Reproducible left anterior-superior hypothalamic enlargement in episodic migraine: evidence from two independent cohorts

Abstract

Background

Growing evidence implicates the hypothalamus as a key structure in migraine pathophysiology; however, our understanding of its precise role and of the specific nuclei involved remains limited. We combined MRI data from our laboratory with publicly available MRI datasets from OpenNeuro to examine hypothalamic subunit volumes in episodic migraine and to assess whether comparable alterations were observed in available chronic pain datasets.

Methods

Structural MRI combined with an automated atlas-based segmentation algorithm was employed to investigate cross-sectional volumetric differences across 5 bilateral hypothalamic subunits in two independent migraine cohorts: DS1-MIG (DS1-MIG-base, n = 111 patients, n = 35 controls) and DS2-MIG (n = 28 patients, n = 31 controls). The adjusted volumes were compared between groups using MANOVA as an omnibus test, followed by Welch t-tests to test univariate follow-up. Longitudinal volumetric changes were additionally assessed in DS1-MIG participants with available follow-up scans using linear mixed models. To examine whether comparable hypothalamic alterations were observed in available clinical comparison datasets, the same cross-sectional pipeline was applied to two chronic pain datasets, one including patients with fibromyalgia (DS-FM, n = 33 patients, n = 33 controls) and the other including patients with trigeminal neuralgia (DS-TN, n = 119 patients, n = 55 controls).

Results

MANOVA revealed significant multivariate group differences in both independent migraine cohorts (DS1-MIG-base: p = .006; DS2-MIG: p = .008). Follow-up univariate analyses identified a consistent enlargement of the left anterior-superior subunit across both cohorts ( in DS1-MIG-base and in DS2-MIG), representing the only cross-cohort replication finding. Beyond this shared signature, DS2-MIG exhibited additional significant enlargements of the right anterior-inferior and right inferior tubular subunits. Longitudinal analyses in DS1-MIG showed no significant group-by-time interactions in hypothalamic subunit volumes over the available follow-up period. No significant volumetric alterations were detected in the fibromyalgia or trigeminal neuralgia cohorts, either in multivariate or univariate analyses.

Conclusions

These findings provide evidence for subunit-specific hypothalamic structural alterations in migraine localized to the left anterior-superior hypothalamic subunit. The absence of detectable differential longitudinal change over the available follow-up period, together with the absence of comparable alterations in the available fibromyalgia and trigeminal neuralgia datasets, supports a migraine-related pattern of hypothalamic structural organization. However, this finding does not establish diagnostic specificity and should be confirmed in harmonized multi-diagnostic cohorts.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s10194-026-02479-9.

Introduction

The pathophysiological mechanisms of migraine, a complex neurovascular disorder affecting over one billion people worldwide [1, 2], remain poorly understood [3, 4].

Current evidence points to the involvement of both the peripheral trigeminovascular system and the central nervous system, with a potential prominent role of the hypothalamus [2]. Anatomically, the hypothalamus is a heterogeneous diencephalic region composed of multiple specialized nuclei that collectively support homeostatic, neuroendocrine, autonomic, circadian, and motivated behavioral regulation [5].

Crucially, characteristic features of migraine, such as rhythmic recurrence of attacks, the presence of prodromal manifestations (albeit their existence remains debated, see Christensen et al. [3]), such as yawning, fatigue, and mood alterations, autonomic symptoms like nausea and vomiting, and sensitivity to specific triggers, such as hormonal fluctuations and stress, point to a major involvement of the hypothalamus [6].

Functional neuroimaging studies have consistently implicated the hypothalamus across the migraine cycle, including sustained activation during spontaneous attacks [7], premonitory-phase activation [8–10], altered perfusion and connectivity with limbic and pain-processing networks around attack onset [11, 12], and interictal dysregulation of hypothalamic functional connectivity [13]. These functional findings are paralleled by preclinical evidence implicating hypothalamic neuropeptides in migraine mechanisms [14].

However, while clinical, functional neuroimaging, and preclinical evidence support a central role of the hypothalamus in migraine pathophysiology, consensus regarding the specific involvement of distinct hypothalamic nuclei and subregions remains elusive. In this context, high-resolution structural MRI provides a non-invasive in vivo approach to investigate whether migraine is associated with hypothalamic macrostructural alterations. An early voxel-based morphometry study reported region-specific alterations with respect to control subjects, showing reduced posterior hypothalamic volume in episodic migraine without aura () scanned at least 3 days after the last migraine attack, while reduced anterior hypothalamic volume in chronic migraine () [15]. Two very recent investigations [16, 17] employed the same FreeSurfer-based algorithm [18], a fully automated segmentation method that has been widely used to investigate hypothalamic structure in both neurological and neuropsychiatric disorders and healthy populations [19–24], to extract the volumes of 5 bilateral hypothalamic subunits: anterior-superior, anterior-inferior, posterior, inferior tubular, and superior tubular.

Xiong et al. [17] investigated a cohort of migraine patients (chronic, ; episodic, ) irrespective of ictal stage and aura status and reported that, compared with healthy controls, patients with chronic migraine exhibited significant volume reductions in the right superior tubular subunit and in inferior hypothalamic regions (specifically, the left inferior tubular and right anterior-inferior subregions). In contrast, patients with episodic migraine did not show statistically significant differences relative to controls, although descriptive analyses suggested a gradual decrease in hypothalamic volumes from controls to episodic and chronic migraine patients within the inferior tubular subunit. The authors speculated that the limited sample size of the episodic migraine group may have resulted in insufficient statistical power to detect significant differences. Notably, the episodic migraine group experienced a mean of ≈ 6 headache days per month. The same study also performed a Mendelian randomization study suggesting that a smaller inferior tubular subregion volume is linked to increased migraine risk [17].

Critically, Giardina et al. [16] found no significant volumetric differences between a group of 30 episodic migraine patients without aura in the interictal state with respect to a well-matched control group. In this case, migraine patients experienced a mean of ≈ 5-6 headache attacks per month.

Altogether, current volumetric findings suggest that hypothalamic alterations may be more evident in chronic migraine or in more severe migraine phenotypes, whereas direct evidence for consistent macrostructural abnormalities in episodic migraine remains limited. However, this evidence is constrained by the small number of available studies and by the relatively limited sample sizes of episodic migraine cohorts.

In light of converging clinical, preclinical, and functional neuroimaging evidence implicating the hypothalamus in migraine, additional studies are therefore required to determine whether the absence of consistent volumetric findings in episodic migraine reflects a true lack of macrostructural involvement or instead cohort-specific effects related to clinical heterogeneity or methodological differences.

It is therefore critical to examine whether hypothalamic structural alterations are present in episodic migraine when evaluated in larger and independent cohorts, whether such alterations represent a reproducible migraine-related feature, and whether comparable patterns are observed in other pain-related conditions. Longitudinal evidence is also needed to clarify whether hypothalamic volumes remain stable over time or vary in relation to clinical burden.

To address these unresolved questions, the present study applied the same hypothalamic segmentation algorithm employed in recent studies [16, 17] to four structural MRI datasets, including two independent cohorts of patients with episodic migraine and two clinical comparison cohorts comprising patients with fibromyalgia or trigeminal neuralgia. Three datasets were publicly available through the OpenNeuro repository, whereas one migraine cohort was acquired as part of our previous work [25].

This design allowed us to test whether episodic migraine is associated with reproducible hypothalamic macrostructural alterations, whether they preferentially involve specific hypothalamic subregions, and whether comparable alterations are observed in the available fibromyalgia and trigeminal neuralgia datasets. In particular, the fibromyalgia dataset was included to assess whether hypothalamic subunit alterations observed in migraine were also present in a chronic widespread pain condition [26], whereas the trigeminal neuralgia dataset was included to determine whether comparable alterations were present in a chronic craniofacial pain condition involving the trigeminal somatosensory system [27]. Finally, longitudinal data available for one migraine cohort [28] were used to examine the temporal stability of hypothalamic subunit volumes over a follow-up period during which patients underwent two treatment interventions and showed a reduction in headache attack frequency. Together, this approach aimed to clarify the reproducibility, regional specificity, and temporal stability of hypothalamic structural alterations in episodic migraine, while examining whether comparable alterations were observed in the available clinical comparison datasets.

Methods

Participants

In the present study, we used four independent MRI datasets. These included two cohorts of patients with episodic migraine, namely DS1-MIG, which comprised both baseline data (DS1-MIG-base) and longitudinal follow-up data (DS1-MIG-long) [29], and DS2-MIG [25], as well as two disease-control datasets: one comprising patients with fibromyalgia (DS-FM) [30] and one comprising patients with trigeminal neuralgia (DS-TN) [31]. Each dataset included a healthy control (CTRL) group. Except for the DS2-MIG dataset, whose data were acquired in our laboratory, all remaining datasets were obtained from the publicly available OpenNeuro repository (https://openneuro.org). Although this repository contains multimodal MRI data, in the present study, we used only the high-resolution 3D T1-weighted images. Demographic characteristics of the four datasets are reported in Table 1.

