San Diego

And I wish I been out in California
When the lights on all the Christmas trees went out

Jagger & Richards

Festive conference blog here, dreaming of my December days in sunny California where I attended Antibody Engineering and Therapeutics 2023. Replete with Stones lyrics, Science and Sunset imagery. Hope y’all enjoy!

Touchdown on Tuesday evening at 5pm Western time. I was staying in the Gaslamp Quarter in Downtown San Diego before moving to the conference hotel the next day. Highly recommended area for food and drink!

The first day began with the pre-conference workshop where I happened to be presenting the latest OPIG antibody related manuscript:

https://www.biorxiv.org/content/10.1101/2023.12.08.570786v1

The workshop focused on the promise and challenges of AI and machine learning in antibody discovery. Phillip Kim (https://www.kimlab.org/) gave an overview of deep learning and the state-of-the-art methods revolutionising antibody, protein and peptide design. He briefly discussed his manuscripts on docking (https://doi.org/10.1093/bioadv/vbad072), conditional generational of protein structure (https://doi.org/10.1038/s43588-023-00440-3) and the use of flow matching methods to design peptides (https://www.biorxiv.org/content/10.1101/2023.06.25.546443v1).

Wei Wang discussed his work with the RESP AI model (https://www.nature.com/articles/s41467-023-36028-8) and the power of quantifying uncertainty in model predictions. Possu Huang discussed antibody generation with Ig-VAE (https://doi.org/10.1371/journal.pcbi.1010271), tandem sequence and structure design with Protpardelle (https://doi.org/10.1101/2023.05.24.542194) and conditional generation with Sculptor (https://doi.org/10.1101/2022.12.22.521698). David Prihoda discussed BioPhi (https://doi.org/10.1080/19420862.2021.2020203), and Karolis Martinkus finished the session discussing AbDiffuser (https://arxiv.org/abs/2308.05027).

Now rather than go into the details of each presentation, which were mainly available in the references, I will try to give an overview of the general themes of this workshop. JP, our host, opened the session by asking us to consider how fast the field is moving and how far we are from having truly de novo therapeutic antibodies making it to the clinic, i.e. without the need for large experimental screens and iterative trial and error development. The general opinion was that one day, in perhaps 5-10 years’ time, this will be possible. At present we remain heavily dependent on experimental methods with the latest deep learning techniques merely augmenting existing pipelines, and perhaps providing incremental progress but certainly not a revolution. However, there are many reasons to be excited and everyone stressed that the pace of progress in deep learning was staggering and hard to keep up with… Each year there is a new technique, for example ‘Diffusion Modelling’ will give way to ‘Flow Matching’ which will give way to ‘Jigsaw Puzzling’ and these will be superseded by ‘Tumbling Dice’ and so on… It is very hard to know when, or if, the next paradigm shift will occur, it could be next week, or it could be never!

That evening was my first experience of jet lag. Talking over drinks to the lovely people from AbSci (shout out to Sharrol and his team) at 6pm but feeling like I was in a club at 2am was strange. In keeping with my warped experience of time, at this point I will break with chronological narrative and instead summarise some of my favourite talks and experiences as well as the recurring themes.

Recurring themes (alphabetically ordered):

• Agonism: Multiple talks were leveraging agonism of immune receptors such as PD1 to turn down unwanted immune responses. The challenge here is developing an agonist antibody, rather than a traditional, and comparatively, easier blocker. Key to this venture are an understanding of things such as cross-linking, ligand competition and the biological mechanisms happening within your screening process.
• AI arms race: If you are not using AI in your antibody discovery pipelines you are getting left behind. Each company got up and told us how they were using different types of AI models and bigger datasets to generate the antibodies of choice for their customers, prospective customers, or their in-house portfolio. The details were often sparse, with presenters careful to not reveal their secrets but instead focus their power and potential. I won’t name drop the various organisations; however it is worth checking out the companies listed on the AET website. This time next year it will be very interesting to see if a dominant player in the antibody development market emerges. Are one of these companies sitting on something truly special in their toolkit or are they simply derivatives and customisations of existing and publicly available tools. Only time will tell.
• Constant region rules: Related to the above, multiple talks provided case studies where they were able to affinity engineer antibodies purely by AI informed mutations in the constant region, i.e. away from the CDRs. This is in keeping with the 2023 paper where the development process was guided by language models (https://doi.org/10.1038/s41587-023-01763-2). It is brilliant to see how a very simple AI driven suggestion appears to have directly resulted in better antibodies that are on their way to the clinic.
• Fc receptor targeting and engineering: This was my first antibody development conference. You may not be aware, but the heart of this author is firmly tied to TCRs. Until joining OPIG I didn’t give a hoot about antibodies, so my awareness of what people usually talk about at these conferences is quite poor. Therefore, I learnt a lot about Fc receptor function and the various ways it can be targeted to alleviate antibody mediated autoimmunity. As well as Fc receptor modifications that improve developability and the importance of this region to all antibody engineers. There were great talks from Prof Sally Ward (https://www.wardoberlab.com/), Xencor and Argenix (https://doi.org/10.1038/s41467-022-33764-1), as well as several others I am forgetting. These talks iterated how much work goes into half-life extension, FcRn targeting and general engineering. An enormous amount of work has been directed at this area of the antibody, and I felt a bit ashamed that I have never given it this much thought before…
• Plug and play: This phrase came up many times. The basic idea is to try different backbones and constructs to reach our desired molecule. Quite quickly this builds up the image of antibody engineers and their molecular toolkit, swapping domains and grafting CDRs until they hit their goal.
• Two-factor killing/cis-targeting/dual-targeting: I cannot wait to see when this technology hits the clinic. The question is simple, how can you target a specific cell subset or tumour type and leave the rest of the body untouched? One ligand may not be specific enough to prevent off target reactivity and associated adverse events (side effects). Therefore, is it possible to direct your depleting antibody, bispecific or antibody drug conjugate to the target tissue using two markers. For example, your drug may only deliver its cytotoxic payload if it binds to two cell surface proteins which are both upregulated on the surface of tumour cells at the same time. Each marker alone may be present in healthy tissue, but if you can develop a bi- or tri- specific which must engage both to trigger some kind of functionality then you drastically increase specificity. Your drug hits the tumour cells, but leaves surrounding healthy tissue and organs intact. Great stuff!

