Investigation of potential descriptors of chemical compounds on prevention of nephrotoxicity via QSAR approach.

Drug-induced nephrotoxicity remains a common problem after exposure to medications and diagnostic agents, which may be heightened in the kidney microenvironment and deteriorate kidney function. In this study, the toxic effects of fourteen marked drugs with the individual chemical structure were evaluated in kidney cells. The quantitative structure-activity relationship (QSAR) approach was employed to investigate the potential structural descriptors of each drug-related to their toxic effects. The most reasonable equation of the QSAR model displayed that the estimated regression coefficients such as the number of ring assemblies, three-membered rings, and six-membered rings were strongly related to toxic effects on renal cells. Meanwhile, the chemical properties of the tested compounds including carbon atoms, bridge bonds, H-bond donors, negative atoms, and rotatable bonds were favored properties and promote the toxic effects on renal cells. Particularly, more numbers of rotatable bonds were positively correlated with strong toxic effects that displayed on the most toxic compound. The useful information discovered from our regression QSAR models may help to identify potential hazardous moiety to avoid nephrotoxicity in renal preventive medicine.

Introduction

Toxic nephropathy is an important topic and relatively general term of kidney damage to cause acute and chronic renal failure by toxic effects of chemicals [1], and the mechanisms of nephrotoxicity include inflammation, tubular cell toxicity, and crystal nephropathy [2]. Nephrotoxicity is frequently induced by a wide spectrum of therapeutic drugs and environmental pollutants [3]. The nephrotoxicity of kidney disease can be classified into two types: Acute kidney injury (AKI) and chronic kidney disease (CKD) [4]. The AKI is characterized by rapid impairment of kidney function and is defined as an increase in serum creatinine concentration and duration ≤ 7 days, leading to an acute decreased glomerular filtration rate (GFR) [5]. Clinically, untreated AKI will lead to CKD and quickly develop to end‐stage renal disease (ESRD) in most cases. The CKD is defined as abnormal kidney function or glomerular filtration rate (GFR) less than 60 mL/min/1.73 m2, albuminuria > 30 mg/24 h, and presence of kidney damage marker for three months [6]. In other cases, the progression of CKD to ESRD is closely related to the accumulation of toxic metabolites in the patient’s blood [7]. The underlying condition such as renal fibrosis, kidney tubular damage, vascular insufficiency, glomerular hypertension, and vascular endothelial cell damage may cause the AKI progression into CKD [8]. Therefore, both AKI and CKD are unifying and can be considered as risk factors for each other.

Recent reports have indicated that drug-induced nephrotoxicity may cause by cumulative dose-dependent toxicity and is recognized as a contributor to kidney injury resulting in AKI and CKD [9]. Clinically, numerous drugs are used to treat multiple diseases that have the nephrotoxic impact on the nephron directly [10], the common problems include renal dysfunction [11], fluid-electrolyte disorders [12], and sepsis [13] after exposure to nephrotoxic drugs during therapy. The nephrotoxic drugs can cause damage to tubules and nephrons by entering proximal tubular cells through cellular uptake or apical drug transport [14]. Drug-induced kidney disease (DIKD) is an origin of kidney disease that constitutes a major cause of AKI and CKD in current clinical practice [15]. Several prospective studies of AKI have demonstrated that approximately 14–26% of critically ill patients with AKI diagnoses were treated with nephrotoxic drugs [16], [17]. Therefore, the prevention of drug-related nephrotoxicity before starting the treatment is an important issue to avoid or diminish renal toxicity. Herein, we focused on toxicity assessment to investigate the chemical information about toxic substances that cause adverse effects on kidney cells.

