The Luffa plant materials were provided by the Vegetable Research Institute of Guangzhou Academy of Agricultural Sciences, including the non-bitter variety WK1 and the bitter variety YK1 derived from the high-generation self-inbred germplasm material of Luffa acutangular. In addition, YK2 is an highly bitter variety from the F₁ generation obtained after interspecific hybridization of Luffa cylindrica and Luffa acutangula, with non-bitter fruits formed in both parents. The three materials were planted at the Nansha base of the Guangzhou Academy of Agricultural Sciences in open-field cultivation and were managed according to local customary practices. After germination, the seeds were sown in 21-hole trays (the matrix ratio was coconut bran:vermiculite = 3:1) and transplanted in the open field when the seedlings reached the two-true-leaf stage. The bottom fertilizer for the cultivated land consisted of a mixture of organic biological fertilizer and chemical fertilizer (5:1 ratio) at a rate of 4000 kg/ha. The 'Z'-shaped artificial vine and pruning of the branches at the bottom of the main vine were carried out when the height of the plant reached approximately 1 m, along with proper water and fertilizer management. Aulacophora indica infestation was controlled by spraying the plants with a mixed imidacloprid 1500 solution and spinetorams 1500 solution; melon silk borer was controlled by spraying plants with broflanilide 2500 solution once every 5 days; and powdery mildew was controlled by spraying plants with a mixed azoxystrobin 2000 solution and methyl thiophanate 800 solution. Finally, the fruits were collected at 16 days after pollination with three samples from each treatment, wrapped in tinfoil, and then immediately frozen in liquid nitrogen. Three biological replicates were derived from at least five fruits.

The bitterness of the developing fruit is focused primarily near the stem(close to the stem 2-3cm), with little bitterness present in other parts of the fruit. The fruit grows in three stages: early (7 days after pollination), middle (12 days after pollination), and late (16 days after pollination). According to previous research methodologies, a group of three volunteers assessed the bitterness phenotype of the fruit during the middle stages, approximately 2–3 cm from the stem. After tasting the fruit, the volunteers washed their lips with clean water to enable tasting subsequent samples without being affected by the previous bitterness to reduce potential error in the tasting process [24, 25]. We divided the bitterness phenotype into four levels as follows: level 0, not bitter (sweet); level 1, slightly bitter; level 2, moderately bitter; and level 3, highly bitter. WK1 and YK1, the homozygous inbred lines of Luffa, were categorized as having fruit phenotypes of not bitter (sweet) and moderately bitter (levels 0 and 2), respectively. The phenotype of YK2, an interspecific hybrid F₁ of Luffa cylindrica and Luffa acutangula, was characterized as highly bitter (level 3).

The fruits (50 mg, approximately 2–3 cm from the stem) were ground for 10 min in a grinder at 30 Hz after being vacuum freeze-dried. The samples were then extracted at 4°C in 1000 L of a methanol-acetonitrile solution (methanol:acetonitrile:water = 2:2:1 v/v). The samples were centrifuged at 12,000 rpm for 15 min at 4°C before being filtered and absorbed for liquid chromatography (LC) and mass spectrometry (MS) analysis to detect metabolites, which was performed at Beijing Biomarker Technologies Co., Ltd.

All metabolite annotations were performed according to the BMK G database. Metabolite quantification was accomplished by multiple reaction monitoring-mode analysis using triple-quadrupole MS. The variation in metabolite abundance among groups was evaluated using orthogonal partial least-squares discriminant analysis (OPLS-DA) with the R language package ropls and 200 permutation tests were performed to verify the reliability of the model. The variable importance in projection (VIP) value of the model was calculated using multiple cross-validations. Differentially accumulated metabolites (DAMs) were screened according to a fold change (FC) > 1, P value < 0.05, and VIP > 1 of the OPLS-DA model. The functions of the DAMs were evaluated according to significant enrichment in Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways based on a hypergeometric distribution test.

