Computational modeling, ligand-based drug design, drug-likeness and ADMET properties studies of series of chromen-2-ones analogues as anti-cancer agents

Abstract

Background

In spite of the significant escalation in the depth of our conception and regulation of breast cancer over the past decades, the malady is still a serious community health challenge globally and poses a substantial tasks. Selective estrogen modulators (SERMs) such as Tamoxifen are approved for the therapy of this illness but developed drug resistance and unwanted side effects such as endometrial cancer caused by the long-term Tamoxifen chemotherapy limit their therapeutic applicability. Hence, developing new ER⁺ drugs with better therapeutic effect is strongly needed. In an attempt to overcome this challenge, this research is aimed at designing novel chromen-2-one analogues with better inhibition capacity against MCF-7 breast cancer cell line via structural modification of the reference compound and predict their activities using a developed QSAR model.

Results

Four models were developed, and the first was selected for the design as it has the highest statistical parameters such as: coefficient of determination (R² = 0.950), cross-validation coefficient (Q_cv² = 0.912), adjusted R² (R_adj² = 0.935), and external validation R² (R_pred² = 0.7485). Twelve (12) new novel chromen-2-one analogs were designed through structural modification of the reference compound. Their activities was predicted using the selected model, and their pIC₅₀ was found to be better than that of the reference compound and standard drug (Tamoxifen) used in the research. Results of pharmacokinetic study of the designed compounds revealed that they possess drug-likeness properties as none of them violated the Lipinski’s rule of five while ADMET studies confirmed designed compounds 6, 8, 11 and 12 as orally safe and non-toxic. Furthermore, molecular docking analysis was performed between these orally safe designed compounds and the active site of the ER⁺ receptor and the result showed that they have higher binding affinities than the reference compound and the standard drug used for this research.

Conclusion

Hence, designed compounds 6, 8, 11 and 12 can be used as novel ER⁺ breast cancer drug candidates after performing in vivo and in vitro studies.

Background

The most frequently diagnosed form of cancer amongst female worldwide is breast cancer (BC) (Torre et al. 2015). In 2012, a 1.67 million estimated new cases of breast cancer were identified worldwide, and the figure is projected to rise to 1.7 million cases by 2020 according to a recent report (Forouzanfar et al. 2011). Estrogen receptor-α (ER-α) is one of the enormous superfamily of nuclear receptors, and overexpression of these receptors is seriously involved in at least 70% breast cancer patients (Sommer and Fuqua 2001). Estrogen receptors (ER⁺) support estrogenic actions in various significant biological progressions and play vibrant role in the discovery of therapeutic agents for the management of breast cancer (Traboulsi et al. 2017; Feitelson et al. 2015). In spite of the significant escalation in the depth of our conception and regulation of breast cancer over the past decades, the disorder is still a serious community health challenge over the world and poses a substantial tasks (Amir et al. 2010). It is a renowned fact that almost 70% of human breast cancers are hormone-dependent and Erα⁺ (Maurer et al. 2017). Endocrine therapy is considered as a promising treatment option as it aims in the blockage of the ER transcription effect. Thus, ERα has provided an ultimate pharmaceutical target and several ERα ligands were established as antagonists against ERα positive breast cancer (Traboulsi et al. 2017). A distinctive group of ligands that serve as antagonist in breast tissue but agonist in other tissues such as bone and cardiovascular system (Maruyama et al. 2013; Bai and Gust 2009). Due to their outstanding mode of action, SERMs are still essential anti-breast cancer agents with benefits in cardiovascular system and bone density maintenance in comparison with aromatase inhibitors and pure anti-estrogen (Kaur et al. 2014; Jordan 2007). The first SERM approved for the treatment of breast cancer is Tamoxifen that has triphenyl ethylene skeleton and a basic side chain (Fig. 1a), it provides indispensable therapies for a number of patients (Jordan et al. 2014). 4-hydroxy [4-OH] Tamoxifen, which is an active metabolite of Tamoxifen (Fig. 1b), displays better binding affinity for the ERα⁺. Several SERMs having numerous frames that mimics Tamoxifen were produced to treat and inhibit breast cancer growth (Wang et al. 2009).

Fig. 1

a Structure of Tamoxifen, b structure of 4-hydroxyTamoxifen

Even though SERMs have enhanced the treatment outcome for ERα⁺ breast cancer patients, unwanted side effects limit their therapeutic applicability relentlessly. For example, long-term Tamoxifen chemotherapy intensifies the occurrence of endometrial cancer due to their partial estrogenic activity on the endometrium (Chen et al. 2014). Another frequent deficiency that restricts their use is inherent and developed drug resistance, in which breast tumors become refractory to endocrine therapies and relapse (Garcia-Becerra et al. 2012). Hence, developing new ER drugs with better therapeutic effect is still needed.

