MSI2 promotes translation of multiple IRES-containing oncogenes and virus to induce self-renewal of tumor initiating stem-like cells

RNA-binding protein Musashi 2 (MSI2) is elevated in several cancers and is linked to poor prognosis. Here, we tested if MSI2 promotes MYC and viral mRNA translation to induce self-renewal via an internal ribosome entry sequence (IRES). We performed RIP-seq using anti-MSI2 antibody in tumor-initiating stem-like cells (TICs). MSI2 binds the internal ribosome entry site (IRES)-containing oncogene mRNAs including MYC, JUN and VEGFA as well as HCV IRES to increase their synthesis and promote self-renewal and tumor-initiation at the post-transcriptional level. MSI2 binds a lncRNA to interfere with processing of a miRNA that reduced MYC translation in basal conditions. Deregulation of this integrated MSI2-lncRNA-MYC regulatory loop drives self-renewal and tumorigenesis through increased IRES-dependent translation of MYC mRNA. Overexpression of MSI2 in TICs promoted their self-renewal and tumor-initiation properties. Inhibition of MSI2-RNA binding reduced HCV IRES activity, viral replication and liver hyperplasia in humanized mice predisposed by virus infection and alcohol high-cholesterol high-fat diet. Together MSI2, integrating the MYC oncogenic pathway, can be employed as a therapeutic target in the treatment of HCC patients. A hypothetical model shows that MSI2 binds and activates cap-independent translation of MYC, c-JUN mRNA and HCV through MSI2-binding to Internal Ribosome Entry Sites (IRES) resulting in upregulated MYC, c-JUN and viral protein synthesis and subsequent liver oncogenesis. Inhibitor of the interaction between MYC IRES and MSI2 reduces liver hyperplasia, viral mRNA translation and tumor formation.

with His-tag was purified from E. Coli with anti-His-conjugated beads and eluted by using imidazole. Fractionated samples in Elution #1-9 were loaded. Wash: represents washing buffer flowing through column. Coomassie Brilliant Blue staining (Top) and immunoblotting (bottom) were performed to verify recombinant proteins. His-tagged protein was detected in peak fractions between #3 to 8 for MSI2 (Left) and MSI2 RNA binding domain (RBD: aa8-193) (Right). Fig. 3C-S3D. Production, purification and verification of His-tagged of recombinant hnRNPK and YB1. Recombinant hnRNPK (Left) and YB1 (Right) with His-tag was purified from E. Coli with anti-Hisconjugated beads and eluted by using imidazole. Fractionated samples in Elution #1-9 were loaded. Wash: represents washing buffer flowing through column. Coomassie Brilliant Blue staining (Top) and immunoblotting (bottom) were performed to verify recombinant proteins. His-tagged protein was detected in peak fractions between #3 to 6 for hnRNPK (Left) and #3 to 9 for YB1 (Right). •

EXPERIMENTAL MODEL AND SUBJECT DETAILS o
In vivo animal studies • Genetically manipulated mouse models • Mice • In the animal studies, NSG-SMG3 mice were bred at the USC mouse facility. The primary mouse fibroblast cultures were prepared from both core transgenic mouse and littermate embryos by trypsinizing the embryonic tissue and plating the dissociated cells. Littermates on FRG mice and NSG-SMG3 mice (Jackson Lab) were intercrossed at least six times. Lieber-DeCarli diet containing 3.5% ethanol or isocaloric dextrin (Bioserv, Frenchtown, NJ) was fed to all the mice in the alcohol feeding arm of the experiment. High-cholesterol high-fat diet was modified from TD.03350 (Harkan Teklad, Inc.) as previously described (Haluzik et al., 2004;Van Heek et al., 1997). Before treatment, mice were randomized to avoid the confounding effects.
Tumorigenicity studies in NSG mice NOD;Shi-scid;Il2rg null (NSG and NSG-SMG3) mice were obtained from Jackson Laboratory and housed under pathogen-free conditions in accordance with approved Institutional Animal Care and Use Committee protocols. MSI2, sh-MSI2 and scrambled shRNA stable-transduced Huh7 cells were generated with the corresponding lentivirus construct. Transduced cells were resuspended in 100 μL PBS and 100 μl Matrigel (BD Biosciences) and injected subcutaneously into the dorsal hind flanks of anesthetized mice as previously described (Carlsson et al., 1983;Siddique et al., 2015;Siddique et al., 2011). Before treatment, mice were randomized to avoid the confounding effects. Tumor-bearing animals were euthanized at day 60 or at humane-end-point and tumors were collected and measured for volumes and weights. Tumor size was measured with calipers, and the volume was calculated according to the formula V = (a × b 2 )/2, where "V" represents tumor volume, 'a' represents the largest and 'b' the smallest superficial diameters (Carlsson et al., 1983).