In consideration of unethical animal procedures, alternatives to animal testing could be used to avoid the drawbacks of animal experiments [18]. According to the 3Rs principle published by Russell and Burch, the efforts to reduce, refine and replace animal tests were developed in recent years to perform more humane animal research [19]. In order to reduce the number of experimental animals, various alternative methods of experimental animals have been suggested including computer-based models [20], cells and tissue cultures [21], invertebrates [22], alternative organisms [23], and microorganisms [24]. The aforementioned methods provide good strategies to overcome the major concern of ethics. Besides, the advantages associated with the alternative methods in the assessment of hazardous chemicals are cost-effectiveness, time efficiency, and less ethical concern [25]. In some cases, combining in silico and in vitro approaches have the potential to estimate the likelihood of hazardous compounds statistically [26], [27], [28]. Therefore, we focused on the relationship between the toxin structures and their nephrotoxicity in this study. The quantitative structure–activity relationship (QSAR) is one of the computer-based drug design methods employed in pharmaceutical sciences and lead optimization [29], which is also called indirect drug designing in the field of computational modeling [30]. The QSAR can provide predictive models based on mathematical methods that can be employed to study the design of known types of chemical compounds to reach an improved activity [31]. In general, the QSAR techniques were regarded as a cost-effective tool in precaution of the toxicants for human health and environmental safety [32], [33]. However, there are still very few studies focused on the nephrotoxic using in silico model such as QSAR modeling in predicting human nephrotoxicity for drug development [34]. To identify for investigation of therapeutic drugs with non-nephrotoxic, we aim to investigate the relationship between chemical structure and its nephrotoxicity.

Here, we explored the potential of developing predictive QSAR models for a series of toxic chemicals acting as inhibitors of renal cells. A series of fourteen different compounds were evaluated for the activity on the HK-2 cell line. Meanwhile, identification of the potential chemical addends on the chemicals was investigated using QSAR models. The QSAR models were constructed with the GA algorithm to construct the relationship between structural changes and nephrotoxicity, which can be applied to predict the nephrotoxicity for new chemical compounds and provide a new guideline for the development of non-nephrotoxic drugs. The information from the experimental data and equations of predictive computer models might guide the potential functional groups on chemicals that are related to toxic nephropathies of kidney cells.

Materials and methods

Cell cultures and test compounds

The human kidney proximal tubular epithelial cell line HK-2 was purchased from American Type Culture Collection (CRL2190). Cells were cultured keratinocyte serum-free (KCSF) medium containing bovine pituitary extract (40 μg/mL) and recombinant epidermal growth factor (5 ng/mL) (Gibco BRL, Grand Island, NY, USA). The cultured cells were maintained in a humidified 5% CO2 chamber at 37℃. The cultured medium replacement was performed two times every week. A density around 1 × 105 for HK-2 cell lines was respectively seeded in each well of 96-well plates then maintained in the humidified incubator for 24 h before treatment with test compounds. All dissolve solutions of Thapsigargin, 4-Deoxy Nivalenol, Ochratoxin A, Ochratoxin B, Fusarenon X, 15-O-Acetyl-4-deoxynivalenol, Cinacalcet, Tacrolimus, Mitomycin C, Pantoprazole (Cayman Chemical, Ann Arbor, MI, USA), Aristolochic acid I (Sigma-Aldrich Corp., St. Louis, MO, USA), Cisplatin, Amphotericin B, and Cyclosporine A (MedChemExpress, Monmouth Junction, NJ, USA) were prepared before cell viability assay.

Cell viability assay

Sulforhodamine B (SRB) assay was used to evaluate the biological activity of HK-2 for test compounds. After 24 h incubation time with each test compound in 96-well plates, cells were washed with Phosphate buffered saline (PBS) solution two times and then fixed with a trichloroacetic acid solution for 1 h. The supernatant was removed, and each well was washed two times before SRB staining (Sigma-Aldrich Corp.) After 1 h staining, the residual dye was removed and then washed two times with 1% acetic acid. Subsequently, Tris-buffer (20 mM) was added to each well and then the absorbance of the colored solution was measured at a test wavelength of 562 nm via an absorbance microplate reader (Molecular Devices, Sunnyvale, CA, USA).

The percentage of cell viability was calculated according to the following equation:where AbsSample indicates the optical density of cells with tested compounds, AbsControl means the optical density of control cells, and AbsBlank is the absorbance of PBS.

