(Paes) on Cell Viability and Nrf2 of Hepg2 and 3D-QSAR Studies

toxics

Article Effects of Phthalate Esters (PAEs) on Cell Viability and Nrf2 of HepG2 and 3D-QSAR Studies

Huan Liu 1,2, Huiying Huang 1,2, Xueman Xiao 1,2, Zilin Zhao 1 and Chunhong Liu 1,2,*

1 College of Food Science, South China Agricultural University, Guangzhou 510642, China; [email protected] (H.L.); [email protected] (H.H.); [email protected] (X.X.); [email protected] (Z.Z.) 2 Guangdong Provincial Key Laboratory of Food Quality and Safety, Guangzhou 510642, China * Correspondence: [email protected]

Abstract: Phthalate esters (PAEs) are a widespread environmental pollutant, and their ecological and environmental health risks have gradually attracted attention. To reveal the toxicity characteristics of these compounds, ten PAEs were selected as research objects to establish a cell model. CCK-8

was used to determine cell viability, Western blots were used to determine the content of Nrf2 in HepG2, and the LD50 collected for the 13 PAEs administered to rats. On this basis, 3D-QSAR models of IC50, LD50 and Nrf2 were established. The experimental results showed that as the time of PAEs exposure increased (24, 48 and 72 h), cell viability gradually decreased. The test concentration (62.5 /125/250 µM) of PAEs exposed for 48 h could significantly increase the content of Nrf2, and the 1000 µM PAEs could inhibit the content of Nrf2. The model is relatively stable and predicts well that the introduction of large and hydrophobic groups may significantly affect the toxic effects of PAEs on cells. The present study provided a potential tool for predicting the LD50 and Nrf2 of new PAEs, and provide a reference for the design of new less toxic PAEs in the future. Citation: Liu, H.; Huang, H.; Xiao, X.; Zhao, Z.; Liu, C. Effects of Phthalate Keywords: phthalate esters; Nrf2; 3D-QSAR; HepG2 Esters (PAEs) on Cell Viability and Nrf2 of HepG2 and 3D-QSAR Studies. Toxics 2021, 9, 134. https://doi.org/ 10.3390/toxics9060134 1. Introduction Phthalate esters (PAEs) are mainly used as plasticisers that can be employed in various Academic Editors: Lidia Caporossi chemical products to promote processing and product flexibility [1,2]. Since their first and Maria Marino application in industrial products in the 1920s [3], their global production and consumption have increased to more than 6 million tons [4]. Historically, the use of phthalic acid (2- Received: 7 May 2021 ethylhexyl) ester (DEHP) in polyvinyl chloride (PVC) was very common, and DEHP was Accepted: 3 June 2021 Published: 5 June 2021 later replaced by didecyl phthalate (DIDP) and diisononyl phthalate (DINP) [5]. Low molecular weight phthalates including dibutyl phthalate (DBP), diethyl phthalate (DEP),

Publisher’s Note: MDPI stays neutral and dimethyl phthalate (DMP), were mostly used in personal care products, pesticides, with regard to jurisdictional claims in glue, and paint solvents [6,7]. However, due to the weak binding force between phthalates published maps and institutional affil- and substrates, the PAEs could be gradually released from the product to the environment iations. during production, use, disposal and recycle [8,9], leading to frequent and undesirable human contact. The persistent existence of PAEs in the environment and their toxicity has aroused widespread concern in society. Many experiments have revealed that some PAEs could cause developmental toxicity, reproductive toxicity, hepatotoxicity, neurotoxicity, and even Copyright: © 2021 by the authors. Licensee MDPI, Basel, Switzerland. carcinogenicity, teratogenicity, and mutagenicity [10]. For example, in a study compared This article is an open access article with the workers who were not exposed to high levels of DEHP and DBP, workers exposed distributed under the terms and to high levels DEHP and DBP had significantly lower free testosterone [11]. In another conditions of the Creative Commons study, exposure of pregnant women to various PAEs led to a shortened anal-genital distance Attribution (CC BY) license (https:// in male infants [12]. creativecommons.org/licenses/by/ Hepatotoxicity is also commonly observed as systemic toxicity in rodents exposed to 4.0/). PAEs. The liver is the main target of PAEs because it is the main detoxification site for the

Toxics 2021, 9, 134. https://doi.org/10.3390/toxics9060134 https://www.mdpi.com/journal/toxics Toxics 2021, 9, 134 2 of 16

decomposition of PAEs in the body. It has been reported that some PAEs caused peroxisome proliferation and liver tumors through peroxisome proliferation factor activated receptors (PPARs) [13]. Similarly, DEHP could disrupt the homeostasis of thyroid hormones by inducing liver enzymes, leading to liver edema [14]. In addition, high-fat diets exposed to DEHP can cause various degrees of non-alcoholic fatty liver in mice [15]. Furthermore, hepatotoxicity caused by PAEs is closely related to oxidative stress, and nuclear factor erythroid 2 related factor 2 (NRF2) is a nuclear transcription factor that regulates oxidative stress [16], which plays a crucial role in liver detoxiﬁcation. Therefore, the impact of PAEs on Nrf2 has been directly related to the liver toxicity, with the toxicity of PAEs varying greatly due to their side chains and the type of analogues [17]. Considering the complexity, three-dimensional quantitative structure-activity relationship (3D-QSAR) model was introduced to explore the toxic effects of different PAEs. The ﬁeld of quantitative structure-activity relationship (QSAR) research in medicinal chemistry has been greatly developed since the 1970s for the quantitative studies om the mathematical relationship between the chemical structure and pharmacological activity or other properties of a series of compounds [18]. Therefore, the aim of this study was to develop a 3D-QSAR model based on the investigation of the effects of PEs on Nrf2, IC50 in human hepatocyte (HepG2) cells, in the hope of predicting the toxicity of new analogues and providing a reference for the design of less toxic PAEs in the future.

2. Materials and Methods 2.1. Cell Culture Human hepatocyte HepG2 was purchased from Procell (Wuhan, China) and cultured in MEM (Procell) complete medium supplemented with 10% fetal bovine serum (FBS), 1% penicillin streptomycin (pen-strep:10,000 U/mL) and 1% L-glutamine. Cells were ◦ maintained in a 37 C, 5% CO2, fully humidiﬁed incubator and passed twice weekly.

2.2. Cell Viability To evaluate the cytotoxicity of DEHP and DMP on the cells, HepG2 cells were cultured in a 96-well plate at a density of 3 × 103 cells/well. After 24 h incubation, cells were taken at the logarithmic stage and treated with DEHP, dihexyl phthalate (DHXP), di-n-octyl phthalate (DNOP), dinonyl phthalate (DNP), bis(2-methoxyethyl) phthalate (DMEP), DEP, and DIDP (62.5, 125, 250, 500 and 1000 µM), or DMSO (solvent control, ﬁnal concentration <0.1%) (purchased from Sigma, St. Louis, MO, USA). After 24, 48 and 72 h exposure, the viability was measured using CCK-8 cell viability assay kit (Nanjing Jiancheng Bioengineer- ing Institute, Nanjing, China). After different treatments, 10 µL CCK-8 solution was added into the culture and incubated at 37 ◦C for 3 h. The generated luminescent signal was captured on a Microplate Reader (Multiskan FC, Thermo, Waltham, MA, USA). The results were expressed as percentage of control, and each value was presented as the mean ± SD of at least three independent experiments.

2.3. Western Blot Analysis Cells were inoculated in six-well plates for 24 h and then treated with DEHP, DEP, DMP, DMEP, DHXP, DNOP, DBP or DMSO for 48 h. After treatment, RIPA lysate (50 mM Tris (pH 7.4), 150 mM NaCl, 1% Triton X-100, 1% sodium deoxycholate, 0.1% SDS, (Be- yotime Biotechnology, Haimen, China) was added to lyse for 30 min to extract the total protein. Then, the cell lysate was centrifuged at 12,000× g for 5 min, and the supernatant was collected. Protein concentration in the extract was determined with the Bicinchoninic Acid assay (Beyotime Biotechnology). Equal amount of protein samples (15 g/lane) were subjected to 10% sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and then adsorbed on polyvinylidene ﬂuoride (PVDF) membranes (Millipore Corporation, Bedford, MA, USA), which were sealed by 1.5 h in 5% (w/v) skim milk. Afterwards, the membranes were incubated with primary antibodies against Nrf1(1:2000, abcam, Cam- Toxics 2021, 9, x FOR PEER REVIEW 3 of 17

Toxics 2021, 9, x FOR PEER REVIEW 3 of 17 Toxics 2021, 9, x FOR PEER REVIEW 3 of 17 Toxics 2021, 9, x FOR PEER REVIEW 3 of 17

