Ethics statement

Bone marrow sampling was performed within the study protocol of the GEBDIS study at the Central Hospital in Augsburg after informed written consent of patients was obtained. The ethics committees of the University of Tübingen and of the University of Regensburg (ethics vote number 07-079) approved bone marrow sampling and genomic analysis of the isolated cells. Additionally, as control and reference samples, we used single cells from the mononuclear cell fraction of peripheral blood obtained from five healthy donors. Three donors provided written informed consent after obtaining approval by the ethics committee of the University of Regensburg (ethics vote number 12-101-0038) and two healthy donors provided verbal informed consent. The latter samples were taken before 2008 when no ethics vote for voluntary blood donations of healthy donors was required.

Cell lines

BT-474 and SKBR3 cell lines were obtained from repository of the German Collection of Microorganisms and Cell Cultures at the Leibniz Institute DSMZ (Braunschweig, Germany), other cell lines including MDA-MB-453 and MDA-MB-361 were obtained from the American Type Culture Collection (ATCC). OE-19 was obtained from the European Collection of Cell Cultures. PT1590 cell line was originally generated at the University Medical Center Hamburg-Eppendorf [28]. All cell lines have been maintained in the conditions recommended by the distributor. Identity of all cell lines was confirmed using a PCR-based fingerprinting.

Detection and isolation of disseminated cancer cells

The procedure for bone marrow preparation has been described elsewhere [12], [29]. Bone marrow or blood cells (1–2×10⁶) were stained using either monoclonal antibody clone A45-B/B3 against cytokeratins 8,18,19 or monoclonal antibody clone CK2 against cytokeratin 18. Control cells (from healthy donors) were stained with monoclonal antibody specific for vimentin (clone V9, Dako). Mouse IgG1 Kappa (MOPC-21) was used as isotype control for all immunocytochemical experiments. Visualization was carried out using alkaline phosphatase/anti-alkaline phosphatase technique with 5-bromo-4-chloro-3-indolyl phosphate and nitroblue tetrazolium (BCIP/NBT) used as substrate.

Primary whole genome amplification (WGA)

All single cells, cell pools (few hundred cells) and microdissected specimens from formalin fixed paraffin embedded (FFPE) tumor tissue were amplified using the Ampli1™ WGA Kit (Silicon Biosystems) [12], [19].

Re-amplification of the WGA products

Re-amplification was performed in a volume of 50 µl. Each PCR reaction was composed of the following ingredients: 5 µl Expand Long Template Buffer 1 (Roche Diagnostic), 1 µM of the LIB1 (5′-TAGTGGGATTCCTGCTGTCAGT-3′) or MseLig-21 primer (5′-AGTGGGATTCCTGCTGTCAGT-3′) – depending on the adapter used in the primary WGA, 1.75 µl dNTPs (10 mM), 1.25 µl BSA (Roche Diagnostic), 2.5 U of Expand-Long-Template DNA Polymerase (Roche Diagnostic) and 1.0 µl of the template DNA. The MJ thermocycler was set as follows: 1 cycle of 94°C for 60 sec, 60°C for 30 sec, 65°C for 2 min, 10 cycles of 94°C for 30 sec, 60°C for 30 sec, 65°C for 2 min (extended by 20 sec/cycle). Typically three reactions were run in parallel, which were pooled and used as template for DNA labeling and aCGH. A negative control was included in every run.

