Methods for the directed evolution of biomolecular interactions

Abstract

Non-covalent interactions between biomolecules such as proteins and nucleic acids coordinate all cellular processes through changes in proximity. Tools that perturb these interactions have and will continue to be highly valuable for basic and translational scientific endeavors. By taking cues from natural systems, such as the adaptive immune system, we can design directed evolution platforms that can generate proteins that bind to biomolecules of interest. In recent years, the platforms used to direct the evolution of biomolecular binders have greatly expanded the range of types of interactions one can evolve. Herein, we review recent advances in methods to evolve protein-protein, protein-RNA, and protein-DNA interactions.

The importance of manipulating biomolecular interactions

Non-covalent interactions between biomolecules – DNA, RNA, proteins, lipids, and sugars – underlie all biophysical processes in the cell. Biological signaling is largely driven by proximity between biomolecules [1, 2]; therefore, whether biomolecules are near one another, interacting, or critically, not interacting, is central to the organization and function of the cell. Moreover, aberrant interactions between biomolecules are often the drivers of disease and can be targeted with inhibitors for therapeutic development. For example, the interactions between BCL-2 family proteins and their binders can be blocked by a protein-protein interaction (PPI) inhibitor to treat cancer [3, 4]. Biomolecular interactions can also be reprogrammed for beneficial purposes, exemplified by the recent explosion of CAR-T cell therapies [5] and PROTACS [6], which engineer cells to respond to novel antigens and redirect protein degradation pathways to target proteins, respectively. As such, methods to understand, reprogram, and create de novo biomolecular interactions are increasingly important to understanding molecular biology and creating biotechnologies.

In this review, we detail the use of directed evolution (see Glossary) as a method to modulate biomolecular interactions. In particular, we highlight continuous evolution, and most prominently Phage-Assisted Continuous Evolution (PACE), as a powerful technology for evolving proteins to interact with proteins and nucleic acids. We also describe additional methods that can engineer proteins to interact with DNA, RNA, and other proteins, as well as experimental campaigns to create multi-partner, or higher-order, interactions.

Directed evolution as a technique to evolve biomolecular interactions

Advances in structural methods, such as cryo-electron microscopy (cryoEM), paired with advances in machine learning-based computational approaches such as AlphaFold 2 and RoseTTAFold, have led to a dramatic increase in our ability to study and predict the structures of biomolecules [7–10]. On the other hand, understanding whether and how a given set of biomolecules interact [11] or reprogramming their interaction though defined mutations, remains challenging. However, one technology for the creation of PPIs, evolved by nature, has proven wildly successful as an engineering tool: the immune system. The mammalian immune system is capable of rapidly creating antibodies that bind to a target antigen, often another protein, as part of the body’s defense system. The basis for the immune system’s capacity to solve these complex biophysical puzzles is its use of evolution, essentially, selecting for antibodies that bind to a given epitope. The rapid diversification, selection, and amplification of antibodies allows for the immune system to identify high-affinity interactions [12, 13]. This process can and has been harnessed to create antibodies for a given epitope of interest, revolutionizing basic science and medicine in turn [14, 15]. However, this natural evolutionary process cannot be used to evolve biomolecules other than antibodies. To fill this technological gap, researchers must engineer experimental evolution approaches in the laboratory.

Early work toward harnessing evolution in the laboratory focused on selection methods for enzymes and PPIs. In recognition of this foundational work, the 2018 Nobel Prize in Chemistry was awarded for “the directed evolution of enzymes” to Francis Arnold and for “the phage display of peptides and antibodies” to George Smith and Sir Gregory Winterⁱ. Arnold’s work, which focuses on engineering enzymes by library generation and screening to catalyze new chemical reactions, illuminated how screening individual mutants can be used not only to endow biomolecules with improved or even novel functions [16–18], but also to uncover fundamental knowledge about how the biomolecules function and evolve [19, 20]. The use of evolution in biocatalysts has been extensively reviewed elsewhere [21–23]. Phage display, invented by Smith and Winter, is analogous to the selection of antibodies by the immune system and uses enrichment of phage-encoded protein libraries as a method for identifying novel binders or ligands [24, 25]. Phage display was subsequently expanded to related technologies, such as yeast display [26], mRNA display [27], and ribosome display [28]. These early approaches toward harnessing evolution in the laboratory highlight the power of technologies that can properly and specifically focus evolution on a desired outcome, which is referred to as directed evolution [29–32].

