Drugbaron Blog

February 16, 2021 no comments

The history – and the future – of antibody discovery technologies

Monoclonal antibodies are now well-established as a mainstay of the therapeutic armoury.   More than a hundred antibodies have now been approved as drugs, making up around a third of all approvals over the last three years.  This reflects a number of inherent advantages of antibodies over “conventional” small-molecule drugs: their inherent specificity allows off-target toxicities to be eliminated; their long-life makes them suitable for infrequent dosing, in some cases as little as twice in a year; and, perhaps most importantly, antibodies can be found against essentially any extracellular target.

Small molecules are well-suited to inhibiting enzymes, such as kinases or proteases, but they struggle to block protein:protein interactions.  Antibodies, by contrast, excel at blocking peptide and protein ligands binding to receptors, and, through antibody-mediating clearance, removing circulating proteins from the plasma altogether.

But antibodies can play even smarter tricks.  They can exploit their exquisite specificity to carry toxic payloads to specific locations, they can mimic the binding of natural ligands to act as agonists and they can even be engineered into bispecific (or multispecific) formats to hit several targets at once. Indeed, sophisticated multispecific antibodies are now fulfilling their potential in the clinic, achieving outcomes that are not possible with other modalities.

Delivering these undoubted advantages of antibodies as therapeutics has been dependent on advances in technology.  Four milestones stand out, all with strong links to Cambridge – where it all began in the mid-1970s.

The first technological leap was the invention of hybridoma technology, by César Milstein together with his post-doc Georges Köhler working at the world-famous Laboratory of Molecular Biology in Cambridge.  By fusing individual B cells to immortal cancer cells, they were able to create cell lines that could make essentially infinite amounts of a single antibody.  Recognised with a Nobel Prize for Physiology or Medicine in 1984, the ability to make monoclonal antibodies at scale laid the foundation for antibodies as therapeutics.

But creating a monoclonal antibody against a target of interest was still laborious, depending on immunising mice and letting nature create the antibody that the scientists eventually cloned out and immortalised.  That changed in the late 1980s, with the invention of phage-display libraries by Greg Winter and John McCafferty, also in the Cambridge-based Laboratory of Molecular Biology.  Recombinant library technology solved two problems in one: it allowed researchers to create human antibodies directly (rather than grafting bits of mouse monoclonal sequences into a human framework) and it allowed rapid screening in vitro, obviating the need to work in mice at all.  This technology formed the basis of the legendary Cambridge Antibody Technology, which began operations at the Babraham Research Campus, where DrugBaron is now based, in 1990.  Winter, together with George Smith and Frances Arnold, was also duly recognised with a Nobel Prize for this work in 2018.

Over the following 20 years, libraries became the go-to technology for antibody discovery, undergoing a slow evolution of the basic library design, with different options emerging with different advantages and disadvantages, from Fab libraries in yeast in place of single-chain antibodies in bacteria, through to the use of camelid and even shark antibodies.  But none of these addressed the fundamental limitation of libraries: size.

Nature solves the problem of creating an antibody specific for any pathogen the body might meet in a combinatorial way.  By shuffling amino acids that make up the “tips” of the antibody protein that interact with its cognate antigen, the mammalian adaptive immune system creates vast numbers of candidates, and selects and optimises any that bind even weakly to the invader.  There are so many possible sequences that you could make more different human antibodies than there are atoms in the universe – about 1080 – which is clearly a lot more than could ever fit in a library housed in a test tube!

To keep a library a practical size for storage and screening, you are limited to about 1013 or so elements (analogous to books on the shelves of a real library).  Such libraries, therefore, can only ever sample a tiny, tiny fraction of all possible antibody sequences.  For simple applications (if you just want an antibody against a given protein, but don’t care how it binds, for example) this may not matter too much – as long as there is one sequence among the 1013 options that binds, if you pull that out, then ‘job done’.

Only, too often, it isn’t ‘job done’.  For a start, antibodies found this way often carry serious “baggage” hidden in their sequence, which hamper development as therapeutics.  They often do not fold well, have poor solubility and a tendency to aggregate.  They often contain unstable amino acid combinations, or are susceptible to post-translational modifications either on storage or once injected.  They often contain sequences recognised by T cells, triggering the development of anti-drug antibodies that neutralise your drug after its injected on chronic use.  In short, the majority of the time, finding a clone that binds your target is just the start, rather than the end, of the discovery journey.

To make matters worse, many libraries with 1013 slots on the bookshelf, fill many of them with sequences that don’t even fold well enough to make a functional antibody.  Not only does these “space-fillers” reduce the effective size of the library, worse they have a tendency to be “sticky” and generate false-positive hits during screening.

