ABSTRACT
GFP labeling by genome editing can reveal the authentic location of a native protein but is frequently hampered by weak GFP signals and broad expression across a range cell types in multicellular animals. To overcome these problems, we engineered a Native And Tissue-specific Fluorescence (NATF) strategy which combines CRISPR/Cas-9 and split-GFP to yield bright, cell-specific protein labeling. We use CRISPR/Cas9 to insert a tandem array of seven copies of the GFP11 β-strand (gfp11x7) at the genomic locus of each target protein. The resultant gfp11x7 knock-in strain is then crossed with separate reporter lines that express the complementing split-GFP fragment (gfp1-10) in specific cell types thus affording tissue-specific labeling of the target protein at its native level. We show that NATF reveals the otherwise undetectable intracellular location of the immunoglobulin protein, OIG-1, and demarcates a receptor auxiliary protein LEV-10 at cell-specific synaptic domains in the C. elegans nervous system.
INTRODUCTION
Reliable localization of a given protein can provide useful clues to its mechanism of action. One way to achieve this goal is to label the protein of interest with tags, such as fluorescent proteins (e.g., GFP)1,2 or small peptides (e.g. FLAG, HA)3. Because tagged proteins are typically expressed with heterologous promoters or from multicopy transgenic arrays, this approach can result in misleading signals due to over-expression4. This problem can be obviated by using CRISPR-Cas9 for single copy labeling of the native protein5,6, but this genome editing strategy suffers from two additional limitations. First, the endogenous expression level of a target protein may be too low for detection. Second, the protein of interest may be expressed in several tissues thus preventing a clear delineation of cell-specific localization in multicellular organisms. Here we describe an experimental approach, NATF (Native And Tissue-specific Fluorescence or “Native”) that exploits a combinatorial strategy to resolve both of these problems.
Our approach relies on the finding that the barrel-like GFP structure can be reconstituted by the spontaneous interaction of two separate GFP peptides derived from the highly stable GFP variant, superfolder GFP (sfGFP). The larger of these fragments is comprised of the first 10 β-strands (GFP1-10). Its smaller complement, a short, 16 amino acid sequence, contains the eleventh β-strand (GFP11). Neither GFP1-10 nor GFP11 fluoresce independently but a strong signal is restored in the hybrid split-GFP that they reconstitute7. Thus, to enhance the GFP signal, a target protein can be tagged with multiple copies of the short GFP11 peptide and then co-expressed with excess GFP1-108,9. In addition to labeling the native protein with smaller covalent tags, this combinatorial approach offers the further benefit of limiting the GFP signal to the specific cell type in which GFP1-10 is expressed (Figure 1a).
In this report, we describe a NATF toolbox that combines split-GFP and CRISPR technology for live-cell imaging of labeled C. elegans proteins expressed at native levels. With this approach, a GFP11 multicopy DNA array (GFP11X7) is inserted into the target gene. The resultant knock-in line can then be crossed with separate reporter lines in which GFP1-10 is expressed in different cell types for tissue-specific visualization of the reconstituted NATF GFP (Figure 1a). We utilized this strategy to demonstrate effective enhancement of an otherwise weak signal from single copy fluorescent protein labeling of a key protein (OIG-1) in the C. elegans nervous system as well as the cell-specific resolution of a receptor accessory protein (LEV-10) at closely spaced but functionally distinct synapses.
RESULTS
Tool box and strategy for NATF GFP labeling
We used a previously described CRISPR/Cas-9 system for genome editing in C. elegans10. In this approach, homology arms flank a self-excising cassette that carries positive selection markers (sqt-1) for identifying transgenic worms (“rollers”) and drug resistance (hygR) for detecting CRISPR/Cas9-induced integrants10. A brief heat shock treatment induces excision of the marker cassette to restore wild-type movement (“non-roller”) (Figure 1b). For split-GFP experiments, we replaced the fluorescent protein sequence in the original repair template plasmid with a gfp11x7 insert8. Homology arms of ~500bps were used for the two genes targeted (oig-1 and lev-10) in this study (Figure S1a). We also constructed plasmids for expressing GFP1-10 in specific cell types including body muscles, all neurons, cholinergic neurons and GABAergic neurons (Figure 1b and Figure S1b). In these lines, the GFP1-10 transgenes are carried as extrachromosomal arrays that are maintained by selecting for pharyngeal coninjection marker (Pmyo-2::mCherry) (Figure 1b). Cell-specific drivers are flanked with multiple cloning sites to facilitate construction of plasmids for GFP1-10 expression in other tissues (Figure S1b). gfp11x7 knock-in strains can be confirmed within two weeks of the initial injection and then crossed with GFP1-10 expressing lines for characterization (Figure 1b).
