The Mode of Action of Migraine Triggers: A Hypothesis

Abstract

Objectives.— To review conjectured modes of action of migraine triggers and to present a new hypothesis about them.

Background.— Migraine attacks are initiated in many migraineurs by a variety of “triggers,” although in some patients no external trigger can be identified. Many triggers provoke attacks with such a short latency that only some kind of neural mechanism can explain the triggering.

Results.— We present here a hypothesis that the pain of migraine has its ultimate origin in the cortex, but that the immediate generator is in the brainstem. Our hypothesis is that most migraines have triggers that produce excitation of cortical neurons and that this directly causes withdrawal of descending sensory inhibition originating in the brainstem. A wide range of evidence from the literature that cortical activation induced by a number of different mechanisms often produces headache is presented to support this notion. Several nuclei in the brainstem appear to participate in the selective control of trigeminovascular sensation through descending inhibitory mechanisms that arise in the cortex. In this review we focus on 2 of them, the periaqueductal gray matter and nucleus raphe magnus. Our own past results and those of others show that this inhibition is specific for craniovascular sensation and involves the neurotransmitter 5-hydroxytryptamine. Finally, we summarize our own recent experiments, which show that cortical activation by migraine triggers (including cortical spreading depression) inhibits neuronal discharge in the brainstem and facilitates trigeminovascular sensation.

Conclusion.— If the hypothesis can be proven and the neurotransmitters involved in the hypothetical trigger pathway can be identified, it may be possible to develop novel migraine preventative therapies.

Abbreviations:

5-HT

5-hydroxytryptamine

CGRP

calcitonin gene-related peptide

CSD

cortical spreading depression

DC

direct current

ECT

electroconvulsive therapy

LC

locus coeruleus

NO

nitric oxide

NRM

nucleus raphe magnus

NTS

nucleus tractus solitarius

PAG

periaqueductal gray matter

PET

positron emission tomography

SP

substance P

TMS

transcranial magnetic stimulation

In this review, we present the hypothesis that most migraines have triggers that produce excitation of cortical neurons and that this excitation directly inhibits neurons in 2 brainstem nuclei: the periaqueductal gray matter (PAG) and the nucleus raphe magnus (NRM). These nuclei then release their ongoing descending control of dura mater sensation in the trigeminal nucleus and cause normal sensory traffic from the dura to be perceived as migraine pain.

Migraine is one of our most common and debilitating organic diseases, yet bafflingly has no apparent pathology. It is a condition involving the “trigeminovascular system,” in which the cranial vasculature, the trigeminal sensory system, and the central nervous system all play a part. Its cardinal features are headache, nausea, and neurological disturbances. The pain is unique in its apparent source, its lack of apparent pathological cause, the variable nature of both the factors that trigger it and the drugs that prevent it, its intensity, its periodicity, and its response to specific pharmacological agents. These are features unique to migraine. No other pain, not even other trigeminal pain, displays such a constellation of qualities.

If migraine could be prevented, suffering could be alleviated and money and lost time could be saved, but we do not yet understand how to prevent migraine. Rational migraine prevention would be greatly improved if we could understand the causes of migraine pain, but we do not yet know the cause. However, we do know that a wide variety of “triggers” may precipitate an attack. How can this multiplicity of triggers always produce the same intense pain? Are all migraines triggered? Why is this pain specific for blood vessels and the dura mater? Is there some way in which we can intercede to stop triggers progressing to a migraine headache?

The pain of migraine is perceived to arise from the large cranial blood vessels and the dura mater. Sensory innervation of these structures is by the trigeminal nerve, the fibers of which terminate in the trigeminal nucleus caudalis. One of the enduring mysteries of migraine is its lack of apparent pathology – a century of research has so far failed to find any convincing defects in any of the sensory or vascular systems believed to be involved in producing the headache. This has led to the belief that migraine is not an organic disease, but it assuredly is an organic disease. If there is no overt pathology in the trigeminovascular system, then could it be that the pain arises in some change in the way in which normal sensory traffic from these structures is misinterpreted?

HYPOTHESES

According to Cornelius Celsus, friend of the emperor Tiberius, migraine is caused by drink or the heat of the sun – quoted by Critchley,¹ who emphasized also the statement of the Persian Avicienna that “little does it concern the patient that there is an underlying cause . . . if the practitioner is unable to relieve his pain.” Two millennia later, this remains an essential dilemma for migraine therapy – patients seek prevention but, since the physician cannot say how the condition arises, prevention falls back upon ad hoc methods.

Broadly speaking, theories of migraine pathogenesis have formed 2 groups, vascular and neural for well over a century, with an offspring – the neurovascular/sensitization theory – springing from them in the late 20th century. Debate between the vascular and neural camps has continued for over a century, much of it oriented toward illuminating the mode of action of anti-migraine drugs.² The pendulum has swung from neural to vascular and back again and the rationale and search for migraine prevention has swung with it. Extensive summaries of possible pathophysiological mechanisms can be found in “Biomarkers for Migraine”^3-9 and in Sanchez-del-Rio et al¹⁰ and Goadsby.¹¹

Vascular Factors.— Pain originating inside the head must necessarily arise from its large blood vessels or the tissue which surrounds the blood vessels of the dura mater, because these are the only structures to receive substantial sensory innervations.¹² A vascular origin for migraine pain is therefore an attractive one and it can be traced back to observations that headache pain was synchronous with arterial pulsation¹³ and could be relieved by compression of the carotid artery,¹³ that there were obvious changes in cutaneous vascular tone such as flushing,¹⁴ or pallor¹⁵ during headaches, and that powerful cranial vasoconstrictors such as the ergot alkaloids were effective in relieving the pain.¹⁶ The work of Ray and Wolff,¹² who demonstrated the remarkable pain-inducing effects of distension of dural blood vessels, added strongly to this impetus.

If the pain arises from cranial vessels, how is it initiated? Various theories have had it that the vessels constrict,¹⁷ that they dilate,¹⁶ that normal circulation is diverted through arteriovenous anastomoses,¹⁸ and that blood or vascular tissue release either dilators, constrictors, or pain-producing substances, such as 5-hydroxytryptamine (5-HT) by platelet aggregation¹⁹ or histamine by mast cell degranulation.²⁰ Theories of the last type have sometimes been described as humoral.

