Metabolomics Combined with Transcriptomics Analysis Revealed the Amino Acids, Phenolic Acids, and Flavonol Derivatives Biosynthesis Network in Developing Rosa roxburghii Fruit

Rosa roxburghii Tratt. is a specific fruit with high nutritional value and antioxidative activities. However, the key metabolites and their biosynthesis are still unknown. Herein, a main cultivated variety, ‘Guinong 5’ (Rr5), was chosen to analyze the metabolomics of the three developmental stages of R. roxburghii fruit by liquid chromatography–tandem mass spectrometry (LC-MS/MS). A total of 533 metabolites were identified, of which 339 were significantly altered. Total phenols, flavonoids, and amino acids were significantly correlated to at least one in vitro antioxidant activity. The conjoint Kyoto Encyclopedia of Genes and Genomes (KEGG) co-enrichment analysis of metabolome and transcriptome was focused on amino acid, phenylpropanoid, and flavonoid biosynthesis pathways. The amino acid, phenolic acid, and flavonol biosynthesis networks were constructed with 32 structural genes, 48 RrMYBs, and 23 metabolites. Of these, six RrMYBs correlated to 9–15 metabolites in the network were selected to detect the gene expression in six different R. roxburghii genotypes fruits. Subsequently, 21 key metabolites were identified in the in vitro antioxidant activities in the fruits at various developmental stages or in fruits of different R. roxburghii genotypes. We found that four key RrMYBs were related to the significantly varied amino acids, phenolic acids, and flavonol derivatives in the network during fruit development and the key metabolites in the in vitro antioxidative activities in the fruits of six R. roxburghii genotypes. This finding provided novel insights into the flavonoid, polyphenol, and amino acid synthesis in R. roxburghii.


Introduction
Rosa roxburghii Tratt. is a newly popular rosaceous plant with uniquely flavored fruits and a high nutritional and medical value. The fruits are rich in L-ascorbic acid (AsA) and organic acids, amino acids, phenols, flavonoids, superoxide dismutase (SOD), and triterpenoids, among which total phenols, flavonoids, AsA, and triterpenoids contribute >80% to the antioxidant activity [1][2][3][4][5]. Jiang et al. identified 723 metabolites in six R. roxburghii genotypes by widely targeted metabolomics and showed that flavonoids, triterpenoids, and phenolic acids have an antioxidant capacity [6]. The flavonoids in R. roxburghii fruit also exhibit critical biological activities, such as anti-inflammation [7], the expansion of coronary arteries, the control of blood pressure, and the protection of blood vessels [8]. Polyphenols from R. roxburghii ameliorate the symptoms of diabetes by activating insulin [9]. Because of its nutritional and medical functions, the economic cultivation area of R. roxburghii has expanded to 140,000 ha in the Guizhou province. Several enterprises have participated in processing R. roxburghii, and >50 products have been explored.

Determination of Bioactive Substance Content and Antioxidant Capacity
The total phenol content was determined using the Folin-Ciocalteau reagent [20]. The flavonoid content was analyzed using the spectrophotometric method, as described by Huang et al. [21]. The AsA was determined according to the method of An et al. [1]. Triterpenoids were determined by vanillin-glacial acetic acid colorimetry [22]. The levels of organic acids and lipids were measured as described in the Principles and Techniques of Plant Physiological Biochemical Experiment [23]. The amino acid content was determined using Suzhou Greis Biotechnology Co., Ltd (Suzhou, China). kits. In order to determine the DPPH (1,1-Diphenyl-2-picrylhydrazyl Free Radical) radical-scavenging activity, the protocol by Andrés et al. [24] was employed with minor modifications. Martinez et al.'s [25] method was used to determine ABTS (2,2 -Azinobis-3-ethylbenzthiazoline-6-sulphonate) cation radical-scavenging activity, and Benzie and Strain's method [26] was applied to determine the ferric ion-reducing activity. The antioxidant activity was expressed in µmol Trolox equivalents (TE)/g fresh mass.

