SNP arrays give information for both copy number and b-allele frequency.  With most copy number analysis, median recentering is selected by default.  This is appropriate when a sample is close to diploid.  However, when the overall ploidy strays due to extensive gains or losses, as is typical in many cancer samples, median recentering can lead to false calling.

Therefore, when evaluating an individual cancer sample, there are a few red flags to watch out for:

- Are there extensive copy neutral (baseline) areas with allelic imbalance (purple)?

- Are there extensive areas of copy number loss (red) without any corresponding allele change (purple allelic imbalance or yellow loss of heterozygosity)?

-  Are there extensive areas of gain (blue) without any corresponding allelic imbalance (purple)?

Here is one such example:

As you can see in this example, there are extensive “gains” of chromosomes 2, 5, 7, 9, 10, 12, 19, 20,and 22, with no corresponding allelic imbalance, while the copy number loss regions all show appropriate loss of heterozygosity.  If we look at the whole genome view, we can get further information:

The top panel shows the copy number calling (note this is an Affymetrix MIP array, so everything is shown in a linear copy number scale).  We can see many areas called as loss (~-0.4) and gain (~0.4), but almost nothing is called at the baseline state (0).  When we look at the bottom B-allele frequency panel, we can see many regions of balanced heterozygosity, as noted by the typical 3 band pattern (spots at 0, 0.5 and 1.0).  These areas include chromosomes 2, 5, 7, 9, 10, 12, 19, 20,and 22; all of the areas that are also marked as “gained”.  Chromosome 14 shows a typical 4 band allelic imbalance pattern, but is marked as high amplitude gain (~1.2).

We can recenter this sample to the balanced regions of heterozygosity identified above, thus defining these chromosomes as diploid.  Within Nexus Copy Number software, this is done by:

1. Adding the factor “Diploid Regions” on the data set tab

2. Naming the diploid regions in the sample (chr2,chr5,chr7, chr9, chr10, chr12, chr19, chr20, chr22)

3. Adjust the settings to Recenter: Diploid Regions

4. Highlight the selected sample and Reset

5. Re-process the sample with the new settings by selecting View

Once recentered, this essentially shifts the top copy number panel while maintaining the identical lower B-allele frequency panel, as shown in the full genome view below:

Now this sample looks as expected: we see areas of copy number loss in the top panel are associated with loss of heterozygosity in the bottom panel for chromosomes 1p, 3, 4, 6, 8, 11, 13, 15, 16, 17, and 21.  We see copy number gain on chromosome 14 is associated with allelic imbalance on the bottom panel.  Furthermore a baseline diploid state is associated with a normal 3 band heterozygous pattern for chromosomes 2, 5, 7, 9, 10, 12, 19, 20, and 22.  This is reiterated in the summary view below:

As you can see, if we had not recentered this sample, we would have had extensive false positive calling.  Again, default median recentering is appropriate in samples that are close to diploid, and it is always a good place to start.  For a sample like the one above, where the actual ploidy is closer to 1.5, further recentering to specified regions may be required.  However, paying attention to the red flags outlined above will help you to recenter the sample and should yield positive results for your downstream analysis.

Drilling down to the whole genome view can also be useful for estimating the percentage of aberrant cells in a sample, or identifying potential areas of chromothripsis.

In the example shown above, we can clearly see several regions of copy number loss (red regions in middle logR panel), which are associated with appropriate loss of heterozygosity (yellow regions in the bottom B-allele frequency panel).  These regions alternate with copy number neutral regions (uncolored baseline in logR panel, uncolored 3 band pattern in bottom B-allele frequency panel) and regions of copy gain (blue regions in logR panel, purple 4 band allelic imbalance pattern in B-allele frequency panel).

Even from a whole genome overview, we can easily identify the alternating copy states and altered heterozygosity of chromosome 6.  This is just one of the many ways BioDiscovery Nexus Copy Number software can be used to identify some of the frequent aberrations found in cancer.

You may also like to read the article here on how to estimate the percentage of aberrant cells in a tumor/normal cell mixture or the article here on how to paired analysis to identify somatic changes in tumor/normal pairs.

