Strategy and efficiency-redefined discovery of novel nanomolar tyrosinase inhibitors: AI de novo molecular generation + expert-guided structural optimization

Graphical abstract

Highlights

• •A reinforcement learning model was designed for de novo molecular generation targeting tyrosinase.
• •Generated molecules exhibited optimal target affinity and good synthetic feasibility.
• •Expert-guided structural optimization was performed based on AI-generated lead V.
• •The activity of optimized compound V-24 increased from micromole to nanomole level.
• •AI generation + Expert-guided optimization represents an efficiency-redefined strategy.

Abstract

Artificial intelligence (AI) has played an excellent supporting role in novel drug discovery and development. This study introduces a reinforcement learning (RL) model based on the Soft Actor-Critic (SAC) algorithm for AI-driven de novo molecular generation targeting tyrosinase. The model facilitates forward molecular generation design by integrating a chemical reaction template and a molecular building block library, concurrently performing molecular docking and assessing drug-likeness. Through sequential decision-making, signal feedback, and a dynamic learning process, the model generates molecules exhibiting potent target affinity, optimal drug-like properties, and good synthetic feasibility. The AI-generated molecules undergo rigorous manual screening, synthesis, and biological evaluation, culminating in the identification of a prioritized lead compound V. Subsequent structural optimization of compound V reveals a series of compounds with significantly enhanced activity, shifting inhibitory potency from the micromolar to the nanomolar range. The optimized compound, V-24, demonstrates low cytotoxicity and significant anti-melanogenic activity both in cell melanogenesis inhibition and zebrafish anti-pigmentation models. Notably, it effectively reduces melanin content in an ultraviolet light-induced human 3D skin pigmentation model, exhibiting the potential to serve as a promising tyrosinase inhibitor for the treatment of skin pigmentation. More importantly, this “AI de novo Molecular Generation + Expert-Guided Structural Optimization” work demonstrates that integrating an AI algorithm with traditional medicinal chemistry experience is a novel approach and efficiency-redefined strategy for drug discovery.

Introduction

The color of human skin is primarily determined by melanin, which consists of two main types: eumelanin (black or dark brown) and pheomelanin (red or yellow).[1] Both types of melanin contribute to the absorption of ultraviolet rays, thereby protecting the skin from environmental damage.[2] However, excessive deposition of pigment can result in a range of skin disorders, including chloasma, butterfly rash, freckles, acanthosis nigricans, Riehl’s melanosis, skin aging and melanoma in severe cases.[[3], [4], [5], [6]] The biosynthesis of melanin relies on tyrosinase (TYR), a crucial catalytic enzyme in the double-loop oxidation process initiated by L-tyrosine.[7,8].

TYR is classified as a type 3 binuclear copper metalloenzyme, characterized by a central structure comprising six highly conserved histidine residues and two copper ion-binding sites.[9,10] It plays a critical role in both the monophenolase catalytic cycle of L-tyrosine and the diphenolase catalytic cycle of L-dopa, ultimately resulting in the formation of dopachrome, which is essential for subsequent melanin synthesis (Fig. 1).[[11], [12], [13]] As the key rate-limiting enzyme in the melanin synthesis pathway, tyrosinase has attracted considerable attention as a primary therapeutic target for inhibiting pigment deposition.

Fig. 1.

The tyrosinase-catalyzed production mechanism of melanin.

Traditional TYR inhibitors predominantly consist of structural analogues of the substrates L-tyrosine and L-dopa, as well as resorcinol analogues.[[14], [15], [16], [17]] These include various natural products, including kojic acid, arbutin, resveratrol, polyphenols, flavonoids, stilbenes, and lignans (Fig. 2A).[[18], [19], [20], [21], [22], [23], [24]] These natural TYR inhibitors typically demonstrate weak activity and often necessitate high concentrations for efficacy. However, high concentrations may elevate the risk of skin irritation and other potential side effects, such as liver burden from kojic acid metabolites. Moreover, these inhibitors tend to possess high hydrophilicity, leading to poor skin permeability, and some are susceptible to degradation from light, heat, or oxidation (e.g., kojic acid and resveratrol), which restricts their practical application. Additionally, the extraction of natural products can incur elevated costs due to low extraction yields, challenges in standardization, and complex formulation requirements, complicating large-scale application. Consequently, there is an urgent need for the development of more efficient, safe, and stable TYR inhibitors.

Fig. 2.

Representative TYR inhibitors.

Recently, structural modification and optimization efforts have led to the discovery of a diverse array of small molecule tyrosinase inhibitors exhibiting high activity, including derivatives of kojic acid, azole derivatives, thiourea derivatives, amide and thioamide derivatives, cinnamic acid derivatives, and benzopentacyclic heterocycle derivatives (Fig. 2B).[[25], [26], [27], [28], [29], [30], [31], [32], [33]] Synthetic TYR inhibitors have demonstrated enhancements in both activity and stability relative to natural TYR inhibitors. Nonetheless, these compounds exhibit several limitations, including structural characteristics that may pose safety risks (e.g., kojic acid, thioureas, and azole derivatives), the potential for resistance development, and complex synthesis pathways for more intricate compounds. Additionally, it is crucial to recognize that these inhibitors are predominantly derived from the structural modification of natural products, resulting in low efficiency in drug discovery and generally limited structural novelty. Consequently, there is an urgent need to implement innovative and efficient drug discovery approaches to develop novel TYR inhibitors.

To accelerate the identification of structurally novel, highly active, and easily synthesizable TYR inhibitors, this study utilized an artificial intelligence (AI)-driven de novo molecular generation strategy targeting TYR.[34] After identifying a promising lead compound, traditional structural modification and optimization approaches were subsequently performed to further advance the lead compound toward becoming a potential candidate compound.

In recent years, AI technologies, particularly the subset of deep learning (DL), have achieved significant advancements in the field of drug discovery.[[35], [36], [37]] This progress has led to the development of multiple DL-based molecular generation algorithms, many of which have been effectively utilized in drug discovery and drug design.[[38], [39], [40], [41]] Among these models, de novo drug design targeting specific protein targets stands out as particularly appealing. This approach effectively reduces the vast chemical space to a more manageable range, facilitating the targeted identification and acquisition of tool compounds.[[42], [43], [44]] Additionally, it serves as an ideal starting point for hit-to-lead optimization, thereby highlighting the formidable molecular design capabilities of AI models.[45].

Many de novo molecular generation methods utilize structure-based approaches to develop ligands within target pockets.[46,47] These methods including FRAME, ResGen, and DiffGui, consider the detailed 3D structures of protein binding pockets, enabling the generation of molecules with high target affinity and favorable drug-like properties.[[48], [49], [50]] However, some methods may not guarantee that the generated molecules exhibited desirable chemical synthesizability, thereby limiting the practical viability of what initially appears to be a highly efficient strategy. To address these limitations, DL algorithms based on synthons were introduced, including BBAR, Synnet, DeepLigBuilder, and the DOGS software developed by Hartenfeller et al.[[51], [52], [53], [54]] These advanced techniques aim to directly generate active compounds with synthetic accessibility, tailored to specific protein targets.

From an AI perspective, molecular synthesis can be conceptualized as a sequential decision-making process that entails a series of steps to construct a target molecule; each step requires the selection of suitable reaction conditions and raw materials. The choices made during synthesis impact not only the feasibility of the final product but also other critical properties. These decisions generate feedback signals that can be utilized as reward signals in reinforcement learning (RL), guiding algorithms to discover optimal synthesis pathways.[45,55] Additionally, molecular synthesis encompasses a vast array of intricate chemical knowledge within a dynamically changing environment, which makes it challenging to represent using deterministic rules. RL is particularly adept at addressing these complex dynamic decision-making problems.[56] It offers an effective modeling framework for molecular synthesis pathway planning, enabling the comprehensive use of feedback signals and facilitating the development of interpretable and controllable optimal synthesis strategies in intricate environments, thereby significantly enhancing the synthesizability of molecules and achieving multi-objective optimization.[57,58].

In this study, we employed an RL-driven forward synthesis algorithm.[34] This approach leverages existing chemical reaction templates and commercially available molecular building blocks as starting points for de novo generation of novel molecules through forward reaction prediction, thereby ensuring the synthetic accessibility of the generated compounds. During the molecular generation process, the model concurrently conducts real-time molecular docking and drug property prediction to guarantee that the generated molecules demonstrate both high target affinity and favorable pharmacological properties. Moreover, RL enables the model to continuously refine the properties of the generated molecules to enhance the quality of the final compounds (Fig. 3).

Fig. 3.

The overall process of lead compound AI-generation and traditional lead optimization.

Using this strategy for de novo molecular generation targeting TYR, we identified a series of AI-generated molecules with novel scaffolds and high TYR affinity. Through TYR inhibitory activity screening of AI-generated molecules, a lead compound was identified with micromolar activity. Subsequently, we conducted systematic structural modifications and structure–activity relationship (SAR) analysis, ultimately obtaining a compound exhibiting nanomolar TYR inhibitory activity. This “AI Generation + Expert Optimization” drug discovery strategy—where AI-driven molecule generation yields lead compounds, followed by expert-guided structural optimization—demonstrates good feasibility and efficiency of integrating AI de novo molecule design with conventional medicinal chemistry approaches.

Materials and methods

De novo molecular generation method

The molecular generation process is systematically categorized into three primary stages: the initial stage, the generation stage, and the post-processing stage, as depicted in Fig. 4. The specific process and algorithm can refer to previous studies.[34].

Fig. 4.

The process of AI de novo molecular generation.

In the initial stage, molecular docking is performed individually using Vina software to evaluate each molecule from a predefined library of starting molecular building blocks against the target protein. The top 200 fragments exhibiting the highest docking scores are then selected to create the initial fragment set. This stage ensures that the molecules generated in subsequent phases possess multiple potential pharmacophores as a foundational element.

In the generation phase, an RL model is utilized to guide the stepwise synthesis of molecules. During this process, each generation round randomly selects a molecule from the initial fragment set to serve as the starting state. Based on the current strategy, an action is determined by selecting an appropriate reaction template (as detailed in previous studies)[59] to perform chemical modifications on the chosen molecule. The resultant products are then docked with the target protein using Vina software to assess their binding affinity, which is subsequently used to compute a reward function. Each operational step of the generation is documented and stored in a replay buffer pool, enabling random sampling during subsequent model training for iterative updates to the model parameters. The pseudocode presented below offers a comprehensive description of the training process for the described model. It is crucial to highlight that the molecular generation process follows specific termination conditions, including reaching a maximum of three synthetic steps, exceeding a molecular weight of 600 Da, or terminating prematurely in cases where a suitable reaction template cannot be identified or when the product fails to be successfully generated.

Finally, in the post-processing phase, an initial screening based on Lipinski’s Rule of Five is performed on a virtual library of 20,000 small molecules generated by the previously mentioned RL model to ensure that the candidate molecules exhibit favorable drug-like properties. Subsequently, the algorithm ranks these small molecules according to their binding affinity scores with the target protein and identifies the top 100 candidates along with their corresponding synthetic pathways. Following a manual evaluation by medicinal chemistry experts, approximately 20 of the most promising candidate molecules are selected for progression into the actual synthesis experimental phase.

The source code of Soft Actor-Critic (SAC) and associated data preparation pseudocode are available at https://github.com/sanomics-lab/RL_based_syn (replace the corresponding target protein with TYR, PDB ID: 2Y9X). The reaction template library is also available at this link.

Chemistry

All chemicals were procured from Bidepharm, Aladdin or Sigma Aldrich and were utilized without any subsequent purification. For both reaction and column chromatography (CC), solvents of analytical grade (Shanghai Chemical Reagents Co., Ltd., China) were employed, while solvents for HPLC were of HPLC grade (J & K Scientific Ltd.). Silica gel (100–200, 200–300 mesh, Qingdao Haiyang Chemical Co., Ltd., China) was utilized for CC. For thin-layer chromatography (TLC), precoated silica gel GF254 plates (Qingdao Haiyang Chemical Co., Ltd., China) were employed. The progress of the reaction was monitored using TLC, and visualization was achieved with UV light or iodine staining. Melting points were determined with an X-4BII + microscope melting point apparatus. ¹H NMR and ¹³C NMR spectra were recorded on Bruker instruments (500 MHz and 125 MHz, respectively), with tetramethylsilane used as the internal standard. High-resolution mass spectra were obtained using an Agilent 1290LC-6530QTOF instrument, employing electrospray ionization as the ion source. The purity of all compounds was evaluated through NMR spectroscopy and HPLC analysis. All compounds showed ≥ 98 % purity by HPLC (SHIMADZU Labsolutions, UV detection at λ = 254 nm), utilizing an Agilent C18 column (4.6 × 250 mm, 5 μm) and eluting at a flow rate of 0.5 mL/min with methanol as mobile phase.

General procedure for the preparation of intermediates 13, 52, 61, 75–76, 78 and compounds I-V, XIII, XIV, V-1, V-2, V-4-V-7, V-9, V-10, V-14, V-16, V-17, V-23-V-34. To a 50 mL round-bottom flask was added amines 1, 5, 6, 11, 7b, 26, 35–37, 44, 47, 50–51, 54, 57, 60, 69, 72 or 85–88 (1.0 mmol), carboxylate 2–4, 12, 27, 34, 39–40, 73–74, 77 or 89–90 (1.2 mmol), HOBt (1.3 mmol), EDCI (1.3 mmol), DIPEA (3.0 mmol) and DMF (5 mL) sequentially. The resulting mixture was stirred at room temperature for 4 h and monitored by TLC. Upon completion of the reaction, EA and H₂O were added and the aqueous layer was extracted with EA (3 × 20 mL). The combined organic layers were washed with saturated aqueous NaCl, dried over Na₂SO₄ and concentrated under reduced pressure. The resulting residue was further purified using thin layer chromatography (DCM:MeOH = 30:1, twice) to obtain intermediates 13, 52, 61, 75–76, 78 and compounds I-V, XIII, XIV, V-1, V-2, V-4-V-7, V-9, V-10, V-14, V-16, V-17, V-23-V-34.

General procedure for the preparation of compounds VI, XV, V-3. To a 50 mL round-bottom flask was added amines 7, 28 or 37 (1.0 mmol), carboxylate 8 or 29 or 38 (1.2 mmol), HOBt (1.3 mmol), EDCI (1.3 mmol), DIPEA (3.0 mmol) and DMF (5 mL) sequentially. The resulting mixture was stirred at room temperature for 4 h and monitored by TLC. Upon completion of the reaction, H₂O was added, the precipitate was washed with EA, filtered and dried to obtain compounds VI, XV, V-3.

General procedure for the preparation of intermediates 19, 22 and compounds VII, XII. To a 25 mL round-bottom flask was added amines 9, 17, 21 or 23 (1.0 mmol), carboxylate 10, 18 or 20 (1.2 mmol), HATU (1.5 mmol), DIPEA (3.0 mmol) and DMF (5 mL) sequentially. The resulting mixture was stirred at room temperature for 2 h and monitored by TLC. Upon completion of the reaction, EA and H₂O were added and the aqueous layer was extracted with EA (3 × 20 mL). The combined organic layers were washed with saturated aqueous NaCl, dried over Na₂SO₄ and concentrated under reduced pressure. The resulting residue was further purified using thin layer chromatography (DCM or DCM:MeOH = 100:1 or DCM:MeOH = 50:1, thrice) to obtain intermediates 19, 22, and compounds VII, XII.

