From animal models to human disease: a genetic approach for personalized medicine in ALS

Mutations in the SOD1 gene have been the first described cause of familial ALS [43]. Since 1993, more than 150 missense mutations have been described in SOD1, all updated on the ALSoD website (http://​alsod.​iop.​kcl.​ac.​uk/​). Unfortunately, individual mutations are poorly correlated with clinical presentation. Furthermore, the clinical phenotype is often variable between members of the same family. Almost all clinical manifestations of ALS can be observed with SOD1 mutations [23, 44‐47]. However, some authors have stated that the SOD1 A4V mutation leads to a rapid death after only one year of symptoms [22] and that people with D90A mutations have slow disease progression [45]. Some specific characteristics for SOD1 mutations are listed in Table 2. Age at onset can vary from 6 to 94 years old with a mean of 40 ± 9.9 to 58.9 ± 12.6 according to studies. Progression of the disease can also vary from 8 months to 18.7 ± 11.4 years. Most of the time, familial SOD1 ALS starts with an asymmetric weakness in a limb with predominantly lower motor neurons signs. Patients usually suffer from weakness, atrophy, fasciculation and reflexes can be either increased or decreased. Babinski’s sign is often absent. Cognitive symptoms are usually not present in familial SOD1 cases. However, FTD has been described with four SOD1 mutations, G41S, L144F, I113T and G141X [48‐51]. Non-motor symptoms have been described mainly with the D90A mutation, such as urinary symptoms, painful muscle spasm, cerebellar ataxia and sensory symptoms [45, 52].

Neuropathological findings described in patients with SOD1 mutations include several classes of cytoplasmic inclusions. Lewy-body like hyaline inclusions (LBHI) are the most described inclusions in SOD1 variants. They have been found in patients with A4V [53], G37R [54], H46R [55], H48Q [56], I113T [57] and L126S [58] mutations. LBHI, in hematoxylin and eosin (H&E) staining, exhibits dense cores with paler peripheral halos. These inclusions are usually composed of granule-coated fibrils, mutated and wild-type SOD1 and ubiquitin [59]. Other bodies, like intracytoplasmic SOD1 inclusions, skein-like inclusion or neurofibrillary tangle, were found in SOD1 familial cases (Table 2). Hyaline inclusions are also present in astrocytes [60].

The first animal model carrying a mutation in human SOD1 was created in 1994 [61]. SOD1^G93A was expressed under the control of the human SOD1 promoter. These mice reproduced most of the clinical and neuropathological findings of ALS. They developed motor deficits in Rotarod and Hangwire tests at 80-90 days and died at 130 days after loss of muscle innervation and motor neuron degeneration. Reports have demonstrated degeneration of the neuromuscular junction before the onset of symptoms, around 40 to 50 days [62, 63]. Gliosis was found before onset and increased in intensity over time [64]. Also, one report has showed an increased proportion of activated microglia from 80 days of age [65] and more recently, positron emission tomography (PET) imaging in the SOD1^G93A mouse showed increased inflammation by 110 ± 33 % in the whole brain [66].

It was originally proposed that the pathological effect of SOD1 in ALS was caused by a loss of dismutase function. However, Sod1 knock-out (KO) mice did not develop ALS up to six months of age [67]. Further studies suggest that Sod1 KO mice develop a significant distal motor axonopathy without any motor neuron loss [68]. Indeed, the role of SOD1^WT in ALS pathology has to be clarified. Some studies have demonstrated that SOD1^WT was not implicated in neurodegeneration of ALS. Mice expressing wild-type human SOD1 developed vacuolization of mitochondria and axonal degeneration in the spinocerebellar tract but without motor neuron loss before 2 years of age [69]. However, co-overexpression of SOD1^WT in fALS-mutant mice succeeded to increase the ALS phenotype and to convert unaffected mice to ALS [69, 70]. More recently, expression of SOD1^WT at same level as seen in SOD1^G93A mouse caused an ALS-like phenotype with similar neuropathological findings [71].

Numerous SOD1 mouse models have been created and they have variable phenotypes, age of disease onset and survival (Table 3). These heterogeneous phenotypes seems to be dependent on specific mutations, expression levels of mutant SOD1, gender and genetic background [72]. Generally, females experienced delayed onset and prolonged survival as compared to males and such differences can also be observed in humans [73]. Most SOD1 mutant mice represent human SOD1 pathology quite well. Mice develop fatal paralysis with motor neuron deficit, gliosis and intracytoplasmic ubiquitinated SOD1 inclusions (Fig. 3). However, poor correlation can be made between age of onset and progression in mouse and human. For example, the A4V mutation, the most frequent mutation in humans which causes a rapid disease, is not pathogenic in mouse before 85 weeks [74]. Moreover, the SOD1 G93A mutation, which leads to an early onset and fast progression in mouse, has a slow rate of progression in human (Tables 1 and 2). Interestingly, the D90A mutation in mice and humans shows strong similarities. Homozygous SOD1^D90A mice exhibit a slow disease progression and bladder disturbance, which are also found in humans with the same homozygous mutation [45, 75]. Cognitive symptoms have been described in some mouse models. Mice exhibiting the G37R mutation have learning deficits in passive avoidance from 8 months of age and pre-symptomatic SOD1^G93A mice exhibit learning delay and long-term memory deficits [76, 77]. Non-motor features such as sensory deficits are described in G37R and D83G [78].

