Background

Migraine is a genetically-driven, chronic brain disease that ranks second on the most disabling neurological disorders of the Global Burden of Disease Study [1, 2]. If not adequately diagnosed and treated, migraine can increase in frequency and transform, through a chronification process, into chronic migraine (CM), which is defined by having 15 or more headache days per month [3], where patients’ burden increases exponentially [4]. Multiple factors, including both genetic and environmental mechanisms seem to play a role in migraine and its chronification [5,6,7] process, although the susceptibility of headache attack recurrence mainly depends on the brain-sensorial adaptability of both internal and external stimuli [8, 9]. Moreover, migraine stills represent a clinical challenge because, up to date, disease diagnosis, evolution and treatment response cannot be predicted due to the lack of reliable biomarkers.

From large-scale genome-wide association studies (GWAS) performed in the last decades, we have learnt that migraine has a polygenic heritability pattern and, up to now, we have identified more than 120 risk loci associated to migraine [10], mainly codifying for neuronal and vascular functions. However, the effect size of each individual reported gene is relatively modest and most of these genes have regulatory effects on gene expression rather than protein coding. Environmental factors may also modulate the expression of DNA. Therefore, the study of epigenetics which are the environmental influenced genetic regulatory mechanisms [11] could provide further understanding on migraine pathophysiology as well as, give insight to the dynamic aspect of the disease specifically contributing to its chronification [12].

Among the epigenetic regulatory mechanisms, microRNAs (miRNA) are a subtype of noncoding RNA of ∼22 nucleotides long that act as key regulators of genetic expression by inhibiting transcription or promoting degradation of selected messenger RNAs (mRNAs). A single miRNA sequence could regulate multiple mRNA, exerting a pleiotropic modulation of cellular processes. miRNA have been found intracellularly and in all body fluids, including plasma, saliva and urine [13]. Accessing the central nervous system (CNS) is too invasive and there is an increasing evidence that a relationship between peripheral blood mononuclear cells (PBMC) and brain-based epigenetic activity exists [14]. Thus, PBMC gene expression may reflect both central and peripheral neurological mechanisms in migraine, induced, for example in migraine by the activation and neuroinflammation of the trigeminocervical pain pathway [15]. In fact, previous studies in headache disorders have shown that transcriptional changes of PBMC can reflect alterations in the sensitization of the peripheral and/or central pain pathway [16, 17].

Several pilot studies have described miRNAs as potential peripheral biomarkers of migraine disease [18,19,20,21,22,23,24,25,26]. Nevertheless, studies are scarcely comparable and reproducible not only because of the lack of standardized protocols, biological matrix or sample size but mainly because of the heterogeneous population and clinical variability included. Increasing evidence shows that disease-related intrinsic factors altogether with lifestyle factors may influence epigenetic mechanisms and identifying them is crucial to find unbiased reliable factors associated with a specific disease [27,28,29].

Hence, the goal of this exploratory study is to find differentially expressed miRNAs (DE miRNA) in PBMC of patients with migraine compared to healthy controls in a well-controlled homogeneous non-menopausal female cohort.

Methods

Participants and protocol approval

The study was approved by our Institutional Review Board (PR(IR)224/2016). All participants provided written informed consent before participation. We recruited patients treated for migraine in the Headache Outpatient Clinic. Patients had a diagnosis of migraine with or without aura, confirmed by a neurologist according to the International Classification of Headache Disorders, 3rd edition (ICHD-3) [3]. Healthy controls (HC) were recruited among hospital staff and non-related acquaintances of patients who denied past or first-degree familial history of any recurrent primary or secondary headache disorder. To diminish confounding factors, the inclusion criteria were: female sex, 18–45 years old (non-menopausal) and Mediterranean ethnicity for all participants. Moreover, patients could not have received prophylactic treatment for migraine prevention or if they received it, it had been done for a brief period of time more than one year or more before the study and without following national guidelines of effective dosage and duration. Due to the strict inclusion criteria, participants were included from August 2016 to September 2017.