Subcutaneous xenograft transplantation of the TICs into immunodeficient mice
All the animal experiments were approved by the IACUC Committee. MSI2, sh-MSI2 and scrambled stabletransduced Huh7 Cells (10,000) in 100 μl PBS were mixed with 100 μl Matrigel (BD Biosciences) and injected into the dorsal flanks of anesthetized NOG mice. The tumor volume was measured with a caliper and calculated according to the formula V= (a × b 2 )/2, where "V" represents tumor volume, "a" presents the largest, and "b" the smallest superficial diameter (Carlsson et al., 1983).

Tumor collection and analysis
Tumor-bearing animals were euthanized at day 35, and tumors were collected the volume and weight measured. The tumor tissues were divided for (1) fixation with neutrally buffered 10% formalin for H&E staining and histological evaluation of the tumor; (2) fixation with 4% paraformaldehyde followed by sucrose treatment for subsequent immune-staining; and (3) snap-freezing for mRNA and protein analysis of the targeted genes with shRNA. •

Human patients' tissue samples
The clinical and pathological characteristics of the patients were summarized in the Table S8. Thirty formalinfixed, paraffin-embedded paired primary HCC tissues or non-tumor (adjacent liver) tissues were obtained. 30 patients were diagnosed as having HCC based on the clinicopathologic findings at the University of Southern California Norris Cancer Hospitals. At the time of surgical resection, the tumor area was dissected from the surrounding tissue. Part of the resected tissue was fixed in formalin and embedded in paraffin for histological diagnosis, another part of the resected material was snap frozen in liquid nitrogen for storage at -80°C for molecular analysis. All tissues were collected with patient informed consent that was granted before surgery by a protocol approved by the Institutional Review Board. These tissues were de-identified prior to use in this study. For immunostaining and immunoblotting analysis of MYC and MSI2 in human HCC, necropsy or surgically excised HCC tissues from 30 patients with or without HCV infection, with or without a history of alcoholism, with or without Obesity/Diabetes/ BMI >30 were obtained as cryo-preserved samples and paraffin embedded tissue sections according to the approved University Institutional Review Board (IRB) protocol. Many of the specimens were obtained from Liver Tissue Cell Distribution System at University of Minnesota. Samples were obtained from both genders between the ages of 42-80. Histologically, they all had varying degrees of steatosis (microvesicular and macrovesicular) and inflammation in addition to different stages of HCC. Completely normal liver tissues from 2 patients with accidental death or stroke, but without an apparent liver pathology, were also obtained for immunostaining or immunoblotting. Metastatic Brain tissue was provided by Dr. Zin Htway at HCA Los Robles Hospital and Medical Center. Allotransplants from 15 cryopreserved different mouse metastatic HCC cell lines were also studied for drug susceptibility.
To increase silencing effects and to reduce off-target effects, a combination of shRNA lentiviruses was used to knock down target genes.

Construction of dual luciferase reporter plasmids
Dual reporter plasmids which contain firefly luciferase gene under control of the IRES from 5'UTR of either JUN or VEGFA mRNA were constructed to assess impact of MSI2 on IRES-driven luciferase activities. Briefly, for JUN IRES constructs, a 1020 bp and a 491 bp 5'UTR fragments spanning from −977 to +43 and -448 to +43 relative to the start codon of JUN mRNA were PCR amplified from HepG2 first-strand cDNA library. The corresponding PCR fragments were digested with SpeI and EcoRI and ligated into the pRL and phpRL linearized by SpeI and EcoRI digestion. For VEGFA IRES constructs, a 496 bp 5'UTR fragments were PCR amplified, digested with SpeI and EcoRI and ligated into SpeI /EcoRI sites of the linearized pRL and phpRL. The sequences of primer sets employed for generating dual luciferase reporter constructs were listed in Supplementary Tables S6.

