*Article* **Mechanistic Insights into Biological Activities of Polyphenolic Compounds from Rosemary Obtained by Inverse Molecular Docking**

**Samo Lešnik <sup>1</sup> and Urban Bren 1,2,\***


**Abstract:** Rosemary (*Rosmarinus officinalis* L.) represents a medicinal plant known for its various health-promoting properties. Its extracts and essential oils exhibit antioxidative, anti-inflammatory, anticarcinogenic, and antimicrobial activities. The main compounds responsible for these effects are the diterpenes carnosic acid, carnosol, and rosmanol, as well as the phenolic acid ester rosmarinic acid. However, surprisingly little is known about the molecular mechanisms responsible for the pharmacological activities of rosemary and its compounds. To discern these mechanisms, we performed a large-scale inverse molecular docking study to identify their potential protein targets. Listed compounds were separately docked into predicted binding sites of all non-redundant holo proteins from the Protein Data Bank and those with the top scores were further examined. We focused on proteins directly related to human health, including human and mammalian proteins as well as proteins from pathogenic bacteria, viruses, and parasites. The observed interactions of rosemary compounds indeed confirm the beforementioned activities, whereas we also identified their potential for anticoagulant and antiparasitic actions. The obtained results were carefully checked against the existing experimental findings from the scientific literature as well as further validated using both redocking procedures and retrospective metrics.

**Keywords:** rosemary; inverse molecular docking; carnosol; carnosic acid; rosmanol; rosmarinic acid

#### **1. Introduction**

Rosemary (*Rosmarinus officinalis* L.), which belongs to the Lamiaceae family, represents an evergreen, perennial, branched shrub that can grow up to three feet tall. It grows fragrant, needle-like, dark green leaves with curved margins and tiny white, pink, purple, or blue flowers [1,2]. The plant is native to the Mediterranean region and its leaves are used extensively in Mediterranean cuisine, mainly as a spice.

Rosemary has been found to possess several bioactive compounds that exert various pharmacological activities, particularly antioxidative [3], anti-inflammatory [4], antidiabetic [5], and antibacterial [6], effects. Moreover, rosemary extracts exhibit promising anticarcinogenic activities in several in vitro [7–9] as well as in vivo studies [10,11].

Carnosic acid (Figure 1a), carnosol (Figure 1b), rosmanol (Figure 1c), and rosmarinic acid (Figure 1d) are most frequently cited in relation to the beneficial pharmacological activities of compounds found in rosemary [12]. Carnosol, carnosic acid, and rosmanol represent polyphenolic diterpenes with similar structures. They consist of the main abietane scaffold, a fused six-membered tricyclic ring system, with one of these rings being aromatic. Carnosic acid represents the major constituent of rosemary and constitutes up to 4% of the dried leaves [13]. However, it is not very stable and, once isolated, undergoes oxidation leading to the formation of the γ-lactone carnosol, which causes it to lose the

**Citation:** Lešnik, S.; Bren, U. Mechanistic Insights into Biological Activities of Polyphenolic Compounds from Rosemary Obtained by Inverse Molecular Docking. *Foods* **2022**, *11*, 67. https:// doi.org/10.3390/foods11010067

Academic Editor: Guowen Zhang

Received: 3 December 2021 Accepted: 24 December 2021 Published: 28 December 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

acidic properties [14]. Oxidation of carnosic acid can alternatively lead to rosmanol, which differs from carnosol in that it has a free hydroxyl group at the C-7 atom and that the γ-lactone is formed via C-20 and C-6 atoms. The three diterpenes form a very effective oxidation cascade, which is vital for the rosemary's potent antioxidative activity. When carnosic acid is oxidized by free radicals, it forms a quinone derivative. This substance can then undergo isomerization, producing carnosol, or a redox reaction, yielding rosmanol. Thus, carnosic acid, while itself a potent antioxidant, can form two additional substances that also exhibit potent antioxidative activities. This mechanism probably represents the main reason behind the extraordinary antioxidative properties of rosemary [15]. Moreover, these compounds also exhibit antibacterial [16], antiviral [17,18], anti-inflammatory [19,20], antiproliferative [7,8,21–28], and antidepressant [29,30] effects. The study by Romo Vaquero, et al. [31] in rats showed that after oral intake, the glucuronide derivatives of these compounds can be found in plasma as early as 25 min after administration, indicating a good bioavailability. Moreover, carnosic acid was also found in the brain tissue of rats, suggesting that it is able to cross the blood–brain barrier, giving credence to a number of studies in which various positive neuroprotective and cognitive effects were established [32,33].

**Figure 1.** Molecular structures and atom numbering of compounds investigated in the inverse docking study (**a**) molecular structure of carnosic acid, (**b**) carnosol, (**c**) rosmanol, and (**d**) rosmarinic acid.

Rosmarinic acid represents an ester of caffeic acid and 3,4-dihydroxyphenyllactic acid [34]. The structure contains two electroactive catechol moieties that can neutralize free radicals through the electron/proton donor mechanism. Examination of the steps reveals that rosmarinic acid is first oxidized at the caffeic acid moiety of the molecule, while the second step corresponds to the oxidation of the 3,4-dihydroxyphenylic acid moiety. Moreover, the hydroxyl and carboxylic oxygens form a system that exerts good metal chelating properties [35]. Rosmarinic acid can also insert itself into lipid membranes where it effectively inhibits lipid peroxidation [36]. Numerous studies describe that rosmarinic acid exhibits also anti-inflammatory [37,38], antimicrobial [39], anticarcinogenic [7,40,41], and neuroprotective effects [42]. However, unlike the diterpenes in rosemary, the oral bioavailability of rosmarinic acid is poor and amounts to only about 1% in rats [43]. This highlights the need to develop novel delivery systems, such as nanoparticles, to improve the poor pharmacokinetic properties of rosmarinic acid [44,45].

Our aim is to identify potential protein targets of carnosic acid, carnosol, rosmanol, and rosmarinic acid using the inverse docking methodology [46], in which a ligand is docked to a multitude of protein binding sites. The method is typically applied to discover new potential protein targets for small molecule drugs [47] or natural products [48–50] and to explain their mechanisms of action in various diseases. To the best of our knowledge such an investigation has never been performed for the major rosemary compounds.

#### **2. Materials and Methods**

*2.1. Starting Coordinates of Rosemary Compounds*

The initial coordinates of carnosic acid, carnosol, rosmanol, and rosmarinic acid were obtained from the ZINC15 database [51], using ZINC IDs ZINC000003984016, ZINC000003871891, ZINC000031157853, and ZINC0000899870, respectively. Prior to performing inverse molecular docking, all molecules were subjected to a quantum mechanical geometry optimization procedure using the MP2/6-31G\* level of theory/basis set combination. This optimization was performed in Gaussian 16 [52].

#### *2.2. In Silico Determination of ADME Properties*

In silico determined ADME/Tox profiles provide a useful tool for predicting the pharmacological and toxicological properties of investigated molecules [53]. To provide a more detailed prediction of the pharmacokinetic properties of carnosic acid, carnosol, rosmanol, and rosmarinic acid, which would complement the known experimental data, we implemented the SwissADME web server [54]. SwissADME represents a freely available tool that enables robust predictions of absorption, distribution, metabolism, and extraction, based on the two-dimensional data of the molecule. In addition, it yields predictions on drug-likeness based on well-established metrics.

All compounds were inputted on the SwissADME webpage (http://www.swissadme. ch/ date accessed: 20 December 2021) using the Simplified Molecular-Input Line-Entry System (SMILES) strings.

#### *2.3. Inverse Molecular Docking*

Our goal was to gain mechanistic insight into the potential mechanism of pharmacological actions of the investigated rosemary compounds using CANDOCK (Chemical Atomic Network based Docking) [55] inverse molecular docking on more than 65,000 protein structures potentially associated with human pathologies. Protein binding sites for small molecules were obtained from the ProBiS-Dock Database [56]. The main advantage of defining binding sites in this way is that multiple spherical centroids are defined in advance to describe a very accurate 3D shape that can be used in conjunction with the CANDOCK algorithm. Moreover, binding sites at the interface of multiple protein chains are also considered for docking.

For docking, the CANDOCK algorithm applies a hierarchical approach to reconstruct small molecules from the atomic lattice using graph theory, while applying a generalized statistical potential function for scoring. The docking scores represent approximations of the relative binding free energies and are expressed in arbitrary units. Specifically, CANDOCK finds the best-docked poses of small-molecule fragments and applies a fastmaximum-clique algorithm [57] to link them together. In the molecular reconstruction, the algorithm uses iterative dynamics for better placement of the ligand in the binding pocket. After the initial docking and reconnection procedure is completed, a minimization procedure based on the Chemistry at Harvard Macromolecular Mechanics (CHARMM) force field [58] is performed to model the induced fit of the ligand binding to the protein binding site.

#### *2.4. Method Validation*

To retrospectively validate our inverse molecular docking procedure, we applied receiver operating characteristic curves (ROC) [59], enrichment curves [60], and predictiveness curves (PC) [61]. Briefly, the ROC metric plot shows a correlation between the true-positive fraction (TPF) on the *y*-axis and the false-positive fraction (FPF) on the *x*-axis. In our case, the TPF represents experimentally confirmed protein targets of rosmarinic acid from the ChEMBL database [62] with the corresponding PDB entries, while the FPF represents all other protein targets from the ProBiS-Dock database. We did not perform an analogous validation for diterpenes as only a small number of confirmed targets is available for them. The area under the ROC curve (ROC AUC) represents a simple measure

to evaluate the overall performance of the inverse molecular docking method. The larger the ROC AUC, the more effective is the method at discriminating true from false targets. The enrichment curve represents the early quantification of target proteins from the TPF. Moreover, PC also provides the early detection quantification of target proteins from the TPF, but in addition, it can be used to define the threshold for potential targets from the inverse molecular docking to be tested experimentally. Contrary to ROC, PC can describe the dispersion of the inverse docking scores well. To quantify the early detection, we applied the enrichment factor of 1% of the compounds tested (EF) [63], the Boltzmann-enhanced discrimination of ROC (BEDROC) [59], and the robust initial enrichment (RIE) [63] measures as well. Using PC, the standardized total gain (TG) [61] was also determined, which summarizes the contribution of the inverse molecular docking scores in explaining the probability of targets over the entire protein dataset. To calculate all of the listed measures, the Screening Explorer web server [64] was implemented.

