The splicing factor DHX38/PRP16 is required for ovarian clear cell carcinoma tumorigenesis, as revealed by a CRISPR‐Cas9 screen

Certain cancers, such as ovarian clear cell carcinoma (OCCC), display high levels of genetic variation between patients, making it difficult to develop effective therapies. In order to identify novel genes critical to OCCC growth, we carried out a comprehensive CRISPR‐Cas9 knockout screen against cell growth using an OCCC cell line and a normal ovarian surface epithelium cell line. We identified the gene encoding DHX38/PRP16, an ATP‐dependent RNA helicase involved in splicing, as critical for the growth and tumorigenesis of OCCC. DHX38/PRP16 knockdown in OCCC cells, but not normal cells, induces apoptosis and impairs OCCC tumorigenesis in a mouse model. Our results suggest that DHX38/PRP16 may play a role in OCCC tumorigenesis and could potentially be a promising therapeutic target.

Certain cancers, such as ovarian clear cell carcinoma (OCCC), display high levels of genetic variation between patients, making it difficult to develop effective therapies. In order to identify novel genes critical to OCCC growth, we carried out a comprehensive CRISPR-Cas9 knockout screen against cell growth using an OCCC cell line and a normal ovarian surface epithelium cell line. We identified the gene encoding DHX38/PRP16, an ATP-dependent RNA helicase involved in splicing, as critical for the growth and tumorigenesis of OCCC. DHX38/PRP16 knockdown in OCCC cells, but not normal cells, induces apoptosis and impairs OCCC tumorigenesis in a mouse model. Our results suggest that DHX38/PRP16 may play a role in OCCC tumorigenesis and could potentially be a promising therapeutic target.
Epithelial ovarian cancer is divided into five major histological subtypes. Of these, the clear cell subtype [ovarian clear cell carcinoma (OCCC)] shows particularly poor prognoses, with higher stage OCCC tumors presenting with severely low 5-year survival rates of between 20% and 30% [1,2]. A major factor underlying these poor prognoses is the poor response (1-8%) OCCC tumors exhibit against conventional platinumbased therapies [3,4].
In recent years, clustered regularly interspaced short palindromic repeat (CRISPR)-Cas9 has come to prominence as the leading tool for in vitro and in vivo gene editing. By employing designer single guide RNAs (sgRNAs), double-strand breaks can be introduced almost anywhere in the genome; by taking advantage of the cell's innate repair machinery one can not only introduce mutations, but also whole expression cassettes and even large deletions. This has led to the development of a number of pooled screening approaches utilizing the CRISPR-Cas9 system to screen for genes critical to a number of cellular processes [9][10][11].
In the current study, we employed a CRISPR-Cas9based screening system to identify novel genes critical to OCCC cell growth in order to better understand the mechanisms underlying OCCC tumorigenesis and help aid the development of novel therapies against it. We show that DEAH-box helicase 38 (DHX38/PRP16), known for its role in priming the spliceosomal C complex for exon ligation through a rearrangement allowing for 3 0 -splice site docking during mRNA splicing [12][13][14][15][16], plays a critical role in the proliferation and tumorigenesis of OCCC.

Ethical statement
Mouse experiments were approved by the Ethics Committee of the Institute of Quantitative Bioscience, The University of Tokyo, and were performed according to 'the Guidelines for Proper Conduct of Animal Experiments' provided by the Science Council of Japan.

Cell culture
Cell lines and composition of culture media can be found in Table 1. All cell lines were cultured at 37°C under a 5% CO 2 atmosphere.

Growth assays
Cell growth was assayed using the Cell-Titer Glo Luminescent Cell Viability Assay (Promega, Madison WI, USA, #G7573) following the manufacturer's protocol. Plates were read on a Berthold Mithras LB 940 Plate Reader.

