Biomarkers¶
TL;DR — No blood, saliva, CSF, imaging, electrophysiological or genetic test diagnoses common migraine or selects routine treatment. The most mature biomarkers are mechanistic rather than clinical: CGRP provocation can trigger attacks and enrich erenumab response probability, but an infusion test is burdensome and not sufficiently validated for care (Al-Khazali 2024, PMID 38859744). A GWAS of 102,084 cases identified 123 loci, confirming polygenic neuronal and vascular biology without yielding an individual diagnostic threshold (Hautakangas 2022, PMID 35115687). CSF meta-analysis found candidate differences but few replicated compounds, small studies and substantial heterogeneity (van Dongen 2017, PMID 26888294). Imaging/EEG classifiers often distinguish carefully selected migraine from healthy controls in the development sample; clinically useful validation must be locked, multisite and include secondary headache and other pain/neurological comparators (Chong 2017, PMID 27306407; Gomez-Pilar 2022, PMID 35927625).
Biomarker jobs are different¶
| Biomarker role | Required question | Minimum validation |
|---|---|---|
| Diagnostic | Does this person have migraine? | Representative intended-use population and mimics |
| Subtyping | Aura, chronicity or mechanism subgroup? | Stability and outcome relevance |
| State/attack | Is an attack beginning/active? | Dense longitudinal within-person data |
| Prognostic | Who will chronify/remit? | Prospective external cohort beyond frequency |
| Predictive | Who benefits more from treatment A than B? | Treatment-by-marker interaction in randomized data |
| Safety | Who develops a rare harm? | Large exposed cohorts and prespecified outcome |
| Pharmacodynamic | Did the target/pathway change? | Dose/exposure relationship, not necessarily clinical prediction |
A correlate of response among treated responders is not a predictive biomarker unless it distinguishes relative benefit versus comparator.
Validation ladder¶
- analytical reliability of assay/acquisition;
- association in a discovery sample;
- replication with prespecified direction and endpoint;
- locked threshold/model in an external site;
- clinically relevant comparator conditions;
- incremental value over free clinical data;
- treatment or decision utility;
- cost, feasibility and equity.
Most migraine candidates remain between steps 2 and 4 (Ashina 2021, PMID 33773610).
Genetics¶
The 2022 meta-GWAS included 102,084 cases and 771,257 controls and identified 123 loci, 86 previously unknown. Subtype analysis used 29,679 cases and found both shared and subtype-specific signals, including vascular and neuronal gene/tissue enrichment (Hautakangas 2022, PMID 35115687).
| Genetic use | Current status | Limitation |
|---|---|---|
| Common-migraine diagnosis | Not clinical | Polygenic overlap and incomplete sensitivity/specificity |
| Aura vs without aura | Subtype signals exist | Smaller phenotype sample and self-report heterogeneity |
| Drug selection | No validated pharmacogenomic rule | Few randomized marker interactions |
| Familial hemiplegic migraine | Clinical testing can identify rare variants | Negative test does not exclude phenotype |
| Biological discovery | Strongest current use | Locus-to-gene-to-target mapping uncertain |
In Finnish migraine families, polygenic risk increased progressively from no headache through non-migraine headache, probable migraine, migraine and hemiplegic aura, supporting biological validity of ICHD symptoms while also showing a continuum rather than a clean diagnostic cutoff (Häppölä 2022, PMID 34648375).
Polygenic scores are sensitive to ancestry and phenotype ascertainment. A score trained in European-ancestry cohorts can widen diagnostic inequity when transported without calibration.
CGRP in blood and saliva¶
CGRP is a validated therapeutic target but a difficult assay biomarker: short half-life, pre-analytic degradation, sampling site, attack timing, medication and assay platform change measured concentration.
| Candidate observation | Evidence | Why not yet clinical |
|---|---|---|
| Salivary CGRP rises during some attacks | Small longitudinal studies | Assay and intra-person variability |
| Higher ictal salivary CGRP predicts rizatriptan response | Exploratory study (Cady 2009, PMID 19788468) | Small sample; no locked replication |
| Salivary CGRP differs with onabotulinumtoxinA | Exploratory randomized study (Cady 2014, PMID 24147647) | Surrogate and assay instability |
| Provoked CGRP attack predicts erenumab response | 139-person prospective provocation/treatment study (Al-Khazali 2024, PMID 38859744) | Test burden, imperfect discrimination and no strategy trial |
An earlier 13-person crossover in prior erenumab-trial participants explored provocation-response concordance but was underpowered for a clinical rule (Christensen 2018, PMID 30409109). Reviews list CGRP among the most plausible fluid candidates while emphasizing lack of standardization (Ferreira 2021, PMID 34210227; Yan 2021, PMID 34003145).
