Chromosomal amplification at 3q is common to multiple human cancers, but has a specific predilection for squamous cell carcinomas (SCC) of mucosal origin. We identified and characterized a novel oncogene, SCC-related oncogene (SCCRO), which is amplified along the 3q26.3 region in human SCC. Amplification and overexpression of SCCRO in these tumors correlate with poor clinical outcome. The importance of SCCRO amplification in malignant transformation is established by the apoptotic response to short hairpin RNA against SCCRO, exclusively in cancer cell lines carrying SCCRO amplification. The oncogenic potential of SCCRO is underscored by its ability to transform fibroblasts (NIH-3T3 cells) in vitro and in vivo. We show that SCCRO regulates Gli1—a key regulator of the hedgehog (HH) pathway. Collectively, these data suggest that SCCRO is a novel component of the HH signaling pathway involved in the malignant transformation of squamous cell lineage. (Cancer Res 2006; 66(19): 9437-44)
Several studies suggest that amplified DNA is unstable. Accordingly, increased prevalence in human cancers implies specific selection for the amplified region (1–3). Amplification at 3q is seen in multiple tumor types but shows a higher predilection for squamous cell carcinoma (SCC) of mucosal origin, including those of the lung, head and neck, esophagus, vagina, vulva, and cervix, where it has been associated with tumor progression and an aggressive clinical course (4–11). The high prevalence, combined with biological implications and clinical associations, suggests that amplification at 3q may play a role in squamous cell carcinogenesis.
Identification of gene(s) driving selection for 3q amplification has been a significant research focus. Unlike deletions, translocations, and mutations, there is no pathognomonic attribute that confirms a gene as a target of amplification. Over 100 candidate genes are present in the minimal common region of 3q amplification (∼40 Mb), defined by comparative genomic hybridization. Given that several genes can be overexpressed in a single amplification locus, it is not surprising that multiple putative targets that drive selection for 3q amplification have been identified. Despite functional assessments of some candidate genes (including PIK3CA, PKCι, LAMP3, and eIF-5A2), the precise target(s) of 3q amplification remains ill defined (12–14).
Most studies suggest that the 3q25-27 region likely harbors the important candidate genes. Fine-resolution mapping show that subpeaks of amplification exist within the 3q25-27 region, and that the gene(s) driving selection for the amplified locus is present within these subpeaks. Lower-resolution positional cloning analyses (using yeast artificial chromosome clones) show overlapping results with an amplification subpeak centered around 3q26.3 (4, 10, 15). Fine-resolution mapping has been less homogenous in defining amplification subpeaks. Our group identified two amplification peaks, one in the region of ∼180 Mb of chromosome 3 and the second at ∼184 Mb in lung and head and neck cancers (9). Snijders et al. (16) also identified two amplification subpeaks at 3q in fallopian tube carcinomas (map positions ∼165-175 and ∼181-185 Mb). Interestingly, these authors excluded PIK3CA (180.4 Mb) as a candidate gene, as it was outside their amplification boundaries. Conversely, Or et al. (17) found the amplification apex around the region covered by bacterial artificial chromosome (BAC) clone RE11-510K16 at ∼181.6 Mb, suggesting that PIK3CA may be the target. Jiang et al. (18) did comparative genomic hybridization on cDNA arrays in nasopharyngeal carcinomas showing three amplified loci at 3q, including the region around TP73L (∼192 Mb) and SOX2 (∼182 Mb) and AP2M1 (185.3 Mb). In contrast, Snijders et al. (19) did not identify any recurrent amplifications at 3q26-27 in their analysis of primary oral cancers. Combined, the fine-resolution mapping suggests that multiple amplification subpeaks may exist, with frequent overlap in the 183 to 186 Mb position at 3q26.3. Here, we report the identification and functional validation of SCCRO/RP42/DCUN1D1 (184.1 Mb) as one potential target driving selection for 3q amplification in SCC of mucosal origin. Moreover, our findings suggest that the oncogenic activity of SCCRO results at least partly from activation of hedgehog (HH) signaling.
