Tumorigenic and Antiproliferative Properties of the TALE-Transcription Factors MEIS2D and MEIS2A in Neuroblastoma

Alternative splicing, the combinatorial use of protein coding exons during the maturation of pre-mRNA to mRNA, facilitates expression of multiple mRNA isoforms from the same gene, thus greatly expanding the protein repertoire in a given cell. In humans, over 90% of multiexon genes are subject to alternative splicing, and the number of different splice isoforms synthesized from a single gene can vary between few and several thousands (1). Genome-wide transcriptome profiling of normal and diseased tissues revealed an unexpected high incidence of conditions involving altered mRNA splicing of disease-associated genes. In fact, defective splicing has been linked to all aspects of cancer biology, including tumor initiation and progression, metastasis and therapy resistance (2). In addition, alternative splicing is particularly prevalent in the mammalian nervous system, with emerging roles in nearly all aspects of neuronal development and function, and dysregulated splicing has been implicated in neurodegenerative pathologies, schizophrenia, and autism spectrum disorders (3, 4).

Neuroblastoma is a pediatric solid tumor that arises from the sympathoadrenal lineage of the neural crest. Neuroblastoma accounts for about 15% of all childhood cancer deaths in the United States and Europe, making it the most frequent extracranial solid tumor in infants and young children (5). This neuroendocrine malignancy displays diverse clinical spectrums, ranging from defined tumors removable by surgery (categorized as INSS stages st1 and st2) to highly metastasized disease (st3 and st4). Patients with st4 tumors frequently exhibit amplification of the MYCN protooncogene and have a poor survival probability, even under aggressive, multi-modal therapy (5, 6). In about 10% of patients, however, tumors and metastases regress spontaneously without any treatment. Such regressing tumors mostly occur in patients under one year of age at diagnosis and are recognized as the distinct tumor stage 4S. 4S tumors show low cell proliferation and contain large numbers of terminally differentiated neurons (7). The underlying cause of this spontaneous tumor regression is still unknown, although favorable expression of neurotrophin receptors, permissive epigenetic modification of differentiation-related genes and telomere shortening have been implicated (7).

Here, we report that the transcription factor MEIS2 affects neuroblastoma proliferation and differentiation in opposing ways, depending on the relative expression of two of its alternative splice isoforms. MEIS2 (myeloid ecotropic integration site 2) belongs to the TALE (three amino acid loop extension) family of atypical, evolutionary conserved homeodomain (HD)-containing proteins. MEIS2 controls a broad range of developmental processes, but its role in onset or progression of cancer appears complex (8). Strong MEIS2 expression correlates with benign forms and a higher survival probability in ovarian cancer, and MEIS2 transcripts are downregulated in poor-prognosis prostate cancers, implicating MEIS2 as tumor suppressor in both malignancies (9, 10). By contrast, MEIS2 appears to be required for M-phase progression in neuroblastoma cell lines, suggesting oncogenic properties (11). Several splice isoforms exist for MEIS2, which differ in the alternative inclusion of 7 amino acids following the HD or of the C-terminal exon 12. Whether these isoforms are functionally distinct, was so far unknown. We now show that two of these isoforms, MEIS2A and MEIS2D, exert antagonistic functions in neuroblastoma.

All 18 human neuroblastoma patient genome-wide mRNA expression profiling datasets available in the public domain were retrieved from the public NCBI GEO GSE (http://www.ncbi.nlm.nih.gov/geo/), EMBL EBI (http://www.ebi.ac.uk/arrayexpress), or TARGET (https://ocg.cancer.gov) websites. The Albino-28, Heiskanen-49, Łastowska-30, Lavarino-23, Maris-101, Schramm-18 (GSE7529, -81895, -13136, -54720, -3960, and -65303), Speleman-24 (E-MEXP-669), and Asgharzadeh-249 (TARGET) datasets were examined but not included in the final analysis because they contained ≤ 30 samples and/or missed clinical annotations. The 10 remaining datasets were Delattre-50, Hiyama-51, Jagannathan-100, Khan-47, Kocak-649, Seeger-102, SEQC-498, Versteeg-88, and Westermann-105 (GSE12460, -16237, -19274, -27608, -45547, -3446, -62564, -16476, -73517), and TARGET-122 (TARGET). In no cases were partial data from non-included datasets in conflict with the final data shown. All datasets were analyzed using the R2 genomics analysis and visualization platform developed in the Department of Oncogenomics at the Academic Medical Center—University of Amsterdam (http://r2.amc.nl) using website default settings. The R2 TranscriptView genomic analysis and visualization tool was used to check if probe-sets uniquely mapped to an antisense position in an exon of the targeted gene that showed high transcription activity. All probes selected meet these criteria. Correlation between MEIS2 exon 12 inclusion frequency in primary neuroblastoma tumor biopsies and their time to engraftment as orthotopic patient-derived xenografts (O-PDX) was made possible through the Childhood Solid Tumor Network (CSTN) repository at St. Jude Children's Research Hospital (Memphis, Tennessee; https://www.stjude.org/research/resources-data/childhood-solid-tumor-network.html) and is based on RNA sequencing results obtained from tumor samples SJNBL_124, SJNBL012407, SJNBL013761 and SJNBL015725 (12, 13). To calculate the MEIS2 exon 12 inclusion rate for each sample, the number of junction reads from MEIS2 exon 11 to exon 12 (ex12-inclusion) was divided by the number of junction reads from exon 11 to exon 13 (ex12-skipping). This value was correlated to the reported time to engraftment following para-adrenal injection of the respective sample into immunodeficient mice (12, 13).

