Disruption of NOTCH signaling by a small molecule inhibitor of the transcription factor RBPJ

NOTCH plays a pivotal role during normal development and in congenital disorders and cancer. γ-secretase inhibitors are commonly used to probe NOTCH function, but also block processing of numerous other proteins. We discovered a new class of small molecule inhibitor that disrupts the interaction between NOTCH and RBPJ, which is the main transcriptional effector of NOTCH signaling. RBPJ Inhibitor-1 (RIN1) also blocked the functional interaction of RBPJ with SHARP, a scaffold protein that forms a transcriptional repressor complex with RBPJ in the absence of NOTCH signaling. RIN1 induced changes in gene expression that resembled siRNA silencing of RBPJ rather than inhibition at the level of NOTCH itself. Consistent with disruption of NOTCH signaling, RIN1 inhibited the proliferation of hematologic cancer cell lines and promoted skeletal muscle differentiation from C2C12 myoblasts. Thus, RIN1 inhibits RBPJ in its repressing and activating contexts, and can be exploited for chemical biology and therapeutic applications.

molecule. To increase the likelihood of selectively targeting RBPJ (as opposed to NOTCH directly, or mediators of its expression, trafficking or cleavage), we developed a primary high throughput screen (HTS) aimed at disrupting the interaction between RBPJ and an unrelated scaffold protein, SHARP, that binds to the same region of RBPJ to mediate its repressive effects 18 . Hits from this screen were triaged by secondary screens that identified inhibitors of RBPJ in the context of NOTCH ICD. One molecule, termed RBPJ INhibitor-1 (RIN1), potently disrupted the functional interaction of RBPJ with NOTCH and functioned as a probe of RBPJ function in cancer and skeletal myogenesis models. RIN1 is the first inhibitor of RBPJ and can be exploited for research and possible therapeutic applications.

Identification of an RBPJ inhibitor.
To identify selective inhibitors of RBPJ, we developed a primary screen to detect inhibitors of a functional interaction between RBPJ and the scaffold protein SHARP that was followed by secondary assays to establish efficacy against NOTCH. The primary screen was adapted from a cellbased 2-hybrid assay that probed the interaction of RBPJ with a minimal RBPJ-binding fragment of SHARP (amino acid residues 2770-3127) 18 (Fig. 1a, upper schematic). To minimize unwanted genomic effects of RBPJ, we used a mutated form of RBPJ that cannot bind to DNA (see Supplemental Methods). A functional interaction between the two proteins activated the UAS-Luciferase reporter gene in stably transfected AD-293 cells. We used siRNA against RBPJ to determine that the assay had an acceptable dynamic range (Z' = 0.84) (Fig. 1b). Z' scores between 0.5 and 1.0 are considered excellent 19 .
www.nature.com/scientificreports www.nature.com/scientificreports/ and Supplementary Table 1). Using a single hit compound from a pilot screening at the Prebys Center for Drug Discovery (La Jolla, California, USA), we determined that the mean Z' for the entire screen was 0.64 ± 0.05. The hit selection cutoff was set at 4 standard deviations of the mean of all control wells across the entire screen, which corresponded to a threshold of ≥40% inhibition of UAS-Luc activity and yielded 18,047 compounds. Some of these were either duplicates or unavailable for retesting, therefore 17,086 primary positives were retested through a dose range to confirm activity (Fig. 1d). To rule out inhibition of the Luciferase reporter or other non-specific effects on the cell assay system, hits were also tested in a counter screen that was identical to the primary assay except that a fusion protein consisting of the Gal4 DNA binding domain covalently linked to the VP16 activation domain was used instead of the separate Gal4-SHARP and RBPJ-VP16 fusion proteins in the primary assay ( Fig. 1a bottom schematic). Of the primary assay hits, 530 compounds confirmed activity (≥50% inhibition at 10 µM or lower concentration) in the confirmatory assay and were considered inactive (≤25% inhibition) in the counter screen. Of these compounds, 130 showed dose responsive inhibition comprising 14 distinct chemical families plus 17 singletons.
