Neural correlates of time-varying functional connectivity in the rat

Abstract

Functional connectivity between brain regions, measured with resting state functional magnetic resonance imaging, holds great potential for understanding the basis of behavior and neuropsychiatric diseases. Recently it has become clear that correlations between the blood oxygenation level dependent (BOLD) signals from different areas vary over the course of a typical scan (6-10 min in length), though the changes are obscured by standard methods of analysis that assume the relationships are stationary. Unfortunately, because similar variability is observed in signals that share no temporal information, it is unclear which dynamic changes are related to underlying neural events. To examine this question, BOLD data were recorded simultaneously with local field potentials (LFP) from interhemispheric primary somatosensory cortex (SI) in anesthetized rats. LFP signals were converted into band-limited power (BLP) signals including delta, theta, alpha, beta and gamma. Correlation between signals from interhemispheric SI was performed in sliding windows to produce signals of correlation over time for BOLD and each BLP band. Both BOLD and BLP signals showed large changes in correlation over time and the changes in BOLD were significantly correlated to the changes in BLP. The strongest relationship was seen when using the theta, beta and gamma bands. Interestingly, while steady-state BOLD and BLP correlate with the global fMRI signal, dynamic BOLD becomes more like dynamic BLP after the global signal is regressed. As BOLD sliding window connectivity is partially reflecting underlying LFP changes, the present study suggests it may be a valuable method of studying dynamic changes in brain states.

Keywords: BLP; BOLD; Dynamic; EEG; FFT; FIR; FWER; Functional connectivity; Global signal; LFP; Neural basis; SGoF; SI; Sliding window; Time varying; band-limited power; blood oxygen level dependent; electroencephalography; fMRI; family-wise error rate; fast Fourier transform; finite impulse response; functional magnetic resonance imaging; local field potentials; primary somatosensory cortex; sequential goodness of fit.

Figure 1

Example of interhemispheric correlation, measured in a sliding window as a function of window start time (“sliding window series”) from one rat, one fMRI run. Values are calculated using a window length of 50s, between left and right SI, for BOLD (dashed line) and each BLP band (solid lines, colors shown in legend), and are plotted versus the time of each window’s start point. Correlation varies substantially over time for all signals, ranging from strongly positive to strongly negative for BOLD, delta, theta, and alpha. Beta and gamma band correlation is rarely negative in this example, yet still varies between zero and high positive values.

Figure 2

(A) Mean correlation (normalized z values) between SI BOLD sliding window series and BLP sliding window series for each frequency band, after global signal regression. Correlation between the SI BOLD sliding window series and the BLP time courses for signals from incorrectly matched runs are shown as a control. Error bars are one standard error. The highest correlation is observed in high beta and gamma bands. Theta, high beta and gamma bands exhibited significant correlation to the SI BOLD sliding window series after correction for multiple comparisons when compared to randomly matched SI BOLD and BLP pairs (t-test, p values: delta 0.272, theta 1.35×10⁻³, alpha 7.65×10⁻², low beta 1.45×10⁻², high beta 8.40×10⁻⁶, gamma 5.18×10⁻⁵). (B) The same calculations were performed for data without global signal regression. All values are slightly lower, but the same three bands remain significance (t test, p values: delta 0.743, theta 7.20×10⁻³, alpha 0.214, low beta 3.59×10⁻², high beta 1.35×10⁻³, gamma 1.81×10⁻³). Error bars are one standard error. This figure was calculated with a 50s long sliding window and with BLP lagged four seconds after SI BOLD. * Statistically significant at 5% passing multiple comparisons correction (Carvajal-Rodriguez et al., 2009).

Figure 3

Mean of normalized z values for interhemispheric SI correlation, over all windows and all runs, for BOLD and each BLP band, plotted versus window length. Error bars are one standard error in terms of inter-trial variance. For visibility purposes, only every 20^th error bar is shown, and they are staggered between plots. As they were calculated from two second long segments, theta and delta include information from up to 1.5s longer than the window length used for correlation. Positive correlation is present for all bands even at the shortest windows, and increases steadily as window length increases. This plot uses normalized z scores, for naïve correlations see Supplemental figure S1.

Figure 4

(A) Mean of normalized z values for correlation between BOLD sliding window series and each BLP sliding window series (shifted 4s prior), over all runs, plotted versus window length. Error bars are one standard error in terms of inter-trial variance. For visibility purposes, only every 20^th error bar is shown, and they are staggered between plots. As they were calculated from two second long segments, theta and delta may include information from up to 1.5s longer than the window length used for correlation. (B) Standard error of z values plotted in (A), these are the same values shown on the error. Note that correlation steadily increases until it reaches a plateau, and error appears to steadily increase with window length. This plot uses normalized z scores, for naïve correlations, see Supplemental figure S2.

Figure 5

Correlation between the sliding window series from SI BOLD where no global regression has been performed versus the “global signal,” i.e. the mean BOLD signal from the whole brain. Ordinate is the normalized z value calculated from the correlation, abscissa is the relative time point within the window (used to calculate the sliding window series) where the corresponding point from the global signal was taken; negative numbers indicate earlier points in the window, positive numbers indicate later points. The darker gray solid line is from correctly matched SI BOLD sliding window series with global signals, the lighter gray dashed line is from incorrect matching. The lighter colored borders around each line represent one standard error. The two lines are significantly different at alignments from 17.5s to 25s (0.0210 ≤ p ≤ 0.0125, threshold of p=0.05 corrected for multiple comparisons with SGoF), this is shown with a black bar. The negative correlation values suggest that increases in global signal decrease amplitude of sliding window correlation, and/or vice-versa. Note that, while theoretically possible (see section 2.12), no statistical significance was seen if global regression was performed on SI BOLD prior to the calculation of the sliding window series (see section 3.4 and Supplemental Figure S4).