Amyotrophic lateral sclerosis (ALS) is a fatal and incurable neurodegenerative disease characterized by the progressive loss of motor neurons in the spinal cord and brain. In particular, autosomal dominant mutations in the superoxide dismutase 1 (SOD1) gene are responsible for ~20% of all familial ALS cases. The clustered regularly interspaced short palindromic repeats (CRISPR)–CRISPR-associated (Cas9) genome editing system holds the potential to treat autosomal dominant disorders by facilitating the introduction of frameshift-induced mutations that can disable mutant gene function. We demonstrate that CRISPR-Cas9 can be harnessed to disrupt mutant SOD1 expression in the G93A-SOD1 mouse model of ALS following in vivo delivery using an adeno-associated virus vector. Genome editing reduced mutant SOD1 protein by >2.5-fold in the lumbar and thoracic spinal cord, resulting in improved motor function and reduced muscle atrophy. Crucially, ALS mice treated by CRISPR-mediated genome editing had ~50% more motor neurons at end stage and displayed a ~37% delay in disease onset and a ~25% increase in survival compared to control animals. Thus, this study illustrates the potential for CRISPR-Cas9 to treat SOD1-linked forms of ALS and other central nervous system disorders caused by autosomal dominant mutations.

Here, we demonstrate that CRISPR-Cas9 can disrupt mutant SOD1 expression in the spinal cord of the G93A-SOD1 mouse model of ALS. Systemic administration of an AAV9 vector encoding the Cas9 nuclease with an sgRNA targeting the human SOD1 G93A gene to neonatal G93A-SOD1 mice reduced mutant SOD1 protein in the spinal cord and enhanced the survival of motor neurons. This therapeutic genome editing strategy delayed disease onset, improved motor function, reduced muscle atrophy, and, critically, extended survival in ALS mice. These results thus establish that CRISPR-mediated genome editing can be used to treat familial ALS and potentially other central nervous system (CNS) disorders caused by autosomal dominant mutations.

Genome editing offers an alternative approach to treat autosomal dominant disorders, including many familial forms of ALS, via the disruption of mutant gene function. In particular, the RNA-guided Cas9 endonuclease ( 17 ) from clustered regularly interspaced short palindromic repeats (CRISPR)–CRISPR-associated (Cas) systems has emerged as a versatile genome editing tool ( 18 – 21 ). Cas9 can be targeted to genomic loci to induce a DNA double-strand break (DSB) via RNA-DNA base complementarity using a single-guide RNA (sgRNA). DSBs induced by Cas9 enable the introduction of frameshift-inducing base insertions and/or deletions (indels) that can disrupt gene expression following DNA repair ( 22 ). Given these features, we reasoned that CRISPR-Cas9 could be used to treat familial ALS via genome editing following its in vivo delivery into an animal model of the disease using an adeno-associated virus (AAV) vector.

Dominant mutations in the Cu-Zn superoxide dismutase 1 (SOD1) ( 3 ) gene account for ~20% of familial forms of the disease and ~2% of all cases. The SOD1 gene encodes a metalloenzyme that converts superoxide anions into oxygen and hydrogen peroxide and is thus critical to cellular antioxidant defense. Although the mechanism behind SOD1 toxicity is not completely understood ( 4 ), transgenic animals that express mutant forms of the gene develop a progressive neurodegenerative disease that emulates the hallmarks of ALS ( 5 , 6 ), including motor neuron degeneration, muscle wasting, and paralysis. Reducing mutant SOD1 expression in the spinal cord using antisense oligonucleotides (ASOs) ( 7 , 8 ) or RNA interference (RNAi) ( 9 – 15 ) can slow disease onset and improve survival in these animal models. However, to date, ASOs have been unable to reduce mutant SOD1 protein in ALS patients ( 16 ), and both ASOs and RNAi can be hampered by incomplete knockdown, which limits their effectiveness as therapeutics.

Amyotrophic lateral sclerosis (ALS; also known as Lou Gehrig’s disease) is an adult-onset neurological disorder ( 1 ) that involves the loss of motor neurons in the spinal cord, brainstem, and motor cortex. ALS leads to progressive muscle weakness and atrophy throughout the body, ultimately leading to paralysis and death within 3 to 5 years of symptom onset. There is no cure for ALS, and the only U.S. Food and Drug Administration–approved medications, riluzole and edaravone, are minimally effective, increasing survival by only 2 to 3 months in the case of riluzole (no survival data for edaravone have been reported to date) ( 2 ).