Neural Factors.— Neural theories of migraine have their origin in a statement by Sir Edward Liveing²¹ that migraine is a tendency on the part of the nervous system centers to the irregular accumulation and discharge of nerve-force– his so-called “nerve storms,” although Samuel Tissot had pinpointed the nervous system 90 years before.²² Generally, “neural theory” means “central nervous system theory,” although neural theories grade into the peripheral nervous system also. Most neural theories have adhered to Liveing's idea that some form of increased neural activity is the ultimate cause of migraine pain. Brain areas or nuclei which have come under suspicion include the cortex,²³ the PAG,²⁴ the raphe nuclei, including the NRM,^25-27 the locus coeruleus (LC),²⁸ the hypothalamus (HYP),^29-31 the superior salivatory nucleus,³² and the thalamus.³³

Neurovascular, Sensitization, and Inflammatory Factors.— In the 21st century, neurovascular theories of migraine have tended to single out the relationship between cranial sensory nerves and the cranial vasculature, but there is also a long history of investigation of neurovascular mechanisms involving the autonomic nervous system as well. The peripheral terminals of trigeminovascular sensory fibers are greatly enriched in pain-producing neuropeptides such as calcitonin gene-related peptide (CGRP) and substance P (SP)^34,35 and can be depolarized by them, thus forming a positive feedback loop, which is ideally suited to initiating runaway nociception in the dura mater and cranial blood vessels. It has been thought that processes occurring at either the peripheral or central fiber terminals of trigeminovascular sensory nerves may produce pain because the terminals become depolarized,³⁶ because they may release inflammatory and pain-producing mediators,³⁷ or because neurotransmission, particularly at the central ends, has become facilitated by the inflammatory or neuropathic processes already initiated.³⁸

In the long debate between these competing hypotheses, the advantage currently seems to lie with the neurological theories. Of these, the cortical and brainstem versions seemed equally attractive to us. Both of these ideas might be correct simultaneously – they may be complementary rather than competing explanations for migraine genesis and they may be causally linked. We theorized that different pathways arising from different neurons in different cortical regions, each itself activated by a different trigger, might converge on the brainstem and inhibit the ongoing activity of its neurons, thereby releasing dural sensation from neuronal inhibitory control and result in the common symptomatology of migraine headache. If there are different convergent pathways in different migraineurs, this might explain why there is no universally effective prophylactic drug for migraine, but it suggests where to look for one to act and what type of drug it might be.

Significant pieces of this puzzle have already been solved,^26,39,40 some through our own published experiments.⁴¹ We are now trying to assemble the remaining pieces into a coherent picture of migraine initiation. If this new hypothesis is proven correct, novel therapies to prevent migraine from developing could perhaps be produced.

MIGRAINE TRIGGERS

Migraine attacks are initiated in many migraineurs by a variety of “triggers”; some three quarters of migraineurs report that their migraines are triggered at least some of the time.⁴² Sometimes the trigger appears to be an internal clock. In other cases the attack seems to be triggered by excessive afferent stimulation such as flickering light, noise, or strong smells, or in response to stress or (paradoxically) the relief of stress. In “migraine with aura,” headache may be triggered by a particular neural event, which is thought to be “cortical spreading depression” (CSD). Food or drink, especially those containing vasodilators or those that affect biogenic amines in the CNS, may also trigger migraine. Importantly, drugs which deplete the brain of the neurotransmitter serotonin are powerful triggers for migraine. Hormonal variations have a strong influence – the periodicity of migraine headache is related to the menstrual cycle in about 60% of female patients. Some trigger factors appear to act primarily on the cranial blood vessels, such as alcohol, glyceryl trinitrate, and vascular irritation (as occurs in angiography). Environmental changes, especially in temperature and barometric pressure, can also trigger an attack.⁴² For many patients no external trigger is apparent. Many triggers provoke attacks with such a short latency that only some kind of neural mechanism can explain the triggering.

According to Lance and Goadsby,⁴³“migraine may be regarded as a hereditary tendency to have headache characterized by associated signature symptoms such as nausea or sensitivity to light. The basis of this predisposition is instability in the control of pain coming from the intracranial structures and sensitivity to changes in the central nervous system.”^11-61 Lance and Goadsby hypothesize that the brain of susceptible subjects has a low “migraine threshold” and attacks are initiated in their brains by a variety of “triggers.” They go on to discuss the predisposing factors, including genetics, magnesium deficiency, changes in excitatory amino acids, neurophysiological changes, the hypothalamic-pituitary axis, the endogenous pain control system, and vascular reactivity – all in light of the possibility that some malfunction in one or other could predispose the brain to migraine. In our view most of this predisposition will be manifested in the cortex.

CEREBRAL CORTEX

That migraine might originate in the cerebral cortex dates from 1873 (Liveing's “brain storms”²¹), but it is only in the past few decades that the idea has been seriously reconsidered and is yet to be rigorously tested in experimental animals. Evidence for the involvement of the cortex has been reviewed by Welch⁴⁴ and in a lecture by Schoenen,⁴⁵ who suggest that there is a strong case implicating the cortex as both a “culprit” and a “bystander” in migraine pathophysiology. Coppola et al⁴⁶ have recently reviewed the notion of cortical hyperresponsivity in migraineurs and have suggested that “a thalamo-cortical dysrhythmia might be the culprit.”

There is abundant evidence that the cortices of migraineurs are “under inhibited”⁴⁷ or hyperexcitable inter-ictally.^48-50 This predisposes them to the development of aberrant cortical discharge or of CSD,^51-53 which occurs more easily in a hyperexcitable cortex,^54-57 especially that of females.^58,59 Migraineurs with aura exhibit a larger cerebrovascular response to repetitive visual stimulation compared with headache-free subjects,⁶⁰ an indication that there is an increase in visual cortical reactivity to light stimuli in migraineurs. Recent evidence for a potential mechanism for hyperexcitability in migraine comes from the demonstration of a number of genetic mutations in some families with familial hemiplegic migraine,⁶¹ including on the α_1A– subunit gene on chromosome 19 which encodes for the neuronal P/Q Ca⁺⁺ channel⁶² and on the ATP1A2v gene, which encodes the catalytic α2 subunit of Na+, K+-ATPase.⁶³ Also, abnormalities in neuromuscular transmission in some patients with migraine with aura further suggest a generalized abnormality of Ca⁺⁺ channels in migraineurs which, if it is also present in the cortex, may possibly predispose them to CSD and other forms of cortical hyperexcitation.

