Validated animal models for antiseizure drug (ASD) discovery: Advantages and potential pitfalls in ASD screening

INTRODUCTION

The early identification of new antiseizure drugs (ASDs) for the symptomatic treatment of epilepsy has, since 1937, largely depended on in vivo testing in one or more rodent seizure and epilepsy models (Bialer and White, 2010). A number of animal models have been developed since the initial identification of phenytoin in 1937 by Merritt and Putnam using the cat maximal electroshock (MES) seizure test (Putnam and Merritt, 1937). Each of these has certainly contributed to a greater ability to differentiate the potential utility of an investigational drug, but only three of the currently available animal models have been “clinically validated”: the MES and subcutaneous pentylenetetrazol (scPTZ) acute seizure tests, as well as the kindled rodent model of chronic hyperexcitability (Table 1). This review will describe the history and utility of these three preclinical models to support the early identification of anticonvulsant activity. We will then discuss the clinical validity of these models as well as define their place in current and future preclinical development to identify more efficacious and better tolerated ASDs for the treatment of drug-resistant focal onset seizures.

Table I:

Clinically Validated Animal Seizure and Epilepsy Models

Model	Species	Behavioral Phenotype	Clinical Correlate	Clinical Validation	Widely used for drug screening	Reference
Maximal Electroshock Seizure (MES) Model	Mice/rats	Tonic Extension of forelimbs and hind limbs, followed by all limb clonus	Generalized Tonic-Clonic Seizures	Contributed to identification and development of phenytoin	Yes	Merritt and Putnam. 1937, 38’
scPentylenetetrazol (scPTZ)	Mice/rats	Myoclonic jerks followed by unilateral forelimbs and bilateral clonus, vibrissae twitching	Generalized myoclonus and spike-wave seizures	Led to discovery of trimethadione, phensuximide, and ethosuximide	Yes	Everett and Richards, 1944; Chen et.al., 1951; Loscher, 2016
Amygdala-Kindling	Rats	Unilateral and bilateral forelimb clonus that progresses to rearing and falling	Focal seizures that evolve to disruption of awareness	Only model to correctly identify anti-seizure activity of levetiracetam	Yes	Loscher and Honack, 1993

While thousands of new chemical entities may enter the initial screening process to define anticonvulsant activity, very few compounds will advance beyond the early identification phase and proceed to advanced Investigational New Drug (IND)-enabling toxicology studies. Even fewer compounds will proceed to clinical candidate status. It certainly reasons that the more closely the animal model reproduces key aspects of any given seizure type or syndrome, the greater the likelihood that an investigational ASD, which is effective in that model, will subsequently demonstrate efficacy in clinical trials. Indeed, we have been fortunate in the epilepsy field to have three rather simple rodent models that display significant predictive validity for efficacy in clinical seizures: the MES, scPTZ, and kindling models.

As discussed below, the use of one or more of these three models has predicted the clinical efficacy of all of the currently available ASDs available for the treatment of epilepsy. Unlike epilepsy, other CNS diseases like migraine and bipolar disorder have not benefitted from such similarly predictive animal models (Loscher, 2016). In this regard, epilepsy has enjoyed significant success in advancing numerous novel therapies to the clinic (Perrin, 2014).

THE “IDEAL” MODEL SYSTEM

The 2014 NINDS Epilepsy Benchmarks area III.C. calls for models that are aligned with etiologies and clinical features of human epilepsies, especially treatment resistant forms (Dlugos et al., 2016). A “true” rodent model of epilepsy should exhibit seizures that evolve spontaneously following a post-insult latent period or arise within a developmental timeframe consistent with the human condition. Furthermore, the ideal model of pharmacoresistant seizures should display a pharmacological profile that is resistant to at least two of the existing ASDs (see (Stables et al., 2003)). Of general relevance to drug development, the ideal model should be amenable to high-throughput screening if it is to be used to evaluate large libraries of investigational compounds. That said, the highly heterogeneous nature of human epilepsy, coupled with the complexity of the numerous seizure phenotypes and syndromes involved, presents a high likelihood that we will never have just one model system that can accurately predict the full therapeutic potential of any given investigational agent. In 1945, Norbert Weiner summed up the state of model systems quite adeptly when he suggested that the “best material model of a cat is another cat, preferably the same cat.” Thus, the epilepsy community should consider the fact that no one model is likely to be perfect when animals are employed for drug discovery for a human condition. Instead, the initial model should be chosen on the basis of “best fit” for the situation (Loscher, 2016). Further evaluation conducted in a battery of well-defined animal models certainly provides predictive value to support the therapeutic potential of an investigational ASD, as well as important proof-of-concept information that could help to define the clinical potential of an ASD. The emergence of a number of etiologically-relevant models of pediatric encephalopathy, infection-induced epilepsy, and trauma-induced epilepsy have provided an important platform for hypothesis testing and target validation to better understand the processes underlying ictogenesis and epileptogenesis; however, the majority of these models are simply not amenable to high-throughput screening of drugs for the symptomatic treatment of epilepsy. In contrast, these syndrome-specific models may play an important role in the identification of therapies capable of preventing, delaying, or modifying epilepsy in susceptible individuals. For the purpose of this review, we are focusing our comments on the three clinically validated animal models for: 1) Generalized Tonic Clonic (GTC) seizures - i.e., the rodent MES test; 2) generalized absence seizures - i.e., the s.c PTZ test; and 3) focal seizures - i.e., the kindling model of mesial temporal lobe epilepsy.

WHAT DOES IT MEAN TO BE “CLINICALLY VALIDATED?”

When it comes to the use of any given animal model for drug development applications for any pathological condition, several factors should be considered. Among these, face and predictive validity for efficacy and tolerability are particularly important. Preclinical testing of a drug in a highly-validated animal model can provide important data on the expected clinical performance of the drug, as well as inform on the potential for adverse effects liability in the target patient population. In addition, proper study design, execution and data reporting will help make the resultant preclinical data more reproducible and translatable to the clinic. Importantly, meaningful results emerge from animal testing when the preclinical strategy is part of a well-developed translational plan rather than a single experiment, often relying on clinical observations to guide preclinical development (Barker-Haliski et al., 2014). With respect to efficacy testing of an investigational ASD, the more validated the animal model, the more likely the preclinical results will predict the clinical outcome.

