Differential Effect of Global Signal Regression Between Awake and Anesthetized Conditions in Mice

Abstract

Introduction:

In resting-state functional magnetic resonance imaging (rs-fMRI) studies, global signal regression (GSR) is a controversial preprocessing strategy. It effectively eliminates global noise driven by motion and respiration but also can introduce artifacts and remove functionally relevant metabolic information. Most preclinical rs-fMRI studies are performed in anesthetized animals, and anesthesia will alter both metabolic and neuronal activity.

Methods:

In this study, we explored the effect of GSR on rs-fMRI data collected under anesthetized and awake state in mice (n = 12). We measured global signal amplitude, and also functional connectivity (FC), functional connectivity density (FCD) maps, and brain modularity, all commonly used data-driven analysis methods to quantify connectivity patterns.

Results:

We found that global signal amplitude was similar between the awake and anesthetized states. However, GSR had a different impact on connectivity networks and brain modularity changes between states. We demonstrated that GSR had a more prominent impact on the anesthetized state, with a greater decrease in functional connectivity and increased brain modularity. We classified mice using the change in amplitude of brain modularity coefficient (ΔQ) before and after GSR processing. The results revealed that, when compared with the largest ΔQ group, the smallest ΔQ group had increased FCD in the cortex region in both the awake and anesthetized states. This suggests differences in individual mice may affect how GSR differentially affects awake versus anesthetized functional connectivity.

Discussion:

This study suggests that, for rs-fMRI studies which compare different physiological states, researchers should use GSR processing with caution.

Impact statement

Global signal regression (GSR) is a commonly used denoising technique, however, it can also introduce specific artifacts. In functional connectivity (FC) studies of functional magnetic resonance imaging, it is crucial to examine the impact of GSR on animals under various anesthesia conditions. Our results showed that GSR had distinct effects on FC and modularity of awake versus anesthetized mice. Therefore, caution should be taken when using GSR to remove noise, especially when comparing different neurometabolic states.

Introduction

Functional magnetic resonance imaging (fMRI) is a noninvasive and robust method for studying brain function (Biswal et al., 1995; Schifani et al., 2022). Due to its ease of signal acquisition, resting-state fMRI (rs-fMRI) is one of the most commonly used fMRI techniques for obtaining information about the continuous and internal neural activity that supports cognitive function (Smitha et al., 2017). rs-fMRI can identify brain networks similar to those observed under task- or stimulation-induced conditions, and has been used to develop biomarkers for numerous neurodegenerative disorders and neurological diseases (Guerra-Carrillo et al., 2014; Harrewijn et al., 2020; Martin et al., 2021).

rs-fMRI recordings of spontaneous fluctuations of blood-oxygenation-level-dependent (BOLD) signals contain contributions from physiological and nonphysiological sources. Denoising aims to separate the neuronal component of the BOLD signal from the physiological and nonphysiological signals (Parkes et al., 2018; Power et al., 2017). Global signal regression (GSR) is one of the most controversial denoising steps (Li et al., 2019; Murphy and Fox, 2017; Uddin, 2017). GSR uses linear regression to remove the average whole-brain signal from each voxel, because the global signal (GS) is associated with motion, respiration, and heartbeat artifacts (Burgess et al., 2016; Liu et al., 2017; Power et al., 2018). Thus, GSR effectively removes global artifacts caused by movement and other physiological sources (Ciric et al., 2017; Power et al., 2014).

However, GSR can also introduce specific artifacts in fMRI data, such as “artificial” anticorrelations in modeled data, which were created without such anticorrelations originally (Anderson et al., 2011; Murphy et al., 2009; Xu et al., 2018). In addition, GSR can lead to false differences in functional connectivity (FC) in studies of psychiatric diseases, for example, schizophrenia (Parkes et al., 2018; Yang et al., 2014). These findings indicate that although advantages exist to using GSR, it may cause unexpected changes, particularly when comparing multiple groups.

The majority of experiments studying FC changes in animal disease models are conducted under general anesthesia to reduce the effects of motion artifacts and stress responses (Grandjean et al., 2014; Lukasik and Gillies, 2003). However, the effects of anesthesia on fMRI signals are substantial (Gao et al., 2017; Liang et al., 2015; Paasonen et al., 2018; Xie et al., 2021). More and more studies have demonstrated that anesthetics can affect neural activity (Bharioke et al., 2022; Yatziv et al., 2020), the source of the fMRI signal. Hamilton et al. (2017) observed that FC was significantly altered within the thalamocortical and frontoparietal networks under awake versus unconscious conditions. Furthermore, anesthesia also affects the physiological function of experimental animals, reducing the brain's energy metabolism, altering the function of neural circuits, and altering the organization of brain networks.

Recent development of an awake mouse model has demonstrated that awake and anesthetic states have specific fMRI characteristics (Sitsen et al., 2022). Liang et al. (2015) showed weaker coactivated patterns in the anesthetized state compared with the awake state. Dinh et al. (2021) compared BOLD responses from anesthetized and awake mice under visual stimulation. They observed that mice in the awake state had faster BOLD responses in active regions and also had stronger responses in subcortical areas, in comparison with the anesthetized state (Dinh et al., 2021). These findings demonstrate that fMRI signals are altered by different anesthetic conditions. Because of the large effects of anesthesia and GSR, it is likely that they may interact in unpredictable ways. In particular, specific networks may be affected by GSR differently whether under anesthesia or not.

In this work, we aim to investigate the impact of GSR on resting-state FC under awake and anesthesia conditions. Moreover, we examine whether and to what extent GSR affects FC between these conditions. We used FC, functional connectivity density (FCD) mapping, and network modularity to quantify the effect of GSR. We observed that both FC networks and brain modularity are altered between awake and under anesthesia and that GSR has a more pronounced influence during anesthesia.

Materials and Methods

Animals

This study uses data from the same experimental animals and imaging sessions as Xu et al. (2022), however, this work examines BOLD-fMRI data that were collected but not analyzed in that previous work.

All animal experiments were approved by the Institutional Animal Care and Use Committee of ShanghaiTech University and were consistent with the governmental regulations of China for the care and use of animals. Fifteen male, adult, C57BL/6 mice, 7–9 weeks old and weighing 20–30 g, were imaged under awake and anesthetized conditions.