Table 1.

Demographic characteristics of patients and healthy controls across datasets. Age is reported as M ± SD. Sex is reported as number of females/males (F/M). age range as reported in Seminowicz et al. [28] which describes a randomized clinical trial including 98 patients with episodic migraine. age ranges as reported in Filimonova et al. [31], which describes 112 patients with trigeminal neuralgia and 48 control individuals. DS1-long is the matched longitudinal subsample of DS1-MIG comprising participants with both baseline and week-20 acquisitions available. All Ns are reported before outlier removal

Dataset	Group	N	Sex (F/M)	Age (years)	Age range	Between-group p-values
DS1-MIG-base (Migraine)	MIG	111	97 / 14	n.r.	18–65	n.r.
	CTRL	35	30 / 4*	n.r.	n.r.
DS1-MIG-long (Migraine, longitudinal)	MIG	79	73 / 6	n.r.	18–65	n.r.
	CTRL	25	23 / 2	n.r.	n.r.
DS2-MIG (Migraine)	MIG	28	21 / 7	28 ± 8.0	18–36	Sex: p = .106; age: p = .208
	CTRL	31	17 / 14	26 ± 7.2	19–47
DS-FM (Fibromyalgia)	FM	33	33 / 0	41.7 ± 6.1	30–50	Age: p = .887
	CTRL	33	33 / 0	41.5 ± 6.0	29–50
DS-TN (Trigeminal Neuralgia)	TN	119	n.r.	n.r.	30–84	n.r.
	CTRL	55	n.r.	n.r.	38–75

Abbreviations: MIG = migraine patients; CTRL = healthy controls; FM = fibromyalgia patients; TN = trigeminal neuralgia patients. n.R., not reported in the repository and/or in the paper of reference. * missing information in the repository

Migraine dataset 1 (DS1-MIG)

The DS1-MIG dataset consisted of MRI data from patients with episodic migraine (MIG) and control participants (CTRL), acquired across three sessions to assess the neural effects of enhanced mindfulness-based stress reduction (MBSR+) and stress management for headache (SMH). The clinical outcomes of these interventions were reported by Seminowicz et al. [28], showing that both treatments reduced the number of headache days at week 20. However, the reduction was significantly greater in the MBSR+ group (from 7.8 to 4.6 days) compared with the SMH group (from 7.7 to 6.0 days; p = .04). Furthermore, 52% of patients receiving MBSR+ achieved a ≥ 50% reduction in headache days, compared with 23% in the SMH group. For this dataset, we performed both cross-sectional and longitudinal analyses. Cross-sectional analyses were based exclusively on the baseline acquisition (DS1-MIG-base), collected before treatment onset. The baseline sample comprised 111 MIG patients (97 females, 14 males) and 35 CTRL participants (30 females, 4 males, 1 participant with missing sex information). For the longitudinal analyses, we included the baseline (DS1-MIG-base) and week-20 acquisitions (DS1-MIG-long), restricting the sample to participants with data available at both time points. This yielded a longitudinal sample of 79 MIG patients and 25 CTRL participants. For this dataset, participant-level sex information was available in the public repository, whereas participant-level age information was not.

Migraine dataset 2 (DS2-MIG)

The DS2-MIG dataset consisted of MRI data acquired in our laboratory and previously described in Ferraro et al. [25]. Briefly, individuals meeting preliminary eligibility criteria (age 18–65 years, at least one headache attack per month, and headache history longer than one year) were invited for diagnostic classification using the Clinical Decision Support System (CDSS 2.0), an automated interview system with demonstrated high sensitivity and specificity for diagnosing migraine according to the International Classification of Headache Disorders, 3rd edition (ICHD-3) criteria (Headache Classification Committee of the [32, 33]. Following CDSS 2.0 classification, 31 patients with MIG were enrolled; 3 were subsequently excluded during the initial image-quality assessment because marked signal loss in their native T1-weighted images, yielding a final sample of 28 MIG patients (28 ± 8.0 years; 21 females / 7 males). Three patients had migraine with aura; the remaining patients were diagnosed with migraine without aura. None of the included patients reported current or previous use of prophylactic migraine medication. Inclusion criteria for the MIG group required a diagnosis of definite migraine with stable headache characteristics for at least one year prior to study entry. Thirty-one CTRL participants (age, 26 ± 7.2 years; 17 females / 14 males) were enrolled as the comparison group. Exclusion criteria for all participants included a history of cardiovascular disease, diabetes, hypertension, psychiatric or neurological disorders, and any contraindications to MRI; additional exclusions for migraine patients included comorbidity with other primary or secondary headache disorders. Between-group comparisons showed no significant differences in age or sex distribution. Demographic and clinical characteristics of the DS2-MIG dataset are reported in Table 2. The study was conducted in accordance with the Declaration of Helsinki and was approved by the local ethics committee of the School of Life Science and Technology, University of Electronic Science and Technology of China. All participants provided written informed consent.

Table 2.

Demographic and clinical characteristics of the DS2-MIG dataset. The peri-ictal status was defined as having experienced a headache attack within the 72 hours preceding the scanning session. VAS indicates the reported level of pain intensity experienced during headache attacks; NRS indicates the level of pain intensity before the MRI scan; disorder duration indicates the reported years since headache onset; headache frequency indicates the number of attacks per month

	MIG (8)	CTRL (n = 31)	Statistic	p-value
Sex (M/F)	7/21	14/17	= 2.61	0.106
Age (M±SD)	28±8.0	26±7.2	U = 351.0	0.208
Disorder duration (years, M±SD)	9.4±6.99	-	-	-
Headache frequency	3.1±1.78	-	-	-
Pain level during attacks - VAS: 0–10 (M ± SD)	6.5±1.55	-	-	-
Pain level before MRI - NRS: 1–9 (M ± SD)	1.0±0	-	-	-
Patients with an attack in the last 72 hours (yes/no)	15/13	-	-	-
Patients without aura/with aura	25/3	-	-	-

Abbreviations: MIG, migraine patients; CTRL, control subjects; VAS, visual analog scale; NRS, numerical rating scale

Fibromyalgia dataset (DS-FM)

The DS-FM dataset consisted of MRI data from 33 FM patients and 33 CTRL participants [34]. FM patients were diagnosed according to the American College of Rheumatology 1990 [35] and [36] criteria. All participants were female, and groups were matched for age (p = .887). Participant-level age information was available in the public repository.

Trigeminal neuralgia dataset (DS-TN)

The DS-TN dataset consisted of MRI data from patients diagnosed with primary trigeminal neuralgia (n = 119) according to ICHD-3 criteria, together with CTRL participants (n = 55) [37]. Participant-level sex and age information was available in the public repository only for a subset of participants. For the present analyses, only the baseline MRI acquisition was used.

MRI data acquisition

For all datasets, 3D T1-weighted structural images were acquired on 3.0 T MRI scanners, with an isotropic voxel size of mm. Detailed scanner, head-coil, and sequence parameters for each dataset are reported in Supplementary Table SM1. DS1-MIG-base included two scanner/head-coil configurations, namely a Siemens Tim-Trio scanner equipped with a 32-channel head coil and a Siemens Prisma scanner equipped with a 64-channel head coil. The potential influence of scanner/head-coil configuration was explicitly assessed through sensitivity analyses in both the DS1-MIG cross-sectional (DS1-MIG-base) and longitudinal analyses (DS1-MIG-long).

Structural MRI processing and hypothalamic segmentation

All the 3D T1-weighted images were processed using FreeSurfer (v8.0.0). The standard recon-all pipeline was used to derive the total intracranial volume estimate (eTIV, expressed in ) for each subject. Following the standard processing, hypothalamic segmentations were performed using the dedicated FreeSurfer module (mri-segment-hypothalamic-subunits). Importantly, across all datasets, the 3D T1-weighted images were acquired with a spatial resolution of mm, at the same resolution for which the FreeSurfer segmentation model was optimized (https://surfer.nmr.mgh.harvard.edu/fswiki/HypothalamicSubunit); therefore, no preliminary resampling was required before segmentation.

The mri-segment-hypothalamic-subunits algorithm provides probabilistic segmentation of the hypothalamus into 5 bilateral subunits: anterior-inferior, anterior-superior, posterior, inferior tubular, and superior tubular [18]. These labels represent macro-anatomical subdivisions based on histological priors. For each subject, the segmentation pipeline generated a mask for each hypothalamic subunit and extracted the corresponding volume, expressed in . A representative example of the hypothalamic subunit segmentation and their anatomical localization is shown in Fig. 1. These volumetric measures were used in the subsequent statistical analyses, which were performed in R version 4.5.3.

Fig. 1.

T1-weighted MRI sagittal section illustrating the segmentation and anatomical localization of the hypothalamic subunits of a representative subject

Hypothalamic subunit volume adjustment for estimated intracranial volume (eTIV)

To minimize the influence of global head-size differences on regional hypothalamic volumes, all volumetric measures were adjusted for estimated total intracranial volume (eTIV) using the residuals method [38–40].