Ideally, any given animal model will display:

1. Face validity: i.e., when there is similarity in biology and symptoms between the animal model of interest and the human disease state being modeled. Although important for the validation of a disease model, assessing face validity is often hampered by a lack of understanding of the biology underlying the disease symptoms. This is particularly true for a highly complex disease state like epilepsy where there is such heterogeneity in the cause, pathology, and response to treatment; in this regard, the acute MES and scPTZ models lack face validity. Mounting evidence is increasingly demonstrating, however, that the kindling models do recapitulate many aspects of human epilepsy, including neuroinflammation (Loewen et al., 2016), cognitive and behavioral disturbances (Albertini et al., 2017; Barker-Haliski et al., 2016b; Koneval et al., 2018; Moller et al., 2018), and even spontaneous recurrent seizures (Brandt et al., 2004), and thus exhibit a greater degree of face validity than the MES/scPTZ models.

2. Predictive validity: i.e., clinically effective interventions demonstrate a similar effect in the model. This is where the MES, scPTZ, and kindling models have been shown to possess the greatest validity. At least one of these 3 models is sensitive to every clinically-available ASD and will be discussed in greater detail below.

3. Target validity: i.e., the target under investigation should have a similar role in the disease model as in the clinical situation. Unfortunately, the MES and scPTZ models for the most part lack “target validity”; these acute seizure models are conducted in neurologically-intact rodents. This is likely why most of the ASDs identified to-date in the MES/scPTZ models largely engage a single neuronal process: excitation/inhibition, which is a universal process within all neurons. However, as with face validity, kindling models demonstrate many of the neuropathological and behavioral changes associated with epilepsy (Albertini et al., 2017; Loewen et al., 2016; Moller et al., 2018; Remigio et al., 2017), which likely results in more epilepsy-specific targets becoming expressed. This more epilepsy-like neuronal substrate has thus been useful to identify mechanistically-novel agents, as with levetiracetam (Klitgaard, 2001; Loscher and Honack, 1993). Overall, kindling models demonstrate more target validity than the MES/scPTZ models and are thus more appropriate models of epilepsy for ASD development.

SCREENING APPROACH TO INITIAL ASD DISCOVERY

For the MES and scPTZ tests, a compound of interest is administered systemically to a minimal number of neurologically-intact normal laboratory mice or rats and then, at the peak time of exposure or pharmacodynamic effect, the animal is challenged with the seizure stimulus (e.g. MES or the chemoconvulsant PTZ) and the seizure outcome evaluated (e.g. tonic hindlimb extension for the MES test or minimal clonus of the forelimbs and vibrissae twitching (scPTZ test; (Barker-Haliski et al., 2018)). Most typically, the outcome is whether the treatment prevented the behavioral seizure. Because neurologically-intact rodents are most often employed in the MES and scPTZ tests for ASD identification, the main conclusion that can be made when a drug is found to be efficacious in one or both of these models is that it is intervening in the ictogenic process by either elevating seizure threshold or preventing seizure spread.

In contrast to the MES and scPTZ tests, the kindling model, first described by Goddard and colleagues (Goddard et al., 1969), is a chronic model of network hyperexcitability that results when repeated stimulation with an initially benign electrical stimulus elicits evoked seizure activity after days to weeks of repeated stimulation. Although there are indeed chemoconvulsant and even optogenetic kindling models (Cela et al., 2019; Krug et al., 1997), we will herein only focus our discussion on the electrical kindling models. There is a myriad of methods to deliver the electrical stimulus in laboratory animals: e.g., via a depth electrode implanted into a limbic brain region; e.g., amygdala, hippocampus, perirhinal cortex, or even with topical stimulation of the corneas (Barker-Haliski et al., 2018). Regardless of the approach, the initially non-convulsive electrical stimulus induces a permanent seizure susceptibility and lasting brain alterations that are similar to that found in human temporal lobe epilepsy (Sato et al., 1990). As discussed below, the pharmacology of the kindled rodent suggests that this model displays the best predictive validity for focal epilepsy. However, it is important to note that at the time that an animal is considered to be fully kindled, seizures are still evoked and not spontaneous. Further, for early drug screening applications, the effect of an investigational agent on the subsequent expression of the evoked seizure is the predominate outcome measure. Due to these technical considerations, the kindled rodent has not historically been considered a true model of epilepsy wherein the animal displays spontaneous, unprovoked seizures suitable for late-stage drug development. Nonetheless, mounting evidence in multiple kindling approaches in both mice and rats increasingly contradicts the notion that kindling is not a useful model of human focal epilepsy. Kindled rodents demonstrate notable cognitive and behavioral deficits consistent with human epilepsy (Albertini et al., 2017; Barker-Haliski et al., 2016b; Meeker et al., 2019; Moller et al., 2018; Remigio et al., 2017), neuroinflammation (Loewen et al., 2016), and there are even pharmacoresistant kindled rodents that may be highly useful to inform preclinical differentiation studies and/or clinical studies (Koneval et al., 2018; Loscher et al., 1993; Srivastava and White, 2013). Thus, these data instead demonstrate that the kindled rodent is the only validated model most frequently used for early ASD screening that recapitulates many comorbid features of human temporal lobe epilepsy, including anxiety-like behaviors, depression, and cognitive deficits. Kindled rodents are thus much more well-suited for early-phase moderate-throughput ASD development than acute seizure models in neurologically-intact rodent models (e.g. MES/scPTZ) because they more closely reproduce human focal epilepsy.

As evidenced by the large number of new ASDs discovered and developed for the treatment of focal onset seizures since 1937 (Figure 1), it is hard to argue against the value that the MES, scPTZ, and kindled rodent models bring to the early discovery phase for the identification of promising anticonvulsant agents. These tests are relatively simple to implement and can provide useful insight into the potential for clinical efficacy of many mechanistically-distinct anticonvulsant agents. For a detailed discussion on the approaches to ASD discovery and development procedures in the MES, scPTZ, and kindling models, the reader is referred to a detailed methodological guideline outlining the use of common data elements for pharmacological studies for epilepsy (Barker-Haliski et al., 2018), extensive reviews (Barker-Haliski and White, 2015; Loscher, 2017; White and Barker-Haliski, 2016), or a practical application (Barker-Haliski et al., 2017a). Through numerous decades of use, adult and pediatric patients worldwide have benefited from the strategic screening approach using these models and thus it reasons that for the continued discovery of potentially impactful anticonvulsant agents, there will continue to be a position for these models going forward.

