Self-motion perception is sensitized in vestibular migraine: pathophysiologic and clinical implications

Vestibular migraine (VM) is the most common cause of spontaneous vertigo but remains poorly understood. We investigated the hypothesis that central vestibular pathways are sensitized in VM by measuring self-motion perceptual thresholds in patients and control subjects and by characterizing the vestibulo-ocular reflex (VOR) and vestibular and headache symptom severity. VM patients were abnormally sensitive to roll tilt, which co-modulates semicircular canal and otolith organ activity, but not to motions that activate the canals or otolith organs in isolation, implying sensitization of canal-otolith integration. When tilt thresholds were considered together with vestibular symptom severity or VOR dynamics, VM patients segregated into two clusters. Thresholds in one cluster correlated positively with symptoms and with the VOR time constant; thresholds in the second cluster were uniformly low and independent of symptoms and the time constant. The VM threshold abnormality showed a frequency-dependence that paralleled the brain stem velocity storage mechanism. These results support a pathogenic model where vestibular symptoms emanate from the vestibular nuclei, which are sensitized by migraine-related brainstem regions and simultaneously suppressed by inhibitory feedback from the cerebellar nodulus and uvula, the site of canal-otolith integration. This conceptual framework elucidates VM pathophysiology and could potentially facilitate its diagnosis and treatment.

Basic characteristics of the VM and control groups. Symptom severity and demographics of the VM and control groups are summarized in Table 1. The principal observations are that VM patients had significantly greater dizziness as assessed with the Dizziness Handicap Inventory (DHI 22 ) than the migraine or normal groups, and motion sickness susceptibility (MSSQ 23 ) was significantly increased in VM patients compared to migraine or normal controls. Headache severity (assessed with the Headache Impact Test 24 ) was more pronounced in the migraine and VM groups than the normal subjects, as expected, with no significant difference between the migraine and VM groups. Headache severity was uncorrelated with vestibular symptoms (DHI, MSSQ) in normal and migraine subjects, but it did correlate with dizziness (DHI) in VM patients (Pearson R test p = 0.02).
Anxiety assessed with the Beck Anxiety inventory 25 did not differ significantly between the VM, migraine, and normal control groups. The VM, migraine, and normal groups did not differ in age, and the migraine and VM groups were gender-matched while the normal group was evenly divided between genders.
Roll tilt perceptual thresholds are reduced in VM at mid and low frequencies. The geometric means of the roll tilt perceptual thresholds for the four subject groups, recorded using standard motion stimuli 26 and psychometric methods 27 , are shown in Fig. 1. All groups showed a similar frequency-dependence whereby displacement thresholds (the smallest position change that was accurately perceived as rightward or leftward) were smallest at the higher frequencies and larger at the lower frequencies -consistent with earlier normative data 28 . Roll tilt thresholds depended significantly on both the subject group (VM, migraine, normal, ANOVA p < 0.001) and on the motion frequency (ANOVA p < 0.001). Furthermore, in the mid-to-low frequency range (0.03, 0.05, 0.1, and 0.2 Hz) VM thresholds were significantly lower than normal subjects (t-test: p = 0.001) and migraine patients (t-test: p = 0.001), while migraine and normal subjects did not differ (t-test: p = 0.81). Over the higher frequency range (1.0 Hz and above) thresholds in the three groups converged and did not differ significantly (ANOVA p = 0.1; t-tests between groups all > 0.05). Migraine patients had thresholds that were equivalent to normal controls except for 0.1 and 0.5 Hz, where their thresholds fell between the normal and VM groups but did not differ significantly from normal (t-tests p = 0.1 for 0.5 Hz, p = 0.4 for 0.1 Hz). Patients with Meniere's disease, who had recurrent vertigo not due to migraine, had normal thresholds at 0.2 Hz, a frequency where VM thresholds were significantly lower than the control groups. The conclusion from these data is that roll tilt thresholds are abnormally small in VM patients over the mid to low-frequency range, but this was not due to migraine (without episodic vertigo) or episodic vertigo (without migraine). activity in the vertical canals, the graviceptive organs (primarily the utricles for the small tilt angles we employed), and the tactile cues applied to the head and trunk. We therefore tested each of these three sensory components in isolation to determine if they were responsible for the threshold reduction in VM during roll tilt. Figure 2 shows perceptual thresholds for roll rotation while supine (vertical canals), inter-aural (IA) translation (utricles), and tactile thresholds over 3 dermatomes on each body side, and in all cases thresholds in VM and control subjects were indistinguishable (roll rotation: 2-way ANOVA, p = 0.79 for group, 0.1 for frequency, and 0.9 for interaction; IA translation: 2-way ANOVA, p = 0.51 for group, p < 0.001 for frequency, and p = 0.97 for interaction; tactile Mann-Whitney p = 0.34). In particular, below 0.5 Hz roll tilt thresholds were abnormally low in VM but roll rotation and IA translation thresholds were normal. Together these results indicate that the roll tilt threshold reduction in VM is not due to sensitization of canal or otolith signals in isolation, nor does it reflect increased tactile sensitivity. Since prior work in normal subjects (e.g. 28 ) demonstrated that roll tilt thresholds at mid and low frequencies require central integration of canal and otolith cues, abnormal roll tilt thresholds in VM appear to result from changes in the synthesis of canal and otolith inputs in the brain.
