Grapevine bud development is modulated by environmental factors such as temperature and day length 25 . In our experimental conditions, Tempranillo cv. latent buds are formed during the first growing season in the young sprouting stems between April and May (APR-MAY) and experience active developmental processes involved in the set up of the vegetative and reproductive growth until the end of the summer of the following year (Figure 1) 32 . Flowering induction takes place within latent buds around the middle of June (JUN) and inflorescence primordia differentiate from lateral meristems developed by the shoot apex. Inflorescence meristems proliferate to generate inflorescence branch meristems in complex inflorescence primordia along July (JUL) and August (AUG) 40 . In our growing conditions, Tempranillo buds are endodormant in the second half of September (SEP) and endodormancy is released by the end of November (NOV), although buds remain in an ecodormant stage from DEC to MAR 41 . Once winter is over, ecodormancy is released and inflorescence branch meristems proliferate to produce flower meristems in April (APR) swelling buds that initiate the second growing season 40 .

High throughput transcriptional analysis was performed along bud development on bud samples collected at 8 different time points during the annual cycle (see Methods). Principal Components Analysis (PCA) was performed on the whole expression dataset (Additional file 1) to verify correlation among different biological replicates and to identify main sources of gene expression variation. The results of the PCA plot showed consistency across biological replicates, as shown in Figure 2. The first two principal components (PC1 and PC2) explained 77.5 percent of the total variability in gene expression (62.2 percent and 15.3 percent respectively). PC1 seems to represent the time course evolution of bud developmental stages and appears to be reset to the original status with bud swelling, since APR bud samples of the second season were neighboring to MAY bud samples of the first season (in the same quadrant), revealing a high transcriptome similarity between those two stages. PC2 highlights major transcriptome differences between JUN, JUL and SEP samples and the remaining time points.

To investigate the biological basis of the principal components, transcripts with the highest contribution to each component in the analysis were identified according to the absolute value of their component score (CS) for PC1 and PC2 (Additional files 2 and 3). Figure 3A shows the expression profiles of transcripts mostly contributing to PC1. Transcripts with negative CS values (547 transcripts, blue color) are up-regulated in non-dormant buds and show declining expression during endodormancy and the lowest expression during ecodormancy. Transcripts with positive CS values (204 transcripts, orange color) follow the opposite trend (Figure 3A). Functional enrichment analyses indicated that the transcripts up-regulated in non-dormant buds were characteristic of actively proliferating and growing cells (Figure 3B). Among them, transcripts related with cell division, cell growth and differentiation (cell-cycle regulation, microtubule-driven movement, chromatin assembly, cell growth, peptidase-mediated proteolysis and cell wall metabolism) and those related with primary and secondary metabolism (photosynthesis, fatty acid biosynthesis, flavonoid biosynthesis and aromatic compounds glycosidation) showed high CS value.

On the other hand, the most significant functional categories contributing to PC1 and enriched in dormant buds (orange color) were those related to stress responses (stilbenoid biosynthesis, "HSP-mediated protein folding", temperature stress response, as well as CCAAT transcription factor family) 42 . These functions could be related to bud responses to dehydration and temperature changes that take place together with dormancy, as previously reported in Populus 24 . Furthermore, the observed up-regulation of genes involved in ABA catabolism could be related to the decay of this hormone previous to dormancy release 5 . Finally, an increase in the expression of starch catabolism genes together with a down-regulation of genes encoding photosynthetic proteins, also observed in Populus 24 , is in agreement with the physiological state of dormant buds.

Regarding PC2, the enriched functional categories contributing to its negative values were those related to stress responses characteristic of JUN, JUL and SEP samples, whereas the functional categories associated with its positive values were those related to cell proliferation that show higher expression in non-dormant buds (Additional file 4). Therefore, PC2 could represent the transcriptional effects of the stress experienced by buds from JUN to SEP with respect to the remaining stages.

In summary, our results suggest that active growth, dormancy and stress responses are major contributors to the gene expression variability observed along the bud annual cycle. Processes characteristic of actively proliferating and growing cells are up-regulated in non-dormant buds and decline during bud dormancy together with the up-regulation of stress response pathways.

In order to identify the developmental stages representing major transcriptome changes during the annual cycle of the bud, we performed a pair wise differential expression analysis between consecutive time points. The number of differentially expressed genes varied strikingly among sample pair comparisons (Figure 4). Major changes were observed between JUL and SEP (involving 3139 transcripts), SEP and NOV (involving 2002 transcripts) and MAR and APR (involving 5658 transcripts). Interestingly, these transitions could be associated, respectively, to the proposed timing for para/endodormancy, endo/ecodormancy and ecodormancy/bud break transitions 28 41 . These results supported the conclusions of the PCA experiment, suggesting that the transitions between bud dormancy and active growth explains most of the variation in bud gene expression profiles.

