`
`Research Overview
`
`Some Practical Considerations and Applications of the
`National Cancer Institute In Vitro Anticancer Drug
`Discovery Screen
`Michael R. Boyd and Kenneth D. Paull
`Laboratory of Drug Discovery Research and Development, Developmental Therapeutics Program,
`Division of Cancer Treatment, National Cancer Institute-Frederick Cancer Research and Development
`Center, Frederick (M.R.B.), and Information Technology Branch, Developmental Therapeutics Program,
`Division of Cancer Treatment, National Cancer Institute, Bethesda (K.D.P.), Maryland
`
`Strategy, Management and Health Policy
`
`Venture Capital
`Enabling
`Technology
`
`Preclinical Development
`Toxicology, Formulation
`Drug Drdivery,
`Pharmacoki netics
`
`Clinical Development
`Phases 1-111
`Regulatory, Quality,
`M,rnufacturing
`
`Postmarketing
`Phase IV
`
`ABSTRACT
`During 1985-1990 the U.S. National Cancer Institute (NCI) phased out its murine
`leukemia P388 anticancer drug screening program and developed as the replacement a new in vitro
`primary screen based upon a diverse panel of human tumor cell lines. For each substance tested,
`the screen generates a remarkably reproducible and characteristic profile of differential in vitro
`cellular sensitivity, or lack thereof, across the 60 different cell lines comprising the panel. Several
`investigational approaches to display, analysis, and interpretation of such profiles and databases,
`derived from the testing of tens of thousands of substances during the past 4-5 years since the NCI
`screen became fully operational, have been explored. A variety of useful, practical applications of
`the in vitro screen have become apparent. As these applications continue to evolve, they are
`proving to be complementary to diverse other anticancer screening and drug discovery strategies
`being developed or pursued elsewhere. Reviewed herein are some practical considerations and
`selected specific examples, particularly illustrating research applications of the NCI screen that may
`be more broadly applicable to the search for new anticancer drug development leads with novel
`profiles of antitumor activity and/or mechanisms of action.
`© 199.'i Wiley-Liss, Inc.•
`
`Key Words: antineoplastics, cancer, drug development
`
`INTRODUCTION
`In simplest terms, the NCI in vitro primary
`screen consists of a panel of 60 different human tumor
`cell lines against which compounds arc tested over a
`defined range of concentrations to determine the rel(cid:173)
`ative degree of growth inhibition or cytotoxicity
`against each cell line. The design and operation of the
`screen is such that for each compound tested, both
`the absolute and relative sensitivities of individual cell
`lines comprising the screen are sufficiently reproduc(cid:173)
`ible that a characteristic profile or "fingerprint" of cel(cid:173)
`lular response is generated. Depending upon the ex-
`
`•This article is a US Government work
`© 1995 Wiley-Liss, Inc.
`and, as such, is in the public domain in the United States of America.
`
`tent of differential cellular response, the profile may
`contain much information which is useful for further
`research. The least interesting or usefol (and expect(cid:173)
`edly most common) response to a random selection of
`chemical structures is none at all; that is, none of the
`cell lines show any evidence of growth inhibition or
`
`Address reprint requests to Michael R. Boyd, Laboratory of
`Drug Discovery Research and Development, National Cancer In(cid:173)
`stitute-Frederick Cancer Research and Development Center, Bldg.
`10S2, Rm. 121, Frederick, MD 21702-1201.
`
`
`
`92
`
`BOYD AND PAULL
`
`cytotoxicity. A similarly featureless profile may be ob(cid:173)
`tained when one or more concentrations of the tested
`compound produce(s) growth inhibition and/or cyto(cid:173)
`toxicity of essentially the same magnitude across the
`entire panel of cell lines. Certainly, the NCI screen is
`capable of identifying highly potent, indiscriminant
`direct cell poisons; however, that is not a unique or
`particularly useful attribute of the screen.
`In contrast, the cell lines comprising the NCI
`panel may show differential sensitivity to a given test
`substance. The degree of differential response be(cid:173)
`tween the most and least sensitive lines typically may
`be relatively small (e.g., 2- to 10-fold), or occasionally
`as great as 3-4 orders of magnitude. Furthermore,
`the cell lines may be widely heterogeneous in re(cid:173)
`sponse to a given compound, or they may be compar(cid:173)
`atively homogeneous, with only a relatively few lines
`showing much greater or lesser sensitivity than aver(cid:173)
`age. Regardless of the magnitude of the differential or
`the degree of heterogeneity of response of the cell
`line panel, it is the reproducibility of the response
`fingerprint that is key to the useful information con(cid:173)
`tained therein. This valuable information can be ex(cid:173)
`ploited productively in its own right, as well as in
`complement to other drug discovery research models
`and strategies.
