Deciphering the genetic determinism of bud phenology in apple progenies: a new insight into chilling and heat requirement effects on flowering dates and positional candidate genes

Authors


Author for correspondence:
J-M. Celton
Tel: +33 628437885
Email: jean-marc.celton@angers.inra.fr

Summary

  • The present study investigates the genetic determinism of bud phenological traits using two segregating F1 apple (Malus × domestica) progenies.
  • Phenological trait variability was dissected into genetic and climatic components using mixed linear modeling, and estimated best linear unbiased predictors were used for quantitative trait locus (QTL) detection. For flowering dates, year effects were decomposed into chilling and heat requirements based on a previously developed model.
  • QTL analysis permitted the identification of two major and population-specific genomic regions on LG08 and LG09. Both ‘chilling requirement’ and ‘heat requirement’ periods influenced flowering dates, although their relative impact was dependent on the genetic background. Using the apple genome sequence data, putative candidate genes underlying one major QTL were investigated. Numerous key genes involved in cell cycle control were identified in clusters within the confidence interval of the major QTL on LG09.
  • Our results contribute towards a better understanding of the interaction between QTLs and climatic conditions, and provide a basis for the identification of genes involved in bud growth resumption.

Introduction

In the context of global climate change, vegetative and floral bud phenology of deciduous tree species is crucial as it may affect their productivity, adaptability and distribution (Chuine & Beaubien, 2001).

In order to adapt to the naturally changing environmental conditions, temperate tree species have developed the ability to establish a dormant state (endo-dormancy) in the rest period, during which meristems are unable to undergo ontogenic development towards bud burst (Doorenbos, 1953), followed by an eco-dormancy phase that occurs at the end of winter and the beginning of spring, during which meristems achieve full growth competence (Hänninen, 1995; Legave et al., 2008). In order for budbreak to occur promptly and uniformly in spring, trees require exposure to cold temperature (chilling requirement, CR), followed by a period of warmth (heat requirement, HR). The time of bloom is an important agronomic trait affecting seed and fruit development (Fan et al., 2010), and is quantitatively inherited in the majority of fruit tree species (Anderson & Seeley, 1993).

Apples (Malus × domestica) are classified as high, intermediate and low CR. The modeling of flowering time, based on mathematical functions to simulate CR and HR, is commonly based on temperature response functions. Recently, three models have been validated under a range of climatic conditions (Legave et al., 2008). Based on modeling and phenological data, CR required to break dormancy can vary from 200 (cv ‘Anna’) to 1500 (cv ‘Wright #1’) (Hauagge & Cummins, 1991). However, the majority of commercial apple cultivars are classified as intermediate to high CR, that is between 600 and 800 hours below 7°C of CR (Ferree & Warrington, 2003), and are poorly adapted to mild climates. Changes in apple tree blooming dates have already been observed throughout Europe (Legave et al., 2009). Results suggest that temperature changes have led to an advance in the mean flowering dates of 7–9 d for cv ‘Golden Delicious’ (GD).

If temperatures were to continue to increase, phenological disorders could be observed, including irregular floral and leaf budbreak and poor fruit set, as is already the case in South Africa (Labuschagnéet al., 2002), Brazil (Petri, 1987) and Florida (Sherman & Sharpe, 1971). Although the current strategy involves the application of dormancy-breaking chemicals, increased awareness of the negative environmental effects has resulted in the need to breed cultivars better adapted to the changing environmental conditions.

In order to understand the genetic determinism of the time of budbreak in apple, several studies have already been conducted using a linkage analysis approach. Conner et al. (1998) identified eight genomic regions influencing the time of budbreak using a population derived from a ‘Wijcik McIntosh’ × NY75441-58 cross. Later, Segura et al. (2007) identified additional quantitative trait loci (QTLs) associated with the time of budbreak using a population derived from a ‘Starkrimson’ × ‘Granny Smith’ cross. More recently, a study using several segregating populations with the low chilling cv ‘Anna’ as a parent (van Dyk et al., 2010) identified LG09 as having a major QTL effect on the time of budbreak.

Despite these efforts, no major QTL has yet been defined against a wide genetic background or across variable environments, and no attempt has been made to identify putative candidate genes (CGs) or regulatory pathways involved in the determinism of these traits. To do this, we combined two complementary approaches: one without a priori assumptions, screening one major zone of interest; the other based on an a priori assumption about the physiological mechanisms that could be responsible for the date of budbreak.

It is recognized that the break of dormancy in buds is based on processes that are intrinsic to the bud itself (Metzger, 1996), and that dormancy breaking is unlikely to be reliant on a linear control pathway (Rinne et al., 2001). Indeed, chilling restores the ability of buds to grow, but does not promote growth itself. The capacity to resume growth under favorable conditions, following CR fulfillment, results mainly from the capacity of cells to divide and elongate (Rohde & Bhalerao, 2007). We hypothesized that the key to unraveling the molecular determinism involved in the date of vegetative or flowering budbreak in apple may resides in genes involved in the cell cycle and its control. The search for CGs was thus refined on genes involved in cell cycle activation, inhibition and regulation (Meijer & Murray, 2001) (Fig. 1). Phytohormones were also considered, primarily auxins and cytokinins, as they influence cell proliferation (del Pozo et al., 2005), as well as genes playing a major role in growth processes, for example, for the organization of the orientation of microtubules during cell expansion (Schwab et al., 2003), or involved in plant cell wall loosening (Cosgrove, 2000). The purpose of our study was first to further our knowledge on the genetic determinism of apple time of budbreak using a QTL mapping approach based on the best linear unbiased predictor (BLUP) of the dates of budbreak, vegetative budbreak and floral budbreak. This study was performed on two independent F1 apple populations derived from the crosses ‘Starkrimson’ × ‘Granny Smith’ (STK × GS) and X3263 × ‘Belrène’. Second, we investigated the interactions between QTLs, CR and HR. Finally, we proposed putative CGs underlying one major QTL by in silico mapping using the recently released apple genome sequence (Velasco et al., 2010).

Figure 1.

Genetic and hormonal control of the plant cell cycle (inspired by del Pozo et al., 2005). Specific cyclin-dependent kinase (CDK)/cyclin complexes control the two major pathways regulating G1/S and G2/M transitions. Hormones (indicated in boxes) control the level of CDK activators and inhibitors, and proteasome degradation mediates the level of cell cycle regulators by either SCF or APC complexes. ABA, abscisic acid; APC, anaphase promoting complex; AXR, auxin resistant; BRs, brassinosteroids; CDK, cyclin-dependent kinases; CK, cytokinins; CYC, cyclin; E2F, family of transcription factors; HBT, HOBBIT; IAA, indol acetic acid; JA, jasmonic acid; KRP, p27kip-related protein; PRZ1, PROPORZ1; RBR, etinoblastoma-related protein; SCF, Skp1-Cullin1-F-box.

