Identifying Phlorofucofuroeckol-A as a Dual Inhibitor of Amyloid-β25-35 Self-Aggregation and Insulin Glycation: Elucidation of the Molecular Mechanism of Action

Both amyloid-β (Aβ) and insulin are amyloidogenic peptides, and they play a critical role in Alzheimer’s disease (AD) and type-2 diabetes (T2D). Misfolded or aggregated Aβ and glycated insulin are commonly found in AD and T2D patients, respectively, and exhibit neurotoxicity and oxidative stress. The present study examined the anti-Aβ25-35 aggregation and anti-insulin glycation activities of five phlorotannins isolated from Ecklonia stolonifera. Thioflavin-T assay results suggest that eckol, dioxinodehydroeckol, dieckol, and phlorofucofuroeckol-A (PFFA) significantly inhibit Aβ25-35 self-assembly. Molecular docking and dynamic simulation analyses confirmed that these phlorotannins have a strong potential to interact with Aβ25-35 peptides and interrupt their self-assembly and conformational transformation, thereby inhibiting Aβ25-35 aggregation. In addition, PFFA dose-dependently inhibited d-ribose and d-glucose induced non-enzymatic insulin glycation. To understand the molecular mechanism for insulin glycation and its inhibition, we predicted the binding site of PFFA in insulin via computational analysis. Interestingly, PFFA strongly interacted with the Phe1 in insulin chain-B, and this interaction could block d-glucose access to the glycation site of insulin. Taken together, our novel findings suggest that phlorofucofuroeckol-A could be a new scaffold for AD treatment by inhibiting the formation of β-sheet rich structures in Aβ25-35 and advanced glycation end-products (AGEs) in insulin.


Introduction
The aberrant aggregation of misfolded proteins within a biological system is responsible for various pathological conditions. Protein aggregates commonly form and accumulate during normal aging, and it remains unclear whether misfolded proteins are a cause or consequence of aging [1]. In any case, protein aggregates are the major hallmarks of numerous neurodegenerative and metabolic disorders, and many central nervous system pathologies are associated with protein aggregation, such as amyloid-β peptide (Aβ) and tau protein aggregates in Alzheimer's disease (AD), α-synuclein in Parkinson's disease (PD), and the huntingtin protein in Huntington's disease [2].

Inhibition of Insulin Glycation by Phlorotannins
Glycated bovine insulin was observed by fluorescence spectroscopy because AGEs are marked by a typical fluorescence emission at 410 nm (excitation at 320 nm). To verify our experimental condition, we used vanillin as a negative control for d-ribose-induced protein glycation [28] and rutin as a positive control for d-glucose-induced protein glycation [29].
As shown in Figure 2C, fluorescence intensity after a 1-week incubation of bovine insulin and d-ribose increased significantly compared to the blank group (p < 0.001). However, in the presence of 100 µM eckol, PFFA, or dieckol, a significant reduction of insulin glycation was detected, as indicated by a decline in fluorescence intensity. Those inhibitory activities of eckol, PFFA, and dieckol were dose-dependent, with IC 50 values of 258.54 ± 10.81, 29.50 ± 0.53, and 63.67 ± 3.83 µM, respectively ( Figure 2D). Phloroglucinol and dioxinodehydroeckol showed weak or no inhibitory activity on d-ribose-induced insulin glycation at 100 µM, and the negative control (vanillin) showed no activity at 500 µM.

