (Altschul et al., 1997). A gene product was assigned to each EST, on the basis of sequence similarity to proteins with known function in the trEMBL and Swiss-prot databases, using blastx with an e-value ≤ 10−10. The whole dataset (sequence and annotation) is available on line at http://cbi.labri.fr/outils/SAM/COMPLETE/index.php under the project name ‘Quercus petraea bud ESTs’. A functional category was assigned to each gene product according to the MIPS (Munich Information Center for Protein Sequences) classification.

Expression level assessment using cDNA-macroarrays

The expression levels of the clones selected by colony array hybridization were evaluated by reverse-Northern blotting using cDNA-macroarrays. First, plasmid DNAs of 801 selected clones were amplified using the M13 forward and reverse primers. High-density nylon filters were prepared by Eurogentec (Liège, Belgium). Each clone was arrayed in triplicate on four filters. Hundreds of control DNAs (Desmine) and water were spotted on each filter to normalize intensity values between filters. cDNAs from each of the five developmental stages were then used to synthesize targets, as described earlier. Filters were hybridized with target 0, I, II, III, IV and V and washed, exposed and digitized as described above. Normalized intensity (using Desmine ratio) values were analysed using a two-way anova model under the following model:

Yijk = µ + Fi + Tj + (FxT)ij + eijk(Eqn 2)

(Yijk is the intensity value of the kth replicate of the clone; µ is mean overall replicates; Fi is the filter effect (i = 1–4), Tj is the target effect (j = 1–6); (FxT)ij is the interaction effect between filter i and target j; eijk is the residual part of observation not accounted for by the above effects (k = 1–12)). Comparisons of mean values were also carried out to rank the target intensity levels for each gene. Expression patterns of the remaining unique transcripts, were clustered, using the program epclust (http://ep.ebi.ac.uk/EP/EPCLUST/, European Bioinformatics Institute, Hinxton, UK).

Real-time RT-PCR

The expression level of 10 highly regulated genes was finally assessed by real-time RT-PCR, using a LightCycler 2000 instrument (Roche Applied Science, Mannheim, Germany). For the reverse transcription, we used SuperScript RNase H Reverse Transcriptase (Invitrogen), supplemented with 2 µm oligo d(T)22 primers, dNTPs at 0.7 mm and 2.5 µg of total RNA. Primer pairs were designed to amplify fragments varying in size from 203 bp to 410 bp (Table S1). Real time RT-PCR was performed with 1 or 2 µl of diluted cDNAs, depending on the level of expression of the targeted product. Serial dilutions of quantified PCR products were used as standard templates. Triplicates of cDNA were used to minimize experimental error, and each PCR run was carried out at least in duplicate for each of the two RNA extraction sets (Fig. 2). The expression levels were computed relative to the stable expression level of a 60S ribosomal protein L13 gene (accession P41127) used as a standard. The PCR reactions were performed in a final volume of 10 µl. The assays contained 1 µl of LightCycler FastStart DNA Master SYBR Green I (Roche Applied Science), 300 nm of each primer, 1.5–4 mm of MgCl2 and 1 or 2 µl of cDNA template.

Results

Identification of differentially expressed genes

To identify ESTs differentially expressed during bud burst, a targeted approach based on subtractive cDNA libraries construction followed by a prescreening step using high-density colony array (Chen et al., 2003) was developed. A total of 1800 clones were arrayed onto nylon filters. Each filter was hybridized with targets corresponding to three mRNA pools: Q (enriched for stage 0 transcripts), E (enriched for stages I and II transcripts) and L (enriched for stages III, IV and V transcripts). In order to ensure the correct assessment of the expression levels of the clones, hybridizations were performed twice, providing eight replicates per clone. Expression of about two-thirds of the clones did not change during bud burst (Fig. 4). By contrast, 604 ESTs were found to be upregulated from the ‘Quiescent’ to the ‘Early’ stage and their level of expression varied from 2- to 39-fold. Similarly, the expression of 513 ESTs increased during the ‘Late’ flushing compared with the ‘Quiescent’ stage, corresponding to a 2- to 13-fold change. Conversely, 98 and 109 clones were down-regulated during the ‘Early’ and ‘Late’ flushing stages, respectively, compared with the ‘Quiescent’ stage. Overall, 900 clones found to be differentially expressed using high-density colony array analysis, and whose expression varied more than twofold (increase or decrease) between at least two of the three conditions, were selected and sequenced.

