Arterial CO2 Fluctuations Modulate Neuronal Rhythmicity: Implications for MEG and fMRI Studies of Resting-State Networks

Abstract

A fast emerging technique for studying human resting state networks (RSNs) is based on spontaneous temporal fluctuations in neuronal oscillatory power, as measured by magnetoencephalography. However, it has been demonstrated recently that this power is sensitive to modulations in arterial CO2 concentration. Arterial CO2 can be modulated by natural fluctuations in breathing pattern, as might typically occur during the acquisition of an RSN experiment. Here, we demonstrate for the first time the fine-scale dependence of neuronal oscillatory power on arterial CO2 concentration, showing that reductions in alpha, beta, and gamma power are observed with even very mild levels of hypercapnia (increased arterial CO2). We use a graded hypercapnia paradigm and participant feedback to rule out a sensory cause, suggesting a predominantly physiological origin. Furthermore, we demonstrate that natural fluctuations in arterial CO2, without administration of inspired CO2, are of a sufficient level to influence neuronal oscillatory power significantly in the delta-, alpha-, beta-, and gamma-frequency bands. A more thorough understanding of the relationship between physiological factors and cortical rhythmicity is required. In light of these findings, existing results, paradigms, and analysis techniques for the study of resting-state brain data should be revisited.

Significance statement: In this study, we show for the first time that neuronal oscillatory power is intimately linked to arterial CO2 concentration down to the fine-scale modulations that occur during spontaneous breathing. We extend these results to demonstrate a correlation between neuronal oscillatory power and spontaneous arterial CO2 fluctuations in awake humans at rest. This work identifies a need for studies investigating resting-state networks in the human brain to measure and account for the impact of spontaneous changes in arterial CO2 on the neuronal signals of interest. Changes in breathing pattern that are time locked to task performance could also lead to confounding effects on neuronal oscillatory power when considering the electrophysiological response to functional stimulation.

Keywords: cortical oscillations; functional connectivity; hypercapnia; magnetoencephalography; physiological noise.

Figure 1.

Physiological measures in response to graded hypercapnia. Shown are example P_ETCO₂ (A) and P_ETO₂ time courses (B), with color areas showing hypercapnia periods. N.B. the order of the four hypercapnia levels (+2, +4. +6 and +8 mmHg above baseline) were randomized across subjects. Each color represents a different hypercapnia level, so the color order (left–right) differs in this plot compared with subsequent plots in which hypercapnia level are displayed increasing sequentially. Gray boxes at the bottom of A indicate periods of the breathlessness rating feedback task. C, Box plot of ratings feedback scores (red line, median across subjects; blue box, interquartile range; dashed whiskers, data range; red crosses, outliers). D, Breathing rate change compared with the normocapnia (0 mmHg) level (mean ± SD across subjects). E, Breathing depth percentage change compared with the normocapnia (0 mmHg) level (mean ± SD across subjects). Asterisks denote a significant change with respect to normocapnia (Bonferroni-corrected p < 0.05) based on a post hoc Wilcoxon signed-rank test for the discrete ratings feedback scores and a two-tailed paired t test for breathing rate and depth.

Figure 2.

The MEG oscillatory amplitude response to hypercapnia. A, Group average Hilbert amplitude response percentage change maps for alpha, beta, and low-gamma bands for the highest (+8 mmHg target) level of hypercapnia relative to normocapnia (p_corr < 0.01, 2-tailed t test across subjects). B, Whole-brain average percentage change in alpha, beta, and gamma amplitude (mean ± SEM across subjects) for each hypercapnia level (relative to the medical air breathing normocapnia periods). Asterisks denote a significant change from normocapnia (Bonferroni-corrected p < 0.05, post hoc 2-tailed t test). C, Line of best fit for each subject (1 line per subject) between whole-brain average Hilbert amplitude and P_ETCO₂ on an epoch-wise basis. Example scatter plots that these lines of best fit are based on are presented later in Figures 4 and 5.

Figure 3.

Top, Frequency spectra for each condition for virtual sensors positioned at the peak response coordinate for alpha (left), beta (middle), and low-gamma (right) bands. MNI coordinates for these virtual sensors are presented in the title of each plot. Bottom, The same spectra plotted as an amplitude percentage change from medical air (normocapnia) periods for each hypercapnia level.

Figure 4.

Spontaneous fluctuations in P_ETCO₂ during periods of medical air and corresponding beta-band amplitude demonstrated in a subject with a high correlation coefficient (A) and a subject with a median correlation coefficient (B). i, P_ETCO₂ time series with normocapnia periods shown in gray. ii, Whole-brain average beta amplitude time series. iii, Plot of beta amplitude against P_ETCO₂ for each time point across the whole time series with corresponding Pearson correlation results displayed above the plot and the line of best fit across the whole time series plotted in black. iv, Plot of beta amplitude against P_ETCO₂ for normocapnia time points only with corresponding Pearson correlation results.

Figure 5.

Spontaneous fluctuations in P_ETCO₂ during periods of medical air and corresponding alpha-band amplitude demonstrated in a subject with a high correlation coefficient (A) and the subject with a median correlation coefficient (B). i, P_ETCO₂ time series with normocapnia periods shown in gray. ii, Whole-brain average alpha amplitude time series. iii, Plot of alpha amplitude against P_ETCO₂ for each time point across the whole time series with corresponding Pearson correlation results displayed above the plot and the line of best fit across the whole time series plotted in black. iv, Plot of alpha amplitude against P_ETCO₂ for normocapnia time points only, with corresponding Pearson correlation results.

Figure 6.

A, Histogram showing voxelwise R² indicating the contribution of P_ETCO₂ fluctuations to the beta-amplitude signal. The gray histogram indicates all voxels across all subjects, whereas the lines indicate the histograms of voxels for each individual subject. B, Windowed version of A to highlight the length of the tail and the shape of each individual subject histogram.