Comparative transcriptome analysis linked to key volatiles reveals molecular mechanisms of aroma compound biosynthesis in Prunus mume

Mei (Prunus mume) is the only woody plant in the genus Prunus with a floral fragrance, but the underlying mechanisms of aroma compound biosynthesis are unclear despite being a matter of considerable interest. The volatile contents of the petals of two cultivars with significantly different aromas, Prunus mume ‘Xiao Lve’ and Prunus mume ‘Xiangxue Gongfen’, were characterised by GC-MS at different flowering periods, and a total of 44 volatile compounds were detected. Among these, the main substances forming the typical aroma of P. mume were identified as eugenol, cinnamyl acetate, hexyl acetate and benzyl acetate, with variations in their relative concentrations leading to sensory differences in the aroma of the two cultivars. We compiled a transcriptome database at key stages of floral fragrance formation in the two cultivars and used it in combination with differential analysis of floral volatiles to construct a regulatory network for the biosynthesis of key aroma compounds. The results indicated that PmPAL enzymes and PmMYB4 transcription factors play important roles in regulating the accumulation of key biosynthetic precursors to these compounds. Cytochrome P450s and short-chain dehydrogenases/reductases might also influence the biosynthesis of benzyl acetate by regulating production of key precursors such as benzaldehyde and benzyl alcohol. Furthermore, by analogy to genes with verified functions in Arabidopsis, we predicted that three PmCAD genes, two 4CL genes, three CCR genes and two IGS genes all make important contributions to the synthesis of cinnamyl acetate and eugenol in P. mume. This analysis also suggested that the downstream genes PmBGLU18-like, PmUGT71A16 and PmUGT73C6 participate in regulation of the matrix-bound and volatile states of P. mume aroma compounds. These findings present potential new anchor points for further exploration of floral aroma compound biosynthesis pathways in P. mume, and provide new insights into aroma induction and regulation mechanisms in woody plants.

weight [3]. To date, over 1700 VOCs have been identified across more than 90 plant families [4]. These compounds play valuable roles in attracting pollinators and as a defence mechanism against other animals and microorganisms [3,5], and can also improve the aesthetic value of ornamental plants, benefit human health, and serve as integral components of cosmetics, fragrances and condiments [6,7]. However, the aroma originally typical of Prunus mume was lost when enhancing its resistance to cold by crossing with closely related species of plum and apricot that are tolerant to such conditions. As a result, the floral fragrance of P. mume has been explored frequently [8], with recent research focusing on temporal and spatial changes of P. mume aroma compounds, and the molecular regulatory mechanisms of these VOCs [9][10][11].
The species P. mume, commonly known as the Chinese plum, is a traditional ornamental plant of the genus Prunus (Rosaceae) that flowers in early spring in China. It is also an important fruit tree with high economic value in China and other Southeast Asian countries, and is the only known member of Prunus that can send out a strong floral fragrance [12,13]. Whereas there are considerable variations in aroma type and composition between different cultivars, the typical aroma substances of P. mume seem to be eugenol, benzyl acetate and hexyl acetate [14]. Current studies appear to focus mainly on the identification of floral aroma components, together with their dynamic temporal and spatial changes [15,16]. For example, the biosynthetic pathways for benzyl acetate and eugenol both belong to the broader phenylpropene metabolic pathway, which has been better characterised in the model plants petunia and Arabidopsis [17,18]-with the synthesis of benzyl acetate catalysed by benzyl alcohol acetyltransferase (BEAT), using benzyl alcohol and acetyl CoA as substrates [19].
Although some studies have addressed the molecular mechanisms regulating flower fragrance in different P. mume cultivars [20], few of them have focused on the differences in floral VOCs between cultivars [21,22]. Furthermore, to the best of our knowledge, there are currently no reports that compare and analyse the transcriptome of different P. mume cultivars at different flowering stages. We therefore considered it important to investigate the diversity of volatile components in different P. mume cultivars, as well comparing differences in aroma composition using headspace volatiles and endogenous extracts [23]. To clearly elucidate the molecular mechanisms of VOC biosynthesis and compare them between cultivars, combined transcriptomic and metabolomic analysis was adopted since this has been established as an effective method and has been applied in many plants, including Chimonanthus praecox [24] and Nymphaea colorata [25]. For sample collection in this study, petals were harvested from P. mume 'Xiao Lve' (LE) and P. mume 'Xiangxue Gongfen' (GF) at three developmental stages (Fig. 1a) and analysed by headspace solid-phase micro-extraction (HS-SPME), combined with gas chromatographymass spectrometry (GC-MS) for headspace volatiles and endogenous extracts. Differential compounds in the two aroma types were then screened as biomarkers using variable important for prediction (VIP) scores. RNA-Seq analysis was performed on petals collected at the three developmental stages to permit screening for differentially expressed genes associated with floral aroma synthesis. In addition, further analysis using odour activity values (OAVs) was able to screen out several biomarkers that make significant contributions to the two aroma types. Ultimately, our comparative molecular-level analysis of the mechanisms underlying differential expression of aroma-related components of the two cultivars-and our search for the synthetic pathways of aroma compounds in P. mume-aimed to provide a theoretical basis for molecular breeding to achieve specific Prunus floral aromas, and to enhance essential oil production.

