Genome-wide association analysis of treatment response to onabotulinumtoxinA in Han Chinese patients with chronic migraine: a pilot study

Abstract

Background/Objectives

OnabotulinumtoxinA (BoNT-A) use is an established preventive treatment for chronic migraine (CM); however, individual response rates vary. Although genetic factors may contribute to this variability, the underlying genetic determinants remain largely undefined, especially in Asian populations. Therefore, we aimed to identify genetic variants associated with the response to BoNT-A treatment in Han Chinese patients with CM.

Methods

We conducted a genome-wide quantitative trait locus (QTL) analysis using the Taiwan Precision Medicine (TPM) array in Han Chinese patients with CM receiving BoNT-A treatment according to the PREEMPT protocol. Treatment efficacy was assessed by calculating the improvement in monthly headache days after 24 weeks compared with baseline. Genotype–phenotype associations were assessed using a linear regression model adjusted for relevant clinical covariates. Functional annotation, Gene Ontology (GO) annotation, and linkage disequilibrium (LD) mapping were performed to interpret the biological importance of significant variants.

Results

We identified four intronic single nucleotide polymorphisms (SNPs) that showed genome-wide significance: rs11147666 in TRPC4, rs3924647 in LINC02197, rs56381512 in GTF2H2, and rs13126813 in SPOCK3. rs11147666 in TRPC4 showed the strongest association with treatment response. Patients with heterozygous variants showed lower improvement rates than those with common homozygotes. The implicated genes function in calcium ion transport, transcriptional regulation, and extracellular matrix interactions. Given the modest sample size and low minor allele frequencies, these findings should be interpreted as exploratory signals.

Conclusion

This pilot genome-wide association study identified genetic variants linked to variability in BoNT-A response among Han Chinese patients with chronic migraine, highlighting potential candidate loci for future validation. Given the limited sample size and absence of an independent replication cohort, the reported loci should be interpreted as exploratory and hypothesis-generating findings.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12883-026-04878-6.

Introduction

Migraine, a debilitating neurological condition consistently ranked among the most disabling disorders worldwide, imposes considerable personal, social, and economic burdens [1]. Chronic migraine (CM) is characterized by headache occurring for 15 or more days per month for over 3 months, with migraine symptoms present for at least 8 days, severely disrupting daily functioning [2]. The associated symptoms, including nausea, vomiting, and hypersensitivity to light and sound, further diminish patient quality of life [3]. CM affects approximately 1%–2% of the general population globally and 7%–8% of individuals with migraine, imposing a threefold higher economic burden than episodic migraine because of medical costs and lost productivity [4]. In East Asian populations, including Han Chinese, the prevalence of CM is slightly lower at 0.6%–1.7%, although the clinical characteristics and disability burden appear to be consistent across ethnicities [5, 6].

OnabotulinumtoxinA (BoNT-A) was approved by the Food and Drug Administration for CM prevention in 2010, following the publication of results of the pivotal PREEMPT clinical trials, which demonstrated a significant reduction in headache frequency and improved quality of life in patients with CM [7]. Although its initial pain-relieving properties were thought to stem from muscle relaxation, multiple mechanisms underlying its efficacy have since been identified [8]. These mechanisms include inhibition of neurotransmitter and neuropeptide exocytosis in both motor and sensory pathways; suppression of proinflammatory mediator production; reduction in the secretion of excitatory neuropeptides, such as substance P, calcitonin gene-related peptide (CGRP), and glutamate; and interference with the soluble N-ethylmaleimide-sensitive factor attachment protein receptor (SNARE) [9].

Clinical evidence consistently supports the efficacy of BoNT-A in managing CM. The landmark PREEMPT trials established its efficacy and tolerability, demonstrating a significant decrease in the number of headache days over 24 weeks alongside meaningful improvements in patient-reported outcomes [10]. Real-world evidence from long-term observational studies has confirmed sustained benefits, including reduction in the number of monthly headache days (MHDs) and acute medication use, and improvement in daily functioning [11]. A meta-analysis of 17 randomized controlled trials with 3,646 patients further validated these findings, showing a significant reduction in CM frequency and enhanced quality of life after 3 months of treatment [12].

Despite established clinical efficacy, patient responses to BoNT-A remain heterogeneous. Studies have indicated that approximately 65%–74% of patients respond to treatment, leaving a substantial proportion of patients with minimal to no improvement [11]. A 3-year follow-up study reported that although 86.1% of patients continued treatment, their response rates varied considerably, highlighting the need for individualized treatment strategies [13]. This variability suggests that genetic factors influence treatment outcomes.

Recent advances in genomics have substantially enhanced our understanding of the genetic basis of CM. Large-scale genome-wide association studies have identified more than 100 genetic loci associated with migraine susceptibility [14]. Notably, polymorphisms in genes associated with CGRP signaling, pain perception, and neurotransmitter synthesis have been implicated in the development and persistence of CM [6]. Although pharmacogenomic studies have examined the relationship between genetic variations and response to CGRP-targeting therapies [15, 16], research on the genetic basis of BoNT-A responsiveness remains limited, and the findings are inconclusive. A previous study identified two single nucleotide polymorphisms (SNPs) potentially linked to BoNT-A efficacy: CALCA rs3781719, associated with CGRP regulation, was more frequent in non-responders; TRPV1 rs222749, involved in pain modulation, correlated with differential responses [17]. Additionally, elevated baseline levels of CGRP and pentraxin 3 (PTX3), an inflammatory marker, have been linked with improved outcomes, suggesting roles in both neurogenic and inflammatory mechanisms [18]. However, genetic associations with BoNT-A efficacy remain inconsistent across studies.

