Cerebrospinal Fluid and Plasma Amine Profiles in Interictal Migraine

Impaired amine metabolism has been associated with the etiology of migraine, that is, why patients continue to get migraine attacks. However, evidence from cerebrospinal fluid (CSF) is lacking. Here, we evaluated individual amine levels, global amine profiles, and amine pathways in CSF and plasma of interictal migraine patients and healthy controls.

of patients experience weekly attacks. 3 Genetic and nongenetic risk and susceptibility-modifying factors are involved. 4 Although the mechanisms of migraine attacks are reasonably well understood, 5,6 little is known about the pathogenesis of migraine disease, that is, the tendency to have recurrent migraine attacks. 1,4 This has hindered the development of a causal prophylactic therapy that prevents the initiation and occurrence of migraine attacks. Even the recently emerging calcitonin gene-related peptide (CGRP) inhibitors have limited efficacy, probably because they most likely mainly act at the end of the initiation cascade of the migraine attack rather than at the beginning. 6,7 Unraveling the neurochemistry of migraine could help elucidate its etiology and the reason people get recurrent attacks. 4,8 Most biochemical studies in migraine have focused on blood and urine. 5,9 Although these fluids are easier to collect, they reflect systemic rather than cerebral changes. 10 Cerebrospinal fluid (CSF) could better reflect brain neurochemistry, 11,12 but its sampling is hampered by logistical and ethical issues.
More than 70 metabolites from different molecular classes such as neurotransmitters, neuropeptides, cytokines, hormones, and ions have been measured in the CSF of migraine patients. 9,13 Several amines such as glutamate, glutamine, γ-aminobutyric acid (GABA), and serotonin have been implicated in the pathogenesis of migraine. 4,9,14 However, glutamate was the only amine measured in the CSF of migraineurs in 3 or more different studies and was found to be elevated during the ictal phase and in the chronic form of migraine. 9 Studies in the CSF of migraineurs were small and had significant chemical detection and clinical methodological limitations. Details of the measurement methods used and their validation were often reported only to a limited extent. 9 Moreover, only a single molecule was usually measured rather than several simultaneously, and in only a limited number of cases were multiple molecules measured simultaneously in both CSF and plasma. 9 This prevented a more detailed analysis of the cerebral and systemic profiles and pathways. In addition, controls were rarely matched for important confounding factors such as sex, age, and diurnal and seasonal timing of sampling. 9 They were also rarely truly healthy, as control CSF was usually obtained from lumbar punctures in people with neurological symptoms to exclude neurological disease. 9 Finally, many studies compared CSF collected between attacks only with CSF collected during attacks, but not with CSF obtained from healthy controls. 9 Thus, these studies focused on changes occurring during and mechanisms of the attack rather than on the pathogenetic mechanisms of migraine disease, to identify mechanisms for why people get recurrent migraine attacks. 9 In our search for mechanisms of susceptibility to getting migraine attacks, we investigated whether amine metabolism is impaired in people with migraine outside of attacks. To this end, we determined and compared the levels of 31 amines and their global profiles and pathways in CSF and plasma of n = 99 interictal patients with migraine with aura, n = 98 with migraine without aura, and n = 96 healthy volunteers. Healthy volunteers had no personal or family history of migraine. The 3 groups were matched for age, sex, and diurnal and seasonal timing of sampling.

