Abstract
Migraine is believed to be initiated by neuronal activity in the CNS, that triggers excitation of nociceptive trigeminal ganglion (TG) nerve fibers innervating the meninges and thus causes a unilateral throbbing headache. Drugs that precipitate or potentiate migraine are known to elevate intracellular levels of the cyclic nucleotides cAMP or cGMP, while anti-migraine treatments couple to signaling pathways that reduce cAMP or cGMP, suggesting an involvement of these cyclic nucleotides in migraine. Members of the HCN ion channel family are activated by direct binding of cAMP or cGMP, suggesting in turn that a member of this family may be a critical trigger of migraine. Here, we show that pharmacological block or targeted genetic deletion of HCN2 abolishes migraine-like pain in three rodent migraine models (in both sexes). Induction of migraine-like pain in these models triggered expression of the protein C-FOS, a marker of neuronal activity, in neurons of the trigeminocervical complex (TCC), where TG neurons terminate, and C-FOS expression was reversed by peripheral HCN2 inhibition. HCN2 block in vivo inhibited both evoked and spontaneous neuronal activity in nociceptive TG neurons. The NO donor glyceryl trinitrate (GTN) caused an increase in cGMP in the TG in vivo. Exposing isolated TG neurons to GTN caused a rightward shift in the voltage dependence of HCN currents and thus increased neuronal excitability. This work identifies HCN2 as a novel target for the development of migraine treatments.
SIGNIFICANCE STATEMENT Migraine is believed to be initiated by localized excitability of neurons within the CNS, but the most disturbing symptom, the characteristic throbbing migraine headache pain, is widely agreed to be caused by activity in afferent pain-sensitive (nociceptive) nerve fibers of the trigeminal nerve. Using a variety of preclinical models of migraine, we identify the HCN2 ion channel as the molecular source of trigeminal hyperexcitability in migraine and we show that pharmacological or genetic inhibition of HCN2 can relieve migraine-like pain symptoms. The work highlights the HCN2 ion channel as a potential pharmacological target for the development of novel analgesics effective in migraine.
Introduction
Migraine, a disabling chronic condition affecting one in seven people worldwide, is a huge unmet clinical need that costs over £2 billion annually in the United Kingdom (Natoli et al., 2010). The pathophysiology driving migraine symptoms, namely sensory disturbances (aura), throbbing headaches, photophobia, phonophobia, and nausea, is largely unknown. The pain of migraine in particular is poorly understood mechanistically and can result in decreased efficiency at the workplace, withdrawal from social activities, and depression (Buse et al., 2009).
Actual or potential promise as therapies in migraine has been demonstrated by the nonsteroidal anti-inflammatory drugs (NSAIDs) family, by the triptan and ditan families of 5HT3 receptor agonists (Cameron et al., 2015), by antagonists to the receptor for calcitonin gene-related peptide (CGRP; Ong et al., 2018) and by monoclonal antibodies against CGRP and its receptors (Tso and Goadsby, 2017; Skljarevski et al., 2018). All have significant disadvantages, including gastric and renal toxicity for NSAIDs, promotion of medication overuse (MO) headaches by extended use of triptans, liver toxicity with some of the early CGRP antagonists, and the expense and need for regular injection of monoclonals. A significant fraction of migraine patients do not achieve relief with any of these treatments (Goadsby et al., 2017b; Tso and Goadsby, 2017; Ong et al., 2018), underscoring the need to develop more targeted treatments.
Some evidence suggests that meningeal vessels may be dilated during a migraine episode, and vasoconstrictors such as triptans are effective antimigraine treatments, observations that led to the “vascular” theory of migraine origin, which proposes that vasodilation of meningeal arteries is the primary trigger of migraine (Jacobs and Dussor, 2016). More recently, a view has gained ground that excitation of meningeal nociceptive (pain-signaling) nerve fibers is instead the primary trigger, while arterial vasodilation is a secondary event caused by release of inflammatory vasodilators, such as CGRP, from nociceptors (Bartsch and Goadsby, 2003; Messlinger, 2009; Olesen et al., 2009; Burgos-Vega et al., 2015). According to this view, augmented activity of trigeminal nociceptive nerve fibers innervating the meninges feeds into ascending central pain pathways and drives migraine pain (Burstein et al., 2015; Goadsby et al., 2017a). The molecular basis of increased trigeminal hyperexcitability, however, remains to be determined.
The HCN ion channel family comprises four isoforms, HCN1–HCN4, that carry an inward current called Ih (also Iq or If; Kaupp and Seifert, 2001). All HCN channel isoforms are activated by membrane hyperpolarization between −60 and –90 mV. The primary isoform driving the cardiac pacemaker potential is HCN4 (Kaupp and Seifert, 2001; Biel et al., 2002, 2009), while HCN2 is expressed in pain-sensing (nociceptive) neurons where its activation plays a crucial role in both inflammatory and neuropathic pain (Emery et al., 2011, 2012; Young et al., 2014; Tsantoulas et al., 2016, 2017). An important property of HCN2 and HCN4 is that their membrane voltage activation range is shifted in the positive direction by direct binding of intracellular cAMP (Kaupp and Seifert, 2001), whose cellular levels are elevated by many inflammatory mediators linked to Gs-type G-protein-coupled receptors (GPCRs). Increases in cellular cGMP also have a similar effect on the activation of HCN2 (Zagotta et al., 2003; DeBerg et al., 2016). The shift in HCN2 voltage dependence promotes opening of the ion channel, thus enhancing inward current flow and promoting action potential (AP) firing in nociceptors. A cyclic nucleotide-dependent increase in HCN2 activity has been documented in several chronic pain conditions (Tsantoulas et al., 2016, 2017). In the present study, we explore whether HCN2 might play a similar role in driving the pain of migraine.
Materials and Methods
Experimental design
The main objective of these controlled laboratory experiments was to test the effect of pharmacological or genetic inhibition of HCN2 ion channels on pain-like behaviors in mice and rats in which migraine-like conditions had been induced. Sample sizes for animal behavior, histology, biochemistry and electrophysiology were chosen based on previous experience with these assays, with the minimum number of independent experiments used to achieve statistical significance at the 5% level. Sample sizes and statistical tests used are described in detail in the corresponding Results sections. All behavioral testing was conducted during the day in a quiet temperature-controlled room by an experimenter blinded to the identity of drug treatment and/or mouse genotype.
Animals
Behavioral experiments on wild-type (WT) animals were conducted using adult male (200–300 g) Sprague Dawley rats, or adult (six to eight weeks old) male C57/BL6 mice (Charles River). For experiments with transgenic mice, we used both male and female NaV1.8-Cre/HCN2−/− (conditional knock-out; cKO) and floxed littermate adults (8–24 weeks old) to maximize cohort sizes. There was no substantial difference in male versus female responses in any of the experiments using transgenic animals (see Results). The f/fHCN2 mice used to generate cKO mice had been backcrossed onto a C57/BL6 genetic background (BG) for at least six generations as described in detail elsewhere (Emery et al., 2011). For in vivo electrophysiology, we used 250–300 g Sprague Dawley rats, while for patch-clamp experiments we used either male C57/BL6 mice (six to eight weeks old) or global HCN2 KO littermates (less than four weeks old, because of the age-dependent lethality of this genotype). All animals were housed under standard conditions (12/12 h light/dark cycle, food and water ad libitum) in a temperature-controlled environment (22 ± 1°C). All experiments conformed to Home Office (United Kingdom) regulations, Animals Scientific Procedures Act (1986) as well as to guidelines from King's College Animal Welfare and Ethical Review Body (AWERB) and the Committee for Research and Ethical Issues of the International Association for the Study of Pain (IASP). Animals were allowed at least one week of acclimatization before behavioral testing.
Migraine models
Acute and chronic GTN model
GTN (Hospira, UK) was diluted in Dulbecco's PBS (PBS; Invitrogen) to the desired concentration. For induction of acute migraine, rats or mice were given a single intraperitoneal injection of GTN (10 mg/kg, injection volume 10 ml/kg). The chronic migraine model was induced by GTN injections every 2 d (10 mg/kg per injection, four injections in total) as noted in the relevant graphs of results. Control animals were subjected to an identical regime of vehicle (PBS) administration.
MO headache (MOH) model
Sumatriptan was diluted in PBS to the desired concentration. For induction of MOH, rats or mice were given a subcutaneous injection of sumatriptan (0.6 mg/kg, injection volume 10 ml/kg) every 2 d (six injections in total as noted in corresponding graphs). Control rats received injections of PBS.
Behavioral testing
Mechanical sensitivity
Rats or mice were acclimatized to the behavioral apparatus, a ventilated Plexiglas chamber on an elevated aluminum screen surface with mesh openings, in the week before experiments (1 h daily × 3 d for plantar and 2 h × 7 d for periorbital measurements), and also on the day of the experiment (at least 60 min). Von Frey (VF) filaments (0.08–26 g) were used to measure facial (application to the periorbital region immediately above the brow) or plantar (hindpaw application) mechanical hyperalgesia. In order to avoid artificially reduced thresholds observed in mice because of sensitization to approach of the VF filament during repeated application, separate mouse cohorts were used for measurement of periorbital and plantar responses. Each filament was applied perpendicularly until it buckled and the stimulus was maintained for 3–5 s; mechanical thresholds were determined using the “up-down” method. During the acclimatization period mice were “trained” with repeated application of the VF hairs in the periorbital and plantar areas. To monitor the development of facial and paw hypersensitivity in migraine models, baseline pain thresholds were measured before induction of the migraine model (before injection of sumatriptan or GTN) and subsequent measurements were taken at regular intervals postinduction as indicated in graphs. For evaluation of the effect of ivabradine, mechanical thresholds were obtained before and 30, 60, and 120 min after ivabradine delivery and were compared with vehicle-injected animals. For the evaluation of the effect of repeated ivabradine treatment, mechanical thresholds were measured 60 min after each injection of the drug for a total of three injections at hourly intervals.