NATF GFP labeling reveals the intracellular localization of OIG-1 in GABAergic motor neurons
oig-1 encodes a soluble protein with a single immunoglobulin domain (Figure 2b) that is temporally regulated in GABAergic motor neurons to antagonize a synaptic remodeling program; in oig-1 mutants, a postsynaptic acetylcholine receptor (AChR) containing the AChR subunit, ACR-12::GFP, is ectopically relocated from dorsal to ventral GABAergic neuron processes. OIG-1 is secreted when over-expressed from multicopy transgenic arrays to produce bright puncta adjacent to clusters of ACR-12::GFP (Figure 2a,b)11,12. To ask if OIG-1 is also secreted when expressed from the native locus, we used CRISPR/Cas-9 to engineer a single copy knock-in of the red fluorescent protein, TagRFP together with a 3XFLAG epitope tag (Figures 2b, S2a-c). We used immunoblotting to confirm expression of TagRFP::3XFLAG::OIG-1 (Figure 2c) but failed to detect TagRFP fluorescence in vivo (Figure 2d-g). To produce a potentially brighter signal, we created a gfp11x7::oig-1 knock-in (Figure 2b) with a sgRNA that targeted the same 5’-N18GGNGG site used for the TagRFP insert13. Successful knock-in of gfp11x7 was confirmed by sequencing (data not shown). The resultant GFP11X7::OIG-1 fusion protein is likely functional as ACR-12::GFP puncta show robust localization to GABA neuron processes in both dorsal and ventral nerve cords as observed in wild type (Figure S2 b-d). We then crossed the gfp11x7::oig-1 knock-in with a pan-neural Prab-3::gfp1-10 transgenic line. Consistent with our previous findings, the OIG-1 signal can be detected in head neurons and in both dorsal and ventral nerve cords (Figure 2h-k)11. Co-localization of the OIG-1 NATF GFP signal with the pan-neural marker Prab-3::NLS::mCherry confirmed OIG-1 expression in neurons (Figure 2l). As an independent strategy to validate OIG-1 expression in GABA neurons, we crossed the gfp11x7::oig-1 line with Pttr-39::gfp1-10 which is selectively expressed in DD and VD class GABAergic motor neurons14. In this case, the OIG-1 NATF GFP signal is limited to VD neurons with either weak or undetectable expression in DD neurons at the L4 larval stage (Figure 2l). This finding confirms previous results obtained with a Poig-1::gfp transcriptional reporter that was expressed in VD but not DD neurons after the L2 larval stage11. Because the GFP1-10 peptide is expressed intracellularly in these strains, the NATF GFP signal likely derives from cytoplasmic OIG-1. Notably, the OIG-1 NATF GFP signal is diffusely visible throughout VD neuron soma and neurites (Figure 2h-m) and does not show the distinctive punctate appearance of over-expressed OIG-1 from a multicopy array (Figure 2a, b). In addition, we failed to detect extracellular NATF GFP when the gfp11x7::oig-1 knock-in was crossed with a transgenic line in which the GFP1-10 peptide is secreted from neurons (Prab-3::ss::gfp1-10) (Figure S3a). As a positive control, we showed that the secreted form of GFP1-10 in the Prab-3::ss::gfp1-10 strain is functional because it robustly labels a GFP11 peptide fused to the extracellular domain of the synaptic membrane protein, NLG-115 (Figure S3b-d). Thus, we conclude that OIG-1 is not secreted when expressed at the native level but localizes intracellularly (manuscript in preparation). This surprising finding critically depended on the use of the NATF strategy to detect low levels of native OIG-1 expression and thereby circumvent artifactual localization due to OIG-1 over-expression from multicopy arrays.