Construction of QSAR models

A set of 14 compounds with measured bioactivity data (IC50 values) were utilized for the QSAR model generation. The 2D structure of each molecule was created using BIOVIA Discovery Studio (BIOVIA, USA), and the molecular properties of the compounds were calculated with the Calculate Molecular Properties module of Discovery Studio software. Before the process of the QSAR model study, all of the IC50 values (in μM) were converted to pIC50 values using eq 2:

The genetic function approximation (GFA) algorithm was used to generate equations of the QSAR model, which contains molecular descriptors correlating with the activity values. The GFA algorithm was employed to select suitable properties related to the activity value. The detailed settings of QSAR model generation were displayed in Table S1.

Statistical analysis

The experimental data were displayed as the means ± standard deviation (SD), and the differences between each group were measured by one-way analysis of variance. Each experimental data was performed independently three times for reproducibility. The value of cross-validation (q2) to evaluate the accuracy of each QSAR model was measured by the following equation:where PRESS indicated the predicted residual error sum of squares, and the TSS means the total sum of squares of the differences.

Results

Cell viability

The molecular structures of fourteen chemical compounds used to determine the biological inhibitory effects for HK-2 cells are shown in Fig. 1. The standard deviation (SD) of each measured half-lethal inhibition concentration (IC50) value was estimated from three independent experimental groups via SRB assay. The calculated IC50 values for HK-2 cells treated with the aforementioned fourteen compounds are listed in Table 1. After treatment with the fourteen chemical compounds for 24 h, the cell viability of HK-2 cells decreased in a dose-dependent. Compounds 1 and 5 displayed significant cytotoxicity on HK-2 cells, which have IC50 values of 0.18 μM and 0.35 μM, respectively (Fig. 2). Compounds 6, 10, and 13 are more effective than the other compounds, and their IC50 values have unit digits for HK-2 cells (Fig. 3). Compounds 2, 3, 4, 8, 9, 10, 12, and 14 have slightly inhibitory effects among the tested chemical compounds, and the estimated bioactivity value has micromolar of tens digit. However, the IC50 value of compound 11 is 299.6 μM due to low toxicity (Fig. 3).

Fig. 1

The chemical structure of compounds 1–14 used in cell viability assay on HK-2 cells.

Table 1

The biological activity of HK-2 cells after exposure to various chemicals for 24 h. The cell viability was measured by SRB assay.

Comp.	Name	CAS	IC50 (μM)
1	Thapsigargin	67526–95-8	0.18
2	4-Deoxy Nivalenol	51481–10-8	12.31
3	Ochratoxin A	303–47-9	46.15
4	Ochratoxin B	4825–86-9	37.86
5	Fusarenon X	23255–69-8	0.35
6	15-O-Acetyl-4-deoxynivalenol	88337–96-6	3.77
7	Aristolochic acid I	313–67-7	102.60
8	Cinacalcet	226256–56-0	17.14
9	Tacrolimus	104987–11-3	49.39
10	Mitomycin C	50–07-7	5.04
11	Pantoprazole	102625–70-7	299.60
12	Cisplatin	15663–27-1	25.45
13	Amphotericin B	1397–89-3	3.70
14	Cyclosporine A	59865–13-3	38.27

Fig. 2

The effects of compounds 1–8 on the viability of HK-2 cells after treatment with compounds in dose-dependent manner measured by the SRB assay for 24 h. Data of cell viability were shown as the mean ± SD of three independent experiments. Cell viability % = [(mean optical density of the sample − blank)/ (mean optical density of the control − blank)] × 100%.

Fig. 3

The effects of compounds 9–14 on the viability of HK-2 cells after treatment with compound measured by the SRB assay for 24 h.

