Hide-and-Seek genome editing reveals that Gephyrin is required for axo-axonic synapse assembly

Significance

We report Hide-and-Seek, a rapid and efficient application of CRISPR-mediated genome editing in neurons in vitro and in vivo. Hide-and-Seek allows for the simultaneous knockout of one gene and knock-in of an epitope tag into a second gene of interest. Using Hide-and-Seek, we show that the scaffolding protein Gephyrin is required for the assembly of axo-axonic inhibitory synapses at the axon initial segment.

Abstract

The visualization and manipulation of proteins in neurons is widely used to deduce their functions. While every experimental approach has limitations, the concurrent knock-in and knockout of two different proteins can be especially challenging. To this end, we developed Hide-and-Seek genome editing, which allows the simultaneous visualization and knockout of proteins in neurons using Adeno-associated viral vectors and the CRISPR/Cas9 system. We demonstrate the efficacy and flexibility of this method for rapid, efficient, and simultaneous knock-in and knockout of proteins in vitro and in vivo, at the synapse, axon initial segment (AIS), nucleus, and mitochondria. Using Hide-and-Seek, we show that the scaffolding protein Gephyrin is required for the proper assembly of axo-axonic synapses at the AIS.

The ability to detect and manipulate neuronal protein expression in situ is essential to define protein functions. Most commonly, proteins are detected using antibodies, while function is tested using knockout mouse models. Antibody-based experiments are fast, flexible, and simple. However, they rely on previously validated and protein-specific primary antibodies which may not be available. The biomedical literature is replete with experiments using poorly validated antibodies leading to spurious conclusions, and much has been said about the importance of rigorous antibody validation (1). Alternatively, expression of tagged versions of neuronal proteins has also been used to study protein localization and function. But this only approximates the native state since it requires overexpression of the protein being studied. In addition, while the introduction of proteins into neurons can be performed by transient transfection in vitro, in vivo approaches including electroporation or viral delivery have many additional experimental challenges including the timing of electroporation and viral packaging limits. Recently, several CRISPR-mediated genome editing strategies (e.g., HiUGE, ORANGE, and HITI) have been developed for neurons that allow tagging of endogenous proteins, thereby bypassing the need for protein-specific antibodies or overexpression (2–4).

Whole-body and cell type–specific knockout mouse models are frequently used for in vivo “loss-of-function” studies to determine a protein’s necessity for neuronal and brain functions. However, knockout mouse models are expensive, time-consuming, and may be lethal. In vitro loss-of-function strategies include silencing of proteins using small interfering RNA (siRNA). However, siRNA can result in only partial knockdown and off-target effects. Thus, flexible, rapid, and reliable alternative methods permitting the simultaneous visualization and manipulation of proteins both in vitro and in vivo would facilitate efforts to define neuronal protein functions at both cellular and circuit scales.

For example, pyramidal neuron (PyN) excitability is regulated mainly by axo-dendritic and axo-somatic inputs from excitatory and inhibitory neurons. However, gamma-aminobutyric acid (GABA)-ergic axo-axonic synapses located at the PyN axon initial segment (AIS) also powerfully modulate the generation of axonal action potentials (5, 6). Despite their important role in neuronal excitability and circuit function, the molecular and circuit mechanisms underlying AIS axo-axonic synapse assembly and function remain incompletely understood (7). AIS axo-axonic synapses can be visualized by coimmunostaining pre- or postsynaptic proteins [e.g., vesicular GABA transporter (VGAT) or Gephyrin (Gphn)] together with AIS proteins (e.g., AnkyrinG or β4 spectrin). However, due to the high density of neurons in the cerebral cortex, it can be challenging to clearly distinguish non-AIS signals from bona fide AIS axo-axonic synapses, and whole-body knockout of components of inhibitory synapses (e.g., Gphn) can be lethal (8). Dysregulation of AIS axo-axonic synapses has been reported in patients with schizophrenia, autism spectrum disorder, and Alzheimer’s disease (9–11). Thus, defining the mechanisms of AIS axo-axonic synapse assembly may reveal how their dysregulation leads to neuropsychiatric and neurodegenerative diseases.

Here, we report “Hide-and-Seek” as an application of CRISPR-mediated genome editing that enables high-efficiency knockout with concurrent knock-in of targeted proteins using Adeno-associated viral (AAV) vectors. This system overcomes a major experimental bottleneck in molecular neurobiology by allowing simultaneous visualization and manipulation of specific proteins. We established and optimized this methodology and illustrate its use by showing that Gphn is required for axo-axonic synapse assembly.

Results

Optimization of CRISPR-Cas9-Mediated Genome Editing.

We previously reported an efficient AAV-mediated multiplexed (triple) single-guide RNA (sgRNA) expression system to knockout proteins in neurons (SI Appendix, Fig. S1A) (12). We designed this vector to be constructed in a single step of cloning. As proof of concept, we targeted NeuN for knockout. We generated AAV vectors and injected them into postnatal day 0 (P0) Cas9-knock-in mice by intracerebroventricular (ICV) injection (SI Appendix, Fig. S1B). The introduction of this vector resulted in the high-efficiency knockout of NeuN in the cerebral cortex (SI Appendix, Fig. S1 C and D). Thus, we developed a highly efficient method for postnatal neuronal gene disruption.

