Brain functional connectivity initiates structured reorganization at a critical oxygen threshold during hypoxia

Abstract

The human brain dynamically adapts to hypoxia, a reduction in oxygen essential for metabolism. The brain's adaptive response to hypoxia, however, remains unclear. We investigated dynamic functional connectivity (FC) in healthy adults under acute hypoxia (FiO₂ = 7.7%, 11.8%) using BOLD fMRI, physiological monitoring (PetO₂, PetCO₂, SpO₂), and a Go/No-Go task. Principal component analysis identified a hypoxia-responsive FC component involving 400 cerebral parcels. This component emerged with a critical drop in PetO₂ (~53 mmHg), preceding changes in SpO₂, BOLD signals, and behavior. These FC changes were network-specific and centered on the default mode network (DMN), which selectively synchronized with other high-level cognitive networks. In contrast, visual networks remained stable and segregated from the DMN. These results suggest that the brain proactively reorganizes its functional architecture in anticipation of oxygen decline, rather than in response to it. FC-based markers may offer early indicators of vulnerability in neurological or neurodegenerative conditions.

Keywords: arterial oxygen pressure; end-tidal oxygen; functional connectivity; hypoxia; proactive organization.

Fig. 1. Multi-domain effects of acute severe hypoxia on respiratory gas dynamics, oxygenation responses, cognition performance, and brain connectivity.

a, Experimental protocol with a 10-minute BOLD fMRI session including 3 minutes of normoxia, 3 minutes of hypoxia (FiO₂ = 7.7%), and 4 minutes of normoxia reoxygenation, performed concurrently with a Go/No-Go cognitive task. b, Time course marking the onset and duration of hypoxic stimulation. c, d, Time course of estimated partial pressures of end-tidal oxygen (PetO₂) and carbon dioxide (PetCO₂), respectively. e, Time course of peripheral oxygen saturation (SpO₂). f, Hypoxia-induced bulk BOLD signal normalized to the 1-minute average proceeding hypoxia onset. g, Behavioral change, expressed in the change in commission error rate (Δcommission error) over a 90-second causal sliding window. h, Temporal dynamics of the global functional connectivity within a 90-second causal sliding window. Sampling intervals: 2 seconds for the panels (a, f, g, h); 10 seconds for (c, d); 30 seconds for (e). Error bar (grey area) denotes standard deviation across subjects (N = 11).

Fig. 2. Hypoxia-responsive changes in functional connectivity are network-specific.

To examine hypoxia-induced temporal dynamics in brain functional connectivity, principal component analysis (PCA) was applied to dynamic connectivity profiles comprising 79,800 pairwise connections among 400 cortical parcels (Schaefer atlas) across subjects. a, Summary table presenting the explained and cumulative variance of the top five principal components (PCs), with PC1 accounting for the largest portion of the variance. b, Group-averaged time series of PC1. The red horizontal bar indicates the period of acute severe hypoxia. Shaded areas denote standard deviation across subjects. The y-axis represents signal intensity in arbitrary units (a.u.). c, Histogram of PC1 coefficients. For further analysis, functional connections (FCs) with the top 5% of largest coefficients were classified as hypoxia-responsive (HR) FCs, and the bottom 10% with the smallest coefficients were defined as stable FCs. d, 400 × 400-ROI FC matrices, visualizing the distribution of HR FCs (left) and stable FCs (right). e, Inter- and intra-network distributions of HR and stable FCs. Cell colors indicate the percentage of possible connections within and between the 17 canonical brain networks. Numbers shown in cells represent significance levels from Monte Carlo (MC) permutation tests (1: p<0.05, 2: p<0.01, 3: p<0.001), corrected for multiple comparisons using the Benjamini–Hochberg false discovery rate (FDR) method. Notably, the Default A network exhibited selective increases in inter-network connectivity, with minimal change in intra-network connectivity. Limbic A and B networks were excluded from this summary due to signal dropout in many of their parcels. Abbreviations of brain networks: Cont = Control; Default = Default Mode; DorsAttn = Dorsal Attention; Limbic = Limbic; SalVentAttn = Salience/Ventral Attention; SomMot = Somatomotor; TempPar = Temporoparietal; VisCent = Visual Central; VisPeri = Visual Peripheral.

Figure 3.. Hypoxia-induced changes in functional connectivity are triggered when PetO₂ crosses a critical threshold.

Normoxia and mild hypoxia experiments were included to validate the hypoxia-responsive (HR) change in FC. a, b, fMRI protocol under normoxia (FiO₂ = 21.0%) and the corresponding dynamics of hypoxia-responsive (HR) and stable FCs. c-e, fMRI protocol under severe hypoxia (FiO₂ = 7.7%) and the corresponding dynamics of HR and stable FCs and interpolated PetO₂ time course. f-h, fMRI protocol under mild hypoxia (FiO₂ = 11.8%) and the corresponding dynamics of HR and stable FCs and interpolated PetO₂ time course. i, j, Scatterplots of mean HR-FC strength versus PetO₂ during severe and mild hypoxia, respectively. Colored dots (blue to red) indicate five time points from 180s to 540s (90-second intervals), illustrating temporal progression alongside red arrows. In panel i, the blue and yellow dots mark the onset and offset of severe hypoxia, respectively. In panel j, all dots represent time points under hypoxia, as mild hypoxia was applied throughout. A vertical dashed line at 53 mmHg indicates the proposed threshold at which HR-FCs begin to emerge. k, Combined view of i and j, showing the consistent threshold effect under both hypoxia conditions, suggesting a common PetO₂ tipping point for triggering FC reorganization.