Abstract
Background
Migraine attacks are believed to unfold through discrete but interrelated phases, among which the premonitory phase has garnered increasing attention. This early phase, occurring hours to days before pain onset, might reflect neurobiological processes that promote subsequent headache generation. Neuroimaging of experimentally induced attacks enables controlled investigation of the brain mechanisms underlying these early symptoms. This systematic review synthesizes and critically appraises current neuroimaging evidence on the experimentally induced premonitory phase in individuals with migraine.
Methods
A systematic literature search of MEDLINE and Embase was performed from database inception through June 1, 2025, using predefined terms related to migraine and premonitory symptoms. Studies were eligible if they reported original neuroimaging findings on experimentally induced migraine attacks and specifically addressed the premonitory phase. Two independent reviewers screened all titles, abstracts, and full-text articles and extracted relevant data. Due to methodological heterogeneity in study design, a narrative synthesis was applied to summarize findings and assess the consistency of reported brain activity patterns across included studies.
Results
Seven studies met the inclusion criteria, all of which used glyceryl trinitrate to induce migraine attacks and applied either positron emission tomography or magnetic resonance imaging to assess brain changes. Across studies, reported alterations in cerebral perfusion and functional connectivity—used as surrogates for neural activity—during the premonitory phase were inconsistent and lacked replication. While some investigations suggested involvement of the thalamus and pons, these findings were derived exclusively from exploratory or uncorrected analyses. Hypothalamic involvement—a hypothesized migraine generator—was only sporadically observed, and reproducibility not investigated. The inconsistencies across studies likely reflect small sample sizes (ranging from 5 to 21 participants), suboptimal definitions of the premonitory phase, absence of appropriate controls, and unconventional use of neuroimaging statistics.
Conclusions
Neuroimaging studies investigating the experimentally induced premonitory phase of migraine have largely produced inconsistent findings, likely due to small sample sizes, non-standardized symptom definitions, and exploratory imaging analyses lacking statistical rigor. Current evidence offers limited support for specific neural correlates of this phase, and its translatability to spontaneous migraine remains uncertain. To improve interpretability and clinical relevance, future research should prioritize standardized definitions, adequate control groups, and use of appropriate neuroimaging statistics.
Trial Registration
PROSPERO - Identifier: CRD42023415959.
This is a visual representation of the abstract.
Introduction
Migraine is a disabling neurological disorder marked by recurrent attacks of moderate-to-severe headache, typically accompanied by photophobia, phonophobia, nausea, and vomiting (1,2). In addition to the headache phase, some people with migraine experience premonitory symptoms—non-painful features such as fatigue, mood changes, and yawning—that can emerge up to 48 h prior to pain onset (2–4). These symptoms, reported by about one-third of individuals with migraine, are garnering increased attention as potential early markers of brain dysfunction preceding the development of headache (3,5,6). Despite their clinical relevance, the underlying neurobiological mechanisms that generate premonitory symptoms remain inadequately characterized (7).
Efforts to define the premonitory phase of migraine have been limited by its variable clinical presentation and the retrospective nature of symptom recognition in most study designs (3,4). Inconsistent recall of early manifestations, combined with frequent departures from standardized criteria, has limited the reliability of these symptoms as predictors of migraine onset (3,4,8). To address these methodological challenges, researchers have used provocation models in which migraine attacks are pharmacologically induced under controlled conditions, (9,10) most often through administration of glyceryl trinitrate (GTN), a nitric oxide donor (11,12). This strategy might capture the occurrence of premonitory symptoms within a defined temporal window and could support systematic neuroimaging assessment of functional brain changes preceding the headache phase (13–19).
Although pharmacologic provocation offers a structured approach to studying early migraine pathophysiology, (9) findings from neuroimaging studies using this model have remained inconsistent and reproducibility untested (13–19). Reports from both positron emission tomography (PET) and magnetic resonance imaging (MRI) have described altered activity in regions such as the hypothalamus, brainstem, and thalamus (13–19). However, these findings have lacked replication, largely because of differences in study designs, small sample sizes, non-standardized symptom definitions, and inadequate correction for multiple comparisons in imaging analyses. These limitations further complicate efforts to clarify the significance of individual premonitory symptoms, such as photophobia or nausea, which are often closely associated with the headache phase and therefore difficult to isolate within a specific temporal window (2,20).
