PBX1 Is a Favorable Prognostic Biomarker as It Modulates 13-cis Retinoic Acid–Mediated Differentiation in Neuroblastoma

This article is featured in Highlights of This Issue, p. 4171

Neuroblastoma is the most common extracranial pediatric solid tumor, representing 8% of childhood cancer diagnoses but 15% of childhood cancer-related deaths (1). The diverse clinical spectrum of neuroblastoma presents numerous therapeutic challenges. Infant patients commonly present with low-grade disease, often with spontaneous regression, and some patients are managed expectantly with observation never requiring surgery (2). However, there are no validated biomarkers to predict which patients can be safely observed in contrast to those who would benefit from surgery and chemotherapy.

In contrast, >40% of patients present at diagnosis with aggressive disease and distant metastases (1). Patients with high-risk disease receive multiagent chemotherapy, surgery, high-dose chemotherapy with autologous hematopoietic stem cell rescue (HDC-aHSCR), radiotherapy, immunotherapy with chimeric anti-GD2 antibody ch14.18, IL2, and GM-CSF, and the differentiation agent 13-cis retinoic acid (13-cisRA). Despite intensive treatment and improved responses with the introduction of immunotherapy (3), up to 40% of high-risk patient tumors progress during induction chemotherapy (4) and 5-year disease-free survival remains <50% (5). Identification of MYCN amplification as a prognostic biomarker of aggressive disease radically improved risk stratification and treatment of neuroblastoma (6). Since the seminal discovery of MYCN, additional clinical and genetic biomarkers have been identified, but connection of these factors to neuroblastoma biology or tumor aggression remains cryptic, particularly in tumors without MYCN amplification (7, 8).

Neuroblastoma is a prototypic embryonic cancer with demonstrated aberrations in normal developmental pathways (9, 10). The HOX genes are master regulators of development in animals; select HOX gene expression, including HOXC6 and HOXC9, has been associated with neuroblastoma differentiation, response to 13-cisRA, and outcome (11, 12). HOX proteins critically interact with cofactors, including members of the three–amino-acid loop extension (TALE) gene family. The TALE family genes (PBX1-4, MEIS1-3, and PKOX1-2), like HOX genes, are critical to tissue differentiation (13), including retinoid-induced differentiation (14). Expression of PBX1-3 is regulated by retinoids in other cell types (15) and has been shown to direct endogenous retinoid synthesis within the nervous system (16). A comprehensive study of the expression of all TALE gene family members has not been previously performed in neuroblastoma.

The functions of TALE family genes are temporospatially specific, with the same gene often having divergent functions in different tissues. For example, PBX1 is necessary for normal pancreatic development (17) and behaves as a tumor suppressor in prostate cancer (18). In contrast, PBX1 is implicated as an oncogene in breast cancer (19) and melanoma (20), and PBX1 is oncogenic in leukemia as part of the E2A–PBX1 fusion protein (21). The functional complexity is increased by gene paralogs (e.g., PBX1-4), which can have distinct or overlapping functions within a given tissue. These varying but important roles make it critical to evaluate these genes specifically in neuroblastoma to define their roles in oncogenesis.

Here, we perform the first comprehensive analysis of TALE family gene expression in neuroblastoma. We demonstrate that, among TALE family genes, only PBX1 expression is associated with responsiveness to 13-cisRA in neuroblastoma. We confirm that PBX1 induces neuronal differentiation and increases sensitivity to 13-cisRA. Furthermore, we show that PBX1 expression is directly associated with decreased proliferation independently of 13-cisRA. Finally, we demonstrate that PBX1 expression in primary human tumors is associated with low tumor grade and patient survival. PBX1 expression may thus serve as a biomarker in low-risk disease by identifying patients who may be observed without intervention. In high-risk disease, PBX1 may stratify those patients for whom current therapies are ineffective, directing them to novel therapies.