Total RNA was extracted from the nine samples ((in the middle stages, three replicates of WK1, YK1, and YK2) using the RNAprep Pure Plant Kit (Tiangen, Beijing, China), following the manufacturer's instructions. The RNA quality was assessed using the RNA Nano 6000 Assay Kit of the Agilent Bioanalyzer 2100 system (Agilent Technologies, CA, USA), and the concentration and purity were quantified using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, USA). The cDNA libraries were sequenced on an Illumina NovaSeq platform, resulting in 150-bp paired-end reads that were processed using Trimmomatic [26]. Clean reads were accurately compared to the reference genome using Hisat2 software to gain information on the location of reads on the Luffa genome [27]. The reads from the comparison were then combined with StringTie [28] to reconstruct the transcriptome for further investigation. Differentially expressed genes (DEGs) were screened using the differential analysis program DESeq2 [29] according to the thresholds of |log2 (FC)| ≥1 and false discovery rate < 0.01 [30]. Functional annotation and analysis of the DEGs was performed using Gene Ontology (GO) [31], Kyoto Encyclopedia of Genes and Genomes(KEGG) [32], and Cluster of Orthologous Groups of proteins(COG).

For heat stress, Luffa seedlings were cultivated at 20–25°C. The 14-day-old seedlings were then exposed to high temperatures of 45°C /28°C with a 12-h/12-h light/dark cycle in a climate-controlled chamber; control plants were cultivated under normal conditions. The leaves were sampled after 5 days of treatment. For the ABA treatment, the 14-day-old seedlings were separated into two groups: the experimental plants were sprayed with a 100 µM ABA solution, while the control plants received water. The leaves were sampled 6 days after treatment. Each experimental treatment comprised five experimental samples and three biological replicates.

Total RNA was extracted from the test samples using the RNAprep Pure Plant Kit, following the manufacturer instructions (Tiangen, China). Total RNA (0.2 µL) was reverse-transcribed into cDNA with SynScript®III RT SuperMix. The qRT-PCR analysis was performed using the ArtiCan^CEO SYBR qPCR Mix; the 18S rRNA gene (gene ID = 58119177)was used as an internal control. The 2^-△△Ct method was used to calculate the relative expression level of the target genes. Three biological replicates were performed. The primers used qRT-PCR are listed in Supplementaty Table S1.

Statistical analysis was performed using the independent samples t-test in SPSS 22.0 (SPSS Inc., USA), with * p ≤ 0.05 and ** p < 0.01. The data were reported as mean ± SD. Each reaction was performed on the biological replicates.

Samples were extracted from fruits of three different bitterness phenotypes in Luffa for qualitative and quantitative analysis of bitter-related metabolites using ultrahigh-performance LC-MS/MS. A total of 633 metabolites were detected, including 572 in WK1, 573 in YK1, and 576 in YK2 (Supplementaty Table S2). These metabolites were classified into 20 functional categories, with the major categories being Others (17.38%); Ketones, Aldehydes, and Acids (15.80%); Alkaloids (12.48%); Flavonoids (8.37%); and Terpenoids (7.42%) (Fig. 1A, Supplementaty Table S3). Sesquiterpenes (31.91%) and triterpenes (25.53%) accounted for the major components among the identified terpenoids.

Principal component anaylsis (PCA) of the detected metabolites was performed to identify differential metabolites. As shown in Fig. 1B, significant differences in metabolites were found among the non-bitter, moderately bitter, and highly bitter Luffa fruits. A total of 364 metabolites were characterized as DAMs, including 162, 242, and 249 DAMs in the WK1 vs. YK1, YK1 vs. YK2, and WK1 vs. YK2 comparisons, respectively (Fig. 1C; Supplementaty Tables S4–6). Among the DAMs, a total of 62, 151, and 128 metabolites were up-regulated, whereas 100, 91, and 121 metabolites were down-regulated in the WK1 vs. YK1, YK1 vs. YK2, and WK1 vs. YK2 comparisons, respectively. Further analysis revealed that the types of compounds in fruits without bitterness did not significantly different from those with bitterness, and some even showed a decreasing trend with increasing bitterness. However, number of terpenoid compounds increased when moving from no bitterness (31 terpenoids) to moderately bitter (37 terpenoids) and then to highly bitter (40 terpenoids) (Supplementaty Tables 7).