Study Design and Participants
We enrolled patients with migraine with or without aura 2 and healthy controls who were group-matched for sex and age by adhering to 5-year age strata; age ranged from 18 to 69 years. Patients were diagnosed according to the International Classification of Headache Disorders criteria (ICHD3) by experienced physicians (supervised by G.M.T. and M.D.F.). 2 For the first few years, we used the ICHD2 version without essential differences between patients. 15 Participants were recruited from the LUMINA (Leiden University Medical Center Migraine Neuro Analysis) project, which includes migraine patients and healthy controls from the general Dutch population who agreed to participate in migraine-related scientific research. Some controls and patients were recruited through a public advertisement or through the Leiden University Medical Center Headache Clinic. Migraineurs did not use acute migraine drugs on more than 8 days per month. CSF and plasma were collected interictally when patients had been attack-free for ≥3 days. Sampling occurred between April 2008 and May 2016. Healthy volunteers had no obvious signs or symptoms of a disease and had no history of headache (except for infrequent tension-type headaches) or other pain syndromes. To exclude any underlying genetic risk, they also did not have first-degree relatives with migraine or trigeminal autonomic cephalalgia. Participants did not have a severe psychiatric disorder, nor a history of oncological disease, or a contraindication for lumbar puncture (signs and symptoms of increased intracranial pressure, local skin infection, or a coagulopathy including use of anticoagulant drugs or platelet inhibitors). The study was conducted according to the criteria of the Declaration of Helsinki and approved by the Leiden University Medical Center institutional ethics committee. All participants provided written informed consent prior to participation and received financial compensation according to standard fees for participation in similar studies.

Sample Collection
Patients and controls were sampled, after standard neurological examination, between 8:30 AM and 1:00 PM and in random order and interchangeably over the time span to avoid diurnal and seasonal variation. They abstained from eating or drinking, other than water, for at least 8 hours prior to sampling to minimize the influence of food or drinks. CSF was sampled by lumbar puncture in the L3/L4, L4/L5, or L5/S1 interspace. Intracranial pressure was measured, and 3.0ml CSF was sampled for routine diagnostics (cell count, glucose and total protein levels). Next, 3.8ml of CSF was sampled directly in a 15ml polypropylene Falcon tube prechilled on ice and centrifuged at 4 C for 5 minutes (2,000rpm, 747 Â g). The supernatant was transferred into a new chilled 15ml polypropylene Falcon tube, inverted several times, and divided into 0.5ml aliquots (1.8ml cryotubes) that already contained 1.0ml of cold ethanol. Cryotubes were inverted several times to mix CSF and ethanol. Samples were placed on dry ice within 30 minutes of sampling and stored at À80 C within 60 minutes. Blood was collected from the median cubital vein in ethylenediaminetetraacetic acid plasma tubes, immediately after lumbar puncture, and centrifuged at 21 C for 20 minutes (2,000rpm, 622 Â g). The supernatant was transferred to a new 15ml polypropylene Falcon tube, inverted several times, and divided into 0.5ml aliquots (1.0ml Nunc [Rochester, NY] cryotubes). Plasma samples were stored at À80 C within 60 minutes from sampling. All CSF and plasma samples remained at À80 C until sample preparation; no extra freezethaw cycles were allowed. See Figure S1 for detailed information on sample processing. To monitor possible complications, participants were followed for 3 days, or longer if necessary.

Amine Measurements
Amines were measured with an ultraperformance liquid chromatography mass spectrometry (UPLC-MS) method specifically developed and validated for amine profiling. 16 Samples were randomized across 5 CSF batches and 5 plasma batches. All batches included calibration lines, blanks, and quality control samples. Blanks were used to subtract background levels from study samples. Quality control samples were analyzed every 10 samples and were used to monitor data quality and to correct for instrument response. 17 For the analysis of CSF, 30μl of CSF/ethanol sample was spiked with an internal standard solution. For the analysis of plasma, 5μl of starting material was used. For the analysis of amino acids, 13 C 15 N-labeled analogues were used. 16 For other amines, the closest-eluting internal standard was added. The target list is included as Table S1. After spiking, proteins were precipitated by the addition of methanol and the mixture was vortexed for 10 seconds and centrifuged (9,400 Â g for 10 minutes at 10 C). The supernatant was transferred to a new Eppendorf tube and taken to dryness in a Speedvac. The residue was reconstituted in borate buffer (pH 8.5) with 6-aminoquinolyl-N-hydroxysuccinimidyl carbamate reagent. Next, vials were transferred to an autosampler tray and cooled to 4 C until injection, when 1.0μl of the reaction mixture was injected into the UPLC-MS system. Chromatographic separation was achieved by an ACQUITY UPLC system (Waters, Milford, MA) on an AccQ-Tag Ultra column (Waters) with a flow of 0.7ml/min over a 11-minute gradient. The UPLC was coupled to electrospray ionization on a triple quadrupole mass spectrometer (Qtrap 6,500; AB SCIEX, Nieuwekerk aan den Ijssel, the Netherlands). Analytes were detected in the positive ion mode and monitored by multiple reaction monitoring (MRM) using nominal mass resolution. Acquired data were evaluated using MultiQuant Software for Quantitative Analysis (v3.0.2, AB SCIEX), by integration of assigned MRM peaks and normalization using internal standards. Unit-free concentrations are reported in this paper (ie, relative response ratios, calculated as "amine target area"/"internal standard area").