Cell lines SK-N-SH, LAN-5, IMR-32, SK-N-BE (2), and SK-N-RA were obtained from Javed Khan; SMS-KAN and SMS-KANR from Joanna Kitlinska [via Children's Oncology Group (COG) MTA with Georgetown University, Washington, D.C.]; NBL-WS and LAI-5S from Susan Cohn (University of Chicago, Chicago, IL); SHSY5Y from ATCC; CHLA-15, CHLA-42, CHLA-90, CHLA-136, and LAN-6 from COG Cell Culture Repository; and HEK293 T from OpenBioSystems. All were previously characterized (22–28). A table of the MYCN status of each cell line is in the Supplementary Methods. CHLA-15, CHLA-42, CHLA-136, and CHLA-90 were grown in Iscove's modified Dulbecco's medium with 20% FBS (Hyclone) and 0.1% insulin, transferrin, and selenium (Corning). All other cell lines were grown in RPMI with 10% FBS. All parental and modulated cell lines were tested and authenticated by PowerPlex16 short tandem repeat (STR) analysis (Promega) by the Nucleic Acids Core laboratory at Nationwide Children's Hospital (NCH), last in December 2013, and January 2014. Note that the SK-N-RA cell line was initially labeled as SK-N-AS when received, but has been characterized by STR analysis to be the SK-N-RA cell line, including aliquots of the original cell line received from the Khan lab, all generated modulated cell lines, and the aliquots of the cell lines at the end of the studies. Dr. Khan's lab has confirmed that the cell line they are using labeled SK-N-AS is, by STR, truly SK-N-AS, so we can only surmise our cell line was mislabeled, but it has been characterized to be consistent with SK-N-RA as published elsewhere. Although we performed the studies under the supposition that the cells were of SK-N-AS origin, the cells were always of SK-N-RA origin, and the presented data have been interpreted accordingly.

PBX1 cDNA plasmid (MHS1768-101376233), PBX1-shRNA set (RHS4531-EG5087), and nonsilencing negative control were purchased from OpenBioSystems. PBX1 cDNA was cloned into pCDNA3.1+/Hygro (LifeTechnologies) or pLPCX (Clontech) and transfected using Lipofectamine 2000 (LifeTechnologies). shRNA plasmids were transfected into HEK293 T cells with pHR-8.2ΔR and pVSVG packaging vectors. Generated viral supernatant was applied to neuroblastoma cells. Cells were selected with hygromycin (200 μg/mL) or puromycin (1–7 μg/mL) for 2 weeks, then retreated monthly.

Primary ganglioneuromas (n = 7) and neuroblastomas (low-risk, n = 11; intermediate-risk, n = 5) were obtained from Johns Hopkins University (JHU) pathology archives under Institutional Review Board (IRB) exemption and waiver of consent. Patient and sample characteristics and tissue and data management are fully described in Supplementary Methods. RNA from high-risk neuroblastoma (International Neuroblastoma Staging System, INSS, criteria, n = 40) was obtained from the COG Neuroblastoma Tumor Bank, 20 from patients alive without disease progression, and 20 from patients died of disease. Patient and treatment characteristics are described in Supplementary Methods. COG samples were obtained after informed consent and IRB approval. Protected Health Information was sequestered by COG and JHU, and IRB exemption granted by Georgetown University and NCH.

Twenty-four hours after seeding in complete media, cell lines were cultured in media with 10 μmol/L 13-cisRA (Sigma-Aldrich) or 0.1% DMSO. At 50% to 75% confluence, 100 cells were counted, in triplicate, and assessed for neurite extension, defined as neurite length ≥ soma length (29). For RNA and protein analysis, cells were cultured for 7 days and then collected in PBS; RNA was extracted as above, and protein was extracted with RIPA buffer.

A total of 5,000 cells per well were plated in 96-well plates in complete media, in triplicate, on 5 plates each. After 24 hours, one plate was treated with WST-1, 10%v/v (Roche) for 1 hour, then absorbance at 450 nm (670 nm reference) measured, and each triplicate was averaged. These values were used to normalize cell count and set as time 0. Media were changed on the remaining plates to complete media with 0.1% DMSO or 10 μmol/L 13-cisRA. Every 48 hours, one plate was treated with WST-1 and absorbance measured as above, then normalized for each derived cell line to time 0 values to calculate percent viability. Normalized absorbances were compared with control cells for each cell line set. Experiments were performed three times.

Of note, 2 μg (cell lines) or 200 ng (primary tumors) of RNA were used in cDNA synthesis by Superscript VILO (LifeTechnologies) as per the manufacturer's protocol. Yields were diluted per protocol and 2 μL of resulting product used in qPCR, using the Eppendorf RealPlex Mastercycler and KiCqStart Mastermix (Sigma Aldrich), as per the manufacturer's protocols. qPCR was performed for 40 cycles, and T_m analysis used to confirm purity of PCR amplification. Primers are listed in Supplementary Methods. Relative gene expression was calculated using the ΔΔC_t method, using 18SRNA as reference, and samples compared with SHSY5Y cells grown in complete media for normalization. Samples were tested in triplicate on three separate experiments.