Mice were prepared for awake imaging in the same manner as in previous studies (Han et al., 2019). Briefly, a head brace was installed on each animal's skull to allow fixation of the head. The animals were allowed to recover for 1 week after surgery. After that, the animals were trained to acclimatize to awake imaging for a week. The mice were first imaged awake and again ∼1 week later under anesthesia. This order was used due to the concern that exposure and recovery from dexmedetomidine (Dex) could cause trauma if lucid imaging was performed later. Notches at the edge of the ear of each mouse were cut with scissors, to identify the same mouse under different conditions. Mice were euthanized after the second (Dex) scan. In total, data were collected from 12 mice, for details on sample size selection and mice, including excluded data, see Xu et al. (2022). This study used the same mice data as that study.

Animal preparation

Mice were given a total concentration of 37% medical air (20% O₂, 80% air) during imaging, and respiration was monitored using pneumatic pillows. When imaging under anesthesia (“Dex” condition), mice were initially induced by 3.5% isoflurane, and 1.5% isoflurane was used in the imaging preparation process. A total of 0.025 mg/kg Dex was injected intraperitoneally, and then, isoflurane was discontinued. After 10 min, a constant Dex infusion rate (0.05 mg/kg/h) was maintained under anesthesia. During anesthesia, a warm water pad was used to maintain the body temperature at ∼37.0°C, and a small animal rectal probe was used to detect the body temperature. The respiratory rate of mice was 277 ± 16 breaths per minute when awake and 150 ± 8 breaths per minute when under anesthesia.

Animal imaging

The MRI measurements were performed using a Bruker BioSpec 9.4T scanner, equipped with an 86-mm volume coil (Bruker), and received using a four-channel phased array cryogenic mouse head coil (Bruker). First, anatomical images were acquired using a T2-weighted MRI (T2WI) rapid acquisition with relaxation enhancement scan, repetition time (TR) 3300 ms, echo time (TE) 33 ms, field-of-view (FOV) 16 × 16 mm², slices: 20, slice thickness 0.5 mm, and matrix size 256 × 256. Then, a gradient-echo echo-planar imaging (EPI) sequence, TR 1000 ms, TE 15 ms, FOV 15.975 × 12.15 mm², slices: 20, slice thickness 0.5 mm, matrix size 71 × 54, and repetitions 600, was used to obtain rs-fMRI BOLD data.

MRI preprocessing

fMRI BOLD data were preprocessed and analyzed using in-house MATLAB code. Motion correction and slice timing were performed in SPM12 (Wellcome Trust Centre for Neuroimaging, London, United Kingdom). Anatomical images were registered to the mouse brain template using a nonlinear registration with BioImage Suite (Yale School of Medicine, 2015; bioimagesuite.yale.edu). Each mouse's corresponding nonlinear registration was then applied to fMRI data from the same mouse. To facilitate group statistics, each fMRI image was spatially smoothed (0.4 mm isotropic Gaussian kernel, full width at half maxima = 5 mm), and each voxel's time series was bandpass filtered from 0.01 to 0.3 Hz (Grandjean et al., 2014; Schroeter et al., 2017; Zerbi et al., 2015).

To reduce noise and improve the quality of the BOLD signal, group independent component analysis was performed in EPI images using the GIFT toolbox, as per our previous work (Han et al., 2019). The number of independent components, as estimated by the minimum description length criteria, was 8. Among them, components 1, 2, 4, and 8 were removed because they were identified as the signals from ventricles and white matter (Supplementary Fig. S1).

To further increase BOLD signal quality, epochs of the BOLD signal with a low signal-to-noise ratio were removed. Because the primary somatosensory cortex of the forelimb (S1FL) has a high correlation between the left and right hemispheres (de Arce et al., 2023), lacking correlation in S1FL is a sign of a lack of brain networks in general, indicating a potential problem in the physiological state. Thus, data whose Pearson's correlation coefficient (PCC) between bilateral S1FL was lower than 0.3 were removed from FC results, as per our previous work (de Arce et al., 2023).

GS calculation

The GS was defined as the average of the time course signals from all voxels in the whole-brain mask. Correlation with the GS was calculated by using the PCC between each mouse's GS and each voxel's time course signal. These calculations were done as per previous work, which examined the GS (Thompson et al., 2016).

The standard deviation was also used to measure the amplitude of the GS (Zarahn et al., 1997). Previous work, which calculated the amplitude in this manner, used a different type of calculation for the GS (Wong et al., 2012), and thus, we used an alternate calculation, which we refer to as GS-%. To calculate GS-%, for each voxel, the percentage change time series (C) was calculated by the preprocessed MR signal (S) subtracting the mean value ( $S_{m e a n}$ ) and then dividing the difference by mean value, shown in Equation (1) (Thompson et al., 2016; Wong et al., 2012). GS-% was defined as the average of all C in the whole-brain mask. $C = (S - S_{m e a n}) ∕ S_{m e a n}$ (1)

Region-of-interest-based FC calculation

From fMRI data, the correlation coefficients between the time courses of regions-of-interest (Supplementary Table S1) were calculated (25 regions, see the Results section). Two types of FC calculations were performed as follows: (i) Calculate the correlation between different brain regions based on unilateral data, such as left posterior parietal association area and left visceral area (VISC). In the figures, these correlations are displayed on the upper triangular half of matrices for the right side of the brain, and on the lower triangular half of matrices for the left side of the brain. (ii) Calculate the correlation in the same brain region based on bilateral data, such as left retrosplenial area (RSP) and right RSP. In the figures, these correlations are displayed along the diagonal of matrices.

FCD calculation

Using PCC, FCD mapping is a metric based on graph theory that uses rs-fMRI data to generate a map of the number of functional connections (either local or global) for every voxel in the brain (Tomasi and Volkow, 2010). FCD mapping evaluates the local functional connectivity density (lFCD) by restricting the functional connections to the nearby cluster of other voxels around a voxel, and determines global FCD by counting the number of connections to each voxel across the whole brain (Tomasi and Volkow, 2011). In this study, we used lFCD to calculate the number of voxels linked with a particular voxel, and Pearson correlation was utilized with correlation thresholds set at R ≥ 0.25. The size of the nearby cluster was radius = 2 mm (Cohen et al., 2018). lFCD was calculated on a per-mouse basis and average results are shown as pseudocolor brain maps in the figures.