Statistical Analysis
Amine concentrations were first corrected based on quality control samples. 17 Only metabolites with relative standard deviations of quality controls below 15% were included in further analyses. Outlier detection was performed using principal component analysis with a 99% confidence interval (Fig S2). Logtransformed concentrations were used for further statistical analyses.
Univariate. To detect univariate metabolite differences, analysis of covariance was performed with age and sex as covariates. Bonferroni was used for multiple testing correction, separately for CSF (30 tests) and plasma (31 tests). Additionally, we analyzed the data using the less conservative false discovery rate (FDR). For significant metabolites (p < 0.05 after multiple testing correction) post hoc comparisons were tested according to Shaffer's method. 18 Multivariate, Global Amine Profiles and Specific Pathways. To detect multivariate differences in global metabolite profiles or amine pathways, the global test approach 19 was used (separately for CSF and plasma). First, global amine profiles of the 3 groups were compared with age and sex as covariates. Second, pathways were tested. We used MetaboAnalyst to obtain Kyoto Encyclopedia of Genes and Genomes (KEGG) identifiers 20 from the 31 amines and subsequently downloaded KEGG metabolite set pathway definitions from Con-sensusPathDB. 21 Because most KEGG pathways were not covered in a coherent fashion by the amines measured on our platform, we filtered the metabolite sets by eliminating those amines that were disconnected from the rest in the pathway diagram. Specifically, we computed a metabolic interaction network of the 31 amines from a curated version of the human Genome-Scale Metabolic Model (GSMM) Human Metabolic Reaction 2.0, 22 using a generic path-finding algorithm that was developed in house. Only biochemical paths of one or two reaction steps were considered, and each path was checked for stoichiometric and thermodynamic consistency. In addition, only substrate-product mappings were considered that involved the transfer of carbon-based moieties, except for S-adenosylmethionine, which was considered a hub metabolite. As a consequence, half-reactions involving the transfer of electrons, amino or phosphate groups were decoupled from the main reaction in the path-finding procedure. For example, in the reaction nicotinamide adenine dinucleotide (NAD)H + pyruvate <=> NAD+ + lactate, only NADH and NAD+ are linked in the network, and pyruvate and lactate are linked. Likewise, in the reaction glutamate + pyruvate <=> α-ketoglutarate + alanine, only glutamate and α-ketoglutarate are linked, and pyruvate and alanine are linked. In this way, we prevented the creation of a crowded, highly connected network in which all amines are interconnected via a few hub metabolites such as H+, H2O, adenosine triphosphate (ATP), and NADH. Finally, transporter reactions that transferred compounds over the cellular membranes and (half) reactions that involved uniquely produced metabolites were given a weight of zero in the path-finding procedure.
