BiFC Magic: Shedding Light on Protein-Protein Interactions

1. Fluorescent proteins for BiFC assay

Since 1994, different fluorescent proteins and their variants have been discovered or developed [4,5,6,7]. In the witness of the recent technical progress, many of them have been adopted for BiFC assays (Figure 2).

Figure 1. Principle of BiFC and multicolor BiFC. (A) Diagram of fragments from mVenus [1] fluorescent protein (VN: 1-172aa and VC:155-238aa) and mCerulean [2] (CN: 1-172aa and CC: 155-238aa); (B) The different combination of fragments for BiFC. The protein A and protein B are fused to mVenus fragments (VN and VC) or mCerulean fragments (CN and CC). The interaction between proteins A and B induces the fluorescent protein to be reconstituted, leading to emission of fluorescence upon excitation; (C) For multicolor BiFC, the fusions of A-VN, B-CC and C-CN are constructed. Three-protein interaction are simultaneously visualized by the reconstitution of two different FPs. Reprinted from figure 1 in [3].

Figure 1. Principle of BiFC and multicolor BiFC. (A) Diagram of fragments from mVenus [1] fluorescent protein (VN: 1-172aa and VC:155-238aa) and mCerulean [2] (CN: 1-172aa and CC: 155-238aa); (B) The different combination of fragments for BiFC. The protein A and protein B are fused to mVenus fragments (VN and VC) or mCerulean fragments (CN and CC). The interaction between proteins A and B induces the fluorescent protein to be reconstituted, leading to emission of fluorescence upon excitation; (C) For multicolor BiFC, the fusions of A-VN, B-CC and C-CN are constructed. Three-protein interaction are simultaneously visualized by the reconstitution of two different FPs. Reprinted from figure 1 in [3].

Figure 2. A timeline of major achievements in BiFC development.

Figure 2. A timeline of major achievements in BiFC development.

The first application of BiFC can be dated back to the study of Regan and colleagues. Two fragments of GFP (NGFP and CGFP), split in a loop between residues 157 and 158, were fused to antiparallel leucine zippers in Escherichia coli [8]. Subsequently, YFP was successfully reconstituted in mammalian cells using fragmented YFP fused to interacting transcription factors [9]. Since then, the BiFC has been more widely used to visualize the PPIs in biological research, taking advantage of the different characteristics of fluorescent reporter proteins (Table 1).

Following the initial development with GFP, a monomeric Kusabira-Green (mKG) with spectral characteristics similar to GFP has been developed for a BiFC assay [10]. Otherwise, a notable advance in BiFC based on GFP is Tripartite split-GFP [11], in which superfolder GFP was used and will be further described in Section 4.1.

In addition to GFP-based BiFC, the YFP fragments for a BiFC assay were first demonstrated to visualize calcium-dependent PPIs in living cells [12]. Over the past decade, different variants of YFP appeared in succession, such as EYFP (S65G, S72A, T203Y) [9], Citrine ( a PH-resistant YFP variant) [13] and Venus (a rapidly-maturing YFP variant) [13]. A series of efforts have been witnessed on the development of CFP variants in BiFC assays. For example, ECFP split between amino acids 154 and 155 (CN155 and CC155) show fluorescence complementation when fused to bJun and bFos [14]. An improved CFP variant, termed Cerulean (S72A, Y145A and H148D), has also been successfully applied for BiFC assay [13].

In line with sequential application of GFP, YFP and CFP-based BiFC, the red fluorescent protein DsRed [15] would further extend the detection range of BiFC. Continuously, arising from its strong tendency to oligomerize [16], a monomeric RFP, mRFP1, was thus generated [17].As further better performance in BiFC, mRFP1-Q66T was developed with improved fluorescence intensity [18]. The mRFP1-Q66T-based BiFC assay was sensitive enough to catch weak and transient PPIs. Another RFP-based BiFC system is the split mutant monomeric RFP, mCherry, with excitation and emission wavelengths at 587/610 nm [6].

