(CANCER RESEARCH 52, 3892-3900, July IS, 1992]
`
`Identification of Novel Antimitotic Agents Acting at the Tubulin Level by
`Computer-assisted Evaluation of Differential Cytotoxicity Data
`Kenneth D. Paull, Chii M. Lin, Louis Malspeis, and Ernest Hamel1
`Information Tec~nology Branch {~. D. P.J and Laboratory of Molecular Pharmacology {C. M. L., E. H.J. Developmental Therapeutics Program, Division of Cancer
`Treatment, National Cancer Institute, NIH, Bethesda, Maryland 10891, and Laboratory of Pharmaceutical Chemistry, Developmental Therapeutics Program
`DMsion of Cancer Treatment, National Cancer Institute, Frederick Cancer Research and Development Center, Frederick, Maryland 11701 {L. M.J
`'
`
`ABSTRACT
`
`Data generated in the new National Cancer Institute drug evaluation
`proaram, which are based on inhibition of cell growth in 60 human
`tumor cell lines, were probed with nine known antimitotic agents usina
`the COMPARE alprithm. Cytotoxicity data were available on approx(cid:173)
`imately 7000 compounds at the time of the analysis. and, based on the
`criteria used, 82 compounds were selected as positive by the computer
`search. Nine were the probe compounds themselves. and 41 were ana(cid:173)
`lopes of known antimitotic qents. Amoq the remaining 32 com(cid:173)
`pounds there were 19 distinct chemical species. Agents in ten of these
`groups (containing 20 compounds) were effective inhibitors of in ,itro
`tubulin polymerization and caused the mitotic arrest of cells grown in
`culture. Two compounds were related natural products binding in the
`Vinca domain of tubulin, and the others were synthetic qents which
`interfered with colchicine binding. The remaining 12 agents (one natural
`product, the remainder synthetic) fell into several groups: two com(cid:173)
`pounds were weak inhibitors of tubulin polymerization, inhibited colch(cid:173)
`icine bindina, and caused mitotic arrest; one compound weakly inhibited
`tubulin polymerization but did not cause an increase in the number of
`cells arrested in mitosis; two compounds caused mitotic arrest at micro(cid:173)
`molar concentrations. but thus far no in ,itro interaction with tubulin
`has been observed; the remainder neither inhibited tubulin polymeriza(cid:173)
`tion nor caused a rise in the number of cultured cells arrested in mitosis.
`Tubulin-clependent GTP hydrolysis was stimulated or inhibited by all
`qents which inhibited tubulin polymerization with the exception of one
`compound. The analysis of differential cytotoxicity data thus appears to
`have p-eat promise for the identification of new antimitotic qents with
`antineoplastic potential.
`
`INTRODUCTION
`
`The National Cancer Institute has screened large numbers of
`compounds for many years in a continuing search for effective
`antineoplastic drugs. Recently the initial in vivo screen (murine
`P388 leukemia) has been replaced with a tissue culture screen of
`60 human tumor cell lines, with a major goal being the devel(cid:173)
`opment of tumor-specific therapeutic agents (1-3). Data ob(cid:173)
`tained are entered into a database on a VAX 9000 computer. At
`the time the study presented here was initiated (late 1990) over
`7000 compounds, including about 200 .. standard" agents of
`known therapeutic utility and/or mechanism of action, had
`been processed through the screen.
`Several analytical approaches to this growing database are
`being explored. We have developed an algorithm, COMPARE,
`which evaluates patterns of cytotoxicity among the cell lines in
`the screen (4). This algorithm permits the pattern of cytotox(cid:173)
`icity against the 60 cell lines obtained with any agent to be
`compared with those obtained with all other agents in the da(cid:173)
`tabase. The algorithm also provides a numerical evaluation
`(Pearson correlation coefficient) of the degree of similarity be-
`
`tween the patterns obtained with two agents. We noted that
`when the probe compound (or .. seed") was a known antimitotic
`agent, many of the compounds identified as having similar dif(cid:173)
`ferential cytotoxicity patterns were known to interfere with tu(cid:173)
`bulin function. We wondered whether other compounds of un(cid:173)
`known mechanism of action identified in these screening probes
`would also interact with tubulin. Since such interactions are
`readily confirmed in biochemical assays, we initiated this sys(cid:173)
`tematic evaluation of the potential of the COMPARE algo(cid:173)
`rithm to provide mechanistic information based on differential
`cytotoxicity data.