Roll tilt thresholds in VM are related to vestibular symptoms. The relationship between vestibular symptom severity (DHI score) and roll tilt thresholds at 0.05 Hz is shown in Fig. 3A for the expanded VM population tested at this frequency. The data appear to be distributed in a pattern that approximates a tilted 'V' with the two limbs converging at the lower left corner of the plot. To facilitate analysis of the data we considered the two limbs separately, dividing patients into the VM1 (grey icons) and VM2 (black icons) clusters ( Fig. 3A) with an intuitive method based on visual inspection of the plot, and a soft clustering method which required no assumptions about the data's distribution except that it was divisible into two subsets: (i) for the visual approach, we calculated the (black dashed) line that connected the average (DHI, threshold) values for all VM patients with the average values for the three data points located in the lower left corner of the plot where the two limbs of the 'V' converge. This line was used to separate the VM patients into VM1 and VM2 clusters and the VM1 and VM2 data were each fit with a linear regression (grey and black lines, respectively) and with 95% confidence ellipses (Fig. 3A); and (ii) to verify this visual classification, we also segregated the VM data using soft clustering, in which the probability of the patient belonging to each cluster is calculated 29 , with points at the edge of a cluster having lower probabilities of membership to that cluster 30,31 . As detailed in the methods, the results of the visual and soft clustering approaches were very similar-ROC AUC analysis comparing these two classification methods yielded a value of 0.93. The principal difference between VM1 and VM2 was the relationship between vestibular symptom severity (DHI) and roll tilt threshold. Using the visual classification as our standard approach, VM1 showed a positive correlation between tilt threshold and symptom severity (Pearson R for DHI and tilt threshold, R = 0.94, p < 0.001) and therefore a negative correlation between tilt sensitivity and dizziness. In contrast, VM2 patients had DHI values that varied independently of their tilt thresholds. Furthermore, tilt thresholds in VM2 were uniformly low (less than normal) but thresholds in VM1 were more widely distributed and included values above, within, and below the normal range. Comparison of other VM1 and VM2 characteristics (Table 2) indicates that they were equivalent except for the roll tilt threshold, which was lower in VM2, and the motion sickness susceptibility score, which was higher in VM1. Otherwise, VM1 and VM2 s shared equivalent dizziness and headache symptom severity and equivalent ages and gender distributions.
Roll tilt thresholds in VM are related to VoR dynamics. Velocity storage is the brainstem mechanism that mediates the dynamic and spatial characteristics of vestibular behaviors 32,33 and in particular it determines the time constant (Tc) of the VOR. Similar to the DHI-threshold data shown in Fig. 3A, the relationship between For all data points, n = 12 for VM, migraine, and normal groups (except for 0.03 Hz where the n is 4 for these three groups) and n = 8 for Meniere's patients. Some error bars are smaller than the associated icons and therefore are not visible. Less data is available for 0.03 Hz tilt because most subjects lacked the attention to perform this task adequately (as each trial lasted 33.3 s).
