Abstract
The cell adhesion molecule leucine-rich repeat transmembrane neuronal protein 2 (LRRTM2) is crucial for synapse development and function. However, our understanding of its endogenous trafficking has been limited due to difficulties in manipulating its coding sequence (CDS) using standard genome editing techniques. Instead, we replaced the entire LRRTM2 CDS by adapting a two-guide CRISPR knock-in method, enabling complete control of LRRTM2. In primary rat hippocampal cultures dissociated from embryos of both sexes, N-terminally tagged, endogenous LRRTM2 was found in 80% of synapses, and synaptic LRRTM2 content correlated with PSD-95 and AMPAR levels. LRRTM2 was also enriched with AMPARs outside synapses, demonstrating the sensitivity of this method to detect relevant new biology. Finally, we leveraged total genomic control to increase the synaptic levels of LRRTM2 via simultaneous mutation of its C-terminal domain, which did not correspondingly increase AMPAR enrichment. The coding region of thousands of genes span lengths suitable for whole-CDS replacement, suggesting this simple approach will enable straightforward structure–function analysis in neurons.
- ALFA tag
- AMPA receptors
- cell adhesion molecules
- conditional reporter
- CRISPR
- knock-in
- LRRTM2
- neuron
- protein structure–function analysis
- protein trafficking
- synapse biology
Significance Statement
Synaptic cell adhesion molecules are transmembrane proteins vital for neurotransmission, and their genes are frequently linked to disease. Mechanistic studies are challenging; however, because overexpression alters their protein trafficking and perturbs neuronal function, traditional gene-editing techniques typically permit manipulation at just single sites. Here, we replaced the entire coding sequence (CDS) of the critical adhesion molecule leucine-rich repeat transmembrane protein 2 (LRRTM2) with a freely editable custom donor sequence and made the approach even more versatile through introduction of knock-in–dependent cell markers. Using whole-CDS replacement, we discover new aspects of LRRTM2 subcellular distribution and test its role in regulating synaptic proteins. The approach should be suited to structure–function analysis of many other neuronal proteins in their endogenous genetic locus.
Introduction
Synaptic organizing molecules, or cell adhesion molecules (CAMs), are unique transmembrane proteins that reach across the synaptic cleft, where they play critical roles in both cis- and transsynaptic biological processes from synaptogenesis to receptor trafficking and plasticity. The trafficking and expression of CAMs are tightly regulated, and mutations in CAM genes can severely disrupt neuronal function and are linked to neurological disease (Südhof, 2008; Leshchyns’Ka and Sytnyk, 2016; Schroeder and De Wit, 2018). Accordingly, alterations of CAM expression levels, such as due to the use of exogenous expression for visualization and mutational analysis, can alter neuronal function and cloud interpretations of CAM biology. It is therefore necessary to measure and manipulate the endogenous protein itself to advance our understanding of CAMs. However, endogenous structure–function studies have been challenging, particularly because one critical approach is to mutate CAM sequences at multiple locations simultaneously. For example, N-terminal, extracellular tagging is ideal for most CAMs to visualize surface-expressed protein, but many critical protein interaction sites are quite distant from the N-terminus and even intracellular, as CAM C-tails play important roles in their anchoring and synaptic signaling. Standard genome editing techniques have largely been insufficient to make these changes simultaneously.
One example of a CAM for which new tools are needed is the synaptic CAM leucine-rich repeat transmembrane neuronal protein 2 (LRRTM2). LRRTM2 resides in the postsynaptic membrane, where it binds to PSD-95 at its C-terminus and interacts transsynaptically with presynaptic neurexins (de Wit et al., 2009; Ko et al., 2009; Siddiqui et al., 2010). LRRTM2 exerts considerable power over synaptic function at mature synapses, largely as a result of its influence over AMPA receptor (AMPAR) anchoring and abundance at synapses (Siddiqui et al., 2010). Knockdown of LRRTM2 affects basal AMPAR synaptic enrichment and synaptic strength (de Wit et al., 2009), and dual knockdown or knock-out of LRRTM2 and its sister protein LRRTM1 abolishes long-term potentiation (LTP) in hippocampal CA1 neurons (Soler-Llavina et al., 2013; Bhouri et al., 2018). Furthermore, enzymatic cleavage of the LRRTM2 extracellular domain in a knockdown–rescue context disorganizes AMPAR subsynaptic distribution within minutes and reduces total synaptic AMPAR content over the following hours (Ramsey et al., 2021). LRRTM2 also plays a dose-dependent role in synaptogenesis, where over- or underexpression leads to corresponding changes in synapse density (de Wit et al., 2009; Ko et al., 2011). These results highlight the central role LRRTM2 plays in regulating synaptic strength at mature synapses and motivate a detailed analysis of its molecular mechanisms.
As is true for many proteins, our understanding of LRRTM2 cellular functions has relied on genetic knock-out and overexpression or antibody-based immunocytochemistry. Each of these approaches has considerable shortcomings. In particular, methods that manipulate LRRTM2 expression levels dramatically alter synaptic development and can engage compensatory mechanisms from related CAMs (de Wit et al., 2009; Ko et al., 2011). The use of knockdown–rescue can reduce overexpression phenotypes (de Wit et al., 2009; Ramsey et al., 2021), but expression via exogenous promoters can nevertheless cause diverse effects that are difficult to adequately monitor. While some studies have utilized antibody detection of endogenous LRRTM2, the low signal-to-noise ratio and lack of possibility to manipulate the protein sequence limit their scope (Linhoff et al., 2009; Lloyd et al., 2023). Knock-in mouse models could bridge this gap; however, LRRTM2 has numerous binding partners, and making a separate mouse model for each mutation is expensive and inefficient. Disappointingly, traditional CRISPR knock-in approaches to tag the LRRTM2 N-terminus have proven infeasible due to a lack of suitable PAM sites in the protein-encoding region, and N-terminal knock-ins could threaten the integrity of the reading frame due to possible indels. C-terminal tagging is less than ideal due to the inability to accurately identify protein present on the cell surface and likely disrupts the PDZ-binding motif. Recently, the CRISPR approach, Targeted Knock-In with Two guides (TKIT), was used to replace a protein-coding exon using two intron-localized guide RNA sequences (Fang et al., 2021). This approach allowed N-terminal tagging of proteins without shifting the reading frame, as any Cas9-induced indels occur in intronic regions. We considered whether this strategy could be expanded and adapted to permit tagging and mutagenesis anywhere within a protein where the entire protein-coding sequence (CDS) can be replaced. To test the feasibility of this approach while investigating LRRTM2 trafficking, we have applied whole-CDS replacement TKIT to the rat Lrrtm2 gene.
In this work, we demonstrate whole-CDS replacement in neurons as well as the simultaneous tagging and mutagenesis of the Lrrtm2 coding region. We utilized endogenously tagged LRRTM2 to measure its synaptic enrichment as well as its relationship with AMPAR enrichment at synapses. To enable rapid identification and measurement of knock-in cells, we also developed a strategy to express a marker conditional on successful knock-in, enabling new and flexible experimental designs. Finally, we demonstrated the power of this technique by combining N-terminal tagging with distal point mutations to increase synaptic surface enrichment of LRRTM2. Together, our findings provide context for our understanding of LRRTM2 function as well as a novel method for future structure–function studies in neurons and other postmitotic cells.
Materials and Methods
CRISPR design and plasmids
The NCBI Rnor 6.0 database LRRTM2 sequence (NC_005117.4) was used as a reference for both guide and donor design. Guides positioned at least 50 bp from the intron–exon splicing boundary (5′ guide) or over 50 bp into the 3′UTR (3′ guide) were identified using the Benchling CRISPR guide ranking tool, targeting those with the best on-target and off-target scores based on previous publications (Hsu et al., 2013; Doench et al., 2016), and then were cloned into the pX330 backbone as previously described (Fang et al., 2021). Guide sequences: 5′ GTTTTAATCTCTCTTATACA 3′ (Guide 1, anneals to Intron 1) and 5′ CTTTTTAAGTAGGAAGCCAG 3′ (Guide 2, anneals to the 3′UTR). Once cloned into the pX330 vector under identical U6 promoters, the guide construct was grown in NEB Stable cells at a reduced temperature of 30°C to prevent bacterial recombination of the promoters. Guide 2 was cloned with guanine at the 5′ end after the U6 promoter, as described by Fang et al. (2021). 3xALFA epitope tag and IRES2-Cre insertions, as well as YACA (Y501A/C504A) mutations, were added to the donor by either Gibson NEB HIFI Assembly or NEB Q5 site-directed mutagenesis as appropriate. The 3xALFA tag was inserted after the signal peptide along with a small linker (TS) after the tag, and IRES2-Cre was added immediately after the stop codon. For lentiviral expression, the guides and donor were combined into the pFW lentiviral backbone (Dharmasri et al., 2023) by the NEB HIFI Assembly. FLEx-mTagBFP2 was generated by NEB HIFI Assembly, replacing the mCherry-KASH in Addgene #139652 (a gift from Harold MacGillavry; http://n2t.net/addgene:139652; RRID:Addgene_139652) with the mTagBFP2 gene and replacing the hSyn promoter with the CAG promoter for improved expression levels. FLEx-IRES-EGFP was made similarly, with a CMV promoter in place of CAG. HA-spCas9 was a gift from Harold MacGillavry (Addgene plasmid #131506; http://n2t.net/addgene:131506; RRID:Addgene_131506), and constructs to make lentivirus (pMD2.G and psPAX2) were gifts from Didier Trono (Addgene plasmid #12259; http://n2t.net/addgene:12259; RRID:Addgene_12259; Addgene plasmid #12260; http://n2t.net/addgene:12260; RRID:Addgene_12260). ALFA and HA-tagged LRRTM2 KDR variants for transient expression in HEK cells were generated by replacing the GFP tag in KDR GFP-LRRTM2 previously described (Ramsey et al., 2021) using NEB HIFI Gibson cloning. All sequences were confirmed by whole plasmid sequencing (Plasmidsaurus) using Oxford Nanopore Technologies with custom analysis and annotation. These plasmid sequences will be deposited in public databases upon publication.
