Mechanism and Pathophysiology

Source: Chapter 1, section 3. Section numbering retained from the source chapter.

From what the disease is to why it happens: the mechanistic picture, from the refuted vascular theories to the current network models, the contested signalling molecules, the genetics, and the clockwork.

3. Mechanism and Pathophysiology

3.1 The hypothalamus enters the picture

The single most influential modern paper in CH pathophysiology is:

May A, Bahra A, Büchel C, Frackowiak RSJ, Goadsby PJ. “Hypothalamic activation in cluster headache attacks.” Lancet. 1998 Jul 25;352(9124):275–278. doi: 10.1016/S0140-6736(98)02470-2. PMID 9690407. peer-reviewed

A correction worth flagging explicitly: this landmark PET study is frequently misattributed to Nature Medicine in popular and even some semi-technical accounts. It was published in The Lancet. A separate, genuinely distinct paper — May A, et al. Nature Medicine 1999, doi: 10.1038/10561, a structural voxel-based morphometry (VBM) follow-up — is often conflated with the 1998 Lancet paper but should be cited separately, since it asks a different question (structural volume, not functional activation).

The 1998 finding — that the posterior/inferior hypothalamic grey matter activates specifically during CH attacks, a finding not seen in other headache types — is what shifted the field from believing CH originated in blood vessels (the older vascular theories, §3.6) to believing it originated in the brain itself, and specifically in a structure that governs circadian rhythm, among other things.

3.2 An important update: the “pure generator” model is now outdated

Even the hypothalamic model itself has moved on since 1998. Goadsby’s own more recent position, and a 2024 review, describe a more distributed picture:

Coppola G, et al. Cephalalgia. 2024. doi: 10.1177/03331024231209317. peer-reviewed — proposes a “central permissive” multi-network model, in which the hypothalamus functions as a “crossroads” or coordinating hub rather than the sole attack generator.

A striking piece of supporting evidence: CH attacks have been documented to continue even after surgical sectioning of the trigeminal nerve — a finding that is hard to reconcile with any model where a single anatomical structure alone is both necessary and sufficient to generate the attack peer-reviewed. VBM findings on hypothalamic volume are also contested rather than settled: an early study found increased hypothalamic grey-matter volume in CH patients, but this was not confirmed by Arkink et al. (2017) or by Coppola et al. (2024). Newer network-topology approaches (Lee et al. 2022, Korean cohort) find altered brain-network organisation rather than a simple volume difference peer-reviewed.

3.3 The trigeminal-autonomic reflex: why pain and tears happen together

The defining clinical puzzle of CH — why does severe unilateral pain co-occur, tightly time-locked, with tearing, a runny nose, and a drooping eyelid on the same side? — has a well-documented anatomical answer, independently confirmed by Western, Japanese, and German sources:

• Pain pathway: trigeminal ganglion → releases CGRP, substance P, and neurokinin A onto cerebral vessels and the dura → signal passes to the trigeminocervical complex → thalamus → cortex, producing the severe unilateral pain.
• Parasympathetic pathway: superior salivatory nucleus → facial nerve → sphenopalatine ganglion → lacrimal gland and nasal mucosa, producing lacrimation, conjunctival injection, rhinorrhoea, and nasal congestion.
• Sympathetic deficit: a concurrent sympathetic deficiency via the internal carotid artery/cavernous sinus region produces an incomplete Horner’s syndrome — ptosis and miosis, but typically without the anhidrosis seen in full Horner’s syndrome.
• A feedback loop runs from the trigeminocervical complex/thalamus back to the hypothalamus, forming a reflex arc that explains why the pain and the autonomic symptoms are so precisely time-locked to each other.

Primary source: Wei DY, Ong JJY, Goadsby PJ, PMC5909131 peer-reviewed, independently corroborated by Japanese Headache Society guideline material and a German (LMU Munich) dissertation.