Mouse models have also been created to understand the role of SOD1 in specific central nervous system (CNS) cells. Neuron-specific high expression of SOD1^G93A under Thy1.2 promoter led to motor neuron degeneration and paralysis [79] but lower expression of SOD1^G37R under the NF-L promoter did not cause motor neuron deficit [80]. This inconstancy can probably be explained by level of hSOD1 expressed and the specific mutation used. A group also took advantage of the Cre recombinase to decrease expression of the SOD1^G37R mutant in astrocytes and they observed delayed microglial activation and slowed late disease progression [81]. Mice expressing the G86R mutant restricted to astrocytes did not develop motor deficit or microglial activation but showed increased Glial fibrillary acidic protein (GFAP) reactivity [82]. These results suggest that glial cells contribute to disease progression but are not sufficient to trigger motor neuron disease.

H46R and G93A are the only two human SOD1 mutations introduced in rat (Table 2). These rats developed ALS features similar to SOD1 mouse models. They develop UMN and LMN degeneration and similar neuropathological findings. SOD1^G93A rats also have motor neuron (MN) loss in trigeminal, facial and hypoglossal nuclei which correspond to features of bulbar ALS [83]. Rats are of particular interest for ALS research because of their size which facilitate intra-thecal or intra cerebro-ventricular injection for preclinical trials.

Canines can be affected by degenerative myelopathy (DM), a progressive neurodegenerative disorder with robust similarities to human ALS, making dogs the only mammals with naturally occurring non-human ALS [84]. DM is particularly similar to the UMN-dominant onset form of ALS. The onset of clinical signs generally occurred at 8 years of age and duration of disease rarely exceed 3 years because of elective euthanasia. T18S and E40K mutations in SOD1 were reported in dogs with DM [85, 86]. Similarities between both diseases include axonal degeneration, astrocytosis, fibrillation and positive sharp waves at electromyography evaluation, muscle atrophy, sensory impairment and SOD1 inclusions [84]. However, MN loss is not clearly established in DM as some authors reported MN loss in dogs with DM [87] whereas others did not [88].

Zebrafish have a simplified vertebrate nervous system which offers accessible motor neurons for in vivo study of ALS. They are particularly promising for fast screening of new potential therapeutic approaches as response to treatment can be rapidly observed. However, even if the zebrafish CNS shares similarities with human CNS, structural differences are important to consider when analyzing these models. Comparative neuroanatomy was recently reviewed in [89]. For example, the absence of corticospinal and rubrospinal tracts in zebrafish make it a poor model for UMN disorders. Zebrafish injected with mutant SOD1 mRNA display abnormal branching and reduced motor neuron axonal length, but did not become paralysed. In contrast to mice, zebrafish injected with the SOD1 A4V mRNA exhibit the most prominent phenotype [90]. Transgenic G93A and G93R zebrafish exhibit MN and neuromuscular junction (NMJ) loss with deficits in swimming [91, 92].

Drosophila melanogaster is a useful tool for genetics research given its low maintenance cost as compared to rodents and other mammals. It also has shorter lifespan which allows for fast initial studies of diseases and treatment. However, the mammalian nervous system is much more complex and additional studies have to be realized in mammals before translating results directly from Drosophila to human. Human SOD1^WT, SOD1^A4V and SOD1^G85R were expressed in Drosophila [93]. All of these transgenic flies exhibit climbing inability at two to four weeks of age, suggesting a motor deficit. A decrease in synaptic transmission was observed without MN loss. The authors also observed a stress response in glial cell through increased immunostaining of Hsp70.

Caenorhabditis elegans is a nematode which can be cultured in agar plates and reach a length of 1 mm [94]. C. elegans possess the advantage of being transparent thus allowing easy detection of fluorescent proteins. Their short lifespan permits the rapid screening of drugs and the extensive study of molecular pathways. However, the simplicity of their nervous system, as in Drosophila and zebrafish, do not allow the direct translation of discoveries and treatment to humans. Human SOD1^WT, SOD1^G85R, SOD1^G93A and SOD1 with mutations in four cysteine residues (C⁴S) were introduced into C. elegans. Motor neuron deficits were measured by the number of “thrashing” movements by the worm, which are periodical changes of direction when placed in physiological buffer. Overexpression of hSOD1^WT and SOD1^G85R caused motor deficit leading to paralysis within 10 to 20 days [95].

In last few years, pigs have been used for modeling disorders such as Parkinson disease [96], Alzheimer’s disease [97] and Huntington disease [98]. Their role in modeling disease is based on their anatomical, genetic and physiological resemblance to human. CNS anatomy is particularly closer to human than the rodents. Minipigs expressing hSOD1^G93A under the CMV promoter were produced [99]. From 3 months of age, the pigs develop hind limb defects resulting in running difficulty and this incapacity became more severe with age. The pigs also exhibit muscle atrophy and accumulation of SOD1 in motor neurons but did not die of the disease up to two years of age.

Interest in biomarkers in ALS research has growth in the last decades. Unfortunately, no biomarkers are validated yet for the diagnosis of ALS. The discovery of biomarkers will help early diagnosis and potentiate therapeutic intervention in a context of personalized medicine. The optimized biomarker will be quantifiable, will be standardized, will have low variability and will be introducible within all centers. Volatile organic compounds (VOC), which are tools for biomarker investigation, have been analyzed in SOD1^G93A mice [100]. Mice, in early stage of the disease, exhibit a pattern of 12 different oxidative stress-related blood VOCs as compared to non-transgenic aged-matched mice. MicroRNAs (miRNAs) are also of particular interest in biomarker development as they are frequently altered in disorders, can be dosed in biological fluids and the pattern of variation in miRNA level is relatively specific. miRNA-206 has been identified in the blood and muscles of SOD1^G93A symptomatic mice [101]. miRNA-206 was also increased in blood of ALS patients, but without any discrimination of familial or sporadic form of disease. However, a recent report suggests that the miRNA pattern of expression in serum is mutation-independent. Consequently, stratified subgroups could increase the significance of miRNA as biomarkers. Indeed, patients with FUS and C9orf72 mutations have a similar serum miRNA profile compared to SOD1 patients, but profiles of miRNA was less similar among SOD1 patients [102]. Actually, among all ALS patients, a serum miRNA signature was observed decade before disease onset as compared to healthy controls. Imaging techniques are also potential biomarkers for ALS. SOD1^G93A mice exhibit a tissue vacuolization in T2 MRI studies. These alterations occurred before MN loss and matched histopathological change in post-mortem analysis [103]. Another group also observed vacuolization and gliosis in T2 MRI as soon as 60 days in SOD1^G93A [104].

