Expanding the Chemical Diversity of Genetically-Encoded Libraries.

Abstract

The power of ribosomes has increasingly been harnessed for the synthesis and selection of molecular libraries. Technologies such as phage display, yeast display, and mRNA display effectively couple genotype to phenotype for the molecular evolution of high affinity epitopes for many therapeutic targets. Genetic code expansion is central to the success of these technologies, allowing researchers to surpass the intrinsic capabilities of the ribosome and access new, genetically encoded materials for these selections. Here we review techniques for the chemical expansion of genetically-encoded libraries, their abilities and limits, and opportunities for further development. Importantly, we also discuss methods and metrics used to assess the efficiency of modification and library diversity with these new techniques.

Graphical Abstract

Introduction.

A number of new ligand selection technologies are increasingly being used in drug and probe discovery efforts. Technologies such as DNA encoded libraries (DELs),¹ mRNA display,² and phage display³ significantly reduce the instrumentation costs and time relative to traditional high-throughput screening (HTS),^4–6 while also increasing the frequency of success against some of the most challenging therapeutic targets.^7–13 These techniques, which rely on genetic barcodes to deconvolute hits from complex mixtures, allow large compound libraries (~10⁹-10¹³ compounds versus ~10⁴-10⁶ compounds for HTS) to be assayed in one pot and have been successfully used to identify high affinity ligands for numerous disease-implicated protein targets.^7,14–23 Indeed, many of these methods are quickly finding broad application in drug discovery efforts in both academia and industry.^24–28

Genetically encoded libraries (GELs), which include mRNA display, phage display, yeast and cell surface display libraries, as well as intracellular SICLOPPS^29,30 and plasmid-encoded libraries, are particularly powerful. GELs are unified as selection methods in that they depend on ribosomes for library synthesis and are inherently highly peptidic as a result. Through the power of ribosomes, GELs can be synthesized rapidly, in high fidelity, and with relatively minimal effort, and present some of the largest libraries (up to 10¹³ molecules) prepared to date. In contrast to DELs and small molecule libraries, GELs can be synthesized directly from their genetic barcodes, allowing for multiple, iterative rounds of GEL screening to be easily performed for the enrichment of high affinity ligands. Additionally, GEL compounds are typically larger, which can aid in allowing them to better disrupt protein-protein interactions due to their increased surface area.^{13,18,22,31–35} Although there are still limitations, most notably in cell permeability and chemical space, the unique benefits of GELs have not gone unnoticed and many major pharmaceutical companies have partnered with or imported GEL-based ligand selection programs for drug discovery campaigns.^36–42

The power of GELs in ligand discovery has inspired numerous efforts to expand GELs beyond their innately peptidic chemical space and enable GEL-based screening with non-natural or small molecule-like functionalities. Such chemical functionality that goes beyond the natural proteinogenic amino acids could further enhance the rates of success in affinity selections while also improving drug-like properties, permeability, and stability of hit compounds.^43–47 To this end, new diversity elements or building blocks have been incorporated into GELs either co-translationally, by hijacking steps in the ribosomal assembly process, or post-translationally, by modifying the ribosomal products of translation. Cell-free translation systems, orthogonal tRNA synthetases, and engineered ribozymes have all been recruited or developed to expand the chemical diversity that can be filed through the ribosome via genetic code reprogramming or expansion.^48,49 Meanwhile, chemical and enzymatic transformations can introduce modifications, such as multiple macrocyclizations^50–52 or backbone modifications^53,54 that otherwise prove challenging to ribosomal incorporation. Together, an array of more than 300 non-canonical groups can now be incorporated into GELs of various kinds and the libraries that can be accessed look less like the simple peptides of early days and more akin to the natural products deployed by many microbes. Such methods are providing a promising new array of GEL libraries for next-generation targets.

In this review, we attempt to comprehensively and concisely summarize advances in expanding GEL diversity, through both translational and post-translational means. We strive to provide an organized presentation of methods used to expand GEL chemical diversity, key barriers to their implementation, and a catalog of the chemical diversity that can be accessed; we eschew exhaustive descriptions of the selections or binders that result from using these diversified GELs. Importantly, with respect to post-translational diversification, this is not a review of biologically compatible chemistry that might be used to modify GELs, but only of chemistries that have been site-specifically validated on GEL members, as several hurdles must be passed before a given reaction can be reliably employed with GELs. These hurdles include: installation of a diversity element on 1) a translated peptide, 2) a displayed molecule, 3) a displayed library, and 4) a ligand identified from a selection. The hurdles to diversifying GELs depend on the selection system in use, and diversification methods are presented with this consideration in mind. Additionally, the proliferation of methods to expand GEL chemical diversity has required concomitant advances in techniques to assess library quality post-modification. Consequently, we also discuss important quality control metrics and provide a summary of the assays that have been used to measure them. Thus, this paper is presented in three distinct sections: section 1 discusses quality control metrics to consider when diversifying GELs and accompanying assays to quantify them, section 2 summarizes methods to expand GEL diversity during translation and their documented substrate scopes, and section 3 summarizes approaches to post-translationally expand GEL diversity. We hope the focused scope of this review will prove a practical guide for those who are looking to incorporate new diversity elements into GELs for selection purposes.

Quality Control and Diversity Metrics.

Quality control is arguably the most important part of a GEL, as poorly compatible diversification strategies may result in library degradation, false positive enrichment, and/or failed selections. For a diversity element to be successfully enriched in selections, it must be sufficiently and specifically incorporated and not interfere with the genetic material by which library members are propagated and identified. Thus, it is important to consider the established metrics through which incorporation can be measured and how they might be assessed to assure high quality libraries for use in benchtop selection campaigns (Figure 1). This section first defines these quality control metrics, then subsequently recounts assays that have been used to quantify them.

Figure 1.

Quality control metrics and accompanying assays to assess GEL diversification strategies. A) Schematic of a chemically diversified GEL selection with applicable quality control metrics (yellow numbers) at each step. B) Common methods to assess GEL quality control metrics and their direct uses.

Library Metrics

Library Size & Chemical Space.

Two metrics used to define all libraries, chemical or genetic, are library size and chemical space (Figure 1A). Both parameters are largely theoretical but are essential in evaluating the utility of a given library. Library size refers to the theoretical number of unique members that could exist in a library. For a GEL, this number is calculated by exponentially multiplying the number of randomized codons by the number of amino acids allowed per codon. Library size is an essential consideration during selection planning because a larger library often correlates with a higher degree of success. Additionally, it may prove beneficial to ensure that libraries present more molecules than their theoretical library size in a naïve selection so that every potential sequence iteration is accounted for. Chemical space should also be considered when designing a library. For small molecule libraries like those used in high throughput screening (HTS), physicochemical properties such as molecular weight, AlogP, number of hydrogen bond donors/acceptors, and polar surface area, are often used to estimate chemical space.^55,56 Since GELs are inherently peptidic, this parameter is more commonly estimated by assessing the distribution of sequence counts throughout a library to establish that no particular sequences are overrepresented and that amino acids are uniformly accounted for. However, unnatural amino acids (UAAs) and other diversity elements can dramatically impact chemical space and even add new dimensions, such as ring size and topology in the case of cyclic peptide libraries. Ultimately, the chemical space of modified GELs is an important indicator of their broader utility.

Metrics for Translational Diversification.

To incorporate new functionalities with translational machinery, UAAs and other building blocks must first be loaded onto tRNAs, and then accepted by the ribosome. Thus, tRNA acylation efficiency and translation efficiency are important metrics used to assess new methods for introducing diversity elements via ribosomal translation (Figure 1A). tRNA acylation efficiency refers to the percent of tRNA that is properly acylated with the desired UAA, while translation efficiency is based on the ability of the translation system, in vitro or in vivo, to incorporate a given UAA into the sequence of a displayed peptide or protein at the desired codon. Several different assays have been developed to measure both metrics, each of which are discussed in further detail below.

Metrics for Post-Translational Diversification.

Chemical and enzymatic transformations can also be performed on translated libraries to introduce diversity. Some of the metrics used to assess post-translational modifications are similar to those employed in solution phase organic synthesis, such as percent yield and regioselectivity (Figure 1A). As might be expected, yield is defined as the percentage of chemical handles in a library that are successfully altered by a post-translational modification (PTM). Percent yield is both the most important and most difficult metric to assess; the number of methods used to measure percent yield is almost as varied as the number of chemistries used to modify GELs. There is no established acceptable value for this metric and yields vary greatly depending on the PTM. Generally speaking, yields within an order of magnitude of the total library size (i.e. ≥10%) are commonly seen in the literature,^53,57–59 but yields as high as 90% modification have been reported.⁶⁰ Regioselectivity refers to selectivity of a modification for its intended site, but more practically can be thought of as an extent of off-target modifications or cross-reactivity with other components of the display system, such as surface proteins or other library members.

