The human tumour antigen PRAME (preferentially expressed antigen of melanoma) is frequently overexpressed in tumours. High PRAME levels correlate with poor clinical outcome of several cancers, but the mechanisms by which PRAME could be involved in tumourigenesis remain largely elusive. We applied protein‐complex purification strategies and identified PRAME as a substrate recognition subunit of a Cullin2‐based E3 ubiquitin ligase. PRAME can be recruited to DNA in vitro, and genome‐wide chromatin immunoprecipitation experiments revealed that PRAME is specifically enriched at transcriptionally active promoters that are also bound by NFY and at enhancers. Our results are consistent with a role for the PRAME ubiquitin ligase complex in NFY‐mediated transcriptional regulation.
PRAME is the founding member of a large family of PRAME genes (Birtle et al, 2005). PRAME genes have been implicated in oogenesis (Dadé et al, 2003), spermatogenesis (Wang et al, 2001) and stemness (Bortvin et al, 2003; Cinelli et al, 2008), but their function is poorly understood. Human PRAME was initially identified as the protein triggering an autologous anti‐tumour immune response in a melanoma patient (Ikeda et al, 1997). Since then, many reports have shown that PRAME is not expressed in most normal adult tissues, while it is overexpressed in a wide variety of solid and haematological malignancies (Kilpinen et al, 2008). Although not prognostic for all cancers (van Baren et al, 1998; Steinbach et al, 2002; Tajeddine et al, 2005), high PRAME expression has been found to correlate with the stage of melanoma lesions (Haqq et al, 2005), and with shorter overall survival of neuroblastoma (Oberthuer et al, 2004) and serous ovarian adenocarcinoma patients (Partheen et al, 2008). High PRAME mRNA levels were also shown to be an independent prognostic factor of poor clinical outcome in breast cancer (Doolan et al, 2007; Epping et al, 2008). In addition, PRAME is significantly activated in BCR‐ABL‐positive chronic myeloid leukaemia (Ph+ CML) patients at the onset of advanced phase and blast crisis (Radich et al, 2006), and hypomethylation of the PRAME locus seems to have a role in this process (Roman‐Gomez et al, 2007; Schenk et al, 2007). A growing body of evidence indicates that PRAME contributes markedly to the oncogenic phenotype of several cancers. PRAME has been recently implicated in retinoic acid signalling and polycomb‐mediated transcriptional repression in melanoma cells (Epping et al, 2005); however, these mechanisms do not seem to have a role in breast cancer (Epping et al, 2008) and acute myeloid leukaemia (Steinbach et al, 2007). While evidence that expression of PRAME is restricted mainly to tumour cells has designated it as a promising therapeutic target, efforts to develop strategies to interfere with the oncogenic activity of PRAME are hindered by a lack of information about its function(s) (Paydas, 2008).
In this study, we combined biochemical and genomics approaches to identify PRAME as a subunit of Cullin2‐based E3 ubiquitin ligases. In addition, we show that PRAME is recruited in cells to epigenetically and transcriptionally active promoter regions bound by nuclear transcription factor Y (NFY), a transcription factor that is essential for early embryonic development (Bhattacharya et al, 2003) and has been implicated in the maintenance of the high proliferative capacity of ES cells (Grskovic et al, 2007) as well as in the inhibition of differentiation (Gurtner et al, 2003).
PRAME associates with Cullin2‐based E3 ubiquitin ligases
To investigate the molecular function of human PRAME, we used a proteomic approach to identify PRAME‐interacting proteins. To this end, we stably expressed epitope‐tagged versions of PRAME suitable for protein purifications in two cellular backgrounds: human K562 leukaemia cells, which endogenously express detectable levels of PRAME, and human embryonic kidney HEK293 cells, which lack detectable endogenous PRAME (Figure 1A). When analysed by size exclusion chromatography, both endogenous and epitope‐tagged PRAME from K562 cells showed a broad elution profile, with a substantial fraction of the protein eluting earlier than expected of free PRAME, which has a predicted molecular mass of 58 kDa (Figure 1B). This observation suggests that PRAME may be part of one or more protein complexes.
We purified PRAME‐associated proteins from K562–TAG–PRAME or HEK293–HA–PRAME cells by anti‐epitope immunoaffinity chromatography and analysed them by mass spectrometry. As shown in Tables I and II, eluates from K562–TAG–PRAME and HEK293–HA–PRAME included a collection of proteins not found in control samples. Among the most highly represented were Cullin family member 2 (Cul2), Elongins B (TCEB2) and C (TCEB1), and RING finger protein Rbx1, all of which are subunits of a large family of multisubunit, Cullin‐based, ubiquitin ligases. In these ubiquitin ligases, the Elongin BC dimer acts as an adaptor that links an Elongin BC‐box containing substrate recognition subunit, or BC‐box protein, to heterodimeric modules composed of either Cullin2 and Rbx1, or Cullin5 and Rbx2 (Kamura et al, 1998, 2004; Lonergan et al, 1998). The RING finger domains of the related Rbx1 and Rbx2 proteins in turn recruit and activate an E2 ubiquitin‐conjugating enzyme for ubiquitination of substrates (Kamura et al, 1999; Petroski and Deshaies, 2005).
The interaction of TAG–PRAME with ubiquitin ligase subunits was confirmed by western blotting of TAG‐eluates. As shown in Figure 1C, these fractions contained PRAME, Elongins B and C, and two forms of Cul2 that most likely correspond to unmodified and a more slowly migrating Neddylated‐Cul2, a hallmark of active ubiquitin ligases. Supporting the physiological relevance of these interactions, endogenous PRAME, Cul2, Elongin B and Elongin C were all detected in anti‐PRAME immunoprecipitates from nuclear extracts of parental K562 cells (Figure 1D). Notably, while PRAME was clearly depleted from the extract after incubation with the specific antibodies (Figure 1D, compare lanes 1, 2 and 4), Cul2, Elongin B and Elongin C were not, consistent with their function as components of many Cullin‐based ubiquitin ligases present in cells (Kamura et al, 1998; Mahrour et al, 2008).