Personalized medicine (PM) is a medical scheme that combines study of genetic, environmental and clinical diagnostic testing, including biomarkers, to individualize patient care. Patients are divided into stratified subgroups to improve their response to treatment [105]. This concept has emerged from the need to understand the variety of response to similar treatment in common illnesses. The application of personalized medicine in ALS first requires genetic screening among ALS patients in clinical trials and in overall neurology clinics and eventually, utilization of biomarkers for personalized diagnosis and treatment. Ideally, subgroups of specific mutations should be applied in clinical trials because of the high phenotypic variability of these mutations (Table 2). Unfortunately, recruitment difficulty is limiting this approach. As better drugs become available, it will be important to take into account the genetic profile status of the patient to determine if individuals with certain mutations would respond better to particular treatments. This has been successful in other diseases such as the use of Vemurafenib in melanoma patients specifically with a BRAF V600E mutation [106].

SOD1 animal models exhibits most clinical features found in humans with SOD1 mutations and this is particularly accurate for rodents and canine. This accuracy suggests that results in SOD1 rodents can potentially be applied to human with SOD1 mutations in a context of PM. However, the review of the literature suggest that SOD1 models are less accurate for modeling general ALS since they exhibit important disparities with sALS and other forms of fALS. SOD1 transgenic mice do not recapitulate the C-terminal and phosphorylated TDP-43 cytosolic inclusions observed in almost all familial and sporadic ALS cases [107, 108]. Likewise, Lewy body-like inclusions are a typical neuropathological finding of SOD1 patients and animal models but are not observed in sALS and SOD1-unrelated fALS [109].

Gene therapy is a promising avenue for personalized medicine in ALS. Recombinant adeno-associated viral (AAV)-mediated gene delivery is the most developed method for gene therapy. These vectors can target specific cells when directly injected in the CNS or by crossing the blood-brain barrier when injected systemically. AAVs expressing the insulin growth factor 1 (IGF-1) and glial cell line-delivered neurotrophic factor (GDNF) injected in muscle succeed to prolong life-span and delay disease in the SOD1^G93A mouse by retrograde action on MN [110]. Intra-thecal injection of AAV expressing a single-chain antibody against misfolded SOD1 also delayed disease onset and extended life-span up to 28 % in SOD1^G93A mice [111]. It is clearly established that SOD1 mutants develop a toxic gain of function. Consequently, silencing the SOD1 gene by delivery of a lentiviral vector that expresses small interfering RNAs (siRNA) was tested in SOD1^G93A. The authors described a delay in disease onset by more than 100 % and a life-span extends by 80 % [112]. A delay in onset and extension of life-span were also observed in SOD1^G93A rats injected with AAV9-SOD1-shRNA in the motor cortex [113] and in SOD1^G93A mice treated with intravenous injection of AAV9-SOD1-shRNA [114]. With the same idea, antisense oligonucleotide targeting the SOD1 gene infused in the lateral ventricle of SOD1^G93A rats extends survival by 10 days and extends disease duration by 37 % [115]. Actually, a recent phase 1 clinical trial of intrathecal injection of antisense oligonucleotide was conducted in SOD1 familial ALS cases and no serious adverse effects were identified [116]. A phase 2 clinical trial is currently conducted but no results are yet available.

TAR-DNA-binding protein 43 (TDP-43) is a multi-functional DNA/RNA binding protein normally found in the nucleus. It is known to play a role in RNA processing and transport and splicing regulation [117]. TDP-43 was found to be a major component of pathologic cytosolic inclusions in ALS and FTD [118, 119]. TDP-43 was subsequently found to be mutated in familial and sporadic ALS patients [120‐123]. While TDP-43 mutations are rare (4 % of fALS), it is notably established that the TDP-43 protein is found in inclusions of most ALS cases. Over 40 mutations are currently identified but G298S, A315T, M337V, G348C, and A382T are the most frequent among patients [124].

It is currently clinically impossible to clearly discriminate TDP-43 patients from other sALS and fALS patients. It is also difficult based on the literature to establish clear phenotype-genotype correlation within each variant of TDP-43 mutation. Clinical characteristics of frequent mutations are outlined in Table 3. However, one publication succeeds to point out some specific clinical characteristic in TDP-43 mutated patients [124]. Age at onset can vary from 20 to 77 years old with a mean age at onset of 54 years old [125]. M337V and G348C have the earliest disease onset. Onset within TDP-43 patients is significantly earlier than sALS patients but no difference can be observed when compared to SOD1 patients, which also have earlier disease onset [124]. The upper limb is the predominant site at onset comprising 60.7 % of patients [124]. Bulbar onset seems more frequent in Asian patients (55 %) as compared to Caucasians (24.7 %) and in patients with a M337V mutation. Small cohort of patients with TDP-43 mutations was used and makes these results difficult to generalize.

