Changes in DNA methylation patterns are a common characteristic of cancer cells. Recent studies suggest that DNA methylation affects not only discrete genes, but it can also affect large chromosomal regions, potentially leading to LRES. It is unclear whether such long-range epigenetic events are relatively rare or frequent occurrences in cancer. Here, we use a high-resolution promoter tiling array approach to analyze DNA methylation in breast cancer specimens and normal breast tissue to address this question. We identified 3,506 cancer-specific differentially methylated regions (DMR) in human breast cancer with 2,033 being hypermethylation events and 1,473 hypomethylation events. Most of these DMRs are recurrent in breast cancer; 90% of the identified DMRs occurred in at least 33% of the samples. Interestingly, we found a nonrandom spatial distribution of aberrantly methylated regions across the genome that showed a tendency to concentrate in relatively small genomic regions. Such agglomerates of hypermethylated and hypomethylated DMRs spanned up to several hundred kilobases and were frequently found at gene family clusters. The hypermethylation events usually occurred in the proximity of the transcription start site in CpG island promoters, whereas hypomethylation events were frequently found in regions of segmental duplication. One example of a newly discovered agglomerate of hypermethylated DMRs associated with gene silencing in breast cancer that we examined in greater detail involved the protocadherin gene family clusters on chromosome 5 (PCDHA, PCDHB, and PCDHG). Taken together, our results suggest that agglomerative epigenetic aberrations are frequent events in human breast cancer. [Cancer Res 2008;68(20):8616–25]
- DNA methylation
- breast cancer
- CpG island
Aberrant levels and patterns of cytosine methylation are ubiquitous events in human cancer ( 1). DNA hypomethylation generally occurs often in repetitive elements, pericentromeric regions, and within the body of genes ( 2– 4). In contrast, DNA hypermethylation is largely limited to CpG island regions ( 5– 7). Some gene targets of DNA hypermethylation are found in a broad spectrum of different tumor types, whereas others occur in a tumor type–specific fashion ( 8). Studies into the biological consequences of hypomethylation and hypermethylation suggest that each plays a participative role in carcinogenesis. Earlier studies suggested that hypermethylation events were focal in nature, usually associated with the transcriptional silencing of one gene. Recent studies show that DNA hypermethylation can extend over regions of the genome from 100 to >1,000 kb in size and is linked to long-range epigenetic silencing (LRES; refs. 9– 11).
These LRES events have practical parallels to genetic lesions, such as deletions and microdeletions, in that in both cases long contiguous stretches of DNA have become functionally inactivated. With respect to the epigenetic events, the long-range silencing is associated with an epigenetic terrain consisting of landmarks linked to a repressed chromatin state, such as aberrant DNA methylation and/or histone modification state ( 9– 12). Considering the gross genetic and epigenetic changes that occur in a cancer cell, it is likely that diverse inappropriate combinations of epigenetic landmarks will partner together to produce aberrant gene silencing. This LRES is metastable and reversible with drugs that target epigenetic control mechanisms ( 9, 10).
We set out to better determine the scope and the spatial organization of aberrant DNA methylation and its relationship to LRES in human breast cancer. Using epigenomic scanning approaches, we analyzed a set of normal and cancerous breast specimens, a limited life span human mammary epithelial cell (HMEC) strain, and breast cancer cell lines to identify differentially methylated regions (DMR) between normal and malignant breast tissue. Locations of DMRs revealed that aberrant DNA methylation is not randomly distributed across the cancer genome, and agglomerates of DMRs can be identified, consisting of agglomerates of both hypermethylated and hypomethylated DMRs that ranged from ∼30 kbp to almost 600 kbp in size. These aberrant agglomerative methylation events can occur multiple times within an individual tumor and are associated with LRES events. The types of promoters and the regions affected by DNA hypomethylation were distinct from those targeted by hypermethylation; however, a feature common to both types was their apparent concentration in gene family clusters.
Materials and Methods
Cell culture. All cell lines were maintained as previously described ( 9). The MDA-MB-231 and BT549 breast cancer cells were obtained from the American Type Culture Collection. Cell line identity was assured by genotypic analysis by DNA fingerprinting using single tandem repeats (STRs). Normal HMECs were obtained from Clonetics and were grown according to supplier's instructions.
