Natural evolution has yielded small molecules and macromolecules with a diverse array of activities, many of which have been harnessed by human society to advance industry and medicine. However, evolution is primarily selective for survival of the organism and/or gene rather than any particular activity directly, and is largely constrained by its cellular environment — interference or negative interactions (such as precipitation or cytotoxicity) between the cell and effector (the RNA or protein product) will generally act as a counter-selective pressure for that effector. Evolution also lacks foresight. In fact, evolution’s main advantage over human efforts to develop novel drugs or catalysts seems to be the sheer number of molecules that it has been able to sample over the past ~3.5 billion years — as well as the fact that the cell is a relatively robust and efficient core platform on which to build.

Ultrahigh-throughput screening offers researchers a powerful tool to begin sampling larger regions of sequence space, and thus partially addresses one of evolution’s main advantages. This approach has particularly grown in promise and scale as DNA sequencing has become faster, cheaper and more accurate in recent years, greatly facilitating identification of macromolecular — and in some cases, small molecule — library members. Combining the ability to screen large libraries with a clear concept of our desired activity can provide useful insights into the relationship between molecular-scale structure and function, in addition to facilitating the development of novel, ‘artificial’ candidates that evolution may not have explored or been exposed to before. This latter feature is more significant than it may first appear, as many of the challenges faced by modern humanity are likely to be the product of survivorship or observation bias — i.e. those problems that natural evolution has already solved or could feasibly solve are less likely to present challenges to us in the first place. Nevertheless, the vast majority of existing approaches are still constrained by cellular expression. Additionally, low-throughput assays for a given activity are often difficult to adapt for ultrahigh-throughput approaches.

To help address these challenges, we have developed a platform which is capable of displaying generic DNA and protein on biologically inert and microfluidic-compatible polyacrylamide microbeads. We envision this as an ultrahigh-throughput-compatible, robust, abiotic tool for maintaining the genotype-effector (“phenotype”) linkage over several experimental steps (e.g. PCR, in vitro expression, and the assay itself), as well as a generally applicable module for protein purification, solid-phase (DNA) synthesis, etc. Our platform’s compatibility with in vitro expression may allow exploration of novel sequence space which is poorly accessible in cellulo, and we hope that this will provide novel opportunities for naïve phenotypic screening and drug lead compound discovery.

In this thesis, I will first present my work in characterising and enhancing protein immobilisation on these beads using a fully covalent, suicide substrate-based linkage module we developed (polyacrylamide-benzylguanine-SNAP-SpyCatcher-SpyTag; Chapter 1). I find that the beads are highly permeable to proteins, and that protein-display capacity is largely determined by methacrylatebenzylguanine’s input concentration, copolymerisation efficiency and accessibility, but can reach at least 100 μM in practice.

Next, we combine this capture method with other protein modules to create two purification and multivalent (up to 30X) assembly workflows for SpyTagged proteins (Chapter 2). I explore the impact of construct valency on the potency of apoptosis induction for two TRAIL-receptor agonists, observing that potency is strongly dependent on agonist valency and therefore likely also on microdomain formation (‘lipid rafting’) for TRAIL-receptor. We use multimerisation to achieve a ~5 pM EC50 for a multivalent assembly of a nanobody, compared to an EC50 of at least 115 nM of the same nanobody in its monovalent form (a more than 2.3×10⁴-fold potency improvement). We simultaneously present clickable modules to control valency which are ‘plug-and-play’ with our protein capture module from Chapter 1, and which can be readily expressed and employed by others.

In Chapter 3, I demonstrate and refine an ultrahigh-throughput-compatible phenotypic screen for bacteriolysis. I show that this assay is sensitive to antimicrobial peptide-induced lysis under treatment conditions which could theoretically be achieved using microfluidics and bead protein capacity levels as demonstrated in Chapter 1, although the low potency of antimicrobial peptides makes this practically non-trivial. I therefore begin the process of optimising the practical steps necessary to effectively deliver such a large on-bead payload, beginning with our protease-based solubilisation step and in vitro transcription-translation in Chapter 4.