Discovery of Potential Antihypertensive Agents from the Marine Microalga Phaeodactylum tricornutum Through Metabolite Profiling and In Silico Analysis

Abstract

Hypertension remains a leading cause of global morbidity and mortality, and angiotensin-converting enzyme (ACE) represents a central therapeutic target within the renin–angiotensin–aldosterone system. Marine microalgae, particularly Phaeodactylum tricornutum, provide an underexplored reservoir of structurally diverse metabolites with potential cardiovascular relevance. In this in silico study, we characterized metabolites putatively annotated by UPLC-ESI-HRMS and evaluated their predicted ACE inhibitory potential. We performed molecular docking with AutoDock 4 and assessed pharmacokinetic and toxicological properties using the SwissADME, PASS, and ProTox platforms. Several metabolites showed favorable binding orientations within the ACE catalytic pocket, including interactions with key residues and proximity to the zinc-binding motif. Lehualide G, Val–Asn–Pro, tanariflavanone B, hydroxyterbinafine, and anhydro-vitamin A exhibited the most favorable docking profiles. PASS predictions indicated vascular-related bioactivity signals for selected compounds, whereas ADMET modeling revealed heterogeneous but classifiable pharmacokinetic and safety characteristics. The convergence of predicted binding compatibility, bioactivity signals, and stratified safety margins supports P. tricornutum as a promising source of candidate molecules for further experimental validation in antihypertensive research.

1. Introduction

Hypertension, also known as high blood pressure, represents a disruption in blood pressure regulation and remains one of the leading causes of mortality worldwide. It is estimated that approximately 1.4 billion adults aged 30–79 years live with hypertension [1]. Most individuals with high blood pressure may remain asymptomatic, even when readings reach dangerously elevated levels. The World Health Organization estimates that 46% of adults with hypertension are unaware of their condition. When symptoms do occur, they may include chest pain, dizziness, edema, seizures, severe headache, and visual disturbances [2].

Based on its origin, hypertension can be classified as primary, when associated with genetic and multifactorial etiologies, or secondary, when it results from underlying pathological conditions [3]. Blood pressure regulation involves multiple interconnected systems, including natriuretic peptides, endothelial function, the sympathetic nervous system, immune mechanisms, and particularly the renin–angiotensin–aldosterone system (RAAS) [4]. Prevention and initial treatment typically focus on lifestyle modifications, such as dietary changes, increased physical activity, reduced alcohol intake, improved sleep, and pharmacological interventions targeting angiotensin-converting enzyme, angiotensin receptors, and calcium channels [5,6,7].

Currently, most antihypertensive drugs are derived from plant compounds. However, microalgae have emerged as a promising alternative source of bioactive molecules with potential health applications, including the development of nutraceuticals and functional foods [8]. Algae are primarily photosynthetic organisms that exist in both micro- and macroscopic forms and include both prokaryotic and eukaryotic members [9]. Microalgae produce a wide range of secondary metabolites with documented biological activities. Despite the identification of thousands of such metabolites, comprehensive characterization and pharmacological evaluation remain limited [10].

Over the past decades, multiple pharmacological strategies have been developed to manage high blood pressure [11]. In the 1970s, captopril was introduced as the first orally active angiotensin-converting enzyme inhibitor (ACEI) [12]. Subsequent structural modifications led to the development of enalapril and lisinopril [13], followed by additional derivatives such as benazepril, perindopril, quinapril, and ramipril, which remain in clinical use [14].

Although these agents are effective, they are associated with adverse effects, including dry cough, angioedema, electrolyte imbalance, dizziness, headache, and, in some cases, a decline in renal function [12,15]. Variability in therapeutic response and reports of reduced efficacy have also been described [16,17]. These limitations highlight the need to explore novel bioactive scaffolds with improved safety and pharmacological profiles. In this context, angiotensin-converting enzyme (EC 3.4.15.1; ACE) remains a strategic molecular target due to its central role in the RAAS. ACE catalyzes the conversion of angiotensin I, an inactive peptide, into angiotensin II, a potent vasoconstrictor that elevates blood pressure [18,19,20]. RAAS remains a principal therapeutic axis in antihypertensive drug development, given its role in blood pressure control and fluid and electrolyte balance [21]. Renin, initially synthesized as prorenin and secreted by afferent renal arterioles, initiates the cascade by cleaving angiotensinogen to produce angiotensin I [22,23].

Angiotensin I is subsequently converted into angiotensin II through ACE-mediated hydrolysis [24]. ACE is a Zn²⁺-dependent metalloprotease located on the membrane of parenchymal cells, where it hydrolyzes angiotensin I by cleaving the histidyl–leucine bond [25].

In the present work, an integrative in silico approach was implemented, using Food and Drug Administration (FDA)-approved ACE inhibitors as reference compounds to systematically evaluate metabolites extracted from Phaeodactylum tricornutum.

The objective was to identify candidate molecules with structural compatibility with ACE, favorable binding characteristics, and acceptable pharmacokinetic and toxicity profiles, thereby supporting their prioritization as potential antihypertensive agents for further experimental validation.

2. Materials and Methods

2.1. Biologic Materials of Phaeodactylum tricornutum

The diatom Phaeodactylum tricornutum is part of the strain collection of the Molecular Physiology Laboratory at the Center for Applied Research in Public Health (CIASAP), Faculty of Medicine, Autonomous University of Sinaloa, Culiacán, Mexico. This strain was provided by the culture collection of the Center for Scientific Research and Higher Education of Ensenada (CICESE) in Ensenada, Baja California, Mexico.

2.2. Microalgae Culture Condition and Biomass Processing

The diatom was cultivated following the methodology previously used by Norzagaray-Valenzuela et al., 2016 [26], Saline water (34 g NaCl L⁻¹) and F/2 medium proposed by Guillard and Ryther [27]. The following chemicals were used for growth medium preparation: NaNO₃, NaH₂PO₄ · H₂O, FeCl₃ · 6H₂O, Na₂ EDTA · 6H₂O, MnCl₂ · 4H₂O, ZnSO₄ · 7H₂O, CoCl₂ · 6H₂O, CuSO₄ · 5H₂O, Na₂MoO₄ · 2H₂O, thiamine, biotin and cyanocobalamin were used; the medium was adjusted to a pH of 7.6. The inoculum was prepared by gradually increasing the culture volume through weekly subculturing, starting from 10 mL and progressing to 20 L of growth medium. Inoculum concentration was 10% (v/v) of the culture volume. The diatom was grown in a polycarbonate carboy with 1% CO₂, allowing gas exchange and the suspension of microorganisms in the liquid medium, under warm white light lamps (continuous irradiance of 120–130 μmol photons m⁻² s⁻¹) and at 22 °C. Batch cultures were grown until the cells reached the stationary phase. Biomass recovery was carried out using a flocculation process with chitosan (C₅₆H₁₀₃N₉O₃₉), sodium hydroxide (NaOH), and acetic acid (CH₃COOH) [28]. Vitamins, chitosan, and other chemical reagents were obtained from Sigma-Aldrich (St. Louis, MO, USA). The culture was harvested in the late log phase of growth. The wet biomass was then placed in an oven at 40 °C for 24 h to remove water, resulting in dry biomass.