Drug administration
Ivabradine hydrochloride (Sigma SML0281) was diluted in PBS and delivered intraperitoneally (5 mg/kg, injection volume 10 ml/kg) either as a single injection (single administration) or three injections over 3 or 6 h (repeated treatment) as described in Results. Control animals received matched PBS injections.
Immunohistochemistry (IHC)
Tissue preparation
Rats were anaesthetized by sodium pentobarbital intraperitoneal injection and transcardially perfused sequentially with saline and 4% paraformaldehyde (PFA) in 0.1 m phosphate buffer (pH 7.4). Tissue [trigeminal ganglion (TG) and cervical spinal cord] was dissected and postfixed for either 4 h (TG) or overnight (spinal cord) at 4°C. Tissue was then stored in 20% sucrose in PBS (4°C, 48 h), embedded in optimum cutting temperature (OCT) compound, frozen and stored until sectioning (−80°C). Tissue sections were cut to required thickness (TG, 15 μm; spinal cord, 20 μm) using a cryostat and mounted onto Superfrost Plus glass slides (VWR).
Antibody staining
Slides were incubated overnight at room temperature (RT) with the primary antibody solution [PBS supplemented with 0.3% Triton X-100 solution, 2% BSA (Sigma), 4% normal goat serum and 4% normal donkey serum (Jackson ImmunoResearch)]. Primary antibodies used were rabbit anti-HCN2 (1:1000, Alomone APC-030), mouse anti-NaV1.8 (1:1000, Neuromab N134/12), chicken anti-β3tubulin (1:1000, Abcam #ab41489), mouse anti-NeuN (1:1000, Abcam #ab104224), and rabbit anti-cfos (1:1000, Cell Signaling #2250). The next day, slides were washed and incubated for 2 h with the appropriate secondary antibodies; donkey anti-mouse IgG-conjugated Alexa Fluor 594, goat anti-chicken IgG-conjugated Alexa Fluor 546, donkey anti-rabbit IgG-conjugated Alexa Fluor 488 or donkey anti-goat-conjugated Alexa Fluor 594 (all 1:1000, Invitrogen). Finally, slides were cover-slipped using Fluorsave mounting medium (Calbiochem) and stored at 4°C until visualization.
Image analysis
Antibody staining was visualized under an Axioplan 2 fluorescence microscope (Zeiss) and images were acquired using a digital camera and Axiovision software. For each experiment, tissue staining and imaging of all slides was performed at the same time to minimize variability, and the same exposure settings were used for image acquisition. Analysis of HCN2 immunofluorescence was conducted with ImageJ software. For each image, three BG intensity measurements were taken from tissue areas with no positive cells and mean intensity of BG as well as its SD were calculated; a cell was only considered positive when its immunoreactivity intensity was higher than 2× BG + 6 SD. For calculation of average intensity of HCN2 staining in TG, measurements of intensity were conducted on individual neurons using a 20× objective from at least five sections per TG (>200 cell profiles/animal), divided by the average BG density. For C-FOS-stained neurons (indicated by co-localization of C-FOS and NeuN fluorescence) were counted per spinal cord hemisection at the trigeminal nucleus caudalis (TNC), C1 and C2 levels (at least two sections/level). All image analysis was performed in a blinded fashion.
cAMP/cGMP ELISA
Cyclic nucleotides (cAMP and cGMP) were quantified using a commercial ELISA kit (Enzo Life Sciences). Briefly, the TG (both sides) were rapidly dissected and flash-frozen in liquid nitrogen (within 5 min of mouse death). Samples were ground using a small Teflon pestle over liquid N2, followed by vortexing in 250 μl of lysis buffer (1% Triton X-100 in 0.1 m HCl to stop endogenous phosphodiesterase activity). Samples were centrifuged at 600 × g to pellet the debris and supernatants were collected and run in duplicate on the ELISA plate, according to the manufacturer's protocol. Calculated cAMP or cGMP levels were normalized to total protein content, measured using the Pierce 660 nm Protein Assay kit (Thermo Scientific).
RNA isolation and qRT-PCR
After animals had been treated chronically with GTN (10 mg/kg, i.p., as in Fig. 1C) or saline, TG were collected 24 h after the last injection. Total RNA was extracted using TRI Reagent (Sigma-Aldrich) according to the manufacturer's instructions. Concentration and purity of RNA was measured with a NanoDrop Spectrophotometer (Thermo Fisher Scientific). For gene expression analysis, total RNA was reversed-transcribed using High-Capacity RNA-to-cDNA kit (Applied Biosystems). Quantitative real-time PCR (qPCR) was performed using TaqMan Gene Expression Master Mix (Applied Biosystems). Target Hcn2 mRNA expression was normalized to a house keeping gene (Hprt). qPCRs were run in a Roche LightCycler 480 PCR machine. Results were analyzed by the standard ΔΔCt method. TaqMan probes (Applied Biosystems) used were Hcn2 (Mm00468538_m1) and Hprt (Mm00446968_m1).
Extracellular single-neuron recordings from TG
Rats were anaesthetized with an intraperitoneal injection of 60 mg/kg pentobarbital sodium (Merial), and general anesthesia was maintained with continuous intravenous infusion of pentobarbital (12–15 mg/kg/h). A tracheotomy was performed to permit ventilation of the animal, and end-tidal expired CO2 was monitored and maintained between 3.5% and 4.5% (CapStar-100; CWE). The carotid artery and left femoral vein and artery were cannulated to allow for GTN delivery, infusion of anesthetic and monitoring of blood pressure, respectively. Adequate anesthesia was judged by the absence of toe-pinch withdrawal and eye-blink reflexes and of gross changes in blood pressure. Core temperature was monitored and maintained near 37°C using a homoeothermic blanket system (TC-1000, CWE). The animal was fixed on a stereotaxic frame (Kopf Instruments). Craniotomies were performed to expose the middle meningeal artery and to allow access for the recording electrode. Extracellular activity from single units in the TG, accessed stereotaxically, was recorded using a glass-insulated tungsten microelectrode (Kation Scientific) with an impedance of ∼1 MΩ. Signals were amplified and filtered as previously described (Andreou et al., 2015). The conditioned signal was digitized for storage using a Micro 1401-3 computer with Spike2 software (CED). Evoked AP activation thresholds of single TG neurons (∼ 5–20 neurons per animal) were assessed by electrical stimulation of the eyelid (1–30 V) via a Grass S88 stimulator (Grass Instrumentation), which activates the same first order neurons of the ophthalmic branch of the trigeminal nerve as those innervating the dura matter but is less invasive than direct dura stimulation. Monitoring of evoked APs allowed assessment of the threshold of AP activation by ramping up the stimulating voltage to a maximum of 40 V while keeping the pulse width constant (0.1 ms).
For the acute GTN model, following baseline recordings from neurons, GTN (50 μg/kg/min) was infused for 30 min via intracarotid delivery and GTN-induced excitability was studied at 30 min after infusion. This dosing/administration route regime was implemented to avoid potential side effect (e.g., low blood pressure) in anaesthetized animals (Ramachandran et al., 2012). Following this, ivabradine (5 mg/kg) or saline was delivered via intraperitoneal injection, and responses were studied 20–40 min after. Where possible, the recording electrode was kept on the same neuron before and after application of GTN or IVA to acquire additional data for paired analysis, before moving to a different neuron. Spontaneous activity (SA) was monitored for a period of 3 min before stimulation of each neuron and firing frequency was calculated from the inverse of the time between each AP. The latency of neuronal activation was also recorded following stimulation of the eyelid, as above, to separate neurons with Aβ (<6 ms), Aδ (7–25 ms), or C fiber (>25 ms) afferents. The extracellular recordings were made from cell bodies and were characterized by the biphasic appearance of the APs. Recordings of SA typically had 2–3 single units visible in traces, as judged from the differing sizes of action potentials, but it was not possible to reliably separate out individual neurons, so firing frequencies were calculated from all clearly visible action potentials. The experimenter was blinded to the identity of the treatment during data acquisition and analysis.
Patch clamp
TG neuron culture
Mice (f/fHCN2 male C57/BL6, effectively WT, six to eight weeks old) or global HCN2 KO littermates (male, up to four weeks old) underwent cervical dislocation and decapitation, the skull was opened and brain removed to expose the TG at the base of the skull. The TG were collected and incubated in papain (2 mg/ml in Ca2+ and Mg2+-free HBSS) for 30 min, followed by incubation in collagenase (2.5 mg/ml in Ca2+ and Mg2+-free HBSS) for 30 min at 37°C. Ganglia were mechanically dissociated in Neurobasal-A/B27 growing medium, which was prepared with Neurobasal-A Medium supplemented with 0.25% (v/v) L-glutamine-200 mm (Invitrogen), 2% (v/v) B-27 supplement (Invitrogen), 1% (v/v) penicillin-streptomycin (Invitrogen), and nerve growth factor (NGF; Sigma-Aldrich) at 50 ng/ml. Finally, dissociated TG neurons were centrifuged and plated overnight onto 10-mm coverslips (VWR) precoated with poly-L-lysine (10 μg/ml) and laminin (40 μg/ml). Electrophysiological recordings were performed at RT (22°C) up to 24 h (Iclamp recordings) or 48 h (Vclamp recordings) after plating.