Fronto-temporal dementia is relatively rare within TARDBP ALS cases. A382T is the most described mutation in FTD-TDP-43 familial cases [126‐129]. One group identified a Sardinian cohort of ALS with an A382T mutation with frequent dementia (30.8 %) [130]. They also identified that the A382T mutation is present in 21.7 % of their FTD cohort [126]. However, a similar frequency of TDP-43 mutations was not reported in other populations. The authors explained that this frequency is probably caused by the isolated population of Sardinia with its associated founder effects. Clinical features of FTD cases with an A382T mutation mainly include irritability, aggressiveness, poor flexibility, fixed ideas, change in eating behaviors, trouble in social behaviors and emotional flatness [126]. Brain imaging studies also suggested a pattern of defect found in A382T related FTD cases. MRIs showed unilateral or bilateral frontotemporal atrophy and frontotemporal hypoperfusion, observed in SPECT brain perfusion. In combination, neuroimaging findings showed prevalent involvement of temporal regions, which is rarely reported within other familial FTD cases [126].

Neuropathological findings in TDP-43 fALS patients tend to represent the majority of sALS and fALS cases. Regardless of the numerous publications on TDP-43, only a few reports are available with neuropathological findings. Indeed, most of our knowledge about TDP-43 neuropathological features comes from mouse models. In patients, neurons and glial cells are positive for TDP-43 cytosolic inclusions [131]. Often, these inclusions are also composed of ubiquitin and P62. Bunina bodies are described in most mutations except in A315T.

The discovery of TDP-43 as a major component of ALS inclusions had led to the generation of many mouse models trying to recreate this pathology (Table 4). In ALS patients, levels of the TDP-43 protein are elevated 1.5-2.5 fold in pathologic neurons [132, 133]. The first strategy for modeling ALS with TDP-43 transgenic mice was to overexpress hTDP-43^WT to similar level found in humans. Many available hTDP-43^WT models are based on high protein expression by the mouse prion (mPrp) or Thy1.2 promoters. The phenotype observed in these mice seems to be dependent on TDP-43 protein level and did not reproduce the age-related degeneration of ALS. Hemizygous mice expressing 3–4 fold levels of hTDP-43^WT did not recapitulate features of ALS [134]. These mice did not develop motor dysfunction and had only mild gliosis and diffuse ubiquitin staining. However, homozygous mouse expressing 2.5 fold increases in TDP-43 levels had rapid onset and low survival [135]. Mice died within approximately 2 months with associated motor deficits. Nevertheless, these mice did not recapitulate classical ALS neuropathological findings. Phospho-TDP-43 inclusions were rare and no motor neuron loss was observed. Mice expressing high level of hTDP-43^WT under Thy1.2 promoter (3.8–5.1 folds) also developed rapid progression of disease and had 30 % UMN and 25 % LMN loss with rare phospho-TDP-43 aggregates. Some reports have proposed that the fast disease progression without important loss of motor neurons could be caused by gastrointestinal complications before appearance of full neurodegeneration [136, 137].

Our group took advantage of genomics fragments including the human endogenous TDP-43 promoter to moderately express hTDP-43^WT, hTDP-43^A315T and hTDP-43^G348C in mouse [138]. These mice develop age-related progressive motor deficit in the rotarod test from 36–42 weeks of age and cognitive impairment suggestive of FTD. We also observed ubiquitin/TDP-43 nuclear and cytosolic inclusions in 10 month old hTDP-43^A315T and hTDP-43^G348C mice (Fig. 3). Nonetheless, these mice did not become paralysed and died at standard age. Interestingly, mPrp-hTDP-43^A315T mice develop disease and develop paralysis within few months, but no TDP-43 neuronal cytosolic inclusions (NCI) were observed [139]. As shown in Table 4, no clear correlation can be made between mutation and mouse phenotypes. Correlation can be observed though between the levels of protein expression and phenotypes. Low protein expression caused age-related motor dysfunction and TDP-43 cytosolic accumulation, but the mice did not get paralysed [138, 140]. On the other hand, high protein expression cause early onset and fast disease progression without clear motor neuron loss and TDP-43 NCI. More recently, double transgenic mice expressing hTDP-43^WT and hTDP-43^Q331K developed rapid progressive limb paralysis starting from 3 weeks of age and leading to death by 8–10 weeks [141]. TDP-43^WT and TDP-43^Q331K single transgenic mice do not develop motor dysfunction up to 24 months of age. These double transgenic mice also exhibited NCI positive for p62, ubiquitin and TDP-43, reproducing features of inclusions found in human cases. They also exhibited MN loss in the anterior horn of the spinal cord with association of muscular atrophy and NMJ loss. These mice recapitulated most ALS features in exclusion of age-related degeneration.

Few TDP-43 rat models have been generated. Mild expression of hTDP-43 with its endogenous promoter did not cause motor deficits in rats [142]. However, rats expressing the M337V mutation under endogenous TDP-43 promoter developed fast motor deficit following by paralysis and death within 10 to 29 days. To have a delayed phenotype and to successfully establish transgenic lines, the authors used a Tet-off system and activated TDP-43 expression 4 days after birth. These rats develop symptoms from 3 weeks of age and die at 5 weeks. TDP-43 was observed within the cytoplasm but no clear inclusions were detected. Interestingly, the rats exhibited degeneration of the ventral root, dorsal root and corticospinal tract. Specific neuronal expression of hTDP-43^M33V under the neurofilament heavy (NFH) promoter with a Tet-off system expressed at 60 days also caused rapid disease onset [143]. Rats became paralysed within 3–4 weeks after doxycycline induction of TDP-43 expression in neurons. Again, no TDP-43 neuronal cytoplasmic inclusions were detected but ubiquitin inclusions were detected in rats when hTDP-43^M337V was expressed in motor neurons.