Looking at a recent paper (http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2919738/), cost-effective sequencing techniques can be used to detect large copy number changes. The low-depth techniques have low resolution (e.g., in the paper 0.04x coverage resolved CNVs of 15kb), but are cost-competitive. At increased, but still low depth (e.g., 0.3x), the resolution and cost compares favorably to some aCGH platforms, but the technique has some limitations. Obviously, focal events smaller than the window will not be resolved. Perhaps less apparent, the technique cannot be used to determine the zygosity of the input DNA. For both constitutional and cancer cases, this limitation makes the technique poorly suited to finding copy-neutral LOH (loss of heterozygosity) regions. However, unlike aCGH, the technique can be used to find the ploidy of the sample by summing the number of reads per chromosome and dividing by the chromosome length, then relative read depths can be used to infer copy number at points along the genome.

Simple CGH arrays also have the problem of determining zygosity, but SNP array technology (such as the Affy CytoScan HD and OncoScan products, Illumina OMNI arrays, and Agilent CGH+SNP) solves it. By using SNP arrays, zygosity of the sample within various regions of the genome can be imputed through the allelic balance (commonly reported as the B-allele frequency). Interpretation of SNP array data is further complicated, especially in cancer samples, due to sample heterogeneity; but in most cases the underlying aberrations can be detected with a high degree of confidence, along with copy-neutral LOH.

The limitations of low-depth sequencing are also circumvented, easily enough, at a sufficient depth of coverage that allows accurate base calling. Here, not only can zygosity be deduced, but small structural variants as well as somatic mutations can be detected, creating a more robust picture than arrays alone. Though, problems still exist. At this level, the cost and analysis of sequencing data may become problematic, particularly if the primary goal of genome-wide sequencing is for copy number detection (as opposed to a by-product result from some other objective). For instance, techniques to detect LOH from sequencing (including exome sequencing) have been developed, but require a depth of coverage which significantly increases the cost. The Exome-CNV paper tested the technique to find copy number and LOH using about a 40x depth of coverage on melanoma samples (http://bioinformatics.oxfordjournals.org/content/early/2011/08/09/bioinformatics.btr462.full.pdf).

Archival FFPE samples pose challenges for both arrays and NGS methods. The low-input DNA, sample heterogeneity (since archival FFPE are commonly cancer samples), and sample degradation all make analyzing FFPE samples difficult. The current wisdom is that sequencing methods ought to have an advantage here, since they don’t require DNA amplification, and the degraded DNA can be sequenced directly. But at least one array method, using molecular inverted probes (MIP), promises to work at least as well on FFPE samples, and perhaps better (http://www.sciencedirect.com/science/article/pii/S2210776212001627)

From where we stand today, the way to arrive at a clear answer as to whether sequencing or arrays makes sense for copy number analysis is to frame the question in terms of the nature of the sample, analysis objectives, and cost constraints. You may conclude, for example, that low-depth sequencing is a cost-effective and robust alternative to a low-resolution aCGH platform. But, if you work in cancer samples, or constitutional samples requiring LOH detection, depending on your goals you may decide the best approach is a high-density SNP array combined with targeted re-sequencing. In any event, the technology is moving at a rapid pace and you can be reasonably assured that whatever is the right answer ‘today’ is likely to change.

Please comment and add your thoughts — there’s so much happening at the moment that this likely (a) contains errors, (b) is out of date, and/or (c) misses some key points.

Now we’ve made it even easier to obtain additional genomes by providing currently available annotation files as zip archive downloads on our Downloads page. If you don’t see your organism here, don’t worry, you can easily create these annotation files yourself and add to Nexus. Check out the Appendix on “Adding a New Organism” in the User Manual. If you still need help, contact BioDiscovery Technical Support and we will help in creating the necessary files. As we generate new annotation files, we will add them to the Downloads page.

Here are some of the organisms currently available for Nexus Copy Number and Nexus Solo: cow, cat, dog, monkey, horse, platypus, chimpanzee, orangutan, rat, pig, C. elegans, cichlid, zebrafish, medaka, corn, chicken, yeast, Streptococcus, P. falciparum, and C. raciborskii.

