A decrease of ripples precedes seizure onset in mesial temporal lobe epilepsy

Full text

Research Paper
A decrease of ripples precedes seizure onset in mesial temporal
lobe epilepsy☆
Cenglin Xu
a,1
, Shuang Wang
b,c,1
,YiWang
a,1
,KangLin
d
,GangPan
d
, Zhenghao Xu
e
, Jorge Gonzalez-Martinez
c
,
Feng Gao
b
, Xiaohua Wu
b
, Shihong Zhang
a
, Juan C. Bulacio
c
,ImadM.Najm
c
, Jianhong Luo
a
,WeiweiHu
a
,
Zhaohui Wu
d
,NormanK.So
c
, Zhong Chen
a,b,
⁎
a
Department of Pharmacology, Key Laboratory of Medical Neurobiology of the Ministry of Health of China, School of Medicine, Zhejiang University, Hangzhou, China
b
Epilepsy Center, Second Affiliated Hospital, School of Medicine, Zhejiang University, Hangzhou, China
c
Epilepsy Center, Neurological Institute, Cleveland Clinic, Cleveland, OH, USA
d
Department of Computer Science, Zhejiang University, Hangzhou, Zhejiang, China
e
Laboratory of Brain Functionand Disease in Chinese Medicine, College of Basic Medical Science, Zhejiang Chinese Medical University, Hangzhou, Zhejiang, China
abstractarticle info
Article history:
Received 25 February 2016
Received in revised form 21 July 2016
Accepted 21 July 2016
Available online 22 July 2016
High-frequency oscillations (HFOs) are promising biomarkers for epileptic foci; however, their characteristic
changes during the preictal period remain unclear. Here, the preictal HFOs were recorded and detected by an au-
tomated HFOsdetection method in the mousepilocarpine model as well as in patients withmesial temporal lobe
epilepsy (mTLE) and neocortical epilepsy. A total of sixteen low-voltage fast (LVF) and fifty-three
hypersynchronous-onset (HYP) seizures were recorded in ten mice. The rate of ripples (80–250 Hz) decreased
during 1 min before the onset of LVF and HYP seizures, which was primarily due to the reduction of type II (in-
dependent of epileptiform discharges) rather than type I ripples (superimposed on epileptiform activities). The
ripple rate decreased until 30 s before HYP seizure, whereas it increased with a peak at 40 s during the 1 min
preictal period of LVF seizures. Furthermore, the “ripple reduction”phenomenon was also observed in all twelve
seizures from nine patients with mTLE but not in neocortical epilepsy. These results indicate that ripples maypo-
tentially be helpful for understanding the mechanisms of ictogenesis in mTLE, and the different modes of ripple
changesduring the minute before LVF and HYPseizures might alsobe beneficial for the diagnosis of seizure types.
© 2016 Elsevier Inc. All rights reserved.
Keywords:
High frequency oscillations
Mesial temporal lobe epilepsy
Preictal
1. Introduction
Patientswith mesial temporal lobe epilepsy (mTLE) have a high risk
of developing pharmacoresistance. The unpredictable nature of seizure
attacks will strongly impair patients' health and social functioning, im-
posing a major burden on their families. Finding specific ictal bio-
markers is necessary and important for patients' safety. Furthermore,
early and reliable ictal biomarkers may be necessary for the timely de-
livery of neuromodulation in a closed-loop manner. It has recently
been reported that using optogenetics with closed-loop seizure detec-
tion for real-time, spatially-restricted therapeutic intervention can
control spontaneous seizures in TLE only when it is delivered immedi-
ately following seizure onset (Krook-Magnuson et al., 2013). Also, we
previously reported that the efficacy of low-frequency stimulation
(LFS) for seizure control is more prominent in the initial few seconds
(0–4 s) following seizure onset while delayed delivery of LFS is less ef-
fective or even aggravates seizures (Wang et al., 2008; Wu et al.,
2008; Xu et al., 2010).
High-frequency oscillations (HFOs, 80–700 Hz) are a promising EEG
biomarker of epilepsy. Accumulating evidence has shown that inter-
ictal HFOs, especially fast ripples (250–700 Hz), are closely associated
with the seizure onset zone (Bragin et al., 2002; Jacobs et al., 2009a;
Jirsch et al., 2006; Wang et al., 2012;Worrell et al., 2004) and the epilep-
togenic zone (Akiyama et al., 2011; Fujiwara et al., 2012; Jacobs et al.,
2010). Recently, an increase of HFOs following seizure onset in vivo
and in vitro has been reported (Khosravani et al., 2005; Levesque et al.,
2012; Timofeev and Steriade, 2004), suggesting HFOs are also associat-
ed with ictogenesis. Therefore, analyzing the dynamic changes of
preictal HFOs may not only help predict seizures, but also shed new
light on the mechanisms of ictogenesis. So far preictal changes of HFOs
have been described in a few reports, but contradictory conclusions
are drawn from the results of Khosravani (an increase in high-frequency
Experimental Neurology 284 (2016) 29–37
Abbreviations: mTLE, mesial temporal lobe epilepsy; HFOs, high frequency
oscillations; LVF, low-voltage fast; HYP, hypersynchronous; ANOVA, one-way analysis of
variance; SE, st atus epilepticus; SOZ, seizur e onset zone; IISs , interictal spikes; PP,
primary propagation.
☆The authors report no conflict of interest.
⁎Corresponding author at: Department of Pharmacology, Key Laboratory of Medical
Neurobiology of the Ministry of Health of China, School of Medicine, Zhejiang University,
Hangzhou 310058, China.
E-mail address: chenzhong@zju.edu.cn (Z. Chen).
