Optimized CRISPR/Cas9-mediated in vivo genome engineering applicable to monitoring dynamics of endogenous proteins in the mouse neural tissues

To analyze the expression, localization, and functional dynamics of target proteins in situ, especially in living cells, it is important to develop a convenient, versatile, and efficient method to precisely introduce exogenous genes into the genome, which is applicable for labeling and engineering of the endogenous proteins of interest. By combining the CRISPR/Cas9 genome editing technology with an electroporation technique, we succeeded in creating knock-in alleles, from which GFP (RFP)-tagged endogenous proteins are produced, in neurons and glial cells in vivo in the developing mouse retina and brain. Correct gene targeting was confirmed by single-cell genotyping and Western blot analysis. Several gene loci were successfully targeted with high efficiency. Moreover, we succeeded in engineering the mouse genome to express foreign genes from the endogenous gene loci using a self-cleaving 2A peptide. Our method could be used to monitor the physiological changes in localization of endogenous proteins and expression levels of both mRNA and protein at a single cell resolution. This work discloses a powerful and widely applicable approach for visualization and manipulation of endogenous proteins in neural tissues.

In order to analyze the expression, localization and functional dynamics of target proteins in situ, it is essential to selectively label target proteins of interest. Several methods are widely used to label target proteins. These methods include the use of antibodies specific to target proteins and the transfection of cDNA encoding "tagged" recombinant proteins into cells. The former cannot be used for live cells and tissues, except for the proteins that are localized on the cell surface. Moreover, antibodies specific to target proteins of interest may not always be available. It is not easy to faithfully reproduce the expression levels and patterns of target proteins by transfecting exogenous genes using the latter method. Moreover, it is possible that overexpression of exogenous genes causes undesirable side effects. For example, in the case of rhodopsin, a light-sensitive G-protein coupled receptor in rod photoreceptors, overexpression of exogenous rhodopsin in transgenic mice alters the structure and function of rod photoreceptors, and results in retinal degeneration [1][2][3][4] .
Inserting a cDNA encoding a tagged recombinant protein into a targeted gene locus via homologous recombination in mouse embryonic stem (ES) cells and subsequently generating transgenic mice is an ideal approach to labeling endogenous proteins 5 . This method could be utilized to overcome the issues stated above. However, gene knock-in experiments in mouse ES cells are usually costly, labor-intensive, and time-consuming. In addition, this technique was only applicable to a limited number of cell lines including mouse ES cells, in which homologous recombination is observed with the high frequency of 1/10 to 1/10 3 . Even in human ES and induced pluripotent stem (iPS) cells, gene knock-in is quite difficult because of low efficiency of spontaneous homologous recombination 6 . Typically, homologous recombination only occurs in cells at a frequency of ~1/10 6 .

Results
Tagging endogenous rhodopsin with GFP in rod photoreceptors in the mouse retina by in vivo gene targeting. To determine whether and to what extent CRISPR/Cas9-mediated gene knock-in could occur in vivo in the postnatal mouse retina, we first chose the rhodopsin locus as a target. Rhodopsin is abundantly and selectively expressed in rod photoreceptors in the retina, and within rod photoreceptors, it is predominantly localized in the outer segments 23 . The CRISPR guide RNA (gRNA) was designed to target the genome sequence just after the stop codon in the last exon of the rhodopsin gene. The homologous recombination targeting vector (Rhodopsin targeting vector (EGFP)) had ~1 kb homology arms at the 5′ and 3′ ends of the EGFP cDNA to produce a rhodopsin-EGFP fusion protein upon insertion into the rhodopsin locus (Fig. 1A, Supplementary  Fig. S1A). We designed the targeting vector not to contain the promoter activity so that it could not express EGFP by itself. Once the targeting vector is correctly inserted into the target locus with the aid of CRISPR/Cas9, a fusion protein between rhodopsin and EGFP is produced from the endogenous rhodopsin locus.
Postnatal day 0 (P0) mouse retinas were co-electroporated in vivo with Rhodopsin targeting vector (EGFP), CAG-mCherry, and the CRISPR construct (CBh-Cas9/U6-gRNA (Rho)) ( Fig. 1B). When the transfected retinas were harvested at P21, many bright EGFP-positive cells were observed in intact retinas only in the presence of the CRISPR construct (Fig. 1C). EGFP signals were exclusively detected in rod photoreceptors in the outer nuclear layer (ONL) in retinal sections and were localized predominantly in the rod outer segments and to a lesser extent in the rod inner segments, with little or no expression in rod cell bodies (Fig. 1D, Supplementary Fig. S1B). There were no mCherry-positive cells that expressed EGFP in the inner nuclear layer (INL) and the ganglion cell layer (GCL). This expression pattern of EGFP was consistent with that of endogenous rhodopsin, suggesting that endogenous rhodopsin was correctly tagged with EGFP in the postnatal mouse retina by in vivo gene targeting. The number of EGFP-positive cells among the transfected cells (mCherry-positive cells) was estimated to be ~10% on average.
