The canine chronic atrioventricular block model in cardiovascular preclinical drug research.

Ventricular cardiac arrhythmia is a life threating condition arising from abnormal functioning of many factors in concert. Animal models mirroring human electrophysiology are essential to predict and understand the rare pro- and anti-arrhythmic effects of drugs. This is very well accomplished by the canine chronic atrioventricular block (CAVB) model. Here we summarize canine models for cardiovascular research, and describe the development of the CAVB model from its beginning. Understanding of the structural, contractile and electrical remodelling processes following atrioventricular (AV) block provides insight in the many factors contributing to drug-induced arrhythmia. We also review all safety pharmacology studies, efficacy and mechanistic studies on anti-arrhythmic drugs in CAVB dogs. Finally, we compare pros and cons with other in vivo preclinical animal models. In view of the tremendous amount of data obtained over the last 100 years from the CAVB dog model, it can be considered as man's best friend in preclinical drug research. LINKED ARTICLES: This article is part of a themed issue on Preclinical Models for Cardiovascular disease research (BJP 75th Anniversary). To view the other articles in this section visit http://onlinelibrary.wiley.com/doi/10.1111/bph.v179.5/issuetoc.

atrioventricular

chronic atrioventricular block

long QT syndrome

spatial dispersion of repolarization

short‐term variability of repolarization duration

INTRODUCTION

Cardiovascular diseases are a major cause of death in the world. In this aspect, cardiac arrhythmias form an important quantitative and qualitative category (Al‐Khatib et al., 2018). There are several types of cardiac arrhythmias, some of which are directly life‐threatening, such as ventricular fibrillation, whereas others increase the risk of secondary conditions, such as atrial fibrillation (Al‐Khatib et al., 2018). Although treatments of cardiac arrhythmias are still improving and expanding, such as device technology and catheter ablation, pharmacotherapy remains an important pillar in the clinic (Al‐Khatib et al., 2018). Many anti‐arrhythmics have been developed that can directly or indirectly inhibit a ventricular or atrial arrhythmia. Moreover, new pharmacological targets are still being discovered for which medication is being developed, although the amount of new anti‐arrhythmics in different phases of development is still rather low (Camm, 2017), which definitively requires the united attention from academia and industry. Importantly, in each drug discovery and development trajectory, in vivo testing using animal systems is still necessary and mandatory by legislation (ICH Guideline S7B, 2005). Many in silico and in vitro systems are currently available, or are being improved or developed as part of the Comprehensive in Vitro Proarrhythmia Assay (CiPA) initiative, that can evaluate one or more factors underlying cardiac arrhythmia. These systems are positioned mostly in the intermediate phases of drug development. Especially in silico action potential models combined with dynamic clamp approaches and/or human‐induced pluripotent stem cell‐derived cardiomyocytes as testing systems almost completely integrate the interplay of the ion channels and transporters that generate the human action potential and their collective responses to new chemical entities (Gintant et al., 2016). Thus far however, only in vivo systems can show the intricate interplay between cardiac electrical activity, pump function, haemodynamic feedback, neural regulation, electrolyte homeostasis, display of spatial dispersion of repolarization (SDR) and their collective response to pharmacological treatment in one model.

The ideal animal system for cardiac drug evaluation does not exist. But such a model should at best mirror human properties such as ion channel expression, both in type and interdependence, haemodynamics, neural control of the heart, pharmacodynamics, cardiac size and so forth. In other words, a model should include all biological aspects that are important or currently perceived to be important, for the occurrence of cardiac arrhythmia. With respect to these factors, the dog is considered a suitable model (Clauss et al., 2019).

Canines have been an important model for (electro) cardiac research since the start of experimental physiology in the 19th century. They have been used for example in research on sinus arrhythmia (Ludwig, 1847), the cardiac conduction system (Tawara, 1906), the ventricular excitation pattern (Durrer & Van der Tweel, 1957) and many more physiological mechanisms involved in the function of the heart.

Here, we will present an overview of different canine models used in preclinical cardiovascular research, followed by an in‐depth exploration of one of the most frequently used canine models in cardiac safety pharmacology and anti‐arrhythmic drug research, the chronic atrioventricular block (CAVB) dog model.

CONCORDANCE BETWEEN HUMAN AND CANINE PHYSIOLOGY

To appreciate the extent to which the results of different cardiac disease models in dogs can be translated to the human situation, it is important to know the similarities and differences between the canine and human heart. First off anatomically, the heart sizes of large dogs are comparable with human heart sizes, but obviously, large dogs display a higher heart weight to body weight ratio. The canine coronary anatomy is different to the human situation, as there is an extensive collateral circulation and left dominant system in dogs. Electrophysiologically, dogs compare reasonably well with humans. The anatomy of the conduction system is quite similar, given the length of the His bundle, the distribution of the Purkinje fibres along the ventricular wall and the location of the atrioventricular (AV) node just above the tricuspid valve. Moreover, almost all major currents found in humans are present in dogs. These similarities explain the close resemblance between canine and human action potential waveform, both portraying typical notching (Clauss et al., 2019). However, the relative contributions of the ionic currents do vary considerably between humans and dogs. I K1 and I Ks densities were found to be threefold and 4.5‐fold larger in dogs than in humans, whereas I Kr density was similar. As these are the major repolarizing currents, the redundancy in human repolarization is thus far smaller than in dogs. In other words, humans have a smaller repolarization reserve, ‘safety reserve’ to withstand challenges on the I Kr current (Jost et al., 2013). We will further elaborate on the concept of repolarization reserve in Section 5. Because of the different composition of canine ionic currents, there are important electrophysiological differences between human and dog hearts. In sinus rhythm during rest, atrial and ventricular action potential duration, and therefore all the ECG parameters, are shorter in dogs than in humans (dogs vs. humans: P wave 40 ms vs. 110 ms, PR interval 60–130 ms vs. 120–200 ms, QRS duration 50–60 ms vs. 84–110 ms, QT interval 150–250 ms vs. 400–430 ms). Dogs also have a higher heart rate and respiratory rate than humans (dogs vs. humans: heart rate 80–160 vs. 60–80 bpm and 15–25 vs. 12–15 rpm). Since the QT interval physiologically shortens as the heart rate increases, it is very important to correct the QT interval appropriately when interpreting canine preclinical data. Several correction formulas have been designed based on large patient cohorts and applied in clinical practice, such as the Bazett, Fridericia and Framingham formulas. However, the Van de Water formula has been specifically designed to correct for QT intervals in anaesthetized dogs. Indeed, when applying different QT correction formulas on dog data, the Van de Water formula gives the best approximation (Patel et al., 2017). We therefore recommend the use of the Van de Water formula when correcting the QT interval in preclinical research in dogs.

The systolic blood pressure (BP) in dogs is higher than in humans, but diastolic BP is lower (136/66 in dogs vs. 120/80 mmHg in humans) (Clauss et al., 2019). A final physiological difference is that there is a more pronounced and more variable respiratory sinus arrhythmia in dogs, which results in a higher heart rate variability. Importantly, increased heart rate variability does not necessarily indicate that the canine heart is more susceptible to arrhythmias in general. In heart failure patients, for example, loss of respiratory sinus arrhythmia is often found and is associated with poor long‐term outcomes, such as sudden cardiac death. The higher heart rate variability in dogs should thus be interpreted as a reflection of a different balance between cardiac sympathetic and parasympathetic innervation and thus a species‐specific trait. Indeed, an increased parasympathetic tone to the canine heart is one of the major contributors to the elevated heart rate variability (Moïse et al., 2020). The autonomic nervous system (ANS) is a well described factor in the development of cardiac arrhythmias. Therefore, the different balance in sympathetic versus parasympathetic innervation between humans and dogs is an important factor to be considered when interpreting preclinical research in canine cardiac disease models (Fukuda et al., 2015).

Other than the increased parasympathetic tonus in the canine heart, the innervation of the human and canine heart are very similar. Sympathetic innervation arises from the stellate ganglion, where preganglionic nerves activate postganglionic nerves by releasing the neurotransmitter acetylcholine (ACh). The postganglionic sympathetic nerves exert their effect on the heart by releasing noradrenaline. The parasympathetic cardiac innervation originates mainly from the nucleus ambiguus, located in the medulla of the brain. The preganglionic, cranial nerve that projects from the nucleus ambiguus to the postganglionic nerves is the vagal nerve or the 10th cranial nerve. ACh release, in case of the parasympathetic division, not only activates of the postganglionic nerves but also transmits the postganglionic nerve activity to the heart (Mizeres, 1955).

All in all, the dog has been described as the most predictive species in the cardiac electrophysiological research field, which underlines the consonance between canine and human cardiac electrophysiology. Other large animals with more comparable coronary anatomy, like pigs, have proven superior in myocardial ischaemia research (Clauss et al., 2019; Gralinski, 2003).

A second important factor to consider for translational validity of canine cardiac disease models, particularly cardiac drug evaluation, is pharmacokinetics. Compared with humans, dogs have some different characteristics that can alter the pharmacokinetics of drugs and therefore make extrapolation between canines and humans difficult. These characteristics include altered gastric pH, intestinal motility and permeability, plasma protein binding, physiological volumes and the substrate specificity of the CYP450 enzymes, that is, the hepatic enzymes that perform catabolic reactions on compounds (better known as ‘first‐pass metabolism’). It is therefore very important to monitor plasma concentrations of administered compounds when investigating the effect of a drug in canine preclinical cardiac disease models (Tibbitts, 2003).

