Anti-tumor effects of rigosertib in high-risk neuroblastoma

Highlights • Analysis of multiple tumor types reveal neuroblastoma sensitivity to rigosertib.• Neuroblastoma 2D and 3D organoid models are sensitive to rigosertib.• Rigosertib causes G2M cell cycle arrest rather than inhibition of Ras binding domains.• Rigosertib prolongs survival of MYCN- amplified patient-derived xenograft tumors.• Combination of rigosertib and vincristine is synergistic against neuroblastoma.


Introduction
Neuroblastoma is responsible for 15% of all pediatric oncology deaths and is the most common solid tumor in children under one year of age. High-risk neuroblastoma is associated with chromosomal alterations including 1p deletion, 11q deletion, 17q gain, MYCN amplification, and ALK-activating mutations. High-dose chemotherapy in the rapid COJEC regimen (i.e., cisplatin, etoposide, vincristine, carboplatin, cyclophosphamide), surgery, and other treatment strategies are usually effective against primary disease, but many patients relapse with metastatic disease [1][2][3]. Novel therapies for high-risk neuroblastoma are urgently required.
High-risk neuroblastoma contains significant inter-and intratumoral heterogeneity [4,5], suggesting that multi-signaling inhibition could be a promising and necessary therapeutic strategy. Rigosertib (ON-01910. Na) was originally described as a multi-kinase inhibitor with effects on PLK1 and the phosphoinositide 3-kinase (PI3K) pathway [6][7][8][9]. Rigosertib has also been proposed to function as a RAS mimetic by affecting the RAS-binding domain of RAS effector proteins, thereby inhibiting MEK-ERK signaling [10]. However, recent studies have shown that rigosertib functions as a microtubule-destabilizing agent [11], therefore its precise mechanism(s) of action remain a matter of debate [12,13]. Nevertheless, rigosertib has shown anti-tumor effects in preclinical in vivo models of solid tumors including breast, pancreatic [6], colorectal, and lung cancer [10]. Results from phase 1 clinical studies have reported treatment responses in several tumor types, including ovarian cancer [14], pancreatic ductal adenocarcinoma, thymic carcinoma [15], and head and neck squamous cell carcinoma [16].
In a recent high-throughput drug screen, we identified rigosertib as a potential anti-neuroblastoma agent that displayed high selectivity towards tumor cells compared to healthy bone marrow-derived cells [17]. Recently, Kowalczyk and co-workers also showed that rigosertib can have anti-neuroblastoma effects in conventional in vitro neuroblastoma models [18].
We previously established neuroblastoma patient-derived xenograft (PDX) models that retain the geno-and phenotypes of their parental tumors [19,20]. We also established PDX-derived 3D neuroblastoma organoids that retain the chromosomal aberrations, protein markers, as well as and tumorigenic and metastatic capacities of their parental tumors in vivo, making them suitable for preclinical drug testing [19,21].
Here we used both in silico analyses and experimental approaches to investigate neuroblastoma sensitivity to rigosertib. Analysis of drug screening data of hundreds of cancer cell lines from two major public datasets showed that neuroblastoma is one of the most sensitive tumor types to rigosertib. In vitro, rigosertib treatment disrupted PDX-derived tumor organoids, decreased cell viability, and induced cell cycle arrest and apoptotic cell death. Transcriptomic analysis (RNA-seq) indicated that rigosertib mainly induces cell cycle arrest. This was accompanied by decreased ERK1/2 (Thr202/Tyr204) and AKT (Ser473) phosphorylation. In vivo, rigosertib treatment of a high-risk MYCN-amplified PDX model delayed tumor growth and prolonged survival. We also identify drugs which, in combination with rigosertib, might improve its therapeutic efficacy.

Exploration of publicly available drug screening data
The current results are partially based upon data generated by the Cancer Target Discovery and Development (CTD 2 ) Network (https://o cg.cancer.gov/programs/ctd2/data-portal) established by the National Cancer Institute's Office of Cancer Genomics. Two publicly available datasets were used: PRISM Repurposing Secondary Screen (BRD-K55187425-236-05-2) and the CTD 2 dataset (CTRP:660,397). The datasets were accessed through DepMap Portal (https://depmap.org/ portal/). Information about the included neuroblastoma cell lines is presented in Table 1. Rigosertib area under the (dose-response) curves (AUCs) of neuroblastoma cell lines were compared with the AUCs of the other cell lines within each dataset.