As miRNAs can be modulated by different lifestyle habits and disease characteristics, recruited participants completed a very thorough and complete questionnaire and structured-interview to collect demographic information including: age, sex, ethnicity, educational level, marital status, gross income; clinical information including: habits of caffeine, tobacco and drugs consumption, physical activity level using the International Physical Activity Questionnaire (IPAQ-SF) [30]; life-time comorbidities, medications use, menstrual status and migraine information including disease duration, acute medication use, monthly headache frequency and family history of headache. We registered headache pain, accompanying symptoms and menstrual cycle the day of the blood sample extraction. We only analyzed samples from migraine patients with at least 48 h of a headache-free period prior day of the blood extraction. In addition, all patients had to fill in the following scales: Allodynia Symptom Checklist (ASC-12) [31], Headache Impact Test-6 (HIT-6) [32], Migraine Disability Assessment Test (MIDAS) [33], Beck Depression Inventory 2nd Edition (BDI-II) [34], State-Trait Anxiety Inventory (STAI) [35] and Perceived stress scale (PSS) [36]. Patients were asked to complete headache diary 30 days before extraction to confirm migraine diagnosis (screening period). We also excluded patients who fulfilled acute medication overuse headache criteria during the screening period.

Blood extraction, PBMC isolation and RNA extraction

For blood extraction patients were asked to fast for 6 h, minimize their exercise/activity and not take analgesic medication 24 h prior to blood extraction. Peripheral blood was collected by antebrachial vein puncture using three, 8 mL, BD Vacutainer® CPT™ Cell Preparation Tubes with Sodium Citrate (BD, Franklin Lakes, NJ, USA) and processed within 2 h following the manufacturer’s instruction to isolate PBMC pellets that were immediately stored at -80ºC till miRNA extraction. To minimize variability due to manipulation, the miRNA extraction was made by the same laboratory technician in the same facilities. For miRNA extraction, we used the Qiagen miRNeasy Mini Kit (Qiagen, Valencia, CA, USA), following the manufacturer’s instruction. RNA concentration was measured using Nanodrop 2000 (NanoDrop Technologies, Wilmington, DE). RNA integrity number (RIN) was determined the kit Agilent RNA 6000 Nano by the Bioanalyzer Agilent 2100 (Agilent Technologies, Santa Clara, CA, USA). We excluded from analysis samples with RIN under 6.

Microarray hybridization

Microarrays service was outsourced and carried out in at High Technology Unit at Vall d’Hebron Research Institute (HTU-VHIR), Barcelona (Spain). For this study, we used the Affymetrix GeneTitan microarray platform and the Genechip miRNA array plate 4.1. Starting material was 550 ng of total RNA of each sample. MiRNA in the sample was labelled using Flash Tag Biotin HSR RNA Labelling kit following the manufacturer’s instructions.

Statistical analysis

All statistical analysis were conducted in R v4.2.0 [37] and figures were produced using the package ggplot2 v.3.3.6 [38]. A statistical power calculation was not conducted prior to the study because the sample size was based on the available data for this exploratory analysis. However, to increase the likelihood of finding differential miRNA expression profiles for this study, we selected participants from a well-controlled homogeneous non-menopausal female cohort. P-values < 0.05 were considered as statistically significant and are reported for a two-tailed test.

Clinical analysis

Nominal variables (aura and presence of allodynia) were reported as frequencies (percentages) while median and interquartile range (IQR) were reported for quantitative variables: age, BMI, IPAQ-SF, caffeine and alcohol consumptions, STAI, BDI-II, PSS, disease evolution time, ASC-12, monthly headache frequency (MHD), monthly acute medication intake (MAMI), MIDAS and HIT-6. Normality assumption of quantitative variables was checked through visual methods (Q-Q plots) and normality tests (Shapiro–Wilk test). Statistical significance between study groups (HC vs. patients with migraine) was assessed by unpaired t-test for BMI, caffeine consumption, STAI and PSS was used and unpaired Wilcoxon rank-sum test was used for the other quantitative variables that did not follow any normality assumption.