Solutions
Manual patch-clamp experiments were conducted using an extracellular solution containing (in mmol/l): 140 NaCl, 4 KCl, 1.8 CaCl2, 1 MgCl2, 10 HEPES, 5 glucose, and 10 D-Mannitol (pH adjusted to 7.3 and osmolality 308 mOsm/l). The intracellular solution contained (in mmol/l): 140 KCl, 1.6 MgCl2, 2.5 MgATP, 0.5 NaGTP, 2 EGTA, and 10 HEPES (pH adjusted to 7.25 and osmolality 290 mOsm/l). GTN was diluted to the desired concentration in extracellular solution on the day of the experiment. All reagents were purchased from Sigma-Aldrich.
Whole-cell patch-clamp recordings
Small (<20 μm in diameter) mouse neurons were used for electrophysiological recordings. Whole-cell patch-clamp recordings were performed using an Axopatch 200B patch-clamp amplifier. Borosilicate patch-clamp pipettes (Science Products GmbH) were pulled using a P-97 horizontal micropipette puller (Sutter Instruments). Before use, pipettes were fire-polished with a Narishige MF-900 microforge, giving a final resistance between 2.5 and 3.5 Mohms. Pipette offset was corrected after immersing the pipette tip in extracellular solution and before approaching the cell. Once a giga-seal was obtained between the pipette and the cell, capacitative transients were cancelled before achieving the whole-cell configuration. Series resistance was compensated by 40–70%. Cells were held at −60 mV in voltage-clamp mode. When working in current-clamp mode, the I-Clamp fast mode was used. Whole-cell current or voltage recordings were sampled at 20 kHz and low-pass Bessel filtered at 2 kHz. Data were acquired using Axon pCLAMP software version 10.4 and analyzed offline with Clampfit 10.4 (Molecular Devices, LLC). Voltage-dependent activation of Ih currents was assessed by applying a family of prepulse voltage steps 1.5-s-long from −40 to −140 mV (Δ = 20 mV) and then calculating the midpoint activation voltage (V½) from tail currents recorded when a 1.5-s-long voltage step to −140 mV was applied. Maximal current density was measured with a single step to –140 mV. To calculate V½, plots of fractional activation versus voltage were fitted using a Boltzmann equation:
Data analysis
All data are presented as mean ± SEM. All results were analyzed using GraphPad Prism version 6.0. Unpaired Student's t test, one-way or two-way ANOVA and linear regression analysis were used to determine statistical significance, as indicated in the results section. Post hoc analysis was conducted using Student-Newman-Keuls (SNK) or Bonferroni post hoc tests, as indicated in the relevant legends; p < 0.05 was considered significant.
Results
Effect of pharmacological inhibition of HCN in rodent models of migraine-like pain
Administering glyceryl trinitrate (GTN; also known as nitroglycerine) induces migraine episodes in ∼80% of human migraineurs (Olesen, 2010). Nitric oxide (NO) is released from GTN in many bodily tissues by the mitochondrial enzyme aldehyde dehydrogenase (Chen et al., 2005), though the reason why an NO donor might trigger migraine has not been clearly established to date and is addressed in experiments described below. Periorbital hypersensitivity is characteristic of many human migraine states, including those induced by GTN (Burstein et al., 2011). A similar paradigm to human GTN-induced migraine has been established in rodents, where GTN administration causes migraine-like mechanical hypersensitivity that can be measured using VF hair stimulation of the periorbital area.
In rats, we found that a single GTN injection (10 mg/kg, i.p) significantly decreased baseline periorbital withdrawal thresholds (point at which animals react to a VF hair applied to the area immediately above the brow; see Materials and Methods) compared with the control (saline) group (from 9.71 ± 0.82 to 4.71 ± 0.79 g at 60 min, p < 0.001; Fig. 1A, red line). This hypersensitivity developed within 60 min, and the effect lasted for at least 2 h (3.08 ± 0.65 at 120 min, p < 0.001 vs baseline). Administration of sumatriptan (0.6 mg/kg, i.p.), a first-line migraine treatment in the clinic, at 60 min post-GTN increased withdrawal threshold compared with vehicle at 90 min (9.38 ± 0.96 vs 2.81 ± 0.93 g; p < 0.001, green) and 120 min (9.78 ± 1.5 vs 3.08 ± 0.65 g; p < 0.001), but thresholds were still significantly lower than baseline (13.71 ± 1.53 g; p < 0.05) or to controls (gray). We next tested the analgesic efficacy of ivabradine, an HCN channel blocker that inhibits all HCN isoforms approximately equally (Stieber et al., 2006). A single dose of ivabradine (5 mg/kg, i.p.) at 60 min post-GTN completely reversed pain thresholds to baseline levels 30 min after injection (from 6.75 ± 0.76 to 15.48 ± 1.68 g; p < 0.001, blue), an effect that was significantly greater than that achieved by sumatriptan treatment (p < 0.01). Note that when administered to control rats, without prior GTN injection, ivabradine did not affect acute mechanical thresholds in either the periorbital or the plantar area (Fig. 1B).
Development of hyperalgesia in GTN models of migraine in rats, and relief by the HCN ion channel blocker ivabradine. A, Acute GTN model in rat. Single injection of GTN (10 mg/kg, i.p., red arrow on x-axis) followed by measurements of peri-orbital mechanical threshold (manual VF hair) at times shown. Hyperalgesia develops over 90 min in the vehicle group. A single administration of ivabradine (5 mg/kg, i.p., black arrow on x-axis) restores periorbital thresholds to baseline levels. The analgesic effect of ivabradine was larger than that of sumatriptan (0.6 mg/kg, i.p., black arrow on x-axis). N = 8/group except GTN+veh where n = 24, two-way repeated measures (RM) ANOVA with Student-Newman-Keuls (SNK) correction (F(9,132) = 9.83, p < 0.001) versus BL (+); versus GTN+vehicle (*) or versus GTN+SUMA (#). B, Ivabradine administration (5 mg/kg, i.p.) in rats with no other treatment does not affect acute mechanical pain thresholds as measured at either the periorbital (top) or plantar (bottom) area; n = 6/group, two-way RM ANOVA; periorbital (F(3,30) = 0.42, p = 0.744); plantar (F(3,30) = 0.39, p = 0.761). C, Chronic GTN model in rat. Top, Repeated injection of GTN (10 mg/kg, i.p.) at 2-d intervals causes gradual development of plantar mechanical hyperalgesia measured by VF hair applied on the periorbital (left panels) or plantar (right panels) areas. Threshold measurements conducted at 2-d intervals, with measurements on days 1, 3, 5, and 7 (before daily GTN administration) and followed by GTN injection immediately after VF testing. GTN, n = 12; PBS, n = 4; two-way RM ANOVA with SNK, periorbital (F(5,70) = 30.2, p < 0.001); plantar (F(5,70) = 19.0, p < 0.001) versus BL (+) or saline (*). D, Analgesic effect of ivabradine in the chronic GTN model. Relief of hyperalgesia developed on day 9 of the chronic GTN model following repeated injections of HCN blocker ivabradine (5 mg/kg, i.p.) but not of vehicle control (PBS). N = 12/group; two-way RM ANOVA with Bonferroni, periorbital (F(4,88) = 10.44, p < 0.001), plantar (F(4,88) = 5.130, p < 0.001) versus BL (+), day 8 post-GTN (#) or saline (*). E, A high dose of ivabradine (15 mg/kg, i.p.) in rats chronically treated with GTN fully reverses both periorbital (top) and plantar (bottom) thresholds back to baseline levels; n = 8/group, two-way RM ANOVA with Bonferroni correction; periorbital (F(4,56) = 15.71, p < 0.001); plantar (F(4,56) = 6.82, p < 0.001) compared with baseline (+), vehicle control (*), or hyperalgesia post-GTN (#).
To explore an additional experimental paradigm of a migraine-like state, as well as to determine whether the continued presence of GTN is necessary to maintain a pain phenotype, we established a chronic GTN model in rats by repeated GTN administration (10 mg/kg, i.p., every 48 h, four injections in total). In this model, rats tested 2d after each GTN injection, immediately before the next injection, developed a progressively increasing periorbital mechanical hypersensitivity that reached a maximum on day 8 (threshold decreased from 8.6 ± 0.2 to 2.4 ± 0.2 g, p < 0.001 vs baseline, p > 0.05 vs day 7; Fig. 1C, left, red trace) and was maintained for several days after the last injection of GTN. This hypersensitivity was also evident in the plantar thresholds (from 13.2 ± 0.2 to 4.2 ± 0.3 g, p < 0.001; Fig. 1C, right, red trace), thought to reflect the extracephalic mechanical hypersensitivity that is also encountered in migraine patients. In the chronic GTN model, repeated administration of ivabradine (5 mg/kg, i.p., every hour for 3 h, starting on day 9 post-GTN) provided a cumulative analgesia that was evident in both periorbital thresholds (from 5.1 ± 1.3 g post-GTN to 10.9 ± 1.3 g at 3 h; p < 0.001 vs saline) and plantar thresholds (from 9.6 ± 2.4 g post-GTN to 15.3 ± 3.2 g at 3 h; p < 0.05 vs saline; Fig. 1D). The cumulative analgesic effect of ivabradine in the chronic GTN migraine model was similar to that previously observed in other persistent pain models such as painful diabetic neuropathy (Tsantoulas et al., 2017). In a separate experiment we injected chronic GTN-induced rats with a single high dose of ivabradine (15 mg/kg) and found that this provided superior analgesia at 60 min (Fig. 1E), reversing both periorbital and plantar thresholds back to baseline levels (p < 0.001).
