High-resolution relaxometry-based calibrated fMRI in murine brain: Metabolic differences between awake and anesthetized states

Abstract

Functional magnetic resonance imaging (fMRI) techniques using the blood-oxygen level-dependent (BOLD) signal have shown great potential as clinical biomarkers of disease. Thus, using these techniques in preclinical rodent models is an urgent need. Calibrated fMRI is a promising technique that can provide high-resolution mapping of cerebral oxygen metabolism (CMR_O2). However, calibrated fMRI is difficult to use in rodent models for several reasons: rodents are anesthetized, stimulation-induced changes are small, and gas challenges induce noisy CMR_O2 predictions. We used, in mice, a relaxometry-based calibrated fMRI method which uses cerebral blood flow (CBF) and the BOLD-sensitive magnetic relaxation component, R₂′, the same parameter derived in the deoxyhemoglobin-dilution model of calibrated fMRI. This method does not use any gas challenges, which we tested on mice in both awake and anesthetized states. As anesthesia induces a whole-brain change, our protocol allowed us to overcome the former limitations of rodent studies using calibrated fMRI. We revealed 1.5-2 times higher CMR_O2, dependent upon brain region, in the awake state versus the anesthetized state. Our results agree with alternative measurements of whole-brain CMR_O2 in the same mice and previous human anesthesia studies. The use of calibrated fMRI in rodents has much potential for preclinical fMRI.

Introduction

Functional magnetic resonance imaging (fMRI) is a key tool in understanding brain function. In particular, the commonly-used blood oxygenation level-dependent (BOLD) signal, can reveal both brain regions activated by tasks or stimuli, and communication among networks at rest. ¹ Such networks, in particular, have been linked to a wide variety of neurological and psychiatric diseases, strongly suggesting the possibility of clinical translation for diagnosis or treatment planning. ²

Complicating translation, however, is that the BOLD signal depends on the deoxyhemoglobin concentration ([dHb]) in a voxel which itself depends on complex interactions between cerebral blood flow (CBF), cerebral blood volume (CBV), and cerebral metabolic rate of oxygen (CMR_O2).^3,4 One way towards better understanding BOLD signal changes is to use rodent models so that genetic alterations and invasive measurements can be used. ⁵ In particular, it is important that methods which increase the number of metabolic parameters that MRI can measure, such as methods to separate CBF, CBV, or CMR_O2 from the BOLD signal, can be translated to rodent models.

As neurons and glial cells require CMR_O2 commensurate with their activity level^6–9 CMR_O2 tracks functional brain activity and thus quantitative measurement of CMR_O2 is a promising direction. Davis et al. and Hoge et al introduced a method designed to disassociate changes in CMR_O2 from changes in CBF (or CBV) and BOLD signal, called “calibrated fMRI”.^10,11 Currently, most calibrated fMRI methods involve “gas challenges” to induce changes in blood levels of O₂ ¹² and CO_2. ¹³ The gas challenge step induces either hyperoxia (inhalation of O₂) or hypercapnia (inhalation of CO₂). The latter is more commonly used since CO₂ is a potent vasodilator and thus can be used to estimate the maximum possible BOLD signal change, denoted as M.^10,11

Because of the promise of calibrated fMRI, its implementation in preclinical rodent models is very important. However, several barriers exist:

1. In small animals the measurement of M by gas challenge is extremely noisy because of the small range of BOLD and CBF data points that are used to derive the equation parameters. To address this, prior work by Kida et al. and by Shu et al. indicated that estimation of M without a gas challenge was less noisy.^14,15 Several other methods can also estimate M without a gas challenge.^16–18

2. Calibrated fMRI studies are often based on local sensory stimulation paradigms. ¹⁹ The changes observed in local sensory stimulation paradigms are small, both in terms of activation magnitude and spatial size. Thus they are more difficult to statistically locate on the small brains of rodents than in they are in humans (even at high magnetic fields). To address this, an alternative is to use an anesthesia paradigm to alter global brain activity. ²⁰ This is similar logic to previous validation of CMR_O2 changes from calibrated fMRI vs. CMR_O2 changes from Carbon-13 Magnetic Resonance Spectroscopy. ¹⁴ The anesthetic dexmedetomidine has been used previously as a global paradigm in human CMR_O2 ²¹ and CBF ²² studies, and can also be used in mouse fMRI. ²³

3. The vast majority of human fMRI studies (including calibrated fMRI) have used awake subjects, whereas the vast majority of animal fMRI studies have used anesthesia. ⁵ To bridge this gap, our previous work has developed an awake mouse fMRI model ²⁴ and this was utilized in the current study.