PRAME is a BC‐box protein
Although we identified all of the other subunits of Cul2‐based ubiquitin ligases as PRAME‐interacting proteins, we did not detect any of the previously identified BC‐box proteins in our fractions. BC‐box proteins are characterized by a sequence motif of about 10 amino acids, called the BC‐box, which binds to Elongin C. In addition, a downstream Cullin‐box specifies binding to either Cul2 or Cul5 (Kamura et al, 2004; Mahrour et al, 2008). Multiple sequence alignments of PRAME with several known BC‐box proteins identified putative BC‐ and Cul2‐boxes near the PRAME N‐terminus (Figure 2A), raising the possibility that PRAME might function as a substrate recognition subunit of a Cul2‐based ubiquitin ligase.
To assess the functionality of the predicted motifs, we introduced point mutations in key residues of the putative BC‐box (L26P/A30F) or the putative Cul2‐box (L48A/P49A). We observed that the BC‐box mutation greatly reduced the binding of PRAME to Elongin C and Cul2 in baculovirus‐infected insect cells (Figure 2B). We next generated K562 and HEK293 cells stably expressing epitope‐tagged PRAME carrying the BC‐box or Cul2‐box mutations and performed immunoprecipitations from nuclear extracts. In these cells, both mutations disrupted the interaction of PRAME with endogenous Cullin2 complex components (Figure 2C; Table II), demonstrating the functionality of the motifs.
In order to test the biochemical activity of the Cullin2–PRAME complex, we performed in vitro ubiquitination assays using purified recombinant proteins. As shown in Figure 2D, we observed significant synthesis of polyubiquitin chains only when all the subunits were combined in the presence of ATP, confirming that the Cullin2–PRAME complex is a catalytically active ubiquitin ligase.
Taken together, these results indicate that PRAME's BC‐ and Cul2‐boxes are essential for the assembly of a PRAME‐containing Elongin BC ubiquitin ligase, and they identify PRAME as a BC‐box substrate recognition subunit of a Cul2‐based ubiquitin ligase (Figure 2E).
PRAME associates with chromatin
Cell fractionation and cesium chloride gradient purification of crosslinked chromatin from K562 cells showed that PRAME associates with chromatin (data not shown). In addition, we observed that PRAME can be recruited to DNA in vitro. Consistent with the absence of recognizable DNA‐binding domains in PRAME and other PRAME complex subunits, purified PRAME complexes did not bind directly to DNA immobilized on agarose beads. When incubated with bead‐bound DNA in the presence of nuclear extract, however, PRAME complexes were retained on the DNA beads, suggesting that activity(s) in the extract are needed to bridge PRAME's interaction with DNA (Supplementary Figure S1A).
To investigate the chromatin‐binding profile of PRAME, we performed ChIP assays followed by deep sequencing (ChIP‐seq) using anti‐PRAME antibodies or preimmune serum as control for the specificity of the signals obtained. Overlapping sequence tags were joined and displayed as tag‐density profiles in the UCSC genome browser. Strikingly, inspection of the profiles revealed that the promoter regions of many protein‐coding genes had specific enrichment for PRAME, as exemplified by the screenshots in Figure 3A. Validation experiments by ChIP–qPCR confirmed the high quality of the sequencing data (Supplementary Figure S1B). Furthermore, ChIP‐seq with stably expressed, epitope‐tagged PRAME resulted in very similar if not identical patterns of PRAME enrichment (data not shown). To substantiate our finding, we used the MACS algorithm (Zhang et al, 2008) at a false discovery rate of 10−6 and identified a total of 14 132 PRAME‐binding sites. Location analysis showed that 43% of the PRAME sites are closer than 1 kb upstream or downstream of a transcription start site (TSS) (Figure 3B). Importantly, promoters were nine‐fold enriched as compared with the background distribution of genomic DNA.
Recently, Heintzman et al (2009) identified enhancer elements in K562 cells based on ChIP‐on‐chip histone modification signatures and binding of the enhancer‐associated acetyltransferase p300. Most enhancers were cell‐type specific, and they were enriched in genomic domains containing genes specifically expressed in K562 cells. When we compared the binding profile of PRAME to the predicted enhancers, we found a significant overlap between promoter‐distal PRAME‐binding sites and enhancer regions: 44% of the distal PRAME sites overlapped predicted enhancers, whereas only 4.1% of random genomic sites did (11‐fold enrichment). Moreover, similarly to the annotated enhancers, the distal PRAME sites were not randomly distributed in the genome, but clustered in the vicinity of PRAME‐bound promoters (Supplementary Figure S1C), suggesting possible promoter–enhancer interactions.
PRAME associates with NFY promoters
We next analysed all PRAME‐binding sites falling within 1 kb of promoters. When we plotted their distance to the closest TSS, we found that 83% of promoter‐proximal sites bound by PRAME are concentrated in a narrow window between −400 bp and +100 bp from TSS, peaking around −87 bp relative to the TSS (Figure 3C).
Promoters are characterized by the presence of transcription factor binding sites that critically contribute to the regulation of gene expression. Since neither PRAME, nor other PRAME complex subunits contain recognizable DNA‐binding domains, we reasoned that other transcription factor(s) might mediate the recruitment of PRAME to these regions. We therefore performed de novo motif discovery with GimmeMotifs (van Heeringen and Veenstra, 2011), using the DNA sequences underlying the promoter regions bound by PRAME. This analysis identified a CCAAT sequence as the top‐scoring motif, which stood out with highly significant P‐values (P<10−20) and the highest enrichment and MNCP scores as compared with a pool of random promoters (Figure 3D; Supplementary Figure S2). The CCAAT sequence can be bound by the NFY complex, a heterotrimeric transcription factor ubiquitously expressed in cycling cells (Maity and de Crombrugghe, 1998). The NFYA subunit harbours the DNA‐binding domain that recognizes the CCAAT sequence, while NFYB and NFYC are characterized by histone‐like folds (Maity and de Crombrugghe, 1998). Similar de novo motif analysis performed on the distal PRAME‐binding sites identified two significant motifs: a GATA/Evi1 motif, and a Fos motif.
