Split-wrmScarlet and split-sfGFP: tools for faster, easier fluorescent labeling of endogenous proteins in Caenorhabditis elegans

Abstract We create and share a new red fluorophore, along with a set of strains, reagents and protocols, to make it faster and easier to label endogenous Caenorhabditis elegans proteins with fluorescent tags. CRISPR-mediated fluorescent labeling of C. elegans proteins is an invaluable tool, but it is much more difficult to insert fluorophore-size DNA segments than it is to make small gene edits. In principle, high-affinity asymmetrically split fluorescent proteins solve this problem in C. elegans: the small fragment can quickly and easily be fused to almost any protein of interest, and can be detected wherever the large fragment is expressed and complemented. However, there is currently only one available strain stably expressing the large fragment of a split fluorescent protein, restricting this solution to a single tissue (the germline) in the highly autofluorescent green channel. No available C. elegans lines express unbound large fragments of split red fluorescent proteins, and even state-of-the-art split red fluorescent proteins are dim compared to the canonical split-sfGFP protein. In this study, we engineer a bright, high-affinity new split red fluorophore, split-wrmScarlet. We generate transgenic C. elegans lines to allow easy single-color labeling in muscle or germline cells and dual-color labeling in somatic cells. We also describe a novel expression strategy for the germline, where traditional expression strategies struggle. We validate these strains by targeting split-wrmScarlet to several genes whose products label distinct organelles, and we provide a protocol for easy, cloning-free CRISPR/Cas9 editing. As the collection of split-FP strains for labeling in different tissues or organelles expands, we will post updates at doi.org/10.5281/zenodo.3993663