2.3. Preparation of the Biomass Methanolic Extract

The methanolic extract was prepared following the methodology as reported by Pereira and coworkers, with minor modifications [29]. Thirty grams of dried biomass was macerated in a 250 mL methanol/water solution (4:1 v/v). The biomass was extracted by maintaining the mixture under continuous agitation for 24 h in the absence of light at 35 °C. The mixture was then separated by centrifugation at 10,000× g/10 min, and the supernatant was recovered. This was then filtered in two stages: first, using Whatman No. 2 filter paper, followed by a second filtration using a 0.22 μm Captiva Econo Filter acrodisc, Agilent^® (Andover, MA, USA). The solvents were removed from the extract using a rotary evaporator at 45 °C, and the extract was finally lyophilized. The lyophilized extract was stored at −80 °C for later use.

2.4. Procedure of Non-Targeted Metabolomic Analysis

Extracts were subjected to non-targeted metabolomic analyses, following the methodology described by Hernández-Peña et al. and Rodríguez-Castillo et al., with minor adjustments [30,31].

Non-targeted metabolomic analyses were performed using an Ultra-Performance Liquid Chromatograph (Acquity^TM UPLC^TM I-Class system, Waters Corporation; Milford, MA, USA) coupled with a high-resolution Q-TOFMS High-Resolution Mass Spectrometer (HRMS) using an electrospray ionization source in positive and negative modes (Synapt^TM HDMS Waters Corporation, Milford, MA, USA).

A reversed-phase Acquity^TM BEH^TM column C18 (Waters Corporation, Milford, MA, USA) (2.1 × 150 mm, 1.7 µm) was used for the chromatographic separation. The mobile phases consisted of water containing 0.1% formic acid (A) and acetonitrile containing 0.1% formic acid (B), both LC–MS grade (JT Baker^®, Phillipsburg, NJ, USA). The gradient program was as follows: 0–0.5 min, 95% A; 0.5–25.5 min, linear gradient to 100% B; 25.5–27.5 min, column wash with 100% B; 27.51 min return to initial conditions (95% A); followed by 3 min for column re-equilibration. The flow rate was set to 0.2 mL/min, the column temperature to 40 °C, and the injection volume to 5 µL.

All samples (including blank extraction) were resuspended in 1 mL of acetonitrile/ultra-pure water (70:30 v/v) and filtered through a 0.2 µm PTFE membrane (Agilent Technologies, Santa Clara, CA, USA). Solvent blanks and extraction blanks were analyzed under the same conditions to monitor background signals. The sample was analyzed by PDA, and ions were generated using an electrospray ionization (ESI) source in negative and positive modes. Both ionization modes were injected separately. Under negative electrospray ionization (ESI-) conditions, the capillary voltage was adjusted to 2 kV and the cone voltage to 40 V.

The source temperature was maintained at 150 °C, with a cone gas flow of 20 L/h. The desolvation temperature was set at 350 °C, and the desolvation gas flow was 600 L/h. For positive electrospray ionization (ESI+), the capillary voltage was increased to 3 kV while the cone voltage was maintained at 40 V. The source temperature was set at 130 °C. The desolvation temperature remained at 350 °C, with a desolvation gas flow of 700 L/h. Data were acquired in continuum mode with an MS scan time of 1.5 s. In both ionization modes, spectra were obtained using MSE acquisition, employing argon as the collision gas. Collision energy in the trap region was fixed at 6 eV for low-energy scans (Function 1) and ramped from 20 to 40 eV for high-energy scans (Function 2). During acquisition, Leucine-Enkephalin was used as a mass reference at m/z 554.2615 for ESI- and 556.2771 for ESI+, and was infused directly at a flow of 5 µL/min at a concentration of 2 ng/mL, enabling internal mass calibration. Positive and negative adducts were considered.

The metabolomic raw data were analyzed using Progenesis^® QI for metabolites (Nonlinear Dynamics version 2.3, Waters Corporation), calibrated with lock mass, and normalized. Putative annotation of compounds was performed using the ChemSpider Database (Version: 1.0.6905.308112) and Progenesis MetaScope (Version: 1.0.6907.37313), with an isotope-similarity filter of 90%. Subsequently, a handpicked database was conducted for each metabolite, and its putative candidates were confidence-filtered using score punctuation ≥ 40 and isotope similarity ≥ 90 as criteria. The metabolites that passed the criteria of score and isotope similarity were chosen for this first stage. According to the Metabolomics Standards Initiative (MSI), these assignments were classified as MSI level 2 annotations, corresponding to putatively annotated compounds, because they were supported by accurate mass, isotopic similarity, and database matching without confirmation using authentic analytical standards under identical experimental conditions.

2.5. Molecular Docking Analysis of ACE

The angiotensin-converting enzyme (ACE) is a zinc-dependent metalloprotease composed of two catalytic domains (N- and C-domains). Molecular docking was performed against the C-domain because it is primarily associated with angiotensin I hydrolysis and is the main target of clinically used ACE inhibitors [32]. The crystal structure of human ACE C-domain (PDB ID: 1UZF) was obtained from the protein data bank (PDB) [33]. The protein chain used for docking (chain A) was selected, and the co-crystallized ligand (D-captopril) was removed prior to docking. Protein preparation was carried out using AutoDockTools (MGLTools 1.5.7).

All crystallographic water molecules were removed. Hydrogen atoms were added, and protonation states of titratable residues were assigned at physiological pH (7.4).