This approach removes variance in regional volume linearly attributable to eTIV and is considered an appropriate strategy for intracranial-volume correction in volumetric MRI analyses, reducing apparent sex-related gray-matter differences after adjustment [39, 40].

For each dataset and subunit, residualization was performed by estimating the eTIV correction model in CTRL participants only; the resulting model was then applied to all participants [39]. Specifically adjusted volumes were computed as follows:

1

where is the ordinary least-squares regression slope of the subunit volume on eTIV estimated within the control group, and is the mean eTIV of the control group. The same control-derived was then applied to all participants (both patients and CTRL) to compute , ensuring that the volume–eTIV scaling relationship was not contaminated by disease-related variance.

This procedure has two important properties: (i) the resulting adjusted values remain interpretable on the original volumetric scale (mm), rather than as standardized residuals, facilitating comparison with raw values and other studies; and (ii) deriving the adjustment slope from the CTRL group reduces the risk that disease-related volumetric variance is partially absorbed into the eTIV correction term, as may occur when the regression is fitted on the pooled patient-control sample [39].

Sensitivity analyses of residual sex- and age-related effects after eTIV adjustment in CTRL groups

Because participant-level demographic information was incomplete across datasets, formal covariate-based analyses, including age and sex, could not be applied uniformly to all samples. Therefore, before conducting the main group analyses, we performed sensitivity analyses restricted to CTRL participants from datasets with available demographic information to evaluate whether eTIV residualization reduced demographic-related variance in hypothalamic subunit volumes. See Supplementary Materials, section ‘Sensitivity analyses of residual sex- and age-related effects after eTIV adjustment in CTRL groups Methods’, for additional details.

Segmentation quality control and outlier handling

Before statistical outlier identification, the native T1-weighted images and the corresponding hypothalamic segmentation outputs underwent a general visual quality assessment by one of the authors (SF), who was not blinded to diagnostic group. The T1-weighted images were inspected for gross motion artefacts, incomplete brain coverage, marked signal loss, or other major image-quality problems that could compromise automated processing. This initial image-quality assessment resulted in the exclusion of three migraine participants from the DS2-MIG dataset because marked signal loss in their native T1-weighted images was considered likely to compromise reliable hypothalamic segmentation, as already reported in the “Participants” section. The corresponding segmentation outputs were subsequently reviewed for overall anatomical plausibility and gross segmentation failures, including complete absence of the hypothalamic segmentation, major anatomical mislocalization, or extensive under- or over-segmentation. No additional participants were excluded following this review, and no participant-level gross segmentation failures were identified. The visual assessment was intended as a broad pass/fail screening procedure to identify segmentations that were clearly unusable for volumetric analysis; it was not designed to verify the precise boundaries or voxel-wise accuracy of each individual hypothalamic subunit. No manual editing of the automatically generated hypothalamic masks was performed. Exact zero-volume estimates for individual hypothalamic labels were identified from the quantitative segmentation outputs. These estimates were considered label-specific segmentation failures, indicating that no voxels had been assigned to the corresponding label, and were recoded as missing values before eTIV adjustment. When either hemispheric estimate was unavailable, the corresponding eTIV-adjusted bilateral measure was also considered unavailable. This handling of label-specific segmentation failures was distinct from the subsequent IQR-based statistical outlier-identification procedure.

To identify and address potentially aberrant non-zero volumetric measurements while preserving statistical power, we implemented a hybrid outlier strategy separately within each dataset. Outlier detection was performed on eTIV-adjusted volumes using Tukey’s 1.5 × IQR rule and was applied independently to each eTIV-adjusted hypothalamic subunit volume included in the subsequent analyses. Importantly, no cross-variable propagation was performed during this statistical outlier-detection step: an outlier flagged on the bilateral measure of a subunit did not automatically result in flagging of the corresponding left- or right-hemisphere measures, and vice versa.

For each subject, we tabulated the number of distinct anatomical subunits flagged as outliers. Here, “root subunit” refers to the anatomical structure irrespective of laterality or aggregation level; for example, a subject flagged on the bilateral, left, or right anterior-superior volume was considered to have one affected root subunit.

We then applied a two-tier decision rule. First, multi-unit outliers, defined as subjects flagged on two or more distinct root subunits, were considered to have potentially unreliable multivariate volumetric profiles and were therefore excluded entirely from the corresponding dataset. Second, single-unit outliers, defined as subjects flagged on only one root subunit irrespective of whether the flagged value involved the bilateral, left, and/or right measure, were retained. For these subjects, only the specific flagged value(s) were set to NA, with no propagation to non-flagged measures of the same subunit or to other hypothalamic subunits.

This hybrid strategy is more conservative than complete subject exclusion (which would discard otherwise valid information) and more robust than retaining all values (which could allow an aberrant volumetric estimate to drive group-level effects).

A complete quality-control and outlier-handling audit trail, including dataset- and group-specific exclusions and missing values, is provided in the Supplementary Materials (see section ‘Segmentation quality control and outlier handling’ and Table SM2).

Cross-sectional group-comparison analyses

In each dataset (DS1-MIG-base, DS2-MIG, DS-FM, and DS-TN), the main cross-sectional analyses followed a hierarchical decomposition framework.

First, group differences (patients vs. CTRL) were tested at the bilateral anatomical level using an omnibus MANOVA including the 5 bilateral eTIV-adjusted hypothalamic subunit volumes (calculated as the sum of the corresponding left and right adjusted values) as dependent variables.

This multivariate test was used to assess whether each dataset showed an overall group effect across hypothalamic subunits. The main analysis was completed with several sensitivity analyses.

Second, bilateral effects were decomposed using univariate group comparisons for each eTIV-adjusted bilateral subunit volume.

Third, to further characterize the spatial pattern of any group differences and assess potential hemispheric specificity, the same univariate analyses were then performed separately for the corresponding left- and right-hemisphere eTIV-adjusted subunit volumes.

Multivariate test statistics were extracted using summary.manova with statistical significance defined as p < .05.

Bilateral hypothalamic subunit volumes: multivariate analyses (MANOVAs)

To test whether patients differed from CTRL participants in the overall bilateral hypothalamic subunit volume profile, we conducted a one-way MANOVA separately within each dataset. The 5 bilateral eTIV-adjusted hypothalamic subunit volumes (anterior-inferior, anterior-superior, posterior, inferior tubular, and superior tubular) were included as dependent variables, with group as the between-subjects factor. Pillai’s trace was used as the primary multivariate statistic because of its robustness to mild violations of multivariate normality and homogeneity of covariance matrices [41].

MANOVA was conducted using listwise deletion: participants with missing values in any of the 5 bilateral dependent variables were excluded from the multivariate test. Thus, in the primary cleaned dataset, subjects with single-unit outliers affecting one or more bilateral measures were excluded from the bilateral MANOVA. Distributional assumptions were inspected using per-group Q-Q plots for each subunit. Minor deviations from normality were observed in some anterior subunits, but no transformation was applied given the relative robustness of Pillai’s trace. Where participant-level demographic information (sex and age) was available, covariate-adjusted MANCOVAs were performed.

Sensitivity analyses

To assess the robustness of the bilateral MANOVA findings, omnibus analyses were repeated using the full uncleaned samples without outlier removal. Where participant-level demographic information (sex and age) was available, covariate-adjusted MANCOVAs were performed.

To evaluate the potential influence of acquisition heterogeneity in DS1-MIG-base, a coil-homogeneous sensitivity analysis was performed by restricting the sample to participants scanned with Coil 1 only, using the same five bilateral eTIV-adjusted hypothalamic subunit volumes and the primary hybrid outlier-handling strategy.

Finally, in DS2-MIG, we assessed whether the omnibus group effect was influenced by peri-ictal status by excluding migraine participants scanned within 3 days of their most recent attack, while retaining all CTRL participants.

Bilateral hypothalamic subunit volumes: univariate follow-up analyses

Following the bilateral hypothalamic subunit volumes omnibus MANOVA, we performed univariate decomposition on each of the 5 bilateral eTIV-adjusted subunits to identify which structures contributed to the multivariate group effect. For each subunit, Welch’s t-test (patients vs. CTRL) was used as the primary test. Effect sizes were quantified using Hedges’ g with 95% confidence intervals [42].

Univariate tests were performed on pairwise-complete observations, excluding only the specific missing value for each subunit. Benjamini–Hochberg FDR correction was applied across the 5 bilateral subunits; significance was set at adjusted p < .05.

Left- and right-hemisphere subunit volumes: univariate follow-up analyses

The same univariate framework was applied separately to the left- and right-hemisphere eTIV-adjusted subunit volumes to assess whether group differences showed hemispheric specificity. Specifically, we tested the 10 side-specific eTIV-adjusted hypothalamic subunit volumes. For each side-specific subunit, Welch’s two-sample t-test was used as the primary group comparison, given its robustness to unequal sample sizes and heteroscedasticity. Effect sizes were quantified using Hedges’ g with 95% confidence intervals. Analyses were performed using pairwise-complete observations, excluding only missing values for the specific subunit being tested.