Introduction
Genetically expressed fluorophores are essential tools for visualizing and quantifying cellular proteins. In Caenorhabditis elegans, fluorescent proteins have traditionally been introduced on extrachromosomal arrays (Kimble et al. 1982;Mello et al. 1991) or via MosSCI-based integration (Frøkjaer-Jensen et al. 2008, 2012. These methods have enabled important discoveries but can also lead to artifacts due to supraphysiological gene-expression levels and lack of endogenous regulatory control. In recent years, the repertoire of C. elegans transgenic tools has expanded (see Nance and Frøkjaer-Jensen 2019, for review), particularly due to advances in CRISPR/Cas9 genome-editing technologies (Paix et al. 2014;Dickinson and Goldstein 2016). CRISPR/Cas9 allows precise transgene insertion by homology-directed repair (HDR) and can be used to label an endogenous gene at its native locus with a fluorescent protein (Friedland et al. 2013;Dokshin et al. 2018;Farboud et al. 2019;Vicencio et al. 2019).
However, relative to CRISPR/Cas9-mediated integration of smaller transgenes, genomic insertion of large DNA fragments like those encoding fluorescent proteins remains a challenge, both because repair with double-stranded templates is less efficient than repair with single-stranded oligodeoxynucleotide donors (ssODN) (Farboud et al. 2019), and because of the requirement for cloning to prepare the HDR donor template. Recent methods such as "hybrid" (Dokshin et al. 2018) and "nested" (Vicencio et al. 2019) CRISPR remove the need for cloning but still require preparation of the DNA template or several rounds of injections and selection of transgenic progeny. As a result, using CRISPR with small ssODN templates is currently faster, easier, cheaper, and more efficient than with large templates. In our lab, we routinely make C. elegans genome edits with short ssODNs with almost guaranteed success. In contrast, in our experience, large edits using double-stranded DNA templates have higher failure rates and are more time-consuming.
Our preferred approach is to combine the utility of full-length fluorescent proteins with the convenience of short genomic edits, by using high-affinity asymmetrically split fluorescent proteins (Cabantous et al. 2005). These fluorophores typically separate a GFP-like protein between the 10th and 11th strands of the beta barrel, splitting it asymmetrically into a large (FP 1-10 ) and a small (FP 11 ) fragment. The fragments are not individually fluorescent, but upon binding one another, recapitulate the fluorescent properties of an intact fluorophore ( Figure 1A). Unlike the low-affinity split fluorescent proteins used in BiFC assays (Hu et al. 2002), high-affinity binding between the fragments is critical here. Our preferred approach for tagging a new cellular protein begins with a C. elegans strain expressing the large FP 1-10 fragment in cells of interest, unattached to any cellular protein. This way, only the small FP 11 fragment (<72 nt) needs to be inserted to tag the target protein, which will only fluoresce in compartments where it can bind the large fragment. These short insertions tend to be faster, easier, and more reliable than inserting a > 600 nt full-length fluorescent protein (Paix et al. 2015;Prior et al. 2017, Dokshin et al. 2018Richardson et al. 2018). Therefore, collections of C. elegans lines stably expressing the large FP 1-10 in different tissues are an invaluable resource allowing rapid fluorescent tagging in a cell type of choice. Stable lines with red FP 1-10 fragments would be especially useful, given C. elegans' substantial autofluorescence in the GFP channel.
Green and red asymmetrically split fluorescent proteins have been used to combine cell and protein specificity in C. elegans neurons and synapses (Noma et al. 2017;Feng et al. 2019;He et al. 2019); however, these strains used extrachromosomal arrays, not stable lines, which are more time-consuming to maintain and can have variable expression levels. To the best of our knowledge, there is only one available unbound FP 1-10 stable C. elegans line, which expresses sfGFP 1-10 in the germline (Hefel and Smolikove 2019), and there are no available lines with red FP 1-10 fragments. Existing red split fluorophores are also much dimmer in C. elegans than green ones, despite recent improvements like split-sfCherry3 (Feng et al. 2019). In addition, we often struggle to express genome-integrated full-length fluorescent protein fusions in the germline, potentially due to generational silencing.
Here, we describe tools that reduce these obstacles for convenient fluorescent labeling of endogenous C. elegans proteins. We engineer split-wrmScarlet, a new split red fluorescent protein based on mScarlet (Bindels et al. 2017;El Mouridi et al. 2017), which is three times brighter in worms than split-sfCherry3 (https://www.addgene.org/138966). We generate and share C. elegans lines carrying single-copy insertions of split-wrmScarlet 1-10 expressed broadly in somatic cells (https://cgc.umn.edu/strain/ CF4582), and specifically in muscle (https://cgc.umn.edu/strain/ CF4610). We also describe a novel approach to make C. elegans lines with robust germline expression of exogenous proteins that appears to be resistant to generational silencing. We use this approach to make a germline-specific split-wrmScarlet 1-10 strain (https://cgc.umn.edu/strain/DUP237).  Figure 1 Engineering and evaluating split-wrmScarlet. (A) Principle of endogenous protein labeling with split-wrmScarlet. The protein structure from split-wrmScarlet was generated using Phyre2 and PyMOL. (B) Schematic of the plasmids encoding split-wrmScarlet and split-sfCherry3. Each plasmid consists of the large FP 1-10 sequence fused to mNeonGreen, and the corresponding small FP 11 sequence fused to mTagBFP2. The T2A sequence ensures that mTagBFP2::FP 11 and the corresponding mNeonGreen::FP 1-10 are separated. The images are representative displays of the ratio of red to green fluorescence intensity from images acquired under identical conditions after background subtraction and masking with the same threshold. Scale bar, 50 mm. (C) Emission intensities from split-sfCherry3 and split-wrmScarlet normalized to mNeonGreen. Mean 6 SD. Circles are individuals (n ¼ 6 for each split fluorescent protein). ****P < 0.0001.
We provide a protocol for an easy, cloning-free method to label endogenous genes with FP 11 s using CRISPR/Cas9, commercially available synthetic ssODN, and microinjection (dx.doi.org/ 10.17504/protocols.io.bamkic4w). We validate this protocol by targeting split-wrmScarlet 11 to six different genes whose products have distinct cellular locations. We show that labeling with tandem split-wrmScarlet 11 -repeats increases fluorescence in vivo, and we provide the plasmid necessary to generate the dsDNA template (https://www.addgene.org/158533). We also generate a strain expressing an integrated copy of sfGFP 1-10 (Pé delacq et al. 2006) broadly in somatic cells (https://cgc.umn.edu/strain/ CF4587), and a strain expressing sGFP2 1-10 (Kö ker et al. 2018) specifically in the germline (https://cgc.umn.edu/strain/DUP223). Finally, we generate a dual-color strain expressing both sfGFP 1-10 and split-wrmScarlet 1-10 in somatic cells (https://cgc.umn.edu/ strain/CF4588), for two-color applications such as colocalization studies or organelle interaction. We hope that these resources will facilitate the study of C. elegans biology. As the collection of split-FP strains and related resources for labeling different tissues, organelles and proteins expands, we will post updates at doi.org/10.5281/zenodo.3993663.