Moreover, the far-red fluorescent proteins are important for imaging deep tissue in animals. To achieve this purpose, a monomeric form of Katushka far-red protein, named mKate [19], was chosen to develop a far-red-based BiFC. This system eventually exhibited high BiFC efficiency in COS-7 cells [20]. Interestingly, the site-mutated mKate-S158A, mLumin, could increase 2-fold the mKate brightness in the same study [20]. Further mutation of mKate is called Neptune, which is the first bright fluorescent protein with an excitation peak reaching 600 nm. The monomeric variant of Neptune, mNeptune [21], has been successfully used for BiFC assays in animal tissues. For whole animal imaging, the near-infrared-based BiFC has been developed in the near past [22]. The near-infrared protein iRFP-based BiFC exhibits high fluorescence intensity and low cytotoxicity and utilizes endogenous concentrations of biliverdin chromophore to acquire fluorescence [23].

Generally, tracking BiFC signals is hindered by repeated capturing. To settle this matter, Dronpa, an artificial GFP-like fluorescent protein cloned from Pectiniidae [24], was used as a BiFC reporter. Dronpa has a reversible photo-switching activity between the fluorescent and non-fluorescent states. The Dronpa-based BiFC was successfully performed in HEK293 cells [25]. It will therefore enable the study of protein complex translocation between various cellular compartments.

Nowadays, the single-molecular PPIs in living cells is coming into focus. Nonetheless, current BiFC is limited by the brightness and photo-stability of fluorescent proteins, resulting in insufficient resolution for single molecule tracking. BiFC coupled with photo-activated localization microscopy (BiFC-PALM) [26,27], largely alleviates this problem, and allows the imaging and tracking of single-molecule PPIs at sub-diffraction resolution in crowded PPI background of living cells, by using split of photo-switchable (mEos3.2) [28] and photo-convertible (PAmCherry1) [29] fluorescent proteins. Likewise, the newly reported TagBiFC [30], which leveraged the split HaloTag system for single-molecule PPI in living cells via super-resolution imaging, provides an alternative approach to tackle this important issue.

2. Advantages of BiFC assay

BiFC is a very sensitive method with minimum background [31], thus it enables direct visualization of PPIs and has been successfully applied in a wide variety of cell types and organisms[32]. Since the inception of PPI study, diverse techniques have been developed, such as FRET, Y2H, and AP-MS. By comparison, BiFC presents several advantages per se over other PPI detection methods.

FRET and BiFC are the two most commonly employed PPI detection methods in cells. Like BiFC, FRET also employs the FPs as reporter signals. It needs energy transfer from an excited donor fluorophore to an acceptor at angstrom distances (10–100Å) [33] and in a permissive orientation. Therefore it is applicable only to analyze bimolecular, direct PPIs [34]. In contrast, BiFC assays may generate positive signals, as long as two tag-fused proteins are present within the same protein complex, including direct and indirect interactions between the two proteins. FRET, moreover, requires confocal image capture of two different FPs, as well as accurate and elaborate computation via time-correlated single-photon counting to predict protein interactions [33].

In an Y2H assay, the system addresses only the qualitative question, yes or no on protein associations. Its experiments drive expression of the target proteins into a rather unnatural context, especially for non-yeast proteins, differing from their native situation. The genetic-manipulated proteins, accordingly, may be misfolded due to the absence of mediating factors. Reciprocally, BiFC conducts context-dependent interaction studies with the native context from which the target proteins derive.

AP-MS method and its variants capture the bait protein complex in the native cellular context, but the weak PPIs will be disrupted by the cell lysis and purification steps [35,36]. Consequently, AP-MS precludes the weak or transient interaction partners from its final candidate list. The proximity-labelling methods, like BioID, have revitalized the detection of transient and low-affinity interactions for the AP-based methods [36]. Nevertheless, paired controls are still necessary to reduce the final false positives [37,38], which would be labor- and time-consuming, as compared to BiFC.

Collectively, these merits make BiFC as one of the most popular methods for the study of PPIs. A whole spectrum of FPs can be used for BiFC analyses, which underpins the multicolor visualization of different protein binding partners at the same time and in the same cell [39]. Throughout the years, BiFC assays have been utilized in high-throughput screens, uncovering novel PPIs in yeast, plant and mammalian cells (see Section 4.2). However, each of these methods has its advantages and limits that make them best suited methods in certain fields. The continuing efforts on BiFC improvement will further extend its application in more extensive circumstances.