Figure 4.

Scatter plots of signal intensities (I-stage) for all expressed sequence tags (ESTs) in the colony array. (a) Contrast between quiescent stage (x-axis) and early stage (y-axis). (b) Contrast between quiescent stage (x-axis) and late stage (y-axis). The solid diagonal line is the bisecting line and the dotted lines represent the cut-off lines at twofold induction or repression of gene expression.

EST analysis and functional classification

After discarding vector sequences and poor quality sequences, 801 sequences remained available: 403, 206 and 192 originated from the Early, Late and Quiescent libraries, respectively. A total of 302 694 nucleotides were sequenced, containing 43.8% of GC and 2.4% of unassigned nucleotides. The average EST length was 376 bp, ranging from 71 to 712 bp. Length frequencies were equally distributed from 200 to 700 bp, except for the Q library which contained 35% of ESTs longer than 500 bp. These 801 ESTs represented 233 unique transcripts (137 consensus, including 705 ESTs and 96 singletons) corresponding to a redundancy (number of ESTs in consensus – number of consensus/total number of ESTs) of 71%.

A functional role was assigned for each transcript on the basis of sequence similarity to proteins with known functions in GenBank using blastx with an e-value ≤ 10−10. Thirty-five per cent of the transcripts fell into the ‘hypothetical’ (8%) or ‘no hit’ (27%) categories. Over the three libraries, the category ‘energy’ was predominant with 21% of the ESTs, followed by ‘protein synthesis’ and ‘cell rescue, defense and virulence’, 11% each. The other categories shared 0.4–5% of the remaining ESTs. The comparison of the categories between the three libraries was much more informative (Fig. 5). Indeed, the ‘energy’ class accounted for 2.6% of the ESTs in the Q library, whereas it represented 30.9% and 20.4% of the ESTs, respectively, in the E and L libraries. The proportion of ESTs allocated to ‘protein synthesis’ was lower in the Q library (5.2%) than in the E and L libraries (8.4% and 19.9%, respectively). A similar trend was observed for the ‘cell cycle’ ESTs. Conversely, the proportion of ‘cell rescue/defense’ ESTs was higher in the Q library (16.6%) compared with the E and L libraries (14.1% and 1.5%, respectively). This was also true for the metabolism category (Q: 6.2%; E: 4%; L: 1.5%). Finally, hypothetical ESTs were more abundant in the Q library (15.5%) than in the E and L libraries (7% and 3%, respectively). Overall, this analysis showed very clearly that the spectrum of genes from each library was quite specific. In other words, the SSH approach succeeded in isolating cDNAs specific to particular stages of bud burst. Interestingly, a larger proportion of ‘cell rescue/defense’ class was previously reported in dormant cambial meristems (Schrader et al., 2004), whereas active cambial meristems exhibited more ‘protein synthesis’-related ESTs. As for our buds libraries, a large fraction of cambium transcripts lacked functional information, illustrating our limited knowledge of molecular factors involved in cambial meristems activity and bud development.

Figure 5.

Functional classification of expressed sequence tags (ESTs) following MIPS (Munich Information Center for Protein Sequences) categories in the different libraries.

Transcript profiling

Gene expression levels were carried out for the six stages of bud burst, using high-density filters prepared with the 801 cDNA clones (representing 233 unique genes). Forty-three genes, among the 233 unique, that exhibited no target effect were discarded. Expression patterns of the 190 remaining genes (Table S2) were clustered to study their relatedness (Fig. 6). We used the most mature stage (#V) as a reference to display expression results. As a result of the hierarchical clustering procedure, the genes were classified into three main groups. These groups reflect the general trends but several genes exhibited expression patterns slightly different from the general tendency. To determine whether particular functions were specific from one group to another, the main functional categories were compared between groups (Table 1). Some sequences were included in different groups under the same annotation, but e-scores were found to be different. These sequences reflect different members of the same multigene family.

