Upregulation of inflammatory gene transcripts in periosteum of chronic migraineurs: implications to extracranial origin of headache

Abstract

Objective

Chronic migraine (CM) is often associated with chronic tenderness of pericranial muscles. In fact, a distinct increase in muscle tenderness prior to onset of occipital headache that eventually progresses into a full blown migraine attack is common. This experience raises the possibility that some CM attacks originate outside the cranium. The objective of this study was to determine whether there are extracranial pathophysiologies in these headaches.

Methods

We biopsied and measured the expression of gene transcripts (mRNA) encoding proteins that play roles in immune and inflammatory responses in affected (i.e., where the head hurts) calvarial periosteum of (a) patients whose CMs are associated with muscle tenderness and (b) patients with no history of headache.

Results

Expression of proinflammatory genes (e.g., CCL8, TLR2) in the calvarial periosteum significantly increases in CM patients attesting to muscle tenderness, whereas expression of genes that suppress inflammation and immune cell differentiation (e.g., IL10RA, CSF1R) decreased.

Interpretation

Because the up-regulated genes were linked to activation of white blood cells, production of cytokines, and inhibition of NFKB, and the down-regulated genes linked to prevention of macrophage activation and cell lysis, we suggest that the molecular environment surrounding periosteal pain fibers is inflamed and in turn activates trigeminovascular nociceptors that reach the affected periosteum through suture branches of intracranial meningeal nociceptors and/or somatic branches of the occipital nerve. This study provides the first set of evidence for localized extracranial pathophysiology in chronic migraine.

INTRODUCTION

Approximately 5% of the population suffers daily or near daily headache ^{1, 2}. When the headaches include migrainous features, they are diagnosed as chronic migraine (CM) and when migrainous features are absent and the headaches are not attributed to medication overuse, they are diagnosed as chronic tension-type headache (CTTH) ^3–5. Regarding migraine, current thinking suggests that attacks are initiated centrally, in brain areas capable of generating the classical neurological symptoms of premonitory symptoms ⁶ and aura ⁷, that the headache phase begins with consequential activation of meningeal nociceptors at the origin of the trigeminovascular system ^{8, 9}, and that the chronification of disease involves structural and functional alterations in multiple cortical and subcortical regions ¹⁰. Regarding TTH, little is known about the origin of disease, whereas chronification is attributed to sensitization of central pain pathways ¹¹. While some clues exist to how intracranial pathophysiologies such as aura can alter the molecular environment in the meninges to an extent that is sufficient for activation of trigeminovascular nociceptors ¹², nothing is known about the mechanisms by which extracranial pathophysiologies such as chronic pericranial muscle tenderness initiate migraine or tension-type headache.

Pericranial muscle tenderness is common in both CM ^13–16 and CTTH ^{13, 14, 17}. Patients whose pericranial muscles and soft tissues are tender, may also perceive their headache as a tight vice around their head ^18–20 and identify the pain as originating outside, rather than inside, their cranium – a perception termed imploding headache ²⁰. Whereas in migraine, muscle tenderness (and potentially imploding headache) are considered secondary to the headache itself, in TTH they have been considered primary - postulated to originate in activation of pericranial nociceptors. Regardless of whether muscle tenderness is primary or secondary to the headache, the possibility that some migraine attacks and TTHs originate outside the head depends on identification of reasons for why pericranial nociceptors become activated in these patients, and what activates them.

Attempting to determine whether or not there are extracranial pathophysiologies in chronic migraineurs, we sought to measure the expression of gene transcripts (mRNA) that play roles in immune and inflammatory responses in the periosteum of CM patients whose headaches are associated with muscle tenderness and imploding headache. We focused on inflammatory genes because anti-inflammatory drugs are effective in treating migraine ^{21, 22}, and in mildly delaying the progression from episodic to chronic headaches ^{23, 24}. We focused on the calvarial periosteum because it is the tissue to which pericranial and neck muscles are inserted to, because muscle neck aches can originate in their tendons ^{25, 26}, and because it is commonly included in the perception of imploding headache.

MATERIAL AND METHODS

Study design

To achieve the goals of this study, we collected calvarial periosteum tissue from patients undergoing clinically indicated surgeries. The periosteum is an irregular connective tissue consisting of fibroblasts and osteoblasts. The tissues were de-identified to protect subjects’ privacy, frozen, processed and analyzed for targeted transcriptome involved in inflammatory and immune responses. All aspects of this study were carried out in compliance with 1983 revision of the 1975 Helsinki Declaration, and according to the ethical standards of the Western Institutional Review Board (Study #1122419, WO #1-699226-1) and the Beth Israel Deaconess Medical Center (BIDMC) Committee on Clinical Investigation on Human Experimentation (2012-P-000097/4).

