In silico screening for clinical efficacy of antiseizure medications: Not all central nervous system drugs are alike

Key Points.

• We found differences between physicochemical properties of antiseizure medications with proven clinical efficacy and those of other CNS drugs.

• Sirolimus and everolimus are exceptionally large molecules, prompting further investigation of their mechanism of action against seizures.

• Drug repurposing for treating epilepsy should consider the currently known chemical structure‐derived limitations for antiseizure medications.

• Drugs of proven efficacy against seizures whose etiologies are characterized by overt BBB disruption would not necessarily be effective in other seizure types.

1. INTRODUCTION

A recent review by Odi et al. provided important information on the physicochemical and biopharmaceutical properties of antiseizure medications (ASMs). ¹ The authors included marketed ASMs, ASMs in development, and those that reached clinical trials but did not make it to the market. Based on that analysis, a set of properties was suggested as a screening tool to be used in the early developmental stages of small molecule ASMs intended for oral dosing. These properties influence the permeability of drugs across membranes, which is critical for their absorption, elimination, and distribution across the blood–brain barrier (BBB), and their interactions with drug‐metabolizing enzymes. The analysis referred to parameters that may be used individually or as components of desirability scoring systems, namely the drug's molecular weight (MW), logP (the logarithm of a drug n‐octanol:water partition coefficient), logD (logP at pH = 7.4), topological polar surface area (TPSA; described in more detail below), and pKa (the negative log of the acid dissociation constant). Among the scoring systems were the “Lipinski rule of five” and the central nervous system (CNS) multiparameter optimization (MPO) desirability tool. The former relates to passive permeation across membranes and predicts oral absorption based on logP, MW, and the number of hydrogen bond donors and hydrogen bond acceptors. ² , ³ The CNS MPO tool, used for designing CNS drugs, is a function that maps the molecule's logP, logD, MW, TPSA, number of hydrogen bond donors, and pKa to desirability scores (Table 1). ⁴ , ⁵

TABLE 1.

Examples of scoring systems that have been utilized in CNS drug development and desired values according to each score.

Score	Lipinski rule of five 2 , 3 desirable value	CNS multiparameter optimization desirability tool 4 , 5
		Desirable value	Less desirable values
MW, Da	<500	≤360	360–500
TPSA, Å2	‐	40–90	20–40, 90–120
LogP	<5	≤3	3–5
LogD	‐	≤2	2–4
Number of hydrogen bond donors	<5	≤1	0–4
Number of hydrogen bond acceptors	<10	‐	‐
pKa	‐	≤8	8–10
Scoring methodology		Score = 1 for meeting each criterion	Score = 0 for each value beyond the range; a desirability score across ranges is determined by a linear function; collective scores: 0–6
Desirable score	≥2 properties	≥5

Abbreviations: CNS, central nervous system; logD, the logP at serum's pH (7.4), indicating lipophilicity at typical blood–brain barrier permeation pH; logP, the logarithm of the drug's un‐ionized species n‐octanol:water partition coefficient (higher logP values indicate more lipophilic drugs); MW, the drug's molecular weight (presented in daltons); pKa, the negative log of the drug's acid dissociation constant; TPSA, the drug's topological polar surface area, i.e., area of molecular surface arising from polar atoms, which affects the molecule's ability to cross membranes by passive diffusion (presented in square angstroms).

Here, we argue that scores that have been developed for CNS drugs as a broader group are not optimal for identification of clinically‐efficacious small molecule ASM candidates. Moreover, we suggest that these comparisons might also provide some insights into drug disposition within the epileptogenic brain. These notions are based on our evaluation of the characteristics of ASMs approved by the US Food and Drug Administration (FDA) and/or the European Medicines Agency for at least one epilepsy indication (other than tuberous sclerosis, as described below). We compared those drugs to ASM candidates that reached clinical trials (≥phase 2) but did not make it to the market due to efficacy considerations, and to other CNS drugs. ⁶ We chose MW and TPSA, which are key determinants of small molecule distribution across the BBB, ⁷ , ⁸ as examples of parameters that require refinement. This does not exclude the need for further assessment of other parameters. In addition, the presented analysis is based solely on clinical efficacy data, and not on results from preclinical seizure models.

