Isolation and genomic analysis of circulating tumor cells from castration resistant metastatic prostate cancer

Clinical samples were obtained from 20 castration resistant prostate cancer patients who were recruited from November 2008 to August 2009 at the University of California San Francisco (UCSF) to participate in our study (Table 1). Twenty patients were selected based on statistical considerations for a pilot study. All patients gave informed consent under a protocol approved by the UCSF Institutional Review Board. Blood was collected in three CellSave™ tubes (Veridex). This allowed us to use one tube for an initial CTC enumeration using the published CellSearch™ method, and the two remaining tubes to be combined for IE/FACS isolation. Microdissection and DNA extraction of archived formalin-fixed paraffin-embedded (FFPE) tumor biopsy samples from two patients were done in a manner as previously described [7]. Using H&E stained slide, the pathologist (J.S.) demarcated tissue containing at least 80% tumor. The H&E stained slide was then used as a guide to microdissect the tumor area using a surgical blade (Feather #15) on non-stained deparaffinized sections (10 sections × 5 μM thickness). In the case of a small tumor (Patient #20), manual microdissection of the selected area was performed by scraping successive tissue sections under a stereomicroscope.

This pilot study was conducted among men with prostate cancer who met the following criteria: histologically-proven castration resistant prostate cancer that has metastasized; prior therapy with ≥ 3 cycles (9 weeks) of docetaxel based chemotherapy; progressive disease within the last 1-2 months as characterized by rising PSA level, new lesions on radionuclide bone scan or computed tomography scans; last dose of docetaxel ≥ 21 days prior to blood collection; no chemotherapy within the previous 21 days and no radiation therapy within 21 days.

To increase the likelihood of isolating CTCs for molecular analysis, we first performed CTC enumeration using the automated CellSearch™ assay to identify patients with > 5 CTCs per 7.5 mL. We then isolated CTCs by IE/FACS (see below) from the remaining blood sample in these patients. Cell counts via CellSearch™ correlated closely with cell counts via IE/FACS (Park, unpublished data). In initial studies, we observed that whole genome amplification of DNA from CTCs was less efficient when CTCs were isolated after 48 h post-phlebotomy. Therefore, enumeration was performed within 24 h and isolation of CTCs via IE/FACS within 48 h after blood draw.

For immunomagnetic enrichment, one-half volume of Cell Buffer (Immunicon) was added to a blood sample. Immunomagnetic particles coated with EpCAM (MJ37) mAb (50 μg/10 mL of indirect conjugate or 60 μg/10 mL for the direct conjugate) and EpCAM (EBA-1) mAb conjugated to phycoerythrin (PE) (400 μL at 5 μg/mL per 10 mL of sample) were added to the sample. The sample was incubated at room temperature for 15 min, mixing at half-way point and again when incubation was completed. The sample was placed in a magnetic separator (Immunicon) at room temp for 15 min, and then aspirated to remove unbound blood cells. The bound cells were eluted in 2 mL of Cell Buffer, moved to a 12 × 75 mm tube and subjected to a second round of magnetic separation for 5 min. The fluid in the tube was aspirated and the remaining cells were resuspended in 150 μL of Cell Buffer. Twenty μL of a solution containing 2 μg of a proprietary (BD Bioscience) nucleic acid dye, and 0.1 μg of a leukocyte-specific CD45 (2D1) mAb conjugated to peridinin-chlorophyll-protein-Cy5.5 were added to resuspended cells. The sample was incubated in the dark for 15 min and then sorted using FACS Aria (BD Biosciences). CTCs were defined as nucleated, EpCAM-positive, and CD45-negative while leukocytes were defined as CD45-positive, EpCAM-negative and nucleated. Cells were sorted into 10 μL of TE (10 mM Tris-HCl 1 mM EDTA pH 8.0) in a 500 μL PCR tube. All conjugated antibodies and reagents for FACS were obtained from BD Biosciences unless otherwise noted. During FACS analysis, we sorted a range of 18-50 CTCs. Table 1 shows the number of "input cells" used for WGA, and the number of replicate analyses performed for each subject. Due to the small numbers of CTCs isolated from patients' blood, it was not feasible to perform FACS analysis after sorting. Also, since blood samples were collected in CellSave™ (Veridex) tubes [6, 8], CTCs were thereby rendered nonviable and fixed to maintain antigenicity prior to immunomagnetic enrichment and FACS sorting. Complete FACS plots from two representative patients are shown in Additional file 1: Figure S1.