Tumor organoid cultures
Neuroblastoma tumor organoids were previously established from PDX mice and cultured as free-floating 3D tumor organoids in serumfree medium with the addition of epidermal growth factor (EGF) and basic fibroblast growth factor (bFGF) as described [19,21]. Tumor organoids were regularly verified by single nucleotide polymorphism (SNP) analysis and tested for Mycoplasma. All in vitro experiments were performed in duplicate unless otherwise stated.

Cell viability assays
LU-NB-1, LU-NB-2, and LU-NB-3 PDX-derived tumor organoids were dissociated into single cells and seeded into opaque 96-well plates (Corning Inc., Corning, NY), 5000 cells per well, and treated immediately with a range (0-100 nM) of rigosertib concentrations. Cells were incubated for 72 h. Cell viability was calculated as a percentage of control wells based on CellTiter-Glo (G7571; Promega, Madison, WI) luminescence. Luminescence was measured with a Synergy2 Multi-Mode plate reader (BioTek, Winooski, VT). Biological triplicates were used.

Cell death assays
Neuroblastoma PDX cells (1 × 10 6 cells/ sample) were treated with an ~IC 90 concentration of rigosertib or control amount of DMSO. Cells were allowed to grow for 72 h, dissociated to single cells using accutase, and stained with cell death markers annexin V and propidium iodide (PI). To investigate the mechanism of cell death, a fluorescent substrate of caspase 3/7 (Nucview 405 1:5, Biotium, Fremont, CA) was used. Flow cytometry was performed with a FACSVerse flow cytometer (BD Biosciences, Franklin Lakes, NJ), and the data were analyzed using the FlowJo software.

RNA sequencing and analysis
LU-NB-2 PDX cells were seeded in T25 flasks, allowed to grow for 24 h, and treated for 24 h with rigosertib (175 nM, n = 6) or DMSO (10 µl, n = 6). Cell pellets were then collected and RNA was extracted using the AllPrep DNA/RNA Mini Kit (Qiagen, Hilden, Germany) and sequenced on a NovaSeq 6000 System (20,012,850, Illumina, San Diego, CA). Data is available at R2 (https://hgserver1.amc.nl/cgi-bin/r2/main.cgi) under: 'PDX Neuroblastoma Rigosertib-in-vitro_LUNB2_2019 -Aaltonen -12 -custom -ensh38e94'. Differentially expressed genes (ANOVA, p<0.01, FDR correction) were subjected to GO term analysis via R2, and obtained significant terms (p<0.05) were grouped into higher ranked GO terms groups with QuickGO (https://www.ebi.ac.uk/QuickGO/). Additionally, all differentially expressed genes were subjected to analysis via Gene Set Enrichment Analysis (GSEA) 4.0.3 (Broad Institute) and a collection of Hallmark gene sets from the Molecular Signatures Database was used as a reference. Normalized enrichment scores were used as an indicator of particular pathway enrichment.

Combination drug testing
Tumor organoids were dissociated into single cells and seeded into 96-well plates (20 000 cells per well in a volume of 80 μl). The cells were allowed to form organoids for 48 h and were then treated with 10 μl of Table 1 Neuroblastoma model characterization. Data acquired from DepMap portal.
-Metastasis rigosertib and 10 μl of the additional drug (cisplatin, vincristine, filanesib, or azacitidine) at varying concentrations (see matrices). The tumor organoids were incubated for 72 h and then analyzed for cell death (released proteases) and cell viability using the CytoTox-Glo (G9290; Promega, Madison, WI) assay according to the manufacturer's instructions. Cell viability matrices were used to calculate most synergistic area scores (MSA) using FIMM SynergyFinder (https://syne rgyfinder.fimm.fi) [22]. SynergyFinder compares the observed drug combination responses to the expected responses calculated using a synergy modeling method (here ZIP scores) [23]. The predicted response is compared with experimental data and the synergy score (δ) is calculated as percent of response beyond expectation.

