Enzymatic Macrolactamization of mRNA Display Libraries for Inhibitor Selection

Abstract

mRNA display is a powerful, high-throughput technology for discovering novel, peptide ligands for protein targets. A number of methods have been used to expand the chemical diversity of mRNA display libraries beyond the 20 canonical amino acids, including genetic code reprogramming and biorthogonal chemistries. To date, however, there have been few reports using enzymes as biocompatible reagents for diversifying mRNA display libraries. Here we report the evaluation and implementation of the common industrial enzyme, microbial transglutaminase (mTG), as a versatile biocatalyst for cyclization of mRNA display peptide libraries via lysine-to-glutamine isopeptide bonds. We establish two separate display-based assays to validate compatibility of mTG with mRNA-linked peptide substrates. These assays indicate that mTG has a high degree of substrate tolerance and low single round bias. To demonstrate the potential benefits of mTG-mediated cyclization in ligand discovery, high diversity mTG-modified libraries were employed in two separate affinity selections: 1) one against the calcium and integrin binding protein, CIB1, and 2) a second against the immune checkpoint protein and emerging therapeutic target, B7-H3. Both selections resulted in identification of potent, cyclic, low nanomolar binders and subsequent structure-activity studies demonstrate the importance of the cyclization to the observed activity. Notably, cyclization in the CIB1 binder stabilizes an α-helical conformation, while the B7-H3 inhibitor employs two bridges, one mTG-derived lactam and a second disulfide, to achieve its potency. Together, these results demonstrate potential benefits of enzyme-based biocatalysts in mRNA display ligand selections and establish a framework for employing mTG in mRNA display.

Graphical Abstract

Peptide macrocycles are increasingly being incorporated into high throughput display techniques, such as yeast, phage, and mRNA display, to facilitate inhibitor and drug discovery efforts.^1,2 Macrocyclization can augment peptide affinity by helping to pre-adopt a lower energy binding conformation and typically makes peptides more resistant to proteolysis and more cell permeable.³ Most display techniques are continuously being improved with new chemistries and methodologies to allow the rapid selection of potent, macrocyclic peptides against targets of interest. Cysteines are particularly adept chemical handles for such cyclization strategies. For example, the co-translational introduction of amino acid building blocks bearing pendent electrophiles, either via orthogonal tRNA synthetase or Flexizyme technologies, allows macrocyclization with upstream Cys-thiol nucleophiles in phage display and mRNA display.^1,4 Alternatively, exogenous electrophiles have been added to bridge two or more Cys-thiols post-translationally.^5,6 Although some excellent methods have been developed to take advantage of other common peptide side-chains and functional groups, the challenges of orthogonality and biocompatibility collude to limit the diversity of peptide macrocycles that can be explored by display technologies.

Peptide-modifying enzymes provide potentially attractive orthogonal routes to create new display libraries of peptide macrocycles.⁷ Enzymes are already biocompatible, work in water, and are tuned to act upon specific functional groups or side-chains within a complex peptide selectively. Additionally, there is a vast and growing catalog of enzymatic chemistries that are known to be carried out on peptide substrates, especially those from ribosomally synthesized and post-translationally modified peptide natural products, also known as RiPPs.⁸ A potential challenge lies in identifying enzymes that exhibit good side-chain specificity but also a high degree of sequence promiscuity, meaning that the given biocatalyst will be able to reliably modify and/or cyclize at a given type of amino acid regardless of the randomized sequence context around it and allow for a broad substrate scope. Forerunners in this area, the RiPP enzymes, ProcM and NisBC have been successfully incorporated into yeast, phage, and bacterial display to form thioether linked macrocycles for inhibitor selections.^9–11 To date, however, there have been very limited reports of the use of enzymes in mRNA display, where they might make use of the expanded synthetic capabilities and control of cell free biosynthesis for inhibitor discovery.^12–14 Integration of enzymatic post-translational modifications into mRNA display libraries could present a highly useful advance in high throughput macrocyclic ligand discovery.

Microbial transglutaminase (mTG), a versatile peptide modifying enzyme with intrinsic macrocyclase activity. mTG is formally a cysteine protease that works on the side-chain carboxamides of glutamine residues, cleaving the carboxamide to an enzyme-linked thioester, which is then condensed with available amine nucleophiles, typically the sigma-amino groups of lysine side chains.^15,16 The net reaction effectively allows crosslinking between two peptide-based functional groups in glutamine and lysine (Figure 1A). This activity has made mTG a versatile tool for a number of commercial applications, including the functionalization of biomacromolecules and as a potent crosslinking biocatalyst^17–19. mTG also has cyclase abilities and can carry out the intramolecular reaction between glutamine and lysine residues to yield isopeptide-bridged macrocycles.^20,21 This isopeptide linkage is present in a number of natural products, such as microviridins, and has been used to stabilize α-helixes in medicinal chemistry applications, but is not currently accessible in mRNA display.^22,23 Early studies have shown that mTG is compatible with several display platforms and has good specificity-to-promiscuity profile.^21,24,25 Despite this foundational work, mTG has not yet been used to cyclize mRNA display libraries for in vitro selection against protein targets of interest.

Figure 1.

A) The reaction catalyzed by mTG to form an isopeptidic bond between a glutamine substrate (blue) and a primary amine (red). This reaction can be carried out in an inter or intramolecular fashion. B) Glutamine activation assay. Peptide-RNA fusions containing a fixed glutamine residue can be modified by reacting a primary amine containing ligand with mTG. This will enable functionalization of the fusion and facilitate isolation assays. C) Lysine cyclization assay. mTG can be used to install a macrocycle on fusions containing both lysine and glutamine residues. This will enable affinity selection assays against targets of interest using enzymatically modified libraries.