2. MARKETED ASMS ARE SMALLER THAN OTHER CNS DRUGS

One of the structural prerequisites of successful ASMs that was identified by Odi et al. ¹ is MW < 375 Da. This value is close to the upper limit of the CNS MPO's desirable MW (≤360 Da). ⁵ The CNS MPO tool was based on an analysis of 119 marketed CNS drugs and 129 Pfizer CNS candidates. ⁴ The drug set used to build the tool included 16 marketed ASMs but also compounds such as aripiprazole and buprenorphine whose MWs are considerably larger than 375 Da. Accordingly, the upper MW cutoff value of this tool for CNS drugs is 500 Da, in agreement with the upper boundary of Lipinski's rule of five (Table 1). ² , ⁹

Since the publication of the analysis by Odi et al., ganaxolone (333 Da) was approved by the FDA for the treatment of seizures associated with cyclin‐dependent kinase‐like 5 deficiency disorder. ¹⁰ In contrast, padsevonil (433 Da) showed no advantage over placebo in its primary end points, despite its efficacy in experimental models of drug‐resistant epilepsy. ¹¹ This adds to the failure of the prodrug VX‐765 (MW of the active metabolite = 481 Da) to perform better than placebo in treatment‐resistant focal epilepsy. ¹² Bumetanide (364 Da), which was suggested to act through both cerebral and extracerebral targets, ¹⁰ did not improve seizure control in newborns with hypoxic ischemic encephalopathy ¹³ and showed no effectiveness in children with autism spectrum disorder, leading to early termination of the trials. ¹⁰ Although the lack of VX‐765 and padsevonil efficacy may not relate solely to the MW of these molecules, it is a missed opportunity to demonstrate the success of compounds larger than the currently used ASMs. It further calls for between‐species comparisons of ASM permeability into epileptogenic brain tissue.

Besides being a predictor of ASM marketability, the 375‐Da cutoff value distinguishes clinically‐used ASMs from other classes of CNS drugs, particularly psycholeptics (Figure 1, Table 2). The MW of many CNS drugs is higher than 375 yet they are below the 500‐Da upper boundary (Figure 1). An outlier is bromocriptine (655 Da), which is both a substrate of uptake transporters at the BBB ¹⁵ and a highly potent compound that can exert pharmacological effect at nanomolar concentrations. ¹⁶

FIGURE 1.

Physicochemical characteristics of new antiseizure medications (ASMs; n = 32) and nonmarketed ASMs (n = 3) in comparison with other central nervous system (CNS) drugs (n = 176). The drugs categorized as nonmarketed ASMs were (1) classified in the original publication by Odi et al. ¹ as “reached clinical trials but did not make it to the market” or “new antiseizure drugs in development” and (2) did not improve seizure control in clinical trials. Accordingly, lack of efficacy was a key consideration in the decision to stop their development as ASMs. This does not necessarily exclude their potential efficacy against other seizure types for which they have not been tested yet. The drugs in this category include VX‐765, padsevonil, and bumetanide. XV‐765, a prodrug, is represented by its active metabolite. (A) Molecular weight (MW). (B) Topological polar surface area (TPSA). Data were retrieved as described for Table 2. The color scheme refers to the desirable and the less desirable ranges defined by the CNS multiparameter optimization score. Horizontal lines denote median values. Marketed ASMs and other CNS drugs were compared by the Mann–Whitney test. **p < .01. Sensitivity analyses in which dual‐use benzodiazepines (clobazam, lorazepam, diazepam, midazolam) were categorized as both ASMs and psycholeptics yielded similar results (not shown)

TABLE 2.

Physicochemical properties of antiseizure medications in comparison to those of other CNS drugs

Property	n	Molecular weight, Da	TPSA, Å2
Marketed ASM	32	251.3 (129.2–375.6)	63.3 (12.0–124.0)
Nonmarketed ASM	3	432.8 (364.4–480.9)	88.0 (127.0–159.0)
Anesthetics [N01]	12	253.3 (104.0–416.5)	58.0 (20.2–90.3)
Analgesics [N02]	13	301.3 (221.3–467.6)	49.8 (23.5–73.2)
Antiparkinsonian drugs [N04]	19	304.0 (151.2–654.6)	32.1 (3.2–118.0)
Psycholeptics [N05]	67	328.5 (7.0–492.7)	48.5 (0–110.0)
Psychoanaleptics [N06]	48	264.3 (133.1–470.0)	37.5 (3.2–102.0)
Other nervous system drugs [N07]	17	228.7 (94.1–467.6)	38.8 (12.5–91.8)