Note that we originally tried to combine the FLEx fluorescent protein with the guides/donor lentiviral plasmid, but full plasmid sequencing revealed that growing a construct with both IRES2-Cre and FLEx genes in bacteria caused a reversal of the flipped gene. Based on previous literature, we suspect that this could be due to bacteria recognizing the IRES2 sequence and expressing the Cre protein themselves (Colussi et al., 2015). Thus, we elected to separate these constructs into two viruses, with the Cas9 virus as a third virus due to its large size and the limitations of lentiviral packaging. The guides and Cas9 were cloned into separate viral constructs to avoid Cas9 activity at the Lrrtm2 locus in the absence of a donor sequence to replace it.
Genomic sequencing
Cultured hippocampal neurons were infected with either knock-in viruses (Cas9 and guides/donor) or guides/donor alone at DIV1. DNA was extracted and purified at DIV21 using the Wizard Genomic DNA Purification Kit (Promega) according to the manufacturer's instructions. The LRRTM2 gene was amplified using primers that anneal on either side of the guide1 Cas9 cut sites (5′ AGCCAGTGAATTCCCGTTTT 3′, 5′ AGGCGAACTGGGATAGTCCGCA 3′). This PCR product was gel-purified, and then PCR was amplified again using a reverse primer that anneals specifically to the knock-in HA sequence (5′ CATTTAGGTGGACAACTAGTAGCGTAGTCTGGTACATCAT 3′) and the same forward primer as in the first PCR run. Gel-purified PCR products from the Round 1 PCR control condition and the Round 2 PCR knock-in condition (no product was made from the control condition due to the lack of knock-in HA sequence in the Lrrtm2 locus) were Sanger sequenced (Genewiz/Azenta) to confirm knock-in in the correct genetic locus.
Lentivirus production
HEK293T cells (ATCC CRL-3216) maintained in DMEM plus 10% FBS supplemented with penicillin/streptomycin were plated at high density and transfected using PEI as we have described (Dharmasri et al., 2023). Cells were incubated for 6 h followed by a media change to standard neuronal culture media. Lentivirus was allowed to accumulate in the media for 2 d before harvesting, and debris was removed by centrifugation at 1,000 × rpm for 5 min. Virus aliquots were used fresh or else frozen at −80°C. Based on prior titrations and to achieve high levels of coinfection, neurons were infected with 100 μl of each virus.
Immunostaining
Dissociated, mixed-sex hippocampal cultures were prepared as previously described (Dharmasri et al., 2023). Knock-in preparations were infected at DIV1 with 100 µl of unconcentrated Cas9, guide/donor, and marker viruses. To immunostain LRRTM2 and synaptic proteins, coverslips were removed from their culture media at DIV21 and blocked for 5 min in 1% BSA in Tyrode's buffer and then live-labeled with ALFA-Alexa Fluor 647 nanobody (NanoTag) and mouse anti-pan GluA (Synaptic Systems) at 1:500 and 1:200, respectively, in the same solution composition for 12 min at room temperature. To avoid background from aggregation, aliquots of nanobody are vortexed thoroughly for 2 min followed by centrifugation at max speed for 1 min. Coverslips were then rinsed twice briefly in phosphate-buffered saline (PBS) and fixed in 4% paraformaldehyde (PFA) plus 4% sucrose in PBS for 9 min at room temperature. The cells were washed with PBS–100 mM glycine (PBS–glycine) three times for 5 min each, permeabilized with 0.3% Triton X-100 in PBS–glycine for 15 min at room temperature, and then incubated with PSD-95 nanobody conjugated to AZ568 dye (NanoTag) at 1:100 or a rabbit anti-Bassoon antibody (Cell Signaling Technology) at 1:500 with goat anti-rabbit Alexa Fluor 488 at 1:750 in 4% TBS–milk for 45 min at room temperature. Secondary labeling for the Bassoon primary antibody was performed with a goat anti-rabbit Alexa Fluor 568 secondary diluted at 1:750 in 4% TBS–milk for 45 min at room temperature. Coverslips were then rinsed in TBS three times for 5 min prior to imaging.
To test different knock-in tags, we transfected HEK cells using Lipofectamine 2000 (Thermo Fisher Scientific) with an EGFP cell fill to normalize for transfection as well as the respective KDR LRRTM2 constructs with 1xALFA, 3xALFA with interstitial linkers, and a 3xALFA without linkers. A 1xHA tag visualized with a traditional primary–secondary antibody approach was included for comparison. We then immunolabeled, fixed, and stained as above. Cells were fixed and surface stained with either the ALFA nanobody or an HA primary antibody followed by a fluorescent secondary. Both approaches utilized the Alexa Fluor 647 dye to detect LRRTM2 as well as identical imaging parameters. Images were analyzed in ImageJ for Alexa Fluor 647 fluorescence intensity normalized to GFP as a transfection control.
Microscopy
Images were acquired on an Andor Dragonfly spinning disc confocal (Andor) attached to a Nikon Ti2 Eclipse inverted microscope base with a 60× Plan Apo 1.49 NA objective. Excitation laser light (405, 488, 561, or 638 nm) from an Andor ILE, flattened by an Andor Beam Conditioning Unit, was passed to the sample by a 405/488/561/640 quadband polychroic (Chroma). Emission light was passed through an appropriate bandpass filter [FF02-447/60-25 (Semrock), ET525/50 and ET600/50 (Chroma), or Em01-R442/647 (Semrock), for 405, 488, 561, and 638 nm emission, respectively] and collected on a Zyla 4.2 sCMOS camera. Cells of interest were imaged with confocal z-stacks with 0.5 µm z-steps at 25–90% laser power with 200 ms exposure (400 ms for 638 nm channel), with each channel imaged sequentially.
Analysis
Image processing was performed using macros and plugins in Fiji/ImageJ (Schindelin et al., 2012), and image file names were blinded during analysis. All z-stacks were converted to maximum intensity projections, and then x–y chromatic aberrations were corrected using the Register Channels tool of the NanoJ-Core plugin (Laine et al., 2019) and four-color TetraSpeck bead images acquired prior to imaging as a standard.
We used a semiautomated ImageJ macro similar to that described by Dharmasri et al. (2023) to detect synapses. In brief, the macro provides user-guided image cropping, followed by automated thresholding to isolate the ALFA-LRRTM2 signal, and then uses the puncta detection plugin SynQuant (Wang et al., 2020) to detect synapses on knock-in neurons from the PSD-95 staining. Detected synapses were converted to ROIs and used to measure synapse area and PSD-95 intensity, as well as intensity of LRRTM2 and/or AMPARs within those puncta. Due to high cellular variability, we took the cellular average for each channel and assigned each synapse ROI a Z-score based on their relative intensity within that channel to the cellular average.
Synapses along knock-in dendrites were categorized as containing or lacking LRRTM2 by comparing the distributions of ALFA labeling along knock-in dendrites or neighboring dendrites (acquired simultaneously and analyzed with the same pipeline as above in parallel):
For detection of LRRTM2 and/or AMPAR-containing puncta within and outside PSD-95 or Bassoon puncta, we handpicked elliptical ROIs around puncta in each channel using Fiji ROI selection tools and then quantified fluorescence intensities as above. At confocal resolution, there is occasionally a small offset between pre- and postsynaptic proteins even at the same synapse, depending on the size and orientation of the synapse. We thus linked Bassoon puncta to the nearest AMPAR or LRRTM2 puncta if their ROIs overlapped to any degree. Raw integrated density was measured within each hand-drawn ROI, followed by background subtraction using similarly sized ROIs drawn nearby in the background region of each image. Each puncta was categorized as “positive” or “negative” for each of LRRTM2, AMPARs, and PSD-95 or Bassoon by utilizing a first local minimum cutoff in the intensity distribution of each protein across the dataset. These categories were derived and plotted as pie charts using a combination of RStudio and GraphPad Prism.
