Cerebral Glutamate Alterations Using Chemical Exchange Saturation Transfer Imaging in a Rat Model of Lipopolysaccharide-Induced Sepsis

Glutamate-weighted chemical exchange saturation transfer (GluCEST) is a useful imaging tool to detect glutamate signal alterations caused by neuroinflammation. This study aimed to visualize and quantitatively evaluate hippocampal glutamate alterations in a rat model of sepsis-induced brain injury using GluCEST and proton magnetic resonance spectroscopy (1H-MRS). Twenty-one Sprague Dawley rats were divided into three groups (sepsis-induced groups (SEP05, n = 7 and SEP10, n = 7) and controls (n = 7)). Sepsis was induced through a single intraperitoneal injection of lipopolysaccharide (LPS) at a dose of 5 mg/kg (SEP05) or 10 mg/kg (SEP10). GluCEST values and 1H-MRS concentrations in the hippocampal region were quantified using conventional magnetization transfer ratio asymmetry and a water scaling method, respectively. In addition, we examined immunohistochemical and immunofluorescence staining to observe the immune response and activity in the hippocampal region after LPS exposure. The GluCEST and 1H-MRS results showed that GluCEST values and glutamate concentrations were significantly higher in sepsis-induced rats than those in controls as the LPS dose increased. GluCEST imaging may be a helpful technique for defining biomarkers to estimate glutamate-related metabolism in sepsis-associated diseases.


Introduction
Sepsis is a complex systemic syndrome characterized by an imbalance between proand anti-inflammatory responses to invading pathogens [1] and is a major cause of morbidity and mortality in intensive care units worldwide [2]. Sepsis and its associated complications may progress to septic shock and multi-organ failure through uncontrolled production of proinflammatory cytokines [3]. Cerebral dysfunction is the initial symptom and frequent complication of sepsis, usually presenting as sepsis-associated encephalopathy (SAE), with an incidence rate of up to 70%, depending on the severity of the patient's symptoms [2,4,5]. Moreover, SAE can affect brain cells and induce blood-brain barrier breakdown, intracellular metabolism dysfunction, brain cell death, and brain injuries [6]. Therefore, biophysical assessment of SAE in cerebral metabolism through observational

Animal Models
All experiments were conducted to minimize the number of animals used in the study and the pain and suffering of all experimental animals. The study complied with the principles outlined in the Animal Research: Reporting of In Vivo Experiments guidelines and Basel Declaration, including the replacement, reduction, and refinement (3R) concept. Moreover, all animal care and experiments were conducted with the approval of the Animal Care and Use Committee of the Asan Medical Center of the University of Ulsan Medical School (approval date: 25 February 2020; permit code: 2020-13-056).
Starting at 8 weeks of age, 21 Sprague Dawley rats were obtained from Orient Bio, Inc. (Seongnam, Kyunggi-do, Korea) and divided into three groups (sepsis-induced groups (SEP05, n = 7 and SEP10, n = 7) and control (CTRL, n = 7) group). The animals were pair-housed in standard plastic cages and maintained on a 12 h light-dark cycle at an ambient temperature of 23-24 • C. Before the start of the experiments, the animals were allowed free access to food and water for 1 week.
Sepsis was induced through a single intraperitoneal injection of LPS (from Escherichia coli O26:B6; L8274; Sigma-Aldrich, St. Louis, MO, USA) at doses of 5 mg/kg (SEP05) and 10 mg/kg (SEP10) [23,24]. An equal volume of vehicle (sterile endotoxin-free PBS) as that in SEP05 and SEP10 groups was administered intraperitoneally to the CTRL rats. All animals underwent MRI 24 h after injection of a single dose of LPS or vehicle.

