Resting-State Functional Connectivity in Animal Models: Modulations by Exsanguination

Abstract

We studied the spatiotemporal characteristics of the resting state low frequency fluctuations in functional MRI (fMRI), blood oxygenation level dependent (BOLD) signal in isoflurane-anesthetized rats. fMRI-BOLD measurements at 9.4 Tesla were made during normal and exsanguinated condition previously known to alter cerebral blood flow (CBF) fluctuations in anesthetized rats. fMRI signal time series were low-pass filtered and studied by spectral analysis. During normal conditions, baseline mean arterial pressure (MAP) was 110 ± 10 mmHg and low-frequency fluctuations in BOLD signal were observed in the frequency range of 0. 01–0. 125 Hz. Following blood withdrawal (exsanguination), MAP decreased to 68 ± 7 mmHg, resulting in an increase in the amplitude of the low-frequency fluctuations in BOLD signal time series and an increase in power at several frequencies between 0.01 and 0.125 Hz. Spatially, the BOLD fluctuations were confined to the cortex and thalamus spanning both hemispheres with sparse presence in the caudate putamen and hippocampus during both normal and exsanguinated states. Spatial distribution of the low frequency fluctuations in BOLD signal, from cross correlation analysis, indicate substantial inter-hemispheric synchrony similar to that observed in the conscious human brain. The behavior of the resting state BOLD signal fluctuations similar to CBF fluctuations during exsanguination indicates a myogenic dependence. Also, a high inter-hemispheric synchrony combined with different phase characteristics of the low frequency BOLD fluctuations particularly in the hippocampus relative to the cortex emphasizes distinct functional networks.

1. Introduction

We first demonstrated a significant temporal correlation of resting-state low-frequency blood oxygenation level dependent (BOLD) signal fluctuations in functional MRI (fMRI) both within and across hemispheres in primary sensorimotor cortex during rest. We defined resting-state connectivity (RSC) as significant correlation signal between functionally related brain regions in the absence of any stimulus or task. This result has been validated by a number of groups using different models (1,2) and is a subject which has been recently reviewed by Fox and Raichle (3). This correlated signal arises from spontaneous low-frequency signal fluctuations (SLFs). These low-frequency spontaneous fluctuations have also been observed by several investigators using animal models and a variety of measurement techniques, including polarographic technique of tissue oxygenation (4, 5), laser Doppler flowmetry (LDF) (6), fluororeflectometry of NADH and cytochrome-aa3 (7).

Testing hypotheses of the role of SLFs has involved attempts to determine their physiological origins. Cooper et al. (8) hypothesized that these fluctuations represent cellular maintenance of an optimum balance between cerebral blood flow (CBF) and oxidative metabolic rate (CMR_O2). Testing this hypothesis has involved manipulation of cerebral metabolism with anesthesia. These studies have involved comparison of activity during waking and anesthetized states in animals using various techniques including LDF (6) and fMRI in humans (9, 10). Signal oscillations in the rodent brain vary with differing levels of halothane anesthesia, carbon dioxide level and nitric oxide synthase (NOS) blockade (11). These results suggest support for the biophysical-origin hypothesis that affects the cerebral vasculature, either directly or indirectly. The neural mechanisms of slow rhythmic fluctuations have not yet been clearly defined, though with recent evidence indicating an active role for glia in neurovascular coupling (12-15), the slow rhythmic fluctuations may have both neuronal and glial origins.

In spectrophotometric studies of the intramitochondrial redox state of enzyme cytochrome aa₃ (CYTox) and cerebral blood volume (CBV), continuous slow oscillations and inter-hemispheric synchrony has been observed between these variables (16). The relationship between CYTox and CBV oscillations seem to be independent of the physiological state as they have been observed during both awake state and sleep (6, 16, 17), anesthesia (5, 18, 19) and during cerebral ischemia (20, 21). Though there have been studies indicating the presence of metabolic oscillations in the absence of CBF oscillations – indicating that metabolic oscillations may be primary in origin – there is no concrete evidence to indicate so. On the contrary, flow oscillations can be linked to metabolic oscillations by the evidence indicating that NADH and/or cytochrome aa₃ oscillations lagged behind CBV oscillations (16, 17). These results indicate that spontaneous oscillations in the intra-mitochondrial redox state may, at least, in part be linked to rhythmic variations in CMR_O2.