C. elegans strains and maintenance
Animals were cultured under standard growth conditions with E. coli OP50 at 20 C (Brenner 1974). Strains generated in this work are listed in the Supplementary Table S3.

Nucleic acid reagents
Synthetic nucleic acids were purchased from Integrated DNA Technologies (IDT), GenScript or Genewiz. For knock-in of a single split-wrmScarlet 11 or sfGFP 11 sequence, 200-mer HDR templates were ordered in ssODN form (synthetic ssODN) from IDT. For knock-in of split-wrmScarlet 11 repeats, HDR templates were ordered in dsDNA form (plasmids) from GenScript or Genewiz. For plasmids injected as extrachromosomal arrays, sequences were synthesized and cloned into the pUC57 vector (Genewiz). The complete set of crRNAs and DNA sequences and plasmids used for the experiments described here can be found in Supplementary Tables S1, S4, and S5.

Strain generation: Mos1-mediated single-copy insertion
The COP1795 strain was generated by NemaMetrix's MosSCI services. The PHX1797 strain was generated by SunyBiotech's MosSCI services, using a codon-optimized sequence of split-wrmScarlet 1-10 with three introns, and engineered to avoid piRNA recognition transgene silencing Zhang et al. 2018; Supplementary Table S1).

Microscopy
Confocal fluorescence imaging was performed using NIS Elements imaging software on a Nikon confocal spinning disk system equipped with an Andor EMCCD camera, a CSU-X1 confocal scanner (Yokogawa), 405, 488, and 561 nm solid-state lasers, and 455/50, 525/26 and 605/70 nm emission filters. Transgenic animals expressing sfGFP 11 or split-wrmScarlet 11 were screened using a Leica M165 FC fluorescent stereomicroscope equipped with a Sola SE-V with GFP and mCherry filters.

Image analysis
Images were analyzed using Fiji. Image manipulations consisted of maximum intensity projections along the axial dimension, rolling ball radius background subtraction, smoothing, and LUT minimum and maximum adjustments. Masks were created by thresholding and setting the pixels under the threshold cutoff to NaN. Plotting of values per pixel was carried out in python 3, using numpy and matplotlib. When performing normalizations for split-sfCherry3 versus split-wrmScarlet, the red channel was divided by the green channel (mNeonGreen::FP 1-10 ) because the localization of both fragments is expected to be the same (cytosolic). For normalization of signals where mTagBFP::FP 11 was targeted to the membrane, the blue channel was used instead of the green channel.

Mounting worms for microscopy
Pads made of 3% agarose (GeneMate) were dried briefly on Kimwipes (Kimtech) and transferred to microscope slides. Approximately 10 lL of 2 mM levamisole (Sigma) was pipetted onto the center of the agarose pad. Animals were transferred to the levamisole drop, and a coverslip was placed on top before imaging.

Brood size analysis
Eight single synchronized adults grown at 20 C were transferred to fresh plates every 24 h until cessation of reproduction, and the number of viable progeny produced by each worm was scored.

Lifespan assays
NGM plates were supplemented with 5-Fluorouracil (5-FU, Sigma, 15 lM) (Goudeau et al. 2011) in order to prevent progeny from hatching and with kanamycin sulfate to prevent bacterial contamination (Sigma, 25 lg/mL). Animals fed with kanamycinresistant OP50 were scored manually as dead or alive, from their L4 larval stage defined as day 0. A worm was considered alive if it moved spontaneously or, in cases where it wasn't moving, if it responded to a light touch stimulus with a platinum wire. Animals that crawled off the plates, had eggs that accumulated internally, burrowed or ruptured were censored and included in the analysis until the time of censorship.