Cue the third milestone in antibody discovery: a mouse with a fully human antibody repertoire.  Another first for Cambridge, as Allan Bradley and his colleagues at the Sanger Centre used state-of-the-art genome editing to put all the human gene components that are shuffled to create antibodies into exactly the right places in the mouse genome, building on the work of earlier pioneers at Mederex.  This mouse, the Kymouse™ was the foundational technology for Babraham-based biotech company Kymab, which was recently acquired by Sanofi for £1.4billion. 

Exploiting the serial way antibodies are evolved in vivo solves the combinatorial problem and biases the system towards the selection of well-behaved human antibody clones – often shortening the path to the clinic.  But in vivo ‘screening’ has its downside too – you cannot impose any functional constraints on the binding profiles that are delivered.  You cannot easily select for multi-specificity, species cross-reactivity or particular functional consequences of binding, such as receptor agonism.

Humanized mice, and similar in vivo selection methods, have therefore complemented libraries rather than replaced them.  Optimising your discovery and development path to the clinic has never stopped relying on guesswork, and no small amount of luck, choosing how best to navigate the trade-offs associated with each kind of discovery platform. As a result, many drug discovery companies have defaulted to regularly using both library and engineered animal technologies in parallel on high priority projects. Until now.

The fourth milestone in antibody discovery is just beginning to unfold.   The holy grail is a platform that offers the advantages of both in vitro AND in vivo screening at the same time.  A platform that rapidly generates high quality fully human antibodies with minimal baggage and no need for immunization cycles in animals. No more compromises, just highly developable antibodies even with rare or complex binding profiles and functional capabilities.

Jonny Finlay

Two years ago, star protein engineer, and founder of Medicxi’s UltraHuman, Jonny Finlay had an idea how to get the benefits of in vivo entirely into a petri dish.  He takes up the story: “To get the best of all worlds, we had to find a way to fill up a library with only high-quality candidates – fill the shelves with real books, for want of a better analogy.  And then we had to mimic the way the immune system iteratively closes in on the optimal sequence, compressed into a single pass.”

“To date, the classic philosophy of existing libraries is that they are pure chemistry and will work by trying to hit the bullseye in one fell swoop, picking the best sequences directly from the library and presenting them as the output.  The trouble is, with only at best 1013 options, and much of that non-functional, the best sequences we find are still usually far from optimal.  What follows is a time-consuming and expensive process of manual engineering to convert the prototype into a usable drug. And many times, we don’t do a very good job even still. All too often, we get caught in an iterative cycle of improving one characteristic while impairing another that then also needs to be fixed. The frequency with which this vicous cycle gets triggered has historically robbed display technologies of much of their promise.”

“We have called our new platform, Galaxy™ to reflect the fact that all the empty space has been taken out of the library, to fill the available shelf space with nothing but the stars. This approach deliberately mimics the stepwise, iterative approach to antibody selection that nature uses in the B cell system.  Instead of hoping the best sequences are in the starting library, Galaxy™ evolves the library during selection until development-ready clones pop out at the end.”

Finlay is keeping the precise details of how Galaxy™ works its magic firmly proprietary.  But there are at least two unique features that are key.  The first he calls “smart randomness” – instead of constructing the initial library with random sequence combinations, Galaxy™ used a proprietary algorithm to prioritise well-behaved sequences that encode highly developable antibodies.  Put another way, all 1013 books in the Galaxy™ library are best-sellers.

The second unique feature is the stepwise evolution of the library that increases the effective combinatorial space being probed out as far as 1040 and beyond.  “Galaxy™ scans billions of times more of the entire probability space than has ever previously been possible with an in vitro library” Finlay told DrugBaron.  “That’s not an incremental improvement – it’s the first time library technology has directly mimicked nature.”

In 2018, Galaxy™ was just a lightbulb moment – a vision of how to take in vitro antibody discovery libraries to the next level.  But now, two years of hard work and a million pounds of investment later, it is a reality.  And before telling the world about his brainchild, Finlay was keen to prove that the theoretical advantages played out in practice.  So he has been busy using his new toy to find novel antibodies against well-known targets – and he is ecstatic with the results.  “Going back to the drawing board and challenging long-standing dogma isn’t easy. When we decide to strip a technology back down to its constituent parts and rebuild, it means we risk being wrong at the end. The fact that the technology is genuinely working how we wanted it to means there really isn’t a reason now for anyone to choose immunizing animals over Galaxy™ to discover new therapeutic antibodies. The antibodies being generated are truly ‘fully human’, built on templates with proven performance in man and ideal for the next generation of multispecifics.” he enthused.

Antibody discovery has come a long way since Köhler and Milstein created the first monoclonal antibody in Cambridge in 1974.  Almost half-a-century later, Cambridge remains the hottest crucible of innovation in antibody discovery technology, and Galaxy™ looks well set to take its place in the history books.

Galaxy™ is now commercially available for discovery projects through RxBiologics, a new subsidiary of RxCelerate created in partnership with Jonny Finlay.

5 1 vote
Article Rating

Yearly Archive

Would love your thoughts, please comment.x