The present study represents the first genome-wide investigation of genetic factors associated with response to BoNT-A therapy in patients of Han Chinese ancestry with CM. We aimed to identify genetic variants associated with treatment outcomes and to explore their potential biological implications through pathway analysis, with the longer-term goal of informing future biomarker research for personalized treatment approaches.

Materials and methods

Standard protocol approval and patient consent

The study protocol was approved by the TSGH Institutional Review Board (TSGHIRB: 2-108-05-038). All individuals provided signed informed consent before enrolment.

Study design and participants

This prospective observational study was conducted at the Headache Clinic, Department of Neurology, Tri-Service General Hospital, Taipei, Taiwan. A total of 64 consecutive Han Chinese patients with CM were enrolled between 2021 and 2024. After quality control procedures, 63 participants were included in the final genetic analysis. Each participant was followed for 24 weeks (two treatment cycles) after initiation of BoNT-A prophylaxis, with clinical assessments at baseline and 12-week intervals.

A clinical and genetic analysis was performed on a cohort of Han Chinese patients with CM from the neurology outpatient department of a tertiary medical center in Taiwan. The eligibility criteria were as follows: (1) age 18 years or older, (2) CM diagnosed based on the International Classification of Headache Disorders, 3rd edition (ICHD-3) criteria (headache occurring for ≥ 15 days per month for > 3 months, with migraine features present for ≥ 8 days) [2], and (3) Han Chinese ancestry. Patients with secondary headaches or other coexisting primary headache disorders were excluded.

Each participant underwent a comprehensive screening process, including a standardized questionnaire survey and evaluation by a board-certified neurologist and headache specialist (FCY) [16]. Patients completed a demographic survey along with the Migraine Disability Assessment (MIDAS) questionnaire to evaluate headache-related disability [19]. Comprehensive clinical data were collected, including patient age, sex, body mass index, education level, migraine duration, presence of an aura, medication overuse headache (MOH), previous preventive treatment failures, triptan response, monthly migraine days (MMDs), MHDs, and MIDAS scores.

Prior medication exposure was systematically collected from medical records and patient interviews. During the 24-week follow-up period, no preventive medications were newly added or changed. MHD and MMD were captured using prospective patient headache diaries, and participants were instructed to record their headache status daily throughout follow-up. All participants had failed multiple guideline-based preventive therapies prior to BoNT-A initiation. Commonly used preventive agents included topiramate, valproic acid, beta-blockers, flunarizine, and tricyclic antidepressants. Acute medications included triptans and NSAIDs.

BoNT-A therapy protocol

All patients received standardized BoNT-A treatment (BOTOX; Allergan/AbbVie, Irvine, CA, USA) according to the PREEMPT fixed-site/fixed-dose protocol. A total dose of 155 units was administered via intramuscular injection across 31 fixed injection sites in seven head and neck muscle regions. No additional follow-the-pain injections were used in this cohort. Each session comprised 31 injections, with 0.1 mL (5 units) per injection site. The targeted injection sites were the forehead, temples, bridge of the nose, occipital region, neck, and upper shoulders. Injections were administered using a fine needle to minimize discomfort.

The treatment was administered by experienced neurologists specialized in headache medicine to ensure consistency. Treatment sessions were scheduled every 12 weeks, with an initial assessment conducted after two treatment cycles to evaluate efficacy. The primary endpoint was the mean reduction in MHD from the baseline to week 24 after two cycles of BoNT-A treatment. Retreatment was scheduled every 12 weeks based on individual response, and follow-up visits were conducted at regular intervals to monitor efficacy and adjust therapy as needed. Treatment efficacy was assessed by calculating the improvement in MHD from baseline to 24 weeks, using the following formula: Improvement rate = (baseline MHD - post-treatment MHD) / baseline MHD × 100. For the genome-wide QTL analysis, the improvement ratio (0–1 scale) was used as the quantitative phenotype; for clinical summaries, results are also reported as percentages.

DNA extraction and genotyping

Peripheral blood samples (5 mL) were collected from all participants in EDTA tubes. Genomic DNA was extracted and purified using the QIAsymphony SP system (QIAGEN, Hilden, Germany). DNA quality was assessed, and samples with 260/280 ratios between 1.8 and 2.0 were selected for genotyping. Genome-wide genotyping was performed using the Taiwan Precision Medicine (TPM) array chip [20], which contains markers optimized for Han Chinese populations. Signals were detected using the Axiom GeneTitan platform (Thermo Fisher Scientific, Sunnyvale, CA, USA). SNP data were subjected to quality control, SNP calling, and sample annotation using the Axiom Analysis Suite (Thermo Fisher Scientific). Quality control procedures were applied to both samples and SNPs. Samples with call rates < 95%, excessive heterozygosity, or evidence of relatedness or population stratification were excluded. SNPs were filtered for call rates < 80%, minor allele frequency < 1%, and significant deviation from the Hardy–Weinberg equilibrium (p < 1 × 10^{− 6}).