ACGH protocols typically use between 100 ng and 1 μg of specimen DNA in the labeling reaction, equivalent to ~20,000-200,000 cells, which does not permit reliable analysis of important samples containing small numbers of tumor cells such as CTCs in blood. Analyzing these samples for copy number aberration requires a high-fidelity whole genome amplification (WGA) step. To select the best strategy for WGA of limiting amounts of input DNA, we tested four WGA protocols: multiple displacement amplification [9], random prime amplification [10], ligation-mediated-PCR [11], and the GenomePlex^® WGA4 [12, 13]. We compared aCGH results generated from diluted BT474 breast cancer cell line genomic DNA (equivalent to ~10,000 to 10 cells) to non-limiting unamplifed BT474 DNA (600 ng). We found GenomePlex^® WGA4 to be most reliable and reproducible (data not shown). We also did not observe artifacts or bias via BAC aCGH analysis. Other groups have also demonstrated the reproducibility and reliability of the GenomePlex^® WGA protocol [12, 14].

To prevent any loss of genomic material, we avoided the use of column purification of genomic DNA. Instead, whole cell lysis of isolated cells and WGA of genomic DNA were performed in the same PCR tube the cells were sorted into. Genomic DNA in the whole cell lysate served as template for WGA which was performed using the GenomePlex^® Single Cell WGA4 Kit (Sigma-Aldrich) following the manufacturer's protocol. Briefly, 1 μL lysis and proteinase K digestion solution was added to cells in 10 μL of TE. The DNA was then fragmented and libraries were prepared. Amplification was performed by adding 7.5 μL of 10 × Amplification Master Mix, 48.5 μL of nuclease-free water and 5 μL WGA DNA polymerase. Samples were amplified using an initial denaturation of 95°C for 3 min followed by 25 cycles, each consisting of a denaturation step at 94°C for 30s and an annealing/extension step at 65°C for 5 min. Amplified DNA was purified using the QIAQuick PCR Purification Kit (Qiagen). DNA concentration was determined by a NanoDrop™ spectrophotometer and stored at -20°C. The average yield after WGA of ~20 CTCs is ~10 μg. Early tests via aCGH analysis of samples with WGA product yield less than < 5 μg revealed non-specific amplification. Therefore samples with low yield (< 5 μg) were excluded from further downstream assays.

In initial studies we performed aCGH analysis using oligonucleotide microarrays (Affymetrix 50 K SNP array), but observed significant experimental variability and background noise with our WGA4 amplified DNA samples. Similarly, studies by Fuhrmann et al [15] and Nowak et al [14] showed that bacterial artificial chromosome (BAC) arrays outperformed oligonucleotide microarrays in detecting small gains or losses [15] and showed far fewer outliers and less technical noise [14]. So, we performed our aCGH analysis using a bacterial artificial chromosome (BAC) array (Version 3.2) containing 2,464 clones printed in triplicate [16]. The BAC arrays are available at the Helen Diller Family Comprehensive Cancer Center Array Core.

The random prime amplification method [16] was used to enzymatically label approximately 600 ng of test DNA (WGA product) or unamplified normal female genomic (Promega) reference DNA with Cyanine 3-dCTP or Cyanine 5-dCTP (Amersham), respectively. Labeled test and reference DNA were co-hybridized to a BAC array. The arrays were imaged using a custom built CCD array imaging system [17] and analyzed using the UCSF Spot Program to calculate Cy3/Cy5 ratios. The UCSF Spot Program is a fully automated system for microarray imaging that locates both subarray grids and individual spots [18]. Ratios are then computed based on explicit segmentation of each spot. The resultant log2 ratios were analyzed by UCSF Sproc to yield the mean and standard deviation of the log2 ratios over replicate spots. The UCSF Sproc is a companion program to UCSF Spot that maps each spot on the array to a specific clone and chromosome position, and averages over replicate spots in order to output a final ratio value for each clone on the array. Clones with a standard deviation of the log2 ratios of the triplicate spots greater than 0.2 for were excluded from analysis.