Breast tumor specimens. Flash-frozen specimens derived from normal or cancerous breast tissue were obtained from patients who underwent surgery for breast cancer, either lumpectomy or mastectomy, at the University Medical Center in Tucson, AZ, from 2003 to 2005. All patients signed surgical and clinical research consents for tissue collection in accordance with the University of Arizona Institutional Review Board and Health Insurance Portability and Accountability Act of 1996 regulations. At the time of surgery, a 1- to 3-cm section of the tumor was immediately snap frozen in liquid nitrogen and stored in our prospective breast tissue bank at −80°C. From each tissue block, a series of 5-μm sections were cut and stained with H&E for pathologic evaluation. All H&E slides were reviewed by two independent pathologists to determine the integrity of the tumor specimen. A partial molecular characterization of these samples have been reported on previously ( 9). Supplementary Table S1 provides the pathologic assessment of each specimen.
Nucleic acid isolation. RNA and DNA were isolated as previously described ( 9).
Bisulfite sequencing and DNA sequence analysis. Genomic DNA was analyzed by bisulfite sequencing as described ( 13). The protocadherin CpG islands were amplified from the bisulfite-modified DNA by two rounds of PCR using nested primers. Primer sequences are available upon request.
DNA methylation analysis by MassARRAY. Sodium bisulfite (NaBS)-treated genomic DNA was prepared according to manufacturer's instructions (Zymo Research). NaBS-treated DNA (5ng) was seeded into a region specific PCR reaction incorporating a T7 RNA polymerase sequence as described by the manufacturer (Sequenom). Resultant PCR product was then subjected to in vitro transcription and RNase A cleavage using the MassCLEAVE T-only kit, spotted onto a Spectro CHIP array, and analyzed using the MassARRAY Compact System MALDI-TOF mass spectrometer (Sequenom). Each NaBS-treated DNA sample was processed in two independent experiments. Data were analyzed using EpiTyper software (Sequenom) and the R-script “Analyze Sequenom Function” using methods described previously ( 14). Primer sequences were designed using EpiDesigner. 5 Primer sequences are available upon request.
Real-time reverse transcription-PCR. Total RNA was converted to cDNA and then amplified in a real-time PCR format using gene-specific primers and fluorescent probes using Roche UniversalProbe technology and the ABI 7500 Real-time Detection System (Applied Biosystems). Results were calculated using the Delta Ct method normalizing to glyceraldehyde-3-phosphate dehydrogenase expression. Primer sequences are available upon request.
Methyl DNA immunoprecipitation microarrays and data analysis. Methylated fraction of DNA was obtained by immunoprecipitation as described ( 15). Two hundred nanograms of enriched fraction and corresponding input DNA were hybridized on Affymetrix GeneChip Human Promoter 1.0R Array according to manufacturer's protocol.
Tiling Analysis Software (Affymetrix) was used to extract signal intensities and map them to corresponding genomic position based on National Center for Biotechnology Information human genome assembly (Build 34). Subsequent data processing and data analysis was done using R programming environment ( 16). Signal intensities were loess normalized, then input intensities were subtracted from corresponding methyl DNA immunoprecipitation (MeDIP) intensities. Resulting MeDIP/Input ratios were again loess normalized. For each probe, a quality weight was calculated as previously described ( 17). To find DMRs in normal and cancerous samples, t statistics was calculated for each probe. To combine information from neighboring probes, a weighted moving average (window, 1,000 bp) was computed as final summary statistics for each probe. The same procedure was done with probe positions randomly permutated to find the appropriate cutoff for false discovery rate of 5%. The positions of all positive regions can be found in Supplementary Data as an annotation track that can be visualized in University of California at Santa Cruz (UCSC) Genome Browser. Multiple occurrences of DMR sequences were analyzed using BLAT alignment tool ( 18). The raw data from these microarray studies have been deposited in the Array Express database, accession E-MEXP-1481. 6
The expected (theoretical) distances of randomly distributed DMRs were calculated by taking the number of DMRs detected in the sample set and randomly distributing this number through the regions covered by the promoter array. Distances between DMRs from these simulations were then calculated; the average results of 50 simulations are shown. The random distribution simulation was performed separately for hypermethylated and hypomethylated DMRs.