`
`MATERIALS AND METHODS
`
`Materials. Tubulin and heat-treated microtubule-associated pro(cid:173)
`teins were prepared as described elsewhere (S). [8-14C]GTP (repurified
`by anion exchange chromatoaraphy) and [3H]colchicine were obtained
`from Moravek Biochemicals and Du Pont, respectively. Ail drup were
`provided by the Drug Synthesis and Chemistry Branch, National Can(cid:173)
`cer Institute. They were dissolved in dimethyl sulfoxide, and equivalent
`amounts of the solvent were included in all control reaction mixtures.
`Monosodium aJutamate (from Sigma) was repurified to remove Mg2+
`(6). HL-60 human leukemia cells were a gift of Dr. T. Breitman (Lab(cid:173)
`oratory of Biological Chemistry, National Cancer Institute).
`The COMPARE Ataorithm. Each time a compound is tested in the
`NCP screen, the data obtained must pass quality control standards
`prior to entry in the database. These data are processed to yield delta
`values for each cell line, defined as the difference obtained when the
`10110 of its TGI is subtracted from the 10110 of the mean of the TGls of
`all successfully evaluated cell lines in the specific test with the com(cid:173)
`pound.
`When a compound is selected as a seed for a probe of the database,
`the delta value for each cell line successfully tested with the seed is
`compared to the delta values with the same cell lines (if successfully
`tested) for all compound entries in the database. For each compound in
`the database a set of pairs of delta values is obtained. The commercially
`available SAS statistical proaram was used to calculate a Pearson prod(cid:173)
`uct moment correlation coefficient for each set of delta value pairs. Ail
`compounds in the data base are rank-ordered in comparison to the seed
`for similarity of pattern of differential cytotoxicity, with a correlation
`coefficient of 1.0 signifying identical patterns.
`Methods. The tubulin polymerization assays have been described
`(7, 8). In brief, varying concentrations of drug were preincubated for 1 S
`min with 1.0 ma/ml (10 µM) tubulin in 1.0 M monosodium aJutamate
`plus MaCl2 as indicated. GTP was added (0.4 mM), and polymerization
`was followed turbidimetrically at 350 nm for 20 min. The drug con(cid:173)
`centration required to inhibit the extent of polymerization by SO% was
`determined from the data. At least three experiments were performed
`with each agent. The binding of (3H]colchicine to tubulin (triplicate
`samples) was determined by the DEAE-cellulose filter technique (9).
`For measurement of GTP hydrolysis (10), samples were applied to
`polyethyleneimine-cellulose thin-layer sheets. Following thin-layer
`chromatoaraphy, product (8- 14C)GDP and residual [8-1"C)GTP were
`
`Received 2/6/92; accepted 5/6/92.
`The costs of publication of this anicle were defrayed in pan by the payment of
`page charges. This article must therefore be hereby marked advertisement in accor(cid:173)
`dance with 18 U.S.C. Section 1734 solely to indicate this fact.
`1 To whom requests for reprints should be addressed, at Building 37, Room
`5C25, NIH, Bethesda, MD 20892.
`
`2 The abbreviations used are: NCI, National Cancer Institute; TGI, lowest drug
`concentration that completely inhibits growth; IC50, concentration of drug required
`to inhibit the increase in cell number by 5()411, relative to the increase in control
`cultures.
`3892
`
`

`

`COMPUTER-ASSISTED EVALUATION OF CYTOTOXICITY DATA
`
`located by autoradiography and quantitated by counting in a liquid
`scintillation spectrometer.
`HL-60 cells were grown at 37•c in a 5% CO2 atmosphere in 5-ml
`suspension cultures in RPMI 1640 containing 17% fetal calf serum
`(both from Gibco-BRL), 0.3% L-glutamine, 0.01 mg/ml gentamicin
`sulfate, varying drug concentrations, and 1% (v/v) dimethyl sulfoxide.
`For determination of IC50 values, cells were counted after 40 h. Since
`cell number did not decrease significantly with drug treatment during
`this period, the IC50 value is defined as the concentration of drug
`required to inhibit the increase in cell number by 50% relative to the
`increase in the control cultures. For the determination of the mitotic
`index, cells were grown for 18 h. They were collected by centrifugation
`at 1000 rpm for 5 min. The cells were first washed with phosphate(cid:173)
`buffered saline (pH 7 .2), recollected by centrifugation, swollen by re(cid:173)
`suspension for 10 min in 0.033 M phosphate buffer, and recollected by
`centrifugation. The cells were fixed by the addition of ice-cold 1.5%
`(v/v) ethanol-0.5% (v/v) acetic acid. After IS min they were recollected
`by centrifugation, resuspended in 0.25 ml of 75% ethanol-25% acetic
`acid, and transferred to a microscope slide. After the solvent had evap(cid:173)
`orated, the cells were stained with Giemsa and examined by bright-field
`microscopy. At least 200 cells were examined for the determination of
`a mitotic index, with control values routinely less than 5%.
