|Home | About | Journals | Submit | Contact Us | Français|
Retinoids play key roles in differentiation, growth arrest and apoptosis and are increasingly used in the clinic for the treatment of a variety of cancers, including neuroblastoma. Using a large-scale RNA interference-based genetic screen we identify ZNF423 (also known as Ebfaz, OAZ or Zfp423) as a component critically required for retinoic acid (RA)-induced differentiation. ZNF423 associates with the RARα/RXRα nuclear receptor complex and is essential for transactivation in response to retinoids. Down-regulation of ZNF423 expression by RNA interference in neuroblastoma cells results in a growth advantage and resistance to RA-induced differentiation, whereas overexpression of ZNF423 leads to growth inhibition and enhanced differentiation. Finally, we show that low ZNF423 expression is associated with poor disease outcome of neuroblastoma patients.
Cancer biomarkers make it possible to foretell cancer outcome (prognosis) or responses to therapy (prediction). Human neuroblastoma is the most common childhood solid tumor with a broad range of clinical outcomes, ranging from spontaneous regression to extremely aggressive disease. We show here that ZNF423 is a prognostic biomarker for human neuroblastoma independent of MYCN amplification. We also establish here a causal role of ZNF423 in RA-induced differentiation and proliferation of neuroblastoma cells. Therefore, ZNF423 may also predict responses to RA-based therapies in the clinic. More generally, our results underscore that the identification of novel components of key signaling pathways using genetic screens can yield biomarkers having clinical utility.
The vitamin A metabolite retinoic acid (RA) is essential for embryonic and adult growth. It plays key roles in development, differentiation and homeostasis. These diverse effects of RA are exerted primarily through the ability to differentially regulate gene expression mediated by the retinoic acid receptors (RARs). RAR belongs to the super family of nuclear hormone receptors that are ligand-regulated transcription factors (Chambon, 1996). RAR functions through hetero-dimerization with retinoid × receptor (RXR), which is the common partner for several other nuclear receptors, such as vitamin D receptor (VDR), peroxisome proliferator-activated receptor (PPAR), thyroid hormone receptor (T3R) and RXR itself (Mangelsdorf et al., 1995; McKenna and O’Malley, 2002). As the critical dimerization component, RXR is therefore a master regulator of many hormone responses.
RAR/RXR heterodimers constitutively associate with retinoic acid response elements (RAREs) in promoters of target genes (Chambon, 1996). In the absence of ligand, RAR/RXR actively represses transcription through association with corepressors NCoR and SMRT and recruitment of histone deacetylases (HDAC) that prevent opening of the chromatin (Chen and Evans, 1995; Horlein et al., 1995). Binding of RA to RAR induces a conformation change of the complex. Subsequently, corepressors are released and coactivator complexes are recruited to activate transcription. This ligand-dependent exchange of coregulators requires adaptor proteins such as TBLR1 (Perissi et al., 2004). Many of the coactivators, including CBP/p300, PCAF, and members of the p160 family (SRC1, TIF-2/GRIP1, and ACTR/RAC3/AIB1), possess histone acetylase (HAT) activity that promotes transactivation of RAR/RXR (Chen et al., 1997; Onate et al., 1995; Voegel et al., 1996). In contrast, ligand-dependent corepressors such as RIP140, LCoR and PRAME recruit HDACs or PcG proteins to ligand-bound RAR/RXR complexes to repress their activities (Cavailles et al., 1995; Epping et al., 2005; Fernandes et al., 2003).
RA signaling through RAR/RXR and the subsequent activation of target genes induce differentiation, cell cycle arrest and apoptosis in many cell types. Consequently, RA displays distinct anti-carcinogenic activities and is currently used and being tested as a therapeutic agent for several human cancers (Altucci and Gronemeyer, 2001; Freemantle et al., 2003). For example, RA is used to treat patients suffering from acute promyelocytic leukemia (APL), where translocations of RARα give rise to the PML-RARα chimeric gene (Grignani et al., 1998; He et al., 1998). The resulting fusion protein is a constitutive repressor that inhibits wild-type RARα activity, thereby preventing myeloblast differentiation at physiological levels of RA. Pharmacological concentrations of RA alleviate this dominant negative block by allowing dissociation of corepressors and recruitment of coactivators to activate transcription. Furthermore, lost expression of RA target gene RARβ, an isoform of RAR, is involved in the progression of a diverse range of solid tumors (Altucci and Gronemeyer, 2001; Freemantle et al., 2003). As a chemoprevention, RA treatment can be used to restore RARβ expression. For example, restoration of RARβ expression by RA therapy in pre-malignant oral lesions was associated with a clinical response (Xu et al., 1995).
