Protein Adsorption on Solid Surfaces: Data Mining, Database, Molecular Surface-Derived Properties, and Semiempirical Relationships

Protein adsorption on solid surfaces is a process relevant
to biological,
medical, industrial, and environmental applications. Despite this
wide interest and advancement in measurement techniques, the complexity
of protein adsorption has frustrated its accurate prediction. To address
this challenge, here, data regarding protein adsorption reported in
the last four decades was collected, checked for completeness and
correctness, organized, and archived in an upgraded, freely accessible
Biomolecular Adsorption Database, which is equivalent to a large-scale,
ad hoc, crowd-sourced multifactorial experiment. The shape and physicochemical
properties of the proteins present in the database were quantified
on their molecular surfaces using an in-house program (ProMS) operating
as an add-on to the PyMol software. Machine learning-based analysis
indicated that protein adsorption on hydrophobic and hydrophilic surfaces
is modulated by different sets of operational, structural, and molecular
surface-based physicochemical parameters. Separately, the adsorption
data regarding four “benchmark” proteins, i.e., lysozyme,
albumin, IgG, and fibrinogen, was processed by piecewise linear regression
with the protein monolayer acting as breakpoint, using the linearization
of the Langmuir isotherm formalism, resulting in semiempirical relationships
predicting protein adsorption. These relationships, derived separately
for hydrophilic and hydrophobic surfaces, described well the protein
concentration on the surface as a function of the protein concentration
in solution, adsorbing surface contact angle, ionic strength, pH,
and temperature of the carrying fluid, and the difference between
pH and the isoelectric point of the protein. When applying the semiempirical
relationships derived for benchmark proteins to two other “test”
proteins with known PDB structure, i.e., β-lactoglobulin and
α-lactalbumin, the errors of this extrapolation were found to
be in a linear relationship with the dissimilarity between the benchmark
and the test proteins. The work presented here can be used for the
estimation of operational parameters modulating protein adsorption
for various applications such as diagnostic devices, pharmaceuticals,
biomaterials, or the food industry.