Circulating microRNAs are promising novel biomarkers for drug-resistant epilepsy

MicroRNAs (miRNAs) open up a new field for molecular diagnosis for cancer and other diseases based on their stability in serum. However, the role of circulating miRNAs in plasma/serum in epilepsy diagnosis is still unclear. The aim of this study was to evaluate whether miRNAs can be used as biomarkers for drug-resistant epilepsy. We measured the differences in serum miRNA levels between 30 drug-resistant patients and 30 drug-responsive epilepsy patients in discovery and training phases using Illumina HiSeq2000 sequencing followed by quantitative reverse transcriptase polymerase chain reaction (qRT-PCR) assays. The selected miRNAs were then validated in 77 drug-resistant epilepsy patients, 81 drug-responsive epilepsy patients and 85 healthy controls by qRT-PCR. We found that circulating miRNAs are differentially expressed between drug-resistant group and drug-responsive group. MiR-194-5p, -301a-3p, -30b-5p, -342-5p and -4446-3p were significantly deregulated in drug-resistant group compared to drug-responsive group and control group. Among these 5 miRNAs, miR-301a-3p had the best diagnostic value for drug-resistant epilepsy with 80.5% sensitivity and 81.2% specificity, and was negatively associated with seizure severity. These provide the rationale for further confirmation studies in larger prospective cohorts and in other ethnics.

Distinct circulating miRNA profilings of drug-resistant epilepsy vs drug-responsive epilepsy in discovery set. In total, genome-wide sequencing identified 10,000,000 raw reads in both drug-resistant group and drug-responsive group. As is shown in Fig. 1A,B, the dominant small RNAs were 22-23nt in length, accounting for 72.77% and 77.65% of the total reads in drug-resistant and drug-responsive group, respectively. After getting rid of low-quality sequences, sequences shorter than 18 nucleotides, and single-read sequences, 9,630,805 (96.65%) clean reads in drug-resistant group and 9,606,969 (96.40%) clean reads in drug-responsive group were remained for further analysis. Among these clean reads, 6192151 (64.3%) reads in drug-resistant group and 5983857 (62.29%) reads in drug-responsive group were perfectly mapped to the human genome in Genbank. Although miRNAs accounted only a tiny fraction of the total small RNAs, the expression levels of individual miRNAs were relatively high. Moreover, both the number of the unique miRNA sequences and the amount of miRNA species were mildly higher in drug-resistant epilepsy patients compared with drug-responsive epilepsy patients (1638vs 1050, 5467036 vs 5253711, respectively) ( Fig. 1C-F). The deep sequencing data and analyses of differentially expressed miRNAs were listed in Supplementary Table S1. Genome-wide sequencing showed that 185 miRNAs were differentially expressed between drug-resistant group and drug-responsive group. The miRNA levels were considered to be significantly different only if they met the following criteria 20 : (1) having at least 10 copies in drug-resistant or drug-responsive groups; (2) showing a fold-change (log-2 drug-resistant/drug-responsive) > 2 or < − 2 between each comparisons (P < 0.05). According to these criteria, we found that 12 miRNAs were downregulated (miR-194-5p, -204-5p, -221-5p, -301a-3p, -30b-5p, -342-5p, -3605-5p, -4446-3p, -598-3p, -874-3p, -889-3p and novel-mir-451) and 3 were upregulated (miR-574-5p, novel-mir-67 and novel-mir-9) in drug-resistant group compared to drug-responsive group (Supplementary Table S1). Among the 15 deregulated miRNAs, 3 miRNAs (novel-mir-451, -67 and -9) were not listed in miRBase (Release 21, http://www.mirbase.org/) and seems to be novel miRNAs.  Table S2). The primers for real-time PCR of each miRNA were listed in Supplementary Table S3. MiRNA levels were normalized to cel-miR-39. All samples were measured in triplicates and the mean values were used for analysis. Only miRNAs with a Cq value < 36, a detection rate > 75% in both groups, and a p value < 0.05 were selected for further analyses 20 . As a result, miR-194-5p, -301a-3p, -30b-5p, -342-5p and -4446-3p were significantly decreased in drug-resistant patients when compared with drug-responsive patients; while novel-mir-67 was increased in drug-resistant patients ( Fig. 2A). The detection rates of miR-3605-5p and novel-mir-9 were less than 75%; miR-221-5p and miR-889-3p displayed poor results of melting curving analysis; no significant difference was observed in the levels of miR-204-5p, -574-5p, -598-3p, -874-3p and novel-mir-451 between drug-resistant patients and drug-responsive patients (P > 0.05).