RA therapy has also been used in the clinic to treat human neuroblastoma, a childhood tumor arising in the peripheral sympathetic nervous system (Brodeur, 2003). Some tumors regress spontaneously, but the majority of neuroblastoma patients have aggressive tumors with poor clinical outcome despite intense therapy. While a subset of these aggressive tumors is identified by genomic amplification of MYCN proto-oncogene, less is known about additional genetic factors that control neuroblastoma tumor progression (Brodeur, 2003). High expression of the neurotrophic receptor TRKA has been identified as prognostic for favorable outcome, whereas TRKB and its ligand, the brain-derived neurotrophic factor (BDNF), are frequently expressed in unfavorable aggressive tumors with MYCN amplification (Brodeur, 2003; Schramm et al., 2005). In addition, the expression of FYN kinase is prognostic for good outcome of patients independent of MYCN amplification (Berwanger et al., 2002). Signaling through FYN kinase controls neuroblastoma cell differentiation and proliferation. RA signaling has also been implicated in human neuroblastoma, because high levels of either all-trans RA or 13-cis-RA induce cell proliferation arrest and morphological differentiation of human neuroblastoma cell lines (Reynolds et al., 1994; Sidell et al., 1983). A phase III randomized trial showed that treatment of 13-cis-RA given after completion of intensive chemo-radiotherapy significantly improved event-free survival in high-risk neuroblastoma (Matthay et al., 1999). This RA therapy has now become standard practice to treat high-risk neuroblastoma patients after marrow or stem cell transplantation.
ZNF423 is a transcription factor containing 30 Krüppel-like C2H2 zinc fingers (Turner and Crossley, 1999). It was first described as a transcriptional partner of Olf-1/EBF, a family of transcription factors in olfactory epithelium and lymphocyte development in the rat (Tsai and Reed, 1997; Tsai and Reed, 1998). Independently, ZNF423 was then identified as a co-factor interacting with SMAD1-SMAD4 complex in BMP-signaling pathways (Hata et al., 2000; Ku et al., 2006). The zinc finger clusters of ZNF423 mediating these two signaling pathways are separable and independent from each other. In both pathways, ZNF423 assembles a transcriptional complex by binding to distinct partners and DNA sequences. ZNF423 is required for cerebellar development, and plays important roles in olfactory neurogenesis and CNS midline patterning in mice (Cheng and Reed, 2007; Cheng et al., 2007; Warming et al., 2006). We identify here ZNF423 as a component of RA signaling and demonstrate that it plays a key role in RA-induced differentiation in several cell types, including neuroblastoma.
In response to RA mouse F9 embryonic teratocarcinoma cells differentiate into extraembryonic endoderm-like cells, recapitulating early stages of mouse embryogenesis (Rochette-Egly and Chambon, 2001). Therefore, F9 cells have been used extensively as a model to investigate RA signaling in vitro. To identify genes involved in RA signaling, we performed a large-scale RNA interference (RNAi)-based loss-of-function genetic screen in F9 cells using a collection of 28,256 short hairpin RNA (shRNA) vectors, which target 14,128 mouse genes (Figure 1A and Supplemental Data). Using retroviral infection, we introduced the entire shRNA library polyclonally into F9 cells. The infected cells were plated in low density and exposed to 1 μM all-trans RA (henceforth referred to as RA). After four weeks of RA selection, the resistant colonies were pooled and total genomic DNA was isolated. The shRNA vectors were recovered by PCR amplification and re-cloned as a pool to construct a mini library. Using this functionally-selected mini library, which was enriched for shRNA vectors that can confer resistance to RA, we performed a second-round selection in F9 cells. To avoid a “passenger” effect from irrelevant shRNA vectors, low multiplicity of infection was used for infection. The resistant colonies from the second-round RA selection were individually isolated. The shRNA inserts were recovered by PCR, re-cloned and subjected to DNA sequence analysis to reveal their identities as described (Berns et al., 2004). As an initial validation, each of the identified shRNA vectors was individually introduced into F9 cells by retroviral infection to retest its ability to confer resistance to RA. As a negative control, a functional shRNA targeting GFP of A. victoria was used throughout this study.
