Download NEW MODEL OF NONCLINICAL CARDIAC RISK ASSESSMENT

NEW MODEL OF NONCLINICAL CARDIAC RISK ASSESSMENT
Kramer JK1, Myatt, GJ.2; Obejero-Paz, C.1, Bruening-Wright, A.1 Brown AM. 1ChanTest Corp., Cleveland, OH, USA.
2Leadscope Inc., Columbus, OH, USA
2.) hERG Is Not The Best Predictor of TdP When Drug Affects Multiple APD30 Shortened
0
APD30 is Shortened
1.E+00
-30
1.E‐01
hERG IC50
1.E‐02
0.9
APD Prolonged
-60
hERG IC50
1.E‐01
Prolonged
-80
-90
0s
1.E‐03
400 ms
0s
1.E‐03
Terfenadine
APD60 is Unchanged
APD90
Prolonged
APD
90 is ETPC
1.E‐02
ETPC
-40
Cav1.2 IC50
400 ms
800 ms
Verapamil
True Positive Ratee (sensitivity)
Ion Chann
nel IC50 (μM)
Ion Chan
nnel IC50 (μM)
0
Contingency Table
40
1.E+01
Nav1.5 IC50
Model Based on Multiple Ion Channel Effects Is More Predictive of TdP 1.0
30
Cav1.2 IC50
1.E+00
4.)
Verapamil Blocks hERG and Cav1.2 Verapamil Blocks hERG and Cav1.2 and has Mixed Effects on SC‐HCM AP
and has Mixed Effects on SC‐HCM AP
1.E+02
Nav1.5 IC50
1.E+01
0.8
0.7
Model 1
0.6
0.5
0.4
No Discrimination; AUC = 0.5
0.3
Model 2; AUC = 0.88
0.2
0.1
Model 2
0.0
Many torsadogenic drugs potently and selectively inhibit hERG ion channel current.
current However,
However hERG
blockers that have effects on multiple ion channels, are not always torsadogenic. Verapamil, which is
used to treat high blood pressure and angina, is a multiple ion channel blocker that inhibits hERG very
potently (IC50 = 143 nM). Verapamil inhibits the two major depolarizing currents ICaL and INa
offsetting the hERG block and avoiding delayed repolarization. There are no reported incidents of TdP
since it has been on the market. Compounds that have multiple ion channel effects should be
identified as early as possible during drug development.
+TdP and –TdP Drugs Characterized by Their Ion Channel g
y
IC50/ETPC Profiles
3.))
Torsadogenic Compounds
1000000
100000
100000
10000
10000
Vanoxerine
Risperidone
Ranolazine
Quetiapine
Propranolol
Propafenone
Phenytoin
Pentobarbital
Nitrendipine
Nifedipine
Mibefradil
Mexiletine
Methadona
Diltiazem
Diazepan
Desipramine
0.001
Cibemzoline
0.01
Amitriptyline
Ziprasidone
Thioridazine
Terfenadine
Tedisamil
Sotalol
Sertindole
Sematilide
Quinidine
Prenilamine
Pimozide
Moxifloxacin
Ibutilide
Haloperidol
Defetilide
Cisapride
Bepridil
0.1
Astemizole
0.001
0.4
0.6
0.8
False Positive Rate (1‐specificity)
False Positive Rate 1.0
We created two predictive logistic regression models: The first model was developed with hERG IC50/ETPC data and
the second with hERG IC50/ETPC + Nav1.5 IC50/ETPC + Cav1.2 IC50/ETPC data. Cross validation of each model was
performed to classify drugs as being +TdP or ‐TdP. The first logistic regression model using only hERG data poorly
classified the drugs, displaying an overlapping of the +TdP and –TdP groups, 11 false positives and 1 false negative.
In contrast, the second model effectively separated the +TdP and –TdP groups and resulted in only 3 false positives
and 2 false negatives.
negatives
1
Imipramine
0.1
0.01
0.2
Conclusions:
10
Fluvoxamine
1
TdP‐
3
False Positives
18
True Negatives
100
Diphenhydramine
30 fold greater than ETPC
IC50/ETPC
10
TdP+
18
True Positives
2
False Negatives
30 fold greater 30
f ld
t
than ETPC
1000
100
Verapamil
1000
TdP‐
11
False Positives
10
True Negatives
Receiver Operator Characteristic (ROC) for Model 1 (green circles) and Model 2 (blue circles) were constructed by
plotting the false positive rate (false positives )/(false positives + true negatives)) and true positive rate (true
positives)/(true positives +false negatives)) at each cross‐validated probability (shown in the box plots above). The
solid line is the theoretical result for a test that does not discriminate between the two groups (the area under the
curve (AUC) = 0.5). The logistic regression model derived from hERG IC50/ETPC alone is not a good TdP predictor
since a 90% True Positive Rate is associated with a 52% False Positive Rate (AUC = 0.74). Encouragingly, the
inclusion of Nav1.5 and Cav1.2 parameters results in a significant increase in predictivity by reducing the False
Positive Rate to 14% while maintaining a True Positive Rate of 90% (AUC = 0.88).
Non‐Torsadogenic Compounds
1000000
TdP+
19
True Positives
1
False Negative
Model 1; AUC = 0.74
0.0
Amiodarone
The challenge is to identify drugs early in development that may
cause torsade de pointes (TdP), a serious polymorphic ventricular
y
that can lead to sudden death. The frequency
q
y of TdP mayy
tachycardia
be <10‐6 for non‐cardiac drugs cannot be reliably detected in clinical
trials. The risk of sudden death may become apparent only during
marketing surveillance. QT prolongation has been chosen as a
surrogate marker for TdP because TdP is always associated with a
prolonged QT interval but the QT‐TdP linkage is incomplete and the
predictivity is weak. Impaired predictivity remains even under the
minimal safety strategy laid out in the ICH S7B industry guidance.