Construction of Vegfa 3' UTR reporter, miR-7014-3P-sponge, and production of lentivirus for lentiviral transduction
To generate Vegfa 3' UTR reporter constructs, the gBlocks gene fragments for 3' UTR of Vegfa mRNA with wildtype or mutant miR-7014-3P binding sites were synthesized by IDTDNA and assembled into pMIR-report vector linearized with MluI and HindIII using NEBuilder HiFi DNA Assembly kit according to manufacturer's instruction. The sponge construct for mmu-miR-7014-3P containing six tandem repeats of the bulge-containing miR-7014-binding motif was generated by synthesizing a set of sense and antisense oligonucleotides with 5′-CTAG-3' overhang and 5′-GATC-3' overhang respectively. The annealed DNA fragments were ligated into XbaI/ BamHI sites of linearized pCDH plasmid. Lentivirus was produced by harvesting culture supernatants obtained upon co-transfection of HEK293T cells with the generated lentiviral constructs and pPAX2 and pMD2.G packaging plasmids. Hepa1-6 cells were plated in a 12-well plate and spin-infected by centrifugation at 1100 × g for 30 min in the presence of lentiviral supernatants and 8 μg/ml polybrene (Sigma-Aldrich Co.). Stably transduced cells were obtained under 2 µg/ml puromycin selection for 48 hours. The sequences of oligonucleotides employed for generating Vegfa 3' UTR reporter, miR-7014-3P sponge, and mimics were listed in supplementary Extended Data Fig. 4E and Tables S6.

RNA-Immunoprecipitation (RIP)
2X10 7 TICs overexpressing Flag-MSI2 and stably transduced shRNA against MSI2 were subjected to RNA immunoprecipitation (RIP) using the MBL RIP kit (MBL International, MA). In brief, cells were washed with cold PBS and lysed with provided RIP lysis buffer. Anti-Flag M2 Ab (Sigma-Aldrich), anti-rabbit Ab, or anti-MSI2 Ab (EMD Millipore), pre-incubated with magnetic beads (5μg), were used to immunoprecipitate Flag-MSI2-RNA complexes. Immunoprecipitated complexes were washed and treated with proteinase K. Finally, RNA was extracted with the reagents provided in the kit.

Bioinformatics analysis of RNA sequencing
Sequencing reads were trimmed of adaptors, and high-quality reads were used for further analysis using Burrows-Wheeler Aligner (BWA) and TopHat essentially following the procedure detailed by Huang et al. (Huang et al., 2013) except that we used a windowed approach to call the enriched regions in the genome. Further reads aligning to mitochondrial DNA, repetitive elements or unassigned sequences were discarded. This resulted in uniquely aligned 13 million reads for further analysis. Briefly in the windowed approach signal density was calculated in sliding 50bp windows, and aligned reads were considered to be within a window of the midpoint of its estimated fragment. Midpoints in each window were counted, and empirical distributions of window counts were created. Genomic bins containing statistically significant regions were identified based on background distribution of randomized reads specific for each sample as detailed previously (Kim et al., 2013). Further, a Mann and Whitney U-test were used to determining the differential enrichment Msi2 IP from input sample. Similarly, for RNA-seq high-quality reads were aligned to mm9 using Top-Hat in conjunction with a gene model from Ensemble release 61. Data were quantitated by counting the number of RPKM (reads per kilo base per million mapped reads) (Mortazavi et al., 2008). The read counts were adjusted globally by matching count distributions at the 75 th percentile and then adjusting counts to a uniform distribution between samples. Differential expression was estimated by selecting transcripts that displayed significant changes (FDR, P < 0.05) using a null model constructed from 1% of transcripts showing the average closet level of observation to estimate experimental noise as explained previously (Mortazavi et al., 2008). Significantly enriched regions of RIP-seq were annotated with closest non-overlapping gene/ ncRNA (downloaded from UCSC genome browser on Jan.5, 2016). For visualization, alignment files were transformed into reading coverage files (25 bp bin and smooth length 50 bp) to generate bigwig or tdf files. IGV was used at different steps to visualize the data. For identification of motif center of RIP-seq window were considered as peak summit the sequence ± 25 bp across this submit was used to identify the consensus motif in different data sets. The link for data to be included: https://drive.google.com/folderview?id=0B3skVboh2QdZ1REUGtwaG11SG8&usp=sharing

