Proximity labeling to detect RNA–protein interactions in live cells

RNA biology is orchestrated by the dynamic interactions of RNAs and RNA‐binding proteins (RBPs). In the present study, we describe a new method of proximity‐dependent protein labeling to detect RNA–protein interactions [RNA‐bound protein proximity labeling (RBPL)]. We selected the well‐studied RNA‐binding protein PUF to examine the current proximity labeling enzymes birA* and APEX2. A new version of birA*, BASU, was used to validate that the PUF protein binds its RNA motif. We further optimized the RBPL labeling system using an inducible expression system. The RBPL (λN‐BASU) labeling experiments exhibited high signal‐to‐noise ratios. We subsequently determined that RBPL (λN‐BASU) is more suitable than RBPL (λN‐APEX2) for the detection of RNA–protein interactions in live cells. Interestingly, our results also reveal that proximity labeling is probably capable of biotinylating proximate nascent peptide.

RNA biology is orchestrated by the dynamic interactions of RNAs and RNA-binding proteins (RBPs). In the present study, we describe a new method of proximity-dependent protein labeling to detect RNA-protein interactions [RNA-bound protein proximity labeling (RBPL)]. We selected the well-studied RNA-binding protein PUF to examine the current proximity labeling enzymes birA* and APEX2. A new version of birA*, BASU, was used to validate that the PUF protein binds its RNA motif. We further optimized the RBPL labeling system using an inducible expression system. The RBPL (kN-BASU) labeling experiments exhibited high signalto-noise ratios. We subsequently determined that RBPL (kN-BASU) is more suitable than RBPL (kN-APEX2) for the detection of RNA-protein interactions in live cells. Interestingly, our results also reveal that proximity labeling is probably capable of biotinylating proximate nascent peptide.
RNA-protein interactions are pervasive in cells [1,2]. RNAs serve as the binding sites for RBPs to form ribonucleoprotein. RNAs function together with specific binding proteins to determine post-transcriptional processes, including translation, RNA splicing, cleavage and polyadenylation, RNA editing, RNA localization and decay [3][4][5][6][7][8][9]. RNA-protein interactions execute numerous roles in cellular functions and diseases. There are diverse methods for detecting and characterizing RNA-protein interactions [2,[10][11][12]. One method of RNA antisense purification (RAP) is to isolate specific long non-coding RNA and its associated proteins [13][14][15]. These associated proteins are then identified by quantitative MS. Further validation and characteristics are performed. In the RAP-MS method, UV cross-linking is used to generate covalent bonds between the contacting RNA and protein, purifying RNAs in denaturing conditions to eliminate non-specific interactions. Such attempts have sufficiently defined the roles of Xist-mediated transcriptional silencing via direct interaction with protein complexes in X-chromosome inactivation [14]. RAP-MS is an applicable method for the biochemical isolation of RNA-binding proteins by enriching specific RNA and its associated RNA complexes from native cell lysis. However, this method only identifies highly abundant RNAs and isolates interactions that are cross-linked in cells. It requires a huge quantity of starting materials to obtain high purification yields of RNA-protein complexes because any individual RNA is probably present at only a very small proportion of the total cellular RNAs.
Chromatin immunoprecipitation has been used to determine chromatin bound DNA motifs. Similarly, immunoprecipitation-derived methods have also been proposed for the detection of RBPs, termed RNA immunoprecipitation (RIP) [16][17][18][19][20][21]. RIP involves the immunoprecipitation of an RBP of interest, which employs a specific antibody. The RNAs that co-immunoprecipitated with the protein are then subjected to sequence for identification. This method is referred to as RIP-seq. If UV cross-linking is applied, it would derive as cross-linking immunoprecipitation (CLIP) [22]. In cells, RNAs are always complexed with RBPs, whereas UV light of wavelength 254 nm (i.e. UV-C) induces covalent bonds between complexes of RBPs and their contacting RNAs. The RBP-RNA complexes could be enriched with a specific antibody, and bound RNAs are eluted from the RBP-RNA complexes and then determined. This procedure is relatively inefficient and time-consuming. RIP-seq and CLIP-seq are important methods for studying RNAprotein interactions. The major concerns are the efficiency of cross-linking and the availability of specific antibodies. In CLIP-seq, UV cross-linking is more specific, although it only links proteins to RNAs that are at near-zero distance [23].
