The study protocol was designed in accordance with the Declaration of Helsinki and approved by the local Ethics Committee (EK-2018-00937). Informed written consent was obtained from each participant before study enrolment.

Twenty-eight patients with chronic traumatic SCI (2 female) and 18 age- and education-matched healthy controls (3 female) were recruited between March 2021 and July 2021 at the Spinal Cord Injury Center, Balgrist University Hospital, for this cross-sectional study. During the prospective recruitment process of the study, we took steps to ensure a comparable proportional distribution of age and education between both groups. This was done to minimize any potential confounding variables that could affect the study’s results and to increase the validity of our findings. The number of subjects was defined based on our previous study on SCI (17). Inclusion criteria for SCI patients were: chronic stage of SCI (more than 6 months post-injury), no additional diagnosed neurological or psychological disorders, and no MRI contraindications. Inclusion criteria for healthy controls were no neurological, cognitive, or psychological disorders, and no MRI contraindications. To guarantee that participants could comprehend the testing materials and instructions as well as to provide accurate responses in German, all study participants were required to be fluent in the language, including proficiency in speaking, reading, and writing. Demographics and injury characteristics of SCI patients are summarized in Table 1. During the study appointment, all subjects were asked about their highest level of education. Graduating from mandatory school, including elementary school or secondary school, was defined as category [I] and graduating from high school or university was ascribed to category [II]. Data on behavioral visuospatial and verbal memory performance, MRS measurements of the right hippocampus, and hippocampal volumetric MRI measurements were obtained from all study participants.

To examine hippocampus-related cognitive function, the visuospatial and verbal working memory functions were assessed using the test ‘Verbaler und Visueller Merkfähigkeitstest’ (VVM) (18). The VVM is a widely used (19–22) and standardized test assessing memory function in German language and includes two tasks as follows: the first task is called ‘Map’ which assesses the visuospatial memory function, and the second task is called ‘Text’ which examines the verbal memory function. For the ‘Map’ task, the SCI and healthy controls were asked to memorize a specific path through a city map within 2 min time. Immediately thereafter, which was time-point 1 (t1), participants were asked to recall the path with drawing it on the same city map without the path within 2 min. Every correctly crossed intersection was scored with one point and a maximum number of 31 points were achieved. For the ‘Text’ task, participants were asked to read and memorize a text including names, numbers, and other facts during 2 min. They had to answer questions related to the text within 4 min immediately after memorizing the text details (t1). Every correct answer was scored with one point and a maximum number of 24 points were scored. The recalling part of both tasks was repeated after the scans– approximately taking 90 min – and represented intermediate memory function at time-point 2 (t2) (18). SCI patients without or with very impaired hand function were assisted filling out the tests to counteract the limited time factor.

We performed the Folstein Mini-Mental State Examination (MMSE) (23) with our study participants to explore potential cognitive deficits and to rule out participants with signs of dementia. We used a validated telephone version of the MMSE as part of the Adult Lifestyles and Function Interview (ALFI-MMSE) (24) to ensure the highest possible participation rate despite the limited availability of study participants. Of 28 SCI patients and 18 healthy controls, we were able to contact 22 patients and all healthy controls for the follow-up telephone interview. This version of the MMSE has a maximum score of 22 points, consisting of the subscale’s orientation to time, orientation to place, registration, attention, and recall. Study participants were categorized into without dementia, with mild dementia, with moderate dementia, or severe dementia, based on their scores.

MRS and MRI measurements were performed on a 3 T MR scanner (Prisma, Siemens Healthcare, Erlangen, Germany) with a 64-channel receive head and neck radio frequency (RF) coil. The protocol consisted of structural T1-weighted (T1-w) and T2-weighted (T2-w) MRI sequences of the whole brain, B₀ shimming, and MRS measurements in the right hippocampus and a reference region within the posterior parietal lobe (total acquisition time ≈60 min). The participants were scanned in a supine, head-first position and foam pads were used to minimize head motion in the RF coil.

As anatomical reference images, three sagittal, transversal, and coronal T2-weighted MRI scans based on a Turbo Spin Echo sequence were performed on the brain for subsequent MRS voxel placement on the hippocampus (echo time (TE) = 96 ms, echo spacing = 11 ms, repetition time (TR) = 5,000 ms, flip angle = 150°, in-plane resolution 0.5 × 0.5mm², echo train length = 11). The T2-w MRI measurements were recorded separately for all three directions (sagittal, transversal, and coronal), first covering the whole brain (slice thickness of 4 mm) and then focally with better slice resolution (1.6 mm) to delineate the hippocampus anatomy in more detail and increase the accuracy for placing the MRS voxel. To accelerate the T2-w scanning, we used GRAPPA mode with acceleration factor of 2.