Figure 1.

The ivPTZ seizure threshold test is useful to differentiate whether an investigational compound has the potential to raise or lower seizure threshold. The effect of ASDs and the local anesthetic lidocaine on seizure threshold has been documented in clinical case reports, suggesting that the ivPTZ model is a valid preclinical model to define whether a compound can influence seizure threshold. A) The prototype ASDs, carbamazepine and valproic acid, when tested at their previously determined time of peak effect (0.25 hour pretreatment time) dose-dependently increases the threshold to first twitch (n = 10 mice/treatment group; VEH = 0.5 % methylcellulose, i.p.). Barker-Haliski and White, unpublished data. B) The prototype ASDs, carbamazepine and valproic acid, when tested at their previously determined time of peak effect (0.25 hour pretreatment time) dose-dependently increases the threshold to sustained clonus of the fore- and hindlimbs (n = 10 mice/treatment group; VEH = 0.5 % methylcellulose, i.p.). Barker-Haliski and White, unpublished data. C) The administration of the analgesic, lidocaine, has been reported to reduce seizure threshold in patients with epilepsy when delivered at high doses. I.p. administration of lidocaine to CF-1 mice dose-dependently reduces threshold to first twitch when tested at the previously determined time of peak effect (0.25 hour pretreatment time; n = 10 mice/treatment group; VEH = 0.5 % methylcellulose, i.p.). From NINDS PANAChE database (https://panache.ninds.nih.gov, accessed June 2019). C) I.p. administration of lidocaine to CF-1 mice dose-dependently reduces threshold to sustained clonus when tested at the previously determined time of peak effect (0.25 hour pretreatment time; n = 10 mice/treatment group; VEH = 0.5 % methylcellulose, i.p.). From NINDS PANAChE database (https://panache.ninds.nih.gov, accessed June 2019).

WHAT DO THE MES, SCPTZ, AND KINDLING MODELS ACTUAL “MODEL?”

As noted above, the MES, scPTZ, and kindling models are currently the only “clinically validated” models available for the early evaluation of investigational ASDs. This conclusion is based almost entirely on the observation that these three models all display “Predictive Validity.”

The MES test has been the most commonly used initial screen employed in the search for a new ASD (Bialer and White, 2010) and it will likely remain a work horse for most ASD development programs going forward. Although once considered a model of focal seizures (Krall et al., 1978a; Krall et al., 1978b), the MES test is primarily now considered a model of GTC seizures (Table 1).

The scPTZ test is generally considered a useful model of generalized non-convulsive myoclonic and generalized spike-wave seizures (Loscher, 2016; White et al., 2006). Validation of the scPTZ test was provided largely by the work of Everett and Richards when they demonstrated that trimethadione was effective against PTZ-induced seizures (Everett and Richards, 1944) . Lennox subsequently demonstrated that trimethadione was effective against absence seizures, but was ineffective or worsened generalized tonic-clonic seizures (Lennox, 1945). Trimethadione’s clinical success and its ability to block PTZ-induced threshold seizures provided sufficient evidence to establish the PTZ test as a model of generalized absence seizures. Five years later, the PTZ test was used to identify phensuximide, methsuximide and ethosuximide (Chen et al., 1951). The much-improved tolerability of the succinimides over the oxazolidinediones led to the rapid replacement of trimethadione for the treatment of generalized absence epilepsy (Chen et al., 1951; Loscher, 2016). In addition, the PTZ test correctly predicted the lack of human efficacy of phenytoin (Everett and Richards, 1944) and thus provided a means to quickly differentiate the potential clinical utility of investigational ASDs.

Subsequent investigations found that valproic acid, phenobarbital, gabapentin, tiagabine, and the benzodiazepines were effective at blocking clonic seizures induced by scPTZ. Based on the above argument, one would predict that these three ASDs would be effective against spike-wave seizures. In contrast, clinical experience has found just the opposite: i.e., the barbiturates, gabapentin, and tiagabine all aggravate spike-wave seizure discharge (Bazyan and van Luijtelaar, 2013; Coenen et al., 1995; Wilder et al., 1971) in patients with spike-wave or generalized absence seizures. However, barbiturates can be used in juvenile myoclonic epilepsy (Mantoan and Walker, 2011); a finding that suggests that the PTZ model may also be useful as a model of myoclonic seizures. The scPTZ model is useful to identify compounds that may be effective in patients with primary generalized spike-wave and myoclonic seizures; however, the ultimate model is the patient and efficacy data in animals should be considered with this in mind. Preclinical efficacy data in any seizure or epilepsy animal model should primarily serve as a guide to inform a clinical study. There are of course certain limitations to preclinical animal data, but when it is considered in context with the clinical population being studied, it can be extremely useful. The final clinical indication can only be determined by assessing efficacy in the seizure type of interest. In this regard, the scPTZ model is adeptly positioned to support the initial identification of potentially promising agents for further differentiation and clinical study.

It is presently unclear whether preclinical studies can extensively predict whether an agent may or may not be effective for spike-wave discharges in absence epilepsy and/or myoclonic seizures purely on the basis of activity in the scPTZ test. For this reason, the WAG/Rij and GAERS models may be more beneficial to ASD development for anti-absence medications than the scPTZ test (Gower et al., 1995; Marescaux and Vergnes, 1995). These rats, clearly exhibit spontaneous spike-wave discharges and as such, may be considered a fourth class of validated ASD screening model; however, the discussion of their utility in ASD development is beyond the scope of this present review as we are primarily focusing on generalized and focal seizure disorders.