the VOR Tc and roll tilt thresholds (Fig. 3B) also approximated a tilted 'V' with the two limbs converging at the lower left corner of the plot. We employed the same two approaches (visual and soft clustering), therefore, to separate the two limbs of the Tc-threshold distribution into VM1 (grey icons) and VM2 (black icons) clusters. Again, as detailed in the methods the results of the visual and soft clustering approaches were very similar with ROC value of 0.99, and all VM patients with DHI and VOR Tc data available (filled icons in Fig. 3A,B) independently received the same label (VM1 or VM2) for both the DHI-threshold and Tc-threshold analyses. Paralleling the DHI-threshold data described above, the principal difference between VM1 and VM2 for the Tc-threshold data was the relationship between the VOR time constant and roll tilt threshold, asVM1 patients showed a positive correlation between the VOR time constant and tilt threshold (  (Table 3) were otherwise comparable except for the roll tilt threshold, which was lower in VM2 (note that some patients had only DHI or Tc data available, open icons in Fig. 3, so values in Tables 2 and 3 differ).  72 at three locations (zygomatic arch, scapula, and iliac tuberosity, with the left and right-sided measurements averaged for each subject and location). Tactile thresholds were measured only in the VM and migraine groups since tactile sensitization related to allodynia was a potential concern in the two migraine populations but not in the normal control subjects.

Discussion
Our primary findings are that roll tilt perceptual thresholds in VM: i) are abnormally low for mid and low frequency motion stimuli; ii) are not due to migraine (when it is unassociated with vestibular symptoms), or due to vestibular symptoms (when they are not due to migraine); iii) reflect abnormal integration of canal and otolith information in the brain; iv) have a 'V' shaped distribution with respect to the severity of vestibular symptoms, with data clustered about one leg displaying a positive correlation between tilt threshold and symptom severity and data clustered about the other leg displaying threshold-independent symptoms but uniformly low tilt thresholds; and v) have a very similar 'V' shaped distribution with respect to velocity storage (VOR time constant). These results are discussed below in terms of their potential diagnostic, pathophysiologic, and therapeutic implications.
Diagnostic implications. VM is the most common cause of episodic vestibular symptoms 6 but lacks any pathognomonic finding and is diagnosed using criteria determined by an expert panel 7,8 . Identifying an approach that segregates VM patients from vestibular patients whose symptoms are not due to migraine would therefore be of great clinical value. Our results show that for populations of patients, VM differs from vestibular (without migraine) patients, migraine (without vestibular) patients, and normal subjects with respect to mid and low frequency roll tilt perceptual thresholds. The mechanisms responsible for these findings are considered below, but from a clinical perspective these results could provide a biomarker that allows individual patients to be classified as VM or non-VM based on their roll tilt threshold when considered in tandem with their DHI and/or VOR time constant. To determine the sensitivity and specificity of this approach, substantially larger patient populations must be studied to more accurately define the VM confidence ellipses and to determine the data structure for vestibular patients whose symptoms are not due to migraine. Ultimately a multivariate analysis that includes the low frequency roll tilt thresholds, DHI, and VOR time constant, and other perceptual (e.g., z-axis translation thresholds 34 ) or VOR (e.g., response variability) measures may be a way to most accurately separate VM and non-VM vestibular patients.
Such a multivariate analysis may be particularly helpful because there are a plethora of possible interactions between migraine and the vestibular system 9 , so a heterogeneous set of mechanisms almost certainly contributes to the vestibular symptoms of VM. For example, we followed the current VM classification criteria 7,8 in this study, so patients with evidence of a peripheral vestibular damage were excluded. Migraine is epidemiologically linked to Meniere's disease 35 , however, some migraine patients develop evidence of labyrinthine damage 36 , and animal models of migraine show evidence of biochemical changes in the labyrinth 37 , implying that vestibular dysfunction in VM may have a peripheral component in some patients 38 . While the current VM diagnostic criteria allowed us to focus on central mechanisms without confounding peripheral effects that could potentially mask central  pathophysiologic implications. To facilitate interpretation of our results, we first summarize the relevant concepts of central vestibular processing (reviewed in 33 ); then consider the implications of our findings when examined within this conceptual framework; and finally provide a hypothesis for mechanisms that could contribute to the vestibular symptoms of VM that is consistent with our results and with known vestibular physiology and migraine pathophysiology.
