Genomic Alterations Indicate Tumor Origin and Varied Metastatic Potential of Disseminated Cells from Prostate Cancer Patients

The metastatic potential of a tumor is presumably contained within cells that are able to escape and disseminate from the primary lesion. Consequently, the identification of cytokeratin-positive cells (i.e., cells of epithelial origin) in the peripheral circulation and bone marrow of patients with clinically localized adenocarcinoma of the prostate has generated much interest. These cells are hypothesized to originate from the primary tumor and represent a very early stage in the metastatic process (1). The presence of these putative disseminated tumor cells (DTC) in the bone marrow is particularly noteworthy because bone metastases are a well-known feature of lethal metastatic prostate cancer (2). It seems likely that these cells are the source of metastatic disease in those patients who develop secondary tumors years after resection of the primary tumor. Thus, DTCs have become a key target for risk-association studies and experiments characterizing molecular features that underlie the phenotypic traits responsible for survival, dormancy, and ultimately proliferation in environments distinct from their origin.

There is a compelling need for improved ways to evaluate patients with clinically localized prostate cancer for their risk of developing metastasis. Currently, risk is estimated by the pathologic features of the primary tumor (e.g., Gleason score and tumor stage) and prostate-specific antigen levels in peripheral blood (3, 4). These predictive features are used to stratify patients into low- and high-risk groups. However, up to 50% of patients classified as high risk for progression do not develop metastasis, and ∼10% of patients classified as low risk for progression subsequently develop secondary disease (5). Thus, the need to more accurately distinguish those men who would benefit from aggressive adjuvant treatment regimens from those who could be safely treated by active surveillance is a powerful motivation to improve on the ability to predict outcome.

To date, some studies have found a positive correlation with the presence of DTCs and risk of relapse following primary therapies for localized prostate cancer (6–10), whereas others have not (11–16). Among the latter studies, Pfitzenmaier and colleagues (2007) found that 75% of presurgery patients without clinical evidence of metastatic spread, and only 11% of normal controls, harbored disseminated epithelial cells in their bone marrow. This frequency far exceeds the expected frequency of biochemical recurrence or metastasis. These results strongly suggest that the mere presence of DTCs is not indicative of risk of disease progression. Why do so many patients possess DTCs, if escape from the primary tumor is a rate-limiting step in the metastatic process? Despite the low incidence in normal controls, it is possible that the disseminated cells are not actually tumor cells. Characterization of the molecular features of disseminated cells might resolve the question of their tumor origin and provide a better understanding of their relationship with disease progression.

Low-resolution and low-throughput methods such as fluorescence in situ hybridization (FISH; ref. 17) or conventional comparative genomic hybridization (CGH; ref. 18) have provided some evidence that DTCs from epithelial cancers possess genomic abnormalities. However, the molecular characterization of DTCs by high-resolution techniques has been hampered by the small number of DTCs obtainable: typically <50 epithelial cells from bone marrow aspirates are available for analysis. To date, high-resolution genomic analysis of prostate cancer DTCs was only possible by first generating cell lines from these cells (19).

Given the proposed role that DTCs play in systemic disease and the paucity of genomic analyses of these cells, we have used array CGH to produce high-resolution genomic profiles of cells expressing epithelial markers isolated from the bone marrow of prostate-cancer patients. We categorize our samples by the status of the patients from which they were acquired. Thus, our samples of disseminated cells obtained from patients with organ-confined (localized) disease are designated as LocDCs. Our samples of disseminated cells from patients with metastatic (advanced) disease are referred to as AdvDCs. Our study of genomic changes in LocDCs and AdvDCs lays the groundwork for a better understanding of the role these cells play in carcinogenesis.

Sample acquisition. All samples were collected from patients undergoing treatment at the University of Washington Medical Center or the Veterans Affairs Puget Sound Health Care System after informed consent was obtained. A total of 59 prostate cancer patients (11 patients with advanced disease and 48 patients with localized disease) participated in this study. Supplementary Table S1 gives the Gleason grade and tumor-node-metastasis stage for 55 of the patients and chemical castration status for all advanced patients. Clinical data on four localized patients was unavailable. The mean (SD) age of these patients was 60.2 (7.4) y. Our use of human samples for this study was approved by the Institutional Review Boards of the participating institutions.