Further, while sodium channel antagonists such as phenytoin, carbamazepine, and lamotrigine display mixed clinical effects against generalized absence seizures, they do display consistent preclinical inefficacy in the scPTZ test. Specifically, phenytoin and carbamazepine are both ineffective against human spike-wave seizures and are both inactive in the scPTZ test. In contrast, lamotrigine is effective against human generalized absence, but it is inactive in the scPTZ test. One potential explanation for this clinical divergence in activity is the hypothesized additional mechanism of lamotrigine on h-channels (Poolos et al., 2002), which may function together with T-type Ca²⁺ channels to promote the synchronized activity of the thalamocortical network characteristic of absence seizures. Further, lamotrigine, which is often used to treat absence epilepsy when other front-line options are not employed; i.e., ethosuximide and valproic acid, may worsen myoclonic seizures. Thus, positive results in the scPTZ test should be confirmed by similarly positive results in one or more of the genetic models of absence epilepsy; e.g., the GAERS and WAG/Rij rat models of spike-wave seizures (Coenen and Van Luijtelaar, 2003) as they are likely more predictive of clinical efficacy in absence epilepsy, before concluding that an ASD has potential utility in treating human spike-wave seizures. For example, lamotrigine is effective against spontaneous spike-wave discharges in the Long-Evans rat (Huang et al., 2012) and in vitro against simple thalamocortical bursts (sTBCs) believed to model absence-like seizures (Gibbs et al., 2002). Carbamazepine, however, aggravates spike-wave discharges in the GAERS rat (Wallengren et al., 2005) and both carbamazepine and phenytoin are ineffective against sTBCs (Zhang and Coulter, 1996). The scPTZ model nonetheless provides an important moderate-throughput test that is distinct from the MES assay and that can identify potentially effective therapies that should be evaluated further for their potential efficacy in more etiologically-relevant models, e.g. genetic models of spike-wave discharges if that is the clinical condition to be targeted.

As eluded to above, the PTZ test can also be used to determine whether a newly-identified drug possesses any liability to ‘unmask’ or worsen spike-wave seizures in a susceptible patient population; i.e., phenytoin, carbamazepine, phenobarbital, vigabatrin, gabapentin, and tiagabine may worsen human absence seizures; an effect that is supported by animal studies wherein EEG spike-wave discharges, and not behavioral seizures, serve as the outcome measure (Coenen and Van Luijtelaar, 2003; Marescaux and Vergnes, 1995). This is an important differentiating point that should be considered when evaluating drugs in the scPTZ test; results from behavioral studies wherein the outcome measure included attenuation of the phenotypic myoclonic jerk and clonus falsely predicted that tiagabine and vigabatrin might be effective against generalized absence epilepsy. Moreover, as noted above, the scPTZ test failed to identify the now established clinical utility of lamotrigine and levetiracetam against generalized absence epilepsy (see (Loscher, 2016) for further discussion). In spite of these limitations, a variation on PTZ delivery via the intravenous route (i.v.PTZ) is an extremely useful test to define the effect of an ASD on seizure threshold (Piredda et al., 1985; Swinyard and Kupferberg, 1985); which may be broadly important for drug development. This test defines the extent to which an investigational agent can either elevate or reduce the amount of PTZ required to induce either a myoclonic jerk and full generalized clonus of the vibrissae and/or limbs (Barker-Haliski et al., 2018; Mandhane et al., 2007). While this acute seizure test is not “clinically validated” like the acute MES or scPTZ models, it does provide incredibly useful insight to understand if an investigational agent, whether intended for the management of epilepsy or any other condition, carries the potential to negatively affect seizure threshold in mice; an effect that is also replicated in humans. For example, lidocaine significantly and dose-dependently reduces seizure threshold in CF-1 mice in the ivPTZ assay (Figure 1; data from NINDS PANAChE database accessed June 2019), consistent with the effects observed with high-dose lidocaine in patients with epilepsy (DeToledo et al., 2002; Usubiaga et al., 1966). Further, the ivPTZ assay provides useful differentiation data to complement studies with the MES and scPTZ test, which may altogether increase the confidence in the clinical anticonvulsant potential of a candidate ASD or other investigational agent that may be concurrently administered to the patient with epilepsy.

In contrast to the MES and scPTZ tests, the kindled rodent model of mesial temporal lobe epilepsy is the only clinically-validated chronic seizure model. While most kindled rodents do not display spontaneous recurrent seizures under common kindling conditions at the point of enrollment into drug screening protocols (Barker-Haliski et al., 2018; Barker-Haliski et al., 2017a; Matagne and Klitgaard, 1998), it is well-established that kindled rodents do exhibit spontaneous seizures (Brandt et al., 2004; Brandt et al., 2003). However, it should be emphasized that spontaneous seizures are not utilized as the primary outcome measure when evaluating ASD efficacy in the kindled rodent. Nonetheless, kindled rodents have become a highly valuable model for drug differentiation. Further, the amygdala kindled rat model was the only model to correctly identify the anticonvulsant activity of levetiracetam (Loscher and Honack, 1993). Since the implanted electrode-induced kindling phenomenon in rats was first described by Goddard and colleagues (Goddard et al., 1969) and later with the corneal stimulation approach in mice first reported by Sangdee and colleagues (Sangdee et al., 1982), kindling models have been useful to identify novel therapies for the clinical management of epilepsy (Barker-Haliski et al., 2017a). Importantly, as illustrated in Table 2, the amygdala-kindled rat specifically has been found to correctly predict the clinical activity of all ASDs currently available for the treatment of human focal epilepsy, including the identification of levetiracetam (Klitgaard et al., 1998).

Table 2.

Efficacy of clinically approved antiseizure drugs to block seizures in animal models and patients with focal epilepsy^a

Drug	Antiseizure effect in rodents (rats or mice) on:			Antiseizure effect on spontaneous focal seizures in patients with epilepsy
	MES (generalized tonic seizures)	PTZ (generalized clonic seizures)	Amygdala kindled focal and/or secondarily generalized tonic-clonic seizures
Benzodiazepines	+/−	+	+	+ (but tolerance)
Brivaracetam	+	+	+	+
Cannabidiol	+	+	ND	+
Eslicarbazepine acetate	+	NE	+	+
Ethosuximide	NE	+	NE	NE
Felbamate	+	+	+	+
Gabpentin	+	+	+	+
Lacosamide	+	NE	+	+
Lamotrigine	+	NE	+	+
Levetiracetam	NE	NE	+	+
Oxcarbazepine	+	NE	+	+
Perampanel	+	+	+	+
Phenobarbital	+	+	+	+
Phenytoin	+	NE	+	+
Pregabalin	+	+	+	+
Primidone	+	+	+	+
Retigabine (ezogabine)	+	+	+	+
Tiagabine	NE	+	+	+
Topiramate	+	NE	+	+
Valproate	+	+	+	+
Vigabatrin	NE	±	+	+
Zonisamide	+	NE	+	+

^{a –}Updated from Klein et al Epilepsia 2017, with permission, with addition of efficacy data from cannabidiol based on Klein et al, Neurochemical Research 2017 and Patra et al, Epilepsia 2019.