Major floral aroma constituents and qualitative and quantitative analysis of the P. mume cultivars
In total, 36 compounds were detected over the three developmental stages of GF, including 12 esters, nine alcohols, four phenols, three terpenoids and two aldehydes. For LE, a total of 28 compounds were detected, including 12 esters, six alcohols, three terpenoids, two aldehydes and two phenols. Of the compounds identified, 15 were unique to GF, and seven to LE ( Fig. 1b and Table  S1). The percentage of each aroma compound in the total volatile substances released differed between the two P. mume cultivars (Fig. 1c). An OPLS-DA model was developed based on the relative contents of headspace volatiles in the two P. mume cultivars at the three developmental stages, which enabled identification of the differential biomarkers in headspace volatiles of the two cultivars, and the cross-validated predictive capability (Q 2 = 0.967) manifested the feasibility of the model (Fig. 1d). Among the VOCs detected, cinnamyl acetate, eugenol, benzyl alcohol, benzaldehyde and benzyl benzoate varied greatly in content between the two cultivars across all flowering development stages, which may be an important factor underlying their differences in floral fragrance. Together with the OAV analysis (Table 1), this confirmed that among the shared aroma compounds, benzyl acetate, eugenol, benzaldehyde, benzyl alcohol and hexyl acetate were present at relatively high levels, consistent with previous studies on the characteristic aroma compounds of P. mume [26].

RNA-Seq analysis of developing flowers of two P. mume cultivars with different floral aromas
To explore the molecular mechanisms underlying the formation of different fragrances between LE and GF, petal samples of the two cultivars collected at the S2, S3 and S4 Fig. 1 a Different developmental stages of P. mume 'Xiao lve' (LE) and P. mume 'Xiangxuegongfen' (GF) flowers: (S2) petal baring period; (S3) early flowering period; (S4) full flowering period. b Heatmap analysis of volatile compounds of two P. mume cultivars at three flowering development stages. The colour of the heatmap ranges from dark blue (value, − 2.5) to magenta (value, + 2.5) on a natural logarithmic scale. c Amounts of the six main floral aroma volatile compounds during three different flowering stages of GF and LE. d Score plot of OPLS-DA model of headspace volatiles of P. mume cultivars with the statistical parameters: R 2 X = 0.999, R 2 Y = 0.994, Q 2 = 0.967 developmental stages were subjected to RNA-Seq analysis, with each sample run in triplicate (N = 3). Therefore, a total of 18 libraries were constructed and analysed, yielding 58.49 GB of clean reads. More than 97% of bases were above the Q20 threshold, and the GC content was above 46% (Table S2), indicating a high overall sequencing quality.
The number of transcripts expressed in each sample, after genes with normalised reads with FPKM < 0.5 were removed, is shown in Fig. 2a To investigate global differences in transcriptome dynamics during flowering in LE and GF, principal component analysis (PCA) was carried out on the RNA-Seq data obtained for the samples taken at different periods ( Fig. 2b), which could be clearly divided into three groups corresponding to S2, S3 and S4. The data for S2 and S3 clustered significantly differently between LE and GF, indicating that these flowering stages differed considerably in their transcriptional programmes between cultivars. The closer grouping of S2 and S3 for LE than for GF indicated that these two developmental periods were more closely transcriptionally related for the former. For stage S4, the data for LE and GF clustered together, indicating a high similarity in their transcriptional programmes at this stage; thus, the greater transcriptional divergence observed at S2 and S3 may explain the different floral scents of the two P. mume cultivars.