Participants

Selection of headache patients

Included in the study were patients who (1) met the criteria of the International Headache Classification Committee for chronic migraine⁵, and (2) were deemed as appropriate candidates for an occipital nerve decompression surgery. These patients were identified by Dr. Pamela Blake, a UCNS-certified Headache Specialist, at the Headache Center of Greater Heights, Memorial Hermann Greater Heights Hospital and referred to Dr. Perry for the surgical decompression procedure. Those deemed good candidates for the surgery were presented with an option to hear about the study and sign the informed consent. Criteria for selecting patients for nerve decompression surgery included: (1) diagnosis of chronic daily headache with or without migraine symptoms, (2) pain that corresponds to the anatomic distribution of the occipital nerves, (3) tenderness to palpation of the occipital nerves and nuchal musculature, and (4) refraining from taking medications and/or herbal supplements that affect blood clotting cascade (e.g., aspirin, coumadin) and platelet functions (e.g., NSAIDs and steroids) for 10–14 days prior to surgery (Table S1). Excluded from the study were patients with medical (cardiac or pulmonary) conditions that increase risk of anesthesia, and those with significant psychological comorbidity. These included depression (anhedonia, sadness, hopelessness) or anxiety requiring psychological and/or psychiatric treatment prior to onset of CM, and potential somatic symptom disorders such as irritable bowel syndrome, fibromyalgia, and interstitial cystitis. (screened using the PHQ-9 and clinical interviews).

Selection of control subjects

Control subjects were selected (1) if they underwent clinically indicated deep brain stimulation surgery for the treatment of Parkinson’s Disease (PD), and (2) if they attested to have no personal or family history of migraine or any other type of headache. As with all surgeries, they were also asked to refrain from taking medications and/or herbal supplements that affect inflammation and/or coagulation.

Nanostring

The targeted transcriptome profiling of tissues were done with the advanced nanostring technologies. Nanostring is a polymerase-free and amplification free nucleic acid quantification platform based on hybridization chemistry that gives it advantage over existing technologies; a lack of enzymatic bias and direct digital readout. The method involves mixing RNA with pairs of capture and reporter probes tailored to each gene, hybridizing, washing away excess probes, immobilizing probe-bound genes on a surface and scanning color-coded bar tags on the reporter probes to calculate expression level or copies of target genes in solution. This solution-phase hybridization result in minimizing background signal and improving detection of low-abundance genes, providing higher sensitivity (<1 mRNA per cell). This platform is capable of detecting as little as 0.5 femtomolar of a specific mRNA, making it a valuable tool for expression, signature generation, and validation in translational studies (which are often limited by the very small amounts of clinical material available).

Tissue preparation

Frozen tissues were homogenized using automated cell homogenizer for extraction of RNA. RNA was extracted using total RNA preparation kit from Qaizol. RNA quantity and quality was determined using Aglient Bioanalyzer. Only high quality RNA, RNA integration score (RIN) > 7, were used for transcriptome profiling.

Targeted transcriptome profiling

The targeted transcriptome profiling was performed by measuring a panel of 540 human genes related to inflammation and immune responses. The profiling was performed using standard assays from nanostring technologies. A 100 ng of high quality RNA was used for labeling and hybridization. The hybridized chips were scanned to count copies of fluorophore linked to each gene.

Quality controls (QC) analysis

Quality measure of each sample was based on Field of View (FOV) counted, binding density, array mean signal, average background, and scaling factor. Reproducibility of the samples was tested using correlation and signal-to-noise ratio (SNR) methods for replicate arrays. Low quality arrays were excluded from further analysis.

Normalization and preprocessing analysis of transcriptome data

Normalization was performed for removing technical noise and making different chips comparable. The normalization was performed in three steps: (i) Positive Spike in controls based normalization that took care of signal variation due to different amounts of input RNA, (ii) Negative Spike in controls based background correction, (iii) Normalization based of signal values of house keeping genes from the experiments which brought all chips to comparable expression level. The different normalizations and QCs were performed using the Bioconductor R packages workflows ^{27, 28}.

Unsupervised analysis

To identify outlier arrays and batch effects, unsupervised learning techniques were performed on the normalized expression profiles of all genes. The unsupervised learning was carried out using the Unweighted Pair Group Method with Arithmetic Mean (UPGMA), also referred to as hierarchical clustering and principal component analysis (PCA) ²⁹. Outlier samples were excluded from further supervised analysis.