1
These authors contributed equally to the paper.
http://dx.doi.org/10.1016/j.expneurol.2016.07.015
0014-4886/© 2016 Elsevier Inc. All rights reserved.
Contents lists available at ScienceDirect
Experimental Neurology
journal homepage: www.elsevier.com/locate/yexnr

band powerat 8 s preictally) and Jacobs (preictal HFOs showing irregu-
lar trends) (Jacobs et al., 2009b; Khosravani etal., 2009). This contradic-
tion may be due to the differences in epilepsy types enrolled in their
studies. Moreover, Levesque et al. have recently reported that ripples
predominate in low-voltage fast (LVF) onset seizures and fast ripples
predominate in hypersynchronous-onset (HYP) seizures in animal
models of TLE (Levesque et al., 2012), which suggests different subtypes
of HFOs change in specific ways in different types of epilepsy. Thus,
preictal changes in different types of HFOs need to be studied carefully
in different forms of epilepsy. Here, we studied the dynamic changes
of HFOs in preictal periods in a classical mouse pilocarpine model of
mTLE (Curia et al., 2008) as well as in patients with mTLE who
underwent invasive EEG recording. We also compared the findings
with those in patients with neocortical epilepsy.
2. Materials and methods
2.1. Animals and surgery
Male ICR mice (30–35 g, Grade II, Experimental Animal Center, Zhe-
jiang Academy of Medical Science, China) were intraperitoneally
injected with scopolamine methylnitrate (1 mg/kg, Sigma) followed
30 min later by pilocarpine hydrochloride (200 mg/kg, i.p., Sigma). Sta-
tus epilepticus (SE) was terminated after 1 h of continuous seizures by
diazepam (1 mg/kg, i.p.). After 15 days of recovery, the surviving ani-
mals were anesthetized with pentobarbital (50 mg/kg) and mounted
in a stereotaxic apparatus (SR-5N, Narishige, Japan). Bipolar electrodes
made of stainless-steel Teflon-coated wires (791,500, A.M. Systems,
USA; diameter 0.125 mm; distance between exposed tips, 0.5 mm)
were implanted into the unilateral CA3 subfield of hippocampus
(AP,−2.8; ML,−3.0; DV,−3.5) and entorhinal cortex (AP: −4.6, ML:
−3.0, V: −3.7). The reference and ground screws were placed in the
bone over the cerebellum. Then the recording and reference electrodes
were welded to a receptacle, whichconnected the electrodes to record-
ing wires for EEG recording. All experiments were carried out in accor-
dance with the ethical guidelines of the Zhejiang University Animal
Experimentation Committee and were in complete compliance with
the National Institutes of Health Guide for the Care and Use of Laborato-
ry Animals.
2.2. EEG recording and data acquisition in mice
Mice surviving three week post-SE were h oused individually in Plex-
iglas boxes (30 × 30 × 40 cm) under controlled conditions (12 h light/
dark schedule). Food and water were provided ad libitum. Mice were
habituated for 24 h before EEG recording by a PowerLab system (AD In-
struments, USA) with sampling rate of 4 kHz and bandwidth of 0.5–
1000 Hz. And a montage was used to diminish those artifacts recorded
from two electrode tips. All mice were recorded 24 h/day for 10 days.
2.3. Intracranial EEG recording in patients
The patients' EEG data were collected in the Epilepsy Center, Second
Affiliated Hospital, School of Medicine, Zhejiang University, and the
Cleveland Clinic Epilepsy Center (study number 10–171). The patients
with medically intractable epilepsy (Supplementary Table 1)
underwent invasive EEG evaluation by either stereotactic EEG or a sub-
dural grid and supplementary depth electrodes. The depth electrodes
were composed of cylindrical platinum electrodes which had several
contacts, ranging from 8 to 16. In addition, the contacts were 2.5 mm
long, 0.8–1.1 mm in diameter, and 4–5 mm apart center-to-center.
Each subdural contact had a diameter of 4 mm with a center-to-center
separation of 10 mm. A three-dimensional reconstruction map was
made to show the locations of the electrode contacts after implantation.
Intracranial EEG was analyzed by at least two epileptologists. The areas
showing the earliest intracranial EEG change in a seizure, such as the
onset of repetitive spikes, background suppression, or paroxysmal fast
activity, were defined as the seizure-onset zone (SOZ).Each selected pa-
tient had a confined SOZ for all recorded seizures either in mesial tem-
poral structures or in neocortical area. All seizures were recorded
routinely on reduced or no anti-epileptic medication at a sampling
rate of 2000 Hz (bandwidth, 0.016–2000 Hz).
2.4. Distinction of electrographic seizure and seizure onset
Based on Bragin et al. (Bragin et al., 2005), an electrographic seizure
was defined by a period of consistent, repetitive changes in the ampli-
tude and frequency of electricalactivity, and thisactivity was clearly dis-
tinct from inter-ictal activity. To exclude the interference from other
neighboring seizures, only EEG seizures that were longer than 30 s
and were seizure-free for at least 1 h before and after the seizure
event were selected. Seizure onset was determined visually by the
first sign of persistent changes distinct from the inter-ictal background.
A consensus was made by two reviewers (C.L.X and Y·W).