Experimental evolution approaches can generally be categorized as screens, which involve performing individual assays on each variant, often using robotics, or selections, where a fitness advantage is used to enrich variants with desired activities (Figure 1A). Formally, evolution is the process of repeated rounds of diversification and selection. By adding a round of mutagenesis or further diversification to a screen and performing a second round of screening, as is often done in the context of enzyme engineering, such a process would then be classified as directed evolution. Selections entail testing a library of variants simultaneously, where some experimental process is used to enrich variants with a given fitness level. Selection methods can be further sub-divided into isolation-based methods, where target variants are physically separated from the population (e.g., display technologies or FACS), or growth-based selection approaches, where organismal or viral fitness and/or growth are directly tied to the fitness of the target molecule. Likewise, selections can be run either with or without re-diversification/mutagenesis to identify fit variants within the starting library. Whether by screening or selection, several rounds of diversification and identification of active variants allows for the directed evolution of a biomolecule towards a new function.

Figure 1.

A) Methods to identify mutants with a specific function include (left) screening individual variants to characterize function and selecting to enrich variants with the desired function either by isolation or growth. Experimental evolution (right) can be carried out by repeating rounds of diversification and identification of active variants either with screens or selections. B) Types of surface display methods. In general, the bait target is immobilized to a surface, and prey (binding entities) are displayed in various manners, e.g. on the surface of phage, yeast, bacteria, etc., and flowed over the immobilized targets. C) Biosensors enable function to be linked to fitness when surface immobilization is not used. Transcription-based biosensors such as the 1-, 2-, and 3- hybrid constructs induce the localization of RNAP to the promoter region of a reporter gene (i.e. fluorescent protein or luciferase) for screens and isolation based selections or a gene required for host fitness (i.e. antibiotic resistance cassettes or gIII in PACE) for growth based selections. Additionally, protein complementation can be used in a similar manner for isolation-based selections (e.g. split fluorescent protein or luciferase) and growth-based selection (split RNAP).

Selections offer several benefits over screens: assaying larger libraries, having tunable fitness thresholds, in some cases featuring negative selections that allow for assessing multiple characteristics of a biomolecule at once, and in general, fewer labor- and instrument-intensive processes. However, the construction of the selection platform is critical to successful directed evolution campaigns. The primary challenges in designing a selection platform are 1) linking the genotype of the evolving biomolecule to a function of interest, and 2) linking the function of interest to the fitness of the genotype. For example, phage are well-suited hosts for the creation of a selection platform to evolve peptide and protein binders: phage genomes are naturally tied to the peptides/proteins they encode (linking genotype to the biomolecule); phage coat proteins can be fused to the evolving peptide/protein to facilitate phage binding to a target protein via the displayed peptide/protein (linking the biomolecule to function); and finally, bound phage can be isolated from non-binding phage (linking function to fitness). While powerful, phage display is an example of non-continuous evolution, requiring researcher intervention to replicate and possibly mutagenize the phage-encoded biomolecules for additional rounds of selection; this constraint limits these evolutions to only a few rounds of selection.

Continually generating diversity while enriching fit variants via growth-based selections is referred to as continuous evolution. Due to the challenges of linking the fitness of a host to a desired function of a biomolecule, continuous evolution approaches were, until recently, limited to selections that directly link fitness to function, such as selecting for antibiotic resistance [33, 34], and in vitro systems, such as self-replicating RNA ligases [35]. However, the past decade has brought about an explosion of new continuous evolution technologies to solve this challenge [36–45]. For instance, PACE links phage propagation (host fitness) in Escherichia coli carrying a mutagenesis plasmid to the phenotype (function) of a gene within the phage (Figure 2A) [38, 46]. The link between phage fitness and target activity is established by the inducible expression of pIII, a required phage protein, which is provided by the host E. coli cells. PACE, once developed for a desired function of interest (that is, once a robust link between target activity and pIII expression is engineered), can enable hundreds of rounds of selection in days with minimal researcher intervention. For additional information on current continuous in vivo evolution methods, see Box 1.

Figure 2.