Note: See Figure 1 legend for the categorization of nonmarketed ASMs. Psycholeptics: antipsychotics, anxiolytics, hypnotics, and sedatives. Psychoanaleptics: antidepressants, psychostimulants, agents used for attention‐deficit/hyperactivity disorder and nootropics, and antidementia drugs. The marketed and nonmarketed ASM drug lists are based on Odi et al. ¹ and includes bumetanide (see text for further details). ¹⁰ The other drug lists are based on the World Health Organization's Collaborating Center for Drug Statistics Methodology, and the numbers in brackets refer to drug classes as defined in this database. ⁶ Drug combinations, gases, local anesthetics, drugs not approved for use by the US Food and Drug Administration (except for nonmarketed ASMs), and those that do not necessarily have to cross the blood–brain barrier to exert their effects (e.g., antiemetics and antimigraine drugs) were excluded. A full drug list is available as Table S1. Shown are median values (minimum–maximum). Molecular weights and TPSA values were retrieved from PubChem's PUG API ¹⁴ using an ad hoc automated script.

The observed difference between ASMs and other CNS drugs might be associated with distinct structural requirements for receptor binding, but is more likely to provide hints into drug disposition within epileptogenic brain tissue, given the multitude of cerebral ASM targets. In the absence of direct comparisons of BBB permeability across neurological and psychiatric disorders, the reasons for the higher MW cutoff value for CNS drugs other than ASMs have yet to be clarified. Overexpression of the BBB efflux transporter P‐glycoprotein (P‐gp), ¹⁷ for which drugs with MW > 400 Da are good substrates, ¹⁸ could have provided an elegant mechanistic explanation for the low MW cutoff in epilepsy. However, some other brain disorders are also characterized by P‐gp upregulation. ¹⁹ , ²⁰ Moreover, recent proteomic analyses suggest the P‐gp is globally underexpressed and not overexpressed in the epileptogenic human brain. ²¹ , ²² Likewise, key tight junction proteins are downregulated in the epileptogenic brain, ²² , ²³ supporting looser rather than tighter barrier than in the healthy brain. Alternatively, seizures might theoretically alter the porosity of endothelial cell membranes, thereby slowing down or preventing drug “jumping” through transitory holes in the membrane. ⁷ , ²⁴

The MW limitation for ASMs appears to conflict with the parenchymal distribution of much larger molecules, for example, albumin (>2000 Da) within epileptogenic tissue. ¹⁷ , ²³ , ²⁵ , ²⁶ The paradox may be explained by heterogeneity of the vasculature in epilepsy ²¹ ; some vessels, for example, those newly formed through vascular remodeling and aberrant angiogenesis ²⁷ , ²⁸ could be “leaky,” whereas others are restrictive. That is, both albumin and large molecule ASMs may extravasate permeable capillaries. However, neither would uniformly distribute across diseased brain tissue, such that ASM concentrations near neurons distant from the permeable vessels could be subtherapeutic.

3. MARKETED ASMS ARE MORE POLAR THAN OTHER CNS DRUGS

ASMs additionally differ from other CNS drugs by generally having more surface area over polar atoms, as indicated by the higher TPSA values of the marketed ASM group relative to the nonmarketed ASMs and other CNS drugs (Figure 1B, Table 2). The TPSA measures the polarity of the compound and is an indicator of the molecule's ability to cross membranes by passive diffusion. TPSA larger than 140 Ų usually restricts a compound from entering the brain, ¹ , ²⁹ and the optimal range on the CNS MPO score is 40–90 Ų. ⁵ Yet simply enhancing the brain penetration of candidate molecules by decreasing their TPSA (or increasing the logP) may be associated with increased risk of in vivo toxicity, ⁵ , ³⁰ , ³¹ particularly for basic molecules. ³¹ In addition, higher lipophilicity leads to unnecessary nonspecific partitioning into brain tissue. ³² , ³³ In two recent indirect comparisons of ASM efficacy in patients with focal seizures, ³⁴ , ³⁵ the relative responder rate was highest for molecules with TPSA value > 90 (Figure 2). The exact nature of the nonspecific drug disposition sites within epileptogenic brain tissue requires further investigation.

FIGURE 2.

