Orexins and primary headaches: an overview of the neurobiology and clinical impact

ABSTRACT

Introduction

Primary headaches, including migraines and cluster headaches, are highly prevalent disorders that significantly impact quality of life. Several factors suggest a key role for the hypothalamus, including neuroimaging studies, attack periodicity, and the presence of altered homeostatic regulation. The orexins are two neuropeptides synthesized almost exclusively in the lateral hypothalamus with widespread projections across the central nervous system. They are involved in an array of functions including homeostatic regulation and nociception, suggesting a potential role in primary headaches.

Areas covered

This review summarizes current knowledge of the neurobiology of orexins, their involvement in sleep-wake regulation, nociception, and functions relevant to the associated symptomology of headache disorders. Preclinical reports of the antinociceptive effects of orexin-A in preclinical models are discussed, as well as clinical evidence for the potential involvement of the orexinergic system in headache.

Expert opinion

Several lines of evidence support the targeted modulation of orexinergic signaling in primary headaches. Critically, orexins A and B, acting differentially via the orexin 1 and 2 receptors, respectively, demonstrate differential effects on trigeminal pain processing, indicating why dual-receptor antagonists failed to show clinical efficacy. The authors propose that orexin 1 receptor agonists or positive allosteric modulators should be the focus of future research.

KEYWORDS:

• Headache
• orexin
• hypocretin
• migraine
• sleep

1. Introduction

Primary headache disorders are highly prevalent [1,2] and disabling conditions [3] which significantly impact quality of life [4]. These include migraine, tension-type headache, and several trigeminal autonomic cephalalgias, including cluster headache (CH) [5]. Whilst treatment advances have been made in recent years, these are largely due to the development of anti-CGRP-based therapies [6,7], and the mechanisms of headache initiation and chronification are still relatively unclear [8,9]. There remains a significant need to lessen the burden for those with primary headache disorders, particularly for those individuals who do not respond to current treatments, experience intolerable side effects, or have contraindications [10]. Recently, several novel neuropeptides and neurotransmitter targets have emerged [11] including the orexins [12], which may prove fruitful for the treatment of these disorders.

The hypothalamus is a small diencephalic region involved in homeostatic regulation including body temperature, hormone release, appetite, and arousal [13]. It has reciprocal connections to the thalamus, brainstem periaqueductal gray (PAG), median raphe nuclei, locus coeruleus, and spinal cord, regions which are involved in pain processing, thus, the hypothalamus and associated areas may be responsible for the underlying pathophysiology of headaches and their related features [5]. Indeed, evidence for the involvement of the hypothalamus in trigeminovascular nociception and headache pathophysiology has been provided by several studies [14,15]. For example, the hypothalamus is shown to be active during the premonitory and attack phases of spontaneous and experimentally triggered migraine [16,17] and CH attacks [18]. Alterations in the functional connectivity between the hypothalamus and pain processing regions have also been observed before the beginning of migraine attacks [19–23]. Moreover, the chronobiological features of several headache disorders [24], the potentially sleep-related attacks of migraine [25–28], and the circannual periodicity of CH bouts [29] implicate alterations to the biological clock, which resides in the suprachiasmatic nucleus (SCN) of the hypothalamus, in their pathophysiology. Furthermore, the premonitory symptoms (e.g. thirst, abnormal fatigue, and frequent urination) associated with migraine headaches, and alterations in sleep architecture and quality in migraine [30] are suggestive of perturbed homeostatic regulation [31]. Highlighting the potential importance of altered hypothalamic signaling in setting the threshold for the initiation of primary headaches, including migraine and CH [32].

This review discusses the neurobiology of the hypothalamic neuropeptide system – the orexinergic system and their potential relevance in the underlying mechanisms and treatment of primary headaches.

1.1. Orexins

Although the hypothalamus contains a variety of neurotransmitter and peptide systems including GABA, glutamate, melanin-concentrating hormone, oxytocin, kisspeptin, and neuropeptide Y [33], one important hypothalamic peptide linked to headache pathophysiology is orexin [34]. The orexin system is comprised of two G protein-coupled receptors (GPCR): OX₁R and OX₂R (Figure 1). There are two neuropeptides, orexin-A (OXA) and orexin-B (OXB; also known as hypocretin 1 and hypocretin 2) which are cleaved from the precursor protein prepro-orexin [36]. Whilst activation of either receptor is excitatory, the OX₁R is selective for OXA, whereas OX₂R is nonselective for OXA and OXB. However, OXB has a 10-fold higher affinity for the OX₂R than the OX₁R [37]. The orexin peptides were originally identified for their role in feeding behavior, and consequently termed them the orexins – named after the Greek for appetite [36], while another research group simultaneously named them hypocretin after their hypothalamic location and sequence homology to secretin [38].