2.5. Automated detection of HFOs with further visual confirmation
Type I ripple (superimposed on an inter-ictal spike or fast activity)
and fast ripple were automatically detected according to Dümpelmann
et al. (Dumpelmann et al., 2012). Briefly, EEG data with manually la-
beled type I ripple, type II ripple and fast ripple were imputed and fil-
tered with a FIR equirriple filter at 80–700 Hz, and then the signals
were divided into 10 ms-long subsets. Three features were derived
and further labeled each subset. The features one and two reflect the in-
creased signal amplitude during an HFO, in which the first feature was
based on the increased signal energy of the signal during an HFO by es-
timating the short time energy; the second feature being sensitive for
the increased signal amplitude was the short time line length. The
third feature reflects the lower and more stable signal frequency of
the signal during an HFO. Then subsets of features were selected as
input vector for SVM for training, and the final model could detect all
the manually labeled type I ripple and fast ripple, thus it could be used
for automated type I ripple and fast ripple detection with high precision.
Type II ripples (independent of epileptiform discharges) were more
likely to be linked with normal EEG activities, and the majority of la-
beled type II ripples couldn't be detected by this SVM classifier. It sug-
gested that SVM classifier was not suitable for type II ripple detection
due to its low precision. Therefore, we used SVM classifier merely for
type I and fast ripple detection. Anda few type II ripples which were de-
tected by SVM classifier were excluded.
For type II ripple only, we carried out another program based on the
principles for visual detection of type II ripple (Wang et al., 2012). First-
ly, EEG data were imputed with band pass filtered at 80–250 Hz. Sec-
ondly, signals containing at least five consecutive oscillations in the
filtered band were separated and selected. Thirdly, selected signals
were under further “denoising”,i.e., signals that had “false”oscillations
such as non-sinusoidal harmonics caused by filtering sharp spikes and
artifacts (Benar et al., 2010) were excluded, and fourth, selected signals
which did not have three times higher in amplitude compared with
background were excluded. All these above procedure were automati-
cally carried out by computer program. Then, the left detected signals
were extracted for further identification under manually time-frequen-
cy analysis, the time-frequency map of a true ripple event must show a
primary isolated peakin the frequency range of 80–250 Hz. At last, type
II ripples were selected based on their characteristics from these true
ripples andlabeled. All manually work was doneby two experienced re-
searchers and a consensus was made (C.L.X and S.W.). By using this
method, we could detect all manually labeled type II ripple. Schematics
of the automated detection method of type II ripple and representative
type I, type II and fast ripple were shown in Fig. 1.
Interictal spikes without HFOs are thought to be closely related with
epilepsy, and might be a different biomarker from HFOs (Levesque et al.,
30 C. Xu et al. / Experimental Neurology 284 (2016) 29–37

2011). So we also detected and analyzed interictal spikes without HFOs
in this study. Interictal spikes without HFOs were detected visually ac-
cording to the Levesque's (Levesque et al., 2011). Briefly, interictal
spikes without co-occurrence of HFOs in a 200 ms window before and
after themselves were selected due to the high threshold (5 times
higher) compared with background, and special attention should be
paid to exclude false positive events created by movement artifacts.
2.6. Quantitative analysis and statistical tests
First, we used 1 min as a time interval, and EEGs at 60th, 50th, 40th,
30th, 20th, 10th, 5th, 4th, 3rd, 2nd, and 1st min preictally were used to
analyze the rates of ripples, fast ripples, and inter-ictal spikes (IISs)
without HFOs in each mouse. The rates were then normalized by the
mean rates of each time interval. To further analyze the dynamic varia-
tion of HFOs within 1 min before seizure onset, we calculated the rates
of ripples and fast ripples during each 10 s duration and then normal-
ized them to the mean rates of these 6 parts. To detect the ripple varia-
tion in the interictal period, consecutive EEGs of the interictal period
were selected, similar to the analysis in the preictal period, 1 min time
interval was used, and the rates were normalized by the mean rates of
each timeinterval. To comparethe reduction of type I and type IIripples,
the reduction ratio was calculated by dividing the difference of rate in
preictal 2nd and 1st min by the rate during the 2nd min preictally.
EEGs at 5th, 4th, 3rd, 2nd and 1st min preictally were used to analyze
the rates of ripples and fast ripples in patients with mTLE. HFOs in the
SOZ and primary propagation area (PP area, the brain region next in-
volved in the seizure within 3 s after seizure onset) were included for
analysis. The rates of ripples and fast ripples during the 5 min period
preictally were normalized to mean value of rate in each time interval.
EEGs at 2 and 1 min preictally were used to analyze the rates of ripples
and fast ripples in patients with neocortical epilepsy, HFOs in the SOZ
were included for analysis. One-way analysis of variance (ANOVA)
followed by the Bonferroni correction was used to compare the rates
of HFOs and IISs at two time points in the preictal and interictal period
(such as rates of events in 2nd min and 1st min preictally). Student's
t-test was used to compare the reduction ratio of ripple subtypes, type
II/type I ratio in 2nd and 1st min preictally and absolute value of HFOs
in 1st min preictally between LVF and HYP seizures. All data are
shown as means ± SEM. Only pb0.05 was considered as a significant
difference.
3. Results
3.1. Preictal changes of HFOs in the mouse pilocarpine model
Sixty-nine electrographic seizures in ten mice were included, and
twenty-seven thousand eight hundred and sixty ripples, two thousand
five hundred and twenty fast ripples, and fourteen thousand and
twenty-one IISs without HFOs were identified in EEG segments lasting
Fig. 1. Schematic of automatedtype II ripple detection method and representative type I ripple, type II ripple,and fast ripple in the mouse pilocarpine model. (A) A selected type II ripple
must undergo four computerized and oneartificial steps. (B–D)shows the representative type I ripple(B), type II ripple (C) and fast ripple (D). Top panelof each shows the three typesof
HFO events. Type I ripples are usually superimposed on an inter-ictal spike or fast activity. Type II ripples are independent on epileptiform discharges. Fast ripples are usually combined
with epileptiform activity. Middle panels of each show the signals after ripple band and fast ripple band filtering. Bottom panels of each show the time-frequency analysis in which true
HFOs show isolated color peaks.