4. These problems may compound. For example, Sicard et al. suggested that the neural response to a stimulus can be reduced by an anesthesia, and also the ability to measure it simultaneously reduced by loss of vascular reactivity due to a gas challenge. ²⁵

We used an extension of the calibrated fMRI method developed by Shu et al. ¹⁵ Shu et al. began this work in anesthetized rats ¹⁵ but (to our knowledge) this method has not been published on mice thus far. The derived M parameter (or specifically, the relaxation rate described below) from this method is no different than the M parameter derived from Hoge et al.’s deoxyhemoglobin-dilution model of CMR_O2. ¹¹ It is based on the fact that the BOLD signal is captured differentially by two different transverse magnetic relaxation rates, R₂* which includes magnetic susceptibility components that can be reversed, and components that cannot be reversed, and R₂ which includes only magnetic susceptibility components that cannot be reversed. The M parameter can be modelled as the product of the fMRI echo time (TE) and R₂′ (the BOLD-based magnetic susceptibility component that can be reversed, given by the difference between R₂* and R₂ under well-shimmed conditions). ¹⁸ We combine the R₂′ measurement with a measurement of CBF to calculate a relative CMR_O2 map (rCMR_O2) between states, which we refer to as relaxometry-based calibrated fMRI (rcfMRI). We applied this rcfMRI to the same mice in the awake ²⁴ and dexmedetomidine anesthetized ²³ states. We also assessed the power-law dependence between BOLD signal and [dHb], modelled as the exponent β. ²⁶ Together, quantitative mapping of CBF and the M and β parameters allowed estimation of brain-wide rCMR_O2 changes induced by anesthesia from the awake state.

The map of rCMR_O2 between states displayed globally higher CMR_O2 in awake versus anesthetized mice. Interestingly, we were also able to observe some regional differences in the global increase, including a cortical versus subcortical difference previously observed (using other techniques) in humans under dexmedetomidine. ²² Our rCMR_O2 result from rcfMRI was corroborated by a global CMR_O2 change measured using an established method, T₂ relaxation under spin-tagging (TRUST) MRI,^27,28 an idea similar to the cortex-wide validation previously performed by Kida et al. ¹⁴

The ability to apply rcfMRI in a preclinical model is critical for the translation of calibrated fMRI as a technique, so that it can be used for preclinical objectives such as assessing drug effects or testing mouse disease models. We have mapped rCMR_O2 on a voxel-by-voxel basis in awake mice, opening new avenues for preclinical studies.

Materials and methods

Summary of the overall method for rcfMRI

The overall method for rcfMRI is shown in Figure 1. Data were recorded in two states using the same sequences.

Figure 1. Summary of the overall method for rcfMRI. Data were recorded in the awake and anesthetized states using the same MRI sequences. The relaxation time map R₂′ was calculated based on R₂ and R₂*, and combined with CBF in both states to calculate rCMR_O2 in the awake divided by the anesthetized state. rCMR_O2 was calculated using a range of values for the α constant, and a per-voxel map of the β constant which was measured experimentally. Global CMR_O2 (gCMR_O2) was calculated from this method (rCMR_O2_M), and also from TRUST MRI sequences by converting the T₂ measured with TRUST to venous oxygen saturation Y_v (rCMR_O2_T).

Theory

In this section, we describe the derivation of the equation we use to calculate rCMR_O2. This is an expansion of previous work by Kida et al. ¹⁴ which showed the basis of relaxometry-based calibrated fMRI method applied in rats.^15,29 We begin with the original model of the relationship between magnetic relaxation and deoxyhemoglobin concentration, shown in equation (E1),^11,30

$$R_2^{\mathrm{\prime }}=A\cdot \mathit{CBV}\cdot {\left[\mathit{dHb}\right]}^{\beta }$$ (E1)

where R₂′ is the (positive) increase in magnetic relaxation rate due to deoxyhemoglobin,^4,15,31 A is a field-dependent magnetic susceptibility constant, ³⁰ CBV is the blood volume, and β is the exponent in the power-law dependence between BOLD signal and the deoxyhemoglobin concentration in blood ([dHb]) which depends on blood vessel size. ¹⁰ β and R₂′ vary across different brain regions. ²⁶ Ignoring all non-susceptibility terms that contribute to transverse relaxation, blood oxygenation (Y) induced intravoxel spin dephasing is reflected differently by gradient-echo (R₂*) and spin-echo (R₂) relaxation rates

$$R_2^{\mathrm{*}}=R_2^{\mathrm{\prime }}\left(Y\right)+R_2^{}\left(Y\right)$$ (E2a)

$$R_2^{}=R_2^{}\left(Y\right)$$ (E2b)

because R₂* contains both reversible (e.g., static magnetic fields inhomogeneity, slow diffusion regime) and irreversible (i.e., intermediate to fast diffusion regime) terms and R₂ contains just the irreversible term.¹⁴ Under well-shimmed conditions, we assume that the difference between R₂ and R₂*, R₂′, is greater than zero primarily due to deoxyhemoglobin, the BOLD-sensitive component.

$$R_2^{\mathrm{\prime }}=R_2^{\mathrm{*}}-R_2^{}$$ (E2c)

