Directed evolution of 2′-fluoro-modified, RNA-supported carbohydrate clusters that bind tightly to HIV antibody 2G12

Abstract

Carbohydrate binding proteins (CBPs) are attractive targets in medicine and biology. Multivalency, with several glycans binding to several binding pockets in the CBP, is important for high-affinity interactions. Herein, we describe a novel platform for design of multivalent carbohydrate cluster ligands by directed evolution, in which serum-stable 2′-fluoro modified RNA (F-RNA) backbones evolve to present the glycan in optimal clusters. We have validated this method by the selection of oligomannose (Man₉) glycan clusters from a sequence pool of ~10¹³ that bind to broadly neutralizing HIV antibody 2G12 with 13 to 36 nM affinities.

Many events in host-pathogen recognition, cell adhesion and cell signaling are mediated by carbohydrate-protein interactions.¹ Therefore, the design or discovery of carbohydrate structures that can faithfully mimic or interrupt interactions with carbohydrate binding proteins (CBPs) has potentially broad applications in medicine, including the development of vaccines,² immunomodulatory agents,³ diagnostics,⁴ and anti-adhesion drugs against cancer and other diseases.⁵

Whereas CBPs typically have low affinity (K_D ~ mM to μM) for individual glycan units, high affinity interactions are achieved through multivalent interactions with glycan clusters.⁶ Rather than using rational design and medchem to optimize multivalency to match the target binding sites (Figure 1a), our lab is developing directed evolution methods to rapidly select optimal multivalent glycan clusters from extremely diverse libraries of 10¹²-10¹³ structures (Figure 1b). Previously, we have clustered carbohydrates on libraries of DNA aptamer-⁷ or peptide backbones and developed methods to genetically encode and amplify these libraries.⁸ RNA has a more diverse structural repertoire than DNA, motivating the display of glycan clusters on RNA backbones.⁹ Although RNA is highly susceptible to nuclease degradation, 2′-deoxy-2′-fluoro RNA (F-RNA) exhibits increased serum stability.¹⁰ Herein, we report a new method for directed evolution of carbohydrate clusters assembled on F-RNA backbones. We demonstrate that this method can evolve tight binders of the HIV antibody 2G12,¹¹ a target with four glycan binding pockets¹² that is of interest in HIV vaccine design.^2b-d

Figure 1.

Design strategies for carbohydrate cluster ligands. a) Rational design or trial and error typically result in mid μM to high nM binding affinities. b) In our approach, carbohydrate clusters presented on an extremely diverse library of ~10¹³ modified F-RNA scaffolds are subjected to selection. Sampling this vast structure space yields potentially highly optimized clusters with low nM binding affinities.

Unlike our previously reported SELMA^8a (SELection of Modified Aptamers) method, in which glycan-modified DNA was covalently displayed on unmodified DNA,^{7b, 13} the present method (Capture SELMA) utilizes a capture strategy (Scheme 1), in which a nascent RNA transcript from a DNA template is non-covalently hybridized to a “capture strand”, and thus travels with its DNA template. The DNA template can be amplified by PCR, regardless of subsequent glycosylation of the RNA, without reverse transcription.¹⁴ Selection begins with a (-)-strand template (Form A) encoding the random library (25 random bases in red). The capture strand is annealed (Form B), then extended to generate dsDNA library template (Form C). A “capture strand rigidifier” is then hybridized to most of the capture strand (Form D) to prevent secondary structure formation and discourage selection of sequences that depend on hybridizing anywhere other than to the 5′ end (blue). T7 R&DNA polymerase¹⁵ and triphosphates are then added to transcribe F-RNA from the template (Form E). The 5′ end of the template strand contains an isodC modified nucleobase (green star) to stall the polymerase and prevent dissociation from the template.¹⁶ The nascent F-RNA (Form E, wavy strand) is then captured on a region of complementary sequence at the 5′ end of the capture strand (Form F).

Scheme 1. Library Construction^a.

^a a) Architecture of the Capture SELMA construct. b) structure of the Man₉ oligomannose glycan. c) Capture SELMA cycle. Translucent colors denote F-RNA and solid colors denote DNA. The stem-loop pictured in the RNA region is illustrative of potentially folded RNA, but library sequences are random in the red N₂₅ region.

Transcription is carried out in the presence of ribopurines but 2′-deoxy-2′-fluoropyrimidines, where uridine is replaced with 2′-deoxy-2′-fluoro(5-ethynyl)uridine, for CuAAC¹⁷ attachment of glycan azides.^{14c, 18} The resulting glycosylated F-RNA library (Form G) then undergoes selection for binding to the target and the bound fraction is amplified by PCR with a biotinylated forward primer (Form H). The biotinylated strand is then removed with streptavidin beads, regenerating Form A of the library to begin the next round of selection.