MRI Scan
MRI experiments were conducted using a 7 T horizontal-bore PharmaScan 70/16 scanner (Bruker BioSpin GmbH, Ettlingen, Germany) with a 400 mT/m self-shielding gradient system and an actively decoupled cross-coil setup, which consisted of a 72 mm body coil for excitation and 25 mm single-loop rat brain surface coil. During the MRI scans, the animals were anesthetized using medical air (1.0 L/min) with isoflurane (1.5-3.0% for induction and 2.0% for maintenance) in a mixture of 70% N 2 O and 30% O 2 administered through a nose cone and were immobilized with ear-bars during scanning. Body temperature was maintained at approximately 37 • C using a warm-water circulation system in a flat bed positioned around the rat's body. Physiological parameters, including heart and respiratory rates, were monitored using an animal respiratory gating system (SA Instruments, Stony Brook, NY, USA).
Before MRI acquisition, the B 0 field was corrected through localized high-order shimming using a region of interest (ROI, 34 × 34 × 34 mm 3 ) covering the whole cerebral area to achieve a homogeneous B 0 field and a high signal-to-noise ratio in the MRI system. Subsequently, the radio frequency (RF) field and center frequency were calibrated using the prescan protocol. GluCEST and 1 H MRS scans were conducted sequentially for each rat.
In vivo GluCEST data were collected from a single slice in which the hippocampal region was well observed, and CEST parameters were applied as follows: fat-suppressed, turbo-rapid acquisition with relaxation enhancement sequence with a continuous-wave RF saturation pulse, repetition time (TR) = 4.2 s, echo time (TE) = 36.4 ms, RF saturation power = 3.6 µT, RF saturation time = 1 s, 96 × 96 matrix size, 30 × 30 mm 2 field-of-view (FOV), and 1.5 mm slice thickness [14,18,25]. Z-spectra with 25 frequency offsets from -6 to +6 ppm (intervals of 0.5 ppm) and an unsaturated reference image (S 0 ) were obtained from the selected single-slice image. To correct the B 0 field inhomogeneity, water saturation shift referencing (WASSR) Z-spectra with 33 frequency offsets from -0.8 to +0.8 ppm (intervals of 0.05 ppm) using 0.3 µT RF saturation power were applied [26]. The B 1 map was generated from two images obtained using the dual flip-angle (30 • and 60 • ) method [18].
For the single voxel placement of 1 H MRS, anatomical T 2 -weighted MR images with TR/TE = 5250/66 ms, FOV 25 × 25 mm 2 , 256 × 256 matrix size, 11 ms echo spacing, and 2 averages were obtained in all three orthogonal planes. A spin-echo-based, point-resolved spectroscopy pulse sequence with water-selective suppression was used for the hippocampal region. All parameters were as follows: variable power and optimized relaxation delays method, volume of interest = 2 × 2 × 3 mm 3 (12.0 µL), TR/TE = 5000/16.3 ms, 5000 Hz spectral width, 256 average, 2048 data points, and total scan time = 23 min 15 s.

Data Analysis
All CEST data processing and analysis were performed using custom-written scripts in MATLAB 2022b (MathWorks, Natick, MA, USA). To correct the B 0 inhomogeneity, the Z-spectra from the WASSR data for each voxel were fitted, and the water signal in the chemical shift was reset to 0 ppm to obtain the frequency shift information in each voxel. The relative B 1 values were calculated using the B 1 map. The GluCEST signal was calculated by subtracting the normalized magnetization signal at 3.0 ppm from the magnetization at the corresponding reference frequency symmetrically up-field from water; GluCEST (%) = 100 × (S sat@−3.0 ppm − S sat@+3.0 ppm ) ÷ S sat@−3.0 ppm [18], where S sat@−3.0 ppm and S sat@+3.0 ppm are the water signals with a saturation pulse at offsets ±3 ppm from the water resonance. GluCEST signals in all rats were carefully quantified based on a manually drawn ROI in the hippocampal region.

Hematoxylin and Eosin Staining
After MRI scans, all animals were anesthetized and fixed through cardiac perfusion using 4.0% paraformaldehyde (PFA) (Biosesang, Seongnam, Republic of Korea). The brain tissues were obtained and then soaked in 4% PFA for 24 h. Paraffin-embedded tissues were sectioned into 3 µm thickness and mounted on slides coated with poly-L-lysine. Finally, the slides were stained with hematoxylin (for 10 min) and eosin (for 5 min) (H&E staining) at room temperature. The pathological features of the hippocampal tissues were observed using an automated quantitative pathology imaging system (Leica, Wetzlar, Germany). H&E staining was performed as previously described [27,28].

