Differences in Cerebral Glucose Metabolism in ALS Patients with and without C9orf72 and SOD1 Mutations

Amyotrophic lateral sclerosis (ALS) is characterized by progressive loss of upper and lower motor neurons. In 10% of patients, the disorder runs in the family. Our aim was to study the impact of ALS-causing gene mutations on cerebral glucose metabolism. Between October 2010 and October 2022, 538 patients underwent genetic testing for mutations with strong evidence of causality for ALS and 18F-2-fluoro-2-deoxy-D-glucose-PET (FDG PET), at University Hospitals Leuven. We identified 48 C9orf72-ALS and 22 SOD1-ALS patients. After propensity score matching, two cohorts of 48 and 21 matched sporadic ALS patients, as well as 20 healthy controls were included. FDG PET images were assessed using a voxel-based and volume-of-interest approach. We observed widespread frontotemporal involvement in all ALS groups, in comparison to healthy controls. The degree of relative glucose metabolism in SOD1-ALS in motor and extra-motor regions did not differ significantly from matched sporadic ALS patients. In C9orf72-ALS, we found more pronounced hypometabolism in the peri-rolandic region and thalamus, and hypermetabolism in the medulla extending to the pons, in comparison to matched sporadic ALS patients. Our study revealed C9orf72-dependent differences in glucose metabolism in the peri-rolandic region, thalamus, and brainstem (i.e., medulla, extending to the pons) in relation to matched sporadic ALS patients.


Introduction
The multifaceted clinical syndrome of amyotrophic lateral sclerosis (ALS) presents with rapidly progressing neurodegeneration in motor and extra-motor regions and is usually fatal within 2-5 years after disease onset [1]. The disorder is characterized by varying phenotypic manifestations with degeneration of both upper motor neurons (UMN) in the motor cortex and lower motor neurons (LMN) in the brainstem and spinal cord [2].

Materials and Methods
The local Ethical Committee of the University Hospitals Leuven approved this study (s50354). Informed consent was obtained from all subjects in accordance with the Declaration of Helsinki.

Healthy Controls
The healthy control population (HC) (n = 20; mean age ± SD, 62.4 ± 6.4 y, 8M/12F) was the same as that used in a previous FDG PET study [21].

Patients
A total number 538 incident patients were recruited from the Neuromuscular Clinic in the University Hospitals Leuven between October 2010 and October 2022. All patients were diagnosed by an experienced neurologist specialized in neuromuscular disorders, using the revised El Escorial and Awaji-Shima criteria [30], and underwent a diagnostic 18 F-2-fluoro-2-deoxy-D-glucose-PET (FDG PET).
In 84 patients with ALS, a pathogenic mutation was found, known to be associated with ALS. Most commonly, we found a C9orf72 HRE (n = 48, 57%), followed by a mutation in the SOD1 gene (n = 22; 26%), a mutation in the TARDBP gene (n = 7; 8%), a mutation in the FUS gene (n = 5; 6%), and lastly, a dual C9orf72 and TARDBP gene mutation (n = 2; 2%). Given the limited sample size of the three latter groups, they were not considered for further analyses.

Propensity Score Matching
The aim of this approach was to balance the selected covariance between the cohorts of observational data to minimize the effect of confounding factors. From an initial group of 449 sporadic ALS patients, we performed propensity score matching (PSM) [31,32] using a logistic regression model based on the following six variables: age at FDG PET, sex, King's disease stage, diagnostic delay, the extent of lower vs. upper motor neuron involvement at diagnosis, and scanner type. Case-control matching for scanner type was performed, such that scanner type would not bias observed differences in the cerebral glucose metabolism. One-to-one pair matching was performed without replacement, using a caliper of 0.20.
The algorithm matched 48 C9orf72-associated ALS patients and 21 out of 22 SOD1associated ALS patients to an equal number of sporadic ALS patients. One patient displayed a distinct phenotype of rapidly progressive respiratory onset ALS at a young age. A similar phenotype has not been observed before in our cohort of sporadic ALS patients.

