|Home | About | Journals | Submit | Contact Us | Français|
Acquired epilepsy (i.e., after an insult to the brain) is often considered to be a progressive disorder, and the nature of this hypothetical progression remains controversial. Antiepileptic drug treatment necessarily confounds analyses of progressive changes in human patients with acquired epilepsy. Here we describe experiments testing the hypothesis that development of acquired epilepsy begins as a continuous process of increased seizure frequency (i.e., proportional to probability of a spontaneous seizure) that ultimately plateaus. Using nearly continuous surface cortical and bilateral hippocampal recordings with radiotelemetry and semi-automated seizure detection, the frequency of electrographically recorded seizures (both convulsive and non-convulsive) was analyzed quantitatively for about 100 days after kainate-induced status epilepticus in adult rats. The frequency of spontaneous recurrent seizures was not a step-function of time (as implied by the “latent period”); rather, seizure frequency increased as a sigmoid function of time. The distribution of inter-seizure intervals was non-random, suggesting that seizure clusters (i.e., short inter-seizure intervals) obscured the early stages of progression, and may have contributed to the increase in seizure frequency. These data suggest that (1) the latent period is the first of many long inter-seizure intervals and a poor measure of the timeframe of epileptogenesis; (2) epileptogenesis is a continuous process that extends much beyond the first spontaneous recurrent seizure; (3) uneven seizure clustering contributes to the variability in occurrence of epileptic seizures; and (4) the window for anti-epileptogenic therapies aimed at suppressing acquired epilepsy probably extends well past the first clinical seizure.
Epilepsy is generally divided into genetic and acquired forms; acquired epilepsy is thought to arise from a brain insult (e.g., status epilepticus) and generally occurs after a latent period. Temporal lobe epilepsy (TLE) is an important form of acquired epilepsy that usually includes mesial temporal sclerosis (Margerison and Corsellis, 1966) with complex partial seizures that may spread secondarily to cause convulsive seizures. The time between a brain insult and onset of clinical seizures (i.e., the latent period) is highly variable, ranging from months to years (e.g., Annegers et al., 1998; French et al., 1993; Mathern et al., 1995). Antiepileptic drug therapy is typically initiated when a patient first reports seizures. The reliance on self-reporting of seizures from patients with post-seizure amnesia, the lack of continuous electroencephalographic (EEG) monitoring, and the use of therapy to suppress seizures have prevented quantitative studies on the natural history of epileptogenesis in humans. Animal models with spontaneous recurrent seizures may provide insights into the temporal features of acquired epilepsy.
The latent period is generally considered to be the time when a brain insult induces molecular and cellular mechanisms that lead to spontaneous recurrent seizures, and is thus considered a measure of epileptogenesis (Walker et al., 2002). The presence of a latent period, and the apparent variability or unpredictability of the subsequent spontaneous recurrent seizures, has led to the view that epileptogenesis occurs only during the time between the insult and the first clinical seizure. Therefore, development of epilepsy is sometimes represented as a “step function” of time after brain injury (Fig. 1A). Many other studies have suggested that the development of acquired epilepsy can be progressive, at least initially (Bertram and Cornett, 1993, 1994; French et al., 1993; Mathern et al., 1995; Hellier et al., 1998; Tasch et al., 1999; Fuerst et al., 2001; Gorter et al., 2001; Wuarin et al., 2001). Epileptogenesis clearly occurs during the latent period, but another hypothesis states that the end of the latent period (i.e., the first clinical seizure) is not a terminal milestone for the mechanisms of epileptogenesis; rather, the process of epileptogenesis is a continuous function of time that persists long after the first clinical seizure. A corollary to this hypothesis is that a prolonged period of increasing seizure probability is hypothetically manifest as a gradual increase in seizure frequency during acquired epileptogenesis, and that a sigmoid function of seizure frequency versus time after the brain insult best represents acquired epileptogenesis (Fig. 1B). This alternative hypothesis would define the latent period as a probabilistic concept, and predicts that the latent period is only the first of several long inter-seizure intervals.
The data in this study – using the repeated, low-dose kainate model of acquired epilepsy - suggest a new perspective on the latent period and provide evidence that acquired epileptogenesis is a continuous or progressive function of time where clusters of seizures may obscure the continuous nature of the increase in seizure frequency, which may be quite difficult to detect without prolonged and continuous electrographic monitoring.
All of the seizures in these experiments were recorded electrographically with radiotelemetry and behaviorally with video monitoring. Electrographic (EEG) seizures in this study refer to any seizure that was recorded electrically, which was virtually all of the seizures. These electrographically recorded seizures could either be non-convulsive seizures or convulsive motor seizures. A modified Racine scale (Racine, 1972; Ben-Ari, 1985) was used to characterize convulsive motor seizure severity as follows: Class III seizures were defined by forelimb clonus and a lordotic posture; Class IV seizures included forelimb clonus and rearing; Class V seizures displayed a Class IV seizure with loss of the righting reflex. All convulsive motor seizures were associated with electrographic (EEG) seizure activity, and would be considered to be generally similar to a clinical seizure in a human patient. The behaviors during non-convulsive seizures were essentially those typically described as Racine (1972) Class I or Class II seizures; because characteristic electrographic seizures (e.g., Fig. 2) were used to identify these non-convulsive seizures, they are often described here as electrographic non-convulsive seizures, and could be a model for complex partial seizures. The EEG-seizure latent period was the time from the day of kainate treatment to the first electrographically recorded seizure, which was typically a non-convulsive seizure. The motor-seizure latent period was the time to the first convulsive motor seizure, which was equal to (one case) or longer than the EEG-seizure latent period.