3. Critical considerations

Despite these advantages mentioned above, there are some potential problems and critical factors that one should consider when performing BiFC.

3.1. Topology of BiFC-tag fusions

A wrong BiFC-tag orientation could mask the true PPI-based BiFC signals and change the protein localization, as well as inducing abnormal protein expression. Therefore, it is advised to test, firstly, the full-length FP-fused protein in a same vector backbone to the BiFC-vectors [40] , for confirming the target protein localization and expression level. If there is no available localization information, both N- and C-terminal fusion should be made for a subsequent co-localization test. Then choose the one same to the locus of its known interacting protein. When both tag orientations are available, the C-terminal tag fusion is preferable, as it has been reported that N-terminal tag fusion may generate higher background [40]. Once the tag orientation is determined, a corresponding split BiFC-tag will be used to generate final BiFC constructs and test the fluorescence intensity of reconstructed FP.

Another strategy is using the split BiFC-tag to determine optimal positions for the fusions [32]. FP fragments and both terminal ends of each interaction partner will be fused to test for all eight possible combinations (Figure 3). Moreover, the available combinations should go through the protein localization test and expression level comparison, by immunofluorescence or Western blot analysis.

3.2. Linker selection

In general, split BiFC-tags and target proteins are separated by a linker sequence to provide the flexibility of flanking proteins. The linker is crucial for independent motion of the FP fragments and the interaction partners, facilitating FP fragment reconstitution and proper folding of fusion proteins [41]. As a result, different linkers with various lengths have been designed, based on the sequences from natural multi-domain proteins [42]. The most frequently applied linkers in BiFC are composed of non-polar glycine and polar serine amino acids (GS) series, developed by Argos [43]. These flexible GS-based linkers have been widely applied by employing the (GGGGS)n or (GGSGG)n template, where n indicates the GS copy number. For BiFC linkers, empirically, GS-based linkers include about 8-15 amino acids, like DGGSGGS [44], GGGSGGGS [45], or GSSGGGGSGGGGSSG [46]. Sometimes, rigid linkers are also implemented to keep a fixed distance between the domains and to maintain their independent functions, for instance ATGLDLELKASNSAVDGTAGPVAT [32], as well as RSIAT and RPACKIPNDLKQKVMNH linker sequences, which have been used in many fusion constructs for BiFC analysis [9,14]. In addition, researchers have designed many other empirical linkers for some special purposes. Therefore, the linker selection may largely vary on a case-by-case basis.

3.3. Improvement of signal-to-noise ratio

The fluorescence intensity of reconstituted fragments should be comparable to that of independent non-fluorescent fragment self-assembly, which makes the true BiFC bright enough to be distinguished from background signals. This signal ratio between true BIFC and background is the so-called signal-to-noise (S/N) ratio, influenced by three main aspects: property of FPs, split-site, and hydrophobic interactions between split non-fluorescent fragments.

All proteins can be split into non-fluorescent fragments, but that does not mean they can all be used in BiFC; those recommended for BiFC analysis are summarized in Figure 2 and Table 1. To reduce the S/N ratio, the spontaneous self-assembly of split fragments, which is endowed by the inherent property of FPs to a large extent, should be avoided. Against this backdrop, two FPs, EYFP and Venus, have been most extensively used in BiFC [31,40].

Notably, applying different split-sites within same or different FPs shows concomitant variances of S/N ratio. Several studies reported that GFP could be split at different positions, which could lie in a loop between the 6th and 7th, the 7th and 8th, or the 8th and 9th β-strands [47]. Given the structural similarity between GFP and EYFP, fragments of EYFP truncated at residue 155 (designated YN155 and YC155), which are split at a position between the 8th and 9th β-strands, exhibit a relatively high S/N ratio, when used in BiFC [9]. Another split site at residue 173, generating YN173 and YC173, cut in the loop between 7th and 8th β-strands [14]. Similarly, Venus, as a mutated GFP with high fluorescence intensity, can be also truncated at either residue 155 or 173 [13]. A great advantage of Venus-based BiFC is that it avoids a short incubation at 30°C to facilitate fluorophore maturation, which is generally required for split YFP fragments [32]. Moreover, from a split-site screening study, Venus can be split at the loop between the 10th and 11th β-strands, with lower background fluorescence in BiFC application [48]. Compared to previously published sfGFP-based BiFC [49], the Venus-based BiFC assay had the lowest background fluorescence, albeit the same split-site imposed. Prospectively, this differential effect of the same split site might be also linked to the target proteins used in the test.