To distinguish compounds that inhibit RBPJ (as opposed to SHARP), we designed a secondary assay to test whether the molecules would block the function of RBPJ in the context of activated NOTCH signaling. In this assay (Fig. 1e), NOTCH2 ICD was transiently expressed in AD-293 cells harboring a Hes1-Luciferase reporter construct. One compound, RIN1, inhibited Hes1-Luciferase activity with an IC 50 of 0.18 µM and E max of 82% (Fig. 1e). RIN1 also inhibited NOTCH3 ICD with similar potency and efficacy (0.19 µM and E max = 88%). We tested whether RIN1 would inhibit a RBPJ-VP16 fusion protein that, because of the Herpes Simplex Virus VP16 transactivation domain, induces transcription independently of NICD and co-activators. Again, RIN1 inhibited RBPJ-VP16-dependent Hes1-Luciferase with the same potency and efficacy (IC 50 = 0.20 µM and E max = 81% inhibition; Fig. 1f). In the same experiment, RIN1 had only a minimal effect on cell viability (Resazurin) (12% inhibition at 10 µM, Fig. 1g) or on the CMV promoter (22% reduction at 10 µM, Fig. 1h) that was used to direct transgene expression in the primary and secondary assays. Thus, these data suggest that RIN1 is a potent inhibitor of RBPJ in the contexts of NICD and SHARP signaling.
Next, we examined if RIN1 would inhibit pre-existing RBPJ or if it must be present during synthesis for inhibition to occur. If RIN1 were required during RBPJ synthesis, its inhibitory activity should be abrogated by treatment with cycloheximide, which blocks translation of nascent proteins. RIN1 (2 µM) decreased RBPJ-VP16-dependent Hes1-Luciferase activity and caused a corresponding decrease in Luciferase reporter mRNA and endogenous RBPJ target gene expression ( Fig. 1i-k) mRNAs. Cycloheximide did not abrogate the ability of RIN1 to inhibit either Luciferase or endogenous gene expression (Fig. 1j,k). Furthermore, RIN1 did not alter the abundance of RBPJ-VP16 protein (Fig. 1l,m) under these conditions. Similarly, when new synthesis was blocked by siRNA silencing in cells, RIN1 did not alter the decay kinetics of endogenous RBPJ (Fig. S1). These data suggesting that RIN1 does not alter the turnover of RBPJ. Instead, the ability to inhibit in the presence of cycloheximide suggests that RIN1 disrupts the function of pre-synthesized RBPJ.

RIN1 inhibits NOTCH-dependent tumor cell proliferation. NOTCH plays a role in the carcinogenesis
and tumor progression, including leukemia, breast and lung cancers 2 . As a bioassay of NOTCH inhibition, we assayed the effect of small molecule NOTCH inhibitors on the proliferation of two cell lines immortalized from T-cell acute lymphoblastic leukemia (T-ALL) patients (Jurkat and KOPT-K1) and in the mantle cell lymphoma (MCL) line REC-1, all of which have activating mutations in the NOTCH1 heterodimerization and/or PEST domain common in hematologic malignancies 20,21 . RIN1, the γ-secretase inhibitor DAPT, and CB-103, a recently described small molecule NOTCH inhibitor with no reported mechanism 22 , all decreased cell proliferation in the three cancer cell lines but with markedly different potencies and efficacies ( Fig. 2a-c). The tumor lines had comparable levels of RBPJ protein (Fig. 2d); therefore, the varying effects of RIN1 on tumor cell proliferation might reflect differential reliance on RBPJ-dependent versus RBPJ-independent NOTCH signaling 23 . RIN1 promotes muscle differentiation. As a second bioassay of NOTCH inhibition, we tested the function of the small molecule inhibitors on the differentiation of muscle progenitor cells into mature myofibers, which is blocked by endogenous NOTCH activation 24 . C2C12 myoblasts were induced to differentiate by www.nature.com/scientificreports www.nature.com/scientificreports/ passaging from low density, high serum culture conditions into high density, low serum conditions causing fusion of the myoblasts into multinucleated myofibers that expressed structural muscle proteins such as α-actinin and myosin heavy chain (MHC). Fusion was quantified by image analysis measuring the number of MHC + cells and the number of nuclei per cell (Fig. 3). Relative to treatment with DMSO vehicle alone, RIN1 (0.6 µM, corresponding to 3 × IC 50 ) decreased the number of MHC + cells and increased the number of nuclei per cell, indicating that it induced the formation of multinucleated myofibers. DAPT (0.6 µM, ~3 × IC 50 ) was less potent in this assay, and CB-103 (0.6 or 2.5 µM) did not affect the formation of myofibers. Thus, RIN1 was active in both the cancer cell anti-proliferation and myoblast differentiation assays.

