Exploring Hippuric Acid’s Putative Protective Role in Aging through In Silico and In Vitro Screening

Abstract

Hippuric acid (HA) is a diet- and microbiome-derived metabolite consistently associated with healthy aging, yet whether it is a passive biomarker or an active effector remains unresolved. To investigate mechanisms underlying its protective role, in silico screening of large bioactivity data sets was performed and integrated with results from in vitro assays. A multifingerprint similarity approach was employed to identify close HA analogues and their reported targets. HA analogues were identified with reported potencies in the low-micromolar or nanomolar range toward transporters (solute carriers, organic anion transporters, multidrug resistance-associated proteins), epigenetic regulators (histone deacetylases), and enzymes linked to inflammation, proteostasis, or vascular homeostasis (carbonic anhydrases, cyclooxygenase-2, soluble epoxide hydrolase, neprilysin). Additional evidence pointed to proteins involved in detoxification, energy metabolism, and neurodegeneration (kynureninase, microtubule-associated protein tau, 2-amino-3-carboxymuconate-6-semialdehyde decarboxylase), supporting their plausibility as candidate HA targets. By contrast, database screenings for direct antioxidant or anti-inflammatory activity were mostly inconclusive, whereas in vitro assays suggested limited radical-scavenging activity and no modulation of nitric oxide production pathways. Results suggest that HA’s protective associations with aging might involve modulation of specific proteins or transporter-mediated mechanisms, rather than acting solely through target-nonspecific pathways. While these findings remain exploratory and require further confirmation, this integrated in silico and in vitro analysis provides an initial, hypothesis-generating strategy to narrow the spectrum of potential molecular interactions, offering candidate targets to guide future functional in silico, in vitro, and in vivo validation experiments.

1. Introduction

Aging is a progressive biological process characterized by the loss of physiological and cognitive function and an increased vulnerability to geriatric syndromes, which collectively reduce quality of life and increase mortality in older adults. −

Hippuric acid (HA) (CAS 495-69-2) is the glycine conjugate of benzoic acid, formed through microbial catabolism of dietary polyphenols in the gut. − Epidemiological studies indicate that HA levels rise with age in healthy individuals, − yet are reduced in geriatric syndromes, including frailty, , sarcopenia, rheumatic disease, hypomobility, and cognitive impairment. In this context, HA has emerged as a promising marker of age-associated decline.

Although epidemiological and metabolomics studies suggest that HA may reflect diet-microbiome interactions relevant to healthy aging, − mechanistic data remain scarce. Current evidence does not clarify whether HA exerts direct biological activity or merely serves as a biomarker of beneficial polyphenol metabolism. At present, no systematic effort has been made to identify downstream molecular targets, such as receptors, signaling pathways, or enzymes, through which HA might act. −

Because in vivo and in vitro studies of HA’s activity are lacking, nonanimal New Approach Methodologies (NAMs) provide an efficient and cost-effective way to generate mechanistic hypotheses. In silico methods are examples of NAMs and are particularly valuable in early-stage research. Virtual screening enables large-scale interrogation of chemical space, comparing compound structures against large chemical libraries to identify analogues with possible biological relevance.

In silico methods offer substantial advantages over traditional experimentation: they require limited resources and can be conducted rapidly, making them particularly suited when an initial mechanistic hypothesis is missing. , Although they cannot replace empirical validation, these approaches are ideally positioned as first-tier strategies to prioritize plausible molecular mechanisms, reduce experimental burden, and narrow down vast numbers of possibilities, guiding subsequent and more focused in vitro and in vivo investigations.

The present study applies an integrated in silico and in vitro strategy to explore putative mechanisms by which HA may exert protective effects against age-related decline. First, a virtual screening of the ChEMBL database was conducted to identify structural analogues of HA and its known biological targets. This approach is designed to infer potential downstream receptors or enzymes with which HA itself may interact with and that could mechanistically explain the observed associations between HA levels and frailty-related conditions. Additional databases were screened for analogues, namely, the Antioxidant Database (AODB), the Anti-Inflammatory Compounds Database (AICD), and InflamNAT, in order to explore beneficial activities of HA that may not be explained by direct protein interactions, with a particular focus on antioxidant and anti-inflammatory properties that are central to the pathogenesis of geriatric syndromes. Ultimately, we performed in vitro assays measuring DPPH radical scavenging and inhibition of nitric oxide by HA to validate the results from the in silico screening (Figure ).

1.

A graphical workflow for evaluating the potential mechanism of action of hippuric acid. Computational screening of large chemical databases (ChEMBL, InflamNAT, AODB, AICD) was performed using 11 orthogonal molecular fingerprints and custom-tuned similarity thresholds to identify potential macromolecular targets and explore antioxidant and anti-inflammatory activity. In vitro validation (DPPH, Griess/NO) was performed, revealing weak antioxidant and no anti-inflammatory activity. A specific array of potential binding partners for hippuric acid was proposed as a basis for future studies.

