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

Moving in synchrony to the beat is a fundamental component of musicality. Here, we conducted a genome-wide association study (GWAS) to identify common genetic variants associated with beat synchronization in 606,825 individuals. Beat synchronization exhibited a highly polygenic architecture, with sixty-nine loci reaching genome-wide significance (p<5×10−8) and SNP-based heritability (on the liability scale) of 13%-16%. Heritability was enriched for genes expressed in brain tissues, and for fetal and adult brain-specific gene regulatory elements, underscoring the role of central nervous system-expressed genes linked to the genetic basis of the trait. We performed validations of the self-report phenotype (through internet-based experiments) and of the GWAS (polygenic scores for beat synchronization were associated with patients algorithmically classified as musicians in medical records of a separate biobank). Genetic correlations with breathing function, motor function, processing speed, and chronotype suggest shared genetic architecture with beat synchronization and provide avenues for new phenotypic and genetic explorations.


51
Our tendency to perceive, create, and appreciate rhythms in a variety of contexts (e.g., speech, music, 52 movement) is a key feature of the human experience 1-3 . Rhythmic patterns provide predictable and 53 robust sensorimotor structure to every-day interactions 4,5 , helping guide our attention to 54 communicatively important moments in time 6,7 . Even young children are sensitive to the social and 55 linguistic signals carried by rhythm 8-10 and parents use rhythmic vocalizations and synchronous 56 movement (e.g., lullabies and rocking) to interact with their infants from birth 11,12 . Rhythmic musical 57 interactions are structured around the percept of a stable periodic pulse (termed the "beat" in Western 58 music and present in music of most cultures 1,13 , though its precise instantiation in musical structure 59 varies cross-culturally 14,15 ). While music in general and rhythmic structures in particular vary globally 15-17 , 60 there is evidence that hierarchical beat structure of most music is robust to cultural transmission 2 and 61 indeed common in many types of music 1 . 62 63 Beat perception and synchronization (i.e. perceiving, predicting, and moving predictively in synchrony to 64 a musical beat 18 ) is an important feature of musical experiences across many human cultures and 65 musical genres 1,19 . The predictive temporal mechanisms afforded by beat structure enhance general 66 perceptual and learning processes in music, including melody perception and production, singing, and 67 joint music-making 3,6 . While some features of rhythm perception and production vary across listeners 68 from different cultures 13,19-21 , the same studies showed considerable consistencies across cultures for 69 other features (e.g., preference for beat-based isochrony). Musicality (broadly encompassing musical 70 behavior, music engagement and musical skill 22 ) impacts society by supporting pro-social behavior 11,23 71 and well-being 24 . Many have proposed that beat perception and synchronization evolved in humans to 72 support communication and group cohesion 18,22,25,26 . In modern humans, beat perception and 73 synchronization are predictive of language and literacy skills 27,28 and are related to cognition, motor 74 function, and social coordination 29 . Thus, the biology of beat synchronization has general importance for 75 understanding human ability to perceive and predict natural rhythms, may have relevance for 76 characterizing phenotypes such as developmental speech-language disorders which demonstrate 77 associations with atypical rhythm 30 , and may further elucidate mechanisms of rhythm-based 78 rehabilitation (e.g., for stroke and Parkinson's 31 ).

146
Phenotype Experiment 1: Rhythm perception task performance. 147 In this experiment N=724 (see Table 1 for demographics) were asked the target question and performed 148 a musical rhythm perception test (Supplementary Figure 2). In each of the 32 trials of the task, 149 participants judged whether two rhythms were the same or different (see Figure 1A), following a 150 standard procedure for assessing musical perception ability 45 and utilizing rhythm sequences with 151 simple (highly metrical) and complex (syncopated) rhythms 46 . The rhythm perception task yielded 152 quantitative scores (d'). Individuals with better performance in the rhythm perception test (higher total 153 d') were more likely to answer Yes (vs. No) to the target question (OR=1.92, p=0.002, McFadden's 154 R 2 =0.06, 95% CI=1.27,2.95; Figure 1B). All tests in both phenotype experiments were two-tailed. 