The catalytic Zn²⁺ ion was retained during receptor preparation to preserve the active site’s integrity. Kollman united-atom charges were assigned to the protein, and atomic coordinates were saved in PDBQT format. Canonical SMILES for metabolites putatively annotated from P. tricornutum extracts were obtained from the PubChem database (https://pubchem.ncbi.nlm.nih.gov, accessed on 3 October 2025) and ChemSpider database (https://www.chemspider.com, accessed on 3 October 2025). Ligand 3D structures were generated and optimized using Open Babel 3.1.0 [34]. Geometry minimization was performed using the MMFF94 force field (~5000 steps). Ligands were protonated at pH 7.4, and Gasteiger partial charges were assigned [35,36,37]. Ligands were saved in PDBQT format. The docking grid was configured to allow ligand accommodation within the catalytic pocket, ensuring proper coverage of the zinc-coordination region [36]. Subsequently, the docking grid was centered on the catalytic site at coordinates X = 36.98, Y = 27.05, and Z = 50.65, while the grid box size was set to 40 × 40 × 40 points with a grid spacing of 0.564 Å, following the methodology proposed by Zarei et al., and Qiu et al., [38,39]. Molecular docking for metalloproteins with zinc was performed using AutoDock 4.2 with the Lamarckian Genetic Algorithm [40]. Multiple independent runs were performed for each ligand, and poses were clustered by RMSD. The lowest-energy conformation from the most populated cluster was selected for analysis. Captopril, lisinopril, and enalaprilat were included as positive controls. Ligands were prioritized based on clustering reproducibility, binding mode within the catalytic site, proximity to Zn²⁺, and relative docking scores relative to controls. A threshold of ≤−7.0 kcal·mol⁻¹ was used as an initial filter, recognizing the approximate nature of docking scores [41].

2.6. Redocking Validation Using RMSD Analysis

Docking validation was performed by redocking ligand D-captopril into the ACE C-domain structure (PDB ID: 1UZF) using the same grid box configuration and docking parameters described above. The resulting docked pose was compared with the experimental crystallographic conformation of D-captopril from the original PDB complex. Root mean square deviation (RMSD) values were calculated in PyMOL (v3.1.6.1) [42]. After structural alignment of the receptor, the protein was kept fixed and only heavy atoms were considered (hydrogen atoms were excluded). An RMSD value ≤ 2.0 Å was considered indicative of successful pose reproduction and adequate docking accuracy (Figure S1) [36,43,44,45].

2.7. Prediction of Biological Activity

Biological activity prediction was performed using the PASS online server (Prediction of Activity Spectra for Substances) (https://way2drug.com/PassOnline/predict.php, accessed on 18 November 2025). PASS estimates the probability of activity (Pa) and inactivity (Pi) for each compound based on structure–activity relationships derived from curated experimental datasets using Bayesian statistical modeling [46]. Compounds were considered to have potential biological activity when Pa exceeded Pi, with higher Pa values indicating a greater likelihood of activity. The analysis was conducted according to previously reported procedures with minor modifications [47].

2.8. In Silico Physicochemical and ADME Profiling, and Toxicity Prediction

The physicochemical and ADME properties of the prioritized ligands were predicted using SwissADME (default settings), including molecular weight, lipophilicity (logP), topological polar surface area (TPSA), hydrogen bond donors and acceptors, gastrointestinal absorption, blood–brain barrier permeability, and Lipinski’s rule-of-five compliance.

Toxicity was independently evaluated using ProTox 3.0, which provides predicted oral LD₅₀ values and toxicity class estimations based on machine learning models (https://tox.charite.de/protox3/, accessed on 25 November 2025) [48,49]. Compounds were prioritized according to predefined drug-likeness and toxicity criteria (TC ≥ 4).

3. Results

3.1. Non-Targeted Metabolomic Analysis

UPLC-ESI-HRMS profiling of the methanolic extract of P. tricornutum resulted in the detection of 662 mass spectral features. After applying annotation confidence criteria and removing redundant signals, 162 putatively annotated compounds were retained for subsequent molecular computational analyses (Table S1). The overall experimental and computational workflow of the study is summarized in Figure 1.

3.2. Molecular Docking Screening

A total of 162 annotated metabolites were subjected to molecular docking analyses targeting the ACE C-domain. Utilizing a predetermined docking threshold of ≤−7.0 kcal·mol⁻¹, 18 compounds emerged with favorable predicted binding energies, warranting further interaction analysis (

Table 1. Molecular docking-derived binding free energies of Phaeodactylum tricornutum metabolites against the human ACE C-domain.

PUBCHEM CID	Molecular Formula	Compound	Binding Affinity ΔG (kcal.mol−1)	Two-Dimensional Structure
71728359	C14H24N4O5	Val-Asn-Pro	−8.178
21917706	C21H25NO	Hydroxyterbinafine	−7.900
71464556	C25H45NO5	3-hydroxylinoleoylcarnitine	−7.380
24771804	C16H24O	(5E,10E)-5,10-Pentadecadien-1-ol	−7.064
13730286	C20H32O	7,8-Dihydroretinol	−7.741
70698171	C15H24O5	Merulin B, (rel)-	−7.374
14165048	C22H30O2	Abieta-6,8(14),9(11),12-tetraen-12-ol	−7.642
91825614	C20H34O4	4-hydroxy-6-(15-hydroxypentadecyl)-pyran-2-one	−7.780
11113726	C30H34O6	Tanariflavanone B	−8.105
20348793	C18H18O4	7C-aglycone	−7.425
85908104	C18H28O4	4-hydroxy-6-(2-oxotridecyl)pyran-2-one	−7.781
16190788	C21H32N2O3	Methyl 4-([4-cyclohexyl-3-(2-hydroxyethyl)-1-piperazinyl]methyl)benzoate	−7.382
0698273	C20H30O2	6β-hydroxyferruginol	−7.759
162952	C17H24O4	Gingerdione	−7.283
54733285	C25H34O4	Lehualide G	−8.429
5287678	C20H28	Anhydrovitamin A	−7.792
5388962	C20H28N2O5	Enalapril	−9.438
5362119	C21H31N3O5	Lisinopril	−8.449
447055	C9H15NO3S	Captopril	−7.651

ΔG values correspond to the lowest-energy binding pose obtained from the most populated docking cluster using AutoDock 4.2. More negative values indicate stronger predicted binding affinity. Reference inhibitors (enalapril, lisinopril, and captopril) were included as positive controls. Note: Atom colors in the 2D structures follow conventional chemical notation: C, black/gray; O, red; N, blue; H, omitted or explicitly shown when applicable.

). The reference inhibitors exhibited binding energies of −9.438 kcal·mol⁻¹ for enalapril, −8.449 kcal·mol⁻¹ for lisinopril, and −7.060 kcal·mol⁻¹ for captopril.

Notably, among the natural metabolites, the most promising docking scores were recorded for lehualide G (−8.429 kcal·mol⁻¹), Val-Asn-Pro (−8.178 kcal·mol⁻¹), tanariflavanone B (−8.105 kcal·mol⁻¹), hydroxyterbinafine (−7.900 kcal·mol⁻¹), and anhydrovitamin A (−7.792 kcal·mol⁻¹).