Chromatin immunoprecipitations. Chromatin immunoprecipitations were performed as previously described ( 19). The antibodies directed toward histone H3 lysine-9 dimethylation (DiMeK9) and H3 lysine-27 trimethylation (TriMeK27) were purchased from Upstate/Millipore. Primer sequences are available upon request.
In this study, we sought to examine the nature and spatial organization of aberrant DNA methylation in invasive human breast cancers and corresponding cell line models. The DNA methylation-state of 16 invasive breast tumors, 5 histologically normal breast specimens, and 4 pure cell line/strain populations derived from cancerous and normal mammary tissue was analyzed using the Affymetrix Human gene promoter array 1.0R, a more comprehensive approach that provided greater and more focused genomic coverage than our previous studies ( 9, 20). Labeled targets from each of the samples for microarray hybridizations were generated from the 5-methylcytosine–specific antibody enriched fraction of the DNA, as well as the total the genomic DNA (Input). Input DNA was used as a reference control ensuring that each sample analyzed for DNA methylation was compared with its own unique genome. This reference approach served to minimize false discovery due to other genomic aberrations, such as gene amplification.
We identified 3,506 cancer-specific DMRs, which were either aberrantly hypermethylated (2,033) or hypomethylated (1,473) in cancerous tissue. A DMR was defined as any region covered by a group of consecutive probes on the tiling array that exhibited statistically significant differences in normalized signal (i.e., MeDIP/Input ratio) between the group of normal and the group of cancerous specimens. A vast majority (99%) of these DMRs covered 8 to 115 tiling array probes and therefore ranged in size from 280 to 4,000 bp; 1,000 bp was the most frequent DMR size. The methylation status of each of these DMRs was analyzed in each individual cancer specimen relative to the normal specimens, which enabled us to determine the number of aberrantly methylated DMRs in each individual tumor as well as the frequency of aberrant methylation for each DMR within the sample population as whole. This analysis revealed that each cancerous specimen exhibits hypomethylation and hypermethylation in ∼25% to 75% of the 3,506 DMRs identified ( Fig. 1A ). Finally, most of the DMRs that we detected are recurrent in the breast cancer. Approximately 90% of these DMRs are aberrantly methylated in at least 33% of our sample set ( Fig. 1B). Overall, these results indicate that there are thousands of recurrent DMRs in a substantial fraction of human breast cancers.
To identify potential long-range events, we analyzed the spatial distribution of the DMRs across all 23 chromosomes ( Fig. 1C). Positive peaks indicate hypermethylated regions, whereas negative peaks indicate hypomethylated regions. The amplitude of each peak is proportional to the number of DMRs found within a 100-kb window, and therefore, the largest peaks in Fig. 1C reveal agglomerates of DMRs—regions with the highest density of aberrant DNA methylation. The agglomerates containing hypermethylated DMRs then represent regions with the potential for DNA methylation-associated LRES.
Examination of the DNA methylation chromosomal plot in Fig. 1C shows that we could detect multiple regions of agglomerative DNA methylation. One such region is the HOXA gene family cluster, which we previously identified in breast cancer ( 9), thereby providing empirical support for the analytic approach used to detect aberrant DNA methylation related to LRES. One novel region of agglomerative DNA hypermethylation discovered is located on chromosome 5 and is associated with the three protocadherin gene clusters (PCDHA, PCDHB, and PCDHG). A heat map of the DNA methylation state of all DMRs ordered by physical position along chromosome 5 is shown in Fig. 1D. The patient samples were ordered by hierarchical clustering and the heat map allows visualization of the hypermethylated DMRs associated with the PCDH gene clusters. We also detected agglomerative hypomethylation events specific to cancer; one of which was localized to the GAGE gene family cluster on the X chromosome. The GAGE genes are tumor-specific antigens not expressed in adult somatic tissues but undergo a transcriptional activation during carcinogenesis that has been associated with DNA hypomethylation ( 21, 22).
We evaluated whether the agglomeration of DMRs in apparent hotspots was simply due to chance or if DMRs clustered in a cancer-driven fashion. One way in which this was assessed was through an analysis of the distributions of genomic distances between neighboring hypermethylated and hypomethylated DMRs. Figure 2A shows the frequency of distances between DMRs in the data obtained (histogram) compared with the frequency of theoretical distances based on a simulated random spatial distribution of DMRs (overlaid solid line). Overall, the results were quite similar for both hypermethylated and hypomethylated DMRs. In both cases, the observed and expected (theoretical) distribution shows the majority of DMRs occur ∼106 bp apart; however, when the observed data are further compared with the expected data, there is an increased frequency of hypermethylated as well as hypomethylated DMRs that have a spacing of <104 bp. Thus, the divergence of the observed DMR distributions from the expected DMR distributions suggests that aberrant DNA methylation is nonrandomly distributed across the breast cancer genome.