Coumarine analogs are imperative class of pharmacologically active skeletons that possess a wide biological activities such as anti-cancer (Luo et al. 2017a, b, c), anti-HIV (Olomola et al. 2013), anti-microbial (Arshad et al. 2011), and anti-inflammatory activities (Chen et al. 2017). Coumarin’s therapeutic applicability depends on the substitution pattern, and as a result of their varied pharmacological effects in recent years they have attracted extreme attention (Bisi et al. 2017). Their anti-cancer properties were the most extensively studied among all their properties (Emami and Dadashpour 2015; Thakur et al. 2015). Anti-angiogenesis and independent induction of apoptosis are the major mechanisms of the anti-cancerous properties of Coumarin’s as revealed by several studies (Sinha et al. 2016). Computational methods of drug discovery were established to accelerate the drug discovery process, as it reduces the time, resources and facilitates the assessment of properties of new compounds such as effectiveness and poisonousness prior to their synthesis. A mathematical frameworks that interfaces the quantitative relationship between the activities of a compounds and their molecular structures in an equation format is referred to Quantitative Structure–Activity Relationships (QSAR) (Abdullahi et al. 2021). Safety and efficacy of a drug to the body system are the major causes that results in the failure of drug candidate. Consequently, it is essential to discover effective compounds with better ADMET and drug-likeliness properties (Abdullahi et al. 2022a, b).

This research is aimed at developing a robust QSAR models for the prediction of the anti-breast cancer activities of novel compounds with chromen-2-ones scaffold against MCF-7 cell line, design new compounds, perform ADMET and drug-likeness predictions and lastly perform molecular docking between the newly designed compounds and the active site of the ER⁺ receptor on the designed compounds to evaluate their drug-likeness characteristics.

Methods

Softwares and online tools employed for this study

This research was performed on HP laptop furnished with a dual-core Intel (R) PENTIUM (R) B940 CPU processor running at 2.0 GHz and 4.0 GB of RAM running on Windows 8. The following softwares were used in this research: Chemdraw 19.1, SPARTAN "14 v 1.1.0, Molegro Virtual Docker (MVD) and Discovery Studio. SwissADME and pkCSM online web tools are utilized in assessing the pharmacokinetics and ADMET properties of the molecules.

Chromen-2-ones data set retrieval and activity normalization

$${\text{pIC}}_{{{5}0}} = - {\text{ log}}_{{{1}0}} \left( {{\text{IC}}_{{{5}0}} \times { 1}0^{{ - {6}}} } \right).$$

(1)

Drawing of two-dimensional structures of the compounds and geometry optimization

Two-dimensional structures of the chromen-2-ones were drawn with the aid of PerkinElmer ChemDraw software utilizing ACS-1996 document settings and then converted to three-dimensional format using Spartan 14.0 software program. Geometry optimization of the analogs was performed on Spartan 14.0 interface by using density functional theory calculations with B3LYP/6-31G* basis set. 2D structures of all the chromen-2-ones analogues are presented in Table 1.

Table 1 2D structures of the Chromen-2-one derivatives and their experimental pIC₅₀

Molecular descriptors generation

The optimized structures in sdf format were imported to pharmaceutical data exploratory (PADEL) software package to compute the molecular descriptors that are responsible for the biological activities of the compounds (Yap 2011).

Data pretreatment and division

The result obtained from PADEL in excel worksheet was pretreated using Kennard-Stone data pretreatment software to eliminate redundant and non-relevant descriptors. Data division software was then utilized to partition the data set into training and test set. This data set was separated into 73% training set and 27% test set. This partitioning confirms that an interrelated principle can be utilized to estimate the biological activities of the test set (Kennard and Stone 1969).

QSAR model building

Regression study was achieved through Genetic Function Algorithm (GFA) in material studio 8.0 software package in which the biological activities (pIC₅₀) are the dependent variable and the physicochemical properties (descriptors) are the independent variables (Khaled 2011). The length of the regression equation was set to 6, and population and maximum generation were set to 1000 and 1500, respectively. The number of top regression models developed was set as 4. Mutation probability was 0.1, and the user defined smoothing parameter was 0.5

Internal validation of the developed QSAR model

The models generated were evaluated by means of Friedman’s Lack of Fit (LOF) which is used to measure the capability of a model. The revised formula for the Friedman’s lack of fit is represented by Eq. 2 below:

$${\text{LOF}} = \frac{{{\text{SEE}}}}{{\left( {1 - \frac{a + bq}{N}} \right)^{2} }}$$

(2)

SEE is the standard error of estimation, q is the total number of physicochemical parameters (descriptors) in the model, b is a user-defined smoothing parameter, a is the number of terms in the model, and N is the number molecules in the modeling (training) set. SEE is the standard error of estimation which is the same as the standard deviation of the model. A good model has lower SEE value. Equation 3 is used to compute SEE values of a model

$${\text{SEE}} = \frac{{\left( { Y_{{{\text{exp}}}} - Y_{{{\text{pred}}}} } \right)}}{{\left( {N - q - 1} \right)^{\frac{1}{2}} }}^{\frac{1}{2}}$$

(3)

Structure of a typical regression equation is of the form:

$$A \, = \, b_{1} \delta_{1} + \, b_{2} \delta_{2} + \, b_{3} \delta_{3} + \cdots + \, c$$

(4)

where A is the biological activity (pIC₅₀), b’s represents the coefficient of regression for the corresponding δ’s which are the independent variables that represents the molecular descriptors of a compound and c is the mathematical constant of regression.

