Genome-wide arrays: Quality criteria and platforms to be used in routine diagnostics

Authors

  • Joris R. Vermeesch,

    Corresponding author
    1. Laboratory for Cytogenetics and Genome Research, Centre for Human Genetics, KU Leuven, University Hospital Leuven, Leuven, Belgium
    • Laboratory for Cytogenetics and Genome Research, Centre for Human Genetics, KU Leuven, UZ Leuven, Campus Gasthuisberg, Herestraat 49, B-3000 Leuven, Belgium
    Search for more papers by this author
  • Paul D. Brady,

    1. Laboratory for Cytogenetics and Genome Research, Centre for Human Genetics, KU Leuven, University Hospital Leuven, Leuven, Belgium
    Search for more papers by this author
  • Damien Sanlaville,

    1. HCL, Cytogenetics Department and INSERM U1028, CNRS UMR5292, TIGER Team, Lyon Neuroscience Research Centre, Lyon, France
    Search for more papers by this author
  • Klaas Kok,

    1. Department of Genetics, University Medical Centre Groningen, University of Groningen, Groningen, The Netherlands
    Search for more papers by this author
  • Rosalind J. Hastings

    1. CEQA and UK NEQAS for Clinical Cytogenetics, Women's Centre, Oxford University Hospitals NHS Trust, Oxford, United Kingdom
    Search for more papers by this author

  • For the Focus on CNV Detection with Diagnostic Arrays

Abstract

Whole-genome analysis using genome-wide arrays, also called “genomic arrays,” “microarrays,” or “arrays,” has become the first-tier diagnostic test for patients with developmental abnormalities and/or intellectual disabilities. In addition to constitutional anomalies, genomic arrays are also used to diagnose acquired disorders. Despite the rapid implementation of these technologies in diagnostic laboratories, external quality control schemes (such as CEQA, EMQN, UK NEQAS, and the USA QA scheme CAP) and interlaboratory comparisons show that there are huge differences in quality, interpretation, and reporting among laboratories. We offer guidance to laboratories to help assure the quality of array experiments and to standardize minimum detection resolution, and we also provide guidelines to standardize interpretation and reporting. Hum Mutat 33:906–915, 2012. © 2012 Wiley Periodicals, Inc.

Introduction

Following the discovery that specific developmental disorders are associated with chromosomal abnormalities, genome-wide analysis for chromosomal rearrangements and imbalances became a routine test in the genetic diagnosis of patients with developmental disorders. For nearly five decades, conventional karyotyping was the only method to perform such an analysis, although over this period cytogenetics has seen some major improvements. In the 1960s, only whole-chromosomal imbalances or large segmental rearrangements could be detected. With the invention of chromosomal banding techniques, the resolution increased dramatically, and it became possible to associate smaller imbalances with novel developmental disorders. The introduction of improved banding resolution as well as several novel banding techniques has led to different levels of quality in the diagnoses made by different laboratories. This observation led to the development of guidelines and quality criteria for laboratories (e.g., European Cytogenetic Guidelines, 2006; Association for Clinical Cytogenetics Professional Guidelines, 2007, 2009). Although conventional karyotyping has been the gold standard for many years, this technology has several limitations: (1) the resolution is limited to the level of the banding quality; (2) it requires short-term culture of the patient sample, which is time-consuming and can lead to selective advantage of specific cell lines; and (3) the technical skills required for karyotyping are substantial and the analyses are time-consuming.

With the development of genome-wide arrays, these limitations have been largely overcome. DNA is usually isolated from a blood sample, but can also be obtained from other cell types, and the analysis can be performed in less than 5 days. Arrays enable the detection of genomic gains and losses with unprecedented resolution, and allow for a higher degree of automation. Most importantly, the use of genome-wide arrays has significantly improved the diagnostic yield over conventional karyotyping in patients with developmental delay, intellectual disability, multiple congenital anomalies, and autism [Menten et al., 2006; Stankiewicz and Beaudet, 2007]. This led to an international consensus statement that genome-wide arrays should be used in the diagnostic workup of such patients [Miller et al., 2010] and, as a consequence, more diagnostic laboratories are now introducing genomic microarrays as the first-tier test or as a supplementary test.

