A comparative transcriptomics and eQTL approach identifies SlWD40 as a tomato fruit ripening regulator

Abstract Although multiple vital genes with strong effects on the tomato (Solanum lycopersicum) ripening process have been identified via the positional cloning of ripening mutants and cloning of ripening-related transcription factors (TFs), recent studies suggest that it is unlikely that we have fully characterized the gene regulatory networks underpinning this process. Here, combining comparative transcriptomics and expression QTLs, we identified 16 candidate genes involved in tomato fruit ripening and validated them through virus-induced gene silencing analysis. To further confirm the accuracy of the approach, one potential ripening regulator, SlWD40 (WD-40 repeats), was chosen for in-depth analysis. Co-expression network analysis indicated that master regulators such as RIN (ripening inhibitor) and NOR (nonripening) as well as vital TFs including FUL1 (FRUITFUL1), SlNAC4 (NAM, ATAF1,2, and CUC2 4), and AP2a (Activating enhancer binding Protein 2 alpha) strongly co-expressed with SlWD40. Furthermore, SlWD40 overexpression and RNAi lines exhibited substantially accelerated and delayed ripening phenotypes compared with the wild type, respectively. Moreover, transcriptome analysis of these transgenics revealed that expression patterns of ethylene biosynthesis genes, phytoene synthase, pectate lyase, and branched chain amino transferase 2, in SlWD40-RNAi lines were similar to those of rin and nor fruits, which further demonstrated that SlWD40 may act as an important ripening regulator in conjunction with RIN and NOR. These results are discussed in the context of current models of ripening and in terms of the use of comparative genomics and transcriptomics as an effective route for isolating causal genes underlying differences in genotypes.


Introduction
Given that seed dispersal is of major ecological and evolutionary importance for all plants and the fact that fleshy fruit plays a vital role in this process, fruit ripening assumes a central importance in the plant life-cycle. It is well documented that hundreds of genes display altered expression during this process (Karlova et al., 2011;Osorio et al., 2011), and that metabolism also undergoes concurrent dramatic shifts to form fruit quality (Carrari and Fernie, 2006). As one of the most important appearance qualities, the accumulation of carotenoids, when combined with naringenin chalcone tainted yellow peel, forms the reddish color of tomato fruits (Solanum lycopersicum) (Ballester et al., 2010;Fernandez-Moreno et al., 2016;Zhu et al., 2018). Among the carotenoid biosynthesis pathway, phytoene synthase (PSY) is the key rate-limiting enzyme of the whole pathway; it catalyzes two molecules of GGPP to form the colorless phytoene. Subsequently, under the catalysis of a series of enzymes, phytoene undergoes dehydrogenation and isomerization reactions to form lycopene, which is the dominant carotenoid of tomato fruit (Bird et al., 1991;Bartley and Scolnik, 1993;Cazzonelli and Pogson, 2010;Chen et al., 2019). In addition, the texture of fruits is affected by the modification of cell walls and pectate lyase (PL), which hydrolyzes pectin and is the most substantially cell wall gene contributor to this process identified to date . Besides appearance and textural qualities, another important quality aspect is taste, which has been attributed to the sugar/organic acid ratio, and volatile and secondary metabolite accumulation. The key genes underlying the levels of these metabolites have been uncovered by a range of quantitative trait loci (Fridman et al., 2004;Tieman et al., 2006;Schauer et al., 2008;Centeno et al., 2011) and genome-wide association studies (Sauvage et al., 2014;Tieman et al., 2017;Ye et al., 2017;Gao et al., 2019).
Moreover, the considerable metabolic changes are coordinated and mediated by transcription factors (TFs) and epigenome dynamics on the metabolic structural genes' expression (Centeno et al., 2011;Rohrmann et al., 2011;Giovannoni et al., 2017;Lu et al., 2018;Li et al., 2020). Over the last 50 years, several mutants, such as ripening-inhibitor (rin), nonripening (nor), Never ripe (Nr), and Colorless nonripening (Cnr) mutations, have been identified as severely impacting the tomato ripening process (Tigchelaar, 1973;Lanahan et al., 1994;Vrebalov et al., 2002;Manning et al., 2006). Among these mutants, rin is the one of the most famous ripening delaying mutants substantially lacking the ethylene burst and hindering the color change and softening processes, which results from the repression of the ripening inhibitor-macrocalyx (RIN-MC) chimeric protein (Robinson, 1968;Vrebalov et al., 2002;Ito et al., 2017). The integrated analysis of chromatin immunoprecipitation (ChIP)-chip and transcriptome indicated that RIN can directly induce the expression of the key ripening-related structural and regulator genes, ACS2/4, SGR1, PSY, Cel2, EXP1, PAL1, C4H, LoxC, AAT1, CNR, NOR, AP2a, and itself (Fujisawa et al., 2012(Fujisawa et al., , 2013Irfan et al., 2016). Furthermore, the transcriptional behavior of 1,000 TFs has been established during tomato ripening  and gene regulator networks have been modeled on the basis of these data (Rohrmann et al., 2012). The stronger acting ripening genes mentioned above were, in contrast, identified in mutant screens intent on isolating strong mutants in order to enhance tomato shelf-life (Tigchelaar, 1973;Lanahan et al., 1994;Vrebalov et al., 2002;Manning et al., 2006). While recent years have resulted in the identification of many additional genes with ripening consequences (Li et al., 2021;Shi et al., 2021), it is probable that additional contributors to this complex process remain to be found. As such, further genome-wide analysis is required to mine the regulator affecting this process and provide more comprehensive knowledge on this process.
Time-series, species/accession specific or tissue specific and comparative transcriptomics studies have previously deciphered gene regulatory networks underlying plant developmental pathways allowing the identification of additional functional genes (Breschi et al., 2017;Chang et al., 2019;Batyrshina et al., 2020;Baranwal et al., 2021). For example, Bolger et al. (2014) compared the transcriptomes of M82 and Solanum pennellii, and identified 100 key candidate genes related to salt and drought stress. Additionally, transcriptomics studies on genetic mapping populations have defined expression QTLs (eQTLs) as genomic loci controlling variation in steady state levels of transcript between individuals (Sønderby et al., 2007) of what has subsequently become a well-characterized mapping population (Ofner et al., 2016;Szymanski et al., 2020). During tomato domestication, many phenotypes (such as the leaf structure and ripening process) of wild species, such as S. pennellii were under strong selection and were substantially different to that of in the cultivar species S. lycopersicum. Based on the eQTL analysis of the 76 introgression lines (ILs) from S. pennellii in the background of S. lycopersicum, Ranjan et al. (2016) identified important genetic regulators of leaf development on chromosomes 4 and 8. The above-mentioned studies demonstrate the power of comparative transcriptomics in combination with ILs; however, limited studies have been carried out using their integrated approach to mine the genes involved in tomato fruit ripening.
As a distant relative of the cultivated tomato S. lycopersicum, S. pennellii has many substantially different phenotypes with the cultivated tomato and one of these is the mature fruit morphology. The mature fruit of S. lycopersicum is red and soft while the mature fruit of S. pennellii is green and hard, which renders this pair the ideal parents to cross and illustrate the genetic landscape of fruit ripening. The core set of 76 S. pennellii ILs, which represent whole-genome coverage of S. pennellii in overlapping segments in the background of M82, have been widely used to identify the key genes of many traits such as yield and metabolic composition (Semel et al., 2006;Alseekh et al., 2013Alseekh et al., , 2015. In the present study, to identify key candidates regulating tomato fruit ripening, an integrated comparative transcriptomics and eQTL approach was taken utilizing S. pennellii ILs (Eshed and Zamir, 1995). We isolated 16 candidates and provided primary validation of eight of them as being involved in the ripening process via the virus-induced gene silencing (VIGS) method. Following this screen, one candidate, SlWD40, was taken for further study. For this candidate stable RNAi and overexpression (OE) lines were generated and characterized. The OE of SlWD40 promoted ripening while its inhibition inhibited it. The co-expression networks, metabolome and transcriptome analysis indicated that SlWD40 acted as a positive regulator of tomato ripening with the key ripening TFs such as RIN, NOR, AP2a, and SlWRKYs. These results are discussed within the context of their implications regarding fruit ripening as well as with respect to the utility of genomic information in filling our knowledge gaps in important biological processes.