Labeling of sample DNA

Random-primed DNA labeling approach (RP labeling). Test and reference DNA samples were labeled using SureTag DNA Labeling Kit (Agilent Technologies) according to the instruction provided by the supplier (Agilent Oligonucleotide Array-Based CGH for Genomic DNA Analysis, version 7.1, December 2011). Briefly, 1.5 – 2.0 µg of the purified input DNA (WGA product or unamplified genomic DNA) was supplemented with 5 µl of Random Primer Mix and filled up with H₂O to 31 µl. Unamplified DNA and WGA products samples were denatured at 95°C for 10 or 3 minutes, respectively. Sample tubes were transferred on ice and incubated for 5 min. The labeling reaction with exo-Klenow fragment consisted of the following ingredients: 31 µl of denatured DNA, 10 µl of 5x Reaction Buffer, 5.0 µl of 10x dNTP Mix, 3.0 µl of Cy5-dUTP (test) or Cy3-dUTP (reference) and 1.0 µl of Exo(–) Klenow fragment. Labeling reaction was run at 37°C for two hours, followed by an inactivation step at 65°C for 10 minutes. Labeled DNA was purified using Ultra 0.5 purification system with a size cut-off of 30 kDa. DNA yields and dye incorporation rates were quantified using the NanoDrop ND-1000 instrument.

PCR-based labeling. PCR-based labeling using dye-conjugated universal primer (PCR-T1). Placement of the dye on the universal primer provides the advantage that all restriction digestion fragments present in the WGA product irrespectively of their size will be labeled with the same amount of dye. To avoid cross-hybridization of adapter sequences flanking amplicons in the WGA products, test and reference samples were labeled using different PCR-adapters. Test samples were labeled with the PCR-adapter incorporated in the Ampli1™ WGA kit, while all the reference DNA samples were amplified using the following adapters: MIB5 (5′-TGAGCTGGTCATTGCGCATGGT-3′) and ddMse XI (5′-TAACCATGCGC-3′). Universal primers used in the labeling reaction were directly conjugated with either Cy5 in the case of Ampli1™ universal primer [5′-TAGTGGGATTCCTGCTGTCAGT-3′] or Cy3 in the cases of MIB5 primer [5′-TGAGCTGGTCATTGCGCATGGT-3′] (underscores indicate the placement of the dye). The labeling PCR was run in a total volume of 50 µl reaction composed of 5 µl of 10x Expanded Long Template Buffer 1 (Roche Diagnostics), 2.4 µM of dye-conjugated LIB1 or MIB5 primer (for test and reference sample, respectively), 350 µM of dNTPs, 0.5 µl BSA (Roche), 3.75 U of Expand-Long-Template DNA Polymerase (Roche Diagnostics) and 1.0 µl of the template DNA. The PCR was programmed as follows: 10 cycle of 94°C for 15 sec, 51°C for 30 sec and 65°C for 3∶30 min, 2 cycles of 94°C for 15 sec, 51°C for 30 sec and 65°C for 3:30 min (extended by 10 sec every cycle), followed by a final elongation step of 7 min at 65°C. Products were purified using the Amicon Ultra 0.5 System (cut-off size 100 kDa) and quantified with the NanoDrop ND-1000 instrument.

PCR-based labeling using incorporation of dye-conjugated dNTPs (PCR-T2). WGA products were labeled by PCR in the presence of Cy5 or Cy3 conjugated dCTP and dUTP. Incorporation of the fluorescent dyes using the dNTPs provided the advantage that adapters flanking each amplicon in the WGA product could be removed prior to aCGH hybridization, thereby avoiding unspecific cross-hybridization of test sample with reference DNA and oligonucleotide probes on the array. For each sample two 50µl PCR reactions were run in parallel each comprising 5 µl of 10x Expanded Long Template Buffer 1 (Roche Diagnostics), 2.4 µM of the universal primer (LIB1 or MseLig-21, depending on the adapter sequence incorporated in the WGA product), 350 µM of dATP and dGTP, 315 µM of dCTP and dTTP, 35 µM of Cy5/Cy3-dCTP and Cy5/Cy3-dUTP (GE Healthcare), 0.5 µl of BSA (Roche Diagnostics), 7.5 U of Expand-Long-Template DNA Polymerase (Roche) and 0.5 µl of the template DNA. The PCR cycler was programmed as follows: 10 cycle of 94°C for 15 sec, 60°C for 30 sec and 65°C for 3:30 min, 2 cycles of 94°C for 15 sec, 60°C for 30 sec and 65°C for 3:30 min (extended by 10 sec every cycle), followed by a final elongation step of 7 min at 65°C. Subsequently, the resulting PCR products were subjected to digestion with Tru1I restriction endonuclease to cleave off the PCR-adapters. For this purpose the PCR product was supplemented with 5.6µl of Buffer R and 30 U of Tru1I. Digestion was performed at 65°C for 3 hours. Resulting products were pooled and purified using the Amicon Ultra 0.5 System (100 kDa cut-off). DNA yields and dye incorporation rates were quantified using the NanoDrop ND-1000 Instrument.