Traditional methods to perform genome editing to knock-in an epitope tag, a fluorescent protein, or some other protein of interest rely on homology-directed repair (HDR). However, the efficiency of HDR is very low in postmitotic cells like neurons. Recently, several homology-independent knock-in methods using Non-Homologous DNA End Joining (NHEJ) have been developed and allow higher efficiency of knock-in in neurons (2–4). This knock-in occurs stochastically and labels neurons sparsely at a maximum efficiency of ~20% (3). We generated single vector-based homology-independent knock-in vectors that can be constructed in one step of cloning (SI Appendix, Fig. S1E). As proof-of-concept, we targeted β2 spectrin, a protein widely expressed in neurons and glia. We generated AAV vectors and injected them into P0 Cas9-knock-in mice by ICV injection. We observed the knock-in of a spaghetti monster fluorescent protein with V5 epitope tag (smFP-V5) into β2 spectrin throughout the brain (SI Appendix, Fig. S1F) including the cerebral cortex, hippocampus, and cerebellum (SI Appendix, Fig. S1G).

Development of Hide-and-Seek Genome Editing.

We developed the Hide-and-Seek genome editing vectors by combining gene disruption with endogenous protein tagging using NHEJ for both. We designed this vector to induce high-efficiency gene disruption (i.e., knockout of one protein) while permitting detection of another targeted gene by knock-in (Fig. 1A). To reduce the number of common promoters contained in the vector, we used the U6 promoter for expression of the sgRNAs for gene disruption, and the H1 promoter for expression of a sgRNA targeting both the gene of interest and the donor recognition sites flanking the tag for the homology-independent knock-in (13). As proof-of-concept, we used primary hippocampal neurons to knockout (KO) PSD95, a postsynaptic scaffolding protein, and knock-in (KI) a 3xV5 tag into β-actin, a cytoskeletal protein distributed throughout neurons and highly enriched at dendritic spines. Transduction of neurons using a high titer of the control AAV vector (KO: control; KI: β-actin) resulted in efficient and sparse labeling of neurons by β-actin knock-in, while PSD95 was strongly detected at spines by immunostaining (Fig. 1 B and F; Ctrl). The transduction of high-titer knockout AAV vectors [KO: PSD95 at 600,000 viral genomes/cell (vg/cell); KI: β-actin] yielded labeling of neurons through knock-in to β-actin, but most neurons, whether they had knock-in to β-actin or not, lacked PSD95 (Fig. 1 B and F; high titer). In contrast, low-titer AAV vector transduction (KO: PSD95 at 6,000 vg/cell), resulted in β-actin knock-in neurons that nearly always lacked PSD95, while surrounding non-knock-in neurons were mostly PSD95-positive (Fig. 1 B and F; low titer).

Fig. 1.

Hide-and-Seek genome editing. (A) Schematic of the Hide-and-Seek vector. The gRNA1, gRNA2, and gRNA3 target the gene of interest for knockout (gene A), and gRNA4 targets the gene of interest for knock-in (gene B) and the donor sequence flanking the donor tag used for homology-independent knock-in. (B) Representative images of primary cultured neurons targeted for N-terminal knock-in of β-actin and knockout of PSD95. The control neurons and the PSD95 knockout neurons with high-titer infection were infected with AAV vectors at a concentration of 600,000 vg/cell. The PSD95 knockout neurons with low-titer infection were infected with AAV vectors at a concentration of 6,000 vg/cell. Neurons were stained using antibodies against V5-tag (knock-in, green), PSD95 (red), and MAP2 (blue). (Scale bars, 20 µm in the Upper panel and 5 µm in the lower magnified panel.) (C) Quantification of the percentage of MAP2+ neurons with knockout of PSD95 using different amounts of viral infection. N = 4 independent experiments. One-way ANOVA with Tukey’s multiple comparisons test. In all violin plots, dotted and dashed lines indicate quartiles and the median, respectively. (D) Quantification of the percentage of β-actin knock-in neurons with knockout of PSD95 using different amounts of viral infection. N = 4 independent experiments. One-way ANOVA with Tukey’s multiple comparisons test. In all violin plots, quartiles and the median are 0 or 100. (E) Quantification from panels C and D, with data fitted using the Sigmoidal 4PL model. Dots represent the mean, and error bars indicate ±SEM. (F) Summary of the Hide-and-Seek genome editing method and results. In control neurons (KO: control; KI: β-actin) infected with high-titer AAV vectors, knock-in occurs without a knockout since there is no PSD95 sgRNA. In knockout neurons (KO: PSD95; KI: β-actin) using high viral titers, both knock-in and knockout are induced. However, in knockout neurons using low viral titers, not all neurons are infected with enough AAV for knock-in and knockout. The knock-in signal is an indicator of higher viral transduction, and these neurons are more likely to be knockout.

By progressively decreasing the AAV vector concentration, we quantified the knockout efficiency in 1) the total MAP2+ neuronal population (Fig. 1 C and E), and 2) the β-actin knock-in neuronal population (Fig. 1 D and E). In the MAP2+ neuronal population, we found that reducing the AAV vector amount by a factor of 10 (60,000 vg/cell) did not affect the knockout efficiency. A further reduction to 6,000 vg/cell significantly reduced knockout efficiency (Fig. 1 C and E). Doses of 600 vg/cell or 60 vg/cell yielded little or no observable knockout. Conversely, when we examined only the β-actin knock-in positive neuronal population, decreasing the vector concentration by a factor of 10 (60,000 vg/cell) or 100 (6,000 vg/cell) did not alter knockout efficiency (Fig. 1 D and E). However, we observed a significant reduction in knockout efficiency at concentrations of 600 and 60 vg/cell. Although the efficiency of knock-in was reduced using low-titer viral transduction, there were enough knock-in neurons for quantification even at 60 vg/cell. On average, 10-fold more vg/cell is required to achieve an equivalent knockout efficiency in the MAP2+ neuronal population as compared to the β-actin knock-in positive neuronal population (Fig. 1E). Although low-titer transduction reduces knockout efficiency, it remains likely that knock-in neurons are also knocked out. Taken together, these findings support the conclusion that if a knock-in event occurred using the Hide-and-Seek AAV vector, the neuron is also a successful knock-out (Fig. 1F).