Another critical metric for post-translationally modified GELs is genetic viability (Figure 1A), or the remaining integrity of genetically encoded material after a PTM is performed. For in vitro display systems this predominantly relates to the ability of DNA to be amplified, sequenced, and/or propagated in a selection round. For in vivo display systems, such as phage display, one must additionally consider retention of infectivity or cellular growth after library modification. Acceptable amounts of genetic loss due to PTM vary depending on the library size. Most reported modifications attempt to create mild reaction conditions that keep library viability as high as possible, but PTMs have certainly been reported that reduce library viability by an order of magnitude (~90% decrease). For example, the well-known phage bicyclization strategy developed by the Heinis group significantly decreases phage viability upon addition of micromolar concentrations of cyclizing reagent, presumably due to the cross-linking of lysine residues between phage particles, leaving them noninfective.^50,61 Despite this loss however, multiple selection campaigns have successfully been performed with this PTM.^14,16,17,19 Indeed, since GELs can be as large as 10¹³ molecules, substantial losses in genetic viability may still be tolerable as long as the remaining library members still encompass theoretical library diversity.

General Definitions & System-Specific Considerations.

Several other metrics are important for both translationally and post-translationally diversified GELs. Sequence bias (Figure 1A), or the change in sequence distribution over rounds of selection, is an essential metric for both strategies as a diversity element may cause artificial enrichment if its incorporation is favored in proximity to certain amino acids. Diversity elements that are incorporated at the amber codon during translation may be subject to additional sequence bias, as sequences containing nonsense codons can be outgrown by those with only sense codons. A degree of sequence bias may also be inherent to certain display systems, such as in phage display, where sequences that aid in infectivity could be enriched for, or in mRNA display, where PCR bias may occur, during amplification between selection rounds. Additionally, display valency (Figure 1A), which refers to the number of peptide or protein “phenotypes” expressed per “genotype”, can be variable in phage and cellular display systems and is relevant when calculating percent yield. Importantly, multivalent display can facilitate cross-reactivity between displayed peptides due to proximity and impact library quality. For phage display one must also consider where to express the library on the phage particle (i.e. which coat protein), which will dictate display valency, as well as several other parameters.

Quality Control

Library Size & Chemical Diversity.

Besides calculating the maximum number of unique library members to estimate theoretical library size, measurements are also usually taken to quantify the number of displayed molecules in a library during a given selection round (Figure 1B). For in vitro selection techniques like mRNA display, the number of peptide-mRNA fusions produced in translation can be quantified through amplification in qPCR and comparison to a standard curve. For phage display, plaque forming assays are generally performed to calculate viral titer, a quantitative measurement of the biological activity of phage virions. Plaque forming units (PFU) are the most common measurement of phage concentration, but quantification of phage particles in solution is also made possible via absorption spectrophotometry given length of viral DNA. Transformation efficiency of plasmid libraries in cellular display systems can be routinely calculated via colony forming assays. These assays only quantify the portion of a library with viable genetic material and do not give a direct indication of whether a library member is successfully displayed; secondary assays are often performed to ensure proper library expression.

Because GELs are innately peptidic and synthesized de novo by ribosomes, they are not frequently assayed for their chemical diversity. The chemical diversity of macrocyclic DNA encoded libraries (DELs), however, has been characterized according to basic physicochemical properties, including those associated with Lipinski’s rule of 5 and by Veber descriptors,^62,63 as well as lowest energy 3D conformation.⁶⁴ These properties have then been compared to published guidelines for oral macrocyclic drugs (i.e. Kihlberg parameters),⁶⁵ or to predict solvation energy and the likelihood of cell permeability.⁶⁴ This characterization however is less frequently done with GELs; instead, the initial chemical diversity of a library is qualitatively evaluated after its construction and/or transformation via next generation or deep sequencing (Figure 1B). The library is typically analyzed for number and distribution of sequence counts, as the overall number of unique counts verifies library size while sequence distribution serves as an indicator for good chemical diversity. Inspection of a library for sufficient sequence coverage, minimal sequence bias, and minimal truncates is a critical QC checkpoint before subsequent library expression and selections.

tRNA Aminoacylation & Translation Efficiency.

There are several different methods for measuring tRNA aminoacylation that are required for introducing diversity elements via translational diversification (Figure 1B). (1) The standard ATP/PPi exchange assay has been used extensively to confirm UAA activation by aaRSs and adenylation domains^66–70, however it cannot inform on tRNA acylation. (2) In the [α³²P]-AMP radioisotope assay, the tRNA is labeled with [α³²P]-AMP at the 3’ end, then acylated, derivatized, and cleaved; acylation is measured by TLC separation and autoradiography.⁷¹ (3) As long as the UAA contains an α-amino group, reaction of the free amino group with a biotinylating agent, such as sulfosuccinimidyl-D-biotin, allows for separation of acylated from non-acylated tRNA via a gel-shift assay or by MALDI-TOF.⁴⁸ In the case of MALDI-TOF, the UAA is derivatized and cleaved from the tRNA by nuclease P1 before analysis.^72,73 (4) Lastly, Suga has developed a microhelix assay, which is currently one of the most robust and versatile methods to assess tRNA aminoacylation via ribozymes. This assay makes use of an RNA microhelix that mimics the acceptor arm of the tRNA. Because the microhelix is small relative to full length tRNA, it can be directly run on an acid PAGE gel without need for prior derivatization to determine the degree of acylation. The microhelix assay has been used extensively to optimize conditions for acylation by Flexizyme (Fx) and is currently the only reported assay used to measure percent acylation of a UAA lacking an α-amino group.^48,74 In general, an acylation efficiency above 10% is accepted as sufficient for further testing in translation efficiency.⁴⁸

Translation efficiency of a UAA is measured slightly differently depending on whether translation is done in vitro or in vivo. In vitro translation efficiency has commonly been measured by equipping a C-terminal tag, such as FLAG or 6x-His, with a radioisotope-labeled amino acid (e.g. ¹⁴C-Asp, ³⁵S-Met, or ³H-His) to a model peptide (Figure 1B). This peptide is then separately translated with and without the UAA-charged tRNA; translated peptides are then affinity purified and run on a tricine SDS-PAGE gel.⁷⁵ UAA translation efficiency is calculated relative to translation of the peptide with canonical amino acids based on autoradiography. The identity of the UAA-containing peptide should also be confirmed via MALDI-MS (Figure 1B) to ensure little or no background incorporation of canonical amino acids.^48,74,76,77 In vivo translation efficiency on the other hand is measured in several different ways depending on the display system. PFU of phage display libraries are often analyzed in the presence or absence of a desired UAA, and fold enrichment in plaque formation is calculated to estimate UAA incorporation.^78,79 Alternatively, in phage, a successfully incorporated UAA can be recovered and quantified via capture assays as discussed below.^80,81 Other in vivo libraries that are not exposed on the cell surface, such as in SICLOPPS and plasmid-encoded libraries, rely on cell lysis and library capture for analysis via MALDI or SDS-PAGE.^15,35 In most cases, efficiency is quantitatively assessed by comparing the amount of peptide(s) from canonical translation to that from translation with the UAA.^15,35,82,83 Translation efficiencies as low as 12%⁸⁴ and as high as 164%⁷⁵ have been reported in mRNA display. The Suga group uses 10% efficiency⁸⁴ or 0.1 uM translated peptide⁸⁵ as sufficient for use in selection while the Szostak group uses 25% efficiency⁷⁶ as sufficient, however no cutoffs for in vivo GELs have been reported.

Percent Yield of PTMs.

Percent yield is likely the most important metric for assessing the utility of PTM strategies for GELs. The methods used to assess percent yield of a PTM are fundamentally similar to those used in assessing translation efficiency of a UAA in that they often involve preliminary isolation of molecules containing the diversity element followed by their subsequent quantitative detection. Yet, the size and complexity of many GELs makes measuring percent yield non-trivial. As such, there are several different tactics that have been used to quantify the yield of a given modification, including: 1) capture assays, 2) subtractive capture assays, 3) MS-based methods, and 4) positive control selections. Each strategy has its benefits and drawbacks. It is important to note that many times, due to the large size of GELs, these strategies are employed in model systems only and results are extrapolated to future libraries, although there are select examples of PTM validation on a library during the course of a selection campaign.

Capture assays involve the direct, quantitative reaction of a newly introduced functional group with an orthogonal detection reagent (capture reagent). The capture then allows for affinity or electrophoretic separation from unreacted materials and quantitation by fluorescence, radioactivity, ELISA, qPCR, or other methods depending on the type of GEL being explored (Figure 1B). A representative example of the capture assay is the biotinylation of a library followed by ELISA with a streptavidin-HRP antibody,⁸⁶ which can then be compared to a standard curve for quantitation relative to wild-type, unreacted phage. Biotinylation can also be followed with SDS-PAGE and western blotting,⁸⁷ or streptavidin pulldown and measurement of either the bound^53,58,59,88 or free^86,89 (supernatant) fraction to calculate yield from an unreacted control. When necessary, this kind of approach can be extended by additional reactions, such as in the case of a phage library where chemically installed boronic acid handles were validated via reaction with a fluorophore-labeled semicarbazide and SDS-PAGE analysis.⁹⁰ Alternatively, if a diversity element is not successfully installed on all library members, a subtractive capture can be used to sequester the unmodified portion of the library. In this case, the unreacted remainder of the library is measured to indirectly calculate the percent yield of a given reaction. Both biotin and FLAG tags have been used as chaser reagents to measure yield in indirect pulldown assays.^{53,59,88,90,91} Importantly, for libraries presenting more than one reactive handle, as is the case with poly-cysteine bicyclization strategies⁵⁰ or orthogonal click-based intracyclization,⁵² two distinct PTM chasers, each with one component of the reaction (i.e. azide/alkyne, thiol/alkyl halide) have been strategically used to measure the unreacted, mono-reacted, or bis-addition products in comparison to cyclized counterparts.^91,92

A number of variations on the capture assay have also employed detection reagents such that the PTM under investigation impacts reagent recovery. For example, Roberts and coworkers placed a ³⁵S methionine downstream of a base-labile ester linkage in an mRNA display library while developing a method for post-translational library cyclization; if cyclization chemistry was unsuccessful, the radiolabel would be lost on workup.⁹³ Alternatively, proteases have been used in tandem with reporter assays to assess the PTM of various GELs.^36,54,94 For instance, van der Donk et al. used a 6x-His tag to monitor lanthipeptide formation in a phage display library; properly formed lanthipeptides could be selectively cleaved with NisP, leading to a decrease in His tag recovery as determined by densitometric analysis.⁹⁴ Similarly, GluC has been used to probe for PTM of mRNA display libraries with the enzyme PaaA by cleaving unmodified peptides from a biotin tag.⁵⁴ Given that GELs are mostly peptidic, proteases can be very powerful in quantifying percent yield of GEL PTMs.