A range of external and internal influences can produce activation of cortical neurons and some of these do indeed seem to produce headache. These include epilepsy, electroconvulsive therapy (ECT), transcranial magnetic stimulation (TMS), and transcutaneous low-voltage direct current (DC) stimulation of the cortex. The headache from most of these procedures could arise from a number of sources, including direct stimulation of sensory fibers, but it often displays features such as a delay in onset, an intracranial origin, and its response to drugs, which suggest that something more than directly produced peripheral nociception is involved.⁶⁴ While the effect of each might be explained away by these other mechanisms, collectively they fit best as jig-saw pieces in the migraine genesis puzzle.

Epilepsy.— The relationship between migraine and epilepsy has long been discussed⁶⁵ and it is thought that they may be related, at least partly, through common calcium channel neuropathies.^66-68 Headache occurs post-ictally in 50% of epileptic attacks⁶⁹ and such headaches are classified as migrainous in 34% of patients and as tension-type in 34%.⁶⁴ Furthermore, the antiepileptic agents topiramate and sodium valproate, which inhibit cortical discharge,^70,71 are among the more successful migraine prophylactic drugs.⁷²

Electroconvulsive Therapy.— Headache, sometimes even described as “migraine,” is a commonly reported side effect of ECT (reviewed by Datto et al⁷³). In 1994, Weiner et al⁷⁴ surveyed 98 patients retrospectively about their experiences with headache prior to and following ECT. Of the 54 patients who submitted properly completed questionnaires, five reported new onset of headaches following ECT, four reported exacerbation of a previous headache problem, and two reported their headaches improved. The patients experienced changes in the character or location of pain, with a tendency to progress from tension-type to migrainous headache. Headache following ECT often does share features with migraine, including its susceptibility to treatment with triptans⁷⁵ and its prevention with known migraine prophylactic agents. In a 2001 report, Hawken et al⁷⁶ described “migraine” occurring in a patient undergoing ECT and reported that the headache could be prevented by prior administration of propranolol, used for many years as a migraine-preventative drug. It is interesting that propranolol did not apparently interfere with the effectiveness of the ECT treatment – meaning, presumably, that propranolol must act as a migraine preventative somewhere other than the cortex, perhaps “downstream” in the pathway hypothesized here.

Direct Current Stimulation.— DC electrical stimulation of the human cortex (for depression) raises the excitability of cortical neurons,⁷⁷ especially those in the human visual cortex.^78,79 In pilot experiments, DC electrical stimulation, which increases the susceptibility to, and the velocity of spread of, CSD⁵⁷ resulted in headache in three quarters of the participants (Colleen Loo, unpublished observations).

Transcranial Magnetic Stimulation.— Recent experiments with TMS have shown that it increases cortical excitability and can also produce headache as a side effect.⁸⁰ TMS has now been advocated as a potential therapy for migraine,⁸¹ because it can interrupt the spread of CSD, supposedly by creating a refractory barrier in the cortex.

Cortical Activation by Known Migraine Triggers.— While it is natural to suppose that many other triggers, particularly sensory triggers, would activate cortical neurons, this supposition needs to be tested. Light flash stimulation definitely produces generalized activation of neurons in the primary visual cortex; this activation is accompanied by vascular changes similar to those which occur in the visual cortex during migraine.⁸² As mentioned below, CSD is a special case of migrating cortical activation, and causes discharge rates of cortical neurons to rise by up to two orders of magnitude.⁸² Evidence that other migraine triggers also activate cortical neurons comes from the imaging literature. Inhalation of odorants, for instance, activates neurons in the primary olfactory cortex.⁸³ Sounds, including speech sounds, activate neurons in many areas of the cortex, including the temporal lobe.⁸⁴ The female brain under estrogen shows marked increases in perfusion in cortical areas involved in cognitive tasks⁸⁵– this is almost certainly a result of increased cortical activity and may be related to higher frequency of migraine in women and to the link between hormonal levels and the initiation of migraine. None of this is surprising – the question really is whether these triggers produce activation of sufficient magnitude to have an influence on pathways which descend through the brainstem. In the case of CSD and light flash, they clearly do,⁴¹ but the case for other triggers is yet to be proven.

Cortical Spreading Depression.— Cortical activation is a prominent feature of CSD which, although it is labeled as “depression,” produces its initial neurological effects such as the scintillating scotoma, through cortical excitation. The aura in migraine with aura has been ascribed to CSD, which is unquestionably an important component of this type of migraine.^86,87 The aura shares many of its characteristics with CSD, such as the velocity of its spread and the pattern of changes in cortical blood flow associated with it. It commonly lasts 15-60 minutes and the headache usually begins after the aura. Some patients who experience both aura and headache may also experience aura without headache. In these patients, the headache may disappear in later life and some 1-2% of patients only ever experience aura without headache.^88,89

The nature of any link between CSD and migraine pain has hitherto been rather mysterious. Past hypotheses have included activation of meningeal afferents by potassium⁹⁰ or other pain-inducing substances and induction of painful ischemia.⁹¹ In 1993, Moskowitz et al reported that repeated waves of CSD in rats induced by multiple cortical micro-injections of KCl increased the expression of c-fos, a marker indicative of activation of sensory neurons, in the trigeminal nucleus caudalis and in the PAG.⁹² Subsequent work by Ingvardsen et al⁹³ raised the possibility that the expression of trigeminal c-fos after CSD may be associated more with the noxious stimulation caused by multiple KCl injections than with any consequent CSD. In our laboratories, we showed that CSD can reduce blood flow in the dura mater, possibly to ischemic levels.⁹¹ More recently, Takano et al have suggested that migraine pain may arise from CSD because of the hypoxia, edema, and inflammation that CSD can produce.⁹⁴ Spreading depression also occurs in the hippocampus and, like CSD, activates neurons in the trigeminal nucleus.⁹⁵ It has been hypothesized recently that the activation of matrix metalloproteases by CSD may be the mechanism by which CSD produces this trigeminovascular edema.^96-98 A recent study by Ayata et al⁹⁹ demonstrated that long-term treatment with a number of migraine prophylactic drugs including topiramate, valproate, propranolol, amitriptyline, and methysergide decreased the frequency of, and increased the threshold for, the occurrence of CSD in an animal model. As noted by the authors, these agents are equally effective in migraine without aura.