Bioinformatic analysis

The statistical analysis of the Genechip miRNA array plate 4.1 was performed using libraries developed for the microarray analysis in the Bioconductor Project [39]. At the moment of the analysis, there were no specific packages for miRNA 4.1 arrays available, thus we programmed our own package. Quality control (QA) was assessed with the package arrayQualityMetrics before the normalization process [40]. Additionally, principal components analysis (PCA) and hierarchical clustering analysis (HCA) were performed to study samples distribution and detect technical problems, including batch effect. Each array (CEL file) was preprocessed and background corrected, normalized and summarized using RMA (Robust Multiarray Average) through Bioconductor packages [41, 42].

The identification of differentially expressed miRNA (DE miRNAs) was based on adjusting linear model with empirical Bayes moderated t-statistic, conducted by the limma package [43] in the comparison between controls and patients with migraine. Models were adjusted by covariates using surrogate variables estimation [44]. DE miRNAs that had more than 0.5-fold increase/decrease relative to controls were chosen for further analysis. The minimum of 0.5-fold expression difference cut-off has been chosen based on the several miRNA profiling studies using microarrays; confirming that subtle changes in miRNA expression, such as a 0.5-fold difference, have a significant biological impact [45,46,47,48]. Results were further corrected for multi-testing using the Benjamini–Hochberg procedure for control the false discovery rate (FDR) [49]. Features that met both criteria (adjusted t-test and ± 0.5 fold change) were combined in a single matrix of DE miRNAs that underwent classification analysis.

We used Random Forests (RF) models for the classification task, a machine learning method widely used in the microarray analysis [50,51,52]. RF algorithm uses an ensemble of classification trees, internally and randomly constructed using a bootstrap sample of the data. Then, the algorithm merges all constructed tree altogether (forest) in order to get an accurate and stable prediction [53, 54]. Once we obtained the forest, the construction of the classifier is performed by a feature reduction algorithm, where the less important features are successively eliminated and out of the bag (OOB) error is continuously analyzed. To evaluate the stability of the results and the prediction error rate, a bootstrap (0.632 + rule) was run through all the procedure. All classification task and bootstrap were performed through the varSelRF R package [55, 56].

Target genes of selected DE miRNAs were predicted by using the miRDB database through the multimiR package [57] and functional and pathway enrichment analysis was performed using the pathfindR package [58]. PathfindR identifies gene sets that form active subnetworks in a protein–protein interaction network (PIN) in order to identify distinct active subnetworks and then perform enrichment analyses on these subnetworks. For assigning a significance value for each predicted target gene (pathfindR input), all Homo sapiens miRNA-target gene predictions were obtained: 1,954,039 miRNA-gene predictions from all H. sapiens miRNAs listed in the miRDB (2,656 miRNAs). Then, the significance for each predicted gene was computed as the probability of observing a score greater than or equal to the score of this target gene over all H. sapiens miRNA-target gene scores [59]. For genes that were targeted by more than one miRNA, the lowest significance was kept.

Results

Clinical characteristics of the patients

Participant recruitment is represented in Fig. 1. Since August 2016, a total of 76 (49 patients with migraine and 27 HC) participants were recruited. For this exploratory study, RNA was extracted from 69 samples that fulfilled inclusion/exclusion criteria. Among these, 37 showed a RIN < 6 and were discarded. Therefore, 32 participants (12 HC, 20 patients) were considered for the analysis with a median age of 33.2 [28.0, 39.8] years old. No statistically significant differences were found neither in demographical variables nor life-habits measurements between groups. In terms of comorbidities, we only found that patients with migraine presented a statistically significantly greater trait-anxiety score (HC: 31.0 [27.0, 38.2] vs. migraine: 39.5 [37.2, 55.8]; p = 0.007). In patients, median MHD was 10.5 [8.0, 24.2] days/month (d/mo) and median MAMI was 7.5 [4.0, 9.0] d/mo. Main demographic and clinical characteristics are shown in Table 1.