Structure prediction and rendering of split-wrmScarlet
Phyre2 was used to predict the three-dimensional modeling in intensive mode with default parameters (Kelley et al. 2015). The 3D model obtained was visualized using PyMOL (v2.2.0).

Statistical analysis
Differences in fluorescence intensity between groups were compared using unpaired t-test with Welch's correction. Data are presented as means 6 SD. Kaplan-Meier estimates of survival curves were calculated using survival (v2.38-3) and rms (v4.5-0) R packages and differences were tested using log-rank test. The number of animals used in each experiment is indicated in the figure legends.

Split-wrmScarlet
To engineer split-wrmScarlet, first we introduced a 32 amino acid spacer between the 10th and 11th b-strands of full-length yeast-codon optimized mScarlet, following a strategy described previously (Feng et al. 2017). We subjected the spacer-inserted mScarlet sequence to several rounds of semi-random mutagenesis in E. coli, generating a version with fluorescence comparable to the full-length mScarlet when expressed in bacteria. However, upon separating the two fragments into two S. cerevisiae plasmids to test for complementation, we observed no detectable fluorescence in yeast. We decided to continue with several rounds of selection of new mutant libraries in yeast using FACS, by fusing the small fragment (without the MDELYK C-terminus residues) from our brightest E. coli clone to a plasma-membrane-targeted blue FP (mTagBFP), and expressing the large fragment from a high-copy number vector containing a strong promoter. The brightest resulting protein, which we named split-wrmScarlet, contained 10 amino acid substitutions relative to the C-terminal truncated mScarlet (Supplementary Figure S1, A and B). Fluorescence microscopy of yeast containing both plasmids corroborated that split-wrmScarlet showed the expected localization and can reach brightness comparable to that of intact mScarlet in yeast (Supplementary Figure S2, A and B).
Split-wrmScarlet is threefold brighter than split-sfCherry3 in C. elegans muscles In order to compare split-wrmScarlet to split-sfCherry3, the brightest published red split-FP at the time of the experiment, we combined the FP 1-10 and FP 11 fragments into a single plasmid for each fluorophore. Specifically, we generated worm-codonoptimized plasmids encoding three nuclear localization signals (NLS), mTagBFP2, FP 11 , a T2A peptide-bond-skipping sequence, mNeonGreen and the corresponding FP 1-10 , driven by the eft-3 promoter (Supplementary Figure S3A). Each FP 11 was linked to mTagBFP2 in order to reduce the risk of proteolysis of the short peptide, and mNeonGreen was linked to FP 1-10 to monitor its expression, and for normalization purposes. Each construct was injected into wild-type animals and fluorescent progeny were analyzed. Unexpectedly, split-sfCherry3 turned out to be toxic when expressed ubiquitously, whereas 99% of split-wrmScarletoverexpressing worms became viable adults (Supplementary Figure S3B).
In an attempt to reduce split-sfCherry3-associated toxicity, we modified our construct by using the muscle-specific myo-3 promoter and removing the NLS sequence ( Figure 1B). We did not detect toxicity associated with the expression of these constructs and were able to compare the fluorescence of split-sfCherry3 and split-wrmScarlet in young adults. Red fluorescence emitted from split-wrmScarlet was 2.9-fold higher than that of split-sfCherry3 when normalized to the mNeonGreen signal (Figure 1, B and C and Supplementary Figure S4). We also observed a 60% higher expression level of mTagBFP2 in the split-wrmScarlet-expressing animals (Supplementary Figure S4). It is worth noting that differences in expression levels could influence both brightness and toxicity comparisons. A more controlled way to compare the split FPs at similar expression levels would be to make single-copy genomic insertions of these constructs at a neutral site in the genome.