The complexity of RNA-protein interaction in specific cellular context is incompletely addressed. Therefore, an efficient detection method for studying RNA-protein interactions is critical to a full understanding of gene expression regulation. Recently, Ramanathan et al. [24] utilized proximity-dependent protein labeling to efficiently identify the proteins that bind RNA transcript of interest in intact live cells. Proximity-dependent labeling employs enzymes that produce reactive radicals to covalently tether proximate proteins with biotin. The biotinylated proteins can then be purified in a denatured condition for further analysis by western blotting or MS. Based on the labeling enzymes used to catalyze reaction, proximitydependent labeling methods can be classified into two mainstreams: engineered ascorbate peroxidase (APEX or APEX2) and bacterial biotin ligase mutant, birA* (bioID or bioID2). APEX uses biotin-phenol as substrates, and H 2 O 2 is required to ignite the enzyme reaction. APEX catalyzes biotin-phenol into biotinphenoxyl radicals that diffuse to the surrounding milieu and react with proximate proteins in electronrich amino acid side chains. BirA* uses biotin as substrates to produce biotinoyl-5ꞌ-AMP radicals, which then react with lysine residues on proximate proteins.
Ramanathan et al. [24] engineered a derived promiscuous biotin ligase enzyme, birA*, from Bacillus subtilis, termed BASU. BASU rapidly biotinylates proteins bound to specific RNA motifs; biotinylated proteins are then separated by streptavidin pulldown, followed by MS to identify RNA interacting proteins. We tested BASU in the detection of RNA-protein interaction by transient transfection. We initially chose a well-known RNA-protein interaction pair: PUF and PUF binding motif [25,26]. This is because the PUF protein recognizes its target RNA motif in a modular manner and binds it with high affinity (K d % 18 nM). Subsequently, we generated an inducible stable cell line of BASU for labeling efficiency and consistency.
To date, birA* and APEX2 have been successfully applied to study a variety of proteins and processes in cells [27][28][29][30][31][32][33]. APEX2 labels proximate proteins within 1 min. APEX2 is probably the most active proximal labeling enzyme, which makes it very suitable for capturing dynamic processes. Two recent studies have showed the utilization of APEX2 in capturing the 'snapshot' of proximate interactions in a fast signaling turnover [27,28]. The technique also works well in confined compartments [29,30]. The prompts the question of whether APEX2 might also be suitable for RNAprotein interaction detection. In the present study, we aimed to establish an optimal RNA-bound protein proximity labeling (RBPL) method for identifying RNA-protein interactions in live cells. We addressed this question by performing a side-by-side comparison of two proximity labeling enzymes, BASU and APEX2, in which we utilized PUF as a point of reference. Stable expression system was applied to overcome the bias of transient transfection. Relative protein levels affinity [25,26]. BoxB RNA motif sequences flank the PUF RNA motif. BoxB RNA motif recruits the RBPL fusion protein (kN-BASU), leading to the biotinylation of proteins proximal to the flanked PUF RNA motif, flag-BFP-PUF protein and peptide dsRed in the vicinity (Fig. 1B). Next, we considered whether the RBPL biotinylation has labeling resolution in cells because proximity labeling is based on the diffusion of reactive radicals. Beta-actin, known as a housekeeping protein, is constitutively and stably expressed at high levels in cells. As we show, beta-actin proteins are not biotinylated in this process (Fig. 1B). We performed semiquantification of biotinylated PUF proteins by RBPL on PUF RNA motifs and scrambled controls. The RBPL with PUF RNA motif yielded an approximately five-fold enrichment of PUF proteins over scrambled controls in transient transfection of HEK-293T cells (Fig. 1C).

Generation of stable expression of RBPL cell lines
Although previous studies have reported that proximity-dependent labeling has a high background in mammalian cells [27,28,34,35], we have achieved a very good signal-to-noise ratio in this transient transfection experiment. Proximity labeling biotinylates proteins in a way that relies on proximal distance and duration time. We have further optimized this proximity labeling procedure by generating stable cell lines (Fig. 2). We generated a stable expression of RBPL under the inducible control of a Tet-On system in HEK-293T cells. The inducible expression of RBPL (kN-BASU) was confirmed by western blotting, as well as RBPL (kN-APEX2) expression ( Fig. 2A). The biotinylation function of RBPL can be fine-tuned with tetracycline-inducible expression. We also determined the RBPL expression with an immunofluorescent cytochemical staining assay. The results obtained reveal that the RBPL is mainly expressed in the cell cytosol (Fig. 2B).