RIN1 treatment resembles RBPJ silencing.
We expected that inhibition of NOTCH signaling at the level of RBPJ would have a different effect on gene expression than would disruption at the level of γ-secretase since RBPJ can both activate and repress NOTCH target genes 25 and γ-secretase cleaves numerous proteins in addition to NOTCH [6][7][8] . Therefore, we compared the transcriptomic effects of treating Jurkat cells with RIN1 (2 µM) to treatment with DAPT (2 µM), CB-103 (10 µM) (all compounds were at ~10x IC 50 ) and siRNA against RBPJ, which reduced Jurkat cell proliferation consistent with the effect of RIN1 (Fig. S2). To compare gene expression profiles, we analyzed transcripts that did not vary (fold change <1.4, p < 0.05) between negative control groups (DMSO vehicle for the small molecules versus control siRNA for siRNA against RBPJ) to remove from analysis any genes that were influenced by the treatment differences per se (siRNA versus small molecule). Transcripts that were differentially expressed (fold change ≥ 2, p < 0.05) are shown in the heatmap (Fig. 4a) and representative examples were confirmed by qRT-PCR (Fig. 4b-e). DAPT repressed known target genes as expected, including HES1, HEY1 and DTX. CB-103 resembled DAPT suggesting that it functions at the level of inhibiting NOTCH. In contrast, RIN1 resembled siRNA silencing of RBPJ, albeit often with a greater response, suggesting that it acts at the level of RBPJ. The differential response of NOTCH target genes to RIN1 and RBPJ siRNA compared to DAPT and CB-103 suggests that NOTCH target gene expression in Jurkat cells is simultaneously sustained by RBPJ-independent signaling and repressed by RBPJ. Figure 4f positions RIN1 in the context of NOTCH signaling and in relationship to other NOTCH pathway modulators.

Discussion
Here we describe RIN1 as the first small molecule inhibitor of RBPJ signaling. RBPJ can either activate genes by forming a complex with the NOTCH ICD when NOTCH is active, or silence an overlapping but non-identical set of genes by recruiting co-repressors in the absence of NOTCH signaling 25 . In the context of NOTCH signaling, RBPJ mediates many processes, including progenitor renewal and cell fate selection during embryogenesis and tissue homeostasis, as well as pathological processes such as tumorigenesis and pulmonary hypertension 1,2,26,27 . In addition to mediating NOTCH signaling, RBPJ can act independently, for instance to attenuate hypoxia signaling through direct physical interaction with HIF1α and 2α proteins 28 . RIN1, therefore, is a unique chemical tool to probe this biologically important protein.
RIN1 treatment induced a profile of transcript changes that was distinct from the changes induced by DAPT, including opposing effects on bonafide NOTCH target genes HES1, HES5, HEY1 and DTX (Fig. 4). Activation of some NOTCH target genes by RBPJ inhibition was not unexpected since RBPJ functions as both a repressor (in the absence of NOTCH signaling) and as an activator in response to NOTCH (Fig. 4f). RIN1 was identified based on the functional inhibition of RBPJ complexed with SHARP (primary screen) and secondarily based on inhibition of NOTCH (secondary screen), with which it forms repressing and activating complexes, respectively. Furthermore, siRNA-mediated knockdown of RBPJ induced a similar profile of transcript changes as did RIN1. Together, we conclude that RIN1 inhibits RBPJ in both its activating (NOTCH) and inhibiting (SHARP) complexes. In contrast, CB-103 induced a profile of gene expression changes that differed from that of RIN1 but resembled the changes induced by DAPT, suggesting that CB-103, whose target is unknown, might act at or near the level of the NOTCH receptor itself. Consistent with targeting RBPJ, RIN1 inhibited the activity of RBPJ-VP16 fusion protein that is a constitutively active transcriptional activator (Fig. 1f). Moreover, RIN1 was not selective for NOTCH isoform, also consistent with it targeting RBPJ. Thus, the transcriptomic and functional data suggest that RIN1 targets RBPJ itself or a closely interacting protein. Studies to evaluate the possible physical interaction with RBPJ are underway.