Although there is consensus that microarrays should be introduced, individual laboratories are confronted with many practical questions on how to implement this novel technology. Over the last few years, many different array platforms have been offered commercially by a plethora of providers. Each platform is different and the underlying methodology also differs, affecting the ability to detect different-sized imbalances. Confronted with these differences, it is not easy to choose between the options available. Each laboratory has the freedom to choose a platform, providing that it fulfils their minimum criteria. In addition, each laboratory is responsible for ensuring the analytical validity of the technique used before introducing it into its routine diagnostic service. To assist in the standardization of the tests applied in different genetic centers and to assure a similar quality of interpretation, the American College of Medical Genetics has recently developed standard guidelines [Kearney et al., 2011a, 2011b]. Our aim here is to complement and reinforce those efforts by presenting a European consensus on the recommended minimum analytical and reporting criteria, and to provide guidelines to assure the validity of the microarray analysis.

Considerations for Array Design

There is much debate about the optimal resolution for genomic arrays. Increasing the resolution has the advantage that even smaller pathogenic copy number variations (CNVs) can be detected, but as a consequence, the number of benign CNVs detected increases exponentially. In other words, while the analytical validity increases, there is a drop in the clinical validity of the information obtained. Cooper et al. (2011) compared the CNV burden in 15,767 patients and 8,329 controls. At a resolution of 500 kb, a 13.5% increase was observed in the number of CNVs detected in the patient population, strongly suggesting that this percentage of CNVs is also causal. At a resolution of 200 kb, there was only a small increase in the CNV detection burden in the patient and control populations. Based on this study, it is fair to state that arrays should at least aim to detect any imbalance larger than 500 kb. We therefore recommend using a resolution of 200 kb. Genomic backbone coverage below 200 kb may increase the number of CNVs of unknown significance to the extent that their clinical validity becomes too low.

Since microarrays are aimed at detecting imbalances across the whole genome, probes should be present at regular intervals across the genome. Any imbalance larger than the reported minimum resolution should be detectable by the platform. This minimum resolution does not mean that smaller imbalances cannot be causal. However, the interpretation of such a smaller imbalance is possible only if additional information on the gene content is available. To enable the detection of these potentially causative, smaller imbalances, enrichment of probes targeting dosage-sensitive genes known to result in genetic disease is recommended. Some experienced laboratories have also proposed “exon array” designs with probes enriched in exonic regions for detection of imbalances encompassing individual or multiple exons [Boone et al., 2010]. The International Standards for Cytogenomic Arrays (ISCA) Consortium (https://www.iscaconsortium.org/) provides guidance on the design of genomic arrays for diagnostic use, and many manufacturers provide “ISCA”-designed genomic arrays [Miller et al., 2010]. Furthermore, it is important to remember that an array analysis is not a sequence analysis screen and therefore it does not rule out the possibility that other unidentified mutations, including single-nucleotide polymorphisms (SNPs), indels, or short tandem repeats (STRs), or CNVs below the resolution of the respective platform are associated with a given phenotype.

Although the first array-based comparative genomic hybridization (CGH) methods were applied only for CNV detection, platforms designed for genome-wide, SNP detection are also applied. SNP arrays were originally designed to detect common SNPs (incidence > 1% in the population), and were mainly used in genotyping individuals for genome-wide association studies of many common multifactorial diseases [Klein et al., 2005; Sladek et al., 2007; Wellcome Trust Case Control Consortium, 2007]. In addition to SNP typing, these platforms can also be used to perform copy number analysis by comparing the patient target signal intensities with reference target signal intensities. Gains and losses of genomic regions can therefore be detected, as is the case for CGH arrays. SNP array manufacturers have added additional oligonucleotide targets not aimed at detecting SNPs, but at providing an even coverage of the genome for improved CNV detection. In addition, the genotype information can be used to analyze structural variation of the chromosomes. SNP identification is performed by calculating the B-allele frequency (BAF), a measurement of the presence of allele “B” and allele “A.” The BAF of the respective SNP targets also allows SNP arrays to detect copy-neutral loss of heterozygosity or absence of heterozygosity, uniparental disomy, and regions identical by descent. However, while SNP arrays will detect uniparental isodisomy, parental samples are required for the detection of uniparental heterodisomy.

Measurement of the BAF can also improve the sensitivity for low-grade mosaicisms [Conlin et al., 2010]. The outcome of a genotype analysis also offers an internal quality control check by ruling out potential sample mismatches due to nonpaternity or sample mix-ups. In addition, detecting regions of homozygosity may pinpoint loci harboring recessive disorders. de Leeuw et al. (2011) have reviewed CNV detection, genotyping, and quality control for the use of SNP arrays in constitutional and cancer diagnostics. Hence, the ability to measure SNPs can be used to improve the diagnosis.

