We recruited study participants for an IRB-approved cross-sectional study at the Stanford University Headache and Facial clinic. All participants gave written informed consent before enrollment and were recruited as non-headache controls (CT) or those with CM based on the International Classification for Headache Disorders, second edition (ICHD-3 beta) criteria (Olesen and Third International Headache Classification Committee of the International Headache, 2011). Exclusion criteria included the use of opioids and other non-migraine-related neurological disorders. Control participants were healthy individuals (>18 years) with no history of headaches. In addition, participants were not fasting when we collected CSF and plasma for both CT and CM.

We obtained CSF by lumbar puncture between 8.00 am and 4.00 pm, centrifuged to remove any cellular debris, and then fractionated into 1 mL aliquots before storage at −80°C.

We collected whole blood by venous puncture into EDTA-treated (lavender tops) anticoagulant tubes. Removal of blood cells/platelet depletion to obtain a plasma layer was achieved by centrifugation for 15 min at 2,000 × g at 4°C. The supernatant fluid (plasma) was immediately transferred in 0.5 ml aliquots into a clean polypropylene tube. The plasma aliquots were stored at −80°C until required for lipid and enzyme analyses.

Cerebrospinal fluid was diluted (2×) with TBS, and protein levels were determined using a Quant-iT fluorescent assay (Invitrogen/Molecular Probes, Eugene, OR, United States) with bovine serum albumin (0–500 ng/ml) as a standard.

Lipids in 25 μl plasma or 1 mL CSF were extracted using a modified Bligh and Dyer method (Bligh and Dyer, 1959). Deuterated fatty acid standards (100 ng each) were added to monitor recovery and GC/MS quantification of fatty acids. We performed derivatization by adding 50 μL each of a solution of 10% N,N-diisopropylethylamine in acetonitrile, and 5% 2,3,4,5,6-pentafluorobenzyl bromide in acetonitrile to the dried lipid extracts, followed by heating to 60°C for 30 min. The solution was dried under a nitrogen stream and then taken up to 100 μL of dodecane for GC/MS analysis.

To 25 μl plasma in 975 μl Tris-buffered saline or to 1 mL CSF, we added 3 drops of 0.9% formic acid to acidify the solutions before adding 100 ng of each deuterated internal standard in the presence of antioxidant (0.25 mg/mL BHT in methanol). Lipids were extracted and derivatized as described above. UFA levels in CSF were quantified using stable isotope dilution GC/MS as previously described (Fonteh et al., 2014).

For EFAs, 100 ng deuterated fatty acid standards were added to 10% of lipids extracted from 25 ul plasma or 1 mL CSF. The mixture was hydrolyzed using 0.5 M HCl in acetonitrile (9:1 v/v) at 100°C for 30 min (Aveldano and Horrocks, 1983). Extracted fatty acids containing UFA and EFA were derivatized as described above and quantified using stable isotope GC/MS.

Glycerophospholipids and SPs in plasma or CSF extracts were separated using a hydrophilic interaction liquid chromatography (HILIC) as previously described (Fonteh et al., 2013). GPs and SPs were eluted from the HILIC column and directly infused to a triple quadrupole mass spectrometer (TSQ Classic, Thermo Fisher Scientific, San Jose, CA, United States). The instrument was operated in the positive mode and scan events for different GPs [phosphatidylcholine (PC), platelet-activating factor (PAF), lyso platelet-activating factor (LPAF), and sphingomyelin (SM) precursor ion scanning m/z 184, lysophosphatidylcholine (LPC) precursor ion of 104, ceramide (Cer), precursor ion of 264, and dihydroceramide (dhCer)] with precursor ion scan of 266). The internal standard (PC 11:0/11:0) peak was acquired using selected reaction monitoring (SRM) of m/z 595 (precursor ion) to 184 (product ion) (Fonteh et al., 2013).

Plasma or CSF equivalent of 10 μg total protein was incubated with a fluorescent lipid cocktail, and PLA₂ activity was kinetically monitored over 1 h (Fonteh et al., 2013). PLA₂ specific activity (RFU/μg/min) was calculated at the linear portion of the activity profile.