The most frequently used internal assessment parameter for the QSAR model is correlation coefficient of the training set (R²). It provides explanation of the fragment of the total deviation of the model. The more closer the R² value is to unity the better the model generated. R² is calculated using Eq. (5) below:

$$R^{2} = 1 - \frac{{\sum (Y_{{{\text{exp}}}} - Y_{{{\text{pred}}}} )^{2} }}{{\sum (Y_{{{\text{exp}}}} - Y_{{{\text{mtraining}}}} )^{2} }}$$

(5)

where Y_exp, Y_pred, and Y_mtraining are the actual experimental activity, the predicted activity and the mean experimental activity of the training set.

Adjusted R² (R_adj²) value varies directly with a rise in the number of molecular descriptors; value of R² alone is insufficient for assessing the stability of a model. R² is adjusted to get a stable and reliable model. The adjusted R² is defined by Eq. (6) below:

$$R_{{{\text{adj}}}}^{2} = \, 1 - \, \left( {1 - \, R^{2} } \right)\frac{N - 1}{{N - P - 1}} = \frac{{\left( {N - 1} \right)R^{2} - q}}{N - P + 1}$$

(6)

where N is the number of samples in the training set, q = number of descriptors in the model (Abdullahi et al. 2021). The ability of a QSAR model to predict the activity of new molecules is determined using the cross-validation coefficient (Q_cv²), and it is calculated using Eq. (7) below:

$${\text{Q}}^{{2}}_{{{\text{cv}}}} = {1} - \frac{{\sum (Y_{pred} - Y_{exp} )^{2} }}{{\sum (Y_{exp} - Y_{mtraining} )^{2} }}.$$

(7)

External validation of a QSAR model

The external validation of the built QSAR models is evaluated based on the value of R_test² as defined in Eq. (8):

$$R_{{{\text{test}}}}^{2} = \, 1 - \frac{{\sum (Y_{{{\text{pred}}}} - Y_{{{\text{exp}}}} )^{2} }}{{\sum (Y_{{{\text{exp}}}} - Y_{{{\text{mtraining}}}} )^{2} }}.$$

(8)

Applicability domain

The determination of influential and outlier compounds used to develop a QSAR model is performed by studying its applicability domain (AD). The robustness and reliability of a model can as well be affirmed using the domain (Tropsha et al. 2003). Technique used to evaluate the AD of a QSAR model is the leverage approach, for a molecule the leverage h_i is defined by Eq. 9:

$$h_{i} = \, y_{i} \left( { \, Y^{T} Y} \right)^{ - 1} y^{T}_{i}$$

(9)

where y indicates the vector descriptor of the referred sample and Y signifies the matrix of the descriptor obtained from the training set descriptor values. The threshold leverage (h*) was computed using Eq. (10) below:

$$h^{*} = \frac{{3\left( {Q + 1} \right)}}{N}$$

(10)

N is the number of training set data and Q is the number of independent variables (descriptors) used in building the model. William’s plot is a plot of standardized residual values against the leverage values of molecules. It facilitates the viewing of AD of a QSAR model. When a leverage of a compound exceeds the threshold value (h*), it is alleged to have influence the performance of the model and the compound may be eliminated from the domain, compounds having residual values within ± 3 regions are not tagged as outliers since points lying within this region cover 99% of the normally distributed data (Abdullahi et al. 2022a, b). Hence, the leverage together with standardized residuals was jointly used to characterize and determine the applicability domain.

Quality assurance of the model

Internal and external validations parameters are used to assess the reliability and predictive ability of a QSAR model. Table 2 gives the general minimum requirement values for the assessment of a QSAR model.

Table 2 Minimum recommended values for acceptable QSAR model

ADMET and drug-likeness studies

Available online web sites such as SwissADME (http://www.swissadme.ch/index.php) and pkCSM (http://structure.bioc.cam.ac.uk/pkcsm) are applied to explore the drug-likeness and ADMET properties of the studied molecules. These sites allow researchers to discover an innovative drug candidate, to minimize the number of empirical experimentations and to upgrade the success rate (Abdullahi et al. 2021). Lipinski’s rule of five (ROF) was employed as the principal screening step for the drug-likeness properties trailed by computing the central ADMET properties which are measures of the pharmacokinetics of the molecules under research.

