1. Introduction
Free radicals are unstable molecules containing at least one unpaired electron in the external orbital [
1]. However, in the 1950s it was discovered that free radicals play a significant role in many pathological and aging processes [
2]. The redox balance is the key factor of all processes occurring in both normal and pathological cells because reactive oxygen species (ROS) could imitate some molecular messengers and severely disturb normal metabolic pathways. The broken oxidation-reduction equilibrium caused by either the excessive production of ROSs or inadequate activity of the endogenous antioxidant protective system inevitably results in the oxidative stress [
3].
There are some exogenous naturally occurring molecules, mainly nutrients, possessing the antioxidant activity. They include carotenoids, vitamins E and K, ascorbic acid, flavonoids and phenol carbonic acids, zinc, selenium and many other chemicals. Humic acids (HAs) also could be considered as the effective antioxidants; their high potential to reduce active free radicals has been reported by many research groups [
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17]. According to the commonly accepted opinion, HAs are naturally occurring high- and super-molecular polyfunctional colloidal bioorganic molecules featuring both amphoteric and amphipathic properties. The antioxidant activity of HAs could be attributed to the abundance of phenolic hydroxyl, quinoid and other chemical groups having a highly delocalized molecular orbital. As a result, HAs can donate protons, catch free radicals and chelate reactive ions under normal physiological conditions. Moreover, being potent antioxidant and free radical catcher substances, HAs do not display any specific toxic activity toward cells, tissues, or organisms [
18,
19,
20,
21]. Owing to this fact, humic substances are a promising natural source of new raw materials for food and pharmaceutical industries [
12,
22,
23]. Though very promising as a potential new drug the chemical structure of HAs has not yet been identified and the quantitative structure-activity relationship (QSAR) remains unmodeled. The main reason for that is the exacerbated chemical complexity of HAs. One more important factor is that the samples of HAs obtained from different sources have quite a remarkable variation in their physical and chemical properties and are accompanied by nonlinear relationships between these parameters and biological activities. As a result, typical statistical models, such as a multiple linear regression, are not relevant enough to produce a practically significant prognosis.
This fact severely limits the possibility of focused research and creation of new potent antioxidants and cell protectors based on HAs. These obstacles, at least in theory, could be overcome through application of artificial neural networks (ANNs) due to their potential to approximate a very wide range of functions, including nonlinear QSARs of a moderate complexity. With some reasonable precautions, ANNs could be considered as a powerful framework to model complex biopharmaceutical systems [
24,
25]. ANNs successfully predicted antioxidant activity of essential oils [
26], beetroot extracts [
27], teas [
28], and phenolic compounds extracted from strawberries [
29]. In addition, neural networks effectively complement common analytic methods, enhancing data processing protocols when it comes to characterization of very complex biochemical systems [
30].
Recently proposed state-of-the-art machine learning techniques, for example, the self-organizing feature map (SOFM), made it possible to find the collective and individual characteristics of HAs obtained from different geographic locations [
31]. One more application of ANNs is to simulate physical and chemical properties of supramolecular structures. A backprojection neural network trained on experimental UV absorption spectra demonstrated high accuracy predicting optical parameters during coagulation of HAs [
32]. Finally, artificial neural networks are among the best tools to optimize parameters of chemical processes that consider humic acids [
33]. Unfortunately, ANNs have several inherent drawbacks that significantly restrain their use as an analytical tool [
34]. Primarily, ANNs tend to over-fit experimental data, which reduces the overall accuracy of the model. One more unwanted phenomenon is lack of transparency and ambiguous structure of the artificial networks. As a result, ANN-based QSAR models in many cases cannot be interpreted in physical, chemical and biological terms. These disadvantages become even more important when one studies relationships between the structure of complex macromolecules and their biological activities.
Recently we reported our results on a new approach to modeling immunotropic effects of humic acids extracted from peat [
35]. The proposed ANN model uses spectral parameters of the samples in visible and near-ultraviolet range to estimate the production of nitric oxide by activated peritoneal macrophages. This approach is useful for screening and preliminary selection of the samples because it requires only basic measurements of the absorbance spectrum.
