5893_01_p525-532 11/7/05 3:05 PM Page 525

ASSAY and Drug Development Technologies

Volume 3, Number 5, 2005
© Mary Ann Liebert, Inc.

Assessing the Minimum Number of Data Points Required for

Accurate IC50 Determination

Robert J. Turner and Steven J. Charlton

Abstract: Combinatorial chemistry technology has enabled the synthesis of large, targeted libraries
consisting of a much higher ratio of pharmacologically active compounds than traditional compound
archives. This has resulted in a higher number of compounds requiring follow-up characterisation in
full 50% inhibitory concentration (IC50) curves after the preliminary screen. The aim of this study
was to develop a new assay format and analysis protocol for IC50 determination from the minimum
number of data points so that more information could be derived from a primary inhibition screen.
Data points from existing 10-point IC50 curves were used retrospectively to test the accuracy of IC50
predictions derived from just one or two compound concentrations. Regression analysis showed that
both methods were useful, although, as expected, two compound concentrations gave more accurate
IC50 predictions. A final experimental data set comparing IC50 values derived from a two- and 10-
point assay format gave highly comparable data (r2
0.89). This study shows that it is possible to
generate IC50 values from an appropriately designed primary inhibition screen using two compound
concentrations, reducing the requirement for follow-up IC50 determinations.

Introduction First, unlike traditional HTS compound archives, li-

braries designed during the hit-to-lead and lead optimi-

A DVANCES IN COMBINATORIAL CHEMISTRY techniques

have led to a dramatic increase in the generation of
compound libraries at all stages of the drug discovery pro-
cess, in particular during hit-to-lead and lead optimisation
phases.1,2 In order to generate an SAR for these com-
pounds, accurate IC50 measurements are required from
which affinity constants can be determined (e.g., Ki from
Cheng-Prusoff analysis3). These libraries are typically too
large to test as IC50 curves in the first instance, so screen-
ing commonly follows the same paradigm as HTS, where
compounds are first tested once or twice at a single high
concentration to identify hits based on percent inhibition
of control activity/binding. Hits (e.g., 40% inhibition)
are then confirmed by running up to 10-point IC50 curves.
This has the advantage of only testing active compounds
in the more resource-intensive IC50 format. However,
whilst proving very successful in most cases, this is not
necessarily the most efficient process, for several reasons.
sation phases are often done so using previous knowl-
edge of SAR at the target. As a consequence, they
commonly consist of a much higher ratio of pharmaco-
logically active compounds than standard diversity li-
braries. This can lead to many of the compounds being
identified as active in the primary screen, necessitating
the repeat of many compounds in an IC50 format. Sec-
ond, although many compounds may appear active in the
first screen (at a high compound concentration), the ma-
jority of “hits” are frequently of relatively poor affinity.
This can result in a large number of the compound dilu-
tions in the IC50 curve being made at inactive concen-
trations, representing a considerable waste of expenditure
in terms of resource and time. Finally, data from primary
screens are often reported as percent inhibition of radi-
oligand binding or maximal functional response. This
makes it difficult to directly compare these data with the
subsequent IC50 values generated on a smaller subset.

Novartis Institutes for Biomedical Research, Horsham, UK.

ABBREVIATIONS: CCR3, human CC chemokine receptor 3; cpm, counts per minute; DMEM, Dulbecco’s modified Eagle medium; FCS,
foetal calf serum; RBL cells, rat basophil leukaemia cells; SAR, structure–activity relationship; SPA, scintillation proximity binding assay; WGA,
wheat-germ agglutinin.