Mass spectrometry (MS) has been used quite extensively to assess yields of PTM chemistries (Figure 1B). Because MS is not strictly quantitative, yields determined in this manner are typically relative yields and require some degree of benchmarking. Additionally, the small quantity of each peptide member in most display libraries means that a separate scale-up step is required for MS characterization of single display genes and results are then extrapolated to the library as a whole. Thus, MS-based strategies are used on fully displayed peptides after separation from their genetic encoding element or else with model peptide precursors that are not yet displayed. Several strategies have been used to cleave or separate peptide “phenotypes” from their genotypes, including equipment of model display peptides with TEV cleavage sites in mRNA display,⁹⁵ reduction of disulfide linkers on the cell surface in yeast display,⁹⁴ and cell lysis followed by affinity purification for other in vivo display methods such as plasmid-encoded libraries and split-intein ligation.^18,96 Notably, for phage display, direct MS of the pIII coat protein on whole phage cannot be done because its copy number is too low (~5 copies/phage) relative to the abundance of other coat proteins like pVIII (~10³ copies/phage). Thus, PTM of purified coat proteins and subsequent MS has been used as a surrogate for chemistry on whole phage particles.^50,60 For chemistry where no change in mass occurs, it is possible to perform a subtractive capture assay with a capture reagent discernable by MS.⁵² In several instances, simple MS validation on translated but non-displayed peptides have also been reported as an initial PTM validation step.^51,97 Although individual peptides can be easily validated by MS, the percent yield of PTMs on small libraries of translated peptides has also been estimated by fitting MALDI-MS data curves to a Gaussian distribution and calculating the curve area for percent conversion.⁹⁸

Finally, in many cases, a positive control selection has been used as a surrogate for yield or else paired with other methods to provide definitive proof of modification. In such selections, the experiment is designed such that enrichment or sequence convergence is only possible if the PTM is successfully performed. For example, in the enzymatic phosphorylation of a phage display library with various kinases, enriched sequences converged on the recognition motifs of each kinase after multiple “selection” rounds.^99,100 Alternatively, chemically equipping a library with a specific binding epitope should selectively enrich for chemically modified sequences when selecting against the complementary protein. This has been done with a biotin PTM and streptavidin selection, mannose PTM and concanavalin A selection, sulfonamide PTM and carbonic anhydrase selection,¹⁰¹ and EGFR peptide PTM and EGFR selection.⁸⁶ If the PTM displays unique biological activity, as is the case for lanthipeptides, testing for antibiotic activity after display can act as another positive control.⁹⁴

Overall, there are many approaches to calculating percent yield for a PTM, and while direct protocols can be case-specific, general tactics can be observed. Notably, many of these assays can be conducted in a pulse-chase approach over multiple time points to also gain insight into the reaction rate of a PTM on a given library, a metric which is rarely thoroughly studied, but worth considering.

PTM Regioselectivity.

Ensuring that a PTM has desired regioselectivity is important for a library’s success in selections, as off-target modifications could result in unintended library structures or cross-reactivity between library members, which in turn could affect genetic viability. Heinis’ phage display bicyclization strategy⁵⁰ for example, while proven in its utility, suffered from some side-reactivity with methionine and lysine side chains and N-terminal free amines when using a cysteine-free M13 phage variant.¹⁰² Fortunately, regioselectivity can often be simultaneously assessed while determining PTM percent yield. Employing fluorescent reporters to measure yield via SDS-PAGE and in-gel imaging (Figure 1B) can also indicate site specificity if done alongside controls lacking the reactive handle, as fluorescence should only appear on the chemically amenable fusions.^{52,90,103,104} For in vivo display systems, observed fluorescence in negative control gel bands can potentially be sourced back to the protein(s) undergoing unexpected modification. Fluorescence reporting has also been used in synergy with affinity tag-containing libraries based on the principle that affinity tag and fluorescence detection should only occur at the same site if a PTM is regioselective.^92,103 For libraries with more than one reactive handle, nuclease digestion and SDS-PAGE analysis has been used to identify the number of different products per PTM.¹⁰⁵ Furthermore, peptides that are sequestered to assess PTM percent yield via MS, as described above, have sometimes been further analyzed by tandem MS to deduce regioselectivity.^18,94 It is also possible to overexpress peptides on a display molecule to achieve appropriate concentrations for LC-MS/MS analysis, as was done for PTM of a special high copy phage vector that expressed increased peptides per phage particle.¹⁰³ Despite these isolated examples, it can be challenging to discern regioselectivity by MS, even on single molecules.

Notably, Derda and coworkers have developed a phage display regioselectivity assay called a “blue/white” screen (Figure 1B). In this assay, chemically amenable phage carry a reporter gene that enables blue plaque formation after PTM, whereas unreactive control phage form clear plaques. A PTM is performed on a mixture of both phage types, and after isolation of modified phage, libraries are spotted on bacterial lawns and the number of blue versus white plaques that form are compared. This assay has been used to assess the regioselectivity of numerous phage display PTMs developed in their lab.^53,59,88,91

Genetic Viability.

It is critical to validate that libraries retain genetic viability after PTM. Because this metric is only concerned with the genotypic portion of a displayed molecule, the assays to measure genetic viability are similar to those used to assess initial library size upon translation, and largely focus on amplification and degradation of genetic material. In regard to amplification, this entails measuring the qPCR cycles,⁹⁵ or colony and PFU (Figure 1B),^{50,53,59,87,88,103,106} generated by displayed molecules in comparison to non-treated or solvent only controls to investigate how amplification differs after subjection to reaction conditions. If the genetic material is directly accessible, it is also possible to visualize the extent of its degradation via SDS or urea PAGE.^52,105 Finally, subjecting a model display gene to multiple rounds of chemistry can reveal whether genetic material remains viable in a mock selection environment.¹⁰⁷

Sequence Bias.

The extent of sequence bias resulting from library PTM can be visualized via the same methods used to assess initial library diversity and chemical space, through next generation and deep sequencing (Figure 1B). While powerful techniques, sequencing results may still be skewed by PCR bias when amplifying low copy number vectors or templates. Some studies exist that investigate the extent of sequence bias inherent to phage display,^108,109 but there are few reports that interrogate this phenomenon in other display systems. Sometimes, researchers will report the number of unique sequences in their naïve library as a percentage of theoretical library size and use this value (usually at least >30%,⁹⁴ up to 99%¹⁸) as an indicator of good sequence distribution. Still, this can only be done with a naïve library as it cannot indicate the extent of sequence bias after any selection rounds. There are even fewer attempts to deliberately evaluate PTM-induced sequence bias, but a recent report by the Derda group revealed that two specific PTMs involving N-terminal oxime ligation and cysteine thiol alkylation on phage display libraries returned sequences that were distinctly lacking in Phe and Ser residues.¹¹⁰ While the exact mechanism behind this bias remains unclear, the authors more readily attribute the bias to processing defects of the library rather than the PTM itself. Importantly, they also observed different residue biases depending on where the naïve library was expressed on pIII, revealing some of the inherent bias that may occur when using phage display versus other display systems. Auspiciously, emulsion amplification¹¹¹ has been used to retain recovery of slow-growing phage clones with sequences that would otherwise be lost to their rapidly growing counterparts during amplification.

Display Valency.