Metabolite KEGG enrichment analysis revealed that the majority with the most significant enrichment found for carbohydrate metabolic pathways, followed by membrane transport and amino acid metabolism pathways (Fig. 1D). The content of triterpenoids in bitter fruits was significantly higher than that in non-bitter fruits and showed an increasing trend, with the highest content found in YK2, including the cucurbitacins cuA, cuB, and cuD(Supplementaty Tables2). Furthermore, seven DAMs were enriched in various alkaloid biosynthetic pathways (ko00996): Salicylic Acid, 2-(Methylamino)Benzoic Acid, ,Nsc 12465, Vanillylamine, Vasicine, CuB and cuD(Supplementaty Tables15). Enrichment of the metabolite cuB and cuD were commonly observed in all three comparisons of WK1 vs. YK1, YK1 vs. YK2, and WK1 vs. YK2 (Supplementaty Tables 4–6).

The global transcriptomes of the Luffa fruits with three different bitterness phenotypes were determined by RNA-seq. A total of 19.42 Gb, 18.36 Gb, and 18.20 Gb clean reads were obtained from WK1, YK1, and YK2, respectively. The percentage of bases with a quality value greater than 20 (Q20) and (Q30) exceeded 91.61% and 91.08%, respectively. The GC content (relative to total bases) of the clean reads ranged from 44.80 to 45.72% (Supplementaty Table S8). Following quality control of the raw sequences, the transcriptome data were deemed to be suitable for subsequent downstream analysis.

A total of 57,454 genes (88.3%) were functionally annotated, including 5725 genes (8.8%) commonly annotated in eight transcriptome sequencing databases (Supplementaty Table S9). The top five species matched in the non-redundant (NR) database, in terms of proportion, were Cucumis melo (12,865, 23%), Momordica charantia (11,047, 19%), Cucurbita moschata (7,877, 14%), Cucurbita pepo (7186, 13%), and Cucumis sativus (6211, 11%), indicating their close relationship with Luffa (Fig. 2). GO analysis further classified these genes into 7884 GO terms, which were categorized into biological process, cellular component, and molecular function categories. Metabolic process and cellular process were the most significant gene-enriched terms in the biological process category. In the cellular component category, cellular anatomical entity had the highest level of gene enrichment, whereas catalytic activity and binding were the top two gene-enriched terms in the molecular function category (Supplementaty Table S10). According to the KEGG pathway analysis, 35,851 genes were divided into five expression branches and 136 metabolic pathway maps. The five branches were metabolism, genetic information processing, environmental information processing, cellular processes, and organismal systems (Supplementaty Tables 11).

Comparison of the transcriptome data among the three bitterness types of Luffa revealed a total of 25,577 significant DEGs, including 7877 DEGs in the comparison of WK1 vs. YK1, 20,942 DEGs in the comparison of YK1 vs. YK2, and 18,657 DEGs in the comparison of WK1 vs. YK2 (Fig. 3A). Therefore, a greater number of DEGs was observed in the comparison of the moderately bitter (YK1) and highly bitter (YK2) Luffa fruit (Fig. 3B).