Results
Integrating comparative transcriptomics and eQTL mapping to mine for genes involved in tomato ripening Our earlier work described a high-quality genome assembly of the parents of the Solanum pennellii IL population as well as identifying candidate genes involved in salt as well as drought stress tolerance (Bolger et al., 2014). Surprisingly, the open-reading frame sequence of most well-characterized ripening-related genes is identical between S. pennellii and S. lycopersicum. We therefore thought to try a comparative transcriptomics approach of the S. pennellii IL population since studies on fruit gene expression of a subset of the ILs has proven highly informative (Baxter et al., 2005;Alseekh et al., 2015) as well as in leaves (Chitwood et al., 2013). For this purpose, as an initial approach, transcriptome data for M82 and S. pennellii mature fruits were sorted as follows: the total 34,727 genes in transcriptome sorted into two different data sets named as Lyco and Penn ( Figure 1A) based on the ratio of their expression values. The Lyco data set contained genes that are highly expressed in M82 (13,521), while the Penn data set contained genes that are highly expressed in S. pennellii (11,781) ( Figure 1A). For the Lyco genes which we reasoned would be more likely to harbor genes underlying the "red" ripe phenotype of cultivated tomatoes, around 300 candidates could be narrowed down by using three independent filters. Firstly, we chose to focus on genes for which expression was at least five times higher in S. lycopersicum with respect to S. pennellii. Secondly, based on the transcriptome profiling of red ripe fruit from S. lycopersicum (M82) parent and a set of lines with distinct introgressed S. pennellii segments (http://ted.bti.cornell.edu/ cgi-bin/TFGD/array_data/home.cgi), large numbers of specific eQTL and nonspecific eQTL have been identified as the former definition that specific eQTL candidates are the candidates whose expression are sharply (exponentially) increased or decreased in its located IL compared with other 75 ILs while nonspecific eQTL candidates' expression does not show such specificity and have expression in all ILs. With the specific eQTL and nonspecific eQTL information, the candidates of Lyco data set were classified into 119 specific and 223 nonspecific eQTLs ( Figure 1B). For the Penn data set, around 105 specific and 202 nonspecific candidates were classified ( Figure 1C). Here, the high number of nonspecific eQTL is attributed to the epistatic interactions between S. pennellii alleles and M82 alleles or, alternatively, the presence of a large number of trans-QTL as previously reported for leaf expression analysis (Chitwood et al., 2013) and fruit enzyme abundance analysis (Steinhauser et al., 2010). Previous studies have indicated that the candidates whose functional categories belong to transcription regulators, oxidase and cytochrome P450 may be involved in regulating tomato fruit ripening and secondary metabolism; therefore, we chose seven candidate genes that were of these three functional categories among the specific eQTL candidates.
In a parallel approach, given that TFs act as important regulators in fruit ripening, we also adopted a TF-centric approach ( Figure 1D). From the total of 34,727 genes, candidates annotated as TFs and displaying more than five times higher expression in S. lycopersicum were selected. Next, eQTL mapping thinned the list to 127 candidates which were then arranged with respect to the ratio of their expression in Breaker + 10 to that in Breaker stage. Based on putative ortholog information (Arabidopsis thaliana and S. lycopersicum) and literature survey concerning their putative functions, a final set of 20 candidates was selected. Finally, on the basis of tissue specific expression and the S. lycopersicum to S. pennellii expression ratio, the 7 candidates from eQTL approach and 20 candidates from TF approach were narrowed down to the 16 potential candidates described in Supplemental Data Set S1.