In vivo study
In vivo studies were conducted according to the guidelines from the regional Ethics Committee for Animal Research in Lund/Malmö (ethical permit no. M11-15). LU-NB-3 PDX tumor cells (2 × 10 6 ) were suspended in medium/Matrigel (2:1) and injected subcutaneously into the flanks of female nude mice. Each mouse was allocated to either the control or treated group when the tumor reached 200-300 mm 3 . Tumor volume in mm 3 was measured with a caliper and calculated according to . Mice were treated intraperitoneally (i.p.) five times a week with PBS (n = 7) or PBS-diluted rigosertib: 200 mg/kg (n = 7). When tumors exceeded 1800 mm 3 , mice were sacrificed and tumor pieces were collected, fixed in 4% paraformaldehyde, and embedded in paraffin.

Immunohistochemistry
PDX tumor sections (4 μm) were stained manually with hematoxylin and eosin (H&E). H&E staining of all tumor sections was assessed by blinded analysis. Tumor sections (n = 5, control; n = 7, treated) were analyzed for cell death using the TUNEL Assay Kit HRP-DAB (Abcam ab206386). Three representative photos of each slide were subjected to CellProfiler 3.1.8 analysis, and the numbers of DAB positive cells were quantified [24].

Statistical analysis
Statistical analysis of the data was performed with GraphPad Prism version 8.0.1 for Windows (GraphPad Software, San Diego, CA) or Excel 2016 (Microsoft). The two-sided unpaired t-test was used to analyze differences between groups. Statistical significance for Kaplan-Meier survival analyses was calculated using the log-rank test. Data from RNA sequencing was analyzed using one-way analysis of variance (ANOVA) with a 0.01 false discovery rate (FDR) correction for multiple testing.

Multiple neuroblastoma models show high sensitivity to rigosertib
We analyzed rigosertib sensitivity in a wide range of cancer types using publicly available screening data: Cancer Target Discovery and Development (CTD 2 , CTPR 660397) [25,26], containing 650 tumor cell lines representing 23 tumor types, and PRISM Repurposing Primary Screen 19Q4 [27], containing 468 tumor cell lines representing 21 tumor types (https://depmap.org/portal/compound/rigosertib?tab=o verview). These datasets include MYCN-amplified and non-MYCN amplified neuroblastoma models representing one relapse, two primary tumors, and 10 metastases ( Table 1). Neuroblastoma had the lowest mean Area Under the Curve (AUC), thus making it the most rigosertib-sensitive cancer type across all tumor types in both datasets (Fig. 1A-B). There was no obvious difference of IC50 values between MYCN-amplified and non-MYCN amplified cell lines in the CTD 2 dataset (Fig. S1). In a recent drug screen, we assessed drug sensitivity scores for various compounds against three neuroblastoma PDX models (LU-NB-1, LU-NB-2, and LU-NB-3) relative to human healthy bone marrow cells [17]. There was a more than three-fold increase in the effect of rigosertib on neuroblastoma cells relative to the effect on bone marrow control cells (Fig. S2). Rigosertib is thus effective in a wide range of preclinical high-risk neuroblastoma models, with low effect on healthy bone marrow cells, suggesting an underlying vulnerability to rigosertib in neuroblastomas (https://depmap.org/portal/compound/rigosertib? tab=overview).

Rigosertib treatment of neuroblastoma organoids results in their disintegration and apoptosis
We next tested rigosertib in PDX-derived neuroblastoma organoids. These tumor organoids originate from high-risk tumors containing 1p loss, MYCN amplification, and 17q gain and are tumorigenic upon implantation in vivo ( Table 2) [19,21]. The average rigosertib IC 50 s for the three PDX neuroblastoma organoids LU-NB-1, LU-NB-2, and LU-NB-3 based on tumor cell viability were 48.5 nM, 39.9 nM, and 26.5 nM,  respectively (Fig. 2A). Rigosertib treatment of neuroblastoma organoids required a higher drug concentration (200 nM for 72 h) and resulted in complete disintegration of their 3D structures (Fig. 2B). Furthermore, treatment (100 nM for 72 h) caused an increased fraction of late apoptotic/necrotic cells (annexin V and PI-positive cells) compared to control DMSO treated sample (Fig. 2C). To examine the possible mechanism of cell death triggered by rigosertib, we performed flow cytometry with fluorescent caspase-3/7 substrate and observed higher levels of reagent activation in the treated samples, indicating apoptosis as an active cell death pathway (Fig. 2D).
Flow cytometric cell cycle analysis showed that rigosertib treatment of LU-NB-1 and LU-NB-2 neuroblastoma cells increased the fraction of cells in G2/M phase (Fig. 3D), indicating inhibition of mitosis. There was also an increase of cells in the sub-G1 phase, representing dead cells, particularly at 24 h (Fig. 3D).