CANINE MODELS IN CARDIOVASCULAR RESEARCH

Due to the aforementioned translational benefits that dogs pose for studying cardiac diseases, a plethora of models have been developed. Table 1 shows an overview of the different types of dog models that have been used in cardiac preclinical research and their effectivity. Please note that the selected references serve as illustrative examples and do not represent a comprehensive list of all studies that exploit the respective model. The table is divided in three major categories: electrical heart disease, heart failure and myocardial infarction. The first category, electrical heart diseases, is subdivided into ventricular and atrial arrhythmias. Notably, the majority of the models require electrical stimulation protocols to evoke episodes of arrhythmias and are therefore performed under anaesthesia. The second category, heart failure, is subdivided into models that approach heart failure with a reduced ejection fraction, the ‘classic’ form of heart failure and models that approach heart failure with a preserved ejection fraction. The latter category has recently gained increasing clinical interest, as subjects display typical symptoms of heart failure but do not demonstrate a loss of ejection fraction. They do however show impaired myocardial relaxation, hampering the filling of the heart during diastole and thereby contributing to the clinical symptoms of heart failure. As this is a relatively new insight in the field of heart failure, far fewer models for the latter subcategory have been described than for ‘classic’ heart failure. Within the subcategory of heart failure with a reduced ejection fraction, the models have been organized based on the pathological condition they simulate, that is, ischaemia, pressure/volume overload and a residual category. Finally, the last major category consists of myocardial infarction. This category displays methods to induce myocardial infarction in the context of developing (pharmacological) interventions to limit the size of the infarcted area. Notably, the induction of ischaemia is also listed under electrical heart disease and heart failure, but in these cases, the aim of these studies was to induce arrhythmias or heart failure, respectively. Therefore, the induction of ischaemia was instrumented to study other endpoints in these cases, whereas infarction size was the primary outcome in the ‘myocardial infarction’ category.

TABLE 1

Overview of canine models for cardiac disease

Category	Model	Procedures	Effectivity	Reference
Electrical heart diseases
Ventricular arrhythmias	Electrical stimulation protocol	Faradic stimulation (A)	Not reported	Wiggers, 1930
		Shock in vulnerable period (A)	Not reported	Wiggers & Wegria, 1940
	Sympathetic modulation	i.v. epinephrine + deep cyclopropane anaesthesia (A)	9/11 spont. VT paroxysms, varying from 11 to 302 s	Wilburne et al., 1947
	MI (acute/chronic)	Acute LAD ligation + chronic two‐stage occlusion (A)	After acute ligation 3/4 VF; after two‐stage occlusion 4/4 VT in large infarct size, 2/6 VT in small infarct size	Harris, 1950
	MI (acute/chronic) + (ESP/compounds)	Acute LAD dissection + shock in vulnerable period (A)	7/7 dogs VT/VF	Wiggers et al., 1940
		Double LAD ligation, after 3–7 days of premature stimulation (A)	40/45 VT (5/45 small infarction due to collaterals)	El‐Sherif et al., 1977
		Transient LAD occlusion + pacing 80 bpm + compounds (A)	Flecainide 4/6 VT/VF, dofetilide 0/6	Damiano et al., 2015
	Chronic AV block + compounds	Formaldehyde injection ventricular septum, diuretics + compounds (A)	Quinidine 2/5 TdPs, propranolol 1/5 VT, sotalol 4/5 TdPs	Weissenburger et al., 1991
	Chronic ventricular tachypacing	VVI 250 bpm for 3–5 weeks	6/25 SCD (polymorphic VT), average 6.7 NSVT runs/dog/24 h	Pak et al., 1997
	Aortic constriction and insufficiency	Aortic leaflet perforation, 6 weeks later banding of abdominal aorta, then repeated 24‐h Holter recordings	After 240 days 14/26 VT, after 720 days 25/26 VT	Zhu et al., 2014
	Kidney disease + ESP	Kidney artery ligation until 50%–60% kidney infarction, 6 weeks of remodelling + premature stimulation (A)	8/8 VF	Tang et al., 2017
	Long QT syndrome + sympathetic modulation	i.v. I Ks blocker (HMR1556) + isoproterenol bolus (A)	17/18 TdPs	Gallacher et al., 2007
	Long QT syndrome + sympathetic modulation + ESP	i.v. caesium chloride + adrenaline + overdrive pacing (A)	8/8 VA (4/8 VT, 3/8 TdPs, 1/8 VF)	Levine et al., 1985
	Hereditary ARVC	Naturally occurring myocardial disease in boxer dogs	9/23 SCD, 19/23 ventricular arrhythmias, all dogs (fibro)fatty replacement (absent in control dogs)	Basso et al., 2004
Atrial arrhythmias	Electrical stimulation protocol	Faradic stimulation of the right auricle (A)	Flutter 15/?, fibrillation 7/?	Eyster & Swarthout, 1920
		Faradic stimulation of the right auricle (A)	Max. 25 s of auricular flutter/fibrillation	Brams & Katz, 1931
	Electrical parasympathetic stimulation + ESP	Bilateral cervical vagal trunk stimulation + rapid pacing bursts (A)	16/16 persistent (>30 min) atrial fibrillation, terminated when vagal stimulation stopped	Wang et al., 1992
	Chemical parasympathetic stimulation	ACh injection in pulmonary vein fatpad containing autonomic ganglia (+ESP)	2/5 spontaneous atrial fibrillation (AF), 3/5 susceptible to single extrastimulus. Avg. AF duration 9.6 min	Po et al., 2005
		Carbachol injection in pulmonary vein fatpad containing autonomic ganglia (+ESP)	2/6 spontaneous atrial fibrillation (AF), 4/6 susceptible to single extrastimulus. Avg. AF duration 37.7 min	Po et al., 2005
	Atrial lesion + ESP	Y‐shaped intracaval lesion, after 2 weeks of rapid burst pacing	5/5 non‐self‐terminating flutter, variable susceptibility	Frame et al., 1986
		Crush injury on right atrial free wall + acute rapid and premature pacing (A)	8/8 dogs sustained (>10 min) flutter, variable susceptibility	Feld & Shahandeh‐Rad, 1992
	Chronic ventricular tachypacing + ESP	3 weeks VVI 240, 2 weeks VVI 220 bpm, then burst pacing (A)	10/18 sustained (>30 min), 8/18 non‐sustained atrial fibrillation	Li et al., 1999
	Chronic atrial tachypacing + ESP	Atrial pacing 400 bpm, after 6 weeks of premature stimulation/rapid burst if necessary (A)	18/22 sustained (>15 min) atrial fibrillation, 2/22 non‐sustained, 2/22 non‐inducible	Morillo et al., 1995
	Chronic atrial tachypacing + mitral regurgitation	Transection of chordae tendinae of mitral valve + atrial pacing 640 bpm, after 6 weeks 24 h of Holter recording	16/16 spontaneous persistent atrial fibrillation	Mitchell et al., 1997
	Mitral regurgitation + ESP	Transection of chordae tendinae of mitral valve, after 3 months of rapid burst pacing if necessary (A)	3/25 spontaneous fibrillation, 16/25 after burst pacing (5/16 non‐self‐terminating), 6/25 non‐inducible	Cox et al., 1991
	Sterile pericarditis + ESP	Pericardiotomy, talcum powder on atrial surface, after recovery rapid premature pacing	23/25 flutter, 17/23 lasting >5 min	Pagé et al., 1986
	Chronic systemic inflammation + ESP	Peridontitis through silk ligatures around premolars, after 60/90 days of extrastimuli (A)	60 days: 5/12 fibrillation, duration avg. 1.55 s, 90 days: 10/12 fibrillation, duration avg. 4.76 s	Yu et al., 2010
	Ischaemic cardiomyopathy + ESP	Permanent LAD occlusion, after 6–8 weeks of burst pacing (A)	5/5 dogs inducible for atrial fibrillation, average 28 s	Ohara et al., 2002
	Atrial infarction (acute/chronic) + ESP	Double ligation of RIAA + acutely burst pacing (A)	9/20 dogs prolonged (>20 min) atrial fibrillation	Sinno et al., 2003
		Double ligation of RIAA, after 8 days of burst pacing (A)	All dogs showed atrial fibrillation, average duration increased from 30 s (sham) to 1146 s	Nishida et al., 2011
Heart failure
Reduced ejection fraction
Ischaemia (acute/chronic effect)	LAD ligation	LAD ligation + intermittent ligation of collaterals until ‘medium‐sized ischaemia’. Then 2–6 h of observation	2 h: LVp decreased from 136 to 98 mmHg, LVEDP increased from <12 to 15 mmHg, LVEF decreased from 68% to 27%; 6 h: 80% mortality due to VF	Feola et al., 1971
	LAD ligation − reperfusion	LAD + major diagonal branches ligation for 180 min, then reperfusion and 4 weeks of remodelling	LVEDP increased from 3.9 to 15.0 mmHg, dP/dt and LVEF decreased (2032 to 1982 mmHg and 64% to 47%)	Saku et al., 2018
	Sequential coronary microembolization	Every 1–3 weeks of injection of 3–6 ml of latex microspheres suspension in LAD/circumflex via implanted coronary catheter until LVEF < 35%. Then 3 months of remodelling	LVEF and CO decreased (64% to 21% and 2.9 to 2.3 L·min−1). LVEDP and LVEDV increased (6 to 22 mmHg and 64 to 101 ml), 30% in‐study mortality, 26% atrial fibrillation	Sabbah et al., 1991
Pressure/volume overload	Chronic aortic constriction	Inflatable cuff around ascending aorta, repeated measurements of LV dimensions	After 19 days of LV wall thickness increased 15%, peak wall stress increased 22%	Sasayama et al., 1976
		Expanding band around ascending aorta of 8‐week‐old puppies, 12 months of follow‐up	6/16 dogs clinical signs of heart failure, LV:body weight 9.8 g·kg−1, LVEDP 25 mmHg (compensated group: 7.7 g·kg−1 and 8 mmHg)	Gaasch et al., 1989
	Mitral regurgitation	Disruption of mitral chordae/leaflets until significant mitral regurgitation. Then 12–17 months of follow‐up	16/22 dogs survived >3 months, 11/15 severe regurgitation (RF > 0.5). Severe group: LVEDV, LVEDP and LV mass increased (48–85 ml, 9–16 mmHg and 71–90 g), CO decreased (2.3 to 1.8 L·min−1)	Kleaveland et al., 1988
	Chronic AV block	Section of the region with the bundle of His; then 3 months of remodelling	5/11 congestive heart failure (hepatomegaly + pulmonary vascular congestion) in resting state, 7/11 under daily exercise, 1/7 dogs SCD	Starzl & Gaertner, 1955
Other	Chronic ventricular tachypacing	VVI 250 until a clear biological endpoint for heart failure	After avg. 5.3 weeks: >25% increase in cardiac size (X‐ray) and/or >10% increase body weight	Armstrong et al., 1986
	Chronic mechanical dyssynchrony + ventricular tachypacing	His bundle ablation + VVI 170 for 4 weeks; left bundle branch ablation + atrial pacing 200 bpm for 6 weeks	9/9 dogs heart failure + broad QRS: LVEF decreased from 58.4% to 21.4%, EDV increased from 49.6 to 71.3 ml	Arita et al., 2007
	Cytotoxic agents (anthracyclines)	Weekly Adriamycin injection in the LAD for 5 weeks, then 3 weeks remodelling	6/6 clinical heart failure (1 required medication), LVEDV increased (76 to 99 ml), LVEF and CO declined (54% to 25% and 5.6 to 3.9 L·min−1)	Magovern et al., 1992
	Chagas cardiomyopathy	Intraperitoneal inoculation of the VL‐10 strain of Trypanosoma cruzi, after 6–9 months + echocardiography (A)	28% of dogs LVEF < 40% (established cut‐off for dilated cardiomyopathy in dogs)	Carvalho et al., 2019
	Hereditary dilating cardiomyopathy	Naturally occurring myocardial disease in Portuguese water dogs	Dogs died 2–32 weeks after birth, all post‐mortem signs of dilating cardiomyopathy	Werner et al., 2008
Preserved ejection fraction	Sequential coronary microembolization	Daily coronary injection of microbead suspension until endpoint heart failure was reached (LVEDP ≥ 18 mmHg). Then 2.5 weeks of remodelling	9/21 heart failure (LVEDP > 18 mmHg and signs) with preserved systolic function (dP/dt > 2650 mmHg), 6/21 systolic heart failure (LVEDP 23 mmHg, dP/dt < 2650 mmHg), 6/21 lesser degree of LVEDP rise	He et al., 2004
	Chronic hypertension	Renal wrapping with cellophane; then 6–8 weeks of remodelling	Increased MAP (170 vs. 140 mmHg), impaired LV relaxation (53 vs. 35 ms), higher LV mass to body weight (5.5 vs. 4.9 g·kg−1), more fibrosis (3.4 vs. 2.0%), preserved ejection fraction (49% vs. 55%)	Munagala et al., 2005
Myocardial infarction
Open chest	Surgical coronary artery ligation	Clamp of proximal circumflex artery for 3 h, then reperfusion and measurements; adjustable snare occluder around left circumflex artery until cyanosis of 75% of the inferior wall, 7 days of follow‐up	3/14 SCD, infarct size 26% of LV; 4/24 SCD, infarct size 22% of LV	Reimer et al., 1985
	Adjustable constrictor	Adjustable constriction of LAD using ameroid constrictor + exercise during 14–21 days	Subendocardial blood flow decreased from 1.19 to 0.51 ml·min−1, systolic wall thickening decreased from 24.3% to 6.0%	Heusch et al., 1987
Closed chest	Embolization	Inject flexible plugs that swell after contact with fluids in a coronary artery, follow‐up for 14–37 days	1/8 dogs died 2 days after procedure (infarct size 26% of LV), other infarctions varied from 2% to 21%, 6/8 infarctions transmural	Herr et al., 1985
		Induce coronary thrombosis by passing a current through an electrode in a major coronary branch until total occlusion; then max 16 days of follow‐up	23/23 total occlusion of major branch, infarction size varied depending on competence of collateral circulation	Salazar, 1961

