Access to this full-text is provided by Wiley.
Content available from CNS Neuroscience & Therapeutics
This content is subject to copyright. Terms and conditions apply.
CNS Neurosci Ther. 2020;26:1021–1030.
|
1021wileyonlinelibrary.com/journal/cns
Received: 15 February 2020
|
Revised: 29 April 2020
|
Accepted: 29 A pril 2020
DOI: 10.1111/cns.13394
ORIGINAL ARTICLE
CpG methylation signature defines human temporal lobe
epilepsy and predicts drug-resistant
Wenbiao Xiao1 | Chaorong Liu1 | Kuo Zhong2 | Shangwei Ning3 | Rui Hou4 |
Na Deng1 | Yuchen Xu1 | Zhaohui Luo1 | Yujiao Fu1 | Yi Zeng5 | Bo Xiao1 |
Hongyu Long1 | Lili Long1
This is an op en access arti cle under the ter ms of the Creative Commons Attribution L icense, which pe rmits use, dis tribu tion and reprod uction in any med ium,
provide d the original wor k is properly cited.
© 2020 The Authors. CNS Neuroscience & Therapeutics publishe d by John Wiley & Sons Ltd
Xiao an d Liu are co ntributed equ ally to this work .
1Depar tment of Neurology, Xiangya
Hospit al, Central South Universit y,
Changsha, China
2Tsinghua-Berkeley Shenzhen Institute,
Tsinghua Universit y, Shenzhen, China
3College of B ioinformatic s Science and
Technolog y, Harbin Me dical U niversity,
Harbin, China
4Shanghai Biotechnology Corporation,
Shanghai, China
5Depar tment of Geriat rics, Second X iangya
Hospit al, Central South Universit y,
Changsha, China
Correspondence
Hongy u Long an d Lili Long, Department of
Neurology, Xiangya Hos pital , Central South
University, Changsha, Hunan 410008, China.
Emails: longhongyu@ csu.edu.cn (HL);
10353654@qq.com (LL)
Funding information
Fundamental Res earch Funds for the Central
Universities of C entral South University,
Grant /Award Number: 2018zz ts248;
Clinical Resea rch Foundation of X iangya
Hospit al, Grant/Award Nu mber: 2016L08;
Hunan Provincial Science and Technology
Department, Grant/Award Number:
2019SK1012; National Natural Science
Foundat ion of China, Gra nt/Award Number:
81671299, 81671300, 81701182 and
81974206
Abstract
Aims: Temporal lobe epilepsy (TLE) is the most common focal epilepsy syndrome in
adults and frequently develops drug resistance. Studies have investigated the value
of peripheral DNA methylation signature as molecular biomarker for diagnosis or
prognosis. We aimed to explore methylation biomarkers for TLE diagnosis and phar-
macoresistance prediction.
Methods: We initially conducted genome-wide DNA methylation profiling in TLE pa-
tients, and then selected candidate CpGs in training cohort and validated in another
independent cohort by employing machine learning algorithms. Furthermore, nom-
ogram comprising DNA methylation and clinicopathological data was generated to
predict the drug response in the entire patient cohort. Lastly, bioinformatics analysis
for CpG-associated genes was performed using Ingenuity Pathway Analysis.
Results: After screening and validation, eight CpGs were identified for diagnostic
biomarker with an area under the curve (AUC) of 0.81 and six CpGs for drug-resistant
prediction biomarker with an AUC of 0.79. The nomogram for drug-resistant predic-
tion comprised methylation risk score, disease course, seizure frequency, and hip-
pocampal sclerosis, with AUC as high as 0.96. Bioinformatics analysis indicated drug
response–related CpGs corresponding genes closely related to DNA methylation.
Conclusions: This study demonstrates the ability to use peripheral DNA methylation
signature as molecular biomarker for epilepsy diagnosis and drug-resistant prediction.
KEY WORDS
biomarker, DNA methylation, machine learning, nomogram, temporal lobe epilepsy
1022
|
XIAO et Al.
1 | INTRODUCTION
Tempo ral lobe epilep sy (TL E) is the most comm on focal epi lep sy sy n-
drome in adults and frequently develops drug resistance,1 requiring
surgical treatment which offers a comparatively favorable progno-
sis.2,3 Moreover, cognitive impairment and psychiatric comorbidities
including depression and anxiety disorders, together with the long-
term actual seizures and accompanying drug usage, often result in
severe effects on the quality of life and individual health.4,5
At present, the diagnosis of epilepsy mainly depends on clini-
cal manifestation, neuroimaging, and electroencephalogram (EEG).
These methods are not only expensive and time-consuming, but also
require professional equipment and trained specialists that are not
accessible to many patients, which result in delayed diagnosis or mis-
diagnosis to some extent.6,7 Furthermore, drug-response prediction
is mainly based on subjective clinical features by experience and has
not come to a conclude.4,8,9 Earlier identification of drug-resistant
patients makes it possible to benefit from epilepsy surgery. Thus,
biomarkers for assisting the current diagnosis and predic ting the
treatment outcome are in urgent need. Preliminar y attempts have
been made in circulating molecules biomarkers of epilepsy, includ-
ing inflammatory cytokines, S100 calcium-binding protein B(S100B),
and matrix metallopeptidase 9(MMP9), and recently miRNA.1 0-1 4
However, the limitations of these studies mainly related to small
sample size and lack of validation, as well as heterogeneit y of epi-
lepsy that prevent the clinical value of these biomarkers.15
DNA methylation, the best-studied epigenetic mechanism, refers
to the covalent attachment of methyl groups to the cytosine resi-
dues (mainly confined in CpG sites) mediated by DNA methyltrans-
ferase (DNMT).16,17 It is mostly stab le throughout the genome and is
associated with transcriptional activation/repression.18, 19 Aberrant
DNA methylation implicated in underlying epileptogenesis and pro-
gression mechanisms of epilepsy has gained considerable attention.