Identification of differentially expressed genes, and enrichment analysis
Pairwise comparisons were made for GF and LE at each developmental stage. In each comparison, differentially expressed genes (DEGs) were screened for by expression level with the constraints FPKM > 5, FDR < 0.01, and log 2 fold change > 1 or < − 1. The results showed that 5884, 2850 and 2336 genes were differentially expressed between LE and GF at stages S2, S3 and S4, respectively (Fig. 2c). In order to screen further for genes related to flower aroma formation, pairwise comparisons were made between the two important flowering stages, S2 and S4, and identified 2829 and 6327 DEGs between these stages for GF and LE, respectively.
According to their expression profiles, 2973 of the DEGs were assigned to one of three groups (I, II, III), consisting of 1031, 1719 and 223 individual genes, respectively. The first two groups contained genes that were upregulated (group I) or downregulated (group II) as the flowering stages progressed, whereas the genes in group III were upregulated between S2 and S3 but then downregulated between S3 and S4 (Fig. 3).
In addition, gene ontology (GO) enrichment analysis was performed on the identified DEGs, with the main GO categories shown in Fig. S1. A total of 50 GO annotations were obtained for 8980 DEGs in the three functional categories of biological process, molecular function and cellular component. The main terms (with the associated number of genes in parentheses) included metabolic process (5667), catalytic activity (5310), cellular process (4945), single organisation process (3880), binding (3800), cell (2901) and cell part (2901). The GO annotations of these genes may help in determining the molecular mechanisms of flower aroma production and release.
To identify relevant genes that were significantly enriched in metabolic or signalling pathways, DEGs for each of the S2, S3 and S4 periods were mapped to the KEGG database. For the three flowering periods, 1618, 1095 and 1059 DEGs, respectively, were enriched in the KEGG PATHWAY database. Among these, the metabolic pathways with more annotations of the DEGs at different periods were metabolic pathways, biosynthesis of secondary metabolites, ribosome, biosynthesis of amino acids, plant hormone signal transduction, phenylpropanoid biosynthesis, and pentose and glucuronate interconversions ( Fig. S2).
As discussed above, plant aroma constituents mainly include terpenoids, aromatic species and aliphatic compounds. Based on known metabolic pathways related to the synthesis of these three substance classes, 739, 614 and 504 DEGs-related to 13 metabolic pathways-were identified at the S2, S3 and S4 periods, respectively, of GF and LE (Fig. 4). For the three flowering periods, metabolic pathways associated with the biosynthesis of aroma compounds that were more extensively annotated with DEGs were: biosynthesis of secondary metabolites; phenylpropanoid biosynthesis; fatty acid metabolism; and phenylalanine, tyrosine and tryptophan biosynthesis.

Identification of DEGs related to regulation of flower aroma metabolism at different developmental stages
More detailed comparisons were then made between the three flowering periods of GF and LE. Genes that were significantly differentially expressed between cultivars at S2 were further screened by comparison with S3 and S4 (Table S3), and the pathway classification of DEGs enriched at S2 is shown in Fig. 5a. At this flowering stage, a large number of transcription factors (TFs) and genes associated with metabolism of floral volatiles were highly enriched in both cultivars-including protein coding genes for phenylalanine metabolism, fatty acid biosynthesis and phenylpropanoid biosynthesis-suggesting that regulatory networks including phenylpropanoid biosynthesis and fatty acid metabolism control the synthesis of aroma compounds in both Prunus varieties. Similar analyses identified genes significantly overexpressed at S3 (Fig. 5b, Table S3) and S4 (Fig. 5c, Table S3); and DEGs that were more highly differentially expressed between cultivars over all three flowering periods were also screened out, then functionally enriched and classified (Fig. 5d, Table S3).