Computations were performed in MATLAB 2019b (MathWorks, Natick, MA). Network analysis of the GSMM resulted in 2 larger connected networks of 9 and 10 amines, respectively, and 2 smaller networks of only ethanolamine and o-phosphoethanolamine and of phenylalanine, tyrosine, and 3-methoxytyrosine. Overlapping the network with the KEGG metabolite sets and eliminating disconnected amines resulted in 10 independent pathways that contained a subnetwork of 3 amines or more: 1. Cysteine and methionine metabolism 2. Glycine, serine, and threonine metabolism 3. Lysine degradation 4. GABAergic synapse (same set definition as alanine, aspartate, and glutamate metabolism) 5. Arginine and proline metabolism (glutathione metabolism is a subset of this pathway) 6. Arginine biosynthesis 7. Aminoacyl-tRNA biosynthesis (same set definition as protein digestion and absorption) 8. Central carbon metabolism in cancer 9. Mineral absorption 10. ATP-binding cassette transporters After inspection of the new list of metabolite sets, we decided to exclude the last 4 pathways, because they involved unspecific biological processes that were not relevant for this study. To test whether the defined pathways differed between the 3 groups, we used the global test function with the pathways as subsets and age and sex as covariates. Bonferroni was used to correct for the final number of pathways (n = 6).
Finally, to determine whether amine profiles could predict diagnosis, a logistic regression model was developed. Migraine diagnosis was set as dependent variable and amines, age, and sex were used as predictors. Overfitting of the model was prevented using L2 ridge penalization. 23 Predictions resulted from cross-validation.

Study Population
We included n = 96 healthy volunteers and n = 197 patients with migraine, n = 99 with aura and n = 98 without aura. After inclusion, n = 5 healthy controls were found to have a first-degree relative with migraine. However, their results were kept in the study because CSF from healthy volunteers is difficult to obtain and a separate analysis of these 5 controls did not yield any abnormal results. In 1 patient with migraine with aura, we were unable to collect plasma. Additionally, 2 outliers were excluded from further data analysis ( Fig S2). There were no differences in clinical characteristics between the final study groups (Table 1), except for a higher monthly attack frequency (2.8 AE 2.6) in migraine without aura versus migraine with aura (2.1 AE 1.9, p = 0.026). In total, 92 (31.4%) participants developed post-dural puncture headache (fewer in patients with migraine than in controls, 31.0% vs 32.3%; see Table 1 24 ), of whom 18 required a blood patch.

Univariate Results
In total, 30 amines were reliably detected in CSF, and 31 in plasma. In CSF, L-arginine levels were different between the 3 groups (Table 2; p = 0.042 after Bonferroni correction). Post hoc analysis showed 10.4% lower levels in migraine with aura (p < 0.001) and 5.0% lower levels in migraine without aura (p = 0.027) versus controls, without difference between migraine subtypes (p = 0.153; Fig 1). Because L-arginine is potentially related to cardiovascular status, 25 we repeated the analysis after exclusion of participants with possible cardiovascular comorbidities (n = 12) or antihypertensive medication for migraine prevention (n = 20), but differences remained significant (Table S2). Participants who developed a migraine attack ≤3 days after sampling (and thus potentially were in a preictal phase), did not show different results (Fig S3). The ability of L-arginine to predict migraine was modest, with an area under the curve of 0.657 for migraine with aura and 0.560 for migraine without aura (Fig S4).
None of the other metabolites in CSF differed significantly after Bonferroni correction (see Table 2). Using FDR correction, CSF L-phenylalanine also differed; concentrations were reduced in migraine with aura (6.9%, p = 0.011) and without aura (8.1%, p = 0.001; Fig S5).  In plasma, there were no metabolite differences after correction for multiple comparisons (Table 3 and Fig S6).

Multivariate Results
In CSF, global metabolite profiles differed between groups (p = 0.035). Whereas there were no differences between migraine subtypes (p = 0.640), profiles for migraine with aura (p = 0.029), migraine without aura (p = 0.020), and all migraine participants combined (p = 0.009) did differ from those in controls (Fig 2). Again, the ability to predict migraine was modest, with an area under the curve of 0.67 for migraine with aura and 0.63 for migraine without aura (Fig S7).