The limitations of ANN-based QSAR models are becoming increasingly more evident as the complexity of the structure-activity relationships increases. There are three major drawbacks that must be overcome to build an adequate ANN model of a complex system with sufficient non-linear dependencies between inputs and output. First, the input variables have to be carefully selected to provide distinguishability of the samples. Second, both training and testing datasets must be representative, meaning their size and variation of values. Last, but not least, the structure of the artificial neural network must be defined before any training and testing runs. In the case of complex multistage processes, for example, antioxidant activity, ANN models could produce practically meaningless results. We believe that this is the main reason why this approach is not the mainstream modeling tool used to describe complex biological processes.
Counterintuitive enough, but simpler methods, for instance, multiple regression occasionally work better and produce meaningful interpretable results. The first systematic study of antioxidant activity of humic acids having different origin and composition was reported in [
7]. The major components of humic acids with the highest antioxidant activity have been identified and described as a linear multiple regression model with only four independent variables: atomic C/N ratio, content of O-substituted methine and metoxy groups and total content of phenols.
One more successful example of the regression-based QSAR model was published in [
36]. The authors built a regression model that involves NMR parameters of humic-like substances extracted from lignin in agro-industrial byproducts. Similar results were reported in [
37], where a principal components model was implemented. This type of model is also relatively simple and features ability to unveil the structure of the underlying parameters. Research has also been performed on potential antiviral activity of humic substances; for example, [
38] reported dependency of the antiviral activity on parameters measured with NMR and mass spectrometry.
Summarizing the published results, it can be concluded that although artificial neural networks are a very powerful tool to model a quantitative structure-activity relationship, they have inherent disadvantages limiting their ability to describe complex multistage biological processes. The simpler models based on a multiple regression or linear vector decomposition occasionally produce more robust and self-explainable predictions.
This paper focuses on new approach to model antioxidant and cell protective activity of humic acids extracted from peat. We propose the so-called ontology-based model that allows integration of heterogeneous models into a single framework [
39]. The ontology-based modeling is a reliable and effective approach that finds application in many research areas, for example, the Gene Ontology Consortium (
http://geneontology.org, accessed on 30 May 2022). There have been no publications found on building QSAR models using ontology-based principles and methods. Here we introduce this approach to model antioxidant and cell protective activity of humic acids extracted from peat.
2. Materials and Methods
2.1. Materials
Humic acids were extracted from nine samples of minerotrophic, mesotrophic and oligotrophic peat types, using basic (NaOH) and pyrophosphate (Na4P2O7) extractions were designated as HAalk and HApyr, respectively.
The nine representative types of peat [
40] were taken from major peat bogs in the Tomsk region, Russia: the oligotrophic bog located in the southern taiga zone between the Iksa and the Bakchar rivers, representing the north-eastern spurs of the Great Vasyugan Mire (56°58′ N latitude and 82°36′ W longitude; peat samples 1–4, 6, 7, and 9), and the eutrophic bogs of Klyukvennoe (56°23′ N latitude and 84°42′ W longitude; peat sample 5) and Tagan (56°21′ N latitude and 84°48′ W longitude; peat sample 8). Different peat profiles in the bogs were chosen, and samples were taking from different depths of the peat sections (
Table 1).
The degree of decomposition of the peat samples was estimated microscopically (at magnification ×56–140) as a ratio of groundmass to volumetric amount of tissues contained in the peat samples. The HAs extracts were prepared following the standard procedure. Firstly, the peat samples were desiccated at room temperature, milled, and incubated with 0.1 M Na4P2O7 (pyrophosphate extraction) or 0.1 M NaOH (basic extraction) for 5–8 h at 30–50 °C under continuous mixing. Then, the precipitate was filtered out and the remaining solution was treated with HCl at pH 1.0–2.0.