For the purposes of our model, we only consider the deoxyhemoglobin-sensitive part of changes in R₂′. However, under poorly-shimmed conditions this may not hold as R₂′ could be dominated by effects of the static magnetic field, B₀. The right side of E3a is taken from E1 with CBV being replaced by CBF to an exponential power. This is done using Grubb’s constant α. ³² Note that this assumes that CBF follows CBV changes and vice-versa with a consistent functional relationship, not that either is necessarily passive. Finally, steady state deoxyhemoglobin concentration in the venous compartment is replaced with CBF and CMR_O2. ¹¹

$$R_2^{\mathrm{\prime }}=A\cdot {\left[\mathit{CBF}\right]}^{\alpha }\cdot \mathrm{ }{\left[\frac{1}{4}\frac{CM{R}_{O2}}{\mathit{CBF}}\right]}^{\beta }$$ (E3a)

$$CM{R}_{O2}={4\cdot \left(\frac{R_2^{\mathrm{\prime }}}{A}\right)}^{\frac{1}{\beta }}\cdot (CB{F)}^{1-\frac{\alpha }{\beta }}$$ (E3b)

Previous work has shown that β is invariant across brain states ²⁶ and the constant α is defined as a shift between two brain states. ¹¹ Thus, defining CMR_O2, R₂′, and CBF in two different states, represented by subscripts P and Q, we can derive equation (E4),

$${\mathit{rCMRO}}_2=\frac{{\mathit{CMRO}}_{2,P}}{{\mathit{CMRO}}_{2,Q}}={\left(\frac{{R2}_P^{\mathrm{\prime }}}{{R2}_Q^{\mathrm{\prime }}}\right)}^{\frac{1}{\beta }}\cdot {\left(\frac{{\mathit{CBF}}_P}{{\mathit{CBF}}_Q}\right)}^{1-\frac{\alpha }{\beta }}$$ (E4)

where A, the only unmeasured field-dependent constant, cancels out. Thus we can model rCMR_O2 based on only R₂′ and CBF in two brain states, with only the constants α and β remaining, both of which are potentially measurable. For further information on α and β, see “Constants of the metabolic model” in supplemental information.

Animals

All animal procedures were approved by the Animal Care and Use Committee of the Institute of Neuroscience, Chinese Academy of Sciences, Shanghai, China (A), following “Laboratory animal – Guideline for ethical review of animal welfare” published by the People’s Republic of China.^33,34 Animal data reporting herein follows the ARRIVE 2.0 guidelines. ³⁵

15 male adult C57BL/6 mice (Shanghai Laboratory Animal Center, Shanghai, China) of 7–9 weeks of age with weight 20–30 g were imaged in both awake and anesthetized states. Mice underwent awake imaging preparations identical to our previous studies.^24,36 Briefly, a head holder was attached to each animal’s skull for head fixation, and animals were allowed to recover for one week after the surgery. Afterwards, animals were habituated for awake imaging over one week. The same mice were first imaged in the awake condition, and then again approximately one week later in the anesthetized condition. This order was maintained as we were concerned that dexmedetomidine exposure and recovery in the same setup as awake imaging would induce trauma if awake imaging was done later. Mice were labeled by cropped ear to match the same mouse between conditions. Initial examination of 4 mice indicated a strong effect in every mouse tested, but n = 15 was chosen to avoid under-sampling. 22 mice were tested in total. 5 mice died prior to completion of the experiment and were excluded, and 2 mice had physiological parameters outside of the acceptable range, so were also excluded; these criteria were established a priori. Mice were euthanized following the second (dexmedetomidine) scanning session.

Image acquisition

For both awake and anesthetized imaging sessions, mice were given 20% O₂/80% air mixture (“medical air” with total ∼37% O₂ concentration), and respiration was monitored through a pneumatic pillow (SAII, Stony Brook, NY, USA). For anesthetized imaging sessions, mice were initially anesthetized using isoflurane (3.5% for induction, 1.5% during set-up) and a bolus of 0.025 mg/kg dexmedetomidine was given by intraperitonial injection. Isoflurane was discontinued after 5 min. Continuous subcutaneous infusion of dexmedetomidine at a dose of 0.05 mg/kg/h was started 10 min after the bolus injection. While under anesthesia, body temperature was monitored through a rectal probe and maintained at 37.0 ± 0.5 °C using a warm water pad. Respiratory rate was between 262–293 breaths per minute while awake and between 142–158 breaths per minute while anesthetized.

MRI data were acquired with a Bruker BioSpec 9.4 T scanner, using an 86-mm volume coil (Bruker) for transmit and a 4-channel phased-array cryogenic mouse head coil (Bruker) for receive. After a Localizer scan for animal positioning, T₂ RARE anatomical images were acquired (TE 33 ms, TR 3300 ms, FOV 16 × 16 mm, matrix size 256 × 256, slice thickness 0.5 mm, number of slices 20, averages 2). The scanner adjustment took >3 min and the localizer and T2 RARE took >7 min. When combined with setup time, and time taken for physiology to stabilize, at least 30 minutes passed between the bolus injection and when calibrated fMRI scans began. As previous studies measuring CMR_O2 with ¹⁵O positron-emission tomography (PET) techniques have assumed CBF is relatively stable for at least 15 min to complete scanning,^37,38 and approximately 15–30 minutes is sufficient for a steady state of the vasculature after administering dexmedetomidine, ³⁹ we assume a relatively steady state under our experimental protocol.