Paraffin-embedded tissue specimens from neuroblastoma tumor biopsies were obtained from the Hospital for Pediatric and Adolescent Medicine, University Hospital Frankfurt. The study was conducted in accordance with the Declaration of Helsinki, approved by the Institutional Review Board and local ethical committee of the University Hospital Frankfurt, Germany (GS 4/09; SNO_05-13), informed written consent was obtained from the subjects.

SK-N-BE(2), a gift from Prof. Dr. Michel Mittelbronn, Institute of Neurology, University Hospital Frankfurt am Main, were cultured in DMEM GlutaMAX (Thermo Fisher Scientific) supplemented with 10% FCS (PAN Biotech). The identity of the cell line was verified by short tandem repeat (STR) profiling (Leibniz Institute DSMZ), absence of Mycoplasma infection was verified by conventional PCR with the PromoKine PCR Mycoplasma Test Kit I/C. Stable, inducible cell lines were aliquoted and frozen after derivation and expansion, experiments were performed within the first 25 passages following thawing. SMS-KCN cells, a gift from Prof. Dr. Jindrich Cinatl, and Lan1 cells, a gift from Prof. Dr. Simone Fulda (both Frankfurter Stiftung für krebskranke Kinder e.V., University Hospital Frankfurt am Main) were grown in IBM (Biochrom) or RPMI (Thermo Fisher Scientific), respectively, both with 10% FCS. BE(2)-MEIS2A and BE(2)-MEIS2D cell lines were grown in DMEM GlutaMAX, 10% FCS, 2.5 μg/mL puromycin (Sigma-Aldrich) and 0.8 mg/mL G418 (Carl Roth). Phase contrast images were captured with an Olympus IX70 microscope (Olympus).

Human MEIS2B and MEIS2D (NCBI NP_733774 and NP_733776, respectively) and mouse MEIS2A (NCBI NP_001153039) were fused to the HA-epitope for immunohistochemical detection. The amino acid sequences of human and mouse MEIS2A are identical. Stable, Tet/doxycycline-inducible cell lines were generated by retroviral transduction with the Tet-regulated transactivator/repressor pLIB-rtTAM2-TRSID and MEIS2A-HA and MEIS2D-HA cloned into the EcoR1 site of the Tet-responsive vector pQ-tetCMV-SV40-Neo (both gifts of Prof. Dr. Michael J. Ausserlechner, Medizinische Universität Innsbruck; refs. 14, 15). For the SK-N-BE(2) cell line, single colonies were picked after one week of selection in 0.8 mg/mL G418 (Carl Roth) and 2.5 μg/mL puromycin (Sigma-Aldrich). Puromycin- and G418-resistant Lan1 or SMS-KCN cells were pooled for further assays. Transient overexpression was performed with pcDNA3 or pZome-1-N, a retroviral vector based on pBABE-Puro. For knockdown of exon 12–containing MEIS2 isoforms, neuroblastoma cells were transfected with siRNAs targeting this exon (5′-guaaagcaaucgcaaagcatt-3′ and 5′-ugcuuugcgauugcuuuacat-3′). No other sequence within the 96nt-comprising exon 12 fulfilled the thermodynamic requirement for siRNA-mediated knockdown. Exon 12-specific siRNAs and control siRNAs (Cat No. 4390846) were purchased from Life Technologies (Silencer Select; 100 pmol transfected per 6-well of cells). Transfection was performed with Metafectene Pro (Biontex) according to the manufacturer's instructions.

Anchorage-dependent growth assays of MEIS2A- or MEIS2D-stable transfectants, were started with 30,000 cells per 24-well plate, either stimulated with 0.5 μg/mL doxycycline (Sigma-Aldrich) or untreated. Cells were counted at the days indicated. For proliferation analysis after transient overexpression, SK-N-BE(2) cells were transfected with pZome-1-N-MEIS2A, pZome-1-N-MEIS2B or pZome-1-N empty vector as control. Two days post-transfection the cells were selected with 2.5 μg/mL puromycin for 3 days before cell proliferation was assessed. For growth analysis after transfection with exon 12-specific siRNAs, 10,000 cells were plated 24 hours after transfection and retransfected with the respective siRNAs after 6 days. Colony formation capacity was determined with 5,000 cells per 6-well plate, mixed in 0.4% DNA-grade agar (DifcoTM Noble Agar, BD Pharmingen) in DMEM growth medium (10% FCS, 1% pen/strep) and transferred onto a solidified bottom layer (0.5% DNA-grade agar in DMEM growth medium). Assays were performed with BE(2)-MEIS2A or -MEIS2D cells or with SK-N-BE(2) cells 24 hours after siRNA transfection. After 6 days (siRNA knockdown) or 14 days (BE(2)-MEIS2A or -MEIS2D cells), living cell colonies were stained with 1 mg/mL MTT (Sigma-Aldrich) in PBS for 5 hours at 37°C. Colonies were photographed with an Axiovert 25 stereomicroscope (Carl Zeiss) and counted. To assess neurite outgrowth, BE(2)-MEIS2A and BE(2)-MEIS2D cells were plated onto cover slips coated with 0.1% gelatin and grown for 5 days, either treated with 0.5 μg/mL doxycycline or untreated. Cells were fixed in 4% PFA, stained with antibodies recognizing neurofilament M and neurite length was measured with the ImageJ plug-in NeuronJ. For the images shown in Supplementary Fig. S6, SK-N-BE(2) cells were transfected with pZome-1-N carrying full-length mouse MEIS2A or full-length human MEIS2B, or with empty pZOME-1-N as control. Cells were grown for times between 14 and 61 days under puromycin-selection. MEIS2D-transfected and non-transfected control cultures were passaged every 2 to 3 days, whereas only medium exchange was required in MEIS2A- and MEIS2B-transfected cultures.