Materials and Methods

Plant material

The first F1 progeny is derived from a cross between apple ‘Starkrimson’ and ‘Granny Smith’. As described in Segura et al. (2008), it comprises 123 seedlings replicated twice and grafted onto ‘Pajam 1’ apple rootstocks. Both replicates were planted in 2004 at the Melgueil INRA Montpellier experimental station (co-ordinates: 43°36′35″N; 3°58′50″E) and grown with minimal training and under irrigated conditions.

The second F1 progeny is derived from a cross between the hybrid X3263 and the cv ‘Belrène’. X3263 is derived from a cross between ‘Red Winter’ and X3177, the latter being itself a hybrid derived from a cross between ‘Idared’ and ‘Prima’ (F. Laurens, pers. comm.). X3263 was bred at the INRA station of Angers and is considered to be of type 3 (intermediate) growth habit according to the classification of Lespinasse (1992), whereas ‘Belrène’ is considered to be class 2 with a more erect growth habit. This population is composed of 324 trees, 50 of which were randomly selected to produce replicates. All 324 trees were phenotyped, and 271, including the replicated trees, were used for linkage map construction.

All trees from the population, including the replicates, were grafted onto ‘Pajam 1’ apple rootstocks and planted in 2005 under similar growing conditions.

All parents are classified as moderate to high CR.

Phenotypic assessment

Three phenological stages were defined, based on Fleckinger’s apple phenological classification (Fleckinger, 1964). These stages comprised: date of green point (GP), corresponding to 50% of all tree buds (either vegetative or flowering) showing some green (C3, D); date of vegetative budbreak (VB), corresponding to 50% of the vegetative buds showing the first deployed leaf (D3); and date of flowering budbreak (FB), corresponding to 50% of the flower clusters having the king flower open.

From the beginning of the budbreak period, trees were observed at the whole tree scale three times a week. Phenotypic assessments were recorded from 3 to 6 yr depending on trait and population (Table 1).

Table 1.   Phenotypic assessment of the apple (Malus × domestica) populations derived from the crosses between ‘Starkrimson’ × ‘Granny Smith’ (SG) and X3263 × ‘Belrène’ (XB) between the years 2005 and 2010
 200520062007200820092010
  1. FB, floral budbreak stage; GP, green point stage; VB, vegetative budbreak stage; –, absence of phenotypic assessment.

GPSGSGSGSG/XBSG/XBSG/XB
VBSGSG/XBSG/XBSG/XB
FBSGSG/XBSG/XBSG/XB

Statistical analyses

Statistical analyses were performed using R software v2.8.1 (R Development Core Team, 2009). The significance of the various effects was estimated using a linear model Pij ∼ μ + Gi + Yj + Gi × Yj + εij, where P is the phenotypic value of tree j of genotype i, μ is the total average of the population, Gi is the effect of genotype i, Yj is the effect of year j, Gi × Yj is the interaction between the genotype and year, and εij is the residual error.

Broad-sense heritability of genotypic means (h²b) was estimated using a balanced subset of the data to allow the estimation of the G × Y effect. This set is defined here as comprising all the replicated individuals for which phenotypic data of the character of interest were recorded for every year studied. Heritability was estimated as: inline image, where F is the Fisher statistic obtained using ANOVA type I (Gallais, 1989). Associated confidence intervals (CIs) were calculated according to Knapp et al. (1985):

image

Variables were considered to be heritable if their h2b value was > 0.2, and if the lower limit for their CI was > 0 (Gallais, 1989).

To further study the environmental effect on the FB variable, each year (from 2007 to 2010 for STK × GS, and from 2008 to 2010 for X3263 × ‘Belrène’) was characterized by the length of the periods required to fulfill CR (starting from November 1 of the previous year) and HR (starting the estimated day following CR fulfillment). Both periods were estimated using a model built for the cultivar ‘GD’ derived from Legave et al. (2008), and thereafter called F1Gold1. This model, based on the hypothesis that chilling and heat temperatures have successive and independent effects on trees, was built using a triangular function in the chilling submodel, and an exponential function in the heat submodel. As described previously, chilling and heat effects, and their interaction with genotypes, were estimated by ANOVA on a balanced dataset.

Finally, mixed linear models were built for all variables, including the year (Y) as a fixed effect, and genotype (G) or genotype by year (G × Y) interaction as random effects. The models were estimated using the residual maximum likelihood (REML) estimation method and effects to be included were selected based on the Akaike information criterion (AIC). For each trait, when G and G × Y effects were included, the BLUP was computed for the G effect. This G BLUP was considered to be independent of climatic year. It was denoted by the trait name and used for QTL detection.

DNA extraction and molecular marker genotyping

Purification of F1 seedling and parental DNA was performed from young leaf tissue using the DNeasy 96 plant kit (Qiagen) following the manufacturer’s instructions. Simple sequence repeat (SSR) marker amplifications were performed as described by Gianfranceschi et al. (1998). Single nucleotide polymorphism (SNP) markers were originally identified within the apple genome sequencing project as heterozygous single nucleotide positions within the cultivar ‘GD’. SNPs were organized and genotyped as five 48-plexed SNP arrays, based on the SNPlex™ (Applied Biosystems Inc., Foster City, CA, USA) technology, as described in Micheletti et al. (2011).

Genetic linkage map construction

For the STK × GS population, in addition to the 107 SSR markers from Segura et al. (2009), a further six SSR and 44 SNP markers were mapped to the consensus linkage map (B. Guitton et al., unpublished).

For the X3263 × ‘Belrène’ population, a framework genetic map was constructed using 271 individuals with 83 SSR markers (Liebhard et al., 2002; Silfverberg-Dilworth et al., 2006), previously tested for their polymorphism on the parents (144 screened for polymorphism on the parents in total), and 128 SNP markers (Velasco et al., 2010).

Linkage analysis was performed using JoinMap version 3.0 (Van Ooijen & Voorrips, 2001) with a logarithm of the odds (LOD) score of four for grouping. Genetic maps were constructed independently for each parent, and a consensus linkage map was built using all segregating markers. Genetic distances between markers were calculated using the Kosambi mapping function, and linkage groups (LGs) were numbered in accordance with Maliepaard et al. (1998).