In plasma, results were similar to those in CSF. Profiles differed between groups (p = 0.021). Whereas there were no differences between migraine subtypes (p = 0.275), profiles for migraine with aura (p = 0.028), migraine without aura (p = 0.018), and all migraine (p = 0.011) did differ from those in controls. The area under the curve of the prediction model was 0.61 for migraine with aura and 0.57 for migraine without aura (Fig S8).
Pathways present in our metabolite set are visualized in Figure 3. In CSF, the pathway "arginine biosynthesis" was significantly associated with migraine with aura (Fig 4; p = 0.004). The "arginine/proline metabolism" pathway showed the strongest association with migraine without aura, but this was not multiple testing resistant. In plasma, the strongest associated pathways were "arginine biosynthesis" with migraine with aura and "arginine/ proline metabolism" with migraine without aura (see Fig  4), but these were only significant before multiple testing correction. Exclusion of L-arginine, to study the effect of other metabolites in these pathways, showed similar results (Table S3).

Discussion
We compared the individual levels, global profiles, and pathways of 31 amines in the CSF and plasma of n = 197 interictal patients with migraine with or without aura with those of n = 96 truly healthy volunteers. The 3 groups were matched for age, sex, and diurnal and seasonal timing of sampling. L-Arginine (and possibly also Lphenylalanine) levels were reduced in the CSF, but not in the plasma of interictal individuals with migraine with or without aura, and global amines profiles were abnormal. Pathway analysis pointed toward a disrupted L-arginine metabolism in migraine with aura. Contrary to previous findings in the literature, no differences were found in the levels of L-glutamic acid (glutamate), L-glutamine, and GABA.
L-Arginine has never been measured in the CSF of migraineurs. 9,13 Studies in serum, plasma, platelets, saliva, and urine have yielded inconsistent findings for and within different biofluids. [26][27][28][29][30][31][32][33] In migraine with and without aura, plasma or serum L-arginine levels were found to be higher, 29,30 whereas in chronic migraine lower levels were found compared to healthy controls. 31 However, 3 other studies failed to detect a difference in L-arginine level between migraine patients and controls. 28,32,33 These inconsistent results can perhaps be explained by different inclusion criteria for patients and controls, different confounding factors, and the lack of adjustment for sex and age. In our study, sex was found to be a highly significant factor determining L-arginine levels.
Because we found no differences in plasma levels but solely in CSF, this could indicate only central metabolism (that is, in the central nervous system [CNS]) of L-arginine is affected and not systemic metabolism. In the CNS, Larginine is unevenly distributed, with the greatest pool in astrocytes, less so in neuronal tissue, and absent in oligodendrocytes. 34 L-Arginine is involved in endocrine activity, immune system modulation, regulation of vascular tone, and peptide and protein production in the CNS. 34 It is also suggested to be neuroprotective; in ischemic stroke models, L-arginine suppresses the elicited proinflammatory response of microglia. 35 In the CNS, the metabolism of L-arginine is closely linked to the metabolism of L-citrulline and L-ornithine (see Fig 3 and Fig 5), which can be synthesized and degraded into each other (see Oja et al 34    L-arginine can be metabolized in L-ornithine (and urea) by arginase. Finally, L-arginine can be metabolized by arginine decarboxylase, resulting in formation of agmatine (and afterward putrescine), or by arginine:glycine amidinotransferase, resulting in guanidinoacetic acid (and eventually creatine). Of the above metabolites, we measured L-citrulline, L-ornithine, and putrescine, and we observed no differences in their concentrations. We studied several ratios (L-citrulline/L-arginine, L-ornithine/L-arginine, and putrescine/L-arginine) to investigate whether one of those was particularly altered in migraine patients compared to controls, but this was not the case (data not shown). Because L-citrulline, L-ornithine, and putrescine levels are also dependent on amines other than L-arginine, this could explain why altered L-arginine levels are not reflected in their concentrations. The link between L-arginine and nitric oxide formation is of special interest in migraine. Glyceryl trinitrate, a nitric oxide donor, can provoke migraine-like attacks in migraineurs but not in nonmigraineurs. 