Molecular docking studies

Ligand–protein docking studies was performed to study the nature of binding interactions between the newly designed chromen-2-one analogs and the binding pocket of the ER⁺ receptor and to gain insight into the amino acid residues that are responsible for the ligand–protein interaction (Abdullahi et al. 2022a, b). The simulation studies was carried out on HP workstation furnished with a dual-core Intel (R) PENTIUM (R) B940 CPU processor running at 2.0 GHz and 4.0 GB of RAM running on Windows 8. X-ray crystallized structure of the ER⁺ protein was downloaded from protein data bank (pdb id = 3ERT) and was prepared using Molegro virtual docker by the elimination of excess water molecules and co-crystallized ligand enveloped in its crystal structure. Residues with structural errors were repaired and rebuilt. The active site of the 3ERT receptor was predicted and set inside a restricted sphere having X, Y, Z coordinates of 24.12, 3.48 and 20.11 Å, respectively. Ligand preparation was performed by optimization using DFT calculations with B3LYP/631G* basis set and then saved in pdb format. The docking algorithm were scored based on MolDock and Rerank scoring functions. Molegro virtual docker was utilized for the docking studies due to its ability to offer better and accurate results in comparison with other docking softwares. Discovery studio software was utilized to view the various ligand/protein interactions in the docked complexes.

Results

Four QSAR models were developed from the training set data using genetic function algorithm (GFA) coupled with multi linear regression (MLR), and their expressions are presented below:

Model 1

• Y = 0.342327907 *  apol + 0.002006877 * ATSC8m + 0.021947183 * ATSC7s − 2.110146447 * SM1_Dzm − 0.027702443 * SpAbs_Dzs + 0.122940438 * ZMIC4 − 9.882891756.

Model 2

• Y = 0.333966562 * apol + 0.001909583 * ATSC8m + 0.019049122 * ATSC7s − 2.079324191 * SM1_DzZ − 0.027112784 * SpAbs_Dzs + 0.119956742 * ZMIC4 − 9.456381109.

Model 3

• Y = 0.342932868 * apol + 0.002004154 * ATSC8m + 0.021734174 * ATSC7s − 2.067273713 * SM1_Dzm − 0.027821824 * SpAD_Dzs + 0.122642435 * ZMIC4 − 9.889860694.

Model 4

• Y = 0.334437304 * apol + 0.001906343 * ATSC8m + 0.018877487 * ATSC7s − 2.033875692 * SM1_DzZ − 0.027216796 * SpAD_Dzs + 0.119277728 * ZMIC4 −  − 9.454115852.

Discussion

Internal and external validations of the developed models

Internal and external validation of the developed QSAR models was performed to reveal their robustness, stability, reliability and predictability, and the results are presented in Table 2. As observed from the table, all the models passed the minimum requirements for an acceptable QSAR model with model 1 having the best statistical parameters. Hence, it was selected for further examination in this study. Values of different types of molecular descriptors that have appeared in model 1 together with experimental predicted pIC₅₀, and residuals for the training set data are presented Table 3.

Table 3 Values of different types of molecular descriptors that have appeared in model 1, experimental and predicted pIC₅₀

Moreover, external validation of the selected model was performed to determine its predictive ability. The value of the external validation regression coefficient (R_ext²) was found to be 0.745, and this value surpasses the minimum recommended value (R_ext² ≥ 0.6). This illustrates that the selected model is capable of providing a valid predictions of the activities of new compounds. Step by step calculation of the external prediction correlation coefficient (R_ext²) is shown in Table 4.

Table 4 Calculations of the external prediction correlation coefficient (R_ext²) of the test set data

Figure 2 represents a plot of the predicted pIC₅₀ for the model building (training) and test sets against the experimental pIC₅₀ values. Furthermore, the residual values of all the data sets were plotted against the experimental activities as illustrated in Fig. 3. The predicted activities of the compounds strongly agree with their corresponding experimental activities as observed from Fig. 2, and this affirms the predictive ability of the selected model. Additionally, as observed from Fig. 3 the residual values of the compounds reside on both sides of zero, and this suggests that the selected model did not demonstrate any relative and systematic error (Abdullahi et al. 2021).

Fig. 2

Plot of predicted against experimental Pic₅₀ of the training and test set

Fig. 3

Plot of standardized residuals against the experimental activities of the training and test set

William’s plot

Williams’s plot of the selected model is portrayed in Fig. 4. It can be observed that only six (1, 13, 19, 25, 36 and 40) compounds from the test set data are found to be beyond the defined domain of applicability, i.e., they have leverage values greater than the threshold value (h* = 0.656). These compounds are termed as influential compounds, and their high leverage values might be related to their differences structurally from the other compounds in the data set.