Altered expression of DNMTs and methylation changes in individ-
ual candidate genes (ie, RELN) have been found in TLE patients.2 0-2 3
Several genome-wide studies using epileptic brain tissue have iden-
tified differential methylation events occurred in genes associated
with inflammation, neuronal development, etc24-27 Moreover, our
previous research reported that dysregulated methylation impli-
cated in both protein-encoding genes and noncoding RNA genes in
peripheral blood DNA from TLE patients.28,29
A substantial number of studies have investigated the value of
peripheral DNA methylation signature as molecular biomarker for
diagnosis or prognosis, especially in c ancer research.15,30-32 The
prognostic value of O6-methylguanine-DNA-methyltransferase
(MGMT) promoter methylation in glioblastoma and methylated
SEPTIN 9(SEPT9) in plasma for detection of asymptomatic colorec-
tal cancer is well-known paradigms,33-35 which have been included
in clinical guidelines and translated into the commercially available
clinical test.15 In addition, there was a trend that researchers fa-
vored combinatorial biomarkers of multiple CpG signature.36-39 DNA
methylation–based biomarkers present advantages with regard to
clinical application: presence in various biofluids, more stable than
other biological materials (such as RNA or protein), easy detection
by well-established methodologies, and cell-type specificity.15,31
However, to date, methylation biomarkers for TLE diagnosis and
pharmacoresist ance prediction have not been explored.
In this study, we aimed to identify and validate disease-related
and drug response–related CpGs in TLE. We initially conducted ge-
nome-wide DNA methylation profiling in TLE patient s; then, we se-
lected candidate CpGs in training cohort and validated those CpGs
in another independent cohort by employing machine learning al-
gorithms. Furthermore, a nomogram comprising DNA methylation
and clinicopathological data was generated to predict the drug re-
sponse in the entire patient cohort. Lastly, mechanistic links were
pursued for all biomarker CpGs corresponding genes by bioinfor-
matics analysis.
2 | MATERIALS AND METHODS
2.1 | Patient cohorts
The study was carried out on a cohor t of 78 patients with TLE and
78 sex- and age-matched healthy controls, from the Department
of Neurology at Xiangya Hospital. And all patients went through
comprehensive medical history, physical examination, cranial mag-
netic resonance imaging (MRI) scans, and EEG. Inclusion criteria of
TLE and drug-resistant epilepsy were accorded to our previous re-
search.28 Written informed consent was obtained from all enrolled
participants. Study was conducted in accordance with the guide-
line for the research involving human and approved by the Ethics
Committee of Central South University, Xiangya School of Medicine
and the affiliated Xiangya Hospital (201303120). The data were
divided into two set s: in the training cohort, 30 TLE patients were
analyzed; in the validation phase, candidate CpGs were validated in
another independent cohort (n = 48).
2.2 | DNA methylation quality
control and processing
Whole blood DNA extraction and quality control were constructed
as in our previous study.28 The discovery and training samples were
run on the Illumina Infinium HumanMethylation450 BeadChip
Kit (450K array). The validation samples were run on the Illumina
Infinium HumanMethylationEPIC BeadChip (850K array). The sam-
ples DNA underwent bisulfite treatment using the EZ-96 DNA
Methylation kit (Zymo Research Corporation, Irvine, CA, USA) and
hybridized to arrays according to Illumina recommended protocols.
All samples passed the Illumina quality control. Methylation at indi-
vidual CpG was reported as a methylation β-value, ranging continu-
ously from 0 (unmethylated) to 1 (completely methylated). The minfi
R package (Version 1.18.1) was used to retrieve raw data of 450K and
850K array. Initially, we excluded probes located on the sex chro-
mosome and null probes. We also removed the failed probes with a
|
1023
XIAO et Al .
detection P-value > .05 in more than 5% samples. The probes with
single-nucleotide polymorphisms of MAF > 5% within 10 bp of the
CpG sites were also rejected. We next performed Subset-quantile
Within Array Normalization (SWAN) methods for normalization.40
The probes of 450K assay are expected to perform similarly on data
from the 850K array. In this study, we removed the set of CpG sites
that were not included in the 850K array.
2.3 | Building a diagnostic model
We included three phases to identify and validate disease-related
CpGs signature for patients with TLE. In the discover y phase, the
logical regression test was performed to obtain differentially meth-
ylated CpG sites (DMCs) between 30 TLE and 30 normal control
samples, with a threshold value of .0 01 for P-value was used sub-
sequently for filtration. The 237 candidate CpGs were analy zed by
Le as t Abso lu te Sh rin kag e an d Sele cti on Op era tor (LASS O) meth ods .
The CpGs were then ranked by the regression parameters. In the
training phase, to further shrink the marker numbers to a reason-
able range, support vector machine (SVM) algorithm was used for
different number of CpGs. As a result, eight CpGs with the highest
pred ict ion acc ura c y we re conf irm ed. SVM al go rit hm s were tu ned by
5-fold internal cross-validation, which implies optimal determination
of parameters of the SVM algorithm. In the validation phase, the
parameters of the SVM model from the training cohort were used
to an ind ep endent co ho rt of 96 samples (4 8 TL E an d 48 nor ma l con-
trols) for validating the diagnostic performance of the model.