For senescence-associated β-galactosidase (β-gal) staining, 15,000 BE(2)-MEIS2A, BE(2)-MEIS2D or SK-N-BE(2) cells were seeded per 200 mm² and grown in medium supplemented with 0.5 μg/mL doxycycline. Medium was exchanged every 3 days. Cells were stained with 250 μL β-gal staining solution (0.1% X-gal, 5 mmol/L K₃Fe(CN)₆, 5 mmol/L K₄Fe(CN)₆, 40 mmol/L HCl/Na₂HPO₄, 2 mmol/L MgCl₂, 150 mmol/L NaCl, pH 6.0) after 5 days of doxycycline treatment following standard procedures and photographed with a Nikon Eclipse TS100 inverted microscope.

Cells were grown on coverslips coated with 0.1% gelatine under standard conditions. Antibodies and staining conditions are listed in Supplementary Table S1. Images were taken with a Nikon 80i microscope and analyzed with the NIS Elements acquisition software (Nikon). A minimum of 2,000 cells per experiment were counted, experiments were performed at least in triplicates, cells were counted blind. Standard error of the mean (SEM) was calculated between biological replicates in Prism 5.01 (GraphPad). If necessary, brightness and contrast were moderately enhanced in Adobe Photoshop CS4 across the entire image. For bromodeoxyuridine (BrdUrd) pulse labeling, cells were incubated with 10 μmol/L BrdUrd (Roche) before fixation for 1 hour. Chromogen staining on 3-μm-thick paraffin sections was performed with either a Ventana Discovery XT automated staining system with Omni-Map horseradish peroxidase detection and counterstaining for hematoxylin or with a BOND Polymer Refine Detection system from Leica (Leica BOND-III). Primary antibodies used were rabbit anti-MEIS2 and mouse anti-HA (Supplementary Table S1). Hematoxylin and eosin staining was performed following standard procedures.

Western blot analysis was performed with 40-μg cell lysates of BE(2)-MEIS2A or BE(2)-MEIS2D cells on polyvinylidene difluoride membranes following standard procedures. Antibodies are listed in Supplementary Table S1. Positive controls for apoptotic cells or cells with autophagic activity were LN229 cells treated with 250 ng/mL TRAIL (Peprotech) for 18 hours or with 100 nmol/L bafilomycin A1 and 250 nmol/L Torin 1 for 2 hours and were kind gifts from Elena Ilina (Neurological Institute, University Hospital Frankfurt am Main).

Total RNA was isolated from 6-well cultures using the RNeasy Mini Kit (Qiagen) and transcribed into cDNA with the RevertAid H Minus First Strand cDNA Synthesis Kit (Thermo Scientific). Quantitative real-time PCR was carried out using ABsolute QPCR SYBR Green Fluorescein Mix (Thermo Scientific) on a CFX96 Real-Time System (Bio-Rad), primer pairs are listed in Supplementary Table S1.

0.3 μg pGL3-RARE-luciferase (Addgene #13458) were cotransfected with 0.2 μg pcDNA3-MEIS2A-HA, pcDNA3-MEIS2D-HA or pcDNA3-HA mammalian expression vectors and 0.05 μg pHef1-Renilla (16). Twenty-four hours later, cells were treated with 10 μmol/L retinoic acid (RA; Sigma-Aldrich) in DMSO (Sigma-Aldrich; positive control) or DMSO alone for 24 hours. Reporter gene activity was measured with a Tecan Infinite M200 pro plate reader with I-Control 1.7 software. Values obtained for luciferase activity were normalized to those obtained for Renilla luciferase.

Chromatin immunoprecipitation (ChIP) was carried out as published (17) from 1 × 10⁶ SK-N-BE(2) cells per ChIP. Cells were fixed and the purified chromatin was sheared by sonication with a Bioruptor Plus (Diagenode) to an average fragment length of 100–300 bp. Chromatin-immune complexes were collected by incubation with 50 μL of a 1:1 mixture of ProteinA and ProteinG Dynabeads (both LifeTechnologies). DNA was purified with Qiagen MinElute columns (Qiagen) and analyzed in triplicates by qPCR. Enrichment of the precipitated DNA was determined relative to the input (1:100) as 100 × 2^{(Ct adjusted Input−CtIP)}. Antibodies and primers are listed in Supplementary Table S1.

Per experiment, 1 × 10⁶ cells were diluted in 20 μL Corning Matrigel containing 0.5 μg/mL doxycycline and grafted onto the chorioallantoic membrane (CAM) of fertilized chicken eggs at E7 (white leghorn, LSL Rhein-Main, Dieburg, Germany). BE(2)-MEIS2A and BE(2)-MEIS2D cells were treated with 0.5 μg/mL doxycycline for 4 hours before the surgery and placed inside a sterile precision O-ring (5 × 1 mm, HUG) onto the CAM. The eggs were allowed to develop for 6 days, 10 μL 0.1 μg/mL doxycycline were added onto the tumor cells daily. Tumors were isolated from the CAM and embedded in paraffin. Three-μm sections were generated and analyzed by automated IHC staining.