Just before radical prostatectomy, bone-marrow aspirates were collected from the upper iliac crest of all patients with localized disease. Bone-marrow aspirates were collected from advanced patients before any treatment for four patients and during the course of treatment for seven patients. Pools of 10 to 20 epithelial cells were isolated and collected into 10 μL H₂O based on expression of EpCAM (CD326; a pan-epithelial cell antigen) from the aspirates of each of 58 patients and 3 biological replicates from the aspirate of 1 metastatic patient, as described previously (14). Twenty of the 59 samples (14 from localized-disease and 6 from advanced-disease patients) were collected using 1:20 dilutions of both an anti-EpCAM (CD326) antibody conjugated with a FITC isomer (Dako) and an anti-CD45 antibody conjugated to R-phycoerythrin (Dako) to label normal hematopoietic cells. This dual detection system was used to exclude normal lymphocytes from disseminated epithelial cell pools and to collect samples of normal cells. These normal-cell samples were analyzed by array CGH and used to define thresholds of loss and gain in disseminated cell arrays (see Supplementary Materials and Methods) and in proof-of-principle tests (see Results).

Prostate tissue samples containing tumor were collected from the patients with localized disease following radical prostatectomy and were embedded in freezing medium (Tissue-Tek OCT Compound, Sakura Finetek) and stored in liquid nitrogen. For nine patients from whom LocDCs were collected, the entire prostate was histologically evaluated (see Supplementary Table S1 for matched pairs). The area selected for molecular analysis was from the tumor, which was not multifocal in any of the cases. Tumor cells were isolated by laser capture microdissection (LCM) from 5-μm-thick tissue sections as described previously (20). Cells (2,000–4,000) were collected from each patient and from the normal stroma of five patients. Pathologist (L.D.T.) review of all LCM images confirmed collection of tumor cells. Arrays of stromal cell samples were used to define thresholds of copy-number change in arrays of primary tumor (see Supplementary Materials and Methods).

All other methods are described in Supplementary Materials and Methods.

Array CGH can be made compatible with small numbers of cells. To analyze samples containing only 10 to 20 cells by array CGH, we first amplified the genomic DNA using a ligation-mediated PCR method. This approach, which we term rare cell genomic amplification (RCGA), was developed by Klein and colleagues (18) to analyze the genome of a single cell by conventional CGH. Our CGH platform was a bacterial artificial chromosome (BAC) array, which contains ∼4,200 BAC clones, each with a known location in the publicly available assembly of the human genome (Build 35), and provides a resolution of ∼0.4 Mbp across the euchromatic portions of the genome (21). For all small-cell-number arrays, the reference is normal female genomic DNA amplified by RCGA from an amount of DNA roughly equivalent to the amount in the test sample.

We conducted tests to establish that RCGA and array CGH when applied to small numbers of cells give (a) consistently low levels of experimental noise, (b) acceptable dynamic range for detecting chromosomal alterations, and (c) reproducibility across replicates. First, we amplified nine samples each composed of 10 to 20 normal CD45⁺ bone marrow–derived cells (from patients with localized prostate cancer) by RCGA and analyzed them by array CGH. We used these normal-cell array results to define thresholds for calling loss and gain (see Supplementary Materials and Methods). Applying these thresholds back to the normal cell arrays, we observed the expected X- and Y-chromosome shifts for all these comparisons of male test to female reference DNA. On average, only 1% (40 of 4,096) and 0.6% (23 of 4,096) of autosomal BACs were encompassed by segments designated as lost and gained, respectively. No site was called a deviation in more than two normal-cell arrays, and only four loci encompassing a total of 28 BACs were called deviant in two of the nine (∼20%) normal-cell arrays. Thus, we report below only deviations detected in >20% of LocDC or AdvDC samples to minimize the risk of reporting false-positive deviant loci.

We next tested the ability of RCGA and array CGH to detect deviations in small numbers of cells of the LNCaP prostate cancer cell line, which has well characterized genomic abnormalities (22). Four samples each composed of 10 or 20 cells collected by micropipetting, and one sample composed of 200 ng of bulk genomic DNA purified from cells of the same passage, were independently amplified by RCGA for array CGH analysis. The analysis of bulk LNCaP DNA identified 13 sites of deviation, all previously reported for this cell line (22). Seven of the 13 (54%) deviant loci observed in the bulk sample were identified in all of the samples of 10 to 20 cells, and 11 of 13 (85%) were observed in at least one of the arrays from the 10- or 20-cell pools. The two deviations observed in the bulk sample but not in the small cell number samples each encompassed only two BACs. Thus, short deviations might be missed in arrays of DNA amplified from a small number of cells.