Quantitative Trait Locus (QTL) analysis

To investigate the association between improvement rate and potential risk SNPs, QTL analysis was performed using a linear regression model in the PLINK 1.9 software [21]. The MHD improvement rate served as the quantitative phenotype, and clinical variables—including sex, age, aura, MMD, MHD, MOH, triptan response, preventive treatment failure, migraine duration—were included as covariates in the regression analysis. MIDAS score was not included in the multivariable models because it was not available for all participants. To evaluate robustness and potential overfitting, we performed prespecified sensitivity analyses using three models with increasing levels of covariate adjustment (Table S1). The steps involved in the QTL analysis are illustrated in Fig. 1. Associations were determined using linear regression with an additive genetic model; a genome-wide significance threshold of p < 5 × 10^{− 8} was applied to control for multiple testing.

Fig. 1.

Study flowchart. This flowchart outlines the analytical workflow, including participant recruitment, TPM-based genotyping, QTL analysis using PLINK 1.9, and post-association steps such as Gene Ontology annotation and linkage disequilibrium analyses

Variant annotation and functional analysis

Significant variants (p < 5 × 10^{− 8}) obtained from QTL analysis and their mapped genes were annotated using the RefSeq Database from Wannovar [22]. Allele frequencies across diverse populations (East Asian and Taiwanese) were assessed using the 1000 Genomes Project [23] and Taiwan BioBank [24]. To provide functional context for the identified associations, we queried the Genotype-Tissue Expression (GTEx) database to examine whether the lead variants were associated with gene expression in brain tissues [25].

The biological implications of identified genes were explored through Gene Ontology (GO) annotation using the DAVID Bioinformatics Resource v6.8 [26]. This analysis revealed descriptive biological processes and molecular functions for associated genes. We also examined the potential functional consequences of the identified variants using in silico prediction tools such as the FORGEdb score to assess regulatory potential based on gene expression patterns in relevant tissues using publicly available databases [27].

Linkage Disequilibrium (LD) analysis

LD analysis was performed to identify genes likely to be functionally relevant to the treatment response and to determine the number and location of contributing genes. The significant variants identified through the QTL analysis were uploaded to the LDproxy Tool, a part of LDlink (https://ldlink.nih.gov/?tab=home). The analysis parameters included the GRCh37 genome build version, Han Chinese in Beijing (CHB) and Southern Han Chinese (CHS) populations, LD measurements D prime (D’) and R-squared (R²) based on allele frequencies, a 10,000-bp window, and regulatory potential predictions from FORGEdb [27]. A 10,000-bp window was selected to capture a broad spectrum of regulatory elements and interactions, providing a detailed map of the genomic landscape surrounding the variants. Pairwise LD was estimated using the r² statistic, with heatmaps generated to visualize the LD patterns. Haplotype association analysis was performed to assess whether specific haplotypes showed stronger associations with treatment outcomes than individual SNPs.

Statistical analysis

Demographic and clinical characteristics were compared between the medication overuse and non-medication overuse groups using the Mann–Whitney U test for continuous variables and Fisher’s exact test for categorical variables, given the modest sample size. Genome-wide QTL analysis was performed in PLINK 1.9 using linear regression under an additive genetic model, with the MHD improvement rate as the quantitative phenotype. Results were visualized using circular Manhattan plots and quantile-quantile (Q-Q) plots generated using rMVP [28] to assess potential genomic inflation. The genomic inflation factor (λ) was calculated to quantify possible population stratification effects. For significantly associated SNPs, improvement rates were visualized using scatter plots, and Wilcoxon rank-sum tests were used only for secondary descriptive comparisons of genotype-group distributions, given the non-normal distribution of improvement rates. These non-parametric comparisons were not used for the genome-wide discovery analysis. The relationship between genotype and triptan response was also investigated to explore potential cross-treatment genetic influences. All statistical analyses were performed using R version 4.0.3 (R Core Team, 2020; Vienna, Austria). Statistical significance was defined as a two-sided p-value < 0.05 unless otherwise specified. For genome-wide analyses, the significance threshold was set at p < 5 × 10⁻⁸. To evaluate the robustness of the associations to potential influential observations-particularly given the modest effective sample size (n = 63) and low minor allele frequencies of the lead SNPs, robust linear regression was performed using the robustbase package in R. Statistical power estimates of the significant SNPs were calculated using the R package genpwr.

Results

Demographic and clinical characteristics

A total of 63 Han Chinese patients with CM were enrolled, including 52 women (82.5%) and 11 men (17.5%) (Table 1). The median age was 50.0 years (IQR, 44.5–60.5). MOH was observed in 52 patients (82.5%); patients with and those without MOH did not differ significantly in terms of age, sex, migraine duration, MMD, MHD, or improvement rate. Migraine with aura was reported by six patients (9.5%). Baseline disease burden was substantial, with a median MMD of 19.0 (IQR, 17.5–30.0), a median MHD of 24.0 (IQR, 18.5–30.0), and a median MIDAS score of 20.0 (IQR, 11.8–45.3). Twenty-seven patients (42.9%) had failed ≥ 3 prior preventive therapies. The mean baseline MHD was 23.9 ± 6.2, and the mean week-24 MHD was 6.5 ± 6.3, corresponding to a mean MHD improvement rate of 70.6% ± 27.2%. Observation of the improvement-rate distribution is shown as a histogram in Supplementary Figure S1, with a mean value of 0.71 and a skewness of -1.49, indicating a negatively skewed distribution with a larger proportion of patients showing higher improvement rates.