Assessment of Clinical Functioning
The clinical assessment at diagnosis and revised Amyotrophic Lateral Sclerosis Functional Rating Scale (ALSFRS-r) were utilized to perform King's clinical staging [33]. PET acquisitions were performed in list mode thirty minutes after injection. Images were acquired dynamically for 30 min on the ECAT HR+ (3D mode) and for 15-20 min on the HiRez and Truepoint (3D and listmode). During the acquisition, the patient's head was immobilized with a vacuum pillow or headholder. All images were corrected for attenuation, scatter, and decay.

Image Analysis
On the ECAT HR+, images were corrected for attenuation with the attenuation map, acquired with a transmission scan using 68 Ge/ 68 Ga rod sources, and projection data were reconstructed using 3-dimensional filtered back projection with a Hanning post-filter. On the HiRez and Truepoint, images were reconstructed using ordered-subset expectation maximization (OSEM) with 5 iterations over 8 subsets and 3 iterations over 21 subsets, respectively, with an 8 mm Gaussian kernel.

FDG PET Image Analysis
All scans were spatially normalized to Montreal Neurological Institute (MNI) space using a FDG PET template (SPM12; Welcome Trust Centre for Neuroimaging), implemented in Matlab (R2020b, The MathWorks Inc., Natick, MA, USA), and nonrigid registration with 16 iterations, in voxels of 2 mm × 2 mm × 2 mm. Isotropic Gaussian smoothing with full width at half maximum of 8 mm was performed, followed by proportional scaling to the mean uptake value within a brain mask, obtained by applying a relative threshold masking of 0.8 in the healthy control group.
Voxel-based findings were corroborated with an atlas-based volume-of-interest (VOI) analysis using the N30R83 Hammers probabilistic atlas and the Automated Anatomical Labeling (AAL) single-subject atlas implemented in PNEURO (PMOD, v3.7, PMOD Inc., Zurich, Switzerland) [34]. The AAL single-subject atlas was used to delineate the midbrain, pons, and medulla, enabling a more detailed delineation of the entire brainstem [35]. Standardized uptake value ratio (SUVR) values were calculated, by scaling the mean uptake in VOIs to the average grey matter activity to obtain relative metabolic activity.

Statistical Analysis
General statistics were done using SPSS 28.0 (SPSS Inc. Chicago, IL, USA). Data are presented as median (interquartile range). Normality of the distributions of the clinical data was assessed with a Shapiro-Wilk test (α = 0.05). For statistical comparisons, we used the χ2 test for categorical variables, or Kruskal-Wallis test for continuous variables. Survival was estimated using the Kaplan-Meier method and compared using log-rank tests, with Bonferroni-corrected post hoc pairwise comparisons.
For the voxel-based analyses, we used an analysis of variance (ANOVA) model, using the SPM12 toolbox, implemented in Matlab, to assess the effect of genetic status on cerebral glucose metabolic uptake. Statistical significance for the voxel-based analyses was set at p uncorr < 0.001, combined with a p FWE < 0.05 at the cluster level, unless otherwise specified.
VOI-based independent Mann-Whitney U tests were conducted to determine if there were statistically significant differences in relative metabolic activity between C9orf72-ALS and C9orf72-matched sALS, and between SOD1-ALS and SOD1-matched sALS. We also compared each disease group to the healthy control group. We evaluated 37 regions, and the Benjamini-Hochberg procedure was used to control the false discovery rate (FDR) at 0.05 [36].
As we know that the disease progression rate between C9orf72-ALS and SOD1-ALS differs significantly, we employed propensity score matching to obtain "artificial sporadic ALS control cohorts", matched to our C9orf72-ALS and SOD1-ALS cohorts. A comparative analysis did reveal a statistically significant difference in diagnostic delay (p < 0.001) between different ALS disease groups. Detailed demographics and clinical characteristics of each group are reported in Table 1.