Relationship between serum levels of miR-301a-3p and clinical characteristics.
In addition to group comparisons, we performed regression analysis to investigate the association between the expression level of miR-301a-3p with clinical parameters. No significant association was observed between miR-301a-3p and disease duration or seizure frequency (P > 0.05, data not shown). Interestingly, we found that miR-301a-3p level was significantly associated with NHS3 score (r = 0.604, P = 6.2 × 10 −9 ) (Fig. 4), indicating that the expression level of miR-301a-3p was negatively associated with seizure severity.

Discussion
To date, the diagnosis and treatment of drug-resistant epilepsy still suffer from a lack of reliable biomarkers, despite ample efforts have been made [21][22][23] . While in recent years, miRNAs have gained significant attention and have been proposed as novel biomarkers for the diagnosis of several diseases including some CNS diseases 4,6-9,20 for several reasons. First, miRNAs have been found stable in serum, and the test of miRNAs in blood is broadly accessible, rapid, noninvasive, and economical. Moreover, the development of powerful detection technologies such as high-throughput sequencing has given a significant boost to the search in miRNAs as biomarkers. Over the past 5 years, several target studies and genome-wide miRNA expression profiling studies [10][11][12]14,15,[24][25][26] have identified changes to over 100 different miRNAs in epilepsy patients and animal models, particularly in mesial temporal lobe epilepsy (mTLE), about 30% of which are pharmaco-resistant 27 , and provided compelling evidence that epilepsy is associated with widespread changes to miRNA expression.
Here, we provided the first study to identify serum-based miRNA biomarkers for detection of drug-resistant epilepsy from drug-responsive epilepsy. Our results revealed that miRNAs are differentially expressed in serum samples from drug-resistant patients compared with those from drug-responsive patients. In particular, the expression of miR-194-5p, -301a-3p, -30b-5p, -342-5p and -4446-3p were significantly decreased in drug-resistant patients compared to drug-responsive patients and healthy controls. Among these miRNAs, miR-301a-3p had the best diagnostic value for drug-resistant epilepsy and yielded AUC of 0.897 with 80.5% sensitivity and 81.2% specificity in discriminating drug-resistant patients from drug-responsive patients, and was negatively associated with seizure severity.
The miRNA expression profiling and candidate miRNA biomarkers identified in our study showed some overlap, even limited, with previous reports. In 2012, Hu and colleagues 11 conducted profiling studies using hippocampus from rat model of TLE, and found miR-301a (previous ID of miR-301a-3p)  and other 14 miRNAs were down-regulated in TLE rat models. Later, Bot et al. 14 profiled miRNA expression levels in dentate gyri from epileptic rat models and sham operated controls, and found that 57 miRNAs including miR-301a-3p and miR-30b-5p were downregulated, while 9 miRNAs were upregulated in epileptic models. Over the same period, Mckiernan et al. 13 profiled mature miRNA levels in hippocampus from pharmacoresistant TLE patients and controls. Their results showed that 37 miRNAs including miR-301a-3p and miR-30b-5p were significantly downregulated in pharmaco-resistant TLE samples. These results showed that changes to miR-301a-3p and miR-30b-5p were consistent with our findings. However, in the study of Kan and colleagues 12 , which first undertook genome-wide profiling of miRNAs in human epilepsy, miR-301a was upregulated in mTLE patients with and without hippocampal sclerosis in comparison to controls. In addition, a set of miRNAs were deregulated in some studies, but not in others. These differences may be explained by the different standards for the selection of TLE patients or varied criteria for the surgery selection of TLE patients. Moreover, different standards to screen for significantly deregulated miRNAs may also contribute to a difference in results. Additionally, limited sample size, different models and/or brain regions, study design, technical factors, extraneous effects including race, BMI, lifestyle, and other individual characteristics may also influence the profiling of miRNA abundance. These need to be validated in the future.