Figure 1. Schematic representation of the orexinergic system.

The orexin system is comprised of two G-protein coupled receptors (GPCR): orexin-1 (OX₁R) and orexin-2 (OX₂R). Two neuropeptides: orexin-A and orexin-B, are cleaved from the precursor protein prepro-orexin. The OX₁R is selective for OXA, whereas OX₂R is nonselective for OXA and OXB. However, OXB has a 10-fold higher affinity for the OX₂R than the OX₁R. The OX₁R couples to the G_q class of G protein, whereas OX₂R couples with G_i or G_o. Adapted with permission from PR Holland (2006) [35].

The two neuropeptides, OXA and OXB, are largely conserved across rats, mice, and humans [39] and secreted from neuronal cell bodies which are almost exclusively located in the lateral hypothalamus (LH) and perifornical regions of the hypothalamus. The orexinergic neurons represent only ~20% of the neurons in the LH, yet have diffuse projections across the central nervous system (CNS) including to the PAG [40], nucleus accumbens [41], hippocampus [42], thalamus [43], sensory trigeminal neurons [44], and the spinal and trigeminal dorsal horns [42,45], with particularly dense input to the locus coeruleus and tuberomammillary nucleus. In addition to orexin, orexinergic neurons also secrete glutamate [46] and dynorphin [47]. In agreement with the widespread orexinergic projections, orexin receptors are also found to be expressed extensively throughout the CNS [39,48]. This diverse projection pattern and receptor expression has implicated the orexins in a variety of functions from sleep-wake [49], neuroendocrine and autonomic regulation [50], reward-seeking behavior [51], stress [52], feeding behavior [53], and nociception [54] (Figure 2).

Figure 2. Orexinergic projections and their proposed functions.

The orexins are almost exclusively synthesized in the lateral hypothalamus (LH) and project to many areas of the CNS including brainstem and cortical regions. The orexins are proposed to be involved in a variety of functions from reward, feeding, cognition, to nociception. LC = locus coeruleus; LH = lateral hypothalamus; vlPAG = ventrolateral periaqueductal gray; VTA = ventral tegmental area; Ox = orexin; NTS = nucleus tractus solitarii; NAc = nucleus accumbens; PVH = posterior ventral hypothalamus; CeA = central amygdala. Adapted with permission from P Sureda-Gibert [55].

1.2. Orexin in sleep-wake regulation

Although their original discovery was linked to appetite and feeding behavior, the orexins gained further attention for their potential role in sleep-wake regulation and arousal. Initial studies demonstrated that electrical stimulation of the LH led to increased wakefulness, and lesions of the LH resulted in sleep [56,57], leading to their identification as a potential arousal-promoting peptide [58]. Orexinergic neurons inhibit sleep promoting regions including the ventrolateral preoptic (VLPO) nucleus of the hypothalamus [59] to induce wakefulness and conversely GABA-ergic neurons in the VLPO innervate the LH to promote sleep [60]. Direct injection of OXA during the light period (sleep) in rats results in an increase in wakefulness [61] again highlighting their function as a wake-promoting or wake-stabilizing neuropeptide. Orexin is thought to mediate its own expression from neurons by forming a positive-feedback circuit which maintains the orexinergic system at a high level of activity over long periods, resulting in the stabilization of arousal [62], further highlighting their role as wake-stabilizing peptides. In support of this role, several dual-orexin receptor antagonists (DORAs) have been developed and have shown promise in sleep initiation and maintenance, preserving the typical sleep architecture in patients with insomnia [63].