Metabolite Extraction and ESI-Q TRAP-MS/MS Analysis
The frozen fruits were crushed with a churn (1.5 min, 30 Hz, three repetitions, 400 mm, rage). Then, an equivalent to 100 mg powder samples was extracted using a 70% methanol solution containing 0.1 mg of lidocaine at 4 • C overnight. The supernatant obtained by centrifugation of the extract at 10,000× g, 4 • C for 10 min, was filtered through a 0.2286-micron hydrophilic Teflon syringe filter (SCAA-104, Amperale, Shanghai, China) and subjected to metabolomics analysis. The quality control samples were injected every two samples (mixtures 1-3) to obtain a dataset, and the repeatability was evaluated.
The mass data were acquired in the electrospray ionization-positive/negative mode using the following parameters: ion spray voltage 5.5 kV; 55 pounds/square inch ion-derived gas iodine (GSI); 60 pounds/square inch gas II (GSII); 25 pounds/square inch curtain gas; 550 • C temperature of the turbine spray; instrument tuning and mass calibration with 10 and 100 µmol/L polypropylene glycol solution in triple quadrupole (QQQ) and linear ion trap (LIT) modes, respectively. The specific diffusion potential and collision energy optimization were carried out to assess the diffusion potential and collision energy of a single multi-reaction monitoring transition.

Qualitative and Quantitative Analysis of Metabolites
Based on the self-built metadata database (MWDB), material characterization was carried out according to the secondary spectral information. Metabolite quantification is a multi-response monitoring model (multiple reaction monitoring) that uses the mass spectrometry data of triple and four-stage rods multiple reaction monitoring (MRM). Based on fold change ≥2 (upregulated) or ≤0.5 (downregulated) and p-value < 0.05, the significantly changed metabolites (SCMs) were screened. SCM principal component analysis was performed on platform (www.r-project.org, accessed on 11 May 2020) to study the variety-specific accumulation of metabolites. Then, these differential metabolites were screened using a threshold variable importance in projection (VIP) value (VIP ≥ 0.8) from the orthogonal partial least squares discriminant analysis (OPLS-DA) model. The annotated metabolites were mapped to the KEGG pathway database (http://www.kegg.jp/kegg/ pathway.html, accessed on 13 May 2020) to determine the pathway associations. Pathway enrichment analysis was performed on the web-based server Metabolite Sets Enrichment Analysis (MSEA; http://www.msea.ca, accessed on 15 May 2020). The pathways with Bonferroni-corrected p-values ≤ 0.05 were considered significantly enriched.

Transcriptome Information and Prediction of Transcription Factors
The data on the lignin of R. roxburghii were obtained from Lu et al. [11]. The transcriptome of the three samples was sequenced at 30, 60, and 90 days after flowering. The fruits were harvested in 2017, and denoted as 30 DAA, 60 DAA, and 90 DAA, respectively. The transcription factors were identified by the predicted peptide sequences of all transcripts searched against the transcription factors in PlantTFDB 3.0 using the Transcription Factor Prediction module (http://planttfdb.cbi.pku.edu.cn/prediction.php, accessed on 14 May 2020) with default parameters. Then, the promoter elements were predicted using the Plantcare website (http://bioinformatics.psb.ugent.be/webtools/plantcare/html, accessed on 8 February 2021).

Quantitative RT-PCR Analysis
Total RNA was extracted from the fruit samples using TRIzol reagent (Invitrogen, Shanghai, China), according to the manufacturer's instructions. The primer sequences are listed in Table S6. qRT-PCR was performed on an ABI ViiA 7 DX system (Thermo Fisher Scientific, MA, USA) using SYBR Premix Ex Taq II (TaKaRa, Dalian, China) with the ubiquitin gene as an endogenous control. The data were analyzed using the 2 −∆∆CT method [27]. The mean expression and standard deviation (SD) were calculated based on the results of three independent experiments.

Statistical Analysis
The data were analyzed statistically with SPSS 25.0 (SPSS Inc., Chicago, IL, USA). Analysis of variance was used to test any difference in bioactive substance content and antioxidant activities resulting from these methods. Duncan's new multiple range test was used to determine significant differences. Correlations among the data obtained were calculated using Pearson's correlation coefficient (r). Heatmap and principal components analysis (PCA) were made using the pheatmap and factoextra packages, respectively, within R v3.5.2 (R Core Team. R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria, 2015).