Western Blots were performed with 50 μg of cell lysate from each sample electrophoresed through 4% to 12% Bis-Tris Bolt gels (LifeTechnologies), then transferred onto polyvinylidene difluoride membranes. Antibodies and full protocols are listed in Supplementary Methods.

Cells were grown in complete media with DMSO or 10 μmol/L 13-cisRA for 5 days, then passaged onto CultureSlides (BDBiosciences) and grown for another 2 days. TUBB3 was evaluated by immunofluorescence as per the manufacturer's protocol (Cell Signaling). Cells were counterstained with Prolong AntiFade Mounting Media with DAPI (LifeTechnologies), and imaged under ×40 magnification.

A total of 5,000 cells per well were suspended in 0.3% low melting point Agarose/RPMI Media +10% FBS in 6-well plates. Cells were plated in triplicate and cultured for 4 weeks, then stained with 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) solution. Wells were photographed and colonies counted.

Experiments were analyzed using GraphPad Prism6 software, except survival analyses of the expression datasets (30–32), described below. Data were tested by the Student t test, one-way ANOVA, or two-way ANOVA as indicated.

Characteristics of the three published datasets have been previously described fully, including sample inclusion, clinical characteristics, and analysis methods (30–32). Survival data for Khan and Seeger datasets were obtained from the Oncogenomics website, annotated by the Oncogenomics section, Pediatric Oncology Branch of the NCI. Survival data for the Oberthuer dataset were obtained from the Journal of Clinical Oncology, directly annotated by the primary researchers, including patient and outcomes data. The primary objective of survival studies was to investigate PBX1 expression association with survival, in the presence of other factors. Event-free survival (EFS) and overall survival (OS) were examined. Data were right-censored as per the patient status. INSS stage was grouped into elementary (stages 1, 2, and 4S) or advanced (stages 3 and 4). Kaplan–Meier plots were generated to compare EFS and OS based on PBX1 expression, classified into “high” and “low” groups. According to a priori knowledge and univariate analysis, five factors—PBX1 expression, histology, age at diagnosis, INSS stage, and MYCN amplification—were chosen for regression analysis with the Cox proportional hazards model. Statistics were calculated by the R Statistical Analysis package.

We evaluated PBX1 expression in neuroblastoma cell lines in response to 13-cisRA, utilizing 6 RA-sensitive and 4 RA-resistant cell lines. RA resistance was defined as no morphologic differentiation or apoptosis after treatment with 10 μmol/L 13-cisRA (Supplementary Fig. S1A). Basal PBX1 expression was variable across cell lines (Figs. 1 and 2), with lowest expression in RA-resistant CHLA-15 cells. Notably, after treatment with 13-cisRA, RA-sensitive cell lines all had significant increases in PBX1 expression, from 2.2-fold to 6-fold increase (Fig. 1B), whereas RA-resistant cell lines had no significant increase in PBX1 expression (P = 0.014). We evaluated the expression of other TALE family genes, and none were consistently altered in association with RA sensitivity (Supplementary Fig. S1B). These findings suggested that PBX1 is unique among TALE family genes in response to 13-cisRA in neuroblastoma.

The PBX1 mRNA changes were corroborated by protein expression. We treated 14 neuroblastoma cell lines (4 RA-resistant and 10 RA-sensitive) with either 13-cisRA or DMSO for 5 days, then performed Western blots for PBX1. We also evaluated MYC and MYCN expression, which are downregulated in RA-sensitive cells by 13-cisRA (33, 34), and NTRK1 and NTRK2, which are upregulated in response to 13-cisRA (35). Expression of these proteins in primary tumors correlates with clinical outcomes (36, 37). We confirmed the increase of PBX1 expression, observed as isoforms PBX1a and PBX1b, in RA-sensitive cell lines treated with 13-cisRA, and associated with changes in MYCN, MYC, NTRK1, and/or NTRK2 (Fig. 2A–C). PBX1 expression did not increase in RA-resistant cell lines (Fig. 2D).