The relaxometry-based calibrated fMRI (rcfMRI) experiment consisted of measurement of CBF, a T₂ map, and a T₂* map. A pseudo-continuous arterial spin labeling (pCASL) sequence, ⁴⁰ including label and control prescans to adjust the phase, was used to measure CBF (TE 22 ms, TR 4000 ms, matrix size 71 × 71, in-plane resolution 0.225 × 0.225 mm, slice thickness 0.5 mm, 20 slices, label duration 3 s, post labeling delay 300 ms, 80 pairs of label/control volumes). All pCASL sequences used 0.59 inversion efficiency, which was measured on the MRI prior to beginning this experiment as per Hirschler et al. ⁴⁰ After field-map-based local shimming within the mouse brain (Figure S1), T₂* and T₂ maps were acquired with MGE (Multiple Gradient Echo, TR 500 ms, TE 2.15-17.23 ms, echoes 8, averages 4, slice thickness 0.5 mm, number of slices 20, FOV 22 × 16 mm, matrix size 110 × 80) and MSME (Multi Slice Multi Echo, TR 3000 ms, TE 10-100ms, echoes 10, averages 1, slice thickness 0.5 mm, number of slices 20, FOV 22 × 16 mm, matrix size 110 × 80) sequences. Finally, a mouse-optimized T₂-relaxation-under-spin-tagging (TRUST) sequence ²⁸ was used to assess the T₂ of venous blood (TE 5.9 ms, TR 5000 ms, inversion slab thickness 1.5 mm, PLD 600 ms, FOV 16 mm × 16 mm, matrix size 128 × 128, slice thickness 0.5 mm, eTE values: 0.27, 7.5, 15 ms).

A separate, anesthetized, cohort of four mice without head holder implantation were used for quantifying β maps based on a previously reported method. ²⁶ Mice were anesthetized using isoflurane followed by dexmedetomidine identical to the main cohort’s anesthetized state (described above). Mice received three intravenous bolus injections of superparamagnetic contrast agent Molday (BioPAL Inc., Worcester, MA, USA, incremental dose of 5 mg/kg). Prior to injecting contrast agents and following each injection, T₂* and T₂ maps were acquired respectively with the same parameters as within the main cohort (described above). This separate cohort of mice was euthanized following scanning. Shu et al. discovered that β maps were consistent between two very different anesthetics which both have completely different mechanisms of action and also induce very different levels of sedation. ²⁶ In addition, the β mapping experiment would have been very difficult to perform in awake mice due to its invasive nature. Thus, we assumed stable β maps between the awake and anesthetized conditions.

Data processing

Data were preprocessed using custom-written scripts in MATLAB and SPM12 (http://www.fil.ion.ucl.ac.uk/spm/). After the pCASL images were converted from Bruker format to NIFTI format, the mouse brain was extracted manually using ITK-SNAP (http://www.itksnap.org/), realigned for motion correction, normalized to a mouse brain template ⁷⁴ (http://imaging.org.au/AMBMC/Model), and spatially smoothed (0.4 mm isotropic Gaussian Kernel) for quantifying CBF. ⁴¹

The T₂* and T₂ maps were calculated using the normalized magnitude multi-echo MGE and MSME data with a mono-exponential fit by using the Levenberg-Marquardt algorithm. R₂′ images were calculated using E2. ¹⁴

To test data normality, we used Kolmogorov-Smirnov (KS) tests. The Lilliefors adjustment was used where N ≤ 15. Most results are presented per-voxel (each voxel averaged across all mice) to follow previous studies of CMR_O2 in humans. ⁴² In addition to per-voxel results, per-mouse results were also calculated. In per-mouse results, to test statistical significance in maps of CBF and R₂, the mean value of the map for each mouse was taken in 15 different brain regions (see Table S1). A two-tailed, paired T-test was done for each region, counting each mouse as a sample and pairing awake vs. anesthetized in that mouse. Resulting p values were corrected for multiple comparisons using Sequential Goodness of Fit ⁴³ at p ≤ 0.05, considering each brain map as a statistical family with 15 tests within it.

β maps were calculated using a previously published method ²⁶ in MATLAB. By measuring R₂′ as a function of the contrast agent concentration, β was experimentally determined on a voxel-by-voxel basis.

A per-voxel rcfMRI was calculated from E4 using maps of R₂′ from each state, maps of CBF from each state, α = 0.2, and the β map calculated from a separate cohort of mice. All results were normalized to the template described above. For further analysis, after registration, only slices 5-16 were selected due to image distortion induced by the head holder. A manual mask was applied to the final result by using the roi_poly function on each slice in MATLAB to exclude any artifacts along the edge of the brain. This helped maintain assumptions of our model by avoiding the higher inhomogeneity of the static magnetic field (B₀) at the edge of the brain. In addition, only the 5th through 16th slices were used for analysis, to avoid possible B₀ inhomogeneity at the far rostral and caudal extents of the brain.

For TRUST data, subtraction between the control and label images was performed. ²⁸ A region of interest (ROI) was manually drawn on the difference image to encompass the sinus confluence at the superior sagittal sinus, then a mono-exponential fitting of the difference signal yielded a T_2v (venous T₂) estimation. A student’s t-test was performed to compare T_2v under different conditions with a statistical threshold of P < 0.05.

