taxane analogues against lung cancer: a quantitative structure–activity relationship study

Taxane Analogues against Lung Cancer: AQuantitative Structure–Activity RelationshipStudy

Rajeshwar P. Verma* and Corwin Hansch

Department of Chemistry, Pomona College, 645 North CollegeAvenue, Claremont, CA 91711, USA*Corresponding author: Rajeshwar P. Verma, [email protected]

Lung cancer is the second most common cancer inboth men (after prostate cancer) and women (afterbreast cancer). The microtubule-stabilizing taxanesuch as docetaxel is the only agent currentlyapproved for both first- and second-line treatmentof advanced non-small cell lung cancer. Althoughdocetaxel has made significant progress in thetreatment of lung cancers either using alone or incombination with various novel targeted agents,its use often results in various undesired side-effects. These limitations have led to the searchfor new taxane derivatives with fewer side-effects, superior pharmacological properties, andimproved anticancer activity to maximize theinduced benefits for lung cancer patients. Herein,four series of taxane derivatives were used to cor-relate their inhibitory activities against lung can-cer cells with hydrophobic and steric descriptorsto gain a better understanding of their chemical–biological interactions. A parabolic correlationwith MRY is the most encouraging example, inwhich the optimum value of this parameter is welldefined. On the basis of this quantitative structure–activity relationship model, six compounds (3-23 to3-28) are suggested as potential synthetic targets.Internal (cross-validation (q 2), quality factor (Q),Fischer statistics (F ) and Y-randomization) andexternal validation tests have validated all thequantitative structure–activity relationship models.

Key words: hydrophobicity, lung cancer, molar refractivity, quantita-tive structure–activity relationship, taxanes

Received 23 September 2008, revised 5 February 2009 and accepted forpublication 20 March 2009

Lung cancer is a form of cancer that begins in the lungs. There aretwo main types of lung cancer: (i) non-small cell lung cancer(NSCLC) and (ii) small cell lung cancer (SCLC). Non-small cell lungcancer and SCLC account for about 85% and 15%, respectively, ofall the people diagnosed with lung cancer. Lung cancer (both small

cell and non-small cell) is the second most common cancer in bothmen (after prostate cancer) and women (after breast cancer). In avery recent study, the American Cancer Society estimates thatabout 215 020 new cases of (114 690 among men and 100 330among women) and 161 840 deaths from lung cancer (90 810among men and 71 030 among women) will be expected to occurin the US in 2008. It means more people will die from lung cancerthan from colon, breast and prostate cancers combined (1).

Tobacco smoke, particularly of cigarette, is the main contributor tolung cancer. However, many patients with lung cancer have neversmoked. Patients with lung cancer who have never smoked aremore likely to have mutations in epidermal growth factor receptortyrosine kinase and have better response to its inhibitors than dopatients with tobacco-associated lung cancer. Furthermore, the prev-alence of mutations in KRAS and P53 differ for patients with lungcancer who have never smoked and those with tobacco-associatedlung cancer (2). Recently, it has been suggested that women arenot more susceptible than men to the carcinogenic effects of ciga-rette smoking. In smokers, incidence rates tended to be higher inmen than women with comparable smoking histories, but differ-ences were modest; smoking is strongly associated with lung can-cer risk in both men and women (3).

In India, the most popular tobacco smoke is a locally made bidi,which is made of 0.15–0.25 g of sun-dried flaked tobacco rolledinto a conical shape in a dried rectangular piece of Temburni leaf(Diospyros melanoxylon) and a thread securing the roll (4,5). Lungcancer incidence is relatively high among Moslem people and thosewith lower educational history. On considering the age, religion andeducation, the relative risk (RR) between current bidi smokers andthose who had never smoked bidis was 3.9. In further analysesusing only those never smoked cigarettes to examine the effect ofbidi smoking alone on lung cancer risk, current smokers of bidishad the RR of 4.6. Lung cancer incidence increased with largeramounts of bidi smoked a day, with longer durations of smokingbidis, and with younger ages starting smoking bidis (6).

The growing research on tea and cancer prevention during the pastdecade has generated strong evidence, mainly from the cell cultureand animal studies, that tea has a protective effect against carcino-genesis. Tea is an aqueous infusion from the dried leaves of Camel-lia sinensis, which is one of the most widely used beveragesthroughout the world and is known to possess various beneficialproperties that may affect carcinogen metabolism, free radical

627

Chem Biol Drug Des 2009; 73: 627–636

Research Article

ª 2009 John Wiley & Sons A/S

doi: 10.1111/j.1747-0285.2009.00816.x

scavenging or formation of DNA adducts (7,8). Green tea, which isprepared with minimal oxidation of polyphenols, has been shown inanimal studies and human epidemiological studies to preventcancer, including lung cancer (9,10). Studies of the molecularmechanism of the anticancer effects of green tea in general andepigallocatechin-3-gallate (EGCG) in particular have focused onantioxidative and anti-inflammatory effects (11). It has been sug-gested that the green tea polyphenol, EGCG inhibits telomerase andinduces apoptosis in drug-resistant lung cancer cells (8). For years,flavonoids are well known to have antimutagenic and anticarcino-genic activities. To investigate the associations between commonlyconsumed flavonoids and lung cancer, Yan et al. (12) conducted apopulation-based case–control study of 558 lung cancer cases anda group of 837 controls. Certain flavonoids, including epicatechin,catechin, quercetin and kaempferol, were found to be associatedinversely with lung cancer among tobacco smokers, but not amongnon-smokers.

Although supplements multivitamins are frequently used by half thepopulation, limited information is available about their specificeffect on lung cancer risk. In a recent study, it has been suggestedthat supplemental multivitamins, vitamin C, vitamin E and folatewere not associated with a decreased risk of lung cancer. Supple-mental vitamin E was even associated with a small increased risk(13). Vitamin E is a group of eight different naturally occurring com-pounds known as tocopherols and tocotrienols as well as syntheticvitamin E. Epidemiological studies have shown both positive andnegative correlations between plasma concentrations of a-tocoph-erol and lung cancer risk. The findings from the a-tocopherol, b-car-otene cancer prevention study do not support the hypothesis thatsupplementation with this molecule would reduce the incidence oflung cancer. Thus, supplementary vitamin E in any form cannot berecommended for prevention of lung cancer (14).

Despite multiple investigations into the association between aspirinuse and lung cancer risk, no clear answer has emerged, althoughresults point toward a protective association. Later on, the fre-quency, duration, and dose of aspirin and their associations withlung cancer risk were established with matched odds ratios andtests for trend using conditional logistic regression analysis (15). Ina very recent study, it has been suggested that the regular use ofnon-steroidal anti-inflammatory drugs is associated with a slightlyor moderately reduced risk for lung cancer (16). Matrix metallopro-teinases (MMPs) are known to play a key role in the breakdown ofextracellular matrix and in inflammatory processes (17). Matrixmetalloproteinase-1 is the most highly expressed interstitial collage-nase degrading fibrillar collagens. Overexpression of MMP-1 hasbeen shown in tumor tissues and has been suggested to be associ-ated with tumor invasion and metastasis. Thus, MMP-1 is associ-ated with an increased risk for lung cancer, which is modified bysmoking (18).