Herein we report our efforts to integrate the peptide cyclization activity of mTG with mRNA display. We develop two distinct high throughput assays to separately assess the substrate tolerance and single round bias of mTG in 1) glutamine activation and 2) lysine cyclization (Figure 1B–C). We then modify a high diversity library (~5e¹¹ molecules) and carry out an affinity selection against two novel cancer targets: the calcium and integrin binding protein, CIB1, and the immune checkpoint protein, B7-H3. These selections prove fruitful, eliciting the identification of novel and potent, cyclic peptide ligands for both targets; the affinity of both ligands is dependent on mTG cyclization, validating the impact of the enzymatic method. This work represents a first application of enzymatic macrocyclization in mRNA display-based ligand selection and shows that mTG is a potent tool for generating and selecting highly diverse macrocyclic peptide libraries.

Results.

mTG compatibility and substrate bias during glutamine activation.

We began our efforts by validating the compatibility of mTG with mRNA displayed substrates in our hands. To this end, we adopted a capture assay previously deployed in two separate reports by Lee et al. and Sugimura et al.^24,25 This assay uses pentylamine biotin (PAB – Figure 2A) to cleave and capture the mTG-linked intermediate thioester of activated glutamine side chains. In this way, sequences that are able to be activated by mTG can be functionalized with PAB and then isolated on streptavidin (SA) resin. In the previous work, the PAB assay had been used over multiple rounds of selection to hone in on preferred mTG substrates from phage libraries. We anticipated that a similar version of this assay could be paired with quantitative PCR (qPCR) to test the efficiency of mTG modification of mRNA-linked substrates and also with next-generation sequencing (NGS) to provide course grain assessment of substrate promiscuity and substrate bias in a single round of selection.

Figure 2.

A) Schematic detailing the process for assessing substrate bias during glutamine activation. B) Table detailing library used and NGS statistics for the assay. The diversity is calculated as 19ⁿ where n is equal to the number of randomized positions. All 20 canonical amino acids were used in this analysis. C) qPCR results of streptavidin captured material. Orange bars show recoveries after mTG treatment while grey bars show recoveries if the library is not treated with mTG. Values are normalized to input samples for each replicate. This isolated material was sent for NGS analysis D) NGS results for average amino acid positional variation at each randomized position of sequences isolated from the glutamine activation assay. E) Bar graph depicting the NGS results of isolated fusions from ligation assay. Values >1 indicate amino acids enriched while values <1 denote a loss of amino acids at each randomized position, N=3. Blue line designates the position of the fixed glutamine residue. A sample key details the color for each amino acid. F) Schematic detailing the process for assessing substrate bias during lysine cyclization. Note, it is possible for trypsin to cleave within the ring of the peptide but the isopeptidic bond maintains the connection the biotin handle and mRNA tag still enabling isolation and analysis. G) Table detailing libraries used and NGS statistics for the assay. The diversity is calculated as 19n where n is equal to the number of randomized positions. Methionine is left out to enable the translation of a biotin handle at the first amino acid position in each fusion. H) Cyclization/trypsin assay on the control sequence. Fusions were treated with increasing amounts of trypsin (0 to 800nM). I) Timecourse recovery assay with the NNK₄ library over 1h mTG treatment. J) Cyclization/trypsin assay on both NNK₄ and NNK₆ libraries after 1h mTG treatment. The isolated material was sent for NGS analysis. Orange bars show recoveries after trypsin digestion of (+) mTG while grey bars show recoveries if the library is not treated with mTG. All values for H-J are a normalized to a non-digested control. K) NGS results for average amino acid positional variation at each randomized position of sequences isolated from the lysine cyclization assay. L) NNK₆ NGS of (+) mTG isolated material. Values >1 indicate amino acids enriched while values <1 denote a loss of amino acids at each randomized position, N=3. The blue and red lines designate the position of the fixed glutamine and lysine residues. A sample key details the color for each amino acid.

To this end, we prepared a focused display library, in which three residues either side of a constant glutamine were comprehensively randomized (Figure. 2B). This library was treated with mTG and submitted to PAB-capture. qPCR revealed that samples treated with mTG recovered better than control, nontreated samples, suggesting that mTG is effective in modifying mRNA linked substrates (Figure. 2C). Both the initial, naïve library and isolated material from the single round capture were submitted for NGS. We analyzed codon percentage in selected material relative to the naïve library (Figure. 2D, E); values >1 indicate enrichment in that amino acid and therefore, likely favorable interactions, while values <1 suggest less activation by mTG. At least two composition trends are apparent: 1) aromatic residues are enriched on both sides of the constant glutamine and 2) amino acids that influence flexibility, such as Gly and Pro, are underenriched at all positions. The deprecation of proline contrasts with the mTG Gln-activation substrates selected by Lee et al.²⁵ This can be readily attributed to the decreased pressure and therefore greater number of returned sequences in our single round assay, as compared to multi-round selections in the Lee report. Additionally, although proline is underenriched in general, several of the Pro-containing substrates from the Lee study are also present in our single round dataset.

With an eye towards use of mTG for cyclization, we also took a close look for the emergence of a ‘WAL’ or similar epitope in the three N-terminal residues. Sugimura and colleagues identified this motif as a directing group for mTG glutamine activation and it has been used in a number of subsequent applications.²⁴ Mining the ligation dataset for this epitope shows an abundance of less <0.01% or only 3 out of >30,000 unique sequences. We further expanded this analysis to look for any analogous sequence comprised of an aromatic residue followed by two aliphatics in this position. This analysis still showed a low enrichment over the single round at 0.5% of the isolated library. These data suggest that while WAL may be a preferred substrate motif, there is little bias towards it early in a selection and other residues could likely be enriched by a target if the substrate were randomized at these positions. Overall, no single sequence or sequence family appears enriched (<0.03% sequence convergence) in this initial round, suggesting low substrate bias during glutamine activation.

mTG compatibility and substrate bias during lysine cyclization.