Bioinformatics analysis of binding motif from MSI2 RIP-seq
The reads were aligned to mm9 genome using BWA (https://www.ncbi.nlm.nih.gov/pubmed/19451168) allowing single mismatch. The genome was divided into non-overlapping bins of 20 bp and a sliding window approach was used to find regions of enrichment. A read is considered to be within a window if the midpoint of its estimated fragment is within the window. The center of the peak will have maximal enrichment. Zero-truncated negative binomial model (ZNTB) was used to determine the statistically significantly enriched peaks / regions and a peak cut off of FDR adjusted p value (p<0.05) was used (represented as P value in Table). Further a Mann and Whitney U test was used determine the differential enrichment Msi2 IP from input sample. Further ranking of peaks/ genes were performed using Fold enrichment (>2 fold).

Reverse transcription and quantitative PCR (RT-qPCR)
Total RNA was extracted by using TRIzol Reagent (Invitrogen) and purified using the RNeasy mini kit (QIAGEN) according to the manufacturer's protocol. RNA concentration and purity were determined by A260 and A260/A280 ratios, respectively. The RNA samples were treated with DNase I (Invitrogen) to remove residual traces of DNA. cDNA was obtained from 1 μg of total RNA, using SuperScript III reverse transcriptase (Invitrogen) and oligo dT primers in a final volume of 20 μl. Quantitative real-time PCR was performed on an ABI 7900 HT Real-Time PCR machine using 2X SYBR Green Master Mix (Applied Biosystems). Specificity of PCR products was tested by dissociation curves. Gene expression was determined using the 2−ΔΔCt method described in the ABI user manual.

SDS-PAGE and immunoblot analysis
Huh7 Cells were lysed with lysis buffer (100 mM NaCl, 50 mM Tris-Cl pH 7.5, 0.5 mM EDTA, 1% NP-40) containing complete protease inhibitor cocktail (Roche Life Science, Indianapolis, IN, USA). Cell lysates were separated by SDS-PAGE and transferred onto PVDF membranes. The membranes were blocked for 30 min with 5% fat-free milk in Tris-buffered saline containing Tween-20 (TBST; 50 mM Tris-Cl pH 7.5, 150 mM NaCl, 0.1% Tween-20) and incubated for 2 h with the indicated antibody in TBST containing 2% BSA. The membranes were subsequently washed in TBST and incubated for 1 h with HRP-conjugated secondary antibody. The immunoreactive bands were visualized using chemiluminescent HRP substrate (Immobilon Western; Millipore, Temecula, CA, USA) and the ChemiDoc MP Imaging System (BIO-RAD; Hercules, CA, USA). The density of bands on images of western blots was analyzed by ImageJ.

Luciferase reporter assays
Huh7 cells were plated on 24-well plates, allowed to adhere overnight. The following day, cells were cotransfected with MYC IRES wildtype or mutant luciferase reporter construct, and expression plasmids for MSI2, as indicated, using BioT (Bioland Scientific, LLC.). Cells were lysed 48 hr post-transfection and luciferase activity was determined. Cell lysates were subjected to Western blotting analysis for detection of relative protein expression level of MSI2. Relative luciferase activities were calculated as fold induction relative to unstimulated control.

UV cross-linking and EMSA
To show an interaction between MSI2 protein and MYC mRNA, UV cross-linking and EMSA was performed with 32P-labeled MYC entire IRES or the MSI2-binding-motif mutant transcripts combined with purified MSI2-His6, in the presence or absence of cold probes as previously described (Paulin et al., 1998). Radiolabeled RNA transcript was generated from pSK-ML (MYC IRES expression construct) and pSK-GAP-L (GAPDH vector for specificity control) (plasmid gifts from Prof. Anne Willes in the UK) (Paulin et al., 1998) by using MAXIscript™ SP6/T7 Transcription Kit (Invitrogen). Radiolabeled RNA transcript was incubated with purified recombinant MSI2 and subjected to EMSA.

