Drug-Like Small Molecules That Inhibit Expression of the Oncogenic MicroRNA-21

Abstract

We report the discovery of drug-like small molecules that bind specifically to the precursor of the oncogenic and pro-inflammatory microRNA-21 with mid-nanomolar affinity. The small molecules target a local structure at the Dicer cleavage site and induce distinctive structural changes in the RNA, which correlate with specific inhibition of miRNA processing. Structurally conservative single nucleotide substitutions eliminate the conformational change induced by the small molecules, which is also not observed in other miRNA precursors. The most potent of these compounds reduces cellular proliferation and miR-21 levels in cancer cell lines without inhibiting kinases or classical receptors, while closely related compounds without this specific binding activity are inactive in cells. These molecules are highly ligand-efficient (MW < 330) and display specific biochemical and cellular activity by suppressing the maturation of miR-21, thereby providing an avenue toward therapeutic development in multiple diseases where miR-21 is abnormally expressed.

Graphical Abstract

INTRODUCTION

Human microRNA-21 (miR-21) is a well-known protooncogene, marker of fibrosis, and molecular link between inflammation and various diseases.^1–9 The mature miR-21 is generated by the canonical microRNA biogenesis pathway and is highly conserved. The functional mature miR21-5p sequence, located on the 5′-strand of pre-miR-21 between residues U8 and A29, is overexpressed in many diseases (Figure 1; residues numbered as per the primary sequence in miRbase). When upregulated in response to pro-inflammatory signals, the mature miR-21 post-transcriptionally silences hundreds of genes that regulate multiple biological pathways, including several tumor suppressors (e.g., PDCD4, PTEN), acting in a pleiotropic fashion.^10–12 In multiple mouse models, miR-21 plays a causal role in malignant transformation, metastatic spread, or resistance to treatment, suggesting that pharmacological inhibition of this “oncomiR” might reverse even late-stage cancer.^1,2,13,14

Figure 1.

Drug-like small molecules bind to pre-miR21. (A) The three miR-21 sequences examined in this study: short (residues G22-C52); medium (G18–56); and full-length (G8-U66, corresponding to the complete pre-miR-21 sequence). For the short RNA sequence, an A23G substitution was used to improve transcription; similarly, in the full pre-miR-21 construct, the G8-U65 pair was swapped to CG to improve transcription. (B) Chemical structures of compounds 45 and 52. (C) 1D ¹H NMR ligand-detected titrations were performed to assess the binding of the candidate compounds. Three small molecules from the series shown in Table 1, compounds 45, 49, and 52, had strong binding activity, in the low to mid-nM range. The best fit for compound 52 corresponds to a K_d of approximately 200 nM, while compounds 45 and 49 both have K_ds of approximately 600 nM.

The direct targeting of mature miRNA sequences such as miR-21 using antisense oligonucleotide chemistries has been reported in numerous studies, with encouraging results in cellular and animal models, but disappointing clinical outcomes and significant toxicities.^2,4,13,15,16 Because of these toxicities as well as well-known limitations in oligonucleotide delivery and distribution,¹⁶ there remains a compelling need to discover alternative compounds that inhibit the biogenesis pathway that generates the pathogenic mature miR-21 and thereby reduce its overexpression in disease.

Many small molecules have been reported to bind to pre-miR-21, but most have shown disappointing affinity or poor specificity. Mitoxantrone,¹⁷ for example, is a cytotoxic PAIN molecule¹⁸ that binds to many RNAs.^19,20 The NMR spectrum of its complex with pre-miR-21 is extensively broadened due to the nonspecific nature of the interaction (data not shown). Similarly, peptoids are typically too flexible to bind specifically,^21–23 while peptides selected by phage display have been too insoluble to evaluate binding.²⁴ Numerous small molecules have also been reported to regulate pre-miR-21 expression in cells²⁵ and generate cellular responses,²⁶ but the evidence for direct cellular engagement has been universally weak. Natural products discovered in enzymatic screens, for example, are unlikely to be specific,²⁷ resulting in activity that is independent of RNA binding.^28,29 Similarly, our 14-mer templated macrocyclic peptides,^30–32 when screened against pre-miR-21, had only weak inhibitory activity in biochemical assays, indicative of a superficial interaction.³³ Although we were subsequently able to improve their affinity and biochemical potency by using noncanonical side chains³² (manuscript in preparation), peptides still present numerous pharmacological challenges.

In efforts targeting other unrelated microRNAs, sub-μM affinity has been reported only for molecules^34,35 which do not satisfy the “rule of 5” criteria of successful pharmaceutical projects,^36–38 effectively rendering such compounds useful only as probes. Finally, Ribotac strategies have been proposed,^39,40 but their application to RNA remains to be demonstrated. Altogether, with the exception of splicing modifiers from Roche and Novartis,^41,42 the identification of pharmaceutically attractive small molecules which bind to miRNA precursors potently and specifically and yet retain drug-like properties remains elusive.^41,42

In the process of investigating a class of RNA-binding small molecules that bind to HIV TAR RNA (BIORXIV/2022/477126), we discovered that compounds incorporating a 2-((5-(piperazin-1-yl)pyridin-2-yl)amino)pyrido[3,4-d]-pyrimidin-4(3H)-one structure with the carbonyl functional group situated within the pyrimidine fragment (Figure 1 and Table 1) display strong affinity for pre-miR-21 and inhibit processing of the precursors by the enzymes that generate mature miR-21 in rigorous biochemical assays and in cells. These “Lipinski” molecules³⁸ are very ligand-efficient (molecular weights of about 330) and target a specific structure at the junction between the apical loop and helical stem of pre-miR-21 which is required for processing by both Drosha and Dicer. This affinity is reduced significantly by conservative single nucleotide changes in the pre-miRNA. The most potent compounds in this small series have binding activities in the mid-nM range, and simple modifications of the structure (such as shifting the position of a nitrogen within a ring) reduce binding significantly, demonstrating clear structure–activity relationships (Figure 1); compounds within this series which lack the same binding profile are inactive in cells.

Table 1.

Structures of the Compounds Investigated in the Present Work

Compound	Structure
44
45
46
49
50
51
52
208

Reducing cellular levels of miR-21 with small molecules targeting its biogenesis is expected to have therapeutic benefits in a variety of cancers^1,2,14,40 as well as fibrotic and inflammatory conditions.^4–6,43 The small molecules described herein, by residing in the favorable “Lipinski” drug-like chemical space,³⁸ hold considerable promise for addressing an unmet need in RNA-targeted pharmacology.

RESULTS

The Conformation of the Dicer Binding Site Is Construct-Dependent.