The DEGs were then subjected to KEGG enrichment analysis and functional classification. We found subtle changes in gene expression in Luffa according to bitterness phenotype, with up-regulated DEGs acting on 128 metabolic pathways from non-bitter fruits (WK1) to moderately bitter fruits (YK1), and up-regulated DEGs enriched in 135 metabolic pathways from moderately bitter fruits (YK1) to highly bitter fruits (YK2), suggesting that the differential genes were more actively expressed in bitter Luffa (Supplementaty Tables 16). The DEGs were clearly different in the three different comparisons. For example, compared with the DEGs obtained in the comparison of WK1 vs. YK2, the DEGs in the comparison of YK1 vs. YK2 were more actively involved in pathways such as terpenoid skeleton biosynthesis (68), sesquiterpenoid and triterpenoid biosynthesis (20), and diterpenoid biosynthesis (29). Few DEGs overall were found in the comparison of WK1 vs. YK1 (Fig. 3C). Interestingly, the number of these DEGs related to bitterness was consistent with the number of metabolites in the metabolic pathways of samples in the three comparisons. Therefore, these DEGs may be key genes involved in the metabolism of different bitter taste compounds.

We identified that the genes involved in the triterpenoid biosynthetic pathway were differentially expressed in fruits exhibiting varying levels of bitterness. (Fig. 4A). Based on the gene expression levels (fragments per kilobase of transcript per million mapped reads values), we classified the gene characteristics and expression levels of enzymes involved in various metabolic processes (Fig. 4B; Supplementaty Tables S12-13). In the MVP pathway, a total of 17 genes, encoding six enzymes [ACAT, HMGCS, HMGC, mevalonate kinase (MVK), phosphomevalonate kinase (PMVK), and mevalonate diphosphate decarboxylase (MVD)], are predicted to be involved in the biosynthesis of triterpenoid metabolites based on homology with genes and pathways identified in Luffa [23]. First, acetyl-CoA is catalyzed by three ACAT-encoding genes to produce acetoacetyl-CoA; subsequently, two HMGCS genes synthesize acetoacetyl-CoA into 3-hydroxy-3-methylglutaryl-CoA, followed by five HMGCR genes that reduce 3-hydroxy-3-methylglutaryl-CoA to mevalonate. Two genes encoding MVKs then convert mevalonate to mevalonate-5-phosphate, which is converted to isopentenyl-5-diphosphate under catalysis of MVD encoded by two genes. Thirteen genes encoding seven enzymes in the MEP pathway were identified, including six genes encoding 1-deoxy-d-xylulose-5-phosphate synthase, two genes encoding 1-deoxy-d-xylulose-5-phosphate reductoisomerase, and one gene each encoding 2-C-methyl-d-erythritol 4-phosphate cytidylyltransferase, 4-diphosphocytidyl-2-C-methyl-d-erythritol kinase, 2-C-methyl-d-erythritol 2,4-cyclodiphosphate synthase, 4-hydroxy-3-methylbut-2-en-1-yl diphosphate synthase, and 4-hydroxy-3-methylbut-2-en-1-yl diphosphate reductase. In addition, four genes encode isopentenyl diphosphate delta-isomerase, and five and four genes encode geranyl pyrophosphate synthase and farnesyl diphosphate synthase, respectively. Moreover, seven genes were enriched that are associated with squalene synthase (SS) and 2,3-oxidosqualene cyclase (SE).

OSC is involved in the first committed step in the modification of plant triterpenoids from a linear to cyclic form. In this study, two Luffa genes (Lac05g013070 and gene.Maker00013650) were annotated as cucurbitadienol synthase (LaCBS1 and LaCBS2) and two genes (Lac08g011790 and gene.Maker00025502) were annotated as β-amyrin synthase (LaBAS1 and LaBAS2). The coding sequence of LaCBS1 with 2717 nucleotides exhibited 97.1% sequence similarity to TcCBS of Trichosanthes cucumerina, 89.2% similarity to SgCBS of Siraitia grosvenorii, and 88.9% similarity to McCBS of Momordica charantia. Similarly, LaCBS2 with 2199 nucleotides showed 99.5% similarity to TcCBS, 91.0% similarity to McCBS, and 89.0% similar to CpCBS of Cucurbita pepo. LaBAS, with a coding sequence of 2890 nucleotides, showed 9.3%, 90.1%, and 89.0% similarity to TcCBS, McCBS, and CpCBS, respectively, while LaBAS2, with a coding sequence of 2424 nucleotides, showed 99.8%, 90.4%, and 89.7% similarity to TcCBS, McCBS, and CpCBS, respectively. Consistently, these four OSCs, LaCBS1, LaCBS2, LaBAS1, and LaBAS2, were phylogenetically categorized in the groups of cucurbitadienol synthase and β-amyrin synthase (Fig. 4C).

