Why IC50’s Are Bad for You And Other Surprises Petr Kuzmič, Ph.D. BioKin, Ltd.

What is enzyme inhibition on the molecular level COMBINATION OF TWO MOLECULES TO FORM AN ENZYME-INHIBITOR COMPLEX “Drugs produce their inhibitory action by combining with the enzyme [molecules].” “One molecule of drug will inhibit the activity of one [molecule] of enzyme.” Easson, L. H. & Stedman, E. (1936) Proc. Roy. Soc. B 121, 142-151. E I + enzyme molecule inhibitor molecular complex IC50's are bad for you

What is the inhibition constant (Ki) DISSOCIATION EQUILIBRIUM CONSTANT OF THE ENZYME-INHIBITOR COMPLEX E + I E·I low Ki (“dissociation”) means high binding activity dimension = concentration (moles/liter, M) “good” inhibitors have Ki’s around 10-9 moles/liter or better (nanomolar) 10-3 mili- mM 10-6 micro- mM 10-9 nano- nM 10-12 pico- pM 10-15 femto- fM 10-18 atto- “better” inhibitor IC50's are bad for you

A thought experiment: Effect of Ki on “fraction bound” Assume equal amounts: [E]0 = [I]0 = 1 nM E + I E·I 10-15 10-12 10-9 10-6 at nanomolar [E]0 = [I]0 and femtomolar Ki, there is no free enzyme left IC50's are bad for you

A thought experiment: Titrate 1 nM enzyme, Ki = 0.000001 nM remaining enzyme activity v recall: one molecule of inhibitor inhibits one molecule of enzyme V0 0.5 V0 recall: at [I]0 = [E]0 and Ki << [E]0, there is no free enzyme left 0.0 0.5 1.0 1.5 2.0 2.5 [E]0/2 [E]0 [I]0, nM IC50's are bad for you

Thought experiment, continued: What is the IC50 ? CONCENTRATION OF INHIBITOR THAT PRODUCES HALF-MAXIMUM INHIBITORY EFFECT remaining enzyme activity [E]0 = 1 nM Ki = 1 fM = 0.000001 nM v V0 IC50 in this case: IC50 ~ 0.5 nM IC50 ~ [E]0 / 2 IC50 in every case: IC50 > [E]0 / 2 0.5 V0 0.0 0.5 1.0 1.5 2.0 2.5 [E]0/2 [E]0 [I]0, nM IC50's are bad for you

What is the difference between Ki and IC50 ? IC50 DEPENDS ON ENZYME CONCENTRATION AND IS ALWAYS HIGHER THAN THE Ki competitive uncompetitive noncompetitive mixed-type Cha, S. (1975) “Tight binding inhibitors. I. Kinetic behavior” Biochem. Pharmacol. 24, 2177-2185. IC50's are bad for you

Implications for drug discovery: “Hitting the IC50 wall” NO MATTER HOW TIGHTLY THE INHIBITOR BINDS, THE IC50 CAN NEVER GET LOWER THAN [E]0/2 Assume: competitive [E] = 5 nM [S]0 = KM Ki(app) = Ki (1 + [S]/KM) Ki, nM IC50, nM 1,000 100 10 1 0.1 0.01 0.001 2,030 230 50 32 30.2 30.02 30.002 competitive [E] = 60 nM [S]0 = KM Ki, nM IC50, nM 1,000 100 10 1 0.1 0.01 0.001 2,002.5 202.5 22.5 The IC50 wall. 4.5 2.6 2.52 2.502 IC50's are bad for you

What is “tight binding” THERE IS NO SUCH THING AS A “TIGHT BINDING INHIBITOR” EXPERIMENTAL CONDITIONS “Tight binding” kinetics applies when the enzyme concentration in any given assay is greater than the inhibition constant.     [E] / Ki ~ 1 tight binding is beginning to show up     [E] / Ki ~ 10 tight binding is definitely present     [E] / Ki ~ 100 highly tight binding     [E] / Ki ~ 1000 extremely tight binding IC50's are bad for you