We found 8.5 additional deviant loci, on average, per array of 10 to 20 LNCaP cells. We selected five of these deviations to verify by FISH using one of the BACs encompassed by each deviation as FISH probe. FISH showed a heterogeneous mix of cells with and without the deviation of interest for three of the five copy-number alterations assessed (data not shown). Thus, random selection from a heterogeneous population contributes to the deviations detected by array CGH in small samples of LNCaP cells.

We next showed the reproducibility of our approach using three biological replicates composed of 20 DTCs each collected from a single patient with metastatic prostate cancer. Each 20-cell sample was independently isolated from bone marrow, amplified, labeled, and analyzed by array CGH. Figure 1A shows the high correlation between the normalized log₂ ratio values of each possible pair of replicates. Numerous sites of copy number change spanning many Mbp of genomic material were identified for this patient's samples. Figure 1B shows the considerable number of consistent deviant segments for the three replicates. Of the 3,077 BACs encompassed by alterations in one or more of these arrays, 72% (2,219 of 3,077) were identified in all three cell samples, and 88% (2,716 of 3,077) were deviant in two or more samples.

The numerous genomic alterations in AdvDCs include those associated with prostate cancer and disease progression. High-level amplifications as well as loss or gain of many segments encompassing several Mbp were detected in all 11 AdvDC samples. On average, each AdvDC array detected 13 gains and 14 losses. All chromosomes showed a deviation in at least one of the AdvDC samples. Figure 2 summarizes the deviations observed in the AdvDC samples. The list of copy-number changes observed in three or more samples (≥27%) is extensive (Supplementary Table S2).

The minimally overlapping regions (MOR) of copy-number change for deviations observed in at least four samples (36%) are given in Table 1. This list of MORs includes many chromosomal locations reported in the literature as frequent alterations for primary or metastatic prostate tumors, including loss on 8p, 10q, 13q, and 16q and gain on 8q and Xq. Our results help narrow in on the regions of interest for these prostate cancer–related alterations. We show five discrete MORs on 10q, three on 13q, three on 16q, and four on Xq.

Genomic alterations support the tumor origin of LocDCs. Based on the lack of a clear association between the mere detection of DTCs in men with clinically localized prostate cancer and subsequent relapse, an important question is if what we call DTCs are in fact tumor cells. To address this question, we compared the LocDC array results with the results from the primary tumor and AdvDC arrays. First, we found that the log₂ ratios of the deviant autosomal segments for the LocDCs have a significantly lower dynamic range than those of the primary tumors and AdvDCs (P < 0.05, Student's t test). The median log₂ ratio of gained segments in LocDCs was 0.13 (SD, 0.07), in primary tumors was 0.21 (SD, 0.11), and in AdvDCs was 0.15 (SD, 0.15). The median log₂ ratio of segments classified as loss for LocDCs was −0.15 (SD, 0.10), for primary tumors was −0.17 (SD, 0.09), and for AdvDCs was −0.17 (SD, 0.15). Second, we found fewer deviations, on average, in LocDCs (average, 14) than in primary tumors (average, 43) or AdvDCs (average, 27).

We determined that normal-cell contamination is unlikely to be responsible for the diminished dynamic range in LocDC arrays. The normalized log₂ ratio values for LocDCs collected using a marker (CD45) to exclude normal hematopoietic cells were not significantly different than those from LocDCs collected without this marker (P = 0.1362). Moreover, AdvDCs were collected by the same methods as LocDCs yet showed no evidence of normal-cell contamination.

Despite the less striking amplitude and number of deviations in LocDCs, we provide evidence that they are tumor cells. We exclude experimental noise as being responsible for the deviations observed in LocDCs. We compared LocDC array results to the normal cell arrays (produced from the nine samples of 10–20 normal CD45⁺ cells collected and amplified in parallel with the LocDC samples). The deviant segment values for gains and losses of autosomal material in the LocDC and normal cell arrays were not significantly different (P = 0.16 and 0.09, respectively, Student's t test). However, the number of alterations and the amount of genomic material altered were significantly greater in LocDC arrays than those found in the arrays of 10 to 20 normal cells. LocDC samples showed an average of 14 deviations (3.5-fold more than normal cells, P = 0.005) and ∼320 Mbp of altered genomic material (7.8-fold more than normal cells, P = 0.039).