ChIP‐seq experiments with antibodies against the NFYA and NFYB subunits revealed that PRAME sites at promoters are indeed bound by NFY in vivo. Screenshots of the sequencing data show clear enrichment of NFY at the representative PRAME sites DNM2 and EIF2AK3, as well as PRAME enrichment at the classical NFY promoters TOP2A and CDC25C (Figure 4A). Consecutive ChIP experiments (ChIP–re‐ChIP) using crosslinked NFYA antibodies detected PRAME and NFY concomitantly on the same regions, indicating that these proteins are simultaneously localized at the same promoter regions (Figure 4B).
Using the MACS algorithm, we identified a total of 28 844 regions that were bound by NFYA and/or NFYB. The distance distributions of both NFYA‐ and NFYB‐binding sites from TSSs showed a distinct peak around −78/−84 bp very similar to PRAME (Figure 4C). We next counted the number of sequence tags within NFY‐binding sites and performed a regression curve analysis (Supplementary Figure S3). The binding of NFYA and NFYB were very closely correlated at the majority of the sites, as expected for stoichiometric binding and in agreement with in vitro experiments, showing that all subunits are required for efficient DNA binding (Sinha et al, 1995).
Globally, 71% of all PRAME‐binding sites overlapped NFY regions. Importantly, virtually all PRAME‐bound promoters were also bound by NFY, corroborating and extending the relevance of the CCAAT motif (Figure 4D). In contrast, a sizeable number of NFY promoters did not contain PRAME peaks (Figure 4D). We performed functional annotation of the associated genes and found that NFY‐only genes as well as NFY genes with PRAME enrichment are associated with metabolic and cancer‐related pathways. Furthermore, NFY‐only genes are specifically enriched for immunological processes and intercellular junctions, while NFY genes with PRAME binding are significantly associated with numerous transcription‐related pathways and cell cycle regulation (Supplementary Table SI).
PRAME is specifically recruited to transcriptionally active NFY promoters
Post‐translational modifications of the histone proteins, like lysine methylations and acetylations, are implicated in the regulation of gene transcription. Therefore, we used histone marks to assess the transcriptional status of PRAME loci. For this purpose we processed ChIP‐seq data sets generated by the ENCODE project for K562 cells (see Materials and methods), and we quantitated the signals at all NFY promoters. We found that the active histone marks H3K9ac and H3K4me3 showed a strong correlation with the binding of PRAME (Figure 5A). In contrast, NFY‐only promoters had near‐background levels of these marks.
To investigate the transcriptional status of the associated genes, we quantitated the occupancy in gene bodies of RNA Polymerase II and of the histone marks H3K36me3, which has been linked to transcriptional elongation (Krogan et al, 2003), and H3K27me3, which has been linked to repression (Barski et al, 2007). As shown in Figure 5B, NFY genes without PRAME binding displayed low levels of PolII and H3K36me3, and high levels of the repressive H3K27me3 mark. In contrast, PRAME‐bound genes had significantly higher occupancy of both PolII and H3K36me3 (Figure 5B and D; Supplementary Figure S4), suggesting that PRAME is associated specifically with transcriptionally active genes. To substantiate these findings, we analysed gene expression levels by RNA‐seq in our K562 cells. In agreement with the ChIP‐seq data, we detected few, if any, transcripts from NFY‐only genes, while PRAME‐bound NFY genes were expressed at high levels (Figure 5C).
We next tested if components of the Cullin2 complex are present together with PRAME on chromatin. In the absence of ChIP‐grade antibodies, we established stable K562 cell lines expressing Elongin C tagged with a double TY1 epitope (TTE–EloC), for which a ChIP‐grade antibody was validated in our laboratory. ChIP–qPCR experiments showed specific binding of EloC at all the PRAME promoters tested (Figure 5E). As a control, we performed similar experiments with TTE–PRAME.
Taken together, our data support a model in which a Cul2/EloBC/PRAME complex is specifically bound to the subset of epigenetically and transcriptionally active NFY promoters and nearby enhancer regions.
Our proteomic data reveal that the human oncoprotein PRAME is a BC‐box substrate recognition subunit of a Cul2‐based E3 ubiquitin ligase, suggesting a functional link between ubiquitination and the oncogenic properties of PRAME. Notably, the Von Hippel‐Lindau tumour suppressor protein is one of the best‐characterized BC‐box proteins, and its tumour suppressor activity is due in part to its role as the substrate recognition subunit of a Cul2‐based ubiquitin ligase (Kaelin, 2007). Intriguingly, all the annotated PRAME‐like proteins display a complete conservation of the BC‐box and Cul2‐box at their N‐terminus (Supplementary Figure S5), suggesting that all members of the large PRAME protein family might function as substrate recognition subunits of Cul2‐based ubiquitin ligases, possibly displaying different degrees of specificity and/or redundancy in substrate selection.