Using this approach, we identified 7 different shRNA vectors that were able to confer RA resistance in F9 cells when expressed. We found one shRNA targeting mouse retinoid × receptor Rxrα (shRxrα) (Figure 1B), which validated the screen, since down-regulation of this receptor leads to decreased transcriptional response to the RA (Figure 3). In addition, we also identified one shRNA targeting the mouse homolog of the human zinc finger protein ZNF423, referred to as Zfp423 (shZfp423) (Figure 1B).
To rule out that the shRNAs conferred RA resistance due to ‘off-target’ effects, we designed and tested additional non-overlapping shRNAs against each of the genes identified in the screen. We only consider a gene identified from the screen as a genuine hit, if at least two non-overlapping shRNAs are able to suppress expression of the intended target gene and confer RA resistance (Echeverri et al., 2006). We found a direct correlation between the knockdown abilities of the vectors and their ability to confer resistance to RA (Figure 1C), indicating that Zfp423 is a genuine hit from the screen. Two of the most potent shRNAs, shZfp423#1 and shZfp423#2 were used throughout this study. We also generated multiple active shRNAs for the gene Rxrα, which we consider as a positive control (Figure 1D). Two of the best shRNAs, shRxrα#1 and shRxrα#2 are used in subsequent experiments. For the remaining five candidate genes, we did not observe a correlation between knockdown of the intended gene and the ability to confer resistance to RA when additional shRNA vectors were tested (data not shown) and were not studied further.
Next, we examined if the RA resistance mediated by suppression of Zfp423 is due to the inhibition of RA signaling or due to a general growth advantage, unrelated to RA. We performed a long-term proliferation assay of F9 cells expressing shRNAs against Zfp423, Rxrα or GFP in the absence or presence of exogenous RA, using the 3T3 protocol (Figure 1E). In the absence of RA, no significant growth difference was detected in all cell lines. When exposed to 1 μM RA, the cells expressing shRNAs targeting Zfp423 or Rxrα continued to proliferate, while the control cells were drastically inhibited. These data argue against a generalized growth advantage for the Zfp423 knockdown F9 cells and suggest a specific resistance to RA-induced growth arrest.
In addition to F9 cells, mouse embryonic stem (ES) cells have been used as another cellular system to study RA-induced differentiation in vitro (Rohwedel et al., 1999). We observed that mouse ES cells also express a high level of Zfp423 (data not shown). Therefore, we examined the role of Zfp423 in RA-induced differentiation in this cellular system. shRNAs against Zfp423, Rxrα or GFP were introduced into E14T mouse ES cells by retroviral infection. As expected, the knockdown abilities of these shRNAs in E14T cells were comparable to those seen in F9 cells (Figure 2B, 1C and 1D). The infected cells were plated in a low density and cultured in the absence or presence of 1μM RA for 1 week. The surviving cells were then stained for alkaline phosphatase (AP), which is a marker of undifferentiated ES cells. In the absence of RA, there was no significant difference in cell growth and maintenance of the undifferentiated state in all knockdown lines (Figure 2A). In the presence of RA, proliferation of the control cells was severely inhibited and the remaining differentiated cells failed to stain for AP (Figure 2A). In contrast, cells stably expressing shRNAs against Zfp423 or Rxrα continued to proliferate and stained positively for AP (Figure 2A). Hence, Zfp423 appears to be also required for RA-induced differentiation in mouse ES cells.
ZNF423 was first implicated as a transcriptional partner of Olf-1/EBF in olfactory epithelium and lymphocyte development in the rat, and was then identified as a transcriptional activator and SMAD1/4 cofactor in regulating bone morphogenetic protein (BMP) signaling. Therefore, we tested whether ZNF423 can also act as a co-factor for RARα/RXRα transactivation. ShRNAs targeting Zfp423, Rxrα or GFP were co-transfected with a reporter gene-containing consensus Retinoic Acid Response Elements (RARE) linked to luciferase (RARE-Luc) into F9 cells (Epping et al., 2005). Both shRNAs against Zfp423 inhibited the reporter gene activation by RA, similar to the shRNAs targeting Rxrα (Figure 3A), indicating that Zfp423 is required for transcriptional activation of Rarα/Rxrα in response to RA. As an independent cellular system, we performed the RARE-Luc reporter assays in NTERA2 human embryonic teratocarcinoma cells that also express high level of ZNF423 (Figure S1). Similarly, shRNAs targeting human ZNF423 and RXRα (see Figure 5C and 5D) also suppressed RA-induced expression of the reporter gene (Figure 3B). Conversely, expression of both the wild type ZNF423 and FLAG-ZNF423ΔN (which lacks the first 60 amino acids but contains all of the 30 zinc fingers) were able to hyper-activate the RARE-Luc reporter in response to RA (Figure 3C). The difference in the ability of ZNF423 and FLAG-ZNF423 to activate reporter expression probably reflects the fact that FLAG-ZNF423 was more highly expressed than wild-type ZNF423 (Figure S1). These results suggest that ZNF423 functions as a cofactor for RARα/RXRα transactivation.