The subcutaneous flank injections were approved by the local animal care committee and the Regierungspräsidium Darmstadt (license V54-19c20/15-FK/1063) and were in accordance with the law for animal experiments issued by the state government as well as German and European Union guidelines. 8-weeks-old male NU/NU mice (Envigo) were deeply anesthetized with a narcotic solution containing analgesic and sedative components. Dissolved Matrigel was mixed with 3 × 10⁶ BE(2)-MEIS2A or BE(2)-MEIS2D cells in a 1:1 ratio in 100 μL and the total volume was injected subcutaneously. Each mouse was transplanted in the right and left flank with the same cell line. Transgene expression was induced by feeding with doxycycline-containing pellets (625 mg/kg doxycycline) 36 hours after implantation to allow both cell populations to initiate tumor growth at comparable rates. Animals were inspected daily and tumor growth was measured every second or third day with a sliding caliper. Mice were euthanized at day 33 or according to the termination criteria defined in V54-19c20/15-FK/1063. Tumors were dissected and incubated in 4% PFA at 4°C. Tumors were embedded in paraffin and 3-μm-thick sections were cut and analyzed by automated IHC staining.

MEIS2 mRNA expression correlation with survival probability was evaluated by Kaplan–Meier analysis using the log-rank test (18). Gene mRNA expression correlations with neuroblastoma clinical and genetic features were determined using the nonparametric Kruskal–Wallis test. Other statistical significance tests were performed with the GraphPad Prism software 5.0b by unpaired or paired, two-tailed Student t test, as indicated. Comparison between three or more groups was carried out by two-way ANOVA followed by Bonferroni Multiple Comparison post hoc test. Normal distribution of the data was determined by Shapiro Wilk tests. Statistical significance was assumed when P < 0.05, with *, P < 0.05; **, P < 0.01; ***, P < 0.001. Unless noted otherwise, data were presented as mean ± SEM. Protein bands in Western blots were quantified by ImageJ and normalized to protein bands corresponding to α-tubulin obtained by reprobing the same blot.

Please see Supplementary Information.

We analyzed the genome-wide mRNA expression profiles of all publicly available neuroblastoma datasets using the web-based R2 Genomics Analysis and Visualization Platform (http://r2.amc.nl) and identified consistent MEIS2 expression patterns in 10 selected sets (Fig. 1A–D, Supplementary Fig. S1A and S1B). High MEIS2 mRNA expression significantly correlated with established predictors of good disease outcome, such as young age at diagnosis and low risk or relapse-free disease in all datasets for which clinical annotations were available. Taking the SEQC-498 dataset as example, MEIS2 expression was found to be prognostic for good survival (Fig. 1B, P = 4.4 e⁻³). In fact, even among patients suffering from st4 disease, survival probability positively correlated with high MEIS2 mRNA levels (Fig. 1C, P = 0.027). Upon categorization according to the five prognostic INSS stages, MEIS2 mRNA expression was lowest in high-risk st3 and st4 tumors, but highest in st4S tumors (Fig. 1D; Supplementary Fig. S1B). Importantly, MEIS2 mRNA levels were not correlated to the MYCN-status in any of the datasets (Fig. 1A). Differential MEIS2 expression is therefore not a consequence of MYCN amplification but rather a MYCN-independent predictor of disease outcome.

In paraffin-embedded sections of biopsies taken from st4 tumors involving MYCN amplification, MEIS2 immunoreactivity varied between samples with a significant portion of the signal detected in the cytoplasm (Fig. 1E–G). This contrasts the strong nuclear MEIS2 staining in MEIS2-expressing cells in the healthy brain, exemplified by the MEIS2-immunoreactive ependyma of the adult human forebrain (Fig. 1H). The reason for this cytoplasmic MEIS2 retention in high-risk neuroblastoma is unknown at present, but considering that transcriptional activity of any transcription factor requires its nuclear localization, the low prevalence of MEIS2 in the nuclei of st4 tumors suggests that weak MEIS2 activity is characteristic for aggressive neuroblastoma. In support of this, the tumor sample of a surviving st4 patient showed substantial nuclear MEIS2 staining, whereas MEIS2 staining was low and mostly cytoplasmic in the samples of two patients who had succumbed to the disease after relapse (Fig. 1E–I; Supplementary Fig. S1C). Yet, these clinical characteristics associated with high MEIS2 expression sharply contrast the recent report that MEIS2 possesses oncogenic properties in neuroblastoma, motivating us to investigate alternative splicing as possible mechanism underlying this discrepancy (11).

Alternative splicing generates MEIS2 isoforms with different C-termini. Skipping of exon 12 results in expression of a longer C-terminus, present in isoforms C and D, whereas inclusion of exon 12 introduces an early stop codon and generates the distinct, shorter C-terminus, found in isoforms A and B (Fig. 1J; Supplementary Fig. S2A and S2B). To investigate the prevalence of these isoforms in different neuroblastoma stages, we analyzed the two public neuroblastoma datasets that contained MEIS2 splice isoform-specific mapping data, Khan-47 (19) and TARGET-122 [the 122 neuroblastoma samples contained in the TARGET-161 dataset (http://r2.amc.nl); unpublished]. For the Khan-47 array study, we expressed the relative frequency of exon 12-containing transcripts in st1 and st4 tumors as ratio (st1:st4) and compared this ratio with that of “universal” probes corresponding to sequences found in all MEIS2 isoforms. Exon 12-specific sequences were significantly overrepresented in st1 tumors (Fig. 1K). Likewise, in the TARGET-122 RNA sequencing study, MEIS2A mRNA levels were significantly lower in st3 and st4 than in st4S tumors, whereas MEIS2D transcripts did not differ between these tumor stages (Fig. 1L and M). With the caveat that only two neuroblastoma datasets contained isoform-specific expression data, these results suggest a tumor-suppressive profile for MEIS2 isoform A.