Toxics 2021, 9, x FOR PEER REVIEWand then adsorbed on polyvinylidene fluoride (PVDF) membranes (Millipore Corpora-3 of 17 tion,and Bedford,then adsorbed MA, USA), on polyvinylidene which were sealed fluoride by 1.5(PVDF) h in 5%membranes (w/v) skim (Millipore milk. Afterwards, Corpora- theandtion, membranes then Bedford, adsorbed MA, were USA),on incubated polyvinylidene which werewith sealedprimaryfluoride by (PVDF)antibodies1.5 h in membranes5% against(w/v) skim Nrf1(1:2000, (Millipore milk. Afterwards, Corpora- abcam, and then adsorbed on polyvinylidene fluoride (PVDF) membranes (Millipore Corpora- Cambridge,tion,the membranes Bedford, United MA, were USA),Kingdom) incubated which and were with tubulin sealed primary al bypha 1.5antibodies (1:6000, h in 5% Affinity (againstw/v) skim Biosciences,Nrf1(1:2000, milk. Afterwards, Chang-abcam, tion,and Bedford, then adsorbed MA, USA), on polyvinylidene which were sealed fluo rideby 1.5 (PVDF) h in 5% membranes (w/v) skim (Millipore milk. Afterwards, Corpora- zhou,theCambridge, membranes China), United followed were Kingdom) incubatedby a horseradish and with tubulin primary peroxidase- alpha antibodies (1:6000, (HRP-) Affinityagainst conjugated Nrf1(1:2000,Biosciences, with goat Chang-abcam, anti- theCambridge,tion, membranes Bedford, United MA,were USA),Kingdom) incubated which and withwere tubulin primarysealed al bypha antibodies 1.5 (1:6000, h in 5% againstAffinity (w/v) skim Nrf1(1:2000,Biosciences, milk. Afterwards, abcam,Chang- Toxics 2021, 9, 134 rabbitzhou, IgGChina), antibody followed (Nanjing by a Jianchenghorseradish Bioeng peroxidase-ineering (HRP-) Institute, conjugated Nanjing, China)with goat 3diluted of 16anti- Cambridge,the membranes United were Kingdom) incubated and with tubulin primary alpha antibodies (1:6000, Affinity against Biosciences, Nrf1(1:2000, Chang- abcam, tozhou,rabbit 1:10,000 China),IgG atantibody room followed temperature (Nanjing by a horseradishJiancheng for 1 h. Finally, Bioeng peroxidase- ineeringthe protein (HRP-) Institute, bands conjugated onNanj theing, membrane with China) goat diluted were anti- zhou,Cambridge, China), Unitedfollowed Kingdom) by a horseradish and tubulin peroxidase- alpha (1:6000, (HRP-) Affinityconjugated Biosciences, with goat Chang- anti- visualizedrabbitto 1:10,000 IgG using antibodyat room ECL temperature (Nanjing western Jiancheng blot for detection 1 h. Finally,Bioeng reagentineering the protein (BioSharp, Institute, bands Technology Nanj on theing, membrane China) Inc., Shang-diluted were rabbitzhou, IgG China), antibody followed (Nanjing by a Jiancheng horseradish Bioeng peroxidase-ineering Institute,(HRP-) conjugated Nanjing, China) with goat diluted anti- hai,tovisualized 1:10,000 China) andatusing room quantified ECL temperature western using blot ImageJfor detection1 h. softwareFinally, reagent the (National protein (BioSharp, Institutes bands Technology on of the Health, membrane Inc., Bethesda, Shang- were torabbit 1:10,000 IgG at antibody room temperature (Nanjing Jiancheng for 1 h. Finally, Bioeng theineering protein Institute, bands on Nanj theing, membrane China) diluted were bridge,MD,visualizedhai, China)USA). UK) and using and tubulin quantified ECL alphawestern using (1:6000, blot ImageJ detection Affinity software Biosciences, reagent (National (BioSharp, Changzhou, Institutes Technology of China), Health, Inc., followed Bethesda, Shang- byvisualized ato horseradish 1:10,000 using at room peroxidase- ECL temperature western (HRP-) blot for conjugateddetection 1 h. Finally, reagent with the goat protein(BioSharp, anti-rabbit bands Technology on IgG the antibody membrane Inc., (Nan-Shang- were hai,MD, China) USA). and quantified using ImageJ software (National Institutes of Health, Bethesda, hai,visualized China) and using quantified ECL western using blotImageJ detection software reagent (National (BioSharp, Institutes Technology of Health, Inc., Bethesda, Shang- jing2.4.MD, Jiancheng D-QSAR USA). Bioengineering Institute, Nanjing, China) diluted to 1:10,000 at room tem- peratureMD,hai, USA). China) for 1 h. and Finally, quantified the protein using ImageJ bands onsoftware the membrane (National were Institutes visualized of Health, using Bethesda, ECL 2.4.1.2.4. D-QSAR CoMFA and CoMSIA for LD50 of PAEs western2.4.MD, D-QSAR blotUSA). detection reagent (BioSharp, Technology Inc., Shanghai, China) and quantified using2.4.2.4.1. D-QSAR ImageJIn CoMFA the softwarequantitative and CoMSIA (National structure-activity for Institutes LD50 of PAEs of rela Health, tionship Bethesda, modeling, MD, USA).the biological activity 2.4.1. CoMFA and CoMSIA for LD50 of PAEs 50 2.4.1.value2.4. D-QSARCoMFA InLD the (collectedquantitative and CoMSIA from structure-activity forPubChem) LD50 of PAEs was rela calculatedtionship inmodeling, mg/kg, asthe shown biological in Table activity 1. In the quantitative structure-activity relationship modeling, the biological activity 2.4.CoMFA D-QSAR and50 CoMSIA methods were used for 3D-QSAR analysis [19], and all operations value2.4.1. CoMFALD (collected and CoMSIA from PubChem)for LD50 of PAEswas calculated in mg/kg, as shown in Table 1. valueIn LDthe50 quantitative(collected from structure-activity PubChem) was rela calculatedtionship modeling, in mg/kg, the as biologicalshown in activityTable 1. 2.4.1.wereCoMFA CoMFA completed and and CoMSIA CoMSIAwith SYBYL2.1.1methods for LD 50were ofsoftware PAEs used for module. 3D-QSAR The analysismolecular [19], dynamics and all operations program valueCoMFA InLD theand50 (collected quantitativeCoMSIA from methods structure-activity PubChem) were used was for relacalculated 3D-QSARtionship in modeling,analysis mg/kg, [19],as the shown and biological all in operations Table activity 1. MinimizewereIn the completed quantitative was used with to optimize structure-activity SYBYL2.1.1 the energysoftware relationship of allmodule. compounds modeling, The molecular to obtain the biological theirdynamics lowest activity program energy CoMFAwerevalue completed LDand50 CoMSIA(collected with methods SYBYL2.1.1from PubChem) were software used was for module. 3D-QSARcalculated The analysisin molecular mg/kg, [19], as dynamics andshown all operations in program Table 1. valueconformation.Minimize LD (collected was In used the fromtooptimizati optimize PubChem)on the process, energy was calculatedtheof all Tripos compounds inforce mg/kg, field, to obtain Powell as shown their energy lowest in Table gradient energy1. wereMinimizeCoMFA completed50 and was CoMSIA used with to SYBYL2.1.1optimize methods the were software energy used of formodule. all 3D-QSAR compounds The analysismolecular to obtain [19], dynamics their and lowest all operationsprogram energy CoMFAandconformation. Gasteiger-Huckel and CoMSIA In the methods optimizati charge wereon usedapplied,process, for 3D-QSARandthe Tripos the termination analysisforce field, [ 19conditions Powell], and all energy and operations max gradient iter- Minimizeconformation.were completed was used In thewith to optimizatioptimize SYBYL2.1.1 theon energyprocess, software of the all module. Triposcompounds forceThe molecularfield,to obtain Powell their dynamics energy lowest gradient programenergy wereationand completed Gasteiger-Huckelconditions with were SYBYL2.1.1 respectively charge were software 0.005 applied, kcal module. /(moland the ×The A)termination and molecular 1000. conditionsOther dynamics parameters and program max were iter- conformation.Minimize was In used the optimizatito optimizeon the process, energy the of allTripos compounds force field, to obtain Powell their energy lowest gradient energy Minimizedefaultandation Gasteiger-Huckel conditions values. was used DMP were to optimize was respectivelycharge used the wereas energythe 0.005applied, template of kcal all and/(mol compoundsmolecule, the × terminationA) and to 1000. obtainall molecules conditions Other their lowestparameters were and superim- energymax were iter- andconformation. Gasteiger-Huckel In the chargeoptimizati wereon applied, process, and the theTripos termination force field, conditions Powell energy and max gradient iter- conformation.posedationdefault conditionsusing values. Inthe theDMP Align were optimization was respectivelyDatabase used as process,method. the 0.005 template thekcalThe Tripos/(mol training molecule, × force A) set and and field, was 1000. all PowellDMP, molecules Other DBP, energy parameters wereDIBP, gradient superim- DHXP, were ationand conditionsGasteiger-Huckel were respectively charge were 0.005 applied, kcal/(mol and the × A) termination and 1000. Otherconditions parameters and max were iter- andDNOP,defaultposed Gasteiger-Huckel using diisooctylvalues. the DMP Align phthalate chargewas Database used (DIOP), were as method.the applied, DEHP, template The DNP, and molecule,training the 1,2-benzenedicarboxylic termination set and was all DMP, molecules conditions DBP, acid,1-decylwere DIBP, and superim- maxDHXP, 2- defaultation conditionsvalues. DMP were was respectively used as the 0.005template kcal /(molmolecule, × A) and and all 1000. molecules Other parameters were superim- were iterationoctylposedDNOP, ester conditionsusing diisooctyl (nDNOP), the Align were phthalate diallyl Database respectively phthalate (DIOP), method. 0.005DEHP, (DAP), kcal/(molThe DNP, and training the1,2-benzenedicarboxylic× test A)set set andwas was 1000.DMP, DMEP, Other DBP, DEP, parametersacid,1-decylDIBP, and DHXP, DIDP 2- poseddefault using values. the AlignDMP wasDatabase used asmethod. the template The training molecule, set andwas all DMP, molecules DBP, DIBP, were superim-DHXP, were(usingDNOP,octyl default ester the diisooctyl method values.(nDNOP), phthalateof DMP randomdiallyl was (DIOP),phthalate assignment used as DEHP, the(DAP), to template DNP,determine and 1,2-benzenedicarboxylicthe molecule, test the set training was and DMEP, allset molecules and DEP, acid,1-decyl test and set). were DIDP Sta- 2- DNOP,posed diisooctylusing the phthalateAlign Database (DIOP), method. DEHP, TheDNP, training 1,2-benzenedicarboxylic set was DMP, DBP, acid,1-decyl DIBP, DHXP, 2- superimposedtisticaloctyl(using ester analysisthe (nDNOP),method using adopted theof diallylrandom Align partial phthalate Database assignment least squares (DAP), method. to determine analysisand The the training test(PLS) the set tr to setwasaining establish was DMEP, set DMP, and theDEP, DBP, testmodel, and set). DIBP, DIDP and Sta- octylDNOP, ester diisooctyl (nDNOP), phthalate diallyl phthalate (DIOP), DEHP,(DAP), DNP,and the 1,2-benzenedicarboxylic test set was DMEP, DEP, acid,1-decyl and DIDP 2- DHXP,then(usingtistical cross-validation DNOP, theanalysis method diisooctyl adopted of was random phthalate partialcarried assignment least (DIOP),out by squares leave-one-out DEHP, to determine analysis DNP, (L 1,2-benzenedicarboxylic(PLS) theOO) tr ainingtomethodology. establish set and the test model, acid,1-set). Sta-and (usingoctyl theester method (nDNOP), of random diallyl phthalateassignment (DAP), to determine and the testthe settraining was DMEP, set and DEP, test set).and DIDPSta- decyltisticalthen 2-octyl cross-validation analysis ester (nDNOP),adopted was partial diallylcarried least phthalate out bysquares leave-one-out (DAP), analysis and the (PLS)(LOO) test to setmethodology. establish was DMEP, the DEP,model, and and tistical(using analysis the method adopted of random partial assignmentleast squares to analysisdetermine (PLS) the trtoaining establish set andthe model,test set). and Sta- Table 1.DIDP Actualthen (using cross-validation value, the predicted method value was of random carriedand error assignmentout of 3D-QSARby leave-one-out to model determine of (LPAEsOO) the lgLD trainingmethodology.50. set and test set). thentistical cross-validation analysis adopted was carriedpartial outleast by squares leave-one-out analysis (L (PLS)OO) methodology. to establish the model, and TableStatistical 1. Actual value, analysis predicted adopted value partial and error least of squares 3D-QSAR analysis model (PLS)of PAEs to lgLD establish50. the model, and Table 1. Actualthen cross-validationvalue, predicted value was andcarried error out of 3D-QSARbyCoMFA leave-one-out model of CoMFA(LPAEsOO) lgLD methodology.50. CoMSIA CoMSIA Names Structures then cross-validation LD50(mg/kg) was carriedExp.(lgLD out50 by) leave-one-out (LOO) methodology. Table 1. Actual value, predicted value and error of 3D-QSARCoMFAPred. model of PAEsCoMFARes. lgLD 50. CoMSIAPred. CoMSIARes. Names Structures LD50(mg/kg) Exp.(lgLD50) CoMFA CoMFA CoMSIA CoMSIA Names StructuresTable 1. Actual value, LD50(mg/kg) predicted valueExp.(lgLD and error50) of 3D-QSARPred. model of PAEsRes. lgLD50. Pred. Res. Table 1. Actual value, predicted value and error of 3D-QSARCoMFAPred. model of PAEsCoMFARes. lgLD 50 . CoMSIAPred. CoMSIARes. Names Structures LD50(mg/kg) Exp.(lgLD50) DMP 6800 3.8325 Pred.3.791CoMFA 0.0415CoMFARes. CoMSIAPred.3.829 0.0035CoMSIARes. Names Structures LD LD50(mg/kg)Exp. Exp.(lgLDCoMFA50) CoMFA CoMSIA CoMSIA Names Structures 50 Pred. Res. Pred. Res. DMP 6800 3.8325 3.791 0.0415 3.829 0.0035 DMP (mg/kg) 6800 (lgLD50) 3.8325Pred. 3.791Res. 0.0415Pred. 3.829 Res. 0.0035 DMP 6800 3.8325 3.791 0.0415 3.829 0.0035

DMP 6800 3.8325 3.791 0.0415 3.829 0.0035 DMPDEP 68008600 3.83253.9345 3.791 3.89 0.04150.0445 3.829 3.942 0.0035−0.0075

DEP 8600 3.9345 3.89 0.0445 3.942 −0.0075 DEP 8600 3.9345 3.89 0.0445 3.942 −0.0075 DEP 8600 3.9345 3.89 0.0445 3.942 −0.0075 DEP 8600 3.9345 3.89 0.0445 3.942 −0.0075 DEP 8600 3.9345 3.89 0.0445 3.942 −0.0075 DBP 7499 3.875 3.955 −0.08 3.896 −0.021

DBP 7499 3.875 3.955 −0.08 3.896 −0.021 DBPDBP 74997499 3.8753.875 3.955 3.955− 0.08−0.08 3.896 3.896 −0.021−0.021 DBP 7499 3.875 3.955 −0.08 3.896 −0.021

DIBPDBP 15,0007499 4.17613.875 4.2413.955 −0.0649−0.08 4.1753.896 0.0011−0.021 Toxics 2021, 9, x FOR PEER REVIEW 4 of 17 ToxicsDIBPDIBP 2021 , 9, x FOR PEER REVIEW 15,00015,000 4.17614.1761 4.241 4.241−0.0649 −0.0649 4.175 4.175 0.00110.00114 of 17 ToxicsDIBP 2021 , 9, x FOR PEER REVIEW 15,000 4.1761 4.241 −0.0649 4.175 0.00114 of 17

DIBP 15,000 4.1761 4.241 −0.0649 4.175 0.0011 DIBP 15,000 4.1761 4.241 −0.0649 4.175 0.0011 DHXPDHXP 29,60029,600 4.4713 4.4713 4.417 4.417 0.0543 0.0543 4.454.45 0.02130.0213 DHXP 29,600 4.4713 4.417 0.0543 4.45 0.0213 DNOPDHXP 47,00029,600 4.6721 4.4713 4.6474.417 0.02510.0543 4.6844.45 −0.02130.0119 DNOP 47,000 4.6721 4.647 0.0251 4.684 −0.0119 DHXPDNOP 29,60047,000 4.4713 4.6721 4.4174.647 0.05430.0251 4.454.684 0.0213−0.0119 DNOP 47,000 4.6721 4.647 0.0251 4.684 −0.0119 DHXP 29,600 4.4713 4.417 0.0543 4.45 0.0213 DIOP 22,000 4.3424 4.374 −0.0316 4.339 0.0034 DIOP 22,000 4.3424 4.374 −0.0316 4.339 0.0034 DIOP 22,000 4.3424 4.374 −0.0316 4.339 0.0034 DIOP 22,000 4.3424 4.374 −0.0316 4.339 0.0034

DEHP 30,000 4.4771 4.475 0.0021 4.477 0.0001 DEHPDEHP 30,00030,000 4.4771 4.4771 4.475 4.475 0.0021 0.0021 4.477 4.477 0.00010.0001 DEHP 30,000 4.4771 4.475 0.0021 4.477 0.0001

DNP 21,500 4.3324 4.458 −0.1256 4.332 0.0004 DNP 21,500 4.3324 4.458 −0.1256 4.332 0.0004

DNP 21,500 4.3324 4.458 −0.1256 4.332 0.0004

DIDP 64,000 4.8062 4.759 0.0472 4.804 0.0022 DIDP 64,000 4.8062 4.759 0.0472 4.804 0.0022

DIDP 64,000 4.8062 4.759 0.0472 4.804 0.0022

nDNOP 45,200 4.6551 4.604 0.0511 4.647 0.0081 nDNOP 45,200 4.6551 4.604 0.0511 4.647 0.0081

