Optimizing Epitope Conformational Ensembles Using α-Synuclein Cyclic Peptide "Glycindel" Scaffolds: A Customized Immunogen Method for Generating Oligomer-Selective Antibodies for Parkinson's Disease

Abstract

Effectively presenting epitopes on immunogens, in order to raise conformationally selective antibodies through active immunization, is a central problem in treating protein misfolding diseases, particularly neurodegenerative diseases such as Alzheimer's disease or Parkinson's disease. We seek to selectively target conformations enriched in toxic, oligomeric propagating species while sparing the healthy forms of the protein that are often more abundant. To this end, we computationally modeled scaffolded epitopes in cyclic peptides by inserting/deleting a variable number of flanking glycines ("glycindels") to best mimic a misfolding-specific conformation of an epitope of α-synuclein enriched in the oligomer ensemble, as characterized by a region most readily disordered and solvent-exposed in a stressed, partially denatured protofibril. We screen and rank the cyclic peptide scaffolds of α-synuclein in silico based on their ensemble overlap properties with the fibril, oligomer-model and isolated monomer ensembles. We present experimental data of seeded aggregation that support nucleation rates consistent with computationally predicted cyclic peptide conformational similarity. We also introduce a method for screening against structured off-pathway targets in the human proteome by selecting scaffolds with minimal conformational similarity between their epitope and the same solvent-exposed primary sequence in structured human proteins. Different cyclic peptide scaffolds with variable numbers of glycines are predicted computationally to have markedly different conformational ensembles. Ensemble comparison and overlap were quantified by the Jensen-Shannon divergence and a new measure introduced here, the embedding depth, which determines the extent to which a given ensemble is subsumed by another ensemble and which may be a more useful measure in developing immunogens that confer conformational selectivity to an antibody.

Keywords: Cyclic peptides; ensemble similarity; epitope scaffolding; molecular dynamics; protein misfolding; virtual screening.

Figure 1

Cyclic peptide renderings for cyclo(CGTKEQGGGG), a scaffold of TKEQ. (a) 2D representation of the cyclic peptide. (b) Three-dimensional rendering of the cyclic peptide in licorice, also showing the surface of the TKEQ epitope. Colors are assigned by residue name, with glycine in white, cysteine in yellow, threonine in dark pink, lysine in cyan, glutamate in light pink, and glutamic acid in orange. (c) Ball and stick (CPK) rendering with color assigned by the atom identity.

Figure 2

Equilibrium ensemble distributions for the TKEQ epitope projected by the multidimensional scaling (MDS) method onto the first MDS dimension. For a given epitope, different cyclic peptide scaffolds possess different distributions, which will result in different overlap with the other three ensembles. By comparison of the degree of ensemble overlap, the conformational selectivity of a scaffold can be assessed. The scaffolds shown are (1,4)TKEQ in (a), and (2,3)TKEQ in (b).

Figure 3

Measures for the rankings of all 16 epitope scaffolds for three overlapping four-residue subepitopes of EKTKEQ in α-synuclein (top of each column). (a) Scaffolded cyclic peptide ensemble dissimilarity to monomer (triangle), fibril (star), and stressed fibril (circle) ensembles, as measured by Jensen–Shannon divergence (JSD), showing the changes in ensemble overlap with varying numbers of flanking glycines. (b) Scaffolded cyclic peptide ensemble embedding depth within the monomer (triangle), fibril (star), and stressed fibril (circle) ensembles, showing the changes in ensemble embedding with varying number of flanking glycines. (c) Normalized off-pathway targeting values (OP) for scaffolds with varying number of flanking glycines. Higher values indicate there is less predicted off-pathway targeting by a given scaffold. The ranks of the top 10 scaffolds are indicated in the figure panels, along with the rank for the highest ranking scaffold (15) for epitope KTKE.

Figure 4

Correlation matrices of the ensemble comparison metrics JSD and and three other scaffold properties: dynamic flexibility (RMSF) of an epitope, total residue length of the cyclic peptide scaffold, and the ranking, for α-synuclein epitopes (a) EKTK, (b) KTKE, and (c) TKEQ. The number in each square is the Pearson correlation coefficient.

Figure 5

Epitope-dependent correlation between (a) RMSF and scaffold length (number of residues in the scaffold), (b) ranking and scaffold length, (c) rank and the quantity , and (d) rank and , for α-synuclein epitope scaffolds. The Pearson correlation coefficient r and the corresponding p-values are given for EKTK (green triangles), KTKE (purple circles), and TKEQ (blue stars) scaffolds. The shaded areas around the fitted lines are the 68% confidence intervals corresponding to the standard errors.

Figure 6

Off-pathway target analysis for (1,4)TKEQ. (a) Structural ensemble distribution of cyclic peptide (1,4)TKEQ in 1D along the first MDS component of the ensemble, along with the projected embedding of potential off-pathway targets. Most of the off-pathway targets are located at the periphery of the scaffold distribution. The actual calculation is performed in 5D. Structures of PDB entries 2KKW and 1XQ8 are rendered in ribbon schematics, and the epitope is rendered in red van der Waals surface. (b) SASA distribution of (1,4)TKEQ, along with the SASA for the off-pathway target structures. (c) Only 1XQ8 and 2KKW show both noticeable structural similarity and SASA exposure (f(SASA_{off-target-exceed-cyclic}) > 5%).

Figure 7

Two replicate experiments (run 1 and run 2) of seeded aggregation, as probed by ThT fluorescence. Curve fitting is obtained from the AmyloFit server.

Figure 8

The α-synuclein epitope, EKTKEQ, is found to be highly exposed in five structurally distinct fibrils. (a–e) Averaged SASA as a function of residue position. A rolling average window of 6 amino acids was applied. The window that contains EKTKEQ (residues 57–62), as indicated by the red line, exceeds more than 80% of the other windows in all fibril structures analyzed. The shaded region contains the rolling average values for residues 57–62. In each panel, a single chain of each fibril structure is aligned and rendered to show their structure heterogeneity. (f) The average SASA across all 5 fibrils. The epitope region has the highest average SASA across the whole structured sequence. (g) The pairwise local distance test (lddt) shows that the analyzed fibrils are all mutually dissimilar.

Figure 9

Workflow of in silico screening.

Figure 10

Collective coordinate epitope prediction for α-synuclein, using three criteria of increased SASA, loss of native contacts, and increased fluctuations (RMSF). Several epitopes were predicted by each criterion; however, only a single consensus epitope EKTKEQ was predicted. Chain E is not shown for ΔSASA because no epitope is predicted.

Figure 11

Illustration of the embedding depth of the point x₀ in a multimodal distribution P(x). The embedding depth of point x₀ is given by the integral over all parts of the distribution with probability less than P(x₀). Note there are four points with the same P(x_i) = P(x₀) and thus the same embedding depth.