We registered a prospective study protocol in the International Prospective Register of Systematic Reviews (PROSPERO, CRD42022373146, https://www.crd.york.ac.uk/PROSPERO/) and used the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines for reporting (26).

We searched PubMed and Ovid EMBASE for relevant publications from inception up to December 19, 2022. See Supplementary Table 1 for the search strings in each of these databases.

We included original publications that reported on any structural brain or spinal cord MRI outcome in MND animal models. Conference abstracts, non-English articles, and publications which reiterated previously reported quantitative data were excluded. Reviews were excluded but retained as potential sources for additional records. Reference lists of these reviews were screened for additional eligible publications.

Titles and abstracts of studies were screened for their relevance in the web-based application Rayyan (27) by two independent reviewers followed by full-text screening. From eligible full texts, the following data was extracted by two independent reviewers: title, authors, publication year, journal, MND model, number of animals in the treatment and control groups, MRI static magnetic field strength, and main findings related to structural neuroimaging.

Risk of bias was assessed against a 3-item checklist according to the consensus statement for good laboratory practice in the modeling of stroke (sample size calculations provided, reporting of animal welfare, statement of a potential conflict of interest) (28), as well as four items on reporting any measure of randomization or blinding (29).

The main objective of this study was to systematically summarize the available evidence on structural CNS MRI features in ALS animal models. Frequent MRI features include brain and spinal cord atrophy, signal changes in brain stem motor nuclei and the motor cortex as well as breakdown of the blood-brain barrier (Table 1). In the following paragraphs, we will compare this phenotype with MRI features of human ALS.

Based on the findings of our systematic review, ALS animal models seem to feature several imaging signs reminiscent of human ALS (Table 1). Among these features is the volume loss of CNS structures with progressive disease. Atrophy in both the motor cortex (37) and the spinal cord (37, 40, 43) has been reported in ALS animal models, similar to the human imaging phenotype (20, 53, 54), which could correspond to the underlying decline of the upper and lower motor neurons (14). These similarities between the human and animal imaging phenotype are particularly interesting since most eligible animal studies used the G93A-SOD1 model thus mimicking familial ALS, a rare clinical phenotype constituting around 10% of ALS patients. It is also noteworthy that, similar to the human population (20), a wide and not always consistent array of CNS structures have been reported to be affected by volume loss in animal models. For example, motor cortex atrophy has not been consistently shown in ALS animal models (38). It is likely that different methodological approaches for the quantification of atrophy patterns between animal studies is in part responsible for these inconsistencies: This has been emphasized by a human study in ALS-FTD patients which found variable atrophy patterns when comparing different software to assess cortical volumes (FSL, FreeSurfer, and SPM) (84). Further confounders could be technical parameters such as intra-/inter-scanner variability and physiological factors such as hydration state of animals during imaging [reviewed in (85)].

ALS rodent models can present with T2 signal changes in the CNS, potentially corresponding to axonal degeneration (23). In rodents, these signal alterations seem to commonly affect brain stem motor nuclei (21, 22). In ALS patients, T2 signal changes are also commonly observed (20, 77), albeit at different locations, i.e., mostly along the corticospinal tract (Figure 1).

Abnormal iron deposition in the motor cortex and spinal cord has been reported by some rodent ALS studies, measured by T2^*-based MRI approaches (35, 37). Although respective publications did not include pictorial examples of iron deposition within the motor cortex, this feature could correspond to the “motor band sign” (linear motor cortex hypointensity) which is commonly observed in the motor cortex of ALS patients on T2^*-based sequences (Figure 1). In ALS, these signal drops seem to correspond to astro- and microglia iron deposition within deep layers of the motor cortex (86).

One imaging feature which seems more specific to rodent ALS models is breakdown of the blood-brain barrier, as visualized by gadolinium enhancement in periventricular and hippocampal regions (22). However, although gadolinium enhancement is not observed in the clinical setting in ALS, several lines of evidence demonstrate damage to the blood–brain and blood-spinal cord barrier in ALS [reviewed in (87)]. Such vascular changes seem to include alterations of tight junction proteins (88) and can be observed already early in the disease process (89). Structural MRI features of preclinical ALS models are summarized in Table 1, alongside with MRI features of human ALS.

To assess the external validity of ALS animal models, we focused our analysis on structural brain and spinal cord MRI features. However, other disease aspects such as patterns of physical disability or also more advanced MRI methods like diffusion-tensor imaging, which are able to more specifically reflect pathogenic disease processes, might enable a more comprehensive comparison between experimental and human phenotypes.

A genuine limitation of this systematic review is that only a limited number of studies employing MRI in ALS animal models was eligible. As a result, it is difficult to map imaging phenotypes of less commonly used ALS models such as cycad toxins or wobbler mice or even for different SOD1^G93A mutants. It is possible that certain ALS rodent models might mimic specific human imaging phenotypes better than others (36), similarly to the situation in experimental autoimmune encephalomyelitis (EAE)—a commonly used animal model for multiple sclerosis (90).

Finally, although seven of the eligible publications tested a putative therapeutic intervention for ALS, no corresponding human MRI studies could be identified. Correlating the impact of therapeutic interventions on neuroimaging phenotypes between rodent models and humans would further enhance understanding of the translational value of experimental ALS models.