C9orf72-mediated ALS and FTD: multiple pathways to disease

C9orf72 transcription and splicing

In C9orf72 transcript variant 2, the repeats are located within the promoter region (Fig. 1), so are not incorporated into variant 2 pre-mRNA but have the potential to affect the expression of this variant. By contrast, in variants 1 and 3, the repeats are within intron 1, so are included in the respective pre-mRNAs. Variant 2 is expressed at higher levels than variants 1 and 3 in CNS tissue^138,139.

Studies have demonstrated reduced levels of one or more of the C9orf72 transcript variants in blood lymphocytes^1,132,140, induced pluripotent stem cell (iPSC)-derived neurons^12,90,94,138, frontal cortex^{1,132,141–145}, cerebellum^{12,16,143–145}, motor cortex¹² and cervical spinal cord¹² from C9orf72 expansion carriers compared with controls. The findings are particularly robust for variants 1 and 2¹⁴⁵. C9orf72 protein levels might be correspondingly reduced in the frontal cortex^110,144. Transcripts upstream of the repeat are increased relative to downstream transcripts in blood lymphocytes and brain and spinal cord tissue⁶⁶ from C9orf72 expansion carriers, possibly owing to abortive transcription in the presence of the repeat expansion⁶⁶. Raised levels of variant 1 in the frontal cortex and cerebellum are associated with increased survival¹⁴⁵ — an important consideration when developing therapies that affect transcript levels.

The intronic location of the repeats in variants 1 and 3 means they should be spliced out of the transcript, but the fact that they are translated into DPRs implies that they are either retained in the transcript or that the spliced intron is translated. Although levels of mature spliced C9orf72 mRNA are reduced in the brains of individuals with C9FTD/ALS, levels of sense and antisense transcripts containing intron 1, where the repeats are located, are increased^14,16, suggesting stabilization of repeat RNAs. Mature C9orf72 transcripts with correct splicing or with retention of intron 1 are both detected in C9FTD/ALS lymphoblasts and brain tissue^138,146, indicating that both species contribute to RAN translation.

The expression of C9orf72 is also modified by epigenetic effects (Box 1).

Autophagy and lysosomal function

Bioinformatic analysis shows that C9orf72 is structurally related to the differentially expressed in normal and neoplastic cells (DENN) guanine nucleotide exchange factor (GEF) proteins, which activate Rab proteins. Rabs are crucial for a wide range of vesicular trafficking events, and multiple lines of evidence from several independent groups point to a role for C9orf72 in autophagy and endolysosomal trafficking and function.

Knockdown of C9orf72 in human cell lines and primary neurons specifically inhibits autophagy induction, but not later stages of the autophagy pathway^91,96, leading to accumulation of p62^91,96 and cytoplasmic aggregation of TDP-43⁹⁶. Consistent with these findings, accumulation of autophagy substrates, including p62, is observed in the spleens of C9orf72-knockout mice^54,56. Conversely, overexpression of C9orf72 can activate autophagy, leading to an increase in autophagosomes in cell lines⁹¹. The role in autophagy seems to be mediated by the long C9orf72 protein isoform (C9orf72-L) rather than the short isoform (C9orf72-S)⁹⁶. The mechanism underlying these changes involves Rab proteins, although no consensus has been reached on which are the most important, with Rab1a, Rab8a and Rab39b all being implicated^{56,78,91,96,154}. C9orf72 interacts with guanine nucleotide exchange protein SMCR8 and WD repeat-containing protein 41 (WDR41)^{56,58,96,101,122,154}, and the effect on autophagy is generally agreed to be mediated through an interaction with the serine/threonine-protein kinase ULK1 complex^{56,91,96,100,154}, a key initiator of autophagy. Potentially, this interaction can also occur via TBK1⁹⁶, another serine/threonine-protein kinase implicated in autophagy. This finding is particularly intriguing, as loss-of-function mutations in TBK1 cause ALS and FTD¹⁵⁵.

Another link to known ALS-associated genes is through ataxin-2 (ATXN2), in which intermediate expansions of polyglutamine increase the risk of ALS¹⁵⁶. C9orf72 knockdown specifically increases the aggregation and toxicity of ataxin-2 protein with intermediate polyglutamine repeats⁹⁶.