Multiple Drosophila melanogaster models for TDP-43 have been generated to understand the cellular role of TDP-43. Many of them do not represent ALS because of their tissue expression and poor motor phenotype. They have been extensively described in a recent review on Drosophila melanogaster and ALS [144]. Interesting models of Drosophila were created with the D42-Gal4 system [145‐147]. These flies expressed the TDP-43 transgene exclusively in motor neurons. hTDP-43^WT overexpression in Drosophila leads to fast paralysis and death in the larval stage [145]. Features of ALS were not clearly demonstrated since these flies died rapidly and no NCI or motor neuron loss was established. Later, hTDP-43^WT and hTDP-43^A315T overexpressing flies were found to have motor deficit in a climbing assay [146]. Interestingly, cytosolic aggregates of TDP-43 were observed in another model of hTDP-43^WT Drosophila [148]. These flies also developed dose-dependent loss of NMJ with adult-onset motor disorder. They exhibit progressive loss of motor ability leading to complete paralysis at 30 days and control flies kept 20 % of motor aptitude at same age. Recently, thoracic motor neuron loss was observed in flies expressing temperature-dependant dTDP-43^WT [147].

Transgenic Caenorhabditis elegans models of TDP-43 are also based on protein overexpression. Two groups have expressed different forms of TDP-43 under control of the snb-1 promoter [149, 150]. They observed a motility defect and degeneration phenotype which was worse with mutant TDP-43. Some worms became paralysed and died within 3 weeks. Phosphorylated TDP-43 was detected but no TDP-43 NCI were noted. Moreover, injection of mutant TDP-43 mRNA in Zebrafish embryos caused motor neuron axonopathy [151, 152]. No phenotype was observed upon overexpression of WT-TDP-43 alone.

Few TDP-43 biomarkers are available in literature and none of them are clearly validated for disease screening in human. ELISA quantification demonstrated an elevated TDP-43 protein level in plasma of ALS cases as compare to healthy subjects [153]. No differences were established between sporadic and familial cases. The same finding was observed in the cerebrospinal fluid (CSF) of ALS patients [154]. Interestingly, TDP-43 CSF level was higher within the first 10 months of disease onset. TDP-43 CSF level was significantly higher than the level measured in other neurologic disorder such as Parkinson’s disease, multiple sclerosis, Guillain-Barré syndrome and progressive supranuclear palsy [155]. A limit dose of 27.9 ng/ml of TDP-43 protein in CSF had a sensitivity of 59.3 % and a specificity of 96.0 %. These results suggest that TDP-43 CSF level could be an interesting tool for ALS diagnosis. Recently, tissue-engineered skins derived from ALS patients demonstrate that skin could possibly be used as biomarker as well. TDP-43 aggregation was observed in sALS-derived skin and in not yet symptomatic patients carrying GGGGCC DNA repeats [156].

According to our review of the literature, most TDP-43 models exhibit only a part of all important ALS features. We used five major criteria to point out the mice models which are more representative of human disease. Mice should have TDP-43 cytosolic inclusions, motor neuron loss, age-related onset of disease, paralysis which cause shorter lifespan, and gliosis. In most models, the rapid onset does not correlate with development of ALS in older patients and thus is difficult to use as a neurodegenerative model. Generally, mice with late onset do not get paralysed and miss one the major aspect of ALS pathogenesis. We have highlighted models which filled most of the criteria (Bold in Table 4). We consider that these models should be preferred for drug testing.

Promising approaches for the treatment of TDP-43 cases have been generated in the last few years. We have highlight therapies which can be applied to specific treatment of TDP-43 cases in a context of personalized medicine. Induced pluripotent stem cells (iPSCs) derived from TDP-43^M337V patients have been used for gene therapy testing [157]. A reduction of 30 % of cytosolic TDP-43 and 45 % of nuclear TDP-43 level was observed in M337V-iPSCs transfected with a siRNA specifically targeting M337V allele. This approach has to be tested in animal models but seems to be a potential treatment of TDP-43 fALS cases. IPSCs will certainly help to develop therapies in the next years. Specific mutations in patients could be targeted by siRNAs. It is well accepted that TDP-43 mutations disturb cellular RNA metabolism. One of the surveillance mechanisms in cells is composed of upframeshift protein 1 (UPF1) which destroys mRNAs with premature codon. AAV-expression of UPF1 in TDP-43 mice ameliorates motor phenotype and blocked paralysis of forelimbs [158]. These results suggest that gene therapy could also be used for overexpression of surveillance mechanisms instead of directly targeting the TDP-43 protein.

A hexanucleotide (GGGGCC) repeat in a non-coding region of C9orf72 was first described in 2011 in ALS, FTD and ALS-FTD familial cases [159, 160]. C9orf72 frequency varies geographically but seems to be present in up to 35–45 % of fALS, making it the most common genetic cause. The normal number of GGGGCC repeats is variable within healthy persons; approximately 90 % have fewer than 10 repeats. Conversely, the number of repeats can reach hundreds to thousands in ALS patients. The role of these repeats in ALS-FTD physiopathology remains unclear. However, recent semi-automated quantification of expansion number exposed a link between the number of G₄C₂ repeats and clinical characteristics [161]. FTD patients have shorter disease duration with a higher number of repeats, but no correlation was observed in ALS patients.

C9orf72-related ALS cases have several distinct characteristics. Nevertheless, like other genes, a wide range of clinical phenotype can be observed in patients carrying C9orf72 expansions. Age at onset of C9orf72 (C9ALS) fALS patients vary between studies [162‐164]. There is good evidence that bulbar onset is more frequent in C9ALS patients than non-C9orf72 patients [162, 164, 165]. Also, there is certainly a higher prevalence of FTD in C9ALS cases. Co-morbid dementia was observed in 50 % of C9orf72 patients and only in 12 % of ALS cases without the expansion [164]. C9orf72-related FTD is mainly characterized by a higher frequency of psychotic symptoms and irrational behaviour as compared to other causes of FTD [166]. Interestingly, hypokinesia/bradykinesia, rigidity and sometimes, tremor, suggestive of parkinsonism, have been described in C9ALS patients [167]. Members of this family were either touched by ALS, FTD and/or progressive supranuclear palsy (PSP). Generally, parkinsonism symptoms seen in these patients are unresponsive to levodopa treatment. C9orf72 expansion was rare in the diagnosis of Parkinson’s Disease (PD) [168, 169].