155 156 Phenotype Experiment 2: Beat synchronization task performance 157 We then validated self-reported beat synchronization phenotype (N=1,412) as a proxy for directly-158 measured beat synchronization ability. Participants (Table 1) completed a questionnaire on musicality, 159 health, and personality, and were asked to tap in real time to the beat of 4 different musical excerpts 160 (Supplementary Figure 3) Figure 1C). Tapping asynchrony was also negatively correlated with responses to a 170 highly similar item ("I can tap in time to a musical beat") when asked on a seven-point Likert agreement 171 scale (1= disagree; 7 = agree), r= -.40, p<.001, 95% CI=-0.47,-0.33] (H1a; Figure 1D). Similarly, individuals 172 with higher self-reported rhythmic ability (from another multi-item questionnaire) were much more 173 likely to respond "Yes" to the target question, OR=7.34, p<.001, McFadden's R 2 =.34, 95% CI=4.90,11.52], 174 ( Figure 1E), and demonstrate lower tapping asynchrony, r = -.41, p <.001, 95% CI=-0.47,-0.33] ( Figure 1F) 175 (H2). Controlling for confidence judgments or confidence as a personality trait did not diminish the 176 associations between self-report and tapping asynchrony (H3; Supplementary Notes). Musical 177 Sophistication 43 was positively associated with the target question, OR=4.16, p <.001, McFadden's 178 R 2 =.18, 95% CI=2.90,6.12 ( Figure 1G) and negatively correlated with tapping asynchrony r= -.36, p <.001, 179 95% CI =-0.43,-0.28 ( Figure 1H; H5  GWAS was conducted using logistic regression  195  under an additive genetic model, while adjusting for age, sex, the first five principal components from  196 genetic data, and genotype platforms (Methods). Seventy "sentinel" SNPs (after two rounds of LD  197 pruning, first at r 2 =0.6 and then at r 2 =0.1, kb = 250) at 69 genomic loci reached genome-wide 198 significance (p<5x10 -8 ; two-tailed; Figure 2, Table 2, and Supplementary Table 2), with a total of 6,160 199 SNPs passing the genome-wide significance threshold. Sixty-seven loci were autosomal and two were on 200 the X chromosome; locus 28 contains two independent sentinel SNPs. QQ-plot is provided in 201 Supplementary Figure 4, and local association plots at each locus are in the Regional Plots Supplemental 202 document. The LD score regression intercept was 1.02 (se=0.01) the ratio was 0.03, indicating that the 203 majority of inflation in test statistics was due to true polygenicity instead of population stratification.

209
Sixty-nine loci (70 sentinel SNPs, with one locus containing two independent sentinel SNPs) surpassed the threshold for genome-wide 210 significance of p<5x10 -8 (dotted horizontal line). For illustration purposes, only 500,000 SNPs with p<0.1 are shown; gene symbols for sentinel 211 SNPs are notated when FUMA provided a gene mapped to nearest sentinel SNP.

213 214
The top-associated locus (rs848293) was mapped at chromosome 2 close to VRK2 (Vaccinia 215 Serine/Threonine Kinase 2 which codes for a protein kinase with multiple spliced isoforms expressed in 216 the brain) and FANCL, within a region previously linked to multiple neurological phenotypes 48,49 . Another 217 strongly associated locus at chromosome 17 (rs4792891) included the Microtubule Associated Protein 218 Tau (MAPT) gene, a Parkinson's disease 50 associated locus. The Mitogen-Activated Protein Kinase 3 219 (MAPK3) gene at 16p11.2, a region known to harbor rare variants which influence neurodevelopmental 220 disorders 51 and language-related phenotypes 52 , was also strongly implicated. We also identified a locus 221 at Glycoprotein M6A (GPM6A), whose gene promoter contains a transcription factor binding site for 222 GATA2, a gene previously related to music phenotypes 37 . 223 224 SNP-based heritability estimates on the liability scale 53 ranged from 13% to 16% when adjusted for a 225 range of estimated population prevalence for atypical beat synchronization (3.0% to 6.5%; 226 Supplementary Table 3; see Supplementary Notes for explanation of prevalence estimates). The  227 observed (unadjusted) genetic variance explained 5% (se=0.002) of the phenotypic variance. 228 229 Gene based GWAS. Gene-based genome-wide association analyses performed with MAGMA yielded 129 230 genes surpassing the threshold of p<2.56x10 -6 (two-tailed; Supplementary Table 4), with top two hits at:  231 CCSER1, in the 4q22 region in proximity to genes previously associated with multiple musicality 232 phenotypes 54 , and VRK2 (converging with the top locus in our SNP-based GWAS). Within these 233 associations, we examined potential replication of 29 genetic associations with musicality in humans 234 from prior reports 37,54,55 ; none reached significance after genome-wide correction (Supplementary Table  235 5, Supplementary Notes), neither independently, nor as a gene-set (p=0.297). 236 237 In silico functional analyses.

