An exciting new study reveals the hidden genetic architecture of quantitative ability, showing how brain wiring and signaling shape math skills independently of general intelligence.
Study: A genetic common factor underlying self-reported math ability and highest math class taken. Image Credit: Aree_S / Shutterstock
In a recent study published in the journal Molecular Psychiatry, researchers identified genetic variants and biological pathways underlying a quantitative-ability factor, distinct from general intelligence (g) and non-cognitive educational skills (NonCog), using multivariate genome-wide data.
Background
One number can change a life: the right math class can open doors to majors, careers, and confidence. Yet comfort with numbers varies, and not all of that difference reflects g. Quantitative ability, the knack for reasoning about numbers, change, and structure, may be distinct. Genome-wide association studies (GWAS) scan deoxyribonucleic acid (DNA) for single-nucleotide polymorphisms (SNPs), but most focus on g. If quantitative ability has a different biology, signals should appear beyond g and schooling.
Understanding biology can guide teaching that matches students’ strengths. Gene-to-tissue analyses can highlight brain and synaptic mechanisms. The authors also note that these indicators can capture interests and educational opportunity as well as ability, and residual variances were fixed for model identification, which can affect relative loadings. Further research should separate ability from interests and opportunity.
About the Study
Researchers modeled a latent quantitative factor beneath two indicators, self-reported math ability and the highest math class taken, using Genomic structural equation modeling (Genomic SEM). GWAS summary statistics were obtained from 23andMe, Inc. participants (n = 564,698; class n = 430,445). Genetic covariances were estimated by bivariate LD Score regression (LDSC).
The model removed variance shared with cognitive performance (CP) and educational attainment (EA), so the quantitative factor was orthogonal to g and NonCog. A genome-wide association analysis of the latent factor was run in Genomic SEM. Lead SNPs were identified with Genome-wide Complex Trait Analysis-Conditional and Joint analysis (GCTA-COJO) at P<5×10⁻⁸; heterogeneity was tested with the Single-nucleotide polymorphism heterogeneity (QSNP) statistic.
Prior associations were queried in the GWAS Catalog. Genetic correlations with phenotypes, including those from the United Kingdom Biobank (UKB) job codes, were estimated using LDSC. Polygenic scores (PGS) were derived using Polygenic Risk Scores-Continuous Shrinkage (PRS-CS) and validated in the Minnesota Center for Twin and Family Research (MCTFR), predicting outcomes on the Wide Range Achievement Test (WRAT); within-family models included parental PGS.
Tissue and pathway enrichment used stratified LD Score regression (S-LDSC) with Genotype-Tissue Expression (GTEx) data, Polygenic Priority Score (PoPS), Expression-Prioritized Integration for Complex Traits (DEPICT), Protein ANalysis THrough Evolutionary Relationships (PANTHER), and Mapping Gene-Set Annotations (MAGMA).
Study Results
Genomic factor analysis showed that the two math indicators were strongly genetically correlated with CP and EA, motivating a model that partialed CP and EA to isolate a quantitative factor orthogonal to g and NonCog. The factor model fit the genetic covariance matrix well (comparative fit index 0.996; standardized root-mean-square residual 0.0195).
In a genome-wide association analysis of the latent factor, 53 SNPs reached genome-wide significance. Inflation was modest (mean χ²≈1.64), and the LD Score regression intercept (0.99) indicated negligible confounding. QSNP flagged one pleiotropic locus, rs13107325 in SLC39A8, whose effects on the two indicators had opposite signs (negative for the highest math class, positive for self-reported ability), consistent with influences not mediated purely through the factor.
Phenome-wide lookups of lead variants in the GWAS Catalog revealed overlaps with traits spanning internalizing symptoms, sleep, and substance use, and identified 16 novel loci for cognitive characteristics.
Genetic correlation analyses supported the construct’s distinctiveness. The quantitative factor was uncorrelated with the first principal component of school grades (general scholastic ability) and showed a strong negative correlation with the language-math tilt axis (better math at lower values). It was positive for mathematician and software engineer/programmer job codes in the UKB, and negative for vocations in verbal persuasion (e.g., writer/poet). These are genetic correlations rather than causal effects and should not be interpreted as deterministic predictions of occupations.
Correlations with anthropometry were minimal, including a small positive association with body mass index. Psychiatric patterns included negative correlations with attention deficit hyperactivity disorder (ADHD), the general factors of externalizing and neuroticism, and major depressive disorder; a negative association with autism spectrum disorder (ASD); a small positive association with schizophrenia; and a slight positive correlation with dyslexia.
PGS for the quantitative factor predicted arithmetic performance on the WRAT in an independent MCTFR sample (ΔR²≈0.39%), but not reading or spelling; similar coefficients appeared in within-family models controlling parental PGS, consistent with limited confounding. The PGS effect was marginal in both size and significance, accounting for less than half a percent of the arithmetic variance (P≈0.01), underscoring its small practical impact.
Biological annotation pointed to brain-based mechanisms. S-LDSC showed enrichment concentrated in central nervous system tissues, with the cerebellar hemisphere and amygdala exceeding a 1.3-fold increase. Although the amygdala surpassed this benchmark, multivariable regression showed a negative weight for amygdala expression, suggesting stronger caution in over-interpreting its rank relative to other cortical and cerebellar tissues.
Gene prioritization with PoPS highlighted processes regulating neuron projection development; prioritized genes included SEMA6D and EFNA5, which are implicated in axon guidance. Feature clusters emphasized the synapse part, messenger ribonucleic acid (mRNA) splicing, which the authors described as a somewhat surprising enrichment, and glutamate receptor activity; related genes, such as GRM8 and NGEF, suggested signaling at excitatory synapses. Consistent with a specialization distinct from global brain size effects, the quantitative factor showed near-zero genetic correlation with brain volume.
Conclusions
This study identifies a genetic signal for quantitative ability separable from g and NonCog. Fifty-three genome-wide loci, minimal stratification, and enrichment in central nervous system tissues support specificity.
Gene-set patterns implicate regulation of neuron projection development, synaptic components, mRNA splicing, and glutamatergic signaling. PGS predicted arithmetic, not reading or spelling, and the factor aligned with science-and-technology occupations and lower liability for internalizing and externalizing behaviors.
The authors emphasize that results should be interpreted cautiously, given the indicator limitations and the small predictive value of PGS. Together, results suggest that neurons connect, communicate, and contribute to quantitative specialization.
Journal reference:
Giannelis, A., Willoughby, E. A., Edwards, T., McGue, M., & Lee, J. J. (2025). A genetic common factor underlying self-reported math ability and highest math class taken. Molecular Psychiatry. DOI: 10.1038/s41380-025-03237-0. Read the study