Construction of gene coexpression networks
Weighted gene correlation network analysis (WGCNA) was then carried out in order to gain a comprehensive understanding of the genes expressed in GF and LE at successive developmental stages, and to identify specific genes associated with the synthesis of floral volatiles. After screening out low-expressed genes (FPKM < 0.05), WGCNA retained 12,401 genes. Coexpression networks were constructed based on pairwise correlation of gene expression across all samples. This analysis identified 15 distinct coexpressed modules (Fig. 6a), seven of which (mediumpeople3, bisque4, green, lightyellow, ivory, midnightblue and white) were significantly associated with changes in floral volatile content during flowering (Fig. 6b). More specifically, cinnamyl acetate and benzyl alcohol content were highly positively correlated with mediumpeople3 module gene expression; cinnamyl acetate and benzyl acetate were negatively correlated with bisque4 module gene expression; benzaldehyde and eugenol were positively correlated with green module gene transcript abundance; benzyl acetate, benzaldehyde, benzyl alcohol and eugenol were negatively correlated with the lightyellow module; and benzyl acetate was highly positively correlated with ivory module gene expression. Finally, hexyl acetate was highly positively correlated with the midnightblue module, but negatively correlated with the white module. Based on the correlations between genes of these seven modules and the floral volatiles, together with the KEGG PATHWAY enrichment results for floral aroma compound anabolism (as described above), we identified a total of 957 structural genes potentially associated with the biosynthesis of floral volatiles. The expression patterns of these genes were correlated with changes in the content of other important aroma compounds to find specific structural genes that may be important for biosynthesis of these volatiles (Supplementary Fig. S3). For example, it was found that the genes LOX, ADH, CAD, HPL and BAHD may be involved in the synthesis of hexyl acetate, which is an important VOC in P. mume.

Analysis of proposed genes associated with biosynthesis of known floral volatiles
Combining the results of GO enrichment, KEGG PATH-WAY enrichment and WGCNA analysis of DEGs, a focused set of 101 genes was obtained that may be directly involved in the biosynthesis of P. mume aroma compounds. To specifically address the biosynthetic pathways of cinnamyl acetate, benzyl acetate and (iso) eugenol-and find the genes that play essential roles in these pathways in P. mume-the key coding enzymes were searched using the phmmer functionality of HMMER 3.0. The biosynthesis of benzene-containing aroma compounds in plants usually starts from shikimic acid, which is converted to phenylalanine through the shikimate pathway (involving aromatisation); a process that crosses the boundary of primary and secondary metabolism. Here, a phylogenetic tree was constructed by comparing the protein sequences of putative essential genes for P. mume fragrance formation with their verified homologues in Arabidopsis thaliana (Fig. 7). The results showed that the two P. mume ADH (PmADH) genes and two reported A. thaliana ADH (AtADH) genes fall into the same branch; that the three PmSDR1-like and AtSDR1 protein sequences are very similar; and that the sequence similarity between Pm4CL2 and At4CL3 is high. Also, PAL genes were generally found to have high sequence homology between Prunus and Arabidopsis species. The genes PmUGT73C6-like and AtUGT73D clustered in one branch, whereas PmBGLU18-like clustered with the AtBGLU45 and AtBGLU46 genes, whose relationship to flower aroma volatilisation has been confirmed in Arabidopsis [27].
Phenylalanine ammonia lyase (PAL) catalyses the deamination of phenylalanine to trans-cinnamic acid, the primary and committed step of phenylpropanoid biosynthesis. There are two PmPAL genes (ncbi_103332355 and ncbi_103342143), with significant differences in their comparative transcriptome (Table S5). Furthermore, utilisation of phenylalanine is divided between two different pathways to synthesise phenylalanine-related compounds or other benzene compounds. We used the characterised A. thaliana genes from these pathways as bait to search the P. mume genome and found three PmCAD, two Pm4CL, two PmCCR , two PmADH, four PmSDR and three PmCYP genes (Table S5). Among these, the expression pattern of PmCCR was highly positively correlated with the content of volatile benzene compounds, and the expression of PmCCR1 in LE was significantly higher than in GF.
β-Glucosidase (BGLU) is known to be widely present in various plant tissues. Its role is mainly in the prehydrolysis of aroma compounds, converting aromatic substances in the plant body from the bound (glycosylated) form  Table 4. The colour of the heatmap ranges from green (value, − 2) to red (value, + 2) on a natural logarithmic scale to the free state and therefore releasing the aroma volatiles. In contrast, UDP-glycosyltransferase (UGT) acts in the opposite direction. The present transcriptomic data revealed that, in total, four β-glucosidases (PmBGLU) and two UDP-glycosyltransferases (PmUGT ) were significantly differentially expressed in LE and GF (Table S5). The expression of all six of these genes was significantly upregulated in GF, relative to LE, at the S3 stage.
Further analysis of the enrichment network of these genes using the Cytoscape revealed that 4CL and CCR were mainly enriched in upstream floral aroma synthesis processes, including phenylpropanoid metabolism, cinnamoyl-CoA reductase (CCR) activity and 4-coumarate: coenzyme A ligase (4CL) activity (Fig. 8). Both PAL and CAD (encoding cinnamyl alcohol dehydrogenase) were clearly enriched in a wider range of processes, including phenylpropanoid metabolism, phenylpropanoid biosynthetic processes, organic hydroxy compound metabolism and glucosinolate metabolism. In addition, UGT and BGLU were mainly enriched in the downstream stages of floral compound synthesis, in key functions of floral aroma formation such as phenol-containing compound metabolic processes and β-glucosidase activity. Taking a more general view, GO annotation of the genes and TFs identified during this analysis is anticipated to help determine possible relationships between key genes in the biosynthesis of P. mume aroma compounds and the molecular mechanisms of floral aroma compound synthesis and volatilisation.

