Sai Pradeep Velagapudi

Sequence-based design of bioactive small molecules that target precursor microRNAs

Oligonucleotides are designed to target RNA using base pairing rules, however, they are hampered by poor cellular delivery and non-specific stimulation of the immune system. Small molecules are preferred as lead drugs or probes, but cannot be designed from sequence. Herein, we describe an approach termed Inforna that designs lead small molecules for RNA from solely sequence. Inforna was applied to all human microRNA precursors and identified bioactive small molecules that inhibit biogenesis by binding to nuclease processing sites (41% hit rate). Amongst 29 lead interactions, the most avid interaction is between a benzimidazole (1) and precursor microRNA-96. Compound 1 selectively inhibits biogenesis of microRNA-96, upregulating a protein target (FOXO1) and inducing apoptosis in cancer cells. Apoptosis is ablated when FOXO1 mRNA expression is knocked down by an siRNA, validating compound selectivity. Importantly, microRNA profiling shows that 1 only significantly effects microRNA-96 biogenesis and is more selective than an oligonucleotide.

Two-Dimensional Combinatorial Screening Identifies Specific Aminoglycoside−RNA Internal Loop Partners

Herein is described the identification of RNA internal loops that bind to derivatives of neomycin B, neamine, tobramycin, and kanamycin A. RNA loop-ligand partners were identified by a two-dimensional combinatorial screening (2DCS) platform that probes RNA and chemical spaces simultaneously. In 2DCS, an aminoglycoside library immobilized onto an agarose microarray was probed for binding to a 3 x 3 nucleotide RNA internal loop library (81,920 interactions probed in duplicate in a single experiment). RNAs that bound aminoglycosides were harvested from the array via gel excision. RNA internal loop preferences for three aminoglycosides were identified from statistical analysis of selected structures. This provides consensus RNA internal loops that bind these structures and include: loops with potential GA pairs for the neomycin derivative, loops with potential GG pairs for the tobramycin derivative, and pyrimidine-rich loops for the kanamycin A derivative. Results with the neamine derivative show that it binds a variety of loops, including loops that contain potential GA pairs that also recognize the neomycin B derivative. All studied selected internal loops are specific for the aminoglycoside that they were selected to bind. Specificity was quantified for 16 selected internal loops by studying their binding to each of the arrayed aminoglycosides. Specificities ranged from 2- to 80-fold with an average specificity of 20-fold. These studies show that 2DCS is a unique platform to probe RNA and chemical space simultaneously to identify specific RNA motif-ligand interactions.

Structure–Activity Relationships through Sequencing (StARTS) Defines Optimal and Suboptimal RNA Motif Targets for Small Molecules

