Reversal of C9orf72 mutation-induced transcriptional dysregulation and pathology in cultured human neurons by allele-specific excision

The C9orf72 repeat region lies between two alternative start sites (1, 2), one in exon 1A, the other in exon 1B (Fig. 1A). Our patient line contains ~250 GGGGCC repeats (28) on the mutant allele, and two on the wild-type (WT) allele. The donor was a 37-y-old female patient of European descent who was asymptomatic at the time of donation. We chose this line because of a rare but advantageous single nucleotide polymorphism (SNP) in exon 2 that allows us to quantify gene expression from each allele independently (Fig. 1B). From here on out, we refer to this line as C9-unedited. Our control cell line was donated by a healthy 30-y-old subject of mixed European and Asian descent. It contains 10 repeats on one allele and two on the other and is referred to as WT-control.

Before carrying out our edits, we engineered our C9 and WT iPSC lines to contain the hNIL transgene cassette with a TET-on system in the CLYBL safe-harbor locus (29, 30) (SI Appendix, Fig. S1). This well-characterized inducible system (29, 30) puts expression of three human transcription factors, NGN2, ISL1, and LHX3, under doxycycline control, which drives the iPSCs to differentiate into motor neurons. The neurons express high levels of motor neuron markers (HB9 and ChAT) compared to iPSCs at the time point investigated in our expression studies (2 wk of age) (SI Appendix, Fig. S1 E and F).

We then introduced Cas9 and gRNA pairs via electroporation into the WT and patient hNIL iPSC lines to generate a variety of excisions of the C9orf72 genomic locus. We used CRISPOR (31) or AlleleAnalyzer (32) to design gRNAs (SI Appendix, Table S1) with the fewest computationally predicted overall off-targets. The genetic change was confirmed by PCR and Sanger sequencing or single-molecule sequencing in sorted single cells or hand-picked clones. All clones had a normal karyotype (SI Appendix, Figs. S1–S10).

We first excised the noncoding region containing the repeat expansion (REx, Fig. 1A and SI Appendix, Fig. S2). Since gRNAs targeting the repeated motif have numerous predicted off-targets throughout the genome, it is not safe to cut within the repeat region itself; instead, we made cuts just 5′ and 3′ to the repeats (Fig. 1A REx, SI Appendix, Fig. S2). These repeat-flanking sequences are highly conserved and do not offer allele-specific gRNA binding sites. Therefore, this excision was expected to be biallelic. However, by chance it only occurred on the mutant allele in the clone we selected from our patient cell line (SI Appendix, Fig. S2), leaving intact the WT allele with its two native repeats. Our WT-control clone with the same edit was biallelic as expected (SI Appendix, Fig. S7).

Newer versions of Cas9 can distinguish between alleles that differ by a single nucleotide (33), which allowed us to use SNPs in cis with the mutation to specifically excise the mutant allele. We phased SNPs to the repeat expansion mutation in our patient line using single-molecule sequencing (28). We used AlleleAnalyzer (32) to design allele-specific gRNA pairs based on common heterozygous polymorphisms found in a reference dataset of over 2,500 human genomes from around the world (34). Using two allele-specific gRNAs, we excised 22.2 kb of the mutant allele in the patient line starting 12.3 kb upstream of exon 1A and stretching all the way through exon 3 (Fig. 1A HET(Mut)x, SI Appendix, Fig. S3). In addition, we made the equivalent 21 kb excision on the WT allele in both the patient and WT lines (Fig. 1A HET(WT)x, SI Appendix, Figs. S6 and S8) (35).

Our fourth edit, excision of exon 1A, was designed to determine the impact of this exon on C9orf72 expression and to test the hypothesis that silencing the mutation would suffice to eliminate C9orf72 pathology. Exon 1A includes a transcriptional start site and controls the expression of the C9orf72 sense-transcript harboring the mutation. We excised it on both alleles (1Ax) in both the patient (Fig. 1 1Ax, SI Appendix, Fig. S4) and WT-control lines (SI Appendix, Fig. S9). As additional controls, we also created homozygous knock-outs of the gene in our patient and WT lines, using biallelic excisions starting 21 kb upstream of exon 1A and ending in exon 3 (patient line; SI Appendix, Fig. S5) or 7 kb upstream of exon 1A and ending in exon 2 (WT line; SI Appendix, Fig. S10). We found no detectable off-targets in the edited clonal iPSC lines REx, HET(Mut)x, 1Ax by comparing Sanger sequencing across computationally predicted potential off-target sites (SI Appendix, Fig. S11).