A) Phage-assisted continuous evolution (PACE) biosensors for the evolution of biomolecular interactions, including B) specific protein-protein interactions (PPIs), C) DNA binders, and D) PPI glues. A) General schematic for how PACE works. Bacteriophage carry a plasmid that encodes an evolving protein of interest. Phage infect host E. coli cells that contain plasmids that encode a transcription-based biosensor that drives pIII to produce infectious phage for positive selection and a dominant negative form of pIII (pIII_neg) to create non-infectious phage for negative selection. B) Phage carry an evolving protein fused to the N-terminus of proximity dependent split RNAP. Host E. coli cells express two proteins fused to two orthogonal C-terminal T7 RNAP fragment variants with different promoter specificities. If the evolving protein variant interacts with the target protein of interest, this reconstitutes an RNAP that binds to the “CGG” promoter, which triggers pIII production and phage replication. If the variant binds the counterselection protein, it reconstitutes an RNAP that binds the “T7” promoter and leads to the production of a dominant negative phage protein, pIII_neg, which lowers phage fitness. C) Proteins that bind DNA, including transcription factors and Cas9 effectors, can be evolved by fusing them to the subunit of an RNAP and encoding the fusion in phage. Positive selection is driven by the protein binding to a specific sequence upstream of the RNAP promoter to drive pIII expression, and negative selection can be driven via non-specific binding triggering pIII_neg production. D) A genetically encoded bifunctional molecule is expressed by phage and can drive pIII production and phage propagation if it binds to two partners fused to split halves of the split RNAP biosensor to reconstitute active RNAP.

Box 1. Eukaryotic continuous directed evolution.

Like phase-assisted continuous evolution (PACE), eukaryotic continuous directed evolution efforts aim to evolve biomolecules by linking a desired activity to survival and, critically, to focus the evolution on a desired gene. One approach that has been developed is to engineer a native retrotransposon to elevate mutation rates of its corresponding cargo [39]. Additionally, OrthoRep uses an orthogonal error-prone DNA polymerase that specifically drives replications and mutations of a gene expressed in a plasmid in yeast [45]. A similar platform is seen in the recently-realized compatibility of EvolvR for use in bacteria and yeast; here, dCas9 is linked to an error-prone DNA polymerase such that targeted mutagenesis is possible [47]. As in PACE, the challenge then becomes linking activity to fitness, which can be done though employing various biosensors to evolve activities such as catalysis and binding. Though the replication rate of yeast is lower than that of phage, meaning the evolutionary process in principle takes longer than PACE to achieve the same number of rounds of mutagenesis and selection. A noteworthy advantage of yeast continuous evolution is that activity can be evolved in a eukaryotic cellular context, which is arguably more suited to evolving biomolecules that can function in humans than the bacterial environment. Advances such as combining OrthoRep with automated continuous culture technologies [48] and with yeast surface display [49] showcase the great potential of using yeast as a conduit for directed evolution.

Continuous directed evolution in mammalian cells is also a fast-growing yet challenging area of research. Like PACE, mammalian cell continuous evolution approaches seek to use viruses as a conduit for directed evolution, in one method by a double-stranded DNA adenovirus [40] and in another by a single-stranded RNA Sindbis alphavirus, the latter of which is known as VEGAS [41]. For a recent perspective on this field, please see [32].

As mature display-based technologies continue to find new applications and novel continuous evolution systems continue to develop, an expansive experimental evolution toolkit for probing and engineering biomolecular interactions is emerging. This review sets out to highlight recent advances in approaches that use evolution to probe the interactions between biomolecules, organized by technologies for engineering interactions with proteins, interactions with RNA, interactions with DNA, and molecules that influence the interactions of pairs of biomolecules (“higher-order” interactions). In each case, these directed evolution approaches are leading to basic insights into how biomolecular interactions have evolved and to novel biotechnologies and therapeutic approaches.

Engineering interactions between proteins

Along with phage display, many other surface display technologies have been developed that can evolve peptides and proteins to bind protein targets. These include mRNA [50], ribosome [51], yeast [49, 52], and bacteria display [53] (Figure 1B). Display technologies have proven particularly fruitful for the evolution of antibodies, as covered in a recent review [54], and each display technology has its advantages and drawbacks. For instance, while phage display can attain larger library sizes, bacteria and yeast display can accommodate larger proteins. Additionally, certain protein targets are more compatible with yeast display, such as those that are insoluble when expressed by bacteria or those containing post-translational modifications specific to eukaryotic hosts. For more details on display technologies, see a recent review [55].