To control for multiple comparisons, p-values were adjusted using the Benjamini–Hochberg FDR procedure at two levels. First, within-side FDR correction was applied separately within each hemisphere across the 5 subunit tests. Second, as a more conservative sensitivity correction, FDR adjustment was also applied across all 10 side-specific subunit tests within each cohort. Findings surviving within-side FDR correction were interpreted as hemisphere-specific effects, whereas findings additionally surviving the all-10 correction were considered the most robust side-specific effects. The side-specific analyses were used as localized anatomical follow-ups to characterize the hemispheric distribution of group differences observed at the bilateral level.

To assess whether the key replicated findings in MIG datasets (DS1-MIG-base and DS2-MIG) depended on the specific censoring strategy, complementary sensitivity analyses were conducted using the full uncleaned samples and winsorized versions of these samples. See Supplementary Materials for additional details (section ‘Sensitivity analyses addressing the influence of extreme observations Methods’).

Cross-cohort meta-analysis of the convergent subunit effect

To formally quantify the consistency of the side-specific hypothalamic subunit effect observed in the same direction in both DS1-MIG and DS2-MIG, we performed an inverse-variance fixed-effect meta-analysis using the cohort-specific Hedges’ g estimates and their corresponding sampling variances. The pooled Hedges’ g estimate and its 95% confidence interval were calculated, and between-cohort heterogeneity was assessed using Cochran’s Q and statistics.

Subunit structural covariance analysis and clinical correlates

To further characterize the anatomical organization of the most consistent hypothalamic subunit volumetric alteration identified in DS1-MIG and DS2-MIG datasets, we performed an exploratory subunit structural covariance analysis. The subunit showing consistent alteration across the migraine cohorts was used as the reference subunit and correlated with all remaining left- and right-sided hypothalamic subunits separately for each dataset and group; structural covariance (inter-subunit volume correlations) was then compared between patients and controls within each dataset.

As a further characterization of the subunit of interest, within DS2-MIG we also examined the association between this subunit’s volume and headache-related clinical measures, including attack frequency, disease duration, and VAS pain intensity during attacks. Full methodological details are reported in the Supplementary Materials, section ‘Subunit structural covariance analysis and clinical correlates Methods’.

Longitudinal analyses on dataset DS1-MIG

To examine whether the hypothalamic volumetric abnormality identified in the baseline univariate analyses persisted after treatment-related reductions in headache frequency, we conducted longitudinal follow-up analyses in the subset of DS1-MIG participants with both baseline (DS1-MIG-base) and week-20 follow-up (DS1-MIG-long) MRI acquisitions. After application of the outlier-handling criteria, the final matched analytic sample comprised 99 participants (75 MIG and 24 CTRL).

The primary longitudinal outcome was defined as the hypothalamic subunit volume that emerged as significant in the baseline univariate analyses.

As a first step, we assessed whether the participants included in the longitudinal analysis retained the baseline MIG versus CTRL difference observed in the full DS1-MIG-base sample. For this purpose, using baseline data only, we repeated the bilateral MANOVA and the univariate Welch’s t-test on the primary longitudinal outcome within the matched longitudinal subsample. The direction and magnitude of the resulting group effect were then descriptively compared with those observed in the full baseline sample.

As a second step, for the primary longitudinal outcome, we fitted a linear mixed-effects model testing the group-by-time interaction:

In this model, time indexed baseline and week-20 follow-up, group indexed MIG versus CTRL, baseline eTIV centered on the CTRL mean was included as a between-subjects covariate, and acquisition coil (Coil1/Coil2) was included as a time-varying covariate because 12 of 99 matched participants changed coil between acquisitions. The main effect of interest was the interaction, testing whether longitudinal volume changes differed between MIG patients and CTRL participants.

For the longitudinal models, raw, non-residualized hypothalamic volumes were used as dependent variables, rather than the eTIV-residualized volumes used in the cross-sectional analyses. This distinction reflects the different inferential focus of the two approaches: cross-sectional residualization removes between-subject variance associated with head size at a single timepoint, whereas the longitudinal analyses were designed to assess within-subject change over time. Applying separate control-based eTIV correction slopes at baseline and follow-up would require timepoint-specific regression estimates derived from finite and noisy samples. Any discrepancy between these estimates would be unlikely to represent a genuine biological change, as intracranial volume is not expected to vary meaningfully within an individual over a 20-week interval. Consistent with this assumption, eTIV was highly stable between baseline and follow-up in the matched longitudinal sample (n = 99; Pearson’s r = 0.996; mean difference = 0.06% of baseline eTIV; paired t(98) = 0.89, p = 0.38). We therefore included baseline eTIV directly as a covariate in the mixed-effects model, thereby accounting for between-subject differences in head size through a common adjustment across timepoints.

Additional analyses

Three additional analyses were conducted.

First, to verify robustness to coil changes, a sensitivity analysis excluding the 12 participants who changed coil between baseline and follow-up was conducted, yielding a reduced matched sample of 87 participants. Second, as a secondary analysis, the same group-by-time model was applied to all 10 side-specific hypothalamic subunit volumes, with p-values adjusted using the Benjamini–Hochberg FDR procedure. Third, as an exploratory analysis within the MIG group, we additionally tested whether the two treatment arms (MBSR+ vs. SMH) differed in their longitudinal trajectories on the primary outcome.

Models were estimated by restricted maximum-likelihood using lme4/lmerTest, with Satterthwaite degrees of freedom. Effect sizes were quantified as Hedges’ for paired observations.

Results

Sensitivity analyses of residual sex- and age-related effects after eTIV adjustment in CTRL groups

See Supplementary Materials, section ‘Sensitivity analyses of residual sex- and age-related effects after eTIV adjustment in CTRL groups Methods’, and Table SM3 for additional details. Briefly, before eTIV adjustment, raw hypothalamic volumes appeared to differ between sexes in several hypothalamic subunits, particularly in DS2-MIG; however, none of these differences reached conventional statistical significance, whereas age-related associations were generally limited. Following control-derived eTIV residualization, the apparent sex-related differences were attenuated and no significant sex-related and age-related associations remained in the considered CTRL samples. Overall, these findings indicate that eTIV residualization reduced sex-related and age-related volumetric scaling effects in the CTRL groups.

Cross-sectional group-comparison analyses

Bilateral hypothalamic subunit volumes: multivariate analyses (MANOVAs)

For the results of this section, see Table 3. Mean eTIV-adjusted bilateral hypothalamic subunit volumes are reported in Supplementary Table SM4. In DS1-MIG-base, the omnibus MANOVA on the 5 bilateral eTIV-adjusted hypothalamic subunits revealed a significant multivariate effect of group (MIG vs. CTRL) in the primary analysis [F(5, 124) = 3.43, p = .006]. Inclusion of biological sex as a covariate preserved the group effect [F(5, 123) = 3.40, p = .007], while sex itself did not significantly contribute to the multivariate model. Subject-specific age was not available for the whole dataset and could therefore not be included as a covariate.

Table 3.

Multivariate analysis of bilateral hypothalamic subunit volumes across datasets. For each dataset, the five bilateral eTIV-adjusted hypothalamic subunit volumes were tested jointly. Results from the primary outlier-handled models are reported; corresponding sensitivity analyses in the full uncleaned samples are provided in Supplementary table SM5. Covariate-adjusted models were fitted only when individual-level demographic information was available for the full analytic sample. Accordingly, biological sex was included in DS1-MIG-base, biological sex and age in DS2-MIG, and age in DS-FM. Biological sex was not included in DS-FM because all participants were female. No covariate-adjusted model was fitted for DS-TN because individual-level demographic information was not available for the full analytic sample. In the MANCOVA models, the notation “effect | covariate(s)” indicates the effect tested after adjustment for the listed covariate(s). Pillai’s trace is reported as the primary multivariate statistic

Cohort	Model	n (CTRL/PAT)	Pillai	F(df)	p
DS1-MIG-base	MANOVA	130 (31/99)	0.121	3.43 (5, 124)	0.006
	MANCOVA, group | sex	130 (31/99)	0.121	3.40 (5, 123)	0.007
	MANCOVA, sex | group	130 (31/99)	0.030	0.75 (5, 123)	0.587
DS2-MIG	MANOVA	52 (28/24)	0.283	3.63 (5, 46)	0.008
	MANCOVA, group | sex	52 (28/24)	0.298	3.82 (5, 45)	0.006
	MANCOVA, sex | group	52 (28/24)	0.035	0.33 (5, 45)	0.894
	MANCOVA, group | sex, age	52 (28/24)	0.336	4.44 (5, 44)	0.002
	MANCOVA, age | sex, group	52 (28/24)	0.117	1.16 (5, 44)	0.343
DS-FM	MANOVA	61 (31/30)	0.084	1.01 (5, 55)	0.418
	MANCOVA, group | age	61 (31/30)	0.088	1.04 (5, 54)	0.404
	MANCOVA, age | group	61 (31/30)	0.099	1.19 (5, 54)	0.325
DS-TN	MANOVA	143 (42/101)	0.043	1.22 (5, 137)	0.303

Abbreviations: CTRL = healthy controls; PAT = patients

In DS2-MIG, the omnibus MANOVA also showed a significant multivariate effect of group in the primary analysis [F(5, 46) = 3.63, p = .008]. Inclusion of biological sex as a covariate preserved and slightly strengthened the group effect [F(5, 45) = 3.82, p=.006], while sex itself did not significantly contribute to the model. Further inclusion of age as an additional covariate yielded a comparable pattern: the group effect was further strengthened [F(5,44) = 4.44, p = .002], while age did not significantly contribute to the model. Biological sex was also non-significant in the sex-adjusted model.