`Routine operation of the NCI in vitro screen
`began in 1990, after 5 years of extensive development
`and pilot evaluations during 1985-89. Reviews of the
`concept, rationale, and technical aspects of develop(cid:173)
`ment of the screen are available elsewhere l c. g., see
`Boyd, 1986, 1989, 1993; Boyd et al., 1992]. From
`1990 to the present, more than 30,000 compounds,
`submitted by cancer researchers worldwide, have
`been tested in the NCI screen. Screening databases
`derived therefrom have provided NCI staff and col(cid:173)
`laborators a unique opportunity to explore a consid(cid:173)
`erable variety of data analysis strategies and methods.
`Reviews and other publications describing such stud(cid:173)
`ies are available [e.g., see Paull et al., 1989, 1995;
`Boyd et al., 1992; Hodes et al., 1992; Weinstein et al.,
`1992, 1994; van Osdol et al., 1994].
`In many if not most of the important respects
`thus far examined, the results and conclusions from
`diverse analytical approaches have been convergent.
`Increasingly sophisticated mathematical and compu(cid:173)
`tational techniques are being developed and applied
`further, and undoubtedly these will add important
`new dimensions to the valuable information that can
`be derived from the in vitro screening panel. This
`seems particularly certain when the data from parallel
`ongoing efforts to further characterize the unique bi(cid:173)
`ology of the individual cell lines can be further inte(cid:173)
`grated into the analyses.
`
`The purpose of this brief review is to offer some
`practical considerations and to describe and illustrate
`some relatively simple and straightforward research
`applications that may be of immediate and consider(cid:173)
`able utility to many current and future users of the
`NCI service screen. In so doing, some selected ex(cid:173)
`amples are drawn from the authors' particular re(cid:173)
`search experiences using the screen. The review is by
`no means intended to be comprehensive; the scope is
`limited to some generally useful applications that can
`be pursued by "nonexpcrts" using relatively simple
`analytical techniques with data generated and sup(cid:173)
`plied routinely for pure compounds submitted for
`testing in the NCI screen.
`
`THE SCREEN
`Detailed descriptions of the screening assay in
`use as of 1990 are available elsewhere [Boyd, 1989;
`Monks et al., 1991; Skehan ct al., 1990]. Some
`changes in the screen subsequently have been made,
`particularly in late 1992. These are noted briefly be(cid:173)
`low. Investigators evaluating recent data in compari(cid:173)
`son with older data from the NCI in vitro screen may
`wish to take these differences into account.
`Cell Line Panel
`The identities, sources, derivation, morpholog(cid:173)
`ical and immunocytochemical characteristics, and
`methods of maintenance of the cell lines comprising
`the NCI 60 cell line panel as of 1990 have been de(cid:173)
`scribed in detail elsewhere [Boyd, 1989; Monks et al.,
`1991; Stinson et al., 1992]. On December 1, 1992, ten
`of the original cell lines were deleted from the panel
`to make way for ten breast cancer and prostate cell
`lines. The lines removed from the panel comprised:
`[lung]
`IIOP-18, LXF-529L, DMS114, DMS273;
`lbrain] XF-498, SNB-78; [colon] KM20-L2, DLD-1;
`[renal] RXF-631L; [melanoma] M19MEL. The lines
`added (and references to the original sources and/or
`corresponding descriptive publications) are as follows:
`[breast] MCF-7 [Soule et al., 1973], MCF-7ADR [Co(cid:173)
`hen et al., 1986], HS578T [Hackett et al., 1977],
`MDA-MB-231 [Cailleau ct al., 1974; Siciliano et al.,
`1979], MDA-MB-43.'5 [Cailleau et al., 1978; Brinkley
`et al., 1980], MDA-N (Steeg, NCI, unpublished
`data), BT-,549 (American Tissue Culture Collection,
`Rockville, MD, unpublished data), T47-D [Keydar et
`al., 1979J; [prostate] DU-145 [Stone et al., 1978;
`Mickey et al., 1977], PC-3 [Ohnuki et al., 1980;
`Kaighn et al., 1979, 1981].