To examine the functional relevance of MEIS2 expression for neuroblastoma growth and differentiation, we generated stable, tetracycline (Tet)-inducible MEIS2A- or MEIS2D-expressing SK-N-BE(2) clones [termed BE(2)-MEIS2A and BE(2)-MEIS2D from here on; refs. 14, 15]. Tet-responsive MEIS2A and MEIS2D carried a HA-tag for immunohistochemical detection. The SK-N-BE(2) cell line expresses low to medium levels of the MEIS2 splice isoforms A-D (Supplementary Fig. S3A; ref. 20). Doxycycline stimulation of BE(2)-MEIS2A and BE(2)-MEIS2D cells induced MEIS2A or MEIS2D expression within hours and at moderate levels, with a 4.4-fold and 8.4-fold increase in MEIS2A and MEIS2D protein, respectively, 48 hours after addition of doxycycline (Supplementary Fig. S3B–S3D). Interestingly, doxycycline-induced MEIS2A expression significantly compromised cell proliferation compared to BE(2)-MEIS2A cells analyzed without doxycycline treatment (Fig. 2A). Conversely, doxycycline-induced MEIS2D expression significantly enhanced proliferation relative to nontreated BE(2)-MEIS2D cells (Fig. 2B). Similar results were also obtained with two other MYCN-amplified neuroblastoma cell lines, Lan1 and SMS-KCN (Fig. 2C and D). In fact, stably transfected Lan1 cells ceased proliferation within days following doxycycline-induced MEIS2A expression (Fig. 2C). MEIS2A-transfected cells grew slower than MEIS2D-transfected cells even without doxycycline-induced transgene expression, likely reflecting a certain leakiness of the vector system. BrdUrd incorporation and the proportion of mitotic cells were also reduced in doxycycline-treated BE(2)-MEIS2A cultures and in stable, inducible MEIS2A-transfected SMS-KCN cells, but increased following doxycycline-induced upregulation of MEIS2D expression in BE(2)-MEIS2D cultures or in stable MEIS2D-SMS-KCN transfectants (Fig. 2E–J; Supplementary Fig. S4A and S4B). In addition, anchorage-independent growth, a readout for a tumorigenic phenotype, was significantly enhanced following MEIS2D, but not MEIS2A overexpression (Fig. 2K). To determine whether cell death contributed to the poor growth of BE(2)-MEIS2A cells, we evaluated activation of caspase-3, an indicator for apoptosis, and phosphatidylethanolamine-conjugation of LC3, LC3-II, a sign of autophagy. Neither increased numbers of cells that were immunoreactive for activated caspase-3 nor enhanced cleavage of procaspase-3 or a substantial increase in LC3-II were detected by Western blot analysis following doxycycline-stimulation of BE(2)-MEIS2A cells (Supplementary Fig. S4C–S4F). We therefore asked whether the decreased growth of BE(2)-MEIS2A cells may instead be due to increased replicative senescence. Indeed, cells staining positive for senescence-associated beta-galactosidase (β-gal) were abundant in BE(2)-MEIS2A cultures 5 days after doxycycline stimulation but virtually absent in doxycycline-treated BE(2)-MEIS2D cells or in the parental SK-N-BE(2) cell line (Fig. 2L–N; ref. 21). We concluded that elevated MEIS2A expression shifts neuroblastoma cells from proliferation to quiescence, whereas MEIS2D does not.

MEIS-type HD-proteins usually function as components of larger multiprotein complexes, yet it is presently unknown whether isoform-dependent differences in MEIS-interactomes exist (22). We isolated MEIS2A- and MEIS2D-containing protein complexes by immunoprecipitation with HA-specific antibodies from doxycycline-stimulated BE(2)-MEIS2A and BE(2)-MEIS2D cells and analyzed these by mass spectrometry. PBX1 and PBX2, established binding partners that contact MEIS family proteins through the N-terminal MEINOX-domain, copurified with both isoforms, but several proteins were selectively enriched in either the MEIS2A- or MEIS2D-specific immunoprecipitates in two independent experiments (Fig. 2O; ref. 22). Among the proteins exclusively coprecipitating with MEIS2A were GATA2, a zinc-finger transcription factor whose expression correlates with low risk disease in neuroblastoma, and CDK11B [cell division cycle 2-like 1 (CDC2L1)], a serine-threonine kinase involved in the regulation of cell proliferation and mitosis (Fig. 2O and P; refs. 23–25). Interestingly, the human CDK11B gene is located on chromosome 1p36.2–36.3, a region of frequent chromosomal deletion in neuroblastoma (26). MEIS2D coprecipitating proteins, by contrast, included several known oncoproteins, such as nucleophosmin (NPM1), nucleolin (NCL), DExD-box helicase 21 (DDX21, RH II/GU) and fibrillarin (FBL; Fig. 2O; refs. 27–30). Consistent with a role in aggressive neuroblastoma, expression of these genes is significantly enriched in st4 tumors (Fig. 2O and Q).

Because the different C-termini of MEIS2A and MEIS2D result from alternative splicing of exon 12, we devised siRNAs targeting this exon. Transfection of these siRNAs profoundly decreased expression of the exon 12-containing isoforms A and B without affecting transcript levels of isoforms C and D (Supplementary Fig. S5A–S5D). Notably, anchorage-dependent and -independent proliferation of SK-N-BE(2) cells were significantly increased following treatment with exon 12-specific siRNAs (Fig. 3A–E). Similar effects were also observed with the Lan1 and SMS-KCN neuroblastoma cell lines, demonstrating that the growth-promoting effect of MEIS2A-depletion was not selective for the SK-N-BE(2) cell line (Fig. 3F and G). Collectively these results establish a tumor-suppressive activity for MEIS2A in neuroblastoma with forced MEIS2A expression and knockdown, respectively, reducing and enhancing neuroblastoma cell proliferation.