We evaluated the effect of each of the edits on C9orf72 RNA expression in induced motor neurons derived from the edited and unedited iPSC lines. The C9orf72 locus is known to produce at least three sense mRNAs: variant 1 (exon 1A-short through exon 5), variant 2 (exon 1B-exon 11), and variant 3 (exon 1A-long through exon 11) (1). Using ddPCR probes spanning either the exon 1A-exon 2 or exon 1B-exon 2 splice junctions, we quantified the two major splice forms of C9orf72, variant 3 and variant 2, which we refer to as 1A-transcript and 1B-transcript from here on (Fig. 1B and SI Appendix, Table S2). We were not able to detect variant 1 in our lines, consistent with its low to undetectable expression in human tissue (36, 37). We also quantified total C9orf72 mRNA using a probe targeting the exon 2-exon 3 junction.

Across all lines, most C9orf72 mRNAs contained exon 1B, while exon 1A-containing transcripts represented only a small proportion of total transcripts (Fig. 1 C and D). However, the sum of 1A and 1B transcripts was inferior to the total amount of C9orf72 transcripts in lines containing the repeat expansion (Fig. 1C; C9-unedited, HET(WT)x). This discrepancy was not observed in patient lines in which the mutation was excised (REx, HET(Mut)x) or silenced (1Ax) (Fig. 1C) or in any of the WT lines (Fig. 1D). We hypothesize the gap corresponds to sense 1A-transcripts that retain the repeat expansion instead of splicing it out; such transcripts cannot be detected by ddPCR, because the repeat expansion disrupts the binding of the ddPCR exon-exon spanning probe and puts too much distance between the primers’ binding sites for a successful PCR. If this hypothesis is correct, the actual proportion of sense 1A-transcripts could reach 30% or more of total C9orf72 transcripts in lines with the repeat expansion, up from <1% in corrected and WT motor neurons (Fig. 1 C and D). We report the amount of sense mutant transcript in human induced motor neurons. This estimate does not include antisense repeat-containing transcripts, which are known to occur but are not captured by our assay, and would further increase the total amount of repeat-containing transcripts. We cannot confirm this amount directly, because it is not yet possible to PCR or sequence transcripts harboring a large repeat expansion.

Interestingly, the number of 1A-transcripts (blue bar) decreased in the C9 lines after removal of the repeat region (REx) or excision of the mutant allele (HET(Mut)x) but not after excision of the WT allele (HET(WT)x) (Fig. 1C). The removal of the repeat expansion in the patient line (C9-REx and C9-HET(Mut)x) decreased the 1A transcript level below that of unedited WT (0.2 vs. 0.5% of UBE2D2 expression, respectively; Fig. 1C vs. Fig. 1D). We are cautious to compare expression from the two lines directly because differences in genetic background can influence baseline transcriptional levels. However, we wondered whether editing affected the methylation state which could influence transcriptional regulation. We compared methylation patterns (38) between C9-unedited and the C9-REx and C9-HET(Mut)x iPSCs. We found no differences in methylation between either lines within a 5 kb region around the repeat expansion and including exon 1A and 1B (SI Appendix, Fig. S12). In summary, these data suggest that 1A-transcription is upregulated in the diseased state, most probably from the mutant allele.