`
`RESULTS
`Initial Evaluation of Compounds the COMPARE Algorithm
`Indicated Were Antimitotic Agents. Substantial numbers of
`agents have now been evaluated in the new human cell cytotox(cid:173)
`icity drug screen of the NCI. Most of the compounds evaluated
`have been newer submissions, but approximately 200 standard
`agents have also been examined. These standard agents in(cid:173)
`cluded a number of antimitotic drugs. The COMPARE algo(cid:173)
`rithm was developed to permit the rapid selection of com(cid:173)
`pounds with similar patterns of cytotoxicity toward the tumor
`panel. The algorithm rank-orders all entries in the database for
`similarity of pattern of differential cytotoxicity relative to the
`seed compound. The database was initially probed with five
`antimitotic compounds, vincristine, vinblastine, colchicine,
`podophyllotoxin, and taxol. We arbitrarily selected the 100
`compounds most similar in pattern to each of the seeds. There
`was considerable overlap among the five lists, and most com(cid:173)
`pounds on these lists were analogues of known antimitotic
`agents. Compounds with novel structures on these lists were
`first evaluated for effects on in vitro tubulin polymerization.
`Positive and negative compounds in the tubulin polymerization
`assay were then evaluated in terms of their cytotoxicity with the
`human tumor cell lines and in terms of their similarity of dif(cid:173)
`ferential cytotoxicity pattern toward the seeds. That is, the
`quantitative correlation coefficients were examined.
`Fig. 1 (previously presented in Ref. 11) is an attempt to
`present a visual image of what the COMPARE algorithm eval(cid:173)
`uates, patterns of differential cytotoxicity. Even though there
`were wide differences in the cytotoxicity of maytansine, the
`halichondrins, and VM-26 (see figure legend for details), some
`cell lines were more sensitive and others less sensitive than
`average toward each drug. In terms of a quantitative compari(cid:173)
`son of patterns, with maytansine as seed, halichondrin B and
`homohalichondrin 8 had correlation coefficients of 0.8, while
`that ofVM-26 was 0.3. In computing the correlation coefficient
`the algorithm does not consider the quantitative deviation from
`the average as much as the qualitative pattern of more resistant
`or more sensitive. The ovarian carcinoma cell line OVCAR-3 is
`· ·
`h
`h
`)"
`000 ti Id
`bo
`a ut 1
`- o more sens1t1ve t an t e average me to may-
`tansine but only 12-fold more sensitive to the halichondrins.
`
`The Computer Search to Evaluate the COMPARE Algo(cid:173)
`rithm. Based on the initial evaluations, we determined that the
`COMPARE algorithm would yield optimal results with anti(cid:173)
`mitotic agents by imposing two restrictions on the compounds
`selected with any seed. First, the correlation coefficient should
`be at least 0.6. Second, our initial results indicated that com(cid:173)
`pounds with low cytotoxicity generally did not greatly affect
`tubulin polymerization. Therefore, we imposed the second cri(cid:173)
`terion that selected compounds be toxic (50% growth inhibi(cid:173)
`tion) at 1 µM or less in the original screen with HL-60 (TB)
`human leukemia cells.
`Originally nine seeds were used. These agents were taxol,
`vincristine, vinblastine, colchicine, podophyllotoxin, may(cid:173)
`tansine, rhizoxin, dolastatin 10, and combretastatin A-4. The
`most potent compounds indicated by these nine seeds to be
`potential antimitotic agents were homohalichondrin B (NSC
`609394) and halichondrin B (NSC 609395). These are complex
`polyether natural products derived from sponges of the genera
`Halichondria (12) and Axinella (13). We confirmed that these
`two agents inhibited tubulin polymerization and that halichon(cid:173)
`drin B was a noncompetitive inhibitor of Vinca alkaloid binding
`to tubulin, as presented elsewhere (11). Since halichondrin B
`and homohalichondrin B may bind at a distinct site on tubulin,
`the database was probed again with halichondrin B as a tenth
`seed, but no additional compounds were obtained.