We studied the effects of PBX1 in neuroblastoma and RA-mediated differentiation using four cell lines: RA-resistant, MYCN-nonamplified SK-N-RA cells, the RA-sensitive, MYCN-nonamplified cell lines SK-N-SH and SHSY5Y (previously derived as an N-type subclone of the parental SK-N-SH), and the RA-sensitive, MYCN-amplified SK-N-BE (2) cells. We specifically used these cell lines to model induction of PBX1 expression, mimicked by exogenous expression, or blockade of that induction by use of shRNAs, using empty vector–transfected or nonsilencing shRNA-infected cells as controls. We measured PBX1 mRNA levels in the generated cell lines by qRT-PCR, and then confirmed protein expression changes by Western blot (Fig. 3). We reduced PBX1 expression to <20% of control in all cell lines (Fig. 3, shPBX1 #3–#5). We increased PBX1 expression in SK-N-RA cells to 10- to 35-fold of control (Fig. 3D), but only 3- to 8-fold of control in the RA-sensitive cell lines, as measured by qRT-PCR (Fig. 3A–C), comparable with levels observed in parental cell lines treated with 13-cisRA (Fig. 1A)

We used the stably-derived neuroblastoma cell lines to assess the effects of the induction or blockade of PBX1 expression upon neuroblastoma differentiation (Fig. 3). The expression of the neuronal differentiation marker TUBB3 (29) and of retinoid-induced differentiation markers MYCN, MYC, NTRK1, and NTRK2 was evaluated in the cells grown in complete media with 13-cisRA or vehicle. In all cell line sets, without 13-cisRA, PBX1 expression correlated with increased expression of either NTRK1 or NTRK2 compared with controls, and decreased expression of MYCN or MYC (Fig. 3, immunoblots), and correlated with increased expression of TUBB3 in the SHSY5Y and SK-N-RA sets. When treated with 13-cisRA, these expression changes were amplified in RA-sensitive cells with increased PBX1 expression (Fig. 3A–C). SK-N-RA cells with exogenous PBX1 expression also had increased protein levels of TUBB3 and NTRK2, but not NTRK1, compared with treated control cells (Fig. 3D, PBX1 lanes). In all cell lines, reduction of PBX1 expression reduced NTRK1 and NTRK2 expression, and, to a lesser extent, TUBB3 expression. Furthermore, blockade of RA-induced increase in PBX1 expression by shRNA prevented increased expression of these proteins when treated with 13-cisRA.

Neuroblastoma differentiation is also characterized by morphologic changes, including neurite extension and/or cell lengthening (29). Microscopy demonstrated increased PBX1 expression was associated with increased neurite extension and cell lengthening, similar to control cells treated with 13-cisRA. In the SK-N-SH vector control cells, <20% cells grow neurite extensions in complete media with vehicle (Fig. 4). Neurite outgrowth is seen after 7 days of treatment with 10 μmol/L 13-cisRA in 76% of control cells. In SK-N-SH cells with 2- to 3-fold increased PBX1 levels, neurite extensions spontaneously develop in 53% of cells without 13-cisRA, and 90% of cells with 13-cisRA. Strikingly, clonal populations with higher PBX1 expression (>5× control cells) formed spindle-shaped cells with long neurites (Fig. 4A, third row) and virtually no proliferation (Fig. 5), without exogenous 13-cisRA. However, they remain fully viable in culture for >8 months, able to readhere after trypsinization and extend neurites without any appreciable mitosis. This phenotype is consistent with a state of terminal differentiation. Of note, <10% of cells with repressed PBX1 expression, in contrast, grew neurite extensions even when treated with 13-cisRA (Fig. 4A, fourth and fifth rows). Similar results were observed in SHSY5Y cells and SK-N-BE (2) cells (Fig. 4B and Supplementary Fig. S2A and S2B).

TUBB3 is a neuron-specific cytoskeletal component (38). In differentiating neuroblasts, it relocates from a generalized cytoplasmic distribution to within elongating neurites. In the three cell lines tested, immunofluorescence for TUBB3 showed a direct association of expression and relocalization with increased PBX1 expression. In SK-N-SH and SHSY5Y cells, exogenous PBX1 expression caused relocalization of TUBB3 along neurites in the absence of 13-cisRA, similar to control cells treated with 13-cisRA (Fig. 4C and Supplementary Fig. S2C and S2D). Cells with repressed PBX1 expression had decreased or absent TUBB3 expression and no relocalization. Though SK-N-RA cells did not show significant morphologic changes with PBX1 expression changes, cells with increased PBX1 levels also had increased TUBB3 expression by immunofluorescence (Supplementary Fig. S2E).