Fig. 1
figure 1

Patients’ recruitment flowchart. We recruited participants from August 2016 to September 2017. HC: healthy controls; RIN: RNA integrity number; QA: quality assessment

Full size image
Table 1 Demographics, life habits, comorbidities and migraine characteristics
Full size table

No statistically significant differences were found in terms of RNA quality (RIN, median [IQR]) between study groups: 6.9 [6.5, 7.8] HC vs. 6.7 [6.3, 7.2] patients; p = 0.420.

MiRNA microarray expression profiling

During microarrays QA, we distinguished two samples (from patients) with higher values of intensity in all their probe sets. Similarly, normalized unscaled standard errors values and relative log expression values for the same two samples diverged from the rest, indicating aberrant expression. Hence, we discarded these 2 samples from further analysis (Fig. 1). Control and migraine samples were perfectly segregated along PCA and HCA (Fig. 2).

Fig. 2
figure 2

Principal component analysis (A) and unsupervised hierarchical clustering (B) of miRNA normalized expression data. Unsupervised hierarchical clustering (B) was computed through euclidian distances between samples. HC: healthy controls; MIG: patients with migraine, PC: principal component

Full size image

Adjusting by STAI-trait values, we found 191 DE miRNA between study groups: 59 over-expressed and 132 under-expressed transcripts in patients (Fig. 3A). Then, RF analysis combined with a feature selecting algorithm was used to build the classifier and to identify the best targets that distinguished between controls and migraine. The OOB for the initial model was 0.067 (91.7% sensitivity, 94.4% specificity) and a 3-miRNA signature was found as best variables without dropping the OBB error substantially: miR-342-3p, miR-532-3p and miR-758-5p. Then, as RF algorithms are deterministic, performing it on different training samples is the only way to generate diversity in the selection. To this end, we performed bootstrapping (+ 0.632) in which the random forest constructed for a certain number of variables was subsampled and compared. The prediction error rate among the bootstrap samples was 0.142. The same 3-miRNA signature were consistently selected (stability) in all the sub-samples, with miR-532-3p being selected in the 60% of cases. Figure 3B shows the heatmap from the selected DE miRNA between groups where the expression from all of them were under-expressed in patients: miR-342-3p (log2FC: -0.551), miR-532-3p (log2FC: -1.109) and miR-758-5p (log2FC: -0.508).

Fig. 3
figure 3

Volcano plot (A) and heatmap (B) of differentially expressed miRNA between healthy controls and patients with migraine. Scattered points from the volcano plot (A) represent miRNAs: the x-axis is the log twofold-change for the ratio healthy controls vs patients with migraine, whereas the y-axis is the log 10 P-value. Colored dots (blue and green) are differentially miRNA statistically significantly over- (right) and under-expressed (left) in patients compared to controls. Blue dots are differentially expressed miRNA with a log twofold-change > 0.50. A heatmap (B) of the 3-miRNA signature selected from the classification algorithm (RF). Columns represent individual arrays, while rows represent specific DE miRNA. The Z-score depicts a measure of distance, in standard deviations, away from the mean. The relative value for each miRNA is depicted by color intensity, with yellow indicating over-expression and red indicating under-expression. HC: healthy controls; MIG: patients with migraine

Full size image

Targets prediction and functional enrichment analysis

Predicted target genes associated with the 3-miRNA signature were retrieved from the miRBD database through the multimiR package. Each target significance was computed according to the prediction score retrieved from the miRDB (see Methods section). Then, a total of 68 target genes were identified: 42 genes from miR-342-3p, 26 genes from miR-532-3p and 9 genes from miR-758-5p. While four common target genes were found between miR-342-3p and miR-532-3p: ATXN7 (ataxin 7), CDKL5 (cyclin dependent kinase like 5), PAPPA (Pappalysin 1) and RRM2 (Ribonucleotide Reductase Regulatory Subunit M2); two common targets were found between miR-342-3p, miR-758-5p: DHRSX (Dehydrogenase/Reductase X-Linked), NUFIP2 (Nuclear FMR1 Interacting Protein 2); and 3 common targets between miR-532-3p and miR-758-5p: DSG3 (Desmoglein 3), PTP4A1 (Protein Tyrosine Phosphatase 4A1), TET3 (Tet Methylcytosine Dioxygenase 3). No common target gene was found for all three miRNAs.