Overview of relevant central vestibular processing. As shown schematically in Fig. 4, the labyrinth transduces high-pass filtered angular velocity in three dimensions (w, canals) and the vector sum of gravity and linear acceleration (GIA, otoliths). The brain estimates angular velocity (w') by combining the direct signal from the canal afferents (g D pathway) with a temporal integration of the canal afferent signal (via velocity storage, with a gain g VS and a time constant T VS ). Velocity storage (light grey box, Fig. 4) is generated in the commissural fibers between the medial and superior vestibular nuclei and hence can be considered a marker of vestibular nuclear activity 39,40 . It is responsible for the dynamics of VOR and perceptual responses induced by head rotation (primarily in yaw) 41,42 . The nodulus and uvula of the cerebellum (NU) are the location of the 'tilt estimator" (dark grey box, Fig. 4) where canal and otolith inputs are first synthesized 43 . Specifically, the NU integrates the off-vertical component of the angular velocity signal to update the estimated orientation of gravity G' 44,45 , and G' is subtracted from the GIA sensed by the otoliths to estimate A' (linear acceleration) 46,47 . The NU tilt estimator has GABAergic inhibitory projection to the vestibular nuclei 48 (Fig. 4, grey arrow #1), including the velocity storage network, as NU ablation lengthens 49 and stimulation shortens 50 the VOR Tc.
Evidence that reduced VM tilt thresholds reflect sensitization of the cerebellar NU. Three types of evidence suggest that the NU is sensitized in VM patients -i) VM thresholds were reduced when canal and otolith cues co-modulated (e.g., during tilt, Fig. 1) but not when either the canals or otoliths modulated in isolation (Fig. 2). Since the NU is the location of canal-otolith integration, the reduction of tilt but not rotational or translational thresholds in VM suggests that central sensitization localizes at least in part to the NU; ii) if the NU is sensitized in VM, a positive correlation is predicted between tilt thresholds and the VOR time constant because of the inhibitory projection (Fig. 4, grey arrow #1) from the NU (where tilt is calculated) to velocity storage in the brainstem (which sets the VOR Tc). Specifically, NU sensitization (reflected by lowered tilt thresholds) should result in increased inhibition of velocity storage (reflected by lowered Tc). Indeed, one cluster of VM patients (Fig. 3B, VM1 = grey icons and line) showed this predicted correlation, as did the non-VM subjects (data not shown). In contrast, the second VM cluster (Fig. 3B, VM2 = black icons and line) had uniformly low tilt thresholds that were independent of the VOR Tc (discussed below); and iii) since the NU projects to and receives reciprocal projections from vestibular nuclei 48,51 where velocity storage resides (Fig. 4), if reduced thresholds indicate sensitization of the NU then the frequency-dependence of the threshold abnormality should mirror that of velocity storage 52 . Figure 5 compares the frequency-dependence of the VM threshold abnormality and velocity storage and demonstrates their close similarity (this figure includes "pseudo-static" DC 28 results from 13 ), supporting our contention that tilt threshold Evidence that the vestibular nuclei contribute to the symptoms of VM and are modulated by (inhibitory) cerebellar and (excitatory) brainstem projections. The relationship between tilt thresholds and the VOR Tc in VM patients (Fig. 3B) is structurally the same as the relationship between tilt thresholds and vestibular symptom severity (DHI, Fig. 3A). Both the threshold-Tc and threshold-DHI plots have two clusters of data, one grouped about a positive regression (grey icons and lines) and one grouped about a line that is nearly parallel to the y-axis with uniformly low thresholds (black icons and lines). Since the VOR Tc is a well characterized measure of vestibular nuclear activity (velocity storage in particular), one logical explanation for the similarity in plots 3 A and 3B is that vestibular symptoms in VM are generated, at least partially, within the brainstem vestibular nuclei in a manner that correlates with the processing of the VOR Tc. Like the threshold-Tc data (Fig. 3B), the positive correlation between tilt thresholds and vestibular symptoms in Fig. 3A (grey regression line) likely reflects the inhibitory projections 48 from the cerebellar NU to the vestibular nuclei (Fig. 4, grey arrow #1). Further, the vestibular nuclei receive multiple projections from brainstem regions associated with migraine initiation and propagation, including the trigeminal nuclei, locus coeruleus, and dorsal raphe nucleus [53][54][55] . We propose, therefore that both the vestibular symptoms of VM and the VOR Tc can also be modified independent of NU activity/tilt thresholds (black regression lines in Fig. 3A,B) because of the effects of excitatory projections from these migraine-related brainstem regions to the vestibular nuclei (Fig. 4, black arrow #2).
