Natural product derivative Gossypolone inhibits Musashi family of RNA-binding proteins

Human colon cancer cell lines HCT-116, HCT-116 β/W and DLD-1, are as described by Lan et al. [27] and tested for mycoplasma contamination [60] before use.

Gossypolone (Gn) was prepared as previously described [61]. (3, 4-Dimethoxyphenyl)methanimine gossypol (MP-Gr) was synthesized from gossypol [27]. The Gn and MP-Gr powder were dissolved in DMSO at 20 mM as stock solutions. L-α-phosphatidylcholine (EPC) and 1,2-distearoyl-sn-glycero-3-phosphoethanolamine-N-[methoxy(polyethylene glycol)-2000] (PEG-DSPE) were purchased from Avati Polar Lipids, Inc. (Alabama, USA) . DiR (1,1′-dioctadecyl-3,3,3′,3′-tetramethylindotricarbocyanine iodide) was purchased from Invitrogen (Carlsbad, CA).

Cell growth, MTT, colony formation, western blot analysis, Caspase-3 activation assay, RT-PCR and quantitative real-time PCR were carried out according to our previous publications [27, 62‐66]. Protein expression and purification, FP competition assay, SPR, NMR, and Wnt luciferase reporter assay were carried out as previously described [27]. The primer sequences, the primary and the secondary antibodies used were from Lan et al. [27]. Live cell imaging was carried out using EVOS FL Auto Cell Imaging System (Invitrogen, Thermo Fisher Scientific) and images were cropped and processed using ImageJ (NIH).

For all cell based studies, the DMSO concentration was 0.1% except where indicated below (for CETSA).

The AutoDock4.2.6 program [67] was used for docking calculations. The three-dimensional structure of Musashi1’s RBD1 in complex with RNA (PDB: 2RS2) was used to dock the gossypolone compound at the MSI1 RBD1 - RNA interface. A grid box of size 40*44*56 Å with 0.375 Å spacing centered around residue F23 was used for docking. A total of 200 docking runs were carried out using the Lamarckian genetic algorithm. The docked conformation with lowest energy was selected as the final predicted binding mode.

CETSA was carried out according to Molina et al. [68]. For Gn dose CETSA, the HCT-116 β/W cell lysates with different concentrations of Gn were incubated for 30 min and heated individually at 52 °C for 3 min (StepOnePlus™ Real-Time PCR System, Applied Biosystems/Life Technologies) followed by cooling for 3 min at 25 °C. The soluble fractions were analyzed by western blot. The concentration of DMSO in each sample is 3.3%. Musashi-1 antibody used for CETSA was anti-MSI1 (01–1041, Millipore, Billerica, MA). Western band intensities were measured using Image Studio Ver 4.0 (LI-COR Bioscience, Lincoln, NE), and normalized to α-Tubulin.

Gn-lip was formed using a mixture of Gn, EPC, PEG-DSPE, and cholesterol in chloroform, at a molar ratio of 30/85/6/9. The solution was dried under vacuum to form a thin film of Gn/carrier mixture, which was then dissolved in DPBS to produce Gn-encapsulated liposomes. Blank liposomes were prepared similarly without the addition of Gn. To prepare the samples for TEM image, both Gn-lip and blank liposomes were diluted in DI water, respectively. The suspensions were applied to a grid and negatively stained by 4% uranyl acetate. Images of liposomes were acquired using FEI Tecnai G2 Polara 200 kV TEM (FEI Company, OR, USA). The size distribution and zeta potential of liposomes in DI water were measured at 25̊ C using a Malvern instrument (Nano-ZS90, Malvern, UK). The size stability of Gn-lip for 3 months was investigated at 4 °C. The drug loading efficiency (DLE%) and drug loading content (DLC%) of Gn were determined using filtration method. Gn-lip solution was filtered using an ultra centrifugal filter unit (MWCO 3000 Da, Amicon®, Merck KGaA, Germany). The concentration of free drug in the filtrate was determined using a UV-vis spectrophotometer. The DLE% and DLC% of Gn were calculated as follows: DLE% = (weight of loaded Gn ÷ total weight of input Gn) × 100%; DLC% = (weight of loaded Gn ÷ total weight of Gn-lip) × 100%.