To test this hypothesis, we determined the proportion of transcripts coming from the WT vs. mutant alleles. We took advantage of a coding SNP (rs10757668) in the exon 2 splice acceptor of our patient line and phased it to the repeat expansion by single-molecule sequencing. Then, using ddPCR probes that targeted either variant of this SNP, we determined the fraction of 1A- and 1B-transcripts derived from each C9orf72 allele (Fig. 1 E and F) as well as the amount of 1A and 1B transcripts produced by each allele relative to a house-keeping transcript (Fig. 1 G and H). Once again and as expected, our assay only detected correctly spliced RNA transcripts and not those retaining the repeat expansion. Nevertheless, most (94%) of the exon 1A-containing transcripts we were able to amplify came from the mutant rather than the WT allele in the unedited patient line (Fig. 1 E and G, C9-unedited). The imbalance was corrected by repeat expansion excision, which reduced expression from the mutant allele without altering expression from the WT allele (Fig. 1 E and G, REx). These findings suggest that at least some of the mutant 1A-transcripts can undergo normal splicing. Furthermore, since the mutant and WT 1A transcripts we detect differ only at the SNP (Fig. 1B), the excess of mutant 1A transcripts in the unedited C9 line must reflect increased transcription from the mutant allele, rather than a difference in RNA stability. According to our quantification, 1A is at least 10-fold more active in the mutant than the WT allele (Fig. 1E, C9-unedited). This is likely an underestimate since we cannot currently measure 1A-transcripts retaining the repeat expansion.

In contrast to 1A transcripts, 1B transcripts came predominantly (>68%) from the WT allele in C9-unedited motor neurons (Fig. 1 F and H). The balance between the two alleles was fully restored by the allele-specific excision of the repeat expansion (REx), suggesting that the repeat expansion partially inhibits 1B-transcription on the mutant allele. Biallelic excision of exon 1A also restored equal production of 1B transcripts by the two alleles (Fig. 1F; 1Ax). However, it reduced expression from the WT allele (Fig. 1G; compare height of the purple bar (WT) between 1Ax and C9-unedited), suggesting a positive interaction between 1A and 1B on the WT allele.

Interestingly, removal of the repeat expansion did not alter the production of 1A and 1B transcripts by the WT allele, and neither did excision of the mutant allele (Fig. 1 G and H; compare the height of purple bars between C9-unedited, REx and HET(Mut)x). Similarly, removal of the WT allele did not alter expression from the mutant allele (Fig. 1 G and H, compare the height of yellow bars in HET(WT)x relative to C9-unedited). Thus, expression from either allele is independently regulated. As expected, excision of either allele resulted in elimination of any detectable transcript from that allele (HET(Mut)x and HET(WT)x, Fig. 1 E and H).

In total, these findings demonstrate that transcripts harboring the repeat expansion can be correctly spliced (i.e., rid of the toxic repeat), an exciting finding that could inspire an additional therapeutic avenue. Second, the repeat expansion increases the activity of the 1A start site and decreases the activity of the 1B start site on the mutant allele but not the WT allele. Finally, sense transcription is allele-independent as removal of either allele did not alter sense expression from the other allele.

To quantify C9orf72 protein, we used the Simple Western system (WES) after validating antibody specificity using our knock-out line (39). None of the edits in the patient line reduced the C9orf72 protein levels in induced motor neurons (Fig. 2 A and C). In WT motor neurons, only exon 1A excision reduced C9orf72 expression (Fig. 2 B and D). Furthermore, unedited WT and C9-patient motor neurons produced equivalent amounts of C9orf72 proteins (Fig. 2E). These results indicate that the repeat expansion itself or large excisions of the C9orf72 locus (such as removal of an entire allele) do not greatly alter total protein levels in motor neurons.

C9orf72 is transcribed off of both the sense and anti-sense strands, in both normal and diseased cells (Fig. 3A) (7, 15, 40, 41). Our data suggest that sense transcription of the repeat region starts from exon 1A, since excision of exon 1A closed the gap in “undetectable” sense transcript (Fig. 1C). However, it is unresolved where anti-sense transcription initiates. Regardless, both sense and anti-sense transcripts harboring the repeat expansion are translated through noncanonical RAN translation to form five DPRs thought to be toxic (6, 7, 10). These DPRs are likely to be variable in size. They deposit in the brains of C9orf72 mutation carriers but not in nondiseased controls and are not found in other neurodegenerative conditions (i.e., DPRs are a specific C9-pathology) (4, 42). The absence of DPRs in WT cells may reflect the rapid splicing out of the repeat region from sense 1A transcripts, the inefficiency of noncanonical RAN translation when there are too few repeats, or a combination of both.

