Over the past decade, bimolecular fluorescence complementation (BiFC) has emerged as a key technique to visualize protein-protein interactions in a variety of model organisms. The BiFC assay is based on reconstitution of an intact fluorescent protein when two complementary non-fluorescent fragments are brought together by a pair of interacting proteins. While the originally reported BiFC method has enabled the study of many protein-protein interactions, increasing demands to visualize protein-protein interactions under various physiological conditions have not only prompted a series of recent BiFC technology improvements, but also stimulated interest in developing completely new approaches. Here we review current BiFC technology, focusing on the development and improvement of BiFC systems, the understanding of split sites in fluorescent proteins, and enhancements in the signal-to-noise ratio. In addition, we provide perspectives on possible future improvements of the technique.

The dawn of BiFC technology

Protein-protein interactions (PPIs) play important roles in various biological processes. In the post-genome era, numerous PPI networks have been identified in various organisms, such as humans, worms, yeast, and plants (1-4). To verify these PPIs identified from genome-wide studies, numerous methods, including canonical yeast two-hybrid assay, in vitro pull-down assay, in vivo immunoprecipitation assay, fluorescence resonance energy transfer (FRET) assay, bioluminescence resonance energy transfer (BRET) assay, and bimolecular fluorescence complementation (BiFC) assay, have been used. Among these methods, the fluorescent protein-based BiFC assay has become widely accepted over the past decade.

The BiFC assay is based on structural complementation between two non-fluorescent N-terminal and C-terminal fragments of a fluorescent protein. Fluorescent proteins, e.g., green fluorescent protein (GFP), consist of 11 antiparallel β-strands forming a β-barrel, with an α-helix inside and several short helical structures (5) (Figure 1, A and B). The chromophore, located in the α-helix within the β-barrel structure, is chemically formed by three residues (6) (Figure 1, A and B). For BiFC analysis, several studies have demonstrated that fluorescent proteins can be split at a loop or within a β-stand (Figure 1, A and B). The two resulting non-fluorescent fragments can then be fused to proteins of interest that may interact (Figure 1C). If the proteins do interact, the non-fluorescent fragments are brought into close proximity and reconstitute an intact fluorescent protein that can be imaged using any fluorescence microscope (Figure 1C). Acquired images can then be used for quantitative analysis. The fluorescence complementation of GFP was initially demonstrated in vitro and in Escherichia coli by using two antiparallel leucine zippers fused to the two non-fluorescent fragments of a split GFP (7). Subsequently, an enhanced yellow fluorescent protein (EYFP)-based BiFC assay was successfully developed for visualizing protein interactions between the basic leucine zipper (bZIP) and NF-κB family of proteins in living mammalian cells (8). This study uncovered interesting localization patterns for several bZIP dimers and demonstrated crosstalk between Fos and Jun proteins and NF-κB family proteins (8). These results stimulated more widespread interest in applying BiFC technology to the visualization of PPIs in mammalian cells, bacteria, worms and plants (9).

Figure 1.  Principle of BiFC. (Click to enlarge)

Various fluorescence complementation-based technologies also have been developed to visualize molecular events involving more than two interacting proteins (10). The availability of these fluorescence complementation-based assays has enabled visualization of protein aggregation, protein folding, protein topology, conformational change, multiple protein complexes (multicolor BiFC), and ternary and tetramer complexes (BiFC-FRET and BiFC-BRET) in an unprecedented manner (10). In the meantime, several laboratories continue to add new fluorescent proteins to the BiFC toolbox. In this review, we will introduce recent improvements of BiFC technology and provide our thoughts on future development.

New fluorescent proteins added to the BiFC toolbox

Many fluorescent proteins possessing distinct spectral and physicochemical properties have been discovered or developed since 1994 (11-14). Among these, 15 fluorescent proteins have been found to work with fluorescence complementation assays (Table 1).

