1282 Current Topics in Medicinal Chemistry, 2012, 12, 1282-1290

The IC50 Concept Revisited

Gary W. Caldwell*, Zhengyin Yan, Wensheng Lang and John A. Masucci

Community of Research Excellence & Advanced Technology (CREATe), Janssen Pharmaceutical Companies of Johnson
and Johnson, Welsh and McKean Roads, Spring House, PA 19477-0776, USA

Abstract: A major strategy used in drug design is the inhibition of enzyme activity. The ability to accurately measure the
concentration of the inhibitor which is required to inhibit a given biological or biochemical function by half is extremely
important in ranking compounds. Since the concept of the half maximal inhibitory concentration (IC50) is used extensively
for studying reversible inhibition enzymatic reactions, it is important to clearly understand the experimental design and
the mathematical modeling techniques used to generate IC50 values. The most important part of the experimental design is
to measure the rate of production of [P] during the linear phase of the time course of the reaction and to prove that the en-
zyme-catalyzed reaction is reversible. The most important part of the mathematical modeling is to select the correct model
and to have a firm understanding on how to handle outliers in the data. These topics are discussed in greater detail along
with a discussion on how much quantitative and mechanistic information can be reasonably deduced from an experiment.
Keywords: IC50, Ki, EC50, kinetic parameters, assay conditions, data fitting strategies.

INTRODUCTION case, the concentration of the agonists or antagonists that

Many endogenous and exogenous molecules have been
discovered that bind to an enzyme and in doing so increase
(i.e., stimulate) or decrease (i.e., inhibit) its catalytic activity
[1]. The stimulation or inhibition of an enzyme-catalyzed
reaction can enable the selective modulation of a variety of
biochemical processes such as cell growth, division and vi-
ability (untenable), or interrupting major metabolic pathways
by blocking the formation of an essential or undesirable me-
tabolite [2]. Inhibition of enzyme activity represents a major
strategy in drug design as noted by the fact that many of the
top drugs by sales are enzyme inhibitors [3]. Enzyme inhibi-
tion is complementary to receptor modulation via antagonists
and in some cases can be used to potentate the activity of a
desirable species by inhibiting its degradation [4]. In com-
paring endogenous and exogenous antagonist inhibitors (i.e.,
small molecules, drugs, or ligands) to a single enzyme, the
concept of the half maximal inhibitory concentration (IC50)
is extensively used in the pharmaceutical world as a measure
of the effectiveness in inhibiting biological or biochemical
functions. The IC50 value indicates the concentration of the
inhibitor which is required to inhibit a given biological or
biochemical function by half. In other words, large IC50 val-
ues denote inhibitors that interact less effectively with an
enzyme than inhibitors that have small IC50 values [5]. The
IC50 is comparable to the half maximal effective concentra-
tion (EC50) for agonist drugs. In this case, EC50 represents
the plasma concentration of the drug which is required for
obtaining 50% of a maximum biological effect. The IC50
concept also can be used for in-vitro or ex-vivo assays where
a comparison between a set of unlabeled antagonists or ago-
nists (i.e., inhibitors) is ranked against a known labeled or
unlabeled selective enzyme ligand (i.e., substrates). In this
*Address correspondence to this author at Janssen Research and Develop-
ment, Welsh and McKean Roads, Spring House, PA 19477-0776, USA; Tel:
(215) 628-5537; Fax: (215) 628-7064; E-mail: gcaldwel1@its.jnj.com
inhibit the binding of a known ligand by 50% is determined.
Under equilibrium conditions, a ligand binding IC50 value
may be converted to a true dissociation constant (Ki). Here
Ki is defined as an equilibrium constant for the dissociation
of the enzyme-inhibitor bound complex [6-8].
At first glance the concept of IC50 data appears to be
simple enough for pharmacology and in-vitro or ex-vivo
assay applications. However, when you actually collect and
fit the data necessary to determine IC50 values, it is discov-
ered that there are several complexities and ambiguities in its
application. If these complexities and ambiguities are not
fully understood, IC50 data can be totally misleading in pre-
dicting the potency of a set of inhibitors against an enzyme.
For example, consider the case of the cyclooxygenase
(COX) enzymes that are involved in the prostanoid biosyn-
thetic pathway for catalyzing the conversion of arachidonic
acid to prostaglandin H2 [9]. The anti-inflammatory effects
of nonsteroidal anti-inflammatory drugs (NSAIDs) result
from the inhibition of the COX-1 and COX-2 enzymes
thereby decreasing the production of prostaglandin H2. To
discover COX-1 and COX-2 inhibitors, selective in-vitro
assays have been developed to rank order inhibitors based on
IC50 data (Table 1).
The range of IC50 values in (Table 1) is significant when
comparing different in-vitro cell assays [10-13]. Piroxicam is
an example of a COX-1 and COX-2 inhibitor with an aston-
ishingly wide range of IC50 values in these assays while Ibu-
profen is an example of an inhibitor with a more consistent
range of IC50 values. Thus, the absolute value of these in-
vitro cell assays is restricted because the IC50 value strongly
depends on the test system, cell types, stimulating agents,
culture conditions and substrate concentration used (i.e., ara-
chidonic acid). Even if relative IC50 data are used to rank
order NSAIDs, it is noted that different NSAIDs would be
chosen from different in-vitro assays. For example, Flur-
biprofen is the most potent inhibitor in the COX-1 whole

