
Psychology
The default network of the human brain is associated with perceived social isolation
R. N. Spreng, E. Dimas, et al.
This groundbreaking study explores how loneliness shapes our brains by leveraging insights from the impressive UK Biobank cohort of around 40,000 participants. Conducted by researchers including R. Nathan Spreng and Alain Dagher, it uncovers how lonely individuals adapt through enhanced neural connectivity, potentially compensating for social absence. Discover the fascinating interplay between isolation and brain function!
~3 min • Beginner • English
Introduction
Human evolution has favored enhanced cooperation, making social interactions crucial for survival and well-being. Loneliness—defined as the subjective perception of social isolation or mismatch between desired and perceived social connection—affects 10–20% of adults and is linked to broad health risks including morbidity, hypertension, immune dysfunction, suicide risk, poorer mental health, cognitive decline, and elevated dementia risk. While prior human and animal studies implicate subcortical reward systems, visual, attentional, limbic, and prefrontal regions in aspects of loneliness and social processing, higher-order social cognitive regions (medial prefrontal, medial temporal, temporoparietal junction, posteromedial cortex) are underexplored. The authors hypothesize that trait loneliness (a chronic, enduring sense of unmet social need) would be associated with distinctive neural signatures, especially in higher association cortices that evolved for sociality. They conduct a large-scale, multimodal assessment in the UK Biobank (~40,000 participants) to identify structural, functional, and white-matter features linked to trait loneliness, and to examine potential sex-specific associations.
Literature Review
The paper reviews evidence that loneliness is prevalent (10–20%) and carries health burdens comparable to or exceeding obesity or smoking. Prior neurobiological research has emphasized subcortical reward systems with dampened mesolimbic responses to social cues in loneliness, and behavioral findings of heightened vigilance to negative social information with reduced cognitive control. Neuroimaging studies have reported alterations in visual cortices, attention networks, limbic structures, and prefrontal cortex. However, higher-order social brain regions overlapping the default network (medial prefrontal, medial temporal, temporoparietal junction, posteromedial parietal cortex) have been relatively neglected despite their roles in mentalizing and social cognition. This gap motivates a comprehensive, multimodal investigation to test whether loneliness relates predominantly to these higher-associative systems.
Methodology
Study design: Multimodal population neuroscience analysis using UK Biobank imaging-genetics data release (Feb/Mar 2020). Participants: 38,701 individuals (47.5% men; 52.5% women), aged 40–69 at recruitment (mean 54.9, SD 7.5). Trait loneliness was assessed via single-item question: “Do you often feel lonely?” (yes=1/no=0). Preliminary genetic validation employed LD score regression to estimate genetic correlations between loneliness and 774 phenotypes using HapMap3 SNPs via LDHUB, with Bonferroni correction.
Imaging modalities and preprocessing: MRI acquired on 3T Siemens Skyra across dedicated sites with standardized protocols. Data underwent uniform preprocessing and QC from FMRIB pipelines.
- Structural MRI (sMRI): T1-weighted MPRAGE (1 mm isotropic). Preprocessing: GDC, brain extraction (BET), linear/nonlinear registration to MNI152 (FLIRT/FNIRT), tissue segmentation (FAST), head-size normalization (SIENAX). Gray matter volumes summarized into 100 cortical parcels using the Schaefer-Yeo atlas (7-network, 100-parcel).
- Resting-state fMRI (rs-fMRI): 2.4 mm resolution, TR=0.735 s, multiband=8. Preprocessing: EPI/GDC unwarping, motion correction (MCFLIRT), intensity normalization, high-pass filtering (sigma=50 s), ICA+FIX artifact removal; transformations merged to reduce interpolation. Regional time series extracted and Pearson correlation computed to obtain 100x100 functional connectivity matrices per participant.
- Diffusion MRI (dMRI): AP encoding; b=1000 and 2000 s/mm²; 2 mm resolution; 50 directions. Preprocessing: eddy-current/motion correction, outlier slice removal, GDC; DTI and TBSS used to compute FA maps and align to a mean FA skeleton. FA summarized within 48 white-matter tracts from the Johns Hopkins University atlas.