Array comparative genomic hybridization

Array CGH was performed on oligonucleotide-based SurePrint G3 Human CGH 4×180K microarray slides (design code: 022060) according the protocol provided by the manufacturer (Agilent Oligonucleotide Array-Based CGH for Genomic DNA Analysis, version 7.1, December 2011). Slight modifications were introduced WGA products processed with the PCR-based labeling approaches. Here, the hybridization mix consisted of 5.0 µg of Cot1-DNA (Roche Diagnostics), 12 µl of 10x Blocking Reagent (Agilent Technologies), 60 µl of 2x Hi RPM Hybridization Buffer, 1% (v/v) of both Tween20 and Igepal and 19 µl of both test and reference DNA. For each hybridization 100 µl of the hybridization mix was applied on the array and hybridized at 65°C for 24 h. Following the hybridization, the slides were washed twice for 2:30 min in Oligo aCGH/ChIP-on-Chip Wash Buffer 1 (Agilent Technologies) at room temperature, twice for 30 sec min in Oligo aCGH/ChIP-on-Chip Wash Buffer 2 (Agilent Technologies) at 37°C. Washed slides were immersed in acetonitrile to remove all remaining traces of the wash buffers. Finally, slides were scanned using an Agilent Microarray Scanner Type C.

Processing and analysis of the aCGH data

Microarray TIFF image files were processed with the Agilent Genomic Feature Extraction Software (version 10.7). The resulting text files were imported and analyzed with Agilent Genomic Workbench Software (version 6.5 Lite). Aberrant regions were recognized using ADM-2 algorithm with threshold set to 6.5. Centralization algorithm was set to a threshold of 6.0 and bin size of 10. To avoid false positive calls, aberration filters were applied to define the minimum log2 ratio (0.25 for unamplified DNA or WGA products from freshly picked cell and 0.3 for WGA product from cell stained for vimentin or cytokeratins) and the minimum number of probes in an aberrant interval (5 and 25 for unamplified DNA and WGA products, respectively). The microarray data presented in this manuscript has been deposited at Gene Expression Omnibus (GEO) database (accession number: GSE52366).

Identification of minimal regions of aberration (MRAs)

MRA refers to the smallest region of genomic copy number alteration overlapping across an analyzed set of aCGH profiles.

Statistical analysis

Identification accuracy was assessed using separate ROC curves for gains and losses. To this end segmentation profiles obtained by the ADM2 algorithm were binarized according to threshold values (Thr): for gains, values ≥ Thr were set to unity and 0 otherwise, for losses: values ≤ –Thr were set to unity and 0 otherwise. Reference profiles were binarized according to a fixed threshold of Thr  =  0.25, while test profiles were binarized for different values of Thr. ROC curves were obtained by comparing a reference binarization to test binarizations for different positive threshold values. For clustering, segmentation profiles were discretized according three levels, i.e. values ≥ Thr were labeled as gains by 1, values ≤ –Thr were labeled as losses by –1, and all other values were labeled by zero. Hierarchical clustering was performed using Euclidean distance and agglomeration by complete linkage. Analyses were done using R [R: A Language and Environment for Statistical Computing, R Core Team,R Foundation for Statistical Computing, Vienna, Austria, 2013, url  =  http://www.R-project.org].