Table 1.

Demographic and clinical data

Patient Characteristic	All chronic migraine	Chronic migraine with medication overuse	Chronic migraine without medication overuse	P-value
Age (years)	50.0 (44.5–60.5)	50.0 (45.0–60.3)	53.0 (41.0–60.0)	0.638
Female/Male	52 (82.5%)/11 (17.5%)	43 (82.7%) / 9 (17.3%)	9 (81.8%) / 2 (18.2%)	1.000
Aura	6 (9.5%)	5 (9.6%)	1 (9.1%)	1.000
Triptan Responders	35 (55.6%)	28 (53.8%)	7 (63.6%)	0.795
Preventive Treatment Failure > 3	27 (42.9%)	25 (48.1%)	2 (18.2%)	0.138
Body Mass Index	22.58 (20.13–25.28)	22.76 (20.08–25.57)	22.54 (21.31–23.07)	0.488
Education (years)	14.0 (12.0–16.0)	14.0 (12.0–16.0)	16.0 (14.0–16.0)	0.260
Improvement Rate (%)	80.6 (62.8–89.7)	80.0 (62.2–89.7)	85.7 (75.5–88.8)	0.568
MMD	19.0 (17.5–30.0)	19.5 (18.0–30.0)	19.0 (16.5–27.5)	0.406
MHD	24.0 (18.5–30.0)	23.0 (18.0–30.3)	26.0 (21.5–29.0)	0.884
Migraine Duration (years)	32.0 (23.0–39.0)	33.0 (22.8–39.0)	29.0 (24.5–40.5)	0.807
MIDAS Score*	20.0 (11.8–45.3)	26.0 (15.0–44.0)	10.0 (7.0–47.0)	0.208

Continuous variables are presented as median (IQR). Categorical variables are shown as absolute number with percentage. Owing to small sample sizes, p-values were calculated using the Mann–Whitney U test for continuous variables and Fisher’s exact test for categorical variables

Abbreviations MMD monthly migraine days, MHD monthly headache days, MIDAS Migraine Disability Assessment Score

^*MIDAS score was available for 51 participants; summary statistics and group comparisons for MIDAS were based on available cases only

Genome-wide association analysis

The QTL association analysis revealed four SNPs that reached genome-wide significance (p < 5 × 10^− 8) in association with BoNT-A treatment response (Table 2). These results are presented in a Manhattan plot (Fig. 2a). The Q-Q plot (Fig. 2b) revealed deviations above the diagonal line at the upper tail, suggesting potential genetic associations with improvement rates beyond what would be expected by chance. The λ value was close to 1 (λ = 1.066), suggesting minimal population stratification bias and supporting the validity of the test statistics.

Table 2.

Significant variant SNPs of BoNT-A therapeutic efficacy in our chronic migraine cohort

Variant SNP ID	CHR	BP	Effect Allele	Other Allele	Genotype Counts (Ref/Het/Alt)	BETA	SE	95% CI	P-valuea	Type	Gene	MAFb	TWc	EASd
rs11147666	13	37,722,946	C	T	TT = 59 CT = 4 CC = 0	-0.662	0.096	-0.851 to -0.473	6.64E-09	intronic	TRPC4	0.032	0.051	0.037
rs3924647	5	71,365,029	C	T	TT = 58 CT = 4 CC = 0	-0.661	0.097	-0.852 to -0.470	9.56E-09	intronic	LINC02197	0.032	0.043	0.032
rs56381512	5	71,054,325	C	A	AA = 57 CA = 6 CC = 0	-0.644	0.099	-0.838 to -0.450	2.53E-08	intronic	GTF2H2	0.048	0.042	0.031
rs13126813	4	166,900,006	G	A	AA = 54 GA = 8 GG = 0	-0.635	0.098	-0.828 to -0.442	3.36E-08	intronic	SPOCK3	0.065	0.078	0.101

The effect allele corresponds to the minor allele (PLINK allele A1). Beta represents the estimated change in improvement rate per copy of the effect allele. Genotype counts are listed as reference homozygote / heterozygote / effect-allele homozygote. 95% CI, 95% confidence interval for beta. For rs3924647 and rs13126813, genotype counts sum to 62 because of one missing genotype call for each variant

^a Filter by p < 5 × 10^-8

^b Minor allele frequency in this study

^c Minor allele frequency in the Taiwan Biobank

^d Minor allele frequency in East Asians from the 1000 Genomes Project

Fig. 2.

Quantitative trait loci (QTL) results for BoNT-A treatment response in patients with chronic mi-graine. a Circular Manhattan plot showing the QTL association results. Each point represents a single nucleotide polymorphism (SNP), with its position on the respective chromosome indicated with the outer circle. The inner circle shows the statistical significance (-log10 p-value) of the association of each SNP with treatment response. The red dashed line indicates the standard genome-wide significance threshold (p < 5 × 10⁻8). Four highly significant variants are highlighted and labeled: rs11147666, rs3924647, rs56381512, and rs13126813. b Quantile-Quantile (Q-Q) plot of the observed versus expected -log10 p-values from the association analysis. The red line represents the null hypothesis of no association. Deviations above this line suggest potential genetic associations with the improvement rate. The genomic inflation factor (λ) is provided; it was calculated based on the median of the observed and expected chi-square values, supporting the validity of the test statistics

All four significant variants were located in intronic regions and demonstrated negative beta coefficients, indicating that the presence of the minor allele was associated with reduced treatment efficacy. The strongest association was observed for rs11147666 (p = 6.64 × 10^− 9) in TRPC4 on chromosome 13, with a substantial negative effect on the treatment response (β = -0.662, SE = 0.096).