Using a similar approach, we found that ZNF423 is required for activation of RARE-Luc reporter induced by RAR selective-agonists for all isoforms: RARα, β and γ (Figure S2). Consistent with these results, co-expression of ZNF423 sensitizes the RA response in RARα,β,γ Triple Knockout (TKO) MEFs reconstituted with each of the three RAR isoforms (Figure S3) (Epping et al., 2007). Furthermore, ZNF423 is also required for both reporter gene activation and growth inhibition induced by an RXR-selective ligand in both mouse F9 and human SH-SY5Y neuroblastoma cells (Figure S4 and S5). Together, these results indicate that ZNF423 is a critical co-factor for all three retinoic acid receptors.
To further substantiate that Zfp423 knockdown inhibits RA-induced transcription, we analyzed the mRNA expression levels of a panel of eight of bona fide RA target genes including in Rarβ and Crabp2 in F9 cells. As expected, all eight target genes were drastically induced in the control cells following exposure to RA (Figure 3D, 3E and S6). However, induction of these genes was significantly inhibited in the cells expressing shZfp423 or shRxrα (Figure 3D, 3E and S6). In addition, it has been shown that Wnt6 expression is down-regulated in F9 cells upon RA treatment (Eifert et al., 2006). We also observed this repression of Wnt6 in control F9 cells in response to RA, but not in the cells stably expressing shZfp423 or shRxrα (Figure 3F). These results further demonstrate that ZNF423 is required for RXRα/RARα transcriptional regulation in response to RA.
In contrast, Zfp423 appears to be not required for RARα-mediated transcriptional repression in the absence of RA in F9 cells (Figure S6). Consistent with this, ZNF423 is not required for transcriptional repression of GAL4-Luciferase by GAL4-RARα in the absence of RA (Figure S7).
To investigate the possible physical interaction between ZNF423 and RARα/RXRα, we performed GST pull-down experiments using recombinant GST fusions of RXRα, RARα and in vitro-translated ZNF423. ZNF423 associated with GST-RXRα and GST-RARα in both the absence and presence of RA (Figure 4A). As controls for proper folding of the fusion proteins, hetero-dimerization between GST-RXRα and RARα or visa versa was found, but homo-dimerization of RARα or RXRα was inefficient under the conditions used (Figure 4A). To ask if ZNF423 binds RARα/RXRα in vivo, we performed co-immunoprecipitation experiments in parental F9 cells. Endogenous Rxrα and Rarα co-immunoprecipitated with endogenous Zfp423 (Figure 4B), indicating that Zfp423 and Rarα/Rxrα form a stable complex under physiological conditions in vivo. Consistent with the in vitro binding results, the association of Zfp423 with Rarα/Rxrα in F9 cells did not require RA, but was slightly enhanced when cells were treated with 1 μM RA. This enhanced interaction in response to RA coincided with the slight increase in Zfp423 protein and mRNA levels after RA exposure (Figure 4B and data not shown).
Since ZNF423 directly associates with RXRα and is required for transactivation of all three RARs, we predicted that ZNF423 would also interact with RARβ and RARγin addition to RARα. Indeed, all RAR isoforms were able to be co-immunoprecipitated with ZNF423 in cells transfected with pSG5-RXRα/RARα and either with or without pCS2-FLAG-ZNF423ΔN (Figure S8).
Since Zfp423 forms a stable complex with Rarα/Rxrα and is required for their transcriptional regulation in response to RA, we examined whether Zfp423 interacts with the promoters of bona fide RA target genes in chromatin. Using antibodies for Zfp423, Rxrα or control IgG, and primers sets flanking the RARE promoter region of Rarβ, we performed chromatin immunoprecipitation (ChIP) assays on F9 cells followed by quantitative PCR (qPCR). Rxrα, used as a positive control, showed strong and specific interaction with the RARE region (Figure 4C). Zfp423 was also significantly associated with the Rarβ promoter in direct proximity of the RARE element both in the presence and absence of RA, while its occupancy in the region distal to the RARE was minimal (Figure 4C). The small increase in the Zfp423 occupancy in the RARE region after RA treatment is consistent with the increased association between Zfp423 and Rarα/Rxrα, and maybe a consequence of the up-regulation of Zfp423 levels as described above (Figure 4B).
