Endoproteolytic processing of C-terminally truncated NF-κB2 precursors at κB-containing promoters

DNA-Binding Activity of p100ΔCs Is Required for Their Constitutive Processing.

It has been demonstrated that p100ΔCs, the constitutive processing forms, are predominantly located in the nucleus (8, 10, 11, 14). However, whether the nuclear translocation of p100ΔCs contributes to their processing and where p100ΔCs are processed are currently unknown. Because the constitutive processing of p100ΔCs also requires their dimerization domain (9), a domain involved in DNA binding of NF-κB, we hypothesized that the nuclear translocation of p100ΔCs enabled binding to κB promoter DNA and that this binding might be necessary for their processing. To test this hypothesis, we initially examined whether p100ΔCs directly bind to DNA by EMSA using a T cell lymphoma cell line, HuT78. HuT78 cells contain a genetically rearranged nf-κb2 gene that expresses a p100ΔC with a molecular mass of ≈80 kDa, therefore named p80HT (Fig. 1A) (10, 11, 15).

As expected, HuT78 cells showed constitutive κB DNA-binding activity (Fig. 1B, also see Fig. 2D by using a different κB site for the assay). It appeared that both p80HT and its processed product p52 bound to κB DNA, because anti-p100N antibody could supershift all of the three NF-κB/DNA complexes (Fig. 1B, lane 2; also see Fig. 2B, lane 2). Because available antibodies for p100 do not exclusively recognize the C termini of p100ΔCs, we generated C-terminally hemagglutinin (HA)-tagged p80HT and p100 (1–665), another p100ΔC capable of constitutive processing (6, 14). As shown in Fig. 1C, anti-HA antibody, like anti-p100N antibody, also efficiently supershifted NF-κB/DNA complex (lanes 6–7 and 13–14), indicating that these p100ΔCs, together with their processed product p52, indeed bound to the κB DNA.

To examine the role of the DNA-binding activity of p100ΔCs in their constitutive processing, we substituted R₅₂FRYGCE_58, the core amino acids within the DNA-binding domain of p100ΔCs, with L₅₂FLFGPV₅₈ (Fig. 1A). In agreement with previous structural studies (16), mutation of these amino acids ablated the DNA binding of p100ΔCs (Fig. 1D, lanes 3 and 6) but had no effect on their nuclear localization (Fig. 1E). Importantly, the constitutive processing of p100ΔCs was completely abolished by this mutation (Fig. 1F, lanes 3 and 5). In sharp contrast, these amino acids were dispensable for the inducible processing of full-length p100 (Fig. 1F, lane 9). In fact, deletion of the first 70 aa from the N-terminal end did not affect the inducible processing of p100 (7). This result is consistent with our previous findings that inducible processing of p100 occurs in the cytoplasm (9). Taken together, these studies demonstrated that p100ΔCs directly bind to the κB promoter and that this binding activity is essential for their constitutive processing.

p100ΔCs but Not the Processed p52 Recruit the Proteasome to the κB Promoter to Form a Stable Proteasome/p100ΔC/κB Promoter Complex.

Because the proteolytic activity of the proteasome is required for the constitutive processing of p100ΔCs (6), the proteasome itself should be recruited to the same promoter if the constitutive processing of p100ΔCs occurs at the κB promoter. To test this possibility, we used the Bcl-xL promoter as a model system (Fig. 2A), because the Bcl-xL gene is a well known target of p52 (17, 18). First, we tested whether p80HT binds to the Bcl-xL promoter by EMSA using nuclear extracts from HuT78 cells or 293 cells expressing a C-terminally HA-tagged p80HT. As expected, both endogenous and transfected p80HT specifically bound to the Bcl-xL promoter in vitro (Fig. 2 B and C).

Next, we performed ChIP assays to determine whether p80HT binds to the Bcl-xL promoter in vivo. We found that the Bcl-xL promoter was immunoprecipitated from HuT78 cells by the anti-p100N (Fig. 2D, lane 3), suggesting that p80HT and/or its processed p52 indeed bound to the κB promoter in vivo. Interestingly, we found that antibodies either for the α6 subunit of the 20S proteasome or for the Rpn7 subunit of the 19S proteasome also specifically precipitated the Bcl-xL promoter from the HuT78 cells (Fig. 2D, lanes 4 and 5), demonstrating that the 26S proteasome was associated with the Bcl-xL promoter under physiological conditions.