The viabilities of HCT-116 and DLD-1 cells in the presence of free Gn or Gn-lip were determined using MTT-based assay, as described previously.

NOD.CB17-Prkdcscid (SCID) mice were purchased from Harlan laboratory (Indianapolis, IN) and bred at the University of Kansas Animal Care Unit. The in vivo tumor-specific distribution of liposomes was studied using DiR, a near-infrared (NIR) fluorescent dye. DiR-loaded liposome was formed using a mixture of DiR, EPC, PEG-DSPE, and cholesterol in chloroform, at a molar ratio of 1/85/6/9. The solution was dried under vacuum to form a thin film of DiR/carrier mixture, which was then dissolved in DPBS to produce DiR-loaded liposomes. Two DLD-1 tumor-bearing SCID mice were used for in vivo fluorescence imaging according to our previous studies with modifications [69, 70]. Briefly, 10 nmol DiR-loaded liposome in 200 μL was intravenously (i.v.) injected into one mouse; 200 μL 10 nmol DiR ethanol/water (1:4 v/v) mixed solvent as the control was i.v. injected into another mouse. At different time points, the biodistributions of DiR in both mice were observed using a Carestream Molecular Imaging System (Carestream Health, Rochester, NY), with excitation at 750 nm and emission at 830 nm using an exposure time of 60 s. Mice were euthanized at 72 h post-injection by CO2 overdose and confirmed by cervical dislocation as recommended by the Panel on Euthanasia of the American Veterinary Medical Association. Organs and tumors of mice were obtained for further ex vivo fluorescence imaging. The fluorescence intensities of tumors at different time point in vivo, and tumors and livers ex vivo, were quantified using the ‘Image Math’ function of Carestream Molecular Imaging Software (Carestream Health, Inc). To produce calibration curves for DiR-lip and free DiR, 50 μL DPBS containing different amount of DiR-lip or free DiR was added in each well of a 96-well plate, followed by in vitro imaging using the same settings with that of the in vivo imaging. The calibration curves were produced using the fluorescence intensity of each well. The amount of dye in each tissue was calculated using its fluorescence intensity and the corresponding calibration curve. The fluorescence percentage of injected dose per gram (%ID/g) of each tissue was calculated using the following formula:in which M_DiR is the amount (nmol) of DiR in the tissue, ID is the injected amount (nmol) of DiR, and W_Tissue is the weight (g) of tissue.

The in vivo experiments were carried out with 5 to 6-week-old female athymic NCr-nu/nu nude mice purchased from the Harlan laboratory (Indianapolis, IN). After alcohol preparation of the skin, mice were inoculated subcutaneously with 200 μL DLD-1 cell suspension (1 × 10⁶ cells) in plain DMEM on both flanks using a sterile 23-gauge needle. When tumors reached 40 mm³ on average, the mice were randomized into 2 groups. Group 1 (10 mice, 20 tumors) was given vehicle as the control; group 2 (5 mice, 10 tumors) was given 10 mg/kg Gn-lip. Gn-lip was administrated intravenously 2 times weekly for 3.5 weeks. Tumor size and body weight of each mouse were measured twice a week, and tumor volumes were determined as a × b²/2, in which a and b represent the longest and shortest diameter of the tumors, respectively. All animal experiments were carried out according to the protocol approved by the Institutional Committee for the Use and Care of Animals of University of Kansas.

Using Prism 5.0 software (GraphPad Prism), one-way ANOVA and t-Test were used to analyze the in vitro data, two-way ANOVA was used to analyze the in vivo data. A threshold of P < 0.05 was defined as statistically significant.