Sucrose density gradient centrifugation and RT-qPCR analysis
Sucrose density gradient (5%-50%) ultracentrifugation was used to separate ribosomes into polysomal and subpolysomal fractions. Briefly, Huh7 cells were treated with 100 ug/ml cycloheximide (CHX) for 15 min at 37°C then collected and lysed in polysome lysis buffer. Following cell disruption and centrifugation, cytoplasmic cell lysates (supernatant) were layered onto sucrose gradients 5-50% and spun at 45,000 rpm for 1.5 h at 4 °C in a SW55Ti rotor (Beckman Instruments). Gradients were then fractionated and total RNA was purified from each fraction using TRIzol (Invitrogen), according to the manufacturer's instructions. Reverse transcription was performed using SuperScript III first-strand synthesis system (Invitrogen) with oligo-dT primer. RT-qPCR was carried out using ABI Step One Plus Real-Time PCR System and TaqMan universal PCR Master Mix (Applied Biosystems) for gene expression analysis. Data were analyzed using the 2 −ΔΔCt method described in the ABI user manual. The expression level of MYC mRNA was normalized to GAPDH. The primer sequences are listed below, MYC Fwd: 5'-ttcgggtagtggaaaaccag-3', Rev: 5'-cagcagctcgaatttcttcc-3'; GAPDH Fwd: 5'gagtcaacggatttggtcgt-3', Rev: 5'-gatctcgctcctggaagatg-3'.

MicroRNA isolation, cDNA synthesis, and RT-qPCR
Small RNAs including microRNAs were extracted by using microRNA Isolation Kit (Sigma) followed by poly-A tailing reactions and miRNA Easy Script cDNA synthesis kit (Abm Good, BC, Canada). 250 ng small cellular RNA was used for cDNA synthesis. The specific microRNA was amplified using primers obtained from Abm Good (cat# MPH01277), The RT-qPCR was performed using Evagreen miRNA qPCR Master Mix (Genomicsonline) for quantitative real-time analysis of miRNA samples. The results were normalized to U6 snRNA (MPH00001). The pre-miR22 was amplified by using the primers Fwd-5'-ctgagccgcagtagttcttc-3'; Rev-5' cagagggcaacagttcttca-3'

Northern blots for miR22hg
Levels of miR-22 in cells under different expression levels of MSI2 were measured by using Northern Blot Assay kit, according to manufacturer's instructions (Signosis). In brief, 10 μg of small RNAs were separated by electrophoresis in 16% Tris-borate-EDTA urea gels, transferred to a nylon membrane, and then immobilized by UV-crosslinking. Membranes were hybridized overnight at 42°C with a biotin-labeled, complementary miR-22 and U6 probes. Membranes were blocked for 30 min and then incubated with streptavidin-HRP at room temperature. After washing, miRNAs were detected with a chemiluminescence substrate and the bands were detected by exposure to X-Ray film. The densities of the bands were measured by ImageJ software.

Serial Spheroid Formation
Huh7 cells were washed with 1x PBS and centrifuged at 150 g for 5 min. Cells were counted using a cell counter. 100 cells were serially diluted to three separate wells in a 6 well Ultra-low attachment cell culture plate by Corning. DMEM F12 media (Sigma) was used containing 10% FBS, Insulin (1 mg/ml), Dexamethasone (1 x 107 M), Nicotinamide (10 mM), HEPES (5 mM), 1% penicillin/ streptomycin, and Epidermal Growth Factor (20 ng/ml). Spheroids were allowed to grow for ten days in the 5% CO2 incubator and then were serially diluted to 100 cells per well twice more. Statistical significance was determined using ANOVA Single Factor. Asterisks (*) indicate a p-value < 0.05.

In silico 3D structure prediction
To simulate the docking site(s) between MSI2 and mRNA regions, local quality was estimated. Human MSI2 structure was simulated by using templates of 2mssA and 2lyvA. Non-redundant sets of PDB structures were compared. The simulated structure has three potential RNA binding sites. To predict the interfaces between RNA and MSI2, each binding site was simulated 1500 times with redundant radii searching for lowest energy, results were then combined to return the lowest energy from these parameters. Cavity 2 had the lowest calculated energy and was chosen for further analysis.
Fluorescence Polarization assay FITC-tagged MYC IRES-MSI2 binding RNA (wild-type) or FITC-mutant IRES RNA was serial diluted in assay buffer (150 mM NaCl, 25 mM HEPES (pH 7.6), 0.05% CHAPS and 1 mg/ml BSA) in a range of 0.5-500 nM and aliquoted into a black 384-well non-binding microplate (Greiner Bio-One, Germany) in triplicates (20 ul). Bufferonly (20 ul) serves as blank controls and probe-only controls for assessing displacement by the unlabeled small molecules. Total fluorescence intensities of probe were recorded with an EnVision multilabels plate reader 2103 (PerkinElmer) using the following settings: mirror -FITC FP, excitation filter -FITC FP 480 nm, emission filter 1 -FITC FP p-pol 535 nm, emission filter 2 -FITC FP s-pol 535 nm. The resulting values were plotted with GraphPad Prism against the respective concentrations of probe.