Finally, the mixture was centrifuged, washed on the filter with water to increase pH up to 7.0, and dried at room temperature. Yield of HAs from the peat samples was measured by the gravimetric method. The yield of the alkaline extraction is 1.5–3 times higher in comparison to the pyrophosphate (Na4P2O7)-based method. The samples of peat-derived HAs are amorphous dark brown odorless powder.
2.2. Characterization of Humic Acids
The optical properties of HAs were studied with visual light spectroscopy of the aqueous solutions (0.001% mass) in quartz cuvettes (1 cm) using an Unico 2800 spectrophotometer (UNICO, Dayton, NJ, USA). The optical absorbance at 465 nm (A
465) and 650 nm (A
650) were measured, then the A
465/A
650 ratio was calculated (
Table S1).
The infrared (IR) spectra of HAs mixed with KBr in the proportion 1:100 were registered in the IR region from 500 to 4000 cm
−1 with IR-Fourier spectrometer FSM 2201 (Infraspek Ltd., Saint-Petersburg, Russia). The IR spectrum does not feature specific peaks and is highly variable among the samples. The averaged spectrum is given in
Figure 1;
Table S2 contains detailed IR-related parameters for each sample. The spectral parameters were selected following the method described in [
41]; relative quantity of HAs functional groups in the peat samples was estimated through optical absorption in spectral ranges associated with oxygen-containing hydroxyl groups (υ
OH, 3400 cm
−1), carbonyl groups (υ
C=O, 1720 cm
−1), ester (υ
C-O, C-O-C, 1225 cm
−1), and alkyloxy groups (υ
C-O, 1035 cm
−1), divided by optical absorption of aromatic (υ
C=C, 1610 cm
−1) and aliphatic (υ
Aliphatic, 2920 cm
−1) fragments of the HA molecular structure.
To measure total acidity, the HAs samples were treated with a Ba(OH)
2 solution under N
2 atmosphere for 24 h. The Ba(OH)
2 remaining in the solution after the reaction was then back-titrated with a standard acid solution. For the titration of carboxylic acid groups, the HA samples were treated for 24 h with calcium acetate solution in excess, which causes the release of acetic acid. The CH
3COOH released was then titrated with a standard base solution. Phenolic OH groups were calculated as the difference between total acidity and acidity of the carboxylic groups (
Table S1).
The elemental (C, H, N, O, S) composition of HAs was determined by combustion using CHNS Flash 2000 (Thermo Fisher Scientific, Cambridge, UK); the O
2 content was calculated as the difference (
Table S1).
The quantitative solution-state
13C-NMR spectrum of HAs was recorded with a Bruker NMR spectrometer AVANCE 400 (400 MHz, Bruker, Billerica, MA, USA) operating at 100 MHz; the mode was INVGATE; the pulse sequence was CPMG; the first sequence pulse was 90°; the registration time of the free induction decay signal was 0.2 s; the relaxation delay time was 7.8 s, and the recording duration of one
13C–NMR spectrum was about 12 h [
42]. The spectra obtained were quantified and the assignments were made (in ppm): 0–48 aliphatic H and C substituted C atoms (C
CHn); 48–59 methoxyl C atoms (C
CH3O); 59–66 O-substituted methylene groups (C
CH2O); 66–91 O-substituted methine groups (C
CHO); 91–108 anomeric double substituted aliphatic C atoms (C
OCO); 108–145 aromatic H- and C-substituted atoms (C
Ar); 145–168 aromatic O-substituted C atoms (C
Ar-O); 168–189 C atoms of carboxylic and esteric groups (C
COO); and 189–220 C atoms of quinone and ketone groups (C
C=O). In addition to the integrals of the given ranges, the sum of O-substituted aliphatic C ΣC
Alk-O, carbohydrate carbon ΣC
Carb and the ratio ΣC
Ar/ΣC
Alk was calculated. The value of ΣC
Alk-O was a sum of C
OCHO+C
CHO+C
CH2O+C
CH3O; ΣC
Carb was a sum of C
OCO, C
CHO, and C
CH2O; and ΣC
Ar and ΣC
Alk were the sums C
Ar+C
Ar–O and ΣC
Alk-O+C
CHn, respectively (
Table S2).