Male Sprague Dawley rats (Harlan, Indianapolis, IN) weighing 180–200 g were kept under 12-h light and dark conditions throughout this study. All procedures in this study were approved by the Institutional Animal Care and Use Committee of Colorado State University. A detailed description of the surgery and recording techniques can be found in Williams et al. (2006). Intrahippocampal electrodes were placed bilaterally in the granule cell layer. The dentate gyrus was chosen as a target for chronic intracranial electrographic recording of seizures because of the pathological and electrographic evidence suggesting that the dentate gyrus is involved in the seizures of patients with TLE, and because an extensive in vivo and in vitro data base exists for this structure in rodent models of TLE, such as the kainate model (e.g., Margerison and Corsellis, 1966; Williamson et al., 1995; see Dudek et al., 2002, 2005; Dudek and Sutula; 2007; Sutula and Dudek, 2007 for reviews). Briefly, the intrahippocampal electrodes were constructed from Teflon-coated steel wire (A-M Systems, Everett, WA) and were implanted using stereotaxic coordinates (bilateral dorsal hippocampus; rostral-caudal −4.0 mm, medial-lateral (ML) ±2.5 mm, dorsal-ventral (DV) −3.3 mm at a bite of −4.0 mm); the position in the cell body layer was confirmed by increased spike activity, detected with an audio monitor. The dural recording electrode was placed on the left hemisphere approximately 2 mm lateral and 1 mm rostral to the left intrahippocampal recording electrode (see Williams et al., 2006 for details). The dural reference electrodes were placed approximately 1.5 mm caudal to the intrahippocampal recording electrodes. The radiotelemetry unit (i.e., the three-channel PhysioTel® Multiplus series transmitter F50-EEE, Transoma Medical, Arden Hills, MN) was positioned in a pocket created subcutaneously in the flank region through a 2-cm skin incision behind the scapula, and the leads tunneled forward to the rostral head incision. All electrodes were fixed in place with dental acrylic. The left and right intrahippocampal leads were connected to the male ends of the radiotelemetry unit, and then covered and fixed to the skull with dental acrylic. The local anesthetic bupivacaine 0.4 ml (7.5 mg/ml) was applied liberally around the open cranial incision. Both incision sites were closed using 4–0 Dermalon (American Cyanamid Co., Danbury, CT). The animal was given 4–6 ml of warmed Ringer’s solution (SC), 0.06 ml buprenorphine (0.1 mg/kg, SC), removed from anesthesia and placed back in its cage under a heat lamp for 30 min. All animals were given 0.06 ml buprenorphine (0.01 mg/kg, SC) and 0.2 ml penicillin (300,000 IU, SQ) for 3 days following surgery. The toenails were cut while the rats were anesthetized, and if necessary, once per week to decrease scratching and irritation at the incision sites. Rats were housed in individual cages within an isolated room after surgery and throughout the study, and were given food and water ad libitum. For more details of this surgery, see Williams et al. (2006).
This study used the Dataquest A.R.T. Analog software from Transoma Medical (Arden Hills, MN), which is an analog acquisition system. The cages were placed on individual radio receiving plates (RPC-1; Transoma Medical, Arden Hills, MN), which captured data signals from the radiotransmitter and sent them to an input exchange matrix. The signal was then sent to the computer running Dataquest A.R.T. Analog software, which converted the digital output of the receiver into a calibrated analog output. Custom-made software (K. Staley) was used to acquire the data with routines written in Visual Basic 6.0 (Microsoft, Seattle, WA). The electrographic data were written to DVD for analysis offline.
Details of the kainate treatment, the response to kainate injections, and an analysis of status epilepticus have been previously published (Hellier et al., 1999; Dudek et al., 2005; Hellier and Dudek, 2005; Williams et al., 2006). One to 2 weeks after the implantation surgery, rats (n=9) were injected with kainate (5 mg/kg/hr; IP; Sigma, St. Louis, MO) diluted in sterile 0.9% saline at 2.5 mg/ml. Rats were continuously monitored for electrographic and convulsive motor seizures. Hourly kainate treatment continued in animals with convulsive seizures until Class III, IV, or V seizures were evoked for at least 3 h (i.e., >10 convulsive seizures per hour). If an animal was nearing its endpoint, half-doses (2.5 mg/kg) were given to avoid excessive toxicity and mortality (for a review of this model, see Dudek et al., 2005). The mean total dose of kainate was 2.98 mg ± 1.34 mg. Control animals (n=6) were treated with an equivalent volume and number of injections of sterile saline. All rats were given 3–6 ml lactated Ringer’s (subcutaneously) and apple slices following treatment. Subcutaneous fluids were continued for 2–3 days if the animals appeared lethargic. Kainate-induced seizures were complete by 24 h, and the animals were eating and drinking within 24–36 h after kainate treatment.