In our previous screen for small molecule inhibitors of MSI1-Numb RNA binding using FP competition assay, we identified and validated (−)-gossypol as an effective inhibitor that disrupts MSI1-RNA binding [27]. We also identified gossypolone (Gn) as a potent disruptor of MSI1-Numb RNA binding, with more than 20-fold higher affinity than that of (−)-gossypol under the same experimental condition (Ki 13 ± 5 nM vs 476 ± 273 nM) [27]. Figure 1a showed that Gn dose dependently inhibits MSI1 from binding to a fluorescein labeled Numb RNA (5’-UAGGUAGUAGUUUUA-3′), with Ki of 12 nM and 62 nM to full length MSI1 (MSI1-FL) and RNA-Binding Domain 1 (RBD1) of MSI1 (MSI1-RBD1) respectively. As a control, MP-Gr, which is structurally related to Gn and (−)-gossypol, showed a Ki of larger than 200 μM (Fig. 1a top panel, [27]). Because of the conserved residues in N-terminal RRMs of MSI1 and MSI2 [43], the same Numb RNA in our MSI1 FP assay also binds to MSI2-FL and MSI2-RRM1 (data not shown). Figure 1a bottom panel showed that Gn disrupted MSI2-Numb RNA binding, with Ki of 7 nM and 37 nM to full length MSI2 and MSI2-RRM1 respectively. Our data demonstrate that Gn disrupts MSI1/MSI2-Numb RNA binding and can potentially work as a MSI1/MSI2 dual inhibitor.

To confirm the direct binding of Gn to MSI1, we carried out additional assays. First, we tested Gn in a SPR-based binding assay. In SPR, GB1-tagged MSI1-RBD1 was immobilized on the sensor chip and the level of response increases with increasing amount of material bound to the surface. As shown in Fig. 1b, at 5 μM, the response was 50, while at 10 μM, the response was 200. The SPR assay showed that Gn binds to MSI1-RBD1 in a dose-dependent manner.

The binding of Gn to MSI1-RBD1 was also confirmed using NMR (Fig. 1c). Results of NMR titrations of ¹⁵N MSI1-RBD1 with Gn showed that the RNA-binding residues (K93, F23, and W29) were primarily affected by Gn (Fig. 1c). The backbone amide peaks of K93, F23, and W29; including the side chain peak of W29, showed changes in peak positions as well as decreased peak intensities with increasing amounts of Gn, whereas most non-RNA binding residues remained unaffected. These NMR results suggested that these residues are involved in the binding of Gn to the RNA-binding pocket of MSI1-RBD1. Using computational docking, we then built a model of Gn bound to the RNA-binding pocket of MSI1-RBD1 (Fig. 1d); this model is in agreement with the NMR observation that these particular residues (K93, F23 and W29) are responsible for the interaction of MSI1-RBD1 with Gn.

To test drug-target engagement in cells, we used the CETSA to determine the thermal stability of target protein MSI1. When a protein is stabilized with addition of a ligand, the bound proteins can stay in solution whereas unbound proteins denature and precipitate with increasing temperatures [68]. The advantage of CETSA is that one can evaluate the compounds in a cellular context, thus allowing us to identify the compounds with poor bioavailability that otherwise have high affinity in biochemical assays. Fig. 1e showed the concentration-dependent target engagement of Gn with MSI1, that is, more MSI1 protein is stabilized at higher Gn concentration.

Previous studies have pointed to a tumorigenic role for MSI1, with overexpression of MSI1 leading to tumorigenesis in a mouse xenograft model [71], and decreased MSI1 leading to reduced tumor progression [3, 4, 10]. Our in vitro biophysical binding studies revealed a role of Gn in disrupting the RNA-binding ability of MSI1. We hypothesized that such disruption would lead to a de-repression of MSI1 target mRNA translation, thus decreased Notch/Wnt signaling and decreased cell growth. To investigate the effect of Gn in cells, we first assayed the overall growth of colon cancer cells with Gn treatment. As shown in Fig. 2, compared to negative controls DMSO or MP-Gr [27], 10 μM Gn treatment led to a significant decrease in cell growth in three colon cancer cell lines tested (Fig. 2a), colony formation assays also confirmed that there were fewer colonies formed with higher concentrations of Gn (Fig. 2b). Gn treatment phenocopied the cell growth assay and colony formation assay results obtained with HCT-116 β/W MSI1 CRISPR knockout clones (Fig. 2c, Additional file 1: Figure S1) and HCT-116 β/W MSI1 shRNA knock down clones (data not shown).