It should be emphasized that the origin of the slow cerebral fluctuations of CBV and CYTox remains to be determined. It is unlikely to be entirely vascular (“vasomotion”), in view of the complex frequency/time and interhemispheric architecture of these fluctuations in cats and rabbits (16, 17). There are a variety of neuronal, glial, and vascular phenomena that may offer their contributions to what finally appears as a measurable “fluctuation” (22). For instance, glutamate-induced intracellular calcium waves within the glial syncytium may represent an energy-dependent indirect reflection of activity within focal neuronal fields. Such factors would need to be carefully dissected by future efforts. An interesting example of such multifactorial components of the slow fluctuations emerges from the study of CBV and CYTox during the transition from slow-wave sleep to REM sleep in the cat, as will be discussed in the following sections.

We first demonstrated a significant temporal correlation of SLFs, both within and across hemispheres, in primary sensorimotor cortex during rest (9, 10). Nearly 74% of the time series from these voxels correlated significantly (after filtering the fundamental and harmonics of respiration and heart rates) while only a few voxel time courses (< 3%) correlated with those in regions outside of motor cortex. Subsequently, Hampson et al. (23) demonstrated the presence of RSC in sensory cortices, specifically auditory and visual cortex. In their studies, signal from visual cortex voxels during rest (first scan) was used as a reference and correlated with every other voxel in the brain. Significant number of voxels from the visual cortex passed a threshold of 0.35, while only a few voxels from outside the visual cortex passed the threshold. They have demonstrated similar results in the auditory cortex (24).

Lowe et al. (25) extended Biswal, Hyde and colleague’s (26) results by showing such correlations over larger regions of sensorimotor cortex (i.e., across multiple slices). Xiong, Fox and colleagues (27) established relationships between motor and association cortex. Similar to earlier results, they observed RSC between sensorimotor cortex areas (primary, premotor, secondary somatosensory). Further, however, they observed RSC relationships between these motor areas and association areas, specifically anterior and posterior cingulate cortex, regions known to be involved in attention. Greicius et al. (28) observed RSC in anterior and posterior cingulate areas. Subsequent observation of activation during a visual attention task indicated similar cingulate activity.

These studies have established the foundation for “resting-state functional connectivity studies” using fMRI (e.g., (23, 25, 26, 28, 29)). Results from these studies form the basis for speculation regarding the functional role of RSC. Bressler et al. (30) have suggested that such correlated signal fluctuations may be a phenomenon representing the functional connection of cortical areas analogous to the phenomenon of “effective connectivity” defined by Friston et al. (31). Thus, a family of cognitive-origin hypotheses (in contrast to biophysical-origin hypotheses) has emerged. Gusnard and Raichle (29), for instance, have suggested that such coherence indicates the presence of a “default mode of brain function” in which a default network continuously monitors external (e.g., visual stimuli) and internal (e.g., body-functions, emotions) stimuli. Other cognitive-origin hypotheses suggest that low-frequency fluctuation is related to ongoing problem-solving and planning (28). Biswal, Hyde and colleagues (26) and Xiong, Fox and colleagues (27) observed that analyses of resting-state physiological fluctuations reveal many more functional connections than those revealed by task-induced activation analysis. They hypothesized that task-induced activation maps underestimate the size and number of functionally connected areas and that functional networks are more fully revealed by RSC analysis.

Here, we discuss various animal studies that affect the CBF oscillations and hence the low frequency fluctuations in BOLD signal. The spatiotemporal characteristics of the low frequency fluctuations in brain oxygenation was mapped non-invasively using BOLD weighted fMRI imaging using physiological perturbations known to alter CBF oscillations. The resting state connectivity maps obtained from normocapnia and changes in mean arterial pressure (MAP) and differences between them are discussed. Low-frequency BOLD signal fluctuations were studied both in the time-domain and frequency-domain. These animal results were also consistent with earlier studies done in humans using fMRI. Functional maps obtained using the power spectrum of the frequencies and cross correlation analysis indicate a significant reduction in the resting state low-frequency BOLD physiological fluctuations in the cortical, sub-cortical and deeper brain structures during normal physiological states. Exsanguination led to an enhancement in the amplitude of the low frequency BOLD fluctuations and spatially expanded to most of the cortical, sub-cortical and deeper brain structures. The behavior of the resting state BOLD signal fluctuations similar to CBF fluctuations, during exsanguination, indicates a myogenic dependence. However, a high intra-hemispheric symmetry in the BOLD fluctuations with similar phase characteristics in other regions in this study except the hippocampus suggests that the fMRI-BOLD signal fluctuations in anesthetized rats may carry information from underlying neural activity.