Measuring display valency is of concern for in vivo display techniques that express libraries on virion or cell surfaces. Known valency is required to accurately calculate percent yield of a PTM, but it is also useful for ensuring that viable virions or cells are successfully displaying a given library. The most widely used phage display system, which utilizes M13 filamentous bacteriophage, expresses libraries on the N-terminus of the pIII coat protein for a maximum display valency of five library members for every phage particle.¹¹² However, depending on the size of the library insert, wild-type pIII may be preferentially incorporated, resulting in a majority of wild-type, non-displayed phage particles and only a small percentage of monovalent phage particles expressing the desired library.¹¹³ This inconsistency in display valency can confound characterization of PTMs on phage displayed libraries. To combat expression issues, a phagemid system employing helper phage can be used to achieve consistent monovalent display.^61,94 Successful cell-surface display has been quantified by equipping the displayed peptide with a recognition domain⁹² or affinity tag¹⁸ for subsequent immunostaining and flow cytometry analysis (Figure 1B). Median fluorescent values can be compared with a standard curve to estimate the number of peptides displayed per cell.⁹² The capture assays used to measure percent yield can be also run in tandem with unreactive controls to measure the extent of library expression. Additionally, the Derda group has used binomial statistics to link reaction yields of a PTM with unreactive or non-expressive M13 phage and estimate the number of modified peptide copies displayed on each particle.⁵³

Exploiting the Promiscuity of Ribosomal Translation.

The ribosomal translation machinery is remarkably flexible and has proven to be one of the most powerful tools for expanding the chemical diversity of GELs. The primary hurdle to exploiting this flexibility is the introduction of a UAA or other acyl group of choice onto the chosen tRNA. As detailed below, four primary methods have been employed to alter and control tRNA loading and are discussed in greater detail below: 1) chemical aminoacylation, 2) native tRNA synthetase/tRNA pairs, 3) ribozyme-mediated acylation, and 4) orthogonal tRNA synthetase/tRNA pairs. Of these methods, only orthogonal tRNA synthetases are currently compatible with in vivo GELs, such as SICLOPPS or phage display. The other three methods take advantage of the extreme control afforded by in vitro transcription/translation cocktails and have thus been employed extensively in mRNA and ribosome display. Many of these methods were first used for genetic code reprogramming, where codon boxes are hijacked, such that a UAA(s) is introduced by replacing a natural amino acid(s). Each technology has its advantage when it comes to introducing new UAAs into the available pool of building blocks for a library but each method of genetic code reprogramming, is limited to reprogramming only the 20 available natural amino acids. However, rapid advances, such as splitting codon boxes and four-base codons, have provided additional control and now allow substantial genetic code expansion, such that the 20 natural amino acids can be maintained or even extensively reprogrammed, while one or more unnaturals are simultaneously introduced.¹¹⁴ Advantages and disadvantages of reprogramming and expansion methods are discussed below, as well as a review of the UAAs that have been incorporated into translated peptides.

Chemical and Chemoenzymatic Acylation of tRNAs

Chemoenzymatic preparation of tRNAs has been used extensively to load tRNAs with UAAs for incorporation into proteins, antibodies, peptides, and aptamers, and for the study of the ribosomal translation system. One of the most common procedures involves chemical acylation of the 3’-end dinucleotide of a tRNA of choice with the UAA of choice. The tRNA is then prepared without the 3’-end dinucleotide and then the acyl-dinucleotide can be ligated on using T4 RNA ligase to afford the full length misacylated tRNA (Figure 2.I.A). This approach was used in many early studies on mRNA display. For instance, Roberts employed a chemoenzymatic approach to incorporate biocytin into mRNA display libraries for streptavidin capture assays.^115,116 More recently, the Ito group has used chemoenzymatic acylation of tRNAs for the development of peptide aptamers and probes containing UAAs in ribosome display selections. They have incorporated UAAs such as a photo-responsive ε-(azobenzoyl)-lysine,¹¹⁷ a fluorogenic 7-nitro-2,1,3-benzoxadiazole conjugated aminophenylalanine,^118,119 the small molecule purvalanol conjugated aminophenylalanine,¹²⁰ and an electroreactive 3,4-ethylenedioxythiophene conjugated aminophenylalanine.¹²¹

Figure 2.

Substrate scope of genetically encoded peptide diversification by (I) chemical aminoacylation, chemical N-methylation, and native aaRSs, and (II) orthogonal aaRSs. (I-A) Chemical scheme for the chemical acylation of tRNAs by T4-RNA ligase, and the chemical transformation of acylated amino acids to N-methylated amino acids acylated on tRNAs. (I-B, C, D, E) Examples of substrates that have been incorporated using the indicated native aaRSs, the structures are not all inclusive. Green refers to the total number of UAAs that were aminoacylated by native aaRSs. Purple refers to the number of UAAs efficiently translated into model peptides with a translation efficiency greater than 25%. (II-A, B) Structures shown are the unnatural amino acids that have been incorporated using MjTyrRS and MmPylRS mutants in a GEL setting. Red indicates the number of substrates for various mutants of the orthogonal aaRSs, but that may not have been used in a GEL. ^aServes as initiator, although inefficiently, ^bLoaded by IleRS and ValRS, however only efficiently translated when loaded by IleRS, ^cEfficient in elongation but not initiation

Alternatively, preacylated tRNAs have been chemically modified to introduce new functionality. The pool of natural amino acids can be permethylated by a reductive amination protocol involving an intermediate o-nitrobenzylamine protection step to prevent deleterious demethylation (Figure 2.I.A).¹²² Charged tRNAs prepared in this way have proven competent substrates for the ribosome and libraries of N-methyl peptides can be prepared and selected. Notably, although N-methyl versions of all proteinogenic amino acids prepared in this manner are accepted by the ribosome, N-methylated leucine, threonine, and valine are most efficiently incorporated into translated peptides.¹²³ Similarly, preacylated proteinogenic tRNAs can be converted to their α-hydroxy acids by deamination with nitrous acid¹²⁴ and incorporated into polyesters via the ribosome, although this chemistry has not been exploited with GELs.

Native tRNA Synthetase/tRNA Pairs

Despite the high specificity and tight editing mechanisms of most natural tRNA synthetases, they have proven unexpectedly promiscuous towards UAAs. This promiscuity no doubt is a result of the natural amino acids not having to compete with UAAs under normal physiological conditions. The relatively recent development of translation systems composed almost entirely of purified components^125,126 has allowed elucidation of the rather broad substrate promiscuity of E. coli aaRSs in particular and this promiscuity has then been exploited in a number of in vitro GELs for genetic code reprogramming. Natural amino acids need only be omitted from the translation mix and appropriate UAAs added in their stead. Cross-compatibility between different UAAs should be taken into account when devising UAA cocktails for this approach. An extensive study of native aaRS/UAA compatibility was carried out by Szostak and Hartman by means of a high-throughput MALDI-TOF assay.^72,73 This work demonstrated that nearly 100 UAAs could be efficiently ligated to tRNAs by a native E. coli aaRS and that many of these could also be efficiently translated into test peptides, allowing the synthesis of peptides containing numerous, consecutive UAAs (Figure 2.I.B–E).⁷⁶ Notably, although some of the promiscuity may have seemed predictable based on structural similarity to the cognate amino acid, there were several cases of aaRSs taking analogs of a non-cognate amino acid. Perhaps most strikingly, the MetRS was seen to readily accept the side chain methyl esters of Asp and Glu.⁷² MetRS was also able to accept the only translated α-hydroxy acid, α-hydroxy methionine, while ValRS was able to accept two α, α-disubstituted amino acids, 1-aminocyclopentanoic acid and 1-aminocyclohexanoic acid (Figure 2.I.B).⁷⁶ This unexpected promiscuity underscores the utility of this kind of study in uncovering the innate flexibility of native aaRSs. In sum, these studies found 17 out of 20 aaRSs⁷² that could take at least one analog and up to 13 different amino acid codons could be simultaneously reprogrammed.⁷⁶ Several groups have since gone on to use the insight from these broad promiscuity assays for mRNA display campaigns with expanded chemical diversity. For example, L-canavanine, L-threo-β-hydroxy aspartic acid, and ten other UAAs were used in one such campaign,⁹⁸ while homopropargylglycine was used to replace methionine to allow “click glycosylation” for preparation of libraries of glycopeptide analogs.¹⁰⁵ In a promising variation on this method, Roberts and colleagues expanded the capabilities of native aaRSs further still by introducing an editing deficient mutant of the ValRS. This mutant allowed the incorporation of various UAAs, including α, α-disubstituted amino acids and constrained β-amino acids, into their libraries (Figure 2.I.B, C).^127,128 Thus, much diversity can be accessed by in vitro translation systems without resorting to exogenous acylation strategies.

Flexizyme Technology

To date, the set of evolved ribozymes known as Flexizymes (or Fx for short) have proven the most robust and versatile tool for non-cognate loading of tRNAs and genetic reprogramming in in vitro translation systems. Fxs have been essential in elucidating the exceptionally broad promiscuity of the ribosome. They nucleate the association between a suitably protected carboxylic acid and a tRNA of choice to effect acylation of the tRNA 3’-OH in a manner similar to class II aaRSs.^129,130 Because Fxs recognize a small epitope, three bases on the 3’ end of the acceptor stem of the tRNA, they are particularly versatile and can be employed with a broad array of tRNAs. Substrate recognition typically requires the presence of an aromatic group on the substrate molecule and is thought to involve pi-pi interactions. Three different Fxs have been developed to date, each recognizing a different side chain or carboxylate ester to allow for a broad variety of UAAs to be incorporated into GELs (Figure 3). The broadly promiscuous dinitro-Fx (dFx) recognizes the dinitrobenzyl esters (DBE) (Figure 3A.1) of amino acids lacking an aromatic side chain. For those amino acids that contain an aromatic side chain, enhanced Fx (eFx) accepts the cyanomethyl ester (CME) group as well as the somewhat more activated 4-chlorobenzyl thioester (CBT) (Figure 3A.2).⁷⁴ Finally, amino Fx (aFx) has been developed to recognize solubilizing aminobenzyl thioesters (ABT) (Figure 3A.2) for use with amino acids with poor aqueous solubility.¹³¹ This array of Fxs has facilitated the incorporation of hundreds of different amino acids into peptides and GELs. While not all amino acids discussed below have been used in GEL selections, there is little barrier to their use in such libraries based on the robust in vitro translation assays that they have already surmounted.