During kainate treatment, the number and duration of each convulsive seizure was measured. Convulsive seizure duration was measured from the start to the end of clonus and lordotic posturing; animals then typically resumed normal motor activity and righted themselves. The mean number of convulsive motor seizures during kainate treatment was 99.2 ± 20.2, and the mean amount of time in convulsive seizures was 66.3 ± 25.2 min. Neither the total number of convulsive motor seizures during treatment nor the total amount of time in status epilepticus was correlated to the dose of kainate (r2 = −0.124, p = 0.56, r2 = −0.125, p = 0.56, respectively).
The seizure-detection analysis was performed both visually (A.M.W.) and in a semi-automated manner with custom-written software that minimized the potential for bias (for greater detail of the EEG analysis, see White et al., 2006). All of the electrographic recordings were examined for artifacts, and all of the electrographic seizures were confirmed by visual inspection. The investigators were not blinded to treatment or time after kainate-induced status epilepticus. Electroencephalographic seizures were differentiated from background noise by the appearance of large-amplitude (at least 3X baseline), high-frequency (minimum of 5 Hz) activity, with progression of the spike frequency (Fig. 2) that lasted for a minimum of 30 sec. The behavioral data were used to confirm EEG seizure activity versus potential animal-generated noise.
Continuous videotaping of individually housed rats was accomplished with the use of two Color Quad Observation Systems SOD14C4LN (Samsung, Korea). Time stamps for these systems were synchronized to the digitizing computer. Night recordings were performed with a Kodak 1A filter (Eastman Kodak, Rochester, NY) over a safelight, and daytime recordings with a diffuse fluorescent light. The behavioral data were used in this study for the determination of the latent period for convulsive motor seizures (versus non-convulsive seizures) and for differentiating EEG seizure activity from electrical noise generated by jaw-movement artifact or grooming.
Similar previous studies with tethered recordings and perforant path stimulation using this same implantation method consistently revealed that the electrodes were located in the dentate gyrus of the hippocampus (Hellier et al., 1999). A histological examination of the hippocampi with cresyl violet staining for confirmation of hippocampal electrode placement showed that in 8 out of 15 rats (53%), a linear or elliptical glial scar (a common indicator of recording site) was present in or near the dentate gyrus of septal hippocampal sections (see Supplementary Data, Fig. 1). Aside from the glial scar, no evidence of damage from the recording electrodes was detected in any of the animals, even though the expected histopathological alterations from the kainate-induced status epilepticus (Buckmaster and Dudek, 1997) were readily apparent.
Two related issues relating to the histological localization of the recording electrodes deserve consideration in terms of the reported conclusions from these studies. The first issue is the possibility that the intrahippocampal recording electrodes were not actually in the cell body layer of the dentate gyrus. Even though it was not always possible to localize histologically the site of the intrahippocampal recording electrodes, all three electrodes (including the dural electrode) recorded a similar profile of electrographic (i.e., EEG) seizures throughout the recording period for all animals (Fig. 2), thus indicating that the exact electrode placement was not an important factor in the data reported here. The second issue is whether the intrahippocampal electrodes induced sufficient damage to cause, or at least influence, the development of spontaneous recurrent seizures after kainate-induced status epilepticus. If damage from the recording electrodes (i.e., associated with the glial scars) caused non-convulsive or convulsive electrographic seizures, then the control animals should have had seizures, since they also had glial scars; none of the control animals showed any seizures. Another issue was the site of seizure onset, which was not analyzed in this study. This question has been addressed in another model of status epilepticus-induced epilepsy, where more recording electrodes were used (see Bertram, 1997; Bertram et al., 1998). The exact time (and thus, site) of seizure onset can be difficult to determine (see Bower and Buckmaster, 2008).
All graphs and statistical analyses were done using Prism software (Graphpad, San Diego, CA), except where noted. Means and distributions for each of the parameters and groups of parameters were determined. Differences among group means were tested for significance using analysis of variance and post-hoc tests (Student-Newman-Keuls multiple comparisons test). Correlations were calculated using Pearson’s r. Differences with p≤0.05 were considered statistically significant. A Poisson cumulative distribution function was generated with Microsoft Excel using the mean number of days for the slow growth, exponential, and plateau phases, and the mean inter-seizure intervals as the expected probability of seizure occurrence during that phase. These distributions were then compared to the cumulative probability functions for the actual inter-seizure intervals derived for the three different phases of seizure progression. A similar analysis was done with the generation of two ranges of random numbers that were matched to the total number of inter-seizure intervals for each animal during each phase of seizure progression, and limited by the total number of minutes for each individual animal’s three different phases. Intervals were generated by taking the difference between the two ranges of randomly generated intervals. Cumulative distribution functions were then generated for the random intervals and compared to the cumulative distribution functions for the actual inter-seizure intervals. Tests for differences between cumulative distribution functions were performed with a Kolomogorov-Smirnov test, which was run using a program located at: http://home.ubalt.edu/ntsbarsh/Business-stat/otherapplets/ks.htm.
