Human Estrogen Receptor Alpha Antagonists, Part 3: 3-D Pharmacophore and 3-D QSAR Guided Brefeldin A Hit-to-Lead Optimization toward New Breast Cancer Suppressants

The estrogen receptor α (ERα) is an important biological target mediating 17β-estradiol driven breast cancer (BC) development. Aiming to develop innovative drugs against BC, either wild-type or mutated ligand-ERα complexes were used as source data to build structure-based 3-D pharmacophore and 3-D QSAR models, afterward used as tools for the virtual screening of National Cancer Institute datasets and hit-to-lead optimization. The procedure identified Brefeldin A (BFA) as hit, then structurally optimized toward twelve new derivatives whose anticancer activity was confirmed both in vitro and in vivo. Compounds as SERMs showed picomolar to low nanomolar potencies against ERα and were then investigated as antiproliferative agents against BC cell lines, as stimulators of p53 expression, as well as BC cell cycle arrest agents. Most active leads were finally profiled upon administration to female Wistar rats with pre-induced BC, after which 3DPQ-12, 3DPQ-3, 3DPQ-9, 3DPQ-4, 3DPQ-2, and 3DPQ-1 represent potential candidates for BC therapy.


3-D Pharmacophore and 3-D QSAR Modeling and Models' Interpretation
SB 3-D pharmacophore hypotheses (3-D Phyp) and atom-based 3-D QSAR models were built with the TR using Schrödinger's PHASE program [88,89] and interpreted as a unique 3-D Phyp/3-D QSAR model ensemble. To derive the best PHASE hypotheses (associated with the highest q 2 values [90,91]), TR molecules were classified into "actives" and "inactives," using a pIC50 threshold value of 7.30, as suggested by the default settings (Tables 1 and 2). While searching for the optimal 3-D Phyp/3-D QSAR model ensemble, all the available pharmacophoric feature combinations were explored, from which both common pharmacophore hypothesis (CPH) and atom-based 3-D QSAR models were built (top hypotheses are displayed in Supplementary Material Table S2). Based on the highest associated q 2 values, the two best hypotheses were selected, ADDHHHP.13 and AD-DRRRP.11 (Table 4, Figure 4), herein named 3-D PhypI and 3-D PhypII, respectively. Both hypotheses consisted of one hydrogen-bond acceptor (A), two hydrogen-bond donators (D1 and D2), either three hydrophobic (H1, H2, and H3) or aromatic rings (R1, R2, and R3), and one with positively ionizable (P) features, which were coupled with the under-developing 3-D QSAR model PLS-coefficients contour maps revealing the areas associated to positive and negative steric (GREENPLS-coefficients and YELLOWPLS-coefficients) and HB bonding (BLUEPLS-coefficients and REDPLS-coefficients) interactions, respectively. Considering that in the PHASE definition, the H features are statistically more important, 3-D PhypI was consequently taken as the base model for the upcoming discussion (Table 4). Only the most important implications of two top hypotheses ( Figure 5 and Supplementary Materials Figures S1-S9) on the potency against ERα were presented, whereas the detailed analyses and comparison with previous hypotheses    Tables S3-S6) [80,92], whereas any lack of chance correlation was confirmed by employing Y-scrambling (Y-S) [

3-D Pharmacophore and 3-D QSAR Modeling and Models' Interpretation
SB 3-D pharmacophore hypotheses (3-D Phyp) and atom-based 3-D QSAR models were built with the TR using Schrödinger's PHASE program [88,89] and interpreted as a unique 3-D Phyp/3-D QSAR model ensemble. To derive the best PHASE hypotheses (associated with the highest q 2 values [90,91]), TR molecules were classified into "actives" and "inactives," using a pIC50 threshold value of 7.30, as suggested by the default settings (Tables 1 and 2). While searching for the optimal 3-D Phyp/3-D QSAR model ensemble, all the available pharmacophoric feature combinations were explored, from which both common pharmacophore hypothesis (CPH) and atom-based 3-D QSAR models were built (top hypotheses are displayed in Supplementary Material Table S2). Based on the highest associated q 2 values, the two best hypotheses were selected, ADDHHHP.13 and AD-DRRRP.11 (Table 4, Figure 4), herein named 3-D PhypI and 3-D PhypII, respectively. Both hypotheses consisted of one hydrogen-bond acceptor (A), two hydrogen-bond donators (D1 and D2), either three hydrophobic (H1, H2, and H3) or aromatic rings (R1, R2, and R3), and one with positively ionizable (P) features, which were coupled with the under-developing 3-D QSAR model PLS-coefficients contour maps revealing the areas associated to positive and negative steric (GREENPLS-coefficients and YELLOWPLS-coefficients) and HB bonding (BLUEPLS-coefficients and REDPLS-coefficients) interactions, respectively. Considering that in the PHASE definition, the H features are statistically more important, 3-D PhypI was consequently taken as the base model for the upcoming discussion (Table 4). Only the most important implications of two top hypotheses ( Figure 5 and Supplementary Materials Figures S1-S9) on the potency against ERα were presented, whereas the detailed analyses and comparison with previous hypotheses    Tables S3-S6) [80,92], whereas any lack of chance correlation was confirmed by employing Y-scrambling (Y-S) [80,92]. 10

3-D Pharmacophore and 3-D QSAR Modeling and Models' Interpretation
SB 3-D pharmacophore hypotheses (3-D Phyp) and atom-based 3-D QSAR models were built with the TR using Schrödinger's PHASE program [88,89] and interpreted as a unique 3-D Phyp/3-D QSAR model ensemble. To derive the best PHASE hypotheses (associated with the highest q 2 values [90,91]), TR molecules were classified into "actives" and "inactives," using a pIC50 threshold value of 7.30, as suggested by the default settings (Tables 1 and 2). While searching for the optimal 3-D Phyp/3-D QSAR model ensemble, all the available pharmacophoric feature combinations were explored, from which both common pharmacophore hypothesis (CPH) and atom-based 3-D QSAR models were built (top hypotheses are displayed in Supplementary Material Table S2). Based on the highest associated q 2 values, the two best hypotheses were selected, ADDHHHP.13 and AD-DRRRP.11 (Table 4, Figure 4), herein named 3-D PhypI and 3-D PhypII, respectively. Both hypotheses consisted of one hydrogen-bond acceptor (A), two hydrogen-bond donators (D1 and D2), either three hydrophobic (H1, H2, and H3) or aromatic rings (R1, R2, and R3), and one with positively ionizable (P) features, which were coupled with the under-developing 3-D QSAR model PLS-coefficients contour maps revealing the areas associated to positive and negative steric (GREENPLS-coefficients and YELLOWPLS-coefficients) and HB bonding (BLUEPLS-coefficients and REDPLS-coefficients) interactions, respectively. Considering that in the PHASE definition, the H features are statistically more important, 3-D PhypI was consequently taken as the base model for the upcoming discussion (Table 4). Only the most important implications of two top hypotheses ( Figure 5 and Supplementary Materials Figures S1-S9) on the potency against ERα were presented, whereas the detailed analyses and comparison with previous hypotheses    Tables S3-S6) [80,92], whereas any lack of chance correlation was confirmed by employing Y-scrambling (Y-S) [