The molecular weight distribution of HAs was determined by HP-SEC using an Ultrahydrogel 250 column (300 × 7.8 mm, 6 µm, pore size 250 Å; Waters, Milford, MA, USA). The mobile phase was 0.1 M Tris-HCl, pH 8.9 (1 mL/min). A Dionex Ultimate 3000 chromatograph (Thermo Fisher Scientific/Dionex, Waltham, MA, USA) with a vacuum degasser was used, LPG-3400SD pump, column thermostat TCC-3000SD, and a spectrophotometric detector DAD-3000 operating at 240 nm. The molecular weights of the fractions were estimated by comparison with the retention times of polystyrene sulfonate standards (PSS Polymer Standards Service GmbH, Mainz, Germany) (
Table S1).
The excitation-emission fluorescent (EMM) spectrum of the samples was recorded with a PerkinElmer 6500 fluorescence spectrometer using standard 10 mm cuvettes for fluorescence measurements. The emission spectra were in range from 200 to 500 nm, step 1 nm. The excitation spectra were from 200 to 475 nm, step 25 nm (
Figure S1).
All spectra were processed using the staRdom package (
https://github.com/MatthiasPucher/staRdom, accessed on 30 May 2022) in the statistical calculation environment Rstudio (version 2021.09.0) statistical calculation environment (
https://www.rstudio.com, accessed on 30 May 2022). As a preprocessing step, first and second order Rayleigh and first order Raman scatter were removed, after which the spectra were normalized.
2.3. Free Radical Scavenging Activity of Humic Acids
The antioxidant activity of HAs was determined using assay of 2,2-diphenyl-1-picryl-hydrazyl-hydrate (DPPH) radical-scavenging capacity test. The bleaching reaction was registered with a Unico 2800 (UNICO, Dayton, NJ, USA) spectrophotometer at 520 nm. The scavenging activity was expressed as the percent of the inhibited radicals (
Table S3).
Additionally, the antioxidant potential was measured as the electrochemical reduction current of O
2 at a mercury film electrode after reaction of the compounds with O
2− in phosphate buffer solution with pH = 6.9 [
43]. The HAs were added to the electrochemical cell at concentration of 1 µg/mL and stirred. Voltammograms of O
2 cathodic reduction were recorded using linear sweep voltammetry with potential scan rate 30 mV/s and potential range from E = 0 to −0.7 V. The electrochemical cell included a working mercury film electrode and a silver/silver chloride reference electrode in saturated KCl (Ag|AgCl|KClsat) solution; HAs did not adhere to the surface of the working mercury film electrode in the range of O
2 reduction [
43]. Test compounds reacted with ROS and changed the electrochemical reduction current of O
2 (first wave at E = −0.3 V). The voltammograms were used to plot time dependences of the function [1 −
I/
I0] in the presence of a test sample. The linear part of the plot and the slope ratio of the tangent to this portion of the curve were used to calculate the kinetic criterion of antioxidant activity, K (μmol/L × min
−1) (
Table S3).
Total antioxidant capacity of HAs was evaluated using ABTS assays and a Unico 2800 (USA) spectrophotometer. Absorption was measured at 734 nm. HAs interact with a stable free radical cation ABTS (diammonium salt of 2,2′-azino-di-(3-ethylbenzthiazolinsulfonic acid) reducing its content in the reaction mixture [
44]. The radical scavenging activity was expressed as the IC
50—the concentration of HAs, at which the concentration of ABTS decreased by 2 times. “Trolox” (Acros Organics, Bratislava, Slovakia) was used as a positive control (
Table S3).
The concentration of semiquinone-type free radicals of HAs was estimated through the number of paramagnetic centers (PMC) by EPR spectroscopy using a Bruker EMX EPR spectrometer (EPR Division, Bruker Instruments, Inc., Billerica, MA, USA) at room temperature (~23 °C). The absolute concentration of the unpaired spins was calculated against a standard CuCl
2·5H
2O solution containing a known number of PMC (
Table S3).