General procedure for the preparation of compounds VIII, X, XI, V-20-V-22. To a 50 mL round-bottom flask was added intermediate 13 (0.3 mmol) and DCM (8 mL) sequentially. The resulting mixture was stirred at −78 ℃ and then dropwise added DCM solution of BBr₃ (3 mmol). The mixture was reacted at room temperature for 3 h and monitored by TLC. Upon completing the reaction, ice H₂O was added and adjust the pH to 7 with saturated NaHCO₃ solution. Removal of DCM by vacuum concentration, the aqueous layer was extracted with EA (3 × 20 mL). The combined organic layers were washed with saturated aqueous NaCl, dried over Na₂SO₄ and concentrated under reduced pressure. The resulting residue was further purified using thin layer chromatography (DCM:MeOH = 20:1) to obtain compound VIII, V-20-V-2. For compounds X and XI, upon completing the reaction, ice H₂O was added and the precipitate was washed with EA, filtered and dried to obtain compounds X and XI.

Synthetic procedure for the preparation of compound IX. To a 50 mL round-bottom flask was added 14 (1.0 mmol), 15 (0.87 mmol), DIPEA (2.6 mmol) and MeCN (18 mL) sequentially. The resulting mixture was stirred at room temperature for 8 h and monitored by TLC. Upon completing the reaction, the mixture was extracted with DCM (3 × 20 mL). The combined organic layers were washed with saturated aqueous NaCl, dried over Na₂SO₄ and concentrated under reduced pressure. The resulting residue was further purified using thin layer chromatography (DCM:MeOH: ammonium hydroxide = 30:1:0.1) to obtain 16.

To a 50 mL round-bottom flask was added intermediate 16 (0.3 mmol) and MeOH (5 mL) sequentially, then dropwise added 10 % NaOH solution (0.7 mL). The resulting mixture was stirred at room temperature for 8 h and monitored by TLC. Upon completion of the reaction, the mixture was concentrated under reduced pressure and added H₂O (3 mL). Adjust pH to 3–4 using 2 mol/L HCl solution at 0 ℃. The precipitate was washed with EA, filtered and dried to obtain compound IX.

General procedure for the preparation of intermediates 25, 43, 46, 49, 56, 59, 81, 84 and compounds V-8, V-11, V-19. To a 100 mL round-bottom flask was added 24, 42, 45, 48, 53–54 or 62 (1.0 mmol), various benzyl bromide or benzyl chloride (1.5 mmol), K₂CO₃ (2.0 mmol) or Et₃N (2.0 mmol) and MeCN (8 mL) sequentially. The resulting mixture was stirred at room temperature for 4 h and monitored by TLC. Upon completing the reaction, the mixture was concentrated under reduced pressure. EA and H₂O were added and the aqueous layer was extracted with EA (3 × 20 mL). The combined organic layers were washed with saturated aqueous NaCl, dried over Na₂SO₄ and concentrated under reduced pressure. The resulting residue was further purified using column chromatography (DCM:MeOH = 100:1–20:1) to obtain intermediates 25, 43, 46, 49, 56, 59, 81, 84 and compounds V-8, V-11, V-19.

General procedure for the preparation of intermediates 26, 44, 47, 50, 53, 57, 60, 62, 72, 85–88. The mixture of 25, 43, 46, 49, 52, 56, 59, 61, 71 or 81–84 (1.0 mmol) in MeOH (5 mL) was cooled to 0 °C. Then, dioxane hydrochloride solution (3 mL) was added dropwise, and the resulting mixture was stirred at 0 °C. When the reaction was complete, the solvent and excess hydrochloride were removed under reduced pressure to obtain the crude product. The crude product was further washed with EA to afford intermediates 26, 44, 47, 50, 53, 57, 60, 62, 72, 85–88.

General procedure for the preparation of compounds XVI-XVII. To a 50 mL round-bottom flask was added 28 (1.0 mmol), di-tert-butyl dicarbonate (1.5 mmol), Et₃N (2.0 mmol) and DCM (6 mL) sequentially. The resulting mixture was stirred at 0 ℃ for 2 h and monitored by TLC. Upon completing the reaction, the mixture was washed with H₂O (2 × 20 mL). The combined organic layers were washed with saturated aqueous NaCl, dried over Na₂SO₄ and concentrated under reduced pressure. The resulting residue was further purified using thin layer chromatography (DCM:MeOH = 20:1) to obtain 30.

To a 50 mL round-bottom flask was added 30 (0.5 mmol), carboxylate 31 or 8 (1.2 mmol), DCC (0.8 mmol), DMAP (0.05 mmol) and DCM (8 mL) sequentially. The resulting mixture was stirred at room temperature for 12 h and monitored by TLC. Upon completing the reaction, the mixture was filtered and washed with H₂O (2 × 20 mL). The organic layers were washed with saturated aqueous NaCl, dried over Na₂SO₄ and concentrated under reduced pressure. The resulting residue was further purified using thin layer chromatography (DCM:MeOH = 100:1–50:1) to obtain 32 or 33.

The mixture of 32 or 33 (1.0 mmol) in MeOH (5 mL) was cooled to 0 °C. Then, dioxane hydrochloride solution (3 mL) was added dropwise, and the resulting mixture was stirred at 0 °C. When the reaction was complete, the solvent and excess hydrochloride were removed under reduced pressure to obtain the crude product. The crude product was further washed with EA to afford compounds XVI-XVII.

General procedure for the preparation of intermediate 71 and compounds V-12, V-13, V-15, V-18. To a 50 mL round-bottom flask was added 54 or 62 (1.0 mmol), various benzaldehyde (1.5 mmol), glacial acetic acid (2.4 mmol) and DCE (6 mL) sequentially. The resulting mixture was stirred at room temperature for 2 h and then added sodium triacetoxyborohydride (1.0 mmol). After 2 h, sodium triacetoxyborohydride (1.0 mmol) was added again and monitored by TLC. Upon completion of the reaction, the mixture was concentrated under reduced pressure and neutralized with saturated NaHCO₃ solution. The aqueous layer was extracted with EA (3 × 20 mL). The combined organic layers were washed with saturated aqueous NaCl, dried over Na₂SO₄ and concentrated under reduced pressure. The resulting residue was further purified using thin layer chromatography (DCM:MeOH = 20:1) to obtain intermediate 71 and compounds V-12, V-13, V-15, V-18.

(S)-N-(1-amino-1-oxo-3-phenylpropan-2-yl)-4-hydroxy-3-methoxybenzamide (I). White solid, yield 8 %, m.p. 208.0–208.8 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.52 (s, 1H), 8.24 (d, J = 8.5 Hz, 1H), 7.50 – 7.48 (m, 1H), 7.37 – 7.28 (m, 4H), 7.24 (t, J = 7.6 Hz, 2H), 7.18 – 7.12 (m, 1H), 7.07 (d, J = 2.1 Hz, 1H), 6.78 (d, J = 8.2 Hz, 1H), 4.61 (ddd, J = 10.6, 8.4, 4.3 Hz, 1H), 3.80 (s, 3H), 3.09 (dd, J = 13.7, 4.3 Hz, 1H), 2.97 (dd, J = 13.6, 10.5 Hz, 1H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 173.6, 165.9, 149.5, 147.0, 138.7, 129.2, 128.0, 126.2, 125.2, 121.0, 114.7, 111.5, 55.7, 54.7, 37.3. ESI-HRMS: m/z calcd for C₁₇H₁₈N₂O₄Na [M + Na]⁺: 337.1159; found: 337.1156. HPLC analysis: retention time = 5.641 min; peak area, 99.04 %.

(S)-N-(1-amino-1-oxo-3-phenylpropan-2-yl)-3-(3-hydroxyphenyl)propanamide (II). White solid, yield 44 %, m.p. 47.0–47.9 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.22 (s, 1H), 8.01 (d, J = 8.5 Hz, 1H), 7.38 (s, 1H), 7.25 (t, J = 7.2 Hz, 2H), 7.22 – 7.15 (m, 3H), 7.05 – 7.00 (m, 2H), 6.58 – 6.52 (m, 3H), 4.45 (td, J = 9.0, 4.8 Hz, 1H), 2.98 (dd, J = 13.7, 4.8 Hz, 1H), 2.74 (dd, J = 13.7, 9.5 Hz, 1H), 2.58 (ddd, J = 8.8, 7.0, 2.4 Hz, 2H), 2.31 (td, J = 7.7, 3.4 Hz, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 173.1, 171.2, 157.2, 142.7, 138.1, 129.1, 129.1, 128.0, 126.1, 118.7, 115.0, 112.8, 53.6, 37.7, 36.8, 31.0. ESI-HRMS: m/z calcd for C₁₈H₂₁N₂O₃ [M + H]⁺: 313.1547; found: 313.1541. HPLC analysis: retention time = 6.232 min; peak area, 98.06 %.

(S)-N-(1-amino-1-oxo-3-phenylpropan-2-yl)-3-hydroxybenzamide (III). White solid, yield 28 %, m.p. 209.6–210.4 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.60 (s, 1H), 8.33 (d, J = 8.5 Hz, 1H), 7.50 (s, 1H), 7.35 – 7.30 (m, 2H), 7.28 – 7.19 (m, 4H), 7.18 – 7.13 (m, 2H), 7.10 (s, 1H), 6.88 (pd, J = 3.9, 2.5 Hz, 1H), 4.60 (ddd, J = 10.5, 8.4, 4.2 Hz, 1H), 3.09 (dd, J = 13.7, 4.2 Hz, 1H), 2.98 (dd, J = 13.8, 10.6 Hz, 1H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 173.4, 166.2, 157.2, 138.6, 135.6, 129.2, 129.1, 128.0, 126.2, 118.1, 117.9, 114.4, 54.7, 37.2. ESI-HRMS: m/z calcd for C₁₆H₁₆N₂O₃Na [M + Na]⁺: 307.1053; found: 307.1051. HPLC analysis: retention time = 5.645 min; peak area, 99.48 %.

N-(1-benzylpiperidin-4-yl)-3-(3-hydroxyphenyl)propanamide (IV). Colorless oil, yield 41 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.21 (s, 1H), 7.70 (d, J = 7.7 Hz, 1H), 7.34 – 7.21 (m, 5H), 7.03 (t, J = 7.7 Hz, 1H), 6.64 – 6.53 (m, 3H), 3.51 (td, J = 7.2, 3.8 Hz, 1H), 3.42 (s, 2H), 2.75 – 2.66 (m, 4H), 2.29 (dd, J = 8.7, 6.9 Hz, 2H), 1.97 (t, J = 11.2 Hz, 2H), 1.66 (d, J = 9.2 Hz, 2H), 1.40 – 1.28 (m, 2H).¹³C NMR (125 MHz, DMSO‑d₆) δ 170.6, 157.3, 142.8, 138.6, 129.1, 128.7, 128.2, 126.8, 118.8, 115.2, 112.8, 62.2, 52.0, 45.9, 37.1, 31.6, 31.2. ESI-HRMS: m/z calcd for C₂₁H₂₇N₂O₂ [M + H]⁺: 339.2067; found: 339.2070. HPLC analysis: retention time = 6.930 min; peak area, 98.85 %.

1-(4-(4-fluorobenzyl)piperazin-1-yl)-3-(3-hydroxyphenyl)propan-1-one (V). White solid, yield 62 %, m.p. 112.7–113.4 °C. ¹H NMR (500 MHz, CDCl₃) δ 7.25 (dd, J = 8.5, 5.5 Hz, 2H), 7.12 (t, J = 7.8 Hz, 1H), 7.02 – 6.96 (m, 2H), 6.74 – 6.67 (m, 3H), 3.62 (t, J = 5.0 Hz, 2H), 3.44 (s, 2H), 3.37 (t, J = 5.0 Hz, 2H), 2.90 (t, J = 10.0 Hz, 2H), 2.61 (t, J = 5.0 Hz, 2H), 2.38 (t, J = 5.0 Hz, 2H), 2.25 (t, J = 5.0 Hz, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 169.8, 162.3 (d, ¹J = 241.3 Hz), 157.3, 142.7, 134.1 (d, ⁴J = 2.5 Hz), 130.7 (d, ³J = 7.5 Hz), 129.1, 119.0, 115.3, 114.9 (d, ²J = 20.0 Hz), 112.8, 60.9, 52.6, 52.2, 44.9, 41.0, 33.9, 30.8. ESI-HRMS: m/z calcd for C₂₀H₂₄FN₂O₂ [M + H]⁺: 343.1816; found: 343.1816. HPLC analysis: retention time = 5.671 min; peak area, 98.46 %.

N-(4-fluorobenzyl)-2-hydroxyquinoline-4-carboxamide (VI). Yellow solid, yield 82 %, m.p. 343.8–344.6 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 11.94 (s, 1H), 9.29 (t, J = 6.0 Hz, 1H), 7.69 (dd, J = 8.1, 1.4 Hz, 1H), 7.53 (ddd, J = 8.4, 7.1, 1.4 Hz, 1H), 7.44 – 7.38 (m, 2H), 7.35 (dd, J = 8.4, 1.2 Hz, 1H), 7.22–7.17 (m, 3H), 6.56 (s, 1H), 4.48 (d, J = 6.0 Hz, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 165.8, 162.2 (d, ¹J = 241.3 Hz), 161.2, 146.0, 139.2, 135.1 (d, ⁴J = 2.5 Hz), 130.9, 129.3 (d, ³J = 8.8 Hz), 125.8, 122.1, 119.8, 116.1, 115.7, 115.2 (d, ²J = 21.3 Hz), 41.7. ESI-HRMS: m/z calcd for C₁₇H₁₄FN₂O₂ [M + H]⁺: 297.1034; found: 297.1033. HPLC analysis: retention time = 7.219 min; peak area, 98.82 %.

(E)-N-(1H-indazol-3-yl)-3-(3-methoxyphenyl)acrylamide (VII). Yellow-green solid, yield 26 %, m.p. 175.5–176.3 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 8.38 (d, J = 8.3 Hz, 1H), 7.94 (d, J = 7.8 Hz, 1H), 7.81 (s, 2H), 7.60 (ddd, J = 8.3, 7.1, 1.2 Hz, 1H), 7.39 (dt, J = 8.2, 7.4 Hz, 2H), 7.34 – 7.28 (m, 2H), 7.04 (ddd, J = 8.2, 2.6, 1.0 Hz, 1H), 6.60 (s, 2H), 3.83 (s, 3H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 162.9, 159.7, 152.9, 143.3, 139.4, 136.1, 130.2, 129.7, 123.9, 120.9, 120.5, 120.4, 118.7, 116.3, 115.5, 113.3, 55.3. ESI-HRMS: m/z calcd for C₁₇H₁₆N₃O₂ [M + H]⁺: 294.1237; found: 294.1233. HPLC analysis: retention time = 9.401 min; peak area, 99.48 %.

N-(3-hydroxyphenethyl)quinoline-5-carboxamide (VIII). White solid, yield 15 %, m.p. 198.4–199.2 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.29 (s, 1H), 8.93 (dd, J = 4.1, 1.7 Hz, 1H), 8.67 (t, J = 5.6 Hz, 1H), 8.46 (ddd, J = 8.6, 1.8, 0.9 Hz, 1H), 8.09 (dt, J = 8.5, 1.1 Hz, 1H), 7.77 (dd, J = 8.5, 7.0 Hz, 1H), 7.65 (dd, J = 7.0, 1.2 Hz, 1H), 7.53 (dd, J = 8.6, 4.2 Hz, 1H), 7.11 (t, J = 7.9 Hz, 1H), 6.70 (dd, J = 7.2, 1.4 Hz, 2H), 6.67 – 6.61 (m, 1H), 3.55 (td, J = 7.3, 5.7 Hz, 2H), 2.81 (t, J = 7.2 Hz, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 167.4, 157.4, 150.8, 147.7, 140.8, 135.1, 133.9, 130.8, 129.3, 128.5, 125.5, 125.1, 121.9, 119.4, 115.7, 113.1, 40.6, 35.0. ESI-HRMS: m/z calcd for C₁₈H₁₇N₂O₂ [M + H]⁺: 293.1285; found: 293.1288. HPLC analysis: retention time = 5.814 min; peak area, 99.57 %.