We assessed the effects of PBX1 on neuroblastoma cell proliferation in monolayer culture. Among the four cell lines, cells with increased PBX1 expression proliferated significantly more slowly than vector control over 96 to 192 hours (Fig. 5; P < 0.01 or less for each cell line). In SK-N-SH cells, clonal populations with >5-fold increase in PBX1 expression demonstrated virtually no proliferation (Fig. 5A; PBX1 Clone 1; P < 0.001 vs. vector control). Cells with decreased PBX1 expression proliferated significantly faster than control (P < 0.05 for each cell line). We noted that rates of proliferation were stable or increased over time in cell lines with reduced PBX1 expression, despite increasing confluency, and decreased with time in cell lines with increased PBX1 expression despite subconfluency (Supplementary Fig. S3).

We originally observed that increased PBX1 expression in neuroblastoma cell lines correlated with RA sensitivity (Figs. 1 and 2). Accordingly, we hypothesized that PBX1 expression was not only induced by 13-cisRA but also critical to responsiveness of the cells to RA. Indeed, we found that repression of PBX1 expression induced resistance to 13-cisRA, with increased proliferation compared with vector control (Fig. 5A, P < 0.01 for all cell lines). Increased PBX1 protein levels significantly increased RA sensitivity in all cell lines. That effect was greater and dose-dependent in natively RA-sensitive cell lines (SK-N-RA, P < 0.05; SK-N-SH, P < 0.05–0.001; SHSY5Y and SK-N-BE (2), P < 0.001). The SK-N-SH and SHSY5Y PBX1 clonal populations with highest PBX1 expression showed no significant proliferation after 8 days of 13-cisRA treatment (P < 0.001). These results show that PBX1 affects differentiation directly and also mediates the effects of RA upon neuroblastoma cells.

We studied the effects of PBX1 expression on anchorage-independent growth in soft agar. In the three cell lines tested, increased PBX1 expression caused significant suppression of colony formation, and reduced PBX1 expression increased colony number (Fig. 5B and C; P < 0.001 for SK-N-RA cells; P < 0.0001 for SK-N-SH and SHSY5Y cells, vs. vector controls). Increased PBX expression caused near-total suppression of colony formation in SK-N-SH cells (Fig. 5B, second row, second column), again consistent with terminal differentiation.

Current neuroblastoma pretreatment risk stratification follows the guidelines established in the INSS (39) or the International Neuroblastoma Risk Group (INRG) Classification system (5). Both systems use anatomic, histologic, and molecular criteria to establish at diagnosis patient risk from disease, which is used to guide the treatment approach for each patient. We evaluated the correlation of PBX1 expression with current pretreatment risk classification and patient outcomes in primary human tumors and in three independent cohorts of human tumor data.

RNA was obtained from pretreatment primary tumors, including benign ganglioneuromas (n = 7), low-risk neuroblastoma without recurrence after surgery (LR-NB, n = 8), low-risk neuroblastoma treated with chemotherapy after recurrence (LRR; n = 3), intermediate-risk neuroblastoma cured with surgery and chemotherapy (IR-NB; n = 5), and high-risk neuroblastoma (HR-NB; n = 40). PBX1 expression, measured by qRT-PCR, was normalized to untreated SHSY5Y cells as a baseline (Fig. 6A and Supplementary Table S1). In these samples, PBX1 expression was associated significantly with risk group and outcome. Expression was highest in ganglioneuromas; most had 2-log fold higher expression than SHSY5Y cells. LR-NB samples had significantly lower expression than ganglioneuromas samples (P < 0.001), but 1-log fold higher than samples from patients who needed chemotherapy for definitive cure (LR-NB vs. IR-NB or LRR; P < 0.001). HR-NB samples had significantly lower expression than the ganglioneuromas or LR-NB groups (P < 0.001), but subset analyses showed no significant difference between survivors and nonsurvivors in MYCN-amplified HR-NB (Supplementary Fig. S4A; P = 0.12) or MYCN-nonamplified HR-NB (P = 0.22).

