Technologies that can safely edit genes in the brains of adult animals may revolutionize the treatment of neurological diseases and the understanding of brain function. Here, we demonstrate that intracranial injection of CRISPR–Gold, a nonviral delivery vehicle for the CRISPR–Cas9 ribonucleoprotein, can edit genes in the brains of adult mice in multiple mouse models. CRISPR–Gold can deliver both Cas9 and Cpf1 ribonucleoproteins, and can edit all of the major cell types in the brain, including neurons, astrocytes and microglia, with undetectable levels of toxicity at the doses used. We also show that CRISPR–Gold designed to target the metabotropic glutamate receptor 5 (mGluR5) gene can efficiently reduce local mGluR5 levels in the striatum after an intracranial injection. The effect can also rescue mice from the exaggerated repetitive behaviours caused by fragile X syndrome, a common single-gene form of autism spectrum disorders. CRISPR–Gold may significantly accelerate the development of brain-targeted therapeutics and enable the rapid development of focal brain-knockout animal models.
Programmable RNA-guided endonucleases have the potential to revolutionize the treatment of neurological diseases because of their ability to cut genes with sequence specificity1,2,3,4,5. Despite their potential, the translational impact of RNA-guided endonucleases on the central nervous system has been limited due to challenges in performing efficient gene editing in adult brains with minimal toxicity. Currently, gene editing in the adult brain is mainly accomplished through the viral delivery of clustered regularly interspaced short palindromic repeats (CRISPR)–CRISPR-associated protein 9 (Cas9)6. However, the translation of viral delivery methods for CRISPR–Cas9 and single guide RNA (sgRNA) in the brain can be challenging because of the immunogenicity of viruses7 and the genomic damage caused from the prolonged expression of CRISPR–Cas9 and sgRNA8. Moreover, the toxicity caused from the continuous expression of foreign proteins in neurons, which frequently causes changes in neuronal phenotypes9, is a challenge. Therefore, there is great interest in developing nonviral methods for delivering RNA-guided endonucleases into the brains of adult mice. However, very little is known about the ability of RNA-guided endonucleases to transfect the brains of adult animals via nonviral methods. Currently, there is only one report of nonviral gene editing in the adult brain, which used an intracranial injection of Cas9 ribonucleoproteins (RNPs) engineered with multiple nuclear localization sequence (NLS) signals10. In addition, there are no reports of nonviral gene editing in the brain with RNA-guided endonucleases beyond Cas9, making it unclear whether other RNA-guided endonucleases, such as CRISPR from Prevotella and Francisella 1 (Cpf1), can perform gene editing in the adult brain via nonviral delivery methods.
A wide number of neurological diseases could potentially be treated with gene editing-based therapeutics. Of these, fragile X syndrome (FXS) is a particularly attractive target for gene editing-based therapeutics because of the current lack of therapeutic options and the high morbidity that this disease causes. FXS is an autism-associated single-gene mutation-based disorder, which is driven by a repeat expansion mutation in the fragile X mental retardation 1 (FMR1) gene, which encodes the fragile X mental retardation protein (FMRP), an mRNA-binding protein11. FXS is the most common inherited form of intellectual disability and a common single-gene form of autism spectrum disorders (ASDs), accounting for ~2.1% of patients12. Current drug treatments, such as psychostimulants, antidepressants and antipsychotics, are ineffective because they do not address the underlying aetiology of FXS; they only target individual symptoms13,14. In addition, current FXS pharmacotherapies cause severe side effects such as weight gain and sedation14. New treatments for FXS are urgently needed; however, developing FXS treatments based on traditional small molecules has been challenging because of the limited number of validated FXS therapeutic targets and the difficulties associated with developing therapeutics that can provide long-term effectiveness without causing cytotoxicity to brain cells.
CRISPR-based editing of the brain, generated by a local intracranial injection, has great potential for treating FXS because it will lead to localized gene editing in the brain. Consequently, the patient is spared the the toxic effects of globally inhibiting neuronal signalling pathways. In addition, CRISPR gene editing is permanent and would make repeated injections unnecessary, making it feasible in a variety of clinical scenarios. There are several genes that are potential targets for a CRISPR-based FXS therapy; of these, mGluR5 is particularly attractive because its exaggerated signalling has been demonstrated to be associated not only with FXS but also with other ASDs15,16,17,18,19,20. The importance of modulating mGluR5 in ASDs triggered several pharmaceutical companies to develop small molecule-based therapies that target mGluR5; however, these small molecules failed in clinical trials21,22. Therefore, knocking out the mGluR5 gene (Grm5) locally in brain regions that are hypothesized to cause the behavioural phenotypes in patients with FXS has great potential for treating FXS and autism-related disorders. However, several questions need to be answered before gene editing can become a therapeutic treatment for FXS. First, the delivery challenges associated with gene editing in the brain with Cas9 RNPs, as described above, need to be solved. Second, it is unclear whether mGluR5-mediated behavioural phenotypes are caused by focal overactivation of mGluR5 signalling and, if so, which parts of the brain need to have the exaggerated mGluR5 signalling reduced to ameliorate a specific behavioural phenotype.
Here, we investigated whether a recently developed nonviral Cas9 delivery vehicle, termed CRISPR–Gold23, could deliver the RNA-guided endonucleases Cas9 and Cpf1 into the brains of adult mice and we performed gene editing in Thy1-YFP and Ai9 mice. We then targeted the mGluR5 gene to reduce the exaggerated mGluR5 signalling in the striatum of a mouse model of FXS. We show that the CRISPR–Gold-mediated mGluR5 reduction rescued striatum-dependent exaggerated repetitive behaviours, as measured by the marble-burying assay and jumping behaviour of the mice. Our results demonstrate that CRISPR–Gold has the potential to significantly accelerate the development of new brain-targeted therapeutics. Moreover, this platform can enable the rapid development of focal brain-knockout models for mechanistic, brain region or preclinical studies given its ability to edit genes in adult brains via the nonviral delivery of Cas9 RNPs.