Brain modularity calculation

Modularity was calculated from unweighted and undirected correlation matrices (as calculated in the previous section) using the modularity_und tool from the Brain Connectivity Toolbox (Rubinov and Sporns, 2010). For valid comparisons between mice and scans, 100 modularity scores were generated. The partition with the largest modularity score was normalized by the average modularity generated from 100 randomly rewired (shuffled) FC matrices (Maslov and Sneppen, 2002). This procedure was repeated separately for each mouse and scan.

Intraclass correlation coefficient

Reliability was measured using the intraclass correlation coefficient (ICC) as per recommendations for rs-fMRI studies (Noble et al., 2021). Two types of ICC were calculated. In the first manner, awake or Dex groups were randomly divided into two groups to calculate ICC with six mice in each group. In the second manner, mice were divided into two groups according to the degree of change of Q value with groups for large and small change. The R language was used for this calculation, the model used was two-way, and the type was consistency.

Results

GS difference between awake and Dex mice

We first examined the GS itself, and the effects of anesthesia on it following the methods of Wong et al. (2012). We calculated the percentage change of time series (C) under awake and anesthesia conditions. Average C within two brain regions, infralimbic area and orbital area, was statistically significantly different between different states (Fig. 1b). However, for the mean of C across the entire brain, GS-%, there was no statistical difference in the standard deviation between the different states (Fig. 1c).

FIG. 1.

Mean of percentage change time series (C) from awake and anesthetized mice. (a) Average value of C across all mice in awake and Dex groups, 20 coronal slices from caudal to rostral shown. (b) Comparison of C averaged within 25 brain regions (* indicates p < 0.05 SGoF corrected, p = 5.56 × 10⁻⁷ in ILA region, p = 8.88 × 10⁻⁵ in ORB region). (c) Standard deviation of the GS-% (average of C across whole brain, no units). Awake mice are green dots, Dex mice are red dots. The difference between awake and Dex mice was not statistically significant (p = 0.068). N = 12 mice (under both conditions). Dex, dexmedetomidine; GS, global signal; ILA, infralimbic area; ORB, orbital area; SGoF, sequential goodness of fit.

FC alteration due to GSR

To explore the effects of GSR on FC under different physiological states, we calculated the FC of mice under awake and Dex conditions. We then observed the effects of GSR by comparing the differences of FC matrices without (GSR⁻) and after GSR (GSR⁺). As shown in Figure 2, GSR decreased FC in both awake and Dex mice. (The difference between these maps is shown in Supplementary Fig. S2.) Our results showed that the decrease of FC was greater in the Dex condition. We also found that without GSR, the FC of mice under Dex was lower than that of awake mice, which is consistent with previous results (Areshenkoff et al., 2021; Paasonen et al., 2018).

FIG. 2.

Mean Pearson correlation coefficient (r-score) matrix of awake and anesthetized (Dex) mice with different GSR conditions. Four correlation matrixes are shown, which include 25 brain ROIs, including GSR⁺ under awake state (a), GSR⁺ under anesthetized state (b), GSR⁻ under awake state (c), GSR⁻ under anesthetized state (d). The lower left of the matrix represents the correlation coefficients between the ROIs of the left brain, the upper right represents the correlation coefficients between the ROIs of the right brain, and the diagonal represents the correlation coefficients between the left and right regions of the same ROI. GSR greatly affects the FC matrix in the Dex condition, versus awake mice. N = 12 for awake and Dex. Full names of brain regions are shown in Supplementary Table S1. FC, functional connectivity; GSR, global signal regression; ROI, region of interest.

Impact of GSR on modularity

Modularity revolves around the extent to which a complicated system is fragmented into distinct and self-contained constituent elements that perform unique functions and can operate in tandem with other elements. We examined the modularity scores of GSR⁻ and GSR⁺ under both awake and Dex mice (Fig. 3a, b). The degree of modularity in mice affected by GSR was represented by calculating the difference in Q between GSR⁺ and GSR⁻ conditions (ΔQ _G) divided by the Q value in GSR⁺ condition, yielding Q _G% (Fig. 3c). It was found that the modularity score increased in GSR⁺ under both awake and Dex states, and the change of degree of modularity score was greater in Dex mice than awake mice.

FIG. 3.

The effect of GSR and anesthesia on brain modularity change. Brain modularity was significantly greater in GSR⁺ (with GSR) both in awake (a) and Dex (b) states, indicating a global alteration in brain network integration due to GSR. The green dots represent the brain modularity score of GSR⁻ (without GSR), and the red dots represent the brain modularity score of GSR⁺ (with GSR). Statistical testing was performed using a paired-sample t-test, p < 0.05 (p _Awake = 4.41 × 10⁻⁴, p _Dex = 5.42 × 10⁻⁹). (c) Brain modularity changes between GSR⁺ and GSR⁻ in the awake and Dex group. Brain modularity score in Dex and awake group under GSR⁺ (d) and GSR⁻ (e) conditions. In both GSR⁺ and GSR⁻ conditions, there were no statistically significant differences in modular scores among mice in different anesthesia states. Statistical testing was performed using a paired-sample t-test, p < 0.05 (p _GSR+ = 0.09 p _GSR− = 0.11). (f) Brain modularity changes between Dex and awake groups under GSR⁺ and GSR⁻ condition. GSR can increase brain modularity score both in the awake and Dex groups, but the increase in the Dex group is greater than in the awake group. N = 12 for all groups.

In GSR⁺, Q was greater in most mice, but Q values in the Dex group were more tightly clustered around a small range of values than Q values in the awake group, and the percentage change was greater than in awake mice (Fig. 3c). By comparing the extent of anesthesia-induced modulation in modularity score, we discovered that mice under GSR⁺ conditions exhibited more prominent alterations in the modularity score than those under GSR⁻ conditions (Fig. 3f).