Dynamic Simulation of Phlorotannins Inhibiting Aβ 25-35 Self-Aggregation
In the absence of inhibitors, the amorphous form of Aβ 25-35 changed into the β-sheet form with many internal hydrogen bonds between strands in a 150 mM NaCl aqueous solution during a 20 ns MD simulation ( Figure 5A-C). We analyzed and visualized the evolution of the secondary structure during that 20 ns MD simulation using VMD. As shown in Figure 6A, the β-sheet began being generated at 5.9 ns and continued forming until 20 ns.
To understand the binding modes between Aβ and phlorotannins, we subjected the most stable phlorotannin-Aβ 25-35 complexes from the docking study to MD simulation.
As shown in Figure 7A, the MD simulation results suggest that eckol interacts favorably with the Asn27-Lys28-Gly29 residues of the peptide. After 20 ns, the hydroxyl moiety of the eckol formed strong hydrogen bonds with the Asn27 and Gly29 residues, and the aromatic ring of this compound interacted with Lys28 via an amide-pi stacked bond. Those interactions between the peptide and eckol increased the β-bridge content during the 20 ns MD simulation. However, unlike Aβ25-35 alone, no β-sheet was observed in the presence of eckol ( Figure 6A).  Dioxinodehydroeckol formed many interactions with most of the peptide residues, which eliminated all β-structures, including β-sheet and β-bridge formations ( Figure 6A). Although it did not interact with Met35 and Ile31, which are assumed to play a key role in the assembly and neurotoxicity of Aβ [31], in the best docked pose ( Figure 4D), dioxinodehydroeckol continuously interacted with those residues via hydrophobic and hydrophilic bonds during the MD simulations. After 20 ns of simulation, dioxinodehydroeckol formed seven hydrogen bonds with the Ser26, Ile32, Gly25, and Met35 residues. In addition, the aromatic ring of this compound interacted with Ala30 and Met35 via several pi-interactions: pi-sigma, pi-sulfur, and pi-alkyl bonds ( Figure 7B). Figure 7C, dieckol strongly interacted with Met35 and Lys28 via hydrogen bonds during the simulations. In addition to Met35, dieckol continuously reacted with the Ser26, Lys28, Ile31, and Leu34 residues. After the 20 ns MD simulation, hetero oxygen atoms and the hydroxyl moiety of dieckol formed hydrogen bonds with Met35 and Lys28. Pi-interactions were also detected between the aromatic rings of dieckol and the Met35, Ile31, Leu34, Ser26, and Lys28 residues. PFFA ( Figure 7D) interacted favorably with the Ile31, Ala30, Gly29, and Gly33 residues during the simulations and effectively interrupted the self-assembly of Aβ [25][26][27][28][29][30][31][32][33][34][35] . After the 20 ns simulation, this compound formed a hydrogen bond with Asn27 and a strong electrostatic interaction (pi-cation) with Lys28. In addition, hydrophobic interactions were observed between PFFA and the Gly29 and Lys28 residues. Furthermore, a secondary structure analysis revealed that the formation of a β-sheet decreased significantly in the presence of dieckol and PFFA, as shown in Figure 8A. We also analyzed the minimum distances between each residue in the peptide and the phlorotannins during the 20 ns simulations. As shown in Figure 6B, dieckol, PFFA, and dioxinodehydroeckol moved to within 0.5 nm of the Aβ 25-35 during the simulations, whereas eckol moved to within 1.0 nm of the peptide. In addition, these four phlorotannins continuously interacted with Asn27 and Ile31 at a close distance during the 20 ns MD simulations. Dioxinodehydroeckol and dieckol interacted closely with Gly33, Leu34, and Met35. PFFA did not interact closely with Met, but it did compactly interact with other hydrophobic residues, including Ile31-Ile32-Gly33-Leu34. However, phloroglucinol could not access the Aβ 25-35 peptide until 3 ns into the 20 ns MD simulation even though the simulation began with a docked phloroglucinol-Aβ 25-35 complex (data not shown).

Docking Simulation for PFFA on Bovine Insulin
The most stable binding site of PFFA on the bovine insulin was analyzed using an in silico automated docking study. PFFA showed a negative binding energy (−5.03 kcal/mol) to the bovine insulin. As shown in Figure 8A and Table 2, the hydroxyl moieties of the PFFA interacted with Phe1 and Asn3 in chain B and Ser12, Gln15, Glu17, and Asn18 in chain A of the insulin via hydrogen bonds. In addition, the dibenzofuran ring of the PFFA interacted with Tyr14 in chain A via a pi-amide stacked interaction.