Several features of the oligonucleotide design were important for optimal library function. First, to avoid early transcript termination in the presence of fluoropyrimidines¹⁹, the beginning of the RNA sequence (blue) was designed to contain only natural purine bases. Additionally, this sequence, which must anneal to the capture strand, was designed to avoid formation of G quadruplexes²⁰ or stable secondary folds that could interfere with efficient hybridization of the RNA transcript to the capture strand (see SI Figures 1-4 for gel analyses illustrating library construction).²¹ A control experiment verified that the captured RNA strand did not significantly exchange with free RNA transcripts in solution (SI Figure 5). A poly(A) spacer (orange) was included to distance the capture annealed region from the random region (red) bearing the glycans. The required ethynyl fluorinated uridine triphosphate (1) was prepared by a brief synthetic sequence (Scheme 2). 1 was readily incorporated by R&D polymerase into the transcription product (SI Figures 1, 3). Denaturing and native polyacrylamide gels exhibited bands of the expected size for each stage of the selection cycle (SI Figures 1, 2).

Scheme 2. Synthesis of 2′-deoxy-2′-fluoro(5-ethynyl)uridine triphosphate^b.

^bReagents and conditions: (a) 2′-fluoro-2′-deoxyuridine, NIS, DMF, 60°C, 7 h, 82%; (b) Pd(PPh₃)₂Cl₂, CuI, ethynyltri-ethylsilane, ACN, Et₃N, 60°C, 1 h, 86%; (c) Proton sponge, (MeO)₃PO, POCl₃, −20°C, 1 h; then [Bu₃NH]₂P₂O₇, Bu₃N, DMF, −20°C, 45 m, 22%; (d) TBAF, ACN, rt, 18 h, 91%.

With an optimized protocol in hand for generating the library, we commenced selection for 2G12 binders. For the first round, a 20 pmol library (~1.2 × 10¹³ library members) was incubated with 150 nM 2G12, and the bound complexes were captured on Protein A magnetic beads. The bead-bound library fraction was eluted by 3.5 M MgCl₂ disruption of the antibody/protein-A interaction and quantified by qPCR (Figure 2a). Although successive rounds of selection included more stringent conditions (lower antibody concentration, higher temperature, increased washes, see Figure 2a and SI Table 2), a small parallel selection was run under the original conditions at each round to assess enrichment (Figure 2b). Prior to each selection with target, the library was incubated with empty beads and the supernatant was transferred to another tube to select against bead binders or plastic binders.

Figure 2.

Library fraction bound and clone frequency at each round of selection. a) Percent of library bound under selection con-ditions, quantified by qPCR. b) Control conditions to monitor library fitness: percent bound to constant 150 nM 2G12, quantified by qPCR. c) Clone frequency over final six rounds of selection.

When qPCR indicated increases in library recovery, stringency was increased by lowering the 2G12 target concentration (eventually down to 2 nM), increasing the number of wash cycles and volume for bead-bound library, increasing selection temperature from room temperature to 37 °C, or switching from Protein A to Protein G beads. Recovery increased dramatically in the first eight rounds (Figure 2b) from < 0.1 % to ~10 %. The increase of selection temperature in round 9 and the switch to Protein G beads in round 13 each resulted in temporary decreases in library recovery (Figure 2a), which typically increased again after multiple rounds under the same conditions.

Throughout selection, the number of glycans per library member, as assessed by PAGE of DNAse-digested library (SI Figure 6), narrowed from a broad distribution to 2-4 glycans/clone, and the melting behavior of the library DNA visibly shifted to a higher temperature distribution (SI Figure 7). 36 clones from rounds 13, 15, 17 and 18 were sequenced and their 2G12 binding behavior assessed (Figures 2c, 3). Nitrocellulose filter binding assays²² were conducted with glycosylated F-RNA transcripts, whereas in an alternative assay, clones were produced as glycosylated F-RNA/DNA hybrids (Form G) and qPCR was used to quantify 2G12-bound complexes captured on protein A beads.

Figure 3.

Predicted fold of clones 15F1 and 15F3 and mutant sequences with binding data. a) Filter binding curves (left Y-axis) and qPCR binding curves (right Y-axis) for clones 15F1 and 15F3. b) 15F3 mutant 1 binding (qPCR assay). c) 15F1 RNAFold-predicted structure. Green U = modified uridine, bearing Man₉-cyclohexyl-triazole at uridine 5 position. The red asterisk indicates a predicted G-U wobble base pair. X = (A)₁₀ spacer (see SI Table 1). d) 15F3 RNAFold-predicted structure. e) Binding data and sequences. Red = glycosylation sites, cyan = mutated glycosylation sites, yellow = auxiliary mutations that maintain stem stability after glycosylation site mutation. nd = not determined. nb = no binding at 300 nM 2G12. Data were obtained in duplicate. Reported errors are the standard error of the curve fit.

Among the major clones observed in the final rounds, 15F1 and 15F3 became most dominant after rounds 17 and 18. They bound (Figure 3a,e, SI Figure 10) with nanomolar affinity to 2G12 (K_Ds of 40 ± 6 and 13 ± 2 nM, respectively) in the filter binding assay, and those results agreed well with the qPCR binding assay results (K_Ds of 40 ± 4 and 19 ± 3 nM, respectively). In the predicted secondary fold (RNAfold)²³ of both clones, the reverse primer constant sequence hybridizes with three bases at the 5′-end of the aptamer where one glycan is present, whereas the remaining 2-3 glycans are on a stem-loop feature (Figure 3c, d).