FCD map alteration due to GSR

Following the methods of Thompson et al. (2016), we examined the effect of GSR on FCD, as well as correlation with the GS itself. In most brain areas under GSR⁺, the FCD value of awake mice was higher than that of anesthetized mice, for example, VISC (p = 2.67 × 10⁻⁶, T = 7.091), CTXsp (p = 1.87 × 10⁻⁷, T = 8.829), and perirhinal area (PERI) (p = 1.93 × 10⁻⁷, T = 7.971) (Fig. 4b). However, for GSR⁻, most of these differences in brain regions were not statistically significantly different, and only the differences in CTXsp (p = 0.0024, T = 3.445) region under different anesthesia states were statistically significant (Fig. 4b). In GSR⁺ condition, the difference of FCD value between the anesthesia group and awake group was enlarged, and the difference of FCD values in different states was more obvious.

FIG. 4.

The effect of GSR on FCD maps and maps of correlation with the GS in awake and anesthetized states. (a) Average brain maps were generated to represent the measured parameters. The first row displays the FCD map of awake and Dex mice under GSR⁺ (with GSR) condition. The second row represents the FCD map of awake and Dex mice under GSR⁻ condition. The third row represents the map of correlation with the GS in awake and Dex groups. (b) Bar graphs of 25 brain regions' average FCD and average correlation with the GS between awake and Dex groups. FCD is shown under GSR⁻ and GSR⁺. Under the GSR⁺ condition, there were statistically significant differences in FCD values in many brain regions between awake and anesthetized mice. However, under the GSR⁻, the number of statistically significant brain areas was greatly reduced. GSR decreased FCD values in both awake and Dex states, but the difference between awake and Dex states was amplified in GSR⁺ in some brain regions. *Indicates p < 0.05, SGoF corrected. N = 12 for awake and Dex groups. FCD, functional connectivity density.

Correlation with the GS has previously also been related to brain physiology (Thompson et al., 2016). The correlation between most brain regions and the GS was higher under anesthesia, and the similarity between VISC (p = 0.0027, T = −3.381), temporal association area (TEa) (p = 0.0034, T = −3.276), PERI (p = 0.018, T = −2.576), and ectorhinal area (p = 0.0056, T = −3.11861) brain regions and GS was statistically significantly different between different anesthesia states.

GSR⁺ had lower FCD values than GSR⁻ in both the awake and Dex groups. However, while most brain regions had decreased FCD values, some brain regions had increased FCD values (Fig. 4). We used the FCD value of GSR⁻ minus GSR⁺ to obtain the difference map of FCD (Supplementary Fig. S3). We calculated the difference values in different brain regions, and made the statistical comparison between the awake and anesthesia states (Supplementary Table S2). We found that the changes under different physiological states, which differ under GSR⁺ performed, can present opposite trends. For example, the difference of FCD with GSR⁺ in VISC region in awake mice was significantly lower than Dex mice (p = 0.006096, T = −3.3843), while the difference of FCD with GSR⁻ in HY region in awake mice was significantly higher than anesthetized mice (p = 0.02694, T = 2.511, Supplementary Table S2).

Thus, it appears that GSR may have completely opposite effects on different brain regions depending on the current physiological state. This suggests that even while GSR reduces the FCD value of the whole brain, it can specifically enhance the FCD values in certain brain regions depending on the current physiological state.

We divided the data into two groups based on the degree of modularity change. Based on ΔQ _G, the highest 50% was designated as the “largest” group and the lowest 50% was designated as the “smallest” group. Thus, the “largest” group was highly influenced by GSR and the “smallest” group was less influenced by GSR (Fig. 5). In GSR⁻, FCD maps in the smallest and largest groups under both awake and Dex had no significant differences. In GSR⁺, some brain regions such as olfactory areas had no significant differences in FCD maps. However, certain brain regions such as somatosensory areas (SS) did have significant differences. In SS, for example, the smallest group was significantly greater under Dex, while there was no significant difference in SS between the smallest and largest groups in the awake state.

FIG. 5.

Differences in FCD maps between awake and Dex (anesthetized) mice with largest versus smallest Q differences in specific brain regions based on GSR⁺ versus GSR⁻. (a) The average of FCD maps under the largest change in Q, the smallest change in Q, and the difference between them in awake and anesthetized states. The “largest” group had a large change in Q, these mice FCD maps were strongly affected by GSR. The “smallest” group had a small change in Q, and these mice FCD maps were weakly affected by GSR. Difference is the average largest FCD map minus the average smallest FCD map. (b) FCD changes in SS and OLF between largest and smallest mice under awake and Dex condition. In the Dex group, there is a significant increase in FCD value in SS between the largest and smallest groups (paired t-test, p = 0.03). However, there was not a similarly significant change in SS for awake mice (paired t-test, p = 0.99) or a change in any case in OLF (paired t-test, p = 0.35). N = 6 for largest group and smallest group. *p < 0.05. OLF, olfactory area; SS, somatosensory areas.

ICC reliability

When we used the degree to which FCD was affected by GSR to sort (“largest” vs. “smallest” groupings), we found that both sets of ICC values are greater than 0.4, which indicates good reliability (Noble et al., 2021). When we used a random ordering, the awake group still had good reliability, with an ICC of 0.599. However, the Dex group had an ICC of 0.138 with a random ordering (Table 1). This showed that the large change and small change groups are reliable within each group, but not between the two groups.

Table 1.

Intraclass Correlation Coefficient Values in Awake and Dexmedetomidine Groups When Using Different Block Methods

	Awake	Dex
Random	0.599	0.138
Large versus small	0.828	0.681

ICC values under two methods of ordering samples (mice) and two kinds of anesthesia. The two sorting methods are “Random,” which is the ICC value under random ordering, and “Large versus Small,” which is sorted according to the degree of Q value change for GSR⁺ versus GSR⁻.

Dex, dexmedetomidine; GSR, global signal regression; ICC, intraclass correlation coefficient.

Discussion

We explored the effect of GSR on FC under awake and anesthesia conditions. We found that (i) there is no statistically significant difference in the amplitude of the GS between states; however, (ii) GSR decreases FC in awake and anesthetic states, but the anesthetic condition is more sensitive to changes due to GSR; (iii) GSR produced a more significant impact on brain network modularity in the anesthetized state; and (iv) within a given physiological state, some mice were affected more by GSR than others, and these effects were heterogeneous across different regions of the brain.