Dynamic Simulation of PFFA on Bovine Insulin
To understand the binding modes between insulin and PFFA, we subjected the most stable PFFA-insulin complex ( Figure 8A) obtained from the docking study to MD simulation.
As shown in Figure 8B, the MD results reveal that PFFA moved slightly closer to Phe25 (chain B) over time, and the PFFA-insulin complex was finally stabilized by intra-interactions between PFFA and insulin residues, including Ser12 and Tyr19 in chain A and Phe1 and Phe25 in chain B.
Furthermore, minimum distances (nm) between PFFA and major glycation site residues including Phe1, Arg22, and Lys29 in chain B and Gly1 in chain A of insulin over MD run times are described in Figure 8C. PFFA formed stable interaction with Phe1 in chain B over the entire runs, whereas PFFA interacted with Gly1 in chain A via H-bond from 5 ns until 10 ns runs (data not shown). In addition, PFFA weakly interacted with Arg22 and could not reach near Lys29 residue during 15 ns MD runs.

Discussion
Phlorotannins, a natural polyphenol found abundantly in brown seaweeds (especially in the Ecklonia species), are known to have a diverse range of pharmacological activity. As neuroscience research progresses, the many neuroprotective effects of various phlorotannins are being reported, including inhibitory activity against enzymes linked to the pathogenesis of AD and PD [22,24,25,32], modulatory activity against G-protein coupled receptors related to neuronal diseases such as PD and psychological diseases [22,23], and free-radical scavenging activity [15]. Although some reports have indicated that phlorotannins have protective effects against the neurotoxicity of the Aβ 1-42 oligomer [32] or Aβ 25-35 peptides [33,34] in neuronal cell-lines, phlorotannins have not previously been studied as an inhibitor of Aβ self-aggregation.
Our molecular docking and MD simulation studies have clearly demonstrated the binding modes between the peptide and phlorotannins. After a 20 ns MD simulation, Aβ 25-35 alone showed β-sheet content and had β-turn content at Gly29-Ala30, in accord with the results of a previous study [40]. However, when bound to eckol, dioxinodehydroeckol, dieckol, or PFFA, the peptide had significantly less β-sheet content and existed in an amorphous state. The MD analysis revealed that these four phlorotannins commonly interacted with the Asn27 and Ile31 residues, though their interactions differed. Eckol and dioxinodehydroeckol favorably interacted with hydrophilic residues, including Asn27-Lys28-Gly29, whereas dieckol and PFFA mainly interacted with the C-terminus hydrophobic residues, including Ile31-Ile32-Gly33-Leu34-Met35. It was previously reported that Aβ [25][26][27][28][29][30][31][32][33][34][35] assemblies are mediated by side-chain to side-chain hydrogen interactions in the Asn27-Ile32 region. In addition, an experimental analysis conducted by Pike and coworkers confirmed that the Leu34-Met35 region is essential to Aβ aggregation, and Met35 is important for the neurotoxicity of Aβ 25-35 / 1-40 [31,38]. Therefore, the different binding aspects of the phlorotannins could explain their different potency against Aβ 25-35 aggregations in vitro. However, it was reported that GROMOS force fields showed strongly biased results toward β-sheet structures [41] and the drug-binding sites of Aβ monomers and small oligomers are very transient [42], thus further extensive MD simulation using other CHARMM, AMBER99-ILDN, or AMBER14SB force fields, which showed good balanced results in structures as well as kinetics, should be conducted to confirm our MD results [42,43].
For many years, it was commonly believed that the brain was insensitive to insulin. However, it is now acknowledged that insulin has vital neuro-modulatory functions, such as the regulation of glucose homeostasis and roles in cognition, learning, and memory, which are impaired in AD [44]. In addition, insulin can prevent the formation of the Aβ 1-42 oligomer and ameliorate the Aβ 1-42 -induced impairment of long-term potentiation in hippocampal slices [45]. But once insulin is glycated under hyperglycemic or diabetic conditions, it cannot bind IR or block Aβ aggregation, which produces a decline in IR-mediated signaling pathways and can facilitate Aβ-mediated brain damage [9,45]. Glucose is the most abundant reducing sugar in vivo with its plasma concentrations ranging from 70 to 140 mg/dL in healthy individuals, while two times higher in T2D patients [46]. In addition, abnormally high doses of d-ribose have been found in the urine of T2D patients [47], and ribosylated insulin was found to exhibit significant cytotoxicity in NIH-3T3 cells [10].