238
Gene property and gene set enrichment analyses. To understand the biological functions and gene 239 expression associations of beat synchronization, we performed gene set analysis (GSA) and gene 240 property enrichment analyses 56 on the gene-based p-values, using MAGMA 57 implemented in FUMA 58 . 241 Results of conditional gene property analysis (based on GTEx data tissue types 59 and controlling for 242 average expression) demonstrated that the genetic architecture of beat synchronization was 243 significantly enriched in genes expressed in brain tissues ( Figure 3A), including cortex, cerebellum, and 244 basal ganglia (putamen, caudate and nucleus accumbens), converging with subcortical and cortical 245 regions supporting beat perception and synchronization 34 . 246 247 To further examine potential biological functions associated with beat synchronization, we performed 248 exploratory GSA 57 (Supplementary Table 6). The genetic architecture of beat synchronization was 249 enriched for two gene sets associated with nervous system function: gene sets for synaptic membrane 250 adhesion (p=1.01 x 10 -7 ) and synaptic adhesion-like molecules (p=8.35x 10 -7 ). 251 252 Partitioned Heritability. Complementing these gene-based enrichment analyses, we also performed 253 stratified LDSC 60 on the GWAS results to partition heritability according to genomic properties, using 254 specific functional categories to gain insight into the types of variation that contribute most to beat 255 synchronization. Among broad SNP annotation categories 61 (Supplementary Table 7), we found 256 enrichment (all p<9.6x10 -4 ) of: regions conserved in mammals (considered under purifying selection 62 ), 257 regulatory regions marked by acetylation of histone H3 at lysine 9 (H3K9ac; a marker for active 258 chromatin, and monomethylation of histone H3 at lysine 4 (H3K4me1; a marker for enhancers), 259 supporting the hypothesis that identified associations may affect gene regulation. We next used LDSC-260 Specifically Enriched Genes (LDSC-SEG 63 ) to determine whether genes expressed in specific cell-or 261 tissue-types (conditional to the other annotations) are enriched for beat synchronization-associated 262 variants. For tissue-specific annotations of active chromatin and enhancers (marked by H3K9ac, 263 H3K27ac, DNase hypersensitivity sites and H3K4me1), heritability was enriched in central-nervous-264 system-and skeletal muscle-specific regulatory regions (Supplementary 6 and 7 respectively. Enrichment in brain-specific regulatory elements, in multiple fetal and adult tissue-267 specific elements as well as CNS-specific cell cultures, are shown in Figure 3B.
Adult brain tissue Cell culture Fetal brain tissue B cell cultures in teal, and in fetal brain tissue shown in dark purple. The graph shows -log-10 p-values are on y-axis, with tissue and marker type 280 on x-axis. The dotted line shows p-value threshold for significant enrichment after Bonferroni correction for number of gene sets tested.

282 283 Evolutionary Analyses 284
Given evolutionary hypotheses about the origins of rhythm 4,18,64 , we evaluated the contribution of 285 regions of the human genome that have experienced significant human-specific shifts in evolutionary 286 pressure, using stratified LDSC 53,60 . In particular, we analyzed the contribution to beat synchronization 287 heritability from variants in genomic loci that are conserved across non-human species, but have 288 elevated substitution rate on the human lineage 65 . Many of these human accelerated regions (HARs) 289 play roles in human-enriched traits 66 , including cognition 67 . Two variants significantly (p < 5x10 -8 ) 290 associated with beat synchronization (rs14316 at locus 66, rs1464791 at locus 20) fall within HARs. This 291 is 11.2 times more overlap than expected by chance (µ=0.178 overlaps; p=0.017, based on 10,000 292 permutations). The rs1464791 variant is near GBE1, a gene associated with neuromuscular disease 68 , 293 reaction time 69 and cognitive impairments 70 . Applying LDSC to consider the full set of association 294 statistics, we find that genetic variants in HARs contribute 2.26 times more to the observed heritability 295 of beat synchronization than would be expected if heritability were distributed uniformly across variants 296 (p = 0.14). Given the small number of common variants within HARs, this stratified heritability analysis is 297 substantially underpowered (0.17% of variants considered are in HARs). The general agreement of these 298 two approaches supports the enrichment of functional variation relevant to beat synchronization in 299 HARs. We also evaluated the contribution of genetic variants detected in the Neanderthal genome to 300 the heritability of beat synchronization (Supplementary Notes and Supplementary Results included positive correlations with motor function (grip strength and usual walking pace) and 317 heaviness of smoking, and negative correlations with risk-taking and smoking initiation. There were two 318 correlations with hearing traits (positive correlation with tinnitus and negative correlation with hearing 319 difficulty). From the cognitive traits, processing speed (faster perceptual motor speed) was genetically 320 correlated with beat synchronization, in addition to executive function -shifting (from a GWAS of trail-321 making, a task that involves complex processing speed). There were multiple associations with other 322 biological rhythms: breathing function traits (positive associations with peak expiratory flow, forced 323 expiratory volume, forced vital capacity, and a negative correlation with shortness of breath) and 324 negative associations with sleep-related traits (insomnia and morning chronotype). 325 326 327 328

337