We explored the distribution of DMRs further and found that the location of individual DMRs are positively correlated with methylation changes of proximate DMR sites (<100 kb apart). This effect is practically null for distant sites (>1 Mb; Fig. 2B). In other words, there is a higher probability that proximate DMR sites will exhibit the same aberrant methylation status in a particular sample. Examples of such correlated regions can be seen on Fig. 1D around positions 20 and 140 Mbp. This phenomenon suggests that there is a common factor affecting methylation status of such aberrant methylation prone regions, and provides further support for the idea that cancer-specific aberrant methylation is distributed across the genome in a nonrandom fashion.
Further examination of the genomic characteristics of the promoter-associated DMRs revealed other features that distinguished hypermethylation from hypomethylation events in the invasive breast cancers. Results in Fig. 3A show that the aberrant hypermethylation events in breast cancer were linked to CpG-rich regions, with 38% of hypermethylated DMRs coincident with CpG islands, whereas only 6% of hypomethylation DMRs were linked to CpG islands. The locations of these DMRs within the 10-kb promoter regions assayed on the gene promoter array were also determined ( Fig. 3B). DMRs associated with aberrant hypermethylation were concentrated in the region near transcription start, whereas hypomethylated DMRs were more evenly distributed throughout the promoter region.
The types of promoter regions targeted by DNA hypermethylation and hypomethylation differed further. Cancer-specific DNA hypomethylation events occurred at a higher frequency in regions of segmental duplications, whereas DNA hypermethylation events were heavily biased toward single copy sequences ( Table 1 ). Taken together, these results suggest that both long-range DNA hypermethylation and hypomethylation events occur in human breast cancer, and that the occurrence of these events show regional selectivity both within promoter regions and throughout the genome.
Table 2 shows the chromosomal regions associated with the highest density of DMRs within the breast cancer samples when compared with normal tissue. The HOXA genes are present and representative of hypermethylation events, whereas GAGE genes are present and representative of hypomethylation events. Table 2 also shows a range of 22% to 89% for the frequency of occurrence of the 15 agglomerative DNA methylation events detected in the breast cancer samples. In general, these results show that DNA agglomerative events occur at frequencies that are comparable with and exceed the frequency of common genetic mutations in breast cancer, such as p53 ( 23). Interestingly, there was a prevalence of gene family clusters in the list of agglomerates of DMRs including the protocadherin gene family clusters (PCDHA, PCDHB, and PCDHG), the HOXA gene cluster, the HOXD gene cluster, and a histone gene cluster.
The PCDH gene family clusters showed a strong hypermethylation signal in the DNA methylation chromosomal plot ( Fig. 1C). Considering their unique genomic organization and the potential for these clusters to be a novel target for epigenetic silencing in human breast cancer, we analyzed this region in greater detail. This family of PCDH genes (α,β, and γ) resides in 3 clusters on chromosome 5, and the genes are expressed in normal human breast tissue and down-regulated in breast cancer ( Fig. 4B ). The α and γ gene clusters share similar characteristics. The genes within each cluster share the same last three 3′ exons and differ only in their unique 5′ exons. In contrast, each gene in the β cluster is distinct, ordered along the chromosome, and consists of only one exon ( 24). Each PCDH gene in these three gene family clusters also has a CpG island promoter.
To confirm and extend the PCDH differential methylation results detected on the gene promoter microarrays, the methylation states of nine 5′ CpG island promoters (three each from the PCDHA, PCDHB, and PCDHG clusters) were analyzed by bisulfite sequencing in a selected set of samples. The genomic organization of the CpG island regions analyzed as well as the results of the bisulfite sequencing are shown in Fig. 4A. The bisulfite sequencing analysis of normal mammary tissue and the nontumorigenic, mortal cell strain HMEC revealed surprisingly high and widespread methylation of the PCDH CpG island promoters. The overall levels of methylation between the normal tissue and the HMEC cell strain were comparable and, based on the reverse transcription-PCR (RT-PCR) data, not at a level sufficient to repress transcription. The results further show that DNA hypermethylation of PCDH CpG island promoters occurs in human breast cancer. Importantly, the breast cancer sample 2845T, which shows LRES of the PCDH gene cluster, also contains an LRES event at the HOXA cluster ( 9). Thus, this tumor is a clear example showing the occurrence of multiple LRES events within a single cancer genome. Overall, the results from this high-resolution analysis confirm that aberrant DNA hypermethylation of multiple PCDH CpG islands is a common event in human breast cancer.