Md (IQR)
C9orf72-ALS As the disease progression rate between C9orf72-ALS and SOD1-ALS differs substantially, we compared survival rate between all four groups. A log rank test (Mantel-Cox) revealed that the survival distribution for the four ALS patient groups was significantly different, χ2(3) = 11.130, p = 0.011 ( Figure 1). Pairwise comparisons revealed a significant difference between the unmatched genetic groups, SOD1-ALS and C9orf72-ALS, χ2(1) = 13.922, p = 0.0002 ( Figure 1, Table 1 A log rank test revealed a significantly different distribution in survival for the different ALS groups, χ2(3) = 11.130, p = 0.011. Log rank pairwise comparisons were done with a Bonferroni correction, with statistical significance accepted at the p < 0.0125 level. There was a significant difference in survival distribution for SOD1-ALS and C9orf72-ALS patients (p = 0.0002). A log rank test revealed a significantly different distribution in survival for the different ALS groups, χ2(3) = 11.130, p = 0.011. Log rank pairwise comparisons were done with a Bonferroni correction, with statistical significance accepted at the p < 0.0125 level. There was a significant difference in survival distribution for SOD1-ALS and C9orf72-ALS patients (p = 0.0002).

Cerebral Glucose Metabolic Changes between Different Genotypes
Voxel-based comparative analyses between C9orf72-ALS and SOD1-ALS patients revealed clusters of relative hypermetabolism in the brainstem (i.e., pons, extending to the midbrain) and cerebellum of C9orf72-ALS patients and relative hypometabolism bilaterally in the basal ganglia of SOD1-ALS patients ( Figure 2 and Table S1).

Cerebral Glucose Metabolic Changes between Different Genotypes
Voxel-based comparative analyses between C9orf72-ALS and SOD1-ALS patients revealed clusters of relative hypermetabolism in the brainstem (i.e., pons, extending to the midbrain) and cerebellum of C9orf72-ALS patients and relative hypometabolism bilaterally in the basal ganglia of SOD1-ALS patients ( Figure 2 and Table S1). The results of the VOI-based Mann-Whitney U test (Table 2, Figure S3) corroborated our findings at the voxel level in the basal ganglia (i.e., lentiform nucleus), brainstem (i.e., pons and midbrain), and cerebellum. In addition, the VOI-based comparative analysis revealed a significant difference in FDG uptake between C9orf72-ALS and SOD1-ALS Figure 2. T-statistical map detailing patterns of relative hypometabolism (yellow) and hypermetabolism (blue) in C9orf72-ALS patients, when compared to SOD1-ALS patients, thresholded at p uncorr < 0.001 at voxel level, and p FWE < 0.05 at cluster level, while correcting for age, sex, scanner type and King's disease stage. Clusters are overlaid on a T1-weighted template.

Differences in Cerebral Glucose Metabolism between ALS Groups and Healthy Controls
We compared all ALS groups to a group of healthy controls. This revealed significantly reduced glucose metabolism in fronto-insular, temporal, parietal, and occipital cortices, the amygdala, basal ganglia, and thalamus in all ALS groups (Figures S1 and S2, Tables S2-S4). Hypermetabolism was observed in the cerebellum in C9orf72-ALS, SOD1-ALS, and C9orf72-matched sALS patients and in the brainstem in C9orf72-ALS patients.

Effect of Genetic Status on Cerebral 18 F FDG Uptake
Voxel-based comparative analyses between C9orf72-ALS and C9orf72-matched sALS patients, and between SOD1-ALS and SOD1-matched sALS patients, revealed two significant clusters of relative hypometabolism and one cluster of relative hypermetabolism in C9orf72-ALS patients in relation to C9orf72-matched sALS patients.
This study demonstrated that C9orf72-ALS was associated with reduced glucose metabolism in the peri-rolandic region, extending to the precuneus, and the right thalamus, as well as increased glucose metabolism in the brainstem (i.e., medulla, extending to the pons), in relation to the C9orf72-matched sALS group (Figure 3; Table S5). Voxel-based analyses between SOD1-ALS and SOD1-matched sALS groups revealed no significant clusters of relative hypo-or hypermetabolism. The results of the VOI-based Mann-Whitney U test (Tables 3 and 4, Figure S3) corroborated our findings at the voxel level.  Voxel-based analyses between SOD1-ALS and SOD1-matched sALS groups revealed no significant clusters of relative hypo-or hypermetabolism. The results of the VOIbased Mann-Whitney U test (Tables 3 and 4, Figure S3) corroborated our findings at the voxel level. Table 3. Volume-of-interest (VOI)-based Mann-Whitney U test of 18F-FDG SUVR between C9orf72-ALS (n = 48) patients and matched sporadic ALS ( C9orf72-matched sALS; n = 48) patients. FDR-corrected p-values < 0.05 for multiple testing were considered significant. Abbreviations: C9orf72-ALS, C9orf72-associated ALS; IQR, interquartile range; Md, median; sALS = sporadic ALS; SUVR = standardized uptake value ratio.