As examples of its utility as a chemical probe, RIN1 suppressed the proliferation of three hematologic tumor cell lines (Jurkat and KOPT-K1 T-ALL, and REC-1 MCL). Interestingly, the potencies and efficacies of RIN1 relative to DAPT and CB-103 varied across lines (Fig. 2). RIN1 and siRNA against RBPJ effectively blocked proliferation of all cells; however, DAPT modestly blocked proliferation of Jurkat cells (Fig. 2), suggesting that RBPJ plays a relatively more important role in controlling Jurkat cell proliferation than does NOTCH cleavage and release of ICD. Also, CB-103 was more effective than either DAPT or RIN1 in Jurkat and KOPT-K1, but not Rec1 lines, suggesting differential roles of the NOTCH pathway components across cancer lines with activating NOTCH mutations. RIN1 also promoted the differentiation of the C2C12 skeletal myoblasts into multinucleated myofibers  The heatmap represents changes in the levels of transcript (fold change > 2, P < 0.05) that were induced by either the small molecules or RBPJ siRNA relative to their respective controls (DMSO vehicle or control siRNA) and varied < 40% between techniques (DMSO-vehicle vs. control siRNA. (b-e) qRT-PCR analysis of RBPJ target gene expression. Data are presented as mean ± standard deviation. Incrementing symbols (* and †) denote P < 0.05, P < 0.01, P < 0.001 and P < 0.0001 respectively, using two-way ANOVA with Dunnett's post-test. *between groups indicated; † relative to respective control treatments (DMSO or control siRNA). ns, not significant. (f) Schematic showing NOTCH pathway inhibitors (red text) in relationship to signaling and summary of RIN1 effects on downstream gene expression.
www.nature.com/scientificreports www.nature.com/scientificreports/ (Fig. 3). CB-103 was ineffective in this context suggesting that its target (which is unknown) is not involved, again illustrating the variable effects of inhibiting different NOTCH pathway components.
In summary, RIN1 inhibits the functional association of RBPJ with SHARP and NOTCH ICD, thereby blocking the transcriptional repression and activation, respectively, of downstream genes. RIN1 is structurally and functionally distinct from existing small molecule NOTCH inhibitors. The only previously known selective inhibitor, IMR-1, acts at the level of NOTCH itself 5 (Fig. 4f). Other small molecule inhibitors (Fig. 4f) are γ-secretase inhibitors (GSIs), which act unselectively to block NOTCH processing, dihydropyridine (DHP) inhibitors of NOTCH trafficking 29,30 and CB-103 22 , which appears to function at the level of NOTCH (Fig. 4) and has recently entered clinical development for treatment of NOTCH-dependent cancers 31 . As the first small molecule RBPJ inhibitor, RIN1 could be exploited for chemical genetics and therapeutic applications.

HTS for small molecule RBPJ inhibitors. Screen and counter screen cell lines. The cell line AD-293
(Agilent # 240085) was used to generate a stable cell line carrying the UAS-Luciferase reporter plasmid pGL4.35 (Promega) by hygromycin selection (50 µg/ml) termed AD-293-UAS-Luc. Subsequently, this line was transfected (followed by clone selection) with the plasmids indicated below to generate the primary screen and counter screen lines. AD-293-UAS-Luc and its derivatives were grown in DMEM 4.5 g/ml glucose, sodium pyruvate, glutamine, pen/strep, 10% fetal bovine serum and selection antibiotic. Assay media was DMEM (Cellgro #17-205-CV) 4.5 g/ml glucose, glutamine, pen/strep, 5% fetal bovine serum, without phenol red.