`Screening Assay
`In routine screening, each agent is tested over a
`broad concentration range against every cell line in
`
`
`
`NCI IN VITRO SCREEN
`
`93
`
`the panel. All lines are inoculated onto a series of
`standard 96-well microtitre plates on day zero, fol(cid:173)
`lowed by a 24 h incubation in the absence of the test
`compound. The inoculation densities employed de(cid:173)
`pend upon the particular cell line and its growth char(cid:173)
`acteristics. Inoculation densities used currently in the
`screen for many of the cell lines are the same as orig(cid:173)
`inally published [Monks et al., 1991]. Exceptions, in(cid:173)
`troduced as of December 1, 1992, are as follows (cur(cid:173)
`rent densities used [cells/well] are
`indicated in
`parentheses): HOP-62 (10,000), UO-31
`(15,000),
`786-0 (5,000), LOX IMVI (7,500), SR (20,000). Inoc(cid:173)
`ulation densities used for the breast and prostate lines
`beginning with their addition on December 1, 1992,
`are as follows: MCF-7 (5,000), MCF-7ADR (1.5,000),
`HS578T (20,000), MDA-MB-231 (20,000), MDA-MB-
`43,5 (1.5,000), MDA-1\" (1.5,000), BT-,549 (20,000),
`T47-D (20,000), DU-145 (15,000), PC-3 (7,500). Fur(cid:173)
`ther exceptions introduced as of July 2,5, 1994, are:
`NCI-H226
`(15,000), RXF-393
`(15,000), ACHN
`(10,000), PC-3 (7,500). Test compounds arc routinely
`evaluated at five 10-fold dilutions starting from a high
`of 10-4M, unless otherwise requested. Following a
`48-h incubation with the test compound, the cells are
`assayed by the sulforhodamine B procedure [Skehan
`et al., 1990; Monks et al., 1991; Rubinstein et al.,
`1990]. Optical densities are measured on automated
`plate readers, followed by computerized data acqui(cid:173)
`sition, processing, storage and availability for display,
`and analysis.
`
`DATA DISPLAY AND ANALYSIS
`A detailed description of the contents and format
`of the data report package routinely provided to sub(cid:173)
`mitters of compounds for NCI screening has been
`published [Boyd et al., 1992]. The "dose-response
`matrix" part of the package is no longer provided or
`routinely used. The "dose-response curves" and the
`"mean-graphs" components of the report are the main
`interest of most investigators. Therefore, following
`are some brief descriptions and comments concerning
`the dose-response curves, calculated response param(cid:173)
`eters, and mean-graphs which are most germane to
`the examples to be presented. Also offered are some
`comments and suggestions as to standards for inves(cid:173)
`tigator reporting of NCI screening data in the scien(cid:173)
`tific literature. The consistency, detail, format, and
`placement of such information has increasingly been a
`concern of some journal editors [e.g., see Editors,
`1994].
`
`Dose-Response Curves
`Each successful test of a compound in the full
`screen generates 60 dose-response curves, which are
`
`printed in the 1'CI screening data report as a series of
`composites comprising the tumor-type subpanels,
`plus a composite comprising the entire panel. Data
`for any individual cell line(s) failing quality control
`criteria, or otherwise deflcient for any cell line(s) not
`tested successfully, are eliminated from further anal(cid:173)
`ysis and are deleted from the screening report. Figure
`1 shows contrasting patterns in the dose-response
`curves obtained from two different compounds. The
`figure was prepared directly from the corresponding
`NCI supplier reports by deleting extraneous or oth(cid:173)
`erwise distracting information, and adding minimal
`scaling and reference information for clarity and sub(cid:173)
`stantial photoreduction. The cell line subpanels are
`identified in the figure legend (Fig. 1).
`The "percentage growth" (PG) term, and mean(cid:173)
`ing of the + 50,0 and -50 response reference lines,
`and the calculated response parameters, GI,50, TGI,
`and LC50 have been defined elsewhere [see Boyd ct
`al., 1992; Monks et al., 1991]. Although the response
`parameters are already calculated by computer and
`provided to the investigator in the data report pack(cid:173)
`age, it is important to appreciate how these values are
`determined, and likewise how this may affect data
`interpretation.
`The 50% growth inhibition parameter (GI50) is
`the concentration of test drug where 100 X (T-T0)/
`(C-T0) = 50 = PG. The optical density of the test
`well after the 48-h drug exposure is T; the optical
`density at time zero is T0 ; and, the control optical
`density is C. The PG is a TIC-like parameter that can
`have values from + 100 to -100. Whereas the GI,50
`may be viewed as a growth-inhibitory level of effect,
`the TGI signifies a "total growth inhibition" or cyto(cid:173)
`static level of effect. The TGI is the drug concentra(cid:173)
`tion where 100 x (T-T0)/(C-T) = 0 = PG. The LC,50 is
`the lethal concentration, "net cell killing" or cytotox(cid:173)
`icity parameter. It is the concentration where 100 X
`(T-T0)/T0 = -50 = PG. The control optical density
`is not used in the calculation of LC,50 .