We tested whether removing the repeat region could abrogate DPR production. We evaluated ten antibodies targeting each of these DPRs using Meso Scale Discovery’s (MSD) enzyme-linked immunosorbent assay (ELISA) immunoassay. We found two antibody combinations that could reliably detect the presence of poly-GA and poly-GP DPRs above the background level defined by our knock-out (KO) line (28, 39). The other antibodies had comparable signal in the KO line and the unedited C9-patient line, suggesting they were not specific to the C9orf72 DPRs. Using these two combinations, we measured poly-GA and poly-GP DPRs in 2-wk-old neurons derived from our C9-isogenic series.

The poly-GA is encoded exclusively on the sense strand, whereas the poly-GP can come from either strand (Fig. 3A). As expected, Poly-GA expression was eliminated by removal of the REx and excision of the mutant allele (HET(Mut)x), or by elimination of sense transcription through excision of exon 1A (1Ax) (Fig. 3B). Also as expected, all poly-GP was eliminated by removal of the repeat expansion or mutant allele. However, some poly-GP remained after excision of exon 1A (Fig. 3C, 1Ax). Since we know that sense 1A transcript is eliminated in the 1Ax line (Fig. 1 C, E, and G), the poly-GP we detect in this line must arise from antisense mutant transcript. We quantify DPR (poly-GP) expression specifically from the anti-sense transcript in patient-derived motor neurons rather than in artificial expression systems. Our data indicate that at least one-third to one-half of poly-GP derives from antisense transcription, and confirm that eliminating toxic DPRs requires the removal of the expanded repeat region. In total, these data also indicate that removal of the repeat expansion, rather than silencing sense expression, is required to completely eliminate DPRs.

A surprising finding is that excision of the WT allele more than doubled the amount of poly-GP expressed from the mutant allele (Fig. 3C, HET(WT)x). By contrast, poly-GA expression was not changed by excision of the WT allele (Fig. 3B, HET(WT)x), consistent with our earlier observation that sense transcription from one allele was unaffected by removal of the other allele (Figs. 1 E–H and 3B). Because poly-GP is translated from both sense and antisense transcripts, we asked whether an increase in antisense transcription from the loss of the WT allele could account for the increase in poly-GP expression in the HET(WT)x induced motor neuron. We measured intron-containing (antisense) RNA using ddPCR across the edited C9-induced motor neurons. Removal of the WT allele lowered antisense transcription by one-third in C9-HET(WT)x compared to C9-unedited (Fig. 3D). Additionally, elimination of sense transcription by excising exon 1A also lowered antisense transcription by one-third and was not statistically different than removal of the WT allele (Fig. 3D, 1Ax vs. HET(WT)x). Altogether, these findings suggest that the upregulation of poly-GP in HET(WT)x induced motor neurons (Fig. 3C) is not driven by an increase in either the sense or antisense transcripts we measured.

The pathological hallmark of C9-FTD/ALS is loss of nuclear TDP-43 and aggregation of cytoplasmic TDP-43 in affected neurons (43). These events are thought to be independent (43) and have been difficult to model in cellular and animal systems. Here we detected a clear loss of nuclear TDP-43 in 57% of TDP-43-positive motor neurons derived from our unedited patient iPSC line (Fig. 4A and SI Appendix, Fig. S13 pink arrows). This effect was age dependent, rising significantly 6 wk after induction of differentiation. The reason we could detect it is that we were able to maintain our motor neurons as a monoculture for 2 mo, after improving published protocols (29) for the generation of doxycycline-induced motor neurons by adding three growth factors (brain-derived neurotrophic factor (BDNF), glial cell line-derived neurotrophic factor (GDNF) and neurotrophin-3 (NT-3)). We quantified the loss of nuclear TDP-43 in the edited C9 lines that eliminated all RNA and DPR pathology (REx and HET(Mut)x) or silenced sense expression of the repeat expansion (1Ax) that at 7 wk postinduction. We found that on average, less than 30% of the TDP-43 positive cells lacked nuclear TDP-43 after removal of the repeat expansion (REx), the mutant allele (HET(Mut)x) or exon 1A (1Ax) (Fig. 4B). As the loss of nuclear TDP-43 was apparent only after aging motor neurons in culture for 7 wk postinduction, our system appears to recapitulate the interaction between age and genotype that is an important aspect of human C9-FTD/ALS. To our knowledge, TDP-43 pathology has not been detected in cultured human C9orf72 mutant cells so far and has only been detectable in two other cell culture systems [GRN mutant (44) and TDP-43 (45) mutant cell lines] without artificially altering the TDP-43 expression and in the absence of drug treatment.