Table 1.  List of fluorescent proteins used in BiFC assays. (Click to enlarge)

The development and subsequent refinement of BiFC-supportive fluorescent proteins can be broadly classified into three stages (Figure 2). From 2000–2003, GFP from the jellyfish Aequorea victoria and its variants (EGFP, EBFP, ECFP, EYFP) were initially employed for BiFC (Figure 2; Table 1). The use of spectral variants was largely driven by the desire to perform multicolor analysis (15-20). These early results using GFP and its spectral variants provided proof of principle that most, if not all, fluorescent proteins could be used for BiFC assays (Figure 2, stage 1). Following the initial developments with GFP, it was realized that the sensitivity of GFP and its variants to the environment prohibited further application of BiFC under physiological conditions. Hence, a search for fluorescent proteins that would yield a bright signal under physiological culture conditions was initiated (Figure 2, stage 2). During this time (2004–2008), fluorescent proteins such as frGFP, Citrine, Venus, and Cerulean were demonstrated to be BiFC competent (21, 22, Table 1). The similar development of brighter BiFC systems has continued to date, e.g., sfGFP- and GFP-S65T-based BiFC systems (23, 24) (Table 1), owing to the availability of improved fluorescent proteins engineered by several labs (25-30). Starting in 2006, researchers began exploring the use of fluorescent proteins with unique spectral and physicochemical properties for BiFC analysis (Figure 2, stage 3). These BiFC-supportive fluorescent proteins provide optical tools for visualizing PPIs in living animals. For example, several red fluorescent proteins have been demonstrated to support BiFC analysis, including red fluorescent protein (RFP) variants from Discosoma sp., i.e., monomeric RFP (mRFP), mCherry, DsRed monomer, and mKate (20, 31-33) (Table 1). These longer wavelength RFP variants enable visualization of PPIs in deep tissues. Given that some of these fluorescent proteins have been used in whole body imaging, it is likely that PPIs can be visualized in small animals using RFP-based BiFC systems.

Figure 2.  History of BiFC developments. (Click to enlarge)

Recently, Dronpa, a photoswitchable GFP-like protein isolated from Pectiniidae, was also demonstrated to support BiFC analysis (34) (Table 1). Dronpa exhibits green fluorescence, with an emission maximum of 518 nm when excited at 503 nm (35). It can be photoswitched to a non-fluorescent state by strong light irradiation at around 490 nm and then switched back to a fluorescent state by minimal irradiation at 405 nm. It was reported that a Dronpa-based BiFC system possesses the same reversible photoswitching characteristic and could be used in repeated photo-switching experiments (34). Since Dronpa was successfully used for visualization of nucleocytoplasmic shuttling proteins (35), and many transcriptional regulatory proteins function as dimers and shuttle between the cytoplasm and nucleus, Dronpa-based BiFC analysis should help uncover how transcriptional regulatory proteins shuttle between the cytoplasm and nucleus. Likewise, Dronpa-based BiFC could also facilitate the study of protein complex translocation between various cellular compartments.

Given the strong evidence that most, if not all, fluorescent proteins are BiFC-competent, it is plausible to predict that other photoactivatable and photoconvertible fluorescent proteins, such as PA-GFP, EosFP, and Kaede, could be good candidates for BiFC analysis (36-38). If true, such fluorescent protein-based BiFC systems could enable new applications, including multicolor photoregulatable BiFC as well as super resolution imaging of BiFC.

