We describe the development of the G2H method and show that it can be used to detect interactions among proteins that cannot be studied using transcription-based two-hybrid systems, namely transcription factors. By virtue of its secretory pathway localization, the G2H also has the potential to be used to study the secretome, a class of proteins that remains poorly characterized 
. Detecting protein interactions via the G2H method relies on robust phenotypic changes that will enable the G2H method to be used in both screening and selection experiments.
By interrogating protein-protein interactions in the Golgi, rather than the nucleus, the G2H provides a powerful complement to transcription-based approaches. We demonstrated two specific examples of the utility of Golgi localization. First, Gal4AD is a transcriptional activator and cannot easily be studied using assays that employ a transcriptional readout 
. By testing Gal4p AD interactions in the Golgi, we avoided off-target transcriptional effects. Furthermore, we were to detect specific Gal4p interactions, even in the face of non-specific binding. Second, by targeting a toxic protein to the secretory pathway, we relieved its negative growth effects. Overexpression of Gal11p normally results in a dramatic decrease in growth 
, yet cells expressing the Golgi-localized LOC-Gal11 construct did not experience this toxicity. Indeed, the opposite was true: co-transformation with the LOC-Gal11 plasmid improved the growth of Gal4AD-CAT-expressing yeast.
In addition to detecting the interactions of Gal4AD with a number of binding partners, we were pleased to observe robust detection of the MyoD:Id2 interaction. Because the LOC-MyoD fusions are sequestered to the luminal face of the Golgi membrane, we were concerned that they might preferentially homodimerize with one another, rather than heterodimerizing with the soluble Id2-CAT fusion. Nonetheless, we are able to observe strong evidence of the LOC-MyoD:Id2-CAT heterodimeric interaction. Competition for binding will also be expected in cases where the prey is endogenously expressed within the Golgi and capable of competing with the LOC fusion protein for binding to the bait-CAT fusion protein. To investigate whether this situation will interfere with interaction detection by the G2H, we plan to conduct experiments to test the ability of the G2H to detect interactions that normally occur within the secretory pathway.
Like the traditional, transcriptional-based Y2H, the G2H is likely to have limitations. Proteins not normally localized to the secretory pathway could misfold in the G2H due to glycosylation of cryptic acceptor sequences or abnormal disulfide bond formation, thereby rendering them unable to engage in their normal protein-protein interactions. Misfolded proteins could also be retained in the ER, potentially leading to false positive signals. We observed reconstitution of Och1 in the Golgi through the MyoD:Id2 interaction and through the interactions of Gal4AD with a number of binding partners; therefore, at least some nuclear protein-protein interactions can assemble the secretory pathway environment, but others may be unable to do so due to differences in pH, ion concentration (Ca2+ in particular), oxidation state, or protein composition. Because the secretory environment is distinctly different from the nucleus, we predict that the G2H will be able to detect many protein-protein interactions that the classical, transcription-based Y2H cannot. Conversely, we expect that many interactions that are readily detected by the classical Y2H will be inaccessible to the G2H.
So far, the only case where the G2H assay failed to detect a well-characterized interaction was the p53:SV40TAg complex (data not shown). We speculate that the dodecameric structure of the SV40TAg:p53 complex 
may be incompatible with the topology of the LOC and CAT fusions or that the high molecular weight complex interferes with correct trafficking of the Och1p fusion proteins, a phenomenon that was observed when another large oligomeric complex was ectopically localized to the secretory pathway 
. If one of these hypotheses is correct, modifications to the G2H may be necessary to adapt it to analyses of proteins that oligomerize into very high molecular weight complexes. For example, inserting longer linkers between LOC and CAT domains and the bait and prey proteins may accommodate complex interaction geometry. Alternatively, using an ER-resident glycosyltransferase, rather than a Golgi-resident one, may enable detection of complexes that cannot exit the ER.
The experiments presented here describe a qualitative relationship between protein-protein interaction affinity and the signal observed in the G2H assay: the interaction between MyoD and Id2 produces strong signals, while introduction of mutations designed to disrupt this interaction decreases the strength of the phenotypic readouts. More comprehensive analysis will be needed to determine whether the signals observed in the growth and WGA binding assays are directly correlated with interaction affinity. A systematic analysis of interactions with varying affinities will enable us to answer this question and to assess the full dynamic range of this new assay.
The use of a glycosyltransferase, rather than a transcription factor, as a reporter enables new screening and selection methods. The Och1p reporter system described here relies on phenotypic changes observed in och1Δ yeast. The growth assay is simple to implement and its sensitivity can be adjusted by altering the concentration of Congo red. The WGA binding assay sensitively detects different levels of Och1p activity and, in principle, could be incorporated into a fluorescence-activated cell sorting (FACS) experiment to separate yeast with an active Och1p from those in which the protein is not reassembled.
By relying on a glycosyltransferase reporter, we envision that the G2H could also be adapted for use in other eukaryotic cells; all that is required is a modular reporter glycosyltransferase that causes a measurable cell surface change. Large families of glycosyltransferases occur in all eukaryotes, with 171 of these enzymes identified in humans 
. In addition the yeast Pichia pastoris
has recently been engineered to have human-like glycosylation patterns 
and may have a secretory pathway better suited to discovering novel mammalian secretome protein-protein interactions 
. For example, one could imagine using a G2H assay that incorporates human-like glycosylation to discover protein ligands for orphan cell surface receptors. Indeed, interactions among extracellular and cell surface proteins are poorly represented in existing protein-protein interaction databases 
and new methods are needed to enable their discovery.
More broadly, glycosyltransferase activity has the potential to be more widely exploited for screening and selection experiments. The utility of glycosyltransferases stems from two key features. First, they are modular enzymes that can be reassembled from their component parts. Second, they have the ability to provide an extracellular report of intracellular events: the activity of secretory pathway glycosyltransferases occurs within the cell, but results in dramatic changes on the cell surface. In the same way that the transcription-based Y2H assay has been adapted to new uses, such as the discovery of protease substrates 
and of protein-protein interactions that depend on post-translational modifications such as acetylation and phosphorylation 
, we anticipate the G2H has the potential to be used to report on biological events beyond simple protein-protein recognition.