Figure 6.

Hierarchical clustering of transcript accumulation between the first five stages of bud burst (S0, S1, S2, S3 and S4). For each stage, the Log2 value of ratio between a specific stage and stage V was represented.

Table 1.  Genes of known function listed by expression profile group
ESTAccession numberAssignmentBLAST e-valueFunctional category
  1. EST, expressed sequence tag.

Group I
Cons 75CR627918Galactinol synthase2 e-39Metabolism
Cons 70CR627933α-Amylase/subtilisin inhibitor4 e-15Metabolism
08D11CR627958Glucan endo-1,3-β-glucosidase, acidic isoform G1 e-22Metabolism
08E12CR627971Putative glycosyl hydrolase family protein1 e-13Metabolism
Cons 68CR627777Dihydroflavonol-4-reductase1 e-19Metabolism
06H01CR627762Cystathionine γ-synthase1 e-21Metabolism
Cons 28CR6275355′-Adenylylsulfate reductase 35 e-18Metabolism
06B10CR627726Catalase isozyme 36 e-91Cell rescue/defense
Cons 50CR62773660 kDa Dehydrin-like protein2 e-14Cell rescue/defense
Cons 52CR62771560 kDa Dehydrin-like protein2 e-14Cell rescue/defense
Cons 51CR62778460 kDa Dehydrin-like protein3 e-14Cell rescue/defense
08B09CR627936Endochitinase PR4 precursor3 e-61Cell rescue/defense
Cons 76CR627925Proteasome subunit α type 72 e-50Cell rescue/defense
Cons 41CR627646Stromal 70 kDa heat shock-related protein1 e-38Cell rescue/defense
Cons 1CR627507S-phase kinase-associated protein 1 A (SKP1)7 e-26Cell cycle
Cons 2CR627502S-phase kinase-associated protein 1 A (SKP1)2 e-26Cell cycle
Cons 72CR627907Hypothetical J-domain protein5 e-11Cell cycle
06E10CR627745NADP-dependent malic enzyme5 e-70Protein with binding function
08A03CR627920Squamosa-promoter binding protein-like 17 e-64Protein with binding function
Cons 71CR627780Tumor-related protein2 e-14Protein fate
Cons 130CR628241Glyceraldehyde 3-phosphate dehydrogenase A9 e-46Energy
Cons 61CR627776Late embryogenesis abundant protein Lea54 e-23classification not yet clear cut
Group II
Cons 19CR62751760S ribosomal protein L133 e-27Protein synthesis
Cons 14CR62750860S ribosomal protein L18a9 e-44Protein synthesis
Cons 134CR628060Chloroplast 50S ribosomal protein L165 e-26Protein synthesis
Cons 82CR62796150S ribosomal protein L142 e-26Protein synthesis
Cons 18CR62751340S ribosomal protein S283 e-20Protein synthesis
Cons 21CR62752340S ribosomal protein S272 e-30Protein synthesis
05D10CR62769140S ribosomal protein S212 e-21Protein synthesis
Cons 24CR627527Ribosomal protein 41, large subunit5 e-34Protein synthesis
Cons 122CR627872Ribosomal protein L335 e-20Protein synthesis
Cons 93CR627881Metallothionein-like protein type 34 e-25Cell rescue/defense
Cons 94CR627809Metallothionein-like protein type 37 e-27Cell rescue/defense
Cons 95CR627847Metallothionein-like protein type 37 e-27Cell rescue/defense
Cons 109CR627834Metallothionein-like protein type 23 e-20Cell rescue/defense
08E06CR627965Probable glutathione S-transferase parC1 e-47Cell rescue/defense
Cons 74CR627917Probable glutathione S-transferase5 e-51Cell rescue/defense
Cons 47CR62771218.2 kDa class I heat shock protein (HSP 18.2)3 e-58Cell rescue/defense
Cons 46CR62795218.2 kDa class I heat shock protein (HSP 18.2)4 e-55Cell rescue/defense
Cons 49CR62771460 kDa dehydrin-like protein1 e-14Cell rescue/defense
