Background

Protein–protein interactions drive and regulate a broad array of cellular processes [1]. Defining individual protein–protein interactions and mapping entire networks as interactomes has become a key tool for the interpretation of cellular function and complex phenotypes [2]. Protein species display tremendous diversity in structure and the chemical environment in which they accomplish their functions. This diversity ensures that no single approach is capable of effectively detecting all forms of protein–protein interactions that take place in any given cell type. In response to this challenge, a wide range of tools have been developed to investigate protein–protein interactions. These include biochemical strategies of co-fractionation, affinity purification, and proximity labeling [3,4]. Genetic approaches of two-hybrid and synthetic genetic arrays have also provided extensive data sets that add to the catalog of protein–protein interactions [5,6].

Any form of protein–protein interaction analysis that depends on protein isolation or purification has limitations regarding the ability to detect weak or transient interactions and the potential to generate false-positive interactions when cellular compartments are mixed following cell lysis. In vivo approaches to the analysis of protein–protein interactions benefit from sensitivity and an improved likelihood that proteins will retain their native structure. The nuclear yeast two-hybrid that detects the interaction between specified "bait" and "prey" proteins that are imported into the nucleus has allowed extensive interactome analysis and is effective for nuclear and many cytoplasmic proteins; however, some cannot interact correctly when removed from their native context and environment [7].

Integral membrane proteins that reside partially or entirely within the confines of a phospholipid bilayer compose around 30% of the proteome of characterized eukaryotic cells [8]. This class of proteins mediates, among other functions, communication, cell–cell interaction, transport of metabolites and small molecules, energy metabolism, and lipid biosynthesis. Considering the diverse functions and, in some cases, the cell surface localization, it is not surprising that membrane proteins are a large portion of drug targets [9]. The wide range of cellular processes controlled or influenced by this class of proteins highlights the importance of understanding their function and regulation. Proteins do not function in isolation, and their activity, localization, and modification state can be governed by their interaction with other proteins. Developing an understanding of these relationships among integral membrane proteins is challenging owing to the phospholipid membrane environment in which these proteins reside. The phospholipid membrane environment influences the structure and function of integral membrane proteins [10]. Investigations of the structure, function, and protein interactions of integral membrane proteins are complicated by this environment, as in many cases, removal from the membrane leads to loss of structural integrity and protein aggregation. Thus, not all interactions among integral membrane proteins can be reliably identified through biochemical methods that rely on protein removal from the native cellular environment.

A variety of techniques can be applied to investigating protein–protein interactions among integral membrane proteins, including affinity purification and co-immunoprecipitation of detergent-solubilized proteins [11,12]. These have limitations, namely the need to remove the proteins from their native lipid bilayer environment with potential for protein aggregation and loss of any low-affinity protein interactions. In vivo approaches, including proximity labeling experiments using fusions of a membrane protein to a biotin ligase, have been reported but have not been widely applied, possibly owing to the extensive analysis required to identify the binding partners [13]. A variety of fluorescence microscopy strategies, including fluorescence resonance energy transfer (FRET) and fluorescence recovery after photobleaching (FRAP), to investigate the interaction between integral membrane proteins, have been discussed in recent reviews [14,15]. These are dependent on high-resolution microscopy capabilities and require fusing the proteins of interest to fluorescent protein tags. They are most effective with larger cells, such as mammalian cells, and can be challenging when applied to small cells such as Saccharomyces cerevisiae.