Moreover, support of the orexins’ involvement in sleep/wake regulation came from studies of canine narcolepsy [64]. Narcolepsy is a sleep disorder characterized by inappropriate transitions between sleep and wake [65] as well as daytime somnolence, sleep paralysis, and cataplexy. Narcolepsy patients exhibit an 80–90% reduction in orexinergic neurons in the LH [66] and patients with narcolepsy with cataplexy – narcolepsy type 1 (NT1), demonstrate decreased levels of OXA in cerebrospinal fluid (CSF) [67]. Of interest, narcoleptic patients display an increased prevalence of migraines compared to the general population (females 44.4%, males 28.3% vs 16–25% and 7–8% in the general population) [68], and children with migraines were demonstrated to show a greater risk for developing narcolepsy in a prospective study [69]. A plausible neuroanatomical basis for this may involve diminished orexinergic regulation of PAG, dorsal raphe nucleus (DRN), and locus coeruleus networks. Indeed, these regions are also thought to be involved in the generation, modulation, and cessation of rapid-eye-movement (REM) sleep [70,71], and alterations in REM sleep are reported in narcolepsy [72,73].

More generally, orexin levels in the hypothalamus show circadian variation in mice and humans [74,75] and are maximal during wake, followed by in rapid-eye-movement (REM) sleep and slow-wave sleep [74,76]. Sleep deprivation increases c-Fos activation in orexinergic neurons [77], and sleep disruption is reported to be a key trigger for migraine and CH attacks [78,79].

1.3. Orexin’s involvement in nociception/trigeminal nociception

Orexinergic neurons project to the brainstem including the periaqueductal gray (PAG) [80], nucleus raphe magnus (NRM), rostroventromedial medulla (RVM), and superficial lamina of the spinal cord [81]. These regions have been shown to be involved in the trigeminal nociceptive processing [82] suggesting that the orexins may also modulate headache pathophysiology. In particular, the observation that the orexinergic system projects from the hypothalamus to the PAG – a key structure involved in descending pain modulation, also suggests the possibility of modulation of descending inhibitory pathways, or alternatively, through the direct release of orexin at the spinal cord [54,83].

1.3.1. Orexin’s involvement in nociception/pain

The role of the hypothalamus, orexin receptors, and orexin peptides in general nociceptive processing in animal models is contradictory. Direct nociceptive activation excites orexinergic neurons in freely behaving mice [54]; however, both anti-nociceptive and pro-nociceptive effects for orexin have been demonstrated. For example, the stimulation of the hypothalamus can be analgesic [84,85], and inhibits the response of spinal cord neurons to noxious stimulation [86]. Destruction of the ventro-medial-posterior hypothalamus results in transient analgesia [87] and triggers increased neuronal activation in the PAG [88]. In behavioral pain models, both OXA and higher doses of OXB delivered intrathecally resulted in increased hind-paw mechanical withdrawal thresholds in a rat diabetic neuropathic pain model, and this effect was blocked by pre-treatment with an OX₁R antagonist. However, this was not the case for healthy rats without diabetic neuropathy [89]. In contrast, intrathecal OXA, but not OXB, normalized hind-paw mechanical withdrawal thresholds in rats with sciatic nerve injury [90]. OXA, but not OXB, was analgesic in the hotplate and formalin test [91] in rats and concomitantly decreased the number of c-Fos positive cells in laminae I-II of the spinal cord [81,92] while intracerebroventricular OXA was shown to have antinociceptive effects in assays of thermal, mechanical, and chemical pain [93].

1.3.2. Orexin’s involvement in trigeminal nociception/head pain

Focussing on trigeminal nociception, microinjection of OXA into the posterior hypothalamus inhibits trigeminal nociceptive responses to dural electrical stimulation as well as spontaneous activity, whereas OXB elicited increased responses to stimulation, suggesting a pro-nociceptive role [94]. Similarly, OXA inhibited electrically induced CGRP-dependent vasodilation in an animal model of trigeminovascular activation, and this was reversed by pre-treatment with an OX₁R antagonist, while OXB had no effect [95]. Whereas conversely, injection of OXA and OXB into the raphe nucleus magnus was shown to facilitate activity in the trigeminal cervical complex, which was thought to be driven mainly by OX₂R activation [96]. In an animal model of migraine-related mechanical hypersensitivity evoked by the administration of the clinically experimental migraine trigger nitroglycerin (NTG) [97], an OX₁R antagonist injected into the amygdala resulted in increased anxiogenic responses to NTG in rats, but had no impact on thermal hyperalgesia [98]. In another study, delivery of OXA into the vlPAG attenuated NTG-induced hyperalgesia and photophobia, and this was prevented with administration of an OX₁R antagonist [99]. Similarly, local OX₁R antagonism in the vlPAG aggravated NTG-induced anxiety and social conflict, whereas administration of OXA attenuated this effect [100]. Therefore, it is evident that activation of the OX₁R or administration of OXA may be antinociceptive or involved in preventing associated headache symptomatology, as blocking this receptor seems to attenuate such effects.