We next tested whether Gn treatment will induce apoptosis and/or autophagy in cells. We examined PARP cleavage and Caspase-3 activation in two colon cancer cell lines. As shown in Fig. 3a-b, at 10 μM, Gn led to increased PARP cleavage (Fig. 3a), as well as augmented Caspase-3 activation (Fig. 3b); while MP-Gr or DMSO did not show any effect, consistent with our early report [27]. These data indicate that Gn induces apoptosis in colon cancer cell lines. Using a live cell image system, we showed that 20 μM Gn induced autophagy in DLD-1 cells and led to cell death via apoptosis (Additional file 2: Video 1 and Additional file 3: Video 2). One representative view from the video in each treatment was presented in Fig. 3c. The cell treated with 20 μM Gn started to accumulate autophagosome after 14 h, and died via apoptotic cell death (Fig. 3c left panel). In contrast, cells with DMSO control proliferated and covered the whole view at the end of the time lapse (72 h) (Fig. 3c right panel). Additionally, autophagy induction was shown by the LC3 conversion and P62 degradation [72] in Gn treated samples (Fig. 3d). When we pretreated the cells with chloroquine (CQ), an autophagy inhibitor that blocks the fusion of autophagosome with lysosome and lysosomal protein degradation [73], p62 degradation was blocked (Fig. 3d). These data indicate that Gn induces efficient autophagic flux, and leads to apoptotic cell death.

As describe above, binding assays showed that Gn bound to RBD1 of MSI1 and potentially blocked MSI1-target mRNAs binding, which would presumably lead to changes in MSI1 downstream targets. To test this idea, we examined the levels of several proteins and mRNAs upon Gn treatment. We noticed an increase in P21 protein level with Gn treatment in both colon cancer cell lines tested (Fig. 4a). P21 is a direct binding target of MSI1 [21], Gn binding to MSI1 would release P21 from its translation repression. However, we saw an increase in P21 mRNA level as well (Fig. 4b), which may potentially be due to other effects of Gn not related to MSI. Gn is an active metabolite of (−)-gossypol, and we and others previously reported (−)-gossypol as a Bcl-2 inhibitor [62, 63, 74‐76]. The increase in P21 mRNA level could be due to Bcl-2 related functions of Gn. With Gn treatment, we observed decreases in other MSI1 downstream targets as well. These included c-MYC, CCND1 (CYCLIN D1) and BIRC5 (SURVIVIN), all of which are downstream of Notch/Wnt pathways. Additionally, we detected decreases in MSI1 protein and mRNA levels, such reductions are results of decreased Wnt signaling, as MSI1 is a Wnt target [19, 71]. To evaluate the Gn’s ability in inhibiting Wnt signaling, we used a TOP/FOP reporter assay. As shown in Fig. 4c, Gn dose-dependently inhibited the reporter activity. Taken together, our data indicate that Gn down-regulates Notch/Wnt signaling.

Compare with (−)-gossypol, Gn was less effective in downregulating Notch/Wnt signaling through MSI1 in cells. For example, Wnt target gene AXIN2 mRNA levels were 50% (10 μM Gn treatment) versus 20% (10 μM (−)-gossypol treatment) compared to DMSO control (set as 1) in HCT-116 cells, and 80% versus 40% in DLD-1 cells (Fig. 4b, [27]). In our biophysical assays, Gn showed a better affinity to MSI1 (Fig. 1a, [27]). We thus sought to introduce a carrier for delivering Gn in vivo.