Seventeen loci showed copy number deviation in >20% of LocDCs (Fig. 3; Table 2). Thirteen of these loci are likely to be real chromosomal deviations because they have no overlap with deviant segments in normal cell arrays. Losses involving 13q22 and 19q12 seen in >20% of the LocDCs encompass significantly more BACs than overlapping deviations registered in the normal cell arrays (18 versus 9, and 9 versus 2 BACs, respectively) and are thus also candidate tumor-related changes. We discount two deviations (gain in 4p16 and 11q13) as probable experimental artifact because they appeared in ∼20% of normal-cell arrays.

A comparison of the genomic profiles of LocDC and primary tumors indicates several overlapping regions of deviation (Fig. 3A; Table 2) supporting their common origin. Nine of the 15 (60%) sites of genomic change detected in >20% of the LocDC samples (excluding the two sites discounted as possible artifact) were also detected as deviant, in the same direction, in >20% of the primary tumor arrays. Moreover, 12 of the 15 loci showing change in >20% of the LocDC samples correspond to previously reported deviations in prostate cancer (Table 2; refs. 23–28), providing additional evidence that the LocDCs are prostate tumor cells.

Direct comparison of concordant sites in the nine matched samples of LocDCs and primary tumors provides additional evidence that LocDCs are tumor cells. A concordant deviation was defined as one in which >30% of BACs in one sample's deviation were encompassed by the other sample's deviation, and vice versa. There were a total of 29 concordant sites of deviation across the nine sample pairs (Table 3), an average of 3.2 concordant loci per pair. For 24 of the concordant sites, >50% of BACs were shared. Only one pair lacked concordant alterations. Using simulated data sets, we found that three of the matched pairs have significantly more concordant changes than the number expected if the deviant segments were randomly distributed across the two genomes.

The degree of concordance we observe is unlikely to be affected by the different amplification schemes used to amplify LocDCs (RCGA) and primary tumors (WGA2; see Supplementary Materials and Methods for detailed comparisons of RGCA and WGA2). In brief, arrays produced from LNCaP DNA amplified by these two methods showed significant concordance in their sites of genomic change (P < 0.0001) and no significant difference in the dynamic range of the deviant segments (P = 0.4820).

Frequent LocDC alterations are often detected in AdvDCs but not vice versa. We compared the array CGH profiles of LocDCs and AdvDCs to look for changes that might illuminate the shared or divergent biology of these cell types. Ten (67%) of the 15 deviations observed in >20% of the LocDC samples (excluding those discounted as artifacts) were also observed in >20% of the AdvDC samples (Table 2). This set of loci includes loss at 8p23, one of the putative tumor suppressor containing regions on 8p (29), which was observed in 27% (3 of 11) of AdvDC samples and in 23% (11 of 48) of LocDC samples. Two of the four most prevalent changes seen in the AdvDC samples were seen in >20% of the LocDC samples. Loss at 10q26 was observed in 55% (6 of 11) of AdvDC samples and in 25% (12 of 48) of LocDC samples. Loss at 16q21 was also observed in 55% (5 of 11) of AdvDC samples and in 31% (15 of 48) of the LocDC samples.

Given the remarkable degree of genomic change in AdvDCs and the more limited deviations in LocDCs, it is not surprising that the majority of deviations that we observed in >20% of AdvDC samples were deviant in <20% of LocDC samples (Fig. 3B; Supplementary Table S2). For example, losses that overlap with 8p12-21 and 8p22 were observed in 4 of the 11 (36%) AdvDC samples, 5 of the 9 (56%) primary tumor samples, but only 1 of the 48 (2%) LocDC samples. Two deviations with considerable frequency differences were a gain of most of 8q (36% of AdvDCs and 10% of LocDCs) and gain of 1q32 (45% of AdvDCs and 8% of LocDCs).

Regions of frequent chromosomal alteration in DTCs and primary tumors are enriched for genes involved in specific biological processes. We sought to determine if the chromosomal regions altered in DTCs and primary tumors affect genes of particular biological processes based on gene ontology (GO) enrichment analysis. We considered all regions altered in >20% of the LocDC, AdvDC, or primary-tumor samples (Supplementary Table S3), identified the genes corresponding to sites of loss or gain, and used hypergeometric tests (30) to determine if any GO categories were enriched (at a significance level of P < 0.001) in those gene sets. The enriched categories are given in Supplementary Table S4 (losses) and Supplementary Table S5 (gains). We also include a second P value, the SimPValue, which is an indicator of those GO categories that might be enriched due to gene clustering. Thus, categories with a SimPValue > 0.05 should be considered with caution.