In order to identify biological functions involved in the three major bud transcriptional changes we performed studies of functional categories enrichment (Figure 5). The para/endodormancy transition (JUL to SEP) (Figure 5A) was characterized by a major reduction in enriched functional categories that mostly contributed to PC1 in Figure 3. Among them, categories related to cell proliferation (including regulation of cell cycle, chromatin assembly and microtubule organization and biogenesis) and cell growth and death, were down regulated from JUL to SEP. These results are consistent with the shutdown of those processes during bud dormancy that has been described in other systems 24 43 . Cell wall organization and biogenesis, also linked to cellular processes, was significantly reduced with endodormancy. Modification of the cell wall, mainly xyloglucans metabolism and cell wall proteins, was an enriched category in JUL samples but not in SEP. However, cell wall biosynthesis (mainly including cellulose biosynthesis transcripts) appeared as a significant category in SEP. Metabolism of carbohydrates was in some aspects related to cell wall metabolism, thus 1,3-β-glucan catabolism predominated in JUL while in SEP carbohydrate metabolism was mainly represented by starch and sucrose metabolism. These results could be related to the required sealing of plasmodesmata with callose to lower their size exclusion limit, a process dependent on 1,3-β-glucansynthase and 1,3-β-glucanase activities. Additionally, cell wall composition must be modified to reduce water and molecules movement among cells during this stage 44 .

Representation of photosynthesis-related functional categories also decreased during para/endodormancy transition, in agreement with results reported in Populus 24 . Lipid metabolism (including fatty acid biosynthesis, glycerolipds metabolism and oxylipins) was also under-represented in this transition. This change could be related to the formation of lipid bodies (LBs) which store triacylglicerols and improve freezing tolerance previous to endodormancy, as has been reported in other systems 44 . However, as far as we know, no cytological evidence on the formation of LBs exists in grapevine buds.

Other significant functional categories that were under-represented from JUL to SEP were those related to stress responses (mainly abiotic and temperature), chaperone-mediated protein folding and aquaporins (mainly TIP). In Populus, genes responsible for adaptation to dehydration and low temperatures have been shown to be expressed in response to SD even in the absence of those stresses 24 . In leafy spurge, genes responsive to cold stress were also up-regulated in fall and winter 43 . Within the signaling pathways functional category, 129 transcripts were down-regulated between JUL and SEP. Among them, salicylic acid-mediated signaling was significantly enriched, what could be related to stress responses as well as elicitation of phytoallexin biosynthesis 45 .

The transcription factor functional category was also significantly enriched among transcripts down-regulated in this comparison. Among them, a significant enrichment was also proven for the ERF subfamily (9 transcripts), within the AP2-like 46 transcription factor family, most of them related to ethylene regulated responses. In Populus, the temporal expression of some ERF-like transcripts at the beginning of SD photoperiods has suggested a role for ethylene in the regulation of dormancy 24 . In fact, Populus, leafy spurge and potato have a transient peak in ethylene or ethylene perception associated with endodormancy induction 5 .

Only a few functional categories were significantly enriched among transcripts showing up-regulation from JUL to SEP. Apart from those concerning cell wall organization and biogenesis and carbohydrate metabolism that were previously described, functional categories related to metabolism of nucleotides and nucleic acids and protein metabolism were also significant. Among them, enrichment of RNA metabolism and translation initiation could reveal the existence of mechanisms relying on stored mRNAs transcripts ready to be translated, as described in dry angiosperm seeds 47 .

The endo/ecodormancy transition (SEP to NOV) (Figure 5B) was characterized by a further decline in transcripts participating in cellular processes as well as primary and secondary metabolism functional categories, similar to what was observed in the para/endodormancy transition. Functional categories related to cell wall metabolism and biogenesis (mainly based on biosynthesis of cellulose, catabolism of pectins and modification of pectins and xyloglucans) were still relevant in SEP and further down-regulated in NOV. Moreover, carbohydrate metabolism-related transcripts that were down-regulated in SEP versus NOV corresponded now to oligosaccharides metabolism and glucans catabolism. A parallel situation has been reported in Vitis riparia during the chilling period required for endodormancy release 34 . Stress responses were also under-represented from SEP to NOV, especially categories related to oxidative stress. An activation of the oxidative stress response machinery preceding endodormancy release would be in agreement with previous reports showing that oxidative stress affecting mitochondrial function could participate in endodormancy release in grapevine 36 37 .

