R-BIND 2.0: An Updated Database of Bioactive RNA-Targeting Small Molecules and Associated RNA Secondary Structures

Abstract

Discoveries of RNA roles in cellular physiology and pathology are increasing the need for new tools that modulate the structure and function of these biomolecules, and small molecules are proving useful. In 2017, we curated the RNA-targeted BIoactive ligaNd Database (R-BIND) and discovered distinguishing physicochemical properties of RNA-targeting ligands, leading us to propose the existence of an "RNA-privileged" chemical space. Biennial updates of the database and the establishment of a website platform (rbind.chem.duke.edu) have provided new insights and tools to design small molecules based on the analyzed physicochemical and spatial properties. In this report and R-BIND 2.0 update, we refined the curation approach and ligand classification system as well as conducted analyses of RNA structure elements for the first time to identify new targeting strategies. Specifically, we curated and analyzed RNA target structural motifs to determine the properties of small molecules that may confer selectivity for distinct RNA secondary and tertiary structures. Additionally, we collected sequences of target structures and incorporated an RNA structure search algorithm into the website that outputs small molecules targeting similar motifs without a priori secondary structure knowledge. Cheminformatic analyses revealed that, despite the 50% increase in small molecule library size, the distinguishing properties of R-BIND ligands remained significantly different from that of proteins and are therefore still relevant to RNA-targeted probe discovery. Combined, we expect these novel insights and website features to enable the rational design of RNA-targeted ligands and to serve as a resource and inspiration for a variety of scientists interested in RNA targeting.

Figure 1.

Reclassification of database ligands based on molecular weight (MW) and associated new search features. (A) Histogram of articles published reporting ligands that are included in the R-BIND database. (B) Histogram of R-BIND 1.2 classified by SM and MV. (C) Histogram of R-BIND 2.0 with new classifications of SM and LM. For histograms, each bin covers 100 Da and the center of the bin is the number listed on the x-axis. For example, bin “200” covers ligands with MW between 150 and 250 Da. (D) Website image of new screening approach “modular assembly” and “multivalent: Y” under the Design and Discovery tab in Single Molecule View. (E) Website image showing the ability to search using multivalency as a criterion under the “Advanced Search”. SM, small molecule; MV, multivalent; LM, large molecule; and Y, Yes.

Figure 2.

Analysis of screening libraries and approaches utilized in R-BIND (SM) 1.2 and 2.0. (A) Number of molecules discovered by various screening approaches by R-BIND (SM) 1.2 and 2.0. (B) Number of molecules discovered by primary screen methods in R-BIND (SM) 1.2 and 2.0. (C) Hit rates of screens by screening approach, primary screen method, and library source for R-BIND (SM) 2.0. FcS, focused screen; HTS, high-throughput screen; and LO, lead optimization.

Figure 3.

Statistical analysis of cheminformatic parameters between R-BIND (SM) 1.2 and 2.0 and the updated FDA library (MW-filtered). The box encompasses 25–75% of variance, while the whiskers describe 10–90%. The mean is indicated by the + symbol, and the line designates the median value. All comparisons are performed using the Mann–Whitney U test with statistically significant differences indicated as *P < 0.05 and **P < 0.001.

Figure 4.

Principal component analysis (PCA) plots describing the variance between libraries. (A) Two-dimensional plot of principal components 1 and 2. (B) Two-dimensional plot of principal components 1 and 3. (C) Two-dimensional plot of principal components 2 and 3. (D) Three-dimensional plot of principal components 1, 2, and 3. R-BIND (SM) 2.0 includes only the new ligands between 1.2 and 2.0. Contributions of each parameter are listed in Table S9, and the loading plots for the first three principal components are shown in Figure S1.

Figure 5.

Analysis and implementation of RNA structure targeting using the R-BIND search tools. (A) Linear discriminant analysis (LDA) by RNA structure class explained by the first three principal components (PCs). (B) Two-dimensional centroid structures and associated loading plot with 20 physicochemical properties. Abbreviations are listed in Figure 3. (C) Workflow of newly implemented “RNA Structure Search” on the R-BIND website showing a theoretical input RNA containing a 3-nt bulge and 4-nt apical loop. Briefly, upon inputting a target RNA sequence with or without a dot-bracket notation, the RNAfold algorithm predicts the minimal free energy structure. The structure is inspected for bulge, internal loop, or apical loop structural elements of any size, which is then compared to the structural elements targeted by compounds in the database. The output represents all compounds in R-BIND that bind structures of the same size and motif.