`The GI,50, TGI, and LC 50 values are calculated
`by interpolation using the tested concentrations that
`give PG values above and below the respective refer(cid:173)
`ence values (e.g., ,50 for GI50). Therefore, a "real''
`value for any of the three response parameters is oh(cid:173)
`tained only if at least one of the tested drug concen(cid:173)
`trations falls above, and likewise at least one falls be(cid:173)
`low, the respective appropriate PG reference value
`(i.e., the dose-response curve for that particular cell
`line must cross the respective PG reference line). If,
`however, for a given cell line all of the tested concen(cid:173)
`trations produce PCs exceeding the respective refer(cid:173)
`ence level of effect (PG value of + 50,0 or -,50 as
`appropriate), then the lowest tested concentration
`
`
`
`94
`
`BOYD AND PAULL
`
`\
`
`' '
`
`C1
`
`100.~7-
`--~·~,
`50
`\',
`0
`-50
`
`A1
`
`c,:,
`
`CD
`
`100
`50
`0
`-50
`-100
`
`··--·- ... \~:~::-,
`--><"'i'-'\
`','.I.:-',
`~-,:--,:\,,"~.,
`', .. __
`-,, '
`
`D1
`
`- -100
`D.. -.c
`-
`ii
`e
`c,:,
`-C
`a,
`C,
`ca
`CJ ... a,
`a.
`
`100
`50
`0
`-50
`-100
`8
`
`G1
`
`A.
`
`7
`
`6
`
`5
`
`4
`
`3
`
`8
`
`3
`4
`5
`6
`7
`( - Log10 Molar)
`
`8
`
`7
`
`6
`
`5
`
`4
`
`3
`
`100 -1 - - - - -~ - -~
`50 f---------c-'...,-,
`Of-----
`A2
`-50 - - - - - - - - - - ,
`-100-~-~-~-~~
`
`a D.. -
`.c - 100 r-----.....==""""r----,
`3 e c,:,
`C, ca -C
`a, e a,
`
`so--------~-,
`0 f-----------'1.....-t
`D2
`- 50 - - - - - - - - - -
`- 100 ' - -~ -~ -~~ - -1
`
`100,----====:::s-ar-7
`50 f - - - - - - - - -~
`o---------
`-so~G_2 ______ -'o
`-100 '--~~--~ ~---
`9
`8
`7
`6
`5
`4
`
`a,
`
`D..
`
`8.
`
`B2
`
`=-==
`
`-• - -.... -i:.. ,,__ --ii
`
`. - ,;.i,_,
`-._ '\\;• •.
`
`\\
`H2
`1 - - - - - - - - - - - - ' i - ,
`\
`4
`5
`6
`7
`8
`( - Log10 Molar)
`
`9
`
`C2
`
`F2
`
`12
`
`9
`
`8
`
`7
`
`6
`
`5
`
`4
`
`Figure 1. The top composite (A) of nine sets of dose-response
`curves is from the testing of halomon (structure 1 of Fig. 7) in the
`NCI in vitro screen. The bottom composite (B) of nine sets of
`dose-response curves is from the testing of a related natural prod(cid:173)
`uct (structure 5 of Fig. 7) in the NCI in vitro screen. Individual cell
`line identifiers have been omitted for clarity. Graphs A1 and A2 are
`from the leukemia/lymphoma subpanel, graphs B1 and B2 are from
`the non-small-cell lung cancer subpanel, graphs C1 and C2 are
`from the small-cell lung cancer subpanel, graphs D1 and D2 are
`
`from the colon cancer subpanel, graphs E1 and E2 are from the
`brain tumor subpanel, graphs F1 and F2 are from the melanoma
`subpanel, graphs G1 and G2 are from the ovarian cancer subpanel,
`graphs H1 and H2 are from the renal cancer subpanel, and graphs
`11 and 12 are composites of all the respective subpanels together.
`Reprinted with permission from the American Chemical Society
`from Fuller et al. [1992]. Copyright 1992 American Chemical
`Society.
`
`
`
`NCI IN VITRO SCREEN
`
`95
`
`(specified in negative log10 units) is assigned as the
`default value. In the screening data report, that de(cid:173)
`fault value is preceded by a"<" sign, signifying that
`the "real" value is something "less than" the lowest
`tested concentration. Likewise, if none of the tested
`concentrations produces the required PG reference
`level of effect or greater, then a">" sign precedes the
`printed default value (which is the highest tested con(cid:173)
`centration or HICONC, specified in negative log10
`units), signifying that the "real" value is something
`"greater than" the HICONC. in any case, either the
`"real" (interpolated) or the default ( < or >) Cl50,
`TGI, and LC,50 for every cell line in the panel are
`printed with the mean-graphs included in the screen(cid:173)
`ing data report. The investigator can, if desired, verify
`visually that for any printed response parameter con(cid:173)
`centration preceded by a"<" or">" for a given cell
`line in the CI,50, TCI, or LC150 mean graphs, there
`must be a corresponding dose-response curve that ei(cid:173)
`ther lies entirely below or entirely above the corre(cid:173)
`sponding PC reference line, respectively.