QSAR modeling

The measured IC50 values of HK-2 cells were regarded as the activity to generate three different simple linear regression models for the fourteen chemical compounds. The molecular descriptors were generated from the QSAR analyzing tools of BIOVIA Discovery Studio. The potent molecular descriptors of the aforementioned chemicals related to pIC50 values were identified via the GFA algorithm. The chemical structures and the measured IC50 values used for the model generation are listed in Table 2. The equation of the suggested QSAR model is described as follows:

Table 2

Top ten QSAR models generated by genetic function approximation (GFA) algorithm for HK-2 cells and ranked by values of correlation coefficient (r2).

Index	QSAR models	r2	r2 (adj)	q2	p-value
1.	GFATempModel_1 = 3.0506–0.1228 * C_Count + 0.014105 * Num_BridgeBonds + 0.94643 * Num_H_Donors + 0.15129 * Num_NegativeAtoms − 2.7209 * Num_RingAssemblies + 2.7209 * Num_Rings3 + 2.9782 * Num_Rings6 + 0.87064 * Num_RotatableBonds	0.9988	0.9968	0.9582	7.61E-07
2.	GFATempModel_2 = -3.2614–0.12267 * C_Count + 0.14995 * Num_BridgeHeadAtoms + 0.9651 * Num_H_Donors + 0.22799 * Num_NegativeAtoms − 2.5999 * Num_RingAssemblies + 2.6853 * Num_Rings3 + 2.9419 * Num_Rings6 + 0.87257 * Num_RotatableBonds	0.9982	0.9954	0.8957	1.85E-06
3.	GFATempModel_3 = -2.6452–0.10623 * C_Count − 0.045306 * N_Count + 0.92158 * Num_H_Donors + 0.11012 * Num_NegativeAtoms − 2.5817 * Num_RingAssemblies + 2.6126 * Num_Rings3 + 2.799 * Num_Rings6 + 0.8085 * Num_RotatableBonds	0.9977	0.9941	0.8721	3.56E-06
4.	GFATempModel_4 = -3.9427–0.05922 * H_Count − 0.20301 * IsChiral + 0.14684 * Num_BridgeHeadAtoms + 0.9517 * Num_H_Donors − 2.5379 * Num_RingAssemblies + 3.0888 * Num_Rings3 + 2.828 * Num_Rings6 + 0.84963 * Num_RotatableBonds	0.9977	0.9939	0.9703	3.78E-06
5.	GFATempModel_5 = -1.7591–0.14381 * C_Count + 0.97518 * Num_H_Donors − 2.9476 * Num_RingAssemblies + 0.72947 * Num_Rings + 1.7158 * Num_Rings3 − 0.94125 * Num_Rings5 + 2.2146 * Num_Rings6 + 0.82905 * Num_RotatableBonds	0.9976	0.9938	0.8242	3.93E-06
6.	GFATempModel_6 = -4.2701–0.12096 * C_Count + 0.06777 * O_Count + 0.032085 * Num_AromaticBonds + 0.97449 * Num_H_Donors − 2.6918 * Num_RingAssemblies + 3.195 * Num_Rings3 + 3.1125 * Num_Rings6 + 0.88274 * Num_RotatableBonds	0.9976	0.9937	0.8598	4.17E-06
7.	GFATempModel_7 = -3.0312–0.1573 * C_Count − 0.048019 * N_Count + 0.0026228 * Molecular_Mass + 0.95237 * Num_H_Donors − 2.7969 * Num_RingAssemblies + 2.6027 * Num_Rings3 + 3.0261 * Num_Rings6 + 0.84102 * Num_RotatableBonds	0.9975	0.9936	0.9237	4.40E-06
8.	GFATempModel_8 = -3.7494 + 0.14322 * O_Count − 0.0059193 * Molecular_Mass + 0.07357 * Num_AromaticBonds + 0.83293 * Num_H_Donors − 2.0552 * Num_RingAssemblies + 3.4906 * Num_Rings3 + 2.4363 * Num_Rings6 + 0.76509 * Num_RotatableBonds	0.9973	0.993	0.9751	5.36E-06
9.	GFATempModel_9 = -2.8723–0.11324 * C_Count + 0.078083 * O_Count − 0.066154 * HBA_Count + 0.85671 * Num_H_Donors − 2.424 * Num_RingAssemblies + 2.7191 * Num_Rings3 + 2.7878 * Num_Rings6 + 0.81744 * Num_RotatableBonds	0.997	0.9923	0.778	6.81E-06
10.	GFATempModel_10 = -3.2288–0.05648 * H_Count + 0.01213 * Num_BridgeBonds + 0.98205 * Num_H_Donors − 0.084766 * Num_H_Donors_Lipinski − 2.5418 * Num_RingAssemblies + 2.8971 * Num_Rings3 + 2.6477 * Num_Rings6 + 0.79772 * Num_RotatableBonds	0.997	0.9923	0.9578	6.82E-06