A genetic approach based on the two-hybrid concept, modified for the detection of protein–protein interactions between integral membrane proteins, was pioneered by Dünnwald et al. [16]. This strategy was based on the small and highly conserved protein ubiquitin, which can be covalently ligated to target proteins by ubiquitin ligases and removed from proteins by ubiquitin peptidases (Ubp) [17]. A remarkable feature of ubiquitin is that the amino-terminal residues 1–34 can be expressed separately from the carboxyl-terminal residues 35–76; if the two halves are brought into close proximity, they will refold into a functional ubiquitin molecule that can be recognized by ubiquitin peptidases [18]. In this "split-ubiquitin" method, two proteins of interest can be fused: one with the amino-terminus of ubiquitin (Nub) and the other with the carboxyl-terminus of ubiquitin (Cub). If the two proteins interact with one another, the two fragments of ubiquitin will fold into a functional molecule [16]. The native yeast ubiquitin has an isoleucine residue at position 13 (Ile 13), providing the Nub fragment referred to as NubI with a very high affinity for Cub and the ability to spontaneously refold. The replacement of Ile 13 with a glycine residue creates NubG. This variant has a significantly reduced affinity for Cub and requires that NubG and Cub be in very close proximity before the two fragments can refold into a functional ubiquitin that can be recognized by ubiquitin peptidases [18]. Ubiquitin Ile 13 forms part of the hydrophobic core of the folded ubiquitin protein. The hydrophobic nature of this amino acid residue is important for the conformational stability of ubiquitin. Based on the crystal structure of ubiquitin, Ile13 is in a position to make hydrophobic interactions with amino acid Phe 3 and Val 5 [19]. A mutation of Ile 13 to glycine (Gly 13), creating NubG, reduces the potential for hydrophobic interactions at the amino-terminus of ubiquitin and reduces conformational stability of the protein molecule. The reduced conformational stability is reflected in a reduced ability to spontaneously refold, thus requiring the amino and carboxyl fragments of ubiquitin to be in very close proximity to allow refolding of Nub and Cub fragments and the reconstitution of Ubp cleavage [18].

A strategy utilizing a split-ubiquitin was directly applied to the analysis of membrane proteins by generating a reporter system that could detect when the two fragments of ubiquitin came into close contact and folded into a functional ubiquitin molecule [20]. This system, referred to as the membrane yeast two-hybrid (MYTH), makes use of a fusion protein composed of the Cub fragment fused to a DNA-binding protein domain from the E. coli LexA protein and a transcriptional activation domain from the Herpes simplex virus VP16 including a nuclear localization signal. This fusion protein Cub-LexA-VP16 is abbreviated to CLV. When the CLV fragment is fused to one protein (a bait protein), and the Nub fragment is fused to another (a prey protein), the fusions remain intact unless the bait and prey proteins bind to one another. If the two proteins bind, this brings CLV and NubG into close proximity, allowing the Nub and Cub fragments to fold and be recognized by cellular ubiquitin peptidases (Ubp). The Ubp can subsequently cleave the LexA-VP16 fragment from Cub of CLV, releasing LexA-VP16 to migrate into the nucleus (Figure 1). The detection mechanism for this event makes use of reporter genes under the regulation of upstream LexA operator sites (LexAop) so that when a LexA-VP16 fusion is released and enters the nucleus, it can bind the LexA operators, allowing the fused VP16 to trigger activation of the reporter genes (Figure 1). Common reporter genes used include HIS3, ADE2, and LacZ [20,21]. This genetic approach to testing for interactions between membrane proteins benefits from investigating the interactions in vivo, with the integral membrane proteins remaining in their native conformation and membrane environment. The approach has found wide application with systematic screens to characterize large interactomes among S. cerevisiae membrane proteins in the endoplasmic reticulum [22]. The strategy has also been applied to discover and characterize protein interactions and interactomes among mammalian and plant cell proteins [23–25].

Figure 1. Split-ubiquitin membrane yeast two-hybrid (MYTH) systems. (A) The bait protein of interest is modified by the addition of a sequence coding for the carboxyl-terminus of ubiquitin (Cub) fused to the DNA binding protein LexA and transcriptional activation domain of VP16. This fusion is labeled as CLV in the figure. The prey protein is modified by the addition of a sequence coding for the amino-terminus of ubiquitin (NubG). The tester S. cerevisiae strain is also engineered by the addition of reporter genes under the regulation of LexA operator sites (LexAop) that can be bound by LexA. Commonly employed reporter genes include HIS3 and ADE2 that confer growth on medium lacking histidine and adenine. An additional reporter gene (LacZ) confers blue color development in the presence of X-Gal. In the absence of an interaction between the bait and prey proteins, the reporter genes are not activated, and yeast colonies will remain white in the presence of X-Gal and will not form in the absence of histidine and adenine. (B) In the case that the bait and prey proteins interact with one another, the Cub and NubG fragments will fold into a functional ubiquitin moiety and will be recognized for cleavage by ubiquitin peptidases (Ubp). This cleavage releases the LexA-VP16 fragment of the CLV fusion to migrate into the nucleus and activate the reporter genes, leading to blue colonies in the presence of X-Gal and growth on medium lacking histidine and adenine.

A variation of the MYTH assay that can be applied to investigations of S. cerevisiae membrane proteins is the integrated membrane yeast two-hybrid assay (iMYTH). In this approach, the endogenous gene encoding an integral membrane protein can be "tagged" with the CLV fusion to create a "bait" protein that can be used to identify interacting prey proteins [26].

Why use iMYTH?

The plasmid-borne split-ubiquitin approach has spawned a collection of plasmids that allow for the expression of amino and carboxyl-terminal fusion of CLV and Nub to bait and prey proteins [20]. These approaches are the only reasonable way to perform library screens where thousands of bait and prey proteins will be tested or when proteins from heterologous organisms are to be tested in S. cerevisiae. As a limitation, protein fusions will be overexpressed in the tester strain. When testing interactions between membrane proteins native to S. cerevisiae, if the CLV-tagged bait and NubG-tagged prey are placed under the regulation of their native promoters in episomal plasmids, both an untagged chromosomal gene and a tagged episomal gene will be expressed in the tester strain. This increases the expression of the bait and prey proteins and creates a situation where the tagged bait must compete with the untagged endogenous protein for binding partners. Additionally, overexpression can also induce mislocalization with the potential to induce false-positive interactions. Some integral membrane proteins are subject to regulation, which retains homeostasis over protein abundance. For example, the abundance of the essential ∆9 desaturase in S. cerevisiae, Ole1, is subject to regulation by the endoplasmic reticulum-associated degradation (ERAD) system; although Ole1 activity is essential, excess Ole1 is rapidly degraded [27]. When a plasmid-borne OLE1-CLV fusion is tested as a bait, this results in elevated auto-activation of the split-ubiquitin system, because ERAD-mediated degradation of the excess OLE1-CLV leads to release of the LexA-VP16 fusion and creates high background signal [28].

The iMYTH protocol is well-suited for testing interactions among protein species that naturally reside in the endoplasmic reticulum, peroxisome, or plasma membranes. The procedure can be effectively used for testing individual protein–protein interactions or for library screening with multiple bait proteins. One limitation of the iMYTH test for protein–protein interaction is that it cannot be employed if the CLV and/or NubG tags disrupt the structure or function of candidate bait and prey proteins. A related issue is that the topology of the bait and prey proteins must allow for both the CLV and NubG tags to be exposed on the cytosolic side of the membrane. Otherwise, they will not be accessible to the ubiquitin peptidases. When testing heterologous proteins not native to yeast in the iMYTH system, if the heterologous proteins require post-translational modifications for them to interact, those modifications may not occur in yeast, creating a limitation on the use of the assay. Similarly, heterologous proteins that interact with partners within the context of multiprotein complexes may not bind to specific partners in the absence of the entire complex. Additionally, heterologous proteins that display instability in S. cerevisiae may yield high background signals. Cases where the iMYTH assay is unsuitable for testing protein–protein interactions may be candidates for other procedures, including co-immunoprecipitation or microscopy-based co-localization studies. These alternative approaches have the advantage of being able to be performed on the cells where the proteins of interest naturally reside.

With the advent of PCR-mediated gene tagging [29], it has become possible to introduce fusions of the Cub-LexA-VP16 bait and NubG prey fusions into the chromosomal copies of the bait and prey genes. A primary advantage of employing an integrated bait or bait and prey approach is that the tagged proteins are retained under the regulation of their native promoters, and only one species of the protein is produced in the cells, rather than one tagged and one untagged version as would be the case if plasmid-borne copies of bait and/or prey were added to the cells. Thus, the tagged bait protein will be the only version participating in any complex formation in the cells. Further, this will aid in retaining stoichiometry within protein complexes and reduce the potential for false-positive signals that can be triggered by overexpression.