Periaqueductal Gray Matter Dysfunction in Migraine: Cause or the Burden of Illness?


Address all correspondence to Dr. K.M.A. Welch, University of Kansas Medical Center, Room 6001 Wescoe, 3901 Rainbow Boulevard, Kansas City, KS 66160-7300.


Objective.—The periaqueductal gray matter (PAG) is at the center of a powerful descending antinociceptive neuronal network. We studied iron homeostasis in the PAG as an indicator of function in patients with episodic migraine (EM) between attacks and patients with chronic daily headache (CDH) during headache. High-resolution magnetic resonance techniques were used to map the transverse relaxation rates R2, R2*, and R2 in the PAG, red nucleus (RN), and substantia nigra (SN). R2 is a measure of non-heme iron in tissues.

Methods.—Seventeen patients diagnosed with EM with and without aura, 17 patients diagnosed with CDH and medication overuse, and 17 normal adults (N) were imaged with a 3.0-tesla magnetic resonance imaging system. For each subject, mean values of the relaxation rates, R2 (1/T2), R2* (1/T2*), and R2 (R2*  R2) were obtained for the PAG, RN, and SN. R2, R2*, and R2 values of the EM, CDH, and N groups were compared using analysis of variance, Student t test, and correlation analysis.

Results.—In the PAG, there was a significant increase in mean R2 and R2* values in both the EM and CDH groups (P<.05) compared with the N group, but no significant difference in these values was demonstrated between the EM and CDH groups, or between those with migraine with or without aura in the EM group. Positive correlations were found for duration of illness with R2 in the EM and CDH groups. A decrease in mean R2 and R2* values also was observed in the RN and SN of the CDH group compared with the N and EM groups (P<.05), explained best by flow activation due to head pain.

Conclusions.—Iron homeostasis in the PAG was selectively, persistently, and progressively impaired in the EM and CDH groups, possibly caused by repeated migraine attacks. These results support and emphasize the role of the PAG as a possible “generator” of migraine attacks, potentially by dysfunctional control of the trigeminovascular nociceptive system.


periaqueductal gray matter


migraine without aura


dorsal raphe nucleus


locus ceruleus

R2, R2*, R2′

transverse relaxation rates


red nucleus


substantia nigra


episodic migraine


chronic daily headache


migraine with aura


healthy normal adults


gradient-echo sampling of free induction delay and echo




Iterative Self-organizing Data Analysis Technique

The midbrain periaqueductal gray matter (PAG) is an anatomically heterogeneous, functionally diverse region of densely layered neurons surrounding the aqueduct of Sylvius.1 Receiving input from the frontal cortex and hypothalamus and projecting to the rostral ventromedial medulla and from there to the medullary and spinal dorsal horn, the PAG is the center of a powerful descending antinociceptive neuronal network.2 Further, the PAG can be considered a major nodal point in the central nervous system (CNS), regulating autonomic adjustments to antinociceptive, autonomic, and behavioral responses to threat.3 Raskin and colleagues first recognized the potential for PAG dysfunction in migraine, observing that patients with PAG electrode implantation for chronic pain sometimes developed headache resembling migraine.4 A discrete sclerotic lesion in the PAG was reported to cause severe headache in a single case report.5 In positron emission tomography (PET) studies during spontaneous migraine without aura (MWOA), increased blood flow was measured in mesencephalic regions that possibly reflected PAG, dorsal raphe nuclei (DRN), and locus ceruleus (LC) activation, raising the possibility of a brain stem “generator” of migraine attacks.6

These observations prompted study of iron homeostasis in the PAG of patients with migraine; elevation or decline in tissue iron is associated with altered cellular function.7 High-resolution magnetic resonance (MR) techniques were used to map the transverse relaxation rates R2 (1/T2), R2* (1/T2*), and R2′ (R2* − R2) in the brain, in particular, in the PAG, red nucleus (RN), and substantia nigra (SN). These measures are sensitive to shifts in the paramagnetic properties of free iron in brain tissue and blood.8 We report increased tissue iron levels in the PAG of patients with episodic migraine (EM) with and without aura and in patients with chronic daily headache (CDH) that further increase with duration of the disorder. We discuss these findings as a possible cause of migraine and the burden of illness.