1873-5294/12 $58.00+.00 © 2012 Bentham Science Publishers

The IC 50 Concept Revisited
insect cell assay, Piroxicam is the most potent inhibitor in
the COX-1 whole cell assay and Diclofenac is the most po-
tent inhibitor in the COX-1 human whole blood and ho-
mogenized insect cell assays. Therefore, an IC50 is a relative
value whose magnitude clearly depends upon many experi-
mental factors. In addition, IC50 values will also depend upon
the enzymatic mechanism of action and the mathematical
model used to extract this information.
The IC50 data in (Table 1) could be complicated by the
fact that different enzymatic mechanisms of action may be in
play. Enzyme inhibitors fall into two broad types: reversible
and irreversible. Reversible inhibitors generally bind via
non-covalent or weakly covalent forces, and can be removed
from the enzyme by dilution, gel filtration, or dialysis. They
form a rapid equilibrium with the enzyme that results in a
fixed degree of inhibition, which depends on their concentra-
tion. Irreversible inhibitors generally bind covalently to the
enzyme and are not removed by dilution or dialysis and the
degree of inhibition is time-dependent. Thus, some NSAIDs
in (Table 1) may have been irreversible or reversible inhibi-
tors of COX-1/2 enzymes while others may have been slow
time-dependent irreversible or reversible inhibitors. It is
sometimes not clear when using literature IC50 values if dif-
ferent mechanisms of action have been considered. For ex-
ample, reversible inhibition of enzyme-catalyzed reactions
can follow complete (full) or partial simple-mixed inhibition
mechanisms. These mechanisms are sometimes referred to as
linear and hyperbolic inhibitors, respectively [14]. A com-
pletely reversible enzyme inhibitor is one where the rate of
formation of the product primarily arises from an enzyme–
substrate complex where as a partially reversible enzyme
inhibitor is one where the rate of formation of the product
primarily arises from a mixture of enzyme–substrate and
enzyme-substrate-inhibitor complexes [15].
Different mathematical modeling techniques of the raw
data (i.e., rate of production of product or velocity) will have
a significant effect on the IC50 values in (Table 1). The most
mathematically accurate manner to fit the raw data is to per-
form a nonlinear regression analysis [16]. It is obvious that
the results are dependent upon the correct model being cho-
sen to emulate the experimental data. Once the model is se-
lected, the user makes an initial guess of the kinetic parame-
ters in the model and a curve is calculated. These kinetic
parameters are then adjusted by computer algorithms until
the calculated curve comes close to the raw data. In this type
of analysis it is typically assumed that the substrate concen-
trations are known more accurately than the raw data and
thus, this assumption leads to the standard method of mini-
mizing the sum of the squares of the vertical distances be-
tween the raw data and the curve. For enzyme kinetic appli-
cations, common models are Michaelis-Menten type equa-
tions [4, 14]. It is also common practice to plot percent activ-
ity as a function of log inhibitor concentration; logistic re-
gression analysis is used to determine the inflection point in
the curve, which represents the IC50 value [16]. Others have
used Dixon graphical methods to estimate IC50 values for
competitive, uncompetitive, and mixed reversible inhibition
mechanisms [17]. Many literature IC50 values have few de-
tails on fitting procedures.
Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 11 1283
In this review, we will revisit the IC50 concept and look at
experimental design complexities and mathematical model-
ing ambiguities for reversible inhibition enzyme-catalyzed
reactions in the hope of alerting the reader to these pitfalls.
We will also review data fitting procedures to convert IC50
values to a true dissociation constant (Ki). It is our hope that
a better understanding of the IC50 concept will lead to better
questioning of the experimental design and analysis of the
data such that it is possible to decide on how much quantita-
tive and mechanistic information can be reasonably deduced
from an experiment.
GENERAL COMMENTS ON INIHIBITOR DATA
COLLECTION AND CURVE FITTING STRATEGIES
If all experiments were carried out with uniform and con-
sistent biological materials, data was collected with perfect
accuracy and the mechanism of action of the enzyme reac-
tion was known, there would be no complexities and ambi-
guities in data interpretation. However, in the real world,
biological materials have variable properties, measurements
made on them are subjected to experimental error and the
mechanism of action is typically unknown. Thus, the reason
why the absolute IC50 data varies for each in-vitro assay in
(Table 1) is due to the biological material used, the assay
culture conditions, the experimental error generated, multiple
binding sites on the enzyme, different fitting models, and so
on. As long as the data in (Table 1) was collected in a vali-
dated statistical manner and analyzed with the correct fitting
model, all the relative IC50 data are correct for that particular
in-vitro assay. Thus, what is the best in-vitro assay to use in
this case? From a pharmacology point of view, the in-vitro
assay that correlates best with the in-vivo case is the pre-
ferred in-vitro assay. The reader is referred to the literature
for more details around this topic [18]. With this imperfect
experimental situation in mind, we will only comment on
general experimental design conditions and data fitting tech-
niques used in reversible competitive inhibition enzyme ki-
netics to generate IC50 or Ki data.
A hypothetical enzyme inhibition reaction is illustrated in
Fig. (1) where the biological activity of compound [P] can be
attenuated via inhibition of the enzyme involved in its bio-
synthesis from substrate [S]. Reversible enzyme inhibitors
typically fall into three major mathematical kinetic models.
The first is competitive inhibitors that compete directly with
the substrate for its binding site on the enzyme, so that either
the substrate [S] or the inhibitor [I] molecule is bound to the
enzyme to produce [ES] or [EI]. The other two most com-
mon mathematical kinetic models are un-competitive and
simple-mixed non-competitive inhibition kinetics which will
be discussed later.
The IC50 or the Ki of an inhibitor for the enzymatic reac-
tion shown in Fig. (1) can be determined by constructing an
inhibitor-response curve. Briefly, to construct this curve, it is
essential to have reliable analytical methods to measure the
concentration of either the disappearance of the substrate [S]
or the appearance of the product [P] with time. A typical
time-course curve for the appearance of [P] is shown in Fig.
(2) where the concentration of [P] is measured over a time
period. It is noted that there is an initial linear phase of the
1284 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 11 Caldwell et al.