Particular features can be seen in imaging studies of C9orf72-related ALS/FTD and are dependent of the presence or absence of FTD symptoms. PET imaging identified hypometabolism in anterior and posterior cingulate, insula, caudate and thalamus, and hypermetabolism in the midbrain, bilateral occipital cortex, globus pallidus and left inferior temporal cortex as compared to sALS patients [170]. Voxel-based morphometry revealed atrophy in right frontal gyrus, in left anterior cingulate gyrus and in right precentral gyrus of C9ALS patients [164].

TDP-43 pathology is largely observed in the motor system of C9ALS patients [171]. There is also a large amount of P62 positive and TDP-43 negative inclusions in the pyramidal cells, hippocampus and cerebellum of C9ALS patients [172]. A recent review of the literature has help to understand the neuropathological findings associated with G₄C₂ repeat cases [173]. It is important to note that while the G₄C₂ repeat occurs within intron 1 of the C9orf72 gene, the repeat expansion has been shown to nonetheless be translated in both sense and antisense orientations to generate proteins with dipeptide repeats that then are found in intracellular inclusions. The prevalence of TDP-43, p62 and dipeptide protein repeat (DPR) inclusions was assessed specifically in the CNS. In the spinal cord, TDP-43 NCI was observed in 84.1 %, P62 NCI in 70.6 % and DPR inclusions in 49.8 % of patients. P62 and DPR were the major component of NCI in the cerebellum (93.8 and 90.9 % respectively) and TDP-43 was found in only 3.9 % of cerebellums. P62 and DPR were more prevalent than TDP-43 in the hippocampus but TDP-43 was more prevalent in the substantia nigra.

The first mice carrying GGGGCC repeats expansions were recently generated [174] (Table 5). These mice carried 80 G₄C₂ repeats controlled by TRE promoter. After doxycycline induction, the mice developed ubiquitin-positive inclusion but no DPR were observed. The authors suggest that this model can further be used for the study of toxicity induced by RNA. However, no behavioral analysis was conducted on these mice. Knockout of C9orf72 in mice did not result in any motor neuron degeneration or reduced survival, which suggests that loss-of-function is not sufficient to cause ALS [175]. More recently, an AAV vector expressing either 2 or 66 repeats of G4C2 were injected into CNS of postnatal day 0 mice [176]. RNA nuclear foci were observed in mice carrying 66 repeats but not in control mice carrying only 2 repeats. Also, rare pTDP-43 aggregates were observed in the nucleus and cytosol of cortex and hippocampus regions. These aggregates were not positive for poly(GA) but 75 % of cells containing TDP-43 NCI were positive for poly(GA) inclusions. (G₄C₂)₆₆ mice exhibit anxiety behavior in an open field test and motor impairment in the second day of rotarod testing at 6 months of age as compared to control mice.

Recently, two different groups generated mice carrying a bacterial artificial chromosome (BAC) with 100 to 1000 GGGGCC repeats [177, 178]. These mice did not exhibit any behavioral or motor phenotypes. However, they developed sense/antisense intranuclear RNA foci and DPR in both neurons and glial cells. Both groups suggest that this number of repeat is sufficient to cause RNA accumulation but not to cause cellular dysfunction leading to motor disease in mice. ALS patients with only a few hundred copies have been reported. Thereby, this discrepancy between mice and human has to be clarified. Very recently, two other groups generated transgenic mice carrying BAC with approximately 450 and 500 GGGGCC repeats [179, 180]. Jiang et al. have generated a mouse model that exhibits cognitive deficits from 12 months of age in working memory test and anxiety evaluation [179]. The mice also exhibit age-related RNA foci and DPR cytosolic accumulation. Liu et al. have generated and well characterized a mice model which developed many features of C9ALS [180]. These mice developed muscle denervation, MN loss, anxiety-like behavior, degeneration in hippocampus, paralysis and decreased survival. RNA accumulation was age-related and the authors also found TDP-43 inclusions in degenerating brain.

Whether nuclear RNA foci or DPR, or both, are toxic for cells remain an important question and this is currently under investigation by many groups. As previously described, Drosophila melanogaster is a superior tool for understanding pathological mechanisms than mimicking disease phenotype. Two recent publications suggested that DPR are more toxic to cells than nuclear RNA foci by generation of transgenic drosophila [181, 182]. In both studies, drosophila containing DPR had decreased survival but no phenotype was observed in flies with expression of RNA-only repeats. However, motor impairment was not described. 30 GGGGCC repeats were expressed under the motor neuron specific promoter Ok371-GAL4 in D. melanogaster [183]. With Drosophila activity monitoring (DAM) system, the authors observed a significant motor impairment at 28 days after eclosion. Neuronal death was observed in cell culture and in the eye but was not confirmed in motor neurons of flies. Another drosophila model was generated with 58 repeats of G₄C₂ in motor neurons using the Ok371-GAL4 promoter [184]. These flies developed impaired locomotor activity in larvae, a decrease in bouton number of NMJ and a low muscle area.

Deletion of the C9orf72 orthologue, alfa-1(ok3062), in C. elegans, caused a motor phenotype [185]. The worms developed an age-related motility defect reaching paralysis in 60 % of worms at 12 days of adulthood. The authors also crossed alfa-1(ok3062) with their FUS^S57Δ and TDP-43^A315T model and observed that the motor phenotype was worse in the double transgenic alfa-1(ok3062);TDP-43^A315T C. elegans as compared to simple TDP-43 mutant worms. However, no synergistic effect was observed in alfa-1(ok3062); FUS^S57Δ worms. The loss of function theory was also studied in zebrafish. Knockdown of the zebrafish C9orf72 orthologue was established using antisense morpholino oligonucleotides [186]. Fish developed motor neuron axon abnormalities, swimming impairment in reaction to touch and also reduced spontaneous swimming 48 h post-fertilization. No information was given on survival of zebrafish.

