The cultivar of E. ferox for the experiment was ‘ZHSQ’, which was planted in the aquatic vegetable experimental field (plot size: 18.9m², water source: tap water) in Yangzhou University and grown under natural conditions (Yangzhou, China, 2019). Seed kernels were collected from 10 days after flowering (DAF10) to 40 days after flowering (DAF40). Arabidopsis thaliana (Columbia) and tobacco (Nicotiana benthamiana) seeds were preserved for our laboratory and grown in a culture chamber (at 25 ± 2°C, under 12 h of light and 12 h of darkness, Yangzhou, China) for subsequent experiments.

The amino acid sequences of four PALs (AT2G37040, AT3G53260, AT3G10340, AT5G04230) from Arabidopsis thaliana (https://www.arabidopsis.org/) were blast in the protein database of E. ferox by Tbtools software (Chen et al., 2020). After that, the hidden Markov model (PF00221, http://pfam.xfam.org/) of PAL domain was also searched in the protein database of gorgonians. 12 candidate PAL genes were obtained by combining the results of both searches. Finally, they were further confirmed by CD-search in the NCBI database (https://www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi). Molecular weight, and isoelectric point were predicted by the ExPASy tool (https://www.expasy.org/).

Conserved motifs were analyzed in MEME Suite (https://meme-suite.org/meme/tools/meme), and gene structure information was obtained from the GFF files of the E. ferox genome, and then all were visualized by Tbtools (Chen et al., 2020).

Amino acid sequence alignment was performed by clustal W (MEGA 7 software), followed by neighbor-joining method to construct phylogenetic tree with 1000 bootstrap replicates, further perfected using EVOLVIEW (https://www.evolgenius.info/evolview/#/login). The amino acid sequences of all PALs are listed in Table S1 . The chromosomal positions of PALs were obtained from annotation files of the E. ferox and N. colorata genome. The syntenic relationships of PALs in E. ferox and N. colorata were analyzed using the MCScanX in TbTools and visualized using Advanced Circos in Tbtools (Chen et al., 2020).

Total RNA extraction from seed kernels at different developmental stages (DAF10-DAF40) of E. ferox and Arabidopsis leaves was performed using Plant RNA extraction kit (Takara, Dalian, China). Then, HiScript^® II Q RT SuperMix (Vazyme, NanJin, China) was applied to reverse transcription into cDNA. The 20 μL qRT-PCR reaction volume consisted of 10 μL 2×ChamQ SYBR qPCR Master Mix (Vazyme, Nanjing, China), 0.4 μL forward primer, 0.4 μL reverse primer, 1.0 μL cDNA template, and 8.2 μL ddH₂O. EfUBQ5 (ID: EF11G001150) and AtActin2 (ID: AT3G18780) were used as the internal reference genes (Feng et al., 2018; Wu et al., 2022c). Primers were designed using Primer Premier 5.0. Supplementary Table S2 for the gene specific primer sequences. The amplification was performed on a CFX-96 Real-time PCR system (Bio-Rad) with a real-time fluorescence quantitative PCR program: 95°C for 30s, 95°C for 10s, and 60°C for 30s, for a total of 40 cycles. The amplification efficiency of qPCR was determined by gradient dilution of the concentration of cDNA from E. ferox seed kernels 30 days after flowering (DAF30). Relative gene expression was calculated using 2^-ΔΔCT (Livak and Schmittgen, 2001). Each amplification reaction contained three biological and technical replicates.

The ORF sequences of EfPAL1 and EfPAL2 were recombined into 16318-GFP vector with the restriction enzymes Sal I and BamH I (Takara, Dalian, China) using ClonExpress II One Step Cloning Kit (Vazyme, Nanjing, China). The leaves of Arabidopsis thaliana were placed in the enzymolysis solution (1.5% Cellulase R10, 0.4% Macerozyme R10, 0.4M Mannitol, 20mM KCl, 20mM MES) and then reacted for 2-3 h on a shaker (40 rpm) protected from light. After termination of the reaction by W5 solution (154mM NaCl, 125mM CaCl₂, 5mM KCl, 2mM MES, pH=5.7), the protoplasts were resuspended by MMG solution (0.2M Mannitol, 15mM MgCl₂, 4mM MES, pH=5.7). 10 μg of recombinant plasmids p35S::PAL1-GFP and p35S::PAL2-GFP were added to 200 μL of Arabidopsis protoplasts and left to stand for 5 min before adding 210 ul of PEG solution (PEG 4000, 0.2M Mannitol, 0.1M CaCl₂) and left to stand for 30 min at 23°C. Then, 800uL W5 solution was added and mixed and centrifuged for 8min (23°C, 100g). The protoplasts were resuspended in 1 mL WI solution (0.5M Mannitol, 20mM KCl, 4mM MES, pH5.7), incubated overnight at 23°C in low light and then centrifuged to remove the supernatant. The protoplasts were suspended by adding 500uL WI solution, mixed with 5uL Dil, and incubated for 20min under dark conditions. The WI solution was washed the protoplasts three times. Finally, 100uL WI solution was added to resuspend the protoplasts, and the fluorescence signals were observed under ultra-high resolution laser confocal microscope (Leica, TCS SP8 STED, Germany). The empty vector was used as a control.