Although display methods can work well for engineering extracellular interactions, as in discovering ligands for GPCRs [56] and plasma proteins [57], one drawback is that they do not evolve proteins to function in an intracellular context. Biomolecular interactions can depend on a variety of cellular factors, from metabolite concentrations to localization; thus, evolution in a more native biological context is advantageous. For this reason, in vivo evolution systems have gained popularity in recent years, and those that select for binding generally use biosensors that adhere to an n-hybrid or protein complementation assay approach (Figure 1C). In these methods, a protein of interest (bait) and the protein under selection pressure (prey) are each fused to additional proteins, and binding of the prey to the bait protein results in bringing these components into close proximity which assemble to form some sort of selection output. As illustrated in Figure 1C, n-hybrid systems typically involve the localization of an RNA polymerase (RNAP) or other transcription inducer to a reporter gene such as GFP. Similarly, protein complementation approaches can utilize split fluorescent proteins or other optical reporters for assays by screening or isolation via FACS for selections, or a DNA-binding protein/transcription factor pair or split RNAP to produce a protein that allows for survival in a growth-based selection. Biosensors have been widely used for analyzing and screening PPIs, which has been highlighted in previous reviews [58, 59]. For example, 2-hybrid based systems have been used to extensively map potential interactions between the proteome [60], and recent split-luciferase reporter systems enabled the rapid screening of SARS-CoV-2 antibodies [61] and endosomal disruption stimuli [62]. Protein complementation technologies have also been employed in directed evolution campaigns to yield useful binders for applied purposes and advanced study of evolution itself, biochemistry, and structural biology [74].

Both display and complementation technologies have been linked to viral replication to create powerful in vivo experimental evolution platforms that can generate protein binders (Table 1). The PACE technology incorporates protein complementation when used to evolve protein binders. For example, by fusing an insect receptor protein to a DNA-binding domain and Bacillus thuringiensis δ-endotoxin (Bt toxin) to an RNA polymerase subunit, the Liu lab used PACE to evolve Bt toxin to bind the insect receptor and overcome resistance [63]. A recently published paper from the lab also now enables the evolution of binders that contain di-sulfide bonds [65]. Additionally, our lab developed a protein binder PACE system based on complementation of split RNAP [75]. We performed deep-mutational scanning of the Ras/Raf interaction, by generating a library of Raf variants and enriching for Ras binding without mutagenesis—a technique we dubbed Phage-Assisted Continuous Selection (PACS) [76]. Technologies are also emerging for experimental evolution in eukaryotic cells. For instance, “autonomous hypermutation yeast surface display,” deemed AHEAD, combines yeast display with OrthoRep to facilitate continuous evolution of protein binders in yeast [49]. It has been used to evolve camelid single-domain antibodies, or nanobodies, that bind to targets such as the SARS-CoV-2 S glycoprotein. Moreover, efforts at directed evolution in mammalian cells (reviewed by Shoulders) are aiming to use adenovirus or RNA virus variants as a vector, analogous to the role of phage in PACE [32]. While the field still faces challenges, these technologies do have potential for evolving a variety of activities, including protein binders. Indeed, in its premier paper, “Viral Evolution of Genetically Actuating Sequences,” or VEGAS, was used to evolve nanobodies that bind to GPCRs [41].

Table 1.

Example continuous evolution methods to evolve binders to nucleic acids and proteins.