Decreasing the hydrophobic interactions between split non-fluorescent fragments has been reported to weaken the BiFC background fluorescence, further improving the S/N ratio [50]. In an informative study, Venus was split in two fragments, VN155 and VC155, which were used to test bFos-bJun interaction in the BiFC assay. To test the impact of hydrophobicity in BiFC, four residues L201, L207, V150, and I152 were replaced with other hydrophobic amino acids (L, I, V, and A respectively) by site-directed mutagenesis (Figure 4A). As a result, the V150A mutation increased by 8.6-fold the S/N ratio in the BiFC assay of bFos-bJun interaction, significantly improving the S/N ratio (Figure 4B).

3.4. Controls for BiFC assay

As mentioned previously, the true BiFC fluorescence signals are brought when the two tagged-proteins are interacting with each other and generating a functional FP. Equally, the random collisions of two FP fragments can also occur when they are co-expressed in a limited space of the subcellular compartment. These random encounters can subsequently generate PPI-independent and non-specific signals. Moreover, many BiFC assays, which were routinely implemented by transfection, involve transient overexpression of the candidate proteins. Coupling with the irreversibility of the BiFC, the artificial non-specific signals will be fixed and then measured together with the true PPI-driven fluorescence, contributing to the overall fluorescence signal in the BiFC assay. Accordingly, it is important to design and choose rigorous controls (summarized in Figure 5) for determining this non-specific fluorescence or technical background signal, and faithfully detecting interactions that are truly biologically relevant.

One of the most stringent controls in BiFC assay is the use of a mutant negative control where a single mutation or small deletion is introduced into one of the two interacting proteins (Figure 5B,C). In most cases, nevertheless, it will be challenging to find an appropriate mutated control, due to lack of protein biochemical or structural information, when examining an interaction between two new novel proteins. If so, a very similar protein from the same protein family can be a surrogate negative control (Figure 5D,E). In case this control protein is not available, an unrelated protein can be used as a negative control, when meeting certain conditions. It is desirable that this unrelated control protein co-localizes with the protein of interest in the same subcellular compartment. Moreover, the proof that this unrelated protein can interact with its known interactors using BiFC method, should be reported (Figure 5F). The last negative control assay is, BiFC competition analysis [9]. One of the binding partners is expressed as an untagged protein, which serves as a competitor when coexpressed with the two fusion proteins, to measure the signal difference with original BiFC, validating the true interactions. Retrospectively, a large number of BiFC-based studies have been published with inappropriate controls (Figure 5G–N). For example, the most frequently used controls are combinations of the individual split FP fragments with the complementary split-fragment tagged proteins (Figure 5K,L). In addition, the orientation of protein fusions (Figure 5O,P), should be taken into consideration, when applying BiFC control, to ideally keep the same topology to the tested interaction. Overall, appropriate controls are crucial to validate BiFC data and establish specificity of the observed PPIs.

3.5. Irreversibility of BiFC

The irreversibility of BiFC has been frequently reported [55], as a long-standing limitation. Related to this issue, most fluorescent protein-based BiFC complex formation is irreversible, which largely impedes the analysis of dynamic interactions. Instead, this seeming blemish facilitates the visualization of transient or weak protein-protein interactions [56] and offers a significant advantage for BiFC-based PPI screening [57,58]. Indeed, the reversible BiFC development would expand the BiFC application in monitoring the dynamic PPI, providing one alternative method in this domain. As a response to this need, a reversible BiFC system [59], based on a reconstituted infrared fluorescent protein IFP1.4, has been used to study the spatiotemporal dynamics of protein complexes in yeast and mammalian cells (Figure 6A). Another system named splitFAST [60], which specifically and reversibly binds fluorogenic hydroxybenzylidene rhodanine (HBR) analogs, demonstrated rapid and reversible complementation, allowing the real-time PPI visualization (Figure 6B).