Viral construction
The Human CRISPR Knockout Pooled Library (GeCKO v2) [17] comprising the lentiCas9-Blast and lentiGuide-Puro two vector system was purchased from Addgene (Watertown MA, USA, #1000000049) and prepared following the manufacturer's instructions. shRNA vectors ( Table 2) were inserted into the CS-Rfa-CG backbone (provided by H. Miyoshi, RIKEN BioResource Research Center, Tsukuba, Japan). Lentiviruses were prepared via cotransfection of 1 9 10 7 HEK293FT cells per 10 cm culture plate with the appropriate vector along with the packaging plasmids psPAX2 (Addgene #12260) and pMD2.G (Addgene #12259) in a 2 : 1 : 1 ratio (15 lg plasmid DNA per plate) using 60 lL of 1 mgÁmL À1 polyethylenimine 'MAX' (Polysciences, Inc., Warrington PA, USA, #24765) in a 4 : 6 mix of Opti-MEM reduced serum medium (Gibco, Waltham MA, USA, #22600050):Dulbecco's Modified Eagle Medium (DMEM; Nissui, Tokyo, Japan, #5919). Lentiviral particles were collected via ultracentrifugation in an Optima XE-90 Ultracentrifuge at an RCF of 76,800 x g (r avg ) / 106,000 x g (r max ) 72 h post-transfection and reconstituted in PBS for at least 24 h before use in downstream experiments. CRISPR-Cas9 screening JHOC5 and ovarian surface epithelium (OSE) cells were infected with the GeCKO v2 two-vector system as described previously [18]. Briefly, JHOC5-Cas9 and OSE3-Cas9 expression cell lines were established by lentiviral transfection of lentiCas9-Blast (Addgene #52962) at an multiplicities of infection (MOI) of 0.3 followed by selection with blasticidin. These cell lines were then further transfected with a lentivirus containing the lentiGuide-Puro A library (Addgene # 1000000049) at an MOI of 0.3. Following this transfection step, cells were maintained in an appropriate concentration of puromycin. One day after infection, a sample of 2 9 10 7 infected cells (back-calculated from the MOI) was pelleted and frozen under liquid nitrogen as a Day 0 sample. Cells were then passaged until they had undergone 8 doublings.
To maintain sufficient sgRNA coverage, the total number of cells was maintained above 2 9 10 7 for the duration of the culture period. A second sample of 2 9 10 7 cells was taken as a final time point. Genomic DNA was extracted from the above-mentioned samples using a QIAGEN Blood and Cell Culture DNA Midi Kit (QIAGEN, Germantown MD, USA, #13343). 125 lg of genomic DNA from each sample was split into 2.5 lg fractions, and sgRNA sequences were amplified to form a sequencing library as described previously [18] in two steps, using the primers in Table 3. Reactions were measured for fragment size using the Agilent 2200 Tapestation and quantified using the KAPA SYBR Fast qPCR Kit (KAPA Biosystems, Wilmington MA, USA, #7959362001). Libraries were sequenced on an Illumina HiSeq2500.

Subcutaneous xenografts
After being trypsinized and washed twice, once with culture medium followed by once with PBS, 1.5 9 10 7 TOV21G cells or 1.0 9 10 6 ES2 cells transferred to a 15 mL falcon tube (VIOLAMO, AS ONE, Osaka, Japan, #VIO-15RN) were mixed with an appropriate volume of lentivirus suspension (MOI = 25 and 50), respectively, brought up to 1 mL with PBS, and allowed to incubate at 37°C for 1 h. Infected cells were plated at an appropriate density (such that cells were no > 70% confluent 2 days after plating) onto 10 cm 2 culture plates. Two days later, cells were trypsinized and washed three times with PBS. A sample of cells was reserved to confirm DHX38/PRP16 knockdown, and the rest were resuspended in 50% Matrigel at a concentration of 1.0 9 10 7 cellsÁmL À1 for TOV21G cells and 6.67 9 10 6 cellsÁmL À1 for ES2 cells. 150 lL of this suspension was injected into the flanks of female Balb/c nu/nu mice (Charles River Laboratories, Yokohama, Japan), and tumor volume was measured periodically.

Western blot
Cells were harvested and lysed using a solution containing 50 mM Tris/HCl, 0.14 mM NaCl, 1 mM EDTA, and 1% NP-40 along with 1x Halt Protease Inhibitor Cocktail (Thermo Fisher, #78438). Protein concentration was measured using the Pierce BCA Protein Assay (Thermo Fisher, #23225) as described by the manufacturer. Proteins were separated by SDS/PAGE and transferred to an Immobilion PVDF Membrane (Millipore, Burlington MA, USA, #IPVH00010) by electroblotting. All primary and secondary antibodies used can be found in Table 4. Protein bands were visualized using Luminata Forte Western HRP substrate (Millipore, #WBLUF0500).

