To achieve high-fidelity FP labeling, we developed the CRISPIE method, which inserts a designer exon module that consists of an exogenous exon encoding the FP flanked by intronic sequences, including the splicing acceptor and donor sites, into the intronic region of the target gene via the NHEJ pathway (Figure 1A, Figure 1—figure supplement 1). NHEJ exhibits a higher editing efficiency than HDR in postmitotic somatic cells, such as neurons (Heidenreich and Zhang, 2016; Heyer et al., 2010; Mao et al., 2008; Saleh-Gohari and Helleday, 2004). Because introns are spliced out of mRNAs, they are often poorly conserved and tolerate stochastic mutations better than exons (e.g., see Figure 1—figure supplement 2). INDELs at the joint junction of the insertion therefore will not result in mutations in the mRNA and thus the encoded protein under most circumstances (Figure 1A). In addition, NHEJ-mediated donor DNA insertion can result in unwanted integration with inverted orientation, and INDELs can occur at loci where the donor fails to insert. However, neither of these events will produce disrupted mRNAs when using CRISPIE (Figure 1A). Conceptually, unlike conventional exonic targeting, CRISPIE only results in either wild-type or successfully labeled mRNAs. For convenience, after a gene is labeled by CRISPIE, we say it has been ‘CRISPIEd’.

As a proof of principle, we first labeled human β-actin (ACTB) at its N-terminus with monomeric EGFP (mEGFP) in human U2OS cells by co-transfecting a plasmid that expresses a single guide RNA (sgRNA) targeting the ACTB gene and Streptococcus pyogenes Cas9 (SpCas9), a plasmid carrying the donor module (Figure 1B and Figure 1—figure supplement 3B1), and a transfection marker plasmid that expresses mRuby3. The sgRNA/SpCas9 complex cuts an editing site at the first intron of ACTB (Figure 1C), as selected using the BROAD Institute sgRNA designer (Doench et al., 2016), and releases the donor module from the plasmid by cutting sgRNA-targeted sequences flanking the module (Figure 1B, lower panel). The released module is then inserted into the cut site at the ACTB intron via the NHEJ pathway. The module-flanking sgRNA-targeted sequences are in the reverse orientation to facilitate the insertion of the module in the forward orientation (Suzuki et al., 2016). The donor module includes an exon in the appropriate translational phase (0–0 for intron 1 of ACTB) encoding mEGFP and peptide linkers (Figure 1—figure supplement 3B1). At the exon-intron junctions, ~100 bp of intronic sequence and ~10 bp of adjacent exonic sequence were taken from an obligatory intron (i.e., no reported alternative splicing events) and its adjacent exons of mouse CaMKIIα (Camk2a), a highly expressed protein in the brain, to include the splicing donor or acceptor site (Wang and Burge, 2008). mEGFP labeling was observed at 3–4 days post-transfection (Figure 1D) and required the presence of both sgRNA/SpCas9 and the donor module. Over 35% (37 ± 5%) of transfected cells, as identified by the expression of mRuby3 (red), were mEGFP (green)-labeled (Figure 1E). This likely represents a lower bound of labeling efficiency, since the co-transfection rate of all three plasmids is unlikely to reach 100%. Confocal and super-resolution structured illumination microscopy (SIM) revealed that the mEGFP signals exhibited the characteristic distribution of actin protein (Fischer et al., 1998; Planchon et al., 2011; Figure 1F). Time-lapse confocal microscopy revealed dynamic actin ruffles at the edges of cells (Figure 1G), which is consistent with the current knowledge of actin dynamics (Fischer et al., 1998; Planchon et al., 2011). Furthermore, cells with CRISPIEd actin grew at a rate indistinguishable from that of unlabeled cells in the same dish (Figure 1H and Figure 1—figure supplement 4). These results are consistent with the predictions for a functional, FP-labeled endogenous actin protein.