525
5893_01_p525-532 11/7/05 3:05 PM Page 526

526 Turner and Charlton

The optimal approach would be to use an analysis method
that is able to generate IC50 values for all compounds in
a library during the initial screen with the minimum num-
ber of data points for each compound.
Methods for deriving IC50 data from a single point
have been described previously, for example, for use in
cytochrome P inhibition screens.4,5 In the present study,
we have evaluated the use of both one and two compound
concentrations to derive IC50 values in a CCR36
chemokine receptor SPA. This was approached in two
steps. Initially, data points (from the compound concen-
trations stated) obtained from 10-point IC50 curves were
selected and analysed in a retrospective manner using
XLfit (ID Business Solutions Ltd., Guildford, UK) to de-
termine the optimal analysis parameters for accurate IC50
determinations from either one or two compound con-
centrations. The predicted optimal analysis parameters
were then tested in a prospective experimental study and
correlated to IC50 values obtained using the traditional
10-point IC50 method.
Materials and Methods
Materials
125I-human eotaxin (25
Ci; specific activity 2,000
Ci/mmol) and WGA-PVT SPA beads were from Amer-
sham Biosciences UK Ltd. (Little Chalfont, UK). Com-
plete™ (protease inhibitor cocktail tablets) was from
Roche Diagnostics Ltd. (Lewes, UK). RBL cells ex-
pressing human CCR3 (RBL-CCR3) were provided by
the Scripps Research Institute (La Jolla, CA). DMEM
(high glucose), cell dissociation buffer (enzyme free,
Hanks based), Hanks’ Balanced Salt Solution, geneticin,
100 mM MEM non-essential amino acids, 1 M HEPES,
200 mM L-glutamine, and FCS were supplied by Invit-
rogen (Paisley, UK). Roller bottles, flasks, sterile tubes,
and cell-scrapers were from Fisher Scientific (Lough-
borough, UK). All other chemicals were of reagent grade
and supplied by Sigma (Poole, UK).
Tris-HCl, pH 7.6, 250 mM sucrose, and 10% glycerol with
1 Complete tablet per 50 ml). A small aliquot was used
to determine the protein concentration using a Bio-Rad®
(Hemel Hempstead, UK) Protein Assay with bovine serum
albumin standards. Membranes were stored at 80°C.
CCR3 SPA
The assay was performed in a final volume of 250
l per
well in white 96-well Optiplates™ (Perkin Elmer, Seer
Green, UK). Compound dilutions were performed on a
Biomek 2000® (Beckman) in assay buffer (25 mM
HEPES-NaOH, 120 mM NaCl, 1 mM CaCl2, and 5 mM
MgCl2, pH 7.6) with 5% (vol/vol) dimethyl sulfoxide
(1% final concentration) in U-bottomed 96-well plates to
give a series of 10 concentrations (10–0.0005
M final
concentration). Fifty microlitres of test compound was
transferred to the Optiplates using a Tomtec (Hamden,
CT) Quadra®. Total binding was determined using 50
l
of vehicle, and nonspecific binding was determined us-
ing 50
l of control compound. After the addition of test
compound, 150
l of membrane (10
g per well final
amount) and WGA-PVT SPA beads (1 mg/ml final con-
centration) mixture in assay buffer were added to each
well of the Optiplate. Fifty microlitres of the radiolabeled
ligand 125I-human eotaxin was added at to give a final
assay concentration of 50 pM. Following the additions,
the plate was incubated for 4 h at room temperature. The
plate was sealed using TopSeal-S™ according to the
manufacturer’s (Packard, Meriden, CT) instructions, and
the radioactivity in the plate was counted using a Packard
TopCount:NXT™, with each well being counted for 1
min. For the two-point IC50 determination the assay for-
mat consisted of the same 10-point control compound,
nonspecific bindings, and totals, but the test compounds
were plated out at only two concentrations in duplicate.

Cell culture and membrane preparation

RBL-CCR3 cells were cultured in roller bottles at 37°C
in DMEM supplemented with 20% FCS, 10 mM HEPES,
2 mM L-glutamine, and 0.5% geneticin in a 10% CO2 hu-
midified atmosphere. Cells were harvested using cell dis-
sociation buffer and pelleted at 800 g and 4°C for 10 min.
Cells were resuspended in 5 ml of hypotonic buffer (20
mM Tris-HCl, pH 7.6, with 1 Complete tablet per 50
ml) and incubated on ice for 30 min. The cells were then
homogenised using a Teflon Dounce homogeniser, and
the homogenate was further centrifuged (48,000 g, 4°C,
FIG. 1. Graphical representation of the parameters used to
30 min; Avanti®, Beckman, Coulter Limited, High generate 50% inhibitory concentration curves from two com-
Wycombe, UK). The supernatant was discarded, and the pound concentrations.CPM, counts per minute; EC50, 50% ef-
pellet was resuspended in homogenisation buffer (20 mM fective concentration.
5893_01_p525-532 11/7/05 3:06 PM Page 527

