Influence of the Nonprotein Amino Acid Mimosine in Peptide Conformational Propensities from Novel Amber Force Field Parameters

Mimosine is a nonprotein amino acid derived from plants known for its ability to bind to divalent and trivalent metal cations such as Zn2+, Ni2+, Fe2+, or Al3+. This results in interesting antimicrobial and anticancer properties, which make mimosine a promising candidate for therapeutic applications. One possibility is to incorporate mimosine into synthetic short peptide drugs. However, how this amino acid affects the peptide structure is not well understood, reducing our ability to design effective therapeutic compounds. In this work, we used computer simulations to understand this question. We first built parameters for the mimosine residue to be used in combination with two classical force fields of the Amber family. Then, we used atomistic molecular dynamics simulations with the resulting parameter sets to evaluate the influence of mimosine in the structural propensities for this amino acid. We compared the results of these simulations with homologous peptides, where mimosine is replaced by either phenylalanine or tyrosine. We found that the strong dipole in mimosine induces a preference for conformations where the amino acid rings are stacked over more extended conformations. We validated our results using quantum mechanical calculations, which provide a robust foundation for the outcome of our classical simulations.


Charge derivation
The methodologies applied for the charge derivation in this work are a compromise to replicating the processes used by the authors on the force fields while working around some limitations imposed by the mimosine (Mms). The methodology corresponding to Amber 99sb used Amber 86 S1 for two consecutive optimizations and the methodology corresponding to Amber 03 used Amber 94 S2 for its first optimization. Since there are no parameters available for the Mms residue in any of the mentioned force fields these optimization steps were substituted by QM method optimizations. Additionally, since the Mms was not included in a statistical analysis done by McGregor et al. S3 on structures of the Protein Data Bank, S4 used to take the data for the starting geometries of residues for Amber 99sb, a mean of the most structurally similar peptides, Phe and Tyr, was used.
The remnant of both processes was identically followed step by step.

Methodology to derive Amber ff99sb charges
Charges of amino acids were derived using dipeptides, an amino acid forming two peptide bonds with the capping groups acetyl (Ace) and methyl-amide (Nme), as models. Two conformations of the peptide were considered, right-handed α-helix (α R ) and β -sheet (β ). Starting geometries had their psi (ψ), phi (φ ) and chi (χ) angles taken from the statistical analysis previously mentioned.
These geometries were then optimized using the HF method and 6-31G basis S5-S7 firstly keeping the ψ, φ and χ angles constrained, and then, restraining the ψ and φ angles of the α R conformation. If the optimized structure differed significantly from the starting one, it was discarded.
The electrostatic potential of the optimized structures (obtained at the same level of theory) was then used in the RESP fitting method. S8 Both conformations were given equal weight and charges were restrained so that the residue and capping groups were neutral. Additionally, the amide atoms charges were set to a set of values which were also used in the original force field on all neutral non-terminal amino acids. S9 S-2

Methodology to derive Amber ff03 charges
For this methodology the same dipeptides were used as models to derive the charges. The α R conformation had their backbone angles constrained at φ = −60 o and ψ = −40 o , and the PPII/β conformation with the backbone angles constrained at φ = −120 o and ψ = 140 o . The optimization was done using the HF/6-31G** level of theory. Electrostatic potential was obtained at the B3LYP/cc-pVTZ S10,S11 level and using the IEFPCM implicit solvent model (ε = 4) S12,S13 The atomic charges were derived using the same methodology used for the Amber ff99SB force field, except that no restrains were applied on the amide atoms to set their charges a particular value.
Based on the results, a 1/1.2 correcting factor was applied on the charges to keep the slope values near 1.

RESP methodology
The charges were obtained using MultiWFN S14 on Gaussian formatted checkpoints corresponding to each studied conformation. Additionally, the weight of each conformation was inputted and a series of restraints was set to keep the charges of some atoms to a fixed value and to keep the charge of symmetric atoms equal.

Hydrogen bonds
Hydrogen bonds were characterised based on the Wernet-Nilsson S15 function as implemented in the mdtraj library S16 for each simulation of the octapeptides. In the following graph we include every hydrogen bond whose frecuency is superior to 0.05 for at least one of the octapeptides for one of the force fields. As it can be observed the prevalence of hydrogen bonds is significantly higher for X = Phe or X = Tyr supporting the idea that this interactions determine the behaviour of these octapeptides.
S-3  Tables   Table S1: Number of water molecules included in every system for both ff99SB and ff03 force fields.