anticancer activity of estradiol derivatives a quantitative structure activity relationship approach

Upload: irwan-hikmawan

Post on 06-Apr-2018

227 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/3/2019 Anticancer Activity of Estradiol Derivatives a Quantitative Structure Activity Relationship Approach

    1/4

    Research: Science and Education

    1390 Journal of Chemical Education Vol. 78 No. 10 October 2001 JChemEd.chem.wisc.edu

    Today it would cost as much as $500 million to bringa new drug to market, and clinical development and theapproval process can take over 10 years while the success ratesof lead compounds remain very low. A systematic and efficientapproach to drug discovery is imperative (13).

    Medicinal chemistry draws on the principles of chemistry,biochemistry, molecular biology, and pharmacology to introducenew therapeutic agents (4). To facilitate the drug discoveryprocess, quantitative structureactivity relationship (QSAR)analysis using state-of-the-art computing technology has becomea standard method of pharmaceutical and agrochemical sci-entists. By QSAR, one can exploit a simple view of ligand

    (drug) structure versus biological or pharmacological activitiesin conventional terms even in the absence of informationregarding the three-dimensional structure (X-ray crystallo-graphic data) of a receptor or an enzyme (57).

    QSAR techniques are very practical in lead optimizationand can produce a serendipitous discovery. Numerous adap-tations and applications are being found in industry. In termsof the pedagogical value of QSAR or QSPR (quantitativestructureproperty relationship) with real-world data, only a fewarticles have appeared in thisJournal(811). Commercial pack-ages to implement modern QSAR techniques are highly priced,but the essence of QSAR can be taught without them. As a class-room example, published data on anticancer activities of estra-

    diol analogs (12) will be analyzed by a QSAR approach usingonly the Lotus function in Microsoft Excel 97 and theMicrosoft Excel 97 statistical tool.W

    StructureActivity Data for Cytotoxicity of EstradiolAnalogs

    The formation of new blood vessels (angiogenesis) is acritical factor in the growths of tumors. Antiangiogenictherapy is a promising concept that has actually been appliedin the treatment of tumors and angiogenic diseases. The firststeroid to have inhibitory effects on angiogenesis was 2-methoxyestradiol, which is believed to be a potent inhibitor of

    endothelial cell proliferation and migration (13). 2-Methoxy-estradiol is a naturally occurring mammalian metabolite of thefemale sex hormone estradiol. It blocks mitosis via inhibitionof tubulin polymerization by binding to the colchicine (anantineoplastic agent) binding site of tubulin (12).

    Analogs of 2-methoxyestradiol (see structure) weresynthesized, and their cytotoxicities and antitubulin activi-ties were examined (12).

    R

    HO

    OH

    1

    2

    34

    6

    17

    A B

    C D

    2-Methoxyestradiol

    The results are summarized in Table 1. Compound numberscorrespond to those used in the original article (12). With theexception of the 17-acetate derivative 13, analogs were modi-fied only at the 2-position. Because compound 13 does not havesignificant inhibitory effect, it was not included in this analysis.

    The common logarithms of the reciprocal values of thetubulin polymerization inhibitory activities, IC50 (M), wereused to define the biological response. This indicates that thesmaller the concentration used to cause the inhibitory effect is,the larger is the log of 1/IC50 and thus larger is the response.

    Building the QSAR Models

    The aim of QSAR studies is to best correlate the physico-chemical properties and structural features of a set of thecongeners with the observed biological responses. Because thestructural modifications were introduced solely at the carbon-2position in the structure shown, appropriate descriptors orparameters for the substituents or molecules can be correlatedto the observed biological or pharmacological responses (depen-dent variable), and can thus be the explanatory (independent)variables in multiple linear regression. The regression models arethe QSAR models that can be used to predict or design theoptimized candidates for the lead compounds.

    Anticancer Activity of Estradiol Derivatives: WA Quantitative StructureActivity Relationship Approach

    Ken Muranaka

    Ks Garden Nishioji, Suite 401, Kasuga Hachijo Sagaru, Minami-ku, Kyoto 601-8312, Japan;[email protected]