Most NMR work on pre-miR-21 has studied short RNA stem-loop models,^33,44 as shown in Figure 1 (“WT short”). In these shorter constructs, the A29 nucleotide at the 5′ Dicer cleavage site is stacked within the RNA helix, and clear nuclear Overhauser effect (NOE) interactions are observed between the amino nitrogen (NH₂) protons of A29 and the imino resonances of both G28 (11.6 ppm) and G45 (12.5 ppm). The A29 NH₂ resonances are seen at 8.9 and 10.1 ppm, consistent with the formation of a protonated A29+:G45 base pair.⁴⁴

However, the native full-length pre-miR-21 is a 59 nt hairpin, and the characteristic NOE pattern that arises from A29 in the short construct is not observed in the full-length pre-miR-21. The upfield-shifted A29 NH₂ signal is seen only in the short construct (Figure S1), demonstrating that the two sequences have different structures and dynamics at the helical-loop junction region, i.e., the Dicer cleavage site.⁴⁵ In the present study, we use either the full-length pre-miR-21 or an intermediate-length sequence (Figure 1) that recapitulates the structural and dynamic properties of the full pre-miR-21 which are missing in the shorter stem-loop model. All sequences studied are listed in Table S1.

Drug-Like Small Molecules Bind to Pre-miR-21 with Mid-nM Affinity and Induce a New Structure near the Dicer Cleavage Site.

We recently reported that the small molecule Palbociclib binds with high affinity to HIV TAR RNA and designed a variant molecule that retains binding activity but does not inhibit kinases (BIORXIV/2022/477126). Because that molecule was toxic, we synthesized additional variants to eliminate toxicity, using pre-miR-21 as a control for specificity during this elaboration. While performing these optimizations, we discovered serendipitously that a 2-((5-(piperazin-1-yl)pyridin-2-yl)amino)pyrido[3,4-d]-pyrimidin-4(3H)-one structure, with the carbonyl functional group positioned within the pyrimidine fragment, instead showed a strong binding affinity for pre-miR-21 (Figure S2), while binding to TAR was significantly decreased (Figure S3).

Based on that unexpected result, we synthesized a series of analogues (see Table 1) and used a simple NMR relaxation assay (Figures 1C and S4) to obtain their approximate binding affinities. This NMR approach has been extensively validated for proteins;⁴⁶ in order to validate the assay for use with RNA, we tested it using neomycin, a control with known affinity for TAR RNA (Figures S5 and S6), obtaining an approximate affinity (1–2 μM) essentially identical to those provided by other methods such as band shifts⁴⁷ and microscale thermophoresis (MST). We repeated the same experiment with mitoxantrone and HIV TAR (Figure S7) and obtained mid-nM affinity by the NMR assay, comparable to the 120 nM affinity measured by MST; the 3–5 fold difference in affinity can be explained by the introduction of a fluorescent label at or near the binding site, as required by the MST experiment. To further validate the method, we repeated the assay while progressively reducing the small molecule concentration from 100 to 50, 25, and finally 10 μM and obtained comparable compound 52:miR-21 affinity at each small molecule concentration (Figure S8; the titrations look very similar regardless of the small molecule concentration).

Since we are not aware of any existing assay for the measurement of small molecule affinities for pre-miR-21 and we were unable to obtain convincing data by SPR even in collaboration with Crelux, we also developed a fluorescence assay based on the incorporation of 2-amino purine (2AP) within pre-miR-21, as was done for HIV TAR.⁴⁸ We introduced 2AP in place of A29 at the Dicer cleavage site because the NMR analysis demonstrates that while this residue is not directly involved in binding (see below), it is close enough to the binding site to potentially manifest a change in fluorescence. As shown in Figure S9, we observe a decrease in 2AP fluorescence; when the data are fit using standard methods (as described in the methods section), we obtain an apparent affinity of 500 nM, within a factor of 2–3 of the NMR relaxation assay; this difference can probably be attributed to the effect of the A-2AP substitution so close to the binding site.

To eliminate the possibility that the optical properties of the small molecules (molecule 52 weakly absorbs and fluoresces in the near UV) would interfere with 2AP fluorescence through inner filtering,⁴⁹ we titrated free 2AP with compound 52 both in the presence and absence of pre-miR-21. In both control experiments, we did not observe any change in fluorescence (Figure S10). We conclude that both the NMR and the fluorescence assay converge to the same affinity value for compound 52, 200–500 nM.

As a further control, when we methylated each of the NHs of compound 52 to generate compound 208, binding to RNA was completely lost (>10 μM) in both fluorescence (Figure S9B) and NMR relaxation (Figure S11). Moreover, this trimethylated compound also failed to inhibit Dicer in a biochemical assay (see below).

Among the small number of compounds synthesized, compounds 45 and 52 showed the most promising affinities, with K_ds of approximately 600 and 200 nM, respectively (Figures 1 and S4). Most significantly, small structural modifications to these compounds, such as shifts in the position of a nitrogen, or removal or addition of a nitrogen to a six-membered ring (see Table 1), resulted in 5–10 fold losses in binding activity (compare compound 52 with compounds 45, 49, and 50, which differ only in the positioning of a nitrogen within the six-membered ring; compounds 45 and 49 are about 5-fold weaker binders, while the very similar compound 50 has approximately 10-fold weaker binding activity; cf. Figure S4). Such pronounced changes in response to fine-tuning of local chemical details are rarely observed among RNA-binding small molecules, where SAR is often shallow and relatively small changes in affinity are instead seen even in response to much more radical chemical changes.^50,51

The strong binding activity of compounds 45 and 52 is correlated with distinctive structural changes in the RNA in response to binding. First, these compounds “close the loop” of pre-miR-21 (Figure 2). Namely, the addition of compounds 45 and 52 results in the emergence of new NMR imino peaks arising from residues in the pre-miR-21 apical loop, corresponding to the formation of two GUs and an AU base pair just above the Dicer cleavage site. Conclusive confirmation that these were GU base pairs was provided by ¹⁵N-edited HSQC spectra of the complexes (Figure S12), as well as by preliminary spectral assignments based on NOE interactions which identify the pair of GU base pairs as the primary binding site for molecule 52 (see below).

Figure 2.

