The physiological component of the BOLD signal: Impact of age and heart rate variability biofeedback training

Fig. 1.. Schematic for the model to determine the physiological component of the BOLD signal. (A) After detrending and normalizing heart rate and respiratory variation, the signals are convolved with CRF and RRF basis functions. For every voxel, a general linear model is used to find beta weights for each of the cardiac and respiration regressors to minimize the error from the original BOLD signal. (B) Example of an original BOLD signal (normalized to percent signal change) and the corresponding physiological component.

Fig. 2.. Log-transformed metrics of HRV (RMSSD, LF Power, HF Power) and low-frequency peripheral physiology across the (A) NKI dataset and (B) HRV-ER dataset. Standard box and whisker plots are overlaid over the violin plots. *p < 0.05.

Fig. 3.. Percent variance of BOLD signal explained by low-frequency peripheral physiology across all voxels. Three separate models were used to determine the variance in the BOLD signal explained by (A) both heart rate and respiration, (B) only heart rate, and (C) only respiration. In each panel, age group averages are shown for young and old participants. Voxels in which percent variance explained in younger adults was statistically significantly greater than older adults (p < 0.05 TFCE-corrected) are outlined in black, and alpha-fading was used to highlight sub-threshold voxels. For visualization purposes in the NKI dataset, p values were transformed using the natural logarithm to improve the contrast between highly significant voxels (e.g. p < 0.01). The brain slices shown are at z = -16 mm, 4 mm, 24 mm, 44 mm, and 64 mm in standard MNI152 space.

Fig. 4.. (A) HR–BOLD and (B) RV–BOLD cross-correlations for different time lags in the NKI dataset. Negative lags indicate that BOLD leads the physiological measure, and vice versa for positive lags. Age group averages for Pearson r coefficients are plotted at each lag. Significant voxels by age group at p < 0.05 (TFCE-corrected) are also outlined in black at each lag, along with alpha fading to show sub-threshold voxels. Red voxels indicate that young adult r values are greater than old adults, and blue voxels indicate that old adult r values are more positive (or less negative) than young adults. For visualization purposes, p values are transformed using the natural logarithm to improve the contrast between highly significant voxels (e.g. p < 0.01). The brain slices shown are at z = -6 mm, 4 mm, 24 mm, 44 mm, and 64 mm in standard MNI152 space.

Fig. 5.. (A) HR–BOLD and (B) CO2–BOLD cross-correlations for different time lags in the HRV-ER dataset at baseline. Negative lags indicate that BOLD leads the physiological measure, and vice versa for positive lags. Age group averages for Pearson r coefficients are plotted at each lag. Significant voxels by age group at p < 0.05 (TFCE-corrected) are also outlined in black at each lag, along with alpha fading to show sub-threshold voxels. Red voxels indicate that young adult r values are greater than old adults, and blue voxels indicate that old adult r values are greater than young adults. The brain slices shown are at z = -16 mm, 4 mm, 24 mm, 44 mm, and 64 mm in standard MNI152 space.

Fig. 6.. Cross-correlations between the BOLD signal and (A) heart rate and (B) respiration (i.e., CO2 for HRV-ER, RV for NKI), averaged across three tissue types: gray matter, white matter, and ventricles. In the HRV-ER data, both BOLD signal and CO2/HR are upsampled to TR = 0.2 seconds before cross-correlation calculation, but no upsampling occurred in the NKI data. Group averages for old and young adults are plotted along with shading for standard error. Lags where the cross-correlations between older and younger adults are statistically significant (p < 0.05) after Bonferroni correction are plotted in dark gray, and lags where the cross-correlation passed a p < 0.05 uncorrected threshold are plotted in light gray.

Fig. 7.. Impact of HRV biofeedback on BOLD–physiological cross-correlations. Cross-correlations between the BOLD signal (averaged across gray matter, white matter, and ventricles) and (A) HR and (B) CO2, before and after the Osc+ condition, are plotted. Cross-correlations between the BOLD signal (averaged across gray matter, white matter, and ventricles) and (C) HR and (D) CO2, before and after the Osc− condition, are plotted. Group averages are plotted along with standard error shading at every time point. Both BOLD signal and CO2/HR are unsampled to TR = 0.2 seconds before cross-correlation calculation. Lags where the cross-correlations between pre- and post-intervention are statistically significant (p < 0.05) after Bonferroni correction are plotted in dark gray, and lags where the difference passed a p < 0.05 uncorrected threshold are shown in light gray.

Fig. 8.. HR–BOLD whole-brain cross-correlation plots for older adults before and after the Osc+ intervention. Group averages for Pearson r coefficients are plotted at each lag. Significant voxels by age group at p < 0.05 (TFCE-corrected) are also outlined in black at each lag, along with alpha fading to show sub-threshold voxels. Red voxels indicate that post Osc intervention r values are greater than pre Osc intervention r values, and vice versa for blue voxels. The brain slices shown are at z = -16 mm, 4 mm, 24 mm, 44 mm, and 64 mm in standard MNI152 space.