B₀ shimming was performed using FASTESTMAP (25) prior to MRS measurement in the right hippocampus and posterior parietal lobe of the brain. The posterior parietal lobe served as an internal reference region since it is less affected by secondary neurodegeneration after SCI. Single-voxel MRS data were collected from a 25 × 12 × 8 mm³ (2.4 mL) voxel positioned in the right hippocampus and from a 20 × 14 × 22 mm³ (6.2 mL) voxel positioned in the parietal lobe. The MRS measurements were performed using a custom-made semi localization by adiabatic selective refocusing (sLASER) sequence (26) combined with metabolite cycling (MC) (27) [TE = 35 ms, TR = 2,500 ms, 1.024 s acquisition time, (only 0.512 s used)]. In total, 256 individually stored acquisitions were recorded across two blocks of 128 scans to reduce potential exclusion of whole datasets. Additionally, unsuppressed water signals with differing echo times (TEs = 35, 1,000, 50, 400, 200, 75, 100, and 140 ms, TR = 6,000 ms) were obtained to determine the parenchymal vs. cerebrospinal fluid (CSF) water signal for metabolite quantification. Outer volume suppression slabs were used to further prevent signal contamination from outer volume signals. Using sagittal, transversal, and coronal T2-w images, the volume of interest (VOI) was aligned parallel to the longitudinal axis of the right hippocampus and placed in a way such that it covered the spatial memory-relevant body and tail but not the head, which had been shown to be bilaterally associated with episodic memory (11) and verbal memory (28). T1-w magnetization-prepared rapid acquisition with gradient echo (MPRAGE) whole brain and upper cervical spinal cord images were acquired (TE = 2.32 ms, TR = 2,300 ms, flip angle = 8°, in-plane resolution 0.9 × 0.9mm², slice thickness = 0.9 mm) to explore injury-induced macrostructural changes within the hippocampus in SCI patients using voxel-based morphometry (VBM) (29).

Phase, frequency, and eddy-current correction as well as removal of motion-related artifacts were performed on the metabolite-cycled, non-water-suppressed MRS acquisitions using a motion compensation (MoCom) scheme in MATLAB 2020b (MathWorks, Natick, MA) (27). Spectra were quantified by linear combination model fitting using the fitting tool for arrays of interrelated datasets (FitAID) (30). The model basis set included simulated spectra of the following metabolites: aspartate, creatine (Cr), γ-aminobutyric acid (GABA), glucose, glutamate (Glu), glutamine (Gln), glutathione, glycine, glycerophosphorylcholine (GPC), lactate, myo-inositol (mI), N-acetylaspartate (NAA), N-acetylaspartylglutamate (NAAG), phosphocreatine (PCr), phosphorylcholine (PCho), phosphorylethanolamine, scyllo-inositol, and taurine. A macromolecular background (MMBG) signal was included in the basis set. It had been created based on a cohort average spectrum of the first 20 subjects by modeling an overall MMBG as a set of equally spaced Voigt lines (5 Hz spacing, 14 Hz Lorentz width, 10.6 Hz Gauss width) in addition to the metabolites’ basis spectra. Results are presented for total N-acetylaspartate (tNAA = NAA + NAAG), mI, choline-containing compounds (tCho = GPC + PCho), total creatine (tCr = Cr + PCr), and combined glutamate and glutamine (Glx = Glu + Gln) as these are relevant biochemicals that are known to be most reliably and reproducibly quantified in hippocampal MRS at 3 T (31, 32). The Cramér-Rao lower bound (CRLB) (33, 34) was calculated for each metabolite estimated. It represents the minimal error of fitting and is often used as an indicator of the data quality (33, 34) because it is affected by signal-to-noise, linewidth and model-based uncertainties. Here, the CRLBs are used in addition to a pure linewidth criterion as a way to filter out spectra of unusual low quality. The exclusion criterion from the CRLBs was that if absolute CRLBs of more than half of the evaluated metabolites were larger than 1.5 times the respective median value from the cohort (27), then the spectra were discarded. In addition, we excluded spectra with a Gaussian linewidth of more than 7.5 Hz in model fitting from our analysis. For absolute quantification, signal intensities were converted into millimolar (mM) concentrations based on parenchymal water derived from the TE series of unsuppressed water acquisitions in FiTAID assuming two water compartments (35). Metabolite concentrations ( M molar ) were estimated based on the following equation (34):

S M and S H 2 O indicate the metabolite and water signal intensities, as estimated in FitAID including the scaling factors for the respective number of protons per molecule. S H 2 O refers to the relaxation-corrected (from TE-series) signal of parenchymal water (i.e., excluding CSF). Metabolite relaxation corrections used relaxation times from literature (36, 37), where T2 values were multiplied by the factor 1.5 to accommodate the slower T2 signal decay for sLASER (31, 38). f G M and f W M refer to the volume fractions of gray matter (GM) and white matter (WM) in the voxel, which are needed to fine-tune the water content to convert to molar concentrations. Small individual differences with even smaller effect on the water content were neglected and values derived from literature (31, 39) were used throughout. Resulting f G M and f W M amounted to 0.67 and 0.33 for hippocampus, and 0.81 and 0.19 for the reference ROI, respectively. d G M and d W M are the GM- and WM-specific water content, taken from literature [0.78 for GM, 0.65 for WM (35)]. H 2 O molar refers to the molar concentration of water, which is 55.01 moles/L (34).