    sisylanARASQrofpetStsriF,elbaTretemaraP.1elbaT

    .oN R2-C CI 05 CI/1(gol 05 ) RM I de I rb

    1 HC 3O 9.2 835.5 20.0 97.0 1 0

    a8 C2H5O 19.0 140.6 83.0 52.1 1 0

    b8 n C- 3H7O 2.4 773.5 50.1 17.1 1 0

    c8 iC- 3H7O 8.4 913.5 63.0 17.1 1 0

    9 HC 3 71 077.4 65.0 65.0 0 0

    a01 H2 HC=C 4.2 026.5 28.0 1.1 1 0

    b01 HC 3 HC=HC 1.1 959.5 22.1 65.1 1 0

    c01 C2H5 HC=HC 6.8 660.5 30.2 1 0

    d01 nC- 3H7 HC=HC 04 893.4

    1 0e01 HC( 3)2 HC=C 4.9 720.5 30.2 1 1

    a11 C2H5 7.7 411.5 20.1 30.1 0 0

    b11 n C- 3H7 9.4 013.5 55.1 5.1 0 0

    c11 n C- 4H9 04 893.4 31.2 69.1 0 0

    d11 n C- 5H 11 04 893.4 2.2 24.2 0 0

    e11 HC( 3)2 HCHC 2 04 893.4 7.1 69.1 0 1

    31 )cAO-71(I 04 893.4

    41 I 8.4 913.5 21.1 93.1 0 0

    51 C2H5S 01 000.5 70.1 48.1 0 0

    22 HC 3 HNOC 04 893.4 79.0 94.1 0 1

    32 C2H6N 3 325.5 80.0 5.1 1 0

    http://jchemed.chem.wisc.edu/Journal/http://jchemed.chem.wisc.edu/Journal/Issues/2001/Oct/http://jchemed.chem.wisc.edu/http://jchemed.chem.wisc.edu/Journal/Issues/2001/Oct/abs1390.htmlhttp://jchemed.chem.wisc.edu/Journal/Issues/2001/Oct/abs1390.htmlhttp://jchemed.chem.wisc.edu/http://jchemed.chem.wisc.edu/Journal/Issues/2001/Oct/http://jchemed.chem.wisc.edu/Journal/
  • 8/3/2019 Anticancer Activity of Estradiol Derivatives a Quantitative Structure Activity Relationship Approach

    2/4

    Research: Science and Education

    JChemEd.chem.wisc.edu Vol. 78 No. 10 October 2001 Journal of Chemical Education 1391

    Numerous physicochemical properties and structuralparameters have been devised for QSAR studies (1520). Hy-drophobicity is the most used property. It is the 1-octanol/water partition coefficient (P); its logarithm, log P, for a mol-ecule or for a substituent (symbolized as ) can characterizethe optimized candidate structures in terms of solubility inwater. The electronic effect (known as the Hammett )may provide a clue to the best substituents on the basis of

    their electron-releasing or electron-withdrawing properties.Tafts steric effect (Es), molar refractivity (MR), or Verloopparameters can define structural features. Indicator variablesto define properties or characteristics by 1 (presence) or 0(absence) can also be included. QSAR models can be empirical(e.g., physicochemical properties), quantum chemical (e.g., netatomic charges), nonempirical (e.g., molecular connectivity),or three-dimensional, and may be classified accordingly.

    Finding the QSAR parameters (independent variables)that best explain the observed activities is important in order tointerpret the data and elucidate the lead compounds. GivenQSAR data, we may have many different QSAR models andmany different interpretations (e.g., given a historical chart

    of the Dow Jones Industrial Average, a variety of econometricmodels can be proposed). Once the QSAR parameters havebeen defined, statistical judgment based on correlation coef-ficient (r) andp (probability) values for these parameters inmultiple regression can be used to propose the best models.

    As indicated in Table 1, the substituents hydrophobicity(), molar refractivity (MR), and two indicator variables (Ied,Ibr) were chosen as the QSAR parameters for this study. Allvalues of and MR were obtained from the literature (17). Ied

    is 1 if oxygen, nitrogen, or a double bond is present to definethe effect of increased electron density; otherwise, Ied is set equalto 0. Similarly, Ibr indicates the effect of branching; it is definedto be 1 for branched substituents and otherwise is 0.

    Table 2 is the correlation matrix for the QSAR parameters.In multiple linear regression, regressors should be independentof one another; otherwise, there is a problem called multi-colinearity, and the proposed model would not be reliable.

    In this QSAR table, and MR are correlated by .55 (55%). Anincrease in MR means larger substituents; hydrophobicity shouldincrease with increasing number of carbon atoms. These twovariables may provide similar information about size. Whenconstructing a multiple linear regression model, inclusion ofboth and MR as explanatory variables must be avoided.