Among novel chemotherapeutic agents, the taxanes have emergedas the most powerful group of compounds. Taxanes such as paclit-axel (PTX, Taxol; 1) and docetaxel (DTX, Taxotere; 2) are the twomost important anticancer drugs currently used in clinics for thetreatment of various types of cancers, including breast, colon, lung,ovarian and prostate (19–22). These drugs bind to the b-subunit of

tubulin polymer in a stoichiometric ratio and promote tubulin poly-merization. This phenomenon disrupts tubulin polymerization dynam-ics, leading to cell cycle arrest and ultimately, cell death byapoptosis (23–26). Docetaxel is the only agent currently approvedfor both first- and second-line treatment of advanced NSCLC. Multi-ple randomized clinical trials have established the efficacy of plati-num–docetaxel regimens for the first-line treatment of advancedNSCLC. Carboplatin-based regimens and non-platinum combinationswith docetaxel also have proven efficacy in first-line therapy. Com-binations of docetaxel with various novel targeted agents have pro-duced encouraging data in phase II trials (27). Although docetaxelhas made significant progress in the treatment of lung cancer, itsuse often results in various undesired side-effects (28). Therefore, itis important to develop new taxanes with fewer side-effects, supe-rior pharmacological properties, and improved anticancer activity tomaximize the induced benefits for lung cancer patients and reducethe large economic and disease burden worldwide. The quantitativestructure–activity relationship (QSAR) paradigm may be helpful inthe design and development of novel taxane molecules as newanticancer agents, which are expected to show improvements inactivity against lung cancers.

Herein, we demonstrate the QSAR studies on five sets of taxaneanalogues with respect to their activities against lung cancer inorder to understand their chemical–biological interactions. TheQSAR approach employs extra-thermodynamically derived andcomputational-based descriptors to correlate biological activity inisolated receptors, in cellular systems, and in vivo. Four standardmolecular descriptors routinely used in QSAR analysis are electronic,hydrophobic, steric and topological indices. These descriptors areinvaluable in helping to delineate a large number of receptor–ligandinteractions that are critical to biological processes. The quality of aQSAR model depends strictly on the type and quality of the data,and not on the hypotheses, and is valid only for the compoundstructure analogues to those used to build the model. Quantitativestructure–activity relationship models can stand alone to augmentother computational approaches or can be examined in tandem withequations of a similar mechanistic genre to establish their authentic-ity and reliability (19,23,29). Since the advent of the QSAR methodol-ogy about 46 years ago (30), it has become increasingly helpful inunderstanding many aspects of chemical–biological interactions indrug-design process, pesticide research and in the areas of toxicol-ogy (31–35). This method is useful in elucidating the mechanisms ofchemical–biological interaction in various biomolecules, particularlyenzymes, membranes, organelles and cells, as well as in virus, bacte-ria and human (17,31,36–44). It has also been utilized for the evalua-tion of absorption, distribution, metabolism and excretion phenomenain many organisms and whole animal studies (45,46).

Verma and Hansch

628 Chem Biol Drug Des 2009; 73: 627–636

Materials and Methods

All the biological data has been collected from the literature (seeindividual QSAR for respective references). IC50 or GI50 is the molarconcentration of a compound, which inhibits 50% of growth of thecancer cell population. log(1 ⁄ IC50) or log(1 ⁄ GI50) is the subsequentdependent variable that defines the biological parameter for QSARdevelopment. Physicochemical descriptors are autoloaded andmultiregression analyses (MRA) are used to derive the QSAR byusing the C-QSAR programa. Selection of descriptors is made onthe basis of permutation and correlation matrices among thedescriptors to avoid collinearity problems. Details about the C-QSAR

program, the search engine, the choice of parameters and their usein the development of QSAR models, have already been discussed(47,48). Descriptors used in this paper have also been discussedpreviously in detail along with their application (31). Briefly, Clog Pis the calculated partition coefficient of a compound in n-octa-nol ⁄ water and is a measure of its hydrophobicity, whereas p is thehydrophobic parameter for substituents only. Calculated molarrefractivity (CMR) for the whole molecule is calculated from theLorentz–Lorenz equation: [(n 2–1) ⁄ (n 2+2)](MW ⁄ d), where n is therefractive index, MW is the molecular weight and d is the densityof the substance. Molar refractivity (MR) is dependent on both thevolume and polarizability. It can be used for a substituent or for thewhole molecule. Molar refractivity is thus a mean of characterizingthe bulk and polarizability of a substituent or compound. Althoughit contains no information about the shape, it has found consider-able use in biological QSAR, where intermolecular effectspredominate. Molar refractivity is scaled at 0.1 to make it equisca-lar with p (19,23,31).

In all of the QSAR equations, n is the number of data points, ris the correlation coefficient between observed values of thedependent and the values calculated from the equation, r 2 is thesquare of the correlation coefficient and represents the goodnessof fit, q 2 is the cross-validated r 2 (a measure of the quality ofthe QSAR model) and s is the standard deviation. The cross-vali-dated r 2 (q 2) is obtained by using a leave-one-out procedure asdescribed by Cramer et al. (49). Q is the quality factor, for whichQ = r ⁄ s. Chance correlation, because of the excessive number ofdescriptors (which also increases the r and s values), can, there-fore, be detected by the examination of the Q value. High valuesof Q indicate the high predictive power of the QSAR models andthe lack of 'overfitting'. F represents the Fischer statistics,F = fr 2 ⁄ [(1 ) r 2)m], where f is the number of degrees of freedom[f = n ) (m + 1)], n is the number of data points and m is thenumber of variables. The F-value is actually the ratio betweenexplained and unexplained variance for a given number ofdegrees of freedom. Thus, it indicates a true relationship, or thesignificance level for MLR models [within parenthesis the figurein each equation refers to the literature F-value at 99% level(50)]. The modeling was taken to be optimal when Q reached amaximum together with F, even if slightly non-optimal F valueshave normally been accepted. A significant decrease in F withthe introduction of one additional variable (with increasing Q anddecreasing s) could mean that the new descriptor is not as sig-nificant as expected, i.e. its introduction has endangered the

statistical quality of the combination. However, the statistical qual-ity could be improved by the introduction of a more convincingdescriptor (19,23,51,52). Compounds were deemed to be outliers onthe basis of their deviation between observed and predicted activitiesfrom the equation (obsd ) pred > 2s) (53–57). Outliers are thosecompounds which have unexpected biological activities and areunable to fit in a QSAR model. The presence of outliers is not onlydue to the possibility that the molecules may act by different mecha-nisms or interact with the receptor in different binding modes butalso due to the intrinsic noise associated with both the original dataand methodological aspects involved in the construction of a QSARmodel (58–60). Each regression equation includes 95% confidencelimits for each term in parentheses.