Diagram description (attack mechanism pathway — described here in place of a rendered figure, per the master brief’s instruction that a described figure is acceptable):

                    Suprachiasmatic nucleus (circadian input)
                                    │
                                    ▼
                 ┌──────  HYPOTHALAMUS (posterior/inferior)  ──────┐
                 │        "crossroads" — coordinating hub,          │
                 │   NOT sole generator (superseded single-site      │
                 │              model, see [§3.2](/mechanism/#32-an-important-update-the-pure-generator-model-is-now-outdated))                    │
                 └──────────────────┬───────────────────────────────┘
                                    │  (reflex feedback loop)
              ┌─────────────────────┴─────────────────────┐
              ▼                                            ▼
   TRIGEMINOVASCULAR PATHWAY                    PARASYMPATHETIC PATHWAY
   Trigeminal ganglion                           Superior salivatory nucleus
   → releases CGRP*, substance P,                → facial nerve
     neurokinin A onto dura/vessels              → sphenopalatine ganglion
   → trigeminocervical complex                   → lacrimal gland / nasal mucosa
   → thalamus → cortex                           → lacrimation, conjunctival
   = SEVERE UNILATERAL PAIN                         injection, rhinorrhoea,
                                                     nasal congestion

                                            ┌───────────────────────────┐
                                            ▼
                              SYMPATHETIC DEFICIT (internal carotid /
                              cavernous sinus)
                              = ptosis + miosis (incomplete Horner's)

   Side annotations:
   * CGRP direction of change is CONTESTED — see [§3.4](/mechanism/#34-cgrp-a-genuinely-contested-signalling-molecule)
   * PACAP/VIP also released at these sites — implicated, less studied than CGRP
   * Orexin: proposed modulator of the hypothalamic hub — evidence CONTESTED, see [§3.5](/mechanism/#35-orexinhypocretin-an-interesting-but-unreplicated-hypothesis)

(Structure mirrors the figure in Wei, Ong & Goadsby 2018, PMC5909131, using their anatomical abbreviations IML, SCG, ICA, SSN, SPG, TCC, HT, SN.)

3.4 CGRP: a genuinely contested signalling molecule

The “classic” view, going back three decades, is that CGRP is elevated during CH attacks:

Goadsby PJ, Edvinsson L. Brain. 1994;117:427–434. peer-reviewed

But a recent, well-powered Danish study directly contradicts the direction of this finding:

Petersen AS, et al. (Danish Headache Center). Cephalalgia. 2024;44(3). doi: 10.1177/03331024231223970. peer-reviewed — n=301, found CH patients (across subtypes) had significantly lower plasma CGRP than controls — the opposite direction to the classic literature.

This is presented here as a genuine, unresolved contradiction rather than smoothed over. It has real clinical consequences: CGRP-targeting drugs have had mixed results in CH trials. Galcanezumab is FDA-approved for episodic CH (NEJM, 2019) but failed its primary endpoint in chronic CH (per EMA documentation) — an important negative result that is easy to miss if one only remembers the positive episodic-CH approval. Eptinezumab (the ALLEVIATE trial) also missed its primary endpoint in episodic CH. The clean, simple “CGRP is the culprit, block it and CH improves” story that holds reasonably well for migraine does not transfer cleanly to CH peer-reviewed.

3.5 Orexin/hypocretin: an interesting but unreplicated hypothesis

Orexin (also called hypocretin), a hypothalamic neuropeptide involved in wakefulness regulation, has been proposed as a modulator relevant to CH’s circadian pattern — but the supporting evidence is low-to-moderate confidence and directly contradictory between the two key CSF studies:

• Cevoli et al. (2011) found normal orexin/hypocretin CSF levels in CH patients.
• Barloese et al. (2015, Danish Headache Center) found significantly reduced levels.

peer-reviewed — this hypothesis should be described as speculative rather than established, pending further replication.