31C. Xu et al. / Experimental Neurology 284 (2016) 29–37

Fig. 2. Preictal changes of ripple rates, fast ripple rates, and rates of inter-ictal spikes without HFOs in CA3 of pilocarpine mice. (A) The distribution of ripples, fast ripples, and inter-ictal
spikes without HFOs. (B–D) Normalized rates of ripples (B), fast ripples (C), and inter-ictal spikes without HFOs (D) from 60 min to 1 min before seizure onset. The rates of ripples
(pb0.001) and inter- ictal spikes with out HFOs (pb0.01) showed significant de creases from 2nd min to 1st min before seizure onset, while th e rate of fast ripples showed no
significant change (pN0.05, One-way ANOVA followed by the Bonferroni correction was used). (E) and (F) Power spectrum density of the ripple band and fast ripple band from
30 min to 1 min before seizure onset. Neither showed any significant change (pN0.05, One-way ANOVA followed by the Bonferroni correction was used).
Fig. 3. Inter-ictal changes of ripplerates in pilocarpinemodel. (A) Shematicof interictal periodselection in the sameday with seizures.The periods whichhas seizure-free before and after
in at least 2 h were regarded as inter-ictalperiod. (B) Five minute consecutive inter-ictal EEG data of 10 mice were analyzed. And rate of ripples showednosignificant change (pN0.05,
One-wayANOVA followed by the Bonferroni correction was used).(C) Interictal periodin 1 day without seizures.(D) Ten minute consecutive inter-ictalEEG data of 9 mice were analyzed.
And rate of ripples showed no significant change (pN0.05, One-way ANOVA followed by the Bonferroni correction was used).
32 C. Xu et al. / Experimental Neurology 284 (2016) 29–37

60 min before seizure onset. Among them, 62.75% were ripples, 5.68%
were fast ripples, and 31.58% were IISs without HFOs (Fig. 2A). The nor-
malized rate of ripples showed no significant changes from 60th min to
2nd min prior to seizure onset (pN0.05, Fig. 2B). However, it decreased
from 1.01 ±0.04 to 0.67 ± 0.07 from 2nd min to 1st min before seizure
onset in CA3 (pb0.001, Fig. 2B). And we further compared the rate of
ripples in 1st min preictal with average rate of ripples in other preictal
time points. The rate of ripples in 1st min was significantly lower than
the average rate of ripples of other preictal time points (from 1.03 ±
0.007 to 0.6689 ± 0.06522, pb0.001, paired t-test, data not shown).
Such a reduction in ripple rate was observed in all sixty-nine seizures
from ten mice. But in the EC, the normalized rate of ripples did not
show any significant change from 2nd min to 1st min prior to 24 sei-
zures of 6 mice (pN0.05, Supplementary Fig. 1). No significant changes
Fig. 4. Changesof ripple rates and fast ripplerates within 1 min before LVF andHYP seizure onset in pilocarpine model.Representative LVF(A) and HYP (B) seizures recorded in the CA3
region of themouse pilocarpine model. Top panel: entire seizure EEGs and spectrums respectively.Middle and lower panels: extended EEGs of seizureonset pattern. From theEEGs, LVF
seizures were characterized by the occurrence of a positive- or negative-going spike that was followed by the appearance of low-amplitude, high-frequency activity, while HYP seizures
started with an increase in the frequency of IIS, and these high amplitude repetitive spikes will slowly transfer into seizure discharges. Asterisks indicate the start of seizure onset in the
EEG. (C) and (E): The ripple rate showed a increase tendency from 60 s to 40 s beforeLVF seizure onset, but decreased significantly in 10s before LVF seizure onset (pb0.01). While the
ripple rate significantly decreased from 30 to 10 s before HYP seizure onset (pb0.001). And showed an increase in ictal period (pb0.001, one-way ANOVA followed by the Bonferroni
correction was used). (D) and (F): the rate of fast ripples showed no statistical significant change before LVF and HYP seizures. But both showed an increase in ictal period (pb0.01
for LVF, pb0.001 for HYP, one-way ANOVA followed by the Bonferroni correction was used).
33C. Xu et al. / Experimental Neurology 284 (2016) 29–37

were observed in the normalized rate of fast ripples within 60 min be-
fore seizure onset at the time points tested (pN0.05, Fig. 2C). The nor-
malized rate of IISs without HFOs also decreased from 1.10 ± 0.08 to
0.57 ± 0.06 from 2nd min to 1st min before seizure onset (pb0.01,
Fig. 2D). The power spectrum density (PSD) of ripple and fast ripple
band showed no significant changes before seizure onset (pN0.05,
Fig. 2E and F). To test whether ripple reduction was specific in preictal
period, we further studied the dynamic change of ripple in the interictal
period, and found the normalized rate of ripples showed no significant
changes in the interictal period both in the days with (ranging from
1.03 ± 0.03 to 0.98 ± 0.05, pN0.05, One-way ANOVA, Fig. 3B) and with-
out seizures (ranging from 1.06 ± 0.05 to 0.94 ± 0.04, pN0.05, One-way
ANOVA, Fig. 3D). Furthermore,sixty-nine seizures in ten mice contained
sixteen LVF and fifty-three HYP seizures. Among them, five mice
showed only HYP seizures and the others had both. In this study, LVF
seizure was characterized by the occurrence of a positive- or negative-
going spike followed by the appearance of low-amplitude, high-fre-
quency activity, while HYP seizures started with an increase in the fre-
quency of IIS, and these high amplitude repetitive spikes will slowly
transfer into seizure discharges (Levesque et al., 2012; Velasco et al.,
2000). Representative EEGs and spectrums of LVF and HYP seizures
are shown in Fig. 4A and B, respectively. The ripple rate showed an in-
crease tendency from 60 s to 40 s before LVF seizure onset, but de-
creased from 1.13 ± 0.09 to 0.61 ± 0.09 during the period 40s to 10s
before LVF seizure onset (pb0.01; Fig. 4C). While in HYP seizures, the
ripple rate significantly decreased from 1.19 ± 0.09 to 0.56 ± 0.07 dur-
ing the period 30 s to 10 s before seizure onset (pb0.01, Fig. 4E). In con-
trast, no significant changes in the rate of fast ripples before either LVF
or HYP seizures were found (pN0.05, Fig. 4DandF).Theabsolute
rates of ripples and fast ripples showed no significant difference in the
preictal period between LVF and HYP seizures (pN0.05, Supplementary
Fig. 2).