ND – no data available

The predictive validity of the kindled rat makes it a highly valuable model for ASD discovery, especially when one considers that the amygdala-kindled rat identified levetiracetam, arguably one of the most widely employed ASDs to be developed in recent decades. Given this, one might ask why the kindled rat is not a primary screen rather than a secondary screen for the early identification and evaluation of novel ASDs? The answer is primarily one of experimental logistics. Historically, kindling has been most frequently performed in rats implanted with chronic recording/stimulating electrodes within discrete brain regions (e.g. hippocampus; amygdala, etc). Kindled rats are thus extremely labor intensive. A well-powered dose-response evaluation to define the median effective and behaviorally-impairing doses of candidate ASDs in chronically kindled rats requires adequate facilities and resources to surgically implant the stimulating/recording electrode, to kindle the animals, and to house these numerous, often quite mature, rats over a long period of time (Barker-Haliski et al., 2018). Furthermore, unlike the acute seizure models, the time required to conduct a drug study with a chronic model such as the kindled rat far exceeds the time required to conduct a similar study with the acute seizure tests (e.g., MES or scPTZ); thereby severely limiting the number of ASDs that can be screened in a timely manner. As a result, chronic seizure models such as the kindled rat have been historically restricted to late-stage differentiation studies well-after a promising investigational agent has been found to demonstrate robust anticonvulsant efficacy.

PHARMACOLOGICAL VALIDATION OF THE MES, scPTZ, and KINDLING MODELS

As noted above and summarized further in Table 2, the pharmacological profile of the MES and scPTZ tests provides some insight into the potential clinical utility of drugs that are found to be active in one or both of these models. Clearly, the pharmacological profile of the MES test supports its utility as a model of human generalized tonic-clonic seizures (GTC). With the exception of levetiracetam, (approved in Europe and the USA for the treatment of primary GTC seizures in patients with idiopathic generalized epilepsy), several other ASDs have proven clinically useful for treating GTC seizures (see (Shorvon et al., 2018) for review and discussion). All are effective in blocking tonic extension seizures in the rodent MES test (Loscher, 2016). Similarly, ASDs that have found little utility in the treatment of GTC seizures of any origin (i.e., ethosuximide, tiagabine and vigabatrin) are ineffective in the MES test (Table 2). Collectively, these data support the important role that the MES test has played as a first-pass screen for anticonvulsant activity against GTC seizures, irrespective of origin.

Given the pharmacology of the MES test, it has been suggested that it may also be highly predictive of activity against focal seizures. A quick review of the data summarized in Table 2 tends to broadly support this conclusion. For example, 18 of the 21 ASDs approved for the treatment of focal seizures have been found to be active in the MES test at non-impairing doses. The three outliers were levetiracetam, tiagabine, and vigabatrin, all of which are clinically effective in the patient with focal epilepsy. This less than perfect correlation between the preclinical and clinical results demonstrates that the MES test is not a perfectly predictive model of focal seizures. Furthermore, the fact that NMDA-receptor antagonists were found to be highly effective against MES-induced tonic extension seizures but failed to demonstrate any notable efficacy in patients with focal onset seizures is consistent with this conclusion (Loscher and Honack, 1991). It is highly likely that the lack of efficacy for the NMDA antagonists that have been evaluated in patients with epilepsy may be more related to the emergence of serious adverse events (Sveinbjornsdottir et al., 1993), than to a pure lack of anticonvulsant effect. That said, all other ASDs that are active against MES seizures have proven effective for the patient with focal epilepsy, including most recently cannabidiol (Klein et al., 2017; Patra et al., 2018), which has been hypothesized to have a relatively unique mechanism of action in the central nervous system (Stella, 2010). Thus, one can argue that there is significant value in the MES test as an initial “screen” for both GTC and focal seizures but that this test is not the only useful model for bringing new compounds to the clinic. Ultimately, an investigator should not be discouraged by a negative result in the MES test and prematurely “kill” a mechanistically novel molecule that targets a pathway thought to be involved in ictogenesis and seizure spread simply on the basis of inefficacy in the MES test. Instead, she should look to other available models, like the scPTZ test and the fully kindled rodent model, to establish, and/or differentiate, the anticonvulsant profile of an investigational agent with a novel mechanism of action. Indeed, like cannabidiol (Patra et al., 2018), many promising compounds with unique mechanisms of action exhibit mixed efficacy in the MES, scPTZ, and kindling models. Tiagabine and vigabatrin, but not levetiracetam, are also active against scPTZ seizures; all three ASDs are active against kindled focal seizures and human focal epilepsy (Table 2). Thus, the MES test clearly provides valuable information to predict clinical potential, but it certainly is not the only model that should be in use for ASD development purposes. Given this discussion, it reasons that the MES test, plus at least a kindling or scPTZ model would be sufficient to a priori establish whether a novel chemical compound has anticonvulsant potential and serve to demonstrate important proof-of-concept before moving into a more resource intensive syndrome-specific model of chronic seizures.

DO THE MES AND KINDLING MODELS DISPLAY PREDICTIVE VALIDITY FOR DRUG RESISTANT EPILEPSY (DRE)?

It is easy to appreciate how the MES, scPTZ, and kindled rodent models employed in the early phase of ASD discovery would be predictive of subsequent efficacy for easy-to-manage generalized tonic clonic and focal seizures. It is likely that this patient population will attain complete seizure control. However, there remains a significant unmet therapeutic need to address the seizures of the patient with difficult to control epilepsy; i.e., 25-40% of patients with focal epilepsy fail to achieve satisfactory seizure control with the currently available ASDs (Chen et al., 2018; Kwan et al., 2010). For the purpose of this review, the following discussion will be focused on the MES and kindling models.

The incidence of DRE has not significantly changed over the decades in spite of the number of ASDs that have been developed and brought to the market (Chen et al., 2018; Kwan et al., 2010). From this perspective, the argument could be made that the MES and kindling models are not predictive of efficacy for DRE (Loscher, 2002). The ILAE defines DRE as “failure of adequate trials of two tolerated and appropriately chosen and used ASD schedules (whether as monotherapies or in combination) to achieve sustained seizure freedom.” When one considers the results from the pivotal placebo-controlled double-blinded clinical trials that formed the basis of the NDA and EMA application for ASD approval, it is clear that the efficacy results from the MES and kindling models are at least somewhat predictive of efficacy in this patient population (Table 2). Specifically, in order to meet enrollment criteria for most, if not all, of the pivotal trials, patients randomized to the active treatment arm had previously failed at least two, if not more, ASDs; i.e., they all met the clinical definition of DRE. Further these ASDs were ultimately approved for clinical use, attesting to the fact that at least some DRE patients attained seizure control with the investigational agent they were exposed to, further emphasizing the predictive utility of these models to identify compounds for at least some patients with clinically-defined DRE. However, it still remains to be determined what specific measure is important to predict efficacy in this population. For the sake of argument, a positive outcome under these conditions suggests drug efficacy in a highly DRE patient population. Admittedly, the outcomes in terms of lasting “seizure freedom” have largely been “incremental” and not “transformative” for this patient population. Nonetheless, a small percentage of the patient population from these trials do achieve “seizure freedom” with each new ASD approved; a finding that lends support to the continued relevance of the current approach.