Materials and Methods
Genomic sequence analysis and prediction. Genomic annotation, computational analyses, and homology searches were done using internet accessible software (see Supplementary Methods).
cDNA cloning, expression, and transfection. The human SCCRO cDNA was generated by reverse transcription-PCR (RT-PCR) of total RNA isolated from head and neck cancer cell line MDA1386 and subcloned into pGEM-T (Promega, Madison, WI), pUSEamp (Upstate Biotechnology, Lake Placid, NY), pCMV-HA (BD Biosciences, San Jose, CA), and pGEX-4T-3 (GE Healthcare Life Sciences, Piscataway, NJ) vectors (see Supplementary Methods).
5′ Rapid amplification of cDNA ends. The sequences of the 5′ end of cDNA were derived using SMART RACE cDNA Amplification kit (BD Biosciences Clontech, Palo Alto, CA) according to protocols of the manufacturer and sequenced (see Supplementary Methods).
Cell lines, tumor tissue, and chemical reagents. The derivation and growth conditions for cell lines were published or are available through the American Type Culture Collection (Manassas, VA; ref. 10). Primary tumors from head and neck and lung cancers were collected from patients undergoing surgical resection, after obtaining informed consent and following institutional guidelines (see Supplementary Methods).
Cell growth and colony formation assays. 3-[4,5-Dimethylthiazol-2-yl]-2,5-diphenyl-tetrazolium bromide (MTT)–based colorimetric assay for cell growth (Sigma-Aldrich, St. Louis, MO) was done as described previously (9). Soft agar colony formation assay was done by plating ∼2,000 cells on 0.9% agarose-coated plates and incubated for 1 week in DMEM containing 250 μg/mL neomycin, essentially as described previously (see Supplementary Methods; ref. 20).
Flow cytometry and fluorescence-activated cell sorting analysis. Cells were collected by trypsinization and resuspended in PBS containing 2 mmol/L EDTA. The cells were then fixed in 70% ethanol, digested with RNase A (0.02 μg/μL), stained with 50 μg/mL ethidium bromide, and analyzed using FACScan and CellQuest software (Becton Dickinson, Franklin Lakes, NJ).
Apoptosis assessment. Apoptosis was quantified using Annexin/7-amino-actinomycin D (7AAD) staining according to protocol of the manufacturer (R&D Systems, Minneapolis, MN).
Quantified in vitro migration and invasion. The migratory potential and invasive capacity of cultured cells were assayed by a modified Boyden chamber method as described previously (21).
Athymic mouse tumorigenicity. pUSEamp-SCCRO-3T3 or pUSEamp-3T3 cells were injected s.c. into each flank of BALB/C athymic nude mice. Autopsy studies were done on all animals at the end of the experiment following institutional guidelines (see Supplementary Methods).
Fluorescence in situ hybridization. Dual-color fluorescence in situ hybridization (FISH) was done essentially as described earlier using DNA from BAC clone 202B22 and chromosome 3 centromeric probes (10).
Southern and Northern blot analyses. Southern and Northern blot analyses were done according to standard protocols using probes corresponding to the entire coding region of SCCRO, generated by PCR (see Supplementary Methods; ref. 9).
Western blot analysis. Western blots were done using rabbit polyclonal anti-SCCRO antibody at a concentration of 1:5,000 (see Supplementary Methods). Antibody against Gli1 was obtained from Chemicon International (Temecula, CA) and used at a concentration of 1:5,000.
Quantification of mRNA levels using real-time RT-PCR analysis. Real-time PCR was done as described in Supplementary Methods. Melt curve analysis was done following amplification (22). The acquisition temperature was set 1°C to 2°C below the Tm of the specific PCR product. The relative quantification of a target gene compared with a reference gene (18S rRNA) was done as described (23, 24). PCR primers and conditions are detailed in Supplementary Table S2.
Immunohistochemistry. Immunohistochemistry was done using polyclonal antibody against SCCRO raised in rabbit (antibody 1B) and goat human Gli1 antibody (Santa Cruz Biotechnology, Santa Cruz, CA). Details for immunohistochemistry are provided in Supplementary Methods.
Protein knockdown. pSUPER RNA interference (RNAi) system (Oligo Engine Inc., Seattle, WA) was used to stably direct endogenous synthesis of short hairpin RNA (shRNA) against SCCRO and Gli1. We designed two RNAi sequences using OligoEngine software.79).
Statistical analysis. All analyses were done using JMP4 statistical software (SAS Institute, Inc., Cary, NC; see Supplementary Methods).