Exploration of key transcription factors
Transcription factors are thought to be critical for the proper control of many secondary metabolic processes. In our study of the comparative transcriptome of P. mume, 613 TFs (including transcriptional regulators) were identified from 76 families classified in the PlantTFDB database. On constructing reciprocal networks of differentially expressed TFs together with key genes (Fig. 8), we were surprised to find that MYB4 transcription factors were enriched in phenol-containing compound metabolism, organic hydroxy compound metabolism, phenylpropanoid biosynthesis and phenylpropanoid metabolism, suggesting that these TFs might be important for regulation of the biosynthesis of volatile compounds [29]. In support of this possibility, PmMYB4 TFs were upregulated in GF.
Furthermore, MYB73 was significantly enriched in glucosinolate metabolism and glycosyl compound metabolism, with PmMYB73 transcriptionally downregulated in LE but consistently upregulated in GF.
Given this discovery of the involvement of core TFs, especially the MYB family, in VOC synthesis, the mechanisms of floral aroma formation in P. mume were interrogated further. Combining the observations regarding TFs with comparison of structural genes with verified protein Fig. 7 Proteins corresponding to genes highly relevant to the biosynthesis of floral aroma compounds in Prunus were screened by nucleotide sequence alignment with the related Arabidopsis proteins function, we constructed a potential gene regulatory network for floral fragrance composition (Fig. 9), laying a foundation for uncovering the biosynthetic pathways of benzene compounds in LE and GF; especially cinnamyl acetate, benzyl acetate and eugenol.

Gene expression validation
To confirm the integrity of the RNA-Seq results, we selected five key genes in the metabolic pathway of floral aroma compounds and one housekeeping gene (PP2A), and analysed relative expression levels in GF and LE by qRT-PCR for samples collected at the three flowering periods (Fig. 10). Pearson correlation analysis was used to evaluate the relationship between the RNA-Seq and qRT-PCR results for the different developmental stages (S2, S3 and S4). From a plot of log 2 (fold change in FPKM) from RNA-Seq against log 2 (2 −ΔΔCq ) from qRT-PCR, Pearson correlation coefficients of 0.95 and 0.84 were obtained for GF and LE, respectively, indicating that the RNA-Seq data were reliable. Fig. 8 The key genes screened out by combining GO enrichment, KEGG PATHWAY enrichment [28] and WGCNA analysis were further analysed using Cytoscape. Nine key genes and MYB TFs were significantly enriched in a total of 18 metabolic processes Xiujun et al. BMC Plant Biology (2022) 22:395

Dynamics of the main floral compounds of GF and LE
Floral fragrance is an important characteristic of ornamental plants, and the key components of this aroma consist of diverse secondary metabolites. The relative composition, amounts and interactions of these volatile compounds determine the specific floral fragrance [8], which is characterised by a high odour value (content/ olfactory threshold) [30]. In the current study, we used GC-MS for relative quantification of 44 major volatile metabolites of P. mume and performed absolute quantification against external standards for six VOCs that were present at widely varying relative concentrations. The main volatiles identified were benzene compounds, From left to right, the first three squares represent the relative log 2 (expression ratio) at the S2, S3 and S4 stages for LE; and the latter three squares represent the equivalent values for GF. The colour scale corresponds to log 2 (expression ratio), as indicated with benzyl alcohol and benzaldehyde in particular being important precursors for the formation of floral aroma compounds typical of P. mume [13,18]. The main components of its distinctive aroma are aromatics (eugenol, cinnamyl acetate and benzyl acetate) [31], and similarly, the floral aromas of many plants contain aromatic esters synthesised from the corresponding aromatic alcohols and acyl donors through reactions catalysed by acyltransferases [32,33]. The amount of benzyl acetate released was higher for LE than for GF at all three flowering stages (Fig. 1c), and the odour threshold was lower, with an aroma of fruit and jasmine [31]. The amounts of benzaldehyde (with a sweet flower and fruity aroma) and eugenol (with a spicy clove odour) released by LE were also higher than for GF, and the interaction of these three volatiles may therefore explain why the aroma of the former is more sweet and fruity than the latter. In contrast, in the late flowering stage of GF, more cinnamyl acetate, cinnamyl alcohol and benzyl alcohol were detected, which manifest themselves with creamy and sweet aroma, and altogether shed the aroma around to be richer, stronger, and more intense. The biosynthesis of all of these floral aroma compounds was mainly influenced by expression of relevant genes in the biosynthetic pathways, enzyme activity, and substrate utilisation. The diversity of volatile aroma compounds and the abundance of P. mume germplasm therefore provide a material basis for the development of a wide range of aroma products.