Genomic Structural Equation Modeling (SEM). We conducted Genomic SEM 73 to examine whether 338 specific associations between beat synchronization and a subset of associated traits (e.g., 339 musculoskeletal strength, walking pace, breathing function, and processing speed 74-76 ) that are known 340 to be related among each other in prior research 74-76 represent distinct genetic relationships or share a 341 common set of genetic influences with beat synchronization. The best fitting model, displayed in Figure  342 5, included a common genetic factor that accounted for genetic correlations among beat 343 synchronization, grip strength, processing speed, usual walking pace, and expiratory flow. This common 344 factor explained 11.6% of total variance in the beat synchronization GWAS and 9.6-25.0% of the 345 variance in the other GWASs (see Supplementary Notes). 346

375
Our GWAS revealed highly polygenic architecture of the human capacity to synchronize to a musical 376 beat, representing a significant advancement of our understanding of the genomic basis of a musicality 377 phenotype. Heritability of beat synchronization is enriched for functions of the central nervous system 378 on a number of dimensions: SNPs involved in neural development and brain-specific regulatory regions 379 of the genome; genes involved in synaptic function; and gene expression in cortical and subcortical brain 380 tissues aligned with auditory-motor regions previously linked beat perception and synchronization 34 . 381 Polygenic scores for beat synchronization were associated with self-identified musicians in a separate 382 cohort, showing that the GWAS taps into the larger construct of musicality. Genetic correlations pointed 383 to pleiotropy between beat synchronization and biological rhythms (including breathing function, 384 walking pace, and chronotype), paving the way to a better understanding of the biological 385 underpinnings of musicality and its health relevance. 386 387 In a series of phenotypic experiments, we also demonstrate that self-reported beat 388 synchronization/rhythm measures can be used in large-scale population-based studies as suitable 389 proxies for measuring individual differences in beat synchronization. Our findings indicate that the 390 GWAS phenotype beat synchronization question was highly related to beat synchronization task 391 performance (i.e., accuracy in tapping along to musical excerpts). Clearly the self-report is an imperfect 392 correlate of beat synchronization; nevertheless, we demonstrate that it is a suitable proxy for very large-393 scale studies in which task administration is impractical. Furthermore, the GWAS phenotype is also 394 significantly associated with: rhythm perception task performance 46 , a multi-item Rhythm questionnaire, 395 and a well-established assessment of musical sophistication 43 . These results also converge with earlier 396 work showing shared variance among task performance of beat synchronization, rhythm perception, 397 and musical engagement/training 44,77-80 . The phenotypic associations were robust to demographic 398 factors and self-confidence, and were not driven by the presence of professional musicians in the 399 sample. These phenotype validation studies represent critical groundwork (see 81 ) enabling brief rhythm 400 self-report questionnaires to be deployed online in large-scale population genetic cohorts. 401 402 With sixty-nine loci surpassing the threshold for genome-wide significance, the polygenic architecture of 403 the beat synchronization GWAS aligns with expectations for complex traits 82 We examined potential mechanisms linking genetic variation to neural architecture of the beat 415 synchronization trait using multiple in-silico enrichment methods. Results showed enrichment of the 416 heritability of beat synchronization in many brain tissues including cerebellum, dorso-lateral prefrontal 417 cortex, inferior temporal lobe, and basal ganglia nuclei (i.e., putamen, caudate, and nucleus accumbens). 418 This pattern of results likely reflects a genetic contribution to subcortical-cortical networks underlying 419 musical rhythm perception and production 32,34 ; furthermore, enrichment of brain-tissue-specific 420 enhancers and active-regulatory regions in tandem with gene expression enrichments in brain tissue 421 suggest that regions of the genome involved in regulation of gene expression within the beat perception 422 and synchronization network contribute to phenotypic variance. Moreover, the partitioning heritability 423 chromatin results showed an enrichment in both fetal and adult brain tissues, suggesting that beat 424 synchronization may be the result of neurodevelopmental or basic brain processes. Gene set 425 enrichments were also observed for synaptic function in the nervous system. Moving in synchrony to a musical beat encompasses beat perception and extraction, motor periodicity, 457 meter perception, and auditory-motor entrainment (see 4,32,98 and Glossary in Supplementary Notes). 458 Despite this complexity, beat is a highly frequent feature of many musical systems 1,3,26 . Indeed, we 459 found that the heritability of beat synchronization is enriched in auditory-motor regions known to be 460 active during rhythm tasks. 34 It should be noted that beat perception and production does not depend 461 on musical training or music genre, and atypical beat synchronization is not linked to lack of music 462 exposure 99 . A limitation of the current work is the restriction of the genetic sample to a European 463 ancestry (due to GWAS methodology constraints); investigating beat synchronization, musicality, and 464 cross-trait correlations in populations of non-European ancestry should be a future priority for capturing 465 the spectra of musicality traits in a wider range of ethnic, cultural and socio-economic contexts (see 100 ). 