2.4 | Building a predictive model for drug response
In the discovery phase, the t test was performed to identify DMCs
between 10 drug-resistant and 20 drug-responsive samples, with a
threshold value of .005 for P-value. After 99 DMCs were obtained,
we used SVM-Recursive Feature Elimination (SVM-RFE) to select
candidate CpG sites. In the training phase, logistic regression was
used to further narrow CpGs. Six CpGs with the highest predic-
tion accuracy were identified, with parameter tuning conducted by
5-fold cross-validation. A risk score was calculated for each patient
using a formula derived from the methylation levels of these 6 CpGs
weighted by their regression coefficient. Validation analyses were
per formed in another cohor t (17 drug-resista nt an d 13 drug-respon-
sive samples). In addition, a nomogram comprising integrated DNA
methylation risk score and clinicopathological data was generated to
predict the drug response. The performance of the nomogram was
explored graphically by calibration plot s.
2.5 | Bioinformatics analysis
Pathway analysis for CpG-associated genes was performed using
Ingenuit y Pathway Analysis (IPA; http://www.ingen uity.com/). For
the pur pose s of this study, the canonical pathw ay and dise as es fun c-
tions analysis available in IPA were applied, which resulted in the
inclusion of CpG corresponding genes and the other identified genes
interacting with in the analysis. The Fisher's exact test was applied to
measure the significance of the association between genes mapped
by IPA and the canonical pathway.
2.6 | Bisulfite pyrosequencing of selected DNA
methylation loci
Bisulfite pyrosequencing is well-established technique that used
for quantitative methylation analysis of genomic regions in single-
nucleotide resolution.41 We selected 4 CpG loci (cg25838818,
cg27564766, cg11954680, and cg26119877) for assay cross-val-
idation by bisulfite pyrosequencing. Blood DNA samples from 10
TLE patients and 10 healthy control cases or 10 drug-responsive
TLE cases and 10 drug-resistant TLE cases were bisulfite con-
verted, followed by PCR amplification of the relevant regions
using the PyroMark PCR kit (Qiagen, CA, USA) according manu-
facturer's instruction. Nucleotide probes with biotinylated ver-
sion can be detected by streptavidin sepharose, as listed below:
(a) cg25838818: GTAGTTGAGGGT TAGGA AAGATGTG (F), ATACA
AATAC CA AC TCCC TCTA ATTC AT ( R ) and GTGA A A AAT T T TAGT
TGGTG (S).
(b) cg27564766: GGAGGGATAGGGGTTGTTT (F), CCA ACCAA
CCACCTCATC (R) and T TGTGGTGGTT TATAGG (S).
(c) cg1195 468 0 : AT TAGAAT TA AGA GTG AT T TAGGA AG TG ( F), AA
AAAAAAATTTTCCTATTTCACCTTCTA (R) and GTGATTTAGGAAG
TGGTTAA (S).
(d ) cg 261198 7 7: A AG ATT G G G TG GT T TATAA G A AA G (F), CC ACA
AATA A A ACACATTT TAC TATAAC AC (R ) and G T TG T TTG GGATTAG
TTG (S).
Pyrosequencing assay, purification, and subsequent processing
of the biotinylated single-stranded DNA were carried out according
to the manufacturer's recommendations.
2.7 | Statistical analysis
In the comparative analysis of clinical characteristics (SPSS18.0),
measurement data (age, disease course, and seizure frequency) were
subject to K–S test following by statistically analyzing with Student's
t test or nonparametric test, and enumeration data (HS, aura, and
SGS) were assessed using chi-square test, with P-value < .05 con-
sidered statistically significant. For the current research, scikit-learn
(Version: 0.20) was used to perform the LASSO, SVMs, RFE, and
logistic regression. The predictability of the model was evaluated
by the area under the receiver operating characteristic (ROC) cur ve
(AUC). The Youden's index was defined for all points of a ROC curve,
the maximum value of which may be used as a criterion for screening
the optimum cutof f point. The nomogram was constructed using the
rms pack age in R (Vers io n: 3.5.1) . Al l stati st ic al test s were tw o- side d.
1024
|
XIAO et Al.
3 | RESULTS
3.1 | Characteristics of individuals
A total of 156 participants (including 10 patients with drug-resist-
ant epilepsy and 20 patients with drug-responsive epilepsy and 30
healthy controls in discovery and training phase, 17 drug-resistant,
13 drug-responsive patients and 18 unclassified patient s and 48
healthy controls in validation phase) were recruited to our study. No
significant dif ferences of age and gender were found in the training
and validation set. The duration of seizures in patients with drug-
resistant epilepsy was significantly longer than that in patients with
drug-responsive epilepsy (P = .01 in training phases and P = .04 in
validation phases). The pathology of hippocampal sclerosis (HS) in
patients with drug-resistant epilepsy was significantly more than
that in patients with drug-responsive epilepsy (P = .002 in training
phases and P = .026 in validation phases). Seizure frequency, the ex-
istence of aura, and secondarily generalized seizure were not related
to drug response of TLE patients. The detailed clinical characteristics
of participants were listed in Table 1.
3.2 | Diagnostic model for TLE
To identify the TLE-associated CpGs, we first studied the global
methylation profiles in the DNA of whole peripheral blood obtained
from 30 TLE patients and 30 healthy controls. Epigenome-wide as-
sociation identified 237 DMCs associated with TLE at P < .001 by
logistic regression (Figure 1A, Table S1).
