|Home | About | Journals | Submit | Contact Us | Français|
RNA aptamers represent an emerging class of biologics that can be easily adapted for personalized and precision medicine. Several therapeutic aptamers with desirable binding and functional properties have been developed and evaluated in preclinical studies over the past 25 years. However, for the majority of these aptamers, their clinical potential has yet to be realized. A significant hurdle to the clinical adoption of this novel class of biologicals is the limited information on their secondary and tertiary structure. Knowledge of the RNA’s structure would greatly facilitate and expedite the post-selection optimization steps required for translation, including truncation (to reduce costs of manufacturing), chemical modification (to enhance stability and improve safety) and chemical conjugation (to improve drug properties for combinatorial therapy). Here we describe a structural computational modeling methodology that when coupled to a standard functional assay, can be used to determine key sequence and structural motifs of an RNA aptamer. We applied this methodology to enable the truncation of an aptamer to prostate specific membrane antigen (PSMA) with great potential for targeted therapy that had failed previous truncation attempts. This methodology can be easily applied to optimize other aptamers with therapeutic potential.
RNA aptamers are short single-stranded oligonucleotide ligands that can bind with high affinity and specificity to target molecules. These RNA ligands are derived through a process of directed chemical evolution known as SELEX (systematic evolution of ligands by exponential enrichment), which was first described 25 years ago by Tuerk and Gold  and Ellington and Szostak . The SELEX approach has since yielded a vast number of aptamers for a wide range of applications, including in vitro diagnostics, biomarker discovery, biosensor technology and therapeutics [3–11]. The first aptamer therapy, Pegaptanib (Macugen), was approved in 2004 by the US Food and Drug Administration for the treatment of wet (neovascular) age-related macular degeneration . Today the pipeline of therapeutic aptamers extends to at least 10 candidates, which are undergoing clinical trials for ocular diseases, cancer, cardiovascular disease, and type II diabetes [7–11]. Despite these successes, the path to the clinic for aptamer biologics is slow given the costs of manufacturing of long RNA sequences and the slow post-selection optimization process, which is primarily based on trial and error.
In a recent study we used computational RNA structural and RNA/protein docking modeling to guide the truncation of the A9 prostate-specific membrane antigen (PSMA) RNA aptamer (70 nucleotides long) [12,13]. This analysis resulted in shorter versions of the PSMA A9 aptamer that retain the ability to bind to and inhibit PSMA, as well as internalize into PSMA-expressing cells, but are now amenable to large-scale chemical synthesis for therapeutic applications. One of these second-generation truncated aptamers, termed A9g, has recently been evaluated in preclinical models of prostate cancer metastasis and shown to reduce metastatic spread in mice .
In conclusion, the methodologies and assays described herein, with the exception of the NALAADase assay, can be broadly applied to determining the structure of other aptamers and are poised to expedite the optimization of aptamers for clinical applications. Aptamer-specific functional assays (e.g. binding assays, inhibition assays) can be developed and optimized for any given aptamer and used in place of the NAALADase assay. Below is a detailed description of the methods for guiding RNA aptamer truncations. We also included methods for the NAALADase assay that was used for the PSMA aptamer. This will be of interest to those who wish to truncate other aptamers to PSMA or aptamers to other protein targets with carboxypeptidase activity. A detailed description of the NAALADase assay can be also found in Dickey et al. 2016 .
The Vfold2D RNA folding computational model was used to predict the two-dimensional (2D) structure of a 70 nucleotide (nt) long RNA aptamer (A9) to prostate specific membrane antigen (PSMA) and identify key sequence/structure motifs (Fig. 1). Based on the predicted secondary structure of A9, we systematically modified the original aptamer sequence to include base dele tions/insertions/mutations that would alter the aptamer’s overall predicted secondary structure. For each modified sequence, we predicted the 2D and three-dimensional (3D) RNA structures using Vfold2D and Vfold3D modeling algorithms (see Note 1). The altered RNA sequences were then in vitro transcribed and validated experimentally to assess activity in an established functional assay (NAALADase assay). From the structure–activity relationship for the different sequences and structures, we identify the sequence and structural motifs that are essential for aptamer activity.