Compounds that bind strongly to pre-miR-21 “close” the apical loop of pre-miR-21 and stabilize a conformation of pre-miR-21 that is processed inefficiently by Dicer-TRBP. (A) Binding of compounds 45 and 52 closes the two GU wobble base pairs and a neighboring AU pair in the miR-21 apical loop. (B) Paired 1D ¹H NMR spectra showing the imino region for pre-miR-21 with (top) and without (bottom) addition of compound 45. New signals appearing for U31, G32, U43, and G44 reveal the closing of the tandem UG/GU wobble pair in the apical loop (upper spectrum). The peaks are split and broadened, indicating the presence of two nearly equally populated and structurally similar conformations. (C) Paired 1D ¹H NMR spectra for pre-miR-21 with (top) and without (bottom) the addition of compound 52. Compound 52 induces a similar but stronger response, closing the U33:A42 base pair as well, in addition to the two new GU pairs (middle spectrum). (D,E) Both compounds also select a unique conformational state of the pre-miR-21 hairpin as demonstrated by NOESY spectra recorded under the same conditions: the U27H3-A47H2 NOESY signal observed in each spectrum shows the close contact between U27 and A47 within that base pair, corresponding to the inefficiently processed A29-bulged-out conformation of the Dicer cleavage site. (F) NOESY spectrum showing direct contacts between aromatic resonances of compound 52 and the two GU base pairs of the apical loop. (G) Schematic illustration of the strong NOE contacts observed between compound 52 and the GU base pairs in the miR-21 apical loop.

A second structural consequence is the stabilization of the bulged-out conformation of A29 that corresponds to a state that Dicer processes inefficiently.⁴⁵ In free pre-miR-21, A29 occupies two comparably populated conformations. This can be clearly seen in the NOESY spectrum for the free RNA, where two distinct U27 NH and A47 H2 NOE resonances are observed, the first corresponding to stacking of the A29 base within the helix, and the second to unstacking and repositioning of the base outside of the helix.⁴⁵ Binding of compound 45 or 52 strongly enhances the downfield shifted U27 NH signal arising from the latter bulged-out A29 state, a conformation which represses pre-miR-21 processing (Figure 2D,E).⁴⁵ Since A29 is the residue that has been substituted with 2AP in the fluorescence assays, this conformational change alone could explain the 2- to 3-fold difference in affinity observed in the fluorescence assay.

The Interaction between the Small Molecules and Pre-miR-21 Is Specific.

To evaluate specificity, we tested binding of compound 52 to HIV TAR and observed a 10-fold decrease in affinity compared to pre-miR-21 (Figure S3). In contrast, two other chemically similar high-affinity TAR ligands that we have recently reported bind 10–20 times less potently to pre-miR-21 than to TAR (BIORXIV/2022/477126).

To more rigorously assess specificity, we introduced conservative mutations at or near sites in pre-miR-21 where the conformational change takes place (Table S1 and Figure 3B). The binding of compound 52 stabilizes three base pairs between loop residues U33 and A42, G32 and U43, and U31 and G44 (Figure 2). We tested the binding responses of constructs containing substitutions in each of these pairs. Remarkably, swapping the identities of residues U33 and A42 (U33 → A and A42 → U) to invert this base pair to A33:U42 reduces binding by about 5-fold (Figure 3; mutant 3). Even more dramatic reductions in affinity are observed when loop residue U43 is mutated to C43 (Figure 3, mutant 1), such that the GU base pair which forms adjacent to U33-A42 in the complex with wild-type pre-miR-21 is changed to GC (mutant 1), or when the potential U31:G44 base pair just below is also changed to CG (Figure 3, mutant 2). These are structurally conservative substitutions; the introduction of much more radical changes is common when testing for specificity, for example the complete removal of the unpaired A29 to generate a perfect A-form helix that is refractory to binding of any small molecule.

Figure 3.

Ligand-detected NMR titration of compound 52 with wild-type pre-miR-21 and three mutant RNAs, and secondary structures of the four RNAs studied. Binding of compound 52 to pre-miR-21 is sensitive to conservative single- or two-nucleotide substitutions. (A) Changing base pairs from GU to GC (mutants 1 and 2), or inversion of an AU base pair to UA (mutant 3), leads to 5- to 10-fold reductions in affinity compared to wild-type pre-miR-21, as demonstrated in ligand-detected NMR titration experiments; (B) RNA sequences of the three mutants studied, with locations of base substitutions.

An explanation for these relatively large effects is provided by the analysis of the NMR spectra: once the double mutation is introduced to invert the U33:A42 base pair induced in the wild-type pre-miR-21 by compound 52, the closed-loop conformation is no longer observed (Figure S13).

To further test for specificity, we challenged several other pre-microRNAs (Table S2) with molecules 45 or 52. With pre-miR-24, pre-miR125b, and pre-miR10b, we observe only very small changes in the NMR spectra upon the addition of stoichiometric amounts of 45 or 52, further demonstrating that the interaction is specific for pre-miR-21 (Figure S14). This is particularly noteworthy for pre-miR-10b, which shares the same UGU element recognized not only by the DGCR8 component of the microprocessor complex⁵² but also by our molecules on pre-miR-21. Notably, we do not observe stabilization of the base pairs in pre-miR-10b, and very few intermolecular NOEs (Figure S15, top) are present in the NOESY spectra. For pre-miR-21, in contrast, numerous intermolecular NOEs are seen between the RNA and the small molecule (>50, Figure S15, bottom), consistent with a high affinity and site-specific interaction.

A preliminary analysis of the pre-miR-21 NOESY data (Figure 2F, and S15, bottom) identifies strong intermolecular NOE contacts between aromatic protons on the compound 52 heterocycle (two distinctive peaks at 8.5–9 ppm) and the NHs of the two consecutive GU base pairs, consistent with the formation of direct intermolecular contacts between the small molecule and the major groove edges of the GU base pairs, which are likely to stabilize the wobble pairs. Intriguingly, these CH signals originate in the 6-membered ring distal from the pyridine-piperazine, and SAR is observed when a single nitrogen is repositioned within this ring (Figure 1C and Table 1); the piperazine (which is likely protonated since its pKa is approximately 7.4) also makes NOE interactions with the RNA, suggesting that the entire 52 molecule is in direct contact with the RNA.

Small Molecules That Bind to Pre-miR-21 Have Specific Biochemical Activity.

The NMR analysis demonstrates that molecules 45 and 52 “close the loop” by stabilizing the GU base pairs and at the same time induce the bulged-out extra-helical conformation of the A29 base that represses miR-21 processing by Dicer.⁴⁵ Encouraged by these observations, which provide a plausible mechanism for the downregulation of processing, we evaluated their biochemical activity using both a commercial enzyme preparation and a highly active recombinant Dicer-TRBP preparation expressed in-house.