nDNOP 45,200 4.6551 4.604 0.0511 4.647 0.0081

DAP 656 2.8169 2.821 −0.0041 2.815 0.0019 DAP 656 2.8169 2.821 −0.0041 2.815 0.0019

DAP 656 2.8169 2.821 −0.0041 2.815 0.0019

DMEP 2750 3.4393 3.436 0.0033 3.419 0.0203 DMEP 2750 3.4393 3.436 0.0033 3.419 0.0203

DMEP 2750 3.4393 3.436 0.0033 3.419 0.0203

2.4.2. CoMFA and CoMSIA for IC50 of PAEs 2.4.2. CoMFA and CoMSIA for IC50 of PAEs 2.4.2.HepG2 CoMFA cells and were CoMSIA used foras experimentalIC50 of PAEs objects to explore the effects of PAEs expo- HepG2 cells were used as experimental objects to explore the effects of PAEs exposure forHepG2 24 h, cells 48 h were and used72 h onas experimentalcell viability, objectsand the to IC explore50 of PAEs the effectsexposure of PAEsfor 72 expo-h on sure for 24 h, 48 h and 72 h on cell viability, and the IC50 of PAEs exposure for 72 h on HepG2 cells was calculated as model biological activity data. Using DEHP, DMEP, DNP, sureHepG2 for cells 24 h, was 48 hcalculated and 72 h as on model cell viability, biological and activity the IC data.50 of UsingPAEs exposureDEHP, DMEP, for 72 DNP, h on DNOP, DIDP as the training set and DPHP, DHXP as the test set to establish CoMFA and DNOP,HepG2 cellsDIDP was as the calculated training as set model and DPHP, biological DH XPactivity as the data. test setUsing to establish DEHP, DMEP, CoMFA DNP, and CoMSIA models. The molecular modeling method and parameters were as elaborated in DNOP,CoMSIA DIDP models. as the The training molecular set andmodeling DPHP, method DHXP asand the parameters test set to establishwere as elaborated CoMFA and in Section 2.4.1. CoMSIASection 2.4.1. models. The molecular modeling method and parameters were as elaborated in Section 2.4.1. 2.4.3. CoMSIA for Nrf2 of PAEs 2.4.3. CoMSIA for Nrf2 of PAEs 2.4.3.Western CoMSIA blots for Nrf2were of used PAEs to determine the content of Nrf2, a key antioxidant protein, Western blots were used to determine the content of Nrf2, a key antioxidant protein, in HepG2 cells exposed to PAEs for 48 h, which was used as the model biological activity in HepG2Western cells blots exposed were toused PAEs to determinefor 48 h, wh theich co wasntent used of Nrf2,as the a model key antioxidant biological protein,activity data. Taking DEHP, DMP, DBP, DNOP, DMEP as the training set, and DEP, DHXP as the indata. HepG2 Taking cells DEHP, exposed DMP, to PAEsDBP, DNOP,for 48 h, DMEP which aswas the used training as the set, model and DEP, biological DHXP activity as the test set to establish the CoMSIA model. The molecular modeling method and parameters data.test set Taking to establish DEHP, the DMP, CoMSIA DBP, model.DNOP, Th DMEPe molecular as the training modeling set, method and DEP, and DHXP parameters as the were as elaborated in Section 2.4.1. weretest set as to elaborated establish thein Section CoMSIA 2.4.1. model. The molecular modeling method and parameters were as elaborated in Section 2.4.1. 2.5. Applicability Domain 2.5. Applicability Domain 2.5. ApplicabilityWe used a Leverge-basedDomain approach to calculate the range of use of the model and We used a Leverge-based approach to calculate the range of use of the model and represented it visually using Williams diagrams. Here leverage denotes the distance be- representedWe used it avisually Leverge-based using Williams approach diagrams. to calculate Here the leverage range ofdenotes use of the the distance model and between the value of the ith observation x and all X and was calculated as: tweenrepresented the value it visually of the ithusing observation Williams x diagrams. and all X andHere was leverage calculated denotes as: the distance between the value of the ith observationhi = x xi (XandT X) all−1 xXiT and was calculated as: hi = xi (XT X)−1xiT i where x is the vector of variables forhi = compound xi (XT X)−1x iiT to be tested and X is the matrix of inde- where xi is the vector of variables for compound i to be tested and X is the matrix of inde- pendenti variables for the training set. The general threshold for whether a sample is ab- wherependent x isvariables the vector for ofthe variables training for set. compound The general i to threshold be tested forand whether X is the amatrix sample of isinde- ab- normal or not is set to h* = 3 k/n, where k is the variable term and n is the number of normalpendent or variables not is set for to the h* training= 3 k/n, set.where The k general is the variablethreshold term for whetherand n is athe sample number is ab- of normal or not is set to h* = 3 k/n, where k is the variable term and n is the number of

Toxics 2021, 9, x FOR PEER REVIEW 4 of 17 Toxics 2021, 9, x FOR PEER REVIEW 4 of 17

Toxics 2021, 9, x FOR PEER REVIEW 4 of 17 Toxics 2021, 9, x FOR PEER REVIEW 4 of 17 Toxics 2021, 9, x FOR PEER REVIEW 4 of 17

DNOP 47,000 4.6721 4.647 0.0251 4.684 −0.0119 DNOP 47,000 4.6721 4.647 0.0251 4.684 −0.0119 DNOP 47,000 4.6721 4.647 0.0251 4.684 −0.0119 DNOP 47,000 4.6721 4.647 0.0251 4.684 −0.0119 DNOP 47,000 4.6721 4.647 0.0251 4.684 −0.0119 DIOP 22,000 4.3424 4.374 −0.0316 4.339 0.0034

DIOP 22,000 4.3424 4.374 −0.0316 4.339 0.0034 DIOP 22,000 4.3424 4.374 −0.0316 4.339 0.0034 Toxics 2021DIOP, 9 , 134 22,000 4.3424 4.374 −0.0316 4.339 0.00344 of 16 DIOP 22,000 4.3424 4.374 −0.0316 4.339 0.0034

DEHP 30,000 4.4771 4.475 0.0021 4.477 0.0001 DEHP 30,000 4.4771 4.475 0.0021 4.477 0.0001 DEHP 30,000 4.4771 4.475 0.0021 4.477 0.0001 DEHP 30,000 Table 1. 4.4771Cont . 4.475 0.0021 4.477 0.0001 DEHP 30,000 4.4771 4.475 0.0021 4.477 0.0001 DNP LD 21,500 Exp. 4.3324CoMFA 4.458CoMFA −0.1256CoMSIA 4.332 CoMSIA0.0004 Names Structures 50 DNP (mg/kg) 21,500 (lgLD50) 4.3324Pred. 4.458 Res. −0.1256Pred. 4.332 Res. 0.0004 DNP 21,500 4.3324 4.458 −0.1256 4.332 0.0004 DNP 21,500 4.3324 4.458 −0.1256 4.332 0.0004 DNP 21,500 4.3324 4.458 −0.1256 4.332 0.0004 DNPDIDP 21,50064,000 4.3324 4.8062 4.458 4.759− 0.12560.0472 4.3324.804 0.00040.0022

DIDP 64,000 4.8062 4.759 0.0472 4.804 0.0022 DIDP 64,000 4.8062 4.759 0.0472 4.804 0.0022 DIDP 64,000 4.8062 4.759 0.0472 4.804 0.0022 DIDPDIDP 64,00064,000 4.8062 4.8062 4.759 4.759 0.0472 0.0472 4.804 4.804 0.00220.0022 nDNOP 45,200 4.6551 4.604 0.0511 4.647 0.0081

nDNOP 45,200 4.6551 4.604 0.0511 4.647 0.0081 nDNOPnDNOP 45,20045,200 4.6551 4.6551 4.604 4.604 0.0511 0.0511 4.647 4.647 0.00810.0081 nDNOP 45,200 4.6551 4.604 0.0511 4.647 0.0081 nDNOP 45,200 4.6551 4.604 0.0511 4.647 0.0081 DAP 656 2.8169 2.821 −0.0041 2.815 0.0019

DAPDAP 656656 2.81692.8169 2.821 2.821− 0.0041−0.0041 2.815 2.815 0.00190.0019 DAP 656 2.8169 2.821 −0.0041 2.815 0.0019 DAP 656 2.8169 2.821 −0.0041 2.815 0.0019 DAP 656 2.8169 2.821 −0.0041 2.815 0.0019

DMEPDMEP 27502750 3.4393 3.4393 3.436 3.436 0.00330.0033 3.4193.419 0.02030.0203 DMEP 2750 3.4393 3.436 0.0033 3.419 0.0203 DMEP 2750 3.4393 3.436 0.0033 3.419 0.0203 DMEP 2750 3.4393 3.436 0.0033 3.419 0.0203 DMEP 2750 3.4393 3.436 0.0033 3.419 0.0203 2.4.2. CoMFA and CoMSIA for IC50 of PAEs 2.4.2. CoMFA and CoMSIA for IC50 of PAEs 2.4.2. CoMFA and CoMSIA for IC50 of PAEs HepG2 cells were used as experimental objects to explore the effects of PAEs expo- 2.4.2.HepG2 CoMFA cells and were CoMSIA used as for experimental IC50 of PAEs objects to explore the effects of PAEs exposure 2.4.2.sure forHepG2CoMFA 24 h, cells and48 h wereCoMSIA and 72used h for onas IC experimentalcell50 of viability, PAEs andobjects the to IC explore50 of PAEs the exposureeffects of forPAEs 72 expo-h on for 24 h, 48 h and 72 h on cell viability, and the IC50 of PAEs exposure for 72 h on HepG2 cells sure2.4.2. HepG2for CoMFA 24 h, cells 48and hwere CoMSIAand used 72 h for ason experimentalIC cell50 ofviability, PAEs objects and the to IC explore50 of PAEs the effectsexposure of PAEs for 72 expo- h on wasHepG2 calculatedHepG2 cells cellswas as model calculatedwere biological used asas model experimental activity biological data. objects Using activity DEHP, to explore data. DMEP, Using the DNP, effects DEHP, DNOP, of DMEP, PAEs DIDP expo-DNP, as sureHepG2 forHepG2 cells24 h, wascells 48 h calculatedwere and used72 h as onas model experimentalcell viability, biological objectsand activity the to IC exploredata.50 of PAEsUsing the exposureeffectsDEHP, of DMEP, PAEsfor 72 DNP,expo-h on thesureDNOP, training for DIDP24 set h, andas48 theh DPHP, and training 72 DHXP h onset celland as the viability,DPHP, test set DH and toXP establish the as theIC50 test CoMFA of setPAEs to and establishexposure CoMSIA CoMFA for models. 72 h and on HepG2DNOP,sure for cellsDIDP 24 h, was as48 thecalculatedh and training 72 h as seton model andcell viability,DPHP, biological DH and XPactivity theas the IC data. 50test of UsingsetPAEs to establishDEHP,exposure DMEP, CoMFA for 72 DNP, h and on TheHepG2CoMSIA molecular cells models. was modeling calculatedThe molecular method as model and modeling parameters biological method wereactivity and as parameters elaborateddata. Using in wereDEHP, Section as DMEP,elaborated 2.4.1. DNP, in DNOP,CoMSIAHepG2 DIDPcells models. was as the calculatedThe training molecular asset model and modeling DPHP, biological method DHXP activity asand the parametersdata. test setUsing to establishwere DEHP, as DMEP, elaboratedCoMFA DNP, and in DNOP,Section DIDP2.4.1. as the training set and DPHP, DHXP as the test set to establish CoMFA and CoMSIASectionDNOP, 2.4.1.DIDP models. as the The training molecular set andmodeling DPHP, method DHXP andas the parameters test set to wereestablish as elaborated CoMFA and in 2.4.3.CoMSIA CoMSIA models. for Nrf2 The ofmolecular PAEs modeling method and parameters were as elaborated in Section 2.4.1. Section2.4.3.CoMSIAWestern CoMSIA 2.4.1. models. blots for were Nrf2The usedmolecular of PAEs to determine modeling the method content and of Nrf2,parameters a key antioxidant were as elaborated protein, in 2.4.3. CoMSIA for Nrf2 of PAEs SectionWestern 2.4.1. blots were used to determine the content of Nrf2, a key antioxidant protein, in2.4.3. HepG2 CoMSIA cells exposed for Nrf2 to of PAEs PAEs for 48 h, which was used as the model biological activity data.2.4.3.in HepG2 Taking WesternCoMSIA cells DEHP, blots forexposed Nrf2 DMP, were of to DBP,used PAEsPAEs DNOP,to fordetermine 48 DMEP h, wh theich as co thewasntent training used of asNrf2, set,the a andmodel key DEP, antioxidant biological DHXP asactivityprotein, the in2.4.3. HepG2Western CoMSIA cells blots forexposed Nrf2were of toused PAEsPAEs to determinefor 48 h, wh theich co wasntent used of Nrf2, as the a modelkey antioxidant biological protein, activity testdata. setWestern Taking to establish DEHP,blots the were DMP, CoMSIA used DBP, to model. determineDNOP, The DMEP molecularthe co asntent the modeling training of Nrf2, set, a method key and antioxidant DEP, and parametersDHXP protein, as the indata. HepG2 Taking cells DEHP, exposed DMP, to PAEs DBP, forDNOP, 48 h, DMEP which aswas the used training as the set, model and DEP,biological DHXP activity as the wereintest HepG2 assetWestern elaborated to establishcells blots exposed in thewere Section CoMSIA to used PAEs 2.4.1 to .model.for determine 48 h, Th whe themolecularich cowasntent used modeling of asNrf2, the amodel method key antioxidant biological and parameters activityprotein, data.testin HepG2 set Taking to establishcells DEHP, exposed the DMP, CoMSIA to DBP,PAEs DNOP,model.for 48 h, ThDMEP whe molecularich as was the used training modeling as the set, model method and DEP, biological and DHXP parameters activity as the data.were Takingas elaborated DEHP, in DMP, Section DBP, 2.4.1. DNOP, DMEP as the training set, and DEP, DHXP as the 2.5.testwere Applicability set as to elaborated establish Domain thein Section CoMSIA 2.4.1. model. The molecular modeling method and parameters testdata. set Taking to establish DEHP, the DMP, CoMSIA DBP, model. DNOP, Th DMEPe molecular as the trainingmodeling set, method and DEP, and DHXP parameters as the weretest set as toelaborated establish inthe Section CoMSIA 2.4.1. model. The molecular modeling method and parameters were2.5.We Applicability as used elaborated a Leverge-based Domain in Section 2.4.1. approach to calculate the range of use of the model and 2.5. Applicability Domain representedwereWe as usedelaborated it visually a Leverge-based in usingSection Williams 2.4.1. approach diagrams. to calculate Here the leverage range of denotes use of the the model distance and 2.5. Applicability Domain between2.5.represented ApplicabilityWe the used value it visuallya DomainLeverge-based of the ithusing observation Williams approach xdiagrams.and to allcalcXulate Hereand was theleverage range calculated denotesof use as: of the the distance model andbe- represented2.5. ApplicabilityWe used it avisually Leverge-basedDomain using Williams approach diagrams. to calculate Here the leverage range ofdenotes use of the the distance model and be- tweenWe the used value a Leverge-basedof the ith observation approach x and toT allcalc− X1ulate andT wasthe rangecalculated of use as: of the model and representedtweenWe the used value it visuallya Leverge-basedof the ithusing observation Williams approachhi = x xi diagrams. (andX toX allcalc) Xxulate iandHere wasthe leverage rangecalculated denotesof use as: of the the distance model andbe- represented it visually using Williamshi = x idiagrams. (XT X)−1xiT Here leverage denotes the distance be- tweenrepresented the value it visually of the ith using observation Williams x diagrams.andT all− 1X Tand Here was leverage calculated denotes as: the distance between ithei value of the ith observationhi = xx iand (X allX) Xx iand was calculated as: wherewherex xis is the the vectorvector ofof variables for for compound compoundT −1 iT ito to be be tested tested and and X isX theis thematrix matrix of inde- of wheretween thexi is valuethe vector of the of ith variables observation forhi = compoundx xi (Xand X) all xX ii andto be was tested calculated and X is as: the matrix of inde- independentpendent variables variables for for the the training training hset.i = set.x Thei (X TheT generalX) general−1xiT threshold threshold for for whether whether a sample a sample is isab- wherependent xi isvariables the vector for of the variables training for hset.i = compound xThei (XT generalX)−1x iiT to threshold be tested andfor whether X is the matrixa sample of inde-is ab- abnormalwherenormal x ior oris thenot not vectoris is set toof h*h*variables == 33 k/n,k/n, for where compound kk isis thethe i to variablevariable be testedterm term and and and X is n nthe is is the matrixthe number number of inde- of of pendentnormalwhere x ori variablesis notthe vectoris set for to ofthe h*variables training = 3 k/n, forset. where compound The kgeneral is the i to variablethreshold be tested term for and whetherand X is n the is amatrixthe sample number of is inde- ab- of samples.pendent If variables the leverage for valuethe training of a compound set. The general is higher threshold than h*, it for is consideredwhether a ansampleX outlier. is ab- normalpendent or variables not is set for to the h* training= 3 k/n, set.where The k general is the variable threshold term for andwhether n is athe sample number is ab-of normal or not is set to h* = 3 k/n, where k is the variable term and n is the number of 2.6.normal Statistical or not Validation is set to h* = 3 k/n, where k is the variable term and n is the number of To test the ﬁtting quality and prediction capability of continuous QSAR model, we used Model Acceptance Criteria in Enalos+ KNIME nodes [20]. According to the criteria proposed by Tropsha A, the following conditions must be satisﬁed:

R2 > 0.6; Rcvext2 > 0.5; (R2-R02)/R2 < 0.1; (R2-R’02)/R2 < 0.1; abs(R02-R’02) < 0.3; 0.85 ≤ k ≤ 1.15; 0.85 ≤ k’ ≤ 1.15

where R2 is the correlation coefficient between the predicted and observed activities, Rcvext2 the external cross validation, R02 the coefficient of determination: predicted versus observed activities, R’02 the coefficient of determination: observed versus predicted Toxics 2021, 9, x FOR PEER REVIEW 5 of 17

samples. If the leverage value of a compound is higher than h*, it is considered an X outlier.