CSF and other soluble markers¶
CSF sampling is invasive and therefore studied in small, selected cohorts. The 2017 systematic review/meta-analysis pooled compounds measured in ≥3 studies and found inconsistent candidate differences across CGRP, glutamate, inflammatory and monoamine-related measures (van Dongen 2017, PMID 26888294).
| Candidate family | Biological rationale | Recurrent problem |
|---|---|---|
| Neuropeptides | Trigeminovascular transmission | Sampling site/timing and short half-life |
| Glutamate/GABA metabolites | Excitability/CSD | Non-specific and state-dependent |
| Cytokines | Neuroimmune signaling | Systemic confounding and assay multiplicity |
| Oxidative/metabolic markers | Attack energetics | Diet, fasting, drugs and comorbidity |
| Hormones | Sex/cycle association | Cycle timing and broad physiological variation |
CSF findings may clarify mechanism without becoming a feasible diagnostic test. A biomarker must outperform clinical criteria enough to justify lumbar puncture risk/cost.
Imaging¶
Resting-state fMRI classification using 33 pain-related seed regions distinguished 58 migraine participants from 50 healthy controls in a development study (Chong 2017, PMID 27306407). Deep-learning structural MRI studies can automatically extract features, but cross-validation within a dataset does not test scanner/site transport (Rahman Siddiquee 2023, PMID 36751567).
| Imaging claim | Necessary control |
|---|---|
| Migraine vs healthy | Tension-type, cluster, chronic pain and secondary headache |
| Chronic vs episodic | Frequency-matched medication use, depression and sleep |
| Treatment response | Baseline locked prediction, not post-treatment group difference |
| Attack state | Repeated within-person interictal/ictal scanning |
| Structural lesion | Longitudinal replication and vascular-risk adjustment |
Systematic review of EEG/MRI/PET chronic-versus-episodic studies found high-frequency network differences but methodological heterogeneity and no evidence-based diagnostic rule beyond monthly-day count (Gomez-Pilar 2022, PMID 35927625). Imaging-treatment review found candidate changes after drugs/devices, while small samples and nonstandard tasks prevent clinical response prediction (Messina 2023, PMID 37221469).
An fMRI study showed differing trigeminal/hypothalamic changes after ligand versus receptor antibody and between responders/nonresponders; post-treatment neural differences are mechanistically interesting but not a baseline selector (Basedau 2022, PMID 35604755).
Electrophysiology¶
Visual, auditory, somatosensory and nociceptive evoked potentials often show altered amplitude or habituation. Results depend on migraine phase, aura, stimulus, averaging and preventive medication (Ulutas 2025, PMID 41364425).
The “lack of habituation” model proposes abnormal synaptic plasticity across repeated sensory stimuli. Narrative syntheses of evoked-potential and cortical-excitability studies support phase-dependent abnormal responsivity, but direction and specificity vary across laboratories (Puledda 2023, PMID 37622421; Brighina 2009, PMID 19209386; Brighina 2013, PMID 23602117).
| Modality | Candidate finding | Translation barrier |
|---|---|---|
| Visual evoked potential | Amplitude/habituation difference | Visual attention, phase and analysis |
| TMS | Altered inhibition/excitability | Protocol heterogeneity, seizure precautions |
| EEG spectral/connectivity | Alpha/slowing/connectivity changes | Artifact and state dependence |
| MEG | Visual-cortex dynamics in episodic/chronic migraine (Chen 2013, PMID 23250794) | Cost and small cohorts |
Aura neurophysiology reviews find abnormal alpha, asymmetry and habituation but no single pattern with clinical diagnostic performance (Coppola 2019, PMID 31035929). Chronic-migraine studies similarly report excitability changes entangled with medication overuse (Coppola 2012, PMID 22076672).
Metabolomics, proteomics and multi-omics¶
Omics can discover pathways without prespecifying molecules, at the cost of high dimensionality and batch/confounding risk. Reviews describe lipid, amino-acid, inflammatory and energy-metabolism candidates, but independent replication and assay harmonization are sparse (Chaturvedi 2022, PMID 35796901).
Mouse familial-hemiplegic-migraine work found plasma lipid changes after experimentally induced CSD; it is a pharmacodynamic model signal, not a human diagnostic signature (Loonen 2022, PMID 35323663). Chronic-pain metabolomics review found no robust clinically deployed signature across conditions, reinforcing comparator requirements (Aroke 2020, PMID 32666804).
Response prediction¶
REFORM prospectively collects clinical data, blood, structural/functional MRI and CGRP provocation around erenumab treatment (Karlsson 2023, PMID 37303034). In 623 treated participants, baseline plasma suPAR did not predict ≥50% erenumab response (OR 0.83, 95% CI 0.64–1.07), illustrating the value of publishing a negative prespecified candidate result (Karlsson 2025, PMID 40275185). This multimodal design addresses replication better than retrospective responder mining, but a useful model must be frozen and tested in new sites, then compared with a simple clinical rule and an alternative-treatment arm.
| Model benchmark | Why required |
|---|---|
| Baseline MMD/prior failures | Strong cheap predictors of observed change |
| Early response after one dose | May outperform expensive baseline omics |
| Clinical-only model | Shows incremental biomarker value |
| Alternative treatment arm | Establishes predictive, not merely prognostic, value |
| Decision-curve/net benefit | Tests whether classification improves care |
Circularity and spectrum bias¶
Migraine has no independent gold standard. A classifier trained against ICHD labels can reproduce criteria-associated features without discovering a distinct disease boundary. Healthy-control designs exaggerate separability; real diagnostic populations include probable migraine, tension-type headache, MOH, epilepsy, TIA, IIH and mixed phenotypes.
