Language network lateralization is reflected throughout the macroscale functional organization of cortex

The human brain displays broad structural homotopy across hemispheres yet exhibits prominent functional asymmetries, notably in language. Most individuals show left-hemispheric dominance for higher-order language functions anchored in anterior and posterior cortical regions (e.g., Broca’s and Wernicke’s areas), although atypical right-dominant language organization occurs in a substantial minority and is more common among left-handers. Developmentally, language lateralization emerges from more symmetric childhood distributions toward typical left dominance by adolescence, but mechanisms and consequences of lateralization remain unclear. Concurrently, macroscale functional gradients organize cortical networks along axes from unimodal sensorimotor/visual to heteromodal association territories, and these gradients exhibit hemispheric differences. The central question addressed here is whether individual variation in language lateralization is reflected throughout these macroscale functional gradients across cortex, and whether genetic factors contribute to both language dominance and gradient asymmetries. Using a higher-order language atlas and large-scale HCP twin/family data, the study tests how typical versus atypical language dominance relates to global gradient organization and heritability.

Prior research delineates a left-lateralized language network supported by anterior (inferior frontal gyrus/Broca’s) and posterior (posterior superior temporal/Wernicke’s) regions, with atypical right-dominant language present in roughly 2–10% of adults depending on handedness. Early development shows relatively symmetric language organization, transitioning toward left dominance during adolescence. Macroscale functional gradients capture cortical organization, with a principal gradient spanning from unimodal sensory/motor and visual cortex to heteromodal association/default-mode regions; secondary gradients differentiate unimodal systems (e.g., somatomotor vs. visual) and contrast control/frontoparietal with default/sensorimotor systems. Hemispheric differences in these gradients have been reported, and gradient positions relate to cortical microstructure, connectivity, and gene expression. Genetic influences have been established on aspects of network size, connectivity, and topography, with emerging evidence that gradient asymmetries and certain structural asymmetries are heritable and differ by handedness. However, links between language lateralization and whole-cortex gradient organization, and their shared genetic bases, remained insufficiently characterized.

Design and data: Cross-sectional analysis of Human Connectome Project (HCP) S1200 participants with complete 3T language task fMRI and resting-state fMRI (n=995; mean age 28.7 years; 527 females; 110 left-handers). Language lateralization phenotypes were derived using a higher-order language atlas (SENSAAS) and validated clustering; gradient analyses used diffusion map embedding of whole-cortex resting connectivity. Twin/family heritability used SOLAR on n=989 with known kinship (130 MZ twins, 70 DZ twins, 479 non-twin siblings, 110 singletons). Covariates included age, sex, handedness, intracranial volume, and interaction terms as specified. Imaging acquisition: Multiband EPI on Siemens 3T Skyra. Resting-state fMRI: two sessions (REST1/REST2), each with two runs (14:33 min; TR=720 ms; TE=33.1 ms; 2 mm isotropic). Task fMRI: language Story>Math contrast (two 3.8 min runs). Structural: 0.7 mm isotropic. HCP minimal preprocessing applied. Language lateralization features: Five functional measures quantified language network organization: (1) task-evoked BOLD asymmetry at the network level (Story–Math), (2) task-evoked BOLD asymmetry at language hubs (Broca’s and Wernicke’s), (3) resting homotopic interhemispheric connectivity within the language network, (4) resting intrahemispheric language-network average strength asymmetry (left minus right), (5) resting intrahemispheric average strength sum. For rs-fMRI, time series were averaged within 18 SENSAAS language ROIs (AICHA-based); Pearson correlations formed connectivity matrices, Fisher z-transformed and averaged across four scans; strength per hemisphere summed correlations, with sum and asymmetry computed; homotopic interhemispheric connectivity computed between homologous pairs. Participant classification: Agglomerative hierarchical clustering (Euclidean distance; Ward’s linkage) on the five standardized features identified language lateralization phenotypes. Cluster number determined via NbClust and pvclust stability; three clusters retained: strong typical (left-dominant), mild typical (moderate left), atypical (right-dominant). ANCOVA tested differences across features controlling for age, sex, handedness, intracranial volume, and handedness×phenotype interaction; post hoc Tukey or t-tests applied with multiple-comparison correction. Gradient computation: Whole-cortex functional connectivity matrices (384 AICHA parcels) per participant were constructed (averaged across four rs-fMRI scans). Top 10% strongest connections retained (sparsity 0.9); normalized angle kernel formed similarity matrices. Diffusion map embedding extracted gradients (BrainSpace toolbox); first three gradients retained. Individual gradients aligned to group-level via Procrustes rotation (10 iterations) and min-max normalized (0–100). Parcels were assigned to the 7 Yeo canonical networks; for each participant and network, gradient asymmetry was computed as mean left minus mean right gradient value across network parcels. Statistical analysis of gradients: ANCOVA tested effects of language phenotype (typical vs atypical; mild and strong merged for primary contrasts) on gradient asymmetry within each canonical network and gradient, controlling for covariates as above; Bonferroni correction for 7 networks per gradient; post hoc t-tests reported. Heritability: SOLAR variance-components modeling estimated narrow-sense heritability (h^2) for (a) binary language phenotype (typical vs atypical) and (b) gradient asymmetry (left-right difference) for each gradient and for each network, covarying age, sex, age^2, age×sex, age^2×sex, handedness, and intracranial volume; Bonferroni correction across networks. Data/code availability: HCP data publicly available via agreements; SENSAAS atlas and analysis code provided via GitHub/Zenodo.