Analytical Validity

The main quality parameters for genomic arrays fall into two categories: (1) specificity and sensitivity, and (2) internal quality control parameters. These are discussed in more detail below. Quality criteria for genomic arrays have been published [Fiegler et al., 2006; Vermeesch et al., 2005, 2007]. These recommendations were based on the use of bacterial artificial chromosome arrays, but the general principles apply to oligonucleotide and SNP arrays too. Brady and Vermeesch (in press) have recently reviewed quality criteria for using copy number and SNP arrays in diagnostics, and the American College of Medical Genetics has published guidelines on the design and performance expectations for genomic arrays [Kearney et al., 2011a].

Determination of Specificity and Sensitivity

Manufacturers of arrays generally provide the array platform, equipment to scan the arrays, and software to analyze the intensity ratios and identify CNVs. Each type of analysis software is different, often using either different statistical approaches and/or different default settings. In addition, each laboratory applies different experimental and analysis protocols. These differences in protocols and analysis software influence the sensitivity and resolution of a test.

Different segmentation algorithms are used to identify copy number variable genomic segments. Different calling algorithms may give different results for an array experiment, and this has been comprehensively discussed [Lai et al., 2005; Pinto et al., 2011]. The sensitivity and specificity are influenced considerably by the choice of the log R-ratio-calling threshold cutoff values for discriminating deletions and duplications from normal diploid copy number. A high calling threshold will give a higher specificity and can strongly decrease the (technical) false-positive rate. However, it will also lower the sensitivity of the test. Hence, every laboratory needs to determine the parameters that offer the best performance for their platform and sample type.

Validation experiments

Any user of the technology should assure the quality and accuracy of the results obtained. Therefore, before offering any new technique or test for diagnostic use, the test should be internally validated. To do this, a number of control experiments can be set up [Fiegler et al., 2006; Vermeesch et al., 2007] to validate the optimal thresholds and to establish the false-positive and false-negative rates of array results. These experiments include self–self hybridizations, replicate experiments, gender-mismatch analyses, and hybridizations using DNA with established copy number changes. A reasonable number of samples to be tested for validation purposes is 50, comprising both normal and known abnormal samples. Normal samples are also beneficial in defining the optimum calling thresholds because apparently normal individuals also harbor benign CNVs.

Self–self hybridizations enable control of the standard deviation (SD) of log R ratios, allowing for technical noise to be assessed by removing biological differences between cohybridized samples. Sex-mismatch experiments enable rapid determination of the dynamic range. In addition, the laboratory should determine the operational resolution of an array experiment, and the false-positive and false-negative rates of the platform, by performing experiments with DNA samples with different-sized known CNVs (preferably <1 Mb in size).

Self–self hybridizations can only be performed when comparative genome hybridizations are used as a platform (array–CGH). DNA from the same individual is labeled with both Cy3 and Cy5 dyes (or suitable substitutes), and they are hybridized together. By definition, there cannot be any CNVs when DNA is hybridized against itself. None of the targets on the array platform should have intensity ratio changes, and no CNVs should be detected. In reality, however, a small number of intensity ratio outliers or false CNV calls will be made due to labeling biases and hybridization artifacts. Such analyses can be used to determine the default parameter settings to avoid false-positive results.

Important Internal Quality Parameters

SD of intensity ratios

The most important parameter in array–CGH is the SD of the log R intensity ratios at regions with similar copy numbers. This is commonly measured as the derivative log ratio (DLR) spread with oligo arrays and as the median absolute pairwise difference with Affymetrix SNP arrays. The higher this SD, the more the information that is lost, and the more difficult it becomes to perform an accurate analysis. It is essential to optimize array protocols in order to achieve low SD/DLR values.

In Figure 1, the consequence of different SDs on the ability to detect copy number changes is shown. Higher SDs lower the operational resolution of an array and can result in an increased number of false-positive calls. With low SDs, a smaller number of targets deviating from normal can be called copy number variable with confidence (for the theoretical approach, see Vermeesch et al., 2005).

Figure 1.

The effect of different SDs of log R ratios on the technical noise of array experiment is shown. Targets are plotted by genomic position on the x-axis, and the normalized log R signal intensity ratios are plotted on the y-axis. On the left is a sex-mismatch array–CGH experiment, in which genomic DNA from a female sample labeled in Cy5 is hybridized against a male reference sample labeled in Cy3. On the right, the experiment is repeated using the same sample, but this time, DNA is isolated from a single cell. The effect on the SD is apparent in the image from a single cell, which displays more technical noise, leading to an increase in false-positive calls.