The other significant variants included rs3924647 (p = 9.56 × 10^− 9) in the intron region of long non-coding RNA LINC02197 on chromosome 5, rs56381512 (p = 2.53 × 10^− 8) in the intron region of GTF2H2 on chromosome 5, and rs13126813 (p = 3.36 × 10^− 8) in the intron region of SPOCK3 on chromosome 4. These variants also presented strong negative associations with the BoNT-A treatment response, showing similar effect sizes (rs3924647: β = -0.661, SE = 0.097; rs56381512: β = -0.644, SE = 0.099; and rs13126813: β = -0.635, SE = 0.098). The minor allele frequencies of these significant variants in our cohort ranged from 3.2% (rs11147666 and rs3924647) to 6.5% (rs13126813), which were comparable to those reported in the Taiwan Biobank and East Asian populations from the 1000 Genomes Project (Table 2).

To address potential overfitting given the modest sample size, we evaluated the lead SNP associations using three pre-specified models with increasing levels of covariate adjustment (Table S1): (1) a minimal model including genotype, age and sex; (2) a clinically adjusted model including genotype, age, sex, triptan response, preventive treatment failure, MMD, MHD, and migraine duration; and (3) a fully adjusted model including genotype, age, sex, MOH, aura, triptan response, preventive treatment failure, MMD, MHD and migraine duration. The SNP effect estimates remained directionally consistent across these models, supporting the robustness of the observed associations while still warranting cautious interpretation given the modest sample size.

We also re-analyzed the four lead SNPs using robust linear regression. As shown in Table S2, the fully adjusted model yielded statistically significant associations for all four SNPs, consistent in direction with the original PLINK results. The model included genotype, age, sex, MOH, aura, triptan response, preventive treatment failure, MMD, MHD, and migraine duration. Overall, the lead SNP associations remained significant after covariate adjustment, supporting the robustness of the observed associations despite the modest sample size and low minor allele frequencies.

Power calculations of these four lead SNPs indicated that, given the sample size of 63 and genome-wide significance (p < 5 × 10⁻⁸), the power was approximately 25–29% (rs11147666: 0.29, rs3924647: 0.29, rs56381512: 0.25, and rs13126813: 0.28) even across a range of MAF values of 0.1, 0.15, 0.2, and 0.25. While these variants reached formal significance, the limited power suggests that these findings should be interpreted as exploratory signals.

Genotype–phenotype associations

Descriptive genotype-group scatter plots showed lower improvement rates among heterozygous carriers than among common homozygotes for each lead SNP (Fig. 3). Given the small number of minor-allele carriers (n = 4–8), these genotype-group comparisons should be interpreted cautiously as supportive descriptive findings requiring independent replication.

Fig. 3.

Descriptive genotype–phenotype visualization for the four significant variants. Individual-level scatter plots show the distribution of MHD improvement rates after BoNT-A therapy according to genotype for rs11147666, rs3924647, rs56381512, and rs13126813. For each variant, heterozygous carriers showed lower improvement rates than the corresponding common homozygotes. These plots are intended as descriptive visualizations of the raw data distribution, whereas the primary association results were derived from the additive linear regression model used in the genome-wide QTL analysis

For rs3924647, patients with the TT genotype demonstrated markedly better improvement rates than those with the CT genotype. Similarly, for rs56381512, the AA genotype was associated with better outcomes than the CA genotype. The most pronounced effect was observed with rs13126813, where the heterozygous GA genotype showed up to a 70% reduction in improvement rates compared to the AA genotype.

Further analysis of individual cases (Table 3) revealed that participants with particularly low improvement rates (defined as < 0.3) carried multiple heterozygous variants. For example, participants 42 and 50, both of whom exhibited no improvement with BoNT-A therapy, carried heterozygous variants in rs13126813 (SPOCK3), with participant 50 also harboring a heterozygous mutation in rs56381512 (GTF2H2).

Table 3.

BoNT-A therapy profile in participants with significant variant SNPs

Participant number	Improvement rate	Triptana	Variants Genotypeb
			rs11147666	rs3924647	rs56381512	rs13126813
50	0.000	N	-	-	Het	Het
42	0.000	N	-	-	-	Het
27	0.000	Y	-	-	-	Het
74	0.032	N	Het	Het	Het	Het
17	0.038	Y	Het	Het	Het	-
38	0.143	N	-	-	-	Het
40	0.226	N	Het	Het	Het	Het
15	0.267	Y	Het	Het	Het	-

^a Triptan response Y self-reported positive response to triptan, N no response

^b Het: heterozygous genotype; -: common homozygote

We also compared improvement rates with BoNT-A treatment between participants who showed a positive triptan response and those who did not (Figure S2). Although triptan non-responders appeared to show a trend toward lower improvement rates post-BoNT-A treatment, particularly with more individuals falling below the 0.25 improvement threshold, the overlap between distributions was considerable, with no significant difference.