Biocompatibility test of CRISPR–Gold in neurons
Nonviral delivery vehicles that can edit genes in the brain have great potential for treating neurological disorders. However, very little is known about nonviral gene editing in the brain. We identified CRISPR–Gold (Fig. 1a) as a potential delivery vector for gene editing in the brain because it was able to efficiently deliver Cas9 RNPs into a variety of cell types in vitro and into mouse muscles23. To investigate the biocompatibility of CRISPR–Gold in neuronal cells, we measured the cytotoxicity and physiological effects of primary cultured hippocampal neurons treated with CRISPR–Gold loaded with Cas9 RNPs (the formation of the CRISPR–Gold complex was verified, as shown in Supplementary Fig. 1). The concentration of CRISPR–Gold applied to the neuronal cells was 25 pmole per 0.5 ml for both Cas9 and sgRNAs, 17 ng per 0.5 ml of gold, and 2.5 µg per 0.5 ml of PAsp(DET). Two hours after treatment, the residual CRISPR–Gold was removed and replaced with fresh growth medium plus conditioned medium (50:50). After 10–14 days of CRISPR–Gold treatment, neurons were patch-clamped or stained with SYTOX Red to count dead cells. We first checked the electrophysiological properties (whole-cell current clamp recording) of pyramidal neurons after treatment with CRISPR–Gold. The results showed that there was no difference in the membrane potentials between control and treated neurons (Fig. 1b). Input resistance, which in part indicates the leakiness of the plasma membrane, in the treated neurons was not significantly different from the untreated neurons (Fig. 1c). Consistent with the input resistance result, the number of spikes generated by a 200 pA current injection did not significantly change in CRISPR–Gold-treated neurons (summarized in Fig. 1d, representative traces in Fig. 1e). Therefore, we conclude that CRISPR–Gold treatment does not have adverse effects on neuronal membrane health or specifically affect neuronal excitability. We then stained dead cells with SYTOX and visualized cell morphology by co-staining actin with a phalloidin stain to check the cytotoxicity of CRISPR–Gold treatment. No significant differences in the number of dead cells and neuronal morphology were found in CRISPR–Gold-treated cells compared with untreated neurons (Fig. 1f). Taken together, our results show that treatment with CRISPR–Gold is neither cytotoxic nor does it affect the physiological function of neurons under the conditions tested here. However, we do not know the elimination kinetics of the gold nanoparticle (GNP) core of CRISPR–Gold from neurons or whether extended retention of the gold core in neurons leads to toxicity.
CRISPR–Gold-mediated gene editing in the brains of Thy1-YFP mice
To test our idea of whether CRISPR–Gold could deliver Cas9 RNPs in an adult mouse brain, we stereotaxically injected CRISPR–Gold into the brains of adult mice (Fig. 2a). We selected neurons as the initial target for gene editing investigations because of their essential function in brain activities and their strong correlation with a wide variety of neurological diseases. The Thy1-YFP mouse model, a transgenic mouse line that expresses YFP only in neurons and not in other types of brain cells24, was used to monitor gene editing in neurons (Fig. 2a). The sgRNAs for Cas9 and CRISPR RNAs (crRNAs) for Cpf1 were designed to target the 5′ region of the YFP gene to induce insertion/deletion (indel) mutations (Supplementary Fig. 2a). These guide RNAs were verified by checking the YFP gene knockout capability in YFP-expressing HEK cells (Supplementary Fig. 2b–d). CRISPR–Gold loaded with Cas9 or Cpf1 RNPs targeting the 5′ region of the YFP gene was then injected into the dentate gyrus in the hippocampus of 1–2-month-old adult mice (Fig. 2b). Figure 2c,d and Supplementary Fig. 3 show that CRISPR–Gold can deliver Cas9 and Cpf1 RNPs and efficiently edit the YFP gene in neurons that are projecting to the molecular layer of the dentate gyrus after the stereotaxic injection of CRISPR–Gold with Cas9 or Cpf1 RNPs. As a result, there was an ~17% and ~28% decrease in YFP+ cells in the granular layer of the dentate gyrus by Cas9 and Cpf1, respectively (Fig. 2c,d). Moreover, there was a 34% and 25% decrease in YFP expression levels (Supplementary Fig. 3a,b) in the molecular layer of the dentate gyrus after the delivery of Cas9 and Cpf1, respectively. These results demonstrate that Cas9 and Cpf1, delivered by CRISPR–Gold, are efficient methods for local in vivo editing in the brain. Interestingly, modulating small subpopulations of neurons (200–300 neurons) in focal brain regions has been shown to change behavioural outputs25,26. CRISPR–Gold may therefore be able to treat neurological disorders and elucidate the function of genes in brain circuits, despite only editing a small fraction of the total number of brain cells after an intracranial injection.
CRISPR–Gold-mediated gene editing in the brains of Ai9 mice
Gene editing via deletion of repeated genetic sequences in a human brain can be a potential cure for disorders such as Huntington disease and FXS, which have repeated sequences that result in brain dysfunctions27,28,29. Therefore, we performed experiments in Ai9 mice to determine whether Cas9 and Cpf1 RNPs could induce genome editing by deleting a target gene in an adult mouse brain (Figs. 3a,b and 4a). The Ai9 mouse is a genetically engineered mouse model that has a fluorescent tdTomato gene with a stop sequence upstream of it30. In this mouse model, tdTomato is silent because of the stop signal, but the deletion of the stop sequences allows transcription of the tdTomato gene, resulting in fluorescence expression (Fig. 3a). sgRNAs for Cas9 and crRNAs for Cpf1 were designed to target both ends of the stop sequences to remove them, leading to the expression of tdTomato (Supplementary Fig. 4a). These guide RNAs were verified in primary fibroblasts cultured from Ai9 mice, and were able to induce the expression of tdTomato (Supplementary Fig. 4b). Cas9- or Cpf1-loaded CRISPR–Gold was stereotaxically injected into two brain regions (the hippocampus and the striatum of 1–2-month-old adult Ai9 mice as shown in Figs. 3b and 4a), and the expression of tdTomato was measured via fluorescence histology.