    Multiple linear regression can be done easily if a Lotustool in Excel is used. Alternatively, the add-in for statisticalanalysis (Data Analysis in Tools) in Excel can be used easilyto implement the regression analysis, as output containingstandard errors andp values are automatically generated. TheLotus tool has more pedagogical value because a student mustcomputep values from standard errors.W

    Results and Discussion

    Two QSAR models were obtained. The values in paren-theses are p values, and a parameter is considered statisticallysignificant ifp < .05; ris the regression correlation coefficient,and n is the number of the observations. The first model op-timizes the structureactivity data with the QSAR param-eters , the quadratic term 2, and the indicator variable Ied.

    log(1/IC50) = 4.963 + 0.384 0.2872 + 0.593Ied (1)

    (0.031) (0.011) (0.006)

    r= .881; n = 16

    All thep values for the regression coefficients are less than.05 (5%), and ris high (88.1%). Because the coefficient ofIed is positive, an increase in electron density should increasethe biological response. This is consistent with the fact thatthe congeners with oxygen or double bond (8aand 10b) showa strong inhibitory effect on tubulin polymerization.

    The quadratic term for the hydrophobic factor was in-cluded in eq 1 so that an optimized value ofcan be obtained.But the optimum can be found only if the regression coefficientof2 is negative, because such a parabola in terms of opensdownward. Taking the partial derivatives of both sides of eq 1with respect to and solving for , we find 0.67 to be the

    optimal value for the hydrophobicity, which suggests thatslightly hydrophobic substituents are favored. The fact that suchan optimum exists and the finding that an increase in electrondensity will increase the response would explain why the responseof compound 8a(the ethoxy substituent containing an oxygen)is so high.

    Figure 1 is a graph indicating how the observed inhibitioncan be explained by hydrophobicity. Two theoretical plots(for Ied= 0, 1) are drawn. The parabolic curves predict thepresence of an optimum hydrophobic structure. However, theQSAR model obtained (eq 1) is valid only for the values thatare between the two extreme values in the data set (Table 1).This is a limitation of QSAR approach as a predictive tool.

    sretemaraPRASQehtrofxirtaMnoitalerroC.2elbaT

    CI/1(gol 05 ) RM I de I rb

    CI/1(gol05) 1 10862.0 28314.0 963775.0 38514.0

    1 869945.0 66173.0 56942.0

    RM 1 5650.0 115172.0

    Ide

    1 63761.0

    Irb

    1

    4.0

    4.5

    5.0

    5.5

    6.0

    6.5

    -1.0 -0.5 0.0 1.51.00.5 2.0

    log(1/IC50

    )

    Observed

    Ied = 1

    Ied = 0

    Figure 1. Parabolic relationship between inhibition and hydro-phobicity.

    http://jchemed.chem.wisc.edu/http://jchemed.chem.wisc.edu/Journal/Issues/2001/Oct/http://jchemed.chem.wisc.edu/Journal/http://jchemed.chem.wisc.edu/Journal/http://jchemed.chem.wisc.edu/Journal/Issues/2001/Oct/http://jchemed.chem.wisc.edu/
  • 8/3/2019 Anticancer Activity of Estradiol Derivatives a Quantitative Structure Activity Relationship Approach

    3/4

    Research: Science and Education

    1392 Journal of Chemical Education Vol. 78 No. 10 October 2001 JChemEd.chem.wisc.edu

    The second QSAR model uses MR, Ied, and Ibr as theindependent variables.

    log(1/IC50) =

    4.137 + 1.640MR 0.665MR2 + 0.555Ied 0.453Ibr (2)

    (0.031) (0.011) (0.000) (0.015)

    r= .911; n = 18

    All the regression coefficients are significant, and the correla-tion rhas improved. Because the regression coefficient of MR2

    is negative, the optimal MR should exist, and it has beencomputed to be 1.23. Again, the regression coefficient ofIedis positive so that increased electron density is advantageous toenhance the inhibitory effect. Branching in the substituentsis not favored, as the regression coefficient ofIbris negative.The molar refractivities of8a, 10a, and 10b are close to theoptimum. These substituents are not branched, and they havean increased electron density due to oxygen or a double bond.

    Table 1 shows that the natural product 2-methoxy-estradiol (1) is a potent inhibitor of tubulin polymerization.But the two QSAR models suggest that more potent analogs

    of this natural product should exit. The optimal value of MRindicates that there is a critical size factor for the substituent.As branching would decrease the activity, the shape of the sub-stituent is important also. Therefore the optimized structureof the substituent at the 2-position must be a 2- or 3-carbonunbranched chain containing a double bond, oxygen, ornitrogen. This should explain the high inhibitory activitiesof8a, 10a, and 10b. The low activity of 10c, though thestructure is unbranched and has a double bond, is then clear:it is too long, or the MR is much larger than the optimum.