Confound control: For all modalities, nuisance variables were regressed out from imaging measures: body mass index, head size, head motion during task and resting-state scans, head position and receiver coil (x,y,z), scanner table position, and imaging site.
Analytical approaches:
- Structural MRI: Bayesian hierarchical models relating parcel-wise GM volumes (z-scored) to loneliness, leveraging the atlas hierarchy (regions nested within 7 canonical networks). Two model specifications estimated (i) region-level effects and (ii) network-level variance components (sigma) quantifying overall network relevance. Priors centered at zero; covariance informed by LKJ prior; parameters estimated via NUTS MCMC (PyMC3 v3.8), 4000 tuning, 1000 samples, convergence assessed via Rhat<1.02. Sex-stratified models examined sex differences.
- Functional MRI: Partial least squares (PLS; sklearn v0.21.3) applied to the lower triangle of standardized connectivity matrices (X) with loneliness labels (+1/-1) as y to identify the dominant mode of covariance linking connectivity patterns to loneliness. Statistical significance assessed via 1000-label permutation test (p<0.05, one-sided, FDR context; only leading mode tested). Sex-stratified analyses repeated.
- Diffusion MRI: Associations between FA in each of 48 JHU tracts and loneliness assessed with Pearson’s rho and Spearman’s rho (to probe non-linear trends). Uncertainty estimated via 100 bootstrap resamples to obtain 5–95% intervals; multiple comparisons controlled using Bonferroni. Sex-stratified analyses performed.
Sensitivity analyses: Findings assessed after additionally adjusting for clinical diagnoses (ICD F32*/F33* depression; F40*/F41 anxiety), trait neuroticism, alcohol consumption, education, higher-order age effects, adiposity, and non-white British ancestry. Cross-validated out-of-sample prediction/effect sizes evaluated within each modality.
Key Findings
- Genetic correlations: Of 774 phenotypes, 264 showed significant genetic correlation with loneliness after Bonferroni correction (p<0.05). Moderate overlap observed with BMI (Rg=0.26), education (Rg=−0.31), depressive disorder (Rg=0.61), anxiety (Rg=0.59), and alcohol intake (Rg=0.37).
- Structural MRI—network level: The default network exhibited the largest variance component linking gray matter volume to loneliness (posterior sigma=0.07; 5–95% HPD=0.04/0.10), followed by limbic (0.06; 0.01/0.14), dorsal attention (0.05; 0.01/0.09), somatomotor (0.04; 0.01/0.08), visual (0.04; 0.01/0.07), frontoparietal control (0.03; 0.01/0.06), and salience (0.02; 0.01/0.05). The default network had the tightest posterior (most informative).
- Structural MRI—region level: Positive associations (higher volume in lonely individuals) in left pSTS (mean=0.14; HPD=0.01/0.27), right pSTS (0.27; 0.10/0.44), left TPJ (0.17; 0.05/0.28), left fusiform gyrus (0.13; 0.01/0.26), right inferior temporal gyrus (0.31; 0.16/0.46), right posterior parietal lobe (0.15; 0.03/0.26), right dACC (0.14; 0.01/0.28). Negative associations in left dACC (−0.14; −0.27/−0.01), left DLPFC (−0.12; −0.24/−0.01), right DLPFC (−0.12; −0.22/−0.02), right central operculum (−0.16; −0.29/−0.05), right inferior parietal lobule (−0.27; −0.42/−0.14), left retrosplenial cortex (−0.19; −0.30/−0.08), inferior visual cortex left (−0.18; −0.33/−0.01) and right (−0.19; −0.34/−0.03).
- Structural—sex differences: Default network relevance was consistent in both sexes (men sigma=0.08; HPD=0.05/0.13; women sigma=0.08; HPD=0.03/0.13). Somatomotor network showed stronger associations in men (sigma=0.10; 0.02/0.16) vs women (0.03; 0.01/0.06). Right dorsomedial prefrontal cortex showed a negative association in men (mean=−0.10; −0.22/−0.01) vs a small, non-reliable positive trend in women (0.04; −0.04/0.13).
