Systems biology of tomato fruit development: combined transcript, protein, and metabolite analysis of tomato transcription factor (nor, rin) and ethylene receptor (Nr) mutants reveals novel regulatory interactions.

Tomato (Solanum lycopersicum) is an established model to study fleshy fruit development and ripening. Tomato ripening is regulated independently and cooperatively by ethylene and transcription factors, including nonripening (NOR) and ripening-inhibitor (RIN). Mutations of NOR, RIN, and the ethylene receptor Never-ripe (Nr), which block ethylene perception and inhibit ripening, have proven to be great tools for advancing our understanding of the developmental programs regulating ripening. In this study, we present systems analysis of nor, rin, and Nr at the transcriptomic, proteomic, and metabolomic levels during development and ripening. Metabolic profiling marked shifts in the abundance of metabolites of primary metabolism, which lead to decreases in metabolic activity during ripening. When combined with transcriptomic and proteomic data, several aspects of the regulation of metabolism during ripening were revealed. First, correlations between the expression levels of a transcript and the abundance of its corresponding protein were infrequently observed during early ripening, suggesting that posttranscriptional regulatory mechanisms play an important role in these stages; however, this correlation was much greater in later stages. Second, we observed very strong correlation between ripening-associated transcripts and specific metabolite groups, such as organic acids, sugars, and cell wall-related metabolites, underlining the importance of these metabolic pathways during fruit ripening. These results further revealed multiple ethylene-associated events during tomato ripening, providing new insights into the molecular biology of ethylene-mediated ripening regulatory networks.

Fruit ripening is a complex developmental program involving the coordinated regulation of numerous metabolic pathways that influence color, flavor, aroma, and texture. Many of these attributes enhance fruit nutritional value and attractiveness, thereby promoting consumption and seed dispersal (Liu et al., 2004;Goff and Klee, 2006) by offering important dietary minerals, vitamins, fibers, and antioxidants to seed-dispersing organisms.
Insights into the genetic mechanisms that mediate fruit ripening-related processes, such as cell wall metabolism, pigment synthesis, and sugar metabolism, have resulted from studies that collectively span a wide range of plant species and indicate that they are broadly conserved (Seymour, 1993;Carrari and Fernie, 2006;Fait et al., 2008;Moing et al., 2011;Zhang et al., 2011). However, tomato (Solanum lycopersicum) has emerged as the primary experimental model in which to study the development and ripening of fleshy fruits (Giovannoni, 2004;Fernandez et al., 2009). This reflects its economic importance and many favorable genetic characteristics, such as small genome size (950 Mb), a relatively short life cycle, and routine transient and stable genetic transformation.
The many studies of the development and maturation of tomato fruits have resulted in the identification of specific genes that participate in ripening (Vrebalov et al., 2002;Manning et al., 2006;Wang et al., 2009;Chung et al., 2010;Nashilevitz et al., 2010;Karlova et al., 2011; for review, see Giovannoni, 2007;Matas et al., 2009). Similarly, much has been learned about the substantial changes in both primary and secondary metabolism that accompany tomato fruit ripening, although much of these data have not been directly related to regulatory events (Carrari and Fernie, 2006) and often target specific pathways. On a more comprehensive scale, tomato development has also been examined at the levels of transcriptome (Alba et al., 2005;Lemaire-Chamley et al., 2005;Vriezen et al., 2008;Wang et al., 2009), proteome (Saravanan and Rose, 2004;Faurobert et al., 2007;Page et al., 2010), and metabolome (Roessner-Tunali et al., 2003;Fraser et al., 2007;Moco et al., 2007) characterization. However, to date, relatively few analyses have combined multilevel approaches Armengaud et al., 2009;Enfissi et al., 2010;Zamboni et al., 2010), and those that have are limited with respect to genotype, developmental stage, or the number of parameters measured.
A number of important advances in our understanding of the mechanisms that regulate ripening have also come from the characterization of monogenic tomato mutants, including ripening-inhibitor (rin), nonripening (nor), colorless nonripening (Cnr), green-ripe (Gr), green flesh (gf), high pigment1 (hp1), high pigment2 (hp2), and never-ripe (Nr; Lanahan et al., 1994;Mustilli et al., 1999;Vrebalov et al., 2002;Liu et al., 2004;Barry and Giovannoni, 2006;Manning et al., 2006;Barry et al., 2008). Although the complete network of ripening regulatory pathways remains to be resolved, cloning of many of the underlying genes has provided insights into their respective roles and hierarchical relationships (Giovannoni, 2004). The rin and Cnr loci encode a MADS box and a SPBP transcription factor, respectively, and are necessary regulators of ripening (Vrebalov et al., 2002;Manning et al., 2006). The Gr gene is suggested to interact with components of the fruit-specific ethylene response (Barry and Giovannoni, 2006), while the Nr mutation has been characterized as an ERS-like ethylene receptor that is impaired in the ability to bind ethylene (Lanahan et al., 1994). The high-pigment mutations influence the role of light in ripe fruit pigmentation via tomato orthologs of the DEETIOLATED and DAMAGED DNA-BINDING PROTEINS1, genes originally defined in Arabidopsis (Arabidopsis thaliana; Mustilli et al., 1999;Liu et al., 2004). The gf gene encodes a senescence-associated STAY-GREEN protein, indicating a role for plastids in both manifesting and regulating ripening phenotypes (Barry et al., 2008).
The identification and characterization of such mutants provide an opportunity to dissect the complex networks of ripening-related pathways at multiple levels. In this study, we have examined three dominant ripening mutants of tomato, nor, rin, and Nr, along the developmental and ripening periods at the transcriptomic, proteomic, and metabolomic levels in order to further extend the analysis carried out by Alba et al. (2005). Use of these specific mutants helps define transcriptional activity as transcriptionally regulated and ethylene response subtranscriptomes. Transcriptional analysis was carried out using the tomato cDNA microarray (TOM1; Alba et al., 2005), microarray-containing ESTs, proteomic data were obtained using isobaric tag labeling (iTRAQ; Wise et al., 2007), and metabolomics data were obtained by gas chromatography-mass spectrometry (GC-MS) according to . The combined results are discussed in the context of current models of ripening and development.

Experimental Design
The experiments described here were designed to span tomato fruit ripening in cv Ailsa Craig, which occurred under greenhouse conditions over a period of 57 d from fruit set to a fully ripe stage. Tomato fruit development can be divided into four major phases: cell differentiation, division, expansion, and ripening (Gillaspy et al., 1993). As described by Alba et al. (2005), these phases correspond to wild-type fruits between 7 and 27 d after pollination (DAP) for the first two phases, to the first visible carotenoid accumulation (42 DAP) in the third phase, and to full red color development (57 DAP) in the ripening phase. Pericarp of anthesis-tagged and equivalent age fruits of the rin, nor, and Nr isogenic mutants was harvested (one biological replicate was considered an individual fruit from different plants). We note that at stages following mature green (39 DAP), the mutant fruits diverge in their development from the wild type, as they are indeed substantially (Nr) or nearly completely (rin and nor) ripening inhibited. Changes resulting from the selected mutations may result in delayed effects, which could be further compounded by occurring in the context of an older fruit, where developmental context is altered due to the normal progression of activities not impacted by the mutations employed. Thus, while an alternative to comparing identically aged fruit would be to compare identical physiological stages such as breaker or red ripe, rin and nor never achieve comparable stages and Nr in the Ailsa Craig background does not achieve red ripe (and breaker is delayed). We have thus selected comparison of identically staged fruits, as this approach allows for comparison of changes in the context of a developmental parameter (age post pollination) that can be measured accurately in all genotypes.
Subsequently, transcriptome analysis was carried out using two-color hybridizations of the TOM1 array, which contains 12,899 different EST clones representing approximately 8,500 tomato genes, using mRNA prepared from each stage. Proteomic data from three ripening stages (39, 42, and 52 DAP) were obtained using the iTRAQ labeling technique. A total of 48 primary metabolites were analyzed by GC-time of flight-MS in fruits harvested between 27 and 57 DAP for nor and rin and from 42 to 57 DAP for Nr, as described in "Materials and Methods."

