Novel predictive approaches for drug-induced convulsions in non-human primates using machine learning and heart rate variability analysis

Abstract

Drug-induced convulsions are a major challenge to drug development because of the lack of reliable biomarkers. Using machine learning, our previous research indicated the potential use of an index derived from heart rate variability (HRV) analysis in non-human primates as a biomarker for convulsions induced by GABA_A receptor antagonists. The present study aimed to explore the application of this methodology to other convulsants and evaluate its specificity by testing non-convulsants that affect the autonomic nervous system. Telemetry-implanted males were administered various convulsants (4-aminopyridine, bupropion, kainic acid, and ranolazine) at different doses. Electrocardiogram data gathered during the pre-dose period were employed as training data, and the convulsive potential was evaluated using HRV and multivariate statistical process control. Our findings show that the Q-statistic-derived convulsive index for 4-aminopyridine increased at doses lower than that of the convulsive dose. Increases were also observed for kainic acid and ranolazine at convulsive doses, whereas bupropion did not change the index up to the highest dose (1/3 of the convulsive dose). When the same analysis was applied to non-convulsants (atropine, atenolol, and clonidine), an increase in the index was noted. Thus, the index elevation appeared to correlate with or even predict alterations in autonomic nerve activity indices, implying that this method might be regarded as a sensitive index to fluctuations within the autonomic nervous system. Despite potential false positives, this methodology offers valuable insights into predicting drug-induced convulsions when the pharmacological profile is used to carefully choose a compound.

INTRODUCTION

Adverse effects in the central nervous system (CNS) consistently present significant impediments to the advancement of pharmaceutical research and drug development because they can potentially be irreversible and/or life-threatening in humans. In particular, convulsions are the most frequently encountered CNS toxicity in non-clinical studies (Authier et al., 2016). The unpredictable nature of these incidents, coupled with their potential to induce sudden mortality, is among the most severe toxicological phenomena requiring thorough evaluation in preclinical studies. Several methods have been investigated to determine the convulsive activity of drug candidates; for instance, in vitro techniques, such as the use of induced pluripotent stem cells (iPSCs), have been utilized for preliminary screening (Grainger et al., 2018; Odawara et al., 2018; Tukker and Westerink, 2021). In addition, in vivo procedures encompassing the measurement of liquid markers and electrophysiological and imaging modalities have been performed for animal seizure potential assessments. However, their invasive nature and intricate implementation often hinder their practicality (Engel and Pitkänen, 2020; Kuga et al., 2023). From the perspective of animal welfare and shortening drug development time, less intricate evaluation methods are required to reasonably integrate testing into the non-clinical development stage of drugs.

Fluctuations in the autonomic nervous system have been identified as pertinent to a range of CNS disorders in humans and animals. Heart rate variability (HRV) analysis is a commonly employed method to evaluate autonomic nervous system function in humans and has demonstrated its utility in elucidating drug effects on the autonomic nervous system in non-clinical studies, primarily via telemetry-derived electrocardiograms (Champeroux et al., 2013). Autonomic assessment of the nervous system is useful in evaluating epileptic seizures (Baumgartner et al., 2019). Recently, a novel method was developed for predicting human epileptic seizures using HRV analysis (Fujiwara et al., 2016). HRV analysis has also exhibited potential for predicting convulsions in non-human primates (NHPs) caused by gamma-aminobutyric acid A (GABA_A) antagonists, pentylenetetrazol (PTZ), and picrotoxin (Nagata et al., 2021). With this method, it is feasible to evaluate convulsions using existing data from cardiovascular evaluation studies on large animals employing telemetry systems, which is a routine procedure in drug development. Continuous electrocardiographic (ECG) data can be obtained in an unanesthetized, unrestrained state and subsequently analyzed. This technique avoids any additional involvement of animals and has no impact on the study duration. Moreover, its capability to retrospectively analyze historical data renders it particularly valuable for drug candidate assessments. However, it is unclear whether this method can also be applied to convulsants other than GABA-related drugs.