Is CSD Just Another Migraine Trigger?— In our hypothesis, CSD is relegated to the status of being another migraine trigger, which produces headache in the same way as do other triggers, because it activates cortical neurons. The only difference is that CSD spreads, while cortical activation produced by other triggers remains stationary. Our hypothesis is that CSD inhibits some PAG and NRM neurons via projections from the cortex, perhaps in only a particular cortical region which it invades, and leads to withdrawal of descending inhibition – ie, that it acts just like other migraine triggers. Thus, there is no need to invoke direct vascular effects peculiar to CSD in order to explain why CSD produces trigeminovascular pain,^96-98 nor to differentiate CSD from other triggers. This is not to say, of course, that CSD occurs in all migraines (it clearly does not) or that CSD invariably leads to a migraine headache (it clearly does not). This second point is addressed later in this review, under “Questions.”

Brainstem Nuclei.— The brainstem attracted our attention about 25 years ago²⁶ and there is increasing evidence that a defect or “perturbation”¹⁰⁰ of the brainstem may cause some of the symptoms of migraine. The evidence for and against the idea that the brainstem may be involved in the modulation of migraine and other pain has been summarized by Welch²⁴ and by Basbaum.^101,102 Two nuclei have been especially implicated in this defect – the PAG²⁴ and the NRM, in which many of the projections from the PAG synapse.^27,103 Both form part of the descending pain control system. The LC is also an important part of this system¹⁰⁴ and, although “defects” in it have so far not been advanced as a factor in migraine pathogenesis, it may well be important in some aspects of nociception.¹⁰⁵ In addition, the hypothalamus has been shown to be involved in the pathogenesis of cluster headache and its involvement in migraine pathophysiology has long been inferred from the association between it and a host of migraine prodromal symptoms.¹⁰⁶ The hypothalamus and the PAG are complex nuclei with internal structures that may be selectively involved in migraine, but the LC and NRM seem to be more uniform. Of these nuclei, 3 are relatively small (NRM, LC, HYP) and it has so far been hard to acquire direct imaging evidence for their involvement in migraine.

There is ample evidence to implicate the PAG and NRM in migraine, but there is little to suggest how they come to be “defective” in migraine, or how so many different triggers could each make them defective. Sensation from cranial structures may usually be kept below the pain threshold by continuous discharge of neurons in the PAG and/or NRM; the raphe nuclei in particular are characterized by slow rhythmical firing of neurons. This discharge produces descending inhibition of sensory input, particularly that from cranial structures, and so prevents the normally low discharge rate of second-order trigeminovascular neurons¹⁰⁷ from “running away” or “winding up.” According to the hypothesis presented here, cortical activation produced by migraine triggers would feed downward to the PAG and NRM and reduce the discharge rate of their neurons. This in turn would reduce the descending inhibition that they produce (selectively for blood vessel and dural sensation) and thus a migraine headache would result.

Periaqueductal Gray Matter.— The PAG plays an important role in pain, analgesia, fear, anxiety, and cardiovascular control. Neuroanatomical and functional evidence points to a connection between the trigeminal sensory system and the PAG and possibly a direct projection to the PAG from primary dural sensory afferents.

The PAG receives descending input from the cortex¹⁰⁸ and ascending primary and secondary sensory trigeminovascular input from nociceptive neurons in the trigeminal nucleus caudalis,¹⁰⁹ and sends projections to thalamic nuclei that process nociception. In the monkey, sensory afferents from the middle cerebral artery terminate in the caudal ventrolateral PAG. Of particular interest is a study that showed that electrical stimulation of the superior sagittal sinus, a vessel of the dura mater, led to a restricted expression of c-fos protein in the same area of the PAG of the cat.¹¹⁰ The PAG has a distinct columnar structure,^111,112 with components which have different functional roles and which interact. Although anti-nociception is produced by electrical stimulation throughout the PAG,^113-117 in caudal PAG there are distinct dorsal and ventral systems with several significant differences. Neurons in the caudal ventrolateral PAG project directly and indirectly to the ipsilateral superficial medullary dorsal horn¹¹⁸ and also receive projections from it.¹¹⁹ Analgesia produced by stimulation in this area of the PAG¹²⁰ has been attributed to inhibition of spinal cord neurons, possibly directly, but most probably through other nuclei, including the serotonin-containing nucleus, the NRM.¹²¹ Knight et al have shown that PAG stimulation inhibits trigeminovascular sensory neurons in the C2 region of the spinal cord.³⁹

In patients undergoing a migraine attack, the area of the brain showing the greatest increase in metabolic activity, as revealed by positron emission tomography (PET) scanning, is in the region of the PAG.¹²² This PAG activation, but not that in other areas such as orbitofrontal cortex, remains after suppression of the attack with anti-migraine drugs,¹²³ suggesting that the PAG could well play a pivotal role in initiating the migraine attack.

Migraine-like headache (sensitive to anti-migraine drugs) can be triggered by lesions, trauma, gliomas, tissue damage, demyelinating plaques, or vascular malformations in the PAG.^100,124-126 All of these “perturbations” might be expected to reduce descending PAG drive. The link between PAG abnormalities and migraine has been discussed recently by Welch,^24,127 but there is, as yet, little evidence as to what might cause such a “perturbation” in migraineurs free of apparent pathology.

The anti-migraine drug dihydroergotamine has been shown to bind to the ventrolateral PAG in the cat,¹²⁸ as has sumatriptan. Both drugs bind strongly with several G-protein-coupled 5-HT receptors including 5-HT_1A, 5-HT_1B, and 5-HT_1D subtypes, which are all present in the PAG. The normal function of these receptors is not known, nor is the origin of the fibers, which contain the 5-HT that activates them, but it is very likely to be in one of the raphe nuclei.

Supporting evidence comes from a study by Bartsch et al,¹²⁹ in which the 5-HT_1B/1D agonist anti-migraine drug naratriptan was microinjected into the ventrolateral PAG and activity in the trigeminal nucleus caudalis was monitored. Naratriptan decreased the A-δ and C-fiber-mediated responses of neurons to electrical stimulation of the dura mater. This decrease was accompanied by an increase in mechanical stimulus intensity thresholds in the dura mater. Responses to stimulation of the face and cornea were not altered by injection of naratriptan. This is the strongest evidence yet that the PAG is responsible for selective mediation of migraine pain and is a logical “pressure point” for pharmacologically selective migraine preventative therapy.

