TL;DR
This study reveals that the global signal (GS) in resting-state fMRI, often treated as noise, actually contains crucial age-related neurobiological information. By analyzing GS topography across a wide age range (6-85 years), researchers found distinct patterns in how subcortical and cortical brain regions contribute to the GS, suggesting a "dual-layer" model of brain function that changes throughout the lifespan. This highlights the need to carefully consider the implications of removing the GS in aging research.
What Problem Does This Paper Address?
A major challenge in functional neuroimaging, particularly with resting-state fMRI (rs-fMRI), is distinguishing true neural signals from noise and artifacts. The "global signal" (GS), which is the average time series across all brain voxels, has been a subject of debate. While it can capture artifacts, emerging evidence suggests it also reflects important neural activity and is linked to individual differences in behavior and cognition. However, how the GS topography, or the spatial distribution of its contributions, changes across the human lifespan, especially in relation to aging, has not been systematically investigated. This gap is critical because changes in brain function are a hallmark of aging, and if the GS reflects these changes, its handling during data preprocessing could confound age-related findings.
Methods
The researchers employed a comprehensive approach using multiple preprocessing pipelines to ensure the robustness of their findings. They utilized data from 601 participants aged 6-85 years. Key methods included:
- Minimally Preprocessed (MP) Pipeline: Standard preprocessing steps using FSL, AFNI, and SPM.
- Covariate Regression (CR) Pipeline: Regression of nuisance variables like motion parameters, white matter, and CSF signals.
- ICA-FIX Denoising: An automated method using Independent Component Analysis (ICA) to identify and remove noise components from spatial maps.
- Temporal ICA (tICA) Denoising: A complementary approach that removes temporally structured noise.
The core analysis involved calculating the global signal (GS) for each participant and then performing linear and quadratic regression analyses to assess the association between GS topography and age. They generated voxel-wise beta maps, converted them to z-statistics, and used general linear models (GLMs) to examine age-related effects. To ensure head motion didn't drive the findings, they also conducted multivariate partial least squares (PLS) analyses and examined the impact of scrubbing (removing high-motion frames).
Datasets
The study utilized data from the Enhanced Nathan Kline Institute - Rockland Sample (NKI-Rockland Sample) publicly available repository. This dataset is a well-established resource for neuroimaging research, containing rs-fMRI and structural MRI data from a large, diverse sample. No new datasets were introduced.
Key Insights and Ideas
- GS as a Window into Lifespan Brain Development: The GS is not just noise; its spatial topography dynamically changes with age, reflecting fundamental shifts in brain organization.
- Dual-Layer Model of GS: The findings support a "dual-layer" model of GS composition, where subcortical regions (like the thalamus) contribute to a background "arousal/vigilance" layer, and cortical regions (like the frontoparietal network) contribute to a foreground "cognition/network integration" layer. These layers interact and change differently with age.
- Subcortical vs. Cortical Age Trajectories: Subcortical regions show distinct linear age associations (e.g., thalamus increasing, putamen decreasing with age), while cortical regions (particularly frontoparietal and dorsal attention networks) exhibit more complex quadratic patterns, being stronger in early and late life and weaker in middle age.
- Robustness Across Preprocessing: The age-related GS topography findings were remarkably consistent across different preprocessing pipelines (MP, CR, ICA-FIX, tICA), underscoring their reliability and suggesting that these effects are not artifacts of specific denoising strategies.
Key Results
- Thalamus: Shows a strong positive linear association with the GS across age, meaning its contribution to the GS increases with age.
- Putamen: Exhibits a negative linear association, with stronger contributions to the GS in younger individuals compared to older ones.
- Nucleus Basalis: Displays a U-shaped quadratic pattern, with weaker GS associations in middle age and stronger associations in young and old age.
- Lateral Frontoparietal and Dorsal Attention Networks: Show positive quadratic associations, with stronger GS coupling in early (<20 years) and late (>60 years) life stages, and weaker coupling in middle age.
- Preprocessing Pipeline Consistency: Temporal correlations between GS time-series across pipelines ranged from r = 0.73 to 0.96, and spatial map correlations were high (r = 0.96), indicating that while denoising affects the GS time-series, the age-related topography remains stable.
Limitations and Future Work
- Cross-Sectional Design: The study uses a cross-sectional design, meaning it compares different age groups at a single point in time. Longitudinal studies are needed to confirm these age-related trajectories.
- Age Range: While covering 6-85 years is extensive, future work could extend this range to include fetal and very late life stages for a more complete picture of development and aging.
- Mechanisms: The study identifies how GS topography changes with age but not the underlying neurobiological mechanisms driving these changes. Future research should explore how differing levels of brain activity might drive these relationships.
- GS Regression Implications: The authors emphasize the need for caution when using GS regression in aging studies, as it may have system-specific implications and could potentially mask or alter true age-related findings. Further research is needed to understand when GS regression is beneficial versus detrimental for identifying age-related cognitive processes.
- Artifact Handling: While tICA effectively removes structured noise, the authors acknowledge that unstructured noise might still influence findings. They also note that respiration and cardiac activity, which contribute to the GS, are intrinsically linked to neural function and may be important to retain in some contexts.
Summary updated