3-D Pharmacophore and 3-D QSAR Modeling and Models' Interpretation
SB 3-D pharmacophore hypotheses (3-D Phyp) and atom-based 3-D QSAR models were built with the TR using Schrödinger's PHASE program [88,89] and interpreted as a unique 3-D Phyp/3-D QSAR model ensemble. To derive the best PHASE hypotheses (associated with the highest q 2 values [90,91]), TR molecules were classified into "actives" and "inactives," using a pIC50 threshold value of 7.30, as suggested by the default settings (Tables 1 and 2). While searching for the optimal 3-D Phyp/3-D QSAR model ensemble, all the available pharmacophoric feature combinations were explored, from which both common pharmacophore hypothesis (CPH) and atom-based 3-D QSAR models were built (top hypotheses are displayed in Supplementary Material Table S2). Based on the highest associated q 2 values, the two best hypotheses were selected, ADDHHHP.13 and AD-DRRRP.11 (Table 4, Figure 4), herein named 3-D PhypI and 3-D PhypII, respectively. Both hypotheses consisted of one hydrogen-bond acceptor (A), two hydrogen-bond donators (D1 and D2), either three hydrophobic (H1, H2, and H3) or aromatic rings (R1, R2, and R3), and one with positively ionizable (P) features, which were coupled with the under-developing 3-D QSAR model PLS-coefficients contour maps revealing the areas associated to positive and negative steric (GREENPLS-coefficients and YELLOWPLS-coefficients) and HB bonding (BLUEPLS-coefficients and REDPLS-coefficients) interactions, respectively. Considering that in the PHASE definition, the H features are statistically more important, 3-D PhypI was consequently taken as the base model for the upcoming discussion (Table 4). Only the most important implications of two top hypotheses ( Figure 5 and Supplementary Materials Figures S1-S9) on the potency against ERα were presented, whereas the detailed analyses and comparison with previous hypotheses    Tables S3-S6) [80,92], whereas any lack of chance correlation was confirmed by employing Y-scrambling (Y-S) [

3-D Pharmacophore and 3-D QSAR Modeling and Models' Interpretation
SB 3-D pharmacophore hypotheses (3-D Phyp) and atom-based 3-D QSAR models were built with the TR using Schrödinger's PHASE program [88,89] and interpreted as a unique 3-D Phyp/3-D QSAR model ensemble. To derive the best PHASE hypotheses (associated with the highest q 2 values [90,91]), TR molecules were classified into "actives" and "inactives," using a pIC50 threshold value of 7.30, as suggested by the default settings (Tables 1 and 2). While searching for the optimal 3-D Phyp/3-D QSAR model ensemble, all the available pharmacophoric feature combinations were explored, from which both common pharmacophore hypothesis (CPH) and atom-based 3-D QSAR models were built (top hypotheses are displayed in Supplementary Material Table S2). Based on the highest associated q 2 values, the two best hypotheses were selected, ADDHHHP.13 and AD-DRRRP.11 (Table 4, Figure 4), herein named 3-D PhypI and 3-D PhypII, respectively. Both hypotheses consisted of one hydrogen-bond acceptor (A), two hydrogen-bond donators (D1 and D2), either three hydrophobic (H1, H2, and H3) or aromatic rings (R1, R2, and R3), and one with positively ionizable (P) features, which were coupled with the under-developing 3-D QSAR model PLS-coefficients contour maps revealing the areas associated to positive and negative steric (GREENPLS-coefficients and YELLOWPLS-coefficients) and HB bonding (BLUEPLS-coefficients and REDPLS-coefficients) interactions, respectively. Considering that in the PHASE definition, the H features are statistically more important, 3-D PhypI was consequently taken as the base model for the upcoming discussion (Table 4). Only the most important implications of two top hypotheses ( Figure 5 and Supplementary Materials Figures S1-S9) on the potency against ERα were presented, whereas the detailed analyses and comparison with previous hypotheses    Tables S3-S6) [80,92], whereas any lack of chance correlation was confirmed by employing Y-scrambling (Y-S) [80,92]. 10.00 [77] a Partial agonist; b H12: closed conformation.