For data reduction, we calculated fatty acid indices that represent desaturation, elongation, the number of carbon, unsaturation, and peroxidation. Desaturase and elongase that measure the desaturation and elongation levels of fatty acids, respectively, were determined by calculating the ratios (desaturase or elongase indices) of products to precursors; D9D = C18:1/C18:0, C16:1/C16:0, or C24:1/C24:0, elongase = (C16:0/C14:0 + C18:0/C16:0 + C20:0/C18:0 + C22:0/C20:0 + C24:0/C22:0)/6, D6D = C18:3n-6/C18:2n-6, D5D = C20:4n-6/homo-γ-C20:3n-6, and D4D = C22:6n-3/C22:5n-3. Fatty acid chain length Chain length index (CLI) = Σ FA × carbon number/total FA concentration (Mocking et al., 2012). The number of double bonds per fatty acid = Unsaturation index (UI) = Σ 1 × Monoenoics + 2 × Dienoics + 3…./Total FA concentration (Mocking et al., 2012). Peroxidation susceptibility = Peroxidation index (PI) = Σ 0.025 × Monoenoics + 1 × Dienoics + 3 × trienoics + 4 × tetraenoics + 6 × pentaenoics + 8 × hexaenoic/Total FA concentration (Mocking et al., 2012).

Plasma and CSF UFA and EFA were normalized to levels from 1 mL (ng/mL) and as the percentage of total UFA or total EFA > C14:0. Mann–Whitney U tests were performed to determine significant differences in fatty acid levels in CT compared with CM. One-way ANOVA on ranks (Kruskal–Wallis test) and correction for multiple comparisons with statistical hypothesis testing using Dunn’s method were performed to determine within-group differences of lipid molecular species. We used the Multiple Mann–Whitney test that compares ranks with multiple comparisons adjustment for False Discovery Rate (FDR) using the Two-Stage step-up method of Benjamini, Krieger, Yekutieli. For plasma fatty acids, data normalization was performed using MetaboAnalyst software by first converting Excel sheet data to tab-delimited text (.txt) before importing the text to the MetaboAnalyst Statistical Analysis platform (Chong and Xia, 2020). Fatty acid data normalization and scaling used globalized logarithm transformation (glog) and mean-centering to obtain a Gaussian distribution and compare fatty acid levels over several orders of magnitude in plasma. Hierarchical clustering data presented in the form of a heatmap used Euclidean for distance measure and Ward for the clustering algorithm. All analyses were performed using GraphPad Prism software v 9. (La Jolla, CA, United States) or MetaboAnalyst. Since this is an exploratory pilot study, sample size, study power, and effect size were not computed a priori. Lipid data were considered significant if p < 0.05 after adjustment for multiple comparisons. We present all lipid measures as the mean ± SEM and the 95% CI.

We collected CSF and plasma from CT (n = 10) and CM (n = 15) participants at the peak of a typical headache for this study. Table 1 shows the demographic distribution and clinical classification of each study group. Age, the proportion of females, and BMI were similar for CT and CM patients. The CT group did not take any medication, but CM patients took both migraine prophylactic and rescue drugs; the most commonly used prescription medications in CM were NSAIDs in 67% of patients (Table 1). The duration of CM ranged from 1 to 36 years (mean ± SD = 11.4 ± 9.2 years), and 10 out of 15 reported comorbid migraine conditions in the CM group.

We quantified UFA and EFA levels in plasma and CSF to determine if changes in lipid metabolism are associated with CM pathology. UFA levels were significantly higher in CM plasma than CT (Figure 1A), while plasma EFA levels were not significantly different in CT versus CM. We did not measure any significant difference in UFA and EFA in CSF from CM than CT (Figures 1C,D, respectively).

To determine if specific groups of fatty acids (SAFA, MUFA, and PUFA) changed in CT compared with CM, we quantified unesterified and esterified SAFAs, MUFAs, and PUFAs in plasma and CSF. For Figures 2–4, the fatty acid levels are normalized (globalized log scale (glog) and mean-centered). The unnormalized data are presented as Supplementary Tables for plasma UFAs (Supplementary Table 1), plasma EFAs (Supplementary Table 2), CSF UFAs (Supplementary Table 3), and CSF EFAs (Supplementary Table 4).

Levels of 3 unesterified even chain saturated fatty acids (eSAFA) (C14:0, C16:0, and C18:0), and the sum of unesterified eSAFA was higher in CM than in CT plasma (Figure 2A). Figure 2B also shows that unesterified levels of two odd chain saturated fatty acids (oSAFAs include C15:0, C17:0) and the sum of oSAFA are higher in CM plasma than in CT. We next examined EFA levels and found an increase in C14:0, C22:0, and C24:0 (Figure 2C), and C15:0, C17:0, and oSAFA in the plasma of CM compared with CT (Figure 2D). These data suggest that levels of some EFAs may change in CM even when total EFA levels do not significantly change.