Bioinformatics
To assess how applicable a whole-CDS replacement of a similar size to our approach would be to other genes, genetic data on Rattus norvegicus were pulled from the search function of the University of California Santa Cruz Genome Database (https://genome.ucsc.edu/cgi-bin/hgGateway). To measure the coding sequence (CDS) spanning region for each gene and splice isoform, we subtracted the stop codon position from the start codon position. In practice, replacement size will depend on the proximity of suitable PAM sites and guide sequences on either side of the CDS. For this analysis, we set the maximum CDS spanning region size as 10,000 bp. Search parameters entered into the UCSC GD were as follows: absolute value(endCDS–startCDS) ≤10,000 bp. Alternative splice isoforms of the same gene were included as separate genes because they likely differ in genomic span and carry differing biological significance. The histogram was produced in RStudio, and the cumulative plot was produced in GraphPad Prism.
Statistics
Data from ImageJ measurements were processed, and statistics were calculated using RStudio (Posit). Frequency distributions were produced for each of the synaptic intensity measurements to evaluate normality. These distributions were not normally distributed as evaluated with a Shapiro–Wilks test, and we therefore elected to utilize Wilcoxon tests to compare LRRTM2-containing and LRRTM2-lacking synapses, as well as YACA and wild-type LRRTM2 mutant conditions. Graphs were produced in RStudio or GraphPad Prism (Dotmatics). Due to the high-throughput nature of our semiautomated synapse detection method, we pruned the number of points displayed in scatter plots by averaging every five synapses together based on ranked data (PruneRows, Prism). While this had very little effect on the slope of the regression lines, it does influence the R2 values, so we have reported the values of regressions performed on the raw data in the text.
Results
Two-guide CRISPR approach TKIT can be adapted for whole-CDS replacement of LRRTM2
To decipher the trafficking patterns of the key synaptic cell adhesion molecule LRRTM2, we designed an approach to simultaneously measure and manipulate the protein while maintaining its endogenous transcriptional regulation and expression level. HITI is a common CRISPR genome editing method used in postmitotic cells such as neurons that allows for DNA insertions that could be useful for tagging native LRRTM2 (Suzuki et al., 2016). However, HITI is limited to insertion at single sites, which for LRRTM2 would only allow either an N-terminal tag or protein mutagenesis, not both. In addition, HITI targets PAMs to make double-stranded breaks within the coding region and therefore may generate indels that alter the reading frame unpredictably (Fig. 1A, left). Therefore, we considered whether we could edit the entire mature LRRTM2 CDS by adapting the exon-replacement technique Targeted Knock-In with Two guides (TKIT; Fang et al., 2021). Instead of targeting a single PAM site within the coding region, TKIT utilizes two guide RNAs targeting PAMs in the intron regions flanking an exon, which directs Cas9 to excise the exon entirely and enables its replacement with a synthetic exon (Fig. 1A, middle). This has the advantage of improving knock-in efficiency and design, since there are often multiple guide RNAs and PAM sites throughout the intronic sequences to choose from, and intron-targeted guides reduce the effects of indels as they will occur outside the coding region. TKIT has previously been used to replace the 5′-most coding exon of diverse neurotransmitter receptors with a synthetic version that includes an N-terminal fluorescent tag (Fang et al., 2021). We wondered whether this approach could be modified such that the entire coding region of a gene, in our case LRRTM2, could be replaced instead (Fig. 1A, right). Notably, nearly the entire LRRTM2 protein is encoded in a single large exon, with the 5′ UTR and only four bases of signal peptide in Exon 1, and the rest of the coding region, plus the 3′ UTR, in Exon 2 (Fig. 1B). This genomic organization offered the attractive possibility of replacing the entire protein-coding sequence of LRRTM2 with an entirely customizable one using TKIT. Therefore, we modified the TKIT approach to replace much of the second exon, including all the coding sequence contained therein. Given that the signal peptide is cleaved cotranslationally, this approach would allow us to replace the entire mature LRRTM2 protein with a donor sequence of our choosing.
Successful whole-CDS replacement of LRRTM2. A, Diagram of traditional HITI, TKIT, and whole-CDS replacement CRISPR methods. Locations of possible Cas9-induced indels are indicated with red zigzag lines. B, Diagram of the Rattus norvegicus Lrrtm2 gene with CRISPR guide locations. Note that the entire coding sequence (orange), minus four bases of the signal peptide that is cleaved from the mature protein, is confined to Exon 2. Guide 1 anneals to the intronic region and Guide 2 to the 3′UTR. The replacement donor (engineered CDS, black) contains an N-terminal epitope tag between the signal peptide and mature protein. C, To determine the best tag for visualizing endogenous LRRTM2, we evaluated three versions of the epitope tag ALFA, which can be stained using a dye-conjugated nanobody: 1xALFA, 3xALFA with linkers between the ALFA tags, and a 3xALFA without linkers. A 1xHA tag visualized with a traditional primary–secondary antibody approach was included for comparison. Left, Measured intensities were normalized for transfection using cotransfected GFP cell fill. Right, Background fluorescence intensity analyzed across conditions from left, normalized to the HA condition. While the primary–secondary antibody amplification provided a higher signal than ALFA nanobody (left), secondary antibody-induced background fluorescence was also higher (right). Interestingly, background ALFA levels were low, and tripling the ALFA tag improved the signal by approximately 50% but only when no interstitial linkers were included. D, Genomic sequencing of the knock-in. Samples of genomic DNA from either the guides/donor virus alone (control) or the guides/donor and Cas9 viruses (knock-in, KI) underwent two rounds of PCR amplification, first to amplify the Lrrtm2 gene and then again to amplify the knock-in tag (no product in the control condition). Sanger sequencing shows the 3xHA tag sequence in the KI condition, without indels. E, Image of a cultured hippocampal knock-in neuron, stained for 3xALFA-LRRTM2. Sparse knock-in permits the identification of synaptic puncta along knock-in versus neighboring dendrites. F, Images of a dendrite from a knock-in neuron, stained for 3xALFA-LRRTM2 (yellow), PSD-95 (magenta), and presynaptic Bassoon (cyan). LRRTM2 puncta were largely synaptic (white arrows), with some LRRTM2 puncta outside PSD-95 synapses (blue arrow). G, Endogenous LRRTM2 (KI, right) is expressed at far lower levels than typical exogenous expression via knockdown–rescue (KDR, left). Matched staining, imaging, and look-up tables show that while both KDR and KI show synaptic localization, knockdown–rescue ALFA labeling was considerably higher than knock-in both at synapses and between (blue and orange, respectively, line scan, lower right). Black arrows indicate examples of spine enrichment, and white arrows indicate dendritic signal. Line scan locations shown with red lines in the respective images.
To replace the LRRTM2 coding sequence using TKIT, we designed a 5′ guide RNA (Guide 1) targeting a PAM site within intron 1 of Rattus norvegicus Lrrtm2 and a 3′ guide RNA (Guide 2) targeting a PAM site after the Lrrtm2 stop codon within the 3′UTR, resulting in a large (2,131 bp) span of replacement that exceeds those of previous efforts (Fang et al., 2021; Arsić et al., 2022; Ziak et al., 2024). The replacement donor sequence was engineered using rat genomic DNA and subcloned to include an epitope tag between the signal peptide and LRRTM2 N-terminal regions. The donor was also flanked by flipped and reversed guide sequences to allow reversed integration to be fixed, as used in both HITI and TKIT (Suzuki et al., 2016; Fang et al., 2021; Fig. 1B). We tested several N-terminal tags for both specific labeling intensity and nonspecific background from the affinity reagent using exogenous LRRTM2 expression in HEK cells (Fig. 1C). We found that the ALFA nanobody (Götzke et al., 2019) displayed lower nonspecific signal than an HA antibody plus secondary antibody. Furthermore, we were able to increase specific staining of the ALFA epitope by adding a triple ALFA tag without interstitial linkers, which had approximately 50% higher fluorescence intensity than either the single ALFA tag or a triple ALFA tag with linkers. Therefore, we utilized a triple ALFA tag without linkers in the donor sequence for fluorescent labeling, resulting in a donor sequence of 2,260 bp (note that for genomic PCR and sequencing experiments, a triple HA tag was used instead for convenience). As the transfection efficiency of cultured neurons is low, we delivered the necessary DNA with two lentiviruses, one to express the guides and donor sequence and a second to express Cas9.