Bioactive Substance Content and Antioxidant Capacity in R. roxburghii during Fruit Development
The content of organic acids, amino acids, flavonoids, total phenols, terpenoids, AsA, lipids, DPPH radical-scavenging activity (DPPH), ABTS cation radical-scavenging activity (ABTS), and the ferric ion-reducing activity (FRAP) were measured at 30 (Rr5-30), 60 (Rr5-60), and 90 (Rr5-90) DAA of R. roxburghii fruits ( Table 1). The content of flavonoids, total phenols, terpenoids, and FRAP showed a similar trend, i.e., decreased significantly with fruit development. Correlation analysis verified that the FRAP was the most positively correlated with total phenols, flavonoids, and terpenoids (p < 0.01) ( Figure 1). Similarly, the content of amino acids decreased with fruit development. The subtle difference was that there was no significant variation between Rr5-60 and Rr5-90 (Table 1). Hence, the content of amino acids also showed a significant positive correlation with the FRAP value (p < 0.05) ( Figure 1). Taken together, the content of amino acids, flavonoids, total phenols, and terpenoids was positively correlated with three in vitro antioxidative activities during R. roxburghii fruit development ( Figure 1).

Figure 1.
Correlation coefficient between active substances and antioxidant capacity. High correlation corresponds to red, and low correlation corresponds to green. When the correlation coefficient is ≥0.95, it is significant at the level of 0.05; when the correlation coefficient is ≥0.99, it is significant at the level of 0.01. In order to better understand the metabolites in the fruit development stage of R. roxburghii fruit, we performed a widely targeted LC-MS/MS-based metabolite profiling of Rr5-30, Rr5-60, and Rr5-90, respectively. Subsequently, principal component analysis ( Figure S2A) and sample correlation analysis ( Figure S2B) were carried out to evaluate the repeatability of the data. The principal component analysis showed that in the PC1×PC2 score chart, three samples related to the quality control (QC) samples were clearly separated, indicating differences in the metabolomes between the sample groups [24], while the sample correlation analysis showed high correlation among biological replicates and low correlation between different biological replicates, suggesting satisfactory data.
Subsequently, we performed a KEGG pathway enrichment analysis to identify the differences in the metabolic pathways between the three groups. The enrichment analysis suggested that the metabolic pathways and biosynthesis of secondary metabolites differed significantly in the three groups (p < 0.05) ( Figure 4). Notably, significant differences were observed in the biosynthesis of amino acids, purine metabolism, and phenylpropanoid biosynthesis between Rr5-60 and Rr5-90 ( Figure 4B). The characteristic metabolites were the intersections of different groups of various metabolites. At this stage, the characteristic metabolites need to meet the condition that the content was more than at other stages. A total of 59 characteristic metabolites were identified in Rr5-30, 6 in Rr5-60, and 120 in Rr5-90 (Table S2). These results suggested that the metabolites differed significantly between Rr5-30 and Rr5-90, while Rr5-60 may be at the middle stage of maturity. The characteristic metabolites were involved in phenolic acids, flavonoids, amino acids and derivatives, organic acids, nucleotides and derivatives, tannins, and lipids. Importantly, flavonoids and amino acids, such as dihydroquercetin (taxifolin), naringenin chalcone, L-serine (Ser), L-asparagine (Asn), L-(−)-threonine (Thr), 2,6-diaminoopimelic acid (2,4-DAP), and L-citrulline (Cit), need to be investigated in depth in Rr5-30. For Rr5-90, phenolic acids and amino acids, such as chlorogenic acid, cinnamic acid, p-coumaric acid, caffeic acid, ferulic acid, γ-aminobutyric acid (GABA), L-(+)-arginine (Arg), 5-aminovaleric acid

Conjoint Analysis of Transcriptome and Metabolome
The conjoint analysis of the transcriptome and metabolome was combined with the transcriptome data [11] to screen out the essential genes and metabolites. The KEGG enrichment analysis showed the co-enrichment pathways of DEGs and SCMs through conducted p-values ( Figure 5). The co-enrichment is involved in nine pathways, including the biosynthesis of amino acids, arginine biosynthesis, alanine, aspartate, and glutamate metabolism, glycine, serine, and threonine metabolism, cysteine and methionine metabolism, and the biosynthesis of valine, leucine, isoleucine, lysine, phenylalanine, tyrosine, tryptophan, phenylpropanoid, and flavonoids. and metabolites between the two groups. Each post represents a pathway: orange shows the transcriptome, and green represents the metabolome. The ordinate represents pathway enrichment, and the abscissa represents the target pathway. The parameter "p-value < 0.05" was used as the threshold to judge the significance of gene expression and the difference in the metabolites.