Ivabradine has few off-target actions (Young et al., 2014), so the analgesic effect in these models supports an involvement of HCN ion channels in GTN-induced migraine pain. Ivabradine does not cross the blood-brain barrier (Young et al., 2014) so the effects can be attributed to actions on HCN ion channels in the peripheral nervous system. The particular isoform(s) involved cannot be identified, however, because ivabradine blocks all four members of the HCN ion channel family with approximately equal potency (Stieber et al., 2006).
Effect of nociceptor-specific HCN2 genetic deletion in mouse migraine models
We next used genetic approaches to tease out the key HCN isoform involved in driving migraine-like pain. To this end we extended the GTN model to mice (Fig. 2A). Acute GTN treatment (10 mg/kg, i.p.) triggered periorbital mechanical sensitivity from 1 h posttreatment (from 0.511 ± 0.044 to 0.081 ± 0.013 g; p < 0.001, red trace; Fig. 2A, left) which was also mirrored in plantar hypersensitivity (from 0.889 ± 0.115 to 0.232 ± 0.045 g; p < 0.001; Fig. 2A, right, red trace). In both cases, ivabradine injection (5 mg/kg, i.p.) elevated thresholds back to baseline levels 30 min after administration (periorbital, 0.483 ± 0.060 g; plantar, 0.671 ± 0.054; p < 0.001 vs vehicle, blue traces), an effect that lasted for at least 60 min (periorbital, 0.388 ± 0.056 g; vs vehicle; plantar, 0.567 ± 0.076; p < 0.001 vs vehicle). In mice, measurement of periorbital thresholds using a VF filament gave variable data, owing to the tendency of the mice to perceive and avoid the filament approach, while plantar testing gave more robust data with smaller variability (Fig. 2A). Plantar testing therefore became our preferred method for assessing hyperalgesia in mice.
Effect of nociceptor-specific genetic deletion of HCN2 on GTN-induced hyperalgesia in mice. A, Acute GTN model in mice. Acute administration of GTN (10 mg/kg, i.p.) in mice decreased VF withdrawal thresholds in the periorbital (left) and plantar (right) areas in a similar fashion, but the variability involved in measuring mouse plantar thresholds was much lower; for this reason, subsequent experiments on mice were conducted using plantar responses. A single injection of ivabradine (5 mg/kg, i.p.) was able to reverse mechanical thresholds back to baseline in both cases, confirming the analgesic effect of HCN block in mouse migraine models. GTN, glyceryl trinitrate; IVA, ivabradine; VEH, vehicle. n = 6/group; two-way RM ANOVA with SNK; periorbital (F(6,45) = 8.19, p < 0.001), plantar (F(6,45) = 8.96, p < 0.001); p < 0.05 versus BL (+), 60 min (#) or vehicle (*). B, Plantar hyperalgesia develops in mouse acute GTN model within 60 min but not when deletion of HCN2 is targeted to nociceptors (n = 14 cKO, n = 7 floxed littermates). cKO mice are NaV1.8-Cre × fHCN2 in which HCN2 is deleted in NaV1.8-Cre-expressing peripheral nociceptive neurons; “floxed” mice carry two copies of the fHCN construct and no Cre construct so are effectively WT. GTN injection 10 mg/kg, i.p., Two-way RM ANOVA with SNK (F(2,28) = 52.59, p < 0.001) versus BL (+) or fHCN2 (*). C, Chronic GTN treatment triggers a robust mechanical sensitivity in floxed mice, evident from day 3, but much less in the HCN2 cKO animals. GTN injection 10 mg/kg, every 48 h, i.p., four injections in total as in Figure 1. cKO (n = 8); fHCN2, floxed HCN2 controls (n = 8); two-way RM ANOVA with SNK (F(4,56) = 17.74, p < 0.001) versus BL (+) or fHCN2 (*). D, fHCN2 is selectively deleted in small neurons of the TG using a NaV1.8Cre construct. Images of neurons in TG stained for NaV1.8 (left), HCN2 (middle), and both images merged (right). Top row, TG neurons from f/fHCN2 mice (effectively WT). NaV1.8 is expressed predominantly in small neurons (arrows), most of which are nociceptor-specific. HCN2 is expressed both in small neurons and some larger neurons. Bottom row, TG neurons from NaV1.8Cre x f/fHCN2 mice. The Cre construct deletes HCN2 from all small neurons, but HCN2 expression in larger neurons is similar to that observed in the absence of the Cre construct. The HCN2 antibody was KO-validated (Tsantoulas et al., 2017).
In order to investigate whether the proalgesic effects seen in the rodent migraine models in Figure 1 are directly linked to the HCN2 isoform, we examined development of GTN-induced mechanical hypersensitivity in transgenic mice where HCN2 had been genetically deleted specifically from NaV1.8-expressing peripheral nociceptive neurons (Emery et al., 2011), including those in the TG (Fig. 2D). Administration of a single dose of GTN in mice that are WT apart from the presence of lox-P insertions flanking exons 2 and 3 of the HCN2 gene (homozygous fHCN2) induced a reduction in plantar VF thresholds, as previously observed in Figure 2A (from 0.975 ± 0.056 to 0.120 ± 0.012 g at 60 min; p < 0.001; Fig. 2B, red). We next tested the effect of GTN injection in HCN2 cKO mice, littermates to the fHCN2 mice, in which the HCN2 gene had been inactivated by expression of Cre in NaV1.8-expressing neurons of the peripheral nervous system. Figure 2B shows that HCN2 deletion prevented the development of hypersensitivity following GTN injection (from 0.983 ± 0.047 to 1.0 ± 0.036 g at 60 min; p > 0.05; Fig. 2B). Investigation of the chronic GTN model gave similar results; floxed controls (fHCN2), effectively WT, developed significant mechanical sensitivity on days 5 through 9 (from 0.975 ± 0.056 g at baseline to 0.038 ± 0.002 g on day 9, p < 0.001), while in cKO mice, this hypersensitivity was much reduced (p < 0.001 vs fHCN2 on day 9; Fig. 2C).
Genetic deletion of HCN2 using a Cre-lox system provides a highly selective means of identifying the particular isoform involved in GTN-induced mechanical hypersensitivity. A second advantage of the use of NaV1.8 as the Cre driver is that this sodium channel isoform is almost exclusively expressed in small and medium-sized nociceptors of the peripheral nervous system (Stirling et al., 2005; Shields et al., 2012). The results shown here therefore identify HCN2 located in peripheral nociceptive neurons as the isoform responsible for mechanical hypersensitivity in the rodent GTN model.
Medication Overuse Headache (MOH)
In order to investigate whether HCN2 function is also involved in other types of migraine-associated headaches, we used a rodent model of MOH. Migraine patients receiving pain medication (typically triptans) more than two to three times a week or 10 times a month can develop a dull constant headache present on most days or on part of every day (Diener et al., 2016). To model MOH in rats, we delivered sumatriptan (0.6 mg/kg, s.c.) every 2 d for up to 12 d (Fig. 3A; De Felice et al., 2010). Treated rats developed progressively increasing sensitivity to VF stimulation in both periorbital and plantar areas, measured 2 d after each injection of sumatriptan, that reached steady-state after four injections, i.e., on day 9 (periorbital, from 13.5 ± 1.3 to 3.8 ± 0.5 g; plantar, from 12.1 ± 0.8 to 3.5 ± 0.6 g; p < 0.001 vs baseline, p > 0.05 vs day 12) and persisted for around 13 d following cessation of treatment. On day 12, rats were injected with ivabradine (5 mg/kg) or vehicle (saline) following the cumulative dosing regime (one intraperitoneal injection every hour, three injections in total) with pain thresholds (periorbital and plantar) measured before each injection (Fig. 3B). As with the previous models, cumulative ivabradine injection caused analgesia that reached maximal levels at 3 h (periorbital, from 3.6 ± 0.1 to 7.4 ± 0.7 g; plantar, from 5.7 ± 0.1 to 10.6 ± 1.8 g; p < 0.01 vs vehicle). Treatment with ivabradine alone, using the same dosing regime as sumatriptan (5 mg/kg, i.p., every 48 h, four injections in total), did not cause any significant change in threshold at any time (Fig. 3C).