Figure 3c,d demonstrates that both Cas9 and Cpf1 CRISPR–Gold complexes can induce deletion of their target sequences and can induce the expression of tdTomato in the CA1 region of the hippocampus of Ai9 mice. For example, fluorescence histology images of the hippocampus area treated with Cas9 or Cpf1 CRISPR–Gold complexes showed clear expression of tdTomato compared with the contralateral control side (Fig. 3c,d). Approximately 10–15% of the cells in the injected area of the hippocampus were tdTomato+ in both Cas9 and Cpf1 RNP-injected brains; these cells were quantified by analysing the percentage of cells stained positive for both tdTomato and 4,6-diamidino-2-phenylindole (DAPI) (Fig. 3c,d). Similar results were found in the striatum (Fig. 4b,c). For example, tdTomato is clearly expressed in the striatum after an injection of Cas9 or Cpf1 CRISPR–Gold complexes compared with the contralateral control side (Fig. 4b,c). Approximately 10% of the cells in the injected area of the striatum were tdTomato+ in both Cas9 and Cpf1 RNP-injected brains (Fig. 4b,c). These results suggest that Cas9 and Cpf1 RNPs can efficiently delete targeted DNA sequences using the CRISPR–Gold delivery vehicle. The gene-edited area was 1–2 mm × 1–2 mm for the hippocampus and striatum from the injection sites, and within these regions, there was no observable change in cell density by counting DAPI+ cells (Fig. 4d,e), indicating that cell viability is not significantly adversely affected. A potential limitation of CRISPR–Gold is its GNP core, which can potentially accumulate in the brain and cause toxicity. However, the gold core constitutes a minor fraction of the total weight of CRISPR–Gold, and the dose of gold used in the mouse experiments was ~2.84 µg kg–1, which is orders of magnitude lower than the toxicity threshold for GNPs in the brain31,32.
CRISPR–Gold-mediated gene editing in multiple cell types
Neurons are the basic working units of the brain, whereas non-neuronal cells play a central role in maintaining, supporting and regulating neuronal functions. Glial cell dysfunction causes multiple brain disorders33, and there is great interest in editing the genes of glial cells. Therefore, we used Ai9 mice to further identify the brain cell types edited by Cas9 or Cpf1 RNPs delivered by CRISPR–Gold and to determine whether glial cells were edited. CRISPR–Gold-injected Ai9 brain sections were stained with the following cell markers: glial fibrillary acidic protein (GFAP), ionized calcium-binding adapter molecule 1 (IBA1) and neuronal nuclear protein (NeuN). GFAP, IBA1 and NeuN identify the astrocyte, microglia and neuronal populations, respectively, in histology sections. Supplementary Figures 5 and 6 demonstrate that gene-edited tdTomato-expressing cells were composed of all three of these cell types (astrocytes, microglia and neurons). In the hippocampus, more than half of the tdTomato+ cells had the astrocyte marker GFAP in Cas9 or Cpf1 RNP-injected brains, and IBA1- and NeuN-stained cells accounted for approximately 40% and 10% of the tdTomato+ cells, respectively (Supplementary Fig. 5b,d, left panels). This result suggests that astrocytes, microglia and neurons are edited by CRISPR–Gold Cas9 and Cpf1 in the hippocampus. Similar results were found in the striatum. In the striatum of brains injected with Cas9 or Cpf1, more than half of the tdTomato+ cells had the astrocyte marker GFAP. In addition, cells stained with IBA1 and NeuN accounted for 10–30% of tdTomato+ cells in brains injected with Cas9 or Cpf1 (Supplementary Fig. 6b,d, left panels). In the hippocampus of brains injected with Cas9, 33% of GFAP+, 19% of IBA1+ and 3% of NeuN+ cells were edited among astrocytes, microglia and neurons (Supplementary Fig. 5b, right panels). For Cpf1, 65% of GFAP+, 21% of IBA1+ and 5% of NeuN+ cells were edited among astrocytes, microglia and neurons (Supplementary Fig. 5d, right panels). Similar results were found in the striatum, with 50% (Cas9) and 46% (Cpf1) of GFAP+, 18% (Cas9) and 14% (Cpf1) of IBA1+, and 3% (Cas9) and 7% (Cpf1) of NeuN+ cells edited among astrocytes, microglia and neurons (Supplementary Fig. 6b,d, right panels). Taken together, these results demonstrate that CRISPR–Gold-delivered Cas9 and Cpf1 RNPs can induce deletion of target genes in the major cell types of the brain, including astrocytes, microglia and neurons.