RNA extraction and qRT-PCR
Total RNA was extracted from at least 1 9 10 5 cells and purified using TRIsure (Meridian Bioscience, Cincinnati OH, USA, #BIO-38033) following the manufacturer's protocol. Synthesis of cDNA was carried out using Prime Script RT Master Mix (TaKaRa, Kusatsu, Saga Prefecture, Japan, #RR036B) following the manufacturer's instructions. An amount of synthesized cDNA corresponding to 10 ng of original RNA was mixed with primers (Table 5) Sub-G1 assays

Data and statistical analyses
Statistical analyses, including Student's t-tests, were performed using R version 3.6.0 (http://www.r-project.org/). Plots in Figs 2-4 and 5A,B were generated in excel. Figure 1A,B was generated from the model-based analysis of genome-wide CRISPR-Cas9 knockout (MAGeCK) software. All other plots were generated in R using the ggplot2 package in conjunction with the cowplots package. The MAGeCK algorithm was carried out as published in Li et al. [19]. Briefly, sgRNAs ranked based on their abundance in cultured samples compared to controls were grouped by gene and assessed for skew to lower abundancies using the MAGeCK test command on the default settings. For detailed commands employed during the MAGeCK algorithm, as well as those used to generate of sgRNA count data, refer to GSE188485 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi? acc=GSE188485). GO statistical overrepresentation analyses were carried out using the Panther Classification System

CRISPR-Cas9 screen identifies genes critical to OCCC proliferation
In order to screen for genes critical to the growth of OCCC, we employed the pooled genome-scale CRISPR Knock-Out (GeCKO) v2 2-vector CRISPR-Cas9 knockout screening system [17] using the OCCC cell line JHOC5, as well as the T-antigen immortalized normal OSE cell line OSE3 as a negative control. After performing next-generation sequencing on the resulting amplified libraries, we used the MAGeCK robust rank aggregation algorithm to aggregate sgRNA depletion data across multiple sgRNAs targeting the same gene [19]. This allowed us to identify genes whose sgRNAs were, on average, significantly depleted in each cell line over the course of the culture period (Fig. 1A,B).
Comparing screening results between JHOC5 and OSE3 cell lines, we were able to identify 462 genes whose targeting sgRNAs were significantly (P < 0.02) depleted in JHOC5 cells after the growth period, compared to 535 genes in OSE3 cells (Fig. 1C), 52 of which were common to both cell lines. As far as we are aware, this is the first study to date to perform a comprehensive genome-wide CRISPR-Cas9 screening against not only an OCCC cell line, but also an OSE cell line in tandem. This distinction is especially important, as the inclusion of a normal sample is critical for the identification of cancer-specific lethalities and growth-promoting mechanisms. Indeed, analysis of the 52 genes common to both JHOC5 and OSE3 using the GO-slim biological function panther classification system ontologies [21,22] revealed them to be significantly (FDR < 0.05) enriched with housekeeping genes related to ribosomal assembly (GO:0000027), tRNA metabolism (GO:0006399), and translation (GO:0006412).
Performing GO statistical overrepresentation analysis on the JHOC5-specific growth-critical genes, we were able to identify seven significantly enriched (P < 0.01) ontologies (Fig. 1D). As expected, the top ontologies were related to cell cycle deregulation. However, among the most significant ontologies was the RNA splicing gene set (GO:0008380). Although splicing has long been implicated in a wide variety of cancer types [23,24], the role it plays in OCCC remains unclear. As such, we decided to focus our attention on this category for further analysis, obtaining a pooled list of 11 splicing-related genes ( Table 6).
To help narrow down our list of candidate genes, we turned to a recently published public dataset of CRISPR-Cas9 growth screenings carried out and curated by the Broad Institute [20,25]. Using an algorithm designed to remove confounding effects from copy number variation, a dependency score is calculated for each gene in each cell line. These scores are normalized so that a score of 0.0 represents a gene nonessential for growth in a particular cell line, and a score of À1.0 represents a gene highly essential for growth in a particular cell line. We first calculated the differential dependency score of each of the previously identified 11 candidate genes (Table 6) from the RNA splicing gene set in 42 epithelial ovarian cancer cell lines, compared to all other cancer cell lines. We next employed a Student's t-test to determine which of these genes have a consistently lower dependency score among ovarian cancer cell lines. Our decision to group OCCC cell lines with those of the other epithelial ovarian cancer subtypes was based on both the small sample size of cell lines belonging to the OCCC subtype as well as the lack of other cancers with a general mutational profile similar to OCCC. The only gene found to be significantly enriched in ovarian cancer cell lines was the splicing factor DHX38/PRP16 (Fig. 1E). To help confirm this result, we carried out binomial regression on the dependency score data for DHX38/ PRP16, using it as a predictor variable to model the likelihood of a cell line belonging to a particular lineage. We found that out of 14 clinically relevant lineages tested, only epithelial ovarian cancer was significantly (P < 0.001) likely to display a more negative DHX38/PRP16 dependency score compared to all other cell lines (Fig. 1F).
Recently, DHX38/PRP16 has been implicated in the proliferative potential of the HCT116 colorectal carcinoma cell line, where cells harboring an active KRAS mutant become responsive to DHX38/PRP16 knockdown [26]. However, as reported in the Cancer Cell Line Encyclopedia, of the three OCCC cell lines employed in the current study, ES2 and JHOC5 have no activating mutations in the KRAS signaling pathway. This suggests that while DHX38/PRP16 growth dependency in colorectal carcinoma cells may rely on KRAS, the link between the two in other cancers, including OCCC, may not be as strong.