4-((4-benzoylpiperidin-1-yl)methyl)benzoic acid (IX). White solid, yield 70 %; m.p. 213.6–214.2 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 7.96 (td, J = 7.9, 1.4 Hz, 4H), 7.71 – 7.61 (m, 3H), 7.52 (t, J = 7.7 Hz, 2H), 4.06 (s, 2H), 3.58 (s, 1H), 3.16 (s, 2H), 2.73 (s, 2H), 1.89 (s, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 201.7, 167.1, 135.3, 133.3, 131.0, 130.7, 129.4, 128.9, 128.3, 59.4, 51.1, 48.6, 26.4. ESI-HRMS: m/z calcd for C₂₀H₂₂NO₃ [M + H]⁺: 324.1594; found: 324.1595. HPLC analysis: retention time = 7.151 min; peak area, 99.80 %.

3-(3-hydroxyphenyl)-N-(2-methylbenzyl)propanamide (X). Light yellow solid, yield 91 %, m.p. 115.8–116.5 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.23 (s, 1H), 8.16 (t, J = 5.7 Hz, 1H), 7.14 – 7.03 (m, 5H), 6.63 – 6.57 (m, 3H), 4.22 (d, J = 5.6 Hz, 2H), 2.75 (t, J = 7.7 Hz, 2H), 2.44 – 2.40 (m, 2H), 2.22 (s, 3H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 171.2, 157.3, 142.7, 137.0, 135.6, 129.8, 129.2, 127.6, 126.8, 125.8, 118.9, 115.3, 112.9, 40.2, 36.9, 31.2, 18.6. ESI-HRMS: m/z calcd for C₁₇H₂₀NO₂ [M + H]⁺: 270.1489; found: 270.1487. HPLC analysis: retention time = 8.350 min; peak area, 99.13 %.

N-(3-hydroxyphenethyl)cinnamamide (XI). White solid, yield 63 %, m.p. 184.1–184.8 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.28 (s, 1H), 8.18 (t, J = 5.7 Hz, 1H), 7.57 – 7.53 (m, 2H), 7.44 – 7.34 (m, 4H), 7.08 (t, J = 7.7 Hz, 1H), 6.66 – 6.59 (m, 4H), 3.41 – 3.36 (m, 2H), 2.69 (t, J = 7.4 Hz, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 164.9, 157.4, 140.8, 138.5, 134.9, 129.4, 129.3, 128.9, 127.5, 122.3, 119.2, 115.5, 113.1, 40.4, 35.21. ESI-HRMS: m/z calcd for C₁₇H₁₈NO₂ [M + H]⁺: 268.1332; found: 268.1330. HPLC analysis: retention time = 7.325 min; peak area, 99.94 %.

(E)-1-(4-(3-hydroxyphenyl)piperazin-1-yl)-3-phenylprop-2-en-1-one (XII). Tawny solid, yield 39 %; m.p. 72.1–73.0 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.17 (s, 1H), 7.74 (d, J = 7.3 Hz, 2H), 7.52 (d, J = 15.4 Hz, 1H), 7.40 (dd, J = 10.9, 7.0 Hz, 3H), 7.31 (d, J = 15.4 Hz, 1H), 7.01 (t, J = 5.0 Hz, 1H), 6.42 (d, J = 5.9 Hz, 1H), 6.35 (d, J = 2.4 Hz, 1H), 6.25 (d, J = 7.9 Hz, 1H), 3.84 (s, 2H), 3.71 (s, 2H), 3.12 (s, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 164.4, 158.1, 152.2, 141.6, 135.1, 129.6, 129.6, 128.8, 128.1, 118.1, 107.0, 106.6, 102.9, 49.0, 48.4, 44.9, 41.5. ESI-HRMS: m/z calcd for C₁₉H₂₁N₂O₂ [M + H]⁺: 309.1598; found: 309.1598. HPLC analysis: retention time = 9.980 min; peak area, 99.86 %.

N-(4-hydroxybenzyl)-2-methyl-1H-benzo[d]imidazole-6-carboxamide (XIII). White solid, yield 60 %, m.p. 260.2–261.1 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 12.40 (s, 1H), 9.25 (s, 1H), 8.85 (t, J = 5.9 Hz, 1H), 8.02 (s, 1H), 7.70 (dd, J = 8.4, 1.7 Hz, 1H), 7.47 (s, 1H), 7.20 – 7.08 (m, 2H), 6.76 – 6.67 (m, 2H), 4.37 (d, J = 5.9 Hz, 2H), 2.50 (s, 3H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 156.1, 143.1, 136.5, 134.0, 130.2, 128.6, 121.2, 120.4, 117.2, 115.0, 110.2, 110.1, 42.2, 14.7. ESI-HRMS: m/z calcd for C₁₆H₁₅N₃O₂ [M + H]⁺: 282.1237; found: 282.1237. HPLC analysis: retention time = 5.663 min; peak area, 99.49 %.

(4-benzyl-1,4-diazepan-1-yl)(2-methyl-1H-benzo[d]imidazol-5-yl)methanone (XIV). White solid, yield 25 %, m.p. 65.9–66.7 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 12.41 – 12.29 (m, 1H), 7.59 – 7.06 (m, 8H), 3.61 (d, J = 29.2 Hz, 4H), 3.42 (s, 2H), 2.76 – 2.53 (m, 4H), 2.49 (s, 3H), 1.75 (d, J = 76.7 Hz, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 171.5, 153.3, 139.6, 130.5, 130.5, 130.4, 129.0, 128.9, 128.7, 127.3, 120.7, 120.5, 61.7, 61.3, 55.9, 55.3, 54.7, 53.8, 49.9, 48.8, 45.9, 45.2, 28.8, 27.1, 15.1. ESI-HRMS: m/z calcd for C₂₁H₂₅N₄O [M + H]⁺: 349.2023; found: 349.2024. HPLC analysis: retention time = 7.689 min; peak area, 99.37 %.

(3-(benzyloxy)phenyl)(4-(4-hydroxyphenyl)piperazin-1-yl)methanone (XV). White solid, yield 92 %, m.p. 215.7–216.5 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 8.89 (s, 1H), 7.48 – 7.43 (m, 2H), 7.42 – 7.36 (m, 3H), 7.36 – 7.31 (m, 1H), 7.10 (ddd, J = 8.4, 2.6, 1.0 Hz, 1H), 7.02 (dd, J = 2.7, 1.4 Hz, 1H), 6.97 (dt, J = 7.5, 1.2 Hz, 1H), 6.83 – 6.78 (m, 2H), 6.69 – 6.64 (m, 2H), 5.15 (s, 2H), 3.72 (s, 2H), 3.39 (s, 2H), 3.00 (s, 2H), 2.91 – 2.83 (m, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 168.6, 158.1, 151.4, 143.9, 137.3, 136.9, 129.7, 128.5, 127.9, 127.7, 119.2, 118.5, 116.2, 115.5, 113.0, 69.3, 54.9, 50.4, 47.2, 41.7. ESI-HRMS: m/z calcd for C₂₄H₂₅N₂O₃ [M + H]⁺: 389.1860; found: 389.1858. HPLC analysis: retention time = 6.692 min; peak area, 98.34 %.

4-(piperazin-1-yl)phenyl quinoline-3-carboxylate (XVI). White solid, yield 64 %, m.p. 176.4–177.2 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.43 (d, J = 2.2 Hz, 1H), 9.20 (d, J = 2.2 Hz, 1H), 8.27 (d, J = 8.1 Hz, 1H), 8.15 (d, J = 8.5 Hz, 1H), 7.96 (t, J = 7.7 Hz, 1H), 7.75 (t, J = 7.5 Hz, 1H), 7.19 (d, J = 8.7 Hz, 2H), 6.99 (d, J = 9.0 Hz, 2H), 3.05 (t, J = 5.0 Hz, 4H), 2.84 (t, J = 4.9 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆)) δ 164.1, 149.7, 149.4, 148.7, 139.5, 132.7, 129.9, 128.9, 127.9, 126.6, 122.4, 118.5, 116.8, 115.7, 47.9, 46.8, 43.4, 33.4. ESI-HRMS: m/z calcd for C₂₀H₂₀N₃O₂ [M + H]⁺: 334.1550; found: 334.1553. HPLC analysis: retention time = 6.707 min; peak area, 99.84 %.

4-(piperazin-1-yl)phenyl 2-hydroxyquinoline-4-carboxylate (XVII). Light yellow solid, yield 31 %, m.p. 249.8–250.6 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 8.14 (dd, J = 8.2, 1.4 Hz, 1H), 7.60 (ddd, J = 8.5, 7.2, 1.4 Hz, 1H), 7.42 (dd, J = 8.4, 1.2 Hz, 1H), 7.28 (ddd, J = 8.3, 7.1, 1.3 Hz, 1H), 7.25 – 7.21 (m, 2H), 7.19 (s, 1H), 7.02 – 6.98 (m, 2H), 3.05 (t, J = 5.0 Hz, 4H), 2.84 (t, J = 5.0 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 164.1, 160.8, 150.0, 142.4, 139.5, 139.3, 131.2, 125.9, 124.8, 122.5, 122.0, 117.8, 116.0, 115.4, 49.5, 45.5. ESI-HRMS: m/z calcd for C₂₀H₂₀N₃O₃ [M + H]⁺: 350.1499; found: 350.1494. HPLC analysis: retention time = 6.843 min; peak area, 99.85 %.

1-(4-(2-fluorobenzyl)piperazin-1-yl)-3-(3-hydroxyphenyl)propan-1-one (V-1). White solid, yield 40 %, m.p. 90.9–91.6 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.21 (s, 1H), 7.40 (td, J = 7.7, 1.9 Hz, 1H), 7.35 – 7.29 (m, 1H), 7.20 – 7.13 (m, 2H), 7.03 (t, J = 7.7 Hz, 1H), 6.64 – 6.60 (m, 2H), 6.56 (ddd, J = 8.1, 2.5, 1.0 Hz, 1H), 3.52 (d, J = 1.3 Hz, 2H), 3.44 (t, J = 5.1 Hz, 2H), 3.38 (t, J = 5.0 Hz, 2H), 2.69 (t, J = 7.7 Hz, 2H), 2.54 (dd, J = 8.7, 6.8 Hz, 2H), 2.30 (dt, J = 6.8, 4.7 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 169.9, 161.8 (d, ¹J = 243.75 Hz), 157.3, 142.8, 131.6 (d, ⁴J = 3.75 Hz), 129.2 (d, ³J = 7.5 Hz), 129.1, 124.3 (d, ²J = 13.8 Hz), 124.2 (d, ⁴J = 3.8 Hz), 119.0, 115.4, 115.3 (d, ²J = 21.3 Hz), 112.9, 54.4, 52.5, 52.2, 44.9, 41.1, 33.9, 30.9. ESI-HRMS: m/z calcd for C₂₀H₂₄FN₂O₂ [M + H]⁺: 343.1816; found: 343.1821. HPLC analysis: retention time = 6.155 min; peak area, 98.85 %.

1-(4-(3-fluorobenzyl)piperazin-1-yl)-3-(3-hydroxyphenyl)propan-1-one (V-2). White solid, yield 23 %, m.p. 244.7–245.5 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.21 (s, 1H), 7.39 – 7.33 (m, 1H), 7.16 – 7.11 (m, 2H), 7.10 – 7.02 (m, 2H), 6.64 – 6.60 (m, 2H), 6.56 (dd, J = 9.2, 2.4 Hz, 1H), 3.48 (s, 2H), 3.44 (d, J = 5.0 Hz, 2H), 3.40 (t, J = 5.0 Hz, 2H), 2.70 (t, J = 7.7 Hz, 2H), 2.58 – 2.52 (m, 2H), 2.32 – 2.25 (m, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 169.8, 163.2 (d, ¹J = 241.3 Hz), 157.3, 142.7, 141.1 (d, ³J = 7.5 Hz), 130.1 (d, ³J = 8.8 Hz), 129.1, 124.7 (d, ⁴J = 2.5 Hz), 119.0, 115.3, 115.3 (d, ²J = 21.3 Hz), 113.8 (d, ²J = 20.0 Hz), 112.8, 61.1, 52.7, 52.3, 44.9, 41.0, 33.9, 30.8. ESI-HRMS: m/z calcd for C₂₀H₂₄FN₂O₂ [M + H]⁺: 343.1816; found: 343.1821. HPLC retention time = 6.158 min; peak area, 99.01 %.

1-(4-(4-fluorobenzyl)piperazin-1-yl)-3-(2-hydroxyphenyl)propan-1-one (V-3). White solid, yield 35 %, m.p. 123.8–124.6 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.35 (s, 1H), 7.32 (dd, J = 8.5, 5.8 Hz, 2H), 7.14 (t, J = 8.9 Hz, 2H), 7.05 (dd, J = 7.5, 1.7 Hz, 1H), 6.99 (td, J = 7.7, 1.8 Hz, 1H), 6.76 (dd, J = 8.0, 1.2 Hz, 1H), 6.69 (td, J = 7.4, 1.2 Hz, 1H), 3.44 (d, J = 4.1 Hz, 4H), 3.40 (t, J = 5.1 Hz, 2H), 2.75 – 2.68 (m, 2H), 2.53 (d, J = 8.3 Hz, 2H), 2.26 (q, J = 5.7 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.3, 162.3 (d, ¹J = 240.0 Hz), 155.2, 134.0 (d, ⁴J = 2.5 Hz), 130.7 (d, ³J = 7.5 Hz), 130.0, 127.3 (d, ²J = 33.8 Hz), 118.9, 115.0, 115.0, 114.8, 60.9, 52.6, 52.2, 44.9, 41.0, 32.7, 25.9. ESI-HRMS: m/z calcd for C₂₀H₂₄FN₂O₂ [M + H]⁺: 343.1816; found: 343.1820. HPLC retention time = 6.163 min; peak area, 98.94 %.

1-(4-(4-fluorobenzyl)piperazin-1-yl)-3-(4-hydroxyphenyl)propan-1-one (V-4). White oil, yield 60 %.¹H NMR (500 MHz, DMSO‑d₆) δ 9.12 (s, 1H), 7.35 – 7.29 (m, 2H), 7.17 – 7.11 (m, 2H), 7.02 – 6.98 (m, 2H), 6.67 – 6.62 (m, 2H), 3.43 (s, 4H), 3.37 (t, J = 5.0 Hz, 2H), 2.67 (dd, J = 8.9, 6.5 Hz, 2H), 2.54 – 2.50 (m, 2H), 2.26 (dd, J = 9.9, 5.0 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.0, 162.3 (d, ¹J = 241.3 Hz), 155.5, 134.1 (d, ⁴J = 3.8 Hz), 131.3, 130.7 (d, ³J = 7.5 Hz), 129.25, 115.00, 114.83, 60.92, 52.63, 52.22, 44.90, 41.01, 34.39, 30.07. ESI-HRMS: m/z calcd for C₂₀H₂₄FN₂O₂ [M + H]⁺: 343.1816; found: 343.1822. HPLC analysis: retention time = 7.550 min; peak area, 98.22 %.

(E)-1-(4-(4-fluorobenzyl)piperazin-1-yl)-3-(3-hydroxyphenyl)prop-2-en-1-one (V-5). White solid, yield 75 %, m.p. 219.6–220.4 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.54 (s, 1H), 7.38 (d, J = 15.4 Hz, 1H), 7.36 – 7.32 (m, 2H), 7.18 (t, J = 7.8 Hz, 1H), 7.16 – 7.13 (m, 2H), 7.13 – 7.10 (m, 2H), 7.05 (t, J = 2.0 Hz, 1H), 6.79 (dd, J = 8.0, 1.5 Hz, 1H), 3.67 (s, 2H), 3.56 (s, 2H), 3.48 (s, 2H), 2.36 (dt, J = 18.1, 4.8 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 164.4, 162.3 (d, ¹J = 240.0 Hz), 157.6, 141.7, 136.4, 134.0 (d, ⁴J = 3.8 Hz), 130.8 (d, ³J = 7.5 Hz), 129.7, 118.9, 118.0, 116.6, 115.0 (d, ²J = 21.3 Hz), 114.5, 60.9, 53.1, 52.2, 45.1, 41.7. ESI-HRMS: m/z calcd for C₂₀H₂₂FN₂O₂ [M + H]⁺: 341.1660; found: 341.1658. HPLC analysis: retention time = 6.163 min; peak area, 98.60 %.