Results and Discussion

Inhibition of growth of A549 (NSCLC) cells bytaxane analogues 3QSAR 1 is based on the data obtained from Baloglu et al. (61) (seeTable 1).

logð1=IC50Þ ¼ 11:23ð�2:60ÞMRY � 2:72ð�0:63ÞMRY2

� 1:16ð�2:65Þð1Þ

n = 19, r2 = 0.839, s = 0.119, q2 = 0.766, Q = 7.697, F2,16 = 41.689(6.226)

Optimum MRY = 2.06(2.03–2.11)

Outliers: X = CH=C(CH3)2, Y = CH=CHCH3; X = CH=C(CH3)2, Y = C6H5

X = 2-thiophenyl, Y = 2-thiophenyl

IC50 is the molar concentration of taxane analogues 3 that kills50% of the growth of A549 cancer cell population after 72 h ofexposure. The F-value is the F-ratio between the variances of pre-dicted and observed activities [within parenthesis the figure refersto the literature F-value at 99% level (50)]. The above QSAR modelis a parabolic correlation in terms of MRY (molar refractivity of Ysubstituents), which suggests that the cytotoxic activities of taxanederivatives 3 first increase with an increase in molar refractivity ofY substituents up to an optimum MRY value of 2.06 and thendecreases. Three compounds (X = CH=C(CH3)2, Y = CH=CHCH3;X = CH=C(CH3)2, Y = C6H5 and X = Y = 2-thiophenyl) were deemedto be outliers on the basis of their deviations (obsd ) pred >2s).One derivative (X = CH=C(CH3)2, Y = C6H5) was considered to be an

Taxane Analogues against Lung Cancer: A QSAR Study

Chem Biol Drug Des 2009; 73: 627–636 629

outlier because it was more active than expected, by 4.8 times thestandard deviation. Possible reasons for its unusually high activityare not clear, although its steric bulk or geometry because of the

presence of a phenyl group at Y position may decrease coplanaritywith the X group (CH=C(CH3)2) and maximize the activity (62). Plac-ing the same aromatic (phenyl) group at both the X and Y positionsdecreases the activity of the compound (3-21; X = Y = C6H5), andthis is well predicted. On the other hand, by placing 2-thiophenylgroup at both the X and Y positions increases the activity of thecompound (3-20; X = Y = 2-thiophenyl), which is an outlier. Toassess the effects of excluding outliers, QSAR models were exam-ined before and after the removal of compound. There is not a verygood correlation between pY (hydrophobicity of Y substituents) andMRY (pY versus MRY; r = 0.644). Thus, pY can not replace MRY.By substituting pY for MRY in the above QSAR (eqn 1), theobtained statistics (r 2 = 0.156, s = 0.274, q 2 = )0.115) were notacceptable.

Inhibition of growth of A549 (NSCLC) cells bytaxane analogues 4QSAR 2 is based on the data obtained from Ojima et al. (21) (seeTable 2).

Table 1: Biological and physicochemical parameters used toderive QSAR 1

No. X Y

log(1 ⁄ IC50) (eqn 1)

MRY pYObsd. Pred. D

3-1 CH=C(CH3)2 OC(CH3)3 10.40 10.42 )0.02 2.10 2.043-2 CH=C(CH3)2 CH=C(CH3)2 10.60 10.42 0.18 2.03 1.433-3 CH=C(CH3)2 2-Furyl 10.22 10.22 0.00 1.79 1.003-4a CH=C(CH3)2 CH=CHCH3 10.52 9.74 0.78 1.56 1.033-5a CH=C(CH3)2 C6H5 10.39 9.82 0.57 2.54 1.823-6 CH=C(CH3)2 O(CH2)3CH3 10.38 10.40 )0.02 2.17 2.393-7 CH=C(CH3)2 2-Thiophenyl 10.38 10.11 0.27 2.40 1.663-8 2-Furyl CH=C(CH3)2 10.43 10.42 0.01 2.03 1.433-9 2-Furyl OC(CH3)3 10.51 10.42 0.09 2.10 2.043-10 2-Furyl 2-Furyl 10.05 10.22 )0.17 1.79 1.003-11 2-Furyl CH=CHCH3 9.79 9.74 0.05 1.56 1.033-12 2-Furyl C6H5 9.66 9.82 )0.16 2.54 1.823-13 2-Furyl O(CH2)3CH3 10.44 10.40 0.04 2.17 2.393-14 2-Furyl 2-Thiophenyl 10.0 10.11 )0.11 2.40 1.663-15 2-Thiophenyl CH=C(CH3)2 10.33 10.42 )0.09 2.03 1.433-16 2-Thiophenyl OC(CH3)3 10.48 10.42 0.06 2.10 2.043-17 2-Thiophenyl 2-Furyl 10.26 10.22 0.04 1.79 1.003-18 2-Thiophenyl CH=CHCH3 9.74 9.74 0.00 1.56 1.033-19 2-Thiophenyl O(CH2)3CH3 10.32 10.40 )0.08 2.17 2.393-20a 2-Thiophenyl 2-Thiophenyl 10.44 10.11 0.33 2.40 1.663-21 C6H5 C6H5 9.89 9.82 0.07 2.54 1.823-22 C6H5 OC(CH3)3 10.28 10.42 )0.14 2.10 2.04

aNot used in the derivation of QSAR 1.

Table 2: Biological and physicochemical parameters used to derive QSAR 2

No. X Y Z

log(1 ⁄ IC50) (eqn 2)

MRY pZObsd. Pred. D

4-1 H 2-Furyl OC(CH3)3 9.30 9.06 0.24 1.81 2.044-2 COCH3 2-Furyl OC(CH3)3 9.30 9.06 0.24 1.81 2.044-3 H CH=C(CH3)2 OC(CH3)3 9.70 9.21 0.49 1.92 2.044-4 COCH3 CH=C(CH3)2 OC(CH3)3 8.85 9.21 )0.36 1.92 2.044-5 COCH2CH3 CH=C(CH3)2 OC(CH3)3 9.30 9.21 0.09 1.92 2.044-6 CO(Cy-C3H5) CH=C(CH3)2 OC(CH3)3 9.30 9.21 0.09 1.92 2.044-7 CON(CH3)2 CH=C(CH3)2 OC(CH3)3 9.22 9.21 0.01 1.92 2.044-8 COCH=CHCH3 (E) CH=C(CH3)2 OC(CH3)3 9.30 9.21 0.09 1.92 2.044-9 COOCH3 CH=C(CH3)2 OC(CH3)3 9.30 9.21 0.09 1.92 2.044-10 H CH=CHCH3 (E) OC(CH3)3 8.92 8.57 0.35 1.45 2.044-11a COCH3 CH=CHCH3 (E) OC(CH3)3 9.40 8.57 0.83 1.45 2.044-12 CO(Cy-C3H5) CH=C(CH3)2 C5H11 7.74 7.87 )0.13 1.92 2.244-13 CON(CH3)2 CH=C(CH3)2 C5H11 8.26 7.87 0.39 1.92 2.244-14 COCH=CHCH3 (E) CH=C(CH3)2 C5H11 7.60 7.87 )0.27 1.92 2.244-15 H CH2CH(CH3)2 OC(CH3)3 8.85 9.24 )0.39 1.94 2.044-16 COCH3 CH2CH(CH3)2 OC(CH3)3 9.15 9.24 )0.09 1.94 2.044-17 CO(Cy-C3H5) CH2CH(CH3)2 OC(CH3)3 8.92 9.24 )0.32 1.94 2.044-18 CON(CH3)2 CH2CH(CH3)2 OC(CH3)3 9.22 9.24 )0.02 1.94 2.044-19 H (CH2)2CH3 OC(CH3)3 8.36 8.61 )0.25 1.48 2.044-20 COCH3 (CH2)2CH3 OC(CH3)3 8.33 8.61 )0.28 1.48 2.04

aNot used in the derivation of QSAR 2

Verma and Hansch


log(1/IC50Þ ¼ 1:37ð�0:85ÞMRY�6:76ð�1:94ÞpZ

þ20:34ð�4:01Þð2Þ

n = 19, r 2 = 0.785, s = 0.283, q 2 = 0.657, Q = 3.131,F2,16 = 29.209(6.226)