Functional enrichment analysis of miRNA-target genes predicted from the miRDB was performed through active-subnetwork-oriented analysis via pathfindR and a total of 24 enriched KEGG pathways were identified (Table 2). Top pathway was calcium signaling pathway and a total of 9 different target genes were involve in all significantly enriched pathways: CACNA1C, HTR2C, EP300, PTGER4, ID4, RICTOR and EEA1 from miR-342-3p; PRKCA and DTX4 from miR-532-3p and ID2 from miR-758-5p.

Table 2 Enriched KEGG pathways of predicted target from the 3-miRNA differentially expressed signature
Full size table

Discussion

In this exploratory study, we found 191 differentially expressed miRNAs (59 over-expressed and 132 under-expressed transcripts) between controls and patients with migraine in a very homogeneous female cohort, controlled by factors as demographics, life-habits and comorbidities (Table 1, Fig. 3A). After transcripts classification task, we finally selected three DE miRNAs that distinguished study groups: miR-342-3p, miR-532-3p and miR-758-5p. All three DE miRNAs were under-expressed in patients compared to controls (Fig. 3B). Sixty-eight target genes were predicted from this 3-miRNA signature and in the functional analysis, we found that they mainly enriched ion channels and signaling pathways (calcium, notch, TRP, TGF-beta, GnRH, HIF-1, thyroid, oxytocin, mTOR and Wnt), neurotransmitter and hormone homeostasis (aldosterone, serotonin, growth hormone, dopamine and adrenaline), infectious diseases (Hepatitis, Influenza and tuberculosis) and circadian entrainment (Table 2).

MiRNAs may have the potential to accurately diagnose disorders at an early stage. For instance, it has been found that levels of miR-182 in the peripheral blood of patients with gliomas increase with disease grade, which had allowed to researchers to patent a diagnostic kit for early diagnosis of human glioma (patent number: CN101792793A) [60]. This is beneficial for patients in order to receive earlier and more individual targeted treatment therapies, a milestone in migraine since we already know that if it is not adequately diagnosed and treated, migraine frequency increases until disease chronification [61]. Moreover, there is increasing evidence that the later we start treating migraine, less effective is the treatment and more difficult it becomes to reverse the disease [5, 62]. In regards to CNS diseases, specific DE miRNAs have also been identified and validated in order to diagnose multiple sclerosis (US9187785B2) [63], Alzheimer’s disease (EP2733220A1) [64, 65] or major depressive disorder (EP2529222A4) [66]. MiRNAs have also been given considerable attention as a therapeutic targets/agent. In recent years, miRNA-based therapeutic approaches have been developed in CNS diseases, especially pre-clinical studies in neurodegenerative disorders [67], although clinical trials have not started yet [68]. Hence, these patents and previous studies in other diseases highlight that investigating miRNA profiles in migraine will lead us not only to a better understanding of their pathogenesis, but also to help develop diagnostic kits, identify specific cluster of patients, explore disease onset, monitor migraine attacks, personalize therapeutic treatment and revert disease chronification.