First, we titrated pre-miR-21 with increasing amounts of compound 52, or with a variant of this molecule generated by methylating the three NH functionalities of the molecule (compound 208). This second molecule does not bind to pre-miR-21 even at 10 μM concentration (Figures 4, S9 and S11) and does not inhibit pre-miR-21 processing even at concentrations of 50 μM or higher. In contrast, molecule 52 inhibits pre-miR-21 processing by Dicer with a single digit μM approximate inhibition constant (Figure 4). This result would seem to suggest a 10-fold difference between affinity (200 nM) and enzyme inhibition (low μM). However, when we re-measured the affinity of compound 52 under conditions closer to those of the enzymatic assay (which cannot be replicated for NMR), we observed high nM binding (Figures S16 and S17), fully consistent with the results of the biochemical assays. Furthermore, the mechanism of inhibition is likely to be allosteric⁴⁵ because the small molecule binds in the major groove, while Dicer performs its chemistry in the minor groove. This allostery might lead to imperfect inhibition as well.

Figure 4.

Pre-miR-21 processing by Dicer in the presence of increasing amounts of compounds 52 and 208. Inhibition of pre-miR-21 processing by Dicer is observed after the addition of compound 52 but is not observed for compound 208, a version of the same compound that has lost RNA binding through methylation of the three NH functionalities (Figures S9 and S11). Processing reactions were run for 1 h in the presence of various concentrations (1.5, 3, 6, 12.5, 25, 50, and 100 μM) of either compound 52 or compound 208 and then analyzed on a 15% sequencing gel. The leftmost lane contains size markers from a 10 bp DNA ladder.

The in-house-purified Dicer-TRBP complex processes pre-Let-7a and pre-miR-21 in minutes, similar to processing times observed within the cell (Figure S18), permitting inspection of the early (minutes) time points in the reaction, before significant product builds up. With pre-Let7a, no significant difference in activity was observed in the presence or absence of 2 μM of compound 45. In contrast, for pre-miR-21 at this same concentration, compound 45 significantly reduced processing by Dicer-TRBP relative to the no-compound control.

Pre-miR-21 Binding Compounds Have Specific Cellular Activity.

Three sets of measurements were conducted in cell lines to evaluate cellular responses after exposure to the pre-miR-21 binding compounds: (i) proliferation of transformed cell lines, (ii) levels of mature miR-21, and (iii) levels of the downstream target PDCD4. All assays were conducted in a blind format at the MD Anderson Cancer Center, without knowledge of the identities or binding activities of the compounds tested.

The anti-proliferative activity of compounds 45 and 52 was determined in two cancer cell lines with high levels of miR-21, gastric adenocarcinoma (AGS), and pancreatic (ASPC1) cells. Compound 52 and Palbociclib (a positive control as a well-known cancer inhibitory drug in both AGS⁵³ and ASPC1 cells,⁵⁴ which also contains the same pyridine-piperazine basic center as our molecules) each showed anti-proliferative activity at both concentrations tested in standard MTS assays conducted over 6 days. Cellular viability in the presence of either 52 or Palbociclib at 10 nmol/mL (Figure 5) or 5 nmol/mL (Figure S19) decreased significantly after approximately 72–96 h of culture, but no significant effect was observed for compound 45. The absence of anti-proliferative activity of compound 45 might be a consequence of its reduced affinity for pre-miR-21 (Figure 1), a suggestion supported by the observation of its reduced activity in regulating mature miR-21 levels in the ASPC1 cell lines (Figure 6).

Figure 5.

Anti-proliferative activity of compounds 45 and 52 against gastric (AGS) and pancreatic (ASPC1) cancer cell lines, measured as a function of time in a blind experimental format and compared to the breast cancer drug Palbociclib (Ibrance). Cell viability was assayed using standard MTS assays in the indicated cell lines, following the addition of each compound at 10 nmoles/mL or 10 μM; error bars show experimental uncertainty from results collected in triplicate. Palbociclib and compound 52 reduce the viability of both cell lines to a statistically significant extent (marked with asterisks) while compound 45 does not have any significant effect.

Figure 6.

Levels of pre-miR-21 (left) and mature miR-21 (right) measured in gastric adenocarcinoma (AGS) and pancreatic cancer cell lines (ASPC 1), normalized to internal controls (U6 snRNA), in the presence of 10 nmoles/1 mL (10 μM concentration) of compound 45, compound 52, or Palbociclib (Ibrance; used here as a negative control not expected to affect microRNA levels). MicroRNA levels were measured 48 h after incubation, relative to DMSO control, and measured by standard quantitative real-time polymerase chain reaction (qRT-PCR). Statistically significant changes are labeled with asterisks.

Levels of mature miR-21 and its precursor pre-miR-21 were measured in the same cancer cell lines under conditions where decreased proliferation was observed (Figure 6). Significant (30%) decreases in mature miR-21 levels were observed in both AGS and ASPC1 cells when challenged with compounds 45 or 52, while inconsistent changes were observed for the control Palbociclib, a kinase inhibitor not expected to reduce miR-21 levels (Figure 6). This level of reduction is as much as can be expected for an inhibitor of microRNA maturation because microRNAs are long-lived (2–5 days)⁵⁵ and only newly produced mature miR-21 would be affected by our compounds. Large reductions of mature miR-21 levels over short time scales (24–48 h), as reported,^17,26 are inconsistent with the inhibition of new microRNA expression, unless these compounds also activate an unknown microRNA-degradation pathway which would be primarily responsible for the reduction in mature miR-21 levels. Reductions in mature miR-21 were also accompanied by a decrease in pre-miR-21 levels, suggesting that at least part of the inhibitory block occurs prior to Dicer processing.

As shown in Figure S20, inhibition was specific to miR-21. Levels of mature miR-16⁵⁶ (a tumor suppressor miRNA), miR-28⁵⁷ (an oncogenic miRNA in intestinal cancers), and miR-182⁵⁸ (an oncogenic miRNA, whose expression is closely related to the clinicopathological features of gastric cancer) were not consistently or significantly affected by the compounds in either gastric (AGS) or pancreatic (ASPC1) cell lines, over multiple time scales (24–72 h), relative to controls where only DMSO was added.

In order to assess the molecular mechanism of inhibition of miR-21 expression, we also investigated some of the lower-affinity compounds in the series (molecules 44, 46, 50, and 51, Table 1). In both AGS and ASPC1 cell lines (Figure S21), we observe no significant changes in miR-21 levels upon the addition of these compounds, which bind to pre-miR-21 less potently than compound 52. Furthermore, levels of both U6 and U8 snRNAs are not affected at all, demonstrating that our compounds do not affect general RNA metabolism.