VIGS analysis of candidate genes
To provide preliminary analysis of the function of the candidate genes in tomato ripening, we carried out VIGS experiment using purple Microtom cv. tomato fruit which accumulate high amount of anthocyanin resulting from the introduction of Del/Ros1 petunia (Petunia hybrida) TFs (Orzaez et al., 2009). Partial fragments of the 16 candidate genes were cloned into pTRV2-Ros/Del/GW vector. Around 10-15 fruits per plant were infected with agrobacterium carrying the respective VIGS vector. After silencing Del/Ros1 (empty vector) in Microtom Del/Ros1 fruits, there was depletion in purple pigments but not in lycopene content due to the silenced part accumulating less purple anthocyanin pigments and thereby being easy to discriminate from nonsilenced (purple pigment rich) tissues ( Figure 2). Phenotypes were scored visually after 15 d of infection for all the 16 validated candidates (Supplemental Data Set S1).
Red color of western tomato cultivars represents the accumulation of lycopene, which is an important indicator of tomato ripening. VIGS for the structural genes encoded by Solyc01g094080 and Solyc03g095900 displayed a red phenotype, indicating that these genes are not associated with  lycopene biosynthesis or the pathways that fuel it. However, as the TFs regulating ripening are generally reported to hinder carotenoid biosynthesis, the yellowish phenotypes of Solyc11g010710 (ethylene response factors, ERF TF) and Solyc07g052700 (MADS-box TF, AGL66) VIGS fruits indicate that they may function as a ripening regulators in line with former studies that implicated SlERFs and MADS-box TFs in tomato fruit ripening (Wang et al., 2014;Liu et al., 2016; Supplemental Data Set S1 and Supplemental Figure S1). Interestingly, the VIGS fruits of a transcription regulator, SlWD40 (Solyc04g005020) also exhibited a yellowish phenotype. Given that ERF and MADS box family TFs are already well-known to be involved in tomato fruit ripening and that SlWD40 was identified as a downstream target gene of RIN (Fujisawa et al., 2013), we selected SlWD40 for in-depth analysis here ( Figure 2).

Co-expression network and VIGS of SlWD40 confirmed its role in tomato fruit ripening
In order to analyze the function of SlWD40 on fruit ripening, we initially identified its potential regulators following cisregulatory element analysis of the promoter of SlWD40. This analysis indicated that the promoter contained several ethylene (AP2, B3, EIN3, and EIL) and ripening-related elements (C2H2, MADS, NF-YB, NF-YA, and NF-YC) in the 1-kb promoter region, which indicated that it may well be induced by the ripening and ethylene burst (Supplemental Figure  S2). Moreover, the evaluation of publicly available expression data with tissue-specific expression analysis of SlWD40 confirmed the hypothesis that SlWD40 was only slightly expressed in the leaf, bud, flower, root, and young fruit but that its expression increases exponentially following mature green stage (Tomato Genome Consortium, 2012). Intriguingly, its expression in different cell types of the tomato fruit revealed that it is highly similar to that of the other known ripening regulators, such as RIN and NOR (Shinozaki et al., 2018).
Moreover, given that assembly of co-expression networks is an efficient method to identify the important interactions and relationship among different genes (Mutwil et al., 2011), available transcriptome data of different organ and fruit development stages were used to construct tomato coexpression network (Figure 3 and Supplemental Table S1; Tomato Genome Consortium, 2012). The co-expression subnetwork containing SlWD40 included 171 structure genes/ regulators, which are involved in chlorophyll and carotenoid metabolism as well as tomato fruit ripening and cell wall metabolic pathways. Among the 171 genes, a total of 62 genes exhibited high co-expression phenotype (jCo-expression Coefficientsj 40.6, P 5 0.05) with SlWD40 (Supplemental Table S1). Consistent with the results of cis-regulatory element analysis, three MADS TFs, including RIN, two AP2s TFs, and one ARF TF were significantly positively co-expressed with SlWD40. Moreover, another vital ripening-related TF, NOR, exhibited a co-expression coefficient of 0.86 with SlWD40 (Supplemental Table S1). Besides the master ripening-related TFs, SlWD40 also highly coexpressed with key carotenoid-related genes (PSY1), as well as ethylene (ACS4) and abscisic acid (NCED3) biosynthesis genes and cell wall modification genes (PL and PMEI) (Supplemental Table S1). All of these results indicate that SlWD40 may act in concert with the better characterized ripening TFs to regulate the ripening processes, including those dependent on changes in pigmentation, hormone levels, and signaling and cell wall modification.