Two subtypes of ripples were classified as described in our previous
study to analyze the characteristics of ripple reduction. The reduction
ratio of type II ripples (0.46 ± 0.04) was higher than type I (0.23 ±
0.02, pb0.001, Fig. 5A). The ratio of ripple rates (type II/type I) de-
creased from 0.40 ± 0.03 to 0.28 ± 0.04 during 2nd min to 1st min be-
fore seizure onset (pb0.05, Fig. 5B).
3.2. Preictal change of HFOs in patients with mTLE or neocortical epilepsy
To verify whether rate of ripple decreases in mTLE patients, we also
analyzedEEG data of SOZ and PP area in mTLE patients. A representative
seizure onset EEG from one patient with mTLE is shown in Fig. 6A. A
total of twelve seizures from nine patients with mTLE were included
in the study. The normalized ripple rate in the SOZ decreased from
1.15 ± 0.06 to 0.74 ± 0.08 from 2nd min to 1st min before seizure
onset (pb0.05, Fig. 6B). However, no significant changes could be
observed in the normalized rate of fast ripples within 5 min before sei-
zure onset among the time-points tested (pN0.05, Fig. 6C). The normal-
ized rate of IISs without HFOs also decreased from 1.32 ± 0.08 to 0.94 ±
0.05 from 2 min to 1 min before seizure onset (pb0.05, Supplementary
Fig. 3). Neither the rate of ripples nor of fast ripples in the PP area
showed any significant change preictally (pN0.05, Fig. 6D and E).
Also, we analyzed ripple classification in TLE patient. Consist with the
findings in animal model, the reduction ratio of type II ripples
(0.67 ± 0.13) was higher than type I (0.28 ± 0.07, pb0.05, Supplemen-
tary Fig. 4).
A total of nine seizures from six patients with neocortical epilepsy
were analyzed in this study. Fast ripples are infrequently detected in
adult neocortical epilepsy (Jacobs et al., 2008), thus only ripples were
counted and analyzed. Only one seizure had a decreased ripple rate in
the SOZ, while seven seizures had an increased ripple rate and one sei-
zure had no change during 2 min to 1 min preictally. The ripple rates
from 2 min to 1 min before 4 seizure onsets in one patient are shown
in Supplementary Fig. 5 A1–A4. The ripple rates from 2 min to 1 min
before 5 seizure onsets in 5 patients are shown in Supplementary
Fig. 5B–F.
4. Discussion
So far, the preictal change patterns of HFOs in TLEhave not been fully
interpreted. In the present study, we have provided the first evidence
that the ripple rate declines within 1 min before mTLE seizure onset in
the mouse pilocarpine model. This “ripple reduction”phenomenon
was further confirmed in mTLE patients but not in neocortical epilepsy
patients. It is indicated that the ripple rate specifically declined before
mTLE seizures. Besides, rate of interictal spikes without HFOs could
also decreased from 2 to 1 min preictally, and showed fluctuant changes
in other preictal period we observed. However, our results are not con-
sistent with Khosravani's finding of an increase in high-frequency band
power at 8 s prior to seizure onset in mTLE patients by calculating the
band power within a 5-s time window (Khosravani et al., 2009). Differ-
ent protocols, including time windows selected for analysis might con-
tribute to this discrepancy. Khosravani et al. studied 30s preceding
seizure onset in TLE patient, while in our study the time periods ana-
lyzed were longer (1 h preictally in pilocarpine mice, 5 min for TLE pa-
tients). In addition, because it is sometimes difficult to determine
exactly the time point of seizure onset, a longer time window, such as
1 min as we used, may minimize the influence of this confounding fac-
tor and ripple variations from other causes. Meanwhile, we also ana-
lyzed the power of the ripple and fast ripple bands 1 min preictally.
Unlike Khosravani's, no specific changes were found. This might be
due to different analysis methods (power spectrum density in ours
and power spectrum amplitude in Khosravani's). As many factors such
as the rate, duration, and amplitude of HFOs together determine the
Fig. 5. Preictal changes inripple subtypes inpilocarpine model.(A) Reduction ratioof type I and type II ripples. Type IIripples had a significant higher ratiothan type I (pb0.001, Student's
t-test was used). (B) Ratio of ripple rates (type II/type I) for 60-s interval in 0–1minand1–2 min before seizure onset. The ratio of ripple rates in 1–2minwassignificant higher than 0–
1 min before seizure onset (pb0.05, Student's t-test was used).