MOH model of migraine. A, MOH development in rats. Induction of MOH via repeated sumatriptan treatment (0.6 mg/kg, s.c., every 48 h as denoted by vertical arrows on x-axis) leads to prolonged mechanical sensitivity to VF stimulation in both the periorbital (left) and plantar (right) areas. Withdrawal thresholds remain low even after cessation of sumatriptan treatment (day 12) and slowly recover to baseline levels by day 35. Up to day 14, SUMA, n = 16; PBS, n = 8; after day 14 n = 4/group; two-way RM ANOVA with SNK, periorbital (F(10,159) = 9.18, p < 0.001); plantar (F(10,160) = 7.62, p < 0.001); p < 0.05 versus BL (+) or PBS (*). BL, baseline; SUMA, sumatriptan. B, Analgesic effect of ivabradine in MOH in rats. Repeated injection of ivabradine (5 mg/kg, i.p.) or vehicle control (PBS) tested on day 12 of MOH model (n = 6 per group). Treatment with ivabradine leads to a delayed increase in both periorbital (left) and plantar (right) mechanical pain thresholds. Two-way RM ANOVA with SNK; periorbital (F(4,40) = 7.25, p < 0.001); plantar (F(4,40) = 5.80, p < 0.001). Symbols denote comparison to baseline (+), vehicle control (*) or hyperalgesia post-MOH (#). C, Repeated injections of ivabradine (5 mg/kg, i.p.) do not cause mechanical hyperalgesia. Periorbital (left) and plantar (right) thresholds after an injection regime similar to that used for sumatriptan (every 48 h for 8 d). Two-way RM ANOVA, periorbital (F(4,40) = 0.60, p = 0.668), plantar (F(4,40) = 0.15, p = 0.964). D, Effect of nociceptor-specific genetic deletion of HCN2 on development of MOH in mice. Induction of MOH via repeated sumatriptan treatment (0.6 mg/kg, s.c., every 48 h, six injections in total) leads to prolonged mechanical sensitivity in floxed mice, but in HCN2 cKO mice, this phenotype is largely attenuated. GTN, glyceryl trinitrate; BL, baseline; cKO, conditional HCN2 KO (n = 15); fHCN2, floxed HCN2 controls (n = 7); two-way RM ANOVA (F(7,98) = 3.83, p = 0.001) with SNK; p < 0.05 versus BL (+) or fHCN2 (*).
Finally, we looked at the development of MOH in HCN2 cKO mice. Prolonged treatment with sumatriptan triggered MOH in control (fHCN2) mice that reached a steady level of hyperalgesia on day 9 (red line, from 0.958 ± 0.049 to 0.36 ± 0.032 g; p < 0.05; Fig. 3D) in a similar way to experiments using rats. In HCN2 cKO mice this mechanical hypersensitivity was largely attenuated (blue line), with pain thresholds in cKO being significantly higher compared with fHCN2 at all time points from day 7 onwards. There was a small drop in mechanical threshold in the cKO mice during the course of the sumatriptan injections, but it was only significant on days 7 and 9 (from 0.942 ± 0.056 g at baseline to 0.673 ± 0.066 g on day 7 and 0.640 ± 0.057 g on day 8, p < 0.01) and all measurements from day 12 onwards were not significantly different from the baseline threshold before the first injection of sumatriptan. Looking at potential sex differences, we found that, overall, there was no substantial difference in male versus female responses of transgenic animals in the migraine models we implemented (Fig. 4).
Male and female responses of transgenic mice were not different in the acute GTN, chronic GTN and MOH migraine models. A, Acute GTN (Fig. 2B). B, Chronic GTN (Fig. 2C). C, MOH (Fig. 3D) models implemented in male and female HCN2 cKO mice. No significant differences in hyperalgesia between sexes of same genotype were observed in GTN-induced models at any point. In the sumatriptan (MOH) model, there was no difference in the fHCN2 mice, but in the cKO group, there was a trend for female mice to develop higher sensitivity on days 5–9; however, this subsided by day 12 with no significant difference compared with male thereafter. N numbers for cKO male, fHCN2 male, cKO female, and fHCN2 female, respectively, were: acute GTN (4, 5, 4, 3); chronic GTN (4, 5, 4, 3); MOH (4, 5, 4, 3); two-way RM ANOVA with SNK (F(6,24) = 17.32, p < 0.001); F(12,48) = 7.08, p < 0.001; F(21,84) = 2.07, p = 0.01; p < 0.05 versus BL (+), versus same-group opposite sex (#) or same-sex fHCN2 (*). Arrows on x-axis denote GTN (A, B) or SUMA (C) injections.
Overall, these results show that peripheral HCN2 inhibition, achieved either by pharmacological block or by genetic deletion, inhibits hyperalgesia in rodent models of MOH. The results suggest that HCN2 block may therefore have analgesic value in treating MOH in humans.
Effect of HCN2 inhibition on activation of second-order neurons in rodent migraine models
Peripheral drive from primary nociceptive sensory neurons activates second-order neurons in outer layers of the dorsal horn of the spinal cord, and this activation drives expression of C-FOS, the protein product of the c-fos immediate-early gene (Hunt et al., 1987; Tsantoulas et al., 2017). Here, we measured C-FOS expression to detect activation of second-order neurons in the trigeminocervical complex (TCC), where TG afferents terminate.
In rats chronically treated with GTN (10 mg/kg, i.p., every 48 h, four injections in total) we quantified C-FOS upregulation in superficial laminae I–II across all levels of the TCC, comprising the TNC and cervical levels C1 and C2 (Fig. 5A), where the majority of trigeminal nociceptive neurons terminate (Tassorelli and Joseph, 1995; Ter Horst et al., 2001). On day 9 after treatment, 2 d after the last GTN injection, but at a time when migraine-like pain is still maintained (Fig. 1C), C-FOS expression was evident (11.6 ± 2.2 C-FOS-positive neurons per hemisection in GTN-treated rats). Note that only a fraction of TCC neurons were C-FOS positive, implying that the SA driving C-FOS expression is present only in a subset of primary afferent TG nerve fibers. In contrast, C-FOS expression was virtually absent from control animals not treated with GTN (4.2 ± 0.3 C-FOS-positive neurons per hemisection, p < 0.05).
Activation of C-FOS expression by chronic GTN model in second-order TCC neurons. A, Migraine was induced in rats via chronic GTN treatment (10 mg/kg, i.p., every 48 h for 8 d); on day 9, rats received three injections of either ivabradine (5 mg/kg, i.p.) or vehicle (PBS) every 2 h; 2 h after the last administration (i.e., 6 h since first injection, to allow enough time for changes in C-FOS expression to take effect), the spinal cord was perfused and the TCC, which comprises the TNC and C1-C2 levels of the cervical spinal cord, was dissected and cut into transverse sections for histologic examination. In control animals, C-FOS expression was virtually absent from the TCC. Chronic GTN treatment induced C-FOS expression in the superficial laminae of the spinal cord and ivabradine treatment significantly reduced C-FOS expression. N = 4/group, p < 0.05 compared with either control or GTN+IVA, Student's t test. Scale bar, 100 μm. B, Chronic GTN treatment also induced C-FOS staining in the TCC of floxed mice, but not in HCN2 cKO mice. N = 4/group, p < 0.001 versus fHCN2+GTN, Student's t test. Scale bar, 100 μm.
To test the hypothesis that C-FOS expression induced in second-order neurons of the TCC may be driven by activity of HCN2 in primary trigeminal nerve fibers, we next examined the effect of peripheral pharmacological block of HCN channels using ivabradine. We induced the chronic GTN migraine model as above, and on day 9 administered ivabradine (5 mg/kg, i.p.) or vehicle every 2 h and investigated C-FOS expression at 6 h after the start of treatment. The longer period of pharmacological inhibition compared with Figures 1–3 was implemented to allow time for changes in C-FOS protein expression. Ivabradine treatment reduced the number of C-FOS-positive neurons back to baseline levels (4.3 ± 1.7, p < 0.05 vs GTN; Fig. 5A), indicating a reduction in the activation of second-order neurons in the TCC.
We next tested the effect on C-FOS expression in the TCC of selective HCN2 gene inactivation in nociceptive primary afferent neurons using HCN2 cKO mice. Examining C-FOS expression in fHCN2 versus cKO mice revealed that although second-order neurons were activated by GTN treatment in the control fHCN2 mice, C-FOS stain was largely absent in cKO littermates following GTN treatment (26.0 ± 0.1 vs 0.9 ± 0.3; p < 0.001; Fig. 5B). Collectively these results link HCN2-dependent activity in peripheral sensory neurons to activation of second-order neurons in the TCC, a key afferent locus of the migraine pathway.
Finally, we examined C-FOS activation in the MOH model (Fig. 6). Repeated sumatriptan injections in rats induced a sustained migraine-like sensitivity to mechanical stimuli (as in Fig. 3A), and on day 12 animals were treated with either ivabradine (3× 5 mg/kg, i.p., every 2 h) or vehicle (saline). Second-order neuron activation determined from C-FOS expression was low in nonmigraine controls (4.3 ± 1.7, data from Fig. 5A), but there was substantial C-FOS upregulation following MOH induction (MOH, 10.8 ± 0.8 neurons/hemisection, p < 0.01 vs control; Fig. 6A). As with the other models investigated, C-FOS upregulation was suppressed by ivabradine treatment (4.2 ± 0.6 neurons/hemisection; p < 0.01 vs MOH). In fHCN2 mice (effectively WT), subjected to the same MOH induction paradigm, the expression levels of C-FOS were elevated (13.8 ± 0.5 neurons/hemisection), while the expression was suppressed by peripheral HCN2 gene deletion in HCN2 cKO mice (3.0 ± 1.0 neurons/hemisection; p < 0.001 vs fHCN2; Fig. 6B).