DHX38/PRP16 knockdown in OCCC induces apoptosis
We next wondered whether DHX38/PRP16 knockdown might be inducing apoptosis in knockdownsusceptible cell lines. Cells were transfected with either a control siRNA or one of two siRNA-targeting DHX38/PRP16 and allowed to proliferate for 4 days, after which DHX38/PRP16 knockdown was confirmed at both the transcriptional and translational level (Fig. 3A,B). We employed RT-qPCR to determine whether we could observe any change in the transcript level of several pro-apoptotic genes [p53 upregulated modulator of apoptosis (PUMA), growth arrest and DNA damage-inducible alpha (GADD45A), NOXA, BAX, BID, IL6, and TNFA] upon DHX38/PRP16 knockdown. We found the potent pro-apoptotic factor PUMA to be transcriptionally upregulated in a cancerspecific manner (Fig. 3C), suggesting possible apoptotic induction by p53. We also detected transcriptional upregulation of GADD45A (Fig. 3D), which is also known to play a role in the induction of apoptosis.
To corroborate these results, we employed propidium iodide staining of OCCC and OSE cells 4 days post-DHX38/PRP16 knockdown using lentivirally delivered shRNAs followed by flow cytometry. DHX38/PRP16 knockdown resulted in a sharp (2-109) increase in the sub-G1 population compared to a cohort expressing shLuc in OCCC cells; we were unable to observe a similar increase in OSE cells (Fig. 3E). Interestingly, although the ES2 OCCC cell line harbors a mutation in one allele of p53, it still shows a level of growth inhibition and apoptotic induction upon DHX38/PRP16 knockdown similar to that of p53 nonmutant OCCC lines [27].
To help elucidate the potential role of p53 in DHX38/PRP16 knockdown-mediated apoptosis, we employed two HCT116 cell lines, one with wild-type p53 (HCT116-p53 wt/wt ), and one with a p53 double knockout (HCT116-p53 À/À ). Knockdown of DHX38/  None  ES2  TP53, BRAF, JAK1  TOV21G  PIK3CA, ARID1A, ARID1B PRP16 via RNAi led to a concomitant decrease in cell viability in both HCT116-p53 wt/wt and HCT116-p53 À/À cells (Fig. 4A,B). Additionally, DHX38/PRP16 knockdown led to an increase in PUMA expression, as well as to an increase in cleaved caspase-3 levels in HCT116-p53 wt/wt cells (Fig. 4C). This effect was much less pronounced in HCT116-p53 À/À cells. Even so, as stated above, HCT116-p53 À/À cells show a similar level of growth inhibition to HCT116-p53 wt/w upon DHX38/ PRP16 knockdown. This suggests that while the cell growth inhibition seen upon DHX38/PRP16 knockdown may proceed in part through the p53 pathway in cells that have an intact pathway, it is not strictly requisite for such growth inhibition.
Overall, the above data provide evidence suggesting that the induction of apoptosis brought about by DHX38/PRP16 knockdown is not necessarily dependent on the presence of wild-type p53, but that it instead proceeds through multiple p53-dependent and p53-independent pathways. In the context of OSE, this would suggest that the lack of growth inhibition seen following DHX38/PRP16 knockdown may not be due to the presence of the large-T antigen and therefore that knockdown of DHX38/PRP16 may result in an OCCC-specific induction of apoptosis.