Table 1. Range of IC50 Values (
M) Generated for Individual NSAIDs Using a Variety of In-Vitro Assay COX-1 and COX-2 Assays

COX-1
IC50 (
M)

Whole Insect Cells Whole Cells Homogenized Human Whole

NSAIDs
[10]1 [11]3 Insect Cells [12]5 Blood [13]

Flurbiprofen 0.001 0.1 0.1 0.44

Diclofenac 0.003 1.6 0.04 0.14

Indomethacin 0.002 0.03 0.1 0.16

Naproxen 0.5 9.6 1.1 7.76

Piroxicam 1 0.002 13 0.76

Ibuprofen 2.9 4.9 3.3 4.75

COX-2
IC50 (
M)

Whole Insect Cells Whole Cells Homogenized Human Whole

NSAIDs
[10]2 [11]4 Insect Cells [12]6 Blood [13]

Diclofenac 0.006 1.1 0.1 0.05

Indomethacin 0.052 1.7 0.9 0.46

Flurbiprofen 0.069 0.1 0.4 6.42

Piroxicam 0.53 0.6 >100 8.99

Naproxen 6.3 5.7 36 73.74

Ibuprofen >50 73 37.5 >30

1. Infected Spodoptera frugiperda cells infected with recombinant human COX-1

2. Infected Spodoptera frugiperda cells infected with recombinant human COX-2
3. Whole cell cultures of bovine aortic endothelial cells
4. Whole cell cultures of J774.2 macrophages treated with E. coli lipopolysaccharide
5. Infected Spodoptera frugiperda cells infected with recombinant human COX-1 and then homogenized.
6. Infected Spodoptera frugiperda cells infected with recombinant human COX-2 and then homogenized.

Fig. (1). Hypothetical illustration of an inhibitor interfering with a biological process.