Transcriptomic Profiling
We investigated differential transcript accumulation using two-color simultaneous hybridizations of the TOM1 array during tomato development for the wild type, Nr, nor, and rin (10 time points: 7,17,27,39,41,42,43,47,52,and 57 DAP). The transcript data for Nr has previously been published (Alba et al., 2005); however, it is included here for comparative purposes and in the broader context of the orthogonal data sets. It is important to note in interpreting our results that Nr in the Ailsa Craig background retains partial ethylene sensitivity and thus partial ripening occurs .
The PageMan  and MapMan  mapping files, described in "Materials and Methods," were used to study the development and ripening of Nr, nor, and rin by identifying significantly overrepresented functional groups, in comparison with the corresponding wild-type control samples, on the basis of Fisher's exact tests and Wilcoxon tests for each category. This facilitated the analysis of the global activation and/or repression of metabolic pathways and gene regulatory networks of the pericarp. We identified metabolic pathways that were enriched during these processes, including several that would be anticipated based on previous studies, including hormone metabolism (Supplemental Fig. S1). The individual gene responses can be viewed in MapMan (Supplemental Data Set S1; Supplemental File S1).
Visual inspection revealed the categories with either qualitatively similar or different responses between the mutants and the control. The general down-regulation of photosynthetic light reactions during the studied stages of Nr is immediately apparent, suggestive of reduced photosynthesis and reduced carbon assimilation in the fruit. In addition, this mutant is characterized by an induction of genes associated with the Suc-to-starch transition, being most prominent at 42 DAP for genes related to Suc degradation and at 17 DAP onward for genes associated with starch synthesis. In rin, transcripts related to starch synthesis appear to exhibit a biphasic response, with a large induction occurring relatively early followed by a secondary increase (43 and 52 DAP). Moreover, these transcripts appear to be up-regulated in nor at 27 and 41 DAP (Supplemental Fig. S1). These results point to the importance of hexoses derived from the degradation of Suc for starch synthesis during early tomato development and at the onset of ripening and are consistent with a recent study highlighting the importance of transitory starch in normal fruit development (Centeno et al., 2011).
A spectrum of genes/proteins involved in cell wall restructuring and disassembly (such as endo-b-1, 4-glucanases from the glycosyl hydrolase [GH] family, expansins, xyloglucan transglucosylase hydrolases [GH16], and a range of pectinases) are canonically associated with fruit development and ripening (Rose and Bennett, 1999;Rose et al., 2004b;Brummell, 2006). In our study, a number of genes associated with cell wall degradation were down-regulated during ripening in nor and rin, while those putatively involved in reversible wall loosening and assembly were downregulated at 27 and 41 DAP in Nr, consistent with the previously observed depolymerization of cell wall components at the onset of storage mobilization (Amor et al., 1995;Reiter, 2008;Supplemental Fig. S1). At the onset of ripening (42 DAP), genes associated with amino acid synthesis were expressed at higher levels, including the Asp-derived Met pathway in nor, rin, and Nr (Supplemental Fig. S1). In contrast, we could infer that amino acid degradation metabolism is generally repressed in all three mutants, and the phytoene synthase-1 gene is down-regulated at the 42-DAP stage (breaker) in nor and at 47 DAP in rin but 1 d before breaker (41 DAP) for Nr (Supplemental Fig. S1).
The expression of genes associated with hormone biosynthetic pathways was altered in Nr and rin, consistent with the lack of climacteric ethylene production and activity in these nonripening fruits. During ripening, transcripts for genes involved in the biosynthesis of ethylene were expressed at high levels in Nr as well as observed in rin, but almost no change was observed in the same time frame for nor, suggesting that nor activity my be upstream of rin in the regulatory hierarchy. Besides the well-characterized participation of ethylene in the control of ripening in climacteric fruits (Giovannoni, 2004(Giovannoni, , 2007Alba et al., 2005), other hormones have more recently been demonstrated to play important roles (Martí et al., 2007;Chaabouni et al., 2009;Yang et al., 2010). These data indicate that a highly complex interactive hormonal network likely regulates fruit ripening. In Nr, transcripts for genes involved in the biosynthesis of auxin and jasmonate were up-regulated at 42 DAP (breaker stage) and 52 DAP, respectively, while nor was characterized by a down-regulation of abscisic acid metabolism at the end of ripening (57 DAP; Supplemental Fig. S1). Interestingly, transcripts associated with ethylene perception and signal transduction changed relatively little but were significantly down-regulated during ripening in nor (Supplemental File S1; Supplemental Fig. S1).
We additionally observed differential expression of transcripts involved in protein synthesis, which were down-regulated during all ripening phases in Nr but were up-regulated during earlier development in nor and also during the period that would correspond to normal ripening in rin. Moreover, genes involved in protein degradation were up-regulated in Nr (Supplemental Fig. S1).
While the changes described above suggest considerable alterations in the programs influencing early development and ripening in the mutants, a linear analysis on a time point-by-time point basis may exaggerate the differences between the genotypes. Therefore, we reanalyzed the data such that gene expression at any time point was compared, within genotypes only, with the first harvested time point ( Fig. 1; Supplemental Fig. S2; Supplemental Table S1). These data revealed a general down-regulation of photosynthetic light reactions during development and ripening in the wild type and the rin mutant ( Fig. 1) and similarly in nor and Nr, but only from 39 DAP onward. In addition, the Nr mutant showed up-regulation of these genes in the first two developmental stages (17 and 27 DAP; Fig. 1). The Calvin-Benson cycle was apparently down-regulated at the 39-and 41-DAP stages during normal tomato fruit development ( Fig. 1A) but across all studied stages in rin (Fig. 1C). A similar pattern of changes was observed from the 39-DAP stage in Nr, which also showed up-regulation in the two earliest stages (17 and 27 DAP; Fig. 1C). The same down-regulation was observed for nor at later stages (43, 47, and 57 DAP; Fig. 1B).
The transcript analysis suggested that starch synthesis was down-regulated from 17 DAP onward Figure 1. Expression analysis in the wild type (WT), nor, rin, and Nr. A condensed PageMap display of altered pathways is shown. Gene expression data are presented as log 2 fold changes in comparison with the first harvested time point (7 DAP) within genotypes. The analyzed time points were 7, 17,27,39,41,42,43,47,52, and 57 DAP. The data were subjected to a Wilcoxon test in PageMan, and the results are displayed in false-color code. Bins colored in red are significantly up-regulated, whereas bins colored in blue are significantly down-regulated. CHO, Carbohydrate. during normal fruit development and ripening, with a similar pattern of changes seen in rin and Nr. However, in Nr, this trend was only significant from the 39-DAP stage onward (Fig. 1, C and D). On the other hand, nor only showed this behavior in later ripening stages (43, 52, and 57 DAP; Fig. 1B). Another process related to sugar metabolism, the degradation of Suc, was suggested to be up-regulated in all three mutants, but with differing temporal patterns: from 42 DAP (breaker) in Nr, 27 and 39 DAP in rin, but only in the last ripening stage (57 DAP) in nor (Fig. 1, B-D). Again, the more pronounced impact of nor on starch and Suc metabolism is consistent with an earlier or more global role in ripening for nor than rin or Nr.
Another well-characterized ripening-related process is the activation of alcohol hydrogenase (Carrari and Fernie, 2006). Analysis of the alcohol hydrogenaseassociated transcripts suggested a similar pattern in nor, rin, and the wild type, but it only occurred at 43 DAP ( Fig. 1B) in nor, at 39, 42, and 47 DAP in rin (Fig. 1C), and in all ripening stages in the wild type (Fig. 1A). We also observed that TCA metabolism was up-regulated across the later stages in Nr (43, 47, and 57 DAP) but down-regulated in rin and nor mutants at this time point (Fig. 1, B -D).
Genes associated with pectin degradation (pectate lyase and polygalacturonase) are clearly up-regulated during normal ripening in tomato fruit, as expected ( Fig. 1A), but this pattern was delayed in Nr, consistent with the residual ethylene response and partial ripening (up-regulated in 47, 52, and 57 DAP; Fig. 1D), and not observed in the other two mutants (Fig. 1, B and C). Several genes related to reversible wall modification were up-regulated at the 42-DAP stage in the wild type, while nor showed no such changes. However, these genes were up-regulated at the 43-and 57-DAP stages in rin ( Fig. 1C) but in Nr only at 57 DAP (Fig.  1D).
The transcript pattern suggested that synthesis of aromatic amino acids was down-regulated from 39 DAP onward in the wild type ( Fig. 1A), while the associated genes showed the same behavior at later stages in the three mutants ( Fig. 1, B-D): 43, 52, and 57 DAP in nor; 47 and 52 DAP in rin; and 57 DAP in Nr. Interestingly, Ser, Gly, and Cys degradation was upregulated at the 47-DAP stage during fruit development in the wild type (Fig. 1A) and throughout all stages of development in Nr, but not in rin or nor (Fig.  1, B-D).
At the onset of ripening (42 DAP), transcripts associated with isoprenoid and carotenoid metabolism were up-regulated in normal ripening (Fig. 1A), whereas in both nor and rin, this was apparent at earlier stages (Fig. 1, B and C). In contrast, in Nr, this pattern was observed in later stages (52 and 57 DAP, again consistent with the partially ripening phenotype), exemplified by phytoene synthase-1, which was down-regulated in earlier development (27 and 41 DAP; Fig. 1D). As was expected from previous studies (Alba et al., 2005;Carrari and Fernie, 2006), ethylene metabolism was altered in all three mutants. In the wild-type fruit, we observed an induction of ethylene biosynthesis at 42 DAP (breaker stage), with a similar pattern of changes seen in Nr (Fig. 1, A and D). Additionally, in the first developmental stages (17 and 27 DAP), a down-regulation of genes related to ethylene signal transduction was apparent (Fig. 1A). In rin, the up-regulation was also displayed at the same stage as in the wild type (42 DAP) but subsequently continued throughout ripening until 52 DAP, as in Nr (Fig. 1C), but in nor, the induction of ethylene biosynthesis was delayed (Fig. 1B). In addition, the Nr mutant displayed an apparent up-regulation of jasmonate metabolism during the ripening process (from 43 to 57 DAP).
During development of wild-type tomato fruit, we observed a down-regulation of polyamine synthesis in the earlier stages (27 and 39 DAP) and also at a later stage (47 DAP; Fig. 1A), while Nr showed up-regulation late in development (57 DAP; Fig. 1D) and no significant changes were observed in rin and nor ( Fig. 1, B and C).
A general up-regulation of transcription regulatory genes was observed at 27 DAP and also from 42 DAP (breaker stage) until 57 DAP during normal tomato fruit development (Fig. 1A). In rin, this up-regulation was only significant at 43 DAP, and the same behavior was observed in the Nr mutant, but again in later stages (52 and 57 DAP; Fig. 1D). However, in nor, the only obvious difference was a down-regulation at 17 DAP (Fig. 1B). Interestingly, the transcription factor family APETALA2 (AP2/EREBP), one member of which (AP2a) has been described as a ripening-related repressor of ethylene response (Chung et al., 2010), was substantially up-regulated in nor (Fig. 1B) at the time point equivalent to the onset of ripening in the wild type (42 DAP).
It is also worth noting that general down-regulation of genes involved in the biosynthesis of ribosomal proteins occurred in both the wild type and rin (Fig. 1, A and C). In nor and Nr, these genes showed the same patterns during ripening (43, 52, and 57 DAP for nor and 52 and 57 DAP for Nr). However, these genes were up-regulated in earlier stages (17, 27, and 39 DAP for nor and 17, 27, 39, 41, and 42 DAP for Nr; Fig. 1