Molecular cloning and analysis of SCCRO. We applied a positional cloning approach to systematically refine the minimal common region of 3q amplification. Analysis of five cancer cell lines with 3q amplification identified two recurrent amplification subpeaks contained within BAC clones, 202B22 and 386M7 (4, 5, 9). PIK3CA, previously identified as amplified in ovarian and cervix cancers, was an obvious candidate within one of these peaks (386M7; ref. 25). Analysis of the genomic insert in BAC clone 202B22 representing the second subpeak showed no known genes. Annotation using the GENSCAN and Genie prediction programs identified several potential genes. RT-PCR done on total RNA from two SCC cell lines (MDA886 and MDA1186) and normal human placenta resulted in a cDNA product for only one of these genes, which upon sequencing revealed a 780 bp open reading frame (ORF; Supplementary Fig. S1A and B). This gene was designated as SCCRO.
Sequence analysis of SCCRO revealed a single ATG at the 5′ end in the context of a sequence that can support translation initiation and a polyadenylation signal at the 3′ end (Supplementary Fig. S1A; ref. 26). In vitro translation of this cDNA yielded a protein of predicted size (∼30 kDa; data not shown). A multi-tissue Northern blot was probed with full-length ORF for SCCRO. Transcripts of different sizes were detected (∼2.0, 4.1, and 4.3 kb). The expression level and relative ratios of these transcripts varied from tissue to tissue (Supplementary Fig. S1C). Gene expression was reconfirmed by real-time RT-PCR (data not shown) in adult RNA with an expression pattern similar to that detected by Northern blot analysis. These observations suggest that the expression and the relative levels of SCCRO transcripts are regulated in a tissue-specific manner. A search of the genome database shows significant evolutionary conservation, suggesting SCCRO has important biological functions (Supplementary Fig. S2).8
R. Ryan et al., unpublished data.
SCCRO is amplified and highly expressed in human tumor. In agreement with the reported frequency of 3q amplification, SCCRO mRNA overexpression was detected in 21 of 44 (48%) primary lung, 16 of 45 (36%) head and neck, and 4 of 9 (44%) cervical carcinomas relative to matched histologically normal lung, oral mucosa, or cervical mucosal controls, respectively (levels >2 SD; refs. 4–8, 28). Moreover, a significant correlation was observed between SCCRO copy number, the corresponding mRNA, and protein levels in cell lines and primary tumors derived from lung and head and neck cancer (Fig. 1A and B; r = 0.81; P < 0.001; see Supplementary Figs. S3 and S4 for anti-SCCRO antibody validation). In addition, non–small cell lung cancers showed SCCRO overexpression in SCC (61%), but not in adenocarcinomas (9%; P = 0.004), reflecting the known predilection of 3q amplification for SCC (Fig. 1C; ref. 29). Survival analysis was done in a larger cohort of previously untreated non–small cell lung cancer cases that underwent primary surgical treatment (n = 79). Although there were no differences in tumor-node-metastasis (TNM) stage, histology, treatment, or follow-up (median 44 months) based on SCCRO expression status, SCCRO expression negatively correlated with cause-specific survival (P = 0.05), indicating that SCCRO overexpression imparts an aggressive phenotype in these tumors (Fig. 1D). The effects of SCCRO overexpression on cause-specific survival remained significant even after controlling for the effects of confounding TNM stage by multivariate analysis (relative risk, 4.38; P = 0.04). These findings are similar to our prior analysis of head and neck SCC (11). To exclude the possibility of an oncogenic mutation in SCCRO, we did RT-PCR and sequenced the coding region of SCCRO from 120 cases of SCC of lung or head and neck origin. No mutations were identified. Combined, these observations suggest that amplification of SCCRO results in its overexpression, which is associated with aggressive behavior in mucosal SCC.