Gene editing in Fmr1 knockout mice with CRISPR–Gold
Our results indicated that CRISPR–Gold has the potential to treat brain disorders that are caused by the overexpression of genes in localized brain regions. We selected Grm5 as a potential target for CRISPR–Gold-based therapeutic gene editing because a wide number of studies have demonstrated that exaggerated mGluR5 signalling can generate FXS pathophysiology. Therefore, knocking out the mGluR5 gene through nonviral CRISPR gene editing in specific regions of the brain may be a therapeutic way to treat patients with FXS. However, it is unclear whether exaggerated mGluR5 signalling-mediated FXS phenotypes are caused by focal overactivation of (versus global) mGluR5 signalling, and, if so, which parts of the brain need to have the mGluR5 gene deleted to ameliorate the specific behavioural phenotypes. Furthermore, the delivery challenges associated with gene editing in the brain with Cas9 RNPs need to be solved to be used as treatments of brain disorders. To address these two unresolved issues in therapeutic brain gene editing, we generated CRISPR–Gold Cas9 sgRNA RNPs targeting Grm5, and investigated their ability to knockout the mGluR5 gene in vivo and rescue mice from the behavioural phenotypes of FXS using Fmr1 knockout mice, a mouse model of FXS. We first confirmed that CRISPR–Gold-mediated mGluR5 gene editing is successful, in vitro and in cells, as shown in Supplementary Fig. 7. We then stereotaxically injected either saline vehicle (control) or CRISPR–Gold targeting the mGluR5 gene (mGluR5–CRISPR) into the striatum of wild-type or Fmr1 knockout mice to determine whether CRISPR–Gold could knockout the mGluR5 gene in vivo in the striatum after a direct local injection (Fig. 5a). Tracking of indels by decomposition (TIDE) analysis indicated that there was a frequency of 14.6% in mGluR5 gene mutations (Supplementary Fig. 8a), and that there were no significant off-target effects (Supplementary Fig. 8b). As a result, the mRNA levels and the protein levels of mGluR5 were reduced by 40–50% both in wild-type and Fmr1 knockout mice, which was confirmed by quantitative PCR with reverse transcription (RT–qPCR) (Fig. 5b) and immunostaining analysis (Fig. 5c). Mice treated with mGluR5–CRISPR also had no evidence of an increased immune response, as measured by mRNA levels of microglial markers in mGluR5–CRISPR-treated brains (Supplementary Fig. 9).
Behavioural rescue in Fmr1 knockout mice by CRISPR–Gold
Commonly known repetitive behaviours of mice with autistic phenotypes include excessive digging behaviour, which can be observed in the marble-burying assay, and increased jumping, which can be observed during empty cage observations34. Given that the striatum is an important brain region in mediating repetitive behaviours35, we wanted to test whether knocking out the mGluR5 gene by CRISPR–Gold delivery of Cas9 RNPs can rescue the exaggerated repetitive behaviours shown in Fmr1 knockout mice36. Figure 6 shows the effect of injecting saline (control) or mGluR5–CRISPR into the striatum on repetitive behaviours in wild-type and Fmr1 knockout mice. In the marble-burying assay, Fmr1 knockout mice injected with saline buried significantly more marbles than wild-type mice. Conversely, injection with mGluR5–CRISPR into the striatum significantly rescued the excessive digging phenotype of Fmr1 knockout mice back to normal, while having no significant effect on wild-type mice (Fig. 6a, representative movies are shown in Supplementary Video 1). Likewise, the same effect was shown in jumping behaviour in Fmr1 knockout mice, whereby those injected with mGluR5–CRISPR showed a phenotype much more comparable to wild-type mice compared with Fmr1 knockout mice injected with saline (Fig. 6b, representative movies are shown in Supplementary Video 2).
To determine whether the rescue of excessive digging and jumping behaviour in Fmr1 knockout mice were the consequences of potentially reduced hyperlocomotor activities induced by mGluR5–CRISPR, locomotor activities in these mice were also assessed. This behaviour can be assessed via line crossings during empty cage observations37, the distance that a mouse travels during an open-field activity assay38, and how long it takes for a mouse to fall in the accelerated rotarod performance test39. While there was a significant difference between wild-type and Fmr1 knockout mice when it comes to these phenotypes, there was no significant change in locomotor activity in Fmr1 knockout mice that had been treated with mGluR5–CRISPR. Fmr1 knockout mice treated with mGluR5–CRISPR performed similarly to control Fmr1 knockout mice in the three locomotor activity experiments (Fig. 6c and Supplementary Fig. 10), suggesting that exaggerated mGluR5 signalling in the striatum does not mediate the hyperlocomotor activity shown in Fmr1 knockout mice. In other words, increased repetitive behaviours were specifically rescued by local treatment of mGluR5–CRISPR to the striatum of Fmr1 knockout mice. Finally, mGluR5–CRISPR-treated Fmr1 knockout mice had similar body weights to control Fmr1 knockout mice (Supplementary Fig. 11). Taken together, these results demonstrate that gene editing via the nonviral delivery of CRISPR into a local brain region can rescue specific behavioural phenotypes in an autism mouse model.
CRISPR–Gold was able to deliver CRISPR–Cas9 and Cpf1 and edit genes in local regions of the brain after an intracranial injection. Although local gene editing may treat numerous behavioural deficits, gene editing in broad areas of the brain may be necessary for a number of neurological disorders. For these types of brain disorders, alternative strategies for delivering CRISPR–Cas9 will need to be developed, such as convection-enhanced delivery of CRISPR–Cas9 or delivery into the cerebrospinal fluid via intrathecal injection. In addition, CRISPR–Gold has broad cell tropism and can edit multiple cell types in the brain after a local injection. Although targeting diverse cell types other than neurons, including astrocytes and microglia, can be useful (as neurons as well as non-neuronal cells are important targets in numerous diseases), a gene-editing therapy that can target a specific cell type, such as neurons, may be also needed. Finally, as the GNP core of CRISPR–Gold can potentially accumulate in the brain after multiple injections, this may prevent CRISPR–Gold from being injected into patients multiple times. However, a single injection into the brain of an extremely low dose of gold in CRISPR–Gold (2.84 µg kg–1) efficiently rescued behavioural deficits in mice, and appears to be well tolerated in the brain.