To date, ubiquitination has been demonstrated to have important roles in gene regulation, affecting several different steps in the transcriptional process (Conaway et al, 2002; Collins and Tansey, 2006). Many activators are subject to ubiquitin‐mediated degradation at a rate that parallels their transactivation abilities (Salghetti et al, 2001). Histones, transcription factors and RNA polymerase II (RNAPII) are targets of ubiquitination, and ubiquitination events are required for the initiation‐competent RNAPII complex to exchange factors and become competent for cotranscriptional mRNA processing (Muratani et al, 2005). A recent study indicated that the stability of NFYA can be modulated by ubiquitination and degradation via the proteasome, and that degradation‐resistant NFYA mutants increased cell proliferation (Manni et al, 2008). Interestingly, some of the lysine residues in NFYA that are targets of ubiquitination could also be acetylated by p300, suggesting a regulatory competition between these post‐translational modifications (Manni et al, 2008).
Our genome‐wide analyses reveal that PRAME is strongly enriched at NFY promoters, and binding of PRAME predicts a permissive epigenetic landscape and transcriptional activity at these regions. Moreover, we found that PRAME is also significantly enriched at adjacent enhancer regions, suggesting that it could participate in promoter–enhancer interactions. Recently, Ceribelli et al (2008) reported a ChIP‐on‐chip analysis of NFYB in HeLa cells with 2.2% coverage of the human genome. Similar to our K562 data, the majority of the NFY‐bound promoters identified in HeLa cells were transcriptionally active and carried H3K9/14ac and/or H3K4me3 marks, while about 27% of NFY‐bound promoters were devoid of these active marks and mostly silent. Further experiments implicated NFY not only in the expression of active genes, but also in the repression of silent ones, revealing an unexpected dual nature of the NFY complex. The basis of this dual functionality, however, is still unclear, and genome‐wide data suggest that the activity and inactivity of NFY promoters is determined by additional factors and mechanisms that are yet to be identified.
PRAME is the first factor reported to discriminate between active and inactive NFY loci. Our genome‐wide analyses, summarized in Figure 5F, reveal that PRAME is specifically recruited to NFY loci associated with the promoters of genes that are highly transcribed and that have high levels of H3K9ac and H3K4me3 at their promoters and high H3K36me3 but low H3K27me3 levels in their gene bodies. These data suggest a role for PRAME in transcriptional regulation, and possibly in the switch between activity and inactivity of NFY loci. It is tempting to speculate that PRAME might mediate ubiquitination, and possibly degradation, of one or more of the factors that are involved in the regulation of NFY promoters. Despite evidence that PRAME and NFYA can co‐occupy the same genetic loci (Figure 4B), we have not been able to detect stable and direct interactions between PRAME and NFY subunits by co‐immunoprecipitation following their expression in transfected mammalian cells or in baculovirus‐infected insect cells (data not shown). We have also observed that PRAME can be recruited in vitro to a DNA fragment containing the NFY‐dependent HSPA5 promoter by a mechanism that depends on activity(s) present in a nuclear extract. It seems likely that PRAME–NFY interactions, if they exist, occur only on DNA and depend on additional, yet to be identified factors, since (i) purified NFY is not sufficient to recruit PRAME to HSPA5 promoter DNA and (ii) supplementing nuclear extract with exogenously added NFY does not further enhance PRAME recruitment (Supplementary Figure S1A).
Interactions have been reported between NFY and the histone acetyltransferases GCN5 and PCAF (Currie, 1998), and single‐gene studies showed that NFY can activate at least some of its target genes in cooperation with the acetyltransferase p300, which is able to acetylate histone tails as well as NFY subunits (Li et al, 1998; Caretti et al, 2003; Salsi et al, 2003; Gurtner et al, 2008). While these acetyltransferases constitute potential targets of PRAME, other unknown factors could also take part to these mechanisms. Notably, p300 is not only recruited to promoters, but it is also a hallmark of enhancer elements. Heintzman et al (2007) used p300 ChIP‐on‐chip binding profiles to identify an enhancer‐specific chromatin signature, which was successfully used to predict enhancer elements in several cell lines, including K562. Significantly, we found that about half of the promoter‐distal PRAME sites are localized at predicted enhancers, suggesting that PRAME might be implicated in promoter–enhancer crosstalk.
In an effort to gain insight into PRAME's potential role in transcription regulation, we knocked down PRAME with retroviral shRNA vectors and measured the mRNA levels of several target genes. Despite a 70% decrease of PRAME transcripts, we did not observe reproducible changes in the transcript levels of the target genes tested (data not shown). This is perhaps not surprising given that PRAME is a member of a large family of closely related proteins that could have overlapping or redundant functions in cultured cells. Further, the fact that the PRAME gene family underwent an extraordinary expansion specifically in rodents and primates (Birtle et al, 2005; Consortium et al, 2007) suggests that they are more likely to fine‐tune transcriptional regulatory processes rather than to be essential for transcription. Regrettably, the lack of a one‐to‐one orthology between the human and mouse PRAME genes (Birtle et al, 2005) has thus far precluded the development of suitable knockout mouse models for studying PRAME function.
Notably, it is becoming clear that many proteins associate with the promoter regions of genes, but their depletion does not seem to cause significant or widespread transcriptional defects. For example, the yeast Set1 histone methyl transferase is widely associated with active promoters, but deletion of the Set1 gene causes expression changes of 1.5‐fold or more in only about 20 genes (Miller et al, 2001). Similarly, the Jmjd3 histone demethylase was recently shown to associate with the promoters of LPS‐induced genes, but the expression of most of them was unaffected by knockout of Jmjd3 (De Santa et al, 2009). These observations are consistent with the model that some factors are not strictly required for the bulk of transcription, but instead have ancillary or regulatory functions that become apparent only in very specific conditions or subpopulations of cells.