Dynamic range

It is not only important to have low SD values, but the distinction between one, two, and three copies should be as close to the theoretical values for deletions and duplications (log2[1/2] =–1 and log2[3/2] = 0.58, respectively) as possible. In reality, a wide range of dynamic values is observed, which varies depending on factors such as the array platform used, the quality of the targets, the quality of the input DNA sample, as well as other technical variables.

Dynamic range can be reduced by poor hybridization, insufficient washing conditions, or saturation of target probes. With lower dynamic ranges, the ability to call a region as copy variable decreases and it becomes more difficult to distinguish an imbalance from normal technical variation. Hence, arrays and array protocols should be optimized to come as close as possible to the theoretical values. Poor dynamic range will particularly affect the ability to detect mosaicism and to place an accurate estimate on the level of mosaicism.

Printing and hybridization artifacts

Occasionally, printing or hybridization and washing artifacts are observed, and some examples are shown in Figure 2. The causes of these artifacts vary widely, and hybridization and washing protocols should be optimized and monitored closely to ensure the quality of array experiments. Clearly, any artifact will reduce the quality of the data. Software packages may provide methods to identify and quantify the number of outliers due to such artifacts. However, since this is not necessarily the case, visual inspection of the scanned images is warranted. As a general rule, these outliers should represent <1% of the probes on the array.

Figure 2.

Scanned images of microarrays with hybridization and washing artifacts displayed. On the left and center, artifacts resulting from large air bubbles during the hybridization process are shown. Poor hybridization is visible as the darker regions, which show reduced signal intensities of the underlying targets. On the right, typical washing artifacts (seen as green fluorescence) are shown.

Signal intensities

Signal intensities should be as high as possible without causing saturation of the probes. As a general guideline, it is advisable to minimize saturated probes to <1% total probes across the array. Low signal intensities are more likely to give inaccurate log2 ratios, and it may be advisable to filter out probes of low intensity. However, caution is needed because a low signal intensity would be expected for homozygous deletions with the respective probes in one of the hybridizations.

Background

The background signal intensities should be as low as possible and of even intensity across the array. Artifacts can inflate the background readings for Cy5 or Cy3 (or both), affecting the processed signals and therefore affecting the calculation of log R ratios. Washing artifacts are typically seen as green fluorescence, so it will affect the background readings for the Cy3 data as well as the Cy3 signal intensity data of the targets. Array washing protocols should be optimized to ensure consistent low background values.

Signal intensity ratio

Due to the increased incorporation and thus fluorescence of Cy3 compared with Cy5, it is necessary to correct the log2 ratios, typically by a global Lowess normalization. Poor labeling (possibly due to a poor-quality DNA sample) may be identified by skewed Cy3–Cy5 signal intensity ratios. Typically, values around 1.5 are expected. Values <1 and >2 are considered poor. It has also been demonstrated that environmental ozone levels affect Cy5 signal intensities, thus reducing data quality [Fare et al., 2003]. To overcome this issue, a novel dye “Hyper5” was developed, which is more ozone stable and photo stable than Cy5 [Dar et al., 2008]. Cabinets fitted with ozone scrubbers are available and used in some laboratories for microarray scanners, hybridization areas, and washing areas to reduce environmental ozone levels.

Signal-to-noise ratio

The signal-to-noise ratio provides a useful measurement of signal quality. Higher signal-to-noise ratios are achieved with good signal intensities and low background readings, providing higher confidence of accurate log2 ratios. The lower this ratio is, the greater the effect of subtracting the background on log2 ratios will be.

Wave

On occasions, an undulating “wave” may be observed across the genome. The increased deviation of log2 ratios from zero correlates to guanine–cytosine (GC)-rich genomic regions (light-stained cytogenetic bands on conventional G-banded karyotype; see Fig. 3). The exact cause of this effect is unknown. Ensuring a good denaturation of the DNA sample during the labeling process is advisable to minimize any contribution this may have. Methods of correction, mainly for percent GC content, have been described [Cheng et al., 2011; Van de Wiel et al., 2009]. Some software platforms provide a “wave” or GC metric, and may, in addition, provide methods to compensate for this effect. Caution should be taken not to overcorrect for this factor and thereby remove true aberrations.

Figure 3.

A wave effect observed in an array CGH experiment. Patient DNA is labeled in Cy5 and hybridized against reference DNA labeled in Cy3, shown in blue dataset, and the same samples are labeled in a dye swap experiment, seen with the yellow dataset. Results are displayed for chromosome 7, where targets are plotted by genomic position on the x-axis, and the normalized log R signal intensity ratio on the y-axis. The increased SD of the log R ratios in regions rich in GC content (light-stained G cytobands) can be seen in comparison with regions of low GC content (dark-stained G cytobands), leading to a number of false-positive calls.