2. Methods

2.1. Surgical Preparation

Seven male Sprague-Dawley rats (250–300 g; Harlan, Indianapolis, IN) were anesthetized with 1.2% isoflurane in oxygen. Body temperature was monitored with a rectal probe and maintained at 37. 0 ± 0. 5 using a homeothermic feedback heating system (Baxter K-MOD100, Gaymar Industries). Femoral arteries were cannulated with PE50 tubing for mean arterial blood pressure (MAP) measurements and blood withdrawal. Blood oxygen saturation was continuously monitored by a pulse oximeter positioned on the hind limb. The physiological parameters in control state were S_aO₂ = 98 ± 1%, pH= 7. 4 ± 0. 08, P_aCO₂ = 34 ± 4 mmHg and MAP= 110 ± 10 mmHg. All procedures were approved by the research animal committee of Massachusetts General Hospital, Harvard Medical School.

2.2. FMRI Studies

fMRI experiments were performed using a 9.4 T/21 cm horizontal bore (Magnex Scientific) using a Bruker Advance console and custom-made surface-RF coil. In order to minimize motion artifacts, the rat was secured to the RF coil by a bite bar resting below the upper hard palate and over the snout along with a mask for the delivery of anesthetic gas. Coronal localization of slices was accomplished using an initially obtained mid-line sagittal slice and comparing it with the sagittal section from a stereotaxic rat brain atlas (32). Five contiguous coronal slices were selected over the region −5 mm to 0 from the Bregma covering the somatosensory cortex, thalamus, caudate putamen and hippocampus. All catheters for mean arterial pressure (MAP) measurement and anesthesia delivery tubes were brought outside the magnet room of the MR scanner. Anatomical images were obtained before fMRI scanning using a RARE sequence with repetition time (TR) = 1 s, echo time (TE) = 19 ms, 256 × 256 matrix and field-of-view (FOV) = 3.0 cm. For fMRI-BOLD measurements, a single shot gradient EPI sequence was used to acquire multiple slices of images using a 128 × 128 matrix, TR/TE=2 s/15 ms, FOV=3.0 cm, slice thickness=1 mm. The resulting BOLD image had a spatial resolution of 0. 24 × 0. 24 × 1 mm³. Four hundred and fifty images were obtained in about 15 minutes in each scanning session. Studies were performed in sequential order starting with the rest scan, followed by exsanguination, where 8 ml/kg of blood was withdrawn from the femoral artery gradually in about 2–3 min. 8 ml/kg of blood was withdrawn since it led to a decrease in MAP almost near the autoregulatory limits in the present anesthetized rat preparation. MR images were acquired 5 min after withdrawal of the required volume. Finally the withdrawn blood was gradually replaced back into the arterial system at a rate slower than withdrawal. MR images were obtained 5 minutes after replacement was complete.

2.3. Data Analysis and Statistics

Low-pass filtered resting BOLD signal time courses from voxels in the sensorimotor cortex were cross-correlated with every voxel time course in the brain. Previously, it was observed that a significant temporal correlation was obtained with voxels from the sensorimotor and its associated cortex in humans (25, 26). Very few voxels outside the sensorimotor cortex were reported to have significant correlation in humans. In the present study using the rat model, the sensorimotor cortex, caudate putamen, hippocampus and thalamus were defined as regions of interest according to the stereotaxic rat brain atlas (32). A gaussian low-pass filter with a cutoff at 0.1 Hz was applied to all voxel time courses (33). This reduced respiratory and cardiac signals and any corresponding aliased signals.

2.4. Frequency Estimation

The frequencies present in the voxel time-series data were calculated using Welch’s averaged periodogram method (34). Briefly, each data set was divided into eight sections, with 50% overlap between adjacent blocks. A Hamming window was then used for each section and the power spectrum computed. After the power spectrum was calculated from each segment, the power spectra were averaged for the eight segments. This was done on a voxel-wise basis for the entire brain. After the power spectrum was calculated, the frequency with the largest amplitude was identified as the dominant frequency noted for all the voxels.