Outlier: X = COCH3, Y = CH=CHCH3 (E), Z = OC(CH3)3

pZ represents the calculated hydrophobicity of Z substituents.According to this QSAR model, the taxane derivative 4 must have amore hydrophilic Z substituent, and a bulkier or more polarizable Ysubstituent for improved cytotoxicity against A549 cancer cells. Onecompound [X = COCH3, Y = CH=CHCH3 (E), Z = OC(CH3)3] was con-sidered to be an outlier because it was more active than expected,by 2.9 times the standard deviation.

Inhibition of growth of A549 (NSCLC) cells bytaxane analogues 5QSAR 3 is based on the data obtained from Miller et al. (63,64)(see Table 3).

log(1/IC50Þ ¼ 0:62ð�0:28ÞMRX�0:20ð�0:06ÞMRY

þ8:56ð�0:86Þð3Þ

n = 12, r 2 = 0.875, s = 0.208, q 2 = 0.760, Q = 4.500,F2,9 = 31.500(8.022)

Outliers: X = C6H4(3-N(CH3)2), Y = COCH3;

X = C6H3(2,5-(OCH3)2), Y = COCH2CH2(OCH2CH2)10SSCH3

IC50 is the molar concentration of taxane analogues 5 that inhibits50% of the growth of A549 cancer cell after 72 h of drug exposure.MRX and MRY are the calculated molar refractivities of X and Ysubstituents, respectively. According to the above QSAR model, thetaxane derivative 5 must have a bulkier or more polarizable X sub-stituent as well as a small or less polarizable Y substituent forimproved cytotoxicity against A549 cancer cells. Two compounds(5-6 and 5-14) were considered to be outliers because these wereless and more active than expected, by 8.32 and 4.62 times,respectively, the standard deviation.

Inhibition of growth of human lung cancer PC-6(SCLC) cells by paclitaxel, docetaxel and taxaneanalogues 6QSAR 4 is based on the data obtained from Takeda et al. (65) (seeTable 4).

log(1/GI50Þ ¼ �2:74ð�1:34ÞClogP þ 0:71ð�0:48ÞCMR

þ 9:16ð�7:24Þð4Þ

n = 9, r 2 = 0.809, s = 0.396, q 2 = 0.649, Q = 2.270, F2,6 =12.707(10.925)

Outliers: 6-1 and 6-3

Table 3: Biological andphysicochemical parameters usedto derive QSAR 3 No. X Y

log(1 ⁄ IC50) (eqn 3)

MRX MRYObsd. Pred. D

5-1 C6H5 COCH2CH2SSCH3 9.10 9.41 )0.31 2.51 3.505-2 C6H4(3-Cl) COCH3 10.52 10.23 0.29 3.00 0.965-3 C6H3(3,5-F2) COCH3 10.00 9.94 0.06 2.54 0.965-4 C6H4(3-OCH3) COCH3 10.40 10.31 0.09 3.13 0.965-5 C6H3(2,5-(OCH3)2) COCH3 10.52 10.69 )0.17 3.74 0.965-6a C6H4(3-N(CH3)2) COCH3 9.00 10.73 )1.73 3.81 0.965-7 C6H4(3-Cl) COCH2CH2SSCH3 9.54 9.72 )0.18 3.00 3.505-8 C6H3(3,5-F2) COCH2CH2SSCH3 9.44 9.43 0.01 2.54 3.505-9 C6H4(3-OCH3) COCH2CH2SSCH3 10.00 9.80 0.20 3.13 3.505-10 C6H3(2,5-(OCH3)2) COCH2CH2SSCH3 10.10 10.18 )0.08 3.74 3.505-11 C6H5 CO-1-piperazine-4-CH2CH2SSCH3 8.96 8.92 0.04 2.51 5.925-12 C6H3(2,5-(OCH3)2) CO-1-piperazine-4-CH2CH2SSCH3 9.52 9.69 )0.17 3.74 5.925-13 C6H3(2,5-(OCH3)2) COCH2CH2(OCH2CH2)4SSCH3 9.54 9.31 0.23 3.74 7.835-14a C6H3(2,5-(OCH3)2) COCH2CH2(OCH2CH2)10SSCH3 8.96 8.00 0.96 3.74 14.31

aNot used in the derivation of QSAR 3.



GI50 is the molar concentration of paclitaxel, docetaxel and taxaneanalogues 6 that inhibits the growth of PC-6 cancer cells by 50% at72 h of continuous drug exposure. Clog P and CMR are the calcu-lated hydrophobicity and molar refractivity of the whole molecule,respectively. Although there is a high correlation between Clog Pand CMR (r = 0.785), both parameters were used in the developmentof the above equation, because a meaningful correlation was notobtained by using either one of the descriptor (Clog P or CMR). Inthis model, 61.7% of the variance in the data is explained by the ste-ric ⁄ polarizability descriptor CMR while the hydrophobic descriptoraccounts for only 19.2% of the variance in the data. As Clog P valuesof the compounds 6 in Table 4 are relatively high (4.08–5.37), thesecompounds may have log P over the optimum value (i.e the equationmay represents only the second half portion of the parabolic or bilin-ear model in terms of Clog P ), deriving the negative coefficient ofthe Clog P term. The negative coefficient of Clog P suggests that theinhibitory activity of these molecules decreases with increasing theirhydrophobicity. On the contrary, the increase in CMR increases theinhibitory activity of these compounds (positive coefficient). Thus,more hydrophilic and bulkier or polarizable compounds 6 would pres-ent better inhibitory activity against PC-6 cancer cells. Two com-pounds (6-1 and 6-3) were considered to be outliers because thesewere less active than expected, by 2.93 and 5.20 times, respectively,the standard deviation.