A Nicotiana benthamiana transient expression system applied to the subcellular localization analysis of EfZAT11 and EfHY5. The empty vector pCAMBIA1300-35S-sGFP was digested with Sac I and Xba I (Takara, Dalian, China). The ORF sequences of EfZAT11 and EfHY5 were reconstituted into the 1300-GFP vector to form p35S::ZAT11-GFP and p35S::HY5-GFP (Vazyme, Nanjing, China). The recombinant plasmids p35S::ZAT11-GFP and p35S::HY5-GFP were transferred into agroinfiltration using the freeze-thaw method, and then their OD₆₀₀ values were adjusted to 0.8-1.0 with the infection solution (100 mM Acetosyringone, 0.5 M MES, 10 mM MgCl₂, pH5.6), mixed with nuclear marker 1:1 (v/v) and left for 2-3 hours. Six-week-old N. benthamiana leaves were injected with a syringe on the back of the leaves until the entire leaf was soaked. The fluorescent signals were observed under a microscope. The primers used above are listed in Table S3 .

The ORF sequences of EfPAL1 and EfPAL2 were inserted into the Pcold-TF vector using a double digestion method (BamH1 and Sal1). The recombinant plasmid was transferred into BL21 (DE3) E. coli and shaken at 37°C to an OD₆₀₀ value of 0.4-0.6, then isopropyl-β-D-thiogalactopyranoside (IPTG, 0.5 mM) was added to induce protein expression at 16°C for 16 h. The cells were collected by centrifugation for 5 min (4°C, 13778g). The cells were resuspended in PBS buffer (137mM NaCl, 2.67mM KCl, 10mM Na₂HPO₄, 2mM KH₂PO₄), added lysozyme (1 mg/ml) and stood for 5 min, followed by sonication on ice for 20 min (run 4s with 8s interval). The supernatant was obtained after centrifugation (13778g, 4°C) for 10 min. The Ni-Agarose column (CWBIO, Jiangsu, China) is equilibrated with 10 mL of binding buffer (50mM Imidazole, 500mM NaCl, 20mM Tris-HCl, pH=7.9), the supernatant is put on the column, then the column is washed with 15 mL of binding buffer, and finally the target protein is eluted with 5 mL of elution buffer (500mM Imidazole, 500mM NaCl, 20mM Tris-HCl, pH=7.9). The purified target proteins were subjected to SDS-PAGE analysis, The concentration of the target protein was determined using Albumin Bovine V as the standard, according to the description of the Bradford protein Assay Kit (Beyotime, Shanghai, China). The primers used above are listed in Table S3 .

The total reaction system was 1 mL, consisting of 1mM L-Phe/L-Tyr, 10μg EfPAL1/EfPAL1 recombinant protein, and 100mM borate buffer (pH=9.0). After 5 min at 70°C, the reaction was terminated by adding 100 μL of 6 mM HCl. The reaction mixture was centrifuged at 13778g at 4°C for 10 min, and the supernatant was analyzed by HPLC (Agilent 1260). A C18 column (4.6×250 mm, 5 µm, Mars) was applied with 0.1% phosphoric acid in ultrapure water (v/v) as solvent A and acetonitrile solvent B. The elution gradients were as follows: 20% B at 0 min, 95% B at 25 min, and 20% B at 25.01 min. A flow rate of 1.0 ml/min and a column temperature of 25°C were applied and the chromatogram was obtained at 290 nm.