2.5. Temporal Correlation

The temporal correlation between voxels was assessed by two different techniques. The first technique used cross-correlation analysis between six seed pixels from the center of region chosen from the function anatomy and was correlated with every voxel in that slice. Since we are only interested in the temporal correlation due to slow periodic spontaneous oscillations, a finite impulse response filter (35) was used to filter the high-frequency components from each of the data sets (as described above). Because of the short data size (110 time points), filter parameters were adjusted to minimize the generation of artifactual frequencies (sidelobes). The correlation coefficient was then calculated using the formula r = X*Y/(XY), where X and Y represent time course pixels from two voxels X and Y, and * represents the dot product. All voxels that passed a correlation of 0.3 were considered significant and locations were noted.

While resting state connectivity maps were generated to show the spatial location of regions that correlated significantly with a representative time course in the brain, they do not necessarily provide much information regarding the interaction between the different subregions. Connectivity patterns between different regions of interest (both for the visual and sensorimotor systems) were obtained using cross-correlation analysis. For each subregion chosen for the sensorimotor or visual systems, principal component analysis was computed (36). Using PCA, a set of orthogonal data sets based on their energy content was determined. Briefly, PCA is a multivariate technique that replaces the measured variables by a new set of uncorrelated variables (principal components), arranged in the order of decreasing standard deviation (SD) or energy distribution. For this study, the first two components accounted for more than 75% of the standard deviation for each region of interest and were therefore used. The resulting time series (consisting of the first PCA) for each subregion was correlated with every other time component to obtain a pair-wise correlation matrix.

While significant temporal correlation typically represents similar time-series structure, it does not give any information regarding the frequency components that may give rise to high correlation coefficients. Therefore, in addition to cross-correlation, coherence maps were generated from which the contribution of specific frequencies could be estimated and compared with the temporal correlation map.

3. Results

BOLD weighted MR images at 9.4 T were obtained in isoflurane-anesthetized rats during various physiological conditions using a gradient echo EPI sequence. During normal physiological conditions, all animals imaged had MAP values ranging between 90 and 120 mmHg with baseline fluctuations in the fMRI-BOLD signal intensity in the frequency range 0.01–0.125 Hz. Figure 12.1a,d shows the spatially averaged power spectrum and the low-pass filtered BOLD signal time series from the whole brain during normal physiological condition for a typical rat.

Fig. 12.1.

Average power spectra of the BOLD signal, average filtered BOLD signal time series and standard deviation of the BOLD signal from the whole brain in a typical rat during normal, exsanguinated and blood replaced conditions. (a–c) average power spectra (d–f) average filtered BOLD signal time series, (g) anatomy (h–j) BOLD signal standard deviation maps during normal, exsanguinated and blood replaced conditions respectively. (See Color Plate)

After MR measurements during normal physiological condition, rats were exsanguinated where a small volume (8 ml/kg) of blood was gradually withdrawn from each animal resulting in hypovolemia and hypotension. MAP dropped from the normal control value of 110 ± 10 – 68 ± 7 mmHg during exsanguination in all rats and led to an enhancement in the amplitude of the low frequency BOLD fluctuations. No significant change was observed in S_aO₂ during exsanguination which remained at 98 ± 1%. This was accompanied by an enhancement in the magnitude of the power spectrum in the low frequency range. Two of the rats studied showed a very minimal (2%) decrease in the mean baseline BOLD signal intensity during exsanguination while five rats showed no significant change. Frequencies centered at 0.02, 0.03, 0.07, 0.10 and 0.125 Hz were significantly enhanced. Figure 12.1b,e shows the spatially averaged power spectrum of the BOLD signal and the low-pass filtered BOLD signal time series respectively from the whole brain in a typical rat during exsanguination.

Figure 12.1c,f shows the average power spectrum and the low-pass filtered BOLD signal time series from the whole brain in the same rat after blood replacement. The total time duration between blood withdrawal and replacement was around 30 minutes in all experiments. As indicated by the standard deviation maps from a typical rat, the enhancement in BOLD signal fluctuations was the maximum in cerebral cortex (Fig. 12.1i). Withdrawn blood volume, when replaced, led to a partial recovery of the BOLD signal fluctuations (Fig. 12.1j). An interesting observation was the enhancement in the fluctuations in the very low frequency range below 0.01 Hz after replacement of blood (Fig. 12.1c).

The regional distribution of the low frequency BOLD fluctuation was analyzed from specific anatomical regions namely cerebral cortex, caudate putamen, hippocampus and thalamus traced according to the rat stereotaxic atlas (32). In any typical rat, the spatially averaged Fourier power of the low frequency BOLD fluctuations in the different anatomical regions were cerebral cortex>hippocampus>thalamus>caudate putamen (Fig. 12.2a-d). The same order in the Fourier power over anatomical regions was observed over all rats. Exsanguination also led to an increased power in most frequencies below 0.1 Hz over all rats.