The notable GO categories significantly enriched in alterations for >20% of the LocDC, AdvDC, or primary-tumor samples with a SimPValue < 0.05 are as follows. Of the four categories enriched in the LocDC regions of gain, one was vitamin transport. This particular GO enrichment included FOLR1, a gene whose transcript and protein product are increased in specific epithelial tumors and are positively associated with tumor stage and grade (31). DNA metabolic process and chromatin modification were 2 of the 6 categories enriched in primary tumor gains, and cell-cell signaling was 1 of the 12 categories enriched in primary-tumor losses. Various immune-related processes made up the six categories enriched in the AdvDC regions of gain. Only electron transport was an enriched category in AdvDC losses (SimPValue = 0.001). One of the electron transport genes in AdvDC regions of loss was the tumor suppressor WWOX (32).

A few enriched gene categories with SimPValue > 0.05 represent biological processes that are sufficiently compelling to note here. Two categories related to adhesion [cell-cell adhesion (SimPValue = 0.051) and homophilic cell adhesion (SimPValue = 0.054)] were enriched in LocDC regions of loss. Transcription, DNA packaging, chromatin assembly, and protein-DNA complex assembly were GO categories enriched in the primary tumor gains. Eight categories related to regulation of transcription, metabolic, cellular, and biological processes were enriched in AdvDC losses.

To the best of our knowledge, this study represents the highest-resolution genome-wide analysis of changes performed directly on prostate-cancer DTCs. The genome analysis of a small number of cells was made possible by combining RCGA, a whole-genome amplification scheme developed by Klein and colleagues (18), and array profiling by CGH. The use of a spotted DNA array as a CGH platform detects loss and gain at a high resolution and provides a direct link to the sequence of the human genome, a considerable improvement over conventional methods. Our proof-of-principle studies show that it is possible to produce genomic profiles of a small number of cells that exhibit relatively low levels of experimental noise, highly concordant results between bulk DNA and samples of a few cells, and excellent reproducibility across biological replicates.

We show that LocDCs have real genomic deviations and are composed of tumor cells, although the genomic deviations in the LocDC samples were generally fewer in number and showed lower dynamic range than those in primary tumors and AdvDCs. First, we see significantly more deviations in the LocDC arrays than in the arrays of a comparably small number of normal cells, arguing against experimental noise as the source of alterations in LocDCs. Second, 80% of the deviant sites detected in >20% of the LocDC samples were previously reported in the literature as altered for prostate tumors. Third, we show significant concordance between a third of the matched pairs of LocDCs and primary tumors and that 60% (9 of 15) of the deviations observed in >20% of all the LocDC samples were identified in our primary tumor samples.

AdvDCs are unmistakably tumor cells, given their numerous genomic alterations. Moreover, AdvDCs frequently possess deviations known to associate with a progressed disease state, deviations that were infrequently observed in LocDCs. A notable example is gain of 8q (36% of AdvDCs and 10% of LocDCs), an alteration associated with advanced disease and poor outcome in patients with organ-confined disease (23, 33, 34). Regions containing the metastasis suppressor genes MKK4 (17p12) and KAI1 (11p11.2) were each deleted in ∼30% of AdvDC samples but very few of the LocDCs. One third of our AdvDC samples and none of the LocDC samples had a high-level amplification of the region on chromosome X encompassing the androgen receptor gene, a well-characterized alteration specific to advanced disease (35, 36). Gain of 1q32 was one of the most common alterations found in AdvDCs (45%), uncommon in LocDCs (8%), and absent in our set of primary tumors. Notably, 50% of the metastatic tumors analyzed by array CGH in a parallel study also have a gain of 1q32,⁶

Changes that we frequently observed in both AdvDCs and LocDCs might reflect their common origin or shared state of dissemination. Arguing for the latter, both DTC types showed frequent losses in 8p23, 10q, 13q, and 16q, alterations that have been frequently identified in prostate cancer (33). Deviations found in both types of DTCs but that were absent from our primary tumor samples might be alterations related to dissemination. These sites include losses in 10q25.1, 13q22.2-32.1, 13q32-34, and 16q21. Interestingly, the loss in 16q21 encompasses two members of the cadherin family of adhesion genes (CDH8 and CDH11). Disruption of the cadherin pathway is implicated in tumor invasiveness and disease progression for several carcinomas (37). Analysis of a much larger set of matched primary and DTC samples will be needed to confirm what alterations are in fact specifically associated with dissemination.