Cons 13CR627506Histone H43 e-40Transcription
Cons 112CR627839Histone H35 e-24Transcription
07B10CR627781DOF zinc finger protein DAG24 e-43Transcription
01E07CR627526Transcription factor MONOPTEROS1 e-11Transcription
Cons 45CR627667Cp10-like protein2 e-11Protein fate
08G03CR627985Cysteine proteinase RD19a precursor2 e-56Protein fate
11H03CR628138Cyanogenic β-glucosidase precursor9 e-30Metabolism
Cons 105CR627826Phosphoenolpyruvate carboxykinase9 e-49Metabolism
07F12CR627903Glyceraldehyde 3-phosphate dehydrogenase9 e-45Energy
Cons 103CR627822Pollen specific protein SF215 e-70Cell cycle
Cons 107CR627830Hemoglobin II5 e-57Cell type differentiation
Group III
Cons 10CR627859Photosystem II reaction center M protein2 e-11Energy
Cons 12CR627504Photosystem II reaction center M protein1 e-11Energy
Cons 7CR627546Photosystem II reaction center M protein8 e-12Energy
Cons 8CR627548Photosystem II reaction center M protein3 e-11Energy
Cons 9CR627815Photosystem II reaction center M protein3 e-11Energy
Cons 11CR627829Photosystem II reaction center M protein3 e-11Energy
12F05CR628202Photosystem II 10 kDa polypeptide1 e-25Energy
Cons 36CR627898Ribulose bisphosphate carboxylase small chain8 e-29Energy
Cons 37CR627808Ribulose bisphosphate carboxylase small chain5 e-28Energy
Cons 38CR627888Ribulose bisphosphate carboxylase small chain5 e-28Energy
Cons 35CR628069Ribulose bisphosphate carboxylase small chain SSU12 e-22Energy
Cons 119CR627860Small subunit of ribulose-1,5-bisphosphate6 e-24Energy
Cons 120CR628024Small subunit of ribulose-1,5-bisphosphate1 e-12Energy
01D08CR627520Cytochrome b6–F complex3 e-26Energy
Cons 106CR627828PLASTOCYANIN7 e-54Energy
Cons 128CR628019Oxygen-evolving enhancer protein 26 e-27Energy
Cons 131CR628030Glyceraldehyde 3-phosphate dehydrogenase A8 e-36Energy
Cons 34CR627577ATP synthase epsilon chain7 e-31Energy
Cons 126CR62800960S ribosomal protein L241 e-16Protein synthesis
Cons 96CR62781060S ribosomal protein L237 e-20Protein synthesis
13C05CR62825460S ribosomal protein L239 e-51Protein synthesis
01A03CR62750140S ribosomal protein S233 e-77Protein synthesis
Cons 136CR62806440S ribosomal protein S11-15 e-35Protein synthesis
Cons 25CR627552ribosomal protein L411 e-15Protein synthesis
Cons 42CR628185tRNA-Leu (trnL)1 e-134Transcription
Cons 43CR627811tRNA-Leu (trnL)1 e-139Transcription
Cons 44CR627807tRNA-Leu (trnL)1 e-149Transcription
03E12CR627611tRNA-Ile (Ile-tRNA)2 e-14Transcription
Cons 90CR628018Metallothionein-like protein type 31 e-26Cell rescue/defense
Cons 91CR628078Metallothionein-like protein type 34 e-26Cell rescue/defense
Cons 92CR628161Metallothionein-like protein type 35 e-27Cell rescue/defense
Cons 129CR628021Cytokinin-specific binding protein2 e-11Cell rescue/defense
Cons 127CR628014Early light-induced protein5 e-52Protein with binding function
Cons 20CR627519Early light-induced protein9 e-49Protein with binding function
Cons 29CR627538Chlorophyll A-B binding protein 78 e-95Protein with binding function
Cons 108CR627833probable aquaporin PIP1.41 e-50Cellular transport
Cons 60CR627750Vesicle-fusing ATPase2 e-33Cellular transport
Cons 110CR627835PVR3-like protein1 e-20Metabolism