3.6 Genetics: heritable, but not simply so

• Family history: Waung, Taylor, Qualmann, Burish (2020, JAMA Neurol, PMID 32310255) peer-reviewed systematic review of 22 cohorts found a median family-history rate of 8.2% (range 0–22%); 69% of pedigrees were consistent with autosomal dominant inheritance; sex ratio among familial cases was 1.39 M:F (notably less male-skewed than sporadic CH — see also the Italian familial data in §4.6).
• Twin studies: Ekbom K, et al. Neurology. 2006;67(5):798–803. doi: 10.1212/01.wnl.0000233786.72356.3e. peer-reviewed — Swedish national twin registry, 37 pairs, concordance only 5.4% (2/37, both monozygotic). This low concordance rate argues against a simple, deterministic single-gene model of inheritance, even though family clustering is real.
• Russell MB, et al. Headache. 1996;36(10):608–612. PMID 8990601. peer-reviewed — first-degree relatives of CH patients have a 14-fold increased risk of CH themselves; segregation analysis is consistent with an autosomal dominant gene of variable penetrance (0.30–0.34 in males, 0.17–0.21 in females), estimated to be present in 3–4% of affected males and 7–10% of affected females.

The HCRTR2 candidate-gene saga — a clear case study in non-replication: Early positive associations between CH and the HCRTR2 (orexin receptor 2) gene were reported by Rainero (2004), Baumber (2006), Schürks (2006, German cohort), and Rainero again (2008). A key negative replication came from:

Weller CM, et al. Cephalalgia. 2015;35(9):741–747. PMID 25398231. peer-reviewed — Leiden LUCA cohort, n=575/874, found no significant association alone (OR=0.91, p=0.319). A pooled meta-analysis across 6 populations did show a significant association (OR=0.69, p=0.006), but this weakened substantially (to OR=0.80) once one outlier study was excluded.

The most authoritative and recent statement on this comes from the largest genome-wide association study to date (below): none of the earlier HCRTR2 candidate-gene associations were replicated at genome-wide significance. This is a useful, concrete illustration of how a plausible, widely-cited early finding can fail to hold up under larger and more rigorous later testing.

The definitive current genetic picture:

Winsvold BS, et al. “Cluster Headache Genomewide Association Study and Meta-Analysis Identifies Eight Loci and Implicates Smoking as Causal Risk Factor.” Ann Neurol. 2023;94:713–726. doi: 10.1002/ana.26743. PMID 37486023. peer-reviewed

This is the largest GWAS conducted to date: 10 European cohorts plus 1 East Asian cohort; European meta-analysis of 4,043 cases vs. 21,729 controls; trans-ancestry meta-analysis of 4,777 cases vs. 31,575 controls.

• SNP-based heritability: 14.5% (SE 1.74%).
• Eight loci identified: seven in the European analysis (DUSP10, MERTK, FTCDNL1/SATB2, FHL5, WNT2, PLCE1, and the novel LRP1), plus one East-Asian-specific locus (CAPN2, also independently found by a Taiwanese Han Chinese GWAS — Chen SP, et al., 2022, J Headache Pain 23:147).
• Genetic correlation with smoking, other pain disorders, and neuropsychiatric diagnoses. Critically, Mendelian randomisation analysis found evidence for a causal (not merely correlational) effect of smoking on CH risk — discussed further, along with the clinical paradox this creates, in Section 4.4.
• Three of the eight loci are shared with migraine, offering a partial genetic-overlap explanation for the two conditions’ occasional clinical overlap. (Editor’s note: per-locus lead SNPs, odds ratios and p-values for these eight loci — and corrections to earlier locus lists in circulation — are tabulated in Part VII, §5.2.1.)

Other candidate genes examined over the years but not consistently verified include ADH4, GNB3, CACNA1A (formally excluded), NOS, MTHFR, PER3, CLOCK, and CRY1, plus a single-patient Japanese mitochondrial finding (MT-TL1, 1994) that was never replicated.