Note: Summarizes the different canine models used for cardiac preclinical research, how these models are induced and their effectivity. The selected references serve as illustrative examples and do not represent a comprehensive list of all studies that exploit the respective model.

Abbreviations:‐ (A), measurements under anaesthesia; ARVC, arrhythmogenic right ventricular cardiomyopathy; AV, atrioventricular; avg., average; CO, cardiac output; ESP, electrical stimulation protocol; LAD, left anterior descending coronary artery; LV, left ventricle; LVEDP, left ventricular end‐diastolic pressure; LVEDV, left ventricular end‐diastolic volume; LVEF, left ventricular ejection fraction; LVP, left ventricular pressure; MAP, mean arterial pressure; MI, myocardial infarction; RF, regurgitation fraction; RIAA, right intermediate atrial artery; SCD, sudden cardiac death; spont., spontaneous; TdP, Torsade de Pointes; VA, ventricular arrhythmia; VF, ventricular fibrillation; VT, ventricular tachycardia; VVI, ventricular demand pacing.

Table 1 shows more overlap, majorly between heart failure and electrical heart diseases, of the use of seemingly similar techniques being used to cause different disease models. The following will provide insight in this concordance and the development of different dog models over the decades.

The use of dog models to study cardiac diseases started early in the 20th century with the induction of atrial and ventricular fibrillation by electrical stimulation. Using several stimulation protocols, short episodes of fibrillation could be induced to study the electrical and mechanical properties of the heart during fibrillation (Brams & Katz, 1931; Eyster & Swarthout, 1920; Wiggers, 1930; Wiggers & Wegria, 1940). A decade later, the severity of the induced episodes was increased by adding acute myocardial infarction or sympathetic modulation to the stimulation protocols (Harris, 1950; Wiggers et al., 1940; Wilburne et al., 1947).

The ischaemic techniques were further developed over the years and came closer to the human pathological conditions that predispose for the development of arrhythmias. Myocardial infarction was applied more chronically, as coronary occlusion was approached less aggressively and was followed by a period of remodelling. Due to the aforementioned extensive collateral coronary circulation, chronic coronary occlusion alone was usually not sufficient to induce arrhythmias. Additional stimulation protocols or compounds remained necessary as an ‘additional hit’ (Damiano et al., 2015; El‐Sherif et al., 1977; Nishida et al., 2011; Ohara et al., 2002; Sinno et al., 2003). Without these ‘additional hits’ as pacing, sympathetic modulation or drugs, different methods for causing cardiac ischaemia were applied in models of ischaemic cardiomyopathy (Feola et al., 1971; He et al., 2004; Munagala et al., 2005; Sabbah et al., 1991; Saku et al., 2018). The same goes for chronic atrioventricular block, mitral regurgitation, increased afterload and tachypacing‐induced cardiomyopathy: on their own, these conditions were induced in the context of studying heart failure (Arita et al., 2007; Armstrong et al., 1986; Gaasch et al., 1989; Kleaveland et al., 1988; Sasayama et al., 1976; Starzl & Gaertner, 1955). In combination with electrical stimulation protocols or drugs, they served as models to study atrial or ventricular fibrillation in the context of a remodelled heart (Cox et al., 1991; Li et al., 1999; Morillo et al., 1995; Weissenburger et al., 1991; Zhu et al., 2014). This introduction of ‘co‐pathology’ is in line with clinical practice, as there is a strong association between heart failure and rhythm disturbances in patients (Al‐Khatib et al., 2018).

Some heart failure‐based models for electrical heart disease are also effective in producing spontaneous arrhythmias without the need for pacing protocols or drugs to evoke episodes of fibrillation/tachycardia. Chronic tachypacing‐induced cardiomyopathy has been applied on its own to study the occurrence of spontaneous ventricular arrhythmias and was proven rather effective: after 5 weeks, six out of 25 dogs had died of sudden cardiac death, with polymorphic ventricular tachycardia as the recorded culprit on chronic Holter recordings (Pak et al., 1997). In combination with mitral regurgitation, chronic tachypacing has also shown effective as a model for atrial fibrillation. After 6 weeks, all dogs showed spontaneous persistent atrial fibrillation (Mitchell et al., 1997).

Of course, there are also models for electrical heart disease without induction of heart failure. Dogs with drug‐induced long QT syndrome (LQTS) and kidney disease achieved by renal infarction are highly inducible for ventricular arrhythmias by pacing protocols or sympathetic challenge (Gallacher et al., 2007; Levine et al., 1985; Tang et al., 2017). LQTS and kidney disease are also well‐known clinical conditions that predispose the occurrence of ventricular arrhythmias. Atrial fibrillation can be induced by pacing protocols quite effectively when combined with, for example, parasympathetic modulation, creation of an atrial lesion, or induction of systemic/ pericardial inflammation. However, the fact that atrial fibrillation can be provoked after electrical stimulation does not necessarily mean that that model is a good reflection of the clinical situation. Atrial fibrillation has been well described in the context of pericarditis and systemic inflammation, but it is questionable whether a surgically created scar in the atrium or stimulation of the cervical vagal trunk is a good clinical reflection (Frame et al., 1986; Pagé et al., 1986; Wang et al., 1992; Yu et al., 2010).