Correct integration of the targeting vector into the rhodopsin locus was verified by single-cell genotyping and Western blot analysis. For single-cell genotyping, electroporated retinas were dissociated into single cells by enzymatic digestion and subjected to fluorescence-activated cell sorting (FACS) to collect EGFP/mCherry double-positive cells in 96 well plates (1 cell per well) (Fig. 1E). Polymerase chain reaction (PCR) analysis of the sorted cells revealed that most of the EGFP-positive cells underwent correct gene targeting (Fig. 1F). Western blot analysis of harvested intact retinas also showed that the rhodopsin locus was correctly modified to produce a fusion protein (~67 kDa) composed of rhodopsin (~39 kDa) and EGFP (~28 kDa) only in the presence of the CRISPR construct (Fig. 1G).
The knock-in efficiency at the rhodopsin locus was dependent on the length of homology arms of the targeting vector. The highest knock-in efficiency was observed when both 5′ and 3′ homology arms of the targeting vector are ~1 kb ( Supplementary Fig. S2).
Biallelic gene knock-in at the rhodopsin locus. To determine whether both alleles of rhodopsin can be modified by in vivo gene targeting, another targeting vector, which was designed to produce a rhodopsin-mCherry fusion protein from the endogenous rhodopsin locus, was constructed ( Fig. 2A). Rhodopsin-targeting vector (mCherry) exhibited comparable knock-in efficiency to that of Rhodopsin targeting vector (EGFP) (Fig. 2C). When these two targeting vectors were co-electroporated together with the CRISPR construct into P0 mouse retinas, about 10% of the rod photoreceptors that expressed the fluorescent reporters were double-positive for EGFP and mCherry, indicating that biallelic gene knock-in occurs at the rhodopsin locus (Fig. 2D). engineering the rhodopsin locus. Taking advantage of the high efficiency of CRISPR/Cas9-mediated in vivo gene targeting at the rhodopsin locus in the mouse retina, we tried to engineer the mouse genome to express foreign genes from the rhodopsin locus. For this purpose, a self-cleaving 2A peptide was used to express multiple proteins from a single locus ( Fig. 2A). When the targeting vector encoding 2A-EGFP was used with the CRISPR construct, EGFP expression was observed exclusively in rod photoreceptors, but the fluorescent signals  www.nature.com/scientificreports www.nature.com/scientificreports/ were barely seen in the outer segments and localized mainly in the inner segments, cell bodies, and synapse regions (Fig. 2E). Western blot analysis confirmed the production of a free EGFP protein (~28 kDa), but not a rhodopsin-EGFP fusion protein (~67 kDa), in the retina (Fig. 2F). On the other hand, when the targeting vector encoding 2A-H2B-EGFP, a fusion between histone H2B and EGFP ( Fig. 2A), was used, EGFP signals were detected only in rod cell nuclei (Fig. 2G). The data show that in vivo gene targeting is a powerful approach to stably express exogenous genes of interest from endogenous gene loci.
Cell type-specific gene knock-in in the retina. The frequency of in vivo gene targeting could be affected by various factors, such as the cell-cycle state of the transfected cells. Using the rhodopsin locus as a model, we tried to determine the optimal timing of CRISPR/Cas9-mediated genome cleavage for gene knock-in during mouse retinal development. To control the timing of the expression of Cas9 nuclease in the retina, the rhodopsin promoter, instead of the ubiquitous CBh promoter 24,25 , was used (Fig. 2B). The rhodopsin promoter is specific to postmitotic rod photoreceptors and is inactive in proliferating retinal progenitors 26,27 ( Supplementary  Fig. S3). Plasmid DNAs electroporated to P0 mouse retina are normally introduced into undifferentiated retinal progenitors, which give rise to neurons (rod photoreceptors, bipolar cells, and amacrine cells) and Müller glial cells after a few rounds of cell division 28 . Therefore, the time course of genome cleavage mediated by Rhodopsin promoter-Cas9 should be entirely different from that by CBh-Cas9. Unexpectedly, the efficiency of EGFP knock-in at the rhodopsin locus mediated by Rhodopsin promoter-Cas9 was almost comparable to that mediated by CBh-Cas9 (Fig. 2H). The data suggest that gene knock-in could occur in postmitotic neurons in the retina and that cell division is not necessarily required for homologous recombination. The data also indicates that "cell type-specific gene knock-in" is possible when cell type-specific promoters are used to express Cas9.