To characterize the insertion events and INDEL rates at the insertion site, we sorted transfected cells into red-only (transfected, but not labeled) and green (transfected and labeled) cells. The actin genomic DNA and mRNA at the editing locus were analyzed using PCR and RT-PCR, respectively (Figure 2A and B; see Figure 2—figure supplement 1 and Supplementary files 1 and 2 for primers, and anticipated PCR products and sizes). As expected, bands corresponding to the successful insertion (i.e., insertion in the forward orientation) at the targeted site are observed in green-labeled cells for both the genomic DNA and the mRNA (Figure 2A). Systematic analyses of green-labeled cells detected all possible editing events at the genomic DNA level, including no label insertion, forward insertion, and inverted insertion (Figure 2B), indicating that insertion events were typically non-homologous (i.e., both chromosomes differentially edited). Had the coding sequence been targeted, an inverted label insertion would inevitably cause disruptive mutations. However, under our conditions, because the donor only contained the splicing signals for exon inclusion in the forward orientation, only wild-type and forwardly inserted mRNAs were detected (Figure 2B), demonstrating the advantage of CRISPIE.

To evaluate how INDELs may affect a targeted gene, the PCR amplicons corresponding to no label insertion events in untransfected, red-only, and green cells, as well as those corresponding to forward insertion events in green cells were subjected to next-generation sequencing (NGS) (Figure 2C and D; sequenced amplicons correspond to gel bands in Figure 2A and B, as indicated by Roman numerals). Remarkably, at the genomic DNA level in red-only cells, over 70% (834/1181 reads) of PCR amplicon sequences contained INDELs (Figure 2C, and second column of Figure 2D), confirming the predicted high rate of INDELs during gene editing. In contrast, at the mRNA level in these cells, 99.7% (2930/2936 reads) of the sequences did not contain INDELs (p<0.001, cf. genomic DNA) (Figure 2D, second column), which was no different from control, untransfected cells (99.7%, 5761/5771 reads) (Figure 2D, first column). Similarly, undesired INDELs were frequently detected in all sequenced amplicons from green cells at the genomic DNA level (Figure 2D). Similar high rates of INDELs were also observed at the genomic DNA level in other insertion locations of ACTB and in another gene (TUBA1B) across two different cell lines (Figure 2—figure supplement 2). However, few errors were detected at the mRNA level (INDELs 24–87% for DNA vs. 0.3–2% for mRNA; Figure 2D). These numbers correspond to correct-to-mutation ratios of 50- to 300-fold for mRNA, which are around two orders of magnitude higher than those for DNA. These results demonstrate that, although inverted insertion and INDEL events may occur at the DNA level—which would have caused mutations or frame shifts when using existing strategies to target coding sequences—CRISPIE is resistant to these disruptive events at the mRNA level.

For insertion strategies that target coding sequences, the options for targeting sites are limited, and not all potential targeting sites allow for high-efficiency editing (Doench et al., 2014). Furthermore, as discussed above, INDELs cause mutations or frame shifts, thereby reducing the likelihood of successful labeling. In contrast, CRISPIE is insensitive to INDELs, and the lengths of introns usually offer ample choices for potential targeting sites. CRISPIE may therefore have the potential to achieve higher labeling efficiency than strategies that target coding sequences. To test this, we first asked whether different intronic insertion sites may exhibit different labeling efficiencies. Five different editing sites (i1 to i5 in Figure 3A) were selected using the BROAD Institute sgRNA designer (with Azimuth v2.0 scores of 0.57–0.69) (Doench et al., 2016). By using a generic donor that was released from the plasmid by a designed sgRNA (called DRS-2; see Figure 1—figure supplement 3B7 for donor plasmid design) (Gao et al., 2019), we found that all five sites permitted the successful labeling of ACTB; however, the labeling efficiencies varied significantly (Figure 3B), with the highest labeling efficiency corresponding to around 20-fold of the lowest.

We also examined FP labeling that directly targets the coding sequence for comparison. There are only two possible SpCas9 editing sites around the start codon of ACTB (referred to as e1 and e2; Figure 3A). Despite using their optimal insertion donors (for both, see Figure 1—figure supplement 3B6), both exonic sites exhibited significantly lower labeling efficiencies compared to the optimized CRISPIE editing (greater than fivefold; Figure 3C). Between the e1 and e2 sites, the one with the higher Azimuth score was in fact less efficient (Azimuth scores = 0.55 and 0.62, respectively), indicating that the labeling efficiency is not yet entirely predictable. Nevertheless, CRISPIE allows for systematic optimization by screening for highly efficient editing sites and, when optimized, can achieve higher labeling efficiency than exonic editing.