Statistical identification of differentially expressed genes

Statistical identification of differentially expressed genes was performed using the Linear Models for Microarray Data (LIMMA) as it employs an empirical Bayes approach for parameter estimation ³⁰. To reduce false positive signal, all confounding factors (e.g. age) were adjusted in linear models to identify only those genes associated with CM. Multiple testing were addressed by controlling the false discovery rate (FDR) using the correction of Benjamini and Hochberg³¹. Genes with multiple test corrected P value < 0.01 and absolute fold change of at least 2 folds were identified as differentially expressed in the periosteum of CM patients.

Gene ontology (GO) enrichment analysis

To identify over-represented GO categories in differentially expressed genes, we used the Biological Processes and Molecular Functions Enrichment Analysis available from the Database for Annotation, Visualization and Integrated Discovery (DAVID)³². DAVID is an online implementation of the EASE software that produces a list of over-represented categories using jackknife iterative re-sampling of the two-tailed Fisher exact probabilities ³³. Here, a p-value gets assigned to each category on the basis of enrichments. Smaller p-values reflect increasing confidence in over-representation. The GO categories with p-values <0.01 and at least 3 genes were considered significant.

Pathway, functional and Interactive Network analysis

Ingenuity Pathway Analysis (IPA 8.0, Qiagen) was used to identify the pathways and functions that were significantly affected by genes associated with CM. The knowledge base of this software consists of functions, pathways and network models derived from systematically exploring the peer reviewed scientific literature. A detailed description of IPA analysis is available at the Ingenuity Systems’ web site (http//www.ingenuity.com). It calculates a p-value for each pathway/function according to the fit of users’ data to the IPA database using right-tailed Fisher exact test. The pathways with p-values <0.01 were considered significantly affected.

To grasp the pathways cross-talk and generate a global system level understanding of alterations due to CM, we performed interactive network analysis. For each network, IPA calculates a score derived from the p-value of right-tailed Fisher exact test [score =-log (p-value)] and indicates the likelihood of focus genes appearing together in the network due to random chance ^34–38. A score of 2 or higher has at least a 99% probability of not being generated by random chance alone. The ability to rank the networks based on their relevance to the queried data sets allows for prioritization of networks with the strongest association to CM.

Identification of CM specific genes by removing PD associated changes

To make sure that differentially expressed genes identified in CM patients were not associated with PD, we performed gene specificity analysis. This analysis was performed on a transcriptome dataset (GSE57475, http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE57475) containing samples from PD and healthy controls. The analysis identified list of genes associated with PD (i.e. PD genes). The PD genes was compared with CM genes to determine any overlap between them. The comparison was performed on the basis of universal gene symbols.

Comparative analysis of CM genes with Lovadopa administration associated genes

To determine the effect of Lovadopa administration on expression of CM associated genes, we performed an external transcriptome analysis by archiving microarray data (GSE39980) from public repository at NCBI ³⁹. The dataset contains the transcriptome data from Parkinson’s Disease rats treated with zonisamide plus Levadopa, Levadopa alone, and saline. Supervised analysis of the transcriptome data from the levodopa alone and saline groups was performed using LIMMA to identify differentially expressed genes. Significantly altered gene due to Levadopa administration were identified on basis absolute fold change ≥ 2 and a P value <0.05. Comparative analysis of CM and Lovadopa associated genes was performed using universal gene symbols.

Comparative analysis of CM genes with isoflurane anesthesia administration associated genes

To determine the effect of isoflurane anesthesia on expression of CM associated genes, we performed transcriptome analysis on publically available microarray datasets (GSE1779, GSE358) ⁴⁰. These datasets contain tissues obtained from animals exposed to prolonged and repeated isoflurane anesthesia and controls. Genes associated with prolonged and repeated isoflurane were identified by performing comparison of control samples with isoflurane exposed samples using Limma approach. The significantly differentially expressed genes were selected on the basis of Fold change and P value cutoff. Comparative analysis of CM and isoflurane anesthesia signatures was performed using universal gene symbols.

Comparative analysis of CM genes with antibiotic modulated genes

To determine whether the cephalosporin (cephaloridine) antibiotic effected the expression of any of the 37 CM gene signature, we performed analysis on publically available microarray datasets (GSE10034) ⁴¹ using the methods described in previous sections. Briefly cephaloridine signature was generated by comparing transcriptome profile of antibiotic and controls groups on basis of fold change and P value.

Comparative analysis of CM genes with fasting signature

To determine the effect of fasting, we generated a fasting signature by comparison of transcriptome profile of non-fasting (0H) vs. Fasting (12H, 24H, 48H and 72H) mice muscle samples ⁴². The comparison of normalized transcriptome profile was performed using LIMMA approach. Significantly differentially expressed genes were obtained on the basis of P value and Fold change. Comparative analysis of CM and fasting signatures was performed using universal gene symbols.