Statistical Analysis
All statistical analyses were performed using PASW Statistics 18 (SPSS Inc., IBM Company, Chicago, IL, USA). The Kolmogorov-Smirnov and Shapiro-Wilk tests were used to test the normality of the GluCEST and 1 H MRS data in the three groups [31]. The p values for GluCEST and 1 H MRS data normality in all groups were above 0.05 (p < 0.169 for the Kolmogorov-Smirnov test and p < 0.594 for the Shapiro-Wilk test), indicating that the data were normally distributed. The results of the present study described one independent variable (dose of LPS) and two dependent variables (GluCEST values and Glu concentrations in the hippocampal region), so one-way analysis of variance was used [32]. As the sample sizes for each group were equal, the Tukey's post hoc test was performed for pairwise comparisons between the group means [33]. Statistical differences were assumed to be significant for p values < 0.05. G*Power 3.1.9.7 software was used to calculate the animal sample size required for the present study. The sample size suitable for the experiment was computed by applying α = 0.05, power (1 − β) = 0.8, and effect size = 0.8. In the G*Power analysis, an actual power of 0.86285 was used. A required sample size of seven animals per group (total sample size = 21) was obtained. Figure 1 shows the magnetization transfer ratio asymmetry (MTR asym ) curves (a) and quantified GluCEST signals at 3.0 ppm (b) in the rat hippocampus in each group (CTRL, n = 7; SEP05, n = 7; and SEP10, n = 7). The MTR asym curves showed distinct differences among the groups. As the LPS dose increased, the MTR asym values were higher than those in the CTRL group. Moreover, as the frequency offset increased (approximately ≥ 4.0 ppm), the MTR asym data in all groups were negative, which might be the effect of upfield nuclear Overhauser enhancement [34,35]. The quantified GluCEST signals differed significantly among the three groups. The mean GluCEST values and standard deviation (SD) are shown, as follows ( Figure 1b): CTRL (2.03 ± 0.45%) vs. SEP05 (2.60 ± 0.43%) (p < 0.05), SEP05 vs. SEP10 (3.62 ± 0.38%) (p < 0.001), and CTRL vs. SEP10 (p < 0.001).