Biosynthesis of Amino Acids, Phenolic Acids, and Flavonol Derivatives
Differential expression analysis was performed using the DESeq R package (1.10.1). The resulting p values were adjusted using Benjamini and Hochberg's approach for controlling the false discovery rates. Genes with an adjusted p-value < 0.05 identified using DESeq were considered as differentially expressed genes (DEGs). We identified 48 enzymes and 74 DEGs, including 10 structural genes involved in the phenylpropanoid biosynthesis, 52 in the biosynthesis of amino acids, and 12 in flavonoid biosynthesis in the transcriptome ( Figure 6, Table S3). Moreover, two potential key structural genes were identified in the phenylpropanoid biosynthesis pathway. The gene (evm.model.Contig428.373) was strongly correlated with cinnamic acid, p-coumaric acid, caffeic acid, and ferulic acid (R 2 = 0.888; R 2 = 0.851; R 2 = 0.959; R 2 = 0.886) (  (Table S4). The results showed that these genes were crucial for the accumulation of naringenin chalcone, dihydroquercetin, and chlorogenic acid, but chlorogenic acid was negatively correlated with F3 H (evm.model.Contig401.234).
In order to further verify the reliability of the data, we randomly selected 15 genes involved in the amino acid, phenylpropanoid, and flavonoid biosynthesis. The qRT-PCR results showed that the expression of 15 genes was consistent with the RNA-seq data ( Figure S4).

RrMYBs in the Regulation of Amino Acid, Phenolic Acid, and Flavonol Derivative Biosynthesis
In order to explore the mechanism that might affect the amino acids, phenolic acids, and flavonoids in R. roxburghii fruit, we analyzed the promoter elements of the above 32 potentially critical structural genes to collect information about the transcription factors. The results showed that the MYB family regulated the biosynthesis of amino acids, phenolic acids, and flavonoid derivatives of R. roxburghii fruits. In addition, Val may be regulated by HD-Zip transcription factors (Figure 7). Based on the transcriptome, we identified 158 MYB transcription factors, of which 75 were DEGs. According to the correlation analysis, 48 DGEs were potentially involved in the regulation of the characteristic metabolites of R. roxburghii fruits. Furthermore, we analyzed the interaction network among the 32 structural genes, 48 RrMYBs, and 23 metabolites using the Cytoscape software, Institute for Systems Biology, Seattle, WA, USA ( Figure 8).
Subsequently, six RrMYBs were selected as potential key regulatory factors from the network that related to the amino acids, phenolic acids, and flavonol derivatives during R. roxburghii fruit development (Figure 8; Table S6). The relative expression of these molecules was detected in the fruits of six R. roxburghii genotypes. After data standardization, the trend clustering was carried out through the heatmap consisting of 21 key metabolites and six key RrMYBs in the fruits of six R. roxburghii genotypes ( Figure 9B). Among these, the expression of four RrMYBs showed a similar variation trend, with some key metabolites, such as evm.model.Contig361.14 (RrMYB105) and the pelargonidin-3,5-Odiglucoside, evm.model.Contig144.143 (RrRAX3) and MetO, evm.model.Contig191.1 (RrLHY), evm.model.Contig100.63 (RrREVEILLE7), and pyridoxine, indicating that these RrMYBs significantly affect various metabolites in the network during fruit development (Figure 8) and the key metabolites related to the in vitro antioxidation activities in the fruits of six R. roxburghii genotypes ( Figure 9B).

