PDZ domains have long been thought to cluster into discrete functional classes defined by their peptide-binding preferences. We used protein microarrays and quantitative fluorescence polarization to characterize the binding selectivity of 157 mouse PDZ domains with respect to 217 genome-encoded peptides. We then trained a multidomain selectivity model to predict PDZ domain-peptide interactions across the mouse proteome with an accuracy that exceeds many large-scale, experimental investigations of protein-protein interactions. Contrary to the current paradigm, PDZ domains do not fall into discrete classes; instead, they are evenly distributed throughout selectivity space, which suggests that they have been optimized across the proteome to minimize cross-reactivity. We predict that focusing on families of interaction domains, which facilitates the integration of experimentation and modeling, will play an increasingly important role in future investigations of protein function.

Evolvability—the capacity to generate beneficial heritable variation—is a central property of biological systems. However, its origins and modulation by environmental factors have not been examined systematically. Here, we analyze the fitness effects of all single mutations in TEM-1 β-lactamase (4,997 variants) under selection for the wild-type function (ampicillin resistance) and for a new function (cefotaxime resistance). Tolerance to mutation in this enzyme is bimodal and dependent on the strength of purifying selection in vivo, a result that derives from a steep non-linear ampicillin-dependent relationship between biochemical activity and fitness. Interestingly, cefotaxime resistance emerges from mutations that are neutral at low levels of ampicillin but deleterious at high levels; thus the capacity to evolve new function also depends on the strength of selection. The key property controlling evolvability is an excess of enzymatic activity relative to the strength of selection, suggesting that fluctuating environments might select for high-activity enzymes.

PDZ domains constitute one of the largest families of interaction domains and function by binding the C termini of their target proteins 1,2 . Using Bayesian estimation, we constructed a threedimensional extension of a position-specific scoring matrix that predicts to which peptides a PDZ domain will bind, given the primary sequences of the PDZ domain and the peptides. The model, which was trained using interaction data from 82 PDZ domains and 93 peptides encoded in the mouse genome 3 , successfully predicts interactions involving other mouse PDZ domains, as well as PDZ domains from Drosophila melanogaster and, to a lesser extent, PDZ domains from Caenorhabditis elegans. The model also predicts the differential effects of point mutations in peptide ligands on their PDZ domain-binding affinities. Overall, we show that our approach captures, in a single model, the binding selectivity of the PDZ domain family.Most efforts to define the binding selectivity of an interaction domain report either a consensus sequence for the domain's peptide ligands 4-6 or a position-specific scoring matrix that captures the domain's binding preferences 7-9 . Although these approaches are clearly useful, they are based on experimental data that are specific to the domain being studied and so are silent with respect to other members of the domain family. A truly general model-one that could be used to predict interactions involving PDZ domains for which no data are available -would take into account the sequence not only of the peptide, but also of the PDZ domain. We reasoned that, if the amino acid identity at specific positions in the PDZ domain's threedimensional structure determines that domain's preferences for amino acids at specific positions in the peptide ligand, it might be possible to capture this information for the entire PDZ domain family in a single model by integrating sequence information, structural information and protein interaction data (Fig. 1a).