Unpublished Work 2: Screening Yeast Displayed Nanobodies against Mammalian Targets with Proximity Labeling

When I joined the Liu lab I was recruited to work on the nanobody evolution system. It was not my preference but I wanted to join the lab so I took it on. My PI wanted me to work on building a screening platform for finding nanobodies based on binding to a specific cell surface protein in mammalian cells, with the ideal of finding a way to pick out and evolve nanobodies that bind GPCRs. My first task was to convince him that the paper he thought showed a way to do this was fraudulent, which was an interesting task as a brand new grad student.

The issue is that just looking for binding between yeast and mammalian cells overexpressing a cell surface protein makes it impossible to filter for binding to your protein of interest, versus binding to some other protein on the cell surface, or anything else on the cell surface. Many have tried, but no one has convincingly shown that they can pick a binder to a specific protein out of a library using an approach like this.

So we came up with an alternative based on proximity labeling. Proximity labeling relies on a reaction which produces a highly unstable molecule, which binds to the first thing it touches. This means it can’t diffuse very far.  By attaching a proximity labeling protein to the outside of a GCPR, it could be possible to significantly clean up the binding signal as only nanobodies binding the correct surface protein would be tagged with the molecule, in theory. I was somewhat trepidatious about this project as the Liu Lab is really specialized in directed evolution rather than antibody screening technology. I was worried we may get scooped.

I demonstrated that attaching a proximity labeling protein to the extracellular domain of a GPCR still allowed for functional recognition of the endogenous ligand (At1R and angiotensin) by the GPCR, by setting up an assay to measure intracellular signaling. This was a nice result.

I established a protocol to mix yeast and mammalian cells which were displaying this APEX-GPCR fusion, carry out a labeling reaction, and detect yeast that had become labeled. The reaction is simple, and uses biotin tyramide – and H2O2 (which I established to be nontoxic at the relevant concentrations) with the peroxidase to biotinylate any yeast bound to the (hopefully) correct protein. I showed that a known At1r (GPCR) binder would be labeled by my system, and labeling would be blocked by inclusion of a known small molecule that inhibits binding, a very nice result! In its final form, streptavidin would be an easy way to pull out yeast displaying a binder nanobody.

The problem was that while for an already specific nanobody (AT1r) there was a signal when compared to a control (PDGFR here), off target nanobodies (5F7, binds to Her2 which is on cell surface but not APEX tagged) would also get labeled. Basically, the labeling radius was too big to distinguish between on-target cell binders vs. off target cell binders.

I tried quite a lot of tuning to decrease that issue, less H2O2, shorter reaction, different cell number per well, ratios of yeast to mammalian cells, adding in an antioxidant to quench signal, no dice. I had some other ideas to try, but around this time Chang informed me that based on some conversations he had we are likely to get scooped for this same idea, with a superior technology (MacMillan labeling) which has a smaller labeling radius than peroxidase-based proximity labeling. So, the impression at the time was that a group was 6-12 months ahead of us with a better tech. As far as I could find that paper never came out and I’m not sure what happened there. I was much more eager to move on than Chang, as my impression was we’d come second with an inferior technology. I may have been wrong. Oops!