`For some applications, the occurrence of many
`default values for the response parameters in a given
`screening test can have a major impact on both the
`accuracy and the interpretation, and therefore the use(cid:173)
`fulness of the data. This problem may be particularly
`prominent, for example, in structure-activity studies
`where both quantitative (e.g., overall or panel-aver(cid:173)
`aged potency) and qualitative (e.g., profile of differ(cid:173)
`ential cytotoxicity) comparisons of compounds are de(cid:173)
`sired. For any given compound, the particular range
`of concentrations tested can be the major determinant
`of the extent of occurrence of"<" or ">" response
`parameter values. Therefore, it may be necessary to
`obtain further testing of a compound in concentration
`rangc(s) other than employed routinely in the screen,
`depending upon the intended use of the data. Indeed,
`in certain instances, data from the testing of a given
`compound in different concentration ranges may yield
`distinctly useful, non-overlapping information. Exam(cid:173)
`ples that follow may provide further clarification of
`these points. Before presentation of specific examples,
`however, some additional background and descriptive
`information concerning the "mean-graph" and the
`COMPARE analysis concepts are pertinent.
`
`Mean-Graph
`A mean-graph is a pattern created by plotting
`positive and negative values, termed "deltas," gener(cid:173)
`ated from a set of GI150, TGI, or LC,,;0 concent;ations
`obtained for a given compound tested against each cell
`line in the I\CI in vitro screen. Figure 2 shows the
`Cl50 , TGI, and LC50 mean-graphs derived from the
`dose-response data of Figure I. This figure was also
`
`prepared directly from the NCI supplier report, by
`manually cropping and editing the original mean
`graphs.
`The deltas are generated from the GI50, TGI, or
`LC 50 data by a three-step calculation. For example,
`the GI50 value for each cell line successfully tested
`with a given compound is converted to its log10 GI50
`value. The mean panel log10 GI50 value is obtained by
`averaging the individual log10 GI50 values. Both "real"
`and default values are used in the calculation. Each
`individual log10 GI50 value then is subtracted from the
`panel mean to create the corresponding delta.
`To construct the mean-graph, the deltas arc plot(cid:173)
`ted horizontally in reference to a vertical line that
`represents the calculated mean panel Gl150• The mean
`panel GI50 may or may not represent, nor even ap(cid:173)
`proximate, a "true" mean, depending upon the extent
`to which defaults were among the values averaged (see
`Dose-Response Curves). In any case, the negative del(cid:173)
`tas arc plotted to the right of the mean reference line,
`thereby proportionately representing cell lines more
`sensitive than the calculated average. Conversely, the
`positive deltas are plotted to the left of the reference
`line to represent the less sensitive cell lines to the
`given agent. Thus, for example, a bar projecting 3 units
`to the right of the vertical reference line in a GI50
`mean-graph indicates that the GI50 concentration for
`that cell line is 1,000 times less than the panel-aver(cid:173)
`aged GI,50 concentration. The TGI and LC50 mean(cid:173)
`graphs are prepared and interpreted similarly.
`In the full standard NCI screening data report
`package, three additional numbers are printed at the
`base of each of the three respective mean-graphs pro(cid:173)
`vided. These numbers are the MG-MID, the Delta
`(not be confused with the "delta" for an individual cell
`line), and the Range. The MG-MID is the calculated
`mean panel GI50, TGI, or LC50. The Delta is the
`number of log10 units by which the delta of the most
`sensitive line(s) of the panel differs from the corre(cid:173)
`sponding MG-MID. Similarly, the Range is the num(cid:173)
`ber of log 10 units by which the delta of the most sen(cid:173)
`sitive line(s) of the panel differs from the delta of the
`least sensitive line(s). The MG-MID, Delta, and
`Range are most meaningful when few if any default
`values are contained in the corresponding mean(cid:173)
`graph; otherwise, they are not particularly meaningful
`or useful, and indeed can be misleading. Further clar(cid:173)
`ification of this point follows in a discussion of data
`presented in Figures 3, 4, 5A and B.
`
`COMPARE
`COMPARE is a computerized, pattern-recogni(cid:173)
`tion algorithm which has considerable utility in the
`evaluation and exploitation of data generated by the
`
`
`
`96
`
`BOYD AND PAULL
`
`J
`
`A1
`
`B1
`
`C1
`
`D1
`
`E1
`
`F1
`
`G1
`
`H1
`
`A.