Fig. 12.2.

Average power spectra of the BOLD signal from different regions of interest namely cortex, hippocampus, caudate putamen and thalamus in a typical rat during (a–d) normal and (e–h) exsanguinated conditions.

Temporal characteristics and spatial distribution of the low frequency physiological fluctuations were analyzed using seed voxels within the previously described anatomical regions of interest (ROI). Randomly selected seed voxels within each ROI were correlated with voxel time courses from the whole brain after low-pass filtering (cutoff frequency of 0.1 Hz). Figure 12.3a,c shows typical activation maps obtained by correlating the time course of a seed voxel from the sensorimotor cortex, hippocampus and thalamus with all other regions of the brain. Though small in area, highly correlated clusters were observed across the sensorimotor cortex from either hemisphere with the seed voxel chosen from the sensorimotor cortex (Fig. 12.3a). However, correlated clusters were not detectable with seed voxels chosen from the hippocampus or thalamus (Fig. 12.3b,c). Exsanguination, which increased the amplitude of BOLD signal fluctuations (Fig. 12.1d), led to an increase in the area of correlated voxels from both hemispheres after cross-correlation with a seed voxel time course chosen from different ROI’s (Fig. 12.3d-f). Correlated voxels in the hippocampus and thalamus across both hemispheres emerged during exsanguination when correlated with a seed voxel obtained from the hippocampus (Fig. 12.3e) and thalamus (Fig. 12.3f). The correlation maps were distinct depending on the anatomical region of choice of the seed voxel in any single rat. The correlation maps were reproducible across all seven rats. Figure 12.3g shows the correlation map obtained after cross-correlating the time course of voxels from the whole brain with a seed voxel from the sensorimotor cortex during exsanguination across all seven rats. The pattern of activation was approximately similar and exhibited bilateral symmetry over each hemisphere over all rats.

Fig. 12.3.

Spatial correlation of low frequency BOLD signals from five contiguous slices from a typical rat. The anatomical underlay consists of a single EPI image of the brain and the functional overlay is the correlation coefficient in the absence of any stimulus during ( a–c) normal and ( d–f) exsanguinated conditions. Voxels from the whole brain were low-pass filtered (cut off 0.1 Hz) on a voxel wise basis and subsequently cross-correlated with the time course of a seed voxel obtained from ( a,d) sensorimotor cortex ( b,e) hippocampus and ( c,f) thalamus. ( g) Typical cross-correlation maps from a single coronal slice during exsanguinated conditions over all rats after cross-correlation with the time course of a seed voxel obtained from the sensorimotor cortex. A threshold (≥0.3 for the correlation coefficient (P< 10⁻⁶) was used to generate all correlation maps. The seed voxel location shows a high correlation coefficient value in the images. (See Color Plate)

Statistical analysis was performed across all rats to determine the extent of spatial correlation between regions. The average correlation coefficients of stereotaxically defined regions namely cerebral cortex, caudate putamen, hippocampus and thalamus (using seed pixels from the cerebral cortex) were determined over each rat and compared among all rats. The average correlation coefficient was the largest in the cerebral cortex, and the least in the hippocampus both during normal and exsanguinated conditions (Table 12.1). The correlation of cerebral cortex was relatively better with caudate putamen and thalamus when compared to the hippocampus. The average correlation coefficient for each of the stereotaxically defined anatomical regions of interest increased during exsanguination when compared to normal physiological conditions. The average correlation coefficients from various anatomical regions of interest over all rats is shown in Table 12.2.

Table 12.1.

Average correlation coefficients in stereotaxically defined anatomical regions of interest after cross correlation with 6 different seed voxels chosen from various spatial locations within the cerebral cortex region of interest from a typical rat (rat2)

Cerebral cortex		Caudate putamen		Hippocampus		Thalamus
Normal	Exsanguinated	Normal	Exsanguinated	Normal	Exsanguinated	Normal	Exsanguinated
0.08	0.28	0.05	0.18	0.02	0.13	0.05	0.22
0.07	0.33	0.01	0.24	−0.03	0.14	0.02	0.28
0.09	0.19	0.03	0.17	−0.02	0.07	0.01	0.11
0.13	0.24	0.04	0.14	0.01	0.13	0.02	0.19
0.10	0.29	0.06	0.17	0.02	0.14	0.05	0.23
0.10	0.28	0.04	0.18	0.00	0.14	0.04	0.24

Table 12.2.