`Results of the Computer Search. Eighty-two compounds (in(cid:173)
`cluding halichondrin B and homohalichondrin 8) met the cri(cid:173)
`teria summarized above. Besides the 9 seeds themselves,3 there
`were 13 analogues of podophyllotoxin, 3 of colchicine, 9 of
`dolastatin 10, 7 of combretastatin A-4, 3 of taxol, 3 carbamates,
`and 2 benzylbenzodioxole derivatives (14). In addition there
`were 32 structurally novel compounds (19 distinct chemical
`species) as summarized in Table 1 and Figs. 2-S (4 compounds,
`indicated as Compounds A-Din Table 1, remain proprietary,
`and we are unable to present their structures).
`These 82 compounds were generally identified with multiple
`seeds. Including the probe with halichondrin 8, only 7 com(cid:173)
`pounds were identified with 1 seed, 2 with 2 seeds, 1 with 4
`seeds, and 1 with 5 seeds. In contrast, 3 compounds were iden(cid:173)
`tified with 6 seeds, 5 with 7 seeds, 25 with 8 seeds, 24 with 9
`seeds, and 14 with 10 seeds.
`Evaluation of the Novel Chemical Structures for ;,, Yitro
`Inhibition of Tubulin Polymerization and u Antimitodc
`Agents. Since we had initially surveyed most of the compounds
`presented in Figs. 2-5 for effects on tubulin polymerization to
`validate the utility of the COMPARE algorithm, we knew that
`most of these agents inhibited the reaction. To obtain quanti(cid:173)
`tative measures of these inhibitory effects we determined the
`IC50 values of these compounds after a 20-min incubation
`(Table 1).
`First we examined all compounds under our "standard" assay
`condition. Tubulin (10 µM) was preincubated with drug in 1.0 M
`monosodium glutamate (commercial preparation) supple(cid:173)
`mented with 1 mM MgC'2 for 15 min at 37•c.4 The polymer(cid:173)
`ization reaction was initiated at 37"C following the addition of
`GTP. The preincubation results in substantial reduction in IC50
`values for drugs which bind slowly to tubulin, especially colch(cid:173)
`icinoids (15, 16).
`
`J When the database is probed using the COMPARE algorithm, the seed com-
`pound always identifies itself.
`' Most commercial preparations of monosodium glutamate that we haw exam-
`ined have been contaminated wit'1-endogenous" Mg2+, in amounts sufficient to
`yield concentrations up to 1 mM in I M glutamate.
`3893
`
`

`

`COMPUTER-ASSISTED EVALUATION OF CYTOTOXICITY DATA
`
`HOMOHALICHONDRIN B
`
`VM-26
`
`•
`
`~
`
`•
`
`HUMAN TUMOR
`CELL LINES TESTED
`
`Leukemia
`
`Non-Small Cell Lung
`
`Small Cell Lung
`
`Colon
`
`Central Nervous System
`
`Melanoma
`
`Ovarian
`
`Renal
`
`MAYTANSINE ;-y = • c._
`9 ...
`~
`~
`-= • 2
`3
`& -• ----• --•
`.....
`
`+
`
`HALICHONDRIN B
`
`•
`
`?
`
`9
`
`r
`r
`t=
`•
`' =-• -F
`=
`!=- ~
`' i=-
`~ £:_
`~
`t
`f
`~ ,.
`j
`}
`~
`r ~
`~
`• -•
`& • • =--
`~
`~
`----.
