GENERAL DISCUSSION 217 10 study,62 and restriction to female participants in another study,92 but no other correctional steps were taken in the design or statistical model for any of the other confounding factors (BMI, alcohol consumption, smoking and depression). In the case of PGE2 less confounders are known, but it seems likely that age is an important factor, as PGE2 plays a role in inflammation, 93-96 and aging is associated with dysregulation of the immune and inflammatory responses. Neuroinflammation has been implicated to contribute to neurodegeneration in normal aging and in age-associated neurological disorders.97 A recent study showed, for the first time, in vivo compelling evidence of neuroinflammation in migraine.98 It has previously been shown that peripheral blood mononuclear cells from elderly human subjects as well as macrophages and splenocytes from old mice make significantly more PGE2 than their young counterparts. 99, 100 This increase was mainly mediated via an increase in COX-2 expression.101 In Chapter 5, where PGE 2 levels were investigated in relation to a provoked migraine attack, our statistical model was adjusted for age. In none of the other studies on PGE2 levels there was a mention of age being taken into account. 63-66 In the most recent study, age was not even reported for the individuals under investigation.66 In Chapter 4, an extensive profile of metabolites was investigated, hence we corrected our regression model for the most commonly known metabolic confounders; age, sex, BMI and smoking status. In conclusion, it is very important prior to starting a metabolomics study and prior to analysing data from such study to consider the role of confounders to be able to meaningfully quantify biological signals of interest. Conflicting results between studies can be the cause of a large heterogeneity in sample collection and analysis and therefore such studies need to be performed with caution. Genetics of migraine Since the last two decades, genome-wide association studies (GWAS) have been able to identify many thousands of low-effect risk DNA variants for numerous diseases.102 An important challenge in GWAS is to determine the exact gene that is affected by a risk SNP, as this is often far from straightforward. A complicating factor is that the location of the associated SNP is often intronic or intergenic, were it functions merely as genomic marker at a distance of, so in linkage disequilibrium (LD) with, the true causal variant. One of the most used methods to link a risk locus/index SNP (i.e., the SNP with the lowest p-value) to a gene is to relate it to the nearest gene. However, in twothirds of cases it was demonstrated that it is not the nearest gene that is affected by the risk SNP.103, 104 Moreover, there can be multiple independent association signals at the same locus (secondary SNPs) that can influence other regulatory features of the same (or nearby) gene. Most risk SNPs do not directly affect amino acid changes, but appear to regulate gene expression indirectly by disrupting enhancer elements. A way to determine the causal gene more effectively is by integrating data from GWAS and expression data in a way that expression quantitative trait loci (eQTL) analyses is able to prioritize likely causal genes.102 Chapter 6, 7 and 8 illustrate that the road ahead to understanding the pathophysiology behind migraine and cluster headache risk loci is far from easy.
RkJQdWJsaXNoZXIy MTk4NDMw