It is worth noting that the aim of this work is not the discovery of novel, uncharacterized chemical targets but rather to map out the network of possible targets (both known and unknown) and generate testable hypotheses regarding the bioactivity of HA, providing a rational framework for future studies. This knowledge will also help assess the potential of HA as a possible target for dietary, gut microbiota, or pharmacological interventions aimed at mitigating frailty and geriatric syndromes.

2. Methods

2.1. Data Set Preprocessing

Four publicly available bioactivity repositories were curated to support the in silico identification of putative HA targets and to explore their potential antioxidant and anti-inflammatory activities. In all data sets, compounds were curated at the SMILES level (simplified molecular input line entry system) following a semiautomated procedure.

• ChEMBL version 35 was used to infer potential HA-interacting receptors and enzymes based on the bioactivity profiles of its structural analogues. Only assays conducted on Homo sapiens and single-protein targets with at least five tested compounds were considered. Data were curated to exclude entries flagged as “outside the typical range,” “potential transcription errors,” “potential missing data,” “potential author errors,” “inconclusive,” or “not determined,” resulting in ∼5.9 million entries. Only binding (B) assays were retained, while inconclusive or indeterminate entries were discarded, yielding ∼5.4 million curated records covering 2807 unique human protein targets as listed in Supporting Material (File 1, Table S1). Compounds were labeled as “Active” if the pChEMBL value ≥ 5, or standard_text_value included “Active,” or standard_units = % and standard_value > 50, or activity_comment included “Active,” “Inhibitor,” or “Substrate,” or standard_relation was “<” and standard_value <10,000 nM. Records were labeled as “Not Active” if they did not meet the criteria above, if standard_text_value or activity_comment were annotated as “Not active,” “No binding,” “Inactive,” or “Not inhibitor/substrate,” or standard_relation was “>” with standard_value >10,000 nM. Only compounds with “Active” outcomes were included in the screening pipeline.

• Antioxidant Database (AODB) provides assay results for small molecules tested across diverse antioxidant mechanisms and drawn from two main sources, namely, the PubChem and ChEMBL databases, enabling the detection of HA analogues with radical-scavenging or redox-modulating potential. Only primary assay data for small molecules (56,666 unique compounds) were considered for analogue screening. The data set used for the analysis is available in Supporting Material (File 1, Table S2). To interpret mechanism classes, assays were grouped into seven categories describing mode of action. As the original categorization was not available, categories were replicated by matching specific keywords:

• Hydrogen-Atom Transfer (HAT) assays: ORAC, TRAP, β-carotene bleaching, crocin, linoleic acid oxidation, chemiluminescence, TOSCA.

• Electron-transfer (ET) assays: ABTS, DPPH, FCR, TEAC, FRAP, DMPD, and CUPRAC.

• Lipid peroxidation inhibition: TBARS, FOX, lipid peroxidation

• Enzyme-based antioxidant/pro-oxidant assays: GST, SOD, CAT, GST, LOX, and xanthine oxidase.

• NRF2-ARE pathway assays: Nrf2, ARE, and antioxidant response elements.

• Cell-based bioassays: cell, HepG2, RAW264.7, SH-SY5Y, PC12, and T67.

• Other assays: superoxide, hydroxyl radical, nitric oxide, FTC, metal chelating, xanthine oxidase, phosphomolybdenum, RP, peroxynitrite, hydrogen peroxide, and singlet oxygen.

• InflamNat database is a collection of 1394 natural products with experimental data referring to various cell-based anti-inflammatory assays measuring inhibition of inflammatory mediators (nitric oxide [NO], prostaglandin E2 [PGE2]), and cytokines (IL-1β, IL-6, IL-12, TNFα), alongside cytotoxicity assessments. Experimental results are reported either as half-maximal inhibitory concentration (IC₅₀) values standardized to micromolar (μM) units or as categorical activity labels (e.g., Active, Inactive). In this study, the database was filtered to keep only compounds with reported NO data. A threshold of 50 μM suggested by the authors was applied to classify IC50 values as Active or Inactive for the NO assay. To account for confounding effects related to cell viability, compounds labeled as Active for both cytotoxicity and NO inhibitory activity, and those whose cytotoxicity IC50 was 100-fold higher than IC50 for NO, were discarded. Duplicates were checked at the InChIKey level, leading to a final data set, including 635 active and 552 inactive entries. The finalized data set is available in Supporting Material (File 1, Table S3).

• Anti-Inflammatory Compounds database (AICD) integrates bioassays for 79,781 small molecules across 232 inflammation-related targets (AICD; Wang et al., 2019). As canonical SMILES representations were not provided in the original AICD publication, chemical structures were reconstructed from compound names and subsequently standardized, yielding a final data set of 79,953 bioactivity records corresponding to 54,226 unique compounds. Bioactivity values are reported as K _i, K _d, or EC50 and standardized to nanomolar units, covering 203 unique inflammation-related protein targets (Supporting Material, File 1, Table S4).