We have showed that CRISPR–Gold can deliver both Cas9 and Cpf1 RNPs in the brain after an intracranial injection and can edit genes. Currently, gene editing in the adult brain is mainly accomplished through the viral delivery of CRISPR–Cas96. However, there is great interest in developing nonviral methods for delivering RNA-guided endonucleases into the brain due to the potential toxicity caused from using viruses to edit neurons7,8,9. Cas9 RNPs engineered with multiple NLS signals can edit genes in the adult brain when intracranially injected10; however, therapeutic gene editing in the brain has still not been accomplished via RNP delivery. CRISPR–Gold-mediated gene editing in adult mouse brains has multiple unique features. CRISPR–Gold was able to edit genes in neurons and in non-neuronal cells important for brain function, including microglia, which have been difficult to edit via other methods, such as viral delivery40,41,42. In addition, CRISPR–Gold was able to inhibit 40–50% of the autism causal gene Grm5 in the striatum after an intracranial injection, and this level of inhibition was able to rescue mice from increased repetitive behaviours, which is one of the core symptoms of ASDs. Importantly, CRISPR–Gold-mediated editing was localized to the striatum, suggesting that global inhibition of neuronal signalling pathways is not needed to rescue autism-associated behavioural phenotypes. Taken together, CRISPR–Gold-mediated gene editing of adult mouse brains, via the nonviral delivery of Cas9 RNPs, can rescue mice from behavioural deficits and has the potential to enable the rapid development of focal brain-knockout animal models.
Oligonucleotides were purchased from Integrated DNA Technologies. GNPs (60 nm) were purchased from BBI Solutions. Sodium citrate and HEPES were purchased from Mandel Scientific. Sodium silicate was purchased from Sigma-Aldrich. Phusion High-Fidelity DNA Polymerase was purchased from NEB. The MEGAscript T7 kit, the MEGAclear kit, the PageBlue solution, the propidium iodide and the PureLink Genomic DNA kit were purchased from ThermoFisher. Mini-PROTEAN TGX Gels (4–20%) were purchased from Bio-Rad. DMEM media, non-essential amino acids, penicillin–streptomycin, Dulbecco's PBS and 0.05% trypsin were purchased from Life Technologies. Amicon Ultra-4 30 kDa was purchased from EMD Millipore. Cas9 proteins with an amino-terminal 6xHis-tag and two SV40 NLS peptides at the carboxyl terminus and Cpf1 proteins were purchased from Macrolab UC Berkeley. The Poly (PAsp(DET)) was a gift from Kazunori Kataoka’s group43,44. The SYTOX Red Dead Cell Stain (excitation/emission of 640/658 nm) and the SuperScript III First-Strand Synthesis kit were purchased from Invitrogen.
The mouse monoclonal RFP antibody (6G6) was purchased from ChromoTek, the chicken polyclonal GFP antibody (GFP-1020) from Aves Labs (Tigard), the rabbit polyclonal GFAP antibody (AB5804) and the mouse monoclonal NeuN antibody (MAB377) from Millipore, the rabbit polyclonal IBA1 antibody (019-19741) from Wako Chemicals, and the rabbit polyclonal mGluR5 antibody (AGC-007) was from Alomone Labs. The goat anti-mouse IgG2a-Cy3, goat anti-chicken-Cy2, goat anti-rabbit-Cy5, goat anti-mouse IgG1-Cy5 and donkey anti-rabbit IgG-Alexa Fluor 647 antibodies were purchased from Jackson ImmunoResearch Laboratories.
In vitro T7 transcription of sgRNA and crRNA
The DNA templates for the in vitro transcription of sgRNAs and crRNAs (sgRNAs for Cas9 and crRNAs for Cpf1) were prepared by PCR. Sequences of the template and primers are listed in Supplementary Tables 1 and 2. Two sequences (left and right) of guide RNAs were used for the Ai9-targeting experiments. A single sequence for sgRNAs and crRNAs for targeting the YFP gene (yfp) and sgRNAs for targeting Grm5 was used. PCR amplification was performed with Phusion Polymerase according to the manufacturer’s protocol. RNA in vitro transcription was performed using the MEGAscript T7 kit, and purification of the resulting RNA was conducted using the MEGAclear kit following the manufacturer’s protocol. The transcribed sgRNAs and crRNAs were eluted into 20 mM HEPES buffer. The concentration of RNAs was determined using a Nanodrop 2000 spectrophotometer, and the final guide RNA products were stored at –80 °C for subsequent experiments. Ai9 sgRNA target sequences were chosen as previously described45 and other target sequences were designed in-house.
Synthesis of CRISPR–Gold
GNPs (60 nm in diameter, 450 nM) were reacted with a 5′ thiol-modified single stranded oligonucleotide (DNA-SH), 26 bases in length (/5ThioMC6-D/GAAATATGCCAGAAATATCTGTCAGA, 200 μM), which had no sequence specificity. The reaction was performed in an Eppendorf tube in nuclease-free water (160 μl). A 100 mM sodium citrate solution (pH 3.5, 40 μl) was added to the reaction, and the reaction was allowed to proceed overnight46. Unconjugated DNA-SH was removed by centrifugation at 3,000g for 10 min, and was washed two times with 20 mM HEPES buffer. The GNP–DNA solution was stored at 4 °C until further use. CRISPR–Gold was synthesized using a layer-by-layer method. Cas9 or Cpf1 (50 pmole in 10 μl) and guide RNAs (50 pmole guide RNA in 10 μl) were mixed in 80 μl of Cas9 or Cpf1 buffer (50 mM HEPES (pH 7.5), 300 mM NaCl and 10% (vol./vol.) glycerol) for 5 min at room temperature, and this solution was then added to the GNP–DNA solution (0.45 pmole of GNPs), generating GNP–Cas9 or Cpf1 RNP. Freshly diluted sodium silicate (6 mM, 2 μl) was added to the GNP–Cas9 RNP solution and incubated for 5 min at room temperature. The mixture was then centrifuged using an EMD Millipore Amicon Ultra-4 30 kDa at 2,000g for 5 min to remove unbound molecules. The recovered GNP–Cas9 RNP–silicate was mixed with 5 μg of PAsp(DET) solution and incubated for 5 min at room temperature to form the last layer of CRISPR–Gold immediately before treatment.