The VM patients have a 'V' like distribution when tilt thresholds are plotted against dizziness severity or the VOR Tc, with the VM1 and VM2 legs merging in the low threshold-low DHI/Tc corner of the plots (Fig. 3). This pattern suggests that VM1 and VM2 could be distinct but overlapping phenotypes or that they could represent two components of a single VM phenotype. Our pathogenic model, discussed above, supports the single phenotype explanation with the VM1 and VM2 clusters representing a continuum of patient states that are determined by the relative contributions of inhibitory (cerebellar) and excitatory (brainstem) projections to the vestibular nuclei. Since patients with less severe vestibular symptoms and/or short VOR Tc always have low tilt thresholds, feedback inhibition supplied by the cerebellar NU to the vestibular nuclei (Fig. 4, grey arrow #1) may be www.nature.com/scientificreports www.nature.com/scientificreports/ sensitized as an early response to the aberrant vestibular nuclei activation produced by migraine (as evidenced by the positive correlation in VM1 between DHI/Tc and tilt thresholds, Fig. 3, grey lines). This inhibitory feedback may be adequate in some patients, namely those with low thresholds but also low DHI or VOR Tc values (bottom left corner of Fig. 3A,B). Cerebellar-mediated inhibition presumably saturates when the NU has reached maximum sensitization, associated with the lowest possible roll tilt thresholds, and then additional stimulation of the vestibular nuclei by brainstem migraine regions (Fig. 4, black arrow #2) increases the DHI and VOR Tc without influencing the already low threshold (VM2 in Fig. 3A,B, black lines).
Two other aspects of the data distributions shown in Fig. 3 warrant further discussion. Firstly, it is evident that, unlike most of the VM patients who have low tilt thresholds, some VM1 patients have tilt thresholds that are in the high-normal range. The exact explanation for this result is not known but presumably the mechanisms underlying vestibular symptoms in migraine are heterogeneous and can result in a distribution of tilt thresholds that are independent of the model described above. We suggest, therefore, that high-normal tilt thresholds in some VM patients reflect both the inherent variability associated with this disorder and suboptimal engagement of the inhibitory cerebellar feedback loop through the NU that we propose is responsible for threshold reduction. Secondly, our results are qualitatively consistent with prior work that suggested the severity of vestibular symptoms in VM (e.g., motion sickness) correlate with the length of the VOR time constant 12 . In particular, comparison of the VM1 clusters in Fig. 3A,B indicates that higher tilt thresholds are associated with both longer time constants and higher DHI scores, implying that longer time constants correlate with greater vestibular symptomatology. While this correlation was not statistically significant in our relatively small VM data set (as it was in the much larger patient population that was previously described 12 ), the general pattern is recapitulated by our results.
This model provides a potential explanation for the relationship between vestibular symptoms and migraine. The vestibular nuclei are a node in a negative feedback loop (Fig. 4), as they project to (and could therefore sensitize) the NU, which then suppresses vestibular nuclear activity through reciprocal inhibitory projections. This feedback loop contributes to the brain's ability to generate accurate motion and orientation estimates and may play a fundamental role in the generation of motion sickness 56 . Persistent dysfunction of this neural circuit in VM could therefore contribute to the characteristic increase in inter-ictal motion sickness susceptibility 3,12 . This neural circuit can oscillate when the NU is damaged and it has been suggested that it is inherently unstable and is stabilized by NU inhibition 50 . In VM patients, a migraine episode could sensitize the vestibular nuclei to the extent that NU suppression is no longer adequate to maintain stability, particularly when NU activity is modulated by changing head orientation, resulting in ictal episodes of vestibular dysfunction that are temporally linked to migraines and induced by head tilts (e.g., are "positional"), both of which are common in VM 4,38 .