tagging endogenous glutamine synthetase in Müller glia in the mouse retina.
To determine if the in vivo gene targeting technique could be applied to other gene loci and other cell types in the retina, we chose the glutamine synthetase (Glul) locus as a target. Glul is expressed exclusively in Müller glia in the retina 29 . The CRISPR gRNA was designed to target the genome sequence around the stop codon of the Glul gene, and the targeting vector was designed to express a Glul-mClover fusion protein upon insertion into the Glul locus ( Fig. 3A, Supplementary Fig. S4). This targeting vector also does not contain the promoter region and thereby cannot express the reporter by itself. In the mouse retina that was co-electroporated with Glul targeting vector (mClover), CAG-H2B-mCherry, and the CRISPR construct (CBh-Cas9/U6-gRNA(Glul)), tens of mClover-positive cells were observed at P21 in a CRISPR construct-dependent manner (Fig. 3B). Fluorescent signals were detected even in one-year-old mouse retinas (at P365). Retinal sections showed that mClover expression was detected only in Müller glia (Fig. 3C). Single-cell genotyping and Western blot analysis confirmed the correct gene knock-in and the production of a Glul-mClover fusion protein in the retina (Fig. 3D, Supplementary Fig. S4). The number of mClover-positive cells among the transfected cells (mCherry-positive cells) was ~2%, as determined by FACS.
Considering that approximately 10% of the retinal cells transfected at P0 become Müller glia 28 , the knock-in efficiency was estimated to be ~20%.

Monitoring light-dependent translocation of endogenous arrestin in the mouse retina.
One of the advantages of in vivo gene targeting is that it can be used to label endogenous proteins of interest with fluorescent reporters to monitor their dynamic movement within cells. To demonstrate this, we chose arrestin as a target. Arrestin is exclusively expressed in photoreceptors in the retina, and its localization within photoreceptors dynamically changes in response to light illumination. The CRISPR gRNA was designed to target the genome sequence around the stop codon of the arrestin gene, and the targeting vector was designed to express an arrestin-EGFP C-terminal fusion protein from the endogenous locus (Fig. 4A, Supplementary Fig. S5). CRISPR/ Cas9-dependent gene knock-in at the arrestin locus occurred at a frequency of ~1% in the electroporated retinas, and EGFP signals were detected only in rod photoreceptors (Fig. 4B,C). Importantly, in the light-adapted retinas, EGFP signals were predominantly localized in the outer segments of rod photoreceptors, while in the dark-adapted retinas, EGFP signals were localized in the inner segments and cell bodies but not in the outer segments of rod photoreceptors (Fig. 4C). The light-dependent dynamic localization patterns of EGFP within rod photoreceptors were consistent well with those of endogenous arrestin 30 . tagging endogenous synaptic proteins in the mouse retina. Another advantage of in vivo gene targeting is that without adverse effects on cells, it can be used to label the endogenous proteins whose overexpression influences cellular functions. Synapses in live neurons are often visualized by transfecting cDNAs encoding fluorescent protein-tagged pre-and post-synaptic proteins, synaptophysin (also called Syp) and PSD-95, respectively. In many cases, however, exogenous expression of tagged Syp and PSD-95 results in "overexpression" and morphology and function of transfected neurons are significantly affected 31,32 . Therefore, to study their dynamic behavior under "physiological" conditions in neurons, it is important to label endogenous Syp and PSD-95 with fluorescent reporters.