- Functional connectivity (PLS): The dominant mode (p<0.05 permutation) showed increased within-network coupling of the default network in lonely individuals. Between-network coupling was up-regulated between default and limbic, dorsal attention, and somatomotor networks (which are more anti-correlated in non-lonely). Visual network showed greater within-network coupling and decreased coupling with several other systems. These connectivity fingerprints were more strongly expressed in men.
- Diffusion MRI: Among 48 tracts, the top associations were fornix-related tracts: fornix (rho=0.06), fornix cres left (rho=0.05), fornix cres right (rho=0.05); all p<0.001 after Bonferroni correction; Spearman analyses confirmed. Sex stratification: stronger effects in men (fornix 0.05; fornix cres left 0.05; fornix cres right 0.03) than women (0.03; 0.02; 0.02).
- Robustness: All findings adjusted for imaging confounds; remained after accounting for depression/anxiety diagnoses, neuroticism, alcohol use, education, higher-order age, adiposity, and ancestry. Associations generalized in cross-validated out-of-sample tests across modalities. Cross-modality analyses showed convergent yet modality-specific features.
Discussion
The study demonstrates that trait loneliness has a coherent neural signature centered on the default network, across gray matter morphology, intrinsic functional connectivity, and white-matter microstructure. While prior accounts emphasized externally oriented processing (visual, attentional, limbic, and control circuits) consistent with vigilance to negative social cues and reduced control, this work reveals that higher-association regions overlapping the social brain—the default network—exhibit larger cortical volumes, increased intra-network functional coupling, and stronger fornix microstructural integrity in lonely individuals. These findings align with the idea that loneliness biases cognition toward internally directed processes: self-referential thought, mentalizing, episodic reminiscence, future thinking, and simulation of social scenarios. The fornix, linking hippocampal memory systems with medial prefrontal/default regions, may support this enhanced reliance on memory-based simulation observed in loneliness. Sex differences emerged across modalities, with men showing stronger default network coupling changes and fornix integrity associations, suggesting potential sex-specific neural expressions of loneliness and possible reporting differences. In the mid- to late-adulthood cohort, increased default network integrity may reflect chronic loneliness and a maladaptive shift toward internally focused network architecture, contrasting with some findings in young adults. Overall, the findings address the research question by identifying a multimodal, default-network-centered neural profile of trait loneliness, highlighting its relevance for social cognition and aging.
Conclusion
This multimodal population study identifies the default network as the core of the neural signature of perceived social isolation: lonely individuals show (i) greater gray matter deviations, (ii) increased intra-network functional coupling and altered inter-network balance, and (iii) higher fornix microstructural integrity linking hippocampal memory systems to default network cortex. These converging findings suggest heightened reliance on internally generated social cognition (mentalizing, reminiscence, future simulation) when external social needs are unmet. The work extends prior models focused on sensory/attentional systems and reveals sex-differentiated patterns more pronounced in men. Future research should use longitudinal designs to track causal pathways and chronicity, examine developmental and age-related differences (including young vs. older adults), clarify mechanisms underlying sex differences and reporting biases, and test whether interventions that enhance social connectedness or target default network dynamics can modify these neural signatures.
Limitations
- Cross-sectional design limits causal inference; most participants had a single imaging session, precluding direct assessment of chronicity and temporal changes.
- Loneliness measured via a single-item binary question; although validated and widely used, it lacks granularity relative to multi-item scales.
- Effect sizes, while robust at population level, are modest; individual-level predictions may be limited.
- Sample is mid- to late-adulthood; findings may not generalize to younger populations without further study.
- Potential residual confounding remains despite extensive adjustment (e.g., unmeasured psychosocial factors, reporting biases); stigma may influence sex differences in self-reported loneliness.
- Imaging limited to specific acquisition parameters and preprocessing pipelines; generalizability to other protocols requires replication.
Related Publications
Explore these studies to deepen your understanding of the subject.