DHX38/PRP16 knockdown abrogates the tumorigenicity of OCCC cell lines
Finally, in order to confirm the applicability of DHX38/PRP16 impairment to the treatment of OCCC, we sought to replicate our findings in vivo. We established TOV21G and ES2 cell lines constitutively expressing either one of two shRNAs targeted to DHX38/PRP16 or to the luciferase gene and implanted them into the flanks of immunocompromised mice. Mice implanted with TOV21G (Fig. 5A) or ES2 (Fig. 5B) cells infected with a control shRNA expression cassette against the luciferase gene (shLuc) all developed tumors, whereas mice implanted with shDHX38/PRP16-expressing cells showed either significantly impaired or no tumor development. Additionally, the average tumor mass of the shLuc cohort upon completion of the study was found to be significantly greater than either of the two shDHX38/PRP16 cohorts for both TOV21G (Fig. 5C) and ES2 (Fig. 5D). These results confirm in vivo that DHX38/ PRP16 expression is critical for the growth of OCCC cells.

Discussion
The present study highlights the importance of DHX38/PRP16 for the tumorigenicity of OCCC; its knockdown induces apoptosis in vitro and prevents tumor formation in vivo. We found that DHX38/ PRP16 knockdown led to an OCCC cell-specific transcriptional upregulation of GADD45A, a target of p53 and BRCA1 which is known to play roles in DNA repair and cell cycle checkpoint control, as well in the induction of apoptosis [28]. The transcriptional upregulation of PUMA observed upon DHX38/PRP16 knockdown in not only OCCC cells but also HCT116-p53 wt/wt cells would indeed support a role for p53 in DHX38/PRP16 knockdown-induced apoptosis.
We also were able to observe PUMA upregulation upon DHX38/PRP16 knockdown in HCT116-p53 À/À cells, though at a reduced level compared to HCT116-p53 wt/wt cells. As both post-knockdown p53 wt/wt cells and p53 À/À cells showed a similar degree of growth inhibition, this would imply the involvement of one or several additional apoptotic pathways unrelated to p53. PUMA, while most frequently associated with its role in apoptosis directly downstream of p53, can also be upregulated by a number of unrelated transcription factors independent of p53 activity [29,30]. In one study, NF-jB was identified as being directly responsible for PUMA upregulation in an HCT116-p53 À/À cell line in response to treatment with TNF-a [31]. Certain proteasome inhibitors have also been found to activate PUMA and lead to apoptosis in a p53-independent manner [32]. Taken together, this hints at the possibility that post-DHX38/PRP16 apoptotic induction is occurring partially independent of p53. Future research will focus on elucidating the p53-dependent and p53-independent mechanisms underlying DHX38/ PRP16 knockdown-induced apoptosis in OCCC. Consistent with our result, amplification of DHX38/PRP16 has been found in as many as 56% of acute myeloid leukemia specimens as well as in established acute myeloid leukemia cell lines [33]. Additionally, several families of splicing factors, including the DDX/DHX family of RNA helicases to which DHX38/PRP16 belongs, have long been implicated in tumor progression and cellular proliferation [24,[34][35][36][37][38][39]. Our future research will therefore also focus on elucidating the splicing-related mechanisms through which DHX38/PRP16 may play a role in the induction of apoptosis upon knockdown in OCCC.DHX38/PRP16 has also been found to have roles outside of the spliceosome; for example, it binds to the protein phosphatase 4 (PP4) protein phosphatase complex, inhibiting the dephosphorylating activity of its PP4C/PP4R2 subunits in both cancer and normal cell lines [40]. Thus, we will also need to consider what mechanisms other than mRNA splicing may be at play. Education, Culture, Sports, Science and Technology (no. 17H06325), Japan.