The relevance of the role of C9orf72 role in autophagy for disease pathogenesis is unclear, but neurons from patients with C9FTD/ALS have impaired basal autophagy^91,93 and increased sensitivity to autophagy inhibition⁹⁰, suggesting that reductions in C9orf72 levels contribute to cellular distress. Furthermore, a Src–c-Abl pathway inhibitor, which increases autophagic processes, rescues survival defects in neurons derived from patients with ALS⁹².

In addition to effects on autophagy, reduced endocytosis was reported in C9orf72 knockdown cell lines⁷⁸, and impaired endosomal and lysosomal trafficking were observed in bone marrow-derived macrophages and microglia from homozygous C9orf72 knockout mice⁵⁴, as well as in patient-derived fibroblasts and neurons⁹³. Moreover, C9orf72 has been shown to reside on lysosomes and can directly affect lysosomal function, which might also explain the effects of C9orf72 loss of function on both endolysosomal trafficking and autophagy¹⁰¹. In iPSC-derived motor neurons, C9orf72 primarily localizes to early endosomes, and iPSC-derived motor neurons from patients with C9FTD/ALS have fewer lysosomes⁹⁴. Both patient neurons and CRISPR–Cas9 C9orf72-knockout iPSC-derived neurons have reduced vesicular trafficking, which can be rescued by C9orf72 overexpression⁹⁴. These cells also have elevated glutamate receptor levels and increased sensitivity to excitotoxicity^94,127. Consistent with this observation, increased glutamate receptor levels were found in spinal cord tissue in a C9orf72-knockout mice⁹⁴, and in spinal cord^94,127 and cortical tissue⁹⁴ from patients with C9FTD/ALS.

Taken together, these data suggest that C9orf72 is involved in multiple cellular trafficking events, and that loss of C9orf72 in both microglia and neurons can sensitize cells to other insults, thereby contributing to neurodegeneration in C9FTD/ALS.

Further insights from loss-of-function mouse models

The mouse C9orf72 orthologue shares 98% homology with human C9orf72 and is expressed in embryonic and early postnatal neurons, various regions of the adult brain and spinal cord, glia, and non-neuronal tissues, including muscle, spleen, kidney and testes^50–54,157. Several C9orf72 knockout or knockdown models have now been reported (Table 1)^51–58.

Transient reduction of C9orf72 expression in the CNS by antisense oligonucleotides (ASOs)¹³ and conditional homozygous knockouts of C9orf72 in neurons and glia⁵¹ do not lead to motor or behavioural phenotypes. By contrast, ubiquitous knockouts of C9orf72^{52–55,57,58} or CRISPR–Cas9-mediated knockouts of C9orf72 isoforms^56,57 throughout development led to immune system dysregulation in homozygous mice. The phenotypes included changes in myeloid and/or lymphoid cell populations in the spleen and lymph nodes, increased levels of inflammatory cytokines, and cervical or systemic lymphadenopathy and splenomegaly, sometimes with reduced body weight^52,53,55,57, neoplasia⁵⁵ or increased autoimmune antibody titres^53,57. In comparison, haploinsufficiency of C9orf72 does not lead to severe phenotypes.

Although some studies reported mild motor or cognitive phenotypes^52,53 or reduced lifespan^{52,53,55,57,58} in homozygous C9orf72-knockout mice, none reported neuronal loss. Transcriptomic analysis confirmed changes in immune pathways^53,54, similar to those observed in CNS tissue from patients with C9orf72 repeat expansions⁵⁴. Transcriptomic analysis in human tissue has shown that C9orf72 transcripts are particularly prevalent in CD14⁺ myeloid cells, which are involved in innate and adaptive immunity¹³⁹. Overall, in line with cellular studies of C9orf72, these findings suggest an important role for C9orf72 in immune regulation, possibly through its effects on autophagosome and lysosome function and/or microglial activity, or through alteration of autoimmune responses. However, in contrast to C9orf72 loss of function, knockout of other autophagy-related genes, including ATG7¹⁵⁸ and ATG5¹⁵⁹, in neurons in mice leads to neurodegeneration. These findings suggest that C9orf72 is not an essential component of the autophagy pathway that mediates neuronal survival.

Crucially, none of the mouse C9orf72 knockouts recapitulate ALS or FTD, suggesting that C9orf72 loss of function is insufficient to precipitate disease. However, given the role of C9orf72 in pathways previously implicated in FTD and ALS¹⁶⁰, haploinsufficiency might contribute to the disease process in combination with gain-of-function mechanisms, and an interesting approach will be to breed loss-of-function mouse models with gain-of-function models.