How prevalent is “tight binding” in a screening campaign? EXAMPLE: A PROTEASE CAMPAIGN A typical data set: Completely inactive: Tight binding: ~ 10,000 compounds ~ 1,100 ~ 400 ... NOT SHOWN about 4% of compounds Points to make: How prevalent is tight binding? For how many compounds is IC50 really meaningless? This is a histogram of Kiapp for 10,000+ compounds. Two groups of compounds, centered at about 200 M and 6 M, respectively. About 400 (or 4%, or one in 25) are "tight binding", so for one compound out of 25, the IC50 has no meaning. Data courtesy of Celera Genomics 2003-2005 IC50's are bad for you

A word about the Cheng-Prusoff Equation IT DOES NOT TAKE INTO ACCOUNT “TIGHT-BINDING”! Cheng, Y.-Ch. and Prusoff, W. H. (1973) “Relationship between the inhibition constant (Ki) and the concentration of inhibitor which causes 50 per cent inhibition (IC50) of an enzymatic reaction” Biochem. Pharmacol. 22, 3099-3108. competitive inhibition: IC50 = Ki (1 + [S]/KM) IC50 = Ki (1 + [S]/KM) + [E]0 / 2 competitive inhibition: Cha, S. (1975) “Tight binding inhibitor. I. Kinetic behavior” Biochem. Pharmacol. 24, 2177-2185. Many J. Med. Chem. papers still use the Cheng-Prusoff equation. If the conditions are tight binding ([E] > Ki) this produces wrong Ki’s. IC50's are bad for you

Misuse of the “fold increase” plot: A case study unnamed kinase compound “X” known ATP competitor MAIN MESSAGE This is one of the earlier ArQule results that will appear (at first sight) to be inconsistent with the most recent data: Here we see a only very slight increase, maybe 25%, going from [ATP]=Km to millimolar [ATP] Later on we will see a ten fold increase. But this is not a contradiction, because the results shown here were obtained under the conditions where the ENZYME CONCENTRATION WAS VERY HIGH. In retrospect, we now know that the Ki is around 0.5 nM. In these AKIP assays the enzyme concentration was 10-20 nM. These are conditions that we call “tight binding”. Under those conditions the Rule of Thumb for IC50 increase completely breaks down. CONCLUSION: Compound “X” is an ATP insensitive inhibitor not IC50's are bad for you

A word about the “fold increase in IC50” plot IT CAN BE VERY MISLEADING – IF “TIGHT BINDING” DOES OCCUR SIMULATED KINASE EXAMPLE: competitive KM(ATP) = 100 µM Ki = 0.5 nM [E]0 = 66 nM Very shallow slope, “unexpected”. [E]0 >> Ki. IC50's are bad for you

Another word about the “fold increase in IC50” plot SOMETIMES IT DOES WORK, BUT ... you don’t know if there is “tight binding” until you get the Ki(app) SIMULATED KINASE EXAMPLE: competitive KM(ATP) = 100 µM Ki = 0.5 nM [E]0 = 0.25 nM Steep slope, as “expected” [E]0 < Ki. IC50's are bad for you

Compound “X” is an ATP-competitive inhibitor after all 10-fold increase in IC50 MAIN MESSAGE: See slide #12, for FGFR1. IC50's are bad for you

Rules of thumb do not always work What is the “IC50 Rule of Thumb”: IC50 for a competitive inhibitor should increase about 10 going from [ATP] = Km to physiological [ATP]. What is “Tight Binding”: Experimental conditions where the enzyme concentration is comparable with the inhibition constant. An important fact: The “Rule of Thumb” does not apply under the conditions of “Tight Binding”. How do we know this: Many journal articles on “Tight Binding” : Morrison (1969), Cha (1974), ..., Kuzmic (2000, 2003, 2011) IC50's are bad for you

Final word about the “fold increase in IC50” plot: Do not use it. SOMETIMES IT WORKS AND SOMETIMES IT DOESN’T. NO WAY OF TELLING IN ADVANCE IF IT WILL. [E]0 = 0.25 nM [E]0 = 66 nM SAME INHIBITOR – DIFFERENT ASSAY CONDITIONS IC50's are bad for you