RA is a neural differentiation agent and RA signaling has been implicated in differentiation of human neuroblastoma cells (Reynolds et al., 1994; Sidell et al., 1983). Clinical trials showed that RA treatment after completion of intensive chemo-radiotherapy significantly improved survival in high-risk neuroblastoma patients (Matthay et al., 1999; Reynolds et al., 2003). Our finding that ZNF423 is a critical cofactor for RXRα/RARα and the notion that Zfp423 is involved in neural development in mice (Cheng and Reed, 2007; Cheng et al., 2007; Warming et al., 2006), suggested a possible role for ZNF423 in human neuroblastoma. Furthermore, ZNF423 is expressed in many of neuroblastoma cell lines (Figure S11 and S12). Therefore, we investigated whether ZNF423 is causally involved in cell growth and differentiation of human neuroblastoma cell lines. In response to RA treatment, many neuroblastoma cell lines exhibit arrest of proliferation and/or morphological differentiation (Figure 5, S9 and data not shown). Similar to the RA-induced up-regulation of Zfp423 in F9 cells, we also observed a small increase of ZNF423 mRNA and protein in human SH-SY5Y neuroblastoma cells after RA treatment (Figure 5C). Down-regulation of ZNF423 expression by multiple non-overlapping shRNA vectors all conferred resistance to the potent effects induced by RA in SH-SY5Y, SK-N-SH and SK-N-FI neuroblastoma cells, which are non-MYCN amplified (Figure 5 and data not shown). Similar results were also obtained in MYCN amplified cells such as SK-N-BE, IMR32 and SK-N-AS cells (Figure S9 and data not shown), indicating the role of ZNF423 in RA signaling in neuroblastoma cells is independent of the MYCN status. Consistent with the RA-resistance caused by ZNF423 knockdown, the induction of the RA target genes RARβ and CRABP2 was significantly inhibited in SH-SY5Y cells stably expressing shRNAs against ZNF423 or RXRα (Figure 6).
To address whether ZNF423 knockdown interfered with neuroblastoma differentiation, we also examined the expression of several critical neural differentiation markers: neurotrophic receptors, TRKA and TRKB, as well as the receptor for the glial cell–derived neurotrophic factor (GDNF), RET. It has been shown that expression of these receptor kinases is induced upon RA treatment and these receptors are causally involved in RA-induced differentiation in neuroblastoma cells in culture (Esposito et al., 2008; Kaplan et al., 1993; Peterson and Bogenmann, 2004). When the control SH-SY5Y cells were exposed to RA for two weeks, the mRNA levels of TRKA, TRKB, and RET were induced (Figure 6). In contrast, the induction of these neurotrophic receptors in cells expressing shRNAs against ZNF423 or RXRα was significantly inhibited (Figure 6).
Conversely, enforced expression of ZNF423 in SH-SY5Y and SK-N-SH cells led to hypersensitivity to RA, as evidenced by growth inhibition (Figure S10A, B and data not shown). As expected, SH-SY5Y cells expressing exogenous ZNF423 induced higher levels of RA target genes and neural differentiation markers in response to RA treatment (Figure S10C). Moreover, expression of ZNF423 in RA-insensitive SHEP cells that have undetectable levels of ZNF423 restored the defective RA response (Figure 7 and S11). Collectively, these results demonstrate that ZNF423 is required for RA-induced differentiation in neuroblastoma cells.
Surprisingly, we also noticed a growth advantage in neuroblastoma cells expressing shRNAs against ZNF423 or RXRα in the absence of exogenous RA (normal growth conditions containing 8% fetal calf serum) (Figure 5 and data not shown). This growth advantage was also observed, but to a lesser extent, in IMR32 and SK-N-AS cells with MYCN amplification (data not shown). Conversely, ectopic-expression of ZNF423 in SHEP, SH-SY5Y and SK-N-SH cells resulted in growth inhibition even when cultured in the absence of exogenous RA (Figure 7, S10 and data not shown). These effects of ZNF423 knockdown and overexpression could be partially due to low concentrations of RA present in the serum-containing culture medium (see also discussion).