Although some reports have suggested that natural products, such as vanillin, rutin, quercetin, and pinocembrin, can act as insulin glycation inhibitors [28], studies about insulin glycation remain inadequate. Therefore, in this study, we evaluated the inhibitory effects of phlorotannins on d-glucose or d-ribose-induced non-enzymatic and irreversible glycation of bovine insulin, and we elucidated the molecular mechanism of action for insulin glycation and its inhibition by phlorotannins.
PFFA and dieckol showed remarkably potent inhibition of d-ribose induced insulin glycation at less than 100 µM. However, other phlorotannins showed weak or no activity at the tested concentrations. In the case of d-glucose induced insulin glycation assay, only PFFA showed significant inhibitory activity at 100 µM. Our results suggest that PFFA might be promising lead compounds against non-enzymatically glycated insulin-mediated pathogenesis.
Glucose-binding sites (Phe1, Val2, Leu17, Arg22, and Lys29 in chain B; Glu1 in chain A) on insulin had already been elucidated via in silico prediction and mass spectrometry studies [48,49]. Our docking and MD simulation analyses clearly show the binding modes between bovine insulin and PFFA. Phe1 in chain B of insulin, which is the major glycation site of insulin, connected strongly with PFFA via H-bond and pi-pi bonds during MD runs. Our computational prediction also indicated that PFFA could interact with Ser12, Tyr19, and Phe25 residues, which could inhibit insulin glycation by disturbing the interaction between glycation site of insulin and d-glucose (or d-ribose).
In the brain, Aβ plaques and AGEs can be major sources of oxidative stress [14,37]. Brains are highly susceptible to oxidative damage because their membranes contain high amounts of polyunsaturated and peroxidable fatty acids and have a high rate of oxygen consumption [50]. In our study, eckol, dioxinodehydroeckol, dieckol, and PFFA dose-dependently reduced MDA levels in whole rat brain tissue homogenate. Dioxinodehydroeckol, dieckol, and PFFA showed strong activity, and eckol had moderate potency. Thus, phlorotannins with more than three repeating phloroglucinol units are required to prevent lipid peroxidation. However, in the homogenates, destroying the structures and cells via a myriad of processes that is started cannot occur in the living organism and almost all unspecific antioxidants could inhibit this nonspecific peroxidation [51,52]. Therefore, further mechanism studies are required to confirm this activity.
AD drugs are required to enter the blood-brain barrier (BBB) to achieve therapeutic levels in the central nervous system (CNS). Pharmacokinetic parameter prediction study indicated that eckol penetrates the CNS moderately [22]. In addition, Kwak et al. reported that dieckol, with a number of hydroxyl groups and high molecular weight, effectively passed the BBB in rats upon intravenous injection [53]. Studies of PFFA in BBB permeability has not been implemented, but properties similar to dieckol may be anticipated. To strengthen the penetration property of BBB, several methods were developed using nanoparticles, aromatic substances (e.g., borneol), and chemical drug delivery systems [54]. Recently, Venkatesan et al. successfully biosynthesized the silver nanoparticles using E. cava [55]. Therefore, we are able to overcome the limitation of phlorotannins over BBB penetration.
We are the first to identify the inhibitory activity of phlorotannins as dual Aβ aggregation and insulin glycation inhibitors, and our computational study clearly shows the mechanism by which PFFA inhibits Aβ self-aggregation and bovine insulin glycation. However, further in vivo experiments are needed to verify this in vitro and in silico prediction.
In conclusion, our results show that dieckol and PFFA derived from marine brown algae strongly reduce Aβ 25-35 self-aggregation and non-enzymatic insulin glycation. In addition, we used docking and MD simulation studies to demonstrate the molecular mechanism by which the active phlorotannins inhibit Aβ aggregation and insulin glycation. Therefore, those phlorotannins can prevent neuronal damage by inhibiting the formation of β-sheet rich amyloid peptide structures and insulin glycation as well as by preventing lipid peroxidation, producing normal insulin and Aβ processing pathways. Taken together, our findings suggest that phlorotannins could be a promising therapeutic lead compound for the treatment of AD and T2D.