The reason for such conflicting results is likely due to the descending inhibitory system being differentially activated during conditions of appropriate stimuli, e.g., under inflammatory or chronic pain conditions and not during acute nociceptive stimuli. Alternatively, the differential functions of orexin receptors could be due to the finding that many brain regions express both OX₁R and OX₂R receptors, yet discrete areas including the locus coeruleus show selective expression of a single receptor subtype [101]. Moreover, discrepancies may be due to the location of orexin or antagonist delivery. Its actions in the amygdala may involve the affective components of pain such as stress and anxiety. Indeed, the amygdala may modulate threat learning through the locus coeruleus [102]. This is of interest as headache is co-morbid with anxiety disorders [103], suggesting potential mediation through orexinergic-noradrenergic pathways. In support of this modulation of noradrenergic signaling in the locus, coeruleus has been shown to exert divergent effects on migraine-related cortical and spinal excitation [104]. Thus, the orexinergic system is potentially involved in the modulation of nociceptive transmission and the pathophysiology of headache, but further work is needed to identify the specificity of these pathways and underlying mechanisms.

1.4. Orexin and cluster headache

It is widely accepted that the hypothalamus plays a role in CH, with hypothalamic activation observed during attacks [18,105,106]. Initially, reports on a link between CH and the orexin system were circumstantial. This was based predominantly on the striking observation that CH attacks show a clear circadian and circannual rhythmicity, highlighting a potential role of the SCN of the hypothalamus [26]. Specifically, most CH attacks occur at the same time each day, and bouts tend to occur at the same time each year (spring and/or autumn) [27]. However, no significant alterations in the CSF orexin levels in CH have been observed [107], with one study reporting slightly lower levels of orexin [108]. However, CH patients do show endocrinological changes and lack typical circadian regulation of certain hormones which are thought to be under hypothalamic control, such as cortisol, melatonin, prolactin, and testosterone [109].

Studies have also highlighted potential genetic links between orexin and the risk of CH occurrence. The G1246A polymorphism of the OX₂R gene HCRTR2 is linked to an increased risk of CH [110–112]. Conversely, alternate studies have reported no such link between the G1246A polymorphism and CH [113] which was not linked to treatment response [114], thus the link between genetic orexin factors and CH is unclear. Interestingly, verapamil, a first-line preventive for CH, results in alterations to the hypothalamic CLOCK gene PER2 expression and sleep timing in mice [115]. However, a study demonstrated that the CLOCK gene T3111C polymorphism was not linked to CH [116], suggesting that other hypothalamic CLOCK genes, such as PER2 or those regulating orexin expressions such as DEC2 [117] may be more important in CH pathophysiology.

Another potential link between the orexin system and CH is through REM sleep. Polysomnographic studies of CH have reported reduced REM sleep and longer REM sleep latency compared to healthy controls [118,119]. However, whether this is due to the pain from CH attacks waking up the individuals during the night is unclear, and some studies have shown no link with REM sleep [120]. Orexin is thought to play a modulatory role in REM sleep via the sublaterodorsal tegmental nucleus [121], suggesting a potential link between REM sleep modulation, CH, and orexin.

1.5. Orexin and migraine

Given the role of the orexins in homeostatic functions and nociceptive processing, they have also been implicated in migraine pathophysiology. More generally, the hypothalamus is active both prior to and during attacks [16,17], and in the premonitory phase of a migraine attack, the hypothalamus is more responsive to trigeminal nociceptive stimulation [122]. More generally, changes to the light–dark cycle, which may be under hypothalamic control such as jet lag and shift-work are known triggers for attacks [78,123,124], while abnormal fatigue, likely involving dysfunctional arousal regulation, is present in up to 83% of the patients [125]. Moreover, mutations in circadian CLOCK-related genes regulated by the SCN have been associated with increased migraine penetrance in specific families and increased migraine-related phenotypes in preclinical models [126]. Similar to CH, migraine patients also display alterations in sleep architecture, with evidence suggesting reduced REM sleep in both children and adults with migraine [30]. This points to potential hypothalamic orexinergic dysfunction in migraine as the orexins are proposed to play a role in REM sleep modulation.