466 Regrettably, early research on individual differences in musicality in the early 1900's was pursued not 467 only using what we now recognize as highly culturally biased assessments, but also explicitly through the 468 lens of eugenics (see 101 ), similar to early research on individual differences in cognition. We strongly 469 condemn the intent and design of those studies, and emphasize that the value of this work arises not 470 from the hypothetical ranking of interindividual differences in beat synchronization (indeed, genomic  471  associations with beat synchronization cannot be used to make deterministic predictions about  472 individual abilities or aptitudes 102, 103 ). Rather the value arises from discovering that the shared 473 experience of rhythm, different though it is across cultures, is, in part, hardwired into our human 474 genome. Furthermore, new knowledge on the genetic basis of musicality must be used ethically and 475 fairly for research discovery and never for harm (e.g., discouraging a child from accessing musical 476 activities). 477 478 We replicated previous findings implicating location 4q22.1 in musicality-related traits 36,55 (CCSER1 was 479 the top-associated gene in our MAGMA analysis), but did not find support for previous gene associations 480 from a set of genes that was drawn from prior candidate-gene, linkage, and GWAS studies with 481 relatively small samples 54 . This is potentially due to well-known methodological problems with these 482 methods particularly when applied to complex traits in small samples 104 . Without a second comparably 483 sized GWAS available within which to conduct replication of the loci discovered in the primary GWAS, 484 we were still able to demonstrate generalizability of these results by showing that PGS for beat 485 synchronization predicts a musical trait in a separate biobank sample. The GWAS results of beat 486 synchronization were nearly identical even after conditioning the results on GWASs of educational 487 attainment and general cognition (g-factor); these results align with twin findings of specific genetic 488 effects of rhythmic aptitude over and above any common genetic influences between rhythm and non-489 verbal cognitive variables 39,105 . Moreover, given both the likely capturing of genetic variation related to 490 SES 106 by the educational attainment GWAS summary stats, and the observation that our beat 491 synchronization GWAS loci are robust to educational attainment, SES is unlikely to play a major role in 492 our findings. 493 494 Our cross-trait explorations revealed pleiotropic effects between beat synchronization and several 495 breathing-related phenotypes (peak expiratory flow, forced vital capacity, forced expiratory volume, and 496 shortness of breath). We demonstrated phenotypically that more accurate beat synchronization task 497 performance was related to lower likelihood of shortness of breath, mirroring the genetic correlations 498 between beat synchronization and breathing function. In light of our genetic correlation between beat 499 synchronization and three categories of traits (breathing, motor, and cognitive functions) previously 500 shown to be genetically interrelated during the aging process 74,75 , we used genomic SEM to uncover 501 shared genetic variance among beat synchronization and enhanced breathing function, greater grip 502 strength, faster walking pace, and faster processing speed. Poor beat synchronization could be tied to 503 certain health risks during aging, in light of other genetic and epidemiological work showing that lung 504 function decline predicts later declines in motor function and psychomotor speed in older adults 107-110 . 505 506 The genetic correlation results suggest that beat synchronization shares common biology with a 507 constellation of health traits, converging with the growing literature on the overlapping biomechanical 508 and perceptual mechanisms of rhythms harnessed during synchronization, communication, muscle 509 tensioning, and breathing; these relationships start very early in development 111,112 . The cerebellum in 510 particular plays important roles in the control of coordinated movement, balance, respiration, dance, 511 and even rhythm perception during passive listening to music 33 . Indeed, our rhythm-related traits multi-512 variate GWAS demonstrated enriched heritability of genes expressed in Cerebellar tissue, potentially of 513 note in relation to experimental findings of functional synchronization of respiratory and upper limb 514 movements during vocalization 5 . Moreover, "beat gestures" in speech involve the cerebellum 113 and are 515 inextricably linked to respiration, upper limb movement, and postural control, all of which may be 516 biomechanically related to tapping or clapping to music. 517 518 Another dimension of biological chronometry was captured in the genetic correlation between 519 chronotype and beat synchronization, which we replicated phenotypically (individuals who self-520 identified as 'evening people' tended to tap more accurately to music, even after removing professional 521 musicians from the analysis). These results complement recent evidence of the increased prevalence of 522 eveningness in musicians 114 , indicating that the relationship between chronotype and musicianship 523 cannot solely be explained by environment (i.e., nocturnal job demands of professional musicians), but 524 that also other shared biological factors may play a role. Given the genetic correlation between beat 525 synchronization and lowered incidence of insomnia, the relationship between regulation of sleep, 526 musicality, and rhythm represents an area for further exploration. 527 528 Our GWAS effectively identified alleles at 69 separate loci differentially associated with typical vs. 529 atypical beat synchronization, complementing existing evidence of underlying neural 530 mechanisms 77,79,80,99 . Future genetic studies could study beat synchronization as a continuous trait, 531 either through self-report or internet-based task paradigms (i.e., REPP 47 ). Prior literature on liability 532 threshold models has shown that case-control GWAS of complex traits yield similar results to those 533 obtained through continuous phenotypic measures (e.g., the genetic architecture of continuous 534 measures of psychiatric symptoms is highly similar to the genetic architecture of cases versus 535 controls 115 ). Moreover, the use here of a population-based control group that is not "super-normal" 536 increases the likelihood that the genetic correlations that we uncovered are reliable and not biased 537 upward 116 . 