The overall stem-loop motif (four helices, a hairpin loop, a bulge loop and two internal loops, shown in Fig. 1) of A9g may be important to the function. To further identify the critical nucleotides and essential structural features, we performed systematic base deletion/ insertion/mutation of A9g and computed the structure for each altered sequence. The altered RNAs were in vitro transcribed as previously described by us [13,16]. The transcribed RNAs were then evaluated in the NAALADase functional assay to identify critical sequential/structural elements necessary that contribute to the inhibitory activity of A9g.
The above analyses suggest that for the A9 PSMA aptamer, the sequence identity of helix S2 is not important but the formation of the loops is required for its inhibitory function.
To gain additional insights into the interaction of the RNA aptamer with its target, we use the Vfold3D [17,23,24] (see Note 10 and 11) model to predict the aptamer 3D all-atom structures. Here we used A9g shown in Fig. 1 to illustrate how to use Vfold3D to predict the 3D structures from the given 2D structure.
The 3D structures are provided to protein–RNA docking program, MdockPP [25,26] (see Note 13 and 14), to further show/verify the key structural features discovered in the above structure–activity analysis. For example, for A9g, the sequence conservation (Uridine) at position 39 may be more important than the overall structure of the L2 loop for conferring the RNA aptamer’s inhibitory function.
A functional assay should be performed in order to confirm that the aptamer retains its activity following truncation. Here we describe a functional assay which measures PSMA enzymatic activity (NAALADase).
1A two-dimensional (2D) structure is defined by the base pairs in the structure and a three-dimensional (3D) structure is defined by the 3D coordinates of all the atoms in the RNA.
2The Vfold RNA folding model relies on a coarse-grained (virtual bond) representation of RNA structures. In the model, each nucleotide has two backbone virtual bonds P-C4-P and a base/sugar virtual bond C4-N1 for pyrimidine or C4-N9 for purine. The model gives RNA loop entropy parameters from explicit conformational sampling in the 3D space. The entropy parameters for the different motifs are precomputed and tabulated [18–22]. Vfold2D is a 2D structure-based folding model. By employing the Vfold-based loop free energy parameters and the experimental thermodynamic data for the different base stacks, Vfold2D predicts RNA 2D structures from the partition function calculations from the RNA sequence.
4RNAs often display multiple, heterogeneous conformational distributions with the formation of multiple stable and metastable structures. Vfold2D model predicts two slightly different 2D structures. The two structures differ by a single nucleotide shift, resulting in the presence/absence of the single-nucleotide B1 bulge in A9g and the resultant rearrangement of the S4 helix (See Fig. 1). Protein binding may stabilize one structure over the other. The experimental data can provide information to help us determine the relevant 2D structure. For A9g, the functional study results for the different mutants, such as the non-functionality of the B1 bulge-free mutant A9g.5 in Fig. 1, support the presence of the B1 bulge and the S4 helix in the functional A9g. Therefore, we use the A9g 2D structure shown in Fig. 1 for the 3D structure prediction. The predicted 3D structure is then used for A9g aptamer-PSMA docking.
5To maintain a T7 transcription start-site (5′GGG) and base-pairing complementarity at the 3′ end, selective base changes at the 5′ and 3′ ends may be needed.
6The experimental conditions as well as the details of other activity assays (filter binding assay, and cell binding/internalization assays) can be found in Rockey et al. 2011 .
7Nucleotide changes in a helix are made by changing A-U pairs to G-C and changing the order of the base pairs.
8For a multi-branched junction, by replacing the helix stems with the closing base pairs of the helices would convert a multi-branched junction into a hairpin loop.
9The nucleotide changes in a helix may not change the global folding, but the change of the nucleotides in loops, especially in junction regions, might lead to a significant change in the global structure.
10To predict the 3D structures for a given 2D structure, the Vfold3D model assembles motif-specific structural templates into a 3D scaffold structure, followed by the refinement of AMBER energy minimization. Currently, due to the limited structural template database, Vfold3D can only predict the 3D structures with hairpin loops, internal/bulge loops, multi-branched junctions and pseudoknots.
11Instead of generating an ensemble of 3D structures, Vfold3D determines the 3D templates according to the best match for the loop size and loop sequence similarity. Therefore, the output from the Vfold3D is deterministic.
12Currently, the energy minimization has not been automated in the Vfold3D server. We manually perform AMBER energy minimization for the Vfold3D predicted structures.
13The crystal structure of PSMA was downloaded from the Protein Data Bank (PDB id: 1z8l).
14If the crystallographic or NMR structure of the protein is not available, one may use protein structure modeling software to build a model for the protein structure.