Hypercapnia Reversibly Suppresses Low-Frequency Fluctuations in the Human Motor Cortex during Rest Using Echo–Planar MRI

Human Subjects

Six healthy human volunteers (five men and one woman, aged 24–32 years) with no current use of medication were scanned. All protocols received Institutional Review Board approval. Written consent was obtained from all subjects. Subjects other than graduate students were paid on an hourly basis. Volunteers were positioned supine on the gantry of the scanner with the head in a midline location in the coil. To limit head movement, foam padding was placed between the forehead and coil. Two axial slices across the motor cortex (Rumeau et al., 1994) were obtained in each EPI scan.

Imaging parameters

Each scanning session began with the acquisition of high-resolution axial anatomic images using a gradient-recalled at steady state (GRASS) pulse sequence, with repetition time (TR) = 600 ms, echo time (TE) = 10 ms, field-of-view (FOV) = 24 × 24 cm, and matrix size = 256 × 256. For functional imaging, several time series of 280 images (280 s) were obtained from multiple axial slices with a gradient-echo EPI sequence, FOV = 24 × 24 cm, matrix size = 64 × 64, slice thickness = 10 mm (corresponding to a spatial resolution of 3.75 mm × 3.75 mm × 10 mm), TE = 40 ms, and TR = 1,000 ms with flip angle = 81°. An additional study was carried out in one of the subjects at TR = 250 ms to control for possible effects from aliasing of the heart rate to low frequencies.

Data acquisition

During resting-state acquisitions, subjects were instructed to refrain from any cognitive, sensory, or motor tasks as much as possible. To minimize the possibility of imagining finger tapping, subjects were falsely told that the experiment involved the auditory cortex and such an experiment was actually performed. For the auditory task-activation paradigm, subjects listened to four cycles of common English words. Each cycle consisted of 30 s of rest followed by 30 s of the auditory task. The paradigm concluded with an additional 40 s of rest. Although the task was auditory, a time course of axial images was, in fact, obtained along the motor cortex. Finally, although subjects had not been informed in advance, they were asked to perform a motor task (bilateral finger tapping). For bilateral finger tapping, subjects were instructed to touch each finger with the thumb sequentially in a self-paced manner without moving their arms. The finger-tapping rate varied between 2 and 0.5 tappings per second. The sequence of the finger-tapping paradigm was the same as for the auditory paradigm, four 60-s cycles followed by 40 s of rest.

Hypercapnia was induced by inhalation of 5% CO₂, 20% O₂, and 75% N₂. This breathing gas mixture was chosen since it has been demonstrated to cause substantial vasodilation without injury (Kwong et al., 1995; Rostrup et al., 1994). Inhalation of 5% CO₂ in artifically ventilated experimental animals increases the arterial CO₂ pressure to ∼70 mm Hg (Lee et al., 1995) and also abolishes low-frequency fluctuations (Hudetz et al., 1992). All subjects practiced breathing 5% CO₂ a few times outside the scanner prior to scanning. A pulse oximeter with infrared sensor on the finger of the subject (Model 504, Criticare Systems, Inc., Waukesha, WI, U.S.A.) was used to monitor the cardiac rate and blood oxygenation throughout the scanning session. The sampling rate of the MR scan at one image per second provided a frequency bandwidth large enough to estimate respiratory frequency without any external monitoring techniques (Biswal et al., 1994).

Each experiment consisted of the following scans: (1) rest while breathing room air, (2) rest during hypercapnia, (3) rest after resuming breathing of room air, (4) auditory task-activation paradigm, and (5) bilateral finger tapping paradigm. For the hypercapnia study, the scan was started 10 s after addition of 5% CO₂ to the breathing gas mixture. Each scan lasted 4 min 40 s and was followed by ∼5 min of rest, during which the subjects breathed room air.

Data analysis

Change of the breathing gas causes change in capillary hemodynamics that, in turn, cause change in MR signal intensity. The approach to physiological equilibrium can be monitored by observing the approach of MR signal intensity to equilibrium. The time dependence of the increase in the signal intensity in grey matter after onset of 5% CO₂ in the breathing gas mixture was fitted to an equation of the form

S ( C O 2 ) − S o S o = A ( 1 − e − t / τ )

(1)

where S_o is the signal intensity in grey matter at the initiation of CO₂ inhalation and S(CO₂) is the signal intensity after several minutes of CO₂ inhalation. The prefactor A is the fractional increase in signal intensity after equilibrium has been reached, and the time constant τ is a measure of the temporal approach to equilibrium. Equation 1 was introduced empirically as a convenient means to characterize the data. Nevertheless, it may give insight into the underlying physiology.