Fig. 4

Williams plot of the selected model

Mean effect of the relevant molecular descriptors

The individual impact and role of the relevant descriptors is designated by their values of mean effects. Key information’s on the effect of the molecular descriptors on a built QSAR model is offered by the mean effect of the descriptors. The magnitude and sign of the molecular descriptors coupled with their mean effect values illustrates their powerfulness in influencing the biological activity of a compound (Abdullahi et al. 2021). Positive mean effect value of a descriptor indicates that biological activity of a molecule rises with the raise in the value of the descriptor, while negative descriptor value suggests that the biological activity of a molecule rises with the decrease in the descriptor’s value. This implies that apol, ZMIC4 and ATSC7s descriptors increase the pIC₅₀ of the compounds when their values are increased, while increase in the values of SpAbs_Dzs, SM1_Dzm and ATSC8m descriptors will lessen the pIC₅₀ of the compounds (Abdullahi et al. 2021).

Mean effect value of descriptors is calculated using Eq. (11) below.

$${\text{MF}}_{j} = \, \frac{{\alpha_{j} \sum\nolimits_{j - 1}^{i - n} {\beta_{ij} } }}{{\sum\nolimits_{j}^{m} {\alpha_{j} } \sum\nolimits_{i}^{n} {\beta_{ij} } }}$$

(11)

where MF_j is the mean effect of molecular descriptor j in a model, α_j denotes the coefficient of the descriptor J in the model and β_ij is the value of the descriptor in the data matrix for each compound in the training set, m demonstrates the number of descriptors found in the model and n is the number of samples in the training set (Abdullahi et al. 2022a, b). Mean effect values of the relevant descriptors that appeared in the selected model are placed in Table 5.

Table 5 Mean effect values of the most relevant descriptors

Ligand-based drug designation

Ligand-based drug design of new novel chromen-2-ones was achieved through virtual screening technique based on the chosen QSAR model. Compound 10 from the training set data was utilized as the reference compound for the design as it has the highest pIC₅₀ (5.344) values and was excellently predicted by the selected model with low residual value (− 0.051), which is within the defined domain of applicability of the model. Twelve (12) new novel chromen-2-one analogs were designed through structural adjustment of the reference compound, and their 3D structures were geometrically optimized on Spartan 14.0 interface using DFT calculations with B3LYP/6-31G* basis set. Their pIC₅₀ was predicted using the selected model, and they were found to have improved pIC₅₀ which ranges from (5.472–8.584) compared to the reference compound and Tamoxifen (pIC₅₀ = 4.843) used in the research. The structure of the reference compound and the template used for the design is shown in Figs. 5 and 6; also, the structure of the newly designed compounds with their predicted pIC₅₀ is shown in Table 6.

Fig. 5

Structure of the reference compound

Fig. 6

Structure of the template used for the design

Table 6 Structure and predicted Pic₅₀ of the newly designed compounds pIC₅₀

Drug-likeness properties of the designed compounds

The standards used in the screening of drug candidates at the early phase of the drug discovery process are the assessment of their drug-likeness parameters. This is accomplished by correlating the physicochemical properties of a given molecule with its bio-pharmaceutical properties in human body, mostly, its impact on oral bioavailability (Bickerton et al. 2012). To affirm that the designed chromen-2-ones analogues are the feasible drugs, their ADMET and pharmacokinetic properties were evaluated using SwissADME. The most inventive and detailed analysis of drug-likeness properties was performed by Lipinski (Lipinski et al. 1997); it results to the prevalent “rule of five,” which propose that a molecule possess drug-likeness properties only when its molecular weight (mol. wt.) < 500, its number of hydrogen bond donors (HBD) < 5, its number of hydrogen bond acceptors (HBA) < 10, and its partition coefficient octanol/ water Log P < 5. Compounds that do not violate more than two (2) of the criteria are deemed to possess drug likeness properties. Results of the drug-likeness properties of the designed compounds are presented in Table 7. All the designed chromen-2-ones possess drug-likeness properties since they violated only one of the Lipinski’s rule of five criteria (Molecular weight > 500). They have optimum profile of permeability and bioavailability as indicated by their bioavailability score of 0.55 (Martin 2005). Furthermore, the synthetic accessibility values of the designed chromen-2-ones were evaluated, based on a scale ranging from 1 (easy to synthesize) and 10 (not easily synthesize). Their predicted synthetic accessibility values range from 4.51 to 5.24 (Table 7), and these values suggested that the designed compounds can be easily synthesized.