Flexible Protein Manipulation by Hide-and-Seek Genome Editing.

We expanded the Hide-and-Seek genome editing strategy to target a range of genes, integrating differential knockout and knock-in. First, we applied the same PSD95 knockout sgRNAs but replaced the knock-in sgRNA with GluA1 for the specific visualization of dendritic spines (Fig. 2A). In control neurons (KO: control; KI: GluA1), PSD95 colocalized with the knock-in GluA1 at the dendritic spines. In contrast, the expression of PSD95 knockout sgRNAs together with the GluA1 knock-in sgRNA (KO: PSD95; KI: GluA1) disrupted the expression of PSD95 while simultaneously labeling GluA1 (Fig. 2 A and F).

Fig. 2.

Application of Hide-and-Seek genome editing to diverse combinations of proteins in vitro. (A) Representative images of primary cultured neurons targeted for knock-in of GluA1 and knockout of PSD95. Neurons were labeled using antibodies against V5-tag (knock-in; red), PSD95 (green), and MAP2 (blue). (Scale bar, 10 µm.) (B) Representative images of primary cultured neurons targeted for knock-in of β4-Spectrin and knockout of Neurofascin. Neurons were stained using antibodies against V5-tag (knock-in; red), Neurofascin (green), and MAP2 (blue). (Scale bar, 10 µm.) (C) Representative images of primary cultured neurons targeted for knock-in of Lmnb1 and knockout of NeuN. Neurons were stained using antibodies against V5-tag (knock-in; red), NeuN (green), and MAP2 (blue). Scale bar, 10 µm. (D) Representative images of primary cultured neurons targeted for knock-in of Lmnb1 and knockout of Mecp2. Neurons were stained using antibodies against V5-tag (knock-in; red), Mecp2 (green), and MAP2 (blue). (Scale bar, 10 µm.) (E) Representative images of primary cultured neurons targeted for knock-in of Pdha1 and knockout of MFF. Neurons were stained using antibodies against V5-tag (knock-in; red), MFF (green), and MAP2 (blue). (Scale bar, 10 µm.) (F–J) Quantification of the percentage of knockout in knock-in neurons. N = 6 (F), 4 (G), 6 (H), 6 (I), and 4 (J) independent experiments. P values were calculated using a nonparametric two-sided Mann–Whitney test. The total number of neurons analyzed is also indicated. Error bars, ±SEM. (K) Quantification of the number of abnormal mitochondria morphology in control or MFF knockout neurons that also have knock-in for pdha1. N = 4 independent experiments for each condition. The P value was calculated using a nonparametric two-sided Mann–Whitney test. The total number of neurons analyzed is also indicated. Error bars, ±SEM.

To illustrate the flexibility of the system not only for different genes but also different subcellular domains, we targeted proteins enriched in nonsynaptic regions of the neuron. The cell adhesion molecule Neurofascin and the cytoskeletal protein β4 spectrin are both located at the AIS. In control neurons (KO: control; KI: β4 spectrin), all β4 spectrin knock-in neurons had overlapping Neurofascin immunoreactivity (Fig. 2B). In contrast, knockout of Neurofascin (KO: Neurofascin; KI: β4 spectrin) led to a profound loss of Neurofascin-labeled AIS (Fig. 2 B and G).

We next designed knock-in sgRNAs to target Lmnb1, a protein found at the nuclear lamina; epitope tagging of this protein serves as a reliable marker to identify knock-in neurons. For knockout, we targeted NeuN and Mecp2, proteins located at the soma and nucleus, respectively (Fig. 2 C and D). In control neurons, NeuN (KO: control; KI: Lmnb1) and Mecp2 (KO: control; KI: Lmnb1) showed strong immunoreactivity overlapping with the knock-in Lmnb1 signal. In contrast, knockout (KO: NeuN; KI: Lmnb1 and KO: Mecp2; KI: Lmnb1) resulted in profound loss of NeuN and Mecp2 (). Finally, we generated knockout sgRNAs for the mitochondrial protein MFF (Mitochondrial Fission Factor) and a knock-in sgRNA for another mitochondrial protein, Pdha1. In control neurons (KO: control; KI: Pdha1), MFF immunostaining colocalized with the epitope-tagged Pdha1 knock-in (Fig. 2E). In contrast, transduction with vectors to express MFF sgRNAs (KO: MFF; KI: Pdha1) significantly reduced immunostaining for MFF (Fig. 2 E and J). Furthermore, since MFF is essential for mitochondrial fission and mitochondrial health, its loss profoundly altered mitochondrial morphology (Fig. 2 E and K). Thus, Hide-and-Seek genome editing is a powerful and flexible strategy to simultaneously knockout genes of interest while knocking-in epitope tags to other proteins of interest.

Hide-and-Seek Genome Editing Reveals Mechanisms of Axo-Axonic Synapse Assembly In Vitro.