In the present study, we sought to validate this method in anticipating drug-induced convulsions. We administered four convulsants of different pharmacological classes (excluding GABA_A antagonists) to NHPs. The convulsants included 4-aminopyridine (4-AP, a potassium channel blocker), bupropion (a dual norepinephrine and dopamine reuptake inhibitor), kainic acid (a kainate type glutamate receptor agonist), and ranolazine (a calcium uptake inhibitor via the sodium/calcium channel). In addition, PTZ data from a previous study (Nagata et al., 2021) were analyzed as positive reference data in this study because some conditions of the analysis in the present study were different from those in the previous study. Furthermore, given that this methodology primarily uses the parameters of the autonomic nervous system, there is a concern that it might also detect autonomic nervous system changes unrelated to convulsive activity. Accordingly, we investigated nonconvulsive autonomic agents to ascertain the specificity of this technique.

MATERIALS AND METHODS

Animals

Male NHPs (cynomolgus monkeys, 4−7 years old, 4–7 kg) were intraperitoneally implanted with a transmitter (TL11M2-D70-PCT, TL11M3-D70-PCTP, or L11; Data Sciences International, Inc., Minnesota, USA). The pressure-sensor catheter of the transmitter was inserted into the femoral artery and placed in the abdominal aorta. A solid-tip-type negative ECG electrode from TL11M2-D70-PCT was inserted into the right jugular vein, and the tip was positioned in the superior vena cava. The positive ECG electrode was fixed to the abdominal side of the diaphragm, near the apex of the heart. Thoracotomy was performed and the negative ECG electrode from the TL11M3-D70-PCTP or L11 transmitter was fixed to the pericardium near the right atrium. A positive ECG electrode was fixed to the pericardium near the apex of the heart. In some animals, a left ventricular catheter was inserted into the apex of the heart. However, the left ventricular pressure was not evaluated in this study. The animals were housed individually in metal cages set on racks in an animal room, with manipulable toys removed from the individual cages during the telemetry recording period. The animal room conditions were as follows: temperature control range: 20°C−27°C, relative humidity control range: 35–75%, air exchange: over 10 times/hr, and a 12-hr light/dark cycle (lights on from 7:00 am to 7:00 pm, over 150 lx at 70 or 85 cm above floor level). Each animal was fed 100 g of a pelleted diet once daily. Food was supplied after completion of the clinical observations at 8 hr post-dose or blood sampling at 4 or 5.5 hr post-dose on the days of dosing and at approximately the same time as the dosing days on the days before dosing. The remaining food was removed the following morning. The animals had free access to tap water.

Animal experiments were performed by Eisai Co., Ltd. and Axcelead Drug Discovery Partners, Inc. This study was approved by the Institutional Animal Care and Use Committee (IACUC), Shonan Health Innovation Park, or Eisai Co., Ltd., and conformed to the Guide for the Care and Use of Laboratory Animals published by the National Institutes of Health.

Test drugs

PTZ (Tront Research Chemicals Inc., Toronto, Canada), 4-AP (Tokyo Chemical Industry Co., Ltd., Tokyo, Japan), bupropion hydrochloride (Tokyo Chemical Industry Co., Ltd.), kainic acid (Hello Bio Inc., Bristol, UK), and ranolazine dihydrochloride (Tokyo Chemical Industry Co., Ltd.) were used as convulsants. Atropine sulfate (Mylan EPD G.K., Tokyo, Japan), atenolol (Sigma-Aldrich Co., LLC, St. Louis, MO, USA), and clonidine hydrochloride (Tokyo Chemical Industry Co., Ltd.) were used as non-convulsants. For 4-AP, bupropion, and kainic acid, range-finding studies have been performed on other male NHPs prior to this study. From the previous reports, 4-AP, bupropion, and kainic acid induced tonic or clonic convulsions at doses of 2.5, 50, and 20 mg/kg, respectively. Hence, a half or one-third of those doses were set as high doses. The dose volume was set at 2 mL/kg for each subcutaneous and intravenous dose, and 5 mL/kg for the oral dose. The dosage settings used in this study are summarized in Table 1.