In the DS-FM dataset, the omnibus MANOVA revealed no significant multivariate group effect (FM vs. CTRL) in the primary analysis. The group effect remained non-significant after adjustment for age. Given the absence of sex variation (all participants were female), no MANCOVA with sex as a covariate was performed.

In the DS-TN dataset, the omnibus MANOVA revealed no significant multivariate effect of group (TN vs. CTRL) in the primary analysis. Because participant-level demographic information was incomplete for the full DS-TN analytic sample, no covariate-adjusted model was fitted.

Overall, the omnibus multivariate analyses revealed a significant group effect on bilateral hypothalamic subunit volumes in both migraine cohorts (DS1-MIG and DS2-MIG), but not in the fibromyalgia (DS-FM) or trigeminal neuralgia (DS-TN) cohorts.

Sensitivity analyses

In DS1-MIG-base, the bilateral multivariate group effect showed a non-significant trend in the full uncleaned sample [F(5,140)=2.25, p =.053], but was significant after adjustment for biological sex [F(5,138)=2.84, p =.018]. In DS2-MIG, the unadjusted group effect in the full uncleaned sample was not significant [F(5,53)=2.03, p =.089]. After adjustment for biological sex, the effect remained non-significant [F(5,52)=2.28, p =.060], but was significant when age was additionally included in the model [F(5,51)=2.42, p =.048]. In DS-FM, both the unadjusted and age-adjusted full uncleaned models remained non-significant. The full uncleaned DS-TN model was also non-significant.

For the results of this section, see Supplementary Table SM5

When the DS1-MIG-base analysis was restricted to participants scanned with Coil 1 only, the bilateral multivariate group effect remained significant [F(5, 112) = 3.21, p = .010], indicating that the main finding was not attributable to head-coil heterogeneity. For the results of this section, see Supplementary Materials, section ‘Subunit structural covariance analysis and clinical correlates’ and Table SM6.

Finally, in DS2-MIG, after the exclusion of migraine participants scanned within 3 days of their most recent attack, the bilateral multivariate group effect remained significant [F(5, 34) = 3.43, p = .013]. For the results of this section, see Supplementary Materials, section ‘Subunit structural covariance analysis and clinical correlates’ and Table SM7.

Bilateral hypothalamic subunit volumes: univariate follow-up analyses

For full results, see Supplementary Table SM8. In DS1-MIG-base, no bilateral hypothalamic subunit showed an FDR-significant group difference. In DS2-MIG, migraine patients showed significantly larger bilateral anterior-inferior and anterior-superior volumes than controls, with both effects surviving FDR correction (both ). No FDR-significant bilateral differences were observed in either DS-FM or DS-TN.

Left- and right-hemisphere subunit volumes: univariate follow-up analysis

For the results of this section, see Table 4 and Figs. 2, 3, and 4. To further characterize the hemisphere-specific anatomical expression of the bilateral multivariate group effect, we examined the 10 side-specific eTIV-adjusted hypothalamic subunit volumes across the different datasets.

Table 4.

Side-specific univariate follow-up analyses. Welch’s t-test on each eTIV-adjusted hypothalamic subunit by hemisphere, with Hedges’ g effect sizes (patients vs. Control; positive g indicates larger volume in patients) and 95% confidence intervals. is corrected within hemisphere (5 tests per family); is corrected across all 10 lateralized variables (sensitivity)

Cohort	Subunit	Side	n (CTRL/PAT)	g	95% CI
DS1-MIG-base	Anterior-inferior	L	35/106	-0.32	[-0.70, +0.06]	0.103	0.172	0.259
	Anterior-inferior	R	34/107	-0.15	[-0.53, +0.24]	0.382	0.828	0.637
	Anterior-superior	L	32/108	+0.45	[+0.06, +0.85]	0.005	.023	.045
	Anterior-superior	R	34/107	+0.11	[-0.27, +0.49]	0.540	0.828	0.675
	Posterior	L	34/107	+0.37	[-0.02, +0.75]	0.056	0.140	0.187
	Posterior	R	35/105	+0.52	[+0.13, +0.90]	0.013	0.066	0.066
	Inferior tubular	L	35/107	-0.13	[-0.51, +0.25]	0.515	0.515	0.675
	Inferior tubular	R	35/106	-0.08	[-0.46, +0.30]	0.677	0.828	0.753
	Superior tubular	L	35/107	+0.18	[-0.20, +0.56]	0.328	0.410	0.637
	Superior tubular	R	34/107	-0.03	[-0.42, +0.35]	0.828	0.828	0.828
DS2-MIG	Anterior-inferior	L	30/27	+0.38	[-0.14, +0.89]	0.160	0.401	0.321
	Anterior-inferior	R	31/27	+0.98	[+0.44, +1.52]	0.0005	.002	.005
	Anterior-superior	L	31/27	+0.70	[+0.18, +1.23]	0.009	.046	.031
	Anterior-superior	R	29/26	+0.53	[-0.001, +1.06]	0.056	0.094	0.140
	Posterior	L	31/26	+0.24	[-0.27, +0.76]	0.356	0.542	0.508
	Posterior	R	31/26	-0.02	[-0.53, +0.50]	0.949	0.949	0.949
	Inferior tubular	L	31/27	+0.20	[-0.31, +0.71]	0.434	0.542	0.542
	Inferior tubular	R	30/26	+0.77	[+0.23, +1.30]	0.004	.011	.021
	Superior tubular	L	30/27	+0.12	[-0.39, +0.64]	0.644	0.644	0.715
	Superior tubular	R	29/26	+0.29	[-0.24, +0.81]	0.286	0.358	0.477
DS-FM	Anterior-inferior	L	33/32	-0.15	[-0.63, +0.33]	0.549	0.958	0.958
	Anterior-inferior	R	33/32	-0.02	[-0.50, +0.46]	0.938	0.938	0.958
	Anterior-superior	L	33/32	-0.01	[-0.49, +0.47]	0.958	0.958	0.958
	Anterior-superior	R	33/31	-0.47	[-0.96, +0.02]	0.061	0.217	0.434
	Posterior	L	32/32	-0.12	[-0.61, +0.36]	0.627	0.958	0.958
	Posterior	R	32/32	-0.09	[-0.57, +0.40]	0.732	0.915	0.958
	Inferior tubular	L	33/32	+0.03	[-0.45, +0.51]	0.903	0.958	0.958
	Inferior tubular	R	33/32	+0.12	[-0.36, +0.60]	0.626	0.915	0.958
	Superior tubular	L	32/32	+0.15	[-0.34, +0.63]	0.546	0.958	0.958
	Superior tubular	R	32/30	+0.43	[-0.07, +0.93]	0.087	0.217	0.434
DS-TN	Anterior-inferior	L	45/108	-0.31	[-0.66, +0.03]	0.038	0.190	0.380
	Anterior-inferior	R	49/103	-0.20	[-0.54, +0.14]	0.241	0.320	0.427
	Anterior-superior	L	51/111	-0.15	[-0.48, +0.18]	0.342	0.571	0.438
	Anterior-superior	R	52/108	-0.16	[-0.49, +0.17]	0.351	0.351	0.438
	Posterior	L	52/111	-0.005	[-0.33, +0.32]	0.977	0.977	0.977
	Posterior	R	52/104	-0.21	[-0.55, +0.12]	0.199	0.320	0.427
	Inferior tubular	L	52/111	+0.23	[-0.10, +0.56]	0.135	0.338	0.427
	Inferior tubular	R	50/108	-0.17	[-0.50, +0.17]	0.256	0.320	0.427
	Superior tubular	L	52/107	-0.12	[-0.45, +0.21]	0.515	0.644	0.573
	Superior tubular	R	52/111	-0.20	[-0.53, +0.13]	0.208	0.320	0.427

Fig. 2.

Effect sizes for group differences in eTIV-adjusted hypothalamic subunit volumes across four cohorts. Forest plots show hedges’ g (patient minus control) with 95% CI for each subunit, separately for the left and right hemispheres, in the migraine (DS1-MIG-base and DS2-MIG), fibromyalgia (DS-FM), and trigeminal neuralgia (DS-TN) cohorts. Red dots indicate FDR-significant effects (); g > 0 reflects larger volumes in the patient group. Significant enlargements were restricted to the migraine cohorts: left anterior-superior in the DS1-MIG dataset, and left anterior-superior, right anterior-inferior, and right inferior tubular in the DS2-MIG dataset. No significant differences were observed in the DS-FM or DS-TN datasets. Abbreviations: Ant-Inf, anterior inferior; Ant-sup, anterior superior; tub-Inf, inferior tubular; tub-sup, superior tubular

Fig. 3.