2.2. Multifingerprint Similarity Screening

To identify structural analogues of HA, a multifingerprint (MFP) similarity approach was adapted from Alberga et al. Multiple molecular fingerprints (FPs) were computed using the KNIME-CDK and the KNIME-RDKit extensions. Highly correlated FPs (Pearson correlation coefficient ≥ 0.80) were removed, leaving 11 orthogonal FPs for downstream analysis. Pairwise similarities were calculated using the Tanimoto coefficient (TC), and then for each fingerprint, empirical TC distributions were obtained from 100,000 random compound pairs drawn from ChEMBL v35 (Figure ). Thresholds corresponding to the top 5% of the TC distributions were established (T5%) and listed in Supporting Material (File 2, Table S7). The structure of HA, represented as SMILES, was screened against the SMILES of ligands from the four selected databases (see Section ). A ligand was considered a close HA analogue if its TC values for all 11 orthogonal FP exceeded the specific T5% thresholds. Less strict rules (requiring fewer TCs exceeding T5%) were used when the strictest criteria failed to identify any analogue. Reported biological targets in ChEMBL associated with these analogues were retrieved for inspection and were considered candidate targets of HA, whereas binary activities (Active/Not Active) associated with analogues from AODB, AICD, and InflamNAT were used to infer antioxidant and anti-inflammatory activity of HA.

2.

Empirical Tanimoto coefficient distributions resulting from 100,000 random compound pairs drawn from ChEMBL v35.

In addition to inspecting reported biological activities of the closest analogues, the 2807 unique human protein targets represented in the curated ChEMBL data selection were screened and prioritized individually by calculating the average TC between HA and the top-scored analogues. Proteins were included in the list of putative HA targets if the average TC of the top 3, top 5, or top 10 analogues was higher than T5% for at least 9, 8, or 7 fingerprints out of 11, respectively.

2.3. In Vitro Evaluation of Anti-Inflammatory and Antioxidant Activity

2.3.1. In Vitro Antioxidant Activity

The antioxidant capacity of HA was evaluated using the QuantiChrom 2,2-diphenyl-1-picrylhydrazyl (DPPH) Antioxidant Capacity Assay Kit (DAOC-120, BioAssay Systems, Hayward, CA, USA), following the manufacturer’s instructions. Briefly, Trolox standards (0–300 μM) were prepared in 70% methanol and dispensed into 96-well clear plates (100 μL/well) to generate the calibration curve. HA standard solutions were freshly prepared at eight concentrations (0.15, 0.3, 0.625, 1.25, 2.5, 5, 10, 50 μg/mL) in pure water, and added to triplicate wells (100 μL/well) together with matched sample blanks containing HA without DPPH to correct for intrinsic absorbance. Immediately before use, the DPPH working reagent was freshly prepared by reconstituting the lyophilized DPPH reagent with methanol, followed by dilution in assay buffer as recommended, after which 100 μL of working reagent was added to each standard and sample well. Plates were gently mixed and incubated at room temperature, and the decrease in absorbance at 517 nm was recorded after 10 min using a microplate reader (TECAN, Spark, Austria). Antioxidant capacity was expressed as Trolox Equivalent Antioxidant Capacity (TEAC, μM), calculated from the standard curve as ΔOD517 nm of water minus ΔOD517 nm of the HA samples after blank subtraction.

2.3.2. In Vitro Anti-Inflammatory Activity

RAW 264.7 murine macrophage-like cells (ATCC TIB-71) were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 2 mM l-glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin at 37 °C in a humidified atmosphere with 5% CO₂. Cells were seeded at densities of 5 × 10⁵ cells per well in 300 μL of medium in 24-well plates and allowed to adhere overnight. The following day, cells were pretreated with HA at concentrations of 5 μg/mL, 10 μg/mL, and 50 μg/mL, alone or in combination with 5 μM and 10 μM 1400W (n = 4 replicates/condition) for 1 h, followed by stimulation with lipopolysaccharide (LPS) at 100 ng/mL for 24 h. After incubation, 100 μL of cell culture supernatant was collected and mixed with an equal volume of Griess reagent (a 1:1 mixture of 0.1% N-(1-naphthyl)­ethylenediamine dihydrochloride in water and 5% sulfanilamide in 5% phosphoric acid) from the Griess Reagent System Kit (Promega, G2930) in a 96-well plate. The plate was incubated at room temperature for 10–15 min, protected from light, and absorbance was measured at 550 nm using a microplate reader (TECAN, Spark, Austria). Nitrite concentration was determined against a sodium nitrite standard curve (0–100 μM) prepared in fresh medium, following the protocol’s instruction. Experiments were performed in quadruplicate, with data expressed as mean ± standard deviation (SD) of nitrite production relative to LPS-stimulated controls. Cell viability was evaluated by crystal violet staining. Briefly, cells were washed with PBS, stained with 0.5% crystal violet solution, extensively rinsed with tap water followed by distilled water, and air-dried. The bound dye was solubilized with a solution containing 1% SDS and 50% methanol in water. Absorbance was measured at 600 nm using a microplate reader (Glomax, Promega). Cell viability was expressed as the percentage of untreated control cells.

2.3.3. Statistical Analysis

All in vitro statistical analyses were conducted using Prism software (version 10.6.1; GraphPad Software, San Diego, CA, USA). Comparisons among experimental groups were performed using the Kruskal–Wallis nonparametric test. Differences were considered statistically significant at p < 0.05.