In this report, we describe the development of an approach that couples computation and experiment to allow the prediction of the affinity of RNA motif-ligand partners identified via two-dimensional combinatorial screening (2DCS).[1] This method, termed Structure-Activity Relationships Through Sequencing (StARTS), uses information from the sequences of the RNA motifs selected to bind a ligand. Sequences are statistically analyzed using the RNA Privileged Space Predictor (RNA-PSP) program to determine features (for example, 5′GC steps) in the selected sequences that occur with ≥95% confidence.[2] The confidence intervals are associated with a Z-score, with a larger value corresponding to a higher confidence level. Each selected RNA motif can have multiple significant features. Therefore, the sum of the Z-scores for all features is also computed. These data are then plotted against the measured binding affinities and can be fit to an inverse first order equation, which allow prediction of the affinity of the ligand for any RNA library member. StARTS has the potential to streamline the identification and scoring of both optimal and suboptimal RNA motif-ligand partners selected via 2DCS. Such information could prove useful in developing rational methods to target RNA with small molecules. RNA represents an important target for small molecule intervention.[3] Potential targets in genomic sequence include mRNAs and non-coding RNAs such as untranslated regions in mRNAs (riboswitches or triplet repeats that cause disease), and pri- and pre-microRNAs.[4] Most of these potential targets, however, have not been exploited in part due to a fundamental lack of understanding of the types of chemical ligands that are specifically bound by RNA and the types of RNA motifs that are specifically bound by chemical ligands. A database of RNA motif-ligand partners and modular assembly strategies are being developed to fill this void.[1,5] These approaches have the potential to enable the rational design of small molecules targeting RNA. A major impediment in the development of a database of RNA motif-ligand partners via 2DCS is its annotation, including determining relative affinities of RNAs selected to bind a ligand. In an attempt to streamline the annotation of the database, we identified the RNA motifs that bind 6′-N-5-hexynoate neamine (1, Figure 1a) using a microarray-based selection method [1] and then analyzed sequencing information from the selection to estimate the binding affinities of RNA motif-ligand partners. In this selection, 1 was immobilized onto azide-functionalized agarose microarrays via a Huisgen dipolar cycloaddition reaction (Figure 1a). Arrays containing serially-diluted loadings of 1 were probed for binding to a 32P-labeled 3×3 nucleotide internal loop library (2, Figure 1c) in the presence of excess competitor oligonucleotides (3-9, Figure 1c). The lowest loading spot of 1 that gave sufficient signal above background was excised from the microarray surface (Figure 1b). After amplification of this population of bound RNAs using biotinylated primers, the products were subjected to a modified ligation-based protocol to increase the information density in sequencing reactions.[6] Figure 1 Schematic of the microarray-based selection protocol used to identify RNA motif-1 partners and the oligonucleotides used in this study. a) anchoring 1 onto azide-functionalized agarose microarrays. b) image of microarray displaying 1 after hybridization ... The randomized regions from selected members of 2 were extracted from the sequencing data and statistically analyzed via RNA-PSP.[2] The output of RNA-PSP is Z-scores for each statistically significant feature in the selected RNAs. Figure 2 illustrates the output of the RNA-PSP analysis for one selected internal loop. Two of the most statistically significant features in the selected internal loops were 5′NAC/3′CGN and 5′NAN/3′CGC (Figure 2). It was then attempted to correlate the statistical analysis with the affinity of the selected RNA motif-1 partners. To do this, the affinities of 15 selected loops were determined as previously described using fluorescein-labeled 1 (1-Fl).[1] The loops and the measured affinities are shown in Figure 3. In addition to studying selected loops, binding affinity was also measured to the entire library, 2, the hairpin cassette into which the randomized region was embedded (10), and the mixture of structures selected to bind 1. Binding analysis shows that 1-Fl does not bind to 2 or to 10 (Kds ≫3000 nM); however, the mixture of selected members binds tightly (Kd = 315 nM). Figure 2 Protocol that was used to determine the feature with the highest Z-score and the summed Z-score of all features using Nea6′AcIL1 as an example. The statistically significant features were determined by analyzing sequencing data via the RNA-PSP ... Figure 3 The secondary structures of the internal loops selected to bind 1; nucleotides shown are derived from the boxed region of 2 (Figure 1c). The top three rows correspond to sequences identified from sequencing data while the bottom row contains members of ... With these measured affinities in hand, two different methods were employed to attempt to correlate the output of RNA-PSP (Z-scores with ≥ 95% confidence) with measured binding affinities. Both methods assume that a statistically significant feature or set of features helps drive association of the RNA-ligand complex. Thus, in principle, a function could be constructed by defining a relationship between these features (Z-scores) and affinity. In the first method, the feature with the single highest Z-score for each loop was plotted as a function of the affinity of the RNA-ligand complex (Figure 4a).[7] There is no global correlation between affinity and a single Z-score. However, loops with the highest Z-score features did exhibit some of the highest affinities measured. In the second method, the sum of the Z-scores for all features displayed by a loop was computed (Figure 3), and those values were plotted as a function of affinity (Figure 4). Plots of these data showed a good correlation (R2=0.85; Figure 4b) when the data are fit to an inverse first-order equation: Figure 4 Correlating Z-scores to the affinity of the RNA motif-ligand interactions. a) plot of the highest individual Z-score feature for each loop as a function of affinity; there is no correlation between these data. b) plot of the sum of each Z-score for the ... Kd=−7.42(±3.9)×10−8+13.7(±1.6)×10−6∑Z−score Eq (1) To test the ability of the Equation (1) to predict the affinity of RNA motif-1 interactions that were not found in sequencing data, six random RNAs that are members of 2 were chosen and tested for binding to 1-Fl. These RNAs have summed Z-scores ranging from 20 to 131. The affinities were then plotted as a function of their summed Z-score that was determined from the analysis of selected structures. Data for these RNAs are shown in Figure 4 as open circles. These data correlated well with the predictive model for affinity that was derived using only 15 selected interactions. Because of the success of this analysis, we modified RNA-PSP to compute summed Z-scores for each member of 2. This new program, RNA-PSP v. 2.0,[8] allows for rapid scoring of the fitness of each library member for a ligand. In summary, by statistically analyzing features in RNAs that are selected to bind a ligand and measuring the affinities of a subset of the selected interactions, a scoring function to predict the affinity of RNA motif-ligand partners was developed. Since 2DCS [1] allows rapid probing of both chemical and RNA spaces to potentially identify large numbers of RNA motif-ligand partners, measuring the affinities of each selected interaction can be intractable. The combination of the computational and experimental approach (StARTS) described herein, however, will allow for the efficient annotation of a growing database of RNA-ligand interactions. Such studies have the potential to enable computational approaches to rationally and modularly design small molecules targeting RNAs present in genomic sequence.