Discussion
The primary objective of this study was to investigate the effect of major gene mutations causative for ALS on cerebral glucose metabolism. The considerable variability in the phenotypic expression of ALS that is reflected in familial and sporadic ALS [8] makes comparative analyses between subgroups of ALS patients challenging.
To ensure that any differences in cerebral glucose metabolism between genotypes and sporadic ALS patients were not driven by population diversity, we matched C9orf72-ALS and SOD1-ALS patients to our cohort of sporadic ALS patients, using propensity score matching.
Comparing two genetic subtypes of ALS (i.e., C9orf72-ALS and SOD1-ALS) and their respective matched cohorts of sporadic ALS patients (i.e., C9orf72-matched sALS and SOD1-matched sALS) readily revealed more pronounced hypometabolism in the peri-rolandic region, extending to the precuneus, and thalamus, as well as hypermetabolism in the brainstem (i.e., medulla, extending to the pons), in C9orf72-ALS, when compared to C9orf72-matched sALS.

The Effect of C9orf72 Status on Cerebral Glucose Metabolism
Previous research has produced contradictory research findings about the glucose metabolic signature in patients with a C9orf72 HRE [21,[26][27][28]. The study by Van Laere et al. produced no significant differences between C9orf72-ALS and sporadic ALS [21], while the study by Diehl-Schmid et al. reported reduced glucose metabolism confined to the thalamus [28] in C9orf72-FTD and sporadic FTD.
Another study by Cistaro et al., on the other hand, described widespread reduced glucose metabolism in the left frontal and superior temporal cortices, the bilateral anterior and posterior cingulate cortex, insula, caudate, and thalamus, in C9orf72-ALS in comparison to sporadic ALS, and relative hypermetabolism in the midbrain, bilateral occipital cortex, globus pallidus, and left middle and inferior temporal cortices [26].
To overcome the challenge of selection bias that might occur due to the nature of retrospective studies and to minimize the effect of confounding factors, we studied the cerebral glucose metabolic pattern of 48 ALS patients with a C9orf72 HRE in relation to a group of matched sporadic ALS patients, using PSM.
The clusters of significantly reduced glucose metabolism in the peri-rolandic region, extending to the precuneus, and thalamus suggest that both the thalamus and the perirolandic region play a key role in C9orf72-mediated disease [26][27][28]37]. Reduced FDG uptake in C9orf72-ALS patients can reflect multiple neurobiological underpinnings, such as reduced excitability [38], synapse loss [39], and neurodegeneration. Mitochondrial dysfunction [40] may also contribute to the observed changes, as a recent study employing the [ 18 F]2-tert-butyl-4-chloro-5-2H-pyridazin-3-one (BCPP-EF) radioligand for mitochondrial complex demonstrated that mitochondrial complex I reduction can be linked to a range of downstream neurodegenerative processes, such as hypometabolism. Follow-up PET studies with [ 18 F]BCPP-EF and [ 18 F]FDG would enable in vivo investigation of dysfunction of mitochondria and its relation to hypometabolism in C9orf72-ALS [41].
Thalamic involvement may not only be one of the earliest features of C9orf72-mediated disease, as demonstrated by PET studies in presymptomatic C9orf72 HRE carriers [37,42], but also one of the regions that is most severely affected by the disorder.
In comparison with healthy controls, we found relative hypermetabolism in the cerebellum, the medial temporal lobe, and the medulla, extending to the pons in C9orf72-ALS. Cerebellar and medial temporal hypermetabolism appears relatively prominent but not specific to C9orf72-ALS, as we found significant changes in the medulla (extending to the pons) when compared to matched sporadic ALS patients.
The neurobiological processes that underlie hypermetabolism in these regions remain to be elucidated. Various theories have been proposed to explain this observation, such as the normalization procedures used [21], recruitment of neural networks as a localized inflammatory response [21,24,25,43], or even a compensatory neuronal mechanism as the frontal regions (including the primary motor cortex) degenerate [44].
Indeed, unlike previous observations in C9orf72-associated FTD [27,28], we observed that the medulla was significantly hypermetabolic in C9orf72-ALS patients when compared to C9orf72-matched sALS patients. A human autopsy study already demonstrated that C9orf72-ALS cases had more severe microglial pathology in the medulla compared to ALS cases without a C9orf72 hexanucleotide repeat expansion [45]. Relative hypermetabolism in this region may thus represent neuroinflammation associated with activated astrocytes or microglia [46] This significant hypermetabolism in the medulla may be associated with a more aggressive disease phenotype in patients with a C9orf72 hexanucleotide repeat expansion [47][48][49], as a recent study with FDG PET/MRI demonstrated that medulla hypermetabolism is associated with shortened survival [50].