Comparison of RBPL (kN-BASU) and RBPL (kN-APEX2) in RNA-protein interaction detection in live cells
The RBPL biotinylation labeling was controlled under the tetracycline induction. The RBPL (kN-BASU) with PUF RNA motif yielded an approximately 17-fold enrichment of PUF proteins over scrambled controls (Fig. 3A,C). The biotinylation labeling by RBPL showed excellent signal-to-noise ratio via the generation of stable cell lines. However, the RBPL (kN-APEX2) biotinylation labeling is very weak, with almost no discernible signal between PUF RNA motif and scrambled controls (Fig. 3B). In comparison, RBPL (kN-BASU) is more suitable than RBPL (kN-APEX2) for the detection of the RNA-protein interaction in live cells.

Discussion
In the present study, we describe RBPL as a method for detecting RNA-protein interactions in live cells. RBPL labels proteins that bound to specific RNA motif with biotin in a cellular environment, which could overcome bias from artificial RNA-protein interactions or post-lysis interactions. RBPL biotinylates proximal proteins within a range of reactive radical diffusion. The biotinylation of proximal proteins by RBPL allows for stringent washing to identify true proximal proteins. Potentially, RBPL permits the study of specific RNA foci associated proteomes in live cells when combined with quantitative proteomics.
Previous work has reported that the detection of RNA-protein interaction only uses transient transfection [24]. Three plasmids were transiently transfected within one culture plate at a time and it was claimed that exceptional signal-to-noise ratio and fast kinetics had been achieved. In the present study, we suspect that this may be highly dependent on efficient transfection or the specific context because proximity-dependent labeling was only effective under the situation when three plasmids simultaneously entered the cells. With transient transfection, this is not reproducible in our experience. We optimized the RBPL method to obtain an improved signal-to-noise ratio by generating a stable cell line. The inducibility of RBPL provides temporal expression for controlling the total docking time during which the labeling enzyme occupies the targeted RNA foci and minimizes the accumulation of excessive labeling activity. In the labeling results readout, we used biotinylated dsRed as an internal control instead of labeling enzyme expression because the dsRed located nearby could be a perfect labeling activity indicator of proximity labeling enzyme and is distributed all over the cell (Fig. S1). This may help to avoid unforeseeable system errors. In addition, our results reveal that proximity labeling by RBPL is capable of biotinylating peptides in its vicinity, as detected by dsRed antibody. These nearby peptides are probably newly synthesized.
The method presented here has some limitations in its current form. For example, RBPL is unable to distinguish between directly protein-binding or indirectly proteins associated with RNA. Moreover, birA* or its derived variant-based proximity labeling biotinylates primary amines of lysines [32,36]. It may not provide an unbiased detection chance as a result of unequal lysine residues appearing on the outer surface of proximate proteins. However, APEX2 catalyzes biotin-phenols into reactive biotin-phenoxyl radicals that diffuse to the surrounding milieu and react with proximal proteins in electron-rich amino acid side chains. Thus, we also tested the APEX2-based RBPL method.
However, APEX2-based labeling suffers from excessive background biotinylation that is unbound to the target RNA motif. Almost no discernible signal is seen between the PUF RNA motifs and the scrambled controls. Based on what we have found in the literature, APEX2 can identify proximate associations, although only with appropriate controls (not unlike birA*) and with SILAC labeling (unnecessary for birA*) [37][38][39][40]. BirA* is the predominant method used to identify candidate protein-protein interactions [32,34,35,41]. APEX2 has predominantly been used for compartmental proteomics, although, if used properly, it can be successful for identifying candidate protein-protein interactions or RNA-protein interactions (Dr Kyle Roux, personal communication). BirA* is a slow enzyme, and so perhaps even if it has a longer half-life in vitro, it is unable to saturate its 'target rich' environment because the low levels of reactive biotin are unable to travel far before reacting with something, regardless of whether this is a protein, a free amino acid or another reactive molecule in the crowded cellular environment. APEX2, on the other hand, generates a massive amount of reactive biotin in a short period of time, thus potentially saturating the proximate environment of reactive molecules and enabling more distal labeling.
Regarding the differential stability of the radicals from birA* and APEX2, we have not identified any compelling comparative in vivo evidence regarding this in the literature. The present study describes, for the first time, a side-by-side comparison of the two enzymes in the detection of RNA-protein interaction in the cellular environment. TTGATATAGGTTCGGTTGATATAGGGTTGATAT AGGGTTGATATACGGTTGATATA. 5X puf seq was located between two 3X Box B seq, generated by overlapping PCR and subcloned into pRiG-SV40 vector non-translated region behind dsRed. pRiG-SV40 vector was a gift from the laboratory of Dr Bin Tian in Rutgers New Jersey Medical School, Newark, NJ, USA. PUF protein cannot recognize base cytosine. Scrambled RNA sequence containing cytosine could be used as a control RNA motif. pRiG-SV40-3X BoxB-5X scrambled control-3X BoxB plasmid was constructed similarly. Scrambled control seq: AGGTAAACCCCAGGTAAACCCCAGGTAAACCCC AGGTAAACCCCAGGTAAAC. pcDNA3.1-Hygro(+) is commonly used for transient transfection studies. To generate BASU labeling plasmid pcDNA3.1-Hygro(+)-HA-kN-BASU, BASU seq was codon-optimized for mammalian cell expression and synthesized, a 66-bp kN (BoxB binding protein) sequence was synthesized, and HA-kN-BASU fragment was generated by overlapping PCR and subcloned into pcDNA3.1-Hygro (+) vector between the BamH1 and EcoRV sites.