3-D Pharmacophore and 3-D QSAR Modeling and Models' Interpretation
SB 3-D pharmacophore hypotheses (3-D Phyp) and atom-based 3-D QSAR models were built with the TR using Schrödinger's PHASE program [88,89] and interpreted as a unique 3-D Phyp/3-D QSAR model ensemble. To derive the best PHASE hypotheses (associated with the highest q 2 values [90,91]), TR molecules were classified into "actives" and "inactives," using a pIC 50 threshold value of 7.30, as suggested by the default settings (Tables 1 and 2). While searching for the optimal 3-D Phyp/3-D QSAR model ensemble, all the available pharmacophoric feature combinations were explored, from which both common pharmacophore hypothesis (CPH) and atom-based 3-D QSAR models were built (top hypotheses are displayed in Supplementary Material Table S2). Based on the highest associated q 2 values, the two best hypotheses were selected, ADDHHHP.13 and ADDRRRP.11 (Table 4, Figure 4), herein named 3-D PhypI and 3-D PhypII, respectively. Both hypotheses consisted of one hydrogen-bond acceptor (A), two hydrogen-bond donators (D 1 and D 2 ), either three hydrophobic (H 1 , H 2 , and H 3 ) or aromatic rings (R 1 , R 2 , and R 3 ), and one with positively ionizable (P) features, which were coupled with the under-developing 3-D QSAR model PLS-coefficients contour maps revealing the areas associated to positive and negative steric (GREEN PLS-coefficients and YELLOW PLS-coefficients ) and HB bonding (BLUE PLS-coefficients and RED PLS-coefficients ) interactions, respectively. Considering that in the PHASE definition, the H features are statistically more important, 3-D PhypI was consequently taken as the base model for the upcoming discussion (Table 4). Only the most important implications of two top hypotheses (Figures 5 and S1-S9) on the potency against ERα were presented, whereas the detailed analyses and comparison with previous hypotheses    Tables S3-S6) [80,92], whereas any lack of chance correlation was confirmed by employing Y-scrambling (Y-S) [80,92].  rescoring; e Site score-an RMDS value for the site points superimposition in an alignment to the pharmacophore of the structures that contribute to this hypothesis; f Vector alignment score; g Volume of the contributing structures' overlap when aligned on the pharmacophore; h Selectivity-the fraction of molecules matching the hypothesis regardless of their potency; i Matches-number of actives that match the hypothesis; j Activity-Activity of the reference ligand (pIC50); k Inactive-Survival score of inactives; l PLS factor, i.e., N/5, where N is the number of ligands present in the training set; m Conventional square-correlation coefficient. n Standard deviation of regression; o Ratio of the model variance to the observed activity variance; p Significance level of variance ratio; q Stability of the model predictions to changes in the training set composition; r Cross-validation correlation coefficient using the leave-one-out (LOO) method. s Cross-validation correlation coefficient using the leave-some-out (LSO) method with 5 random groups; t Average cross-validation correlation coefficient using the leave-one-out (LOO) method obtained after Y-scrambling process. u Average cross-validation correlation coefficient using the leave-some-out (LSO) method with 5 random groups obtained after the Y-scrambling process.
The H1 (R1) feature/GREENPLS-coefficients/YELLOWPLS-coefficients ( Figure 5 and Supplementary Materials Figures S1-S9) suggested that the 1 st PhOH and 2 nd PhOH should be interconnected with five-membered (1ERR, Table 1, Figure 6A and Supplementary Materials Figure S5A [13]) or six-membered heterocyclic aliphatic bridge (1XP1, Table 1, Figure 6C and Supplementary Materials Figure S5C [64]), to interact with H6 Met388 H6-to-H7 loop residues Phe404, Ile424, and Leu428, maintaining the voluminosity toward distinct residues as low as possible [66]; according to the BLUEPLS-coefficients, the bridge may be improved by means of an HBD, to face H3 Glu353 or H3 Thr347 (see 1XP1, Table 1, Figure 5C [64]).  rescoring; e Site score-an RMDS value for the site points superimposition in an alignment to the pharmacophore of the structures that contribute to this hypothesis; f Vector alignment score; g Volume of the contributing structures' overlap when aligned on the pharmacophore; h Selectivity-the fraction of molecules matching the hypothesis regardless of their potency; i Matches-number of actives that match the hypothesis; j Activity-Activity of the reference ligand (pIC50); k Inactive-Survival score of inactives; l PLS factor, i.e., N/5, where N is the number of ligands present in the training set; m Conventional square-correlation coefficient. n Standard deviation of regression; o Ratio of the model variance to the observed activity variance; p Significance level of variance ratio; q Stability of the model predictions to changes in the training set composition; r Cross-validation correlation coefficient using the leave-one-out (LOO) method. s Cross-validation correlation coefficient using the leave-some-out (LSO) method with 5 random groups; t Average cross-validation correlation coefficient using the leave-one-out (LOO) method obtained after Y-scrambling process. u Average cross-validation correlation coefficient using the leave-some-out (LSO) method with 5 random groups obtained after the Y-scrambling process.
The H1 (R1) feature/GREENPLS-coefficients/YELLOWPLS-coefficients ( Figure 5 and Supplementary Materials Figures S1-S9) suggested that the 1 st PhOH and 2 nd PhOH should be interconnected with five-membered (1ERR, Table 1, Figure 6A and Supplementary Materials Figure S5A [13]) or six-membered heterocyclic aliphatic bridge (1XP1, Table 1, Figure 6C and Supplementary Materials Figure S5C [64]), to interact with H6 Met388 H6-to-H7 loop residues Phe404, Ile424, and Leu428, maintaining the voluminosity toward distinct residues as low as possible [66]; according to the BLUEPLS-coefficients, the bridge may be improved by means of an HBD, to face H3 Glu353 or H3 Thr347 (see 1XP1, Table 1, Figure 5C [64]). Site score-an RMDS value for the site points superimposition in an alignment to the pharmacophore of the structures that contribute to this hypothesis; f Vector alignment score; g Volume of the contributing structures' overlap when aligned on the pharmacophore; h Selectivity-the fraction of molecules matching the hypothesis regardless of their potency; i Matches-number of actives that match the hypothesis; j Activity-Activity of the reference ligand (pIC 50 ); k Inactive-Survival score of inactives; l PLS factor, i.e., N/5, where N is the number of ligands present in the training set; m Conventional square-correlation coefficient. n Standard deviation of regression; o Ratio of the model variance to the observed activity variance; p Significance level of variance ratio; q Stability of the model predictions to changes in the training set composition; r Cross-validation correlation coefficient using the leave-one-out (LOO) method. s Cross-validation correlation coefficient using the leave-some-out (LSO) method with 5 random groups; t Average cross-validation correlation coefficient using the leave-one-out (LOO) method obtained after Y-scrambling process. u Average cross-validation correlation coefficient using the leave-some-out (LSO) method with 5 random groups obtained after the Y-scrambling process.
The D 1 /RED PLS-coefficients (Figures 5 and S1-S9) emphasized that the ERα binder should possess the mixed hydrogen bond donating (HBD)/hydrogen bond accepting (HBA) functional group (like the frequently present aromatic hydroxyl group, i.e., 1st PhOH, as in 1ERR, Table 1, Figure 5A, [13]), to form hydrogen bonds (HBs) with H3 Glu353 and H6 Arg394, at the same time not too voluminous, according to the YELLOW PLS-coefficients maps.
The H 1 (R 1 ) feature/GREEN PLS-coefficients /YELLOW PLS-coefficients (Figures 5 and S1-S9) suggested that the 1st PhOH and 2nd PhOH should be interconnected with five-membered (1ERR,  [64]), to interact with H6 Met388 H6-to-H7 loop residues Phe404, Ile424, and Leu428, maintaining the voluminosity toward distinct residues as low as possible [66]; according to the BLUE PLS-coefficients , the bridge may be improved by means of an HBD, to face H3 Glu353 or H3 Thr347 (see 1XP1, Table 1, Figure 5C     For the clarity of presentation, only the H12 helix is presented in a cornflower blue ribbon, as a crucial delimiter for partial agonists, SERMs, and SERDs. The H 2 (R 2 ) feature/GREEN PLS-coefficients /YELLOW PLS-coefficients (Figures 5 and S1-S9) indicated that the chemical linker between the 1st PhOH and the 2nd PhOH should not be further degraded (for instance toward the ethyl group of 3ERD [69], Table 1, Figures 5B and S5A), to avoid ERα partial agonism and pure ERβ antagonism and that the bulkiness of 2nd Ph-OH toward H6 Met388 and H6-to-H7 loop residues Phe404, Ile424, and Leu428 is sufficient as is.
TS CRY 's experimentally available binding conformation's pK i values (herein improperly assumed as pIC 50 s) were thereafter predicted with an average absolute error of predictions (AAEPs) of 0.66 and 2.35 for the model optimized with LOO and LSO CVs, respectively (Table 5) and associated predictive q 2 (q 2 pred ) values were 0.51 and 0.39, respectively. Interestingly and as expected, the SB re-aligned molecules were predicted with lower errors (q 2 pred /AAEP values of 0.46/1.27 and 0.46/1.27 for LOO and LSO derived models) than those LB re-aligned (q 2 pred /AAEP values of 0.29/1.37 and 0.31/1.40 for LOO and LSO derived models). These values indicated the good predictive ability [108][109][110] of the 3-D PhypI/3-D QSAR model ensemble and support the goodness of the realignment methodology.   Table S17). Compound coded as NCI89671, a naturally occurring compound Brefeldin A (BFA, Figure 6A) [111], as the most potency predicted, did exert promising activity against ERα (IC 50 of 8.34 µM) and the MCF-7 cell line (IC 50 of 9.01 µM), and selectivity against the MDA-MB-231 cell line (selectivity index (SI) of 11.10), although less potent than the references E 2 [13], 4-hydroxytamoxifen (4-OHT) [32], and raloxifene (Ral) [13] (Supplementary  Materials Table S17). Previously assessed anti-BC properties of BFA and its derivatives were associated with the apoptosis and the compounds' ability to disrupt the cis-Golgi apparatus [112,113]. Interestingly, C4-and C7-esters of BFA exerted nM antiproliferative activity against MCF-7 cell lines [114], C4-succinyl, glutaryl BFA analogs, and C7-long lipids derivatives showed µM to nM potencies against MCF-7 cell lines [115], whereas the sulfide-and sulfoxide-conjugated BFA analogs were active against MDA-MC-435 cell lines as µM and sub-micromolar ranges [116].
BFA binding mode analysis showed an interaction profile as a putative partial agonist, likely inducing the H12 in a closed conformation ( Figure 6B) [13]. Thus, the BFA's cyclopentane ring and the C7-OH group formed H-bonds with H3 Glu353 and H6 Arg394 (d HB = 2.855 and 2.990 Å, respectively). Moreover, the C4-OH portion established the electrostatic interactions with H3 Glu353. On the other hand, the close contact of the C15-CH 3 with H11 His524 was accounted as unfavorable by the 3-D PhypI/3-D QSAR model ensemble, suggesting the insertion of either HBA or HBD functionality. Consequently, the C1-to-C4 carbon atoms were interfaced to H12, whereas the C9-to-C15 skeleton was engaged in van der Waals interactions with H6 Met388 and H6-to-H7 loop residues Ile423 and Leu428. Finally, the C1 carbonyl group was observed away from any interesting interactions, not satisfying any 3-D PhypI/3-D QSAR model features, indicating it as a possible substitution point into an HBA group. Hence, the 3-D PhypI/3-D QSAR model ensemble indicated that the modification of the C15-CH 3 into C15-OH could endow BFA's horizontal flip toward Glu353/Arg394, at the same time positioning the cyclopentane ring's C7-OH group toward the His524 (an alignment comparable to the E 2 s D ring and C17-OH group experimental conformation [13]). In such a scenario the C1 carbonyl group would face Glu353 and the C-4 OH group would become a further anchor point for the implementation of a Ph-containing scaffold. engaged in van der Waals interactions with H6 Met388 and H6-to-H7 loop residues Ile423 and Leu428. Finally, the C1 carbonyl group was observed away from any interesting interactions, not satisfying any 3-D PhypI/3-D QSAR model features, indicating it as a possible substitution point into an HBA group. Hence, the 3-D PhypI/3-D QSAR model ensemble indicated that the modification of the C15-CH3 into C15-OH could endow BFA's horizontal flip toward Glu353/Arg394, at the same time positioning the cyclopentane ring's C7-OH group toward the His524 (an alignment comparable to the E2′s D ring and C17-OH group experimental conformation [13]). In such a scenario the C1 carbonyl group would face Glu353 and the C-4 OH group would become a further anchor point for the implementation of a Ph-containing scaffold.