Boxplots show the distribution of eTIV-residualized volumes (mm) for control (blue) and migraine (pink) groups in the DS1-MIG dataset and DS2-MIG dataset, separately for left and right hemispheres. Subunits reaching FDR significance are marked with ⋆. Abbreviations: Ant-Inf, anterior inferior; Ant-Sup, anterior superior; tub-Inf, inferior tubular; tub-Sup, superior tubular

Fig. 4.

Boxplots show the distribution of eTIV-residualized volumes (mm) for control (blue) and patients (pink) groups in the DS-FM dataset and DS-TN dataset, separately for left and right hemispheres. No subunit reached statistical significance in either cohort after FDR correction. Abbreviations: Ant-Inf, anterior inferior; Ant-Sup, anterior superior; tub-Inf, inferior tubular; tub-Sup, superior tubular

In DS1-MIG-base, the left anterior-superior subunit was significantly larger in MIG patients than in CTRL participants and survived both within-side and all-10 FDR correction (within-side ; all-10 ). No remaining left- or right-hemisphere subunit survived FDR correction.

In DS2-MIG, three side-specific subunits survived both within-side and all-10 FDR correction, all showing larger volumes in MIG patients than in CTRL participants. These included the right anterior-inferior subunit (within-side ; all-10 ), the right inferior tubular subunit (within-side ; all-10 ), and the left anterior-superior subunit (within-side ; all-10 ). No other side-specific subunit survived FDR correction.

In DS-FM, no side-specific hypothalamic subunit showed a significant group difference after either within-hemisphere or all-10 FDR correction. Results were non-significant for all subunits in both hemispheres, and all confidence intervals crossed zero.

Similarly, in DS-TN, no side-specific hypothalamic subunit showed a significant group difference after either within-hemisphere or all-10 FDR correction. All confidence intervals crossed zero.

Sensitivity analyses for the replicated left anterior-superior finding in migraine datasets

For additional details regarding the sensitivity analyses addressing the influence of extreme observations, see Supplementary Materials, section ‘Sensitivity analyses addressing the influence of extreme observations Methods’ and Table SM9. In the full uncleaned samples, the left anterior-superior subunit patient–control difference showed a consistent patient-greater-than-control direction across the two MIG cohorts, although the non-parametric within-cohort comparison reached significance in DS2-MIG (p =.039) but not in DS1-MIG-base (p = 0.136).

However, the pooled (DS1-MIG-base+DS2-MIG) full-sample analysis supported a significant diagnosis effect [F(1, 199) = 4.98, p = .027], without evidence of a diagnosis-by-dataset interaction. Following winsorization, the patient–control difference was significant within both cohorts (DS1-MIG-base, p =.024), DS2-MIG (p = .028) and remained significant in the pooled analysis [F(1, 199) = 7.44, p = .007], again with no diagnosis-by-dataset interaction. Overall, these findings indicate that the replicated left anterior-superior difference was not solely attributable to the primary outcome-specific censoring strategy.

Cross-cohort meta-analysis of the convergent subunit effect

A formal inverse-variance fixed-effect meta-analysis was performed across DS1-MIG and DS2-MIG for the left anterior-superior subunit. The effect was positive in both cohorts, with Hedges’ g estimates of 0.45 in DS1-MIG and 0.70 in DS2-MIG. The pooled effect estimate was Hedges’g = 0.54, 95% CI [0.22, 0.86], with no evidence of between-cohort heterogeneity, Q(1) = 0.54, p = .463, %. These findings support a consistent patient-greater-than-control left anterior-superior effect across the two independent migraine cohorts.

Subunit structural covariance analysis and clinical correlates

Exploratory structural covariance analyses using the left anterior-superior subunit as reference revealed no FDR-corrected group differences between MIG and CTRL in either DS1-MIG or DS2-MIG. See Supplementary Materials, section ‘Subunit structural covariance analysis and clinical correlates Methods’ and Table SM10.

Within DS2-MIG, left anterior-superior volume showed no significant association with attack frequency or pain intensity during attacks. Disease duration showed a positive trend that did not reach significance either at the uncorrected level or after FDR correction (ρ =.38, uncorrected p =.052, ).

Longitudinal analyses on the DS1-MIG dataset

For the results of this section, see Table 5 and Fig. 5. The MIG vs. CTRL group difference observed at baseline in the full DS1-MIG-base sample was also present, in the same direction and of comparable magnitude, within the matched longitudinal subsample [F(5,84) = 3.10, p = .013]. At the univariate level, the left anterior-superior difference was also significant [t(52.66) = 2.39, p =.021; Hedges’ g = 0.47], closely matching the effect observed in the full baseline sample (Hedges’ g = 0.45 - see Table SM11). These findings indicate that the longitudinal subsample preserved the principal baseline group difference (see Supplementary Table M11).

Table 5.

Longitudinal analysis of lateralized hypothalamic subunit volumes in DS1-MIG (DS1-MIG-base and DS1-MIG-long). Linear mixed model: volume∼time×group + eTIV + coil + (1|subject), estimated in 99 participants with paired baseline and follow-up acquisitions (24 controls, 75 migraineurs; follow-up interval: acq1 → acq3). Hedges’ quantifies paired within-group change (follow-up - baseline); positive values indicate volume increase over time. FDR correction (benjamini–Hochberg) applied across all 10 lateralized subunits for the group × time interaction term. The left anterior-superior subunit was the primary outcome; therefore, no FDR correction was applied to this test. Sensitivity analysis excluding the 12 participants who changed the acquisition coil between timepoints (n = 87: 23 controls, 64 migraineurs) yielded , [-0.13, +0.36], confirming robustness to the coil change. An exploratory analysis restricted to migraineurs tested the treatment-arm × time interaction for the primary outcome

Subunit	Side	n(CTRL/MIG)	Migraine		Control
				95% CI		95% CI
Primary outcome (no FDR)
Anterior-superior	L	24/75	+0.09	[-0.13, +0.32]	+0.18	[-0.23, +0.58]	0.754	—
Secondary outcomes (FDR corrected across 10 lateralized subunits)
Anterior-inferior	L	24/75	+0.08	[-0.15, +0.30]	+0.13	[-0.28, +0.53]	0.977	0.977
Anterior-inferior	R	24/75	+0.16	[-0.06, +0.39]	-0.09	[-0.49, +0.31]	0.195	0.651
Anterior-superior	R	24/75	+0.07	[-0.16, +0.29]	-0.04	[-0.44, +0.36]	0.570	0.815
Posterior	L	24/75	-0.14	[-0.37, +0.09]	+0.21	[-0.20, +0.61]	0.007	0.068
Posterior	R	24/75	-0.10	[-0.33, +0.12]	+0.08	[-0.32, +0.48]	0.170	0.651
Inferior tubular	L	24/75	+0.07	[-0.15, +0.30]	+0.07	[-0.33, +0.47]	0.904	0.977
Inferior tubular	R	24/75	+0.05	[-0.17, +0.28]	+0.19	[-0.22, +0.59]	0.421	0.751
Superior tubular	L	24/75	-0.12	[-0.35, +0.10]	-0.32	[-0.73, +0.09]	0.451	0.751
Superior tubular	R	24/75	-0.12	[-0.35, +0.11]	+0.04	[-0.36, +0.45]	0.338	0.751
Exploratory treatment-arm analyses on ANT-SUP L (migraineurs only)
MBSR+ (n = 37)			Δ = +0.07 mm (SE = 0.53), p = .893
SMH (n = 38)			Δ = +0.59 mm (SE = 0.50), p = .241
Treatment × time interaction			β = -0.52 mm (SE = 0.70), p = .460

Fig. 5.

Longitudinal changes in the left anterior-superior hypothalamic subunit volume. Raw volumes (mm) are shown at baseline and follow-up for the MIG (pink) and CTRL (blue) groups within the matched longitudinal sample (n = 99: 75 MIG and 24 CTRL). Standardized within-group changes were small in both MIG participants (, 95% CI [-0.13, +0.32]) and CTRL participants (, 95% CI [-0.23, +0.58]). The group-by-time interaction was not significant ()

The primary longitudinal analysis revealed no significant group-by-time interaction for the left anterior-superior subunit (p =.754), the primary outcome. Within-group standardized volume changes from baseline to follow-up were small in both MIG patients and CTRL participants, with confidence intervals including zero in both groups. Overall, these findings indicate that no differential longitudinal change in the left anterior-superior hypothalamic subunit was detected between migraineurs and controls over the available follow-up period.