Compared to control animals, immunofluorescent staining of spinal cord sections from end-stage mice revealed that animals treated by CRISPR-Cas9 had ~50% more ChAT + neurons in the lumbar (P < 0.05) and thoracic (P < 0.001) regions of the spinal cord ( Fig. 3 ). This suggests that CRISPR-Cas9 conferred protection to some individual motor neurons. In addition, consistent with our earlier findings (fig. S7), we observed limited SaCas9 expression in GFAP + astrocytes in end-stage spinal cord sections from CRISPR-treated animals (fig. S13), as well as immunoreactive mutant SOD1 inclusion bodies in many of the same cells (fig. S14). Optimization of AAV-mediated gene delivery and/or SaCas9 expression in astrocytes ( 35 , 40 ), microglia ( 34 , 46 , 48 ), and oligodendrocyte precursors ( 49 , 50 ) may further enhance efficacy. Collectively, these results demonstrate that CRISPR-Cas9–mediated disruption of mutant SOD1 expression in G93A-SOD1 mice enhances the survival of spinal cord motor neurons and improves motor function and life span.

( A and B ) Disease onset, ( C and D ) survival, ( E ) rotarod performance, and ( F ) weight of G93A-SOD1 mice injected with AAV9-SaCas9-hSOD1 (n = 7), AAV9-SaCas9-mRosa26 (n = 7), or AAV9-EGFP (n = 7) via facial vein at P0-P1. (E and F) Mean rotarod times and weights were normalized to average 56-day values for each group. Wild type (n = 6) indicates litter-matching control mice. Values are means and error bars indicate (A to C) SD and (E and F) SEM. ***P < 0.0005; **P < 0.005; (A, C) one-way or (E and F) two-way analysis of variance (ANOVA) followed by Tukey’s post hoc analysis.

To test whether in vivo disruption of mutant SOD1 by CRISPR-Cas9 provides therapeutic benefit, we monitored motor function, weight loss, and atrophy in G93A-SOD1 mice injected with AAV9-SaCas9-hSOD1, AAV9-SaCas9-mRosa26, or AAV9-EGFP at P0-P1. Compared to control animals, disease onset (that is, peak weight) in mice infused with AAV9-SaCas9-hSOD1 was delayed by 33 days (hSOD1, 126 ± 5.7 days; mRosa26, 93 ± 9.6; EGFP, 92 ± 8.1; uninjected, 92 ± 8.8; P < 0.0001) ( Fig. 2A ) and ranged from 119 to 133 days in treated animals compared to 77 to 98 days in control mice (P = 0.0001) ( Fig. 2B ). Mean survival also increased by 28 to 30 days in treated mice (hSOD1, 152.4 ± 7.9 days; mRosa26, 122.8 ± 6.1; EGFP, 124 ± 3.8; uninjected, 125 ± 3.5; P < 0.0001) ( Fig. 2C ) and ranged from 142 to 167 days in treated animals compared to 114 to 136 days in control animals (P < 0.0001) ( Fig. 2D ). Compared to age-matched controls, animals treated by CRISPR-Cas9 displayed improved motor function (P < 0.0001) based on rotarod performance ( Fig. 2E ) and maintained and/or gained weight (P < 0.0001) for 28 to 35 days beyond the expected point of disease onset ( Fig. 2F ). Treated mice also exhibited reduced muscular atrophy, as evidenced by a slower rate of weight loss after disease onset [hSOD1, −0.42 ± 0.015 weight (%) per day; mRosa26, −0.67 ± 0.05; EGFP, −0.63 ± 0.067; P < 0.005] (fig. S11). Notably, following the eventual onset of disease, we observed no slowing in disease progression (that is, the length of time between disease onset and end point) (fig. S12), likely due to insufficient gene editing in astrocytes, which contribute to disease progression in SOD1-linked forms of ALS ( 34 , 35 , 45 – 47 ).