The 70 genes belonging to group I were induced at stages 0 and/or I and II and, indistinctly, up- or down-regulated at stages III and IV. Moreover, most of these genes exhibited no differential expression between stage 0, stage I and stage II. This cluster mainly comprised metabolism-related genes, especially carbohydrate metabolism, such as ESTs encoding for galactinol synthase, glucan endo-1,3-β-glucosidase and α-amylase/subtilisin inhibitor. This group of genes preferentially expressed in the quiescent bud and/or at the onset of bud burst (but also at stage IV), also contained cell rescue/defense-related genes such as those encoding for catalase isozyme 3, dehydrin-like, stromal heat-shock proteins or Late Embryogenesis Abundant type 5 (LEA5). The ESTs involved in the cell cycle and corresponding to protein with binding function were also represented.

Conversely, the 55 genes of group II exhibited a differential expression between stage 0 and stages I and II, and were mainly downregulated at stages III and IV. As a result, genes allocated to group II may be important when considering early signals regulating bud burst. Protein synthesis and cell rescue/defense-related genes accounted for the majority of genes allocated to this group. Most of the ESTs exhibited a differential expression level between quiescent stage and the onset of flushing, such as those encoding for ribosomal proteins, as well as metallothionein-like protein type 3 and glutathione S-transferase. The transcriptional control and chromatin organization categories were also important in group II, such as transcripts encoding for histones H4 and H3, DOF zinc finger protein DAG2, or transcription factor MONOPTEROS (MP). Finally, protein fate, metabolism, energy, cell cycle and cell type differentiation categories were represented in group II.

The 65 genes in group III were largely repressed, compared with stage V, during the entire bursting process. This group mainly comprised energy-related genes, such as photosystem II reaction center M, ribulose biphosphate carboxylase, cytochrome b6–F complex, plastocyanin, or ATP synthase. Ribosomal proteins transcripts were also represented in this group, as well as tRNA genes, metallothionein-like type 3, or cytokinin-specific binding protein genes. Moreover, protein-binding, cellular transport and metabolism-related genes were found in group III, as exemplified by early light induced protein, chlorophyll A–B binding protein 7, probable aquaporin PIP1.4 or vesicle-fusing ATPase transcripts. Genes allocated to group III are mainly involved in the energy supply chain, especially photosynthesis. This result is correlated with the leaf development occurring from stages III to IV.

Real-time RT-PCR

To validate the expression data based on reverse-Northern analyses, we used real-time RT-PCR to quantify transcript accumulation of 10 genes chosen among the most regulated ones. We used at least six replicates per biological repetition for each cDNA template to ensure the reproducibility of the experiments. In our hands, PCR efficiencies ranged from 77 to 99%, which is consistent with a correct Taq polymerase activity. Among all genes, expression levels ranged from 0 to 108,000% when compared with that of the standard (60S ribosomal protein L13). There was no significant differential expression (t-test) between the two repetitions of RNA extraction (data not shown). Figure 7 shows transcripts accumulation values and standard errors among the two sets of RNA extractions. In all cases, results of real-time RT-PCR (Fig. 7) confirmed the trends obtained with macroarrays, although the absolute amounts were different and some discrepancies occurred. For each gene, we compared the relative amounts of transcripts at different stages between the two techniques used (e.g. macroarrays and real time RT-PCR). Despite some deviations, expression ratios between the two techniques were closed to 1 for each gene as exemplified in Fig. 8 for putative S-phase kinase associated protein 1A.

Figure 7.

Expression levels during bud bursting of 10 genes assessed by real-time reverse transcriptase polymerase chain reaction (RT-PCR). Expression is relative to that of a 60S ribosomal protein. blast e-value is indicated in brackets. Vertical bars represented standard deviations around the mean value.

Figure 8.

Expression levels of S-phase kinase-associated protein 1 A assessed in real time reverse transcriptase polymerase chain reaction (qPCR) (x-axis) and in macroarray experiments (y-axis). Expression levels are relative to that of the maximum amount of transcripts for each technique used (i.e. stage 0 for macroarrays and stage IV for qPCR).