Metabolic Profiling
To follow the repertoire of metabolic changes that occur during tomato development in the three mutants, we carried out extensive metabolic profiling for primary metabolism using an established GC-MS method (Fernie et al., 2004). For the analysis, we selected several stages that span development and ripening for nor and rin (27,39,41,42,43,47,52,and 57 DAP) and from 42 DAP onward for Nr.  presented an exhaustive metabolic study of fruit development and ripening in the Moneymaker tomato cultivar, which shows similarities and also differences in metabolite behavior compared with the cultivar used in this study (Ailsa Craig). In the previous study, it was reported that Suc decreased strongly during the first developmental stages, until 28 DAP, and in par-allel Glc and Fru accumulated in an essentially linear manner. However, these changes were not observed in Ailsa Craig ( Fig. 2; Supplemental Data Set S3). Instead, Suc levels decreased during the transition to ripening Figure 2. Primary metabolite levels in the wild type (WT), nor, rin, and Nr. Time points presented are 17,27,39,41,42,43,47,52, and 57 DAP. Data are normalized to the mean response calculated for the wild type at 7 DAP. Values presented are means SE of five replicates. Asterisks Q28denote differences that were determined to be significant by Student's t test analysis (P < 0.05) compared with wild-type samples harvested at the same stage. (41, 42, and 43 DAP) and later increased (52 and 57 DAP). Glc and Fru also exhibited a reduction during the development-ripening transition (41 and 42 DAP) and increased in the same stages as Suc (Fig. 2). We observed that the levels of Glc, Fru, and Suc were reduced in the three mutants across the time course. Although starch degradation has not been subjected to detailed characterization in tomato pericarp tissue, it would thus appear that tomato fruit, like Arabidopsis leaves (Niittylä et al., 2004), are largely reliant on the hydrolytic pathway of degradation.
We were also able to detect the levels of a total of 11 amino acids across all genotypes. The three mutants and the wild type all revealed marked alterations during development and ripening (Fig. 2). Two main patterns of amino acid accumulation were apparent: (1) those increasing during ripening, and (2) those decreasing exclusively during ripening. Group 1 was dominated by Phe, Asp, Ile, Glu, and Ala, while group 2 consisted of g-aminobutyrate (GABA), b-Ala, Thr, Ser, Gly, and Val. A comparison of these patterns with those observed by  in Moneymaker revealed that the two cultivars display strikingly similar shifts in amino acid levels. That said, a few cultivar-specific differences were apparent, such as a decrease in Ala during development in Moneymaker  but not in Ailsa Craig (Fig. 2). Interestingly, when comparing the metabolic shifts displayed by the mutants, we observed that in all three mutants, Asp, Glu, and Ala, which were characterized by increases on ripening in the wild type, were slightly increased or unaltered during earlier development and ripening (Fig. 2). Moreover, during ripening, nor accumulated considerable amounts of b-Ala and GABA, whereas rin and Nr showed the opposite behavior at the same temporal stages. Additionally, rin and Nr showed lower levels of Thr compared with the wild type, while it was unaltered in nor.
During normal Ailsa Craig fruit maturation, the tricarboxylic acid (TCA) cycle intermediates, citrate, isocitrate, succinate, and malate, were highly variable but generally displayed the same behavior as observed in Moneymaker . During ripening, we observed a strong increase in the level of succinate. In contrast, malate and isocitrate strongly decreased. All three mutants displayed no alteration in their levels of malate and succinate across the studied period. Furthermore, levels of citrate and isocitrate changed significantly only in nor, which displayed the same behavior, albeit to a less dramatic extent compared with the wild type (Fig. 2).
Cell wall-related metabolites, namely Man, Gal, Rha, Rib, Ara, Xyl, and GalUA, increased during normal tomato ripening. This trend is consistent with earlier work, which indicated the particular importance of the degradation of pectin-derived arabinan and galactans during fruit ripening (Sakurai and Nevins, 1993). The levels of these sugars, however, were unaltered in all stages in the mutants, which is again consistent with the characteristic postharvest physiology of these mutants, as they all show substantially reduced pectin degradation (Hobson et al., 1983).