Correlation between the topological polar surface area (TSPA) of antiseizure medications and the odds ratio versus placebo for reduction of seizure frequency in patients with focal seizures. Values are based on data from Privitera et al. (A,B) ³⁵ and Lattanzi et al. (C). ³⁴ (A,B) Correlation between the TPSA of antiseizure medications and ≥50% responder rate across all studied doses (A) or the ≥50% responder rate at maximal dose in each study (B). ³⁵ (C) Correlation between the TPSA and the surface under the cumulative ranking curve (SUCRA). Similar results were obtained for an analysis that involved the odds ratios of seizure response versus placebo. ³⁴ Shown are linear regression with 95% confidence interval, and the Spearman correlation coefficient (r). Oxcarbazepine was assigned the TPSA of its active metabolite eslicarbazepine. B, brivaracetam; C, cenobamate; E, eslicarbazepine; LC, lacosamide; LE, levetiracetam; LT, lamotrigine, P, perampanel; T, topiramate

4. MAMMALIAN TARGET OF RAPAMYCIN INHIBITORS ARE OUTLIERS

The MW and the TPSA values of the mTOR inhibitors everolimus and sirolimus, used in the treatment of tuberous sclerosis, are well beyond those of all other CNS drugs: 958.2 Da and 914.2 Da (as compared to a maximum of 654.6 Da for CNS drugs), and 205 Ų and 195 Ų (vs. 124 Ų), respectively. In mice, the distribution of everolimus and sirolimus across the BBB is poor (brain:plasma ratio = .0057 for sirolimus and .016 for everolimus). ³⁶ If this is also the case in humans, the antiseizure activity of those ASMs in tuberous sclerosis might be explained by extracerebral sites of action, permeable BBB, or both. Protective effects of sirolimus on the BBB were demonstrated in experimental models of temporal lobe epilepsy, suggesting the BBB as a target. ³⁷ , ³⁸ The activity of these drugs in tuberous sclerosis might also involve targets beyond the BBB; imaging studies with gadolinium‐based contrast agents showed enhancement of white matter lesions, subependymal nodules, subependymal giant cell astrocytomas, and some tubers, particularly cerebellar tubers. ³⁹ , ⁴⁰ , ⁴¹ Because gadolinium complexes are large and polar molecules (gadolinium–diethylenetriamine pentaacetic acid, MW = 1031 Da, TPSA = 315 Ų; gadopentetate dimeglumine: MW = 548 Da, TPSA = 205 Ų), it is plausible that their distribution might correlate with those of sirolimus and everolimus and might even serve as a biomarker for such treatment. Similar considerations apply to other seizure etiologies associated with considerable BBB disruption, such as neurocysticercosis. ⁴² , ⁴³

Sirolimus and everolimus are also being evaluated for epilepsies with etiologies other than tuberous sclerosis, some of which are without overt BBB disruption (e.g., focal cortical dysplasia ⁴⁴ – NCT03198949, NCT02451696 ⁴⁵ ). The outcomes of these trials can add to our understanding of how mTOR inhibitors exert their activity in patients with epilepsy. Currently, the majority of efficacy data for such epilepsies are based on case series and case reports (Table 3).

TABLE 3.

Everolimus and sirolimus efficacy in epilepsies with etiologies other than tuberous sclerosis

Drug	Study type	Outcomes
Sirolimus	Single‐arm, open‐label, multicenter clinical trial	Nonsignificant reduction in seizure frequency in 16 pediatric and adult patients with FCD type II (focal seizure frequency reduced by 25% in all patients during the maintenance therapy period, response rate 33%; focal seizure frequency in the external control group reduced by .5%) 46
Sirolimus	Case report	Temporary but complete seizure control for up to 3.5 months in a neonate with NPRL3 deletion 47
Sirolimus	Case report	In a newborn with a loss‐of‐function NPRL3 gene variant, no change in seizure frequency over the 17‐day treatment 48
Everolimus	Observational open‐label study	Of 4 individuals with drug‐resistant epilepsy and GATOR1 mutations (3 DEPDC5; 1 NPLR3), 2 have experienced a >50% improvement in seizure frequency 49
Sirolimus	Case report	In an infant with hemimegalencephaly, refractory seizures, and a somatic mosaic variant in MTOR, 50% reduction in seizures after 1 week of treatment 50
Everolimus	Case report	In a girl with a mosaic MTOR mutation and intractable seizures, no beneficial treatment effect despite rapamycin‐sensitivity of the activating MTOR mutation of the patient 51
Sirolimus	Case series	Reduction in seizure frequency in 5 patients with PMSE (caused by LYK5/STRADα deletion) 52
Sirolimus	Case report	Seizure freedom for 23 months in a preterm newborn with Sturge–Weber syndrome; effect attributed, at least in part, to antiangiogenic activity 53
Sirolimus	Case series	Seizures controlled in all 6 patients with Sturge–Weber syndrome who were refractory to ASMs; longest follow‐up = 26 months 54

Note: The data are based on PubMed and Embase searches combining the terms "everolimus" or "sirolimus" with "epilepsy", "seizures", or "focal cortical dysplasia".

