One Proline, Two Lost Charges: Simulations Reveal Divergent Early Paths of Amyloid-Beta Fragments

A short slice of the amyloid-beta peptide has become a surprisingly revealing test bed for one of the most stubborn questions in Alzheimer’s research: how do tiny chemical changes at the molecular level steer a protein fragment toward, or away from, the beta-sheet structures that seed toxic plaques? A new molecular dynamics study published in Discover Chemistry examines the ten-residue fragment Aβ(14–23) and shows that two very different mutations produce two very different early structural behaviors — even when the fragment contains the same hydrophobic core that drives amyloid assembly.

The fragment in question, with the sequence HQKLVFFAED, is a compact piece of the full amyloid-beta protein. It contains the central hydrophobic LVFFA motif, widely regarded as a self-recognition region that helps amyloid chains stick together, followed by two negatively charged residues, glutamate 22 and aspartate 23. Decades of experimental work have shown that this region is intimately involved in aggregation: the related seven-residue Aβ(16–22) fragment forms antiparallel beta-rich assemblies, and structures of longer amyloid-beta peptides place the central hydrophobic segment squarely within the beta-strands of mature fibrils.

To probe how local chemistry shapes early assembly, the study compared three versions of the fragment in six-peptide clusters, or hexamers. The wild-type sequence served as the reference. The F20P variant swapped the phenylalanine at position 20 for proline, a residue famous for its rigid backbone geometry and its inability to donate a hydrogen bond through an amide group. The E22A–D23A variant replaced both acidic residues with alanines, stripping away two negative charges per chain while leaving the backbone free. These were chosen as deliberately contrasting physicochemical perturbations rather than as models of familial disease mutations, though substitutions at positions 22 and 23 in longer amyloid-beta systems are known to alter aggregation and fibril architecture.

Each system was built from a single packed arrangement of six peptides generated with PEP-FOLD4 and PACKMOL, solvated in explicit TIP3P water with 0.15 M NaCl, and simulated for 100 nanoseconds with GROMACS using the AMBER99SB-ILDN force field at physiological temperature. Three independent velocity replicates were run for each sequence, giving nine trajectories in total. The author is careful about what this design can and cannot show: because every replicate began from the same coordinates, the simulations measure sensitivity to velocity assignment but not to alternative monomer conformations or packing arrangements, and they do not establish spontaneous oligomerization or fibril formation.

The clearest signal emerged from secondary-structure analysis. In the final 20 nanoseconds of simulation, the wild-type hexamers averaged a beta-content of 0.284, meaning more than a quarter of peptide residues adopted extended-strand or beta-bridge conformations. The double charge-removal variant E22A–D23A came in at an intermediate 0.166, while the proline mutant F20P lagged far behind at 0.083. Residue-level analysis sharpened the picture: wild-type beta-propensity peaked around residues 18 and 19, in the heart of the hydrophobic self-recognition region, whereas the proline mutant sampled low beta-propensity across residues 16 through 20. This is chemically consistent with proline’s restricted backbone sampling and its missing amide hydrogen donor, and it echoes experimental proline-scanning studies showing that proline substitutions reduce fibril formation in amyloid-derived sequences.

But the story is not simply that proline suppresses beta-structure — that much was already known. The more interesting finding is that the two perturbations decoupled properties that might naively be expected to move together. The E22A–D23A hexamers combined intermediate beta-content with the largest radius of gyration of the three systems, 2.41 nanometers in the late window, and with nearly complete operational six-chain connectivity: in the audited contact-graph analysis, 99.9 percent of late-window frames had all six chains connected, compared with 86.9 percent for wild type and only 68.5 percent for F20P. Connectivity was defined operationally, with two chains considered linked when at least one interchain alpha-carbon contact fell within 0.80 nanometers.

That F20P average concealed dramatic replicate dependence. Two of its three trajectories remained fully connected throughout the late window, but one trajectory spent only 5.4 percent of its frames with all six chains in a single connected component, frequently sampling a four-chain largest component instead. Wild type showed a milder version of the same phenomenon, with one replicate fully connected in only 60.7 percent of frames. The lesson is sobering: low beta-content, compact global geometry, and complete aggregate connectivity are not interchangeable properties, and system-level averages can mask trajectory-level behavior that would be invisible in a single simulation.