Multiple biomarkers have been studied as diagnostic tool or disease progression marker in C9orf72 pathology despite the relatively recent discovery of its implication in ALS. One of the most characterized cellular findings within C9orf72 patients is the presence of nuclear RNA foci. RNA foci are found in many cell types such as skin biopsy-derived fibroblasts, leukocytes and lymphoblasts and can possibly be utilized as a disease progression marker [187]. Also, a recent report suggests that the level of 5’ methylation of the G₄C₂ repeat can inversely correlate with disease duration in 34 C9orf72 patients [188]. This epigenetic modification can be found in blood, spinal cord and frontal cortex. Another potential biomarker tool is the level of poly-GP proteins in a patient’s CSF [189]. When compared to healthy subjects and ALS patients without C9orf72 repeat expansion, C9-ALS patients have significantly increased levels of poly-GP peptide in their CSF. Further studies have to be performed to clarify peptide levels in each disease progression stage.

It is premature to establish which C9orf72 model is better for mimicking human disease. Animal models of C9ALS repeat expansion have to be optimized and extensively characterized. However, based on clinical presentation, some specific characteristics are central features of C9 cases and should be found in animals. This includes RNA foci, DPR and TDP-43 positive aggregates, age-related disease and reduced survival. Cognitive symptoms are also commonly found in C9ALS patients. Most generated mice develop cellular features of C9ALS patients such as RNA foci and DPR [176‐178]. Recently, two groups have generated promising models which exhibits most of C9ALS features and might be exploited for drug testing [179, 180].

Although many therapeutic targets are conceivable, only a few have been currently tested. First, targeting RNA with antisense oligonucleotide (ASO) therapeutics should be a potential avenue because of RNA toxicity in C9ALS models. ASO were tested in fibroblasts and iPSCs derived from C9ALS ALS patients and succeed to decrease RNA foci without reducing overall RNA levels [190, 191]. ASO managed to reduce glutamate toxicity and increased survival of iPSCs by 30 % [190]. Tolerability of C9orf72 reduction in mouse was then tested and no pathological effects were observed on motor ability, strength and anxiety up to 17 weeks after ASO treatment by intracerebroventricular injection [191]. More recently, a single-dose ASO injection which targets repeat-containing RNAs caused a reduction in RNA foci and DPR in mice expressing BAC C9ORF72 RNA with 450 repeats [179]. Reduction in anxiety and cognitive function were also observed at 9 months and a positive effect on behavioral was sustained 6 months after injection. These results suggest a promising treatment possibility for C9orf72 patients.

Genetic screenings have identified mutations in the gene encoding fused in sarcoma (FUS) [192‐194]. Similarly to TDP-43, the main function of FUS is linked to RNA metabolism. More than 50 mutations have been identified and mutations in FUS represent around 4 % of fALS and 1 % of sALS. Unfortunately, there are only a few clinical descriptions of FUS pathology and phenotypic correlation is difficult to establish. Many reports suggest that juvenile onset is more frequent in FUS cases. A German study reports a series of families with age at onset ranging from 21 to 76 years old and with many cases before the age of 40 [195]. They also suggest that truncating mutations have a more severe disease course than missense mutations. LMN signs appear to be the dominant features in FUS cases. Cognitive symptoms are particularly rare with only few cases reported in the literature and bulbar/spinal onset are both frequent with a mutation in FUS [196].

Neuropathological findings in FUS-related disease include basophilic inclusions which are round or oval often similar in size to the nucleus. These inclusions are mainly found in the anterior horn of the spinal cord and sometimes in the motor cortex [197]. These inclusions are commonly found in juvenile ALS with mutations in FUS. NCI are also present in pathological analysis and are similar to those observed in TDP-43 pathology. They are often positive for p62 but negative for TDP-43 [32].

FUS KO mice were reported to investigate the effect of FUS deletion. Mice either died in the first 24 h of life or had important deficits including sterility and chromosomal instability [198]. Another group suggested that knock-out FUS mice were developing a neuropsychiatric disorder that did not correspond to ALS [199]. Overexpression of hFUS^WT under the mouse prion promoter leads to an aggressive phenotype in homozygous mice [200] (Table 6). These mice had approximately 1.9 times the level of FUS expression. They develop motor impairment leading to paralysis from 8 weeks of age and were euthanized when 10–13 weeks old. They exhibit increased cytoplasmic FUS without ubiquitinated-FUS inclusions, MN loss in the anterior horn of the lumbar spinal cord, impaired NMJ and gliosis. Mice expressing FUS^WT or FUS^R521G driven by the ubiquitously expressed enhancer-chicken β-actin hybrid (CAG) promoter were also generated [201]. The mice exhibited early mortality before post-natal day 30 without any FUS inclusions or degeneration of lateral columns and ventral horns. However, gliosis and motor impairment in rotarod tests were observed.

An encouraging hFUS^R521C rat model was generated using a tetracycline-inducible system [202]. After doxycycline withdrawal, the mutant FUS protein was expressed in offspring. The rats developed many features of ALS pathology. Degenerating axons were observed in the spinal cord ventral roots, dorsal roots, corticospinal tracts and frontal cortex of hFUS^R521C rats but not in hFUS^WT rats. Muscle atrophy was observed and electromyography shown fibrillation potential which is suggestive of muscular atrophy. FUS-negative ubiquitin aggregates were detected in transgenic hFUS^R521C rats. Glial activation was also detected in the brain and spinal cord. Nevertheless, the rats developed fast paralysis leading to death within 30 to 70 days of age.