A BLAST nucleotide search and comparison with data from other Cucurbitaceae plants such as cucumber and melon, Luffa showed similar expression patterns of related gene clusters involved in the biosynthesis of cucurbitacins. Interestingly, no cucurbitacin synthesis-related genes were expressed in WK1 and one gene cluster (one LaBi(CBS)gene, six CYP450 genes, and one LaACT gene) was involved in cucurbitacin biosynthesis in YK1, whereas two gene clusters (cluster 1: one LaBi gene, five CYP450 genes, and one LaACT gene; cluster 2: one LaBi(2) gene, six CYP450(2) genes, and one LaACT(2) gene) were found to be involved in the biosynthesis of cucurbitacin in YK2. These two gene clusters are co-expressed with the cucurbitadienol synthase module involved in the synthesis of cucurbitacin. Combined with metabolomics analysis, these gene clusters may be involved in the biosynthesis of cuB and cuD. First, LaBi (Lac05g013070) encodes the enzyme that catalyzes the cyclization of 2,3-oxidosqualene to form cucurbitadienol, which is the first committed step in cucurbitacin biosynthesis. Second, under the catalysis of several CYP450-encoded enzymes (including the genes Lac05g013040, Lac07g007990, Lac05g013060, Lac05g013090, Lac06g003140, Lac02g004580), various intermediates of cucurbitadienol are formed and cuD is produced. Finally, cuD undergoes acetylation under the action of LaACT (Lac05g013080) to ultimately produce cuB(Shang et al, 2014; Shibuya et al, 2004). Notably, both parents of YK2 have a no-bitterness phenotype. These findings thus suggest that the highly bitter phenotype that emerged after hybridization is related to a shift toward a similar expression pattern to that of the moderately bitter fruit YK1. Another gene cluster (one LaLaBi(2): gene.Maker00013650; six CYP450s: NEW 5702, gene.Maker00001220, NEW 5703, gene.Maker00014196, gene.Maker00036685, gene.Maker00038613; and one LaACT(2): gene.Maker00015020) was introgressed from Luffa cylindrica, and the original two gene clusters in the two parents were in a silent state. After hybridization, these two gene clusters were activated and co-expressed, resulting in greater production of cuD and cuB (Fig. 5). This can explain why the fruit of the hybrids of non-bitter Luffa parents becomes highly bitter after hybridization.

In cucurbit plants, bHLH transcription factors play important roles in plant growth, development, and physiological processes, and some bHLH transcription factors are even directly involved in regulating cucurbitacin biosynthesis [33]. Some are present in gene clusters, acting as bHLH transcriptional activators to regulate the biosynthesis of cuC, cuB, and cuE in Cucumis sativus, Cucumis melo, and Citrullus lanatus, respectively. A total of 746 transcription factors were identified as DEGs in Luffa, with 103 of them being bHLH transcription factors. Furthermore, 286 transcription factors were identified as DEGs in the comparison of WK1 vs. YK1, including 42 bHLH transcription factors. In contrast, 632 and 564 transcription factors were identified in the comparisons of YK1 vs. YK2 and WK1 vs. YK2, respectively, of which 78 and 74 were bHLH transcription factors, respectively. These findings suggested that the difference appears to be differential expression of the transcription factors (Fig. 6A).