Histology & Immunohistochemistry
Five micrometer-sections were stained with hematoxylin and eosin (H&E) or processed for other staining. Tissue samples were fixed on 10% neutral buffered formalin, embedded in paraffin, sectioned, and stained with hematoxylin and eosin or immunohistochemistry staining was performed using primary antibodies against MSI2 and/ or MYC based on the standard protocol with their respective secondary antibodies. Slides were then mounted using xylene based mounting media including hematoxylin for nuclei counterstaining (Vector Laboratories) according to the manufacturer's recommendations. The staining was subjected to morphometric analysis. To determine the specificity of immunohistochemistry staining, serial sections were similarly processed, except primary antibodies were omitted in controls. The areas of interest were quantified using MetaMorph software.

Immunofluorescence Staining
The staining intensity, percentage of stained cells and immuno-reactive score was determined as reported earlier (Koomagi and Volm, 1999). In brief, the intensity of staining was scored on a scale of 0 to 3, in which, negative staining = 0, weakly positive staining = 1, moderately positive staining = 2, and strongly positive staining =3. The extent of distribution of positive cells was estimated on a scale of 0 to 4, in which negative = 0, positive staining in 1-25.0% of cells = 1, positive staining in 26-50.0% =2; positive staining in 51.0-75.0% =3; and positive staining in more than 75.0% =4. The results were evaluated as immunoreactivity score (IRS), where the IRS = (percentage of positive cells) X (staining intensity). Two investigators separately and independently evaluated the immunohistochemical staining without knowledge of the clinical data. Significant differences in interpretation were resolved in conference. In cases without significant discordance, scores of two observers were averaged.Normal and Tumor tissues were provided by the University of Minnesota. Microscopy services were provided by the Cell and Tissue Imaging Core of the USC Research Center for Liver Diseases, HIH grants No. P30 DK048522 and S10 RR022508.

RT-qPCR analysis of mature microRNA expression
Total RNA including small RNA species was purified using TRIzol (Invitrogen) following the method provided in the TRIzol user manual. Reverse transcription and quantitative PCR (RT-qPCR) was carried out using miR-X miRNA qRT-PCR TB Green kit (TAKARA BIO) according to the manufacturer's instructions, and ABI StepOnePlus™ Real-Time PCR System for microRNA detection. Data was analyzed using the 2 −ΔΔCt method. The expression of microRNA was normalized to RNU6. The primer sequences are listed in Supplementary Supplementary Table 6.

Flow cytometric analysis
For intracellular staining, HepG2 spheroid cells were first dissociated into single cells by StemPro Accutase (GIBCO, Carlsbad, CA, USA). The cells were subsequently fixed and permeabilized with Fix and Perm Buffers following instruction described in True-Nuclear™ Transcription Factor Buffer Set (Biolegend, San Diego, CA, USA), and were incubated in 1 × Perm Buffer containing MSI2 Ab, Alexa Fluor 647-conjugated NANOG, and Alexa Fluor 488-conjugated MYC Abs at 4 °C overnight. The next day, the cells were washed and incubated with Alexa Fluor 350-conjugated anti-rabbit Ab at 4 °C for 1 hour. The cells were washed and resuspended in PBS containing 0.2% BSA, and were subjected to flow cytometric analysis on a FACSCanto II flow cytometer with FACSDiva software (Becton Dickinson, San Jose, CA, USA).

Statistical Analysis
Statistical significance was estimated by unpaired, two-tailed Student's t-test. Bars represent the mean± S.D. *, P < 0.05. Paired t-tests were used to compare the marker expression levels between tumor vs. non-tumor tissues. For each of the markers, percent of staining and intensity of staining, as well as the product of the two measurements (immunoreactivity score: IRS), were presented as dot plots (Koomagi and Volm, 1999). The differences in survival among the three groups of mice (scrambled, MSI2 overexpressing and sh-MSI2) were compared using log-rank tests. Given the small numbers of mice in this experiment, permutation tests were conducted to evaluate the significance of the observed differences among the three groups of mice. Statistical analyses were performed using STATA software (version 11.0; Stata Corp LP College Station, TX). Detailed analyses are explained in Supporting Materials.