1-(4-(4-fluorobenzyl)-1,4-diazepan-1-yl)-3-(3-hydroxyphenyl)propan-1-one (V-6). Yellow oil, yield 35 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.22 (s, 1H), 7.32 (dd, J = 8.6, 5.9 Hz, 2H), 7.12 (td, J = 8.9, 2.8 Hz, 2H), 7.04 (t, J = 7.7 Hz, 1H), 6.63 (ddd, J = 8.5, 5.1, 1.8 Hz, 2H), 6.56 (dq, J = 8.1, 1.9 Hz, 1H), 3.55 (d, J = 2.6 Hz, 2H), 3.51 – 3.43 (m, 4H), 2.72 (td, J = 7.6, 2.0 Hz, 2H), 2.58 – 2.51 (m, 4H), 2.49 – 2.44 (m, 2H), 1.79 – 1.64 (m, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.9, 162.2 (d, ¹J = 241.3 Hz), 157.3, 142.9 (d, ⁴J = 3.8 Hz), 135.3, 130.3 (t, ³J = 8.8 Hz), 129.1, 119.0, 115.4, 114.8 (dd, ²J = 20.0 Hz), 112.8, 60.2, 60.0, 55.4, 54.9, 54.0, 53.5, 47.0, 46.0, 44.4, 44.0, 34.2, 34.1, 30.9, 30.8, 27.8, 26.9. ESI-HRMS: m/z calcd for C₂₁H₂₆FN₂O₂ [M + H]⁺: 357.1973; found: 357.1978. HPLC analysis: retention time = 6.166 min; peak area, 98.67 %.

1-(5-(4-fluorobenzyl)hexahydropyrrolo[3,4-c]pyrrol-2(1H)-yl)-3-(3-hydroxyphenyl)propan-1-one (V-7). White solid, yield 81 %, m.p. 140.5–141.0 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.23 (s, 1H), 7.34 – 7.28 (m, 2H), 7.15 – 7.09 (m, 2H), 7.04 (t, J = 7.7 Hz, 1H), 6.65 – 6.61 (m, 2H), 6.59 – 6.55 (m, 1H), 3.58 – 3.48 (m, 4H), 3.21 (dt, J = 12.1, 4.2 Hz, 2H), 2.83 – 2.65 (m, 5H), 2.49 – 2.42 (m, 3H), 2.33 (ddd, J = 9.3, 3.8, 2.1 Hz, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 169.4, 162.1 (d, ¹J = 240.0 Hz), 157.3, 142.9, 135.3 (d, ⁴J = 2.5 Hz), 130.1 (d, ³J = 8.8 Hz), 129.2, 118.9, 115.3, 114.9 (d, ²J = 21.3 Hz), 112.8, 59.7, 59.4, 57.8, 51.8, 50.9, 41.8, 40.0, 35.6, 30.5. ESI-HRMS: m/z calcd for C₂₂H₂₆FN₂O₂ [M + H]⁺: 369.1973; found: 369.1977. HPLC analysis: retention time = 6.194 min; peak area, 98.16 %.

N-(2-((4-fluorobenzyl)amino)ethyl)-3-(3-hydroxyphenyl)propanamide (V-8). Orange oil, yield 19 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.23 (s, 1H), 7.76 (t, J = 5.7 Hz, 1H), 7.38 – 7.30 (m, 2H), 7.16 – 7.08 (m, 2H), 7.05 – 7.00 (m, 1H), 6.60 – 6.57 (m, 2H), 6.55 (ddd, J = 8.0, 2.5, 1.1 Hz, 1H), 3.64 (s, 2H), 3.12 (q, J = 6.3 Hz, 2H), 2.69 (dd, J = 8.7, 6.8 Hz, 2H), 2.50 – 2.47 (m, 2H), 2.31 (dd, J = 8.7, 6.9 Hz, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 171.4, 162.0 (d, ¹J = 240.0 Hz), 157.3, 142.8, 136.8 (d, ⁴J = 2.5 Hz), 129.8 (d, ³J = 7.5 Hz), 129.1, 118.8, 115.1, 114.8 (d, ²J = 20.0 Hz), 112.8, 51.8, 48.0, 38.6, 37.0, 31.1. ESI-HRMS: m/z calcd for C₁₈H₂₂FN₂O₂ [M + H]⁺: 317.1660; found: 317.1663. HPLC analysis: retention time = 6.127 min; peak area, 98.53 %.

N-(3-((4-fluorobenzyl)amino)propyl)-3-(3-hydroxyphenyl)propanamide (V-9). Red oil, yield 18 %.¹H NMR (500 MHz, DMSO‑d₆) δ 7.85 (t, J = 5.7 Hz, 1H), 7.42 – 7.35 (m, 2H), 7.16 – 7.11 (m, 2H), 7.04 (t, J = 7.9 Hz, 1H), 6.61 – 6.52 (m, 3H), 3.70 (s, 2H), 3.08 (q, J = 6.5 Hz, 2H), 2.69 (dd, J = 8.7, 6.8 Hz, 2H), 2.50 – 2.46 (m, 2H), 2.30 (dd, J = 8.8, 6.8 Hz, 2H), 1.55 (p, J = 6.9 Hz, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 171.4, 162.2 (d, ¹J = 240.0 Hz), 157.3, 142.8, 135.8, 130.3 (d, ³J = 7.5 Hz), 129.2, 118.8, 115.2, 115.0 (d, ²J = 21.3 Hz), 112.9, 51.7, 45.8, 37.1, 36.5, 31.2, 28.9. ESI-HRMS: m/z calcd for C₁₉H₂₄FN₂O₂ [M + H]⁺: 331.1816; found: 331.1812. HPLC analysis: retention time = 6.120 min; peak area, 98.90 %.

1-(4-(3-chlorobenzyl)piperazin-1-yl)-3-(3-hydroxyphenyl)propan-1-one (V-10). Colorless oil, yield 35 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.22 (d, J = 1.6 Hz, 1H), 7.35 (d, J = 7.5 Hz, 2H), 7.31 (d, J = 8.1 Hz, 1H), 7.26 (d, J = 7.2 Hz, 1H), 7.04 (t, J = 7.7 Hz, 1H), 6.65 – 6.60 (m, 2H), 6.59 – 6.55 (m, 1H), 3.44 (d, J = 5.0 Hz, 2H), 3.39 (t, J = 4.9 Hz, 2H), 3.35 (d, J = 2.8 Hz, 2H), 2.70 (t, J = 7.7 Hz, 2H), 2.57 – 2.52 (m, 2H), 2.28 (h, J = 4.7 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 169.9, 157.3, 142.8, 140.7, 133.0, 130.1, 129.1, 128.5, 127.5, 127.0, 119.0, 115.3, 112.8, 61.0, 52.7, 52.3, 44.9, 41.1, 33.9, 30.8. ESI-HRMS: m/z calcd for C₂₀H₂₄ClN₂O₂ [M + H]⁺: 359.1521; found: 359.1522. HPLC analysis: retention time = 8.125 min; peak area, 98.42 %.

3-(3-hydroxyphenyl)-1-(4-(4-methylbenzyl)piperazin-1-yl)propan-1-one (V-11). Gray solid, yield 73 %, m.p. 84.5–85.3 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.22 (s, 1H), 7.18 – 7.15 (m, 2H), 7.12 (d, J = 7.9 Hz, 2H), 7.03 (t, J = 7.7 Hz, 1H), 6.64 – 6.59 (m, 2H), 6.57 (d, J = 7.9 Hz, 1H), 3.43 (t, J = 5.0 Hz, 2H), 3.37 (t, J = 5.0 Hz, 2H), 3.33 (d, J = 2.2 Hz, 4H), 2.70 (t, J = 7.7 Hz, 2H), 2.56 – 2.51 (m, 2H), 2.27 (s, 3H), 2.25 – 2.23 (m, 2H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 169.9, 157.3, 142.8, 136.1, 134.8, 129.2, 128.9, 128.8, 119.0, 115.4, 112.8, 61.6, 52.7, 52.3, 44.9, 41.1, 33.9, 30.8, 20.7. ESI-HRMS: m/z calcd for C₂₁H₂₇N₂O₂ [M + H]⁺: 339.2067; found: 339.2067. HPLC analysis: retention time = 8.212 min; peak area, 98.60 %.

1-(4-(4-ethylbenzyl)piperazin-1-yl)-3-(3-hydroxyphenyl)propan-1-one (V-12). Colorless oil, yield 28 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.21 (s, 1H), 7.19 (d, J = 8.1 Hz, 2H), 7.15 (d, J = 8.1 Hz, 2H), 7.03 (t, J = 7.7 Hz, 1H), 6.64 – 6.59 (m, 2H), 6.56 (ddd, J = 8.0, 2.5, 1.0 Hz, 1H), 3.46 – 3.41 (m, 2H), 3.41 (s, 2H), 3.39 – 3.35 (m, 2H), 2.70 (t, J = 7.7 Hz, 2H), 2.60 – 2.52 (m, 4H), 2.28 – 2.21 (m, 4H), 1.16 (t, J = 7.6 Hz, 3H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.3, 157.4, 142.9, 142.7, 135.1, 129.4, 129.2, 127.8, 119.3, 115.5, 113.1, 61.8, 52.8, 52.4, 45.1, 41.3, 34.1, 31.0, 28.0, 15.8. ESI-HRMS: m/z calcd for C₂₂H₂₉N₂O₂ [M + H]⁺: 353.2224; found: 353.2224. HPLC analysis: retention time = 7.311 min; peak area, 98.15 %.

3-(3-hydroxyphenyl)-1-(4-(4-methoxybenzyl)piperazin-1-yl)propan-1-one (V-13). Brown oil, yield 72 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.23 (s, 1H), 7.19 (d, J = 8.6 Hz, 2H), 7.04 (t, J = 7.7 Hz, 1H), 6.87 (d, J = 8.6 Hz, 2H), 6.64 – 6.60 (m, 2H), 6.57 (dd, J = 8.1, 1.1 Hz, 1H), 3.73 (s, 3H), 3.42 (d, J = 5.3 Hz, 2H), 3.38 (s, 4H), 2.70 (t, J = 7.7 Hz, 2H), 2.56 – 2.51 (m, 2H), 2.25 (d, J = 6.4 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 169.8, 158.3, 157.3, 142.8, 130.1, 129.6, 129.1, 119.0, 115.3, 113.6, 112.8, 61.3, 55.0, 52.6, 52.2, 44.9, 41.1, 33.9, 30.8. ESI-HRMS: m/z calcd for C₂₁H₂₇N₂O₃ [M + H]⁺: 355.2016; found: 355.2013. HPLC analysis: retention time = 7.598 min; peak area, 98.99 %.

4-((4-(3-(3-hydroxyphenyl)propanoyl)piperazin-1-yl)methyl)benzonitrile (V-14). White solid, yield 34 %, m.p. 150.8–151.4 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.22 (s, 1H), 7.82 – 7.77 (m, 2H), 7.51 (d, J = 8.0 Hz, 2H), 7.03 (t, J = 7.7 Hz, 1H), 6.65 – 6.53 (m, 3H), 3.55 (s, 2H), 3.45 (t, J = 5.0 Hz, 2H), 3.40 (q, J = 4.4 Hz, 2H), 2.70 (t, J = 7.7 Hz, 2H), 2.55 (dd, J = 8.6, 6.8 Hz, 2H), 2.28 (q, J = 5.6 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 169.9, 157.3, 144.1, 142.7, 132.2, 129.6, 129.1, 119.0, 118.9, 115.4, 112.8, 109.8, 61.1, 52.7, 52.3, 44.9, 41.0, 33.9, 30.8. ESI-HRMS: m/z calcd for C₂₁H₂₄N₃O₂ [M + H]⁺: 350.1863; found: 350.1866. HPLC analysis: retention time = 5.840 min; peak area, 98.16 %.

(E)-1-(4-(3-(4-fluorophenyl)allyl)piperazin-1-yl)-3-(3-hydroxyphenyl)propan-1-one (V-15). White solid, yield 27 %, m.p. 148.9–149.6 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.22 (s, 1H), 7.53 – 7.45 (m, 2H), 7.18 – 7.10 (m, 2H), 7.06 – 7.01 (m, 1H), 6.65 – 6.60 (m, 2H), 6.58 – 6.49 (m, 2H), 6.29 – 6.20 (m, 1H), 3.49 – 3.43 (m, 2H), 3.42 – 3.38 (m, 2H), 3.08 (d, J = 6.3 Hz, 2H), 2.71 (t, J = 7.7 Hz, 2H), 2.59 – 2.53 (m, 2H), 2.32 (q, J = 5.2 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 169.9, 162.5 (d, ¹J = 242.5 Hz), 157.3, 142.8, 133.2 (d, ⁴J = 2.5 Hz), 131.1, 129.2, 128.1 (d, ³J = 7.5 Hz), 126.7, 119.0, 115.5 (d, ²J = 21.3 Hz), 115.4, 112.8, 59.9, 52.8, 52.3, 44.9, 41.1, 33.9, 30.8. HRMS: m/z calcd for C₂₂H₂₆FN₂O₂ [M + H]⁺: 369.1973; found: 369.1973. HPLC analysis: retention time = 6.176 min; peak area, 99.54 %.

3-(3-hydroxyphenyl)-1-(4-phenethylpiperazin-1-yl)propan-1-one (V-16). Organe oil, yield 65 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.22 (s, 1H), 7.29 – 7.25 (m, 2H), 7.21 (d, J = 6.6 Hz, 2H), 7.19 – 7.15 (m, 1H), 7.04 (t, J = 7.7 Hz, 1H), 6.66 – 6.60 (m, 2H), 6.57 (dd, J = 7.6, 2.2 Hz, 1H), 3.47 – 3.42 (m, 2H), 3.38 (t, J = 5.0 Hz, 2H), 2.71 (td, J = 7.8, 4.0 Hz, 4H), 2.56 (t, J = 7.7 Hz, 2H), 2.51 (d, J = 4.7 Hz, 1H), 2.50 – 2.48 (m, 1H), 2.39 – 2.32 (m, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 169.9, 157.3, 142.8, 140.4, 129.2, 128.7, 128.3, 125.9, 119.1, 115.4, 112.9, 59.6, 52.9, 52.4, 45.0, 41.1, 33.9, 32.6, 30.9. HRMS: m/z calcd for C₂₁H₂₇N₂O₂ [M + H]⁺: 339.2067; found: 339.2067. HPLC analysis: retention time = 7.868 min; peak area, 99.06 %.

1-(4-(furan-2-ylmethyl)piperazin-1-yl)-3-(3-hydroxyphenyl)propan-1-one (V-17). Yellow oil, yield 24 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.21 (s, 1H), 7.58 (dd, J = 1.8, 0.9 Hz, 1H), 7.03 (t, J = 7.7 Hz, 1H), 6.61 (dd, J = 8.0, 5.2 Hz, 2H), 6.56 (ddd, J = 8.0, 2.5, 1.0 Hz, 1H), 6.39 (dd, J = 3.1, 1.8 Hz, 1H), 6.27 (d, J = 3.0 Hz, 1H), 3.49 (s, 2H), 3.43 (t, J = 5.1 Hz, 2H), 3.39 – 3.36 (m, 2H), 2.71 – 2.67 (m, 2H), 2.56 – 2.51 (m, 2H), 2.29 (q, J = 4.6 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.3, 157.4, 151.5, 142.9, 142.6, 129.4, 119.3, 115.5, 113.1, 110.6, 109.1, 53.9, 52.4, 52.0, 45.0, 41.2, 34.0, 31.0. HRMS: m/z calcd for C₁₈H₂₃N₂O₃ [M + H]⁺: 315.1703; found: 315.1701. HPLC analysis: retention time = 6.782 min; peak area, 98.90 %.