Studies have demonstrated that physiological state, cardiac and respiration rate, and vigilance are contributors to GS (Birn et al., 2006; Chang et al., 2016; Evans et al., 2015; Shmueli et al., 2007). The GS contains information both related and unrelated to neural activity. Wong et al. (2012) found that caffeine administration leads to widespread decreases in GS amplitude. However, using the same method (Fig. 1c), we did not find a significant change between anesthesia and awake states. We further calculated the PCC between GS and each brain voxel as per Thompson et al. (2016). We found that anesthesia can increase the correlation in some regions, and these regions overlapped with those in which with the effect of anesthesia on the difference between largest change and smallest change occurred.

Thompson et al. (2016) also observed that increased glucose consumption in brain regions correlated with increased correlation with the GS and increased FCD, but the FCD result was only observed if GSR was not performed. While Dex increases blood glucose in general (Matsumoto et al., 1993), the effect of Dex on cerebral glucose metabolism has had inconsistent effects observed in previous studies, either increased (Nummela et al., 2022) or decreased (Laaksonen et al., 2018) by Dex. Without GSR, we only observed one brain region where FCD significantly changed, being decreased under Dex. However, correlation with the GS was significantly higher in many brain regions under Dex (Fig. 4b). When using data from the same mice as were used in the present study, the cerebral metabolic rate of oxygen consumption was observed to be decreased under Dex (Xu et al., 2022), but the cerebral metabolic rate of glucose consumption was not measured.

Therefore, it is unclear whether FCD or correlation with the GS could be reflecting glucose consumption in our study. Future work such as combined simultaneous positron emission tomography and fMRI to measure aerobic glycolysis (Zhang et al., 2022a) or magnetic resonance spectroscopy (Zhang et al., 2022b; Zou et al., 2022) could potentially answer this question.

Previous studies have found that the default mode network (DMN) had higher correlation with the GS (Cole et al., 2010). We found, similar to their work, that the ACA prelimbic area (part of rat brain's DMN) (Lu et al., 2012; Upadhyay et al., 2011) was also highly correlation with the GS. Moreover, these regions also have high FCD in both awake and anesthetized conditions without GSR. With GSR, the FCD value in Dex group was decreased, and there were more regions that had significant differences between awake and Dex groups. This may indicate that anesthesia increases the GS effect on overall brain activity, while GSR removes this effect, highlighting more local changes in individual brain regions. This is similar to comments by Thompson et al. (2016) about ALFF-based methods under GSR.

Interestingly, we found that FC changes were more sensitive to GSR when the animals were anesthetized. Similar to previous studies, GSR can create or reveal negative FC that was not visible before using the method (Allen et al., 2014; Li et al., 2019; Xu et al., 2018). Previous studies have shown that anesthesia can also reduce FC in mice (Liang et al., 2015; Paasonen et al., 2018; Wu et al., 2016). In addition, our work suggests that GSR has differential effects on FC even in the same mouse under different physiological states. Without GSR, we found that FC decreased significantly under anesthesia compared with awake mice. Data without GSR showed anesthesia reducing FC in mice, but if GSR was used, FC in many brain regions is instead lower in the awake state.

This indicates that GSR has different effects on functional connections in mice under different physiological states and has more impact on functional connections under anesthesia. Thus, when we explore FC changes in fMRI studies using mice under different anesthetic states, if GSR is used, it needs to be considered that it will affect different physiological states differently.

Brain modularity is a tool to quantify how brain networks change in response to tasks or situations (Rubinov and Sporns, 2010). Brain modularity is very sensitive to network changes (Chaddock-Heyman et al., 2021; Newman, 2006). Pullon et al. (2022) observed a decrease in global network efficiency, and the studied brain networks became more segregated under propofol-induced unresponsiveness. Areshenkoff et al. (2021) also found an increase in brain modularity and the number of network nodes under anesthesia. Recent studies demonstrate that anesthesia affects brain networks (Areshenkoff et al., 2021; Pullon et al., 2022), and alters measured modularity scores. Based on our work, we found that GSR also can influence modularity; GSR increased the modularity score both in awake and anesthetized states.

However, the changes in the modularity score caused by GSR were unidirectional under the anesthesia condition and bidirectional under the awake condition. The change in brain modularity in the anesthetized state was more obvious than in the awake state, as brain networks became more segregated in the anesthetized state under GSR.

We divided the mice into “largest” and “smallest” groups based on the amount of modularity change between with and without GSR. Under GSR, within the same anesthesia condition, FCD maps differ significantly between the largest and smallest groups. In mice highly affected by GSR (“largest” group), FCD values within the somatosensory brain network decreased under anesthesia, indicating that the effect of GSR on the brain networks of mice was heterogeneous (Xu et al., 2018). This suggests that perhaps GSR is removing networks with weaker overall connectivity. Although GSR can help find particularly strong networks, researchers are advised to be cautious using GSR to remove noise when planning to perform brain modularity analysis.

Some caveats exist beyond the scope of this work. First, previous studies have demonstrated that FC changes greatly in task condition (Freitas et al., 2020; Rocca et al., 2022). Aguirre et al. (1997) found a significant correlation between the GS and their behavioral paradigm. Task-based fMRI analysis thus needs more caution regarding GSR; whether GSR can remove the signal caused by the task is essential. Second, in this work, we studied the specific effects of GSR on awake and anesthetized states. The extent to which GSR affects different brain networks under different anesthetized states varies. Several studies have found negative correlations between the DMN and task-positive network, which demonstrated the anticorrelations without GSR (Chai et al., 2012; Chang and Glover, 2009). However, the reason for these effects is still unclear (Fox et al., 2009).

The reported results may come from the mathematic characteristics of GSR, underlying neurobiological features and so on. Overall, our work suggests that attention should be focused on the effect of GSR, especially in specific brain regions, as it is used in fMRI studies. To further better use GSR for fMRI data analysis, it is necessary to conduct in-depth research on the reasons for the different effects of GSR.