Figure 3.

Substrate scope of genetically encoded peptide diversification by Flexizymes. (A) Ester functional groups appended onto the C-terminus of amino acids for recognition by the various Fxs. (B-H) Substrates that have been shown to be accepted by one of the three different Fxs and shown to be efficiently incorporated into a model peptide (>10% translation efficiency). Not all amino acids shown have been used in GELs. Red indicates the number and details of amino acids that have been incorporated by reprogramming the initiator codon and tRNA. Blue indicates the number and details of amino acids that have been incorporated via elongation codons and tRNAs.

Fx is particularly well suited to reprogramming translation initiation. Since the α-amine of the initiator amino acid is not required for peptide bond formation, there is seemingly less penalty for exotic groups at this position and significant opportunity for structural and functional diversity. Accordingly, the initiator tRNA can be used as a testing ground for the steric and functional limitations of a wide array of carboxylic and unnatural amino acids that can be loaded via Fxs. The long list of acids incorporated into the initiator position of GELs via Fxs includes: both L- and D- stereoisomers of all 20 natural amino acids,^132,133 amino acids with large, bulky substituents such as L-carboranylalanine,¹³⁴ N-alkyl-amino acids such as D- and L- N-Me amino acids and β-proline,³⁴ thiol-reactive carboxylic acids such as m-chloromethylbenzoic acid,³⁴ peptides containing UAAs, β-, and γ-amino acids,^135,136 among others (Figure 3B–D, F, G).^137–139 Notably, preacylation of the α-amino group of initiating amino acids, as in dipeptides,¹³⁶ often improves the translation efficiency.

One of the most useful functional groups on initiating amino acids incorporated with Fxs is the N-terminal chloroacetyl group. This functional group is used frequently to set up a cyclization reaction with downstream Cys residues via a spontaneous SN2-type reaction that can occur during translation. Libraries of cyclized peptides have been generated using ClAc^L/DF,^{21,84,140–142} ClAc^L/DW,^{43,132,143–146} ClAc^L/DY,^45,132,147 ClAc-L-carboranylalanine,¹³⁴ and p-(N-chloroacetylacetoamido)benzoyl-^L/DF,²¹ among others (Figure 3B.2). Notably, libraries of randomized peptides can frequently display more than one cysteine for cyclization. In this case, the N-terminal chloroacetyl groups typically undergo cyclization with the nearest Cys, unless it is directly adjacent to the chloroacetyl group, in which case the next Cys downstream will serve as the nucleophile.¹⁴⁸ Because of the robustness and reliability of this chemistry, the chloroalkane cyclization has been used in several other contexts, including side chain-to-side chain macrocycles and bicyclic and tricyclic macrocycles to increase structural diversity.^{135,149–151} One other cyclization technique that is accessible via initiator reprogramming is intramolecular oxazole formation between a 5-hydroxytryptophan and an N-terminal benzylamine-phenylalanine under oxidizing conditions (Figure 3B.2).¹⁵² Given the versatility of the initiator position, other N-terminal modifications can still be imagined.

Fx has also proven competent for introducing a variety of other UAAs at the level of translation elongation. As with initiation, elongation can accept D-amino acids although some modifications to standard translation protocols are required to incorporate multiple consecutive D-amino acids (Figure 3C.2). Under standard conditions, 12 of the 19 D-stereoisomers of natural amino acids can be incorporated into multiple peptide sequences at least once at or above 10% translation efficiency.¹⁵³ Consecutive incorporations of D-amino acids can be achieved through the addition of EF-P, an elongation factor, which is normally employed by eubacteria to facilitate translation of poly-proline tracts.¹⁵⁴ Additional changes to the T-stem and D-arm of tRNAs are necessary to facilitate EF-P interactions. As an example of the limits of this approach, D-serine can be incorporated simultaneously at up to ten consecutive residues.¹⁵⁵

Fx can charge tRNAs with N-methyl amino acids as well. Although all N-methyl analogs of proteinogenic amino acids can be loaded by Fx, only aromatic, non-charged, non-branched, and sterically small amino acids, such as Tyr, Phe, His, Leu, Ser, Gly, and Ala achieve greater than 10% translation efficiency. This subset is well tolerated and up to 10 can be incorporated sequentially into model peptides to give short, hyper-methylated peptides (Figure 3F.1).⁸⁴ In a dramatic extension of this proficiency, Fx has been exploited to incorporate N-alkyl glycine peptoids into mRNA display libraries. Branched alkyl, non-branched alkyl, and functionalized N-alkyl,⁸⁵ as well as cyclic N-alkylated residues could all be incorporated efficiently (Figure 3F.2 and 3F.3). Of 21 cyclic N-alkyl residues tested, all could be incorporated a single time, and 15 could be incorporated consecutively, up to four residues in a row. As in the case of D-amino acid incorporation, EF-P could be added to enhance sequential incorporation of the N-alkyl residues.¹⁴² Charged groups, such as amines and carboxylic acids prove challenging substituents for these peptoids, as has also been seen with the more traditional N-methyl amino acids. Side chain protecting groups can be employed to circumvent the poor tolerance of charged functional groups. Fx incorporation of azides and methyl or benzyl esters allow the unmasking of the respective amines and carboxylates (Figure 3F.4),⁷⁵ thus effectively completing the set of amino acid peptoid analogs that can be accessed by Fx.

Remarkably, translation elongation also allows changes to the spacing between the carboxylate electrophile and amino nucleophile and even direct replacement of the native amine with alternative heteroatom nucleophiles. Suga has shown incorporation of both β- and γ-amino acids via Fx-mediated tRNA loading. 13 of 16 β-amino acids tested could be incorporated a single time with efficient translation, including amino acids with L- and D- stereochemistry, and even one charged amino acid, although with low translation efficiency compared to other amino acids.¹⁵⁶ Recently, Suga and colleagues showed the incorporation of the different stereoisomers of the cyclic β-amino acids, 2-aminocyclohexane-carboxylic acid and 2-aminocyclopentane-carboxylic acid.¹⁵⁷ EF-P, again, enhanced consecutive incorporations (Figure 3H.1).^157,158 γ-Amino acids are most efficiently incorporated at the N-termini of dipeptides during translation initiation.¹³⁶ However, select cyclic γ-amino acids can be incorporated during elongation by using the EF-P helper strategy (Figure 3H.2).¹⁴¹ Additionally, α-hydroxy acids and α-thioacids have been incorporated during elongation (Figure 3E.1 and 3E.2). Sequential incorporation of multiple α-hydroxy acids has allowed the translation of polyesters,¹⁵⁹ while the site selective incorporation of thioacids enabled novel native chemical ligation strategies¹³⁵ discussed in detail below.

Orthogonal and Engineered aaRS/tRNA Pairs

The use of orthogonal aaRS/tRNA pairs has become perhaps the most common way to incorporate UAAs into proteins in vivo. Liu and Schultz have defined an orthogonal aaRS/tRNA pair as an aaRS that is specific for only its cognate amino acid and no other natural amino acids, and that will only recognize its tRNA partner.⁴⁹ In most cases, these orthogonal pairs have been adapted from widely different organisms in order to predispose orthogonality, however mutagenesis and selection is often still necessary to ensure orthogonality to the host organism. The thermophilic methanogenic archaea, Methanococcus jannashii and Methanococcus mazei have provided two of the most commonly exploited orthogonal aaRS/tRNA pairs in the M. jannaschii TyrRS (MjTyrRS) and M. mazei PylRS (MmzPylRS),⁴⁹ although more orthogonal pairs are being discovered as informatics and technology improve.¹⁶⁰ Both can be made to accept an amber anticodon on their respective tRNAs^161,162 and thus can incorporate UAAs at a dispensable codon for genetic code expansion. The MjTyrRS is a class I aaRS that has been used to incorporate more than 38 different primarily aromatic UAAs at multiple different codons into proteins (Figure 2.II.A).⁴⁹ The plasticity of the substrate binding pocket and active site allows for mutations that expand the space available for binding large aromatic substrates.^49,163,164 The MmzPylRS is a class IIc aaRS, which exhibits broad substrate specificity that can readily be expanded by mutations to its C-terminal catalytic domain (Figure 2.II.B). For example, the PylRS Y384W mutant has an expanded catalytic domain that allows for the incorporation of multiple lysine derivatives.¹⁶⁵ Other PylRS mutants, that accommodate various sizes of non-canonical amino acids have been identified, and more than 100 non-natural amino acids and their analogs have been incorporated into proteins in vivo and in vitro.¹⁶² UAA substrates tend to be aromatic or lysine derivatives for MjTyrRS mutants and MmzPylRS mutants, respectively, as they best mimic the natural substrates of the two aaRSs. However, substrates as complex as tetrazine-phenylalanines, hydroxyquinolines, and coumarin UAAs via MjTyrRS mutants,¹⁶³ and a cyclooctyne-lysine derivative, phenylalanine derivatives, and a-hydroxy acids via PylRS mutants¹⁶² have been successfully incorporated into proteins in vivo. Both orthogonal aaRS/tRNA pairs have since been transitioned to use in GELs to add in UAAs at the amber codon position. This allows for genetic code expansion of a broad array of aromatics and lysine derivatives in GELs and is readily available for in vivo GELs because the tRNA and aaRS can be expressed in the organism of the display technology.