The interaction of HAs with the superoxide-anion radical (O
2−) was studied through measuring the concentration of O
2−. To maintain stable concentration of O
2−·a non−enzymatic O
2−·generation system was used [
45], where electrons with NADH+H+ are transferred via phenazine metasulfate (FMS) to molecular oxygen forming superoxide that reduces nitro blue tetrazolium (NST) into formazan [
46]. A Unico 2800 (USA) spectrophotometer was used to measure absorption at 560 nm. The radical scavenging activity was expressed as the indicator IC
50–the concentration of HAs, at which the concentration of NST decreased by 2 times. Ascorbic acid (Sigma Aldrich, Merck Life Science LLC, Moscow, Russia) was a positive control (
Table S3).
The specific iron chelating activity was estimated through reaction between HAs and ferrosine-Fe
2+ complex [
46] detecting optical absorption with a Unico 2800 (USA) spectrophotometer at 562 nm. This method evaluates the ability of HAs to bind Fe
2+ ions, thereby suppressing lipid peroxidation (POL) and a lipid oxidation process caused by free radicals. Ferrozine (monosodium salt of 3-(2-pyridyl)-5,6-diphenyl-1,2,4-triazine-p,p′-disulfonic acid) forms a purple complex with Fe
2+ ions. The iron chelating activity was expressed as the indicator IC
50—the concentration of HAs, at which the ferrozine-Fe
2+ complex in a model system decreased by 2 times. Ethylenediaminetetraacetic acid (EDTA) (Sigma Aldrich, Merck Life Science LLC, Moscow, Russia) was a positive control (
Table S3).
Hydroxyl radical (HO·) is a very powerful oxidizing agent that reacts with almost all biomolecules found in living cells. HO generation was carried out during the Haber–Weiss reaction in the presence of deoxyribose. During this reaction deoxyribose was degraded to malondialdehyde (MDA). The latter was determined by the reaction with thiobarbituric acid (TBA), which at high temperature and acidic pH proceeds with the formation of a colored trimethine complex with an absorption maximum at a wavelength of 532 nm [
46]. Owing to the high chelating activity of HA samples, it is necessary to take into account the possibility of Fe
3+ binding by HA molecules, which can lead to a decrease in the concentration of the hydroxyl radical. It is known that EDTA binds Fe
3+ ions into a complex that is capable of generating HO [
46], so the ability of HA to bind HO• was studied in a model system with and without EDTA. The samples contained 20 mM KH
2PO
4-KOH buffer (pH 7.4), 0.1 mM ascorbic acid, 2.8 mM deoxyribose, 1 mM H
2O
2 and 0.1 mM FeCl
3 or 0.1 mM FeCl
3 and 1 mM EDTA, pre-mixed in equal volumes. In the experimental samples solutions HAs samples were added at final concentrations: 0.5, 1, 1.5, and 2 mg/mL. The control and experimental samples were incubated for 1 h at 37 °C, after which 1 mL of 0.5% TBA and 1 mL of 10% TCA (Trichloroacetic acid) were added to 0.5 mL of the reaction medium, placed in a boiling water bath for 15 min, cooled, and centrifuged. The optical density of the supernatant was measured using an SF2000 spectrophotometer (OKB Spectr LLC, Saint-Petersburg, Russia) at a wavelength of 532 nm. Based on the dose–response curve, the concentration of the HA sample at which 50% inhibition of MDA formation from deoxyribose was calculated. Mannitol (Thermo Scientific Acros, Loughborough, UK), a classical trap of hydroxyl radicals, was used as a standard (
Table S1).
2.4. Cytotoxicity Study
The 3T3-L1 fibroblasts cell line was obtained from the State Scientific Center of Virology and Biotechnology “Vector”, Novosibirsk, Russia. The effect of HAs on the viability of 3T3-L1 normal fibroblast cell line was assessed using a neutral red test as described [
47].