The least-squares fitting R2, adjusted R2, and cross-validated correlation coefficient q2 values were three criteria for QSAR model validation. The top 1 equation of the QSAR model, which had R2 of 0.9988, adj R2 of 0.9968, and q2 of 0.9582, were defined as a reasonable model due to the best match between the predicted and experimental activity as displayed in Fig. 4. In order to evaluate the effectiveness of QSAR models constructed by the data set of all test compounds, the top ten QSAR models with acceptable statistics for both training and test sets were displayed in Table S2. To develop an external validation of QSAR models, the dataset was randomly split into training (70% = 10 chemicals) and external (30% = 4 chemicals) subsets [35], [36]. The best reasonable QSAR model contained training and test compounds are plotted in Figure S1.

Fig. 4

Correlation between the actual and predicted activities of HK-2 cells from the recommended QSAR model.

The value of cross-validation q2 beyond 0.6 indicated the individual generated equations of the HK-2 cells with accuracy. The residual errors between the predicted and observed values of all chemical compounds were summarized in Table 3. The data showed that each value of the residual error was below 0.1, which illustrated the recommended QSAR model with high predicted power.

Table 3

The predicted data from the recommended equation of QSAR model.

Comp.	Bioactivity (μM)	Actual value (pIC50)	Predicted value (pIC50)	Residual error
1	0.18	6.74	6.75	0
2	12.31	4.91	4.9	0.01
3	46.15	4.34	4.38	−0.05
4	37.86	4.42	4.38	0.04
5	0.35	6.46	6.4	0.06
6	3.77	5.42	5.45	−0.03
7	102.6	3.99	3.99	0
8	17.14	4.77	4.78	−0.02
9	49.39	4.31	4.33	−0.02
10	5.04	5.3	5.35	−0.06
11	299.6	3.52	3.49	0.04
12	25.45	4.59	4.59	0.01
13	3.7	5.43	5.43	0
14	38.27	4.42	4.41	0.01

Quantitative descriptors of the QSAR equation

As for the best reasonable equation of the toxicity for HK-2 cells, the molecular descriptors of the fourteen chemical compounds relative to bioactivities of HK-2 cells include C_Count, Num_BridgeBonds, Num_H_Donors, Num_NegativeAtoms, Num_RingAssemblies, Num_Rings3, Num_Rings6, and Num_RotatableBonds. In order to investigate relation, the quantitative data of the aforementioned descriptors for the fourteen chemical compounds are computed and displayed in Table 4. Here, C_Count means the number of carbon atoms in the structure of a chemical compound. Num_BridgeBonds indicates bonds in a bridgehead ring system and is defined as how many rings share one bond in common. Num_H_Donors is the number of hydrogen bond donors including heteroatoms (Oxygen, Nitrogen, Sulfur, or Phosphorus) to form hydrogen bonds. Num_NegativeAtoms means the number of atoms with a negative charge. Num_RingAssemblies is defined as the number of fragments remaining when all non-ring bonds are removed from a chemical structure. For instance, the two ring assemblies indicate two benzene rings joined directly by single bonds, while one ring assembly has one or without benzene rings. Num_Rings3 is defined as the number of three-membered rings in one organic compound. Num_Rings6 indicates the number of six-membered rings among the structure of a chemical compound. As for the last descriptor, Num_RotatableBonds is the number of single bonds between heavy atoms that are attached to only hydrogens.