Sample spike-ins

Some manufacturers now provide spike-ins comprising unique DNA sequences that hybridize to targets on the array and assist in identifying sample mix-ups during the array experiment. The use of spike-ins may partly reduce the need for transfer checks throughout an entire array experiment, provided there is an initial check against sample and patient name at the start of the procedure.

Technical operator variation

Genome-wide array techniques involve many processing stages, and therefore variation in the internal quality may arise due to different operators carrying out the procedures. Internal processes should be put in place to check for individual operator variability.

Analysis and Interpretation Quality Criteria

Data analysis is typically performed with the manufacturer's software or other commercially available software platforms. The large number of oligo probes on commonly used genomic arrays requires statistical algorithms to be used to identify aberrant regions. Circular binary segmentation is one commonly used segmentation algorithm, which divides regions into segments of similar deviation from log2 ratio of zero [Olshen et al., 2004; Venkatraman and Olshen, 2007]. Analysts should be aware that different algorithms may produce different array results [Lai et al., 2005; Pinto et al., 2011]. Using multiple different calling algorithms can increase confidence, and they may complement each other.

Calling thresholds should be carefully optimized using control samples with known imbalances of different sizes to assess the operational dynamic range and the most suitable settings in order to maximize the detection of true positives and minimize false-negative and false-positive calls. The number of aberrant probes used to make a call is another factor to consider. Large imbalances containing many probes are more easily identified, but false-positive calls (particularly duplications) are more likely to occur when calling with <5 oligonucleotide targets or with data of poor quality.

When using SNP arrays, only a single hybridization is performed for the patient DNA (single channel/color), and the signal intensities are then compared with a reference dataset. This reference dataset is provided by the companies, but can also be generated by the laboratory. The generation of a large in-house reference dataset has been shown to improve the quality of results in comparison with the reference datasets provided by the manufacturer. For example, Pinto et al. (2011) found differences in the quality control values for Affymetrix SNP arrays depending on the reference dataset used. A disadvantage of the short targets (25-mer oligonucleotides) used for SNP detection is the lower signal-to-noise ratios than those achieved with longer targets.

Whether to apply filters based on size may be a consideration. As a general rule, large imbalances of >1 Mb in size are more likely to be causal, whereas small imbalances of <200 kb in size are more likely to be benign. The use of arrays with different backbone and target resolutions may overcome the need to apply size filters. It is possible to generate automated filters that mask unwanted regions from detection. In practice, it is preferable to fully analyze the array data and for the laboratory to manually filter out regions rather than relying on automated filters.

Detection of Mosaicism

Mosaicism can and should be detected with the use of genomic arrays [Hoang et al., 2011; Scott et al., 2010; Valli et al., 2011]. Individual laboratories should determine the minimum level at which mosaicism can be detected on their array platform. This can be achieved with a dilution series using DNA from an individual with a known imbalance mixed with DNA from a normal individual. The ability at which mosaicism can be detected depends on various quality parameters (discussed above). It is recommended that laboratories should be able to detect a minimum of 30% mosaicism for microscopically visible chromosomal imbalances (>10 Mb) when using genomic arrays, although mosaicism may still be detected at levels below 10%.

Another factor to consider is how to approach discordance between array results obtained from uncultured material and conventional cytogenetic results from cultured material. There may be occasions when the result from direct material differs from that from cell culture. There are several reports that array–CGH can detect mosaic aberrations that are not observed on cultured material [Ballif et al., 2006; Cheung et al., 2007; Gardner and Sutherland, 2004]. However, when a discordant result is found, confirmation of the abnormality is advisable, preferably by fluorescence in situ hybridization (FISH) analysis on uncultured material if possible.