`
`+
`
`+
`
`+
`DIFFERENTIAL CYTOTOXICITY
`(Relative Deviation from Mean)
`Fig. I. Patterns of differential cytotoxicity toward human tumor cell lines. Drugs entered into the new National Cancer Institute screen are evaluated against 60
`different human tumor cell lines. Their cytotoxic effects are evaluated and entered into a database on a VAX 9000 computer. The data summarized here are modified
`to present a visual image consistent with the COMPARE screening algorithm. For each cell line the TGI is obtained. For each agent a mean TGI log is determined,
`defined as the mean of the log111s of the individual TGI values. For each agent the difference between the log1o of each cell line and the mean TGI log for that agent
`is determined, to yield positive values for cell lines more sensitive than average (ban profeding to IU right) and negative values for cell lines less sensitive than average
`(ban projttting to the left). The algorithm permits these values to be compared for all agents in the database, yielding Pearson correlation coefficients. For the fipre,
`the data for each agent have been normali:zed against the maximally sensitive cell line for that agent, with the maximum deviation of equal length with all agents. The
`NCI screening data have yielded average TGI values of about 0.1 nM for maytansine, 7 DM for halichondrin B and homobalichondrin 8, and S l'M for VM-26. The
`maximally sensitive cell lines for each agent relative to the average TGI for that agent were as follows: about I 000-fold with maytansine; about 12-fold with balicbondrin
`8; about 12-fold with homohalichondrin B; and about 8-fold with VM-26. A circle on axis indicates that a cell line was not suc:cessf'ully tested with the agent; a lflUITt
`centered on an axis indicates that the agent yielded a TGI value equivalent to the average value. The cell lines used in the NCI screen (order as in the fipre): leukemia
`lines CCRF-CEM, HL-60 (TB), K-562, MOLT-4, RPMl-8226, and SR; non-small cell lung carcinoma lines AS49/ATCC, EKVX, HOP-18, HOP-62, HOP-92,
`NCI-H226, NCI-H23, NCI-H322M, NCI-H460, NCI-HS22, and LXFL-S29L; small cell lung carcinoma lines OMS 114 and OMS 273; colon carcinoma lines COLO
`205, OLD-I, HCC-2998, HCT-116, HCT-15, HT29, KMl2, KM20L2, and SW-620; central nervous system cancer lines SF-268, SF-295, SF-539, SNB-19, SNB-75,
`SNB-78, U2S I, and XF 498; melanoma lines LOX IMVI, MALME-3M, M 14, M 19-MEL, SK-MEL-2, SK-MEL-28, SK-MEL-S, UACC-257, and UACC-62; ovarian
`carcinoma lines IGROVI, OVCAR-3, OVCAR-4, OVCAR-S, OVCAR-8, and SK-OV-3; and renal carcinoma lines 786-0, A498, ACHN, CAKl-1, RXF-393,
`RXF-631, SNl2C, TK-10, and U0-31.
`
`Recently, with benzylbenzodioxole analogues, we noted little
`progression of inhibition of turbidity development at higher
`drug concentrations (8). We attributed this to aberrant polymer
`formation analogous to that described with colchicine in the
`presence of higher concentrations of Mg2+ ( 17, 18). A similar
`problem was observed with at least two of the compounds ex(cid:173)
`amined here (NSC 624285 and NSC 635478). All apparently
`, The compounds which yielded JC50 values only in the sensitive assay all had
`inactive and weakly active compounds were also examined in a
`second polymerization reaction condition (the .. sensitive" as-
`readily apparent effects on the standard polymerization reaction, as compared with
`the control reaction. In particular, there were sipif'ICllllt reductions in the rate of
`say) (8). The monosodium glutamate used in these experiments
`turbidity development. These agents failed to adequately suppress the rise in tur-
`was repurified to remove Mg2+ (confirmed by atomic absorp-
`bidity in the standard assay to permit determination of IC50 values.
`3894
`
`tion spectroscopy), the reaction mixture was supplemented with
`0.25 mM MgCJi, and the preincubation and incubation took
`place at 30"C. This second reaction condition permitted ready
`quantitation of IC50 values for additional compounds which
`interfere with tubulin polymerization. 5
`
`

`

`COMPUTER-ASSISTED EVALUATION OF CYTOTOXICITY DATA
`
`Table I Effects of novel chemical structures predicted to be antitubulin agents by the COMPARE algorithm on tubulin polymerization and on mitosis in HL-60
`human leukemia cells
`Inhibition of tubulin polymerization6 (IC~ µM)
`
`Inhibition of HL-60 cell growth'
`
`Compound
`(NSC no.)