3. Results

3.1. ChEMBL Screening

The MFP screening of the ChEMBL data set yielded two complementary lines of evidence that prioritized a short list of putative human targets for HA. The inspection of ChEMBL chemicals characterized by TC higher than T5% across all 11 orthogonal FPs returned seven close analogues, namely CHEMBL461 (HA itself), CHEMBL1681 (iodohippurate), CHEMBL500596 (4-hydroxyhippuric acid), CHEMBL447627 (3-hydroxyhippuric acid), CHEMBL463 (aminohippuric acid), CHEMBL1743121 (betamipron), and CHEMBL586 (salicyluric acid). Reported activities of the seven analogues revealed recurrent associations with epigenetic regulators, solute transporters, and enzymes implicated in inflammation and vascular function, as summarized in Figure and, in more detail, in Supporting Material (File 1, Table S5). Several analogues carry experimentally measured bioactivity against solute carrier (SLC) families. Five of the seven analogues (CHEMBL461, CHEMBL500596, CHEMBL463, CHEMBL1231530, CHEMBL586) showed low-micromolar activity (K _m /IC₅₀ = 6.02–66 μM) and inhibition values of up to ∼95% activity as substrates/inhibitors of SLC22 transporters (mainly members 6 and 8). Similarly, CHEMBL463 and CHEMBL1743121 showed inhibition values in the range of 30–100% for several members of the organic anion transporter (OAT) subfamilies (1B1, 1B3, 2B1, and 3A1).

3.

Predicted human targets of hippuric acid were inferred from the closest analogues. The Sankey diagram shows the seven closest hippuric acid analogues (left) and their reported protein targets (right). Edge thickness reflects activity strength (IC50/K _i/K _d).

Histone deacetylase (HDAC) 6 was a reported target among the identified analogues, with low to moderate inhibition values (19.94 to 36.62%) for CHEMBL1231530 and CHEMBL1681, indicating possible epigenetic mechanisms. Single-molecule entries among the analogues showed activity versus other protein classes, namely 2-amino-3-carboxymuconate-6-semialdehyde decarboxylase (ACMSD), canalicular multispecific organic anion transporter 1 (MRP2), endoplasmic reticulum-associated amyloid beta-peptide-binding protein (ERAB), kynureninase, microtubule-associated protein tau (MAPT), and multidrug resistance-associated protein 1 (MRP1). The broader protein-level screening across the 2,807 curated human ChEMBL protein targets produced a higher-level prioritization that converges on a limited set of targets, as summarized in Figure and, in more detail, in Supporting Material (File 1, Table S6). Specifically, the presence of close analogues with activity toward HDAC (IC50s in the sub- to low-μM range [pIC₅₀ = 5.4–6.4] for isoforms 6 and 8) and MAPT confirmed the findings of the previous analyses, whereas carbonic anhydrases (CA I/II), cyclooxygenase-2 (COX-2), neprilysin, and soluble epoxide hydrolase (EPHX2/sEH) were additionally flagged. Of these, the strongest experimental potency and data availability were seen for CA, where analogue records include IC₅₀s in the nM range (pIC₅₀ = 6.6–7.07).

4.

A summary plot ranks putative targets of hippuric acid reaching critical T5% thresholds by averaging Tanimoto coefficients of the top 3 (blue), top 5 (red), and top 10 (gray) analogues. The x-axis reports the number of fingerprints for which the T5% threshold was reached, whereas the minimum number of fingerprints required is indicated by dashed lines. Starred targets were already identified from the previous analysis of hippuric acid close analogues.

3.2. In Silico Evaluation of HA Antioxidant and Anti-Inflammatory Activity

Five close HA analogues were identified in the AODB with TC higher than T5% for all 11 orthogonal FPs (Supporting Material, File 2, Table S8). Analogues, namely CID 464 (HA itself), 2184, 8614, 71651, and 91637, had average TC values = 1.00–0.764 and were tested in the “qHTS assay for small molecule agonists of the antioxidant response element (ARE) signaling pathway” (AID 743203) and the “Yeast HTS for caloric restriction mimetics that inhibit age-related superoxide” (AID 2690), being consistently classified as inactive. In addition, categorization of AODB assays into mechanistic classes showed that HA analogues generally displayed higher average similarity to inactive compounds than to active ones for all seven assay categories (Figure ).

5.

Distribution of Tanimoto coefficients across active versus inactive compounds in the seven mechanistic classes of AODB antioxidant assays. ET: electron transfer; HAT: hydrogen-atom transfer; ARE: antioxidant response elements.

The inspection of the most frequently reported end point, the DPPH assay, showed the absence of structurally related analogues (TC surpassing the T5% for fewer than five FPs) (Supporting Material, File 2, Table S9). Most similar compounds yielded negative or negligible activity, with inhibitory values ranging from 4.9% to 22% at 1 mM after 30 min. This suggests that HA-like chemotypes are unlikely to be potent direct DPPH radical scavengers, at least by analogy with available AODB chemicals.