`
`3
`
`Log10 TGI
`+ 3;,-------'____..___.___.___._ ___ 3 I +3
`
`-3
`
`-3
`
`A2
`
`B2
`
`C2
`
`D2
`
`E2
`
`F2
`
`G2
`
`H2
`
`B.
`
`+3
`
`-3
`
`+3
`
`-3
`
`+3
`
`-3
`
`.. Log10 TGI ................................ Log10 LC50 ..
`
`Figure 2. The top composite (A) shows the Gl 50 , TGI, and LC50
`mean-graphs derived from the dose-response data of Figure 1A;
`the bottom composite (B) shows the corresponding mean-graphs
`
`derived from the dose-response data of Figure 1 B. Individual cell
`line identifiers have been omitted for clarity. The subpanel iden(cid:173)
`tifiers are the same as for Figure 1A, B.
`
`NCI screen. In essence, COYIPARE is a method of
`determining and expressing the degree of similarity,
`or lack thereof, of mean-graph profiles generated on
`
`the same or different compounds. An early impetus
`for the creation of such a tool during the development
`phase of the screen was the need to standardize the
`
`
`
`NCI IN VITRO SCREEN
`
`97
`
`Loa GJSO
`
`GISO
`
`Loa TG[
`
`TGI
`
`Lcukcm&a
`CCRf.Q:M
`lll.-60(TB)
`K-562
`MOLT-4
`RPMI-8226
`SR
`No11-Small Cell I.,1.1111 Cuc:er
`AS49/ATCC
`EKVX
`HOP-18
`HOP-62
`HOP-92
`N(cid:143)
`-II226
`N(cid:143)
`-H23
`N(cid:143)
`-H322M
`N(cid:143)
`-H460
`N(cid:143)
`-1-1522
`I.XFl.529
`SmallCellL.rn1C&nc.er
`DMSll4
`DMS27J
`Colotl Ca.nceJ
`COLO :WS
`DLD-1
`HCC-2998
`HCT-ll6
`HCT-15
`HT29
`KM12
`KM20L2
`SW-620
`CNSC.a.noer
`SJl-l68
`SF-295
`SF-539
`SNB-19
`SNB-75
`SNB-78
`U25I
`Xf'498
`MclilDoma
`l.DXIMVI
`MALME-JM
`M14
`Ml9-MEL
`SK-MEL-2
`SK-MEL-28
`SK-MEL-S
`UACC2S7
`UACC-62.
`
`IGROVI
`OVCAR-l
`OVCAR-4
`OVCAR-S
`OVCAR-8
`SK-OV-3
`Ren.J Cancer
`786.0
`A49II
`A(cid:143)
`IN
`CAKT-l
`RXF-393
`RXF-631
`Sh'J2C
`TK-10
`llO-31
`
`MG_MID
`Dd•
`Range
`
`< -1000
`< -1000
`< -1000
`< -1000
`< -1000
`< -1000
`
`< -10,00
`
`"'
`
`<-10.00
`
`< -10.00
`< -10.00
`
`< -10.00
`< -10.00
`< -10.00
`
`< -10.00
`< -10.00
`< -10 00
`< -10.00
`< -1000
`< -ID 00
`< -1000
`< -1000
`< -10.00
`
`< -10.00
`< -10.00
`< 1000
`< -10.00
`< -1000
`-9.71
`< -10.00
`< -10.00
`
`< -10.00
`< -10.00
`< -10.00
`< -10.00
`< -10.00
`< -10.00
`< -10.00
`< -10.00
`< -10.00
`
`<-1000
`< -l0.00
`-9.17
`
`< -10.00
`< -10.00
`
`< -10.00
`< -10.00
`< -10.00
`
`< -10.00
`< -10.00
`< -10.00
`
`9 95
`
`9.91
`
`0.09 - - - - - - ·
`~
`.,
`
`.,
`
`+2
`
`.,
`
`.3
`
`960
`-11.61!
`>
`-6 00
`> -6.00
`> -6 00
`< -10 00
`
`> -600
`> -600
`> -6.00
`-600
`>
`> -6.00
`-929
`< -10.00
`> -6.00
`-8.96
`< -10.00
`.,oo
`
`< -10.00
`< -I0.00
`
`-9.lll
`> -6.00
`<-10.00
`-1!.9&
`-7.13
`<-10.00
`-6.92
`<-10.00
`> -600
`
`6.04
`> -6.00
`< -10.00
`-6.00
`>
`> -6.00
`> -6.00
`< -10.00
`:,,
`-6.00
`
`:,,
`
`-6.00
`-6.00
`
`>
`-6.00
`> 6.00
`> -600
`
`.,.,.