The ORF sequences of EfPAL1 and EfPAL2 were inserted into the pCAMBIA1300-35S-sGFP vector by homologous recombination and then transferred into A. tumefaciens GV3101 cells. The infection solution (5% sucrose solution supplemented with silwet-77) was used to resuspend A. tumefaciens cells to reach an OD₆₀₀ value of 0.8. The 5-6-week-old Arabidopsis thaliana (with the opened flowers and fruit pods removed) were selected and their flower buds were immersed into agroinfiltration infection solution for 20-30s, incubated in the dark for 24h and then cultured normally. The infection was repeated once a week later. Positive plants were screened on 1/2 MS solid medium containing hygromycin B (50 mg L^-1) to T3 generation for subsequent experiments. The plant DNA extraction kit (Tiangen, Beijing) was used to extract DNA from 5-week-old Arabidopsis leaves. The extracted DNA was used as a template for PCR amplification with the target gene’s forward primer (EfPAL1-F/EfPAL2-F) and vector primer (GFP-R). The primers used above are listed in Table S3 .

2g of 5-week-old Arabidopsis leaves were weighed, dried and ground into powder. The content of total flavonoids in 0.02 g leaves was determined by spectrophotometry according to the protocol of Plant Flavonoid Assay Kit (Suzhou Comin Biotechnology Co., Ltd., Suzhou, China).

0.1 g leaves of five-week-old Arabidopsis thaliana were ground on ice until homogenized, and the enzyme activity was determined according to the protocol of the Phenylalanine ammonialyase kit (Suzhou Comin Biotechnology Co., Ltd., Suzhou, China).

Seed kernels (10, 20, 30 and 40 DAF) (Wu et al., 2021)and leaves (diameter range 15-200 cm) (Wu et al., 2022b) of E. ferox were subjected to cDNA library construction (oebiotech, Shanghai, China). The promoter of EfPAL2 (1000 bp immediately upstream of the EfPAL2 open reading frame) was reconstituted into the pAbAi vector using the restriction enzymes Kpn I and Sal I. The recombinant plasmids were linearized using restriction endonuclease BstbI. The linearized pPAL2-1000-PAbAi was transferred into YIH yeast using the Yeastmaker™ Yeast Transformation System 2 (Clontech) and grown on plates of SD/-Ura for three days. The colonies were resuspended with 0.9% NaCl, its OD₆₀₀ value was adjusted to 0.002, and then coated on plates containing different concentrations (100-1000 ng/ml) of Aureobasidin A (AbA) and screened to the lowest AbA concentration that inhibited its auto-activation. The competent cell was produced with Y1HGold containing pPAL2-1000-PAbAi, then transferred to the library plasmid and incubated on SD/-Leu+AbA¹⁰⁰ plates at 30°C for 3-5 days. Positive clones verified by PCR were sequenced and then annotated by blastx in NCBI (https://www.ncbi.nlm.nih.gov/). The primers used above are listed in Table S3 .

The ORF sequences of ZAT11 and HY5 were inserted into the pGADT7 AD vector via restriction enzymes EcoR I and BamH I. The two recombinant plasmids were then transferred into a competent made with Y1HGold cells containing pPAL2-1000-PAbAi and cultured in SD/-Leu+AbA¹⁰⁰ medium at 30°C for 3-5 days.

ProPAL2 was cloned into pGreenII 0800-LUC vector to fuse luciferase (LUC) reporter gene, EfZAT11 and EfHY5 were cloned into pGreenII 62-SK vectors, and then they were transformed into the A. tumefaciens GV3101 strain. A. tumefaciens cultures containing pGreen-ProPAL2 were mixed with those strains harboring pGreenII62-Sk-EfZAT11 or pGreenII62-Sk-EfHY5 at a ratio of 1:1 (v/v) and infiltrated into the N. benthamiana leaves after standing in darkness for 3 hours. Firefly LUC and Renilla LUC activity was detected with the Duo-LiteTM Luciferase Assay System after 3 days of incubation. Meanwhile, the leaves were sprayed with 100mM luciferin for CDD imaging. The primers used above are listed in Table S3 .

Experimental data were analyzed using GraphPad Prism 8 software. Statistical significance were tested using multiple t tests (* P < 0.05, ** P < 0.01).