3.3. Interaction of the Ligands with the Amino Acid Residues of ACE

Detailed interaction analysis of the selected ligands showed that several compounds occupied both the S1 and S2′ pockets and extended toward the HEXXH–Zn²⁺ catalytic motif, indicating structural compatibility with the ACE active site (Figure 2, Table S2). Compounds such as hydroxyterbinafine, 3-hydroxylinoleoylcarnitine, 4-hydroxy-6-(15-hydroxypentadecyl)-pyran-2-one, and (5E,10E)-5,10-pentadecadien-1-ol established interactions within S1 and S2′ while positioning functional groups in proximity to the catalytic zinc ion. Hydrogen bonds within the 2.2–3.2 Å range were observed with residues including Ala354, Glu384, Tyr523, Gln281, and Lys511. Our results also showed interactions with His383, His387, and Glu411, which form the canonical HEXXH motif. In addition, hydrophobic contacts were identified with Val380, Phe457, Phe527, and Tyr523 (3.3–5.0 Å), contributing to stabilization within the catalytic channel.

Reference inhibitors, captopril and lisinopril, displayed the expected binding orientations in S1 and S2′, and proximity to the Zn²⁺ catalytic site, consistent with their known crystallographic binding modes. Several natural metabolites showed comparable spatial occupation of the catalytic cavity, although differences in interaction density and zinc proximity were observed among compounds. Based on the extent and distribution of predicted contacts, hydroxyterbinafine and 3-hydroxylinoleoylcarnitine exhibited the highest number of simultaneous interactions across S1, S2′, and the catalytic region. Other compounds demonstrated more localized binding profiles, which were primarily restricted to S1 and hydrophobic contacts. These variations suggest differential binding orientations among the selected metabolites within the ACE active site.

3.4. PASS Biological Prediction of Vasoactive and Antihypertensive Potential

The PASS-based prediction of biological activity revealed different probabilities for vasodilatory and antihypertensive mechanisms among the evaluated metabolites and the reference ACE inhibitors (

Table 2. PASS biological prediction of vasoactive and antihypertensive activity of P. tricornutum metabolites.

		VP		VC		V		AHT
Molecular Formula	Compound	Pa	Pi	Pa	Pi	Pa	Pi	Pa	Pi
C14H24N4O5	Val-Asn-Pro	0.234	0.222	0.372	0.071	-	-	0.268	0.095
C21H25NO	Hydroxyterbinafine	0.255	0.198	0.371	0.071	-	-	-	-
C25H45NO5	3-hydroxylinoleoylcarnitine	0.975	0.002	0.231	0.204	0.905	0.004	0.454	0.028
C16H24O	(5E,10E)-5,10-Pentadecadien-1-ol	0.270	0.180	-	-	-	-	-	-
C20H32O	7,8-Dihydroretinol	0.296	0.150	0.269	0.153	-	-	-	-
C15H24O5	Merulin B, (rel)-	0.244	0.211	-	-	-	-	-	-
C20H28O	Abieta-6,8(14),9(11),12-tetraen-12-ol	0.388	0.088	-	-	-	-	-	-
C20H34O4	4-hydroxy-6-(15-hydroxypentadecyl)-pyran-2-one	0.504	0.025	-	-	0.409	0.036	-	-
C30H34O6	Tanariflavanone B	-	-	-	-	0.193	0.155
C18H18O4	7C-aglycone	0.314	0.135	0.296	0.123	0.366	0.048	-	-
C18H28O4	4-hydroxy-6-(2-oxotridecyl)pyran-2-one	0.255	0.197	0.383	0.065	0.497	0.022	-	-
C21H32N2O3	Methyl 4-{[4-cyclohexyl-3-(2-hydroxyethyl)-1-piperazinyl]methyl}benzoate	0.317	0.132	0.267	0.156	0.197	0.151	0.208	0.144
C20H30O2	6β-hydroxyferruginol	0.516	0.041	-	-	-	-	-	-
C17H24O4	Gingerdione	0.642	0.016	0.339	0.090	0.427	0.032	-	-
C25H34O4	Lehualide G	-	-	0.462	0.033	0.414	0.035	-	-
C20H28	Anhydrovitamin A	0.456	0.055	-	-	-	-	-	-
C9H15NO3S	Captopril	-	-	0.661	0.010	0.297	0.073	-	-
C21H31N3O5	Lisinopril	-	-	0.534	0.021	-	-	-	-
C20H28N2O5	Enalapril	-	-	0.740	0.006	-	-	-	-

Abbreviations: Pa—probability of activity; Pi—probability of inactivity; VP—peripheral vasodilator; VC—coronary vasodilator; V—vasodilator; AHT—antihypertensive activity. Activity values were obtained using the PASS prediction platform for each compound. (-) denotes no predicted activity.

). Activity probabilities (Pa) were interpreted relative to their corresponding inactivity probabilities (Pi), with Pa > Pi indicating a putative biological association under the similarity-based model.

PASS analysis identified 3-hydroxylinoleoylcarnitine as the most consistently predicted vasoactive compound, exhibiting the highest probabilities for peripheral vasodilation (Pa = 0.975), general vasodilatory activity (Pa = 0.905), and a moderate antihypertensive signal (Pa = 0.454), all with Pa values markedly exceeding Pi. This profile distinguishes it as the strongest endpoint candidate within the natural metabolite panel.

Among additional metabolites, gingerdione, 6β-hydroxyferruginol, and both pyran-2-one derivatives demonstrated coherent peripheral and/or general vasodilatory probabilities (Pa > Pi), suggesting structural similarity to compounds involved in vascular modulation. Lehualide G showed notable coronary vasodilator activity (Pa = 0.462), while Val–Asn–Pro presented moderate but consistent signals across peripheral and antihypertensive endpoints.

In contrast, the reference ACE inhibitors displayed a distinct pattern dominated by coronary vasodilator activity, with enalapril (Pa = 0.740), captopril (Pa = 0.661), and lisinopril (Pa = 0.534) showing strong and selective VC predictions. These compounds did not demonstrate high peripheral vasodilator activity in the PASS model, suggesting a pharmacological profile that differs from that of several natural metabolites that display stronger structural similarity to known vasoactive pharmacophores.

3.5. In Silico ADME and Drug-Likeness Evaluation

The in silico ADME evaluation revealed heterogeneous physicochemical and pharmacokinetic profiles among the prioritized metabolites and reference ACE inhibitors (

Table 3. In silico prediction of drug-likeness and pharmacokinetic parameters of P. tricornutum metabolites and standard drugs.