IC50 Determination from Primary Screen Data 527

Fitting IC50 curves to one or two
compound concentrations
A standard 10-point IC50 assay was used to generate
data for 510 compounds. Data (in cpm) from either one
or two compound concentrations were then selected from
these 10-point curves and fitted using a standard sigmoidal
curve (Eq. 1) with the top (B) fixed to the total binding
in the absence of competing compound and the bottom of
the curve (A) fixed to the nonspecific binding (all in cpm).
The effect of fixing the slope (D) was also tested. Figure
1 shows a graphical representation of the parameters used:
Y
A  {(B  A)/(1  [(C/X)  D])}
where Y is the amount bound (in cpm) and X is the con-
centration of unlabelled ligand (in log M). The other pa-
rameters are represented as follows: A is the minimum Y
(nonspecific binding, in cpm); B is the maximum Y (spe-
cific binding in the absence of competitor, in cpm); C is
the IC50 (in log M); and D is the slope.
Results
The accuracy of IC50 predictions from a single com-
pound concentration. It was predicted that the key vari-
able in this analysis method, having the greatest influ-
ence on the IC50 determination, would be the concentra-
tion of competitor ligand selected. Compound concen-
trations of 10, 1, and 0.1
M were selected to test the
method as these were the most likely to fall on the slope
of the concentration–response curve in this library. In ad-
dition, the influence of the slope factor was examined by
either fixing it to 1, thus assuming simple competitive
binding to a single site, or by allowing a variable slope
determined as part of the data fitting process. The IC50
predictions from this simulated data set were compared
to those derived from the 10-point IC50 curves in two dif-
ferent ways using GraphPad Prism® (GraphPad Software,
(1)
Inc., San Diego, CA). First, a simple linear regression
analysis showing the best fit of the data was performed,
with the slope factor providing an indication of the de-
parture from unity. Second, the data were fitted to a first-
order polynomial equation with slope and intercept con-
strained to 1 and 0, respectively, to represent the line of
unity. The 95% prediction bands for each data set were
also calculated (Fig. 2). Where r 2 values are quoted in
the text, they indicate the goodness of fit of the data to
the line of unity.
In each case, compounds with IC50 values close to the
concentration tested were well predicted by the single-

FIG. 2. Comparison between 50% inhibitory concentration (IC50) values obtained for 510 compounds using either a full 10-
point IC50 curve or a simulated single-point assay. (A) A one-point model where the slope is not fixed. (B) A one-point model
where the slope was fixed to 1. The single point concentrations selected were: (i) 10
M, (ii) 1
M, and (iii) 0.1
M. A line
of unity is shown with the 95% prediction band calculated for each data set. The r2 value represents goodness of fit of the data
to this line of unity. n signifies the number of successful predictions made in each condition.
5893_01_p525-532 11/7/05 3:06 PM Page 528

528 Turner and Charlton

FIG. 3. Comparison between 50% inhibitory concentration (IC50) values obtained from a 510-compound 10-point screening scin-
tillation proximity binding assay and a theoretical two-point analysis. (A) A two-point model where the slope is not fixed. (B) A
two-point model where the slope was fixed to 1. The two concentrations selected were (i) 10 and 1
M, (ii) 10 and 0.1
M, and
(iii) 1 and 0.1
M. A line of unity is shown with the 95% prediction band calculated for each data set. The r2 value represents
goodness of fit of the data to this line of unity. n signifies the number of successful predictions made in each condition.

point, variable slope method. There was, however, signif- sult was an overall poor correlation of those IC50 values
icant deviation from unity as the affinity of the compounds predicted using a single compound concentration with vari-
increased, with the single-point predictions underestimat- able slope factor, and those obtained using full 10-point
ing the IC50 values for higher-affinity compounds. The re- determinations (see Table 1 for details).

TABLE 1. SUMMARY DATA FROM THE RETROSPECTIVE ANALYSIS OF THE ORIGINAL 10-POINT 50% INHIBITORY CONCENTRATION
(IC50) DATA (N
510), USING EITHER THE ONE-POINT OR THE TWO-POINT METHOD

Variable IC50 slope IC50 slope fixed to 1

Variance Variance
from unity Predictions from unity Predictions
Concentration (
M) (r2) Slope (%) (r2) Slope (%)

One-point
10 0.865 0.504 99.6 0.414 0.670 98.8
1 0.100 0.529 67.6 0.765 0.762 74.1
0.1 1.386 0.373 60.2 0.601 0.657 29.0
Two-point
10 and 1 0.798 0.822 95.5 0.843 0.827 92.5
10 and 0.1 0.741 0.884 89.8 0.878 0.928 85.9
1 and 0.1 0.838 0.852 59.6 0.853 0.835 55.5

