Genome-wide association study of musical beat synchronization demonstrates high polygenicity

The study investigates the genetic architecture of human beat synchronization, a core aspect of musicality involving perceiving, predicting and moving in time with a musical beat. Prior behavioral, developmental and cross-cultural research indicates rhythm’s importance for communication, cognition, and social coordination. Twin and family studies suggest moderate heritability of rhythm-related traits, yet molecular genetic underpinnings have been largely unexplored due to phenotyping challenges. The purpose of this work is to validate a scalable self-report proxy for beat synchronization, perform a large-scale GWAS to identify associated loci, quantify SNP-based heritability, characterize biological enrichment (tissues, pathways, regulatory elements), assess evolutionary annotations, test polygenic prediction of musical engagement in an independent biobank, and evaluate shared genetic architecture with other traits via genetic correlations and genomic structural equation modeling.

Previous studies demonstrate rhythmic abilities emerge early and support language and social behaviors. Neuroimaging highlights auditory-motor networks and neural entrainment during rhythm processing. Twin studies report moderate heritability for rhythm and timing phenotypes (duration and rhythm discrimination, isochronous production, off-beat detection). Prior molecular genetics work on musicality involved small samples, linkage or candidate approaches, with limited reproducibility. A locus at 4q22.1 had been implicated in musical aptitude. The need for large-scale GWAS and well-validated phenotypes is emphasized to resolve the complex polygenic nature of musical traits and to relate genetic findings to neural and regulatory contexts.

Design overview: (1) Validate a self-report beat synchronization item ('Can you clap in time with a musical beat?') against task-based measures of rhythm perception and beat tapping; (2) Conduct a GWAS in N=606,825 23andMe participants of European ancestry using the self-report item; (3) Estimate SNP-based heritability, perform gene-based tests, gene-property and gene-set enrichments, partitioned heritability, and tissue/cell-type specific enrichment; (4) Perform evolutionary analyses for human accelerated regions (HARs) and Neanderthal introgressed variants; (5) Validate GWAS via polygenic scores (PGS) predicting musician status in an independent biobank; (6) Conduct genetic correlation analyses with 64 traits and genomic SEM to model shared genetic factors; (7) Multivariate GWAS on a latent factor and phenotypic replications for selected traits.

Phenotype validation:

• Phenotype Experiment 1 (N=724; Amazon Mechanical Turk). Measures: self-report target question; rhythm perception discrimination test (32 trials; simple vs complex rhythms). Analysis: logistic regression predicting Yes vs No from d′, controlling for age, sex, education; McFadden’s R².
• Phenotype Experiment 2 (Questionnaires N=1,412; tapping subset N=542). Measures: self-report target question; 7-item rhythm scale; Likert item 'I can tap in time with a musical beat'; Gold-MSI short form; health traits (chronotype, smoking, shortness of breath, tinnitus); Beat synchronization tapping task using REPP online platform measuring standard deviation of asynchrony while tapping to 4 musical excerpts (with calibration, practice and main tapping trials). Analyses preregistered: logistic and linear regressions (H1–H6), controlling for age, sex, education; alternative outcome (vector length) as sensitivity.

GWAS:

• Sample: 555,660 cases (Yes) and 51,165 controls (No) of European ancestry from 23andMe; unrelated; adjusted for age, sex, top 5 principal components, and genotyping platform. SNP QC: MAF≥0.01, imputation R²≥0.3, indels excluded; 8,288,850 SNPs analyzed.
• Model: logistic regression additive genetic model. LD pruning to define independent 'sentinel' SNPs (r²=0.6 then r²=0.1 within 250 kb). LDSC used for intercept and SNP-heritability (observed and liability-scale with assumed population prevalence 3.0–6.5%).

Post-GWAS analyses:

• FUMA for locus definition, Manhattan/QQ plots, and gene mapping.
• MAGMA gene-based association, gene property enrichment (GTEx v8; 54 tissues, conditional on average expression), and exploratory gene-set enrichment (Gene Ontology; Bonferroni corrected).
• Stratified LDSC to partition heritability by 52 functional categories (e.g., conserved regions, histone marks H3K9ac, H3K4me1), and LDSC-SEG for tissue/cell-type-specific enrichments (multi-tissue chromatin and gene expression; Bonferroni corrected).