2.6. Statistical Validation To test the fitting quality and prediction capability of continuous QSAR model, we Toxics 2021, 9, 134 used Model Acceptance Criteria in Enalos+ KNIME nodes [20]. According to the criteria5 of 16 proposed by Tropsha A, the following conditions must be satisfied: R2 > 0.6; Rcvext2 > 0.5; (R2–R02)/R2 < 0.1; (R2–R’02)/R2 < 0.1; abs(R02–R’02) < 0.3; 0.85 ≤ kactivities, ≤ 1.15; 0.85 k= ≤ slope:k’ ≤ 1.15 predicted versus observed activities regression lines through the origin and k’ = slope: observed versus predicted activities regression lines through the origin. where R2 is the correlation coefficient between the predicted and observed activities, Rcvext2.7. Statistical2 the external Analysis cross validation, R02 the coefficient of determination: predicted versus 2 observedAll rawactivities, data were R’0 the collated coefficient in a Microsoft of determination: Excel database, observed whilst versus SPSS predicted 18.0 (SPSS activ- Inc., ities,Chicago, k = slope: IL, USA) predicted and GraphPad versus observed Prism 8.0 ac (GraphPadtivities regression Inc., La Jolla, linesCA, through USA) the were origin used andfor statisticalk’ = slope: analysis.observed Experimental versus predicted data ac weretivities expressed regression as means lines through± standard the origin deviation (means ± SD). Differences among groups were compared with one-way analysis of variance 2.7.with Statistical the significant Analysis difference acceptance at p < 0.05, and the extremely significant difference acceptanceAll raw atdatap < were 0.01 confidencecollated in levels,a Microsoft respectively. Excel database, whilst SPSS 18.0 (SPSS Inc., Chicago, IL, USA) and GraphPad Prism 8.0 (GraphPad Inc., La Jolla, CA, USA) were used for3. Resultsstatistical and analysis. Discussion Experimental data were expressed as means ± standard deviation (means3.1. Cell ± ViabilitySD). Differences among groups were compared with one-way analysis of variance withPAEs the are significant a class of exogenous difference compoundsacceptance at and p < HepG2 0.05, and cells the are extremely derived from significant human differencehepatocytes, acceptance which have at p similar< 0.01 confidence metabolic functionslevels, respectively. to hepatocytes and are a common cell model for studying the toxicity of exogenous compounds, hepatotoxicity and mitochondrial 3.toxicity. Results Meanwhile, and Discussion HepG2 is probably considered to be one of the first (one of the first) 3.1.hepatocyte Cell Viability lines to be isolated and widely used. It is relatively inexpensive to culture, relatively well maintained and readily available. Therefore, we chose HepG2 cells as the PAEs are a class of exogenous compounds and HepG2 cells are derived from human subject of our study. The present study utilized CCK-8 to determine the effects of seven hepatocytes, which have similar metabolic functions to hepatocytes and are a common PAEs exposed for different times (24, 48 and 72 h) on the viability of HepG2 cells. cell model for studying the toxicity of exogenous compounds, hepatotoxicity and mito- The results presented in Figure1 show that the cell viability decreased significantly chondrial toxicity. Meanwhile, HepG2 is probably considered to be one of the first (one of (p < 0.05), particularly at 72 h of exposure for all the tested PAEs, except DMP. PAEs had a the first) hepatocyte lines to be isolated and widely used. It is relatively inexpensive to certain accumulation. With the extension of exposure time, the accumulation of PAEs in culture, relatively well maintained and readily available. Therefore, we chose HepG2 cells cells increased, so the toxicity increased significantly. Further, the results in this study are as the subject of our study. The present study utilized CCK-8 to determine the effects of agreement with that of previous ones wherein DEHP was observed to induce toxicity in sevendifferent PAEs types exposed of cells for by different severely times inhibiting (24, 48 the and proliferation 72 h) on the of viability Hep3B [of21 HepG2]. cells.

FigureFigure 1. 1. TheThe influence inﬂuence of of PAEs PAEs on on cell cell viability. viability. Cell Cellss were treated with 62.5,62.5, 125,125, 250,250, 500500 andand 10001000µ μMM PAEsPAEs for for 24 24 h, h, 48 48 h hand and 72 72 h. h. (A (A):): DEHP, DEHP, (B ():B DHXP;): DHXP; (C ):(C DMEP;): DMEP; (D ):(D DPHP;): DPHP; (E): ( DNP;E): DNP; (F): ( DNOP;F): DNOP; (G): ( DIDP.G): DIDP. Data Data are expressed are expressed as mean as mean± SD , ±n SD, = 6. n *=p 6.< * 0.05,p < 0.05, ** p **

3.2. Nrf2 Protein Content Nrf2, a ubiquitous transcription factor that could mediate and regulate the expression of more than 200 antioxidant enzymes and cytoprotective proteins [22] was used as the main mediator in this study. Previous studies identiﬁed the Nrf2 signalling pathway as the main host defense pathway in the body. This transcription factor, Nrf2 enters the nucleus and interacts with the downstream HO-1 protein to activate the oxidative stress pathway to protect the cell that is not coerced when the intracellular ROS increases [23]. Toxics 2021, 9, 134 6 of 16

It is observed that the inhibitory effect of Nrf2 enhances the lipid peroxidation of the liver, which indicates that the activation of Nrf2 plays an important role in the antioxidant defense mechanism of the liver [24]. However, some recent studies have shown that DEHP Toxics 2021, 9, x FOR PEER REVIEW 6 of 17 exposure caused oxidative stress by interfering with the transcripts of the Nrf2 signaling pathway and its downstream genes [25]. After DEHP exposure, INS-1 cells produced a large amount of ROS leading to the imbalance in the NRF2-dependent antioxidant defense protectionThe [results26]. In thepresented present in study, Figure Nrf2 1 show was observed that the tocell have viability undergone decreased similar significantly changes (Figure(p < 0.05),2). After particularly 48 h of exposureat 72 h of toexposure DEHP, DMP,for all etc.,the tested Nrf2 wasPAEs, inhibited except DMP. in the PAEs cells of had thea certain high-dose accumulation. group, which With causes the extension a certain of degree exposure of oxidativetime, the accu stress.mulation However, of PAEs low in concentrationcells increased, PAEs so inthe the toxicity test concentration increased significantly. increased the Further, expression the results of Nrf2 in after this 48 study h, the are resultagreement was accord with withthat otherof previous studies ones which wherei revealedn DEHP that was DEHP observed induced to testicular induce toxicity toxicity in throughdifferent oxidative types of stresscells by and severely up-regulated inhibiting Nrf2 the expressionproliferation [27 of]. Hep3B Considered [21]. together, under the stimulation of low concentration PAEs in the test concentration, the Nrf2 protein content3.2. Nrf2 increased, Protein Content indicating that HepG2 cells can trigger a defense mechanism to combat the toxicityNrf2, ofa lowubiquitous concentration transcription PAEs in factor the test that concentration could mediate [28 and]. Meanwhile regulate the with expres- the increasedsion of more dose than and 200 the antioxidant exposure time, enzymes the Nrf2 and signalingcytoprotective pathway proteins was inhibited,[22] was used and as subsequentlythe main mediator the production in this study. of free Previous radicals studies and the identified antioxidant the Nrf2 defense signalling system pathway were outas ofthe balance, main host making defense the pathway cells enter in athe state body of. oxidativeThis transcription stress. However, factor, Nrf2 due enters to the the complexitynucleus and of theinteracts branched with chains the downstream of PAEs, the HO-1 changes protein were alsoto activate different, the and oxidative even some stress PAEspathway did not to causeprotect a signiﬁcantthe cell that decrease is not coerce in Nrf2.d when the intracellular ROS increases [23].

Figure 2. µ Figure The2. The inﬂuence influence of of PAEs PAEs on on cell cell Nrf2 Nrf2 content. content. Cells Ce werells were treated treated with with 62.5, 62.5, 125, 125, 250, 250, 500 500 and and 1000 1000M μ PAEsM PAEs for 48 for h. 48 (Ah.): Western(A): Western blot stripsblot strips of PAEs; of PAEs; (B): DEHP, (B): DEHP, (C): DMP; (C): DMP; (D): DNOP; (D): DNOP; (E): DHXP; (E): DHXP; (F): DMEP; (F): DMEP; (G): DEP; (G): (DEP;H): DBP. (H): DataDBP. areData expressedare expressed as mean as ±meanSD, ±n SD,= 3. n * =p 3.< * 0.05,p < 0.05, ** p <** 0.01p < 0.01 versus versus the the control control group. group.

Toxics 2021, 9, 134 7 of 16

3.3. LD50 3D-QSAR Additionally, a 3D-QSAR method was used to quantitatively analyze the toxic effects of PAEs on rats. The results obtained, reported in Table2, show that both the CoMFA model and CoMSIA1 model have good predictive ability with cross-validation coefficient (q2) greater than 0.50 [29,30] with values equal to 0.522 and 0.621, respectively. Other performance metrices evaluated were the standard error of estimate and Fischer’s value which were 0.067 and 192.02 and 0.015 and 1528.55 for CoMFA and CoMSIA1 models respectively. Furthermore, the non-cross validation coefficients were all greater than 0.90, indicating strong adaptability and robustness of the developed models [31,32]. The correlation between the predicted value and the actual value of training set and test set r2 were 0.9896 and 0.9997 in CoMFA (Figure3F), and correlation between the predicted value and the actual value of training set and test set r2 were 0.9996 and 0.9997 in CoMSIA1, indicating that the model had good predictive ability. In addition, the contribution rates of the steric field (S) and the electrostatic field (E) in the CoMFA model were 58.4% and 41.6%, respectively. The contributions of the S, E, hydrophobic field (H), hydrogen bond donor field (D) and acceptor field (A) in the CoMSIA1 model were 18.8%, 26.6%, 50.8%, 3.8% and 0 respectively (Table2). As can be observed from the results, contrastingly the structure-activity relationship between the hydrogen bond donor field and acceptor field was extremely small. Hence a CoMSIA2 model was established without considering the D and A fields. Subsequently, this approach was observed to improve the predictive performance with a q2 of 0.631 and r2 of 1 (Table2). In this case, the contribution rates of S, E and H were 19.3%, 28.4% and 52.3% (Table2), respectively. Further as can be observed from the figure the r2 of the model was 1.000 and 0.999 for the training and test sets, respectively.

Table 2. 3D-QSAR models’ parameters.

3D-QSAR Model. Styles Q2 R2 F SEE S E H D A CoMFA 0.522 0.991 0.067 192.025 58.4 41.6 LD50 CoMSIA1 0.621 1 0.015 1528.55 18.8 26.6 50.8 3.8 0 CoMSIA2 0.631 1 0.015 1552.090 19.3 28.4 52.3 IC50 CoMFA 0.69 0.98 0.021 48.513 79 21 IC50 CoMSIA1 0.687 0.926 0.036 24.907 20.5 22.3 36.1 0 21.2 Nrf2 CoMSIA1 0.512 0.966 0.075 42.754 23.2 11.8 51.9 0 13.1

In the CoMFA, CoMSIA1 and CoMSIA2 models evaluated in this study, the contribution rate of the S and the H accounted for 58.4%, 69.6%, and 71.6% (Table2), respectively, suggesting that the non-specific reactions account for a larger proportion in the toxic effects of PAEs. The results of this study were consistent with other QSAR study on PAE biotoxicity and flammability which has shown that the hydrophobicity of molecules was an important parameter to characterize the toxicity of compounds [33]. This study also tried to apply the CoMSIA method to include the hydrogen bond donor and acceptor properties of the molecule. It was found that the above two parameters contributed very little to the structure-activity relationship, further confirming that non-specific reactions accounted for a relatively large proportion. However, whether it was CoMFA or CoMSIA1 and CoMSIA2 models, the electrostatic field had a greater contribution, accounting for 41.6%, 26.6% and 28.4% (Table2), respectively. The QSAR model indicates that after PAEs enter cells, there may be electrostatic interactions between ester bonds, which would cause toxic effects through biochemical reactions. The contour maps of CoMFA and CoMSIA models could also be used to explore the toxicity mechanism of compounds to a certain extent [34,35]. In the CoMFA model, two red areas appeared above the ether group of the electrostatic field equipotential diagram (shown in Figure3B), and one red area appeared above the carbonyl group. These appearances indicate the introduction of negatively charged groups at the positions of the ether and the carbonyl groups, respectively which were responsible for the decreased toxicity of the compound. This effect was significant at the ether group wherein a significant electrostatic Toxics 2021, 9, 134 8 of 16

interaction with the possible acceptor was observed. In fact, in the stereo field equipotential diagram in Figure3A, 12 yellow areas at the double bond positions on both sides of the DAP were observed. These regions indicated the introduction of large groups at the double bonds of the positively charged groups which increased the compound toxicity. As for the large groups introduced at the middle position of the double bonds on both sides and the right carbonyl group might have decreased the compound toxicity (Figure3A). Similar additions on the bonds and equipotential results of the electrostatic fields were observed Toxics 2021, 9, x FOR PEER REVIEW for the CoMSIA1 model as well. In fact, the presence of white and yellow areas on the8 of side 17 chains in the hydrophobic field (Figure3E) confirm the influence of bond additions and electrostatic fields on the side chains and subsequent toxicity.