Molecular Formula	Compound	MW	HBA	HBD	RB	iLOGP	LP	PAINS	BRENK
C14H24N4O5	Val-Asn-Pro	328.36	6	4	9	0.87	0	0	0
C21H25NO	Hydroxyterbinafine	307.43	2	1	5	3.88	0	0	1
C25H45NO5	3-hydroxylinoleoylcarnitine	439.63	5	1	20	3.17	2	0	1
C16H24O	(5E,10E)-5,10-Pentadecadien-1-ol	232.36	1	0	2	3.07	0	0	1
C20H32O	7,8-Dihydroretinol	288.47	1	1	6	3.07	0	0	1
C15H24O5	Merulin B, (rel)-	284.35	5	2	1	1.98	0	0	1
C20H28O	Abieta-6,8(14),9(11),12-tetraen-12-ol	284.44	1	1	1	3.58	1	0	0
C20H34O4	4-hydroxy-6-(15-hydroxypentadecyl)-pyran-2-one	338.48	4	2	15	4.09	0	0	0
C30H34O6	tanariflavanone B	490.59	6	3	6	4.98	0	0	1
C18H18O4	7C-aglycone	298.33	4	1	5	2.33	0	1	1
C18H28O4	4-hydroxy-6-(2-oxotridecyl)pyran-2-one	308.41	4	1	12	3.17	0	0	0
C21H32N2O3	Methyl 4-{[4-cyclohexyl-3-(2-hydroxyethyl)-1-piperazinyl]methyl]benzoate	360.49	5	1	7	3.97	0	0	0
C20H30O2	6β-hydroxyferruginol	302.45	2	2	1	3.31	0	0	0
C17H24O4	Gingerdione	292.37	4	1	10	3.14	0	0	0
C25H34O4	Lehualide G	398.54	4	1	12	4.83	0	0	1
C20H28	Anhydrovitamin A	268.44	0	0	4	4.26	1	0	1

Abbreviations: MW—molecular weight; HBA—hydrogen bond acceptors; HBD—hydrogen bond donors; RB—rotatable bonds; iLOGP—predicted lipophilicity (logP); LP—Lipinski rule violations; PAINS—pan-assay interference structural alerts; BRENK—structural alerts associated with reactivity or medicinal chemistry liabilities.

; Figure 3 and Figure 4). Most compounds complied with Lipinski’s rule of five, with molecular weights below 500 g/mol and acceptable hydrogen-bond donor (HBD) and acceptor (HBA) counts. Only one compound (3-hydroxylinoleoylcarnitine) exhibited two Lipinski violations, while two additional compounds showed a single violation. Several metabolites demonstrated moderate lipophilicity (iLOGP 2–4), whereas a subset, including tanariflavanone B, lehualide G, and anhydrovitamin A, displayed elevated iLOGP values (>4), suggesting increased hydrophobic character. Rotatable bond counts varied considerably, ranging from 1 to 20, with long-chain derivatives such as 3-hydroxylinoleoylcarnitine and substituted pyran derivatives showing the highest conformational flexibility. Most compounds were free of PAINS alerts, with only 7C-aglycone presenting a PAINS flag. SwissADME analysis identified BRENK structural alerts in several metabolites, associated with potential reactivity or medicinal chemistry liabilities.

Predicted gastrointestinal absorption indicated that 18 of the evaluated compounds were classified as having high oral absorption, whereas only three metabolites were predicted to exhibit low absorption (Figure 5a). Compounds displaying reduced predicted absorption were generally characterized by increased conformational flexibility and extended aliphatic chains, as observed for 3-hydroxylinoleoylcarnitine and certain long-chain pyran derivatives.

Solubility predictions according to the Ali classification demonstrated a heterogeneous distribution across categories. Six compounds were classified as poorly soluble, five as soluble, five as moderately soluble, three as very soluble, and two as highly soluble (Figure 5b,c).

Long-chain pyran derivatives and lehualide G were predominantly associated with poor solubility profiles, consistent with their elevated lipophilicity and rotatable bond counts. In contrast Val-Asn-Pro exhibited more favorable solubility rankings, aligning with their moderate iLOGP values and lower structural complexity (Table 3).

Blood–brain barrier permeability predictions revealed that several lipophilic metabolites, including hydroxyterbinafine, (5E,10E)-5,10-pentadecadien-1-ol, 3-hydroxylinoleoylcarnitine, 7,8-dihydroretinol, 7C-aglycone and 6β-hydroxyferruginol, were predicted to cross the BBB (Figure 5d). Conversely, lehualide G, anhydrovitamin A, merulin B, (rel)-, and the reference ACE inhibitors captopril, enalapril, and lisinopril were predicted as non-permeant. These findings suggest that structural lipophilicity and reduced polarity were major determinants of predicted central distribution.

Efflux transport predictions indicated that only a limited subset of compounds was identified as P-glycoprotein substrates, specifically merulin B (rel), lisinopril, and enalapril, whereas many natural metabolites were classified as non-substrates. This distribution implies that efflux-mediated disposition is more likely to influence certain reference drugs than most of the evaluated metabolites.

Metabolic assessment based on cytochrome P450 inhibition predictions revealed distinct compound-specific interaction patterns (Figure 6). Several natural metabolites, including 7C-aglycone, 4-hydroxy-6-(15-hydroxypentadecyl)-pyran-2-one, 3-hydroxylinoleoylcarnitine, gingerdione, and anhydrovitamin A, were predicted to inhibit one or more CYP isoforms, with CYP2D6 and CYP1A2 being the most frequently affected enzymes. Tanariflavanone B and lehualide G were predicted to inhibit CYP3A4, suggesting potential interactions with the principal drug-metabolizing pathway.

In contrast, Val-Asn-Pro and merulin B (rel) showed no predicted inhibition across all evaluated CYP isoforms. Notably, the reference ACE inhibitors captopril, enalapril, and lisinopril were predicted to be inactive across all CYP endpoints.

Furthermore, these findings indicate that while most metabolites demonstrate favorable predicted oral absorption and variable solubility, the potential for metabolic interactions differs substantially across compounds. These predictions provide a structured pharmacokinetic framework for prioritizing compounds with balanced absorption and limited metabolic liability for further investigation. Therefore, these results indicate variability in predicted drug-likeness properties among the putatively annotated metabolites, with some compounds occupying classical small-molecule drug space and others exhibiting structural complexity that may influence their predicted pharmacokinetic behavior.

3.6. Prediction Toxicological Profile and Acute Oral Toxicity Assessment

The computational toxicity profiling revealed a diverse but classifiable safety landscape among the evaluated metabolites and reference ACE inhibitors (Figure 3 and Figure 4). When integrating endpoint toxicity flags with predicted acute oral toxicity (LD₅₀), compounds could be categorized into three safety tiers.

