Translating QT interval prolongation from conscious dogs to humans

Abstract

Aim

In spite of screening procedures in early drug development, uncertainty remains about the propensity of new chemical entities (NCEs) to prolong the QT/QTc interval. The evaluation of proarrhythmic activity using a comprehensive in vitro proarrhythmia assay does not fully account for pharmacokinetic–pharmacodynamic (PKPD) differences in vivo. In the present study, we evaluated the correlation between drug‐specific parameters describing QT interval prolongation in dogs and in humans.

Methods

Using estimates of the drug‐specific parameter, data on the slopes of the PKPD relationships of nine compounds with varying QT‐prolonging effects (cisapride, sotalol, moxifloxacin, carabersat, GSK945237, SB237376 and GSK618334, and two anonymized NCEs) were analysed. Mean slope estimates varied between −0.98 ms μM^–1 and 6.1 ms μM^–1 in dogs and −10 ms μM^–1 and 90 ms μM^–1 in humans, indicating a wide range of effects on the QT interval. Linear regression techniques were then applied to characterize the correlation between the parameter estimates across species.

Results

For compounds without a mixed ion channel block, a correlation was observed between the drug‐specific parameter in dogs and humans (y = −1.709 + 11.6x; R ² = 0.989). These results show that per unit concentration, the drug effect on the QT interval in humans is 11.6‐fold larger than in dogs.

Conclusions

Together with information about the expected therapeutic exposure, the evidence of a correlation between the compound‐specific parameter in dogs and in humans represents an opportunity for translating preclinical safety data before progression into the clinic. Whereas further investigation is required to establish the generalizability of our findings, this approach can be used with clinical trial simulations to predict the probability of QT prolongation in humans.

What is Already Known about this Subject

• The use of model‐based approaches for the screening of molecules in safety pharmacology is limited.

• A previous investigation has shown the feasibility of characterizing pharmacokinetic–pharmacodynamic (PKPD) relationships as the basis for evaluating the risk of QT prolongation in dogs and humans.

• The magnitude of drug effects on the QT/QTc interval varies across species as a result of the underlying differences in pharmacokinetics, pharmacodynamics and homeostasis.

• In spite of extensive screening procedures in early drug discovery, uncertainty remains about the propensity of non‐antiarrhythmic drugs in prolonging the QT/QTc interval in humans.

What this Study Adds

• We showed evidence of the correlation between PKPD model parameters describing the drug‐induced QT‐prolonging effect in preclinical species and humans.

• Drug‐specific model parameter estimates obtained in dogs can be used to predict drug effects in humans, reducing the attrition due to false‐positive and false‐negative results during the screening of candidate molecules.

• The interspecies difference in the slope of the correlation between drug concentration and QTc interval prolongation may provide the basis for go/no go decisions regarding the progression of novel drug candidate molecules into clinical development.

Tables of Links

TARGETS
Voltage‐gated ion channels 2
Cav1.2
Kv11.1
Nav1.8

LIGANDS
cisapride	sotalol

These Tables list key protein targets and ligands in this article that are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY 1, and are permanently archived in the Concise Guide to PHARMACOLOGY 2015/16 2.

Introduction

An important matter of concern in drug development is the proarrhythmic potential of pharmaceutical compounds 3. A number of drugs have had to undergo labelling revision or market withdrawal owing to postmarketing reports of sudden cardiac death linked to the development of torsades de pointes (TdP) 4, 5, 6, 7, 8, 9, 10. It has been demonstrated that an excessive lengthening of cardiac repolarization (measured by QT interval prolongation) may induce these life‐threatening ventricular tachyarrhythmias 11.

A wide range of drugs from various therapeutic classes has been associated with QT prolongation and TdP, including both antiarrhythmic drugs and non‐antiarrhythmic cardiovascular drugs (n = 20), and non‐cardiovascular drugs (n = 50). Examples include antihistamine, antipsychotic, antidepressant, antifungal, anti‐infective and gastrointestinal prokinetic drugs 6, 12, 13, 14. However, predicting the risk of these types of serious side effects has proven a major challenge in cardiac safety pharmacology, as not all drugs showing QT‐prolonging effects are associated with TdP. In fact, a new proposal is being considered which shifts the focus from evaluating QT prolongation to evaluating proarrhythmic activity using a comprehensive in vitro proarrhythmia assay (CiPA), but this new approach does not take into account the underlying pharmacokinetic–pharmacodynamic (PKPD) relationships and overlooks the implications of differences between in vitro and in vivo experimental protocols 15, 16.

Whereas the characterization of the relationship between drug concentration and QT interval prolongation in vivo in preclinical species may not be considered a surrogate for the risk of TdP, such protocols can be informative and provide the basis for predicting drug effects in humans in a strictly quantitative manner. In a previous investigation, in which a general PKPD model was used to assess the QT‐prolonging effects of reference compounds, we showed that there are differences in the concentration of moxifloxacin associated with the probability of a ≥10 ms increase in QT(c) prolongation between dogs, monkeys and humans 17. In contrast to data‐driven approaches, our method relies on a PKPD model with a generic parameterization, which disentangles drug‐specific properties from biological or physiological system properties, enabling extrapolation of the estimates across species. In addition, the approach readily allows for the incorporation of historical data on system‐related parameters describing circadian variability as well as the effect of heart rate on the QT interval. Given the Bayesian nature of the analysis, it also offers the possibility to estimate posterior parameter distributions, which reflect all acknowledged sources of uncertainty 18. Here, we attempted to demonstrate the feasibility of establishing an interspecies correlation between drug‐specific parameter estimates, which can be used subsequently as a scaling factor or predictor of the clinical effects before candidate molecules enter clinical development.