Categories of Classification for CNVs

The laboratory should interpret and classify any CNV detected into one of the following four general categories: (1) pathogenic or clinically relevant, (2) clinically significant but unrelated to the phenotype, (3) uncertain clinical significance, or (4) likely benign polymorphic CNV. This is naturally a spectrum with differing degrees of confidence depending on a number of factors, including prior knowledge, size, gene content, and inheritance status [de Ravel et al., 2007]. It is essential that copy number state is taken into account because deletions may be causal, whereas the reciprocal duplication may be benign [Hannes et al., 2009]; amplifications may display a phenotype whereas duplications may not [Balikova et al., 2008]; and homozygous deletions may be causal whereas the hemizygous state may represent only carrier status for a recessive disorder [Balikova et al., 2009]. There are ongoing large-scale efforts to collect and classify CNVs [Cooper et al., 2011; de Leeuw et al., 2012; Kaminsky et al., 2011]. Care should be taken at the reporting stage (discussed later) since many interchangeable terms are used to describe the different clinical consequences of CNVs. The terminology used for classifying CNVs is discussed further by de Leeuw et al. (2012). It should be noted that the interpretation of CNVs depends on the clinical indications. Clinicians should be urged to provide as good a clinical phenotype as possible with the referral request.

Pathogenic or clinically relevant CNV

A CNV may be classified as pathogenic if well documented in the literature, for example, known microdeletion/microduplication syndromes. In the absence of peer-reviewed publications, external databases such as Database of Chromosomal Imbalance and Phenotype in Humans using Ensembl Resources (http://decipher.sanger.ac.uk/) [Firth et al., 2009] or European Cytogeneticists Association Register of Unbalanced Chromosome Aberrations (http://umcecaruca01.extern.umcn.nl:8080/ecaruca/ecaruca.jsp) [Feenstra et al., 2006] may be useful if a clear correlation can be made between the observed imbalance and phenotype of the patient and the respective imbalances and associated phenotypes of at least two reported patients.

Clinically significant CNVs unrelated to phenotype

Situations may arise when particular CNVs are identified that are of clinical significance for the patient and/or parents or other family members, but are unrelated to the initial referral request. These findings may include susceptibility or predisposition loci [Girirajan and Eichler, 2010]. The possibility of array testing revealing such asymptomatic/presymptomatic results should be discussed with the patient during the pretest counseling session, and informed consent should be granted. It is desirable to allow the patients to decide for themselves whether they wish or do not wish to be informed of such findings. However, it is also advisable that the patient should be told of a presymptomatic finding for which early treatment can be offered in order to facilitate appropriate treatment. It is essential that the patient should be informed of this possibility during the informed consent and pretest counseling session.

CNVs of uncertain clinical significance

There is a broad spectrum of variants of uncertain clinical significance, from likely pathogenic CNVs to likely benign CNVs. If it is not possible to confidently classify a particular CNV as pathogenic or benign, it should be reported as a variant of uncertain clinical significance (VOUS). However, the laboratory should try to subclassify any VOUS detected based on additional factors such as the occurrence in a single affected/normal individual, gene content, size, and so on. The subcategories are:

  • VOUS—likely pathogenic: for example, the CNV has been reported in only a single affected individual with a similar phenotype, or the CNV overlaps only partially with reports from a small number of affected individuals and the causal gene has not yet been identified. The CNV has not been previously reported, but encompasses a gene(s) whose function is well characterized and likely causal for the patient phenotype.

  • VOUS—unknown significance: for example, the particular CNV has not been previously observed. The affected gene(s) function is unknown. There is currently ongoing debate as to the exact clinical significance of the CNV (observed in small numbers of both normal and affected individuals).

  • VOUS—likely benign: for example, the CNV has not been previously observed, but is inherited from an apparently normal parent. The CNV contains no genes or is intronic. The CNV has only been observed in a small number of apparently normal individuals or by a single study and, as such, is not a common polymorphic CNV.

Benign polymorphic CNVs

To be considered a common benign polymorphism, a CNV should be observed in >1% apparently normal individuals. The Database of Genomic Variants (http://projects.tcag.ca/variation/) is a valuable Web resource that catalogues CNVs observed from studies of normal individuals. However, caution should be taken because particular CNV may have only been observed in a single study, or with a single platform, and also may not have been validated by an alternative means. The curation of an internal laboratory- and platform-specific list of common benign CNVs can assist with the interpretation process (see de Leeuw et al., 2012, for recommendations on databases). Further recommendations on CNV terminology and methods to assist in interpreting CNVs of uncertain clinical significance are also discussed in de Leeuw et al. (2012).

Inheritance Status/Familial Studies

In many cases, parental samples are necessary for a full interpretation of the proband array result. Array testing may identify private familial variants that have not been previously observed in studies of apparently normal individuals or in patients [Itsara et al., 2009; Mencarelli et al., 2008]. The ability to investigate parental samples assists determination of the inheritance status of any CNV and therefore in the classification process. Additional familial samples may also be required to determine whether a particular CNV is segregating with the phenotype within a family. However, caution should be taken not to automatically regard an inherited CNV as benign [Buysse et al., 2009]. There are numerous reports of inherited pathogenic deletions/duplications that display reduced penetrance and/or variable expression [Cooper et al., 2011; Fernandez et al., 2009; Kaminsky et al., 2011; Ou et al., 2008]. In addition, a deletion inherited from an apparently normal parent may in fact reveal a recessive disorder in the proband due to a mutation on the remaining allele that is not detectable by the array.

