Using the amino acid sequences of LiAATase1 (UniProt code A0A0K0LCG5) and LiAATase2 (UniProt code A0A0K0LBP0) (Figure 1), we predicted their secondary structures via the NPS@ server (Combet et al., 2000). The predicted secondary structures of LiAATase1 and LiAATase2 consist of alpha helices (37.17% and 33.78% of residues, respectively), along with multiple strands and coils (Figure 2). LiAATase1 contains 155 residues in helices, while LiAATase2 has 151.

The three-dimensional (3D) structures of LiAATase1 and LiAATase2 were predicted using AlphaFold2 (Jumper et al., 2021; Wayment-Steele et al., 2023), a deep learning-based tool that provides highly accurate and reliable protein structure predictions, surpassing traditional homology modeling methods.

To evaluate the quality of the predicted structures (Figures 3a,d), we employed the Ramachandran plot to analyze the dihedral angles of the protein backbones, ensuring they fell within acceptable regions, which indicates a valid protein conformation. For LiAATase1, 88.7% of the residues were in the most favored region, 10.8% in the additionally allowed region, 0.5% in the generously allowed region, and none in the disallowed region (Figure 3b; Table 1). For LiAATase2, 90.5% of residues were in the most favored region, 9.2% in the additionally allowed region, 0.3% in the generously allowed region, and none in the disallowed region (Figure 3e; Table 1).

ProSA analysis of the models showed Z-scores of −10.90 for LiAATase1 and -10.79 for LiAATase2 (Figures 3c,f). Both Z-scores were well below −10.00, demonstrating excellent model quality. However, the slight difference between the two Z-scores does not suggest a significant structural discrepancy in the model.

Although the overall fold of LiAATase1 closely resembles that of LiAATase2 (Figure 4), the root mean square deviation (RMSD) for all atoms was 2.25 Å, and the sequence identity was 27.47% (Figure 4).

Using the predicted models (Figures 3a,d, 4, 5a,b), we employed the GalaxyWEB program (Ko et al., 2012; Heo et al., 2013; Heo et al., 2016; Seok et al., 2021) to identify the active sites of LiAATase1 and LiAATase2. The analysis revealed that the active site residues of LiAATase1 include I156, R225, P238, S239, R240, V241, H273, A274, V275, N276 (Figures 5a,c). For LiAATase2, the active site residues are T254, S256, K257, F258, N285, T286, V287, N288, K334, and D368 (Figures 5b,c). These residues are likely involved in interactions with the alcohol substrate, potentially forming bonds with the substrate’s side chain atoms.

To determine the optimal conditions for alcohol acetyltransferase activity of LiAATase1 and LiAATase2, we examined the effects of temperature and pH on enzyme activity (Figure 6). Activity was tested across a temperature range of 21–39°C, revealing an increase in activity up to 30°C, after which it declined (Figures 6a,b). The enzymes showed optimal activity within a pH range of 5.5–9.5, with peak activity occurring at pH 7.5 before decreasing (Figures 6c,d). Based on these results, we selected pH 7.5 and 30 °C as the optimal conditions for further activity assessments.

Alcohol acetyltransferases are well-known for their role in producing acetate esters and are often associated with ethylene-dependent or ripening-specific processes in many plants (Beekwilder et al., 2004; Defilippi et al., 2005; Aschariyaphotha et al., 2024). The activity of LiAATases was determined under the optimal conditions (pH 7.5 at 30 °C). We found that both LiAATase1 and LiAATase2 demonstrated high activity with medium-chain alcohols, particularly 1-hexanol, while showing minimal activity with short-chain alcohols like methanol and ethanol. The enzyme activity increased with the length of the alcohol chain, with 1-hexanol exhibiting 4.1 to 5.5 times higher activity than methanol (Figure 7). This trend may be attributed to alcohol acyltransferase’s preference for utilizing acyl-CoA residues, as the enzyme tends to react more readily with acyl-CoA substitutes other than acetyl-CoA.

To establish a detailed spatial and temporal expression profile, RT-qPCR (real-time quantitative polymerase chain reaction) was used to quantify the expression levels of the target genes LiAATase1 and LiAATase2 across different flower organs and developmental stages. The results showed that LiAATase1 expression was highest in the petals (Figure 8a), while LiAATase2 expression peaked in the pistils (Figure 8b). Regarding developmental stages, LiAATase1 expression was greatest at stage 4 (Figure 8c), whereas LiAATase2 expression was highest at stage 1 (Figure 8d). This tissue- and stage-specific expression pattern highlights the significant role of LiAATase1 and LiAATase2 in lavender essential oil biosynthesis, with flower tissues being the primary site of expression.

The amino acid sequences of LiAATase1 (UniProt code A0A0K0LCG5) and LiAATase2 (UniProt code A0A0K0LBP0) were analyzed using the ProtParam online server (https://web.expasy.org/protparam/) to predict their chemical properties and physicochemical parameters.

Structural predictions for LiAATase1 and LiAATase2 were performed using the AlphaFold2 program (Jumper et al., 2021; Wayment-Steele et al., 2023). Secondary structures were predicted with the NPS@ server (Combet et al., 2000), while active site residues were identified using the GalaxyWEB program (Ko et al., 2012; Heo et al., 2013; Heo et al., 2016; Seok et al., 2021). Multiple sequence alignment data were obtained from the LSQKAB program within the CCP4 suite (Collaborative Computational Project, 1994), and the root mean square deviation (RMSD) for Cα atoms was calculated. Structural images were generated using PyMOL 2.3.4 (https://www.pymol.org/2/).