`56030
`
`No. of Seeds"
`10
`
`37"C/I mM Mg2+
`system
`11±0.7d
`
`30"C/0.25 mM Mg2+
`system
`
`83292
`
`609394r
`609395h
`
`619859
`
`622691
`
`Compound Ai
`624285
`624544
`624545
`624546
`624547
`
`Compound Bi
`Compound Ci
`Compound Di
`
`630032
`
`631583
`
`9
`
`8
`8
`
`10
`
`10
`8
`9
`8
`8
`6
`
`9
`9
`9
`
`9
`
`8
`
`>100
`
`~15'
`7.2±0.3
`
`>100
`
`>100
`
`2.5±0.3
`>100
`2.4±0.6
`4.2±0.6
`2.3±0.1
`2.1±0.5
`
`6.3±0.3
`8.7±0.2
`7.9±0.3
`
`12±2
`
`51±3
`
`2.5±0.3
`
`7.6±0.6
`
`2.5±0.5
`
`0.94±0.08
`3.0±0.6
`
`S.8±0.5
`
`IC,o (l'M)
`0.2
`
`0.3
`
`0.001•
`O.OOOJR
`
`3
`
`0.5
`
`0.4
`0.07
`
`0.4
`
`0.9
`
`2
`
`% mitoses
`(l'M drug)
`26(2)
`
`25(3)
`
`22(0.0l)R
`23(0.00l)R
`
`26(10)
`
`14(2.S)
`
`29(4)
`27(0.7)
`
`24(4)
`
`32(8)
`
`30(20)
`
`635477
`635478
`635479
`
`626391
`
`627777
`
`625538
`
`83265
`
`622093
`
`76455
`
`376265
`
`628301
`
`8
`8
`8
`
`9
`
`7
`
`8
`
`8
`
`9
`
`3.3±0.3
`>100
`S.5±0.S
`
`90±8
`
`>100
`
`53±2
`
`>100
`
`>100
`
`>100
`
`>100
`
`>100
`
`1.0±0.1
`1.6±0.2
`
`15±1
`
`21±2
`
`16±0.2
`
`>100
`
`>100
`
`>100
`
`>100
`
`>100
`
`0.2
`
`7
`
`0.6
`
`0.04
`
`2
`s
`s
`
`0.03
`
`>30
`
`29(2)
`
`14(60)
`
`19(2)
`
`0(0.2)
`
`29(20)
`
`21(40)
`
`2(20)
`
`8(1)
`
`2(100)
`
`>100
`I
`>100
`633268
`2
`>100
`>100
`633270
`I
`>100
`>100
`633272
`I
`>100
`>100
`633274
`" Number of seeds which identified each agent as a potential antitubulin compound. The tenth probe with halichondrin B is included in this tabulation.
`6 Tubulin polymerization reactions: 1.0 mg/ml (10 µM) tubulin; IM monosodium glutamate (pH 6.6 with HO); drugs with 4% (v/v) dimethyl sulfoxide; MgCl2 as
`indicated; preincubation for I 5 min at the indicated temperature; 0.4 mM GTP added; and 20 min incubation at the indicated temperature. lC50 value represents graphical
`determination of drug concentration which inhibits extent of polymerization by SO% (at least three independent determinations). Commercial monosodium glutamate was
`used in the 37"C/I mM Mg2+ system; repurified glutamate was used in the 30"C/0.2S mM Mg2+ system.
`< IC50 values were determined as described in the text, using multiple drug concentrations. The mitotic index for each drug was determined at a single concentration,
`as indicated for each agent in parentheses following the percentage of mitotic cells observed (see text for experimental detail).
`d SDs presented for data from three independent determinations.
`r Homohalichondrin B; data from Bai et al. ( 11 ).
`f Supplies of homohalichondrin B were inadequate for an accurate determination of its IC 50 value ( 11 ).
`• Cytotoxicity data for homohalichondrin B and halichondrin B obtained with LI 210 murine leukemia cells ( 11 ).
`h Halichondrin B; data from Bai et al. ( 11 ).
`1 Proprietary compounds, the structures of which we are unable to reveal at the present time.
`
`3(2)
`
`0.3
`
`In at least one of these two assays we were able to demon- A-4, 1.5 µM for vinblastine, 3.5 µM for maytansine, and 6.8 µM
`strate significant inhibition of tubulin polymerization for 23 of
`for rhizoxin. In the present studies a value of 2.6 µM was ob(cid:173)
`the 32 novel compounds the COMPARE algorithm had indi-
`tained for podophyllotoxin. In the sensitive assay a concurrent
`cated were antitubulin agents. With 20 compounds, IC50 values value of 0.86 µM was obtained for podophyllotoxin, and previ(cid:173)
`substoichiometric to the tubulin concentration (10 µM) were ously values of0.64 and I.I µM were obtained for thiocolchicine
`obtained. For comparison with the agents summarized in Table
`and combretastatin A-4, respectively.
`I, in previous studies (8, 19, 20) we have obtained IC50 values
`As in our previous study (8), IC50 values in the sensitive assay
`in the stand~d assay of 1.2 µM for dolastatin 10, 1.4 µM for were always significantly lower than those obtained in the stan(cid:173)
`thiocolchicine, 2.4 µM for colchicine, 2.5 µM for combretastatin dard assay, when values could be obtained in both systems.
`3895
`
`

`

`COMPUTER-ASSISTED EVALUATION OF CYTOTOXICITY DATA
`
`Fig. 2. Structurally novel compounds se(cid:173)
`lected by the COMPARE algorithm with an(cid:173)
`timitotic agents as seeds: the most potent in(cid:173)
`hibiton of tubulin polymerization.