As shown in Figure 1, decisions about the progression of a candidate molecule during the drug development path demand a good understanding of arrhythmogenic signals. In the current paper, an integrated approach is proposed, along the same principles as suggested by Lowe et al. 19 and Pollard et al. 20. According to these authors, to scale between in vivo species, multiple compounds need to be tested, where quantifiable differences can be detected. This implies an iterative process between models and experiments 19, 20. Similarly, such an iterative process can be applied when considering findings in preclinical species and humans. For the scaling of drug effects on the QT interval from preclinical species to humans, a correlation can be derived and continuously updated as new compounds progress to clinical development. Assuming that differences in QT‐prolonging effects across species reflect varying degrees of target engagement (activation or inhibition) and/or homeostatic mechanisms, one could use such a correlation to predict the magnitude of the drug effects in humans as well as to improve the design of clinical study protocols aimed at the characterization of QT‐prolonging effects.

Figure 1.

Opportunities (indicated by stars) where pharmacokinetic–pharmacodynamic (PKPD) modelling can help in decision making in cardiovascular safety. The possibility of discriminating between system‐ and drug‐specific properties allows preclinical data to be used as basis for translating drug effects from animals to humans. Adapted from Pollard et al. 20. FTIM, first‐time‐in‐humans study; hERG, human Ether‐a‐go‐go Related Gene; TQT, Thorough QT study; ECG, Electrocardiogram

We envisage, therefore, that model‐predicted estimates can provide the basis for go/no‐go decisions before taking the compound into clinical development, while reducing attrition due to false‐positive and false‐negative results, as often observed in the analysis of individual experimental protocols.

Methods

Data

Data on the QT interval, heart rate and plasma concentrations of nine different compounds were collected from multiple partners within the safety pharmacology workgroup of the TI‐Pharma PKPD platform. All available data were used for the purpose of PKPD modelling. An overview of the available data and the study designs is presented in Table 1. Whereas no predefined selection criteria were applied to the compounds included in this analysis, the data set consisted of compounds with different mechanisms of action – namely, moxifloxacin (DNA gyrase inhibitor), d,l‐sotalol (nonselective β‐adrenoreceptor antagonist), cisapride (5‐HT₄ receptor agonist), carabersat (SB204269, a benzopyran derivative with an undefined central nervous system binding site), SB237376 (potassium and calcium channel blocker), GSK945237 (topo‐isomerase II inhibitor), GSK618334 (dopamine D₃ receptor antagonist), NCE03 and NCE04 (unknown mechanism of action) 21.

Table 1.

Summary of protocol and study designs used for pharmacokinetic–pharmacodynamic modelling of the QT‐prolonging effects in dogs and humans

