Understanding and predicting aggregation in biopharmaceuticals

Lead Research Organisation: University of Manchester
Department Name: Chem Eng and Analytical Science

Abstract

Currently, one of the bottle necks to developing cheaper protein therapeutics is the cost of the downstream bioprocessing and formulation steps. A key problem is the loss of active protein therapeutic to irreversible aggregation throughout the bioprocess. Other problems can arise during chromatography or filtration when encountering high protein concentrations which could lead to high viscosites or even precipitation. The focus of this work is to develop predictive methods for identifying problematic conditions early on in the bioprocess. These could then be used for identifying changes to the protein to minimize the problems. Alternatively, the method could be used for optimizing the solvent properties (pH, buffer type and concentration) or finding other small molecule additives to be used in order to avoid aggregation or increase protein solubility. We benchmark our approach by studying antibodies and antibody fragments due to their growing importance as human therapeutics.

Technical Summary

The principal objective of this proposal is to develop methods for predicting protein bioprocessability based on knowledge of protein structural properties, a database of protein behaviour, and a set of high throughput data obtainable using accessible methods requiring small amounts of protein. The main emphasis is in part on predicting the intrinsic aggregation propensity of a protein, but also in predicting how the range of solvent conditions accessible in a bioprocess alters the aggregation behaviour. This tool could then be used for choosing from a set of mutants based on their bioprocessability, developing a bioprocess through informed decisions of process conditions (including excipients) chosen to minimize aggregation, improve solubility, and minimize viscosity, and in formulation, where the approach could be used as an excipient pre-screen for minimizing the solvent space to be sampled in formulation design. Our approach differs from others in that we probe the self-association step in the aggregation pathway and then integrate this knowledge into the predictor. The interactions between the aggregating precursors (partially folded proteins) are determined using static light scattering under native 'bioprocessing' conditions and using a novel approach based on probing interactions under destabilizing conditions. Using this approach we can isolate how the rate limiting steps of protein-protein association and unfolding depend upon co-solvent composition. This knowledge is then used to develop an improved aggregation predictor which incorporates the influence of different solvent conditions and protein structural properties. The tool is designed such that the predictive ability can be refined by incorporating experimental data obtained from high throughput experiments which are also developed as part of this work.

Planned Impact

This application is being submitted to a call by the Bioprocessing Research Industry Club (BRIC). One of the goals of BRIC is to bring together fundamental science for solving problems in the production of biopharmaceuticals. As such the impact of the proposed work can be defined by the research priorities defined by BRIC. 3 out of the 5 of these priorities can be identified with tools utilized in our study. These areas include 1) developing predictive tools based on protein structural properties for identifying the manufacturability of a protein product early on in the bioprocess. 2) the use of high-throughput technology to guide bioprocess development 3) effective modelling of whole bioprocesses. In particular, the main focus of our work is in developing a protein aggregation propensity predictor which incorporates protein structural properties and is also benchmarked using experimental data obtained from high-throughput experiments to be developed in the proposal. Furthermore, the tool is not only linked to predicting protein aggregation, but also can be utilized for predicting protein solubility or viscosity. The advantage is that each of these solution properties are linked to each other as they all depend on protein-protein interactions. In the future, understanding this link can lead to improved bioprocess design, where knowledge is transferable across different scales (lab to pilot plant) and across different operations. More specifically, the tool we develop can have impacts at various levels in bioprocessing. These include: 1) In antibody production, very often a set of mutant proteins are now produced with similar therapeutic potential. The tool we develop could be used for choosing one of the mutant proteins based on its bioprocessability. This could lead to identifying troublesome proteins early on in bioprocess development leading to large savings. 2) The predictor could be used in developing refolding operations where protein aggregation can lead to significantly reduced yields. In this case, the tool could be used to identify troublesome proteins, suggest mutations to improve the refoldability, and identify solvent conditions which can optimize the refolding yields. Along these lines, the PI and coPI (JPD) have submitted for a BBSRC funded case studentship with MSD to study refolding. 3) The work is also especially relevant in protein formulation. The big issue is predicting long term storage stability of the protein, where one of the main degradative pathways is by aggregation which can lead to an immunogenic response. In order to prevent aggregation, certain molecules (additives) are included in the formulation. Identifying the optimal conditions requires a pre-screening step usually done with accelerated storage studies which usually takes weeks. Our method could be used as a replacement pre-screening step reducing the time for the formulation design. Each of these proposed impacts can lead to savings in downstream bioprocessing which contributes most of the cost to commercialized therapeutics. Thus, the extended impact of the work is in reducing the cost of biopharmaceuticals. The predominant mechanism for dissemination of results is via the bioprocessUK meetings which are heavily attended by both industrialists and academics. These meetings foster interactions between academics and industrialists, via various mechanisms such as 'speed-networking'. This gives a good forum for exchange of ideas and developing the projects within the industrial interests. In addition, these meetings often lead to initial collaborative studies, which can take the form of MSc and MEng based projects.