On the other hand, the SEP to NOV transition was marked by an enrichment of ABA catabolism, in agreement with the role of ABA in endodormancy and its decay during endo/ecodormancy transition 5 . ABA levels correlate with bud dormancy in several species and decay throughout the transition to ecodormancy. Both in Populus and leafy spurge, ABA content peaked after few weeks of SD and decayed later 24 43 . Moreover, an ABA-related transcript has also been reported to be down-regulated during the chilling period required for endodormancy release in grapevine 34 . Buds of NOV also showed an enrichment of stilbenoid biosynthesis that in grapevine usually responds to biotic and abiotic elicitors 48 . Finally, enrichment of nicotinate and nicotinamide metabolism in NOV as well as nucleotide and amino acid transport might suggest the initiation of certain metabolic activity paralleling the endo/ecodormancy transition. Interestingly, the ERF subfamily of transcription factors that was significantly enriched in JUL versus SEP was also found in NOV versus SEP. Moreover, five of these transcripts were common in JUL and NOV, suggesting common functions before para/endodormancy and after endo/ecodormancy transitions.

The ecodormancy/budbreak transition (MAR to APR) (Figure 5C) was characterized by down-regulation of specific functional categories involved in starch and sugar catabolism or signaling and stress responses mainly related to previous periods. Similar underrepresentations were also observed in transcription factor families known to be involved in stress adaptive responses such as HSF 49 , NAC 50 and WRKY 51 families, as has been reported in other species 19 43 . In contrast, bud break in APR was characterized by the presence of most functional categories that were enriched in JUL before bud para/endodormancy transition. Those categories were related to cellular processes required for cell proliferation (chromatin assembly, microtubule organization and biogenesis, cell cycle regulation) and cell growth (cell growth and death, cell wall organization and biogenesis, photosynthesis and primary and secondary metabolism). Up-regulation of cell wall organization and biogenesis and carbohydrate metabolism categories (mono, oligo and polysaccharides and more specifically glucans and 1,3-β-glucan catabolism), could also be related to restoring cell wall properties and cell communication throughout callose hydrolysis at plasmodesmata. Induction of 1,3-β-glucanase after the application of dormancy-release agents has also been reported in grapevine 38 . Secondary metabolism categories included enrichment of flavonoid biosynthesis (mainly anthocyanins) as well as aromatic compound and shikimate metabolism as probable flavonoid precursors. Interestingly, the significantly enriched hormone signaling functional categories were those related to auxin and salicylic acid, that likely have a role in cell proliferation and expansion 52 as well as in the response to biotic stress 45 .

Flowering induction in grapevine takes place in latent buds at the beginning of the summer whereas flower meristem differentiation and flower development take place in the second growing season 10 30 31 32 33 . To identify putative genes involved in flowering induction and flower development, we examined in detail the expression profiles of reproductive development key genes such as VFL, the MIKC-type MADS-box, the SPL and the FT-TFL1 gene families. Hierarchical clustering based on bud expression values of these transcripts along the annual cycle were represented in Figure 6. Consistently with the main bud transcriptional profiles described in the previous section, expression analysis of these transcripts identified three major distinct clusters. The first cluster grouped transcripts up-regulated in non-dormant buds and down-regulated during dormancy. The second cluster contained transcripts with highest expression level during bud break. The third cluster grouped transcripts up-regulated in dormant buds with an opposite expression to those of cluster 1.

Cluster 1 showed transcripts expressed in latent buds when inflorescence primordia are initiated and proliferate. A well characterized gene within this cluster is the MADS-box gene VvSOC1.1 39 , which expression pattern suggests that it could play a crucial role in flowering induction, as does SOC1, its Arabidopsis homolog 53 . Other MADS-box genes in cluster 1 were members of the FLC (VvFLC1) and SVP subfamilies (VvSVP1, VvSVP3 and VvSVP5) 39 . VvFLC1 showed high expression during the first season, decreased during dormancy and increased again during the second growing season (Additional file 5). Expression patterns of the three grapevine SVP homologs were distinct than that observed for SVP in Arabidopsis 54 . Their expression levels were high at flowering induction, reduced during dormancy and increased again at flower meristem formation. No significant expression changes of SVP homologs were observed along grapevine bud dormancy, what would not justify their role in that process, as has been reported in other species 18 19 20 21 22 23 24 . Cluster 1 also included VvMFT1 31 , a member of the FT-TFL1 family of transcriptional regulators 55 56 , with highest expression during the first season. In addition, detection within this cluster of APETALA3.2, PISTILLATA and AGAMOUS homologs suggests that these genes, involved in the specification of flower organs identity, could already be expressed in the inflorescence meristems of first season buds in preparation for the flower organ specification that takes place during the second growing season. Finally, cluster 1 included many homologs of A. thaliana SPL genes (SPL2, SPL3, SPL4, SPL5, SPL8 and SPL9) what will be discussed later when considering the expression patterns observed within this gene family.