In conclusion, GSR had a notably different effect on awake and anesthetized mice. The FC and modularity change in anesthetized mice was more remarkable than in awake mice. GSR tended to remove networks that had lower connectivity, and highlight differences in individual brain regions. Considering the difference in the effects of GSR on functional connectivity between anesthesia and wakefulness, caution is advised when using GSR to remove noise, particularly when comparing different neurometabolic states.

Footnotes

Acknowledgment

The authors are thankful for the help of the 9.4 T Small Animal MRI Platform of ShanghaiTech University.

Authors' Contributions

D.W., H.L., and G.J.T. designed and implemented this project. M.X., B.B., M.P., and Z.L. collected and preprocessed the data. D.W. finished the draft of the article, and H.L. and G.J.T. helped to revise it. G.J.T. supervised the analysis of animal data and the writing of the article.

Author Disclosure Statement

No competing financial interests exist.

Funding Information

This study was supported by ShanghaiTech University, the Shanghai Municipal Government, and the National Natural Science Foundation of China Grant 81950410637 (Garth J. Thompson), Grant 3210055 (Hui Li), and Grant 81771821 (Zhifeng Liang), Strategic Priority Research Program of the Chinese Academy of Sciences (Grant No. XDB32030100 to Zhifeng Liang), the Shanghai Municipal Science and Technology Major Project (Grant No. 2018SHZDZX05 to Zhifeng Liang) and CAS Pioneer Hundred Talents Program (to Zhifeng Liang).

Supplementary Material

Supplementary Table S1

Supplementary Table S2

Supplementary Figure S1

Supplementary Figure S2

Supplementary Figure S3

References

Aguirre

, Zarahn

, D'Esposito

. Empirical analyses of BOLD fMRI statistics. II. Spatially smoothed data collected under null-hypothesis and experimental conditions. Neuroimage, 1997; 5(3):199–212; doi: 10.1006/nimg.1997.0264

Allen

, Damaraju

, Plis

, et al. Tracking whole-brain connectivity dynamics in the resting state. Cereb Cortex, 2014; 24(3):663–676; doi: 10.1093/cercor/bhs352

Anderson

, Druzgal

, Lopez-Larson

, et al. Network anticorrelations, global regression, and phase-shifted soft tissue correction. Hum Brain Mapp, 2011; 32(6):919–934; doi: 10.1002/hbm.21079

Areshenkoff

, Nashed

, Hutchison

, et al. Muting, not fragmentation, of functional brain networks under general anesthesia. Neuroimage, 2021; 231:117830; doi: 10.1016/j.neuroimage.2021.117830

Bharioke

, Munz

, Brignall

, et al. General anesthesia globally synchronizes activity selectively in layer 5 cortical pyramidal neurons. Neuron, 2022; 110(12):2024–2040.e2010; doi: 10.1016/j.neuron.2022.03.032

Birn

, Diamond

, Smith

, et al. Separating respiratory-variation-related neuronal-activity-related fluctuations in fluctuations from fMRI. Neuroimage, 2006; 31(4):1536–1548; doi: 10.1016/j.neuroimage.2006.02.048

Biswal

, Yetkin

, Haughton

, et al. Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magn Reson Med, 1995; 34(4):537–541; doi: 10.1002/mrm.1910340409

Burgess

, Kandala

, Nolan

, et al. Evaluation of denoising strategies to address motion-correlated artifacts in resting-state functional magnetic resonance imaging data from the human connectome project. Brain Connect, 2016; 6(9):669–680; doi: 10.1089/brain.2016.0435

Chaddock-Heyman

, Weng

, Loui

, et al. Brain network modularity predicts changes in cortical thickness in children involved in a physical activity intervention. Psychophysiology, 2021; 58(10):e13890; doi: 10.1111/psyp.13890

10.

Chai

XQJ

, Castanon

, Ongur

, et al. Anticorrelations in resting state networks without global signal regression. Neuroimage, 2012; 59(2):1420–1428; doi: 10.1016/j.neuroimage.2011.08.048

11.

Chang

, Glover

. Effects of model-based physiological noise correction on default mode network anti-correlations and correlations. Neuroimage, 2009; 47(4):1448–1459; doi: 10.1016/j.neuroimage.2009.05.012

12.

Chang

, Leopold

, Scholvinck

, et al. Tracking brain arousal fluctuations with fMRI. Proc Natl Acad Sci U S A, 2016; 113(16):4518–4523; doi: 10.1073/pnas.1520613113

13.

Ciric

, Wolf

, Power

, et al. Benchmarking of participant-level confound regression strategies for the control of motion artifact in studies of functional connectivity. Neuroimage, 2017; 154:174–187; doi: 10.1016/j.neuroimage.2017.03.020

14.

Cohen

, Tomasi

, Shokri-Kojori

, et al. Functional connectivity density mapping: comparing multiband and conventional EPI protocols. Brain Imaging Behav, 2018; 12(3):848–859; doi: 10.1007/s11682-017-9742-7

15.

Cole

, Pathak

, Schneider

. Identifying the brain's most globally connected regions. Neuroimage, 2010; 49(4):3132–3148; doi: 10.1016/j.neuroimage.2009.11.001

16.

de Arce

, Ribic

, Chowdhury

, et al. Concerted roles of LRRTM1 and SynCAM 1 in organizing prefrontal cortex synapses and cognitive functions. Nat Commun, 2023; 14(1):459; doi: 10.1038/s41467-023-36042-w

17.

Dinh

TNA

, Jung

, Shim

, et al. Characteristics of fMRI responses to visual stimulation in anesthetized vs. awake mice. Neuroimage, 2021; 226:117542; doi: 10.1016/j.neuroimage.2020.117542

18.

Evans

, Kundu

, Horovitz

, et al. Separating slow BOLD from non-BOLD baseline drifts using multi-echo fMRI. Neuroimage, 2015; 105:189–197; doi: 10.1016/j.neuroimage.2014.10.051

19.

Fox

, Zhang

, Snyder

, et al. The global signal and observed anticorrelated resting state brain networks. J Neurophysiol, 2009; 101(6):3270–3283; doi: 10.1152/jn.90777.2008

20.