Discussion
In the present study, organic acids, amino acids, flavonoids, total phenols, terpenoids, AsA, and lipids were measured at 30 (Rr5-30), 60 (Rr5-60), and 90 (Rr5-90) days after anthesis (Table 1). Significant variation was observed at different developmental stages except for lipids. Organic acids increased from 0.58% at Rr5-30 to 1.12% at Rr5-90, AsA increased from 70.78 mg/100g at Rr5-30 to 861.13 mg/100g at Rr5-90, which was similar to the results found by An et al. [1]. Amino acids were the highest at Rr5-30, but there was no significant decrease between Rr5-60 and Rr5-90. Flavonoids, phenolic acids, and terpenoids decreased sharply in terms of fruit growth. Previous studies also showed the same results [5]. Meanwhile, we detected DPPH, ABTS, and FRAP ( Table 1). The results showed that the changes in the three antioxidant activities were different in the R. roxburghii growth, but these activities were the highest in Rr5-30. Correlation analysis verified that the FRAP was most positively correlated with total phenols, flavonoids, and terpenoids (p < 0.01) (Figure 1). The results of the computing correlations presented in Figure 1 are in accordance with previous studies on R. roxburghii fruits [5]. The content of total phenols and flavonoids was 1701 mg/100 g and 456 mg/100 g, respectively, in mature fruit, and the DPPH, FRAP, and ABTS antioxidant activity was 66.04 mmol TE/g, 141 mmol TE/g, and 66 mmol TE/g (Table 1), respectively. The results showed that the values of total phenol, flavonoids, DPPH, FRAP, and ABTS in R. roxburghii fruit were higher than common fruits rich in these bioactive substances, such as citrus [28], grape [29,30], and Lonicera caerulea [31].
Moreover, two lipids (lysoPC 20:2 and 1-linoleoylglycerol), two organic acids (4-acetamidobutyric acid and shikimic acid), and four others (4-pyridoxic acid, pyridoxine, menatetrenone (vitamin K2), and glucarate O-phosphoric acid) were selected as key metabolites in both the developing R. roxburghii "Rr-5" fruits and six other fruits of different genotypes ( Figure 9A). These were not consistent with the analysis of physiological data (Figure 1), reflecting the differences between the whole and the individual (Table 1; Figure 2). A recent study showed that lysoPC 20:2 relieves the symptoms of gastrointestinal disorder by increasing the bacteria Ruminococcaceae and decreasing Streptococcaceae, Erysipelotrichaceae, and Lachnospiraceae in the fecal microbiota [42]. In addition, 1-linoleoylglycerol showed a high level of antioxidant activity in Morchella esculenta Pers [43], while 4-acetamidobutyric acid was involved in chronic fatigue syndrome [44] and pediatric overweight and obesity [45]. Shikimic acid has antibacterial, anti-inflammatory, hair-growth-stimulating, and anti-aging effects as well as antifungal properties [46]. Menatetrenone significantly decreases undercarboxylated osteocalcin to osteocalcin and improves lumbar bone mineral density in osteoporotic patients [47]. High 4-pyridoxic acid/pyridoxine ratio is independently associated with global cardiovascular risk [48]. Metabolomics and transcriptomic analysis highlighted that Klebsiella oxytoca P620 application decreases the intensities of pyridoxine and glucarate O-phosphoric acid in PHBA-stressed leaves, and downregulates the expression of genes related to these metabolites [49]; evm.model.Contig191.1 (RrLHY), evm.model.Contig100.63 (RrREVEILLE7); pyridoxine exhibited the same variation trend ( Figure 9B), indicating that they might upregulate the expression of the genes related to pyridoxine.
The content of amino acids was significantly related to at least one in vitro antioxidant activity during "Rr-5" fruit development (Figure 1), but not related to three in vitro antioxidant activities in six R. roxburghii fruits [6]. Typically, amino acids have been reported as antioxidant substances. Fodor et al. reported that free amino acids contributed to FRAP [50].
Cit has the characteristics of a hydroxyl radical scavenger [51]. Glutamic acid and histidine decreased the generation of the hydroxyl radicals [52]. Hwang et al. demonstrated a strong antioxidant activity of methionine and lysine [53]. The results of Tian et al. showed that the electronic and hydrogen-bonding properties of the amino acids in the tripeptide sequences, and the steric properties of the amino acid residues at the C-and N-termini, play a major role in the antioxidant activities of the tripeptides. Tripeptides exhibit the highest FRAP to contain Cys (C) and Trp (W) residues [54]. Finally, in this study, in both the developing R. roxburghii "Rr-5" fruits and six different genotype fruits, two amino acid derivatives (MetO and L-lysine-butanoic acid) were confirmed to be significantly related to at least one in vitro antioxidant activity. Interestingly, MetO could not effectuate as an antioxidant. However, Met can reduce reactive oxygen species (ROS) levels through the activity of the Met sulfoxide reductase (MSR) system, in which MSRs act as natural scavenging systems for ROS by catalyzing the conversion of MetO to Met [55]. The phenomenon that the content of MetO is correlated to three in vitro antioxidant activities ( Figure 9A) might be attributed to the high content of MetO stimulating the MSR system, thereby playing an indirect role in the in vitro antioxidant activities. In the current study, one of the MYBs, evm.model.Contig144.143 (RrRAX3), and MetO had a similar variation trend ( Figure 9B), indicating that evm.model.Contig144.143 (RrRAX3) positively regulates the MSR gene expression and the MSR activity.
