1) General study characteristics
Eligible publications
In total, 10,571 original publications were retrieved from our comprehensive database search. After automated deduplication as well as abstract and title screening, 521 publications were eligible for full-text search. After screening the full text of these publications, 223 publications (4% of deduplicated references) were included for qualitative synthesis and 21 publications (0.4%) for the quantitative synthesis (Figure 1).
Diseases of eligible publications
Of the eligible publications, 191 investigated neuroimaging findings in ALS (n=7235 subjects), 35 in ALS-FTD or FTD-MND (n=1808 subjects), 14 in PLS (n=272 subjects), 3 in familial ALS (i.e., with mutations in SOD1), 1 in monomelic amyotrophy (n=109 subjects), 1 in O’Sullivan–McLeod syndrome (n=7 subjects), 1 in pre-symptomatic carriers of C9orf72 (n=15 subjects) and 1 in ALS plus (n=30 subjects). These publications together included a total of 5704 control subjects.
Risk of bias assessment and data mining
A majority of studies showed a low risk of bias for the selection domain, i.e., whether patients and controls were defined according to acknowledged diagnostic criteria; see Supplementary Table 3. Many publications did not report on adjusting their statistical analyses for subject age, sex, or other potential confounders (comparability domain), thus potentially inducing biases.
The R text mining function for risk of bias assessment in animal studies performed with high accuracy on all items: reporting any measure of randomization (recall 96.9%) or blinding (100%), animal studies conducted according to animal welfare guidelines (93.8%), statement of a potential conflict of interest (90.6%), prior sample size calculation (96.9%), and study conducted in accordance with the ARRIVE guidelines (100%).
2) Magnetic resonance imaging acquisition and assessed imaging biomarkers
MRI findings are summarized in Table 1. The most commonly assessed imaging biomarker in motor neuron diseases were whole-brain or regional CNS volumes, i.e., white and/or gray matter atrophy (156 publications) or spinal cord atrophy (6 publications). Another frequently assessed MRI feature was hyperintensities along the corticospinal tracts (CST) on T2-weighted (T2w) imaging (40 publications) or proton density-weighted (PDw) imaging (10 publications). Less commonly assessed imaging biomarkers were signal loss in the motor cortex on susceptibility-weighted (SWI) or T2*-weighted (T2*w) imaging, also termed the “motor band sign” (16 publications), motor cortex hypointensities (on T1w or T2w imaging, 5 publications), quantitative susceptibility mapping (2 publications), iron deposition (based on T2*w contrast, 2 publications), cerebral microbleeds (1 publication), microinfarcts (1 publication), and ADC values (1 publication).
Out of 223 eligible publications, most publications (202, 91%) acquired MRI at conventional field strengths, i.e., ≤ 3 tesla (T). 7 publications acquired MRI at 7T (3%), including 2 publications acquiring at both 7 and 3T. The remaining 16 publications did not report on the used static magnetic field strength.
CNS atrophy in motor neuron diseases
A majority of publications (166 out of 223; 74%) also reported whole-brain or regional CNS volume measures. Together, these studies demonstrated volume loss in a wide variety of CNS regions as assessed by a plethora of different automated and manual volumetric methodologies. Therefore, we restricted the analysis to studies employing automated volumetric segmentation methods and comprising a healthy control group as comparator; more comprehensive study findings are reported in Supplementary Table 5.
Whole-brain and spinal cord atrophy
A reduced total brain volume [18] or reduced brain parenchymal fraction (BPF) have been reported in ALS patients [19, 20]. One study only noted BPF reduction in ALS-FTD but not in non-demented ALS [21]. One longitudinal study showed higher rates of brain atrophy (around 1.4% per year) in C9orf72+ ALS-FTD patients [22]. In ALS, several publications showed frontal and temporal lobe atrophy [23, 24] as well as longitudinal temporal lobe atrophy [24]. In ALS/bvFTD [25] and C9orf72+ ALS [26, 27], cerebellar atrophy has been described. This was not confirmed in ALS patients in another study [28]. Several publications also described hippocampal volume loss in ALS [29, 30].
In ALS, brain stem atrophy has been shown cross-sectionally [31] and longitudinally [32]. Similarly, reduced spinal cord volumes/cross-sections have been observed in ALS [33, 34].