LD and functional analyses

The analysis revealed high LD among several variants in each gene region. Multiple SNPs near the lead variant rs11147666 in TRPC4 showed strong correlations, suggesting a potential regulatory region in this intron (Fig. 4a). Similarly, rs13126813 in SPOCK3 demonstrated high LD with other intronic SNPs in the same gene (Fig. 4d).

Fig. 4.

Linkage disequilibrium (LD) analysis of intronic variants. LD plots and functional annotations for the significant intronic variants: (a) rs11147666 in TRPC4 (chromosome 13), (b) rs3924647 in LINC02197 (chromosome 5), (c) rs56381512 in GTF2H2 (chromosome 5), (d) rs13126813 in SPOCK3 (chromosome 4). Each panel shows the LD structure around the focal SNP. The color gradient, ranging from white to deep blue and red, visualizes the spectrum of D' (D prime) and R² (R-squared) values ranging from 0.0 to 1.0 in the relative chromosome region. White numbers in purple squares represent FORGEdb scores, indicating the likelihood of variants functioning as regulatory elements. The red text indicates the query intronic variants from the QTL analysis. The genes in this region are shown below each plot; thin lines represent intron regions and bold lines indicate exon regions

Notably, variant rs3924647 in LINC02197 showed high LD with nearby SNPs in its region (rs375624366, rs373554921, and rs2000994706), as well as with variants in the GTF2H2 region (Fig. 4b). Similarly, rs56381512 exhibited a high correlation with rs200246435, rs199928774, and rs376037226 in the intron region of GTF2H2 (Fig. 4c). This pattern suggests potential functional relationships between these genomic regions.

The FORGEdb analysis predicted moderate regulatory effects for several variants, including multiple SNPs in TRPC4 (rs7320883, rs12583718, rs17056504, rs75411262, rs76267737, rs78705242, rs12584260, rs17263892, and rs4941878), one in GTF2H2 (rs56381512), and two in SPOCK3 (rs17520609 and rs67647350). These SNPs were preliminarily inferred to be concentrated within the same gene region, exhibiting potential for functional impact as indicated by their FORGEdb scores, and potentially exerting a synergistic influence on gene functionality.

GTEx interpretation and GO annotation analysis

To provide functional context for the identified associations, we queried the GTEx database and found that two lead SNPs, rs3924647 and rs56381512, exhibited significant eQTL effects on GTF2H2 expression in multiple brain regions (Table S3). Given the neurological relevance of the studied phenotype, these results provide biologically plausible supportive context for the observed genetic associations.

The DAVID-based GO annotation provided descriptive biological process and molecular function terms for the genes mapped to the lead variants (Table 4 and S4). These annotations are presented as supportive functional context rather than formal enrichment results and should therefore be interpreted cautiously. TRPC4 is primarily involved in calcium ion transport-related functions (GO:0070509), as well as gamma-aminobutyric acid secretion (GO:0014051) and oligodendrocyte differentiation (GO:0048709). Its molecular function includes calcium channel activity (GO:0005262 and GO:0015279). GTF2H2 is associated with transcription processes by RNA polymerase II (GO:0006357), G protein-coupled receptor internalization (GO:0002031), and nucleotide-excision repair (GO:0006289). The descriptive molecular function terms included metal ion and zinc ion binding capabilities (GO:0046872 and GO:0008270). SPOCK3 is involved in regulating cell motility (GO:2000146) and endopeptidase activity (GO:0010951). Notably, its molecular functions include calcium ion binding (GO:0005509), suggesting a potential role in calcium signaling pathways, which may be relevant to the mechanism of action of BoNT-A.

Table 4.

Gene ontology (GO) biological process (BP) annotations of genes associated with significant variant SNPs

Variant	Gene	Relative GO terms	Relative GO IDs
rs11147666	TRPC4	calcium ion import calcium ion transport calcium ion transmembrane transport gamma-aminobutyric acid secretion oligodendrocyte differentiation regulation of cytosolic calcium ion concentration	GO:0070509 GO:0006816 GO:0070588 GO:0014051 GO:0048709 GO:0051480
rs56381512	GTF2H2	DNA-templated transcription G protein-coupled receptor internalization nucleotide-excision repair regulation of transcription by RNA polymerase II response to UV transcription by RNA polymerase II transcription initiation at RNA polymerase II promoter	GO:0006351 GO:0002031 GO:0006289 GO:0006357 GO:0009411 GO:0006366 GO:0006367
rs13126813	SPOCK3	negative regulation of cell motility negative regulation of endopeptidase activity	GO:2,000,146 GO:0010951

rs3924647 (LINC02197) is not listed because LINC02197 is a long non-coding RNA gene without GO biological process annotation in DAVID

Discussion

This study represents, to our knowledge, the first genome-wide investigation of genetic factors associated with response to BoNT-A therapy in Han Chinese patients with chronic migraine. Using a QTL-based exploratory approach, we identified four SNPs—rs11147666, rs3924647, rs56381512, and rs13126813—that reached genome-wide significance in this dataset and were mapped to or located near TRPC4, LINC02197, GTF2H2, and SPOCK3, respectively. However, given the modest sample size, low minor allele frequencies, and absence of an independent replication cohort, these findings should be interpreted as preliminary and hypothesis-generating, rather than confirmatory evidence of causal or clinically actionable associations.