GABA_A receptors (GABA_ARs) are enriched at inhibitory synapses through binding to the scaffolding proteins Gephyrin (Gphn) and Collybistin. Gphn interacts strongly with the α1 and α3 GABA_AR subunits while Collybistin preferentially interacts with α2 GABA_AR subunits (α2-GABA_ARs) (14); AIS axo-axonic synapses are enriched with α2-GABA_ARs (15). Previous work suggests that Collybistin is necessary for AIS axo-axonic synapses (14), and both Gphn and Collybistin form clusters at the AIS (16, 17). Consistent with the notion that Collybistin is essential for AIS axo-axonic assembly, a mouse model with a human pathogenic variant in ARHGEF9 (the gene that encodes Collybistin) showed loss of inhibitory AIS axo-axonic synapses (17). However, the specific contribution of Gphn to axo-axonic synapses remains unknown.

To further demonstrate the utility of Hide-and-Seek genome editing to identify molecular mechanisms of nervous system development, we investigated the role of Gphn in assembly of axo-axonic synapses. Neuroligins (Nlgns) are cell adhesion molecules that interact transsynaptically with neurexins to regulate synapse formation and function. Nlgn2 interacts directly with Gphn to initiate the formation of GABA_A-R clustering at GABAergic perisomatic synapses (18). Given its importance as a postsynaptic organizer of perisomatic inhibitory synapses, we used endogenously tagged and clustered Nlgn2 as a potential readout of assembled inhibitory axo-axonic synapses. We first confirmed Nlgn2 localization in cultured neurons using homology-independent knock-in (SI Appendix, Fig. S2A); we designed two sgRNAs targeting Nlgn2’s C-terminus. After transduction, we labeled neurons for the epitope tag and confirmed labeling occurred at synapse-like puncta on the soma and AIS (SI Appendix, Fig. S2B). We then performed in vivo knock-in to Nlgn2, and coimmunostained brain sections for V5 (tagged Nlgn2) and parvalbumin (marker of inhibitory neurons; SI Appendix, Fig. S2C). We counted V5+ neurons in the cortex (249, 269, and 334 neurons from three mice) and found that only 3.5% of V5+ neurons were also PV+. These results indicate that our knock-in is strongly biased toward excitatory neurons and may reflect intrinsic differences in knock-in efficiency or viral tropism.

Next, we generated Hide-and-Seek vectors targeting Nlgn2 for knock-in and Gphn for knockout. To simplify the vector construction, we generated a two-component version of the Hide-and-Seek vectors that are made by 1) a “KO + donor vector” containing three U6 promoter-driven knockout sgRNAs and a donor epitope tag, and 2) a “KI vector” with two U6 promoter-driven knock-in sgRNAs—one for the donor recognition sequence and another for the genomic target Nlgn2 (Fig. 3A). Each plasmid can be constructed in a single step, enhancing flexible targeting. Although the vector was separated into two parts, successful knock-in requires the concurrent transduction of both vectors, ensuring that knock-in neurons likely have Gphn knocked out. In control neurons (KO: control; KI: Nlgn2), Nlgn2 knock-in puncta were enriched on the soma and the AIS, and colocalized with the presynaptic protein Vgat (Fig. 3 B and C). In Gphn knockout neurons (KO: Gphn; KI: Nlgn2), the number of Nlgn2 puncta located at any given AIS was significantly reduced (Fig. 3 B–D). Although the number of Nlgn2 puncta on the soma was also significantly reduced, the reduction was less dramatic compared to the AIS (Fig. 3E). Together, these results suggest that Gphn is required for AIS axo-axonic synapse assembly in vitro.

Fig. 3.

Gphn is required for axo-axonic and axo-somatic synapse assembly in vitro. (A) Schematic of the two-component Hide-and-Seek vector. The gRNA1, gRNA2, and gRNA3 target the gene of interest for knockout (Gphn). The gRNA4 targets the donor recognition sequence flanking the tag sequence, and the gRNA5 targets the gene of interest for knock-in (Nlgn2). (B) Representative images of primary cultured neurons targeted for knock-in of Nlgn2 and knockout of Gphn. Neurons were immunostained using antibodies against V5-tag (knock-in; red), Neurofascin (blue), and Vgat (green). The dashed boxes are magnified in panel C. (Scale bar, 20 µm.) (C) Magnified images from B showing immunostaining for Nlgn2 knock-in (red), Neurofascin (blue), and Vgat (green). An outline of the AIS and location of each site for Nlgn2 and Vgat labeling is shown below each microscopy image. (Scale bar, 10 µm.) (D and E) Quantification of the number of Nlgn2 knock-in puncta at the AIS (D) and on the soma (E). N = 3 independent experiments. One-way ANOVA with Tukey’s multiple comparisons test. In all violin plots, dotted and dashed lines indicate quartiles and median, respectively. The total number of neurons analyzed is also indicated.

Hide-and-Seek Genome Editing Reveals Mechanisms of Axo-Axonic Synapse Assembly In Vivo.

Chandelier cells (ChCs) are a unique type of inhibitory neuron that exclusively form AIS axo-axonic synapses, and these inhibitory synapses are preferentially located at the distal end of the AIS (19, 20). In the mouse primary visual cortex, ~60% of AIS axo-axonic synapses are from ChCs (21). Although other inhibitory neurons may also innervate the AIS, it is not their exclusive target (22–24). However, ChCs are typically absent in neuronal cultures, and Gphn knockout may yield different results in vivo. Furthermore, Gphn whole-body knockout mice die at birth due to respiratory failure (8). To determine the role of Gphn in vivo and its potential contribution to ChC-mediated axo-axonic synapses, we prepared high-titer AAV vectors from the plasmids we previously used in culture experiments (Fig. 3). We performed ICV injection of AAV into P0 Cas9 knock-in mouse brain. We killed mice at P30, collected brains, and examined layer 2/3 PyN for any changes to axo-axonic synapse density (Fig. 4A). We first confirmed efficient tagging of endogenous Nlgn2 in cortical neurons using homology-independent knock-in. Immunostaining of the epitope-tagged Nlgn2 showed knock-in neurons were relatively sparse and distributed randomly throughout the cortex, with clear separation of synaptic signals on the soma and AIS (Fig. 4B).