“Global” rCMR_O2 was calculated for both rcfMRI results and TRUST results, referred to respectively as rCMR_O2_M and rCMR_O2_T. To calculate rCMR_O2_M, we used the relaxometry-based calibrated fMRI map (as described above, calculated on a per-voxel basis), and took the average from a conservative estimate of voxels known to be within the watershed area of the superior sagittal sinus based on existing atlases (Figure S2),^44,45 from the mouse brain template. (This area was chosen to match the TRUST results as closely as possible.) To calculate rCMR_O2_T, we used the T₂ values from TRUST, as described above, converted it to oxygen content of venous blood (Y_v) as per Lu et al, ²⁷ then, to match methods described elsewhere, ¹⁹ we used the CBF data averaged from the watershed area (Figure S2) from the pCASL sequence described above to convert it to rCMR_O2, based on the Fick principle. (A full derivation is given in Supplemental Information.) The watershed areas were used instead of the whole brain as we hypothesized these areas had the greatest effect on TRUST results.

Mice with values for either rCMR_O2_T or rCMR_O2_M greater than 2 standard deviations from the respective group mean were removed. (Note: this was done only for the comparison analysis.) This resulted in removal of one mouse, leaving N = 14. The comparison analysis included both a test based on linear similarity, Pearson correlation, and a test which can also find non-linear similarities, mutual information (base 2 for logarithms, 5 bins of values), to compare rCMR_O2_T and rCMR_O2_M. Statistical significance for Pearson correlation was calculated using the permutation test from the built-in corr function in MATLAB. Statistical significance for mutual information was calculated by randomly permuting which mouse’s rCMR_O2_T corresponded to which mouse’s rCMR_O2_M and calculating mutual information between them for 100,000 random permutations. A one-tailed p value was calculated using the distribution of the randomized data to calculate a percentile, assuming the mutual information of the actual data was greater than the randomized data.

Results

Measurement of R₂′ and CBF

Brain maps of the parameters measured in our experiment are displayed in Figure 2. We measured relaxation times R₂ (1/T₂) (Figure 2(b)) and R₂* (1/T₂*) (Figure 2(c)). R₂′ (Figure 2(d)) was calculated as per E2. CBF was measured using pCASL (Figure 2(e)). R₂ and CBF maps showed a statistically significant difference in some brain regions (Table S2) with shorter R₂ (indicating longer T₂) and higher CBF in the awake condition versus the anesthetized condition. R₂* and R₂′ maps did not show a statistically significant difference between awake and anesthetized.

Figure 2.

Comparison of R₂, R₂*, R₂′ and CBF maps in awake and anesthetized states. Average brain maps of measured parameters. A. The 5th to 16th coronal slices from co-registered anatomical data. In B-E Mean values from awake and anesthetized groups (N = 15 mice) are shown on the top and bottom of each map respectively. B. R₂ relaxation time indicating transverse relaxation with refocused inhomogeneities. C. R₂* relaxation time with non-refocused inhomogeneities. D. R₂′ relaxation time of only the inhomogeneities (see E2) E. Cerebral blood flow (CBF) generated using pCASL. Units are s⁻¹ for relaxation maps, mL/100g/min for CBF.

We compared our mouse CBF results to a previous study of CBF in awake versus dexmedetomidine-anesthetized human subjects which used PET. ²² Similar to humans, mice show a greater change from awake to anesthetized in the subcortex than the cortex, with the ratio being very similar between the two species (Table S3).

α and β constants

The α constant, representing the exponential power relationship between CBV and CBF, was primarily set to 0.2 to follow the original literature in human subjects,^46,47 measurements of the non-activated state in rats,^46,48 and finally because this later gave a good correspondence between rcfMRI and TRUST (see “Comparison of rCMR_O2 from rcfMRI and rCMRO2 from TRUST” below). However, as α has not been directly measured in mice, we also tested the full range of 0.1–0.4 ²⁹ in increments of 0.1 (see “Effect of varying of α on rcfMRI” below).

The β constant, representing the non-linear susceptibility change due to deoxyhemoglobin, was measured using a previously published method, ²⁶ which requires multiple injections of the contrast agent Molday. Figure 3(a), first four rows, shows the R₂′ maps up to the third dose of Molday (mean of N = 4 mice). After each dose, R₂′ increased substantially. β maps were generated based on the dose-relaxation curve, and are shown in the last row of Figure 3(a). Eight anatomical regions were chosen as ROIs, to represent the major areas based on a mouse brain atlas (http://atlas.brain-map.org/). See Table S1 for a list of ROIs and Figure S3 for locations of different ROI. ROI-specific quantification is shown in Figure 3(b).

Figure 3.

Dose-dependent R₂′ for β-mapping. A. Mean R₂′ images from four mice (N = 4) before, and after one, two, and three doses of Molday, to which β was fitted. B. Top row, ROI definition, bottom row, [5%,95%] mean β values distribution from each ROI. The colored outline of the bar matches the color of the equivalent ROI. One dot shown per voxel. The mean value for all shown regions is 0.80. Full ROI names are given in Table S1. Compare to Shu et al., 2016 Figure 3. ²⁶ (See Figure S3 for ROI locations in grayscale).

For further information regarding α and β, see “Constants of the metabolic model” in supplemental information.