Publications

10 25 50
publication icon
Austerberry JI (2017) The effect of charge mutations on the stability and aggregation of a human single chain Fv fragment. in European journal of pharmaceutics and biopharmaceutics : official journal of Arbeitsgemeinschaft fur Pharmazeutische Verfahrenstechnik e.V

publication icon
Corbett D (2017) Coarse-Grained Modeling of Antibodies from Small-Angle Scattering Profiles in The Journal of Physical Chemistry B

publication icon
Curtis R (2015) Depletion forces due to image charges near dielectric discontinuities in Current Opinion in Colloid & Interface Science

publication icon
Hebditch M (2017) Protein-Sol: a web tool for predicting protein solubility from sequence. in Bioinformatics (Oxford, England)

 
Description The main overarching objective for the work was to improve upon predictive methods for assessing the manufacturability of protein therapeutics (or biopharmaceuticals). Prediction is especially important early on in the discovery phase when not much protein material is available. Some of the key characteristics that control the manufacturability of a protein product are the protein stability against physical degradation (through irreversible aggregation) and the possibility of protein solutions becoming viscous, phase separating, and opalescent or cloudy. These characteristics are all linked either directly or indirectly to the protein-protein interactions (PPIs) or what is now commonly referred to as reversible self-association (RSA). The emphasis of our work was to elucidate the nature of protein-protein interactions and to understand how to use their knowledge in developing improved predictors of aggregation, which is one of the most troublesome problems in developing a therapeutic. The project also focused in particular on tools for formulation. In formulation development, solution conditions can be manipulated by changing the pH, using different buffers or salts, and other small molecule additives called excipients that are chosen to reduce aggregation or phase separation for instance. As such, we aimed at predicting how aggregation and protein-protein interactions are controlled both by protein structural properties and solution conditions.

Below we list the key findings of the work in order of importance.

1. A rational mutagenesis strategy for reducing aggregation propensity. We used an initial structural informatics study on a proteome wide solubility database to show that there is a strong correlation between the lysine to arginine content (KR ratio) of proteins and their solubility or aggregation propensity, that is a high KR ratio appear to correlate with reduced aggregation propensity (Warwicker et al., 2014). Subsequently, we generated a series of 10 mutants for an antibody fragment protein (scFv) with varying KR ratios. A detailed protein-protein interaction and aggregation study demonstrated that increasing the lysine content did reduce the aggregation propensity indicating the KR ratio could be used as a rational design strategy for improving protein manufacturability (Austerberry et al., 2017).