Fig. 4.

Gphn is required for axo-axonic synapse but not for axo-somatic synapse assembly in vivo. (A) AAV vectors were delivered by ICV injection at P0 into Cas9 knock-in mice. Mice were killed at P30, and the brains were coronally sectioned. The viral concentration of the KI vector was 6.64E+13 GC/mL. The viral concentration of the “KO + donor vector” was 8.77E+12 GC/mL for sgRNA control, 2.51E+13 GC/mL for sgRNA Gphn knockout #1, and 2.90E+13 GC/mL for sgRNA Gphn knockout #2. (B) Immunostaining of brain sections from mice with knock-in for Nlgn2 and control knockout sgRNAs using antibodies against the V5-tag (green), Nlgn2 (red), and Trim46 (blue). The knock-in of Nlgn2 reveals axo-axonic synapses along the Trim46-labeled AIS. (Scale bar, 10 µm.) (C) Representative images of PyN in the brain targeted for knock-in of Nlgn2 and knockout of Gphn. Neurons were labeled using antibodies against the V5-tag epitope inserted in Nlgn2 (red, knock-in), Vgat (green, inhibitory presynapse), and β4 spectrin (blue). (Scale bar, 10 µm.) (D) Quantification of the number of Nlgn2 knock-in puncta at the AIS. N = 3 mice for control and 4 mice for Gphn knockout using sgRNAs #1 and #2. One-way ANOVA with Tukey’s multiple comparisons test. In all violin plots, dotted and dashed lines indicate quartiles and the median, respectively. The total number of neurons analyzed is also indicated. (E) Quantification of the number of Nlgn2 knock-in puncta at the proximal (first half of the AIS from the soma) and distal AIS (last half of the AIS). N = 3 mice for control and four mice for Gphn knockout using sgRNAs #1 and #2. P-values between the proximal to distal segments were calculated using a parametric two-sided t test. P-values between the groups were calculated using a two-way ANOVA with Sidak’s multiple comparisons test. Error bars, ±SEM. The total number of neurons analyzed is also indicated. (F) Quantification of the number of VGAT knock-in puncta at the AIS. N = 1 mouse for control and 1 mouse for each of the different Gphn knockout constructs (sgRNAs #1 and #2). One-way ANOVA with Tukey’s multiple comparisons test. In all violin plots, dotted and dashed lines indicate quartiles and the median, respectively. The total number of neurons analyzed is also indicated. (G) Quantification of the number of Nlgn2 knock-in puncta at the soma. N = 3 mice for control and four mice for Gphn knockout using sgRNAs #1 and #2. One-way ANOVA with Tukey’s multiple comparisons test. In all violin plots, dotted and dashed lines indicate quartiles and the median, respectively. The total number of neurons analyzed is also indicated.

Using the two-component vector system described above (Fig. 3A) and in contrast to Gphn whole-body KO mice, mice transduced with Gphn knockout sgRNAs (KO: Gphn; KI: Nlgn2) survived at least until P30. These mice had clear motor deficits and even hind limb paralysis (Movie S1). In contrast to the Gphn KO mice, the survival of mice transduced with the Hide-and-Seek vectors likely reflects only a subset of neurons exhibiting complete knock-out of Gphn. We found that in Layer 2/3 PyN, Nlgn2 puncta were present at the soma and AIS in control neurons (KO: control; KI: Nlgn2) (Fig. 4 C–E, G). However, Gphn knockout (KO: Gphn; KI: Nlgn2) significantly reduced Nlgn2 puncta at the AIS (Fig. 4 C–E, G). Although the number of Nlgn2 puncta on the soma was significantly reduced in vitro (Fig. 3E), it remained unchanged in the brain (Fig. 4G), possibly reflecting the wider variety of inhibitory neurons in vivo compared to in vitro. This suggests diverse inhibitory neurons in vivo may compensate for inhibitory synapse formation on the soma even in the absence of Gphn, but inhibitory AIS axo-axonic synapses are not rescued. Additionally, we found that persisting axo-axonic synapses in Gphn knockout neurons were primarily located at the proximal AIS (Fig. 4E). Thus, the loss of Gphn preferentially affects the assembly of AIS axo-axonic synapses, and especially the ChC axo-axonic synapses that are normally found along the distal AIS.

Discussion

Here, we developed Hide-and-Seek genome editing and showed the simultaneous combination of reliable and flexible knockout and knock-in. The strategy permits the analysis of protein localization in various subcellular regions and in knockout neurons without the use of protein-specific antibodies. For example, we showed the successful knock-in of β-actin (ubiquitous), GluA1 (spines), β4 spectrin (AIS), Lmnb1 (nuclear envelope), Pdha1 (mitochondria), and Nlgn2 (inhibitory synapses), all in the context of knockout for another protein (e.g., Gphn, Neurofascin, PSD95, MFF, NeuN, or Mecp2). Importantly, the knock-in labeling illuminates not only the protein and the structures of interest but also serves as a reliable indicator of successful knockout. This is especially useful to identify single knockout neurons in vivo where transduction events may be infrequent.

