All experiments were approved by the Committee on Animal Experimentation for the Canton de Vaud, Switzerland (VD3022.1). Adult male Wistar rats underwent bile duct ligation (BDL; Charles River Laboratories, L’Arbresle, France, 175–200 g at surgery) to create a model of chronic liver disease-induced type C HE, as previously described (Braissant et al., 2019; DeMorrow et al., 2021), and SHAM surgery as controls for histology assessment only. BDL animals were scanned longitudinally under isoflurane anesthesia (~1.5%, in a mixture of 50% oxygen and 50% air). A first scan was performed before surgery (n = 5, “week 0”), and a second scan between 6- and 7-week post-BDL (same animals, n = 5, “week 6”); the long duration of the MRI/MRS experiments did not allow us to scan all animals on the same day. For the MRI/MRS scans, all animals were placed in an in-house-built holder, with the head fixed in a stereotaxic system using a bite bar and a pair of ear bars. A small-animal monitor system (SA Instruments, New York, NY, USA) was used to monitor the body temperature (maintained at 37.7 ± 0.2°C by warm circulating water and measured with a rectal thermosensor) and the respiration rate. Blood samples of bilirubin (Reflotron Plus system, F. Hoffmann-La Roche Ltd.) and ammonium (PocketChem™ BA PA-4140) were performed before the MRI/MRS scans (before BDL and at week 6–7 post-BDL) to validate the model of chronic liver disease.

All experiments were performed on a 9.4 T, actively shielded MRI system with a 31-cm horizontal bore (Magnex Scientific, Oxford, United Kingdom), featuring a 12-cm gradient coil insert (400 mT/m, 120 μs) interfaced to an Agilent/Varian Direct Drive console (Palo Alto, CA, USA). An in-house-built ¹H quadrature transceiver was used (25-mm inner diameter).

Fast T₂-weighted images (multislice turbo-spin-echo sequence, TR = 4,000 ms, TE_eff = 52 ms, echo train length = 8, field of view (FOV) = 23 × 23 mm², slice thickness = 1 mm, 15 slices, matrix size = 256 × 256, two averages) were acquired in the axial direction to position the volumes of interest (VOIs) for ¹H-MRS. First, a ¹H-MRS scan was performed in the hippocampus (2 × 2.8 × 2 mm³, 11.8 μL) using the SPECIAL sequence (TE = 2.8 ms, TR = 4 s, 160 shots) to measure neurometabolism, as previously described (Braissant et al., 2019). Then, dMRS data were acquired using a localized STEAM-based spectroscopic pulse sequence (Kunz et al., 2010; Ligneul et al., 2024) (TE/TM = 15/112 ms, 5 kHz spectral width, 4,096 spectral points, single shot acquisitions) in a voxel ranging from 162 to 245 μL depending on the animal. The dMRS voxel size was increased compared to the hippocampus MRS voxel due to the lower signal-to-noise ratio (SNR) in the diffusion experiments as compared to a simple MRS acquisition. FASTESTMAP (Gruetter, 1993; Gruetter and Tkác, 2000) was used for shimming, leading to water linewidths of 9–10 Hz in the hippocampus and of 18–20 Hz for the dMRS VOI (Figure 1A). Outer volume suppression blocks were interleaved with the VAPOR water suppression module. Diffusion gradients were applied simultaneously along three orthogonal directions (gradient duration δ = 6 ms, diffusion time Δ = 120 ms, direction [1,1,1]). A total of nine b-values (in ms/μm², corrected for cross-terms (Kunz et al., 2010; Mosso et al., 2024)) with the following number of shots were acquired: 0.4 (160), 1.5 (160), 6.0 (160), 7.6 (160), 9.3 (160), 13.3 (320), 15.6 (480), 20.8 (480), and 25.1 (480).