Discussion

Combining SSH, HDCA, cDNA macroarrays and real-time RT-PCR to identify genes regulated during bud burst

We used SSH and HDCA to generate and isolate cDNA clones of genes regulated during bud burst in oak. This approach was reported as very efficient to rapidly identify differentially expressed genes (e.g. SSH and cDNA microarray used by Yang et al. (1999), or SSH, SH and differential screening used by Nakata & McConn, 2002)). In our case, the combination of SSH, HDCA and cDNA macroarrays has the obvious advantage of reducing the number of differentially expressed clones allowing for a closer look on the remaining ones. First, SSH was performed to ensure the isolation of specific and rare transcripts. Then, the HDCA permitted to reduce by half the number of clones, by selecting those that showed a significantly different expression between different stages of bud development. Finally, cDNA macroarrays highlighted genes highly regulated during bud burst. By using this stepwise procedure, we combined the advantages of global and more targeted gene discovery approaches. While thousands of clones were retained by the SSH, HDCA screening procedures reduced this number to several hundred (801 clones). Molecular pathways highly regulated during bud burst were thus highlighted.

Real-time RT-PCR experiments validated the results obtained by reverse-Northern blot. Despite problems inherent in quantification using this technique (Bustin, 2002; Vandesompele et al., 2002), real-time RT-PCR has been used as a powerful tool to validate arrays results (Chtanova et al., 2001; Jones et al., 2002; Pfaffl et al., 2003). Although some discrepancies occurred, differences of expressions were of similar magnitude to those obtained using reverse-Northern blot. The use of RT-PCR has been reported in animal physiology, endocrinology, immunology, virology, microbiology (Hopkins, 2002; Pfaffl et al., 2003; Schams et al., 2003; Pinzani et al., 2004; Stram et al., 2004; Thomas et al., 2004) and plant science (Khan & Shih, 2004; Thomas et al., 2004). It has been described as a technique of choice to assess the expression levels of gene transcripts, and preferred to the more traditional macroarray methods (Walker, 2002). In our study, the observed differences of expression patterns between both techniques could be explained by the amplification phase used to synthesize double-stranded cDNA in the case of macroarray radioactive targets. Although PCR reactions on single cDNAs were stopped during the exponential phase, some transcripts could have been artificially favoured or disfavored. Moreover, some differences could be caused by cross-hybridizations between genes from multigene families that probably occur on arrays, contrary to real-time RT-PCR, where primers are gene-specific.

As described in the original paper by Diatchenko et al. (1996), SSH is efficient ‘for generating cDNAs highly enriched for differentially expressed genes of both high and low abundance’. On one hand, six out of the 10 genes analysed by real-time RT-PCR were highly differentially expressed between stages 0 and I, namely LEA5 (Fig. 7a), galactinol synthase (Fig. 7b), DAG2 (Fig. 7c), alpha-amylase/subtilisin inhibitor (RASI) (Fig. 7f), At2g14910 (Fig. 7h) and At4g24120 (T19F6.8) (Fig. 7i). On the other hand, RASI and LEA5 transcripts were sixfold and 1000-fold more abundant than the control gene in the quiescent stage, respectively, whereas DAG2 or Zwh21.1 transcripts were 55-fold less abundant than the 60S ribosomal protein transcript used as control. These results provide candidate genes with potential roles in bud burst. It should be noted that several genes described as regulating seed germination, dormancy or bud burst were identified (e.g. DAG2 and SKP1) among the 10 genes assessed by real-time RT-PCR. This finding constituted the first level of validation for these candidate genes. Indeed, the DAG2-like zinc finger protein isolated in our study, has been hypothesized to act on a maternal switch that controls seed germination in Arabidopsis thaliana (Gualberti et al., 2002), and seed and bud dormancy have been hypothesized to involve similar processes (Rohde et al., 2000). For its part, S-phase kinase-associated protein 1 A (SKP1) is a component of the SCF complex necessary to trigger the G1 to S-phase transition in yeast, and possibly in plants (Horvath et al., 2003). Knowing that Devitt & Stafstrom (1995) found that cells in paradormant buds are arrested in G1 and at the G2/M boundary, SKP1 constitutes a relevant candidate gene for bud burst.