Cluster 2 grouped transcripts showing their highest expression during bud break in the second growing season and likely associated with the events of flower meristems and flower organs differentiation. Consistently with the developmental processes taking place in bud break, cluster 2 contained both VvFT and VFL, which have homologs in Arabidopsis that are required for flower induction and flower meristem specification. VvFT was over-expressed during flowering induction, decreased during bud dormancy and increased again during the second season, which is compatible with the roles proposed for its homologs in Populus and other woody plants 7 11 13 43 . Although the putative TFL1 homolog (VvTFL1A) was not present in the Grapegen GeneChip®, our previous analysis of this gene family 31 showed that it is down-regulated during dormancy and up-regulated with dormancy release. This expression pattern parallels what has been reported for TFL1 homologs in Populus 13 . VvSPL13-L and VvAG3 on one side and VvFUL-L, VvAGL6.2 and VvSEP4 on the other, showed a similar expression pattern and could also participate in the processes of flower meristem and flower organ specification.

Cluster 3 included transcripts showing their highest expression during dormancy. Interestingly, this cluster included two MADS-box genes, VvFLC2 and VvAGL15.1, which Arabidopsis homologs have been involved in flowering repression. In addition, it contains several SPL genes (VvSPL6-L, VvSPL12-L and VvSPL14-L). The remaining genes also showed some expression in non-dormant stages. The two FLC homologs found in grapevine (VvFLC1 and VvFLC2) 39 showed an opposite expression pattern in buds (Additional file 5). VvFLC1 behaved as PEP1, the Arabis alpina FLC homolog. PEP1 expression is reduced during the winter and increases when growth is resumed after the cold period 57 . Fluctuations in the FLC transcript levels in this perennial species would allow some meristems to undergo flowering transition while others maintain the vegetative growth of the plant. A similar role for VvFLC1 in maintaining vegetative growth of the young meristems that will give rise to the new latent buds could be proposed in grapevine. In contrast, the opposite expression pattern of VvFLC2, resembles expression reported for FLC-like genes in Poncirus 16 and leafy spurge 17 , where they could be involved in the regulation of dormancy. Similarly, a role as repressor of flower meristem initiation has been proposed for AGL15 and AGL18 in Arabidopsis 58 . Interestingly, their putative grapevine homolog, VvAGL15.1, has the same expression pattern as VvFLC2. The search of FLC gene homologs with opposite expression pattern in other polycarpic plant species could help to elucidate their role in these processes. In addition, two members of the FT/TFL1 gene family (VvTFLC1 and VvMFT2) 31 were also found in cluster 3, opening the possibility that their function could be related with the control of dormancy in grapevine. In agreement, high expression of a member of the FT/TFL1 family (PaFT4) in Norway spruce (Picea abies L.) correlates with growth cessation and bud set 59 , in contrast to that observed for the PtFT1 gene of Populus 11 . Indeed, comparative sequence and expression analysis of the FT/TFL1 family in gymnosperm and angiosperm species lead to speculate that the original function of this gene family could be related to the regulation of growth arrest and/or dormancy 60 .

Different members of the SPL family of transcription factors were found within the three clusters. This family of transcription factors is known to participate in the regulation of diverse plant developmental processes such as plant phase transition, flower and fruit development and plant architecture 61 62 63 64 . Ten of the 16 Arabipdopsis SPL genes are post-transcriptionally regulated by miR156, which incorporates endogenous age/development signals into vegetative phase transition and flowering 61 62 . This vegetative phase regulatory mechanism is also conserved in woody perennials 65 . Interestingly, cluster 1 contains transcripts homologous to SPL3, 4, 5 and 9, all belonging to the miR156/7-targeted SPL subfamily, which act as positive regulators of juvenile-to-adult phase change transition and flowering in Arabidopsis 63 64 66 and are regulated by SOC1 67 . Cluster 1 also included two SPL2 and one SPL8 homologs. Arabidopsis SPL2 is also miR156 targeted and seems to be involved in lateral organ development within the reproductive phase 68 . The miR156/7 non-targeted SPL8 gene is involved in pollen sac development 69 and required for male fertility 70 . Finally, several members of this gene family were found in Cluster 3 displaying an expression pattern more restricted to the dormancy period (VvSPL1-L, VvSPL6-L, VvSPL7-L, VvSPL12-L, VvSPL14-L and VvSPL13-L2). Most Arabidopsis counterparts of these genes belong to the miR156/7 non-targeted SPL subfamily (SPL1, 7, 12, 14 and 16) that comprises the larger proteins in the family 70 . Little is known about the functions of these putative transcriptional regulators with the exception of SPL14, which seems to regulate plant architecture and the length of vegetative phase. An Arabidopsis mutant with reduced SPL14 expression, had elongated petioles, serrated leaf margins and accelerated vegetative phase change 71 , suggesting that this gene could play a role as a negative regulator of phase transition and flowering 72 , having antagonistic function to other SPL proteins that promote vegetative phase change. Interestingly, both VvSPL14-L and VvSPL12-L showed an expression pattern very similar to VvFLC2 and VvAGL15.1 which could also suggest a role for these SPL genes in dormancy maintenance. Little is known about the role of SPL-like genes in woody species. Two genes (SPL-like 3 and 6) were detected during dormancy in Populus 24 with SPL6-like increasing and SPL3-like decreasing along dormancy. In addition, an SPL-2 homolog has also been found in grapevine which seems to be regulated by photoperiod 35 . Further studies will be needed to elucidate the possible role of the SPL gene family in bud dormancy.