Other people still use the “fold increase in IC50” plot A RECENT PAPER FROM NOVARTIS What is the main difference between Novartis and ArQule? Chene, P. et al. (2009) BMC Chem. Biol. 9:1, doi:10.1186/1472-6769-9-1 IC50's are bad for you

Review: Measures of inhibitory potency THE INHIBITION CONSTANT IS AN INTRINSIC MEASURE OF POTENCY: DG = -RT log K i Depends on [S] [E] Example: Competitive inhibitor DEPENDENCE ON EXPERIMENTAL CONDITIONS 1. Inhibition constant 2. Apparent K i 3. IC50 NO YES NO YES K i K i* = K i (1 + [S]/KM) IC50 = K i (1 + [S]/KM) + [E]/2 Points to make: Ki is the best, but laborious to get (need to vary both [S] and [I]). Kiapp is the next best. It "costs" as much as getting IC50 (vary [I] only), but is more informative than IC50 (does not depend on [E]). Of these three, IC50 is the least appropriate measure of potency, because... ... for tight binding inhibitors, IC50 shifts with [E]! That means, with a new protocol, a new staff person who changes [E] for some reason, in a different lab, etc., everyone will be getting a different IC50 for the same tight binding inhibitor (e.g., staurosporine) under otherwise identical conditions! Remember, "tight binding" has nothing to do with any particular value of Ki (picomolar, nanomolar, etc.). "Tight binding" is not a property of an inhibitor, but rather a characteristic of experimental conditions: whenever [E] is numerically close to Ki. For example, an inhibitor with Ki = 1 M is "tight binding" when [E] = 500 nM (this could happen, if your substrate is really slow). [E] « K i: IC50  K i* "CLASSICAL" INHIBITORS: [E]  K i: IC50  K i* "TIGHT BINDING" INHIBITORS: IC50's are bad for you

Summary: IC50’s are bad for your drug discovery efforts You can hit the “IC50 Wall” You could be looking at picomolar Ki inhibitors and yet they would look “nanomolar” to you, if you are screening at nanomolar [E]. A thousand-fold difference in true vs. “apparent” potency. You could get very confused about the “mode” of inhibition A competitive inhibitor can give you almost no “fold increase” in IC50 if the experiment is done under tight binding conditions. Muddled communication channels within your enterprise The question “What is the biochemical IC50 for compound X?” does not make sense for a competitive inhibitor, without specifying [ATP] level. Don’t use IC50 as measure of potency in biochemical assays. IC50’s are perfectly good for cell-based assays. IC50's are bad for you

What else is there, other than the IC50? Use intrinsic molecular measures of potency (Ki, ...) IC50's are bad for you

Example: Correlation between kinact/Ki and cellular IC50 “Molecular Underpinnings of Covalent EGFR Inhibitor Potency [...]” SUBMITTED inhibition of EGFR-L858R/T790M autophosphorylation in H1975 tumor cells cell-based potency Points to make: Ki is the best, but laborious to get (need to vary both [S] and [I]). Kiapp is the next best. It "costs" as much as getting IC50 (vary [I] only), but is more informative than IC50 (does not depend on [E]). Of these three, IC50 is the least appropriate measure of potency, because... ... for tight binding inhibitors, IC50 shifts with [E]! That means, with a new protocol, a new staff person who changes [E] for some reason, in a different lab, etc., everyone will be getting a different IC50 for the same tight binding inhibitor (e.g., staurosporine) under otherwise identical conditions! Remember, "tight binding" has nothing to do with any particular value of Ki (picomolar, nanomolar, etc.). "Tight binding" is not a property of an inhibitor, but rather a characteristic of experimental conditions: whenever [E] is numerically close to Ki. For example, an inhibitor with Ki = 1 M is "tight binding" when [E] = 500 nM (this could happen, if your substrate is really slow). kinact Ki IC50's are bad for you