C9orf72 repeat RNA

In vitro, GGGGCC repeat RNA forms secondary structures, including hairpins^45,66 and highly stable G-quadruplexes^{45,66,161–163}. Other secondary structures, including DNA–RNA heteroduplexes, RNA duplexes and i-motifs^164–167, might arise from the sense and antisense repeat RNA and DNA sequences. In vivo, such secondary structures are likely to mediate the sequestration — and, as a consequence, depletion — of RNA-binding proteins (RBPs)^66,161 (Box 2), thus providing a clear potential route to RNA toxicity (reviewed extensively elsewhere^164,168).

DPRs: insights from post-mortem studies

The results of post-mortem studies have raised suspicions that DPR inclusions are not the primary culprit in C9FTD/ALS pathogenesis^{29,35–40,43,181–183}. TDP-43 pathology and neurodegeneration co-occur in affected regions of the CNS in ALS and FTD^{29,35,37–39,43}. By contrast, DPR pathology does not coincide neuroanatomically with TDP-43 pathology43 and is generally not found in the same neurons as TDP-43 inclusions^16,29,35,40. Furthermore, DPR inclusions do not differ in neuroanatomical distribution between FTD and ALS cases^29,37,38 and are rare in the spinal cord in C9orf72-associated ALS, whereas TDP-43 pathology is common^{14,17,18,29,36–40}. DPR inclusions are frequent across several brain regions^{17,35–37,39}, including structures that are thought to be minimally affected in ALS and FTD, such as the cerebellum and occipital and parietal lobes. However, one study showed that poly-GR inclusions — but not the other DPRs — correlated with areas of neurodegeneration in C9orf72-associated ALS, and also, of all the DPRs, uniquely colocalized with TDP-43 pathology in a small sample of brains that were obtained shortly after death⁴⁴. These data suggest that further investigation in larger, deep-phenotyped post-mortem cohorts will provide important insights.

Despite these observations, strong counterarguments to support a pathogenic role for DPRs in C9FTD/ALS have been put forward. Post-mortem studies tend to represent the final stages of the disease process and might not reflect the early pathogenicity of DPRs. Aggregates observed at post-mortem could represent protective species, and correlations with inclusions might be misleading if soluble species mediate neurotoxicity. CNS regions that have extensive DPR pathology but are unaffected by neurodegeneration might contain protective factors; indeed, selective vulnerability is a frequent observation in neurodegenerative diseases¹⁸⁴.

TDP-43 pathology is likely to be downstream of DPR pathology, probably explaining why it correlates more closely with neurodegeneration. This idea is consistent with downstream effects on TDP-43 in some experimental models expressing expanded repeats^109,185 or pure DPRs^106,185, suggesting that one or both of these gain-of-function mechanisms are linked to TDP-43. Affected individuals with DPR pathology but relatively mild or absent TDP-43 pathology have been reported^{16,21,43,141,182,186,187}. These individuals include a young patient with evidence of intellectual disability¹⁸⁶, a patient with C9orf72-associated ALS who had little extramotor TDP-43 pathology but showed evidence of cognitive impairment and high cerebellar DPR levels⁴³, and patients with pathological or clinical diagnoses of FTLD or FTD^{16,141,182,186,187}, some of whom died prematurely from other causes¹⁸². Therefore, DPR pathology without substantial TDP-43 pathology seems to be sufficient for disease to develop in some cases. Furthermore, as discussed below, considerable evidence from model systems indicates that DPRs can cause neurodegeneration.

Evidence for C9orf72 repeat RNA toxicity

The effects of C9orf72 repeat RNA were modelled in primary cortical and motor neurons transfected with expanded GGGGCC repeats within an artificial intronic region of the green fluorescent protein gene, reflecting the intronic human genomic context of C9orf72 expansions⁴². These neurons demonstrated nuclear RNA foci and reduced survival. Dot blots and immunocytochemistry revealed no DPRs in these cells, suggesting that the reduced survival was attributable to repeat RNA. Interestingly, however, co-expression of poly-PR and the intronic expanded GGGGCC repeats had a synergistic detrimental effect on neuronal survival.