Although we used smFP-V5 and 3xV5 tags to demonstrate successful knock-in, the strategy is flexible and “donors” may be fluorescent proteins, other epitope tags, and even enzymes such as TurboID for use in proximity biotinylation (25). For example, experiments where the investigator wishes to examine fluorescent signals, or changes to protein interactomes in the context of a knockout neuron are feasible using Hide-and-Seek. This greatly accelerates the pace of investigation since it bypasses the need to generate combinations of knockout, knock-in, and transgenic mouse models. In addition, the knock-in may target the N terminus or C-terminal regions of the protein of interest.

Traditional knockout vectors often use overexpression cassettes (e.g., green fluorescent protein (GFP)) to label transduced cells. This can be misleading since the overexpressed protein may be detectable even with poor vector transduction that does not result in knockout. In contrast, Hide-and-Seek genome editing ensures that both knock-in and knockout have sufficient levels of Cas9 and guide RNA (gRNA), and confirms that knock-in neurons are also effectively knocked out. Nevertheless, there may be several concerns associated with this genome editing strategy. For example, knock-in tags could be integrated into genes targeted for knockout, resulting in the detection of tagged proteins intended for knockout. However, we never observed this situation. We use three gRNAs (gRNA1-3) per target and our knockout gRNAs were not designed to target the last exons. In addition, the C-terminal donor cassette is designed with multiple stop codons after the tag sequence corresponding to each reading frame but omitting a poly-A signal. If this cassette is integrated anywhere upstream of the last exon, the stop codons generate a premature termination codon that induces the degradation of the mRNA by nonsense-mediated decay. For N-terminal tagging there is an upstream stop codon cassette restricting expression of the tag to the N terminus. In our experiments, we used N-terminal tagging of β-actin targeting the first exon based on Willems et al. (4), who showed that this gRNA has no effect on nonlabeled neurons. Thus, using both N- and C-terminal tagging, the tagged proteins are produced only when the cassette is inserted precisely in the first or last exon.

An additional concern is that the relatively low knock-in efficiency (~20%), could result in many nonlabeled neurons undergoing knockout due to the formation of indels from the knock-in gRNA. However, we did not observe this for C-terminal knock-in since the location of the knock-in CRISPR sites we used is in the last exon where most nonintegrated indels are benign. Since the integration site is located in the last exon, the addition of the tag truncates the targeted protein.

An important limitation to this targeting strategy is that if the N- or C-terminus of the protein plays important roles in protein localization or protein–protein interactions, even correct targeting may appear as a knockout or partial loss-of-function. Nevertheless, as reported by Gao et al. (3), 24/25 targeted genes were successfully knocked in. Thus, for most proteins, knock-ins are well tolerated. When C-terminal knock-in is not tolerated, an N-terminal knock-in may be successful. Although for most proteins, successful knock-in does not alter protein levels, in some instances correct knock-in may result in reduced protein levels (4). Thus, there may be gene-specific changes in protein expression levels.

Finally, we always generated multiple gRNAs for each target to confirm that identical results were obtained with different gRNAs, thereby mitigating concerns about off-target integration, knockout, or changes in protein localization or expression levels. The use of higher efficiency Cas9 nucleases may also limit off-target events.

To illustrate the utility of Hide-and-Seek, we demonstrated that Gphn is required for assembly of AIS axo-axonic synapses but not for axo-somatic synapses in the brain. This suggests that the mechanisms underlying formation of axo-somatic synapses and some proximal axo-axonic synapses are similar, but that mechanisms of assembly for most axo-axonic synapses are different from those found on the soma. This difference may reflect the important barrier properties of the AIS excluding some proteins from this domain. The reduction in VGAT-labeled presynaptic terminals along AIS after loss of Gphn is also consistent with the reduction in AIS axo-axonic synapses. ChCs form axo-axonic synapses mainly at the distal AIS (19), while the synapses formed at the proximal AIS are mainly from non-ChCs (26). Although some specific cell adhesion molecules, including L1CAM and Contactin-1 have been shown to play important roles for the targeting and innervation of ChCs to the distal AIS (12, 27), the presynaptic partners facilitating the binding and subsequent assembly of axo-axonic synapses remain unknown.

Hide-and-Seek genome editing is a versatile, rapid, and powerful methodology to investigate protein functions across different subcellular domains in neurons. The in vivo approach, induced by the postnatal injection of AAV vectors, enables structural and functional analyses in specific knockout cells within a single generation of mice. By eliminating the time required for the generation and maintenance of specific mouse lines, Hide-and-Seek genome editing accelerates the pace of neuroscience research and permits the identification of molecular mechanisms.

Materials and Methods

Animals.

All experiments were performed in accordance with the NIH Guide for the Care and Use of Laboratory Animals and were approved by the Baylor College of Medicine Animal Care and Use Committee (protocol no. AN4634). Timed pregnant Sprague-Dawley rats were obtained from Charles River Laboratories. Rats were euthanized for embryo collection at E18. Neonatal (P0-P2) Cas9 transgenic mice (catalog #Jax: 027650, RRID: IMSR_JAX:027650) were used for intraventricular injection of AAV for knockout and knock-in of targeted genes. There was no consideration of sex for primary neuronal cultures or AAV injection.

Cell Culture.

Primary cultures of hippocampal and cortical neurons were obtained from E18 Sprague-Dawley rat embryos. Hippocampi and cortices were dissected and dissociated. Neurons were plated onto Poly-D-Lysine (Sigma) and laminin-coated glass coverslips (Life Technologies) at a density of ~1.25 × 10⁴ cells/cm² for imaging. Neurons were maintained in Neurobasal medium (Life Technologies) containing 1% Glutamax (Life Technologies), 1% penicillin and streptomycin (Life Technologies), and 2% B27 supplement (Life Technologies) in an incubator at 37 °C with 5% CO₂. Half of the medium was removed and replaced every 5 d.