Seven kainate-treated rats and 6 saline-treated rats were recorded for at least 100 days, and 2 other kainate-treated rats were recoded for <100 days (kainate-treated rats, n=9; saline controls, n=6). Figure 2 shows an example of an electrographic recording of a spontaneous seizure. The mean and standard deviation for the number of days to the first electrographically recorded seizure in the kainate-treated rats was 11.0 ± 9.8 days with a median of 8 days and a range of 7 to 37 days (Fig. 3A and Fig. 4A). The high degree of variance in the latent period to the first electrographic seizure was due to a measurement of latent period from a single animal. The mean and standard deviation for the electrograpahic-seizure latent period without this one value (i.e., a possible outlier) was 8.25 ± 0.88 days with a median of 8 days and a range of 7–9 days. Aside from some of the earliest seizures, both hippocampal leads and the dural EEG lead (Fig. 2) showed qualitatively similar activation throughout the recording period for all kainate-treated rats. In all but 1 kainate-treated rat, the first electrographically recorded seizure was non-convulsive. The saline-treated controls were recorded and studied the same as the kainate-treated rats, and no seizures (neither non-convulsive nor convulsive seizures studied with both continuous video and EEG recordings; see below) were detected in any of the animals in the control group. In 5 out of the 9 kainate-treated animals, one 24-h period (i.e., day 7) was not recorded between the first and the eighth day after kainate treatment. Thus, the true latent period (i.e., the time to the first seizure if all the time preceding the first seizure was known to be seizure free) for electrographically recorded seizures was in error by less than 1 day for 5 out of 9 kainate-treated rats. No significant correlations between the total kainate dose during treatment or the total number of convulsive motor seizures observed during kainate treatment and the latent period for electrographic seizures were observed (r2 = 0.007, p = 0.9 and r2 = 0.08, p = 0.6, respectively). These data indicate that the latent period for electrographic seizures was about 1 week (Fig. 3A and Fig. 4A) for this model of TLE. Furthermore, the latent period to the first electrographic seizure was not detectably dependent on the severity of the status epilepticus.
The mean number of days to the first observed convulsive motor or “clinical” seizure (i.e., a behavioral score of Class III or above; see Racine, 1972) in all 9 rats was 18.3 ± 10.2 days with a median of 14.5 days and a range of 10 to 37 days (Fig. 4A). In only 1 out of 9 rats was the first electrographically recorded seizure a convulsive motor seizure, with a mean of 4.0 ± 1.1 electrographic non-convulsive seizures occurring prior to the first convulsive motor seizure. A behavioral score was determined for the first observed convulsive motor seizure in all 9 rats and the mode was a Class V (i.e., Class V is the most severe seizure type). When the video record of the first 20 electrographically recorded seizures was examined for each animal, 63% of the first 20 electrographic seizures across the 9 animals were classified as convulsive motor seizures, and the mode of the behavioral scores for these convulsive motor seizures was also Class V. These data show that non-convulsive electrographic seizures usually preceded convulsive motor seizures by about 1 week, and that multiple non-convulsive electrographic seizures almost always occurred prior to the appearance of the first convulsive motor seizure. This finding is similar to what has been shown previously by Bertram and Cornett (1993). Furthermore, in the kainate model, convulsive motor seizures tended to be severe once motor seizures occurred, and the convulsive motor seizures generally remained severe for the first 20 seizures and beyond.
In order to address the hypotheses that electrographically recorded seizures occur as a step function (Fig. 1A) or as a continuous function of time (Fig. 1B), the latent periods (for both the first electrographically recorded non-convulsive seizure and the first convulsive motor seizure) were compared to the longest inter-seizure interval for the first 20 seizures (Fig 4A). The successive inter-seizure intervals for the first 20 seizures were surprisingly variable, and thus required a semi-log plot to encompass the range of values (Fig 4B). The mean and standard deviation of the longest inter-seizure interval across all detected EEG seizures per animal was 4.36 ± 2.65 days, with a median of 4.1 days and a range of 0.94 to 8.5 days (Fig. 4A, in this analysis, data from the periods without recordings were excluded). Because the longest inter-seizure intervals were approximately 4 to 8 days (Fig. 4B), these inter-seizure intervals were only slightly less than the latent periods to the first electrographically recorded non-convulsive seizures (Fig. 4A). Although a positive trend was present, no significant correlation was found between the latent period to the first electrographic seizure and the longest inter-seizure interval (r2 = 0.47, p = 0.06). This result indicates that the latent period to the first electrographically recorded seizure was similar across the population of kainate-treated rats to the longest inter-seizure interval observed during the early stages of epileptogenesis, which implies that the onset of epileptogenesis – as assessed by measuring the time after kainate-induced status epilepticus of the initial spontaneous recurrent electrographic seizures (i.e., both non-convulsive and convulsive) – was gradual or virtually continuous. This finding does not challenge the concept that an actual latent period exists, but it does support the hypotheses that (1) the progression of epileptogenesis is a “continuous process,” and (2) the duration of the latent period is a suboptimal measure of epileptogenesis. On the other hand, the shortest inter-seizure intervals (Fig. 4B) were as short as 1–4 h, nearly two orders-of-magnitude less than the longest intervals, which indicates that the initial spontaneous recurrent seizures often occurred in clusters (i.e., inter-seizure intervals of at least a few days separated by inter-seizure intervals of only a few hours). Thus, in the comparison of the latent periods and the longest inter-seizure intervals, the longer inter-seizure intervals conceptually and practically represent the inter-cluster intervals while the shorter inter-seizure intervals represent the intra-cluster intervals, as will be clarified in subsequent sections.