#### **3. Results**

#### *3.1. Inverse Molecular Docking of Diterpenes*

Because of their similar structure and good agreement, the docking results for carnosic acid, carnosol, and rosmanol were combined and analyzed together: the diterpene ligand with the best score for the individual protein was considered. The 0.05% (3.5σ) top scoring proteins from the entire docked database were selected (Figure 2) and among them, those with implications for human health were chosen. Human and mammalian proteins as well as proteins from pathogenic bacteria, viruses, and parasites were considered. Moreover, mammalian proteins were considered in order to increase the protein space available for docking, where we assumed that within the class of mammals, analogous proteins and their binding sites are similar enough so that our findings from non-human mammals are transferable to human proteins.

**Figure 2.** Normal distribution fit of the inverse docking scores. (**a**) Combined distribution of docking scores obtained by inversely docking the rosemary diterpenes carnosic acid, carnosol, and rosmanol to the whole ProBiS-Dock Database. (**b**) Distribution of docking scores for rosmarinic acid. A cut-off criterion of 3.5 σ was used to select the most promising protein–ligand complexes to be further investigated in more detail.

In Table 1, we present the highest-scoring protein–ligand complexes based on the cut-off criterion of 3.5 σ. Moreover, where data were available, we redocked ligands/drugs that are known to bind to the presented targets using an analogous procedure as the one applied for inverse docking. These results, presented in Table S1, show that in all cases except for K-Ras G12C and enhanced intracellular survival protein, the docking scores of the known ligand/drugs are worse than the ones of the rosemary diterpenes. This indicates an already strong binding affinity of the rosemary compounds, although they have not yet been rationally optimized for these specific protein targets.

**Table 1.** Best scoring human, mammalian, and pathogen protein targets of rosemary diterpenes carnosic acid, carnosol, and rosmanol. Docking scores independent of the organism or type of protein are collected in Supplementary Materials in Table S2.


**Rank PDB ID with Chain Ligand Predicted Ligand Docking Score (arb. Units) Protein Name Organism Protein Function and Disease Correlation Reported Experimental Correlation of Protein and Ligand** 14 5ilqC Carnosic acid −63.0 Aspartate carbamoyltransferase *Plasmodium falciparum* Enzyme involved in pyrimidine biosynthesis, crucial for *Plasmodium falciparum* (causative agent of malaria) survival and replication. 15 4iv5A Carnosic acid −62.8 Aspartate carbamoyltransferase *Trypanosoma cruzi* Enzyme involved in pyrimidine biosynthesis, crucial for *Trypanosoma cruzi* (causative agent of Chagas disease) survival and replication

**Table 1.** *Cont.*

#### 3.1.1. K-Ras

K-Ras is a GTPase responsible for relaying signals from outside the cell to the nucleus. It represents a part of the rat sarcoma/mitogen-activated protein kinase (RAS/MAPK) pathway, and K-Ras signaling leads to cell growth, proliferation, and differentiation. K-Ras is of utmost clinical importance as it represents the most frequently mutated oncogene in pancreatic, colon, and lung cancers [72]. Numerous attempts have been made to develop compounds that inhibit the function of K-Ras, but with limited success only [73]. The non-druggability of K-Ras is mainly due to the lack of a well-defined binding pocket, as well as the high affinity for guanosine triphosphate (GTP), with which alternative drug molecules have difficulty competing. Nevertheless, progress has been made in recent years in modulating K-Ras with small-molecule ligands. Fell, et al. [74] developed a potent inhibitor of the oncogenic K-Ras G12C mutant that induces the formation of a new binding pocket near the nucleotide (GTP) binding site (Figure 3a). Binding to this new pocket results in signal inhibition by arresting the enzyme in its inactive state. Interestingly, this induced binding pocket was ranked most favorable of all the protein binding sites tested by our method for carnosic acid (Table 1). Carnosic acid docks at this induced binding site where it forms two hydrogen bonds with Thr58 side chain and two hydrogen bonds to the backbone atoms of Ala59 and Gly60 (Figure 3b, Table S4). A strong salt bridge with a distance of 4.1 Å is additionally created between the carboxylate of carnosic acid and Arg68. Finally, the relatively large hydrophobic ring system of carnosic acid forms hydrophobic interactions with Glu62, Tyr96, and Gln99. Although none of the diterpenes have been previously reported to bind directly to K-Ras, rosemary extracts have indeed been shown to lead to the down-regulation of K-Ras expression in colon cancer cells [75]. This suggests an interesting potential of carnosic acid for a two-pronged attack on the protein by down-regulating its expression and by inhibiting it directly.

**Figure 3.** (**a**) K-Ras protein structure highlighting the GTP- and induced-binding site. (**b**) The induced binding site of the K-Ras protein (blue) with docked carnosic acid (carbon atoms depicted in grey). Orange dotted lines represent salt–bridge interactions, and blue dotted lines H-bonding interactions. Amino acid residues forming hydrophobic interactions are denoted with yellow sticks.

#### 3.1.2. Glucosamine/Fructose-6-Phosphate Aminotransferase

In humans, infection with pathogenic strains of *Escherichia coli* leads to various diseases such as gastroenteritis, septic shock, and urinary tract infections. In addition, some strains have been linked to colon cancer because they can synthesize substances that damage DNA [76]. While most *Escherichia coli* infections can be treated with existing antibiotics, such as fluoroquinolones, the proliferation of multidrug-resistant strains produces the need to identify new compounds with antimicrobial activity. Although specific binding of rosemary diterpenes to glucosamine/fructose-6-phosphate aminotransferase (GlmS) is not reported in the scientific literature, a number of studies shows that rosemary compounds indeed exhibit activity against *Escherichia coli* [66–68]. Since no mechanism of this inhibition has yet been reported, we speculate that carnosic acid may bind to GlmS, which catalyzes the first step in hexosamine metabolism by converting fructose-6P to glucosamine-6P using glutamine as a nitrogen source [77], yielding N-acetylglucosamine an essential building block of bacterial cell walls. Therefore, targeting this enzyme could lead to the inhibition of bacterial growth [78]. Predicted interactions between carnosic acid and GlmS are presented in Table S5.

#### 3.1.3. Pyruvate Kinase 2–Muscle Isoform

Cancer cells often rely on glycolysis to meet their high energy demands, whereas normal cells derive most of their energy from oxidative phosphorylation [79]. This difference in cell metabolism can be, therefore, exploited to target cancer cells. The muscle isoform of pyruvate kinase 2 (PKM2) is universally expressed in cancer cells and catalyzes the final step of glycolysis by transferring a phosphate group from phosphoenolpyruvate (PEP) to adenosine diphosphate (ADP), resulting in one molecule of pyruvate and one molecule of adenosine triphosphate (ATP). On the other hand, the remaining isozymes of pyruvate kinase are expressed in most normal tissues, so targeting PKM2 represents a viable way to selectively inhibit glucose metabolism in cancer cells [80]. Carnosic acid binds at the site where variations in two amino acid residues are present compared to PKM1, namely Ile389Met and Gln393Lys (Table S6). These variations result in a significant decrease in docking score as the best PKM1 isoform scores −65.1 A.U compared to −68.1 for the M2 isoform (Table 1), which may indicate that carnosic acid is indeed selective towards PKM2.

#### 3.1.4. Hemagglutinin HA1

Influenza virus hemagglutinin (HA) represents a surface glycoprotein that is critical for viral infectivity. It has multifunctional activity, allowing entry of the virus by binding to sialic acid at the surface of host cells, while also being responsible for the fusion of the viral envelope to the endosomal membrane [81]. Due to its importance, this protein forms a key target for neutralizing antibodies [82]. However, it is also possible to target it with small molecules such as arbidol [83]. Carnosic acid docks to a cavity in the HA trimer stem at the interface between the three protomers. This binding site is separate from the conserved epitope targeted by the neutralizing antibodies. The drug arbidol is known to stabilize the conformation of HA, thereby preventing the large conformational changes required for membrane fusion. This could potentially also be the case with carnosic acid, as it forms three hydrogen bonds, one with each protomer, and could thus act as a so-called molecular glue that binds the protomers together, making them nonfunctional (Figure 4, Table S7).

**Figure 4.** Carnosic acid glues together chains A, C and D of the HA glycoprotein. Carbons of carnosic acid are displayed as teal sticks, chain A in orange, chain B in green, chain C in sky blue and chain D in dark blue pipes and planks. Amino acids forming hydrogen bonds (denoted with red dotted lines) are displayed as sticks of matching colors.