Compared with the Bt transcription factor of cucumber and melon, the homologous gene LaBt (Lac02g016030) was not detected in any of the three samples. However, LaBt is located within a gene cluster containing two other bHLH transcription factors(Lac02g016040, Lac02g016080), and Lac02g016040 was also not detected in the three samples. In contrast, Lac02g016080 had significantly higher expression in the bitter fruits (YK1 and YK2) than in the non-bitter fruit (WK1) and showed an up-regulation trend. In addition, gene.Maker00003045 encodes a specific transcription factor found only in bitter samples (YK2), which is highly homologous to Lac02g016080 (Supplementaty Table 14). Referring to the cucumber transcription factor CsBt, Bt is a fruit-specific transcription factor that evolved from the original wild ancestor (with a bitter taste) through artificial domestication into cultivated varieties (with a non-bitter taste). In this study, LaBt was not detected in the three samples of Luffa, which may be due to the fact that LaBt is only expressed in wild species, while the three samples selected for this study are all cultivated or hybrid varieties. Therefore, the two bHLH transcription factors, Lac02g016080 and gene.Maker00003045, may be involved in the biosynthesis of cuB and cuD.

Finally, we used Luffa seedlings to investigate whether heat stress and ABA treatment activates cucurbitacin biosynthesis genes. The LaBi (Lac05g013070), LaACT (Lac05g013080), and LaCYP450 (Lac05g013040, Lac07g007990, Lac05g013050, Lac05g013060, Lac02g004580) genes in the cucurbitacin biosynthesis pathway were selected among the DEGs identified in the RNA-seq data for validation with qRT-PCR analysis. The results showed that all of the selected genes cucurbitacin biosynthesis were significantly differentially expressed between the heat stress and control groups, with up-regulated expression in the stress condition. Similarly, for ABA treatment, all cucurbitacin synthesis-related genes showed significantly higher expression levels compared to those of the control group (Fig. 6B, C). These results indicated that abiotic stresses can activate cucurbitacin biosynthesis genes, leading to increased production of cucurbitacin, confirming the trends found in the RNA-seq data.

Cucurbitaceae plants contain a special and representative class of metabolic compounds known for their bitterness and toxicity called cucurbitacins. As the name implies, cucurbitacins were initially discovered only in Cucurbitaceae plants; however, they have also been extracted from plants in other families such as Brassicaceae, Malvaceae, and Primulaceae [34, 35]. There are many types of cucurbitacins, which vary according to modification of the synthesis backbone (cucurbitadienol) at different positions, including oxidation, hydroxylation, and acetylation. Curcurbitacins are classified into 12 types, named cucurbitacin A-T [36]. Therefore, different Cucurbitaceae plants may contain different bitter substances. For instance, cuC is the main bitter compound found in cucumber [20], whereas the main bitter compound in melon is cuB [22], and the bitter taste in watermelon is mainly attributed to cuE [15, 37]. In this study, we conducted quantitative and qualitative analysis of bitter metabolites in Luffa using HPLC-MS, which identified 633 metabolites, including 17 triterpenoids or triterpenoid glycosides (Supplementaty Table 2). Among the triterpenoid compounds, cuA, cuB, and cuD were detected, and quantitative analysis revealed that the contents of these three cucurbitacins in bitter fruits (YK1 and YK2) were significantly higher than those in the non-bitter fruit (WK1). Similarly, Zhao et al.[23] discovered several cucurbitacins (including cuA, cuD, cuF, and iso-cuB) in bitter Luffa fruits. In this study, the content of cuA was relatively lower in the three samples than in other plants belonging to the Cucurbitaceae family [23]. CuA is a toxic and bitter medicinal compound that exhibits effective activity against ovarian cancer cells [38]. The abundance of cuB and cuD was higher in the bitter fruits of Luffa, especially in YK2 fruits. CuD is synthesized from the triterpene precursor 2,3-oxidosqualene to form the cucurbitacin skeleton cucurbitadienol, which is then oxidized by members of the CYP450 family (CYP87D, CYP81Q32, and CYP705A5) to eventually form cuD. CuD is then acetylated to form cuB by ACT. Both cucurbitacins can be interconverted under the action of CoA and ACT[22].

