The topomer CoMFA method provides a means for an alignment-independent 3D-QSAR approach, which is advantageous because it does not require alignment and it provides a means for automated activity searches in fragment libraries. BA derivatives in the literature have a common pentacyclic triterpenoid core, and most studies have focused on modifications at the C-3 and C-28 sites. In this topomer CoMFA study, we designated the groups at the two sites as R-groups. We split the molecules into three pieces by cutting at the C-3 and C-28 sites with a consideration toward structural diversity (Figure 2). In this study, Gasteiger-Hückel charges were used for charge calculations.

A database alignment was used to establish topomer CoMFA 3D-QSAR models using the methods described in the Experimental Topomer CoMFA setup section. The predicted values of compounds in the training and test sets using topomer CoMFA are listed in Table 1. The experimental IC₅₀ value of compound 10 is >100 μM, and compound 10 was selected as a test molecule with the pIC₅₀ value set at 4.0 to expand the diversity of structures and the range of pIC₅₀. Therefore a bigger residual was found between the predicted and experimental pIC₅₀s for compound 10.

The topomer CoMFA model was optimized. Cross-validation q² value of 0.722 and a non-cross-validation r² value of 0.974 with an optimized component of 5 were obtained, which suggested that the model has good predictive ability (q² > 0.5) (F values = 151.669). The predictive correlation coefficient (r²_pred) of 0.675, the R²_m value of 0.726, and the root mean square error of prediction (RMSEP) value of 0.201 were determined to be externally validated. The pIC₅₀ values of the test set were predicted, and the test set r²_pred value of 0.675 was obtained in the evaluation of the predictive ability of the model. The correlation plot of the experimental versus predicted pIC₅₀ values is shown in Figure 3, and a good correlation between the predicted and the experimental values of the compounds was observed. The predictive power of the model is high but only within the addressed range, i.e., truly high-affinity compounds (pIC₅₀ > 8) cannot be explained by the model.

The topomer CoMFA 3D contour maps around R1 and R2 (shown in Figure 4 and Figure 5, respectively) were generated by plotting the coefficients from the model. It is better to choose the molecule with highest activity as the reference molecule which makes it is easier to explain the contour map. Compound 13 showed the best biological activity against HT 29 cells in all 37 compounds. Hence these maps are shown using compound 13 as a reference structure. In the steric field, the green contours located at the C-28 site indicate that a bulky substituent would be favorable, and the yellow contours denote where bulky substituents would not be tolerated (Figure 4a). In the electrostatic field, the blue contours located at the C-28 site indicate that electropositive groups would be favorable, and the red contours indicate that electronegative groups would be favorable (Figure 4b). Green contours dominated the R1 group in the steric field, blue contours occupied the middle of the electrostatic field, whereas red contours were located at the end of the substituent; this suggested that bulky groups with electronegative potential at the end of the side chain at the C-28 site would be favorable for activity. Regarding the contours of the R2 group, green contours (Figure 5a) were located near the C-3 site, and yellow contours were located in farther away; the red contours were located in the middle of the electrostatic field, and blue contours were located far away from the C-3 site (Figure 5b); this demonstrated that a moderately bulky group with electronegative potential at the R2 site would improve the anti-tumor activity.

To verify the topomer CoMFA model, molecules were designed. Since BE is the 28-hydroxyl analog of BA and shares the same pentacyclic triterpenoid core with BA, both BA and BE are good starting materials for further derivatization. BE shows better availability and much cheaper price because it is present in large amounts (up to 30% dry weight) in the outer bark of birch trees, so BE was selected as the starting material for the experimental verification. Because a moderately bulky group with electronegative potential was required at the C-3 site, we left this position unchanged, as the hydroxyl group at the C-3 site in BE perfectly matches the requirements. To introduce electronegative potential at the end of the side chain at the C-28 site, we employed dicarboxylic anhydride to form a carboxylic group at the end of side chain via esterification of the C-28 hydroxyl group of BE. To meet the needs of a sterically bulky group, the cyclic dicarboxylic anhydrides phthalic anhydride and hexahydrophthalic anhydride were selected, and maleic anhydride was chosen as a representative fatty acid. The chemical reaction and the substituent groups are illustrated in Scheme 1. The bulky electronegative groups (maleyl, phthalyl, and hexahydrophthalyl groups) were directly introduced at the C-28 position of BE. The anti-cancer activity of these three compounds against HT29 cells were predicted and tested (Table 2). The experimental pIC₅₀ were compound 1 < compound 2 < compound 3, which were consistent with the tendency of predicted values. The residues between the experimental and predicted values were much bigger in Table 2 than that in Table 1, because the experimental IC50 results from our lab may be significantly different from Kommera’s lab, which is also the reason we should select the data from the same research group to establish a reliable model. The consistent tendency further confirmed the established topomer CoMFA method as a predictive model for the structure-activity relationships of BA derivatives. The hexahydrophthalyl and phthalyl derivatives displayed better or equal bioactivity to BA, which is much more active than the parent molecule BE.

To better understand the anti-tumor activities of these three derivatives, their IC₅₀ values were also assayed in five other tumor cell lines, including pancreatic cancer (MPC2), prostate cancer (DU145), lung carcinoma (NCI-H520), cervical cancer (HeLa), and ovarian cancer (2774) cells. The results are presented in Table 3. Among these three structures, compound 3 (containing the hexahydrophthalyl group) exhibited the greatest anti-cancer activities; its IC₅₀ values were lower than those of BA in all cell lines excluding DU145 cells. Following the instructions of the topomer CoMFA model, additional molecular design can be conducted, and more promising and potent agents can be synthesized.