The 3T3-L1 cells cultivation was under standard conditions (5% CO2 atmosphere, DMEM/F-12 medium (GibcoTM, Billings, MT, USA) + 10% FBS (GibcoTM, Billings, MT, USA) + 2 mM L-glutamine (GibcoTM, Billings, MT, USA) and 1% gentamicin (GibcoTM, Billings, MT, USA). Aqueous solutions of the studied HAs samples were added in the concentration range 5.5–700 µg/mL. Cell plates were placed into a CO2 incubator for 24 h. After washing cells with 1 × PBS, 40 μg/mL neutral red working solution was added into wells for 2 h at 37 °C. To extract the dye, 150 μl mixture of 96% ethanol: deionized water: glacial acetic acid (50:49:1) was used. Optical density was measured at 540 nm and a reference wavelength of 650 nm using a multifunctional microplate reader Tecan Infinite 200 pro mplex (Tecan Group Ltd, Männedorf, Switzerland).
2.5. Intracellular Humic Acids Distribution Assay
The assay was conducted as described with minor modifications [
48]. 3T3-L1 fibroblasts cells were cultured in complete medium (DMEM/F-12 (Gibco
TM, Billings, MT, USA) +10% FBS (Gibco
TM, Billings, MT, USA) + 2 mM L-glutamine (Gibco
TM, Billings, MT, USA) and 1% gentamicin (Gibco
TM, Billings, MT, USA) for at least 3 passages. After seeding the 24 well plate (0.05 × 10
6 cells per well) and achieving 70–80% confluence, aqueous solutions of the HA samples were added at a final concentration of 25 μg/mL. After 24 h of incubation, cells were transferred to separate Eppendorfs, pelleted by centrifugation (2000 rpm, 5 min), and smeared on Cytoslide coated slides using a Thermo Scientific Cytospin 4 cytocentrifuge (Thermo Scientific, Waltham, MA, USA) (800 rpm, 3 min). Cells were fixated by 4% paraformaldehyde solution, permeabilized with 0.01% Triton solution, and covered with DAPI mounting media. Fluorimetric detection of intracellular HA distribution was carried out using a Leica DMi8 fluorescence microscope (Leica Microsystems, Wetzlar, Germany) by intrinsic fluorescence of HAs at λ
ex = 532–558 nm, and λ
em = 570–640 nm. Cell nuclei were visualized using the DAPI channel (λ
ex = 325–375 nm, and λ
em = 435–385 nm).
2.6. The Effect of Humic Acidss on the Action of Prooxidants In Vitro
Assessment of intracellular ROS production was carried out using a 2,7-dichlorodihydrofluoresceindiacetate (DCFDA) fluorescent probe. Intracellular ROS production was induced using two common prooxidants (tert-butyl hydroperoxide (t-BHP) and Fe2+ ions (FeSO4)).
The 3T3-L1 cells were cultured under standard conditions (5% CO2 atmosphere, DMEM/F-12 medium (GibcoTM, Billings, MT, USA) + 10% FBS (GibcoTM, Billings, MT, USA) + 2 mM L-glutamine (GibcoTM, Billings, MT, USA) and 1% gentamicin (GibcoTM, Billings, MT, USA)) and seeded on black 96-well culture plates for fluorescence measuring (1 × 104 cells/well). The HAs (12.5 μM) or Trolox (10 μM) were added to the corresponding wells and incubated for 24 h. Then the cells were washed from the samples, and a working solution of DCFDA (10 μM) (Sigma-Aldrich, USA, D6883) was added to the wells. The plates were incubated in a thermostat for 20 min at 37 °C, washed from DCFDA, and stimulated with prooxidants 25 μM t-BHP or 10 μM FeSO4. After incubation for 60 min at 37 °C, the fluorescence in the wells was determined at λex = 485 nm and λem = 530 nm using a multifunctional microplate reader, Tecan Infinite200 pro mplex (Switzerland).