NATF GFP labeling reveals discrete locations for the CUB domain protein LEV-10 in different cell types
Having shown that NATF could detect a soluble protein (OIG-1), we next targeted, LEV-10, a CUB domain transmembrane protein that clusters AChRs at postsynaptic sites in body muscles16. First, we created a CRISPR/Cas9 knock-in line in which a single copy of GFP was fused to the intracellular C-terminus of LEV-10 (see Figure 4a). We detected LEV-10::GFP in both ventral and dorsal nerve cords as predicted for a protein that localizes to body muscle synapses16. LEV-10::GFP puncta were also detected in the head region where motor neurons synapse with body muscles on the inside surface of the nerve ring17,18 (Figure 3a). For NATF GFP labeling of body muscle synapses, we generated a lev-10::gfp11x7 knock-in and crossed it with a muscle-specific transgenic line expressing GFP1-10 (Pmyo-3::gfp1-10). The LEV-10 NATF GFP signal in the head region and axial nerve cords (Figure 3b) mimics that of the single copy lev-10::gfp knock-in but is noticeably brighter (Figure 3a). We quantified the GFP signal for each marker at the nerve ring muscle synapses to confirm that the LEV-10 NATF fluorescence is ~3X brighter than the GFP signal from the lev-10::gfp single copy insertion8 (Figure 3c). In addition to determining that the lev-10::gfp11x7 array yields a stronger signal than that of the single copy lev-10::GFP insert, we also showed that NATF GFP is more resistant to photobleaching as previously demonstrated for reconstituted split-GFP from measurements in vitro8 (Figure 3d).
In addition to expression in muscle, our independent studies have shown that LEV-10 is also expressed in ventral cord neurons where it co-localizes with AChRs at postsynaptic sites in GABAergic motor neurons (manuscript in preparation). In the motor neuron circuit, cholinergic motor neurons form dyadic synapses that innervate closely spaced postsynaptic domains in body muscle and GABA neurons (see Figure 4g)17. Both of these postsynaptic regions in the ventral nerve cord region should be labeled in the lev-10::gfp knock-in and thus, cannot be unambiguously identified (Figure 3a). To resolve this problem, we crossed the lev-10::gfp11x7 knock-in with transgenic lines that express GFP1-10 in either body muscles (Pmyo-3::gfp1-10) or in DD and VD GABAergic motor neurons (Pttr-39::gfp1-10). NATF GFP puncta can be readily detected in both cases (Figures 4c,d) but are brighter in muscles than in GABAergic neurons (data not shown). Expression of a TagRFP-labeled AChR subunit, UNC-29, in muscle confirms co-localization of UNC-29::TagRFP and LEV-10 NATF GFP (Figure 4c). Expression of GFP1-10 in DD and VD neurons produces LEV-10 NATF GFP puncta that overlap with a cytoplasmic GABA neuron mCherry marker (Punc-47::mCherry) as predicted for LEV-10 protein that localizes to GABA neuron synapses (Figure 4d). To confirm the postsynaptic location of LEV-10 in GABA neurons, we used a DD-specific construct (Pflp-13::gfp1-10) to generate a LEV-10 NATF GFP signal. In this case, super-resolution imaging resolves distinct LEV-10 NATF GFP puncta at the tips of postsynaptic spine-like projections that have been recently described in the ventral processes of mature DD neurons (Figure 4e and Figure S4)19. Notably, we have also observed that the AChR marker, ACR-12::GFP, is positioned in the same distal location in DD dendritic spines and that these spines are aligned with presynaptic cholinergic vesicles (manuscript in preparation)19. In addition to resolving LEV-10 localization at distinct postsynaptic locations in muscle vs GABA neurons, we also used a cholinergic motor neuron driver (Pacr-2::gfp1-10) to detect a separate LEV-10 NATF signal in ventral cord cholinergic neurons. In this case, LEV-10 NATF GFP is diffuse (Figure 4f) and asymmetrically localized to the ventral but not dorsal nerve cord (data not shown), a labeling pattern that closely resembles the perisynaptic position of the AChR subunit ACR-12::GFP in cholinergic motor neurons20. Because LEV-10 is expressed at its native level and retains its AChR clustering function (data not shown) when fused to the GFP11X7 adduct, it seems likely that each of the three distinct, cell-specific LEV-10 NATF signals that we have detected with our combinatorial approach (i.e., muscle, GABA neurons, cholinergic neurons) marks authentic subcellular locations for the endogenously expressed LEV-10 protein.