Rules for the Rational Design of Novel Brefeldin A Derivatives as SERMs
The BFA structural optimization toward novel ERα SERMs (Table 6) was thereafter performed by applying the guidelines from the 3-D PhypI/3-D QSAR model ensemble, applicable only for the rational design of SERMs. The partial agonist-to-SERM conversion was undertaken by applying the following strategies: 1. The BFA's C15-CH3 group was converted to C15-OH as a mixed HBA/HBD functional group to increase the compounds' capacity for establishing hydrogen bonds with either H3 Glu353 and H6 Arg394 (or H11 His524) and hopefully the solubility (data not shown). 2. The BFA's C4-OH was substituted with 3-acetyl-4-hydroxybenzoic acid to provide interactions with H6 Trp383 and H3 Thr347, as well as to stabilize the H3 Thr347-Leu525-H12 Leu536 hydrophobic network, and consequent H12 dislocation. Choosing 3-acetyl-4-hydroxybenzoic acid as a BFA's C4-OH substituent was an experimentally-guided decision since the tentative attempts to synthetically incorporate (see further text) the 1-(1,4-dihydroxynaphthalen-2-yl)ethenone as a fragment, perhaps more suitable to target H6 Trp383 by means of steric interactions, failed.

Rules for the Rational Design of Novel Brefeldin A Derivatives as SERMs
The BFA structural optimization toward novel ERα SERMs (Table 6) was thereafter performed by applying the guidelines from the 3-D PhypI/3-D QSAR model ensemble, applicable only for the rational design of SERMs. The partial agonist-to-SERM conversion was undertaken by applying the following strategies: 1.
The BFA's C15-CH 3 group was converted to C15-OH as a mixed HBA/HBD functional group to increase the compounds' capacity for establishing hydrogen bonds with either H3 Glu353 and H6 Arg394 (or H11 His524) and hopefully the solubility (data not shown).

2.
The BFA's C4-OH was substituted with 3-acetyl-4-hydroxybenzoic acid to provide interactions with H6 Trp383 and H3 Thr347, as well as to stabilize the H3 Thr347-Leu525-H12 Leu536 hydrophobic network, and consequent H12 dislocation. Choosing 3-acetyl-4-hydroxybenzoic acid as a BFA's C4-OH substituent was an experimentallyguided decision since the tentative attempts to synthetically incorporate (see further text) the 1-(1,4-dihydroxynaphthalen-2-yl)ethenone as a fragment, perhaps more suitable to target H6 Trp383 by means of steric interactions, failed.  (Tables 1 and 2) in a way that their HBD functional groups could primarily engage H3 Asp351, thus influencing, alongside the steric pressure, the H12 s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues. pressure, the H12′s induced fitting, whereas the existing HBA functional groups could produce additional favorable interactions with the surrounding residues.  Figures S23 and S24) and the pIC 50 prediction procedures against ERα (Table 6). This way, the designed compounds composed the ultimate prediction set [109,110] for the 3-D PhypI/3-D QSAR model ensemble, in which the SB and LB models' associated q 2 pred and AAEP values were 0.858/0.045 and 0.732/0.1, respectively. Indeed, even eight compounds, namely 3DPQ-12, 3DPQ-3, 3DPQ-9, 3DPQ-4, 3DPQ-2, 3DPQ-1,  3DPQ-7, and 3DPQ-11 were predicted as more potent than 1ERR [13] (the most potent TR compound; see further text).