Split-wrmScarlet 11 -mediated tagging in all somatic tissues or specifically in muscles
Our protein-tagging approach was analogous to existing split-FP methods developed for human cells , Leonetti et al. 2016) and C. elegans (Hefel and Smolikove 2019). It requires split-wrmScarlet 1-10 (i.e., just the large fragment of split-wrmScarlet without the 11th b-strand) to be expressed in the cell or tissue of interest, and the small split-wrmScarlet 11 fragment to be inserted at an endogenous locus to tag a protein of interest ( Figure 1A).
To build strains expressing single-copy insertions of split-wrmScarlet 1-10 , we first optimized its sequence for C. elegans codon usage (Redemann et al. 2011) and included three introns (Supplementary Table S1). The strain expressing split-wrmScarlet 1-10 throughout the soma (driven by the eft-3 promoter and unc-54 3'UTR) was generated by editing the genome of the existing MosSCI line CA1200 (Zhang et al. 2015) and replacing the sequence encoding tir-1::mRuby with split-wrmScarlet 1-10 using CRISPR/Cas9 and hybrid DNA templates (Paix et al. 2015;Dokshin et al. 2018;Supplementary Table S4). In order to perform tissuespecific labeling, we generated a strain expressing musclespecific split-wrmScarlet 1-10 using the SKI-LODGE system in the strain WBM1126 (Silva-García et al. 2019; Supplementary Table  S4). The expression of split-wrmScarlet 1-10 in these two lines did not affect the number of viable progeny (Supplementary Figure  S5A) nor lifespan (Supplementary Figure S5B and Table S6), suggesting that the expression of split-wrmScarlet 1-10 had no deleterious effect. To tag a gene of interest with the split-wrmScarlet 11 fragment, we used microinjection of preassembled Cas9 ribonucleoproteins, because this method enables high-efficiency genome editing in worms (Paix et al. 2015). The most efficient insertion of short sequences in C. elegans was previously shown to be achieved using ssODNs (Paix et al. 2015;Prior et al. 2017;Dokshin et al. 2018). A great advantage of this strategy is that all of the components required for editing are commercially available or can be synthesized rapidly in the lab (Leonetti et al. 2016). Synthetic ssODNs have a typical size limit of 200 nt. The small size of split-wrmScarlet 11 (18-24 a.a.) is key: 200 nt can encompass split-wrmScarlet 11 (66-84 nt, including a 4 a.a. linker) flanked by two homology arms >34 nt (up to 67-58 nt) for HDR. In principle, a few days after the somatic and/or muscle-specific split-wrmScarlet 1-10 strain(s) are microinjected, progeny can be screened for red fluorescence, genotyped and sequenced to check the accuracy of editing (Figure 2; a detailed protocol is available at dx.doi.org/10.17504/protocols.io.bamkic4w). If desired, co-CRISPR strategies such as dpy-10(cn64) (Paix et al. 2015) or coinjection with pRF4 (Dokshin et al. 2018) can be used to screen for correct candidates and to control for microinjection efficacy and payload toxicity.
To test our approach, we used it to tag six proteins with distinct subcellular localizations. Starting with the somatic split-wrmScarlet 1-10 parental strain CF4582, we introduced split-wrmScarlet 11 at the N-terminus of TBB-2, FIB-1 or VHA-13 or at the C-terminus of EAT-6, HIS-3, and TOMM-20 (Supplementary Table  S4). These proteins mark the cytoskeleton, nucleoli, lysosomes, plasma membrane, nuclei, and mitochondria, respectively. Importantly, for tagging transmembrane proteins, the split-wrmScarlet 11 tag was introduced at the terminus exposed to the cytosol. Split-wrmScarlet fluorescence from all six proteins matched their expected subcellular localization in somatic cells (Figure 3 Pglh-1::glh-1::T2A::wrmScarlet 1-10 ::glh-1 3'UTR OR Germline-specific expression Figure 2 Split-wrmScarlet 11 -mediated tagging. Schematic representation of the split-wrmScarlet tagging workflow to visualize endogenous proteins specifically in muscles, germline, or throughout the soma. Some illustrations were created with BioRender.com. A-F). To test the muscle-specific split-wrmScarlet 1-10 line CF4610, we tagged the N-terminus of endogenous FIB-1 with split-wrmScarlet 11 and confirmed the fluorescence from nucleoli in muscle cells ( Figure 3G). Together, our results show that split-wrmScarlet enables rapid fluorescent tagging of proteins with disparate cytoplasmic or nuclear locations expressed from their endogenous loci. The 18 a.a. split-wrmScarlet 11 sequence used for these experiments ends with two glycines. In mammalian cells, C-terminal gly-gly sequences have been reported to function as degradation signals (Koren et al. 2018). Our TOMM-20::split-wrmScarlet 11 had a spontaneous mutation of the last glycine to a stop codon (Supplementary Table S4), which could be problematic if the protein degradation mechanism, DesCEND (destruction via C-end degrons) operates in C. elegans. However, we do not detect differences in protein abundance of HIS-3 versus HIS-3::split-wrmScarlet 11 by western blots in C. elegans (Supplementary Figure S11). We also do not detect differences in protein abundance of mScarlet truncated to end in gly-gly compared to mScarlet ending in MDELYK via fluorescence in yeast. Nonetheless, we recommend using a 24 a.a. split-wrmScarlet 11 sequence YTVVEQYEKSVARHCTGGMDELYK when labeling proteins at their C-terminus to avoid the possibility that split-wrmScarlet 11 ending in gly-gly could function as a degron. This modified sequence still fits within the 200 nt ssODN synthesis limit and works at least as well as the 18 a.a. split-wrmScarlet 11 sequence (Supplementary Figure S6).

