Through adenosine antagonism, caffeine enhances neural activity by blocking the inhibitory affects of adenosine activation . In addition, by inhibiting adenosine binding to receptors on smooth muscles cells, caffeine reduces the ability of blood vessels to dilate and causes an overall reduction in baseline cerebral blood flow . All of these factors can lead to BOLD signal changes. Previous work by our group assessing caffeine’s effect on resting-state BOLD fluctuations has shown that it reduces both the stationary correlation and amplitude of the fluctuations in the motor cortex . While it is difficult to determine the underlying physiological mechanisms behind this effect, recent studies suggests that it may stem primarily from changes in neural activity coherence. For example, preliminary work by our group with magnet oencephalography found similar reductions in the correlation of MEG power fluctuations in the motor cortex, which do not have the same vascular confounds that are present in the BOLD fMRI signal . In addition, caffeine has been shown to impair motor learning compared to a placebo . Since it has been shown that the strength of resting-state functional connectivity is related to memory performance , these findings suggest that the caffeine-induced reduction in BOLD correlation may represent underlying neural changes. In this study, we employed a non-stationary analysis approach to gain further insight into the mechanisms of caffeine’s effect on functional connectivity. Specifically, we used a sliding window correlation analysis to assess whether caffeine consistently weakens the correlation over time or if transient periods of strong correlation still exist, vertical grow system albeit less frequently. A consistent decrease in correlation could be caused by an overall change in the vascular system induced by caffeine.
However, it is unlikely that a shift in the state of the vascular system would give rise to an increase in the non-stationarity of the correlations, when viewed on a time scale of tens of seconds. Instead, a caffeine induced increase in the temporal variability of the correlations would tend to support the existence of greater temporal variability in the coherence of the underlying neural fluctuations. The data used in this study were collected for a previous experiment examining the effects of caffeine on resting-state BOLD connectivity as assessed with stationary correlation measures . Participants were instructed to refrain from ingesting caffeine for at least 12 hours prior to being scanned. The estimated daily caffeine usage for each subject based on self-reports of coffee, tea, and soda consumption is presented in Table 3.1. The assumed caffeine contents for an 8-oz cup of coffee, an 8-oz cup of tea, and a 12-oz soda were 100 mg, 40 mg, and 20 mg respectively . Each subject participated in two imaging sessions: a caffeine session and a control session, in that order. The two imaging sessions were separated by at least 6 weeks.The caffeine session consisted of a pre-dose and a post-dose imaging section, each lasting around 45 minutes. Upon completion of the pre-dose section, participants ingested a 200 mg caffeine pill and then rested for approximately 30 minutes outside of the magnet before starting the post-dose section. The first resting-state scan of the post-dose section began approximately 45 minutes after the caffeine pill was ingested to achieve approximately 99% absorption of caffeine from the gastrointestinal tract . Control sessions used the same protocol, but without the administration of caffeine between sections, similar to the protocol used in . Subjects were not given a placebo during the control session. However, for convenience, we will still refer to the two scan sections as the “pre-dose” and “post-dose” sections.
Each scan section included a high-resolution anatomical scan, a bilateral finger tapping block design, and two five-minute resting-state BOLD scans. Bilateral finger tapping was self- paced and the block design run consisted of 20s rest followed by 5 cycles of 30s tapping and 30s resting. Subjects were instructed to tap while a flashing checkerboard was displayed and then to rest during the display of a control image, consisting of a white square situated in the middle of a gray background. During resting state scans, the control image was displayed for the entirety of the scan and subjects were asked to maintain attention on the white square.Images from each scan section were co-registered using AFNI software . In addition, the anatomical volume from each post-dose section was aligned to the anatomical volume of its respective pre-dose section, and the rotation and shift matrix used for this alignment was then applied to the post-dose functional images. The outer two slices of the functional data were discarded to minimize partial volume effects associated with the rotation of the post-dose data, and the first 10s of each functional run were not included. In addition, voxels from the edge of the brain were not included in the analysis in order to minimize the effects of motion.The second echo data from the finger tapping scans were used to generate BOLD activation maps of the motor cortex. This was accomplished using a general linear model approach for the analysis of ASL data . The stimulus-related regressor was produced by the convolution of the square wave stimulus pattern with a gamma density function . Constant and linear trends were included in the GLM as nuisance regressors. In addition, the data were pre-whitened using an autoregressive model of order 1 . The statistical maps were based on the square root of the F-statistic, which is equal to the t-statistic in the case of one nuisance term .