We first validated the success of LRRTM2 whole-CDS replacement by infecting dissociated embryonic rat hippocampal neurons with either Cas9 and LRRTM2 guides/donor lentiviruses or with guides/donor virus only as a control and then sequenced genomic DNA after 3 weeks in culture. Briefly, we utilized PCR to isolate the Lrrtm2 gene and amplified knock-in–positive alleles using sequential PCRs and sequenced the resulting amplicons via Sanger sequencing. Positive sequences for the 3xHA tag were identified in the knock-in condition (Fig. 1D, middle) but not in the control condition lacking Cas9 (Fig. 1D, bottom), indicating that the tag was successfully incorporated into the correct gene locus and that knock-in was successful. Note that we did not introduce indels at the signal peptide–tag junction or at the tag–protein junction, as this region was contained entirely within the engineered donor sequence as is true for TKIT. We next evaluated whether the knocked-in 3xALFA tag would allow us to visualize the endogenous expression of LRRTM2. We performed surface fluorescent labeling of 3xALFA-LRRTM2 on live, dissociated hippocampal cultured neurons infected 3 weeks prior with the knock-in constructs, and then fixed the cells and additionally immunolabeled for PSD-95 and the presynaptic scaffold protein Bassoon to visualize synapses. Fluorescent confocal imaging showed isolated cells with anti-ALFA labeling in discrete puncta along the cell body and dendrites (Fig. 1E). Close inspection revealed these puncta typically occurred at synapses, as indicated by colocalization with both Bassoon and PSD-95, consistent with previous studies of LRRTM2 (de Wit et al., 2009; Linhoff et al., 2009; Fig. 1F, especially white arrowheads), with some LRRTM2 puncta localized outside PSD-95 (blue arrowhead). In line with previous TKIT-based knock-ins, we commonly observed 20–30 ALFA-stained cells per coverslip with this method; note that this may be an underestimate of the knock-in efficiency as some cells may be knocked-in but not express LRRTM2, and some cells may not be coinfected by both viruses. These results together demonstrate that our knock-in was successful and could be used to visualize endogenous LRRTM2 in cultured neurons.
The previous best practice for visualizing LRRTM2 has been knockdown–rescue (KDR), which is frequently assumed to minimize overexpression due to the knockdown of endogenous expression. However, this method still typically relies on exogenous or unregulated promoters, which often express proteins at higher than endogenous levels. To compare knock-in LRRTM2 expression levels with those induced by KDR, we performed live labeling of surface ALFA on cells infected with knock-in constructs and on cells transfected with a knockdown–rescue 3xALFA-LRRTM2 construct. As expected, ALFA staining was clearly far higher in the KDR condition than the knock-in (Fig. 1G). Furthermore, while the expression pattern of both was largely synaptic, the LRRTM2 signal in the KDR condition showed higher and more consistent levels of LRRTM2 at each spine (Fig. 1G, line scan, dark arrows) and also showed higher ALFA labeling along the dendrite outside of dendritic spines (Fig. 1G, line scan, white arrows). This indicates that overexpression of LRRTM2, even following knockdown, can alter its surface and subcellular trafficking, highlighting the importance of labeling and measuring endogenous protein.
Rapid identification of whole-CDS knock-in cells
As is true for many endogenously labeled proteins, the low endogenous expression levels of LRRTM2 made visualizing knock-in cells or tracing their morphology for analysis difficult. We therefore leveraged our whole-CDS replacement system to design a knock-in conditional marker. The conditional effect was achieved by adding an internal ribosome entry sequence (IRES) and Cre gene after the stop codon in the LRRTM2 donor containing the 3xALFA tag. This donor could allow for the expression of a FLExible marker protein (Schnütgen et al., 2003), conditional upon successful knock-in of 3xALFA-LRRTM2 (Fig. 2A). To test the utility of this approach, we infected neurons with three lentiviruses expressing the 3xALFA-LRRTM2-IRES-Cre donor and guide RNAs, Cas9, and FLEx-mTagBFP2, expecting that when the 3xALFA-LRRTM2-IRES-Cre donor is knocked into the Lrrtm2 locus and the mRNA translated, Cre will also be expressed and able to induce FLEx-mTagBFP2 expression (Fig. 2A, bottom). This strategy successfully produced cells expressing both knock-in 3xALFA-LRRTM2 and knock-in–dependent mTagBFP2 cell fill, facilitating rapid visual identification on the microscope (Fig. 2B). Notably, the 3xALFA tag and IRES-Cre sequences are 1,449 bp away from one another on the donor sequence, a simultaneous knock-in impossible to make by other methods. The donor used was also 3,925 bp long, over two times larger than the largest TKIT donor previously reported (Fang et al., 2021) and among the largest we are aware of successfully being knocked in using nonhomologous end joining (NHEJ)-based CRISPR methods. We observed a fraction of neurons that were positive for mTagBFP2 without detectable surface ALFA labeling (16%), presumably reflecting cell types with successful knock-in that express very little or only transiently expressed LRRTM2. Cells in this category were excluded from further analysis. A small number of neurons (8%) were positive for surface ALFA staining without detectible mTagBFP2 expression, potentially due to a lack of coinfection by all three viruses. Nevertheless, the expression of the marker protein was routinely high enough to rapidly identify presumed knock-in cells for image acquisition. An important advantage of this approach is the potential to use alternative conditional reporters. To illustrate this, we utilized a FLEx-IRES-EGFP lentivirus in place of the FLEx-mTagBFP2 and observed EGFP-positive knock-in cells (Fig. 2C). Conceivably, any FLEx fluorescent protein, sensor, or optogenetic tool could be used to accommodate numerous experimental approaches, demonstrating the utility of our whole-CDS control.
Rapid identification of whole-CDS knock-in cells. A, Schematic of knock-in conditional marker implementation. An internal ribosome entry site (IRES) sequence and a Cre gene were inserted into the donor construct between the LRRTM2 STOP codon and the 3′UTR (IRES-Cre, orange). Thus, Cre is expressed once the donor is successfully incorporated into the LRRTM2 gene. A separate lentiviral construct was used to deliver FLEx-mTagBFP2 cell fill, which can be targeted by the knocked-in Cre. B, Knock-in neuron expresses the mTagBFP2 cell fill. Surface immunostained 3xALFA-LRRTM2 (yellow), cytosolic mTagBFP2 (cyan). C, The IRES-Cre approach permits flexibility in the conditional marker. Due to the flexible multivirus approach, it is simple to swap mTagBFP2 for any other FLEx protein marker, such as EGFP (magenta). Three viral constructs were used: Cas9, fuides/donor, and FLEx marker.
Endogenous LRRTM2 labeling enables investigation of LRRTM2 trafficking and synaptic enrichment
We next set out to deploy our LRRTM2 knock-in and conditional reporter system to characterize LRRTM2 expression and synaptic trafficking at endogenous levels. We first quantified the average synaptic LRRTM2 expression level per cell and found a wide range of intensities across cells (Fig. 3A). This suggests that LRRTM2 expression varies within the transcriptomic profile of neurons in these CA1-enriched hippocampal cultures, indicating possible cell-intrinsic control of LRRTM2 expression. We considered that some of the cells with higher LRRTM2 intensity might be homozygous for the 3xALFA-LRRTM2 knock-in and that those with lower expression might be heterozygous. However, we observed a smooth, rather than bimodal, distribution of average synaptic intensity (Fig. 3B), consistent with the regulation of LRRTM2 expression levels by activity history or other cell-specific mechanisms.
Variation in endogenous synaptic LRRTM2 content reflects key markers of synaptic strength. A, Exemplar hippocampal neurons expressing 3xALFA-LRRTM2 at low, medium, and high levels. B, Histogram of synaptic LRRTM2 expression across the dendrite, averaged by cell. C, Image of an example stretch of dendrite expressing knock-in 3xALFA-LRRTM2 (yellow) immunolabeled for PSD-95 (magenta). LRRTM2 is highly synaptic in localization (exemplars, white arrows), but did not appear in every synapse on the dendrite (blue arrow). D, Distribution of synaptic LRRTM2 mean intensity along knock-in dendrites, normalized for cellular variability using a per-cell Z-scoring method. E, Cumulative frequency plot of ALFA mean intensity at synapses in knock-in neurons (pink) or neighboring non–knock-in dendrites (blue). Z-scores for LRRTM2 mean intensity measures were calculated from the mean and standard deviation of synaptic ALFA signal within neighboring non–knock-in synapses. F, Bar plot of synapses along knock-in dendrites categorized as containing (pink) or lacking (blue) LRRTM2. The 95th percentile of non–knock-in mean intensity Z-scores was used to delineate the cutoff between LRRTM2-containing and LRRTM2-lacking synapses. G, Violin plot comparing synapse area along knock-in dendrites that do or do not contain LRRTM2 (pink or blue, respectively; p < 2.2 × 10−16). H, Scatter plot correlating synapse area with synaptic LRRTM2 content (raw integrated density Z-score) in LRRTM2-containing synapses. I, Violin plot comparing PSD-95 content within synapses along the knock-in dendrites that do or do not contain LRRTM2 (p < 2.2 × 10−16). J, Scatter plot correlating PSD-95 content with synaptic LRRTM2 content at LRRTM2-containing synapses. In violin plots, green lines represent means, and black lines represent medians for each distribution. Scatter plots have been averaged across every five ranked data points for visibility; statistics are calculated from raw data. Lines show linear regression with gray dotted lines representing 95% confidence intervals. Sample size throughout: 54 neurons from three culture replicates and 8,550 synapses.