We next evaluated PBX1 as a prognostic marker in neuroblastoma in three independent gene expression datasets of clinical neuroblastoma samples, with matching outcome data. The first dataset (hereafter referred to as the “Oberthuer” dataset) included data from 251 primary tumors from patients across the clinical spectrum, including all stages and MYCN amplification status, treated on German Neuroblastoma clinical trials NB90-NB2004 (30). High PBX1 expression was significantly prognostic of OS and EFS in univariate analyses (Fig. 6B; OS, P = 0.0004; Supplementary Fig. S4B; EFS, P = 0.00016). Data on INSS criteria were available for the Oberthuer dataset. Multivariate analysis comparing PBX1 expression with INSS risk factors confirmed that high PBX1 expression is independently a favorable prognostic marker in these patients (EFS HR, 0.49; 95% confidence interval, CI, 0.28–0.87; P = 0.0144 and OS HR, 0.55; 95% CI, 0.31–0.96; P = 0.03653; Supplementary Table S2), and a stronger biomarker than most factors currently used in INRG risk classification.

The second dataset (the “Khan” dataset) consisted of primary tumors from patients presenting with low- and high-risk disease, including tumors with either MYCN amplification or nonamplification (31). The dataset included 30 patients <18 months of age and 19 patients >18 months of age; 12 samples were also included in the Oberthuer dataset (3 low-risk, 4 intermediate-risk, and 6 high-risk, per INSS). All patients received treatment similar to current approaches, including 13-cisRA but not immunotherapy, based on risk classification. In these patients, high PBX1 expression was prognostic of OS in univariate analysis (Fig. 6C, P = 0.00125). INSS classification data, including age at diagnosis, stage, and MYCN amplification, were available for these patients. In multivariate analysis, PBX1 was a favorable prognostic marker for OS that trended toward statistical significance (OS HR, 0.4639; 95% CI, 0.15–1.43; P = 0.18; Supplementary Table S3). Among patients <18 months of age, the 3 patients who died of disease had significantly lower PBX1 expression than the 27 patients who survived (Supplementary Fig. S4C; P = 0.005). Patients with high-risk, MYCN-nonamplified disease represent a group for which there are no validated biomarkers predictive of response to treatment. Among these patients in the Khan cohort, survivors had significantly higher PBX1 expression than nonsurvivors (Supplementary Fig.S4D; P = 0.0145).

The third dataset used for survival analysis (the “Seeger” dataset; ref. 32) consisted of 102 patients with metastatic, MYCN-nonamplified disease at diagnosis. These patients (intermediate-risk, n = 28; high-risk, n = 74) were treated with multiagent chemotherapy, including a subset of patients who also had HDC-aHSCR with or without 13-cisRA (n = 23). High PBX1 expression strongly correlated with OS in univariate analysis (Fig. 6D; P = 0.0002); annotated patient data were not available for multivariate analysis.

We evaluated if other TALE family gene expression correlated with clinical course. No other TALE gene significantly stratified outcome in all three databases in univariate analysis (Supplementary Table S4).

Neuroblastoma is a cancer with inherently aberrant differentiation (9, 40). This is the first comprehensive evaluation in neuroblastoma of the TALE gene family, known regulators of differentiation. In low-risk neuroblastoma, there are no validated biomarkers that identify patients who can be safely observed without surgery or those who require chemotherapy to prevent recurrence. Two challenges in high-risk patient management are (i) identification of patients with RA-resistant disease and therapeutic alternatives and (ii) identification and therapeutic targeting of mechanisms of disease aggression in MYCN-nonamplified disease. Our experimental findings combined with the key findings from three clinical datasets identify PBX1 as a potential physiologic node in both low-risk neuroblastoma and high-risk MYCN-nonamplified disease. As a network node, PBX1 would serve as a novel prognostic biomarker and therapeutic target.

Prior investigations into RA signaling in neuroblastoma have identified few clinically relevant biomarkers of RA resistance, including HOXC9 (12) and the NF1/ZNF423 axis (41). In contrast to those studies, our conclusions are supported by experiments in multiple human neuroblastoma cell lines and in multivariate analysis of three independent clinical datasets, supporting the role of PBX1 as a biomarker of disease severity across clinical presentations of neuroblastoma. More importantly, we show that PBX1 expression not only is induced by 13-cisRA in vitro, but also potentiates its effects, suggesting that therapeutic approaches to increase its expression in neuroblastoma would improve the efficacy of 13-cisRA.