Given the clinical importance of early migraine detection and the potential for preemptive intervention, characterizing the neural mechanisms underlying premonitory symptoms has become a subject of growing scientific interest. Neuroimaging studies employing pharmacologic provocation, particularly with GTN, offer a controlled framework to investigate these early-phase brain changes (13–19). This systematic review synthesizes existing evidence from such studies, critically examines methodological variability, and highlights areas in need of greater analytic rigor to inform future research directions in migraine pathophysiology.
Methods
Definitions and terminology
The definition of premonitory symptoms used in this review is based on the International Classification of Headache Disorders (ICHD), which outlines this symptomatic phase as occurring prior to pain or aura onset (2). In the first three editions of the ICHD (ICHD-1, ICHD-2, and ICHD-3β), the term premonitory symptoms was favored (21–23), whereas the most recent edition (ICHD-3) recommends the use of prodrome to describe the same phase (2). Despite this shift, the two terms remain functionally synonymous, both referring to the constellation of non-painful symptoms that precede migraine headache. Notably, in an editorial published by the incoming Chair of the ICHD Classification Committee, a preference was expressed for reinstating premonitory symptoms in future editions (24). Given its broader representation in the scientific literature and comparable meaning, the term premonitory symptoms is used throughout this review for consistency and clarity.
Search strategy and selection criteria
This systematic review adhered to the PRISMA reporting guidelines (25), and the protocol was registered with PROSPERO (Identifier: CRD42023415959). A systematic literature search was performed to identify neuroimaging studies investigating the provoked premonitory phase in participants with migraine. The electronic databases MEDLINE (via PubMed) and Embase were searched from inception to June 1, 2025.
The following search string was used: migraine AND (premonitory OR prodromal OR prodrome). Eligible studies were original neuroimaging investigations reporting on premonitory symptoms (any case definition) in relation to provoked migraine attacks (Table 1). Exclusion criteria were reviews, case reports, case series, and conference abstracts (Table 1). Reference lists of relevant sources were hand-searched to identify any additional relevant studies.
Inclusion and exclusion criteria.
ICHD: International Classification of Headache Disorders.
After removal of duplicate records, titles and abstracts were independently screened for relevance by two reviewers (NV and RHC). Full-text articles meeting preliminary eligibility were then evaluated, and data were extracted using a standardized form.
Data analysis
Data extraction was performed using a standardized template by two independent reviewers (NV and RHC). Discrepancies in extracted variables or interpretations were resolved through discussion or, if necessary, adjudication by a third reviewer (HA). Variables collected included study design, sample size, demographic characteristics of participants with migraine, migraine subtype, provocation protocol, and explicit criteria for defining both the premonitory phase and associated symptoms. Neuroimaging-specific variables encompassed the imaging modality employed, timing of scans in relation to symptom onset, analytical framework, statistical models, significance thresholds, and methods used for correction of multiple comparisons. When applicable, additional design-dependent parameters such as imaging sequence, radiotracer identity, experimental task type, and predefined regions of interest or seed regions for connectivity analyses were also recorded.
Due to the heterogeneity of the study designs of the included studies, the conduct of a quantitative analysis was not considered to be feasible. Therefore, each study is presented descriptively. Qualitative analysis is used for interpretation. A formal risk of bias assessment was not performed, since validated tools are not available for neuroimaging studies and to avoid compromising comparability due to the considerable variation in study design. Instead, we assessed the quality of the study designs by reviewing the selection of participants, methodological approaches, and resulting outcomes. Finally, attention was paid to the use of exploratory analyses, especially in studies using post-hoc subgrouping or uncorrected thresholds. The presence of these features was used to assess the interpretability and reliability of findings across studies.
Results
The systematic literature search identified 1089 records, of which 731 remained for title and abstract screening following the removal of duplicates (Figure 1). Screening of these records resulted in 32 articles being selected for full-text evaluation, during which each was assessed against the predefined inclusion criteria. Seven of these 32 studies met eligibility requirements by employing neuroimaging techniques to investigate provoked premonitory symptoms in participants with migraine and were therefore included in the qualitative synthesis (13–19). Full texts were mainly excluded for not investigating premonitory symptoms or not conducting neuroimaging. The principal methodological features and primary neuroimaging findings of each study are summarized in Tables 2 and 3.