Some naturally occurring diseases in dogs are also suitable for studying heart failure and electrical heart disease. Firstly, hereditary arrhythmogenic cardiomyopathy in boxers closely resembles the human syndrome. Nineteen out of 23 dogs showed spontaneous ventricular arrhythmias and post‐mortem investigation demonstrated typical fibrofatty replacement of the myocardium (Basso et al., 2004). Hereditary dilating cardiomyopathy in Portuguese water dogs also showed typical post‐mortem signs of dilating cardiomyopathy (Werner et al., 2008). Both breeds can be applied to study (early) interventions that can alter the course of the disease.

Finally, there is a residual category of less common causes of heart failure. Intraperitoneal infusion of Trypanosoma cruzi can be performed to induce Chagas cardiomyopathy or, secondly, intra‐coronary infusion of the cardiotoxic Adriamycin® (doxorubicin) can be performed to bring about drug‐induced cardiomyopathy. However uncommon, these heart failure models are reliable and have a phenotype that is close to its human counterpart (Carvalho et al., 2019; Magovern et al., 1992).

One dog model that has contributed enormously to our understanding of cardiac pro‐ and anti‐arrhythmogenicity in response to drug application is the chronic atrioventricular block (CAVB) dog, which will be the further focus of this review.

A SHORT HISTORY OF THE CHRONIC ATRIOVENTRICULAR BLOCK (CAVB) DOG

Shortly after the discovery of auricular–ventricular conduction pathway and its muscular nature by Wilhelm His Jr. and Sunao Tawara in the late 90s of the 19th and beginning of the 20th century (His, 1893; Tawara, 1906), attempts were made to create heart block, or currently named as atrial‐ventricular block, in dogs in Europe (e.g. Hering, 1905) and the United States (e.g. Erlanger, 1906) (Figure 1). The first chronic AV block (CAVB) dogs were produced by Erlanger and Blackman (1910) in the years 1906 and 1907, as an experimental model of Stokes–Adams disease. In humans, this disease is characterized by periods of low pulse rates and syncope, due to AV conduction disturbance. They succeeded to create CAVB in four animals, which lasted up to 343 days in one dog. Of note are their observations that the animals behaved normally after they had recovered from the invasive procedure, however the AV block was irreversible and the hearts became hypertrophic. Furthermore, they observed that ‘animals with complete heart block exhibit many interesting heart irregularities …’ and spells of tachycardia. Two animals experienced syncopal attacks ‘probably not determined by central processes’ and sudden death, suggestive of lethal ventricular arrhythmia in retrospect. Remarkably, several of the basic observations made from the CAVB dog by Erlanger and Black are now, 100 years later, still not completely understood on a mechanistic level.

FIGURE 1

Timeline of chronic atrioventricular block (CAVB) dog development and usage in pharmacological research. Inserts: (1) Title page of the landmark Erlanger and Blackman (1910) paper and description of the creation of AV block in dog number 7 on 17th of March 1906. This dog experienced ‘syncopal attacks’ and was found dead in the morning of 15th of April 1906, after 4 weeks of permanent AV block. (2) ECG recording (Einthoven III) of a self‐terminating drug‐induced Torsade de Pointes arrhythmia

During the following years, methods for inducing AV block became less invasive and more selective (Boucher & Duchene‐Marullaz, 1985). Whereas earliest procedures required thoracotomy combined with compression, sectioning and/or crushing of the His bundle area, later methods used (catheter) assisted injection of formalin or alcohol. Currently, radiofrequency ablation is the standard method for creating complete and irreversible AV block.

Since the Erlanger and Blackman paper, we could not obtain any reports on canine CAVB animals until the study of Starzl and Gaertner from 1955. They produced CAVB in dogs to establish a heart failure model. The majority of animals developed ‘clinical, laboratory, and pathological evidence of heart failure spontaneously or after … exercise’ (Starzl & Gaertner, 1955). In the 1960s, the CAVB model was used for testing compounds that either inhibited or stimulated noradrenergic (sympathetic) drive of the heart. Haemodynamic and, to some extent, electrophysiological effects were studied. In the 1980s, the conscious CAVB dog was used in toxicity and anti‐arrhythmic research for which ventricular tachycardia induced by digitalis or ouabain infusion combined with ventricular pacing was taken as readout (Gorgels et al., 1987; Vos et al., 1989). From the early 1990s onwards, the model was used for systematic studies on drug‐induced pro‐arrhythmia with Torsade de Pointes (TdP) as an endpoint (Weissenburger et al., 1991) (Figure 1). Torsade de Pointes is a polymorphic ventricular arrhythmia characterized on the ECG by a twisting QRS morphology around the isoelectric line (Vos et al., 1998). The latest developments that make use of the CAVB model are found in implantable device development and testing (e.g. Smoczyńska et al., 2020).

CHARACTERISTICS OF THE CAVB DOG AND ITS USE IN PHARMACOLOGY RESEARCH

Blocking the electrical conduction between the atria and ventricles results in an acute and permanent decrease in ventricular beating rate to an idioventricular rhythm (Figure 2). Due to the immediate drop in cardiac output as a result of bradycardia, a series of ventricular adjustments are initiated. The Frank–Starling principle and accompanying neurohumoral activation leads to a rapid increase in contractile force and hence stroke volume (Vos et al., 1998). At a slower timescale, structural remodelling associates with a gradual development of biventricular hypertrophy that contributes to the need for an increased stroke volume (Verduyn et al., 2001). Electrical remodelling, i.e. adjusting functional expression of ion channels and transporters that include a decrease in I Na, I Kr and I Ks and an increase in Na+/H+ exchanger activity, results in lengthening of the action potential duration, increased intracellular Na+ levels, (Na+–Ca2+ exchanger [NCX] dependent) calcium (over)load and thus to increased contraction force and duration (Sipido et al., 2000; Van Borren et al., 2013; Verdonck et al., 2003; Volders et al., 1999). Whereas cardiac output is successfully restored to up to 90%, the remodelling processes come at a price: an increased vulnerability for drug‐induced Torsade de Pointes arrhythmias (Vos et al., 1998). The decrease in expression of channels that conduct repolarizing currents challenges the repolarization reserve. The concept of repolarization reserve postulates that impairment of one type of ion channel does not excessively change repolarization duration and thus is not (easily) recognizable using standard diagnostic tools, like ECG parameters. Only multiple hits, for example, application of an ion channel inhibiting drug on top of hypokalaemia, dietary components, genetic predisposition or certain disease states, would excessively prolong repolaration and induce cardiac arrhythmia (for a comprehensive review on this topic, see Varró & Baczkó, 2011). The decrease in repolarization reserve can be quantified using short‐term variability of repolarization (STV). STV is a measure of beat‐to‐beat changes in action potential duration, that is, temporal dispersion of ventricular repolarization, over a certain number of beats. Arrhythmia severity was found to associate with the steepness of STV increase just before Torsade de Pointes occurrence (Smoczyńska et al., 2019). In addition to displaying a clear increase prior to Torsade de Pointes occurrence, STV can identify CAVB dogs at risk for development of arrhythmias. The dogs that have proven inducible for Torsade de Pointes have a higher baseline STV than their non‐inducible counterparts. This makes STV highly sensitive and specific in assessing the pro‐arrhythmic risk of drugs and identification of individuals at risk. Compared with QT, which is still most commonly used for identifying drugs at risk for inducing ventricular arrhythmias, STV is superior (Varkevisser et al., 2012). Moreover, several clinical studies demonstrate the usefulness of STV as a predictor of human arrhythmias (Hinterseer et al., 2009). In addition to down‐regulation of potassium channels, more factors contribute to the abnormal repolarization in the CAVB dog. The aforementioned alterations in calcium handling lead to increased intracellular calcium concentrations. This further enhances the prolongation of the action potential duration, as also brought about by the down‐regulation of cardiac potassium channels. If the intracellular calcium surpasses a threshold, it can trigger early after depolarizations, which can evoke Torsade de Pointes under the right circumstances. Moreover, the bradycardic conditions during CAVB further enhance variability in repolarization. Experimental variables, for example, anaesthesia or low potassium plasma levels, challenge the delicate electrical balance of the remodelled ventricles even more (Dunnink et al., 2010; Thomsen, Volders, et al., 2006; Weissenburger et al., 1991). Furthermore, the ectopic foci that initiate each ventricular beat alter the normal ventricular activation pattern. This exaggerates electrical remodelling and arrhythmia susceptibility (Winckels et al., 2007).

FIGURE 2

Depiction of cardiac changes in the chronic atrioventricular block (CAVB) dog in time following AV block, compared with the sinus rhythm (SR) animal. See text for further explanation. Bar graphs represent Torsade de Pointes(TdP) inducibility. CO, cardiac output; HR, heart rhythm; HW/BW, heart weight/body weight ratio; LV dP/dt, left ventricular pressure

Another parameter to quantify the increased sensitivity for Torsade de Pointes in the CAVB dog model is spatial dispersion of repolarization (SDR). SDR appraises local differences in repolarization duration within the ventricle. It develops, among other factors, in the setting of biventricular hypertrophy and heterogeneous electrical remodelling. By detailed ventricular mapping, it was found that dofetilide‐induced Torsade de Pointes initiates in a region with highest SDR (Dunnink et al., 2017). Moreover, SDR is one of the factors that is believed to promote the formation of re‐entry circuits. This is highly relevant in the formation of severe arrhythmias. Short lasting Torsade de Pointes are maintained by focal activity, whereas non‐self‐terminating Torsade de Pointes are maintained by re‐entry mechanisms (Vandersickel et al., 2017). Van Weperen et al. (2019) underlined the importance of SDR in arrhythmogenesis as they demonstrated that the anti‐arrhythmic efficacy of four highly effective anti‐arrhythmic drugs was best reflected by SDR. STV, however still superior to the QT interval, only partially reflects the susceptibility to Torsade de Pointes. They concluded that SDR and STV are separately involved in arrhythmogenesis. The exact interplay however between spatial and temporal dispersion of repolarization in the formation of arrhythmias remains to be elucidated (Van Weperen et al., 2019).