Although slightly higher, the levels of unesterified and esterified SAFAs are not significantly altered in CM compared with CT (Supplementary Table 3).

Five unesterified MUFAs (C14:1n-5, C16:1n-7, trans C16:1 (C16:1Tn-7), C18:1n-9, C20:1n-9, C24:1n-9, Figure 3A) and C17:1n-9, C19:1n-9, and the sum of oMUFA levels are higher in CM than in CT (Figure 3B). Even though plasma EFAs did not change in CM, we found a significant decrease in C16:1Tn-7 and an increase in C24:1n-9 in CM compared with CT (Figure 3C). There was no significant change in esterified oMUFA levels (Figure 3D).

Of all the UFAs detected in CSF (Supplementary Table 3), only the levels of C16:1n-7 was significantly higher in CM (1.8 ± 0.5, mean ± SEM, ng/mL, 95% CI = 0.7–3.0) than in CT (1.0 ± 0.1 mean ± SEM, ng/mL, 95% CI = 0.7–1.2, ROC AUC = 0.75 ± 0.1, p = 0.0375). Esterified C16:1 levels were also higher in CSF of CM (449.8 ± 92.0 mean ± SEM, ng/mL, 95% CI = 252.4–647.2) than in CT (225.6 ± 22.7 mean ± SEM, ng/mL, 95% CI = 174.3–276.9, ROC AUC = 0.84 ± 0.08, p = 0.0037) while the levels of other EFAs were not altered (Supplementary Table 4).

Plasma unesterified levels of 4n-3 PUFAs (C18:3n-3, C20:3n-3, C22:5n-3, C22:6n-3) and the sum of n-3 PUFAs were higher in CM than CT (4A). Plasma unesterified levels of 3 n-6 PUFAs (C20:2n-6, dihomo-g-C20:3n-6, and C22:4n-6) were higher in CM than CT (4B). There were no significant changes in the levels of esterified n-3 PUFAs (Figure 4C) and esterified n-6 PUFAs (Figure 4D) between CM and CT.

None of the unesterified n-3 PUFAs detected in CSF were different between groups, although their levels were generally lower in CM (Supplementary Table 3), and only the percent of C18:3n-3 is higher in CM. CSF esterified n-6 PUFA fatty acids were similar in CT and CT (Supplementary Table 3).

To determine if there is higher hydrolysis or an abnormal buildup of fatty acids compared with EFA storage sites in plasma or CSF, we calculated the ratio of unesterified to the EFA. Plasma UFA to EFA ratio in CT (4.0 ± 1.0 mean ± SEM, 95% CI = 2.9–5.2) is significantly lower than in CM (6.3 ± 2.5 mean ± SEM, 95% CI = 4.9–7.8, p = 0.0475). CSF UFA to EFA in CT (0.8 ± 0.2 mean ± SEM, 95% CI = 0.7–0.9) is significantly lower than plasma (p = 0.0004) but does not significantly differ from CM (0.7 ± 0.3 mean ± SEM, 95% CI = 0.6–0.9, p = 0.3472). These data show a higher turnover of plasma fatty acids than CSF and an increase in their hydrolysis in CM plasma but not in CSF.

To determine fatty acid metabolism in plasma or CSF, we calculated total desaturase indices for CT and CM in plasma and CSF. We also examined the elongase indices to determine if there is a change in the fatty acid chain length.

Of the four desaturase indices (D4D, D5D, D6D, and D9D) measured in plasma and CSF, only the D5D (AA/DGLA) ratio changed in CM. The plasma unesterified D5D index is lower in CM than CT (Figure 5A). A decreasing trend for plasma esterified AA/DGLA ratio (Figure 5B) is not significant, probably because of 3 outliers in the CM samples. The unesterified CSF D5D is not significantly different (Figure 5C). In contrast, CSF esterified D5D index is significantly lower in CM than in CT (Figure 5D).

We calculated the elongase index in plasma and found no difference between groups for UFAs (Figure 5E). However, there is a significant decrease in plasma esterified elongase index in CM compared to CT (Figure 5F). On the other hand, there is no difference in unesterified (Figure 5G) and esterified (Figure 5H) elongase indices in the CSF.

The metabolism of GPs and SPs generate signaling lipids that may impact CM pathology. Therefore, we quantified the principal GP and SP classes in plasma and CSF from CT and CM participants.