Table 4

The quantitative values of the chemical descriptors from the recommended equation of QSAR model.

Comp.	Activity		Molecular property
	IC50 (μM)	pIC50	C_Count	Num_BridgeBonds	Num_H_Donors	Num_NegativeAtoms	Num_RingAssemblies	Num_Rings3	Num_Rings6	Num_RotatableBonds
1	0.18	6.74	34	0	2	0	1	0	0	17
5	0.35	6.46	17	9	3	0	1	1	2	3
13	3.7	5.43	47	40	11	1	2	0	2	3
6	3.77	5.42	17	9	2	0	1	1	2	3
10	5.04	5.3	15	0	4	0	1	1	1	4
2	12.31	4.91	15	9	3	0	1	1	2	1
8	17.14	4.77	22	0	1	0	2	0	3	7
12	25.45	4.59	0	0	2	2	1	3	0	0
4	37.86	4.42	20	0	2	1	2	0	3	5
14	38.27	4.42	62	0	5	0	1	0	0	15
3	46.15	4.34	20	0	2	1	2	0	3	5
9	49.39	4.31	44	25	3	0	2	0	3	7
7	102.6	3.99	17	0	0	2	1	0	3	3
11	299.6	3.52	16	0	2	0	2	0	2	7

Discussion

Animal ethics and animal consciousness has been considered an important issue in recent years [37]. To avoid unethical procedures, in vitro cell-based methods and in silico approaches are useful strategies of alternatives to minimize the experimental animals used in the assessment of kidney toxicity to reduce costs and time. The QSAR method can provide a useful strategy to identify key features based on the relationship between the Physico-chemical properties of each chemical and their biological activities for the investigation of nephrotoxic. The GA algorithm applied in QSAR studies can generate interpretable equations with numerical estimation to link molecular descriptors to the biological activity, which has been applied in various areas of drug design such as leading drug optimization [38] and predictive toxicology [39].

In comparison with early reports, 3D-QSAR and 4D-QSAR approaches were regarded as useful strategies to construct pharmacophore features, and the most frequently used methods include Comparative Molecular Field Analysis (CoMFA), Comparative Molecular Similarity Indices Analysis (CoMSIA) and GRID/GOLPE program. The regression of the 3D-QSAR approach was applied using a partial least squares (PLS) algorithm to establish the optimal number of descriptors in structure-based pharmacophore modeling under PHASE software [40], [41]. To build the 4D-QSAR model, the electron-conformational genetic algorithm (EC-GA) method was used to identify pharmacophore groups and bioactivity prediction based on electronic structure and conformational parameters [42], [43]. However, 3D or 4D QSAR requires a relatively large and chemically diverse training set, and the predictive ability of QSAR was limited by the determination of quantum chemical calculation in three-dimensional space [44], [45], [46].

In other cases, the QSAR model with the predictive ability provides toxicological guidance on unknown hazardous substances [47]. However, the potential molecular descriptors of the nephrotoxic drugs are still under investigated. Therefore, we investigated fourteen marked drug and their nephrotoxicity to generate reasonable QSAR models for hazardous properties identification in this study. As for the result of QSAR modeling, we analyzed the descriptors suggested from the ten different generated equation models responsible for the toxic effects of HK-2 cell. According to the best reasonable linear regression model of HK-2 cells, the suggested molecular descriptors were C_Count, Num_BridgeBonds, Num_H_Donors, Num_NegativeAtoms, Num_RingAssemblies, Num_Rings3, Num_Rings6, and Num_RotatableBonds. We noted that descriptors such as Num_RingAssemblies, Num_Rings3, and Num_Rings6 with high value of the estimated regression coefficients among the best reasonable equation, and the value of the above three coefficients are − 2.7209, 2.7209, and 2.9782, respectively. Therefore, these three properties were regarded as critical features and be used to further study the difference among all tested compounds.