We next sought to modify the capture assay to examine potential substrate bias during mTG catalyzed macrocyclizations onto lysine residues (Figure 2F). The macrocyclase activity of mTG is well established and was readily recapitulated in in vitro translation with a model substrate (Figure S1). To test this activity in a display context, we prepared libraries of Gln-Lys-containing peptide substrates, all of which were individually fitted with N-terminal biotin tags by Flexizyme-based codon reprogramming. These substrates could be sequentially treated with mTG to effect cyclization and then trypsin protease to discriminate cyclized and linear substrates. Trypsin should cleave at the unmodified lysine in substrates that are not cyclized; trypsin cleavage would remove the N-terminal biotin tag and prevent streptavidin recovery of the displayed substrates in a final step. As with the PAB-Gln activation assay, efficiency could be assessed by qPCR and bias by NGS. We prepared two libraries for screening in the cyclization assay: one of lower diversity with four randomized positions flanked by Gln and Lys (NNK₄, 1.3e⁵ sequences), and one of medium diversity with six randomized positions flanked by Gln and Lys (NNK₆, 4.7e⁷ sequences). Although our Gln activation data suggested that an N-terminal ‘WAL’ was not necessary, we included this epitope at the N-termini of both cyclization libraries for its potential to anchor the Gln residue(Figure 2G).^20,24 Under optimized conditions of trypsin loading (Figure 2H) and incubation time (Figure 2I), we saw strong recovery (~50%) for mTG treated versus untreated libraries in this assay (Figure 2J), suggesting that mTG has little difficulty accepting the RNA-tagged peptide substrates.

NGS analysis showed similar conserved trends between the two libraries, NNK₄ and NNK₆ with some notable similarities and differences to the Gln-activation data. Compositional analysis for the recovered sequences revealed high overall sequence diversity for both datasets at <0.06% and <0.01% sequence enrichment for the NNK₄ and NNK₆, respectively. This aligns well with the ligation data and further indicates a lack of single round bias for mTG. Generally, amino acid-level analysis shows a similar overall low degree of variance at all positions for both libraries (Figure 2K and S3). Similar to the Gln-activation assay, proline is the most underenriched amino acid, although to a lesser degree here. Aromatic residues are not as highly enriched as in the Gln-activation dataset. Glycine appears particularly well enriched at position X₂. Differences in reaction type (ligation vs cyclization) and the presence of the WAL motif may contribute to these differences. One other notable trend in the cyclization data is the slight enrichment of Gln residues closer to the C-terminal Lys: positions X₁, X₂, and X₃ in the NNK₄ library and X₃, X₄, and X₅ in the NNK₆ library. Conversely, internal Lys residues seem somewhat underenriched (Figure 2L, Figure S2). These effects, though minor, may reflect a move towards smaller macrocycles. Despite these potentially smaller lactam bridges, a global analysis of the selected sequences showed no proclivity towards secondary structure, especially alpha-helicity (see Supplemental Information). Overall, both the activation and cyclization data combine to show that mTG works well on the displayed libraries and suggest that substrate bias may be sufficiently low to be out-competed by the selective pressure of target binding. Thus, we proceeded with inhibitor selections.

Selection of mTG cyclized CIB1 Inhibitors.

With an mRNA display-compatible and seemingly promiscuous macrocyclase in hand we next sought to validate that enzymatically modified mRNA display libraries could be used to identify novel macrocyclic peptide inhibitors. In a first test of this method, we screened enzymatically modified libraries against the calcium and integrin binding protein CIB1. CIB1 is an appealing target because it has a deep, ligandable pocket that nucleates many of its key interactions and we have previously established a robust biochemical assay for disruption of binding to this pocket.^26–28 Thus, libraries of two different sizes, NNK₆ and NNK₉, were pooled, treated with mTG, and screened together against CIB1 over four rounds of selections (Figure 3A). Enrichment began in round three and grew to 11% of the input in round 4 (Figure S5), at which point NGS revealed at least three CIB1 peptide families (C1–3): families C1 and C2 came from the NNK₉ library whereas C3 came from the NNK₆ library (Figure 3B). Interestingly, a number of Lys residues emerged in these hits, in particular at the i+4 position relative to the static Gln residue in the parent library, presenting a potential alternative site for cyclization. Additionally, a V to A mutation had occurred in the WAL tag of the C3 sequences, likely due to PCR error during selection.

Figure 3.

A) Schematic detailing the selection process against CIB1. B) NGS results after four rounds of selections against CIB1. Sequences are grouped by family type. The sequence selected for validation efforts is indicated by a red arrow. C) TR-FRET results for all variants of the selected hit. IC₅₀ values (nM) are depicted in the table under the TR-FRET curves. Each peptide was analyzed 9 times. D) in vitro mTG assay comparing the linear or unmodified peptide vs the product of the mTG rxn. These traces are show against the two linkage types (i,i+4 and i,i+7). EIC trace of the corresponding masses were extract from the LCMS results and shown above. E) Circular dichroism analysis of the linear, i,i+4 and i,i+7 hit peptide variants. F) modeling the lowest energy conformation for both the i,i+4 and i,i+7 in solution. G) Docked structure of the i,i+4 (most active variant) into the crystal structure of CIB1.