Design of a small molecule against an oncogenic noncoding RNA

Significance The goal of precision medicine is to identify selective drugs that modulate disease-causing biomolecules. This slow process often involves developing a high-throughput screen to test millions of potential drugs to find a few that affect the biomolecule. Here, we describe a facile approach using a disease-causing biomolecule’s sequence to enable design of specific drugs, eliminating arduous and time-consuming screens. By using the sequence of a non–protein-coding, oncogenic RNA, we designed a drug specifically targeting the RNA’s folded structure. In cells and animals, the drug inhibits its target, killing cancer cells while leaving healthy cells unaffected. Thus, a preclinical anticancer drug candidate can be quickly designed from sequence. The design of precision, preclinical therapeutics from sequence is difficult, but advances in this area, particularly those focused on rational design, could quickly transform the sequence of disease-causing gene products into lead modalities. Herein, we describe the use of Inforna, a computational approach that enables the rational design of small molecules targeting RNA to quickly provide a potent modulator of oncogenic microRNA-96 (miR-96). We mined the secondary structure of primary microRNA-96 (pri-miR-96) hairpin precursor against a database of RNA motif–small molecule interactions, which identified modules that bound RNA motifs nearby and in the Drosha processing site. Precise linking of these modules together provided Targaprimir-96 (3), which selectively modulates miR-96 production in cancer cells and triggers apoptosis. Importantly, the compound is ineffective on healthy breast cells, and exogenous overexpression of pri-miR-96 reduced compound potency in breast cancer cells. Chemical Cross-Linking and Isolation by Pull-Down (Chem-CLIP), a small-molecule RNA target validation approach, shows that 3 directly engages pri-miR-96 in breast cancer cells. In vivo, 3 has a favorable pharmacokinetic profile and decreases tumor burden in a mouse model of triple-negative breast cancer. Thus, rational design can quickly produce precision, in vivo bioactive lead small molecules against hard-to-treat cancers by targeting oncogenic noncoding RNAs, advancing a disease-to-gene-to-drug paradigm.

Defining the RNA internal loops preferred by benzimidazole derivatives via 2D combinatorial screening and computational analysis.

RNA is an important therapeutic target; however, RNA targets are generally underexploited due to a lack of understanding of the small molecules that bind RNA and the RNA motifs that bind small molecules. Herein, we describe the identification of the RNA internal loops derived from a 4096 member 3 × 3 nucleotide loop library that are the most specific and highest affinity binders to a series of four designer, druglike benzimidazoles. These studies establish a potentially general protocol to define the highest affinity and most specific RNA motif targets for heterocyclic small molecules. Such information could be used to target functionally important RNAs in genomic sequence.

Small molecule chemical probes of microRNA function

MicroRNAs (miRNAs) are small, non-coding RNAs that control protein expression. Aberrant miRNA expression has been linked to various human diseases, and thus miRNAs have been explored as diagnostic markers and therapeutic targets. Although it is challenging to target RNA with small molecules in general, there have been successful campaigns that have identified small molecule modulators of miRNA function by targeting various pathways. For example, small molecules that modulate transcription and target nuclease processing sites in miRNA precursors have been identified. Herein, we describe challenges in developing chemical probes that target miRNAs and highlight aspects of miRNA cellular biology elucidated by using small molecule chemical probes. We expect that this area will expand dramatically in the near future as progress is made in understanding small molecule recognition of RNA.

Probing a 2-Aminobenzimidazole Library for Binding to RNA Internal Loops via Two-Dimensional Combinatorial Screening

There are many potential RNA drug targets in bacterial, viral, and human transcriptomes. However, there are few small molecules that modulate RNA function. This is due, in part, to a lack of fundamental understanding about RNA-ligand interactions including the types of small molecules that bind to RNA structural elements and the RNA structural elements that bind to small molecules. In an effort to better understand RNA-ligand interactions, we diversified the 2-aminobenzimidazole core (2AB) and probed the resulting library for binding to a library of RNA internal loops. We chose the 2AB core for these studies because it is a privileged scaffold for binding RNA based on previous reports. These studies identified that N-methyl pyrrolidine, imidazole, and propylamine diversity elements at the R1 position increase binding to internal loops; variability at the R2 position is well tolerated. The preferred RNA loop space was also determined for five ligands using a statistical approach and identified trends that lead to selective recognition.