We demonstrated that PBX1 is a critical component in neuroblastoma differentiation, unique among TALE family genes. In cell lines treated with 13-cisRA, PBX1 mRNA and protein expression was induced only in those cell lines sensitive to the effects of the drug, as demonstrated by associated morphologic and expression changes (Figs. 1 and 2). We did appreciate a discordance between mRNA and protein expression within a given cell line (e.g., SK-N-RA) suggesting additional posttranscriptional regulation of expression. Nonetheless, the pattern of induction of PBX1 expression was consistent between mRNA and protein, supporting a transcriptional induction in response to 13-cisRA

We modeled the effects of RA-mediated induction of PBX1 by modulating its expression in neuroblastoma cell lines. Our data showed that PBX1 expression correlated with cell growth and differentiation in RA-sensitive and -resistant cell lines. Specifically, reduced PBX1 expression increased proliferation and abrogated responsiveness to 13-cisRA in three RA-sensitive cell lines. Exogenous PBX1 expression, mimicking RA-induced increase of PBX1 levels, decreased proliferation and induced RA-independent differentiation in those cell lines. SK-N-SH cells differentiate with a markedly different phenotype compared with SHSY5Y and SK-N-BE (2) cells when exposed to 13-cisRA, although all are N-type cell lines (Fig. 4 and Supplementary Fig. S2). SK-N-SH cells differentiate into spindle-shaped cells that express neuronal proteins, in marked contrast to the subcloned cell line SHSY5Y. SHSY5Y and SK-N-BE (2) cells differentiate into neuronal cells with narrow neurites, typical of other N-type cells. SK-N-BE (2) cells are MYCN-amplified, in contrast to the other cell lines used. PBX1 induced differentiation similar to that induced by 13-cisRA in each cell line, suggesting that PBX1 acts through a common mechanism in neuroblastoma even with different biology.

We observed in SK-N-RA cells that increased PBX1 expression caused significant increases in TUBB3 expression, decreased proliferation, and increased RA sensitivity, though less profoundly as in the other cell lines. We further observed no significant change in NTRK1 expression in SK-N-RA due to PBX1, which may prevent differentiation (Fig. 3D). Thus, PBX1 may be necessary but not sufficient for terminal differentiation in neuroblastoma. A similar role for PBX1 in terminal differentiation has been observed in adipocytes (42) and osteoblasts (43). Investigations into how PBX1 functions in normal neuroblasts and neuroblastoma will elucidate the differentiation process in neuroblastoma, and how differentiation is avoided in RA-resistant neuroblastoma.

Neuroblastoma cells express two PBX1 isoforms, PBX1a and PBX1b. The longer PBX1a isoform was expressed more strongly qualitatively in most RA-sensitive cell lines, and increased with 13-cis RA, but not the RA-resistant cell lines. Differential functions of PBX isoforms are found in normal biology (44) and cancers (45, 46). PBX1b acts as a transcriptional activator through recruitment of chromatin remodeling proteins (47), including CREB binding protein, a histone acetyltransferase (44). PBX1b may act to maintain an undifferentiated phenotype in normal neuroblasts and neuroblastoma, as it is generally expressed in embryonic tissues (45). PBX1a, in contrast, acts as a transcriptional repressor through recruitment of complexes including NCOR2 and HDACs (44). Increased differential PBX1a expression may drive differentiation, consistent with its expression in mature tissues (45). Studies into PBX1 isoform functions and the regulation of differential PBX1 splicing will be important in understanding their roles in neuroblast development. PBX1 paralogs PBX2-4 share homology and can also be induced by retinoids in other tissues (15), so studies into their functions will further define their effects on neuroblastoma differentiation. For example, these paralogs could functionally replace or competitively inhibit PBX1 in neuroblastoma differentiation.

Our findings suggest that PBX1 acts to inhibit proliferation in neuroblastoma and may induce spontaneous differentiation in patients with low-risk disease. This is substantiated by the significantly higher PBX1 expression levels in benign ganglioneuromas and low-risk neuroblastoma samples cured surgically, compared with low-risk and intermediate-risk tumors needing chemotherapy for definitive cure. Subgroup analysis of the Khan dataset similarly demonstrated significantly lower PBX1 expression in patients <18 months of age with recurrent disease compared with those cured surgically. In both sample sets, there is ≥30-fold difference in median expression between recurrent low-risk tumors and nonrecurrent tumors with no overlap in the range of expression between groups.