To narrow hits for synthesis we carried out a qPCR-based orthogonal sequence validation assay (Figure S5). The top hit from C1.1 (0.95%) and second most enriched hit from C3.2 (0.44%) performed best in this assay, which recapitulates the selection on a single sequence. We opted to pursue hit C3.2 for synthesis by solid phase peptide synthesis (SPPS) and further analysis because of the additional synthetic complexity and difficulty in deconvolution of the multiple Lys residues in the C1.1 hit. Since the C3.2 hit has two Lys residues and therefore, two potential sites of cyclization, we prepared both regioisomers as well as the linear substrate (synthetic route, Figure S10). These synthetic peptides were then compared by liquid chromatography/mass spectrometry (LCMS) to small scale in vitro reactions with mTG carried out under similar conditions to the mRNA display reaction. Interestingly, the enzymatic reaction gave exclusively a single regioisomer, which lined up with the smaller (i,i+4) of the two potential macrocycles, suggesting a preference in reactivity for mTG (Figure 3D). We next screened all three peptides, the linear, i,i+4, and i,i+7, in our previously reported CIB1 TR-FRET assay to assess inhibitory activity. The i,i+4 macrocycle, which accords to the enzymatic product, exhibited the best inhibitory activity in this assay at 68nM IC₅₀. The linear peptide was almost an order of magnitude less potent at 422nM and the i,i+7 macrocycle exhibited low μM inhibition. We further prepared a truncated version of the i,i+4 macrocycle, in which the N-terminal MWVL was removed. Removal of this motif completely ablates activity, indicating that the hit evolved around this fixed recognition element (Figure 3C).

We next looked at the potential implications of the macrocyclization site for the structure and inhibitory activity of the cyclic peptide. The i,i+4 positioning of the active macrocycle is well known for stabilizing α-helixes. Additionally, CIB1 is a known binder of α-helixes.²⁶ Therefore, we measured circular dichroism (CD) spectra for all three peptides to assess their degree of helicity. The linear peptide showed low ellipticity from 210 to 250 nm, suggesting an unstructured or disordered peptide and the larger, i,i+7 macrocycle displayed a slight negative band at 220nm but a positive signal at 210 nm, suggesting a small degree of helicity but also some other secondary structure. In contrast, the smaller, i,i+4 macrocycle displayed two strong negative bands at 210 and 220 nm and a positive shift at the lower ends of the spectra, indicating strong helical secondary structure (Figure 3E).²⁹ Molecular dynamics (MD) simulations of the two macrocycles either in complex with CIB1 or separately agrees with the CD data. An x-ray structure of CIB1 in complex with UNC10245092, a helical peptide identified from a phage display pan (PDB: 6OD0) was used as a structural template to generate starting conformations for MD.²⁷ The MD simulations of solvated peptides suggest that the i,i+4 lactam bridge likely lowers the entropic cost of binding by stabilizing this helicity prior to binding, while the i,i+4 bridge rather destabilizes the helical conformation (Figure 3F). Moreover, peptide-CIB1 simulations show how the linear version of the C3.2 hit might adopt a tight helical structure, which fits cleanly into the CIB1 binding pocket (Figure 3G).

Selection of mTG cyclized B7-H3 Binders.

We next sought to use mTG to discover new peptide ligands for a currently unliganded therapeutic target in the immune checkpoint protein B7-H3.^30,31 Multiple clinical trials are currently underway for biologic-based therapies, such as chimeric antigen receptor T (CAR T) cell and monoclonal antibody-based agents, against B7-H3-expressing cancers.³² However, no small molecule or low molecular weight scaffolds have been identified as binders for this target. To identify mTG-cyclized peptides against B7-H3, we deployed the same conditions and same libraries as used in the CIB1 pans. The combined NNK₆ and NNK₉ libraries were screened over six rounds against biotin immobilized B7-H3. Enrichment was observed in round 6 at 1.3% of the input (Figure S8A). NGS revealed strong enrichment of a single B7-H3 family (B1), all members of which came from the NNK₉ library. Sequences in this family displayed cysteines at positions 7 and 14 relative to the N-terminal methionine, suggesting the potential selection of a bicyclic disulfide-bridged macrocycle (Figure 4A). We prepared the top hit (B1.1) in both linear and bicyclic (both lactam and disulfide) forms by SPPS. To simplify synthesis, we opted not to include the N-terminal methionine in the synthetic peptides, as it is prone to oxidation and also prepared the Met9Nle in parallel because of similar concerns (Figure S10). Notably, while there are potentially multiple different interlocking conformations possible with the bicyclic B1.1, the synthetic material has an identical retention time via LCMS as enzymatically-cyclized material prepared under the display conditions (Figure 4B). We assessed binding affinities for each peptide by surface plasmon resonance (SPR) against biotinylated B7-H3. Pleasingly, B1.1 displayed a 43.5nM binding affinity towards B7-H3 and the Met9Nle variant was similar. Meanwhile, the linear variant showed no detectable binding interaction (Figure 4C). This data demonstrates the necessity of cyclization for binding and strongly suggest its influence in the identification of our B7-H3 hits. Overall, these results show that this method is capable of identifying potent, nanomolar cyclic binders of new targets.

Figure 4.

A) NGS results after six rounds of selections against B7-H3. The sequence selected for validation efforts is indicated by a red arrow. B) EIC ([M+2], 948.91m/z) traces corresponding to the product of the mTG reaction with linear B1.1 peptide and the synthetic lactam, disulfide peptide. C) Surface Plasmon Resonance (SPR) analysis of the selected peptide (B1.1, black), the Met9Nle variant (blue), and the linear variant (red). Top shows the fitted curves for each peptide. The selected B1.1 and the Met9Nle variant were analyzed at analyte concentrations of 7.5, 15, 30, 60, 120nM. The linear variant was analyzed at analyte concentrations of 50, 100, 200, 400, 800nM. Bottom shows binding kinetics of each peptide.