Despite the fact that the MES and kindling models do continue to identify effective treatments for patients with epilepsy, regardless of whether their seizures are drug resistant or drug sensitive, there is still a need to identify and characterize more etiologically-relevant models of DRE. To this end, the epilepsy community has seen a significant expansion in the number of rodent models that demonstrate a pharmacoresistant acute or chronic seizure phenotype. We have highlighted a number of recent mouse models of pharmacoresistant seizures and epilepsy that are well-positioned for high- to moderate-throughput ASD development applications in Table 3. For the purposes of drug development, these models are themselves resistant to at least 2 ASDs, consistent with the clinical definition of DRE. As illustrated in Table 3, one of the best-characterized models is the 6 Hz 44 mA 6 Hz psychomotor seizure model in mice (Barton et al., 2001; Leclercq and Kaminski, 2015), and the equivalent 6 Hz model in rats (Metcalf et al., 2017). The 6 Hz model itself provides a very useful model to identify potentially effective ASDs in a high-throughput screening approach, but this model is conducted in neurologically-intact rodents and thus fails to reproduce the behavioral comorbidities associated with epilepsy.

Table 3.

Mouse models of pharmacoresistant chronic seizures in use for ASD development.

Seizure Model	Acute vs. Chronic Seizures	Evoked vs. sponta neous seizure	Induction Protocol	Seizure Endpoints	Pharmaco resistant Model?	Translation al Relevance	Effective ASDs at Non- Motor Impairing Doses	Notes
6 Hz	Acute	Evoked	6 Hz 3 second current of varying intensity (22 mA to 44 mA for male CF-1 mice from Charles River), delivered to anesthetized corneas	Presence of stunned posture, Straub tail, vibrissae twitching and forelimb clonus; duration of these behaviors may also be useful	Yes at 44 mA current intensity a	Model of partial psychomotor seizures	BDZ c; TGB c; RTG c; CBZ c; VPA a,c	Pharmacoresistance is highly strain-dependent
6 Hz Corneal Kindled Mouse b	Chronic	Evoked	6 Hz 3 sec stimulation of various current intensity (strain-dependent) delivered to anesthetized corneas for 2-4 weeks	Forelimb clonus	Yes	Model of pharmacoresistant chronic secondarily generalized focal seizures	CLZ b; LEV b, VPA b; CBZ b (shift in ASD potency was considered the outcome measure for pharmacoresistance)	Limited pharmacology
Lamotrigine-Resistant Corneal Kindled Mouse e	Chronic	Evoked	60 Hz 3 mA 3 second stimulus delivered to anesthetized corneas; mice receive lamotrigine pretreatment during 3-4 week kindling acquisition period	Presence of Racine stage 3-5 seizures and/or mean seizure score	Yes	Model of pharmacoresistant chronic secondarily generalized focal seizures	VPA e; DZP e; LEV e
Intrahippocampal Kainic Acid Mouse Model	Chronic	Spontaneous	Intrahippocampal infusion of KA; spontaneous hippocampal paroxysmal discharges (HPDs) and spontaneous, albeit infrequent, generalized seizures present 2-4 weeks later and persist throughout the life of the animal	Frequency of HPDs during acute treatment period	Yes	Model of acquired mesial temporal lobe epilepsy with hippocampal sclerosis	BDZ d; TGB d; PB d; VGB d	Generalized spontaneous seizures are infrequent and efficacy is measured as effects of agents on HPD frequency.

^{a –}Based on Barton et al, Epilepsy Research 2001

^{b –}Based on Leclercq et al, Epilepsy Research 2014

^{c –}Based on Barker-Haliski et al, Neurochemical Research. 2017

^{d –}Based on Duveau et al, CNS Neuroscience and Therapeutics 2016

^{e –}Based on Koneval et al, Epilepsia 2018

Patients with epilepsy experience chronic seizures, thus prioritizing frontline ASD screening in more etiologically-relevant models of DRE with chronic seizures may provide a greater likelihood of advancing promising ASDs (Table 3). There are several drug-resistant kindling models, including the phenytoin-resistant kindled rat (Ebert et al., 1999; Loscher et al., 1998; Loscher et al., 2000), which are naturally selected for phenytoin sensitivity, as well as the lamotrigine-resistant amygdala kindled rat model (Srivastava and White, 2013), which is kindled in the presence of anticonvulsant doses of the ASD, lamotrigine. More recently, the lamotrigine-resistant kindled rat has been scaled down to accommodate moderate-throughput drug screening applications with the development of the lamotrigine-resistant corneal kindled mouse (Koneval et al., 2018). Interestingly, the pharmacological profile of the lamotrigine-resistant corneal kindled mouse suggests a highly drug resistant preclinical model, which has the capacity to identify agents that may aggravate acute secondarily generalized focal seizure severity. Notably, the severity of seizures of the lamotrigine-resistant corneal kindled mouse are worsened by sodium channel-blocking drugs, such as phenytoin, carbamazepine, and lamotrigine (Koneval et al., 2018). Corneal kindled mice normally demonstrate dose-related reductions in Racine stage seizure severity in the presence of sodium channel-blocking ASDs (Rowley and White, 2010); lamotrigine-resistant kindled mice show dose-related “popcorn-like” seizures, indicative of generally worsened seizures in the presence of sodium channel blocking ASDs (Figure 2, (Koneval et al., 2018)). Interestingly, the seizures of the highly drug-resistant pediatric epilepsy, Dravet’s syndrome, are also known to be worsened by these same sodium channel-blocking ASDs. Dravet’s syndrome arises due to the genetic mutation in the Scn1a gene encoding neuronal sodium channel subunits, particularly within inhibitory interneurons (Bender et al., 2012), thus it reasons that agents that block sodium channels could further aggravate seizures. Seizures in the lamotrigine-resistant corneal kindled mouse are dose-dependently controlled with valproic acid, benzodiazepines (diazepam), and levetiracetam, drawing further similarities to the preferred clinical treatment options for Dravet’s syndrome patients (topiramate and stiripentol have not to-date been tested in this rodent DRE model). Whether the lamotrigine-resistant corneal kindled mouse is able to accurately predict agents that may be effective in DRE patients, and in particular Dravet’s syndrome patients, certainly remains to be defined but offers an intriguing preclinical model to better understand the nature of DRE, while simultaneously providing a useful model for moderate-throughput drug screening applications.