SlWD40 affects the tomato fruit transcriptome
To confirm the accuracy of our approach and to assess in detail the function of SlWD40 in the tomato ripening process, we chose the fruit specific patatin B33 promoter which has been widely used for fruit specific expression to carry out the stable transformation (Rocha-Sosa et al., 1989;Vallarino et al., 2020). T0 transformants of RNAi and OE lines were characterized by NPT-II-specific polymerase chain reaction (PCR). Real-time quantitative polymerase chain reaction (RT-qPCR) was also carried out using fruit samples from promising T0 transgenics to select high OE and knockdown lines to raise T1 generation (Supplemental Figure S3). Fruits from all generations were analyzed and phenotype was stable over T0 and T1 generations. Before the T1 generation plant transplant to soil, we also used the NPT-II-specific PCR to confirm that the plants are transgenic. Based on the expression of SlWD40, two independent lines of RNAi (RNAi-1 and -2 lines) and OE (OE-1 and -2 lines) were chosen for subsequent experiments (Supplemental Figure S3).
To analyze fruit phenotype at the identical stage, fruits of each genotype were labeled upon anthesis and harvested for phenotyping, transcriptome, and metabolite profiling at mature green (MG, 34 DPA), breaker (Br, 37DPA), and pink (Pink, 45 DPA) stages of the wild type (WT). As seen in Figure 4, the development and ripening process were substantially hindered in the RNAi fruit while the ripening process was significantly accelerated in comparison to the OE lines. Moreover, especially at the Br stage of WT fruits, the size of RNAi fruits was significantly smaller than that of the OE and WT lines and the RNAi fruits were still at the mature green stage while the OE fruit were almost at the pink stage. The contents of chlorophylls and carotenoids, some of the most important parameters of fruit ripening, also confirmed the positive function of SlWD40 on tomato ripening process: The degradation of chlorophylls and synthesis of the predominant carotenoid, lycopene, were significantly hindered in RNAi fruits but accelerated in the OE fruits ( Figure 4B).
In order to estimate the effect of SlWD40 on the global difference of gene expression during the different fruit developmental stages, we additionally analyzed the differentially expressed genes (DEGs) among the RNAi, OE lines, and WT fruit at MG, Br, and Pink stages. For this purpose, we used FPKM (fragments per kilobase per million mapped fragments) and identified genes with jlog 2 (fold change) j 5 1 and false discovery rate (FDR) (corrected P value) 5 0.05 (Supplemental Table S2). Firstly, we checked the SlWD40 expression among the different genotypes at the MG stage. Given the low expression level of SlWD40 of WT fruit at MG stage, RNAi fruit did not exhibit significantly different expression from WT fruit. However, as the ripening process was initiated, the expression of SlWD40 was significantly induced and its expression was remarkably lower in the RNAi fruit compared with that of WT fruit at Br and Pink stages without affecting the expression of other WD40 family genes (Supplemental Table S2). In the OE fruits, SlWD40 expression was 5.64-and 5.71-fold higher than that of WT fruit at the MG stage. That said owing to the massive induction of the endogenous SlWD40 expression, the OE effect of B33 promoter was concealed and SlWD40 expression was not significantly different between the OE and WT fruit at the Br and Pink stages (Supplemental Table S2). These results were further confirmed by the principal component analysis (PCA) and cluster analysis based on the transcriptome data of different samples. RNAi samples were closely grouped with WT at MG stage but substantially separated samples at Br and Pink stages. Conversely, OE samples were clustered with WT sample especially at the Pink stage and subsequently separated from the WT sample at the MG stage  ( Figure 5, A and B). In order to further mine the important DEGs under the effect of SlWD40, we further analyzed the overlapping DEGs of OE-WT fruit at MG stage and RNAi-WT fruits at Br and Pink stage. The results indicate that 244 genes were stably downregulated in the OE-MG and upregulated in the RNAi Br and Pink stages, while 60 genes were stably upregulated in the OE-MG and downregulated in the RNAi Br and Pink stages (Supplemental Table S3). To further mine the functional categorization of DEGs, AgriGO v2.0 analysis tools (http://bioinfo.cau.edu.cn/agriGO/) by singular enrichment analysis has been used based on the conserved DEGs Supplemental Table S4). Among the GO terms included in the "Molecular Function" category of the DEGs upregulated in the OE-MG and downregulated in the RNAi Br and Pink stages, the pathways that affected the lysis and enzyme activity, such as lyase activity (FDR = 0.0014), oxidoreductase activity (FDR = 0.011), and monooxygenase activity (FDR = 0.013), were enriched. Moreover, in the "Biological Process" category of DEGs downregulated in the OE-MG and upregulated in the RNAi Br and Pink stages, several cell-wall-related pathways were significantly enriched (Supplemental Table S4). Interestingly, in the SlWD40-OE lines, the expression of SlWRKY75 SlWRKY6 (Solyc02g080890.3.1), SlWRKY17 (Solyc07g051840. 4.1), SlWRKY31 (Solyc06g066370.4.1), and SlWRKY79 (Solyc02g072190.4.1) was increased by 5.7-, 3.4-, 2.6-, 2.6-, 2.4-, 2.2-, 2.2-, and 1.8-fold, respectively. Moreover, since fruit size of SlWD40-RNAi fruits was smaller and IAA content directly affects organ size, we found that the expression of SlGH3-2 (gene regulating auxin homeostasis) was increased by three-fold in OE lines while the level of the same gene was decreased in RNAi lines by three-to six-fold.
Additionally, given that SlWD40 was strongly co-expressed with RIN and NOR, the conserved DEGs of rin, nor mutants, and SlWD40 transgenic fruits were analyzed and 31 genes were found as conserved DEGs across the genotypes ( Figure 5C and Supplemental Table S5; Fujisawa et al., 2012;Gao et al., 2020). Among the 31 genes, several important ripening-related genes, such as ACS4 for ethylene biosynthesis, PSY and Z-ISO for carotenoid biosynthesis, PL and PMEI for cell wall modification, INV for sugar metabolism, and branched chain amino transferase 2 (BCAT2) and THA1 for amino acid metabolism, were significantly downregulated in the SlWD40-RNAi, rin, and nor fruits and upregulated in the SlWD40-OE fruits, which indicates that SlWD40 may participate in some or all of the primary regulatory functions of these important regulators, such as RIN and NOR. As the transcriptome data indicated that SlWD40 exhibited substantially changed expression in the MG stage of two OE lines and Br and Pink stage of two RNAi lines compared with that of in the WT, to further confirm the expression difference, we checked the expression of SlWD40 and other 17 ripening-related genes such as RIN, NOR, PSY, and PL in these samples. The results indicate that all of these genes were significantly upregulated in the two OE lines at the MG stage but significantly downregulated in the two RNAi lines at Br and Pink stages, which further confirm the positive function of SlWD40 on tomato ripening with the strongly co-expressed TFs (Figure 6).