In silico screening was additionally conducted against the InflamNAT and AICD databases to explore potential anti-inflammatory activity of HA. Only analogues with low similarity (TC surpassing T5% for three FPs; average TC ≈ 0.46–0.48) were retrieved, with reported IC50 values for NO production in macrophages or microglia in the midmicromolar range (55 μM and <50 μM, respectively) (Supporting Material, File 2 Table S10). Given the low similarity to HA and the borderline activity levels, these analogues cannot be considered reliable proxies for HA bioactivity. Like InflamNAT, no compounds from the AICD database met the strictest similarity criterion (TC surpassing T5% for 11 FPs); however, six analogues displayed moderate similarity (TC surpassing T5% for seven FPs; average TC = 0.57–0.70) to HA. Their reported activities spanned IC₅₀ or K_i values in the 6–20 μM range against inflammation-associated enzymes, including neprilysin (P07943), COX-2 (P15144), and endothelin-converting enzyme-1 (Q9GZT9) (Supporting Material, File 2, Table S11).

3.3. In Vitro Assessment of HA Antioxidant and Anti-Inflammatory Activity

Evaluation of antioxidant activity was performed using the QuantiChrom DPPH Antioxidant Capacity Assay Kit, following the manufacturer’s instructions. HA was tested over a concentration range of 0.15–50 μg/mL (n = 3 replicates per concentration), covering the human plasma concentration. ,, Across this 333-fold concentration interval, the measured Trolox Equivalent Antioxidant Capacity (TEAC) values remained unchanged, with a mean TEAC of 80.28 ± 2.06 μM (CV = 2.56%; Figure A). Linear regression analysis of TEAC values versus log₁₀ (HA concentration) revealed no significant concentration-dependent relationship (R ² = 0.08, p = 0.156), and the Kruskal–Wallis test confirmed the absence of significant differences among the eight concentration groups. The observed TEAC values corresponded to approximately 26.8% of the assay’s upper calibration limit (300 μM) and were located within the lower region of the quantification range (11–300 μM). Collectively, these findings indicate that HA exhibits minimal direct antioxidant activity, in agreement with the in silico screening.

6.

A) Trolox Equivalent Antioxidant Capacity (TEAC) of hippuric acid at 0.15 μg/mL, 0.3 μg/mL, 0.625 μg/mL, 0.125 μg/mL, 2.5 μg/mL, 5 μg/mL, 10 μg/mL, and 50 μg/mL. Absorbance was measured after 10 min of incubation. The kit’s limit of detection was set at 11 μM. B) Nitrite production in LPS-stimulated RAW 264.7 macrophages (5 × 10⁵) treated with hippuric acid at 5 μg/mL, 10 μg/mL, and 50 μg/mL, and with 1400W at 5 μg/mL and 10 μg/mL. *p < 0.05 (Kruskal–Wallis).

In vitro HA anti-inflammatory activity was assessed at three different concentrations (5, 10, and 50 μg/mL). According to the Griess Reagent System kit protocol (Promega), all tested HA concentrations showed no significant inhibition of nitrite production in LPS-stimulated RAW 264.7 cells, compared with the statistically significant reduction in NO generation induced by 1400W at both concentrations (5 and 10 μM) (Figure B).

Importantly, HA treatment did not affect RAW264.7 cell viability at any tested concentration or experimental condition, as assessed by crystal violet staining (data not shown), indicating that the nitrite measurements were not influenced by cytotoxic effects. This outcome aligns with the lack of anti-inflammatory activity of HA in macrophages.

4. Discussion

4.1. MFP Similarity Screening

In the present work, in silico similarity-based screening of large chemical databases was combined with in vitro assays to evaluate the potential target-specific (protein interactions) and target-nonspecific (antioxidant, anti-inflammatory) bioactivities of HA, aiming to formulate a set of hypotheses justifying the potential role of HA as an active effector in healthy aging.

The in silico screening adopted an MFP similarity strategy to identify structural analogues with reported experimental bioactivities from the ChEMBL, AODB, AICD, and InflamNAT databases. Combining multiple orthogonal molecular FPs ensured that diverse structural and chemical features were captured and overcame a recognized limitation of using a single FP, which may emphasize only certain aspects of chemical topology or functional groups and thus deliver biased or incomplete assessments of similarity. In contrast, the MFP framework reduces the risk of overlooking structurally meaningful analogues and enhances confidence in downstream target prioritization. Furthermore, the statistical instability of similarity metrics applied to a small scaffold characterized by a few active bits (such as in the case of HA) is mitigated when multiple independent FPs consistently prioritize the exact same targets.