Inhibition of growth of PC-6 ⁄ VP1-1 (resistantSCLC) cells by paclitaxel, docetaxel and taxaneanalogues 6QSAR 5 is based on the data obtained from Takeda et al. (65) (seeTable 4).

log(1/GI50Þ¼�1:53ð�0:87ÞClogPþ0:83ð�0:33ÞCMR�1:77ð�5:40Þð5Þ

n = 9, r 2 = 0.863, s = 0.297, q 2 = 0.616, Q = 3.128,F2,6 = 18.898(10.925)

Outlier: 6-2

QSARs 4 and 5 represent the cytotoxicities of paclitaxel, docet-axel and a series of taxane analogues 6 against two differentcancer cell lines (PC-6 and PC-6 ⁄ VP1-1), and are very similar toeach other, which suggests that these compounds may act by asimilar mechanism and ⁄ or target the microtubules of these twocancer cell lines.

Validation of the QSAR modelsValidation is always a crucial step in the development of statisti-cally robust QSAR models because the real utility of a QSAR modelis in its ability to predict accurately the modeled property for newcompounds not present in the data set. Criteria of validation for theQSAR models have already been discussed previously (66–71).Table 5 lists the regression coefficients and statistical parametersfor all the QSAR models. The predictivity of the QSAR models wasevaluated by using both 'internal validation' and 'external valida-tion'.

Internal validation

Fraction of the variance (r 2)The values of r 2 for these five QSAR models (eqns 1–5) areranging from 0.785 to 0.875, which suggests that these QSARmodels explain 78.5–87.5% of the variance in the data. Accordingto the literature, the predictive QSAR model must have r 2 > 0.6(70,71).

Cross-validation testThe cross-validated r 2 (q 2) values for these QSAR models (eqns 1–5) are ranging from 0.616 to 0.766. According to the literature, thepredictive QSAR model must have q 2 > 0.5 (70,71).

Standard deviation (s)The smaller the value of s, the better the QSAR model. The valuesof s for these QSAR models (eqns 1–5) are from 0.119 to 0.396.

Table 4: Biological and physico-chemical parameters used toderive QSARs 4 and 5No. X

log(1 ⁄ GI50) (eqn 4) log(1 ⁄ GI50) (eqn 5)

Clog P CMRObsd. Pred. D Obsd. Pred. D

1 Paclitaxel 11.83 11.77 0.06 8.93 9.30 )0.37 4.73 22.132 Docetaxel 12.49 12.51 )0.02 9.26 9.08 0.18 4.08 20.66

6-1a H 12.54 13.70 )1.16 10.67 10.92 )0.25 4.34 23.376-2b 5-OH 13.66 13.03 0.63 9.20 10.60 )1.40 4.63 23.526-3a 3-OH 11.10 13.16 )2.06 ND 10.68 ND 4.58 23.526-4 5-CH3 12.43 12.66 )0.23 10.81 10.54 0.27 4.84 23.836-5 5-C2H5 11.37 11.54 )0.17 10.10 10.11 )0.01 5.37 24.296-6 5-F 12.47 13.09 )0.62 10.84 10.58 0.26 4.57 23.386-7 5-Cl 12.08 11.86 0.22 10.38 10.10 0.28 5.14 23.866-8 5-CF3 11.22 11.24 )0.02 9.67 9.76 )0.09 5.37 23.886-9 5-OCH3 13.13 12.99 0.14 10.51 10.78 )0.27 4.76 23.98

ND, not determined.aNot used in the derivation of QSAR 4.bNot used in the derivation of QSAR 5.

Verma and Hansch


Quality factor (Q)High values of Q (2.270–7.697) for these QSAR models (eqns 1–5)suggest their high predictive power.

Fischer statistics (F )The larger the value of F, the greater the probability that the QSARmodel is significant. The F-values for all the QSAR models (eqns1–5) are ranging from 12.707 to 41.689, which are statistically sig-nificant at the 99% level.

Y-Randomization testIn this test, the dependent-variable vector (Y vector) is randomlyshuffled, and a new QSAR model is developed using the originalindependent variable matrix. The process is repeated several times.It is expected that the resulting QSAR models should have low r 2

and low q 2 values. This is a widely used technique to ensure therobustness of a QSAR model. The statistical data of r 2 and q 2 forfive runs are listed in Table 6 (eqns 1–5). The poor values of r 2

and q 2 in the Y-randomization test confirm the robustness of allthe five QSAR models (53,71–73).

Lack of overfittingA QSAR model overfits if it includes more descriptors than required.The lack of overfitting for all the QSAR models (eqns 1–5) waschecked by using the following conditions: (i) number of datapoints ⁄ number of descriptors ‡4; (ii) high values of Q; (iii) compari-son of the original QSAR model with that of fewer descriptors; (iv)Y-randomization test (Table 6) suggests that the high r 2 values ofall the QSAR models (eqns 1–5) are not because of the chancecorrelation or overfitting (23,74).

External validation

Selection of the training and test setsThe original data set of QSAR model (eqn 1) was divided intotraining [n = 14 (�75%)] and test [n = 5 (�25%)] sets in a randommanner. The QSAR model for this training set was generated byusing the same descriptors as those of QSAR 1 and validated onthe basis of their statistics (acceptance criteria: r 2 > 0.6 andq 2 > 0.5). The predictive capacity of this model was judged fromtheir predictive R2 (R2

pred) value, which was calculated by eqn 6:

R 2pred ¼ 1�

PðYpredðtestÞ � YtestÞ2

PðYtest � Y trainingÞ2

ð6Þ

In eqn 6, Ypred(test) and Ytest are the respective predicted andobserved activities of the test set compounds and Y training is theobserved mean activity of the training set compounds. Similarly, theQSAR models (eqns 2–5) were also divided into training and testsets for their external validation. A random selection pattern of thetest sets as well as the regression coefficients and the statisticalparameters of their respective training sets for all these QSAR mod-els (eqns 1–5) are given in Table 7.

New molecule predictionQSAR 1 (parabolic correlations in terms of MRY) is one of the mostencouraging examples in which the optimum value of MRY is welldefined. We believe that this model may prove to be an adequatepredictive model for providing guidance in design and synthesis,and for yielding very specific compounds 3 that may have highanti-lung cancer activities. On the basis of this QSAR model, sixtaxane analogues (3-23 to 3-28) are suggested as potential syn-thetic targets. Clog P and CMR of the molecules can be collinear

Table 5: Comparison of the regression coefficients and the statistical parameters obtained from the multiregression analyses process forQSARs 1–5

QSAR no. System n

Regression coefficients Statistical parameters

Hydrophobic Molar refractivity Intercept r 2 q 2 s Q F a

1 A549 cells 19 11.23 MRY ) 2.72 MRY2 )1.16 0.839 0.766 0.119 7.697 41.689(6.226)

2 A549 cells 19 )6.76 pZ 1.37 MRY 20.34 0.785 0.657 0.283 3.131 29.209(6.226)3 A549 cells 12 0.62 MRX ) 0.20 MRY 8.56 0.875 0.760 0.208 4.500 31.500(8.022)4 PC-6 cells 9 )2.74 Clog P 0.71 CMR 9.16 0.809 0.649 0.396 2.270 12.707(10.925)5 PC-6 ⁄ VP1-1 cells 9 )1.53 Clog P 0.83 CMR )1.77 0.863 0.616 0.297 3.128 18.898(10.925)

aThe figure within parenthesis referred to the literature F-value at 99% level.