Finally, when levels of PDCD4 (a known target of miR-21^11,59,60) were measured, an increase in the levels of the tumor suppressor was induced by both compounds 45 and 52 after 48 h of incubation (Figure S22), showing that the decrease in mature miR-21 levels also leads to an increase in the PDCD4 downstream target of miR-21.

Taken together, these data indicate that compound 52 induces decreased cell proliferation for AGS and ASPC1 cells, and also produces RNA-specific responses in cells, decreasing levels of mature miR-21 and restoring levels of the tumor suppressor PDCD4, while leaving other microRNAs unaffected. Furthermore, inhibition of miR-21 processing by compounds 45 and 52 is specific (miR-16, -28, and -182 are not affected) and not simply a result of reduced proliferation (Palbociclib does not significantly affect miR-21 levels) or of general effects on RNA metabolism (U6 and U8 snRNAs are not affected).

Compounds 45 and 52 Show Only Minimal Kinase or Receptor Activity in Panel Screening Assays.

While the cellular results are supportive of on-target activity, it is always very difficult to rule out the possibility that inhibition of other unknown cellular receptors might be responsible for the observed responses. We conducted two experiments to investigate whether binding to protein targets could be responsible for the observed effects. First, we executed a KinomeEdge assay, measuring inhibition of the activity of approximately one hundred human kinases; this assay represents the industry standard in evaluating kinase inhibition by small molecules. Only weak inhibition (at >10 μM) of a small number of kinases was observed for both compounds 45 and 52, very strongly suggesting that the effects we observe are not due to kinase inhibition (Figure S23).

Second, we conducted a CEREP assay to measure the activation or suppression of the activity of about 30 classical drug receptors (Figure S24). Only weak activation of the cannabinoid receptor was observed, with no other significant effects in agonizing or antagonizing any other receptor. This assay is generally used to flag potential toxicities early in drug discovery programs; here, the results indicate no broad activity against classical receptors, in addition to a safe early pharmacological profile. While these experiments do not conclusively rule out effects mediated by other intracellular receptors, in conjunction with the absence of effects on other cellular RNAs and noncognate microRNAs, these results are supportive of on-target activity.

DISCUSSION AND CONCLUSIONS

We report the discovery of a small series of drug-like small molecules that satisfy all Lipinski rules of pharmacologically viable small molecules³⁸ and bind to the precursor of the prooncogenic and pro-inflammatory noncoding RNA miRNA-21 with mid-nM affinity and specificity. Approaches to inhibiting microRNAs based on peptides,³³ engineered proteins,⁶¹ or oligonucleotide chemistries, have well-known limitations with regards to bioavailability, distribution, and toxicity,¹⁶ which do not apply to drug-like small molecule chemistry. However, with the notable exception of successful splicing modifiers developed at Novartis and Roche,^41,42,62 the discovery of small molecules that bind to RNA potently and elicit an on-target cellular response, has been challenging in both Academia and the Biotech industry. Most academic reports of RNA-targeting describe small molecules with affinities in the 10s of μM range, highly charged cationic ligands, or “probe” molecules, far outside the established realm of pharmaceutically successful small molecules.³⁸

The molecules reported herein are unlike previously reported ligands for pre-miR-21 or other microRNAs, which typically had a much weaker affinity,²⁹ were natural products with unclear cellular targets,^29,63 were cytotoxic molecules lacking specific RNA-binding activity,^17,27 or were much larger than drug-like small molecules.^34,39,40 In other cases, small molecules reported to have effects on miR-21 maturation had other known cellular targets for which they had greater affinity than they do for pre-miR-21, raising the question of whether the observed cellular effects were due to RNA binding or to other well-established mechanisms.²⁶

The molecules we report bind to pre-miR-21 potently and specifically, with clear SAR (changes in activity are observed when small changes are made to the chemical structure; Figures 1, S4 and Table 1). The two most potent compounds induce specific changes in the structure and dynamics of pre-miR-21 at the Dicer cleavage site (Figure 2), pushing the RNA into an inefficiently processed state.⁴⁵ As a consequence, compound 52 reduces processing by Dicer both in vitro (Figures 4 and S18) and in cells (Figure 6) and does so without affecting other microRNAs or cellular RNAs (Figures S20 and S21). We observe concomitant decreases in levels of mature miR-21 in cell lines of gastric and pancreatic cancer origin after compound addition (Figures 5 and S19), as well as increases in the well-known target of miR-21, the tumor suppressor protein PDCD4 (Figure S22).

It is generally difficult (and requires resources beyond what is typically available in academia) to conclusively attribute cellular effects to interaction with a specific cellular target. However, in order to address potential off-target effects, we evaluated the inhibition of kinases and classical receptors, using assays which are gold standard in pharma and the biotech industry. Our two compounds do not inhibit kinases (Figure S23) and have a clean profile against classical receptors (Figure S24), indicating that they reside in a pharmacologically safe chemical space and do not target some obvious potential intracellular targets. While these assays do not obviously rule out the possibility that other intracellular targets are affected, they go further than most academic reports of anti-miR-21 activity; we are in fact very surprised that these straightforward assays are not routinely required of all academic publications on RNA-targeting small molecules.

The compounds we have identified are potent and, unlike previously reported examples,^34,39 they are very ligand-efficient. It is thus quite conceivable that their biochemical and cellular activity can also be further improved by structural biology or medicinal chemistry approaches aimed at improving their binding activity and cell penetration; the small molecular weight (330 Da) provides significant headroom to do so without compromising pharmacological potential and abandoning Lipinski space. Thus, these compounds are prime candidates for further therapeutic development aimed at the many diseases where overexpression of miR-21 plays a causative role.

METHODS

RNA Transcription.

All RNAs for NMR or biochemical experiments were prepared in-house using in vitro transcription on a large scale (typically 10 mL)^64,65 or synthesized by IDT when required for more efficient ³²P labeling. RNA transcription and purification protocols used purified DNA oligonucleotide templates (IDT) and T7 RNA polymerase. Briefly, 1 mL of 8 μM DNA (5’-CTATAGTGAGTCGTATTA-3′), corresponding to the phage T7 RNA polymerase promoter region, was annealed to 80 μL of 100 μM template sequences with 13 mM MgCl₂, heated to 95 °C for 4 min, and then allowed to cool to room temperature over 20 min. Following annealing, the mixture was incubated in transcription buffer, 8% PEG8000, and 35 mM magnesium chloride, with 5 mM of each of the four NTPs (ATP, GTP, UTP, and CTP, from Sigma) and 0.4 mg/mL T7 RNA polymerase expressed and purified in-house.