Figure 3. DAP’s three-dimensional contour maps of steric field (A), the electrostatic field (B) in the CoMFA model and steric Figure 3. DAP’s three-dimensional contour maps of steric field (A), the electrostatic field (B) in the CoMFA model and stericfield (fieldC), the (C electrostatic), the electrostatic field (D field), hydrophobic (D), hydrophobic field (E )field in the (E CoMSIA2) in the CoMSIA2 model. The model. relation The schema relation between schema predictedbetween predictedvalues and values experimental and experiment values onal PAEs’values lgLD50on PAEs’ ((F lgLD50): CoMFA; (F: (CoMFA;G): CoMSIA1; G: CoMSIA1; (H): CoMSAI2). H: CoMSAI2). Structural Structural superimposition superim- positionof PAEs(I). of PAEs(I).

In the CoMFA, CoMSIA1 and CoMSIA2 models evaluated in this study, the contribution rate of the S and the H accounted for 58.4%, 69.6%, and 71.6% (Table 2), respectively, suggesting that the non-specific reactions account for a larger proportion in the toxic effects of PAEs. The results of this study were consistent with other QSAR study on PAE biotoxicity and flammability which has shown that the hydrophobicity of molecules was an important parameter to characterize the toxicity of compounds [33]. This study also tried to apply the CoMSIA method to include the hydrogen bond donor and acceptor properties of the molecule. It was found that the above two parameters contributed very little to the structure-activity relationship, further confirming that non-specific reactions accounted for a relatively large proportion. However, whether it was CoMFA or CoMSIA1 and CoMSIA2 models, the electrostatic field had a greater contribution, accounting for

Toxics 2021, 9, x FOR PEER REVIEW 9 of 17

Toxics 2021, 9, x FOR PEER REVIEW 9 of 17 Toxics 2021, 9, x FOR PEER REVIEW 9 of 17

Toxics 2021, 9, x FOR PEER REVIEW 9 of 17 41.6%, 26.6% and 28.4% (Table 2), respectively. The QSAR model indicates that after PAEs enter41.6%, cells, 26.6% there and may 28.4% be (Tableelectrostatic 2), respectively. interactions The between QSAR estermodel bonds, indicates which that would after causePAEs toxicenter41.6%, effectscells, 26.6% there through and may 28.4% biochemicalbe electrostatic(Table 2), reactions. respectively. interactions The between QSAR ester model bonds, indicates which that would after cause PAEs 41.6%, 26.6% and 28.4% (Table 2), respectively. The QSAR model indicates that after PAEs toxicenterThe effects cells, contour there through maymaps biochemical be of electrostatic CoMFA reactions. and intera CoMSIActions models between could ester also bonds, be used which to would explore cause the enter cells, there may be electrostatic interactions between ester bonds, which would cause toxicitytoxicThe effects mechanism contour through maps of biochemical compounds of CoMFA reactions. and to a CoMSIA certai n extent models [34,35]. could Inalso the be CoMFA used to model,explore two the toxic effects through biochemical reactions. redtoxicity areasThe mechanism appearedcontour maps aboveof compounds of theCoMFA ether andtogroup a CoMSIAcertai of then extent electrostaticmodels [34,35]. could field In also the equipotential be CoMFA used to model, explore diagram two the Toxics 2021, 9, 134 The contour maps of CoMFA and CoMSIA models could also be used to explore9 of 16 the (shownredtoxicity areas in mechanism appearedFigure 3B), above of and compounds onethe redether area groupto appeareda certai of then extent aboveelectrostatic the[34,35]. carbonyl field In the equipotential group. CoMFA These model, diagram appear- two toxicity mechanism of compounds to a certain extent [34,35]. In the CoMFA model, two ances(shownred areas indicate in appearedFigure the 3B), introduction above and one the redether of areanegatively group appeared of chargedthe above electrostatic groups the carbonyl atfield the equipotential positionsgroup. These of the diagramappear- ether red areas appeared above the ether group of the electrostatic field equipotential diagram andances(shown the indicate carbonylin Figure the groups, 3B),introduction and respectively one red of negativelyarea whichappeared chargedwere above responsible groups the carbonyl at for the the positions group. decreased These of the toxicityappear- ether ances(shown indicate in Figure the 3B), introduction and one red of areanegatively appeared charged above groups the carbonyl at the positionsgroup. These of the appear- ether 3.4.ofand IC the50 the3D-QSAR compound. carbonyl groups,This effect respectively was significant which at were the ether responsible group whereinfor the decreased a significant toxicity elec- andances the indicate carbonyl the groups, introduction respectively of negatively which chargedwere responsible groups at for the the positions decreased of the toxicity ether trostaticofTo the further compound. interaction explore This thewith effect toxic the effectswaspossible significant of PAEs,acceptor at an the wasin vitroether observed. cellgroup experiment whereinIn fact, in a to significantthe determine stereo fieldelec- ofand the the compound. carbonyl groups,This effect respectively was significant which at were the ether responsible group whereinfor the decreased a significant toxicity elec- the 50%equipotentialtrostatic lethal interaction dose diagram of PAEs with in to Figurethe HepG2 possible 3A, cells 12 acceptoryellow was established. areas was atobserved. the The double experimental In bondfact, inpositions the and stereo model on bothfield trostaticof the compound. interaction This with effect the waspossible significant acceptor at the was ether observed. group whereinIn fact, in a significantthe stereo fieldelec- predictedsidesequipotential of values the DAP ofdiagram 72 were h-lgIC50 observed.in Figure and 3A,These the 12 difference regionsyellow areasindicated between at the themthe double introduction are bond detailed positions of in largeTable ongroups3 both. equipotentialtrostatic interaction diagram with in Figurethe possible 3A, 12 acceptoryellow areas was atobserved. the double In bondfact, in positions the stereo on bothfield Onatsides the the basis ofdouble the of DAP thebonds results,were of observed.the a positively CoMFA These modelchar regionsged and groups indicated a CoMSIA which the increased model introduction were the built compoundof large which groups tox- sidesequipotential of the DAP diagram were observed.in Figure 3A,These 12 regionsyellow areas indicated at the the double introduction bond positions of large ongroups both respectivelyicity.at the As double for had the abonds q large2 of of 0.69 groups the and positively introduced 0.687(Table char ged2at), the and groups middle r 2 of which 0.980 position increased and of 0.926 the the (Tabledouble compound2). bonds The tox- on atsides the ofdouble the DAP bonds were of observed.the positively These char regionsged groups indicated which the increased introduction the compoundof large groups tox- non-crossbothicity. sidesAs validation for and the thelarge coefﬁcients right groups carbonyl wereintroduced allgroup greater atmi theght than middlehave 0.9, decreased indicating position of thatthe the compound the double model bonds wastoxicity on at the double bonds of the positively charged groups which increased the compound tox- relatively(Figurebothicity. sides stable.As 3A). for and Similar Thethe the large training rightadditions groups setcarbonyl correlation introducedon the group bonds r2 miatbetween anghtthed middleequipotentialhave the decreased predicted position results valueofthe the compoundof and double the the electrostatic actual bonds toxicity on icity. As for the large groups introduced at the middle position of the double bonds on valuefields(Figureboth in CoMFA sideswere 3A). observedand Similar and the CoMSIA rightadditions for the carbonyl were CoMSIA1 on allthe group 0.98 bonds model (Figure mi anght asd 4 equipotentialwell.haveG,H), Indecreased respectively, fact, theresults thepresence indicating compoundof the of electrostatic white that toxicity and both sides and the right carbonyl group might have decreased the compound toxicity theyellow modelfields(Figure were hadareas 3A). goodobserved onSimilar the predictive side additionsfor chains the ability.CoMSIA1 inon the the In hydrophobic bonds addition,model an asd thewell.equipotential field contribution In (Figure fact, the 3E)results ratespresence confirm of of the S theof and electrostatic white influence E in and (Figure 3A). Similar additions on the bonds and equipotential results of the electrostatic theof CoMFAyellowfields bond were areas additions model observed on werethe and side 79% forelectrostatic chains andthe CoMSIA1 21% in the (Tablefields hydrophobic model 2on), respectively.the as side well. field chains In(Figure The fact, and contribution the 3E)subsequent presence confirm rates toxicity.theof whiteinfluence of S, and E, H,ofyellowfields D,bond and were areasadditions A inobserved on the the CoMSIA1and side forelectrostatic chains themodel CoMSIA1 in the fields were hydrophobic model on 20.5%, the as side 22.3%, well. field chains In 36.1%,(Figure fact, and 0the subsequent3E) and presence confirm 21.2% (Tabletoxicity.theof white influence2), and ofyellow bond areas additions on the and side electrostatic chains in the fields hydrophobic on the side field chains (Figure and 3E)subsequent confirm toxicity.the influence respectively.3.4. IC50 3D-QSAR Like the 3D-QSAR model of LD50, the contribution rates of S and H were of bond additions and electrostatic fields on the side chains and subsequent toxicity. relatively3.4. IC50 high, 3D-QSAR especially the contribution rate of S in the CoMFA model reached 79% 3.4. IC50 Table 3.(Table Actual2), value, which 3D-QSAR predicted has shown value thatand error non-speciﬁc of 3D-QSAR reactions model of accounted PAEs 72 h-lgIC for a50 relatively. large 3.4. IC50 3D-QSAR Table 3.proportion Actual value, with predicted a high value probability and error of chargeof 3D-QSAR transfer. model of PAEs 72 h-lgIC50. Table 3. Actual value, predicted value and error of 3D-QSARCoMFA model of PAEsCoMFA 72 h-lgICCoMSIA50. CoMSIA Names Structures 72 h-IC50 (μM) Exp.(lgIC50) Table 3. Actual value, predicted value and error of 3D-QSARCoMFAPred. model of PAEsCoMFARes. 72 h-lgICCoMSIAPred.50. CoMSIARes. Names Table Structures 3. Actual value, predicted 72 h-IC value50 (μ andM) errorExp.(lgIC of 3D-QSAR50) model of PAEs 72 h-lgIC . CoMFAPred. CoMFARes. 50 CoMSIAPred. CoMSIARes. Names Structures 72 h-IC50 (μM) Exp.(lgIC50) CoMFAPred. CoMFARes. CoMSIAPred. CoMSIARes. Names Structures 72 h-IC50 (μM) Exp.(lgIC50) CoMFA CoMFA CoMSIA CoMSIA Names Structures 72 h-IC (µM) Exp. (lgIC ) Pred. Res. Pred. Res. 50 50 Pred. Res. Pred. Res. DEHP 1160 3.0645 3.053 0.0115 3.069 −0.0045 DEHP 1160 3.0645 3.053 0.0115 3.069 −0.0045 DEHP 1160 3.0645 3.053 0.0115 3.069 −0.0045 DEHPDEHP 11601160 3.06453.0645 3.0533.053 0.01150.0115 3.0693.069 −0.0045−0.0045

DNP 1513 3.18 3.206 −0.026 3.2 0.025 DNP 1513 3.18 3.206 −0.026 3.2 0.025

DNPDNP 15131513 3.183.18 3.2063.206 −0.026−0.026 3.23.2 0.025 0.025 DNP 1513 3.18 3.206 −0.026 3.2 0.025

DIDP 1722 3.236 3.232 0.004 3.094 −0.0052 DIDPDIDP 17221722 3.2363.236 3.2323.232 0.0040.004 3.0943.094 −0.0052−0.0052 DIDP 1722 3.236 3.232 0.004 3.094 −0.0052 DIDP 1722 3.236 3.232 0.004 3.094 −0.0052 Toxics 2021, 9, x FOR PEER REVIEW 10 of 17 Toxics 2021, 9, x FOR PEER REVIEW 10 of 17

Toxics 2021, 9, x FOR PEER REVIEW 10 of 17

DMEPDMEP 975.1975.1 2.9892.989 2.9842.984 0.0050.005 2.9922.992 −0.003−0.003 DMEP 975.1 2.989 2.984 0.005 2.992 −0.003 DMEP 975.1 2.989 2.984 0.005 2.992 −0.003 DHXPDMEP 1227975.1 3.08882.989 3.0912.984 −0.00220.005 2.9722.992 − 0.0080.003 DHXP 1227 3.0888 3.091 −0.0022 2.972 0.008 DHXPDHXP 12271227 3.08883.0888 3.0913.091 −0.0022−0.0022 2.9722.972 0.008 0.008

DNOP 1678 3.225 3.206 0.019 3.233 0.003 DNOPDNOP 16781678 3.2253.225 3.2063.206 0.0190.019 3.2333.233 0.0030.003 DNOP 1678 3.225 3.206 0.019 3.233 0.003

DPHPDPHP 954.9954.9 2.982.98 2.9922.992 −0.012−0.012 3.2033.203 −0.023−0.023 DPHP 954.9 2.98 2.992 −0.012 3.203 −0.023 DPHP 954.9 2.98 2.992 −0.012 3.203 −0.023