Drug	Mechanism of action	Species	Sample size	Dosing regimen	Route	Dose	PK sampling (h)	PD measurement	PD sampling (h)
Moxifloxacin	DNA gyrase inhibitor	Beagle dogs	N = 8	Single dose min. 1 week between changes of dose	Oral gavage	10, 30, 100 mg kg–1	0, 0.5, 1, 2, 4, 8, 24, 36, 48	Implanted telemetry measurements: blood pressure ECG	Every 1 min, averages over 48 h
Moxifloxacin	DNA gyrase inhibitor	Humans (healthy subjects)	F = 88 M = 49	Single dose	Oral	Placebo, 400 mg	–1, −0.5, −0.83, 0.25, 0.5, 1, 1.5, 2, 2.5, 3, 4, 6, 8, 10, 12, 24	–	–1, −0.5, −0.83, 0.25, 0.5, 1, 1.5, 2, 2.5, 3, 4, 6, 8, 10, 12, 24
Sotalol	Nonselective β‐adrenoreceptor antagonist	Beagle dogs	N = 6	Single dose min. 1 week between changes of dose	Oral gavage	Vehicle, 4, 8 mg kg–1	0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 6, 8, 24	Implanted telemetry measurements: blood pressure ECG	Every 5 min, averages over 24 h
Sotalol	Nonselective β‐adrenoreceptor antagonist	Humans (healthy subjects)	F = 12 M = 18	Single dose	Oral	Placebo, 160 mg	–0.5, −0.25, 0.83, 0.25, 0.5, 0.75, 1, 2, 4, 8, 10, 18, 24	–	–2, −1.75, −1.5, −1, −0.75, −0.5, −0.25, 0.5, 1, 1.25, 2, 4, 6, 8, 10, 12, 18, 24
Cisapride	5‐HT4‐receptor agonist	Beagle	N = 8	Single dose min. 1 week between changes of dose	Oral gavage	Vehicle, 0.6, 2, 6 mg kg–1	0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 5, 6, 8, 24	Implanted telemetry measurements: blood pressure, ECG	Every 30 s, averages over 24 h
Cisapride	5‐HT4‐receptor agonist	Humans (healthy subjects)	F = 10 M = 14	Dose escalation	Oral	Placebo, 10, 20, 40, 80, 120 mg	0, 1, 1.5, 2, 3, 4, 6, 12, 24	–	–24, −23, −22.5, −21, −20, −18, −12, 0, 1, 1.5, 2, 3, 4, 6, 12, 24
Carabersat (SB204269)	Benzopyran derivative	Beagle dogs	N = 4	Single dose min. 1 week between changes of dose	Oral	Vehicle, 10, 30, 100, 1000 mg kg–1	0, 0.5, 1, 2, 4, 8, 24, 36, 48	Implanted telemetry measurements: blood pressure , ECG	Every 30 min, for 20 h
Carabersat (SB204269)	Benzopyran derivative	Humans (healthy subjects)	M = 35	–	Oral	Placebo, 100, 200, 400, 800, 1600, 2800, 4000, 5000 mg	0, 0.5, 1, 1.5, 2, 3, 4, 6, 8, 10, 12, 24, 30, 48	–	0, 0.5, 1, 1.5, 2, 3, 4, 5, 6, 8, 10, 12, 24
NCE03	NA	Beagle dogs	N = 4	Single dose min. 1 week between changes of dose	Oral	Vehicle, 2.15, 4.3 mg kg–1	Predose, 0.5, 1, 1.5, 2, 6, 24	Implanted telemetry measurements: blood pressure, ECG	–1, –0.75, –0.5, –0.25, 0, 0.5, 1, 1.5, 2, 1.5, 2, 2.5, 3, 4, 5, 6, 10, 12, 16, 20
NCE03	NA	Humans	M = 29	–	Oral	Placebo, 10, 30, 70, 90, 180, 360, 430, 500 mg	0, 0.33, 0.67, 1, 1.33, 1.67, 2, 2.5, 3, 4, 5, 6, 8, 12, 24, 36, 48	–	–1, 0.17, 0.33, 0.5, 0.67, 0.83, 1, 1.25,1.5, 1.75, 2, 2.5, 3, 3.5, 4, 6, 8, 10, 12, 24, 36
NCE04	NA	Beagle	N = 6	Single dose min. 1 week between changes of dose	Oral/s.c.	Vehicle, 4, 20, 100 mg kg–1	0, 1, 3, 6, 17, 24	Implanted telemetry measurements: blood pressure, ECG	0.5, 1, 2, 3, 4, 6, 8, 12, 16, 20, 24
NCE04	NA	Humans	M = 64	–	Oral/s.c.	Placebo, 3, 6, 12, 24, 48, 95, 190 mg	0, 0.33, 0.67, 1, 1.33, 1.67, 2, 2.5, 3, 4, 5, 6, 8, 12, 24, 36, 48	–	Frequent sampling up to 2.5 h then at, 3, 3.5, 4, 6, 8, 12, 24
SB237376	Potassium‐calcium channel blocker	Beagle dogs	N = 4	Single dose min. 1 week between dose levels	Oral	0, 10 ,80 mg kg–1	–	Implanted telemetry measurements: blood pressure, ECG	–
SB237376	Potassium‐calcium channel blocker	Humans (healthy subjects)	F = 9 M = 30	Single and repeated dose	Oral	Placebo, 25 , 50 mg	–	–	–
GSK945237	Topo‐isomerase II inhibitor	Beagle dogs	F = 6 M = 6	Twice daily, half of dose, approx. 6 h between doses	Oral	Placebo, 30, 100, 300 mg kg–1 day–1	0, 0.25, 0.5, 1, 3, 6 6.25, 6.5, 7, 9, 24 (Days 1 and 14)	–	–
GSK945237	Topo‐isomerase II inhibitor	Beagle dogs	M = 4	Twice daily, half of dose, approx. 6 h between doses, min. 1 week between dose levels	Oral	30, 100, 300 mg kg–1 day–1	–	Implanted telemetry Measurements: HR, blood pressure, ECG, body temperature	0, 1, 2, 3, 4, 5, 7, 9, 10, 11, 12, 13, 15, 18, 21, 24
GSK945237	Topo‐isomerase II inhibitor	Humans (healthy subjects)	45	Single dose	Oral	(6/dose) 50, 250, 500, 1000, 1750 mg day–1	0, 1, 2, 3, 4, 8, 24, 48	12‐lead ECG, dual‐lead cardiac monitoring	0, 1, 2, 3, 4, 8, 24, 48
GSK618334	Dopamine D3 receptor antagonist	Beagle dogs	M = 9 F = 9	‐	Oral	2, 5, 15 mg kg–1	0.5, 1, 2, 4, 6, 8, 24	–	–
GSK618334	Dopamine D3 receptor antagonist	Beagle dogs	M = 4	Min. 6 days between doses	Oral	2 ,5, 15 mg kg–1	–	Implanted telemetry measurements: blood pressure ECG	0–24 h continuous ECG, every 30 s
GSK 618334	Dopamine D3 receptor antagonist	Humans (healthy subjects)	M = 20 Divided into 2 cohorts	Single ascending doses, min. 2 weeks between doses	Oral	Cohort 1: 2.5, 25, 100, 400 mg Cohort 2: 10, 50, 200, 600 mg	0, 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, 8, 10, 12, 16, 24, 48, 72, 96	Lead II ECG; continuous 12‐lead ECGs Measurements: blood pressure, HR, ECGs	Continuous from 0 to 6 0, 1, 2, 3, 4, 6, 12, 24, 48,