`
`NSC 624285: R=N02
`NSC624544: R=CF3
`NSC 624545: R=CH3
`
`NSC624546:R1=N02, R2=CH2CH3
`NSC624547:R 1=H, R2=CH3
`
`NSC6354n: R1=R2=H
`NSC635478: R1 =H, R2=0CH3
`NSC635479: R1=CH3, R2=H
`
`Furthermore, there was generally a greater relative drop for less
`active agents. Thus, there was about a 3-fold reduction in the
`IC50 values for NSC 624285 (from 2.5 to 0.94 µM) and NSC
`635477 (from 3.3 to 1.0 µM), but a 6-8-fold reduction for NSC
`631583 (from 51 to 5.8 µM) and NSC 626391 (from 90 to 15
`µM).6
`Nine compounds which the COMPARE algorithm predicted
`would be antimitotic agents, however, did not affect tubulin
`polymerization in vitro. Moreover, these nine compounds also
`had no significant effect on polymerization dependent on mi(cid:173)
`crotubule-associated proteins.
`Since the COMPARE algorithm uses data generated by the
`evaluation of drug effects on cell growth, as opposed to a cell(cid:173)
`free biochemical assay, we evaluated the antimitotic effects of
`both the inactive compounds and those that inhibit tubulin
`polymerization (Table I).
`First, IC50 values were determined for effects on the growth
`of HL-60 cells in culture ( cell number was measured as opposed
`to protein in the original screening studies). The IC5o values for
`most compounds fell within the expected range (0.01-1.0 µM).
`In six cases (see Table 1) the IC50 value we obtained fell in the
`range of 2-7 µM, representing both compounds which inhibited
`tubulin polymerization and compounds which were inactive as
`inhibitors. One agent (NSC 628301), which did not inhibit
`polymerization, was noncytotoxic in our hands, as opposed to
`the findings in screening studies. The reason for the discrepancy
`is unknown, but it may indicate that the compound is chemi(cid:173)
`cally unstable.
`Next we evaluated each drug at one or more concentrations,
`generally at a concentration 3-10 times its IC50 value, for in(cid:173)
`hibition of mitosis (a mitotic cell defined morphologically as a
`cell with condensed chromosomes and no nuclear membrane).
`All agents examined which inhibited tubulin polymerization
`with IC50 values substoichiometric to the tubulin concentration
`(i.e., less than 10 µM) caused a significant rise in the mitotic
`index (range, 14-32%). Furthermore, two of three superstoichi(cid:173)
`ometric inhibitors also caused a significant increase in the mi(cid:173)
`totic index. The only inhibitor of tubulin polymerization which
`
`failed to increase the mitotic index was NSC 625538. It is
`perhaps of interest that this compound was selected by only a
`single seed.
`In addition, two compounds which failed to inhibit tubulin
`polymerization, NSC 83265 and NSC 622093, caused the ap(cid:173)
`pearance of substantial numbers of cells arrested in mitosis.
`Both of these agents were selected by eight seeds. A borderline
`increase in mitotic cells was also observed with NSC 376265.
`The remaining agents which failed to inhibit tubulin poly(cid:173)
`merization, although cytotoxic (with the exception of NSC
`628301, see above), did not cause a significant change in the
`mitotic index. They thus seem to represent false positive pre(cid:173)
`dictions of the COMPARE algorithm.
`Evaluation of the Novel Chemical Structures for Effects on
`Tubulin-dependent GTP Hydrolysis. We have previously pro(cid:173)
`posed (21) that tubulin-dependent GTP hydrolysis (in 1 M
`glutamate) could be used as a simple in vitro screen to identify
`compounds which interact with tubulin, since antimitotic
`agents all seem either to stimulate or to inhibit this reaction (22,
`23). The COMPARE algorithm has permitted us to identify a
`new group of antimitotic agents, and it was of interest to de(cid:173)
`termine whether the GTP hydrolysis assay would have also
`permitted the identification of these compounds. In addition,
`the GTPase assay is technically more suitable for large-scale
`biochemical screening than the tubulin polymerization assay. It
`therefore might prove useful for indicating which compounds
`selected by the COMPARE algorithm represent false positives.
`Accordingly, representatives of each class of new agent, both
`active and inactive, were examined for effects on tubulin-depen(cid:173)
`dent GTP hydrolysis in 1.0 M glutamate (Table 2). [Halichon(cid:173)
`drin 8 (11) has previously been shown to inhibit the reaction.]
