Emmanuel Lacroix

De novo designed peptide-based amyloid fibrils

Identification of therapeutic strategies to prevent or cure diseases associated with amyloid fibril deposition in tissue (Alzheimers disease, spongiform encephalopathies, etc.) requires a rational understanding of the driving forces involved in the formation of these organized assemblies rich in β-sheet structure. To this end, we used a computer-designed algorithm to search for hexapeptide sequences with a high propensity to form homopolymeric β-sheets. Sequences predicted to be highly favorable on this basis were found experimentally to self-associate efficiently into β-sheets, whereas point mutations predicted to be unfavorable for this structure inhibited polymerization. However, the property to form polymeric β-sheets is not a sufficient requirement for fibril formation because, under the conditions used here, preformed β-sheets from these peptides with charged residues form well defined fibrils only if the total net charge of the molecule is ±1. This finding illustrates the delicate balance of interactions involved in the formation of fibrils relative to more disordered aggregates. The present results, in conjunction with x-ray fiber diffraction, electron microscopy, and Fourier transform infrared measurements, have allowed us to propose a detailed structural model of the fibrils.

Computer-aided design of a PDZ domain to recognize new target sequences.

PDZ domains are small globular domains that recognize the last 4–7 amino acids at the C-terminus of target proteins. The specificity of the PDZ–ligand recognition is due to side chain–side chain interactions, as well as the positioning of an α-helix involved in ligand binding. We have used computer-aided protein design to produce mutant versions of a Class I PDZ domain that bind to novel Class I and Class II target sequences both in vitro and in vivo, thus providing an alternative to primary antibodies in western blotting, affinity chromatography and pull-down experiments. Our results suggest that by combining different backbone templates with computer-aided protein design, PDZ domains could be engineered to specifically recognize a large number of proteins.

The design of linear peptides that fold as monomeric β-sheet structures

Current knowledge about the determinants of β-sheet formation has been notably improved by the structural and kinetic analysis of model peptides, by mutagenesis experiments in proteins and by the statistical analysis of the protein structure database (Protein Data Bank; PDB). In the past year, several peptides comprising natural and non-natural amino acids have been designed to fold as monomeric three-stranded β-sheets. In all these cases, the design strategy has involved both the statistical analysis of the protein structure database and empirical information obtained in model β-hairpin systems and in proteins. Only in one case was rotamer analysis performed to check for the compatibility of the sidechain packing. It is foreseeable that, in future designs, algorithms exploring the sequence and conformational space will be employed. For the design of small proteins (less than 30 amino acids), questions remain about the demonstration of two-state behavior, the formation of a well-defined network of mainchain hydrogen bonds and the quantification of the structured populations.

Conformational strain in the hydrophobic core and its implications for protein folding and design

We have designed de novo 13 divergent spectrin SH3 core sequences to determine their folding properties. Kinetic analysis of the variants with stability similar to that of the wild type protein shows accelerated unfolding and refolding rates compatible with a preferential stabilization of the transition state. This is most likely caused by conformational strain in the native state, as deletion of a methyl group (Ile→Val) leads to deceleration in unfolding and increased stability (up to 2 kcal mol−1). Several of these Ile→Val mutants have negative φ‡−U values, indicating that some noncanonical φ‡−U values might result from conformational strain. Thus, producing a stable protein does not necessarily mean that the design process has been entirely successful. Strained interactions could have been introduced, and a reduction in the buried volume could result in a large increase in stability and a reduction in unfolding rates.

Combinatorial approaches: A new tool to search for highly structured β-hairpin peptides

Here we present a combinatorial approach to evolve a stable β-hairpin fold in a linear peptide. Starting with a de novo-designed linear peptide that shows a β-hairpin structure population of around 30%, we selected four positions to build up a combinatorial library of 204 sequences. Deconvolution of the library using circular dichroism reduced such a sequence complexity to 36 defined sequences. Circular dichroism and NMR of these peptides resulted in the identification of two linear 14-aa-long peptides that in plain buffered solutions showed a percentage of β-hairpin structure higher than 70%. Our results show how combinatorial approaches can be used to obtain highly structured peptide sequences that could be used as templates in which functionality can be introduced.