Nucleus Raphe Magnus.— It was suggested many years ago that in migraineurs “the build-up of pain characteristic of migraine could result from reduction of NRM inhibition of pain-transmission neurons in the trigeminal nucleus.”¹⁰² The NRM is a nucleus of the rostroventromedial medulla, which also contains several other nuclei that are important in modulating pain perception, but the NRM is probably the most important of these nuclei for migraine because of its 5-HT content. It has been viewed as another candidate nucleus likely to relay the descending signals of the intrinsic pain control system and receives afferents from the cortex.¹³⁰ Electrical or chemical stimulation of the NRM influences sensory input at a spinal and trigeminal level and, in particular, selectively inhibits C-fiber input.¹⁰² This stimulation inhibits release of sensory neurotransmitters such as glutamate, CGRP, and SP,¹³¹ actions which can also be produced by serotonin or serotonin agonists.¹³² A similar inhibition of glutamate release in the cortex is mediated by 5-HT_1D receptors.¹³³

Two types of neurons in the NRM respond to noxious input from the periphery: silent ON cells respond with an increase in firing prior to withdrawal from the noxious input; spontaneously active OFF cells respond by cessation of firing; there are also Neutral cells in which the discharge rate is unchanged by noxious stimulation.^134,135 Responses of all three classes of neurons appear to depend on the location of the noxious stimulus site. For instance, neurons which are Neutral to noxious stimulation of the tail may react differentially to stimulation of the dura mater (becoming 43% ON, 14% OFF, 43% Neutral) or the facial skin (becoming 48% ON, 38% OFF, 14% Neutral).¹³⁶ It is reasonable to assume that the activity of OFF cells produces tonic inhibition of peripheral sensation. It is also reasonable to assume that ON neurons produce sensory facilitation, but this assumption has not been tested. ON cells are known to have an excitatory output to spinal and medullary dorsal horn second-order sensory neurons (not necessarily, but predominantly, those with nociceptor-specific input¹¹³) and OFF cells are inhibitory on these same neurons. The 2 effects act synergistically to facilitate withdrawal from noxious stimulation. Both types may relay signals from the PAG.^103,121

The NRM and 5-HT.— Serotonin infusions relieve migraine headache,¹³⁷ an observation that quickened interest in the idea that defects in serotonergic control may be responsible for the initiation and persistence of the pain of migraine. The reported greater pain-sensitivity of migraineurs has been linked to changes in serotonergic neuronal systems.¹³⁸ Serotonin is undoubtedly involved in the elaboration of migraine and the most reliable acute anti-migraine drugs are agonists at 5-HT₁ receptors.¹³⁹ The highest concentration of binding sites for anti-migraine drugs in the brain and the highest concentrations of 5-HT_1B/1D receptors are in the areas associated with processing of trigeminovascular sensation, especially the PAG and NRM.^128,140 The latter was identified as containing neurons, which use 5-HT as a neurotransmitter in the 1960s.^141,142

5-hydroxytryptamine depletion induces migraine symptoms in humans¹⁴³ and migraineurs have been shown to have low serum and platelet 5-HT levels between attacks.^144,145 Blood platelet and cerebrospinal fluid (CSF) levels of 5-HT and its metabolites further decrease during a migraine attack,^144,145 which supports the hypothesis of a decrease in 5-HT neuronal drive. Migraineurs have greater pain sensitivity in general, a symptom that has also been linked to changes in serotonergic neuronal systems originating in the NRM.¹³⁸ 5-HT is regarded as being excitatory in the periphery and it can even induce vascular pain,¹⁴⁶ but in the CNS, it is another matter – 5-HT is always a hyperpolarizing inhibitor of neuronal function and, as already mentioned, can itself relieve headache.

Furthermore, drugs that decrease the discharge rate of NRM neurons such as ipsapirone,¹⁴⁷ methyl chlorophenyl piperazine,¹⁴⁸ and 5-methoxydimethyltryptamine,¹⁴⁹ precipitate migraine-like headaches with a high degree of reliability.^150-152 Conversely, drugs like methiothepin, methysergide, and cyproheptadine, which increase the discharge rate of raphe neurons,¹⁵³ have been used to prevent migraine.

Drugs that are 5-HT_1B/1D receptor agonists, including the ergot alkaloids and the triptans, are the most effective drugs in the acute treatment of migraine.^154,155 These drugs inhibit sensory neurotransmission in experimental animals at the first trigeminovascular sensory synapse when administered systemically or directly to the neurons by iontophoresis.^107,156 The relative efficacy of the triptans and ergot alkaloids correlates better with their 5-HT_1B/1D binding affinities than with their affinities for any other receptor.¹³⁹

NRM-Trigeminal Projections.— Intracranial pain sensitive structures are innervated by fibers of the first and second trigeminal divisions, which synapse with neurons in the trigeminal nucleus caudalis and upper cervical spinal cord. Nearly all these neurons also receive convergent fibers from the face and scalp, usually with wide dynamic range modality of input. The way these neurons process trigeminovascular sensation is quantitatively and qualitatively different from the way they process cutaneous sensation. Differences include the high CGRP content of sensory nerves innervating the dura, the unique ability of nitric oxide (NO) donors to induce pain in only the trigeminovascular system, and the unique susceptibility of trigeminovascular pain to 5-HT_1B/1D receptor agonists. Neurotransmitters that have been implicated in the transmission or modulation of sensory information to second-order neurons include glutamate, NO, CGRP, SP, and 5-HT. While all are probably involved, it is the modulatory roles played by 5-HT itself, 5-HT₁ agonist drugs, and 5-HT₁ receptors, that are of greatest relevance for migraine pathophysiology and therapy.

