The WISE strategy described here has the merit of combining the strengths of a selective approach (such as panning combinatorial peptide libraries displayed on phage) with a quantitative analysis that can be achieved by screening a large number of peptides arrayed at high density on a solid support. This makes it possible to identify rapidly and directly the tighest ligands of a peptide-binding receptor among all the peptides in an entire proteome. We have demonstrated the approach by applying it to the family of SH3 domains. However, WISE can also be extended to all those domain families (WW, PDZ, EH, GYF, VHS, SH2 PTB, 14-3-3, FHA, WD40, etc.) that mostly recognize short peptides in their partner proteins.
Our approach, as any in vitro approach, suffers from some simplifications when it comes to inferring the physiological partners from the domain–peptide interaction data. According to a naive strategy, we would assimilate the cell to a cellulose membrane, where all the peptides are equally represented and accessible to the bait domain, and we would be tempted to conclude that all the proteins containing the identified peptide ligands were likely to be physiological partners. In the real cell, however, the target peptides may be hidden inside the core of the folded proteins, and the protein partners may not be equally represented. Furthermore, the partner proteins may be expressed in different cell types or segregated in different macromolecular complexes or cell compartments. In order to obtain more reliable inferences, the peptide interaction information obtained by a WISE approach should be complemented by information about peptide accessibility obtained by structural predictors (Garner et al. 1998
; Linding et al. 2003
) and data about mRNA and protein concentrations in different physiological and subcellular contexts (Simpson et al. 2000
; Kumar et al. 2002
Nevertheless, the average number of peptides in the yeast proteome that have the potential to bind SH3 domains with an affinity that may have physiological relevance was found to be surprisingly high, ranging from a few peptides, in the case of the Abp1 and Boi2 SH3 domains, to several tenths, in the case, for instance, of the Yfr024w SH3 domain. Given the hypothesis that all (or most of) these peptides are equally expressed inside the cell and exposed to the solvent in the folded protein structure as most Pro-rich peptides are, these findings raise the question of whether the observed binding promiscuity has any physiological implication. Recent proteome-wide analyses of yeast protein complexes have revealed that many proteins are organized in discrete complexes (Gavin et al. 2002
; Ho et al. 2002
). Yet this approach has failed to identify a large number of interactions whose physiological relevance was validated by traditional single (or few) protein studies, implying that many physiologically relevant protein interactions do not lead to the formation of stable complexes. SH3-mediated interactions may belong to this latter class. This is consistent with the observation that SH3-containing proteins have a connectivity significantly lower than average (2.33; average, 4.00) in the yeast complexosome (Gavin et al. 2002
; Ho et al. 2002
), in contrast with the observed connectivity in the interaction network derived from high-throughput two-hybrid experiments (average connectivity of SH3-containing proteins, 5.05; average connectivity for all proteins, 1.53) (Uetz et al. 2000
; Ito et al. 2001
; L. Montecchi-Palazzi and G. Cesareni, unpublished data). SH3-mediated interactions are much less likely to be detected by coimmunoprecipitation assays than by solid-state (or two-hybrid) assays, because relatively weak interactions are almost certainly lost in the extensive washing needed for coimmunoprecipitation experiments. Our approach has made it possible to rediscover most of the SH3-mediated protein interactions that were previously described for these proteins. Admittedly, though, few clearly characterized protein interactions of this type have yet been reported in the literature. The few failures of our approach (false negatives) are due to weaknesses in the design of the relaxed consensus used to search for matching peptides in the protein databases.
All the same, we have identified a larger number of target peptides that bind with affinities that are comparable to the ones of the validated physiological targets. Some of these peptides may never encounter the cognate SH3 domain, while some will only meet partners in specific physiological conditions. Others may contribute to add specificity to the formation of a complex by cooperating with other associated low-specificity binding domains. Finally, we have to consider a new scenario in which proteins, even when not forming stable complexes, are seldom isolated in solution, but navigate in the cell by moving from one weak partner to another. These weak interactions may be important in modulating cell architecture even when they are not instrumental in the nucleation of a stable complex. Although this is difficult to prove, the semiquantitative data provided by our approach, complemented with the results of large-scale expression and localization studies, may eventually allow one to model these different settings.
The in vitro approach that we have described, albeit limited to interactions in which one of the partners can be reduced to a relatively short peptide, presents a number of interesting features that complement other strategies aimed at revealing the details of the protein interaction network within cells.
First, it takes full advantage both of the genomic information that is being accumulated and of the array format in which all the possible targets are equally represented. Second, it is comprehensive and provides a high level of detail on the interaction topology. Third, it is not affected by protein concentrations inside the cell and is very sensitive (interactions up to 100 μM can be detected). Fourth, interactions that depend on peptide modifications, for instance, phosphorylation, can also be studied. Fifth, the output is semiquantitative. Finally, the identified target peptide can be used as a lead to develop tighter binding molecules in order to interfere with complex formation in vivo.
We have shown that the current implementation of the SPOT synthesis technology is sufficient to carry out WISE screening of a proteome as complex as the one of a mammalian organism. Foreseeable technological improvements of the SPOT synthesis technology will permit the assembly of relatively cheap microarrays containing up to 15,000 peptides. This will extend the approach's power by relieving, in some cases, the requirement for an experimental filtering step, as performed here by the phage display approach, thereby allowing more freedom in the design of the relaxed pattern.