All RNA samples were purified from crude transcriptions by 20% denaturing polyacrylamide gel electrophoresis (PAGE), electroeluted, and concentrated by ethanol precipitation. The samples were re-dissolved in 12 mL of high salt wash (700 mM NaCl, 200 mM KCl, in 10 mM potassium phosphate at pH 6.5, with 10 μM EDTA to chelate any divalent ions), then concentrated using Centriprep conical concentrators (3000 kDa MWC, Millipore). The RNA was then slowly exchanged into low salt storage buffer (10 mM potassium phosphate at pH 6.5, with 10 mM NaCl and 10 μM EDTA). Prior to NMR experiments, all RNA samples were desalted using NAP-10 gravity columns, lyophilized, re-dissolved in buffer (see below), and then annealed by heating for 4 min to 90 °C followed by snap cooling at −20 °C.

Sequences for all of the RNAs used in this study are shown in Tables S1 and S2.

Small Molecule Preparation.

Sources.

Palbociclib was purchased from Selleckchem. Starting materials were purchased from Absyn Chemicals (compound 52) or Combi-blocks, PharmaBlock, and Astatech.

General Procedure, Route 1 (Figure S25).

The aniline (B, 1.05 equiv) was taken up in dry 1,4-dioxane (0.1 M) in a microwave vial with a magnetic stir bar. 2-Chloropyrimidine (A, 1.0 equiv) was added followed by K₂CO₃ (3.0 equiv) and X-Phos (0.2 equiv). The reaction mixture (RM) was sparged with nitrogen gas for 5–10 min. Finally, Pd₂(dba)₃ (0.1 equiv) was added, and the RM was sparged with nitrogen for another 5 min, after which the microwave vial was sealed, and microwave irradiated at 120 °C for 1.5 h. The RM was cooled to ambient temperature, filtered through a pad of Celite, rinsed with ethyl acetate, and the solvent was removed under reduced pressure. The crude RM was purified on silica gel using 0–80% ethyl acetate in hexanes as the eluent. Relevant pure fractions were evaporated in vacuum to give intermediate-C, which was dissolved in DCM/TFA (4:1, 0.05 M) and stirred for 1–2 h at RT. The crude reaction was concentrated and purified on high-performance liquid chromatography (HPLC) using water/acetonitrile as eluent. Relevant pure peak fractions were lyophilized to generate compound 52 and its analogues (overall yield for the two-step reactions: 20–50%).

General Procedure, Route 2 (Figure S25).

1-Boc-4-(4-aminophenyl)piperazine (1.05 equiv) and the corresponding 2-chloropyrimidine (A, 1.0 equiv), TFA (3.0 equiv) were taken in a sealed tube with n-BuOH (0.05 M). The RM was heated at 145 °C overnight (16–20 h). The RM was cooled to ambient temperature and excess TFA was quenched with triethylamine (TEA). The crude compound was purified by HPLC using water/acetonitrile as the eluent. Relevant pure peak fractions were lyophilized to give the corresponding 52 analogues (yield 35%). Mass spectra and 1D NMR spectra of compounds 45 and 52 are provided in Supplementary Figures S26–S28.

Ligand-Detected NMR Binding Assays.

Compounds were first dissolved to 1–10 mM in pure water or DMSO, depending on their solubility, then prepared to 100 μM in 490 μL in 50 mM deuterated Tris buffer at pH 6.5, containing 11.1 μM sodium 4,4-dimethyl-4-silapentane-1-sulfonate (DSA) as chemical shift reference, all dissolved in 99.99% D₂O. The nonbinding internal DSA reference allows normalization of the small molecule spectra (in the absence or presence of RNA) during ligand-detected titrations. The 9 protons on the internal reference also provide a control for ligand concentration and flag compounds that aggregate or precipitate. All ligand-detected experiments were conducted using the 1D-¹H NMR excitation sculpting water suppression scheme (Bruker sequence “zgesgp”); a free ligand reference spectrum was collected followed by titrations from stock RNA solutions up to 5 μM or 10 μM RNA concentration in the NMR tube at the end of the titration. Each experiment was collected with 16 scans, 16 k data points with a recycle delay of 1.0 s to increase throughput, requiring only 1.5 min in total per experiment, for an overall acquisition time of about 5 min, including the experimental setup.

We also generated titration curves under high salt buffer conditions that mimic the conditions prevalent in the cell. A 10 mL stock of each compound at 100 μM concentration was prepared in high salt buffer (50 mM deuterated Tris buffer at pH 6.5, containing 11.1 mM DSA, 250 mM NaCl, 50 mM KCl, and 2 mM MgCl₂). Compounds were divided into 12 1.5 mL microcentrifuge tubes at 490 μL each and titrated with 10 μL of RNA (0.5 to 250 or 500 μM). The final RNA concentrations in each tube ranged from 0.01 to 5 or 10 μM, while a tube with no added RNA was used as the control.

The titration of RNA into samples of the small molecules shown in Table 1 results in decreases in free ligand linewidth if binding occurs, with corresponding decreases in amplitude (e.g., Figures S2, S3, S5–S8, S11, S16, and S17), which are then plotted against the relevant RNA concentration (Figures 1, 3, and S4). For proteins with a compact and globular shape, these curves can be directly fit using eq 1 below, to extract binding constants.⁴⁶ This expression is also appropriate for an RNA of the size of the pre-miR-21 sequence used by us (Figure 1A),⁶⁶ whose shape approximates an ellipsoid of rotation with aspect ratio between 1.1 and 1.2 when hydration is considered. In the expressions that follow, $I_{\mathrm{B}}$ is the intensity of the bound ligand peak height, $I_{\mathrm{F}}$ is the free ligand peak height, $P_{\mathrm{t}}$ is the total RNA concentration, and $L_{\mathrm{t}}$ is the total ligand concentration. The constant $c$ is the ratio of the bound peak width ${\nu }_{\mathrm{B}}$ and the free peak width ${\nu }_{\mathrm{F}}$:

$$B=1-\frac{I_{\mathrm{B}}}{I_{\mathrm{F}}}=1-\left(\frac{1}{1+\left(\frac{{}^{c\left[P_{\mathrm{t}}\right]}}{\left[L_{\mathrm{t}}\right]+K_{\mathrm{D}}}\right)}\right)$$ (1)

$$c=\frac{v_{\mathrm{B}}}{v_{\mathrm{F}}}-1$$ (2)

For proteins, the bound peak width $\left({\nu }_{\mathrm{B}}\right)$ can be approximated by the molecular weight of the protein multiplied by a shape-related constant $\rho$ as follows:⁴⁶

$${\nu }_{\mathrm{B}}=\rho \times \mathrm{M}\mathrm{W}$$ (3)

An RNA of the size of pre-miR-21 has a diameter of about 25A, accounting for hydration, and a length of 30–35A. The use of eqs 1–3 might not be appropriate when a ligand induces large changes in structure and shape upon binding; under these circumstances, variations in the line width constant $\mathrm{c}$ might not reliably allow measurement of the bound ligand line width ${\nu }_{\mathrm{B}}$ because it would change between the free and bound ligand state. While the changes in the hydrodynamic shape are not likely to be large, given the small size of the RNA and its shape, changes in this constant would nevertheless lead to uncertainties in the binding affinities.