The INRG risk classification system defines “very-low” and “low-risk” patients expected to have >75% 5-year EFS after surgery alone (5). No biomarkers identify patients within those groups who do not need tumor resection for disease control, or those patients who will need adjuvant treatment. Our research indicates that PBX1 may serve as such a biomarker in low-risk neuroblastoma, despite the small sample number, particularly given the large difference in PBX1 expression. A larger study of PBX1 expression in low-risk disease is warranted to validate the threshold of stratification and correlate mRNA expression with protein expression measured by IHC. Ideally, after vigorous validation of our findings, PBX1 expression would be used as a biomarker of differentiation in patients risk-stratified as “low” or “very-low” risk by INRG criteria. Those patients' tumors would be tested for PBX1 expression, by qRT-PCR or IHC. Those with “high” expression would be managed with observation without surgery, and those with “low” expression would be recommended for surgery and adjuvant chemotherapy.

It is noteworthy that, unlike in neuroblastoma cell lines, “basal” PBX1 tumor expression correlates with tumor grade and treatment outcome. This may be due to the origination of most neuroblastoma cell lines, including all used in this study, from patients with stage 4 disease. Tumors from those patients would be expected to have low PBX1 expression, consistent with our analyses of primary human tumors. Furthermore, neuroblastoma inherently has perturbations in differentiation; we found that the degree of differentiation correlated with PBX1 expression in cell lines and tumors (i.e., the more PBX1 observed, the more differentiated the cells). Accordingly, it is not surprising that low-risk surgically-cured neuroblastoma and terminally differentiated ganglioneuromas have significantly higher PBX1 expression than more aggressive disease; differentiated tumors have responded to endogenous factors normally expressed in children, and more aggressive tumors have not. The lack of differential PBX1 expression in the subsets of high-risk disease may reflect the heterogeneity of the cohort, as they ranged from stage 2 to 4 and nonsurvivors progressed at various points in therapy. Alternately, PBX1 may exert its effects in tumor cells at a more differentiated state than in high-risk disease, or in an isoform-specific manner, further emphasizing our future studies into PBX1 isoforms. A pilot study examining only patients with recurrent disease, after receiving 13-cisRA, may clarify if PBX1 is most important in terminal differentiation.

Evaluation of PBX1 expression in three clinical neuroblastoma cohorts demonstrated its potential utility as a biomarker in high-risk disease. PBX1 is particularly powerful prognostically in the Seeger dataset of metastatic MYCN-nonamplified neuroblastoma, the subgroup of patients that would most benefit from a validated biomarker. PBX1 expression also was independently prognostic in multivariate analysis in the larger heterogeneous Oberthuer cohort, including MYCN-amplified and nonamplified patients. It was more significant than age, stage, and MYCN amplification, well-established biomarkers of aggressive disease even in patients with localized tumors (5). Although multivariate analysis was limited in the Khan dataset due to small sample size and homogeneous patient characteristics, PBX1 expression was prognostic of survival even in this cohort. The significant survival difference appreciated in these sets based on PBX1 expression and including high-risk patients may be explained by the microarrays used. The probes used hybridize to the 3′ end of PBX1, where the differences in the isoforms are, and the analysis may have unveiled an isoform-specific difference (e.g., the arrays identify PBX1a specifically). Future studies could reasonably evaluate if PBX1 expression distinguishes high-risk patients for whom current therapies are ineffective from those who benefit from standard therapy, in a similar approach to that described for low-risk disease. It may be particularly useful for high-risk MYCN-nonamplified disease. Despite INRG risk classification criteria (48), there are no validated predictors of therapeutic resistance for these patients and recently identified biomarkers have proved ineffective (49).

Our work has identified low PBX1 expression as a novel biomarker of disease aggression in neuroblastoma, and laboratory findings define a role for PBX1 in neuroblastoma differentiation. This biomarker may serve to better define risk for patients with both low-risk disease and MYCN-nonamplified high-risk disease through future prospective trials. These data anticipate a therapeutic potential for patients by modulating the pathways that affect PBX1 expression.

No potential conflicts of interest were disclosed.

Conception and design: N. Shah, J. Toretsky

Development of methodology: N. Shah, J. Selich-Anderson, H. Siddiqui

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Shah, J. Selich-Anderson, G. Graham, H. Siddiqui

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Shah, J. Wang, J. Selich-Anderson, H. Siddiqui, X. Li, J. Khan

Writing, review, and or revision of the manuscript: N. Shah, J. Selich-Anderson, G. Graham, H. Siddiqui, J. Khan, J. Toretsky

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Shah, G. Graham

This study was supported by Conquer Cancer Foundation Young Investigator Award and NCH Internal Fund 187213 (to N. Shah), Burroughs Wellcome Clinical Scientist in Translational Research, and NIH grants R01CA133662, R01CA138212, and RC4CA156509 (to J. Khan).

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.