Based on the hidden Markov model (HMM) and blastp function in Tbtools, 12 candidate PAL genes were identified in the E. ferox genome. The 12 putative PAL genes were distributed on 7 chromosomes, with PAL1, PAL4, PAL11, and PAL12 located on EF13, EF24, EF27, and EF29, respectively, PAL3, PAL5, and PAL6 were located on EF05, PAL2, PAL7 and PAL8 were located on EF20, and PAL9 and PAL10 were located on EF23 ( Figure 1A ). The full-length coding sequences (CDS) of the 12 putative EfPALs ranged from 1926 bp to 2286 bp with amino acid numbers varying from 641aa to 761aa. There was a wide variation in their molecular weights (Mw), ranging from 69.20 KDa to 83.71 KDa, with isoelectric points (pI) between 6.08 and 6.60 ( Table 2 ).

Obviously, these 12 putative PALs with strictly conserved motifs, all of which contained motif 1, motif 2, motif 3, motif 4, motif 5, motif 6, motif 7 and motif 10. Only PAL5 had no motif 8 and PAL6 had no motif 9 ( Figure 1B ). Meanwhile, these 12 PALs all possessed five conserved structural domains, including PLN02457, phe_am_lyase, Lyase_aromatic, PAL-HAL, and HutH ( Figure 1B ). This well reflects the conservation of the PAL structural domain in E. ferox. In addition, gene structure analysis showed that PAL1, PAL2, PAL3, PAL4, PAL9 and PAL10 had no introns, PAL7, PAL8, PAL11 and PAL12 contained one intron, and PAL5 and PAL6 had two introns ( Figure 1B ). It is hypothesized that differences in gene structure may result in differences in function.

Phylogenetic tree was constructed by using amino acid sequences of the 12 PALs in E. ferox with those in Arabidopsis thaliana, Vitis vinifera, Oryza sativa and Nymphaea colorata. The EfPALs are most closely related to N. colorata, in which NcPAL4 and NcPAL2 correspond to the three PAL genes in E. ferox, respectively, suggesting that these six PALs (EfPAL1-6) have been fully conserved in evolution and may have critical functions ( Figure 1C ). Besides, VvPAL3 and VvPAL4 clustered with NcPAL1 and EfPAL1. Moreover, synteny analysis was performed for PALs in E. ferox and N. colorata. As a result, there were four replication events in PALs from E. ferox, including 8 whole-genome duplications (WGD), 2 dispersed duplications (DSD), 1 tandem duplication (TD) and 1 proximal duplication (PD). There are 4 WGD and 1 PD presented in the PALs of N. colorata ( Table S4 ). This suggested that whole-genome duplications (WGD) was likely to be the major driving force for PAL gene evolution in E. ferox and N. colorata. Meanwhile, the microcollinearity analysis demonstrated that 13 colinearity gene pairs were existed between E. ferox and N. colorata, among which NcPAL2 and NcPAL4 exhibited a one-to five relationships with EfPAL, respectively ( Figure 1A , Table S5 ).

A differential genes PAL (c55946) identified in the transcriptome were aligned with 12 putative PALs before EfPAL1 and EfPAL2 were selected for further study (Liu et al., 2018b). First of all, the expression patterns of EfPAL1 and EfPAL2 during seed kernel development (DAF10-DAF40) were analyzed by qPCR. EfPAL1 and EfPAL2 increased from DAF10 to DAF30 and decreased from DAF30 to DAF40, while the relative expression of EfPAL2 was significantly higher than that of EfPAL1 ( Figure S1 , Table S6 ). Interestingly, the expression trends of these two genes were similar to the previously measured flavonoid metabolite profiles (Wu et al., 2021), further suggesting that EfPAL1 and EfPAL2 are likely to be key genes for flavonoid biosynthesis in E. ferox seed kernels.

To characterize the function of EfPAL1 and EfPAL2, their CDS regions were ligated into Pcold-TF vector and transformed into E. coli BL21(DE3) for prokaryotic expression. The expression product with six His tags was purified by nickel column and confirmed by SDS-PAGE ( Figure S2A ). Enzyme activity analysis was performed using phenylalanine and tyrosine as substrates. With phenylalanine as the substrate, EfPAL1 and EfPAL2 could catalyze the production of trans-cinnamic acid, while with tyrosine as the substrate, the production of trans-cinnamic acid did not occur ( Figure 2 ). This result revealed that EfPAL1 and EfPAL2 possessed enzymatic activity, and only for phenylalanine. Besides, the enzyme activities of purified EfPAL1 and EfPAL2 proteins were compared. As a result, the enzymatic activity of EfPAL2 was significantly higher than that of EfPAL1, which was 2.37 times higher ( Figure S2B , Table S7 ). This suggested that EfPAL2 might play a leading role in the biosynthesis of flavonoids in E. ferox.