Hydrogen-bond analysis added another layer of nuance. Raw backbone hydrogen-bond counts in the late window were highest for wild type at 13.45 interpeptide bonds on average, followed by E22A–D23A at 10.42 and F20P at 9.21. Because proline removes one backbone donor per chain, the author normalized interpeptide bonds by the number of available donor nitrogen atoms, yielding 0.224 for wild type, 0.174 for E22A–D23A, and 0.171 for F20P — a much narrower spread that underscores how easily unnormalized counts can mislead. Salt-bridge counts between lysine and carboxylate groups also differed across systems, and a full-system ion-coordinate reanalysis revealed system-dependent local distributions of sodium and chloride ions around defined peptide sites, including markedly lower sodium occupancy at the C-terminal carboxylate in the charge-removed variant. The author is explicit that these ion patterns are consistent with the altered charge environments but do not establish an ion-specific causal mechanism.

Conformational clustering of the late-window frames reinforced the theme of heterogeneity. Wild-type trajectories varied substantially in how concentrated their sampled coordinates were, with dominant clusters containing anywhere from 9 to 51 percent of frames. The proline mutant was even more extreme, with one trajectory concentrating 92 percent of frames in a single cluster while another spread its frames across hundreds of clusters. The charge-removal variant produced 235 to 266 clusters per trajectory with dominant-cluster fractions of only 7 to 26 percent, consistent with its extended, diffuse contact maps. None of these clustering results should be read as thermodynamic stability or equilibrium populations — the author repeatedly cautions that they describe concentration within the sampled coordinates of individual trajectories only.

The study’s limitations are stated with unusual candor, and they matter for how the results should be used. Each sequence was represented by a single packed coordinate set, so starting-structure bias cannot be separated from mutation-specific effects. The AMBER99SB-ILDN force field was not optimized for disordered amyloid ensembles, and amyloid simulations are known to be force-field dependent, so the observed ordering is force-field-specific. The 100-nanosecond trajectories are far too short to establish equilibrium populations, and the contact-graph connectivity is an operational geometric definition, not a measure of thermodynamic stability. The author frames the work as a controlled, hypothesis-generating comparison that requires independent packing replicates, second-force-field validation, longer or enhanced sampling, and experimental measurements before any of the trends can be considered robust.

Even with those caveats, the central insight stands: a backbone-disrupting substitution and a double charge-removal substitution did not produce one common structural response in these early-stage hexamers. Wild type sampled the highest beta-content and the strongest donor-normalized interpeptide hydrogen bonding; the charge-removal variant combined intermediate beta-structure with greater spatial extension and near-complete connectivity; and the proline mutant sampled the lowest beta-content with the greatest replicate-to-replicate variability in connectivity. For a field where toxic oligomers are believed to form through fleeting, hard-to-capture intermediates, the message is that the earliest structural tendencies of amyloid fragments are exquisitely sensitive to local sequence chemistry — and that understanding them will require exactly the kind of matched, carefully audited, statistically honest comparisons this study sets out to model.

Subject of Research: Early structural behavior of wild-type and mutant amyloid-beta 14–23 hexamers in molecular dynamics simulations

Article Title: Early structural tendencies differ among amyloid beta 14 to 23 hexamers with F20P and E22A D23A substitutions

Article References: Zein, H. F. (2026). Early structural tendencies differ among amyloid beta 14 to 23 hexamers with F20P and E22A D23A substitutions. Discover Chemistry, 3(1), Article 540. https://doi.org/10.1007/s44371-026-00998-7

Image Credits: AI Generated

DOI: 10.1007/s44371-026-00998-7

Keywords: amyloid-beta, Aβ(14–23), molecular dynamics, F20P, E22A–D23A, beta-strand propensity, peptide aggregation, Alzheimer's disease, protein structure, hydrogen bonds, salt bridges, force field dependence