2. An aggregation predictor tool: We have used the series of mutants to test commonly used approaches for predicting aggregation propensity using measurements of surrogate parameters reflecting protein conformational stability (melting temperatures, free energies of unfolding) and colloidal stability (native-state PPI parameters, aggregation temperatures). The most stable mutants do not score well according to any of these commonly used predictors. The reason is that aggregation is controlled by interactions between the unfolded or partially folded proteins that are difficult to probe as partially folded states occur very transiently. We have subsequently shown interactions between partially or unfolded proteins can be probed indirectly by PPI measurements under chemically denaturing conditions. As such, we are currently developing the method as a complementary aggregation predictor tool. Another key outcome of the mutant scFv work is in understanding when and why measurements of native-state PPIs correlate with aggregation. We showed that only the electrostatic contribution is reflective of aggregate growth rates, but not rates of aggregation formation. As such, the native-state PPI measurements need to be applied with care when using them as a predictive tool.

3. Developing high throughput methods for quantifying PPIs. As alluded to in the previous section, native state PPI interaction parameters provide insight into aggregation processes. As part of the work, we have demonstrated how to interpret these parameters from dynamic light scattering (DLS) measurements. The traditional approach of using static light scattering (SLS) requires substantial amounts of protein and time, which often is a limiting factor in applying the method during development. The DLS approach however can be implemented on a multi-well plate reader in a minimal consumption high-throughput format. We have provided the largest available dataset relating SLS to DLS measurements, which has allowed us to derive a molecular basis for correlating the measurements (Roberts et al., 2014; Roberts et al., 2015).

4. A mechanistic understanding of how formulation components impact PPIs and aggregation: Through the detailed PPI measurements complemented with electrophoretic light scattering, we have been able to deduce how formulation components interact with proteins and alter PPIs. This has allowed us to develop rules for choosing stable formulation conditions and for finding novel excipients with improved effectiveness. In formulation development, aggregation is often suppressed by using a formulation at low ionic strength at pH conditions where proteins carry a sufficient net charge, in which case, repulsive electrostatic interactions reduce aggregate growth rates. Through our systematic PPI study, we have demonstrated that commonly used anionic buffers bind to proteins and neutralize the repulsive interactions thereby increasing aggregate growth (Roberts et al., 2015). Conversely, aggregation can be enhanced in solutions near to the protein isoelectic pH due to interactions between oppositely charged surfaces. We have shown novel excipients such as dipeptides can be used to suppress isoelectic aggregation through a combination of charge screening effects and reducing hydrophobic association between proteins (Nuhu et al., 2014).

5. Coarse-grained models for antibody molecules: Simplified spherical models have been used successfully to describe the solution behaviour of small globular proteins. However, antibody molecules (which comprise the largest class of protein therapeutics) have a highly anisotropic y-shape with significant molecular flexibility, which raises a question of whether simplified models can still be used for describing their solution behaviour. We have shown these simplified descriptions are applicable for interpreting the dilute solution measurements of PPIs (Roberts et al., 2014). The main utility of the models is for extrapolating the dilute solution measurements to calculate concentrated solution properties such as phase separation, solution cloudiness, or rheological properties.
Exploitation Route Being able to predict manufacturability from protein sequence or structural properties is an especially challenging problem. Making progress requires a large dataset for aggregation or solubility behaviour for a range of proteins, some of which have well-defined mutations. Towards this end, our suite of scFv mutant proteins provides a good model system for further developing and testing predictive algorithms. For instance, we are continuing the work with scFv mutant proteins in a BBSRC/LINK grant (BB/M006913/1), where we are studying their behavior under chemically denaturing conditions. We have shared the mutants as part of a collaborative project with Professor Chris Roberts at the University of Delaware for more detailed aggregation studies. In addition, the mutants are also useful for understanding mechanisms of cellular production in a collaborative project led by Professor Colin Robinson (University of Kent) investigating the structural determinants for protein transport across bacterial membranes (Jones et al., 2016).

The knowledge of buffer effects on PPIs and protein aggregation has led to changes in industrial formulation development groups. Initial buffer screens used by formulation development involved mixtures of citrate and phosphate buffers due to their buffering capacity over a range of pH values. However, both buffers, due to their multivalent charges, are especially effective at neutralizing protein charged groups and increasing aggregation. As such groups have moved away from using these buffers during development.

