Imaging genetics of language network functional connectivity reveals links with language-related abilities, dyslexia and handedness

The study investigates the genetic architecture underlying functional connectivity of the human brain’s language network and its left-right asymmetry. Language processing is typically left-lateralized in ~85% of adults, and language-related abilities, disorders (e.g., dyslexia), and hemispheric specialization show substantial heritability and early developmental origins. The authors aim to identify common and rare genetic variants associated with resting-state functional connectivity among bilateral regions of a consensus sentence-level language network, to test genetic associations with hemispheric differences in connectivity, and to relate polygenic dispositions for language-related abilities, dyslexia, and left-handedness to variation in connectivity and asymmetry. Understanding these genetic influences may illuminate prenatal developmental mechanisms that shape language network organization and individual differences in related cognitive and behavioral traits.

Prior work shows high heritability for language-related cognition and genetic contributions to language-related neurodevelopmental disorders (e.g., dyslexia, developmental language disorder). GWAS have implicated loci for reading/language performance, dyslexia, brain structural asymmetry, and handedness, with many implicated genes most strongly expressed prenatally, suggesting early developmental mechanisms. Task-free (resting-state) functional connectivity predicts task activations and often yields more heritable measures than task-based fMRI, making it suitable for genetics. A previous GWAS of functional language connectivity (Mekki et al., 2022) used mostly left-hemisphere regions without explicit hemispheric analyses. The present study leverages a bilateral atlas (SENSAAS-derived regions with homotopies) to assess intra- and interhemispheric connections and hemispheric differences, addressing gaps regarding lateralization. Prior exploratory work suggested rare variants (e.g., in actin cytoskeleton genes) might influence hemispheric dominance, but large-scale exome-based analyses of functional connectivity were lacking.

Participants: 29,681 UK Biobank participants (ages 45–82) with SNP array data, exome sequences, and resting-state fMRI, restricted to a white British ancestry cluster after extensive sample-level QC (imaging quality, sex concordance, aneuploidy, heterozygosity/missingness, relatedness). Imaging: Resting-state fMRI collected on Siemens 3T Skyra (TR=0.735 s; 6 min; 490 timepoints). UK Biobank preprocessing included motion correction, normalization, high-pass filtering, unwarping, gradient distortion correction, ICA-FIX denoising, and registration to MNI152 NLin6. Connectivity derivation: Using the AICHA atlas with defined homotopies and focusing on 18 left-hemisphere SENSAAS core language regions and their right homotopes (36 ROIs). Pearson correlations (Fisher z) computed for all ROI pairs, yielding 630 edges (intra- and interhemispheric). For intrahemispheric homotopic pairs, 153 hemispheric differences (L−R) were computed. Outliers (6×IQR) removed. Heritability: GREML in GCTA on 783 IDPs; only heritable IDPs (p≤0.05) were retained: 629 edges and 103 hemispheric differences. Median SNP-heritability: edges 0.070 (min 0.018, max 0.165); hemispheric differences 0.026 (min 0, max 0.070). Common variant mvGWAS: MOSTest applied to 8,735,699 imputed bi-allelic SNPs (hg19) for 629 edges and separately for 103 hemispheric differences, covarying sex, age, age^2, age×sex, 10 ancestry PCs, genotype array, and scanner-related measures (positions, IT-SNR, head motion). Genome-wide significance: p≤5×10^-8. Post-GWAS annotation: FUMA for loci/genes (positional, eQTL, chromatin interactions), MAGMA for gene/tissue/developmental expression analyses (BrainSpan, GTEx), and gene set analyses. Descriptive directionality analyses: t-tests on univariate betas to summarize global effects (overall connectivity up/down; left vs right intrahemispheric; intra- vs interhemispheric). Polygenic score analyses: PRS-CS with external GWAS summaries for language-related abilities (GenLang multivariate: word reading, nonword reading, spelling, phoneme awareness), dyslexia (23andMe), and left-handedness (UKB non-imaging). Scores were residualized for covariates (as above), quantile-normalized, then related to brain phenotypes using canonical correlation analysis (CCA) with 10,000 permutations to assess significance; loadings interpreted and directionality tested with t-tests. Rare variant analyses: Exome data processed with stringent QC; variants annotated with snpEff and CADD. Two variant filters: Strict (high-impact or moderate with CADD≥20; excluding 5% tail ends of canonical proteins for high-impact) and Broad (including additional moderate/modifier variants with CADD≥1). REGENIE step 1 whole-genome ridge regression with LOCO predictors; step 2 gene-based SKAT-O tests across >18,000 genes for 629 edges and 103 hemispheric differences, separately under Broad/Strict filters. Multivariate exome testing per domain used Tippet’s method (minimum p across phenotypes). Exome-wide empirical significance: p≤2.5×10^-7. Follow-ups: burden tests for direction and variant-level checks. Sensitivity: Additional models included mean whole-brain connectivity/asymmetry as covariates (results unchanged).