To investigate whether p80HT was responsible for recruitment of the proteasome to the Bcl-xL promoter, we repeated the ChIP assay using 293 cells expressing the C-terminally HA-tagged p80HT or p52. Consistent with the EMSA (Fig. 2 B and C), our ChIP assay demonstrated that both p80HT and p52 bound to the Bcl-xL promoter in vivo (Fig. 2E Top and Middle, lane 3). However, the proteasome associated with the Bcl-xL promoter only in cells expressing p80HT but not in those expressing p52 (Fig. 2E Top and Middle, lane 4). Thus, it is p80HT, but not p52, that specifically recruits the proteasome to the Bcl-xL promoter. We further demonstrated that p80HT-mediated recruitment of the proteasome to the Bcl-xL promoter requires its κB DNA-binding activity, because the proteasome failed to associate with the Bcl-xL promoter in the cells expressing the p80HT DB mutant, a DNA-binding-deficient mutant of p80HT (Fig. 2E Bottom, lane 4). Similar results were obtained when different NF-κB target promoters and p100ΔCs, such as IL2Rα promoter and LB40, were used (Fig. 2F and data not shown), suggesting a general mechanism for the recruitment of the proteasome to the κB promoters. LB40 is the p100 (1–702) mutant that was originally identified in B cell chronic lymphocytic leukemia (B-CLL) (15). Taken together, these results clearly demonstrated that recruitment of the proteasome to the κB promoters is mediated by p100ΔCs but not by their processed p52 and that the recruitment depends on specific binding of p100ΔCs to the κB promoters.

To investigate the mechanism whereby the proteasome is recruited to the κB promoter by p100ΔCs, we assessed possible direct interactions between the proteasome and p80HT. In vitro-translated p80HT bound to the purified proteasome weakly in the absence of κB DNA, similar to the weak interaction observed with the DB mutant (Fig. 2G, lanes 3 and 5). However, addition of a κB DNA oligonucleotide, but not Oct-1 DNA oligonucleotide, significantly enhanced the interaction between the proteasome and p80HT (Fig. 2G, lanes 3–4 and 7–8). On the other hand, the κB DNA oligonucleotide had no effect on the association between the proteasome and p80HT DB mutant (Fig. 2G, lane 6). In agreement with its inability to recruit the proteasome to the κB promoter (Fig. 2 E and F), p52 also failed to interact with the proteasome in the presence of the κB DNA oligonucleotide (Fig. 2G, lane 1). These data demonstrated that recruitment of the proteasome to the κB promoter and subsequent formation of the stable proteasome/p100ΔC/κB promoter complex requires binding of p100ΔCs to both proteasome and κB promoter.

Constitutive Processing of p100ΔCs Requires the κB Promoter DNA.

To investigate whether the formation of the stable proteasome/p100ΔC/κB promoter complex is essential for the processing of p100ΔCs, we used the purified proteasome and in vitro-translated p100ΔC proteins to recapitulate the in vivo processing of p100ΔCs. As expected, both 20S and 26S proteasomes failed to mediate the processing of p100 (1–665) and p80HT in the absence of the κB DNA (Fig. 3, lanes 1). However, addition of the κB DNA oligonucleotide significantly stimulated proteasome-mediated processing of both p100 (1–665) and p80HT in a dose-dependent manner (Fig. 3 A, lanes 2–5; B, lane 3; C, lanes 2–5). The role of the κB DNA is specific, because the Oct-1 DNA had no effect on the processing of p100ΔCs (Fig. 3A, lane 6; B, lane 2; C, lane 6). These results were further confirmed by immunoblotting (IB) assay using the anti-p100N antibody (Fig. 3D). Collectively, these data provided direct evidence demonstrating that the proteasome-mediated processing of p100ΔCs occurs in association with κB promoter DNA.

Constitutive Processing of p100ΔCs Is Initiated by a Proteasome-Mediated Endoproteolytic Cleavage.