Plots of successive inter-seizure intervals (i.e., the reciprocals of instantaneous seizure frequency) for the first 20 seizures did not reveal clear evidence of a progressive increase in seizure frequency (Fig. 4B). In order to better assess the temporal characteristics of the onset of epileptogenesis over a longer period of time, the number of electrographically recorded seizures per day was determined, and the daily seizure frequency was calculated over the 100-day period after kainate treatment (Fig. 5A, see Supplemental Data Fig. 2). Previous studies, based on regular but discontinuous behavioral monitoring of convulsive motor seizures, suggested that seizure frequency generally increases in this model for several months, but then reaches a steady state or plateau (Hellier et al., 1998). Although a sigmoid curve best fit the average increase in electrographically recorded seizure frequency after kainate treatment (i.e., group data, Fig. 5A), similar plots of seizure frequency versus time were more variable for individual animals (Fig. 5B and 5C). A Boltzmann sigmoid curve modeled the time-dependent increase in seizure frequency in 6 of 9 rats by 100 days after kainate-induced status epilepticus, and extended to 114 days in 1 rat (see Supplemental Data Fig. 2). The equation describing this relationship had the form
where Szrate was the EEG seizure rate at time t, Inrate was the initial EEG seizure rate, Finrate was the EEG seizure rate at the plateau phase, 1/2max was the time at which the EEG seizure rate reached half-maximum and slope was the maximum slope of the curve. An exponential growth curve provided the best fit to the data in the other 3 rats (i.e., panels A, B, and H in Fig. 2 of the Supplementary Data); however, a sigmoid growth curve could also be well fit to these data if one assumed that the EEG seizure frequency eventually reached a plateau, which was considered a reasonable assumption given previous data (see Fig. 2 of Hellier et al., 1998) and the view that an exponential growth in EEG seizure frequency without some type of plateau phase seems untenable (in some animals, the end of the exponential growth phase was equivalent to 12–24 h periods of status epilepticus). The ability to fit the progression of EEG seizures over time to a sigmoid curve allowed us to measure parameters of EEG seizure progression (i.e., the slope factor and the time required to reach the half-maximum of the sigmoid curve). A correlation analysis between the number of days to the first detected EEG seizure and the first convulsive motor seizure (i.e., measures of latent period) and the two measures of EEG seizure progression were performed on those animals that could be best fit by a sigmoid curve. The EEG latent period did not significantly correlate with either the half-maximum (r2=0.73, p=0.09, n=6) or the slope (r2=0.17, p=0.4, n=6). The latent period to the first motor seizure also did not correlate with the half-maximum time (r2=−0.12, p=0.8, n=6), although the latent period to the first convulsive motor seizure did have a significant negative correlation with the slope (r2=0.72, p=0.03, n=6). Furthermore, correlations between the amount of time in status epilepticus and either the half-maximum or the slope of EEG seizure progression (r2=0.23, p=0.4, r2=0.26, p=0.37, respectively), and between the total number of convulsive motor seizures during kainate treatment and either the half-maximum or the slope of EEG seizure progression were not significant (r2=0.47, p=0.3, r2=0.55, p=0.25, respectively). Therefore, the key findings were that (1) non-convulsive electrographic seizures usually began after a relatively short latent period (about 1 week), (2) convulsive motor seizures generally began to occur after the appearance of non-convulsive EEG seizures, (3) spontaneous recurrent seizures continuously increased in frequency with time after the latent period, (4) nearly all animals experienced an exponential growth in EEG seizure frequency at some point after kainate treatment, and (5) the latent periods (i.e., to non-convulsive electrographic and convulsive motor seizures) did not correlate well with measures of seizure progression.
Because a Boltzmann sigmoid curve could model the group mean increase in electrographically recorded seizure frequency (i.e., both non-convulsive and convulsive seizures) for the population of kainate-treated rats (Fig. 5A, F test for goodness-of-fit, p=0.017, n=9), four stages of EEG seizure progression were apparent. The first distinguishable stage of EEG seizure progression was the latent period, during which no seizures were detected. Although the occurrence of the first clinical seizure is usually considered to be the end of the latent period, the observation of the first electrographically recorded seizure was viewed as the end of the first stage, or seizure-free time. The second stage began with a slow growth phase, which was characterized by a low but gradually increasing EEG seizure rate with highly variable inter-seizure intervals (Fig. 3B, ,4B4B and Fig. 5B,C). In this second phase, the longest inter-seizure intervals were 4–8 days (see above) and nearly as long as the latent periods to the first EEG seizure. Although this phase had a positive slope, the increase in seizure frequency in some animals was barely detectable for as much as 2 months (Fig. 5B, 5C; see also Supplementary Data, Fig 2). This result suggests that some of the rats with kainate-induced epilepsy would have been described as “non-progressive” (e.g., see Gorter et al., 2001), if they had only been subjected to 2 months of continuous monitoring, and if quantitative measures of seizure frequency had not been used. The end of the slow growth phase for each rat was arbitrarily defined as the first day in which the daily seizure rate was ≥0.2 electrographically recorded seizures per hour or 5 EEG seizures per day. The mean number of days (for all kainate-treated rats) after treatment to obtain this EEG seizure rate was 49 ± 8.5 (Fig. 5A). Stage 3 of EEG seizure progression consisted of an exponential growth phase in which the number of daily EEG seizures rapidly increased (Fig. 5A). The end of this stage was mathematically defined as the time in which the fitted sigmoid curve attained 95% of its maximal value. The mean time to the end of the exponential growth phase was 79 ± 10 days (Fig. 5A). Stage 4 was termed the plateau phase, in which the EEG seizure rate remained relatively stable (Fig. 5A) with a mean EEG seizure rate over this time period of 0.87 ± 0.02 EEG seizures per hour (i.e., approximately 21 seizures per day) and a short mean inter-seizure interval. These data show that the progression of the daily EEG seizure rate for individual rats was variable, but ultimately all animals experienced a continuous exponential rise in EEG seizure frequency. These data also indicate that the EEG latent period may best be viewed as the first of several long inter-seizure intervals, which eventually become shorter as the EEG seizure frequency continuously increases with time.