While numerous reports established a role for PRAME in human cancers, several studies indicated that PRAME‐like genes have roles in the early stages of spermatogenesis (Wang et al, 2001) and oogenesis (Dadé et al, 2003), as well as in embryonic development and embryonic stem cells (Bortvin et al, 2003). Recently, overexpression of Pramel7 in E14 embryonic stem cells was reported to maintain a pluripotent state in the absence of the anti‐differentiation factor LIF (Cinelli et al, 2008). These findings are consistent with the notion that stem cells and neoplastic tissues share many properties, and several oncogenic pathways were recently found to also regulate self‐renewal mechanisms in stem cells (Reya et al, 2001). Interestingly, recent studies suggest important roles for NFY in stem cell biology. Knockout mice demonstrated that NFYA is essential for early mouse development (Bhattacharya et al, 2003), and NFYA was found to promote the self‐renewal of haematopoietic stem cells (Zhu et al, 2005). Moreover, a systematic analysis of cis‐regulatory elements identified the CCAAT box and NFY as crucial regulators of both mouse and human stem cells‐specific genes, and NFY was involved in the maintenance of the proliferative capacity of ES cells (Grskovic et al, 2007).
Taken together, our data unambiguously show that the PRAME oncoprotein is part of a Cullin2‐based ubiquitin ligase complex and that PRAME localizes to active NFY promoters, paving the road for understanding the functions of PRAME in normal cells and in human malignancies. Future experiments will address the role of the PRAME ubiquitin ligase at TSSs, its role in transcription of NFY‐regulated genes, and its contribution to the oncogenic properties of PRAME.
Materials and methods
Cell culture and stable cell lines
HEK293 cells were cultured in DMEM and K562 in RPMI medium (Gibco, Invitrogen) at 37°C in 5% CO2. Both media were supplemented with 5% Glutamax, 10% fetal bovine serum, 100 U/ml of penicillin, 100 μg/ml of streptomycin (Gibco, Invitrogen). Stable HEK293–HA–PRAME cells were generated using pcDNA5/FRT/HA–PRAME and the Flp‐in system (Invitrogen). K562 stable cell lines were generated with retroviral constructs according to the following 5 days protocol. On the first day, HEK293T cells at 50–60% confluence in a 6‐cm dish were transfected using Lipofectamine2000 (Invitrogen) and 1.8 μg of MD‐MLV, 300 ng VSV‐G, 3.6 μg retroviral construct, 100 ng pGFP. After 24 h (day 2), the virus‐containing supernatant was harvested, centrifuged for 10 min at 3000 r.p.m. and used for a first round of transduction of 106 K562 cells in the presence of 8 μg/ml of Polybrene (Sigma) for 24 h. A second round of transduction was performed on day 3, and on day 4, the target cells were transferred to fresh RPMI and selection was started on day 5 with either 6–10 μg/ml of puromycin (Sigma) or 1–2 mg/ml Zeocin (Invitrogen).
Antibodies and western blot
PRAME antisera and affinity‐purified antibodies were generated by immunizing rabbits with the previously described peptide FPEPEAAQPMTKKRKVDG (Epping et al, 2005) (V27 serum). Mouse monoclonal HA‐12CA5 and Myc‐9E11 were produced in‐house from hybridoma cultures. Commercial antibodies used were mouse monoclonal FLAG‐M2 (Sigma), c‐Myc monoclonal (Roche Applied Science), rabbit HA (Bethyl Laboratories, or Abcam ab9110), mouse monoclonal HA.11 (Covance MMS‐101P), rabbit Cul2 (Zymed Laboratories 51‐1890), mouse ElonginC (BD Transduction Laboratories 610760), goat ElonginB (Santa Cruz P‐16, sc‐23407), α‐tubulin (Santa Cruz B‐7, sc‐5286), HDAC2 (Biomol SA‐402).
Proteins were separated by conventional SDS–PAGE or NuPAGE 4–12% Bis–Tris gels (Invitrogen) run with MES buffer. Western blots were visualized by ECL (GE Healthcare) or Odyssey (LiCor).
Plasmids and cloning
Full‐length human PRAME was PCR amplified and cloned into NotI and XhoI sites of pcDNA5/FRT/TO/StrepII‐2 × TEV–Myc‐3 × HA (Le Guezennec et al, 2006) to generate pcDNA5–TAG–PRAME. A BamHI–XhoI fragment encoding TAG–PRAME was further subcloned into LZRS(Zeo) to generate the retroviral expression vector LZRS(Zeo)–TAG–PRAME. Full‐length human PRAME (ATCC) was subcloned with N‐terminal HA epitope tags into pcDNA5/FRT (Invitrogen). Mutagenesis to generate PRAME mutants was performed using the QuikChange II XL site‐directed mutagenesis kit (Stratagene) according to the manufacturer's instructions. To create the pTTE retroviral vector for ChIP, the ProtA‐2 × TEV–Myc cassette of the pZXN plasmid (Le Guezennec et al, 2006) was removed by digestion with BamHI and SalI. The resulting backbone was ligated to overlapping synthetic oligos with BamHI and XhoI overhangs, encoding two consecutive TY1 epitopes, a single‐ERalpha epitope and an extended MCS. PRAME was cloned using the EcoRI site to generate pTTE–PRAME plasmid; EloC was cloned using BglII and XhoI to generate pTTE–EloC plasmid. All constructs were checked by sequencing.
Gel filtration analysis
Size exclusion chromatography was performed with a Superose 6 column on a SMART system (Amersham Pharmacia Biotech) with buffer containing 300 mM NaCl, 10% glycerol, 20 mM TRIS, 0.2 mM EDTA, 1 mM DTT and 1 mM PMSF. In all, 50 μl nuclear extract was separated in 100 μl fractions and analysed by SDS–PAGE and western blot.