Antibodies.

The following antibodies were used for immunofluorescence studies (dilutions for each antibody are indicated in parentheses): mouse monoclonal anti-PSD95 (1:200; Invitrogen Cat# MA1-046, RRID:AB_2092361), mouse monoclonal anti-V5 (1:500; Invitrogen Cat# R960CUS, RRID: AB_159298), chicken polyclonal anti-MAP2 (1:1,000; EnCor Biotechnology Cat# CPCA-MAP2, RRID:AB_2138173), rabbit polyclonal anti-β4 spectrin (1:1,000, Rasband lab, RRID:AB_2315634), chicken polyclonal anti-neurofascin (1:500; R&D Systems Cat# AF3235, RRID:AB_10890736), mouse monoclonal anti-NeuN (1:500; Millipore Cat# MAB377, RRID:AB_2298772), rabbit polyclonal anti-Mecp2 (1:200; Sigma Cat# 07-013, RRID:AB_2144004), rabbit polyclonal anti-MFF (1:500; Proteintech Cat# 17090-1-AP, RRID:AB_2142463), rabbit polyclonal anti-GFP (1:1,000; Invitrogen Cat# A11122, RRID:AB_221569), guinea pig polyclonal anti-VGAT (1:500, Synaptic Systems Cat# 131004, RRID:AB_887873), guinea pig monoclonal anti-Trim46 (Synaptic Systems Cat# 377308, RRID:AB_2924929, 1:500), rat monoclonal anti-HA (Millipore Sigma Cat# 11867423001, RRID:AB_390918, 1:1,000), rabbit polyclonal anti-Parvalbumin (PV, Novus Biologicals Cat#NB120-11427, RRID:AB_791498, 1:500), mouse monoclonal anti-Gephyrin (Synaptic Systems Cat#147011, RRID:AB_2810215, 1:500), and mouse monoclonal anti-Nlgn2 (UC Davis/NIH NeuroMab Facility clone L107/39, RRID:AB_2877595, 1:500). A nanobody against PSD95 conjugated with Alexa Fluor 647 (Nanotag Biotechnologies Cat# N3702-AF647-L, RRID: AB_2936216, 1:500) was used for immunofluorescence studies. The following secondary antibodies were used: Aminomethylcoumarin (AMCA) anti-mouse IgG1 (1:250 Jackson Immunoresearch labs 115-155-205), AMCA anti-chicken IgY (1:250 Jackson Immunoresearch labs 103-155-155), AMCA anti-guinea pig (1:250 Jackson Immunoresearch labs 106-155-003), Alexa Fluor 488 anti-mouse IgG1 (1:1,000 Thermo Fisher Scientific A21121), Alexa Fluor 488 anti-mouse IgG2a (1:1,000 Thermo Fisher Scientific A21131), Alexa Fluor 488 anti-rabbit (1:1,000 Thermo Fisher Scientific A11034), Alexa Fluor 488 anti-chicken IgY, (1:1,000 Jackson Immunoresearch labs 103-545-155), Alexa Fluor 488 goat anti-guinea pig (1:1,000, Thermo Fisher Scientific A11073), Alexa Fluor 594 anti-mouse IgG1 (1:1,000 Thermo Fisher Scientific A21125), and Alexa Fluor 594 anti-mouse IgG2a (1:1,000 Thermo Fisher Scientific A21135). Hoechst fluorescent reagent (1:100,000; Thermo Fisher Scientific Cat# H3569, RRID:AB_2651133) was used to label nuclei.

Immunofluorescence Labeling.

Cultured rat primary hippocampal neurons were fixed in 4% paraformaldehyde (PFA, pH 7.2) for 15 min at 4 °C. Acutely dissected brains were drop fixed in 4% PFA (pH 7.2) for 60 min at 4 °C. Brains were then equilibrated in 20% and 30% sucrose in 0.1 M Phosphate Buffer (PB) overnight at 4 °C. Brains were then sectioned at 12 to 25 µm and mounted on coverslips. Fixed neurons and brain sections were permeabilized and blocked in 0.1 M PB with 0.3% Triton X-100 and 10% normal goat serum(PBTGS) for 1 h. Cells and sections were then incubated in primary antibodies diluted in PBTGS overnight at room temperature or 4 °C. Tissues and cells were then washed three times using PBTGS for 5 min. each. Fluorescent secondary antibodies were then diluted in PBTGS and added to cells and tissues for 1 h. Coverslips were then washed once using PBTGS, 0.1 M PB, and finally 0.05 M PB for 5 min. each. Coverslips were then mounted using Vectashield plus (Vector Labs) anti-fade mounting media.

Plasmid Construction.