Molecular mechanisms revealed at the onset of bud burst

This work also provided tools to dissect molecular mechanisms regulated during bud burst. Although a majority of isolated transcripts were expressed during the entire process, some were specific for a particular stage or at least significantly up- or down-regulated at a specific stage. Because the monitoring of bud burst extended from the fully quiescent stage (0) to the stage of elongating shoots (V), transcript accumulation along the sequence not only changed due to bud burst but also due to elongating tissues. However, the morphological and cytological variation (Figs 1 and 3) of the bud clearly suggested that the transition between stage 0 and I corresponded to the release from ecodormancy. Potential roles of specific transcripts of stages 0 and I (i.e. the quiescent and the swelling bud) are discussed later. Conversely, genes preferentially expressed at the end of bud burst will not be considered as candidate genes for bud burst, as they appeared to be more related to leaf development than to bud burst.

Cell rescue and defense-related genes were expressed during the quiescent stage and at the onset of bud burst. Desiccation stress obviously occurred at the quiescent stage, as exemplified by LEA5 expression pattern (Fig. 7a). Indeed, although their precise biological function remains obscure (Wise & Tunnacliffe, 2004), transcripts encoding LEA proteins have been isolated in a wide range of vegetative tissues of plants under moisture stress. In addition to LEA proteins, dehydrin-like and heat-shock transcripts were found to accumulate at the onset of flushing, probably acting to protect cells from desiccation and temperature stresses, through accumulation of dehydrins and heat-shock proteins. Transcripts of galactinol synthase also increased at the onset of bud burst (Fig. 7b). Galactinol synthase catalyses the first step in the biosynthesis of raffinose family oligosaccharides (RFO) from UDP-galactose. These RFOs are thought to play a role in the desiccation tolerance of seeds, and galactinol synthase has already been reported to be involved in drought and heat-stress tolerance (Pukacka & Wojkiewicz, 2002; Taji et al., 2002; Zhao et al., 2003). As a result of these different mechanisms, the integrity of both membranes and proteins may be maintained in the bud tissues. It is also well known that water stress enhances the production of reactive oxygen species and increases susceptibility to pathogens (Bray et al., 2000). On one hand, we found that oxidative stress, counteracted by the activation of catalase, acting in free radical removal, was activated during bud break. Moreover, the regulation of the osmotic potential by the accumulation of dehydrins, LEA, or sugars could also contribute to the protection of cells against oxidation (Rhodes & Hanson, 1993). On the other hand, the Endochitinase PR4 precursor gene was upregulated during quiescent and early stages and may contribute to defense against pathogens.

Expression of histones H3 and H4, as well as that of putative transcription factors related to DAG2 (Fig. 7c) and MONOPTEROS from Arabidopsis thaliana, were found to be induced at the onset of bud burst. While elevated expression of histones H3 and H4 appears to reflect increased cell division activity in general, induction of some distinct transcription factor-like genes might provide some clues about developmental processes specifically taking place during bud burst. DAG2, by contrast, was shown to act as a transcription factor specifically involved in the maternal control of seed germination (Gualberti et al., 2002). In buds, as in seeds, it may potentially act on dormancy release, supporting the existence of a common basis for the control of seed and bud dormancy (Rohde et al., 2000), as first hypothesized by Wareing (1956). In line with this model, we also found that expression of a homologue of MONOPTEROS appears to be induced during bud burst. MP is an auxin response factor (ARF) that seems to act as a transcriptional activator, required for the control of axis formation in the embryo and in auxin-dependent cell expansion (Hardtke & Berleth, 1998; Hardtke et al., 2004). A similar role could be attributed to the MP homologue found in oak buds. By analogy to its counterpart in Arabidopsis, this gene might transduce auxin signals, essential for early developmental events occurring during bud burst. Expression of presumptive cell cycle-regulators was also affected by bud burst. Indeed, SKP1 (Fig. 7d) and pollen-specific protein SF21 (Fig. 7e) were upregulated at stage IV. This result may indicate that cell division and differentiation are shifted during bud burst: cell division occurring during the first stages and both cell division and differentiation at stage IV. Conversely, expression of SKP1 remained constant between quiescent and early active stages, as previously observed for the transition to dormancy for cambial meristems (Schrader et al., 2004). There was no gene related to epigenesis/cold requirement found to be regulated in our experiment.