LASSO algorithm was used to select the most significant CpGs
from 237 candidate CpGs. We then used SVM algorithm to further
narrow down the marker numbers. As a result, eight CpGs were
identified (cg25838818, cg27564766, cg07782795, cg09383187,
cg09293614, cg09270525, cg09197288, and cg08664849), corre-
sponding to SU LT1C 2, TP73, BAIAP2, CLIP2, MUM1, PTPRN2, IFI27L1,
and TB C1D24, respectively (Figure 1B, Table S2). Subsequent valida-
tion, using a separate validation cohort (n = 96), yielded a sensitivity
of 71%, a specificity of 73%, and an accuracy of 77%, with an AUC of
0.81 to detec t TLE (Figure 1C, Table S3).
3.3 | Prediction model for drug response
T test was used to analyze the DMCs between 10 drug-resistant and
20 drug-responsive patients, which identified 99 DMCs at P < .005.
(Figure 2A, Table S4).
99 DMCs were analyzed by SVM-RFE algorithm to selec t signif-
icant CpGs. Logistic regression was used to further narrow CpGs.
Six CpGs were identified (cg15999964, cg08768218, cg11954680,
cg17706086, cg21761639, and cg26119877), corresponding
to ZNF608, DLC1, PCDHA, MEST, and SLC25A21, respectively
(cg21761639 has no corresponding gene) (Figure 2B; Table S2).
To better investigate the performance of CpGs signature in
predicting drug response, a methylation risk score was built with
the coefficients weighted by the logistic regression model in the
validation cohort (17 drug-resistant, 13 drug-responsive). The
methylation risk score was calculated as follows: risk score = 19.3
*cg15999964 − 43.5 *cg08768218 + 54.9 *cg11954680 + 26.3
*cg17706086 + 66.8 *cg21761639 + 29.9 *cg26119877 − 86.3, with
a cutoff value of 0.78. Applying the model yielded a sensitivity of
77%, a specificity of 71%, and an accuracy of 73%, with an AUC of
0.79 in the validation cohort, to distinguish drug-responsive from
drug-resistant patients (Figure 2C, Table S5).
3.4 | Building a predictive nomogram
We performed the multivariate analysis of the methylation risk score
and clinicopathological charac teristics with dr ug response in the en-
tire TLE cohor t (Figure S1). The methylation risk score and HS were
TABLE 1 Clinical participants of individuals
Training set Validation set
Drug-
responsive
Drug-
resistant
P-
value Control
Drug-
responsive
Drug-
resistant
P-
value Control
No. 20 10 30 13 17 48
Age, mean ± SD (y) 28.6 ± 10.9 31.4 ± 16.5 .65 31.3 ± 10.3 18.8 ± 10.2 31.6 ± 11.0 .32 31.5 ± 10.3
Female: male 7:13 5:5 .46 12:18 8:5 9:8 .72 25:23
Disease course,
mean(range) (y)
5 (1-24) 13 (4-29) .01 NA 7 (1-13) 12 (1-26) .04 NA
Seizure frequency,
mean(range) (/
month)
3 (0.1-120) 1 (0.3-90) .62 NA 2 (0.01-75) 7.5 (1-300) .28 NA
HS 2 (10%) 7 (70%) .002 NA 2 (15%) 10 (59%) . 026 NA
Aura 13 (65%) 5 (50%) .46 NA 7 (54%) 10 (59%) 1.0 NA
SGS 12 (60%) 5 (50%) .71 NA 10 (77%) 14 (82%) 1.0 NA
Abbreviations: HS, hippocampal sclerosis; NA , not applicable; SD, standard deviation; SGS, secondarily generalized seizure.
|
1025
XIAO et Al .
significantly associated with drug response. To develop a clinically
applicable method that could predict drug-resistance probability of
an individual TLE patient, a nomogram was used to built a predic-
tive mod el in the entire TL E co ho r t, ta ki ng into co ns idera ti on cli nico -
pathological factors (Figure 3). The predictors included methylation
risk score, disease course, seizure frequency, and HS. Applying the
model yielded a sensitivity of 94% and a specificity of 89%, with an
AUC of 0.96 in the entire TLE patient cohor t, to distinguish drug-
responsive from drug-resistant patients (Figure 4). The calibration
plots for drug-response nomogram model were predicted well in the
entire TLE patient cohort. (Figure S2).
3.5 | Bioinformatics analysis
All biomarker CpGs corresponding genes were uploaded to IPA for
the canonical pathway and diseases functions analysis, and network
generation for defined molecular interactions. Results were visual-
ized as networks (Figure S3, Figure S4) and ranked as diseases func-
tions and canonical pathways involved (Table S6). The IPA analysis
showed that “cell death and survival” and “cellular development”
were the top-ranked diseases functions. DLC1, IFI27L1, TP73, and
PTPRN2 involve in “cell death and survival” and T BC1D24, BAIAP2,
CLIP2, and SU LT1C 2 involve in “neurological disease.” Fur thermore,
“DNA methylation and transcriptional repression signaling” were the
top-ranked canonical pathways of 6 drug response–related CpGs
corresponding genes, and BAIAP2 involves in “axonal guidance sign-
aling” pathway. Notably, DLC1 is clos ely rel ate d DNMT gene (DN MT1
and DNMT3B) in the gene-interaction network of 6 drug response–
related CpGs corresponding genes.
3.6 | Cross-validation of methylation with bisulfite
pyrosequencing
To evaluate the accuracy of DNA methylation data from methylation
beadchip, a subset of CpG loci was selected for additional methyla-
tion validation by the pyrosequencing. Blood DNA samples of TLE
patients (n = 10) and controls (n = 10) were subjected to methyla-
tion detection at 2 loci (cg25838818, cg27564766), and drug-re-
sponsive TLE (n = 10) and drug-resistant TLE (n = 10) were subjected
to methylation detection at 2 loci (cg11954680 and cg26119877).