Synthesis of Brefeldin A Derivatives 3DPQ-1 to 3DPQ-12
Designed compounds 3DPQ-1 to 3DPQ-12 were synthesized in high yields and purities (Scheme 1). The synthetic protocols and associated 1 H NMR, 13 C NMR, 15 N NMR, and 17 O NMR spectral data, as well as the HPLC spectra confirming compounds' purity of 95% and higher, are reported in Supplementary Materials (Synthetic protocols for the preparation of compounds 3DPQ-1 to 3DPQ-12, Synthesized Compounds spectral data interpretation, Supplementary Materials Figures S26-S190).
Compared to BFA, in all the synthesized compounds, the C15-CH 3 to C15-OH conversion seemed to participate in an ERα's LDB main core horizontal flipping (Figures 7 and S195). Thus, the C15-OH faced the H3 Glu353 and H6 Arg394 to establish two further HBs (see Supplementary Materials Table S18 for details). Consequently, the C1 carbonyl portion produced weak electrostatic interactions with H6 Trp383 s indole ring nitrogen. The C8-C15 carbon skeleton was observed to be sterically attracted by H6 Met388 and H6-to-H7 loop residues Ile423 and Leu428. The inverse alignment of the main core influenced the spatial positioning of the cyclopentane ring's C7-OH, as well, which produced HBs with H11 His524 (see Supplementary Materials Table S18 for details). The remaining C1-C4 carbon backbone participated in steric hindrance with H6 Trp383. Furthermore, the esterification of the C4-OH portion with 3-acetyl-4-hydroxybenzoic acid influenced the H3 Thr347-H11 Leu525-H12 Leu536 hydrophobic network [13,69] formation: the ester oxygen electrostatically targeted the H11 His524 side chain, while the p-carbonyl group made H-bonds with H3 Thr347 s side-chain hydroxyl (see Supplementary Materials Table S18 for details); the incorporated o-Ac-Ph moiety formed eclipsed (i.e., edge to edge) van der Walls interactions with the H3 Thr347 s side chain methyl group using its own methyl group, as well as the additional HBs with H3 Thr347 s side chain hydroxyl group (see Supplementary Materials Table S18 for details) by the acetyl group carbonyl portion. The unsubstituted 3-acetyl-4-hydroxybenzoic ac carbons faced the H12 Leu536 in a Tshaped fashion. Furthermore, the p-O-CH 2 -CH 2 -bridge bore the 3DPQ-1 s to 3DPQ-12 s functionalities that forced the H12 drifting, at the same time establishing the electrostatic attraction with H3 Thr347 s hydroxyl group via the oxygen atom and the steric interactions between the methylene carbons and the Leu536 isobutyl group.
The activity and SERM pharmacology [13] of 3DPQ-12 (Table 7, Supplementary Materials Figure S191A, Figure 7A, potency 1.85-fold higher than Ral) could be also ascribed to the 3-hydroxypiperidin-2-one portion: positioned beneath the Asp351-Leu536 plane, its hydroxyl group established an HB with Asp351 (the d HB = 3.112 Å), stabilizing ERα with H12 in the open conformation; the carbonyl group electrostatically interfered with the Thr347 s side chain hydroxyl group, whereas the carbon skeleton was in the proximity of Leu536 isobutyl group. A slightly less potent SERM, for just 0.04 nM, was the 3DPQ-3 (Table 7, Supplementary Materials Figure S191B, Figure 7B, potency 1.68-fold higher than Ral), whose 1,2,5,6-tetrahydropyridine-3-carboxylic acid scaffold formed an HB with Asp351 (the d HB = 3.222 Å) via the carboxyl group, whereas the carbon skeleton behaved similarly as in 3DPQ-12. Furthermore, the potency of 3DPQ-9 (Table 7, Supplementary Materials Figure S191C, Figure 7C, 1.64-fold stronger binder than Ral), decreased by 0.01 nM related to 3DPQ-3 with the introduction of the carbonyl portion at position C6 of 1,2,5,6-tetrahydropyridine-3-carboxylic acid, which electrostatically attracted the Trp383 s indole ring nitrogen, having a consequence in C3-COOH group dispositioning and a weaker HB with H3 Asp351 (the d HB = 3.314 Å). Table 7. Antagonistic potencies (IC 50 s) and the logarithm of the relative binding affinities (RBA) against ERα and ERβ of the newly synthesized compounds. Isoform affinity preferences and respective antagonist constants are also reported.

Comp.
ERα a Concentration that antagonizes the 50% of ERα signaling activity; b Concentration that antagonizes (inhibits) the 50% of ERβ signaling activity; c Logarithmic value of the percentage of relative binding affinity toward the ERα; d Logarithmic value of the percentage of relative binding affinity toward the ERβ (for both c values and d values relative binding affinity (RBA) values where calculated related to estradiol with an affinity of 100%, logRBA values higher than 0 refer to strong binders, logRBA values between −2 and 0 refer to moderate binders, logRBA values below −2 refer to weak binders); e Calculated antagonistic (i.e., inhibitory) constants against ERα; f Calculated antagonistic (i.e., inhibitory) constants against ERβ; g Results are presented as mean value ± standard deviation; h 17β-estradiol; i 4-hydroxytamoxifen; j Raloxifene; k No ligand (0.9% NaCl). l Not available. * p < 0.05 when compared with control group; † p < 0.05 when compared with E 2 ; ‡ p < 0.05 when compared with 4-OTH; § p < 0.05 when compared with Ral.

Synthesized Compounds Antiproliferative Activity against ERα(+)-and ERα(-)-Dependent Breast Cancer Cell Lines as Well as against ERα(+)-Dependent Endometrial Cancer Cell Lines
Synthesized compounds were evaluated as antiproliferative agents against MCF-7 ( Table 8, Supplementary Materials Figures S196 and S197), and MDA-MB-231 (Table 8, Supplementary Materials Figures S198 and S199) cells lines [130], respectively, as well as for the ability to induce ERα downregulation in MCF-7 cells (Table 8) [15,21,131,132] and to antagonize the progesterone receptor (PR) ( Table 8) [126]. Table 8. Synthesized compound antiproliferative activity and selectivity index against hormonedependent MCF-7, hormone-independent MDA-MB-231 breast cancer cell lines, normal MRC-5 human lung tissue fibroblasts cell lines, and Ishikawa endometrial adenocarcinoma cell lines, as well as the downregulation of ERα in MCF-7 and PR antagonism in MCF-7 cell lines. Compounds-proposed bioactive conformations anticipated a SERM-like profile, which was experimentally confirmed as they induced no ERα degradation, at the same time exerting no antagonism against PR (Table 8) [125]. Therefore, the further focus was on the antiproliferative activity, where even eight derivatives showed antiproliferation against MCF-7 better or comparable to Ral (Table 8). 3DPQ-12 (Table 8, Supplementary Materials Figure S196A) was the most potent MCF-7 cell growth inhibitor with an IC 50 value equal to 560 pM and a selectivity index (SI) relative to MDA-MB-231 cell lines of 147.93. Similar antiproliferation profiles were also exerted by 3DPQ-3 (Table 8, Supplementary Materials Figure S196B, potency 1.11-fold lower than 3DPQ-12 but 1.43-fold higher than Ral, SI equal to 131.66) and 3DPQ-9 (Table 8, Supplementary Materials Figure S196C, potency 1.09-fold lower than 3DPQ-12 but 1.46-fold more potent than Ral, SI equal to 142.02).
As SERMs profile is often associated with the stimulation of endometrial cell proliferation and an increase in the incidence of endometrial cancer (EC) [130], the herein compounds were therefore evaluated against Ishikawa endometrial adenocarcinoma cells ( Table 8, Supplementary Material Figures S200 and S201). At this stage of evaluation, the herein SERMs significantly inhibited Ishikawa cell lines growth. However, future experimental elaboration, currently beyond the authors' experimental facilities, is required to confirm compounds' promising profiles in terms of no EC induction [130].

