Introduction
Chromosome analysis remains one of the most commonly performed diagnostic genetic tests, being offered for a wide variety of indications in obstetrics and gynecology, pediatrics and oncology. At the cytological level, banded human chromosomes show a consistent and similar pattern in clinically healthy individuals. Hence, balanced and unbalanced chromosomal aberrations can serve as informative markers for a clinical phenotype. Although healthy individuals do not show variation at most cytogenetic landmark bands, a few chromosomal heteromorphisms have been identified and have long been considered benign variants without clinical significance (for example, refs. 1–3).
With the advent and application of array-based comparative genomic hybridization (array CGH), which allows analysis of the genome at a significantly higher resolution than previously possible, scientists have demonstrated that humans are much more genetically variable than previously appreciated. Indeed, the genomes of unrelated individuals can differ from one another with respect to the copy number of thousands of loci. This has led to new challenges for the clinical cytogeneticist when interpreting vast amounts of information gathered from high-resolution array CGH assays performed in clinical diagnostic settings.
Blurring the boundaries of normality
On April 14, 2003, 50 years after the landmark discovery of the structure of the double helix, leaders from six countries proclaimed that the human genome project was complete (see the Nature Human Genome Collection (Nature S1, 4; 2006)). This massive undertaking was especially rewarding because of the implication that with one complete human DNA sequence, we now had the DNA sequence of essentially every human being. This implication was based on the notion that the genomes of healthy individuals were 99.9% identical and that the major genetic differences that existed between individuals were in the form of scattered single–base pair changes accounting for 0.1% of the genome. This scientific dogma was recently challenged when two independent studies analyzed the genomes of unrelated, healthy individuals using genome-wide array CGH platforms and found hundreds of genomic regions that varied—not significantly with respect to the actually DNA sequence, but significantly with respect to the number of copies an individual had of each DNA segment4, 5. These submicroscopic copy number variants (CNVs) are both intriguing and of particular concern to clinical cytogeneticists, who have relied on a 'standardized' genome (represented at the cytogenetic level as the human karyotype) to identify 'abnormal' chromosomal alterations that could be implicated in the etiology of a individual's disease or disorder.
A CNV is now operationally defined as a DNA segment, longer than 1 kb, with a variable copy number compared with a reference genome6, 7. This broad definition for CNVs makes no reference to the clinical impact of a given genomic imbalance and can be confusing for clinical geneticists and clinical cytogeneticists who have traditionally understood chromosomal 'variants' as being alterations that are not clinically significant. This overarching use of the term CNV is also being endorsed by scientists who note that our understanding of the functional impact of genomic imbalances continues to evolve as genotype-phenotype correlative data accumulate and as significant biomarkers are obtained from large-scale disease association studies. Hence, what may have once been thought of as clinically insignificant might later turn out to be a CNV that confers differential susceptibility to a disease or is even causative of a genomic disorder with late onset or variable penetrance. Therefore, 'CNV' is now used to describe copy number differences in studies of both disease and normal controls (reviewed in ref. 8) as well as imbalances that cause well-known microdeletion and microduplication syndromes (for example, ref. 9). To minimize confusion, it may be justified to use qualifiers for the term 'CNV' when discussing functional or clinical significance. The terms 'pathogenic CNV', 'benign CNV' or 'CNV of unknown clinical significance' may be useful for this purpose.
Array CGH diagnostic testing
It has been estimated that in the US alone, more than 10,000 array CGH tests are now performed clinically every year. The vast majority of this testing is performed using 'targeted' array platforms that specifically assess relative copy number for critical regions of well-defined genetic syndromes10, 11. Genome-wide arrays, on the other hand, mimic the human genome at a given resolution by placing probes at defined intervals and/or with specific parameters12, 13, 14, 15.
There has been some hesitation in North America to use genome-wide arrays in a clinical setting. This stems in part from the knowledge of the large extent of CNVs in the genomes of healthy individuals and also from concerns about the difficulties in accurately and efficiently discriminating benign CNVs from pathogenic CNVs. Many laboratories in Europe and elsewhere seem to have taken the alternate approach, viewing genome-wide array CGH testing as analogous to banded chromosome analysis (karyotypic analysis) and favoring the ability to obtain more genomic information in a single test. Those that favor genome-wide platforms argue that even an array with 1-Mb resolution has the potential to capture up to twice as many clinically significant imbalances than most targeted arrays, leading to many more diagnoses and identifying more individuals with increased reproductive genetic recurrence risks.
Which genomic imbalances contribute to a genomic disorder?
When analyzing array CGH data in a clinical setting, clinical cytogeneticists should categorize CNVs into those that are likely to be benign, those that are likely to be pathogenic and those of unknown clinical significance. CNVs that overlap critical regions of known microdeletion or microduplication syndromes (or that overlap other genomic regions that are defined as clinically significant, such as subtelomeric regions) are likely to be pathogenic in nature. These genomic imbalances are then often validated with an alternate molecular detection technology. Deletions >40 kb can be confirmed by standard metaphase or interphase FISH analyses using a clone within the deletion interval (Fig. 1a,b). Duplications that are tandem in nature may also be verified by standard metaphase FISH, provided that the duplication is extremely large (>10 Mb; Fig. 1c,d) or, in some cases, may be verified by interphase FISH (Fig. 1e,f). Smaller duplications (on the order of 40 kb) may also be detectable by metaphase FISH if the duplication is distributed to other chromosomal locations (for example, Fig. 1g,h). Otherwise, these duplications and deletions (or imbalances <40 kb) could also be validated by customized multiplex ligation-dependent probe amplification (MLPA) assays, quantitative PCR methods, microsatellite genotyping, long-range PCR or, in some cases, with the use of another array CGH or genotyping platform. Fiber-FISH methods can be extremely informative, especially for determining chromosome-specific copy number information, but they are technically challenging, time consuming and usually outside the means of most clinical cytogenetic laboratories (see ref. 6 for further information on the use of these and other technologies to validate CNVs).
