1
Continuous Evolution of Proteins In Vivo

Alon Wellner1, Arjun Ravikumar1, and Chang C. Liu1,2,3

1University of California, Department of Biomedical Engineering, 3201 Natural Sciences II, Irvine, CA, 92697, USA

2University of California, Department of Chemistry, 1102 Natural Sciences 2, Irvine, CA, 92697, USA

3University of California, Department of Molecular Biology and Biochemistry, 3205 McGaugh Hall, Irvine, CA, 92697, USA

1.1 Introduction

Directed evolution is a powerful approach for engineering new biomolecular and cellular functions [1-3]. In contrast to rational design approaches, directed evolution exploits diversity and evolution to shape the behavior of biological matter by applying the Darwinian cycle of mutation, selection, and amplification of genes and genomes. By doing so, the field of directed evolution has generated important insights into the evolutionary process [4-6] as well as useful RNAs, proteins, and systems with wide-ranging applications across biotechnology and medicine [7-11].

To mimic the evolutionary process, classical directed evolution approaches carry out cycles of ex vivo diversification on genes of interest (GOIs), transformation of the resulting gene libraries into cells, and selection of the desired function (Figure 1.1). Each iteration of this cycle is defined as a round of evolution, and as selection stringency increases over rounds, either automatically through competition or manually through changing conditions (or both), this process can lead GOIs closer and closer to the desired function. This overall process makes practical sense for a number of reasons, especially for the goal of protein engineering (i.e. GOI encodes a protein). First, ex vivo diversification is appropriate, because test tube molecular biology techniques such as DNA shuffling, site-directed saturation mutagenesis, and error-prone (ep) polymerase chain reaction (PCR) [2] are capable of generating exceptionally high and precise levels of sequence diversity for any GOI. Second, transforming diversified libraries of the GOI into cells is appropriate, because each GOI variant needs to be translated into a protein in order to express its function, and cells, especially model microbes, are naturally robust hosts for protein expression. Third, carrying out selection inside cells is appropriate, because (i) cells automatically maintain the genotype-phenotype connection between the GOI and expressed protein that is necessary for amplification of desired variants, (ii) we often care about a GOIs function within the context of a cell, especially as metabolic engineering and cell-based therapy applications mature, and (iii) the use of cell survival as the output for a desired protein function allows millions or billions of GOI variants to be simultaneously tested - it is easy to culture billions of cells under selection conditions - in contrast to ex vivo screens that are much lower throughput. Survival-based selections are not always immediately available, but one can often find a way to reliably link the desired function of a protein to cellular fitness.

Figure 1.1 A schematic illustration of a typical directed evolution setup. (a) A GOI is diversified ex vivo, typically by applying an error-prone PCR to generate a GOI library. (b) The library is then cloned into an expression vector and transformed/transfected into cells that are subjected to (c) outgrowth and selection for enhanced protein activity. (d) Plasmid DNA that is enriched for library members with increased properties is extracted and (e) subjected again to diversification and selection. The directed evolution cycle is iterated until the desired outcome is achieved or until diminishing returns (a plateau is reached).

While sensible, the practical requirement that diversification should occur in vitro but expression and selection should occur in vivo in this classical directed evolution pipeline creates significant suboptimalities. First, the number of steps that can be taken along an adaptive path becomes few, since each round of in vitro mutation, transformation, and in vivo selection takes several days or weeks to carry out. Second, limited DNA transformation efficiencies result in strong bottlenecking of diversity that can mitigate the probability of finding the most optimal solutions in sequence space. Third, the number of evolution experiments that can be run simultaneously is minimal, because in vitro mutagenesis, cloning, and transformation are experimentally onerous, demanding extensive researcher intervention [12]. These shortcomings keep two highly promising categories of experiments largely outside the grasp of classical methods: first is the directed evolution of genes towards highly novel functions that likely require long mutational paths to reach (e.g. the optimization of multi-gene metabolic pathways or the de novo evolution of enzyme activity); and second is the large-scale replication of directed evolution experiments, needed in cases when many different functional variants of a gene are desired (e.g. the evolution of multiple synthetic receptors for a collection of ligands) or when statistical power is required in order to understand outcomes in experimental evolution (e.g. probing the scope of adaptive trajectories leading to resistance in a drug target).

An emerging field of in vivo continuous directed evolution seeks to overcome these shortcomings by performing both continuous diversification of the GOI and selection entirely within living cells [13]. In this way, GOIs can be rapidly and continuously evolved through basic serial passaging of cells under selective conditions. This removes the labor-intensive cycling between in vitro and in vivo steps and the DNA transformation bottlenecks associated with the classical pipeline, creating a new paradigm for directed evolution that is limited only by the generation time of the host cell and the number of cells that can be cultured. These limitations are usually negligible - in most host organisms for directed evolution such as Escherichia coli and Saccharomyces cerevisiae, generation time is fast (20-100 minutes) and the number of cells that can be cultured is massive (108-109 ml-1) - so the potential power of continuous systems is enormous. Moreover, in vivo continuous directed evolution is amenable to high-throughput experiments, because serial passaging is straightforward and can be automated at scale or converted to continuous culture using bioreactors [14-16]. In this chapter, we discuss various systems that partially or fully achieve in vivo continuous evolution (ICE).

1.2 Challenges in Achieving In Vivo Continuous Evolution

Before discussing how ICE can be realized, we shall first clarify why this is a challenging problem. The difficulty of achieving ICE of GOIs lies in the fundamental relationship between how fast one can mutate an information polymer and its length. Several theories predict that organisms face an "error threshold" at mutation rates on order 1/L (where L is the length of the genome), near which selection cannot maintain fitness, leading to gradual decline towards low fitness, or above which one is nearly guaranteed a lethal mutation every cycle of replication, leading to rapid extinction [17-20]. Because cellular genomes are large (e.g. ~5 × 106 in E. coli, ~1.2 × 107 in S. cerevisiae, and ~ 3 × 109 in humans), this implies that evolution strongly favors low genomic mutation rates (e.g. ~5 × 10-8 substitutions per base [s.p.b.] in E. coli, ~10-10 s.p.b. in S. cerevisiae and ~3 × 10-9 in human somatic cells) [21-23]. Experiment confirms this prediction. Drake observed empirically that mutation rates scale as 1/L across many organisms [17]; evolution experiments have shown that when mutator phenotypes do arise, they are accompanied by fitness costs and only transiently persist [19,24-26]; and more direct tests in yeast find that there is indeed a mutation-induced extinction threshold at ~1/L, above which yeast cannot propagate [18]. Yet individual GOIs are small in comparison with genomes, so they are capable of tolerating much higher error rates. In fact, they require much higher per base error rates than genomes to generate the same amount of total mutational diversity, because they have fewer bases. Following the 1/L scaling, a typical 1 kb GOI should be able to tolerate mutation rates on order ~10-3 s.p.b.

Therefore, the primary challenge in achieving rapid ICE is how to develop molecular machinery or other strategies that target rapid mutagenesis to only GOIs, allowing the host genome to replicate at mutation rates below its low error thresholds but driving the GOI at the high mutation necessary for fast generation of sequence diversity. When considering the level of targeting in the ideal case, the formidability of this challenge...