Mapping the end-tidal CO₂ response function in the resting-state BOLD fMRI signal: Spatial specificity, test–retest reliability and effect of fMRI sampling rate

Highlights

• We estimate the resting-state BOLD response to intrinsic PETCO2 fluctuations.

• We include CRV and RVT variations in the model to account for interaction between them and PETCO2.

• PETCO2 response has a peak around 7 seconds and its effect is more dominated in gray matter.

• fMRI sampling rate has minimal effect on the shape of the estimated responses.

• Test–retest reproducibility of the responses is higher for short-TR data.

Abstract

The blood oxygenation level dependent (BOLD) signal measures brain function indirectly through physiological processes and hence is susceptible to global physiological changes. Specifically, fluctuations in end-tidal CO₂ (PETCO₂), in addition to cardiac rate variation (CRV), and respiratory volume per time (RVT) variations, have been known to confound the resting-state fMRI (rs-fMRI) signal. Previous studies addressed the resting-state fMRI response function to CRV and RVT, but no attempt has been made to directly estimate the voxel-wise response function to PETCO₂. Moreover, the potential interactions among PETCO₂, CRV, and RVT necessitate their simultaneous inclusion in a multi-regression model to estimate the PETCO₂ response. In this study, we use such a model to estimate the voxel-wise PETCO₂ response functions directly from rs-fMRI data of nine healthy subjects. We also characterized the effect of sampling rate (TR = 2 seconds vs. 323 ms) on the temporal and spatial variability of the PETCO₂ response function in addition to that of CRV and RVT. In addition, we assess the test–retest reproducibility of the response functions to PETCO₂, CRV and RVT. We found that despite overlaps across their spatial patterns, PETCO₂ explains a unique portion of the rs-fMRI signal variance compared to RVT and CRV. We also found the shapes of the estimated responses are very similar between long- and short-TR data, although responses estimated from short-TR data have higher reproducibility.

Introduction

The involvement of physiological processes in the resting-state fMRI signal and in the resulting functional connectivity measures is an important question, and has yet to be fully understood. Thus far, the most substantial contributors to the resting-state fMRI signal relevant to functional connectivity measurements are thought to be respiratory volume per unit time (RVT) variability, cardiac rate variability (CRV) and end-tidal CO₂ (PETCO₂) fluctuations. Constituting a consistent source of physiological noise in the BOLD fMRI signal (Wise et al., 2004), PETCO₂ has been associated with respiration RVT variations (Birn et al., 2006a, Chang and Glover, 2009b); however the BOLD signal variations explained by the two variables are not identical (Chang and Glover, 2009b). Therefore, accounting for the RVT variation in the BOLD signal would not necessarily eliminate the PETCO₂ effect, making it essential to study the relationship between PETCO₂ and the BOLD signal. Previous studies have either relied on cross-correlation (Chang and Glover, 2009a, Chang and Glover, 2009b, Peng et al., 2013, Wise et al., 2004) or assumed a standard gamma shaped hemodynamic response (Birn et al., 2008b, Wise et al., 2004). Wise et al. (2004) were the first to estimate the PETCO₂ response function (RFco₂) in the context of MRI, but used transcranial Doppler ultrasound of the middle cerebral artery. Thus, the hemodynamic response of the resting-state BOLD fMRI (rs-fMRI) signal to PETCO₂ has not been estimated directly through deconvolution.

Arterial pressure of CO2 is generally reflected in end-tidal pressure of CO2 (PETCO2), assuming arterial blood gases and lung gases have equilibrated (Peebles et al., 2007). Changes in PETCO2, in turn, trigger changes in vascular tone via the arterial baroreflex (Ainslie et al., 2005a, Badra et al., 2001, Harris et al., 2006). However, even without external modifications of PETCO2, PETCO2 fluctuates over the natural breathing and cardiac cycle (Sin et al., 2012). A robust relationship between PETCO2 fluctuations and the rs-fMRI signal was first reported by Wise et al. (2004) and reproduced by Chang and Glover (2009b), with gray matter exhibiting more of this effect than white matter. However, as interactions with cardiac pulsation and respiratory volume were not fully accounted for, the unique modulatory effect of PETCO2 on the rs-fMRI signal remains unclear.