The IC 50 Concept Revisited
Fig. (2). Hypothetical illustration of an enzyme-catalyzed reaction
for the time dependent product of [P].
curve followed by a nonlinear phase at longer times. The
nonlinearity of an enzymatic reaction is caused by many
factors including instability of the enzyme, changes in assay
conditions (i.e., volume, temperature, pH) as the reaction
progresses, inhibition of the enzyme by the product formed,
and so on. Thus, to have any validity, it is essential to meas-
ure the rate of production of [P] during the linear phase
where these nonlinear effects are negligible. That is, the
change in the concentration of [P](final) -[P](initial) divided by
the time period t(final) - t(initial) (i.e.,
P/
t = V0) where the
quantity V0 is referred to as the rate of formation or the initial
velocity of the reaction.
A typical procedure that is followed to ensure that the
velocity is measured in the linear phase is to keep
t down to
a period of time in which less than 10% of the total substrate
[S] is used (i.e., initial velocity approximation). However, it
is more prudent to ensure that
t is chosen within the linear
initial rate period by carrying out measurements after a series
of increasing incubation times. Furthermore, the reversibility
of the enzyme-catalyzed reaction needs to be demonstrated.
One way to test for reversibility of an enzyme-catalyzed re-
Fig. (3). Hypothetical illustration of an inhibitor vs. response for a biological process.
Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 11 1285
action is to compare enzymatic activity at a fixed concentra-
tion of [I] with the results obtained when the enzyme is pre-
incubated with the same concentration of [I] but in a smaller
volume and diluted. Enzymatic activity in the latter case
should be greater than the former since the inhibitor should
dissociate from [EI] on dilution [2]. It is clear that V0 can be
expected to be proportional to the enzyme concentration [E]
provided that [E] is significantly less than the concentration
of the substrate [S] which is typical for most enzyme kinetic
reactions.
The V0 (inhibitor) vs. response curve is constructed by
measuring the initial velocity of the reaction V0 in the pres-
ence of an increasing concentration of inhibitor. An example
of this data is shown in Fig. (3) where the black dots repre-
sent individual experimental V0 measurements as outlined
above and the solid line connecting the dots is generated
from the mathematical model used to fit the data.
In the ab-
sence of an inhibitor ([I] = 0), the maximum concentration of
[P] is produced or the maximum velocity (V) for the reaction
is obtained Fig. (3A). As greater concentrations of inhibitor
are added, the concentration of [P] is reduced since the in-
hibitor [I] is binding to [E] instead of the substrate [S] and
thus, reducing the concentration of [P] (i.e., the velocity of
its production is reduced). Typically six to eight inhibitor
concentrations [I] are used with only one substrate concen-
tration [S] to generate reliable IC50 data [19]. In other words,
to generate IC50 data requires a minimal of 18 - 24 individual
V0 measurements of [P] if the experiment is done in tripli-
cate. When the inhibition data is re-plotted using log units on
the x-axis the familiar symmetrical sigmoidal curve is pro-
duced Fig. (3B).
The goal at this point is to determine the IC50 value of the
inhibitor from the inhibitor-response curve Fig. (3); that is,
the concentration of [I] which is required for obtaining 50%
of a maximum effect. Before processing the raw velocity
data and extracting the IC50 value, the data should be care-
fully checked for any large variations in apparent precision
and unexpected patterns in the numbers. Thus, plotting the
data is essential since the human eye is excellent at recogniz-
ing unexpected patterns. The IC50 values can be extracted
from the inhibitor-response curve for a given inhibitor by
using several different fitting techniques. The raw velocity
data can be scaled before analysis or not, just as long as there
1286 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 11
is no loss in data precision. It should be remembered that
different curve fitting techniques will give different IC50 val-
ues for the same raw data being processed. Some of the more
common methods used are outlined below. For example, the
IC50 value can be approximately estimated by eye-balling the
sigmoidal curve in Fig. (2B) or using a simple exploration
technique [20]
IC50 = (50%-low percentage)*(high conc.-low conc.) + (low conc.)
(high percentage-low percentage)
where, low percentage = highest percent inhibition less than
50%; high percentage = lowest percent inhibition greater
than 50%, low concentration = concentration of test com-
pound corresponding to the low percentage inhibition, and
high concentration = concentration of test compound corre-
sponding to the high percentage inhibition. While these
methods are acceptable in some cases, it is more advanta-
geous to determine the IC50 value of an inhibitor by nonlin-
ear regression of a plot of enzyme activity versus inhibitor
concentration using a nonlinear regression fitting software
package [21]. An important point to remember here is that
the correct mathematical model must be chosen by the user
in order to determine the IC50 value. For example, a simple
exponential decay equation (2) can be used to fit the data
when the data is plotted as illustrated in Fig. (3A) [21].
V0 = (V - Plateau)e -K*[I] + Plateau
In equation (2) V0 is the rate of formation (or initial velocity)
of [P] as a function of the concentration of [I], [I] is the con-
centration of the inhibitor, V is the rate of formation of [P]
when [I] = 0, the plateau is rate of formation of [P] when [I]
=
, the K value is the rate constant, and the IC50 value is
equal to ln(2)/K. Note that [I] is an experimental value,
whereas V, K and Plateau are variables in equation (2).
When the data is plotted as illustrated in (Fig. 3B) a simple
sigmoidal equation (3) can be used to fit the data [21].
(Top - Bottom)
V0 = Bottom +
(1+10(LogIC50-[I]))*Hill Slope
In equation (3) the Top and Bottom parameters are the pla-
teaus (V0 values) at [I] = 0 and [I] =
in (Fig. 3B), respec-
tively. The Hill Slope parameter is the slope of the linear
portion of the curve in (Fig. 3B) between the top and bottom
plateaus. This parameter can be either set to -1.0 or allowed
to be a variable parameter. When the Hill Slope is set to -1 it
is assumed that the ligand binds to an enzyme following the
law of mass action; that is, the enzyme-catalyzed reactions
are at equilibrium. Note that [I] is an experimental value
whereas Top, Bottom, LogIC50 and Hill Slope are variables
in equation (3). When all four variables are used to fit the
raw data, equation (3) is referred to as a four-parameter lo-
gistic model. This model is typically used when there are lots
of experimental data points (i.e., >12 individual rate meas-
urements of [P]). However, equation (3) can also be used by
setting the variables to constants; that is, setting the Hill
Slope to -1 and the Top plateau to the V0 value at [I] = 0 or
setting the Hill Slope to -1 and the Bottom plateau to the V0
value at [I] =
and so on. Equations (2) and (3) will give the
same IC50 values as along as there is very little error in the
experimental data. Unfortunately, in the real world, there is
significant experimental error and thus, depending on the
Caldwell et al.
amount of scatter in the data all of these methods will give
different IC50 values. In some cases, the IC50 values can have
2 - 4 fold-differences in their values when using different
fitting procedures.
It is clear that the IC50 values vary with experimental
design and they are not directly comparable unless identical
assay conditions are used; they provide no indication of the
mechanism of action for enzyme inhibition; they are not a
direct indicator of binding affinity since they are dependent
(1)
upon the substrate concentration used; time-dependent inhi-
bition and multiple binding events must be understood to
interpret IC50 values [22]. Therefore, why are IC50 values
measured instead of an intrinsic thermodynamic quantity
such as the enzyme-inhibitor dissociation constant (Ki)? To
gain some understanding of this situation, it is interesting to
compare the experimental effort of generating a true disso-
ciation constant Ki to that of an IC50 value. To determine a Ki
values for an inhibitor, individual rate measurements of [P]
are made while independently varying the concentration of
the substrate [S] and the concentration of the inhibitor [I].
This is shown in (Fig. 4) where the velocity of the enzyme
reaction illustrated in (Fig. 1) is measured for a range of sub-
strate concentrations in the absence and presence of one con-
centration of the inhibitor.
(2)
(3)
Fig. (4). Hypothetical illustration of the Rate of Formation of [P]
vs. [S] substrate concentration for a biological process in the pres-
ence and absence of an inhibitor.
Here a simple Michaelis-Menten approach is used to fit
the top curve in (Fig. 4) to calculate the kinetic parameters
VMax and Km. Briefly, equation (4) is used to fit the data
V0 = VMax [S]0
Km + [S]0
(4)
where VMax is the maximal velocity of the enzymatic reac-
tion and occurs at the highest substrate concentration; Km is
the Michaelis constant and is a measure of the substrate con-
centration when the enzymatic reaction is at half-maximal;
[S]0 is the initial substrate concentration. To determine Ki,
the user must select the enzymatic mechanism of action. For
example, we could assume that the enzymatic reaction fol-
lows a competitive mechanism and equation (5) would be
used to fit the data
The IC 50 Concept Revisited Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 11 1287