Table 1. Dose settings of drugs used in this study.

Data collection

The vehicles used in this study were water for the injection for ranolazine and saline for injection for the other compounds. Appropriate amounts of the compounds were weighed, dissolved in vehicle, and then administered to the animals. We titrated the doses for at least a 3-day interval in the experiment for each drug, and an interval of at least 1 month was interposed between experiments. PTZ, 4-AP, bupropion, and kainic acid at four dose levels (including vehicle) were administered using an escalating dose design, and ranolazine at two or three dose levels (including vehicle) and non-convulsants were administered using a crossover dose design. Because convulsions occurred in two NHPs after high-dose kainic acid administration, all NHPs were administered diazepam at 8 hr post-dose to suppress convulsions. Convulsions also occurred in one NHP at a high dose of ranolazine on the first day of dosing. However, the ranolazine-induced convulsions were brief and transient; thus, no anticonvulsant was administered. Thereafter, the administration of a high dose of ranolazine was discontinued for the remaining two NHPs. All non-convulsants were included in the crossover study. Before dosing with each drug, approximately 10 mL of saline subcutaneously or 20 mL of tap water orally was administered to the animals once daily for 1−4 days to habituate them to the dosing procedure and acquire training data.

Blood pressure, ECG waveforms, body temperature, and activity counts were continuously recorded from 16−24 hr pre-dose to approximately 24 hr post-dose using a telemetry system; human access to the animal room was limited during the recordings. Cardiovascular parameters were not evaluated at doses at which the animals convulsed. The animals’ behaviors and clinical signs were recorded using a charge-coupled device (CCD) camera and a digital recorder. All parameters were recorded using a transmitter and receiver (RMC-1 or TRX-1, Data Sciences International), and a telemetry data collection and analysis computer system (Ponemah Physiology Platform ver. 5.0, 5.2, and 5.3; Data Sciences International). After removing abnormal waveforms from the ECG data, all RR intervals (RRIs) obtained in this study (including the pre-dose period) were output to an Excel file for each beat. The timing of convulsions, dose (test agent or diazepam), feeding, and technician entry and exit were also recorded. ECG signals were recorded at 1000 Hz.

HRV is known to be affected by sleep (Gong et al., 2016; Gosselin et al., 2002); therefore, only HRV parameters were evaluated within 6 hr of administration to avoid including the dark period in the evaluation. As previously alluded, PTZ data were not collected for this study, but were secondary to those from a previous study (Nagata et al., 2021).

Convulsion prediction

In this study, we employed a multivariate statistical process control (MSPC) to predict convulsions. MSPC is a correlation-based anomaly-detection technique extensively utilized in the field of artificial intelligence for medical applications (Fujiwara et al., 2016). This method identifies samples that deviate from the predominant trend in the modeling dataset. The Q-statistic serves as an index for measuring the dissimilarity between a given sample and the modeling data, emphasizing the correlation among variables. When the Q-statistic surpasses a predefined control limit, MSPC flags the sample as an anomaly. The proposed convulsion prediction method, which consists of model training and prediction, was based on a previous study (Nagata et al., 2021). The MSPC model was built on a compound-by-compound basis for each test drug, while a common MSPC model was built for three negative control drugs because pre-dose data were common. The model training procedure was carried out as follows:

1. ECG data was collected without using convulsion-inducing drugs.

2. RRIs were extracted from the ECG data.

3. HRV features were calculated (as described in section “Analysis procedure” and Table 2) from the RRI data.

4. Principal component analysis (PCA) was applied to the HRV data.

5. An MSPC model was built for use in the prediction.

Convulsion prediction was performed following the procedure below:

1. ECG of an animal was measured.

2. RRIs were extracted from the ECG data.

3. HRV features were calculated from the RRI data.

4. Q-statistics were calculated from the HRV features using the trained MSPC model.

5. Q-statistics were modified using the median.

6. Q-statistics were standardized using the Q-statistics of the entire dosage case.

7. The sum of the areas under the curve was calculated, which is the time-series function of the Q-statistics in the range in which the value of the function exceeds the threshold.