Discussion
In a recent high-throughput drug screen, we identified rigosertib as one of the top tumor-selective drugs against neuroblastoma [17]. In the present study, we analyzed data from hundreds of cancer cell lines covering multiple tumor types and identified neuroblastoma as a tumor type sensitive to rigosertib. Treatment of PDX-derived neuroblastoma organoids disrupted the 3D structure, decreased cell viability, increased cell death, induced cell cycle arrest, and downregulated mediators of the RAS and AKT signaling pathway. In a MYCN-amplified neuroblastoma PDX model, rigosertib delayed tumor growth and prolonged mouse survival. We also identified rigosertib and vincristine as a potential drug combination. The mild side-effects of rigosertib in patients [16,29], its availability as oral formulations, and the results presented here suggest that rigosertib could form part of novel anti-neuroblastoma drug combination strategies.
Initial findings described rigosertib as a multi-kinase signaling inhibitor with effects on PLK1, PI3K pathways, and the RAF-MEK-ERK pathway [7,[30][31][32], but the precise mechanism of action of rigosertib is still being debated [11][12][13]. Our results showing decreased pAKT (Ser473) and pERK1/2 (Thr202/Tyr204) levels indicate the involvement of the AKT and RAS signaling pathways in its mechanism of action. However, our RNA-seq results are consistent with recent CRISPRi/a-based chemical genetic screens and targeted cell biological, biochemical, and structural assays [11,13] suggesting that rigosertib functions as a microtubule-destabilizing agent. Our finding that c-RAF (Ser338 phosphorylated) levels were unchanged by rigosertib might indicate that the decreases in pAKT (Ser473) and pERK1/2 (Thr202/204) were not caused by upstream blockage of RAS effectors but rather indirect effects due to other mechanisms like microtubule inhibition and cell cycle arrest. This is consistent with recent findings suggesting that rigosertib has tubulin-interacting effects rather than RAS signaling inhibitory effects in neuroblastoma [18]. It has also recently been reported that several cancer drugs in clinical trials do not function as previously thought but rather through off target-effects [33], and this cannot be ruled out for rigosertib. Clarification of the precise mechanism of action will help to design rational therapeutic combination strategies and companion biomarkers for patient identification and selection.
Using a MYCN-amplified neuroblastoma PDX model, we observed that rigosertib can delay tumor growth in vivo, accompanied by an increase in apoptotic cells. This is in contrast to a recent study by Kowalczyk et al., in which rigosertib had no in vivo effects when using the conventional cell line SK-N-AS [18]. This discrepancy might reflect the fact that neuroblastoma is very heterogenous and drug responses can vary between patients. As shown in preclinical and clinical studies, rigosertib is quickly eliminated from plasma (t1/2 ~30 min in mice, ~2 h in humans) which might contribute to its modest response as a single drug [34,35]. Development of extended release formulations could increase tumor exposure of the drug and improve anti-tumor efficacy. Our findings point to further need for therapy optimization as complete regression is rarely obtained with single agent treatment. We investigated clinically relevant drugs in combination with rigosertib and identified synergistic effects between rigosertib and vincristine, an antimitotic agent included in the rapid-COJEC regimen.
Taken together, we show that rigosertib has anti-neuroblastoma effects in multiple preclinical models. Rigosertib decreased neuroblastoma cell viability and caused cell cycle arrest and cell death through apoptosis. In vivo, treatment was well-tolerated, delayed tumor growth and increased survival of mice. Additionally, combination therapy with vincristine might improve responses and should be further investigated.
The mode of action of rigosertib is still being debated and needs to be resolved. Nevertheless, our findings suggest that rigosertib may add benefit as part of a combination treatment for MYCN-amplified high-risk neuroblastomas.

Declaration of Competing Interest
The authors declare no conflicts of interest.