To achieve fluorescent labeling of endogenous Syp and PSD-95 in the mouse retina, the CRISPR gRNAs were designed to target the exons containing the stop codons of the Syp and PSD-95 genes, and the targeting vectors were designed to express Syp-EGFP and PSD-95-EGFP fusion proteins from the endogenous loci (Fig. 5, Supplementary Figs S6, S7). As we found that the use of two distinct gRNAs whose target sequences are adjacent to each other significantly enhances the knock-in efficiency compared to that of a single gRNA ( Supplementary Fig. S8), two gRNAs were used together to increase the knock-in efficiency. We observed CRISPR/Cas9-dependent gene knock-in at both gene loci in the retina, with a frequency of ~6% and ~3%, respectively. When endogenous Syp, which is localized in the outer plexiform layer (OPL) and the inner plexiform layer (IPL) of the retina, was labeled with EGFP by in vivo gene targeting, fluorescent signals were predominantly (2019) 9:11309 | https://doi.org/10.1038/s41598-019-47721-4 www.nature.com/scientificreports www.nature.com/scientificreports/ detected in the synaptic terminals of rod photoreceptors, bipolar cells, and amacrine cells. Weaker fluorescent signals were also detected in the cell bodies of these retinal neurons (Fig. 5B). Occasionally, fluorescently labeled rod photoreceptor and bipolar cells connected to each other were observed. On the other hand, when endogenous PSD-95, which is predominantly localized in the OPL of the retina, was labeled with EGFP, most of the fluorescent signals were detected in the synaptic terminals of rod photoreceptors (Fig. 5D). The synaptic terminals of bipolar cells in the OPL were also labeled though at much lower frequencies.
Monitoring protein dynamics of PSD-95 in vivo. Finally, we examined if in vivo gene targeting can be used to label endogenous proteins of interest with fluorescent reporters to monitor their expression changes in specific subpopulations of neurons at a single cell resolution. It is well known that experience-dependent plasticity has been thought to involve increases in the expression of synaptic proteins 33 . But such studies were based on global expression analysis using tissue lysates. Taking advantage of our labeling strategy developed in this study, we tried to label endogenous PSD-95 in vivo in the developing mouse brain (Fig. 6, Supplementary Fig. S9).
To test whether the expression of endogenous PSD-95 tagged with mClover is increased by experience-dependent plasticity, electroporated mice were separated into two groups at P21. One group was kept in normal cages, and the other group was placed in the enriched cages with running wheels (Fig. 6E). In the mice exposed to the enriched environment, brighter mClover expression was observed at P65, and the density of mClover-positive puncta was also increased (Fig. 6F). Western blot and qPCR analyses showed that PSD-95-mClover protein and mRNA were both up-regulated several-fold by enrichment (Fig. 6G-I). These results show that our in vivo genome engineering technique enables quantitative analysis of not only endogenous proteins but also mRNAs in specific subpopulations of neurons at a single cell level.

Discussion
Our results showed that endogenous proteins, including rhodopsin, glutamine synthetase, arrestin, synaptophysin, and PSD-95 can be efficiently labeled with fluorescent proteins in vivo in the mouse retina and brain, by using the combined methods of CRISPR/Cas9 and in vivo electroporation. We also showed that using this technique, physiological changes in localization and expression levels of endogenous proteins can be monitored in situ. Moreover, we provided a strategy to engineer the mouse genome in vivo to express foreign genes from the endogenous gene loci using a self-cleaving 2 A peptide.
Compared with the existing protein-labeling methods, this method is superior for the following reasons. First, this labeling method is easy and rapid. In contrast to the conventional knock-in methods using mouse ES cells, this method significantly shortens the necessary time for tagging of endogenous proteins in vivo because of the direct modification of genomic DNA in the internal organ of living mice. Second, this labeling method is accurate. As endogenous proteins are labeled via modification of the genome coding for them, but not via direct chemical modification of their amino acid residues, the specificity of labeling is extremely high. Third, this method is more versatile. In this study, we used the mouse retina and brain as in vivo model systems, but the established labeling technique should be widely applicable to endogenous proteins in other organs. We also believe that this method can be used in other animal models besides mice. In addition, since this method does not require the fixation of cells, it is suitable for the observation of the dynamic behavior of labeled proteins in living cells and tissues. Fourth, this labeling method is ideal for imaging of neurons with complicated morphology at a single-cell resolution. As only a limited number of transfected cells undergo CRISPR/Cas9-mediated homologous recombination, this method facilitates "sparse cell labeling", which is an important approach to study the neural circuit.
During preparation of the manuscript, other groups reported that endogenous proteins in the developing mouse brain can be tagged with the reporters using the approach similar to ours 34-36 . Mikuni et al. reported that several endogenous proteins can be labeled with the HA tag in neurons in the developing mouse brain using ~200 base single-stranded oligodeoxynucleotides (ssODNs) as knock-in donors at an efficiency of 0.6-7.5%. They also reported that endogenous CamKIIα and CamKIIβ can be labeled with EGFP using donor plasmids carrying 0.7-0.9 kb homology arms, although the knock-in efficiency was less than 1% 34   www.nature.com/scientificreports www.nature.com/scientificreports/ that endogenous β-actin (Actb) can be tagged with EGFP in neurons in the developing mouse brain using a donor plasmid carrying 1.4 kb and 0.7 kb homology arms at an efficiency of ~1.7% 35 . Tunekawa et al. reported that endogenous βIII-tubulin (Tubb3) in neurons in the developing mouse brain can be tagged with EGFP using a donor plasmid carrying 1 kb and 1.8 kb homology arms at an efficiency of ~20% 36 .