LRRTM2 is synaptically localized and plays a key role in synaptic function (de Wit et al., 2009; Soler-Llavina et al., 2013; Bhouri et al., 2018). However, our observation that LRRTM2 puncta were more discrete and variable with our knock-in than with KDR due to overexpression artifacts (Fig. 1F) indicated its endogenous distribution may be more nuanced. We therefore investigated in detail the variability of endogenous LRRTM2 content at individual synapses in 3xALFA-LRRTM2 knock-in neurons, immunolabelled for PSD-95 with a fluorescently labeled nanobody (Fig. 3C). Despite prominent colocalization at synapses (Fig. 3C, white arrows), we observed that the quantity of LRRTM2 varied substantially between synapses (coefficient of variation 1.514) and that, in fact, not every synapse along dendrites from knock-in cells contained LRRTM2 (Fig. 3C, blue arrow). To better understand the synaptic variation of LRRTM2 content while accounting for cellular variability of LRRTM2 expression levels (Fig. 3A), we measured LRRTM2 intensity within synapses and normalized for variable cellular expression by calculating cell-based intensity Z-scores for each synapse. We identified synaptic puncta as PSD-95 regions of interest (ROIs) using a semiautomated synapse detection analysis SynQuant (Wang et al., 2020; Dharmasri et al., 2023) of images from 54 knock-in neurons across three culture replicates and calculated Z-scores for each synapse. We measured the area and fluorescence intensity of each labeled protein within these synapse ROIs and normalized for their variable cellular expression by calculating cell-based intensity Z-scores for each synapse using the cellular average and standard deviation for LRRTM2 or PSD-95, respectively (Fig. 3D). This analysis clearly showed that synaptic LRRTM2 content varied considerably within individual neurons, even when normalized for cellular variability.
To define a synapse as lacking LRRTM2, we sought to establish a baseline by examining the off-target ALFA staining. This was done by measuring ALFA intensity at neighboring synapses from non–knock-in neurons, reflecting nonspecific staining and noise, which as expected had Z-scores near 0 (Fig. 3E, neighboring synapses). To compare these two distributions, we recalculated the Z-scores of knock-in LRRTM2 synaptic intensity using the mean and standard deviation of ALFA signal in neighboring PSDs and found the synapses along the knock-in dendrite were, as expected, far brighter in ALFA signal than the neighboring background staining (Fig. 3E). We then used the 95th percentile of non–knock-in synapse ALFA staining Z-scores to delineate a cutoff between LRRTM2-containing and LRRTM2-lacking synapses within the knock-in dataset. Using this cutoff, approximately 80% of the synapses along knock-in dendrites contained LRRTM2 (Fig. 3F). This is slightly higher than a previous estimate utilizing LRRTM2 antibody staining in DIV13–15 mouse hippocampal cultures (Lloyd et al., 2023), which not only could represent a developmental or species difference but also could be a reflection of our ability to more precisely estimate background noise from cells that unequivocally lacked the knock-in and therefore accurately identify true signal.
Our observation of a synapse subtype lacking LRRTM2 raises additional questions about what other differences these synapses exhibit relative to their LRRTM2-containing neighbors. Given that LRRTM2 plays a role in synaptic plasticity and binds to PSD-95, one explanation of its synaptic variability is that synapses with higher LRRTM2 levels might be larger and contain more PSD-95. Indeed, we found that synapses with LRRTM2 exhibited larger PSDs than those without (Wilcoxon test, p < 2.2 × 10−16; r-statistic, −0.142; Fig. 3G). Furthermore, within the population of LRRTM2-containing synapses, larger synapses contained more LRRTM2 protein (slope, 7.061; R2, 0.226; Fig. 3H). Our interpretation is further supported by examining PSD-95 content synapses with LRRTM2 contained more PSD-95 than those without LRRTM2 (Wilcoxon test, p < 2.2 × 10−16; r-statistic, −0.149; Fig. 3I), and PSD-95 and LRRTM2 intensities were positively correlated (slope, 0.442; R2, 0.183; Fig. 3J). These data indicate that the subcellular trafficking of LRRTM2 to synapses is correlated with both synapse size and the amount of PSD-95 present. While previous studies have indicated that overexpressed LRRTM2 does not require PSD-95 binding to be enriched at synapses (Linhoff et al., 2009), our data indicate that nonetheless LRRTM2 trafficking scales with these key markers of synaptic strength.
Synaptic AMPA receptor content scales with LRRTM2 content
Previous studies utilizing LRRTM1/2 KDR or LRRTM1/2 KO replacement with LRRTM2 have established that LRRTM2 stabilizes AMPARs in synapses at baseline and anchors new AMPARs after LTP (Soler-Llavina et al., 2013; Bhouri et al., 2018). Despite these findings, it has been difficult to disentangle how AMPAR trafficking compares with endogenous expression levels of LRRTM2. Furthermore, our ability to visualize endogenous LRRTM2 and identify synapses lacking it gives us the opportunity to compare AMPAR content between these two groups of synapses. We hypothesized that synapses lacking endogenous LRRTM2 would also have reduced AMPAR content relative to synapses with LRRTM2. To examine the relationship between the synaptic enrichment of LRRTM2 and AMPARs, we immunolabeled surface AMPA receptors simultaneously with ALFA-LRRTM2, fixed and stained the neurons for PSD-95 to identify synaptic puncta, and measured synaptic LRRTM2 and surface AMPAR intensity from the same synapses (Fig. 4A). As in Figure 3, we normalized both the LRRTM2 and AMPAR raw integrated density signals using Z-scores based on respective means in each cell and then assessed synapses along knock-in dendrites for the presence or absence of LRRTM2. Consistent with the role of LRRTM2 in AMPAR anchoring, LRRTM2-lacking synapses contained less surface AMPAR signal than synapses with LRRTM2 (Wilcoxon test, p < 2.2 × 10−16; r-statistic, −0.136; Fig. 4B). While LRRTM2-lacking synapses still exhibited AMPAR staining, this is likely due to stabilization by other mechanisms. When LRRTM2 was present, synapses with more LRRTM2 also had more AMPAR content on the surface (slope, 0.3513; R2, 0.1326; Fig. 4C). While these results do not support a model where endogenous LRRTM2 is solely responsible for synaptic AMPAR anchoring, the correlation between AMPAR content and LRRTM2 content supports the hypothesis that LRRTM2 plays a strong role in control of AMPAR trafficking (Soler-Llavina et al., 2013; Bhouri et al., 2018).
Synaptic AMPAR content scales with LRRTM2 content. A, Images of a knock-in neuron labeled for 3xALFA-LRRTM2 (yellow) and surface AMPARs (cyan). Arrows show high coenrichment of AMPARs with LRRTM2. B, Violin plot comparing synaptic surface AMPAR content along the knock-in dendrites that do or do not contain LRRTM2 (pink and blue, respectively; p < 2.2 × 10−16). Green lines represent means and black lines represent medians for each distribution. C, Scatter plot correlating AMPAR content (raw integrated density Z-score) with synaptic LRRTM2 content (raw integrated density Z-score) at LRRTM2-containing synapses. Scatter plot has been averaged across every five ranked data points for visibility; statistics are calculated from raw data. Lines show linear regression with gray dotted lines representing 95% confidence intervals. Sample size: 54 neurons from three culture replicates and 8,550 synapses. D, Images of triply labeled neurons with 3xALFA-LRRTM2 (yellow), PSD-95 (magenta), and surface AMPARs (green). Arrows indicate LRRTM2 clusters that do (white) or do not (red) coenrich with PSD-95. E, Left, Pie chart depicting the proportions of all puncta types from D. Center, Subset of the left, separating LRRTM2, AMPAR, and/or PSD-95-containing puncta types along knock-in dendrites. Right, Subset of the left, showing only AMPAR and/or LRRTM2 clusters without PSD-95 enrichment. Notably, LRRTM2 puncta outside of PSD-95 often also contain AMPAR clusters. Sample size: 22 neurons from three culture replicates and 1,743 puncta. F, Images of triply labeled neurons with 3xALFA-LRRTM2 (yellow), presynaptic marker Bassoon (magenta), and surface AMPARs (green). Arrows indicate LRRTM2 and AMPAR clusters that do (white) or do not (red) coenrich with Bassoon. G, Left, Pie chart depicting the proportions of all puncta types from F. Center, Subset of left, separating LRRTM2, AMPAR, and/or Bassoon-containing puncta types along knock-in dendrite. Right, Subset of the left, showing only extrasynaptic AMPAR and/or LRRTM2 clusters (without Bassoon enrichment). Sample size: 10 neurons from two culture replicates and 1,251 puncta.