Biomarker identification, predictor development and validation using linear and artificial intelligence based approaches

The differentially expressed genes from the supervised transcriptome analysis were used for bioinformaker identification of CM. The classifier development was performed using linear and non-linear classfier approaches such as linear discriminant analysis (LDA), support vector machine (SVM), and nearest neighborhood (KNN) based method. The classifier were training on normalized expression data and validated using leave one out cross-validation approaches. Validation performance was measured using threshold dependent (e.g. sensitivity, specificity, accuracy, positive predictive value (PPV), Negative predictive value (NPV), and threshold independent approaches (e.g. Area under Curve (AUC)). We developed biomarkers panels ranging from 2–10 gene to identify biomarker panel with most accuracy and minimum number of genes.

RESULTS

Demographics

Patient demographics and headache history are shown in Table 1. They were 15–59 y/o with an average of 4.6 years (range 1–15) of daily headache. Most (61%) had a family history of migraine and their headaches started as episodic and gradually became daily. Based on the International Classification of Headache Disorders (3^rd edition), 15/18 patients were diagnosed with CM. The remaining were diagnosed with probable migraine. Relevant to our study, 94% testified to having chronic tenderness of neck and occipital muscles, and 83% described their headache as imploding. Control subjects were 60 y/o (range 19–77) and had no history of headache (Table 2). They were PD patients that underwent deep brain stimulation (DBS) procedure for alleviation of stiffness and tremor.

Table 1.

Characteristics of participants with daily headache

Characteristics	n (%)	Participants
Identification no.		1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18
Age (years), mean (SD)	38 ± 13	15	35	23	20	25	28	30	37	32	40	52	53	59	42	47	61	44	35
Sex
Female	12 (66)		✓		✓	✓		✓	✓	✓				✓	✓	✓	✓	✓	✓
Male	6 (33)	✓		✓			✓				✓	✓	✓
Months of daily headache, mean (SD)	56 ± 48	12	11	55	26	21	119	58	180	19	74	66	150	36	15	19	48	36	60
History of episodic headache	11 (61)	✓	✓	✓		✓			✓		✓		✓	✓	✓	✓		✓
Family history of migraine	11 (61)	✓		✓	✓	✓		✓		✓	✓		✓			✓	✓		✓
Sudden onset of CM	7 (39)				✓		✓	✓		✓		✓					✓		✓
Headache characteristics
Unilateral	3 (17)							✓		✓							✓
Bilateral	15 (83)	✓	✓	✓	✓	✓	✓		✓		✓	✓	✓	✓	✓	✓		✓	✓
Frontal	3 (17)				✓	✓		✓			✓							✓
Temporal	3 (17)	✓	✓			✓	✓						✓		✓	✓		✓
Occipital	15 (83)	✓	✓	✓		✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
Photophobia	15 (83)	✓			✓	✓	✓	✓	✓	✓	✓		✓	✓	✓	✓	✓	✓	✓
Phonophobia	12 (67)	✓	✓		✓	✓	✓	✓	✓	✓	✓	✓			✓				✓
Osmophobia	8 (44)	✓			✓	✓	✓	✓	✓				✓		✓
Nausea/vomiting	11 (61)			✓	✓	✓	✓	✓	✓	✓	✓			✓	✓	✓
Throbbing	7 (39)	✓			✓		✓		✓	✓			✓				✓
Activity worsens headache	14 (78)	✓	✓		✓			✓	✓	✓	✓		✓	✓	✓	✓	✓	✓	✓
Severity
Mild	3 (17)							✓							✓		✓
Moderate	14 (78)	✓	✓		✓	✓	✓		✓	✓	✓	✓	✓	✓		✓		✓	✓
Severe	1 (5)			✓
Aura	4 (22)						✓		✓						✓			✓
Eye tearing	2 (11)						✓	✓
Nasal congestion	2 (11)						✓	✓
Allodynia	11 (61)		✓		✓	✓		✓	✓	✓	✓			✓	✓	✓	✓
Muscle tenderness	17 (94)	✓	✓	✓	✓	✓	✓		✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
Imploding headache	15 (83)	✓	✓	✓	✓	✓	✓	✓			✓	✓		✓	✓	✓	✓	✓	✓
Exploding headache	3 (17)								✓	✓			✓
Headache classification
Chronic migraine	15 (83)	✓	✓		✓		✓	✓	✓	✓	✓	✓	✓	✓		✓	✓	✓	✓
Probable migraine	3 (17)			✓		✓									✓
Treatment
Response to triptans
(Good, Moderate, None)	5 (28)	N	M	N	N	N	N	N	N	N	M	N	M	N	G	M	N	N	N
Response to NSAIDs	6 (33)	N	M	N	N	N	M	N	N	N	N	M	N	M	N	M	M	N	N
Response to Botox	0 (0)	na	na	na	N	na	na	na	N	na	N	na	na	na	N	na	na	N	N
# of preventative tried		1	3	5	4	4	5	4	4	4	4	0	3	1	4	2	4	9	4

N = none, M = mild, G = good

Table 2.