DISCUSSION
Although our results have determined that fusion with the GFP11X7 domain does not disrupt either OIG-1 or LEV-10 activity, other proteins may be less tolerant. In that event, smaller adducts with fewer copies of GFP11 could be attempted21. In that case, GFP signal augmentation will be diminished but tissue-specific labeling is still possible. Because the gfp11×7 insert is stably integrated at the native locus and is thus limiting, the complementing GFP1-10 peptide can be provided from multicopy transgenic arrays without risk of inducing over-expression artifacts. Thus, a given GFP11X7 split GFP insert can be rapidly tested with multiple tissue specific GFP1-10 transgenic lines which can be readily generated using conventional methods. A similar combinatorial approach should also be useful for tissue-specific protein labeling in other model organisms. We note that NATF can be modified for multicolor split-GFP imaging with cyan (CFP) and yellow (YFP) GFP variants or with the sfmCherry marker11,12.
AUTHOR CONTRIBUTIONS
SH, ACC and DMM conceived the project. SH and ACC performed the experiments and analyzed the data. SH, ACC and DMM wrote the manuscript.
COMPETING INTERESTS
The authors declare no competing interests.
MATERIALS AND CORRESPONDENCE
David M. Miller, Ph.D, david.miller{at}vanderbilt.edu
Methods
C. elegans strains
C. elegans strains were maintained at room temperature on NGM plates seeded with OP50 as previously described22. Some strains were obtained from the Caenorhabditis Genetics Center (CGC). The N2 Bristol strain was used as the wild-type reference. Transgenic animals were generated using standard microinjection techniques. Unless noted otherwise, 100 ng/μL total DNA injection samples were prepared with the pBluescript plasmid, as carrier. The strains used in this study are described in Supplementary Table 1.
Molecular biology
sgRNA/Cas-9 plasmid design
A 200bp DNA sequence that contains the desired cut site was submitted to opitimized CRISPR Design online tool (http://crispr.mit.edu/) to predict sgRNA sequences. To enhance gene editing efficiency, we selected a 5’ N18GGNGG sequence13 as a sgRNA targeting site for both oig-1 and lev-10. For oig-1, GGAGAGAAAGACGAAAATGG was cloned into pDD162(Addgene #47549), a plasmid that contains the sgRNA backbone and Cas-9 expression system using Q5 site directed mutagenesis (NEB). Similarly, for lev-10, ACGAATCGACTGGTGGCCGG was used as the sgRNA binding sequence, which is ~80bp upstream of the lev-10 stop codon.
CRISPR repair template for oig-1 and lev-10
To create the SEC repair template for oig-1 TagRFP CRISPR knock-in, flanking ~500bp genomic DNA regions immediately upstream and downstream of the desired insertion site were amplified by PCR using the following primers (Primer 1 and Primer 2 for upstream homology arm and Primer 3 and Primer 4 for downstream homology arm) with overlap regions to the target plasmid pDD284(Addgene #66825):
OIG-1 Primer 1: 5’ gacgttgtaaaacgacggccagtcgacctaaccattccaaaagat
OIG-1 Primer 2: 5’ tgagctcctctcccttggagaccatcgcatttattccaactgata
OIG-1 Primer 3: 5’ ttacaaggatgacgatgacaagagaaaatcttcgcatatagaaga
OIG-1 Primer 4: 5’ caggaaacagctatgaccatgttatccaagtcggagtactgttca
The amplified DNA fragments were cloned into plasmid pDD284 using Gibson cloning (NEB) to create the repair template. The corresponding PAM sequence in the repair template plasmid was mutated from AGG to CCC using Q5 site-directed mutagenesis to produce the final plasmid, pSH30. Correct insertions and mutations were confirmed by sequencing.
To create the SEC repair template (pSH55) for the oig-1 GFP11x7 CRISPR knock-in, the GFP11x7 coding sequence was amplified from a previously published plasmid (Addgene #60910) and inserted into pSH30 to replace the TagRFP sequence by In-Fusion cloning (Takara) with the following primers.