Correlations with imaging and other behavioral studies. Our model is supported by the one ictal functional imaging study of VM patients 57 which showed prominent activation of the cerebellum and brainstem during vestibular episodes. Interestingly, the primary anatomic changes in VM appear to be supratentorial, and these changes in the thalamus 58 and cerebral cortex 59 may be due to primary migraine-related dysfunction in these brain regions or could in part reflect their role as targets for infratentorial projections. More generally, symptoms of VM are almost certainly heterogeneous in origin and most likely are generated by mechanisms that localize to both infra-and supratentorial 60 brain structures and possibly to the vestibular labyrinth as well. Only infratentorial mechanisms were investigated in our study so our findings do not preclude significant contributions from other brain regions or from the inner ear. therapeutic implications. There are no placebo-controlled prospective clinical trials that assess the efficacy of different pharmacologic treatments for VM 61 and there is no rationale to guide the choice of medication for individual patients. Clearly, placebo-controlled prospective clinical trials are necessary to develop evidence-based guidelines for VM medical therapy, and we suggest that our results (and perceptual threshold testing prior to therapy) could help provide a framework for future clinical trials. Specifically, vestibular symptoms in VM patients with relatively high thresholds (e.g., falling near the VM1 grey cluster on Fig. 3A) may respond best to drugs that activate GABA receptors, since these would increase inhibition of the brainstem vestibular nuclei and potentially suppress ictal and inter-ictal vestibular symptoms. Potential drugs would include those that increase synaptic GABA such as tiagabine 62 or the aminopyridines that augment GABA release by cerebellar Purkinje cells 63 . This approach has not been tried but positive results from a trial of acetazolamide 64 , which augments GABAergic neuronal activity as one of its properties, offer support for the contention that increasing suppression of brainstem vestibular nuclei may benefit VM patients. Conversely, patients with low tilt thresholds (falling near the VM2 black cluster on Fig. 3A) may have saturated GABAergic inhibition of the vestibular nuclei and may therefore respond best to medications that affect other neurotransmitters such as beta blockers 61 .
While their effects on vestibular symptoms due to migraine have not yet been studied, drugs that inhibit the action of calcitonin gene-related peptide (CGRP) in the trigeminal ganglion such as erenumab (reviewed in 65 ) could be important in preventing or aborting vertigo episodes in patients with VM. In particular, since migraine-related sensitization of the trigeminal nuclei appears to affect the sensitivity of structures that receive trigeminal nuclear projections (e.g. 66 , for the visual system), it is possible that the well-characterized trigeminal nuclei → vestibular nuclei projections in the brainstem 55 could contribute to vestibular sensitization in migraine patients who develop VM. Attenuating the effects of this pathway with drugs the inhibit CGPR activity in the trigeminal ganglion could potentially minimize trigeminal-mediated sensitization of the vestibular nuclei and thereby stabilize the central vestibular feedback loop (Fig. 4).

Materials and Methods
All subjects provided informed consent in accordance with the Declaration of Helsinki, and all studies were approved by the Massachusetts Eye and Ear Infirmary (MEEI) IRB.
Experimental design. The objective of this study was to measure self-motion perceptual thresholds and other information about vestibular function and symptoms in patients with VM and in the appropriate control groups, with the goal of elucidating the pathophysiology of VM. Blinding was not employed for the vestibular threshold tests since the threshold values were calculated using an automated "expert advisor" that chose the motion parameters and converged on the threshold in a manner that was independent of the experimenter 67 . For tactile threshold tests, the experimenter was blinded to the subject category. Test order was randomized as described below. A priori power calculations were performed using data from our prior preliminary study 13 , with an α = 0.05 and a power of 0.8. The thresholds during roll tilt at 0.01 Hz were 0.55 for VM and 1.4 for migraine, with standard deviations of 0.8 and 0.7, so 8 subjects in each group were calculated as the minimum to achieve the desired power.
Subject selection/characterization. (a) VM patients were culled from the otoneurology clinics at MEEI.