Binding partner	Method	Application	Evolution Environment	Reference
Protein	PACE	Evolve a binder of an insect receptor to prevent antibiotic resistance	E. coli	[63]
Protein	PACE	Evolve specific BCL-2 family protein binders to study the evolution of PPI specificity	E. coli	[64]
Protein	PACE	Evolve a di-sulfide bondcontaining protein binder to Her2	E. coli	[65]
Protein	OrthoRep	Evolve SARS-CoV-2 S glycoprotein nanobodies	Yeast	[49]
Protein	Adenoviral PACE/VEGAS	Evolve GPCR nanobodies	Mammalian cells	[40, 41]
DNA	PACE	Evolve RNAPs with different promoter specificities	E. coli	[66, 67]
DNA	PACE	Evolve DNA binding proteins to bind various sequence motifs	E. coli	[68–70]
DNA	PACE	Evolve dCas9 variants with broadened PAM compatibility	E. coli	[71, 72]
Higher Order Interactions	PACE	Evolve bifunctional binder to a zipper peptide and ULK1	E. coli	[73]

All of the above approaches measure, select for, or evolve a desired PPI. However, preventing interactions with an undesired protein is often just as critical as interacting with a desired protein, both in terms of understanding the emergence of molecular recognition through evolution and for developing selective biotechnologies. Counterselections or secondary screens can be deployed, but they are then decoupled from activity evolved in the primary evolution/screen. For example, in recent work, we developed a new PACE-based system for evolving selective PPIs [64] (Figure 2B). To accomplish this, we employed two separate RNAP-based protein complementation systems using our group’s proximity-dependent split RNAP biosensor technology [75, 77]. In this system, the phage-encoded protein of interest is simultaneously and continuously evolving to interact with a target protein and to not interact with a non-target protein. We used this platform to explore the roles of chance and contingency in the evolution of binding specificity of the BCL-2 family proteins. However, the core platform can, in principle, be used to engineer novel specificity into PPIs of interest. Another recent example of PACE with negative selection was used to reprogram the binding specificity of proteases [78], a class of proteins that both interact with and cleave specific proteins based on sequence motifs. Though improving the enzymatic activity of proteases is often a focus, engineering specificity in their interactions with their protein targets has proven difficult to achieve until now.

Engineering interactions with RNA

In its simplest form, engineering interactions between nucleic acids like RNA and DNA can be very straightforward. For instance, binding to a single-stranded DNA (ssDNA) sequence can be achieved via a complementary single-stranded DNA or RNA molecule through easily programmable Watson-Crick-Franklin base-pair interactions. However, engineering protein-nucleic acid interactions can be quite challenging, as there is no easily decipherable “code” to program proteins to interact with a specific nucleic acid. In nature, proteins have evolved to recognize specific RNA motifs through interactions with specific base sequences, the chemical modification states of bases, and through interactions with RNA structures. These RNA binding proteins are involved in regulating RNA turnover, translation, localization, splicing, and post-transcriptional modifications [79]. Directed evolution techniques have been applied to alter the binding specificity of proteins with each type of protein-RNA interaction using both in vitro binding methods, such as phage display, but also using in vivo three-hybrid biosensors (Figure 1C). These studies have been used to engineer novel RNA specific binding interactions, as well as to study the evolution of RNA binding proteins.

Pumilio and FBF homology proteins (PUF proteins) can recognize single-stranded RNA (ssRNA) through sequence-specific interactions. Typically, PUF proteins are composed of eight 36-amino acid repeat Pumillo homology domains (PUM domains) flanked by N- and C- terminal regions; these domains form a crescent shape with eight ssRNA nucleotides binding to the concave face of the protein. Each PUM domain recognizes a specific RNA base via three conserved side chains (a t–ipartite recognition motif - TRM), and thus, the specific PUM domains and their order dictate the sequence specificity of the PUF protein [80]. Site-directed mutagenesis of the TRM followed by screening variants via electrophoretic mobility shift assays (EMSAs) has facilitated the interconversion of PUM domains that bind to adenine, uracil, and guanine [81]; however, naturally occurring cytosine PUM domains have not been discovered. Flipovska et al. used site-directed random mutagenesis paired with a yeast 3-hybrid growth-based screening assay to identify PUM domain mutants capable of binding to cytosine, and thus, created a universal code for RNA recognition by PUF proteins [82]. These sequence specific PUF proteins have been used in a wide variety of applications, such as the development of a sequence specific RNA endonuclease [83]. Selection-based techniques have been used to determine the RNA sequence specificity for other PUF proteins [84] and to study the evolution of homologs that recognize different length RNA sequences (8–10 nt) [85].