Finally, the autonomic nervous system is a known factor that contributes to cardiac arrhythmias. Remodelling of the diseased heart is associated with nerve sprouting, which leads to a more heterogeneous cardiac innervation. As the autonomic nervous system affects the electrophysiology of the heart directly to adapt its frequency, contractility, conduction velocity and relaxation to altering conditions, it is no surprise that an altered innervation has an influence on arrhythmogenic susceptibility (Cao et al., 2000). More specifically, sympathetic activity shows a causal relation with the emergence of arrhythmias. This is attributed to the ability of the sympathetic overexcitation to trigger early afterdepolarizations. It is however unknown whether the autonomic nervous system is a factor that modulates STV and SDR, or is a separate factor in the complex equation that leads to the formation of arrhythmias (Shen & Zipes, 2014).

All in all, the CAVB dog is mainly considered as a model of triggered activity, although re‐entry‐based arrhythmias are present too.

In the remodelled dog heart, drugs that either block or activate, for example, certain ion channels, transporters or release channels can easily and rapidly induce Torsade de Pointes arrhythmias, which forms the basis of the successful application of the CAVB dog in safety pharmacology (Tables 2 and 3). On the other hand, anti‐arrhythmic compounds that can target the same sets of proteins improve electrical stability or interfere with Torsade de Pointes progression and can prevent or suppress drug‐induced arrhythmia, making the CAVB dog a versatile and valuable model in anti‐arrhythmic drug development (Table 4).

TABLE 2

Cardiac safety testing of drugs and compounds in the awake canine CAVB model

Drug a	Dose b,c	Peak plasma	TdP b,c	Reference
Amiodarone	3 mg·kg−1 p.o.	<50 ng·ml−1	0/4	Yoshida et al., 2002
	30 mg·kg−1 p.o.	1082 ± 188 ng·ml−1	0/4	Yoshida et al., 2002
	200 mg·day−1 p.o. for 7 days + 100 mg·day−1 p.o. for 21 days	Not reported	0/4	Takahara et al., 2008
Amlodipine	0.25 mg·kg−1 p.o. 4 weeks	24.22 ± 3.99 ng·ml−1	0/8	Takahara et al., 2009
Apomorphine	1 mg·kg−1 10 min−1 i.v.	3.5 ± 0.8 μg·ml−1	0/4	Watanabe et al., 2015
Astemizol	3 mg·kg−1 p.o.	Not reported	0/4	Izumi‐Nakaseko et al., 2016
	30 mg·kg−1 p.o.	Not reported	1/4	Izumi‐Nakaseko et al., 2016
Azithromycin	30 mg·kg−1 10 min−1 i.v.	68.5 ± 4.9 μg·ml−1	0/4	Ohara et al., 2015
Bepridil	3 mg·kg−1 p.o.	Not reported	0/4	Takahara et al., 2008
	30 mg·kg−1 p.o.	Not reported	3/4	Takahara et al., 2008
Candesartan	1.2 mg·kg−1 p.o. 4 weeks	Not reported	0/7	Takahara et al., 2009
Cilnidipine	0.5 mg·kg−1 p.o. 4 weeks	9.16 ± 1.44 ng·ml−1	0/7	Takahara et al., 2009
Cisapride	1 mg·kg−1 p.o.	Not reported	1/6	Sugiyama, Ishida, et al., 2002
	10 mg·kg−1 p.o.	Not reported	6/6	Sugiyama, Ishida, et al., 2002
	10–20 mg·kg−1 p.o.	Not reported	4/8	Wijers et al., 2018
E‐4031	0.03 mg·kg−1 10 min−1 i.v.	16.5 ng·ml−1	0/4	Goto et al., 2018
	0.1 mg·kg−1 10 min−1 i.v.	60.5 ng·ml−1	1/4	Goto et al., 2018
	0.3 mg·kg−1 10 min−1 i.v.	182.5 ng·ml−1	4/4	Goto et al., 2018
Donepezil	0.1 mg·kg−1 10 min−1 i.v.	Not reported	0/4	Hagiwara‐Nagasawa et al., 2021
	1 mg·kg−1 10 min−1 i.v.	Not reported	0/4	Hagiwara‐Nagasawa et al., 2021
Famotidine	1 mg·kg−1 5 min−1 i.v.	Not reported	0/4	Sugiyama et al., 2003
	10 mg·kg−1 5 min−1 i.v.	Not reported	0/4	Sugiyama et al., 2003
Flecainide d	1.5 mg·kg−1 10 min−1 + 0.9 mg·kg−1 60 min−1 i.v.	0.7 ± 0.1 mg·L−1	0/6	Weissenburger et al., 1991
	3 mg·kg−1 10 min−1 + 1.8 mg·kg−1 60 min−1 i.v.	Not reported	0/6	Weissenburger et al., 1991
Gatifloxacin	10 mg·kg−1 p.o.	4.1 ± 0.3 μg·ml−1	0/4	Chiba et al., 2004
	100 mg·kg−1 p.o.	11.3 ± 1.6 μg·ml−1	2/4	Chiba et al., 2004
Haloperidol	3 mg·kg−1 p.o.	Not reported	0/4	Izumi‐Nakaseko et al., 2017
	30 mg·kg−1 p.o.	Not reported	4/4	Izumi‐Nakaseko et al., 2017
Lapatinib	3 mg·kg−1 10 min−1 i.v.	2.358 ± 0.424 μg·ml−1	0/4	Ando et al., 2020
Levofloxacin	6 mg·kg−1 p.o.	1.81 ± 0.45 μg·ml−1	0/4	Chiba et al., 2000
	60 mg·kg−1 p.o.	17.74 ± 02.57 μg·ml−1	0/4	Chiba et al., 2000
Lidocaine d	3 mg·kg−1 10 min−1 + 3 mg·kg−1 60 min−1 i.v.	3.1 ± 0.6 mg·L−1	0/6	Weissenburger et al., 1991
Mexiletine d	4.5 mg·kg−1 10 min−1 + 1.5 mg·kg−1 60 min−1 i.v.	Not reported	0/8	Chézalviel‐Guilbert et al., 1995
Moxifloxacin	10 mg·kg−1 p.o.	2.1 ± 0.3 mg·ml−1	0/4	Chiba et al., 2004
	100 mg·kg−1 p.o.	12.6 ± 1.0 μg·ml−1	3/4	Chiba et al., 2004
Nifekalant	3 mg·kg−1 p.o.	4.66 ± 0.21 μg·ml−1	0/5	Satoh et al., 2004
Oseltamivir	3 mg·kg−1 10 min−1 i.v.	Not reported	0/4	Nakamura et al., 2016
(Tamiflu®)	10 mg·kg−1 10 min−1 i.v.	Not reported	0/4	Nakamura et al., 2016
	30 mg·kg−1 10 min−1 i.v.	Not reported	0/4	Nakamura et al., 2016
Propranolol	0.5 mg·kg−1 70 min−1 i.v.	0.21 ± 0.03 mg·L−1	0/6	Weissenburger et al., 1991
Quinidine d	10 mg·kg−1 10 min−1 + 1.8 mg·kg−1 60 min−1 i.v.	4.2 ± 0.8 mg·L−1	0/6	Weissenburger et al., 1991
Risperidone	3 mg·kg−1 10 min−1 i.v.	Not reported	0/4	Nunoi et al., 2020
Sematilide	3 mg·kg−1 p.o.	Not reported	0/4	Yoshida et al., 2002
	30 mg·kg−1 p.o.	Not reported	3/4	Yoshida et al., 2002
d‐Sotalol	3 mg·kg−1 p.o.	Not reported	1/4	Goto, Hagiwara‐Nagasawa, Chiba, et al., 2019
d,l‐Sotalol	3 mg·kg−1 p.o.	Not reported	3/4	Goto, Hagiwara‐Nagasawa, Chiba, et al., 2019
	10 mg·kg−1 p.o.	Not reported	3/4	Goto, Hagiwara‐Nagasawa, Kambayashi, et al., 2019
	2.25 mg·kg−1 10 min−1 + 0.75 mg·kg−1 60 min−1 i.v.	Not reported	1/5	Weissenburger et al., 1991
	4.5 mg·kg−1 10 min−1 + 1.5 mg·kg−1 60 min−1 i.v.	4.5 ± 0.2 mg·L−1	5/6	Weissenburger et al., 1991
d‐Sotalol d	4.5 mg·kg−1 10 min−1 + 1.5 mg·kg−1 60 min−1	Not reported	6/7	Chézalviel‐Guilbert et al., 1998
d‐Sotalol +	4.5 mg·kg−1 10 min−1 + 1.5 mg·kg−1 60 min−1 i.v.	Not reported	1–3/7	Chézalviel‐Guilbert et al., 1998
Quinidine d	10 mg·kg−1 10 min−1 + 1.8 mg·kg−1 60 min−1 i.v.	Not reported
Sitafloxacin	10 mg·kg−1 p.o.	1.7 ± 0.4 μg·ml−1	0/4	Chiba et al., 2004
	100 mg·kg−1 p.o.	9.8 ± 1.7 μg·ml−1	0/4	Chiba et al., 2004
Sparfloxacin	6 mg·kg−1 p.o.	1.56 ± 0.15 μg·ml−1	0/4	Chiba et al., 2000
	60 mg·kg−1 p.o.	3.89 ± 1.39 μg·ml−1	4/4	Chiba et al., 2000
Sulpiride	6 mg·kg−1 p.o.	Not reported	0/4	Sugiyama, Satoh, et al., 2002
	60 mg·kg−1 p.o.	Not reported	1/4	Sugiyama, Satoh, et al., 2002
	120 mg·kg−1 p.o.	Not reported	2/4	Sugiyama, Satoh, et al., 2002
Terfenadine	3 mg·kg−1 p.o.	Not reported	1/6	Takahara et al., 2006
	30 mg·kg−1 p.o	Not reported	5/6	Takahara et al., 2006

In case a drug was tested in multiple studies, publications were selected that provided the most comprehensive dataset.