Given the role of ZNF423 in neuroblastoma growth and differentiation, we investigated a possible correlation between ZNF423 expression and progression-free survival in human neuroblastoma patients. The initial analysis was performed using a cohort of 88 human neuroblastoma patients (72 without MYCN amplification and 16 with MYCN amplification; see also Tables S2 and S3) from the Academic Medical Center (AMC) in Amsterdam, Netherlands. The expression data was obtained by microarray analysis (Affymetrix platform) of the primary tumor samples of these patients at the time of diagnosis prior to any treatment. Using this cohort as a “training set”, a cut-off value of ZNF423 expression was determined using the leave-one-out cross-validation scheme (see also Experimental Procedures and Table S3). The patients were then classified into two groups based on ZNF423 expression in their primary tumors using this cut-off value and Kaplan-Meier analysis for progression-free survival was performed. Interestingly, high level of ZNF423 expression was associated with good outcome of patients (all stages combined) and low ZNF423 expression was associated with poor outcome (Figure 8A, left; p=2.9e-03). This potential prognostic value of ZNF423 expression was also significant in the subset of 72 patients with tumors lacking MYCN amplification (Figure 8A, right; p=4.5e-03), but appeared to be not statistically significant in the 16 patients with tumors having MYCN amplification (p=0.494; data not shown), which is likely due to the small number of patients in this subgroup.
Next, we validated the prognostic value of ZNF423 using a second independent cohort of 102 neuroblastoma patients with metastatic neuroblastoma tumors lacking MYCN amplification (Asgharzadeh et al., 2006). Since the gene expression data of the second cohort was also based on the same microarray platform, the ZNF423 expression from both datasets were normalized and the same ZNF423 cut-off value determined in the AMC cohort was used for the validation study (see (supplemental) Experimental Procedures). Similar results were obtained in the second cohort: high level of ZNF423 expression was associated with good outcome of patients in progression-free survival and low ZNF423 expression was associated with poor outcome (Figure 8C; p=3.3e-04). Thus, these results validate the prognostic value of ZNF423 in neuroblastoma. Furthermore, using this validated ZNF423 cut-off value, multivariate analysis of the AMC cohort with other clinically-relevant parameters showed that expression of ZNF423 predicted survival independently of MYCN amplification and patient age (Figure 8B).
This prognostic value of ZNF423 is further supported by a third independent dataset of 101 neuroblastoma patients (81 without MYCN amplification and 20 with MYCN amplification) (Wang et al., 2006), where ZNF423 expression is high in tumors of early stage disease, but lower in more advanced stage disease (Figure 8D, left; p=3.1e-0.5). Similar correlation was also observed in the subset of 81 patients with tumors lacking MYCN amplification (Figure 8D, right; p=3.1e-03). Since MYCN amplified tumors existed only in stage 4 of this cohort, such correlation analysis of ZNF423 expression and stages could not be performed in this subgroup.
When we expressed ZNF423 in human neuroblastoma cells and performed global gene expression analysis, we found that 852 genes were greater than 2-fold regulated by ZNF423 (GEO Submission GSE14627). Of these 852 genes, 142 (see Table S4) were also significantly differentially expressed in the “ZNF423 high” versus the “ZNF423 low” cohort of 102 neuroblastomas (Asgharzadeh et al., 2006). Randomization analyses demonstrate that the genes that respond to ZNF423 expression in vitro are also statistically significantly associated with ZNF423 expression in patient tumors (Figure S13).
In summary, these multiple expression analyses in patient tumors indicate that expression of ZNF423 is prognostic for outcome in human neuroblastoma patients.
We identified ZNF423 in an unbiased genome-wide RNAi screen for additional components of the RA signaling pathway and found that it is a critical cofactor of RA-induced transcription that interacts with RARα/RXRα. Subsequently, we demonstrated that ZNF423 is critical for growth and differentiation of neuroblastoma cells and discovered the prognostic value of ZNF423 expression in long-term survival of human neuroblastoma patients.
Previous studies have indicated that ZNF423 is a multi-functional transcriptional regulator, which uses distinct sets of its zinc fingers to regulate different signaling pathways. In this study, we provide several lines of evidence for a novel function of ZNF423 to positively regulate cellular responses to RA. First, suppression of ZNF423 expression by RNAi conferred resistance to RA-induced differentiation in mouse teratocarcinoma cells, mouse ES cells and a panel of human neuroblastoma cells. Conversely, ectopic expression of ZNF423 in RA-insensitive neuroblastoma cells that have undetectable levels of ZNF423 restored the RA response. Second, suppression of ZNF423 by RNAi in mouse teratocarcinoma and human neuroblastoma cells impaired the full induction of endogenous RA target genes in response to RA treatment, indicating that ZNF423 is a critical cofactor required for RARα/RXRα transactivation. Third, ZNF423 interacts with RARα/RXRα both in vitro and in vivo and in chromatin immunoprecipitation experiments ZNF423 associates with RARE elements of the promoter region of RA target genes.