International Headache Society (IHS) classification guidelines were used to diagnose patients with episodic MWOA, patients with migraine with aura (MWA), and patients with near daily headache (more than 15 headaches per month) due to medication overuse who had previously suffered episodic MWOA.9 Although not an IHS classification, we used the term chronic daily headache to categorize these patients. Three groups of subjects were imaged. The institutional review board of Henry Ford Health Sciences Center approved the imaging protocol. All subjects who participated in the study provided written informed consent prior to the study. The first group who were imaged consisted of 17 patients with EM, 9 with MWOA and 8 with MWA; 2 men and 15 women, aged 23 to 52 years (mean ± SD, 33 ± 10 years). No migraine attack occurred in any of the patients with EM and no medication was taken, either abortive or preventative, within 1 week of study. The second group who were imaged consisted of 17 patients with CDH preceded by MWOA; 3 men and 14 women, aged 21 to 57 years (mean ± SD, 36 ± 10 years). This group was studied prior to any medication adjustment, and patients were taking pain medications that included codeine and butalbital, as well as miscellaneous migraine preventative drugs. All patients were experiencing headache at the time of imaging. The third group studied consisted of 17 healthy normal adults (N), 7 men and 10 women, aged 20 to 64 years (mean ± SD, 38 ± 12 years). The difference in the mean age of the three groups was not statistically significant (P = .89, repeated-measures analysis of variance [ANOVA], see Table 1). The duration of illness was recorded for the EM group; the duration of CDH, not including previous years of EM, was also recorded.

Table 1.—.  Subject Characteristics*
CharacteristicControl Subjects
(n = 17)
Episodic Migraine Group
(n = 17)
Chronic Daily
(n = 17)
  • *

    Values are mean ± SD unless otherwise indicated.

  • Significantly increased (P<.05) compared with N.

  • Significantly decreased (P<.05) compared with N.

  • The difference in the mean age of the three groups studied was not statistically significant (P = .34, ANOVA). Differences in the relaxation rates amongst the three groups were tested using one-way ANOVA, with the significance level set at P<.05. There was a significant increase in R2′ and R2* values in the PAG matter of patients with episodic migraine (EM) or chronic daily headache (CDH), compared with the control group (N). There was a significant decrease in the R2′ and R2* values of the RN and SN of the CDH group, compared with the N and EM groups.

Ratio of women to men10:715:214:3
Age, y38 ± 1241 ± 1036 ± 10
Periaqueductal gray matter (PAG)   
R2′4.33 ± 0.976.11 ± 0.886.36 ± 1.29
R2*21.1 ± 1.2523.1 ± 1.5323.6 ± 1.90
R217.0 ± 1.1117.1 ± 1.1817.2 ± 1.65
R2′14.3 ± 2.1813.8 ± 1.8610.7 ± 2.33
R2*39.1 ± 2.8539.1 ± 3.0734.2 ± 3.02
R224.1 ± 1.6225.2 ± 1.9823.6 ± 1.45
Substantia nigra  (SN)   
R2′15.1 ± 2.2015.0 ± 1.9411.6 ± 2.96
R2*42.5 ± 3.0142.2 ± 3.2736.5 ± 3.35
R224.1 ± 1.6225.2 ± 1.9823.6 ± 1.45


All MR images were acquired with a 3.0-tesla, 80-cm (inner diameter) magnet (Magnex Scientific, Abingdon, England) with a maximum gradient strength of 18 mT/m and 250 microseconds ramp time. A quadrature birdcage head coil was used for imaging. Spin-echo sagittal images were obtained to align the imaging plane so that it was parallel to the plane encompassing both the inferior colliculus and mamillary body.

Multislice measurements of R2, R2′, and R2* were performed in a single acquisition using the gradient-echo sampling of free induction decay and echo (GESFIDE) sequence.10 The timing of the echoes in this sequence was identical to that used previously.8 Briefly, two slice-selective 90° and 180° radiofrequency (RF) pulses were used in this sequence. Five gradient echoes were acquired between the 90° and 180° RF pulses, followed by acquisition of six echoes after the 180° pulse. The final echo produced a spin echo at 98 milliseconds. Sixteen thin contiguous 2.2-mm slices from the pontomedullary border to slightly above the superior border of the putamen were obtained within 10.7 minutes. A 128 × 128 imaging matrix with a 220-mm field of view and a repetition time of 2500 milliseconds was used for image acquisition. Even and odd slices were obtained in separate scans to avoid interference from adjacent slices. The imaging time of the entire protocol, inclusive of positioning and shimming, was approximately 20 minutes.

Image Analysis.