The CRISPIE method should be broadly applicable for use with different FPs or with other functional domains, different protein targets, and different animal species. To test different FP labels for potential multiplex imaging studies, we successfully labeled ACTB in U2OS cells using cyan (mTurquoise2), green-yellow (mNeonGreen), and red (mRuby3) FPs (Figure 4A, donor plasmids illustrated in Figure 1—figure supplement 3B2-4). Notably, when both mTurquoise2 and mNeonGreen donors were included simultaneously, double-labeled cells could be found, which appeared to be capable of dividing (Figure 4—figure supplement 1), indicating that FP labeling does not affect cell viability in diploid CRISPIEd cells. To test the applicability of CRISPIE to different protein targets, we labeled several additional important subcellular structures in U2OS cells, namely microtubules, mitochondria, focal adhesion complexes, and the endoplasmic reticulum, by labeling tubulin alpha 1B (TUBA1B), translocase of outer mitochondrial membrane 20 (TOMM20), vinculin (VCL), and calreticulin (CALR), respectively (Figure 4B and Figure 1—figure supplement 3A and B; Lam et al., 2012; Rizzo et al., 2009; Roberts et al., 2017; Shroff et al., 2008). In a growth test, cells with CRISPIEd TUBA1B grew at a rate indistinguishable from that of unlabeled cells in the same dish (Figure 1—figure supplement 4), suggesting that such labeled TUBA1B is functional. Notably, the TOMM20 and VCL genes were labeled at their respective last introns (Figure 1—figure supplement 3A) because their encoded proteins are best tagged at their C-termini (Rizzo et al., 2009; Roberts et al., 2017; Shroff et al., 2008). This demonstrates that CRISPIE is also readily compatible with C-terminal labeling. For some genes, such as VCL, there are coding sequences of significant lengths present at the last exon, downstream of the possible CRISPIE labeling site. To prevent potential disruption of protein function caused by an insertion in the middle of the protein, the coding sequence of the last exon of VCL was included in the insertion donor sequence, and a stop codon was introduced at the end of the mEGFP coding sequence (Figure 1—figure supplement 3B). To test the applicability in different animal species, we further labeled β-actin (Actb) in mouse Neuro 2A cells. A mouse-specific sgRNA was used because introns are degenerative across species (e.g., see Figure 1—figure supplement 2). The labeling was successful and was dependent on the correct combination of the species and the corresponding sgRNA (Figure 4C). CRISPIE labeling was also successful in rat neurons (Figure 5A, see below). Overall, these results indicate that CRISPIE is versatile and readily applicable to many different FP labels, protein species, and animal species, and can be used to tag proteins at their N- and C-termini.

To determine whether CRISPIE can be used in postmitotic cells, such as neurons, and whether it is compatible with different transfection methods, we applied it to label proteins in neurons in cultured slices and in vivo using three common transfection methods. First, we used the biolistic method (i.e., ‘gene gun’) to deliver the sgRNA and donor for rat β-actin (rActb), and the cytosolic marker mRuby3, to cultured rat hippocampal slices. Notably, although the biolistic transfection method is widely used in studying the cellular functions of neurons, CRISPR-based labeling has not yet been demonstrated using this method, likely due in part to the relatively low labeling efficiencies. For this same reason, high-contrast double labeling—for example, the target protein together with a cell morphology marker—in brain tissue remains challenging. In our experiments, 6–45 neurons were transfected per brain slice, as determined by their red mRuby3 fluorescence. Among them, on average 15% (15 ± 4%, n = 6 slices) exhibited successful labeling of actin (Figure 5A). In some slices, as many as six to seven successfully labeled neurons of different neuronal types, including dentate gyrus neurons, CA1 and CA3 pyramidal neurons, and interneurons, could be detected (Figure 5—figure supplement 1). Glial cells could also be labeled (Figure 5—figure supplement 1B). The labeled neurons exhibited normal miniature excitatory postsynaptic currents (mEPSCs) in frequency and amplitude when compared to adjacent untransfected neurons (Figure 5—figure supplement 2). These results indicate that CRISPIE provides a viable option for protein labeling using gene gun in cultured slices. Similar results were obtained in cultured mouse hippocampal slices from an SpCas9 heterozygous genetic background, which constitutively expresses SpCas9 and thereby simplified the experimental design (Figure 5B). Other synaptic proteins, including synapsin (SYN1), tubulin (TUBA1B), and CaMKIIα, were also successfully labeled (Figure 5A and B). In many labeled cells, both the mRuby3 cytosolic marker and mEGFP were readily detectible, allowing us to compare the subcellular distribution of three proteins—β-actin, α-tubulin 1B, and CaMKIIα—in neuronal dendrites (Figure 5C). By quantifying their distribution between the dendrites and spines using a previously described spine enrichment index (SEI) (Zhong et al., 2009), we found that the three proteins exhibited distinct subcellular distributions: actin was highly enriched in dendritic spines, and CaMKIIα was moderately enriched, whereas tubulin was excluded from spines (Figure 5C; see Materials and methods for quantification details).