Evidence for direct alterations of the orexinergic system in migraine, however, is lacking. Higher CSF concentrations of orexin have been reported in chronic migraine [127]. However, this evidence is inconsistent as lower OXA levels have also been observed [128]. Whilst, the HCRTR2 G1246A polymorphism has been potentially linked to CH, studies have shown no significant contribution to the pathophysiology of migraine [129,130]. However, studies have linked the risk of migraine to SNPs in the HCRTR1 gene [131,132].

1.6. Orexin’s involvement in accompanying and non-pain symptoms of headaches

Importantly, the hypothalamus has diverse projections and is involved in a range of non-pain related symptoms attributed to headache disorders. For example, patients with migraine may experience associated symptoms such as anxiety, depression, abnormal fatigue, and food cravings, highlighting potential homeostatic dysfunction. The orexins activate the hypothalamic–pituitary axis, which is involved in the body’s response to stress and secretion of hormones [50,133]. Orexins, particularly OXA, are implicated in anxiety behavior [134,135]. Interestingly, CSF OXA levels correlate with anxiety measures in chronic migraine [136], but importantly in this study, the observed orexin levels were not different to controls. This may reflect that only OXA peptide was studied, whereas other studies have looked at orexin levels more generally.

As mentioned previously, the orexins are thought to be involved in feeding behavior and appetite regulation. Descending hypothalamic inputs are integrated with peripheral inputs from the gastrointestinal system [137]. Orexins modulate metabolism as well as nociceptive transmission [138], thus dysfunction to the orexinergic pathway may be a risk factor for co-morbid obesity and migraine [139] as well as being involved in the food cravings associated with migraine attacks.

1.7. Clinical impact

Whilst it is evident that orexin may be important in headache pathophysiology and therefore as a potential therapeutic target, very few clinical studies have been conducted based on orexin, despite their approval and efficacy for associated conditions such as insomnia and narcolepsy [63,140,141]. In preclinical models, a precursor of suvorexant, a DORA (DORA-12) was shown to inhibit trigeminal nociception in response to electrical stimulation of dural trigeminal afferents and increased the threshold for inducing cortical spreading depression, the neurophysiological correlate of migraine aura [142], suggesting their potential utility in treating headaches [143]. However, a small clinical trial found no significant reduction in monthly migraine days after the administration of a DORA – filorexant [144]. However, this negative finding could be the result of a variety of factors. For example, the DORA was given at night when orexinergic signaling is the lowest. It is possible that orexin antagonists delivered at other circadian times could be beneficial; however, issues with somnolence persist, supporting the use of selective OX₁R agonists. Indeed, the orexin itself and peptides involved in headaches such as CGRP show clear diurnal variation [75,145]. Moreover, the short half-life of orexin could be responsible for this null finding. Furthermore, DORAs, as the name suggests, antagonize both orexin receptors, while OXA which preferentially binds to the OX₁R, has demonstrated antinociceptive actions in preclinical models of migraine, whereas OXB demonstrated largely pronociceptive effects. Therefore, the use of more selective orexin receptor agonists may prove fruitful, especially those targeting the OX₁R.

Understanding the precise conditions under which orexins are antinociceptive or pronociceptive and the contribution of different orexin receptors to nociception is required to fully elucidate the relationship between orexin, pain, and headaches. Critically, future studies should also explore the most appropriate route/time for delivery, tailored to different individual chronotypes (migraine patients are more likely to have extreme chronotypes that predict attack occurrence timing) [146] and endogenous circadian periods, especially given the symptoms and severity of insomnia (which DORAs were originally developed for) differ depending on chronotype [147]. As such, the antagonism of selective orexin receptors may prove useful in headache patients with co-morbid insomnia, a fact suggested in a post-hoc analysis of clinical data [144]. Chronotherapy studies [148] should be conducted to establish this, thus further research investigating the temporal dynamics of orexin’s antinociceptive properties is warranted.