SCCRO is oncogenic. To further validate SCCRO as an amplification target in cancer, we tested its oncogenic potential in transformation assays. We used NIH-3T3 cells for the initial experiments, as this cell system is well established for assessment of transforming activity. We stably transfected NIH-3T3 cells with a plasmid expressing SCCRO under the control of a cytomegalovirus (CMV) promoter. Two independent clones (clones 14 and 28) were specifically selected, as they expressed SCCRO protein at levels comparable with human cancers cell lines carrying SCCRO amplification (Fig. 2A). SCCRO-transfected NIH-3T3 cells acquired a dedifferentiated morphology, with increased nuclear size and higher nuclear-to-cytoplasmic ratio. Moreover, focus formation was seen in both SCCRO-transfected clones, but not in the one transfected with vector (data not shown). Cell growth, as determined by MTT cell viability assay, was significantly higher in SCCRO-transformed cells (Fig. 2B). In addition, SCCRO-transformed cells did not senesce in absence of growth signals (serum deficient conditions) in contrast to vector-transfected cells (Supplementary Fig. S5A). SCCRO-transfected cells showed an increase in in vitro invasive potential (53 ± 12.3% and 69.5 ± 8.7% invasive fraction for clones 14 and 28, respectively), compared with the vector-transfected 3T3 cells (18.5 ± 6.7% invasive fraction) as determined by the modified Boyden chamber invasion assay (P < 0.001). In contrast to vector-transfected NIH-3T3 cells, both SCCRO-transformed clones also displayed anchorage-independent growth, as tested on soft agar colony formation assay, indicating a transformed phenotype (Fig. 2C). Finally, in vivo xenograft assay in NIH BALB/c nude mice showed that both SCCRO-transfected clones were oncogenic, resulting in tumor formation in six of six mice within 8 weeks. In contrast, no tumors developed in six mice injected with vector-transfected 3T3 cells even after 12 weeks (Fig. 2D). Autopsy and histopathologic analyses revealed the presence of poorly differentiated tumors (Supplementary Fig. S5B) with metastases to pelvic lymph nodes identified in three of six tumors.
Because SCCRO is purported to play a role in keratinocyte transformation, we did transformation experiments in cells of epithelial derivation (HaCaT). For indeterminate reasons, stable colonies could not be established with adequate SCCRO expression levels. Accordingly, we assessed colony-forming ability in HaCaT cells transiently transfected with SCCRO. SCCRO-transfected HaCaT cells formed 352 ± 97 colonies per 12-well plate on the soft agar forced suspension culture assay, compared with ≥50 colonies in wild-type and empty vector–transfected cells (P < 0.01; see Supplementary Fig. S6 for basal expression). These results show that SCCRO induces transformation in two distinct mammalian cell lines.
SCCRO is required for survival of cancer cell lines carrying amplification. To more firmly establish a role for SCCRO amplification/overexpression in the pathogenesis of SCC, we assessed the requirement of SCCRO expression for tumor maintenance. Transfection of shRNA against SCCRO into cancer cell lines resulted in a 3- to 5-fold decrease in SCCRO protein levels relative to empty vector and scrambled shRNA controls (Fig. 3A). Significant cell death was observed following SCCRO knockdown in representative cancer cell lines carrying SCCRO amplification/overexpression, but not in those without it (Fig. 3B). Fluorescence-activated cell sorting (FACS) analysis showed a significant increase in the sub-G1 fraction, and Annexin/7AAD staining suggested that reduction of SCCRO protein levels by shRNA transfection resulted in death by apoptosis in MDA1386 cells (Fig. 3C). Conversely, no apoptosis was seen in MDA1386 cells transfected with the control shRNA (Fig. 3C) or in cell lines with normal SCCRO copy number (data not shown). Soft agar colony formation assay showed a significant reduction in the number of colonies in shRNA-transfected cells compared with the control (Fig. 3D). Because the specificity of RNAi-based knockdown has been questioned, even when unique sequences are targeted, we repeated the cell viability experiments after knocking down SCCRO by transfection of antisense in the same cell lines. This produced identical results in vitro and in vivo in xenograft models,9
I. Ganly et al., unpublished data.
SCCRO is involved in HH signaling. As a step toward understanding the functional role of SCCRO, we did gene expression analysis using Affymetrix oligonucleotide microarrays on NIH-3T3 cells stably transfected with SCCRO (clones 14 and 28). After normalizing for expression changes in vector-transfected NIH-3T3 cells, the expression of several genes was altered in both SCCRO-transfected clones, including many linked to growth and malignant transformation (Supplementary Table S1). Real-time RT-PCR of eight representative genes was done to internally validate the results from the gene array analysis (Supplementary Tables S1 and S2).