The image sets were Fourier-transformed and zero-filled to yield 256 × 256 in-plane images for each of the 176 two-dimensional images (16 slices × 11 echoes). Maps of R2, R2′, and R2* were obtained as described in previous studies.8 The procedure involved construction of R2* maps from the first five echoes and R2 (R2 = R2 − R2′) maps from the last six echoes of the GESFIDE sequence.10 Further, the R2* and R2 maps were converted to R2 and R2′ maps using the expressions given below:

R2 = (R2*+ R2)/2

2R′ = (R2*− R2)/2

Brain tissue segmentation was performed using the Iterative Self-organizing Data Analysis Technique (ISODATA) clustering technique.11 It is a semiautomated algorithm based on techniques of multivariate statistical analysis. The algorithm is based on Euclidean measures of pattern similarity. A vector feature is constructed at each spatial location from the set of input data. The clusters are determined in such a way that the intraset distance in each cluster is kept to a minimum, and the interset distance between two clusters is made as large as possible. The number of source images determines the dimension of the Euclidean (feature) space in which the clustering is carried out. To improve tissue specificity, qualitatively different images were used to increase the likelihood of separating two tissues with similar signature profiles.

For each subject, all image slices reconstructed from the final echo (TE/TR = 98 /2500 millisecond) of the GESFIDE sequence were reviewed visually to identify and localize slices containing the PAG, RN, and SN. The PAG, RN, and SN were visible in two or more slices for all cases reported in this study. The ISODATA segmentation was used to accurately delineate and identify the entire volume of the PAG, RN, and SN. Eleven images (one from each echo of GESFIDE sequence) were used as source images for the ISODATA segmentation of each slice. This technique ensured that anatomic borders were not crossed. An operator classified the resulting zones into gray matter, white matter, cerebrospinal fluid (CSF), PAG, RN, and SN. For each subject, the R2 (1/T2), R2* (1/T2*), and R2′ (1/T2 − 1/T2) relaxation rates of the RN, SN, and PAG were obtained for the left and right sides separately by projecting the corresponding segmented zones onto the maps. Measurements derived from multiple slices for each subject were expressed as a weighted average.

Statistical Analysis.

The relaxation rates R2, R2′, and R2* of the PAG, RN, and SN for each subject are presented as mean ± SD. One-way ANOVA was used to test the null hypothesis of equality of population means amongst the three groups; the Student paired t test was used to compare the relaxation rates for the right versus left sides. In addition, t tests were performed intragroup to compare differences in relaxation rates between the genders. Correlation analysis also was performed intragroup to determine the relationship between variables, and Pearson product moment correlation coefficient was calculated to determine the extent of linear relationship. Significance levels were set at P<.05.


Representative images containing the PAG, RN, or SN are shown in Figure 1. The RN and SN are readily identified as hypointense regions and the PAG is adjacent to the cerebral aqueduct (hyperintense area), as illustrated in Figure 1. The ISODATA segmentation and classification of the different zones is illustrated in Figure 2. Representative maps of R2, R2′, and R2* for a section through the midbrain are shown in Figure 3.

Figure 1.—.

Spin-echo magnetic resonance images (echo time/repetition time [ TE/TR] = 98/2500 millisecond) of three contiguous 2.2-mm, thin slices obtained with the gradient-echo sampling of free induction decay and echo sequence are shown in the top panel. Magnified view of the center image is presented in the bottom panel. Arrows pointing to the hypointense areas correspond to the substantia nigra (SN) and the red nucleus (RN). The cerebral aqueduct (CA) is the hyperintense area, and surrounding the CA is the periaqueductal gray matter (PAG).

Figure 2.—.

Results of Iterative Self-organizing Data Analysis Technique (ISODATA) segmentation; the segmented zones superimposed on the T2 image (enlarged view) are displayed in the top and bottom panels, respectively, on the right. The T2-weighted image (TE/ TR = 98/2500 millisecond) in the left panel is a spin echo showing a region of the midbrain containing the cerebral aqueduct (CA), periaqueductal gray matter (PAG), red nucleus (RN), and substantia nigra (SN). Eleven images, one from each echo of the grant-echo sampling of free induction decay and echo pulse sequence were used in the computer cluster analysis (ISODATA) to segment brain tissue. Each color of the ISODATA segmented image is representative of a different cluster with unique properties in feature space.

Figure 3.—.