The interaction between ZNF423 and RARα/RXRα does not require RA both in vitro and in vivo and consistent with this, we found Zfp423 constitutively associated with the RARE element of the Rarβ promoter. However, ZNF423 does not appear to be required for repression by the unliganded RAR (Figure S7). Moreover, in line with previous observations (Hata et al., 2000), we did not detect any intrinsic ability of ZNF423 to activate transcription (data not shown). Taken together, we propose a role for ZNF423 as a transcriptional intermediary factor for RARα/RXRα: it constitutively associates with the promoters of RA target genes, and may prime the RA receptors to recruit other essential factors required for RARα/RXRα transactivation in response to RA. Identification of these critical factors recruited by ZNF423 will yield further insights into the regulation of RARα/RXRα transactivation.
Previous studies have indicated that Zfp423 plays an important role in neural development in mice: deletion of Zfp423 resulted in severe cerebellar defects, and impaired olfactory neurogenesis as well as CNS midline patterning in mice (Cheng and Reed, 2007; Cheng et al., 2007; Warming et al., 2006). These effects of Zfp423 appear to coincide with a central role of RA signaling in neuronal differentiation and CNS patterning in animals (Maden, 2002; McCaffery et al., 2003). RAR-knockout mice exhibit neural crest defects and disorganized hindbrain and these effects can be mimicked by the use of a pan-RAR antagonist. These in vivo phenotypes are entirely consistent with the effects of ZNF423 on RA signaling in neuronal cell types reported here.
In the absence of exogenous RA, we also observed effects on cell proliferation and differentiation by ZNF423 knockdown and over-expression in many of the neuroblastoma cell lines. It is possible that the endogenous amount of RA in the regular 8%FCS growth medium is sufficient to activate RA signaling to a low level in these human neuroblastoma cells. Alternatively, other pathways such as BMP and Olf/EBF pathways, in which ZNF423 is also causally involved, could be important for neuroblastoma genesis. BMP2 treatment resulted in growth arrest and differentiation in human neuroblastoma cell lines and Olf/EBF transcription factors were also implicated in neuroblastoma differentiation (Nakamura et al., 2003; Persson et al., 2004). In fact, a synergistic effect of RA and BMP6 on differentiation of human neuroblastoma cell has been demonstrated in culture (Sumantran et al., 2003). As the common mediator for both RA and BMP pathways, ZNF423 might be the critical factor contributing to this synergy. We therefore cannot rule out the possibility that other RA-independent pathways are also responsible for the prognostic value of ZNF423 expression.
Since RA induces growth arrest, differentiation and apoptosis, one might speculate that ZNF423, being an essential mediator of the RA response, has tumor suppressor like properties in neuroblastoma. However, we have not seen loss of heterozygosity at chromosome 16q12, the ZNF423 locus, in a panel of 88 human neuroblastomas, nor did we see that human neuroblastoma cell lines that do not express ZNF423, can be made to re-express the gene after treatment with DNA de-methylating agents (data not shown).
RA-based therapies are also used in leukemias, most notably in Acute Promyelocytic Leukemia (APL). However, we did not observe any ZNF423 expression in two leukemia cell lines, including one APL (Figure S11). This is in good agreement with the neural-restricted expression pattern of ZNF423 (http://symatlas.gnf.org/SymAtlas/) and the neural-specific phenotype of the Zpf423 knockout mice (Cheng and Reed, 2007; Cheng et al., 2007; Warming et al., 2006). Thus, ZNF423 may not be a regulator of the RA response in leukemia. However, since the ZNF gene family consists of some 700 members, it is well possible that another ZNF family member is required for RA responses in other cell types.
Remarkably, in three independent gene expression data sets of over 300 neuroblastomas, we found that low expression of ZNF423 predicts poor outcome. Multivariate analysis shows that ZNF423 prognostic value is independent of MYCN status (Figure 8). These findings are significant since MYCN amplification only accounts for a subset of aggressive neuroblastoma tumors, while the remainder lacks consistently identifiable biomarkers. ZNF423 expression could therefore serve as a prognostic biomarker for neuroblastoma tumors independent of MYCN amplification.
