Phage display has had a good run. After bursting onto the antibody engineering scene in the 1990s, phage display rapidly disrupted the field. The in vitro high throughput screening and selection offered by phage display hit the jackpot in 2002 with the approval of adalimumab (Humira®), which we now know as the world’s best-selling therapeutic antibody.  

However phage display’s crown has been slowly slipping. Despite early promise, only 10 of the 100 or so FDA-licensed monoclonal antibodies were developed using phage display (1, 2). So what are the pitfalls of phage?

Antibodies on display in phage

Phage, or bacteriophage to give them their full name, selectively infect bacteria and hitchhike the bacterial machinery to replicate themselves. Biologists in turn have hijacked phage by inserting genetic code into the phage which is then manufactured into a protein by the bacterial hosts’ protein expression system and displayed on the phage outside coat. This link between the genotype (the genetic sequence inserted into the phage) and the phenotype (the antibody manufactured based on the code) is the most important principle of phage display.

Antibody engineers exploit this system for antibody discovery and for affinity maturation of lead antibody candidates. The genetic sequence of the starting antibody is diversified into a library of variants. Each genetic variant is inserted into a phage, in turn inserted into a bacterial factory that pumps out phage replicas bearing the antibody variant in protein format.

In a high-throughput process called biopanning, the antibody target (the epitope) is immobilised and a sample of each variant in the library is washed over the target. Any variants that bind to the target are considered hits.

Phage display libraries – big but have they delivered?

The “display” principle linking genotype and phenotype isn’t limited to phage. Numerous display systems exist, where the antibody of interest is displayed on ribosomes, or on the cell surface of bacteria, yeast or mammalian cells. Each system has its pros and cons, but phage display has several advantages. First the bacterial libraries hosting the phage are relatively fast, cheap and easy to grow up, compared to slower growing yeast and mammalian cells. Phage display has another big advantage – and that is big, big libraries of up to 10^12 variants.

However these large phage libraries come with a price, favouring quantity over quality. While fast, synthetic and semi-synthetic methods of creating library diversity create artificial variants that would never be generated by natural B cell diversification. In addition, synthetic random libraries can be skewed with certain bases showing up too often (or not often enough) at certain positions in the DNA sequence (3). The knock-on effect is the presence of amino acids in “unnatural” spots in the antibody. Such variants may have good enough affinity to register as hits in phage display biopanning, but these “hits” may turn out to have folding or expression problems.

Express yourself

Expression bias and folding errors are two of the biggest drawbacks of phage display. The prokaryote machinery of bacteria is simply not equipped to fold complex human antibodies and cannot make crucial post-translational modifications such as forming disulphide bonds. E.coli, the most popular phage host, can be picky about which antibodies it makes, preferring to express proteins rich in certain amino acids such as methionine and lysine. Phage display handles expression and folding of smaller proteins such as functional snippets of antibodies much better. Single-chain variable fragments (scFv) of antibodies are most suited to phage display, followed by antigen-binding fragments (Fab) (4). However even scFv expression is not without its problems. For example, these antibody snippets are prone to aggregation which can cause false negatives or positives. Another limitation with antibody fragments such as scFv occurs if they need to be reformatted to full length IgG or other formats.   After the high-throughput screening phase of phage display, reformatting promising scFv variants can bring an antibody development project skidding back to a crawl. Plus, a promising scFv can turn out to have disappointing affinity once the whole antibody is expressed and folded properly in a mammalian cell.

The eukaryote protein expression and post-translational machinery of yeast display can overcome some of the expression problems of prokaryote phage display. But amongst the trade-offs are smaller library sizes and the larger volumes needed for growing enough yeast cells (5).

Alternatives on display

Some of phage display’s limitations can be overcome by pairing phage display with other technologies. Some groups pair phage display with yeast display (6), or screen phage display hits for functionality with flow cytometry (7) while others batch reformat scFv to IgG before further functionality analysis. But all of these approaches take time, money and resources.  

Another option is to bypass phage display altogether. Here at Fusion Antibodies, our rational affinity maturation platform (RAMPTM) sidesteps most of the issues with phage display. Our rationally designed in silico libraries are magnitudes larger than phage libraries, with around 10^25 variants. Diversity is introduced rationally, following the example of somatic hypermutation, which minimises the risk of finding amino acids in unexpected places. Rapid in silico screening of the library yields a micro-library of the strongest candidates which are then expressed directly as full length IgGs in mammalian CHO cells – neatly avoiding expression bias, folding problems and the need to reformat fragments to full IgGs. RAMP™ can also be used in combination with phage display. For example for performing affinity maturation of a phage-derived lead antibody candidate, or for sequence optimisation to improve expression and stability in CHO cells.


Learn how RAMPTM can boost or replace phage display for affinity maturation and sequence optimisation of your antibody

Today, therapeutic antibodies are reaching the clinic in unprecedented numbers. However the path from laboratory to the clinic is far from smooth. Antibodies face a range of obstacles during development, from insufficient efficacy, to manufacturing difficulties to immunogenicity – any of which can spell the costly end of an antibody development program.

Since 2012, Fusion Antibodies has delivered over 160 successful antibody humanization projects to customers worldwide. For many of these projects, we and our customers observed that our humanized antibodies retained and even improved antigen affinity, compared to the parent antibody.

Antibody candidates originate from many sources, using ever-improving discovery technologies. However, achieving a balanced antibody profile can be challenging. Our customers need to identify and eliminate those initially promising antibodies that bind tightly, but later turn out to have problems with manufacturing, stability or immunogenicity. They need to be sure they have the right antibody before establishing a stable cell line, and be confident that their lead candidate can get through CMC testing and make it to clinic.