RNA can form well-defined structures that proteins in nature have evolved to recognize, and directed evolution can be used to evolve proteins that bind specific RNA structures. For example, phage display has been used to evolve antibody fragments (Fabs) that bind to the internal ribosome entry site (IRES) of hepatitis C virus (HCV) [86] and competitively inhibit binding of ribosomal proteins to the HCV IRES. Yeast display was utilized to reprogram the human protein U1A to bind a structured element from the HIV viral RNA genome [87], and this protein-RNA interaction serves as a key building block for our group’s “CIRTS” platform for engineering RNA regulatory proteins [88, 89]. A weakly active dCas13a variant was diversified using random mutagenesis and then selected by FACS to improve the ability of dCas13a to target and repress translation of mRNA targets [90]. Lastly, with similar goals as the counter selections employed to evolve specificity into PPIs, novel selection strategies, such as library-vs-library selection, have been used to evolve orthogonal RNA-RNA binding protein pairs [91].

In addition to evolving proteins that bind to mRNA, directed evolution has also been used to evolve tRNA-aminoacyl-tRNA synthetase (aaRS) pairs [92]. In this study, the authors first computationally identified potential orthogonal tRNAs and tested for orthogonality against natural aaRS proteins in E. coli and for function with their cognate aaRS in an E. coli host. However, when the tRNAs that passed these screens were recoded for an amber supression codon, they no longer functioned with their cognate aaRS. Thus, directed evolution was performed, generating libraries of aaRS variants followed by screening for fluorescence, which occurs if an aaRS interacts with the modified tRNA to enable translation though a GFP stop codon. This yielded additional orthogonal tRNA-aaRS pairs that can be used to further assist genetic code expansion efforts.

Engineering interactions with DNA

Categories of proteins that naturally interact with DNA include polymerases, nucleases, and transcription factors. In the initial report on PACE, the system was shown to be capable of evolving several protein-DNA interactions, including a recombinase and T7 RNA polymerase [38]. In each example, the key property needed to drive pIII production, and thus phage replication, is binding of an evolving protein to DNA. Since these initial studies, PACE has been used to evolve RNAPs with orthogonal promoter specificity [66, 67] and sequence specific DNA binding proteins called TALENs [68]. PACE has also recently been used in combination with rational design to engineer a DNA E-box motif binder based on the Myc/Max transcription factors [69, 70] (Figure 2C). In this paper, the PACE platform used a 1-hybrid approach to evolve a DNA binder, which proved superior to previous yeast and bacterial 1-hybrid non-continuous evolution campaigns in that increasing the selection pressure over time was used to combat false positives.

CRISPR-Cas proteins have found widespread use in the development of DNA editing technologies, yet they tend to have substantial off-target activity and require a PAM-recognition motif at the target DNA [93–96]. While RNA guide optimization and the crystallization of Cas9 have enabled rational design efforts to improve specificity [97–101], several directed evolution campaigns have also been carried out to improve the utility of these systems. The more specific Sniper-Cas9 variant was generated by creating Cas9 libraries via error-prone PCR and XL1-red competent cells and screening them in E. coli cells for cell survival, which depended on them targeting an exact guide match and not targeting a close mismatch that was encoded in the genomic DNA [102]. Screening for on- and off-target activity among error-prone PCR generated Cas9 libraries was also used to create evoCas9 with improved specificity [103]. In this report, screens were done in yeast where targeting an exact guide match lead to cell survival and targeting a mismatch lead to white colonies, whereas no off-target activity created red colonies. PACE has also been used to evolve Cas9 with increased specificity for a specific PAM and for variants with broader PAM compatibility using a similar 1-hybrid approach as was employed for the DNA box motif directed evolution [71, 72] (Figure 2C).

Though generating technologies that regulate DNA has been a large focus in the field of evolving DNA interactions, progress has also been made in understanding the evolution of natural biological interactions with DNA. For example, multi-replicate PACE evolutions of T7 RNAP to achieve novel promoter specificities led to insights about how path-dependence impacts evolutionary trajectories [66, 67]. In another example, ancestral sequence reconstruction combined with deep mutational scanning revealed how an ancient transcription factor evolved to achieve novel DNA specificities [104]. This study found that many alternative protein sequences conferred the given functions, further highlightI role of contingency through permissive mutations that emerged in history.