To generate the primary screen cell line, AD-293-UAS-Luc cells were transfected with plasmid pBI-CMV (Clonetech) that contained two expression cassettes. Multiple cloning site 1 (ClaI-EcoRV) contained the Gal4 DNA binding domain (Gal4DBD) in frame with the nucleotide sequence corresponding to SHARP amino acids 2770-3127, which is a minimal fragment that retains RBPJ binding 18 . Multiple cloning site 2 (BglII-XbaI) contains the sequence of full-length human RBPJ containing 4 mutations that block binding to recognition sites in DNA 32 (to avoid any effects of RBPJ directly interacting with DNA) fused to the transactivating domain of Herpes simplex virus VP16 and a MYC epitope tag.
To generate the counter screen cell line, AD-293-UAS-Luc cells were transfected with a plasmid that contained the CMV promoter driving the GAL4 DBD fused to the VP16 transactivation domain. This plasmid was made by replacing the DsRed2 coding region (AgeI-NotI fragment) in pDsRED2-C1 (Clontech) with a bicistronic DNA fragment encoding Gal4 DBD-VP16 separated from a UAS-eGFP cassette by a stop codon and poly adenylation sequence. This sequence was derived from pBSEGVUG obtained from Scott Fraser, Caltech.
High throughput (primary) screening. HTS was performed in white tissue culture treated 1536 well plates (Corning Cat. #3727), 500 cells per well in 5 µl volume. Compounds were dispensed by either Labcyte acoustical dispenser (100,000 compound pilot screen) or pintool (1.78 MM compound screen) and incubated 17 hours at 37 °C and 5% CO 2 . Luciferase substrate was Britelite Plus (Perkin Elmer). The assay had excellent properties (Z' = 0.74 ± 0.06; S/B = 153 ± 19; hit rate of 0.16%, Supplementary Table S1) determined in a single point determination (10 µM) pilot screen of ~100,000 diverse small molecules at the Prebys Center for Drug Discovery (PCDD, La Jolla, CA) that yielded 15 structurally similar compounds after counter screening. After transfer to Sanofi-Tucson, the primary screen was performed identically against 1,780,000 compounds (10 µM) using the PCDD hit as a positive control and DMSO vehicle as negative control. The mean Z' of the entire screen was 0.64 ± 0.05 with a 1% primary hit rate. Hits were evaluated through two confirmatory screens (Supplementary  Table S1), first 4-point (10, 2.5, 0.625, and 0.156 µM) and then 8-point (20, 10, 5, 2.5, 1.25, 0.625, 0.313, 0.156 µM) in the same assay as used for primary screening, and concurrently in counter screens that measured inhibition of luciferase constitutively driven by Gal4-VP16 (same concentration ranges). Secondary screening. Hes1-luciferase reporter line. The UAS sequence in pGL4.35 was replaced by the murine Hes1 promoter (−194 to +160 relative to TSS of mouse Hes1 gene) 33 to generate the Hes1-Luciferase reporter construct that was then stably transfected into AD-293 cells to yield AD-293 Hes1-Luciferase line. This cell line was transiently transfected in 384 well plates (reverse transfection) with plasmids to direct expression of NOTCH2 ICD (pAdloxN2ICD 28 ), NOTCH3 ICD (pCDNA3.1 + Hygro N3ICD-HA 28 ) and RBPJ-VP16myc fusion protein (pCDNA3.1 RBPJ-VP16-myc 28 ). Luciferase signal was read out as for primary screening using the steadylite plus substrate (Perkin Elmer).
Secondary screen controls. The parental AD-293 cell line was transfected with plasmid pCNDA3.1-Firefly Luciferase [made by transferring the Firefly Luciferase gene from pGL3 (Sigma-Aldrich)]. For cell viability, resazurin (Sigma R7017) was added to the cells at 5 µg/ml final concentration and incubated 2 hours at 37 °C before fluorescence signal detection (Perkin Elmer Envision plate reader).