Favourite talks:

• In line with the last point, two of my favourite talks at the conference were from Asher Bio and Cartography. Both dealt with the cis/dual targeting of relevant T cell subsets, or tumour tissue respectively. Asher showed that if you can hit the “better effector” T cells with IL2 then you limit off-target effects and boost up the cell types which can do the killing. Cartography stressed the importance of using single cell sequencing and showed how they had identified numerous combinations of targets that better separate tumour tissue from healthy.
• Like this, another fascinating talk came from Radiant who have developed a potentially game changing technology which combines multiple Fab fragments of different specificity into a kind of sticky ball called a “Multabody.” The idea being the specificity of the combination of Fabs provides a new platform for dual/cis targeting and effector function.
• Of the many excellent keynote talks, those from Xencor, Moderna and Genentech most grabbed my attention. Xencor relayed the journey of a young biotech as it focused on developing new Fc domain technology to become a key player in the antibody market. Moderna (longitudinal responses to vaccines and reinfection) and Genentech (the role of fibroblasts and the stroma in tumours) delivered talks that felt incredibly exploratory and academic. It was great to see how basic science aimed at understanding the human response to vaccination or disease underpinned the direction of therapeutics in these large organisations.
• Finally, Astra Zeneca flipped the idea of anti-CD3 bispecific therapeutics on its head. Rather than blanket targeting CD3+ T cells, they presented a strategy where the use of a bispecific with common HLA presenting an immunodominant viral peptide (instead of anti-CD3) can engage healthy/functional anti-viral CD8+ T cells to elicit anti-tumour activity. Similar to the Asher Bio talk, the idea here is to bring specific subsets of T cells (non-exhausted anti-viral effectors) that deliver the best immune response to the heart of the tumour and thus improve upon existing T cell engagement.

Food:

You wouldn’t have gone hungry at this conference. The food was spectacular and never seemed to run out. However, if you needed even more calories to fuel your intellectual exertions then you could ladle “half & half” into your coffee to make a delicious caffeine, lactose, and lipid bomb. You get the creamy coffee, but there is also some milk in there to ensure you have the easier consistency when pouring. Genius. I so wish we had “half & half” in the UK…

Exploration:

In the evenings you could walk from the conference hotel up to, and beyond, the USS Midway to get your steps in and let your mind wander. This same walk could be done at 6am if you were still considerably jet lagged and couldn’t sleep, or your co-workers back in the UK forgot to let you into a Zoom meeting on TCRs that you woke up at 5am to join. The morning sun coming up over San Diego Bay was stunning. Downtown San Diego was quite nice, but really for the views and to get closer to the Pacific it was necessary to head up the coast. I stayed a night in Point Loma and used that as a base to get to Sunset Cliffs, La Jolla and then finish off at Ocean Beach, before heading to the airport the next day. Unfortunately, I never made it to Coronado, Chula Vista or Cabrillo National Monument. Hopefully, next year.

Sunset Cliffs was beautiful. La Jolla, meaning “The Jewell” in Spanish, is a stunning place. However, my favourite was Ocean Beach, or “O.B.” as the locals call it. The walk along the length of Newport Avenue felt a bit like being in a film from 1960’s America, with some Blackpool vibes thrown in. From Newport Avenue I made it to the closed off pier. Then for over an hour I walked along the cliffs, looked out at the Ocean and watched the waves hit the rocks as the sun went down. Pretty special. I was thinking about science the whole time of course.