`
`-6.00
`-6.00
`
`>
`:,,
`
`-7.10
`< -10.00
`> -6.00
`> -6.00
`>
`-6.00
`< -10.00
`
`> -6.00
`< -10.00
`> -6.00
`-6.00
`>
`
`-6.16
`> -6.00
`> -6.00
`>
`-6.00
`
`-
`
`-7.38
`
`2.62
`
`.,
`
`+I
`
`+3
`
`.,
`
`.,
`
`.J
`
`Figure 3. Gl 50 and TGI mean-graphs from the initial testing of spongistatin 1 (structure 1 of Fig. 6)
`in the NCI in vitro screen.
`
`screen and to establish and monitor its consistency
`and reproducibility over time. This was, and still is,
`accomplished by the regular testing of standard com(cid:173)
`pounds which are expected to generate the same or
`very similar profiles when screened repetitively
`against the same panel of cell lines.
`Further in the course of standardizing the
`screen, we selected as reference compounds approx(cid:173)
`imately 170 agents for which a considerable amount of
`information was available about their preclinical
`and/or clinical antitumor properties and presumed
`mechanism(s) of action. These compounds included
`commercially marketed anticancer drugs, investiga-
`
`tional anticancer drugs, and other candidate antitu(cid:173)
`mor drugs which \Vere or had been in preclinical de(cid:173)
`velopment at NCI based upon activities in other
`cancer-related test systems. The repetitive periodic
`screening of these prototype "standard agents" has
`continued to the present, and remains the basis for
`calibration and standardization of the screen.
`The standard agent database also is the key to
`many useful research applications of the NCI screen.
`For example, the response profile fingerprint of a se(cid:173)
`lected standard agent may be used as the "seed" to
`probe any other available mean-graph database to sec
`if there are any closely matching profiles contained
`
`
`
`98
`
`BOYD AND PAULL
`
`Pand/Cell Lin,
`
`Log GISO
`
`Log TGI
`
`Lcu.kcmia
`CCRF-CEM
`IIL-60(TB)
`K-562
`MOLT4
`RPMl-8226
`SR
`Non-Small Cell Lung Cancer
`AS49/ATCC
`EKVX
`HOP-62
`HOP-9:2
`KCI-H23
`NCI-Hl22M
`NCI-H522
`ColonC.,ncer
`COLO20S
`HCC-2998
`IICl'-116
`11er.1s
`UT29
`SW-620
`CNS Cancer
`SF-268
`Sl'-295
`SF-539
`SNll-19
`SNH-75
`l1'251
`Melannma
`MAI.ME-3M
`M14
`SK-MEl.-28
`SK-MEL-5
`UACC-257
`UACC-62
`0¥•.nan Cancer
`IGROVI
`OVCAR-3
`OVCAR-4
`OVCAR-8
`SK-OV-3
`Ren11l C1mcu
`78Ml
`M98
`ACHN
`CAKl-1
`SNl2C
`U0-31
`Pror.1.:.cCancu
`l'C-3
`DU-i.:.5
`Urc.astC1r.cer
`MCF7
`MCl-1/ADR-RES
`MDA-~U-231/ATCC
`MDA-\18-435
`MDA-~
`BT-549
`T-47D
`
`MG MID
`Delta
`llangt
`
`-1037
`-10.49
`-10.52
`-10.12
`-10.31
`-1075
`
`-1004
`-9.74
`-!0.02
`> -9.00
`-1000
`-9.55
`-1059
`
`>
`
`-1067
`-990
`-1043
`-959
`-1056
`-1035
`
`-10.18
`-10.60
`-10.46
`-9.17
`-10.67
`-9.87
`
`-10.42
`-1037
`-1013
`-10.60
`-10.06
`-1052
`
`-10.01
`-10 22
`-900
`-971
`-1035
`
`-9.90
`-10.22
`-9.60
`-9"4
`-9.76
`-9.51
`
`.10;0
`-9.81
`
`-l0.23
`-960
`-908
`-1:12
`-1096
`-906
`935
`
`1009
`l 03
`2 12
`
`> -9.00
`-9.86
`-9.53
`> -9.00
`> -9.00
`-10.06
`
`> -9.00
`-9.00
`>
`-902
`> -9.00
`-9.05
`> -9.00
`-10.05
`
`-10_29
`9.20
`-9.70
`-9.00
`-10.16
`> -9.00
`
`>
`
`> ~9.00
`-10.12
`-9.78
`:> -9.00
`-10.18
`> -9.00
`
`:,,
`
`-9.00
`-9.05
`-9.00
`>
`> 9.00
`·9.00
`>
`-9.00
`>
`
`>
`
`-9.00
`-9.48
`> -9.00
`-911
`> -900
`
`-9.03
`-909
`> -9,00
`> -9.00
`> -900
`·900
`>
`
`-954
`-912
`
`>
`>
`>
`
`.900
`-900
`-9.00
`-10.48
`-10.49
`> -9.00
`> -9.00
`
`-9.28
`1.21
`L49
`
`.2
`
`-1
`
`-2
`
`-3
`
`Figure 4. Gl50 and TGI mean-graphs from further testing of spongistatin 1 in different concentra(cid:173)
`tion ranges in the NCI in vitro screen.