Figure 2.

Mice kindled in the presence of VEH (0.5% MC, b.i.d. for 17 days) or LTG (8.5 mg/kg, b.i.d. for 17 days) demonstrate varied sensitivity to subsequent acute pharmacological intervention with prototype ASDs. A) Escalating doses of LTG are effective in VEH- (n=12-13/dose), but not LTG-kindled mice (n=18-20/dose). B) Escalating doses of LEV are effective in both VEH- (n=8-15/dose) and LTG-kindled mice (n=15/dose). C) Escalating doses of DZP are effective in both VEH- (n=6-8/dose) and LTG-kindled mice (n=8/dose). D) Escalating doses of PB are effective in both VEH- (n=6-9/dose) and LTG-kindled mice (n=7-9/dose). E) Escalating doses of CBZ are effective in VEH- (n=6-9/dose), but not LTG-kindled mice (n=7-10/dose). F) Escalating doses of VPA are effective in VEH- (n=6-8/dose), but not LTG-kindled mice (n=8-9/dose). G) Escalating doses of RTG are effective in VEH- (n=6-7/dose), but not LTG-kindled mice (n=6-8/dose). H) Escalating doses of PHT are ineffective in both VEH- (n=7-14/dose) and LTG-kindled mice (n=14-16/dose). * indicates significantly different from same kindling group baseline MSS, p<0.05. # indicates significantly different from VEH-kindled group of same drug dose, p<0.05. Reproduced from Koneval et al, Epilepsia 2018 with permission (John Wiley and Sons).

Pharmacoresistant kindling models are certainly beneficial for ASD development. Some investigators have adapted the 6 Hz stimulation to corneal kindle mice, resulting in a pharmacoresistant chronic seizure model that is also suitable for moderate-throughput screening in a chronically-seizing substrate (Leclercq et al., 2014). This corneal kindled model certainly exhibits a more drug-resistant profile, but the published pharmacological data is presently limited (Leclercq et al., 2014), albeit this model has the potential to also identify compounds with novel mechanisms of action (Leclercq et al., 2019). However, the lamotrigine-resistant kindled mouse of Koneval and colleagues carries several advantages over the 6 Hz corneal kindled mouse due to the easily identified Racine stage seizures, the clinically-relevant ASD monotherapy paradigm to induce pharmacoresistance during the epileptogenesis process, and the ability to detect seizure worsening with specific ASDs (e.g. carbamazepine, phenytoin, discussed above) and tool compounds (e.g. fluoxetine (Koneval et al., 2018). Nonetheless, use of a chronically-seizing model with a drug resistant seizure phenotype may ultimately prove beneficial to identify mechanistically-novel therapies for the DRE patient population. That these models are developed on the backbone of kindling further attests to the continued relevance and utility of kindling models, in general, for ASD development. Whether these DRE models are able to identify novel compounds remains to be defined but certainly carries the potential to more realistically approximate human DRE than the MES and scPTZ models.

Additional approaches to develop new models that may exhibit a pharmacoresistant profile also carry significant potential to one day benefit the patient with DRE. For example, the Theiler’s murine encephalopathy virus (TMEV) model of encephalitis is one of the few rodent models available to study the natural progression of infection induced-acute seizures and chronic epilepsy (Libbey et al., 2008; Stewart et al., 2010). The acute seizures of the TMEV model indeed demonstrate a phenotype that is highly resistant to ASD administration (Barker-Haliski et al., 2015; Barker-Haliski et al., 2016a; Barker-Haliski et al., 2017b), but this model also carries significant potential to identify mechanistically novel therapies to modify the course of epilepsy. For example, disease modification in the TMEV model appears to be highly sensitive to agents with anti-inflammatory mechanisms (Barker-Haliski et al., 2016a; Kaufer et al., 2018; Waltl et al., 2018). Whether any one of the agents that demonstrates promise in the TMEV model will also demonstrate efficacy in the MES, scPTZ, or kindling models and in a relevant clinical population certainly remains to be demonstrated.

Lastly, there is significant use of the post-status epilepticus (SE) rat model of temporal lobe epilepsy in the field due to the high translational relevance of a rodent model with spontaneous recurrent seizures to inform on the processes underlying ictogenesis and epileptogenesis. However, these models are generally not suitable for moderate- to high-throughput drug screening and so a discussion of these post-SE models is truly beyond the scope of this review. Nonetheless, each of these pharmacoresistant seizure models has proven useful for hypothesis testing and drug-differentiation, but to date, no drug has advanced to the clinic solely on the basis of activity in any one of these new pharmacoresistant models. Thus, at present, these pharmacoresistant models all lack any evidence of clinical predictive validity.

Medication adherence is one additional aspect of DRE that remains under-studied but may provide further value to supplement anticonvulsant efficacy identified in the MES, scPTZ, or kindling models. Specifically, modeling medication non-adherence in a preclinical model system may provide valuable insight into whether dose-escalation or just correction of adherence is sufficient to improve seizure control. Recognizing the potential link of nonadherence to a diagnosis of DRE, Hill and colleagues described a novel approach for evaluating the impact of poor adherence on seizure control in an animal model of post-status epilepticus-induced epilepsy (Hill et al., 2019). Whether this approach will be able to identify informative strategies to better treat DRE remains to be determined, but the important caveat is to understand that the doses used to test this hypothesis have all been defined in the MES test (Bialer et al., 2004). In addition to poor adherence, other factors can contribute to a diagnosis of DRE and include inadequate dosing of ASDs and lifestyle factors that lower seizure threshold and increase seizure frequency; e.g., drug and alcohol abuse, stress and sleep deprivation. No animal model or testing paradigm effectively considers these factors during the drug evaluation process, whether it is performed in a validated or novel animal model of acute seizures or epilepsy.