This guidance for evaluating delayed repolarization emphasizes
testing for inhibition of the hERG potassium channel and QT
prolongation in an in vivo assay.
assay However,
However the S7B preclinical cardiac
safety strategy may give internally discordant results and imperfect
clinical predictivity making it difficult to determine a compound’s TdP
risk. Discordances may be resolved and TdP predictivity increased by
following an improved thorough cardiac safety testing strategy that
determines effects on multiple cardiac ion channel currents. In this
study we demonstrate that characterizing the effects of drugs on
Nav1.5 and Cav1.2, in addition to hERG, better predicts whether a
drug is torsadogenic.
Terfenadine Selectively Blocks hERG Terfenadine Selectively Blocks and Prolongs SC‐HCM AP
hERG and Prolongs SC‐HCM AP
1.E+02
Ajmaline
Introduction:
Ion Channels
IC50/ETPC
Abstract
Drug‐induced inhibition of the cardiac hERG potassium channel is assumed to
predict delayed cardiac repolarization (DR). The consequent QTc prolongation is
a surrogate marker of torsade de pointes (TdP), a potentially lethal iatrogenic
outcome. Drugs with effective therapeutic plasma concentrations (ETPC) within
30‐fold of their hERG IC50s are thought to be dangerous despite the fact that
multiple ion channel effects (MICE) can mitigate DR. Here we demonstrate that a
logistic regression model, which integrates MICE, predicts TdP with much greater
certainty than the hERG safety ratio (hERG IC50/ETPC) alone. Safety ratios were
calculated for 41 drugs (20 +TdP and 21 ‐TdP)
TdP) from multiple classes by dividing
their hERG, Nav1.5 and Cav1.2 IC50 values by each drug’s ETPC. Two logistic
regression models were constructed; one using the hERG IC50/ETPC ratio alone
(Model 1), the other integrating hERG IC50/ETPC + Nav1.5 IC50/ETPC + Cav1.2
IC50/ETPC data (Model 2). The predictive power of each model was evaluated by
performing leave‐one‐out cross validations. Each model’s accuracy for
discriminating +TdP and ‐TdP drugs was determined by comparing their
receiver–operator characteristics (ROC, true vs. false positive rates). Model 1 had
a 52% False Positive Rate associated with a 90% True Positive Rate and a ROC
area under the curve (AUC) of 0.74. Model 2 significantly improved accuracy
showing a 14% False Positive Rate associated with a 90% True Positive Rate and
a ROC AUC of 0.88. We propose that models that incorporate quantitative drug
effects on multiple cardiac ion channels will be robust nonclinical predictors of
cardiac risk.
1) Some hERG blockers (IC50/ETPC values lower than 30) are safe drugs when they also block calcium and/or
sodium currents.
2) Logistic regression models that include multiple ion channels are better predictors of cardiac risk than models
including hERG block alone.
1.) The ECG and Cardiac Action Potential are Formed by Many Different Ion Channel Currents.
Cardiac Channel Panel™
Nav1.5 (INa)
Phase 1
Cav1.2/β2, α2δ, (L‐type)
0
Cav3.2 (T‐type)
Phase 2
Kv4.3 (ITO1)
‐30
30
Phase 0
Membrane P
Potential (mV)
30
Phase 3
‐60
Phase 4
‐90
0 s
100 ms
200 ms
300 ms
Time
400 ms
500 ms
AP Phase
0, 2
2
1
1
KvLQT1/minK (I
Kv
QT /minK (IKs)
2 ‐ 3
hERG (IKr)
2 ‐ 3
Kv1.5 (IKur)
2 ‐ 3
Kir2.1 (IK1)
4
HCN2 (pacemaker, If)
4
HCN4 (pacemaker, If)
4
Kir3.1/3.4 (IK,ACh)
4
Kir6.2/SUR2A (IK,ATP)
4
NCX1 (Na‐Ca exchange)
2
Ion channel (hERG, Cav1.2 and Nav1.5) IC50 values for 41 drugs were collected from the literature and
ChanTest’s internal database. The average Effective Therapeutic Plasma Concentrations (ETPC) were
collected from the literature or calculated from Cmax and % protein binding values. In the
logarithmic plots above, the hERG IC50/ETPC value is the vertex of each line segment, the Nav1.5
IC50/ETPC value is the point at the end of each blue line segment and the Cav1.2 IC50/ETPC is the point
at the end of each orange segment. Drugs were characterized as torsadogenic (left panel) or non‐
torsadogenic (right panel) according to information found in Redfern et al. (Categories 1‐3 = +TdP;
C
Categories
i 4‐5
4 5 = ‐TdP),
TdP) the
h Arizona
Ai
CERT database
d b
and
d reported
d cases. The
Th dashed
d h d line
li shows
h
Redfern’s criteria of 30 for torsadogenic drugs. One can see that not all of the torsadogenic drugs
have hERG IC50 values that are less than 30 fold above the ETPC and some non‐torsadogenic drugs
have hERG IC50/ETPC ratios that are less than 30. The plots above reveal that with very few
exceptions, the +TdP and –TdP drugs can be qualitatively distinguished from each other by their multi
ion channel IC50/ETPC profiles. In general safe drugs have IC50/ETPC values for the depolarizing
currents close (within one order of magnitude) to hERG IC50/ETPC values.
3) Drug misclassification is most probably due to inadequacy of TdP classification or drug effects on other channels
involved in cardiac excitability.