3-(3-hydroxyphenyl)-1-(4-(thiophen-2-ylmethyl)piperazin-1-yl)propan-1-one (V-18). Brown oil, yield 24 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.22 (s, 1H), 7.43 (dd, J = 4.7, 1.6 Hz, 1H), 7.03 (t, J = 7.7 Hz, 1H), 6.98 – 6.94 (m, 2H), 6.64 – 6.59 (m, 2H), 6.56 (ddd, J = 8.1, 2.5, 1.0 Hz, 1H), 3.67 (s, 2H), 3.46 – 3.37 (m, 4H), 2.70 (t, J = 7.6 Hz, 2H), 2.54 (dd, J = 8.6, 6.8 Hz, 2H), 2.31 (dt, J = 8.0, 5.0 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 169.9, 157.3, 142.8, 141.3, 129.1, 126.6, 126.3, 125.6, 119.0, 115.3, 112.8, 56.1, 52.4, 52.0, 44.8, 41.0, 33.9, 30.8. HRMS: m/z calcd for C₁₈H₂₃N₂O₂S [M + H]⁺: 331.1475; found: 331.1475. HPLC analysis: retention time = 6.136 min; peak area, 98.64 %.

3-(3-hydroxyphenyl)-1-(4-(pyridin-2-ylmethyl)piperazin-1-yl)propan-1-one (V-19). Yellow oil, yield 27 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.22 (s, 1H), 8.48 (ddd, J = 4.9, 1.8, 0.9 Hz, 1H), 7.76 (td, J = 7.7, 1.8 Hz, 1H), 7.43 (d, J = 7.8 Hz, 1H), 7.30 – 7.23 (m, 1H), 7.04 (t, J = 7.7 Hz, 1H), 6.65 – 6.59 (m, 2H), 6.58 – 6.54 (m, 1H), 3.59 (s, 2H), 3.46 (t, J = 5.0 Hz, 2H), 3.40 (t, J = 5.0 Hz, 2H), 2.70 (t, J = 7.7 Hz, 2H), 2.55 (dd, J = 8.6, 6.8 Hz, 2H), 2.34 (q, J = 4.5 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.1, 158.1, 157.4, 148.9, 142.8, 136.7, 129.3, 123.0, 122.4, 119.1, 115.4, 113.0, 63.6, 53.0, 52.6, 45.0, 41.2, 34.0, 30.9. HRMS: m/z calcd for C₁₉H₂₄N₃O₂ [M + H]⁺: 326.1863; found: 326.1867. HPLC analysis: retention time = 5.963 min; peak area, 98.92 %.

3-(3,5-dihydroxyphenyl)-1-(4-(4-fluorobenzyl)piperazin-1-yl)propan-1-one (V-20). Orange solid, yield 35 %, m.p. 90.4–91.3 °C. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.04 (s, 2H), 7.35 – 7.30 (m, 2H), 7.17 – 7.11 (m, 2H), 6.04 (dd, J = 12.1, 2.2 Hz, 3H), 3.44 (s, 4H), 3.38 (t, J = 5.0 Hz, 2H), 2.59 (dd, J = 8.8, 6.5 Hz, 2H), 2.50 – 2.47 (m, 2H), 2.30 – 2.23 (m, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.0, 162.3 (d, ¹J = 241.3 Hz), 158.3, 143.3, 134.1 (d, ⁴J = 3.8 Hz), 130.8 (d, ³J = 7.5 Hz), 115.0 (d, ²J = 20.0 Hz), 106.5, 100.3, 61.0, 52.6, 52.2, 44.9, 41.1, 33.9, 31.0. HRMS: m/z calcd for C₂₀H₂₄FN₂O₃ [M + H]⁺: 359.1765; found: 359.1764. HPLC analysis: retention time = 6.170 min; peak area, 98.63 %.

3-(2,5-dihydroxyphenyl)-1-(4-(4-fluorobenzyl)piperazin-1-yl)propan-1-one (V-21). Orange oil, yield 26 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 8.62 (s, 1H), 8.54 (s, 1H), 7.33 (dd, J = 8.3, 5.6 Hz, 2H), 7.14 (t, J = 8.7 Hz, 2H), 6.55 (d, J = 8.5 Hz, 1H), 6.48 (d, J = 2.9 Hz, 1H), 6.40 (dd, J = 8.5, 3.0 Hz, 1H), 3.48 – 3.36 (m, 6H), 2.63 (t, J = 7.6 Hz, 2H), 2.49 (d, J = 2.0 Hz, 2H), 2.27 (s, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.4, 162.3 (d, ¹J = 241.3 Hz), 149.8, 147.5, 134.0, 130.8 (d, ³J = 8.8 Hz), 128.1, 116.6, 115.6, 115.0 (d, ²J = 21.3 Hz), 113.2, 60.9, 52.6, 52.2, 44.9, 41.0, 32.8, 26.1. HRMS: m/z calcd for C₂₀H₂₄FN₂O₃ [M + H]⁺: 359.1765; found: 359.1765. HPLC analysis: retention time = 6.447 min; peak area, 98.08 %.

(E)-3-(2,4-dihydroxyphenyl)-1-(4-(4-fluorobenzyl)piperazin-1-yl)prop-2-en-1-one (V-22). Light green solid, yield 11 %, m.p. 108.2–108.6 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.88 (s, 1H), 9.68 (s, 1H), 7.68 (d, J = 15.4 Hz, 1H), 7.46 (d, J = 8.6 Hz, 1H), 7.35 (dd, J = 8.6, 5.7 Hz, 2H), 7.18 – 7.12 (m, 2H), 6.93 (d, J = 15.4 Hz, 1H), 6.34 (d, J = 2.4 Hz, 1H), 6.24 (dd, J = 8.5, 2.4 Hz, 1H), 3.58 (d, J = 29.1 Hz, 4H), 3.48 (s, 2H), 2.35 (s, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 165.4, 162.3 (d, ¹J = 241.3 Hz), 160.0, 157.7, 137.5, 134.1 (d, ⁴J = 3.8 Hz), 130.8 (d, ³J = 7.5 Hz), 129.5, 115.0 (d, ²J = 20.0 Hz), 113.6, 112.8, 107.4, 102.5, 60.9, 53.1, 52.3, 48.6, 41.7. HRMS: m/z calcd for C₂₀H₂₂FN₂O₃ [M + H]⁺: 357.1609; found: 357.1609. HPLC analysis: retention time = 5.750 min; peak area, 98.54 %.

3-(2,4-dihydroxyphenyl)-1-(4-(4-methylbenzyl)piperazin-1-yl)propan-1-one (V-23). Yellow solid, yield 23 %, m.p. 86.7–87.5 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.18 (s, 1H), 8.95 (s, 1H), 7.17 (d, J = 8.0 Hz, 2H), 7.12 (d, J = 7.9 Hz, 2H), 6.79 (d, J = 8.1 Hz, 1H), 6.24 (d, J = 2.4 Hz, 1H), 6.11 (dd, J = 8.1, 2.4 Hz, 1H), 3.44 – 3.41 (m, 2H), 3.40 (s, 2H), 3.38 (d, J = 5.3 Hz, 2H), 2.59 (dd, J = 9.1, 6.2 Hz, 2H), 2.44 (dd, J = 8.8, 6.6 Hz, 2H), 2.27 (s, 3H), 2.25 (q, J = 5.7 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.5, 156.6, 155.8, 136.0, 134.7, 130.3, 128.9, 128.8, 117.8, 106.0, 102.5, 61.6, 52.7, 52.3, 44.9, 41.0, 33.2, 25.4, 20.7. HRMS: m/z calcd for C₂₁H₂₇N₂O₃ [M + H]⁺: 355.2016; found: 355.2016. HPLC analysis: retention time = 4.377 min; peak area, 98.88 %.

(E)-3-(2,4-dihydroxyphenyl)-1-(4-(4-methylbenzyl)piperazin-1-yl)prop-2-en-1-one (V-24). Yellow solid, yield 45 %, m.p. 119.7–120.5 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.86 (s, 1H), 9.66 (s, 1H), 7.68 (dd, J = 15.4, 1.6 Hz, 1H), 7.46 (d, J = 8.6 Hz, 1H), 7.19 (d, J = 7.7 Hz, 2H), 7.13 (d, J = 7.8 Hz, 2H), 6.92 (d, J = 15.4 Hz, 1H), 6.33 (dd, J = 2.4, 1.4 Hz, 1H), 6.24 (dd, J = 8.5, 2.4 Hz, 1H), 3.57 (d, J = 27.9 Hz, 4H), 3.44 (s, 2H), 2.34 (s, 4H), 2.28 (s, 3H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 165.3, 160.0, 157.7, 137.4, 136.0, 134.7, 129.4, 128.9, 128.8, 113.6, 112.8, 107.4, 102.5, 61.6, 53.1, 52.4, 44.9, 41.6, 20.7. HRMS: m/z calcd for C₂₁H₂₅N₂O₃ [M + H]⁺: 353.1860; found: 353.1860. HPLC analysis: retention time = 8.334 min; peak area, 98.83 %.

3-(2,4-dihydroxyphenyl)-1-(4-(4-methoxybenzyl)piperazin-1-yl)propan-1-one (V-25). Light yellow solid, yield 36 %, m.p. 96.7–97.4 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.19 (s, 1H), 8.96 (s, 1H), 7.19 (d, J = 8.7 Hz, 2H), 6.87 (d, J = 8.7 Hz, 2H), 6.79 (d, J = 8.2 Hz, 1H), 6.24 (d, J = 2.4 Hz, 1H), 6.11 (dd, J = 8.1, 2.4 Hz, 1H), 3.73 (s, 3H), 3.42 (s, 2H), 3.39 – 3.36 (m, 4H), 2.59 (dd, J = 8.9, 6.5 Hz, 2H), 2.47 – 2.42 (m, 2H), 2.25 (q, J = 4.4 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.6, 158.3, 156.6, 155.8, 130.3, 130.2, 129.6, 117.8, 113.6, 106.0, 102.5, 61.3, 55.0, 52.6, 52.2, 44.9, 41.0, 33.2, 25.4. HRMS: m/z calcd for C₂₁H₂₇N₂O₄ [M + H]⁺: 371.1965; found: 371.1963. HPLC analysis: retention time = 7.191 min; peak area, 97.98 %.

(E)-3-(2,4-dihydroxyphenyl)-1-(4-(4-methoxybenzyl)piperazin-1-yl)prop-2-en-1-one (V-26). Yellow solid, yield 67 %, m.p. 125.9–126.5 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.87 (s, 1H), 9.68 (s, 1H), 7.68 (d, J = 15.4 Hz, 1H), 7.46 (d, J = 8.5 Hz, 1H), 7.23 – 7.18 (m, 2H), 6.92 (d, J = 15.4 Hz, 1H), 6.88 (d, J = 8.6 Hz, 2H), 6.33 (d, J = 2.4 Hz, 1H), 6.24 (dd, J = 8.5, 2.4 Hz, 1H), 3.73 (s, 3H), 3.56 (d, J = 22.7 Hz, 4H), 3.42 (s, 2H), 2.33 (s, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 165.4, 160.0, 158.4, 157.8, 137.5, 130.2, 129.6, 129.5, 113.6, 113.6, 112.8, 107.5, 102.5, 61.3, 55.0, 53.2, 52.3, 45.0, 41.6. HRMS: m/z calcd for C₂₁H₂₅N₂O₄ [M + H]⁺: 369.1809; found: 369.1807. HPLC analysis: retention time = 7.311 min; peak area, 98.64 %.

3-(2,4-dihydroxyphenyl)-1-(4-(4-ethoxybenzyl)piperazin-1-yl)propan-1-one (V-27). Yellow solid, yield 34 %, m.p. 124.5–125.2 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.18 (s, 1H), 8.95 (s, 1H), 7.18 (d, J = 8.5 Hz, 2H), 6.85 (d, J = 8.6 Hz, 2H), 6.79 (d, J = 8.2 Hz, 1H), 6.24 (d, J = 2.4 Hz, 1H), 6.11 (dd, J = 8.1, 2.4 Hz, 1H), 3.99 (q, J = 7.0 Hz, 2H), 3.42 (s, 2H), 3.37 (s, 4H), 2.61 – 2.56 (m, 2H), 2.44 (t, J = 7.7 Hz, 2H), 2.28 – 2.21 (m, 4H), 1.31 (t, J = 7.0 Hz, 3H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.6, 157.7, 156.6, 155.8, 130.3, 130.2, 129.5, 117.8, 114.1, 106.1, 102.5, 63.0, 61.3, 52.6, 52.3, 45.0, 41.0, 33.2, 25.4, 14.7. HRMS: m/z calcd for C₂₂H₂₉N₂O₄ [M + H]⁺: 385.2122; found: 385.2126. HPLC analysis: retention time = 7.299 min; peak area, 99.49 %.

(E)-3-(2,4-dihydroxyphenyl)-1-(4-(4-ethoxybenzyl)piperazin-1-yl)prop-2-en-1-one (V-28). Yellow solid, yield 67 %, m.p. 118.7–119.2 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.77 (s, 2H), 7.68 (d, J = 15.4 Hz, 1H), 7.46 (d, J = 8.5 Hz, 1H), 7.19 (d, J = 8.7 Hz, 2H), 6.92 (d, J = 15.4 Hz, 1H), 6.86 (d, J = 8.5 Hz, 2H), 6.33 (d, J = 2.4 Hz, 1H), 6.24 (dd, J = 8.5, 2.4 Hz, 1H), 3.99 (q, J = 7.0 Hz, 2H), 3.56 (d, J = 26.4 Hz, 4H), 3.41 (s, 2H), 2.33 (s, 4H), 1.31 (t, J = 7.0 Hz, 3H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 165.3, 160.0, 157.7, 157.6, 137.4, 130.2, 129.5, 129.4, 114.0, 113.6, 112.8, 107.4, 102.5, 62.9, 61.3, 53.1, 52.3, 45.0, 33.6, 14.7. HRMS: m/z calcd for C₂₂H₂₇N₂O₄ [M + H]⁺: 383.1965; found: 383.1969. HPLC analysis: retention time = 5.689 min; peak area, 98.05 %.

1-(4-(3-chlorobenzyl)piperazin-1-yl)-3-(2,4-dihydroxyphenyl)propan-1-one (V-29). Brown oil, yield 42 %. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.19 (s, 1H), 8.96 (s, 1H), 7.38 – 7.34 (m, 2H), 7.31 (dt, J = 8.2, 1.6 Hz, 1H), 7.27 (dt, J = 7.4, 1.5 Hz, 1H), 6.80 (d, J = 8.2 Hz, 1H), 6.25 (d, J = 2.4 Hz, 1H), 6.12 (dd, J = 8.1, 2.4 Hz, 1H), 3.47 (s, 2H), 3.42 (dt, J = 19.9, 5.3 Hz, 4H), 2.60 (dd, J = 9.1, 6.3 Hz, 2H), 2.45 (dd, J = 8.9, 6.6 Hz, 2H), 2.28 (q, J = 5.4 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.6, 156.6, 155.8, 140.7, 133.0, 130.3, 130.1, 128.5, 127.5, 127.0, 117.8, 106.0, 102.5, 61.0, 52.7, 52.3, 44.9, 41.0, 33.2, 25.4. HRMS: m/z calcd for C₂₀H₂₄ClN₂O₃ [M + H]⁺: 375.1470; found: 375.1471. HPLC analysis: retention time = 7.627 min; peak area, 98.96 %.

