^{1}

^{*}

^{2}

^{*}

Our aim is to determine the cause(s) of migraine auras. Our understanding of how migraines and migraine auras originate is very imperfect. An important observation is that migraines occur more frequently in women at reproductive age than in men at a similar age. This suggests that gonadal hormones may be relevant triggers. The occurrence of classical (typical auras without headache) auras in one author (WB) has been recorded. Every aura in six years (85) has been noted, studied and analysed statistically. The auras occur predominantly in Spring and Autumn, especially in the longer reproductive season of Spring. This association is supported statistically. The results support the idea of gonadal hormones as relevant triggers, the strongest candidate being estrogen. Basic mechanisms underlying the auras are discussed, especially the phenomenon of cortical spreading depression. We also propose that both auras and migraines depend upon previous injury to the head or to the brain, giving rise to a condition of “deafferentation hypersensitivity”.

Headaches may be the longest-known medical condition and, among them, migraines may be the best known. In spite of this long history, there is no agreement as to the cause of the migraines. This is probably because migraines are a heterogeneous group. Many are accompanied, usually preceded, by hallucinations mostly either in the visual or somatosensory modalities. These hallucinations are termed “auras” and can occur in the absence of the headache. Auras are of many different types. Recently [

WB has been having visual auras without headache for about 13 years. These auras are of the HIS 1.2.3 type (Typical aura without headache), previously known as “classical” or “fortification”. These commenced at a time when he experienced a macular hole in his right eye and a macular edema in his left eye. This condition initially led to hallucinations known as the Charles Bonnet Syndrome [

The dates of the Equinoxes were taken to be 21 March and 23 September. In Australia Spring is defined as 1 September to 30 November and Autumn as 1 March to 31 May. Because of the possibility that the ambient temperature might influence the occurrence of the auras, the daily and average monthly temperatures were obtained from the Australian Bureau of Meteorology (www.bom.gov.au).

Both authors have had full access to all study data.

Statistical ApproachIn order to examine a seasonal effect on aura, we propose to consider a generalised linear model approach. It is assumed that the number of occurrences of auras in a month follows a Poisson distribution with mean an exponential function of a linear regression using the explanatory variables: years, temperature and months (see for example, Chapter 6 of McCullagh and Nelder, 1989) [

peak frequencies at month 3 (March), month 8 (August), month 10 (October) and month 11 (November). These months have an apparent relationship to the climate seasons of Spring and Autumn, especially the equinoxes.

Can the frequency pattern of WB’s auras be related to the equinoxes?

Although the equinoxes are strongly associated with reproduction and survival in both animals and plants, these functions are also dependent on several other factors. Therefore, we examined the possibility that the ambient temperature might be a triggering factor.

The best fitting model considered includes as predictor variables the factors Year, Temperature, an effect for comparison of first and second half year, and a seasonal effect modelled by a cosine function with a seasonal cycle corresponding to a biannual or six-monthly effect and a phase parameter determining the start of the cycle. This has the smallest AIC value compared to a series of models obtained by an extensive search and in particular is shown to be preferable to a complete model based on the factors Year and Month (P-value for the test 0.135 and AIC values 211.2 and 209.6). The P-values for including the linear effect of Temp, the biannual seasonal effect with phase parameter and half year effect (hY) are 0.1255, 0.0404 and 0.0345, respectively.

We have described the monthly occurrence of classical auras in one person (WB) over six years and analysed the data statistically.

The conclusion we have reached from our statistical analysis is that the auras have occurred at the times of the year closely associated with the equinoxes, i.e. in the Spring and Autumn seasons. The data in this paper were obtained from a single person, so no general conclusions can be drawn from them. We are not aware of any similar data published anywhere else Nevertheless, there are other different data already available that do sup-

port our proposals and will encourage research along similar lines.