Having demonstrated that the proteasome-mediated processing of p100ΔCs occurs at the κB promoter, we next addressed the mechanism by which the C-terminal fragments of p100ΔCs are degraded by the proteasome, namely, whether the constitutive processing of p100ΔCs is initiated by proteasomal endoproteolytic cleavage or through exoproteolytic degradation from their C-terminal ends. For these studies, we fused the hyperstable GFP or p100 N-terminal region p100 (1–405) into the C terminus p100 (1–665) to prevent the possible exoproteolytic degradation from the C-terminal end (Fig. 4A), because the proteasome prefers to degrading unfolded protein peptides instead of highly folded proteins. If the constitutive processing of p100ΔCs is initiated by the C-terminal nibbling, the chimeras should not be processed. If the processing is initiated by proteasome-mediated endoproteolytic cleavage, it is possible that the ankyrin repeats located in the middle of the chimeras might still be degraded, whereas the fused GFP or p100 (1–405) may be resistant to proteasomal degradation. To distinguish two possible processed products of the p100 (1–665)-p100 (1–405) chimera: p52 from the N terminus and p100 (1–405) from the C terminus, we fused HA-tag at the C terminus of the chimera (Fig. 4A).

As shown in Fig. 4B, the p100 (1–665)-GFP chimera was processed to p52 and GFP, because both p52 and GFP were simultaneously detected in cells expressing p100 (1–665)-GFP. Similarly, the p100 (1–665)-p100 (1–405)-HA chimera was processed to p52 and p100 (1–405)-HA. Of note, the molar ratios of p52/p100 (1–405)-HA generated from chimera p100 (1–665)-p100 (1–405)-HA and p52/GFP from chimera p100 (1–665)-GFP were ≈1:1 (Fig. 4C), suggesting no exoproteolytic degradation from the C-terminal ends. These results suggested that the constitutive processing of p100ΔCs is initiated by proteasome-mediated endoproteolysis, followed by the progressive degradation of the cleaved C-terminal fragments.

Amino Acid D₄₁₅ of p100ΔCs Is Required for Their Constitutive Processing.

To identify potential proteasome-mediated endoproteolytic cleavage sites within p100ΔCs, we analyzed the amino acid sequence of p100ΔCs using ExPASy proteomics tools (www.expasy.org/tools/). We found that the amino acid sequence S₄₁₃RDSGEEAAEPASPS₄₂₇ within p100ΔCs was a preferential cleavage site for the proteasome. To investigate the significance of these amino acids in the processing of p100ΔCs, we deleted them from p100 (1–665) by mutagenesis (Fig. 5A). As expected, deletion of these amino acids completely blocked the constitutive processing of p100 (1–665) in vivo (Fig. 5B, lane 3). To further pinpoint the minimal amino acids required for the processing, we performed progressive deletions, and found that four amino acids, S₄₁₃RDS₄₁₆, were actually required (Fig. 5B, lanes 4–6). Among these four amino acids, D₄₁₅ is the most common proteasomal cleavage site and the only conserved amino acid among different species (ref. 19 and Fig. 5A). Remarkably, substitution of this single amino acid with alanine was sufficient to prevent the constitutive processing of p100ΔCs both in vivo and in vitro (Fig. 5 B, lane 7; C, lanes 2 and 3). D₄₁₅ was also required for the constitutive processing of the p100 (1–665)-GFP and p100 (1–665)-p100 (1–405)-HA chimeras (Fig. 5D, lanes 3 and 6). However, deletion of S₄₁₃RDS₄₁₆ did not affect p100ΔC nuclear translocation and recruitment of the proteasome to the bound κB promoters to form a stable complex (Fig. 5 E–G). Thus, amino acid D₄₁₅ is essential for the constitutive processing of p100ΔCs and may function solely as the endoproteolytic site for the initiation of p100ΔC constitutive processing.

Processed p52 Controls Expression of a Subset of Tumor-Associated Genes and Induces Anchorage-Independent Cell Growth.