Ki(app) takes into account “tight binding” IT IS CLOSE TO BEING AN INTRINSIC, MOLECULAR MEASURE OF POTENCY Depends on [S] [E] Example: Competitive inhibitor DEPENDENCE ON EXPERIMENTAL CONDITIONS 1. Inhibition constant 2. Apparent K i 3. IC50 NO YES NO YES K i K i* = K i (1 + [S]/KM) IC50 = K i (1 + [S]/KM) + [E]/2 Points to make: Ki is the best, but laborious to get (need to vary both [S] and [I]). Kiapp is the next best. It "costs" as much as getting IC50 (vary [I] only), but is more informative than IC50 (does not depend on [E]). Of these three, IC50 is the least appropriate measure of potency, because... ... for tight binding inhibitors, IC50 shifts with [E]! That means, with a new protocol, a new staff person who changes [E] for some reason, in a different lab, etc., everyone will be getting a different IC50 for the same tight binding inhibitor (e.g., staurosporine) under otherwise identical conditions! Remember, "tight binding" has nothing to do with any particular value of Ki (picomolar, nanomolar, etc.). "Tight binding" is not a property of an inhibitor, but rather a characteristic of experimental conditions: whenever [E] is numerically close to Ki. For example, an inhibitor with Ki = 1 M is "tight binding" when [E] = 500 nM (this could happen, if your substrate is really slow). What do we need to do to move from IC50 to Ki(app)? IC50's are bad for you

The most “obvious” method to get Ki(app) will not work THIS DOES NOT WORK Could we get the IC50 by our usual method, and then just subtract one half of [E]0? Not in general: - This would work very well for non-tight situations, [E]0 << Ki(app) - In non-tight situation we have IC50 = Ki(app) - However under tight-binding the Ki(app) could become “negative” if our enzyme concentration is lower than we think it is. Points to make: Ki is the best, but laborious to get (need to vary both [S] and [I]). Kiapp is the next best. It "costs" as much as getting IC50 (vary [I] only), but is more informative than IC50 (does not depend on [E]). Of these three, IC50 is the least appropriate measure of potency, because... ... for tight binding inhibitors, IC50 shifts with [E]! That means, with a new protocol, a new staff person who changes [E] for some reason, in a different lab, etc., everyone will be getting a different IC50 for the same tight binding inhibitor (e.g., staurosporine) under otherwise identical conditions! Remember, "tight binding" has nothing to do with any particular value of Ki (picomolar, nanomolar, etc.). "Tight binding" is not a property of an inhibitor, but rather a characteristic of experimental conditions: whenever [E] is numerically close to Ki. For example, an inhibitor with Ki = 1 M is "tight binding" when [E] = 500 nM (this could happen, if your substrate is really slow). IC50's are bad for you

Problem: Negative Ki from IC50 FIT TO FOUR-PARAMETER LOGISTIC: K i* = IC50 - [E] / 2 Points to make: Can't fix the problem by the "naïve" method: 1. Measure IC50 as usual (four-parameter logistic) 2. Subtract [E]/2 from resulting IC50. PROBLEM! The Kiapp is negative, which is a nonsensical result. We never know exactly what the enzyme concentration really is, so that's why the above "naïve" method fails. __________________________ If there is time: Physical meaning of parameters in the four-parameter logistic equation: The Hill exponent (nHill = 1.4 in this case) has not physical meaning at all for a tight binding inhibitor. We do know independently that this compound is very "clean" mechanistically, with exactly 1:1 stoichiometry, a pure competitive inhibitor. The value nHill = 1.4 simply means that the original Hill equation (transformed into the "four parameters logistic") was derived under the assumption that no tight binding occurs! In other words, the basic assumptions of the four- parameter logistic equation are violated. It should not be used for tight binding. Data courtesy of Celera Genomics IC50's are bad for you

Solution: Do not use four-parameter logistic equation FIT TO MODIFIED MORRISON EQUATION: P. Kuzmic et al. (2000) Anal. Biochem. 281, 62-67. P. Kuzmic et al. (2000) Anal. Biochem. 286, 45-50. Points to make: Here is how to fix the problem: - forget about IC50's completely - forget about the four-parameter logistic equation (4PLE) Instead of 4PLE, use a modification of the Morrison equation (paper #1 above). This will get you the Kiapp directly (it is, of course, positive). As a bonus, you may learn what your enzyme concentration really was. The conditions under which this simultaneous determination of [E] and Kiapp is possible are described in paper #2 above. In this case, the enzyme appears about 65% active (4.5 nM actual concentration, as opposed to 7.0 nM nominal concentration). Data courtesy of Celera Genomics IC50's are bad for you