QTL analysis

QTL analysis was performed using MapQTL®4 (van Ooijen et al., 2002) on the extracted G BLUP values and on the mean value per genotype for each year. QTL analysis was carried out using the interval mapping (step size 1 cM) and multiple QTL mapping (MQM) functions. QTLs were declared significant if the maximum LOD, obtained following multiple rounds of MQM, exceeded the genome-wide LOD threshold calculated (1000 permutations, mean error rate of 0.05). Each QTL was characterized by its LOD score and percentage of phenotypic variation explained. QTLs were graphically displayed as bars next to the LG on which they were identified using MapChart version 2.0 (Voorrips, 2001), and CIs were estimated in cM and corresponded to a LOD score drop of one or two on either side of the likelihood peak.

Allelic effects were estimated for female and male additivity, and for dominance.

A global model including all cofactors and their interactions, considered as fixed effects, was built to test epistatic effects between QTLs when several QTLs were detected for a trait BLUP. The construction of such a model allowed the estimation of the global percentage of phenotypic variation (global R2) explained by all the QTLs.

In silico mapping of CGs

Predicted protein sequences derived from contigs underlying the major QTL on LG09 were downloaded from the IASMA genome browser (http://genomics.research.iasma.it/gb2/gbrowse/apple).

Gene ontology (GO) annotations were performed using BLAST2GO. The peptide sequence was loaded into the program, and the BLASTP function was run against the GenBank nonredundant protein database. A minimum E-value of < 10−3 was used before mapping and annotation into GO terms.

A similar analysis was performed on an equivalent number of gene sequences selected randomly throughout the genome. A χ2 test was then performed to compare the gene composition of the LG09 QTL with this random subset, and to identify classes of genes differentially present within the QTL interval.

Further BLAST searches were performed locally with the latest SWISS-PROT database. To identify putative CGs, we used two distinct strategies. The first involved the identification of genes with potential involvement in phenotypic variation. For this purpose, in view of the physiology of vegetative and floral bud development, we chose to focus our attention on genes involved in cell cycle and division, and on its hormonal control. The second strategy involved the identification of putative CGs located closest to the LOD peak, as it is statistically the most likely location of the gene(s) responsible for the variation.

Results

Construction of genetic maps

The integrated map constructed for the STK × GS population enabled the positioning of 157 markers, including 113 SSR and 44 SNP markers, over 17 LGs and covering 1027 cM (data not shown).

A total of 211 genetic markers was mapped on the integrated X3263 × ‘Belrène’ genetic map, which covered 1068 cM over 17 LGs (data not shown). Each of the 83 SSR markers tested were successfully mapped, and 128 SNPs, of the 240 tested (53%), were polymorphic and mapped to the consensus map.

Both consensus linkage maps were aligned to the reference map (Silfverberg-Dilworth et al., 2006) and no inversion in marker order was observed.

Phenotypic trait assessment

Data distribution and significance of CR and HR effects  The analysis of the dates of GP, VB and FB revealed a normal distribution for both populations across most of the years phenotyped. The distribution of the variable FB is shown as an example (Fig. 2). Both populations showed a similar pattern of distribution over the years, and the estimated dates of FB for ‘GD’ were included in the FB dates recorded for both populations, except for 2010.

Figure 2.

Distribution of the flowering budbreak dates (in days) for the apple (Malus × domestica) populations ‘Starkrimson’ × ‘Granny Smith’ and X3263 × ‘Belrène’. Average parental floral budbreak (FB) dates (three parental replicates) are represented by arrows. A full line represents the female parent of each cross, and the male parent is represented by a dashed line. The x-axis represents the date of FB (in days, starting from November 1 in the year before); the y-axis represents the proportion of trees that reached the FB stage at their respective dates. The simulated FB date for cultivar ‘Golden Delicious’ (GD) (in days, starting from November 1), as well as the estimated number of days necessary to reach chilling and heat requirements, is given below the graphs corresponding to the four years studied. In 2007, only one replicate of each parent flowered for ‘Starkrimson’ (STK) and ‘Granny Smith’ (GS).

As shown in Fig. 2, the average dates of FB were identical in 2007 and 2010 for the STK × GS population. However, these two years had different climatic conditions: 2007 was characterized by a mild winter (100 d CR), followed by a warm spring (65 d HR), whereas 2010 was characterized by a normal winter (90 d CR) but cold spring conditions (83 d HR). Similar FB dates indicate that the warm spring conditions might have compensated for the long CR period observed in 2007, whereas the cold conditions of spring 2010 resulted in a later than expected FB date based on CR length period. Furthermore, 2010 showed an atypical pattern of data distribution (L shape), with most genotypes flowering within a short period. The years 2008 and 2009 were representative of average winter and spring conditions in the south of France.

Significance of the effects and broad sense heritability  For both populations, results showed highly significant G, Y and G × Y interaction effects (< 0.01) on all variables (data not shown). To refine these effects, the effects of the length of CR and HR periods on FB were estimated (Table 2). The results indicated that both periods have an effect on FB. For the STK × GS population, the CR effect on FB was stronger (= 1066) than the HR effect (= 353), and the G × CR effect was close to significant ( 0.1). For the X3263 × ‘Belrène’ population, both CR and HR periods had a similar effect on FB, with = 155 and = 153, respectively. The G × CR interaction had a significant effect on FB ( 0.01). No significant G × HR interaction effect was detected.

Table 2.   Significance (P) of the genotype effect (G), chilling requirement effect, heat requirement effect and the interactions G × chilling and G × heat for the date of flowering budbreak (FB), estimated by ANOVA on balanced data for the apple (Malus × domestica) populations ‘Starkrimson’ × ‘Granny Smith’ (STK × GS) and X3263 × ‘Belrène’
 STK × GSX3263 × ‘Belrène’
PF valuePF value
  1. The significance of each effect is indicated, followed by the F value: *,  0.1; ***,  0.01; ns, not significant.

G***5.38***6.12
Chilling requirement***1066***155
Heat requirement***353***153
G × chilling*1.26***2.54
G × heatns0.75ns0.88

High heritability values were estimated for GP, VB and FB (Table 3) for both populations. Values were between 0.83 (FB, STK × GS population) and 0.92 (GP, STK × GS and X3263 × ‘Belrène’ populations).