used while equilibrium inhibition Ki values are constant. To

V0 = VMax [S]0
1+[I]0 Km + [S]0
Ki
(5)
Note that when [I] = 0, equation (5) reduces to the
Michaelis-Menten equation (4). Comparing equations (4)
and (5), we see that when an inhibitor follows a competitive
mechanism it has the effect of modulating the Km value but
not affecting the VMax value. In other words, a competitive
inhibitor acts by reducing the concentration of the free en-
zyme available for substrate binding. In doing this, the initial
rate of production of [P] is slower in the presence of [I] when
compared to the rate of production of [P] when there is no
inhibitor present. However, given enough time both reac-
tions will produce the same maximum amount of [P].
Typically six to eight substrate concentrations [S] are
used with only one inhibitor concentration [I] to generate
reliable Ki data [18]. Thus, to generate Ki data requires a
minimal of 36 - 48 individual rate measurements of [P] if the
experiment is done in triplicate. Thus, IC50 values can be
generated with approximately 50% less effort than a K i
value. This is the reason why IC50 values are preferred over
Ki values when large numbers of compounds must be as-
sayed.
The rate of formation V0 data shown in Figs (3 and 4) is
describe this conversion mathematically requires that the
mechanism of action for the enzymatic reaction under con-
sideration is known. In Fig. (6) is shown the possible mecha-
nisms of action for an enzyme-catalyzed reaction. The
scheme in Fig. (6) illustrates a simple-mixed inhibition
mechanism, which is a balance between competitive (i.e., Ki
=
; [ESI]t = 0) and uncompetitive (i.e., Ki =
; [EI]t = 0)
inhibition. The subscript t in Fig. (6) is to remind us that the
concentrations of these various species are time dependent
and equilibrium is rapidly established. The Ks term is de-
fined as the disassociation equilibrium constant for substrate
[S].
[E]eq [S]eq
Ks =
[ES]eq (6)
Note that we are not using Km since we are assuming a
rapid equilibrium approximation. The Ki term is defined as
the disassociation equilibrium constant for substrate [I].
[E]eq [I]eq
Ki =
[EI]eq (7)

and the Ki term is defined as the disassociation equilibrium
constant for substrate [I].