Maps of the transverse relaxation rates R2, R2, and R2* from a single 2.2-mm slice are shown in the top panel. Image reconstruction and calculation of these maps is detailed in the text. The bottom panel shows enlarged views of the corresponding map detailing the brain stem structures that were studied. The red nucleus (RN) and substantia nigra (SN) appear hyperintense on the three relaxation rate maps. The isointense area surrounding the cerebral aqueduct (CA) is the periaqueductal gray matter (PAG).

The mean values for R2, R2′, and R2* obtained for the PAG, RN, and SN of the left and right hemispheres for the N, EM, and CDH groups are presented in the Table 1. Using the Student paired t test, no intragroup side-to-side differences were noted in the relaxation rates of the PAG, RN, and SN of any of the groups. Therefore, results from both sides were pooled. No gender-related differences were demonstrated in the relaxation rates of these structures for all three groups. The mean R2 values in the PAG, RN, and SN of the three groups were not significantly different. Similar results were observed for both the left and right sides.

In the PAG, there was a statistically significant increase in R2′ and R2* values in both the EM and CDH groups (P<.05), compared with the N group (Table 1). No statistically significant differences were noted between patients with MWOA and patients with MWA so the results were pooled as the EM group. No significant differences were demonstrated in either the R2′ or R2* values for the EM and CDH groups. For the N, EM, and CDH groups, we performed correlation analysis of the age of the subject with the R2′ value; no association was observed for any group. A positive correlation was noted between the duration of illness and the increase in R2′ for the EM and CDH groups (Figure 4). The intercepts of the two slopes were nearly identical and suggest that R2′ values may have been high at the outset of illness.

Figure 4.—.

Correlation analysis of the duration of illness with R2 value. For the episodic migraine (EM) group (top panel: circles), the correlation coefficient (r) is 0.65 (P = .0052) and for the chronic daily headache (CDH) group (bottom panel: triangles), r is 0.71 (P = .0053). The intercept of the regression line for EM is 5.6698 ms1; for CDH, it is 5.5659 ms1. There is no significant difference between the slopes of the regression lines of the EM and CDH groups.

No differences in R2′ and R2* were observed in the RN and SN of the patients with EM (Table 1). A significant decrease in R2′ and R2* values of the RN and SN was found in the CDH group, compared with the N and EM groups (P<.05), but no significant difference in the relaxation rates of either the SN or RN for the N and EM groups was demonstrated.


Using high-resolution MR techniques, we have mapped transverse relaxation rates R2, R2′, and R2* in patients with EM and patients with CDH and compared them with normal controls. An increase in R2′ is specific to deposition of non-heme iron in tissues.8 A decrease in R2′ and R2* reflects the influence of free iron from deoxyhemoglobin.12,13 No alteration of R2 was measured in any of our subject groups, providing important structural evidence of no change in tissue water content.

To date, these imaging data provide the most specific evidence of disturbed PAG function in patients with migraine. Iron levels were equally elevated in the PAG of subjects with migraine with or without aura. Not only was a significant increase in tissue non-heme iron found in both EM and CDH groups compared with controls, but levels increased with the duration of illness. Because there was no correlation with age in any group, our data suggest that iron accumulation over time in the EM and CDH groups may be caused by repeated headache attacks. The intercept values of the correlation analyses in both migraine groups also raise the possibility that tissue iron values were higher than normal at the outset in migraine-susceptible individuals, supporting cause as well as consequence.

No increased iron levels were measured in the RN or SN of patients with EM, with or without aura. Instead, decreased R2′ and R2* in the RN and SN were recorded in the CDH group, best explained by flow activation and hyperoxia of these structures caused by headache during the study. This was supported by unchanged R2′ and R2* in patients with EM without headache. Activation of the RN and SN was observed previously using fMRI-BOLD12; the precise reasons remain to be established but include both nociceptive and autonomic dysfunction. A study of these patients with CDH when headache-free would be required to evaluate if there is increased tissue iron in their RN and SN. The PAG showed no evidence of similar activation to the RN and SN in the CDH group, possibly because of dysfunction or because high levels of tissue non-heme iron may have attenuated changes in fMRI-BOLD signal.