8. Convulsions were predicted using abnormal scores for each drug dosage based on the sum of the areas under the curve.

The reason for correcting the Q-statistics by the median in Step 5 of convulsion prediction is to remove sharp changes that are unlikely to be drug-induced effects. Moreover, the standardization in Step 6 was performed to uniquely determine the threshold.

Table 2. HRV features used in this study

Analysis procedure

The R-waves in the collected ECG data of all the NHPs were detected using a first-derivative-based peak detection algorithm, and each RRI was calculated. We used twelve standard HRV features (Table 2) for convulsion prediction. The frequency-domain analysis must be modified for NHPs because their heart rate (approximately 90–150 beats/min) is considerably higher than that of humans (approximately 60–80 beats/min). The frequency-domain features were defined as the powers of a specific frequency range in the power-spectrum density of the RRI data. In this study, the powers in 0.01−0.2 Hz, 0.2−0.8 Hz, and ≤ 0.01 Hz were adopted as low frequency (LF), high frequency (HF), and very low frequency (VLF), respectively, as previously published in a study on NHPs (Shively et al., 2007). A rectangular moving window with a size of 3 min was used for analysis. For the frequency-domain feature calculation, the RRI data were resampled, arranging the sampling points at equal intervals, which were interpolated by means of a third-order spline, and 4 Hz resampling was adopted. An autoregressive model of order 40 was used in this study. The convulsion prediction model was trained using HRV data obtained during the acclimatization of animal groups scheduled to receive the same drug.

Vehicle and drug dosing data were divided into two datasets: validation and testing. As the drugs were administered for only 1 day for each dosage, we considered two methods for splitting the data. (1) Data were divided into 5-minute intervals and removed alternately to produce two datasets from 1-day data and used for validation and testing. (2) Low and medium doses were used for validation, and the vehicle and high doses were used for testing. However, both these methods have drawbacks. In Method 1, the validation and test data were similar. These test data may not be appropriate for evaluating the generalization capability of this analysis approach. In Method 2, the dosage of the drug was different between validation and test. Therefore, it is not clear whether the validation was done correctly. To determine the influence of these drawbacks on the analysis, we validated each method for a specific drug. Consequently, the hyperparameters of both methods converged to close values, and we concluded that both methods could be used for validation. Because the solvents specific for some drugs were not available in this study, we decided to perform validation using Method 1.

Validation data were used to tune the hyperparameters, that is, the number of principal components denoted by R, which is the threshold used to calculate the sum of the areas denoted by S. In addition, validation data were used to determine the statistics that should be used to calculate the sum of the areas under the curve. The hyperparameters R and S, as well as the selection of statistics, were newly explored owing to the different data set from the previous study (Nagata et al., 2021), and were determined so that the alarm frequency was lower for the vehicle and low dose and higher for medium and high dose groups. When R=8, S=2, and Q-statistics were used, the clearest differences in the drug dose were observed. The cumulative proportion exceeded 99% under any condition in this study. The T² statistic was also calculated; however, it was not used for evaluation in this study because the addition of this parameter did not change the prediction performance.

A convulsion alarm is triggered when the Q-statistic surpasses its control limit for a duration exceeding 1 second. The number of times and cumulative time during which the Q-statistic surpassed the predefined control limit were evaluated as indices for convulsion prediction. HF and LF/HF were examined as indices to determine changes in autonomic nervous activity. The 6-hr total values of HF and the LF/HF were normalized by setting the value at the time of vehicle dosing for each individual and test drug to 1.0.

Due to the exploratory and descriptive nature of this study, no statistical analyses were conducted.

RESULTS

A dose-dependent increase in the mean convulsive indices (number of alarms and duration) was observed with PTZ (Fig. 1), confirming that the results obtained with the present methodology are similar to those reported in a previous study (Nagata et al., 2021).

Fig. 1

Number of alarms and total duration of alarms after pentylenetetrazol dosing. Each plot indicates values by the individual, and horizontal bars indicate mean values. Analyses were performed until 6 hr post-dose.