Results
, 13, x FOR PEER REVIEW 5 of 11 Figure 1 shows the magnetization transfer ratio asymmetry (MTRasym) curves (a) and quantified GluCEST signals at 3.0 ppm (b) in the rat hippocampus in each group (CTRL, n = 7; SEP05, n = 7; and SEP10, n = 7). The MTRasym curves showed distinct differences among the groups. As the LPS dose increased, the MTRasym values were higher than those in the CTRL group. Moreover, as the frequency offset increased (approximately ≥ 4.0 ppm), the MTRasym data in all groups were negative, which might be the effect of upfield nuclear Overhauser enhancement [34,35]. The quantified GluCEST signals differed significantly among the three groups. The mean GluCEST values and standard deviation (SD) are shown, as follows ( Figure 1b): CTRL (2.03 ± 0.45%) vs. SEP05 (2.60 ± 0.43%) (p < 0.05), SEP05 vs. SEP10 (3.62 ± 0.38%) (p < 0.001), and CTRL vs. SEP10 (p < 0.001).    Figure 3 shows the voxel placement in the hippocampal region (Figure 3a,b) and presents the spectral fitting results of 1 H-MRS data in a representative rat from each group (Figure 3c-e). Visual inspection of the proton spectra indicated that the Glu peak area was slightly different among the three groups and showed higher intensities with an  3. Results Figure 1 shows the magnetization transfer ratio asymmetry (MTRasym) curves (a) and quantified GluCEST signals at 3.0 ppm (b) in the rat hippocampus in each group (CTRL, n = 7; SEP05, n = 7; and SEP10, n = 7). The MTRasym curves showed distinct differences among the groups. As the LPS dose increased, the MTRasym values were higher than those in the CTRL group. Moreover, as the frequency offset increased (approximately ≥ 4.0 ppm), the MTRasym data in all groups were negative, which might be the effect of upfield nuclear Overhauser enhancement [34,35]. The quantified GluCEST signals differed significantly among the three groups. The mean GluCEST values and standard deviation (SD) are shown, as follows ( Figure 1b): CTRL (2.03 ± 0.45%) vs. SEP05 (2.60 ± 0.43%) (p < 0.05), SEP05 vs. SEP10 (3.62 ± 0.38%) (p < 0.001), and CTRL vs. SEP10 (p < 0.001).    Figure 3 shows the voxel placement in the hippocampal region (Figure 3a,b) and presents the spectral fitting results of 1 H-MRS data in a representative rat from each group (Figure 3c-e). Visual inspection of the proton spectra indicated that the Glu peak area was slightly different among the three groups and showed higher intensities with an   (Figure 3a,b) and presents the spectral fitting results of 1 H-MRS data in a representative rat from each group (Figure 3c-e). Visual inspection of the proton spectra indicated that the Glu peak area was slightly different among the three groups and showed higher intensities with an increasing dose of LPS than that in the controls. The quantified Glu concentrations differed significantly among the three groups. The mean Glu concentrations and SD are shown as follows (Figure 3f): CTRL (n = 7; 6.401 ± 0.450 µmol/g) vs. SEP05 (n = 7; 7.303 ± 0.671 µmol/g) (p < 0.01), SEP05 vs. SEP10 (n = 7; 7.773 ± 0.812 µmol/g) (p < 0.05), and CTRL vs. SEP10 (p < 0.001). Overall, the 1 H-MRS data demonstrated a similar trend to the GluCEST data, showing high Glu concentrations as the LPS administration dose increased.   Figure 4 shows the histological approach of the hippocampal region in a representative rat from each group. Representative H&E staining and immunofluorescence show the biological characteristics of sepsis-induced cerebral metabolism in the hippocampal region (Figure 4b).  7; and SEP10, n = 7), and the vertical lines on each of the bars represent the standard deviation (ppm, part per million; VOI, volume of interest; significance levels: * p < 0.05, ** p < 0.01, and *** p < 0.001). Figure 4 shows the histological approach of the hippocampal region in a representative rat from each group. Representative H&E staining and immunofluorescence show the biological characteristics of sepsis-induced cerebral metabolism in the hippocampal region (Figure 4b).