Several authors have made an effort in previous studies to identify potential miRNAs as peripheral biomarkers of migraine [12]. Numerous miRNAs have been described associated with pain-free patients with migraine in comparison to healthy controls: miR-382-5p (over-expressed in serum and PBMC) [21, 23]; miR-34a-5p (over-expressed in serum, saliva and PBMC) [21, 24]; miR-27b (over-expressed in plasma) [25]; miR-181a, let-7b, and miR-22 (under-expressed in PBMC) [25]; miR-155, miR-126, and let-7 g (over-expressed in plasma) [18]; miR-30 (over-expressed in PBMC) [69]; miR-5189-3p, miR-3613-5p, miR-99a-3p, miR-542-3p (over-expressed in PBMC) and miR-96-5p (under-expressed) [26]. Results are heterogenous and study-dependent but not only because there are differences in lab procedures, biological matrix or sample size but mainly because of the large degree of heterogeneity between study groups. Published studies only controlled by demographic factors (age and sex) but did not consider neither participants selection nor analysis with an adjustment by covariates other factors such as comorbidities and lifestyle habits. Epigenetics have been established as a molecular link between genetic function and the environment. Hence, epigenetic marks constitute a malleable language by which internal and external factors can influence the genome and, for that reason, it is crucial to be able to separate correlative from causal changes in order to understand and quantify the effect of the epigenetic function in diseases [70]. Nowadays, we have more evidence that lifestyle factors such as diet [71], smoking [72], alcohol [73, 74], stress [75], physical activity [29, 76], treatments [77,78,79] and comorbidities [80, 81] have a great impact on the dynamisms of epigenetic marks. Controlling them will allow us to identify transcripts associated with a disease that are not unmasked for other factors.

Functional analysis showed that targeted genes associated to the 3-miRNA signature enriched pathways have been previously described and associated with migraine pathophysiology. One of the most enriched targets is CACNA1C which encodes for the alpha-1 subunit of a voltage-dependent calcium channel. Calcium channels mediate the influx of calcium ions into the cell upon membrane polarization and they are encoded by different subunits of the CACNA1 gene (CACNA1A-S) [82, 83]. From migraine pathophysiology, we already know that the trigeminovascular system (TVS) can be activated by cortical spreading depression (CSD). Mutations in CACNA1A can disrupt ionic homeostasis, increasing the susceptibility to CSD and inducing migraine attacks [84]. Moreover, a CACNA1A variant has also been associated with migraine with aura in the last migraine association study [10]. So, a down-regulation of CACNA1C gene could be also associated with these ionic disturbances in neuronal synapses; indicating that miR-342-3p has a reliable effect on migraine pathophysiology.

Other enriched pathways codifying for vascular functions which may have and important role in migraine etiopathogenesis were the notch signaling pathway, the inflammatory mediator regulation of TRP channels and the HIF-1 signaling pathway. Notch receptors (Notch1-4) are a family of type-1 transmembrane proteins which are essential for embryonic development. Mutations in the Notch pathway genes, like NOTCH2 and JAG1, are associated with vascular abnormalities [85]. In migraine, there is evidence that distinct polymorphisms of NOTCH4 gene significantly modify clinical characteristics of migraine phenotype [86]. The Transient Receptor Potential (TRP) channels are a superfamily of integral membrane proteins well-described in pain conditions that functions as cation channels: TRPA, TRPC, TRPM, TRPP, TRPL, and TRPV [87,88,89]. These channels contribute to a several physiological processes ranging from thermosensation and pain to the regulation of Ca2 + levels in the endoplasmic reticulum. TRP channels are also expressed in the meningeal nociceptors and it has been proved that their activation, promotes the release of the neuropeptide calcitonin gene related peptide (CGRP) [90]. The role of CGRP in migraine has been exhaustively described in both preclinical and clinical studies during the last decades [91] and the CGRP antagonism has shown to be efficacious for the treatment of migraine [92,93,94]. However, the mechanism by which CGRP is released during migraine is unclear, and for that reason, TRP channels remain a focus of interest for their potential contribution to attacks [95]. Moreover, from our experience in clinical trials and real-world studies there are patients with migraine who do not respond to treatments, even target-driven one such as anti-CGRP therapies [96], suggesting that CGRP is not the only mechanism driving migraine attacks [97, 98]. In regards to hypoxia-inducible factor 1 (HIF-1), it is a transcription factor that functions as a master regulator of oxygen homeostasis. Since HIF-1α is implicated in neuroprotection and inflammation inhibition, a recent study has demonstrated that roxadustat administration, an anti-anemia medication, ameliorated migraine-like behaviors and inhibited central pain sensitization in nitroglycerin-injected mice, which was mainly mediated by HIF-1α/NF-κB/inflammation pathway [99].