Surprisingly, close examination of these triply labeled knock-in cells revealed occasional LRRTM2 puncta that fell outside of PSD-95-demarcated synapses (Fig. 4D, red vs white arrows). It is well documented that extrasynaptic AMPAR pools play a role in regulating synapse strength, particularly during LTP (Penn et al., 2017; Choquet, 2018). Given the role of LRRTM2 in LTP and AMPAR stabilization, we therefore wondered whether these LRRTM2 puncta outside of PSD-95 might also contain AMPARs. To address this, we selected isolated dendrites from 23 knock-in cells across three culture replicates and identified puncta in each channel (surface AMPARs, PSD-95, and 3xALFA-LRRTM2) as ROIs for measurement. As the semiautomated detection was trained to detect puncta that are sized and shaped like synapses and ignore nonsynaptic puncta, these ROIs were hand-selected. We then measured the raw integrated density within these ROIs in all three channels and qualitatively labeled each as positive or negative for each of the labeled proteins based on its intensity, where intensities above the first local minimum of the distribution were described as “positive” for that corresponding protein (Fig. 4E, left pie chart). Consistent with our previous automated analysis, when the hand-selected PSD-95 puncta were pooled together, we found similar numbers of synapses that were LRRTM2-positive and LRRTM2-negative (by hand, 79.7% positive and 20.2% negative; automated, 80.5% positive and 19.5% negative), validating our approaches. When we separately grouped puncta by whether they contained LRRTM2, AMPARs, or PSD-95, it became clear that the majority of LRRTM2 and AMPAR puncta were at PSD-95-containing synapses, yet with small proportions of each either alone or together without PSD-95 (Fig. 4E, middle). We then broke down the non–PSD-95-containing puncta into groups based on their protein expression (Fig. 4E, right). Most such puncta contained both LRRTM2 and AMPARs together (52%), more than the puncta with either protein on its own (29% AMPAR only and 19% LRRTM2 only). The existence of LRRTM2/AMPAR puncta outside PSD-95 synapses could include synapses that do not contain PSD-95 and instead have another scaffold such as SAP-102 or PSD-93 (Elias et al., 2006; Metzbower et al., 2023). However, the presence of LRRTM2/AMPAR puncta outside of PSD-95 synapses raises the possibility that along with its known localization within synapses, LRRTM2 could codiffuse with extrasynaptic AMPARs.
To test whether these LRRTM2/AMPAR puncta were associated with presynaptic terminals, we immunolabeled knock-in cells for 3xALFA-LRRTM2, surface AMPARs, and the presynaptic marker Bassoon (Fig. 4F), which is present at synapses regardless of their postsynaptic scaffold composition. This staining showed a high proportion of synaptic LRRTM2 and AMPAR puncta that were visibly colocalized with presynaptic Bassoon (Fig. 4F, white arrows), along with LRRTM2 and AMPAR puncta that did not colocalize with nearby Bassoon puncta (red arrows), suggesting they were truly extrasynaptic. To quantify this, we performed the same ROI selection analysis as above on isolated stretches of dendrite from 10 neurons across two culture replicates (Fig. 4G). Bassoon puncta far from a postsynaptic punctum (approximately 3.8% of the 1,301 puncta) were removed from the dataset to permit easier comparison with the analysis in Figure 4E. These most likely represent inhibitory synapses that would not have been found in the PSD-95 dataset, and would not contain excitatory-exclusive LRRTM2 or AMPARs (de Wit et al., 2009; Ko et al., 2009; Fig. 4G, left). Comparing the two experiments, we observed that a similar proportion of the puncta that contained both LRRTM2 and AMPARs colocalized with Bassoon (70.7%) as with PSD-95 (66.4%; Fig. 4, compare G, left, and E, left). The small difference between the two experiments could be explained by some LRRTM2 and AMPAR puncta localizing to synapses containing a non-PSD-95 scaffold such as SAP-102 or PSD-93. Proportionally, most LRRTM2 puncta were at synapses, with a small proportion at extrasynaptic sites with or without AMPARs (Fig. 4, compare G, middle, and E, middle). Importantly, however, the apparently extrasynaptic population of puncta containing LRRTM2 and AMPAR together was observed (Fig. 4G, right) regardless of whether PSD-95 or Bassoon was used as the synaptic marker (Fig. 4, compare G, right, and E, right). Together, these results suggest a potential role for LRRTM2 in AMPAR trafficking outside of synapses.
Simultaneous mutagenesis and tagging enabled by whole-CDS replacement permits analysis of mutation-induced changes in LRRTM2 surface expression
A major advantage of whole-CDS replacement is the potential it offers to easily edit a protein of interest at multiple points, as we have already demonstrated by adding IRES-Cre at the C-terminus simultaneously with the N-terminal 3xALFA tag. To demonstrate the power of whole-CDS replacement of LRRTM2, we modified our synthetic donor sequence to include, along with the IRES-Cre, a functional mutation near the C-terminus expected to alter LRRTM2 trafficking. Previous work has shown that surface enrichment of exogenously expressed LRRTM2 can be manipulated through two point mutations in the C-terminal domain, Y501A and C504A (Minatohara et al., 2015; Liouta et al., 2021), which are hypothesized to alter the membrane-targeting trafficking mechanisms of LRRTM2 (Minatohara et al., 2015). Termed “YACA,” these mutations induce a 400% increase in surface LRRTM2 levels in a knockdown–rescue context (Liouta et al., 2021); however, as we established, the KDR protein trafficking pattern may not be the same as endogenous. Our whole-CDS replacement model offers the ideal platform on which to test this. We introduced the YACA mutations into our donor construct to create a 3xALFA-LRRTM2-YACA-IRES-Cre knock-in (Fig. 5A), infected neurons with lentivirus to generate wild type or YACA knock-ins, and then measured mean AMPAR and LRRTM2 content in PSD-95-labeled synapses using SynQuant as above (54 wild type and 62 YACA neurons across three culture replicates; Fig. 5B). Normalized to wild type, the average synaptic LRRTM2-YACA content was 24.6% higher than wild type (Wilcoxon test, p < 0.005; r-statistic, 0.280; Fig. 5C). When we examined the per synapse distribution of LRRTM2 intensities between the two conditions, we found a similar trend where the distribution of YACA-containing synapses was right-shifted relative to wild type (Wilcoxon test, p < 2.2 × 10−16; r-statistic, 0.102; Fig. 5D). Therefore, while the YACA mutations did increase endogenous LRRTM2 surface trafficking as expected, the percent change was dramatically smaller than that observed in a KDR model. These findings reinforce the value of our whole-CDS replacement approach and demonstrate the power of simultaneous genomic editing at multiple sites within a protein.
Simultaneous mutagenesis and tagging enabled by whole-CDS TKIT permits analysis of mutation-induced changes in LRRTM2 surface expression. A, N-terminal 3xALFA tags were inserted in LRRTM2 as in Figure 2, as well as two point mutations in the C-terminus (Y501A and C504A, or YACA). Control condition knock-ins contain tagged wild-type (WT) sequence, and YACA condition knock-ins contain both the tag and the two point mutations. B, Exemplar images of WT or YACA-mutated LRRTM2 (yellow), costained for PSD-95 (magenta) and surface AMPARs (green). C, Violin plot of mean synaptic LRRTM2 content, averaged by cell and normalized to WT, shows the YACA mutation (navy) of LRRTM2 resulted in higher synaptic surface intensity than WT (tan; p < 0.005). Green lines represent means and black lines represent medians for each distribution. D, Distribution of synaptic enrichment of WT or YACA LRRTM2 normalized to WT, showing a rightward shift of the YACA distribution (navy) relative to the WT (tan; p < 2.2 × 10−16). Purple lines represent the mean, and cyan lines represent the median of the respective distributions. E, Violin plot showing mean synaptic surface AMPAR enrichment was not significantly higher in the YACA-mutated condition relative to WT. Green lines represent means and black lines represent medians for each distribution. Sample size: 54 wild-type neurons and 62 mutant neurons from three culture replicates, 8,550 wild-type synapses, and 10,521 mutant synapses.
Previous studies have established that removing LRRTM2 reduces synaptic AMPAR content (Soler-Llavina et al., 2013; Bhouri et al., 2018; Ramsey et al., 2021), and our data indicate that AMPAR content scales with LRRTM2 content at synapses. While we have shown that the YACA mutations moderately increase LRRTM2 content, it is unclear whether this small change would follow the pattern in our correlational data and drive an increase in AMPAR content. We therefore measured synaptic AMPAR staining in each condition and found that the YACA mutations did not have a positive effect on synaptic AMPAR content (Wilcoxon, p = 1; r-statistic, −0.035; Fig. 5E). This suggests that despite the correlation between LRRTM2 synaptic content and AMPAR synaptic content, increasing the LRRTM2 synaptic content by 24.6% via YACA is insufficient or unable to induce a similar increase in AMPAR content.