34 C. Xu et al. / Experimental Neurology 284 (2016) 29–37

band power, variations in rates seem more sensitive, which might cause
the disassociation of changes in the rate and band power. Besides, most
of pilocarpine seizures start at the hippocampus, where the recording
site was located in our study. And it has been also reported that pilocar-
pine seizures may start outside the hippocampus, and then propagate
there after onset (Levesque et al., 2011, Levesque et al., 2012). There-
fore, the observation of reduced ripples prior to seizure onset may re-
flect the global attenuation that is also present in lower frequency
bands. Although our findings show that “ripple reduction”had a high
sensitivity in preictal period, all the data we have so far may be not suf-
ficient for seizure prediction. Due to the fact that fully-automated HFOs
detector is still unavailable, we couldn't test our findings on unselected
continuous long-term EEG recordings. It is accepted that the value
of a seizure predictor rests as much on specificity (i.e. number of
false positives) as sensitivity (Mormann et al., 2007), while in our
study, we found that in interictal period, rate of ripple showed no
significant change. But due to the fact that interictal period is very
long, whether this “ripple reduction”is specificin2ndto1stmin
preictally still cannot be entirely confirmed. This limits the practicability
of “ripple reduction”for seizure prediction purpose to some extent. But
our results at least suggest that the decline of ripple activity might
be followed by the onset of a seizure sometimes, thus attention should
be paid when this occurs when using HFOs for seizure prediction
purpose.
However, our findings were interesting from the point of view of
ictogenesis mechanisms. We previously reported that ripple classifica-
tion is helpful in diagnosis of epilepsy such as SOZ localization (Wang
et al., 2012). Type I ripples, which are superimposed on an inter-ictal
spike or fast activity, are likely to be more pathological and SOZ specific;
whereas type II ripples, which are independent of inter-ictal spike or
fast activity and are not specific markers of SOZ, seem to be less patho-
logical and some might play a physiological role (Barth, 2003;
Matsumoto et al., 2013; Nagasawa et al., 2012), such as in memory con-
solidation in the hippocampus (Axmacher et al., 2008; Girardeau et al.,
2009). It has been reported in several studies that pathological ripples
might increase in ictogenesis process (Bragin et al., 2005; Khosravani
et al., 2005; Levesque et al., 2012), but few study has reported the role
of physiological ripple in this process, especially in the preictal period.
Fig. 6. Preictal changes of HFOs in mTLE patients. (A) Representative seizure onset EEG and spectrums in one patient. The arrow indicates seizure onset in the EEG; “*”indicates seizure
onset zone channel (SOZ); “#”indicates the primary propagation area channel (PP area). (B) and (C) Normalized rates of ripples and fast ripples in the SOZ, respectively. (D) and (E)
Normalized rates of ripples and fast ripples in the PP area, respectively. Ripple rate shows a significant decrease from 2 min to 1 min before seizure onset in the SOZ (pb0.01), but not
in the PP area. Fast ripples show no significant change before seizure onset in the SOZ and PP area (pN0.05, one-way ANOVA followed by the Bonferroni correction was used).
35C. Xu et al. / Experimental Neurology 284 (2016) 29–37

In this study, we are surprised to find that the rate of type II ripples de-
creased more prominently preceding seizure onset than type I ripples.
As selective elimination of physiological ripples in the hippocampus
can impair memory consolidation (Girardeau et al., 2009), we speculate
that the reduction of type II ripples may reflect the disruption of physi-
ological brain functions in the preictal-to-ictal transition state. Indeed,
symptoms such as headache and psychosis prior to seizure onset in
some patients are likely due to this disruption (Fanella et al., 2012;
Shukla et al., 2008). The hippocampal ripples are strongly associated
with the brain circuit activities. The diversity of interneurons, innervat-
ing distinct domains ofpyramidal cells, emerged to coordinate the activ-
ity of pyramidal cells in a temporally distinct manner, and contribute
differentially to ripple oscillations (Klausberger et al., 2003). It has
been reported that when a seizure occurs in rodent TLE model, hippo-
campal circuits activities might change, and the activities of those inter-
neurons in distinct classes may transform into a sequence of
homologous behavior preictally (Ewell et al., 2015; Grasse et al.,
2013). This interruption of physiological brain circuits and different in-
trinsic firing pattern in interneurons preictally might contribute to the
“ripple reduction”, which can further lead us to speculate that an imbal-
ance between inhibition and excitation caused by abnormal function of
interneurons triggers a seizure, and may cast new light on the study of
ictogenesis.
The majority of seizures in both patients and animal models of mTLE
are classified into two common types based on the seizure-onset pat-
tern, i.e., HYP and LVF (Bragin et al., 2005; Levesque et al., 2012;
Velasco et al., 2000). We found that both LVF and HYP showed similar
“ripple reduction”1 min preictally in mice. And interestingly, the pat-
tern of the ripple decrement differed between LVF and HYP seizures
(ripple activity decreased until 30 s before HYP seizure onset, and dur-
ing the 1 min preictal period of an LVF seizure it increased with a peak
at 40 s). Meanwhile, differences in the distribution of hippocampal atro-
phy and HFOvariation between LVF and HYP seizures have been report-
ed (Levesque et al., 2012; Ogren et al., 2009). Combining with the
previous findings, our results support the hypothesis that LVF and HYP
seizures have different mechanisms of ictogenesis. This short time win-
dow for analysis may help to distinguish the subtypes of mTLE seizures
through retrospective analysis and thereby help to select specific treat-
ments for certain seizure types in future.