Characteristics of control participants

Characteristics	n (%)	Participants
Identification no.		1	2	3	4	5	6	7
Age (years), mean (SD)	60 ± 20	19	68	54	77	76	62	65
Sex
Female	2 (41)	✓		✓
Male	5 (59)		✓		✓	✓	✓	✓
Condition
Parkinson’s disease	7 (100)	✓	✓	✓	✓	✓	✓	✓
Headache history	0 (100)

Quality control and unsupervised analysis of transcriptome data

Twenty five processed periosteum samples (18 headache patients and 7 control subjects) yielded high quality RNA (RNA integration score > 7). The unsupervised clustering (Figure 1A) and the principle component analysis (PCA, Figure 1B) of all 524 genes showed that the immunological transcriptome profile of the periosteum of the 18 CM patients differs significantly from that of the 7 control subjects.

Figure 1.

Targeted transcriptome analysis of periosteum tissue from patients with chronic daily headache/chronic migraine and Controls. A) Unsupervised clustering of transcriptome data from CM patients and controls B) Principle component analysis (PCA) of gene expression profiling data of periosteum tissue of CM and control subjects (unsupervised analysis). This analysis shows clusters for controls (green), and CM subjects (red). Note the clear separation along principle component 1. C) Heat map of 36 significantly differentially expressed genes (multiple test corrected p value <0.01 and absolute fold change >2) identified in the periosteum of CM patients compared to control subjects using supervised analysis. Heatmap rows depict differentially expressed genes and columns depict individual control subjects and CM patients. The relative expression level of genes is depicted using a pseudocolor scale from −3 to +3 (green represents down regulation and red represents up regulation). D) Functional enrichment analysis of genes that are differentially expressed in periosteum from CM patients compared to controls. This analysis depicts up regulation of genes linked to inflammatory response and immune cell trafficking as well as down regulation of genes linked to cell death or survival and cellular development. The activation (orange) and inhibition (blue) of functional categories is computed by calculating the Z score for each function, a formula representing the concordant up and down regulation of genes related to such a function. E) Pathway enrichment analysis of genes that are differentially expressed in periosteum from CM patients compared to controls. This analysis depicts the significance of the effect of CM on key inflammation-related pathways that include IL-10, IL-6, NFkB and glucocorticoid receptor signaling.

Identification of genes significantly associated with CM

A linear modeling approach was used to perform the supervised analysis of CM and control tissues. After adjusting for age, we identified 37 differently expressed genes (multiple test corrected p value<0.01, absolute fold change >2) in the periosteum of the CM vs. the control groups (Figure 1C). Of these, 26 genes were upregulated and 11 genes were downregulated. Most of the upregulated genes encode proinflammatory proteins (e.g. IL6, SOCS3, CCL2) whereas most of the downregulated genes encode cell death, apoptosis and osteoclast differentiation (e.g. CD14, TNFSF10, CYBB, CSF1R).

Functional enrichment analysis

Functional enrichment analysis of the differentially expressed genes in CM periosteum revealed dysregulation of genes linked to hematological system development and function (BH p value = 4.94E-24, Selected Genes = CCL8, SOCS3, IL6, CX3CL1, TLR2), cellular movement (BH p value = 2.57E-20, Selected Genes = CCL8, SOCS3, IRAK3, TLR2, NFKBIZ), cell-to-cell signaling and interaction (BH p value = 2.57E-24, Selected Genes = CLEC4E, CXCR4, IL1R2, CCL8, NFATC2), inflammatory response (BH p value = 6.1E-23, Selected Genes = ABCB1, IL6, CLEC4E, IRAK3, NFKBIA, CXCR4, IL1R2, IFNB1), and immune cell trafficking (BH p value = 6.1E-23, Selected Genes = ABCB1, SOCS3, IL6, CSF3R, CXCR4, KIR2DL3 (Figure 1D).