#### 3.1.5. HIV-1 and HIV-2 Protease

Human immunodeficiency viruses (HIV) protease is a retroviral aspartyl protease involved in the hydrolysis of several peptide bonds, which is essential for the life cycle and replication of HIV [84]. Small molecule inhibitors of HIV protease play a critical role in the effective treatment of acquired immunodeficiency syndrome AIDS, as they represent part of the highly active antiretroviral therapy (HAART). While HIV-1, carrier of the HIV-1 protease isoform, forms the most common subtype worldwide, HIV-2 remains mainly confined to West Africa and is also spreading in India [85,86]. However, the treatment of HIV-2 is more difficult than that of HIV-1, as most antiviral drugs have been developed for the HIV-1 isoform. HIV-2 proteases have also been found more resistant to small-molecule inhibition [87]. Moreover, dual infection with both isoforms is possible as well [88]. Consequently, novel inhibitors for both HIV proteases would be of great benefit. It has been shown that carnosic acid exhibits potent inhibition of the HIV-1 protease isoenzyme with an IC90 = 0.08 μg/mL [69]. Inhibition has not yet been experimentally demonstrated for the HIV-2 isoform; however, our studies suggest that carnosic acid is also capable of inhibiting this isoform. This finding can also be corroborated by the fact that the binding sites of both isoforms are very similar, with a sequence identity close to 70% and the ProBiS Z-score of 3.76 [58]. ProBiS Z-score measures the statistical and structural

significance of local binding site similarity. Binding site alignments leading to ProBiS Z-Scores higher than 2 are considered to be very similar. In addition, all the equivalent binding site amino acid residues are of the same charge and polarity type. Overall, carnosic acid could prove to be a valuable starting point for the development of antivirals that would be effective against both strains of HIV.

Interestingly, the inverse docking results also suggest a high binding ability of carnosic acid to the HIV-related feline immunodeficiency virus (FIV) protease, which causes an AIDS-like syndrome in cats. HIV-2 and FIV proteases possess a binding site similarity of 1.90, expressed by the ProBiS Z-score, and a general sequence similarity of 26%. The relatively different binding sites result in different binding positions of carnosic acid in HIV-2 and FIV proteases. In HIV-2, the ligand is positioned deeper in the major binding site, which is located between the two protomers (Figure 5, Tables S8, S13 and S15).

**Figure 5.** Comparison of HIV-2 (grey cartoon) and FIV (purple cartoon) protease. Carnosic acid in HIV-2 binds deep into the protease binding site and forms a hydrogen bond with the backbone nitrogen of Ile50 (red dotted line). On the other hand, carnosic acid is located closer to the protease surface in FIV and forms a hydrogen bond H-bond with the backbone nitrogen of Asp29 (red dotted line).

#### 3.1.6. Enhanced Intra-Cellular Survival Protein

Tuberculosis represents the leading cause of infectious death worldwide, primarily due to the emergence of multidrug-resistant tuberculosis and due to extensively drug-resistant strains of *Mycobacterium tuberculosis* [89]. Up-regulation of the enhanced intra-cellular survival (Eis) protein was found to be the sole cause of resistance to the aminoglycoside of last resort-kanamycin in approximately one-third of *Mycobacterium tuberculosis* isolates. Specifically, Eis represents an acetyltransferase responsible for *Mycobacterium tuberculosis* resistance to multiple aminoglycoside drugs. A distinctive property of Eis is that it acetylates the aminoglycoside drugs at multiple amine functional groups, preventing them from binding to their target, the ribosome. The simultaneous use of Eis inhibitors with anti-tuberculosis drugs may therefore provide a way to combat this resistance by restoring aminoglycoside drug activity [90]. Carnosic acid docks to the aminoglycoside binding pocket formed by the N-terminal domain to which also tobramycin binds, thereby suggesting the possibility of competitive inhibition of Eis by carnosic acid (Table S9) [91].

#### 3.1.7. Peroxisome Proliferator-Activated Receptor δ

The peroxisome proliferator-activated receptor (PPARδ) functions as a sensor for dietary and endogenous fats [92]. It regulates the transcription of genes associated with lipid and glucose metabolism. Specifically, it controls lipid degradation, transport, and storage, while also being associated with insulin secretion and resistance. PPARδ agonists have been shown beneficial in models of metabolic disorders in primates and may thus possess therapeutic potential in hyperlipidemia, atherosclerosis, obesity, and diabetes [93,94]. PPARδ is also associated with cancer by promoting chronic inflammation through increasing cyclooxygenase-2 (COX-2) expression and prostaglandin E2 production, leading to an increase in proinflammatory cytokine concentrations. Moreover, the ability of PPARδ to promote the use of fatty acids as the energy source may enhance cell survival and proliferation under harsh metabolic conditions often found in tumors. Therefore, PPARδ agonists may be useful in treating metabolic disorders, while antagonists may reduce inflammation-related disorders and slow down cancer progression.

Whereas there are no experimental data that carnosic acid, carnosol, or rosmanol bind to PPARδ, it is known that both carnosol and carnosic acid represents agonists of the PPARγ isoform with half maximal effective concentration (EC50)values of 41 and 20 μM, respectively [95]. Carnosic acid docks to PPARδ in the same Ω-pocket where serotonin binds to PPARγ which also acts as agonists at PPARδ. The binding site possesses 62% amino acid sequence identity and a ProBiS Z-score of 3.36. From the superposition of PPARδ (with docked carnosic acid) and PPARγ (with serotonin) Ω-pockets, we observe that PPARγ produce steric clashes with carnosic acid (Figure 6, Table S10). However, due to experimental evidence, that carnosic acid indeed binds to PPARγ, we can predict that induced fitting effects play an important role. Because PPARδ possesses a smaller threonine in this place and because the overall binding site is similar, we can hypothesize that carnosic acid could bind even stronger to the PPARδ isotype as preliminary induced fitting would not be required.

**Figure 6.** The Ω-pocket superimposition between PPARδ (orange cartoon and sticks) and PPARγ (blue cartoon and sticks). The first amino acid residue numbering corresponds to PPARδ, and the second to PPARγ. Serotonin is displayed using blue balls and sticks and the docked carnosic acid using orange balls-and-sticks. We emphasize the difference in amino acid residues Thr252 versus Arg288. Compared to Thr252 in PPARδ, the large Arg288 in PPARγ would lead to stearic clashes with carnosic acid.

#### 3.1.8. Glycogen Phosphorylase

Glycogen phosphorylase (GP) is an enzyme that cleaves the non-reducing ends in the chain of glycogen to produce glucose-1-phosphate monomers which can be further converted to free glucose [96]. Because glycogen is an important source of blood glucose, GP represents a promising target for the treatment of type II diabetes, and its inhibitors have been shown effective in controlling blood glucose concentrations in animal studies [97]. GP can exist in two different forms that bind different regulatory molecules: the active phosphorylated (on Ser14) GPa and the non-phosphorylated GPb form. In addition, GP has been reported to bind compounds at four different binding sites, identified as: (a) the catalytic, (b) the allosteric (indole), (c) the novel allosteric, and (d) the inhibitory site (caffeine) [98]. In our study, carnosic acid docked to the catalytic site (a) of the GPb form, specifically to the α-D-glucose binding site, therefore, it might act as a competitive inhibitor with respect to glucose-1-phosphate (Figure 7) [96,99]. Glucose-1-phosphate forms hydrogen bonds with Glu672, Asn284, Ser674, His337, and Asn484, while the docked pose of carnosic acid binds to the cofactor pyridoxal phosphate with two hydrogen bonds, but also forming hydrogen bonds with Lys574 and Thr676 (Table S11).

**Figure 7.** Glucose-6-phosphate (G6P, light blue sticks) overlapped with carnosic acid (brown sticks) in the catalytic binding site of glycogen phosphatase. Important amino acid residues are shown in grey sticks. Hydrogen bonds formed by carnosic acid are presented with red dotted lines, while the hydrogen bonds formed by glucose-6-phosphate are shown with blue dotted lines. The cofactor pyridoxal phosphate (PLP) is presented in orange sticks.

#### 3.1.9. Tubulin

Tubulins represent protein monomers of microtubules, which form an essential component of the eukaryotic cytoskeleton [100,101]. They are involved in cell division as microtubules form mitotic spindles that are used by the cell to pull the chromosomes apart. Microtubules are produced by the polymerization of dimers of α- and β-tubulin that join together to form long hollow tubes called microtubules. Microtubule targeting agents such as chemotherapeutics vinblastine, colchicine, and paclitaxel bind to tubulin and disrupt microtubule dynamics, leading to a loss of function and to subsequent cell arrest or apoptosis. They can be classified into subgroups based on their binding site within the tubulin dimer: (a) the paclitaxel site at the β-tubulin in the microtubule lumen; (b) the vinblastine site at the interdimeric interface of two heterodimers; and (c) the colchicine site at the β-tubulin at the intra-subunit interface of a heterodimer. In our study, carnosic acid docked to the colchicine-binding site (c) (Table S12) and could therefore, like colchicine, potentially lead to microtubule depolymerization.

#### 3.1.10. Phosholipase A2

Phospholipases A2 (PLA2) represent enzymes that catalyze the hydrolysis of glycerophospholipids at the sn-2 position, releasing free fatty acids, including arachidonic acid. The action of PLA2 forms a crucial upstream step that increases free arachidonic acid levels

and triggers the storm of eicosanoids, especially after inflammatory cell activation. Due to their involvement in the inflammatory response, PLA2 are thought to be associated with various diseases such as arthritis [102], cancer [103], coronary heart disease [104], and neurological disorders such as Alzheimer's disease and multiple sclerosis [105]. In our study, carnosol docks between the two subunits of the dimer and forms a large hydrophobic and desolvated surface that is buried. Most of the carnosol molecule is located within the B subunit. (Figure 8, Table S14). Binding to identical active site as the alkyl portion of the tetrahedral mimic inhibitor MJ33.

**Figure 8.** Carnosol (carbons denoted with green sticks) docks between subunits A (blue surface) and B (orange surface) of the phospholipase A2.

