Although many previous studies have considered the impact of increasing the source/sink ratio on grape berry composition, ours is the first investigation to look at the consequences of cluster thinning on global gene expression profiles, which is the basis of most of the observed physiological and biochemical changes.
Although there is only limited evidence for a strict relationship between yield and quality , in this study the effectiveness of thinning reflected the suboptimal leaf area to yield ratio of control vines (0.6 m2/kg). Increasing this ratio to 1.2 m2/kg by cluster thinning boosted the levels of sugars and anthocyanins and reduced acidity at harvest. Several authors have reported similar effects on sugars and anthocyanins [35, 36], but there is disagreement on the impact of thinning on titratable acidity, with reports suggesting that acidity is unaffected [7, 37], slightly increased  or decreased by yield reduction [2, 38]. In all cases, the effect on acidity seems to be related to the impact of cluster thinning on ripening, particularly the soluble solids content, because thinning reduces acidity only when the soluble solids concentration is strongly and positively affected ([38, 39] and this study).
The grouping of gene expression profiles by principal component analysis showed that C and CT berries could be distinguished as early as two weeks after treatment. Microarray data were then analyzed by two different approaches. Genes that were differentially expressed in C and CT berries were initially clustered on the basis of their expression profiles, and then we carried out a direct comparison of C and CT transcriptomes at each time point.
The first approach revealed such a large number of genes modulated during ripening that we applied a > 5 fold change cut-off threshold before assigning a gene to the modulated group. This resulted in 2466 transcripts that were considered to be modulated in CT berries and 567 that were modulated in C berries, including 447 that were common to both treatments. Approximately half of the common genes were more strongly modulated in CT berries, including several downregulated genes involved in photosynthesis, carbon utilization, carbohydrate metabolism, cell wall modifications and hormone metabolic processes that are already known to have a role in berry ripening [22, 40], and several upregulated genes involved in the normal ripening process (e.g. genes related to secondary metabolite biosynthesis). These data strongly suggest that the entire course of berry ripening is enhanced by the cluster thinning.
An interesting and unexpected result of the microarray analysis was the relatively large number of genes (2019) highly modulated only in CT berries. More than a half of these genes were never expressed in control berries at any ripening stage, and were activated uniquely by the cluster thinning treatment. This shows that thinning is able to trigger the transcription of genes that otherwise would not be activated in untreated berries and therefore that the effect of thinning goes beyond the simple enhancement or acceleration of the normal ripening process. This appears to affect many different metabolic and cellular processes because the CT highly modulated genes are distributed throughout all 18 functional categories we considered.
The CT high modulation of genes involved in carbohydrate metabolism supports the impact of cluster thinning on sugar accumulation commencing at the EV stage. This reflects the achievement of an optimal balance between leaf area and yield in thinned vines (1.2 m2/kg) compared to controls (0.6 m2/kg) as previously reported [2–7]. Interestingly, all CT highly modulated genes involved in sugar transport (including VvSUC11 and VvSUC27) were downregulated in CT berries but expressed at a constant level throughout ripening in controls. The downregulation of sucrose transporters in CT berries contrasts with the increase in sucrose transporter mRNA during berry development reported by Davies et al.  and the general enhancing effect of thinning on the entire ripening process. One possible explanation is that the higher sugar concentration triggers negative feedback that affects the sucrose transporters. Indeed the presence of sugar-response elements in the promoters of various sucrose transporter genes, potentially acting as cis-regulatory elements involved in sugar signaling, has recently been reported .
Despite the significant decline in starch concentration following veraison, several genes involved in starch biosynthesis and modification are modulated during ripening . We detected both highly up- and down-regulated genes in CT involved in starch degradation, and their role is unclear given that developing CT berries accumulate large amounts of sugar. Simultaneous starch synthesis and degradation may facilitate the unloading and storage of sugars in the ripening fruit . Starch-degrading enzymes might also provide carbon backbones for the biosynthesis of secondary metabolites, which could also act as signaling molecules in the regulation of genes controlling phenylpropanoid synthesis [17, 42].