From a biological perspective, these loci may provide supportive functional clues regarding pathways potentially relevant to variability in BoNT-A response. In particular, genes related to calcium signaling (TRPC4 and SPOCK3) and transcriptional regulation (GTF2H2) may suggest the potential involvement of neuronal excitability and synaptic regulatory mechanisms. In addition, GTEx-based annotation showed that rs3924647 and rs56381512 were associated with GTF2H2 expression in several brain tissues, providing further supportive—though not confirmatory—functional context. These interpretations remain speculative and require independent replication and further experimental validation.

Among the four significant SNPs, rs11147666, located within the intronic region of TRPC4, exhibited the strongest association with diminished response to BoNT-A treatment (β = -0.662, SE = 0.096, p = 6.64 × 10^{− 9}). TRPC4 encodes a calcium-permeable non-selective cation channel that plays a central role in regulating the intracellular calcium influx in response to G protein-coupled receptor activation [29]. This function may be relevant to neuronal excitability, synaptic plasticity, and neurotransmitter release, all of which are pivotal in migraine pathophysiology. DAVID-based GO annotation linked TRPC4 to terms related to calcium ion transport, channel activity, and neurotransmitter regulation. This may suggest potential relevance to migraine-related neuronal signaling pathways, including cortical spreading depression (CSD) and trigeminovascular activation, although this interpretation remains speculative and requires further validation [30]. Specifically, CSD can trigger CGRP release and neurogenic inflammation, ultimately activating meningeal nociceptors—processes in which calcium signaling is critically involved.

Recent preclinical studies provide functional validation for the role of TRPC4 in migraine. Cohen et al. [31] demonstrated that TRPC4 is highly expressed in trigeminal ganglia neurons, and its pharmacologic inhibition using ML204 significantly reduced CGRP release, mechanical hypersensitivity risk, and migraine-like behaviors in nitroglycerin-induced mouse models. These findings support a possible role of TRPC4 in peripheral CGRP-mediated pain signaling and indicate its potential as a therapeutic target. Additionally, TRPC4 may act upstream of CGRP release through intraganglionic transmission. Zhang et al. [32] demonstrated that chemical messengers such as CGRP, ATP, and glutamate can facilitate cross-excitation among neighboring trigeminal neurons via gap junctions, contributing to peripheral sensitization and sustained trigeminovascular activation. This mechanism aligns with the action of BoNT-A in reducing CGRP release and attenuating neurogenic inflammation, suggesting that genetic variations in TRPC4 may influence the therapeutic effects.

The variant rs56381512 is located within GTF2H2, encoding general transcription factor IIH subunit 2, which is associated with transcription initiation and DNA repair processes [33]. The descriptive GO terms included transcription by RNA polymerase II, transcription initiation at the RNA polymerase II promoter, and response to UV, indicating the cellular responses to DNA damage caused by ultraviolet radiation, particularly in the context of nucleotide excision repair [34]. It is also interesting that rs3924647 is very far from the GTF2H2 (more than 297,500 bp) but a strong effect of GTF2H2 gene expression. This observation may suggest a regulatory relationship requiring further functional validation.

Although GTF2H2 is not directly implicated in established BoNT-A-related mechanisms, its role in transcriptional regulation may modulate the molecular pathways involved in pain signaling and neuroinflammation. Notably, GTF2H2 has been identified as part of a transcriptional signature associated with chronic neuropathic pain, particularly within immune and glial pathways that regulate inflammatory responses and neuronal excitability. As BoNT-A acts partially through suppression of CGRP-mediated neurotransmission and synaptic plasticity, genetic variants in transcriptional regulators such as GTF2H2 may influence the baseline expression levels of pain-related genes, thereby affecting downstream responses to treatment [35].

The rs13126813 variant in the SPARC/osteonectin, CWCV, and Kazal-like domains proteoglycan 3 (SPOCK3) gene is particularly interesting given its role in proteolytic regulation and calcium binding. SPOCK3 encodes a secreted extracellular matrix glycoprotein that is primarily involved in regulating cell-matrix interactions, neuronal differentiation, and calcium channel modulation [36]. The protein belongs to the testican family of proteoglycans, which influence synaptic plasticity, axon guidance, and neuronal excitability—all processes relevant to migraine pathophysiology [37].

Although direct evidence linking SPOCK3 to migraine biology is limited, its extracellular matrix and calcium-binding properties may warrant further investigation in relation to neuronal signaling and treatment response [38]. As the therapeutic effects of BoNT-A involve modulation of synaptic transmission and neuropeptide release, SPOCK3 variants may potentially affect neuronal sensitivity, synaptic remodeling, or vesicle release dynamics, thereby influencing treatment outcomes.

The variant rs3924647 is located in LINC02197, a long non-coding RNA whose regulatory roles in gene expression are still being elucidated. The identification of a significant variant in a non-coding RNA gene highlights the potential role of epigenetic and transcriptional regulation in shaping individual treatment responses [6]. Considering that BoNT-A may indirectly influence gene expression through long-term modulation of afferent input and neuroinflammatory mediators, regulatory loci such as LINC02197 may affect downstream pathways involved in pain processing [39].