Pyrosequ encing revealed methy lation of cg25838818, cg27564766,
cg11954680, and cg26119877 was correlated with the data from
beadchip array (Figure 5A-D).
4 | DISCUSSION
In this study, we used the methylation array to screen differential
CpGs and selec ted significant CpGs by applying machine learning
algorithms in the training cohort. Subsequently, we validated the
FIGURE 1 Screening and validation of disease-related CpGs. A, Cluster analysis of 237 DMCs associated with TLE at P < .001 by logistic
regression in the training cohort. B, SVM algorithms in the training cohort. C , Receiver operator characteristic cur ve of 8 significant CpGs
prediction of TLE patients or healthy controls. The area under the ROC curve in training cohort was 0.90 and 0.81 for validation cohort
1026
|
XIAO et Al.
candidate CpGs and built the methylation-based signature in the
validation cohor t. Finally, a nomogram comprising integrated DNA
methylation risk score and clinicopathological data was generated to
pre di ct the dru g re sponse in the ent ire patie nt cohor t. Our dat a in di-
cated that the methylation-based signature could define human TLE
and predict drug-resistant. The methylation-based biomarker may
have clinical applications for individualized diagnosis and treatment
outcome prediction for patients with TLE. This study introduced a
methodological framework to screen and validate biomarker and
demonstrates the ability to use machine learning as a potential clini-
cal tool for epilepsy diagnosis and drug-response prediction after
more comprehensive validation.
While great effort s have been made in understanding the under-
lying pathogenic and drug-resistant mechanisms of epilepsy, there
are no existing treatment s to prevent or disease-modify the devel-
opment.10 It is believed that the complex and multifactorial features
FIGURE 2 Screening and validation of drug response–related CpGs. A, Cluster analysis of 99 DMCs associated with TLE drug response at
P < .005 by t test in the training cohort. B, Logistic regression algorithms in the training cohort. C , Receiver operator characteristic curve of 6
significant CpGs prediction of TLE patients drug-responsive or drug-resistant. The area under the ROC curve in training cohort was 0.99 and
0.79 for validation cohort
FIGURE 3 Nomogram to predict the
drug response in the entire TLE patient
cohort. The nomogram is used by adding
up the points identified on the scale for
four variables. The sum is located on the
“Total points” scale, and a line is drawn
downward axes to determine the risk of
resistance
|
1027
XIAO et Al .
of epilepsy have led to hampered progress in these areas.10 Re ce nt ly,
epigenetics as a mediator of gene-environment interactions had ap-
pealed to growing interesting in understanding the potential role in
complex diseases.42,43 Notably, DNA methylation is attractive to ex-
plain the underlying epileptogenesis and pharmacoresistance mech-
anism in chronic human epilepsy.27, 4 4- 4 6 Exploring peripheral DNA
methylation alterations in epilepsy is considered a direction with
translational significance, given that blood samples are available in
most clinical settings. Therefore, this study has several strengths
should be noted. First, our focus on the specific epilepsy syndrome
(TLE), the comparatively homogeneous groups of patients, advan-
tages over the study involving different epilepsy phenotypes. This
study features a genome-wide, dual-platform approach to screen
TLE biomarkers followed by interrogation of several CpGs in a co-
hort of samples. Moreover, we replicated these DNA methylation
sig na tu re s in ano th er ind ependent cohort for val id at io n, and the ma-
chine learning model was performed well.
There are additional mechanistic links between all biomarker
CpGs corresponding genes and epilepsy. For example, Tre2/Bub2/
Cdc16 ( TBC)1 domain family member 24 (T BC1D24) gene is one of
the more recently discovered pathogenic mutations of familial epi-
lepsy, of which associated disorders range from severe epileptic en-
cephalopathy to nonsyndromic hearing loss.47, 4 8 T BC1D24 has been
implicated in normal neural development and survival and plays
an essential role in neurotransmission and presynaptic function.49
In sporadic mesial TLE-HS, whole-exome sequencing has iden-
tified nonsynonymous de novo variants in BAIAP2 gene, of which
knockout model might produce aberrant neurological phenotypes
listed by Mouse Genome Informatics (MGI).50 Interestingly, ade-
nosine treatment reversed the DNA hypermethylation (including
Sult1c2 gene) status in the brain of TLE mo del, inhibited sprouting of
mossy fibers in the hippocampus, and prevented epileptogenesis.51
Further characterization of molecules such as TBC1D24, BAIAP2,
and SULT1C 2 will provide new insights into TLE development and
progression.