The characteristic metabolites of R. roxburghii were analyzed at the development stages. For Rr5-30, 59 characteristic metabolites were identified, including seven key metabolites (5galloylshikimic acid, glucarate o-phosphoric acid, gallic acid, tercatain, 4-acetamidobutyric acid, shikimic acid, 1-linoleoylglycerol). No key metabolites were found in Rr5-60 and Rr5-90. From the perspective of antioxidation, Rr5-30 has high antioxidant properties, and can be used as a raw material for antioxidant products. The characteristic metabolites were also used to perform a query on the traditional Chinese medicine database and analysis platform (TCMSP; tcmsp-e.com, accessed on 7 August 2021). For Rr5-30, Ser and Thr were identified as medicinal components. The eight medicinal substances identified from the characteristic metabolites of Rr5-90 were GABA, Met, Val, Lys, His, Tyr, Hcy, and Arg. No medicinal components were found in Rr5-60. Reportedly, Met and Val can be used to synthesize glucose during starvation, Lys can be used to manufacture ketones as an alternative energy source for the body [56], and Arg and Thr regulate the immune response in the body [57], among which Arg has the best effect [58]. In addition, Arg can stimulate insulin and regulate human blood glucose levels [59]. From a medicinal point of view, the Rr5-90 has a significant value. We can select R. roxburghii fruit at a specific development stage for processing according to its medicinal value or antioxidant value so as to improve the utilization rate of sparse fruits in planting.
The amino acid synthesis was complex. According to the difference in the synthetic skeleton, amino acids could be divided into six categories: glutamate amino acids, aspartic acid, serine family, histidine, aromatic amino acids, and alanine. The analysis of the amino acids of the glutamate family revealed that GPT (evm.model.Contig339.70) affects GABA, Arg, and 5-AVA. GLT1 (evm.model.Contig386.230) and NAGS (evm.model.Contig273.43) affect the accumulation of Cit, and ACY1 (evm.model.Contig290.142) affects the accumulation of Cit, GABA, and Arg. The ACY1 enzyme catalyzes the production of ornithine from N-acetylornithine. Fremont et al. showed that ACY1 has a significant effect on Arg accumulation [60]. Under drought stress, ornithine accumulation in watermelon increases and the expression of N-acetylornithine aminotransferase (clcg09003180) increases significantly [61]. These studies showed that ACY1 was the key gene of the Arg synthesis pathway. The results showed that the upstream gene had a maximal influence on the metabolites of the glutamate family. COT1 was the first enzyme synthesized in the aspartic acid family. The high expression of COT1 (evm.model.Contig120.160, evm.model.Contig283.202) leads to the overall upregulation of the aspartic acid family pathway. L-homocysteine, Met, SAH, and Lys were affected by the synthesis pathway genes, while Thr was affected by ThrC Aromatic amino acids are precursors of polyphenols. Tyr, TA, and L-phenylalanine were substrates for the synthesis of phenolic acids. Therefore, the high accumulation of Tyr and TA guaranteed the high accumulation of phenolic acids. Flavonoids and phenolic acids were the representative metabolites of R. roxburghii, with critical roles in the antioxidant activity of R. roxburghii. In this study, we elucidated the importance of 4-coumaric acid: coenzyme A ligase (4CL) (EC 6.2.1.12) to phenolic acids. 4CL synthesized cinnamic acid and its hydroxyl or methoxy derivatives, such as 4-coumaric acid, caffeic acid, ferulic acid, and 5-hydroxyferulic acid. Erucic acid was the substrate. The activity of 4CL influenced the accumulation of phenolic acids, and it was one of the major enzymes in the phenylpropanoid biosynthetic pathway [62,63]. Flavonoids were affected by many genes in this pathway, and any single gene could not be used to explain the changes in flavonoids.
Complex biochemical pathways were regulated by polygenic reactions and could not be explained by structural genes alone. Moreover, transcription factors play a critical role in biochemical pathways. The analysis of the promoter elements of structural genes identified many binding sites of the transcription factors on structural genes, especially MYB. Herein, we identified 158 RrMYBs in the transcriptome of R. roxburghii fruit, of which 75 showed a significant differential expression. In previous studies, 30 CcMYBs were involved in flavonoid and lignin biosynthesis in pigeon pea [64]. In pears, MYB was considered a key regulator affecting flavonoid biosynthesis [65]. TaMYB-A1 also regulated anthocyanin biosynthesis in Triticum aestivum [66]. The MYB transcription factor exerted a significant influence on the biosynthetic pathway of plants. The correlation between metabolites, structural genes, and transcription factors was presented in a network diagram, and six RrMYBs were identified as the potential key regulatory factors. Lu et al. demonstrated that 15 RrMYBs regulated the lignin of R. roxburghii, among which evm.model.Contig 191.1 (RrLHY) was reported by Lu et al. [11] to regulate lignin biosynthesis, further indicating evm.model.Contig 191.1 (RrLHY) as the key transcription factor. Silencing GhRAX3 reduced the ability of cotton to eliminate the hazardous effects of ROS [67]. In the present study, evm.model.Contig144.143 (RrRAX3) showed a specific effect on scavenging free radicals. We also speculated that evm.model.Contig361.14 (RrMYB105) and evm.model.Contig100.63 (RrREVEILLE7) are key RrMYBs in R. roxburghii. Herein, four key MYBs play a critical role in plants. REVEILLE7 and LHY are circadian clock-related genes, while PbRVE7 upregulated the expression of PbDFR and PbANS and promoted the accumulation of anthocyanins in pear peel [68]. Recent studies have demonstrated that LHY participates in cold stress by regulating the expression of DREB1 [69]. Zhang et al. showed that RAX3 regulates cuticle biosynthesis and is a key regulator of trichome initiation in sand rice (Agriophyllum squarrosum) [70]. OsMYB105 participates in the drought resistance of rice [71].