A few pixels in each data set were identified by inspection of the anatomic images as lying in grey matter, and average values of A and τ determined from the corresponding pixel time courses. These values were inserted into Eq. 1, and the resulting waveform was cross-correlated with every pixel time course. All time courses with correlation coefficients >0.7 (corresponding to a significance value p < 10^–7) were selected for further analysis. Visual inspection of the anatomic images confirmed that these selected pixels were, almost without exception, in grey matter. In addition, very few pixels apparently lying in grey matter, failed to pass the 0.7 threshold. The choice of 0.7 as a threshold was based on the close correspondence of selected pixels with anatomically defined grey matter. Analysis of physiological fluctuations was performed on data from these pixels after the signal had approached 90% equilibrium, corresponding to a time equal of about ∼2τ.

A Gaussian low-pass filter at 0.08 Hz (Hamming, 1983) was applied to all pixel time courses in the EPI data sets of resting brain (both room air and 5% CO₂ scans). Although this eliminates fluctuations at respiration and cardiac rates along with the first harmonic of the cardiac rate for data obtained with a TR of 250 ms, only the respiration rate is reliably eliminated for a TR of 1,000 ms. For this longer TR value, fluctuations at the heart rate would be aliased into lower-frequency components and appear as a low-frequency signal if the heart rate of the individual lay in the range of 57–63 cycles/min. Heart rates outside this range would be aliased into frequency components that would not pass the 0.08 Hz cutoff of the low-pass filter. For each individual, heart rate variability exists that further blurs the aliased frequencies and could, in principle, mimic low-frequency fluctuations. Aliasing problems were a concern in the study presented here, and we are confident that the data were not contaminated by aliasing.

The task-activation data acquired in scan 5 of the protocol was used to determine the borders of the sensorimotor cortex in each of the other four scans. Data from scan 5 were not otherwise used. In all pixel time courses, the first and last three data points from each 30-s period during transition from one state to the other were neglected. All pixel time courses in scan 5 were normalized and cross-correlated with the box-car waveform on a pixel-by-pixel basis (Bandettini et al., 1993). From a total of ∼1,000 pixels from the brain (of the possible 4,096 pixels), ∼60 pixels passed a correlation-coefficient threshold value of 0.35. After a Bonferroni correlation, a threshold of 0.35 corresponds to a statistical significance of p < 10^–3. This threshold was used in our earlier study (Biswal et al., 1995b). All pixels that passed this threshold were identified as lying in the sensorimotor and associated cortices. Pixels from the sensorimotor cortex were further categorized into left and right motor cortices along with the supplementary motor areas (Roland et al., 1980).

For each pixel in scan 1 (resting state) that had been determined from the task-activation data (scan 5 of the protocol) to be within the borders of the motor cortices, the corresponding pixel time course was used as a reference waveform and correlated with all other pixel time courses of the scan. A threshold of 0.35 was used, and ∼60 correlation-coefficient images (Bandettini et al., 1993) which are called functional-connectivity maps in the following, were produced. The procedure was repeated for scans 2 and 3, as well as scan 4 (mock scan).

Pixels from the task-activation data (scan 5) with correlation coefficients with respect to the box-car waveform >0.35 were classified into three regions: left motor cortex, L, right motor cortex, R, and other brain tissue, O. The number of pixels in each region was designated n_L, n_R, and n_O, respectively. There were (n_L + n_R) functional-connectivity maps produced for each of the three resting-state scans (scans 1, 2, or 3) and also for the mock scan (scan 4). Consider the subset of functional-connectivity maps produced from a given resting-state scan using pixel time courses only in the right motor cortex. There were n_R such maps. The quantity n̅_LR is defined as the average number of pixels in region L (left motor cortex) in those n_R maps. Similarly, n̅_RR is the average number in region R. Using the remaining n_L functional-connectivity maps, quantities n̅_RL and n̅_LL were determined. This procedure was repeated for each resting-state scan as well as for the mock scan. A measure of correlations within a region is given by n̅_RR/n_R, n̅_LL/n_L, and between regions by n̅_LR/n_L, which is necessarily identical to n̅_RL/n_R. The quantity (n̅_OL + n̅_OR)/n_O serves as a control. This is the average number of pixels in O (other brain tissues) that passes the threshold when using a motor cortex pixel time course as a reference. This methodology is discussed in more detail in Biswal et al. (1995b).