(E)-1-(4-(3-chlorobenzyl)piperazin-1-yl)-3-(2,4-dihydroxyphenyl)prop-2-en-1-one (V-30). Yellow solid, yield 26 %, m.p. 124.4–124.7 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.77 (d, J = 85.7 Hz, 2H), 7.68 (d, J = 15.4 Hz, 1H), 7.46 (d, J = 8.5 Hz, 1H), 7.39 – 7.34 (m, 2H), 7.34 – 7.27 (m, 2H), 6.93 (d, J = 15.4 Hz, 1H), 6.34 (s, 1H), 6.24 (d, J = 8.5 Hz, 1H), 3.59 (d, J = 27.9 Hz, 4H), 3.51 (s, 2H), 2.36 (s, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 165.8, 160.5, 158.2, 141.1, 138.0, 133.4, 130.6, 129.9, 129.0, 128.0, 127.5, 114.1, 113.2, 107.9, 102.9, 61.5, 53.6, 52.9, 49.1, 45.4. HRMS: m/z calcd for C₂₀H₂₂ClN₂O₃ [M + H]⁺: 373.1313; found: 373.1313. HPLC analysis: retention time = 8.744 min; peak area, 99.01 %.

1-(4-(4-chlorobenzyl)piperazin-1-yl)-3-(2,4-dihydroxyphenyl)propan-1-one (V-31). Yellow oil, yield 33 %.¹H NMR (500 MHz, DMSO‑d₆) δ 9.18 (d, J = 1.1 Hz, 1H), 8.96 (s, 1H), 7.38 (d, J = 8.5 Hz, 2H), 7.32 (d, J = 8.5 Hz, 2H), 6.79 (d, J = 8.2 Hz, 1H), 6.24 (d, J = 2.3 Hz, 1H), 6.11 (dd, J = 8.1, 2.4 Hz, 1H), 3.44 (d, J = 9.6 Hz, 4H), 3.39 (d, J = 5.0 Hz, 2H), 2.59 (dd, J = 9.0, 6.4 Hz, 2H), 2.45 (dd, J = 8.9, 6.6 Hz, 2H), 2.26 (q, J = 5.4 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.6, 156.6, 155.8, 137.0, 131.6, 130.7, 130.3, 128.2, 117.8, 106.0, 102.5, 60.9, 52.7, 52.3, 44.9, 41.0, 33.2, 25.4. HRMS: m/z calcd for C₂₀H₂₄ClN₂O₃ [M + H]⁺: 375.1470; found: 375.1474. HPLC analysis: retention time = 7.665 min; peak area, 98.47 %.

(E)-1-(4-(4-chlorobenzyl)piperazin-1-yl)-3-(2,4-dihydroxyphenyl)prop-2-en-1-one (V-32). Yellow solid, yield 49 %, m.p. 156.8–157.0 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.87 (s, 1H), 9.67 (s, 1H), 7.68 (d, J = 15.4 Hz, 1H), 7.46 (d, J = 8.6 Hz, 1H), 7.38 (d, J = 8.3 Hz, 2H), 7.34 (d, J = 8.3 Hz, 2H), 6.93 (d, J = 15.4 Hz, 1H), 6.33 (d, J = 2.3 Hz, 1H), 6.24 (dd, J = 8.5, 2.4 Hz, 1H), 3.58 (d, J = 28.6 Hz, 4H), 3.49 (s, 2H), 2.36 (d, J = 8.9 Hz, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 165.4, 160.0, 157.8, 137.5, 137.0, 131.6, 130.7, 129.5, 128.2, 113.6, 112.8, 107.5, 102.5, 60.9, 53.10, 52.4, 48.6, 45.0. HRMS: m/z calcd for C₂₀H₂₂ClN₂O₃ [M + H]⁺: 373.1313; found: 373.1312. HPLC analysis: retention time = 8.574 min; peak area, 98.81 %.

4-((4-(3-(2,4-dihydroxyphenyl)propanoyl)piperazin-1-yl)methyl)benzonitrile (V-33). Yellow solid, yield 46 %, m.p. 98.0–98.6 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.19 (s, 1H), 8.96 (s, 1H), 7.79 (d, J = 7.9 Hz, 2H), 7.51 (d, J = 7.9 Hz, 2H), 6.79 (d, J = 8.1 Hz, 1H), 6.24 (d, J = 2.4 Hz, 1H), 6.11 (dd, J = 8.1, 2.4 Hz, 1H), 3.56 (s, 2H), 3.41 (dd, J = 14.3, 9.4 Hz, 4H), 2.59 (dd, J = 9.1, 6.3 Hz, 2H), 2.45 (dd, J = 8.8, 6.6 Hz, 2H), 2.29 (s, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 170.6, 156.6, 155.8, 132.2, 130.3, 129.6, 118.9, 117.7, 109.8, 106.0, 102.5, 93.2, 61.1, 52.7, 52.3, 44.9, 41.0, 33.2, 25.4. HRMS: m/z calcd for C₂₁H₂₄N₃O₃ [M + H]⁺: 366.1812; found: 366.1813. HPLC analysis: retention time = 5.631 min; peak area, 98.20 %.

(E)-4-((4-(3-(2,4-dihydroxyphenyl)acryloyl)piperazin-1-yl)methyl)benzonitrile (V-34). Yellow solid, yield 53 %, m.p. 129.9–130.3 ℃. ¹H NMR (500 MHz, DMSO‑d₆) δ 9.88 (s, 1H), 9.68 (s, 1H), 7.80 (d, J = 8.2 Hz, 2H), 7.68 (d, J = 15.4 Hz, 1H), 7.54 (d, J = 8.2 Hz, 2H), 7.46 (d, J = 8.5 Hz, 1H), 6.93 (d, J = 15.5 Hz, 1H), 6.33 (t, J = 1.9 Hz, 1H), 6.24 (dd, J = 8.8, 2.3 Hz, 1H), 3.62 (s, 2H), 3.59 (s, 2H), 3.56 (s, 2H), 2.37 (s, 4H). ¹³C NMR (125 MHz, DMSO‑d₆) δ 165.4, 160.1, 157.8, 144.2, 137.6, 132.2, 129.6, 129.5, 118.9, 113.6, 112.8, 109.8, 107.5, 102.5, 61.2, 53.2, 52.5, 45.0, 41.6. HRMS: m/z calcd for C₂₁H₂₂N₃O₃ [M + H]⁺: 364.1656; found: 364.1656. HPLC analysis: retention time = 6.792 min; peak area, 98.77 %.

Mushroom tyrosinase inhibition assay

The TYR inhibitory activity assay was performed employing methods established in a previous study.[59] Mushroom TYR (EC 1.14.18.1) was procured from Sigma Chemical Co. Ltd. (T3824). Kojic acid, α-arbutin and β-arbutin were utilized as positive controls, while L-dopa and L-tyrosine served as substrates for the diphenolase and monophenolase catalytic cycles, respectively. All compounds and positive controls were dissolved in DMSO to form stock solutions at a concentration of 100 mM. These stock solutions were then diluted with PBS (pH = 6.8) to achieve the desired concentrations, consisting of eight to ten gradient levels, as a work solution (1.25 × test concentration). A total of 160 μL of compound work solution and 20 μL of TYR solution in PBS (100 U/mL for L-dopa and 300 U/mL for L-tyrosine) were added to a 96-well plate, and incubated at 25 °C for 10 min. Subsequently, 20 μL of substrate solution (0.85 mM L-dopa or 1 mM L-tyrosine) was added and incubated at 25 °C for an additional 20 min. Absorbance was measured immediately at 475 nm. The mixture of TYR and substrate was served as the control, while PBS was used as the blank. The inhibitory rate was calculated using the following formula, and the IC₅₀ values were determined using GraphPad Prism.

$$\mathit{Tyrosinase}inhibition\left(\%\right)=\frac{O{D}_{\mathit{Control}}-O{D}_{\mathit{Sample}}}{O{D}_{\mathit{Control}}-O{D}_{\mathit{Blank}}}\times 100\%$$

Molecular docking

The X-ray crystal structure of the TYR enzyme (Agaricus bisporus) was downloaded from the Protein Data Bank (PDB ID: 2Y9X). Molecular docking was performed using Discovery Studio® 2023. Initially, the “Prepare Proteins” module was employed to preprocess the obtained TYR crystal structure. This process involved repairing missing residues, optimizing the hydrogen bond network, assigning the correct protonation states, and resolving structural conflicts, ultimately leading to the generation of an energetically favorable protein model suitable for computational analysis. Subsequently, to delineate the docking site, we defined the center of the binding pocket based on the position of the ligand in the original crystal structure and the key catalytic residues. This resulted in the coordinates XYZ: −9.99476, −28.9201, −43.9379, which served as the core for determining the spatial extent of the active site. In the preprocessing phase, we removed the original ligand and water molecules from the crystal structure to prevent any interference with the subsequent docking process. The initial structures of ligand compounds V, VII, XIV, and V-24 were imported in MDL SD file format and processed with the “Prepare Ligands” module. This procedure facilitated the generation of their low-energy 3D conformations, as well as the assignment of their correct ionization states and chirality. High-precision docking of the compounds with the protein was conducted using the CDOCKER module within the docking module. The algorithm utilizes the CHARMM force field to dock flexible ligands into the previously defined rigid receptor active pocket, centered at the coordinates XYZ: −9.99476, −28.9201, −43.9379. Ligand conformations are optimized through molecular dynamics simulated annealing, and the binding energies between the ligands and the target protein are subsequently calculated. Finally, a comprehensive analysis of all docking results is performed. This includes preliminary ranking and screening based on the binding energies (CDOCKER Energy) and interaction energies (CDOCKER Interaction Energy) calculated by CDOCKER. Emphasis is placed on visualizing and analyzing the optimal binding conformations of each compound within the specified active pocket, centered at coordinates XYZ: −9.99476, −28.9201, −43.9379. By meticulously examining the key interaction patterns between the ligands and the target’s amino acid residues—including hydrogen bonding, hydrophobic interactions, and π-π stacking—the potential mechanisms of inhibition and variations in activity are elucidated from a structural perspective.

Cytotoxicity assay

The cytotoxicity of compounds V and V-24 was determined using B16F10 and A375 cells purchased from Hangzhou Jesimo Biotechnology Co., Ltd. Kojic acid and β-arbutin were used as positive controls. The cells were cultured in a DMEM high-glucose medium containing 10 % fetal bovine serum. Cells in the logarithmic growth phase were digested, counted, and seeded at a density of 8000 cells/well in a 96-well plate. Following a 24-h incubation at 37 °C in a 5 % CO₂, solutions of the compounds at different concentrations (200, 100, 50, 25, 12.5, 6.25, 3.125, 1.5625, 0.78125 µM) were added, and incubation continued for an additional 72 h. Subsequently, the drug-containing medium was removed, after which 100 µL of fresh medium and 10 µL of CCK-8 solution were added. After incubating for 2 h, the absorbance was measured at 450 nm. Cell viability was calculated using the following formula, with the untreated cell wells serving as controls and PBS used as blanks.

$$\mathit{Cell}viability\left(\%\right)=\frac{O{D}_{\mathit{Sample}}-O{D}_{\mathit{Blank}}}{O{D}_{\mathit{Control}}-O{D}_{\mathit{Blank}}}\times 100\%$$

Intracellular melanogenesis inhibition assay

The efficacy of compounds V and V-24 in inhibiting melanin production was evaluated using B16F10 and A375 cells. Kojic acid and β-arbutin were used as positive controls. Cells in the logarithmic growth phase were digested, counted, and seeded at a density of 5000 cells/well in a 96-well plate. Following a 24-h incubation at 37 °C in a 5 % CO₂, melanotan II (100 nM, control group) and varying concentrations of the compounds (100, 50, 25, 12.5, 6.25, 3.125, 1.5625, 0.78125 µM) were added, and incubation continued for 72 h. Subsequently, the drug-containing medium was removed, and 100 µL of RIPA cell lysis buffer was added and mixed thoroughly. Absorbance readings were taken at 405 nm. The melanin inhibitory rate was calculated using the following formula, with wells containing only α-MSH serving as controls.

$$\mathit{Melanin}inhibitoryrate\left(\%\right)=\frac{O{D}_{\mathit{Control}}-O{D}_{\mathit{Sample}}}{O{D}_{\mathit{Control}}}\times 100\%$$

To macroscopically observe the inhibitory effect of these compounds on intracellular melanin, B16F10 cells in the logarithmic growth phase were digested, counted, and seeded at a density of 3 × 10⁴ cells/well in a 24-well plate. Following a 24 h incubation at 37 °C in a 5 % CO₂, melanotan II (100 nM, control group) and varying concentrations of the compounds (100 µM) were added, and incubation continued for 72 h. Take photographs of the 24-well plate. Subsequently, the drug-containing medium was removed and washed twice with PBS. Then the cells were digested, centrifuged and photographed again. Finally, remove the supernatant, add 100 μL of 1 M NaOH solution, and incubate in a 60 °C water bath for 1 h before taking photographs once more.

Melanogenesis inhibition of zebrafish in vivo

Wild-type AB zebrafish [age: 6 h post-fertilization (6 hpf)] were utilized for in vivo anti-pigmentation assay. The breeding and husbandry of adult fish were carried out in compliance with the standards established by international AAALAC certification (NO: 001458). Zebrafish were randomly selected and placed in a 6-well plate, with 15 fish per well. The samples (V and V-24) and positive controls (kojic acid and β-arbutin) were dissolved in water (60 µM), with a total volume of 3 mL for each well, and a normal control group was also established. Following a 45 h dark incubation at 28 °C, 8 healthy zebrafish from each group were randomly selected and photographed using a dissecting microscope. Advanced image processing software was employed to analyze and collect data on the intensity of melanin signals intensity (S) in the zebrafish heads. A specific formula was applied to calculate and assess the whitening effects of the samples.

$$\mathit{Anti}\text{-}pigmentationeffect\left(\%\right)=\frac{S_{\mathit{Control}}-S_{\mathit{Sample}}}{S_{\mathit{Control}}}\times 100\%$$

3D human skin model assay

The 3D melanin skin model was obtained from CYBERIAD (Shanghai) Intelligent Technology Co., Ltd. (CY-RHME01). This model incorporates human-derived normal epidermal keratinocytes and melanocytes that have been proliferated and differentiated through inoculation, immersion culture, and air–liquid culture, mimicking the stratified tissue structure of normal human epidermis. The model was divided into four groups: blank group, control group, positive drug (kojic acid) group, and compound V-24 group, with six samples in each group (n = 6). The blank group received no light exposure and had the culture medium replaced daily. In contrast, the control group, positive drug group, and compound V-24 group underwent UVB irradiation using a UV light therapy device (SIGMA SH4B) starting on day 4 of air–liquid culture. They were irradiated for a total of seven consecutive days at a dosage of 50 mJ/cm² per session, after which the culture medium was replaced. From day 6 of air–liquid culture onward, drug administration was performed every 24 h for a total of six doses (100 µM). Following the completion of drug administration, the bottom surface of the skin model was wiped with a sterile cotton swab to eliminate any residual test substances.

Subsequently, a digital single-lens reflex (DSLR) camera (SONY ILCE-7M3 a7m3, lens SONY FE3.5–5.6/28–70) was positioned directly above the model to capture images for colorimetric analysis (parameter settings: manual mode, aperture f/8, exposure F22, shutter speed 1/80 s, and ISO 16000). The L value was then determined by making a circumferential incision along the model’s edge with a surgical blade, after which the model was carefully placed on photo paper using clean tweezers, ensuring the stratum corneum faced upward. A color difference meter (3nh NR10QC) was calibrated, and the detection aperture was pressed vertically against the model surface to obtain three readings, with the average value recorded as the final L value for each model.