Although miR-301a-3p has been demonstrated to be deregulated in drug-resistant epilepsy patients in present and previous studies, at this stage, we could not come to the conclusion that miR-301a-3p is ready to be used as a diagnostic biomarker for several limitations. First of all, the sample size in our study, although much larger than previous studies, is still limited. It influences the accuracy of the results. Moreover, participants in our study are from a confined geographic area with less heterogenous background. In addition, miR-301a-3p is also deregulated in some other diseases, including some types of tumors 28 and Alzheimer's disease 29 . Therefore, it is not specific for the diagnosis of drug-resistant epilepsy. Last, it is still not clear whether the deregulation of miRNAs is a cause or a consequence of epileptogenesis. So, large-scale prospective cohort studies in different ethnic populations are necessary to verify our findings. Nevertheless, some advantages make our study a reliable rationale for future studies. First, we employed a rigorous approach including a high-throughput sequencing of pooled serum samples followed by multiple qRT-PCR validation sets at the individual level. This approach has been widely utilized to identify a particular disease-specific serum miRNA profile. The high-throughput sequencing could detect the genome-wide miRNA expression, but ignored the individual discrepancies. Thus 2 stages of qRT-PCR were following to verify the different expression levels of selected miRNAs at individual level. Compared to other methods of measuring miRNA expression levels, RT-PCR assay is not affected by genomic DNA contamination, and is a sensitive and accurate method for assessing miRNA expression. Moreover, to make the result of qRT-PCR more accurate, we measured all samples in triplicates and used the mean value for analysis. In addition, our study is specially designed to explore different expression levels of miRNAs between drug-resistant and drug-responsive patients, and we also verified the differences between drug-resistant patients and controls.
The understanding of miRNA expression patterns as potential biomarkers for diagnosis of drug-resistant epilepsy is still in its infancy, and the miRNA targets and the molecular mechanisms concerning how miRNAs regulate epileptogenesis are not fully understood. Individual miRNAs can have several targets within the same cell and impact more than one pathway. In neurons, miRNAs have been found to regulate translation of a wide range of proteins 30 , including proteins involved in neuronal morphology 31 , channels 32 , neuronal migration 33 among others. A functioning miRNA system is also required in astrocytes with loss of miRNA biogenesis producing neurodegeneration and seizures 34 . In order to obtain a further understanding of the differentially expressed miRNAs in drug-resistant epilepsy, we predicted the potential targets of selected miRNAs using the miRNA target prediction databases-RNAhybrid and miRanda. A network of miRNAs and mRNAs of target genes is presented (Supplementary Fig.  S1). Within the network, many genes are related to inflammation and apoptosis, such as MAPK1, ATM, MYD88, RBL1, TRAF6, PIK3CD, IFNAR2, etc, indicating that these miRNAs may play a role in drug-resistant epilepsy through inflammation and apoptosis. These pathways may represent interesting novel targets for mechanism investigation and therapeutic interventions. In addition, miR-301a-3p is a potential biomarker in our study. It has been revealed that miR-301a-3p was involved in inflammatory response through impacting NF-κ B signaling pathway in cancer 28 . Further research is necessary to explore how miR-301a-3p function in epilepsy, involved in inflammation or some other mechanisms.
In conclusion, we first performed a comprehensive investigation of circulating miRNAs in drug-resistant epilepsy. In this report, we identified 5 serum miRNAs to distinguish drug-resistant epilepsy patients with drug-responsive epilepsy patients and healthy controls. Among these miRNAs, miR-301a-3p has strong potential to discriminate drug-resistant epilepsy from drug-responsive epilepsy with 80.5% sensitivity and 81.2% specificity. Our results contribute to the new avenue of miRNA biology in drug-resistant epilepsy and provide the rationale for larger prospective cohort studies in different ethnic populations, which are certainly needed to further confirm our preliminary results of deregulated serum-based miRNAs. age, gender and BMI between October, 2013 and May, 2014. A multiphase case-control study was designed to identify serum miRNAs as biomarkers for drug-resistant epilepsy (Fig. 5). In the discovery phase, we subjected pooled serum samples from 30 drug-resistant patients and 30 drug-responsive patients to Illumina HiSeq 2000 technology to select miRNAs whose expression were altered in drug-resistant patients compared to drug-responsive patients. Subsequently, we refined the number of serum miRNAs included as the drug-resistant signature by a 2-stage experimental procedure using real-time quantitative reverse transcriptase polymerase chain reaction (qRT-PCR) assays. The training phase used serum samples from the 30 drug-resistant patients and 30 drug-responsive patients that had been assessed by Illumina HiSeq 2000 technology, whereas the validation phase used serum samples from additional 77 drug-resistant patients, 81 drug-responsive patients and 85 healthy controls.