Gel electrophoresis to measure Cas9 and Cpf1 RNP loading in CRISPR–Gold
Strong non-covalent interactions allow the interaction of GNP–DNA with Cas9 and Cpf1 RNPs. Gel analysis was conducted to visualize Cas9 and Cpf1 RNP loading on CRISPR–Gold. CRISPR–Gold was synthesized using the above-described method with Cas9, sgRNA_Ai9_L, sgRNA_Ai9_R, and Cpf1 and crRNA_yfp. The synthesized CRISPR–Gold was purified using a Vivaspin 300 kDa concentrator at 2,000g for 5 min to remove unbound molecules, and then one step of washing was conducted. Each sample collected before and after the purification step was analysed by gel electrophoresis using a 4–20% Mini-PROTEAN TGX Gel, stained with SYBR Green (ThermoFisher). Additionally, the same gel was stained with Coomassie Blue to visualize the Cas9 and Cpf1 proteins. Images were taken with ChemiDoc MP using the ImageLab software (Bio-Rad).
Animal care and use
Ai9 (in C57BL/6J background), Thy1-YFP (in C57BL/6J background), mdx, wild-type (in FVB background: FVB.129P2-Pde6b+ Tyrc-ch/AntJ) and Fmr1 knockout (in FVB background: FVB.129P2-Pde6b+ Tyrc-ch Fmr1tm1Cgr/J) mice were obtained from The Jackson Laboratory. The use and care of animals in this study followed the guidelines of the UTHSCSA and UC Berkeley Institutional Animal Care and Use Committee.
Primary culture of hippocampal neurons from wild-type FVB mice
Hippocampal neurons isolated from embryonic day 17 mouse brains (wild-type FVB mice) were plated at a density of 1–3 × 105 cells per well as described previously47. Cells were kept at 37 °C in a humidified, CO2-controlled (5%) incubator. Primary cultured hippocampal neurons were cultured for 7 days and were treated with either neurobasal medium only or CRISPR–Gold complexes including RNPs (25 pmole of Cas9 or 25 pmole of sgRNAs) with 2.5 µg of PAsp(DET) added in the neurobasal medium to test for electrophysiological properties and toxicity. Fourteen days after CRISPR–Gold treatment, the cells were stained with SYTOX Red and fixed with 4% paraformaldehyde, followed by phalloidin-Alexa488 staining with DAPI.
After 10–14 days of CRISPR–Gold treatment, primary cultured hippocampal neurons were patched and recorded. The extracellular solution contained NaCl (124 mM), KCl (2 mM), MgSO4 (2 mM), NaH2PO4 (1.25 mM), CaCl2 (2 mM), NaHCO3 (26 mM), D-dextrose (10 mM) and vitamin C (0.4 mM). The recordings were conducted at room temperature with an internal solution of potassium gluconate (120 mM), KCl (20 mM), MgCl2 (2 mM), HEPES (10 mM), ATP (2 mM), GTP (0.25 mM) and EGTA (0.1 mM, pH 7.4). Pyramidal cells were identified at ×60 magnification using a water-submersible objective and differential interference contrast/infrared optics on a BX51WI Olympus microscope. Recordings were conducted at 25 °C. Voltage recordings were performed using bridge balance compensation via the amplifier (HEKA, EPC10). If slow capacitance changed by more than 20% during recordings, the cell was excluded from further analysis. A hardware filter of 3 kHz was used for data collection. Input resistance was measured using a −20 pA current injection from the resting voltage. Action potentials were evoked from a holding current to maintain −80 mV, and spikes were generated by rectangular current injection (1 s duration, 200 pA). Action potential frequency was measured 3 min after break-in into the whole cell.
Culture of YFP-expressing HEK cells
YFP-HEK cells were generated by infection of HEK293T cells (from the Cell Culture Facility at UC Berkeley, authenticated using short tandem repeat (STR) profiling) with a YFP-containing lentivirus, and clonal selection was performed for cells expressing YFP. YFP-HEK cells were cultured in the culture medium (DMEM with 10% FBS, 1× MEM, non-essential amino acids and 100 μg ml–1 penicillin–streptomycin). The cells were tested for mycoplasma contamination and the result was negative. Nucleofection was conducted on the YFP-HEK cells.
Primary culture of fibroblasts from Ai9 mice
Primary fibroblasts were obtained from the muscles of Ai9 mice. Collagenase-treated tissues were minced with scaffold and digested in a collagenase and trypsin mixture48. The cells were plated in 10-cm culture dishes with the culture medium. Cells that were not firmly attached were removed during media changes that were conducted every 24 h. Fibroblasts were passaged with Accutase, and transfection with Cas9 or Cpf1 was conducted with fibroblasts within 14 days of culture.
Cells were detached using Accutase, spun down at 600 g for 3 min and washed with PBS. Nucleofection was conducted using an Amaxa 96-well Shuttle system following the manufacturer’s protocol using 10 μl of Cas9 RNPs (100 pmole of Cas9, 120 pmole of guide RNAs) or Cpf1 RNPs (100 pmole of Cpf1, 120 pmole of crRNAs)49. Cells (105) were transfected using the EH-100 Lonza program. After nucleofection, medium (500 μl) was added and the cells were incubated at 37 °C in tissue culture plates. The cell culture medium was changed the next day, and the cells were then incubated for 7 days before flow cytometry analysis.
Flow cytometry analysis fluorescence microscopy
Flow cytometry was used to quantify the expression levels of YFP in YFP-HEK cells or the expression levels of tdTomato in primary Ai9 fibroblasts after transfecting with Cas9 or Cpf1. The cells were analysed 7 days after transfection. The cells were washed with PBS and detached using Accutase. YFP and tdTomato expression was quantified using BD LSR Fortessa X-20 and Guava easyCyte, and analysis was conducted with FlowJo.