Table 3.   Estimation of the heritability value (h2) of the phenotypic traits green point stage (GP), vegetative budbreak stage (VB) and floral budbreak stage (FB) calculated for the apple (Malus×domestica) populations ‘Starkrimson’ × ‘Granny Smith’ (STK × GS) and X3263 × ‘Belrène’
 Heritability (h2)
STK × GSX3263 × ‘Belrène’
  1. Heritability values are presented in bold and are followed by the estimated confidence intervals in parentheses.

GP0.92 (0.90–0.94)0.92 (0.88–0.96)
VB0.88 (0.84–0.91)0.91 (0.86–0.95)
FB0.83 (0.76–0.88)0.84 (0.71–0.92)

Correlation analysis  Correlation coefficients between phenotypic traits measured in consecutive years varied from 0.47 (GP07–GP08) to 0.06 (VB09–VB10; FB08–FB09; FB09–FB10) for the STK × GS population, and from 0.42 (GP08–GP09) to zero (VB09–VB10) for the X3263 × ‘Belrène’ population (Table 4). Correlation coefficients also varied among phenotypic traits measured in the same year. GP and VB were correlated in 2007 and 2008 (R2 = 0.32 and 0.28, respectively), whereas the correlation between these two traits was low in 2009 and 2010 (R2 of 0.07 and zero, respectively), for the STK × GS population. A similar trend was observed for the X3263 × ‘Belrène’ population, with correlation coefficients among traits decreasing for the years 2009 and 2010 (Table 4).

Table 4.   Correlation coefficients indicating varying phenotypic association between consecutive years and among traits measured in the same year (from 2007) Thumbnail image of

QTL analysis

STK × GS population GP  One QTL was identified using BLUP of the genotypic effect. This QTL, located on the top half of LG08 and explaining 23.1% of the variability, displayed mainly female additive effects. For 2005 (Fig. 3), two QTLs were identified on LG06 and LG08, and explained 15% and 17% of the variability, respectively (Table 5). Both QTLs were characterized by female additivity and a dominance effect. For GP06, one QTL identified on LG10 explained 23% of the variation and was characterized by female additivity and a dominance effect. One QTL with female additivity was detected for GP07 on LG08 and explained 27% of the variation. For 2008 and 2009, QTLs were also detected on LG08, and explained 18% and 19% of the variability, respectively. For both years, these two QTLs were characterized by female additivity. Finally, two QTLs, located on LG09 and LG12, and explaining 16% and 21% of the variability, respectively, were identified for the variable GP10. The LG09 QTL was characterized by a dominance effect, whereas the LG12 QTL was characterized by female additivity. Together, both QTLs explained 21.2% of the variability.

Figure 3.

Genomic positions of quantitative trait loci (QTLs) detected on the linkage groups (LGs) of the apple (Malus × domestica) ‘Starkrimson’ × ‘Granny Smith’ integrated map by multiple QTL mapping (MQM) for the variables green point (GP), vegetative budbreak (VB) and floral budbreak (FB). QTLs are represented on the right side of LGs by boxes extended by lines representing the logarithm of the odds (LOD)-1 and LOD-2 confidence intervals. The numbering of LGs is according to Maliepaard et al. (1998).

Table 5.   Parameters associated with the quantitative trait loci (QTLs) detected in the apple (Malus × domestica) ‘Starkrimson’ × ‘Granny Smith’ population by multiple QTL mapping (MQM) for the phenological traits described as green point (GP), vegetative budbreak (VB) and floral budbreak (FB)
TraitLGPosition (cM)LocusLODaR2bAllelic effectscGlobal R2dAmeAffDg
  1. Each trait is followed by the year (2005–2010) in which the QTLs were detected. BLUPs (best linear unbiased predictors) indicate QTLs detected for the genetic effect.

  2. aMaximum logarithm of the odds (LOD) score value of the QTL with the considered threshold in parentheses.

  3. bPercentage of the phenotypic variation explained by the QTL.

  4. cSignificance of the allelic combinations estimated by the global model based on genotypic information from the locus used as co-factor.

  5. dVariation explained by all QTLs estimated by the global model.

  6. eMale additive effect computed as [(μac + μbc) − (μad + μbd)]/4, where μab, μad, μbc and μbd are the estimated phenotypic means associated with each of the four possible genotypic classes ab, ac, ad and bd, deriving from an <ab×cd> cross.

  7. fFemale additive effect computed as [(μac + μad) − (μbc + μbd)]/4, where μab, μad, μbc and μbd are the estimated phenotypic means associated with each of the four possible genotypic classes ab, ac, ad and bd, deriving from an <ab×cd> cross.

  8. gDominance effect computed as [(μac + μbd) − (μad + μbc)]/4, where μab, μad, μbc and μbd are the estimated phenotypic means associated with each of the four possible genotypic classes ab, ac, ad and bd, deriving from an <ab×cd> cross.

  9. *, P = 0.01; **, P = 0.001; ***, P = 0.

GP05635.1CH03d125 (4.3)0.152  −0.2061.7374.173
86.0CLV14.71 (4.3)0.173No effect −0.813−2.3302.206
GP06106.7Hi02d045.7 (4.6)0.236  −0.656−4.3014.631
GP07827.6CH02g098.29 (4.2)0.271  −1.744−3.5082.781
GP08827.6CH02g095.14 (4.7)0.182  −1.102−2.5542.288
GP09822.4Hi04b124.98 (3.9)0.188  −0.987−4.013−0.312
GP1097.7CH01f03b5.42 (4.2)0.159ad**, bc*0.2120.2570.350−2.624
1243.7CH01f025.88 (4.2)0.213ad**, bd** −0.6601.050−0.105
GP_BLUPs825.4CH02g096.35 (4.0)0.231  −0.420−0.7890.514
VB07No QTL        
VB08564.2CH04e034.58 (4.0)0.144bc***, bd*0.262−0.224−4.146−0.464
1035.8CH03d116.58 (4.0)0.264lm*** 0.197−4.822−2.005
VB09No QTL        
VB10229.7AP25.45 (4.2)0.216  −0.435−6.260−0.178
VB_BLUPs230.7AP26.62 (4.1)0.165 0.6620.0000.0000.000
356.0Hi07e08y6.55 (4.1)0.159bd 0.0000.0000.000
565.0CH04e034.33 (4.1)0.099ad 0.0000.0000.000
69.9VRN24.65 (4.1)0.120  0.0000.0000.000
  AP2:CH04e03  lm:bd*    
  Hi07e08:VRN2  ad:eg**, bc:eg*    
  AP2:Hi07e08: VRN2  lm:ad:eg*    
     lm:bc:eg**    
     lm:bd:eg**    
FB0763.0HB09TC5.76 (3.8)0.208lm*0.337−1.041−3.267−3.101
645.1MFT4.86 (3.8)0.233lm*** 0.410−3.4693.067
80.0CLV1a7.74 (3.8)0.236eg***, fg* −1.457−0.8860.534
FB0880.0CLV1a6 (4.0)0.209  −1.341−0.211−1.273
FB09819.4Hi04b122.51 (2.3)0.087lm**0.179−0.955−6.857−4.113
1237.7CH01f023.11 (2.3)0.105ad*** −2.002−5.256−3.956
FB10No QTL        
FB_BLUPs813.5CH01e125.1 (4.0)0.195  −1.796−0.513−0.831