Table 7 Drug-likeness properties of the designed chromen-2-ones

ADMET properties of the designed compounds

ADMET properties of a molecule deals with its absorption, distribution, metabolism, excretion, and toxicity, in and through the human body. These properties constitute the pharmacokinetic profile of a drug molecule and is very essential in evaluating its pharmacodynamics activities. Many online tools and offline software programs are widely available for the prediction of ADMET properties of a molecule. In this study, pkCSM online server was used for this purpose. The results of predicted ADMET properties of the designed compounds are presented in Table 6. All the designed chromen-2-one analogues exhibit excellent human intestinal absorption between 92.520 to 100%, and these values exceed the minimum recommended percentage of absorption (30%), as such they are well absorbed by the human intestine. The allowed range of Blood–brain barrier (BBB) and the central nervous system (CNS) permeability is > 0.3 to <  − 1 log BB and >  − 2 to <  − 3 log PS; thus, the designed compounds have a high possibility of crossing the blood–brain barrier (BBB) and central nervous system (CNS) as their log BB and log PS are within the acceptable range (Umar et al. 2019). The biotransformation of a drug in a body is demonstrated by its enzymatic metabolism; hence, it is very essential to consider the drug’s metabolism. A category of super enzymes, cytochrome P450, plays a critical role in drug’s metabolism. CYP families accountable for the drug’s metabolism include 1A2, 2C9, 2C19, 2D6, and 3A4, among which the most essential is the 3A4 enzyme. All the designed chromen-2-ones are the substrate as well as inhibitors of the 3A4 enzyme. The relationship between the elimination rate of a drug and its concentration is explained by a parameter called total clearance; all the designed analogues possess high values of this parameter which are within the acceptable range of a drug candidate in a human body. Additionally, it is necessary to evaluate the toxicity and adverse side effects of drug candidate at the preclinical and clinical phase as the safety of the drug one the most important issue. Results from Table 8 show that compounds 6, 8, 11 and 12 are non-toxic. These compounds exhibit promising pharmacokinetics and ADMET properties; thus, they can be recommended as ER⁺ inhibitors and breast cancer drug candidates for further analysis.

Table 8 Predicted ADMET properties of the designed chromen-2-ones

Moreover, to further affirm drug-likeness properties of the orally safe compounds (6, 8, 11 and 12), their bioavailability radar and boiled egg plot were analyzed. The Bioavailability Radar allows a first glance at the drug-likeness of a molecule. The pink area signifies the optimum limit for each properties (lipophilicity: XLOGP3 between − 0.7 and + 5.0, size: MW between 150 and 500 g/mol, polarity: TPSA between 20 and 130 Ų, solubility: log S ≤ 6, saturation: fraction of carbons in the sp3 hybridization ≥ 0.25, and flexibility: ≤ 9 rotatable bonds (Daina et al. 2017). Designed compounds 6 and 11 are the most orally bioavailable (Fig. 7), since most of their predicted properties are within the pink region while 8 and 12 having most of their predicted properties placed outside the pink region are deemed to be not orally bioavailable. Similarly, as illustrated from their boiled egg plot (Fig. 8), they possess high gastrointestinal values as they are all located within the white area of the plot and are all P-gp substrate as indicated by their blue colors.

Fig. 7

Bioavailability radar designed compounds 6 (A), 8 (B), 11 (C) and 12 (D)

Fig. 8

Boiled egg plot of designed compounds 6, 8, 11 and 12

Molecular docking studies

Designed chromen-2-ones 6, 8, 11, and 12 being orally safe were docked into the active site of the ER⁺ receptor using Molegro virtual docker to reveal the nature of interactions with the amino acid residues at the active pocket of the receptor (pdb id = 3ERT). They have better docking scores compared to the reference compound, and the standard drug (Tamoxifen) utilized in this research. 3D structures of the prepared ER⁺ receptor and compound 10 are shown in Figs. 9 and 10 while the Docking scores and various kinds of interactions between the docked chromen-2-ones and the active site of the ER⁺ receptor are presented in Table 9, respectively.

Fig. 9

3D structure of the prepared ER⁺ receptor

Fig. 10

3D structure of compound 10

Table 9 Docking scores and various kinds of interactions of the designed chromen-2-ones and the active site of the ER⁺ receptor

Interpretation of the docking result

Designed compound 6 ( MolDock score = − 142.117, Rerank score = − 99.5654) interacted with the active site of the ER⁺ receptor via seven (7) Carbon–Hydrogen bonds and six (6) Pi-Alkyl hydrophobic Pi-Alkyl interactions. Carbon–Hydrogen bonds are between TYR526 and the Oxygen atom attached directly to the ethylpiperidine group at distance 2.83 Å, LYS529, CYS530, LYS531 and VAL 533 with Oxygen and Hydrogen atoms of the methoxy group attached to the chromen-2-one scaffold at distances 2.84 Å, 2.455 Å, 2.64 Å, and 2.839 Å. MET522 forms the other Carbon–Hydrogen bonds with Hydrogen atoms of the ethylpiperidine group at distances 1.586 Å and 2.287 Å. TYR526, LEU354, LEU536, VAL533 and MET522 forms Hydrophobic Pi-Alkyl interactions. 3D and 2D interactions of designed compound 6 with the active site of the ER⁺ receptor is shown in Fig. 11.