To confirm the validity of the MFP similarity screening, we cross-validated our target predictions against SwissTargetPrediction (https://www.swisstargetprediction.ch/) and the Polypharmacology Browser (PPB3) (https://ppb3.gdb.tools/) (Supporting Material, File 2 Table S12). The comparative analysis revealed a reassuring concordance between the three methodologies. Specifically, a subset of the targets prioritized by our pipeline, namely SLC22A6, CA I/II, COX-2, HDAC6, EPHX2, and neprilysin, were independently identified among SwissTargetPrediction’s candidate targets. Notably, both SLC22A6 and CA I/II were ranked within the top ten most probable targets by the server. Analogously, PPB3 ranked OATP1B3, OATP1B1, and ACMSD among the top six Homo sapiens single proteinS, with HDAC also listed as a protein family. Furthermore, as an internal positive control, the MFP screening accurately identified SLC22A6 among the top-ranked targets for HA. This direct retrieval strongly aligns with the literature, which characterizes HA as a well-documented substrate for SLC22A6. ,

4.2. Biological Relevance of HA Putative Targets

The close-analogue inspection and the broader protein-level screening of ChEMBL data converge on two candidate HA targets, i.e., HDAC isoforms and MAPT. Additional candidates include transporters of the SLC/OAT systems, CA I/II, EPHX2/sEH, neprilysin, COX-2, and kynureninase (Table ). HDACs (particularly isoforms 6 and 8) appear across both ChEMBL screenings. HDAC modulation is directly linked to various physiological functions whose decline was traditionally associated with aging, such as prevention of cellular senescence, , attenuation of inflammation, modulation of vascular calcification during atherosclerosis, and central neuron protection and regeneration during cognitive decline. , MAPT is a central mediator of Alzheimer’s pathology. Tau normally stabilizes microtubules, but becomes hyperphosphorylated and aggregates in neuropathies, impairing neuronal function. Beyond disease, age-related declines in tau levels were observed with age in cognitively normal individuals, and some isoforms have been proposed as molecular markers of cellular aging. Moreover, tau-kinase interactions were observed to directly modulate lifespan in animal models. The SLC/OAT result is supported by multiple close analogues (including amino- and hydroxyhippuric derivatives and betamipron) with reported substrate or inhibitory activity against SLC22 family members and several SLC/OAT transporters. These findings suggest a plausible mechanism whereby HA (or its structural relatives) could influence the clearance of endogenous substances (e.g., bile salts, bilirubin, steroid and hormone conjugates, gut microbiome-derived toxins) and xenobiotics (e.g., numerous drugs such as statins, anticancer agents, and kinase inhibitors). , Transporter engagement could explain associations between HA levels and frailty by altered systemic exposure to toxic microbial metabolites and drugs that are central to renal, metabolic, and neurovascular aging. , Notably, the broader screening also flagged MRP1 and MRP2, two transporters expressed respectively in lung, testis, placenta, and immune cells (MRP1) and in liver, kidney, and intestinal epithelia (MRP2). Both serve as key detoxification pumps, actively transporting drug conjugates and other xenobiotics that may induce tissue injury.

1. Summary of Putative HA Targets, Their Role in Aging, and Supporting References.

Target	Role in aging	References
SLC/OAT	Regulate clearance of endogenous metabolites and xenobiotics, impacting renal, metabolic, and neurovascular aging.	Alam et al., 2018; Nigam et al., 2015; Li et al., 2024; Flynn et al., 2013; Schäfer et al., 2021; Liu et al., 2025
MRP1, MRP2	Detoxification pump, exports xenobiotics, protecting against tissue injury.	Nies and Keppler, 2007
HDAC (2, 6, 8)	Epigenetic regulators; linked to senescence prevention, inflammation attenuation, vascular calcification modulation, neuronal protection/regeneration.	Warnon et al., 2021; Yao et al., 2012; Chang et al., 2021; Kee et al., 2024; Gomez-Sanchez et al., 2022; Baltan, 2012
MAPT	Mediator of Alzheimer via hyperphosphorylation/aggregation; influence lifespan and decline with age.	Federico et al., 2021; Chatterjee et al., 2023; Veselkina et al., 2023; Singh et al., 2024
CA I/II	Dysregulation implicated in mitochondrial dysfunction and neurodegeneration	Alim et al., 2024; Eldehna et al., 2020; Pollard et al., 2016; Jang et al., 2010; Provensi et al., 2019
Neprilysin	Critical for amyloid-β clearance; reduced activity associated with Alzheimer’s pathology.	Nalivaeva et al., 2020; Ali et al., 2024; Wang et al., 2003
COX-2	Inflammatory mediator; expression increases with age due to ROS, impacting systemic aging.	Kim et al., 2016; Luo et al., 2011
sEH/EPHX2	Inhibition reduces neuroinflammation and improves vascular resilience in aging models.	Nelson et al., 2014; Davis et al., 2022
ERAB	Interaction with amyloid-β leads to oxidative stress and neuronal death.	He et al., 2022
ACMSD	Diverts NAD+ biosynthesis away from toxic quinolinate, reducing inflammation.	Egashira et al., 2007
Kynureninase	Implicated in lifespan regulation, metabolic and neurodegenerative diseases.	Bakker et al., 2024; Castro-Portuguez et al., 2020