Coordinated Changes of Related Metabolites
In addition to the metabolite changes, we performed a combinatorial analysis of metabolites by subjecting all data points to pairwise correlation analysis. The nor, rin, and Nr mutants showed a total of 520, 864, and 360 significant correlations (P , 0.05), respectively. Off these, 450, 686, and 360 were positive (r 2 . 0.70) and 70, 178, and 36 were negative (r 2 , 20.70), respectively. The full data set of correlation coefficients is presented in the heat maps shown in Figure 3. To simplify the interpretation, metabolites are grouped by compound class. For the Nr mutant, the metabolite correlations were performed only in ripening stages (42, 47, 52, and 57 DAP), while for nor and rin, the analyses were extended to 39, 41, 42, 43, 47, 52, and 57 DAP. In both wild-type sets, all metabolites measured showed significant correlations to compounds outside of their compound class, as was observed previously in cv Moneymaker . The individual metabolites with the highest number of correlations were Gly and GABA, with 21 associations in nor; Ala, Rib, and Xyl, with 28 associations in rin; b-Ala, with 18 associations, and Rha, Ara, and Val, with 17 associations in Nr. In general, Nr and nor displayed far fewer correlations than rin or the wild type. Nr and nor had a considerable number of positive correlations between the levels of various amino acids at different stages of development as well as between amino acids and sugar phosphates, sugar alcohols, and fatty acids, similar to those correlations previously described across development for the wild type Do et al., 2010). That said, the correlations observed for the rin data set were much more similar to those seen in the wild type Do et al., 2010), where cell wall-related metabolites, organic acids, and amino acids were highly correlated. As stated above, the "core" metabolite correlations seen in all four genotypes are reminiscent of the metabolites defined to be reflective of development in our previous study of wild-type tomato development . Furthermore, the tight correlation of the amino acids is in keeping with both surveys of tomato pericarp metabolite levels in populations of introgression lines resulting from wide crosses of tomato Schauer et al., 2006;Do et al., 2010) and of diverse Arabidopsis genotypes (Sweetlove et al., 2008;Sulpice et al., 2010).

Proteomic Analysis
Pericarp tomato proteins from the wild type and three ripening stages (39, 42, and 52 DAP) in nor, rin, and Nr mutants were treated and labeled using iTRAQ reagents and separated by strong cation-exchange fractionation prior to protein identification and quantifica-tion. This approach, in combination with nano-liquid chromatography (nLC)/electrospray ionization-tandem mass spectrometry (MS/MS), resulted in the identification of 158 differentially expressed proteins ( Fig. 4; Supplemental Table S2). As anticipated, nor and rin are phenotypically similar in terms of fruit and fail to produce climacteric ethylene and responsiveness to ethylene at the molecular level (Lincoln and Fischer, 1988). Combined with the inability to induce ripening in either mutant via exogenous ethylene a regulatory network is defined in which ethylene regulates a subset of ripening genes either directly or in concert with developmental signals influenced by the nor and rin gene products (Giovannoni, 2004). Based on this regulatory relationships between the nor and rin genes, a greater number of proteins displayed a similar pattern when comparing nor with rin than when comparing nor with Nr or rin with Nr (Supplemental Fig. S3). In general, we observed that across the three stages analyzed, fruits at the 52-DAP stage showed the highest number of differentially abundant proteins in the mutants compared with the wild type. This total can be subdivided into the 123, 71, and 29 proteins that vary in abundance in nor, rin, and Nr mutants compared with the wild type, respectively. This stage also showed the highest number of common proteins (17) that were detectable in all three genotypes (Supplemental Fig. S3). Of these common proteins, some were related to stress (pathogenesis-related protein, heat shock proteins, glutathione S-transferase (GST)-like protein, catalase protein), cell wall metabolism (polygalacturonase), hormone biosynthesis (1-aminocyclopropane-1-carboxylate [ACC] oxidase), and secondary metabolism (flavonoid glucosyltransferase).  39,41,42,43,47,52,and 57 DAP (A), and for Nr, samples were harvested at 42, 47, 52, and 57 DAP (B). Metabolites were grouped by compound class, and each square represents the correlation between the metabolite heading the column and the metabolite heading the row. Correlation coefficients and significances were calculated by applying the Pearson algorithm using R. Here, only significant correlations are presented (P , 0.05). Positive and negative correlations are presented in red and blue, respectively. WT, Wild type.
Given that the previously discussed transcript data were obtained from the exact same powdered pericarp samples, we next determined the number of differentially identified proteins for which corresponding transcript probes were represented on the TOM1 array. Of the 158 identified proteins, 133 had representative transcript probes on the microarray. Scatterplot anal-ysis of the log 2 -transformed ratios showed the distribution of the corresponding mRNA-to-protein ratios (Fig. 4). At 39 and 42 DAP for all three mutants, almost all mRNA-to-protein ratios were concentrated at the center of the plot (quadrant e), wherein protein and mRNA levels did not vary above 1.5-and 2-fold, respectively. Off center, a total of 45 mRNA-to-protein , and Nr (C) mutants. Each color denotes an mRNA-to-protein ratio that was determined to be significant by Student's t test analysis (P , 0.05) compared with wild-type samples harvested at the same stage: white, Sn-1 (CAA55812); gray, Sn-2 (CAA55813); fluorescent green, heat shock protein (AAD30452); red, polygalacturonase 2 (P05117); blue, pectin esterase (P14280); deep red, Cys proteinase (Q40143); pink, gibberellin 20 oxidase (XP_00251754); dark green, ACC oxidase (ABP68407); dark gray, pathogenesis-related protein P2 (P32045); green-blue, lipoxygenase A (P38415); light gray, lipoxygenase (AAB65766); orange, oxygen-evolving enhancer protein 2 (P29795); yellow, alcohol dehydrogenase (P28032); dark blue, pyruvate decarboxylase (BAC23043); dark lilac, heat shock protein 83 (P51819); dark pink, Met sulfoxide reductase (P54153); dark blue, GST-like protein (AAL92873); brown, pyruvate decarboxylase (AAZ05069); gray-blue, phosphoenolpyruvate carboxylase 2 (CAB65171); light yellow, ASR4 (AAY98032). ratios across all four genotypes were found in which both the mRNA and protein levels exceeded this level of variation (Table I; Fig. 4). At the 39-DAP stage, it was apparent that all three mutants had fewer mRNA-toprotein ratios outside the central quadrant. In Nr, several ratios fell in quadrants b and h, indicating that protein levels were likely up-or down-regulated, but mRNA levels were unchanged. In contrast, nor and rin were characterized by significant down-regulation of Sn-1 and Sn-2 proteins in both protein and mRNA levels (accession nos. CAA55812 [Sn-1] and CAA55813 [Sn-2]). At 42 DAP, we observed that mRNA-to-protein ratios also mainly fell in quadrants b, e, and h. It follows, therefore, that only a few mRNA-to-protein ratios reflected significant changes at both transcript and protein levels. Nr displayed up-regulation at both levels for four proteins: polygalacturonase2 (accession no. P05117), pectinesterase (P14280), Cys proteinase 2 (Q40143), and gibberellin 20 oxidase (XP_00251754). However, contrasting levels were also observed for one gene: up-regulated expression of ACC oxidase (accession no. ABP68407) yet downregulation of its protein abundance. nor and rin additionally shared the commonly up-regulated protein, SGRP-1 (accession no. CAA73034), whose transcript was also significantly up-regulated.
The 52-DAP stage was characterized by a large number of mRNA-to-protein ratios falling in the quadrants a, c, g, and i, wherein mRNA-to-protein ratios are substantially different. Moreover, seven down-regulated proteins, which reflected significant down-regulation at the transcript level, were detected in common between nor and rin, namely pyruvate decarboxylase (BAC23043), GST-like (AAL92873), Met sulfoxide reductase (P54153), polygalacturonase 2 (P05117), heat shock protein 83 (P51819), pyruvate decarboxylase (BAC23043), and alcohol dehydrogenase 2 (P28032). The last three on this list were similarly down-regulated in Nr. In addition to these common abundant proteins, aquaporin-like protein (AALD30452) was also down-regulated in rin and oxygen-evolving enhancer protein 2 (P29795) was downregulated in the Nr mutant. Additionally, nor displayed up-regulation in both protein and transcript levels for Sn-1 (CAA55812) and Sn-2 (CAA55813), as did Nr. Moreover, rin showed up-regulation of two additional proteins, phosphoenolpyruvate carboxylase 2 (CAB65171) and ASR4 (AAY98032). Other up-regulated proteins that corresponded to up-regulated transcript expression were also detected in Nr, such as pectinesterase 1 (P14280), pathogenesis-related protein P2 (P32045), lipoxygenase A (P38415), and lipoxygenase (AAB65766). Additionally, polygalacturonase 2 (P05117) protein and a small heat shock protein (AAD30452) were down-regulated but showed significant up-regulation at the transcript level in Nr and rin, respectively. Also, the up-regulated Sh-RNase (BAE92268) protein in Nr was down-regulated at the transcriptional level. In summary, these results suggest a substantive degree of posttranscriptional regulatory activity during fruit maturation that has not been described previously in reports focused solely on transcript or protein analysis.