In vitro and in-cell cleavage assays
The mGluR5 template was PCR amplified (mGluR5 forward: CCTTAATGCACCACTCAGCA, mGluR5 reverse: GGCTTCCACTCTCTGAATGC) from mouse genomic DNA. For in vitro cleavage assays, the template DNA was incubated with Grm5 sgRNA and Cas9 proteins (Grm5 Cas9 RNPs) in a 1.5-ml tube. Gel electrophoresis was performed to verify cleavage of the template. For cleavage assays in the myoblasts from mdx mice, Grm5 Cas9 RNPs were introduced into the cell by electroporation. The mGluR5 gene was then PCR amplified from the myoblast, and a surveyor assay was conducted to check targeted gene editing.
Stereotaxic injection of CRISPR–Gold into the mouse brain
Mice (1–2 months old) were anaesthetized via an intraperitoneal injection of 100 mg kg–1 ketamine and 10 mg kg–1 xylazine. Preemptive analgaesia was given (Buprenex, 1 mg kg–1, intraperitoneally). Craniotomy was performed according to approved procedures, and 2 μl of CRISPR–Gold (50 pmole of Cas9 and 50 pmoles of guide RNAs, or 50 pmole of Cpf1 and 50 pmole of crRNAs) with 5 µg of PAsp(DET) for Thy1-YFP or Ai9 mice was injected into a single hemisphere of the striatum and/or the dorsal dentate gyrus of the hippocampus. The uninjected contralateral side was used as a control. For stereotaxic injection with mGluR5–CRISPR, 2 μl of saline (control) or CRISPR–Gold loaded with Cas9–mGluR5 RNPs (50 pmole of Cas9 and 50 pmole of sgRNAs) was injected into the striatum of both hemispheres of wild-type or Fmr1 knockout mice. Injection was given separately into three spots in each hemisphere with a 0.4 mm interval. The incision was clipped and proper post-operative analgaesics were administered for 6 days following surgery.
Deep sequencing analysis of CRISPR–Gold-treated brain tissue
The target sequences of the genomic region were amplified by PCR using Phusion High-Fidelity Polymerase according to the manufacturer’s protocol. Target genes were amplified first with primer sets and then amplified again with the deep sequencing primers listed in Supplementary Table 3. Barcode primers were designed as previously described50. The amplicons were purified using the ChargeSwitch PCR clean-up kit (ThermoFisher). PCR with barcode primers was conducted to attach Illumina adapters for deep sequencing. PCR clean-up was performed an additional time. The Berkeley Sequencing facility performed DNA quantification using a Qubit 2.0 Fluorometer (Life Technologies). BioAnalyzer for size analysis and qPCR quantification followed. The library was sequenced using the Illumina HiSeq2500 in the Vincent Coates Genomic Sequencing Laboratory at UC Berkeley. The analysis was conducted using the CRISPR Genome Analyzer51.
The mGluR5 target gene was amplified by PCR using Phusion Polymerase. The PCR amplicon was sent to Quintara Bioscience for sequencing. The sequencing result was analysed with TIDE software to quantify indel mutation efficiency (https://tide-calculator.nki.nl/)52.
Off-target prediction was conducted using Cas9-OFFinder (http://www.rgenome.net/cas-offinder/). Only the top two off-target sites had more than ten base matches. Genomic DNA extracted from mouse brains that were injected with mGluR5–CRISPR (CRISPR–Gold complex) was used to amplify two off-target sites. A surveyor assay and polyacrylamide gel electrophoresis were conducted to detect cleaved products.
RNA extraction from mouse brains and RT–qPCR
Mice were perfused with ice-cold PBS at 11 weeks after injection. Brains were cut into 1-mm sections using a brain slicer matrix (Zivic Instrument) around the injection sites. The brain slices were washed with ice-cold PBS and the injection region (1-mm thick, ~2-mm wide, ~2-mm long) was cut out. After adding TRIzol (800 μl), the brain slices were homogenized, treated with chloroform (160 μl) and centrifuged for 15 min at 4 °C. The aqueous phase of the sample was removed by pipet and 100% isopropanol (400 μl) was added. After being centrifuged for 10 min, the supernatant was removed from the tube, and the pellet was washed with 75% ethanol and centrifuged for 5 min. Afterwards, the supernatant was removed and the pellet was dissolved in DNase- and RNase-free water. RNA (1 μg) was reverse-transcribed using the SuperScript III First-Strand Synthesis kit. RT–qPCR analysis was performed using PowerUp SYBR Green Master Mix (Applied Biosystems) with the primers listed in Supplementary Table 4. The relative expression from RNA samples was analysed using the 2-ΔΔCT method. Values were normalized against the gene expression of the housekeeping gene PPIA.
Two weeks after stereotaxic injection of 1–2-month-old adult mice, the mice were anaesthetized by isoflurane and the left ventricle perfused with ice-cold PBS followed by 4% paraformaldehyde in PBS. The brains were post-fixed for 4 h in 4% paraformaldehyde, washed once with PBS and then transferred to 30% sucrose in PBS at 4 °C. Hippocampal sections from the mice were obtained by cryostat (CM3050S; Leica Microsystems). Before sectioning, brains were cryoprotected by incubating in 30% sucrose, embedded in O.C.T. compound, frozen and then stored at −80 °C until slicing. Slices were cut on the coronal plane at 20 μm, mounted on glass slides and stored at 4 °C. The sections to be immunostained were washed three times in PBS. Antigen retrieval was performed by steaming in a citrate buffer (0.294% sodium citrate, 0.05% Tween 20 in distilled water, pH 6.0) for 15 min with subsequent cooling over ice for 10 min. The sections were rinsed in PBS and blocked (5% goat serum, 0.2% Triton X-100 in PBS) overnight at 4 °C. The sections were incubated in the same blocking solution with primary antibodies at room temperature for 1 h. The sections were washed in PBS before incubation with secondary antibodies for 2 h. After rinsing once more in PBS, the sections were mounted in Prolong Gold Antifade Reagent with DAPI and imaged using a Zeiss confocal microscope.