Spectra were collected as single shots (consecutive ISIS acquisitions from SPECIAL were directly combined, with each combination being labeled as “shot” in this article) and corrected for eddy current distortions and phase and frequency drifts. Outlier shots with manifest signal drops (>50%) were removed, and all shots were averaged. Metabolite signals were quantified using LCModel (Version 6.3-1 N) combined with an in vitro measured metabolite basis set for ¹H-MRS hippocampus spectra and a simulated metabolite basis set for dMRS spectra using published values of J-coupling constants and chemical shifts (Govindaraju et al., 2000; Govind et al., 2015). The following metabolites were included in the basis sets: alanine (Ala), ascorbate (Asc), aspartate (Asp), β-hydroxybutyrate (bHB), glycerophosphocholine (GPC), phosphocholine (PCho), creatine (Cr), phosphocreatine (PCr), glycine (Gly), GABA, glucose (Glc), Gln, glutamate (Glu), glutathione (GSH), Ins, lactate (Lac), N-acetylaspartate (NAA), N-acetylaspartylglutamate (NAAG), phosphoethanolamine (PE), scyllo-inositol (Scyllo), and Tau. In addition, an in vivo macromolecule spectrum acquired under the same conditions as in vivo ¹H-MRS and dMRS data was included in each of the corresponding metabolite basis sets (single inversion recovery, TI = 750 ms, TR = 2,500 ms, no diffusion weighting, and metabolite residuals eliminated as described previously (Cudalbu et al., 2021; Simicic et al., 2021)). For ¹H-MRS, metabolites with relative Cramer Rao Lower Bounds (CRLB) below 25% at week 0 for all animals were reported (selecting reported metabolites but not individual values), a purposely loose criterion to avoid filtering out low-concentrated metabolites (Kreis, 2016). For dMRS, to limit error propagation, only metabolites with relative CRLBs below 6% at the lowest b-value for all animals were reported.

Metabolite signal decays as a function of b-value were then fitted using two different approaches, using a non-linear least squares algorithm in MATLAB (fit function, Trust-Region method): First, Callaghan’s model of randomly oriented sticks (Callaghan et al., 1979) (mimicking neurites or processes) with metabolite diffusivity D_intra along the neurite/process: (1) S S 0 = π 4 b D intra e r f b D intra

Second, the cumulant expansion at second order: (2) ln S S 0 = − b D + 1 6 b D 2 K

Yielding the apparent diffusion coefficient D and kurtosis K, where b refers to the b-value and erf refers to the error function. Assuming an underlying isotropic distribution of sticks (Callaghan’s model) (Equation 1), the radius of convergence of the cumulant expansion is given by the first zero of the error function in the complex plane (Kiselev and Il'yasov, 2007), whereby assuming a diffusivity of about 0.3 μm²/ms, b_c = 19 ms/μm². b-values up to b_c were used for this fit. For each metabolite, the fits were performed on the individual animal diffusion signal decay, and the fitted parameters were reported as mean and SD across animals. The fits were also performed on the group-averaged diffusion signal decay for each metabolite, yielding a mean coefficient.

A detailed table of the acquisition and processing parameters following the experts’ consensus recommendations on minimum reporting standards in in vivo MRS (Lin et al., 2021) is presented in Supplementary Table S1.

Data are presented as mean ± SD. For MRS, differences in metabolite concentration estimates were assessed with a repeated measure two-way analysis of variance (ANOVA) (Prism 5.03, GraphPad, La Jolla, CA, United States), with metabolites and disease (weeks 0 and 6) factors. For dMRS, differences in diffusion parameters based on individual animal fitting were assessed with a repeated measures two-way ANOVA on each parameter individually (ADC, AKC, D_intra), with metabolites and disease (weeks 0 and 6) factors. For both MRS and dMRS, Bonferroni’s multi-comparisons post-hoc test was applied, where the number of comparisons was set to the number of metabolites passing the CRLB criteria (n = 17 comparisons for MRS and n = 7 for dMRS). For the histological measures, a two-way ANOVA with post-hoc Tukey HSD was used to test for statistical significance. All tests were two-tailed. The significance level in all tests was attributed as follows: ^*p < 0.05, ^**p < 0.01, ^***p < 0.001, ^****p < 0.0001.

The increase in brain Gln in the hippocampus, mainly due to increased blood ammonium as a consequence of chronic liver disease, led to an osmotic imbalance resulting in a gradual decrease of other brain osmolytes (Ins), and contributed to morphological astrocytic alterations (shortening of process length together with a decrease in their number) (Häussinger et al., 2000), among other mechanisms. Of note, osmotic stress is not the sole mechanism involved in type C HE; oxidative stress and inflammation are complementary mechanisms acting synergistically (Simicic et al., 2022; Andersen and Schousboe, 2023; Pierzchala et al., 2023). Consistent with our previous findings (Braissant et al., 2019), we also identified some trends (not significant) in the changes of other metabolites. These changes included a decrease in Tau and tCho due to the osmotic response, the neurotransmitter Glu, and the antioxidants Asc and GSH. Moreover, the EM-observed intracellular accumulation of lipofuscin aggregates is a sign of lipid peroxidation and thus the presence of oxidative stress, which is in agreement with our previous studies (Pierzchala et al., 2022; Simicic et al., 2022), and we observed herein a decrease in antioxidants, a sign of redox homeostasis alterations. The same trends in metabolite changes were also observed in the bigger VOI used for dMRS. The quantification of these data was not used for characterizing the brain metabolism due to the longer TE of the dMRS sequence and the lack of T₂ corrections for water and metabolites.