The morphology of Gn-loaded liposomes (Gn-lip) and blank liposomes was observed using transmission electron microscopy (TEM). Both Gn-lip and blank liposomes exhibited a similar spherical shape (Fig. 5a and b on the left). Some shrinkage was also observed in larger liposomes. No obvious difference was found between the two samples. Also, both liposomes possessed similar dynamic sizes around 56 nm (55.89 ± 0.34 nm for Gn-lip and 56.03 ± 0.42 nm for blank liposomes) and Zeta-potential around zero mV (− 0.04 ± 0.06 mV for Gn-lip and 0.83 ± 0.46 mV for blank liposomes) as determined by Dynamic Light Scattering (DLS). The morphology and surface charge of liposomes were not affected by the Gn encapsulation. The DLE% and DLC% of Gn were 80.74% ± 0.77 and 13.22% ± 0.11%, respectively, respectively. After storage at 4 °C for 3 months, the particle size of Gn-lip was 62.97 ± 1.65 nm, which was near the particle size (55.89 ± 0.34 nm) of fresh sample. This result demonstrated a good size stability of Gn-lip.

The viabilities of cells in the presence of free Gn or Gn-lip were similar, for both HCT-116 and DLD-1 cells (Fig. 5c). The IC₅₀ value of free Gn and Gn-lip were 12 μM and 13 μM for DLD-1 cell, respectively, and were 8 μM and 11 μM for HCT-116 cell, respectively. Although the IC₅₀ values of Gn-lip was a little higher than that of free Gn, cytotoxicity of Gn was not significantly compromised by encapsulation. The increased IC₅₀ values of Gn-lip might be due to the sustained release of Gn from the liposomes.

The in vivo tumor-specific accumulation of the liposomes was confirmed using DLD-1 tumor-bearing SCID mice. In the mouse that was given NIR dye-loaded liposomes, DiR signal increased in the tumor regions over time and became the strongest 24 h after the injection (Fig. 6a). DiR signal was much weaker in control mouse that was given free DiR at the same time, due to non-specific distribution, fast clearance, and quenching of free DiR molecules. The ex vivo imaging results were shown in Fig. 6b, and the fluorescence %ID/g tissue is shown in Fig. 6c. Compared with in vivo imaging, ex vivo imaging does not have the masking effects of the skin and hairs on the fluorescence. Consistent with the in vivo results, quantifications using %ID/g also showed more DiR in tumors of DiR-lip group than in tumors of free DiR group. In addition, more DiR existed in the liver of free DiR group. Since liver is the main organ for Gn metabolism, this result indicates that the drug in liposomes may have a long-term effect as compared with the free drug. The results are also consistent with our previous findings, which indicated the elongated retention and protection of DiR in the body brought about by the encapsulation of liposomes [77].

The in vivo tumor suppression effect of Gn-lip was compared with vehicle control (Fig. 7a). Significant tumor growth inhibition was observed in Gn-lip group compared with the vehicle (P < 0.01). Gn-lip also showed better efficacy compared to gossypol treated group (Additional file 1: Figure S2). The mice body weight of Gn group kept stable during the whole experimental time (Fig. 7b), indicating the low systemic toxicity of Gn-lip treatment. To investigate whether Gn induces apoptosis and inhibit Notch/Wnt signaling in tumors, tumor samples were collected and processed for western blotting analysis. With Gn-lip treatment, there was an increase in PARP cleavage, indication of apoptosis (Fig. 7c, left panel). We also probed the Gn-lip treatment group for several Notch/Wnt downstream targets, as shown in Fig. 7c right panel, Gn-lip treatment led to decreases in MSI1, activated Notch, CYCLIN D1 and SURVIVIN protein levels, indication of decreased Notch/Wnt signaling in the tumor tissues. Overall, Gn loaded in the liposomes had a significant tumor inhibition efficacy and a good biocompatibility; it also induced apoptosis and inhibited Notch/Wnt signaling in tumors.