PRISMA flow diagram.
Combined provocation and PET studies. Characteristics and major findings of the included studies. Results are reported only if significant, unless otherwise stated.
CBF – cerebral blood flow, CGRP – calcitonin gene related peptide, FDR – false discovery rate, GTN - glyceryl trinitrate, HC – healthy control, MA - migraine with aura, MO - migraine without aura, PAG – periaqueductal gray, PET – positron emission tomography, PS – premonitory symptoms.
*Total sample eligible for analysis. Participants with both migraine with and without aura were counted only in the migraine with aura group.
Combined provocation and MRI studies. Characteristics and major findings of the included studies. Results are reported only if significant, unless otherwise stated.
ANOVA – analysis of variance, BOLD - blood oxygen level-dependent, CBF – cerebral blood flow, CGRP – calcitonin gene related peptide, FC - functional connectivity, FDR – false discovery rate, fMRI – functional magnetic resonance imaging, FWE – familywise error, GTN - glyceryl trinitrate, HC – healthy control, MA - migraine with aura, MO - migraine without aura, MRI – magnetic resonance imaging, pCASL – pseudo continuous arterial spin labeling, PS – premonitory symptoms, ROI – region of interest, SVC – small volume correction.
*Total sample eligible for analysis. Participants with both migraine with and without aura were counted only in the migraine with aura group.
Methodology of the included studies
Across the seven included investigations, (13–19) the mean number of participants with migraine was 14, with individual study sample sizes ranging from 5 to 25, as detailed in Tables 2 and 3. Participants were adults with mean ages spanning 32 to 50 years and predominantly female. Five studies exclusively enrolled participants with migraine without aura (13,14,17–19), whereas two included individuals with either migraine with aura or without aura (15,16). All studies focused on episodic migraine, although one also included participants with chronic migraine (15).
Inclusion criteria regarding premonitory symptoms varied considerably: five studies required a documented history of such symptoms (13–16,19), one reported the proportion of participants with a relevant history (18), and one provided no details on this aspect (17). Regarding study design, four investigations used a single-arm format (13,14,17,19), two applied double-blinded, 2-way crossover, placebo-controlled protocols (15,16), and one implemented an open-label, 2-way crossover design without placebo during the control day (18). Only one study incorporated a healthy control group (18).
All neuroimaging studies used GTN to provoke migraine attacks under controlled conditions. Three studies employed PET to measure cerebral blood flow changes before headache onset (13,14,19). One applied pseudo-continuous arterial spin labeling to quantify regional cerebral blood flow (15), and another used task-based functional MRI to assess the blood oxygen level–dependent signal (18). Two studies used resting-state functional MRI to examine functional connectivity, with one of these performing wavelet coherence analysis (16,17).
Case definitions of premonitory symptoms
Across the included investigations, case definitions of premonitory symptoms differed in both scope and operational detail. Maniyar et al. described premonitory symptoms as symptoms typically preceding the subject's migraine headache, which were present after the GTN induced headache had completely subsided, and before the migraine headache had appeared (13,14,19). In contrast, Karsan et al. adopted a broader operational definition, classifying any symptom that emerged after GTN infusion and resembled those typically experienced before migraine as premonitory (15,16). Martinelli et al. considered symptoms premonitory if they were included in a list of possible features typical of the premonitory phase (this list was not disclosed) (17), while Van Oosterhout et al. classified premonitory symptoms as those reported more than 30 min after GTN infusion, provided they corresponded to the participant's usual premonitory symptoms (18).
Case definitions of premonitory phase
The definitions of the premonitory phase likewise differed substantially across the included studies, reflecting variation in both temporal parameters and the number of required symptoms. The ICHD-3 defines the prodromal phase as the occurrence of one or more prodromal symptoms preceding pain onset (2), yet individual studies applied more restrictive or operationally distinct criteria. Maniyar et al. characterized the premonitory phase as the interval following resolution of initial GTN-induced headache during which participants experienced symptoms consistent with their usual premonitory profile on at least two separate enquiries (13,14,19). Karsan et al. required the presence of at least three such symptoms, reported after GTN infusion, with confirmation on a minimum of two separate assessments (15,16). Martinelli et al. defined the phase as the occurrence of at least two symptoms from a predefined list of typical premonitory features (18). Van Oosterhout et al. defined only premonitory symptoms, without relying on a distinct definition of the premonitory phase (18).