Birn et al. (2006a) first investigated RVT in the context of fMRI. Believed to be initiated by random dynamics in the autonomic nervous system (Bruce, 1997), this slow-wave (~ 0.03 Hz) breath-to-breath variation was correlated with the BOLD signal in a pattern similar to that of PETCO₂ (Chang and Glover, 2009b), and the maps of signal changes due to breath-holding (Birn et al., 2006a), suggesting (at least in part) a shared mechanism. Although PETCO₂ was found to be strongly correlated with RVT convolved with the respiratory response function (RRF) (Chang and Glover, 2009b), the action of RVT on the rs-fMRI signal is less straight forward, while the effect of PETCO2 on the BOLD signal may be at least partly distinct from those of RVT. Given the previous results, it is of interest to directly probe the PETCO₂ response function in rs-fMRI data.

Variability in cardiac pulsation has also been found to strongly modulate the rs-fMRI signal (Chang et al., 2009a, de Munck et al., 2008, Shmueli et al., 2007) not only near the cerebrospinal fluid and major blood vessels, but also throughout the gray matter. It is well established that respiration and cardiac pulsation are related through respiratory sinus arrhythmia, a naturally occurring variation in heart rate during a breathing cycle driven by the parasympathetic nervous system. Respiratory rate scales with heart rate (Pitzalis et al., 1997) as well as heart-rate variability (Princi et al., 2006), and inhalation is known to temporarily increase heart rate and reduce cardiac rate variability. Interestingly, the effects of RVT and cardiac rate variations on the rs-fMRI signal were found to mirror each other (Petridou et al., 2009b) and partially colocalize (Chang et al., 2009b). On the other hand, while PETCO2 varies with the cardiac cycle through sinus arrhythmia (Sin et al., 2012), the relationship between PETCO2 and cardiac rate variability remains unclear.

In light of these potential interactions, we believe that to avoid possible confounding effects among the various physiological variables, the response function to PETCO₂ should be estimated in a multi-regression model involving cardiac as well as respiration variations. In addition, previous studies have reported global average response functions for PETCO₂ (Wise et al., 2004), RVT (RRF) (Birn et al., 2008b, Falahpour et al., 2013) and cardiac variability (CRF) (Chang et al., 2009b, Falahpour et al., 2013), respectively. Although the shape of the BOLD PETCO₂ response function may be debated, the percent-BOLD signal variance explained by PETCO₂ has been mapped previously (Chang and Glover, 2009a, Chang and Glover, 2009b, Wise et al., 2004), as well as its response latency (Chang and Glover, 2009b). Nonetheless, the high inter-subject variability made it difficult to generalize the findings. It should be noted that previous experiments used a low sampling rate, and high-frequency physiological noise, particularly cardiac noise, is likely to be aliased into lower frequencies and difficult to remove. Although methods such as RETROICOR are typically applied, using recorded cardiac and respiratory phase information to remove these physiological effects, they are not likely to be removed completely. Thus, it is unclear how this residual aliased signal affects the estimation of response functions for the low-frequency physiological fluctuations. Given the limitations mentioned herein, an important avenue of investigation would be to examine spatial variability in the response functions derived using ultra-fast fMRI acquisitions, and to see to what extent this variability is common across different individuals.

The aim of this study are: (1) to derive and validate the PETCO₂ response function in rs-fMRI data; (2) to spatially characterize the variability in the response to PETCO₂ fluctuations in a group of healthy subjects, while accounting for cardiac and respiratory volume variations; and (3) to assess the comparability and reproducibility of PETCO₂, respiratory and cardiac response estimates in slowly sampled fMRI data in reference to accelerated simultaneous multislice (multiband) fMRI data. The contributions of this study are: 1) we assess the effect of PETCO₂ exclusively of the other physiological signals, by simultaneously including the three physiological signals in the model; 2) we estimate physiological response functions and their effects at a voxel-wise level, to account for regional differences in cerebrovascular and metabolic properties (Peng et al., 2013, Xu et al., 2011); 3) we use multi-slice accelerated BOLD fMRI acquisition with ultra-short TR to assess the effect of sampling rate on noise removal and response function estimation; and 4) we investigate the test–retest reliability of all response functions obtained at different sampling rates.