Of the genes identified by array analysis, up-regulation of Gli1 was of particular interest, as it plays an essential role in the development of lungs, trachea, and esophagus—the same organs where SCCRO is purported to have oncogenic activity (30). The expressions of multiple genes known to be downstream of Gli1 was also altered, including Gli2, cyclin D1, FGFR, IGF, IGFR, and ceruloplasmin (31–33). To validate the findings of the gene expression analysis, we did Western blot (Fig. 4A; Supplementary Fig. S6), real-time RT-PCR (Fig. 4B), and immunohistochemical analyses (Fig. 4C) in SCC cell lines and several primary tumors. Results suggest that there is a positive correlation between SCCRO and Gli1 levels in the analyzed cases (Fig. 4A-C; r = 0.75; P < 0.001). This association was further tested in an independent MEF/3T3 tetracycline-repressible SCCRO expression system (Tet-off). As expected, removal of tetracycline from the medium resulted in an increase in SCCRO expression and a resultant increase in Gli1 levels as detected by Western blot analysis and real-time PCR analyses (5.0-fold; P = 0.03; Fig. 4D). Conversely, no change in SCCRO or Gli1 expression was seen in the presence of tetracycline in the medium. In addition, mRNA levels of Gli2 (3.5-fold; P = 0.05) and PTCH (5.4-fold; P = 0.01), genes known to be activated by the HH pathway, were also up-regulated. In contrast, no significant expression changes were detected in Gli3 (P > 0.05) or SMO (P > 0.05), both of which are not targets of the HH pathway. Finally, SCCRO expression in the developing mouse embryo was similar to Gli1 (Supplementary Fig. S7). Taken together, these results strongly suggest that SCCRO is a positive regulator of the HH signaling pathway.
SCCRO binds to and activates the Gli1 promoter. To determine if Gli1 is a transcriptional target of SCCRO, we did a reporter assay using a construct containing luciferase gene under the control of Gli1 promoter. Transfection of the Gli1-luciferase reporter construct into cells overexpressing SCCRO (either stably or transiently) resulted in a significant increase in luciferase expression relative to control (Fig. 5A). Moreover, in transient cotransfection experiments, the level of luciferase expression correlated with the amount of SCCRO expression plasmid transfected (Fig. 5A).
To further elucidate the mechanism of SCCRO-mediated effects on Gli1 transcription, we did modified McKay and chromatin immunoprecipitation (ChIP) assays to assess SCCRO-DNA interactions. Modified McKay experiment showed SCCRO to interact with a fragment that spans the region +122 to +496 of Gli1 promoter (Fig. 5B and C). SCCRO-DNA interaction was further confirmed in vivo by ChIP assay. Three primer sets, spanning overlapping segments of the Gli1 promoter (−962 and +496), were used to detect the in vivo binding region (Supplementary Table S2). ChIP assay validated the results from the modified McKay assay, showing that SCCRO binds to a region spanning +122 to +496 of Gli1 promoter (Fig. 5D). Taken together, it seems that Gli1 is a direct transcriptional target of SCCRO.
Gli1 contributes to SCCRO-induced transformation. To determine the role of Gli1 in SCCRO-driven oncogenesis, we assessed the effects of Gli1 knockdown on the viability of SCCRO-T1 cells (derived from pUSEamp-SCCRO-3T3 after passage through a mouse). Specific silencing was achieved with transfection of pSUPER-Gli1-1502, whereas there was no change in Gli1 protein levels with transfection of either scrambled RNAi (pSUPER-control) or empty vector (pSUPER) in SCCRO-T1 cells (Fig. 6A). Silencing of Gli1 in SCCRO-T1 cells resulted in a 48% mean reduction in colony formation relative to pSUPER-control–transfected cells (Fig. 6B; P = 0.01) These findings suggest that Gli1 partially mediates SCCRO-induced transformation, but does not exclude involvement in other pathways.