`If we use the criterion suggested previously (21) (that the hy(cid:173)
`drolytic reaction in the presence of the test compound deviating
`by more than 15% from the control reaction is to be considered
`positive), then all the substoichiometric polymerization inhibi(cid:173)
`tors were positive. A borderline result was obtained only with
`NSC 83292, while unequivocal stimulation occurred with the
`others. Stimulation of GTP hydrolysis occurred as well with
`NSC 626391 and NSC 625538, two of the three superstoichi(cid:173)
`ometric inhibitors, while the other agent in this group (NSC
`' For NSC 625538, however, the reduction was about 3-fold (from 53 to 16 µM).
`3896
`
`

`

`COMPUTER-ASSISTED EVALUATION OF CYTOTOXICITY DATA
`
`NSC 56030
`
`NSC 83292
`
`NSC 619859
`
`Cl
`
`0
`
`CH3
`~c....-
`1 o-
`
`0
`
`NSC 622691
`
`NSC 625538
`
`NSC 626391
`
`NSC 631583
`NSC 630032
`NSC 627777
`Fig. 3. Structurally novel compounds selected by the COMPARE algorithm with antimitotic agents as seeds: inhibitors of tubulin polymerization.
`
`627777) did not have a significant effect on tubulin-dependent
`GTP hydrolysis. None of the compounds which failed to inhibit
`tubulin polymerization affected the GTPase reaction. This in(cid:173)
`cluded the two compounds which caused an increase in mitotic
`cells without affecting polymerization (NSC 83265 and NSC
`622093).
`These data suggest that the GTPase assay should be useful in
`distinguishing compounds likely to interact with tubulin from
`those which will not, despite their having similar differential
`cytotoxicity patterns (i.e., distinguishing true positives selected
`by the COMPARE algorithm from false positives).
`
`Evaluadon of the Novel Structures for Effects on the Bindin1
`of Radiolabeled Vinblasdne and Colchicine to Tabulin. Aside
`from halichondrin B ( 11 ), and presumably homohalichondrin
`B, none of the agents identified by the COMPARE algorithm
`inhibited the binding ofradiolabeled vinblastine to tubulin (data
`not presented). Conversely, halichondrin B did not inhibit the
`binding of radiolabeled colchicine to tubulin (11), but all the
`other inhibitors of tubulin polymerization identified by the
`COMPARE algorithm did inhibit this reaction (Table 2).
`We examined all compounds as potential inhibitors of colch(cid:173)
`icine binding in 1.0 M glutamate. Those which failed to inhibit
`
`Fig. 4. Structurally novel compounds se(cid:173)
`lected by the COMPARE algorithm with an(cid:173)
`timitotic agents as seeds: inhibitors of mitosis
`which do not interact with tubulin in vitro.
`
`CH3QY)_,,
`
`0
`
`CH30~
`
`0CH3
`
`NSC 622093
`
`NSC 83265
`3897
`
`

`

`COMPUTER-ASSISTED EVALUATION OF CYTOTOXICITY DATA
`
`Structure
`Unknown
`
`0
`
`0
`
`NSC 76455
`
`NSC 628301
`
`Fig. 5. Structurally novel compounds selected
`by the COMPARE algorithm with antimitotic
`agents as seeds: compounds which inhibit nei(cid:173)
`ther mitosis nor tubulin polymerization.
`
`S 0
`0-C=N-NH-C-N
`
`f
`
`CH3
`~ I
`N
`
`II / \
`N
`\__/
`
`NSC 376265
`
`NSC 633268
`NSC 633270
`NSC 633272
`NSC 633274
`
`y
`
`z
`X
`C-H C-H
`N
`C-H C-CH3 N
`N
`C-H C-H
`C-H
`C-H
`N
`
`Table 2 Ef/«ts of non/ chemical structures predicted to be antit11b11/in agents by
`the COMPARE algorithm on t11b11/in-dependent GTP hydrolysis and the binding
`of colchicine to tllbulin
`
`merization in glutamate inhibited colchicine binding
`in
`glutamate. When the reaction condition was changed to 0.1 M
`4-morpholineethanesulfonate, some inhibition of colchicine
`binding was observed with the remaining inhibitors of polymer(cid:173)
`ization (NSC 619859, NSC 627777, and NSC 625538), as well
`as with two compounds which did not inhibit tubulin polymer(cid:173)
`ization (NSC 376265 and NSC 633270).