Orthogonal aaRS/tRNA pairs have led to diversified in vivo display strategies. SICLOPPS and plasmid-encoded libraries have both been modified to use MjTyrRS mutants for incorporation of various aromatic residues. In a first application, Schultz and coworkers incorporated p-benzoylphenylalanine (pBzF) (Figure 2.II.A) into a SICLOPPS library using a mutant MjTyrRS specific for pBzF.^15,166 Notably, in the subsequent selection against HIV protease, pBzF served as a covalent modifier of an active site lysine.¹⁵ Orthogonal aaRSs have also allowed the direct in vivo cyclization of plasmid-encoded libraries without need for split inteins, which are required for cyclization in SICLOPPS libraries. In the earliest work of this kind, Fasan and colleagues introduced novel nucleophilic UAAs N-terminal to a C-terminal intein to allow cyclization onto the intein-derived intermediate thioester. Early nucleophiles were amine side chains, such as p-amino phenylalanine (pAmF) or O-2-aminoethyl-tyrosine (O2ameY),^167,168 (Figure 2.II.A) but later iterations employed pendant thiols, such those present in 3-amino-4-mercaptomethyl-phenylalanine (AmmF) and 3-(2-mercapto-ethyl)-aminophenylalanine (MeaF) to accelerate the cyclization in an NCL-like fashion.^82,167 This approach obviates the need for the N-terminal intein fragment and allowed novel diversity elements that impact macrocycle conformation, such as the aryl rings of the pAmF and tyrosine derivatives. Still more recent iterations of this approach have incorporated electrophilic UAAs, such as O-(2-bromoethyl)-tyrosine (O2beY),^81,169 (Figure 2.II.A) at the N-terminus of plasmid-encoded libraries that can cyclize with downstream cysteine residues. Several mutant MjTyrRSs have been the workhorses of this approach.

Phage display has also been enhanced by the use of MjTyrRS and PylRS mutants to incorporate UAAs. The Schultz group first published the use of MjTyrRS mutants specific to UAAs in phage display in 2004 to incorporate O-me-tyrosine,¹⁷⁰ p-azidophenylalanine,¹⁷¹ p-acetylphenylalanine,¹⁷² pBzF,¹⁶⁶ and 3-(2-naphthyl)alanine¹⁷³ (Figure 2.II.A) in a library of peptides.⁷⁸ The group expanded on this technology when they used UAA-specific MjTyrRS mutants to incorporate sulfotyrosine,¹⁷⁴ p-acetylphenylalanine, bipyridal-alanine,¹⁷⁵ and p-boronophenylalanine¹⁷⁶ (Figure 2.II.A) in a phage display library of antibody fragments.¹⁷⁷ Use of multivalent hyperphage phagemid display rather than wild-type M13 phage was essential to these efforts. The hyperphage present a wild-type pIII phenotype, but lack a functional pIII gene while still successfully infecting E coli, ensuring that the sole source of pIII is the pIII-library member fusion in the phagemid.¹⁷⁸ These hyperphage enabled targeted selections, alongside the orthogonal, engineered MjTyrRS mutants specific to p-boronophenylalanine and bipyridal-phenylalanine to incorporate the UAAs into phage display libraries.^79,179 More recently, Fasan and colleagues used the O2beY specific MjTyrRS mutant¹⁶⁹ with a similar phagemid strategy to incorporate O2beY into a phage display library for cyclization with a downstream cysteine residue.⁸⁰ Moving away from large aromatic groups, N^ε-butyryl-lysine (Figure 2.II.B) was incorporated via the PylRS Y384W mutant¹⁶⁵ to mimic natural PTMs that are read by epigenetic regulators in the cell.⁸³ The same group also utilized the same PylRS mutant to translate N^ε-acryloyl-lysine (Figure 2.II.B) into a phage display library for spontaneous cyclization with a downstream cysteine residue.⁸¹

Advances in Genetic Code Expansion

Most of the technologies mentioned above allow significantly expanded translational diversity through codon reprogramming, but even greater diversity can be achieved by splitting redundant codons or creating new, expanded codon sets, i.e. genetic code expansion. Fx has been used to effectively split codon boxes and increase the number of amino acids available for each position of a randomized mRNA display library. By using a stripped down in vitro translation system with only artificial tRNAs supplied in trans, Suga and colleagues reduced the codon alphabet to 32 minimally tRNASNN transcripts (S = G or C) that removed wobble-based redundancy. 29 of these tRNAs could be used to encode the 20 proteinogenic amino acids plus a stop codon, while the remaining three could be utilized for UAA incorporation.¹⁴³ In this way, libraries containing 23 unique amino acids at variable positions could be made and used in selections.¹⁴⁷ Alternatively, Sisido and colleagues have expanded the genetic code by employing four base codons.^180,181 In this case, two Arg codons, and a Gly codon were selected for reprogramming based on their rare usage in the E. coli genome and therefore low gene copy and transcription levels. Artificial four-base pair anticodon tRNAs could be chemoenzymatically acylated with UAAs and it was found that these competed effectively with triplet decoding by the native tRNAs in a cell-free translation system. While there is some correlation between translation efficiency and the fourth base, in most cases all four bases can be used in the fourth position,¹⁸² in essence allowing one three-base codon to be replaced by four four-base codons. This approach allowed mRNA display selections with libraries exhibiting an additional three UAAs on top of the natural amino acids.¹⁸³ More recently, this approach has been extended to the incorporation of UAAs into phage using an evolved PylRS. Chin and coworkers showed that the UAA p-propargyloxy-L-phenylalanine could be incorporated at a four-base codon.¹⁸⁴ Notably, to the best of our knowledge, this report is also the first example of the use of an orthogonal ribosome in a GEL. This could greatly enhance the ability to utilize four base codons in future GEL designs.

Limits of Translational Incorporation of Diversity Elements

Cumulatively, between chemoenzymatic strategies, both native and orthogonal aaRSs, and Fxs, there are now a number of tools for pushing non-natural building blocks through the ribosomal translation machinery and into genetically encoded libraries. Each of these diverse strategies presents its own advantages and limitations that recommend it for specific GELs. Perhaps the most salient distinction between these loading methods is between orthogonal aaRSs, which can be used with in vivo, cell-based GELs such as SICLOPPs and phage display, and chemoenzymatic and Fx-based methods, which, thus far, are only compatible with in vitro GELs, such as mRNA display. Orthogonal aaRSs by definition exhibit the selectivity necessary to discern between the native, proteinogenic amino acids and UAAs but also, presumably, benefit from increased cellular stability relative to RNA-based Fxs. aaRSs have benefits for in vitro display libraries as well: they accept readily accessible, unprotected amino acids, act catalytically to reacylate tRNAs as they are released from the ribosome and they can be employed in situ, at lower concentrations than Fx and without arduous prior purification of the acyl-tRNAs. That said, the substrate scope of aaRSs with respect to both amino acids and tRNA partners is severely limited and finding or engineering new variants is still non-trivial. From the standpoint of versatility, Fxs are powerful tools: Fxs now exist for a wide array of amino acids and they can easily be adapted to essentially any tRNA, since they recognize a minimal 3-base pair epitope. Still, the need for super-stoichiometric quantities of acyl tRNAs means that Fxs will inevitably be limited in terms of library diversity, since the translation mixtures will only be able to accommodate so much RNA in solution.^48,74 A hybrid strategy, that marries the flexibility of Fxs and the efficiency of aaRSs, along with advances in genetic code expansion strategies, would doubtless allow a diversity of GELs not yet seen in the literature.

Post-Translational Modifications.

This section reviews site-specific PTMs that have been made on genetically encoded molecules and libraries. Both chemical (cPTMs) and enzymatic (ePTMs) transformations have been harnessed to diversify GELs post translationally. To date, non-enzymatic chemistries have been used for most of the diversity elements post-translationally installed on GELs, although enzymes have gained traction in recent years. Additionally, while there is a large and growing arsenal of biocompatible chemical transformations^185–189, only a modest number have been adapted for modifying GELs. Chemistries that have been used to modify GELs are necessarily robust and orthogonal to the genetic material and exploit a limited number of reactive handles present in canonical or else unnatural amino acids and building blocks. Cysteine thiols are easily the most utilized natural reactive handles for cPTM of GELs and have been employed in numerous applications (Figure 4). Similarly, N-terminal amines also present discernable reactivity and have been selectively modified or converted into handles for further PTM on display libraries (Figure 5A). Other canonical amino acid functional groups, such as lysine side chain amines, the carboxylic acids of aspartic or glutamic acid residues, or the phenol groups of Tyr residues, have also been used as sites for cPTM on display molecules, but provide significantly less selectivity.^190–196 While these PTMs may be beneficial for imaging and drug delivery efforts, they are likely unconducive towards affinity selections, as over modification may confound selection results and/or compromise the integrity, infectivity, or growth of library members; thus, these nonspecific modifications are not reviewed here. Indeed, because site selectivity is such an important consideration in the PTM of GELs, non-natural, bioorthogonal functional groups, such as azides and alkynes, are also popular handles for post-translational diversification and have been used in a number of applications (Figure 5B).