The high MEIS2 expression in 4S tumors prompted us to examine whether neuronal differentiation, known to contribute to the spontaneous regression seen in 4S tumors, was also influenced by MEIS2. Five days after doxycycline stimulation, long, neurofilament M- (NFM) positive neurites were formed by BE(2)-MEIS2A cells, but not by BE(2)-MEIS2D or nontransfected SK-N-BE(2) cells (Fig. 3H–J). Thereby the proportion of NFM-immunoreactive cells and the average length of their processes were significantly increased in MEIS2A-overexpressing cultures (Fig. 3K and L). Elevated numbers of NFM-positive cells and formation of neurite-like structures were also seen 5 days after doxycycline-induced expression of MEIS2A in stably transfected Lan1 or SMS-KCN cells (Supplementary Fig. S6A–S6D). Although not further quantified, typical morphological features associated with neuronal differentiation, such as small round cell bodies and compacted cell nuclei, were also frequent in BE(2)-MEIS2A cultures (Fig. 3I). Transcript levels of microtubule-associated protein 2 (MAP2), a marker of differentiated neurons, and tyrosine hydroxylase (TH), the enzyme that catalyzes the rate-limiting step in catecholamine production, were elevated following MEIS2A overexpression compared with MEIS2D-overexpressing cells or SK-N-BE(2) control cells (Fig. 3M). Although TH expression in the sympathetic nervous system is not restricted to postmitotic neurons, TH expression levels correlate with the differentiation grade of neuroblastoma tumors (31, 32). MYCN expression, by contrast, was not altered when either MEIS2A or MEIS2D were overexpressed, strengthening the notion that MEIS2 modulates neuroblastoma cancer cell growth independently of MYCN (Fig. 3M).

We also used two additional vector systems, one based on the expression vector pcDNA3 and one based on the retroviral vector pBABE-Puro, to generate clones with constitutive MEIS2A overexpression. Irrespective of the vector system used, MEIS2A-transfected SK-N-BE(2) cells ceased proliferation and began to extend neurite-like processes, which grew very long when the cells were kept in culture for prolonged periods of time (Supplementary Fig. S6E–S6H). We also generated SK-N-BE(2) cell clones stably overexpressing MEIS2B, an isoform that shares the inclusion of exon 12 with MEIS2A but similar to MEIS2D lacks the N-terminal extension of exon 11 (Supplementary Fig. S6I). These cells exhibited a growth and differentiation behavior comparable with cells transfected with MEIS2A in principle (Supplementary Fig. S6J–S6L). These results argue that the antiproliferative activity of MEIS2 splice isoform A primarily resides within the short C-terminus and hence results from alternative inclusion of exon 12.

The phenotypes of MEIS2A transfection in neuroblastoma cells, cell-cycle arrest and neurite outgrowth, are remarkably similar to those observed following treatment of neuroblastoma cells with pharmacological doses of retinoic acid (RA; ref. 33). RA is a potent inducer of neuronal differentiation during development and can induce irreversible differentiation of neuroblastoma tumors when given in combination with chemotherapy and autologous hematopoietic stem cell transplantation (34). We therefore considered the possibility that MEIS2 may modulate the production of RA. We cotransfected SK-N-BE(2) cells with MEIS2A or MEIS2D expression vectors and a RA-dependent reporter vector in which expression of luciferase is under the control of three RA-response elements (RARE; ref. 16). Reporter activity was significantly enhanced upon MEIS2A but not MEIS2D overexpression (Fig. 4A). MEIS2A can, hence, induce a net-increase in cellular RA.

Mammals synthesize biologically active RA from dietary vitamin A, which involves a final conversion of retinaldehyde to RA by retinal dehydrogenases (ALDH). Transcript levels of ALDH1A2 were significantly elevated in MEIS2A-overexpressing SK-N-BE(2) cells but significantly reduced in MEIS2D-overexpressing cells (Fig. 4B). MEIS2A and MEIS2D thus act on ALDH1A2 expression in opposing ways. MEIS2A and MEIS2D also differentially affected mRNA levels of the RA-metabolizing enzyme CYP26A1, with MEIS2A, but not MEIS2D, reducing CYP26A1 expression (Fig. 4B). Transcript levels of other RA synthesis pathway genes, such as ALDH1L1 and ADH5, were not altered, as was expression of PBX1 (Fig. 4B). In the embryonic mouse hindbrain, direct regulation of ALDH1A2/RALDH2 by MEIS2 through a binding site in an intronic enhancer of the RALDH2 gene controls spatiotemporal RA production and correct rhombomere segmentation (35). This intronic enhancer is conserved in the human ALDH1A2 gene (Fig. 4C). Chromatin immunoprecipitation with MEIS-specific antibodies followed by quantitative real-time PCR (ChIP-qPCR) in SK-N-BE(2) cells revealed significant association of MEIS2 with the ALDH1A2 intronic enhancer, supporting the notion that ALDH1A2 is under direct control of MEIS2, not only in the mouse embryonic hindbrain but also in SK-N-BE(2) cells (Fig. 4D).