Additional mechanisms involving transcriptional repressors could be required during the dormancy period to prevent premature flower meristems formation from the inflorescence meristems. The establishment of annual and perennial life has independently arisen several times in flowering plants 73 , so it is likely that mechanisms involved in the control of bud dormancy or repression of flower meristem formation have recruited different regulatory genes in different botanical families. Complementary experiments will be needed to assess the biological function of the members of these transcriptional regulators families in such processes.

Grapevine (Vitis vinifera L. cultivar Tempranillo) buds were obtained from an experimental vineyard at the Instituto Madrileño de Investigación y Desarrollo Rural, Agrario y Alimentario (IMIDRA, Alcalá de Henares, Madrid). Samples were collected from triplicate blocks in the same vineyard during two consecutive years. Plants and buds developmental stages were classified following the developmental series of Baggiolini (1952) 74 and modified E-L system 75 . Buds were collected at equivalent stem positions from the base and always at the same time of the day. Bud samples were frozen in liquid nitrogen and stored at -80°C before RNA extraction. Samples corresponding to May (MAY), June (JUN), July (JUL), September (SEP), November (NOV), January (JAN), March (MAR) and April (APR) buds were analyzed. Meteorological data were obtained from a station at Finca El Encin (IMIDRA, Alcalá de Henares).

Total RNA was extracted from frozen bud samples according to Reid et al., 2006 76 . RNA purification was performed using the RNeasy Mini Kit (QIAGEN) according to manufacturer's protocols. To remove DNA traces in RNA samples, DNase I digestion was carried out with the RNase-Free DNase Set (QIAGEN). RNA integrity and quantity were assessed by Agilent’s Bioanalyzer 2100. Microarray hybridizations were performed at the Genomics Unit of the National Centre for Biotechnology (CNB-CSIC, Madrid). Raw microarray data from the reported experiments are publicly available at the Plant Gene Expression Database (PlexDB) 77 and labelled as “VV36: Time course of grapevine bud development”.

Monitoring of bud transcriptional activity was performed in three biological replicates for each time point using Affymetrix Grapegen GeneChip®. Raw Affymetrix CEL files were imported to Robin software suite 78 to perform data normalization using the RMA method. Principal component analysis was performed using Acuity software 79 (Molecular Devices, LLC, CA, US). The score matrix was used to select probe-sets that best fit the first principal component (PC1) and those with PC1 scores greater than 7 or lower than -7 were chosen. Likewise, probe-sets that best fitted PC2 were those with component score greater than 3 or lower than -3.

Differential expression analyses were performed in Multi Experiment Viewer 80 using LIMMA, applying a 0.01 cut-off for P-value and log₂ fold ratio greater than 1 and lower than -1. P values were corrected using the Benjamini-Hochberg test.

To identify the biological functions over-represented within selected probe sets functional enrichment analyses were performed using FatiGO 81 (P-value< 0.05). Functional categories were based on manual annotation of the custom made GrapeGen GeneChip®, based on 12X v1 grape genome assembly, described in Grimplet et al., 2012 82 .

Key regulators of reproductive development were selected according to their functional annotation 82 . Expression values were extracted from the whole experiment normalized data matrix (averaged from the triplicates per sample and probe set). When more than one probe set matched a single gene transcript, only one was selected. Hierarchical clustering was performed using MultiExperiment Viewer 80 based on Pearson's correlation and using the complete linkage option.