37 Nitric oxide increases cyclic guanosine monophosphate. This second messenger is believed to play a central role in migraine pathophysiology, because it is the common pathway of not only glyceryl trinitrate but also other migraine provocation models. 37 The nitric oxide pathway is probably more upstream in the migraine cascade than the CGRP pathway, because CGRP levels were found to be elevated in migraine attacks provoked by nitric oxide 38 and premonitory symptoms are frequently reported in glyceryl trinitrate-provoked attacks 39,40 while being rare in CGRPprovoked attacks. 41 L-Arginine serves as the only available nitrogen-containing substrate for NOS. 34 The observed lower L-arginine concentrations might therefore reflect overactivity of NOS, elevating nitric oxide levels and making migraine patients susceptible to a next attack, although this hypothesis could not be studied in our study design. There are 3 isoforms of NOS: endothelial, neuronal (n), and inducible (i) NOS. 34 A nonselective NOS inhibitor was reported to be effective in aborting spontaneous migraine attacks, 42,43 whereas specific iNOS inhibitors failed, and trials of combinations of an nNOS blocker and a triptan produced conflicting results. 37,43,44 Future research should aim to study the various NOS activities, especially that of neuronal NOS, and L-arginine concentrations in the CSF of different types of migraine.
At first glance, the biological significance of only a 10% decrease seems small. However, we must remember that this was found in the CSF of interictal migraineurs, that is, outside of migraine attacks, when they are symptom-free. Patients with paroxysmal brain disorders such as episodic migraine only experience neurological symptoms during attacks. Outside of these, they are clinically completely normal. A biochemical dysregulation that might underlie susceptibility to having recurrent migraine attacks probably fluctuates in severity and is only minor and subtle outside of attacks. Moreover, migraine is probably caused by a local (not generalized) brain abnormality. Focal biochemical abnormalities are expected to be diluted in the CSF. The regional change in L-arginine is likely to be much larger than 10%. Therefore, finding any biochemical abnormality in the CSF of interictal migraineurs that also survives strict statistical Bonferroni correction seems all the more remarkable and should warrant further focused investigation.
Using the less conservative FDR correction, Lphenylalanine levels were also lower in the CSF of both migraine subtypes. Humans cannot synthesize Lphenylalanine de novo. 34 Therefore, the concentrations are fully dependent on dietary intake. Although participants were fasting, long-term dietary differences cannot be excluded, and therefore findings must be interpreted with caution. However, that we found no difference in plasma levels argues against a dietary cause. We observed no difference in L-tyrosine, the hydrolyzation product of L-phenylalanine. 34 FIGURE 4: Results from the global test approach on pathways. The global test is aimed at associations between sets of connected metabolites and disease status and does not test the direction of the association, that is, whether sets of metabolites are up-or downregulated. Therefore, there is no mean difference to plot, and one can only visualize the test statistic and corresponding standard deviation (SD). The full black symbols indicate p < 0.05 after multiple testing correction with Bonferroni, and the half-black symbols indicate p < 0.05 before multiple testing correction. CSF = cerebrospinal fluid; GABAergic = γ-aminobutyric acidergic; MO = migraine without aura; MA = migraine with aura; Migraine = both subtypes combined.
With our strict methodology and inclusion criteria, we could not confirm the previously reported changes in L-glutamic acid (glutamate), L-glutamine, or GABA. 9 However, most of these findings have been made in chronic migraine, and these compounds have never previously been studied in the CSF of interictal episodic migraineurs. A meta-analysis of several small studies with heterogeneity in study quality did show a modest increase in glutamate in the blood. 9 Altogether, the data suggest that glutamate may be elevated during or immediately before or after attacks, but not outside of migraine attacks.