ECG, electrocardiogram; F, female; HR, heart rate; M, male; PD, pharmacodynamic; PK, pharmacokinetic; s.c., subcutaneous

Drug concentrations

Time‐matched concentration and QT interval values were required for the characterization of the PKPD relationships. When direct measurements were not available, individually predicted concentrations relative to each electrocardiogram (ECG) recording time were simulated or interpolated using either nonlinear mixed‐effects modelling techniques in NONMEM VII or VI (ICON, Gaithersburg, MD, USA) or deconvolution in WinNONLIN 4.2 (Pharsight Co., Cary, NC, USA). Pharmacokinetic models were validated based on graphical and statistical criteria, including goodness of fit (GOF) and normalized prediction and distribution errors (NPDE) where applicable. Details on the analysis of pharmacokinetic data for relevant compounds can be found elsewhere 17, 22, 23.

PKPD modelling

A previously published model based on an adaptation of the work of Piotrovsky et al. 24 was used to analyse QT interval and concentration data in dogs and humans independently. In contrast to numerous approaches where PKPD modelling has been used, this model relies on a common set of parameters to describe drug effects both in dogs and in humans. Model parameters, which discriminate between system‐ and drug‐specific properties, were estimated using WinBUGS version 1.4.3 25, 26 (see supporting information for model code). Details of the model can be found in Chain and Dubois et al. 22 and Dubois et al. 23. In brief, the model consists of three main components: (i) an individual correction factor to account for variability in the RR‐interval; (ii) an oscillatory function describing diurnal variation in QT; and (iii) a slope describing the linear relationship between concentration and drug effect on the QT interval 27. The model equation, including all these elements, is shown in Figure 2, along with a diagram illustrating the interspecies differences in the linear relationship between drug levels and QT interval prolongation.

Figure 2.

(A) Equation showing the three model components used to characterize drug‐induced QT prolongation in dogs and humans. QT _c0 (ms) is the individually corrected baseline QTc (gender is included as a covariate in humans); RR (s) is the interval between successive R peaks; α is the individual RR correction factor; A (ms) is the amplitude of the circadian rhythm; t is the clock time (decimal h); Ø (h) is the phase; and Slope is the drug‐specific parameter describing the relationship between drug concentration and QT interval. (B) and (C) illustrate how linear regression techniques were applied to the drug‐specific model parameters to define the interspecies correlation. PKPD, pharmacokinetic–pharmacodynamic

It should be noted that such a linear relationship seems to contrast with the theoretical views that PKPD relationships are best described by a sigmoidal function (e.g. Hill equation). However, linear relationships ensure that focus is given to the lower part of the concentration–effect relationship – i.e. the region of the curve which triggers clinical concern. Another practical limitation for the implementation of a sigmoidal function is the fact that safety considerations may prevent the estimation of maximum QT prolongation in humans. Usually, protocol stop criteria require subject withdrawal when the QTc interval exceeds 500 ms. Likewise, in preclinical studies adverse events often prevent accurate estimation of maximum prolongation 17, 22.

Model diagnostics

Details on the WinBUGS model code and evaluation procedures can be found elsewhere 22, 23. In brief, to assess the adequacy of the Bayesian PKPD model, two Markov Chain Monte Carlo (MCMC) chains 24 were run independently until at least 12 500 samples had been obtained. Estimates from these runs were subsequently pooled and summarized, not only in terms of their point estimates (population mean), but also as posterior distributions and credible intervals. In Bayesian statistics, the availability of posterior distributions allows direct comparison between model predictions and observed values. In addition, GOF and model performance criteria included the deviance information criterion 28 and chain convergence. This latter criterion was assessed visually by monitoring the dynamic traces of Gibbs iterations and numerically by computing the Gelman‐Rubin, Geweke, Raftery‐Lewis and Heidelberger‐Welch test statistics for all population parameters 28, 29, 30.

Interspecies correlation: linear and nonlinear regression

Linear and nonlinear regression methods were used to characterize the correlation between the estimates of the slope parameter (Slope) in dogs and humans (Figure 2). The ultimate goal of the analysis was to establish whether such a correlation can serve as a scaling factor for the differences between the two species and consequently enable the extrapolation of the drug effects on the QT interval from dogs to humans. During model selection, linear regression was prioritized to allow the evaluation of negative slopes, which reflects compounds with QT‐shortening activity. R2.12.3 was used for the purposes of the analysis, including statistical and graphical summaries. In addition to the regression coefficient (r²), the slope of the correlation was selected as the parameter of interest for subsequent evaluation of the predictive performance of the interspecies correlation.

Results

PKPD modelling of the QT interval

All model parameters used as input in the present analysis were derived from a PKPD model based on a generic parameterization of drug‐ and system‐specific properties. A summary of the parameter estimates per compound is presented in Tables 2 and 3. In addition to the typical Bayesian criteria for parameter convergence and model acceptance, graphical summaries using GOF plots (i.e. observed vs. predicted QT values) are shown in Figure 4 for each compound and species. These results are complemented by a graphical summary of the model performance – i.e. how well the PKPD model predicts the experimental data (Figure 5). Model predictions accurately describe the time course of the observed QT interval for the different compounds, both in dogs and humans.