Perfusion imaging
Maniyar et al. investigated alterations in regional cerebral blood flow (CBF) during a GTN-induced premonitory phase using H215O PET (14). The study enrolled eight participants with migraine without aura who reported experiencing premonitory symptoms in relation to their spontaneous attacks. No placebo control was applied, and the authors did not include any comparator group. Each participant received a 20-min intravenous infusion of GTN, after which H215O PET–MRI scans were performed at three predefined time points: before GTN infusion, during the premonitory phase, and during the headache phase of the induced migraine attack. The premonitory phase was defined as the period beginning after resolution of the initial GTN-induced non-migraine headache, during which participants reported symptoms consistent with their usual premonitory symptoms. This phase was further subdivided into an ‘early’ and a ‘late’ premonitory phase, corresponding to the first and second scans obtained during this interval, although no explicit temporal thresholds were specified for these subdivisions.
In the ‘early’ premonitory phase, increased regional CBF was observed in 18 brain regions compared with scans obtained before GTN infusion. These regions included, among others, the dorsal pons, periaqueductal gray, occipital cortex, cerebellum, and the posterior portion of the right hypothalamus. During the late premonitory phase, increased CBF was detected in five regions, with the dorsal pons being the only structure demonstrating perfusion changes in both phases. When all premonitory-phase scans were compared with those obtained before GTN infusion, increased CBF was identified in 13 brain regions. The statistical analysis used false discovery rate correction; however, the authors did not specify whether the single threshold reported was used at a voxelwise or clusterwise level, or subsequently for peak voxel P values.
In two subsequent publications, Maniyar et al. conducted exploratory analyses that expanded upon findings from their initial investigation (13,19). These later studies included two additional participants, increasing the total sample size to ten individuals with migraine who reported premonitory symptoms associated with their spontaneous attacks.
The first exploratory analysis compared participants who developed photophobia during the induced premonitory phase with those who did not (13). After GTN infusion, five participants exhibited photophobia during scanning, whereas five remained free of this symptom. From baseline to the premonitory phase, the photophobic group demonstrated greater increases in perfusion within the right cuneus (containing parts of the visual cortex), and the precentral gyrus compared with the non-photophobic group. During the premonitory-like phase, participants with photophobia exhibited increased perfusion in the bilateral visual system, including the striate and extrastriate visual cortices, the right precuneus, and bilateral frontal cortex. Additional increases were observed in the left parietal lobule and left parahippocampal gyrus relative to baseline. In contrast, those without photophobia showed greater perfusion in the right visual cortex, bilateral frontal cortex, and left posterior cingulate cortex when compared with baseline. Statistical analyses in this study used voxelwise thresholds uncorrected for multiple comparisons.
The second exploratory analysis evaluated perfusion changes in three participants experiencing nausea during the induced premonitory phase and seven participants without nausea (19). Initial uncorrected voxelwise analysis identified increased perfusion within the brainstem, which subsequently underwent post-hoc small volume correction based on false discovery rate. Between baseline and the premonitory phase, the nausea group demonstrated greater perfusion increases in the rostral dorsal medulla and periaqueductal gray compared with the non-nausea group. Within the nausea group, perfusion was increased in the same brainstem regions when compared with baseline scans, whereas no brainstem perfusion increases were observed among participants without nausea.
Karsan et al. explored changes in regional CBF during the premonitory phase of migraine attacks provoked by GTN (15). The study initially enrolled 53 participants with migraine and applied a double-blind, randomized, placebo-controlled imaging design using pseudo-continuous arterial spin labeling to measure regional CBF. Participants who did not develop both premonitory symptoms and headache during the screening visit, or who developed premonitory symptoms without subsequent headache, were excluded from the imaging protocol. The final analyzed dataset included 20 participants who exhibited both features at screening and subsequently underwent imaging on two separate occasions. Each participant completed a baseline scan before GTN administration and a premonitory-phase scan after GTN administration. These sessions were scheduled at least two weeks after the screening visit and separated by a minimum of two weeks to minimize any potential GTN carryover effects.