Activation of expression of C-FOS in second-order neurons of the TCC in the MOH model. A, Induction of MOH in rats by repeated sumatriptan treatment (0.6 mg/kg, every 2 d for 10 d) induces C-FOS expression in the TCC. Treatment with ivabradine (3 × 5 mg/kg, i.p., injections every 2 h) on day 12 reduces C-FOS staining at 6 h (i.e., 2 h after last injection), indicating attenuation of second-order neuronal activation by primary afferent nerve fibers. Control data in bar graph is taken from Figure 5A. N = 4/group, Student's t test, **p < 0.01 versus both control and MOH + IVA. Scale bar, 100 μm. B, Chronic sumatriptan treatment induced C-FOS staining in the TCC of floxed mice (fHCN2, effectively WT; top), but not in HCN2 cKO mice in which HCN2 has been deleted in NaV1.8-positive primary afferent neurons (bottom). N = 4/group, Student's t test, ***p < 0.001. Scale bar, 100 μm.
Spontaneous and evoked neuronal firing in TG neurons in migraine models
Induction of C-FOS expression provides a readily measured index of afferent nerve activity but is indirect, so we sought a more direct measure of neuronal activity in primary sensory neurons of the TG. We used extracellular recording to quantify afferent firing in rat TG neurons following induction of a migraine-like state with GTN, and we recorded both evoked firing thresholds following electrical stimulation of the periorbital area immediately above the eyelid, and ongoing SA in TG neurons.
Acute GTN administration (1.5 mg/kg, i.v.) reduced evoked firing thresholds in TG neurons recorded 1 h post-GTN (from 23.75 ± 0.84 to 14.34 ± 0.43 V, p < 0.001, results for all neurons regardless of conduction velocity; Fig. 7A, top). Treatment with ivabradine (5 mg/kg, i.p.) restored thresholds back to baseline levels at 20–40 min after treatment (24.31 ± 0.82 V, p < 0.001 vs GTN; Fig. 7A). This change was also evident in all three of the fiber categories identified from conduction velocity, i.e., Aβ, Aδ, and C fibers (Fig. 7A, bottom panels). In a smaller number of paired recordings, in which the recording position was maintained before and after administration of ivabradine, significant increases in threshold following administration of ivabradine were observed in both the Aδ and C fiber classes (p < 0.001 for each; Fig. 7B; Extended Data Fig. 7-1).
In vivo extracellular recordings from trigeminal neurons of GTN-treated rats. A, Intraarterial administration of GTN (1.5 mg/kg) reduces the threshold for firing in response to an electrical stimulus in the periorbital area delivered via a stimulating electrode attached to the eyelid, 1 h following treatment. Subsequent administration of ivabradine (5 mg/kg, i.p.) elevated firing thresholds back to baseline levels. Top, All fibers. Control, n = 67; GTN, n = 115; IVA, n = 83; ANOVA with SNK (F(2,262) = 73.52; ###, ***p < 0.001 vs control and GTN, respectively). Bottom, Break-down of results by fiber type, as determined from calculation of conduction velocity following shock administered via stimulating electrode. ANOVA with SNK (F(2,53) = 15.13; F(2,151) = 58.3; F(2,52) = 22.98 for Aβ, Aδ, and C fibers, respectively; ##, **p < 0.01; ###, ***p < 0.001 vs control and GTN, respectively). B, Paired analysis following responses of neurons recorded continuously before and after injection of ivabradine shows increased firing thresholds in Aδ and C fiber neurons following ivabradine (***p < 0.001, Student's paired t test. C, n = 8; Aδ, n = 11; Aβ, n = 3). All activation threshold values reported in Extended Data Figure 7-1. C, Chronic treatment with GTN (as in Fig. 1C) reduces the threshold for evoked firing in response to electrical stimulation of the periorbital area (compare with control values in A). Administration of ivabradine (5 mg/kg, i.p.) elevated firing thresholds (GTN, n = 122; IVA, n = 112; p < 0.001, Student's t test). Bottom panels, Break-down of results by fiber type determined from conduction velocity. D, Paired analysis following responses of single neurons as in B shows that ivabradine significantly increased firing thresholds in C fiber and Aδ fiber neurons (*p < 0.05, Student's paired t test. C, n = 4; Aδ, n = 6; Aβ, n = 2). All activation threshold values reported in Extended Data Figure 7-2.
Extended Data Figure 7-1
Activation threshold values for individual fibers from rats treated with acute GTN, before and after ivabradine. Paired fiber analysis, p < 0.05, Student's t test. Download Figure 7-1, DOCX file.
Extended Data Figure 7-2
Activation threshold values for individual fibers from rats chronically treated with GTN, before and after ivabradine. Paired fiber analysis, p < 0.05, Student's t test. Download Figure 7-2, DOCX file.
Similar results were obtained following induction of the chronic GTN model (10 mg/kg, i.p., every 48 h, four injections in total; Fig. 1C). 24 h after the final GTN administration, activation thresholds for TG neurons were reduced to 16.61 ± 0.57 V (Fig. 7C, p < 0.001 compared with controls in Fig. 7A). Ivabradine treatment (5 mg/kg, i.p.) returned thresholds to pretreatment levels both in analysis of recordings from all fibers (27.32 ± 0.92 V, p < 0.001 vs GTN; Fig. 7C) and from analysis of paired-fiber recordings before and after ivabradine treatment (Fig. 7D; Extended Data Fig. 7-2). This effect was significant across Aδ and C fibers but not in Aβ fibers (Fig. 7C, bottom panels, 7D).
We also investigated changes in neuronal SA induced by GTN treatment. Several units were typically identified in each recording (Fig. 8A,D), but it proved impossible to reliably separate them into individual units so only total firing rates are reported. We found significantly increased levels of SA following acute (Fig. 8A–C) or chronic (Fig. 8D–F) treatment with GTN. Analysis of SA recorded from all neurons showed that SA increased significantly both in the acute GTN model (from 0.17 ± 0.02 to 0.77 ± 0.14 spikes/s, p < 0.01) and in the chronic GTN model (from 0.17 ± 0.02 in the control recordings shown in Fig. 8B, to 0.36 ± 0.04 spikes/s postchronic GTN, p = 0.001; Fig. 8E). In both migraine models, inhibition of HCN channels by ivabradine (5 mg/kg, i.p.) decreased the rate of SA (acute GTN, from 0.77 ± 0.14 to 0.31 ± 0.06 spikes/s, p < 0.01; chronic GTN, from 0.36 ± 0.04 to 0.19 ± 0.03 spikes/s, p < 0.01, both 20–40 min after treatment with ivabradine).
SA recorded from TG neurons following acute (A–C, n = 6 rats) or chronic (D–F, n = 7 rats) GTN treatment, and the effect of ivabradine. Traces in A, D show two representative recordings from TG for each condition (60-s continuous segments), in the acute and chronic GTN models, respectively. B, Quantification of SA frequency in different neurons showing an increase of SA in all units following acute GTN and reduction following IVA (BL, n = 19; GTN, n = 38; GTN+IVA, n = 32 individual recordings); p < 0.01, one-way ANOVA with Bonferroni; F(2,86) = 8.6, p < 0.001). C, Paired analysis of SA in single maintained recordings, before/after application of GTN or IVA showing a statistical difference in firing frequency in some units (***p < 0.001, paired Student's t test). E, Quantification of SA frequency in different neurons in the chronic GTN model showing a reduction of SA in all units following IVA (GTN chronic, n = 40; IVA, n = 17; **p < 0.01, student's t test). F, Paired analysis of SA in single maintained recordings in the chronic GTN model, before/after application of IVA (*p < 0.05, paired Student's t test).
These results were obtained by comparing all neurons, but effects of GTN in enhancing SA, and of ivabradine in suppressing it, were also observed in paired recordings from a smaller number of neurons in which the recording position was maintained before and after application of GTN or ivabradine (Fig. 8C,F). Interestingly, recordings with lower levels of spontaneous firing were in general unaffected by acute administration of GTN, while a significant increase in firing was found in one recording with a higher initial level of spontaneous firing (Fig. 8C). Conversely, in two recordings with elevated levels of spontaneous firing following acute administration of GTN, a highly significant decrease in firing rate was observed following administration of ivabradine (Fig. 8C), while in four units with lower levels of spontaneous firing no effect was seen. Similarly, following chronic administration of GTN, ivabradine caused a significant decrease of firing in one active recording while others with lower SA were unaffected (Fig. 8F).
The relatively small fraction of neurons whose firing rate is enhanced by GTN, and the correspondingly small fraction whose firing rate is suppressed by ivabradine, correlates with the observation that an increase in C-FOS expression, in response to GTN or induction of the MOH model, is only observed in a small fraction of second-order neurons (Figs. 5, 6). In another long-term pain condition, painful diabetic neuropathy, we also found that the number of C-FOS-expressing second-order neurons was significantly less than the number induced by a maximal acute pain stimulus (Tsantoulas et al., 2017). Thus, even agonizing long-term pain conditions may involve activation of repetitive firing in only a relatively small fraction of nociceptive neurons.