We quantified choline-containing lipid classes in plasma and determined the difference in CT compared with CM participants. The proportion of PC, LPC, LPAF, and PAF in plasma was not significantly different in CT compared to CM (Supplementary Table 5A). We quantified SM, Cer, and dhCer in plasma and found a significant decrease in the proportion of Cer in CM (3.5 ± 1.2, mean ± SEM,%, 95% CI = 2.8–4.1) than CT (7.4 ± 4.5, mean ± SEM,%, 95% CI = 4.2–10.6). The AUC for Cer is 0.85 ± 0.09, 95% CI = 0.67–1.0, p = 0.0039, Figure 6A). Analyses of the two main clusters (#1 and #2, Figure 6B) show that plasma GPs and SPs are not good classifiers of CT and CM (Fisher’s exact test p > 0.9999).

We observed heterogeneity in GP levels in CSF for CT compared to CM patients. In general, PC and LPC levels were similar in CM than in CT (Supplementary Table 5B). However, the proportions of PAF compared to all lipids were higher in CT (0.3 ± 0.2, mean ± SEM%, 95% CI = 0.2–0.5) than CM (0.15 ± 0.023, mean ± SEM%, 95% CI = 0.1–0.2). The AUC for PAF in CSF is 0.89 ± 0.06 (mean ± SEM,%, 95% CI = 0.75–1.0, p = 0.0013), (Figure 6C). The SP levels did not attain statistical significance when calculated as ng/mL or expressed as a percent of GP/SP levels in CSF (Supplementary Table 5B). Cer levels were also higher in CM, but dhCer levels were similar for both groups, resulting in lower dhCer in CM (Supplementary Table 5B). Cluster analyses (Figure 6D) show that CSF GPs and SPs are good CT and CM identifiers (Fisher’s exact test p = 0.0154).

To determine if there is a difference in GP and SP metabolism in CSF compared with plasma, we calculated the ratio of the proportion of lipid classes. The PAF CSF/plasma ratio was significantly lower in CM (0.3 ± 0.2, 95% CI = 0.2–0.5) than in CT (0.7 ± 0.2, 95% CI = 0.5–0.8) and the Cer ratio is significantly higher in CM (1.8 ± 0.8, 95% CI = 1.3–2.3) than CT (1.0 ± 0.4, 95% CI = 0.7–1.2). The AUC for PAF and Cer CSF/plasma ratios are 0.87 ± 0.07 (95% CI = 0.7–1.0, p = 0.0019), and 0.82 ± 0.09 (95% CI = 0.7–1.0, p = 0.0076), respectively (Figure 6E). The ratios of the other GP and SP classes were not different (Supplementary Table 5C). However, cluster analyses show that the ratios of GPs and SPs metabolism in CSF to plasma are excellent classifiers of CT and CM (Fisher’s exact test p = 0.0002) (Figure 6F). These data show differences in PAF and Cer formation in CSF and plasma for CT versus CM but similarities in the proportion of the other lipid classes.

To determine if there was an increase in GP hydrolysis in CM, we measured PLA₂ activity in plasma and CSF from CT and CM participants. CSF calcium-dependent PLA₂ activity did not significantly differ in CT (7.2 ± 0.5 RFU/min/μg protein) and CM (7.9 ± 2.3 RUF/min/μg protein, p = 0.6047) CSF. Similarly, plasma PLA₂ activity did not differ in CT (1.3 ± 0.3 RFU/min/μg) and CM (1.2 ± 0.3 RFU/min/μg, p = 0.4863).

Total plasma UFAs and the levels of several unesterified SAFA and esterified SAFA species are higher in CM (Figures 1, 2). We summarize the implication of these changes in Table 2. The physiological implications include energy production (Lindsay, 1975; Jumpertz et al., 2012; Nakamura et al., 2014; Alcock and Lin, 2015; Cucchi et al., 2019), the trigger of the inflammatory response through via TLR4 signaling pathway associated with pain pathogenesis (Zhang and Ma, 2017), worsening pain (Sekar et al., 2020), and a link to insulin resistance that is proposed to be a part of migraine pathophysiology (Rao and Pearce, 1971; Perciaccante and Perciaccante, 2008; Bhoi et al., 2012; Ozcan and Ozmen, 2019). The levels and de novo synthesis of SAFAs are linked with CNS and peripheral nervous system myelination (Salles et al., 2002; Montani et al., 2018; Dimas et al., 2019), and fatty acid elongation is increased during active myelination in rodents (Morita et al., 2015). The changes in SAFA suggest a mirage of biochemical abnormalities in overall metabolism in CM participants.