RNA toxicity has also been implicated in eye and motor neuron degeneration in a Drosophila model that expresses 30 GGGGCC repeats with a 6 bp (CTCGAG) interruption in the middle of the repeats ^109,172. DPRs were not detected in the eyes or neurons, and were only detected when the repeats were strongly induced in all tissues¹⁰⁹.

One caveat for the interpretation of both studies is that the inability to detect DPRs is not sufficient to exclude a role for these proteins. In our experience, poly-GR can be difficult to detect even in flies that express this protein at high levels and show overt toxicity. Therefore, more sensitive detection assays for DPRs will be required to unpick the relative contributions of RNA and DPRs in these models.

A study in developing zebrafish found that both sense and antisense RNA repeats could mediate toxicity, leading to a motor axonopathy phenotype¹³⁴. No DPRs were detected in this model. In a Drosophila model, however, neither sense nor antisense repeats with similar lengths to those found in patients with C9FTD/ALS led to degeneration of adult neurons¹⁸⁹. One potential explanation for this difference is that developing neurons are more susceptible to RNA toxicity than adult neurons. Another possibility is that RNA foci in humans sequester RBPs that are not present in Drosophila, thereby limiting the utility of Drosophila as a model for RNA toxicity.

Evidence for DPR toxicity

Two studies in Drosophila models have provided evidence that neurotoxicity is attributable to DPRs rather than repeat RNA^138,188. In the first study, overexpression of expanded GGGGCC repeats in Drosophila eyes or adult neurons led to neurodegeneration¹⁸⁸. This effect was inhibited when the repeats were interrupted by stop codons in each reading frame that prevented translation of the repeats into DPRs. The second study involved ubiquitous overexpression of 160 GGGGCC repeats in an intronic context, flanked by human C9orf72 sequence¹³⁸. The intron containing the repeat was spliced out and formed large numbers of sense RNA foci in neurons and glia, without production of DPRs. This model showed no evidence of neurodegeneration, reduced survival or widespread transcriptomic changes. Increasing transgene expression in this model led to DPR production and lifespan reduction, although the frequency of sense foci remained the same¹³⁸, supporting the idea that DPRs rather than sense foci mediate neurodegeneration.

It should be noted that although the RNA sequence interrupted by stop codons in the first study forms the same G-quadruplex secondary structure as uninterrupted C9orf72 repeat RNA¹⁸⁸, the tertiary and quaternary structures are not necessarily identical, which might affect the dynamics of RBP sequestration. However, the intronic model did not have an interrupted repeat sequence and exhibited multiple sense foci, yet still showed no evidence of toxicity¹³⁸.

Toxicity of individual DPRs

In numerous studies in cell models^{14,42,69,87,104–106,190}, Drosophila ^{42,72,108,115,123}, zebrafish^133,134,191 and mice^63,64, the repetitive GGGGCC sequence was altered to generate coding sequences that expressed each DPR in isolation, which was often sufficient to produce neurotoxicity and implicate several downstream mechanisms. The main limitation of these models is that DPR overexpression might not reflect the endogenous mechanisms that are seen in patients.

Studies in cultured neuronal or non-neuronal cell lines or primary neuronal cultures indicate that poly-GR and poly-PR are the most toxic of the DPRs^42,69,87,106. These arginine-rich DPRs — in particular, poly-PR — are toxic to yeast.⁸⁹ Poly-GA and poly-PA are also toxic, but to a lesser extent⁸⁹. Synthetic poly-PR and poly-GR are highly toxic when exogenously applied to cultured human astrocytes. Poly-PR has a longer half-life than poly-GR, and is especially potent in this context⁶⁸. Synthetic poly-GR and poly-GA are also toxic to primary neurons¹⁹². Furthermore, overexpression of poly-PR was found to be toxic to control iPSC-derived neurons⁴².

Of all the DPRs, poly-GR had the greatest detrimental effect on development, locomotor activity and survival in a zebrafish model¹³³. In Drosophila models, poly-GR and poly-PR were neurotoxic when expressed in the eyes^{42,72,108,115,123,188}. In addition, flies expressing these proteins exhibited reduced survival^{42,108,123,188} and locomotor phenotypes^42,123,135. Most of these studies found that poly-GA^42,115,123, poly-PA^42,188 and poly-GP¹⁰⁸ were not toxic in Drosophila, although one study reported a mild reduced survival phenotype when poly-GA was expressed in adult neurons¹⁸⁸. Consistent with this finding, poly-GA overexpression in cultured cells^104,105, primary neurons¹⁰⁴, zebrafish^133,191 or mouse brain⁶³ leads to toxicity, and synthetic poly-GA exogenously applied to human cells¹⁹⁰ or primary neurons¹⁹² is also toxic. In addition, cryo-electron tomography revealed that poly-GA forms twisted ribbon structures that sequester the 26S proteasome¹⁰⁷.