In addition to these agglomerative DNA methylation events, analysis of the microarray results also allowed for the detection of focal events previously determined to be aberrantly methylated in human breast cancer specimens. For example, similar to earlier reports we detected aberrant DNA methylation in the promoters of p16, WT-1, RASSF1A, DSC3, SIM1, PCDH10, and FOXA2 ( 25– 29). Furthermore, it is worthwhile to note that some focal events previously associated with breast tumorigenesis are contained within agglomerative DNA methylation events such as HOXA5, HOXA9, HOXD11, and PCDHGB6 ( 12, 28, 30). Taken together, these results indicate that the human promoter microarrays are able to detect common focal methylation changes as well as agglomerates of aberrantly methylated DMRs in human breast cancer specimens even using relatively small sample populations.
The prevalence of hypomethylated DMRs in the breast cancer samples detected by differential methylation microarray led us to validate these results. We analyzed 4 hypomethylated DMRs that are part of agglomerative epigenetic aberration located within a segmental duplication on chromosome 8. This region was also chosen because it was recently found to be hypomethylated in lung cancer ( 31). Quantitative analysis of DNA methylation patterns was determined using the MassArray pipeline of bisulfite PCR, in vitro transcription, base-specific cleavage of the RNA products, and mass spectrometry ( 32). Results from the analysis of three noncancer and three cancer samples are presented in Supplementary Fig. S2 and confirm the microarray data. Overall, these results show that both agglomerative DNA hypomethylation and hypermethylation events occur in breast cancer and represent a facet of long range epigenetic aberrations linked to human carcinogenesis.
Using the Affymetrix Human gene promoter array 1.0R, we took a genome-wide approach to profile the DNA methylation state of gene promoter regions in normal and cancerous human breast specimens and cell line models. We extended earlier observations that showed LRES in cancer and now find multiple agglomerative DNA hypermethylation events and, in the case of the PCDH gene family cluster, we confirmed that this event led to LRES. We also found agglomerative events characterized by DNA hypomethylation. The regions of hypermethylation and hypomethylation were distinct from one another. DNA hypermethylation events concentrated in CpG islands near transcription start sites, whereas hypomethylation events were found typically in CpG-poor regions and were not concentrated near transcription start, but rather the hypomethylated DMRs were more evenly spread across the 10-kb promoter regions analyzed by the array. Interestingly, the hypomethylation events also were disproportionately found in regions of segmental duplication, and it seems that a region we identified on chromosome 8 as hypomethylated in breast cancer also maps to a region of hypomethylation recently described in lung cancer ( 31).
A feature common to both the long-range DNA hypermethylation and hypomethylation events was their frequent occurrence in gene family clusters. The functional significance of the agglomerative DNA hypermethylation events, as opposed to the hypomethylation events, is more evident and is clearly linked to gene repression. One gene family cluster we identified as a new target of aberrant hypermethylation in breast cancer was the protocadherin cluster (PCDHA, PCDHB, and PCDHG) on chromosome 5. Further analysis showed that an overall decrease in the expression of the gene family cluster in breast cancer was correlated with the increased methylation of the PCDH cluster's CpG islands.
One mechanism that may be involved in the aberrant hypermethylation of entire gene family clusters is by disrupting the function of transcription factors and gene regulators that participate in the control of their expression. With respect to the protocadherin cluster, a mechanism involving this possibility exists. Recent studies indicate that binding sites for the insulator protein, CTCF, preferentially concentrate to the PCDH cluster ( 33, 34). Because CTCF DNA binding activity is reported to be attenuated in breast cancer ( 35– 37) and loss of its DNA binding activity has been linked to the spreading of aberrant DNA methylation into regions once protected by CTCF ( 38– 40), the aberrant DNA methylation of the PCDH gene cluster may be a result of compromised CTCF function. An alternative possibility is that aberrant DNA methylation appears first in the PCDH regions, and the increased levels of DNA methylation in CTCF recognition sites directly block CTCF interaction with the DNA because CTCF binding is inhibited by DNA methylation and therefore its functional role in regulating gene transcription ( 33, 34, 41, 42).