Plasma levels of MUFAs were higher in CSF from CM subjects (Figure 3) while esterified C16:1T is decreased and esterified C24:1 is increased in CM (Figure 3). Palmitoleic acid (C16:1) is a proposed novel lipokine (Frigolet and Gutierrez-Aguilar, 2017) that decreases inflammation and is involved in glucose homeostasis and insulin resistance (Nunes and Rafacho, 2017; de Souza et al., 2018). C16:1 is derived from dietary sources or lipogenesis when palmitic acid formed by fatty acid synthase is converted to C16:1 by delta-9 desaturase or stearoyl-CoA desaturase (D9D or SCD1) activity. There are disagreements on the function of C16:1, but its roles on glucose homeostasis and insulin resistance (Nunes and Rafacho, 2017) are not disputed. Detrimental and beneficial effects of C16:1 have been reported in animal and human studies, likely related to dietary sources versus endogenous synthesis by the liver or adipose tissue (Mozaffarian et al.,2010a,b). In addition to insulin resistance and type-2 diabetes, C16:1 is implicated in mitochondrial permeability (Kurotani et al., 2012; Oyanagi et al., 2015), higher plasma levels are biomarkers of triglyceridemia and abdominal adiposity (Paillard et al., 2008), and increased risk of heart failure (Djousse et al., 2012). Higher C16:1 levels are also linked to carbohydrate intake and alcohol usage and are associated with metabolic risk factors. In contrast, consumption of purified C16:1 reduces C-reactive protein and improves metabolic risk factors (Bernstein et al., 2014).

In addition to palmitoleic acid, C18:1 (oleic acid), C20:1, and C24:1 are other MUFA species that increase in CM plasma. C18:1 is a neurotrophic factor (Velasco et al., 2003) that promotes brain development and neuronal growth (Polo-Hernandez et al., 2014), protects against oxidative stress (Guzman et al., 2016), and reversibly opens the blood–brain barrier (BBB) in rodents (Sztriha and Betz, 1991; Han et al., 2013). In addition, C20:1 and C24:1 are the primary components of sphingolipids involved in myelination. oMUFA are involved in vitamin B12 deficiency, resulting in an increase in odd-chain MUFAs in GPs isolated from myelin (Ramsey et al., 1977). Thus, higher oMUFAs in CM may suggest vitamin imbalance (Kishimoto et al., 1973).

Another MUFA that increased in CM is nervonic acid (C24:1). Nervonic acid is a very long chain (VLC) fatty acid and a significant brain myelin fraction component, accounting for 40% of total SM fatty acids (O’Brien and Rouser, 1964). C24:1 is formed from C18:1 by three successive elongation reactions, and lower C24:1 levels are reported in demyelinating diseases (Sargent et al., 1994). While the mitochondria oxidize most medium and long-chain fatty acids, peroxisomes are the oxidation site for VLC fatty acids (Sandhir et al., 1998). C24:1 levels are regulated by biosynthesis/dietary uptake and by β-oxidation, and the higher C24:1 in CM suggests altered VLC fatty acid metabolism in plasma. Plasma esterified elongase ratios were lower in CM, suggesting that the higher plasma C24:1 may be derived from cytosolic lipolysis. These data justify a further examination of these enzymes for a role in CM pathology.

Another crucial fatty acid change in plasma is unesterified homo-γ-C20:3n-6 (DGLA). DGLA is formed by the elongation of GLA and is the intermediate in arachidonic acid (C20:4n-6, AA) formation. The D5D index (C20:4n-6/homo-γ-C20:3n-6) that measures the conversion of homo-γ-C20:3n-6 to C20:4n-6 is lower in CM compared to CT for unesterified plasma and esterified CSF. D5D is a rate-limiting enzyme in long-chain PUFA synthesis, and polymorphism has been reported in different races (Al-Hilal et al., 2013; Abdelmagid et al., 2015). The D5D decrease in plasma unesterified and CSF esterified D5D ratio has not been previously reported in CM. A higher level of homo-γ-C20:3n-6 is associated with insulin resistance, diabetes, obesity, contraceptive hormones usage, thyroid hormone status, and eating disorders (Araya et al., 2010; Swenne and Vessby, 2013; Pickens et al., 2016; Matsuda et al., 2017). DGLA also forms series-1 prostaglandins that are less potent than series-2 prostaglandins formed from arachidonic acid. PGE₁ attenuates leukotriene B₄ formation in human neutrophils (Chilton-Lopez et al., 1996; Kakutani et al., 2010), suggesting that higher DGLA in CM may have benefits. Our study justifies the design of intervention studies that alter the n-3 to n-6 ratios in headache and pain syndromes (Ramsden et al., 2010, 2011).