Network Analysis of Transcript, Metabolite, and Protein Levels during Ripening of nor and rin Tomato Fruits
As is typical of climacteric fruit, ripening in tomato is regulated to a substantial degree by the hormone ethylene. To better understand the ethylene regulatory network, we performed an association study of individual metabolites, a subset of transcript data, and proteins in both the nor and rin mutants (Supplemental Data Set S2). For the integration analysis, we selected three different stages that covered the ripening process in the wild type: 39 DAP (mature green), 42 DAP (breaker), and 52 DAP (ripe). The data sets used for the correlation comprised 48 metabolites from primary metabolism and a subset of transcripts showing significantly differential expression (178 transcripts for nor and 191 for rin). The latter included those associated with cell wall metabolism, starch synthesis and degradation, redox stress, polyamine metabolism, transcription factors, fermentation metabolism, TCA cycle, hormone metabolism and signaling, and secondary metabolism. All proteins showing significant differential abundance in the developmental stage studies were selected (154 proteins for nor and 121 for rin). We analyzed coresponses between metabolic pathway genes, proteins, and metabolites in each genotype (nor and rin) using a Pearson's correlation coefficient approach at a strict stringency threshold (P , 0.01). In each correlation matrix ( Fig. 5; Supplemental Fig. S4), transcripts, proteins, and metabolites are represented as nodes (different node shapes represent the different data sets: significant correlations are represented as links that connect the nodes). Network analysis emphasized links both between and within the various types of data. We built the network from each mutant with data relative to wild-type values (transcripts, proteins, and metabolites) in exactly the same way. Interestingly, nor revealed markedly higher connectivity than rin (nor showed a total of 889 links between 357 nodes; rin showed a total of 792 links between 354 nodes), suggesting again that NOR may represent a higher level in the regulatory hierarchy as compared with RIN.
Based on our previous results, where a high number of common proteins were found between nor and rin in the ripe stage (52 DAP), we expected to have a highly significant overlap between the nor and rin networks. Surprisingly, taking into account the comparative network analysis of three ripening stages when comparing both networks (39, 42, and 52 DAP), we observed different regulation between transcripts, proteins, and metabolites within the three ripening stages (Fig. 5). Comparative network analysis also suggested that in nor, associations between nodes were largely modular, while in rin, we observed more robust associations between specific metabolites, namely malate, fumarate, Glu, Pro, and Gly (Fig. 5B). The nor network analysis revealed two interesting clusters, one of which includes the Nr transcript (3.3.1.9; TOM1 identifier SGN-U590044; cluster 1 in Fig. 5A) and another of which Table I. Correlation between the expression ratios (mutant/wild type) The relative change in abundance (mutant/wild type) is shown in log 2 scale from proteins from 39-, 42-, and 52-DAP tomato fruits. Boldface values denote differences that were determined to be significant by Student's t test analysis (P , 0.05) compared with wild-type samples harvested at the same stage.  Fig. 5A). Interestingly, in the nor network (cluster 1), the Nr transcript showed coexpression with hormones and cell wall metabolism regulators. The Nr transcript level (3.3.1.9; TOM1 ID SGN-U590044) displayed positive correlations with lipoxygenase protein abundance, which is, in turn, positively correlated to lipoxygenase transcript levels (6.4.16.5; TOM1 ID SGN-U578028). The Nr transcript also showed positive correlations to the pectin degradation-related genes pectate lyase (1.1.6.7; TOM1 ID SGN-U585252) and polygalacturonase (1.4.12.4; TOM1 ID SGN-U577423). In turn, these pectin degradation-related gene transcripts (pectate lyase, 1.1.6.7; TOM1 ID SGN-U585252; polygalacturonase, 7.4.12.12; TOM1 ID SGN-U577423) showed positive correlation with the hormone-related genes lipoxygenase (6.4.16.5; TOM1 ID SGN-U578028) and ACC oxidase (3.3.4.14; TOM1 ID SGN-U578607). This small regulated cluster was clearly related to primary metabolism, with all these nodes showing a negative correlation with a phosphoglycerate kinase protein.
Moreover, the phosphoglycerate kinase protein displayed a positive correlation with PSI reaction subunit II and PSII proteins that are negatively correlated with the Nr transcript as well as with cell wall-related genes and lipoxygenase protein.