- Language lateralization phenotypes: Three clusters were identified among 995 participants: strong typical (n=480; strong leftward asymmetry), mild typical (n=433; moderate leftward asymmetry), and atypical (n=82; rightward asymmetry), corresponding to about 8% atypical. Left-handers were overrepresented in the atypical group (26/82). - Feature validation: Atypicals showed rightward task asymmetry at both network and hub levels (Δ_network = −0.96 ± 0.18; Δ_hubs = −1.16 ± 0.26; both p < 1e−4). Strong typicals showed greater leftward asymmetry than mild typicals (e.g., Δ_network strong = 1.74 ± 0.14 vs. mild = 0.70 ± 0.12; p < 1e−4; Δ_hubs strong = 2.64 ± 0.19 vs. mild = 1.17 ± 0.17; p < 1e−4). Resting strength asymmetry indicated bilateral organization in atypicals (mean ≈ 0.009 ± 0.18) vs. leftward in typicals (strong 1.02 ± 0.13, p < 1e−10; mild 0.85 ± 0.12, p < 1e−8). Strength sum and interhemispheric connectivity were higher in strong typical and atypical vs. mild typical (strength sum strong = 12.16 ± 0.39, atypical = 12.16 ± 0.53, mild = 9.40 ± 0.35; interhemispheric r_z strong = 0.61 ± 0.02, atypical = 0.61 ± 0.03, mild = 0.49 ± 0.02; all p < 1e−4; strong vs atypical p > 0.93). - Gradients: The first three group-level gradients explained 22%, 21%, and 14% of the variance (57% total). Gradient 1 spanned unimodal to association/default cortex; Gradient 2 differentiated somatomotor/auditory from visual; Gradient 3 contrasted frontoparietal control vs default/somatomotor. - Typical group gradient asymmetries: For Gradient 1, 5/7 networks were left-lateralized (mean L−R from ~1.1 to ~4.1), somatomotor symmetric (~0.01), and control strongly right-lateralized (mean L−R ≈ −6.21). Gradient 2 showed right-lateralization in control (−2.93), somatomotor (−2.52), limbic (−0.63), default (−0.54), symmetry in visual (−0.24), and left-lateralization in dorsal (0.88) and ventral attention (1.26). Gradient 3 was mostly right-lateralized or symmetric; only visual was left-lateralized (0.53). - Atypical vs typical differences concentrated in association networks: • Gradient 1: Significant main effects in 5/7 networks (p < 0.002). Default (mean L−R atypical ≈ −1.62) and salience/ventral attention (≈ −1.65) shifted from left to right dominance. Dorsal attention (≈ 1.56) and visual (≈ 0.67) showed weakened left dominance. Control showed stronger right dominance (≈ −9.21). Changes largely reflected increased right-hemisphere gradient values (except default, which decreased in left). • Gradient 2: Significant effects in 3 networks (p < 0.006). Default (≈ −1.58) and control (≈ −4.26) increased right dominance due to increased right and decreased left gradient values. Salience/ventral attention became symmetric (≈ −0.17) instead of left-dominant, via decreased left values. Somatomotor (≈ −1.28) and limbic (≈ −1.77) were right-lateralized; dorsal attention (≈ 0.52) and visual (≈ 0.29) were symmetric. • Gradient 3: Four networks showed significant effects (corrected p < 0.01). Default (≈ 9.05), control (≈ 8.51), and limbic (≈ 1.57) became left-asymmetric via increased left and decreased right values; salience/ventral attention bilateralized (≈ 1.25) via decreased right values. Dorsal attention (≈ 1.50), somatomotor (≈ 0.84), and visual (≈ 1.12) showed no significant differences and were slightly leftward/symmetric. - Heritability: Language lateralization phenotype was heritable (h^2 = 11.2%, SE = 6%, p = 0.038). Gradient asymmetry heritability at whole-hemisphere level: G1 h^2 = 14.4% (SE 6%, p = 0.007); G2 h^2 = 2.0% (SE 5%, p = 0.36); G3 h^2 = 24.0% (SE 6%, p < 1e−6). Network-level heritability of gradient asymmetry was higher in heteromodal association networks (mean h^2 ≈ 18.5%, SD 7.7%) than in unimodal networks (mean h^2 ≈ 5.5%, SD 3.8%; p < 1e−3). Somatomotor and visual gradient asymmetries were not significantly heritable; several association networks (e.g., default, control, salience/ventral attention) showed significant h^2 after correction.