#### 3.1.11. Vascular Endothelial Growth Factor Receptor 2

Vascular endothelial growth factor receptors (VEGFR) represent tyrosine kinase receptors for vascular endothelial growth factor (VEGF), a signaling protein critical in angiogenesis [106]. Because solid cancer tumors require an adequate blood supply to grow and metastasize, the inhibition of VEGFR signaling with small molecule drugs such as sorafenib or pazopanib is used as a well-established treatment in various cancers, since tumors cannot grow more than 2 mm without angiogenesis. VEGFR-2 plays an important role in cell migration and proliferation-two crucial steps of angiogenesis. Carnosic acid docks to the binding site representative of type II kinase inhibitors. In general, type II inhibitors, such as sorafenib and lenvatinib, are often more specific than those targeting only the ATP binding site [107,108]. They represent a class of compounds that capture kinases in an inactive form and occupy both the adenine region (of ATP) as well as a hydrophobic pocket adjacent to the ATP binding site [109]. However, due to the small size of carnosic acid, only the hydrophobic binding site is actually occupied (Figure 9, Table S16). This is consistent with the experimental finding that carnosic acid or carnosol actually do not possess a measurable inhibitory activity against VEGFR2 [109]. However, given the strong interaction measured between carnosic acid and VEGFR2 applied in the inverse docking method, carnosic acid could potentially serve as a base compound to which a specific ring system region would be added to also target the adenine binding site.

**Figure 9.** VEGFR2 binding sites. The VEGFR2 enzyme is presented with pink surface; the amino acid residues of the adenosine binding site are shown in pink sticks and of the lipophilic allosteric binding site in brown sticks. The docked carnosic acid located in the allosteric site displayed in brown balls and sticks and the ATP molecule in pink balls and sticks. The typical type II inhibitor imatimib, binding to both sites concurrently is depicted in yellow balls-and-sticks.

#### 3.1.12. Aspartate Carbamoyltransferase

*Plasmodium falciparum* and *Trypanosoma cruzi* represent parasites that cause malaria and Chagas disease, respectively [110]. Aspartate carbamoyltransferase is an enzyme involved in pyrimidine biosynthesis that catalyzes the formation of phosphate and Ncarbamoyl L-aspartate from carbamoyl phosphate and L-aspartate. Reproduction of both parasites requires a sufficient supply of purines, as they form the building blocks of nucleic acid molecules. Recent studies in *Plasmodium falciparum* have shown that aspartate carbamoyltransferase represents a suitable drug target, as its inhibition leads to a reduction in parasite growth [111]. Carnosic acid docks in the aspartate carbamoyltransferase of both *Plasmodium falciparum* and *Trypanosoma cruzi* at the interface between the protomers in the carbamoyl phosphate domain, where the carbamoyl phosphate substrate binds (Tables S17 and S18) [112].

#### *3.2. Inverse Docking of Rosmarinic Acid*

Table 2 lists the top-scoring protein–ligand complexes based on the cut-off criterion of 3.5 σ. We focus only on protein targets related to human health, i.e., we present only proteins from humans and mammals, as well as proteins from pathogenic microorganisms. As before, where data were available, we redocked ligands/drugs known to bind to the presented protein targets using a procedure analogous to inverse docking. These results, presented in Table S1, show that the docking scores of known ligand/drugs are in all cases worse than those of rosemarinic acid. Again, this may indicate an already strong binding affinity of rosmarinic acid, although it has not yet been rationally optimized for these specific protein targets.

**Table 2.** Best scoring mammalian, human, and pathogen protein targets of rosmarinic acid. Docking scores independent of the organism or type of protein are collected in Supplementary Materials in Table S3.


#### 3.2.1. Coagulation Factor X

Factor X represents an enzyme involved in the coagulation cascade that, when activated by the hydrolysis of factor Xa, claves prothrombin to the active thrombin, which in turn converts soluble fibrinogen to insoluble fibrin strands [115]. The role of factor X is particularly important because it is the first enzyme where the intrinsic and extrinsic coagulation pathways converge. Drug manipulation of the coagulation cascade is extremely important in modern medicine, since reducing excessive coagulation is critical for preventing diseases such as myocardial infarction and ischemic stroke, which belong among the leading causes of death and disability in the Western world [116,117]. Oral inhibitors of factor X, such as rivaroxaban, are already successfully used in clinical practice [118]. Rosmarinic acid docks in an analogous manner to a number of sulfonamide

factor X inhibitors (Table S19) [119]. Its caffeic acid ring binds to the S1 pocket, while its 3,4-dihydroxyphenyllactic acid moiety binds to the S4 pocket (Figure 10).

**Figure 10.** Factor Xa binding site. Factor Xa is shown in blue ribbons and surface, its important amino acid residues in blue sticks. Rosmarinic acid (green carbons) is docked in the same binding site as the one occupied by a known inhibitor (orange sticks) with a PDB ID: D01. The caffeic acid part of rosmarinic acid docks to the S1 pocket, and the 3,4-dihydroxyphenyllactic acid moiety to the S4 pocket.

#### 3.2.2. Phospholipase A2

Similar to the case of carnosol, our inverse docking algorithm also detected a strong binding to the enzyme phospholipase A2. This is consistent with existing experimental evidence, as the PDB contains a snake toxin phospholipase A2 homolog (PDB ID: 3QNL) bound with rosmarinic acid, and this complex was also applied later on in our study to validate the inverse docking algorithm by redocking. Compared to the main active site, its binding site is located in a different region between the dimer site where the MJ33 inhibitor was reported to bind and where carnosol was docked in this study (Table S20). We have here an interesting case where two rosemary compounds potentially inhibit the same enzyme.

#### 3.2.3. Matrix Metalloproteinase-3

Matrix metalloproteinase-3 (MMP-3) represents a zinc-dependent endopeptidase Matrix metalloproteinase-3 (MMP-3) represents a zinc-dependent endopeptidase involved in extracellular matrix remodeling [120]. It is, therefore, required for physiological processes such as embryonic development and reproduction and is also involved in various pathological processes. Moreover, MMP-3 can also activate other metalloproteinases, enter cell nuclei, and control gene expression. Excessive activation of MMPs can lead to excessive degradation of the extracellular matrix and to numerous pathological conditions such as arthritis, multiple sclerosis, aneurysms, as well as the spread of metastatic cancer. Furthermore, it has been shown that following a traumatic brain injury, MMP-3 levels can also increase and cause additional damage to the blood–brain barrier [70]. The discovery of novel small molecule inhibitors of MMP-3 is, therefore, of great importance for the treatment of numerous diseases [120]. The 3,4-dihydroxyphenyllactic moiety of rosmarinic acid docks to the catalytic region, but it is too far from the catalytic zinc ion to form direct interactions with it (Figure 11, Table S21). The caffeic acid moiety docks to the S1' pocket that delimits the active site. The S1' pocket is known to confer the selectivity of compounds towards different matrix metalloproteinases. Therefore, compounds that interact within the S1' pocket and not with the catalytic zinc could selectively inhibit one particular MMP without affecting the activities of the remaining ones.

**Figure 11.** Rosmarinic acid docked in MMP3 (blue ribbons). The rosmarinic acid docks near the catalytic zinc ion and one of the catechol groups positions inside the S1 selectivity pocket. Important amino acid residues are shown in blue sticks, hydrogen bonds are denoted with dotted blue lines and coordinative bonds with dotted purple lines.

#### 3.2.4. Farnesyl Pyrophosphate Synthase

*Leishmania major* represents an intracellular, pathogenic, parasitic organism that causes cutaneous leishmaniasis. The World Health Organization stated that leishmaniasis is one of the most neglected diseases. Moreover, 350 million people are considered at risk of contracting the disease, approximately 12 million people are infected worldwide, and an estimated two million new cases occur each year [121]. Farnesyl pyrophosphate synthase (FPPS) represents an important enzyme involved in the biosynthesis of ergosterol in Leishmania parasites. Antiparasitic compounds targeting the ergosterol biosynthesis play an important role in the treatment of leishmaniasis, and the inhibition of FPPS has been shown largely effective against the related *Leishmania donovani* [122]. Interestingly, a study [114] showed that carnosic acid and carnosol form potent inhibitors of human FPPS, with IC50 values of 20.0 and 13.3 μM, respectively. It also demonstrated that inhibition of the human form of the enzyme leads to the induction of apoptosis in pancreatic cell lines by downregulating RAS prenylation. *Leishmania major* FPPS is not among the 0.05% best scoring proteins of rosemary diterpenes and is not listed in Table 1. However, it still scored extremely high with carnosol (−60.4), which is within the 3.0σ. Thus, as with phospholipase A2, we have yet another interesting case of two rosemary polyphenols potentially inhibiting the same enzyme. Both rosmarinic acid and carnosol bind approximately to the same protein space, with portions of the ligands occupying the same region as the reported *Leismania minor* FPPS inhibitor 1-(2-hydroxy-2,2-diphosphonoethyl)-3-phenylpyridinium (300B) (Figure 12, Table S22). Part of the rosmarinic acid enters the substrate-binding region where the substrate isopentenyl pyrophosphate is present in an uninhibited enzyme.

**Figure 12.** Comparison of ligand binding to farnesyl pyrophosphate synthase (blue surfaces). (**a**) Comparison between the crystal ligand 300B (blue carbons) and rosmarinic acid (green carbons) binding. (**b**) Comparison between the crystal ligand 300B (blue carbons) and carnosol (brown carbons) binding. The enzyme substrate isopentenyl pyrophosphate (IPP) was not present during the inverse docking but is shown for comparison purposes using white carbons.