2. Conclusions

Orexins are important neuropeptides involved in sleep-wake regulation and a variety of other homeostatic functions. Their relevance for headache pathophysiology arises from studies of perturbed homeostatic modulation during migraine attacks, genetic links, as well as the circadian and circannual rhythmicity of headache disorders. This rhythmicity points to the role of the SCN which may influence the orexinergic system. The link between the orexin-deficient sleep disorder narcolepsy and headache disorders also suggests a common pathophysiology related to the orexin system. Taken together, the evidence suggests a prominent role of the orexins in headache pathophysiology. However, further research is needed to understand the specificity of individual orexin peptides for nociceptive processing and headache neurobiology. Further large-scale clinical trials of selective orexin receptor agonists, specifically the OX₁R are required, taking into account circadian fluctuation of endogenous orexin, patient chronotype, and co-morbid sleep disorders.

3. Expert opinion

Primary headaches remain one of the most disabling neurological disorders, especially in young women where migraine alone is ranked the number one cause of years lost to disability [149]. The body of evidence available points to a critical role of orexinergic signaling in primary headaches, most prominently migraine and CH, although different receptor subtypes may play a critical role in both. Of particular importance, migraine-related fatigue and the repetitive circannual/circadian rhythmicity of CH remain as key untapped phenotypes in terms of novel therapeutic discovery. Focussing on migraine-related abnormal fatigue, which can occur hours to days before the headache phase [150], it is likely that dysfunctional arousal networks are involved. Patients commonly report feeling washed out and lethargic, which significantly impacts normal everyday tasks and worsens the attack-related disability. Development of novel arousal promoting and antinociceptive compounds, such as OX₁R agonists or positive allosteric modulators therefore hold significant promise, aided by the recent design and synthesis of such compounds [151], which previously proved unsuccessful. Overall, novel therapies including those targeting calcitonin gene-related peptide have shown some potential in tackling migraine-related fatigue [152], in contrast to the ditans (5-HT_1F receptor agonists) which result in somnolence. However, there remains a major therapeutic gap for primary headaches, and we propose that targeted orexinergic modulation is a promising avenue for future clinical development, in conjunction with carefully considered administration routes. Systemic delivery can be hampered by poor CNS penetration, degradation, and unwanted peripheral side effects. This is particularly relevant for intranasal routes which provide rapid CNS access, although this can be impacted by several factors including peptide size and lipophilicity. Indeed, intranasal delivery of orexin A in rats resulted in increased tissue-to-blood ratios when compared to systemic delivery [153], and critically, orexin A levels were found to be highest in the trigeminal nerve, olfactory bulb, and hypothalamus, key areas involved in the pathophysiology of primary headaches.

Article highlights

• The orexinergic system is postulated to play a role in nociceptive transmission as well as in a range of homeostatic functions including arousal, endocrine regulation, and appetite.

• Modulation of the orexinergic system in preclinical animal models may directly facilitate or attenuate trigeminal nociceptive processing, highlighting the relevance of the orexins in headache pathophysiology.

• Patients with migraine and cluster headache show potential alterations in orexin levels and genetic factors related to the orexinergic pathway.

• There is limited clinical evidence for the efficacy of orexin-based therapeutic targets in primary headaches.

• Future research should focus on selective activation of the orexin 1 receptor via selected routes.

Declaration of interest

PR Holland reports, unrelated to this work, grants from Amgen, Eli Lilly and Company, Kallyope and Bristol-Myers Squibb as well as honoraria and travel expenses in relation to educational duties from Allergan, Novartis, Teva, Pfizer and Almirall. J Hoffmann reports honoraria for consulting activities and/or serving on advisory boards and/or for giving lectures/presentations from AbbVie, Allergan, Autonomic Technologies Inc., Cannovex BV, Chordate Medical AB, MD-Horizonte, Eli Lilly and Company, Hormosan Pharma, Lundbeck, Novartis, Sanofi and Teva. He also holds stock options from Chordate Medical AB and has received personal fees for Medico-Legal work from NEJM Journal Watch, Oxford University Press, Quintessence Publishing, Sage Publishing and Springer Healthcare. He also reports a research grant from Bristol Myers Squibb. All these activities are unrelated to the submitted work. EC Stanyer received PhD funding from the King’s College London MRC-DTP Programme grant no: MR/N013700/1. EC Stanyer also reports honoraria from Lindus Health Ltd and FCLabs Ltd unrelated to this work. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Reviewer disclosures

Peer reviewers in this manuscript have no relevant financial or other relationships to disclose.

Additional information

Funding

The authors are funded by the Medical Research Council via grants MR/P006264/1 and MR/N013700/1.