Only studies with a defined drug dosing regimen and TdP as explicit endpoint were selected.

Hypokalaemic conditions.

TABLE 3

Cardiac safety testing of drugs and compounds in the anaesthetized canine CAVB model

Drug a	Dose b,c	Peak plasma	TdP b,c	Reference
Almokalant	0.12 mg·kg−1 10 min−1 i.v.	Not reported	9/14	Verduyn et al., 1997
Amiodarone	40 mg·kg−1 for 28 days p.o.	3.5 ± 0.6 mg·L−1	0/7	Van Opstal, Schoenmakers, et al., 2001
AVE0118	0.5 mg·kg−1 5 min−1 i.v.	0.55 ± 0.1 μg·ml−1	0/5	Oros et al., 2006
	3 mg·kg−1 60 min−1 i.v.	1.9 ± 0.5 μg·ml−1	0/5	Oros et al., 2006
	10 mg·kg−1 60 min−1 i.v.	6.1 ± 1.2 μg·ml−1	0/5	Oros et al., 2006
AZD1305	1.08 mg·kg−1 30 min−1 i.v.	1.77 ± 0.29 μM	4/11	Johnson et al., 2012
Azimilide	10 mg·kg−1 5 min−1 i.v.	Not reported	5/9	Van Opstal, Leunissen, et al., 2001
Azithromycin	2 mg·kg−1 5 min−1 i.v.	5.4 ± 1.3 μg·ml−1	0/5	Thomsen, Beekman, et al., 2006
	8 mg·kg−1 5 min−1 i.v.	20.8 ± 4.9 μg·ml−1	0/5	Thomsen, Beekman, et al., 2006
DHE	0.33 mg·kg−1 5 min−1 i.v.	1.2 μM	2/4	Baburin et al., 2018
	0.5 mg·kg−1 5 min−1 i.v.	2.3 μM	0/4	Baburin et al., 2018
Dofetilide	25 μg·kg−1 5 min−1 i.v.	80 nM	10/13	Thomsen et al., 2003
Dronedarone	2 × 20 mg·kg−1 for 28 days p.o.	1.3 ± 0.3 mg·L−1	4/8	Van Opstal, Schoenmakers, et al., 2001
Flunarizine	2 mg·kg−1 2 min−1 i.v.	Not reported	0/8	Oros et al., 2010
Ibutilide	25 μg·kg−1 5 min−1 i.v.	Not reported	6/10	Boulaksil et al., 2011
Istaroxime	180 μg·kg−1 60 min−1 i.v.	Not reported	0/7	Bossu, Kostense, et al., 2018
K201	0.1 mg·kg−1 2 min−1 + 0.01 mg·kg−1 30 min−1 i.v.	450 ± 100 nM	0/7	Stams et al., 2011
	0.3 mg·kg−1 2 min−1 + 0.03 mg·kg−1 30 min−1 i.v.	1080 ± 350 nM	3/7	Stams et al., 2011
Lidocaine	3 mg·kg−1 2 min−1 i.v.	Not reported	0/7	Antoons et al., 2010
LUF7244	2.5 mg·kg−1 15 min−1 i.v.	2.34 ± 1.57 μM	0/7	Qile et al., 2019
Moxifloxacin	2 mg·kg−1 5 min−1 i.v.	4.6 ± 2.5 μg·ml−1	0/6	Thomsen, Beekman, et al., 2006
	8 mg·kg−1 5 min−1 i.v.	22.0 ± 6.8 μg·ml−1	0/6	Thomsen, Volders, et al., 2006
NS‐7	3 mg·kg−1 5 min−1 i.v.	1.6 ± 1.9 μg·ml−1	3/6	Detre et al., 2005
PA‐6	2.5 mg·kg−1 10 min−1 i.v.	5.33 ± 0.63 μM	0/9	Ji et al., 2017
Ranolazine	4 mg·kg−1 0.5 min−1 + 5.6 mg·kg−1 25 min−1 i.v.	22.8 ± 2.3 μM	0/6	Antoons et al., 2010
SEA0400	0.4 mg·kg−1 5 min−1 i.v.	5 ± 1 μM	0/3	Bourgonje et al., 2013
	0.8 mg·kg−1 5 min−1 i.v.	11 ± 2 μM	0/4	Bourgonje et al., 2013
Sertindole	0.2 mg·kg−1 5 min−1 i.v.	Not reported	0/9	Thomsen, Volders, et al., 2006
	1.0 mg·kg−1 5 min−1 i.v.	1.3 ± 0.2 μM	3/5	Thomsen et al., 2003
d‐Sotalol	2 mg·kg−1 5 min−1 i.v.	Not reported	2/8	Thomsen et al., 2004
	4 mg·kg−1 5 min−1 i.v.	Not reported	6/8	Thomsen et al., 2004
Verapamil	0.3 mg·kg−1 5 min−1 i.v.	Not reported	0/3	Bourgonje et al., 2013
	0.4 mg·kg−1 3 min−1 i.v.	Not reported	0/7	Oros et al., 2010
Vernakalant	2 mg·kg−1 10 min−1 i.v.	Not reported	0/8	Varkevisser et al., 2013

In case a drug was tested in multiple studies, publications were selected that provided the most comprehensive dataset.

Only studies with a defined drug dosing regimen and TdP ( Torsade de Pointes)as explicit endpoint were selected.

TABLE 4

Overview of tested anti‐arrhythmic drugs in the CAVB dog model

Compound	Ion channel target	Dose (i.v.)	TdP occurrence		QTc (ms)		STV (ms)		Reference
			Challenge	Suppression/prevention	Challenge	Suppression/prevention	Challenge	Suppression/prevention
Flunarizine (S)	Calcium channel inhibitor	2 mg·kg−1 2 min−1	10/10	0/10	553 ± 40	425 ± 38*	4.5 ± 1.5	1.5 ± 0.6*	Oros et al., 2010
Verapamil (S)	Calcium channel inhibitor	0.4 mg·kg−1 3 min−1	7/7	0/7	566 ± 87	516 ± 90	3.2 ± 1.1	1.5 ± 0.7*	Oros et al., 2010
W‐7 (S)	Calmodulin inhibitor	50 μmol·kg−1 5 min−1	6/6	2/6	523 ± 69	489 ± 88	3.1 ± 1.1	2.7 ± 1.6	Bourgonje et al., 2012
SEA0400 (S)	Sodium–calcium exchanger inhibitor	0.8 mg·kg−1 5 min−1	4/4	0/4	549 ± 95	702 ± 45	12.0 ± 6.4	7.3 ± 3.2	Bourgonje et al., 2013
GS‐458967 (S)	Late sodium current inhibitor	0.1 mg·kg−1 5 min−1	7/7	0/7	609 ± 44	551 ± 77*	4.2 ± 2.5	2.7 ± 0.9	Bossu, Houtman, et al., 2018
Lidocaine (S)	Sodium channel inhibitor	3 mg·kg−1 2 min−1	6/6	2/6	489 ± 41	503 ± 72	3.6 ± 0.8	2.3 ± 0.9*	Antoons et al., 2010
Ranolazine (S)	Sodium channel inhibitor	4 mg·kg−1 5 min−1 + 5.6 mg·kg−1 25 min−1	5/5	2/5	523 ± 69	489 ± 88	4.5 ± 0.8	3.2 ± 0.5*	Antoons et al., 2010
Mexiletine (S)	Sodium channel inhibitor	4.5 mg·kg−1 + 1.5 mg·kg−1·h−1 for 2.5 h	6/8	2/8	nr	nr	nr	nr	Chézalviel‐Guilbert et al., 1995
LUF7244 (P)	Delayed rectifier potassium channel activator	2.5 mg·kg−1 15 min−1	7/7	2/7	498 ± 44	544 ± 48	3.3 ± 2.0	7.6 ± 6.7	Qile et al., 2019
Levcromakalim (S)	ATP‐sensitive potassium channel opener	3 μg·kg−1 3 min−1	7/7	2/7	460 ± 69	481 ± 120	4.9 ± 2.1	2.6 ± 0.9*	Thomsen, Volders, et al., 2006
Nicorandil (S)	ATP‐sensitive potassium channel opener	1–1.5 mg·kg−1 5 min−1	4/10	0/10	nr	nr	nr	nr	Watanabe et al., 2011

Note: Summarizes the different anti‐arrhythmic drugs tested in the CAVB dog model. The selected references serve as an illustrative example and do not represent a comprehensive list of all the studies that have tested the drug in the CAVB dog model.

Abbreviations: CAVB, chronic atrioventricular block; nr, not reported; (P), prevention experiment; (S), suppression experiment; STV, short‐term variability of repolarization; TdP, Torsade de Pointes arrhythmia.

* denotes significant difference between challenge and prevention/suppression.