Split-wrmScarlet 11 -mediated tagging in the germline
Our initial attempt to use split-wrmScarlet in the germline failed. We made a single-copy integrated Psun-1::split-wrmScarlet 1-10 ::sun-1 3'UTR strain via MosSCI, but when we injected a plasmid encoding mNeonGreen::split-wrmScarlet 11 , we observed green fluorescence, but no red fluorescence (Supplementary Figure S7,  A and B), suggesting the absence of split-wrmScarlet 1-10 expression. We suspected germline silencing of the germline-expressed split-wrmScarlet 1-10 , so we attempted an alternative expression approach by taking advantage of the germline-helicase protein GLH-1, which is highly expressed and germline-specific (Marnik et al. 2019). We fused a T2A::split-wrmScarlet 1-10 sequence to the Cterminus of the endogenous glh-1 gene using CRISPR/Cas9. The high expression of GLH-1 yielded high expression of split-wrmScarlet 1-10 , and the T2A separated split-wrmScarlet 1-10 from GLH-1 (Liu et al. 2017). The glh-1::T2A::split-wrmScarlet 1-10 strain (https://cgc.umn.edu/strain/DUP237) can be used like our other tissue-specific strains for germline-specific tagging. To demonstrate this, we tagged the N-terminus of endogenous FIB-1 with split-wrmScarlet 11 , and we observed red fluorescence localized to the nucleoli specifically in the germline and embryos, as we hoped ( Figure 3H; Supplementary Figure S8, A and C). Finally, we note that the strategy used to express wrmScarlet 1-10 or sGFP2 1-10 in the germline, by tagging the 3' end of the endogenous glh-1 with T2A::FP 1-10 with CRISPR/Cas9, could be used to express any other protein of choice.

Split-wrmScarlet 11 tandem repeats increase fluorescence
To benchmark the fluorescence intensity of split-wrmScarlet against its full-length counterpart, we first generated endogenously-expressed wrmScarlet::vha-13 (El Mouridi et al. 2017) transgenic animals and compared their fluorescence to split-wrmScarlet 11 ::vha-13 in worms expressing split-wrmScarlet 1-10 somatically (Figure 4, A and B). At the vha-13 locus, split-wrmScarlet was about half as bright as a full-length fluorophore (48%), a ratio comparable to that of split-mNeonGreen2 and its full-length counterpart in human cells (Feng et al. 2017).
Since visualizing endogenous proteins of low abundance can be challenging, it is key to address this limitation. Increasing the number of FP 11 domains tagged to an endogenous protein multiplies the number of the corresponding FP 1-10 s recruited, increasing the overall fluorescent signal in human cells (Leonetti et al. 2016) and in C. elegans (He et al. 2019;Hefel and Smolikove 2019). To demonstrate that split-wrmScarlet fluorescence is enhanced by split-wrmScarlet 11 tandem repeats, we introduced two split-wrmScarlet 11 domains at the N-terminus of VHA-13 and three split-wrmScarlet 11 domains at the C-terminus of HIS-3 in animals expressing somatic split-wrmScarlet 1-10 . Compared to animals carrying a single split-wrmScarlet 11 at the identical locus, carrying two split-wrmScarlet 11 s increased overall fluorescence by 1.5-fold, while carrying three increased it by 2.3-fold (Figure 4, C and D). Note that our three-split-wrmScarlet 11 tandem sequence exceeds the 200 nt ssODN synthesis limit, so we used dsDNA donor templates for these constructions (Supplementary  Table S4).