Optimization and validation of the single-cell aCGH assay

First, we developed two customized PCR-based approaches for labeling Ampli1™ single-cell WGA products and compared them with the recently published protocol using random-primed isothermal incorporation of fluorescent dyes (RP labeling) [27]. The first PCR-based protocol employed a dye-conjugated universal primer (PCR-T1) that binds to the adapter sequence utilized in the primary WGA. The second approach was based on the incorporation of dye-conjugated dNTPs (PCR-T2) during a short re-amplification step. Both techniques utilized the same universal primer as incorporated in the Ampli1™ WGA procedure, thereby enabling comprehensive labeling of the complete representation of single-cell DNA. Both labeling approaches were tested for their accuracy and reliability using PT1590 esophageal cancer cell line cells [28]. We generated three single-cell and two WGA products of cells pooled from PT1590 cell line. Paired samples (single cell and cell pool) for both PT1590 cell line and healthy donor were labeled in parallel with both techniques and subsequently hybridized on Agilent SurePrint G3 Human 4×180k arrays. As benchmark, we used aCGH profiles obtained with unamplified genomic DNA. Samples processed with PCR-T1 provided consistently higher dye incorporation rates, whereas PCR-T2 generated higher DNA yields (Table S1 in the Text S1). More importantly, signal intensities and signal-to-noise ratios obtained for both dyes (Cy5-red/test and Cy3-green/reference), were higher for samples processed with PCR-T2. The direct comparison of PCR-T1 and PCR-T2 revealed striking disparities between single-cell aCGH profiles of the same individual cell (Figure S1A-C). When compared to bulk DNA we found that multiple genomic alterations could be detected only if labeling PCR-T2 was applied (Figure S1A-C). The better fit of PCR-T2 aberrations with results from bulk genomic DNA was quantified by subsequent ROC-analysis (Figure S1D). In addition, repeated hybridization of single-cell WGA products processed with PCR-T2 gave highly reproducible results (Figure S2).

To further improve the performance of the assay we searched for the best performing reference sample. To this end we compared two sample types: (i) WGA products generated from cell pools and (ii) samples obtained by pooling four single-cell WGA products. Direct comparison revealed that derivative log2 ratio spread (DLRS) values, a measurement of hybridization noise, were consistently higher when WGA products from cell pools were used as reference (Figure S2). As a consequence all subsequent experiments included a pool of four single cell WGA products originating from one healthy individual as a reference sample.

Next we compared the performance of PCR-T2 with the recently published RP labeling procedure [27]. We started with using freshly picked single cells of OE-19 esophageal cancer cells. Irrespectively of the labeling approach used, two aberrations of only 0.1 Mb in size (one homozygous deletion and one amplification) could be reliably and reproducibly detected in all OE-19 single cells (Figure 2A-B) and the aCGH profiles of single OE-19 cells were highly concordant with the results of unamplified DNA (Figure 2C-D). In addition, we tested both labeling techniques when applied on re-amplified WGA products of OE-19 cells. Independently of the labeling technique applied, aCGH profiles of both re-amplified and primary single-cell WGA products as well as with unamplified DNA showed high level of concordance (Figure S3A-B). This was also true for single cell WGA products obtained from healthy donors displaying balanced profiles (Figure S3C). Therefore, the second round of amplification did not introduce a significant amplification bias. Since fixation and staining protocols used for detection of DCCs may affect the quality of single cell DNA [30], [31], we also compared PCR-T2 and the RP labeling for clinical samples. Here, we could show that when applied to immunostained cells PCR-T2 is more robust and less susceptible to generate technical bias than it is the case with RP labeling. Therefore, PCR-T2 became the method of choice for processing the Ampli1™ WGA product in all subsequent experiments (Figure S6 and Figure S7; text S1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. PCR-based PCR-T2 labeling technique vs. RP labeling.