5. CONCLUSIONS

Current screening tools lack specificity and selectivity to effectively prioritize small molecules for development as ASMs. For instance, the CNS MPO score and TPSA would likely exclude cannabidiol and fenfluramine from further development as ASMs and place topiramate at a lower priority (Figure 3). Accordingly, these parameters are not sensitive enough for ASM candidates. On the other hand, the desirable values on commonly used scores for CNS drugs such as Lipinski's rule of five are of low specificity for ASMs and encourage the testing of many compounds that are less likely to become systemically administered ASMs.

FIGURE 3.

Antiseizure medication (ASM) properties with regard to the desirable central nervous system (CNS) multiparameter optimization (MPO) scoring system. ASMs were ranked by their respective characteristics and color‐coded according to their desirability on each scale from light turquoise (desirable) to crimson (nondesirable). For the total CNS MPO score and molecular weight (MW), the desirable values are >5 and ≤360 Da, respectively. The desirable topological polar surface area (TPSA) values range 40–90 Å. Values of 20–40 Å and 90–120 Å are less desirable. Values of <20 Å or >120 Å are not desirable. *Oxcarbazepine is rapidly converted to licarbazepine.

Exemptions to the aforementioned scores or a refined scoring system could include (1) candidates for seizure prevention in tuberous sclerosis, neurocysticercosis, and other seizure etiologies characterized by overt contrast enhancement; (2) highly potent compounds that exert pharmacological effects even at nanomolar or submicromolar concentrations near their targets, or compounds with little systemic toxicity even at high doses; (3) drugs, prodrugs, and drug formulations that can cross the BBB by using specialized carriers and receptors; and (4) treatments that target extracerebral sites, including the BBB itself and inflammation of the neurovascular unit. Examples include interleukin‐1 and interleukin‐6 antagonists (anakinra and tocilizumab, respectively) for treating febrile infection‐related epilepsy syndrome ⁵⁵ and microbiome‐based therapies, taking advantage of microbiotal effects on the brain through secreted metabolites and other communication routes. ⁵⁶ , ⁵⁷ , ⁵⁸ In addition, drugs might better distribute into the brain during epileptogenesis if the epileptogenic insult involves BBB disruption, as has been reported for traumatic brain injury. ⁵⁹ , ⁶⁰ , ⁶¹ Antiepileptogenic drugs might also treat processes that do not necessarily require their distribution across the BBB (e.g., inflammation). Accordingly, a wider assortment of drugs might be repurposed as antiepileptogenic versus antiseizure medications, but this would require thorough consideration of the epileptogenic etiology.

When interpreting CNS MPO scores for potential drug repurposing in the treatment of epilepsy, researchers should consider that the model was developed for the screening of novel compounds. The score may penalize or reward drugs based on an estimation of properties, such as safety and absorption, distribution, metabolism, and excretion (ADME) profiles, that are already well established for those candidates. For example, the TPSA desirability score hump function used in the CNS MPO model assigns low desirability scores to molecules with low TPSA values such as fenfluramine, due to correlations with increased toxicity, not because of decreased CNS activity. Toxicity and ADME screening are desirable in assessment of novel drug candidates but may exclude repurposing candidates for which the aforementioned screening is redundant. In addition, they may promote others with properties favorable for successful safety and ADME profiles but of reduced relevance in prediction of CNS activity. Importantly, even refinement of efficacy‐related parameters alone can minimize animal testing and increase the success of ASM candidates at preclinical stages of development or during repurposing.

CONFLICT OF INTEREST

S.E has served as a paid consultant for BioPass, Israel. The other authors have no conflicts of interest.

Abbreviations: ASM, antiseizure medication; DEPDC5, pleckstrin domain‐containing protein 5; FCD, focal cortical dysplasia; GATOR1, GAP activity toward RAGs 1 complex; LYK5, protein kinase LYK5; MTOR, mechanistic target of rapamycin; NPRL, nitrogen permease regulator‐like; PMSE, polyhydramnios, megalencephaly, and symptomatic epilepsy syndrome; STRADα, STE20‐related kinase adaptor α.

Supporting information

Table S1.

Hamed R, Eyal AD, Berman E, Eyal S. In silico screening for clinical efficacy of antiseizure medications: Not all central nervous system drugs are alike. Epilepsia. 2023;64:311–319. 10.1111/epi.17479