To determine whether CRISPIE can be used to label endogenous proteins in vivo, in utero electroporation (IUE) was used to introduce the sgRNA, the donor, and the cytosolic marker mRuby3 into the somatosensory cortex of SpCas9 mice to label β-actin or tubulin in separate experiments (Figure 5D). Among the transfected red cells, the labeling efficiencies were 43 ± 5% for actin and 30 ± 5% for tubulin (Figure 5E), both of which are significantly higher than previously reported for CRISPR-based endogenous labeling of any protein in vivo (Mikuni et al., 2016; Nishiyama et al., 2017; Suzuki et al., 2016; Uemura et al., 2016). This high level of labeling efficiency allowed us to routinely perform in vivo imaging in the brains of living mice (Figure 5F and Figure 5—figure supplement 3). Finally, as viruses have become an increasingly prevalent method of transfection, we generated an adeno-associated virus (AAV) harboring the sgRNA targeting β-actin, the donor, and mRuby3 (Figure 1—figure supplement 3C). When injected into the cortex of SpCas9 mice, this virus successfully transduced neurons in vivo and labeled the actin cytoskeleton (Figure 5G). Overall, CRISPIE can be readily applied using a variety of widely used methods to transfect neurons both in vitro and in vivo with high efficiency.

Reversible gene editing may offer significant benefits in research, and in gene therapy in the future. However, existing CRISPR-based methods for endogenous labeling and large-fragment DNA insertions are not readily reversible. CRISPIE allows for the inclusion of functional elements at the flanking intronic regions of the donor module without affecting the mRNA. Taking advantage of this, we included additional designer sgRNA targeting sequences (DRS-1; Gao et al., 2019) at both ends of the module (Figure 6A). These added sites allowed for the later excision of the inserted module guided by the DRS-1 sgRNA. To demonstrate this reversibility, we isolated a U2OS cell clone of CRISPIEd ACTB. All cells derived from this clone uniformly exhibited green-labeled actin structures (Figure 6B, left panel). When DRS-1 sgRNA and SpCas9 were co-transfected with the transfection marker mRuby3, the green label was absent or greatly attenuated in over three quarters (76 ± 6%) of the mRuby3-positive cells at 5 days after transfection (Figure 6B and C). As a control, transfection with an sgRNA targeting mouse β-actin did not result in detectible change in the green labeling, indicating that CRISPIE is a readily reversible gene editing approach at the DNA level.

Plasmid constructs were generated using gene synthesis or standard subcloning methods. All previously unpublished constructs and their corresponding sequences will be deposited to Addgene .

For most experiments, sgRNAs were constructed by cloning the targeted sequences into the pX330 vector (Addgene #42230) via the BbsI sites. sgRNAs were designed using the GPP sgRNA designer from the BROAD Institute (https://portals.broadinstitute.org/gpp/public/analysis-tools/sgrna-design). The expression of each sgRNA was driven by a U6 promotor. The same plasmid also expressed a human codon-optimized SpCas9 protein. The insertion donors were constructed based on the pUC57-Amp vector (Genewiz). Depending on the specific downstream experiments, the donors may be flanked by the same gene-targeting cutting site and/or by DRS-1 or DRS-2 sites (see Gao et al., 2019 for the latter two; and see Figure 1—figure supplement 3 for specific donor designs), which allows the donor to be excised by SpCas9 guided by the gene-targeting sgRNA or by an additional DRS-1 or DRS-2 sgRNA. For some templates, a U6 promoter-driven DRS-2 sgRNA was integrated into the same plasmid of the donor to ensure co-expression with the donor and efficient excision during transfection. mRuby3 (Bajar et al., 2016) was expressed from a pKanCMV or a pCAGGS vector.