In another scenario, aberrant methylation may occur because of an altered chromatin structure that is secondary to changes in the protein composition of the chromatin interactions. We have previously observed DNA methylation–linked LRES in the HOXA gene cluster ( 9). This situation represents a potential case where changes in DNA-protein interactions may play a role in changes in DNA methylation. The histone methyltransferases EZH2 and MLL1 play important and opposing roles in HOXA gene regulation and their dysregulation has been linked to cancer etiology ( 43– 46). Considering their role in the maintenance of epigenetic control, their disruption may allow for or participate in the appearance of regional aberrant methylation. Another factor that may ultimately participate in the LRES seen in the HOXA cluster in breast cancer is retinoid receptors because retinoids are important participants in HOXA transcriptional control and important regulators of breast carcinogenesis ( 47, 48).
Additional mechanisms may participate in the establishment of the agglomerative DNA hypermethylation and likely act in concert with one another to produce stable LRES. It has been proposed that the low level of stochastic DNA methylation found in many CpG islands may serve as seeds or nucleation sites that trigger de novo methylation of the surrounding area leading to a more stable epigenetic silencing event ( 49). This model would be consistent with the results seen in the PCDH and HOXA gene clusters because there is a surprisingly high level of DNA methylation in some of their CpG islands in normal tissue and finite life span mammary epithelial cells. Indeed, it is further possible that these sequences themselves are more prone to DNA methylation ( 50).
In summary, recent epigenome-wide scanning approaches have allowed for the identification of new types of epigenetic lesions in the cancer genome ( 9– 12). One of these epigenetic lesions, LRES, is characterized by epigenetic mechanisms that work together to functionally inactivate long contiguous stretches of the genome; however, it seems that the combination of epigenetic landmarks associated with the long range silencing may arise from different epigenetic partnering schemes. For instance, Frigola and colleagues ( 10) and Hitchins and colleagues ( 11) found aberrant increases in DNA methylation and histone H3 K9 dimethylation that were linked to long range silencing, and pharmacologic reactivation studies implicated histone acetylation state, as well. Stransky and colleagues ( 12) found long-range silencing associated with H3 K9 dimethylation in the absence of aberrant DNA hypermethylation. In contrast, in our previous studies ( 9), as well as the ones reported here, we found long range silencing associated with aberrant increases in DNA methylation and loss of the histone acetylation modifications linked to a permissive or active chromatin state, but we found no consistent increases in the repressive histone modification marks, K9 or K27 methylation (Supplementary Fig. S3).
The differences in epigenetic landmarks associated with long-range gene silencing observed between the separate studies could be technical in nature where the ability to detect long-range epigenetic aberrations currently is unrefined and limited in power. It is likely that increases in the sensitivity and precision of epigenomic research tools and analytic schemes will help identify and more precisely define long-range epigenetic aberrations in human disease. Alternatively, these differences may reflect unique tumor-specific mechanisms because the studies by Frigola and colleagues ( 10) and Hitchins and colleagues ( 11) analyzed colon tumors, whereas Starnsky and colleagues ( 12) analyzed bladder tumors, and our studies focused on breast cancer. We speculate that different external stimuli (e.g., carcinogens and environmental agents) may disturb different nodes of epigenetic control that then combine to create aberrant epigenetic terrains that are cancer- or disease-specific and not seen in normal, nondiseased cells.
Disclosure of Potential Conflicts of Interest
The authors declare that there are no conflicts of interests.
Grant support: R01CA65662 and R33CA0951 (B.W. Futscher); Center Grants P30ES06694 and P30CA023074, and the BIO5 interdisciplinary biotechnology center at the UA supported the Genomics Shared Service; and training grants ES007091 and CA09213 (T. Jensen).
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 Jose Munoz-Rodrigues for outstanding technical support.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
- Received April 15, 2008.
- Revision received August 6, 2008.
- Accepted August 13, 2008.
- ©2008 American Association for Cancer Research.