In summary, our data indicate that low ZNF423 expression is associated with poor outcome in neuroblastoma and clearly implicates ZNF423 in RA signaling. The expression levels of ZNF423 could significantly affect the responses to both endogenous and pharmacological concentrations of RA in cancer patients, which in turn may influence the outcome of neuroblastoma. Therefore, ZNF423 may also be a biomarker predicting responses to RA-based therapies, which are increasingly being used to treat neuroblastoma. Additional clinical studies are needed to validate this potential predictive power of ZNF423 expression for retinoid-based therapies in human neuroblastoma.
The NKI mouse shRNA library (containing 28,256 shRNA vectors that target 14,128 mouse genes) was constructed into pRISC retroviral vector, which is derived from pRETRO-SUPER with additional chloramphenicol resistance marker under regulation of TET promoter (Brummelkamp et al., 2002). The details of this library can be found in the Supplement Data and can also be viewed at www.screeninc.nl.
Transfections were carried out using calcium phosphate precipitation. RARE-luciferase reporter assays were performed in DMEM supplemented with charcoal-stripped FCS (HyClone, Logan, UT) essentially as described (Epping et al., 2007).
These experiments were performed according the protocols as described (Epping et al., 2005). Normal rabbit IgG, anti-RXRα (D-20) and anti-ZNF423 (H-105) were used for ChIP. The sequences of the primer sets used for ChIP-qPCR analysis are listed in table 1 of the Supplemental Data.
For all three independent studies presented, the expression data were obtained using Affymetrix micro-array analyses on the untreated primary tumor samples at the time of diagnosis. For the cohort of 88 neuroblastoma patients from the Academic Medical Center (AMC; Amsterdam, Netherlands), material was obtained during surgery and immediately frozen in liquid nitrogen. MYCN amplifications and LOH 1p were all determined using Southern blot analysis of tumor material and lymphocytes of the same patient. The patient samples for the other two cohorts are as described (Asgharzadeh et al., 2006; Wang et al., 2006). Written informed consent was obtained from patients’ parents or guardians in accordance with institutional review board policies and procedures for research dealing with tumor specimen and clinical information. The institutional review board at Childrens Hospital Los Angeles and the medical-ethics committee of the Academic Medical Center in Amsterdam approved the study.
To determine the optimal value to set as a cut-off for ZNF423 expression in the AMC cohort, the leave-one-out cross-validation scheme was used. The Neuroblastoma patients were sorted based on the expression of ZNF423 and subsequently divided into 2 groups based on the expression value of every patient. For every group separation (Higher or lower than the current ZNF423 expression), the logrank significance was calculated. The best p-value out of the sequence was then used to represent the final gene expression cut-off value for the ZNF423. This cut-off value was validated using a second independent set of 102 patients (Asgharzadeh et al., 2006) (supplemental experimental procedures).
Progression-free survival was measured for all outcome analysis presented in this study using the same ZNF423 cut-off value. The definition of an event for the AMC cohort (all stages) is ‘progressive disease’, ‘recurrent disease’ or ‘death’ of the patient. For the second cohort with metastatic neuroblastomas lacking MYCN amplification, the disease progression was defined as development of any new lesion, increase of any measurable tumor mass by greater than 25%, or previously negative bone marrow becoming positive for tumor cells.
For the multivariate analysis, Cox regression calculations on progression-free survival were performed in SPSS version 15.0. For these calculations single covariates (ZNF423, MYCN, LOH1p, stage, age) as well as double covariates in a non-sequential model (MYCN, LOH1p, stage, age) in combination with the ZNF423 cut-off were used. The variable stage consisted two groups based on INSS: 1) ST1, 2, 3 and 4S; 2) ST4. The variable age (of diagnosis) consisted two groups: 1) <=18 months; 2) > 18 months. Hazard ratio’s, p-values and 95% confidence intervals were part of the SPSS output.
The gene expression data for SH-SY5Y cells with and without over-expression of ZNF423 are available at GEO, submission GSE14627.
We thank Koen Braat for help with the ES cell experiments, Lodewyk Wessels for statistical advice, Hinrich Gronemeyer for the gift of RAR ligands and RAR Triple Kockout MEFs, the NKI microarray facility for supporting gene expression studies, Jasper Mullenders and other members of the Bernards lab for discussion, and Tobias Otto and Martin Eilers for cell lines. We are grateful to Cinzia Pochet for support. This work was supported by the EU 6th framework integrated project INTACT, The Netherlands Genomics Initiative (NGI) and a grant from the US National Institutes of Health.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.