#### 3.2.5. Glutamate Dehydrogenase 1 and Glutaminase

Glutaminase and glutamate dehydrogenase 1 (GDH1) represent enzymes that are both part of the glutaminolysis pathway. Glutaminolysis begins with the conversion of glutamine to glutamate by glutaminase, while the next step is catalyzed by GDH, which converts glutamate to 2-oxoglutarate. The two enzymes play a crucial role in nitrogen and carbon metabolism, as the product 2-oxoglutarate feeds the citric acid cycle. Numerous cancer cells rely on increased glutaminolysis to meet their energy requirements. It has thus been shown that the inhibition of glutaminase and GDH1 by small molecules leads to a decreased viability of cancer cells in vivo and in vitro. Consequently, they form promising targets for cancer treatment [123,124]. It has been already shown that the plant compounds from green tea epigallocatechin gallate and epicatechin gallate strongly inhibit GDH [125–127]. According to our inverse docking procedure, rosmarinic acid is located at a different binding site than the green tea compounds. It binds at hexameric 2-fold axes between the dimers of the GDH subunits, where known inhibitors such as bithionol are also located (Table S23) [127].

Rosmarinic acid binds to the allosteric pocket formed at the interface between the two dimers of glutaminase (Figure 13). In numerous crystal structures of glutaminase in the PDB, co-crystallized inhibitors have occupied this binding site, e.g., 3UO9, 3VOZ, and 3VP1 (Table S24) [128,129].

**Figure 13.** Glutaminase binding sites. Two protomers forming glutaminase are shown on blue surfaces. The main glutamine substrate-binding pockets are highlighted in orange surfaces, whereas rosmarinic acid (carbons denoted with green sticks) docks into the allosteric binding site (green surfaces) formed between the two glutamase protomers.

#### *3.3. Method Validation*

#### 3.3.1. Redocking Procedure

To validate the inverse molecular docking procedure, a redocking study was performed using all available protein complexes from the PDB containing rosmarinic acid (PDB structures: 6MQD, 3QNL, and 4PWI). An analogous redocking procedure using the investigated diterpene structures could not be performed because protein structures containing carnosic acid, carnosol, or rosmanol do not yet exist in the PDB. Redocking of rosmarinic acid was performed by first removing the ligand from the binding site. Then, the CANDOCK algorithm was used with identical settings for inverse molecular docking to bind rosmarinic acid to the binding site defined by the crystal structure. The actual binding site definition was again identical to the one found in the ProBiS-Dock Database. To evaluate the success of the redocking procedure, the root-mean-square deviation (RMSD) of all heavy atoms between the co-crystallized and the redocked rosmarinic acid was measured.

From a molecular docking perspective, rosmarinic acid represents a problematic molecule, because it contains a high number, namely seven, rotatable bonds, which makes it difficult for the docking algorithms to consistently identify the correct conformer of this molecule. This problem is reflected in the fact that we obtained a low RMSD value of 1.3 Å only with the PDB structure 3QNL, which is a snake venom-derived phospholipase A2 structure [113], compared to the original crystal structure (Figure 14). The redocking procedure was not successful for 4PWI or 6MQD structures with significantly larger RMSD values (not shown), implying that the correct pose was not detected with the CANDOCK docking algorithm. However, based on a successful redock with 3QNL and on the fact that the docking algorithm identified numerous targets that have already been also experimentally confirmed for both rosemary diterpenes as well as rosmarinic acid, we are confident that the method is capable of recognizing correct protein targets to large extent.

**Figure 14.** A successful redocking of rosmarinic acid to the crystal structure of phospholipase A2 (PDB ID: 3QNL) from the snake venom (depicted in blue ribbons and transparent surfaces). The stick structure of rosmarinic acid with blue carbons represents the native ligand position found in the crystal structure, while the structure with orange carbons displays the redocked structure. Hydrogen atoms are not shown for clarity. The RMSD between the two rosmarinic acid structures is 1.3 Å.

#### 3.3.2. Validation Using ROC, EF, and PC

We performed the inverse molecular docking using the CANDOCK algorithm on all proteins from the ProBiS-Dock database, including 206 experimentally confirmed targets of rosmarinic acid, whose measured IC50 values were < 10 μM [62]. The ability of our protocol to distinguish the confirmed protein targets from proteins that reportedly do not bind rosmarinic acid was evaluated using the metrics established in the virtual screening community (Figure 15). It was successful in discriminating between the true and false targets of rosmarinic acid (ROC AUC of 0.627). The early detection of protein targets was assessed by the BEDROC of 0.071, by the RIE of 1.403, and by the EF 1% of 1.46, which is satisfactory. The inverse molecular docking protocol based on the CANDOCK algorithm resulted in score variations for the detection of true target proteins (TG determined from PC has a value of 0.171), which in combination with ROC AUC above 0.6 indicates that the protocol provides good results in agreement with the experiments [61]. Moreover, our inverse molecular docking protocol has been already extensively validated by Fine and Konc et al. [55], Furlan et al. [50,130], Kores et al. [49,131], and Jukiˇc et al. [132] using different molecules of interest.

**Figure 15.** Validation of the inverse molecular docking protocol using rosmarinic acid: (**a**) the ROC curve, with the ROC AUC of 0.627; (**b**) predictiveness curve, from which the TG of 0.171 is determined; (**c**) enrichment curve.

#### *3.4. In Silico Prediction of Pharmacokinetic Properties*

In concurrence with experimental findings [31], the SwissADME web server [54] indeed predicts that carnosic acid, carnosol, and rosmanol all exhibit high gastrointestinal absorption (data shown in Supplementary Materials). On the contrary, the server predicts low gastrointestinal absorption for rosmarinic acid, which is again in line with the available in vivo data [43]. All compounds are predicted to be moderately soluble. Diterpenes are overall predicted as quite lipophilic, with a consensus score of logP above 3.5 for carnosic acid and carnosol, and 2.9 for rosmanol. Rosmarinic acid, as expected, due to the large number of polar functional groups, exhibits a much lower logP value of 1.2. Interestingly, carnosol is the only molecule predicted to penetrate the blood–brain barrier, however experimental studies on rat animal models show that carnosic acid also effectively penetrates the blood–brain barrier [32]. The SwissADME output is presented in its entirety in the Supplementary Materials.

#### **4. Discussion**

Natural plant-based compounds play an important role in the development of novel drugs as they may possess several advantages over conventional synthetic compounds, namely, fewer side effects, lower long-term toxicity, and versatile biological effects [130]. We report the potential targets of the major compounds from *Rosmarinus officinalis*, including the diterpenes carnosic acid, carnosol, and rosmanol, as well as the polyphenolic ester rosmarinic acid. Their targets were identified in silico using an inverse molecular docking approach. All four compounds were individually docked to all non-redundant holoproteins available in the PDB. To identify the binding sites of each protein in advance, we applied the recently developed ProBiS-Dock Database—a freely available repository of binding sites between small ligands and proteins. Thereby, the docking procedure

was limited to binding sites already known to bind at least one drug-like small-molecule ligand or to binding sites exhibiting a high similarity with the already known binding sites. Moreover, we used the novel CANDOCK algorithm, which employs a fragment-based docking approach with maximum clique and a knowledge-based scoring function.

Due to the similar molecular structure and docking/scoring values, we combined the results of all three investigated diterpenes into a single set (Table 1). We identified numerous human/mammalian proteins that could explain the observed anticarcinogenic activities of rosemary diterpenes. The best docking score was obtained for the complex between carnosic acid and the proto-oncogene K-Ras G12C. Moreover, the anticarcinogenic activities can also be explained by the potential binding of rosemary diterpenes to pyruvate kinase, PPARδ, tubulin, VEGFR2, and phospholipase A2. In general, phospholipase A2 has also been strongly implicated in inflammation-related disorders, so its inhibition may be likewise beneficial in arthritis, coronary artery disease, or dementia. Due to the identification of potential binding of the investigated diterpenes to glycogen phosphorylase, which facilitates glucose production, these compounds may be also useful in the treatment of type II diabetes. Furthermore, rosemary diterpenes exhibit antiviral activities.

From previous experimental studies, it is known that carnosol strongly inhibits HIV-1 protease. However, we also found out that rosemary diterpenes may bind strongly to the HIV-2 enzyme isotype. These compounds therefore likely represent a good starting point for the development of drugs against AIDS that could treat concurrent infections with HIV-1 and HIV-2. Interestingly, all diterpenes also yield good docking scores when bound to the feline immunodeficiency virus (FIV) protease, which is strongly related to HIV proteases, suggesting their potential utility in veterinary medicine. Finally, we have also found out that these compounds can bind to HA1 of the influenza A virus, potentially reducing its infectivity.

The antibacterial activity of investigated diterpenes can be explained by our discovery that they can bind to the enzyme glucosamine-fructose-6-phosphate aminotransferase in *Escherichia coli*, which is critical for the first step of hexosamine metabolism responsible for bacterial growth. Encouragingly, we have also found out that they can bind to the Eis protein of *Mycobacterium tuberculosis*, which confers resistance to aminoglycoside drugs, rendering them inactive. Therefore, the inhibition of this enzyme in conjunction with tuberculosis treatment could be beneficial in reducing the bacterial resistance to these drugs.

The investigated diterpenes also displayed binding to aspartate carbamoyltransferase of two different pathogenic parasites-*P. falciparum* and *T. cruzi*. *P. falciparum* represents the causative agent of malaria, while *T. cruzi* causes Chagas disease. Inhibition of this enzyme results in the inability of the two parasites to produce pyrimidines, limiting their biosynthesis of new nucleic acids.