Engineering higher-order interactions

Thus far in this review, we have focused on evolution technologies that engineer two-partner interactions. However, evolution is also able to solve more complex problems, such as engineering multi-partner interactions [105]. One recent example from our lab introduced an experimental evolution system to engineer protein-protein interaction glues—molecules that bind to two different target proteins to bring them in proximity to each other [73]. The evolution strategy, termed re-PPI-G, mimics our previous PACE designs: two proteins are fused to the N-terminal and C-terminal halves of the split RNAP biosensor (Figure 2D). These fragments are both expressed in the E. coli host cell, and thus, do not undergo evolution. Rather, another “glue” fragment is encoded in phage, which evolves to bring the two proteins together, promoting RNAP reassembly and subsequent pIII production and phage replication. After optimization, we tested the system by evolving a zipper peptide fused to ULK1 to better interact with both the partner zipper peptide and ULK1’s partner, GABARAP. This technology holds promise as a means to rewire protein-protein interaction networks, just as current small-molecule PROTAC technologies do [6].

Co-evolving biomolecular interactions has been an exciting field in recent years as well. Co-evoultion has primarily been achieved through mutating each partner individually followed by screening to assess changes in activity, though selection platforms are emerging as well [106], and advances in deep mutational scanning technologies have enabled further progress. To date, co-evolution technologies have advanced biochemical knowledge of PPIs and allowed the generation of orthogonal binding partners and signaling pathways [107, 108]. However, though real-time whole organism evolution experiments by nature enable continuous co-evolution [109, 110], directed experimental continuous co-evolution has yet to be achieved.

Additional higher-order interactions include protein packaging, such as occurs when viral capsids encapsulate their RNA or DNA genomes. Directed evolution was used to create a non-viral protein cage based on Aquifex aeolicus lumazine synthase (AaLS) building blocks that could package a tagged HIV protease [111]. To do this, error-prone PCR was used to create a library of AaLS building blocks, which were then screened in E. coli for their ability to encapsulate HIV protease, which is otherwise toxic to the cell. The cage was subsequently evolved using similar directed evolution campaigns to package its own mRNA [112] and more efficiently protect the enclosed RNA from nucleases, resulting in a structure that mimics those found in natural viruses [113]. Computational approaches have also been used to generate de novo non-viral icosahedral capsid scaffolds based on viral structures to address the basic scientific question of what is necessary for capsid formation [114, 115]. Computationally derived scaffolds were then mutagenized by Kunkel mutagenesis and screened in E. coli for the ability to encapsulate and protect their own RNA from challenges such as heat and other environments, where survivors could then be harvested and sequenced to link genotype to successful phenotype. The evolution campaigns found capsids that achieved RNA packaging activity in vivo and in mouse models [115], and an additional deep mutational scanning library provided insight into biochemically important characteristics of nucleocapsids, e.g., hydrophobic cores and positively charged capsid interiors. This series of events spectacularly highlights a trend present in this review: the synergy between computational approaches followed by experimental evolution to generate robust biomolecular interactions. Just as molecular docking has proved a powerful technique in small molecule inhibitor development, so too could computational methods provide potential starting points for evolving new interactions.

Although this review has focused on evolving proteins to interact with components of the central dogma (Figure 3), it is also worth noting a few related advances. Biomolecules can be “decorated” with various modifications that go beyond nucleotides and natural amino acids. Examples where directed evolution has been used to generate proteins that recognize such modifications include proteins that can bind proteins with glycan and sulfotyrosine additions [116, 117] and the reprogramming of RNA reverse transcriptases to interact with specific nucleic acid methylation sites [118]. Moreover, Systematic Evolution of Ligands by EXponential enrichment (SELEX) technologies create DNA and RNA aptamers capable of binding proteins and small molecules and are often used for detection purposes [119]. Just as recent advances have highlighted additional roles that nucleic acids can play other than the traditionally ascribed role of information storage and transfer, progress has also been made in engineering nucleic acid-nucleic acid interactions that go beyond that of simple base-pairing. Efforts to expand the genetic code have required experimental evolutions of tRNA-ribosome interactions to enable quadruplet codons [120, 121]. While others have focused on engineering orthogonal tRNA-tRNA synthetase pairs, self amino-acylating tRNA ribozymes, dubbed Flexizymes, were evolved, which allow for genetic code expansion [122], and evolution of the ribosome itself is being done to allow further expansion of what types of synthetic proteins can be made [120, 123, 124]. Furthermore, DNA enzymes, which are not known to exist in nature, have been evolved for a variety of purposes over the past 20–30 years, as reviewed elsewhere [125].