Conclusions
In this study, total phenols, flavonoids, triterpenoids, and 21 monomers were suggested to be key metabolites, contributing to the in vitro antioxidant activities, irrespective of the different developmental stages of fruits or in fruits of different R. roxburghii genotypes. Phenolic acid, flavonoid, and amino acid biosynthesis networks were constructed with 32 structural genes, 48 RrMYBs, and 23 metabolites by the conjoint analysis of the metabolome and transcriptome in R. roxburghii fruits at different development stages. Among these, four key RrMYBs affected the significantly varied amino acids, phenolic acids, and flavonol derivatives in the network during fruit development and the metabolites related to in vitro antioxidative activities in fruits of six R. roxburghii genotypes.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/foods11111639/s1. Figure S1: Rosa roxburghii fruits of three developmental stages. Scale bar = 1 cm; Figure Figure S4: Expression analysis of 15 genes related to the amino acid, phenylpropanoid, and flavonoid biosynthesis during fruit development of R. roxburghii. UBQ was used as the internal control. The error bars represent the standard error of the three biological replicates. The numbers above the graphics correspond to values obtained with Pearson's correlation; Table S1: The list of metabolites identified in Rosa roxburghii fruits; Table S2: The list of characteristic metabolites identified in each growth. Table S3

Data Availability Statement:
The metabolomic data of R. roxburghii 'Guinong 5' fruits at three different developmental stages: 30, 60, and 90 DAAs, are included in the present article. The mature fruit metabolomic data of the other five R. roxburghii genotypes, Rr-1, Rr-3, Rr-4, Rr-7, and Rr-f, can be found in Ref. [6]. The transcriptomic data of 'Guinong 5' fruits at three developmental stages: 30, 60, and 90 DAAs can be found in Ref. [11]. The RNA sequencing RAW data can be found in the Sequence Read Archive (SRA) database at NCBI (https://submit.ncbi.nlm.nih.gov/subs/sra, accessed on 20 March 2022) and are available under study accession number PRJNA533675.