Together, the extracellular recordings from primary sensory neurons in the TG show that induction of a migraine-like state with GTN can sensitize nociceptive neurons in the TG and that this increase in firing is reversed by the HCN blocker ivabradine.
Cyclic nucleotide levels in TG neurons as potential drivers of migraine pain
The activity of the HCN2 ion channel is known to be enhanced by direct binding of cAMP or cGMP to a cyclic nucleotide binding domain (CNBD) in the C-terminal region of the channel (DeBerg et al., 2016). Chronic pain states are driven by HCN2 activity, and consistent with this, an elevated level of cAMP has been found in somatosensory neurons in pain models of painful diabetic neuropathy (Tsantoulas et al., 2017). In the present study, we found that GTN, an NO donor, can generate a migraine-like state, suggesting that NO released from GTN may activate neuronal guanylate cyclase, leading to increased cellular levels of cGMP and thus to activation of HCN2. We therefore quantified cAMP and cGMP levels in the TG in the GTN model of migraine.
In the acute GTN model, rats were dosed with GTN (10 mg/kg, i.p.) and 1 h later the TG were rapidly dissected and snap-frozen in liquid nitrogen. Quantification of cGMP levels demonstrated an ∼4-fold increase in the GTN-treated animals compared with controls (1.09 ± 0.16 vs 0.27 ± 0.1 pmol/mg protein; Fig. 9, left). Quantifying cAMP levels in the same samples indicated there was no difference in cAMP content before and after GTN treatment (5.13 ± 1.57 vs 6.11 ± 1.68 pmol/mg protein; Fig. 9, right). This result supports the idea that increased cGMP levels drive HCN2 activation in migraine models.
Cyclic nucleotide regulation in the TG in migraine models. Acute GTN treatment (10 mg/kg, i.p.) caused a 4-fold increase in cGMP concentration in TG, assessed 1 h after treatment via ELISA (left panel, n = 5/group; Student's t test, p < 0.01). In contrast, when examining the same samples for cAMP content, there was no difference between groups (right panel). ns, non-significant.
HCN2 expression in chronic migraine-like states
The involvement of HCN2 in chronic migraine-like states, as shown above, raises the question of whether upregulation of HCN2 expression could be important in maintaining pain in chronic migraine. We used the chronic GTN model to ascertain whether there is any upregulation of HCN2 expression in this long-term migraine model. Twenty-four hours after the last injection of GTN, at which time hyperalgesia was clearly established (Fig. 1C), quantitative PCR showed no difference in HCN2 mRNA levels in whole TG between control and GTN-treated rats (1.06 ± 0.15 vs 1.04 ± 0.3, expression normalized to housekeeping gene Hprt, p > 0.05; Fig. 10A). Examining HCN2 protein expression in the TG using an antibody against HCN2, validated by the absence of stain in global HCN2 KO mice (Tsantoulas et al., 2017), also showed no difference in either HCN2-specific staining intensity (2.1 ± 0.1 vs 1.9 ± 0.35; p > 0.05) or percentage of HCN2-positive neurons (70.4 ± 4.1% vs 68.9 ± 6.4%, p > 0.05; Fig. 10B) between control and GTN-treated rats. We conclude that there is no evidence to suggest that upregulation of expression of HCN2, as opposed to an enhancement of ion channel open probability driven by cAMP or cGMP, drives migraine-like pain.
HCN2 is not upregulated in the TG in migraine models. A, Chronic GTN treatment (10 mg/kg, i.p., as in Fig. 1C) did not affect HCN2 mRNA levels in rat TG as assessed by qPCR 24 h after the last injection (control, n = 6; GTN, n = 5; Student's t test, p > 0.05). B, IHC using an HCN2 KO-validated antibody in the rat TG shows that chronic GTN treatment does not change the overall percentage of neurons positive for HCN2 protein or the intensity of HCN2-dependent fluorescence (n = 4/group; Student's t test). n.s., non-significant. Scale bar, 50 μm.
Effect of GTN on HCN currents of TG neurons in vitro
The work in Figures 7 and 8 above shows that TG neuronal excitability is enhanced in vivo following acute and chronic treatment with GTN, and that block of HCN2 ion channels with ivabradine suppresses the enhanced excitability. In order to gain a more mechanistic insight into regulation of neuronal excitability by HCN2 in vitro, we performed whole-cell patch clamping experiments on acutely dissociated mouse TG neurons. We recorded from the small neurons (<20 µm in diameter) that are most likely to perform a nociceptive function, either in control conditions or up to 20 min following acute GTN treatment (0.1–10 μm). A hyperpolarizing shift in HCN2 channel activation (“rundown”) has been reported previously (Pian et al., 2006) and was observed in the present experiments with a long duration of whole-cell patch clamp, so the duration of all experiments, following gaining whole-cell access via the patch clamp pipette, was limited to 10 min. Figure 11A, left, shows a typical example of enhanced neuronal excitability caused by treatment with 1 μm GTN. Following current injection (0–80 pA), neurons treated with 0.1–10 μm GTN showed increased AP firing frequency compared with control neurons, indicating GTN-dependent hyperexcitability (p < 0.01 vs control, linear regression analysis; Fig. 11A, right). Note that the firing frequency was significantly enhanced by GTN at all concentrations up to 10 μm; at 10 μm GTN the increase in firing frequency (green) appears slightly less than at 1 μm (red), though the difference was not statistically significant.
In vitro excitability and HCN conductance of trigeminal neurons after acute GTN treatment. A, Representative traces from a single neuron (left, 1 μm GTN) and quantification of AP firing in response to current injection (right) in acutely dissociated small trigeminal neurons (diameter < 20 µm) following acute (up to 20 min) GTN treatment (0.1–10 μm). Control, n = 5; GTN 0.1 μm, n = 7; GTN 1 μm, n = 6; GTN 10 μm, n = 18; p < 0.05 versus control, linear regression analysis, F = 4.189 (3,122), R2 = 0.15, 0.48, 0.43, 0.35, respectively. B, HCN currents recorded in patch-clamp mode using a previously described hyperpolarization protocol (Young et al., 2014). C, Activation of HCN current as a function of membrane voltage. There was a small but significant depolarizing shift in V50 following GTN treatment: GTN 1 μm (p < 0.001), 10 μm (p < 0.01), and 100 μm (p < 0.05); two-way ANOVA with Bonferroni (F(4,1255) = 4.58, p < 0.001). V50 values: control, −91.63 ± 0.77 mV (n = 65); 0.1 μm GTN, −88.88 ± 2.98 mV (n = 17); 1 μm, −87.68 ± 1.22 mV (n = 65); 10 μm, −89.07 ± 1.37 mV (n = 57); 100 μm, −86.31 ± 2.5 mV (n = 19). D, HCN currents recorded from HCN2 gKO mouse neurons did not exhibit any shift in voltage dependence following GTN treatment. Two-way ANOVA (F(1,39) = 0.22, p = 0.65); gKO, −93.84 ± 2.46 mV (n = 19); gKO+GTN, 94.03 ± 3.04 mV (n = 22, p > 0.05). E, GTN decreased the fast component of the activation time constant (Tau fast) indicating accelerated channel kinetics (control, n = 53; GTN 1 μm, n = 56; two-way ANOVA with Bonferroni; F(3,301) = 5.41; p < 0.01). No significant change in Tau fast was observed in neurons from KO mice (gKO, n = 16; gKO + GTN, n = 19; two-way ANOVA with Bonferroni; F(3,95) = 0.55), p > 0.05).
We next used the voltage-clamp mode to measure the voltage dependence of activation of HCN ion channels (Fig. 11B), using an established hyperpolarizing protocol to activate all HCN channels (Young et al., 2014). HCN currents in small somatosensory neurons are carried by an approximately equal mix of HCN2 and HCN3 (Lainez et al., 2019). Current activation has a similar time course in these two isoforms and they are therefore difficult to separate. A critical difference, however, is that in the case of HCN2 the voltage dependence of current activation shifts in the depolarizing (positive) direction in response to intracellular cAMP or cGMP (DeBerg et al., 2016), while HCN3 is relatively insensitive to changes in the intracellular levels of cyclic nucleotides (Mistrík et al., 2005). The presence of current carried by HCN3 in the mixed HCN2-HCN3 current recordings from TG neurons therefore “dilutes” the voltage shift of HCN2, when compared with that recorded with HCN2 alone (Lainez et al., 2019).
The voltage dependence of Ih activation was analyzed in separate neurons in the presence or absence of GTN. GTN concentrations of 1 μm or higher caused a small but significant rightward shift in the voltage dependence of HCN2/3 activation (Fig. 11C), consistent with HCN2 sensitization by intracellular cGMP. The V50 value for control was −91.63 ± 0.77 mV, whereas after GTN the shifts in V50 values (ΔV50) were 2.75 mV (p > 0.05), 3.95 mV (p < 0.001), 2.56 mV (p < 0.01), 5.32 mV (p < 0.05) for 0.1 μm, 1 μm, 10 μm and 100 μm GTN respectively (see legend to Fig. 11). GTN treatment did not significantly alter the maximum current density of HCN currents at –140 mV (control, −16.61 ± 1.59 pA/pF; GTN 1 μm, −23.13 ± 2.13 pA/pF; GTN 10 μm, −17.08 ± 2.19 pA/pF; GTN 100 μm, −13.46 ± 3.36 pA/pF; all p > 0.05 vs control, one-way ANOVA).