sfGFP 11 -mediated tagging in somatic cells
Split-sfGFP has been used successfully in worms before (Noma et al. 2017;He et al. 2019;Hefel and Smolikove 2019). However, there is still a need for a strain that ubiquitously expresses sfGFP 1-10 in the soma from an integrated single-copy insertion in order to avoid heterogeneous expression and time-consuming manual maintenance. To build this strain, we codon-optimized the original sfGFP 1-10 sequence for C. elegans and included one intron (Cabantous et al. 2005;Redemann et al. 2011;Supplementary Table S1). We initially generated a strain expressing sfGFP 1-10 driven by the let-858 promoter and unc-54 3'UTR using MosSCI, but later replaced the let-858 promoter with the eft-3 promoter using CRISPR/Cas9 and hybrid DNA donor template because we observed that Peft-3 resulted in significantly higher levels of gene expression (Paix et al. 2015;Dokshin et al. 2018;Supplementary Table S4). To validate this strain, we inserted sfGFP 11 at the N-terminus of lysosomal VHA-13 or at the C-terminus of nuclearlocalized HIS-3 ( Figure 5, A and B). Both strains yielded relatively bright signals in accordance with their predicted subcellular localization. We generated eGFP::VHA-13 transgenic animals and compared their fluorescence to sfGFP 11 ::VHA-13 in worms expressing sfGFP 1-10 somatically (Supplementary Figure S9, A  and B). At the vha-13 locus, split-sfGFP was about a third as bright as a full-length eGFP. It is worth noting that this comparison is not perfect, in part due to the presence of the six superfolder mutations S30R, Y39N, N105T, Y145F, I171V, and A206V in sfGFP but not in eGFP.

The current split-wrmScarlet is not detectable in mammalian cells
We failed to detect split-wrmScarlet in mammalian cells, despite our efforts to rescue its fluorescence by screening a mammalian- Figure 4 Split-wrmScarlet 11 tandem repeats increase fluorescence. (A) Images of animals carrying either wrmScarlet, split-wrmScarlet 11 or two tandem repeats of split-wrmScarlet 11 inserted at the endogenous VHA-13 N-terminus. (B) Emission intensities of animals carrying wrmScarlet, split-wrmScarlet 11 or dual split-wrmScarlet 11 inserted at the VHA-13 N-terminus. Mean 6 SD. Circles are individuals. ****P < 0.0001, ***P < 0.001, **P < 0.005. (C) Images of animals carrying either a single split-wrmScarlet 11 or three tandem repeats of split-wrmScarlet 11 inserted at the HIS-3 C-terminus. (D) split-wrmScarlet emission intensities from animals carrying a single split-wrmScarlet 11 or three tandem repeats of split-wrmScarlet 11 knock-in at the HIS-3 C-terminus. Mean 6 SD. Circles are individuals. ***P < 0.001. Images from each comparison were taken under identical instrument conditions using confocal microscopy and are shown using identical brightness and contrast settings. Images shown are from a single confocal plane. Scale bars, 50 mm.