Metabolome analysis of SlWD40 transgenic fruits
Given that metabolites are important parameters for estimating progression of the fruit ripening process , we analyzed their levels in the different genotypes at a time concurrent to the MG, Br, and Pink stages of WTs. To obtain the global metabolome variation of different samples, a PCA analysis was carried out based on the primary metabolites, lipids, and secondary metabolites. The results indicate substantial differences among RNAi, OE, and WT fruits, which were particularly prominent at the Pink stage. RNAi pink fruits were closest to Br fruits and separated clearly from OE and WT fruits (Figure 7, A and B). Given that SlWD40 was strongly co-expressed with RIN and NOR, we analyzed the overlapping differentially enriched primary metabolites between SlWD40-RNAi lines, rin and nor mutants at the Pink stage . The results indicated that as the principal free amino acid that most contributes to the "umami" flavor of ripe tomato fruits, glutamate exhibited a conserved lower accumulation in the SlWD40-RNAi lines, rin and nor pink fruits ( Figure 7C). In detail, given that several amino acids are increased and organic acid decreased during WT tomato ripening , it is of interest that we observed that aspartic acid, glutamic acid, and tryptophan levels were significantly lower in both RNAi lines compared with those exhibited by the OE and WT pink fruits (Figure 8 and Supplemental Table S6). Moreover, nicotinic acid and glyceric acid levels were significantly higher in at least one RNAi line compared with those observed in OE and WT pink fruits (Figure 8 and Supplemental Table S6). Moreover, the former results indicated that the representative lipid components, triacylglycerols (TAGs) are significantly decreased in avocado fruit ripening (Rodriguez-Lopez et al., 2017). The same trends also have been found in the present study that the content of TAG 48:0, TAG 52:5, TAG 52:6, TAG 54:6, TAG 54:7, and TAG 54:8 was remarkably higher in the RNAi fruits than that of OE and WT pink fruits (Figure 9 and Supplemental Table S6). As the secondary metabolites (such as naringenin chalcone, naringenin hexose, and chlorogenic acid derivatives) dramatically accumulate during ripening and represent an important quality of fruit, the significantly higher and lower content of them in the OE MG fruits and RNAi pink fruits compared with that of WT fruits, respectively, further confirmed the positive influence of SlWD40 on the fruit ripening process (Figure 9 and Supplemental Table S6).

Discussion
We adopted a computational approach to mine key candidate genes involved in tomato fruit ripening by integrating comparative transcriptomics and eQTL analysis. In doing so, we identified 16 previously uncharacterized candidate genes for involvement in ripening. Among them, we chose to follow up on SlWD40 since it was identified in the ChIP seq screen as a direct target gene of RIN and relatively little is known concerning its regulatory role (Fujisawa et al., 2013). Intriguingly, cross-tissue co-expression networks for SlWD40 strongly suggest that it may act with the other important ripening regulators, such as RIN and NOR in affecting tomato fruit ripening, and the detailed analysis of DEGs and Integrating comparative transcriptomics with the eQTL approach to mine for ripening regulators Several tomato genes with strong ripening phenotypes have been identified via mutagenesis-based breeding programs including rin, nor, NR, and Cnr (Tigchelaar, 1973;Lanahan et al., 1994;Vrebalov et al., 2002;Manning et al., 2006). Moreover, the recombinant inbred line (RILs) population is also useful to identify key loci regulating quantitative traits including fruit ripening as well as aroma, color, and disease resistance (Kimbara et al., 2018). However, since RILs usually harbor more than one introgression, the possibility of the introgressed loci having beneficial/inhibitory interactions with the genes in the genetic background renders gene function elucidation more complex in RILs. In contrast, ILs are high resolution in that they normally carry only single introgressions and as such epistatic interactions masking single gene effects are largely minimized (Ofner et al., 2016). Here, we used available transcriptome data from red ripe fruit of S. lycopersicum (M82) parent and a set of lines with distinct introgressed S. pennellii segments (http://ted.bti.cor nell.edu/cgi-bin/TFGD/array_data/home.cgi) and optimized a candidate gene filtration pipeline. Moreover, on the basis of the distance between QTL and the target transcript, eQTL can either be classified as cis-eQTL (where the gene encoding the transcript resides within the QTL interval) or trans-eQTL with both types being prominent in tomato (Rockman and Kruglyak, 2009;Zhu et al., 2018). Combining the gene filtration pipeline and the comparative transcriptomics, we filtered genes responsible for the red ripe phenotype of tomato fruits (i.e. genes highly expressing only in "red" ripe fruits). Since gene duplication and subsequent functional diversification creates novel metabolic pathways and regulation, we focused on TF families that were evolved by gene duplication and already reported to be regulating tomato fruit ripening (e.g. MADS box and basic helix loop helix (bHLH) TF families) (Hileman et al., 2006;Waseem et al., 2019). Out of the 16 candidate genes, 10 were found to be generated either through tandem or block duplication (https://bioinformatics.psb.ugent.be/plaza). In the VIGS experiment, green and yellow phenotypes have been obtained for the candidates Solyc11g010710 (AP2 like) and Solyc07g052700 (MADS TF, AGL66). Our previously published work implicated AP2a in regulating tomato fruit ripening via regulation of ethylene biosynthesis and signaling (Chung et al., 2010;Karlova et al., 2011). MADS box TFs such as SlCMB1 (Solyc04g005320), TAGL1 (Solyc07g055920),