Surprise #1: Getting the Ki(app) does not require any additional experiments We just need to change the way the “IC50 data” are analyzed. IC50's are bad for you

The “Morrison Equation” for tight binding inhibition v dependent (“Y”) and independent (“X”) variables v initial reaction rate (“Y” axis) [I] inhibitor concentration (“X” axis) V0 model parameters [E] enzyme concentration V0 control / uninhibited rate ([I] = 0) K* apparent inhibition constant [I] [E] IC50's are bad for you

Caveat: Sometimes we need to fit the same data twice PROBLEM: WE NEVER QUITE KNOW WHAT THE ENZYME CONCENTRATION REALLY IS Fit data to the Morrison Equation while keeping [E]0 constant  Ki(app) Algorithm: Ki(app) < [E]0 ? NO YES Fit the same data to the Morrison Equation while “floating” [E]0  improved Ki(app), [E]0 Report Ki(app) and optionally also [E]0 Kuzmic, P. et al. (2000) Anal. Biochem. 286, 45-50. IC50's are bad for you

Surprise #2: Ki(app) can be guessed from a single concentration plus control IC50's are bad for you

The "single-point" method for Ki(app) AN APPROXIMATE VALUE OF THE INHIBITION CONSTANT FROM A SINGLE DATA POINT Relative rate Vr = V/V0 "control" V0 Ki = 12 nM Ki = 9 nM Ki = 11 nM Ki = 8 nM Single-point formula: V [I] Kuzmic et al. (2000) Anal. Biochem. 281, 62–67 IC50's are bad for you

Weighted average to make the initial estimate of Ki(app) DATA POINTS VERY NEAR “TOP” AND “BOTTOM” ARE LESS TRUSTWORTHY weighted average Error Propagation Theory Weight: IC50's are bad for you

Estimate K*: Example Weighted average. 1 2 3 4 5 6 7 8 [I]max = 50 µM 1 2 3 4 5 6 7 8 Weighted average. [I]max = 50 µM 4:1 dilution series 8 points + control IC50's are bad for you

Surprise #3: We don’t need “fully developed” IC50 curves ... when using the Morrison Equation for Ki(app) IC50's are bad for you

It’s OK to have at most 20%-30% inhibition if necessary HIGHLY PRECISE Ki(app) EVEN WITH LESS THAN 50% INHIBITION AT MAXIMUM [INHIBITOR] [I]max = 50 µM 4:1 dilution series 8 points + control Ki(app) = (43 ± 2) µM IC50's are bad for you

Can we do even better than Ki(app)? IC50's are bad for you

Getting the “true” Ki from a single dose-response curve IF POSSIBLE, SCREEN AT [SUBSTRATE] MUCH LOWER THAN THE MICHAELIS CONSTANT Depends on [S] [E] DEPENDENCE ON EXPERIMENTAL CONDITIONS Competitive Inhibitor 1. Inhibition constant 2. Apparent K i 3. IC50 NO YES NO YES K i K i* = K i (1 + [S]/KM) IC50 = K i (1 + [S]/KM) + [E]/2 At [S] << KM ... Ki(app) ≈ Ki IC50's are bad for you

Getting the “true” Ki from a single dose-response curve FOR “NONCOMPETITIVE” INHIBITORS Ki(app) = Ki(true) ALWAYS Depends on [S] [E] DEPENDENCE ON EXPERIMENTAL CONDITIONS Noncompetitive Inhibitor 1. Inhibition constant 2. Apparent K i 3. IC50 NO NO YES K i K i* = K i IC50 = K i (1 + [S]/KM) + [E]/2 IC50's are bad for you

Summary and Conclusions Biochemical IC50’s are misleading in multiple ways. They are a relic of previous eras (1950-1980) in pre-clinical research. Even Big Pharma is now gradually moving toward Ki(app) and Ki. Small to medium-size companies should not “follow the Big Guys”. They should lead. IC50's are bad for you