Insights into the Origin of the Tendency of the PI3-SH3 Domain to form Amyloid Fibrils

The SH3 domain of the p85alpha subunit of phosphatidylinositol 3 kinase has been found to form amyloid fibrils in vitro under acidic conditions. PI3-SH3 is peculiar due to a large insertion of 15 amino acid residues in the n-Src loop when compared with more canonical members of the family. Spectrin-SH3 (SPC-SH3) with a shorter loop does not form fibrils under any of our conditions tested. Thus, it could be that the longer loop could play a role in amyloid formation. To investigate this we have engineered two chimeras containing the common core of the PI3-SH3 and SPC-SH3 with an exchanged n-Src loop. Thermodynamic and kinetic analyses show that the two chimeras are less stable than the parent proteins, but useful for our comparative purposes they have similar stability. Neither stability, nor folding rates, or pH transition can be invoked as being responsible for the amyloid formation in the PI3-SH3 domain. Substitution of the long n-Src loop in PI3-SH3 by that of SPC-SH3 does not prevent fibril formation. The SPC-SH3 with the PI3-SH3 n-Src loop is in an A-state at low pH and forms beta-sheet amorphous aggregates, but not amyloid fibrils. Thus, we conclude that, for a protein to form ordered fibrils, a delicate balance between solubility of non-native states to allow efficient nucleation and the formation of amorphous aggregates, must be achieved. It is the amino acid residue sequence of the protein and probably its parts that play a determinant role in shifting this balance in one direction or the other.

Reading protein sequences backwards

BACKGROUND Reading a protein sequence backwards provides a new polypeptide that does not align with its parent sequence. The foldability of this new sequence is questionable. On one hand, structure prediction at low resolution using lattice simulations for such a protein provided a model close to the native parent fold or to a topological mirror image of it. On the other hand, there is no experimental evidence yet to tell whether such a retro protein folds (and to which structure) or not. RESULTS In this work, we have analysed the possibility of a retro protein folding in two different ways. First, we modelled the retro sequence of the alpha-spectrin SH3 domain through distance geometry and molecular dynamics. This contradicted the plausibility of a mirror image of the native domain, whereas basic considerations opposed the likelihood of the native fold. Second, we obtained experimental evidence that the retro sequences of the SH3 domain, as well as the B domain of Staphylococcal protein A and the B1 domain of Streptococcal protein G, are unfolded proteins, even though some propensities for the formation of secondary structures might remain. CONCLUSIONS Retro proteins are no more similar to their parent sequences than any random sequence despite their common hydrophobic/hydrophilic pattern, global amino acid composition and possible tertiary contacts. Although simple folding models contribute to our global understanding of protein folding, they cannot yet be used to predict the structure of new proteins.

Computational estimation of specific side chain interaction energies in α helices

We have used a structure energy‐based computer program developed for protein design, Perla, to provide theoretical estimates of all specific side chain–side chain interaction energies occurring in α helices. The computed side chain–side chain interaction energies were used as substitutes for the corresponding values used by the helix/coil transition algorithm, AGADIR. Predictions of peptide helical contents were nearly as successful as those obtained with the originally calibrated set of parameters; a correlation to experimentally observed α‐helical populations of 0.91 proved that our theoretical estimates are reasonably correct for amino acid pairs that are frequent in our database of peptides. Furthermore, we have determined experimentally the previously uncharacterized interaction energies for Lys–Ile, Thr–Ile, and Phe–Ile amino acid pairs at i,i + 4 positions. The experimental values compare favorably with the computed theoretical estimates. Importantly, the computed values for Thr–Ile and Phe–Ile interactions are better than the energies based on chemical similarity, whereas for Lys–Ile they are similar. Thus, computational techniques can be used to provide precise energies for amino acid pairwise interactions, a fact that supports the development of structure energy–based computational tools for structure predictions and sequence design.

Computer-assisted re-design of spectrin SH3 residue clusters.

We have developed a protein design computer program, called Perla, which performs searches in sequence space to uncover optimal amino acid sequences for desired protein three-dimensional structures. Optimal sequences are localised at the minima of a sequence-structure energy landscape defined using a complex scoring function (an all-atom molecular mechanics force field plus statistical terms including entropy and solvation) measured with respect to a reference state simulating a denatured protein. Sequence choices eventually optimise side chain packing, secondary structure propensities, and hydrogen bonding and electrostatics interactions. Perla was used to re-design clusters of residues of the SH3 domain of alpha-spectrin. Several mutant proteins were produced and characterised. Some of our designed proteins have significantly higher stabilities (stability enhancements about 0.25, 0.70 and 1.0 kcal mol(-1)) than the wild-type protein. These successful protein re-designs, and similar examples found in the literature, establish the quality of the structure-based computational approach to protein design.