The initial analysis compared changes in CBF from baseline to the premonitory phase after GTN administration with those observed at the corresponding time point after placebo. This comparison revealed no significant differences between GTN and placebo. The investigators then conducted additional post-hoc exploratory analyses by excluding eight participants who were receiving preventive treatment. Among the remaining 12 participants, CBF within the right anterior cingulate cortex differed between GTN and placebo. When applying a more liberal, exploratory statistical threshold, additional differences emerged in the caudate nucleus, midbrain, lentiform nucleus, amygdala, hippocampus, and parahippocampal gyrus. In this post hoc–selected group, altered CBF was also observed in the hypothalamus, although only in region-of-interest analyses. All tests were conducted as analyses of variance without post-hoc pairwise comparisons to determine the direction of the effects. However, descriptive reporting indicated increased perfusion in the anterior cingulate cortex and hypothalamus during the premonitory phase.
In a further post-hoc exploratory analysis, Karsan et al. examined changes from baseline to the premonitory phase after GTN without accounting for whether similar changes occurred after placebo. This analysis revealed increased CBF in the bilateral orbitofrontal and frontal cortices, including the medial frontal, superior frontal, and rectal gyri, along with decreased perfusion in the occipital cortex. The reduction in occipital cortex perfusion was also present when analyses were restricted to participants with aura.
Functional MRI
Van Oosterhout et al. investigated hypothalamic activity during the premonitory phases of both GTN-induced and spontaneous migraine attacks (18). The study used task-based functional MRI to assess the blood oxygen level–dependent response to glucose ingestion within the hypothalamus of 19 female participants with migraine without aura and 15 age- and BMI–matched controls. Imaging was conducted on two separate days: the first without prior GTN administration and the second following GTN administration, during the premonitory phase of an induced migraine attack. Premonitory symptoms were defined as symptoms reported more than 30 min after GTN infusion that were recognized by participants as their usual premonitory features.
For analysis, Van Oosterhout et al. applied statistical tests which were uncorrected, given the small number of comparisons with distinct sub-hypotheses. The study investigated a manually segmented hypothalamic ROI, which makes reproduction challenging compared with standardized atlases. Following GTN infusion, the lateral hypothalamic blood oxygen level–dependent response within the migraine group was greater on the GTN day than on the non-GTN day. Participants with migraine who developed an attack after GTN infusion also exhibited a significantly faster and steeper recovery compared with those who did not develop an attack. In contrast, there was no statistically significant difference between participants who developed an attack and controls, or between controls and the total group of participants with migraine. Finally, five participants were also scanned during spontaneous migraine attacks. In this subgroup, the lateral hypothalamic response did not differ significantly from that observed after GTN infusion or from the non-GTN day.
Using an overlapping cohort from the previously described investigation, Karsan et al. conducted a randomized, double-blind, placebo-controlled crossover trial employing resting-state functional MRI (16). The objective was to examine alterations in brain functional connectivity during the GTN-triggered premonitory phase, relative to both baseline and placebo. The study comprised eligible data from 25 participants with migraine, including 10 without aura and 15 with aura, all of whom reported experiencing premonitory symptoms preceding spontaneous attacks.
Each participant received a 20-min intravenous GTN infusion to provoke a migraine attack. Functional MRI scans were performed at baseline, during the premonitory phase, and during the migraine headache. Additional functional MRI scans were performed following placebo infusion. The premonitory phase was defined as the occurrence of at least three premonitory symptoms after GTN infusion in the absence of migraine headache, with symptoms confirmed at two separate time points during structured questioning.
During the premonitory phase, functional connectivity between the bilateral thalami and the right precuneus and cuneus was altered compared with both baseline and placebo. Descriptively, only losses of connectivity were observed, though pairwise comparisons of directionality were not performed. No significant differences in connectivity were observed when comparing participants with and without aura or when stratifying by use of preventive treatment.