Analysis and statistics for immunostaining
Quantification of fluorescence was performed on images taken with the same exposure times and within the same experiment. To determine relative YFP intensities in Thy1-YFP mice, a defined region of interest (ROI), which was the same size for all images analysed, was traced in the hippocampus using Image J software (NIH). To quantify YFP gene knockdown, YFP+ cells were also counted and normalized to DAPI+ cells in a defined ROI. To determine the percentage of edited cells or to present the number of DAPI+ cells in Ai9 mice, tdTomato+ cells were counted and normalized against the number of DAPI+ cells or the number of DAPI+ cells were presented itself (analysed from a defined ROI, which was the same size for all images analysed for comparison). To determine the percentage of cell types among edited cells, GFAP+, IBA1+ or NeuN+ cells were counted in only tdTomato+ cells. The percentage of the tdTomato+ cells among the cell types was also analysed by counting GFAP+, IBA1+ or NeuN+ cells co-stained with tdTomato among the total GFAP+, IBA1+ or NeuN+ cells. Each cell marker was stained with tdTomato and analysed independently. To compare the uninjected group and the injected group, Student’s unpaired t-test (two-tailed) was used. For mGluR5 immunostaining analysis, the number of mGluR5 immunofluorescent+ cells was quantified by using a minimum intensity threshold of 110 (range 0 to 255) using Image J software. The mGluR5+ cells were normalized to all DAPI+ cells in the field. Images were quantified from each brain hemisphere (two images per brain), corresponding to the control or CRISPR–Gold-injected hemispheres. The size of the brain region quantified was 212.55 µm2 around the injection site, in either the control-injected striatum or the CRISPR–Gold-injected striatum (~200 cells were counted). One-way analysis of variance (ANOVA) was used for the statistical analysis. All the statistics showed that variances are similar between the groups that are being statistically compared. No statistical methods were used to pre-determine sample sizes, but all sample sizes are similar to those generally employed in the field. Sample size (n) is indicated in each figure legend. The injection side was randomized (left or right side of the brain) for each experiment.
Behavioural test sequence and statistics
Behavioural tests were performed on mice at 2 weeks post-injection in the following order: empty cage observations, marble-burying assay, open-field activity assay, and rotarod performance test, followed by weighing and perfusion of all animals (Supplementary Table 5). A total of 47 mice (n = 11–12 per group) were used as a cohort. Identification of each animal was determined after testing to ensure that the experimenter remained blinded to the genotype or treatment of the test subject. Unless otherwise noted, data were analysed by one-way ANOVA using GraphPad Prism7 software.
Empty cage observations
The test was performed as previously described53 with minor modifications. The mice were brought into a testing room with normal lighting 1.5 h before testing, after which they were individually placed in a test cage identical to their home cage (30-cm wide, 19-cm long, 13-cm high) without bedding to prevent digging for 10 min of habituation. Videotaping of the cage was recorded using a DV Cam camera for 12 min while the mouse was allowed to freely explore. The last 10 min of the video was analysed by investigators who were blinded to the genotype or treatment of the test subject for line-crossing (crossing an imaginary line at the centre of the cage) and jumping behaviours.
The test was performed as previously described36,54 with minor modifications to evaluate repetitive digging behaviour. The mice were brought into the testing room with normal lighting 1.5 h before testing, after which they were individually placed in a cage identical to the test cage (30-cm wide, 19-cm long, 13-cm high) for 30 min of habituation. The test cage was filled with 3 cm of Teklad Sani-Chip bedding, and 20 dark blue marbles (15 mm diameter) were placed in a 5 × 4 pattern equidistant to each other and the side of the cage. The mice were then placed into the test cage and allowed to freely explore for 30 min. A marble was considered buried if two-thirds of the marble’s surface was covered by bedding. The percentage of marbles buried was then scored by four different investigators who were blinded to the genotype or treatment of the test subject.
Open-field activity assay
The open-field activity assay was performed as previously described36 with minor modifications. The mice were brought into the dark testing room with dim red lighting 1 h before the test, after which they were individually placed in a cage identical to their home cage for 30 min. The mice were then placed in the right corner of a clear acrylic chamber and allowed to freely explore for 30 min. Specific parameters, such as total distance travelled for 30 min, were measured to determine motor activity.
Rotarod performance test
The rotarod apparatus was used to measure locomotor activity36. During the training period, mice were allowed to explore the cylinder of the rotarod for 2 min with constant rotation at a speed of 4 r.p.m. After 5 min of rest, they were put on the rotarod and the speed was accelerated to 4–40 r.p.m. over a period of 300 s. The latency to fall off the rotarod within this time period was recorded (up to 300 s). Mice underwent four trials, and the mean latency to fall off the rotarod for all four trials was combined and calculated.
Further information on experimental design is available in the Nature Research Reporting Summary linked to this article.
The authors declare that all data supporting the findings of this study are available within the paper and its Supplementary Information.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank J. Doudna for advice, B. Staahl for discussions and technical support, and H. Kim, A. Rao and K. Kataoka for technical support. We thank M. A. Bhat and members of the Bhat Lab for technical support. We thank M. West in the CIRM/QB3 Shared Stem Cell facility for technical support. This work was supported by the National Institutes of Health grant R01EB023776 to N.M, and by the National Science Foundation grant 1456862 to R.B.