Identifying peptides that bind specifically to protein targets requires either large, diverse libraries or small, focused libraries based on detailed prior knowledge of the target. In this study, we took a different approach. Recognizing that biological surfaces are composed of a unique complex mixture of lipids, polysaccharides, and proteins rich in potential peptide binding sites 
, we hypothesized that a rationally-designed matrix of peptides with ranging biochemical characteristics varying with defined periodicities would be sufficient to interact with these surfaces. We have demonstrated here that a library of as few as 36 peptides, when designed with a sufficient diversity of charge and hydrophobic distributions, can be used to successfully identify lead peptides that actively bind to biological surfaces.
Our hypothesis states that significant binding activity against biological surfaces can be found in a very small library: however, this requires that the level of specificity in any given hit should be relatively low. As we have shown, refinement of these hits by generating similar sequences in an ordered fashion allows much higher levels of specificity to be achieved. The refinement mechanism that we chose was suggested by the directionality of biochemical characteristics varied in the matrix, in that identification of a peptide with modest binding activity actually defines a set of new peptide sequences, bounded by the sequences on either side of the “hit,” that are likely to bind with equal or greater activity. Including orthogonal refinement in this process allowed us to consider other possible periodicities, providing refinement of both the magnitude and the spatial charge distribution of the candidate peptides.
Given the success we encountered in targeting bacterial surfaces, it became obvious that the pilot matrix could also be used for the detection and identification of bacterial strains. Because the library consists of peptides that vary greatly in their charge and hydrophobicity, it is likely that any given bacterium will show some level of interaction with at least one of the peptides. At the same time, the variety that exists among bacterial surfaces ensures that any single peptide will rarely show an identical level of binding to two different bacteria, and thus the relative level of binding of peptides within the library should provide a unique “fingerprint” for each species or strain. Microarray methods have been previously proposed for the identification of microorganisms, primarily utilizing arrays of antibodies against species-specific surface proteins 
. While this approach allows the robust and highly sensitive detection of well characterized pathogens, its usefulness is limited to the identification of pathogens that are well enough known to have had antibodies derived against them; emerging pathogens are unlikely to be identified. Fingerprinting methods, by comparison, rely on the differential interaction of compounds in a library with the desired target and have been demonstrated to differentiate between specific proteins using libraries of 100–1000 compounds 
. The representative bacterial surface binding profiles presented in this report suggest that each bacterium does show a distinctive binding profile, and the development of a set of profiles for known bacteria will likely allow the development of this technique for the rapid identification of unknown and emerging bacterial strains and will merit further development as diagnostic elements.
Interestingly, applying the sparse matrix to a sectioned human tooth revealed the presence of peptides that bound specifically to tooth enamel or dentin. This may reflect the fact that mineral surfaces, while composed of a relatively restricted set of features (such as charged regions, hydrophobic regions, and topological elements), present those features in such a way as to allow multiple peptide binding modes. This makes the problem of binding to bioinorganic surfaces accessible, in principle, to very small peptide libraries such as the one presented here, where distinctions can be made between such similar surfaces as the differing forms of hydroxyapatite present in dentin and enamel. It is intriguing that the peptide sequence identified in this experiment does not resemble known mineral binding motifs, which generally make use of repetitive sequences rich in Asp or Glu to interact with the exposed positively charged calcium ions present on the crystal surface 
. Interestingly, peptides E4 and E6 share identical hydrophobicity, but differ widely in charge: E4 is uncharged, while peptide E6 carries a net charge of -2. These parameters suggest a corresponding difference in charge density, but not necessarily the hydrophobicity, of these two distinct tissue layers of the tooth 
. The identification of an uncharged tissue-specific mineral binding peptide opens up new possibilities in the consideration of mechanisms by which peptides may interact with inorganic surfaces. Extending this method to other surfaces may allow us to identify additional novel sequences to bind to minerals, polymers, or metals.
The ability to identify compounds that bind to specific surfaces is central to the development of therapeutics, diagnostics, and imaging agents that can target bacterial surfaces, mineralized tissues, and implanted devices. The sparse-matrix method described here places the means of producing and identifying these compounds well within the reach of modern high capacity peptide synthesizers. By utilizing simple free-solution screening methods, highly specific surface-binding peptides can be identified without the complex deconvolution schemes or high-throughput screening equipment required by large random libraries. By providing a simple and rapid means of developing peptides that specifically bind to desired surfaces, the sparse matrix method may provide a step forward in the development of rapid diagnostics and targeted therapies.