We offer a number of explanations for the high PAG iron levels in our migrainous patients. The concentration of transferrin receptors that transport iron in and out of the PAG is the highest of any brain region.7 Transferrin receptor density may be a marker of the cellular requirements for iron during oxidative metabolism14 and may be influenced by nociceptive function. For example, morphine injected into the PAG down-regulates transferrin receptors.15 High tissue iron levels might suggest that the PAG is abnormally metabolically active in migraine, even between EM attacks, or has a higher density of metabolically active neurons. This would not explain accumulation of iron in the PAG with duration of illness and presumably the increasing burden of attacks. Those neurons that have high resting iron levels and the potential for iron-induced oxidative stress may be selectively vulnerable to free radical damage.16 Overexpression of transferrin receptors could lead to tissue iron accumulation and iron-catalyzed free radical cell damage, accentuated by repeated episodes of hyperoxia during migraine with or without aura.12,13 Compensatory increases in neuronal metabolism of surviving neurons might also lead secondarily to increased iron uptake.14 Glial cells have a high iron content,17,18 but there was no evidence of gliosis on MR anatomical images in our patients to account for high tissue iron levels. Thus, we believe that the increased R2′ values in our migraine groups reflect impaired iron homeostasis, possibly associated with neuronal dysfunction or damage.

Increased PAG iron levels were common to migraine with or without aura. Currently conceived as due to spreading depression triggered in hyperexcitable occipital cortex,13 migraine aura is difficult to explain on the basis of PAG dysfunction. Directly implicating PAG dysfunction as a cause of headache through interruption of its normal antinociceptive function seems more plausible. Nevertheless, experimental PAG stimulation produces a widespread increase in cortical blood flow that is mediated by nitric oxide.19 Also, the dorsal raphe component of the PAG sends diffuse serotonergic projections to the cerebral cortex and vessels that could influence neuronal excitability and vasomotor control.20 It remains to be determined if these ascending projections contribute to hyperexcitability or in some way alter the threshold for triggering migraine aura.

The PAG iron levels were equally abnormally high in the patients with EM and those with CDH. Our original concept was that repeated episodes of hyperoxia of brain stem structures activated during migraine attacks might make these regions at risk for iron-catalyzed free radical damage.12 We reasoned that iron accumulation would be a marker of progressive PAG dysfunction and the evolution of episodic to chronic head pain. However, we have not been able to show differences between the EM and CDH groups with respect to levels of tissue iron and its accumulation with the duration of illness to support this hypothesis. Nevertheless, because PAG is involved in behavioral manifestations of opiate withdrawal, PAG dysfunction could enhance the susceptibility of patients with migraine to develop drug-induced rebound or CDH.21

Problems with this study include the lack of an episodic or chronic pain control group, which could establish the specificity of the findings to migraine. Second, if the PAG is persistently dysfunctional in EM, patients might experience increased pain sensitivity generally. Although patients with migraine often complain of body and limb pain, we did not specifically question or examine our patients to document this. Conversely, components of the PAG selectively projecting to the trigeminal system alone might be dysfunctional, since presumably these are the regions repeatedly activated during the migraine attack. Unfortunately, our imaging techniques do not have the resolution to evaluate the iron content of select subregions of the PAG. Third, because there were no side-to-side differences in PAG iron, we could not relate any laterality of head pain to laterality of PAG iron content. Finally, the highest tissue iron levels were measured in those patients who suffered prolonged illness with severe and frequent EM or daily headache. Epidemiological evidence for decreased prevalence of migraine attacks in the majority of older age subjects might argue against permanent structural change in the PAG22 and, accordingly, a critical role in migraine. However, the reasons for reduced migraine prevalence in later life remain to be determined. Shifts in factors that govern the threshold for migraine attacks, such as estrogen fluctuations or altered responsiveness of the neurovascular networks usually activated during migraine, might be extrinsic to PAG dysfunction.23 Studies of the PAG iron levels of patients recovered from migraine might resolve these issues in part.

During attacks of MWOA, Weiller et al observed consistent increases in regional cerebral blood flow in brain stem regions that included the PAG, midbrain reticular formation, and LC.6 Because activation continued despite alleviation of headache with treatment, they postulated that these centers were “generators” of migraine headache. It was beyond the resolution of the PET method to identify distinct brain stem nuclei. Our methods of image processing were able to select the PAG for specific analysis of tissue iron, providing evidence that iron homeostasis in the PAG is persistently and progressively impaired due to iron-catalyzed free radical injury from repeated migraine attacks. Because elevation in tissue iron is a marker of disturbed neuronal function,7 these results support and emphasize the role of the PAG as a potential “generator” of MWOA, possibly through dysfunctional control of the trigeminovascular nociceptive system.

Acknowledgments:  This work was supported by NIH grant P50-NS32399 to Dr. Welch. Appreciation is expressed to Dr. Lonnie Schultz for statistical advice.