With these caveats, changes in the ligand peak height vs added RNA concentration provide a robust semi-quantitative estimate of affinity. Therefore, we fit the data to single-site binding curve models using eq 4, where $B_{\mathrm{m}\mathrm{a}\mathrm{x}}$ is the maximum binding capacity and represents the fully titrated or broadened ligand signal, $R_{\mathrm{t}}$ is the total RNA concentration, and NS is the slope of the nonlinear regression (nonspecific binding is assumed to be linear with respect to RNA concentration):

$$B=1-\frac{I_{\mathrm{B}}}{I_{\mathrm{F}}}=\frac{B_{\mathrm{m}\mathrm{a}\mathrm{x}}\times \left[R_{\mathrm{t}}\right]}{K_{\mathrm{D}}+\left[R_{\mathrm{t}}\right]}+\mathrm{N}\mathrm{S}\times \left[R_{\mathrm{t}}\right]$$ (4)

Fluorescence Binding Assays.

Following the approach successfully used for HIV TAR,⁴⁸ we substituted the fluorescent nucleoside analogue 2-aminopurine (2AP) for residue A29, adjacent to the small molecule binding site in miR-21, but not overlapping with it. The mir21-ap29 RNA was purchased from IDT (Coralville, IA) and used without further purification. RNA samples for fluorescence assays were freshly dissolved in 50 mM Bis-tris buffer at pH 6.5, snap-cooled by heating at 90 °C for 5 min, and then immediately transferred to ice for 10 min.

Fluorescence measurements were executed on an Edinburgh Instruments FLS1000 Luminescence Spectrometer operating in the emission scan mode. The emission wavelength was monitored from 320 to 420 nm, with the excitation wavelength set at 308 nm; both the excitation and emission bandwidths were set at 4 nm. Fluorescence intensity data were then measured with excitation set at 308 nm and emission set at 380 nm. Each experiment was performed at 50 nM concentration of mir21-ap29 RNA, which was titrated with ligand stock solution at the indicated concentrations. Samples were allowed to incubate at room temperature for 5 min before each measurement.

Since compound 52 weakly absorbs and fluoresces in the near UV, we also titrated compound 52 into a solution containing free 2AP, or a mixture of free 2AP and unlabeled pre-miR-21, to address the possibility of significant inner-filtering effects.⁴⁹ Consistent with the weak absorption properties of this molecule, we do not observe any significant changes in 2AP fluorescence.

The observed fluorescence signals were plotted as a function of added small molecule concentrations and the resulting fluorescence binding isotherms were fit using eq 5:

$$S=\text{So}+\left(\text{Ssat}-\text{So}\right)\times \frac{\left[RL\right]}{\left[R_{\text{t}}\right]}$$ (5)

where $S$ is the observed fluorescence signal, So is the fluorescence signal in the absence of any added ligand, Ssat is the signal after saturation of the binding site, $\left[L_{\mathrm{t}}\right]$ is the total concentration of the added ligand present at each titration step, and $\left[R_{\mathrm{t}}\right]$ is the total concentration of RNA in the sample. $\left[RL\right]$ is the concentration of ligand-RNA complex present at a given titration point/step and is given by eq 6:

$$\left[RL\right]=\frac{\left(K_{\text{d}}+\left[L_{\text{t}}\right]+\left[R_{\text{t}}\right]\right)-\sqrt{{\left(K_{\text{d}}+\left[L_{\text{t}}\right]+\left[R_{\text{t}}\right]\right)}^2-4\left[L_{\text{t}}\right]\left[R_{\text{t}}\right]}}{2}$$ (6)

Values for $K_{\mathrm{d}}$ were obtained by nonlinear least squares fitting of the experimental data sets in Matlab, which also used Ssat as a fitting parameter.

RNA-Detected NMR Binding Analysis.

The RNA samples were dissolved in 300–500 μL of NMR buffer (50 mM d₉ bis-Tris pH 6.5, 50 mM NaCl), heated to 95 °C for 4 min, and then snap-cooled at −20 °C prior to NMR measurements. Interactions were monitored through changes in the chemical shifts in NMR experiments, primarily 1D ¹H and 2D ¹H–¹H TOCSY or ¹H–¹H NOESY, which were used to map the binding site of the ligand on the RNA. All pulse programs used in these experiments were standard pulse sequences provided by Bruker. Experiments in H₂O were used to detect changes in exchangeable proton signals to assess the RNA secondary structure, while experiments in D₂O were used to improve spectral resolution by reducing the overlap of exchangeable proton signals.

After pre-miR-21 was fully titrated with a small molecule, initial RNA assignments for the complex were obtained by comparing TOCSY and NOESY spectra of the pre-miR-21–small molecule complex with those of free pre-miR-21.

Biochemical Assays.

RNAs.

All pre-miRNA substrates were chemically synthesized by Integrated DNA Technologies (IDT) for efficient 5′-end labeling, which was done using a modified T4 PNK reaction for the sequences listed in Tables S1 and S2. Dicer was purchased from Creative Bio-mart (45–1 Ramsey Road, Shirley, NY 11967, USA; used for assays shown in Figure 4) or prepared in-house and assembled with TRBP, as described below (used for assays shown in Figure S18).

RNA processing substrates were ³²P-radiolabeled using T4 polynucleotide kinase and gamma ³²P ATP as per the standard protocol. Unincorporated ATP was removed by either G25 spin columns or Zymo Research oligo clean-up columns, followed by purification by denaturing polyacrylamide gel (15 or 20%) electrophoresis.

The titrations presented in Figure 4 were done as follows. RNA stocks in reaction buffer without DTT were folded by heating to 95 °C for 2 min and cooling to RT. RNA was added to reaction mixtures containing either compound 52 or 208 at the indicated small molecule concentrations and incubated for 10 min at 37 °C, followed by enzyme addition and further incubation at 37 °C for 1 h. Aliquots of the completed reaction were run directly on a 15% sequencing gel before phospho-imaging. Reaction conditions were 20 mM MES pH 6.5, 25 mM NaCl, 5 mM MgCl2, 1 mM DTT, 1% glycerol, 25 nM Dicer, and approximately 250 to 500 nM RNA. Dicer concentrations were chosen to provide sufficient processing activity, while remaining under conditions of >10-fold RNA excess as observed in the cell, and as is customarily done when performing enzymatic reactions.