The Effect of SOD1 Status on Cerebral Glucose Metabolism
In our study, we found no significant differences in cerebral glucose metabolism between SOD1-ALS patients and matched sporadic ALS patients. Moreover, a comparison to healthy controls revealed a pattern of widespread relative hypometabolism in frontal, temporal, parietal, and subcortical regions (including basal ganglia and thalamus) in SOD1-ALS, resembling the glucose metabolic pattern observed in sporadic ALS [21,23,51,52].
Our findings are in line with several postmortem studies that described lesions of the (extrapyramidal) motor and nonmotor systems in SOD1-positive cases, including subcortical structures as well as frontotemporal cortices [53,54]. A recent study also revealed that behavioral changes are more commonly found in patients carrying SOD1 variants than previously thought [18].
In past years, a few in vivo neuroimaging studies have been conducted in SOD1associated ALS. Two volumetric imaging studies demonstrated similar cerebral volumetric measurements between SOD1-ALS patients and sporadic ALS patients [55,56]. Two PET imaging studies using [ 11 C]Flumazenil and [ 11 C]WAY100635 described a more focal loss of GABAergic cortical inhibition and serotonergic binding differences in patients with a SOD1 D91A mutation than in sporadic ALS patients [57,58]. More recently, a study employing 11 C-PK11195 PET revealed widespread, diffuse cortical and subcortical increased glial activation in SOD1-associated ALS [59], comparably to what was observed in sporadic ALS [60].
A recent FDG PET study by Canosa et al. described a cluster of relative hypermetabolism in the peri-rolandic region in SOD1-ALS patients in relation to sporadic ALS patients and healthy controls [29]. Differences found between both FDG PET studies (i.e., our study and the study by Canosa et al. [29]) may be explained by the differences between the SOD1-ALS cohorts. A quarter of SOD1-ALS patients in the study by Canosa et al. had pure lower motor neuron involvement [29], a phenotype we did not observe in our cohort (Table S6). Moreover, both cohorts differed in the distribution of genetic variants, as in our cohort, the majority of SOD1 patients harbored an SOD1 D91A (45%; n = 10) or SOD1 G94C genetic variant (27%; n = 6) ( Table S6).
Differences may also be attributed to the use of propensity score matching to minimize the impact of confounding variables. Unlike sequential bivariate matching, this technique accounts for potential multivariate interactions between the matching variables, which may introduce bias in the participant selection [31].
Nevertheless, large multicenter approaches investigating the cerebral glucose metabolism associated with different pathogenic SOD1 variants are needed to better understand the variability in disease presentations of different SOD1 mutations.
Our study has several limitations. Firstly, subjects were scanned on three different PET cameras. Slight differences associated with CT-based versus 68Ge/Ga-based attenuation correction (especially in infratentorial areas), as well as spatial resolution may have contributed to an increased variance. However, the spatial resolution of the three scanners is highly similar. Nevertheless, case-control matching for scanner type was performed, such that scanner type would not bias observed differences in the cerebral glucose metabolism.
Secondly, the absence of individual MR images for each subject did not allow for partial volume correction. Cerebral glucose metabolism measured by PET may capture both implicit cellular metabolic functions and atrophy. However, our findings on metabolic signatures are in line with those reported in previous studies [21,23,24,[26][27][28].
Thirdly, we tried to correct for confounding factors using PSM, but bias exposures are inevitable due to the respective nature of the study. PSM can account for multivariate interactions, minimizes selection error, and is better suited for smaller sample sizes [31,61].
Fourthly, due to the small sample size, we excluded the following carriers of a monogenic mutation causative for ALS in our cohort (TARDBP, n = 7; FUS, n = 5; TARDBP and C9orf72, n = 2) from all analyses.
Fifthly, we observed no significant differences between SOD1-ALS and sporadic ALS patients. We cannot exclude that between-group differences may have been washed out, due to within-group heterogeneity. This may have also reduced the power of our analyses. However, a post hoc power analysis on our cortical VOIs revealed that there was sufficient power, which suggests that the lack of significant differences may be due to a small effect size rather than to a shortage of power.
Finally, there were some discrepancies between VOI-based Mann-Whitney U tests and our voxel-based findings, possibly due to VOI dilution or our approach to multiple comparisons correction (i.e., familywise error rate versus false discovery rate (FDR)). Due to the lack of contemporaneous MRIs, the VOIS included in this analysis comprise different distributions of tissue classes at the individual level, which may potentially confound the interpretation. Secondly, the FDR is a less conservative approach to multiple comparisons correction and may result in a false-positive finding.