Strategic cortical structures
Motor cortex thinning has been described in both ALS and PLS [35, 36], also in a longitudinal study [37]. However, two publications did not confirm this finding [18], one of them only observing motor cortex thinning in lower or upper motor neuron dominant ALS phenotypes but not in classical ALS [38]. Two publications also showed cortical volume loss in both pre- and postcentral gyri [39, 40]. Yet, several publications did not observe a difference in cortical volumes between ALS and healthy controls [41-44]. One study also described decreased cortical thickness in prefrontal regions [45].
Strategic subcortical gray matter structures
Thalamic volume loss has been described by several publications and in different clinical phenotypes, i.e., ALS/bvFTD [25, 29], C9orf72+ ALS [26], and FTD-MND [46]. Longitudinal atrophy of the basal ganglia has been observed in sporadic ALS[47]. One study in ALS patients described longitudinal atrophy of several strategic subcortical gray matter structures, including the thalamus, caudate nucleus, putamen and amygdala [48]. However, another longitudinal study did not find changes in gray matter volumes in ALS [49]. Other cross-sectional publications were also not able to find differences in subcortical volumes in ALS [41-44].
Strategic white matter structures
In sporadic ALS, corpus callosum volume loss has been described in one study [47], while another failed to corroborate the finding [50]. Yet another study reported subregional corpus callosum volume loss in PLS but not in ALS [51]. Volume loss of the CST has been observed in C9orf72+ ALS [27] and sporadic ALS [47]. One study only described minimal loss of white matter in ALS [52].
Disease phenotypes
Due to the various clinical phenotypes/genotypes of MND, several publications assessed distinctive atrophy features between different classes.
Overall, C9orf72+ MND seems to show more pronounced atrophy compared to C9orf72- MND, e.g., in cortical brain regions [53-55] (such as the motor cortex [56]) or the thalami [54-57]. Along these lines, a 5-year follow up study observed more precuneal atrophy in C9orf72+ compared to C9orf72- [58]. However, C9orf72+ MND patients showed relative sparing of insular, orbitofrontal, anterior cingulate and temporal pole regions. One study showed a partly overlapping atrophy pattern between these two genotypes with the C9orf72+ patients showing volume loss in the accumbens nucleus and C9orf72- patients in the thalami and putamina[59]. Interestingly, also presymptomatic C9orf72+ carriers above 40 years of age show more pronounced thalamic, cerebellar, parietal and temporal volume loss [60]. This patient subgroup also displays higher cervical spinal cord volume loss [61].
Limb onset ALS patients have been shown to have lower cortical volumes in the limb part of the motor homunculus and similarly the corresponding regions for bulbar onset ALS [62]. Furthermore, limb onset ALS patients seem to show volume loss in adjacent pre- and postcentral regions. In contrast, bulbar onset ALS seems to have more widespread volume loss also extending to the bilateral frontotemporal and left superior temporal and supramarginal gyri [63].
ALS and PLS seem to present with distinct atrophy patterns: Whereas ALS patients show more volume loss in the postcentral gyrus, lateral parts of the primary motor cortex, genu of the corpus callosum, amygdala and putamen, PLS patients have an atrophy predominance in the medial parts of the primary motor cortex, splenium of the corpus callosum, cerebellum and thalami [64, 65]. One study only described corpus callosum volume loss in PLS but not in ALS [51]. Intriguingly, pre-PLS subjects also show subtle thinning of the right precentral gyrus [66].
Partly overlapping but also distinct imaging phenotypes based on the clinical phenotype along the ALS-FTD spectrum are also evident. BvFTD and non-fluent variant primary progressive aphasia (nfvPPA) patients show predominant motor cortex and CST degeneration [67]. C9orf72+ ALS-FTD patients show widespread extra-motor pathology and precentral gyrus atrophy compared to ALS patients without cognitive disability. Both bvFTD and ALS exhibit volume loss in orbital and cingulate cortices, CST and corpus callosum, but bvFTD patients have higher orbitofrontal and frontomedial volume loss whereas ALS patients show higher atrophy motor pathways in ALS [68]. One study also found widespread volume loss in frontal and temporal regions in FTD-ALS but no volume loss in ALS [69]. A more pronounced volume loss in various motor and premotor cortices in ALS-FTD compared to ALS has been confirmed by another study [70].
Of note, while 3 publications did include familial ALS patients [71-73], no imaging data specific for this sub-population was reported.