Because of the strong female additivity displayed by all the QTLs identified on LG08 for the variable GP, these QTLs were also detected for all years at the same location on the female parental genetic map (data not shown).

Vegetative budbreak  Four QTLs were identified for the genotypic effect BLUP. These QTLs were located on LG02, LG03, LG05 and LG06, and explained 21.6%, 16.5%, 15.9% and 9.9% of the variability, respectively. The global model build with markers closest to the LOD peak included interactions between markers only and accounted for 66.2% of the variation (Table 5). For 2008, two QTLs were mapped on LG05 and LG10 (Fig. 3), and explained 14.4% and 26.4% of the variation, respectively. Both QTLs displayed a female additive effect, and together explained 26.2% of the total variability. No QTL was detected for VB09. For 2010, one QTL was identified on LG02. It explained 21.6% of the variability and displayed a female additive effect.

Floral budbreak  One QTL explaining 19.5% of the variability and displaying a significant male additive effect was mapped for G BLUP on LG08. In addition, QTLs were detected for three of the four years investigated for FB. For 2007, three QTLs were identified, two on LG06 and one on LG08 (Fig. 3), explaining 20.8%, 23.3% and 23.6% of the variability, respectively. Both QTLs on LG06 displayed dominance and female additive effects, whereas the LG08 QTL was characterized by a male additive effect. For 2008, a single QTL was mapped on LG08 and explained 20.9% of the variability. As for the previous year, this QTL was characterized by a male additive effect, as well as a dominance effect. For FB09, two QTLs were identified on LG08 and LG12, and explained 8.7% and 10.5% of the variability, respectively. The global model estimation showed a significant effect for both QTLs, with no interaction, and explained 17.9% of the variability. Both QTLs were characterized by a female additive effect.

X3263 × ‘Belrène’ population GP  Three QTLs were detected for G BLUP of the GP variable on LG01, LG03 and LG09. These QTLs explained 5.2%, 10.9% and 32.1% of the variability, respectively. All three QTLs were characterized by a female additive effect, as well as a dominance effect for the LG09 QTL. Together, the three QTLs explained 23% of the variability (Table 6).

Table 6.   Parameters associated with the quantitative trait loci (QTLs) detected in the apple (Malus × domestica) population X3263 × ‘Belrène’ by multiple QTL mapping (MQM) for the phenological traits described as green point (GP), vegetative budbreak (VB) and floral budbreak (FB)
TraitLGPositionLocusLODaR2bAllelic effectscGlobal R2dAmeAffDg
  1. Each trait is followed by the year (2005–2010) in which the QTLs were detected. BLUPs (best linear unbiased predictors) indicate QTLs detected for the genetic effect.

  2. aMaximum logarithm of the odds (LOD) score value of the QTL with the considered threshold in parentheses.

  3. bPercentage of the phenotypic variation explained by the QTL.

  4. cSignificance of the allelic combinations estimated by the global model based on genotypic information from the locus used as co-factor.

  5. dVariation explained by all QTLs estimated by the global model.

  6. eMale additive effect computed as [(μac + μbc) − (μad + μbd)]/4, where μab, μad, μbc and μbd are the estimated phenotypic means associated with each of the four possible genotypic classes ab, ac, ad and bd, deriving from an <ab×cd> cross.

  7. fFemale additive effect computed as [(μac + μad) − (μbc + μbd)]/4, where μab, μad, μbc and μbd are the estimated phenotypic means associated with each of the four possible genotypic classes ab, ac, ad and bd, deriving from an <ab×cd> cross.

  8. gDominance effect computed as [(μac + μbd) − (μad + μbc)]/4, where μab, μad, μbc and μbd are the estimated phenotypic means associated with each of the four possible genotypic classes ab, ac, ad and bd, deriving from an <ab×cd> cross.

  9. *, P = 0.01; **, P = 0.001; ***, P = 0.

GP08114.0GD_SNP003455.17 (4.2)0.051lm***0.249−0.6943.6910.749
331.5GD_SNP001948.17 (4.2)0.102No effect 1.1754.3810.609
90.4GD_SNP0118921.77 (4.2)0.375np*** 1.9796.7736.604
GP09114.0GD_SNP003454.83 (4.2)0.086lm***0.129−0.0492.1531.862
90.4GD_SNP011895.93 (4.2)0.121np*** 0.7792.3551.571
GP10331.5GD_SNP001944.72 (4.8)0.097No effect 0.5683.617−0.010
90.4GD_SNP011897.68 (4.8)0.169  0.9782.2353.504
GP_BLUPs114.0GD_SNP003454.3 (4.3)0.052lm***0.230−0.1651.1110.048
331.5GD_SNP001947.75 (4.3)0.109kk* 0.2801.375−0.027
90.4GD_SNP0118917.6 (4.3)0.321np*** 0.5171.6121.859
VB08140.3GD_SNP000667.59 (4.9)0.138lm***0.226−0.6968.836−4.523
34.1GD_SNP005886.77 (4.9)0.096np* 0.4605.5152.157
90.4GD_SNP011896.46 (4.9)0.112np*** 1.092−0.0125.331
150.0GD_SNP005506.05 (4.9)0.065kk***, hk** −1.131−2.970−0.643
VB09NO QTL     0.0000.0000.000
VB101084.3COL_XB3.25 (4.1)0.055  1.249−0.5390.211
VB_BLUPs138.7GD_SNP000875.34 (3.9)0.149lm*0.2232.1084.996−2.957
90.4GD_SNP011894.47 (3.9)0.083np 0.233−0.1351.202
1082.1GD_SNP003605.23 (3.9)0.067kk* 0.305−0.6910.384
159.7NZ02b01_XB5.95 (3.9)0.072kk***, hk** −0.305−0.783−0.133
FB08118.0GD_SNP000745.89 (5.5)0.102np*0.210−0.0773.5000.176
90.4GD_SNP011897.45 (5.5)0.16np** 0.8592.8852.575
1259.9GD_SNP003764.22 (5.5)0.069kk***, hk** 0.3892.9600.284
  SNP01189:SNP00376  np:hk *    
  SNP01189:SNP00376  np:kk *    
  SNP00074:SNP00376  np:hk *    
  SNP00074:SNP00376  np:kk *    
  SNP01189:SNP00074:SNP00376np:np:hk *    
  SNP01189:SNP00074:SNP00376np:np:kk *    
FB0990.4GD_SNP0118913.18 (4.1)0.291  0.9871.2073.298
1713.4GD_SNP015844.15 (4.1)0.077No effect −0.1792.686−1.169
FB10118.0GD_SNP000748.47 (4.7)0.188  −0.1054.467−0.833
FB_BLUPs17.0Hi02c07_XB5.83 (4.3)0.109bc***, bd**0.158−0.0110.3510.059
90.4GD_SNP011894.32 (4.3)0.093np*** 0.0650.0320.266