SAFETY PHARMACOLOGY

Due to its high sensitivity for drug‐induced Torsade de Pointes arrhythmias, the CAVB dog model has been used in many safety pharmacology studies, either in the awake (Table 2) or anaesthetized state (Table 3). In conscious dog studies, the electrophysiological parameters, including Torsade de Pointes, were mostly determined using a Holter recording and analysis system. Drugs were either applied p.o. or i.v. at the start of a 24‐h recording period. Unfortunately, according to the current British Journal of Pharmacology guidelines, studies in the awake conditions are often underpowered by inclusion of less than five animals. In anaesthetized dog studies, electrophysiological parameters are mostly determined by 6‐lead surface ECG and catheter‐mediated endocardial monophasic action potential recording. In these studies, drugs are mostly applied i.v. Many of the studies indicate a dose–response relationship for cardiotoxic drugs, whereas safe drugs are free of Torsade de Pointes even in 10‐fold increase of clinical dosing. In some studies, using non‐anaesthetized animals, drug sensitivity is further enhanced by diuretics‐mediated hypokalaemia. Only three drugs, lidocaine, moxifloxacin and , have been tested in both the awake and anaesthetized conditions in a safety set‐up. Lidocaine and d‐sotalol had similar outcomes. However, the moxifloxacin studies gave conflicting outcomes, that is, absence of Torsade de Pointes in the anaesthetized state versus Torsade de Pointes induction in the awake state, with reported plasma levels even higher in the former condition. However, the means of application, i.v. versus p.o., respectively, differed. Finally, many outcomes from the studies in anaesthetized animals were obtained in a prevention set‐up of anti‐arrhythmic drug testing (e.g. flunarizine, K201, ranolazine and SEA0400), prior to dofetilide infusion.

In general, there is a very good concordance between the absence or occurrence of drug‐induced Torsade de Pointes in the CAVB dog, either awake or anaesthetized and clinical findings (e.g. lidocaine, e Silva et al., 2018; cisapride, Hennessy et al., 2008; sotalol, Haverkamp et al., 1997; and dofetilide, Torp‐Pedersen et al., 1999).

ANTI‐ARRHYTHMICS TESTED IN THE CAVB DOG MODEL

In addition to safety pharmacology testing, the CAVB dog model has proven useful in testing the efficacy of anti‐arrhythmic drugs. These experiments can be performed as either suppression or prevention experiment. In suppression experiments, dogs are challenged with a drug (usually a potassium channel blocker) to evoke Torsade de Pointes. After the occurrence of a predefined arrhythmic endpoint (e.g. three Torsade de Pointes within 10 minutes or one Torsade de Pointes overall, depending on the experiment), the tested anti‐arrhythmic substance is infused to test if the substance can abrogate the rhythm disturbances. For prevention experiments, an initial experiment is necessary to test if a dog indeed shows Torsade de Pointes after the infusion of a challenging drug. If so, this dog is considered ‘inducible’ and will proceed to the actual prevention experiment. During this experiment, the tested anti‐arrhythmic substance is infused shortly before, or co‐infused with, the challenging drug to test if the occurrence of Torsade de Pointes can be prevented.

All drugs that have been specifically developed to diminish ventricular arrhythmias and that have been tested in the CAVB dog model are presented in Table 4. As can be appreciated, arrangement of the drugs in Table 4 is based on which ion channel is targeted to exert its mode of action. Moreover, the majority of the listed experiments are performed in a suppression set‐up (S). Drugs targeting calcium channels directly (e.g. verapamil, Oros et al., 2010) are highly effective in suppressing Torsade de Pointes, whereas the efficacy of downstream targeting of calcium handling is lower (Bourgonje et al., 2013). Overall, the anti‐arrhythmic effect of sodium channel inhibition is good, but anti‐arrhythmic efficacy varies between these drugs. The potassium channel activators, like I Kr (Qile et al., 2019) and I KATP targeting drugs (Thomsen, Volders, et al., 2006; Watanabe et al., 2015), lastly, have also proven to be effective in suppressing and preventing Torsade de Pointes.

Interestingly, the QT interval only shortened significantly after anti‐arrhythmic therapy in a minority of the anti‐arrhythmic drug interventions. This underlines the shortcoming of the QT interval as a measure for risk of arrhythmic events, as it does not reflect the reduction in the incidence of Torsade de Pointes in these experiments (Bossu et al., 2017). STV performs better as an indicator for arrhythmic risk in these experiments, as it decreased significantly after anti‐arrhythmic treatment in the majority of the interventions. However, as STV does not decrease significantly in all experiments, it is not perfect. Apparently, and not unexpected, additional factors play a part in the determination of the arrhythmic threat.

As mentioned above, Table 4 summarizes the efficacy of drugs that have been designed to treat ventricular arrhythmias. Therefore, some drugs that have also been tested in a prevention or suppression set‐up in the CAVB, but were not primarily designed to target ventricular arrhythmias, have been omitted. An example is istaroxime, a positive inotropic drug. As current inotropic agents are notorious for inducing ventricular arrhythmias, this study should be primarily considered as safety pharmacology testing rather than anti‐arrhythmic testing. It is thus not surprising that istaroxime displayed only a mild preventive effect on the occurrence of Torsade de Pointes (6/6 vs. 4/6) (Bossu, Kostense, et al., 2018). The same goes for the drugs K201 and AVE0118, which were designed to target atrial fibrillation. As their modes of action might predispose to Torsade de Pointes, these studies should also be primarily considered as safety pharmacology testing (Oros et al., 2006; Stams et al., 2011).

STRENGTHS, WEAKNESSES AND PERSPECTIVE OF THE CAVB DOG MODEL

Due to its concordance to human cardiac physiology, the dog is one of the species recommended by the FDA for in vivo cardiotoxicity evaluation (ICH S7B). When comparing a single drug in multiple systems, it was found that the safe I Kr blocker moxifloxacin, at therapeutical concentrations, did not induce Torsade de Pointes arrhythmias in the anaesthetized CAVB dog and methoxamine‐sensitized rabbit, whereas it did induce early after depolarization (considered as an arrhythmogenic endpoint) in isolated adult cardiomyocytes from rabbit and the CAVB dog (Nalos et al., 2012). Furthermore, the CAVB dog is a sensitive model for non‐I Kr blocking drugs, like those that inhibit the inward rectifier channel (e.g. PA‐6, Ji et al., 2017), sodium channel (e.g. lidocaine, Weissenburger et al., 1991), L‐type calcium channel (e.g. verapamil, Oros et al., 2010) or even multiple channel blockers (e.g. flunarizine, Oros et al., 2010). Another benefit of the model is its applicability for serial testing due to its high reproducibility of, for example, dofetilide‐induced Torsade de Pointes arrhythmias between subsequent experiments (Oros et al., 2008). This characteristic can be exploited in a prevention set‐up, in which a test drug is infused in an ‘inducible’ CAVB dog, followed by dofetilide. Furthermore, multiple compounds can be tested subsequently for arrhythmia inducibility and other electrophysiological parameters within the same animal, allowing robust drug comparisons. Thirdly, in line with inducibility is the remarkable finding that approximately 20%–25% of the CAVB dogs display no Torsade de Pointes following dofetilide infusion, whereas all animals undergo similar experimental treatments (Oros et al., 2008). This reflects, to some extent, the human observation that drug‐induced arrhythmia is only apparent in predisposed people, according to the repolarization reserve concept. The CAVB dogs mostly have a mixed genetic background and, in that aspect, mirror the human population. Given the available dog genome sequence, one can envision genotype–phenotype comparisons between inducible and non‐inducible animals for a defined set of genes involved in cardiac electrophysiology, which may provide genetic information that determines susceptibility for drug‐induced arrhythmia in these animals and potentially in humans. Fourthly, in vivo and in vitro testing can be performed using the same animal. Whereas arrhythmogenicity scoring can display different outcomes, as indicated above, effects on repolarization duration and many other parameters can be compared. Furthermore, in vivo observations can be further analysed using isolated cells from the same animal. For example, ranolazine was tested as preventing dofetilide‐induced arrhythmia in vivo and subsequently was demonstrated to inhibit late sodium current in isolated cardiomyocytes, despite a partial down‐regulation of late sodium current in the CAVB dog (Antoons et al., 2010). Moreover, the predictive value of cardiac safety testing in the CAVB dog model is, after so many years of experience, well characterized (Gralinski, 2003). Finally, a practical benefit of the CAVB model is its size. This allows use of standard clinical tools like catheters, pacemakers, implantable cardioverter‐defibrillators (ICDs) and echocardiography probes. It will also provide sufficiently large plasma amounts for multiple tests and isolated cardiomyocytes for subsequent analysis.

Obviously, the CAVB model also comes with several limitations. First of all, it is an expensive system. A single purpose bred experimental dog, including housing and experimentation, will cost approximately €4000. In an academic setting, these costs per animal obviously will not allow large n‐numbers, despite the ability for serial testing. Not only is the use of dogs in research costly, but it is also labour intensive requiring specific expertise, the remodelling process is time consuming and there are many demands to comply with appropriate housing for dogs. Many studies with dogs are being performed in collaboration with the pharmaceutical industry, and therefore, study variables and outcomes could be confidential until scientific publication of the results, which hampers rapid academic exchange of preliminary results. Secondly, societal acceptance for the use of dogs in research is challenged, in the past but even more today and requires continuous stringent justification within ethical committees (obviously) and to the public. This challenged societal acceptance can lead to a diminished openness on research in dogs to the public. The ethical and cost constraints have most likely also contributed to the reluctance for developing genetically engineered dog models, whereas these have been developed in other larger animal models such as rabbit, pigs and goats (Baczkó et al., 2020; Li et al., 2016; Polejaeva et al., 2016). This could limit future developments of dog models that mimic the clinical situation more accurately, such as atherosclerosis‐associated ischaemia and rhythm disturbances.