The convulsive index, a measure of convulsion prediction after a dose of convulsant, is shown in Fig. 2, while changes in the autonomic index (HF and LF/HF) are shown in Fig. 3.

Fig. 2

Number of alarms and total duration of alarms after dosing of convulsants. Each plot indicates values by individual, and horizontal bars indicate mean values. Analyses were performed until 6 hr post-dose. The same animals were used for 4-AP, bupropion, and kainic acid, and the same markings in the plots indicate the same animals. Kainic acid induced convulsions in two males at the high dose as indicated by a red circle and blue triangle. For ranolazine, a different animal was used, and the markings are not consistent with the other graphs. Ranolazine induced convulsions in one male at the high dose as indicated by a blue triangle.

Fig. 3

Normalized autonomic responses to convulsants. Each plot indicates values by the individual, and horizontal bars indicate mean values. Analyses were performed until 6 hr post-dose. The same animals were used for 4-AP, bupropion, and kainic acid, and the same markings in the plots indicate the same animals. For ranolazine, a different animal was used, and the markings are not consistent with the other graphs. The data are normalized to each animal’s control value.

For 4-AP, dose-dependent increases in the mean convulsive indices were observed at medium and high doses, whereas no changes were detected at the low dose. At middle and high doses, a decrease in mean HF and an increase in LF/HF were observed, suggesting sympathetic dominance increases in blood pressure were recorded at 2 hr post-dose at the medium dose and between 0.5 and 8 hr post-dose at the high dose. The high dose also increased heart rate and decreased body temperature 1−4 hr post-dose. Moreover, vomiting was observed at 8 hr post-dose at medium and high doses.

Bupropion did not exhibit an increase in the mean convulsive indices, even at the high dose, but instead showed a decreasing trend. The convulsive index values observed in the control group were higher than those in the low/middle/high dose groups, which was attributed to the standardization process conducted on a compound-by-compound basis. This process resulted in a relative increase in the values for the control group due to the reduction in values observed in the dosing groups. At medium and high doses of bupropion, the mean HF tended to decrease, but the LF/HF did not change. There were increases in blood pressure and heart rate at 1 and 2 hr at the high dose, and no changes in behavior or clinical signs were noted at any dose.

With kainic acid, the mean convulsive indices tended to increase at low and high doses but remained unchanged at the medium dose. No clinical signs were observed up to the medium dose, but convulsions and some clinical signs, including tremors, lateral or prone positioning, and vomiting, were observed at the high dose. At the low dose, HF increased and heart rate and activity decreased after 2−4 hr. At the medium dose, decreases in heart rate and activity, and increases in blood pressure were observed at 0.5−2 hr. At the high dose, a decrease in HF and an increase in LF/HF were noted, indicative of a sympathetically dominant condition.

A high dose of ranolazine induced transient convulsions approximately 30 min after administration in one male and increased convulsive indices. There was an increasing trend in mean convulsive indices, even at the low dose. At the low dose, there was a decrease in blood pressure from 0.5−1 hr. The mean HF and LF/HF increased in a dose-dependent manner.

The convulsion index results after the administration of non-convulsants are shown in Fig. 4, whereas the results for the autonomic indices are shown in Fig. 5. For all non-convulsants, the mean convulsive indices were elevated. Both atenolol and clonidine increased HF, whereas clonidine decreased LF/HF, indicating parasympathetic activation. In contrast, atropine induced a decrease in the HF and an increase in the LF/HF in all animals, suggesting sympathetic activation. Atenolol caused a reduction in heart rate. Six hours after the dose, atropine resulted in increased blood pressure and heart rate; a decrease in body temperature; and mydriasis. With clonidine, a decrease in blood pressure; an increase in heart rate; a decrease in body temperature; and reduced activity were observed.

Fig. 4

Number of alarms and total duration of alarms after dosing of non-convulsants. Each plot indicates values by the individual, and horizontal bars indicate mean values. Analyses were performed until 6 hr post-dose. The same markings in the plots indicate the same animals among the graphs.