538 539 Taken together, our results advance knowledge of the biological basis of beat synchronization by 540 identifying genomic regions associated with individual differences in beat synchronization, estimating its 541 cumulative SNP-based heritability, successfully applying a polygenic score model in a separate genetic 542 sample, and exploring the enrichment of heritability in genes tied to central nervous system function. 543 Movement in synchrony with a musical beat is a fundamental feature of music, and sensitivity to the 544 beat emerges early in development, supporting childhood development in numerous ways 3,11,27,30 and 545 with importance over the lifespan 117 . We have elucidated the genetic architecture of beat 546 synchronization and revealed its health relevance through cross-trait analyses. This study also provides a 547 solid foundation for future exploration of how specific genetic variants contribute to neural mechanisms 548 of entrainment, prediction, and reward harnessed during musical interactions 118 . 549 550

551
Phenotype validation studies 552 Phenotype Validation Experiment 1. 553 Overview. Phenotype Validation Experiment 1 was designed to determine if self-reported rhythm 554 abilities measured with the question used in the GWAS (i.e., 'Can you clap in time with a musical beat?') 555 would be associated with task-based rhythm perception performance. The study was conducted in 556 Amazon's Mechanical Turk and received ethical approval from the Columbia University Institutional 557 Review Board; participants gave their written informed consent, and the research complied with all 558 relevant ethical regulations. We selected the Beat-based Advantage paradigm as a rhythm 559 discrimination (perception) test due to its design of stimuli with simple and complex meter 119 and prior 560 history investigating individual differences in rhythm perception in a variety of brain and behavioural 561 studies in adults and children with typical and atypical development 46,120-122 as well as feasibility for 562 internet-based adaptation. A questionnaire (self-report questions) was administered prior to the 563 perception task, to avoid biasing participant self-report responses by how they perceived their own task 564 performance. See Supplementary Notes for additional details on procedure, compensation, and self-565 report questionnaire. 566 567 Participants. The study sample was N=724 participants recruited anonymously in Amazon's Mechanical 568 Turk. All consented and passed a common headphone check 123 that guarantees good listening 569 conditions and the ability to follow basic instructions; this test also effectively filters out bots. 570 Participants (333 females; 387 males; 4 self-reported "other") were 18-73 years old (mean = 36.1 years, 571 SD=10.9) with 0-45 years of self-reported musical experience (mean 3.7 years, SD=5.7), representing an 572 average degree of musical experience (see norms in 43 ); demographics are reported in Table 1 (note that  573 n=3 did not report their age). 574 575 Rhythm Perception Task. Stimuli for the rhythm perception task consisted of 32 rhythms drawn from 576 prior work 46,119 . For each participant, we randomized with probability of one half the occurrence of 577 "simple" rhythms (strong beat-based metrical structure and generally easier to discriminate) and 578 "complex" rhythms (weaker metrical structure due to syncopation and generally more challenging to 579 discriminate). Each rhythm was presented using pure tone stimuli in one of 6 frequencies (294, 353, 411, 580 470, 528, and 587 Hz, selected at random), and one of 4 durations (ISI of 220, 230, 240, and 250 ms). 581 Each trial consisted of 3 rhythms separated by 1500 ms of silence; there were 32 trials of the task. The 582 two first presentations were always identical, and in half of the trials (counterbalanced) the third rhythm 583 was also identical (standard condition); in the other half of the trials, the rhythm differed by having one 584 interval swapped (deviant condition). The pairings and structure of standard and deviant trials were 585 taken from 46 . Participants were instructed that in each trial, they would listen to the series of three 586 rhythms (the first two were always identical, and the third could be the same or different), and they had 587 to indicate if the third rhythm was the same or different (see Supplementary Figure 2 Overview. 616 The aims of Phenotype Experiment 2 were two-fold: 1) to validate self-reported beat synchronization 617 phenotype as a proxy for objectively measured beat synchronization ability, and 2) to explore 618 phenotypic associations between rhythm/beat synchronization and assorted traits found to be 619 genetically correlated with beat synchronization. Phenotype Experiment 2 was pre-registered with Open 620 Science framework (https://osf.io/exr2t) on July 8, 2020, prior to data collection. This internet-based 621 study consisted of a beat synchronization task to assess the accuracy of participants' tapping in time 622 with musical excerpts, and a series of questionnaires assessing self-reported rhythm, musicality/music 623 engagement, selected health traits, confidence as a personality trait, and demographics. We used 624 REPP 47 to measure participants' tapping responses online with high temporal fidelity. The item from the 625 GWAS study, "Can you clap in time with a musical beat?" with possible responses: Yes/No/I'm not sure, 626 is referred to as the "target question." 