Clustering of seizures in this model and the pilocarpine model of status epilepticus has previously been observed (e.g., Grabenstatter et al., 2006; Goffin et al., 2007), but not quantified over long periods. Seizure clusters were either readily observable as distinct entities, or could be seen as slow increases and decreases in seizure frequency (i.e., they could appear as a “waxing-and-waning” of the seizures). During the first month after kainate treatment, EEG seizures appeared to occur in multiples on the same day (i.e., short inter-seizure intervals) and often within a close time frame of each other (see Fig. 3B, 3C1 and Fig. 4B). A Poisson distribution describes the probability of events (e.g., EEG seizures) occurring over a fixed period of time (e.g., 100 days) if these events occur at a known average rate (i.e., mean EEG seizure frequency) and are independent of the time since the last event (i.e., random or non-clustering). Thus, EEG seizures that occur randomly should have a distribution of inter-seizure intervals that are fit by a Poisson distribution (Binnie et al., 1984), and the actual distribution of inter-seizure intervals should not be significantly different from randomly generated intervals with identical mean values. Clustering of the EEG seizures (i.e., non-random occurrence with inter-seizure intervals dependent on each other) would make the distribution of the inter-seizure intervals differ significantly from a Poisson distribution and randomly generated intervals with identical means. Statistical analyses of inter-event intervals for detection of the presence of clusters have been referred to as tests of the uniformity hypothesis and the random-position hypothesis (Bock, 1995; Jain and Dubes, 1988, respectively). These analyses rely upon statistical differences from a Poisson distribution and from matched, randomly generated data. In order to test the hypothesis that EEG seizures were occurring randomly, a Poisson function was generated for each stage, and random intervals were generated around the mean inter-seizure interval for each individual animal during each stage of EEG seizure progression (see Methods). The Kolomogorov-Smirnov test was used to test for differences in the cumulative distribution function between the actual inter-seizure intervals, the Poisson distribution, and the randomly generated intervals. In each of the different stages of EEG seizure progression, the inter-seizure intervals were significantly different from the Poisson distribution (stage 2, p<0.0001, n=9 [see Fig. 6A]; stage 3, p<0.0001, n=9; stage 4, p<0.0001, n=6), and from randomly generated intervals (stage 2, p<0.002, n=9; stage 3, p<0.0001, n=9; stage 4, p<0.0001, n=6). Figure 6A illustrates the distribution of inter-seizure intervals during the slow growth phase on two scales of the ordinate (A1 = 0–160 intervals, and A2 = 0–15 intervals); Figure 6A1 shows that >160 inter-seizure intervals (i.e., the first bin) during the slow growth phase were ≤250 min (or about 4 h; i.e., the bin duration), while Figure 6A2 shows a peak to the right where >25 intervals were between about 2,000 and 6,000 min (or roughly 1–4 days). Therefore, this analysis of the distribution of inter-seizure intervals during the slow growth phase suggests that many of the inter-seizure intervals during a cluster were <4 h and many of the inter-cluster intervals were about 1–4 days. Qualitatively similar relationships for the inter-seizure intervals were present during stages 3 and 4 (see Fig. 3C2), but the family of intervals showed a distinct shortening that corresponded to the overall higher seizure frequency. The distributions for the plateau phase used only 6 of the 9 rats, because 3 animals did not achieve a plateau phase prior to termination of the recording. These data reject the hypothesis that the electrographically recorded seizures were simply occurring in a random manner, and instead, support the hypothesis that the seizures occur in clusters and that an increase in EEG-seizure clustering may contribute to the time-dependent increase in EEG seizure frequency. These data also indicate that EEG seizure clusters were a likely source of variability throughout all phases of seizure progression, contributing to the variable rates of seizure progression seen in individual animals in Figure 5B and 5C (i.e., the intra-animal variability).
As the EEG seizure rate progressively increased, EEG seizures tended to occur in multiple clusters with very short inter-seizure intervals during any given 24-h period (see Fig. 3B1-3, C2 vs C1).