To determine copy number alterations, microarray data (sproc files) was uploaded and analyzed using Nexus 5.1 software (Biodiscovery). Data was subjected to rank segmentation, a proprietary (Biodiscovery) variation of the circular binary segmentation (CBS) [19]. The thresholds of log2 ratio values for gains and losses were 0.2 and -0.2 respectively; the thresholds for high copy number gains and homozygous deletions were 0.6 and -0.6 respectively. Copy number gains or losses were considered recurrent if present in > 50% of the CTC samples from 9 patients.

Since sex-mismatch hybridization was performed, i.e. male test (tumor) versus female (normal) reference, we analyzed copy number alterations on chromosome X by using a feature available in the Nexus 5.1 software for automatic threshold adjustment on the X chromosome. Theoretically, the intensity of a normal male test sample on the X chromosome is 1/2 that of the normal female reference (i.e. 1 versus 2 copies of chromosome X). If the same threshold used for autosomal chromosomes was applied on X, the segmentation result will show "one copy number loss" But when corrected for gender, the segmentation analysis will no longer pick up the "copy number loss on X" because the adjusted threshold takes into account that the male sample will have 1/2 the intensity on the X chromosome compared to that of the female reference. Therefore, the automatic threshold adjustment feature allows the segmentation analysis to take into account the sex-mismatch hybridization so only copy number changes on the X chromosome that deviated from the expected normal ratio was reported. Chromosome Y was excluded in the analysis.

Concordance between aCGH profiles was measured using weighted Pearson correlation. A weighted Pearson measure was used because, in aCGH comparisons, unweighted Pearson correlation is driven by the number of ratios that differ from "normal." Since the observations in this case are the clones, one can use the spatial relationship of the clones to estimate the weights. Based on this idea, the weight of a clone was estimated by considering its deviation and that of its adjacent clone from "normal" The formula used for calculating weighted correlation coefficient (corr) is given below:

where x and y are the log2 ratio values of the two samples under consideration; w is the weight vector; w i = 1 2 ( m e d i a n ( | x j - x r e f | ) + m e d i a n ( | y j - y r e f | ) ); i = 1,2,...,n-1; j = i, i + 1; n = total number of clones; and x _ref and y _ref are the medians of the n clones for samples x and y, respectively. Only autosomal clones were considered. Correlation coefficients falling in the intervals 0 to < 0.20, 0.20 to < 0.40, 0.40 to < 0.60, 0.60 to < 0.80 and 0.80 to 1 were said to be un-, lowly-, fairly-, moderately- and highly- correlated, respectively.

For proof of principle studies using a spiked tumor cell model system, we chose LNCaP, PC3, and VCaP cell lines because they have well-characterized copy number aberrations [20]. Of note, initial tests on diluted genomic DNA and isolated spiked cells using breast cancer cell line BT474 showed that the optimal input for WGA which yielded robust aCGH results was ~20 cells (data not shown). For each prostate cell line, 1000 trypsinized cells were spiked into 10 mL of whole blood from healthy volunteers. Twenty (20) tumor cells were collected in duplicate via IE/FACS. Isolated cells were then subjected to WGA and aCGH analysis. The resulting genomic profiles demonstrated the expected known copy number aberrations (Figure 1A-C), as did aCGH profiles from amplified genomic DNA (50 ng) and non-amplified genomic DNA (600 ng) performed as a positive control. Profiles derived from 20 tumor cell-inputs revealed moderate to high correlation with the positive control (Table 2) indicating minimal dilution from contaminating normal leukocyte DNA, which would have otherwise dampened dynamic range [21]. Furthermore, consistent with previous genomic studies on VCaP cells [20], a focal amplification in the X-chromosome region containing AR was observed (Figure 1C).