In the BDL group at week 6, increased diffusivities of brain Gln and of the main brain osmolytes (Ins, Tau) were observed when compared to week 0. This observation is a possible consequence of the osmotic stress caused by intra-astrocytic Gln increase: brain osmolytes may be temporarily present in the extracellular space, therein experiencing freer diffusion compared to intracellular space, before being cleared out, as supported by the steady-state net decrease of osmolyte concentrations observed with MRS. The electron microscopy results revealed a loss of tissue integrity and enlarged extracellular spaces in the hippocampus, indicating increased extracellular water content in the area surrounding the astrocytes.

Furthermore, the increased diffusivities of Gln and Ins in the BDL rats, metabolites assumed to be glial markers (Martinez-Hernandez et al., 1977; Häussinger et al., 1994), may also reflect astrocyte alterations following the strong Gln increase. Changes in diffusivity in vivo are usually associated with microstructural changes (Najac et al., 2014, 2016; Palombo et al., 2018; Ligneul et al., 2019; Genovese et al., 2021), and the diffusion time used in the present study (120 ms, characteristic 2D diffusion length of 4 D Δ = 8.5 μm , assuming D = 0.15 μm²/ms) is likely to probe metabolite diffusion along fibers (astrocytic processes or neuronal dendrites) rather than confinement in cell bodies, as shown with dMRS in the human (Najac et al., 2016) and macaque brain (Najac et al., 2014). These dMRS findings are supported by the GFAP histological observations (i.e., decreased length and number of astrocytic processes) pointing toward a less restricted and ramified cellular architecture, explaining increased diffusivities for astrocytic metabolites in the BDL rats. A recent study showed increased serum GFAP levels in cirrhotic patients (Gairing et al., 2023), suggesting that the presence of astrocyte injury and astrocyte activation are two mechanisms that may lead to increased serum GFAP concentrations. It is worth mentioning that, using dMRS in a model of reactive astrocytes, Ins has been revealed as a specific intra-astrocytic marker whose diffusion closely reflects astrocytic morphology, enabling the non-invasive detection of astrocyte hypertrophy (Ligneul et al., 2019). In addition, the diffusion of astrocytic metabolites can mirror their altered morphology and pro-inflammatory phenotype (de Marco et al., 2022), since, during neuroinflammation, both astrocytes and microglia undergo metabolic, functional, and morphological changes (Heneka et al., 2014). In our previous studies, significantly elevated levels of IL-6 and reactive oxygen species were observed in the brains of BDL rats as compared to the SHAM animals (Pierzchala et al., 2022), suggesting the presence of neuroinflammation. IL-6 levels, together with oxidative stress, have also been associated with increased blood–brain barrier permeability, allowing neurotoxins to enter the brain and impair neurological functions (Simicic et al., 2022). Recent studies promoted dMRS as a tool sensitive to glial cytomorphological changes induced by inflammation following LPS administration in humans (de Marco et al., 2022) or in cuprizone-fed mice (Genovese et al., 2021), where Ins and tCho apparent diffusion coefficients were significantly elevated. Following these studies, tCho diffusivity changes were related to the presence of inflammation, even though tCho has a limited specificity for glial cells. Similarly, an increased diffusivity for tCho (a trend of ~20% increase) was measured in this study, possibly reflecting here also the presence of neuroinflammation, as shown previously in BDL rats ex vivo (Pierzchala et al., 2022). Neuroinflammation will impact the astrocyte cytoskeleton, which may lead to an increase in intracellular and extracellular space, as observed in our study. Furthermore, the EM data depicted a breakdown of myelin sheaths and myelin outfolding formation in BDL rats, which could also impact tCho diffusivity as tCho is required for membrane phospholipid synthesis and myelination (Zeisel et al., 1986), although only a few studies have validated the association between myelin status and tCho (Laule et al., 2007; Rae, 2014, Skripuletz et al., 2015).