Compounds in this category exhibited the widest predicted acute safety margins. Merulin B, (rel)- (10,000 mg/kg) and lisinopril (8500 mg/kg) showed the highest LD₅₀ values, indicating lower predicted acute toxicity. (5E,10E)-5,10-pentadecadien-1-ol (4600 mg/kg) and 7,8-dihydroretinol (3389 mg/kg) also demonstrated more favorable predicted toxicity profiles. Notably, 7,8-dihydroretinol showed a completely inactive toxicity endpoint pattern in the heatmap, representing the lowest number of predicted toxicity endpoint alerts among all evaluated compounds. Although (5E,10E)-5,10-pentadecadien-1-ol exhibited isolated neurotoxicity, its high LD₅₀ value suggests a relatively low acute risk.

This group included 3-hydroxylinoleoylcarnitine (3300 mg/kg), enalapril (2973 mg/kg), and gingerdione (2580 mg/kg), along with several metabolites clustered around ~2000 mg/kg.

These compounds displayed limited but specific endpoint activity, most frequently nephrotoxicity and/or immunotoxicity. The presence of focused toxicity signals without widespread multi-organ involvement suggests moderate predicted risk requiring prioritization refinement rather than immediate exclusion.

Compounds with the lowest LD₅₀ estimates included lehualide G (315 mg/kg) and 7C-aglycone (500 mg/kg). These metabolites also displayed endpoint activity in some cases. Within the computational framework, these compounds represent the highest acute toxicity risk among the evaluated panel.

Across all tiers, nephrotoxicity emerged as the most recurrent predicted endpoint, affecting both natural metabolites and the three ACE inhibitors, whereas hepatotoxicity and general cytotoxicity were not broadly distributed. Importantly, the reference ACE inhibitors did not uniformly outperform natural metabolites in safety ranking; certain metabolites, particularly 7,8-dihydroretinol and merulin B, (rel)-, showed computational toxicity estimates comparable to those of the reference ACE inhibitors.

4. Discussion

Marine metabolites remain substantially underexplored, due to the extensive diversity of organisms inhabiting these ecosystems. Among these sources, microalgae, particularly diatoms, represent a group with significant biochemical potential that has yet to be fully investigated [50]. In this context, the marine diatom Phaeodactylum tricornutum has received considerable attention, mainly for its lipid profile and its exploitation for the production of high-value fatty acids and as a nutraceutical [51,52,53].

In the present study, non-targeted metabolomic profiling expanded the chemical repertoire associated with P. tricornutum, revealing a complex distribution of structurally diverse metabolites. Amphipathic and lipophilic compounds are known to exhibit enhanced affinity for membrane-associated proteins and metalloproteins due to favorable hydrophobic and metal-coordination interactions, as seen with angiotensin-converting enzyme [54]. This chemical diversity provided the foundation for subsequent structure-based prioritization and functional prediction analyses.

Molecular docking simulations provide a structure-based framework for estimating the binding orientation and relative affinity of small molecules toward biological targets by scoring functions that approximate Gibbs free energy changes (ΔG) [55,56,57,58,59]. In this context, more negative ΔG values indicate a stronger predicted binding affinity within the scoring function approximation.

In the present study, ligand prioritization was performed according to the binding energy hierarchy, identifying Lehualide G, Val–Asn–Pro, tanariflavanone B, hydroxyterbinafine, and anhydrovitamin A as the top-ranked metabolites. All selected compounds exhibited binding energies ≤ −7.0 kcal·mol⁻¹, a threshold frequently considered indicative of relevant interaction strength for ACE inhibitors in computational screening studies [60,61,62].

Previous reports support this cutoff as biologically meaningful. For instance, Wang et al. identified ACE-interacting peptides with binding energies around −7.4 kcal·mol⁻¹ [63], while Xu et al. reported efficient soy-derived peptides exhibiting docking scores approaching −9.8 kcal·mol⁻¹ [64]. Similarly, Wei et al. described cheese-derived ACE inhibitory peptides with binding energies near −9.5 kcal·mol⁻¹, highlighting the association between increasingly negative ΔG values and stronger predicted inhibitory potential [65]. Within this comparative framework, Lehualide G (−8.429 kcal·mol⁻¹) and Val–Asn–Pro (−8.178 kcal·mol⁻¹) displayed binding affinities that approach those reported for experimentally validated ACE-inhibitory peptides, positioning these metabolites as structurally competitive candidates. Although docking energies represent approximations rather than absolute thermodynamic measurements, the relative ranking observed here supports the hypothesis that selected P. tricornutum metabolites may achieve stable accommodation within the ACE catalytic cavity.

Detailed interaction analysis was performed, focusing on the canonical ACE catalytic architecture, including the S1, S1′, and S2′ subsites and the conserved HEXXH zinc-binding motif, as well as structural elements essential for enzymatic activity and inhibitor recognition [33,66,67,68]. In zinc-dependent metalloproteases such as ACE, ligand engagement with catalytic residues and/or the Zn²⁺ center is frequently associated with inhibitory potential [69]. Among the top-ranked ligands, Lehualide G exhibited extensive engagement within the catalytic region.

The compound established stabilizing contacts with Tyr523 in the S1 pocket and His513 in S2′, while simultaneously interacting with residues of the HEXXH motif (His383, His387, Glu411) and the catalytic Zn²⁺ ion. Notably, the hydrogen bond with Glu411 (1.94 Å) falls within a geometrically favorable range for strong stabilization, suggesting potential interference with the catalytic environment.

Although the Zn²⁺ interaction (4.04 Å) exceeds typical coordination distances, it remains consistent with electrostatic stabilization observed in non-chelating zinc-binding ligands [70]. Collectively, this binding pattern supports stable accommodation within the catalytic cavity rather than definitive enzymatic inhibition.

The tripeptide Val–Asn–Pro displayed a binding mode consistent with that of previously reported peptide-derived ACE inhibitors [71,72]. The ligand formed multiple hydrogen bonds within S1 (Ala354, Tyr523) and S2′ (Gln281, Lys511, Tyr520, His353), including a short hydrogen bond with Glu411 (2.13 Å). This interaction network reflects a polar anchoring mechanism characteristic of peptide inhibitors that rely predominantly on hydrogen bonding and electrostatic complementarity rather than hydrophobic stabilization [73,74].

Tanariflavanone B, a polyhydroxylated flavonoid, demonstrated a multimodal interaction profile. The compound formed strong hydrogen bonds with Tyr523, Glu384, His513, and Asp415 (1.8–2.9 Å) and electrostatic contacts with Zn²⁺ and Glu411. Such short-range polar interactions are consistent with the metal-chelating behavior reported for flavonoids acting on zinc-dependent enzymes [75,76,77,78], supporting a potential competitive binding mode within the catalytic site.

In contrast, hydroxyterbinafine and anhydrovitamin A exhibited predominantly hydrophobic interaction patterns. Hydroxyterbinafine combined polar contacts in S1 and S2′ with an extended hydrophobic network surrounding the zinc-binding region but lacked direct Zn²⁺ coordination, suggesting a steric or substrate-access interference mechanism consistent with non-chelating ACE inhibitors. Anhydrovitamin A interacted mainly through van der Waals contacts (3.4–4.9 Å) involving residues near the catalytic motif, indicating a lipophilic stabilization strategy rather than direct catalytic disruption.