We next investigated whether SaCas9 induced OT modifications in vivo. Using Cas-OFFinder ( 44 ), we identified 12 potential OT sites in the mouse genome that differed from the on-target SaCas9 cleavage site by up to four mismatches (fig. S10A). Deep sequencing revealed no significant increase in indel formation at each candidate OT site in CRISPR-treated mice compared to control animals (P > 0.1 for all) (fig. S10B).

To evaluate indel formation in vivo, we deep-sequenced hSOD1 G93A transgenes amplified from dissected spinal cord tissue, which included transduced motor neurons, as well as nontransduced nerve and glial cells from the white and gray matter. These latter cell populations, which are much more numerous than motor neurons, were not expected to be gene-modified on the basis of our immunohistochemistry results. According to CRISPResso (a software pipeline that analyzes genome editing outcomes from deep sequencing data) ( 43 ), indels were present in ~0.2 and ~0.4% of sequenced hSOD1 G93A transgenes from the lumbar and thoracic spinal cord of CRISPR-treated mice, respectively, corresponding to a ~7-fold (P = 0.01) and ~14-fold (P < 0.05) increase over control animals ( Fig. 1F ). Consistent with Western blot results showing no difference in mutant SOD1 protein in the cervical spinal cord of gene-edited mice ( Fig. 1E and fig. S8), we also observed no significant increase in indel formation in cervical spinal cord samples from CRISPR-modified animals (0.05% modification frequency; P > 0.05 compared to controls).

Compared to control animals, mutant SOD1 was reduced in transduced spinal cord cells in G93A-SOD1 mice infused with AAV9-SaCas9-hSOD1 ( Fig. 1D ). Western blot analysis of spinal cord lysate indicated that mutant SOD1 protein was decreased in CRISPR-treated mice by ~3-fold (P = 0.001) and ~2.5-fold (P < 0.05) in the lumbar and thoracic regions, respectively ( Fig. 1E and fig. S8). Consistent with in vitro studies in NSC-34 cells (fig. S2), no significant difference (P > 0.5) in mouse SOD1 protein was observed in the spinal cord lysate from treated versus untreated animals (fig. S9). Intriguingly, despite its efficient transduction, we also observed no significant difference (P > 0.5) in mutant SOD1 protein in the cervical spinal cord of gene-edited G93A-SOD1 mice (fig. S8), potentially because of variability among treated animals.

We injected G93A-SOD1 mice with AAV9 encoding SaCas9 and an sgRNA targeting either the hSOD1 gene (AAV9-SaCas9-hSOD1) or the mRosa26 locus (AAV9-SaCas9-mRosa26) via the facial vein at postnatal days 0 and 1 (P0-P1) ( Fig. 1C and fig. S3). We used a double tyrosine mutant of AAV9 that we ( 41 ) and others ( 42 ) previously developed to enhance gene delivery to the CNS. Spinal cord sections were analyzed by immunohistochemistry for expression of (i) the motor neuron marker choline acetyltransferase (ChAT), (ii) SaCas9 via a genetically fused HA epitope, and (iii) mutant SOD1, which we detected using an antibody that preferentially recognizes human protein (fig. S4). We observed SaCas9 expression primarily in the ventral horn of the spinal cord (fig. S5) and found that ~74% of all ChAT + cells examined throughout the anterior gray column expressed SaCas9, indicating broad motor neuron transduction. We also observed SaCas9 expression in fibers in the ventral horn, which we detected using the neurite marker β3-tubulin (fig. S6). As indicated by immunostaining for the astrocytic marker glial fibrillary acidic protein (GFAP), we observed little SaCas9 expression in gray and white matter astrocytes (fig. S7), potentially due to limited transduction by the AAV vector and/or suboptimal expression from the CMV promoter.