Table 6: Y-Randomization data for QSARs 1–5

QSAR no.

NOR-1a NOR-2 NOR-3 NOR-4 NOR-5

r 2 q 2 r 2 q 2 r 2 q 2 r 2 q 2 r 2 q 2

1 0.460 0.268 0.093 )0.189 0.564 0.329 0.042 )0.248 0.280 )0.0032 0.225 0.035 0.053 )0.376 0.118 )0.082 0.126 )0.020 0.063 )0.1613 0.259 )0.305 0.207 )1.637 0.472 0.018 0.264 )0.148 0.424 )0.5724 0.026 )0.616 0.039 )0.448 0.254 )2.100 0.333 )0.005 0.081 )1.0185 0.612 0.215 0.443 )0.509 0.737 0.371 0.452 )0.161 0.483 )0.333

aNOR, number of Y-randomization.



unless care is taken in substituents selection. Of course Clog P isimportant in bioavailability; however, usually values in the range of1.50–3.00 suffice for this purpose. The proposed compounds (3-23

to 3-28) have Clog P range from 1.54 to 2.98, suggesting its possi-ble good bioavailability. There is not a good correlation betweenClog P and CMR for these proposed six compounds (r = 0.326). Thepredicted log(1 ⁄ IC50) values of these proposed taxane analogues(3-23 to 3-28) obtained from the QSAR model (eqn 1) are givenin Table 8 along with their physicochemical parameters.

Conclusion

The success of docetaxel in the treatment of lung cancers eitherusing alone or in combination with various novel targeted agentshas been tempered by the development of various undesired side-effects. These limitations have led to the search for new taxaneswith improved biological activity. We believe that the QSAR para-digm may be helpful in the design and development of novel taxanemolecules that are expected to have improved anti-lung canceractivities. Our QSAR results from this study suggest that the inhibi-tory activities of taxane analogues against lung cancers are depen-dent on both the molar refractivity and hydrophobic descriptors ofthe compound ⁄ substituents, with a major contribution coming fromthe molar refractivity of the compound ⁄ substituents. On the basisof the parabolic QSAR model (eqn 1), we suggested six taxaneanalogues (3-23 to 3-28) as potential synthetic targets with pre-dicted log(1 ⁄ IC50) values of 10.42. Further QSAR studies should notonly enlarge the areas of their application, but also increase ourunderstanding of the mechanisms of chemical–biological interac-tions.

References

1. Jemal A., Siegel R., Ward E., Hao Y., Xu J., Murray T., ThunM.J. (2008) Cancer statistics, 2008. CA Cancer J Clin;58:71–96.

2. Subramanian J., Govindan R. (2008) Molecular genetics of lungcancer in people who have never smoked. Lancet Oncol;9:676–682.

3. Freedman N.D., Leitzmann M.F., Hollenbeck A.R., Schatzkin A.,Abnet C.C. (2008) Cigarette smoking and subsequent risk of lungcancer in men and women: analysis of a prospective cohortstudy. Lancet Oncol;9:649–656.

4. Gupta P.C., Murti P.R., Bhonsle R.B. (1996) Epidemiology of can-cer by tobacco products and the significance of TSNA. Crit RevToxicol;26:183–198.

5. Reddy S., Gupta P.C. (2004) Report on Tobacco Control in India.New Delhi, India: Ministry of Health and Family Welfare, Gov-ernment of India.

6. Jayalekshmy P.A., Akiba S., Nair M.K., Gangadharan P., Rajan B.,Nair R.K., Sugahara T. (2008) Bidi smoking and lung cancer inci-dence among males in Karunagappally cohort in Kerala, India.Int J Cancer;123:1390–1397.

7. Zhang H., Spitz M.R., Tomlinson G.E., Schabath M.B., MinnaJ.D., Wu X. (2002) Modification of lung cancer susceptibility bygreen tea extract as measured by the comet assay. CancerDetect Prev;26:411–418.

8. Sadava D., Whitlock E., Kane S.E. (2007) The green tea polyphe-nol, epigallocatechin-3-gallate inhibits telomerase and inducesapoptosis in drug-resistant lung cancer cells. Biochem BiophysRes Commun;360:233–237.

9. Fujiki H., Suganuma M., Matsuyama S., Miyazaki K. (2005) Can-cer prevention with green tea polyphenols for the general popu-lations and for patients following cancer treatment. Curr CancerTher Rev;1:109–114.

10. Clark J., You M. (2006) Chemoprevention of lung cancer by tea.Mol Nutr Food Res;50:144–151.

11. Yoneda T., Hiramatsu K., Sakamoto M., Togasaki K., KomatsuM., Yamaguchi K. (1995) Antioxidant effects of catechins. Bio-chem Mol Biol Int;35:995–1008.

12. Cui Y., Morgenstern H., Greenland S., Tashkin D.P., Mao J.T., CaiL., Cozen W., Mack T.M., Lu Q.-Y., Zhang Z.-F. (2008) Dietary fla-vonoid intake and lung cancer – a population-based case–con-trol study. Cancer;112:2241–2248.

13. Slatore C.G., Littman A.J., Au D.H., Satia J.A., White E. (2008)Long-term use of supplemental multivitamins, vitamin C, vitamin

Table 7: Random selection pattern of the test set compounds as well as the regression coefficients and statistical parameters of theQSAR for their respective training sets obtained from the division of the original QSARs 1–5

QSAR no.a Test set compd Training set compd

Regression coefficients Statistical parameters

pZ Clog P MRX MRY MRY2 CMR Intercept r 2 q 2 s R2

pred

1 3-2, 3-7, 3-11, 3-12, 3-18 Rest of the compd (n = 14) 11.11 )2.68 )1.12 0.813 0.580 0.088 0.8502 4-5, 4-7, 4-10, 4-12, 4-18 Rest of the compd (n = 14) )6.42 1.76 18.92 0.750 0.536 0.310 0.8083 5-1, 5-5, 5-11 Rest of the compd (n = 9) 0.60 )0.21 8.71 0.823 0.548 0.196 0.8974 6-5, 6-6, 6-8 Rest of the compd (n = 6) )3.07 0.83 7.98 0.835 0.539 0.356 0.7775 1, 6-4, 6-6 Rest of the compd (n = 6) )1.12 0.63 0.81 0.919 0.805 0.199 0.752

aQSAR for the respective training set of the original QSAR.

Table 8: Predicted biological activities [log(1 ⁄ IC50); mol ⁄ L]obtained from QSAR 1 along with physicochemical parameters ofthe proposed molecules 3-23 to 3-23

Compound X Y Clog P CMR MRY

log(1 ⁄ IC50)(eqn 1)

3-23 C6H5 CH=CHCOCH3 2.94 22.87 2.11 10.423-24 CH=C(CH3)2 CH=CHCOCH3 2.98 22.19 2.11 10.423-25 2-Furyl CH=CHCOCH3 2.11 22.08 2.11 10.423-26 2-Thiophenyl CH=CHCOCH3 2.58 22.68 2.11 10.423-27 2-Pyrrolyl CH=CHCOCH3 1.54 22.30 2.11 10.423-28 2-Pyrrolyl CH=C(CH3)2 2.78 22.19 2.03 10.42

Verma and Hansch


E and folate does not reduce the risk of lung cancer. Am JRespir Crit Care Med;177:524–530.