Our knowledge gained about how excipients interact and alter PPIs is being used in developing novel excipients as part of our EPSRC Future Formulation of complex products grant (EP/N025105/1). In particular, we are examining a much broader range of excipients with different charged properties. As part of the initial screen, we have discovered poly-phosphates exhibit desirable stabilizing properties for antibody formulations.
Sectors Healthcare,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology

 
Description The primary impact of the work has been and will be in biopharmaceutical research and development sectors, in particular, formulation groups, but also to a lesser extent, the drug discovery and downstream processing specialists. Industrial collaborations provided an avenue for establishing impact during the grant. During the secondment of Dr Curtis at MedImmune there was exchange of academic expertise and industrial know how that for instance helped guide choice of buffers during formulation development and informed upon the choice of screening experiments. The strong relationship has led to additional projects (case studentships, MedImmune funded PDRAs, and a BBSRC Impact award). A collaborative project with Aecor, an SME, established the molecular mechanism for a novel excipient in stabilizing a therapeutic antibody. Instrument manufacturers Wyatt Technology and Avacta (now Unchained Labs) provided free loans of equipment (a dynamic light scattering plate reader and the Optim), in exchange for demonstrating novel applications of the kit, which have now been disseminated through publications (for instance see Roberts et al., 2015; Nuhu et al., 2015; and Austerberry et al., 2017). Some of the discoveries and key findings from the work are being integrated into industrial research groups and will eventually lead to cost savings through more efficient formulation development. Examples include the predictive tools for identifying stable formulations through use of chemical denaturants, a rational design strategy for improving therapeutic stability by increasing the lysine to arginine content, and a publicly available webserver for solubility predictions (http://protein-sol.manchester.ac.uk/about). The results have been and will be disseminated and/or communicated in an effective way such that industrial end users see clear benefits and potential savings of implementing the discoveries. Some of these advances still need further development before translation to an industrial setting through on-going projects that involve academic and industrial collaborators (BBSRC Impact Award with MedImmune, EU ITN training grant, EPSRC future formulation of complex products grant led by UCL). These large consortia type grants provide access to a broad range of therapeutic molecules (with company partners), which can be used for illustrating how the discoveries can be implemented. As an example, in a current BBSRC funded studentship project with MedImmune, we have been provided access to a suite of biotherapeutics with varying aggregation propensities to demonstrate the utility of our chemical denaturation method. Our close relationships with industry and academic experts has facilitated the dissemination and communication of results for instance through meetings with the EPSRC industrial advisory board (including GSK, UCB, FujiFilm Diosynth, Arecor, Wyatt, Albumedix) or bi-annual project meetings involving all partners of the EU ITN training grant (including Roche, Wyatt, MedImmune, Novozymes, Danish Technical University, LMU Munich, University of Lund). Further, we have recently organized a protein aggregation meeting at University of Manchester (13th Febuary 2017) (funded through a grant from BBSRC/BioProNET) to bring awareness to industry for the state of the art in controlling and predicting protein aggregation (title of workshop: Recent Breakthroughs and New Perspectives of Protein Aggregation). The pharmaceuticals sector in the UK consists of over 500 companies and includes the top 20 global pharmaceutical companies with an estimated turn-over of £32.4 billion. The business employs over 70,000 people, where 23,000 are employed in highly skilled research and development. The benefits gained by the industry from academic research, in terms of expertise, know-how, and intellectual property, have contributed significantly to the strength of the sector in the past. It is imperative to continue the collaborative efforts. The academic and industrial network of contacts we established through the BRIC community provided the foundation for follow-on funding to establish larger consortia. This concerted effort is needed to maximize the impact from academic research in addressing the key challenges faced by the industry. A good illustration is given by the need for a database comprising experimental data for a large number of therapeutic molecules in different formulation conditions taken with state of the art methods used in development. This data would be invaluable for developing improved and more efficient screening protocols and enable predictive algorithms. While such a large dataset does exist, it is not publicly available due to industrial confidentiality agreements. Through the consortia, more than 30 therapeutics have now been made available for a systematic screen that is currently being undertaken. The data is being used to populate our own publicly available webserver, which is being refined throughout the process for predicting therapeutic manufacturability. The screening programme is providing immediate benefits to the industrial members of the consortia in terms of enhanced knowledge of their therapeutics, but the webserver also provides a route for exploitation of the findings by the wider community. We envision the consortia are setting an international standard for the benefits of academic research driven by industrial needs, at the very least within the biopharmaceuticals sector. Lastly, a gap of young bioformulation scientists has been emphasized by industry directors at recent UK formulation meetings. The grant has trained two PDRAs, Dorota Roberts who is currently in industry (Lonza) and James Austerberry, whom is employed as a researched on the BBSRC Impact award at University of Manchester. The project is also indirectly related to training a large number of scientists through helping establish the follow-on projects and numerous case studentships.
First Year Of Impact 2014
Sector Healthcare,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology
Impact Types Economic