To further explore the toxic effects of PAEs, an in vitro cell experiment to determine the 50%To furtherlethal dose explore of PAEs the toxic to HepG2 effects cells of PAEs was established., an in vitro cellThe experimentexperimental to anddetermine model predictedthe 50%To furtherlethal values dose explore of of72 PAEs h-lgIC the toxicto50 HepG2 and effects the cells ofdifference PAEs was established., an between in vitro themThecell experimentexperimental are detailed to and indetermine Table model 3. Onpredictedthe the50% basis lethal values of dose the of results,of72 PAEsh-lgIC a to 50CoMFA HepG2 and the modelcells difference was and established. a between CoMSIA themThe model experimental are were detailed built andin which Table model re- 3. spectivelyOnpredicted the basis valueshad of a the q of2 of results,72 0.69 h-lgIC and a 50CoMFA 0.687(Table and the model difference 2), and ra 2 between CoMSIAof 0.980 andthem model 0.926 are were detailed(Table built 2). inwhich The Table non- re- 3. 2 2 spectivelycrossOn the validation basis had of a thecoefficientsq of results, 0.69 and awere CoMFA 0.687(Table all greater model 2), than andand 0.9,ra ofCoMSIA indicating 0.980 and model 0.926that werethe (Table model built 2). whichwasThe non-rela- re- 2 2 tivelycrossspectively validation stable. had The a coefficientsq training of 0.69 andset were correlation0.687(Table all greater r 2),2 betweenthan and 0.9,r of theindicating 0.980 predicted and that0.926 value the (Table model and 2). the wasThe actual rela-non- 2 tivelyvaluecross validationinstable. CoMFA The coefficients andtraining CoMSIA set were correlation were all allgreater 0.98 r between(Figuresthan 0.9, 4G,H),theindicating predicted respectively, that value the model indicatingand the was actual rela-that 2 thevaluetively model instable. CoMFA had The good and training predictive CoMSIA set correlationability.were all In 0.98 additi r (Figuresbetweenon, the 4G,H),the contribution predicted respectively, rates value of indicatingand S and the E actualin that the CoMFAthevalue model in modelCoMFA had good were and predictive 79% CoMSIA and 21% ability.were (Table all In 0.98 additi 2), (Figuresrespectively.on, the 4G,H), contribution The respectively, contribution rates of indicating S rates and Eof in S,that the E, H,CoMFAthe D, model and model Ahad in goodthewere CoMSIA1 predictive79% and model 21% ability. (Table were In additi 20.5%,2), respectively.on, 22.3%, the contribution 36.1%, The contribution0 and rates 21.2% of S (Tablerates and Eof 2),in S, there- E, H,CoMFA D, and model A in thewere CoMSIA1 79% and model 21% (Table were 20.5%,2), respectively. 22.3%, 36.1%, The 0contribution and 21.2% (Tablerates of 2), S, re- E, spectively. Like the 3D-QSAR model of LD50, the contribution rates of S and H were rela- tivelyspectively.H, D, high,and ALike especially in thethe CoMSIA13D-QSAR the contribution modelmodel wereof rate LD 20.5%, 50of, theS in contribution22.3%, the CoMFA 36.1%, ratesmodel 0 and of reached 21.2%S and H(Table 79% were (Table2), rela- re- 2),tivelyspectively. which high, has Likeespecially shown the 3D-QSARthat the non-specific contribution model reactionof rate LD 50of, stheS accounted in contribution the CoMFA for a ratesmodelrelati velyof reached S andlarge H 79%proportion were (Table rela- with2),tively which a high,high has probabilityespecially shown that the of non-specific chargecontribution transfer. reaction rate ofs S accounted in the CoMFA for a relatimodelvely reached large proportion79% (Table with2), which Froma high has the probability shown three-position that of non-specific charge equipotential transfer. reaction maps accounted of the CoMFA for a relatimodel,vely we large found proportion that the yellowwith Froma higharea the probabilitycovered three-position both of chargesides equipotential of transfer. the DMEP map molecule of the CoMFA (Figure model, 4A), indicating we found that the introductionyellowFrom area the coveredof three-position large bothgroups sides equipotentialwould of the enhance DMEP map themolecule of activity the CoMFA (Figure of the model, 4A),molecule indicating we foundand lead thatthat to the a introductionyellow area coveredof large bothgroups sides would of the enhance DMEP the molecule activity (Figure of the 4A),molecule indicating and lead that to the a decrease in IC50. The equipotential diagram in the electrostatic field shows that adding decreasenegativelyintroduction in charged IC of50 .large The groups equipotentialgroups near would the diagramcarbonyl enhance ingroup the the activity elandectrostatic the of ether the field moleculegroup shows below and that couldlead adding tode- a creasenegativelydecrease the in biological charged IC50. The groups activity equipotential near of the the molecule carbonyldiagram (Figure ingroup the 4B),elandectrostatic therebythe ether enhancingfield group shows below the that biologicalcould adding de- toxicitycreasenegatively the of biological thecharged molecule, groups activity whereas near of the the the molecule carbonyl introduc (Figure grouption of 4B),and negatively therebythe ether chargedenhancing group belowgroups the biologicalcould near thede- secondtoxicitycrease the etherof biologicalthe group molecule, would activity whereas cause of the the molecule oppositeintroduc (Figuretion effect. of 4B), negativelyThe thereby position chargedenhancing of the groupsCoMSIA the biological near model the secondwastoxicity like ether of that the groupof molecule, the electrostaticwould whereas cause field thethe oppositeintroducequipotential tioneffect. of diagram, Thenegatively position except charged of thatthe CoMSIAgroupsthere were near model two the redwassecond regionslike ether that at ofgroup the the end electrostaticwould of the cause side field the chain oppositeequipotential in the effect. electrostatic diagram, The position field except eq ofuipotential thatthe CoMSIAthere werediagram, model two redwhichwas regions like show that atthat of the the the end electrostatic introduction of the side field ofchain negative equipotential in thlye electrostaticcharged diagram, groups field except was equipotential notthat beneficial there werediagram, to twothe whichbiologicalred regions show activity. atthat the the end introduction of the side of chain negative in thlye electrostaticcharged groups field was equipotential not beneficial diagram, to the biologicalwhichIn showthe activity. hydrophobic that the introduction field equipotential of negative diagly chargedram with groups the largest was not proportion beneficial (Figure to the 4D),biological Inwe the were activity. hydrophobic surprised tofield find equipotential that the whit diage arearam was with very the large,largest almost proportion covering (Figure the entire4D), weIn side the were chain, hydrophobic surprised which indicated tofield find equipotential that that the the whit incr diageease arearam of was hydrophilicwith very the large,largest groups almost proportion in the covering side (Figure chain the isentire4D), closely we side were related chain, surprised whichto the IC50indicated to find of PAEs that that theon the HepG2whit increease area cells, of washydrophilic and very this large,also groups confirmed almost in the covering theside longer chain the isentire closely side related chain, towhich the IC50 indicated of PAEs that on the HepG2 increase cells, of hydrophilic and this also groups confirmed in the theside longer chain the fatty acid chain of PAEs, the lower IC50. In the hydrogen bond acceptor field (Figure 4F),theis closely fattywe observed acid related chain toa ofredthe PAEs, IC50area ofonthe PAEs thelower left on IC sideHepG250. In of the the cells, hydrogen ether and group this bond also of acceptorDAP,confirmed indicating field the (Figure longer that adding4F),the fattywe hydrogenobserved acid chain bonda redof PAEs, acceptorsarea onthe the lowercould left ICincreaside50. Inofse thethe molecular hydrogen ether group activity, bond of acceptor DAP,but the indicating carbonylfield (Figure thatpo- sitionadding4F), we on hydrogenobserved the right bondhada red a acceptorspurplearea on area, the could leftwhich increaside ha ofdse the molecular oppositeether group activity, effect of to DAP,but the the hydrogen indicating carbonyl bond thatpo- acceptorsitionadding on hydrogen onthe the right left. bondhad a acceptorspurple area, could which increa hadse the molecular opposite activity, effect to but the the hydrogen carbonyl bond po- acceptorsition on onthe the right left. had a purple area, which had the opposite effect to the hydrogen bond acceptor on the left.

Toxics 2021, 9, 134 10 of 16 Toxics 2021, 9, x FOR PEER REVIEW 11 of 17

Figure 4. DMEP’s three-dimensional contour maps of steric field (A), the electrostatic field (B) in the CoMFA model and Figure 4. DMEP’s three-dimensional contour maps of steric field (A), the electrostatic field (B) in the CoMFA model and steric field (C), the electrostatic field (D), hydrophobic field (E), Acceptor field (F) in the CoMSIA model (G): CoMFA; steric field (C), the electrostatic field (D), hydrophobic field (E), Acceptor field (F) in the CoMSIA model (G): CoMFA; (H): (HCoMSIA).): CoMSIA. From the three-position equipotential map of the CoMFA model, we found that 3.5. Nrf2 3D-QSAR the yellow area covered both sides of the DMEP molecule (Figure4A), indicating that the introduction of large groups would enhance the activity of the molecule and lead After exploring the effect of PAEs on the cell viability of the HepG2 cells, the to a decrease in IC . The equipotential diagram in the electrostatic field shows that mechanism was analyzed.50 The content of Nrf2, a key antioxidant protein, was determined, adding negatively charged groups near the carbonyl group and the ether group below and a 3D-QSAR model of the effect of PAEs on Nrf2 was established. On this basis, we could decrease the biological activity of the molecule (Figure4B), thereby enhancing the can predict the content of Nrf2 protein in HepG2 cells with different PAEs, and then biological toxicity of the molecule, whereas the introduction of negatively charged groups understand the effect of PAEs on Nrf2 protein. On this basis, we established CoMSIA near the second ether group would cause the opposite effect. The position of the CoMSIA model, where q2 was 0.512 and r2 was 0.966 (Table 2). model was like that of the electrostatic field equipotential diagram, except that there were two redThe regions non-cross at the endvalidation of the sidecoefficients chain in were the electrostatic all greater fieldthan equipotential 0.9, indicating diagram, that the whichmodel show was thatrelatively the introduction stable. The oftraining negatively set correlation charged groups r2 between was not the beneficial predicted to value the biologicaland the actual activity. value was 0.9818 (Figure 5E), indicating that the model has good predictive ability.In the In hydrophobicaddition, the field contribution equipotential rates diagram of S, E, H, with D and thelargest A in the proportion CoMSIA1 (Figure model4 wereD ), we23.2%, were 11.8%, surprised 51.9%, to find0 and that 13.1% the (Table white area2), respectively. was very large, almost covering the entire side chain, which indicated that the increase of hydrophilic groups in the side chain is

Toxics 2021, 9, 134 11 of 16

closely related to the IC50 of PAEs on HepG2 cells, and this also conﬁrmed the longer the fatty acid chain of PAEs, the lower IC50. In the hydrogen bond acceptor ﬁeld (Figure4F), we observed a red area on the left side of the ether group of DAP, indicating that adding hydrogen bond acceptors could increase molecular activity, but the carbonyl position on the right had a purple area, which had the opposite effect to the hydrogen bond acceptor on the left.

3.5. Nrf2 3D-QSAR After exploring the effect of PAEs on the cell viability of the HepG2 cells, the mechanism was analyzed. The content of Nrf2, a key antioxidant protein, was determined, and a 3D-QSAR model of the effect of PAEs on Nrf2 was established. On this basis, we can predict the content of Nrf2 protein in HepG2 cells with different PAEs, and then understand the effect of PAEs on Nrf2 protein. On this basis, we established CoMSIA model, where q2 was 0.512 and r2 was 0.966 (Table2). The non-cross validation coefﬁcients were all greater than 0.9, indicating that the model was relatively stable. The training set correlation r2 between the predicted value Toxics 2021, 9, x FOR PEER REVIEWand the actual value was 0.9818 (Figure5E), indicating that the model has good predictive13 of 17 ability. In addition, the contribution rates of S, E, H, D and A in the CoMSIA1 model were 23.2%, 11.8%, 51.9%, 0 and 13.1% (Table2), respectively.

FigureFigure 5. 5.DMEP’s DMEP’s three-dimensional three-dimensional contour contour maps maps of of steric steric field field (A (A),), the the electrostatic electrostatic field field (B (),B), hydrophobic hydrophobic field field (C (),C), AcceptorAcceptor field field (D ()D in) in the the CoMSIA CoMSIA model. model. The The relation relation schema schema between between predicted predicted values values and and experimental experimental values values on on Nrf2 Nrf2 contentcontent (( E(E):: CoMFA).CoMFA).

TheMolecular-level experimental toxic and modelaction mechanisms predicted values are generally of Nrf2 and divided the difference into two categories: between themspecific are detailedtype and in non-specific Table4. Like type; the 3D-QSAR among them model, the of specific LD 50 and type IC refers50, the to contribution the presence of reactive substituent groups in the molecular structure of compounds, which were bio- chemically related to biological receptor molecules such as enzymes and proteins, which involved static electricity. Non-specific mechanism meant that molecules with similar structures could produce biological reactions with similar properties, and the reaction process was less dependent on special chemical structures. Organic matter enters the organism playing a role through the interaction with biofilms. The main controlling factor was the distribution ratio of chemical molecules in the organism and the water phase, related to the molecule’s hydrophobicity [36]. The addition of hydrophobic groups makes PAEs easier to enter cells. Generally speaking, the hydrophobicity of a molecule was closely related to the stereo field. Our results found that the changes in LD50, IC50 or Nrf2 were closely related to the hydrophobic field and the steric field, and the proportions were very large with the results like previous studies [33]. In fact, the hydrophobicity and volume of the groups had the most obvious effect on toxicity towards the HepG2 cells and Nrf2 expression. However, it is worth considering that the results for IC50 and Nrf2 were not similar, and that this opposite result may be due to the duration of exposure to PAEs However, it is worth considering that the IC50 results were not similar to those of Nrf2. This generalization of the result can help in predicting the toxicity of PAEs on the HepG2 cells and Nrf2 expression based on the sidechain and branch length of the PAEs without the necessity of experimentation.