We next evaluated whether CRISPR-Cas9 could reduce mutant SOD1 expression in vivo following delivery to G93A-SOD1 mice using an AAV vector. AAV is a promising in vivo gene delivery vehicle found to be safe and effective in an increasing number of clinical trials, including those for hemophilia B ( 26 ), choroideremia ( 27 ), Leber’s congenital amaurosis type II ( 28 ), and lipoprotein lipase deficiency ( 29 , 30 ). When administrated systemically to neonatal mice, AAV9 ( 31 ) can cross the blood-brain barrier and, within the spinal cord, transduce motor neurons, sensory fibers, and, to a lesser extent, astrocytes ( 32 , 33 ). Both motor neurons (determinants of onset and early disease progression) ( 34 ) and astrocytes (secrete neurotoxic factors that can selectively kill motor neurons) ( 35 – 40 ) play an important role in SOD1-linked ALS. In light of this pathology, and considering a recent report highlighting the benefits of early gene therapy intervention in G93A-SOD1 mice ( 12 ), we used neonatal AAV9 delivery for this proof-of-concept study.

We evaluated the ability of SaCas9 and the designed sgRNA to disrupt the mutant SOD1 expression in mouse neuroblastoma–spinal cord–34 (NSC-34) cells stably transfected with complementary DNA (cDNA) encoding hSOD1 G93A . NSC-34 cells are frequently used to study certain aspects of ALS in vitro because they share several morphological and physiological similarities with motor neurons ( 25 ). The most efficient sgRNA targeted exon 2 of the hSOD1 G93A coding sequence ( Fig. 1B and fig. S1) and reduced mutant protein by ~92% in SaCas9-expressing cells, which we enriched using fluorescence-activated cell sorting (FACS) following cotransfection with an enhanced green fluorescent protein (EGFP)–encoding surrogate reporter plasmid (fig. S2A). Sanger sequencing revealed the presence of indels in ~94% of hSOD1 G93A transgenes analyzed from these cells (fig. S2B), confirming that mutant SOD1 expression was disrupted by SaCas9-mediated genome editing. Despite the potential for off-target (OT) disruption of the mouse SOD1 gene, we observed no significant difference (P > 0.5) in mouse SOD1 protein in NSC-34 cells expressing SaCas9 with an sgRNA targeting either the hSOD1 gene or the mouse Rosa26 locus, a genomic “safe harbor site” often used for mouse transgenesis (fig. S2A). These results indicate that SaCas9 and its sgRNA discriminate between human and mouse alleles.

( A ) AAV vector schematic. ITR, inverted terminal repeat; NLS, nuclear localization signal sequence; 3×HA, three tandem repeats of the human influenza hemagglutinin (HA) epitope tag. ( B ) Schematic representation of the human SOD1 locus and the sgRNA target site. The arrowhead depicts approximate position of the G93A mutation. TSS, transcriptional start site; PAM, protospacer-adjacent motif. ( C ) Experimental timeline for in vivo studies. ( D ) Immunofluorescent staining of lumbar spinal cord sections and ( E ) Western blot of lumbar, thoracic, and cervical spinal cord lysate 4 weeks after G93A-SOD1 mice were injected with AAV9-SaCas9-hSOD1 (S; n = 3) and AAV9-SaCas9-mRosa26 (R; n = 3) via facial vein (quantitation of Western blot results in fig. S7). Arrowheads indicate ChAT + and SaCas9 + cells with (upward) high or (downward) low hSOD1 expression. Images were captured using identical exposure conditions. Scale bar, 50 μm. ( F ) Indels from whole spinal cord tissue 4 weeks after G93A-SOD1 mice were injected with AAV9-SaCas9-hSOD1 (n = 3) via facial vein. Indels are colored dark green. Wild-type sequences are colored gray. The arrowhead indicates predicted SaCas9 cleavage site (D to F). All injections were performed at P0-P1.

We used the Cas9 nuclease from Staphylococcus aureus (SaCas9) ( 23 ) to target the mutant SOD1 gene in the G93A-SOD1 mouse model of ALS ( 5 ), which carries ~25 tandem repeat copies of the hSOD1 G93A transgene and recapitulates many aspects of the disease, including progressive muscle atrophy and impaired motor function. Unlike the larger Cas9 nuclease from Streptococcus pyogenes ( 17 – 21 ), SaCas9, along with its sgRNA and a full-length cytomegalovirus (CMV) promoter, can fit into a single AAV particle to drive expression in vivo ( Fig. 1A ). More than 100 mutations in SOD1 have been identified in patients with familial ALS. Because of this heterogeneity, we designed several sgRNAs that do not overlap with the G93A mutation and could be applicable to other ALS-linked SOD1 variants, including wild-type SOD1, which has been associated with sporadic ALS ( 24 ). Notably, no adverse effects were observed in a clinical trial for ALS using an ASO that targets both mutant and wild-type SOD1 ( 16 ).