    The branching index proposed by Randic (11, 21, 22) maybe used in place of the indicator variable Ibr, but eq 2 would notbe statistically valid (i.e.,p values for the QSAR parameters will

    be greater than .05). Though controversial, this topologicalindex was shown to be highly correlated to molar refractivity(and possibly to the partition coefficient) and thus would notgive medicinal chemists any additional information for findingmore potent inhibitors (2326). Molar refractivity and hydro-phobicity are correlated (Table 1), and both can be a measureof how bulky the substituent R is. However, to really determinethe effect of branching in R on inhibition, the indicator vari-able should be an appropriate choice in this QSAR analysis.The Randic branching index can be computed by hand, andthe method is well explained in thisJournal(9, 11). Proposinganother QSAR model with this index as a variable may be agood exercise. If a student has access to molecular modelingsoftware, net atomic (oxygen) charge on the C3-OH positionmay be a better parameter than theindicator variable Ied.

    Chance correlation may occur when there are more vari-ables than the actual number of observations ( 27, 28).Multicolinearity can be a problem. Outliers can profoundlyaffect the analysis. Building a multiple linear regression model without being aware of these problems should be avoided(29). Partial least squares (PLS) regression may be used, butthis will require a software package other than Excel (3033). PLS can produce a model even when there are moreQSAR parameters to be screened than there are observations;cross-validation technique evaluates the resulting model interms of how well itpredicts(rather than how well itfits) the

    data. PLS is used in the three-dimensional QSAR approachCoMFA (comparative molecular field analysis).

    More information about QSAR/QSPR and relevant soft-ware is available on the Web (30, 31, 3446).

    Conclusion

    An underlying concept in rational design and discovery

    of new drugs is the pharmacophore, a three-dimensionalarrangement of key molecular features that would formspecific interactions with a target receptor (18). Biologicalresponse is elicited upon formation of a drugreceptorcomplex for which there is one rate-determining reaction atthe active site. The true effectiveness of a drug is then itsability to selectively hit the target so that adverse reactions will be minimized. QSAR models linearly correlate thestructural features and physicochemical properties of a drugmolecule with the biological activity by multiple linearregression technique (19).

    Biological response cannot be explained by a one-stepreaction. It is also highly likely that a biological system is a

    nonlinearsystem. Molecular descriptors used in QSAR must beincomplete. The QSAR approach has been criticized as being aconceptual offshoot of what used to be called absolute reactionrate theory and as not having been very successful, even whensupplemented by a host of empirical and semiempirical infor-mation coming directly from bioassays (47).

    A pharmacophore, however, can be used to suggest newcompounds that might exhibit an improved biological response.For example, a proposed pharmacophore for tubulin polymer-ization inhibitor has the following structural features at the2-position: (i) increased electron density, (ii) slightly hydro-phobic group, and (iii) an unbranched chain with the lengthof 2 or 3 carbon atoms. Further, relative distance information

    given by the 3-hydroxyl, the 2-position, and the center ofring A in the steroid backbone may be an additional searchquery. Items to define a pharmacophore are like the keywordsfor a database search. One can search through the databasesthat contain three-dimensional structures of known chemicalsand drugs to devise novel structures that best match theproposed pharmacophore. Pharmacophore identification can bedone without crystallographic data on the target receptor orenzyme. The QSAR approach is therefore an effort to reducethe time and the money in lead generation. Not only that,the QSAR/QSPR approach is applied in many areas to reduceanimal testing (38), to protect the environment by predictingtoxicity (37, 48), to estimate drug absorption in humans (49),and to predict chromatographic retention coefficients (50),to mention a few.

    Numerous structureactivity data are published in journalssuch asJournal of Medicinal Chemistry. CA Selectsof ChemicalAbstracts Service has the title StructureActivity Relationships(51). Although three-dimensional QSAR techniques arepopular (18), one can still perform a two-dimensional QSARanalysis on published data by multiple linear regression usingExcel or Lotus. It would be a challenging exercise to comparethe results of the two-dimensional analysis and the publishedthree-dimensional QSAR models. The major structural re-quirements revealed should have a good agreement betweenthe two-dimensional and three-dimensional analyses (52).

    http://jchemed.chem.wisc.edu/Journal/http://jchemed.chem.wisc.edu/Journal/Issues/2001/Oct/http://jchemed.chem.wisc.edu/http://jchemed.chem.wisc.edu/http://jchemed.chem.wisc.edu/Journal/Issues/2001/Oct/http://jchemed.chem.wisc.edu/Journal/
  • 8/3/2019 Anticancer Activity of Estradiol Derivatives a Quantitative Structure Activity Relationship Approach

    4/4