In our study, we showed that endogenous proteins in the developing mouse retina and brain can be labeled with fluorescent reporters at an efficiency of 1-10%. We also found that endogenous proteins labeled with fluorescent reporters can be readily visualized in glial cells as well as in neurons in intact mouse retinas without making sections if the target proteins are relatively abundantly expressed. In addition, we further demonstrated that the dynamics of the tagged endogenous proteins can be easily monitored in situ. For example, we observed light-dependent translocation of endogenous arrestin tagged with EGFP within photoreceptors (Fig. 4). We also observed that the expression of endogenous PSD-95 tagged with mClover is increased in mice exposed to the enriched environment at both mRNA and protein levels (Fig. 6). Our current study, together with the other groups' recent reports, not only presents the versatility of the labeling method among different neural tissues including brain and retina, but also guarantees the accuracy with the "C-terminal fusion approach".
Although the labeling technique developed in this study is very powerful, there exist several issues that must be overcome for the dissemination and popularization of this technique. The first challenge is that this labeling method is not necessarily suitable for N-terminal tagging of target proteins with fluorescent reporters. In this study, we used the knock-in targeting vectors that correspond to the C-termini of target proteins and do not contain detectable promoter activities. However, when a targeting vector is designed against the genomic DNA encoding the N-terminus (ATG start codon) of protein, there is a high probability that the promoter region is included in the homology arm of the targeting vector 37 . In that case, the targeting vector would express a fluorescent reporter independently of CRISPR/Cas9-mediated homologous recombination, and it becomes hard to distinguish the true signals derived from fluorescent protein-tagged endogenous proteins. For this reason, careful attention is required when one wishes to fuse a fluorescent protein to the N-terminus of a target protein by genome editing. www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/ The "C-terminal fusion approach" is also beneficial to reducing the risk of unintended mutations in the regulatory regions, including enhancers and promoters, of target genes. Cas9 might induce unintended indel mutations mediated by NHEJ in transfected cells if homologous recombination does not occur. This problem matters particularly in the case of N-terminal tagging since altered sequence near the N-terminal region can change the transcription and translation of the target genes. On the other hand, in the case of C-terminal tagging, such a risk is relatively low.
Even with the "C-terminal fusion approach", however, unintended indel mutations mediated by NHEJ could affect the functions of target genes, especially when target PAM sequences are located within coding sequences. This may cause non-cell-autonomous effects to cells with knock-in alleles. Moreover, for some proteins, C-terminal fusion may abolish normal protein functions. For example, most of the ras superfamily small GTPases have the CAAX motif important for membrane anchoring at their C-termini, and therefore the "C-terminal fusion approach" cannot be used for these proteins.
The second challenge is that off-target effects of the Cas9 nuclease cannot be entirely excluded. When the CRISPR/Cas9-mediated protein tagging technology is applied to the developing or mature mouse tissues wherein cells do not proliferate infinitely, it is difficult to analyze precisely the genomes extracted from the cells that underwent homologous recombination. In this study, overexpression of CRISPR/Cas9 in the mouse retina for one year did not cause any apparent cytotoxicity, suggesting that even if there were some adverse effects on cell functions, it would be minimal. However, it is necessary to consider the experimental conditions carefully, depending on the target sequences. Recently, engineered Cas9 variants, including eSpCas9 38 , SpCas9-HF1 39 and HypaCas9 40 , with reduced off-target activity have been reported. Use of such Cas9 variants may partly overcome the off-target effects.
The third challenge is that the knock-in efficiency is relatively low. In this study, we found that the use of two adjacent gRNAs targeting one locus increased the knock-in efficiency (Fig. S8) A similar approach was used to increase the knock-in efficiency in mouse fertilized eggs and mouse ES cells [41][42][43] , although the molecular mechanism underlying the enhanced gene knock-in efficiency by two gRNAs remains unknown. Since simultaneous use of two gRNAs exhibited higher DNA cleavage activity compared to each of the single gRNAs 41,42 , two gRNAs may ensure cleavage of target genome, and thereby enhance the knock-in efficiency.