Applying whole-CDS replacement to other genes
We conclude that whole-CDS replacement TKIT will be a valuable approach for protein structure–function analysis. The structure of the rat Lrrtm2 gene, being essentially contained on a single exon, is undoubtedly highly advantageous for the technique, but other genes with a small CDS but different exon/intron structures are likely amenable to the same approach. With whole-CDS replacement TKIT, internal intronic sequences can simply be included in the donor to preserve endogenous splicing, and the TKIT guide selection strategy has been shown to not disrupt exon/intron splicing (Fang et al., 2021). In fact, the length of sequence replaced, rather than its content or functional complexity, may be more critical for successful knock-in, as donor sizes will likely be limited due to practical DNA size restrictions for transfection or transduction, and local chromosomal conditions near or between the guide binding locations may potentially limit long donor insertion. Therefore, the viability of whole-CDS replacement likely depends largely on the CDS spanning length of a gene (the distance between the start and stop codon positions along the chromosome) and could accommodate a variety of exon/intron structures.
To evaluate the possible application of whole-CDS replacement to other genes, we explored the Santa Cruz Genome Browser for rat coding sequences with a reasonably small CDS spanning length. Out of the rat coding sequences whose CDS spanning lengths are under a practical limit of 10 kb, approximately 42.6% (39,459) are smaller than Lrrtm2 (Fig. 6A,B). This analysis includes genes of suitable size with multiple introns and exons, as well as splice variants, which we speculate should be equally amenable to whole-CDS replacement. Note that while the CDS spanning length for Lrrtm2 is 1,882 bp (Fig. 6A), the distance between the guides we selected is 2,131 bases. While of course many important genes have long genomic spans, this simple analysis shows that numerous genes may be targetable for whole-CDS replacement. In addition, the insertion sizes demonstrated in this paper (3,925 bp) are much greater than typically amenable to knock-in via HITI (Suzuki et al., 2016), suggesting larger genes than Lrrtm2 may be targeted; our conservative estimate reflects the fact that we have not systematically explored the upper range of effective donor sizes, suggesting the actual number of suitable genes may in fact be much higher. Overall, the power of whole-CDS replacement will enable new research on endogenous protein localization, trafficking, and function in neurons, avoiding the constraints of overexpression and single-locus CRISPR editing techniques.
Applying whole-CDS replacement to other genes in Rattus norvegicus. A, Bioinformatics analysis of CDS spanning length within the Rattus norvegicus genome. Data procured from the Rnor 6.0 sequence via the UCSC Genome Browser. Results were restricted to coding sequences whose total CDS spanning length (absolute value of the difference between the start and stop codon positions) does not exceed 10,000 bases. Arrow shows the size of the LRRTM2 CDS spanning length, which covers a total of 1,882 bp from the start codon in Exon 1 to the stop codon in Exon 2. B, Cumulative frequency distribution of data in A, showing a considerable proportion of the sequences have a CDS spanning length less than that of LRRTM2 and therefore are potential targets for this method.
Discussion
In this work, we have demonstrated whole-CDS replacement in neurons and shown its power to simultaneously tag and mutate a protein at widely separated points of its sequence while maintaining native genetic regulation. The protein-coding portion of thousands of genes span lengths suitable for whole-CDS replacement, suggesting that this simple approach for total control over protein sequence will be a straightforward method for structure–function analysis in diverse systems. We used the approach here to identify new characteristics of expression and trafficking of the critical synaptic adhesion molecule LRRTM2. LRRTM2 levels in cultured hippocampal neurons were unexpectedly variable between neurons, and while its expression correlated with both PSD-95 and AMPAR content, the protein appeared absent from 20% of synapses. We also observed LRRTM2 outside of synapses at puncta that contained AMPARs but lacked PSD-95, suggesting a previously unappreciated role for LRRTM2 outside of PSD-95-containing synapses. Finally, utilizing our ability to manipulate the endogenous LRRTM2 sequence, we were able to increase synaptic LRRTM2 content without affecting AMPAR content, suggesting that this relationship may not be intrinsic or bidirectional.
Whole-CDS replacement is a general and flexible approach for simultaneous multisite manipulations
While CDS replacement strategies have been demonstrated in dividing cells (Allen et al., 2023), this work is to our knowledge the first demonstration of whole-CDS replacement in neurons. A major advantage of this method is the ability to make multiple, simultaneous modifications to a gene at disparate locations along it in a single editing step. In theory, other CRISPR technologies that allow for DNA replacement could be used for whole-CDS replacement, though likely TKIT, which we relied upon here, is the most generalizable approach. For example, CDS replacement has been achieved in dividing cells for a CDS spanning length of similar size to LRRTM2 (Allen et al., 2023). However, this relied on homology-directed DNA repair, which makes it intractable in postmitotic cell types such as neurons. This is due to the necessary reliance in postmitotic cells on nonhomologous end joining (NHEJ), a method of donor DNA incorporation that religates blunt ends of DNA together. PRIME editing, which utilizes neither nonhomologous end joining nor homology-directed repair (Anzalone et al., 2019), allows for short substitutions [∼100 bp (Anzalone et al., 2022)], but this is too small to replace the vast majority of genes. Designer exon approaches such as CRISPIE (Zhong et al., 2021) could also be used to essentially “knock-in” an engineered CDS in place of a gene's first exon and bypass the later exons, though this would eliminate key intronic regulation and alternative splicing. Finally, while making two separate edits using CRISPR is possible (Anzalone et al., 2022; Droogers et al., 2022), this remains extremely challenging and less efficient than whole-CDS replacement. Furthermore, there is as yet no path for expanding to more than two editing sites, whereas whole-CDS replacement offers near-infinite flexibility within the large donor sequence. Due to the permanent nature of genetic modifications, this whole-CDS replacement approach could simplify the process of generating multiple stem cell lines or animals with multiple genetic mutations in the same gene, as there would be no need to select and validate new guides for each modification that is introduced. Given these factors and the proven efficiency and accuracy of TKIT (Fang et al., 2021), we are confident it presents a flexible tool to manipulate entire coding sequences in diverse systems.
The limit on donor size for TKIT is not known, though previous research would suggest that large donor sizes have generally been a struggle for CRISPR in neurons and other models reliant on NHEJ. Previous studies suggest NHEJ may have greater difficulty with larger DNA sizes (Marks et al., 2021), though it is also more efficient (Suzuki et al., 2016). It is also conceivable that larger replacements may be facilitated by fusing repair machinery onto Cas9, which has been recently demonstrated to impact its efficiency (Richardson et al., 2023). Practically, donor size and efficiency likely depend in part on the genetic and epigenetic context as well as practical limits such as plasmid size or viral packaging and must therefore be determined on an individual basis. The former can be addressed with TKIT due to the flexibility of guide locations within the intronic/noncoding regions, which permits higher efficiency guides with lower off-target effects to be selected. We were able to efficiently knock-in a large donor sequence (3,925 bp, including IRES-Cre), and while we have demonstrated that a replacement of this size is feasible, replacement of larger regions may certainly also be possible. Furthermore, we have identified a sizeable group of genes whose CDS spanning length is smaller than this and thus appear amenable to whole-CDS replacement. Note that for many genes of interest, the difficulty of applying the approach will not be the size of the translated protein but the presence of introns so large as to be impractical to supply in a donor. In such cases, it may still be possible to adapt this technique to replace regions on one side or the other of a large intron, including multiple small, contiguous exons and introns, without modifying the entire CDS. This approach would largely preserve the regulation intrinsic to intronic sequences such as transcription and alternative splicing.
Whole-CDS replacement permits versatile tagging and reporter strategies
Ready identification of knock-in cells is key to experimental performance in many techniques, including microscopy and flow cytometry. However, knocked-in tags can be difficult to visualize because endogenous protein expression is often far lower than typical overexpression, and reliance on abundant target protein translation for tag detection is a potential source of bias. While empirical determination of the brightest protein tags can help improve utility, a knock-in conditional marker independent of the target protein's expression level is highly useful and essential in many cases. By integrating existing Cre-FLEx conditional marker technologies into noncoding regions simultaneously with our modifications to the LRRTM2 coding sequence, we were able to identify, tag, and manipulate knock-in cells in a single step. Importantly, this strategy has previously only been available for C-terminal knock-ins or N-terminal using a Cre-P2A sequence (Gao et al., 2019; Willems et al., 2020), whereas whole-CDS replacement allows combining this manipulation with other mutations throughout the CDS. The FLEx system is highly adaptable, and the cell fills we utilized here can be exchanged for any genetically encoded protein marker, sensor, or optogenetic tool desired. This combination of powerful tools in an efficient CRISPR environment should permit elegant, high-throughput studies of protein function in their cellular context in the future.
Endogenous LRRTM2 expression varies across and within neurons
One surprising result from this study was the cellular variability of LRRTM2 expression even within the relatively restricted range of neuron types found in hippocampal culture. LRRTM2 mRNA levels have been found to vary between cell types in the neocortex and hippocampus (Földy et al., 2016; Paul et al., 2017). The correlation in single cells between mRNA and protein levels is frequently poor, yet analysis of population protein expression levels at single-cell resolution is challenging. Here, sparse knock-in allowed systematic evaluation of single-neuron expression as well as the measurement of expression levels of mutant protein. Note that such analysis is aided substantially by the presence of a knock-in dependent marker, which we expect will be particularly useful for analysis in vivo where the range of expression levels may be even greater. The mechanisms of cell-specific LRRTM2 expression level remain unknown, but one attractive possibility is that the activity history of the neuron drives the regulation of LRRTM2 as the number of AMPARs and synapses is modulated.