In addition, we noted that drug response–related CpGs cor-
responding genes closely related to DNA methylation, which
implicated that DNA methylation play an essential role in phar-
macoresistance mechanisms of epilepsy. Tumor suppressor gene
deleted in liver cancer 1 (DLC1) is shown to induce apoptosis, fre-
quently silenced by methylation and negative correlation with
DNMT expression.52,5 3 Interestingly, previous research found in-
creased expression of DNMT1 and DNMT3A in patients with in-
tractable TLE.22 Hypermethylation of gene promoters was also the
predominant effe ct in TL E pati en ts and rode nt mod el s as well. 24 ,2 7-29
Given the well studied of epigenetic pathomechanisms underlying
drug resistance in cancer, Kobow proposed that the methylation hy-
pothesis of pharmacoresistance could open such new avenue in the
field of epilepsy.45
We produced a nomogram including DNA methylation risk score,
disease course, seizure frequency, and HS for estimation of individ-
ualized outcomes of the drug response in TLE patients. The AUC of
predictive model is as high as 0.96, which suggests that this model
is promising to be applicable in clinical practice. In the previous re-
search, some seizure-related characteristics, including the prior
number of seizures and disease course, were reported to be related
with the risk of drug resistance.4,9 Another factor strongly linked to
the increased risk of intractable epilepsy is HS, which is consistent
wit h the findings in our stu dy. Howeve r, limited st ud ie s ap pl ie d com-
binatorial biomarker signatures to predict drug response in epilepsy
patients. A model composed of clinical variables (the presence of HS)
in combination with genetic information (SNP genot ypes located in
11 genes influencing drug transport and metabolism) improved pre-
dictive accuracy for medical intractability in mesial TLE.54 With the
advent of high-throughput technologies and the availabilit y of mul-
tidimensional data sets, we suggest the need to combine compound
molecular approaches to achieve higher predictive performance for
clinical usefulness and better comprehend the knowledge about the
relevant underlying pathomechanism.
The study also has certain limits and constraint s that should be
noted when interpreting the results. First of all, although our classi-
fication results were promising, we should point out that high dimen-
sional data with small sample size may result in misclassifications and
biased predictors. A larger set of patients can enhance the robust-
ness of the predictive model. Second, due to the single-center study,
the results of this pilot study warrant further validation in samples
from several neurological centers. Third, TLE patients participated in
this study were all being treated with antiepileptic medication, and
it is still unknown whether these might affect the DNA methylation
status. Furthermore, other confounding factors, such as cellular
composition in whole blood, should also be taken into consider-
ation.55 Fourth, two different platforms of methylation data set were
hired that is 450K array for training while 850K array for validation.
Several studies have reported that overall correlations of matched
samples running both on the 450K and 850K array were quite high
FIGURE 4 Receiver operator characteristic curve of 6
significant CpGs combined and not combined clinicopathological
factors prediction of drug-responsive or drug-resistant in the entire
TLE patient cohort
1028
|
XIAO et Al.
(r > .90 for all assessed samples).56, 57 Last, longitudinal study includ-
ing the patients before antiepileptic medication is recommended to
warrant clinical significance of the predictive biomarkers. The bio-
logic mechanisms of the candidate markers are still little known, and
thorough in vivo and vitro experiments are also needed to future
investigate.
5 | CONCLUSIONS
For the first time, we demonstrated that DNA methylation signa-
ture could define human TLE and compound with clinicopathologi-
cal factors to improve the prediction of response to drug treatment.
Furthermore, this study introduced a methodological framework to
screen and validate biomarker and demonstrated the ability to use
machine learning as a potential clinical investigative tool. Despite the
limited pathomechanism contributions, we highlight the utilization
of promising biomarkers in clinical practice for decision-making.
ACKNOWLEDGMENTS
Foremost, we thank the patients, their families, and healthy controls
for their cooperation and participation in this study. This study was
supported financially by the National Natural Science Foundation of
China (No.81671299 and No.81974206 to BX, No.81671300 to LLL
and No.81701182 to YJF), Project from Hunan Provincial Science
and Technology Department (No.2019SK1012 to BX ), Clinical
Research Foundation of Xiangya Hospital (No.2016L08 to LLL), and
Fundamental Research Funds for the Central Universities of Central
South University (No. 2018zzts248 to WBX).
CONFLICT OF INTEREST
The authors declare no conflict of interest.
FIGURE 5 Cross-validation of DNA methylation with the pyrosequencing. Shown are degrees of methylation of 4 CpG loci reported by
methylation BeadChip (Y axis, ratio) and pyrosequencing (X axis, ratio) assays. For cg25838818 (A), cg27564766 (B), cg11954680 (C), and
cg26119877 (D), the degrees of methylation detected by the two methods were positively correlated (P < .05) in reference to individual
samples
|
1029
XIAO et Al .
ETHICAL APPROVAL
The study was approved by the Ethics Committee of Central South
University, Xiangya School of Medicine and the affiliated Xiangya
Hospit al (201303120). Informed con se nt was obt ained fro m the par-
ents/legal guardians of all patients.
ORCID
Wenbiao Xiao https://orcid.org/0000-0002-8922-4654
Bo Xiao https://orcid.org/0000-0001-5204-1902
Hongyu Long https://orcid.org/0000-0001-5641-8850
REFERENCES
1. Tellez-Zenteno JF, Hernandez-Ronquillo L. A review of the
epidemiology of temporal lobe epilepsy. Epilepsy Res Treat.
2012;2012:630853.
2. Blumcke I, Thom M, Aronica E , et al. International consensus clas-
sification of hipp ocampal sclerosis in temporal lobe epilepsy: a task
Ffrce repor t from the ILAE commission on diagnostic methods.
Epilepsia. 2013;54(7 ):1315-1329.
3. Janszky J, Janszk y I, Schulz R, et al. Temporal lobe epilepsy with
hippocampal sclerosis: predictors for long-term surgical outcome.
Brain. 2005;128(Pt 2):395-404.
4. Bjorke AB, Nome CG, Falk RS, Gjerstad L, Tauboll E, Heuser K.
Evaluation of long-term antiepileptic drug use in patient s with tem-
poral lobe epilepsy: assessment of risk factors for drug resistance
and polypharmacy. Seizure. 2018;61:63-70.
5. Kanner AM. Management of psychiatric and neurological comor-
bidities in epilepsy. Nat Rev Neurol. 2016;12 (2):10 6-116.