To examine whether the splice isoform-dependent regulation of cell proliferation is recapitulated in vivo, BE(2)-MEIS2A and BE(2)-MEIS2D cells were tested for their capacity to form tumors in two animal models. In the chick chorioallantoic membrane (CAM) tumor model, only 7 out of 10 embryos inoculated with BE(2)-MEIS2A cells developed tumors. By contrast, tumors efficiently formed from BE(2)-MEIS2D cells, which were also significantly larger than tumors grown from BE(2)-MEIS2A cells (Supplementary Fig. S7A–S7G). We also implanted BE(2)-MEIS2A and BE(2)-MEIS2D cells subcutaneously into the flanks of NU/NU mice. BE(2)-MEIS2D–derived tumors grew exponentially and reached an average tumor volume of 809.7 ± 242.5 mm³ by day 33 after implantation, whereas BE(2)-MEIS2A tumors remained at a relatively constant volume of 73.3 ± 18.6 mm³ (Fig. 5A–D). BE(2)-MEIS2D cells also gave rise to large, heavily vascularized tumors with frequent signs of hemorrhagic bleeding, central areas of necrosis and a fibrovascular, collagenous tumor stroma, whereas BE(2)-MEIS2A cells produced only small, nonvascularized tumor nodules (Fig. 5E–L). Finally, we took advantage of the recently published CSTN collection of O-PDX models for pediatric solid tumors to compare MEIS2 exon 12 inclusion in primary neuroblastoma biopsies with the ability of the respective samples to form tumors following orthotopic implantation in mice (13). Of the 41 primary neuroblastoma tumors that were orthotopically injected in this study, 9 engrafted successfully, of which 4 tumors exhibited substantial MEIS2 expression as determined by RNA sequencing. Interestingly, the time necessary for engraftment positively correlated by trend with the relative frequency of MEIS2 exon 12-inclusion in the tumor sample (Fig. 5M). With the caveat that only four samples possessed sufficiently high MEIS2 transcript levels to be included into the analysis, these results support the notion that MEIS2 alternative splicing is a critical factor contributing to neuroblastoma tumor aggressiveness, not only in a subcutaneous but also in an orthotopic tumor model.

Neuroblastoma is a devastating malignant disease. Although childhood cancer mortality from neuroblastoma has been declining during the last decade, the harsh multimodal therapy currently used to treat high-risks neuroblastoma has severe side effects for most surviving patients (36, 37). There is therefore great need for more effective, less harmful therapies. The spontaneous regression of metastatic disease seen in st4S neuroblastoma raises hope that neuroblastoma may be particularly amenable to treatment strategies that aim at inducing cellular differentiation. We show here that the TALE-HD protein MEIS2 is prominently expressed in st4S neuroblastoma and prognostic for good survival, even in high-risk disease. In addition, we provide experimental evidence that an MEIS2 splicing switch controls the balance between aggressive growth and cellular differentiation in neuroblastoma.

The existence of Meis2 splicing isoforms that differ in their C-termini was already noted when the gene was first cloned from mouse, but reports on functional differences associated with alternative Meis2 splicing remained sparse (38). Here, we establish that different MEIS2 splice isoforms can, in fact, exert antagonistic functions. Specifically, our work shows that even modest MEIS2A overexpression impairs anchorage-dependent and -independent growth, induces cellular senescence and initiates neuronal differentiation in neuroblastoma cell lines in vitro and halts neuroblastoma tumor growth in two xenograft models in vivo. By contrast, isoform-specific knockdown of MEIS2A or moderate overexpression of MEIS2D increased the aggressiveness of several st4 neuroblastoma cell lines. Because cells transfected with MEIS2B, an isoform that shares the short C-terminus of MEIS2A, largely mimicked the antiproliferative behavior of MEIS2A-transfected cells, the oncogenic or tumor suppressive properties of MEIS2A and MEIS2D have to reside in their different C-terminal domains. Mass spectrometry identified several proteins that specifically coprecipitated with MEIS2A or MEIS2D, suggesting that functional differences may arise through differential recruitment of cooperating proteins to the alternative MEIS2 C-terminal domains. Among the proteins exclusively found in the MEIS2A-precipitate were GATA2 and CDK11B, which are both linked to good disease outcome in neuroblastoma (23, 26), whereas the MEIS2D-precipitate included the oncoproteins NPM1, NCL, FBL and DDX21 (27–30). Interestingly, NPM1 and NCL have known functions in centrosome duplication and depletion of either gene leads to mitotic defects and ultimately aneuploidy (39, 40). Our results thus reveal an unexpected link between MEIS2D-interacting proteins and genome homeostasis during M-phase. This is notable in light of the recent report that shRNA-mediated depletion of all MEIS2 isoforms in SK-N-BE(2)-C cells causes cell-cycle arrest in M-phase with multipolar mitotic spindles and centrosome amplification (11). These defects were ascribed to dysregulated expression of RBBP4-BMYB-FOXM1, known regulators of M-phase progression, and proposed to reflect an oncogenic function of MEIS2 in neuroblastoma. Here we show that MEIS2 can have either oncogenic or tumor suppressing properties in neuroblastoma, depending on the prevailing splice isoform. In addition, we uncovered an additional mechanism by which MEIS2 knockdown may lead to mitotic failure, namely through imbalanced centrosome duplication resulting from dissociation of an MEIS2D-NPM1-NCL containing protein complex.

It is presently unknown how MEIS2 alternative splicing is regulated in neuroblastoma. Massive parallel RNA sequencing of neuroblastoma tumor samples identified a set of RNA-binding proteins, which regulate a splicing signature that correlates with neuroblastoma cancer hallmarks (41). Several of these splicing factors are direct targets of MYCN and expression of some of them is significantly associated with poor survival of st4 neuroblastoma patients (41). Whether MEIS2 is subject to alternative splicing by these MYCN-regulated factors remains to be investigated, but the consistent lack of correlation between MEIS2 and MYCN expression seen in 10 neuroblastoma gene-expression datasets argues that MEIS2 acts independently of MYCN.