Our data have shown conclusively that LRRTM2 is present at higher levels in larger synapses. LRRTM2 binds directly with PSD-95 via its PDZ-binding motif (de Wit et al., 2009; Linhoff et al., 2009), and endogenous LRRTM2 was indeed most abundant at synapses with more PSD-95 content. Furthermore, we were able to increase the synaptic content of LRRTM2 via the YACA mutations. Notably, this increase in synaptic content was smaller than expected from mutant overexpression (Liouta et al., 2021), highlighting the importance of manipulating LRRTM2 trafficking, and protein trafficking more generally, with endogenous regulations intact. While the mechanisms by which the YACA mutations influence endogenous LRRTM2 trafficking are unclear, the intracellular location of the YACA mutations suggests intracellular interactions may be involved in trafficking LRRTM2, including possibly PSD-95 whose interaction domain is close by. However, we also found that LRRTM2 was not present at nearly 20% of PSD-95-containing synapses and was also present at dendritic locations lacking PSD-95, suggesting there are additional mechanisms beyond binding to PSD-95 responsible for LRRTM2 localization. Several overexpression studies have found that removing the PDZ-binding motif does not affect the ability of LRRTM2 to traffic to synapses or facilitate LTP (Linhoff et al., 2009; Bhouri et al., 2018), suggesting it is not exclusively responsible for LRRTM2 trafficking. Furthermore, it appears that LRRTM2 is trafficked to extrasynaptic puncta that frequently contain AMPARs. What would prompt the enrichment of LRRTM2 at these points is unclear. The neurexin-binding domain of LRRTM2 is required for rescuing the effects of removing LRRTM1/2 on LTP via overexpression (Soler-Llavina et al., 2013; Bhouri et al., 2018), but it is unclear whether neurexins, or indeed any presynaptic scaffold, would be present outside of PSD-95 and/or Bassoon-demarcated synapses. One hypothesis is that these LRRTM2 puncta are indeed synapses but lack PSD-95 and contain other synaptic scaffolds whose interaction with LRRTM2 has not been investigated. Given the homology in PDZ motifs across synaptic scaffolds, it is certainly likely that the LRRTM2 PDZ-binding motif also interacts with other scaffolds such as SAP-102 or PSD-93 (Lim et al., 2002). However, the presence of LRRTM2 puncta outside of synapses regardless of PSD-95 or Bassoon labeling raises the possibility that these puncta represent a diffusible pool of LRRTM2 that can be recruited to the synapse as needed, much the way that AMPARs are (Penn et al., 2017; Choquet, 2018). This model would include two pools of LRRTM2 protein, one synaptic and one extrasynaptic, where the loss of neurexin binding (specifically) reduces the size or functionality of the synaptic pool.
LRRTM2 and its sister protein LRRTM1 share high sequence homology and are expressed in the same broad regions of the hippocampus (Lein et al., 2007), and many studies have relied on their simultaneous deletion (Soler-Llavina et al., 2013; Bhouri et al., 2018). This is done because CAMs frequently compensate for one another, and dual knock-outs can better illustrate their roles than either knock-out alone, as has been recently shown for CAMs LRRTM1 and SynCAM1 in hippocampal synapses (de Arce et al., 2023). Interestingly, the effects of dual knock-out can be rescued with overexpression of LRRTM2 alone (Soler-Llavina et al., 2013; Bhouri et al., 2018). We found endogenous LRRTM2 in 80% of synapses, which raises the possibility that the remaining 20% may be LRRTM1-positive. This is potentially an interesting case of differential subcellular trafficking of highly homologous proteins. A more extreme segregation between members of a single family of CAMs has been observed previously in vivo; for example, the neuregulin proteins 1 and 3 are differentially sorted to the somatic and axonal domains of pyramidal neurons, respectively (Exposito-Alonso et al., 2020). LRRTM1 and 2 are spatially segregated within the CA1 hippocampus, with LRRTM1 largely localized to the strata radiatum and oriens layers and LRRTM2 localized to the stratum lacunosum moleculare layer (Linhoff et al., 2009; Schroeder et al., 2018; Nozawa et al., 2022). Whether this segregation is input-specific or cell-intrinsic, for instance, as a scaling of expression dependent on the synaptic position along the dendrite, has not been tested. Our knock-in of LRRTM2 shows the protein trafficking to both proximal and distal synapses, suggesting the in vivo pattern may reflect the engagement of specific LRRTM subtypes based on presynaptic factors. The functional implications of such segregation are unknown at this point, but given that many studies have relied on their simultaneous deletion, future work is needed to tease out their individual roles. While it is currently extremely challenging to simultaneously target multiple genes via CRISPR, future developments could make it possible to tag and manipulate both LRRTM1 and LRRTM2 in the same cells to further investigate their relative trafficking.
Relationship of endogenous LRRTM2 with AMPARs
Numerous previous studies have established that LRRTM2 controls synaptic AMPAR content (de Wit et al., 2009; Soler-Llavina et al., 2013; Bhouri et al., 2018; Ramsey et al., 2021). One possibility is that changing LRRTM2 levels would be determinative in establishing AMPAR levels. Our data show that synapses lacking LRRTM2 are smaller and contain fewer AMPARs than those with LRRTM2, though the latter correlation was weaker than predicted by this hypothesis. However, when we increased the content of LRRTM2 with the YACA mutations, there did not appear to be a similar increase in synaptic AMPAR content. Several possible explanations exist for this finding. One possibility is that while YACA mutations may increase LRRTM2 surface content by altering membrane-targeting trafficking mechanisms (Minatohara et al., 2015), they also block the as-yet-unknown mechanisms by which LRRTM2 anchors AMPARs at the surface. However, given that previous studies have removed the entire C-terminal domain of LRRTM2 in a replacement context with no effect on LTP (Soler-Llavina et al., 2013; Bhouri et al., 2018), this appears unlikely. Another possibility is that the YACA mutations, which lead to a cell-wide increase in LRRTM2 content, may not significantly alter AMPAR trafficking at the subcellular level due to potential compensatory mechanisms. Instead, it's plausible that AMPAR trafficking could be influenced by localized increases in LRRTM2 content on a synapse-by-synapse basis. Furthermore, it is possible that the mechanism connecting AMPAR enrichment to LRRTM2 levels is not sensitive enough to respond to the small increase that the YACA mutations induce in LRRTM2 content. Unfortunately, these latter two possibilities are difficult to test without a greater understanding of what induces LRRTM2 trafficking to synapses so that it can be manipulated more directly and in a physiologically relevant manner. Finally, it is important to note that we measured the LRRTM2-YACA mutation at a static baseline state, whereas it is possible that the role of the domain containing the YACA mutants is engaged primarily during synaptic potentiation.
Previous studies have established that surface diffusion of AMPARs to synapses is critical during plasticity (Penn et al., 2017; Choquet, 2018). Our data have shown that LRRTM2 and AMPARs localize to puncta outside of synapses and appear together more frequently than either protein alone. This suggests the exciting possibility that LRRTM2 may traffic together with or potentially shepherd extrasynaptic AMPARs as they are trafficked to synapses during plasticity. This would give LRRTM2 the ability to retain new AMPARs at synapses via its interaction with presynaptic neurexins in a timely fashion. If LRRTM2 recruits AMPARs to the synapse via this surface trafficking mechanism, it would explain why a whole-cell increase in LRRTM2 content, as induced via the YACA mutation, is insufficient to induce a similar increase in synaptic AMPAR content. LRRTM2 facilitates synaptogenesis (de Wit et al., 2009), and extrasynaptic LRRTM2-AMPAR puncta thus also suggest a means by which LRRTM2 could enable AMPAR trafficking to new synapses. The presence of LRRTM2 clusters outside of PSD-95 synapses, and their potential role in local AMPAR trafficking, represents an intriguing new development in our understanding of LRRTM2 function.
Together, our approach for whole-CDS replacement facilitates labeling, imaging, and manipulating endogenous proteins, including LRRTM2, and represents a potent methodological advance in the field of cellular neurobiology. While we have illustrated many new findings regarding the endogenous trafficking of LRRTM2, the approach of whole-CDS replacement will assist in determining the mechanisms by which this cell adhesion molecule controls such critical processes as synaptic AMPAR retention and plasticity.
Footnotes
This work was supported by the National Institute of Mental Health (S.L.P., F32MH130106; M.C.A., F31MH12428330; T.A.B., R37MH080046, R01MH119826).
The authors declare no competing interests.
- Correspondence should be addressed to Thomas A. Blanpied at tblanpied{at}som.umaryland.edu.