Key regulatory genes in the biosynthesis of cinnamic acid
As mentioned above, PAL is the first and key enzyme of the phenylpropanoid metabolic sequence, and is involved in secondary metabolic pathways in higher plants [34]. Phenylalanine produced via the shikimate pathway is deaminated by PAL to produce trans-cinnamic acid, which then enters the phenylpropanoid pathway [35]. In this work, comparative transcriptomic analysis revealed that two PmPAL genes were significantly more expressed in LE than in GF during S2, and vice versa during S4 (Table S6). High expression of PmPAL in LE during the S2 period would produce more trans-cinnamic acid, a precursor for the synthesis of volatile compounds, and may therefore explain why considerably greater amounts of VOCs were measured for LE than for GF at this stage (Fig. 1c). A small number of plant TFs have previously been associated with the formation of floral fragrance. Petunia has four MYB members that regulate benzenoid volatiles [36], and by overexpressing the MYB genes from A. thaliana in petunia flowers, it was demonstrated that regulating the flow of precursors to phenylpropanoid metabolism affects volatile synthesis [37]. The MYB-like transcription factor PhMYB found in petunia indirectly inhibits the synthesis of p-coumaric acid derivatives by repressing the upstream enzyme, cinnamate-4-hydroxylase (C4H) [21]. Analysis of our transcriptomic data identified a PmMYB4 gene (Table S6) that was significantly more expressed in GF than in LE, and was enriched in the metabolic processes of phenolic-containing compound metabolism and phenylpropanoid biosynthesis. These observations suggest that PmMYB4 is also likely an important TF in P. mume, responsible for the significant differences in floral aroma composition between cultivars. Further exploration of the regulatory role of TFs in the synthesis of plum volatiles is therefore required, if the molecular mechanisms involved in biosynthesis of these aroma compounds are to be fully elucidated.

Key genes for the induction of cinnamyl acetate and eugenol biosynthesis
The enzyme 4-coumarate: coenzyme A ligase (4CL) catalyses the third and final step of the phenylpropanoid metabolic pathway. It thus occupies a central position in the plant metabolic network, playing an important role in the biosynthesis of eugenol and cinnamyl acetate [38]. Although long explored, the biosynthetic pathway of benzaldehyde is still unresolved, despite it being a key intermediate in the synthesis of benzenoid compounds in P. mume aroma volatiles. In petunia, isotopic labelling was used to establish that benzaldehyde is derived from phenylalanine [39]; however, the benzenoid profile was unaffected if expression of Ph4CL gene was inhibited, suggesting that Ph4CL is not a major factor in the regulation of benzyl acetate biosynthesis in this species [40]. As stated above, and exemplified by previous studies, 4CL is an entry and control point for the phenylpropanoid and terpenoid biosynthetic pathways [41]; for example, its protein and mRNA expression profiles in basil were shown to be consistent with the production of phenylpropanoids and terpenoids in isolated glandular capillaries [42]. Here, correlation analysis between our comparative transcriptomic data and the aroma compound assay results revealed that two Pm4CL genes were significantly more highly expressed in LE than in GF, meaning they are likely involved in the synthesis of the volatile compounds cinnamyl acetate and eugenol.
Cinnamyl alcohol dehydrogenase (CAD) catalyses the conversion of cinnamaldehydes to the corresponding cinnamyl alcohol via an NADPH-dependent redox reaction [43], which is also the final step in the biosynthesis of lignin precursors [44]. Interestingly, the functions of different CAD protein members of the same species differ considerably [45]. Two CADs have been characterised in A. thaliana, AtCAD4 and AtCAD5, which show different substrate specificities but with the latter usually catalysing the synthesis of lignin monomers [46]. Comparison of the differential expression of CAD genes in LE and GF identified three key genes (Table S6) that might be involved in the synthesis of aroma compounds in P. mume. Correlating the transcriptomic data with the results of the aroma compound assay, it was hypothesised that PmCAD1, which is highly expressed in LE, is more likely to be involved in the synthesis of cinnamyl alcohol.
Biosynthesis of the phenylpropanoids eugenol and isoeugenol shares the same pathway as G-type lignin monomers in the early stages, up until the lignin monomer coniferyl alcohol is produced by CAD, at which point the phenylpropanoid pathway diverges from the lignin pathway [47]. Subsequent esterification produces coniferyl acetate, which is then converted by either of two enzymes-eugenol synthase (EGS) or isoeugenol synthase (IGS)-into eugenol or isoeugenol, respectively, which essentially belong to the same class of compounds [25]. Current studies on the molecular aspects of these floral volatiles have focused on the final step of (iso) eugenol anabolism, with 18 EGS and IGS genes having been identified in 10 plants including basil and petunia [48][49][50][51]. Analysis of the transcriptomic data obtained in this work revealed two genes annotated as isoeugenol synthases that were both expressed in LE at higher levels than in GF, consistent with the observed trend in metabolites and suggesting that these two PmIGS genes (Table S6) are required for the final step of isoeugenol biosynthesis.