Ripening Mutations Provide Tools to Parse Transcription and Hormonal Control
The availability of well-characterized genetic mutants in fruit development and ripening has promoted considerable research on tomato (Giovannoni, 2007). These include two ripening-associated transcription factors, rin, which encodes a SEPALLATA MADS box gene (Robinson and Tomes, 1968;Vrebalov et al., 2002), and nor (GenBank accession no. AY573802; Tigchelaar et al., 1973), and a two-component His kinase ethylene receptor, ETR3 (Nr; Wilkinson et al., 1995;Yen et al., 1995).
The RIN transcription factor is necessary to induce ripening-associated increases in respiration and ethylene concentration, resulting in fruits that fail to complete normal ripening (Vrebalov et al., 2002;Giovannoni, 2007). Furthermore, fruits remain responsive to ethylene in that ethylene-responsive genes are induced by exogenous ethylene. The ripening regulator transcription factor NOR encodes a NAC domain protein (Tigchelaar et al., 1973, J. Vrebalov and J. Giovannoni, unpublished data). nor is phenotypically similar to rin in that nor fruit fail to produce climacteric ethylene yet show responsiveness to ethylene at the molecular level via ethylene-induced gene expression while remaining unripe in response to ethylene (Lincoln and Fischer, 1988). While nor and rin may well act together in a cascade to control ripening (Giovannoni, 2007), a number of molecular differences were observed when the expression of ripening-related genes in the presence and absence of exogenous ethylene was examined. Specifically, in mature nor fruit, transcripts corresponding to both phytoene synthase (PSY1) and the E8 gene, which are known to be up-regulated during normal ripening (Lincoln et al., 1987), are absent and expression of these genes is not restored by exposing the fruit to ethylene Thompson et al., 1999). Similar behavior was observed for the polygalacturonase transcript in the presence and absence of ethylene Thompson et al., 1999). However, in rin, the effect was not as strong as in nor, with rin displaying normal induction but reduced levels of expression of PSY1 and E8 in mature fruits (Knapp et al., 1989), which could be restored with exogenous ethylene application . These data suggest that nor may have a more global effect on ethylene/ripening-related gene expression than rin, in that reduced expression in rin reflected the lack of ethylene in these fruit but the lack of expression in nor, and the inability to restore expression with ethylene suggests a higher level of control. In the model proposed by Thompson et al. (1999), both genes operate upstream of crucial ripening activity and nor is upstream of rin, since ethylene/ripening-related gene expression appears to be more extensively inhibited in nor.
Ethylene is perceived by two-component His kinase ethylene receptors (Chang and Shockey, 1999), and a mutation in the receptor LeETR3 (Nr) interferes with ethylene perception and, therefore, significantly delays fruit ripening (Lanahan et al., 1994;Wilkinson et al., 1995) and represents a tool to define the fruit ethylene-regulated transcriptome/proteome/metabolome/phenome. Previous transcriptomic, phenotypic, and targeted metabolic analysis revealed that Nr influences fruit morphology, seed number, ascorbate accumulation, carotenoid biosynthesis, and ethylene evolution (Alba et al., 2005). Here, we exploited the rin, nor, and Nr mutations to define systems under the influence of transcriptional (rin and nor) and ethylene (Nr) control.

Systems Analysis of Ripening Control Using Ripening Mutations
Here, we utilized an extensive array of tools to characterize the various molecular entities (transcriptome, proteome, metabolome) of the cell (Alba et al., 2004;Fernie et al., 2004;Rose et al., 2004b) in the three ripening tomato mutants, rin, nor, and Nr, with the aim of better understanding transcriptional and ethylenedependent ripening control. Transcriptional analysis on the three ripening mutants at 10 different developmental stages spanning fruit and ripening revealed how ethylene, via NOR, RIN, and Nr, influences the expression of hundreds of genes during development both prior to the onset of and during ripening. These results corroborate the central role that ethylene plays in governing biochemical, physiological, and developmental processes throughout tomato development.

Different Regulation of Ethylene-Related Genes in nor, rin, and Nr
PageMan analysis of microarray data revealed significant changes in different functional gene categories across the three mutants, supporting overlapping and independent regulatory effects. Regarding ethylene metabolism, and in agreement with previous studies (Wilkinson et al., 1995;Vrebalov et al., 2002), we observed that ethylene-related genes were differentially expressed in the three mutants and as compared with the wild type. During normal development, ethylene synthesis-related genes were up-regulated in breaker stage (42 and 43 DAP) fruits and ethylene signal transduction genes in late ripe stages (47 and 57 DAP). In Nr, ethylene synthesis-related genes were strongly up-regulated in a time frame consistent with the normal ripening process (42 DAP) as described by Alba et al. (2005). Interestingly, those genes showed similar upregulation in rin but were unaltered in nor, suggesting a response to nor-specific activities. Downstream in the ethylene signaling cascade, ethylene signal transduction was apparently unaltered across development and ripening in the three mutants. These data thus support the model postulated by Thompson et al. (1999) where nor is a higher order regulator that rin or Nr (ethylene). Recent chromatin precipitation analysis of RIN protein targets indicates that ethylene synthesis (ACC synthase 2 and 4) gene promoters directly interact with RIN (Fujisawa et al., 2011). This would suggest a linear order of NOR-RIN-NR activity.