Evolutionary analyses:

• HARs: overlap of significant loci and HARs (permutation-based enrichment; 10,000 shuffles), and stratified LDSC heritability enrichment within HAR annotations. Neanderthal introgression assessed similarly.

PGS validation:

• Independent cohort from Vanderbilt University Medical Center’s BioVU with algorithmically identified musicians (N=1,259) and matched controls (N=4,893); European ancestry only; logistic regression predicting musician status from standardized PGS (PRS-CS) with covariates (age, sex, top 10 PCs).

Cross-trait analyses:

• Genetic correlations via LDSC with 64 traits across domains: cognitive, psychiatric, neuroimaging/neurological, motor, sleep/respiratory/circadian/heart rhythms, overall health/well-being, and hearing; Bonferroni corrected.
• Genomic SEM to model a common factor linking beat synchronization with grip strength, usual walking pace, peak expiratory flow, processing speed; fit indices reported; multivariate GWAS on latent factor; tissue enrichment for latent factor.

Phenotypic extensions:

• In Phenotype Experiment 2, linear regressions tested associations between tapping accuracy and selected traits (chronotype, shortness of breath, smoking, tinnitus; preregistered), with covariate adjustments and musician status sensitivity analyses.

• Phenotype validation:

  • Experiment 1 (N=724): Rhythm discrimination performance (d′) predicted answering Yes to self-report target question (OR=1.92; P=0.002; McFadden’s R²=0.06; 95% CI 1.27–2.95).
  • Experiment 2 (N=1,412; tapping N=542): Yes to target question associated with lower tapping asynchrony (better synchronization) (OR=0.28; P<0.001; McFadden’s R²=0.24; 95% CI 0.18–0.42). Likert self-report correlated with tapping accuracy (r=−0.40; P<0.001). Rhythm questionnaire associated with target question (OR=7.34; P<0.001; McFadden’s R²=0.34; 95% CI 4.90–11.52) and with tapping accuracy (r=−0.41; P<0.001). Gold-MSI associated with target question (OR=4.16; P<0.001; McFadden’s R²=0.18; 95% CI 2.90–6.12) and with tapping accuracy (r=−0.36; P<0.001). Effects robust to age, sex, education.

• GWAS (N=606,825):

  • 70 sentinel SNPs at 69 loci reached genome-wide significance (P<5×10−8); 6,160 SNPs surpassed genome-wide threshold overall. LDSC intercept=1.02 (s.e.=0.01), ratio=0.03, indicating inflation due largely to polygenicity.
  • Top loci include rs848293 near VRK2 (2p), rs4792891 in/near MAPT (17q), and signal at MAPK3 (16p11.2), among others (for example, GPM6A, GBE1, FOXO6/FOXO3, PPP1CB, SND1, SLC6A13, CDH12, PTPRD, PICALM, CREB1).
  • SNP-based heritability: observed-scale h2≈0.05 (s.e.=0.002); liability-scale 13%–16% assuming 3.0–6.5% prevalence of atypical beat synchronization.

• Enrichment and functional analyses:

  • MAGMA identified 129 genes at P<2.56×10−6; top genes included CCSER1 and VRK2.
  • Gene property enrichments: significant enrichment for genes expressed in brain tissues (cortex, cerebellum, basal ganglia: putamen, caudate, nucleus accumbens).
  • Gene-set enrichment: synaptic membrane adhesion (P=1.01×10−7) and synaptic adhesion-like molecules (P=8.35×10−7).
  • Partitioned heritability: enrichment in mammalian-conserved regions and regulatory regions marked by H3K9ac and H3K4me1; enrichment in CNS- and skeletal muscle–specific regulatory elements; fetal and adult brain regulatory elements enriched.