We noted that REx and HET(Mut)x cultures still harbored some motor neurons with nuclear loss of TDP-43, even though they lack the repeat expansion, and that removing ameliorated nuclear TDP-43 loss, even though it maintains antisense expression of the expanded repeat. Although a hallmark of C9-FTD/ALS, TDP-43 pathology is not specific to this gene mutation. It is also a classic pathological finding for other genetic and sporadic forms of FTD/ALS. In our cell culture system, we hypothesize the nuclear loss of TDP-43 is driven by the stress of aging to 7 wk, and exacerbated by the expression of the mutation. The take-away message is that the C9-unedited motor neurons had a greater pathological response to aging (i.e., more TDP-43-positive cells lost their nuclear TDP-43 expression) than did the edited cell motor neurons.

We next measured the electrophysiologic effect of removing the repeat expansion in motor neurons. We examined population-level electrophysiology using multielectrode arrays (MEAs) at 3 wk postinduction and single neuron function using whole-cell patch-clamp at 6 wk postinduction. The difference in timepoints used for these analyses was necessitated by differing rates of electrophysiological maturation in iPSC-derived neuronal cultures maintained at different culture densities; high densities for MEA recordings and low, subconfluent densities for whole-cell patch-clamp recordings.

On MEAs, burst firing is defined as a period of activity in which the time between detected depolarizing spikes (interspike interval) is less than 100 ms for a minimum of ten consecutive spikes. Increased burst activity is an indicator of increasing functional maturity in cultured neurons (46–48). Furthermore, network bursts indicate detection of simultaneous burst activity across multiple electrodes in a single array and suggest the formation of synaptic networks in cultured neurons.

The C9-unedited and 1Ax motor neurons showed no network burst activity at 3 wk of age; in contrast, both REx and HET(Mut)x motor neurons did demonstrate network burst activity (Fig. 5A), indicating that removing the repeat expansion, but not just silencing sense expression, enhances functional maturity and network forming capabilities in cultured motor neurons. Despite a lack of network activity, C9-unedited and 1Ax motor neurons were still spontaneously active (Fig. 5B). Spontaneous mean firing rate activity was nearly doubled (1.4 ± 0.17 Hz) in REx motor neurons and tripled (2.92 ± 0.48 Hz) in HET(Mut)x motor neurons compared to C9-unedited motor neurons (0.93 ± 0.11 Hz) (Fig. 5B).

It has been previously shown that C9orf72 mutant iPSC-derived motor neurons have a decreased capacity to fire repetitive action potential trains compared with controls (49). Such data indicate that poor firing capacity is a phenotypic identifier for C9-ALS neurons in culture and suggest that interventions that increase firing capacity may therefore have therapeutic benefit. Using whole-cell patch clamp, we found that 500 ms, 2 nA current injections led to an increase in the percentage of induced motor neurons capable of repetitive firing among motor neurons in which the mutant repeat had been removed compared to C9-unedited [Fig. 5C, dark gray, REx, HET(Mut)x)] but not among motor neurons in which sense transcription is silenced (Fig. 5C, 1Ax). This increase in repetitive firing capacity occurred at the single-cell level and was not dependent on network activity. Nor was it dependent on individual action potential characteristics, as the motor neurons in each group showed similar resting membrane potentials (Fig. 5E), spike amplitudes (Fig. 5F), and action potential durations (Fig. 5G).

We did not compare the C9-isogenic series to the WT-series on electrophysiology (or TDP-43 pathology) because the genetic backgrounds between the two series would be confounding. We have shown previously that baseline electrophysiology differs significantly between iPSC-derived motor neurons with different genotypes, even in the absence of disease-causing mutations (48). These differences necessitate electrophysiological comparisons be performed between isogenic groups. Previous work has shown that iPSC-derived motor neurons harboring C9orf72 and TDP-43 mutations exhibit reductions in repetitive firing behavior by patch (49) and burst firing on MEAs (45). Therefore, we interpret the capacity for targeted edits that remove the repeat expansion to improve these metrics in our C9-mutant line as an improvement in overall functional performance.