Purification of protein complexes from K562 cells
K562 cells were grown in spinner flasks and harvested at a confluence of about 106 cells/ml. For each protein extraction, about 1010 cells were centrifuged for 15 min at 1600 r.p.m. and washed with PBS. Cell pellets were resuspended in 2.5 volumes of hypotonic buffer T1 (10 mM Hepes pH 7.9, 10 mM KCl, 0.1 mM MgCl2, 0.1 mM EDTA pH 8, 1 mM DTT, 10% glycerol) supplemented with phosphatase and protease inhibitors (1 mM NaVO4, 1 mM NaF, 1 mM PMSF, Protease inhibitor cocktail (Roche)). After 15 min of incubation on ice, the cell membranes were disrupted with 20 strokes of Pestle A, or until 80–90% of the cells were stained by trypan blue. After centrifugation for 20 min at 3000 r.p.m. at 4°C, the supernatant was harvested and regarded as the cytoplasmic fraction. The pellet containing the nuclei was resuspended in 1 volume of T2 buffer (20 mM Hepes pH 7.9, 420 mM NaCl, 1.5 mM MgCl2, 0.1 mM EDTA pH 8, 1 mM DTT, 10% glycerol) supplemented with inhibitors, and incubated in rotation for 1 h at 4°C to extract the nuclear proteins. The suspension was cleared by ultracentrifugation for 30 min at 100 000 g at 4°C (Sorvall RC M150 GX). The supernatant was the nuclear fraction. The quality of the cytoplasmic/nuclear extractions was determined by western blot of α‐tubulin (cytoplasmic marker) and HDAC2 (nuclear marker).
About 5–10 ml (60–120 mg) of nuclear extracts were incubated overnight at 4°C with anti‐Myc (9E11) antibodies covalently crosslinked to protA‐sepharose CL‐4B (GE Healthcare) (see below). The following day, beads were washed three times with 10 ml buffer IPP300/0.5 (300 mM NaCl, 10 mM Tris–HCl pH 8, 0.5 mM EDTA and 0.5% NP‐40), and the protein complexes were eluted with 20 μg/ml of Myc peptide in a thermoshaker for 30 min at 28°C.
Crosslinking of antibody to beads
ProtA‐sepharose CL‐4B (GE Healthcare) or protG‐sepharose (Sigma) beads were incubated with antibody in PBS at room temperature for 1–2 h, then washed once with 10 volumes of sodium borate solution (0.2 M, pH 9). Chemical crosslinking of the antibodies to the beads was performed resuspending the beads in a sodium borate solution containing 5.2 mg/ml DMP (dimethyl‐pimelimidate; Sigma) and incubating for 30 min at room temperature with gentle rotation. The beads were washed once in 10 volumes ethanolamine (0.2 M, pH 8), incubated with ethanolamine for 2 h at room temperature, washed twice with glycine (0.1 M, pH 3) to remove uncrosslinked antibodies, and eventually with PBS to neutralize the pH.
Mass spectrometry and emPAI calculation for K562 complexes
The eluates of large‐scale immunoprecipitations from K562 cells were concentrated by speedvac and run shortly (∼1 cm) on SDS–PAGE gels to remove detergents and the excess of peptide. The gel lane was then fixed, cut in small pieces, reduced and alkylated. Proteins were digested overnight with trypsin (Promega) and eluted from the gel with trifluoroacetic acid. Peptides were sequenced using a nano‐high‐pressure liquid chromatography Agilent 1100 nanoflow system connected online to a 7‐T linear quadrupole ion trap‐Fourier transform mass spectrometer (Thermo Electron, Bremen, Germany) essentially as described (Olsen et al, 2004). Proteins were identified using the Mascot search algorithm (Matrix Science) with IPI human database v3.37, which contains 69164 unique entries. First ranked peptides (Mascot peptide scores ⩾20) were parsed from Mascot database search html files with MSQuant (http://www.msquant.sourceforge.net) to generate unique first ranked peptide lists that were further filtered for absolute calibrated mass error <5 and peptide δ score >5. To remove redundancy at the protein level and to uniquely assign peptides to one IPI Db entry, the peptides were remapped to the IPI human Db using the program Protein Coverage Summarizer (http://ncrr.pnl.gov/software/) as described (Lasonder et al, 2008). To determine the protein abundance, we computed the exponentially modified Protein Abundance Index (emPAI) values for our data (Ishihama et al, 2005; Lasonder et al, 2008); emPAI values for all proteins were calculated as 10PAI–1 (PAI=observed peptides/observable peptides). The number of observable peptides was calculated by in silico tryptic digestion of the IPI human Db v3.37 using the program Protein Digestion Simulator (http://ncrr.pnl.gov/software/).
Purification of protein complexes from 293 cells and MudPIT analysis
To prepare protein samples for mass spectrometry, 293‐HA–PRAME and control wild‐type cells were grown to 70–80% confluence in 10–20 roller bottles then washed, lysed and subjected to anti‐HA agarose immunoaffinity chromatography as described (Mahrour et al, 2008). TCA‐precipitated proteins were urea‐denatured, reduced, alkylated and digested with endoproteinase Lys‐C (Roche) followed by modified trypsin (Roche) as described (Washburn et al, 2001; Florens and Washburn, 2006). Peptide mixtures were loaded onto 100 μm fused silica microcapillary columns packed with 5 μm C18 reverse phase (Aqua, Phenomenex), strong cation exchange particles (Partisphere SCX, Whatman), and reverse phase. Loaded microcapillary columns were placed in‐line with a Quaternary Agilent 1100 series HPLC pump and a LTQ ion trap mass spectrometer equipped with a nano‐LC electrospray ionization source (ThermoFinnigan). Fully automated 10‐step MudPIT runs were carried out on the electrosprayed peptides, as described (Florens and Washburn, 2006). Tandem mass (MS/MS) spectra were interpreted using SEQUEST against a database of 61 430 sequences, consisting of 37 742 human proteins (downloaded from NCBI on 2008‐03‐04), 177 usual contaminants (such as human keratins, IgGs and proteolytic enzymes), and to estimate false discovery rates, 30 716 randomized amino‐acid sequences derived from each non‐redundant protein entry. Peptide/spectrum matches were sorted and selected using DTASelect with the following criteria set: spectra/peptide matches were only retained if they had a DeltCn of at least 0.08, and minimum XCorr of 1.8 for singly, 2.0 for doubly and 3.0 for triply charged spectra. In addition, peptides had to be fully tryptic and at least 7 amino acids long. Combining all runs, proteins had to be detected by at least two such peptides, or one peptide with two independent spectra. Under these criteria, the final false discovery rate at the peptide level is 0.151±0.009. Peptide hits from multiple runs were compared using CONTRAST. To estimate relative protein levels, we took advantage of evidence that normalized spectral abundance factors (NSAFs) vary with the relative abundance of proteins in a sample. NSAFs were calculated for each detected protein, using the formula:
Expression of recombinant proteins in Sf21 insect cells
cDNAs encoding wild‐type or mutant PRAME, Cul2, Elongins B and C were subcloned into pBAcPAK8 and recombinant baculoviruses were generated. SF21 cells were cultured at 27°C in SF‐900 medium supplemented with 10% fetal bovine serum. SF21 cells were co‐infected with the recombinant baculoviruses indicated in the legend to Figure 2. At 50 h after infection, cells were collected, lysed and processed as described (Mahrour et al, 2008).