Recent computational studies have similarly reported that ligands with predicted ACE-inhibitory activity frequently engage catalytic residues, such as Tyr523, Glu384, His387, Glu411, and Zn²⁺ within the conserved HEXXH motif. Comparable interaction profiles involving hydrogen bonding, electrostatic contacts, and occupation of S1/S2′ subsites have been described in other docking-based investigations of ACE inhibitors [79,80,81], supporting the structural plausibility of the binding modes observed here.

Previous studies have demonstrated that marine algae and microalgae are relevant sources of both ACE-inhibitory peptides and non-peptidic metabolites [82]. For instance, peptides isolated from Ulva intestinalis, including FGMPLDR and MELVLR, showed ACE-inhibitory activity and molecular interactions involving hydrogen bonds and Zn²⁺-related contacts within the ACE active site [83]. Similarly, peptides from Gracilariopsis lemaneiformis, such as FQIN[M(O)]CILR and TGAPCR, exhibited low IC₅₀ values, docking interactions with ACE active pockets, gastrointestinal stability, and antihypertensive effects in spontaneously hypertensive rats [84]. Marine microalgae have also been reported as sources of ACE-inhibitory peptides, including Val–Glu–Gly–Tyr from Chlorella ellipsoidea and peptide-rich fractions from Nannochloropsis oculata [85,86]. In addition to peptides, non-peptidic marine metabolites have shown ACE-inhibitory potential; Phlorotannins from Ecklonia stolonifera, particularly phlorofucofuroeckol A, dieckol, and eckol, displayed marked ACE inhibition, whereas meroterpenoids from Sargassum macrocarpum inhibited ACE through interactions involving Zn²⁺ and hydrogen bonds [85,87]. In this context, the predicted ACE-binding compatibility observed for the putatively annotated metabolites from P. tricornutum, including the peptide-like candidate Val–Asn–Pro and structurally diverse lipophilic metabolites such as Lehualide G, is consistent with the broader evidence supporting marine-derived molecules as a chemically diverse reservoir of potential ACE-interacting candidates.

The PASS-based activity trends suggest that 3-hydroxylinoleoylcarnitine may represent the most structurally consistent vasoactive candidate within the natural metabolite panel. Its high predicted probabilities for peripheral and general vasodilation, together with a moderate antihypertensive signal, indicate alignment with molecular features commonly associated with modulation of vascular smooth muscle. Long-chain acyl derivatives that are structurally related to carnitine esters have previously been implicated in endothelial signaling and membrane-associated enzymatic regulation, supporting the pharmacological relevance of a vascular-associated activity profile [88,89].

Similarly, gingerdione, 6β-hydroxyferruginol, and the substituted pyran-2-one derivatives displayed coherent Pa > Pi distributions across vascular-related endpoints. Such convergence may reflect shared physicochemical characteristics, particularly an amphipathic balance and moderate lipophilicity, that facilitate interactions with membrane-associated or enzymatic targets involved in vascular tone regulation. Comparable structural motifs have been described in natural multi-target vasoactive compounds capable of modulating nitric oxide pathways, calcium signaling, or endothelial reactivity [90,91].

In contrast, the reference ACE inhibitors exhibited a more restricted PASS pattern dominated by coronary vasodilator predictions. This selectivity aligns with their well-characterized mechanism of action in the renin–angiotensin system, in which inhibition of angiotensin-converting enzyme reduces angiotensin II formation and indirectly promotes vasodilation [92].

The comparatively broader peripheral vasodilator signals observed for several natural metabolites may therefore indicate structural compatibility with multi-target vascular modulation rather than exclusive ACE-directed inhibition.

Although PASS predictions do not provide functional validation, they rely on structure–activity relationship modeling derived from curated experimental datasets. Previous evaluations have demonstrated that Pa > Pi trends frequently correlate with experimentally confirmed bioactivity when supported by complementary in silico or biochemical assays [93,94,95]. Accordingly, the observed probability patterns support the hypothesis that selected P. tricornutum metabolites contain structural determinants compatible with vascular and antihypertensive activity. However, experimental validation will be required to clarify whether these compounds act through direct ACE inhibition, endothelial modulation, ion-channel interaction, or alternative vasoactive mechanisms.

Oral bioavailability depends primarily on intestinal permeability, metabolic stability, and physicochemical balance [96]. Because most approved drugs are administered orally [97,98], early prediction of gastrointestinal absorption remains a critical parameter in compound prioritization.

In this study, most selected metabolites showed high predicted gastrointestinal absorption, whereas only three compounds exhibited low absorption. Solubility predictions under the Ali classification revealed a heterogeneous distribution, ranging from poorly soluble to highly soluble compounds [99]. Long-chain derivatives, including 3-hydroxylinoleoylcarnitine and selected substituted pyran derivatives, consistently showed reduced predicted absorption and lower solubility. This pattern aligns with the well-established inverse relationship between lipophilicity and aqueous solubility, as well as with the influence of molecular flexibility on membrane permeability and absorption efficiency [100,101]. Most compounds met established drug-likeness criteria, including Lipinski’s rule-of-five, PAINS, and BRENK filters [98]. Molecular weights generally remained below 500 g/mol, and hydrogen bond donor (HBD) and acceptor (HBA) counts stayed within accepted limits. Lipophilicity values spanned from low to moderately high. Tanariflavanone B, lehualide G, and anhydrovitamin A exhibited elevated iLOGP values (>4), reflecting increased hydrophobicity. Rotatable bond counts ranged from 1 to 20, with long-chain metabolites displaying greater structural flexibility. Previous analyses have demonstrated that molecular weight, hydrogen-bonding capacity, and rotatable bond count collectively influence oral bioavailability and permeability [102].

Val–Asn–Pro showed a relatively high topological polar surface area. TPSA values exceeding 140 Ų correlate with reduced passive intestinal diffusion [100]. However, its limited hydrogen bond donor and acceptor counts may partially offset this constraint, as hydrogen-bonding capacity and conformational adaptability also modulate permeability [103]. Distribution profiles also differed among compounds. Blood–brain barrier permeability depends on molecular size, lipophilicity, polarity, and transporter interactions [104,105,106].