14. Cohen V., Dawood S., Khuri F.R. (2007) Tocopherols and lungcancer. In: Preedy V.R., Watson R.R., editors. Encyclopedia ofVitamin E. Wallingford, UK: CABI Publishing Corp; p. 915–921.

15. McGovern A.L., Hoffman S.C., Helzlsouer K.J., Alberg A.J. (2007)Aspirin use and risk of lung cancer. In: Alvarez C.M., editor.Focus on Cancer Research. Hauppauge, NY: Nova Science Pub-lishers Inc; p. 51–62.

16. Olsen J.H., Friis S., Poulsen A.H., Fryzek J., Harving H., Tjoenne-land A., Soerensen H.T., Blot W. (2008) Use of NSAIDs, smokingand lung cancer risk. Br J Cancer;98:232–237.

17. Verma R.P., Hansch C. (2007) Matrix metalloproteinases (MMPs):chemical–biological functions and QSARs. Bioorg MedChem;15:2223–2268.

18. Sauter W., Rosenberger A., Beckmann L., Kropp S., MittelstrassK., Timofeeva M., Woelke G. et al. (2008) Matrix metalloprotein-ase 1 (MMP1) is associated with early-onset lung cancer. Can-cer Epidemiol Biomarkers Prev;17:1127–1135.

19. Verma R.P., Hansch C. (2008) Taxane analogues against breastcancer: a quantitative structure–activative relationship study.ChemMedChem;3:642–652.

20. Roh E.J., Kim D., Lee C.O., Choi S.U., Song C.E. (2002) Struc-ture–activity relationship study at the 3¢-N-position of paclitaxel:synthesis and biological evaluation of 3¢-N-acyl-paclitaxel ana-logues. Bioorg Med Chem;10:3145–3151.

21. Ojima I., Slater J.C., Kuduk S.D., Takeuchi C.S., Gimi R.H., SunC.-M., Park Y.H., Pera P., Veith J.M., Bernacki R.J. (1997) Synthe-ses and structure–activity relationships of taxoids derived from14ß-hydroxy-10-deacetylbaccatin III. J Med Chem;40:267–278.

22. Arbuck S.G., Blaylock B.A. (1995) Taxol: clinical results and cur-rent issues in development. In: Suffness M., editor. Taxol: Sci-ence and Applications. Boca Raton, FL: CRC Press; p. 379–415.

23. Hansch C., Verma R.P. (2008) Understanding tubulin ⁄ microtubule-taxane interactions: a quantitative structure–activity relationshipStudy. Mol Pharmaceutics;5:151–161.

24. Wang T.-H., Wang H.-S., Song Y.-K. (2000) Paclitaxel-inducedcell death: where the cell cycle and apoptosis come together.Cancer;88:2619–2628.

25. Blagosklonny M.V., Fojo T. (1999) Molecular effects of paclitaxel:myths and reality (a critical review). Int J Cancer;83:151–156.

26. Schiff P.B., Fant J., Horwitz S.B. (1979) Promotion of microtubuleassembly in vitro by taxol. Nature;277:665–667.

27. Saloustros E., Georgoulias V. (2008) Docetaxel in the treatmentof advanced non-small-cell lung cancer. Expert Rev AnticancerTher;8:1207–1222.

28. Hanna N., Shepherd F.A., Fossella F.V., Pereira J.R., De MarinisF., von Pawel J., Gatzemeier U. et al. (2004) Randomized phaseIII trial of pemetrexed versus docetaxel in patients with non-small-cell lung cancer previously treated with chemotherapy. JClin Oncol;22:1589–1597.

29. Selassie C.D., Mekapati S.B., Verma R.P. (2002) QSAR: then andnow. Curr Top Med Chem;2:1357–1379.

30. Hansch C., Maloney P.P., Fujita T., Muir R.M. (1962) Correlationof biological activity of phenoxyacetic acids with Hammett sub-stituent constants and partition coefficients. Nature;194:178–180.

31. Hansch C., Leo A. (1995) Exploring QSAR: Fundamentals andApplications in Chemistry and Biology. Washington, DC: Ameri-can Chemical Society.

32. Bermudez-Saldana J.M., Cronin M.T.D. (2006) Quantitative struc-ture–activity relationships for the toxicity of organophosphorusand carbamate pesticides to the rainbow trout onchorhynchusmykiss. Pest Manag Sci;62:819–831.

33. Hansch C., Telzer B.R., Zhang L. (1995) Comparative QSAR intoxicology: examples from teratology and cancer chemotherapyof aniline mustards. Crit Rev Toxicol;25:67–89.

34. Schultz T.W., Hewitt M., Netzeva T.I., Cronin M.T.D. (2007)Assessing applicability domains of toxicological QSARs: defini-tion, confidence in predicted values, and the role of mechanismsof action. QSAR Comb Sci;26:238–254.

35. Selassie C.D., Garg R., Kapur S., Kurup A., Verma R.P., MekapatiS.B., Hansch C. (2002) Comparative QSAR and the radical toxic-ity of various functional groups. Chem Rev;102:2585–2605.

36. Winkler D.A. (2002) The role of quantitative structure–activityrelationships (QSAR) in biomolecular discovery. Brief Bioin-form;3:73–86.

37. Fujikawa M., Nakao K., Shimizu R., Akamatsu M. (2007) QSARstudy on permeability of hydrophobic compounds with artificialmembranes. Bioorg Med Chem;15:3756–3767.

38. Verma R.P., Hansch C., Selassie C.D. (2007) Comparative QSARstudies on PAMPA ⁄ modified PAMPA for high throughput profil-ing of drug absorption potential with respect to CaCO-2 cellsand human intestinal absorption. J Comput Aided MolDes;21:3–22.

39. Kodithala K., Hopfinger A.J., Thompson E.D., Robinson M.K.(2002) Prediction of skin irritation from organic chemicals usingmembrane–interaction QSAR analysis. Toxicol Sci;66:336–346.

40. Verma R.P. (2007) Cytotoxicity of heterocyclic compounds againstvarious cancer cells: a quantitative structure–activity relationshipstudy. Top Heterocycl Chem;9:53–86.

41. Garg R., Gupta S.P., Gao H., Babu M.S., Debnath A.K., HanschC. (1999) Comparative quantitative structure–activity relationshipstudies on anti-HIV drugs. Chem Rev;99:3525–3601.

42. Verma R.P., Hansch C. (2008) Combating the threat of anthrax: aquantitative structure–activity relationship approach. MolPharmaceutics;5:745–759.

43. Verma R.P., Kurup A., Mekapati S.B., Hansch C. (2005) Chemi-cal–biological interactions in human. Bioorg Med Chem;13:933–948.

