Validation of Copy Number Analysis

Application of the WGSA array to determine genomic copy number will only be possible if the fluorescent intensity from each feature shows a dosage response to variations in copy number. This was tested in two ways, by applying samples with varying numbers of the X-chromosome and by spiking a series of identical DNA aliquots with varying concentrations of PCR product to increase the copy number of 42 SNPs from twofold (control sample) to 1000-fold.

In the X-chromosome copy number experiment, we used the average value from males to represent the case of 1X and the average value of females to represent 2X; we also collected data from 3X-, 4X-, and 5X-containing cell lines. Using (I) to indicate chip intensity, the dosage-response assumption can be written I_a C_ab × I_b, where I_a is the intensity for a region with copy number a, I_b is the intensity on the same region with copy number b, and C_ab is a constant determined by a and b. (see Methods) can be viewed as an approximation of log intensity (all references to the log function refer to the natural log e unless otherwise stated). Therefore, if the assumption is true, a log transformation leads to _{a b} + _ab. The results using average female values as a baseline (173 SNPs on the X-chromosome) give estimated intensity ratios of 0.584, 1, 1.484, 1.822, and 2.243, equating to copy numbers of 1.17, 2, 2.97, 3.64, and 4.49, with a correlation of 0.981, 1, 0.943, 0.935, and 0.939 for onefold, twofold, threefold, fourfold, and fivefold actual copy numbers, respectively (baselined on female values with a copy number of two; therefore, the estimated copy number equals 2 × the estimated intensity ratio). When plotting estimated copy number against actual copy number, a straight-line response is seen (R² = 0.9976). The slope of the relationship, 0.83, differs from the ideal value of 1 but is similar to data from spotted array CGH experiments (Pinkel et al. 1998; Pollack et al. 1999).

The response of the copy number estimation at higher copy number values was simulated by spiking aliquots of the same DNA (NCI-BL2126) with PCR products for 42 SNPs selected because of their high call rates. The copy number was increased from 2 (control) by 2, 5, 10, 25, 50, 75, 100, 250, 500, 750, and 1000 copies. Of the 42 SNPs tested, 40 gave an increase in fluorescence in response to increased copy number up to and including the 1000-fold spike with a correlation between log intensity and log copy number of 0.92 (Fig. 1). Two of the SNPs showed no increase in fluorescence in response to increasing copy number; this subset therefore indicates that only a small fraction of SNPs on the array would fail to report copy number change.

Open in a separate window

Figure 1

Plot of log(copy number) against log(intensity) for the spiking experiment in which 18 aliquots of the same DNA were spiked with varying concentrations of 42 SNPs from twofold up to 1000-fold. The black dots indicate the results for the individual SNPs across the 12 spiking concentrations (spiked with an extra 0, 2, 5, 10, 25, 50, 75, 100, 250, 500, 750, and 1000 copies); the green dots and line show the mean for all 42 SNPs. Two SNPs did not report increased fluorescence with increased copy number, and these are highlighted in red.

Copy Number Changes in Cancer Cell Lines

Having confirmed the utility of the oligonucleotide array for detecting increases in copy number, we analyzed a set of 20 cancer cell lines. The analysis of the fluorescence data from the tumor samples identified a total of 14 putative high-level amplifications in which a minimum of three consecutive SNPs reported ratios >2.5 (equivalent to a copy number of 5, also visible in the unprocessed fluorescence data). Of these, 12 loci were tested, and each was shown to have a copy number in excess of 5 by qPCR (Table 1). An example of genomic amplification of the c-MYC locus in COR-L96-CAR can be seen in Figure 2A; the profile of this amplification was also obtained by qPCR (SYBR Green) using amplicons designed to SNPs from the array; the comparison with the data from the p501 array is shown in Figure 3. A total of 10 putative homozygous deletions were also identified (again reported by three consecutive SNPs and visible in the unprocessed fluorescence data), all of which were assessed and confirmed by conventional PCR (Table 1). Figure 2B shows an example of a homozygous deletion of the p16/INK4 locus in LB1047-RCC.