Discussion.

This work shows the potential power of merging enzymatic post-translational modifications with mRNA display. We developed two display-based assays that separately show the compatibility of mTG activation and cyclization with mRNA display libraries and begin to assess the substrate bias of the enzyme. mTG adds a novel, structurally-distinctive, and pharmaceutically-relevant isopeptide bond to the small but growing list of macrocyclization chemistries that are now accessible to mRNA display libraries. Notably, this chemistry is distinguished by its orthogonality to substrate cysteine-residues, which are commonly employed as cyclization handles; mTG cyclization leaves substrate cysteines open for alternative modification chemistries that can be used to broaden library diversity. We have further demonstrated the ability of this strategy to identify new ligands: the CIB1 inhibitory peptide, C1.1 constitutes a new and unique ligand for this investigational target and the B7-H3 binding peptide, B1.1 and its Met9Nle variant, provide novel scaffolds for development as diagnostic probes or therapeutic delivery systems to B7-H3 expressing cancers. Both the substrate display results and the target selections provide insights into the opportunities and challenges of enzyme-modified display campaigns.

Our substrate display data suggests that mTG is broadly promiscuous, although there is a level of substrate bias that may be able to work its way into selections. It is important to note the limitations of this current promiscuity assessment: due to sequencing constraints, we did not cover the full theoretical diversity of several of the libraries in these experiments. Future experiments, employing more extensive, deep sequencing analysis or other methods will be necessary to comprehensively analyze the substrate bias of this and similar candidate enzymes. Still, most of the reads recovered for the substrate display experiments are for unique sequences and some changes in position level composition are statistically significant, contributing to several identifiable trends. One significant finding from the activation (PAB) assay is that a recognition element (e.g. the WAL epitope) directly around the Gln residue undergoing activation may not be required and Gln residues in diverse peptide contexts may be able to undergo activation by mTG. Similarly, in the cyclization assays there may be a bias towards new cyclization partner residues (Gln and Lys) in closer proximity to the fixed residues. The latter result suggests that partner spacing and thus ring size might have a stronger impact on cyclization pairs than sequence context. Substrate conformation and/or the general electrostatic character of the enzyme active site may play roles in selecting which ring sizes are preferred. A deep sequencing analysis of these assays could further elucidate these potential trends and begin to distinguish kinetic versus thermodynamic control in the biocatalytic context. Designer DNA sequences and libraries could be used to control these substrate biases especially in placement of the Gln and Lys cyclization partners.

Substrate bias did not prevent selection of diverse peptide macrocycles in pans against CIB1 and B7-H3. The CIB1 selectant bears no clear sequence homology to previously selected CIB1 peptides, although the helical conformation allowed a docking pose that situates hydrophobic Leu and Ile residues in the same CIB1 pockets as aromatic residues on the previously identified phage peptide. The selection clearly favored an i+4 Lys residue in several of the hit families, which was capable of undergoing mTG-catalyzed cyclization to stabilize a helical conformation. Enrichment of this helical motif may have been caused by mTG or else by CIB1, which natively binds a number of helical partners, or both. The B7-H3 selectant is likely conformationally more complex that the CIB1 selectant, but it is doubtful that the bicyclic molecule can achieve the same kind of extended helix while accommodating both bridges. Cumulatively, these molecules foretell of the extremities of structural diversity that might be accessible through mTG libraries in particular and enzymatic libraries in general.

Conclusion.

Together, these steps, substrate display followed by ligand selection, suggest a model framework for introducing an enzyme into the mRNA display toolkit and pinpoint several of the knowledge gaps to be traversed in the process. Enzymatic post-translational modifications present a rich and appealing toolset for mRNA display diversification.^7,8 Although further studies may provide insights into preferred cyclization methods for a given target class, we anticipate that access to more cyclization methods and thus greater structurally diversity can only benefit naïve panning campaigns.³³ Perhaps in future applications, multiple enzymatic modifications could be mixed or layered together as in natural product biosynthesis, to enable screening of ever more complex and privileged scaffold libraries. We anticipate that access to such libraries would aid target engagement, serum stability, and cell penetrance of next generation inhibitors and macrocyclic therapeutics.

Materials and Methods.

Full materials and methods are provided in the supplemental information.

Reconstitution of microbial transglutaminase (mTG)

Lyophilized mTG (T001) was purchased from Zedira and reconstituted following manufacture’s protocol. See supplemental information for full protocol.

General translations and reverse transcription procedure for mRNA display.

Translations for mRNA display were carried with a customized NEB PURExpress kit (-aa, -tRNA, -RF123) -E6850Z. A solution containing 1x SolA (supplied as 5x stock), 0.5mM tRNAs, 1.2mM Plinked mRNA, 0.5mM Amino acid mix, MQ-H₂O (to reach final volume), and 1x SolB (added last to initiate translation, supplied as 3.3x stock) was prepared. The solution was incubated at 37°C for 30mins, followed by a 10min incubation at RT to facilitate fusion of peptide to its mRNA strand. Finally, EDTA was added to a final concentration of 17mM, to dissociate the ribosome, and the mixture was incubated at 37°C for 20mins. Next, complementary DNA was added by a reverse transcription reaction containing all the translation product, 0.6mM dNTPs, 5uM reverse primer (P9), 62.5mM Tris-HCl pH 8.3, 37.5mM Mg(OAc)₂, 25mM KOH, 2.5x M-MLV reverse transcriptase H (−), point mutant (Promega, M3681, supplied at 40x), and MQ-H2O (to reach final volume). This solution was incubated at 42°C for 1h. After incubation an analytical sample (0.5uL of translation/reverse transcription product diluted into 250mL MQ-H₂O) was taken for qPCR analysis of initial translation.