- Heritable phenotypes: 629 language-network connectivity edges and 103 hemispheric differences retained (median SNP-h2: 0.070 for edges; 0.026 for hemispheric differences). - Common variant mvGWAS (edges): 14 independent genomic loci reached genome-wide significance (p≤5×10^-8). FUMA mapping implicated 111 genes (40 protein-coding). Developmental expression (MAGMA) showed prenatal enrichment, significant at 21 weeks post-conception. • Example lead SNPs and patterns: - rs35124509 (chr3; exonic in EPHA3): minor allele (C; MAF=0.39) associated on average with reduced connectivity across edges (t=−6.673, p=5.52×10^-11), with heterogeneous edgewise effects. - rs2279829 (chr3; upstream of ZIC4): minor allele (T; MAF=0.21) associated with increased connectivity on average (t=14.606, p=8.27×10^-12). - rs2274224 (chr10; exonic in PLCE1/PLCE1-AS1): C allele (MAF=0.44) showed stronger effects on intra- than interhemispheric connectivity (t=4.588, p=5.41×10^-5). - Common variant mvGWAS (hemispheric differences): 3 independent loci on chromosome 3 were significant. Gene-based mapping implicated EPHA3, TBC1D5, ZIC1, and ZIC4; prenatal expression enrichment replicated. • Lead SNP examples: - rs7625916 (chr3; intergenic in RP11-91A15.1, within EPHA3 locus): heterogeneous hemispheric difference effects; minor allele A (MAF=0.40). - rs2279829 (chr3; upstream ZIC4): minor allele associated with heterogeneous hemispheric difference changes. - rs13321297 (chr3; intronic near TBC1D5): minor allele A (MAF=0.31) associated with a broadly rightward shift in asymmetry (t=−8.767, p=4.314×10^-14). - Polygenic score associations (CCA with permutation testing): • Language-related abilities PS: significant associations with edges (r=0.160, p=3×10^-4) and hemispheric differences (r=0.076, p=9.9×10^-5). Loadings indicated stronger left-hemisphere connectivity and an overall leftward asymmetry shift (t=7.700, p=1.924×10^-11). • Dyslexia PS: significant associations with edges (r=0.177, p=9.9×10^-8) and hemispheric differences (r=0.078, p=2×10^-4). Higher PS linked to increased interhemispheric connectivity (t=7.701, p=5.278×10^-12) and a broadly rightward asymmetry shift. • Left-handedness PS: significant associations with edges (r=0.154, p=2.16×10^-7) and hemispheric differences (r=0.067, p=2.44×10^-3). Higher PS associated with increased interhemispheric (t=−8.583, p=7.258×10^-17) and right intrahemispheric connectivity, corresponding to a rightward asymmetry shift. - Rare variant (exome) gene-based associations: • Language-network edges (Broad filter): NIBAN1 (p=2.356×10^-6), MANEAL (p=1.338×10^-7), SLC25A48 (p=4.263×10^-7), DUSP29 (p=2.494×10^-7), TRIP11 (p=2.183×10^-7). • Hemispheric differences (Strict filter): WDCP (p=2.064×10^-6), DDX25 (p=2.011×10^-6). • Burden analyses: - MANEAL: greater rare variant burden associated with generally decreased connectivity (t=−31.542, p=1.356×10^-131). - DDX25: greater burden associated with a broadly rightward shift of intrahemispheric asymmetry (t=−11.809, p=8.458×10^-21). - Many common-variant loci overlap prior brain connectivity/structure GWAS; prenatal expression enrichment suggests early developmental influences on the language network.

The findings demonstrate that both common and rare genetic variation contributes to individual differences in the functional organization and lateralization of the adult human language network. Multivariate GWAS identified 14 loci for network connectivity and three loci for hemispheric differences, including associations near EPHA3, ZIC genes, and TBC1D5—genes involved in neurodevelopmental processes such as axon guidance and receptor trafficking. Gene expression analyses indicate that implicated genes are preferentially expressed prenatally (notably around ~21 weeks post-conception), supporting the hypothesis that heritable aspects of language network architecture and asymmetry are established early in development. Polygenic dispositions for language-related abilities, dyslexia, and left-handedness align with shifts in lateralization: higher ability PS relates to stronger left-lateralized connectivity, whereas dyslexia and left-handedness PS relate to more rightward asymmetry and increased interhemispheric coupling. These results bridge genetic influences on cognition/behavior with neurobiological correlates in a large general-population cohort and suggest partially shared genetic architecture between functional and structural connectivity. Although locus-level overlap with GWAS of behavioral traits was limited, polygenic analyses reveal broader shared influences. Overall, the study advances understanding of the genetic basis of language network connectivity and its asymmetries, situating much of the influence in prenatal brain development.

This large-scale imaging genetics study of resting-state language network connectivity and its hemispheric differences identified 14 common-variant loci and suggested seven rare-variant genes associated with functional organization of the language network. Polygenic tendencies toward lower language-related abilities, dyslexia, and left-handedness were linked to reduced leftward asymmetry and increased interhemispheric/right-hemisphere connectivity, whereas higher language-ability PS related to stronger left-lateralization. Developmental expression profiles point to prenatal origins of these genetic effects. Future work should: replicate findings in independent and more diverse cohorts; extend analyses to younger samples to clarify developmental trajectories; develop multivariate rare-variant methods for exome analyses; integrate task-based fMRI and alternative connectomic representations; and explore mechanistic pathways (e.g., ephrin signaling) in model systems.

- Resting-state connectivity is an indirect proxy for language lateralization; full correlations may include indirect effects via third regions. - Hard parcellation can allow anatomical variability to influence functional measures. - Cross-sectional design precludes causal inference regarding relationships among genetic disposition, connectivity asymmetry, and behavior. - The sample consists of middle-aged to older adults; developmental and aging effects cannot be fully addressed. - Analyses focused on a white British ancestry cluster, limiting generalizability across ancestries. - No discovery-replication split due to maximizing power in a single large sample; some findings, especially rare-variant associations, are tentative and require replication. - Limited overlap at genome-wide significant loci with behavioral GWAS may reflect power and phenotype differences. - Multivariate frameworks for rare variants are lacking; exome tests used mass univariate SKAT-O with Tippet’s method. - UK Biobank volunteer bias may limit representativeness.