 
Description EPSRC Future formulation of complex products
Amount £2,380,872 (GBP)
Funding ID EP/N025105/1 
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Academic/University
Country United Kingdom
Start 07/2016 
End 06/2021
 
Description EU Innovative Training Networks, Call: H2020-MSCA-ITN-2015
Amount € 4,067,740 (EUR)
Organisation European Commission 
Department Horizon 2020
Sector Public
Country European Union (EU)
Start 02/2016 
End 02/2021
 
Description LINK
Amount £920,000 (GBP)
Funding ID BB/M006913/1 
Organisation Biotechnology and Biological Sciences Research Council (BBSRC) 
Sector Public
Country United Kingdom
Start 01/2015 
End 12/2018
 
Title A publicly available database for solubility prediction of biopharmaceuticals 
Description A key stumbling block to predicting solubility or aggregation propensity is the lack of comprehensive database relating behaviour to structure along with a suite of tools/algorithms for mining the data and testing predictions. Towards this end, we have developed a publicly accessible web server that uses our in-house and freely available software and algorithms for solubility prediction, in a user-friendly graphics window. The software is used to calculate a variety of different properties from protein amino acid sequence and/or protein structural data. Properties calculated for sequences include: amino acid compositions; combinations of charges to give net predicted charge, fraction charged, predicted isoelectic point, ratio of conservative amino acids; propensity for disorder; and ß-strand/sheet forming propensity. These properties are calculated as averages over the entire sequence and as minimal/maximal values in windows of 21 and 51 amino acids along the sequence. For structures and structural models, properties calculated include: patch non-polarity, calculated as non-polar solvent accessible surface area (SASA) divided by polar SASA, size distribution of charged patches (positive and negative, including the largest patches) from Finite Difference Poisson Boltzmann (FDPB) calculations; average contact order for heavy atoms in the protein, relating to the degree of close packing as an empirical proxy for structural stability. Charge patches are dependent on ionic strength and pH, features that are also part of FDPB calculations, which allows us to account for ionic strength and/or pH effects on solubility. The philosophy of our website is to make all software available both to use online, and as downloadable code for in-house use. 
Type Of Material Antibody 
Year Produced 2016 
Provided To Others? Yes  
Impact none as yet 
URL http://www.protein-sol.manchester.ac.uk/
 
Description European Network Collaboration 
Organisation Ludwig Maximilian University of Munich (LMU Munich)
Country Germany 
Sector Academic/University 
PI Contribution We helped write the successful proposal for the Marie Curie ITN training grant, which supports 15 PhD students across the consortium. 4 of the PhD students will be enrolled at Manchester.
Collaborator Contribution Each of the members of the consortium is contributing towards generating an unprecedented dataset for use in biopharmaceutical formulation design. Contributions cover skills and expertise, as well as access to experimental equipment and protein availability.
Impact The collaboration has just commenced this year. PhD students will start in July 2016
Start Year 2014
 
