Introduction
Technological advancement in cell culture has led to an increase in
therapeutic protein titers over the last decade. This advancement has
shifted the attention towards downstream processes as the bottleneck in
the manufacture and production of biopharmaceuticals (Guiochon &
Beaver; Hanke & Ottens, 2014). For biologics, antibodies are extremely
complex and provide an almost unlimited design space to engineer binding
with target molecules (antigens) (Maier & Labute). This results in a
diverse array of biophysical properties and a challenge for separation.
The relative ease with which new antibody candidates are generated and
low toxicity of their degradation products make this class of molecules
an excellent therapeutic agent (Tiller et al., 2008). Yet, the
structural complexity of these molecules and diverse biophysical
properties (including conformational flexibility) pose a significant
challenge to the selection of candidates and development of appropriate
purification modalities. An approach is proposed that can be applied to
determine feasibility of separation of specific degraded products (e.g.
oxidation, deamidation sites) via in silico screening of ligands
to a targeted location. This in silico screening approach is
analogous to the experimental developability (Lorenz et al., 2014; Yang
et al., 2013) approach and hence can be viewed as in silicodevelopability.
The process of developing therapeutic antibodies require the generation
of many variants. The number of molecules generated make empirical
biophysical characterization of each molecule difficult, resource
intensive, and time consuming (Jarasch et al., 2015; Shah et al., 2015).
In-depth biophysical characterization of antibodies requires a
significant amount of protein mass to determine stability profiles,
purification, and formulation conditions. In addition, this process can
be very time consuming even with the advent of high-throughput screening
(HTS) approaches (Petroff et al., 2016). Thus, the ability to usein silico computational modeling will not only save time, but
also optimize candidate selection for the best therapeutic candidate and
accelerate the path from discovery to first-in-human clinical trials.
Computational approaches may also be able to assess mechanisms and
pathways that are not readily accessible experimentally and hence
provide additional insights into molecular development.
Recently published work has demonstrated the utility of computational
algorithms to isolate antibody complementary determining region (CDR)
loops and associated structural features that infer biological and
biophysical properties when bound to receptors and antigens (Morea,
Lesk, & Tramontano, 2000). Further, computational modeling algorithms
to generate 3D structure of any novel antibody based on currently
available protein databank (PDB) structures are becoming increasingly
routine (Morea, Leplae, & Tramontano, 1998; Morea, Tramontano, Rustici,
Chothia, & Lesk, 1997). In silico docking studies have yielded
results that are consistent with relative chromatographic retention
(k’), and surface properties for a range of biologics (Insaidoo et al.,
2015). The next step is to combine these advances in antibody homology
modeling and in silico computational docking to understand
chromatographic separation and select appropriate ligands for bioprocess
development for a specific separation.
In this paper, we investigate the biophysical principles governing
protein-chromatographic ligand interactions. We show in silicodocking studies in combination with molecular dynamics simulations can
inform bio-process development and biologics purification. Specifically,
we connect antibody biophysical properties to ligand selection and
impact of ligand density on retention and selectivity.
In silico models of agarose fibers functionalized with
chromatographic ligands showed increased binding affinity as a function
of increasing number of interacting ligands (N). The observed increase
is consistent with experimental correlation between resin ligand density
and k’. These results have implications not only for bioprocess
development but also for resin design and our understanding of the
principal criteria for biologics design for therapeutic efficacy and
successful downstream development.
Currently, the selection of lead candidates from a range of highly
potent efficacious antibodies relies on limited downstream process
development data (Jarasch et al., 2015; Lorenz et al., 2014). Even with
high through-put screening, it would still require significant resources
to fully screen and develop the operational design space for the
process. This paper maps the interaction between chromatographic ligands
and monoclonal antibodies at an atomic level using a general, extendable
computational approach. This in silico approach to process
development allows us to discern the impact of subtle differences on
candidate selection and process development.