Human U2OS cells were ordered from ATCC (#HTB-96) and grown in McCoy’s 5A medium (ATCC #30–2007) supplemented with 10% fetal bovine serum. Mouse Neuro 2A cells were originally from ATCC (#CCL-131) and were obtained as a gift from Dr Kevin Wright’s laboratory. Note that all cells from ATCC have been authenticated by morphology, karyotyping, and PCR-based approaches. These include an assay to detect species-specific variants of the cytochrome C oxidase I gene (COI analysis) to rule out inter-species contamination and short tandem repeat profiling to distinguish between individual human cell lines and rule out intra-species contamination. These cells are also tested for mycoplasma by ATCC. Cell aliquots were kept frozen in liquid nitrogen until used and were further authenticated based on their morphology. Once thawed, each aliquot of cells was passed and used for no more than 3 months. Both cells were grown in a tissue culture incubator at 37°C and 5% CO₂. Cell culture transfections were carried out using Lipofectamine 2000 (Thermo Fisher) following the standard protocol, except that only half of the recommended Lipofectamine amount (~5 µl/35 mm dish) was used with a lipofectamine to DNA ratio of 2–3:1. For confocal and SIM imaging, the cells were re-plated onto a cover-glass bottom dish (FluoroDish, WPI, #FD35-100) at 2–4 days post-transfection at an appropriate density and imaged after another 2–3 days. For Figure 4C, Neuro 2A cells were serum-starved (in 0.1% FBS) and treated with all-trans retinoic acid (12.5 µM) at 3 days post-transfection to induce cell differentiation and then imaged after an additional 2–3 days.

Transfected U2OS cells were subjected to fluorescence-activated cell sorting (FACS) at the OHSU flow cytometry core using an Influx sorter (BD). Red label-only (i.e., transfected but not successfully labeled) and green-labeled cells were collected separately. For cell cloning, one to eight green-labeled cells were sorted into individual wells of a 96-well plate containing U2OS culture medium. Cell growth was visually inspected daily. Wells with only one to three dividing cells were monitored and later expanded into larger wells. A single-cell clone was used for Figure 6.

FACS cells (30–100 k per reaction) were used. For genomic DNA preparation, cells were lysed in 0.05M NaOH and 0.1 mM EDTA at 95°C for 15 min and then neutralized using 0.1 M (final concentration) Tris-HCl. RNA was isolated following an RNeasy Micro (Qiagen) protocol utilizing a QIAcube isolation robot at the OHSU Gene Profiling Shared Resource Core. mRNA was then reverse-transcribed using the GoScript reverse transcriptase enzyme (Promega) and random primers. Nested PCR was then carried out using OneTag DNA polymerase with GC buffer and Q5 polymerase (NEB) for genomic DNA and cDNA, respectively. GC buffer was necessary for genomic DNA amplification because the relevant intron contains highly GC-rich sequences. See Figure 2—figure supplement 1 and Supplementary files 1 and 2 for primer location, sequences, and combinations. A 4 bp barcode was added to the end of second-round primers to distinguish between individual samples. The first round of PCR for different products of each sample was carried out in a single reaction by including all relevant primers and was amplified for 23 cycles with an annealing temperature of 60°C. Approximately 2 µl of 10-fold diluted PCR products (to minimize the carryover of first-round primers) was added to 50 µl reactions for second-round PCRs with nested, specific primer pairs for 23 cycles. The exact added amount of first-round PCR product was adjusted so that different samples gave comparable band intensities for the control PCR for TUBA1B. The PCR amplicons from different samples were mixed and submitted for NGS (Genewiz). The sequencing results were unmixed by the barcodes and by the presence of specific primers. INDELs were identified using custom code written in MATLAB.