Open in a separate window

Open in a separate window

Open in a separate window

Open in a separate window

Figure 2

Examples of amplification and deletions together with genotyping data generated by the p501 array compared with BAC CGH data. (Top panel) The fluorescence ratio plot from the p501 array showing the smoothed fluorescent intensity data of the sample divided by the figure from the reference sample. (Second panel) The p-value plot for the individual SNPs calculated by comparison to the mean and standard deviation of 29 normal DNAs with deletions represented by red lines and amplifications in green. (Third and fourth panels) The genotypes for the tumor and matched normal samples, respectively; homozygous SNPs are represented by a red line below the center point whereas heterozygous markers are represented by a green line above the center point. (Bottom panel in C and D only) The results for BAC-array-based CGH (where available). (A) Genomic amplification at the C-MYC locus (arrow) on Chromosome 8 in the prostate line COR-L96-CAR. (B) A homozygous deletion (arrow) at the p16/INK4 locus on Chromosome 9 in the renal cell carcinoma cell line LB1047. (C) Chromosome 18 from the breast cancer cell line HCC1937. (i) This chromosome has a copy number of 2 (intensity ratio of 1) from the pter until 18q12.1. (ii) The copy number drops to 1 (intensity ratio of 0.5) from 18q12.1 until the 18qter, although the genotyping data indicate that this line is heterozygous across the full length of Chromosome 18. (D) Chromosome 5 from the small cell lung cancer line NCI-H209. This chromosome shows a complex pattern with partial amplification of the p-arm (i), followed by a drop in fluorescent intensity to 0.5 with corresponding LOH determined from the genotyping data (ii), until 5q23.2 (iii), where the intensity ratio recovers to 1 (copy number of 2); however, this region still represents LOH as determined by the genotyping data and therefore represents duplication of a single parental chromosome. There is also a homozygous deletion at 5q14.3 (arrow). This was not detected with the BAC array, as the array does not have a clone that covers this region.

Open in a separate window

Figure 3

Comparison of copy number estimation for the amplification of the c-MYC proto-oncogene in COR-L96-CAR using the p501 array (A; where copy number equals two times the intensity ratio) and qPCR (B).

The p501 array also detected regions with more subtle copy number changes, namely, amplification events to 3 copies or a reduction in copy number from 2 to 1. For two of the lines (HCC1937 and NCI-H209), we were able to compare the p501 array results with data from BAC-array-based CGH. Of the 44 autosomal chromosomes from these two lines, 34 showed consistent copy number patterns in the two analysis protocols when compared for low copy number changes over extended regions. Chromosomes 17, 19, 20, and 22 gave poor resolution in both samples using the p501 array, thereby accounting for 80% of the inconsistencies. These chromosomes had the lowest SNP density with 1 in 0.71, 1.25, 0.77, and 1.25 Mb, respectively, compared with the average density for the rest of the genome of 1 in 0.44 Mb. For the remaining 18 autosomal chromosomes, the data from the p501 array tended to show a greater variability; however, the underlying pattern was discernable and consistent with that of BAC-array-based CGH (Fig. 2C,D).

Array Design

The p501 array contains allele-specific hybridization probes complementary to 8473 SNPs predicted to be in the fraction of the genome represented by 400–800-bp fragments of XbaI-digested genomic DNA. Oligonucleotides corresponding to these SNPs were synthesized used a photolithographic methodology, and each SNP is represented by 56 different oligonucleotide probes. The oligonucleotides are 25-mer sequences that interrogate the site of polymorphism on both the sense and antisense strands and contain both a perfect match (PM) and mismatch (MM) sequence to allow signal-to-noise measurements. Additional oligonucleotides that are offset from the site of polymorphism by 1–4 nt are synthesized to allow for data redundancy and to maximize genotype accuracy.

Target Preparation

WGSA relies on genomic representation to reduce the complexity of the genome by ~98% and thereby improve hybridization kinetics (Lucito et al. 1998; Kennedy et al. 2003). The sample DNA was digested to completion using Xba1 prior to ligation of adaptors and amplification of the ligation products using an adaptor-specific primer. The reagents and protocol used were taken from Kennedy et al. (2003), except 400 ng of sample DNA was used instead of 250 ng. Because of the increased input DNA all reactions volumes involved in the digestion, linker ligation and amplification stages were increased by 60%, thereby maintaining the reaction conditions. After PCR, the samples were concentrated using a QIAGEN minielute PCR purification kit (catalog #28006) and eluted in 30 μL of EB buffer, using one column per PCR, giving a final elution volume of 180 μL. This was reduced to 50 μL with a Microcon YM-30 filter (catalog #42410). The final concentrated DNA was quantified using a Hoeffer DyNA Quant capillary cuvette kit; 20 μg of the PCR product was fragment, reducing the average product size down to 50–150 bp.

Hybridization and Scanning

Hybridization and staining of the probe to the array together with scanning of the chip were carried out as in Kennedy et al. (2003). Basically, the fragmented DNA was labeled with biotin-N6-ddATP using terminal transferase before being mixed with hybridization solution and added to the p501 arrays for hybridization. After hybridization, the array was washed and stained using the Affymetrix Fluidics Station. The staining procedure was designed to amplify the signal from the annealed probe. The sample was first stained with straptavidin followed by treatment with biotinylated anti-streptavidin and finally Streptavidin R0-phycoerythrin conjugate. Scanning was carried out using the Agilent GeneArray Scanner.