6. Raoof R, Jimenez-Mateos EM, Bauer S, et al. Cerebrospinal fluid
microRNAs are potential biomarkers of temporal lobe epilepsy and
status epilepticus. Sci Rep. 2017;7(1):3328.
7. Wang J, Yu JT, Tan L, et al. Genome-wide circulating microRNA
expression profiling indicates biomarkers for epilepsy. Sci Rep.
2015;5:9522.
8. Kwan P, Brodie MJ. Early identification of refractory epilepsy. N
Engl J Med. 20 00;342(5):314-319.
9. Huang L, Li S, He D, Bao W, Li L. A predic tive risk model for medic al
intrac tability in epilepsy. Epilepsy Behav. 2014;37:282-286.
10. Pitkanen A , Loscher W, Vezzani A , et al. Advances in the develop-
ment of biomarkers for epilepsy. Lancet Neurol. 2016;15(8):843-856.
11. Pitkanen A, Ekolle NX , Lapinlampi N, Puhakka N. Epilepsy bio-
markers - Toward etiology and pathology specificity. Neurobiol Dis.
2019;123:42-5 8.
12. Engel JJ, Pitkanen A, Loeb JA, et al. Epilepsy biomarkers. Epilepsia.
2013;54(Suppl 4):61-69.
13. Raoof R, Bauer S, El NH, et al. Dual-center, dual-platform microRNA
profiling identifies potential plasma biomarkers of adult temporal
lobe epilepsy. Ebiomedicine. 2018;38:127-141.
14. Wang J, Tan L, Tan L, et al. Circulating microRNAs are promising
novel biomarkers for drug-resist ant epilepsy. Sci Rep. 2015;5:10201.
15. Koch A , Joosten SC , Feng Z, et al. Analysis of DNA methylation in
cancer : location revisited. Nat Rev Clin Oncol. 2018;15(7):459-466.
16. Kinde B, Gabel HW, Gilbert CS, Grif fith EC, Greenberg ME. Reading
the unique DNA methylation landscape of the brain: Non-CpG
methylation, hydroxymethylation, and MeCP2. Proc Natl Acad Sci
USA. 2015;112(22):6800-6806.
17. Bestor TH. The DNA methyltransferases of mammals. Hum Mol
Genet. 2000;9(16):2395-2402.
18. Bernstein BE, Meissner A , L ander ES. The mammalian epigenome.
Cell. 2007;128(4):669-681.
19. Jones PA. Functions of DNA methylation: islands, start sites, gene
bodies and beyond. Nat Rev Genet. 2012;13(7 ):484-492.
20. Belhedi N, Perroud N, Karege F, Vessaz M, Malafosse A , Salzmann
A. Increased CPA6 promoter methylation in focal epilepsy and in
febrile seizures. Epilepsy Res. 2014;10 8(1):14 4-148.
21. Unal Y, Kara M, Genc F, et al. The methylation status of NKCC1
and KCC2 in the patients with refractory temporal lobe epilepsy.
Ideggyogy Sz. 2019;72(5-6):181-186.
22. Zhu Q, Wang L, Zhang Y, et al. Increased expression of DNA meth-
yltransferase 1 and 3a in human temporal lobe epilepsy. J Mol
Neurosci. 2012;4 6(2):420- 426.
23. Kobow K, Jeske I, Hildebrandt M, et al. Increased reelin promoter
me t hy lat ion is ass oci ate d wit h gra nul e ce ll di spe rsi on in hu man te m-
poral lobe epilepsy. J Neuropathol Exp Neurol. 2009;68(4):356-364.
24. Kobow K, Kaspi A, Harikrishnan KN, et al. Deep sequencing re-
veals increased DNA methylation in chronic rat epilepsy. Acta
Neuropathol. 2013;126(5):741-756.
25. Debski KJ, Pitkanen A, Puhakka N, et al. Etiology matters - genomic
DNA methylation pat terns in three rat models of acquired epilepsy.
Sci Rep. 2016;6:25668.
26. Miller-Delaney SF, Das S, Sano T, et al. Differential DNA methyl-
ation patterns define status epilepticus and epileptic tolerance. J
Neurosci. 2012;32(5):1577-1588.
27. Miller-Delaney SF, Bryan K, Das S, et al. Differential DNA meth-
ylation profiles of coding and non-coding genes define hippocam-
pal sclerosis in human temporal lobe epilepsy. Brain. 2015;138(Pt
3): 616 -6 31 .
28. Long HY, Feng L, Kang J, et al. Blood DNA methylation pattern is
altered in mesial temporal lobe epilepsy. Sci Rep. 2017;7:43 810.
29. Xiao W, Cao Y, Long H, et al. Genome-wide DNA methylation pat-
terns analysis of noncoding RNAs in temporal lobe epilepsy pa-
tients. Mol Neurobiol. 2018;55(1):793-803.
30. Urdinguio RG, Sanchez-Mut JV, Esteller M. Epigenetic mechanisms
in neurological diseases: genes, syndromes, and therapies. Lancet
Neurol. 2009;8(11):1056-1072.
31. Li L, Choi JY, Lee KM , et al. DNA methy lation in peripheral blood: a
potential biomarker for cancer molecular epidemiology. J Epidemiol.
2012;22(5):384-394.
32. Sina AA, Carrascosa LG, Trau M. DNA methylation-based point-of-
care cancer detection: challenges and possibilities. Trends Mol Med.
2019;25(11):955-966.
33. Wick W, Weller M, van den Bent M, et al. MGMT tes ting–the
challenges for biomarker-based glioma treatment. Nat Rev Neurol.
2014;10(7):372-385.