Serotonergic neurons of the NRM project to the superficial laminae of the spinal cord and trigeminal nucleus caudalis and there make contact with incoming sensory fibers and with trigeminothalamic projection neurons.¹⁵⁷ Work from our laboratories was the first to show that anti-migraine drugs with agonist activity at 5-HT receptors might act at this synapse;¹⁰⁷ our later work demonstrated that this site of action was relatively specific for processing of sensory information from the dura mater and involved the 5-HT_1D receptor subtype.¹⁵⁶ A similar mechanism may occur in the cornea because triptans have been reported to ease pain following keratotomy,¹⁵⁸ an effect that may also be mediated via 5-HT_1D receptors.¹⁵⁹ Serotonin infusions themselves also suppress discharges of these first-order sensory neurons in cats.^160,161 Pronociceptive effects of serotonin antagonists may also be mediated at such a site.¹⁵¹

The 5-HT receptor agonists such as the triptans act predominantly via reduction of the effectiveness of incoming sensory drive.¹⁶² They may do this by reducing the release of sensory neurotransmitters through an action at presynaptic 5-HT_1D receptors, which exist on these fibers.¹⁶³ The 5-HT_1D receptors responsible are relatively selectively confined to dural sensory fibers and are much less common on fibers that innervate the face or non-cranial sensory distribution, although there is one report that 5-HT_1D receptors can be found in dorsal root ganglia as well as the trigeminal ganglia.¹⁵⁹ In contrast, fibers which innervate the face tend to have 5-HT_1B receptors.¹⁵⁶ This would explain the considerable selectivity of modern anti-migraine drugs for headache pain compared with other trigeminal pain¹⁶⁴ and non-trigeminal pain. There is some evidence that 5-HT_1F receptors might be involved in this effect, because a clinical trial of the 5-HT_1F receptor agonist LY334370 showed promising results in migraine.¹⁶⁵ The authors suggested that a likely site of action of LY334370 was at the first sensory synapse but, unfortunately, there are little data from humans on whether 5-HT_1F receptors can be found there. There have been no further reports of clinical trials of 5-HT_1F agonists.

A recent article by Burstein et al¹⁶⁶ has raised again the phenomenon of a triptan anti-migraine drug producing a seemingly anomalous initial exacerbation of migraine in humans and the facilitation of responses in an animal model of migraine. We have seen such enhancement after intravenous administration of both naratriptan¹⁶⁷ and eletriptan (G.A. Lambert, K.L. Hoskin, A.S. Zagami, unpublished observations). These observations are of great interest because they shed some light on the mode of action of triptans, particularly in distinguishing their CNS actions from their peripheral vascular actions. Transient allodynia, paresthesias, and worsening of headache following treatment with triptans have been reported several times since the drugs were first introduced. A common feature of treatment is an initial numbness or tingling, warm sensations, flushing, and paresthesia predominantly in the trigeminal sensory distribution. In some patients there may be a temporary worsening of the headache in association with these symptoms.^168-170 In humans subcutaneous injection of sumatriptan leads to allodynia 20 minutes after injection, but this largely disappears by about 40 minutes.¹⁶⁸ Allodynia and paresthesia following administration of other 5-HT₁ agonists, the ergot alkaloids, was reported as long ago as 1931,¹⁷¹ and there was anecdotal evidence from hundreds of years ago of the occurrence of facial pain associated with outbreaks of ergotism – the “fire” of St Anthony's Fire may well have been allodynia and its initial transient “disagreeable tintillation” of the extremities may well have been paresthesia.¹⁷²

The time course of the responses where a transient facilitation is replaced with a long-lasting suppression suggests a pharmacokinetic mechanism in which the initial facilitation arises from easy early access to one population of receptors (in the periphery perhaps), followed by a later occupation of more difficult to access receptor sites (in the CNS perhaps). In our own experiments, enhancement of responses generally occurred only in the first 5-15 minutes after intravenous administration. Plasma concentrations of triptans peak at 12-13 minutes following subcutaneous administration, with a half-life of several hours.¹⁷³ An alternative explanation for the time course of this dual effect on sensory processing would be a pharmacodynamic one whereby an initial antagonist effect at one population of receptors would be transformed to an agonist effect at the same receptors at higher concentrations. We can find no evidence for this in our own experiments, however, where triptans applied locally by iontophoresis invariably produce a monotonic dose-response curve.¹⁷⁴ Alternatively, the switch from facilitation to inhibition might be related to differential actions of the triptans on “ON” and “OFF” neurons, their dendrites, or their projecting axon terminals.

Locus Coeruleus.— In the context of the descending pain control system, the LC is often seen as a companion or parallel nucleus to the NRM,¹⁰⁴ and its role in control of pain sensation and other functions has been extensively reviewed (see eg, Stamford¹⁷⁵ and Berridge and Waterhouse¹⁷⁶). The LC receives input from, and sends output to, the cortex¹⁷⁷ and a host of other nuclei.^178,179 In the context of the theory presented here, the most important of the afferent inputs appears to be that coming from the PAG; this is unusual in that these projections make contact with the “cloud” of dendrites which surround the LC, rather than penetrate deep into it.¹⁸⁰ Descending projections from the LC innervate other brainstem structures such as the NRM, with which it is reciprocally innervated,¹⁷⁸ and dorsal horn neurons in the spinal cord^175,181,182 and the trigeminal nucleus,^183-187 but it is not known whether those which receive input from the dura mater also receive inhibitory input from the LC. The development of neuropathic pain is influenced by descending projections from the LC, and lesions in the LC exacerbate neuropathic pain states and the excitability of dorsal horn neurons that occurs in such states.¹⁸⁸ It has been hypothesized that adrenergic neuronal mechanisms, driven by the LC, may be important in the therapeutic effects of analgesics and anti-inflammatory drugs.^105,182 Electrical stimulation of the LC produces analgesia in animals.¹⁸⁹ Projections from the LC reach the visual cortex,¹⁹⁰ a connection that may be of relevance to the visual aura of migraine; stimulation of the LC inhibits most visual cortex neurons.¹⁹¹ An unusual projection of the LC is the direct projection to intracranial and extracranial blood vessels;¹⁹² this appears to be involved in the control of these vessels^193,194 and may therefore be of great relevance to the pathophysiology of neurovascular headaches such as migraine.²⁶ LC stimulation produces vasoconstriction in the cortical circulation and it was once thought that this might be of relevance to the vasoconstrictive phase of some migraines and perhaps even trigger CSD.²⁶ The main neurotransmitter in the LC appears to be noradrenaline; this may be of relevance in pharmacotherapy and prevention of migraine, given that several acute treatment preventative drugs have actions at adrenergic receptors and some have actions on the LC itself.^195-199

Despite this weight of circumstantial evidence, there is only limited direct evidence (eg, from imaging) that the LC is involved in the pathogenesis or therapy of migraine; the LC is too small to distinguish easily in such studies.^123,200 Interest in the role of the LC in migraine, and in pain control in general, has consequently waned. In more recent times the function of the LC has been seen as one which involves modulation of behavioral state and cognitive processes;¹⁷⁶ these functions are generally thought to involve the ascending projections of the LC to higher brain centers.