The reduced amplitude of the genomic changes in LocDCs compared with AdvDCs and primary tumors could have several explanations. Copy-number changes can be dampened in the array CGH readout by normal-cell contamination, abnormal ploidy, and heterogeneity. As we have shown, normal-cell contamination of our DTC samples is unlikely. Extra ploidy is also unlikely to be a factor in the LocDCs, as it is well established that most early prostate cancers, from which the LocDCs are presumably disseminated, are approximately diploid (38).

Genomic heterogeneity of LocDCs, as is seen in prostate tumors (39), is likely to contribute to a reduction in amplitude of LocDC alterations. Moreover, early dissemination might contribute additional heterogeneity to LocDCs relative to what might be found in the tumor itself. Inclusion of cells that migrated before acquiring multiple genome changes in the pool of LocDCs will dilute and reduce the overall number of detected alterations relative to the primary tumor at the time of resection, which we observe. Early dissemination of LocDCs might also explain their notably infrequent possession of 8p12-22 loss, the most common copy-number alteration reported in prostate cancer (35). Loss in this region was seen in only 1 of our 48 LocDC samples, in contrast with 54% of our primary tumors and 55% of our AdvDC samples.

How do our LocDC findings fit the current models of metastasis? The two predominant models are that (a) rare, highly metastatic cells within the primary tumor give rise to metastases or (b) particular tumors, not rare cells, are prone to metastasis. In either case, the implication is that cells leave the primary tumor in possession of the molecular alterations necessary for secondary growth and survival. However, it is hard to imagine that the cells that generate secondary disease in patients many years after the removal of their primary tumor possess full metastatic potential (40). Theories of tumor cell dormancy have been proposed recently (41–43), suggesting that DTCs from primary tumors might represent a dormant population of cells that can, through selection and mutation over time, acquire key molecular changes that will permit metastatic growth (41). Dormancy is supported by our GO analyses, which showed an enrichment of genes related to proliferation in sites of frequent alteration in our primary tumor and AdvDC samples, but not in our LocDCs, indicating that growth is not a prevailing theme in LocDC alterations. Thus, additional genomic hits would seem to be required, if LocDCs are to become overt metastases.

The high fraction of prostate-cancer patients with localized disease in whom DTCs can be detected and the relatively low level of genomic disorder shown here for such cells are persuasive evidence that LocDCs have not, in the majority of cases, reached full metastatic potential. If a subset of these cells can produce clinically significant metastases, then it seems likely, from the large number of studies showing gross chromosomal deviations in metastatic tumors, that these cells must eventually undergo additional genomic events. However, a subset of the genomic abnormalities that we identified in LocDCs may prove to be harbingers of progressive disease. Follow-up clinical studies of the patient population studied here along with additional molecular characterization of LocDCs could provide insight into the metastatic potential of DTCs.

Klein and coworkers recently reported array CGH analysis of single disseminated breast-cancer cells isolated from bone marrow (44).

No potential conflicts of interest were disclosed.

Grant support: Pacific Northwest Prostate Cancer Specialized Program of Research Excellence grant CA97186 (I.N. Holcomb, M. Kinnunen, C.L. Friedman, I.S. Gallaher, T.M. Morgan, P.S. Nelson, P.H. Lange, W.J. Ellis, L.D. True, B.J. Trask, and R.L. Vessella), U.S. Department of Defense grant DAMD 17-03-2-0033 (I.N. Holcomb, D.I. Grove, P.S. Nelson, L. Hsu, B.J. Trask, and R.L. Vessella), NIH/National Human Genome Research Institute Genome Training grant T32 HG00035 (I.N. Holcomb and B.J. Trask), National Cancer Institute (NCI) grant P50-CA97186 (D.I. Grove and L. Hsu), National Institute on Aging grant RO1-AG14358 (D.I. Grove and L. Hsu), Fred Hutchinson Cancer Research Center Institutional Funds (C.L. Sather, J.J. Delrow, and B.J. Trask), NCI grant P30-CA15704-31 (C.L. Sather and B.J. Trask), NIH grant RO1-DC004209 (J.M. Young), NIH grant RO1 CA098415 (L. Hsu), NIH grant RO1-CA95717 (L. Hsu and B.J. Trask), and NCI grant U24-CA80295 (B.J. Trask).

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 the members of the Trask, Vessella, and Nelson laboratories, especially Roger Coleman, Ilsa Coleman, Edward Arfman, Christine Wang, Bryce Lakely, and Mengchu Wu, for their support and technical expertise.