MEIS2A overexpression elevated expression of the RA-producing enzyme ALDH1A2. Treatment with pharmacological doses of RA, specifically with all-trans RA (ATRA) and its isomer 13-cis RA, can induce cell-cycle arrest and neurite outgrowth in neuroblastoma cell lines and lead to irreversible differentiation of neuroblastoma tumors in children with high-risk disease (33, 34, 42). Nevertheless, ALDH1A2 activity seems to be a double-edged sword in neuroblastoma, as its expression is particularly high in aggressive, sphere-forming cells, suggesting that elevated ALDH1A2 activity may confer cancer stem cell properties in neuroblastoma (43). However, the patient-derived xenograft cell lines established in that study expressed ALDH1A2 at levels exceeding those seen here after MEIS2A overexpression by several orders of magnitude (43). Because excessive ALDH1A2 expression facilitates clearance of xenobiotics, extreme upregulation of ALDH1A2 may confer acquired drug resistance, while increased RA production may in fact require only a moderate elevation of cellular ALDH1A2 activity (44, 45). In this context it is important to consider that RA initiates DNA demethylation events and epigenetic chromatin changes during stem cell- and neuroblastoma differentiation (46). The subtle increase in ALDH1A2 mRNA and the resulting mildly elevated cellular RA content seen here in response to mildly elevated MEIS2A expression may therefore be perfectly sufficient to initiate long lasting antiproliferative effects in neuroblastoma.

RA downregulates the expression of ALDH1A2 and increases expression of the RA-degrading enzyme CYP26A1, so that under physiological conditions increasing cellular RA levels are immediately answered by reduced RA production and enhanced RA clearance (47). As we show here, MEIS2A overexpression simultaneously elevated ALDH1A2 expression and decreased that of CYP26A1. CYP26A1 transcript levels in SK-N-BE(2) cells rise steeply following RA administration, demonstrating that the normal RA feedback regulation is still intact in this cell line (48). Consequently, the drop in CYP26A1 transcripts seen in response to MEIS2A overexpression must result from transcriptional repression by MEIS2A, directly or indirectly through downstream effectors. RA-treatment of neuroblastoma cell lines causes replicative senescence, suggesting that the MEIS2A-induced senescence may result from cell intrinsic production of RA (49). Together, the dual regulation of RA synthesizing and catabolizing enzymes may allow MEIS2A to uncouple the negative feedback regulatory loop that normally restricts accumulation of cellular RA and thereby initiate RA-mediated differentiation- and maturation programs in neuroblastoma.

All cell lines used in this study, SK-N-BE(2), Lan1, and SMS-KCN were derived from patients suffering from st4 neuroblastoma with massive MYCN amplification (50). Considering these highly malignant features, it was astonishing to notice that moderate overexpression of MEIS2A was sufficient to force the cells into growth arrest, replicative senescence, and neuronal differentiation, de facto abolishing the effects of the strong oncogenic driver MYCN. Conversely, siRNA-mediated isoform-specific knockdown targeting MEIS2 exon 12 enhanced the aggressive growth of several MYCN-amplified st4 neuroblastoma cell lines. These differences are reflected in the growth of MEIS2A- and MEIS2D-overexpressing tumors in vivo, as only BE(2)-MEIS2D cells gave rise to aggressive, anaplastic tumors with central necrosis and a fibrovascular stroma, the latter being associated with malignant progression and aggressive subtypes in various cancers, including neuroblastoma (51). Considering that all four MEIS2 isoforms are present in neuroblastoma cells, a delicate balance between individual isoforms seems to exist, which when overbalanced affects disease outcome. Consistent with this notion, a higher relative frequency of MEIS2 exon 12 inclusion in primary neuroblastoma patient biopsies correlated with a longer latency before tumor development in an orthotopic PDX model. Modulating the MEIS2 splicing process towards the production of MEIS2A may thus offer new therapeutic approaches for neuroblastoma.

Taken together, the present study establishes MEIS2 as MYCN-independent predictor of disease outcome in neuroblastoma and reveals previously unrecognized antagonistic functions of the MEIS2A and MEIS2D splice isoforms in tumor growth and progression, thereby reconciling some of the contradictory views that currently exist on the role of MEIS in cancer. In addition, because alternative splicing of MEIS2 and MEIS1 is similar in principle and given that MEIS1 has been implicated in different malignancies, including leukemia, our findings will likely have implications for cancer diseases that go beyond neuroblastoma (52).

No potential conflicts of interest were disclosed.

Conception and design: D. Schulte

Development of methodology: A. Groß, C. Schulz, H. Rohrer, D. Geerts, D. Schulte

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Groß, J. Kolb, S. Wehner, S. Czaplinski, A. Khilan, P.N. Harter, T. Klingebiel, J.D. Langer, D. Geerts

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Groß, C. Schulz, J. Koster, S. Wehner, A. Khilan, P.N. Harter, J.D. Langer, D. Geerts

Writing, review, and/or revision of the manuscript: A. Groß, J. Koster, S. Wehner, S. Czaplinski, H. Rohrer, T. Klingebiel, D. Geerts, D. Schulte

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Wehner

Study supervision: D. Schulte

We are grateful to M. Anders-Maurer for excellent technical assistance, S. Thom for help with the animal experiments, to M.J. Ausserlechner, A. Buchberg, J. Cinatl, S. Fulda, E. Ilina, V.M. Lee, and M. Mittelbronn for reagents, and to M.A. Dyer and B. Xu for making O-PDX–derived sequencing data available through the Childhood Solid Tumor Network. The work was supported by grants 2012/072.1 and 2012/072.2 from the Wilhelm Sander Stiftung für Krebsforschung (to D. Schulte). A. Groß was a predoctoral fellow supported by the Wilhelm Sander Stiftung, J. Kolb was supported by a Ludwig Edinger Fellowship.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.