Hypothalamus.— About one-third of migraines have prodromes^201,202 and the headache in patients who have them appears to be more intense and lasts longer.²⁰¹ Prodromal symptoms are often of a nature (mood changes, appetite and thirst changes, disturbances of taste and smell,^201,203 circadian rhythmicity,²⁰⁴ and changes in renal function), such as to suggest the involvement of the hypothalamus. Suggestions of this nature have been made frequently²⁰⁵ (see also the review by Matharu¹⁰⁶). The hypothalamus has a very complex structure and subserves a myriad of regulatory functions;²⁰⁶ those of relevance to migraine include its participation in the pain control system¹⁰² and its role in cardiovascular regulation, especially in vascular tone.²⁰⁷ PET studies have shown an increase in blood flow in the hypothalamus during traumatic nociceptive pain²⁰⁸ and other pain states. Stimulation of the superior sagittal sinus, an intracranial pain-sensitive structure, increases c-fos expression in the posterior hypothalamus of cats.²⁰⁹ Electrical stimulation of the hypothalamus produces analgesia in animals, an effect which is mediated by spinal 5-HT_1A,1B&3 receptors.²¹⁰ This suggests that the stimulus-induced analgesia may relay through the NRM, as it apparently does with the PAG. Evidence for the direct involvement of the hypothalamus in migraine is not as strong as it is for cluster headache and there appears to be only one report of the occurrence of activation of the hypothalamus in spontaneous migraine.²¹¹ Although the authors were unsure whether this activation shown in this PET study was due to neuronal excitatory or inhibitory processes, they were able to show that it persisted after treatment of patients with sumatriptan and they suggested that the activation could not have been merely a reaction to the pain. If the hypothalamus is a generator of migraine, the question arises as to whether the generation takes place via its downstream or upstream projections. Many prodromal symptoms of migraine seem to be cortical perceptual manifestations of drives which ascend from an activated hypothalamus. As such, they may involve cortical excitability changes and represent the conscious manifestations of cortical migraine triggers, similar in many ways to the sensory, CSD, or other triggers. Simply put, the hypothalamus may be just another source of migraine triggers for the cortex.

The case for hypothalamic involvement in cluster headache, SUNCT, and hemicrania continua is much stronger than it is for migraine,^212-218 and hypothalamic stimulation has been trialed as a therapy for cluster headache^219-221 and may possibly function through the orexigenic neuronal system.²²²

PREVENTING MIGRAINE

There are as yet no universally reliable prophylactic drugs for migraine. Drugs to treat acute attacks have already been developed, but they do not prevent attacks although naratriptan has been recommended for this purpose.²²³ Of all of the classes of preventative drugs, few can be said to have been a great deal different from placebo. The antiepileptics and the β-blockers may be the exceptions. A dismayingly large number of these preventative drugs have only been discovered to have migraine prophylactic effects by serendipity²²⁴– usually physicians have noted a decrease in migraine frequency after placing patients on a particular drug for other purposes. In at least one instance, the use of monoamine oxidase (MAO) inhibitors, the use of the drug for even the primary condition (in this case, hypertension) also had little basis in theory – a case of double serendipity. Evidence-based medicine has thus made little headway in this field. The “discovery” of the (rather weak) migraine-preventing effects of magnesium ion was an exception. It arose from a bibliographic technique, in which a search of 2 apparently unrelated fields revealed a “hidden connection” between magnesium deficiency and migraine prevalence.²²⁵

Few preventative drugs seem to work in the same way that the modern acute treatment drugs work, and few of them succeed in preventing migraine with the success rates achieved by the acute treatment drugs. Exactly why, how, and where the present preventative drugs act is unknown. While it is tempting to assume that they prevent migraine by acting on the same neurotransmitter receptors for which they were developed, this is by no means certain. If the hypothesis is correct, each drug might act on different components of the hypothetical trigger system and through different receptors, depending upon which trigger factors produce migraine in each particular patient. In our research work, we are attempting to discover a common location or a single receptor type in the cortex-PAG-NRM-trigeminal pathway to overcome this variability of effectiveness.

QUESTIONS—ANSWERED AND UNANSWERED

This review has addressed only some of the questions that could be posed regarding the genesis of migraine headache. Obvious questions, which so far remain unanswered because of the lack of experimental evidence, include:

• 1
  How are migraineurs different from non-migraineurs?
  Up to one-third of women will develop migraine at some stage of their life;²²⁶ with men, the lifetime prevalence is lower. It may be that the person who never has a single migraine-like headache in their entire life is rare. If this were so, then the factors that predispose humans to migraine must be very subtle and perhaps vary through life in each individual. Genetics is one, at least for some headache forms²²⁷ and, as discussed above, such genetic changes may predispose the cortex to increased excitability. What might be termed “sub-pathological” small differences exist between chronic migraineurs and the normal population. These differences include differences in relevant biochemical markers such as 5-HT and its metabolites,^3-8,228 neurophysiological measures which are seen in tests of contingent negative variation,²²⁹ and a few minor CNS abnormalities.²³⁰ But all of these markers are subtle and it is not even clear whether they are the cause or the consequence of migraine.²³¹
• 2
  In what way does the trigeminovascular system differ from other sensory systems?
  It seems that there is no equivalent to migraine in any other sensory system. We have reviewed briefly the overt differences between migraine pain and other pain, but elucidating a mechanism for this difference has so far escaped us. This hypothesis cannot explain why the dura mater seems to have been singled out for special treatment by a cortex-triggered disinhibition mechanism. It may be of relevance that the sensory fibers of the dura are preferentially enriched in neuropeptides such as CGRP or SP.²³² On the other hand, the selectivity may have a teleological explanation, in which only the most critical structures, such as the brain and its enclosing envelope, have been provided with a hair trigger response mechanism.
  In our laboratories, we have attempted to answer some other questions which could put specific parts of this hypothesis to the test. These questions include:
• 3
  Do the PAG and NRM receive functional input from the cranial vasculature?
  Stimulation of the dura mater and skin activates neurons in the NRM and PAG. Neurons were tested for responsiveness to peripheral stimulation. Most neurons were spontaneously active and discharged in response to electrical stimulation of the dura and of the skin and to mechanical stimulation of the skin.^41,233,234
• 4
  Do the PAG and NRM selectively suppress trigeminovascular neurons?
  Others have shown that electrical and chemical stimulation of the PAG does this.³⁹ In our own experiments,⁴¹ conditioning stimuli applied to the NRM at 10-500 microseconds prior to peripheral stimulation suppressed responses to dural mechanical and electrical stimulation but much less so to cutaneous electrical stimulation.
• 5
  Does NRM-induced suppression involve 5-HT?
  Iontophoretic application of the 5-HT_1B/1D receptor antagonist GR127935 antagonized the ability of NRM stimulation to suppress craniovascular sensation.⁴¹
• 6
  Do migraine triggers inhibit the discharge of neurons in the PAG and NRM?
  We made recordings from spontaneously active neurons in cats, under control conditions and during the potential migraine triggers CSD and light flash. The discharge rate of PAG and NRM neurons decreased significantly following the initiation of CSD and that of NRM neurons also decreased following the initiation of repetitive light flash.⁴¹ Discharge rates of PAG neurons were not altered by light flash.^233,234
• 7
  Does CSD affect the ability of the NRM and PAG to influence craniovascular sensation?
  Four waves of CSD gradually antagonized the ability of electrical stimulation of the NRM to suppress the responses of trigeminovascular second-order neurons to mechanical stimulation of the dura mater but not the suppression of their responses to mechanical stimulation of the skin.⁴¹ Recent unpublished experiments have extended this finding to the PAG.
  Our hypothesis must also be able to explain as many of the generalized features of migraine as possible and answer as many of the unanswered questions as possible. Our research program is continuing and some of these further questions will have to be answered via further experiments. There are some questions, however, which could be explained tentatively by appeal to the underlying hypothesis, as follows.
• 8
  Why is there no pathology in migraine?
  Because migraine is not a pathological condition, aberrant cortical activation is mostly a “noise” phenomenon, but in some cases this noise rises to levels where it can influence the brainstem through descending connections.
• 9
  Why do different migraineurs have different triggers?
  It is our belief that trigger pathways can originate in different areas of the cortex and only converge at the brainstem level. If there were differences between migraineurs in (say) the susceptibility of neurons in different areas of the cortex (genetically determined perhaps), then triggers that had a stronger influence on that particular cortical area would be the critical trigger for that migraineur. This might be true even in migraine with aura, because, while CSD activates neurons in all areas which it invades, only one particular cortical region may be the critical one and it may be different in each patient.
• 10
  Why does headache not always follow aura or triggers?
  The existence of cortical activation does not guarantee brainstem inhibition. Even in our own experimental animals, neither CSD nor light flash invariably inhibited brainstem neurons, nor did they invariably reverse the suppressant effects of brainstem stimulation on sensory processing. One reason for these clinical and experimental observations may be that the brainstem neurons involved are more or less susceptible to depolarization at different times due to biological rhythms or input converging on the nuclei from elsewhere.
• 11
  Is there CSD or cortical activation in migraine without aura or in migraine without an apparent trigger?
  By our hypothesis, there would almost always have to be. In migraine without a visual aura, CSD or non-migrating cortical excitation may nevertheless occur in a “non-eloquent” area of the cortex and so produce no perceptible symptoms and there is at least one report of CSD apparently occurring during migraine without aura.^235,236 By the same token, “imperceptible” triggers may produce neuronal activation or even a “silent aura” that nevertheless activates cortical neurons sufficiently to initiate descending influences on the brainstem, without producing other perceptible neurological symptoms.
• 12
  Can the hypothesis explain the origin of nausea in migraine?
  This may involve the nucleus tractus solitarius (NTS), which has been shown to receive direct projections from the raphe nuclei²³⁷ and the cranial vasculature^109,238 and to be involved in the elaboration of nausea and vomiting in painful states.²³⁹ Our own experiments demonstrated that naratriptan and eletriptan blocked trigeminovascular sensory input to the NTS in a manner analogous to its effects in the trigeminal nucleus.¹⁷⁴ Thus, the development of nausea in migraine could parallel the development of headache and might occur via the same mechanism.
• 13
  Why are successful migraine preventatives so diverse in their pharmacological nature?
  Simply put, because they all act in different locations or they all act on the brainstem, but by different neural mechanisms. If our hypothesis is correct, each existing prophylactic drug might act on different components of our hypothetical trigger system and through different receptors, depending upon which trigger factors produce migraine in each particular patient. Antiepileptics, for instance, may act at the cortical level to “damp down” excessive cortical discharge.²⁴⁰ Many preventative drugs also arrest, inhibit, or raise the threshold for CSD,⁹⁹ but they may of course, act elsewhere also. Preventatives like methysergide may act by increasing the drive of neurons in the NRM. Those drugs that cannot be identified as acting on the cortex, PAG, NRM, or first trigeminovascular synapse, may either be acting at a diverse range of cortical sites with diverse neuropharmacology, or else the older ideas about peripheral effects may, after all, be valid for them. From the literature, we perceive that there are at least 3 putative neurotransmitters in the cortex-PAG neuraxis (glutamate, aspartate, GABA) and 2 extra ones in the cortex/PAG-NRM neuraxis (somatostatin, neurotensin). Receptors for these neurotransmitters in the PAG and NRM would seem to be natural targets for preventative drugs. Drugs acting on these receptors would block influences descending from the cortex. Both nuclei also have neurons with dendrites which contain autoreceptors to their own neurotransmitters;^241-243 antagonists to the autoreceptors in these nuclei may therefore also prove to be sites at which particular preventative drugs might act. Drugs acting on autoreceptors would work by preventing self-inhibition of the neurons.
• 14
  Is there a link between the nature of a patient's trigger and the nature of the prophylactic drug that works best for them?
  This is a very interesting question that perhaps can be answered from the existing literature and case studies. Being able to discern a link between (1) triggers, (2) cortical neurophysiology, and (3) the nature of a particular migraine preventative would greatly strengthen the cortico-brainstem hypothesis (Fig.). What is needed is a thorough meta-analysis of the migraine epidemiological literature, to see if such associations can be uncovered.

Figure 1.—. Hypothesis: Trigger factors cause neurons in the cortex to become overactivated and this leads to migraine headache by convergent pathways relaying via the brainstem that disinhibit incoming sensory information from the dura mater at the level of the trigeminal nucleus. Existing preventative drugs act at the diversity of receptors above the level of the periaqueductal gray matter (PAG) (1, 2, 3, cortical spreading depression [CSD]). The ideal migraine preventative drug would act at the point, or after, the pathways converge – PAG and nucleus raphe magnus (NRM).

Acknowledgments

Acknowledgment: This study was supported by a University of NSW Goldstar Award.