Figure 3.

Directed evolution has been employed to evolve proteins that interact with DNA, RNA, and other proteins to facilitate biomolecular interactions in the above areas.

Concluding Remarks

As our understanding of the diverse interactomes of the biomolecules of life continues to expand, so too has our ability to evolve biomolecular interactions. Experimental evolution can approximate “replaying the tape of life” and thus give insight into the molecular basis of evolving interactions, and it can also generate new or improved interactions for use in studying biology and creating novel biotechnologies. Despite significant successes and advances over the past 30 years, the full potential of evolution has still not been harnessed as a design approach for evolving biomolecular interactions. The development of biosensor technologies has spurred advances in evolving different interaction types and promises to continue to do so (see Outstanding Questions). Novelty is certainly important for progress, yet efficiency and selectivity are also paramount. Easily deployable methodologies, robust continuous evolution systems, advanced automation, and computational approaches for improved library design will ensure that experimental evolution becomes more accessible and successful in the coming years.

Outstanding Questions.

• How does nature compare to laboratory evolution and how can the lessons of natural evolution inform how to better deploy evolutionary principles in the laboratory?

• Can we employ multiple negative and/or positive selections simultaneously to evolve more than one characteristic in a molecule (i.e. allostery)?

• What methods can be used to evolve biomolecular interactions not yet amenable to directed evolution, such as interactions with lipids, in vivo continuous co-evolution, and protein-protein interaction inhibitors?

• Can we evolve biomolecular interactions radically different than those found in nature (e.g. synthetic translational machinery for sequence defined polymers and protein materials, control of phase transition/protein condensates)?

• Current continuous evolution techniques generally require some small level of activity—how can one generate truly novel function without pre-existing function (e.g. with robust de novo libraries) using continuous evolution techniques?

• As a continuous evolution technology, PACE has been expanded to evolve a variety of interactions- can we adapt similar selection strategies for yeast and mammalian continuous technologies?

Highlights.

• Directed evolution can be used to improve the ability of a protein to perform a specific function, including binding to a biomolecule of interest.

• Growth-based selections are a powerful evolutionary approach but require careful construction and validation of the biosensor that links function to fitness.

• Phage-assisted continuous evolution (PACE) enables hundreds of rounds of evolution in days and can be employed with parallel positive and negative selections to evolve specific binding interactions to proteins and nucleic acids.

• Continuous directed evolution can now be performed in eukaryotes including yeast (OrthoRep) and mammalian cells (VEGAS).

• Directed evolution allows for the modulation of many types of cellular activities and can increasingly be used to manipulate even higher-order interactions.

Glossary

Biosensor

a system which converts one sort of biological function into a different output, e.g. protein binding into luminescence

Continuous evolution

an experimental evolution technique where a gene or organism undergoes iterative rounds of diversification and selection as driven by an automated, often biological system, i.e. without researcher intervention

Directed evolution

specifically focusing evolution on a desired outcome; diversification and subjecting a particular gene of interest or organism to a selection pressure that links the desired function to individual fitness

Diversification

creating variety within a population, often achieved through mutating/varying nucleotides in a gene

Evolution

the process of iterative rounds of diversification (e.g. mutagenesis) and selection (e.g. natural selection) leading to changes in the gene sequences of individuals within a population

Experimental evolution

a laboratory technique designed to mimic natural evolution in a laboratory setting (i.e. by allowing or causing mutations to accumulate in a population and applying a selection pressure that determines survival)

Negative selection

subjecting a population to a condition that lowers the fitness of variants with a given property

Phage-Assisted Continuous Evolution (PACE)

a continuous evolution technique that links the fitness of a gene to the survival and propagation of bacteriophage on E. coli

Screen

separating individual variants in a population and assaying each for a given function

Selection

identification of individual variants within a population that have a given property by separating variants with the property from variants without the property

Surface display

a technique in which a biomolecule is tethered to its genetic code, either directly or indirectly, and flowed over an immobilized target such that binders can be separated from non-binders