Although p100ΔCs are oncogenic, the mechanisms involved remain unknown. Specifically, it is not clear whether oncogenesis is mediated directly by p100ΔCs themselves or indirectly by their processed product p52. To address this important issue, we analyzed the transforming ability of LB40, LB40Δ413–416 (defective in processing) and p52 by performing colony-formation assays. We used both rat embryonic fibroblast (REFs) and mouse FL5.12 pro-B cells for these assays. As expected, REFs stably expressing an empty retroviral vector failed to form any foci, whereas the LB40-expressing cells generated a significant number of colonies [see supporting information (SI) Fig. 6A and Table 1]. No foci were formed by the cells expressing LB40Δ413–416, implying an essential role of proteolytic processing in transformation. Furthermore, the number and size of foci produced by p52-expressing cells were comparable with those of LB40 expressing cells. Similarly, the cloning efficiency of the LB40 FL5.12 cells was significantly greater than that of the LB40Δ413–416-expressing or control cells, although a low background of colony formation was observed. Moreover, the colonies in the LB40 populations were much larger than those observed in the LB40Δ413–416 or control cells. Surprisingly, the colony forming efficiency of p52 in the B cells was two to three times higher than that of LB40. The different clonogenic abilities of LB40, LB40Δ413–416, and p52 in both REF and FL5.12 cells were not due to their expression levels, because they were comparable (SI Fig. 6B). These results indicated that the processed p52, but not the p100ΔCs themselves, can induce transformation anchorage independent growth.

To confirm that the oncogenicity of p100ΔCs is mediated indirectly by their processed product, p52, we performed microarray analysis to compare the gene expression profiles in the FL5.12 cells stably expressing LB40, LB40Δ413–416, and p52. Consistent with our soft-agar assay, the global gene expression profile in the LB40 FL5.12 cells was much more similar to that of p52-expressing cells than to that of cells expressing LB40Δ413–416 (SI Fig. 7). Most importantly, many tumor-associated genes were significantly induced by LB40 or p52 but not by the LB40Δ413–416 mutant (SI Table 2), suggesting that induction of these genes in the LB40-expressing cells is actually controlled by the processed p52. Consistent with the ChIP assays shown in Fig. 2, the expression levels of Bcl-xL and IL2Rα were also up-regulated in the LB40- or p52-expressing cells but not in the LB40Δ413–416 expressing cells. Taken together, these results strongly suggested that the processed product, p52, is responsible for induction of a number of tumor-associated genes, which in turn contribute to the oncogenicity of p100ΔCs.

A Unique Processing Mode of p100ΔCs.

Using HuT78, p100 (1–665) and LB40 as model systems, we have demonstrated a mechanism for p100ΔC constitutive processing. We have shown that p100ΔCs, like their processed product, p52, specifically interact with the κB promoter DNA in the nucleus (Figs. 1 and 2). Importantly, p100ΔCs, but not p52, are able to recruit the proteasome into the bound κB promoter (Fig. 2 E and F). The recruitment of the proteasome to κB promoter DNA, in turn, significantly enhances the interaction between the proteasome and p100ΔCs (Fig. 2G). Formation of a stable proteasome/p100ΔC/κB promoter complex is essential for the constitutive processing of p100ΔCs, because the p100ΔC mutants defective in DNA binding but still able to translocate to the nucleus and bind to the proteasome fail to recruit the proteasome to the κB promoter and undergo processing (Figs. 1 and 2). In further support, the constitutive processing of p100ΔCs in vitro requires κB DNA (Fig. 3).

Interestingly, the processing of p100ΔCs at the κB promoter is initiated by a proteasome-mediated endoproteolysis (Fig. 4). It seems that amino acid D₄₁₅ is the endoproteolytic cleavage site for the proteasome. Because amino acid D₄₁₅ is required for the endocleavage and subsequent processing of p100ΔCs but dispensable for p100ΔC nuclear translocation and subsequent recruitment of the proteasome to a κB promoter (Fig. 5), which are all essential for the constitutive processing of p100ΔCs (Figs. 1–4).

Of note, D₄₁₅, nuclear translocation, and DNA binding sequences of p100 are not required for its inducible processing (Fig. 1, SI Fig. 8, and refs. 8, 9, and 14), suggesting a fundamentally different mechanism for the constitutive processing of p100ΔCs, which is presented in SI Fig. 9. In brief, deletion of the C-terminal PID and/or ankyrin repeats of p100 leads to the translocation of the p100ΔC proteins into the nucleus and subsequent recruitment of the proteasome to form a stable complex at a κB promoter. Formation of the proteasome/p100ΔC/κB promoter complex may cause conformational changes of the p100ΔCs such that the region spanning amino acid D₄₁₅ is exposed and falls into the catalytic chamber of the proteasome. The proteasome then internally cleaves the p100ΔC proteins into two fragments at D₄₁₅. The N-terminal fragment, the processed product, p52, is left intact, whereas the C-terminal fragment is further degraded by the proteasome. The protection of the newly processed p52 from undergoing degradation could be due to occupation of the proteasome by the newly produced C-terminal fragment and/or the tight association of p52 with its NF-κB-binding partners as well as the κB promoter DNA. In further support of this model, we find that binding to DNA indeed makes p52 more stable (SI Fig. 10).