Grand mean of the correlation coefficients in different anatomical regions of interest over all rats. The signal time series from six seed voxels in various spatial locations within the cerebral cortex region of interest was cross-correlated with all voxels from the cerebral cortex and other anatomical regions of interest namely caudate putamen, hippocampus and thalamus. Significance was tested considering 6 rats, which were “successful” under the hypothesis that exsanguination increases the average correlation coefficient in the cerebral cortex

Animal	Cerebral cortex		Caudate putamen		Hippocampus		Thalamus
	Normal	Exsanguinated	Normal	Exsanguinated	Normal	Exsanguinated	Normal	Exsanguinated
Rat1	0.043	0.077	0.018	0.050	−0.013	−0.042	0.033	0.044
Rat2	0.098	0.266	0.038	0.136	0.001	0.128	0.027	0.212
Rat3	0.071	0.142	−0.035	0.038	0.043	0.031	0.095	0.055
Rat4	0.080	0.330	0.051	0.243	0.006	0.146	0.095	0.262
Rat5	0.056	0.049	−0.030	0.036	−0.021	−0.002	0.036	0.042
Rat6	0.075	0.352	0.038	0.236	0.026	0.131	0.035	0.278
Rat7	0.083	0.271	0.020	0.216	0.003	0.125	0.021	0.194
Mean	0.07	0.21	0.01	0.14	0.01	0.07	0.05	0.16
SD	0.13	0.12	0.04	0.10	0.02	0.08	0.03	0.11

Note: P < 0. 001 cortex with respect to hippocampus during normal condition. P < 0. 0003 cortex with respect to hippocampus during exsanguinated condition. P < 0. 008 cortex with respect to caudate putamen during normal condition. P < 0. 001 cortex with respect to caudate putamen during exsanguinated condition. P < 0. 19 cortex with respect to thalamus during normal condition. P < 0. 0004 cortex with respect to thalamus during exsanguinated condition.

A sign and binomial test was performed under the hypothesis that an increase in mean correlation coefficient in the cortex with exsanguination was a “success” and a decrease in mean correlation coefficient was a “failure”. In six out of a total of seven rats imaged, exsanguination led to an increase in the mean correlation coefficient in the cortex. Thus, with six “successful” trials out of a total of seven, and for a probability of success in each trial as 0.95, the chance of observing six or more successes in seven trials was 95.562%. This implied our underlying hypothesis to be valid. In six rats where our hypothesis was valid, the difference between the mean correlation in different anatomical regions was tested using a paired t-test during normal and exsanguinated conditions. As shown in Table 12.2, the error probability in the paired hypothesis test for significant difference between regions decreased during exsanguination when compared to normal condition.

Spatial distribution of the low frequency fluctuations in the whole brain was generated from specific frequency bands from the power spectra of the signal time series on a voxel wise basis. Figure 12.4 shows color-coded maps of the Fourier power of specific frequency bands 0.01, 0.02, 0.03, 0.07 and 0.1 Hz from a typical rat. Low-frequency fluctuations, as observed from the distinct frequency bands, originated predominantly from the cortical region spanning both hemispheres. The power of all frequencies below 0.1 Hz increased during exsanguination in the cerebral cortex.

Fig. 12.4.

Spatial maps of low frequency fluctuations in the BOLD signal at distinct frequencies in a typical rat during normal (N) and exsanguinated (E) conditions. Images were derived from specific frequencies namely 0.01, 0.02, 0.03, 0.07 and 0.10 Hz respectively from the power spectra of the BOLD signal time series on a voxel wise basis. Fluctuations in most of the low frequency bands increase spatially in the cerebral cortex during exsanguination. (See Color Plate)