Glutamine activation assay analysis.

Plinked mTG-NNK₆ Ligation mRNA (Lib1. see supplemental information) was translated using a full complement amino acid mix (20/20AA). A 10μL reaction (set up in triplicate) was prepared, translated, and transcribed as stated above. Once complete the fusions were purified using anti-HA magnetic beads at a concentration of 4mL bead slurry to 1mL of IVT. This mixture was then diluted 10x from the original translation volume (eg. 100mL) and incubated, while rotating, at RT for 0.5h. Once complete, the beads were washed 3x with 1xTBST (50mM Tris-HCl pH 8.0, 200mM NaCl, 0.05% Tween-20). The fusion bound bead solution was then concentrated to 40μL and then split into two equal portions. Each portion was treated (+mTG) or left untreated (−mTG). To carry out a 100μL reaction was set up containing 4μM mTG, 20μL of fusion-bead sample, 1mM pentylamine biotin (ThermoFisher Scientific, 21345 – prepared in MQ-H₂O as a 10mM stock solution), and diluted to the final volume with 1xTBST. The untreated sample was diluted to the sample volume as the (+) mTG sample (100μL) with 1xTBST. Both samples were then incubated on a 37°C benchtop shaker (to prevent the anti HA beads from clumping) for 1h. After incubation each solution was washed 3x with 1xTBST. The samples were eluted from the anti-HA beads in a 50μL solution containing 25μL of 4mg/mL HA synthetic peptide (prepared in MQ-H₂O) and 2xTBST. Elution was carried out for 1h at RT. Once complete magnetized Streptavidin beads (4mL of streptavidin beads/ 1mL original translation reaction in each sample – eg. 20μL of SA beads to 5μL of IVT) were added to each sample to capture any biotinylated material. Samples were incubated while rotating at 4°C for 0.5h then washed 3x with 1xTBST. After the final wash each sample was brought up in 1xPCR buffer (10mM Tris-HCl pH 9.0, 50mM KCl, and 0.1% Triton X-100) and incubated at 95°C for 5mins to dissociate cDNA from each fusion. The supernatant was collected and the amount of recovered cDNA was determined by qPCR (see below). Recoveries were analyzed as a function of the HA purified input amount. Isolated cDNA was PCR amplified based on the CT value from the qPCR results using standard Q5 polymerase protocols and primers (P3, P9 – annealing temperature of 57°C). Once sufficient amplification was observed then the DNA was crashed out using a 0.1x (v/v) of 3M NaCl and 2x (v/v) of 100% ethanol. This was spun down at 15000rpm for 10mins, supernatant removed, washed with 70% EtOH and then allowed to dry.

Trypsin titration selection assays – control sequence.

Plinked mTG-Control mRNA was translated with a 19AA amino acid mix (0.5mM in solution) lacking methionine. The translation was supplemented with 50mM Bio-Phe-itRNA (prepared as stated above and solubilized in 1mM NaOAc pH5.2). A 16mL reaction was prepared, translated, and transcribed as stated above. Once complete the fusions were purified using anti-HA magnetic beads at a concentration of 4mL bead slurry to 1mL of IVT. This mixture was then diluted 10x from the original translation volume (eg. 160mL) and incubated while rotating at RT for 0.5h. Once complete, the beads were washed 3x with 1xTBST (50mM Tris-HCl pH 8.0, 200mM NaCl, 0.05% Tween-20) and brought up in 0.5x of original binding volume (80mL). This was then split into two equal portions. To one portion mTG was added to a final concentration of 4mM (saturating equivalence of mTG to fusion) and then diluted to 80mL with 1xTBST. The other portion was left untreated and diluted to a final volume of 80uL. Both samples were incubated on a 37°C stand mixer (to ensure the beads remain suspended) for 1h. After incubation, both samples were washed with 1xTBST, to remove mTG from solution, then resuspended with 40mL (0.5x volume) of 2xTBST and 40uL of 4mg/mL HA peptide (to elute the fusion from the anti-HA bead). This was carried out at RT for 1h. The final elution volume for each portion was 80mL. Both samples were split out into 10mL aliquots (~1mL of original translation volume). Trypsin, prepared at 10mM stock in 50mM Acetic Acid, (MS/MS Pearce, 90057) was added at concentrations ranging from 0 to 800nM. Samples were rotated at 4°C for 0.5h. Once complete magnetized Streptavidin beads (4mL of streptavidin beads/1mL original translation reaction) were added to each sample followed by 2x volume of 1xTBST (to dilute trypsin). Samples were incubated while rotating at 4°C for 0.5h then washed 3x with 1xTBST. After the final wash each sample was brought up in 1xPCR buffer (10mM Tris-HCl pH 9.0, 50mM KCl, and 0.1% Triton X-100) and incubated at 95°C for 5mins to dissociate cDNA from each fusion. The supernatant was collected and the amount of recovered cDNA was determined by qPCR (see below). Recoveries were analyzed as a function of the non-trypsin digested sample (0nM).

mTG timecourse lysine cyclization assay.

