Simulations and Experiments: How close can we get?

The interactions between biomolecules and their environment can be
studied by experiments and simulations. Results from experiments
and simulations are often interpretations based on the raw data.
For an accurate comparison of both approaches, the interpretation
of the raw data from experiments and simulation have to be in
compliance. The design of such simulations and interpretation of
raw data is demonstrated in this thesis for two examples;
fluorescence resonance energy transfer (FRET) experiments and
surface adsorption of biomolecules on inorganic surfaces like gold.
FRET experiments allow to probe molecular distances via the
distance-dependent energy transfer efficiency from an excited donor
dye to its acceptor counterpart. In single molecule settings, not
only average distances, but also distance distributions or even
fluctuations can be probed, providing a powerful tool to study
flexibilities and structural changes in biomolecules. However, the
measured energy transfer efficiency does not only depend on the
distance between the two dyes, but also on their mutual
orientation, which is typically inaccessible to experiments. Thus,
assumptions on the orientation distributions and averages have to
be employed, which severely limit the accuracy of the distance
distributions extracted from FRET experiments alone. In this work,
I combined efficiency distributions from FRET experiments with dye
orientation statistics from molecular dynamics (MD) simulations to
calculate improved estimates of the distance distributions. From
the time-dependent mutual dye orientations, the FRET efficiency was
calculated and the statistics of individual photo-absorption, FRET,
and photo-emission events were determined from subsequent Monte
Carlo (MC) simulations. All recorded emission events were then
collected to bursts from which efficiencies were calculated in
close resemblance to the actual FRET experiment. The feasibility of
this approach has been tested by direct comparison to experimental
data. As my test system, I chose a poly-proline chain with Alexa
488 and Alexa 594 dyes attached. Quantitative agreement of
calculated efficiency distributions from simulations with the
experimental ones was obtained. In addition, the presence of
cis-isomers and specific dye conformations were identified as the
sources of the experimentally observed heterogeneity. This
agreement of in silico FRET with experiments allows employment of
the dye orientation dynamics from simulations in the distance
reconstruction. For multiple levels of approximation, the dye
orientation dynamics was used in dye orientation models. At each
level, fewer assumptions were applied to the dye orientation model.
Each model was then used to reconstruct distance distributions from
experimental efficiency distributions. Comparison of reconstructed
distance distributions with those from simulations revealed a
systematically improved accuracy of the reconstruction in
conjunction with a reduction of model assumptions. This result
demonstrates that dye orientations from MD simulations, combined
with MC photon generation, can indeed be used to improve the
accuracy of distance distribution reconstruction from experimental
FRET efficiencies. A second example of simulations and
interpretation in compliance with experiments are the studies of
protein adsorption on gold surfaces. Interactions between
biomolecules and inorganic surfaces, e.g. during the
biomineralization of bone, are fundamental for multicellular
organisms. Moreover, understanding these interactions is the basis
for biotechnological applications such as biochips or nano-sensing.
In the framework of the PROSURF project, a multi-scale approach for
the simulation of biomolecular adsorption was implemented. First,
parameters for MD simulations were derived from ab initio
calculations. These parameters were then applied to simulate the
adsorption of single amino acids and to calculate their adsorption
free energy profiles. For the screening of adsorbed protein
conformations, rigid body Brownian dynamics (BD) docking on
surfaces was benchmarked with the free energy profiles from the MD
simulations. Comparison of the protein adsorption rate from surface
plasmon resonance experiments and BD docking yielded good agreement
and therefore justifies the multi-scale approach. Additionally, MD
simulations of protein adsorption on gold surfaces revealed an
unexpected importance of positively charged residues on the surface
for the initial adsorption steps. The multi-scale approach
presented here allows the study of biomolecular interactions with
inorganic surfaces consistently at multiple levels of theory:
Atomistic details of the adsorption process can be studied by MD
simulations whereas BD allows the extensive screening of protein
libraries or adsorption geometries. In summary, compliance of
simulation and experimental setup allows benchmarking of the
simulation accuracy by comparison to experiments. In contrast to
employing experiments alone, the combination of experiments and
simulations enhances the accuracy of interpreted results from
experimental raw data.