Fragment.FOR 221 5’ gttggaataaatgcgatgcgtgaccacatggtcctt
Fragment.REV 222 5’ aaagtacagattctcggtgataccggcagcat
Vector.FOR 223 5’ gagaatctgtactttcaatccggaaaggtaag
Vector.REV 224 5’ cgcatttattccaactgatagaaagcataaaagtagt
To create the GFP and GFP11x7 knock-in repair template for LEV-10, we designed a two-step In-Fusion cloning method. ~500bp of DNA sequences upstream and downstream of the lev-10 stop codon were selected for flanking homology arms. DNA was amplified and then sequentially cloned into pSH30 or pSH55 to replace the original oig-1 homology arms. The resultant plasmids were then used as templates for site-direct mutagenesis to create the final repair template plasmid with sgRNA binding sequences mutated, pSH84(GFP knock-in) and pSH85(GFP11x7 knock-in). The primers for these cloning steps were designed using the same strategy as described above. Primer sequences are available on request.
GFP1-10 reporter plasmids
The DNA sequence of GFP1-10 was amplified from pcDNA3.1-GFP1-10 (Addgene #70219) and cloned into pGH8 (Prab-3:: mCherry) using infusion cloning to create pSP1(Prab-3::gfp1-10). The DD and VD GABAergic neuron-specific promoter, Pttr-39, the cholinergic specific promoter, Pacr-2, and muscle-specific Pmyo-3 promoter were amplified to replace the Prab-3 promoter in pSP1 to create pSH79(Pttr-39::gfp1-10), pSH88(Pacr-2::gfp1-10), pSH86(Pmyo-3::gfp1-10) and pSH87(Pflp-13::gfp1-10). To create a secreted GFP1-10 construct, infusion cloning was used to add the first 114bp of the oig-1 sequence including the signal sequence11 prior to the start codon of GFP1-10 in pSP1. The combined sequence was analyzed using SignalP 4.1 Server to confirm that the predicted signal peptide was intact. The final plasmid, pSH69 (Prab-3::ssGFP1-10), was confirmed by sequencing.
Confocal microscopy and image processing
Images of fluorescently-labeled worms were captured at room temperature in live C. elegans using a Nikon A1R confocal microscope. Nematodes were immobilized with 15mM levamisole/0.05% tricaine on a 2% agarose pad in M9 buffer. All images for ACR-12::GFP fluorescence quantification were obtained with the same settings using the 40X oil, NA 1.3 objective and Nyquist collection. Constant laser power was used to compare the LEV-10::GFP fluorescence intensity to that of the NATF GPF signal produced by the combination of LEV-10::GFP11x7 with Pmyo-3::GFP1-10. Images in Figure 4 were 3D-deconvolved with NIS-Elements with Automatic algorithm. For other images, ND2 files generated with NIS-Elements were imported into Fiji for analysis. Maximum intensity projections were generated by selecting stacks that have both ventral and dorsal signals. The mean fluorescence intensity of each animal after subtracting background was used for statistical analysis. To measure the stability of GFP signal in lev-10::gfp and lev-10::gfp11x7; Pmyo-3::GFP1-10 lines, a region of interest (ROI) of the same size was bleached with a 405nm laser for 15 seconds at 50% laser power. Images of the ROI were collected and compared before and after photo bleaching.
Airy Scan Imaging
Worms were mounted on 10% agarose pads and immobilized with 15mM levamisole/0.05% tricaine dissolved in M9. A Zeiss LSM880 microscope equipped with an AiryScan detector and a 63X/1.40 Plan-Apochromat oil objective lens was used to acquire super resolution images of the DD neuron (Figure 4e). Images were acquired as a Z-stack (0.19μm/step), spanning the total volume of DD neuron and submitted for AiryScan image processing using ZEN software.
Statistical Analysis
For all experiments, sample numbers were n >10. Student’s t-test was used for comparison between two groups. P < 0.05 was considered significant. Prism 6 was used for statistical analysis.
ACKNOWLEDGEMENTS
We thank members of the Miller lab for critical reading of the manuscript, Sierra Palumbos and Alice Siqi Chen for plasmid construction, Lakshmi Sundararajan for help with confocal imaging and Oliver Hobert for sharing strains. Some nematode strains used in this work were provided by the Caenorhabditis Genetics Center, which is funded by the NIH National Center for Research Resources (NCRR). Super-resolution imaging was acquired at the Vanderbilt Cell Imaging Shared Resource (1S10OD201630-01). This work was supported by National Institutes of Health grants to DMM (R01NS081259 and R01NS106951). ACC is supported by an AHA predoctoral fellowship (18PRE33960581).