Pathway enrichment analysis

Pathway enrichment analysis of differentially expressed genes in CM periosteum identified 12 pathways that were significantly (p<0.0001, multiple test corrected) affected (selected pathways are shown in Figure 1E). These include pathways linked to cell adhesion (granulocyte adhesion and diapedesis), innate immune activation (Toll-like receptor signaling, TREM1 signaling), interleukins (IL-10 signaling, IL-6 signaling, IL-17A signaling) and inflammatory response (NF-κB signaling and glucocorticoid receptor signaling). Further directionality analysis (activation/inhibition) of significantly effected pathways depicted significant activation of IL6, Toll like receptor (TLR) signaling, and TREM1 signaling, and significant inhibition of NF-κB and LXR/RAR signaling pathways.

Systems biology interactive network

Interactive network analysis of genes that are differentially expressed in CM periosteum generated a very cohesive interaction network in which >95% of the genes interact with each other (Figure 2A). The network is significantly associated with activation of immune response and inflammatory processes (P value = E-29). The analysis identified the following molecules as key drivers in the inflammatory pathophysiology of the periosteum: IL6, SOCS3, IFNB, CXCR4, CCL2, and NFKBIA. The mathematical analysis of the network – intended to identify a key regulatory molecule – showed that IL6 interacts with the largest amount of analyzed regulatory molecules and as such, is critical for the stability and functioning of the network. Complimentary regulatory analysis of differentially expressed genes also depicted significant activation of downstream targets of IL6 and TLR2 - indicating their potential role in driving the headache pathophysiology (Fig 2B–C).

Figure 2.

A) Interactive network-based system biology analysis of genes that are differentially expressed in periosteum from CM patients compared to controls. This analysis depicts a very cohesive interaction (p=e10⁻³⁰) in between all the molecules that are dysregulated in the CM patients. In this network, each node represents a gene and each arrow represents an interaction between genes. The red and green colors depict up and down regulation of genes, respectively, in the CM patients. B–C) Interactive network of TLR2 and IL6, the two top activated key regulators found to be significantly upregulated in CM patients compared to controls. In each of these networks, the central molecule is a key regulator (orange=activated, blue=inhibited). The red and green colors depict up and down regulation of the target genes by the upstream key regulator.

Biomarkers

To identify biomarker genes, we performed predictor development analysis using linear (e.g. LDA, KNN) and non-linear (SVM) approaches, and validated their performance using a robust leave-one-out cross-validation approach. The analysis identified a predictor based on four genes (CLEC4E - C-type lectin domain family 4, IL1R2 - interleukin 1 receptor type II, NF-κBIA - nuclear factor of kappa light polypeptide gene enhancer in B-cells inhibitor alpha, and TNFAIP3 - tumor necrosis factor alpha-induced protein 3) that are up-regulated in the periosteum of all CM patients but in none of the control subjects (Fig 3A–B). The 4-genes predictor achieved 100% sensitivity and 100% specificity in separating CM patients from controls (Fig 3A). The complete separation between CM patients and control subjects is further supported by the predictor Area Under Curve 1 (Fig 3C). These results are preliminary as they are not validated on independent validation set but appear promising.

Figure 3.

Putative biomarkers associated with CM. A) Performance of 4 genes used as biomarkers on training sets determined using leave-one-out cross-validation (LOOCV). The biomarkers were identified using linear and non-linear approaches including Linear Discriminant analysis (LDA), K Nearest neighborhood method (KNN), and Support Vector Machine (SVM). B) Heatmap of biomarker genes. C) Non-threshold dependent performance of Biomarkers measured using Area Under Curve of Receiver Operative Curve (ROC).

Parkinson’s disease genes

To ensure that the identified 37 genes are associated with CM and not PD, we performed a comparison with a set of genes identified from a publically available PD signature dataset (GSE57475, http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE57475). None of the 37 genes associated with CM are reported to be dysregulated by PD (Fig 4).

Figure 4.

Comparison of CM associated genes with Parkinson disease signature. The Parkinson disease signature was obtained after preprocessing, normalization and supervised analysis of external published dataset (i.e. GSE57475), containing samples from PD and healthy controls. Parkinson disease signature was obtained using stringent (P value <.01, blue circle) and relaxed (P value <.05, red circle) cutoff.

Medications, anesthesia and fasting effects on CM genes

The analysis of the transcriptome data from the levodopa alone and from the saline groups identified 1,719 genes that are significantly altered due to Levadopa administration (absolute fold change ≥ 2 and a P value <0.05). Interestingly, only 3 of these genes overlap with the CM genes (Figure 5A).

Figure 5.

Comparisons of CM associated genes with confounding factors (e.g. medications, anesthesia and fasting) associated gene signatures. Comparison of CM associated genes with signatures of A) Lovadopa, B) Cephalosprin, C) Fasting, D–E) Isoflurane. The signatures for confounding factors were generated by comparing experimental group with control group on the basis of P value and Fold change. The comparative analysis of CM and confounding factors associated genes was performed on basis of universal gene symbols using Venn Diagrams.