Although BBB penetration is essential for central nervous system-targeted therapies, peripherally acting antihypertensive agents such as ACE inhibitors do not require central access [107,108]. While most evaluated metabolites had molecular weights below 500 Da, molecular size alone does not determine BBB permeability. Lipophilicity and polar surface area frequently exert a stronger influence on CNS penetration than molecular mass per se [109,110]. Consistently, several lipophilic metabolites, including hydroxyterbinafine, 7,8-dihydroretinol and 6β-hydroxyferruginol, showed predicted BBB permeability, whereas lehualide G, anhydrovitamin A, merulin B (rel)-, and the reference ACE inhibitors showed non-permeant profiles.

Transporter-mediated processes further refined the distribution profile. P-glycoprotein (ABCB1) functions as an ATP-dependent efflux transporter that restricts intestinal absorption, promotes biliary and renal elimination, and limits central nervous system penetration at the BBB [111,112,113,114]. In this dataset, only tanariflavanone B and merulin B (rel) showed predicted P-gp substrate behavior. Previous studies describe flavonoids and structurally related polyphenols as P-gp interactors whose activity depends on hydroxylation patterns and polarity [115,116], supporting the plausibility of these predictions.

Finally, cytochrome P450 inhibition predictions characterized metabolic liability. CYP3A4 and CYP2D6 play central roles in drug metabolism and frequently mediate clinically relevant drug–drug interactions [117,118,119]. CYP3A4 alone metabolizes approximately half of marketed drugs, underscoring its relevance in pharmacokinetic risk assessment. Several natural metabolites, including 7C-aglycone, 3-hydroxylinoleoylcarnitine, gingerdione, and 4-hydroxy-6-(15-hydroxypentadecyl)-pyran-2-one, showed inhibition across one or more isoforms. In contrast, Val–Asn–Pro and merulin B (rel)- showed no predicted inhibition across evaluated isoforms, and captopril, enalapril, and lisinopril similarly showed no CYP inhibition signals.

The in silico toxicity assessment performed in this study aligns with contemporary computational drug discovery pipelines that integrate machine learning-based toxicity prediction tools during early-stage candidate prioritization. ProTox-II and its updated version, ProTox 3.0, have been widely adopted for estimating oral LD50 values, assigning compounds to Globally Harmonized System (GHS) toxicity classes (I–VI), and predicting organ-specific toxicity endpoints, including hepatotoxicity, immunotoxicity, carcinogenicity, mutagenicity, cytotoxicity, and neurotoxicity [49,120]. These platforms are trained on large, curated toxicological datasets and provide probabilistic predictions that support rational filtering prior to experimental validation.

Several recent virtual screening studies targeting cardiovascular, anticancer, and antimicrobial pathways have employed ProTox-based toxicity profiling as a standard ADMET filtering step, combining predicted LD₅₀ values with endpoint activity mapping to stratify compounds into safety tiers [121,122,123,124,125]. In these frameworks, compounds exhibiting predicted LD₅₀ values above ~2000 mg/kg (typically corresponding to toxicity classes IV–VI in the GHS system) are generally considered to possess acceptable acute toxicity margins and are retained for further investigation. The present findings are consistent with this benchmark, as Merulin B, (rel)-, 7,8-dihydroretinol, and (5E,10E)-5,10-pentadecadien-1-ol demonstrated high predicted LD₅₀ values alongside limited or absent multi-endpoint activation, supporting their prioritization within a low-risk tier [123,125].

Importantly, comparative ADMET analyses reported in the literature indicate that clinically approved drugs do not invariably demonstrate superior predicted toxicity profiles when evaluated using standardized in silico and machine learning-based platforms [48,49]. Computational toxicity models are trained on structurally diverse chemical spaces, including approved drugs and natural compounds, which may lead to comparable safety rankings across different molecular classes [126]. The present study supports this observation, as certain diatom-derived metabolites, particularly 7,8-dihydroretinol and Merulin B (rel)-, displayed predicted safety margins comparable to or exceeding those of reference ACE inhibitors.

Despite the robustness of ProTox 3.0 models and their validated performance in acute oral toxicity classification [49,120]. Toxicity predictions remain probabilistic and dependent on training-set representation, chemical similarity domains, and algorithmic generalization limits [126]. In addition, regulatory acute toxicity classifications, such as those defined by the Globally Harmonized System (GHS), are based on experimentally determined LD₅₀ thresholds [127], highlighting that computational estimates should be interpreted as supportive rather than definitive evidence. Therefore, while the integration of LD₅₀ estimation and endpoint profiling provides a rational and literature-supported prioritization strategy, experimental validation through in vitro and in vivo assays remains essential to confirm acute and organ-specific toxicity risk.

Study Limitations

The main limitation of this study is that the proposed antihypertensive candidates were prioritized from putatively annotated metabolites and computational predictions rather than from experimentally confirmed bioactivity. Because authentic standards were not analyzed under identical experimental conditions, the metabolite assignments correspond to MSI level 2 annotations and should be interpreted as putative rather than definitive chemical identifications [128].

This limitation also extends to the pharmacological interpretation of the prioritized candidates. Molecular docking, PASS, SwissADME, and ProTox provided a coherent framework to compare ACE-binding compatibility, predicted vasoactive signals, pharmacokinetic behavior, and toxicity risk.

However, these approaches do not demonstrate enzymatic inhibition, antihypertensive efficacy, or toxicological safety. Thus, the present findings should be considered a prioritization framework for selecting candidates for further validation. Future work should confirm the chemical identity of the selected metabolites using authentic standards and evaluate the methanolic extract and top-ranked compounds through in vitro ACE inhibition assays and subsequently in cell culture to in vivo models of antihypertensive activity.

5. Conclusions

In conclusion, the present study suggests that Phaeodactylum tricornutum is a relevant source of structurally diverse, putatively annotated metabolites with potential antihypertensive properties. In silico analyses prioritized multiple putatively annotated compounds that engage the angiotensin-converting enzyme catalytic site, exhibiting binding orientations compatible with potential interaction within the zinc-dependent catalytic environment and predicted affinities comparable to those of reference inhibitors. Rather than a single dominant candidate, the results reveal a differentiated profile across metabolites. Lehualide G and tanariflavanone B showed strong docking compatibility with the catalytic pocket, 3-hydroxylinoleoylcarnitine exhibited the most consistent predicted vasoactive signal, and 7,8-dihydroretinol and Merulin B (rel)- demonstrated the widest predicted acute safety margins. Val–Asn–Pro displayed a balanced profile combining ACE-binding compatibility with acceptable pharmacokinetic predictions, although its elevated polarity may influence permeability. Cytochrome P450 inhibition patterns and toxicity modeling further highlighted variability in metabolic liability across compounds, underscoring the importance of multi-parameter prioritization. Overall, the convergence of docking compatibility, probabilistic vascular activity, balanced ADME characteristics, and stratified safety predictions supports advancing selected metabolites toward targeted experimental validation to confirm antihypertensive efficacy and safety.