In summary, a good experimental design is a prerequisite for a successful assay. BiFC is a technically straightforward and time-saving PPI detection method. Nonetheless, there are still step-to-step or case-to-case differences in practice. To set up a general standard guide as reference, one protocol was given in [3], which focuses on the visualization of protein interactions in cultured mammalian cells via the BiFC assay, using a true example coupling with published data.

4. Implementation of BiFC

In the last few years, although several other genuine methods have been developed to analyze PPIs, classical methods and their variants are still widely used by scientists, such as the BiFC-based methods.

4.1. Low-throughput BiFC-based applications

BiFC method has been widely applied to detect binary PPIs in many living systems, as discussed previously, especially EYFP- and Venus-based BiFC assays (Figure 7A). In addition to its generic usage, the use of two distinct FPs with different spectra, such as YFP-based BiFC combined with mCherry-based BiFC, enables the visualization of quaternary protein complex in living cells, or, at least simultaneous visualization of two independent binary PPIs [62]. This double BiFC combined method is designated as coBiFC (Figure 7B).

Besides coBiFC, a lot of attention has been paid on multicolor BiFC, named mcBiFC (Figure 7C), which provides an effective assay to compare the subcellular distributions of protein complexes formed with different binding partners [31]. In several studies, the split-Venus and Cerulean were used to construct the N-terminal part of Cerulean (1-172aa, Cerulean^N173) fused Protein A, the C-terminal part of Cerulean (155-238aa, Cerulean^C155) fused Protein B, and the N-terminal part of Venus (1-172aa, Venus^N173) fused Protein C [63,64,65]. Interaction of proteins A/B produces a Cerulean signal, whereas proteins B/C interaction generates a Venus signal. With imaging via different excitation and emission wavelengths, the mcBiFC allows studying ternary complexes, and investigates the interactions between three different proteins within the same cells. Alternatively, BiFC-based BRET or FRET (Figure 7D), which involves co-expression of two interacting proteins tagged to YFP- or Venus fragments with one interacting protein tagged to Renilla reniformis luciferase (RLuc) or Cerulean, can be also used for the analyses of ternary protein complexes [66,67]. To take a step further, theoretically, the luciferase can still be split for BiLC assay, enabling a secondary binary PPI detection. By combination of BiLC and BiFC-based BRET (Figure 7E), reconstituted Gaussia princeps luciferase (GLuc) was used for BRET on reconstituted Venus and enabled analyses of quaternary protein complex [68].

Moreover, the applications of sfGFP have to be mentioned here. Since the discovery of its third split site [69], tripartite split-sfGFP have been reported in studies of both binary and ternary PPI analyses. sfGFP was split into three parts: sfGFP1-9, sfGFP10, and sfGFP11. Each part can be fused to one of the target proteins. The reconstitution of the FP requires all three parts to be brought into proximity, and then demonstrating a ternary PPIs (Figure 7F) [70,71]. When binary PPIs is needed, only two twenty amino-acids long GFP tags, GFP10 and GFP11, are fused to interacting protein partners, and coexpressed the GFP1-9 fragment as complementary BiFC tag to finally fulfill the GFP-based BiFC investigation [11]. In addition to detecting PPIs, split-sfGFPs (sfGFP1–10 and sfGFP11) were also used for self-assembly, allowing visualization of single protein localization and imaging [72].

4.2. Large-scale applications of BiFC

A large number of high-throughout studies have been performed for scrutinizing complex protein interactomes in diverse organisms, thanks to current advances in various technologies, including that of BiFC-based screening method.

Since its invention in 2000, upon advantages of simplicity and low-cost, BiFC has become a widely used approach for PPI detections, and thereby suitable for large-scale screens. Indeed, several efforts have been witnessed over the past decade, coupling with the availability of nearly complete hORFeome collections [73,74,75,76], which enables prospecting PPIs at unparalleled scale in two different formats, arrayed or pooled BiFC screens. To date, large-scale screens using BiFC assays have been reported in yeast, plant and mammalian cells (Table 2).