Given similarities in BMI for our CT and CM population (Table 1), it is likely that increased lipolysis (SAFAs, MUFAs, PUFAs) and enhanced synthesis (SAFAs) occurs in CM. Genome-wide association studies (GWAS) of subjects of European ancestry show that FADS1 and FADS2 polymorphism results in higher levels of MUFAs (C16:1, C18:1) and lower SAFA (C18:0) (Wu et al., 2013; Guan et al., 2014). These levels of MUFAs mirror the measurements we see in our study but not the changes in MUFA. These data suggest that different mechanisms, including genetic heritability and epigenetics, may alter our CM cohort’s fatty acid profiles.

We found lower levels of Cer in CM plasma compared to controls (Figure 6A and Supplementary Table 5). Cer is important in energy metabolism, metabolic syndrome, body weight regulation (Yang et al., 2009) and ameliorates energy homeostasis, lipid profile, and antioxidant systems (Zanieri et al., 2020). Therefore, lower Cer levels may indicate an abnormality in these vital physiologic functions in CM participants.

Platelet-activating factor metabolism involves PLA₂ activity, acetyltransferase, and hydrolysis by PAFA. Both PC hydrolysis and lipid remodeling are known to regulate PAF formation (Chilton and Murphy, 1986). Lower remodeling in CM is consistent with bigger lipid pools and diminished capacity to form LPAF and PAF. Lower PAF levels may also be due to increased lipoprotein-associated phospholipase A₂ (Lp-PLA₂), an enzyme shown to increase in plasma of migraineurs implicated in increased cardiovascular risks and brain development (Clark, 2015). Future studies are necessary to examine the remodeling and modulation of lp-PLA₂ pathways in CM compared with CT. Venous blood measurements have shown increased PAF levels during migraine without aura, and migraineurs have heightened sensitivity to PAF (Joseph et al., 1988; Sarchielli et al., 2004). However, LPC and PAF effects on the brain have been studied primarily on rodents. LPC is known to disrupt the BBB, will disturb brain lipid homeostasis, and PAF transiently increases BBB permeability and is a neuromodulator linked to brain injury (Yue and Feuerstein, 1994; Liu et al., 2001). PAF promotes neuroplasticity, so lower PAF levels in CM may point to habituation that mitigates the long-term cellular injury linked with these inflammatory mediators.

Our studies show changes in lipid pathways and identify potential enzyme mechanisms accounting for these changes in CM. We show different plasma and CSF lipids changes, suggesting a manifestation of peripheral and central migraine pathology abnormalities in CM. We summarize the significant lipid changes and their likely significance in CM pathophysiology (Table 2). Our study’s major findings in lipid changes are in plasma lipolysis (SAFAs, MUFAs, PUFAs), lower Cer in plasma, and decreased CSF platelet-activating factor levels. An underlying outcome in the metabolic changes in CM is an imbalance with energy regulation, inflammatory pathways, and insulin resistance (Table 2). With a similar BMI for CT and CM subjects, metabolism and genetics rather than dietary differences likely account for these changes. These new findings can inform clinical care through dietary interventions and guide long-term pharmaceutical research. Since CM participants have abundant fatty acid levels, dietary intervention attempts at normalizing lipids should focus on D5D and elongase in CM rather than random dietary supplementation. For dietary, lifestyle modifications, or enzyme inhibitor studies, plasma UFA measures can be used to monitor efficiency. However, it should be noted that genetic differences and epigenetics may require personalized treatment strategies for people with metabolic diseases based on lipid profiles (Sergeant et al., 2016).

As with any human studies, diet, medication usage, and genetic heterogeneity are extremely difficult to control, and our study is underpowered by the number of plasma and CSF samples used to detect several metabolic changes. Participants did not fast before sample collection, and we did not obtain dietary history, and hormonal influence was not determined. More CM subjects are on NSAIDs, antioxidants and have more comorbid conditions, but most of these are not known to influence CSF lipid composition. As is with migraine prevalence that favors women, more women participated in our study (Table 1). Even with these limitations, CSF lipid differences between CM and CT that we identify underscore the central role of altered lipid metabolism in CM disorders.