Our study has several strengths. The study groups were significantly larger than those in other studies, 9 and they were matched for important potential confounding factors such as sex, age, and diurnal and seasonal timing of sampling. Control CSF was as normal as possible. Unlike in other studies, 9 the controls were truly healthy. For example, control CSF had not been collected as part of a diagnostic process due to neurological symptoms. In addition to having no personal history of migraine, controls also had no first-degree relatives with migraine. This minimized the genetic risk of later developing migraine. 4,45 We also made an extra effort to obtain strictly interictal values without any potential effect of attacks. Patients were examined at least 3 days after and at least 3 days before an attack. For this reason, patients with highfrequency or chronic migraine were excluded from participation. Finally, whereas other studies measured only one or a few molecules, and only in CSF or in plasma, 9 our study measured multiple amines simultaneously in both CSF and plasma using strict, very high-quality control standards. Only amines for which the relative standard deviation of the quality control samples was <15% were included. Because the concentrations of amines are lower in CSF than in plasma or serum, measuring amines in CSF is also more difficult. Therefore, we used a specifically validated UPLC-MS platform for measuring amines in CSF. This also allowed us to perform additional profile and pathway analyses.
We used a network-based pathway analysis approach to investigate which pathways underlie the altered CSF and plasma amine profiles of migraine patients. We found this approach especially useful for metabolomics data, where coverage of the pathway definitions is low compared to, for example, transcriptomics data. Consequently, it is only possible to make assertions about part of the pathway, that is, not the full KEGG pathway. If studied metabolites only reside on extreme ends of the pathway diagram or are only linked through enzymatic steps that do not occur in humans, it would be incorrect to draw statistical conclusions regarding the full KEGG pathway. Therefore, we used the described filtering step. Combining classic pathway definitions with knowledge about how the measured amines are connected through human metabolism, we were able to define metabolite sets for pathway analysis that are more accurate and relevant from a biochemical point of view.
Our study also has limitations. Ideally, we should have replicated our findings in an independent study population, but collecting a second, sufficiently large, matched sample of CSF and plasma from migraineurs and healthy volunteers is logistically challenging, expensive, and time-consuming. Alternatively, we could have divided the total study population into two portions and used a second portion for replication. However, statistical analysis showed that this would have resulted in a lower power to detect and replicate statistically significant findings. Moreover, to assess the specificity of our findings, we should also have investigated other nonmigraine headaches, for example, tension-type headache. Furthermore, extended pathway coverage by measuring additional amines and other pathways is needed to improve our understanding of migraine biochemistry. However, because we considered reliable quantification more important than a broader detection range, and because we had not previously focused on L-arginine but rather on a broad screening of amines in interictal migraine, we did not measure all amines and metabolites involved in nitric oxide metabolism. A detailed, hypothesis-driven, follow-up study on this should definitely be conducted. Another limitation is the crosssectional nature of our study. Repeated sampling, ideally across the entire migraine attack cycle, would have afforded more detailed insight into migraine biochemistry, but seems ethically impossible. Finally, the effects in our study were modest and overlap; hence, the observed differences are unlikely to be useful as diagnostic tests or offer direct clinical implications. Notwithstanding the aforementioned limitations, we believe that our study is a good illustration of the many methodological aspects that must be considered when studying biochemical changes in the CSF of people with a paroxysmal brain disorder such as migraine. Our study shows how metabolomics could be studied in the CSF of patients with paroxysmal brain disease and the specific methodological problems and interpretation issues that may arise.
In conclusion, extensive analyses of amine profiles and pathways in the CSF and plasma of interictal migraineurs with or without aura show that L-arginine is reduced in the CSF of migraineurs outside of attacks. This suggests a permanent dysfunction of nitric oxide signaling underlying susceptibility to having recurrent migraine attacks. After confirmation and additional analysis of other amines and metabolites involved in nitric oxide metabolism during the different phases of the migraine attack cycle, this may serve as a potential more etiological new target for preventive therapy of migraine attacks.