As standard of care, all CM and all PD patients received the 15–33 mg/kg β-lactam antibiotic cephalosporin prior to skin incision. The dataset contains tissues obtained 24 hours after animal exposure to toxic doses (150–600 mg/kg) of this antibiotic, and controls. The compartive bioinformatics analyses showed an overlap of 6 genes that are commonly altered in our CM patients and the cephalosporin treated animals (Figure 5B).

Also as standard of care, all CM and all PD patients were not permitted to eat or drink overnight before the procedure. The comparison of non-fasting (0H) vs. Fasting samples (12H, 24H, 48H and 72H) identified 585 significantly differentially expressed genes (i.e. fasting signature). The comparative analysis of CM and fasting signatures depicted an overlap of 3 genes (Figure 5C).

While propopfol anesthesia was administered briefly to all CM and all PD patients, the inhaled desflurane (isoflurane) anesthesia was administered only to the CM patients. The comparative analyses on the basis of gene symbols showed insignificant overlap between our CM genes and the isoflurane altered genes. As shown in Figure 5D–E, only 1 upregulated gene (IFNB1) is common.

DISCUSSION

By using targeted transcriptome analysis to profile periosteal biopsies of CM patients, our primary data showed that patients attesting to chronic muscle tenderness and imploding headache have significantly increased expression of pro inflammatory genes (e.g., CCL8, TLR2) and decreased expression of genes responsible for the suppression of inflammation and immune cell differentiation (e.g., IL10RA, CSF1R) in the calvarial periosteum. Because the up-regulated genes were linked to activation of mast cells, T cells and natural killer cells ^{43, 44}, production of cytokines, cell adhesion, cell-cell signaling, glycoprotein production ⁴⁵ and disruption of proper NF-kB function ^{46, 47}, whereas the down-regulated genes were linked to repression of TNF-alpha, IL-6, and macrophage activation ^{48, 49}, and to prevention of cell lysis through inhibition of NK, T, and B cells ⁵⁰, we interpret these findings as suggesting that the molecular environment in which periosteal pain fibers exist is inflamed and that this localized inflammation can activate, or lower the activation threshold of trigeminal nociceptors that reach the affected periosteum through suture branches of intracranial meningeal nociceptors ⁵¹ and somatic branches of the occipital nerve ⁵². Accordingly, we conclude that the localized extracranial pathophysiology we identified (i.e., periosteal inflammation) in the CM patients should be considered as evidence that some migraine attacks can originate outside the head. To put our study in a proper perspective, we acknowledge that based on the sampled patients it is not possible to know whether similar findings might be present in the acute remitting phase.

Common to the CM patients we studied was that their headaches were predominantly bilateral and occipital. Because migraines are typically unilateral and frontal/periorbital/temporal, our findings must be interpreted with the perspective that periosteal inflammation may be at the origin of headache only in those patients exhibiting chronic bilateral, occipital headaches accompanied by chronic muscle tenderness and the perception of imploding headache. It is worth noting, however, that these two different types of headaches (the chronic occipital and the episodic frontal) could co-exist in the same patient. Although we did not include in this study patients fulfilling diagnostic criteria for TTH, the findings may also be relevant to this class of headaches as they share features such as bilateral occipital pain, muscle tenderness, perception of imploding headache, and aggravation of headache when bending the head down and stretching muscles of the neck ^{4, 5, 15, 53}.

Concerning muscle tenderness and bilateral occipital headache, our findings do not contradict the widely accepted view that in most cases (i.e., migraines that start with prodromes or aura, lack of sleep or changes in levels of ovarian hormones, and attacks associated with no muscle tenderness or muscle tenderness that appears after onset of headache) they are secondary, resultant of pain referral by activated and sensitized second-order trigeminovascular neurons in the spinal trigeminal nucleus ^{54, 55}. In perspective, our results suggest that activation of pericranial nociceptors can be a primary pathophysiology and a common denominator for various migraines and localized muscle tendernesses – involving activation of (a) the trigeminovascular pathway through small branches of meningeal nociceptors that traverse the calvarial sutures and reach the periosteum ^{51, 56, 57}, and (b) primary afferent nociceptors that travel to the occipital periosteum through the occipital nerve.