Because the identified variants are intronic and no direct functional validation was performed, these interpretations should be regarded as supportive and hypothesis-generating rather than confirmatory. Nevertheless, the observed associations are consistent with the known role of BoNT-A in reducing peripheral and central sensitization by inhibiting neurotransmitter release [40]. Second, variants affecting extracellular matrix interactions or channel function can modulate BoNT-A uptake, diffusion, or downstream neuromodulatory actions [41]. Previous studies have indicated that the pericellular environment and axonal architecture may affect BoNT-A binding and internalization, contributing to variability in treatment responses [42].

Although previous studies have demonstrated that triptan responsiveness is associated with outcomes of treatment with CGRP-targeting monoclonal antibodies, our findings do not support its predictive value for BoNT-A treatment efficacy. Furthermore, although triptan non-responders tended to show lower BoNT-A effectiveness than responders, the difference was not significant. This discrepancy may reflect fundamental differences in the mechanisms of action between these treatment classes. Triptans and CGRP monoclonal antibodies primarily act on acute pain pathways via the activation of 5-HT1B/1D receptors and inhibition of CGRP release, respectively [43, 44]. In contrast, BoNT-A exerts its therapeutic effects by broadly inhibiting vesicle exocytosis of multiple neuropeptides and may further attenuate both peripheral and central sensitization through retrograde transport and central nervous system mechanisms [45]. The lack of a common association suggests that distinct biological mechanisms shape individual outcomes of different migraine therapies.

The observation of genome-wide significant associations despite the modest sample size is noteworthy but must be interpreted cautiously. From a clinical perspective, an early identification of patients less likely to respond to BoNT-A may help optimize treatment sequencing and avoid prolonged exposure to an ineffective invasive preventive therapy. In the context of expanding treatment options for CM, including CGRP-targeted therapies, such information could facilitate efficient allocation of healthcare resources. Nonetheless, any predictive interpretation from the current dataset must remain exploratory until validated in larger cohorts.

This study has several limitations that should inform future research. First, the modest sample size limits statistical power for detecting variants with small-to-moderate effect sizes; future multi-center studies should aim for substantially larger cohorts to achieve adequate power for genome-wide discovery. Second, the absence of an independent replication cohort precludes definitive validation of the reported associations. Therefore, the present findings should be considered exploratory signals only. Future studies should employ a two-stage discovery–replication design and incorporate additional supportive analyses, such as expression QTL (eQTL) annotation, colocalization, and experimental validation, to better assess biological relevance. Third, functional characterization of the identified intronic variants was not performed; future studies should integrate expression eQTL analyses using resources such as GTEx, followed by experimental validation through reporter assays or CRISPR-based perturbation. Finally, although MHD and MMD were collected using prospective daily headache diaries, some entries may have been recorded later in the same day rather than in real time. Therefore, a small degree of recall-related measurement bias cannot be entirely excluded.

In conclusion, this pilot pharmacogenomic study provides preliminary evidence that common genetic variation may contribute to heterogeneity in the BoNT-A response in CM. Given the limited sample size and absence of an independent replication cohort, the reported loci should be interpreted as exploratory and hypothesis-generating findings. Validation in adequately powered, multi-center studies with independent replication cohorts is essential before any clinical translation can be considered.

Abbreviations

BoNT-A

OnabotulinumtoxinA

CGRP

Calcitonin Gene-Related Peptide

CM

Chronic Migraine

GO

Gene Ontology

ICHD-3

International Classification of Headache Disorders, 3rd edition

LD

Linkage Disequilibrium

MHD

Monthly Headache Days

MMD

Monthly Migraine Days

MIDAS

Migraine Disability Assessment Score

MOH

Medication Overuse Headache

PTX3

Pentraxin 3

QTL

Quantitative Trait Locus

SNARE

Soluble N-Ethylmaleimide-Sensitive Factor Attachment Protein Receptor

SNP

Single Nucleotide Polymorphism

TPM

Taiwan Precision Medicine

CHB

Han Chinese in Beijing

CHS

Southern Han Chinese

D'

D Prime

R²

R-Squared

Q-Q

Quantile-Quantile

λ

Genomic Inflation Factor

CSD

Cortical Spreading Depression

Authors’ contributions

The study was conceived and designed by Y.-C.A. and F.-C. Y. Data curation and statistical analyses were performed by Y.-C.A., K.-S.H., and F.-C.Y. First draft of this manuscript was written by Y.-C.A., K.-S.H., and F.-C.Y. All authors conducted the research and assisted with methodology development. The study was supervised by F.-C.Y. All authors contributed to manuscript revision and have read and approved the submitted version.

Funding

This research was funded by the Tri-Service General Hospital, Taiwan, grant numbers TSGH-D113092, TSGH-D113108, TSGH-D114105, and TSGH-D115115; Cheng Hsin General Hospital, Taiwan, grant number CHNDMC-114-03 and CHNDMC-115-12; and Academia Sinica, grant numbers AS-40-05-GMM and AS-GC-110-MD02.

Data availability

The GWAS summary results generated during the current study are accessible via the **LocusZoom** results page at: [https://my.locuszoom.org/gwas/563503/](https:/my.locuszoom.org/gwas/563503).

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki, and approved by TSGH Institutional Review Board (protocol code TSGHIRB: 2-108-05-038).

Informed consent was obtained from all subjects involved in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.