Hippocampi were dissected from P6–P7 Sprague Dawley rat pups (both sexes) and sectioned to 400 µm slices in dissection medium containing (in mM) 1 CaCl₂, 5 MgCl₂, 10 D-glucose, 4 KCl, 26 NaHCO₃, and 248 sucrose, with addition of 0.00025% phenol red. The slices were then seeded to a cell culture insert (Millipore, #PICM0RG50) and cultured at 35°C with 5% CO₂ in medium containing 7.4 g/l MEM (Thermo Fisher #11700–077) with the following supplements (in mM unless labeled otherwise): 16.2 NaCl, 2.5 L-Glutamax, 0.58 CaCl₂, 2 MgSO₄, 12.9 D-glucose, 5.2 NaHCO₃, 30 HEPES, 0.075% ascorbic acid, 1 µg/ml insulin, and 20% heat-inactivated horse serum. Organotypic hippocampal slice cultures were transfected using the biolistic method. Plasmids were coated onto 1.6 μm gold beads, and the slices were transfected at 4–7 days in vitro using a Helios gene gun (Bio-Rad). Slices were examined at 1–2 weeks post-transfection.

Animal handling and experimental protocols were performed in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and were approved by the Institutional Animal Care and Use Committee (IACUC) of the Oregon Health and Science University (#IS00002792). SpCas9 homozygous mice were crossed with C57BL/6 mice (Charles River). The resulting heterozygous offspring of both sexes were transfected using either IUE (at E16) or by stereotaxic viral injection (p14–60). In both cases, the somatosensory cortex was targeted. For in vivo imaging in living mice, a glass window was installed on the skull of the mouse at 1–4 months of age, as described previously (Ma et al., 2018). In vivo imaging was performed immediately after the window was installed.

IUE was performed at E16 as described previously (Ma et al., 2018) by injecting plasmid DNA (1 µl/embryo, total DNA concentration ~3–4 µg/µl, mixed with a 0.2% final concentration of fast green for visualization) into the lateral ventricle of mouse embryos, which were then electroporated with five 100 ms pulses (38 V) using an electroporator (BEX #CUY21). Stereotaxic viral injections were performed as previously described (Hunnicutt et al., 2014) at ~p21 using a custom-modified David Kopf system. Typically, 20–50 nl of AAV was injected.

Mice (p18–50) with neurons transfected using IUE were cardiac perfused with ice-cold, gassed (with 95% O₂/5% CO₂) artificial cerebral spinal fluid (ACSF) containing (in mM) 127 NaCl, 25 NaHCO₃, 25 D-glucose, 2.5 KCl, 1.25 NaH₂PO₄, 2 CaCl₂, and 1 MgCl₂. The brain was then dissected and coronal slices were prepared using a vibratome (Leica VT1200S) in a choline cutting solution (gassed with 95% O₂/5% CO₂) containing (in mM) 110 choline chloride, 25 NaHCO₃, 25 D-glucose, 2.5 KCl, 7 MgCl₂, 0.5 CaCl₂, 1.25 NaH₂PO₄, 11.5 sodium ascorbate, and 3 sodium pyruvate. The slices were the incubated in gassed ACSF at 35°C for 30 min and subsequently kept at room temperature for up to 6 hr.

All imaging experiments were carried out on live samples. Cultured cells were imaged using a custom-built two-photon microscope (up-right) (Ma et al., 2018), a Zeiss LSM 880 Airyscan laser scanning confocal microscope, or an Elyra 7 lattice-based structured illumination microscope. Cultured and acute brain slices were imaged in a chamber perfused with gassed ACSF. In vivo imaging was performed using a custom-built two-photon microscope based on the Janelia MIMMS design through a cranial window over transfected neurons in the mouse cortex. For two-photon microscopy, samples were excited using a Maitai HP Ti:Sapphire laser (Newport) at 960 or 990 nm, and green and red fluorescence were unmixed using a dichroic (Chroma 565DCXR) and band-pass filters (Chroma ET500/40 for green and Semrock FF01-630/92 for red). For Airyscan confocal microscopy and SIM, setups were configured according to the manufacturer’s suggestions for Nyquist spatial sampling, and raw data was auto-processed. Throughout the manuscript, red fluorescence is presented as magenta. For the cell growth assay, transfected cells were re-plated to coverslip-bottomed six-well plates, and were maintained and imaged using a Zeiss Celldiscoverer imaging system.