Activity of glycosyl hydrolases, such as glucan endo-1,3-β-glucosidase and cyanogenic β-glucosidase precursor, were also enhanced at the quiescent and early stages. These enzymes are induced by gibberellic acid and are known to play a role in both cell-wall mobilization and cell elongation (Hrmova & Fincher, 2001), through hydrolysis of glycosidic bonds linking cell-wall components. Induction of these enzymes during bud burst could thus reflect the initiation of outgrowth taking place in early stages of bud burst. By contrast, another regulator of carbohydrate modification, namely an α-amylase/subtilisin inhibitor was highly expressed at the quiescent stage only (Fig. 7f), suggesting that hydrolysis of storage starch or glycogen is repressed in the quiescent bud. The observed reduction in the expression of this gene upon bud swelling would indicate the onset of starch mobilization at this developmental stage.

In general, an increase in the expression of genes essential for energy supply could be observed at the end of bud burst, as shown by expression patterns of transcripts encoding for photosystem II reaction center M protein, cytochrome B6-F complex, plastocyanin, oxygen-evolving enhancer protein 2, ATP synthase ɛ -chain, that were all classified into the Group III. In parallel, RubisCo expression was also upregulated, as well as that of glyceraldehyde-3-phosphate dehydrogenase (Fig. 7g), as previously reported by Wang et al. (1991). Indeed, these authors demonstrated that enzyme activity of the glycolytic pathway increased at the release of dormancy. Regarding the developmental stages used in this study (Fig. 1), this increase in the expression of energy-related transcripts is undoubtedly caused by leaf development, which starts at stage III, but mainly develops at stages IV and V.

In addition to these known genes, several unknown transcripts were shown to be differentially expressed using real-time RT-PCR (Fig. 7h–j). Two of these hypothetical proteins, T19F6.8 and Zwh21.1, contain domains specific for oligo-peptide transporters and F-box proteins, respectively. A presumptive function of these genes in the control of bud burst remains to be determined.

Conclusion

In summary, this study has confirmed the usefulness of combining subtractive cDNAs libraries and macroarrays to identify relevant candidate genes, but also to dissect the molecular mechanisms during a developmental process such as bud burst. This study has provided new insights in the understanding of gene expression during bud burst. It is one of the first to monitor transcript accumulation during the bud bursting process under natural conditions. This study showed that transcription factors, such as DAG2, known to be involved in the control of seed germination, is also highly expressed in quiescent bud of oak. This original result supports the existence of a common basis for the control of seed and bud dormancy. This study also provides novel sequences that are most likely involved in bud burst and clues for the identification of candidate genes for further analysis. Several unknown transcripts have also been isolated and their differential expression validated by real-time RT-PCR (Fig. 7h–j). Cloning entire genes coding for these unknown transcripts is a logical next step. The implication of the identified candidate genes on the timing of bud burst is on going in association studies.

Acknowledgements

We acknowledge E. Bertocchi, G. Roussel, J. M. Louvet and B. Montoussé for their technical help in the nursery and for the field collections. We thank G. Le Provost, J. Paiva, A. Ramboer, N. Ladouce and K. Silhavy for useful discussions and help in the laboratory. We also thank B. Salin and A. Marpeau-Bezard for their helpful work and comments on the microscopy work. This work was partly funded by French Ministère des Affaires Etrangères (programme Egide), Fondation Dufresnoy, the European Union (TREESNIPS: QLK3-CT-2002–01973, FEDER: 2003227), and the Aquitaine Region (2004-03-05–003FA). We thank INRA for funding a PhD fellowship to J.D.

Ancillary