Discussion
In this study, we describe several new tools for rapid CRISPRmediated labeling of endogenously expressed proteins using split fluorophores. While these tools are powerful and relatively easy to implement, several considerations should be taken into account when using this method.
First, detection of a given protein labeled with an FP 11 can only occur in a cellular compartment where the corresponding FP 1-10 is present. Proteins tagged with split-wrmScarlet 11 or sfGFP 11 generated in this work were either exposed to the cytosol or nucleoplasm (nuclei or nucleoli), where split-wrmScarlet 1-10 and/or sfGFP 1-10 were present. For proteins or epitopes located within the lumen of organelles, such as mitochondria or the endoplasmic reticulum, one might need to generate and validate C. elegans lines expressing split-wrmScarlet 1-10 or sfGFP 1-10 containing a mitochondrial localization sequence or ER signal peptide and retention signals, respectively. These approaches have been used successfully in mammalian cells with split-sfGFP when tagging ER-resident polypeptides ) and with split-sfCherry2 to detect proteins present in the mitochondrial matrix (Ramadani-Muja et al. 2019).
Second, when labeling proteins with split-wrmScarlet at the C-terminus, we recommend using the 24 a.a. split-wrmScarlet 11 sequence YTVVEQYEKSVARHCTGGMDELYK. As described in the Results section, our 18 a.a. split-wrmScarlet 11 fragment ends in gly-gly, which has been shown to be a degradation signal in mammalian cells. We cannot exclude the possibility that ending in gly-gly can be detrimental in C. elegans. The 24 a.a. split-wrmScarlet 11 still fits within a 200 nt ssODN donor template with a 12 nt linker and up to 58 nt homology arms and is at least as bright as the 18 a.a. split-wrmScarlet 11 (Supplementary Figure  S6).
Third, as for any other protein tag, it is important to select, when possible, a site that is unlikely to interfere with protein folding, function or localization (Snapp 2005; Nance and Frøkjaer-Jensen 2019). For example, N-termini of membrane-and organelle-resident proteins often contain signal peptides or localization signals, and C-termini may contain degron sequences that regulate protein turnover. Interestingly, there are examples sfGFP 1-10 ; sfGFP 11 ::vha-13 B B A sfGFP 1-10 ; his-3::sfGFP 11
of proteins that become toxic when tagged with a full-length GFP, but tolerate labeling with a split fluorescent protein. For example, SYP-4 was reported to be mostly functional when endogenously tagged with sfGFP 11 in a strain expressing sfGFP 1-10 specifically in the germline, but not functional when labeled with full-length GFP (Hefel and Smolikove 2019). Fourth, for proteins of interest present at low levels, we provided an alternative protocol to insert an additional two or three split-wrmScarlet 11 fragments, which increases the overall fluorescence substantially. However, the number of split-wrmScarlet 11 fragments could likely be increased further, to at least seven tandem repeats, based on approaches used successfully with split-sfGFP in human cells (Feng et al. 2017) and C. elegans (Noma et al. 2017;He et al. 2019;Hefel and Smolikove 2019).
Fifth, we would like to emphasize differences between our technique and the bimolecular fluorescence-complementation (BiFC) assay. When used together, high-affinity green and red split fluorescent proteins can provide information on colocalization, but unlike BiFC split proteins (Hu et al. 2002), they are not intended to assess protein-protein interactions directly. This is because BiFC split proteins require finely tuned weak affinities that do not disrupt the underlying interaction being studied. In our approach, only the split-wrmScarlet 11 fragment is attached to a protein of interest, the split-wrmScarlet 1-10 fragment is expressed in excess and unattached.
Finally, we would like to note that despite being three times brighter than the latest split-sfCherry3 in worms, our current split-wrmScarlet was not visible in the mammalian cell line we examined (Supplementary Figure S10). Its ability to fluoresce is not restricted to worms, because it can reach wild-type levels of brightness in yeast. We do not know the basis for this discrepancy. It is possible that the concentration of the split-wrmScarlet 1-10 fragment in mammalian cells is too low to drive complementation with split-wrmScarlet 11 . This could potentially be overcome by further mutagenizing split-wrmScarlet and screening for fluorescence at low expression levels in mammalian cells.
In conclusion, we believe our system can substantially increase the speed, efficiency, and ease of in vivo microscopy studies in C. elegans. We expect it to facilitate two-color and colocalization experiments and to find wide use in the worm community. We believe that these strains could facilitate novel or large-scale experiments, such as efforts to tag the entire genome of C. elegans.

Conflicts of interest
None declared.