Biomarker publication also selects positive pipelines. Multiplicity correction, preregistration, negative replication and shared raw data are essential.
Open questions¶
- Can CGRP provocation improve outcomes enough to justify testing before erenumab? (Al-Khazali 2024, PMID 38859744)
- Will any imaging classifier remain calibrated across scanners, sites, ancestry and clinical mimics? (Chong 2017, PMID 27306407)
- Can pharmacogenomics predict relative response to CGRP blockade versus topiramate rather than migraine risk alone? (Hautakangas 2022, PMID 35115687)
- Which fluid marker is stable across attack phase and pre-analytic platforms? (van Dongen 2017, PMID 26888294)
- Does a multimodal biomarker add net benefit over an early empirical treatment trial? (Karlsson 2023, PMID 37303034)
Related pages¶
- Trigeminovascular biology and CGRP — target biology behind CGRP candidates.
- Aura and cortical spreading depolarization — imaging/electrophysiology of aura.
- Preventive treatment — response-selection problem.
- Clinical trials landscape — biomarker-enriched development.
- Classification and diagnosis — clinical gold-standard limitations.
References¶
- Ashina M, et al. Migraine: disease characterisation, biomarkers, and precision medicine. Lancet. 2021. PMID 33773610
- van Dongen RM, et al. Migraine biomarkers in cerebrospinal fluid: systematic review and meta-analysis. Cephalalgia. 2017. PMID 26888294
- Hautakangas H, et al. Genome-wide analysis of 102,084 migraine cases identifies 123 loci. Nat Genet. 2022. PMID 35115687
- Häppölä P, et al. Polygenic risk provides biological validity for ICHD-3 criteria. Cephalalgia. 2022. PMID 34648375
- Ferreira KS, et al. Potential biomarkers in migraine: review and new insights. Expert Rev Neurother. 2021. PMID 34210227
- Yan Z, et al. Biomarkers in migraine. Neurol India. 2021. PMID 34003145
- Cady RK, et al. Salivary CGRP during acute migraine predicts rizatriptan response. Headache. 2009. PMID 19788468
- Cady RK, et al. Salivary CGRP and onabotulinumtoxinA in chronic migraine. Headache. 2014. PMID 24147647
- Christensen CE, et al. CGRP migraine induction in patients from erenumab trials. J Headache Pain. 2018. PMID 30409109
- Al-Khazali HM, et al. CGRP hypersensitivity as predictor of erenumab effectiveness. Cephalalgia. 2024. PMID 38859744
- Chong CD, et al. Migraine classification using resting-state functional connectivity. Cephalalgia. 2017. PMID 27306407
- Rahman Siddiquee MM, et al. Headache classification and biomarker extraction from structural MRI using deep learning. Brain Commun. 2023. PMID 36751567
- Gomez-Pilar J, et al. EEG/MRI/PET biomarkers differentiating chronic and episodic migraine. J Headache Pain. 2022. PMID 35927625
- Messina R, et al. Imaging brain and vascular reactions to headache treatments. J Headache Pain. 2023. PMID 37221469
- Basedau H, et al. CGRP antibodies change brain activity by ligand/receptor target. eLife. 2022. PMID 35604755
- Ulutas M, et al. Evoked-potential studies in migraine: systematic review. Cephalalgia. 2025. PMID 41364425
- Puledda F, et al. Electrophysiological findings and abnormal synaptic plasticity in migraine. Cephalalgia. 2023. PMID 37622421
- Coppola G, et al. Clinical neurophysiology of migraine with aura. J Headache Pain. 2019. PMID 31035929
- Demarquay G, et al. Neurophysiological evaluation of cortical excitability in migraine. Rev Neurol (Paris). 2013. PMID 23602117
- Brighina F, et al. Cortical inhibition and habituation to evoked potentials in migraine. J Headache Pain. 2009. PMID 19209386
- Chen WT, et al. Magnetoencephalography in chronic migraine. Curr Pain Headache Rep. 2013. PMID 23250794
- Coppola G, et al. Cortical excitability in chronic migraine. Curr Pain Headache Rep. 2012. PMID 22076672
- Chaturvedi S, et al. Role of omics in migraine research and management. Mol Neurobiol. 2022. PMID 35796901
- Loonen ICM, et al. Plasma lipid changes after cortical spreading depolarization in FHM mice. Metabolites. 2022. PMID 35323663
- Aroke EN, et al. Metabolomics of chronic pain conditions: systematic review. Biol Res Nurs. 2020. PMID 32666804
- Karlsson WK, et al. REFORM biomarker study methodology and baseline characteristics. J Headache Pain. 2023. PMID 37303034
- Karlsson WK, et al. Plasma suPAR and therapeutic response to erenumab: a REFORM study. J Headache Pain. 2025. PMID 40275185