Glu and NAA, both expected to be preferentially located in neurons (MOFFETT et al., 2007; Fendt and Verstreken, 2017), exhibited a trend of increased diffusivity in BDL rats at week 6 post-BDL. Golgi-Cox measures probed an increased soma surface of CA1 hippocampal neurons and a loss of dendritic spines density (which is made of filamentous actin cytoskeleton; Hering and Sheng, 2001) in BDL rats. Numerical simulations (Palombo et al., 2018) have suggested that decreased dendritic spines density would increase the ADC of neuronal metabolites, consistent with the trend of increased Glu and NAA diffusivity observed herein. Furthermore, additional studies using two-photon microscopy on brain slices (Santamaria et al., 2006, 2011) have shown that the diffusional characteristics of dendrites are greatly affected by the dendritic spines density, being slower in the dendrite with the higher density of spines due to anomalous diffusion, and significantly faster in smooth/low spines density dendrites. Of note, Glu is the main precursor of Gln synthesis in the astrocytes, and the observed increased diffusivity trend can also reflect the reduced number and shortening of astrocyte processes. Total Cr also showed a trend of increased diffusivity in this study. Creatine is located in most cell types, has different roles in energy metabolism and cytoprotection, and also appears to act in osmoregulation and neurotransmission (Rackayova et al., 2017; Braissant et al., 2019).

Metabolite kurtosis coefficients overall tended to decrease in the BDL group, suggesting that the intracellular space might be less heterogeneous with reduced structural disorder. Although previous numerical simulations have shown that the number of processes departing from the soma has almost no influence on the measured ADC at any diffusion time (Palombo et al., 2016), the former, observed here by histology, might have an influence on D_intra, which is higher in HE rats compared to control rats for most metabolites. Overall, we believe that increased diffusivities in type C HE rats versus control rats reflect (1) intracellular space alteration with reduced structural disorder, supported by decreased neuronal spines density, decreased length of astrocyte processes and number of ramifications shown by histology, and (2) a higher contribution of extracellular space diffusion in BDL rats compared to controls due to osmolytes leaving the cells counteracting intracellular Gln increase.

dMRS is a challenging measurement, and different factors might affect the estimated diffusion metrics (Ligneul et al., 2024). In the present study, motion artifacts due to simple linear translational motion were compensated on individual shots by phase correction, while data affected by rotational and compressive motion were discarded in the outlier removal process. Consequently, we do not expect any significant effect of motion on the calculated metabolite diffusion metrics, i.e., an overall overestimation. Additionally, a change in metabolite concentration is unlikely to affect diffusivity through a change in cytosol viscosity given the small metabolite concentrations (1–10 M) compared to water (45–50 M) (Kinsey et al., 2011).

For some metabolites (i.e., Tau, tCr, and tCho), the sticks model (Equation 1) showed some discrepancies with the measured data at high b-values. These discrepancies may result from a poorer LCModel spectral fit quality at higher b-values, exemplified by the high inter-animal variability of estimated concentrations for the highest b-values (Supplementary Figures S3,S4). Such variability might be partially alleviated by the use of simultaneous 2D fitting of the spectral and diffusion dimensions (Adalid et al., 2017).

The strict CRLB criterion for dMRS ensures a fair comparison between the groups. The use of each animal as its own control with a scan before the BDL surgery was beneficial as it ruled out possible inter-animal differences in brain microstructure or metabolism that could have biased the group comparison. A good concordance between individual and group-average fit for diffusion estimates was obtained in the current study. The possibility of fitting diffusion coefficients on individual animal signal decays provides an error estimation better representing the group dispersion than the one evaluated from the group-averaged signal decay. dMRS is also characterized by an overall low signal-to-noise ratio (SNR) compared to a simple MRS acquisition. In the present study, the LCModel SNR ranged from 15–20 to 45 depending on the b-value, guiding the choice of the randomly oriented sticks model (Equation 1) instead of the randomly oriented cylinder model (i.e., fitting, in addition, the radius of processes) (Vangelderen et al., 1994). This finding is highlighted in a recent dMRS consensus article showing that fitting the cylinder model would require a higher SNR and higher b-values than what was accessible in the present study (Ligneul et al., 2024). Finally, the increased brain Gln combined with the relatively short TE (15 ms) and high magnetic field allowed us to report brain Gln diffusivity for the first time.

Our study reports an overall increased diffusivity for all investigated metabolites, which was confirmed by histological measures. However, additional studies with an increased number of samples would be required to confirm this trend together with EM data on SHAM animals. dMRI could provide additional information with respect to dMRS, the former also informing on the extracellular space and on exchange between intracellular and extracellular spaces. An increased membrane permeability in BDL rats would also contribute to reduced compartmentalization (intracellular vs. extracellular) of metabolites (Ins, Gln, and Tau) and faster diffusion, which could be evaluated from joint dMRS and dMRI acquisitions in future studies. Future dMRS studies in this animal model should focus on targeting a specific brain region (the dMRS voxel here included several brain regions): brain regional differences in the neurometabolic profiles of BDL rats have been suggested, with the cerebellum exhibiting a stronger Gln increase than other brain regions (Simicic et al., 2019).