Conclusions
In conclusion, our findings demonstrate that the distribution of metabolic activity, using FDG PET, in SOD1-associated ALS resembles the changes observed in matched spo-radic ALS patients, when compared to healthy controls. To enable more sensitive analyses, large multicenter approaches may aid in investigating the cerebral glucose metabolism associated with different pathogenic SOD1 variants.
Patients with a C9orf72 HRE, on the other hand, did present with significant differences in FDG uptake in the peri-rolandic region, thalamus, and brainstem (i.e., medulla, extending to the pons) compared to matched sporadic ALS patients and healthy controls. The results from the current study extend previous findings by identifying that, in addition to the thalamus, the peri-rolandic region and medulla play a key role in C9orf72-mediated ALS. Our findings support the hypothesis that ALS patients with a C9orf72 HRE demonstrate more central nervous system involvement than sporadic ALS and SOD1-associated ALS patients do.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/cells12060933/s1; Figure S1: T-statistical map detailing patterns of relative hypometabolism (yellow) and hypermetabolism (blue) (A) in C9orf72-ALS, when compared to healthy controls, and (B) in C9orf72-matched sALS, when compared to healthy controls; Figure S2: T-statistical map detailing patterns of relative hypometabolism (yellow) and hypermetabolism (blue) (A) in SOD1-ALS, when compared to healthy controls, and (B) in SOD1-matched sALS, when compared to healthy controls; Figure S3: Regional relative glucose metabolism in key volumes-of-interest regions in four ALS groups; Table S1: Global Cluster peak coordinates variants and statistics of relative glucose metabolism comparing C9orf72-ALS with SOD1-ALS; Table S2: Global Cluster peak coordinates variants and statistics of relative hypo-and hypermetabolism when comparing groups of ALS patients (i.e., SOD1-matched sALS, C9orf72-matched sALS, SOD1-ALS, and C9orf72-ALS) to healthy controls; Table S3: Volume-of-interest (VOI)-based Mann-Whitney U tests of 18 F-FDG SUVR between healthy controls and C9orf72-ALS (n = 48) and matched sporadic ALS ( C9orf72-matched sALS; n = 48) patients; Table S4: Volume-of-interest (VOI)-based Mann-Whitney U tests of 18 F-FDG SUVR between healthy controls and SOD1-ALS (n = 22) and matched sporadic ALS ( SOD1-matched sALS; n = 21) patients; Table S5: Global Cluster peak coordinates variants and statistics of relative glucose metabolism comparing C9orf72-ALS with C9orf72-matched sALS; Table S6: Clinical and genetic data of ALS patients with a SOD1 gene mutation.