Split sites

Circular permutation of proteins is a technique to change the order of amino acids in a protein sequence (39). The original N terminus of a protein is ligated with its original C terminus, and new N and C termini are generated by splitting the circularized protein in other structural regions. One of the applications of circular permutation is to determine the position where an insertion can be introduced without affecting the overall structure and protein folding. Several studies provided evidence that circularly permutated GFP could be split at positions in the loop between the 6th and 7th β-strands, in the 7th β-strand, in the loop between the 7th and 8th β-strands, in the 8th β-strand, and in the loop between the 8th and 9th β-strands (40). These positions were later employed to split EYFP for the development of BiFC (8). The identified split sites have been successfully used for BiFC analysis with several fluorescent proteins. For example, there are YN173/YC173 from EYFP, CN173/CC173 from ECFP, and RN169/RC169 from mRFP (Q66T), which are all split at a position between the 8th and 9th β-strands. Another split site is the loop between 7th and 8th β-strands. Examples are YN155/YC155 from EYFP and CN155/CC155 from ECFP (Figure 1, A and B). This split site is likely available for the GFP variants from jellyfish Aequorea victoria, but not for RFP variants from Discosoma sp. (31) (Figure 1, A and B). Although the above two split sites are widely used as canonical split sites of BiFC assay, other split sites have been reported for BiFC assays. For example, superfolder GFP can be split at the loop between the 10th and 11th β-strands (23) (Figure 1B). Because of high background fluorescence resulting from fluorescence complementation using this split site, several point mutations were introduced into these fragments. More recently, an extensive screen was performed by splitting Venus at 13 different sites in the hopes of developing a Venus-based BiFC assay with low background fluorescence (41). Interestingly, this screen identified the same loop between the 10th and 11th β-strands as the best split site for BiFC analysis (41). Contrary to the sfGFP-based BiFC, the Venus-based BiFC assay had the lowest background fluorescence among all combinations tested (41). Given the overall structural homology between sfGFP and Venus, it remains to be determined whether the differential effect of the same split site on these two proteins might be related to the proteins of interest used or the design of their BiFC plasmids (23, 41). Nevertheless, the fact that the C-terminal fragment from the 11th β-strand, which consists of 15 or 29 amino acids (sfGFP: a.a. 215–229, Venus: a.a. 210–238) (23, 41) (Figure 1B), should have less effect on the structure and folding of target proteins, and should be useful for BiFC analysis of proteins that are more sensitive to structural alterations.

Improvement of signal-to-noise ratio

The assembly of two complementary non-fluorescent protein fragments when each is fused to one of a pair of interacting proteins is essential for the BiFC assay. However, self-assembly of the non-fluorescent protein fragments independent of the interacting proteins can contribute to false-positive fluorescence, reducing the signal-to-noise (S/N) ratio (10). Recently three groups have reported improvements of the S/N ratio in the Venus-based BiFC assay in mammalian cells (42-44). All three groups split the Venus fluorescent protein into two non-fluorescent fragments at a position between the 7th and 8th β-strands, focusing on the interaction between the 7th and 10th β-strands of Venus (Figure 3; Table 2). Four residues (V150 and I152 within the 7th β-strand, and L201V and L207V within the 10th β-strand) were reported to affect the S/N ratio in the Venus-based BiFC, with substitutions at these residues significantly improving the S/N ratio (Figure 3).

Figure 3.  Mutation sites that increase the S/N ratio in Venus-based BiFC. (Click to enlarge)

Table 2.  Mutations that increase the signal-to-noise (S/N) ratio in Venus-based BiFC. (Click to enlarge)

“V150” - Lin et al. identified the V150L mutation as having a higher S/N ratio based on the structure, protein stability and solvent accessibility of the 7th β-strand of Venus (42) (Figure 3; Table 2). To determine the effect of V150L on the S/N ratio, the interaction between Bcl-XL and Bak BH3, two peptides involved in apoptosis, was examined in mitochondria (45). Although V150L slightly reduced the BiFC signal (fluorescence intensity) with the peptides, it increased the S/N ratio (referred to as “BiFC specificity” in the paper) by 2-fold (42).

Based on the structure, the solvent accessibility of the 7th β-strand, and the properties of amino acid residues, we also identified the V150L mutation as a substitution that increased the S/N ratio by 4.5 fold (43) (Figure 3; Table 2). Using the interaction between the bZIP domains of Jun (bJun) and Fos (bFos) in the nucleus as a model system (43) (Table 2), we found the V150L mutation did increase the S/N ratio, although the signal was too low to be useful for BiFC analysis. Because the interaction between Bcl-XL and Bak BH3 peptides occurs in mitochondria, whereas the interaction between bJun and bFos takes place in the nucleus (Table 2), it is possible that the V150L mutation might only be useful for BiFC analysis of proteins localized in relatively small organelles such as mitochondria, peroxisomes, and plastids, where BiFC complexes can be locally concentrated.

Nakagawa et al. reported another mutation, V150A, for BiFC analysis, that was identified by predictions of hydrophobic amino acids and by screening of amino acids for substitutions (44) (Figure 3; Table 2). To determine the effect of V150A on BiFC analysis, the interaction between bJun and bFos was again used (Table 2). The results indicate that V150A decreases false-positive interactions more than V150L; thus the V150A mutation might be useful to improve the S/N ratio.