Freitas

LGA

, Bolton

TAW

, Krikler

, et al. Time-resolved effective connectivity in task fMRI: Psychophysiological interactions of co-activation patterns. Neuroimage, 2020; 212:116635; doi: 10.1016/j.neuroimage.2020.116635

21.

Gao

, Ma

, Zhang

, et al. Time to wake up: Studying neurovascular coupling and brain-wide circuit function in the un-anesthetized animal. Neuroimage, 2017; 153:382–398; doi: 10.1016/j.neuroimage.2016.11.069

22.

Grandjean

, Schroeter

, Batata

, et al. Optimization of anesthesia protocol for resting-state fMRI in mice based on differential effects of anesthetics on functional connectivity patterns. Neuroimage, 2014; 102(Pt 2):838–847; doi: 10.1016/j.neuroimage.2014.08.043

23.

Guerra-Carrillo

, Mackey

, Bunge

. Resting-state fMRI: A window into human brain plasticity. Neuroscientist, 2014; 20(5):522–533; doi: 10.1177/1073858414524442

24.

Hamilton

, Ma

, Zhang

. Global reduction of information exchange during anesthetic-induced unconsciousness. Brain Struct Funct, 2017; 222(7):3205–3216; doi: 10.1007/s00429-017-1396-0

25.

Han

, Chen

, et al. Awake and behaving mouse fMRI during Go/No-Go task. Neuroimage, 2019; 188:733–742; doi: 10.1016/j.neuroimage.2019.01.002

26.

Harrewijn

, Abend

, Linke

, et al. Combining fMRI during resting state and an attention bias task in children. Neuroimage, 2020; 205:116301; doi: 10.1016/j.neuroimage.2019.116301

27.

Laaksonen

, Kallioinen

, Langsjo

, et al. Comparative effects of dexmedetomidine, propofol, sevoflurane, and S-ketamine on regional cerebral glucose metabolism in humans: A positron emission tomography study. Br J Anaesth, 2018; 121(1):281–290; doi: 10.1016/j.bja.2018.04.008

28.

, Kong

, Liegeois

, et al. Global signal regression strengthens association between resting-state functional connectivity and behavior. Neuroimage, 2019; 196:126–141; doi: 10.1016/j.neuroimage.2019.04.016

29.

Liang

, Liu

, Zhang

. Dynamic resting state functional connectivity in awake and anesthetized rodents. Neuroimage, 2015; 104:89–99; doi: 10.1016/j.neuroimage.2014.10.013

30.

Liu

, Nalci

, Falahpour

. The global signal in fMRI: Nuisance or information?. Neuroimage, 2017; 150:213–229; doi: 10.1016/j.neuroimage.2017.02.036

31.

, Zou

, Gu

, et al. Rat brains also have a default mode network. Proc Natl Acad Sci U S A, 2012; 109(10):3979–3984; doi: 10.1073/pnas.1200506109

32.

Lukasik

, Gillies

. Animal anaesthesia for in vivo magnetic resonance. NMR Biomed, 2003; 16(8):459–467; doi: 10.1002/nbm.836

33.

Martin

, He

, Chang

. State-related neural influences on fMRI connectivity estimation. Neuroimage, 2021; 244:118590; doi: 10.1016/j.neuroimage.2021.118590

34.

Maslov

, Sneppen

. Specificity and stability in topology of protein networks. Science, 2002; 296(5569):910–913; doi: 10.1126/science.1065103

35.

Matsumoto

, Zornow

, Rabin

, et al. The alpha(2) adrenergic agonist, dexmedetomidine, selectively attenuates ischemia-induced increases in striatal norepinephrine concentrations. Brain Res, 1993; 627(2):325–329; doi: Doi10.1016/0006-8993(93)90337-M

36.

Murphy

, Birn

, Handwerker

, et al. The impact of global signal regression on resting state correlations: Are anti-correlated networks introduced?. Neuroimage, 2009; 44(3):893–905; doi: 10.1016/j.neuroimage.2008.09.036

37.

Murphy

, Fox

. Towards a consensus regarding global signal regression for resting state functional connectivity MRI. Neuroimage, 2017; 154:169–173; doi: 10.1016/j.neuroimage.2016.11.052

38.

Newman

. Modularity and community structure in networks. Proc Natl Acad Sci U S A, 2006; 103(23):8577–8582; doi: 10.1073/pnas.0601602103

39.

Noble

, Scheinost

, Constable

. A guide to the measurement and interpretation of fMRI test-retest reliability. Curr Opin Behav Sci, 2021; 40:27–32; doi: 10.1016/j.cobeha.2020.12.012

40.

Nummela

, Laaksonen

, Laitio

, et al. Effects of dexmedetomidine, propofol, sevoflurane and S-ketamine on the human metabolome A randomised trial using nuclear magnetic resonance spectroscopy. Eur J Anaesth, 2022; 39(6):521–532; doi: 10.1097/Eja.0000000000001591

41.

Paasonen

, Stenroos

, Salo

, et al. Functional connectivity under six anesthesia protocols and the awake condition in rat brain. Neuroimage, 2018; 172:9–20; doi: 10.1016/j.neuroimage.2018.01.014

42.

Parkes

, Fulcher

, Yucel

, et al. An evaluation of the efficacy, reliability, and sensitivity of motion correction strategies for resting-state functional MRI. Neuroimage, 2018; 171:415–436; doi: 10.1016/j.neuroimage.2017.12.073

43.

Power

, Mitra

, Laumann

, et al. Methods to detect, characterize, and remove motion artifact in resting state fMRI. Neuroimage, 2014; 84:320–341; doi: 10.1016/j.neuroimage.2013.08.048

44.

Power

, Plitt

, Gotts

, et al. Ridding fMRI data of motion-related influences: Removal of signals with distinct spatial and physical bases in multiecho data. Proc Natl Acad Sci U S A, 2018; 115(9):E2105–E2114; doi: 10.1073/pnas.1720985115

45.

Power

, Plitt

, Laumann

, et al. Sources and implications of whole-brain fMRI signals in humans. Neuroimage, 2017; 146:609–625; doi: 10.1016/j.neuroimage.2016.09.038

46.