Measurement of rCMR_O2 using rcfMRI

Using the rcfMRI method, we used measurements of R₂′ and CBF with constants α and β, and calculated rCMR_O2 based on E4. Figure 4(a) shows the mean rCMR_O2 map (N = 15 mice) in the awake divided by the anesthetized state. 96.7% rCMR_O2 voxels are >1, indicating that CMR_O2 is globally higher in the awake state versus the anesthetized state. Furthermore, we observed this increase as appearing to be non-uniform across the brain. ROI-based analysis (Figure 4(b), see Figure S4 for ROI locations in grayscale) suggests a higher ratio in subcortical regions than cortical regions, and more variation in rCMR_O2 among cortical regions. The trend of higher subcortical than cortical rCMR_O2 was present in 13 out of 15 mice. Voxel values within each brain region were significantly non-normally distributed (p ≤ 0.034, KS test with Lilliefors adjustment). A Wilcoxon's rank sum test indicated that Cortical rCMR_O2 values were found significantly lower than the subcortical regions (p = 0.0062, mean of cortical regions = 1.64, mean of subcortical regions = 2.11). The auditory cortex exhibited the greatest change among the cortical ROIs, potentially indicating a higher baseline neural activity level in the awake state due to the conscious processing of acoustic noise during imaging. Indeed, a Wilcoxon's rank sum test indicated that the rCMR_O2 value of the auditory cortex was significantly higher than other cortical regions (p = 0.0344, mean of AUD = 2.26, mean of other cortical regions = 1.61).

Figure 4.

Relative CMRO₂ (rCMRO₂) image results and regional analysis. A. rCMR_O2 map calculated using the rcfMRI method, for the awake condition divided by the anesthetized condition (mean of N = 15 mice). Note that there is a global increase in the awake condition. B. Top row, ROI definitions, bottom row, [5%,95%] ROI-wise mean values distribution for rCMRO_2. The colored outline of the bar matches the color of the equivalent ROI. One dot shown per voxel. While there is a global increase, it does not appear to be uniform across the brain, or even appear to be uniform across the cortex. Full ROI names are given in Table S1. (See Figure S4 for ROI locations in grayscale).

Effect of varying of α on rcfMRI

To the best of our knowledge, α has not been directly measured in mice (and such a measurement is substantial work, beyond the scope of our study). Thus, we tested four different values for alpha from 0.1 to 0.4 based on the possible range ²⁹ for calculating rCMR_O2 using the rcfMRI method (as was done above in “Measurement of rCMRO2 using rcfMRI” for α = 0.2). These results are shown in Figure S5. While greater values for α produced slightly lower rCMR_O2 values, results were generally similar across the full range of α. No value of α changed our result of observing a non-uniform, whole-brain increase in the awake condition versus the anesthetized condition.

Measurement of T_2v using TRUST

To validate the rcfMRI method, we used a different MRI method, TRUST, ²⁷ to measure global oxygen metabolism in the awake and anesthetized states, in the same mice and imaging sessions where we measured rcfMRI. TRUST works by the subtraction of a control from a labeled image at multiple TE values, as shown in Figure 5(a), left side. Then, a monoexponential fitting of the signal intensity as a function TE yields the T₂ relaxation time of pure venous blood (T_2v), as shown in Figure 5(a), right side.

Figure 5.

TRUST global rCMR_O2 agrees with rcfMRI. A. Illustration of the TRUST method using a representative mouse from our study. The signal was calculated within the sinus confluence of the superior sagittal sinus at different echo times to allow the calculation of T_2v from regions which drain from it. B. Comparison of Y_v converted from T_2v between awake and anesthetized states(N=15), mean±SD for venous oxygen saturation. One dot shown per mouse. C. Comparison of CBF values between awake and anesthetized states(N=15), mean±SD with error bar. One dot shown per mouse. D. Comparison between global rCMR_O2 from TRUST (rCMR_O2_T) and averaged rcfMRI result from the superior sagittal sinus (rCMR_O2_M). (N = 14) (Scatter plot shown in Figure S6).

We observed a change in the mean T_2v in the awake condition (T_2v = 6.84 ± 1.53 ms, mean ± SD.) when compared to the anesthetized condition (T_2v = 5.80 ± 1.34 ms, mean ± SD.). T_2v in each condition was potentially normally distributed (p > 0.05, KS test with Lilliefors adjustment). This change was statistically significant (paired T-test, one tail assuming the same direction as rcfMRI: awake > anesthetized, p = 0.025, T = 2.14).

Comparison of rCMR_O2 from rcfMRI (rCMR_O2_M) and rCMR_O2 from TRUST (rCMR_O2_T)

To directly compare between rcfMRI and TRUST, both were converted into global rCMR_O2 measurements. For the former, the mean value for all voxels in the watershed area of the superior sagittal sinus (Figure S2)^44,45 was taken from the mean (N = 15 mice) rCMR_O2 map created using the rcfMRI method. This was referred to as rCMR_O2_M. For the latter, values for T_2v were first converted to Y_v as per Lu et al., ²⁷ shown in Figure 5(b). Y_v was 61.0 ± 4.8% for awake and 57.6 ± 5.4% for anesthetized (mean ± SD., same test as T_2v above, p = 0.039, T = 1.90). Mean CBF values from the watershed area (Figure S2) for each mouse and each state are shown in Figure 5(c). CBF was 90.1 ± 7.3 mL/100g/min for awake and 52.3 ± 10.1 mL/100g/min for anesthetized (mean ± SD., same test as T_2v above, p = 5.66 × 10⁻⁹, T = 11.8). Combining Y_v and whole-brain CBF, as per Ciris et al. and Lu et al.,^27,49 we obtained global rCMR_O2 from TRUST. This was referred to as rCMR_O2_T (Figure 5(d)).