One definition of a seizure cluster in humans is three or more motor seizures per day (Haut et al., 1999), and this measure applied to EEG seizures showed a clear increase over time after kainate-induced status epilepticus (data not shown); however, this measure did not seem appropriate in this model because of the relatively high overall EEG seizure rate, and no clear progression pattern in EEG seizure clusters was seen when this measure was normalized to the EEG seizure frequency (Fig. 6B1). Because many of the shortest inter-seizure intervals were measured to be roughly 1 h soon after the onset of spontaneous recurrent seizures (Fig. 4B), and because these 1-h inter-seizure intervals represented obvious seizure clusters when separated by inter-cluster intervals that could be as long as a few days, we determined the frequency of occurrence of inter-seizure intervals that were ≤1 h as a function of time after kainate-induced status epilepticus. When the number of ≤1-h inter-seizure intervals per day (i.e., defined as a seizure cluster) was normalized to the daily EEG seizure frequency in order to control for the effect of increasing EEG seizure frequency on the analysis (Fig. 6B2, n=9), the increase in seizure clusters over time closely followed the pattern of EEG seizure progression (Fig. 5A). A Boltzmann sigmoid curve (F-test, p=0.02) could be fit to this progressive increase in seizure clusters (i.e., inter-seizure intervals of ≤1 h), and this function began a steep increase between weeks 5 and 7 after kainate treatment, which coincided with the beginning of the exponential growth phase (i.e., stage 3). Thus, this analysis of ≤1-h inter-seizure intervals, as a measure of seizure clusters, showed that the exponential increase in EEG seizure frequency occurred concurrently with an enhanced likelihood of electrographically recorded seizure clusters.
The key results of this study are: non-convulsive electrographic seizures almost always preceded the first convulsive motor (i.e., clinical) seizure (Bertram and Cornett, 1993, 1994); the longest inter-seizure interval after the first clinical seizure was nearly as long as the latent period to the first electrographically recorded seizure (Fig. 4A); in the initial period after the onset of spontaneous recurrent seizures, the inter-seizure intervals were highly variable (i.e., spanned two log units; Fig. 4B); individual and group data indicated that a Boltzman sigmoid function best fit the data on seizure frequency versus time after status epilepticus (Fig. 5); plots of the number of intervals as a function of the duration of the inter-seizure interval did not fit a Poisson distribution (i.e., were non-random), and had two peaks during the early, slow growth phase of epileptogenesis (Fig. 6A); and, a graph of the number of inter-seizure intervals that were ≤1 h (normalized for seizure frequency) increased as a function of time after kainate-induced status epilepticus (Fig. 6B). These data support the hypothesis that the early phase of epileptogenesis is a smooth, continuous process whereby seizure frequency (and thus, seizure probability) slowly increases with time, but seizure clusters can obscure the progression. The data do not support the hypothesis that seizure probability is a step function of time after a brain insult (see Fig 1).
These data offer a new perspective of the latent period. Backward extrapolation from the slow growth phase of epileptogenesis (i.e., the second stage) to the latent period (i.e., the first stage) suggests that the latent period is a time of slowly but continuously increasing seizure probability (Fig 5A). Because our data suggest that seizure frequency/probability gradually increases after a brain insult, the first non-convulsive EEG seizure and the first convulsive motor seizure are likely probabilistic events. That the EEG-seizure latent periods appear only slightly longer than the longest inter-seizure intervals during the early growth phase (Fig. 4A) further supports this hypothesis. Thus, the latent period for the first EEG-recorded seizure may be viewed as the longest of the electrographically recorded inter-seizure intervals that occur early in the process of epileptogenesis, where all of the longer inter-seizure intervals (i.e., not the short intervals that comprise seizure clusters, see below) represent a time-dependent continuously-increasing probability of the occurrence of seizures after status epilepticus. This result leads to the hypothesis that the rate of increasing seizure probability determines the duration of the latent period; and thus, the latent period may be viewed hypothetically as the time point when the increasing seizure probability asymptotically departs from a steady baseline of low seizure probability. In conclusion, therefore, the latent period may poorly reflect the epileptogenic increase in seizure probability as a function of time, and may not be an appropriate measure of epileptogenesis, because: (1) theoretically, an asymptotic interception is difficult to measure accurately, and (2) practically, measurement requires continuous electrographic recording until the first seizure.