For consistency with our prior study , active voxels were defined using a method based on activation mapping as a percentage of local excitation . In summary, the √ F maps were separated into left and right hemispheric regions. The highest value in each region was identified and then every voxel was converted to a percentage of the peak statistical value for the region ×100. Active voxels were required to exceed an AMPLE value of 45% and a √ F value of 2 . The final activation maps were defined from the intersection of voxels active in both pre-dose and post-dose scan sections. Regions of interest were then defined for the left and right motor cortices from these activation maps. Thus, the same ROIs were used in the comparison of pre-dose and post-dose functional connectivity within an imaging session. Nuisance terms were removed from the BOLD resting-state data through linear regression. Nuisance regressors included constant and linear trends, 6 motion parameters obtained during image co-registration, physiological noise contributions , low frequency variations in cardiac and respiratory rate , and a version of the global signal that we will call the “regional” signal term. The regional signal term was calculated as the mean signal from the anterior portion of the brain in order to minimize bias that can occur when all the voxels in the brain are used to define a global mean signal as a nuisance regressor . Data were then temporally low-pass filtered using a finite impulse response function with a cutoff frequency of 0.08 Hz. This cutoff frequency was chosen for consistency with previous functional connectivity studies . To quantify functional connectivity strength, we extracted average BOLD signals from the left and right motor cortices. A stationary measure of inter-hemispheric functional connectivity was calculated as the correlation coefficient between the right and left motor BOLD signals computed over the entire length of each resting run. To assess temporal variations in inter-hemispheric motor cortex connectivity,vertical grow rack system we applied a sliding window over the length of each resting run and calculated the correlation between the left and right motor BOLD signals within each window. Correlation variability was quantified as the standard deviation of the sliding window correlation time series. To assess the affect of window length on correlation variability, we varied the window length from 10 seconds to 100 seconds. We found that significant caffeine induced increases in variability occurred for window lengths of 31 seconds and less. For all subjects, metrics were averaged across the two resting runs in each scan section. Two-tailed paired t-tests were performed between the pre-dose and post-dose results to assess caffeine-induced changes. Stationary measures of functional connectivity are shown for each subject before and after caffeine ingestion in Figure 3.1a, where the solid line represents equality between the two states. Consistent with our previous study , we find that caffeine significantly reduces inter-hemispheric BOLD connectivity in the motor cortex = 3.2, p = 0.012. Figure 3.1b shows the pre-dose and post- dose functional connectivity measures obtained in the control session for each subject.
There were no significant changes in these metrics = -1.2, p = 0.25. Windowed BOLD signal correlations between the left and right motor cortices are shown as a function of time for three representative subjects in Figure 3.2. While correlation varies with time in both the pre-dose and post-dose scan sections, temporal variability generally appears larger in the caffeinated state. However, extended time periods of strong correlation still exist in the post-dose measures. The scatter plots in Figure 3.1c and 3.1d show correlation variability using sliding window lengths of 30s and 20s for each subject during the caffeine and control sessions, respectively. Caffeine ingestion significantly increased variability | ≥ 2.5, p ≤ 0.04, while the control session data do not display significant changes in variability between scan sections for either window length | ≤ 0.55, p ≥ 0.6. To show that the caffeine-induced increase in correlation variability is not an artifact of using a specific window length, caffeine-induced changes in correlation variability are shown for different window lengths in Figure 3.3, which plots paired t-statistics between the post- and pre-dose scan sections versus window length. T-statistics are shown for both the caffeine session and control session . In this case a positive t-statistic indicates that variability is larger in the post-dose state, and data points above the top dashed line represent significant post-dose increases. Significant caffeine-induced increases in correlation variability are present in the caffeine session for window lengths of 31 seconds and shorter. Longer windows tend to smooth out correlation variations, making it more difficult to detect the effects of caffeine on correlation variability. In contrast, the control data do not show significant | < 0.39, p > 0.7 changes in correlation variability between the pre- and post dose scan sections for any window length. If the regional signal term is not included as a nuisance regressor, we find that correlation variability for the caffeine session data is significantly greater in the caffeinated state for window lengths of 27 seconds and shorter. BOLD time courses from the left and right motor cortices are shown before and after caffeine ingestion for a representative subject in the top panel of Figure 3.4. To visualize temporal variability in cross magnitude and phase, we created time-frequency plots of the windowed cross power spectra, which are shown below the BOLD time courses in Figure 3.4. These were created by computing the cross power spectrum between the left and right motor cortex BOLD signals for each 30- second sliding window period and displaying the resulting spectrum as a column in the time-frequency plot. For visualization purposes, we “increased” frequency resolution by zero-padding to 4 times the window length. In the plot, the color scale represents magnitude in units of normalized log-spectrum log2 , where σx and σy are the standard deviations computed over the entire length of the two BOLD time courses.Arrows are used to represent phase, with each arrow pointing to a position along the unit circle given by its phase angle. A 90◦ phase would have an arrow pointing up, a 180◦ phase would have an arrow pointing to the left, and so on. The frequency axis is restricted to frequencies less than the 0.08 Hz cut-off frequency that was used in the processing of the data. The plots in the bottom row of Figure 3.4 show sliding window time courses for correlation , cosine of the average phase cos , and average cross power magnitude M0XY . The power spectra in Figure 3.4 show that periods of low joint BOLD signal power correspond in time with reductions in correlation, shown in the plots below. This relationship is also captured by the M0XY time courses , which appear to track the correlation time courses fairly well, particularly in the pre-dose state. In addition, non-zero phase differences between the two signals also correspond with decreases in the correlation time series shown in the plots below, especially in the post-dose state.