The mean value for the rCMR_O2_T is 1.63 ± 0.33 while rCMR_O2_M’s mean value is 1.52 ± 0.36 (Mean ± SD.) at α = 0.2. rCMR_O2_M and rCMR_O2_T were significantly non-normally distributed (p ≤ 0.032, KS test with Lilliefors adjustment). Due to this non-normality, we decided to compare both linear similarity (Pearson correlation) and similarity which includes non-linearity (Mutual Information).

Comparison analysis was not statistically significant for linear similarity measured with Pearson correlation, but was statistically significant when non-linear similarities were considered using mutual information. Pearson correlation between rCMR_O2_T and rCMR_O2_M on a per-mouse basis was r = 0.23, this was not statistically significant (p = 0.43). Mutual information was 1.5 for actual data, with a median of 1.1 for random data with 5th percentile of 0.80 and 95th percentile of 1.4, resulting in p = 0.013. This suggests a statistically significant, but non-linear relationship between CMR_O2 as measured with TRUST and CMR_O2 as measured with calibrated fMRI.

Discussion

Main findings

Calculating rCMR_O2 maps using relaxometry-based calibrated fMRI (rcfMRI) required measurement of relaxation time R₂′ and CBF for the same mouse subject, as well as the parameters α, which we tested across a range, and β, which we measured in a separate cohort of mice. We succeeded in measuring R₂′ and CBF in 15 mice in both the awake and anesthetized states. Considering awake versus anesthetized states, we observed globally higher rCMR_O2 in the awake state. Moreover, the increase we observed was non-homogenous; the cortex showed a trend of a smaller change than the subcortex with high rCMR_O2 in auditory cortex. The observation of globally higher rCMR_O2 in the awake state versus an anesthetized state follows many previous studies which used PET to measure glucose metabolism or CBF in human subjects given anesthetics.^7,50–55 While care must be taken as gray matter/white matter balance differences between the cortex and the subcortex may alter signal-to-noise (SNR) of our measurements, the difference in the change between cortical and subcortical regions matched previous studies of dexmedetomidine anesthesia in humans (Table S3).^22,56

Interestingly, while our work indicates that rcfMRI can measure the metabolic shift from awake to anesthetized, we also found a non-linear relationship between rcfMRI and TRUST rCMR_O2 results. Results from TRUST were in the same direction as from rcfMRI (Figure 5(d)), but the two global measurements only had a weak linear relationship on a per-mouse basis. There was, however, a statistically significant non-linear relationship between global rCMR_O2 measurement with TRUST versus with rcfMRI. This suggests that modeling deoxyhemoglobin from susceptibility changes may be more complex than expected for rCMR_O2_M by rcfMRI. Whereas some speculation is given in Figure S6 for rCMR_O2_T by TRUST due to variable inflow to large vessels, to understand this non-linear relationship will require further study.

Effects of anesthesia, and CBF

We chose anesthesia as a method of creating a large alteration to brain metabolism. ¹⁴ Anesthesia affects neural activity in an agent- and dose-dependent manner. Most anesthetic agents, including dexmedetomidine, suppress neural activity (though other anesthetic agents, such as midazolam, did not significantly).^20,56,57 Studies done in rodents using ketamine-xylazine and urethane-xylazine have shown disturbances to temperature, heart rate, and breath rate, stressing the importance of maintaining physiological parameters under anesthesia. ⁵⁸ Studies done in rodents under isoflurane, (where dose may be easily altered) have shown a strong dose-dependence on the neural suppression induced by anesthesia. ⁵⁹

Our calibrated fMRI method attempts to model CMR_O2, and thus does not distinguish CMR_O2 changes due to purely vascular vs. purely neural activity, both of which can be induced by anesthesia. Vascular effects of dexmedetomidine are relatively stable 30 minutes after the bolus injection. ³⁹ Thus we expect vascular changes we observed to be due to the relatively stable state which the anesthesia elicits.

A recent meta-analysis investigated human PET studies which measured glucose metabolism ⁶⁰ suggested that results from studies which used anesthesia were dominated by a global decrease in glucose metabolism in the anesthetized state. As cerebral glucose and cerebral oxygen metabolism are tightly coupled,^42,61 our results support this observation. Indeed, Qiu et al., who measured neural activity directly, did observe a decrease under dexmedetomidine.^57,62,63 While this may suggest an effect which is not agent-specific, ⁶⁰ possibly due to reduced neural activity, dexmedetomidine hypothetically may also bind to α2-adrenergic receptors in vasculature. ⁶⁴ If this is occurring, some of the vasoconstriction under dexmedetomidine could be caused by this effect, in addition to vasoconstriction caused by reduced neural activity.