Fig. 11

3D and 2D interactions of designed compound 6 with the active site of the ER⁺ receptor

Designed compound 8 (MolDock score = − 166.475 Rerank score = − 102.019) interacted with the active site of the ER⁺ receptor through a conventional Hydrogen bond, three Carbon–Hydrogen bond, double Pi-Sulfur interactions, Pi-Pi stacked and amide-Pi stacked interactions, and several alkyl as well as Pi-Alkyl hydrophobic interactions. Para methoxy oxygen atom forms a conventional Hydrogen bond with LEU536 at distance 2.147 Å, VAL534 forms two carbon–hydrogen bonds with the methoxy group Hydrogen atoms at 3.04 and 2.82 Å, and the last Carbon–Hydrogen bond is between GLU380 and Hydrogen atom of the piperidine group at 3.011 Å. MET343 and MET522 form Pi-Sulfur interactions, and TRP383 and LEU525 form hydrophobic Pi-Pi stacked and Amide-Pi stacked interactions. Finally, LYS529, CYS530, VAL533, MET528, ALA350, LEU525, and LEU536 residues formed Alkyl and Pi-Alkyl Hydrophobic interactions. 3D and 2D interactions of designed compound 8 with the active site of the ER⁺ receptor are shown in Fig. 12.

Fig. 12

3D and 2D interactions of designed compound 8 with the active site of the ER⁺ receptor

Designed compound 11 (MolDock score = − 154.303 Rerank score = − 104.610) is found to have interacted with the binding pocket of the ER⁺ receptor through two (2) conventional Hydrogen bonds, five (5) Carbon–Hydrogen bonds, Pi-Sulfur, Amide-Pi stacked, Alkyl and Pi-Alkyl hydrophobic interactions. Carbonyl Oxygen atom and Hydrogen atom of the dimethyl amine groups form two conventional Hydrogen bonds with TRP383 and MET343 at distances 2.25 and 2.42 Å. THR347 forms two Carbon–Hydrogen bonds with Hydrogen atom of the piperidine and dimethyl amine group Hydrogen atom at 2.83 and 2.62 Å, methoxy group Hydrogen atoms attached to the chromen-2-ones scaffold forms two additional Carbon–Hydrogen bonds with ASN591 at 3.04 and 2.877 Å, VAL534 forms the last Carbon–Hydrogen bond with para methoxy group Hydrogen atom at distance 2.86 Å. MET522 forms a pair of Pi-Sulfur interactions, LEU525 forms an Amide-Pi stacked hydrophobic interactions, while ALA350, TRP383, MET522, LEU536, LEU525 and LYS529 residues formed Alkyl and Pi-Alkyl hydrophobic interactions. 3D and 2D interactions of designed compound 11 with the binding site of the ER⁺ receptor is portrayed in Fig. 13.

Fig. 13

3D and 2D interactions of designed compound 11 with the binding site of the ER⁺ receptor

The binding mode of designed compound 12 (MolDock score = − 170.357 Rerank score = − 171.531) with the active site of the ER⁺ receptor is through a single conventional Hydrogen bond, seven (7) Carbon–Hydrogen bonds, Pi-Sulfur and Amide-Pi stacked interactions, and several alkyl as well as Pi-Alky hydrophobic interactions. Methoxy group oxygen atom attached to the chromen-2-one frame forms a conventional Hydrogen bond with HIS524 at 2.34 Å distance, GLY520 forms a pair of Carbon–Hydrogen bonds with Oxygen and Hydrogen atoms of the methoxy group attached to the chromen-2-one scaffold at 2.71 Å and 1.54 Å, GLU533 residues forms another pair of Carbon–Hydrogen bonds with p-methoxy group Hydrogen atoms at distances 2.02 and 2.95 Å, ASP351 forms triple Carbon–Hydrogen bonds with Hydrogen atoms of the ethyl piperidine group at distances 2.87, 2.49 and 2.78 Å, respectively. MET343 forms double Pi-Sulfur interactions, LEU346 forms an Amide-Pi stacked and ALA350, LEU354, LEU536, LEU387, MET388, LEU346, MET421, LEU525 and LEU391 residues formed Alkyl and Pi-Alkyl Hydrophobic interaction. 3D and 2D binding modes of designed compound 12 with the active pocket of the ER⁺ receptor are shown in Fig. 14, respectively.

Fig. 14

3D and 2D binding modes of designed compound 12 with the active pocket of the ER⁺ receptor

The reference compound (Moldock score = − 142.022 Rerank score = − 108.524) interacted with the binding pocket of the ER⁺ receptor through two (2) conventional Hydrogen bonds, single Carbon–Hydrogen bond, and several Alkyl and Pi-Alkyl interactions. Carbonyl Oxygen atom of the chromen-2-one scaffold forms a conventional Hydrogen bond with CYS530 at distance 2.26 Å, and LEU525 forms the other conventional Hydrogen bond with Hydrogen atom of the amino group attached to the chromen-2-one frame at 2.54 Å. THR347 forms Carbon–Hydrogen bond with ethyl Hydrogen atom attached to the piperidine group at 2.67 Å. LEU346, ALA350, LEU525, LYS529, CYS530 and LEU525 residues formed Alkyl and Pi-Alkyl hydrophobic interactions. 3D and 2D interaction mode of the reference compound with the active site of the ER⁺ receptor is shown in Fig. 15.