DISCUSSION

Genome editing technologies can be used to introduce precise genomic modifications into mammalian cells and model organisms (51) and thus hold tremendous potential for treating the genetic causes of many diseases. Here, we have shown that CRISPR-Cas9 can reduce mutant SOD1 protein in the spinal cord following systemic delivery using an AAV vector. This therapeutic genome editing strategy delayed disease onset, improved motor function, and, critically, increased survival, illustrating the utility of genome editing to treat SOD1-linked ALS and potentially other CNS disorders caused by autosomal dominant mutations, such as Huntington’s disease or the spinocerebellar ataxias. These results build on basic research studies by our laboratory (52) and others (53, 54) showing that CRISPR systems can induce gene editing in the mammalian brain and demonstrates that therapeutic genome editing can be achieved in tissues beyond the liver (55–58) and muscle (59–62) in animal models of human disease.

Both motor neurons (determinants of onset and early disease progression) (34) and astrocytes (secrete factors that selectively kill motor neurons) (35–40) play an important role in SOD1-ALS. Because of its innate capacity to cross the blood-brain barrier and transduce spinal cord motor neurons and, to a lesser extent, astrocytes, we used AAV9 to deliver our CRISPR gene editing system to G93A-SOD1 mice at P0-P1. Because neonatal delivery has proven effective for evaluating RNAi-based gene therapies in G93A-SOD1 mice (10, 12, 14), we reasoned that this strategy would enable us to assess the impact of genome editing on disease onset and progression. Notably, AAV administration to neonatal mice previously facilitated the validation of zinc-finger nuclease– and CRISPR-based strategies for correcting hemophilia B (55) and ornithine transcarbamylase deficiency (57), respectively. In addition, a phase 1 clinical trial based on AAV9-mediated delivery of the survival of motor neuron 1 (SMN1) gene for spinal muscular atrophy (63, 64) in infants is under way, underscoring the therapeutic potential of this serotype.

Immunohistochemistry analysis of spinal cord sections from CRISPR-treated mice revealed that SaCas9 expression, as well as mutant SOD1 gene disruption, was confined primarily to ChAT+ motor neurons. This finding was reinforced by end-stage histological analysis, which showed limited SaCas9 labeling in GFAP+ astrocytes and the presence of presumably toxic mutant SOD1 inclusion bodies in many of the same cells. These results collectively support the hypothesis that inefficient SOD1 disruption in astrocytes could contribute to the lack of slowing in disease progression that we observed in gene-edited mice. Advancing this first-in-class CRISPR-based gene therapy for a CNS disorder toward the clinic will therefore require optimizing gene delivery vehicles for human use (65) and enhancing Cas9 expression in motor neurons, astrocytes, and other cell types implicated in SOD1-ALS. The rapidly expanding catalog of Cas9 orthologs (several of which are smaller than SaCas9) (66, 67) may also enable the use of specialized and/or larger promoter sequences incompatible with the all-in-one AAV vector strategy used here, or additional components that can drive self-inactivation of Cas9, and thereby enhance safety.

Because of its rapid and robust phenotype, the G93A-SOD1 mouse model of ALS is among the most widely used transgenic models of the disease. This mouse carries ~25 tandem repeat copies of the hSOD1G93A transgene, and previous reports have indicated that a reduction in hSOD1G93A copy number can have a profound effect on the ALS phenotype (68). Despite the challenge that high-copy repetitive elements can pose from the perspective of genome editing, the Cas9 nuclease has facilitated the disruption of up to 62 copies of the porcine endogenous retrovirus in a kidney epithelial cell line (69), indicating its capacity for multiplexing and targeting high-copy elements. Although it is difficult to determine the number of gene-edited hSOD1G93A transgenes in transduced spinal cord cells, we showed that SaCas9 modified ~94% of hSOD1G93A alleles in transfected NSC-34 cells.