AD-293 transfections and western blot analyses. AD-293 were transfected with RBPJ siRNA (Dharmacon J-007772-06) or non-targeting control (D-001810-01) using Lipofectamine RNAiMax with the following protocol per 4 cm 2 well (12 well format): 100 µl Optimem containing 1.2 ul siRNA at 25 µM was combined with 100 µl Optimem containing 3 µl Lipofectamine and briefly vortexed. After 5 minutes incubation the mixture was added to 550 ul complete culture media containing 330,000 cells and 250 µl media 8 µM compound (or DMSO). Final siRNA concentration was 30 nM and final compound concentration was 2 µM. After 12, 24 or 48 hours of culture, cells in wells were rinsed with cold PBS and lysed with 175 µl RIPA buffer containing protease inhibitor cocktail. Protein concentration was measured with BCA protein assay (Pierce). 6 µg of total protein was loaded per well in polyacrylamide gel (Miniprotean TGX 10% gel) and transferred to PVDF membrane. Antibodies used were: anti-RPBJ (Cell Signaling 5313, rabbit) and anti-α-Tubulin (T5168, mouse). Secondary antibodies were Image analysis. DAPI and Alexa488 image Z-stacks image stacks consisting of 20 images with step size of 1 µm were obtained by structural illumination microscopy using an IC200 automated microscopy system (Vala Sciences, San Diego, CA) at 20× (0.75 N.A.). Each stack was then projected (maximum intensity projection) and analyzed to calculate the following parameters in ImageJ/Fiji. Nuclei count: A nuclear mask and count was created from the DAPI images by size thresholding (400 pixels) using the Analyze Particle function in ImageJ. MHC + object count: The MHC + object mask (to identify cells) was created as for the nuclear mask but using the Alexa488 images and a minimum size threshold of 3000 pixels and a constant signal intensity threshold for all images. This value is reported in Fig. 3b as the "Number of cells". MHC + nuclei count: MHC + object and nuclei masks were overlaid. Nuclear objects with >80% MHC signal were counted as MHC + nuclei. Average nuclei count per MHC + object: This value is calculated as the ratio MHC + nuclei/MHC + objects) and reported in Fig. 3c as "Nuclei/cell". qRT-PCR. Total RNA was extracted using Quick-RNA miniprep kit (Zymo Research) following kit protocol including a DNAse treatment step. RNA was quantified using a Nanodrop and 500 ng of total RNA was used for reverse transcription, which was performed with the Quantitect RT klt (Qiagen), which includes a genomic DNA removal step. cDNA was diluted to avoid PCR inhibition by contaminants. 0.2 µl of cDNA was used per 10 µl of qRT-PCR reaction for all genes tested except for HES5 that required 2 µl of cDNA per 20 µl qRT-PCR reaction. qRT-PCR reactions were performed using iTaq Universal SYBR Green (BioRad) in ABI 7900HT (Applied Biosystems) following manufacturers' protocols.
RNA sequencing was performed by Novogene (Illumina HiSeq) to obtain 20 million paired-end ~150 bp reads per sample in biological duplicate. FasQC (v0.11.5) and MultiQC (v1.3) were used to assess read quality. Adapter and quality trimming of reads were performed with Trimmomatic (v0.36). Reads were mapped to genome GRCh37 (hg19) using STAR (v2.5.3a) with UCSC gene annotations, and on average 19.1 million uniquely mapped reads were counted. Uniquely mapped reads were summarized at the gene level with featureCounts (v1.28.1) from the Rsubread module. Differential expression was performed with DESeq2 (v1.17.39). Genes having a fold-change of greater than 1.5 and a p-value of less than 0.05 were considered significant. Analysis filters are described in the text and legend to Fig. 4.

Statistical analyses.
Statistical analyses were performed on GraphPad Prism software using two way ANOVA with Dunnett's post-test or by an unpaired Student's T-test for calculation of P-values as indicated in the figure legends. The primary and counter screens were performed at a single dose and determination for each compound. The initial confirmatory screen was performed at 4 doses, single determination each, per compound. The second confirmatory screen was performed at 10 doses, single determination for each compound. The secondary screens and subsequent studies using RIN1 were repeated with multiple determinations each (n indicated in the figure legends) and a minimum of three times with similar results. All experiments were repeated as indicated in the figure legends.

Data Availability
The Jurkat cell RNAseq datasets are available on NCBI GEO (Accession number GSE134401). All other data generated or analyzed during this study are included in this published article (and its supplementary information files).