Constitutive Processing of p100ΔCs in Oncogenesis.

Partial loss of the C-terminal region of p100 leads to its constitutive processing and oncogenesis; however, the link in between has not been established yet. In this study, we find that the anchorage-independent growth induced by p52 is similar to or even higher than that by LB40, depending on the cell types used for the assay (SI Fig. 6 and SI Table 1). In sharp contrast, the LB40 mutant defective in processing but still fully capable of binding to DNA to regulate gene expression (Fig. 5, SI Fig. 7, and SI Table 2), completely loses this potential (SI Fig. 6). These results strongly suggest that the anchorage-independent growth, and likely oncogenicity of p100ΔCs is actually mediated indirectly by their processed product, p52.

Indeed, our microarray analysis indicates that many genes are significantly induced by both LB40 and p52 but not by the processing-deficient mutant of LB40, although the mutant is also involved in certain gene expression (SI Fig. 7 and SI Table 2). All these genes induced by p52 and LB40 are known to play roles in oncogenesis, and many of them have already been identified as NF-κB target genes. For instance, Bcl-xL, interleukin 2 receptor (IL2R), lymphotoxin α(LTα), integrin α6, regulator of G protein signaling 1 (Rgs1), and decay-accelerating factor 1 (DAF1) are well known NF-κB targets and have also been linked to tumorigenesis. Interestingly, cathepsin E, dopa decarboxylase (DDC), and Ia-associated invariant chain (Ii), three marker genes of gastric tumor (20–22), are also specifically induced in p52- or LB40-expressing cells (SI Table 2). Of note, constitutive expression of p52 in p100 knockin mice causes significant gastric and lymphocyte hyperplasia, although the early postnatal death of the mice may prevent further formation of tumors (13). These studies thus strongly suggest that, through induction of tumor-associated genes, the processed p52 functions as the mediator of p100ΔC-mediated oncogenesis (SI Fig. 9).

Transcriptional Regulation by Proteasomal Endoproteolysis at Promoter.

Through controlling protein stability, the proteasome is involved in a broad array of basic cellular processes, one of which is the regulation of gene transcription. Emerging evidence indicates that the proteasome can directly regulate gene transcription at promoters (23). However, in all cases studied, proteasome-mediated proteolysis invariably leads to the complete destruction of the transcription factors (23). In the present study, we have presented another mode of transcription regulation by proteasomal proteolysis. We have shown that the proteasome selectively degrades the C-terminal IκB-like domain of p100ΔCs to generate p52 at κB promoters (Figs. 1–4). Given the different abilities of p100ΔCs and their processed product p52 in gene transcription (SI Fig. 7 and SI Table 2), the endoproteolytic processing will directly result in the modulation of transcription at the promoters. These studies not only demonstrate the physiological and pathogenic significance of the endoproteolytic activity of the proteasome but also establish an immediate link between transcription and proteasomal endoproteolysis.

Plasimids and Reagents.

The antibodies for the α6 subunit of the 20S proteasome and the Rpn 7 subunit of the 19S proteasome as well as the purified human 20S and 26S proteasomes were from Biomol (Plymouth Meeting, PA). The p100 DB, D415A, and D415 internal deletion mutants were generated by routine cloning strategies or site-directed mutagenesis as described (24, 25). Other reagents and expression vectors have been described before (6, 14, 26).

Cell Culture and Transfection.

HeLa and 293 cells were cultured in DMEM supplemented with 10% FBS and 2 mM l-glutamine. HuT78 cells were maintained in suspension in RPMI medium 1640 supplemented with 10% FBS and 2 mM L-glutamine. FL5.12 pro-B cells were cultured in RPMI medium 1640 supplemented 10% FBS, 10% WEHI-3B cell-conditioned media (as a source of IL-3), 20 mM Hepes, and 2 mM L-glutamine. The 293 and HeLa cell transfections were described previously (9, 14).

Stable Transfectants.

FL5.12 pro-B cells were infected with pCLXSN empty virus or virus expressing p52, LB40, LB40Δ413–416, or LB40 DB. The stable transfectants were obtained by selection with G418 as described previously (27).