In summary, we found an improvement of in vitro electrophysiologic markers of neuronal functional capacity for network activity and repetitive firing after removal of the repeat expansion biallelically (REx) or in an allele-specific manner (HET(Mut)x). This effect highlights that deficits due to the C9orf72 mutation affect adaptive neuronal function, as the C9-unedited motor neurons demonstrated normal single-spike action potential properties at the single-neuron level.

We used a deidentified patient iPSC line harboring the C9orf72 mutation and a control cell line without mutation (WT-control). We first knocked in the inducible motor neuron transcription factor transgene cassette in the CLYBL safe-harbor locus. We then edited the lines using HiFi spCas9 (Macrolabs, UC Berkeley) and two gRNAs (SI Appendix, Table S1). We used PacBio single-molecule sequencing to size the repeat expansion, detect repeat expansion excision in C9-REx and detect methylation after editing across C9-patient lines. Further details on iPSC cell line generation (SI Appendix, Figs. S1–S11), maintenance, single-molecule sequencing, and motor neuron differentiation are detailed in SI Appendix, Supplemental Methods.

500 ng of RNA from 2-wk-old induced motor neurons was run on the QX100 Droplet Reader (Bio-Rad 186-3002) with three technical replicates of each of three biologic replicates (independent wells of differentiated motor neurons) using primers and probes in SI Appendix, Table S2. See SI Appendix, Supplemental Methods for full details.

C9orf72 protein quantification by WES (Bio-Techne) was performed according to the manufacturer’s protocol and primary antibodies mouse anti-C9orf72 (GeneTex, GTX634482 at 1:100 and rabbit anti-GAPDH (AbCam, AB9485) at 1:1,000. Duplexed secondaries included 9.5 µL of mouse (ProteinSimple, DM-002) and 0.5 µL of 20× anti-rabbit (ProteinSimple, 043-426). See SI Appendix, Supplemental Methods for full details.

We used the Small Spot Streptavidin Plate (L45SA, MSD). Poly-GA was detected using anti-GA antibody (MABN889, Millipore) at 1 µg/mL (capture) and 2 µg/mL (detect) final concentration and 18 µg total protein per sample (blocking buffer A, solution PBS). Poly-GP was detected using anti-GP antibody (affinity purified TALS828.179 from TargetALS, purification lot A-I 0757 and stock concentration 1.39 mg/mL) at a final concentration of 2 µg/mL capture and 4 µg/mL detected with 18.5 µg total protein per sample (blocking buffer A, solution TBS). See SI Appendix, Supplemental Methods for full details.

TDP-43 immunocytochemistry and quantification was performed in fixed 7-wk-old induced motor neurons. Primary antibodies: rabbit anti-TDP43 (10782-2-AP, Proteintech) at 1:500, beta-III-tubulin (480011, Invitrogen) at 1:250 incubated overnight at 4 °C. Secondary antibodies: goat anti-rabbit Alexa Fluor 488 nm and Goat anti-mouse Alexa Fluor 594 nm incubated at room temperature for 1 h. DAPI (D1306, ThermoFisher Scientific) for 5 min at room temperature. See SI Appendix, Supplemental Methods for full details.

Detailed protocols for induced motor neuron MEA and whole-cell patch clamp are available in SI Appendix, Supplemental Methods.

B.R.C. and C.D.C. conceived the study; C.D.C. designed research; A.S., K.G., M.S., A.M.B., O.A.A., K.A.B., R.S.C., E.P., M.T.B., K.C.K., Y.-C.T., A.S.T.S., and C.D.C. performed research; H.L.W., K.C.K., and C.D.C. contributed new reagents/analytic tools; A.S., K.G., M.S., A.M.B., K.A.B., P.O.I.-L., K.C.K., Y.-C.T., A.S.T.S., and C.D.C. analyzed data; H.L.W. trained on analysis tools, provided feedback on assay development; and C.D.C. wrote the paper.

B.R.C. is a founder of Tenaya Therapeutics, a company focused on finding treatments for heart failure, including genetic cardiomyopathies. B.R.C. holds equity in Tenaya. C.D.C. is a founder, with equity, in Ciznor Co., a CNS therapeutics company. B.R.C. and C.D.C. are inventors on a patent application for this work. Y.-C.T. is a full-time employee at PacBio, a company developing single-molecule sequencing technologies. The other authors declare no competing interests.