The relationship between hormones and neurotransmitters with migraine is not entirely clear, but we do know it exists and plays a fundamental role in its pathophysiology. On one hand, thyroid hormones, along with PTX3 and, insulin resistance share common mechanisms in the development of endothelial dysfunction [100], a vascular pathophysiological genetically driven mechanism of migraine [101]. Moreover, differentially pattern levels have been found between controls and patients with migraine of the hypothalamic-tuberoinfundibular system (prolactin and growth hormone) [102]. On the other hand, the gonadotropin releasing hormone (GnRH) controls the release of sexual hormones although estrogens can negatively regulate the GnRH from the hypothalamus at the time of the preovulatory surge, suggesting that estrogen could be related to the increase in migraine prevalence in relation with puberty and ovulation in women [103, 104]. Other hypothalamic hormone linked to migraine is the Oxytocin (OT). OT has been used to relieve migraine attacks through intranasal or intravenous administration [105] since OT receptors included in the trigeminal ganglia and neurons can inhibit the release of CGRP [106].

Altogether, the 3-miRNA signature found could be a reliable biomarker for migraine diagnosis not only because they are differentially expressed between female patients and controls, but they are also involved in enriched biological processes which have been previously described in migraine pathophysiology. So, the under-expression of these transcripts may cause a dysregulated expression of genes associated with the vascular dysfunction and neurohormonal imbalances characterized by patients with migraine.

This study has limitations. First, RNA-seq is the most commonly used NGS technique to explore differences at transcript level to date. RNA-seq is able to detect novel transcripts with higher specificity and sensitivity. However, at the moment of designing and planning the current study, arrays where the more standardized technology. In addition, the RNA-Seq procedure has a few disadvantages compared to arrays: a lack of optimized and standardized protocols for analysis [107, 108]; the size of RNA-Seq files is considerably larger; and RNA-seq requires highly intensive and expensive computation infrastructure and analytics, as well as, longer analysis times [108, 109]. Nevertheless, arrays technologies are still a reliable technology to study gene expression with results comparable to RNA-seq [110]. Second, we did not evaluate the hypothesis that the miRNA profile could be related to the phenotype of circulating PBMC phenotype, rather than a differential miRNA expression. Several changes in PBMC profile related to migraine have been reported but, as different methodologies were used, the overall results are inconclusive. For instance, a meta-analysis of studies from 1966 to 1999 addressing immunological changes related to migraine did not find clear alterations in circulating leukocytes or other immunological parameters [111]. More recently, a relative lymphocytosis has been reported in chronic migraine associated with medication overuse headache, that was considered probably related to a chronic inflammatory state and stress response rather than specifically to migraine [112, 113]. Finally, the sample was small, and highly selected and homogeneous, only considering non-menopausal women in order to reduce sample size and confounding factors. Moreover, the group of individuals diagnosed with migraine was not totally homogeneous in terms of disease pathophysiology (e.g., either episodic or chronic migraine and either the presence or absence of aura). Hence, this 3-miRNA signature should be validated in a larger sample size and more heterogeneous cohort. However, data gathered in this exploratory study will be used to calculate the required sample size for the subsequent discovery and validation (RT-qPCR) studies. Due to lack of power in small sample size experiments, true effects are often missed and many of the detected effects will probably not be further validated. Therefore, the objective of this study was to to perform an exploratory study using the same in- and exclusion criteria for a miRNA high-throughput screen [114].

Conclusion

A 3-miRNA signature (miR-342-3p, miR-532-3p and miR-758-5p) has been found differentially expressed between controls and patients with migraine in a very homogeneous cohort of non-menopausal women, controlled by factors such as demographics, life-habits and comorbidities. Enrichment analysis show that these three miRNAs are closely associated with migraine pathophysiology, maybe becoming the first reliable biomarker of disease. Further studies will be performed to validate them in a larger and more heterogeneous sample.