In vitro ubiquitination assays
The PRAME–Elongin BC complex and the His‐T7‐Xpress‐Cullin2/Myc–RBX1 complex were obtained by co‐infecting Sf21 cells with baculoviruses encoding for each individual protein and purifying through Flag tag or nickel‐agarose chromatography, respectively. Different combinations of these complexes were then incubated for 1 h at 37°C with ∼50 nM Uba1, ∼0.2 μM UbcH5a and 0.25 mg/ml ubiquitin, in 20 μl reactions containing 50 mM Tris–HCl pH 7.6, 50 mM NaCl, 5 mM MgCl2, 0.5 mM EDTA (pH 7.9), 5% glycerol, 2 mM ATP, 0.5 mM dithiothreitol, 0.3 units/ml pyrophosphatase, 60 mM creatine phosphate and 0.3 mg/ml creatine phosphokinase. N‐terminally (His)6‐tagged mouse E1 was expressed in Sf21 insect cells and purified by nickel‐agarose chromatography as described. Human UbcH5a with an N‐terminal 6‐histidine tag and a C‐terminal Flag tag were expressed in Escherichia coli strain BL21(DE3) and purified by nickel agarose.
ChIP and qPCR
Chromatin harvests, ChIPs and qPCR analyses were performed as previously described (Denissov et al, 2007), with the following minor adjustments. Cells were crosslinked for 15 min in 1% formaldehyde, they were resuspended in ChIP incubation buffer at a concentration of 16 millions/ml and sonicated using a Bioruptor sonicator (Diagenode) for 15 min with settings: high power, 30 s ON, 30 s OFF. For every ChIP reaction, 200 μl of sonicated chromatin were used. The immunoprecipitated chromatin was eluted with 200 μl of elution buffer at room temperature for 20 min and DNA was purified with QIAquick PCR purification kit (Qiagen). For ChIP‐seq, DNA eluted from 5 to 7 ChIP reactions was pooled and the DNA concentration was measured with a Qubit fluorometer and Quant‐iT dsDNA HS Assay (Invitrogen).
The following antibodies were used for ChIP: preimmune serum (preV27, ‘no antibody’ control) and PRAME antiserum (V27); NFYA (Santa Cruz H‐209, sc‐10779); NFYB (pAb‐007‐100) and BB2 (against TY1 epitope) were from Diagenode.
For sequential ChIP experiments (ChIP–re‐ChIP), in order to avoid antibody carry‐over, the first ChIP was performed with antibody covalently crosslinked to ProteinA/G‐Sepharose beads (Santa Cruz) with DMS as described above. Immunoprecipitated chromatin was then eluted in a smaller volume and further processed as a normal ChIP with PRAME immune serum or the preimmune serum as negative control. Real‐time qPCR was performed using the SYBR Green mix (Bio‐Rad) with the MyIQ thermocycler (Bio‐Rad). Specific primers for qPCR (Biolegio) were designed as previously described (Denissov et al, 2007) and are listed in Supplementary Table SII. Primers amplifying exon 12 of albumin were used as negative control.
RNA‐seq library preparation and data analysis
Total RNA was extracted from K562 cells with the RNeasy kit and on‐column DNase treatment (Qiagen). In all, 100 μg of total RNA were subjected to two rounds of poly(A) selection using the Oligotex mRNA Mini Kit (Qiagen) and the concentration of poly(A)+ RNA was measured with a Qubit fluorometer (Invitrogen). In all, 200 ng of purified poly(A)+ RNA were diluted in 160 μl RNAse‐free water and fragmented by addition of 40 μl 5 × fragmentation buffer (200 mM Tris acetate pH 8.2, 500 mM potassium acetate and 150 mM magnesium acetate) and incubation at 94°C for 120 s. After purification with RNeasy Minelute Kit (Qiagen), the quality of fragmented RNA was analysed on an Experion (Bio‐Rad). Double‐stranded cDNA was synthesized from the fragmented RNA with SuperscriptIII (Invitrogen) using random hexamers, and then used for Illumina sample prepping and sequencing (see below).
A total of 7 246 489 RNA‐seq reads were uniquely mapped to hg18 and used for bioinformatical analysis. Normalized gene expression values (NE) for RefSeq genes were computed with Genomatix Region Miner tools.