Triterpenoids are highly diverse natural compounds present throughout the Plant kingdom, with their diversity stemming from rich carbon skeletons and various oxidation and glycosylation modifications at different skeleton positions [39, 40]. Almost all triterpenoid skeletons are derived from the biosynthetic precursor 2,3-oxidosqualene, which undergoes cyclization by OSCs to form various triterpene skeleton forms [41]. Moreover, multiple different triterpene skeletons can coexist in the same plant or tissue, which are catalyzed by different OSCs encoded by different genes [42]. In this study, we identified two common triterpenoid compounds, β-amyrin and lupeol and their respective triterpene synthases β-amyrin synthase and lupeol synthase. β-amyrin is catalyzed by two OSCs: LaBAS1 (Lac08g011790) and LaBAS2 (gene.Maker00025502), which catalyze the cyclization of 2,3-oxidosqualene to form the tetracyclic dammarenyl cation, followed by further oxidation and expansion to form the pentacyclic lupeol cation, and then another cyclization to form β-amyrin. Lupeol is catalyzed by one OSC (LUP:Lac08g011750), directly forming lupeol after formation of the intermediate lupeol cation. β-amyrin was only detected in the fruit of YK2, while lupeol was only found in WK1, with relatively low levels of these metabolites. β-amyrin can serve as a precursor of triterpenoids such as triterpene glycosides, oleanolic acid, and other triterpenoid compounds with various pharmacological effects. Lupeol, as a component of the cell membrane structure, participates in the growth and development of plants.

Cucurbits show specificity in the composition of metabolic compounds that are associated with the bitter taste of the Cucurbitaceae family, which are biosynthesized by folding the 2,3-oxidosqualene C-B-C conformation to form the cucurbitacin skeleton [16, 43]. Based on our gene expression analysis, we hypothesized that two genes, LaCBS1 (Lac05g013070) and LaCBS2 (gene.Maker00013650), may play vital roles in the formation of bitterness compounds in Luffa fruits. The two cucurbitacin metabolites cuB and cuD were found to be significantly up-regulated in the metabolite quantification analysis of bitter fruits and were barely detectable in non-bitter fruits. Therefore, cuB and cuD are likely the major contributors to the bitterness in Luffa fruits. Accordingly, the biosynthetic pathways of cuB and cuD are worth investigating further.

2,3-Oxidosqualene represents a branch point between primary metabolism of membrane sterols and secondary metabolism, acting as an important intermediate that has attracted the attention of researchers in broad fields of biology and chemistry. Bi/CBS is is a critical enzyme in the biosynthetic pathway of cucurbitacin, situated within a gene cluster that includes other functionally related genes implicated in cucurbitacin biosynthesis. For example, a gene cluster comprising one Bi/CBS, seven CYP450, and one ACT gene was identified in cucumber; a gene cluster comprising one Bi/CBS, seven CYP450, and one ACT was identified in melon; and a similar gene cluster comprising one Bi/CBS, 10 CYP450, and one ACT was identified in watermelon [21, 22]. In Luffa, the biosynthesis of cuB and cuD is predicted to be initiated by the cucurbitadienol synthase gene (Bi/CBS) based on transcriptional and metabolic analyses, which first cyclizes 2,3-oxidosqualene to form cucurbitadienol as the first step in cucurbitacin synthesis. This distinguishes CBS from other triterpenoids in terms of its specificity and criticality, and from which all the other triterpenoid compounds derive their basic cucurbitodienol backbones. Subsequently, the oxidation of carbon 11 and carbon 3-β hydroxylation catalyzed by members of the CYP450 family (e.g., CYP87D18) is followed by carbon 20-β hydroxylation catalyzed by CYP87D18 to form 11-carbonyloxy-20 β-hydroxycucurbitadienol, and by carbon 2-β hydroxylation catalyzed by CYP81Q59 to form 11-carbonyloxy-2 β,20 β-dihydroxycucurbitadienol. Carbon 2 is then oxidized to form cuD. Finally, cuD is acylated by ACT to form cuB. Interestingly, in this study, we found two gene clusters involved in cucurbitacin biosynthesis in Luffa. In the bitter fruit of YK1, one LaBi(Lac05g013070), six LaP450s (Lac05g013040, Lac07g007990, Lac05g013060, Lac05g013090, Lac06g003140, Lac02g004580), and one LaACT (Lac05g013080) are present in a gene cluster involved in the biosynthesis of cuB and cuD. Among them, Lac05g013090 and Lac06g003140 are members of the CYP87A family; Lac05g013040 and Lac05g013060 belong to the CYP81Q family subfamily; and Lac02g004580 belongs to the CYP705A family. However, in the bitter fruit of YK2, two gene clusters were involved in the biosynthesis of cuB and cuD, with the other gene cluster consisting of one LaBi2(gene.Maker00013650), six LaP450s (NEW 5702, gene.Maker00001220, NEW 5703, gene.Maker00014196, gene.Maker00036685, gene.Maker00038613), and one LaACT2 (gene.Maker00015020). These two gene clusters may jointly participate in cucurbitacin biosynthesis, resulting in greater accumulation of cucurbitacin and a highly bitte taste in the Luffa fruit.