Methodology for CNS brain volumetry
It is noteworthy that the eligible publications employed a plethora of different approaches to assess whole/regional CNS volumes. Out of 166 publications assessing CNS volumetry, most commonly used automated approaches were SPM (various versions, 44 publications, 27%), FreeSurfer (36 publications, 22%) and FSL (35 publications, 21%). Twenty-four publications used visual rating of atrophy or manual segmentation of regions of interests (14%). Five publications did not report how they quantified CNS volume loss.
Due to considerable heterogeneity in the methodological approach to quantify whole and regional CNS volumes, as well as the various clinical MND phenotypes, we deemed it unfeasible to perform a proper meta-analysis of atrophy patterns of MND patients.
Corticospinal tract hyperintensities
Hyperintensities have been observed along the entire CST projection (Figure 2), i.e., in the corona radiata, internal capsule, cerebral peduncles and pons [74]. The diagnostic role of these CST hyperintensities, as seen on T2w, PDw or T2*w imaging, is conflicting. Some publications found them to be present at higher proportions in ALS patients compared to healthy controls [75], while others failed to confirm this finding [76, 77]. Interestingly, CST signal abnormalities were more common in advanced stage ALS (63%) and PLS (72%) and less commonly in early stage ALS (17%) [78]. Yet the proportion of these CST signal abnormalities among ALS patients varies substantially [40, 79, 80].
The variability of CST signal abnormalities between differnt MND phenotypes is less clear. One study, including 21 non-demented ALS patients and 3 demented ALS patients, found 6/21 of non-demented ALS patients (29%) with CST hyperintensities, while none of the demented ALS patients demonstrated this imaging feature [81]. Of note, one study, comprising 122 non-ALS patients, confirms that these CST hyperintensities are not an ALS-specific feature and that their frequency increases with age [82].
One study in ALS found a diagnostic specificity of 76% and a sensitivity of 48% for this sign, albeit with higher specificity in the subcortical white matter, centrum semiovale and medullary pyramids [83]. Of note, a comparative multi-sequence study found T2w-FLAIR as the most sensitive sequence to detect CST hyperintensities in comparison with T1w, T2w and PDw. T2w-FLAIR could even capture a longitudinal increase in signal intensity in CST [84]. Additionally, T2w-FLAIR shows high inter-rater agreement for evaluating these hyperintensities [85].
Based on the conflicting result of presence of CST hyperintensities among ALS patients and healthy controls, we set out to provide summary estimates for odds ratios (OR). The meta-analysis of CST hyperintensities on T2w image contrast showed an overall OR of 2.21 [95%-CI: 1.40-3.49] (including 13 publications, nALS=519, nCtrl= 389) (Figure 3A). The direction of effect remained after a sensitivity analysis only including studies with low risk of bias and comprising a healthy control group (Figure 3B). Neither the Egger’s regression test nor the rank correlation test suggested publication bias (p=0.10 and 0.25, respectively).
Motor cortex hypointensity
In ALS, motor cortex hypointensity, also termed the “motor band sign” or “black ribbon sign”, can be observed on T2w, T2*w, T2w-FLAIR or SWI (Figure 2). The proportion of ALS patients displaying a motor band sign varies considerably in the literature The reported proportion ranges from over 90% (on T2*w) [86], to much lower numbers [87]. Some publications reported higher proportions of the motor band sign as detected on T2w contrast in ALS patients compared to healthy controls [76]. This sign also seems to remain stable longitudinally [86]. One study in 25 upper motor neuron ALS and 23 healthy controls found 100% specificity and 20% sensitivity for the motor band sign [88].
SWI has been reported as the most sensitive imaging approach to detect the motor band sign (vs. T2w and T2*w) [89]. This was confirmed by another study reporting that only 5% of patients displayed the motor band sign on clinical routine imaging (i.e., T2w, T2*w, T2w-FLAIR, DWI) whereas 78% of patients were deemed positive on SWI [90]. Another study found that T2*w imaging was superior to other clinically employed sequences (not including SWI) [91]. Also, ultra-high-field imaging (7T) seems to have a higher sensitivity to capture the motor band sign compared to imaging at 3T [92].
Based on the conflicting result on motor band sign proportions among ALS patients and healthy controls, we set out to provide summary estimates for odds ratios (OR). The meta-analysis of motor band sign showed an overall OR of 10.85 [95%-CI: 3.74-31.44] (including 10 publications, nALS=197, nCtrl= 209) with substantial heterogeneity across publications (I2=68%, p<0.01) (Figure 4A). The direction of the effect remained after a sensitivity analysis only including publications with low risk of bias and comprising a healthy control group (Figure 4B). Both the Egger’s regression test and the rank correlation test indicated publication bias (p<0.001 for both tests).