Asymptomatic Testing of Minors

There are guidelines for the asymptomatic genetic testing of minors [Borry et al., 2009; Cornel et al., 2009], which should be considered at the interpretation and reporting stage.

Internal Laboratory Databases

The ability to classify and store legacy data is an essential aspect of diagnostic work. This allows for the generation of laboratory-specific likely benign CNV loci, as well as easy comparison of patient genotype–phenotype information with previously identified aberrations within a laboratory. An internal database of array quality criteria is also advisable to monitor array performance over time, changes to standard protocols, and different reagent- or batch-induced changes. Many software manufacturers provide means to store legacy data and to view these data within a custom track in the analysis software. Alternatively, commercial platforms are available, which offer the advantage of storing data from multiple array types as well as clinical patient information (e.g., the BENCH platform from Cartagenia n.v. (Leuven, Belgium); www.cartagenia.com). To improve the data interpretation, it is essential that different laboratories include detailed clinical phenotypes with their results in such databases (see de Leeuw et al. in press).

Reporting Quality Criteria

Array guidelines are available at national levels in many countries, including UK [Association for Clinical Cytogenetics, 2007, 2009] as well as European levels [European Cytogenetics Association, 2011]. Reports must be issued in a standardized manner and be written for a nonspecialist, so they can be clearly understood by the recipient/clinician. The report will be inserted into the patient's notes and may be seen not only by the referring clinician, but also by healthcare workers, unless national legislation specifically states who is authorized to be informed. When writing a report, it is important to remember that it may also be made available to the patient at some stage. Authorization of reports must be carried out by a senior laboratory scientist. If applicable, a clinical geneticist may countersign the report.

It is the responsibility of the laboratory geneticist to provide a clear and unambiguous description of the microarray findings and an explanation of the clinical implications of the results. The clinical significance of a genomic imbalance depends on the size, genetic content, position, whether it is de novo or inherited, and, if inherited, also the carrier parent's phenotype. The literature and (public) databases should be used to help evaluate the clinical significance of the detected imbalance.

Although proband samples may be referred singly or with parental blood samples, depending on local referral policy, it is recommended that urgent proband samples are referred with parental blood samples. A preliminary report with the detection of a significant imbalance (as defined by the laboratory criteria) may be issued on detecting an imbalance in a proband without parental studies. However, comments on the clinical significance may be made in the preliminary report if a phenotypic association is supported in the published literature, for example, a deletion, del(4)(p16.3), involving the WHS1 gene can be reported in the absence of parental blood samples, although follow-up studies are required to exclude a balanced parental rearrangement.

Depending on the probability that a constitutional imbalance is causal, follow-up confirmatory experiments on the patient and their parents should be performed whenever appropriate. Array follow-up studies do not differentiate between a normal and balanced carrier status in the parents. Since both a deletion and a duplication may result from the unbalanced segregation of a balanced rearrangement in one of the parents, it is mandatory to perform additional follow-up tests (e.g., FISH, chromosome analysis) on the parental blood samples to exclude telomeric/insertional translocations or inversions and to provide an assessment of the recurrence risk for future pregnancies. Laboratories should be aware that duplications can result in a recessive disorder [Parri et al., 2010].

When preliminary results are given to the physician and/or patient, a verified printed copy must be issued with a clear indication that the analysis is provisional and stating that a final interpretation will be issued later. Any verbal communications must be documented on the patient's laboratory record, stating the information given, to whom and by whom, with the time and date and followed up with a printed copy. A final report of an imbalance must be issued after completing follow-up studies on both proband and parental samples. It is acknowledged that some reports will be complex and may only be fully understandable to the referring clinical geneticist.

Microarray reports must comply with OECD guidelines [Organisation for Economic Co-operation and Development, 2007] and include an interpretation of the results (unless national legislation specifically states this must be done by the referring medical professional). Handwritten alterations must never be made to the report. Accreditation standards insist that internal authorization procedures are in place to ensure no alteration of reports can be made after they have been issued.