• Evolutionary analyses:

  • Two significant variants overlapped HARs (rs14316 at locus 66; rs1464791 at locus 20 near GBE1). Overlap enrichment 11.2× vs chance (P=0.017; 10,000 permutations). LDSC indicated 2.26× heritability enrichment in HARs (P=0.14; underpowered due to few variants in HARs).

• Polygenic score validation (BioVU):

  • Beat synchronization PGS predicted musician status (OR=1.34 per s.d. increase; P<2×10−16; Nagelkerke’s R²=2%; 95% CI 1.26–1.43).

• Genetic correlations (LDSC; significant at P<7.8×10−4):

  • Positive with motor traits (grip strength; usual walking pace), breathing traits (peak expiratory flow, FEV1, FVC), processing speed, tinnitus, smoking heaviness; negative with risk-taking, smoking initiation, hearing difficulty, insomnia, morning chronotype, shortness of breath.

• Genomic SEM:

  • Best-fitting model included a common genetic factor underlying beat synchronization, grip strength, usual walking pace, peak expiratory flow, and processing speed; common factor explained 11.6% of variance in beat synchronization GWAS and 9.6–25.0% in other GWASs. Multivariate GWAS of this factor identified 130 genome-wide significant loci; heritability of latent factor 7.27% (s.e. 0.25%); enriched expression in cerebellum.

• Phenotypic extensions (Experiment 2):

  • Better tapping accuracy associated with eveningness (r=−0.10; P=0.015), less shortness of breath (r=−0.16; P<0.001), and smoking abstinence/fewer lifetime cigarettes (inverse of ≥20 cigarettes: r=−0.11; P<0.001), consistent with directions of genetic correlations.

The findings demonstrate that beat synchronization is a highly polygenic trait with thousands of common variants of small effect contributing to individual differences. The discovery of 69 genome-wide significant loci, along with moderate SNP-based heritability, validates a genetic basis for rhythm synchronization. Enrichment in brain-expressed genes, synaptic adhesion pathways, and brain-specific regulatory regions links genetic associations to neural systems known to support auditory-motor entrainment. The PGS association with musician status in an independent biobank shows the GWAS captures variance related to broader musicality. Genetic correlations and genomic SEM reveal shared genetic influences between beat synchronization and biological rhythms (respiration, circadian preferences), motor abilities (walking pace, grip strength), and cognitive processing speed, suggesting common biological mechanisms that may impact health across the lifespan. Together, these results support a model in which genetic variation influences neural development and regulatory mechanisms in cortical–subcortical networks underlying rhythm perception and production, with broader connections to motor and respiratory functions.

This study provides the first large-scale genomic dissection of beat synchronization, identifying 69 loci and demonstrating moderate SNP-based heritability with enrichment in brain tissues and synaptic pathways. Validated self-report measures enable scalable phenotyping for musical traits. Polygenic prediction connects beat synchronization genetics to real-world musical engagement, and cross-trait analyses highlight shared biology with motor, respiratory, sleep, and cognitive traits. Future work should: (1) extend genomic studies to diverse ancestries; (2) employ continuous, task-based phenotypes at scale (e.g., online tapping) to refine genetic architecture; (3) integrate multi-omic and neuroimaging data to map variants to neural circuits; (4) dissect specificity versus pleiotropy across musical, auditory, and motor domains; and (5) explore developmental trajectories and health implications of rhythm-related genetic factors.

• Phenotype based on self-report (Yes/No) introduces measurement error and may imperfectly capture continuous beat synchronization ability.
• Liability-scale heritability depends on uncertain prevalence estimates for atypical beat synchronization.
• Sample restricted to individuals of European ancestry; results may not generalize across ancestries; PGS portability is limited.
• Potential residual population stratification cannot be fully excluded, as with most GWAS.
• Limited direct replication due to absence of an independent similarly powered GWAS; validation performed via PGS in an external cohort.
• Overlap with HARs and evolutionary annotations is underpowered due to few common variants in these regions.
• Associations are not deterministic; environmental factors play a major role; ethical considerations caution against individual prediction or selection based on PGS.