Figure 4.

GEL diversification through reactions with cysteine, including (1) alkylation, (2) cyclization, (3) peptide extension, (4) DHA installation, and (5) DHA conjugate addition. Light grey boxes depict representative reagents that have been used to chemically diversify cysteines on GELs for a given type of reaction. Colored dots indicate the display systems where these GEL modifications have been made, where purple = phage display, green = cell-surface display, and yellow = mRNA display.

Figure 5.

GEL diversification through reactions with (A) N-terminal amines, (B) click reactions, and (C) enzymatic transformations. (A) Reaction of N-terminal amines with: 1. sodium periodate to form aldehydes and subsequent diversification with chemical reagents outlined in light grey boxes, 2. chemical linkers to form macrocycles, or 3. an N-terminal remainder of a split intein (Int^N) to enable synthetic peptide ligation. (B) Copper mediated click reactions between translated azides/alkynes of GELs with their complementary chemical reagent to enable either: 1. side chain modifications or 2. cyclizations. (C) Promiscuous enzymatic reactions that diversify GELs through: 1. Thioether formation, 2. Peptide ligation and 3. indolizidinone formation. Colored dots indicate the display systems where these GEL modifications have been made, where purple = phage display, green = cell-surface display, and yellow = mRNA display.

Chemical Modifications (cPTMs)

Side Chain Modifications

Cysteine alkylation or arylation with alkyl or aryl halides has been used extensively with various GELs, including phage, mRNA display, and cell-surface display libraries. This basic SN2 chemistry has been used to fit libraries with various pharmacophores,^57,104 fluorophores,^103,197 and carbohydrate-based epitopes (Figure 4.1).⁸⁷ In a recent, novel application, a phage library was fitted with a boronic acid functionality in order to covalently target bacterial cell walls.⁹⁰ Importantly, when applied to phage, this chemistry has been employed in conjunction with a mutant M13 phage that lacks cysteines in the pIII coat protein.¹⁹⁸ This disulfide-free pIII variant can improve regioselectivity in the cPTM of cysteine containing phage display libraries. However, it also reduces phage infectivity, thus requiring large culture volumes to generate sufficient infective phage to encompass library diversity.⁶¹ In addition to alkylation, cysteine residues in phage libraries have been coupled with synthetic peptides via conjugate addition to maleimide linkers (Figure 4.3).⁸⁶ Cysteines on mRNA displayed peptides have also been post-translationally eliminated via α,α’-dibromoadipic-bis-amide to furnish dehydroalanines (DHAs) (Figure 4.4), which can subsequently be reacted with carbohydrate thiols to form glyclosylated macrocycles (Figure 4.5).⁹⁵ Overall, canonical side chain cPTM of GELs has almost exclusively been limited to cysteine thiols, as other potentially reactive amino acids bearing amines or activated aryl positions are less likely to provide favorable site selectivity.

Some UAAs have also undergone PTM to expand GEL diversity, with the most common example being a copper-mediated click reaction between a translated azide or alkyne UAA and the complementary reactive handle on the PTM. Notably, there have been some reports of reduced phage infectivity¹⁹⁹ or oxidative damage to DNA²⁰⁰ during Cu-Activated Azido Alkyne Cycloaddition (CuAAC); this reduction in genetic viability however may be attenuated with radical scavengers like DMSO²⁰⁰ or reducing agents like sodium ascorbate,^201,202 and several publications report successful PTM of GELs using CuAAC.^52,92,105 For example, Krauss and coworkers reprogrammed Met with homopropargylglycine and, by capitalizing on the natural promiscuity of the native Met-RS, translated a Met-deficient mRNA display library with reactive alkyne handles. They then performed CuAAC to append branched mannose-azide moieties to the library, creating a library of glycopeptides (Figure 5B.1).¹⁰⁵ Strain-promoted click chemistry has also been performed on azide-containing pVIII proteins of M13 phage, suggesting feasibility of click reactions on phage libraries.²⁰³ Click chemistry has also been used to expand structural diversity of GELs with mono-, bi-, and α-helical cyclizations, all of which will be discussed in further detail. Additionally, selenium containing UAAs have also undergone PTM, specifically oxidation, to afford DHAs. Selenalysine for instance has been incorporated into an mRNA display library using native Lys-RS translation machinery, treated with hydrogen peroxide to furnish a DHA, and spontaneously reacted with downstream cysteines to generate a lanthipeptide library for functional selections (Figure 4.5).²⁰⁴ Other UAA scaffolds have also been accessed through chemical modification; for example, the post-translational deprotection of several Fx-mediated N-alkyl UAAs to reveal their electrically charged derivatives has been performed on model mRNA display genes.⁷⁵

N-Terminal Modifications

The N-termini of display libraries are often chemically distinguishable because they can present uniquely reactive free amines that are significantly less sterically hindered than other library portions, making them readily accessible for cPTM. In GELs that are generated in vivo in bacteria, N-terminal amines are innately available for modification, because the formyl group of the initiator formyl methionine is typically removed by deformylases or aminopeptidases immediately after translation. However, GELs produced by in vitro bacterial translational machinery, such as those in mRNA display, retain this formyl group, so deliberate addition of a deformylase or aminopeptidase is needed to reveal this N-terminal residue for these methods. Use of eukaryotic lysate for library translation is an alternative approach to immediately generate N-terminal amines poised for cPTM. Regardless, N-terminal amines are frequent targets for cPTM.

Chemical modification of N-terminal serines and threonines has been particularly useful for diversifying phage libraries. These residues can be converted to highly reactive and biorthogonal aldehydes by oxidation with sodium periodate. This group can in turn be reacted with a number of different functional handles (Figure 5A.1). In one example, thus formed aldehydes were reacted with alkoxylamines to yield the corresponding oximes.⁵³ A variety of alkoyloxylamine substituents could be tethered to the libraries by this chemistry, including biotin,^53,110 sulfonamides,¹⁰¹ and carbohydrates.^{53,101,205–207} As seen with other oxime ligations, the reaction rate of this chemistry on phage can be significantly enhanced by addition of aniline as a nucleophilic catalyst. Synergizing oxime ligation with a nucleophilic aniline group in a 2-amino benzamidoxime derivative of biotin allowed intramolecular ring formation on a phage library via a modified Pictet-Spengler reaction.⁵⁸ In an alternative approach, serine or threonine-derived N-terminal aldehydes have also been reacted with phosphonium ylides in a Wittig reaction to form internal olefins, which have subsequently been derivatized by conjugate addition with thiol nucleophiles or Diels-Alder reaction with cyclopentadiene.⁵⁹

N-terminal cysteines have also been exploited for site selective reactivity in display libraries. In early work, N-terminal native chemical ligation (NCL) of a UAA-containing synthetic peptide with a small phage library bearing an N-terminal cysteine was used to install proteins on phage.¹⁰⁷ More recently, timed deprotection of mRNA displayed peptides bearing an N-terminal thiazolidine has allowed for an intramolecular NCL to form backbone cyclic macrocycles.¹³⁵ In further thiazolidine chemistry, a phage library with an N-terminal cysteine, selectively revealed through a Factor Xa cleavage site, has been reacted with a 2-formylphenylboronic acid biotin moiety to afford N-terminal thiazolidino boronates.²⁰⁸

Finally, split inteins have been expressed on display surfaces to enable trans splicing with synthetic peptide or protein fragments (Figure 5A.3). Using a monovalent phagemid system, Mootz and coworkers split the Ssp DnaB intein between the pIII coat protein of M13 phage and a maltose-binding protein or a biotinylated synthetic peptide for trans splicing.²⁰⁹ This auto-catalytic ligation required initial expression of a solubility tag and a relatively large intein sequence on pIII, but successful reaction with the complementary synthetic intein results in removal of these unwanted domains, making it more feasible as a library diversification strategy. Additionally, the M86 intein has been used to ligate a small molecule fluorophore and hydroxylamine-bearing synthetic peptide fragments onto cell-surface displayed peptides.⁹⁶ The relatively small 11 residue N-terminal portion of the M86 intein is readily accessible by solid-phase peptide synthesis, but also requires the use of solubility tags for successful splicing.