Key genes inducing the biosynthesis of benzyl acetate
Short-chain dehydrogenase/reductase (SDR) genes are involved in many aspects of primary and secondary metabolism and are notable in carbonyl-alcohol redox [52]. The SDR family is important for biosynthesis of some compounds derived from phenylalanine, such as in Rosa damascena flowers, where 2-phenylethanol (2PE) is synthesised from phenylalanine via phenylacetaldehyde as an intermediate. Reduction of the latter to 2-PE is catalysed by phenylacetaldehyde reductase (PAR), an SDR family member [53]. In this study, four PmSDR genes were detected in the transcriptomes of LE and GF (Table S6) and found to be involved in the biosynthesis of benzyl acetate. Their expression was consistently upregulated, with more benzyl alcohol detected in both cultivars during S4. Significant expression of PmPAL and PmCNL genes was detected during the benzaldehyde production phase in a recent study of P. mume, suggesting that endogenous benzaldehyde might originate from two pathways [25]. In P. mume, two P450s were coexpressed to produce the cyanogenic glycosides prunasin and amygdalin, which are derived from phenylalanine with benzaldehyde implied as a biosynthetic intermediate. The plant cytochrome P450 gene family catalyses a large number of mono-oxidation/hydroxylation reactions in plant primary and secondary metabolism [54], and thus, as an integral part of the benzenoid pathway, P450s might also have an important influence on the benzaldehyde synthesis pathway [27]. An in-depth study of the SDR and CYP gene families in P. mume may therefore help unravel the synthesis of benzyl acetate and the accumulation of benzyl alcohol as it occurs in these species.

Key genes affecting the volatilisation of aroma compounds
Glycosylation of plant small molecules is a very common physiological phenomenon, and is one of the main mechanisms by which plant cells maintain metabolic homeostasis [55]. Previous studies on the UGT gene family have focused on its involvement in plant biotic stress response [56]. By comparative transcriptomic analysis, we found three UGT genes (PmUGT71A16, PmUGT73C6, PmUGT74F2) with potential involvement in the biosynthesis and volatilisation of P. mume aroma compounds. Of these, PmUGT74F2 was significantly enriched in phenol-containing compound metabolism and organic hydroxy compound metabolism, suggesting its involvement in regulation of the volatilisation of P. mume aroma compounds.
A total of five DEGs annotated as β-glucosidases were detected in LE and GF (Table S6), all showing downregulation as flowering development progressed, and with expression in GF significantly higher than that in LE. β-Glucosidases are found in a wide range of plant tissues and, in contrast to UGTs, act mainly in the prehydrolysis of aroma compounds, converting aromatic substances in the plant from the bound to the free state and releasing aroma volatiles [57,58]. Hydrolysis of cinnamon using β-glucosidase was found to release 'hidden' aroma compounds and to significantly increase the aroma concentration [27]. Two A. thaliana glucosidases, AtBGLU45 and AtBGLU46, clustered together with Pinus contorta β-glucosidase, and phylogenetic analysis suggested they were all able to hydrolyse eugenol and cinnamyl glycosides [59]. It is interesting to note that although the aroma of P. mume gradually became more intense from stages S2 to S4-and the aroma volatile content was consistently higher in LE than in GF-the PmBGLU gene was progressively downregulated in both cultivars over the three development stages, and it was expressed significantly more in GF than in LE. These observations may reflect a series of upstream processes that synthesise more aroma compounds in LE, making more substrates for prehydrolysis reactions. In addition, we found that PmUGT71A16 was continually upregulated in LE and downregulated in GF, implying glycosylation of more of the aroma compounds produced in GF, and thus lower levels of these compounds in the volatile state. This may be a major reason why more aroma compounds were released from LE than GF, and it is also speculated that glycosylation takes priority over prehydrolysis in P. mume.