The sensitivity analysis restricted to participants scanned with the same coil across acquisitions yielded the same null results (p =.873), indicating that the results were not driven by coil changes. The left posterior subunit showed the strongest exploratory signal, with a nominal group-by-time interaction that did not survive FDR correction across the 10 lateralized subunits (). However, within-group standardized changes were small, and their confidence intervals included zero, indicating no robust longitudinal change in either MIG patients or CTRL participants. The exploratory comparison of treatment arms showed no significant treatment-by-time interaction for the primary outcome (β = -0.52 mm, SE = 0.70, p = .46), indicating no evidence of differential longitudinal trajectories between MBSR+ and SMH.

Overall, these results indicate that the left anterior-superior hypothalamic enlargement observed in MIG patients at baseline was not accompanied by a detectable differential longitudinal change relative to CTRL participants over the available follow-up period.

Discussion

The present study examined volumetric changes in hypothalamic subunits in two independent cohorts of patients with episodic migraine (DS1-MIG and DS2-MIG), each compared with its respective CTRL group.

To examine whether comparable hypothalamic alterations were observed in other chronic pain conditions, including a widespread pain disorder and a trigeminal pain disorder, we further analysed two clinical pain comparison cohorts comprising patients with fibromyalgia (DS-FM) and primary trigeminal neuralgia (DS-TN).

In addition, to assess the stability of the observed hypothalamic alterations over time, we examined longitudinal MRI data available for a subsample of DS1-MIG participants.

Three main findings emerged. First, both migraine cohorts showed significant multivariate differences in hypothalamic subunit volume profiles relative to healthy controls, whereas no effects were observed in the fibromyalgia or trigeminal neuralgia cohorts. Second, lateralized follow-up analyses identified a reproducible enlargement of the left anterior-superior hypothalamic subunit across the two migraine datasets. A formal cross-cohort meta-analysis further supported this finding, yielding a significant pooled effect of moderate magnitude with no evidence of between-cohort heterogeneity. Third, in the longitudinal subsample of DS1-MIG, this left anterior-superior enlargement was not accompanied by a detectable differential longitudinal change relative to CTRL participants over the available follow-up period, despite patients undergoing treatment interventions associated with reduced headache attack frequency.

These results were broadly consistent across several sensitivity analyses confirming that the eTIV residualization procedure adequately accounted for potential age- and sex-related effects, and assessing the potential influence of coil type in DS1-MIG (both cross-sectionally and longitudinally) and peri-ictal status in DS2-MIG.

By contrast, in the full uncleaned samples, the bilateral multivariate group effect did not reach significance in either DS1-MIG-base or DS2-MIG. Covariate-adjusted full-sample models, however, reached significance in DS1-MIG-base after adjustment for biological sex and in DS2-MIG after adjustment for biological sex and age. Following application of the hybrid outlier-handling procedure, the omnibus effects were significant in both cohorts. Overall, these findings indicate that the multivariate signal showed some sensitivity to both extreme observations and covariate adjustment.

Importantly, the replicated left anterior-superior finding was also examined using complementary analyses addressing the influence of extreme observations. Although the full uncleaned within-cohort evidence varied in statistical strength, the patient-greater-than-control pattern was directionally consistent across both MIG cohorts. Importantly, the pooled ANOVA of the full uncleaned data showed a significant diagnosis effect, with no evidence of a diagnosis-by-dataset interaction. Following winsorization, the difference was significant within both cohorts and was again supported by a significant pooled ANOVA, without evidence that the estimated diagnosis effect differed between datasets. These findings indicate that the replicated left anterior-superior effect is unlikely to be attributable solely to the primary outcome-specific censoring strategy.

Taken together, these findings indicate hypothalamic macrostructural alterations in episodic migraine, while refining current models of hypothalamic involvement by suggesting that these changes are not expressed as a global hypothalamic abnormality, but rather as a region-specific alteration localized to the left anterior-superior hypothalamic subunit, with the longitudinal findings being compatible with persistence over the observed follow-up period. Comparable alterations were not observed in the available fibromyalgia and trigeminal neuralgia datasets, providing supportive, but not definitive, evidence that the observed pattern may not reflect non-specific effects of chronic pain or trigeminal pain.

These results align with the functional neuroimaging evidence summarized in the Introduction implicating the hypothalamus across the migraine cycle, and more specifically with studies pointing to anterior hypothalamic involvement. Stronger anterior hypothalamic activation has been reported in chronic compared with episodic migraine during painful trigeminal stimulation, whereas more posterior regions appeared more closely related to the acute headache state [43], with increased anterior hypothalamus–spinal trigeminal nucleus connectivity also reported in chronic migraine [44]. Notably, a 30-day longitudinal neuroimaging study localized pre-attack activation maxima to the lateral anterior hypothalamus [45], broadly consistent with the anterior subunit identified here, albeit with opposite lateralization.

Against this functional background, the interpretation of our findings becomes more complex when considered alongside recent structural investigations using the same hypothalamic segmentation approach. Giardina et al. [16] found no significant volumetric differences in episodic migraine without aura during strictly interictal periods, while also controlling for menstrual-cycle phase by scanning female participants at mid-cycle. Xiong et al. [17] reported hypothalamic volume reductions primarily in chronic migraine patients, specifically in the right superior tubular subunit and in inferior hypothalamic regions, including the left inferior tubular and right anterior-inferior subunits. Descriptive analyses suggested a gradual decrease in hypothalamic volumes from controls to episodic and chronic migraine patients within the inferior tubular subunit, although the episodic migraine group did not reach statistical significance, likely due to limited sample size. Notably, in the same study, a Mendelian randomization analysis suggested that smaller volumes of the inferior tubular subregion are linked to increased migraine risk.

These marked discrepancies with our results may reflect both clinical and methodological differences across cohorts. From a clinical perspective, differences in disease burden and timing relative to the migraine cycle may influence the detectability and direction of hypothalamic volumetric alterations. Direct comparisons of disease burden across the reported studies [16, 17] and the datasets employed in the present investigations (i.e., DS1-MIG and DS2-MIG) are limited by heterogeneous clinical reporting. Notably, only Giardina et al. [16] strictly controlled for timing relative to the migraine cycle, excluding patients who experienced an attack within three days before or after the MRI, and for hormonal status, scanning female participants at mid-cycle. From a methodological perspective, it is plausible that the small size of hypothalamic subunits makes volumetric estimates sensitive to MRI acquisition protocol and quality-control procedures.

Although MRI segmentation cannot resolve individual hypothalamic nuclei with histological precision, the anterior-superior hypothalamic subunit is anatomically associated with the paraventricular nucleus (PVN) and the preoptic area [18].

The PVN projects to trigeminal nociceptive pathways and has been implicated, via oxytocinergic signaling, in trigeminovascular pain modulation [46–49]. Although acute peripheral oxytocin receptor blockade did not increase migraine-like attacks in a recent provocation study, this does not exclude a modulatory role of oxytocin within the trigeminovascular system, given the short blockade duration and uncertain central availability of the antagonist [50]. This anatomical correspondence should be regarded as a hypothesis-generating consideration rather than a mechanistic conclusion, since MRI-based segmentation cannot resolve individual nuclei, cell types, or neurochemical signaling.

The biological interpretation of hypothalamic subunit enlargement remains necessarily cautious, as T1-weighted MRI cannot distinguish among underlying cellular or tissue-level mechanisms. In other contexts, such as elevated body mass index (BMI), larger hypothalamic volume has been discussed as potentially compatible with inflammation-related mechanisms, including gliosis, astrocytic hypertrophy, microglial proliferation, vascular changes, or blood-brain barrier alterations [51].

Moreover, the consistent left-lateralization of our findings should be interpreted cautiously. Although sample-specific influences cannot be entirely excluded, the independent replication of the left-sided effect across two datasets substantially reduces the likelihood that it reflects an idiosyncratic feature of a single cohort. Nevertheless, acquisition- and segmentation-related variability cannot be entirely excluded as potential contributors to the observed lateralized pattern. From a clinical perspective, the meaning of this lateralized structural finding could not be established with the data available in the present study. Information on migraine pain laterality, aura status, handedness, menstrual-cycle phase, and attack timing was not systematically available across the migraine cohorts, precluding a direct test of whether the left-sided hypothalamic enlargement relates to specific clinical features. From a biological perspective, emerging evidence suggests that PVN neuropeptidergic organization more broadly may exhibit side- and subregion-specific asymmetries, including asymmetric coordination of vasopressin neuronal populations [52]. This evidence provides a potential biological context for, but does not explain, the left-sided pattern observed in the present study.

Importantly, the replicated enlargement of the lateralized anterior-superior hypothalamic subunit observed here converges with previous evidence from chronic cluster headache, where an enlargement of the same subunit ipsilateral to the pain was reported [21]. This convergence is notable because, although migraine and cluster headache are distinct primary headache disorders, both show prominent hypothalamic involvement related to attack cyclicity, autonomic-endocrine regulation, and homeostatic modulation [6, 53, 54]. At the same time, the clinical interpretation differs across disorders: the anterior-superior alteration was observed in chronic cluster headache but not in episodic cluster headache, whereas in our study, it was detectable in episodic migraine.