For the RNA processing experiments shown in Figure S18, the Dicer/TRBP enzyme complex was expressed and purified using published methods^67,68 and stored at −80 °C in 20 μL aliquots until needed. Specifically, the gene encoding human full-length Dicer was inserted into a modified pFastBac1 vector (Invitrogen), which has an N-terminal glutathione-S-transferase (GST) tag. The GST fusion proteins were expressed in HiFive insect cells following standard procedures and isolated by glutathione affinity chromatography using buffer A (20 mM Tris–HCl pH 8.0, 150 mM NaCl, 2 mM DTT). After on-column cleavage by Tobacco Etch Virus (TEV) protease overnight, the proteins were eluted, concentrated, and loaded onto a Superdex 200 10/300 GL column equilibrated in buffer A.

Sequences coding for human full-length TRBP2 were cloned into a modified pET28a vector with an N-terminal GB1 tag. The clones were transformed into Rosetta 2 cells and the transformants were grown in LB media at 37 °C until the cells reached a density corresponding to an OD reading of 0.8 at 600 nm. Protein expression was induced by the addition of 0.2 mM IPTG at 18 °C for 20 h. The cells were harvested, pelleted, and resuspended in 20 mM Tris–HCl pH 8.0, 500 mM NaCl, and 25 mM imidazole, with 5 mM β-ME. Cells were lysed by sonication on ice and then pelleted by centrifugation. The crude lysate was applied to a nickel affinity column and eluted in 20 mM Tris–HCl pH 8.0, 500 mM NaCl, and 250 mM imidazole buffer, with 5 mM β-ME. TEV protease was added to remove the His-GB1 tag at 4 °C overnight. The sample was concentrated and loaded onto a Superdex 200 10/300 GL column equilibrated in buffer A. The purified Dicer and TRBP2 proteins were mixed at a molar ratio of 1:1.3 and loaded onto a Superdex 200 10/300 GL column equilibrated in buffer A. The fractions containing the complex were pooled and concentrated to about 250 nM and then flash-frozen for further use in processing assays.

Time-dependent Dicer/TRBP assays (Figure S18) were conducted in 60 μL total volume, with 25 nM RNA and 1 nM Dicer/TRBP in 1× reaction buffer (20 mM Tris at pH 7.5, 25 mM NaCl, 5 mM MgCl₂, 1 mM DTT, and 1% glycerol). Enzyme concentrations were chosen to give sufficient processing activity, while remaining under RNA excess conditions. Assays were conducted in 96-well plates in a PCR thermocycler to maintain a constant temperature and improve reproducibility. Prior to initiating the reactions, 10 μL of ligand was added to the RNA wells and allowed to equilibrate at 4 °C for 30 min. The temperature of the thermocycler was then raised to 37 °C and samples were incubated for 15 min to reach thermal equilibrium. Reactions were initiated by adding Dicer-TRBP to the RNA substrate wells (with or without small molecule ligands). The final enzyme concentration for these experiments was ~1 nM, the final RNA concentrations were ~ 25 nM, and the final ligand concentrations during titration experiments ranged from zero to 100 μM. For the time-dependent experiments of Figure S18, the small molecule concentration was 2 μM. The reactions were stopped by removing 5 μL from the reaction well and adding it to stop buffer wells (2× RNA load buffer, 95% formamide, 18 mM EDTA, 0.025% SDS, 0.1% xylene cyanol, and 0.1% bromophenol blue) at the time points indicated. Reactions were resolved by 20% 19:1 denaturing (8 M urea) polyacrylamide gel and visualized using a Typhoon Gel imaging system (GE).

Cell-Based Experiments.

Cell Line Maintenance.

Using guidance from the Cancer Cell Line Encyclopedia (CCLE), cancer cell lines available in-house were screened to measure levels of miR-21 expression. The human-derived cell lines AGS (gastric cancer) and AsPC-1 (pancreatic cancer) were chosen for our experiments because of the very high levels of miR-21 present in each line; we hypothesized that it would be difficult to regulate miR-21 in cell lines where it is not well-expressed in the first place.

Cell lines were obtained from the American Type Culture Collection and maintained per instructions from the supplier. Namely, AGS cells were cultured in F12K medium supplemented with 10% FBS, while AsPC-1 cells were cultured in RPMI, with 10% FBS added. All cells were kept at 37 °C in a humidified atmosphere of 5% CO₂ and underwent regular mycoplasma testing. All experiments were conducted when cells were 70–80% confluent.

Proliferation Assays.

The impact of the compounds on cellular proliferation was assessed with 5000 cells in each cell line seeded in replicates of six in a 96-well plate and treated with DMSO or with compounds dissolved in DMSO. After 48 h, cells were treated with the MTS reagent and optical density was read to generate dose–response curves.

Quantitative Real-Time PCR Analysis.

RNA was isolated utilizing Trizol (Invitrogen) and the Direct-zol RNA Miniprep kit (Zymo Research). RNA concentration was assessed with a NanoDrop ND-1000 spectrophotometer (ThermoFisher Scientific). Expression levels for all microRNAs were tested in the two representative cell lines using qRT-PCR. Cells were seeded in 6-well plates 24 h before treatment, at a cell confluency of 50–60%. Each cell line was treated with 10 μM or 5 μM of the compound or with only the DMSO solvent. RNA was collected at the 48 h time point. Expression of miR21 was assessed utilizing the TaqMan microRNA assay (Applied Biosystems). The complementary DNA (cDNA) was synthesized using the TaqMan Reverse Transcription Reagents kit (Applied Biosystems) and was used with the TaqMan probes and Supermix (BioRad) for qRT-PCR analysis. U6 and U48 were used as internal controls for RNA levels. All experiments were performed in triplicate, with all samples normalized to the internal controls and relative expression levels calculated using the 2^−ΔΔCt method.

Western Blot Analysis.

Western blot analysis was conducted according to standard protocols. Briefly, proteins were collected from cells and lysed, and the Bradford assay was used to measure protein concentrations. Quantification of protein expression was conducted using the image analysis software ImageJ.

KinomeEdge and CEREP.

These experiments were conducted at the CRO Eurofin using standard protocols as reported in the company’s material; data were analyzed at Eurofin and presented in the standard graphic format generated using the company software.