Loading evaluation data…
General cognitive function ('g') reflects a broad capacity for flexible information processing, yet how it is supported by brain-wide structural connectivity remains unclear. We mapped this relationship in 38,824 individuals (26-84 years) across three cohorts, showing that g is supported by a widely distributed white matter network whose macroscopic wiring capacity, microstructural organisation, and age sensitivity contribute in distinct ways. Across streamline count (SC), fractional anisotropy (FA), and mean diffusivity (MD) weightings, meta-analytic associations with g were widespread at global, nodal and edge levels, spanning all cerebral lobes and key subcortical structures and dependent on both inter- and intra-hemispheric connectivity, particularly ipsilateral long-range inter-lobar connections. White matter node-g associations spatially mirrored independent cortical morphometry-g associations, indicating convergence of grey and white matter contributions to cognitive performance. Effect sizes increased with age: MD associations became more negative and FA more positive, particularly in frontal regions. Edge-level findings replicated across cohorts and predicted g in a hold-out sample. Together, these findings indicate that g reflects a distributed structural communication backbone whose integrity becomes increasingly relevant across adulthood.
g is supported by a widely distributed white matter network whose macroscopic wiring capacity, microstructural organisation, and age sensitivity contribute in distinct ways.
functional-support language and age sensitivity outrun contemporaneous observational associations
Meta-analytic associations with g were widespread at global, nodal and edge levels, spanning all cerebral lobes and key subcortical structures and dependent on both inter- and intra-hemispheric connectivity, particularly ipsilateral long-range inter-lobar connections.
widespread associations are reported, but preferential long-range topology was not normalised or directly measured
Effect sizes increased with age: MD associations became more negative and FA more positive, particularly in frontal regions.
cross-sectional moderation was small and significant only in the large UKB cohort
Edge-level findings replicated across cohorts and predicted g in a hold-out sample.
UKB hold-out prediction was stable, although cross-cohort edge agreement was moderate
White matter node-g associations spatially mirrored independent cortical morphometry-g associations, indicating convergence of grey and white matter contributions to cognitive performance.
coefficient correlations use the same cohorts and are partly volume-driven rather than independent evidence
Derived from the full evaluation — not a separate score.
Strengths
The study constructs networks for 38,824 participants and meta-analyses 20,182 across UKB, STRADL, and LBC1936, substantially increasing power over earlier tract-focused work.
↳ Results, opening paragraph; Methods, Participants
SC, FA, and MD associations are examined at global, nodal, and 818-edge scales using latent-g SEMs, FDR correction, and random-effects meta-analysis.
↳ Methods, Cognitive Testing and Statistical Analysis
The paper reports cross-cohort coefficient agreement, a streamline-density analysis addressing SC–brain-volume dependence, neurological exclusions, and UKB hold-out prediction.
↳ Results, Regional node-g associations; Results, Replication; Supplementary sensitivity analysis
Limitations
Edge-level cross-cohort agreement is only r=0.10–0.48, LBC1936 has few FDR-significant edges, and the hold-out is another random half of UKB rather than an independent population.
↳ Abstract; Results, Network edge-g associations; Results, Replication
Counts of significant intra-hemispheric and inter-lobar edges are not normalised to category availability, and streamline length is not analysed. The preferential long-range interpretation therefore exceeds the reported topology summary.
↳ Results, A New Reference Brain Network; Results, Network edge-g associations; Discussion, edge-level interpretation
Moderation is significant only in UKB, with small coefficients, while STRADL shows no FDR-significant effects. The conclusion that white-matter relevance changes across adulthood cannot distinguish ageing from cohort differences.
↳ Results, Age moderation of network-g; Discussion, age paragraph and final summary
The study’s scale, multilevel mapping, latent-g modelling, and sensitivity analyses support a strong contribution score. Methodological Rigour remains in the competent band because selection is not examined, only three heterogeneous cohorts enter the random-effects meta-analysis, and the ageing analysis is cross-sectional. The most consequential issue is interpretive: moderate edge-level agreement and a within-UKB hold-out do not support an unqualified claim of replication across cohorts. Likewise, unnormalised edge-category counts and absent tract-length analysis do not establish preferential dependence on long-range connections.
Nabu’s assessment, alongside the field’s view.
Are you an author of this paper?
Sound3.6
Confidence mediumThe 38,824-participant analysis extends prior tract-focused work through global, nodal, and 818-edge mapping across three network weightings and three cohorts. Its main advance is a comprehensive descriptive map and shared reference network rather than causal identification.
We conducted the most comprehensive analysis to-date of the relationship between cognitive functioning and structural brain connectivity.
Latent-g SEMs, FDR correction, random-effects meta-analysis, hold-out prediction, and brain-volume and neurological sensitivity analyses provide competent execution. Selection is not examined, cross-cohort meta-analysis uses only three heterogeneous cohorts, and age moderation remains cross-sectional.
A three-way meta-analysis was then performed using a random effects model.
The global-to-node-to-edge progression is easy to follow and effect directions are transparently reported. The Abstract and final Discussion overstate replication, ageing relevance, and preferential long-range connectivity relative to the underlying analyses.
Edge-level findings replicated across cohorts and predicted g in a hold-out sample.
The paper engages P-FIT, tract-based dMRI, connectome prediction, cortical morphometry, scanner heterogeneity, cognitive measurement, and SC–brain-volume dependence. It does not clearly carry the cross-sectional nature of age moderation or healthy-volunteer selection into its concluding interpretation.
Age reshapes which aspects of neurobiology matter for cognitive functioning.
Caveats4 of 4 checks
The reported association patterns are broadly coherent, but several prominent interpretations are stronger than the analyses establish. Replication, age moderation, long-range topology, and SC–brain-volume interpretations require qualification.
Informed consent and named ethics approvals are declared for UKB, STRADL, and LBC1936. No text-grounded research-conduct concern was identified; supplementary resources and unavailable code or registration information remain assessability gaps only.
Flags: 2 declared / 5 total
91 references in manuscript 88 of 91 checkable references found in an index
No retraction notice found in Retraction Watch.
Sources: Retraction Watch ✓
Where this paper’s evidence sits on the path from initial observation to real-world use.
The evidence remains an observational association map with hold-out correlations of approximately 0.13–0.26 rather than a validated clinical or educational tool. No intervention, threshold, workflow, or operational decision rule is evaluated.
Predictions remained stable between the UKB main and hold-out samples.
AI-generated, human-governed. Something look off? Contact us to request a review.