Comparing the numerical properties of the fourteen compounds, compound 1 had a similar quantitative value of suggested descriptors to compound 14. These two compounds with similar numbers of Num_RingAssemblies, Num_Rings3, and Num_Rings6. However, the compound 1 have more inhibitory effects than compound 14 due to the difference in C_Count, Num_H_Donors, and Num_RotatableBonds descriptors. The compound 1 has less numbers of carbon atoms and H-bond donors than compound 14. In contrast, compounds 1 have more numbers of rotatable bonds than compound 14, suggesting that Num_RotatableBonds descriptors is favored property while C_Count and Num_H_Donors are disfavor properties to increase the toxic effects. In an earlier report, Yinping Shi et al. indicated that molecular weight, molecular polar surface area, AlogP, number of hydrogen bond acceptors, molecular solubility, the number of rotatable bonds, and the number of aromatic rings were critical physical–chemical properties in the identification of nephrotoxic drugs [48]. In comparison with the aforementioned properties, our suggested equation of the QSAR model had identical or similar features such as rotatable bonds, organic rings, and hydrogen bonds. In addition, Yinping Shi et al.’s study demonstrated that chemical hydrogen bonding ability is an important feature responsible for toxicity and bioactivity. Meanwhile, their data indicated that the number of hydrogen bond acceptors was obviously related to nephrotoxicity but the number of hydrogen bond donors was not. Interestingly, the number of hydrogen bond donors was also considered a disfavor property based on the analyzed results. Therefore, our suggested molecular properties might have responsible for nephrotoxic structures.

Note that compounds 9, 11, and 13 each have similar values of Num_RingAssemblies, Num_Rings3, and Num_Rings6 descriptors. For the less toxic effects of compounds 9 and 11, the compound 13 has more numbers of carbon atoms, H-bond donors, and bonds in bridgehead ring systems than compound 11. Interestingly, the compound 13 has one Num_NegativeAtoms descript than compounds 9 and 11, indicating that the negative atom might be the critical property increased significantly the toxic effect for HK-2 cells.

Among our tested drugs, the compounds 2, 5 and 6 have similar moiety on chemical structure. Regarding the compounds with more numbers of C_Count, Num_H_Donors, and Num_RotatableBonds descriptors, compound 5 had more toxic effects on HK-2 cells than compounds 2 and 6. The compound 5 and 6 have seventeen carbon atoms and three rotatable bonds, while compounds 2 has fifteen carbon atoms and one rotatable bond. The finding indicated that a greater number of carbon atoms and rotatable bonds may increase the toxic effect on renal cells.

It should be emphasized that we report the first attempt to link the experimentally revealed toxic effects of marked drugs to the peculiarities of their molecular properties based on QSAR analysis. The QSAR model developed in this study provide very useful information on relationships between the chemical structure of the marked drugs and their toxic effects, which can help to analyze the nephrotoxicity for more therapeutic drug.

Conclusion

To assess kidney toxicity, the equations of QSAR modeling were generated based on fourteen therapeutic drugs and their inhibitory effects discovered critical structural features on the renal cells. The aim of this study is not only employing the in vitro cell-based studies, but also include the in silico methods as supporting tools that used for identification of the potential toxicity from chemical properties. According to the suggested equation of QSAR models, the most toxic compound 1 with more numbers of rotatable bonds had stronger toxicity effects on HK-2 cells. In addition, increase in the number of bridge bond, H-bond donors, and negative atoms in the structures of the test compounds could promote the toxic effect. The current study demonstrated that the current used drug with no common structure can be observe a tendency on nephrotoxic effect based on QSAR modeling, which may help to provide useful information on discovery of hazardous property for renal preventive medicine in the future.

CRediT authorship contribution statement

Hung-Jin Huang: Conceptualization, Methodology, Data curation, Writing – review & editing. Yu-Hsuan Lee: Conceptualization, Methodology. Chu-Lin Chou: Conceptualization, Methodology. Cai-Mei Zheng: Conceptualization, Writing – review & editing, Supervision. Hui-Wen Chiu: Conceptualization, Writing – review & editing, Supervision.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.