The plasmid vectors were generated following strategies as described previously (12). Briefly, the triple gRNA knockout vector used in the supplemental experiments expressed gRNA1-3 from three independent U6 promoters as described (28). The homology-independent knock-in vectors used in the supplemental experiments expressed gene-specific gRNA (GS-gRNA) and donor-specific gRNA (DS-gRNA) from two independent U6 promoters. The smFP-V5 and 3xV5 were used as knock-in donors (Addgene plasmids #59758). The following sequence was used for the donor recognition site: GCGATCGTAATCACCCGAGT-GGG. The Hide-and-Seek vectors used in Figs. 1 and 2 expressed gRNA1-3 for KO from three independent U6 promoters as described above. Another gRNA (gRNA4) for KI was expressed from the H1 promoter. The target sequence of gRNA4 was also inserted to flank the tag sequence for homology-independent knock-in. The two-component Hide-and-Seek vectors used in Figs. 3 and 4 expressed gRNA1-3 for KO from three independent U6 promoters as described above. The smFP-V5 was used as a knock-in donor. DS-gRNA (gRNA4) and GS-gRNA (gRNA5) were expressed from two independent U6 promoters from the KI vector. DNA fragments were ligated together using an In-Fusion Snap Assembly Master Mix (Takara). The AAV-SpCas9 plasmid (Addgene plasmid #60957) was modified by removing the HA tag and used in all in vitro CRISPR experiments. The sgRNA sequences for knock-in and knockout are listed in SI Appendix, Table S1. The sgRNA sequences were selected using CRISPOR: https://crispor.gi.ucsc.edu/crispor.py. All DNA constructs were verified by sequencing (Genewiz and plasmidsaurus).

AAV Vector Production.

Small-scale AAV cell lysates were produced following protocols as described (12). Briefly, Human Embryonic Kidney (HEK)293T cells were triple-transfected with AAV plasmid, helper plasmid (Agilent Technologies, Cat # 240071), and serotype PHP.S plasmids (Addgene plasmids #103006) with PEI Max (Polysciences, Cat # 24765). The medium was changed the next day of transfection, and cells were incubated for 3 d after transfection. HEK cells were then collected and lysed with citrate lysis buffer (38 mM citric acid, 75 mM sodium citrate, 75 mM sodium chloride, and 100 mM magnesium chloride, pH 5). The extracted solution was centrifuged at 10,000× g for 10 min to remove debris and mixed with 10% vol/vol of 2 M Tris-HCl (pH 9.5). This small-scale AAV solution was stored at 4 °C and used for neuronal transduction into cultured neurons.

AAV vectors for in vivo transduction were produced as follows: HEK293T cells were triple-transfected with AAV plasmid, helper plasmid (Agilent Technologies, Cat # 240071), and serotype PHP.S plasmids (Addgene plasmids #103006) with PEI Max (Polysciences, Cat # 24765) in two 15-cm dishes. The medium was changed the next day of transfection, and cells were incubated for 5 d after transfection. HEK cells were then collected and lysed with citrate lysis buffer (38 mM citric acid, 75 mM sodium citrate, 75 mM sodium chloride, and 100 mM magnesium chloride, pH 5). The extracted solution was centrifuged at 10,000× g for 10 min to remove debris and mixed with 10% vol/vol of 2 M Tris-HCl (pH 9.5). Meantime, the HEK cell culture medium was mixed with 20% vol/vol of Polyethylene glycol (PEG) solution (40% PEG 8000, 0.5 M NaCl) and incubated overnight at 4 °C. The supernatant was then centrifuged at 3,000× g for 15 min to pull down a viral pellet. The viral pellet was dissolved in the AAV solution obtained from the cell lysates described above. This mixture was then concentrated using the Amicon Ultra-15, Ultracel-100 Membrane (#UFC910024) by centrifugation at 5,000× g for ~60 min. The AAV solution was washed with PBS three times by centrifugation at 5,000× g for ~60 min. At the final spin, The AAV solution was concentrated to ~150 μL, and this AAV solution was used for the in vivo injection. This viral solution was aliquoted and stored at −80 °C until use.

Viral Transduction of Neurons.

For viral transduction of cultured neurons, 10 µL of AAV-Cas9 and 10 µL of AAV-sgRNA and donor, AAV-3x-sgRNA-smFP, or AAV-Hide-and-Seek were added into a well of a 12-well plate at 0 to 1 DIV. The medium was replaced 2 d after infection. For viral transduction of neurons in vivo, AAV vectors were injected into the lateral ventricles of neonatal mice as described previously (12). Briefly, P0 to P2 pups were anesthetized on ice, and 1 to 2 µL of AAV vectors was injected bilaterally. The pups were placed in a heated cage until the animals recovered and then returned to their mother. Tissues were collected 4 to 8 wk after infection.

Image Acquisition and Analysis.

Images of immunofluorescence were captured using an Axio-imager Z2 microscope fitted with an apotome attachment for structured illumination (Carl Zeiss MicroImaging) and a Nikon Eclipse Ni2. 20× (0.8 NA), 40× (0.95 NA), and 63× (1.4 NA) objectives were used. Images were taken using Zen 3.2 (Zeiss) or NIS-Elements (Nikon). All measurements were taken with the same exposure times, and immunolabeling was also performed at the same time. Images were exported to Fiji, Adobe Photoshop, and Adobe Illustrator for figure presentation. Immunostained images of either vGat or V5 (Nlgn2) were processed in Fiji using difference of Gaussian filtering (σ = 1, 2) to reduce background noise and clarify puncta signals. These processed images were then merged with the AIS marker (βIV spectrin) image. Some figures were generated using Biorender.

Statistics and Reproducibility.

Unpaired, two-tailed Student’s t test was used for all statistical analyses unless otherwise indicated. Data were analyzed using Microsoft Excel and GraphPad Prism. All error bars are ±SEM unless otherwise indicated. Colors were added in Adobe Illustrator.

Supplementary Material

Appendix 01 (PDF)

Movie S1.

A mouse that received intracerebroventricular injection at P1 of Hide-and-Seek AAV to knockout Gephyrin and knock-in Neuroligin 2.

Data, Materials, and Software Availability

All study data are included in the article and/or supporting information.