627 628 We tested the following hypotheses: H1: Self-report responses to the target question will be correlated 629 with beat synchronization task performance (i.e., accuracy of tapping to the beat of music), such that 630 individuals who respond Yes to the "target question" are predicted to tap more accurately to the beat of 631 musical excerpts (i.e., they will have lower standard deviation of asynchrony than individuals who 632 respond No to the target question). H1a: Self-report on a highly similar self-report question ("I can tap in 633 time with a musical beat") with responses on a 7-point agreement Likert scale are predicted to be 634 correlated with tapping accuracy. H2a: The target question will be associated with broader rhythm 635 ability/engagement (measured with a rhythm scale from seven other self-report questions). H2b: Beat 636 synchronization task performance reflects broader self-reported rhythm ability/engagement. H3: To 637 examine whether confidence (either as a personality trait or sureness in one's own task performance) 638 affects the reliability of self-reported beat synchronization. H4: Selected traits found to be genetically 639 correlated with beat synchronization in the GWAS will be phenotypically correlated with beat 640 synchronization task performance and the Rhythm Scale. Specifically: better beat/rhythm is correlated 641 with evening chronotype (H4a), less shortness of breath (H4b), more tinnitus and loud music exposure 642 (H4c), and more smoking (H4d); and that these associations would survive controlling for age, sex, and 643 education (H4e). H5. Responses to the target question will be positively correlated with musical 644 engagement measured with the Gold-MSI. H6. The associations in H4 would interact with being a 645 musician, or more generally, with musical engagement. 646 647 Participants. A total of N=1,412 individuals met participation criteria outlined in the pre-registration 648 (including passing the attention check item and not abandoning the study before completion). The study 649 took place in Amazon Mechanical Turk and all participants provided informed consent in accordance 650 with the Max Planck Society Ethics Council's approved protocol; the research complied with all relevant 651 ethical regulations. Participants (728 females; 678 males; 6 prefer not answer) were 18-77 years old 652 (mean=36.3 years, SD=11.9) and had of 1-2 years of self-reported musical experience (Table 1). To 653 ensure that the tapping technology measured beat synchronization with high temporal fidelity, it was 654 crucial that participants complied with instructions to perform the tapping task (e.g., using the laptop 655 speakers instead of headphones, with minimal background noise, etc.), and also used hardware and 656 software without any technical issues that would preclude the recording signal (e.g., malfunctioning 657 speakers or microphones, or the use of strong noise cancellation technology; see 47 ). Thus, several 658 precautions, including calibration tests and practice trials, were taken to make sure the tapping 659 technology would work effectively, excluding cases that did not meet the requirements (see 660 Supplementary Notes for details). A subset of n=542 had appropriate hardware to complete all parts of 661 the study (including the tapping tests). Questionnaires were administered in the full sample of 662 participants. Sample demographics are reported in Table 1. Demographics of the participants that 663 completed the tapping experiment was highly similar to the full sample, as shown in the table;  664 furthermore, 65.3% of the full sample and 64.9% of tapping sample had a Bachelor's degree or higher. 665 666 Data collection for Phenotype Experiment 2. 667 The first questionnaire included self-report items, including the "target question," and also covering a 668 variety of musical, health, and interest phenotypes. The health phenotype questions were chosen from 669 phenotypes (chronotype, smoking, shortness of breath, and tinnitus) found to be genetically correlated 670 with beat synchronization in our genetic analyses. Rhythm questions were selected for their particular 671 relevance to various aspects of interacting/engaging with musical rhythm. The order of the questions 672 was fixed for all participants. In addition, we used an attention check item 126 between item 10 and 11, in 673 order to exclude fraudulent responders, such as computer bots or disengaged participants responding 674 randomly to the experiments. The end-questionnaire consisted of items covering the following 675 additional self-report topics: another question about being a musician, a task confidence rating 676 question, a Confidence scale, a 16-item short version of the Gold-MSI 43 (items were chosen due to their 677 high reliability scores: reliability omega = 0.92), and a Demographic questionnaire. Questionnaire items 678 for Phenotype Experiment 2 are listed in the Appendix of the Supplementary Notes. 679 680 Tapping technology. Beat synchronization is particularly challenging to study with online research, 681 where variability in participants' hardware and software can introduce delay in latency and jitter into 682 the recorded time stamps 127,128 . Here we used REPP (see 47 for full details and a validation study of the 683 technology), a robust cross-platform solution for measuring sensorimotor synchronization in online 684 experiments that has high temporal fidelity and can work efficiently using hardware and software 685 available to most participants online. To address core issues related to latency and jitter, REPP uses a 686 free-field recording approach: specifically, the audio stimulus is played through the laptop speakers and 687 the original signal is simultaneously recorded with participants' tapping responses using the built-in 688 microphone. The resulting recording is then analyzed using signal processing techniques to extract and 689 align timing cues with high temporal accuracy. 