Fig. 5

Normalized autonomic responses to non-convulsants. Each plot indicates values by the individual, and horizontal bars indicate mean values. Analyses were performed until 6 hr post-dose. The same markings in the plots indicate the same animals among the graphs. The data are normalized to each animal’s control value.

DISCUSSION

The present method demonstrated increases in the convulsive indices following the administration of convulsive doses for all four convulsant drugs tested, except bupropion, suggesting potential utility in predicting convulsions for 4-AP, kainic acid, and ranolazine, in addition to PTZ. Changes in either or both the HF and LF/HF were also noted. The non-convulsive drugs atenolol, atropine, and clonidine changed HF and LF/HF as expected, reflecting their individual pharmacological effects. The results of this study suggest that HF and LF/HF serve as reliable indicators of the autonomic nervous system. Regardless of whether the test compound was convulsant, the convulsive index appeared to correlate with, or even predict, alterations in autonomic nerve activity (HF and LF/HF), enabling the detection of changes even in the absence of autonomic-derived findings in clinical observation. This suggests that the convulsive index could be a candidate biomarker and a reliable predictor of drug-induced convulsions.

Regarding 4-AP, the applied method successfully predicted convulsions, with both HF and LF/HF manifesting sympathetic changes, and convulsive indices seemingly shifting in alignment with autonomic nervous system effects. 4-AP is known to cause seizures mainly via glutamate receptors and sodium channels, and is resistant to seizure suppression by GABAergic drugs (Peña and Tapia, 2000). This provides evidence that our method can be used to predict convulsions induced by mechanisms independent of the GABA-related pathways. The convulsion index did not change after bupropion administration. However, the detailed mechanism of bupropion-induced seizures remains unclear. Bupropion has been reported to inhibit GABAergic neurons (Amirabadi et al., 2014); hence, bupropion-induced convulsions may be GABA-related. In contrast, bupropion suppressed 4-AP-mediated glutamate release (Lin et al., 2011). It has been reported that 30 mg/kg bupropion administered intraperitoneally showed anticonvulsant effects against electroshock, while 120 mg/kg induced convulsions in mice (Tutka et al., 2004). Therefore, we hypothesize that, in NHPs, at lower doses, the neuroexcitatory inhibitory effects of bupropion would predominate over its neuroexcitatory effects. We posited that the balance between the inhibitory and excitatory effects undergoes rapid reversal as the dose approaches the convulsive threshold, thereby complicating the prediction of convulsions. Although the dose of bupropion used in this study was up to one-third of the convulsive dose, future investigations are needed to ascertain whether higher bupropion doses can induce discernible changes in convulsive indices and autonomic nervous system parameters. Kainic acid, a well-known convulsant that stimulates glutamate receptors and provokes convulsions (Victor Nadler, 1981), displayed elevated convulsive indices, even in animals that did not manifest convulsions at the convulsive dose, suggesting that changes are detectable at or near this dose. Given that the medium dose of kainic acid set in this study was one-tenth of the convulsion dose, changes might have been discernible if the dose was half or one-third of the convulsion dose. The convulsive index and HF of kainic acid showed increasing trends towards at the low dose but remained unchanged at the medium dose. At the high dose of kainic acid, an elevation in LF/HF was observed, indicating a reversal in autonomic balance between low and high doses. This suggests that kainic acid may function as a complex modulator of autonomic balance. Although the convulsion-inducing mechanism of ranolazine remains unclear, it has been reported that ranolazine can cause seizures at high doses (Akil et al., 2015). Paradoxically, ranolazine exhibits antiepileptic effects by inhibiting abnormal currents via blocking sodium channels (Kahlig et al., 2014). Similar to bupropion, ranolazine may exhibit a sudden switch between anticonvulsant and convulsant effects. Further studies should be conducted using higher doses of ranolazine. A limitation of this study is that although the experiments were conducted at two different facilities, the same compounds were not evaluated; therefore, the effects of inter-facility differences are not clear. In addition, because of the large individual differences and small number of animals available, the results of this study may not be generalizable in other animals. Another limitation is that electroencephalogram assessments were not conducted on NHPs in the present study. Elucidating the correlation between the convulsive index and electroencephalogram outcomes may be important in the future.