Recently, it was reported that CRISPR/Cas9-mediated DNA cleavage on both ends of homology arms of a donor plasmid, to which gRNA target sites had been introduced, significantly increased the knock-in efficiency 44 . A combination of two gRNAs and in vivo linearization of donor plamids may further enhance the knock-in efficiency.
The fourth challenge that must be circumvented is that fluorescent signal intensity may not be high enough for live cell imaging in cases where the expression levels of target proteins are low. With the development of more sensitive fluorescent signal detection systems and brighter fluorescent proteins, the genome editing-based protein tagging technology will become a more powerful tool for monitoring the expression, localization, and functional dynamics of endogenous proteins in situ.

experimental procedures
Animals. All animal experiments were approved by the Institutional Animal Care and Use Committees at Kyoto University and University of Hyogo. All procedures were performed in full compliance with the Fundamental Guidelines for Proper Conduct of Animal Experiments and Related Activities in Academic Research Institutions issued by the Japanese Ministry of Education, Culture, Sports, Science and Technology. Mice were housed under temperature-controlled conditions (temperature: 24 °C) and maintained on a light/dark cycle (12 h each) with ad libitum access to food and water. Timed pregnant ICR mice were purchased from Japan SLC (Shizuoka, Japan). cRiSpR constructs. The CRISPR target sequences (20-nucleotide sequence followed by a protospacer adjacent motif (PAM) of 'NGG') were selected manually, or by using the prediction software 45 (http://www.broadinstitute.org/rnai/public/analysis-tools/sgrna-design). To make plasmids co-expressing S. pyogenes Cas9 and gRNA, following synthetic oligonucleotides were annealed and cloned into the BbsI sites of pX330 14 (Addgene #42230). Rhodopsin; 5′-CACCGTAAGCCTGGCCAGAGACTG-3′ and 5′-AAACCAGTCTCTGGCCAGGCTTAC-3′.
the electroporated mice were kept in normal cages ((−) running-wheel) or in cages with running wheels ((+) running-wheel). Brains were harvested at P65, sectioned and stained with anti-GFP (green) and anti-RFP (red) antibodies. Cell nuclei were visualized with DAPI (blue). A white arrow indicates a mClover-positive punctate along a dendritic shaft. (G) Western blot analysis of brains shown in (F). Six independently electroporated brains were analyzed (1 brain per lane), and membranes were probed with indicated antibodies. Full-length gel images are presented in Supplementary Fig. S10B. The amount of PSD-95-mClover was determined by quantifying the intensity of Western blot bands, and relative expression levels of PSD-95-mClover were shown. Data represent mean +/− SD (n = 3; P = 0.024, Student's t test). (H,I) RT-PCR analysis of brains shown in (F). Three independently electroporated brains were analyzed (1 brain per lane). PCR primer sets were used to detect cDNAs encoding PSD-95-mClover, tdTomato, and endogenous mouse β-actin (Actb), respectively. Full-length gel images are presented in Supplementary Fig. S10C. Relative expression levels of PSD-95-mClover determined by qPCR and normalized to those of tdTomato were shown in (I).
Housing mice in an enriched environment. The expression of tdTomato in the brains of electroporated mice was checked after the birth using a handheld LED light (OPTOCORD Co. Tokyo, Japan, LED-EXHD/RFP). Only the pups strongly positive for tdTomato in the somatosensory cortex were used in this study. After weaning at P21, mice were separated into two groups, and five mice were maintained per cage (17 cm × 27 cm × 12 cm). One group was put in regular cages, and the other group was in the cages with a running wheel (FAST-Trac, Mouse Igloo, Bio-Serv). Genotyping. For single-cell genotyping, FACS-purified single cells in 15 μl of 50 μg/ml proteinase K were incubated for 1 h at 50 °C, followed by 4 min at 99 °C. Then, nested PCRs were performed using PrimeSTAR GLX DNA polymerase (Takara Bio, Shiga, Japan) or KOD FX neo DNA polymerase (Toyobo, Osaka, Japan). For genotyping of retinal and brain cells transfected with PSD-95 targeting vector (EGFP), PSD95 targeting vector (mClover), or Syp targeting vector (EGFP), genomic DNAs were purified from the harvested mouse retinas or brains using QIAamp DNA mini kit (Qiagen). Then, nested PCRs were performed to detect the knock-in alleles. For detail, see Supplemental Experimental Procedures.