Twenty (20) sequential CRPC patients were assayed for CTC level by CellSearch™ (see Additional file 2: Figure S2 for study schema). Fifteen (15) patients showed ≥ 5 CTCs per 7.5 mL, the cutoff used in prior clinical trials of this assay in metastatic prostate cancer [6], with mean = 12.2 and median = 3.0 CTC/mL (Table 1). Of these 15 patients who were positive for CTCs, twelve (12) were subjected to CTC isolation via IE/FACS. Three patients were excluded from further analysis because of insufficient amount of remaining blood. Nine (9) were successfully analyzed for copy number aberrations. Three of the 12 samples that met the established parameters to move forward with IE/FACS did not yield informative aCGH results (25% failure rate) due to FACS instrument error, inadequate WGA yield, or failed microarray hybridization. Two of the samples (patients #15 and #19) were processed 2 days after blood draw (Table 1). Based on this observation, we suggest implementing IE/FACS within 24 h after blood draw.

Isolated CTCs from these CRPC patients exhibited copy number aberrations indicative of malignant origin. Due to the paucity of FACS sorted CTCs, re-analysis by FACS to confirm purity and the correct expression of biomarkers was not performed; hence, the isolation of falsely positive "EpCAM+/CD45-" cells cannot be ruled out. However, the resulting aCGH profiles showed unequivocal evidence of cancer-associated genomic alterations, and did not appear to be significantly dampened by false positive normal cells. For example, copy number analysis of replicate CTC isolates from patient #9 revealed gains including focal amplification in 11q13 (includes CCND1) and high-level gains in a region on chromosome X including the AR locus (Figure 2A). Also, replicates were moderately correlated (r = 0.78) indicating the reproducibility of the assay. As expected, these copy number aberrations were not observed in 100 leukocytes isolated from the same enriched sample (Figure 2B, Additional file 3: Figure S3C).

ACGH analysis of CTCs from all 9 patients revealed a wide range of copy number aberrations, including those that have been previously reported in prostate tumors ([22, 23] Figure 3 and Additional file 3: Figure S3). Copy number aberrations present in > 50% of the samples include gains at 1q, 4q, 7q, 8q, 9p, 9q, 11q, 19q, Xp, and Xq and losses in 1p, 4q, 5q, 8p, 10q, 11q, 13q, 14q, 15q, and 16q (Additional file 4: Table S1). Of note, high-level gains in the region of chromosome X containing the AR locus were observed in 7 (78%) of the 9 cases, while low level copy number gains were observed in the remaining 2 (22%) cases (Figure 4). These results suggest that CTCs may reveal evidence for AR amplification, which has been associated with disease progression in CRPC [24].

In patients #17 and #20, archival tissue previously obtained at time of primary tumor diagnosis, and prior to any cancer therapy, was available. These tumor specimens were subjected aCGH analysis and compared to the corresponding CTC samples.

Patient #17 is a 48-year old CRPC patient initially diagnosed with prostatic adenocarcinoma with Gleason score 9 (4 + 5), and with local extension to the urinary bladder. The aCGH analysis of his CTCs revealed genomic aberrations including high-level gains at 18p, 19q and the AR locus of chromosome X; and losses at 8p and 18q (Figure 5A, Additional file 3: Figure S3H). The archival primary tumor specimen involving the bladder from this same patient, which dated to 1 year and 3 months prior to CTC analysis, was obtained for aCGH analysis as well. Only tumor from the bladder yielded informative aCGH profile (Additional file 3: Figure S3I). Comparison of the respective genomic profiles revealed shared copy number aberrations indicative of common lineage, including losses in 8p and 18q as well as gains in 8q, 18p and 19q (Additional file 3: Figure S3A). Conversely, new copy number aberrations were observed in CTCs but not in the primary tumor, including copy number gain at the AR locus (Figure 4).

Patient #20 is a 67-year old CRPC patient initially diagnosed with prostatic adenocarcinoma with Gleason score 8 (4 + 4). ACGH analysis of his CTCs revealed genomic aberrations including high-level gains in 8q and in a region on chromosome X containing AR, as well as loss of 13q (Figure 5B; Additional file 3: Figure S3K). ACGH analysis of the corresponding primary tumor (Additional file 3: Figure S3L) obtained via core biopsy 5 years and 1 month prior to CTC analysis revealed similarities in copy number alterations when compared with the CTCs, including losses in 8p, 10q, and 11p and gains in 8q and 9q (Additional file 5: Figure S4B). In addition to these conserved genomic aberrations, new copy number changes were evident, particularly the AR region gain shown in CTCs but not in the primary tumor (Figure 4).