Figure 9
The difference of representative lipids and secondary metabolites of T1 transformants of RNAi and OE lines compared with WT fruits. The values in each column are the mean of at least three biological replicates. Error bars indicate SD. The asterisks indicate statistically significant differences determined by the Student's t test (two-tail): *P 5 0.05; **P 5 0.01. and the canonical TF RIN (Solyc05g012020) have previously been reported as positive regulators of tomato fruit ripening . Therefore, VIGS for AP2 like and MADS box candidates (Solyc11g010710 and Solyc07g052700) acted as positive control supporting the efficacy of our approach. Moreover, among the 16 genes, VIGS for two bHLH TFs (Solyc03g044460 and Solyc12g098620) showed green and light red phenotypes (Supplemental Figure S1). Here, light red phenotype for Solyc12g098620 is in line with the recently published work by D' Amelia et al. (2019) in which the authors reported that this bHLH TF regulates carotenoid biosynthesis. Additionally, one of the candidates, Solyc12g010950 (alcohol dehydrogenase), obtained a whitish green fruit phenotype, and another candidate, SlWD40, obtained a yellowish fruit phenotype (Figure 2 and Supplemental Figure S1). While the link between the alcohol dehydrogenase and ripening is currently unclear, that for SlWD40 is not without precedence since it is a transcriptional regulator which has been linked to rin and nor in ChIP experiments but not characterized in detail in its own right.
The eQTL for SlWD40 can be defined as a cis-eQTL since SlWD40 is located in the region which has been introgressed by S. pennellii chromosomal architecture in the IL4-1 and IL4-1-1 and SlWD40's expression is only dramatically lowered for IL4-1 and 4-1-1 (Supplemental Table S7). At the onset of fruit ripening process (MG stage), SlWD40 is moderately expressed in S. lycopersicum (RPKM mean value 31) but negligibly expressed in S. pennellii (RPKM mean value 0.17) and consistent with the ripening stage, the expression of SlWD40 was substantially induced in concert with fruit ripening. Considering the dramatically lowered SlWD40 expression in IL4-1 and 4-1-1 and substantial difference of the red/green colored fruits of S. lycopersicum and S. pennellii, respectively, we hypothesized that SlWD40 may participate in the S. lycopersicum ripening process. Furthermore, VIGS for SlWD40 in MicroTom inhibited normal red coloration. These results demonstrate the utility of our candidate gene filtration pipeline integrating comparative transcriptomics with eQTL in identifying candidate ripening genes.

SlWD40 is an important regulator of tomato fruit ripening
Among the eight genes which we validated by VIGS, the information concerning the role of the WD40 family in regulating tomato ripening is the most limited. Ripening function of this gene was further confirmed by the stable OE and RNAi transformation (Figures 4 and 5). WD40 proteins contain a signature WD (Trp-Asp) dipeptide and 40 amino acids in single repeats that then fold into fourstranded anti-parallel b-propeller sheets and are highly promiscuous interactors, being both platforms for protein-DNA and protein-protein interactions (Xu and Min, 2011;Mishra et al., 2012;Chen et al., 2022). In the canonical MYB-bHLH-WD40 protein complex, WD40 acts as a recruiter and stabilizer of the MYB and bHLH protein which enhances the regulation of the complex during anthocyanin biosynthesis (Ramsay and Glover, 2005;Zhang et al., 2014). In the present study, the green phenotype of fruits of SlWD40-RNAi lines clearly demonstrated a positive role of SlWD40 in tomato fruit ripening (Figure 4). Moreover, given that the global gene co-expression analysis is a powerful approach to identify the important interactions among different genes during the development of certain organs, it is important to note that the co-expression network of SlWD40 revealed links with RIN and NOR, while transcriptome analysis indicated they shared conserved regulated genes associated with ethylene (ACSs) and carotenoid (PSY and Z-ISO) biosynthesis as well as cell wall degradation (PL and PMEI) and sugar (INV) and amino acid (BCAT2 and THA1) metabolism (Supplemental Table S5). Interestingly, based on available ChIP and transcriptome data, the RIN protein can directly bind to the promoter of SlWD40 and thereby increase its expression (Martel et al., 2011;Fujisawa et al., 2013).
Moreover, as an important TF family member involved in ethylene signal, AP2a belonging to the AP2/ERF superfamily has been shown to inhibit ethylene biosynthesis as well as positively regulate chlorophyll degradation and carotenoid biosynthesis in tomato . And in Setaria italica, SiAP2 can bind to the SiWD40 promoter to mediate abiotic stress responses (Mishra et al., 2012). In the present study, the promoter region of SlWD40 was found to harbor eight different AP2 TF binding sites and AP2a was significantly induced in SlWD40-OE-MG fruits while repressed in the WD40-RNAi-Breaker fruits (Supplemental Table S2). Additionally, it has been well-documented that several SlWRKYs regulate tomato fruit ripening and lycopene accumulation (Cheng et al., 2016;Wang et al., 2017). In our analysis, SlWD40 OE resulted in upregulation (by two-to five-fold) of a broad number of SlWRKYs (SlWRKY75, SlWRKY37, SlWRKY23, SlWRKY30, SlWRKY6, SlWRKY17, SlWRKY31, and SlWRKY79) in mature green fruits of OE lines. Moreover, SlWRKY17, which was shown to interact with RIN, SlERF2b, and SlERF7, is strongly co-expressed with ELIP2 and RIN in our analysis Supplemental Table S1). All of these results indicate that SlWD40 might act as a junction point and facilitate binding of one or more above mentioned co-expressed TFs such as RIN, NOR, AP2a, and SlWRKYs to be involved in tomato ripening process.
Besides the important ripening-related TFs, hormone signaling also plays a vital role in the ripening process. As an important regulator of auxin-ethylene homoeostasis which affects fruit ripening (Kumar et al., 2012;Sravankumar et al., 2018), the expression of SlGH3-2 (Solyc01g107390) increased in SlWD40-OE lines by three-fold while decreased in RNAi lines by three-to six-fold ( Figure 6 and Supplemental Table  S2). Combined with the induction and repression effect of ACS4 in the SlWD40-OE and -RNAi fruit, respectively, these results collectively indicate that SlWD40 may also affect the tomato ripening process through the regulation of the ripening-related hormone homoeostasis. However, considerably further experimentation will be required in order to test this hypothesis.