Overall, these studies show that poly-GR and poly-PR are potently neurotoxic and poly-GA also exerts toxicity. The remaining DPRs, poly-PA and poly-GP, are unlikely to be toxic species. As poly-GR, poly-PR and poly-GA can all be damaging when overexpressed, a key aim is to establish the levels of each of these species in patient tissue, particularly at early disease stages. New techniques to extract and measure DPRs from patient tissue and assess their toxicity will be essential for this endeavour.

Mouse gain-of-function models

Several mouse models of C9orf72 gain of function have been characterized. Adeno-associated virus-mediated CNS overexpression of 66 GGGGCC repeats leads to motor and behavioural phenotypes by 6 months, with RNA foci, DPRs, phospho-TDP-43 inclusions and neuronal loss being observed⁵⁹. The same approach was used to generate mice specifically overexpressing poly-GA in the CNS⁶³. These mice developed neuronal cytoplasmic inclusions of fibrillar poly-GA, as well as neurodegeneration and motor, cognitive and behavioural phenotypes. However, these effects were not observed when the poly-GA sequence was mutated to a sequence that was unable to aggregate. Phospho-TDP-43 inclusions were rarely found; therefore, this model does not fully recapitulate the (GGGGCC)₆₆ repeat mouse phenotype, indicating that species other than poly-GA are the main drivers of phospho-TDP-43 accumulation. A further poly-GA model, with expression levels more comparable to those in the cortex in patients with C9FTD/ALS, demonstrated motor deficits, but overall a less severe phenotype than the viral poly-GA model ⁶⁴.

Bacterial artificial chromosome transgenic mouse models

Bacterial artificial chromosome (BAC) transgenic mice have the advantages of expressing the human C9orf72 gene, with surrounding regulatory regions and flanking sequences, at more physiological levels. Three BAC transgenic models have produced similar findings. Two of these models used BAC constructs^52,60, containing exons 1–5 of the gene and 300–500 repeats (Table 2). The third BAC model expressed the full gene and protein with 100–1000 repeats⁶¹. All three models developed RNA foci and DPRs in the CNS, but none demonstrated TDP-43 inclusion pathology, neuronal loss or reduced survival. One model developed memory impairment and loss of hippocampal neurons, with an increase in levels of phosphorylated TDP-43, but was not fully reflective of ALS or FTD⁵². In this model, a single injection of an ASO targeting C9orf72 RNA led to sustained reductions in RNA foci and DPR pathology, and improved cognition. Repeat length had a strong influence on the formation of RNA foci, with 100-repeat mice developing no foci and 400-repeat mice developing many foci, despite considerably higher transgene expression in the 100-repeat mice⁵².

Only one study has reported BAC transgenic mice with a striking neurodegenerative phenotype⁶². The mice expressed the full gene and different repeat lengths in different lines. Three independent lines, two with ~500 repeats and one with high expression levels of 36 repeats, developed RNA foci, DPR aggregates, TDP-43 pathology and neurodegeneration. A subset of female mice developed an acute motor phenotype, with paralysis, weight loss and decreased survival, with other female and male mice showing slower progressive neurodegeneration. Antisense transcripts and foci were observed at especially high levels in some CNS regions that showed neurodegeneration, whereas sense foci were more evenly distributed. These data show that high expression of short repeats can be toxic, indicating that large repeats are not the critical factor for toxicity in this mouse model. The reason why female mice are particularly susceptible and only a subset develops acute disease is currently unclear, although factors such as methylation might be important.

In-depth molecular and phenotypic comparisons between all the BAC models should further our understanding of C9orf72-associated pathogenic mechanisms. One model that expresses the full-length human gene with the repeat expansion recapitulates disease⁶² whereas the other does not⁶¹. These mice feature different genetic backgrounds and flanking C9orf72 sequences (Table 2), which might be contributory factors. Backcrossing the models that did not show a phenotype into different genetic backgrounds, so as to determine whether certain backgrounds facilitate disease, would provide new insight.