4. Discussion

4.1. Baseline BOLD Signal Decrease with Hypotension

Exsanguination led to an enhancement in the magnitude of the low frequency fluctuations (Fig. 12.1e) despite variation in the baseline BOLD signal decrease. Two of the rats studied indicated a very minimal (2%) decrease in the mean baseline BOLD signal intensity during exsanguination while five rats showed no significant change. The small variation in the mean BOLD signal change observed in two rats may be due to inherent variability in autoregulatory limits in normal anesthetized rats (37). A change in hematocrit as a result of hypovolemia can also cause a minimal decrease in the baseline BOLD signal. Variation in the increase in arteriolar CBV with autoregulated CBF during hypotension can influence the decrease in baseline BOLD signals at high fields (38). However, the rate of MAP change determines the capacity of cerebral vasculature to maintain mean CBF levels. Barzo et al. (39) have observed that baseline CBF remains unchanged if hypotension is induced at a rate less than 24 mmHg/min in urethane anesthetized rats. The rate of decrease in MAP in our study was less than 24 mmHg/min as exsanguination was carried out by passive bleeding of the rat over a period of three minutes and is very unlikely to have contributed to the decrease in the baseline BOLD signal. As hypotension does not significantly affect the BOLD signal response to neural activation (40), exsanguination-induced hypotension may not confound the hemodynamic fluctuations if influenced by underlying neural activity.

Amplitude of the BOLD signal fluctuations decreased substantially after replacement of withdrawn blood. While the MAP returned to the normal control levels, the amplitude of the BOLD signal fluctuations did not completely decrease to control levels. The hysteresis in the BOLD signal fluctuation amplitude cannot be attributed to hypotension alone since alternate methods that induce hypotension such as lower body negative pressure in anesthetized rats have shown to be relatively non-invasive where CBF fluctuation amplitude completely returned to normal control levels prior to hypotension (41). Events such as plasma volume refill can occur during exsanguination by transfer of extravascular fluid into the circulation that can transport brain extracellular compounds into the blood stream (42). It is possible that such circulating compounds may have influenced the flow fluctuations even after replacement of withdrawn blood. Further, it is also possible that the return to normal levels may be beyond the measurement window of the present study.

4.2. Dependence of BOLD Signal Fluctuations on CBF and CBV

LDF studies indicate that a decrease in MAP to threshold of autoregulatory limits does not decrease mean blood flow values but increases amplitude of CBF fluctuations, which fluctuates across the same mean value prior to drop in MAP (5, 43). Furthermore, spontaneous fluctuations in BOLD and CBF signals and their dependence on MAP have been observed under different anesthesia (43, 44). The hypocapnia dependent modulation in the amplitude of the low frequency BOLD signal fluctuations, suggests that they are strongly connected to spontaneous CBF fluctuations that have a similar dependence on MAP (5).

BOLD signal response when acquired with a gradient-echo sequence is sensitive to vascular caliber and density. Differences in metabolic regulation and vascular density in different regions affect the BOLD contrast to noise ratio (Fig. 12.1h). The rat cortex has a larger blood volume than the thalamus and other deeper structures of the brain (45). The Fourier power of the low frequency physiological fluctuations in the different anatomical regions and their enhancement in response to exsanguination were cerebral cortex>hippocampus>thalamus>caudate putamen (Fig. 12.2), following a vascular density or CBV weighted dependence of the BOLD signal fluctuations. The enhancement in the amplitude of the low frequency fluctuations in BOLD signal with a decrease in intravascular pressure and tone similar to LDF fluctuations during exsanguination (5) indicates a strong link between fluctuations in brain oxygenation and CBF fluctuations in the microvascular network. While this suggests a myogenic component in the generation of the observed low frequency fluctuations in BOLD signal, some of the results also support underlying neural activity. Though simultaneous measurements of neuronal activity were not carried out in the present study, direct measurement of neuronal signals (spike rate, LFP or EEG) from both hemispheres of the sensorimotor cortex during exsanguination would help clarify the myogenic and neuronal contribution to the observed BOLD signal fluctuations.

4.3. Implications for Resting State Connectivity

Connectivity maps in humans have been generated by cross-correlating the signal time course of every voxel in the brain with a seed voxel chosen from a region where brain activation from the respective sensory or motor stimuli is expected (26, 46, 47). Using a similar analysis, during normal resting conditions in the anesthetized rat, cross-correlating a seed voxel from a region of interest encompassing the sensorimotor cortex, hippocampus and thalamus with the rest of the brain indicated sparse activation with coherent voxels in both hemispheres of the rat cortex (Fig. 12.3a-c). Exsanguination led to an increase in the number of correlated voxels across both hemispheres of the brain (Fig. 12.3d-f). The spatial extent of the correlation maps was dependent on the anatomical region of choice of the seed voxels that were used for the cross-correlation indicating distinct temporal characteristics of the fluctuations. The correlation maps indicate mostly similar phase characteristics of the low frequency BOLD signal fluctuations in the cortex and thalamus but a different phase in the hippocampus (compare Fig. 12.3d,e,f). Most of the cortical regions were negatively correlated with the hippocampal seed voxel (Fig. 12.3e). Within the sensitivity of the present experimental protocol, the distinct phase synchrony observed in the hippocampus may represent contribution from specific extravascular factors, which may be neurometabolic and/or neural signaling in origin. Thus, hypotension near the autoregulatory limits can sufficiently enhance the MR-sensitivity and, to a certain extent, may reflect the ‘resting state’ functional networks in the anesthetized rat model.