SCC of mucosal origin include tumors arising from the head and neck, lung, esophagus, and cervix, and account for over one third of all cancers and cancer-related deaths annually in the United States (34). Although these tumors are anatomically diverse, they share a causal association with tobacco exposure and a common genetic composition (35). Amplification of 3q26-27 locus is a common and crucial event in squamous cell carcinogenesis, associated with both tumor progression and an aggressive clinical course (4, 5, 9–11, 25, 36–39). In this study, we identified a novel gene (SCCRO) within a recurrent amplification subpeak at 3q26.3 in SCC. Interestingly, at least three other genes, namely LAMP3, AP2M1, and eIF-4γ, are in the same relative genomic location as SCCRO, supporting the presence of an amplification subpeak in the region (14, 18, 40). To validate SCCRO as an amplification target, we first correlated genomic amplification with mRNA and protein expression in primary tumors and cell lines derived from head and neck and lung carcinomas. Because overexpression of an amplified gene in itself is insufficient to support its role in cancer pathogenesis, we did a series of experiments to establish SCCRO as an amplification target. We found that SCCRO overexpression was associated with an aggressive clinical course in primary SCC of the lung. We showed that SCCRO transforms cells of both fibroblastic and keratinocytic lineage, suggesting that it functions as an oncogene. Finally, we showed that silencing using shRNA against SCCRO results in apoptosis only in cancer cell lines carrying SCCRO amplification, suggesting that SCCRO not only induces transformation but is also required for maintenance of the malignant phenotype. This phenomenon, termed “oncogene addiction,” suggests that the gene is directly involved in the maintenance of the malignant state, strongly suggesting it is an oncogenic signal rather than noise in malignant progression (41). Although it remains possible that other important cancer gene(s) may be present, our data strongly suggest that SCCRO is a candidate oncogene that drives selection for 3q amplification.
To begin to ascertain its function, we screened the effects of SCCRO transfection on gene expression in NIH-3T3 cells and identified multiple targets. Given the magnitude of up-regulation of Gli1 and several of the Gli1 transcriptional targets in SCCRO-transformed cells, we focused on assessing the role of SCCRO in HH signaling. The HH signaling pathway is known to play an essential role in both embryonic development and cancer pathogenesis (42, 43). Gli1 is also an important primary target in carcinogenesis, as evidenced by its activation by amplification in gliomas (44). It is important to note that prior studies have shown a prominent absence of aberrant HH signaling in SCC (45). However, all of these studies have focused on tumors of cutaneous origin, which are etiologically and pathogenetically distinct from SCC of mucosal origin. Interestingly, recent studies have suggested that the HH signal is activated in head and neck cancer (19). This activation is not induced by receptor-ligand activity, as is present in small cell lung cancers, but rather involves other mechanisms. In a small percentage of cases, the activation is related to amplification of Gli2 (19). The association between SCCRO and Gli1 expression and activation of the HH pathway in both constitutive (stable) and regulated (tetracycline repressible) SCCRO expression systems suggests that SCCRO amplification may play a role in activating HH signaling in SCC. Moreover, SCCRO and Gli1 mRNA and protein are coexpressed in primary tumors derived from head and neck and lung carcinomas. We also found that the expression of known Gli1 targets, including Gli2 and PTCH, is increased in response to SCCRO overexpression, whereas the expression of HH pathway components known not to be effected by Gli1 remains unchanged. Analysis of mouse embryo containing lacZ cassette inserted into SCCRO gene (see Supplementary Fig. S8)10
A. Kaufman et al., in preparation.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
I. Sarkaria, P. O-charoenrat, and S.G. Talbot contributed equally to this work.
This work is dedicated to the memory of Dr. P.G. Reddy, who passed away unexpectedly during the preparation of this article.
Grant support: This work was supported in part by grants from George H.A. Clowes, Jr., MD, FACS, Memorial Research Career Development Award from the American College of Surgeons; Falcone Fund; Clinical Innovator Award, Flight Attendant Medical Research Institute (to B. Singh).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
We thank Dr. Lorenz Studer for assistance with the immunofluorescence analysis, Dr. Agnes Viale for assistance with gene expression profiling, Dr. Nicholas Socci for help with analysis of gene expression data, Dr. Susan Naylor (Department of Cellular and Structural Biology, University of Texas Health Science Center, San Antonio, TX) for providing the BACs used in these experiments, Dr. Shunsuke Ishii (Laboratory of Molecular Genetics, RIKEN Tsukuba Institute, Ibaraki, Japan) for providing the Gli1 promoter constructs, Drs. Chris Sander and Boris Reva for bioinformatics analyses, Swarna Gogneni and Michael Wyler for technical assistance, Drs. Joan Massague and Andrew Simpson for the critical review of the manuscript, and Nancy Bennett for excellent editorial assistance.