The study demonstrates that individual differences in language dominance are mirrored across macroscale functional gradients of the cortex, indicating that hemispheric specialization for a specific cognitive domain (language) is reflected in global network organization. Atypical rightward language lateralization is associated with systematic shifts in gradient lateralization preferentially within heteromodal association networks (default, control, salience/ventral attention, dorsal attention), rather than unimodal systems. This aligns with theories positing that association cortices, which expanded during primate evolution and support complex, internally and externally oriented cognition, exhibit prominent asymmetries that can co-vary with language dominance. Twin-based analyses indicate that both language lateralization and gradient asymmetries have genetic underpinnings, with stronger heritability in association networks, suggesting that aspects of hemispheric specialization are under genetic control. These findings bridge localized hemispheric specialization and distributed cortical gradients, supporting a model in which lateralization of one cognitive system covaries with broader hemispheric differences in information processing streams. The results have implications for understanding typical development, variability linked to handedness, and the neural bases of neurodevelopmental and psychiatric conditions associated with atypical lateralization.

This work establishes that language network lateralization is reflected throughout macroscale cortical gradients, with atypical (rightward) language dominance accompanying widespread, network-specific shifts in gradient asymmetry primarily within association cortex. Both language dominance and gradient asymmetries show significant heritability, especially across heteromodal networks, indicating genetic contributions to hemispheric specialization and global cortical organization. These findings advance a unified framework linking the lateralization of specific cognitive functions with whole-brain gradient architecture. Future research should delineate causal developmental pathways, examine additional lateralized domains (e.g., attention), leverage longitudinal designs to chart developmental trajectories, and explore clinical populations to determine how disruptions in lateralization relate to alterations in global network organization and cognitive outcomes.

- Causality cannot be inferred from cross-sectional data; developmental trajectories linking language lateralization to gradient organization remain unknown. - The HCP includes only one language comprehension task (Story>Math); other language domains (production, reading) and attentional lateralization could not be assessed directly, though comprehension lateralization is a validated marker. - The study focuses on healthy young adults; generalization to clinical populations or different age ranges requires further work. - While analyses show associations across gradients and networks, specific biological mechanisms underlying atypical lateralization and gradient shifts were not identified.