[ES]eq [I]eq
hypothetical data; that is, it contains no experimental error.
Real data has both systematic and random experimental error
K i
=
and thus, statistical analysis needs to be performed. Many
experimentalists are familiar with the situation in which [ESI ]eq (8)
some of the data appears to be significantly different from
the other data points. For example, consider Fig. (5) where The kcat (i.e., catalyze) term describes the overall rate
the plots contain experimental error; in Fig. (5A), the data constant for the entire number of chemical steps in the reac-
has been fitted using equation (3) excluding the single out- tion necessary to generate [P]t.
lier. If this outlier data point was included in the calculation, To obtain a mathematical expression for the IC50 term,
the IC50 value would be significantly different than the value we use the method of Cheng-Prusoff [6]; that is, for a given
obtained by leaving it out. What is the best approach when concentration of [S]0, what concentration of [I]0 is required
dealing with outliers – should we take them out or leave to yield 50% inhibition? We can express the percent inhibi-
them in? For Fig. (5A), it is tempting to assume that this out- tion (i%) as:
lier is due to a systematic error (i.e., experimental mistake)
and the outlier could be removed in an ad hoc manner. This
informal approach is typically very subjective and difficult to i
i% = (1
) *100
document; thus, this approach is not recommended. For Fig.  (9)
(5B), it is not obvious if any of the data are outliers and the
scatter in the data is due to random experimental error or Where
i is the rate in the presence of a complete non-
multiple enzymatic mechanisms of action. In many cases, competitive inhibitor and
is the rate in the absence of an
when dealing with enzymatic data, not enough replicate data inhibitor.
VMax [S ]0
sets are acquired to use standard statistical tests to eliminate
outliers [23]. In addition, using a weighted regression ap-
proach is not recommended for nonlinear regression analysis  i
K s +  [S ]0 K + [S ]0 K s + [S ]0
= = s =
[21, 24]. It has been suggested by Motulsky and Brown [24]  VMax [S ]0
K s +  [S ]0 (1 + [ I ]0 ) K + (1 + [ I ]0 )[S ]
K s + [S ]0
that the best approach to outlier detection and elimination is
K i
s 0
to use a method that contains both a robust nonlinear regres- Ki (10)
sion method and a False Discovery Rate approach [21]. The
Combining equations (9) and (10) leads to equation (11)
reader is referred to their paper [24] for more details.
 
GENERAL COMMENTS ON GENERATING KI-  
NETIC PARAMETERS i
i% = (1
) *100 = 1

K s + [S ]0  *100
  
 (1 + K ) K s + (1 + K  )[S ]0 
A typical problem in enzymatic data analysis is how to [ I ]0 [ I ]0
convert an IC50 to a Ki value. As pointed out above, IC50  i i (11)
values depend upon the concentration of the substrate being
1288 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 11 Caldwell et al.

A B

Fig. (5). Hypothetical illustration of a Log inhibitor vs. V0 plot for a biological process where there is experiment error in the data.

Fig. (6). Scheme for the possible mechanisms of action for an enzyme-catalyzed reaction.

and finally, after performing a little algebra on equation (11) K s + [ S ]0

leads to equation (12). IC50 =
K s [ S ]0
  +
  K i K i
(15)
  
  For a competitive inhibition reaction (Fig. (6); Ki =
;
[ I ]0
i% =   *100 [ESI]t = 0), and an uncompetitive inhibition reaction (Fig.
 [ I ] + (K s + [S ]0 )  (6); ., Ki =
; [EI]t = 0), the IC50 term can be derived in the
 0  K [S ]   same manner, respectively.
  s +
0  
  Ki Ki  

IC50 = K i (1 +
[S ]0 )
(12)
Ks (16)
If
K s + [ S ]0 Ks
[ I ]0 = I C 50 = K i
(1 + )
K s [ S ]0
+
[S]0 (17)
K i K i

(13) Thus, we can see that the relationship between Ki and the
then equation (12) becomes IC50 values is complicated. The IC50 value depends on the
mechanism of action (i.e., competitive, uncompetitive or
some type of mixed competitive mechanism), and the con-
centration of the substrate. In addition, the concentration of
[ I ]0 1
i% = ( ) * 100 = ( ) * 100 = 50% the substrate [S]0 in equations (15) – (17) is not the initial
[ I ]0 + [ I ]0 2 (14)
experimental concentration but is a values 5 - 10% less in
concentration. Remember that we generated all these
or in other words, the IC50 value is equal to: mathematic models under the initial velocity assumption.
Thus, it is difficult to accurately convert IC50 to Ki values.
The IC 50 Concept Revisited Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 11 1289