All three QTLs were recurrently detected over the three years investigated (Table 6). The LG01 QTL was identified in 2008 and 2009 (Fig. 4), and explained 5.1% and 8.6% of the variability, respectively. The LG03 QTL was identified in 2008 and 2010, and explained 10.2% and 9.7% of the variability, respectively. Finally, the LG09 QTL was identified for all three years, and explained between 12.1% and 37.5% of the variability. As for G BLUP, QTLs were characterized by a female additive effect. Together, the QTLs explained between 12.9% (2009) and 24.9% (2008) of the variation.

Figure 4.

Genomic positions of quantitative trait loci (QTLs) detected on the linkage groups (LGs) of the apple (Malus × domestica) X3263 × ‘Belrène’ integrated map by multiple QTL mapping (MQM) for the variables green point (GP), vegetative budbreak (VB) and floral budbreak (FB). QTLs are represented on the right side of LGs by boxes extended by lines representing the logarithm of the odds (LOD)-1 and LOD-2 confidence intervals. The numbering of LGs is according to Maliepaard et al. (1998).

Vegetative budbreak  Four QTLs were detected for G BLUP of the VB variable. These QTLs were identified on LG01, LG09, LG10 and LG15, and explained 14.9%, 8.3%, 6.7% and 7.2% of the variability, respectively. The global model built explained 22.3% of the variability. QTLs detected on LG01, LG10 and LG15 resulted from a female additive effect, whereas the LG09 QTL resulted primarily from a dominance effect.

Three of these QTLs were identified for VB08 on LG01, LG09 and LG15 (Fig. 4), and explained 13.8%, 11.2% and 6.5% of the variation, respectively (Table 6). One additional QTL was identified for VB08 on LG03 and explained 9.6% of the variability. Together, the QTLs explained 22.6% of the variability. The QTLs mapped on LG01, LG03 and LG15 resulted from a female additive effect, whereas the QTL located on LG09 resulted from a dominance effect. No QTL was detected for VB09. In 2010, one QTL explaining 5.5% of the variability was detected on LG10.

Floral budbreak  Two QTLs were detected for G BLUP of the FB variable on LG01 and LG09, which explained 10.9% and 9.3% of the variability, respectively. The LG01 QTL resulted from a female additive effect, whereas the LG09 QTL resulted from a dominance effect. The global model built for G BLUP of this variable explained 15.8% of the variability.

Both QTLs were recurrently detected over the 3 years investigated (Table 6). The LG01 QTL was identified in 2008 and 2010, and explained 10.2% and 18.8% of the variability, respectively. The LG09 QTL was identified in 2008 and 2009, and explained 16% and 29.1% of the variability, respectively. Two additional QTLs were detected for FB on LG12 (2008) and LG17 (2009), and explained 6.9% and 7.7% of the variability, respectively. In 2008, significant interactions occurred between QTLs located on LG09 and LG12, LG01 and LG12, and among all three QTLs (Table 6). The global model built for QTLs detected in 2009 did not permit the confirmation of the effect of the LG17 QTL, which may be a result of the low percentage of variability (7.7%) explained.

CG selection by in silico mapping

Of the two major QTLs detected on LG08 and LG09, we chose to explore the function of predicted genes underlying the LG09 genomic portion based on the importance of the percentage of explanation of this QTL and the relatively small size of its CI. The region spanned the length of the largest QTL interval (15.9 cM) and was located between the top of chromosome 9 (above GD_SNP01891) and GD142. The interval investigated represented 4 039 768 bp and comprised 983 predicted genes.

BLAST results permitted the annotation of 710 genes (72%), whereas 273 (28%) had no match in the databases. A similar proportion was annotated for the set of genes selected randomly. For both sets of genes, the GO terms were distributed into nine main GO categories (Fig. 5), with cellular process and metabolic process being the main categories of biological process. A χ2 test revealed that five of these GO categories were over-represented in the LG09 QTL: response to stimulus, biological regulation, signaling, death and cell cycle (data not shown).

Figure 5.

Gene ontology (GO) distribution charts by second-level GO terms. (a) The distribution of GO terms of predicted proteins located within the LG09 quantitative trait locus (QTL) confidence interval. (b) The distribution of GO terms of an equivalent number of predicted proteins picked randomly within the apple genome. GO terms indicated in bold represent genes over-represented in the LG09 QTL interval according to χ2 test results.

Throughout our investigation of the genes present within the over-represented GO categories, we identified 74 predicted proteins with potential involvement in the cell cycle and its regulation, including 20 located within 900 kb of the QTL peak (Table 7).