The Impact of Targeted ERα Antagonists on the MCF-7 Cells Signaling
The exerted antiproliferation against MCF-7 cell lines was further inspected for the inner mechanisms of action. BFA is known for inducing the endoplasmic reticulum stress within the MCF-7 cell lines, as well as for increasing the expression of p53, a major BC suppressor [132]. Nonetheless, ERα binds to p53, resulting in the inhibition of transcriptional regulation by p53, p53-mediated cell cycle arrest, and apoptosis [133], raising the question of whether the ERα antagonists herein described could have also inhibited MCF-7 cells' growth by decreasing the ERα recruitment and by stimulating the p53 s transactivation function. To investigate this hypothesis, the conventional and sequential site-specific ChIP assays were employed to reveal the mechanisms by which the 3DPQ-1 to 3DPQ-12antagonized ERα influenced the p53-mediated transcriptional activation of the p21 gene (a prototypic p53-target gene) [133]. Experimentally, all the compounds except 3DPQ-5, 3DPQ-6, and 3DPQ-8 have been re-administered in 0.1 and 1 nM to MCF-7 cells (i.e., two concentrations encircling the IC 50 values against MCF-7 cells, Table 8); for the marked compounds, the concentrations were 1 and 10 nM.
Upon the addition of primers specific to the p53-binding site of the p21 promoter, the chromatin was immunoprecipitated with the anti-p53 antibody and re-immunoprecipitated with the anti-ERα antibody, enabling the conclusion that the p53 expression occurred after the ERα has been antagonized by compounds ( Figure 8A). The final round of reimmunoprecipitation was performed with NCoR and SMRT corepressors, guided by the premise that 3DPQ-1 to 3DPQ-12 as antiestrogens could promote their binding to ERα, followed by the recruitment of HDACs and leading to transcriptional repression [134,135]. Nonetheless, as NCoR, SMRT, and HDAC1 had been not recruited to the p21 promoter when ERα was knocked down ( Figure 8B), ERα-3DPQ-1 to ERα-3DPQ-12 complexes, conversely to ERα, stimulated the p53-mediated transcriptional activation without recruiting the distinct corepressors.
Furthermore, the quantitative ChIP (qChIP) analysis measured the strength of 3DPQ-1 to 3DPQ-12 to affect the ERα's ability to bind to p53. Contrary to E 2 , 3DPQ-1 to 3DPQ-12 disrupted the receptor's interaction with the p21 promoter ( Figure 8A) and stimulated the p53 transcriptional activity. The highest rate of p53 promoter activity was induced upon the 3DPQ-12, 3DPQ-3, and 3DPQ-9 administration, 0.65-fold and 0.55-fold, 0.68-fold and 0.61-fold, as well as 0.68-fold and 0.66-fold higher than the one provoked by Ral in lower and higher concentrations, respectively ( Figure 8B). The 3DPQ-4 was similarly potent to 3DPQ-9, exerting 0.70-fold and 0.68-fold higher potency than Ral, respectively, whereas 3DPQ-2 and 3DPQ-1 exerted the matching potency, 0.733-fold and 0.66-fold higher than Ral ( Figure 8A). Conclusively, as ERα and SERMs, 3DPQ-1 to 3DPQ-12 have indeed decreased ERα recruitment and stimulated the p53 (p21) pathway, as another way of preventing the growth of MCF-7 cells. 0.68-fold and 0.61-fold, as well as 0.68-fold and 0.66-fold higher than the one provoked by Ral in lower and higher concentrations, respectively ( Figure 8B). The 3DPQ-4 was simi larly potent to 3DPQ-9, exerting 0.70-fold and 0.68-fold higher potency than Ral, respectively, whereas 3DPQ-2 and 3DPQ-1 exerted the matching potency, 0.733-fold and 0.66fold higher than Ral ( Figure 8A). Conclusively, as ERα and SERMs, 3DPQ-1 to 3DPQ-12 have indeed decreased ERα recruitment and stimulated the p53 (p21) pathway, as another way of preventing the growth of MCF-7 cells.  1 nM (for 3DPQ-5, 3DPQ-6, and 3DPQ-8 the concentrations were 1 and 10 nM) with primers specific to the p53-binding site of the p21 promoter. The primary ChIP was performed with anti-p53 antibody, and the immunoprecipitate was subjected to a second ChIP with anti-ERα antibody; (B) The immunoprecipitate from the ERα ChIP was then subjected to the third ChIP with antibodies against NCoR, SMRT, and HDAC1 antibodies; (C) qChIP was per formed to analyze the ERα-p53 interaction on the p21 promoter in MCF-7 cells saturated with 3DPQ-1 to 3DPQ-12. Cells were grown in media with dextran-coated charcoal-treated FBS for 4 d and treated with E2 (1 and 10 nM) with or without 3DPQ-1 to 3DPQ-12 for 3 h. * p < 0.05 when compared with control group; † p < 0.05 when compared with E2; ‡ p < 0.05 when compared with 4 OTH; § p < 0.05 when compared with Ral.  1 nM (for 3DPQ-5, 3DPQ-6, and 3DPQ-8 the concentrations were 1 and 10 nM) with primers specific to the p53-binding site of the p21 promoter. The primary ChIP was performed with anti-p53 antibody, and the immunoprecipitate was subjected to a second ChIP with anti-ERα antibody; (B) The immunoprecipitate from the ERα ChIP was then subjected to the third ChIP with antibodies against NCoR, SMRT, and HDAC1 antibodies; (C) qChIP was performed to analyze the ERα-p53 interaction on the p21 promoter in MCF-7 cells saturated with 3DPQ-1 to 3DPQ-12. Cells were grown in media with dextran-coated charcoal-treated FBS for 4 d and treated with E 2 (1 and 10 nM) with or without 3DPQ-1 to 3DPQ-12 for 3 h. * p < 0.05 when compared with control group; † p < 0.05 when compared with E 2 ; ‡ p < 0.05 when compared with 4-OTH; § p < 0.05 when compared with Ral.