`
`therein. Conversely, a profile selected from any avail(cid:173)
`able mean-graph database can be used to probe the
`standard agent database to determine if there are any
`dosely matching standard agent profiles. Databases
`used for such studies may be constructed or defined
`as desired and may be relatively small (e.g., a selected
`congeneric series of compounds) or very large (e.g.,
`the entire database from all pure compounds tested in
`the NCI screen to date).
`Initial NCI studies with COMPARE led quickly
`to the observation that compounds matched by their
`mean-graph patterns often had related chemical
`struchues. Closer examination of this phenomenon
`
`led further to the realization that compounds of either
`related or unrelated structures and matched by mean(cid:173)
`graph patterns frequently shared the same or related
`biochemical mechanisms of action [e.g., see Boyd,
`199:3, and Paull et al., 1995, and references therein].
`Before preceding to more specific examples illustrat(cid:173)
`ing some of the intriguing research applications of this
`phenomenon, further description of the COMPARE
`calculation methodology is in order.
`
`Method of COMPARE Calculations
`COMPARE analyses may be performed using
`the mean-graph deltas calculated from either the
`
`
`
`NCI IN VITRO SCREEN
`
`99
`
`100.-----,----~-----,----.------r----,----,----,
`
`50
`
`-s
`~
`0 ...
`0 ..,
`0 &
`ti u ... ..,
`
`Cl..
`
`-50
`
`A.
`
`100
`
`-9
`
`-7
`-8
`Log JO of Sample Concentration (Molar)
`
`-6
`
`50
`
`-s
`:l'
`0 0 ..,
`0 ~
`El ..,
`~ ..,
`
`Q.,
`
`-50
`
`-lO?l'--4-----_-'-13------'_l'-2-----_.,__ll _ __ ___ _.10 _____ _._9
`
`B.
`
`Log 10 of Sample Concentration (Molar)
`
`Figure 5. The top set (A) of dose-response curves is that from which the mean-graphs of Figure 3
`were derived. The bottom set (B) is that from which the mean-graphs of Figure 4 were derived.
`
`GI50s, the TGis, or the LC50s. When a selected par(cid:173)
`ticular mean-graph profile or "seed" is used to probe
`a given database, the appropriate delta value for each
`cell line is compared to the corresponding delta value
`for the same cell line for every mean-graph entry in
`the specified database set. If either delta value is
`missing for any cell line (e.g., due to test failure or
`quality control deletion), then that cell line is elimi(cid:173)
`nated entirely from the calculation for that particular
`seed/mean-graph and database/mean-graph pair.
`Thus, for each mean-graph in the specified database a
`set of pairs (maximum of 60) of delta values is ob(cid:173)
`tained. The commercially available SAS statistical
`program is used to calculate a Pearson product mo(cid:173)
`ment correlation coefficient (0. 0-1. 0) for each set of
`
`delta value pairs. Then the mean-graphs of all com(cid:173)
`pounds in the specified database can be rank-ordered
`for similarity to the seed mean-graph.
`
`Impact of Default Values for Response
`Parameter Concentrations
`Default GI.~0 , TGI, or LC50 values (see defined
`above) are included in the mean-graph and COM(cid:173)
`PARE because they represent valued information
`even though the information is less exact than the
`measured values would he if the measured values
`were available. However, this can sometimes lead to
`peculiarities, of which the investigator should be
`aware. For example, in an extreme case where a com(cid:173)
`pound has essentially no effect at the highest tested
`
`
`
`100
`
`BOYD AND PAULL
`
`concentration, and therefore the GI 50s, TGis, and
`LC 50s are all represented by the HICONC default,
`the corresponding mean-graphs appear as straight
`vertical lines, and COMPARE has no meaningful pat(cid:173)
`terns to correlate. In an opposite extreme case, the
`tested compound is sufficiently potent that the lowest
`concentration tested is the default value for all of the
`GI50s, TGis, and LC,50s. In this instance, the corre(cid:173)
`sponding mean-graphs are also straight vertical lines,
`and COMPARE has nothing to meaningfully corre(cid:173)
`late. Between such extremes are examples of mean(cid:17