Martinelli et al. investigated alterations in functional connectivity in five participants with migraine without aura during the GTN-triggered premonitory phase, the migraine attack, and the subsequent recovery phase using resting-state functional MRI (17). The authors did not specify whether participants had a history of premonitory symptoms preceding spontaneous attacks. All participants received a 20-min intravenous GTN infusion. The premonitory-like phase was defined as the presence of at least two symptoms from a predefined list of typical premonitory features occurring before a GTN-induced migraine headache. Although the authors did not explicitly address whether headache was permissible during this phase, a pain intensity score of 0 to 1 on a numerical rating scale from 0 to 10 was indicated in the corresponding figure.
Seed-based correlation analysis, using the thalamus as the seed region, revealed alterations in functional activity confined to the right thalamus when the premonitory phase was compared with the recovery phase. These changes appeared to involve connectivity with the pons, insula, and cerebellum, although the precise loci and direction of effects were not reported in detail. Complementary wavelet coherence analysis, which quantifies synchronization of neural activation frequencies while accounting for phase delays between regions, demonstrated a progressive loss of phase synchronization bilaterally between the thalamus and the salience network during both the premonitory and headache phases.
Discussion
This systematic review represents the first structured synthesis of neuroimaging studies examining the experimentally induced premonitory phase of migraine. All seven included investigations used GTN provocation, enabling temporal capture of early-phase brain changes (13–19). Reported alterations in cerebral perfusion and functional connectivity were discrepant and frequently derived from exploratory analyses lacking appropriate statistical correction. While some findings implicated regions such as the thalamus, pons, or hypothalamus, results were inconsistent and reproducibility untested (14–18). Methodological constraints—including small sample sizes, heterogeneous phase definitions, absence of adequate controls, and unconventional imaging analyses—substantially limit interpretability. The current evidence base provides only weak, non-reproduced support for specific neural correlates of the induced premonitory phase.
Existing evidence and added value
Perfusion imaging studies have yielded widely divergent results. The first imaging study was by Maniyar et al., who reported that premonitory symptoms after GTN were linked to hyperperfusion in multiple regions, including the dorsal pons, ventral tegmental area/substantia nigra, occipital cortex, and cingulate cortex, using H215O-PET (14). A subsequent analysis distinguishing ‘early’ and ‘late’ phases identified 23 hyperperfused regions overall, of which one was the right hypothalamus. However, ambiguities in statistical thresholds and correction methods raise some concerns – for example, it is unclear if areas of hypoperfusion were assessed at all, or if the single threshold reported pertains to a voxelwise or clusterwise approach. This has important consequences for the actual false-positive rate in neuroimaging (26). The small sample size of eight is another key concern in interpreting this oft-cited study (27).
A more recent study by Karsan et al. observed no significant perfusion differences during GTN induced premonitory symptoms versus placebo, using pseudo-continuous arterial spin labeling (15). Yet excluding participants on preventive medication revealed altered perfusion in the right anterior cingulate cortex and superior temporal gyrus, and further exploratory region-of-interest analysis suggested increased hypothalamic perfusion—though the effect's direction was not statistically tested.
Symptom-specific analyses have also been attempted. In Maniyar et al.'s subgroups, premonitory photophobia was linked to increased perfusion in the right visual and precentral cortices, whereas nausea was associated with rostral dorsal medulla and periaqueductal gray activation (13,19). Both analyses used uncorrected thresholds, involved very small subgroups (nausea: n = 3), and applied post hoc small-volume correction—introducing a considerable risk of spurious associations (28). It is important to note that, photophobia and nausea occur in over 90% of headache phases but are less common in the premonitory phase, where they almost invariably persist into the headache (20,29). Their presence might therefore suggest meningeal involvement (30), just as they commonly occur in disorders such as meningitis (31).
Functional MRI studies have similarly reported heterogeneous results. Van Oosterhout et al. examined hypothalamic blood oxygen level–dependent (BOLD) responses to glucose ingestion (18). In this paradigm, the hypothalamic response (a decrease) and its recovery is dose-dependently related to the amount of glucose ingested, at least in certain parts of the hypothalamus (32). The response is therefore thought linked to satiety (32). In Oosterhout's study, those who developed a migraine attack after GTN displayed slower recovery rates of the BOLD signal than those who did not – potentially suggesting increased satiety signals or an aversion to food intake. Yet sample imbalance (only three without attacks) limits generalizability, and no significant differences were found between those who subsequently developed migraine attacks and healthy controls (33,34). This challenges attribution of the findings to a premonitory dysfunction.