Fig. 15

3D and 2D interaction mode of the reference compound with the active site of the ER⁺ receptor

Validation of docking protocol

For the validation of the docking results, the reference drug (Tamoxifen) which is also the co-crystallized ligand was redocked into the original binding pocket of the ER⁺ receptor. The redocked modes were compared with the original docking modes of the Tamoxifen with the binding site of the receptor. The original docking modes of interaction of Tamoxifen are through three (3) conventional Hydrogen bonds with water molecules, additional pair of conventional Hydrogen bonds, single Carbon–Hydrogen bond, Pi-Sulfur and Amide-Pi stacked interactions and several Pi-Alkyl interactions. Hydroxyl Oxygen atom and Hydrogen atoms of the trimethyl amine group form the three (3) conventional Hydrogen atoms with water molecules. Two (2) conventional Hydrogen bonds are between Hydroxyl group oxygen atom and ARG394 and GLU353 residues at 3.02 and 2.42 Å distances. ASP351 forms the Carbon–Hydrogen bonds with Nitrogen atom of the trimethyl amine group at 3.20 Å, phenyl ring intercalated in space and forms a single Pi-Sulfur interaction with MET343 and Amide-Pi stacked interaction LEU346. ALA350, LEU387, LEU346, LEU525 and MET421 residues formed Pi-Alkyl hydrophobic interactions. 3D and 2D original interaction modes of Tamoxifen with the active site of the 3ERT receptor is presented in Fig. 16.

Fig. 16

3D and 2D original interaction modes of Tamoxifen with the active site of the 3ERT receptor

The redocked Tamoxifen (MolDock score = − 155.184 Rerank score = − 144.415) interaction mode with the active pocket of the ER⁺ is through a conventional Hydrogen bond, five (5) Carbon–Hydrogen bonds and many Alkyl and Pi-Alkyl hydrophobic interactions. GLY420 forms a conventional Hydrogen bond with Hydrogen atom of the Hydroxyl group at 1.79 Å distance, GLY421 forms a carbon–hydrogen bond with Hydroxyl group oxygen atom at distance 2.37 Å, THR347 and ASP351 form three other Carbon–Hydrogen bonds with Hydrogen atoms at 2.89, 2.20, and 2.62 Å, ALA 350 forms another with trimethyl amine Hydrogen at 2.79 Å. PHE404, MET421, LEU346, LEU387, LEU 349 and LEU525 form hydrophobic Alkyl and Pi-Alkyl interactions. 3D and 2D interaction modes of redocked Tamoxifen with the active site of the ER⁺ receptor are shown in Fig. 17.

Fig. 17

3D and 2D interaction modes of redocked Tamoxifen with the active site of the ER⁺ receptor

Furthermore, the original and redocked Tamoxifen complexes were superimposed and aligned using Discovery studio. The root mean square deviation (RMSD) values between the superimposed proteins were found to be 1.15 Å, and this confirms the stability of the protein and that the Tamoxifen binds perfectly well to the binding pocket of the 3ERT receptor in the redocked complex (Umar et al. 2019). Thus, the docking protocol is reliably validated. The docking outcomes of the designed molecules suggested that Hydrogen bonds, electrostatic and hydrophobic (Alkyl and Pi-Alkyl) interactions are the central driving forces that regulate the binding interactions of the designed chromen-2-ones analogs and the active site residues of the ER⁺ receptor, and that the docking scores increase as the number of interactions increases. The designed compounds have higher docking scores than the reference and the standard drug (Tamoxifen), and this is related to the increase in the number of Hydrogen bonds and other interactions due to the presence of more substituent groups in the designed compounds (Abdullahi et al. 2022a, b).

Conclusions

In this study, a predictive QSAR model capable of explaining the structural requirements accountable for the anti-cancer activities of chromen-2-one analogs was developed using genetic function algorithm (GFA). Model 1 was selected for this research based on its excellent statistical parameters. Twelve new novel Chromen-2-one analogues were designed through the structural adjustment of compound 10 adopted as reference compound, and their activities was predicted using the selected model. They have improved pIC₅₀ which ranges from (5.472 to 8.584), compared to the reference compound (pIC₅₀ = 5.344) and Tamoxifen (pIC₅₀ = 4.843) utilized as control drug in the research. Moreover, the designed chromen-2-ones possess drug likeness properties since they do not violate the Lipinski’s rule of five, but ADMET studies showed that designed compounds 6, 8, 11 and 12 are orally safe and non-toxic. These orally safe designed compounds were subjected to molecular docking studies with the active site of the ER⁺ receptor kinase, and they were found to show promising binding scores compared to the reference compound and the standard drug used in the study. Hence, these compounds can be utilized as novel ER⁺ breast cancer drug candidates after performing in vivo and in vitro studies.