Conclusion
Our study aimed at better understanding the molecular mechanisms underlying tomato fruit ripening. Comparative transcriptomics of small green fruited wild species with red fruited S. lycopersicum alongside eQTL mapping allowed the identification of key candidate genes involved in tomato fruit ripening. Utilizing co-expression networks alongside detailed metabolome and transcriptome analysis indicated that SlWD40 has a positive impact on tomato ripening process and suggest that it may act in concert with strongly co-expressed TFs such as RIN, NOR, AP2a, and SlWRKYs ( Figure 10). Beyond these insights into ripening, we believe our study also acts as a proof-of-concept study whereby the transcriptome of phenotypically divergent wild relatives, alongside eQTL mapping, can be used to identify causal genes underlying trait variance.

Narrowing down candidate genes involved in fruit ripening
For M82 and S. pennellii (Penn) fruit, RNA-seq data are available (Bolger et al., 2014). Moreover, RNA-seq data for S. pennellii ILs fruit were also available (http://ted.bti.cornell. edu/cgi-bin/TFGD/array_data/home.cgi). Starting with the RNA-seq data for M82 and S. pennellii fruit (Bolger et al., 2014) all 34,727 genes in the transcriptome were sorted in two ways. Firstly, the ratio of their expression value in M82 relative to that in S. pennellii and secondly, by ratio of their expression value of S. pennellii to that of M82 in order to get genes that are highly expressed in M82 and S. pennellii, respectively. Candidates from these two lists were further filtered by using three different criteria namely (1) that the relative fold change (FC) in the expression was at least 45, (2) the presence of an eQTL for the gene, and (3) the presence of a functional annotation. Subsequently, these candidates were sorted into specific and nonspecific eQTL based on their expression in respective IL. Here, specific and nonspecific eQTL were defined based on the expression of a particular S. pennellii candidate in a specific IL or several ILs, respectively. Specific eQTL candidates were then focused and classified based on their function. For promoter analysis, the 1-kb upstream sequence from the start codon of SlWD40 was retrieved from the SGN tomato genome web browser and the cis-regulatory elements were analyzed using PlantCARE (http://bioinformatics.psb.ugentbe/webtools/ plantcare/html/) and PlantPAN 2.0 (http://plantpan2.itps. ncku.edu.tw/index.html) web tools.
SlWD40 co-expression network construction with tomato ripening pathway genes We listed 171 target genes involved in carotenoid biosynthesis, tomato fruit ripening, and cell wall metabolic pathways and their regulation. Some of these genes were well characterized. For all 171 genes, expression values were extracted from Tomato Genome Consortium (2012). The R script written by Contreras-Lopez et al. (2018) was used to calculate correlation values and P values, both positive and negative correlation values were calculated and cytoscape was used to visualize network (Shannon et al., 2003).

Virus-induced gene silencing
Vector construction, infiltration, and fruit harvesting procedures were performed as previously described (Orzaez et al., 2006(Orzaez et al., , 2009. Briefly, an approximately 300-bp fragment of the candidate gene was amplified from tomato M82 fruit cDNA using gateway compatible primers and recombined into the GATEWAY vector pDONR207 (Invitrogen, http:// www.invitrogen.com/) by the BP reaction following the manufacturer's protocol to generate an entry clone. An error-free entry vector was confirmed by sequencing and then Figure 10 Proposed schematic overview of network of regulatory factors controlling tomato fruit ripening. recombined with the pTRV2-Ros/Del/GW destination vector using an LR reaction to produce the expression clones pTRV2-Ros/Del/GW-Respective Gene ID. Agrobacterium tumefaciens strain GV3101:pMP90 was then transformed with sequenced expression vectors by electroporation. In order to infiltrate fruit for VIGS, purple MicroTom tomato was used and agroinfiltration was performed as previously described (Alseekh et al., 2015).

Development of OE and RNAi lines
The sequence encoding Solyc04g005020 was amplified from S. lycopersicum cv. Moneymaker (MM) cDNA by using genespecific primers and inserted into the pDONR207 by attB recombination to generate entry clone. Primer sequences are provided in Supplemental Table S8. An error-free entry clone was confirmed by sequence analysis before recombination into destination vector B33BinAR for fruit-specific OE and named as B33BinAR_SlWD40. Additionally, artificial miRNA (amiRNA) cassette was designed for Solyc04g005020. For this, Solyc04g005020 cDNA sequence was used as target sequence, employing the WMD3 program (http://wmd3.wei gelworld.org/cgi-bin/webapp.cgi) to design corresponding amiRs. An overlapping PCR (polymerase chain reaction) strategy was employed with in-hand precursor DNA, following the WMD3 protocol (http://wmd3.weigelworld.or/down loads/CloningofartificialmicroRNAs.pdf). The pre-amiRs obtained from overlapping PCR (using the athmir-319a backbone) were cloned into the pENTR/D-TOPO vector and the clones were confirmed by DNA sequencing. Subsequently, these sequences were cloned into B33BinAR via Asp718 and BamHI digestion and cohesive end ligation. Primer sequences are provided in Supplemental Table S8. This and other final LR plasmids were then introduced into A. tumefaciens strain GV2260 by electroporation and subsequently submitted for transformation into MM plants using the leaf disc transformation method (McCormick et al., 1986).