Toxics 2021, 9, x FOR PEER REVIEW 12 of 17

Toxics 2021, 9, x FOR PEER REVIEW 12 of 17 Toxics 2021 , 9, 134 12 of 16 Toxics 2021, 9, x FOR PEER REVIEW 12 of 17

Toxics 2021, 9, x FOR PEER REVIEW 12 of 17

Toxics 2021, 9, x FOR PEER REVIEW Table 4 Actual value, predicted value and error of 3D-QSAR model of PAEs on cell Nrf2 content.12 of 17 Toxics 2021, 9, x FOR PEER REVIEWTable 4 Actual value, predicted value and error of 3D-QSAR model of PAEs on cell Nrf2 content.12 of 17 rates of S and H were relatively high, especially the contributionCoMSIA rate of H in the CoMFACoMSIA Toxics 2021, 9, x FOR PEER REVIEWTable Names 4 Actual value,Structures predicted value and error Exp. of (Nrf2) 3D-QSAR model of PAEs on cell Nrf2 content.12 of 17 CoMSIAPred. CoMSIARes. modelTable reachesNames 4 Actual 51.9% value, (TableStructures predicted2), which value has and shown error Exp. thatof (Nrf2) 3D non-speciﬁc-QSAR model reactions of PAEs on accounted cell Nrf2 content. for CoMSIAPred. CoMSIARes. a relativelyTableNames 4 large Actual proportion value,Structures predicted with probabilityvalue and error Exp. of charge (Nrf2)of 3D-QSAR transfer. model of PAEs on cell Nrf2 content. CoMSIAPred. CoMSIARes. TableNames 4 Actual value,Structures predicted value and error Exp. of(Nrf2) 3D-QSAR model of PAEs on cell Nrf2 content. Table 4 Actual value, predicted value and error of 3D-QSAR modelCoMSIA Pred.of PAEs on cell Nrf2CoMSIARes. content. Table 4. ActualNames value, predictedStructures value and error of Exp. 3D-QSAR (Nrf2) model of PAEs on cell Nrf2 content. CoMSIAPred. CoMSIARes. NamesDEHP Structures Exp.0.6449 (Nrf2) CoMSIA0.646 CoMSIA−0.0011 NamesDEHP Structures Exp.0.6449 (Nrf2) CoMSIA0.646Pred. CoMSIA−0.0011Res. Names Structures Exp. (Nrf2) Pred. Res. DEHP 0.6449 Pred.0.646 Res.−0.0011 DEHP 0.6449 0.646 −0.0011 DEHP 0.6449 0.646 −0.0011 DEHP 0.6449 0.646 −0.0011 DEHPDEHPDEP 0.64490.64491.2220 0.6460.6461.140 −0.0011−0.00110.082 DEP 1.2220 1.140 0.082 DEP 1.2220 1.140 0.082

DEP 1.2220 1.140 0.082

DEP 1.2220 1.140 0.082 DMP 1.1534 1.151 0.0024 DEPDEP 1.22201.2220 1.1401.140 0.082 0.082 DMP 1.1534 1.151 0.0024 DEP 1.2220 1.140 0.082 DMP 1.1534 1.151 0.0024

DMP 1.1534 1.151 0.0024 DMP 1.1534 1.151 0.0024 DMPDMEP 1.15341.1691 1.1511.170 0.0024−0.0009 DMEPDMP 1.16911.1534 1.1701.151 −0.00240.0009 DMP 1.1534 1.151 0.0024 DMEP 1.1691 1.170 −0.0009

DMEPDMEP 1.16911.1691 1.1701.170 −0.0009−0.0009 DMEP 1.1691 1.170 −0.0009 DMEPDHXP 0.68821.1691 0.6731.170 −0.01520.0009 DMEP 1.1691 1.170 −0.0009 DHXP 0.6882 0.673 0.0152 DHXPDHXP 0.68820.6882 0.6730.673 0.01520.0152 DHXP 0.6882 0.673 0.0152 DHXP 0.6882 0.673 0.0152 DHXP 0.6882 0.673 0.0152 DHXPDNOP 0.68820.7129 0.6730.720 −0.01520.0071

DNOPDNOP 0.71290.7129 0.7200.720 −0.0071−0.0071 DNOP 0.7129 0.720 −0.0071

DNOP 0.7129 0.720 −0.0071 DNOP 0.7129 0.720 −0.0071 DBPDNOPDBP 0.73630.73630.7129 0.7460.7460.720 −0.0097−0.00970.0071 DNOPDBP 0.73630.7129 0.7460.720 −−0.00970.0071 DBP 0.7363 0.746 −0.0097

DBP 0.7363 0.746 −0.0097

DBP 0.7363 0.746 −0.0097 TakingThe DMEP experimental as an example, and model wefound predicted that the valu greenes of area, Nrf2 yellow and the area difference and red area between DBPThe experimental and model predicted0.7363 valu es of Nrf2 and0.746 the difference−0.0097 between in thethem stericDBP are field, detailed hydrophobic in Table field,4. Like and the hydrogen 3D-QSAR0.7363 bond model acceptor of LD0.74650 field and wereIC50, the very contribution−0.0097 large (Figurethemrates5A,C,D).The areof S experimentaldetailed and The H more were in Table hydrogen relativelyand 4. model Like bondhigh, thepredicted acceptors 3D-QSARespecial valuly on modelthees the contributionofbranched Nrf2of LD and50 and chains therate IC differenceof50 of, Hthe PAEs, in contribution the thebetween CoMFA fewerthemratesmodel largeThe ofare groupsreaches S detailedexperimentaland H and 51.9% were in the Table (Table morerelatively and 4. hydrophilicmodel2), Like which high, the predicted especial3D-QSARhas groups, show valulyn thethemodel thates more contribution of non-specific ofNrf2 pronounced LD 50and and rate thereactions IC theofdifference50, Hthe decrease in accountedcontribution the betweenCoMFA in for The experimental and model predicted values of Nrf2 and the difference between Nrf2ratesmodelthema content relatively of are reaches S in anddetailed HepG2 large H 51.9% were proportion in cells Table(Table relatively (Figure 4. 2), withLike which5 high,A,C,D). probabilitythe has especial3D-QSAR Addingshow lyofn thechargethatmodel negatively contribution non-specific oftransfer. LD charged50 and rate reactions IC groupsof50 ,H the in accounted contribution onthe theCoMFA for themThe are experimentaldetailed in Table and 4. model Like thepredicted 3D-QSAR valu modeles of Nrf2of LD and50 and the IC difference50, the contribution between right-sidemodelarates relatively TheTakingof branch reaches S experimentaland large DMEP in H51.9% the wereproportion electrostaticas (Table relativelyan and example, 2), modelwith which fieldhigh, probability wepredicted equipotentialhas especialfound show thatofvalulyn chargethethat thees diagram contribution non-specificofgreen transfer.Nrf2 area, couldand yellow ratereactionsthe increase ofdifference areaH inaccounted the andthe Nrf2 betweenCoMFAred area for amodelthemrates relatively ofare reaches S detailedand large H51.9% proportionwere in Table(Table relatively 4. 2),with Like which high, probability the has especial3D-QSAR show oflyn charge thatthemodel contribution non-specific transfer. of LD50 and ratereactions IC of50, Hthe inaccounted contribution the CoMFA for content,themin the whileTaking are steric detailed adding DMEPfield, positively inhydrophobic as Table an example, 4. charged Like field, thewe groups foundand3D-QSAR hydrogen nearthat modelthe the greenbond carbonyl of LDarea,acceptor50group and yellow ICfield could50 area, thewere increase andcontribution very red largearea ratesmodel of reaches S and H 51.9% were (Table relatively 2), which high, hasespecial showlyn thethat contribution non-specific rate reactions of H in accounted the CoMFA for the Nrf2inarates(Figures relatively the contentTaking of steric S 5A,C,D). and large (FigureDMEPfield, H wereproportion hydrophobic The5asB). anrelatively more example, withhydrogen field,high, probabilitywe andfoundespecial bond hydrogen that ofacceptorsly charge thethe contributiongreenbond transfer.on acceptorarea,the branched yellow rate field of area H werechains in and the very ofred CoMFA PAEs,large area modela relatively reaches large 51.9% proportion (Table 2), with which probability has show ofn chargethat non-specific transfer. reactions accounted for in(FiguresMolecular-levelmodelthe the fewerTaking steric reaches 5A,C,D). large field,DMEP 51.9% groups toxic hydrophobicThe as (Table action anmore and example, 2), mechanismsthehydrogen whichfield, more we and hasfoundhydrophilic bond show arehydrogen thatacceptors generallyn thatthe groups, bond greennon-specific on divided acceptor thearea,the branchedmore intoyellow reactions field twopronounced area werechains categories: accounted and very of red thePAEs, large area de-for a relativelyTaking large DMEP proportion as an example, with probability we found that of charge the green transfer. area, yellow area and red area specific(Figurestheinacrease relatively the typefewer stericin and5A,C,D). largeNrf2 largefield, non-specific groupscontent proportion Thehydrophobic more and in type; HepG2the hydrogenwith amongfield,more probability cells andhydrophilic them,bond (Figureshydrogen acceptors theof charge specific groups,5A,C,D). bond ontransfer. type theacceptorthe Adding refersbranchedmore field topronouncednegatively the chainswere presence very of chargedthe PAEs, large de- in theTaking steric DMEPfield, hydrophobic as an example, field, we andfound hydrogen that the bondgreen acceptorarea, yellow field area were and very red large area of reactivethecrease(Figuresgroups fewerTaking substituentin on 5A,C,D). largeNrf2 the DMEP right-side groupscontent The groupsas an more andin example,branch in HepG2the hydrogen the more in molecular wethecells hydrophilicfound electrbond (Figures ostaticthat structureacceptors the groups,5A,C,D). field green ofon equipotential compounds, thethearea, Adding morebranched yellow pronouncednegatively diagram whicharea chains and were ofcould charged redthe PAEs, areade- in- in(Figures the steric 5A,C,D). field, hydrophobicThe more hydrogen field, and bond hydrogen acceptors bond on acceptorthe branched field werechains very of PAEs, large biochemicallycreasegroupstheincrease the fewer steric intheon related Nrf2 largeNrf2the field, right-side contentcontent,groups to hydrophobic biological and in whilebranch HepG2 the receptor adding field,morein thecells molecules andhydrophilic positielectr (Figures hydrogenvelyostatic such charged 5A,C,D).groups, field asbond enzymes equipotential groups acceptor theAdding more andnear field proteins,pronouncednegatively thediagram were carbonyl which very couldcharged the group large de-in- (Figuresthe fewer 5A,C,D). large groups The more and thehydrogen more hydrophilic bond acceptors groups, on thethe branchedmore pronounced chains of the PAEs, de- involvedgroupscrease(Figurescould static increasethe inon 5A,C,D). Nrf2the electricity. right-side thecontent,content TheNrf2 Non-specificmore while contentinbranch HepG2 hydrogen adding (Figurein thecells mechanism positi electr bond5B). (Figuresvely ostatic acceptors charged meant 5A,C,D). field thaton equipotentialgroups the moleculesAdding branched near negativelythe diagram with chainscarbonyl similar could ofcharged groupPAEs, in- thecrease fewer in largeNrf2 groupscontent andin HepG2the more cells hydrophilic (Figures groups,5A,C,D). the Adding more pronouncednegatively charged the de- structurescreasecouldgroupsthe fewer couldincreasethe on Nrf2 largethe produce right-sidethe content, groups Nrf2 biological contentwhileand branch the adding (Figure reactionsmorein the positihydrophilic electr5B). with velyostatic similarcharged groups, field properties, groupsequipotential the more near and pronouncedthe diagram the carbonyl reaction could thegroup de-in- creasegroups inon Nrf2 the right-side content inbranch HepG2 in thecells electr (Figuresostatic 5A,C,D). field equipotential Adding negatively diagram couldcharged in- processcouldcrease was increase thein less Nrf2 dependent the content,content Nrf2 content while onin specialHepG2 adding (Figure chemicalcells positi 5B). (Figures vely structures. charged 5A,C,D). groups Organic Adding near matter negativelythe carbonyl enters thecharged group groupscrease the on Nrf2the right-side content, while branch adding in the positi electrvelyostatic charged field equipotentialgroups near the diagram carbonyl could group in- organismcouldgroups playing increase on the a roleright-sidethe Nrf2 through content branch the interaction(Figure in the electr5B). with ostatic biofilms. field Theequipotential main controlling diagram factor could in- creasecould increasethe Nrf2 the content, Nrf2 contentwhile adding (Figure positi 5B). vely charged groups near the carbonyl group was thecrease distribution the Nrf2 content, ratio of while chemical adding molecules positively in thecharged organism groups and near the the water carbonyl phase, group relatedcould to the increase molecule’s the Nrf2 hydrophobicity content (Figure [36]. 5B). The addition of hydrophobic groups makes could increase the Nrf2 content (Figure 5B). PAEs easier to enter cells. Generally speaking, the hydrophobicity of a molecule was closely

Toxics 2021, 9, 134 13 of 16

related to the stereo field. Our results found that the changes in LD50, IC50 or Nrf2 were closely related to the hydrophobic field and the steric field, and the proportions were very large with the results like previous studies [33]. In fact, the hydrophobicity and volume of the groups had the most obvious effect on toxicity towards the HepG2 cells and Nrf2 expression. However, it is worth considering that the results for IC50 and Nrf2 were not similar, and that this opposite result may be due to the duration of exposure to PAEs However, it is worth considering that the IC50 results were not similar to those of Nrf2. Toxics 2021, 9, x FOR PEER REVIEW This generalization of the result can help in predicting the toxicity of PAEs on the HepG214 of 16 cells and Nrf2 expression based on the sidechain and branch length of the PAEs without the necessity of experimentation.

3.6. Applicability Domain We usedused normalisednormalised residualsresiduals andand levers,levers, visualisedvisualised inin WilliamsWilliams plotsplots withwith trainingtraining and validationvalidation sets, to evaluateevaluate AD.AD. AsAs shownshown inin FigureFigure6 ,6, the the horizontal horizontal and and vertical vertical dashed lineslines indicateindicate thethe boundariesboundaries ofof normalnormal values.values. TheThe horizontalhorizontal dasheddashed lineslines werewere thethe YY (i.e.,(i.e., normalised normalised residual) residual) outliers, outliers, which which were were set ±set2.5. ± 2.5. An unknownAn unknown compound compound was characterisedwas characterised as an as outlier an outlier if its if normalised its normalised residual residual exceeded exceeded the thresholdthe threshold value value of 2.5. of The2.5. The vertical vertical dashed dashed line madeline made the X the (i.e., X leverage)(i.e., leverage) outliers, outliers, warning warning of leverage of leverageh* = 0.692 h* = (A)0.692 or (A) 1.28 or (A, 1.28 B). (A, Figure B). 6FigureA shows 6A thatshows all that compounds all compounds were within were thewithin AD rangethe AD except range forexcept DMEP for (hiDMEP > h*) (hi in the> h*) training in the training set. Although set. Although methane methane shown high shown leverage, high itleverage, has small it normalisedhas small normalised residuals, suggestingresiduals, suggesting that it is a “goodthat it is high a “good leverage high point” leverage [37 ].point” [37].

Figure 6. Williams plot for verifying the applicability domain of the model. (A): LD50;(B): IC50;(C): Nrf2. Figure 6. Williams plot for verifying the applicability domain of the model. (A): LD50; (B): IC50; (C): Nrf2.

3.7. Statistical Validation

The results of the the test test are are shown shown in in Figure Figure 7.7 .The The predicted predicted pLD pLD50, 50IC,50 IC and50 and Nrf2 Nrf2 val- valuesues correlated correlated with with the the observed observed values values within within the the tolerable tolerable error error range. range. The The values values for 2 for the external2 R test were 1 (LD ), 0.767 (IC ) and 0.979 (Nrf2), supporting a good the external R2 test were 1 (LD50), 0.76750 (IC50) and50 0.979 (Nrf2), supporting a good correla- correlationtion between between the predicted the predicted and observed and observed values. values. However, However, one oneof the of Tropsha’s the Tropsha’s test test re- results for IC did not pass, indicating that the model for IC has some shortcomings in sults for IC5050 did not pass, indicating that the model for IC50 has some shortcomings in termsterms ofof predictivepredictive ability.ability.