The equinoxes occur near the beginnings of the seasons of Spring and Autumn, which are both associated with reproduction. The significant statistical linkage between the auras and the reproductive seasons encourages us to suggest a hormonal trigger for the auras. Although the seasonal dependence of reproduction is strong in most mammals [

Migraine is three times as common in women of reproductive age as in men of similar age; in women this commonly occurs at menstruation. Attacks of migraine without aura are correlated with “estrogen withdrawal”

[

There is fairly wide agreement that the underlying basis for the auras is a pathophysiological event called cortical spreading depression (CSD). Although the original account of this phenomenon was due to Leão [

It is well established that the cortices of migraineurs are hyperexcitable [

We suggest also that a feasible trigger for the auras is estrogen. No tests were applied to WB at the time of an aura, simply because it has only recently been realized that a seasonal rhythm existed and that estrogen might be a trigger. We are uncertain how estrogen might act as a trigger to initiate the aura. However, we note that estrogen can elicit seizures when applied to an epileptogenic focus on the cortex [

We examined the possibility that temperature might be a trigger for auras but our statistical analysis did not show a significant effect. Nevertheless, with more data and perhaps a different method of analysis the possibility still exists.

In his classical book on Migraine [

Statistically there is a significant correlation between the auras in WB and the seasons of Spring and Autumn. The results support the idea of gonadal hormones as relevant triggers, the strongest candidate being estrogen. The neural mechanism underlying the auras is believed to depend on the phenomenon of cortical spreading depression and in WB’s case this is due to deafferentation hypersensitivity. We also propose that both auras and migraines depend upon previous disorder in the head or brain.

1. It is highly desirable that more data be collected on the frequency of occurrence of both auras and migraines. Because this will take up much time it requires many very reliable patients. They should be provided with special diaries and perhaps given some incentives. The easiest task is to collect data on classical (HIS 1.2.3) auras, because these are non-graded, all-or-nothing events. Nevertheless, even an examination of data already collected in clinics might reveal a relationship with Spring and Autumn seasons.

2. The medical history of all migraine patients should be examined to determine what neurological diseases or injuries to the head or brain they have sustained. Is there any evidence of deafferentation hypersensitivity? Attempts should be made t o find correlates between the data and the auras or migraines.

3. Although much research has been done on the role of estrogen in migraine, most of this work has been on women and female animals. The role of gonadal hormones in migraines in men needs extended research, especially the analysis of hormones during auras and migraines.

4. The role of temperature in the occurrence of auras and migraines needs more investigation.

5. An important lesson from the paper of Bronson [

Neither author has any conflict of interest.

We are not in receipt of any research funding. But we are grateful to the University of Sydney for the provision of space and sundry clerical assistance.

We are grateful to Dane King for assistance with the illustrations.

The data are number of aura per month collected over a 6 year period, together with year, month and maximum monthly temperature. A plot of Aura against Month in

The number of aura per month might be expected to be from a Poisson distribution with mean depending on the variables: year, average maximum temperature of month (Temp), and effects related to seasons. The data are set out as vectors Aura, Year, Temp and Month, each of length 72. So the logarithm of the mean of the Poisson is taken as a linear form:

where hY takes values 0 for Months 1 to 6 and 1 for months 7 to 12 and the final term represents a biannual effect with starting point a months before mid-January. We note that a seasonal effect cos(2*pi*Month/12) is highly correlated (r = 0.81) with Temp and so is excluded. We fitted the parameters b0, ..., b8 using the glm package in the R language, then maximised the likelihood over the phase parameter a. Many sub-models were fitted to check that we had this best model. The effect of Temp could be removed from the model as this provides effectively the same AIC and a test for the additional effect of fitting Temp is not significant.

A check on the fit can be made visually as in Supplementary

The analysis was performed as shown in the following output from R. Note that the estimate of a as 0.7 was made separately by minimising the deviance for different values of a. So, the AIC for the model should be 209.6.

Call:

glm(formula = Aura ~ as.factor(Year) + Temp + hY + cos(2 * pi * (Month − 0.7)/6), family = poisson)

Deviance Residuals:

Coefficients:

Estimate Std. Error z value Pr(>|z|)

(Dispersion parameter for poisson family taken to be 1).

Supplementary

Null deviance: 124.973 on 71 degrees of freedom

Residual deviance: 85.736 on 63 degrees of freedom

AIC: 207.6

plot(Month,(Aura-BFMod$fit)/sqrt(BFMod$fit),ylab="Standardised Residuals",main="Figure6")

abline(c(0,0))