Table 2.

System‐ specific parameter estimates for dogs () and humans (), where α is the RR correction factor; A is the amplitude; φ is the circadian oscillator; and QT _c0 is the corrected QT intercept, as defined by the model equation in Figure 2. N is the number of dogs/subjects in each experimental protocol or clinical trial. Population mean parameter estimates are shown, along with 95% credible intervals

		Moxifloxacin	Cisapride	Sotalol	NCE03	NCE04	GSK945237	SB237376	Carabersat	GSK618334
	N	8	8	6	4	6	4	4	4	4
	α	0.28 (0.22–0.35)	0.26 (0.2–0.33)	0.18 (0.11–0.3)	0.23 (0.14–0.38)	0.2 (0.31–0.48)	0.3 (0.22–0.4)	0.25 (0.17–0.34)	0.26 (0.17–0.42)	0.28 (0.17–0.46)
	A (ms)	4.6 (3.1–7.0)	5.6 (3.9–8.1)	6.6 (2.1–19.8)	9.2 (4.3–18.1)	4.3 (1.8–8.6)	8.6 (4.5–15.2)	7.4 (3.1–14.6)	4.2 (2.5–7.1)	2.13 (0.56–7.67)
	φ (h)	23.1 (15–35)	19.9 (16–26)	12.2 (7–36)	16.4 (10–26)	4.4 (0.6–9.1)	16.2 (9–28)	14 (10–19)	9 (5–16)	31.4 (16–55)
	QT c0 (ms)	240 (238–242)	238 (237–240)	255 (253–257)	244 (239–249)	261 (252–270)	258 (188–345)	246 (180–328)	250 (248–252)	246 (155–390)
	N	137	24	30	29	64	45	39	35	20
	α	0.4 (0.38–0.42)	0.18 (0.14–0.23)	0.27 (0.24–0.3)	0.3 (0.27–0.33)	0.22 (0.2–0.24)	0.22 (0.17–0.27)	0.33 (0.29–0.38)	0.24 (0.22–0.26)	0.28 (0.25–0.31)
	A (ms)	2.4 (1.7–2.9)	3.3 (1.1–6.0)	3.3 (2.4–4.3)	5.75 (2.6–10.6)	7.9 (6.8–9.3)	3.1 (2.1–4.2)	3.5 (1.9–5.3)	4.9 (3.8–6.3)	2.7 (1–4.6)
	φ (h)	10 (7–13)	4.3 (2.4–8.7)	6.22 (5.1–7.6)	28.2 (22.2–39.3)	4.4 (3.3–5.7)	7.2 (4.9–9.5)	9.2 (6.6–11.8)	9.7 (8.1–10.9)	8.3 (3.9–25.2)
	QT c0 (ms)	399 (394–403)	386 (382–390)	387 (383–392)	379 (371–386)	380 (378–382)	394 (360–431)	386 (371–402)	385 (379–392)	405 (377–435)

Table 3.

Drug‐specific parameter estimates for dogs () and humans (), where CP₅₀ = concentration associated with a 50% probability of a QT increase ≥10 ms. N is the number of dogs/subjects in each experimental protocol or clinical trial. Population mean parameter estimates are shown, along with 95% credible intervals

	Moxifloxacin	Cisapride	Sotalol	NCE03	NCE04	GSK945237	SB237376	Carabersat	GSK618334
N	8	8	6	4	6	4	4	4	4
N	137	24	30	29	64	45	39	35	20
Slope [ms μM–1]	0.56 (0.02–1.4)	4.5 (0.96–9.8)	1.9 (0.6–8)	6.1 (2.2–16)	–0.98 (−2.1–0.6)	0.0098 (−0.01–0.03)	0.092 (0.07–0.11)	0.64 (−0.91–4.3)	0.814 (−0.3–3.4)
Slope [ms μM–1]	3.9 (3.3–4.4)	90 (87–120)	21 (17–26)	70 (50–80)	–10 (−13––7)	0.0114 (0.008–0.02)	0.301 (0.297–0.304)	–0.2 (−1–0.7)	7.2 (5.3–9.2)
CP50	6.4	2.2	46.2	1.6	NA	5000	108.9	16.9	12.4
CP50	2.64	0.14	0.47	0.17	NA	4005	33.2	>9000	1.3

Figure 4.

Plots describing the goodness of fit for the pharmacokinetic–pharmacodynamic relationships. Individual predicted and observed QT interval for all nine compounds after administration to dogs (left) and humans (right). The panels show data for moxifloxacin (A and B), sotalol (C and D), cisapride (E and F), NCE03 (G and H), NCE04 (I and J), carabersat (K and L), SB237376 (M and N), GSK618334 (O and P) and GSK945237 (Q and R). Different colours indicate different dose levels (for details, see Table 1)

Figure 5.