Plinked mTG-NNK₄ mRNA (Lib2., see supplemental information) was translated with a 19AA amino acid mix lacking methionine. The translation was supplemented with 50mM biotin-Pheinitiator tRNA (solubilized in 1mM NaOAc pH5.2). A 25mL reaction was prepared, translated, and transcribed as stated above. Once complete the fusions were purified using anti-HA magnetic beads at a concentration of 4mL bead slurry to 1mL of IVT. This mixture was then diluted 10x from the original translation volume (eg. 250mL) and incubated while rotating at RT for 0.5h. Once complete, the beads were washed 3x with 1xTBST (50mM Tris-HCl pH 8.0, 200mM NaCl, 0.05% Tween-20) and brought up in 250mL of 1xTBST. Then 40mL (~4mL original IVT) was taken out washed 3x with 1xTBST and stored on ice in 1xTBST – this was the 0min timepoint sample. The remaining 210mL was then concentrated and resuspended in 1xTBST supplemented with 4μM of mTG. This was added to a 37°C stand mixer and 40μL samples taken at 3.75, 7.5, 15, 30, and 60mins. Each sample was washed in a similar fashion to the 0min timepoint sample. After the timecourse incubation each aliquot was concentrated and resuspended in ½ volume (eq 20μL) 2xTBST and ½ volume solution of 4mg/mL HA synthetic peptide (prepared in MQ-H₂O). This was mixed on a rotator at RT for 1h. Following incubation, the supernatant was collected (contains eluted fusion) and each sample split into equal portions (~2μL IVT, 20μL total). To one of each sample a solution of 20μL of 0.2nM Trypsin (MS/MS Pearce, 90057) was added. The trypsin was prepared as a 10μM stock in 50mM Acetic Acid. To the other sample 20μL of 1xTBST was added. Both sample types were incubated on a rotator at 4°C for 0.5h. Once complete magnetized Streptavidin beads (4mL of streptavidin beads/1mL original translation reaction) were added to each sample followed by 2x volume of 1xTBST (to dilute trypsin). Samples were incubated while rotating at 4°C for 0.5h then washed 3x with 1xTBST. After the final was each sample was brought up in 1xPCR buffer (10mM Tris-HCl pH 9.0, 50mM KCl, and 0.1% Triton X-100) and incubated at 95°C for 5mins to dissociate cDNA from each fusion. The supernatant was collected and stored at 4°C until ready for qPCR analysis (see qPCR analysis section for sample preparation). Recoveries were analyzed as a function the corresponding non-trypsin digested sample (eg. 1h mTG [+ trypsin]/1h mTG [− Trypsin]). This analysis was carried out in triplicate.

Lysine cyclization assay.

mTG-NNK₄ and mTG-NNK₆ plinked RNA (Lib2. And Lib3., see supplemental information) were carried out in similar fashion. Translation occurred (in triplicate) with a 19AA amino acid mix lacking methionine. The translation was supplemented with 50mM biotin-Phe-initiator tRNA (solubilized in 1mM NaOAc pH5.2). An 8.0μL reaction was prepared, translated, and transcribed as stated above. Once complete the fusions were purified using anti-HA magnetic beads at a concentration of 4mL bead slurry to 1mL of IVT. This mixture was then diluted 10x from the original translation volume (eg. 80mL) and incubated while rotating at RT for 0.5h. Once complete, the beads were washed 3x with 1xTBST (50mM Tris-HCl pH 8.0, 200mM NaCl, 0.1% Tween-20) and brought up 80mL of 1xTBST. The sample was then split into equal portions (40μL), concentrated, and then brought back up in 1xTBST supplemented with or without 4μM mTG (40μL total solution). This slurry was mixed at 37°C for 1h on a benchtop shaker. After incubation, both samples were washed with 1xTBST, to remove mTG from solution, then resuspended with 20mL (0.5x volume) of 2xTBST and 20uL of 4mg/mL HA peptide (to elute the fusion from the anti-HA bead). This was carried out at RT for 1h. The final elution volume for each portion was 40mL. Each sample (both treated or untreated with mTG) was then further split into equal portions (20μL). One sample was digested with 100nM trypsin (prepared as reported above) in a final reaction volume of 50μL (1xTBST). The other sample was left untreated but diluted to 50μL in 1xTBST. All samples were incubated at 4°C for 0.5h while rotating. Once complete magnetized Streptavidin beads (4mL of streptavidin beads/1mL original translation reaction) were added to each sample followed by 2x volume of 1xTBST (to dilute trypsin). Samples were incubated while rotating at 4°C for 0.5h then washed 3x with 1xTBST. After the final wash each sample was brought up in 1xPCR buffer (10mM Tris-HCl pH 9.0, 50mM KCl, and 0.1% Triton X-100) and incubated at 95°C for 5mins to dissociate cDNA from each fusion. The supernatant was collected and the amount of recovered cDNA was determined by qPCR (see below). Recoveries were analyzed as a function of their corresponding non-digested sample. Isolated cDNA was PCR amplified based on the CT value from the qPCR results using standard Q5 polymerase and primers (P3, P9 – annealing temperature of 57°C). Once sufficient amplification was observed then the DNA was crashed out using a 0.1x (v/v) of 3M NaCl and 2x (v/v) of 100% ethanol. This was spun down at 15000rpm for 10mins, supernatant removed, washed with 70% EtOH and then allowed to dry. The (+) mTG and (+) trypsin samples were sent for NGS analysis (below)

CIB1 and B7-H3 seelctions

CIB1 - Round 1:

A 70uL in vitro translation (IVT) reaction which contained plinked RNA for both mTG-NNK₆ and mTG-NNK₉ (Lib3. And Lib4., see supplemental information) was prepared, translated, and transcribed as above. Once complete the fusions were purified using anti-HA magnetic beads at a concentration of 4iL bead slurry to 1uL of IVT. This mixture was then diluted 10x from the original translation volume (eg. 700mL), incubated while rotating at RT for 0.5h, then washed 3x with 1x CIB1 selection buffer (50mM Tris-HCl pH 8.0, 150mM NaCl, 0.2mM CaCl₂, 0.05% Tween-20). The fusion bound-HA beads were brought up in 200μL of 2xCIB1 selection buffer and then 200μL of HA synthetic peptide (prepared as a 4mg/mL solution in MQ-H₂O) was added. This mixture was rotated at RT for 1h and upon completion the supernatant, containing purified fusion (400μL), was collected. To this solution mTG (50μM) and 1x CIB1 selection buffer was added to a final concentration of 4μM of mTG in solution (final volume 450μL). This was incubated at 37°C for 1h. During incubation CIB1 was loaded onto SA beads at a concentration of 2μL of SA beads/1pmol of CIB1. This conversion was used to set up the selection against 0.2μM CIB1 in 450μL total solution. To prepare, the requisite amount SA beads (180μL) were aliquoted and washed with 1xTBST+CaCl₂. Then biotinylated CIB1 was added for a final concentration of 0.2μM in 450μL. This was rotated at 4°C for 15mins. Then excess biotin (25μL of 0.5M in MQ-H₂O) was added and the solution was incubated for an additional 15mins -this step was to block any unreacted SA. Once complete the solution was removed from the beads and the fusion solution from the mTG reaction was added to the CIB1-SA mixture. This was incubated while rotating at 4°C for 0.5h. After incubation the SA beads were washed 3x with 1x CIB1 selection buffer. On the final wash the SA beads were resuspended into 10mM Tris-Cl pH8.3, 50mM KCl, 0.1% Triton X100 (1xPCR Buffer). To isolate the cDNA the SA bead solution was heated at 95°C for 5mins and the supernatant collected. The amount of recovered cDNA was determined by qPCR (see below) and analyzed as a function of the HA purified input amount. The cDNA was amplified based on the CT value from the qPCR results using standard Q5 polymerase protocol and primers (P3, P9 – annealing temperature of 57°C). Once sufficient amplification was observed then the DNA was crashed out using a 0.1x (v/v) of 3M NaCl and 2x (v/v) of 100% ethanol. This was spun down at 16000xg for 10mins, supernatant removed, washed with 70% EtOH and then allowed to dry. A 20μL transcription was set up using standard T7 RNA polymerase (NEB) to prepare material for the second and subsequent rounds.

CIB1 - Subsequent rounds:

Isolated RNA from the first round of selections was then Plinked (as stated above) and carried forth into subsequent rounds. A 5μL IVT reaction was prepared, translated, and transcribed as above. This was then HA purified at 4μL of HA bead slurry to 1μL of original IVT input (eg. 20μL of anti-HA beads to 5μL of IVT input). This mixture was then diluted 10x from the original translation volume (eg. 50mL) and incubated while rotating at RT for 0.5h. Once complete, the beads were washed 3x with 1x CIB1 selection buffer (50mM Tris-HCl pH 8.0, 150mM NaCl, 0.2mM CaCl₂, 0.05% Tween-20). The fusion bound-HA beads were brought up in 25μL of 2x CIB1 selection buffer and then 25μL of HA synthetic peptide (prepared as a 4mg/mL solution in MQ-H₂O) was added. This mixture was rotated at RT for 1h and upon completion the supernatant was collected, containing purified fusion (50μL). To this solution mTG (50μM) and 1x TBST+CaCl₂ was added to a final concentration of 4μM of mTG in solution. This was incubated at 37°C for 1h. Next, negative selections against SA beads were performed to remove background binders – this was carried out for all rounds after the initial round. Three background selections were carried out at an equal volume of SA beads to input IVT (eg. 5μL of IVT to 5μL of SA bead). To prepare 15μL of SA beads were aliquoted, washed with 1x CIB1 selection buffer, and then split into equal portions. To one portion biotin was added to a final concentration of 10mM. Both portions were rotated at 4°C for 0.5h, then washed, resuspended in 1x CIB1 selection buffer, and combined (final volume of 15μL). This sample was then aliquoted into 3×5μL portions. The HA purified and mTG treated fusion was then placed over these beads, incubated while rotating at 4°C for 0.5h, then removed. This process was repeated for every background selection. On the final background selection, after the fusion solution was removed, the beads were washed 3x with 1x CIB1 selection buffer and then suspended in 1xPCR buffer. The sample was heated at 95°C for 5mins and the supernatant (containing bead binding cDNA) was collected for qPCR analysis. After screening through background selections, the library was incubated against 0.2mM CIB1 bound to SA (prepared as above but scaled down for this selection). This was carried out at 4°C for 0.5h, washed 3x with 1x CIB1 selection buffer, and then suspended in 1xPCR. The cDNA was eluted by heating at 95°C and collecting the supernatant. qPCR analysis, PCR amplification, and subsequent round RNA transcription was carried out as described in round 1. This process was repeated until sufficient enrichment was observed. For this selection (CIB1) we observed an spike in round 4 recovery to 11% of the HA purified input. The material from round 4 was then sent for NGS analysis.

B7-H3 selection:

The B7-H3 selection was carried out according to the same procedure as for CIB1 selection rounds (described above). B7-H3 selection buffer was comprised of 1xPBS (prepared from 10x stock, Gibco 14200–075) supplemented with 0.1% Tween-20. Selections were carried out until enrichment was observed. In this case, significant enrichment was observed in round 6 (1.3% of the HA purified input). Material from round 6 was sent for NGS analysis.