Conclusion
In this study, OPLS-DA and OAV analyses identified six main floral aroma components for the two P. mume cultivars across three development stages: benzyl acetate, eugenol, benzaldehyde, benzyl alcohol, hexyl acetate and cinnamyl acetate. This composition of VOCs may define the specific aroma characteristic of P. mume. On combining these volatile compound measurements with transcriptome analysis, we found that the PmPAL enzymes and PmMYB4 TFs may play prominent roles in regulating the accumulation of key precursor substrates (trans-cinnamic acid and p-coumaric acid derivatives) of the major P. mume aroma compounds. Furthermore, we obtained evidence supporting involvement of PmCYP genes and PmSDRs in controlling the biosynthesis of some key precursors, such as benzaldehyde and benzyl alcohol, in the benzyl acetate synthesis pathway. Association analysis of transcriptomic data and the key aroma substances revealed that three CAD genes, two 4CL genes, three CCR genes and two IGS genes are likely to be important for synthesis of cinnamyl acetate and isoeugenol. Finally, the PmUGT71A16 and PmUGT73C6 genes are proposed as key factors controlling the morphology (bound and volatile states) of aroma compounds. These results have established a practical basis for further exploration of the regulatory mechanisms underlying the formation of aroma compounds in P. mume.

Plant materials
Two P. mume cultivars with different flower fragrances, 'Xiao Lve' (LE) and 'Xiangxue Gongfen' (GF), were chosen as the experimental materials. The two P. mume cultivars are not wild plants; instead, we have collected the plant materials in the Mei Garden of Jiufeng National Forest Park (39°54′ N, 116°28′ E) from March to April 2019. We have also obtained specific permissions regarding our collections and experiments with respect to institutional approval. The flowers were selected in advance from the upper part of the branches at 10:00 am on a clear day, and specimens at three stages of flowering (open petal, first bloom and full bloom) were selected as samples, cut off, then wrapped in aluminium foil and frozen in liquid nitrogen until testing of the petals. Transcriptome sequencing analysis (RNA-Seq) was performed on samples corresponding to the petal baring (S2), early flowering (S3) and full flowering (S4) periods of LE and GF (Fig. 1a). Each experiment was run in triplicate, making a total of 18 samples analysed, and the mean value of replicates was taken by default for subsequent analyses.
conserved domains were searched in the EMBL-EBI database (http:// pfam. xfam. org/). Sequence alignment was performed using the MUSCLE tool, and phylogenetic tree construction using the FastME tool. Adobe Illustrator CC 2021 was used for visual presentation of graphics.

qRT-PCR validation
The real-time quantitative reverse transcription polymerase chain reaction (qRT-PCR) method was employed to detect the expression of six genes related to floral biosynthesis in the petals of LE and GF (Table S7), using the fluorescencebased quantitative PCR instrument, ABI Step One Plus (Applied Biosystems, USA). According to the manufacturer's instructions provided with the reverse transcription kit (Novizan R223, China), 50 ng-2 μg total RNA was used to establish each reaction in a volume of 20 μL. Amplification proceeded as follows: predenaturation at 95 °C for 90 s, followed by 40 cycles of 95 °C for 5 s, 65 °C for 30 s (annealing), and 72 °C for 20 s. Dissociation curves were recorded from 65 to 95 °C. Each reaction was repeated three times. The relative expression levels of target genes were calculated using 2 −ΔΔCq , and the data were analysed with SPSS 26.0 and Origin 8 software. The primer sequences used for qRT-PCR analysis are listed in Supplementary Table 7.