Discussion
Although the pathological process of SAE remains unclear, early diagnosis and treatment are important to reduce mortality and disability due to cerebral injury in patients with sepsis. The present study showed that GluCEST imaging and 1 H MRS techniques have a high sensitivity to changes in Glu signals in vivo in the hippocampal region of rats with SAE. Glutamate distribution in the rat hippocampal region was also determined. In addition, we tested cerebral neuroinflammation status and biochemical changes during the SAE using immunohistochemical and immunofluorescence staining.
Since its introduction, CEST has been used in numerous clinical applications for various diseases, such as epilepsy, ischemia, tumors, and psychiatric disorders, to detect in vivo signal changes of specific endogenous agents (for example, proteins and neurotransmitters, such as Glu) [14,[18][19][20][21][22]. However, to date, only a small number of in vivo studies have assessed signal changes in cerebral metabolites using MR-based imaging techniques in patients with SAE or animal models. To our knowledge, no previous studies have evaluated changes in the GluCEST signal caused by SAE. The present study clearly showed in vivo Glu changes in sepsis-induced lesions, demonstrating higher GluCEST values and Glu concentrations ( 1 H MRS) than those in the CTRL group as the LPS dose increased. Regarding the GluCEST signals quantified in this study, we attempted to minimize the factors that influence GluCEST signal formation. As several previous studies have indicated [18,25,26], B 0 and B 1 field inhomogeneities may affect signal drift by overall Z-spectrum shift, resulting in inaccurate signal quantification. To address these factors, we applied the WASSR method and relative B 1 map calculation, respectively, to correct for field inhomogeneity in the whole frequency offset of the GluCEST data. We used a relatively high B 1 saturation power (3.6 µT in the present study) and short saturation time to observe the Glu-weighted signals in the 7 T system. In addition, to minimize the change in pH, which would affect GluCEST signaling, the body temperature of the experimental animals was maintained (37 • C) in this study. Therefore, based on the factors mentioned above, there were significant differences in GluCEST values among the three groups, which can be solely attributed to the differences in Glu concentrations.
Several studies using 1 H MRS have reported quantitative evaluations of cerebral metabolite changes in SAE [36][37][38]. Wen et al. [38] demonstrated a significant decrease in the NAA/Cr ratio in sepsis-induced rats based on the LPS administration model compared to controls, although there was no difference in T2-weighted MR images. Several sepsisinduced animal studies using the cecal ligation and perforation surgical model showed significant differences in the values of various metabolite ratios compared to the baseline or controls, such as increased Glx/Cr [37] and mI/Cr [37] ratios and decreased NAA/Cr [37] and NAA/Cho [36] ratios. In previous conventional 1 H MRS studies, cerebral metabolite changes in various diseases were quantified using the (specific metabolites)/Cr ratio method [39,40]. However, in addition to the relative quantification of these metabolites, the evaluation of absolute concentration based on internal standards (water scaling) might be helpful in interpreting the biological characteristics of the disease [41]. Contrary to the results of previous studies, the present study only quantified changes in Glu concentration using internal standards and observed distinct differences among the three groups. Kitagawa et al. [42] also demonstrated increased Glu levels in LPS-treated rats compared to controls and validated the increased microglial activation using immunofluorescent staining using an Iba-1 antibody. As is well known, the morphology of microglia can be visualized through the staining of Iba-1, which is a specific marker for pan-microglia [43]. Microglia are macrophages present in the central nervous system that orchestrate the inflammatory response in the brain [44]. Therefore, neuroinflammation is necessary to maintain immune homeostasis by inducing repair mechanisms [45]. However, failure to maintain homeostasis might be interpreted as a potentially harmful mechanism of brain injury [45,46]. Recent studies have demonstrated that increased Glu concentrations are often observed in neuronal inflammatory diseases and may be generated by an activated immune system and microglial reactivity during neuroinflammation [47][48][49]. In addition, Moraes et al. [46] suggested that activated microglia constitute a hallmark of SAE and may be involved in associated synaptic deficits.
However, this study presented some limitations. First, to extend the anatomical coverage of GluCEST imaging for detecting signals in SAE regions, multi-slice or threedimensional GluCEST imaging is required. Further studies dealing with multiple regions of the brain susceptible to SAE, including gray and white matter regions, may provide information on the role of in vivo Glu in sepsis-related diseases and corroborate the results of this study. Second, because we focused on detecting changes in GluCEST values and Glu concentrations in a rat model of sepsis-induced brain injury, neurochemical profiles other than those of Glu were not evaluated. As previously reported [36][37][38], further studies are required for the quantitative evaluation of various neurochemical profiles, such as Cr, NAA, Glx, and Cho, in order to identify new biomarkers. Third, several previous studies have identified key cerebral markers in SAE using dynamic 18F-fluorodeoxyglucose PET with computed tomography, multi-parametric MR imaging (including diffusion-weighted or tensor imaging), and resting-state functional MRI [5,37]. Therefore, further studies are needed to evaluate the origin of Glu signals, as well as the dynamic metabolic activity, and to determine the severity of SAE at various time points, including during the early stages of sepsis progression. Fourth, since we did not quantitatively analyze the signal intensity in the immunofluorescence staining data, statistical significance could not be determined. Quantitative evaluation of neuroinflammation markers in the blood, as well as in various histological stainings of various brain regions, should be conducted in future studies to corroborate these results. Finally, the number of experimental animals used in this study was relatively small, although the present results showed that the changes in Glu signals were significant among the three groups. Further studies with a larger number of animal samples modeling SAE are required to generate more reliable results.

Conclusions
Glutamate-weighted 7 T MRI was adequately sensitive in detecting in vivo changes in the hippocampal region in sepsis-induced cerebral injuries. Significant contrast changes in Glu-weighted metrics coupled with the histologic characteristics of sepsis-induced cerebral injuries indicate the potential usefulness of GluCEST imaging in observing neuroinflammatory processes.  Informed Consent Statement: Not applicable.

Data Availability Statement:
The data that support the findings of this study are available from the corresponding author upon reasonable request. Data is not publicly available due to privacy or ethical restrictions.

Conflicts of Interest:
Yu Na Hee is an employee of QuBEST BIO Inc. The paper reflects the views of the scientists and not the company.