Also consider relative CBF (rCBF), calculated by dividing CBF in the awake state by the anesthetized state, shown in Figure S7. rCBF shows a global trend similar to rCMR_O2, as would be expected from human dexmedetomidine studies, ²¹ however rCBF has less variation between brain regions as it lacks the influence of the deoxyhemoglobin-sensitive component R₂′.

Anesthesia is used in most animal fMRI studies. ⁶⁵ However, local metabolic differences between awake and anesthetized states are rarely considered. By using dexmedetomidine, which is an agonist of α2-adrenergic receptor, ⁶⁶ we were able to examine this difference in the mouse brain. In addition to globally lower CMR_O2 in the anesthetized state, regional variability was also apparent. While we speculate some variability may be due to conscious processing of stimulus while awake (e.g. activation of auditory cortex shows a significant difference, likely due to scanner noise), overall the sources of regional variability may be highly complicated. We speculate that one of these sources may be variation of α2-adrenergic receptor expression levels across the brain, as has been observed using genetically-encoded calcium imaging, ⁶² and seen in human studies of CBF. ²² This may cause differential reductions in information processing, or differential vascular effects, due to the dexmedetomidine. Future studies comparing multiple anesthetics with the awake state would be needed to clarify this, though independent measurement of CBV may be needed for anesthetics which alter vasculature tone, such as isoflurane. ⁶⁷ Finally, while mice did undergo habituation prior to the experiment, the awake state may still be stressful, which could affect oxygen metabolism versus true “awake resting” as well.

rcfMRI in context of other MRI-based CMR_O2 measurements

As measuring the change in cerebral oxygen metabolism is a quantitative way to measure neural activity at a whole-brain scale, many MRI-based methods of quantifying CMR_O2 have been developed. These can be classified as extravascular methods, which include most calibrated fMRI methods, and intravascular methods, such as the TRUST method used in our study. ¹⁹ Calibrated fMRI methods can also be divided into gas-challenge and gas-free calibrated fMRI. Traditional calibrated fMRI required a gas challenge to determine M. This was done either by applying hypercapnia through administering CO₂ to subjects, or (less commonly) applying hyperoxia through administering O₂ to subjects. ¹⁹ High levels of CO₂ or low levels of O₂ can suppress the ability to measure neural activity as vascular reactivity is lost. This may thus present a particularly difficult problem in animals under anesthesia, as the anesthesia has already lowered the level of neural activity. ²⁵ In addition, even at CO₂ levels and stimulation levels where gas challenges are effective, they may create problems for translation including the potential for anxiety or panic, ⁶⁸ and altered electroencephalogram rhythms,^69,70 which creates a risk for patients with neurological conditions. Altered neural activity is also a problem from a scientific perspective; in addition local activity is also altered.^71,72

Recently-developed gas-free calibrated fMRI methods, including the rcfMRI method we used, are exciting as they may provide a way to apply calibrated fMRI broadly. Our work used the method developed by Shu et al. ¹⁵ A similar method to Shu et al. was used by Göttler and Kaczmarz et al. who calculated rCMR_O2 based on using R₂′ and independent CBV and CBF measurements to find the oxygen extraction fraction (OEF). ⁷³ Berman et al. used a gas-free calibrated fMRI method which has similar assumptions to Shu et al. (see E3a in the present study), but instead of measurement of traditional R₂ and R₂* maps they quantified R₂′ at multiple echo times using asymmetric spin echo sequences. ¹⁶ It is worth noting that the asymmetric spin-echo sequences used in Berman et al. to record R₂ maps underestimated the R₂′ value, whereas standard spin-echo EPI sequences (including the MSME sequence we used) do not appear to underestimate R₂′. However, standard spin-echo EPI sequences have more heterogeneity than non-EPI sequences as per Figure 4 from Shu et al. ¹⁵ Thus the choice of the relaxometry sequences will influence measurement of R₂′ and be a tradeoff versus time, accuracy, and under- or over- estimation.

Conclusion

We were able to successfully use rcfMRI in awake and anesthetized mice to observe the global metabolic shift due to dexmedetomidine anesthesia. As expected based on human studies, the anesthetized state was lower than the awake state. This global shift was recapitulated with TRUST results from the same mice.

In addition, the global shift was non-homogeneous, indicating that rcfMRI in mice can be useful for measuring per-voxel changes in brain metabolism. Observed local differences included a cortical-subcortical difference that was similar to what was observed in CBF in humans under dexmedetomidine, ²² and also potential greater activation of the auditory cortex in the awake state due to acoustic scanner noise.

Our results thus suggest that per-voxel mapping of calibrated fMRI is possible in the preclinical mouse model. Further study is needed to apply this preclinical imaging tool, as well as to determine translation to human subjects.

Supporting information

All relevant data are available from the authors and included in the manuscript or supplementary information. Supplementary Figures S1–S7 and supplementary Tables S1–S3 are included in the supplementary information. Correspondence and requests for materials should be addressed to gthompson@shanghaitech.edu.cn or zliang@ion.ac.cn.