To further understand the structure and glycan-dependency of binding of the two clones that displayed nanomolar binding, we tested single U->X transition or transversion point mutations to remove each possible glycosylation site (three for 15F1 and four for 15F3) while the complementary base in the predicted stem structure was mutated to retain the predicted fold (Figure 3e, SI Tables 3,4). The qPCR binding method was used to test each mutant at 0 nM and 300 nM 2G12 (SI Figure 11, 12). Only 15F3-mutant 1 (“M1”) displayed any binding to 300 nM 2G12, suggesting that three glycosylation sites are critical in both clones, with only the fourth glycan of 15F3 being superfluous for binding. A full concentration series was performed to measure the K_D of 15F3-M1, revealing a modest affinity reduction (31 ± 4 nM versus 19 ± 3 nM) and a significant decrease in maximum concentration bound, compared with the parent clone. (Figure 3b, e).

The observed Fb_max values were low (< ~20%); to assess whether this was because of incompletely glycosylated aptamer, we purified the fully glycosylated 15F1 and 15F3 F-RNA by urea PAGE and repeated the nitrocellulose filter binding assay. Although glycoDNA aptamers from SELMA and natural RNA aptamers derived from our capture method exhibited excellent binding after PAGE purification,^{8a-c, 24} in this case, fully glycosylated 15F1 and 15F3 exhibited minimal binding signal after PAGE purification (Figure 4a). We hypothesized that the active 15F1 and 15F3 aptamers might be in kinetically trapped folding states after transcription and/or glycosylation, which convert to thermodynamically favored but inactive states upon denaturing PAGE purification and refolding.

Figure 4.

Filter binding of gel-purified 15F3 and stabilized stem variant. a) Filter binding curves for 15F3 crude, heat refolded and gel purified, and 15F3-M1-Mod gel purified. Data were obtained in duplicate. b) Long-stem mutant of 15F3-M1 predicted secondary structure. X = (A)₁₀ spacer, Y = reverse primer (see SI Table 1).

In the case of 15F3, further inspection of the predicted fold (Figure 3d) suggested a route to stabilization of the structure. Three of the four glycans in 15F3 reside near the end of a stem or in a GNRA tetraloop (specifically GUGA), which is a common natural RNA folding motif.^9c Although this fold has not to our knowledge been reported in 2′-fluoro-pyrimidine aptamers, the major hydrogen bonding interactions that confer stability to the tetraloop are all between the purines at the first, third and fourth positions, with base stacking contributing to stability between the 2-4 bases in the loop in some variations.²⁵ Therefore, the 2′-fluoro uridine could reasonably be accommodated in the canonical GNRA structure (SI Figure 13). However, following attachment of the bulky Man₉ moieties to the loop and two stem positions, the structure might be kinetically stable in selection, but not retain enough thermodynamic stability to refold to the same structure after denaturing urea-gel purification. Based on mFold²⁶ predictions, another possible fold with similar thermodynamic stability contains a completely different stem/loop not involving any of the glycosylation sites (SI Figure 14).

To increase the stability of this potentially important stem tetraloop motif, we extended the stem (15F3-M1-Mod, Figure 4b, SI Table 5 and SI Figure 15). This modification necessitated the loss of the “non-essential” glycosylation site (see 15F3-M1). This long-stem variant was transcribed, click glycosylated, gel purified, then radiolabeled and refolded. In the nitrocellulose filter binding assay, we now observed strong binding (K_D 36 ± 4 nM) after gel purification/refolding, and a significantly higher Fb_max (51%) compared to the parent clone, which displayed no binding at all after gel purification (Figure 4a). These data support our hypothesis that the active structure contains this stem-GNRA tetraloop, which is thermodynamically unstable in the original 15F3 aptamer.

It is worth noting that 15F3 and 15F3-M1-Mod, with K_Ds of 13 ± 2 and 36 ± 4 nM respectively, exhibit 14,000- and 5000-fold binding enhancements relative to monovalent Man₉ (K_D 180 μM). Moreover, this affinity enhancement is about 17–50-fold greater than that achieved by unstructured glycan clusters of comparable multivalency, such as the trivalent Man₉ dendrimer described by Wong.^27a

In summary, we have devised Capture SELMA, a selection method for evolving glycan-modified 2′-fluoro-RNA carbohydrate cluster ligands capable of presenting oligomannose Man₉ glycans in a cluster that leads to very tight (mid- to low- nanomolar K_D) binding for the carbohydrate-binding antibody 2G12. This selection method is applicable to the discovery of F-RNA aptamers containing other modifications, appropriate for binding to diverse targets. Further studies along these lines will be reported in due course.

ABBREVIATIONS

CBPs

carbohydrate binding proteins

HIV

Human immunodeficiency virus

CuAAC

Copper(I)-catalyzed azide/alkyne cycloaddition

K_D

equilibrium dissociation constant

PAGE

polyacrylamide gel electrophoresis