Due to the millennia of coexistence and breeding programmes, the dog is considered as man's best friend. However, its physiology as indicated above is still different in several aspects compared with humans, which still gives rise to a translational gap. Moreover, Torsade de Pointes arrhythmias are a very specific type of polymorphic ventricular arrhythmias that are clinically only observed in a setting of prolonged repolarization and bradycardia (Al‐Khatib et al., 2018). Torsade de Pointes is the classic type of arrhythmia that is associated with the congenital LQTS. Congenital LQTS is a cardiac repolarization disorder characterized by a prolonged QT interval, caused by a critical mutation in one of the cardiac ion channels, with a prevalence of about 0.5% (Schwartz et al., 2009). The concordance between the CAVB dog model and the human congenital LQTS is thus a diminished repolarization capacity that can give rise to Torsade de Pointes arrhythmias. However, the contractile and structural remodelling that occur in the CAVB dog are different from this patient population. Heart failure, whether compensated or not, is no typical trait of the congenital LQTS patients (Vos et al., 1998). A transgene LQTS dog would closer approximate these patients. The structural remodelling leading to ventricular hypertrophy in the CAVB dog does show some phenotypical overlap with hypertrophic cardiomyopathy (HCM). Hypertrophic cardiomyopathy is caused by a mutation in contractile sarcomeric proteins, which leads to cardiac hypertrophy with a wide array of clinical manifestations. Patients with hypertrophic cardiomyopathy, however, do not typically show QT prolongation, bradycardia or AV dyssynchrony, nor do they develop Torsade de Pointes arrhythmias. In fact, ventricular tachycardia, ventricular fibrillation and atrial fibrillation are the most common arrhythmias in this patient population. Moreover, hypertrophic cardiomyopathy patients display a decrease in cardiac contractility and develop cardiac fibrosis, whereas the opposite holds true for CAVB dogs (Elliott & McKenna, 2004). Also in this case, a genetically engineered dog model would give a better approximation of this patient population. All in all, the canine CAVB model does reproduce some aspects of hereditary LQTS and hypertrophic cardiomyopathy, but the differences are too extensive to reproduce these syndromes completely. Clinically, ventricular tachycardia and ventricular fibrillation are observed far more often than Torsade de Pointes and therefore make up a larger contribution to sudden cardiac death than Torsade de Pointes. It is thus questionable to which extent obtained insights into the mechanisms underlying Torsade de Pointes are applicable to target sudden cardiac death in humans in general (Al‐Khatib et al., 2018). Contributing to this uncertainty is that the evoked Torsade de Pointes in the CAVB dog model are majorly triggered activity based, as can be explained by the alterations in calcium handling and sympathetic innervation as explained above. As the formation of re‐entry circuits is also a well‐established factor in the formation and perpetuation of ventricular arrhythmias in humans, this could further enhance the translational gap that the CAVB dog model poses in finding targets for diminishing life‐threatening arrhythmias (El‐Sherif et al., 1981).

The use of the CAVB dog in preclinical research has dropped significantly over the past years. Taking the costs, required expertise, ethical constraints, lack of transgenic opportunities and translational obstacles into account, its use will probably drop even more. It will become more and more difficult for the academia to cover the costs that using the CAVB dog model entails. In this regard, collectives such as the Health and Environmental Sciences Institute (HESI) are interesting, which is currently supporting the CiPA initiative. Their aim is to centrally unite academic, industry and government and regulatory parties to resolve health challenges. This would encompass combining the expertise of the academy with the funds of the industry and, by means of centralization, improving the generalizability of the dog model.

In this regard, it is also important to place the CAVB dog in the perspective of other available animal models in preclinical cardiac research. Firstly, mouse models have the major advantage that they are very suitable for genetic modification. Given the short gestation and large litters, they can produce many of transgene animals in a short time. However apt to develop transgene models and gain insight in certain mechanisms, the mouse is not very suitable further on in the translation process in the arrhythmogenic research field. The mouse action potential is very different from the human, due to a different expression of cardiac ion channels and altered relative contributions of calcium and potassium currents to the action potential. Moreover, mice have a resting heart rate of 600–800 bpm and a different cardiovascular anatomy (Clauss et al., 2019). The electrophysiological properties of rabbits however are rather similar to the human situation, especially in the ventricles. Almost all major human ion currents are also present in rabbit ventricular myocytes, with some differences in relative contribution and therefore a fairly comparable ventricular action potential waveform. The atrial electrophysiology however differs significantly from humans. This renders rabbits very suitable for studies on ventricular repolarization. Several transgenic LQTS rabbit models have been developed, in addition to a CAVB rabbit model. The CAVB rabbit displays a high incidence of spontaneous Torsade de Pointes (approximately 70%) and the transgenic LQTS rabbits vary, depending on the type of mutation, from exerting spontaneous VT to a reduced repolarization capacity, which needs an additional hit to promote Torsade de Pointes (Baczkó et al., 2020; Clauss et al., 2019; Tsuji et al., 2002). The costs of rabbits are lower and appropriate housing is easier to arrange than for dogs. Additionally, transgenic models are an option, even though not many are available yet. However, no data on reproducibility of arrhythmia induction in the same animal are available, and therefore, serial testing is not possible, contrary to the CAVB dog. This reduces the financial benefit of the initially cheaper rabbits and limits certain study set‐ups. Moreover, the rabbit CAVB model encompasses a failing heart and requires immediate pacing following acute AV block, contrary to the compensated canine CAVB model and the differences in size and anatomy further limit the rabbit's translation into clinical practice. Although built almost 20 years ago (e.g. Tsuji et al., 2002), the number of publications using the CAVB rabbit model in preclinical drug research is scarce when compared with the dog CAVB model, for unknown reasons. Finally, the predictive value of preclinical research in rabbits has not been characterized, contrary to the CAVB dog model. Rabbit models could however be a future option for cardiac safety and repolarization‐related research but need to be further developed. The use of pigs in preclinical research is also more socially accepted than dogs and there is an option to make transgene models as well. The pig does have a less similar anatomy of the conduction system and make‐up of ion currents to humans, compared with dogs. The coronary anatomy and atrial electrical properties on the other hand are more similar to humans (Clauss et al., 2019). They could be considered to validate data from rabbit models, to bridge the translational gap that the rabbits' size, anatomy and atrial electrophysiology pose. Finally, as mentioned before, human cardiomyocytes derived from induced pluripotent stem cells carrying rare diseases could be applied to evaluate cellular and molecular mechanisms. As they lack the complex integration of cardiac electrical activity, pump function, haemodynamic feedback, neural regulation, electrolyte homeostasis, display of SDR and their collective response to pharmacological treatment, the known targets for these cardiomyocytes should be further investigated from here on.

Underlining the complexity of advancing in the anti‐arrhythmic research field, only a few papers have been published on new anti‐arrhythmic targets in the last decade. For example, SK2 (K channels and connexin43‐based gap junctions were added to the potential anti‐arrhythmic repertoire (Hamilton et al., 2020; Lucero et al., 2020). Obviously, knowledge is progressing but at its best at a modest pace.

CONCLUSIONS

Understanding of the mechanistic processes underlying cardiac arrhythmia in the CAVB dog is steadily progressing, but slowly. The complex and multifactorial cardiac compensatory remodelling process upon CAVB provides multiple targets on which drugs may act to induce cardiotoxicity. The CAVB model is placed in the latter stages of the drug development process and its integrated outcome, that is, cardiac arrhythmia, can be mirrored against more reductionistic endpoints from dedicated in silico and in vitro models. Outcomes of cardiac safety testing in the CAVB dog demonstrate good concordance with clinical observations on adverse drug effects. The CAVB dog is amenable for evaluating anti‐arrhythmic compounds with diverse molecular targets. The combination of determining in vivo and in vitro parameters from the same model strengthens mechanistic insights and potentially translating power. Use of the model is challenged by its costs, societal acceptance and the limited occurrence of Torsade de Pointes (the specific type of arrhythmia that is evoked by the CAVB dog model) in clinical practice. Therefore, the mechanistic insights into the occurrence and perpetuation of arrhythmias that have arisen from the CAVB dog might be limited outside cardiac safety and anti‐arrhythmic testing. Moreover, the lack of (opportunity to develop) transgene models will limit future developments that can improve the translational gap. We therefore expect that the use of the CAVB model will be under increasing pressure. Nevertheless, the CAVB model deserves its important role in drug‐related cardiac arrhythmia research. Compared with other available in vitro and in vivo systems, the CAVB dog model is best characterized in relation to clinical observations. The other available models, however promising, should be further developed and validated with clinical data before they can replace the CAVB dog model completely. In the meantime, centralized collaborations between academic, industrial and government parties can help cover the increasing costs, combine expertise and improve generalizability of drug target validation and safety screening using the CAVB dog model.

Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in the IUPHAR/BPS Guide to PHARMACOLOGY http://www.guidetopharmacology.org and are permanently archived in the Concise Guide to PHARMACOLOGY 2019/20 (Alexander et al., 2019).

AUTHOR CONTRIBUTIONS

V.L. and M.v.d.H. designed the review, collected and analysed the literature and wrote the draft manuscript. M.V. critically reviewed the draft version and contributed to the final version. All authors read and approved the final version of the manuscript.

CONFLICT OF INTERESTS

M.V. is supported by collaborations with Medtronic and Amgen. V.L. and M.v.d.H. declare no conflicts of interest.