Supplementary Discussion
MYC is a potent oncogene of liver tumors and inactivation of MYC is sufficient to induce sustained regression of MYC-initiated liver tumors in mice (Ryan and Birnie, 1996). MYC regulates several cellular processes and is crucial for stem cell maintenance. Down-regulation of MYC both in vitro and in vivo induces growth inhibition, and differentiation of HCC (Ryan and Birnie, 1996). Elevated expression of MYC occurs through multiple mechanisms in tumor cells. One such mechanism is IRES-mediated trans-activation. There are several IREStrans acting factors (ITAFs) that bind to the MYC IRES such as-PSF (PTB-associated splicing factor) and YB-1 (Y-box binding protein 1) (Cobbold et al., 2008;Paulin et al., 1998). The 5'-UTR of the MYC is highly structured as it plays an important role in modulating levels of MYC protein. Transcription initiation of MYC starts from multiple promoters -P0, P1, P2, and P3 but the majority of mRNAs initiate from the P2 promoter (Zakaria et al., 2014). Two major MYC proteins MYC-1 (67kDa) and MYC-2 (64 kDa) are translated from the P2 promoter transcript. The MYC IRES is 341 nucleotide long (nt 2500-2840 exon 1) and is located downstream of the P2 promoter. In the present study, we observed that MSI2 selectively modulated the expression of the longer form of MYC, thus initiating translation from within exon1 and allowed IRES-dependent translation. MSI2 promotes EMT in liver cancer through LIN28 expression (Fang et al., 2017). Other ways to target MSI2 includes antisense oligonucleotides (ASOs) (Fox et al., 2016) or small molecule inhibitor for MSI2 (Ro 08-2750) targets MYC (Minuesa et al., 2019). It was reported that MSI2 with HuR inhibits the processing of miR-7 (Choudhury et al., 2013). Another study showed that LIN28 blocks processing of pre-let-7g to mature let-7g (Viswanathan et al., 2008). Although we do not know if MSI2 interacts physically with the miRNA processing machinery or exerts its influence via other factors, this study represents yet another component of the complex regulation of MYC. In this context, MSI2 warrants further examination, especially if it functions similarly to LIN28, either directly or indirectly on MYC. Finally, the MYC-3'-UTR also interacts with different miRNAs expressed in other cancers; thus it will be of interest to learn whether MSI2 regulates MYC translation and similarly regulates other miRNAs. Although the functions of many identified long noncoding RNAs remain unknown, we demonstrated that miR-22 availability is antagonized by MSI2. Our work identified the MSI2-MIR22HG-miR-22-MYC axis as an important regulator of HCC.
• QUANTIFICATION AND STATISTICAL ANALYSIS Statistical Considerations: For this study, HCC from three etiological backgrounds was used for PDX models: HCC can be attributed to ALD and NASH as these metabolic liver diseases are increasingly contributing to HCC incidence and are primary diseases of the center's interest. HCC from HCV or HBV patients are not included in this study because different viral genotypes may exhibit confounding different phenotypes and HCV-or HBVinfected PDX models pose a significant logistical challenges from the biosafety standpoint. To account for patient variability, we collected HCC from three patients per etiology. We expanded one patient HCC into at least 48 NSG mice by 2nd or 3rd level passages, collectively generating 144 or more mice from three patients per etiology. These 144 mice were randomly assigned to the two groups. The initial analysis considered the two etiology groups separately and the two treatment groups were compared by two-tailed, Student's t-test or Mann-Whitney U-test if abnormal data distribution exists. For in vitro culture studies, Student's t-test was used to analyze data using Statistical software. Statistical Analysis: Experimental data are presented as the mean ±standard deviation (SD). All statistical analysis was performed using a two-tailed Student's t test and Chi squared test. Differences were considered statistically significant when P values were less than 0.05. Error bars reflect standard errors.
• DATA AND CODE AVAILABILITY o NCBI tracking system number: § The data used in this study has been deposited to NCBI under GSE61435 (Microarray) and GSE68237 (ChIP-Seq).