The facile genetic manipulation on yeast is one of main advantages for BiFC assay. As the first effort on genome-wide screen, Sung et al. initially developed BiFC-tagged fusion plasmids, which allow expression of tagged target proteins in S. cerevisiae [77]. Subsequently, the construction of a S. cerevisiae fusion library expressing each ORF fused with the N-terminal fragment of Venus (VN) was achieved by Huh lab [78]. To perform a genome-wide BiFC screen for the SUMO interactome, 5,911 VN-tagged fusion (≈95% known yeast proteins) strains were mated with the strain expressing VC-tagged Smt3. Finally, 367 out of 5911 ORFs were identified as Smt3-interacting candidates, by fluorescence microscopy in arrayed format. Similarly, these BiFC-based screenings were also applied to interrogate ABC (ATP-binding cassette) transporter and TORC1 interactomes in yeast [79,80].

In the plant, a different BiFC-tagged ORF delivery system was used in a study of core cell cycle protein interactions [81]. The GFP-fragment fused constructs were transiently co-expressed in leaf epidermal cells of tobacco by A. tumefaciens-mediated leaf infiltration. Then the high-throughput BiFC assays were performed to test a total of 917 PPIs for 58 cell cycle-related proteins. As result, 341 PPIs were identified as BiFC-positive. In another BiFC-based screen, an Arabidopsis cDNA library comprising ~2×10⁵ cDNAs was fused to C-terminal fragment of YFP (CYFP) for PPIs screening of subsets of NYFP-fused baits in Arabidopsis leaf protoplasts [82]. This screen identified single cDNA clones encoding proteins that interact with bait proteins, VirE2 and VirD2, by co-transfection manner in an arrayed format.

Different from the large-scale screen by BiFC with fluorescence microscopy in yeast and plant, many BiFC screens in the mammalian cells were coupled with fluorescence-activated cell sorting (FACS). For instance, one split GFP-based BiFC system was designed to identify proteins interacting with protein kinase B (PKB), in which the BiFC-positive cells were collected by FACS [83]. In this study, C-terminal fragment of GFP (CGFP) was fused to bait PKB, proceeding a large-scale screen with a human cDNA library that expressed NGFP fusions in COS-1 cells, by pooled cotransfection. The resulting DNA, including genome DNA and ORF-containing plasmids, was extracted from BiFC-positive cells, followed by bacterial transformation, and then the single colony derived plasmids were retransfected in COS-1 cells, to perform the second BiFC for removal of false positive candidates. However, usage of the pooled DNA transfection greatly simplified the screening process compared to the one-by-one arrayed format, albeit further improvement is needed. Moreover, application of viral BiFC vectors allows efficient BiFC-based analysis in mammalian cells. Adenoviral BiFC vectors have been generated based on adenovirus high-throughput system and have been used to monitor G protein-coupled receptor (GPCR) activation in human cells by an adenovirus-based β-arrestin BiFC assay [84]. In parallel, a retrovirus-based protein-fragment complementation assay, termed RePCA, was developed to identify protein–protein interactions in mammalian cells [57]. In this study, a host cell line was made for stably expressing the N-terminal fragment of Venus (VN) fused protein AKT1. The screen was executed by infecting this stable bait cell line with prey VN-fusion lentivirus, following single fluorescent cell sorting by FACS. Large-scale screens by BiFC may also facilitate drug discovery. As published, mKG-based BiFC was used to screen for PPI inhibitors in a natural product library based on a cell-free system [85], which provided a deeper understanding of potential drug actions, therefore demonstrating great potential for high-throughput BiFC screening in drug discovery.

Taken together, recent application of BiFC in large-scale studies has demonstrated its potential for uncovering protein interactomes in live cells. In particular, BiFC-tagged cDNA or ORFeome-based cell libraries have allowed BiFC assays to be more widely applied in high-throughput studies. Upon the well-developed genome-wide screens, such as pooled overexpression screen, and drop-out screens (e.g. RNAi/shRNA screen or CRISPR-Cas9 screen), the arrayed large-scale BiFC screens are more popularly used, relying on combining multiple individual BiFCs in a microplate rather than a pooled format. Though, in few studies, pooled transfection or infection was used in BiFC screen, which is far away from one-gene-one-cell output, the current next-generation sequencing (NGS) is hardly implemented during the deconvolution step. This to-be-improved pooled screening to some degree impaired its original intention. Whereas continuous efforts have paved the future for a versatile and easy-taking BiFC-based screening, there is a long way to realize a rigorous quantification-based pooled method.