Associations Between Age, Heart Rate Variability, and BOLD fMRI Signal Variability

Numerous studies report that BOLD fMRI signal variance (SDBOLD) decreases with age. However, these associations may partly reflect cardiovascular contributions to the BOLD signal. For example, heart rate variability (HRV) has been positively associated with Resting State Fluctuation Amplitude (RSFA), which captures low frequency components of BOLD fMRI variability. HRV is also negatively associated with age, which could potentially confound age-SDBOLD associations. Yet, limited research has examined HRV-SDBOLD associations or tested within-person HRV-SDBOLD coupling using sliding window analyses of simultaneous HRV and SDBOLD. We analyzed resting-state fMRI data from two independent Midlife in the United States (MIDUS) samples: Core at M3 (n=115) and Refresher at MR1 (n=101). Partial Least Squares (PLS) analyses revealed significant positive HRV-SDBOLD associations (Core: permutation p=0.018; Refresher: permutation p<0.001). Whole brain age-SDBOLD PLS associations were non-significant via permutation tests across several models (Core: permutation p=0.201; Refresher: permutation p=0.121). We found age-related decreases in SDBOLD across [~]70% of voxels in both samples. Concordance analyses showed 67-69% of brain voxels exhibited negative age-SDBOLD but positive HRV-SDBOLD relationships, suggesting that regions showing age-related decreases in SDBOLD also showed HRV-related increases in SDBOLD. Sliding-window analyses demonstrated robust positive within-person associations between person-centered HRV and SDBOLD via different HRV metrics: SDNN (Core: p < 0.001; Refresher: p < 0.001), RMSSD (Core: p = 0.072; Refresher: p = 0.009), and low frequency (Core: p < 0.001; Refresher: p < 0.001), with non-significant effects of high frequency (Core: p = 0.516; Refresher: p = 0.12) HRV. Thus, regardless of baseline levels, windows with higher HRV corresponded to higher SDBOLD, suggesting that cardiovascular factors partially explain age-SDBOLD associations and HRV may mechanistically influence SDBOLD. These results suggest that controlling for HRV, especially low-frequency HRV or SDNN, may be necessary when analyzing SDBOLD to isolate neural effects.