The report of a normal microarray must contain

  • patient demographics;

  • reason for referral and patient phenotype;

  • material from which DNA is isolated;

  • description of array (manufacturer, array version);

  • the real array resolution (practical not theoretical);

  • identification of the genome build used;

  • limitations of the test used (e.g., low-level mosaicism, balanced rearrangements, triploidy);

  • summary statement, if no clinically significant imbalance was detected;

  • any relevant details of the practical processing.

CNVs considered to be common benign polymorphic CNVs do not need to be mentioned in the patient report, but should be kept on file and made available to clinicians if required. It is the duty of the clinician/clinical geneticist to ask a laboratory to reanalyze the data and/or recall patients if a previously reported “benign” CNV is later found to be associated with a pathogenic disorder. All CNV results must be retained by the laboratory, if possible in the patient's file and in a CNV database (local, national, or international).

The report of an abnormal microarray must, in addition, contain, if appropriate,

  • a summary statement and/or karyotype designation using ISCN nomenclature (latest version) if genomic imbalances are detected [International Standing Committee on Human Cytogenetic Nomenclature, 2009];

  • a clear written description of the genomic imbalance detected;

  • the location of genomic imbalances (also reporting the position of the first and last significantly aberrant probes, and ideally those of the first and last flanking normal probes);

  • the size, both minimum and maximum (first abnormal and first normal flanking targets);

  • the gene content of the genomic imbalance including the name of any known syndrome(s) in the region. “Gene content” may refer to specific genes that are clinically relevant, for example, Online Mendelian Inheritance in Man (OMIM) listed genes. Where there are few genes involved, they may be listed, or alternatively, a quantitative statement can be given such as “there are many genes in this region” or “there are no genes in this region.”

  • reference to other investigations to clarify significance;

  • the name of any associated syndrome/disease, if applicable;

  • whether the result is consistent with the clinical findings, and/or an indication of the expected phenotype, if it is possible to determine;

  • identification of methods used in follow-up studies;

  • whether there is a risk of recurrence, if it is possible to determine;

  • prenatal diagnosis in a future pregnancy, if applicable;

  • onward referral for genetic counseling, if the referral has not been initiated by a clinical geneticist;

  • where appropriate, a request for follow-up of family members at risk of the abnormality, starting with the closest available relatives.

The laboratory should have a written policy for reporting times and for identifying urgent results. Laboratories' reporting times should take into account the reason for referral and level of urgency. There should not be any delay in reporting results due to insufficient staffing or administrative procedures. It is, however, recognized that it may not be possible to evaluate the true clinical significance of an imbalance detected in the proband without parental studies and that the time taken to obtain parental blood samples may vary (if these have not been sent at the same time as the proband's sample). Recommended maximum reporting times for 90% of referrals are given in Table 1.

Table 1. Recommended Reporting Times for Selected Referrals
  1. These reporting times include weekends and public holidays.

Prenatal results17 days
Urgent results (e.g., neonatal)17 days
Other referrals (no follow-up required)6 weeks
Other referrals (where follow-up is required)Within 4 weeks of receipt of parental blood samples

For array analysis, the technical parameters of the array as well as data files and array results should be stored digitally in the laboratory and/or in the patient records for at least 30 years, and, if possible, indefinitely.

Clinical Indications for Investigation by Genome-Wide Array

Postnatal

Patients with:

  • clinically significant abnormal growth, for example, short stature, excessive growth, microcephaly, macrocephaly, and dysmorphism;

  • multiple congenital abnormalities;

  • intellectual disability, developmental delay, autistic spectrum disorder;

  • suspected deletion/duplication syndrome;

  • X-linked recessive disorder in a female.

Prenatal

See Vetro et al. (2012) for specific guidelines on the use of genomic arrays for prenatal diagnosis.

Acquired

  • See Simons et al. (2012).

Considerations for Prenatal Diagnosis

These guidelines are intended for use in postnatal constitutional genetic diagnosis, but, in general, they also apply to the use of arrays for prenatal diagnosis. Further guidance on the use of genomic arrays for prenatal diagnosis is provided by Vetro et al. (2012). There are important, additional factors for consideration in the prenatal setting, particularly at the interpretation and reporting stages.

Conclusions

Arrays are now mainstream diagnostic tools. Given the many variables that affect analytical validity, all laboratories must internally validate their array technique before offering the service for diagnostic purposes. Access to databases is paramount in the interpretation of array results, as well as an understanding of the possible etiology when requesting follow-up tests or samples. With the proper internal and external quality control measures firmly in place, genetic diagnostic laboratories can and should implement these tools in the diagnosis of patients with developmental anomalies.

Ancillary