At least two key results from previous work suggest that a simple step function is inadequate to describe acquired epileptogenesis: (1) non-convulsive electrographic seizures (often several) generally precede convulsive motor seizures (e.g., Bertram and Cornett, 1993, 1994), and (2) progressive increases in the frequency of spontaneous recurrent seizures can often be detected after status epilepticus (e.g., Bertram and Cornett, 1993, 1994; Hellier et al., 1998; Gorter et al., 2001). Because of the quantitative analyses of the longest inter-seizure intervals after the latent period (Fig. 4) and the plots of seizure frequency versus time (Fig. 5) from continuous recording of the spontaneous seizures, our data emphasize that the transition from the latent period to the slow growth phase is best modeled as a smooth function. This hypothetical reasoning further argues that the presence of seizure clusters (see below) obscures this smooth transition from the latent period to the slow growth phase, and seizure clusters can also obscure the exponential growth phase (see below). Bertram and Cornett (1994) also noted that the increase in seizure frequency could be quite variable, and our data suggest that an important source of variability is seizure clusters (see below). Gorter et al. (2001) emphasized the presence of progressive and non-progressive animals; although all of the kainate-treated rats in the present study ultimately showed a progressive increase in seizure frequency, some animals spent prolonged periods (roughly 2 months) with a relatively steady seizure frequency (i.e., appeared “non-progressive”; see Fig. 5B and C; and Fig. 2 of the Supplemental Data), and then later underwent a profound increase in seizure frequency over a few weeks. Bertram and Cornett (1994) found that seizure frequency ultimately reached a plateau; a distinct plateau phase was not always seen in the present data, presumably because the recordings were conducted for only 100 days, and previous data using discontinuous behavioral monitoring suggest that more time after status epileticus is required for the plateau (Hellier et al., 1998). Therefore, a Boltzman sigmoid curve with a slow early increase in seizure frequency followed by an exponential growth phase and a plateau phase overall provided the “best fit” as a model of the observed changes in seizure frequency over time. Two parameters characterize the sigmoid function of seizure progression: (1) the time to reach 50% of the maximum seizure frequency, and (2) the slope factor, which measures the maximum increase in seizure rate. These parameters provide relatively straightforward measurements of the time course of seizure progression, from the latent period to the time of maximal maintained seizure frequency. Further, these measures can be obtained using short infrequent observations; if the plateau seizure frequency is also measured, the Boltzman function can be fit and used to provide a description of the initial time course, which removes the need for continuous EEG analysis of the period of very low seizure probability. Although the data presented in this study do not support the hypothesis that the latent periods for either the non-convulsive EEG seizures or the convulsive motor seizures are directly related to the progressive increase in seizure frequency, this may reflect the inherent conceptual and technical problems with the latent period described above.
The evidence for seizure clusters comes from several parts of the data set. The early inter-seizure intervals were both extremely long and quite short (Fig. 4B), and appeared to represent the early onset of seizure clustering, which was better documented by the observation of a non-Poisson (i.e., non-random) distribution of inter-seizure intervals (Fig. 6A). Individual variability can obviously confound analyses of seizure progression, and can prevent statistical analyses that assume randomness (e.g., a normal distribution). Bertram and Cornett (1994) noted that seizure rates can be quite variable not only across animals, but also through time in the same animal; it is unclear how much of the variability is due to seizure clusters. This variability can confound analyses of the effects of hypothetical therapeutic interventions. Whether seizure clustering contributed mechanistically to the exponential increase in seizure frequency, or was an unrelated consequence of the mechanisms of epileptogenesis, is also unclear.
Whether the progressive increase in seizure rate is common to all types and models of acquired epileptogenesis (including humans), or may simply reflect a particular animal model, also requires further investigation. Some studies in models of TLE have not observed progressive increases in seizure rate (Gorter et al., 2001; Mazarati et al., 2002; Bragin et al., 2004), but our data strongly suggest that both continuous and prolonged seizure monitoring is required to detect the progressive increase in seizure frequency. A recent study using discontinuous monitoring on pilocarpine-treated rats (Sung et al., 2007) reported the presence of seizures during the first week after status epilepticus, suggesting the lack of a latent period in this model. Our own recent observations suggest that the electrographic status epilepticus after pilocarpine treatment is far more intense than with the repeated, low-dose kainate model (Dudek et al., 2002, 2005), and the pilocarpine model may be more prone to having subsequent seizures in the few days after intense status epilepticus (e.g., see Fig. 3, particularly panel B, in Hellier et al., 1999). Thus, additional work is required to determine the generality of the concepts described here.
These data suggest several possible conclusions concerning acquired epileptogenesis: (1) seizure probability increases continuously after a brain insult until a steady state is achieved; (2) the latent period does not define the duration of epileptogenesis, and is probably a poor measure of it; (3) seizure clusters can obscure a progressive increase in seizure frequency, and yet they may contribute to increases in seizure frequency; and, (4) the fitting of seizure rates after a brain insult with a sigmoid curve may be useful for future anti-epileptogenesis studies. Why some animals have a non-progressive period, why clusters occur, and why seizure frequency reaches a plateau remains unclear. Understanding the mechanisms that contribute to the slow continuous build-up in spontaneous recurrent seizures and the clustering of seizures may provide important therapeutic insights into potentially altering the outcome of patients with acquired epilepsy.
Histological section of the hippocampus from a rat with kainate-induced epilepsy that was implanted with recording electrodes. The section shows a linear glial scar (two arrows) extending through the molecular layer (ML) of the dentate gyrus to the surface of the granule cell layer (GCL). This glial scar was secondary to the placement of a recording electrode into the dentate gyrus and demonstrates the accurate placement of the recording electrode. The scale bar shows 200 µm.
Individual examples of the progressive increase in the electrographic seizure rate as a function of time after status epilepticus for the 9 kainate-treated rats. In nearly every case, the EEG seizure latent period (ESL; i.e., the latent period for the first non-convulsive electrographically recorded seizure) was shorter than the motor-seizure latent period (MSL; i.e., the latent period for the first convulsive motor seizure). Values are the mean seizure frequency over a 1- or 2-week period, plus the standard error of the mean. The curve through the mean values in each panel shows the best fit, and is usually a sigmoid function, or in some cases (e.g., B and H), an exponential function (i.e., the lower part of a sigmoid curve). The data indicate that all of the kainate-treated rats showed a progressive increase in seizure rate with time, and that all kainate-treated rats experienced an exponential increase in seizure rate. In most animals, a sigmoid curve best described the increase in seizure rate.
Supported by NS045144 from the NIH.