CA I/II emerge as additional high-interest candidates. The protein-level prioritization flagged CA because several analogues show inhibitory potency in the nM range in ChEMBL records, yet only modest activity has been reported in the literature for HA and its close derivatives. , If CA modulation proves biologically meaningful, it would align with evidence implicating CA dysregulation in age-related mitochondrial dysfunction and neurodegeneration. − The detection of neprilysin, COX-2, sEH/EPHX2, and ERAB among the prioritized targets warrants further investigation given their recognized roles in inflammation, proteostasis, or vascular homeostasis. Neprilysin is one of the most important metalloproteases involved in amyloid-β clearance, and reduced neprilysin activity has been strongly associated with Alzheimer’s pathology. − COX-2 is an inflammatory enzyme that has been reported to increase with aging in most tissues, in response to the increased production of reactive oxygen species. , Similarly, sEH/EPHX2 inhibition has been consistently linked to reduced neuroinflammation and improved vascular resilience in aging models, , whereas ERAB, a mitochondrial dehydrogenase that interacts with amyloid-β, contributes to oxidative stress and neuronal death when dysregulated. Finally, more speculative candidates such as kynureninase and ACMSD suggest additional mechanistic routes. ACMSD reduces inflammation by diverting NAD+ biosynthesis toward benign metabolites and away from the production of quinolinate, which is associated with the pathogenesis of multiple disorders. Kynureninase is part of the NAD+-generating kynurenine pathway that has been implicated in lifespan regulation across multiple organisms, and its alteration has been associated with several age-associated diseases, including metabolic and neurodegenerative ones. ,

4.3. Target Accessibility

To evaluate whether the interaction with hypothesized targets is biologically plausible under in vivo conditions, the pharmacokinetic profile of HA was characterized based on ADMET predictions of the ADMETLab 3.0 web service (https://admetlab3.scbdd.com/). The predictions reported a fraction unbound in plasma of 21.5%, indicating that a free fraction of the compound was available for tissue distribution. Conversely, limited passive membrane permeability was estimated, as reflected by a Caco-2 permeability coefficient of −5.329 log cm/s and an MDCK permeability of 0.0 cm/s. However, HA and its analogues are reported to be substrates for SLC/OATs, allowing HA to circumvent low passive membrane permeability and access intracellular compartments. ,

Regarding systemic exposure, the literature shows that plasma concentrations of HA in normal conditions typically reside at a low baseline range of 5 to 15 μM in individuals on standard diets. ,, These values undergo profound, acute surges following the consumption of polyphenol-rich foods. For instance, controlled nutritional interventions with wild blueberries, cranberries, coffee, or green and black teas demonstrate that dietary polyphenols are rapidly metabolized by the gut microbiota into HA, driving plasma concentrations upward to peak levels of up to 100 μM and beyond. − Moreover, variations in human gut taxonomy strongly influence the interindividual variability observed in peak plasma micromolar values. These diet-induced systemic exposure data align with the hypothesis that HA can realistically reach levels sufficient to trigger prioritized target engagement.

4.4. Antioxidant and Anti-Inflammatory Activity of HA

The combined in silico and in vitro screening did not provide conclusive evidence that HA exerts its protective effects through direct antioxidant (radical scavenging) or anti-inflammatory (NO pathway inhibition) activity. The AODB screening yielded no evidence for direct radical scavenging, as HA and its closest analogues were consistently inactive in DPPH and other electron transfer assays. Regarding the anti-inflammatory potential, the InflamNAT screening retrieved only distant HA analogues with weak and inconsistent activities. In contrast, the AICD screening identified six moderately similar analogues with reported activities against inflammation-related targets such as neprilysin, COX-2, and endothelin-converting enzyme-1. Although these potencies are modest, they fall within a biologically relevant micromolar range and are consistent with activity levels observed for other prioritized targets in the broader ChEMBL-based screening. This pattern suggests that, if present, HA’s anti-inflammatory effects may derive from direct interaction with specific protein targets rather than from generalized suppression of inflammatory mediators. While the in silico evidence was uncertain and, in some cases, uninformative, the in vitro DPPH radical scavenging and NO production inhibition assays supported classifications of HA as inactive for antioxidant and anti-inflammatory activity. Particularly, HA showed only minimal antioxidant activity (26.8% of the assay’s maximum detection capacity) and did not significantly inhibit NO in LPS-stimulated macrophages. Such results were in line with the modest antioxidant and anti-inflammatory activities reported in the literature for HA , and suggested that, at present, HA’s protective role appears more plausibly explained by indirect mechanisms involving modulation of specific protein targets.

4.5. Study Limitations

Despite screening results and literature evidence supporting the involvement of the identified targets in aging diseases, structural similarity cannot be considered proof of target engagement. Even minor chemical modifications, such as changes in polarity or steric hindrance, can dramatically alter protein binding, transforming an otherwise nonbinding scaffold into a ligand or vice versa. Likewise, the MFP similarity approach, although more robust than single-FP analyses, still reduces chemical space to a mathematical representation that cannot fully capture three-dimensional conformations essential for ligand-protein overlay. Another concern is the statistical instability of Tanimoto similarity when applied to small scaffolds with few FP sets, although this aspect is mitigated when multiple FPs systematically prioritize the same targets. Another concern is the biological context: ChEMBL assay data used for the present analysis are inherently heterogeneous, spanning different assay formats, detection methods, and concentration ranges. This variability complicates direct comparisons across targets and is further increased by the fact that well-studied targets, such as HDACs, CAs, and transporters, are disproportionately represented in ChEMBL, which may inflate their apparent relevance while underrepresenting poorly tested proteins. Finally, with regard to the available assay data for HA itself coming from ChEMBL and the literature, these activity values often come from isolated biochemical assays that may not reflect HA’s in vivo distribution, concentrations, and metabolism, which are strongly influenced by gut microbiota, diet, and interindividual variability.