RAMP™ – a 2 step affinity maturation platform

To help customers face these challenges, we have poured our expertise into creating RAMP™ – our rational affinity maturation platform designed to accelerate and optimize selection of your lead antibody candidate.

RAMP™ combines innovative library design with stringent in silico screening of variants. This sieves out the strongest candidates into a micro-library that can be expressed in mammalian cells for further characterization.

Rational library design

Taking the parent antibody, we create a massive library of around 10^20 variants. Our proprietary rational design approach takes a leaf from nature, inspired by how B cells use somatic hypermutation to generate antibody diversity.

RAMP™ introduces mutations in both the CDR and framework regions to create diversity, allowing only amino acids that can naturally occur at each position in the human antibody sequence. This “natural” approach reduces the likelihood of hydrophobic patches of amino acids and the downstream risks of aggregation and immunogenicity.

At the same time, strict sequence checks are applied to screen out primary sequence liabilities such as deamidation sites, cleavage sites and free cysteines. These checks and balances help create a library of variants that are “pre-screened” for manufacturing and clinical use.

In silico refinement

The library is then refined using in silico software that rapidly models variant-antigen binding and predicts affinity and stability. Over the 3-week in silico phase, the initial library is funneled down into a micro-library of the 100 strongest variants. At this stage we either express the micro-library as full length IgGs in CHO mammalian cells or hand the micro-library back to the client for further in-house testing.

RAMP™ up your chances of getting to the clinic

RAMP™ for affinity maturation is a fast, reliable method for improving affinity and selecting your lead antibody candidate. In a promising performance test, RAMP™ improved the affinity of the best-selling breast cancer drug trastuzumab in silico and we’re currently validating the best variants experimentally.

RAMP™ can also be applied to “rescue” molecules, with promising functional activity but poor developability profiles, where finessing of the sequence is required. Our novel library design approach can also open up new sequence space to potentially build on your patent family and increase the value of your program.

Want to select the best possible antibody for the clinic?

Design, development and clinical testing of therapeutic antibodies is a race against the clock, and against competitors. Companies are increasingly turning to in silico approaches to turbo boost the process. At Fusion Antibodies, we were early adopters of in silico techniques. We’ve seen first hand how embracing this technology accelerates antibody design and development, which reduces costs.

AI and machine learning

Despite the buzz, the holy grail of true artificial intelligence (AI) – machines that can “think” and reason like humans do – remains elusive. In the meantime, deep machine learning is the established star of the in silico show. Machines are “trained” by feeding them large sets of experimental data and teaching them (via algorithms) how to perform a task. With each repetition of the task, the machine adds the experience to their knowledge bank, improves performance and comes up with results that humans didn’t necessarily expect. So how do in silico techniques, including machine learning, fit in to antibody development?

In silico antibody development

Antibody engineers have a plethora of options in their in silico toolkit for optimising antibodies. These bioinformatics tools include homology modelling to predict antibody structure, molecular docking to identify antibody-antigen interactions and algorithms to calculate energy changes in mutated versions of the antibody. Each of these processes can benefit from machine learning to speed up predictions and improve decision making. Stitching these processes together into an automating streamlined workflow saves even more valuable time.

Focus on libraries and screening

In silico techniques in antibody development have been described as third generation, following second generation in vitro and first generation in vivo methods1. Affinity maturation is a good example of how throughput has soared with in silico methodology. In silico libraries of around 10^25 variants have smashed through the experimental library size ceilings of mammalian display (10^10 variants) and phage display (10^12 variants).

Similarly, the time needed to screen through these libraries for the best sequence has decreased drastically from 3 months with mammalian cells or phage display, down to just 3-4 weeks in silico.

An added bonus is that in silico libraries and screening avoid the potential expression and biophysical issues related with phage display systems, while leaving the door open to promising variants that may not have expressed well in phage.

Challenges remain

The in silico approaches currently used in antibody design remain overwhelmingly knowledge-based. However, machine learning is only as good as the data you train it with. A current challenge is the scarcity of robust experimental open-access datasets, and a lack of widely accepted standards for validating data quality.

Another challenge is the changing skillsets needed by today’s biologists. Data scientists, programmers and technologists are now staple members of biology teams, and everyone needs to learn to speak a common language. Training university students in such interdisciplinary working will ensure the teams of tomorrow are well placed to harness the power of machine learning and in silico techniques.

Future applications

An exciting application of in silico technologies would be to open avenues of investigation previously hampered by experimental roadblocks. For example, G-protein coupled receptors (GPCRs) are a rational antibody target for many disease processes, but are notoriously difficult proteins to isolate out of the cell membrane where they are firmly embedded. In silico modelling could sidestep the difficulty posed by purification and antibody development. This means that proteins that were previously very difficult to raise antibodies against can now be targeted.

RAMP™Rational Affinity Maturation Platform

At Fusion Antibodies, we are firm believers in integrating bioinformatics and in silico techniques to accelerate our workflows and therefore your journey to the clinic. That’s why we developed RAMP™ – our rational affinity maturation platform. It takes RAMP™ just 3 weeks to create a massive library of around 10^25 variants of the parent antibody and to select out the best candidates using rapid in silico screening. The result is a micro-library of the 100 best candidates, selected for binding potential and stability.

Harness the power of our in silico RAMP™ technology to optimise your lead antibody faster.