Table 7.   Predicted proteins located within 900 kb of the marker GD_SNP01189 on LG09 and displaying sequence similarity to genes involved in the plant cell cycle
Protein numberSequence similarityNumber of gene copiesE-valueIn silico position (bp)
MDP0000914555Cyclin-A318e-14chr9:606059..606724
MDP0000268652Auxin signaling F-BOX 320chr9: from 41177 to 446464
MDP0000375039Cytokinin-N-glucosyltransferase 213e-122chr9:1179208..1182844
MDP0000303239Myb-related protein Pp212e-56chr9:1120204..1127872
MDP0000241327E3 ubiquitin-protein ligase142e-36chr9: from 244643 to 1509459
MDP0000167621Phytosulfokine receptor 113e-39chr9:978792..981826

The closest CG to the LOD peak had sequence similarity with a cyclin-A3 sequenced in Oryza sativa. This gene was located 16 500 bp from GD_SNP01189. Further away from the LOD peak were two genes with high sequence similarity to an Arabidopsis thaliana auxin signaling F-BOX 3 gene. Other genes located close to the LOD peak included a phytosulfokine receptor 1 (E-value = 2e-39), a Myb-related protein Pp2 (E-value = 2e-56) and a cytokinin-N-glucosyltransferase 2 (E-value = 3e-122) (Table 7). Finally, 14 genes with high sequence similarity to E3 ubiquitin-protein ligase were identified in several clusters, with the main cluster including 10 genes within 400 kb and comprising the region of the highest LOD peak.

Additional proteins located further away from the LOD peak and within the QTL CI were also identified. These proteins had high functional similarity with cullin-3A (×1), cyclin-dependent kinases (×1), cytokinin-O-glucosyltransferase (×14, in two clusters), MYB transcription factors (×12), E3 ubiquitin-protein ligase (×10), aquaporin (×4), actin (×3), expansin (×4) and phytosulfokine (×5).

Discussion

Heritability and significance of the effects

The present study highlighted that all three variables investigated were characterized by a significant genotypic effect and a high heritability value. Heritability estimates have often been shown to be specific to the population and environment analyzed (Souza et al., 1998). However, in a previous study by van Dyk et al. (2010), the heritability value for the trait initial VB (IVB) was estimated to be between 0.62 and 0.92. In our study, similar estimates were found for VB, which suggest that, despite differences in genotype and the environment in which trees were grown, the date of budbreak remains a highly heritable trait, as suggested previously by Anderson & Seeley (1993).

Phenotypic correlations, QTL detection and co-localization among traits and populations

Both populations showed similar patterns of data distribution, comparable correlation coefficients between consecutive years and among traits. However, correlation coefficients between traits decreased notably between consecutive years (Table 4), which can be attributed to either tree age, climatic conditions or a combination of both. In the present state, our experimental design does not permit us to discriminate between these effects.

Despite the similarities between the two populations, QTL analysis revealed that the genetic determinism regulating GP, VB and FB was different. No common QTL was identified between the two populations.

In the STK × GS population, QTL analysis permitted the identification of one major genomic region influencing two of the three variables, with no stable QTL being identified for VB. This major QTL region, located on the proximal part of LG08, influenced GP and FB independently of the year studied. It is interesting to note that the moderate to high correlation values obtained between GP and FB in 2007–2009 coincided with the years for which the LG08 QTL co-located. For 2010, the correlation between the two traits decreased, and no QTL was detected for FB10. The absence of correlation and common QTL in 2010 might be explained by the atypical weather conditions observed during spring, and may be the consequence of the cold spring temperatures observed for this year (Fig. 2). The major QTL on LG08 confirms the preliminary results obtained by Segura et al. (2007). This QTL co-locates with QTLs for the biennial-bearing index, total number of inflorescences (B. Guitton et al., unpublished) and QTLs controlling leaf ecophysiological traits (Regnard et al., 2009) and hydraulic conductance in the xylem (Lauri et al., 2011). One hypothesis explaining the co-localization among these variables could be an increased capacity of the plant to transport water, carbohydrates and sugar to the growing organs.

In the X3263 × ‘Belrène’ population, we identified three QTLs independent of the environmental effect for GP G BLUP. The first, located on the proximal part of LG09, was consistently identified over the 3 years investigated. The other two QTLs identified were located on LG01 and LG03.

The correlation between VB and the other two variables in the X3263 × ‘Belrène’ population was dependent on the year studied. The highest correlation was obtained in 2008, whereas it was low in 2009 and 2010 (Table 4). The high correlation in 2008 might explain the common QTL identified on LG09 for all three variables for this year. Furthermore, a QTL on LG09 was also identified for the G BLUP VB variable, indicating that this QTL was independent of environmental effects. The importance of the LG09 QTL was confirmed by the analysis of the FB data. In a study performed by van Dyk et al. (2010), a major QTL located in a similar genomic region was found to influence the date of IVB. The phenotypic variation was shown to be associated with alleles inherited from the parent ‘Anna’. Co-localization between QTLs was rendered possible by the use of common markers (GD142 and NZmsCN943946). The presence of a common QTL between two unrelated populations suggests that similar sets of genes might be involved in the control of budbreak dates for both populations. However, the allele sizes amplified by NZmsCN943946 were different, which might be caused by the high linkage disequilibrium expected in tree species (Plomion & Durel, 1996). Unlike in van Dyk et al. (2010), the influence of the LG09 QTL did not increase during the consecutive years of phenological trait assessment in our conditions in which CR was fulfilled, suggesting that results from our QTL analysis were more dependent on climatic conditions than tree age. In addition, among the seven genomic regions identified by Conner et al. (1998) as influencing the date of budbreak, one region later identified on LG09 (Kenis & Keulemans, 2007) was found to influence this trait.

In both populations, the major genomic regions identified by QTL analysis were previously shown to be nonhomologous (Celton et al., 2009; Velasco et al., 2010), indicating that different genes or regulatory pathways might be involved. Experimental results in tree crops have demonstrated the inconsistency of QTL expression across populations. The identification of a common QTL among populations derived from an unrelated genetic background, and grown under different environmental conditions, is an unusual finding. Such QTLs, defined against a wide genetic background, could be more useful in marker-assisted breeding (MAB) than QTLs defined in a specific background (Plomion & Durel, 1996). The identification of a second major and stable QTL on LG08 could allow breeders to engage in a strategy of pyramiding favorable QTLs and beneficial alleles into a unique hybrid seedling better suited to future warmer climatic conditions.

Interactions between QTLs and climatic conditions

All models built for GP, VB and FB integrated G, Y and G × Y effects. Decomposition of the Y effect into an estimated CR and HR period allowed us to demonstrate a strong effect of both periods on FB.

For almond (Prunus dulcis) growing in cold climatic conditions, HR was described as more important than CR for the regulation of blooming time because of the early completion of chilling (Alonso et al., 2005). In warmer areas, almond flowering time was more influenced by chilling than heat (Egea et al., 2003). Similar results were found for ornamental peaches (Prunus persica), for which variation in flowering time over the years was mainly a consequence of CR, with HR contributing a smaller effect (Pawasut et al., 2004).