Pullon

, Warnaby

, Sleigh

. Propofol-induced unresponsiveness is associated with a brain network phase transition. Anesthesiology, 2022; 136(3):420–433; doi: 10.1097/ALN.0000000000004095

47.

Rocca

, Schoonheim

, Valsasina

, et al. Task- and resting-state fMRI studies in multiple sclerosis: From regions to systems and time-varying analysis. Current status and future perspective. Neuroimage Clin, 2022; 35:103076; doi: 10.1016/j.nicl.2022.103076

48.

Rubinov

, Sporns

. Complex network measures of brain connectivity: Uses and interpretations. Neuroimage, 2010; 52(3):1059–1069; doi: 10.1016/j.neuroimage.2009.10.003

49.

Schifani

, Hawco

, Nazeri

, et al. Relationship of neurite architecture to brain activity during task-based fMRI. Neuroimage, 2022; 262:119575; doi: 10.1016/j.neuroimage.2022.119575

50.

Schroeter

, Grandjean

, Schlegel

, et al. Contributions of structural connectivity and cerebrovascular parameters to functional magnetic resonance imaging signals in mice at rest and during sensory paw stimulation. J Cerebr Blood F Met, 2017; 37(7):2368–2382; doi: 10.1177/0271678x16666292

51.

Shmueli

, van Gelderen

, de Zwart

, et al. Low-frequency fluctuations in the cardiac rate as a source of variance in the resting-state fMRI BOLD signal. Neuroimage, 2007; 38(2):306–320; doi: 10.1016/j.neuroimage.2007.07.037

52.

Sitsen

, Khalili-Mahani

, de Rover

, et al. Effect of spinal anesthesia-induced deafferentation on pain processing in healthy male volunteers: A task-related fMRI study. Front Pain Res (Lausanne), 2022; 3:1001148; doi: 10.3389/fpain.2022.1001148

53.

Smitha

, Akhil Raja

, Arun

, et al. Resting state fMRI: A review on methods in resting state connectivity analysis and resting state networks. Neuroradiol J, 2017; 30(4):305–317; doi: 10.1177/1971400917697342

54.

Thompson

, Riedl

, Grimmer

, et al. The whole-brain “Global” signal from resting state fMRI as a potential biomarker of quantitative state changes in glucose metabolism. Brain Connect, 2016; 6(6):435–447; doi: 10.1089/brain.2015.0394

55.

Tomasi

, Volkow

. Functional connectivity density mapping. Proc Natl Acad Sci U S A, 2010; 107(21):9885–9890; doi: 10.1073/pnas.1001414107

56.

Tomasi

, Volkow

. Functional connectivity hubs in the human brain. Neuroimage, 2011; 57(3):908–917; doi: 10.1016/j.neuroimage.2011.05.024

57.

Uddin

. Mixed signals: On separating brain signal from noise. Trends Cogn Sci, 2017; 21(6):405–406; doi: 10.1016/j.tics.2017.04.002

58.

Upadhyay

, Baker

, Chandran

, et al. Default-mode-like network activation in awake rodents. PLoS One, 2011; 6(11):e0027839; doi: 10.1371/journal.pone.0027839

59.

Wong

, Olafsson

, Tal

, et al. Anti-correlated networks, global signal regression, and the effects of caffeine in resting-state functional MRI. Neuroimage, 2012; 63(1):356–364; doi: 10.1016/j.neuroimage.2012.06.035

60.

, Mishra

, Wang

, et al. Effects of isoflurane anesthesia on resting-state fMRI signals and functional connectivity within primary somatosensory cortex of monkeys. Brain Behav, 2016; 6(12):e00591; doi: 10.1002/brb3.591

61.

Xie

, Chen

, Ma

, et al. Brain connectivity analysis in anesthetized and awake states: An ECoG study in monkeys. Annu Int Conf IEEE Eng Med Biol Soc, 2021; 2021:117–120; doi: 10.1109/EMBC46164.2021.9631095

62.

, Su

, Qin

, et al. Impact of global signal regression on characterizing dynamic functional connectivity and brain states. Neuroimage, 2018; 173:127–145; doi: 10.1016/j.neuroimage.2018.02.036

63.

, Bo

, Pei

, et al. High-resolution relaxometry-based calibrated fMRI in murine brain: Metabolic differences between awake and anesthetized states. J Cereb Blood Flow Metab, 2022; 42(5):811–825; doi: 10.1177/0271678X211062279

64.

Yang

, Murray

, Repovs

, et al. Altered global brain signal in schizophrenia. Proc Natl Acad Sci U S A, 2014; 111(20):7438–7443; doi: 10.1073/pnas.1405289111

65.

Yatziv

, Yudco

, Dickmann

, et al. Patterns of neural activity in the mouse brain: Wakefulness vs. general anesthesia. Neurosci Lett, 2020; 735:135212; doi: 10.1016/j.neulet.2020.135212

66.

Zarahn

, Aguirre

, D'Esposito

. Empirical analyses of BOLD fMRI statistics. Neuroimage, 1997; 5(3):179–197; doi: 10.1006/nimg.1997.0263

67.

Zerbi

, Grandjean

, Rudin

, et al. Mapping the mouse brain with rs-fMRI: An optimized pipeline for functional network identification. Neuroimage, 2015; 123:11–21; doi: 10.1016/j.neuroimage.2015.07.090

68.

Zhang

, Qin

, Zhang

, et al. Aerobic glycolysis imaging of epileptic foci during the inter-ictal period. EBioMedicine, 2022a;79; doi: 10.1016/j.ebiom.2022.104004

69.

Zhang

, Gao

, Fang

, et al. Proton/deuterium magnetic resonance imaging of rodents at 9.4T using birdcage coils. Bioelectromagnetics, 2022b;43(1):40–46; doi: 10.1002/bem.22382

70.

Zou

, Ruan

, Li

, et al. A new deuterium-labeled compound [2,3,4,6,6’-2H5]-D-glucose for deuterium magnetic resonance metabolic imaging. NMR Biomed, 2022; 36(7):e4890; doi: 10.1002/nbm.4890

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.01 MB

1.13 MB

0.12 MB

0.27 MB