Methods
All the patients were recruited from the Department of Neurology at Qingdao Municipal Hospital, and several other hospitals in Shandong Province. And all patients went through comprehensive clinical examination, including a medical history, physical and psychiatric examination, laboratory examination, cranial magnetic resonance imaging scans and electroencephalogram. Major exclusion criteria were a history of autoimmune diseases, allergic response, immune deficiency disorder, diabetes, heart disease, stroke, atherosclerosis, psychiatric illness, malignancy, severe cognitive impairment, or a systemic or central nervous system (CNS) infection 2 weeks before sample collection. All patients with drug-resistant epilepsy were evaluated for seizure frequency using seizure diaries and seizure severity using the National Hospital Seizure Severity Scale (NHS3) 35 . The control subjects were recruited from the Health Examination Center of the Qingdao Municipal Hospital, and were confirmed healthy and neurologically normal by medical history, general examinations, laboratory examinations, and have no history of seizures or exposure to AEDs. An informed consent to participate in this study was obtained from each subject, and the study protocol was approved by the Ethics Committee of Qingdao Municipal Hospital. All the experiments described here were in accordance with the guidelines and regulations issued by the Ethics Committee of Qingdao Municipal Hospital.
In this study, epilepsy was diagnosed as idiopathic or cryptogenic epilepsy according to the criteria proposed by the International League Against Epilepsy in 2001 36 . Drug-resistant epilepsy was defined as failure of adequate trails of two tolerated and appropriately chosen and used AED schedules (whether as monotherapies or in combination) to achieve sustained seizure freedom. In our study, all the patients with drug-resistant epilepsy were still on medications at the time of serum testing. Drug-responsive epilepsy was defined as freeing from seizures for the period of at least 12 months 37 .
Blood processing. Up to 6 ml whole blood was collected from each participant, and was processed for serum isolation within 3 hours of collection by centrifugation at 3,000 r.p.m. for 5 min at room temperature, followed by a 5 min centrifugation at 12,000 xg at 4 °C 20 . The serum samples were stored at − 80°C and were not thawed until use. The hemolytic serum samples were excluded.
Serum small RNA library construction and sequencing. We mixed 300 μ l of each serum sample from 30 drug-resistant and 30 drug-responsive patients separately. Total RNA of each mixed serum was isolated using a scaled-up version of the mirVana™ PARIS™ Kit (Ambion, USA) protocol 6 . The final RNA was eluted in 100 μ l of preheated (95 °C) elution solution. The concentration and purity of RNA solution were examined by measuring the absorbance at 260-280 nm using the NanoDrop Lite Spectrophotometer (Thermo, Germany). After that, the 18-to 30-nt small RNAs were fractionated, and then were ligated to a 5' and a 3' adaptor sequentially. Next, the 5'-, 3'-ligated small RNA solution was reverse-transcribed to cDNA, followed by PCR with primers complementary to the adaptor sequences. Finally, the two generated libraries were sequenced using the Illumina Cluster Station and Genome Analyze (Illumina Inc, CA, USA) at BGI according to the manufacturer's protocol.
MiRNA quantification by real-time qRT-PCR. Twenty μ l total RNA solution was isolated from 400 μ l serum of each sample using the mirVana™ PARIS™ Kit according to the manufacturer's protocol. To allow for the normalization of sample-to-sample variation in RNA isolation, synthetic C. elegans miRNA cel-miR-39 38,39 was added (25 fmol in a 5 μ l total volume) to each denatured sample after combining the serum sample with 2× Denaturing Solution 6 . Then the total RNA was reversely transcribed into cDNA in a final volume of 20 μ l using One Step PrimeScript miRNA cDNA Synthesis Kit (Takara, Japan). Quantitative real-time PCR was conducted for each sample using SYBR Premix Ex Taq™ II (Takara, Japan) and CFX96 real-time PCR detection system (Bio-rad, Germany) in a final 25 μ l reaction volume according to the manufacturer's protocol. All miRNA primers were purchased from Takara and Tiangen (Beijing, China). At the end of PCR cycles, melting curve analyses were performed to validate the specific generation of the expected PCR products. Each sample was run in triplicates for analysis. Statistical analysis. The expression levels of miRNAs for qRT-PCR were normalized to cel-miR-39 40 , and were calculated utilizing the 2 −ΔΔCt method 41 . Expression levels of miRNAs were compared using the Kruskall-Wallis test or the Mann-Whitney U test. Receiver Operator Characteristic (ROC) curves and area under the ROC curve (AUC) were established to evaluate the diagnostic value of serum miRNAs for differentiating between drug-resistant group and drug-responsive group. The correlations between the variables were assessed with the Pearson's correlation coefficient. Clinical characteristics were compared using χ 2 test of independence for qualitative variables, ANOVA or t-test of quantitative variables with normal distribution, the non-parametric Kruskall-Wallis test or the Mann-Whitney U test of quantitative variables with skewed distribution. A p value of less than 0.05 was considered statistically significant. All analyses were performed by SPSS 17.0 software (SPSS, Chicago, IL, USA) or Graphpad Prism (version 5.0; Graphpad software).