In the present study, the utility of the convulsive index to predict convulsions induced by certain convulsant drugs was demonstrated. However, all non-convulsive drugs also increased the convulsive index. Drugs that affect the autonomic nervous system can be detected irrespective of the presence of convulsions, necessitating careful consideration of false positives. Therefore, this method might not be regarded as a direct and specific convulsion detector but as a sensor sensitive to fluctuations within the autonomic nervous system. For practical applications in safety studies in NHPs in drug development, it may be advisable to apply this method to drug candidates already confirmed to possess convulsant potential based on previous studies. The index could be used effectively as a convulsive marker if the correlation between the actual convulsive dose and the dose that increases the convulsive index is confirmed for each compound. Future studies testing a broader array of drugs may help elucidate the applicability of this method. In future endeavors, this method can be applied to humans for the proactive prevention of unexpected seizures, thereby contributing to the implementation of safer clinical trials in the development of new drugs.

ACKNOWLEDGEMENTS

This study was conducted in cooperation with Yoshiyuki Furukawa, Ryota Hayashi, and Tomoki Shimada of Axcelead Drug Discovery Partners Inc. We would like to thank Editage (www.editage.jp) for English language editing.

Conflict of interest

Kauhiro Kuga and Harushige Ozaki are employees of Takeda Pharmaceutical Company Ltd. Motohiro Shiotani, Hiroshi Mizuno, and Yusaku Matsushita are employees of Eisai Co. Ltd. Kentaro Hori is an employee of Quadlytics Inc. Kohei Hayashi is an executive of Quadlytics Inc.