OTHER CONSIDERATIONS

In addition to their reasonable predictive validity for GTC, myoclonic, and spike-wave seizures, the MES and scPTZ tests provide some insight into blood-brain-barrier penetration of a given chemical entity following systemic administration. Further, both models are well suited for initial ASD screening because both models are mechanistically independent; i.e., neither model assumes that efficacy is dependent on a drug’s molecular mechanism of action. Importantly, both model systems display clear and definable seizure endpoints and require minimal technical expertise; two properties that make them well suited to screen large libraries of structurally-diverse chemical entities.

Ultimately, it is unlikely, however, that a single model will be the defining model to bring a new investigational agent to the clinic. The importance of employing multiple models in the evaluation and differentiation of an investigational agent cannot be overstated. For example, levetiracetam is inactive in the traditional MES and scPTZ tests, yet it demonstrates excellent efficacy in kindled rodents (Klitgaard, 2001; Klitgaard et al., 1998). Likewise, efficacy of tiagabine and vigabatrin against human focal seizures was not predicted by the MES test, but by the kindled rat model (Loscher, 2016; Rogawski and Porter, 1990; Suzdak and Jansen, 1995). Furthermore, exacerbation of spike-wave seizures by tiagabine, vigabatrin, and phenobarbital was not predicted by the scPTZ test, but rather by models with clear spike-wave seizure (i.e., GHB, GAERS, and the lh/lh mouse; (Hosford and Wang, 1997)). These examples demonstrate the importance of evaluating each investigational ASD in a variety of seizure and epilepsy models. Only then will it be possible to gain a full appreciation of the overall spectrum of activity for a given investigational drug.

The reverse can also be true in that a single compound may indeed demonstrate efficacy in a range of preclinical models, thus leaving a preclinical development team to question which model is, in fact, the most appropriate one to use when considering how to advance a new therapy to the clinic. As an example, cannabidiol (CBD), was approved for the treatment of seizures associated with Dravet’s Syndrome and Lennox-Gastaut Syndrome in 2018. While there is clear evidence that CBD is highly effective in mice with the Scn1a^+/− genotype associated with Dravet’s syndrome (Kaplan et al., 2017), the preclinical studies in this relevant mouse model are highly resource- and labor-intensive. Scn1a+/− mice exhibit high mortality early in life (Kaplan et al., 2017; Oakley et al., 2009) and are known to breed poorly in preclinical animal colonies such that obtaining sufficient numbers of animals for large scale drug development applications is challenging (R.E. Westenbroek, personal communication). Using a mouse model with a Dravet’s syndrome-associated genotype is certainly appealing in theory, but in practice this approach becomes infinitely more challenging when the need to identify promising lead compounds or obtain sufficient funds to support drug development are considered. Resource limitations may ultimately restrict the number of investigational compounds that can be initially profiled in syndrome-specific models. Until a syndrome-specific model thus becomes amenable to high-throughput testing, they will fail to be useful to early ASD discovery efforts; but will instead remain largely relegated to “hypothesis testing.” Further, CBD is effective in the MES, scPTZ, and kindling models (Tables 2) (Klein et al., 2017; Patra et al., 2018). Interestingly, Patra and colleagues have even demonstrated that CBD may be disease-modifying in an adult rat population with spontaneous recurrent seizures (Patra et al., 2018), and altogether causes one to question the value provided by further screening of any given candidate drug in a syndrome-specific genetically-engineered model with the Scn1a+/− genotype associated with Dravet’s syndrome. While it is true that these syndrome-specific models provide the important proof-of-concept assurance to in-theory support clinical study in a relevant patient population, the data to-date with available ASDs would suggest that the MES, scPTZ, and kindling models are instead all that is necessary to support a first-in-man clinical evaluation in the relevant patient demographic of interest (e.g. Dravet’s syndrome patients).

OPPORTUNITIES TO IMPROVE PRECLINICAL TESTING

The MES, scPTZ, and kindling models certainly will continue to offer significant value to investigators interested in developing novel pharmacotherapies for the symptomatic treatment of seizures in epilepsy. Integration of more chronic seizure models that demonstrate a highly drug-resistant profile will also afford more opportunities to select only agents that may be useful in this patient demographic. However, there still remains a significant opportunity to further improve preclinical testing practices to better address patient needs. While the animal models of epilepsy have indeed been incredibly beneficial to advance nearly two dozen ASDs for the patient with epilepsy worldwide, rodents are not humans. Recent efforts to develop human-derived induced pluripotent stem cells and/or tissues from resected brain tissues from patients with epilepsy will likely offer new avenues to better understand the pathways and processes underlying ictogenesis and epileptogenesis in man (Du and Parent, 2015; Liu et al., 2013). Whether these human cell-derived organoids and tissues will be integral to ASD development awaits further evaluation and validation, but our ability to-date apply these tissues to basic epilepsy research practices will at the very least improve our understanding of the relevant targets underlying epilepsy to improve target-based drug development practices. As of this writing, human-derived organoids are not yet at a sufficient level to represent the circuitry of a patient with epilepsy to adequately support ASD development. Further, acquired epilepsies (e.g. post-TBI) are likely going to be much more challenging to model with iPSCs/human organoids than the genetic epilepsies, but these tissues will provide significant value to better appreciate how epilepsy arises in discrete patient populations (e.g. genetic epilepsies).

VI. SUMMARY

Given that seizures are the ultimate clinical expression of epilepsy, prevention of seizures as an endpoint for any initial ASD discovery program is appropriate. As such, the MES, scPTZ, and kindling models continue to offer strong value to investigators interested in the initial screening of a structural series of candidate ASDs in order to identify and optimize a lead compound for further development. There is certainly a myriad of additional models that can be utilized to further differentiate the anticonvulsant profile of a given drug from ASD standards-of-care, and we have only highlighted a few of these in the present manuscript (drug-resistant kindled rodent models, post-SE models of spontaneous recurrent seizures, etiologically-relevant models of pediatric encephalopathies, and models of viral encephalitis). More information on the overall anticonvulsant profile of an investigational ASD is always good to have, but in order to address whether any given “lead compound” differentiates itself in terms of better tolerability, safety, and efficacy against human DRE will ultimately require testing the lead drug in the most appropriate model system; i.e., the patient with epilepsy.