Hybrid vigor or heterosis is a common phenomenon in nature, which has sparked great interest among scholars studying hybrid genetic relationship and biological genetic variation [44, 45]. When the kinship between two parents is too distant, a barrier to hybridization occurs, resulting in reproductive isolation [46]. Interspecific hybridization in Luffa can also lead to hybrid vigor, significantly enhancing the growth and development process of Luffa; increasing fruit retention ability, yield, and resistance to adversity and disease; and even introducing the advantageous traits of wild species into cultivated varieties, thereby improving targeted traits for genetic breeding and providing a wide range of interspecific materials for Luffa breeding and molecular marker development [5]. However, interspecific hybridization in Luffa also encounters some hybridization barriers such as the failure of flowering, pollen sterility, and even very bitter fruits. In previous research work, it found that almost all F₁ hybrids of Luffa acutangula and Luffa cylindrica had highly bitter fruits, whereas neither parent had a bitterness phenotype [3, 47]. Therefore, we sought to use the fruits of F₁ hybrids as materials to investigate whether the bitterness traits of other Luffa fruits (moderately bitter) differs from that of F₁ hybrids (highly bitter), and explore how interspecific hybridization in Luffa affects fruit bitterness, aiming to elucidate the biosynthesis and molecular mechanism of bitterness regulation in fruits.

In the current study, metabolomic analysis revealed a total of 633 metabolites for the three samples. In the comparisons of WK1 vs. YK1 and YK1 vs. YK2, there were 162 and 242 DAMs, respectively, of which 62 and 151 DAMs were up-regulated, respectively. Furthermore, the bitter compounds cuA, cuB, and cuD were significantly up-regulated. Additionally, metabolic analysis showed that the contents of cuB and cuD in the interspecific F₁ hybrid YK2 (with a highly bitter fruit) were markedly increased compared to those of YK1 Luffa fruits with moderate bitterness. The gene clusters involved in the cucurbitacin biosynthesis of Luffa (LaBi/CBS1, P450, LaACT) were all up-regulated in bitter fruits. The expression levels of bitter genes in YK2 were significantly higher than those in YK1. This indicates that interspecific hybridization of Luffa has a significant impact on the growth and development of the fruit, resulting in stronger bitterness. According to previous studies, interspecific hybridization between Luffa acutangula and Luffa cylindrica does not result in any issues with respect to fruit set, but reproductive isolation still occurs [3]. Interspecific hybridization within the Luffa genus may result in introgression of genetic material from Luffa cylindrica into hybrid progeny.