Description European Network Collaboration 
Organisation Lund University
Country Sweden 
Sector Academic/University 
PI Contribution We helped write the successful proposal for the Marie Curie ITN training grant, which supports 15 PhD students across the consortium. 4 of the PhD students will be enrolled at Manchester.
Collaborator Contribution Each of the members of the consortium is contributing towards generating an unprecedented dataset for use in biopharmaceutical formulation design. Contributions cover skills and expertise, as well as access to experimental equipment and protein availability.
Impact The collaboration has just commenced this year. PhD students will start in July 2016
Start Year 2014
 
Description European Network Collaboration 
Organisation MedImmune
Country United States 
Sector Private 
PI Contribution We helped write the successful proposal for the Marie Curie ITN training grant, which supports 15 PhD students across the consortium. 4 of the PhD students will be enrolled at Manchester.
Collaborator Contribution Each of the members of the consortium is contributing towards generating an unprecedented dataset for use in biopharmaceutical formulation design. Contributions cover skills and expertise, as well as access to experimental equipment and protein availability.
Impact The collaboration has just commenced this year. PhD students will start in July 2016
Start Year 2014
 
Description European Network Collaboration 
Organisation Novozymes
Country Denmark 
Sector Public 
PI Contribution We helped write the successful proposal for the Marie Curie ITN training grant, which supports 15 PhD students across the consortium. 4 of the PhD students will be enrolled at Manchester.
Collaborator Contribution Each of the members of the consortium is contributing towards generating an unprecedented dataset for use in biopharmaceutical formulation design. Contributions cover skills and expertise, as well as access to experimental equipment and protein availability.
Impact The collaboration has just commenced this year. PhD students will start in July 2016
Start Year 2014
 
Description European Network Collaboration 
Organisation Technical University of Denmark
Country Denmark 
Sector Academic/University 
PI Contribution We helped write the successful proposal for the Marie Curie ITN training grant, which supports 15 PhD students across the consortium. 4 of the PhD students will be enrolled at Manchester.
Collaborator Contribution Each of the members of the consortium is contributing towards generating an unprecedented dataset for use in biopharmaceutical formulation design. Contributions cover skills and expertise, as well as access to experimental equipment and protein availability.
Impact The collaboration has just commenced this year. PhD students will start in July 2016
Start Year 2014
 
Description European Network Collaboration 
Organisation University of Copenhagen
Country Denmark 
Sector Academic/University 
PI Contribution We helped write the successful proposal for the Marie Curie ITN training grant, which supports 15 PhD students across the consortium. 4 of the PhD students will be enrolled at Manchester.
Collaborator Contribution Each of the members of the consortium is contributing towards generating an unprecedented dataset for use in biopharmaceutical formulation design. Contributions cover skills and expertise, as well as access to experimental equipment and protein availability.
Impact The collaboration has just commenced this year. PhD students will start in July 2016
Start Year 2014
 
Description European Network Collaboration 
Organisation Wyatt Technology Corporation
Country United States 
Sector Private 
PI Contribution We helped write the successful proposal for the Marie Curie ITN training grant, which supports 15 PhD students across the consortium. 4 of the PhD students will be enrolled at Manchester.
Collaborator Contribution Each of the members of the consortium is contributing towards generating an unprecedented dataset for use in biopharmaceutical formulation design. Contributions cover skills and expertise, as well as access to experimental equipment and protein availability.
Impact The collaboration has just commenced this year. PhD students will start in July 2016
Start Year 2014
 
Description MedImmune 
Organisation MedImmune
Country United States 
Sector Private 
PI Contribution I have spent a part time sabbatical visiting MedImmune during 2013. We recently were awarded a BBSRC LINK grant at the end of 2014 for studying biopharmaceuticals formulation.
Collaborator Contribution MedImmune are supporting 2 PDRAs at Manchester over a four year period
Impact See publication outcomes
Start Year 2011