Plots describing the model predictions and observed QT interval vs. time for all nine compounds after administration to dogs (left) and humans (right). The panels show data for moxifloxacin (A and B), sotalol (C and D), cisapride (E and F), NCE03 (G and H), NCE04 (I and J), carabersat (K and L), SB237376 (M and N), GSK618334 (O and P) and GSK945237 (Q and R). Different colours and lines indicate different dose levels (for details, see Table 1)

Our results clearly indicate that the so‐called system‐specific parameters – i.e. baseline QT (QT _c0), the QT‐RR correction factor (α), the amplitude (A) and phase (Ф) – are almost all within the same range of values for the different compounds within each species (Table 2). However, we should emphasize that mean differences (95% credible intervals) for parameter values describing system‐specific properties in dogs show larger variability than the estimated obtained in humans. Of interest is QT _c0 in dogs, which ranged between 238 (237–240) ms (cisapride) and 261 (252–270) ms (NCE04), and in humans, which ranged between 379 (371–386) ms (NCE03) and 405 (377–435) ms (GSK618334). Similarly, the values of α in dogs ranged from 0.18 (0.11–0.3) (sotalol) to 0.3 (0.22–0.4) (GSK945327) and in humans from 0.18 (0.14–0.23) (cisapride) to 0.4 (0.38–0.42) (moxifloxacin). Estimates obtained for the other two model parameters describing the circadian rhythm – namely, A and Ф – appeared to be affected by experimental protocol design, with values for A in dogs ranging from 2.13 (0.56–7.67) ms (GSK618334) to 9.2 (4.3–18.1) ms (NCE03) and in humans from 2.4 (1.7–2.9) ms (moxifloxacin) to 7.9 (6.8–9.3) ms (NCE04). For Ф, values in dogs ranged from 4.4 (0.6–9.1) h (NCE04) to 31.4 (16–55) h (GSK618334) and in humans from 4.4 (3.3–5.7) h (NCE04) to 10 (7–13) h (moxifloxacin).

Interspecies correlation

As indicated previously, the PKPD analysis showed that the main difference between compounds, as well as between species, was the slope of the linear concentration–effect relationship and consequently the concentration associated with a 50% probability of a QT increase ≥10 ms (CP₅₀) (Table 3).

The correlation between the slope parameter in dogs and in humans was assessed by linear regression (r² = 0.989) (Figure 3). However, data from cisapride were not included in the estimation steps owing to the known differences in QT‐prolonging effects – i.e. it not only blocks a single ion channel type, it also interacts with other ion channels (see Figure S1 in the supplemental material for the relationship, including cisapride). For the compounds without a mixed ion channel block, the slope and intercept parameters describing the linear regression were 11.58 ms μM^–1 and −1.71 ms μM^–1, respectively. From a translational perspective, these estimates can be considered as a scaling factor and, as such, used to extrapolate drug effects from dogs to humans in the drug‐specific effect. Of particular interest is the slope of the regression, which suggests that at comparable drug levels, dogs are, on average, approximately 12‐fold less sensitive to the drug‐induced QT effects than humans.

Figure 3.

Unweighted linear correlation of the slope (ms μM^–1) in dogs and humans. Compounds are shown in different colours: cisapride (green), moxifloxacin (red), sotalol (blue), NCE04 (orange), NCE03 (purple), carabersat (grey), GSK618334 (pink), GSK945237 (dark green), SB237376 (black). Dashed lines represent the 95% confidence interval around the mean; linear correlation (red line) = y = −1.709 + 11.586x, R² = 0.989

Given the uncertainty in the parameter estimates obtained during the initial PKPD modelling, 95% credible intervals of the mean were also calculated to ensure that a worst‐case scenario is considered for subsequent scaling purposes (Figure 3).

Discussion

Evidence of limited or no proarrhythmic properties is critical for the progression of compounds into clinical development. A myriad of tests are performed both in vitro and in vivo before progressing into humans, during which human Ether‐a‐go‐go Related Gene (hERG) channel blockade, electrophysiological measures of drug activity on heart conductivity and telemetred QT interval are assessed in preclinical species. However, no quantitative measure is available that allows direct extrapolation and prediction of drug effects at therapeutically relevant concentrations in humans. The lack of a scaling factor to translate drug effects in humans therefore represents a major challenge for the assessment of the safety profile of molecules entering phase I trials, during which doses are escalated to supratherapeutic levels 31, 32, 33.

The difficulties in predicting QT prolongation in humans have been attributed to numerous factors, including the lack of suitable measures to characterize prodromic effects in vitro and the poor specificity of the QT interval as a marker of proarrhythmic activity 34, 35. Despite such difficulties, for more than a decade, QT prolongation has been recognized by regulators as the most appropriate marker of the risk of TdP and of the potential implications of hERG or ion channel inhibition on heart conductivity. In fact, the current regulatory requirements have contributed to efforts aimed at the assessment of the relationship between drug concentrations and QT interval prolongation. There are various examples in the published literature where PKPD modelling has been used as a tool to overcome the shortcomings of the traditional approaches based on statistical hypothesis testing. However, most PKPD models are data driven and descriptive – i.e. the primary objective of such analyses is to obtain accurate parameter estimates and establish the magnitude of the drug‐induced effect 36. Less common is the use of PKPD modelling as the basis for translational purposes.

Here, we attempted to assess the correlation between drug‐specific parameter estimates obtained by PKPD modelling of QT‐interval data in dogs and humans. As shown in Figure 3, the identification of a linear correlation for the slope of the PKPD relationships provides an opportunity to revisit the approach for assessing the QT‐prolonging potential of novel molecules.