Figure 7. Enalos Model Acceptability Criteria KNIME node screenshot. (A): LD50; (B): IC50; (C): Nrf2. Figure 7. Enalos Model Acceptability Criteria KNIME node screenshot. (A): LD50;(B): IC50;(C): Nrf2.

The Y-randomisation test is used to assess the robustness of the QSAR model. How- ever, the small amount of data in our study makes it difficult to apply the Y-randomisation test. Therefore, our study did not perform Y-randomisation tests, which is where our paper is flawed, and this would have some impact on the robustness of our model. Finally, the models developed within this study can be utilized for predicting the behavior of PAEs upon interaction with cells as well as to the design of new, less toxic PAEs in the future.

4. Conclusions In this paper, in vitro experiments found that PAEs after 72 h can inhibit the viability of the HepG2 cells to varying degrees, and at the same time, because of their molecular

Toxics 2021, 9, 134 14 of 16

The Y-randomisation test is used to assess the robustness of the QSAR model. How- ever, the small amount of data in our study makes it difﬁcult to apply the Y-randomisation test. Therefore, our study did not perform Y-randomisation tests, which is where our paper is ﬂawed, and this would have some impact on the robustness of our model. Finally, the models developed within this study can be utilized for predicting the behavior of PAEs upon interaction with cells as well as to the design of new, less toxic PAEs in the future.

4. Conclusions In this paper, in vitro experiments found that PAEs after 72 h can inhibit the viability of the HepG2 cells to varying degrees, and at the same time, because of their molecular differences, PAEs can increase or inhibit the content of Nrf2 protein. On this basis, this study established PAEs 3D-QSAR model for LD50, IC50 and Nrf2. The models (LD50 and Nrf2) built for prediction were relatively stable and had a good predictive effect. In terms of structure-effect analysis, it was found that the steric field and the hydrophobic field had the greatest impact, but the electric field also had a certain contribution value. The results show that the introduction of large and hydrophobic groups on the side and branched chains may significantly affect the toxic effects of PAEs on cells. The results observed in this study present a potential tool for predicting the LD50 and Nrf2 of new PAEs based on the structure of the compounds, and also provide a reference for the design of new less toxic PAEs in the future. However, the model in our article suffers from a lack of data, which is the key to the overall model not being as perfect as it could be. We hope to improve this issue in future experiments.

Author Contributions: H.L.: Conceptualization; Formal analysis; Methodology; Investigation; Writing—original draft. H.H.: Validation. X.X.: Methodology. Z.Z.: Formal analysis. C.L.: Conceptu- alization; Methodology; Project administration; Supervision; Writing—review & editing. All authors have read and agreed to the published version of the manuscript. Funding: This work was supported by National Key Research and Development Program of Thir- teenth Five-Year Plan (No. 2017YFC1601702). Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Data Availability Statement: Data presented in this study are available on request from the Authors. Conﬂicts of Interest: There are no conﬂict of interest to declare.

References 1. Schecter, A.; Lorber, M.; Guo, Y.; Wu, Q.; Yun, S.H.; Kannan, K.; Hommel, M.; Imran, N.; Hynan, L.S.; Cheng, D.; et al. Phthalate concentrations and dietary exposure from food purchased in New York State. Environ. Health Perspect. 2013, 121, 473–494. [CrossRef][PubMed] 2. Ventrice, P.; Ventrice, D.; Russo, E.; de Sarro, G. Phthalates: European regulation, chemistry, pharmacokinetic and related toxicity. Environ. Toxicol. Pharmacol. 2013, 36, 88–96. [CrossRef][PubMed] 3. Cirillo, T.; Fasano, E.; Castaldi, E.; Montuori, P.; Amodio Cocchieri, R. Children’s exposure to di (2-ethylhexyl) phthalate and dibutylphthalate plasticizers from school meals. J. Agric. Food. Chem. 2011, 59, 10532–10538. [CrossRef][PubMed] 4. Niu, L.; Xu, Y.; Xu, C.; Yun, L.; Liu, W. Status of phthalate esters contamination in agricultural soils across China and associated health risks. Environ. Pollut. 2014, 195, 16–23. [CrossRef][PubMed] 5. Wittassek, M.; Koch, H.M.; Angerer, J.; Brüning, T. Assessing exposure to phthalates—The human biomonitoring approach. Mol. Nutr. Food Res. 2011, 55, 7–31. [CrossRef] 6. Buckley, J.P.; Palmieri, R.T.; Matuszewski, J.M.; Herring, A.H.; Baird, D.D.; Hartmann, K.E.; Hoppin, J.A. Consumer product exposures associated with urinary phthalate levels in pregnant women. J. Expo. Sci. Environ. Epidemiol. 2012, 22, 468–475. [CrossRef] 7. Schecter, A. Dioxins and Health: Including Other Persistent Organic Pollutants and Endocrine Disruptors; John Wiley & Sons: Hoboken, NJ, USA, 2012. 8. Fromme, H.; Kuchler, T.; Otto, T.; Pilz, K.; Muller, J.; Wenzel, A. Occurrence of phthalates and bisphenol A and F in the environment. Water Res. 2002, 36, 1429–1438. [CrossRef] Toxics 2021, 9, 134 15 of 16

9. Wang, J.; Luo, Y.; Teng, Y.; Ma, W.; Christie, P.; Li, Z. Soil contamination by phthalate esters in Chinese intensive vegetable production systems with different modes of use of plastic film. Environ. Pollut. 2013, 180, 265–273. [CrossRef][PubMed] 10. Meng, X.Z.; Wang, Y.; Xiang, N.; Chen, L.; Liu, Z.; Wu, B.; Dai, X.; Zhang, Y.H.; Xie, Z.; Ebinghaus, R. Flow of sewage sludge-borne phthalate esters (PAEs) from human release to human intake: Implication for risk assessment of sludge applied to soil. Sci. Total Environ. 2014, 476–477, 242–249. [CrossRef][PubMed] 11. Pan, G.; Hanaoka, T.; Yoshimura, M.; Zhang, S.; Wang, P.; Tsukino, H.; Inoue, K.; Nakazawa, H.; Tsugane, S.; Takahashi, K. Decreased serum free testosterone in workers exposed to high levels of di-n-butyl phthalate (DBP) and di-2-ethylhexyl phthalate (DEHP): A cross-sectional study in China. Environ. Health Perspect. 2006, 114, 1643–1648. [CrossRef] 12. Swan, S.H.; Main, K.M.; Liu, F.; Stewart, S.L.; Kruse, R.L.; Calafat, A.M.; Mao, C.S.; Redmon, J.B.; Ternand, C.L.; Sullivan, S. Decrease in anogenital distance among male infants with prenatal phthalate exposure. Environ. Health Perspect. 2005, 113, 1056–1061. [CrossRef][PubMed] 13. Ward, J.M.; Peters, J.M.; Perella, C.M.; Gonzalez, F.J. Receptor and nonreceptor-mediated organ-specific toxicity of di (2-ethylhexyl) phthalate (DEHP) in peroxisome proliferator-activated receptorα-null mice. Toxicol. Pathol. 1998, 26, 240–246. [CrossRef] 14. Ye, H.; Ha, M.; Yang, M.; Yue, P.; Xie, Z.; Liu, C. Di2-ethylhexyl phthalate disrupts thyroid hormone homeostasis through activating the Ras/Akt/TRHr pathway and inducing hepatic enzymes. Sci. Rep. 2017, 7, 1–12. [CrossRef][PubMed] 15. Chen, H.; Zhang, W.; Rui, B.B.; Yang, S.M.; Xu, W.P.; Wei, W. Di(2-ethylhexyl) phthalate exacerbates non-alcoholic fatty liver in rats and its potential mechanisms. Environ. Toxicol. Pharmacol. 2016, 42, 38–44. [CrossRef] 16. Osburn, W.O.; Kensler, T.W. Nrf2 signaling: An adaptive response pathway for protection against environmental toxic insults. Mutat. Res. 2008, 659, 31–39. [CrossRef] 17. Gray, L.E., Jr.; Ostby, J.; Furr, J.; Price, M.; Veeramachaneni, D.R.; Parks, L. Perinatal exposure to the phthalates DEHP, BBP, and DINP, but not DEP, DMP, or DOTP, alters sexual differentiation of the male rat. Toxicol. Sci. 2000, 58, 350–365. [CrossRef] 18. Dearden, J.C. The history and development of quantitative structure-activity relationships (QSARs). In Oncology: Breakthroughs in Research and Practice; IGI Global: Hershey, PA, USA, 2017; pp. 1–55. [CrossRef] 19. Desai, N.C.; Bhatt, N.B.; Joshi, S.B.; Jadeja, K.A.; Khedkar, V.M. Synthesis, antimicrobial activity and 3d-qsar study of hybrid oxazine clubbed pyridine scaffolds. ChemistrySelect 2019, 4, 7541–7550. [CrossRef] 20. Afantitis, A.; Tsoumanis, A.; Melagraki, G. Enalos suite of tools: Enhancing cheminformatics and nanoinformatics through knime. Curr. Med. Chem. 2020, 6523–6535. [CrossRef][PubMed] 21. Chen, X.; Qin, Q.; Zhang, W.; Zhang, Y.; Zheng, H.; Liu, C.; Yang, Y.; Xiong, W.; Yuan, J. Activation of the PI3K–AKT–mTOR signaling pathway promotes DEHP-induced Hep3B cell proliferation. Food. Chem. Toxicol. 2013, 59, 325–333. [CrossRef] 22. Cho, H.Y.; Kleeberger, S.R. Nrf2 protects against airway disorders. Toxicol. Appl. Pharmacol. 2010, 244, 43–56. [CrossRef][PubMed] 23. Naidu, S.D.; Kostov, R.V.; Dinkova-Kostova, A.T. Transcription factors Hsf1 and Nrf2 engage in crosstalk for cytoprotection. Trends. Pharmacol. Sci. 2015, 36, 6–14. [CrossRef] 24. Gong, P.; Cederbaum, A.I. Nrf2 is increased by CYP2E1 in rodent liver and HepG2 cells and protects against oxidative stress caused by CYP2E1. J. Hepatol. 2006, 43, 144–153. [CrossRef][PubMed] 25. Luo, Y.; Li, X.N.; Zhao, Y.; Du, Z.H.; Li, J.L. DEHP triggers cerebral mitochondrial dysfunction and oxidative stress in quail (Coturnix japonica) via modulating mitochondrial dynamics and biogenesis and activating Nrf2-mediated defense response. Chemosphere 2019, 224, 626–633. [CrossRef][PubMed] 26. Sun, X.; Lin, Y.; Huang, Q.; Shi, J.; Qiu, L.; Kang, M.; Chen, Y.; Fang, C.; Ye, T.; Dong, S. Di (2-ethylhexyl) phthalate-induced apoptosis in rat INS-1 cells is dependent on activation of endoplasmic reticulum stress and suppression of antioxidant protection. J. Cell. Mol. Med. 2015, 19, 581–594. [CrossRef][PubMed] 27. Tang, X.; Wu, S.; Shen, L.; Wei, Y.; Cao, X.; Wang, Y.; Long, C.; Zhou, Y.; Li, D.; Huang, F. Di-(2-ethylhexyl) phthalate (DEHP)- induced testicular toxicity through Nrf2-mediated Notch1 signaling pathway in Sprague–Dawley rats. Environ. Toxicol. 2018, 33, 720–728. [CrossRef] 28. Liu, X.; Zhu, Q.; Zhang, M.; Yin, T.; Xu, R.; Xiao, W.; Wu, J.; Deng, B.; Gao, X.; Gong, W.; et al. Isoliquiritigenin Ameliorates Acute Pancreatitis in Mice via Inhibition of Oxidative Stress and Modulation of the Nrf2/HO-1 Pathway. Oxid. Med. Cell. Longev. 2018, 2018, 1–12. [CrossRef] 29. Golbraikh, A.; Tropsha, A. Beware of q2! J. Mol. Graph. Model. 2002, 20, 269–276. [CrossRef] 30. Babu, S.; Sohn, H.; Madhavan, T. Computational Analysis of CRTh2 receptor antagonist: A Ligand-based CoMFA and CoMSIA approach. Comput Biol. Chem. 2015, 56, 109–121. [CrossRef] 31. Liu, S.; Sun, S. Combined QSAR/QSPR, Molecular Docking, and Molecular Dynamics Study of Environmentally Friendly PBDEs with Improved Insulating Properties. Chem. Res. Chin. Univ. 2019, 35, 478–484. [CrossRef] 32. Hennig, C.; Cooper, D. Brief communication: The relation between standard error of the estimate and sample size of histomor- phometric aging methods. Am. J. Phys. Anthropol. 2011, 145, 658–664. [CrossRef] 33. Zhang, H.; Zhao, C.; Na, H. A Modified 3D-QSAR Model Based on Ideal Point Method and Its Application in the Molecular Modification of Plasticizers with Flame Retardancy and Eco-Friendliness. Polymers 2020, 12, 1942. [CrossRef][PubMed] 34. Fan, D.; Liu, J.; Wang, L.; Yang, X.; Zhang, S.; Zhang, Y.; Shi, L. Development of quantitative structure–activity relationship models for predicting chronic toxicity of substituted benzenes to daphnia magna. Bull. Environ. Contam. Toxicol. 2016, 96, 664–670. [CrossRef] Toxics 2021, 9, 134 16 of 16

35. Salahinejad, M.; Ghasemi, J.B. 3D-QSAR studies on the toxicity of substituted benzenes to Tetrahymena pyriformis: CoMFA, CoMSIA and VolSurf approaches. Ecotoxicol. Environ. Saf. 2014, 105, 128–134. [CrossRef] 36. Jaworska, J.S.; Schultz, T.W. Mechanism-based comparisons of acute toxicities elicited by industrial organic chemicals in procaryotic and eucaryotic systems. Ecotoxicol. Environ. Saf. 1994, 29, 200–213. [CrossRef] 37. Jaworska, J.; Nikolova, N.; Aldenberg, T. QSAR applicabilty domain estimation by projection of the training set descriptor space: A review. Altern. Lab. Anim. 2005, 33, 445–459. [CrossRef][PubMed]