5. Conclusion

The present study applied an integrated in silico and in vitro approach that combines MFP, similarity-based screening with complementary experimental assays to investigate the potential protective mechanisms of HA in aging-related contexts. The results suggested that HA is unlikely to exert protective effects through direct antioxidant or anti-inflammatory chemistry, as the in silico screening was mostly uninformative, and in vitro models did not support such activities. Instead, convergent data from structural analogues and protein-level prioritization identified endogenous proteins and enzymes as candidate targets for HA, particularly a series of transporters (SLC/OAT), epigenetic modulators (HDAC), and enzymes implicated in inflammation, detoxification, mitochondrial metabolism, and vascular homeostasis.

While the literature supports a correlation between these prioritized targets and aging or geriatric syndromes, these predictive findings should not be interpreted as definitive evidence of HA’s actual molecular targets but rather as an exploratory, hypothesis-generating framework. Focused experimental validation is needed to move from theoretical considerations to confirmed mechanistic hypotheses, such as molecular docking simulations, targeted in vitro binding assays, or functional cell-based experiments.

Nevertheless, this study is valuable as an initial step in a broader exploratory strategy, helping to restrict the pool of prospective targets. By prioritizing a restricted set of potential targets, the present work provides a structural foundation to accelerate experimental validation and guide targeted testing to clarify whether HA functions as an active effector of healthy aging.

All data generated or analyzed during this study are included in this published article and its Supporting Material. The Supporting Material contains the curated input data sets, similarity screening outputs, and additional tables and figures supporting the results reported in the main text.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.6c03636.

• Table S1. List of protein targets from the curated ChEMBL v35 data set used for analogue-based target identification. The table lists 2,807 unique human single-protein targets retained after filtering for Homo sapiens binding assays and excluding flagged or inconclusive records. Table S2. Curated Antioxidant Database (AODB) data set used for similarity-based screening of antioxidant activity. Assays were grouped into seven mechanistic categories (HAT, ET, lipid peroxidation inhibition, enzyme-based assays, NRF2-ARE pathway, cell-based bioassays, and other antioxidant-related assays) based on keyword matching. Table S3. Curated InflamNat data set for anti-inflammatory activity modeling and similarity screening. The data set includes compounds with reported nitric oxide (NO) inhibition data from cell-based assays. IC₅₀ values were classified as Active or Inactive using a 50 μM threshold. Compounds with confounding cytotoxicity profiles were excluded. Table S4. Curated Anti-Inflammatory Compounds Database (AICD) data set used for protein-level screening of inflammation-related targets. Chemical structures were reconstructed from compound names to obtain canonical SMILES and subsequently standardized. The final data set comprises 79,953 bioactivity records corresponding to 54,226 unique compounds, with activity values reported as K _i, K _d, or EC₅₀ (nM) across 203 inflammation-related protein targets. Table S5. List of bioactivities retrieved from ChEMBL for HA close analogues (similarity surpassing the T5% for each of the 11 fingerprints used for comparison). Table S6. List of bioactivities retrieved from ChEMBL for HA top 10 analogues for targets from the protein-level screening (XLSX)

• Table S7: List of FPs and the corresponding TC threshold defining the top 5% of the distribution (T5%) used to identify close analogues. Additional information (type, bit length, features captured, reference) is reported for each fingerprint. Table S8. AODB records associated with HA close analogues surpassing the T5% for each of the 11 FPs used for comparison. Table S9. AODB records associated with HA close analogues tested specifically for DPPH free radical scavenging activity (at 1 mM after 30 min). Records disclosed a low average TC ≈ 0.53–0.46 and none of them surpassed the T5% threshold for more than four FPs. Table S10. InflamNAT records associated with HA close analogues tested for nitric oxide production in macrophages or microglia. Records disclosed a low average TC ≈ 0.58–0.46 and none of them surpassed the T5% threshold for more than three FPs. Table S11. AICD records associated with HA close analogues. Records disclosed moderate average TC ≈ 0.70–0.57 and none of them surpassed the T5% threshold for more than seven FPs. Table S12. Consensus targets identified by both the MFP screening and SwissTargetPrediction or the Polypharmacology Browser (PPB3). The table lists the predicted macromolecular targets of HA retrieved by the MFP strategy that independently achieved matching hits on the SwissTargetPrediction or the PPB3 servers (using the “DNN/Consensus” prediction method), including their respective screening ranks and calculated target probabilities (PDF)

◆.

sofiaghironi.nutrizionista@gmail.com

D.G.: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing– original draft; S.G.: Data curation, Investigation, Writing–original draft; F.M., M.T., F.F., A.P., E.C.: Investigation, Methodology; Writing–review and editing; A.R., E.B.: Writing–review and editing; L.B.: Funding acquisition, Project administration, Supervision, Writing–review and editing.

The authors declare no competing financial interest.