Image analyses were performed using custom software written in MATLAB. Labeling efficiencies were calculated as successful CRISPIEd cells (green) divided by total transfected cells (red) in random field of views (FOVs) (2 p imaging; 330 × 330 or 165 × 165 µm² FOV) selected based on the red channel only. In some experiments, the numbers of total CRISPIEd (green) and transfected (red) cells were visually counted from nine FOVs of predetermined arrangement (a 3 × 3 grid separated by 600 µm) under a 40× objective. All conditions to be compared were transfected and imaged side-by-side, and data were normalized to the averaged cell counts per FOV of the best labeled transfections. For IUE experiments, only the centers of the transfected regions were imaged and quantified. For mTurquoise2 and mNeonGreen labeling, the images were taken under the same conditions as mEGFP but were pseudo-colored in the color of turquoise and neon green, respectively. For the cell growth assay, labeled cells were manually counted. To control for potential cell damage/death during passaging of the cells before imaging, only CRISPIE-labeled cells that divided at least once were included for analyses. Untransfected cells within a region of interest (ROI) (~1 × 0.8 mm; 13–65 cells at time zero) with the same FOV as labeled cells were also counted.

SEI measurements were calculated as described previously (Zhong et al., 2009). Briefly, averaged green and red intensities were measured from manually drawn ROIs on the spines and their adjacent parental dendritic shaft in 2 p images. SEI was then calculated as log₂[(F_green/F_red)_spine/(F_green/F_red)_shaft], where F is the average fluorescence intensities of individual ROIs. Positive values indicate relative spine enrichment, whereas negative values indicate relative spine exclusion.

Whole-cell patch-clamp was performed in CRISPIEd CA3 pyramidal neurons and adjacent untransfected neurons (within 50 µm) in organotypic cultured slices. Voltage-clamp recordings were performed using a MultiClamp 700B amplifier (Molecular Devices) controlled with custom software written in MATLAB. Electrophysiological signals were acquired/digitized at 20 kHz and filtered at 2 kHz. Slices were perfused at room temperature with ACSF containing (mM): NaCl (127), NaHCO₃ (25), NaH₂PO₄ (1.25), glucose (25), KCl (2.5), CaCl₂ (4), MgCl₂ (4), pH ~7.3, and 310 mOsmol/kg when gassed with carbogenic mixture (95% O₂, 5% CO₂). Recording pipettes (3–5 MΩ) were pulled from borosilicate glass (G150F-3; Warner Instruments) using a model P-1000 puller (Sutter Instruments). Series resistance was 10–25 MΩ. The internal solution contained (in mM): Cs-gluconate (126), CsCl (10), HEPES (10), Na-phosphocreatine (5), Na-GTP (0.5), Mg-ATP (4), TEA-Cl (10), EGTA (5), and QX-314 bromide (4) with an osmolarity of 295–300 mOsmol/kg and pH ~7.2 adjusted with KOH. The liquid junction potential (ca. −14 mV calculated using JCal from the Clampex software, Molecular Devices) was not corrected. mEPSCs were measured at −70 mV in the presence of 1 μM TTX, 10 μM CPP, 100 μM picrotoxin. Ten-minute traces were recorded and mEPSCs events were detected using the template matching feature of Clampfit (Molecular Devices).

Sample sizes, as indicated in figure legends, were determined based on the observed variability across measurements. All experiments have been replicated in multiple independent trials/animals (typically >3) with the exception that only one round of NGS was carried out. Cells and animals were randomly allocated to each experimental groups. All groups to be compared were carried out side-by-side. The researchers were not blinded to group allocation during data collection and analysis. However, the FOVs were selected either predetermined or in the absence of knowledge of the color channel under investigation.

Quantification and statistical tests were performed using MATLAB scripts. p-Values were obtained from one-way ANOVA tests, unless noted otherwise. In all figures, *p≤0.05 and is statistically significant after Bonferroni correction for multiple tests, **p≤0.01, and ***p≤0.001.