Several Drosophila melanogaster models were generated for investigating the pathological role of FUS. First, the Gal4-UAS system was used for expression of WT, R524S or P525L mutations in motor neurons [203]. Morphological changes were observed in the cell body and at NMJ leading to functional deficits of MN in transgenic flies. The authors observed significant locomotive impairment in larval movement in flies expressing either of three FUS variants. Conditional expression of FUS variants with the elav-GeneSwitch system demonstrated a shorter lifespan of transgenic flies [204]. The authors observed a faster rate of death and presence of cytosolic accumulation of FUS protein in flies expressing mutated FUS as compared to flies expressing WT-FUS. Interestingly, the level of non-aggregated but insoluble FUS protein positively correlated with the level of neurodegeneration in transgenic Drosophila [205].

Caenorhabditis elegans was also used for modeling FUS pathology and interesting results were obtained from this model. A group expressed different forms of FUS including WT, R514G, R521G, R522G, P525L and two truncated FUS [206]. Only R522G, P525L and the truncated forms of FUS caused aggregates and only these transgenic worms exhibited a motor phenotype. Motor impairment started at 3 days of age and C. elegans became paralysed at 8 days of age. Survival was reduced by 12.8 days in worms expressing R522G, P525L or truncated FUS as compared to non-transgenic C. elegans. These results were partially confirmed by expression of either WT or S57Δ FUS in GABAergic neurons [207]. Motor phenotypes were observed only in FUS^S57Δ worms which expressed FUS cytosolic inclusions.

Zebrafish harboring mutations in FUS gene were also generated. FUS^R521H but not FUS^R521C or FUS^S57Δ fishes exhibited motor impairment when compared to FUS^WT [208]. High-speed video analyses of touch-evoked locomotor activity revealed shorter swim duration, swim distance and swim velocities in fish expressing FUS^R521H as compared to wild type fish [209]. NMJ synaptic transmission was also reduced.

Our review of the literature has pointed out rat and mouse models with many phenotypic and neuropathological features of ALS [200, 202]. However, they do not possess age-related characteristic of ALS. We consider that these models could be adequate models for drug testing relative to FUS pathology. There is sparse literature which can be applied to a personalized medicine approach in FUS cases. One group has tried to identify biomarkers in skin derived fibroblasts from sporadic ALS patients. Unfortunately, the study failed to identify differences in FUS pattern of expression in skin between healthy control and ALS cases [210]. Like previously described, mi-RNA are potential biomarkers in ALS. FUS is known to be implicated in metabolism of a subset of micro-RNAs such as miR-9, miR-125b and miR-132 [211]. Their value as biomarkers remains to be tested. Methylation is an important process in nuclear-cytoplasmic shuttling of FUS. One group tried to reduce FUS cytosolic accumulation by reducing their methylation with shRNA targeting the FUS methylation enzyme [212]. Reduction of FUS cytoplasmic inclusions was noted.

Ubiquilin-2 plays a central role in the ubiquitin proteasome system (UPS). Mutations in UBQLN2 have been linked to ALS and FTD [213] though screening in different populations has revealed that mutations in UBQLN2 are generally rare within ALS cases [214, 215]. However, the UBQLN2 protein is found in cytosolic inclusions of both familial and sporadic ALS and appears to have an important role in pathological processes such as aggregate formation and proteasome impairment [213, 216]. No clear genotype-phenotype correlation can be established in patients. FTD appears to be frequent in ALS caused by mutation in UBQLN2. Both males and females can be affected despite the X-linked transmission [217]. Some studies have suggested early age at onset in familial UBQLN2 ALS and site at onset was described in lower limbs, upper limbs and bulbar regions [218]. A specific pattern of UBQLN2 staining has been observed in C9orf72 patients and the authors proposed that this staining could be used as biomarker for identification of C9orf72 cases [219].

Mice carrying a hUBQLN2^P497H mutation under control of the UBQLN2 endogenous promoter were recently generated [220]. These mice develop UBQLN2/ubiquitin/p62 positive inclusions in the brain, dendritic spinopathy and cognitive deficits at 11–13 months of age which suggest features of FTD. However, the mice do not develop motor impairment or motor neuron loss. Conversely, AAV expression of three different UBQLN2 variants by intracerebroventricular injection in mice caused a phenotype at 3–4 months of age [221]. Transgenic UBQLN2^P497H or knock-out rats were also created [222]. While cognitive and neuronal loss was observed in UBQLN2^P497H rats, no phenotype developed in knock-out rats. Astrocytes and microglial activation was observed but no information was given about motor function.

The Optineurin (OPTN) protein is mainly implicated in the autophagy processes. A mutation in the OPTN gene was first linked to ALS in a consanguineous Japanese family [223]. A neuropathological study suggested that OPTN was present in skein-like inclusions and round hyaline inclusions in the spinal cord of sALS patients [224]. However, another study mentioned that OPTN inclusions are rare and restricted to a minority of sALS cases [225]. Clinical characteristics were investigated in patients carrying Q398X and E478G mutations [226, 227]. These patients exhibit slowly progressive motor dysfunction, unusual finger malformations and personality changes. Observed NCI were positive for TDP-43, p62 and ubiquitin but negative for OPTN. OPTN inclusions were described in a patient with both C9orf72 hexanucleotide repeats and an OPTN mutation [228]. Both spinal and bulbar onsets are described but no strong conclusions can be made about duration of the disease and age at onset. Generation of a transgenic mouse with mutation in OPTN gene failed to demonstrate any motor phenotype [229]. Similarly, loss of OPTN in Zebrafish results in cell death but no motor phenotypes were noted [230].