The in vitro anti-cancer activities of 37 BA and BE derivatives (reported as IC₅₀s) against HT29 human colon cancer cells were used for the present study. All compounds and associated inhibitory activity data were obtained from the literature [10,11,12,13] reported by the same research group within 2 years. These 37 molecules were divided into a training set of 26 compounds and a test set of 11 compounds. IC₅₀ values were converted to pIC₅₀ values by taking log (1/IC₅₀). The compounds in the test set were randomly selected with the goal of choosing a range of both biological activity data and structural diversity. The structures of the compounds in the training and test sets are listed in Table 1.

The Topomer CoMFA program in sybyl-x1.1 software package was used to perform all calculations. Topomer CoMFA is an alignment-independent 3D-QSAR method that combines the topomer search method [23] (a fragment alignment approach) with the conventional CoMFA method. A 3D-QSAR model was generated by splitting the molecules into fragments, topomerically aligning each fragment, and calculating steric and electrostatic field descriptor values for the topomerically aligned fragments to create a CoMFA table with the field descriptor values. Besides the core of the molecule, we split side functional group into two R-groups that refer to the R1 and R2 groups of BA (Figure 2).

To evaluate the predictive ability of the model, the pIC₅₀ values of the test set were predicted using the model developed using the training set. The optimization and splitting methods of the test set were the same as those described for the training set.

To evaluate the predictive ability of the model, structure optimization of the test set was performed as previously described for the training set. The pIC₅₀ values of the test set were predicted on the basis of the constructed model. r²_pred, which measures the predictive performance of a PLS model, is defined by equation 1 as follows: r²_pred = (SD − PRESS)/SD (1) where SD is the sum of the squared deviations between the predicted biological activities of the test set and the mean activities of the training molecules and PRESS is the sum of the squared deviations between the predicted and experimental activity values for every molecule in the test set.

Furthermore, the r² value between the observed and predicted activity values of the compounds in the test set with (r²) or without an intercept (r²₀) and the RMSEP [defined by Equation (2)] [24] for the test set compounds were also calculated. The models were validated using an additional parameter, R²_m [defined by Equation (3)], which is a good indicator of the external predictability of QSAR models [25]: (2) (3)

BE was extracted from the bark of Betula platyphylla Suk., and its purity exceeded 95% as determined by HPLC. All other chemicals (reagent grade) were commercially available. Separation of compounds 1–3 was performed by column chromatography using silica gel 60 (70–230 mesh). Thin-layer chromatography was performed using silica gel-coated aluminum sheets (silica gel 60 GF254, E. Merck, Darmstadt, Germany). The melting point (m.p.) of each compound measured on an X-6 micro melting-point apparatus (Beijing Tech instrument Co., Beijing, China). The FTIR spectra were recorded using a Nicolet Mattson Infinity Gold FTIR spectrometer using KBr pellets. ¹H-NMR spectra (300 MHz) and ¹³C-NMR spectra (75 MHz) were recorded in CDCl₃ at 25 °C with TMS and solvent signals used as internal standards using a Bruker AVANCE-300 MHz. Chemical shifts are reported in ppm (δ). Mass spectra were obtained using a Finnigan Mat 95 apparatus. Elemental analyses were performed on a Thermo Finnigan EA 1112 Series Flash Elemental Analyzer, and the obtained values were within 0.4% of the theoretical values.

BE (1 mmol) and acid anhydride (1 mmol) were refluxed in methylene dichloride (50 mL) under stirring for 30 hours. After completion of the reaction, the reactant was washed with saturated sodium bicarbonate and brine solutions, dried over anhydrous magnesium sulfate, filtered, and evaporated in vacuo. The product was purified by column chromatography with a mobile phase of ethyl acetate/petroleum ether (1/3).

HT29, MPC2, DU145, NCI-H520, HeLa, and 2774 cells were cultured in RPMI-1640 medium (GIBCO, CA, USA) supplemented with 10% fetal bovine serum (FBS; PAA, Yeovil, UK), l-glutamine (0.29 mg/mL), and 1% penicillin-streptomycin in a humidified atmosphere of 5% CO₂ at 37 °C. Cells were sub-cultured every 4–5 days by total replacement using 0.25% (w/v) trypsin. In vitro cytotoxicity assays were performed according to a previously reported method [26,27]. Briefly, cells were seeded in 96-well culture plates in 0.2 mL of growth medium per well and allowed to attach for 24 h. The culture medium was replaced with either BA or the derivatives at different concentrations in 4 replicates followed by 72 h of incubation. After incubation, 20 μL of 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide (MTT; Amresco, Solon, OH, USA) solution at a concentration of 5 mg/mL was added to each well followed by incubation for 4 h. The MTT solution was then aspirated, and 150 μL of DMSO was added to each well to dissolve the dark blue crystals thoroughly. The absorbance was measured at 490 nm using a microplate reader (Infinite 200 NanoQuant, Tecan, Grödig, Austria). The relative growth rate (%) was calculated as (mean absorbance of the sample/mean absorbance of the control) × 100%, considering the optical density of the control as 100%.