In general, equations (15), (16), and (17) suggest that the completive inhibition mechanisms that can be devised from a
IC50 value is great than Ki. Under special mechanisms of single substrate and inhibitor, we will formulate the general
action or experimental design, the complexity between Ki case.
and the IC50 values is reduced. For example, if the disasso-
This scheme is the general partially non-competitive in-
ciation equilibrium constant Ki
is equal or approximately
hibition mechanism for a single substrate and a single inhibi-
equal to Ki, then equation (15) reduces to equation (18).
tor. When K i = K s = K i
  and *
k cat = 0 the
I C 50 = K i (18) reaction scheme reduces to a pure enzyme reaction. Remem-
If the experimental conditions are set up such that [S]0 is ber that when a dissociation equilibrium constant (e.g.,
equal to Ks, then equations (16) and (17) reduce to equation K i = [E ]eq [I]eq [EI]eq ) approaches infinity, this implies
(19) and (20), respectively.
that the enzyme complex (e.g., [EI]eq) approaches zero con-
I C 50 = 2 K i (19) centration. When K s = K i
  and *
k cat = 0 the
I C 50 = 2 K i
(20)
reaction scheme reduces to a pure competitive inhibition
enzyme reaction. When
Here we are assuming that a substantial amount of his- K i = K s
  and *
k cat = 0 the reaction scheme
torical enzyme kinetic data is available such that the Ks value reduces to a pure un-competitive inhibition enzyme reaction.
is known for the substrate. Using the rapid equilibrium approach, the velocity equation
Next consider the situation where the enzyme forms an can be derived for this reaction scheme.
active-complex after binding to the inhibitor [15, 25]. This
 * [ I ]0 
type of mixed non-competitive inhibition enzymatic reaction VMax + VMax [ S ]0
is shown in Fig. (7) and is referred to as partially non-  K i
 (21)
V0 =
 [I ]   [I ] 
competitive inhibition enzyme reactions. Partially non-
competitive inhibition is the most important type of inhibi- K s 1 + 0  + [ S ]eq 1 +
0 
tion for a variety of regulation cell functions.  Ki   Ki 

Where VMax = k cat [ E ]0 *

and VMax = k cat
*
[ E ]0 .
We can express the percent inhibition (i%) as:
  * [ I ]0 

 VMax + VMax  
[ S ]0 
  Ki 
 
 K s 1 + [I ]0  + [ S ]eq 1 + [I ]0  
   K   K  
i% = (1
i ) *100 = 1
 i  i
[ ]  *100
  VMax S 0 
 K s + [S ]0 
 
 
 
 (22)
Fig. (7). Hypothetical illustration of a Log inhibitor vs. relative It is not possible to obtain a simple algebraic equation for
response plot for a biological process where Compound A follows IC50 as a function of Ki and [S]0 using equation (22). Com-
the enzymatic mechanism outlined in Fig. (6) and Compound B paring equations (11) and (22), it is clear that the % inhibi-
follows the enzymatic mechanism outlined in Fig. (8). tion values are complicated and cannot be directly used to
Since there are a large number of partially non- convert IC50 to Ki values. In general, you could fit the data in