When trigeminal neurons from HCN2 gKO mice were used in similar experiments, a hyperpolarization-activated inward current, carried by HCN3 in the absence of HCN2, was still observed. However, when these gKO cells were treated with GTN, no significant shift in HCN current activation was observed (ΔV50 = −0.19 mV, p > 0.05, two-way ANOVA; Fig. 11D), pinpointing HCN2 as the key isoform in the GTN-induced shift in the voltage dependence of HCN ion current activation and trigeminal neuron excitability.
Finally, we analyzed the fast component of the activation time constant of HCN2/3 ion currents (Tau fast) as a function of voltage, to determine whether GTN affects HCN current kinetics (Fig. 11E). Interestingly, 1 μm GTN decreased Tau fast at –80 mV (from 489.2 ± 22.0 to 391.5 ± 26.9 ms; p < 0.001), at –100 mV (from 361.7 ± 16.0 to 283.5 ± 17.4 ms; p < 0.01), and at –120 mV (from 234.6 ± 11.0 to 176.6 ± 10.0 ms, p < 0.05, two-way ANOVA). No change in Tau fast was detected in HCN2 gKO neurons (483.7 ± 40.1 vs 472.1 ± 20.4 ms at –80 mV, p > 0.05, two-way ANOVA; Fig. 11E), suggesting that GTN may have an effect of accelerating HCN2 but not HCN3 ion channel activation that is additional to its effect on the voltage dependence of activation.
Discussion
Five major findings arise from this study: (1) blocking the function of HCN2 ion channels, either pharmacologically or genetically, provides analgesia in three distinct animal models of migraine; (2) migraine-like hyperalgesia is precipitated by hyperexcitability of neurons in the TG; (3) TG neuronal hyperexcitability is mediated by increased activity of HCN2 channels, as it is abolished by HCN2 inhibition; (4) an elevation of cGMP in trigeminal neurons is observed in the migraine-like state induced by GTN; (5) GTN promotes TG neuronal excitability by causing a positive shift in the voltage range of activation of HCN2.
Despite its high prevalence and impact on quality of life, the pathophysiology of migraine at the molecular level remains largely unknown. Theories based on dilation of meningeal arteries as a causative factor for migraine have largely been abandoned, partly because the ditan family of antimigraine drugs do not cause vasoconstriction (Rubio-Beltrán et al., 2018). Several lines of evidence suggest that CGRP, a neuropeptide secreted by trigeminal nociceptive neurons, plays a key role in the induction of migraine in humans. First, elevated CGRP levels have been reported in serum and saliva during migraine attacks, and these levels are reduced by anti-migraine treatment (Ho et al., 2010); second, intravenous injection of CGRP triggers migraine-like attacks in migraineurs (Hansen et al., 2010); and third, removing CGRP with monoclonal antibodies, or blocking its receptor with antagonists, is effective in relieving migraine in at least some patients (Tso and Goadsby, 2017; Skljarevski et al., 2018).
Our results provide further support for the idea that the pain of migraine is peripherally-mediated. Using in vivo electrophysiology in the GTN model of migraine, we observed increased peripheral hyperexcitability, both evoked and spontaneous, in trigeminal neurons. The HCN ion channel blocker ivabradine, which is peripherally restricted and therefore will have no direct effect on CNS neurons, inhibited both evoked and spontaneous activity. In addition, in all migraine models examined there was a marked increase in C-FOS expression in second-order neurons of the TCC, reflecting enhanced spontaneous peripheral input from neurons of the TG. Both migraine pain and second-order neuronal activation were diminished by blocking peripheral activity, either with ivabradine or by selective HCN2 gene KO in peripheral neurons of the TG.
In two chronic migraine models that we tested [chronic exposure to GTN (Fig. 1D) and sumatriptan MO model (Fig. 3B)], ivabradine was more effective when administered repeatedly, but crucially, long-term HCN2 block with ivabradine did not cause any hypersensitivity (Fig. 3C). This is in contrast to other anti-migraine treatments, including triptans, ergotamines, opiates, NSAIDs, and paracetamol, whose repeated use can trigger MOH. Gene KO experiments showed that the key HCN isoform responsible is HCN2, because genetically deleting HCN2 in peripheral neurons of the TG conferred analgesia both in the acute GTN migraine model and in chronic GTN or MO models. These findings suggest that HCN2-targeted treatments could be effective either preemptively, to minimize the risk of migraine attacks, or to dampen migraine pain once an episode has been initiated.
In many chronic pain conditions, the recruitment of inflammatory mediators (e.g., PGE2) at the site of injury or inflammation causes activation of adenylyl cyclase via the Gs-coupled receptor pathway, increasing intracellular cAMP, which then potentiates HCN2 channel opening and results in more frequent AP firing (Emery et al., 2011; Tsantoulas et al., 2016, 2017). Here, we extend these findings by demonstrating that HCN2 channels are also the downstream effectors of migraine-like pain in rodent models.
An important role for intracellular cAMP as a convergence point in migraine is supported by the cellular mechanisms of action of a number of migraine treatments, as follows:
NSAIDs, which offer relief to a significant fraction of migraineurs, inhibit the production of PGE2, which couples via the EP4 receptor to the G protein alpha subunit Gαs, activating adenylate cyclase (AC) and leading to the production of intracellular cAMP in nociceptive afferent nerve fibers (Zeilhofer and Brune, 2006). Thus, the analgesic action of NSAIDs in migraine may be due to a reduction of cAMP in meningeal nociceptive afferent nerve fibers, as has previously been found in neurons of the DRG (Emery et al., 2011).
Triptans and ditans, potent analgesics in migraine, are agonists at 5HT1B/D receptors and 5HT1F receptor, respectively. These receptors all couple to Gi/o, and therefore, their activation will tend to reduce intracellular cAMP in nociceptors (Alexander et al., 2017).
The gepant family of anti-migraine drugs block the receptor for CGRP, while anti-CGRP monoclonal antibodies inhibit activation of the same receptor by removing its principal agonist. Both have recently enjoyed success as anti-migraine treatments. The receptor for CGRP couples to Gs and therefore increases cAMP (Alexander et al., 2017); thus, both gepants and anti-CGRP monoclonals will act to reduce cAMP.
Taken together, these considerations support the proposal that elevated intracellular cAMP in trigeminal afferents is a critical initiator of migraine (Schytz et al., 2010). Here, we extend this proposal by demonstrating that the downstream target of cAMP in primary sensory neurons of the TG is the HCN2 ion channel, to which cAMP binds, thus causing enhanced channel opening.
Migraines are also known to be precipitated by the NO donor GTN, used clinically as an anti-anginal (Schytz et al., 2010). NO activates guanylate cyclase and thus elevates intracellular cGMP (rather than cAMP), and here we have demonstrated an increased cGMP, but not cAMP, in trigeminal neurons following GTN administration in rats. Importantly, cGMP can bind to the CNBD domain of HCN2 channels, at the same site as cAMP, to potentiate ion channel function (DeBerg et al., 2016). Using patch-clamping we demonstrate that GTN increases AP firing in trigeminal neurons, concurrently with a depolarizing shift in the voltage dependence of HCN activation and an acceleration of channel kinetics. These in vitro changes correlate with the increased neuronal excitability and spontaneous firing rate that we observed in GTN migraine models in vivo and are therefore likely to provide a mechanistic explanation for GTN-induced migraine episodes. Furthermore, these results emphasize an underappreciated role of HCN2 modulation by cyclic nucleotides other than cAMP that may be of relevance to other pain conditions.
Additional lines of evidence supporting a role for cAMP and cGMP in precipitating migraines come from human studies with cilostazol, which causes headache by inhibiting the bifunctional phosphodiesterase PDE3 and therefore elevating both cAMP and cGMP (Birk et al., 2006); with sildenafil, which inhibits the cGMP-selective phosphodiesterase PDE5 and therefore elevates cGMP (Kruuse et al., 2003); and with PACAP38, which activates Gs and thus elevates cAMP (Guo et al., 2017).
Together, our results suggest that HCN2 ion channels are the final common initiator of the pain of migraine. Because HCN2 is the final convergence point downstream of either cAMP or cGMP elevation, on which multiple pathways meet, as noted above, selective targeting of this ion channel isoform has the potential to provide analgesia in migraine with few side effects. Developing HCN2 inhibitors that are both peripherally restricted, to avoid the adverse effects of HCN2 inhibition in the CNS, and HCN2-selective, to avoid bradycardia caused by block of HCN4 that is the main isoform driving the heart rate, is therefore likely to give safe and effective analgesia in migraine patients.
Footnotes
This work was supported by the Brain Research United Kingdom Grant 201718-16 and the Wellcome Trust Grant 205006/Z/16/Z. We thank Joseph Lloyd for assisting with single-unit recording and Bruno Vilar for helpful discussions and assistance with patch clamp.
P.A.M. is involved in a drug discovery program in collaboration with Merck & Co, Inc, to develop HCN2-selective molecules as novel analgesics. All other authors declare no competing financial interests.
- Correspondence should be addressed to Christoforos Tsantoulas at chris.tsantoulas{at}kcl.ac.uk or Peter A. McNaughton at peter.mcnaughton{at}kcl.ac.uk