IB and Co-IP.

IB analysis was performed essentially as described (6). The amounts of cell lysates used for IB were ≈7 μg. For the in vitro protein–protein interaction, the translated p100ΔC proteins were incubated in the reaction buffer (50 mM Tris-HCl, pH 7.5/25 mM KCl/1 mM DTT/10 mM MgCl₂/5% glycerol) at room temperature in the presence or absence of the κB or Oct-1 DNA. After 25 min of incubation, the proteasome was added and the reactions were rotated at 4°C for 2 h, followed by normal Co-IP assay (6).

Immunofluorescence Assay.

HeLa cells were transfected with the indicated p100 constructs. After 24 h, the recipient cells were directly fixed, permeabilized, and subsequently incubated with anti-HA, followed by Texas red-conjugated anti-mouse secondary antibodies. The subcellular localization of stained proteins was detected by using an inverted fluorescence microscope. The cells were also counterstained with Hoechst 33258 for nuclear staining by detecting DNA (9).

EMSA.

Nuclear extracts from the indicated cells were prepared essentially as described (9, 28). EMSA was performed by incubating 5 μg of nuclear extracts with the [γ-³²P]ATP radiolabeled κB probes (5′-AGTTGAGGGGACTTTCCCAGGC-3′ for the consensus κB site; 5′-CGATAAAGGGACTTCCAAGAT-3′ for the κB site of the Bcl-xL promoter), or Oct-1 probe (5′-TGTCGAATGCAAATCCTCTCCTT-3′) at room temperature for 25 min, followed by resolving the DNA–protein complexes on native 6% polyacrylamide gels (28). For the supershift assay, 0.5 μl of the indicated antibody was added to the EMSA reaction for another 20 min before electrophoresis.

In Vitro p100 Processing Assay.

The pcDNA p100ΔCs were transcribed and translated in vitro with the TNT T7-coupled transcription/translation system (Promega, Madison, WI) in the presence of [³⁵S]methionine (PerkinElmer, Wellesley, MA) or cold methionine as described (4). The translated proteins were incubated in the reaction buffer (50 mM Tris·HCl, pH 7.5/25 mM KCl/1 mM DTT/10 mM Mg²⁺-ATP/5% glycerol) at room temperature in the presence or absence of the κB or Oct-1 DNA. After 25 min of incubation, the proteasome (25 nM proteasome for each reaction) was added. The reaction mixtures were incubated at 37°C and then stopped at the indicated time points by immediately adding the SDS loading buffer. The p100ΔC proteins and their processed products were fractionated by SDS/PAGE and visualized by autoradiography or IB using anti-p100N.

ChIP Assay.

ChIP assay was performed essentially as described (29). The primers used were: human Bcl-xL promoter, forward 5′-CGATGGAGGAGGAAGCAAGC-3′, reverse 5′-GCACCACCTACATTCAAATCC-3′; human and mouse actin promoter, forward 5′-TGCACTGTGCGGCGAAC-3′, reverse 5′-TCGAGCCATAAAAGGCAA-3′; mouse Bcl-xL promoter, forward 5′-ACAGATCCGAGGCTGTCTTC-3′, reverse 5′-CCCGGAGGTATGGGTTTAGT-3′; mouse IL2Rα promoter, forward 5′-CACGACCTTGCTTCTCAGTCT-3′, reverse 5′-TGTACAAGGAAAGGGGGATTC-3′. All the ChIP assays presented in this study were repeated by at least two independent experiments.

Colony Formation Assays.

Soft-agar assays were performed essentially as described (30). All of the colony formation assays presented in this study were repeated in at least three independent experiments, and each independent experiment included three different cell doses: 0.5 × 10³, 1 × 10³, and 2 × 10³ for FL5.12 cells and 5 × 10³, 5 × 10⁴, and 5 × 10⁵ for REFs.

Affymetrix Microarray Analysis.

Total RNAs were extracted from FL5.12 cells stably infected with pCLXSN or pCLXSN expressing p52, LB40, or LB40Δ413–416 (4). The integrity of the RNAs was examined by using the Agilent platform (2100 Bioanalyzer; Agilent Technologies, Palo Alto, CA). Biotin-labeled cDNA and subsequent hybridization and screening were performed by the Transcriptional Facility Shared Resource of the Cancer Institute of New Jersey (New Brunswick, NJ).