r2 signifies the goodness of fit to unity (where data are fitted to a line with the slope constrained to 1.0). Slope signifies the
slope factor where the data are fitted to an un-constrained line.
5893_01_p525-532 11/7/05 3:06 PM Page 529
IC50 Determination from Primary Screen Data
FIG. 4. Comparison of 50% inhibitory concentration curves calculated for two representative compounds (A and B) using ei-
ther 10 (i) or two (ii) concentration points. cpm, counts per minute.
Fixing the slope factor to 1 significantly improved
the correlation, although this method still tended to under-
estimate the IC50 of higher-affinity compounds. The con-
centration of compound used in the single-point analysis
had little impact on the overall quality of predictions, but
had a large influence on the number of predictions made.
The highest concentration tested (10
M) returned results
for 99% of the compounds, whilst it was possible to de-
rived IC50 values for only 29% of compounds when they
were tested at 0.1
M. In order to improve the quality of
predictions, data from two different compound concen-
trations were fitted using the same analysis parameters.
The accuracy of IC50 predictions from two
compound concentrations
For the two-point analysis, three different compound
concentration combinations were tested; (a) 10 and 1
M,
(b) 10 and 0.1
M, and (c) 1 and 0.1
M. The predicted
IC50 values from this analysis were then compared di-
rectly to those derived from the full 10-point curves
(Fig. 3).
The r2 values obtained for each of the two-point linear
regression analyses were higher than those using a single
concentration (Table 1). As observed in the single-point
model, closest correlations were obtained when the slope
factor was fixed to 1. This was most apparent with com-
pound concentrations of 10 and 0.1
M, where the r2 was
0.88 when the slope was fixed to 1, as compared to 0.74
529
with a variable slope. As observed previously, the num-
ber of successful curve fits reduced with the lower com-
pound concentration ranges, whilst the accuracy of IC50
determinations was largely unaffected.
The single-point analysis method has the advantage of
being applicable to current screening strategies without
requiring alterations in assay format or plate layout. It
also enables a retrospective calculation of IC50 values
from existing data sets. The weakness of this method is
the relatively poor IC50 predictions in comparison to
those obtained using two concentrations. When screen-
ing new libraries, it could be argued that accurate IC50
values are more important than obtaining no curve fit, as
those compounds for which predictions were not possi-
ble could be repeated. The two-point method was there-
fore selected to experimentally validate the robustness of
this approach during the screening of a small compound
collection.
Screening a compound collection using the
two-point IC50 method
In order to experimentally validate the two-point
method, a collection of 129 compounds was screened at
10 and 1
M. The IC50 of these compounds was also de-
termined in a standard 10-point IC50 curve. Figure 4
shows examples of data obtained in each method for two
different compounds. In both cases, the predicted curve
fits were very similar, despite the minimal number of data
5893_01_p525-532 11/7/05 3:06 PM Page 530
530
FIG. 5. Correlation of 50% inhibitory concentration (IC50) re-
sults from 129 compounds screened using both the 10-point and
two-point scintillation proximity binding assay formats. A line
of unity is shown with the 95% prediction bands calculated.
The r2 value represents goodness of fit of the data to this line
of unity. n signifies the number of successful predictions made
in each condition.
points. The two-point method successfully predicted IC50
values for 126 compounds, which was 98% of those
tested. The 10-point curve successfully fitted curves for
all 129 compounds. Figure 5 shows the correlation of
IC50 predictions from both the 10- and two-point meth-
ods. Comparison of the two data sets using linear re-
gression gave a slope of 0.93. When the linear regression
slope was fixed to unity, thereby measuring the direct
correlation of the two methods, an r2 value of 0.89 was
obtained. Of the seven compounds that fell outside the
95% direct prediction band, only four had greater than
threefold variation in the IC50 predictions from 10-point
and two-point methods. In addition, only 21 compounds
from the whole data set had greater than twofold varia-
tion between the two assay methods. These results dem-
onstrate that accurate IC50 predictions can be made us-
ing an assay format that generates data from just two
concentration points.
Discussion
One of the most widely used methods for the mea-
surement of a compound’s pharmacological affinity is the
determination of its IC50 at equilibrium. This is com-
monly achieved by constructing full concentration–inhi-
bition curves, but this approach is not practical when