Implications of interspecies differences in system‐specific parameters

Whereas intrinsic differences are known to exist between species, such as in baseline QT _C0, changes in this parameter are also caused by physiological factors such as age and basal heart rate 37. In fact, QT _C0 in dogs varied from 238 ms for cisapride to 261 ms for NCE04. Under comparable experimental conditions, these differences are most likely explained by differences in the age of the dogs. It is well known that heart rate varies with age in dogs, and such changes occur over a shorter time span relative to humans. By contrast, as only adult subjects were used in the clinical studies, such effects were not visible in the clinical data (Table 2). Similarly, the individual values of α were larger and had different ranges in humans, as compared with dogs. Despite such differences between species, the observed variability in the estimates of α should have no impact on the slope parameter, which captures the drug effect on the QT interval.

The other system‐specific parameters in dogs and in humans – i.e. the A and Ф of the 24‐h circadian rhythm – were estimated within a similar range for all compounds (Table 2). The variation that was observed in these estimates is likely to be an artefact of the ECG sampling procedures. In fact, for some experiments, the Ф parameter showed values greater than 24 h, which exceed the physiological boundaries of the circadian rhythm. These findings were probably due to the limited sampling scheme used in preclinical protocols, which were sparse between 6 h and 24 h after dosing.

Implications of interspecies differences in the drug‐specific parameter

In our analysis, the main differences between compounds and between species were found in the estimates of the slope of the PKPD relationships (Table 3). Such a systematic difference (i.e. >11‐fold ratio between dogs and humans) prompted us to explore the correlation between species further. We have also noticed that some compounds produce an exposure‐dependent change in RR, yielding a different QT‐RR correlation during treatment, as compared with the normal physiological changes in the QT interval due to variability in the RR interval 38. As most compounds included in the analysis were known to have minor or no intrinsic effect on heart rate, there were no separate steps to distinguish drug‐induced changes in heart rate from drug effects on the QT interval. Eventually, it will be possible to consider a two‐step approach in which predicted RR values in the absence of drug are used when characterizing the QT interval prolongation.

In principle, the larger the value of the slope of the concentration–effect relationship, the stronger the QTc‐prolonging effect of the drug will be. This was observed for cisapride, sotalol, moxifloxacin, NCE03 and GSK6183343, for which a distinct QT‐prolonging effect was detected within the (putative) therapeutic concentration range of these compounds (Table 3). On the other hand, if the slope estimates were around zero, there was no, or borderline, QTc‐prolonging effect. This was observed for carabersat, SB237376 and GSK945237. In addition, we showed that a negative value of the slope indicates a shortening of the QT interval. This phenomenon was observed for NCE04 in dogs and humans. As most compounds have not been used in clinical practice, confirmatory data are not available to corroborate the predictive performance of the findings. We therefore envisage that the analysis of new compounds using the same methodology will provide further insight into the generalizability of the correlation as a scaling factor between dogs and humans.

Translational relevance: extrapolation from animals to humans

One of the main features of the approach proposed by our group was the identification of a measure that describes in a strictly quantitative manner the drug effect on the QT interval. As shown in Figure 3, we were able to demonstrate a correlation between PKPD parameters in dogs and humans using data from nine compounds with different mechanisms of action. The estimates of the slope of the PKPD relationship in dogs appeared to be correlated linearly with the parameter estimates in humans. Based on the linear regression defining the interspecies correlation, our findings indicate that at comparable drug levels, humans are, on average, >11‐fold more sensitive to the drug‐induced effect on the QT interval. Clearly, further investigation will be needed to assess whether this difference can be used as a scaling factor to predict the drug‐induced effects on the QT interval in humans. However, the evidence for such a correlation emphasizes the relevance of dogs as the species of choice for predicting drug‐induced QTc interval prolongation in humans.

Most importantly, model predictions show that when QT prolongation occurs in dogs, drug effects may be observed at a different (somewhat lower) exposure range in humans. It also reveals that without a clear understanding of the expected pharmacokinetic profile and therapeutic exposure in humans, the absence of QT prolongation within a given concentration range in dogs does not allow one to conclude that QT prolongation will not occur in humans.

From a clinical perspective, another important point to consider is that our approach offers the flexibility to explore different thresholds for the QT‐prolonging effect, including a range from >1 ms to >10 ms. Interestingly, the curve describing the probability of QT prolongation in human shows a steeper increase across the therapeutic concentration range of each compound, indicating the higher sensitivity of human subjects to QT prolongation ≥10 ms at comparable levels in dogs 22. Moreover, the maximum probability of QT prolongation is consistently observed at lower exposure in humans than in dogs.

Conceptually, our approach contrasts with previous studies in which PKPD relationships were characterized for drugs with known proarrhythmic activity 35, 36, 39, 40. Whereas other authors have also described interspecies differences in PKPD relationships 41, 42, to date there have been no clear examples of the translation of drug effects from animals to humans. In fact, a potential limitation of these earlier studies is that most models were too specific for the compound of interest or required different experimental data to allow the assessment of the underlying PKPD relationships. Instead, our results demonstrated the feasibility of utilizing a single set of parameters and standardized experimental protocols as the basis for predicting the effect of new compounds in humans. In practice, this implies that those involved in the analysis and interpretation of preclinical data can reuse the same model every time a new compound is screened. Inferences about the magnitude of QT prolongation in the clinic can be extrapolated from the estimates of the slope of the PKPD relationship in dogs based on the interspecies correlation.