Plant material and growth conditions
Transgenic plants for each genotype were selected on kanamycin containing MS medium (50 mgÁL -1 ). SNN and MM (WT) were germinated on MS medium without kanamycin. Both transgenic lines and WT were selected and transferred to soil pot for cultivation under long-day conditions (16-h/ 8-h day/night cycle) at 22 C and 50% humidity, as described previously in the literature (Carrari et al., 2003). Upon anthesis, flowers were labeled with that particular date.

Transcriptome analysis
Two biological replicate samples from two independent plants of each genotype of MG, Br, and Pink stages have been harvested. Total RNA was extracted using the NucleoSpin RNA Plant kit (Macherey-Nagel) and sent to the Novogene Company (Beijing, China) for Illumina HiSeq PE150 sequencing. The cDNA library was constructed following the manufacturer's recommendations and then purified to remove the low-quality sequences. The clean data are available in Zenodo (https://zenodo.org/) (doi:10.5281/zenodo.5525948 and 10.5281/zenodo.5525946). The RNA seq data were analyzed using LSTrAP (Proost et al., 2017). The clean reads of each sample were aligned to the Tomato Genome version SL4.0 and Annotation ITAG4.0 (ftp://ftp.sol genomics.net/tomato_genome/annotation/ITAG4.0_release/). The DEGs between transgenic fruit and WT fruit were identified under the parameter of FC 5 2 and FDR 5 0.05.

Metabolic profiling
Fruit pericarp samples were harvested, immediately frozen in liquid nitrogen, and stored at -80 C until further analysis. Samples were then powdered by using retsch mill at 30 LÁs -1 , for 30 s. Extraction of pigments, primary metabolites, lipid, and secondary metabolites was performed as described previously (Salem et al., 2016). In brief, 500 mL of the upper lipid and pigments containing phase was dried in a SpeedVac concentrator and resuspended in 250 mL acetonitrile: 2-propanol (7:3, v/v) solution. Two microliters of the solution were analyzed by the Waters Acquity ultra-performance LC system coupled with Fourier transform MS in positive ionization mode. Moreover, 150 and 300 lL of the polar phase were dried in a centrifugal vacuum concentrator for primary and secondary metabolite profiling. The primary metabolite pellet was resuspended in 40 lL of methoxyaminhydrochloride (20 mgÁmL -1 in pyridine) and derivatized for 2 h at 37 C. Afterward, 70 lL of N-methyl-N-[trimethylsilyl] trifluoroacetamide was added containing 20 lLÁmL -1 fatty acid methyl esters mixture as retention time standards. The mixture was incubated for 30 min at 37 C at 400 rpm. A volume of 1 lL of this solution was used for injection. The gas chromatography-mass spectroscopy system comprised a CTC CombiPAL autosampler, an Agilent 6890N gas chromatograph, and a LECO Pegasus III time of flight mass spectrometry (TOF-MS) running in EI + mode. The secondary metabolite pellet was resuspended in 200-mL 50% (v/v) methanol in water and 2 mL was injected on RP high strength silica T3 C 18 column using a Waters Acquity UPLC system. The analysis workflow included peak detection, retention time alignment, and removal of chemical noise following the method of Salem et al. (2016). For metabolites and transcriptome data processing, the PCA and heat map analysis were performed by MetaboAnalyst 5.0 (https://www. metaboanalyst.ca/).

RT-qPCR analysis
Total RNA was extracted from fruit using TRIzol reagent (Invitrogen, Waltham, MA, USA). And the first-strand cDNA synthesis was carried out as the manufacturer's instructions of PrimeScript RT Reagent Kit with gDNA Eraser (Takara, Shiga, Japan). RT-qPCR was analyzed on an ABI Prism 7900 HT real-time PCR system (Applied Biosystems/Life Technologies, Darmstadt, Germany) in 384-well PCR plates. The RT-qPCR data were analyzed using the 2 -DDCt analysis method according to Bustin et al. (2009) and all primers are listed in Supplemental Table S8.

Statistical analysis
Student's paired t test was performed to assess whether the differences between different genotypes were statistically significant. The asterisks indicate statistically significant differences determined by the Student's t test (two-tail): *P 5 0.05; **P 5 0.01.

Accession numbers
Sequence data from this article can be found in the GenBank/EMBL data libraries under accession numbers SlWD40, Solyc04g005020.

Supplemental data
The following materials are available in the online version of this article.
Supplemental Figure S3. Genotyping of the SlWD40 of T0 transformants.
Supplemental Table S1. Co-expression network of SlWD40 with tomato ripening pathway specific genes.
Supplemental Table S3. The overlapped DEGs of SlWD40 OE and RNAi fruit.
Supplemental Table S4. Functional categorization of DEGs of SlWD40.
Supplemental Table S5. The overlap DEGs of SlWD40, rin, and nor mutants.
Supplemental Table S7. Expression of SlWD40 in ILs. Supplemental Table S8. Primer sequences used in this study.
Supplemental Data Set S1. Finalized potential candidates from both eQTL and TF approaches and their VIGS phenotypes.