Like diterpenes, rosmarinic acid also shows binding to proteins involved in carcinogenesis, namely matrix metalloproteinase-3 and phospholipase A2. Interestingly, all four compounds display very favorable binding scores for the enzyme phospholipase A2, which could provide a possible explanation for the strong anti-inflammatory effects of rosemary. According to our results, rosmarinic acid may also interfere with the glutaminolysis pathway, as it forms top-scoring complexes with two related enzymes—glutaminase and glutamate dehydrogenase. Inhibition of this pathway by small-molecule drugs has been indeed shown to reduce cancer cell viability. Moreover, the complex between rosmarinic acid and coagulation factor X yielded the best scoring result. Regulating blood clotting with drugs is of utmost importance, as reducing excessive blood clotting is crucial in preventing diseases such as heart attacks and ischemic strokes, which belong among the leading causes of death and disability in the Western world. Furthermore, rosmarinic acid might also possess antiparasitic activity as its binding to farnesyl pyrophosphate synthase (FPPS) of *Leishmania major* obtained a favorable docking score. This parasite causes zoonotic cutaneous leishmaniasis, and inhibition of the FPPS prevents the biosynthesis of ergosterol.

The results of this study will facilitate future molecular dynamics studies. Therein, we plan to investigate the dynamic binding patterns of prior parameterized rosemary compounds to the notable protein targets identified here. The molecular dynamics observations will be extended with the linear interaction energy as well as linear response approximation calculations to obtain the binding free energy values, which will then be compared with drug ligands already known to bind to the protein targets described here.

#### **5. Conclusions**

Using an in silico inverse molecular docking procedure, we identified protein targets that could explain the observed pharmacological activities of rosemary or its major polyphenolic constituents. By identifying protein structures to which carnosic acid, carnosol, rosmanol, and rosmarinic acid can bind, we provide possible explanations for the observed anticarcinogenic, anti-inflammatory, antidiabetic, antiviral, and antibacterial activities of rosemary. In addition, using this methodology we were able to predict new effects of these compounds that have not yet been reported, namely their anticoagulant and antiparasitic activities. Lastly, we believe that our research can form the basis for the development of novel drugs, where the rosemary compounds studied here could serve as a starting point for efficient drug design.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/foods11010067/s1. Table S1: Best docking scores of ligands/drugs known to bind to a specific target. Table S2: Best docking scores for carnosol, carnosic acid and rosmanol. Table S3: Best docking scores for rosmarinic acid. Table S4–S24: Interaction of compounds with respective targets presented in Table 1; Table 2. The SwissADME output file can be found in the file swissadme\_output.xlsx supplied as part of the Supplementary Materials.

**Author Contributions:** Conceptualization, S.L. and U.B.; methodology, S.L. and U.B.; validation, S.L.; formal analysis, S.L.; investigation, S.L.; resources, U.B.; Writing—Original draft preparation, S.L.; Writing—review and editing, S.L. and U.B.; visualization, S.L.; supervision, U.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** Financial support through Slovenian Research Agency (ARRS) grants project and programme J1-2471 and P2-0046 as well as through Slovenian Ministry of Education, Science and Sports project grants C3330-19-952021 and AB FREE is gratefully acknowledged.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** All data generated or analyzed during this study are included in the published article.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Abbreviations**



#### **References**


## *Article* **Inhibition of Biofilm Formation of Foodborne** *Staphylococcus aureus* **by the Citrus Flavonoid Naringenin**

**Qing-Hui Wen 1,2, Rui Wang 1,2, Si-Qi Zhao 1,2, Bo-Ru Chen 1,2 and Xin-An Zeng 1,3,\***


**Abstract:** Taking into consideration the importance of biofilms in food deterioration and the potential risks of antiseptic compounds, antimicrobial agents that naturally occurring are a more acceptable choice for preventing biofilm formation and in attempts to improve antibacterial effects and efficacy. Citrus flavonoids possess a variety of biological activities, including antimicrobial properties. Therefore, the anti-biofilm formation properties of the citrus flavonoid naringenin on the *Staphylococcus aureus* ATCC 6538 (*S. aureus*) were investigated using subminimum inhibitory concentrations (sub-MICs) of 5~60 mg/L. The results were confirmed using laser and scanning electron microscopy techniques, which revealed that the thick coating of *S. aureus* biofilms became thinner and finally separated into individual colonies when exposed to naringenin. The decreased biofilm formation of *S. aureus* cells may be due to a decrease in cell surface hydrophobicity and exopolysaccharide production, which is involved in the adherence or maturation of biofilms. Moreover, transcriptional results show that there was a downregulation in the expression of biofilm-related genes and alternative sigma factor *sigB* induced by naringenin. This work provides insight into the anti-biofilm mechanism of naringenin in *S. aureus* and suggests the possibility of naringenin being used in the industrial food industry for the prevention of biofilm formation.

**Keywords:** naringenin; biofilm formation; cell surface hydrophobicity; confocal laser scanning microscopy; biofilm-related genes

#### **1. Introduction**

*Staphylococcus aureus* is a common pathogen and is responsible for food poisoning through the production of thermally stable enterotoxins in various kinds of food [1,2]. A microbial biofilm is an aggregation of bacteria that is composed of extracellular polymeric substances, which are attached on the surface of microorganisms [3]. The most common feature of microbial lifestyles is attachment onto a surface by biofilm formation. Notably, *S. aureus* can form biofilms on different surfaces in food processing plants and is very adaptable to various environmental stressors including acids, salts, antibiotics, and detergents [4–6]. The presence of *S. aureus* biofilm on food contact surfaces creates serious problems for the food industry because it can lead to food spoilage and disease transmission [7,8]. Therefore, it is important to inhibit the formation of *S. aureus* biofilms on food contact to surfaces ensure the manufacture of safe food products.

Flavonoids from fruits and vegetables have been shown to has a range of biological activities [9,10]. For example, the citrus flavonoid naringenin have beneficial effects on human health by preventing various diseases, including diabetes, hypertension and cancer [11–13]. Moreover, naringenin has wide antibacterial activity and can prevent the growth of numerous microorganisms [14,15]. Specifically, naringenin from bergamot peel

**Citation:** Wen, Q.-H.; Wang, R.; Zhao, S.-Q.; Chen, B.-R.; Zeng, X.-A. Inhibition of Biofilm Formation of Foodborne *Staphylococcus aureus* by the Citrus Flavonoid Naringenin. *Foods* **2021**, *10*, 2614. https:// doi.org/10.3390/foods10112614

Academic Editor: Gary Dykes

Received: 23 September 2021 Accepted: 25 October 2021 Published: 28 October 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

inhibits *Escherichia coli*, *Lactococcus lactis*, *Salmonella* Enteritidis, and *Pseudomonas putida* with minimum inhibitory concentration (MIC) values ranging from 0.25 to 1.0 g/L [9]. A small number of studies have reported that naringenin inhibits the biofilm formation of microorganisms (*Escherichia coli*, *Vibrio harveyi* and *Streptococcus mutans*) by affecting the expression of bacteria related genes and surface hydrophobicity [16,17]. In our earlier study, we determined that naringenin has strong antibacterial activity against *S. aureus* via such mechanisms of action as disrupting the bacterial cytoplasmic membrane and binding to its genomic DNA [18]. Moreover, we also found that naringenin has a strong effect in suppressing the biofilm formation of S. aureus on the surface of glass and plastic well plates. However, to the best of our knowledge, research into biofilm inhibition by naringenin is limited, and its anti-biofilm mechanism is also unclear.

Hence, we aimed to study the effect of naringenin on the inhibition of the biofilm formation of *S. aureus* at different temperatures (25 and 37 ◦C) using confocal laser and scanning electron microscopy techniques and exopolysaccharide production (EPS) and hydrophobicity assays. Furthermore, our study also investigated the genes (*sigB*) related to *S. aureus* biofilms using RT-qPCR, which is the main regulator of gene transcription and expression under the stress conditions induced by naringenin.

#### **2. Materials and Methods**

#### *2.1. Bacterial Strain and Biofilm Formation*

The foodborne strain *S. aureus* was obtained from the Microbial Culture Collection Center of Guangdong Institute of Microbiology (Guangzhou, China) and activated by culturing twice in 100 mL of sterile tryptic soy broth (TSB, Beijing Aoboxing Biotechnology Co., Ltd., Beijing, China) at 37 ◦C for 24 h. The effect of naringenin (purity ≥ 98%, Aladdin Chemical Co., Shanghai, China) on *S. aureus* growth was evaluated by transferring precultured *S. aureus* cells into fresh TSB liquid medium (OD600 nm ≈ 0.08) and cultivating in 96-well polystyrene plates at different temperature (25 and 37 ◦C) with gentle shaking. In order to measure the absorbance value of *S. aureus* growth, a FilterMax F5 multifunctional microplate reader (American molecular devices company, Sunnyvale, CA, USA) was used.

The biofilm assay was performed under similar conditions without shaking. The volume of DMSO, that was used to dissolve naringenin was equal in all of the samples, while the final concentration of naringenin varied from 0 to 60 mg/L. The crystal violet staining method was used to quantify *S. aureus* biofilm according to a relevant publication [19].