Our results suggest that, in our climatic conditions, the CR period has a stronger effect than the HR period on the variation of the time of floral budbreak across genotypes of a population. However, the relative importance of these two factors may also be dependent on the genetic background as a higher heat effect was found in X3263 × ‘Belrène’ than in STK × GS.

The next section of this discussion aims to decipher the interactions between QTLs, HR and CR for FB over the course of the 4 yr studied.

As described earlier, the distribution of FB dates in 2010 in the STK × GS population was very narrow compared with other years. This reduction in data distribution might explain the absence of QTL detected for this year. Furthermore, the results indicate that the effect of the LG08 QTL on FB dates over the years decreased from 23% in 2007 to 9% in 2009, concomitantly with an increase in the length of the HR period. This suggests that the warmer the climate (in winter and spring), the more important is the influence of LG08 on FB.

A similar conclusion can be drawn for the LG01 and LG09 QTLs in the X3263 × ‘Belrène’ population. When considering the average spring temperature of 2008 and 2009 (70–80-d HR period), a decrease in the CR period led to an increase in the percentage of explanation of LG09 QTL for FB. In addition, the longer the HR period (following moderate winter conditions, i.e. 90-d CR), the more important was the influence of the LG01 QTL on the FB date. This suggests that the LG01 QTL influences FB dates in cold spring conditions, whereas the LG09 QTL takes over from LG01 in warmer spring conditions.

Finally, as most of the variation was a result of female additive effects, we conclude that alleles derived from X3263 can regulate FB dates in both cold and warm winter conditions by switching regulation from LG01 to LG09, whereas alleles derived from ‘Starkrimson’ influence FB dates through the LG08 QTL only in mild winter conditions. For both populations, more years and climatic conditions should be studied to confirm the effect of the respective QTLs on FB dates.

In silico CG selection

The conservative approach required to determine GO annotations (Dwight et al., 2002) resulted in almost 28% of all the genes falling into the unknown function category. Despite this, we identified five GO categories differentially present in the LG09 region, including one identified as cell cycle. However, the QTL effect could be a result of either one or a multitude of genes within the CI of the QTL, from any of the GO term categories. Thus, the identification of particular categories of genes over-represented within this interval does not allow us to conclude that the genes responsible for the phenotypic variation are present within one of these GO categories.

From this first screening, we defined a subset of genes whose functions are putatively involved in the cell cycle. Most of the 74 genes thus identified were involved in the two major pathways regulating G1/S and G2/M transitions (Fig. 1). The closest CG to the LOD peak had sequence similarity with an O. sativa cyclin-A3 (GO: cell cycle), which is part of the CycA plant cyclins (Renaudin et al., 1996). Cyclins are positive regulators that bind to the cyclin-dependent kinases, and are essentially expressed in the DNA replication phase (S phase) (Hemerly et al., 1992). A variation in the expression of this gene could have consequences on the rate of DNA synthesis, and thus cell division. Further results indicated the presence of 12 Myb-related (GO: response to stimulus) proteins within the CI. The closest to the LOD peak was identified as Pp2. Leech et al. (1993) demonstrated that the maximum level of transcription of this gene correlated with the time of maximum mitotic index in moss. This protein, which probably functions as a transcriptional activator, could have a major effect on the cell mitosis rate during bud growth resumption and its regulation, even though the GO term may refer to a different category.

Various other genes with homology to known proteins were also identified within the CI of the QTL, such as MDP0000252231, identified as cullin-3A (GO term: cell cycle), whose loss-of-function mutant in A. thaliana was shown to induce a late-flowering phenotype (Dieterle et al., 2005). Other genes include multiple copies of actin (GO: cellular process), expansin (GO: cellular component organization) and phytosulfokine (GO: metabolic process), which are all necessary for plant cell mitosis.

In our study, we also identified genes associated with phytohormones. An auxin signaling F-BOX 3 (AFB3) (GO: reproduction/signaling/developmental process) gene was identified close to the LOD peak. AFB genes encode related F-box proteins that assemble into SCF complexes that are required for auxin-dependent degradation of auxin/indole-3-acetic acid proteins (Dharmasiri et al., 2005). Mutations in the AFB3 gene could impair the auxin response in the plant and influence the cell cycle during the mitotic phase (Fig. 1). Underlying the QTL, a total of 15 genes with homology to cytokinin-O- and cytokinin-N-glucosyltransferase activity were identified (no homology identified using BLAST2GO). The closest was located 600 kb away from the LOD peak. The remaining 14 were located in two clusters along this chromosomal region. In A. thaliana, cyclin-dependent kinases were reported to induce cyclin D3, necessary for entry into the DNA replication phase (Fig. 1). Variation in the level of cyclin-dependent kinases could impair the capacity of cells to enter this phase.

The ubiquitin-proteasome pathway is also recognized to play a crucial role in the cell cycle. This pathway targets specific proteins for programmed destruction in response to internal or external stimuli (Hershko & Ciechanover, 1998). In A. thaliana, E3 ubiquitin protein ligase forms a complex with SCF and plays a major role during the cell mitosis phase (Fig. 1). We identified 24 E3 ubiquitin protein ligase (GO: metabolic process) copies in several clusters.

In recent years, it has become apparent that copy number variants (CNVs) are common within the human genome (Stranger et al., 2007), and can positively or negatively correlate with gene expression levels. The presence of CNV of genes involved in the cell cycle within the CI of the QTL could therefore point to a possible regulation of cell cycle capacity by these CNVs, although the study of CNV gene expression could prove to be challenging. Indeed, our current knowledge of the apple genome does not permit us to discriminate between nonfunctional pseudogenes and functional genes.

In the present study, QTL analysis permitted the identification of major and population-specific genomic regions influencing the dates of GP, VB and FB. The genetic dissection of phenological characters was performed over 3–4 years for each population, and the results suggested strong interactions between CR, HR and genetic effects. Variations in the influence of QTLs over the phenological variables suggested that trees were capable of adapting to climatic conditions by regulating the expression of genes involved in bud growth and located in different genomic regions.

Acknowledgements

We thank J. J. Kelner and P. E. Lauri for fruitful discussions, and F. Laurens (INRA Angers) for providing the vegetative material of the progeny X3263 × ‘Belrène’. We also thank B. Guitton, who kindly provided CG markers to improve the STK × GS map, and G. Droc, who helped with BLAST2GO. J.M.C. was funded by Montpellier SupAgro Agronomic School.

Ancillary