“I152” – Compared with V150L, we found the I152L mutation significantly improved the S/N ratio while preserving fluorescence brightness (43) (Figure 3; Table 2). To determine the effect of the mutation on the S/N ratio, an interaction between bJun and bFos was similarly used (Table 2). I152L decreased false-positive fluorescence and increased the S/N ratio by 4-fold (43). The I152L mutation not only increased the S/N ratio in the VN155/VC155-based BiFC, but also increased the S/N ratio in the VN173/VC173-based BiFC. The effect of this mutation has also been evaluated in other model organisms such as plant (Allium cepa) and worm (Caenorhabditis elegans), as different experimental systems using these model organisms could potentially affect the S/N ratio of BiFC (46, 47). For the purpose of comparison with the S/N ratio in mammalian cells (COS-1), bJun/bFos and bJun/ΔbFos were similarly used as positive and negative PPIs, respectively. Using the same VN155/VC155-based BiFC system, the S/N ratios in A. cepa and C. elegans were only 1 and 2, respectively, and there was almost no difference between positive and negative PPIs in A. cepa. Nevertheless, when the VN155-I152L mutation was introduced, S/N ratios increased by 2-fold, to 2 and 4 in A. cepa and C. elegans, respectively (Kodama, Duren, and Hu, unpublished observations). These results suggest that this improved Venus-based BiFC system may be applicable to multiple model organisms.

“L201 and L207” - Nakagawa et al. also reported the identification of two residues (L201 and L207) in the 10th β-strand that, when mutated, could increase the S/N ratio in Venus-based BiFC (44) (Figure 3; Table 2). Based on the three-dimensional structure of Venus, it was predicted that L201 in the C-terminal fragment of Venus interacts with Y74, F84, V150, and I152 in the N-terminal fragment of Venus, and that L207 in the C-terminal fragment of Venus interacts with V61, Y143, and Y145 in the N-terminal fragment of Venus (44). The two leucines appear to form a hydrophobic core between the N-terminal and C-terminal fragments. When the L201V and L207V mutations were introduced, the S/N ratios increased by 4.2- and 3.0-fold, respectively (44). Because L201 interacts with V150 and I152, the underlying mechanism of the L201V mutation on the S/N ratio might be similar to that of the V150L, V150A, and I152L mutations.

In summary, the search for mutations to reduce self-assembly has demonstrated that the V150L, V150A, I152L, L201V, and L207V mutations can increase S/N ratios in the Venus-based BiFC system. Because different animal model systems were used in these studies, it is difficult to conclude which one is better. Nevertheless, results from these three independent studies provide evidence that the S/N ratio can be improved by a single substitution (Table 2). Since these studies focused on the interactions between the 7th and 10th β-strands, it would be interesting to test whether introduction of mutations into other sites could show increased improvement. Alternatively, introduction of multiple mutations might be necessary to achieve higher S/N ratios. It is interesting to note that the V152L mutation had little effect on the S/N ratio in the Cerulean-based BiFC assay (43), indicating other fluorescent protein-based BiFC systems should be individually tested for improvement by other mutations.

Irreversibility of BiFC

The irreversibility of BiFC complexes has been well documented (10), and it would appear that most, if not all, fluorescent protein-based BiFC systems are irreversible (8, 33, 43, 48-51). Although this irreversibility offers a significant advantage for analysis of transient or weak protein-protein interactions (48) and when performing BiFC-based PPI screening (49, 50), the irreversibility does limit the use of the BiFC assay for dynamic interactions. Given that PPIs are tightly regulated, development of reversible BiFC systems would provide opportunities to use BiFC for many biological studies. It should be noted that several groups have reported that BiFC complexes are reversible (52-54). These claims are largely based on observations that EYFP-based BiFC fluorescence was slightly decreased after chemical treatments (52-54). However, we and others only observed slight decreases in fluorescence intensity when using a rapamycin-inducible PPI system (43, 51). It is also interesting to note that the mutations that were identified to increase the S/N ratio had little effect on irreversibility (43). Therefore, it will be necessary to take a systematic approach (e.g., screening or directed evolution) to the development of a reversible BiFC system. If successful, it would not be surprising to see a full blooming of the BiFC assay.