To validate the tertiary structures, Ramachandran plots for LiAATase1 and LiAATase2 were generated using the PDBsum database (Laskowski, 2004; de Beer et al., 2014; Laskowski et al., 2017; Laskowski, 2022). This tool assesses protein structure quality by detecting geometric errors, thereby improving the accuracy of the models. The Ramachandran plot specifically evaluates the stereochemical properties of the structures by displaying the dihedral angles of amino acid residues, highlighting the allowed conformational regions and identifying disallowed orientations.

Additionally, ProSA (Protein Structure Analysis) is a widely used tool for analyzing and validating predicted protein models (Wiederstein and Sippl, 2007). The z-score reflects overall model quality and is plotted against the z-scores of all experimentally determined protein chains in the current PDB. The plot distinguishes between structure groups (e.g., X-ray, NMR) using color-coding. This allows assessment of whether the input structure’s z-score falls within the expected range for native proteins of comparable size.

LiAATase1 and LiAATase2 were isolated and purified with slight modifications to previously described methods (Egea-Cortines et al., 2019; Nazeer et al., 2019). Fresh flowers were first washed with tap water, then with distilled water, and dried in the shade for 5–6 days at room temperature. The dried samples were powdered in liquid nitrogen using a mortar and pestle. Total protein was extracted using a Tris-buffer saline solution (150 mM NaCl, 20 mM Tris-HCl, PVP, pH 7.4), with a slight adjustment to the extraction buffer ratio (1:7, w/v). The sample was filtered through three layers of muslin cloth, and the filtrate was stirred magnetically at 4 °C overnight. The sample was then centrifuged at 18,000 rpm for 30 min at 4°C, and the supernatant was collected while discarding the pellet.

For protein purification, the crude extract was precipitated using varying ammonium sulfate saturation levels (20%, 35%, 55%, 75%, and 90%). A 75% saturated ammonium sulfate solution yielded the best quality pellet after centrifugation at 18,000 rpm for 30 min at 4 °C. The supernatant was then precipitated with 75% ammonium sulfate saturation and stored at −20 °C overnight to ensure complete protein precipitation. The following morning, the sample was centrifuged again at 18,000 rpm for 30 min at 4°C, with the supernatant discarded. The resulting pellet was washed six times with acetone and then dried. The protein was dialyzed against distilled water for 12 h at 4°C, and purified using size exclusion chromatography (SEC).

Enzyme assays were conducted in 25-mL glass syringes with Luer lock caps, following a previously described method with minor modifications (Hu et al., 2016; Stribny et al., 2016; Egea-Cortines et al., 2019). The reaction mixtures contained a glycerol buffer (50 mM potassium phosphate, pH 7.5, 10% (w/v) glycerol), which included higher alcohol as a co-substrate, acetyl-CoA as the second co-substrate, and a cell extract. The volume ratio of higher alcohol, acetyl-CoA, and cell extract was 10:1:4, with a final reaction volume of 1.5 mL. Isoamyl alcohol (0.01–100 mM final concentration), isobutanol (60 mM), or 2-phenylethanol (30 mM) were used as substrates, along with 0.8 mM acetyl-CoA. Substrate concentrations were carefully measured. After adding all components, entrapped air was removed using the plunger, and the syringe was attached to an orbital shaker. Following a 30-min incubation, 1.5-mL samples were transferred to 15-mL vials containing 0.35 g NaCl for ester quantification via gas chromatography. Enzyme activity was terminated by adding 60 mL of a saturated KSCN solution.

Gene expression of LiAATase1 and LiAATase2 was quantified using real-time quantitative polymerase chain reaction (RT-qPCR) with PowerUp SYBR Green Master Mix (Applied Biosystems). Total RNA was extracted from various developmental stages and tissues using the Universal Plant Total RNA Extraction Kit (Bioteke, Beijing, China) according to the manufacturer’s instructions. cDNA was synthesized from the RNA samples using the PrimeScript first Strand cDNA Synthesis Kit (Takara, Kyoto, Japan). The primers used in the experiments are listed in Table 2. RT-qPCR analysis was performed on an Applied Biosystems QuantStudio five instrument. Data were analyzed using the 2^−ΔΔCT method (Livak and Schmittgen, 2001; Schmittgen and Livak, 2008; Hawkins and Guest, 2017; Green and Sambrook, 2018; Liu et al., 2024a; Liu et al., 2024b; Liu et al., 2025a; Liu et al., 2025b; Liu et al., 2025c), and relative expression ratios were presented as log₂ values in histograms. A ratio greater than zero indicated upregulation, while a ratio less than zero indicated downregulation. Beta-actin was used as a reference gene for normalization, and a positive control with beta-actin was included in the analysis.

All experiments were conducted at least in triplicate. The data were expressed as mean ± SD. Statistical analysis was conducted using Origin 8.5, Microsoft Excel 2013 and SPSS 19.0. In the all statistical evaluations, p < 0.05 was considered statistically significant, and p < 0.01 was considered high statistically significant.