690 691 Beat synchronization task. The beat synchronization task procedure consisted of three parts: calibration 692 tests, practice phase, and main tapping phase. Participants started with the calibration tests, including a 693 volume test to calibrate the volume of the laptop speakers to a level sufficient for detection by the 694 microphone, a background noise test to make sure participants were in a quiet environment, and a 695 tapping test to help participants practice how to tap on the surface of their laptop in the right level and 696 location to be detected by the microphone. Participants were then presented with the practice phase, 697 which consisted of four 15-second trials of isochronous tapping to a metronome beat (two with inter-698 onset interval of 500 msec and two with inter-onset interval of 600 msec). Following the practice phase, 699 participants were presented with the main tapping task consisting of eight trials (4 musical excerpts, 700 each played twice), with each trial 30 seconds long.  Figure 2 illustrates the instructions for participants. 712 713 Data Analysis. 714 Beat synchronization task performance: Tapping accuracy analysis 715 Let St and Rt be the stimulus and response onsets, respectively. In case of the metronome St are the 716 metronome onset (practice phase) and for music clips St is the target beat location based on the 717 annotations. We define the asynchrony as at=Rt -Rt. Based on prior work 130 , we chose the standard 718 deviation of the asynchrony (std(at)) as our main target interest variable, as this appears to be a robust 719 measure of individual performance and tightly linked to musical abilities 131 . We used metronome onsets 720 to mark the beat metric level in an unambiguous way 132 . We emphasize that the metronome onsets 721 were only physically present during the beginning and end of each clip. We used only the participant-722 produced asynchronies during the epoch at beats in which the guiding metronome was not present, in 723 order to test the ability of the participants to synchronize to music without the metronome sounds 724 (results were nearly identical when we included all onsets including the ones where physical metronome 725 onsets were present). For the main test scores, we used the asynchronies computed relative to the 726 virtual beat locations computed from prior human annotators in MIREX. We also computed vector 727 length in order to confirm key associations of interest between the target question and beat 728 synchronization accuracy (See Supplementary Notes). 729 730 Regression analyses 731 In accordance with the OSF preregistration, we examined whether responses to self-reported beat 732 synchronization phenotype were associated with objectively-measured tapping accuracy, other self-733 reported measures of rhythm ability, confidence, and/or musical sophistication using logistic regression 734 and McFadden's R 2 (for H1, H2a, H3, and H5) and linear regression (for H1a and H2b). Likewise, we used 735 linear regression to examine potential replication of cross-trait associations uncovered by genetic 736 analyses (H4a-d), to examine whether musical background interacted with the above associations (H6). 737 Analyses were conducted in R version 3.5.1 133 . As described in our preregistration, individuals were 738 recruited using MTurk and were included unless they failed and attention check item or abandoned the 739 experiment before completing the study (N=1,412). Usable tapping data was available for n=542 740 individuals. The majority of exclusions were due to technical reasons detected by REPP's signal 741 processing pipeline during the practice trials (e.g., poor signal, noisy environment, wearing headphone, 742 issues with laptop microphone, or people not tapping at all), but some additional subjects (n=19) were 743 excluded for not having enough usable trials during data analysis. Missing covariates were handled using 744 pair-wise deletion. Exclusion criteria are detailed in the Supplementary Notes. 745 746 GWAS of beat synchronization. 747 Genome-wide association study summary statistics were generated from data acquired by personal 748 genetics company 23andMe, Inc. Phenotypic status was based on responses to an English-language 749 online questionnaire in which individuals self-reported "Yes" (cases) or "No" (controls) to the question 750 'Can you clap in time with a musical beat?". Individuals who responded "I'm not sure" were excluded 751 from the genomic dataset as their data was not available.  Table  794 8). The X chromosome was not included in these analyses or any subsequent analyses using LDSC, given 795 that the file that is required for LDSC analysis (w_hm3_snplist) does not include chromosome X SNPs. 796 797 Evolutionary analyses. 798 The set of human accelerated regions (HARs) was taken from 65 . All variants in perfect LD (r 2 = 1.0 in 1000 799 Genomes European participants) with variants in HARs were considered in the analysis. Similarly, 800 variants tagging Neanderthal introgressed haplotypes were defined as in 136 . All variants in perfect LD 801 with a Neanderthal tag SNP were considered Neanderthal variants. For each set, we performed 802 stratified LDSC (v1.0.0) with European LD scores and the baseline LD-score annotations v2.1. The 803 heritability enrichment is defined as the proportion of heritability explained by SNPs in the annotation 804 divided by the proportion of SNPs in the annotation. Standard effect size (), which quantifies the effects 805 unique to the annotation, is the proportionate change in per-SNP heritability associated with a one 806 standard deviation increase in the value of the annotation, conditional on other annotations in the 807 baseline v2.1 model 62 . To determine the expected number of overlaps between the N loci significantly 808 associated with beat synchronization and HARs, we computed all overlaps between these sets of 809 genomic regions (in hg19 coordinates) using bedtools2 137 . We then randomly shuffled the locations of 810 HARs around the genome choosing segments of equal lengths and avoiding gaps in the genome 811 assembly. We repeated this process 10,000 times and for each iteration computed the number of 812 overlaps observed with the significantly associated loci.