To investigate potential volumetric changes in the right hippocampus, VBM was performed in SPM12 (University College London, London, UK) using T1-w MPRAGE images (29). First, MPRAGE images were segmented into GM, WM, and CSF applying a unified segmentation method (40). This was followed by template generation for image registration with the Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra (DARTEL) algorithm (41). Based on this template, all subjects were spatially normalized to Montreal Neurological Institute (MNI) space representing the common space. Normalized images were smoothed using a Gaussian kernel with full width at half maximum of 3 mm. Next, an anatomical mask was created with the SPM anatomy toolbox (42) for the voxel-wise statistical analysis.

Statistical analyses were performed using RStudio software (RStudio PBS, version 1.4.1106). Voxel-wise statistical analysis of the right hippocampus VBM was conducted in SPM12 (29). An unpaired two-tailed t-test was used to explore the difference between the mean age of SCI patients and healthy controls. To test for differences in the distribution of sex and highest graduation, the Fisher’s exact test and the chi-squared test were applied, respectively. These tests were used as they both allow comparing the frequency of a categorial variable (sex and graduation) between groups. The threshold for significant differences was set to p < 0.05. SCI patients were compared to healthy controls regarding differences in immediate memory performance (t1), intermediate memory performance (t2), and delta score between both time-points using unpaired two-tailed t-tests or Wilcoxon rank-sum tests (Mann–Whitney U test) in case of a non-normal distribution (tested with the Shapiro Wilk test). Potential differences in memory performance were also explored between the two educational stages groups (elementary school or secondary school vs. high school or university). These tests were furthermore used to investigate differences in metabolite concentrations between SCI patients and healthy controls for the right hippocampus and the reference region. Besides the classical t-test and Wilcoxon rank-sum test, Bayesian unpaired t-test and Bayesian Mann–Whitney U test were conducted, which allow estimating the evidence in favor of the null hypothesis. If the Bayes Factor was above 3, substantial evidence in favor of the alternative hypothesis was assumed, and a Bayes Factor below 0.33 was judged as substantial evidence in favor of the null hypothesis. Bayesian statistic was performed in the software JASP (version 0.14.1, University of Amsterdam, Amsterdam, NL). Pearson correlation was performed to investigate the correlation between memory performance and age. Correlation between hippocampal metabolite concentrations and time since injury was assessed with Pearson correlation for tNAA, mI, and tCr, and with Kendall rank correlation (for non-normal distribution) for tCho and Glx. To investigate brain volume changes at the group level, voxel-wise statistical analysis (29) was performed in the right hippocampus. In SPM12 an unpaired t-test was conducted in which total intracranial volume (TIV), age, and sex were included as covariates of no interest. Family-wise error (FWE) correction was applied, and peak-voxel threshold was set to p < 0.05. A post-hoc power analysis was conducted to evaluate minimal metabolite concentration alterations that would have been detected with this study population using as assumptions n = 26 for patients and n = 18 for controls, normal data distribution for both cohorts, 80% of statistical power, within group standard deviation (SD) of 10% for both groups (as for instance found for tNAA or mI).

This study included 28 chronic traumatic SCI patients and 18 age-, sex-, and education-matched healthy controls. Their mean age ± SD was 47.8 ± 12.8 years and 49.4 ± 12.5 years, respectively (age range of 21–70 years). There was no difference in the mean age between the two groups (p = 0.67). Neither sex (p = 0.37) nor graduation levels (p = 0.50) were different between SCI patients and healthy controls. The median time since injury of SCI patients was 12.5 years with a minimum of 2 years and a maximum of 37 years. Eighteen patients were tetraplegic and 10 patients were paraplegic, as defined by the International Standards for the Neurological Classification of Spinal Cord Injury (ISNCSCI) (43). Twelve patients had a sensorimotor complete lesion according to the American Spinal Injury Association (ASIA) Impairment Scale (AIS) (AIS grade A) and 16 had an incomplete (AIS grades B-D) lesion (Table 1). None of the study participants with ALFI-MMSE assessment showed signs of dementia. Only one patient was at the border of mild cognitive impairment (MCI).