Characterization of the Fruit Ethyleneome Demonstrates Regulated Metabolomic Flux toward Ethylene Synthesis
Climacteric ethylene biosynthesis includes the conversion of Asp to Met, the conversion of Met to ethylene, and the Met recycling pathway (Yang and Hoffman, 1984). Ethylene is synthesized from S-adenosylmethionine (SAM) by the sequential action of two ethylene biosynthetic enzymes, namely, ACC synthase and ACC oxidase. SAM also has the potential to participate in polyamine or ethylene biosynthesis or both (Mattoo et al., 2007;Mattoo and Handa, 2008). From the transcriptomic and metabolomic data in this study (Figs. 1 and 2), it is apparent that ethylene biosynthesis may dominate the flux through SAM. During normal tomato ripening at the initiation of ethylene biosynthesis, Asp, the precursor of Met, increased in addition to putrescine, one of the three major plant polyamines . Asp and putrescine levels were higher in rin and Nr; however, the changes in Asp levels in the two mutants were far less dramatic than those observed in the wild type, while the change in putrescine was more substantial. Interestingly, these changes in Asp and putrescine correlated with the up-regulation of ethylene synthesis-related genes in rin and Nr. In contrast, no significant changes in Asp and putrescine levels were observed when comparing nor with the wild type, which was mirrored by the lack of change in the expression of ethylene synthesis-related genes in this genotype. These changes are in agreement with a previous study (Mattoo et al., 2007) in which detailed analysis of polyamine accumulation in transgenic tomato lines deficient in the expression of enzymes of polyamine biosynthesis revealed a shift in SAM flux concomitant with an increase of ethylene and a decrease of Asp levels.
nor, rin, and Nr Displayed Metabolic Shifts Levels of malate, the precursor of Asp, decreased during ripening in the wild type, as reported previously . Malate levels also decreased in rin and Nr, but the change observed in these mutants was less than that observed in the wild type. In contrast, malate content was unaltered during fruit development in nor. These data suggest that ethylene feedback regulates its own synthesis not only at the transcriptional level but also the metabolic level. Another general trend was a general up-regulation of genes related to Suc degradation in rin and Nr but not in nor as well as a down-regulation of aromatic amino acid synthesis-related genes across the late stages of ripening among the three mutants. These changes were reflected in lower Glc and Fru levels, the major hexoses in tomato fruits, in the three mutants in comparison with the wild type. Levels of Phe, the only aromatic amino acid detected in this study, gradually increased during normal ripening and also in the three mutants, but to a lesser extent. Interestingly, aromatic amino acids can act as alternative respiratory substrates when carbohydrates are not abundant (Ishizaki et al., 2005). As such, this decline may reflect a partial reliance of the mitochondrial electron transport chain during later stages of fruit development, when carbon demand is provided entirely by source leaves of the plants, as opposed to a partially autonomous supply of photoassimilates during early fruit development.
Cell Wall Metabolism Is Affected in nor, rin, and Nr One of the most evident ripening-related changes in tomato fruit is the progressive and extensive loss of firmness, resulting in part from cell wall disassembly (Brummell, 2006;Vicente et al., 2007). Many reports have described modifications to pectic polysaccharides in primary cell walls and middle lamellae, which include the removal of methyl ester groups by pectin esterases, the solubilization of GalUA-rich polysaccharides, and the loss of Gal from pectic fractions (Pressey, 1983;Redgwell et al., 1997). The disassembly of hemicellulosic and pectin cell wall polysaccharides is known to be severely reduced in rin fruit and apparently is diminished in nor, since they also have only limited softening (Seymour et al., 1987;Giovannoni et al., 1989;Dellapenna et al., 1990;Maclachlan and Brady, 1994). The cell wall-associated transcript, protein, and metabolite data in this study corroborate and extend these previous results. At the transcript level, we observed up-regulation of the cell wall degradation-related genes polygalacturonase and pectate lyase during ripening in the wild type but not in equivalent-stage nor and rin fruits, and the up-regulation of these genes was delayed in Nr, consistent with the partial ripening response of the mutant. However, at the cell wall-related metabolite level, all three mutants displayed similar patterns during ripening: galacturonate, Gal, Ara, Xyl, Rib, and Rha levels increased substantially during ripening in the wild type, but no such changes were observed in nor, rin, and Nr. These transcript and metabolite data, therefore, indicate that normal ripening-related cell wall degradation is compromised in all three mutants and further suggest that this is an ethylene-influenced phenomenon. Moreover, at the protein level, it has already been shown that in cell wall preparations from normal tomato fruit during ripening, salt-soluble proteins (cell wall-related proteins) were increased (Hobson et al., 1983). However, polygalacturonase-solubilized protein from rin was less, and that from the Nr cell walls was more, than that from normal wall preparations (Hobson et al., 1983). Similar results were observed in our data: in Nr, polygalacturonase and pectin esterase (cell wall degradation-related proteins) were more abundant as compared with the wild type at the breaker stage (42 DAP); however, the opposite behavior was observed at the same stage in rin. Moreover, in late ripening stages (52 DAP), a reduction of these proteins was apparent in all three mutants, reflecting that cell wall degradation is likely compromised in these mutants Giovannoni et al., 1989). Furthermore, we observed that the reduction in the level of polygalacturonase protein was stronger in nor and rin as compared with Nr at the 52-DAP stage, suggesting that polygalacturonase protein accumulation is influenced by both ethylene and additional transcriptional effects.
Others Metabolic Changes Are Shown by nor, rin, and Nr In addition to cell wall metabolism per se, the closely related metabolic pathway leading to ascorbate biosynthesis has also been described in ripening tomato fruits . We observed an upregulation of ascorbate-related genes during ripening as well as ascorbate accumulation, as described previously (Alba et al., 2005), only in Nr. Interestingly, neither nor nor rin showed changes in this pathway, suggesting that ascorbate metabolism is influenced by ethylene perception.
PSY1 catalyzes the formation of phytoene, the first C40 carotene intermediate in carotenoid biosynthesis. This is an essential step in the production of the carotenoids, which give the fruit its red color. Nr, nor, and rin all showed some carotenoid production, although the synthesis of lycopene was reduced, delayed, or absent depending on the mutation (Tigchelaar et al., 1973). These phenotypes probably can be explained by the different time frame of expression of PSY1. During normal tomato ripening, this gene was induced from breaker stage (42 DAP); however, in nor and rin, this up-regulation was at earlier stages (39 and 41 DAP, respectively), and in Nr, it was at a later stage (52 DAP), again indicating complex ethylene modulation of PSY1 gene expression.
To date, several studies have indicated that levels of organic acids correlate strongly with genes associated with ethylene and cell wall metabolism-associated pathways, underscoring the importance of these metabolic intermediates in ripening Centeno et al., 2011). In this study, we report than in nor, levels of all organic acids, mainly TCA cycle intermediates, were strongly affected across ripening when compared with the wild type, while in rin and Nr, relatively small changes were observed. These results support the hypothesis that TCA cycle intermediates are regulated at the transcriptional level , although we cannot exclude the possibility that in plants, as in animals (He et al., 2004), these intermediates could play a key role in mediating retrograde-regulated gene expression (Zarkovic et al., 2005). Furthermore, the fact that organic acids were more strongly affected in nor, a mutant in which ethylene biosynthesis and signaling are also strongly af- fected, based on our transcriptomic data, suggests that these metabolites are developmentally regulated at the level of ripening-associated transcripts. This observation has important biotechnological implications, given that the manipulation of organic acids in tomato ripening is relatively facile (Centeno et al., 2011).
Protein Variations as Compared with Transcriptomic Data in nor, rin, and Nr Transcriptomic profiling allowed the identification of hundreds of genes that were differently regulated during the 10 stages of tomato fruit development among the three mutants. While by no means comprehensive, in this study a total of 158 proteins showed variations in abundance during ripening or equivalent temporal stages of development in nor, rin, and Nr when compared with the wild type. We also found that a similar number of proteins showed decreased or increased expression among the three mutants. The highest number of differently expressed proteins with respect to the wild type were at the ripe stage (52 DAP) among the three mutants. Moreover, nor and rin shared a greater number of common proteins, especially at 52 DAP (Supplemental Fig. S3). These data further support our conclusion that nor and rin act together in the ethylene signaling cascade. Of the 158 identified varying proteins, 133 corresponded to gene sequences that figure on the TOM1 microarray, allowing a comparison of ripening-related or genotypic differences in specific transcripts or cognate proteins. In this study, 32% (42 out of 133) of protein and transcript pairs decreased or increased in parallel during ripening. This correlation was particularly apparent at the late ripening stage (52 DAP). A high number of proteins that showed no correlations with the corresponding transcripts belonged to primary metabolism, such as Ser malate dehydrogenase, enolase, glyceraldehyde-3-phosphate dehydrogenase, acid invertase, hydroxymethyltransferase, Glu decarboxylase, and Asp aminotransferase, suggesting that posttranscriptional regulation plays an important role in primary metabolism gene expression and highlighting the need for combined transcriptomic and proteomic analyses.

CONCLUSION
Network analysis revealed key candidate regulators that may play defining roles in tomato fruit ripening. This combination of transcriptome, proteome, and targeted metabolite analysis has helped to refine the ethylene-regulated transcriptome of tomato fruit and added to our knowledge the role of ethylene in both protein and metabolite regulation in tomato ripening. Furthermore, in this study, we were able to draw several important conclusions concerning transcriptomic/metabolic regulation during tomato ripening. First, our data support the contention that nor and rin act together in a cascade to control ripening Thompson et al., 1999) and also suggest that nor has a more global effect on ethylene/ripening-related gene expression than rin, which indicates that nor operates upstream of rin. The data also support that rin and nor act upstream of Nr, which would be logically anticipated given the necessity of rin and nor for ethylene synthesis. A potential genetic regulatory network centered on ethylene governing tomato fruit development and ripening is presented in Figure 6. Third, the metabolite abundance of specific compound classes, such as TCA cycle organic acids and cell wall-related metabolites, appears to be strictly controlled, with specific compounds influenced by ethylene, transcriptional control, or both. Based on an integrated analysis of transcripts, proteins, and metabolites, we were able to identify areas of metabolism that seem to be of high importance to the ripening process, such as hormones and cell wall metabolism in ethylene perception. Therefore, the integrated analysis enables us to uncover additional information for the comprehensive understanding of biological events relevant to metabolic regulation during tomato fruit development. These outputs will provide potential targets for the engineering of metabolism to facilitate the controlled modulation of ripening in tomato fruit.