To verify whether the preictal “ripple reduction”is unique to mTLE,
we analyzed the preictal changes of ripples in patients with neocortical
epilepsy and found that the majority of seizures showed increases
in ripple rate within 1 min preictally (seven out of nine seizures),
which was consistent with our previous findings in neocortical epilepsy
patients (Wang et al., 2012; Worrell et al., 2004). This discrepancy
might be due to the different mechanisms of seizure onset between
TLE and neocortical epilepsy. Meanwhile, Jacobs et al. have analyzed
the EEGs from both mTLE patients and cortical epilepsy patients,
and detected both preictal increases and decreases in HFO rates and
band power (Jacobs et al., 2009b). A possible explanation of their find-
ing is that different epilepsy types were included in their analysis. So,
our results suggest that ripple reduction within 1 min before seizure
onset might be more predominant in mTLE, and different types of
seizures should be taken into account when using HFOs for their
prediction.
5. Conclusion
In summary, a decrease in ripples occurs within 1 min before mTLE
seizure onset, both in the mouse pilocarpine model and in patients.
And type II ripples decreased more prominently in this process. Thus,
this study suggests that ripple alteration may be a biomarker for
ictogenesis.
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.expneurol.2016.07.015.
Conflict of interest
Nothing to report.
Acknowledgements
This work was supported by grants from the National Natural Sci-
ence Foundation of China (grant number 91332202, 81271435,
81273492, and 81271624); the China Science Fund for Creative Re-
search Groups (grant number 81221003); and the Program for Zhejiang
Leading Team of S&T Innovation (grant number 2011R50014).
References
Akiyama, T., McCoy, B., Go, C.Y., Ochi, A., Elliott, I.M., Akiyama,M., Donner, E.J., Weiss, S.K.,
Snead 3rd, O.C., Rutka, J.T.,Drake, J.M., Otsubo, H., 2011. Focal resection of fast ripples
on extraoperative intracranial EEG improves seizure outcome in pediatric epilepsy.
Epilepsia 52, 1802–1811.
Axmacher,N., Elger, C.E., Fell, J., 2008. Ripplesin the medial temporal lobe arerelevant for
human memory consolidation. Brain 131, 1806–1817.
Barth, D.S., 2003. Submillisecond synchronization of fast electrical oscillations in neocor-
tex. J. Neurosci. 23, 2502–2510.
Benar, C.G., Chauviere, L., Bartolomei, F., Wendling, F., 2010. Pitfalls of high-pass filtering
for detecting epileptic oscillations: a technical note on "false" ripples. Clin.
Neurophysiol. 1 21, 301–310.
Bragin, A., Azizyan, A., Almajano, J., Wilson, C.L., Engel Jr., J., 2005. Analysis of chronic sei-
zure onsets after intrahippocampal kain ic acid injection in freely moving rats.
Epilepsia 46, 1592–1598.
Bragin, A., Mody, I., Wilson, C.L., Engel Jr., J., 2002. Local generation of fast ripples in epi-
leptic brain. J. Neurosci. 22, 2012–2021.
Curia, G., Longo, D., Biagini, G., Jones, R.S., Avoli, M., 2008. The pilocarpine model of tem-
poral lobe epilepsy. J. Neurosci. Methods 172, 143–157.
Dumpelmann, M., Jacobs, J., Kerber, K., Schulze-Bonhage, A., 2012. Automatic 80–250 Hz
"ripple" high frequency oscillation detection in invasive subdural grid and strip re-
cordings in epilepsy by a radial basis function neural network. Clin. Neurophysiol.
123, 1721–1731.
Ewell, L.A., Liang, L., Armstrong, C., Soltesz, I., Leutgeb, S., Leutgeb, J.K., 2015. Brain state is
a major factor in preseizure hippocampal network activity and influences success of
seizure intervention. J. Neurosci. 35, 15635–15648.
Fanella, M., Fatto uch, J., Casciato, S., Lapenta, L. , Morano, A., Egeo, G., Manfredi, M.,
Prencipe, M., Giallonardo, A.T., Di Bonaventura, C., 2012. Ictal epileptic headache as
"subtle" symptom in generalized idiopathic epilepsy. Epilepsia 53, e67–e70.
Fujiwara, H., Greiner, H.M., Lee, K.H., Holland-Bouley, K.D., Seo, J.H., Arthur, T., Mangano,
F.T., Leach, J.L., Rose, D.F., 2012. Resection of ictal high-frequency oscillations leads
to favorable surgical outcome in pediatric epilepsy. Epilepsia 53, 1607–1617.
Girardeau, G., Benchenane, K., Wiener, S.I., Buzsaki, G., Zugaro, M.B., 2009. Selective sup-
pression of hipp ocampal ripples impairs spati al memory. Nat. Ne urosci. 12,
1222–1223.
Grasse, D.W., Karunakaran, S., Moxon, K.A., 2013. Neuronal synchrony and the transition
to spontaneous seizures. Exp. Neurol. 248, 72–84.
Jacobs, J., Levan, P., Chatillon, C.E., Olivier, A., Dubeau,F., Gotman, J., 2009a. High frequency
oscillations in intracranial EEGs mark epileptogenicity rather than lesion type. Brain
132, 1022–1037.
Jacobs, J., Zelmann, R.,Jirsch, J., Chander, R., Dubeau, C.E., Gotman, J., 2009b. High frequen-
cy oscillations (80–500 Hz) in the preictal period in patients with focal seizures.
Epilepsia 50, 1780–1792.
Jacobs, J., LeVan, P., Chander, R., Hall, J., Dubeau, F., Gotman, J., 2008. Interictal high-fre-
quency oscillations (80–500 Hz) are an indicator of seizure onset areas independent
of spikes in the human epileptic brain. Epilepsia 49, 1893–1907.