Both somatic copy number alterations (CNAs) and germline copy number variants (CNVs) that are prevalent in healthy individuals can appear as recurrent changes in comparative genomic hybridization (CGH) analyses of tumors. This represent a challenge during the analysis since researchers are keen to identify regions that are frequently gained or lost in patients with a particular cancer.

What strategies can be implemented to distinguish between CNA and CNV changes?

One possible solution would be to collect data from healthy and tumor tissue samples from the same individual and run a matched paired analysis.

When performing matched paired analysis, generally a tumor sample is compared to its matched normal sample to identify only those events in the tumor that are not present in the normal tissue. Two files are loaded for each sample: data from the tumor tissue and the data from the matched normal tissue. (i.e. normal tissue from the same patient). The two files are combined into one result after the values in the normal sample are subtracted from the corresponding probes in the tumor sample. In the resulting profile, the Log ratios and B-Allele Frequencies (BAF) are shown after subtraction.

In BioDiscovery Nexus Copy Number software, researchers can perform matched paired analysis by loading both the tumor and normal data for each sample. After the values from the normal sample are subtracted from the tumor sample, these corrected probe level values are used to perform segmentation and calling. To see how to load tumor/normal paired samples in Nexus, please view the How To Load and Process Data for Matched Paired Analysis Guide from our How To Guides collection.

Nexus Copy Number (copy number and allelic events from microarray and NGS)

Nexus Expression (miRNA and mRNA expression analysis)

ImaGene (image analysis)

Affymetrix CEL files (intensity values) need to be normalized to generate logRatios before proceeding to copy number analysis.  A normal reference file generated from the HapMap pooled samples is used for this purpose.  Depending on the Affymetrix array types, different reference files are generated for 500K, SNP 6.0, and CytoScan HD.  Each CEL file from the test samples is used against the corresponding reference file to get logRatios for all the probes on the array.  A systematic correction process follows to straighten up the systematic waviness pattern of the data distribution, which results mainly from the G-C contents of the probes and the samples, as well as from other factors.  The data is finally analyzed to get the copy number calls using the default SNP-FASST2 segmentation algorithm.

LogRatio data, such as Agilent Feature Extraction results files or Illumina final report files, are directly processed by FASST2 (for aCGH arrays) or SNP-FASST2 (for SNP arrays) to get copy number calls, usually preceded by the systematic correction step for possible data waviness.

A special case is Affymetrix OncoScan data files, Copy_Number.txt and Assays.txt.  The “Raw Data” are probe copy number state values, which are linear values rather than logRatios.  However, Nexus Copy Number processes the data similarly to the logRatios, i.e. systematic correction followed by copy number segmentation with SNP-FASST2.

As described above, the last step for the data analysis workflow is the copy number segmentation, which results in the copy number calls for the copy number segments.  What this step ultimately does is to decide whether a group of probes should be in a distinct segment or stay with the current segment of neighboring probes.  The default copy number segmentation algorithm in Nexus Copy Number is FASST2 (for aCGH arrays) or SNP-FASST2 (for SNP arrays), which is based on the popular Hidden Markov Model (HMM) algorithm.  However, unlike the conventional HMM, no integer copy number states (e.g. 0, 1, 2, 3, and 4) are used.  Instead, the states are defined by the copy number calling thresholds, which are based on logRatios (linear values in Affymetrix OncoScan Data) and can be easily adjusted by the end user.  The number of segments is determined by the Significance Threshold setting, which is equivalent to a P-value cut-off and is also adjustable to the user.  The probability that a group of probes belong to a certain copy number state is compared to this Significance Threshold and a new segment is generated if the Significant Threshold is surpassed.  The value of the new segment, median value from the logRatios (linear values in Affymetrix OncoScan Data) of the group of probes in the segment, is then compared to the calling thresholds (one copy gain, one copy loss, gain with two or more copies, and homozygous loss) to get the corresponding copy number calls.

In summary, there are three major steps in the data analysis workflow from “Raw Data” to copy number calls:

1. Probe intensities to logRatios
2. Systematic correction for data waviness
3. Copy number segmentation for copy number calls

And on behalf of everyone at BioDiscovery, Happy Holidays!