Effects of Synthesized Compounds on Cytotoxicity and Cell Cycle Distribution of MCF-7 Cell Lines
The above data encouraged further analysis of the cell cycle of MCF-7 cells treated by 3DPQ-1 to 3DPQ-12 (Table 9, Supplementary Material Figures S202-S213) [130], administered at the same concentrations used for the cell signaling assay. Thus, compounds induced the MCF-7 cells' arrest in the G 0 /G 1 phase, i.e., the phase in between the nondivision, post mitosis (viz., G 0 ), and DNA replication (viz., G 1 ). The G 0 /G 1 phase arrest was accompanied by a decrease in the S phase, suggesting that compounds stopped the MCF-7 proliferation before the DNA replication induced by the transcriptional machinery. The results agreed with previous findings that SERMs block MCF-7 cell cycle progression in G 0 /G 1 [136]. It is worth emphasizing that for all the compounds, applied in both concentrations, the contribution of the G 0 /G 1 phase to the MCF-7 cells' arrest was higher than 70%. The distribution of 3DPQ-12 (Table 9, Supplementary Material Figures S202A,E), and 3DPQ-4 (Table 9, Supplementary Material Figures S205A,E) within the cell cycle mostly affected the cells' proliferation, reaching 77 to 80% of the contribution of the G 0 /G 1 phase upon administering either 0.1 or 1 nM of the compound, respectively. On the other hand, 3DPQ-3 (Table 9, Supplementary Material Figures S203A,E), 3DPQ-9 (Table 9, Supplementary Material Figures S204A,E), 3DPQ-2 (Table 9, Supplementary Material Figures S206A,E), and 3DPQ-1 (Table 9, Supplementary Material Figures S207A,E) blocked the MCF-7 cycle in the initial phase between 71 and 76%. The cell cycle arrest in the G 0 /G 1 phase may be a key mechanism by which targeted antiproliferative agents inhibit MCF-7 cell proliferation.
Experimentally, the adult female Wistar rats were pretreated intraperitoneally (i.p.) with methyl nitrosourea (MNU) with a dose of 50 mg/kg of each rat's body weight (bwt) to induce the BC, after which the compounds herein described were administered per os in two doses, 5 and 50 mg/kg of bwt [81]. The compounds were evaluated employing latency period (i.e., the time passed between the rats being exposed to MNU and the BC detection), tumor burden (i.e., the number of cancer cells), and tumor volume.
Hence, 3DPQ-12, 3DPQ-3, and 3DPQ-9 induced the longest latency period, 12 to 15 weeks depending on the concentration applied, followed by its low burden and volume, overpowering the efficiency of Ral (Table 11). The 3DPQ-4 induced a latency period between 9 and 12 weeks. The remaining leads, 3DPQ-2 and 3DPQ-1, were slightly less efficient tumor suppressants, with tumor latency between 7 to 12 weeks and more emphasized tumor burdens and volumes, but were still more potent than Ral. Of course, the safety of the compound during administration was confirmed with liver enzyme catalytic activities and redox status [147][148][149][150][151][152][153][154][155] (Supplementary Materials Tables S19 and S20), where no significant harm was detected.
The impact of selected leads on BC tissue was registered after their administration to experimental animals with MNU-induced BC (Figures 9 and S211-S218) [159]. Thus, compared to the normal pathological finding of animals treated with saline, reflected in photomicrographs revealing lobuloalveolar unit (LaU) and cuboidal epithelial cells (CE) ( Figure 9A), MNU provoked ductal mammary gland carcinoma and massive proliferation of neoplastic epithelial cells (EC) ( Figure 9B), changes found within the terminal ductal-lobular unit, that formed discrete clusters with duct-like morphology. In contrast to this, the administered leads were harmless in both concentrations, neutralizing the MNU-induced changes, judging by the lobuloalveolar units and cuboidal epithelial cells found ( Figures 9C,D and S214-S218). These compounds were safer than 4-OHT, which caused severe necrosis (NEC) ( Figure 9E,F), and Ral, which caused extralobular ducts (ED) ( Figure 9G,H).  Figure 9.  Finally, the compounds were assayed for the maximum tolerated dose (MTD) or maximum feasible dose (MFD, in the absence of MTD) and weight loss (WL) studies (Table 11). Compounds and controls were daily re-administered per os in five doses, 5, 50, 100, 500, and 1000 mg/kg bwt [160] for 5 days. On the 5th day, the body weights were measured, and the postmortem evaluations were performed by means of a gross examination of all the animals at the terminal necropsy, as well as the histopathological examination of lungs, spleen, liver, kidneys, heart, and colon (Supplementary Materials Figures S219-S224, respectively). Hence, except for MNU, with an MTT of 100 mg/kg bwt, no mortality was observed in the treatment groups for 5 days even at the highest dose (Table 11). The orally administered compounds 3DPQ-12, 3DPQ-3, 3DPQ-9, 3DPQ-4, 3DPQ-2, and 3DPQ-1 did not produce significant changes in body weight. Moreover, no obvious pathologic changes were observed based on histology or necropsy compared to placebo-treated controls. Therefore, given that the Food and Drug Administration (FDA) recommends 1000 mg/kg bwt as the high limit dose for acute, subchronic, and chronic toxicity studies in rodents and non-rodents [160], MTDs were not explicitly determined, and the 1000 mg/kg bwt could be considered as MFD (https://www.fda.gov/drugs/guidance-compliance-regulatoryinformation/guidances-drugs, accessed on 1 March 2022) for 3DPQ-12, 3DPQ-3, 3DPQ-9, 3DPQ-4, 3DPQ-2, and 3DPQ-1 [160]. All the compounds were proven safe for further pre-clinical and clinical trials at a concentration of 50 mg/kg bwt.

ERα LBD-Partial Agonists/Antagonists Complexes Structures Preparation
The 39 complexes of ERα partial agonists and antagonists, co-crystallized with either wild-type (WT) or mutated (MUT) receptors, retrieved from PDB (TR,  Table 3: 13 WT and MUT ERα binders with the activities reported as pK i s) were prepared [93,161] using the validated procedures described elsewhere [80,92] (see the Supplementary Materials: Crystal structures compilation and preparation and Supporting Information Table S1 for detailed information).

3-D Pharmacophore Hypotheses and 3-D QSAR Models Generation
A set of 3-D pharmacophore hypotheses and atom-based 3-D QSAR models were generated using the PHASE software [88] as implemented in Schrödinger's suite [89], using the default setup (see the Supplementary Materials: Pharmacophore modeling and 3-D QSAR modeling for detailed information). For the statistically best hypotheses/models (endowed with the highest q 2 values), robustness was confirmed by means of leave-oneout (LOO) and leave-some-out (LSO) cross-validations (CV) [80,92] while lack of chance correlation was checked by a Y-scrambling procedure [80,92]. Models were graphically interpreted by means of UCSF Chimera [93].

SB Alignment Assessment
All the scoring functions of the Glide software [104][105][106], as implemented in Schrödinger's Suite [89], were evaluated to select the best one to perform an SB alignment assessment on TR compounds. The SB procedure was assessed through four methods, similar to those previously described in [80,92]: experimental conformation re-docking (ECRD), randomized conformation re-docking (RCRD), experimental conformation cross-docking (ECCD), and randomized conformation cross-docking (RCCD). The experimental protocols and Glide's settings [105,106] are reported in the Supplementary Materials: Alignment assessment rules, Ligand's experimental conformations randomizations, and Glide settings.

LB Alignment Assessment
To rule out the LB molecular alignment of TR compounds, all the available scoring functions of the flexible ligand alignment tool (FLA) [89], as implemented in Schrödinger's Suite [89], were evaluated. The LB alignment procedure assessment was conducted at different levels of difficulty, similar to those previously described in [80,92]: experimental conformation re-alignment (ECRA), randomized conformation re-alignment (RCRA), experimental conformation cross-alignment (ECCA), and randomized conformation crossalignment (RCCA). The experimental protocols and FLA setup [89] are reported in the sections Supplementary Materials: Alignment assessment rules and Flexible Ligand Alignment tool settings.

The SB/LB Alignment Accuracy
The alignment fitness was then quantified by evaluating both the RMSD and the subsequent docking accuracy (DA) and alignment accuracy (AA), as previously reported [80,92]. Both DA and AA were used to evaluate how the algorithms used could predict the ligand poses as closely as possible to the experimentally observed ones, by separating the correctly (RMSD ≤ 2 Å) and partially (2 Å ≤ RMSD ≤ 3 Å) docked/aligned poses for those mis-docked/mis-aligned (RMSD ≥ 3 Å). The rules for DA and AA calculation are reported in Supplementary Materials Alignment assessment rules section.