A) Chromosome specific aCGH profiles of chromosome 8. B) Chromosome specific aCGH profiles of 17. Each panel represents aCGH profiles generated with unamplified and single-cell gDNA (PCR-T2 labeling) – left and middle plot, respectively. The right plot of each panel represents magnified graphical overview of genes within loci recognized as aberrant. C) ROC-curves (corresponding to profiles presented in panel A) depicting the accuracy of single cell aCGH assay for PCR-T2 or RP labeling. The array CGH profile generated using unamplified gDNA of OE-19 cells was taken as reference for the comparison. ROC analysis was performed on a genome-wide basis. D) Genome wide aCGH profiles of OE-19 cells generated using unamplified gDNA (upper panel) and a single-cell WGA product labeled with PCR-T2 (middle panel) or RP labeling approach.

https://doi.org/10.1371/journal.pone.0085907.g002

Detection of cell-to-cell heterogeneity in single cells

We evaluated the capability of our assay based on PCR-T2 labeling to assess genomic heterogeneity at the single cell level. We analyzed three single-cell WGA products, cell-pool WGA product and corresponding unamplified DNA of PT-1590 cell line. PT-1590 cells were previously shown to harbor amplification of ERBB2 gene [11]. Surprisingly, we found by aCGH analysis that one out of three analyzed PT1590 cells did not display the amplification of ERBB2 (Figure 3, Figure S4). To determine whether this finding was a technical artifact or reflecting true cellular heterogeneity, we performed FISH analysis. We found that about 10% of the PT1590 cells indeed displayed a balanced copy number ratio at the ERBB2 locus (Figure 3B). CNAs found in single cells only may thus represent true-positive events that escape detection in the bulk DNA where they are masked by the heterogeneity of the cell population. (Figure 3B, Figure S4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Assessment of the cellular heterogeneity.

A) Vertical aCGH profiles of chromosome 17 for unamplified gDNA and single-cell WGA products of PT1590 cells. Red arrow indicate ERBB2 locus. Note, one single cell (cell 3) shows a balanced profile at this site. Black arrows indicate private genetic alteration detected only in individual cells. B) Representative FISH images of PT1590 cells. Red signals indicate ERBB2 locus and green CEP17. White arrows label cells with balanced copy number of ERBB2 vs. CEP17. Yellow arrowhead shows cells with high-level amplification of the ERBB2 locus.

https://doi.org/10.1371/journal.pone.0085907.g003

Quantitative assessment of CNAs in single cells

To further challenge our method, we asked whether we could correctly quantify copy number changes, i.e. assess the level of gene amplification in single cells. For this, we analyzed four different breast cancer cell lines (MDA-MB-453, MDA-MB-361, SK-BR-3 and BT-474) each representing different copy number of the ERBB2 locus (FISH ERBB2 to CEP17 ratios equal 2.15, 3.60, 4.42, 5.79, respectively). For each cell line both unamplified DNA and single-cell WGA products were analyzed by aCGH giving highly concordant profiles (Figure 4A). The average log2 ratios of probes representing the ERBB2 locus in single-cell aCGH experiments closely correlated with corresponding log2 values obtained for unamplified DNA and FISH ratios of ERBB2 (Pearson’s correlation coefficient 0.94 and 0.97, respectively, Figure 4 B-C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Quantitative assessment of copy number changes in tumor cells by single cell aCGH.

A) Vertical aCGH profiles of chromosome 17 of four breast cancer cell lines with increasing copy number of ERBB2 locus (MDA-MB-453, MDA-MB-361, SKBR3 and BT474) generated using unamplified DNA (upper row) or single-cell WGA products (lower row). Red brackets indicate the position of the ERBB2 locus. Corresponding FISH ratios (ERBB2 vs. CEP17) of all cell lines are indicated in blue brackets. B) Correlation of average log2 values of ERBB2 specific probes obtained in single-cell aCGH experiments (Y-axis) vs. corresponding values obtained with unamplified DNA (X-axis). DNA samples from four breast cancer cell lines (MDA-MB-453, MDA-MB-361, SKBR3, BT474) have been included in the analysis. Pearson’s correlation coefficient 0.94. C) Correlation of average log2 values specific for ERBB2 locus obtained in single-cell aCGH experiments (Y-axis) vs. FISH ratios (ERBB2/CEP17) calculated for four breast cancer cell lines: MDA-MB-453, MDA-MB-361, SKBR3, BT474. Pearson’s correlation coefficient 0.97.