Controls for BiFC assay

As noted previously, BiFC complementation occurs when two complementary non-fluorescent fragments of an FP are brought together by the pair of interacting proteins they are fused to, thus reconstituting a functional FP (Figure 1C and 4A). Alternatively, two FP fragments also can come together by random collision when co-expressed in the same subcellular compartment even though the two proteins may not interact. Fluorescence signals generated through random collisions are considered non-specific signal, but in these cases, both the interaction-driven fluorescence complementation and the interaction-independent non-specific assembly contribute to the overall fluorescence in the BiFC assay. In order to reveal specific protein interactions, the contribution of non-specific fluorescence in BiFC experiments must be determined. Because the expression level and subcellular localization of the non-fluorescent fragments that can self-assemble may differ from the fusion proteins, co-expression of the two non-fluorescent fragments may not provide such information (Figure 4). In many cases, the fluorescent signal resulting from co-expression of the two non-fluorescent fragments not fused to interacting proteins is even stronger than that resulting from co-expression of the two fusion proteins. Because of these concerns, we and others have recommended the use of negative controls, in which a mutation or small deletion is introduced into the interaction interface in one of the two proteins (55-61). However, a review of published BiFC experiments has led to the disappointing finding that some BiFC experiments do not include appropriate negative controls (Figure 4, B and C). Examples of inappropriate negative controls include the use of one fragment only, of a third protein that has a different structure or localization from the two proteins under study (Figure 4, D-L), or of mutants with different localization patterns than the wild-type protein or that have lower expression levels (e.g., because of shorter half-life or decreased stability). Several BiFC protocols have provided detailed discussions on the design and use of negative controls (55-58). It may be possible to forgo mutant controls in BiFC experiments if the same two fusion constructs are used under different experimental conditions and the change in BiFC signal is the end result. For example, the increase in BiFC signal for two proteins under study with and without the addition of a chemical (e.g., in the rapamycin-inducible PPI system) (43, 51) can be compared, provided that the chemical does not increase the expression level of the two fusion proteins.

Figure 4.  Negative controls for BiFC assays. (Click to enlarge)

One challenge facing BiFC users is the difficulty in identifying appropriate negative controls (Figure 4, B and C). This is often the case when the two proteins are new and no prior biochemical or structural information is available. To overcome this challenge, BiFC competition analysis should be considered as an alternative approach (Figure 5). The methodology behind BiFC competition analysis was originally reported using purified bJun-YN155 and bFos-YC155 fusion proteins (8). When differing amounts of free bJun were added to the assay, fluorescence complementation was inhibited in a dose-dependent manner (8) (Figure 5). The competitive nature of bJun interactions was also demonstrated using the multicolor BiFC assay to analyze interactions among bJun, bFos, and bATF2 (15). BiFC competition analysis has been successfully used in other BiFC experiments as well (17, 19, 62). However, it is important to note that the competitor must be co-expressed with the two fusion proteins, if not before the expression of the two fusions, since the BiFC complex is essentially irreversible once formed (8, 33, 43, 48-51). As demonstrated in the same bJun-YN155/bFos-YC155 system, addition of bJun at later times failed to efficiently inhibit fluorescence complementation (8).

Figure 5.  Principle of the BiFC competition assay. (Click to enlarge)

In summary, we strongly recommend the use of a mutant control where a single mutation or small deletion is introduced into the interaction interface in one of the two interacting proteins. If such a mutant control is not available, one should at least perform BiFC competition analysis. In either case, quantification of the BiFC signal is desired unless all-or-none results are obtained.

Here we have summarized the recent development of BiFC technology and provided a perspective on future developments. Since the first use of BiFC for visualization of PPIs in mammalian cells a decade ago, various BiFC systems and applications have been reported for use in a wide range of cells and model organisms. Continued development of new BiFC systems will likely further expand and enhance applications to biological research.

Competing interests

The authors declare no competing interests.