#### *2.2. Cell Surface Hydrophobicity of S. aureus*

The effects of naringenin on *S. aureus* cell surface hydrophobicity were evaluated at 25 and 37 ◦C by analyzing cells adhesion to xylene, as previously described [20]. After cultivation to the stationary-phase (48 h for *S. aureus* at 25 ◦C, and 12 h for *S. aureus* at 37 ◦C), *S. aureus* cells were collected by refrigerated centrifugation at 4000× *g* for 5 min. The *S. aureus* pellet was washed twice using distilled water and resuspended in 3 mL of a 0.85% NaCl solution (OD600 nm ≈ 0.3), which defined as A1. Xylene (1 mL) was added to a 3 mL suspension of *S. aureus* and then incubated for 15 min at 25 ◦C. After vortexing for two minutes and then incubating for 15 min, the mixture separated into a xylene/water bilayer system. The OD600 nm of the aqueous phase of the bilayer was recorded as A2. The index of cell surface hydrophobicity (I) was determined using Equation (1):

$$I = (1 - A\_2/A\_1) \times 100\% \tag{1}$$

#### *2.3. Quantification EPS Production of S. aureus*

After growing in the various subminimum inhibitory concentrations of naringenin, the *S. aureus* cells were centrifuged (12,000× *g* for 15 min at 4 ◦C) and the supernatant of *S. aureus* was then filtered through glass fiber filters. An equal volume of absolute ethanol was then added to this supernatant of *S. aureus* and incubated overnight at 4 ◦C to precipitate EPS. The precipitate was resuspended in water with gentle heating (50 ◦C) and then quantified using a phenol–sulfuric acid procedure [21]. The percentage of EPS reduction upon exposure to naringenin was evaluated using Equation (2):

$$\text{Reduction of EPS (\%)} = (1 - E\_2/E\_1) \times 100\% \tag{2}$$

where *E*<sup>1</sup> and *E*<sup>2</sup> are the absorbances at 490 nm for *S. aureus* cells grown in the absence and presence of naringenin, respectively.

#### *2.4. CLSM and SEM of S. aureus Biofilms*

A 3 mL aliquot of inoculum (OD600 nm ≈ 0.1) was transferred into the wells of a 6-well plate containing 13 mm-diameter sterile glass coverslips. After incubation at 25 and 37 ◦C for 48 h and 12 h respectively, the medium with free-floating *S. aureus* cells was removed and the coverslips washed thrice in sterile 0.85% saline solution. *S. aureus* biofilms on glass coverslips were then stained for 20 min in the dark at room temperature with diluted 5(6)-carboxy fluorescein diacetate succinimidyl ester (Aladdin Chemical Co., Shanghai, China). The stained biofilms were adjusted and photographed using a confocal laser scanning microscope (Leica, Wetzlar, Germany). At least ten pictures were taken from different locations for each sample, and the image data were then processed and analyzed.

*S. aureus* biofilms were prepared for SEM analysis as previously described, with minor modifications [18]. Glutaraldehyde (2.5% in 0.01 M phosphate buffer, pH 7.2) was added to the samples and incubated overnight at 4 ◦C and then dehydrated using a series of ethanol solutions (20 min each time) of increasing concentration (30~100%). The dehydrated biofilms were then incubated in tertiary butanol twice for 20 min each, followed by airdrying overnight. After gold-coating by ion sputtering (Jeol JFC-1100, Tokyo, Japan), *S. aureus* biofilms were photographed by scanning electron microscopy (SEM, Zeiss EVO18, Germany) with operation at 10.0 kV.

#### *2.5. RNA Extraction and Real-Time Quantitative PCR (RT-qPCR) Analysis*

TRIzol reagent (Invitrogen, CA, USA) was used according to the kit instructions to extract the *S. aureus* RNA. To check the concentration and purity, the RNA was measure at OD260 and OD280 using an 1800 UV spectrophotometer (Shimadzu Corporation, Kyoto, Japan). cDNA was reverse transcribed from 800 ng RNA with 4.0 μL of 5 × reaction buffer, 0.5 μL Thermo Scientific RiboLock RNase Inhibitor (20 U) and 1.0 μL RevertAid Premium Reverse Transcriptase (200 U), following the protocol of RevertAid Premium First Strand cDNA Synthesis Kit (Thermo Scientific™ EP0733, Thermo Fisher Scientific, Waltham, MA, USA).

RT-qPCRs was performed on an Applied Biosystems StepOne Plus™ thermocycler (Life Technologies Inc., Milano, Italy) using the SybrGreen qPCR Master Mix, following the kit instructions. Reactions were carried out in a system which was composed of 10 μL Master Mix, 0.4 μL of 0.25 μM solutions of each primer (Table 1), and 2 μL cDNA, diluted to a final volume of 20 μL using double-distilled water (DNase-free). The following thermal profile was used: a holding step for 3 min at 95 ◦C, followed by a cycling step consisting of 45 cycles at 95 ◦C for 7 s (to melt), 57 ◦C for 10 s (to anneal) and 72 ◦C for 15 s (to extend). The endogenous reference gene of 16S rRNA was used to evaluate the changes in transcriptional levels of the *S. aureus* RNA.

#### *2.6. Statistical Analysis*

Results are expressed as means ± standard deviation (SD), and data graphics were drawn using OriginPro 7.0 (OriginLab, Northampton, MA, USA). SPSS software (IBM, Armonk, NY, USA) was used to analyze the variance (ANOVA) by Tukey's test, and *p* < 0.05 was represented for significant difference.


**Table 1.** Sequences of the primers used for RT-qPCR.

#### **3. Results**

#### *3.1. Effects of Naringenin on S. aureus at Different Growth Temperatures*

Figure 1 shows the effect of naringenin on cell growth profiles of *S. aureus* at different temperatures as reflected by the optical density (OD) at 600 nm. Subminimum inhibitory concentrations (MIC for 0.5 g/L) [22] values in the range of (5~60 mg/L) did not decrease the cell density of *S. aureus* at 25 ◦C (Figure 1a) and 37 ◦C (Figure 1c). However, the time taken to reach the stationary phase for *S. aureus* was significantly affected by temperatures. S. aureus cultured at 37 ◦C takes 12 h to reach the stationary phase, as compared to 48 h for *S. aureus* grown at 25 ◦C.

The effect of naringenin on the biofilms formed at different temperatures was measured by crystal violet staining, expressed as OD570 nm. Due to different growth rates, *S. aureus* at 25 and 37 ◦C were incubated for 48 and 12 h, respectively. *S. aureus* biofilm formation (measured at OD570 nm) decreased with increasing concentrations of naringenin. For example, the OD570 nm of S. aureus grown at 37 ◦C was reduced by 0.94 (32.3%) with 10 mg/mL and 0.61 (56.1%) with 20 mg/mL naringenin (*p* < 0.05). A further decrease was observed after the addition of 30 mg/mL or higher naringenin concentrations (Figure 1d). By contrast, *S. aureus* cultivated at 25 ◦C showed a larger decrease (*p* < 0.05) in biofilm formation (as measured by OD570 nm) in the presence of naringenin. For example, OD570 nm decreased from 1.21 to 0.22 (81.8%) after exposure to naringenin at a concentration of 20 mg/L (Figure 1b).

#### *3.2. Changes in Cell Surface Hydrophobicity and EPS Production of S. aureus*

Surface hydrophobicity of *S. aureus* cells was determined and expressed as hydrophobicity index (I). Figure 2a shows the significant dose-related reduction (*p* < 0.05) in cell hydrophobicity of *S. aureus* with increasing naringenin concentration at both 25 and 37 ◦C. Naringenin at a concentration of 10 mg/L dramatically reduced the surface hydrophobicity of *S. aureus* cells grown at 25 and 37 ◦C by 40.6% and 57.2% (*p* < 0.05), respectively. The respective values further decreased to 14.4% and 21.2% (*p* < 0.05) when the concentration of naringenin was increased to 40 mg/L.

The effect of naringenin on the EPS production of *S. aureus* was also investigated. The results reveal that *S. aureus* treated with various concentrations of naringenin (20, 40, and 60 mg/L) show a significant reduction (*p* < 0.05) in EPS compared to that of the control (Figure 2b). For *S. aureus* grown at 25◦C, naringenin at 20 and 60 mg/mL reduced EPS by 59% and 5%, respectively. At 37 ◦C, the same concentrations of naringenin reduced EPS by 67% and 18%, respectively.

**Figure 1.** Cell growth of *S. aureus* in the presence of naringenin (0~60 mg/mL) was measured at OD600 nm in 96-well plates at 25 ◦C (**a**) or 37 ◦C (**c**). Biofilm formations of *S. aureus* with naringenin concentration of 0~60 mg/mL at 25 ◦C (**b**) or 37 ◦C (**d**) for 48 and 12 h in 96-well plates, respectively. Biofilm OD values are processed as mean ± SD and a–g indicate significant differences between different columns (*p* < 0.05).

#### *3.3. Microscopic Observations of S. aureus Biofilm*

Direct visual information, including the surface coverage and thickness of *S. aureus* biofilms, were obtained by CLSM analyses. As shown in Figure 3a,b, *S. aureus* formed thick and compact biofilms covering the surface of glass coverslips at 25 and 37 ◦C when grown in the absence of naringenin. The confocal stack images show that the thick coating of *S. aureus* biofilms represented by cell aggregations became thinner and looser on the surfaces in the presence of 20 mg/L naringenin (Figure 3d). At 40 mg/L of naringenin, there was a visible reduction in the numbers of microcolonies for *S. aureus* cells grown at 37 ◦C (Figure 3f). Compared to 37 ◦C, the cells grown at 25 ◦C had a more obvious decrease associated with naringenin exposure, and the bacterial density was significantly decreased (Figure 3c,e). These results were further confirmed by SEM images.

The SEM images show that naringenin inhibited the bacterial growth of *S. aureus* at subminimum inhibitory concentrations (sub-MICs) values of 0, 20 and 40 mg/L. As the concentration of naringenin increases, the total number of bacteria obviously decreased, especially at high concentrations (Figure 4e,f) [17]. Compared to the incubation temperature of 37 ◦C (Figure 4b,d,f), the total number of *S. aureus* cultivated at 25 ◦C (Figure 4a,c,e) showed a greater decrease, demonstrating that naringenin has a significant effect in suppressing the biofilm formation of bacteria.