The strong induction of genes encoding malate-degrading enzymes in CT berries, together with the induction of dicarboxylate-tricarboxylate carriers (which transport malate across the mitochondrial membrane thus supplying substrates for the Krebs cycle) supports the specific modulation of malate metabolism by cluster thinning. Malate, whose catabolism is considered responsible of total acidity reduction in grape berry after véraison  is liberated from the vacuole in post-véraison becoming available for catabolism through various avenues [44, 45] and, with the advance of ripening, malic acid is likely a vital source of carbon.
Although advanced malate degradation is heavily dependent upon the extent to which berry temperature is elevated [46, 47] in response to increased sunlight exposure, this condition was not tested in our experiment as the cluster microclimate was not modified by the removal of one cluster of each shoot, maintaining an unchanged canopy structure.
In this respect, our microarray data suggest that malic acid catabolism is accelerated by cluster thinning following the general enhanced ripening process as seen in other researches .
Genes involved in the metabolism of phenylpropanoids and aromatic compounds were strongly modulated in CT berries. Several transcripts involved in phenolic acid, stilbene, flavonoid and isoflavonoid metabolism were affected by the treatment, suggesting that these distinct branches of the phenylpropanoid pathway are directly affected by the increased source/sink balance. However it is possible that the synthesis of phenolic compounds such as stilbenes or isoflavonoids could be part of a systemic response to wounding resulting from the removal of berry clusters, since these compounds are normally produced by the plant in response to stress conditions such as wounding or interactions with pathogens [49, 50].
The anthocyanin content of CT berries was higher than that of controls (Figure 1B and Table 1) but the only CT highly upregulated transcript related to the flavonoid/anthocyanin pathway that could explain this result was DFR. However, looking specifically for known anthocyanin-related transcripts, we found that VvGST4 (VIT_04s0079g00690) and VvMYBA1 (VIT_02s0033g00410) were also more strongly upregulated in CT berries compared to controls (Additional File 2). The high induction of three flavonoid glucosyltransferases and the large number of CT highly upregulated GSTs and transporters of the ABC and MATE families may also play an important role in triggering anthocyanin biosynthesis, although their precise functions in berry ripening remain to be determined. Several putative flavonoid-related transcripts such as F3H, LDOX, LAR1 and UDP-glucose:flavonoid glucosyltransferase were highly downregulated in CT berries, which may reflect the slowing down of the synthesis of non-anthocyanin flavonoid compounds such as proanthocyanidin. Several reports indicate that agronomic treatments such as water stress and cluster thinning can increase total anthocyanins and also produce a shift in their profile [3, 4, 51, 52]. The anthocyanin composition of CT berries differed from controls in our study, with higher relative levels of peonidin 3-glucoside and total 3'4'-OH anthocyanins, in agreement with earlier reports [3, 51]. This might reflect the specific upregulation of F3'Hb in CT berries during the EV and H phases. Although F3'5'Hi and F3'5'Hk were upregulated in CT berries and not expressed at all in controls, there was a significant decrease in the levels of 3'4'5'-OH anthocyanins at harvest, probably because this expression pattern was restricted just to the EV stage. This suggests that the biosynthesis of 3'4'-OH and 3'4'5'-OH anthocyanins is controlled independently .
Cluster thinning is known to change the aromatic profile of grape berries and increase the resveratrol content in wine [53, 54]. In agreement, our microarray data reveal the induction of many putative aroma-modifying transcripts (e.g. geraniol 10-hydroxylase, germacrene D-synthase, as well as two isoforms each of ADH and ALDH) and the specific upregulation of 19 STSs.
The number of genes highly modulated in control berries (567) was small in comparison with previous studies [22, 40] but this reflects our stringent application of the requirement for a fold change > 5. Only 120 of the control highly modulated genes were not highly modulated in CT berries, none of which were involved in carbohydrate or secondary product metabolism. Therefore we assume they play a relatively minor role in the control of berry quality traits at harvest.
The direct comparison of C and CT transcriptomes at each time point indicated that the greatest number of genes differentially expressed in C and CT berries were found at the EV stage, whereas only minor differences were detectable at harvest. This generally supported the cluster analysis results because almost all the differentially expressed genes had already been detected. The novel data obtained by direct comparison related mainly to genes whose expression was only slightly modulated during ripening.