Preparation of Phlorotannins
Five phlorotannins-phloroglucinol, eckol, dioxinodehydroeckol, dieckol, and PFFA-were isolated from the ethyl acetate fraction of E. stolonifera ethanolic extract as described by Yoon et al. [23]. The cemical structures of the isolaed phlorotannins are shown in Figure 1.

Assay for Non-Enzymatic Insulin Glycation
The insulin from bovine pancreas was dissolved in third grade distilled water to a concentration of 6 mg/mL and acidified to pH 2.4 using phosphoric acid to produce monomeric insulin. Then the insulin solution was neutralized to pH 7.0 using 10 M NaOH. Non-enzymatic glycation of insulin was initiated by mixing the insulin solution, 0.5 M d-ribose (or 1.5 M d-glucose) in 50 mM NaH 2 PO 4 buffer (pH 7.0), and 10% dimethyl sulfoxide (DMSO) or the test phlorotannins in a 1:8:1 ratio and incubating the mixture at 37 • C for 6 days (2 weeks for d-glucose induced insulin glycation). Vanillin was used as a negative control for d-ribose induced insulin glycation [27], whereas rutin was used as a positive control for d-glucose induced insulin glycation [28]. After incubation, the reaction mixture was measured at an excitation wavelength of 320 nm and emission wavelength of 410 nm using a fluorescence microplate reader (Gemini XPS).

Preparation of Rat Brain Homogenates
Whole rat brain homogenates were prepared from freshly killed Sprague Dawley rats (male, 6-months old) and provided by the Aging Tissue Bank of Pusan National University. One g of whole rat brain was homogenized in 10 mL of cold 20 mM sodium phosphate buffer containing 140 mM KCl (pH 7.4) and centrifuged at 1300× g for 15 min.

Lipid Peroxidation Assay
A TBA reactive species (TBARS) assay was used to evaluate the antioxidant activity of the phlorotannins in the rat brain homogenates [57]. An aliquot of the supernatant fraction of the homogenates was mixed with phlorotannins dissolved in 10% DMSO, freshly prepared ferric sulfate (250 µM), and distilled water in a 10:5:3:12 ratio, and then the reaction mixture was incubated at 37 • C for 1 h. The reaction was completed by adding a color reagent containing 0.8% TBA, 8.1% SDS, and a 7.5% (final concentration) acetic acid-NaOH solution (pH 3.4) in a 2:1:2 ratios. The mixture was boiled in a water bath for 1 h. After cooling, the reaction mixture was mixed with an equal volume of n-butanol and centrifuged at 1300× g for 10 min. The absorbance of the upper layer was measured at 532 nm using a spectrophotometer (Molecular Devices). The formation of TBARS was expressed as MDA nmol/mg protein using 1,1,3,3-tetramethoxypropane as a standard. Trolox was used as a positive control.

Statistical Analysis
The 50% inhibitory concentration (IC 50 ) values (µM) obtained from the dose-inhibition curves are expressed as the mean ± SD (n = 3). The Student's t-test (two-tailed) was used to determine the significant differences between the blank and control or the phlorotannin-treated groups and control in Figures 2 and 3.