Fig. (8). Scheme for the possible mechanisms of action for an enzyme-catalyzed reaction.
1290 Current Topics in Medicinal Chemistry, 2012, Vol. 12, No. 11
Fig. (7) using equation (3) to generate two IC50 values. How-
ever, since compound A and compound B are operating un-
der two different enzymatic mechanisms, the IC50 values
cannot be compared to one another to rank compounds.
CONCLUSION
In this review, we have examined the IC50 concept for
reversible inhibition enzymatic reactions highlighting the
complexities in their experimental design and the ambigui-
ties in their mathematical modeling. The most important part
of the experimental design is to measure the rate of produc-
tion of [P] during the linear phase of the time course for the
reaction and to prove that the enzyme-catalyzed reaction
under study is reversible. The most important part of the
mathematical modeling is to select the correct model and to
have a firm understanding on how to handle outliers in the
data. In addition, the accuracy of IC50 values will depend
upon the percent conversion rate of the substrate [26]. In
other words, a high percent conversion rate of the substrate
[S] will provide higher signal-to-noise ratio for [P] and better
dynamic range in the assay; however, a high percent conver-
sion rate of the substrate [S] will lead to the measured IC50
values deviating from the true values. This problem arises
because the mathematical models assume that the substrate
concentration is the same as its initial concentration. When
we examined the problems in converting IC50 values to a true
dissociation constant (Ki) it becomes clear that this process is
complicated and is not recommended for obtaining highly
accurate data. To obtain accurate Ki values, it is advanta-
geous to use Log inhibitor vs. V0 plots and the correct kinetic
model. The Ki values require approximately 50% more effort
to generate than IC50 values. It is this reason why Ki values
are not preferred over IC50 values when large numbers of
compounds must be assayed. Finally, for reversible inhibi-
tors that bind to the enzyme with very high affinity, it will be
necessary to experimentally use a low concentration of in-
hibitor. In these cases, the dynamic range of the assay will be
compromised and it is probably better to measure the Ki val-
ues directly. That is,
Fig. (9). Scheme for the combination of an enzyme to an inhibitor.
REFERENCES
[1] Goodman and Gilman’s The Pharmacological Basis of Therapeu-
tics, Editors, Hardman, J.G., Limbird, L.E., and Gilman, A.G Tenth
Edition, McGraw Hill Medical Publishing Division, New York,
New York, USA, 2001.
[2] Sandler, M.; Enzyme Inhibitors as Drug. University Park Press,
Maryland, USA, 1980.
[3] Aitken, M.; Berndt, E.R.; Cutler, D.M.; Prescription Drug Spend-
ing Trends in The United States: Looking Beyond The Turning
Point. Health Aff., 2009, 28(1):151-160.
Received: December 18, 2011 Revised: February 14, 2012 Accepted: March 26, 2012
Caldwell et al.
[4] Bisswanger, H.; Enzyme Kinetics: Principles and Methods, Wiley-
VCH, Verlag GmbH, Weinheim, Federal Republic of Germany,
2002.
[5] Neubig, R.R.; Spedding, M.; Kenakin, T.; Christopoulos, A. Inter-
national Union of Pharmacology Committee on Receptor Nomen-
clature and Drug Classification. XXXVIII. Update on Terms and
Symbols in Quantitative Pharmacology. Pharmacol. Rev., 2003,
55:597–606.
[6] Cheng Y.; Prusoff, W.H.; Relationship between the inhibition
constant (KI) and the concentration of inhibitor which causes 50
per cent inhibition (IC50) of an enzymatic reaction. Biochem.
Pharmacol., 1973, 22:3099-108.
[7] Munson P.J., Rodbard D.; An exact correction to the "Cheng-
Prusoff" correction. J. Recept Res., 1988, 8: 533-46.
[8] Cheng H.C.; The power issue: determination of KB or Ki from
IC50. A closer look at the Cheng-Prusoff equation, the Schild plot
and related power equations. J. Pharmacol. Toxicol. Methods,
2001, 46: 61-71.
[9] Smith W.L.; Prostanoid biosynthesis and mechanisms of action.
Am. J. Physiol. 1992, 263: 181-191.
[10] Cromlish W.A.; Kennedy B.P. Selective inhibition of COX-1 and -
2 using intact insect cell assays. Biochem. Pharmacol., 1996, 52:
1777-1785.
[11] Vane J.R.; Botting R.M. New insights into the mode of action of
anti-inflammatory drugs. Inflamm. Res., 1995, 44: 1-10.
[12] Gierse J.K.; Hauser S.D.; Creely D.P.; Koboldt, C.; Rangwala,
S.H.; Isakson, P.C.; Seibert, K. Expression and selective inhibition
of the constitutive and inducible forms of human cyclo-oxygenase.
Biochem. J., 1995, 305: 479-484.
[13] Dannhart, G.; Kiefer, W. Cyclooxygenase inhibitors-current status
and future prospects. Eur. J. Med. Chem., 2001, 36: 109-126.
[14] Segel, I. Enzyme Kinetics, John Wiley, New York, 1973.
[15] Palatini, P. The Interaction between Full and Partial Inhibitors
Acting on a Single Enzyme. Mol. Pharmacol., 1983, 24: 30-41.
[16] Cornish-Bowden, A. Analysis of Enzyme Kinetic Data, Oxford
Science Publications, New York, 1995.
[17] Burlingham, B.T.; Widlanski, T.S. An Intuitive Look at the Rela-
tionship of Ki and IC50: A More General Use for the Dixon Plot. J
Chem., Ed. 2003, 80(2): 214-218.
[18] Yan, Z.; Caldwell, G.W. Optimization in Drug Discovery: In-vitro
Methods, in Methods in Pharmacology and Toxicology, Humana
Press, Totowa, New Jersey, USA, 2004.
[19] Kakkar, T.; Pak, Y.; Mayersohn, M. Evaluation of a Minimal Ex-
perimental Design for Determination of Enzyme Kinetic Parame-
ters and Inhibition Mechanism. J. Pharmacol. Exper. Ther., 2000,
293: 861-869.
[20] Yan, Z.; Caldwell, G.W. Optimization in Drug Discovery: In-vitro
Methods, in Methods in Pharmacology and Toxicology, Evaluation
of Cytochrome P450 Inhibition in Human Liver Microsomes, Hu-
mana Press, Totowa, New Jersey, USA, 2004, pages 231-243.
[21] GraphPad Software Inc: All data was analyzed using Prism 5 for
Windows, Version 5.01; August, 7, 2007.
[22] Yan, Z.; Rafferty, B.; Caldwell, G.W.; Masucci, J.A. Rapidly Dis-
tinguishing Reversible And Irreversible Cyp450 Inhibitors By Us-
ing Fluorometric Kinetic Measurements” Eur. J. Drug Metab.
Pharmacokinet., 2002, 27(4): 281-287.
[23] Miller, J.C.; Miller, J.N. Statistics for Analytical Chemistry, 2nd
Edition, Ellis Horwood Limited, Chichester, West Sussex, England,
1992.
[24] Motulsky, H.M.; Brown, R.E. Detecting outliers when fitting data
with nonlinear regression – a new method based on robust nonlin-
ear regression and the false discovery rate, BMC Bioinformatics,
2006, 7(23):1-20.
[25] Whiteley, C.G. Enzyme Kinetics: Partial and Complete Competi-
tive Inhibition. BioChem. Ed., 1997, 25(3): 144-146.
[26] Yang, X.; Dynamic Ranges of Detection-Coupled Assays and Their
Effect on IC50 Measurements for Inhibition of Enzymatic Reac-
tions, J. Biomol. Screening, 2010, 15: 556-561.