The visuospatial ‘Map’ task revealed no differences in the mean memory performance between SCI patients and healthy controls for immediate (t1) recall (21.0 ± 6.0 vs. 20.2 ± 6.8, p = 0.70, Figure 1A) or intermediate (t2) recall (19.4 ± 6.8 vs. 18.5 ± 8.0, p = 0.70, Figure 1B). Furthermore, Bayesian analysis resulted in Bayes Factors smaller than 0.33 for both time-points (BF₁₀ = 0.317 at t1, BF₁₀ = 0.317 at t2). There was also no difference in memory performance in the verbal ‘Text’ task between SCI patients and healthy controls at t1 (6.8 ± 4.0 vs. 8.1 ± 5.0, p = 0.38, Figure 1C) or t2 (5.6 ± 4.0 vs. 6.5 ± 4.8, p = 0.60, Figure 1D). The corresponding Bayes Factors were slightly above and below the threshold of 0.33, respectively (BF₁₀ = 0.350 at t1, BF₁₀ = 0.312 at t2).

The difference in memory performance over time, calculated as the score for intermediate (t2) recall minus the score for immediate (t1) recall, was not different for the ‘Map’ task (p = 0.98) or the ‘Text’ task (p = 0.50). Age was negatively correlated with the memory score for the immediate (t1) memory (R² = 0.14, p = 0.01) and intermediate (t2) memory (R² = 0.18, p < 0.01, Figure 2). There was no significant correlation between age and ‘Text’ memory score at the immediate (t1) time-point or intermediate (t2) time-point.

There was a significant difference in memory performance between the two educational stages groups (elementary school or secondary school vs. high school or university), irrespective of SCI. The group with a higher education performed significantly better on average in the ‘Map’ task at immediate (t1) recall (p = 0.04) and at intermediate (t2) recall (p = 0.01). This group also revealed higher memory performance in the ‘Text’ task at immediate (t1) recall (p = 0.03) and at intermediate (t2) recall (p = 0.04).

Representative metabolite spectra acquired in the right hippocampus (Figure 3A) of a healthy control and a SCI patient are shown in Figures 3B,C, respectively. MRS data of the reference region was missing for one participant as the scan had to be interrupted. The mean Cramer-Rao lower bound (CRLB) was below 0.5 mM for all metabolites of interest in the right hippocampus, this was the case even for metabolites that were not reported individually and were instead included as parts of lumped compounds due to overlapping metabolite patterns. Due to their greater fitting uncertainties, only the total contents of correlated summed metabolites were used in the statistical analyses. Figure 4 provides a visual representation of these findings.

The CRLBs were consistently about four times lower for the reference region due to the larger VOI (hence SNR) and smaller linewidth. No significant differences were observed between the CRLBs in SCI patients and healthy controls. Based on the exclusion criteria, three subjects (two SCI patients, one healthy control) were excluded from the right hippocampus analysis and two subjects (one SCI patient, one healthy control) were excluded from the reference region analysis.

The estimated concentrations and the group SD for all estimated metabolites in the hippocampus and reference region are listed in Supplementary Table 1.

There were no significant differences between SCI patients and healthy controls in the mean concentrations of tNAA (p = 0.53, Figure 5A), mI (p = 0.09, Figure 5B), tCr (p = 0.59, Figure 5C), tCho (p = 0.27, Figure 5D), or Glx (p = 0.41, Figure 5E) in the right hippocampus. A sub-group analysis comparing SCI patients with a sensorimotor complete lesion (AIS grade A, n = 11) against healthy controls did not reveal significant differences in the mean concentration of tNAA (p = 0.54), mI (p = 0.33), tCr (p = 0.36), tCho (p = 0.61), or Glx (p = 0.54). Similarly, SCI patients with a cervical lesion (tetraplegics, n = 17) did not show significant differences with healthy controls in the mean concentration of tNAA (p = 0.65), mI (p = 0.06), tCr (p = 0.97), tCho (p = 0.24), or Glx (p = 0.56). Within the SCI patients’ group, no significant correlation was found between neurological level of injury (NLI) and concentrations of tNAA (R = −0.19, p = 0.19), mI (R = −0.05, p = 0.72), tCr (R = 0.10, p = 0.47), tCho (R = 0.04, p = 0.79), or Glx (R = 0.04, p = 0.79) and between time since injury and concentrations of tNAA (R = 0.37, p = 0.08), mI (R = 0.24, p = 0.26), tCr (R = 0.26, p = 0.22), tCho (R = 0.14, p = 0.33), or Glx (R = −0.05, p = 0.75).

Quantification of metabolite concentrations in the posterior parietal lobe did not reveal significant differences in tNAA (p = 0.97), mI (p = 0.06), tCr (p = 0.87), tCho (p = 0.43), or Glx (p = 0.56) when comparing SCI patients with healthy controls.

Macrostructural VBM analysis of the right hippocampus did not show significant differences in hippocampal volume between SCI patients and healthy controls.