Jacobs, J., Zijlmans, M., Zelmann, R., Chatillon, C.E., Hall, J., Olivier, A., Dubeau, F., Gotman,
J., 2010. High-frequency electroencephalographic oscillations correlate with outcome
of epilepsy surgery. Ann. Neurol. 67, 209–220.
Jirsch, J.D., Urrestarazu, E., LeVan, P., Olivier, A., Dubeau, F., Gotman, J., 2006. High-fre-
quency oscillations during human focal seizures. Brain 129, 1593–1608.
Khosravani, H., Mehrotra, N., Rigby, M., Hader, W.J., Pinnegar, C.R., Pillay, N., Wiebe, S.,
Federico, P., 2009. Spatial localization and time-dependant changes of electrographic
high frequency oscillations in human temporal lobe epilepsy. Epilepsia 50, 605–616.
Khosravani, H., Pinnegar, C.R., Mitchell, J.R., Bardakjian, B.L., Federico, P., Carlen, P.L., 2005.
Increased high-frequency oscillations precede in vitro low-Mg seizures. Epilepsia 46,
1188–1197.
Klausberger, T., Magill, P.J., Marton, L.F., Roberts, J.D., Cobden, P.M., Buzsaki, G., Somogyi,
P., 2003. Brain-state- and cell-type-specificfiring of hippocampal interneurons in
vivo. Nature 421, 844–848.
Krook-Magnuson, E., Armstrong, C., Oijala, M., Soltesz, I., 2013. On-demand optogenetic
control of spontaneous seizures in temporal lobe epilepsy. Nat. Commun. 4, 1376.
Levesque, M., Bortel, A., Gotman, J. , Avoli, M., 2011. High-frequency (8 0–500 Hz)
oscillations and epileptogenesis in temporal lobe epilepsy. Neurobiol. Dis. 42,
231–241.
Levesque, M., Salami, P., Gotman, J., Avoli, M., 2012. Two seizure-onset types reveal spe-
cific patterns of high-frequency oscillations in a model of temporal lobe epilepsy.
J. Neurosci. 32, 13264–13272.
36 C. Xu et al. / Experimental Neurology 284 (2016) 29–37

Matsumoto, A., Brinkmann, B.H., Stead, S.M., Matsumoto, J., Kucewicz, M., Marsh, W.R.,
Meyer, F.B., Worrell, G.A., 2013. Pathological and Physiological High Frequency Oscil-
lations in Focal Human Epilepsy. J. Neurophysiol.
Mormann, F., Andrzejak, R.G., Elger, C.E., Lehnertz, K., 2007. Seizure prediction: the long
and winding road. Brain 130, 314–333.
Nagasawa, T., Juhasz, C., Rothermel, R., Hoechstetter, K., Sood, S., Asano, E., 2012. Sponta-
neous and visually driven high-frequency oscillations in the occipitalcortex: intracra-
nial recording in epileptic patients. Hum. Brain Mapp. 33, 569–583.
Ogren, J.A., Bragin, A., Wilson, C.L., Hoftman, G.D., Lin, J.J., Dutton, R.A., Fields, T.A., Toga,
A.W., Thompson, P.M., Engel Jr., J., Staba, R.J., 2009. Three-dimensional hip pocampal
atrophy maps distin guish two common temp oral lobe seizure-onset patterns.
Epilepsia 50, 13 61–1370.
Shukla, G., Singh, S., Goyal, V., Gaikwad, S., Srivastava, A., Bal, C.S., Gupta, A., Behari, M.,
2008. Prolonged preictal psychosis in refractory seizures: a report of three cases. Ep-
ilepsy Behav. 13, 252–255.
Timofeev, I., Steriade, M., 2004. Neocortical seizures: initiation, development and cessa-
tion. Neuroscience 123, 299–336.
Velasco, A.L., Wilson, C.L., Babb, T.L., Engel Jr., J., 2000. Functional and anatomic correlates
of two frequently observed temporal lobe seizure-onset patte rns. Neural Plast. 7,
49–63.
Wang, S., Wang, I.Z., Bulacio, J.C., Mosher, J.C., Gonzalez-Martinez, J., Alexopoulos, A.V.,
Najm, I.M., So, N.K., 2012. Ripple classification helps to localize the seizure-onset
zone in neocortical epilepsy. Epilepsia.
Wang, S., Wu, D.C., Ding, M.P., Li, Q., Zhuge, Z.B., Zhang, S.H., Chen, Z., 2008. Low-frequen-
cy stimulation of cerebellar fastigial nucleus inhibits amygdaloid kindling acquisition
in Sprague-Dawley rats. Neurobiol. Dis. 29, 52–58.
Worrell,G.A., Parish, L., Cranstoun, S.D., Jonas, R., Baltuch, G., Litt,B., 2004. High-frequency
oscillations and seizure generation in neocortical epilepsy. Brain 127, 1496–1506.
Wu, D.C., Xu, Z.H., Wang, S., Fang, Q., Hu, D.Q., Li, Q., Sun, H.L., Zhang, S.H., Chen, Z., 2008.
Time-dependent effect of low-frequency stimulation on amygdaloid-kindling sei-
zures in rats. Neurobiol. Dis. 31, 74–79.
Xu, Z.H., Wu,D.C., Fang, Q., Zhong, K., Wang, S., Sun,H.L., Zhang, S.H., Chen, Z., 2010. Ther-
apeutic time window of low-frequency stimulation at entorhinal cortex for amygda-
loid-kindling seizures in rats. Epilepsia 51, 1861–1864.
37C. Xu et al. / Experimental Neurology 284 (2016) 29–37