3 4. Bell EH, Zhang P, Fisher BJ, et al. A ssociation of MGMT pro-
moter methylation status with survival outcomes in patients with
high-risk glioma treated with radiotherapy and temozolomide: an
analysis from the NRG oncology/RTOG 0424 trial. Jama Oncol.
2018 ;4 (10) :1405-14 09.
35. Church TR, Wandell M, Lof ton-Day C , et al. Prospec tive evaluation
of methylated SEPT9 in plasma for detection of asymptomatic col-
orect al cancer. Gut. 2014;63(2):317-325.
36. Qiu J, Peng B, Tang Y, et al. CpG methylation signature predicts
recurrence in early-stage hepatocellular c arcinoma: results from a
multicenter study. J Clin Oncol. 2017;35(7):734-742.
37. Widschwendter M, Zikan M, Wahl B, et al. The potential of circu-
lating tumor DNA methylation analysis for the early detection and
management of ovarian cancer. Genome Med. 2017;9(1):116.
38. Li Y, Gu J, Xu F, Zhu Q, Ge D, Lu C. Novel methylation-driven genes
identif ied as prognostic indicators for lung squamous cell carci-
noma. Am J Transl Res. 2019;11(4):1997-2012.
39. Wang L, Shi J, Huang Y, et al. A six-gene prognostic model pre-
dicts overall survival in bladder cancer patients. Cancer Cell Int.
2019;19:2 29.
40. Maksimovic J, Gordon L, Oshlack A. SWAN: subset-quantile within
array normalization for illumina infinium HumanMethylation450
BeadChips. Genome Biol. 2012;13(6):R44.
1030
|
XIAO et Al.
41. Delaney C , Garg SK, Yung R. Analysis of DNA methylation by py-
rosequencing. Methods Mol Biol. 2015;1343:249-264.
42. Gal-Yam EN , Saito Y, Egger G, Jones PA. Cancer epigenetics: modifi-
cations, screening, and therapy. Annu Rev Med. 2008;59:267-280.
43. Qureshi IA, Mehler MF. Understanding neurological disease mech-
anisms in the era of epigenetics. Jama Neurol. 2013;70(6):703-710.
4 4. Kobow K, Blumcke I. The methylation hypothesis: do epigenetic
chromatin modifications play a role in epileptogenesis? Epilepsia.
2011;52(Suppl 4):15-19.
45. Kobow K, El-Osta A, Blumcke I. The methylation hypothesis of
pharmacoresistance in epilepsy. Epilepsia. 2013;54(Suppl 2):41-47.
46. Kobow K, Blumcke I. Epigenetics in epilepsy. Neurosci Lett.
2018 ;667: 40 -46.
47. Balestrini S, Milh M, Castiglioni C, et al. TBC1D24 genotype-phe-
notype correlation: epilepsies and other neurologic features.
Neurology. 2016;87(1):77-85.
48. Lozano R, Herman K, Rothfuss M, et al. Clinical intrafamilial variabil-
ity in lethal familial neonat al seizure disorder caused by TBC1D24
mutations. Am J Med Genet A. 2016;170(12):3207-3214.
49. Finelli MJ, Aprile D, Castroflorio E, et al. The epilepsy-associated
protein TBC1D24 is required for normal development, sur vival
and vesicle trafficking in mammalian neurons. Hum Mol Genet.
2019;28(4):584-597.
50. Wong J, Gui H , Kwok M, et al. Rare variants and de novo variants
in mesial temporal lobe epilepsy with hippocampal sclerosis. Neurol
Genet. 2018;4(3):e245.
51. Williams-Karnesky RL , Sandau US, Lusardi TA, et al. Epigenetic
changes induced by adenosine augmentation therapy prevent ep-
ileptogenesis. J Clin Invest. 2013;123(8):3552-3563.
52. Seng TJ, Low JS, Li H, et al. The major 8p22 tumor suppre ssor DLC1
is frequently silenced by methylation in both endemic and sporadic
nasopharyngeal, esophageal, and cervical carcinomas, and inhibits
tumor cell colony formation . Oncogene. 2007;26(6):934 -944.
53. Fang QL, Yin YR, Xie CR, et al. Mechanistic and biological signifi-
cance of DNA methyltransferase 1 upregulated by growth factors
in human hepatocellular carcinoma. Int J Oncol. 2015;46(2):782-790.
5 4. Silva-Alves MS, Secolin R, Carvalho BS, et al. A prediction algo-
rithm for drug response in patients with mesial temporal lobe
epileps y based on clinical and genetic information. PLoS ONE.
2017;12 (1):e169214.
55. Jaffe AE, Irizarry RA. Accounting for cellular heterogeneity is
critical in epigenome-wide association studies. Genome Biol.
2014;15(2):R31.
56. Logue MW, Smith AK , Wolf EJ, et al. The correlation of methylation
levels measured using Illumina 450K and EPIC BeadChips in blood
samples. Epigenomics-UK. 2017;9(11):1363-1371.
57. Pidsley R, Zotenko E, Peters TJ, et al. Critical evaluation of the
Illumina MethylationEPIC BeadChip microarray for whole-genome
DNA methylation profiling. Genome Biol. 2016;17(1):208.
SUPPORTING INFORMATION
Additional supporting information may be found online in the
Supporting Information section.
How to cite this article: Xiao W, Liu C, Zhong K, et al. CpG
methylation signature defines human temporal lobe epilepsy
and predicts drug-resistant. CNS Neurosci Ther.
2020;26:1021–1030. ht tps://doi. or g/10.1111/cns.133 94
Available via license: CC BY 4.0
Content may be subject to copyright.