Although the pool of compounds used for the current analysis included only molecules for which the prodromic activity was directly linked to the levels of the parent drug, the same PKPD model may be applied to describe drug‐induced effects when QT prolongation is caused by a different moiety or mechanism. In other words, the same model components can be used to assess the PKPD relationship even when delayed effects occur (e.g. in presence of metabolites with QT‐prolonging effects). In these circumstances, a putative effect site compartment can be used to account for nonlinearities between drug exposure and the QT‐prolonging effect.

Limitations and recommendations

In traditional conscious in vivo cardiovascular safety studies, PK sampling is often limited as it may interfere with the QT, RR and blood pressure measurements. Blood samples in these experiments are therefore only obtained after the maximum plasma concentration is reached, and vary between 3–8 samples per animal per study arm. Such a sampling scheme in dogs can lead to uncertainty in drug disposition parameters, yielding considerably large confidence intervals for the parameter estimates arising from PKPD modelling, as proposed here. It should become evident to the reader that uncertainty in drug levels represents an important confounding factor for the assessment of causality, and most importantly for the accurate quantification and extrapolation of drug‐induced QT effects. This uncertainty in preclinical pharmacokinetic data is much larger than is commonly observed in humans. We therefore recommend that attention is paid to protocol design, so as to facilitate the collection of pharmacokinetic sampling in dogs. Appropriate estimation of pharmacokinetic variability will have a considerable impact on the precision of the slope parameter of the PKPD relationships – i.e. high variability will result in large credible intervals, making the prediction of drug effects in humans difficult.

We would also like to emphasize that a sensitivity analysis may be required further to explore the implications of variability in parameter estimates, as well as the impact of the drug‐induced effect on heart rate. Our findings suggest that the QT‐RR relationship may vary across compounds, especially if chronotropic effects are observed at exposure levels that are relevant to humans. In other words, one needs to disentangle the chronotropic effect from the dromotropic effect to ensure accurate prediction of the magnitude of drug‐induced changes in the QT interval for drugs that have a direct effect on RR.

Lastly, we acknowledge the potential limitations of a linear pharmacodynamic model to describe the concentration–effect relationship across the therapeutic and supratherapeutic exposure range. This choice is due to the fact that the maximal observable QT prolongation in clinical trials is likely to be censored by safety stopping criteria. Therefore, a linear concentration–effect relationship was required to avoid underestimation of the QT effect in vivo. Moreover, earlier publications that attempted to fit drug‐induced QT(c) prolongation data did not see major differences in model performances between a linear and a maximum response model (E_max) model, and showed good performance using a linear relationship 24, 41, 42.

Conclusions

In summary, the evidence of a correlation between the slope of the PKPD relationship in dogs and humans for compounds with no, or varying, proarrhythmic characteristics represents an opportunity for reducing attrition rates in the progression of compounds from preclinical to clinical development. Whereas further investigation is required to establish the generalizability of the correlation for a wider range of compounds, we anticipate the use of this correlation as a tool to predict the probability of QT interval prolongation in humans using clinical trial simulations. At this stage, it is unclear whether different correlations can be identified for compounds with distinct mechanisms of action and the extent to which the affinity for other ion channels may result in false‐positive or false‐negative rates. Nevertheless, our approach provides the basis for further integration, and effective extrapolation of preclinical data, enabling the effects of drugs on the QT interval to be predicted before exposing humans to new chemical and biological entities.

Competing Interests

All authors have completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: VFSD had support from the TI Pharma consortium for his PhD fellowship; no financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years; no other relationships or activities that could appear to have influenced the submitted work.

The authors would like to thank the various members of the TI Pharma Cardiovascular Safety Project (http://www.tipharma.com/pharmaceutical‐research‐projects/completed‐projects/pk‐pd‐modeling‐platform.html) for their contribution to the identification of experimental protocols and clinical trials across each organization and for the fruitful discussions that led to the research presented in this manuscript: Sandra Visser, Dinesh De Alwis, Jackie Bloomer, Nick McMahon and Phil Milliken, David Gallagher and An Vermeulen; and Piet van der Graaf Mark Holbrook.

Supporting information

Figure S1 Unweighted linear correlation of the slope (ms μM^–1) in dogs and humans. Compounds are shown in different colours: cisapride (green), moxifloxacin (red), sotalol (blue), NCE04 (orange), NCE03 (purple), carabersat (grey), GSK618334 (pink), GSK945237 (dark green), SB237376 (black). Dashed lines represents the 95% confidence interval around the mean; linear correlation (red line) = y = −1.5 + 14.5x; R2 = 0.855

Supporting info item

Dubois, V. F. S. , Smania, G. , Yu, H. , Graf, R. , Chain, A. S. Y. , Danhof, M. , Della Pasqua, O. , on behalf of the Cardiovascular Safety Project Team, and TI Pharma PKPD Platform (2017) Translating QT interval prolongation from conscious dogs to humans. Br J Clin Pharmacol, 83: 349–362. doi: 10.1111/bcp.13123.