**Figure 2.** Effects of naringenin in the concentrations of 0, 20 and 40 mg/L on cell surface hydrophobicity (**a**) and EPS production (**b**) of *S. aureus*. Values are mean ± SD and there are significant differences between the values of columns marked with different letters (a–d) and (A–D), as indicated (*p* < 0.05).

**Figure 3.** Confocal laser scanning microscopy (CLSM) analysis of biofilms formed by *S. aureus* incubated with different concentrations of naringenin. (**a**,**c**,**e**) for *S. aureus* cells were grown at 25 ◦C with naringenin at 0, 20 and 40 mg/L, respectively; (**b**,**d**,**f**) for *S. aureus* cells were grown at 37 ◦C with naringenin at 0, 20 and 40 mg/L, respectively.

**Figure 4.** SEM images showing inhibitory activity of naringenin on biofilm formation of *S. aureus* cells. (**a**,**c**,**e**) for *S. aureus* cells grown at 25 ◦C with naringenin at 0, 20 and 40 mg/L, respectively; (**b**,**d**,**f**) for *S. aureus* cells grown at 37 ◦C with naringenin at 0, 20 and 40 mg/L, respectively.

#### *3.4. Transcriptional Analysis of Biofilm-Related Genes in S. aureus Cells*

The effect of naringenin on the level of expression of biofilm-related genes, including *cidA*, *icaA*, *dltB*, *agrA*, *sortaseA*, *sarA* and sigma factor *sigB* in *S. aureus* cells, were studied by RT-qPCR. Among the seven tested genes, *icaA*, *agrA*, *sarA* and *sigB* demonstrated significantly down-regulated (*p* < 0.05) gene expression when treated with naringenin, while *cidA* and *dltB* were up-regulated. Specifically, *icaA*, *agrA*, *sarA* and *sigB* were significantly

down-regulated (*p* < 0.05) by 0.47-, 0.49, 0.58- and 0.63-fold for *S. aureus* grown at 25 ◦C in a culture medium with 20 mg/L naringenin (Figure 5a), and further decreased by 0.22-, 0.10-, 0.11- and 0.44-fold after the concentration of naringenin was increased to 40 mg/L, respectively. Under the same conditions, the expression of *cidA* and *dltB* were mildly up-regulated by 1.29- and 2.08-fold when *S. aureus* cells were exposed to naringenin at 40 mg/L. The expression of genes, including *cidA*, *dltB*, *icaA*, *agrA*, *sarA* and *sigB*, exhibited a similar trend for *S. aureus* cells cultivated at 37 ◦C (Figure 5b).

**Figure 5.** Effect of naringenin on the expression levels of biofilm-related genes in *S. aureus* at 25 ◦C (**a**) and 37 ◦C (**b**), where 16S rRNA was used as a reference gene. Data are presented as means ± standard deviations. Relative expression values are processed as mean ± SD and a–c indicate significant differences between different columns (*p* < 0.05).

By contrast, the levels of *sortaseA* expression in *S. aureus* cells grown at 25 ◦C were down-regulated in the presence of naringenin at concentrations of 20 and 40 mg/L, exhibiting 0.82- and 0.78-fold decreases (*p* < 0.05), respectively. At 37 ◦C, naringenin had no obvious effect on the transcription level of the studied *sortaseA* genes.

#### **4. Discussion**

In this work, no significant inhibitory effect of naringenin on S. aureus biofilm formation at 25 and 37 ◦C was found. However, the growth rate of S. aureus at 25 ◦C was significantly lower than that at 37 ◦C, without the final biomass being affected. Thus, different incubation times, i.e., 48 and 12 h, were used for 25 and 37 ◦C, respectively, to compensate for variations in the time required to reach the stationary phase. At subminimum inhibitory concentrations (sub-MICs) values ranging from 5 to 60 mg/L, naringenin dramatically inhibited *S. aureus* biofilm formation, with the biofilm formation further decreasing as the concentrations of naringenin with increased. Compared to an incubation temperature of 37 ◦C, the presence of naringenin resulted in a more significant effect on *S. aureus* biofilm formation for cultivation at 25 ◦C. There are various incubation temperatures that affect the growth of *S. aureus* and a series of changes caused by differences in subsequent growth.

Cell surface hydrophobicity is an important physical-chemical property of bacteria that facilitates their attachment to surfaces. Previously, it was reported that bacterial adherence of oral streptococci to the tooth surface was significantly suppressed by the reduction in cell surface hydrophobicity after treatment with constituents in cranberry juice and tea extract polyphenols [23,24]. Our results indicated that the cell surface hydrophobicity of *S. aureus* is reduced by treatment with naringenin (Figure 2a).

EPS is also important for biofilm production, forming multiple layers that help pathogens to adhere to surfaces and maintain biofilm architecture and acting as a protective barrier to prevent the entry of antibacterial agents into bacterial cells [25,26]. Therefore, the substantial decrease in EPS production by *S. aureus* after treatment with naringenin (Figure 2b) is consistent with this scenario. Since EPS production and surface hydrophobicity are important factors for biofilm formation, the decreases in these two properties following treatment with naringenin are the likely reasons that naringenin decreases biofilm formation in *S. aureus* cells.

Naringenin is not an antibiotic and, thus, the results obtained from CLSM and SEM are inconsistent with some previous studies that have reported that low concentrations of βlactam and aminoglycoside antibiotics often facilitate biofilm formation by bacteria [27,28]. However, the inhibition of biofilm formation by naringenin is in agreement with the behavior of other flavonoids, including morin, rutin, quercetin and phloretin [25,29–31]. Notably, naringenin has a better effect on inhibiting the formation biofilms of *S. aureus* at the incubation temperature at 25 than at 37 ◦C.

To elucidate the underlying molecular mechanism for the inhibition of *S. aureus* biofilm by naringenin, we investigated the expression of some biofilm-related genes using RTqPCR, including *dltB*, *sarA*, *sortaseA*, *agrA*, *icaA*, *cidA* and *sigB*. The *dltB* gene is responsible for D-alanylation of teichoic acids and the translocation of D-alanine through the cell membrane. It has been previously reported that *dltB* deficiency results in a higher negative net charge on the bacterial cell surface and defects in the initial binding of bacteria to a surface in the process of biofilm formation [32]. The expression of *cidA* was reported to be associated with extracellular DNA release, which is essential in the formation of *S. aureus* biofilm [33]. Our results revealed that the *dltB* and *cidA* genes are mildly up-regulated in *S. aureus* cells exposed to naringenin at 25 and 37 ◦C. Thus, it can be inferred that the suppression of biofilm formation by naringenin is probably not through *dltB* and *cidA* in *S. aureus*.

By contrast, the *icaA*, *agrA* and *sarA* genes are down-regulated in the presence of naringenin. The *ica* operons encode enzymes involved in the biosynthesis of polysaccharide intercellular adhesion [34]. Since it is well-known that the ability of *S. aureus* to form biofilm is dependent on polysaccharide intercellular adhesion, down-regulation of *icaA* likely decreases the production of polysaccharide intercellular adhesion, leading to reduction of biofilm formation. Thus, the decreased expression of *icaA* by naringenin might lead to a reduction in the biosynthesis of polysaccharide intercellular adhesion, which hinders the attachment of *S. aureus* cells to solid surfaces and subsequent biofilm formation. Both the *agr* (accessory gene regulator) and *sar* (staphylococcal accessory regulator) operons are two regulatory elements that control the production of virulence factors, as well mediate *S. aureus* biofilm formation by regulating the quorum sensing and polysaccharide intercellular adhesion production [35]. Thus, it can also be inferred that the down-regulated gene expression of *agrA* and *sarA* negatively affects biofilm formation by *S. aureus*. These data are in agreement with previous studies showing that *agr* down regulation has a negative impact on biofilm development by *S. aureus* [33,36].

Additionally, the expression levels of *sigB* were reduced for *S. aureus* cells exposed to naringenin. *sigB* is an alternative sigma factor modulating various stress responses in several Gram-positive bacteria, including *S. aureus* via a large regulon [37]. Overexpression of these genes can confer more resistance to heat, oxidative and antibiotic stresses. In summary, down-regulating the above genes may be the mechanism of action by which naringenin inhibits biofilm formation.

In some Gram-positive bacteria, including staphylococci, enterococci and streptococci, the *sortaseA* gene is responsible for coding a membrane enzyme that plays an important role in grappling some surface-exposed proteins to the cell wall envelope [38]. Recent reports have shown that *sortaseA* upregulation significantly increases the levels of biofilm formation in staphylococcal strains [39]. Our results show a mild down-regulation of *sortaseA* in the presence of naringenin for *S. aureus* grown at 25 ◦C. By contrast, at 37 ◦C no obvious

change in the transcription level of *sortaseA* was observed. Since adhesion to the surface is an essential step for biofilm formation, the larger decrease in biofilm formation observed in *S. aureus* cultivated at 25 ◦C may be attributed to the down-regulated expression of *sortaseA* in *S. aureus* cells exposed to naringenin.

#### **5. Conclusions**

The data from our investigation indicates the potential of naringenin as a natural agent to prevent biofilm formation of S. aureus and possibly reduce health risks related to biofilm-formation in the food industry. However, more studies are necessary to gain a better understanding whether of there is any anti-biofilm activity toward other foodborne pathogens in food industry, and the efficiency when considering stainless steel and plastic surface.

**Author Contributions:** Conceptualization and methodology, Q.-H.W. and R.W.; software, data curation and formal analysis, Q.-H.W. and S.-Q.Z.; writing—original draft preparation, Q.-H.W. and R.W.; writing—review and editing, S.-Q.Z.; visualization, B.-R.C.; supervision, X.-A.Z.; funding acquisition, X.-A.Z. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the S&T projects of Guangdong Province (2019B020212004) and the 111 Project (B17018).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