Macrocyclization Chemistries

Chemical macrocyclization has been used heavily with GELs. This approach generally involves reacting a bivalent chemical linker with a library comprised of members with two or more reactive handles to conformationally constrict or staple the library members. Notably, many of the same chemistries have been used repeatedly to afford different types of scaffolds simply by altering the genetic composition of the library or by synergizing orthogonal cyclization strategies. By far the most common macrocyclization chemistry for GELs is cysteine alkylation. Libraries with fixed cysteines flanking the randomized portions can be reacted with bifunctional alkyl halide linkers to form side chain to side chain macrocycles (Figure 4.2). A wide array of linkers have been used for this kind of chemistry,^{60,91,98,128,210} including some more exotic examples such as dichloroacetone, which can be reacted with hydroxylamines in situ to yield the corresponding oximes,⁸⁸ or 1,5-dichloropentane-2,4-dione, which can be reacted with hydrazines to further increase diversity of the resulting modified libraries via Knorr pyrazole synthesis.²¹¹ An additional cyclization involves the reaction of the N-terminal free amine and side chain amine of a fixed C-terminal lysine with the cross-linking agent disuccinimidyl glutarate (DSG) to form head to side chain cycles (Figure 5A.2).^23,46,93,212 DSG has been used primarily with mRNA display libraries,^23,46 where library translation with rabbit reticulocyte lysate provides immediate access to the free N-terminal amine which would otherwise be formylated in prokaryotic translation systems, but modification of yeast display libraries with DSG has been recently reported.²¹² An alternative N-terminal amine cyclization strategy involves reacting a divalent linker, sporting both a 2-((alkylthio)(aryl)methylene)malononitrile (TAMM) group and alkyl chloride, with a library containing fixed N- and C-terminal cysteines to afford 2-aryl-4,5-dihydrothiazole (ADT) containing macrocycles (Figure 5A.2).²¹³ Several other cyclization approaches have exploited the reactivity of translated UAAs: the UAA p-acetylphenylalanine for example has been translated with an orthogonal synthetase on phage display particles and reacted with an N-terminal hydroxylamine to cause cyclization through oxime bond formation⁹⁶ and click chemistry has been used to connect azide and alkyne-containing UAAs.⁵²

Cyclizations can also induce library α-helicity by bringing specific peptide fragments in close proximity and encouraging α-helix formation. A partially randomized phage display library sporting two fixed cysteines four residues apart was modified using the same cysteine thiol/alkyl halide linker strategy described above to impose an α-helical conformation.²¹⁴ Employing the use of copper-mediated click chemistry, a bacterial cell-surface display library containing two fixed azide UAAs was conformationally stabilized with a propargyl ether linker (Figure 5B.2).⁹² Other types of scaffold stabilization have been applied as well; GEL cyclization with an aza-benzene photo-responsive linker has enabled cyclic peptides that can switch between cis- and trans- isomers upon exposure to light, rendering peptides active or inactive depending on their conformation (Figure 4.2).^60,91,215

Many of the same chemistries have also been adapted to allow bicyclization and tricyclization of display libraries, providing access to highly rigidified and unusual three-dimensional structures that prove potent ligands. For example, the cysteine thiol alkylation chemistry has been used to form bicyclic phage libraries, wherein linear libraries containing three fixed cysteine residues are reacted with trivalent aryl or alkyl halides, such as tris-(bromomethyl)benzene (TBMB) (Figure 4.2).^14,50 Bicyclization of phage libraries with a tetravalent adamantane-based linker has also been reported, with subsequent linker functionalization made possible via the remaining unreacted alkyl halide.⁹⁷ Vast scaffold diversity can be achieved by designing libraries to allow for multiple, randomized macrocyclization patterns. For instance, an NNK10 tetravalent cysteine library, with three fixed cysteine residues and one randomized, was displayed on phage.⁵¹ By reacting two equivalents of a thiol-reactive linker with this new library, simple SN2 chemistry generated bicyclic libraries with 30 possible macrocyclic scaffolds per linker. Reacting the library with six distinct chemical linkers afforded up to 180 structurally unique bicycles for a library with ten randomized residues. This semi-randomized cysteine bicyclization strategy has also been performed with two fixed position and two randomized position cysteines.²¹⁶ Both provide unique examples of exploiting a tried and true cyclization method to access a wealth of scaffold diversity. Notably, a similar tetravalent cysteine library has also been post-translationally oxidized to create structurally diverse bicycles linked through simple disulfide bonds.¹⁰⁶

Combining two cyclization strategies that are orthogonal to each other has been used as an alternative way to form bicyclic peptide libraries. In this vein, Hartman and coworkers reported the use of sequential thiol alkylation and click reactions on an mRNA display library.⁵² In this study, Met and Phe were reprogrammed with L-β-azidohomoalanine and p-ethynyl-L-phenylalanine, respectively, and translated using cognate Met-RS and Phe-RS A294G into libraries containing two fixed cysteine residues. The libraries first underwent monocyclization with α,α’-dibromo-m-xylene, then copper-mediated azide-alkyne bicyclization to afford overlapping, θ-bridged peptide scaffolds.

Enzymatic Modifications

Although there are relatively few reports of enzymatic diversification of GELs, enzymes are attractive tools for mild, biocompatible chemistry that are gaining traction. In early work, the kinases Blk, Lyn, c-Src, and Syk were used to phosphorylate tyrosine residues in a phage display library in order to enrich for and identify peptide sequences similar to the recognition motifs for these kinases.^99,100 Since then, several other reports have also used GELs to enrich for substrate sequences for kinases, proteases, and other enzymes.^89,217–223 Alternatively, in an approach to create more metabolically stable peptides, several libraries have been “pre-selected” against proteases before target panning to enrich for protease resistant and thus serum stable peptides.^46,216 Additionally, Sortase A from Staphylococcus aureas has been used to diversify phage particles with synthetic peptides (Figure 5C.2); phosphorylated²²⁴ and biotinylated²²⁵ peptides, as well as pre-folded GFP,²²⁵ all containing LPXTG recognition motifs, were site-specifically ligated onto N-terminal glycines in the M13 phage pIII coat protein. In the same work, another Sortase variant from Streptococcus pyogenes, was also used to modify N-terminal alanines of high copy pVIII coat proteins with a fluorophore and single domain antibody.²²⁵ Although these examples have limited potential to incorporate new diversity elements into GELs, they provided key validation of enzyme compatibility with GELs and important tools for subsequent efforts.

Promiscuous enzymes from natural product biosynthesis may prove a rich source of new PTMs for diversifying GELs. Enzymes from ribosomally synthesized and post-translationally modified peptide (RiPP) biosynthetic pathways have proven particularly useful in this regard. In recent work, the RiPP lanthipeptide biosynthetic machinery has been used to form relatively large, randomized libraries with cell-surface^94,226 and phage display^36,94 technologies, as well as plasmid-encoded libraries (Figure 5C.1).¹⁸ The cell surface display of a thioether stabilized peptide library was initially achieved by Moll and coworkers on the bacterial cell wall of L. lactis, which supports the dehydratase (NisB), cyclase (NisC), and transporter (NisT) enzymes for the formation of class I lanthipeptide nisin. By translating their library with a cell-wall anchoring motif containing a Sortase recognition sequence, the authors could translocate and retain their library within the cell membrane, then covalently attach it to the peptidoglycan with Sortase to enable cell-surface display.²²⁶ More recently, the van der Donk lab achieved display of lanthipeptide libraries on both yeast and phage surfaces. They fused their yeast display library to the C-terminus of the yeast surface protein agglutinin. To overcome library translocation issues, they targeted a bifunctional lanthipeptide dehydratase and cyclase, LctM, to the endoplasmic reticulum so that thioether PTM could be performed after the peptide library crossed the cell membrane. In phage, the authors constructed a phagemid that fused the C-terminus of their library and a Tat cell-penetrating peptide to the N-terminus of a truncated pIII coat protein of M13 phage. The Tat peptide was introduced to facilitate library export into the periplasm of E. coli after PTM. This phagemid was co-expressed with a modification plasmid containing NisB and NisC, as well as a helper phage that supplied remaining required phage proteins. Post-display addition of the leader peptidase NisP then yielded a phage display library of class I lanthipeptide variants.⁹⁴ A recently discovered promiscuous lanthipeptide synthetase ProcM has also been exploited to generate a large lanthipeptide plasmid-encoded libraries¹⁸ and phage display libraries.³⁶ Our lab recently disclosed the used of the RiPP enzyme PaaA to convert adjacent glutamic acid residues into a bicyclic indolizidinone motif in mRNA display libraries (Figure 5C.3).⁵⁴ These studies revealed that PaaA is broadly promiscuous outside of its glutamic acid binding epitope, suggesting that indolizidinone structures could be readily incorporated into display libraries for affinity selection purposes, and that RiPP enzymes may be broadly applicable to the diversification of GELs.

Conclusion.

GELs are inherently peptidic due to their dependence on ribosomes for synthesis. This typically limits the chemical space they can access to canonical amino acids, which in turn puts limitations on important drug properties of resulting identified ligands, such as cell permeability. Development or adaptation of several new technologies now allow for the incorporation of hundreds of new diversity elements, significantly expanding the chemical space of in vitro and in vivo GELs. Still only a fraction of the chemical potential of GELs has been tapped with these new methods. The continued investigation and optimization of orthogonal aaRSs and ribozymes, as well as the development of new GEL-compatible chemistries will continue to press the capabilities of GELs in the near term. Additionally, the application of peptide modifying enzymes, as mild, biocompatible reagents could still open new frontiers in GEL diversity. However, the need for robust quality control assays and metrics to assess modifications on such a massive molecular scale is still a key bottleneck in this development. Metrics and QC methods, like those discussed in this review will likely need to continue to evolve to meet the growing complexity of GEL libraries.