Aliasing of frequencies greater than the critically sampled ones can be a concern in the low frequency region. The frequencies of respiration and heart rate in the isoflurane-anesthetized rats were in the region of 1.2 Hz and over 5 Hz respectively and the TR has to be less than 100 ms to theoretically avoid aliasing (48). Recently, using very low and high sampling rates (TR=125 ms and TR=3 s), DeLuca et al. (49) have studied the aliasing effects of physiological processes such as cardiac and respiratory fluctuations on the resting state fluctuations in the BOLD signal. The cardiac and respiration frequencies were spatially distinct from the resting state functional network and do not significantly affect the low frequency BOLD fluctuations. In order to minimize any aliasing effects, low-pass filtering (0.1 Hz cutoff) was applied to all data before the correlation analysis. Further, spatial distribution of prominent frequencies in the range between 0.01 and 0.1 Hz was ubiquitous across various anatomical regions with a strong presence in the cortex. No distinct presence of specific frequencies was observed in any one anatomical region either during normal or exsanguinated conditions (Fig. 12.4). As much of the fluctuation signal is located over multiple frequencies in the superficial part of the cortex, a strong fluctuation signal is also consistent on the base of the brain, midline and third ventricle (Fig. 12.4). Ventricular spaces are susceptible to artifacts from CSF pulsations from high frequency sources such as respiratory and cardiac cycles that may be aliased into the low frequency region in the power spectrum.

It is interesting to note, however, that a similar high correlation of the low frequency BOLD signal fluctuations in both hemispheres of the cerebral cortex existed for at least five different frequencies, which get augmented during exsanguination (Fig. 12.4). While this is unlikely due to aliasing effects in the low frequency range, it is possible that connected neuronal networks may be oscillating at multiple frequencies which is more consistent with neuronal than myogenic origin of the BOLD signal fluctuation.

4.4. Implication for Anesthesia Studies in Humans

The increase in the very low frequency below 0.01 Hz in the BOLD signal fluctuations during exsanguination, which remained after replacement of withdrawn blood (Fig. 12.1b and c), may indicate an alteration in autoregulatory equilibrium. Though the reasons behind the loss of CBF autoregulation are still controversial, anesthesia can have a major influence on autoregulation (43, 50). The increase in the gain of the fluctuating feedback system that led to an increase in BOLD signal fluctuations during exsanguination may indeed be a result of an alteration of autoregulatory equilibrium (43, 51, 52). These results are in concurrence with studies on anesthetized children undergoing fMRI which have shown significantly altered baseline BOLD signal fluctuations in the very low frequency range 0.02–0.04 Hz (53) when compared to BOLD signal fluctuations at 0.1 Hz in awake human subjects (26). Prominent, very low frequency fluctuations (0.02–0.04 Hz) are suggested to occur in the absence of autoregulation (18). A complete physiological monitoring and biophysical characterization of physiological fluctuations using the anesthetized rat model can be very useful in translation studies using anesthesia and fMRI in humans where invasive physiological monitoring can be a problem due to ethical reasons (53).

5. Conclusions

Exsanguination significantly enhances the amplitude and spatial spread of the low frequency BOLD fluctuations to most of the cortical, sub-cortical and deeper brain structures in the isoflurane-anesthetized rat brain. The behavior of the resting state BOLD signal fluctuations similar to CBF fluctuations during exsanguination indicates a myogenic dependence. On the other hand, a high intra-hemispheric symmetry in the BOLD fluctuations with similar phase characteristics in most other regions except the hippocampus suggests that the fMRI-BOLD signal fluctuations in anesthetized rats may carry information from distinct functional networks. Hypotension near the autoregulatory limits in anesthetized rats can be used to improve the detection of distinct resting state neural networks despite the myogenic artefact.