Inforna 2.0: A Platform for the Sequence-Based Design of Small Molecules Targeting Structured RNAs

The development of small molecules that target RNA is challenging yet, if successful, could advance the development of chemical probes to study RNA function or precision therapeutics to treat RNA-mediated disease. Previously, we described Inforna, an approach that can mine motifs (secondary structures) within target RNAs, which is deduced from the RNA sequence, and compare them to a database of known RNA motif-small molecule binding partners. Output generated by Inforna includes the motif found in both the database and the desired RNA target, lead small molecules for that target, and other related meta-data. Lead small molecules can then be tested for binding and affecting cellular (dys)function. Herein, we describe Inforna 2.0, which incorporates all known RNA motif-small molecule binding partners reported in the scientific literature, a chemical similarity searching feature, and an improved user interface and is freely available via an online web server. By incorporation of interactions identified by other laboratories, the database has been doubled, containing 1936 RNA motif-small molecule interactions, including 244 unique small molecules and 1331 motifs. Interestingly, chemotype analysis of the compounds that bind RNA in the database reveals features in small molecule chemotypes that are privileged for binding. Further, this updated database expanded the number of cellular RNAs to which lead compounds can be identified.

Small Molecule Inhibition of microRNA-210 Reprograms an Oncogenic Hypoxic Circuit

A hypoxic state is critical to the metastatic and invasive characteristics of cancer. Numerous pathways play critical roles in cancer maintenance, many of which include noncoding RNAs such as microRNA (miR)-210 that regulates hypoxia inducible factors (HIFs). Herein, we describe the identification of a small molecule named Targapremir-210 that binds to the Dicer site of the miR-210 hairpin precursor. This interaction inhibits production of the mature miRNA, derepresses glycerol-3-phosphate dehydrogenase 1-like enzyme (GPD1L), a hypoxia-associated protein negatively regulated by miR-210, decreases HIF-1α, and triggers apoptosis of triple negative breast cancer cells only under hypoxic conditions. Further, Targapremir-210 inhibits tumorigenesis in a mouse xenograft model of hypoxic triple negative breast cancer. Many factors govern molecular recognition of biological targets by small molecules. For protein, chemoproteomics and activity-based protein profiling are invaluable tools to study small molecule target engagement and selectivity in cells. Such approaches are lacking for RNA, leaving a void in the understanding of its druggability. We applied Chemical Cross-Linking and Isolation by Pull Down (Chem-CLIP) to study the cellular selectivity and the on- and off-targets of Targapremir-210. Targapremir-210 selectively recognizes the miR-210 precursor and can differentially recognize RNAs in cells that have the same target motif but have different expression levels, revealing this important feature for selectively drugging RNAs for the first time. These studies show that small molecules can be rapidly designed to selectively target RNAs and affect cellular responses to environmental conditions, resulting in favorable benefits against cancer. Further, they help define rules for identifying druggable targets in the transcriptome.

Defining RNA–Small Molecule Affinity Landscapes Enables Design of a Small Molecule Inhibitor of an Oncogenic Noncoding RNA

RNA drug targets are pervasive in cells, but methods to design small molecules that target them are sparse. Herein, we report a general approach to score the affinity and selectivity of RNA motif–small molecule interactions identified via selection. Named High Throughput Structure–Activity Relationships Through Sequencing (HiT-StARTS), HiT-StARTS is statistical in nature and compares input nucleic acid sequences to selected library members that bind a ligand via high throughput sequencing. The approach allowed facile definition of the fitness landscape of hundreds of thousands of RNA motif–small molecule binding partners. These results were mined against folded RNAs in the human transcriptome and identified an avid interaction between a small molecule and the Dicer nuclease-processing site in the oncogenic microRNA (miR)-18a hairpin precursor, which is a member of the miR-17-92 cluster. Application of the small molecule, Targapremir-18a, to prostate cancer cells inhibited production of miR-18a from the cluster, de-repressed serine/threonine protein kinase 4 protein (STK4), and triggered apoptosis. Profiling the cellular targets of Targapremir-18a via Chemical Cross-Linking and Isolation by Pull Down (Chem-CLIP), a covalent small molecule–RNA cellular profiling approach, and other studies showed specific binding of the compound to the miR-18a precursor, revealing broadly applicable factors that govern small molecule drugging of noncoding RNAs.