Metabolic stability study

Plasma stability study: At designated time points (0, 5, 15, 30, 60, and 120 min), 99 µL of pre-warmed plasma was added to specified wells. In the 0-min well, 400 µL of MeOH:ACN (1:1) containing the internal standard (IS) was introduced, alongside 1 µL of pre-warmed compound solution (0.1 mM). For the remaining wells, 1 µL of pre-warmed compound solution (0.1 mM) was added first, followed by the introduction of 400 µL of MeOH:ACN (1:1) containing the internal standard (IS) at the designated time points to terminate the reaction. Following reaction quenching, the plate was shaken for 5 min at 600 rpm and centrifuged for 15 min at 6000 rpm. Finally, 100 µL of supernatant and 100 µL of ultrapure water were added to a 96-well test plate for LC-MS/MS analysis.

Microsomal stability study: At designated time points (0, 5, 15, 30, and 45 min), 30 µL of pre-warmed compound solution (1.5 µM) was added to specified wells. In the 0-min well, 150 µL of stop solution was introduced, alongside 15 µL of NADPH working solution (6 mM). For the remaining wells, 15 µL of NADPH working solution (6 mM) was added first, followed by the introduction of 150 µL of stop solution at the designated time points to terminate the reaction. Following reaction quenching, the plate was shaken for 10 min at 600 rpm and centrifuged for 15 min at 4000 rpm. Finally, 80 µL of supernatant and 140 µL of ultrapure water were added to a 96-well test plate for LC-MS/MS analysis.

Compliance with ethics requirements

All experiments involving animals were approved by the Experimental Animal Welfare Ethics Review Committee of Icas Testing Technology Service (Shanghai) Co., Ltd. (No. SHZ24090241).

Results and discussion

De novo molecular generation

The AI molecular design model based on RL presented in this article leverages DL algorithms and chemical knowledge. Its core components consist of an Agent network (Fig. 4), a Replay buffer, and an environment that evaluates various properties of the generated molecules. The Agent network employs the Soft Actor-Critic (SAC) algorithm, which incorporates an Actor-Critic structure. The Actor network is tasked with generating new molecular structures, while the Critic network assesses the quality of these generated molecules. Trained using the SAC algorithm’s target function, the Agent network effectively balances exploration and exploitation, thereby enhancing the model’s learning efficiency. The SAC algorithm incorporates chemical reaction templates (discussed in previous research)[59] and a molecular building block library as starting points, to ensure that the generated compounds possess good synthetic accessibility. It subsequently conducts forward reaction predictions, selecting appropriate combinations from the molecular building block library and reaction template library to produce novel molecules. During the molecular generation process, the Actor network receives evaluation scores from the molecular docking software AutoDock Vina, which serve as reward feedback from the Critic network. This interaction allows for an assessment of the docking compatibility of the generated molecules with the target protein, along with their drug-like properties, ensuring that the generated molecules exhibit not only high target activity but also favorable drug characteristics. After evaluation by the Critic network, high-quality molecules are stored in the Replay buffer, allowing for continual adjustments to the molecular generation strategy based on the reward signals from the RL framework. Through iterative training, the model learns the pathways for generating high-quality target molecules.

Utilizing this model, we focused on TYR as the target and established filtering parameters. We executed seven molecular generation tasks, each producing the top 100 molecules. From the resulting pool of 700 molecules, 17 compounds were selected for actual synthesis and activity evaluation. The criteria for selection were based on the following aspects: (1) Compound Structure: We prioritized compounds with pharmacophoric features, including phenolic groups, benzoic acids, formamides, or quinoline carboxylic acids. Additionally, compounds with superior structural characteristics were chosen for synthesis, informed by empirical research and parameter predictions. (2) Drug-like Properties: According to Lipinski’s rule of five, Veber’s rules, and the analysis and scoring of pharmacokinetic parameters ADMET (absorption, distribution, metabolism, excretion, and toxicity) for each compound in the model, molecules with poor pharmacokinetic properties are eliminated. This process facilitated the selection of compounds exhibiting favorable drug-like properties for subsequent synthesis. (3) Synthesis Routes: Preference was given to compounds with straightforward and efficient synthesis routes, as recommended by the model based on the corresponding raw material information.

Synthesis of de novo generated compounds

The synthetic routes for compounds I-XVII are illustrated in Fig. 5, which mainly involves condensation reactions, substitution reactions, as well as protection and deprotection processes. Specifically, compounds I-VI, XIII and XV were obtained by performing an amide condensation reaction with diisopropyl ethylenediamine (DIPEA), hydroxybenzotriazole (HOBt) and 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDCI) as reaction conditions. While compounds VII and XII were achieved under condensation reaction conditions using DIPEA and N,N,N’,N’-tetramethyl-O-(7-azabenzotriazol-1-yl)uronium hexafluorophospate (HATU).

Fig. 5.

The adopted synthetic strategies of de novo generated compounds I-XVII.

Compounds VIII, X and XI were synthesized through an initial condensation reaction, followed by demethylation utilizing BBr₃. Compounds 14 and 15 were subjected to a nucleophilic substitution reaction in the presence of DIPEA to yield intermediate 16. Then, the methyl ester of 16 was removed by 10 % NaOH solution to afford compound IX. Commercially available tert-butyl 1,4-diazepane-1-carboxylate 24 underwent nucleophilic substitution reactions with benzyl bromide to form 25. Subsequently, tert-butoxycarbonyl (Boc) group of 25 was removed using dioxane hydrochloride solution to provide the corresponding amine hydrochloride salt 26, which performed an amide condensation reaction to generate compound XIV.

Intermediate 30 was synthesized through an amido protection reaction of 28 in the presence of di-tert-butyl decarbonate. Subsequently, intermediate 30 underwent an esterification reaction with the corresponding carboxylic acid in the presence of dicyclohexylcarbodiimide (DCC) and dimethylaminopyridine (DMAP), ultimately resulting in the removal of the Boc group to yield compounds XVI and XVII.

Tyrosinase inhibitory activity screening of de novo generated compounds

The TYR inhibitory activity of all synthesized de novo compounds was assessed using mushroom TYR, with L-dopa and L-tyrosine serving as substrates. The results are presented in Table 1. Kojic acid, α-arbutin, and β-arbutin were employed as positive control agents. Over half of the tested compounds exhibited varying degrees of inhibition, with greater inhibitory activity observed against monophenolase compared to diphenolase. Among the 17 compounds examined, 10 compounds exhibited moderate to high inhibitory activity, with 7 compounds demonstrating inhibition levels surpassing those of α-arbutin (L-dopa: no inhibition; L-tyrosine: IC₅₀ = 332.7 ± 26.2 μM) and β-arbutin (L-dopa: IC₅₀ > 800 μM; L-tyrosine: IC₅₀ = 136.0 ± 12.2 μM), confirming the accuracy of the AI de novo molecular generation model. Notably, the most potent compound V (L-dopa: IC₅₀ = 18.5 ± 2.4 μM; L-tyrosine: IC₅₀ = 9.6 ± 1.1 μM) exhibited the highest inhibitory effect, exceeding that of kojic acid (L-dopa: IC₅₀ = 36.5 ± 5.5 μM; L-tyrosine: IC₅₀ = 18.8 ± 0.9 μM). Finally, a promising lead compound (V) with micromolar TYR inhibition was successfully identified through the AI de novo molecular generation. The subsequent work involves structure optimization of the AI-generated lead compound V to further enhance the inhibitory activity to achieve the nanomolar range.

Table 1.

Tyrosinase inhibitory activity of de novo generated compounds.

Compound	IC50 value ± SD (μM, n = 3)		Compound	IC50 value ± SD (μM, n = 3)
	L-dopa	L-tyrosine		L-dopa	L-tyrosine
I	> 800	> 800	XI	> 800	621.6 ± 82.8
II	> 800	423.5 ± 38.3	XII	199.2 ± 22.7	35.9 ± 1.2
III	> 800	> 800	XIII	> 800	99.7 ± 15.3
IV	> 800	197.7 ± 7.6	XIV	> 800	449.1 ± 36.1
V	18.5 ± 2.4	9.6 ± 1.1	XV	> 800	> 800
VI	> 800	>800	XVI	137.5 ± 5.1	16.0 ± 1.6
VII	> 800	>800	XVII	303.0 ± 47.6	164.3 ± 19.8
VIII	> 800	>800	kojic acid	36.5 ± 5.5	18.8 ± 0.9
IX	> 800	>800	α-arbutin	/	332.7 ± 32.0
X	461.5 ± 50.4	60.9 ± 1.9	β-arbutin	> 800	136.0 ± 12.2

Expert-Guided design and synthesis of structural optimization compounds based on the AI-generated lead compound

Based on the structure of the most potent de novo generated compound V and previous research experience,[59] the structural optimization is divided into five parts (Fig. 6). The Ar part primarily focuses on monosubstituted benzene rings and aromatic heterocycles, including furan, thiophene, and pyridine rings. In this section, substituents on the benzene ring mainly include electron-withdrawing groups at the meta and para positions (notably fluorine, chlorine, and cyano groups), and several electron-donating groups (methyl, methoxy, and ethoxy) were introduced.[60,61] The A and C parts involve the introduction of two to three different connecting chains to explore their impact on biological activity, with chains resembling cinnamic acid anticipated to enhance activity.[29,62,63] Based on previous findings, the introduction of the piperazine ring can facilitate hydrogen bonding interactions at the catalytic site, significantly increasing inhibitory activity.[59,64,65] Consequently, part B primarily focuses on piperazine rings while also assessing various straight-chain and cyclic diazole groups. Finally, the moiety D incorporates either mono- or disubstituted phenolic groups, which are crucial for chelating copper ions at the catalytic site.[15,59,66] These comprehensive strategies culminate in the design and synthesis of a total of 34 structurally optimized compounds (V-1-V-34).

Fig. 6.

The design of structural optimization compounds based on lead compound V.

Compounds V-1-V-34 were synthesized according to Fig. 7. Similarly, in the presence of DIPEA, HOBt and EDCI, compounds V-1-V-7, V-9-V-10, V-14, V-16-V-17 and V-23-V-34 were obtained through an amide condensation reaction. The amine intermediates participating in the amide condensation reaction, including compounds 44, 47, 50, 57, 60, 72 and 85–88, must be synthesized through nucleophilic substitution reaction in the presence of potassium carbonate or triethylamine, or reductive amination reaction using NaBH(OAc)₂, followed by Boc deprotection. Additionally, compounds V-8, V-11-V-13, V-15, and V-18-V-19 can be synthesized by initially conducting the amide condensation reaction, followed by Boc deprotection. Subsequent reactions involve a nucleophilic substitution reaction or a reductive amination reaction. Unlike the other compounds, compounds V-20-V-22 are generated via an amide condensation reaction, followed by a demethylation reaction in the presence of boron BBr₃.

Fig. 7.

The synthetic strategies of structural optimization compounds V-1-V-34.

Tyrosinase inhibitory screening of structural optimization compounds and SAR studies

The TYR inhibitory activity of all structural optimization compounds was evaluated using a consistent methodology, with the results presented in Table 2. Variability in TYR activity may arise from differences in the batches purchased; therefore, to ensure comparability, the inhibitory activities of compound V (L-dopa: IC₅₀ = 41.8 ± 3.0 μM; L-tyrosine: IC₅₀ = 11.9 ± 0.5 μM) and the positive control, kojic acid (L-dopa: IC₅₀ = 40.3 ± 2.9 μM; L-tyrosine: IC₅₀ = 20.8 ± 0.6 μM), were reassessed. All designed compounds demonstrated varying degrees of inhibitory activity, with more than half exhibiting more potent inhibition than compound V, and their efficacy significantly surpassed that of kojic acid. Remarkably, 13 of the compounds (V-22-V-34) showed nanomolar IC₅₀ values, ranging from 18 nM to 600 nM. Among them, compound V-24 (L-dopa: IC₅₀ = 0.018 ± 0.0007 μM; L-tyrosine: IC₅₀ = 0.020 ± 0.001 μM) exhibited the most potent inhibitory activity, with its IC₅₀ value being enhanced by 600 to over 2300 times compared to that of compound V.

Table 2.

Tyrosinase inhibitory activity of structural optimization compounds.

Compound	Structure	IC50 value ± SD (μM, n = 3)
		L-dopa	L-tyrosine
V		41.8 ± 3.0	11.9 ± 0.5
V-1		> 800	204.0 ± 3.9
V-2		> 400	294.9 ± 24.2
V-3		66.1 ± 4.6	60.6 ± 4.5
V-4		26.1 ± 4.6	2.9 ± 0.1
V-5		> 200	147.6 ± 25.4
V-6		296.1 ± 7.9	88.9 ± 2.9
V-7		119.1 ± 5.2	25.8 ± 1.1
V-8		> 800	298.2 ± 16.6
V-9		> 800	328.2 ± 4.9
V-10		19.0 ± 2.2	7.4 ± 0.4
V-11		> 400	20.1 ± 0.7
V-12		> 800	> 800
V-13		103.3 ± 5.6	131.8 ± 5.4
V-14		72.6 ± 3.1	83.2 ± 3.2
V-15		140.0 ± 15.6	41.8 ± 3.2
V-16		> 800	> 300
V-17		> 800	> 600
V-18		> 800	> 300
V-19		> 800	> 600
V-20		21.3 ± 0.8	9.1 ± 0.6
V-21		21.0 ± 0.82	8.2 ± 0.4
V-22		0.085 ± 0.016	0.122 ± 0.008
V-23		0.134 ± 0.008	0.183 ± 0.013
V-24		0.018 ± 0.0007	0.020 ± 0.001
V-25		0.147 ± 0.013	0.165 ± 0.008
V-26		0.020 ± 0.002	0.023 ± 0.002
V-27		0.596 ± 0.026	0.497 ± 0.070
V-28		0.019 ± 0.0002	0.021 ± 0.001
V-29		0.189 ± 0.005	0.216 ± 0.019
V-30		0.022 ± 0.001	0.021 ± 0.0003
V-31		0.153 ± 0.011	0.187 ± 0.003
V-32		0.021 ± 0.002	0.024 ± 0.002
V-33		0.237 ± 0.011	0.299 ± 0.007
V-34		0.032 ± 0.004	0.041 ± 0.005
kojic acid		40.3 ± 2.9	20.8 ± 0.6

Important SAR studies (Fig. 8) reveal that when the Ar moiety is a monofunctionalized phenyl group. It generally demonstrates superior activity, irrespective of whether the substitution occurs at the para or meta position. Linker A is ideally a methylene group, while Linker C is optimally a vinyl group. Furthermore, ring B is preferentially a piperazine ring. Notably, ring D serves as a key structural feature (pharmacophore) that significantly influences biological activity. When ring D is configured as 2,4-dihydroxyphenyl, the IC₅₀ value shifts from the micromolar to the nanomolar range, thereby greatly enhancing inhibitory activity. This enhancement is markedly superior to that observed with 2,5-dihydroxyphenyl, 3,5-dihydroxyphenyl, and mono-hydroxy-substituted phenyl compounds. These findings indicate that structural modifications based on compound V have successfully yielded a series of more potent TYR inhibitors, holding promise for further improving their anti-hyperpigmentation efficacy.

Fig. 8.

The SARs of structural optimization compounds V-1-V-34.

Table 2. Tyrosinase inhibitory activity of structural optimization compounds.

Structure-property relationship (SPR) studies

We also employed the AI drug discovery model utilized in this study to predict the drug-like properties and ADMET characteristics of compounds V and V-1–34. Furthermore, we conducted a discussion on the structure–property relationships (SPR). The specific predictions for drug-like and ADMET properties are available in Tables S1-S4 of the Supporting Information.