Molecular cloning
RAR3G-APEX2-FLAG plasmid [42] (from the laboratory of Dr Ruijun Tian in Sust, Shenzhen, China) expresses APEX2-flag, containing Tet-on doxycycline-inducible expression vector, and could be used for lentivirus infection and packaging for the generation of stable cell line. To construct APEX2 labeling plasmid RAR3G-APEX2-FLAG-kN, a 66-bp kN (BoxB binding protein) sequence  Biotin-phenol (BP) labeling was performed as described previously [29,30]. Plasmids were transfected into HEK293T cells with Lipofectamine 2000. For transient transfection, 293T cells were analyzed 24 h after transient transfection with respect to labeling enzymes BASU-N plasmid and PUF protein expression plasmid, along with plasmid containing RNA motif. In BASU stable 293T cells, transfect PUF protein expression plasmid and RNA motif expression plasmid. RNA transcripts contain PUF binding sites. Proximity labeling enzymes BASU biotinylate proteins that bind to these RNA transcripts, and streptavidin-beads pull down biotinylated proteins. PUF was detected by its fusion flag. The APEX2 transfection procedure is the same as BASU. After 24 h of cell culture, BP was added to the culture media for 30 min. The medium was prewarmed to 37°C to facilitate BP dissolution. Thirty minutes later, H 2 O 2 was added at a final concentration of 1 mM, and the cells were incubated at room temperature for 1 min. Then, the labeling medium solution was quickly discarded and the reaction was stopped with cold quenching buffer (10 mM sodium ascorbate, 10 mM sodium azide and 5 mM Trolox). Cells were washed twice with quenching buffer, followed by three washes with PBS. Cells were then scraped onto the plate with 1 mL of RIPA buffer and detected by western blotting or MS.

Immunofluorescent staining
Cells were grown on coated coverslips. After dox induction for 24 h, cells were washed with ice-cold PBS and fixed with 4% paraformaldehyde for 15 min at room temperature. They were then washed three times followed by permeabilization with 0.5% Triton X-100 in PBS (PBST) for 5 min. Next, they were blocked for 30 min with 3% BSA in PBS at room temperature or at 4°C overnight. The APEX2 was detected with flag antibody, whereas BASU was detected with anti-HA. All primary antibodies were diluted 1 : 100 and incubated overnight at 4°C. After three washes with PBS, cells were incubated with secondary Alexa Fluor 488-conjugated Affinipure Goat Anti-Mouse IgG(H+L) for 1 h at room temperature. Secondary antibody was used at a dilution of 1 : 3000. Cells were washed three times with PBST. Cell-containing coverslips were mounted using a small drop of mounting medium with 4 0 ,6-diamidino-2-phenylindole. Examinations were performed by fluorescence microscopy (Ti-E; NiKon, Tokyo, Japan).

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article.