https://doi.org/10.1371/journal.pone.0085907.g004

Application to single DCCs

To test the assay on archival patient samples, we analyzed DCCs and corresponding metastatic tissue of an individual breast cancer patient. At the time of primary tumor resection (July 1998) this patient was diagnosed with stage IV breast cancer (PT4b, N1bii, pM1, G3, ER+, PR+, HER2-) with metastases detected in bones and lymph nodes and was subsequently treated with high dose chemotherapy (Figure 5A). During the course of treatment bone marrow was aspirated four times and screened for the presence of DCCs (Figure 5B). The DCC count was steadily decreasing throughout the treatment (Figure 5B), however it never reached zero. From each time point we analyzed two DCCs using our aCGH method. Additionally, DNA was collected from the primary tumor and a lymph node metastasis, amplified with Ampli1™ WGA and subjected to aCGH. Subsequent aCGH analysis revealed multiple highly penetrant alterations among DCCs and corresponding tumor tissue samples (Figure S5). Systematic analysis of CNAs revealed 43 minimal regions of copy number alteration (MRAs), nine of which were shared by samples from the primary lesion, the lymph node metastasis and at least one of the DCCs (Figure 5C, Table S2 in the Text S1). Among the shared alterations, we identified five core MRAs shared by all samples (Figure 5E) – gains on 17q and 8q and losses on chromosomes 6, 12 and 13 – harboring multiple oncogenes and tumor suppressor genes, i.e. MYC, MTDH, RB1 and CDKN2A, that were previously associated with the progression of breast cancer [32], [33]. Additional three MRAs were shared between the metastatic tissue and all DCCs but not with the PT (Figure 5E). Hierarchical clustering revealed that 7 out of 8 DCCs were genetically different from the primary tumor and displayed more similarities to the lymph node metastasis (Figure 5D). The detailed inspection revealed substantial heterogeneity for various CNAs in the genomes of DCCs (Figure S5). One of the DCCs collected at the time point of the third bone marrow sampling (DCC #1, time point 3) lacked multiple aberrations shared by the other sister cells (Figure S5). This cell showed the highest genomic similarity to the primary tumor (Figures 5D), however it harbored fewer genomic alterations than the primary lesion (Table S3 in the Text S1). Hierarchical clustering confirmed the genomic divergence of cell DCC #1 from time point 3 in comparison with the remaining DCCs and the metastatic tissue. Lack of otherwise recurrent CNAs was also observed in DCC #2 from time point 4. In this case, however, the cell harbored additional unique copy number changes (both gains and losses), indicating the existence of subclones within the DCC population (Figure S5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Molecular findings in individual DCCs over the course of systemic treatment.

A) Chemotherapy regime: first patient was subjected to three cycles of 75 mg/m² Taxotere® (T) and 50 mg/m² Doxorubicin (D) in three week intervals. Subsequently followed two cycles of high dose chemotherapy treatment with an intermediate interval of 4-6 weeks. In the first cycle 500 mg/m² of Vepesid (V), 4000 mg/m² of Isofamid (I) and 500 mg/m² of Carboplatin (C) was administered. The last cycle consisted of 1500 mg/m2 of Cyclophosphamid (Cy) and 200 mg/m2 of Thiotepa (T). Both cycles of high dose chemotherapeutic treatment were accompanied by addition of granulocyte colony-stimulating factor (G-CSF), and autologous transplant of peripheral blood stem cells (PBSCs). B) The course and outcome of bone marrow sampling. DCC count indicated the number of identified DCCs in 1.0×10⁶ mononuclear cells. C) Venn diagram depicting distribution of MRAs (gains and losses) across three types of clinical samples. D) Hierarchical clustering (distance: Euclidian; linkage: average) of samples including DCCs, the primary tumor and a metastatic lesion. E) Table depicting core MRAs that were found in all DCCs, the primary tumor and the metastatic lesion or in the metastatic compartments (DCCs and the lymph metastasis) only.

https://doi.org/10.1371/journal.pone.0085907.g005