Plant Material and Sampling
Tomato (Solanum lycopersicum 'Ailsa Craig') and isogenic lines for nor, rin, and Nr mutations (cv Ailsa Craig, backcross parent) were grown in the greenhouse at 26°C under 12 h of supplemental lighting followed by 12 h at 20°C. To collect stages prior to ripening, fruits were tagged at 7 DAP (fruit less than 1 cm; Giovannoni et al., 1989) and harvested at one of the following 10 time points: 7, 17, 27, 39 (mature green), 41 (breaker 2 1), 42 (breaker), 43 (breaker + 1), 47 (breaker + 5), 52 (breaker + 10), and 57 (breaker + 15) DAP. The first signs of carotenoid accumulation on the external surface of the fruit were taken to define the breaker stage. To maximize developmental synchrony, harvested fruit were visually inspected externally and internally (e.g. size, shape, pigmentation, seed development, and development of locular jelly), and only fruits appearing developmentally equivalent were kept for the analysis. Transcriptome, proteome, and metabolome analyses were performed in the same material. For the Nr mutant, analysis was performed in three separate pools of fruits of 15 to 30 fruit each. All fruits were collected from four individual plants. For nor and rin mutants, eight to 10 individual fruits for each time point were considered as biological replicates. These individual fruit replicates came from 10 to 15 rin, nor, and wild-type plants (all of which were grown in the greenhouse crop; the genotypes were randomly distributed in the greenhouse).
For transcriptome, proteome, and metabolome analyses, eight, four, and five biological replicates were used. After tissue selection, pericarp tissue was collected, frozen in liquid nitrogen, and stored at 280°C until further analysis.

Transcriptome Analysis
Total RNA from tomato pericarp was extracted as described by Alba et al. (2004Alba et al. ( , 2005. cDNA synthesis, labeling, hybridization, washes, and scanning were conducted as described (Alba et al., 2004). Glass slides containing arrayed tomato ESTs were obtained directly from the Center for Gene Expression Profiling at the Boyce Thompson Institute, synthesized in the Giovannoni laboratory as described by Alba et al. (2004). The tomato array (TOM1) contains approximately 12,000 unique elements randomly selected from cDNA libraries isolated from a range of tissues, including leaf, root, fruit, and flower and representing a broad range of metabolic and developmental processes. Nucleic acid sequence and annotation data pertaining to the TOM1 microarray are available via the Tomato Functional Genomics Database (http://ted.bti.cornell.edu). Data acquisition and filtering were conducted as described (Alba et al., 2005). The raw intensity values were normalized using Robin's default settings for two-color microarray analysis (Lohse et al., 2010). Specifically, background intensities were subtracted from the foreground values, and subsequently, a print tip-wise Loess normalization (Yang et al., 2002) was performed within each array. To reduce technical variation between chips, the logarithmized red and green channel intensity ratios on each chip were subsequently scaled across all arrays (Yang et al., 2002;Smyth and Speed, 2003) to have the same median absolute deviation. Statistical analysis of differential gene expression was carried out using the linear model-based approach developed by Smyth (2004). The obtained P values were corrected for multiple testing using the strategy described by Benjamini and Hochberg (1995) separately for each of the comparisons made. Genes that showed an absolute log 2 fold-change value of at least 1 and a P value lower than 0.05 were considered significantly differentially expressed. The log 2 foldchange cutoff value was imposed to account for noise in the experiment and to make sure that only genes that show a marked reaction were recorded. For visualization, the data were loaded into MapMan , which displays individual genes mapped on their pathways as false-color-coded rectangles. MapMan bins were assigned to each cDNA on the glass based on the SGN tomato unigene mapping. Wilcoxon rank sum tests were performed to test whether there were bins that were significantly and consistently different from the other bins in the MapMan ontology using the built-in function in MapMan. The PageMan software package  was then used to select and display biologically relevant details of these data sets. The transcript data for Nr has previously been published (Alba et al., 2005). The design of nor and rin microarrays was performed using RNA pooled from different tissues (a combination of tissues from different developmental stages of pericarp, leaf, and stem tissue) as a reference.

Proteome Analysis
Proteins were extracted from pericarp tissues of mature green, breaker, and breaker + 10 stages from four biological replicates of the wild type and the nor, rin, and Nr mutants. Samples were prepared as described (Isaacson et al., 2006). Extracted proteins were quantified using a modified Bradford method (Bradford, 1976). In order to confirm the quantification, 20 mg of each sample was loaded on a 12% acrylamide gel and run under denaturing conditions. The gel was stained with 0.04% Coomassie blue R-250 staining solution and analyzed using ImageJ software (Abramoff et al., 2004). Protein preparation and identification were done as described by G. López-Casado, P.A. Covey, P.A. Bedinger, L.A. Mueller, Z. Fei, J.J. Giovannoni, and J.K.C. Rose (unpublished data). Proteins were labeled using the iTRAQ (Isobaric Tag for Relative and Absolute Quantitation) Reagents Multiplex kit (Applied Biosystems) according to the manufacturer's instructions and digested with trypsin. Labeled peptide mixtures were first analyzed by matrix-assisted laserdesorption ionization time of flight MS (ABI 4700; Applied Biosystems) to confirm digestion and labeling. A strong cation-exchange fractionation LC apparatus using an Agilent 1100 HPLC device with a UV detector (Agilent Technologies) was followed by nLC-MS/MS. All nLC-MS/MS experiments were performed using an LC Packings Ultimate integrated capillary HPLC system equipped with a Switchos valve switching unit (Dionex) connected in line to a hybrid triple quadrupole linear ion-trap mass spectrometer (4000 Q Trap; ABI/MDS Sciex) and equipped with a Micro Ion Spray Head ion source. MS data acquisition was performed using Analyst 1.4.2 software (Applied Biosystems) in the positive ion mode for information-dependent acquisition analysis. MS data were interrogated using the paragon method (Shilov et al., 2007) using ProteinPilot software 2.0 (Applied Biosystems) and searched against a translated custom unigene database derived from the RNA-sEquation (454 pyrosequencing; GS FLX; Roche) database (Assembly2009_protein; ftp://ted.bti.cornell.edu/pub/tomato_454_unigene/). Only proteins showing a protein score higher than 95% (or 1.3 in ProteinPilot) with two or more unique peptides identified with a P value of 0.05 or less, and false discovery rate of less than 0.15, were further analyzed.

Metabolome Analysis
Metabolite analysis by GC-MS was carried out essentially as described (Lisec et al., 2006) but with modifications specific for tomato . The mass spectra were cross-referenced with those in the Golm Metabolome Database (Kopka et al., 2005). Analyses of primary metabolites were done in pericarp tissue from 27,39,41,42,43,47,52, and 57 DAP in lines for nor and rin mutations and from 42, 47, 52, and 57 DAP from the Nr mutation. Each sample point was analyzed with five biological replicates.

Data Analysis and Statistics
Data normalization, visualization, heat maps, and correlation analysis based on Pearson correlation were performed using R software (Gentleman, 1996). Visualization of transcript-protein-metabolite correlation was conducted using Cytoscape (Cline et al., 2007).

Supplemental Data
The following materials are available in the online version of this article.
Supplemental Figure S1. A condensed PageMap display of changed pathways.
Supplemental Figure S2. A condensed PageMap display of altered pathways that are not represented in Figure 1.
Supplemental Figure S3. Venn diagrams of proteins that increased or decreased among the three mutants.
Supplemental Table S2. Total significantly detected proteins among the three mutants relative to the wild type.
Supplemental Data Set S1. Transcriptome data set.
Supplemental Data Set S2. Selected transcripts and proteins for network analysis.
Supplemental Data Set S3. Primary metabolite levels in the wild type, nor, rin, and Nr (metabolites not presented in Fig. 2).