At the molecular level multiple innate and adaptive immune response pathways were significantly effected by the chronic headache. Of these, IL6, TLR and NF-κB signaling were most notable pathways activated in the transcriptome profile of the CM periosteum. To produce pathophysiology, however, members from different pathways must cross-talk to an extent that creates a biological network. Of all identified members of the biological network we associated with extracranial pathophysiology of CM, IL6 was most interactive, rendering this molecule most critical in stabilizing the headache state. IL-6 signaling is a key adaptive immune pathways for induction of cellular and intracellular signaling cascades for production of inflammatory cytokine in response to trauma and tissue damage ^{58, 59}. It exerts its pathological effects by facilitating cell proliferation, apoptosis and immune responses ⁶⁰. In the past, IL6 has been linked to diseases such as diabetes, lupus, rheumatoid arthritis, stress and cancer ^61–65. The current study support the view that it can also play a role in the pathophysiology of certain CMs ^{66, 67}, such as those caused by mild trauma to the head.

Toll-like receptors 2 (TLR2), which is a membrane receptor, was also observed in the periosteal biopsies of CM patients. Anatomically, TLR2 could be found in both peripheral blood leukocytes⁶⁸ and neurons⁶⁹. When activated in neurons, TLR2 facilitates the production of proinflammatory cytokines in NF-κB dependent (and independent) manner, which leads to the development and maintenance of inflammatory and neuropathic pain ^70–72. Regarding its potential role in headache, it’s worthwhile noting that activation of TLR2 has been linked to meningeal inflammation, influx of leukocytes in the meninges, increased blood flow and intracranial hypertension⁷³ – all capable of triggering headaches of intracranial origin. On a broader note, because TLR activation has been linked to a variety of peripherally-occurring infection and immune diseases such as sepsis, inflammatory bowel disease, atherosclerosis, and asthma ^74–77, it is also reasonable to speculate that this inflammatory pathway may be involved in migraines driven by abnormal localized peripheral inflammatory and immune responses.

Remarkably, our transcriptome data also depicted significant downregulation of genes linked to the NF-κB pathway - a pathway of immense importance in suppression of inflammation ^78–81. We interpreted this findings as suggesting that suppression of NF-κB activation which is critical in resolving the inflammation⁸⁰ (e.g., by regulating macrophage activation – thought to play a role in migraine^{82, 83}), and consequently the pain must be involved in the pathophysiology of the studied CM patients. The data suggest that the suppression of NF-κB is achieved due to significant upregulation of the NF-κB inhibitor NF-κBIA in the patients’ periosteum. Collectively, the findings point to a dual mechanism in the chronification of occipital CM: the first is activation of proinflammatory pathways and the second is suppression of pathways involved in resolving the inflammation.

Other proinflammatory pathways activated in CM periosteum include the granulocytes and agranulocyte adhesion and diapedesis pathways. These pathways regulate the process through which neutrophils, eosinophils, basophils, mast cells, B cells, T cells and natural killer cells - which normally circulate in the blood unattached - adhere to the surface of the endothelium (diapedesis) and pass between neighboring endothelial cells (transmigration) to reach infected/irritated tissues ^84–86. In the context of CM, the invasion of extracranial tissues by these white blood cells is likely to contribute to the formation of an inflamed environment in tissues where the pain fibers exist.

In total, 11 of the 37 CM-associated genes overlapped with the genomic signature of confounding factors influenced by anesthesia, medications and fasting. Of these, 4 were uniquely associated with exposure to toxic doses of antibiotics for at least 24 hours. Given that the CM and PD patients received a much smaller dose of the antibiotics and that the biopsies were obtained within 1 hour from exposure, it is unlikely that the expression of these 4 genes was altered by the antibiotics. The study does not allow us to determine whether the expression of the remaining 7 genes was altered due to CM or the other confounding factors. Furthermore, to remove the effect of confounding factors on migraine associated genes, we had to perform analysis on multiple external transcriptome datasets obtained from tissues other than human periosteum due to lack of transcriptome availability. These include peripheral blood mononuclear cells for Parkinson’s Disease,, brain tissue for anesthesia and levodopa, kidney tissue for antibiotics and muscle tissue for fasting. Since the tissue used to remove the effect of confounding factors were not periosteum, we acknowledge that they cannot be considered as the ultimate control. Accordingly, the validity of our findings should remain an open query until such time as appropriate control tissue (i.e., periosteum) is available.

Finally, the identification of 4 genes that are abnormally expressed in all CM patients but in none of the control subjects may provide novel understanding of the disease as well as a diagnostic tool for occipital headaches associated with muscle tenderness that is localized to the neck. Given that 3 of the 4 biomarkers genes - the NF-κBIA, the TNFAIP3 and the ILR2 - are tightly related to the NF-κB family pathway, and that the first 2 inhibit NF-κB activation ^{87, 88}, it is tempting to propose that these biomarkers point to suppression of normal anti-inflammatory processes common to the pathophysiology of these headaches.