screening large compound libraries. The present study
has demonstrated that accurate IC50 predictions can be
made from just two concentration points and is therefore
possible from primary screening data. This has the dis-
Turner and Charlton
tinct advantage of reducing the number of samples and
therefore the cost of screening such libraries. The method
is easily adaptable to robotics and can be analysed using
widely available non-linear graph drawing packages.
It is apparent that the correct choice of test compound
concentrations is critical to the degree of success of this
method. Although in many early hit expansion libraries
the compound affinity may often fall in the 1–10
M
range, it is clear that lower compound concentrations
should be selected if libraries are designed later in the
drug discovery process, using a good knowledge of the
target’s SAR where predicted affinities are below 1
M.
An important assumption made when fixing the slope
of the IC50 curve to 1 was that the compounds bind re-
versibly to a single site in direct competition with the ra-
dioligand. Unpublished studies have shown this to be the
case for these compounds. Where the mode of action is
less well understood, for example, with a series of novel
scaffolds or in hit-validation experiments, this assumption
may be inappropriate. In such situations, it would be more
suitable to also fit the data to variable slope IC50 curves.
If the slope factor from the variable curve fit was signif-
icantly different from 1 (either higher or lower), then
the mode of action is likely to be complicated, and IC50
predictions should be taken from these fits rather than
from curves with a fixed slope. It is important to note that
in such cases, the IC50 values should not be used to de-
rive Ki values using the Cheng-Prusoff equation as this
derivation assumes the ligands are competitive in nature.7
In addition to the IC50, the slope factor itself may be a
very useful result to report, as it can provide additional
clues on the mechanism of action of each compound.8
To conclude, the increased use of combinatorial chem-
istry in hit-to-lead and lead optimisation phases of drug
discovery projects has led to the generation of more li-
braries with high proportions of pharmacologically ac-
tive compounds. This trend has necessitated the re-eval-
uation of traditional screening approaches. Providing the
assumptions made in this method are understood and ap-
propriate consideration is taken during the assay set-up,
we have demonstrated that accurate IC50 values can be
generated from primary screening data, presenting an ap-
proach to improve screening capacity with little com-
promise in data quality.
Acknowledgments
The authors would like to acknowledge David Beer at
the Novartis Institute for Tropical Diseases, Singapore,
for the original development of the CCR3 SPA used in
this study. Thanks also to Martin Gosling, Novartis In-
stitutes for BioMedical Research, Horsham, UK, for his
helpful comments and suggestions for improving this
manuscript.
5893_01_p525-532 11/7/05 3:06 PM Page 531
IC50 Determination from Primary Screen Data
References
1. Bunin BA, Dener JM, Livingston DA: Application of com-
binatorial and parallel synthesis to medicinal chemistry.
Annu Rep Med Chem 1999;34:267–286.
2. Hunter D: Life in the fast lane: High-throughput chemistry
for lead generation and optimisation. J Cell Biochem Suppl
2001;37:22–27.
3. Cheng YC, Prusoff WH: Relationship between the inhibi-
tion constant (Ki) and the concentration of inhibitor which
causes 50 percent inhibition (I50) of an enzymatic reaction.
Biochem Pharmacol 1973;22:3099–3018.
4. Gao F, Johnson DL, Ekins S, Janiszewski J, Kelly KG,
Meyer RD, West M: Optimizing higher throughput methods
to assess drug-drug interactions for CYP1A2, CYP2c9,
CYP2C19, CYP2D6, rCYP2D6, and CYP3A4 in vitro us-
ing a single point IC50. J Biomol Screen 2002;7:373–382.
5. Moody GC, Griffin SJ, Mather AN, McGinnity DF, Riley
RJ: Fully automated analysis of activities catalysed by the
major human liver cytochrome P450 (CYP) enzymes: as-
sessment of human CYP inhibition potential. Xenobiotica
1999;29:53–75.
531
6. Kitaura M, Nakajima T, Imai T, Harada S, Combadiere C,
Tiffany HL, Murphy PM, Yoshie O: Molecular cloning of
human eotaxin, an eosinophil selective CC chemokine, and
identification of a specific eotaxin receptor, CC chemokine
receptor 3. J Biol Chem 1996;271:7725–7730.
7. Lazareno S, Birdsall NJ: Estimation of antagonist Kb from
inhibition curves in functional experiments: alternatives to
the Cheng-Prusoff equation. Trends Pharmacol Sci 1993;14:
237–239.
8. Motulsky HJ, Christopoulos A: Fitting Models to Biologi-
cal Data Using Linear and Non-Linear Regression. A Prac-
tical Guide to Curve Fitting. GraphPad Software Inc., San
Diego, CA, 2003. Available at: www.graphpad.com.
Address reprint requests to:
Steven J. Charlton, Ph.D.
Novartis Horsham Research Centre
Wimblehurst Road
Horsham, West Sussex, RH12 5AB, UK
E-mail: steven.charlton@novartis.com