Feature Extraction

The p501 array design uses 28 probe pairs for each SNP, 14 for allele A and 14 for allele B. The 14 probe pairs for each allele are equally divided between the sense and antisense strands. A probe pair includes a perfect match cell and a mismatch cell. We use

as the basic measurement for any given SNP, where PM_i is the intensity of the perfect match cell of probe pair i and MM_i is the intensity of the mismatch cell of probe pair i. This value measures the average intensity difference between a perfect match and a mismatch on a log scale. The log transformation makes the distribution more Gaussian. After S is calculated for all the SNPs on a given chip, it is scaled to have a mean of zero. In other words:

j = 1,..., J are all the autosomal SNPs on the chip.

Copy Number Estimation

When estimating copy number, the output from the feature extraction was smoothed, first outliers were removed by taking the median of five SNPs—the test SNP together with the two flanking SNPs for each data point—and the mean of the five SNPs was then taken. This was then converted into copy number by calculating the fluorescence ratio with respect to the mean reading from a series of 29 normal DNAs. Therefore, if an SNP is from a diploid region of the test sample, it will give a reading of 1 when divided by the normal reference sample, representing a copy number of 2.

Significance Calculation

To estimate the significance of the copy number variation in the target cancer cell line, we compare it with a reference set consisting of 29 normal DNA samples. For any given SNP j, we assume that _j follows a Gaussian distribution; the mean and variance are estimated using the 29 normal reference samples.

where k = 1,..., K represents the normal reference set. Assuming the target cancer cell line has value on SNP j, the significance of the difference of from the normal reference distribution is measured by the p-value:

This probability indicates how likely the normal population will have values as extreme as the cancer cell line. The smaller it is, the more significant the difference between the cancer and the normal cell lines.

BAC Array CGH

BAC array CGH was carried out using a 1-Mb array containing ~4100 publicly available BACs. Array construction was based on a modification of the protocol by Hodgson et al. (2001). DNA was extracted from 15-h cultures using the QIAGEN REAL prep kit and amplified using degenerate oligonucleotide primers (5′-CCGACTCGAGNNNNNNATGTGG-3′). Each BAC was amplified under two different annealing temperatures (58°C and 60°C) to improve coverage. PCR products were pooled, purified (QiaQuick kits, QIAGEN), and lyophilized before being resuspended in 50% DMSO/water to a final concentration of ~300 ng/μL. A minimum of two replicates of each clone was printed on each slide.

DNA was labeled (random prime labeling kit, Invitrogen) and hybridized together with Cot-1 DNA for 48–72 h at 37°C. Slides were washed as described by Hodgson et al. (2001) and scanned using the Affymetrix428 microarray scanner. Analysis was carried out using the Genepixsoftware (Axon) with adjustments made to account for variations in labeling efficiency (print-tip-based data adjustment; Yang et al. 2001).

CA Repeat Genotyping

Genotyping was carried out using the 10-cM microsatellite marker set from ABI (LMS-MD10). The markers were amplified from 12 ng of DNA using ABI True Allele PCR premix(catalog #403061) in a total volume of 10 μL. Cycling was performed on a Kbiosystems “Duncan” thermocycler using a 40-cycle program of 94°C denaturation, 60°C annealing, and 72°C extension each for 30 sec, proceeded by a 10-min soak at 94°C to activate the Amplitaq Gold, with a final 10-min soak at 72°C for complete extension. Samples were prepared and loaded onto either an ABI 3700 or ABI 3100 DNA analyzer as per the manufacturer's instructions. The data were analyzed using the ABI Genescan (V5.1) and Genotyper (V3.6) software. Fragment sizing was standardized between runs by comparison of data from a control DNA (CEPH 1347-02) that was included in every sample set.

Confirmation of Homozygous Deletions

Homozygous deletions were confirmed by PCR. Primers were designed for SNPs reporting the deletion together with flanking SNPs. If possible, primers were designed either side of the SNP; otherwise, primers were placed as near to the SNP as possible. PCRs were carried out in duplicate for the test DNA together with a normal control. PCR products were visualized by gel electrophoresis.

The work of the Cancer Genome Project was supported by the Wellcome Trust. Other funding included Cancer Research UK (M.G.) and the Abramson Family Cancer Research Institute (B.L.W.). We thank the staff of the Cancer Genome Project for their contribution to this work.

The publication costs of this article were defrayed in part by payment of page charges. This article must therefore be hereby marked “advertisement” in accordance with 18 USC section 1734 solely to indicate this fact.