REFERENCES

•  Akil,  N.,  Bottei,  E. and  Kamath,  S. (2015): Ranolazine overdose-induced seizures. Am. J. Emerg. Med., 33, 1843.e5-1843.e6.
•  Amirabadi,  S.,  Pakdel,  F.G.,  Shahabi,  P.,  Naderi,  S.,  Osalou,  M.A. and  Cankurt,  U. (2014): Microinfusion of bupropion inhibits putative GABAergic neuronal activity of the ventral tegmental area. Basic Clin. Neurosci., 5, 182-190.
•  Authier,  S.,  Arezzo,  J.,  Delatte,  M.S.,  Kallman,  M.J.,  Markgraf,  C.,  Paquette,  D.,  Pugsley,  M.K.,  Ratcliffe,  S.,  Redfern,  W.S.,  Stevens,  J.,  Valentin,  J.P.,  Vargas,  H.M. and  Curtis,  M.J. (2016): Safety pharmacology investigations on the nervous system: an industry survey. J. Pharmacol. Toxicol. Methods, 81, 37-46.
•  Baumgartner,  C.,  Koren,  J.,  Britto-Arias,  M.,  Schmidt,  S. and  Pirker,  S. (2019): Epidemiology and pathophysiology of autonomic seizures: a systematic review. Clin. Auton. Res., 29, 137-150.
•  Champeroux,  P.,  Martel,  E.,  Jude,  S.,  Laigot,  C.,  Laveissière,  A.,  Weyn-Marotte,  A.A.,  Fowler,  J.S.,  Maurin,  A.,  Richard,  S. and  Babuty,  D. (2013): Power spectral analysis of heart rate variability in cynomolgus monkeys in safety pharmacology studies: comparative study with beagle dogs. J. Pharmacol. Toxicol. Methods, 68, 166-174.
•  Engel,  J. Jr. and  Pitkänen,  A. (2020): Biomarkers for epileptogenesis and its treatment. Neuropharmacology, 167, 107735.
•  Fujiwara,  K.,  Miyajima,  M.,  Yamakawa,  T.,  Abe,  E.,  Suzuki,  Y.,  Sawada,  Y.,  Kano,  M.,  Maehara,  T.,  Ohta,  K.,  Sasai-Sakuma,  T.,  Sasano,  T.,  Matsuura,  M. and  Matsushima,  E. (2016): Epileptic Seizure Prediction Based on Multivariate Statistical Process Control of Heart Rate Variability Features. IEEE Trans. Biomed. Eng., 63, 1321-1332.
•  Gong,  X.,  Huang,  L.,  Liu,  X.,  Li,  C.,  Mao,  X.,  Liu,  W.,  Huang,  X.,  Chu,  H.,  Wang,  Y.,  Wu,  W. and  Lu,  J. (2016): Correlation analysis between polysomnography diagnostic indices and heart rate variability parameters among patients with obstructive sleep apnea hypopnea syndrome. PLoS One, 11, e0156628.
•  Gosselin,  N.,  Michaud,  M.,  Carrier,  J.,  Lavigne,  G. and  Montplaisir,  J. (2002): Age difference in heart rate changes associated with micro-arousals in humans. Clin. Neurophysiol., 113, 1517-1521.
•  Grainger,  A.I.,  King,  M.C.,  Nagel,  D.A.,  Parri,  H.R.,  Coleman,  M.D. and  Hill,  E.J. (2018): In vitro models for seizure-liability testing using induced pluripotent stem cells. Front. Neurosci., 12, 590.
•  Kahlig,  K.M.,  Hirakawa,  R.,  Liu,  L.,  George,  A.L. Jr.,  Belardinelli,  L. and  Rajamani,  S. (2014): Ranolazine reduces neuronal excitability by interacting with inactivated states of brain sodium channels. Mol. Pharmacol., 85, 162-174.
•  Kuga,  K.,  Ozaki,  H. and  Fujiki,  M. (2023): Motor-evoked potentials after focal electrical stimulation predict drug-induced convulsion potentials in rats. Toxicol. Mech. Methods, 33, 123-130.
•  Lin,  T.Y.,  Yang,  T.T.,  Lu,  C.W. and  Wang,  S.J. (2011): Inhibition of glutamate release by bupropion in rat cerebral cortex nerve terminals. Prog. Neuropsychopharmacol. Biol. Psychiatry, 35, 598-606.
•  Nagata,  S.,  Fujiwara,  K.,  Kuga,  K. and  Ozaki,  H. (2021): Prediction of GABA receptor antagonist-induced convulsion in cynomolgus monkeys by combining machine learning and heart rate variability analysis. J. Pharmacol. Toxicol. Methods, 112, 107127.
•  Nadler,  J.V. (1981): Minireview. Kainic acid as a tool for the study of temporal lobe epilepsy. Life Sci., 29, 2031-2042.
•  Odawara,  A.,  Matsuda,  N.,  Ishibashi,  Y.,  Yokoi,  R. and  Suzuki,  I. (2018): Toxicological evaluation of convulsant and anticonvulsant drugs in human induced pluripotent stem cell-derived cortical neuronal networks using an MEA system. Sci. Rep., 8, 10416.
•  Peña,  F. and  Tapia,  R. (2000): Seizures and neurodegeneration induced by 4-aminopyridine in rat hippocampus in vivo: role of glutamate- and GABA-mediated neurotransmission and of ion channels. Neuroscience, 101, 547-561.
•  Shively,  C.A.,  Mietus,  J.E.,  Grant,  K.A.,  Goldberger,  A.L.,  Bennett,  A.J. and  Willard,  S.L. (2007): Effects of chronic moderate alcohol consumption and novel environment on heart rate variability in primates (Macaca fascicularis). Psychopharmacology (Berl.), 192, 183-191.
•  Tukker,  A.M. and  Westerink,  R.H. (2021): Novel test strategies for in vitro seizure liability assessment. Expert Opin. Drug Metab. Toxicol., 17, 923-936.
•  Tutka,  P.,  Barczynski,  B. and  Wielosz,  M. (2004): Convulsant and anticonvulsant effects of bupropion in mice. Eur. J. Pharmacol., 499, 117-120.