Other MRI features
One study employing T2*w contrast at 7T did not find cerebral microbleeds in ALS [93]. Yet, a post-mortem study at 7T found increased numbers of cortical cerebral microbleeds in frontal and temporal lobes of ALS compared to non-ALS controls [94]. The latter study also found increased iron deposition in ALS brains in deep subcortical gray matter structures, such as the caudate nucleus and subthalamic nuclei, compared to healthy controls and subjects with other neurodegenerative diseases [95]. Interestingly, in the same study, cortical microinfarcts were more common in healthy controls compared to ALS.
3) Magnetic resonance imaging histopathology correlation
Fourteen publications presented concomitant histopathology data, but only 7 of them directly correlated MRI findings with histopathology. Three publications assessed the tissue signature of CST/white matter hyperintensities in ALS, comprising a total of 15 brains from ALS patients, and consistently reported myelin pallor, gliosis as well as depletion of (large) axons, albeit at different CNS locations; internal capsule [96, 97] or temporal subcortical white matter [81]. This latter study investigated one brain from a demented ALS patient and also found severe neuronal loss and gliosis in the adjacent temporal cortex.
Three publications assessed the underlying histopathology of low signal in the motor cortex and included a total of 10 brains from ALS patients. All publications found iron accumulation in the deep cortical layers [98], sometimes located within astrocytes and/or microglia/macrophages [89, 99].
Finally, one study aimed at predicting motor cortex neuron density based on white matter volume estimates. The authors indeed report a linear function modelling motor neuron density, albeit only in a subgroup of sporadic ALS patients [100].
4) Translational value of motor neuron disease animal models
As a final aim, we also attempted to address the translational value of animal MND models for capturing MRI imaging findings of human disease. To this end we performed a comprehensive literature search in Pubmed and EMBASE, retrieving 175 unique publications out of which 33 were deemed eligible for the qualitative synthesis (Figure 5).
Most publications showed a low risk of bias in the animal welfare (29/33 publications) and conflict of interest domain (18/33). Yet many publications did not report randomization (7/33), blinding (6/33) or sample size calculations (3/33).
The most frequently used animal MND model was the SOD1G93A transgenic animal model, mimicking familial ALS (26 publications, 79%). Mice were the most commonly used species (29, 88%) followed by rats (4, 12%). The used static magnetic field strengths ranged from 1.5T to 17.6T, with most publications employing 7T (16, 48%). More detailed information can be found in Supplementary Table 6.
Similar to the human paradigm, local CNS tissue volume loss also occurs in animal MND models. Thus, one-year old mice overexpressing both APP and SOD1 mutations exhibited gray matter atrophy, most pronounced in the hippocampi, entorhinal as well as cingulate cortices. In contrast, mice only overexpressing SOD1 exhibited atrophy specifically in cortical regions (cingulate, retrosplenial, and temporoparietal cortex) but not in the hippocampi [101]. A decrease in motor cortex volume as well as spinal cord volume has also been observed in the murine SOD1G93A model at postnatal day 100 [6]. The latter study also found iron accumulation in the cervical spinal cord, that, however, disappeared with progressing disease. Another study found that mice fed with cycad toxins resulting in motor neuron loss also show decreased volumes in the lumbar spinal cord gray matter, substantia nigra, striatum, basal nucleus/internal capsule, and olfactory bulb [102]. One study exploiting the TARDBPQ331K transgenic mouse strain, i.e., a model for ALS-FTD, found a more prominent atrophy in the entorhinal cortex compared to the motor cortex [103].
Furthermore, T2w hyperintensities have been described in the MND model rodent brain, albeit at different locations compared to human subjects, i.e., mainly in the brain stem [104-106]. These hyperintensities seem to parallel or even precede first behavioral signs of a MND [107, 108]. Histopathological correlations found associated vacuolar degeneration [109, 110] as well as micro- and astroglial activation [111]. Interestingly, magnetic resonance microscopy was able to also detect hyperintensities in the ventral motor tracts within the murine spinal cord [112]. Higher T2 values, mainly in the ventral portions of the spinal cord, have also been observed using conventional sequences at 7T [113].
Finally, in conflict with human disease, overt breakdown of the blood-brain barrier was described in a rat ALS model [104], which was congruent with T cell infiltration [114].