Resting-state connectivity findings also vary considerably, with some findings having pointed towards the thalamus and pons. Karsan et al. related the induced premonitory phase to reduced connectivity between the bilateral thalami and occipital regions, as well as between the pons and limbic lobe (16). However, the reductions in connectivity appeared driven equally by baseline differences between placebo and GTN. Moreover, a five-participant pilot study by Martinelli et al. found altered connectivity between the right thalamus and the insula, pons, and cerebellum, as well thalamic-salience networks, during premonitory and headache phases (17). These exploratory methods, combined with small samples, limit reproducibility, and it should be noted that findings do not clearly implicate the thalamus in the spontaneous premonitory phase though some do during migraine headache (35). Early imaging work suggested activation of especially the dorsolateral pons during migraine attacks (36,37), whereas one later study implicated the region during the spontaneous premonitory phase (6). However reassessment of prior evidence have raised questions about the exact role of the pons in generating migraine attacks, rather highlighting a potential modulatory role (38).
Despite hypothalamic interest (6,7,39,40), it is apparent that this region was at best inconsistently implicated. Only isolated analyses reported perfusion or connectivity changes, many of which were exploratory or used liberal statistical approaches (14,15,18). Altogether, across modalities, findings lacked attempts at replication, statistical rigor, and control for confounding by GTN side effects. A general concern of current functional MRI studies on the premonitory phase has also been limited sample sizes, with study comparison groups often ranging from three to five participants. Recent findings suggest that sufficient sample sizes are fundamental to avoid false positives in group-comparisons of functional MRI data, even to the extent that small studies may show effects opposite of ground truth (33,34). Taken together, this review adds value by integrating these heterogeneous results into a cohesive appraisal, highlighting the core methodological weaknesses that underpin discrepancies. It makes explicit that current inconsistencies reflect structural limitations in study design, not merely variability in migraine biology.
Implications and future directions
The lack of reproduced neural signatures raises two possibilities: one option is that the premonitory phase might not represent a uniform neurobiologic state across individuals, as different premonitory symptoms likely involves distinct brain circuits. Findings linking premonitory photophobia and nausea to separate cerebral regions supports to this notion (13,19). The other option is that current study designs lack the ability to detect consistent changes, a plausible explanation given the substantial variability in design. A number of steps might be taken to mitigate this heterogeneity.
First, standardizing definitions is essential. Uniform operational criteria for the premonitory phase should be applied to maintain comparability across studies. Proposed criteria might require I) the presence of one or more non-headache symptoms excluding overt side effects such as flushing or tachycardia, II) clearly separate from headache onset, with exclusion of concurrent head pain, III) followed by migraine headache or aura, IV) prospectively logged to avoid recall bias. This would likewise preserve core similarities with the ICHD definition, while excluding premonitory-like symptoms following trigger administration in participants who do not develop migraine headache. Second, increasing statistical power is critical. Larger sample sizes are needed to detect reproducible effects (34). Achieving adequate power might require multicenter collaborations that specifically recruit participants with migraine who reliably experience premonitory symptoms that precede their spontaneous attacks. Third, rigorous imaging methodology must be adopted (26,41). This includes pre-registration of analytic pipelines, and appropriate and transparent correction for multiple comparisons. Effect sizes should most appropriately be reported with statistical maps or within a priori defined regions of interest, to avoid distorted presentations of data (28,42,43). Lastly, careful control for potential confounds is necessary. Distinguishing GTN-induced premonitory symptoms from general non-headache side effects of GTN poses certain challenges, especially given the vague nature of premonitory symptoms such as fatigue, yawning, or mood changes. The highly heterogenous definitions of premonitory symptoms across studies therefore likely contributes to inconsistencies in the findings. Inclusion of healthy control groups and the use of alternative provocation agents would help distinguish migraine-specific changes from non-specific effects of GTN. Other approaches include comparing participants who develop premonitory symptoms to those who do not. These considerations are not trivial, given that the main effect of GTN upon functional and perfusion imaging remain relatively unelucidated. Some findings suggest an influence on PET perfusion signals (44), whereas functional MRI effects might be context dependent; for example, GTN did not affect the BOLD response to visual stimulation in a small sample of healthy volunteers (45). Likewise, it remains speculative whether the provoked premonitory phase aligns pathophysiologically with the spontaneous. While the premonitory phase is thought to reflect central changes preceding peripheral activation, evidence suggests GTN and other triggers initiate peripheral events that secondarily activate the CNS—undermining induced premonitory symptoms as true facsimiles of spontaneous ones (10,11,30). Whereas some imaging findings have pointed towards regions sometimes implicated in the spontaneous premonitory phase, such as the hypothalamus (6), this same region is also activated by peripheral trigeminal pain (46). Direct comparisons of provoked and spontaneous premonitory phases in the same individuals could clarify whether the induced state is mechanistically relevant. Until such refinements are implemented, the induced premonitory phase should be regarded as a provisional research construct, which might differ in key pathophysiological aspects from its spontaneous counterpart.