Illumina high‐throughput sequencing
A starting amount of 10–20 ng of ChIP DNA or double‐stranded cDNA was used for sample prepping according to the standard protocols. Briefly, end repair was performed with T4 DNA ligase, Klenow DNA polymerase and T4 PNK; a 3′ protruding adenosine was added using Klenow DNA polymerase, and T‐overhang adaptors were ligated. DNA molecules with a size of 270–300 bp were selected from gel, purified and amplified with 14 cycles of PCR. Quality controls were performed by qPCR of DNA before and after sample prepping, and by analysis of the PCR products with an Experion (Bio‐Rad). Cluster generation and sequencing‐by‐synthesis (35 bp) were performed using the Illumina Genome Analyser II according to standard protocols (Illumina). The image files were processed to extract DNA sequence data, and sequences were mapped to the human hg18 reference genome using the eland program allowing at most one base mismatch. Sequence reads were filtered based on quality controls and sequences that could not be mapped uniquely on the reference genome were discarded. For visualization purposes, ChIP‐seq reads were directionally extended to the length of the fragments used for sequencing and the number of overlapping sequence reads was determined for each basepair in the genome, averaged over a 10‐bp window, and visualized in the University of California Santa Cruz genome browser (http://genome.ucsc.edu). ChIP‐seq and RNA‐seq data are published in the GEO database (GSE26439), and in the published history ‘PRAME’ in Galaxy (http://main.g2.bx.psu.edu/).
ChIP‐seq analysis and peak detection
ChIP‐seq experiments generated the following number of uniquely mapped reads: 5.5 millions for PRAME, 5.6 millions for preimmune (control), 8.9 millions for NFYA and 8.8 millions for NFYB. In order to compensate for the small difference in sequencing depth between the PRAME and preimmune data sets, the number of tags of preimmune ChIP‐seq was equalized to the number of tags of PRAME ChIP‐seq by uniform removal of tags as described (Nielsen et al, 2008). PRAME‐binding sites (peaks) were subsequently identified with the MACS peak calling algorithm (Zhang et al, 2008) using the two‐sample method with a P‐value (e) of 10−6, which compared the PRAME profile to the preimmune (control) profile. NFYA‐ and NFYB‐binding sites were determined with MACS using the standard method with extended lambda (10 kb, 50 kb 100 kb) and P‐value (e) of 10−6. MACS identified a total of 14 132 PRAME peaks, 25 151 NFYA peaks and 16 350 NFYB peaks. NFYA and NFYB peaks were combined in a total of 28 844 NFY‐binding sites that were tested for tag density. The number of sequence reads were counted in 500 bp regions centred on the summits of the 28 844 NFY‐binding sites (with the summit being the site of highest enrichment as determined by MACS), and the data were visualized in dot plots using R.
Location analysis and motif discovery
To analyse the genomic distribution of the binding sites identified, we first determined the fraction of peak summits localized within a 2‐kb window centred around the TSS of primary transcripts (TSS±1 kb) using the Genomatix Genome Inspector. The genomic distribution of the binding sites located outside promoter regions (‘distal sites’) relative to genes was analysed with PinkThing (http://pinkthing.cmbi.ru.nl/). The background distribution of genomic DNA was determined applying the same approach to pools of random genomic sites (n>33 000).
The distances of promoter‐associated binding sites from TSSs were computed with the Genomatix Genome Inspector tool, and plots were generated with R.
The localization of PRAME at enhancer elements was determined by overlaps between PRAME peaks summits and enhancer regions recently predicted for K562 cells (Heintzman et al, 2009). The genomic coordinates of the published enhancer regions were converted from hg17 to hg18 with the Galaxy Lift‐over function, resulting in 24 545 enhancers.
De novo motif discovery was performed using the GimmeMotifs pipeline (van Heeringen and Veenstra, 2011), which incorporates a number of widely used motif analysis tools, including MDmodule, MEME, Weeder, MotifSampler and BioProspector. Based on the distribution of PRAME‐binding sites, we performed motif searches on the DNA sequences underlying PRAME summits located within −400 bp and +100 bp of a primary TSS (n=4981). A CCAAT motif scored the highest values of enrichment and MNCP scores, as compared with pools of background sequences with the same genomic distribution.
ENCODE data sets
We retrieved and analysed publicly available ChIP‐seq data sets of H3K9ac, H3K4me3, H3K36me3, H3K27me3 and RNA‐PolII generated from K562 cells by the ‘ENCODE Chromatin Group at Broad Institute and Massachusetts General Hospital’ (http://hgdownload.cse.ucsc.edu/goldenPath/hg18/encodeDCC/wgEncodeBroadChipSeq/).
Starting from 27 903 RefSeq genes with a RNA‐seq expression value, we identified 7214 genes longer than 1 kb with a NFY peak summit in the proximal promoter region. Next, we quantified the signals of PRAME in a 500‐bp window under the NFY‐binding site, H3K9ac and H3K4me3 in a 3‐kb window, H3K36me3 and H3K27 in the gene body and RNA‐PolII in the gene body excluding the first 500 bp.
Supplementary data are available at The EMBO Journal Online (http://www.embojournal.org).
Conflict of Interest
The authors declare that they have no conflict of interest.
We are grateful to B Scheijen for plasmids and protocols, A Kaan for technical assistance with DNA cloning, S Hall for assistance with protein expression and purification, A Gottschalk for recombinant NFY, E Janssen‐Megens and KJ Francoijs for the Illumina sequencing. We are indebted to S van Heeringen, A Brinkman, H Marks and W Welboren for scripts and support with the bioinformatical analyses. We also thank our colleagues for helpful discussions and suggestions. Work in the authors’ laboratories was funded by Dutch Cancer Society KWF (to HS), the Stowers Institute for Medical Research and grant R37 GM41628 (to RCC) from the National Institute for General Medical Sciences (NIGMS).
Authors contributions: AC and NM designed and performed the experiments and contributed equally to this study. ET, RS, MAS and DS performed the experiments. PWJ, SM‐B, MPW and LF performed mass spectrometry analyses. JWC, RCC and HGS conceived strategies and supervised the project. AC drafted the manuscript, which was edited by NM, JWC, RCC and HGS.
- Copyright © 2011 European Molecular Biology Organization