Generation of Modeled and Designed Compounds
Either TS MOD1 s, TS MOD2 s, and TS MOD3 s (Supplementary Materials Tables S10-S15) or the designed compounds (Table 8) were drawn through the Chemaxon's msketch module [103] by means of the optimization of the molecular mechanics using the MMFF94 force field and the default settings, upon which the hydrogen atoms were assigned at pH 7.4. Upon structures' generation, compounds were uploaded into previously described best-performing SB and LB protocols to obtain the bioactive conformations (see Supplementary Materials: Alignment assessment rules, Structure-based alignment assessments, and Ligand-based alignment assessments).

Test Sets and Designed Compounds Alignment
The TS MOD1 , TS MOD2 , and TS MOD3 (Supplementary Materials Tables S10-S15), as well as all the designed compounds (Table 6), were aligned applying either the best performing SB or LB protocols (see Supplementary Materials: Test sets alignment, Alignment assessment rules, Structure-based alignment assessments, and Ligand-based alignment assessments).

Virtual Screening
The virtual screening of NCI compound libraries (486 compounds from Natural Products Set 3 and 1574 and 2351 compounds from the Diversity Sets 2 and 3), taken from the NCI (NCI, https://www.cancer.gov/, accessed on 1 October 2015) was conducted following the guidelines as described elsewhere [90,91]. The compounds were retrieved in structure data file (sdf) format, split into individual files, imported in Chemaxon's msketch module [103], and energy minimized by means of molecular mechanics' optimization using the MMFF94 force field and the default settings, upon which the hydrogen atoms were assigned at pH 7.4. Upon the generation of the structures, compounds were uploaded into previously determined best-performing SB and LB protocols to perform cross-docking and cross-alignment and obtain the bioactive conformations against ERα (see Supplementary Materials: Virtual screening, Alignment assessment rules, Structure-based alignment assessments, and Ligand-based alignment assessments).

Synthesis of Compounds 3DPQ-1 to 3DPQ-12
All the experimental work regarding the conventional synthesis of designed compounds 3DPQ-1 to 3DPQ-12, as well as regarding spectral data interpretation and purity, is described in detail as Supplementary Materials under the Experimental and Results and discussion sections, respectively.

Conclusions
The reported investigation summarizes the usage of rational drug design protocol by means of the SB and LB techniques to disclose new potent and selective antagonists against ERα as in vitro and in vivo anticancer agents, which emerged upon the lead optimization of the virtually screened compound Brefeldin A. The SB 3-D pharmacophore/QSAR models, coupled with molecular docking and ligand-based alignment, were revealed to be effective tools in the design of new Brefeldin A derivatives and were used for the very first time to describe their potency against ERα in physiological conditions, using the ERα antagonists and partial agonists co-crystallized within both wild-type or mutated receptors. Notably, the models emerged from a wide-ranging molecular diversity within the training set, consisting of a variety of antagonists and partial agonists associated with SERDs, SERMs, and naturally occurring sub-groups of compounds. The best ADDHHHP.13 hypothesis (3-DPhypI), alongside the derived 3-D QSAR model, differentiated full antagonists from partial agonists and provided some guidelines for the selectivity toward ERα, describing all the important 3-D pharmacophoric properties desired for a powerful SERM to occupy the natural hormonal environment and to invoke in perspective the complete shut-down of estrogen-initiated basal transcriptional machinery. Moreover, the ADDHHHP.13 hypothesis was used to virtually screen NCI datasets disclosing BFA as an interesting hit, which was structurally optimized by engineering twelve innovative SERMs, 3DPQ-1 to 3DPQ-12, that were synthesized, and broadly biochemically evaluated as ERα antagonists, as prospective BC suppressants. From determining the antagonistic potential against ERα, to elaborating the antiproliferative activity in ERα(+) BC cell lines, including the impact on the inner mechanisms of cancer development and toxicity predicted in silico, all of the designed and synthesized hits exerted notable potency, where slight differences in the activity can be understood from the structure-based point of view. The in vivo administration to adult Wistar rats discriminated the lead compounds by means of their impact on mammary tumorigenesis. Hence, 3DPQ-12, 3DPQ-3, 3DPQ-9, 3DPQ-4, 3DPQ-2, and 3DPQ-1 were indeed found to be as potent as Ral, the most potent compound listed in the TR, at any stage of evaluation. By exerting more-than-promising anticancer activity, a favorable preclinical profile, and notable safety, 3DPQ-12, 3DPQ-3, 3DPQ-9, 3DPQ-4, 3DPQ-2, and 3DPQ-1 can be considered candidates for pre-clinical and clinical trials as the future of SERM-related BC clinical therapy. In a future study, a model for the ERβ antagonists will be also developed to design selective antagonists.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/ 10.3390/molecules27092823/s1. This material contains the Introduction (i.e., The Genomic classical pathway, Genomic indirect pathway, Tethered pathway alternative routes, Non-genomic pathways, Abbreviations, ERα 3-D pharmacophore models generation overview), Results and discussion (i.e., Tables and Figures describing data sets compilation, 3-D pharmacophore models, 3-D QSAR models, SB and LB alignment assessments, activity prediction of test sets, virtual screening, designed compounds SB and LB alignments, synthesized compounds spectral data interpretation, Figures of 1H NMR, 13 C NMR, 15 N NMR, 17 O NMR, and HPLC spectral data of synthesized precursors and bioactive compounds, related tables with biochemical data), and Experimental section (i.e., the training set selection, preparation of antagonists-ERα complexes, interpretation of 3-D QSARs, SB and LB alignment assessment rules definition, virtual screening, equipment, commercial compounds supply, synthetic protocols, the in vitro and in vivo experimental protocols). Figure S1-S9: Data associated with the 3-D pharmacophore and 3-D QSAR model interpretation, Figures S10-S19: Data associated with the structure-based and ligand-based alignment assessments, Figures S20-S22: Data associated with the virtual screening, Figures S23-S25: Data associated with designed compounds binding conformations, Figures S26-S177: Data associated with synthesized compound 1 H NMR, 13 C NMR, 15 N NMR, 17 O NMR spectra, Figures S178-S190: Data associated with synthesized compound HPLC spectra, Figures S191-S224: Data associated with synthesized compound biological activity in vitro and in vivo, Tables S1-S6: Data associated with the 3-D pharmacophore and 3-D QSAR models development, Tables S7-S9: Data associated with the structure-based and ligand-based alignment assessments, Tables S10-S15: Data associated with the external validation of 3-D pharmacophore and 3-D QSAR models predictive abilities, Tables S16-S17: Data associated with the virtual screening, Table S18: Data associated with designed compounds binding conformations, Tables S19-S20: Data associated with the synthesized compounds' toxicity.

Informed Consent Statement: Not applicable.
Data Availability Statement: All the experimental complexes used to build the 3-D pharmacophore and 3-D QSAR models, as well as the structure-based and ligand-based alignment assessments, can be retrieved free of charge from Protein Data Bank (https://www.rcsb.org/, accessed on 1 October 2015). All the compound structures used as test sets can be found in the Protein Data Bank or retrieved from the cited literature (see Supplementary Materials for specifics). All the computational results from 3-D pharmacophore and 3-D QSAR models studies and structure-based/ligandbased alignment assessments, as well as the UCSF Chimera sessions, are available from Milan Miladenović (files in machine-readable formats, e-mail: milan.mladenovic@pmf.kg.ac.rs). All the computational results regarding the design of new compounds can be obtained from Rino Ragno