They met the currently accepted diagnostic criteria for definite VM 8 which includes a history of episodic vestibular symptoms that are temporally associated with headaches that meet the International Headache Society (IHS) criteria for migraine 20 ; and an absence of any other neurologic or otologic dysfunction responsible for the vestibular symptoms, as assessed with history, physical examination, audiogram, MRI of the brain, and standard clinical rotational testing 68 . 12 VM patients were tested on the full range of perceptual tests, while an additional 17 VM patients were tested only on roll tilt at 0.05 Hz, since that motion profile was identified in the initial cohort of 12 subjects as one that showed a large difference between VM and control subjects and was feasible to test in this larger population (details below); (b) Migraine (n = 12) patients met the IHS criteria for migraine with or without aura, lacked a history of any vestibular symptoms (other than motion sickness), lacked evidence of any other neurologic or otologic dysfunction as described above, and had normal brain MRI and rotational testing; (c) Normal (n = 12) subjects similarly had no evidence of neurologic or otologic disease, no history of migraine or vestibular symptoms, and had normal rotational testing; and d) Meniere's Disease (n = 8) patients who met diagnostic criteria for definite Meniere's Disease 21 , had episodic vertigo but no migraine history, and had no evidence of permanent peripheral vestibular damage as assessed with rotational testing and cervical vestibular evoked myogenic potentials. other patient features. No migraine or VM patient was on prophylactic migraine medication and all tests were performed during inter-ictal periods and followed the most recent migraine or vestibular episode by at least two weeks. The VM, migraine, and normal groups were age-matched and the VM and migraine groups were sex-matched. The normal control group did not share the female preponderance of the two migraine groups but a large recent study of 105 healthy asymptomatic subjects found no sex-related differences in self-motion perceptual thresholds for any of the five motion stimuli tested 69 , including roll tilt thresholds at both 1 Hz and 0.2 Hz. non-threshold data. All subjects filled out standard questionnaires to quantify the severity of their dizziness (Dizziness Handicap Inventory, DHI 22 ), motion sickness susceptibility (MSSQ using the Golding revised questionnaire 23 ), headache severity (Headache Impact Test, HIT 24 ) and anxiety (Beck Anxiety Inventory 25 ). Yaw-axis VOR responses were characterized with standard low-frequency (0.01 to 1.0 Hz) sinusoidal testing which yielded phase, gain, and bias values for each frequency, and these were used to calculate the overall VOR time, gain, and bias constants 68,70 for each subject. threshold testing. Three separate test sessions were used to obtain the complete perceptual threshold data set for each subject. Each session lasted a maximum of three hours with breaks provided at a regular schedule to minimize fatigue. The testing order was randomized to eliminate possible order effects and each of the three test sessions were separated by a minimum of one week.
Perceptual task. Self-motion perceptual thresholds were determined using a standard forced-choice, one-interval, direction-discrimination method as used previously in our and other labs for numerous vestibular threshold studies (e.g. 26 ). Briefly, after each motion the subjected responded by pressing one of two buttons to indicate if they perceived the motion towards their right or left, and if they were unsure they were required to guess. An adaptive three-down, one-up protocol was used (magnitude of motion reduced after three correct responses and increased after one incorrect response), the direction of motion was randomized, and the test terminated when the coefficient of variation of the fitted threshold parameter reached a value of less than 0.2. A psychometric function was fit to these binary response data using published methods 71 and the width (sigma) of fitted psychometric function was defined as the threshold 27 which is the stimulus magnitude at which 84% of responses are predicted to be correct.
Motion profiles. With the exception of "pseudo-static" roll tilt (see below), all motions were based on a single-cycle sinusoidal acceleration yielding a unidirectional bell-shaped velocity profile 26 . The frequency of the motion was defined as the inverse of the period and ranged from 0.03 Hz (period of 33.3 sec) to 5.0 Hz (period of 0.2 sec). It was not feasible to test subjects at frequencies below 0.03 Hz because the long duration of each trial impaired attention and precluded acquisition of an adequate data set. Each type of motion (roll tilt while upright about a naso-occipital, earth-horizontal axis centered between the ears; roll rotation while supine about an earth-vertical, naso-occipital axis centered between the ears), and inter-aural (IA) translation while upright along an earth-horizontal axis) was tested in separate blocks. Static (DC) thresholds (in Fig. 5) were provided for  (Fig. 3A). The x axis shows the probabilities calculated using the intuitive, visual approach based on a linear regression separating the data into two groups (dashed black line in Fig. 3A) and the y axis shows the probabilities calculated with the fuzzy clustering method. (B) Membership probabilities of belonging to the black cluster for all subjects in the VOR time constant vs. roll tilt threshold plot (Fig. 3B). Perfect agreement between the classification probabilities produced by the two methods would result in data points falling along the y = x (dashed) line.