The SPR study reveals that all compounds conform to Lipinski’s Rules and Veber’s Rules in terms of drug-like property predictions. Analysis of the QED values presented in Table S1 shows that the para-methoxyphenyl and para-ethoxyphenyl groups (V-26–28) in the Ar moiety slightly diminish druggability. Additionally, when Linker B is a straight-chain diamine linker (V-8–9), a notable decrease in druggability is observed, whereas other substituent variations have minimal impact on this value. Concerning solubility and permeability, the use of Linker B as a straight-chain diamine linker (V-8–9) results in increased solubility but decreased permeability. Conversely, when Linker A and Linker C are represented by vinyl groups, they cause reduced solubility but enhanced permeability, while the remaining compounds maintain acceptable levels of solubility and permeability.

In terms of absorption and distribution property predictions (Table S2), all compounds show medium permeability (−5.7 < Caco2 value < -4.7), high human intestinal absorption rates (> 99 %), favorable blood–brain barrier (BBB) permeability (66 %-93 %), and low plasma protein binding rates (1 %). The majority of compounds demonstrate high P-glycoprotein efflux inhibition rates (> 60 %), suggesting a minimal probability of efflux; however, several compounds display moderate or low efflux inhibition rates (e.g., V-5, V-8, V-9, V-19, V-21, V-34). Eleven compounds are characterized by poor bioavailability (< 60 %), primarily those substituted with 2,4-dihydroxyphenyl groups, although this is not definitive, as the preferred compound V-24 exhibits good bioavailability (87.71 %). The VD_SS values indicate that compounds with electron-withdrawing substituents generally possess moderate in vivo tissue distribution levels (1–2 L/kg), while other compounds show lower levels of tissue distribution.

With respect to metabolism and excretion property predictions (Table S3), most compounds act as moderate inhibitors of CYP2D6, CYP3A4, CYP2C9, and CYP2C19. They show a moderate likelihood of being substrates for CYP2D6 and CYP3A4, while the probability of being substrates for CYP2C9 is very low. Regarding CYP1A2, compounds containing a vinyl group in the connecting chain or those with a dihydroxy-substituted phenyl group at ring D exhibit moderate inhibition rates, whereas other compounds present lower inhibition rates. The majority of compounds possess suitable half-life ranging from 2 to 4 h, although some have shorter half-life (< 2h, such as V-8 and V-17) or longer half-lives (> 4h, such as V-10, V-15, V-26, V-28, and V-30). All compounds exhibit high clearance rates exceeding 25 mL/min/kg.

In terms of toxicity predictions (Table S4), none of the compounds exhibit toxicity associated with polycyclic aromatic hydrocarbons, nor do they possess carcinogenic or eye-corrosive properties, which indicates a low probability of causing liver injury. However, the majority of compounds demonstrate significant hERG toxicity (> 90 %), with only ten compounds exhibiting relatively low hERG toxicity (V-18–19, V-24–28, V-30, V-32, V-34). Most compounds possess moderate to high LD₅₀ values (> 500), suggesting relatively low toxicity; however, a few compounds have lower LD₅₀ values (< 500), and these are not considered preferred candidates.

Overall, the optimized compound V-24 demonstrates favorable drug-like properties (QED = 0.83) and permeability (log D = 3.28, log P = 3.18, log S = −2.97). It exhibits high bioavailability (87.71 %), moderate BBB permeability (66.89 %), low plasma protein binding rate (1 %), and a low volume of steady state distribution (V_DSS) (0.81 L/kg). The compound shows moderate inhibition rates of metabolic enzymes (36 %-70 %), a moderate half-life (2.91 h), and a relatively high clearance rate (48.41 mL/min/kg), which suggests low toxicity. As such, V-24 is a candidate that warrants further in-depth in vitro and in vivo studies.

Molecular docking

Subsequently, molecular docking of the AI-optimized compound V and the structurally modified optimized compound V-24 was performed using Discovery Studio® 2023 to explore their interaction patterns with TYR. Additionally, two low-activity compounds, VII and XIV, were docked for comparative analysis. The optimal binding model is shown in Fig. 9. The two low-activity compounds, VII and XIV, exhibited negligible interactions with the active cavity of TYR, which explains their minimal inhibitory activity. However, compound V can effectively enter the active cavity of TYR, where the phenolic hydroxyl group at position 3 on the D ring can interact with the copper ion at the active center of TYR, a factor essential for its potent enzyme inhibition. Additionally, the benzene ring forms Pi-Pi stacked, Pi-alkyl, and Pi-sigma interactions with HIS263, ALA286, and VAL283, respectively. In contrast, compound V-24 demonstrates a greater ability to penetrate the active center of TYR. The D ring contains phenolic hydroxyl groups at both positions 2 and 4, with the hydroxyl group at position 4 also capable of interacting with the copper ion at the active center of TYR, while the hydroxyl group at position 2 forms a hydrogen bond interaction with ASN260. The benzene ring in the D moiety and the benzene ring in the Ar moiety form Pi-sigma, Pi-Pi stacked, and Pi-alkyl interactions with VAL283, SER282, HIS263, and PRO277, respectively. Furthermore, the carbonyl group in the linker can also establish a hydrogen bond interaction with HIS244. The increase in the number of hydrogen bond interactions, along with other interactions, likely contributes to the significant enhancement of its inhibitory activity. Moreover, molecular dynamic assay has been performed to validate that the binding mode of compounds V and V-24 were all stable, with compound V-24 demonstrating a more potent binding affinity to TYR (Fig. S1). The result of Surface Plasmon Resonance (SPR) assay validate that compounds V and V-24 were really target the TYR, and their K_D value is 24.2 ± 20.3 µM and 13.8 ± 4.2 µM (tyrosinase concentration: 20 µg/mL, pH 4.5, 1:1 binding ratio), respectively (Fig. S2).

Fig. 9.

The best binding mode of compounds V, VII, XIV and V-24 with tyrosinase (PDB ID: 2Y9X).

Cytotoxicity assay

The CCK-8 assay was employed to evaluate the cytotoxicity of kojic acid, β-arbutin, compounds V and V-24 in B16F10 and A375 cells (Fig. 10A). In B16F10 cells, kojic acid, β-arbutin and compound V exhibited no cytotoxicity at 200 μM, while compound V-24 demonstrated no cytotoxicity at 100 μM. In A375 cells, kojic acid and β-arbutin also showed no cytotoxicity at 200 μM; however, compounds V and V-24 displayed weak cytotoxicity at 200 μM, although no cytotoxicity was detected at 100 μM. In summary, the compounds did not show significant cytotoxicity at a concentration of 100 μM.

Fig. 10.

A. The cell viability of kojic acid, β-arbutin, compounds V and V-24 on B16F10 and A375 cells at various concentrations; B. The intracellular melanogenesis inhibitory rate of kojic acid, β-arbutin, compounds V and V-24 on B16F10 and A375 cells at various concentrations; C. Photograph of a 24-well plate following a 72h incubation of various compounds with B16F10 cells; D. Photographs of centrifuged and dissolved centrifuge tubes following a 72h incubation of various compounds with B16F10 cells.

Intracellular melanogenesis inhibition assay

To evaluate the inhibitory effects of the compounds on melanin synthesis in cells, we utilized an enhanced method to determine the melanogenesis inhibitory rates of kojic acid, β-arbutin, compounds V and V-24 in B16F10 and A375 cells.[66] Melanotan II (100 nM) served as a stimulant for melanin production in the cells, which were then treated with varying concentrations of the compounds. The results, illustrated in Fig. 10B, indicated that the inhibitory rate of melanin synthesis displayed a dose-dependent relationship: as the concentration of the compound increased, the melanin content significantly decreased. In B16F10 cells, the inhibitory effect of the positive control β-arbutin surpassed that of kojic acid, while the inhibitory effect of compound V was comparable to that of kojic acid and exceeded it at lower concentrations. The inhibitory effect of compound V-24 was akin to that of β-arbutin. Conversely, in A375 cells, compound V-24 demonstrated the most pronounced inhibitory effect, significantly outpacing kojic acid, β-arbutin, and compound V across all tested concentrations.

To observe the inhibitory effects of these compounds on intracellular melanin at a macroscopic level, B16F10 cells were incubated with 100 μM of these compounds after stimulation with melanotan II (100 nM). Comparisons were made by capturing images of the 24-well plates, cells after digestion and centrifugation, and the resulting dissolved cell solutions. As shown in Fig. 10C-D, the cells treated with melanotan II exhibited a significant darkening colour. Upon the addition of different compounds, the cells displayed varying degrees of color lightening, indicating that all compounds possess a certain capacity to inhibit melanin synthesis. The order of inhibitory effectiveness was observed as follows: compound V-24 > β-arbutin > kojic acid ≈ compound V.

The above results significantly confirm that through manual structural optimization of the AI-generated lead, the resulting compound V-24 exhibited a potent enhancement in the melanogenesis inhibitory effects. These findings further demonstrate that combining AI de novo molecule generation with expert-guided structural optimization represents a promising strategy for discovering high-potency candidate compounds.

Anti-pigmentation assay of zebrafish in vivo

The in vivo anti-pigmentation experiment utilized wild-type zebrafish at 6 h post-fertilization (6 hpf). Due to their transparency during early development, these zebrafish are easily observable. By 24 h post-fertilization, melanin begins to develop from the retinal epithelium and progressively distributes to various body regions, ultimately forming stable and visible melanin spots by 48 h. Throughout this process, drug interventions were administered to inhibit melanin synthesis, thereby enabling the evaluation of the whitening efficacy of the compounds. After 45 h of treatment, the melanin signal intensity in the zebrafish’s head was measured and analyzed. Kojic acid, β-arbutin, compound V, and V-24 all demonstrated significant anti-pigmentation effects (Fig. 11A). Compound V (6.83 ± 5.40 %) exhibited a relatively weak effect, comparable to that of β-arbutin (11.39 ± 3.45 %), while Compound V-24 (16.32 ± 2.13 %) displayed superior depigmentation results, on par with those of kojic acid (18.54 ± 6.11 %).

Fig. 11.

A. The anti-pigmentation of compounds V and V-24 in Zebrafish. Each treatment group included 8 fish; B. The anti-pigmentation effect of compound V-24 in a 3D human melanin skin model and the ΔL value is the difference between the L values of different treatment groups and the control group; C. Metabolic stability study of compound V-24 in human plasma and microsomes. (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, Control group vs. Treatment groups; Please refer to the Supporting Information for detailed explanations of the parameters in panel C.).

The examination of melanin accumulation in the heads of zebrafish revealed that the control group exhibited the highest and most pronounced melanin deposition. In contrast, the treatment groups administered kojic acid, β-arbutin, compound V, and V-24 displayed varying degrees of melanin reduction and relaxation. Notably, zebrafish treated with kojic acid and compound V-24 showed the least melanin deposition in their heads. These findings indicate that both compound V and V-24 possess varying levels of in vivo anti-pigmentation activity, with compound V-24 demonstrating superior efficacy.

3D human skin model assay

The optimized compound V-24 and the positive drug kojic acid were selected to evaluate their anti-pigmentation effects using a 3D human skin model. This model was created through the co-cultivation of melanocytes and epidermal cells in vitro, accurately mimicking the tissue structure of human skin, which comprises the basal layer, spinous layer, granular layer, and stratum corneum. In this model, melanocytes are situated in the basal layer and exhibit a uniform distribution of melanin granules. The model’s appearance closely resembles that of normal human skin, rendering it suitable for simulating the skin-darkening process caused by ultraviolet (UV) irradiation. For this experiment, the blank group received no treatment, while both the control and drug groups were subjected to continuous UVB irradiation starting on the fourth day of culture for a duration of seven days. The drug group commenced treatment on the sixth day of culture and continued for six days at a concentration of 100 µM. Following the treatment period, colorimetric assessment and L-value measurements were conducted.

As shown in Fig. 11B, the apparent color photographs demonstrated that the control group displayed a darker color than the blank group, while the kojic acid group appeared lighter than the control group. This indicated that kojic acid exerted an effective anti-pigmentation activity, thereby confirming the successful establishment of the model. Notably, the compound V-24 group exhibited the lightest color, significantly lighter than the kojic acid group, suggesting that compound V-24 displayed a superior anti-pigmentation effect compared to kojic acid. The assessment of L-values supports this trend: the control group (65.10 ± 0.48) showed the darkest color, followed by the blank group (68.42 ± 0.88), then the kojic acid group (70.98 ± 0.68), with the compound V-24 group (83.38 ± 0.40) yielding the lightest color. The ΔL value, which represents the difference in L-values between the treatment groups and the control group, revealed a significant increase for the compound V-24 group relative to the control group, as well as a pronounced elevation compared to the kojic acid group. These findings underscore that compound V-24 not only demonstrates a marked whitening effect but also surpasses the effectiveness of kojic acid at the same concentration.

Metabolic stability study

We further investigated the metabolic stability of compound V-24 in human plasma and liver microsomes, using procaine and verapamil as controls to validate the effectiveness of the testing method, respectively. As illustrated in Fig. 11C, compound V-24 demonstrates significant stability in human plasma, exhibiting resistance to metabolic degradation. After a 120 min incubation with plasma at 37°C, the retention rate of the intact compound was 89.06 %, with a half-life (t_1/2) of 524.17 min. This indicates that the compound can achieve an adequate therapeutic concentration at the site of administration. In contrast, the compound exhibited a moderate clearance rate in human liver microsomes, with its retention rate declining to 41.96 % after 45 min, a t_1/2 of 34.79 min, and an E_h value of 69.00 %. These findings suggest that V-24 is predominantly metabolized in the liver, supporting the notion that it is unlikely to accumulate in the body and induce systemic toxicity.

Conclusion

This study focuses on the rate-limiting enzyme TYR involved in melanin synthesis, employing a reaction-template-based RL model derived from the SAC algorithm for de novo molecular generation aimed at discovering novel TYR inhibitors. The model utilizes a sequence of decision-making processes, signal feedback mechanisms, and dynamic learning techniques to produce molecules with potent target affinity and synthetic accessibility, thereby achieving multi-objective optimization. Expert-guided pre-screening of 700 AI-generated molecules led to the final synthesis and evaluation of 17 compounds. Notably, compound V (L-dopa: IC₅₀ = 18.5 ± 2.4 μM; L-tyrosine: IC₅₀ = 9.6 ± 1.1 μM) demonstrated the highest inhibitory activity against TYR, which was more potent than the reference drug kojic acid.

In the subsequent work, structural modifications were performed based on compound V as the lead compound, resulting in the design and synthesis of 34 structurally optimized compounds V-1-V-34. Subsequent evaluation of their TYR inhibitory activity revealed that over half of these compounds exhibited enhanced inhibition compared to compound V, with 13 compounds achieving IC₅₀ values in the nanomolar range. Notably, the most optimized compound, V-24 (L-dopa: IC₅₀ = 0.018 ± 0.0007 μM; L-tyrosine: IC₅₀ = 0.020 ± 0.001 μM), displayed IC₅₀ values that were 600 to 2300 times lower than those of compound V. Cellular experiments demonstrated that both compounds V and V-24 exhibited no cytotoxicity at concentrations of 100 μM and effectively inhibited the synthesis of intracellular melanin in a dose-dependent manner. Furthermore, zebrafish studies confirmed that these compounds successfully suppressed melanin production in vivo. In the human 3D melanin skin model, compound V-24 effectively inhibits melanin synthesis and demonstrates more potent anti-pigmentation activity than kojic acid.

In conclusion, this study successfully employed an AI-driven de novo molecular generation model to identify a lead compound with micromolar TYR inhibitory activity. Subsequent expert-guided structural optimization based on this lead compound led to a series of compounds with nanomolar TYR inhibition. This “AI de novo Molecular Generation + Expert-Guided Structural Optimization” work demonstrates that integrating an AI algorithm with traditional medicinal chemistry experience is a novel approach and efficiency-redefined strategy to accelerate drug discovery.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: