Antibody targets are characterized using the two-hybrid system with the scFv antibody as bait, a validated scoring system to remove non-specific binders, and determination of selected interacting domains
Investigators studying and selecting recombinant antibodies have mostly used the Y2 H technology to select antibody populations that fold correctly under the reducing conditions of the cytosol [26–31]. In such experiments, the epitope of a given antigen is defined and used as a bait to sample a diverse scFv prey library. Here, we reversed the system, using as baits known scFv antibodies that had been previously selected in phage display screens, and, as preys, proteins expressed from high-complexity libraries of randomly primed cDNA (Figure 1A).
Individual scFv antibodies were exposed to a cDNA library with full coverage, derived from either human placenta or from Drosophila. More than 50 million potential interactions were tested during each screen. Prey fragments of positive clones were identified using sequence analysis and comparison with GenBank databases using the BLAST algorithm [32]. Next, a Predicted Biological Score (PBS) was computed for each clone as described before [33], enabling classification according to interaction reliability (Figure 1B). This permitted ranking of the clones into categories from "A" to "D" in decreasing probability of having a specific interaction with the scFv-bait. Two additional categories were designated: "E", interacting clones, which were likely retained due to highly connected prey domains; and "F", clones that were experimentally proven Y2 H artifacts. In a third step, overlapping prey fragments originating from the same gene were designated as clusters and their translated amino acid sequences were aligned and superimposed onto the open reading frame (Figure 1C). Overlapping regions shared by all fragments were designated as "selected interacting domain" (SID), as described before [33].
For the conformation-specific antibodies AA2 and ROF7, only large fragments are obtained, suggesting a non-linear antibody binding-site
In the first screen, the scFv AA2 was used as bait. AA2 is a well-described conformation-specific antibody that exclusively binds the activated (GTP bound) form of the small GTPase Rab6 [17]. From the human cDNA prey library only three specific binders were retained by Y2 H (Additional file 1). Two (67%) encoded for Rab6 (one for Rab6a and one for Rab6b, Figure 2A, top). The screen was repeated using a Drosophila cDNA library. Eight specific clones were retained (Additional file 2), four (50%) representing Drosophila Rab6 (dRab6) (Figure 2A, bottom).
The result of the Drosophila screen suggested that AA2, which had been selected against human Rab6a, can also recognize the fly Rab6a homologue. In our experience, polyclonal anti-Rab6 antibodies (generated by immunizing rabbits with the full-length mammalian protein) have failed to label dRab6 in immunofluorescence (unpublished observations). In contrast, and in agreement with the Y2 H data, AA2 labeling of dRab6 on Golgi membranes was seen by immunofluorescence (Figure 2B, a-c). Furthermore, analogous to what had been shown in mammalian cells [17], AA2 was also functional as an intrabody and labeled Golgi stacks in living Drosophila S2 cells (Figure 2B, d-f). Our Y2 H approach thus revealed that in addition to mammalian Rab6, AA2 also detects the Drosophila homologue. This is likely due to the fact that the three-dimensional structure of GTP-bound Rab6 remained highly conserved throughout evolution.
In both Y2 H screens using AA2 as bait, only prey fragments spanning most of the ORF were retained (Figure 2A, and Additional files 1 and 2). Apparently, the entire core region of Rab6 had to be expressed to allow binding of this conformation-sensitive antibody. This further suggested that this method of target determination could validate non-linear epitopes. To confirm this, we next characterized the target of another recently obtained scFv, which was also observed to be conformation specific (own unpublished observations). This antibody, called ROF7, detects the small GTPase Rab1a and/or Rab1b only after their activation through GTP-loading. When the screen was performed using ROF7 as bait, a total of 191 specifically interacting clones were retained from the mammalian cDNA library. Eighty percent (152 clones) encoded for Rab1a or Rab1b, once again all encompassing the full core region of the ORF (Figure 3 and Additional file 3).
In the two Y2 H screens using conformational sensors as bait (AA2 and ROF7) and comprehensive prey libraries, only targets expressing a sizable portion of the protein (amino acids 2-178 for Rab1a, and 13-174 for Rab6a) were retained. Likely, fragments had to be large enough to allow folding into proper tertiary structure and GTP loading. The first few amino-terminal residues and the last 30 or so amino-acid long carboxy-terminal hypervariable tail are known to be dispensable for correct three-dimensional assembly of small GTPases. In fact, the crystal structure of Rab6b was solved by expressing a recombinant protein that contained only the core region of the protein (amino acids 6-181) [34]. We believe that the residues, which make up the epitopes for the binding of AA2 and ROF7, respectively, are non-adjacent in the primary amino acid sequence and only come together after correct folding of the polypeptide into tertiary structure and subsequent activation through GTP binding. Lack of reactivity of both antibodies on immunoblotting, and results from the Y2 H screens strongly supports this hypothesis.
In summary, our approach showed that for the conformation-specific sensors AA2 and ROF7 correct three-dimensional antigen structures are needed to present the antibody binding sites. It furthermore revealed that the AA2 epitope is conserved in evolution since dRab6 is recognized by the scFv.
For the anti-giantin antibody TA10, a small binding site within the large coiled-coil protein giantin is determined
Next, we characterized the target of TA10. This scFv is a recombinant antibody directed against a very large (>350 kDa) Golgi-associated matrix protein called giantin. TA10 was originally selected via phage display using intact, purified, Golgi stacks as target [16].
The mammalian cDNA prey library was screened using TA10 as bait. Sixty interacting clones were retained (Additional file 4). Only one cluster, containing seven clones, obtained a high PBS ranking; they all fell within the central portion of the giantin ORF. Alignment of the seven clones defined a ~28 kDa-large SID (Figure 4A) located in the center portion of giantin. Binding of TA10 within this region of 250 amino acids was confirmed via overexpression experiments (Figure 4B).
TA10 is a blotting antibody and thus, in contrast to AA2 and ROF7, should detect the denatured protein at a small linear binding site. Taking advantage of the flexibility of yeast genetics, we therefore attempted to narrow the binding domain of TA10 further using a method based on gap repair. Primers were designed along the shortest of the seven prey fragments and PCR fragments were generated (Figure 5A). The Y2 H screen was then repeated, this time using the resulting small PCR products as prey (Figure 5B). In this manner, the TA10 binding region could be narrowed further to a single coiled coil domain measuring 9kDa (fragment "a" in Figure 5C). Binding of TA10 to this region was confirmed in overexpression experiments using both immunofluorescence and Western blotting (Figure 5D, E). Finally, carrying out additional biochemical and overexpression experiments we narrowed the antibody-binding region even further (within a 30 amino acid stretch, data not shown).
In summary, the Y2 H screen enabled rapid and accurate target characterization for TA10. A precise region of 79 amino acids, representing less than 2.5% of the entire protein, was determined as its binding site.
For the non-muscle myosin IIA-targeting antibody SF9, a precise epitope within a long coiled-coil protein is determined and cross reactivity with human and other, non-human, homologues is revealed
SF9 was selected alongside TA10 in a phage display screen using purified rat liver Golgi stacks as the antigen. SF9's target, non-muscle myosin IIA, was identified using immunoblotting followed by mass spectrometry analysis [16].
From the Y2 H screen with SF9 as bait a total of 352 specifically interacting clones were retained (Additional file 5). More than half (196 clones) encoded for non-muscle myosin IIA and their alignment allowed identification of a very small SID of only 35 amino acids (Figure 6A). An additional 116 clones represented one of two very closely related myosins: non-muscle myosin IIB and smooth-muscle myosin (55 and 61 clones, respectively). Their alignments yielded similarly precise SID (Figure 6B, C). All three SID overlapped with one another inside a conserved stretch of the tail region of the myosin protein family, further narrowing the putative binding domain for SF9 (Figure 6D).
Binding of the antibody to the 29-amino-acid long common SID was confirmed in a series of overexpression experiments. GFP-tagged non-muscle myosin IIA constructs, which either included or lacked the SID (Figure 7A), were overexpressed in mammalian cells. In addition to the endogenous myosin pool, only the overexpressed full-length but not the truncated protein was recognized in immunofluorescence by SF9 (Figure 7B). Immunoblot analysis revealed that a recombinant protein containing the 29-amino-acid long SID was detected while no binding was seen with the construct lacking this short region (Figure 7C). Together, this confirmed our Y2 H results.
Non-muscle myosins are evolutionarily conserved proteins. In addition, SF9 was found to bind to a region of the protein that shares near-identical amino acid sequences between human homologues (Figure 6D). Alignment of the SID with non-muscle myosins of other species revealed a high degree of evolutionary conservation in this region even with the Drosophila counterpart (Figure 7D). We thus speculated that SF9 might also detect the insect form of myosin. Indeed, when using total insect cell lysates a single ca. 200-kDa-large band was detected by SF9. This signal completely disappeared after specific anti-myosin RNAi-mediated gene silencing (Figure 7E).
In summary, our Y2H-approach allowed precise characterization of the SF9 target, defining a very small (<3kDa) epitope within a ~200kDa protein. In addition, it revealed cross-reactivity with closely related proteins of the same and across species.
Results of screens using antibodies that do not work as intrabodies
Our novel approach yielded important additional information about the targets of several distinct monoclonal antibodies. However, some variability in the strength and reliability of the data could be noted when comparing individual screens. For instance, when screening using yet another scFv, named F2C (an antibody which binds α-tubulin [16]), results were less conclusive. While a total of 380 interacting clones were obtained, only 4 encoded for α-tubulin (data not shown). Their alignment allowed narrowing of the epitope-containing region to the C-terminal portion of the protein. However, many of the remaining 376 clones were also classified as possible binders using the PBS scoring method. Nonetheless, it is unlikely that they represent secondary targets for F2C given a highly specific signal of F2C in both immunofluorescence and Western blotting [16]. Another scFv originating from the above mentioned Golgi-stack screen, called TE5 [16], was tested in our approach. While many interacting clones were retained, none could be confirmed in overexpression experiments and the TE5 target remained elusive so far (data not shown). In sharp contrast to the just mentioned examples (F2C, TE5) our ROF7- and SF9-screens not only confirmed the antibodies' targets but also gave detailed information about their respective epitopes: in each case almost 90% of clones aligned within its target(s) (see Figures 3 and 6). In agreement with the more classical use of scFv in Y2 H [26] there seems to be a direct correlation between the ability of an antibody to fold correctly under the reducing conditions of the eukaryotic (yeast or mammalian) cytosol and their use in Y2 H. While SF9, TA10, AA2 and ROF7 efficiently bind their respective targets when expressed as intrabodies, F2C is only faintly staining microtubules in living cells and TE5 never yielded specific staining as an intrabody. We thus hypothesize that the same relative efficiency for target recognition is present during the yeast two hybrid experiments. To make full use of our method, it will therefore be important to obtain good intrabodies, for example via screening of dedicated libraries [35, 36] or through the use of single domain antibodies, like camelidae antibodies, that seem to be more resistant to the reducing conditions of the intracellular milieu [37].