Perhaps most importantly, the absence of comparable hypothalamic alterations in fibromyalgia and, especially, trigeminal neuralgia provides supportive evidence that the observed enlargement may not reflect a generic hypothalamic alteration related to chronic pain or trigeminal pain per se. The trigeminal neuralgia results are particularly informative for interpreting the migraine findings because, although both conditions involve the trigeminal system, they differ substantially in their broader pathophysiological architecture. Migraine is characterized by cyclic fluctuations, premonitory symptoms, and autonomic and neuroendocrine manifestations, all of which strongly implicate hypothalamic circuits [6]. In contrast, trigeminal neuralgia is primarily characterized by recurrent severe paroxysmal pain occurring within the trigeminal nerve distribution, reflecting involvement of the trigeminal somatosensory system [27]. Thus, the absence of similar hypothalamic changes in trigeminal neuralgia supports the interpretation that our findings may be more closely associated with migraine-related hypothalamic mechanisms than with trigeminal nociception alone, although these findings do not establish diagnostic specificity.

Beyond the replicated left anterior-superior effect, the DS2-MIG cohort also showed additional enlargements of the right anterior-inferior and right inferior tubular subunits. Because these additional effects did not replicate in DS1-MIG, they should be interpreted cautiously, in particular for the anterior-inferior subunit, which showed the lowest reproducibility among all the subunits in the original validation study [18]. As discussed above in relation to discrepancies with previous studies, between-cohort differences in clinical characteristics (such as migraine frequency, attack duration, and disease duration), timing relative to the migraine cycle, and MRI acquisition parameters may increase variability in hypothalamic subunit estimates and may partly explain why these additional effects were observed only in DS2-MIG. Thus, unlike the left anterior-superior enlargement, which was observed across both migraine cohorts, the DS2-MIG-specific anterior-inferior and inferior tubular effects should be considered exploratory until replicated in larger independent samples.

The absence of detectable differential longitudinal change in the DS1-MIG dataset does not resolve whether the left anterior-superior hypothalamic enlargement represents a consequence of migraine or a predisposing factor. One possibility is that recurrent migraine attacks, or migraine-related homeostatic and autonomic dysregulation more broadly, induce long-lasting structural adaptations in this region. Alternatively, this alteration may precede migraine onset and reflect a vulnerability marker, whereby individuals with specific hypothalamic macrostructural features are more susceptible to developing migraine. Because the available longitudinal data were acquired after disease onset and did not include pre-morbid imaging, the present study cannot distinguish between these scenarios. Although this finding does not establish biological stability, the absence of detectable differential longitudinal change over the available follow-up period is compatible with a relatively persistent migraine-related structural difference.

Particularly relevant to the present findings, a recent study in chronic insomnia disorder [22] identified enlargement of the left anterior-superior hypothalamic subunit using the same automated hypothalamic segmentation framework. The authors further reported associations between hypothalamic hypertrophy, symptom severity, and circulating CRH levels, highlighting a potential role of HPA-axis dysfunction. Given the well-established links between migraine, sleep disturbances, stress responsivity, and neuroendocrine regulation, these observations are consistent with the hypothesis that enlargement of the anterior-superior hypothalamic territory may reflect alterations in stress–homeostatic systems.

Limitations

Several limitations should be acknowledged.

First, we were not able to assess the potential effects of medication use and headache burden due to the limited availability and heterogeneity of clinical information across the migraine datasets. Regarding medication, the original DS1-MIG study [28] reported that only a minority of patients were using preventive migraine treatments at baseline, and individuals using opioid medications were excluded. In addition, in our cohort (DS2-MIG), participants were not receiving prophylactic medications and had not taken acute migraine treatments in the 24 hours preceding MRI acquisition. Altogether, these observations reduce the likelihood that the observed hypothalamic alterations were primarily driven by pharmacological effects. Regarding headache burden, it could not be harmonized or jointly modelled as a covariate across the two migraine cohorts due to differences in data granularity and metric definition. For DS2-MIG, individual-participant headache frequency (attacks/month) was available and is reported in Table 2. For DS1-MIG, however, only group-level, treatment-arm-aggregated headache-day reductions were available from the original clinical trial publication [28], precluding individual-level analysis or direct comparison with the attack-frequency metric used in DS2-MIG.

However, the presence of left anterior-superior enlargement in the same direction in both migraine cohorts (DS1-MIG-base and DS2-MIG), despite likely differences in clinical burden, supports the reproducibility of this group-level alteration across cohorts. In the longitudinal DS1-MIG data, no differential change in this subunit was detected between migraine patients and controls over the available follow-up period, despite reductions in mean headache days from approximately 7.8 to 4.6 days per 28-day period in the MBSR+ group and from 7.7 to 6.0 days in the SMH group between baseline and week 20. Nevertheless, these findings do not establish that the alteration is biologically stable, trait-like, or independent of current headache burden. In addition, no detectable differences in longitudinal volumetric change were observed between the two treatment groups, despite their different post-treatment headache burden.

Second, the timing of MRI acquisition relative to the migraine cycle was not available for the DS1-MIG dataset. This represents a relevant limitation, as hypothalamic activity and connectivity are known to fluctuate across the migraine cycle. Importantly, in the DS2-MIG dataset, no migraine participant was scanned during an ongoing attack, and sensitivity analyses (excluding patients with an attack occurring 3 days before the MRI) supported the consistency of the reported findings. However, headache occurrence after the MRI session was not systematically assessed; therefore, we cannot exclude that some participants classified as outside the prespecified peri-ictal window may have been scanned during a premonitory phase.

Third, BMI and other metabolic variables were not available and were therefore not included as covariates. This represents a relevant limitation because the hypothalamus is strongly involved in energy balance, appetite regulation, endocrine function, and metabolic homeostasis. Accordingly, we cannot exclude the possibility that unmeasured metabolic factors may have contributed to inter-individual variability in hypothalamic subunit volumes. In addition, information on sleep, stress, circadian factors, hormonal status and menstrual-cycle phase, as well as clinically relevant variables such as disease duration, migraine frequency, aura status, and preventive and acute medication use, was not consistently available or harmonizable across datasets. Although eTIV adjustment addressed head-size-related scaling and the available sex- and age-adjusted analyses provided some reassurance regarding the influence of sex and age where these variables were available, these analyses could not account for the potential influence of the other unmeasured or incompletely characterized factors.

Fourth, information on menstrual-cycle phase and hormonal contraceptive use was not systematically available across all cohorts, representing a relevant limitation given the predominantly female composition of the migraine patient groups (87% female in DS1-MIG and 75% female in DS2-MIG) and the exclusively female composition of DS-FM. Although direct evidence for menstrual-cycle-phase-related changes in hypothalamic volume remains limited, neuroimaging studies suggest that the hormonal milieu may influence hypothalamic measures. Transient hypothalamic microstructural and metabolic changes have been reported during an artificial menstrual cycle [55], while oral contraceptive use has been associated with smaller hypothalamic volume in healthy women [56]. Therefore, unmeasured variation in menstrual-cycle phase and hormonal status may have contributed to variability in hypothalamic measures in the present study.

Finally, inference regarding diagnostic specificity is limited by the heterogeneity of the available clinical pain comparison datasets. The fibromyalgia and trigeminal neuralgia cohorts differed from the migraine datasets in demographic composition, clinical characterization, MRI acquisition parameters, and covariate availability. In addition, each patient group was compared with its own healthy control group; thus, no formal diagnosis-by-group interaction was tested. Accordingly, the absence of comparable alterations in the available fibromyalgia and trigeminal neuralgia datasets should be interpreted as supportive, but not definitive, evidence for a migraine-related pattern.

Conclusion

In conclusion, our findings indicate that episodic migraine is associated with a reproducible, subunit-specific enlargement of the left anterior-superior hypothalamus. Comparable effects were not observed in the available fibromyalgia and trigeminal neuralgia datasets, providing supportive, but not definitive, evidence that this alteration may not simply reflect chronic pain or trigeminal involvement more broadly. Rather, it may represent a migraine-related macrostructural alteration of an anterior hypothalamic territory involved in homeostatic, autonomic, endocrine, and trigeminovascular regulation. Its replication across two independent migraine cohorts, together with the absence of detectable differential longitudinal change over the available follow-up period, supports further investigation in harmonized multi-diagnostic datasets. Importantly, the present findings do not establish diagnostic specificity or individual-level predictive utility, which will require formal comparative and predictive analyses.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Authors' contributions. SF and DL conceived and designed the study, performed the primary analyses, and drafted the manuscript. KK, BB, and DY supervised the work and critically reviewed the analyses and the manuscript. SP, LY, XW, and BH contributed to participant recruitment, data acquisition, and analyses. All authors read and approved the final manuscript.

Funding

This study was supported by the National Natural Science Foundation of China (NSFC; U24A20274) and the Sichuan Science and Technology Program (2024ZDZX0014).

Data availability

Data are available upon request from the corresponding authors.

Declarations

Competing interests

The authors declare no competing interests.