Direct Comparison of Amino Acid and Salt Interactions with Double-Stranded and Single-Stranded DNA from Explicit Solvent Molecular Dynamics Simulations (vol 13, pg 1794, 2017)

Casey T. Andrews; Brady A. Campbell; Adrian H. Elcock

doi:10.1021/acs.jctc.8b00596

Sometime after the publication of our paper,1 we became aware of a problem with our implementation of modified dihedral parameters for nucleic acids in simulations performed with the particular version of GROMACS (4.6.5) used in our work. Specifically, for reasons described below, simulations that we thought used the χOL3 and χOL4 parameter sets-which were designed to improve the modeling of the glycosidic dihedral angle in RNA and DNA, respectively-did not in fact correctly use those parameters. The problem is a relatively subtle one. In GROMACS, dihedral parameters are listed, along with the other bonded parameters, in a file called "ffbonded.itp". In attempting to incorporate the χOL3 and χOL4 dihedral parameters, we added the new entries at a line in this file af ter all pre-existing dihedral parameters. The pre-existing dihedral parameters include a number that contain wild cards (denoted "X"), such as "X-CT-N∗-X", that can be used to assign generic parameters to dihedral angles involving a variety of atom types. At the time that we conducted our work, we were under the impression that entries containing wild cards would be overridden by more specifically defined dihedral entries that appeared later in the list of dihedral parameters. For example, we assumed that the wild card-containing dihedral entry "X-CT-N∗-X" would be overridden by χOL3s entries for the more specifically defined dihedral entry "OSCT- N∗-C4". Unfortunately, however, we have recently become aware that in all versions of GROMACS prior to version 5.1.3, it is the first set of parameters that matches to a given dihedral angle that is used, with all subsequent matches being ignored, even if those matches are more specific. This means that dihedral entries that are listed after wild card-containing entries in the "ffbonded.itp" file are generally at risk of being ignored, without warning, in older versions of GROMACS. A full description of this feature of GROMACS, as discussed and resolved by its developers, can be found here: https://redmine. gromacs.org/issues/1901; there, the developers note that in versions of GROMACS prior to version 5.1.3 the importance of dihedral parameter ordering "does not seem to be documented". The relative ordering of wild card-containing and specific dihedral entries is no longer an issue in versions 5.1.3 and later: in these newer versions, the entry with the most specific match to the atom types in question is used. With regard to our paper, this means that the simulations we reported were actually performed with a force field much closer to the original AMBER parm99 force field (supplemented with bsc0 corrections) than to the intended combination of this force field with the χOL4 corrections for DNA. The one difference between our implementation and the original parm99 + bsc0 force field is that zero dihedral energy appears to be applied to rotation around the glycosidic bond in our simulations, which represents an unfortunate reversion to what was used in versions of the AMBER force field prior to parm94. A very similar issue affects two other papers from our group.2,3 Based on recently reported results from the Cheatham group,4 it is apparent that our incorrect implementation of the χOL3 parameters for RNA results in stacking free energies for dinucleoside monophosphates (DNMPs) that are significantly more favorable than those computed with the correct implementation. It seems reasonable to expect that our incorrect implementation of the χOL4 parameters for DNA might have similar consequences. Given that the focus of the present study1 was on the thermodynamics of interactions between amino acid side chains and double- and single-stranded DNA (and not on the conformational behavior of the DNA), we expect that the impact of this issue on our findings is likely to be very slight and that essentially all of the conclusions we reported should remain unchanged. The one result that conceivably could change substantially is our observation that no intercalation of aromatic or aliphatic side chains occurred between the bases of double-stranded DNA (dsDNA) on the time scale of our simulations. If base stacking were to be weakened by the correct implementation of the (DNA) χOL4 parameters in the same way that the Cheatham group has shown it is weakened by the correct implementation of the (RNA) χOL3 parameters, it is possible that some degree of intercalation of amino acid side chains into dsDNA could occur during the simulations. If this were to happen, however, it would be more likely to improve, not worsen, the already good agreement with experiment for the relative preferences for binding to ssDNA versus dsDNA that is apparent in Figure 4 of our paper. In particular, such an effect would likely move the data points plotted for aliphatic and aromatic side chains further into line with those plotted for all other side chains in Figure 9 of our paper.

Direct Comparison of Amino Acid and Salt Interactions with Double-Stranded and Single-Stranded DNA from Explicit Solvent Molecular Dynamics Simulations (vol 13, pg 1794, 2017)

Files and links (1)

Abstract

Details

Metrics