44. Hall L.H., Hall L.M. (2005) QSAR modeling based on structure-information for properties of interest in human health. SARQSAR Environ Res;16:13–41.

45. Hansch C., Leo A., Mekapati S.B., Kurup A. (2004) QSAR andADME. Bioorg Med Chem;12:3391–3400.

46. Khan M.T.H., Sylte I. (2007) Predictive QSAR modeling for thesuccessful predictions of the ADMET properties of candidatedrug molecules. Curr Drug Discov Technol;4:141–149.

47. Hansch C., Hoekman D., Leo A., Weininger D., Selassie C.D.(2002) Chem-Bioinformatics: comparative QSAR at the interfacebetween chemistry and biology. Chem Rev;102:783–812.

48. Verma R.P., Hansch C. (2007) Development of QSAR modelsusing C-QSAR program: a regression program that has dual data-bases of over 21 000 QSAR models. Nat Protoc. doi: 10.1038/



nprot.2007.125. Available at: http://www.natureprotocols.com/2007/03/05/development_of_qsar_models_usi_1.php

49. Cramer R.D. III, Bunce J.D., Patterson D.E., Frank I.E. (1988)Cross validation, bootstrapping and partial least squares com-pared with multiple regression in conventional QSAR studies.QSAR;7:18–25.

50. Bennett C.A., Franklin N.L. (1967) Statistical Analysis in Chemis-try and the Chemical Industry. New York: John Wiley & Sons; p.708–709 (5th printing).

51. Pogliani L. (2000) From molecular connectivity indices to semi-empirical connectivity terms: recent trends in graph theoreticaldescriptors. Chem Rev;100:3827–3858.

52. Pogliani L. (1996) Modeling with special descriptors derived froma medium-sized set of connectivity indices. J PhysChem;100:18065–18077.

53. Selassie C.D., Kapur S., Verma R.P., Rosario M. (2005) Cellularapoptosis and cytotoxicity of phenolic compounds: a quantitativestructure–activity relationship study. J Med Chem;48:7234–7242.

54. Verma R.P., Hansch C. (2006) Cytotoxicity of organic compoundsagainst ovarian cancer cells: a quantitative structure–activityrelationship study. Mol Pharmaceutics;3:441–450.

55. Hansch C., Verma R.P. (2007) 20-(S)-Camptothecin analogues asDNA topoisomerase I inhibitors: a QSAR study. ChemMed-Chem;2:1807–1813.

56. Verma R.P., Hansch C. (2007) Understanding human rhinovirusinfections in terms of QSAR. Virology;359:152–161.

57. Verma R.P., Hansch C. (2008) Investigation of DNA-binding prop-erties of organic molecules using quantitative structure–activityrelationship (QSAR) models. J Pharm Sci;97:88–110.

58. Verma R.P., Hansch C. (2005) An approach toward the problemof outliers in QSAR. Bioorg Med Chem;13:4597–4621.

59. Polanski J., Bak A., Gieleciak R., Magdziarz T. (2006) Modelingrobust QSAR. J Chem Inf Model;46:2310–2318.

60. Ekins S., Mestres J., Testa B. (2007) In silico pharmacology fordrug discovery: methods for virtual ligand screening and profil-ing. Br J Pharmacol;152:9–20.

61. Baloglu E., Miller M.L., Roller E.E., Cavanagh E.E., Leece B.A.,Goldmacher V.S., Chari R.V.J. (2004) Synthesis and biologicalevaluation of novel taxoids designed for targeted delivery totumors. Bioorg Med Chem Lett;14:5885–5888.

62. Verma R.P., Kapur S., Barberena O., Shusterman A., Hansch C.H.,Selassie C.D. (2003) Synthesis, cytotoxicity, and QSAR analysisof X-thiophenols in rapidly dividing cells. Chem Res Toxi-col;16:276–284.

63. Miller M.L., Roller E.E., Wu X., Leece B.A., Goldmacher V.S.,Chari R.V.J., Ojima I. (2004) Synthesis of potent taxoids fortumor-specific delivery using monoclonal antibodies. Bioorg MedChem Lett;14:4079–4082.

64. Miller M.L., Roller E.E., Zhao R.Y., Leece B.A., Ab O., Baloglu E.,Goldmacher V.S., Chari R.V.J. (2004) Synthesis of taxoids withimproved cytotoxicity and solubility for use in tumor-specificdelivery. J Med Chem;47:4802–4805.

65. Takeda Y., Uoto K., Chiba J., Horiuchi T., Iwahana M., Atsumi R.,Ono C., Terasawa H., Soga T. (2003) New highly active taxoidsfrom 9b-dihydrobaccatin-9,10-acetals. Part 4. Bioorg MedChem;11:4431–4447.

66. Jaworska J.S., Comber M., Auer C., Van Leeuwen C.J. (2003)Summary of workshop on regulatory acceptance of (Q)SARs forhuman health and environmental endpoints. Environ Health Per-spect;111:1358–1360.

67. Eriksson L., Jaworska J., Worth A.P., Cronin M.T.D., McDowellR.M., Gramatica P. (2003) Methods for reliability and uncertaintyassessment and for applicability evaluations of classification- andregression-based QSARs. Environ Health Perspect;111:1361–1375.

68. Cronin M.T.D., Walker J.D., Jaworska J.S., Comber M.H.I., WattsC.D., Worth A.P. (2003) Use of QSARs in international decision-making frameworks to predict ecologic effects and environmen-tal fate of chemical substances. Environ Health Per-spect;111:1376–1390.

69. Cronin M.T.D., Jaworska J.S., Walker J.D., Comber M.H.I., WattsC.D., Worth A.P. (2003) Use of QSARs in international decision-making frameworks to predict health effects of chemical sub-stances. Environ Health Perspect;111:1391–1401.

70. Golbraikh A., Tropsha A. (2002) Beware of q2! J Mol GraphModel;20:269–276.

71. Tropsha A., Gramatica P., Gombar V.K. (2003) The importance ofbeing earnest: validation is the absolute essential for successfulapplication and interpretation of QSPR models. QSAR CombSci;22:69–77.

72. Wold S., Eriksson L. (1995) Statistical validation of QSAR results.Validation tools. In: van de Waterbeemd H., editor. ChemometricsMethods in Molecular Design (Methods and Principles in Medic-inal Chemistry, Vol 2). Weinheim, Germany: VCH; p. 309–318.

73. Melzig M.F., Tran G.D., Henke K., Selassie C.D., Verma R.P.(2005) Inhibition of neutrophil elastase and thrombin activity bycaffeic acid esters. Pharmazie;60:869–873.

74. Zhang S., Golbraikh A., Oloff S., Kohn H., Tropsha A. (2006) Anovel automated lazy learning QSAR (ALL-QSAR) approach:method development, applications, and virtual screening ofchemical databases using validated ALL-QSAR models. J ChemInf Model;46:1984–1995.

Note

aC-QSAR Program, Claremont, CA, USA: BioByte Corp. Available at:

http://www.biobyte.com

Verma and Hansch


taxane analogues against lung cancer: a quantitative structure–activity relationship study

Documents