3.7 Circadian and circannual rhythmicity

CH’s clockwork timing is one of its most distinctive — and mechanistically informative — features:

• A meta-analysis of 16 studies / 4,953 participants (Benkli et al., summarised in Neurology 2023, PMC10259280) peer-reviewed found 70.5% of patients show a circadian attack pattern, peaking between 21:00 and 03:00, with 2 a.m. the single most common hour (6.5% of 8,856 recorded attack times).
• An older analysis (Manzoni, cited in a Canadian Journal of Neurological Sciences review) found a trimodal pattern (peaks around 01:00–02:00, 13:00–15:00, and ~21:00). The midday peak is explicitly flagged by the reviewing source’s own caveat as potentially confounded by the Italian custom of a midday lunch/rest break — a good concrete example of a real, specific data point that should not be over-generalised as pure neurobiology divorced from cultural or occupational schedule [PEER-REVIEWED, with an explicit cultural-confound caveat].
• Nocturnal attacks are linked to REM sleep: in one recorded series, roughly 60% of nocturnal attacks followed REM sleep, despite REM comprising only about 20% of total sleep time — a striking overrepresentation. (Editor’s note: the REM association is contested in more recent work — see the fuller sleep-architecture discussion in Part IV, §4.)
• Melatonin: a meta-analysis (Liampas et al. 2020, Acta Neurol Scand, PMID 32677039) peer-reviewed found significantly reduced nocturnal melatonin during active bouts (mean difference −29.89 pg/mL), but not significantly reduced serum melatonin during remission — though the urinary metabolite aMT6s is reduced even during remission. This is a nuanced, partially-state-dependent finding, not a simple “CH patients always have low melatonin.”
• Circannual pattern — and a genuine correction to common assumptions: bouts cluster in spring and autumn, not, as commonly assumed, around the solstices. Riederer F & Schankin C (2024, Headache, doi: 10.1111/head.14830) peer-reviewed found circannual clustering correlates specifically with the square of the velocity of daylight change (Spearman R=0.762, p=0.006) — meaning attacks cluster around the equinoxes, when day length is changing fastest, not around the solstices, when day length is nearly static. This is worth flagging explicitly since “solstices” is the more intuitive (and incorrect) assumption.

3.8 Neuroimaging and historical/competing theories

The refuted vascular theories: Horton’s original histamine/vascular theory (§1.4) was historically influential but is now understood as, at best, a partial or secondary phenomenon rather than the primary driver. A separate vascular-adjacent theory — the cavernous sinus hypothesis — went through a full cycle of proposal and refutation that is instructive about how the field progressed toward the current central/hypothalamic model:

• Gawel et al. (1990) — SPECT gallium-67 imaging showed increased cavernous sinus uptake, initially seeming to support a cavernous-sinus origin.
• Sianard-Gainko (1994) — showed the same imaging finding is non-specific, occurring in migraine too.
• Schuh-Hofer (2006) — SPECT findings did not support the hypothesis.
• High-resolution MRI found no cavernous sinus dimension differences between CH patients and controls.

[PEER-REVIEWED, historical refutation documented in Silvestro et al. 2022, PMC9314615] — this refutation was a key transition point that helped redirect the field toward the central/hypothalamic model that dominates today (§3.1–3.2).

Non-English contributions to mechanism research (tallied explicitly per the task brief): Japanese sources (Japanese Headache Society guideline material on reflex anatomy and melatonin/cortisol chronobiology); German sources (an LMU Munich dissertation confirming the parasympathetic/sympathetic pathway anatomy, and Schürks 2006 on HCRTR2); Danish sources (the Danish Headache Center’s substantial body of work on CGRP, PACAP/VIP, and hypocretin — Snoer, Petersen, Pellesi, Barloese); Italian sources (Leone 1998, Cephalalgia, on melatonin excretion); Chinese sources (Qiu 2013, PLoS One, fMRI hypothalamic connectivity; Fan 2018; Chen 2022 Taiwan GWAS); Korean sources (Lee 2022, J Neurol; Kim 2025, Sci Rep, on structural covariance and white-matter markers). No dedicated Italian-language pathophysiology-specific primary research beyond the single Leone 1998 melatonin study was located — flagged as a gap.