The CRISPR-based genome editing strategy used here cannot distinguish between mutant and the wild-type human SOD1, which performs important functions in cells. However, in a phase 1 clinical trial, no adverse effects were observed in ALS patients treated with an ASO targeting both mutant and wild-type SOD1 (16). If needed, an allele-specific CRISPR system configured to target disease-causing SOD1 mutations (including A4V, G37R, H46R, G85R, and G93A) or a gene knockout-and-replace therapy (70) could be used to overcome the toxicity arising from a lack of allele specificity. Because the PAM sequence recognized by SaCas9 (that is, NNGRRT) occurs less frequently than those for the more commonly used Cas9 from S. pyogenes (that is, NGG), it may be necessary to alter the PAM specificity of SaCas9 (71) to discriminate between the mutant and wild-type SOD1 alleles.

We used the Cas-OFFinder algorithm (44) to identify candidate OT cleavage sites for the SaCas9 nuclease. This software is not limited by the number of mismatches in the sgRNA sequence and allows for subtle variations in the PAM recognized by Cas9. Using deep sequencing, we observed no increase in indel formation at each candidate OT site in CRISPR-treated mice versus control animals. However, the use of unbiased genome-wide approaches that rely on the in situ capture of adapter sequences at nuclease-induced DSBs (23, 72) or cell-free digestion of genomic DNA using Cas9 ribonucleoprotein (73) could facilitate the formation of a comprehensive portrait of SaCas9 cleavage specificity. In addition, whole-exome genome sequencing of CRISPR-treated motor neurons differentiated from human-induced pluripotent stem cells derived from SOD1-ALS patients would yield critical insight into the specificity of SaCas9 in a more therapeutically relevant context. Given that the OT activity of a nuclease is proportional to its concentration and the amount of time it is exposed to the cell (74), future work elucidating the kinetics of Cas9 expression in vivo could shed light on its function and inform methods for improving its specificity.

Finally, there are several limitations to this study that should be discussed. First, we observed a ~2.5-fold reduction in mutant SOD1 protein in the lumbar and thoracic spinal cord but detected indels in only 0.2 to 0.4% of sequenced hSOD1G93A transgenes. Although the underlying reason for this difference requires further exploration, studies have demonstrated that genome editing can elicit a phenotypic effect that exceeds the measured indel frequency (23, 75). In our case, this difference could potentially be explained by CRISPR interference (76), which may function alongside SaCas9-mediated genome editing to reduce mutant SOD1 protein. Furthermore, we deep-sequenced hSOD1G93A transgenes amplified from whole spinal cord tissue, which included both transduced motor neurons and nontransduced cells from the white and gray matter, the latter of which are more numerous than motor neurons and may not express as much mutant SOD1. The Allen Spinal Cord Atlas online database indicates that, in P56 mice, SOD1 is more strongly expressed in the anterior horn of the spinal cord, where lower motor neurons reside, as well as the posterior horn. This is consistent with our own immunostaining in 28-day-old G93A-SOD1 mice (fig. S4). Thus, higher SOD1 gene expression in transduced versus nontransduced cells at the time of analysis (that is, 28 days) could possibly contribute to the observed discrepancy. Second, because of the difficulty in isolating spinal cord motor neurons from juvenile and adult mice (77), we could not measure the frequency of hSOD1G93A gene modification in transduced cells. Future work using transgenic mice expressing a GFP reporter in motor neurons (78) could facilitate analysis of indel formation or SOD1 messenger RNA following FACS enrichment. Third, we observed variable and inefficient editing in the cervical spinal cord of G93A-SOD1 mice. Further optimization of SaCas9 expression in preclinical large animal models using increased samples sizes can help resolve this point. Finally, we did not administer the AAV vector to adult animals in this proof-of-concept study. Assessing the efficacy of CRISPR-mediated disruption of mutant SOD1 in adult ALS mice both before and after disease onset will be critical in establishing the potential of this approach for clinical translation.

In conclusion, we have demonstrated that CRISPR-Cas9–mediated gene editing provides therapeutic benefit to the G93A-SOD1 mouse model of ALS. This work establishes genome editing as a possible therapy for ALS and paves the way for this technology to treat other forms of the disease, including those caused by a hexanucleotide repeat expansion in the C9orf72 gene (79, 80), which could potentially be excised by Cas9 following its coexpression with a pair of sgRNA flanking the repeat expansion.