To further demonstrate the role of EfPAL1 and EfPAL2 in flavonoid biosynthesis, we overexpressed EfPAL1 and EfPAL2 in A. thaliana and obtained three overexpression lines (OE-lines), respectively. Above all, target bands of 2000-3000 bp were obtained by RT-PCR amplification in three OE-lines of EfPAL1 and EfPAL2, respectively ( Figures S3A, B ). And the qPCR results showed that the expression of EfPAL1 and EfPAL2 in three OE-lines was significantly higher than that of the wild type (WT) ( Figure 3A , Table S8 ). Subsequently, the total flavonoid content was determined and it was found that OE-lines were significantly higher than WT, indicating that EfPAL1 and EfPAL2 promote the biosynthesis of flavonoids ( Figure 3B , Table S8 ). In addition, the PAL enzyme activity was significantly higher in the transgenic lines than in the WT ( Figure S3C , Table S8 ). These results further suggested that EfPAL1 and EfPAL2 were the main enzymes for flavonoid biosynthesis in E. ferox. Then, the ORF of EfPAL1 and EfPAL2 were inserted into the 16318-GFP vector and the recombinant plasmids were transformed into A. thaliana protoplasts. The GFP signal was observed under laser scanning confocal microscope, with the empty vector as the control. In A. thaliana protoplasts transfected with p35S::PAL1-GFP and p35S::PAL2-GFP, the GFP fluorescent signal was observed mainly in the cytoplasm, while the empty vector was expressed throughout the cell ( Figure 3C ). These results indicated that EfPAL1 and EfPAL2 proteins were localized in the cytoplasm.

The EfPAL2 promoter (1000 bp) was used as a bait to screen in the E. ferox membrane system yeast library by yeast one-hybrid (Y1H) system, and a total of 22 colonies were obtained and sequenced ( Table S9 ). Two transcription factors, ZAT11 and HY5, were identified and the YIH assay further validated the interaction of ZAT11 and HY5 with the EfPAL2 promoter. The results showed that EfZAT11-AD+EfPAL2 promoter and E fHY5-AD+EfPAL2 promoter could grow on SD-Leu¹⁰⁰ medium, but AD+EfPAL2 promoter could not grow ( Figure 4A , Figure S4 ). This indicated that EfZAT11 and EfHY5 could directly bind to the promoter of EfPAL2 to regulate its expression. Dual luciferase assays were performed to detect the regulatory effects of EfZAT11 and EfHY5 on EfPAL2. When EfZAT11 was co-injected with ProPAL2::LUC, LUC activity was significantly increased (up-regulated 2.41-fold), while EfHY5 was co-injected with ProPAL2::LUC, LUC activity was significantly decreased (down-regulated 0.57-fold) ( Figures 4B, C , Table S10 ). This suggested that EfZAT11 promoted EfPAL2 expression and EfHY5 repressed EfPAL2 expression. Furthermore, bioluminescence images of firefly LUC confirmed the results ( Figure 4D ).

Further, expression analysis of EfZAT11 and EfHY5 was performed by transcriptome and qPCR. In the transcriptome, the expression of EfZAT11 decreased from DAF10 to DAF20 and increased from DAF20 to DAF40, which is similar to the expression trend of PAL (c55946) ( Figures S1 , S5 , Table S6 ). EfHY5 increased from DAF10 to DAF30 and decreased from DAF30 to DAF40, which is opposite to the expression trend of c55946 ( Figures S1 , S5 , Table S6 ). Meanwhile, qPCR analysis showed that the relative expression of EfZAT11 increased from DAF10 to DAF20, slightly decreased from DAF20 to DAF30, and increased from DAF30 to DAF40, which is roughly similar to the expression trend of EfPAL2 ( Figurs S1 , S5 , Table S6 ). EfHY5 increased from DAF10 to DAF20 and decreased from DAF20 to DAF40, which is opposite to the expression trend of EfPAL2. ( Figures S1 , S5 , Table S6 ) These results indicated that EfZAT11 may promote flavonoid biosynthesis and EfHY5 inhibits flavonoid biosynthesis. Finally, the fluorescent signals of p35S::ZAT11-GFP and p35S::HY5-GFP were observed only in the nucleus and overlapped with nuclear markers, suggesting that EfZAT11 and EfHY5 are nuclear-localized transcription factors ( Figure 4E ).