Limitations
This review has several limitations. First, the available literature is sparse—only seven studies met inclusion criteria—and marked heterogeneity in design, imaging modality, and statistical approach precluded quantitative meta-analysis. Second, the absence of a formal risk-of-bias assessment reflects the current lack of validated risk of bias tools for assessing neuroimaging studies, although our qualitative appraisal addressed potential risk domains. Third, - aligning with the scope of the review- restriction to pharmacologic provocation models means findings might not generalize to spontaneous premonitory phases, where underlying neurobiologic processes could differ (10). Lastly, the small participant numbers across studies and reliance on exploratory analyses substantially limit the certainty of mechanistic inferences. These constraints reflect not only the limitations of this review but also persistent design weaknesses in the existing literature.
Conclusions
Neuroimaging studies of the experimentally induced premonitory phase of migraine have produced widespread and largely discrepant findings. Specific regions such as the thalamus, pons, and hypothalamus, have only inconsistently been implicated. Methodological weaknesses, including small sample sizes, heterogeneous definitions, and insufficient statistical rigor, limit interpretability. Advancing this field requires standardized definitions, adequate sample sizes, and rigorous analytical approaches.
Key findings
Neuroimaging evidence supporting specific neural correlates of the induced premonitory phase was heterogenous and scarce.
Most studies employed limited sample sizes, exploratory analyses, and variable case definitions potentially confounded by the provoking agent.
Future work should strive to combine well-designed and adequately powered studies with validated analytical approaches.
Footnotes
Acknowledgments
Not applicable.
Ethical considerations
Not applicable.
Consent to participate
Not applicable.
Consent for publishing
The authors agree to publish the accepted paper in Cephalalgia
Author contributions
Rune Häckert Christensen – Project administration, data curation, investigation, writing – original draft, writing – review & editing.
Nina Vashchenko – Data curation, investigation, writing – original draft, writing – review & editing.
Anna Kristina Eigenbrodt – Data curation, writing – original draft, writing – review & editing.
Afrim Iljazi – Data curation, writing – review & editing.
Messoud Ashina – Conceptualization, funding acquisition, project administration, supervision, writing – review & editing.
Håkan Ashina – Conceptualization, funding acquisition, project administration, supervision, validation, writing – review & editing.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by research grants from Lundbeck Foundation (R310-2018-3711 to MA and R403-2022-1352 to HA).
Declaration of Conflicting Interests
The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: RHC has received personal fees from AbbVie, Lundbeck, Pfizer, and Teva, outside of the submitted work, and serves as Section Editor of the Journal of Pain Research. MA has received personal fees from AbbVie, Astra Zeneca, Eli Lilly, GlaxoSmithKline, Lundbeck, Novartis, Pfizer and Teva, outside of the submitted work. MA also serves as an Associate Editor of Brain and The Journal of Headache and Pain. HA has received personal fees from AbbVie, Lundbeck, Pfizer and Teva, outside of the submitted work. HA also serves as an Editorial Board Member of The Journal of Headache and Pain. The remaining authors report no competing interest.
Data availability statement
Data supporting the current findings will be made available upon reasonable request from a qualified investigator.
Open practices
Not applicable.