Figure 1: Detection of CNVs using a dye-reversal strategy on a 1-Mb resolution array platform (Spectral Genomics) and subsequent validation using multicolor FISH.
(a,b) A deletion of four BAC clones at proximal chromosome 14q (a) and metaphase FISH confirmation of deletion of one of these BAC clones, RP11-91K19 (b, green). The chromosome with the deletion is indicated by an arrow, and a control FISH probe for the 14q subtelomeric region is in red. (c,d) A duplication of 8p chromosomal material, detected by 13 BAC clones (c) and metaphase FISH confirmation of the 8p duplication (d), suggesting that the duplication is tandem in distribution. One of the BAC clones that is on this array platform and within the duplicated region, RP11-1K11 (red), was used in the FISH confirmation study, along with a chromosome enumeration probe for chromosome 8 (CEP 8; green). The duplicated region is denoted by an arrow. (e,f) A duplication of 1p chromosomal material, detected by six BAC clones (e) and interphase FISH of one of the duplicated BAC clones, RP5-820O16 (f, red) showing three loci as opposed to only two in control experiments (data not shown). (g,h) A duplication of distal 15q chromosomal material detected by at least six BAC clones (g) and metaphase FISH confirmation of the 15q duplication using a commercially available chromosome 15 subtelomeric probe (h; Vysis; yellow). The third copy of chromosome 15 subtelomeric material is found at the short arm of a derivative chromosome 13 (arrow).
Full size image (410 KB)For CNVs that are initially of unknown clinical significance, clinical cytogeneticists should first try to determine if the CNV is inherited or de novo in nature. Because of the importance of parental testing for determining the pathogenicity of most CNVs, it could be argued that array CGH testing (especially genome-wide testing) is unwarranted when it is foreknown that parental samples will not be available for follow-up studies. However, the potential for identifying imbalances that overlap critical regions of well-defined genomic disorders still remains, and as mentioned above, cytogeneticists can be more confident of the pathogenicity of these genomic imbalances even without parental testing. For other CNVs that are of unknown clinical significance, clinical cytogeneticists should try to exclude non-paternity as well as non-maternity (in the form of sample mix-up). Then, both parents could be tested with the same array CGH platform as used for the initial diagnostic test. This would provide some indication on whether the CNV is inherited (Fig. 2). Alternatively, for cost effectiveness, other clinical laboratories may use targeted assays to validate the genomic imbalance in an individual (for example, FISH, MLPA or quantitative PCR) and investigate the inheritance of the CNV. Whenever possible and interpretable, FISH studies should be considered, as they may provide chromosomal distribution information (not available from other molecular-based testing) that may provide important information on recurrence risk to the parents.
Figure 2: Array CGH results on an Affymetrix 500K-EA array show a 266-kb deletion CNV at chromosome (chr) region 22q11.23 in a child (denoted by the arrow on the child's chromosome 22 profile) and a similar 266-kb deletion CNV in the mother (denoted by the arrow on mother's chromosome 22 profile).
These results are consistent with the inheritance of this CNV. The x-axes represent position along the length of the chromosome.
Full size image (75 KB)If the genomic imbalance is seen both in the affected individual and in a healthy parent, it is more likely to be a benign CNV. Such a criterion for determining the pathogenicity of a genomic alteration has long been the standard for banded chromosome analysis and has been applied more recently for array CGH analyses13, 16, 17. Moreover, recent analyses of CNVs in healthy individuals have uncovered a very low mutation rate for genotypeable CNVs (
0.2%)18, consistent with the notion that benign CNVs are often stably inherited. However, most current array CGH methodologies are not ideal for determining inheritance of CNVs that have a multiallelic or complex population genetic distribution. For example, multiallelic CNVs can appear to be de novo in nature simply as a result of inheritance of allele-specific CNVs7. Obtaining allele-specific copy number information is a limitation of most currently used array CGH-based tests, with the exception of assays such as the molecular inversion probe tests19.
In the event of non-paternity (or when one or both parents are unavailable for testing), array CGH profiling or molecular testing on relatives may also prove to be informative. For example, 'unaffected' relatives carrying a similar CNV may suggest that the CNV in question is clinically insignificant. Alternatively, an 'affected' individual with developmental issues harboring a similar CNV may raise suspicions that the CNV in question is pathogenic. Of course, mounting evidence for the benign nature of a CNV should always be tempered with consideration of issues such as mosaicism, incomplete penetrance and variable expression, among others, during genetic counseling sessions. Genetic counseling should also consider the fact that apparently inherited CNVs sometimes have different breakpoints when analyzed with higher-resolution assays (C.L., unpublished observations) and thus might result in different functions or phenotypes.
Non-inherited (or de novo) CNVs should subsequently be cross-referenced to catalogs of CNVs found in affected as well as healthy individuals to further assess the possible pathogenicity of the genomic imbalance. To obtain a list of 'benign' CNVs, clinical laboratories may develop an internal database that includes CNVs obtained during validation of an array CGH platform before clinical implementation. Such validation studies ideally would use DNAs from healthy control individuals. However, some laboratories may also choose to include data from individuals that have clinical presentations. In these cases, the assumption is that a single, large CNV is deemed pathogenic and that the other imbalances observed in the affected individual (which are usually smaller in size) are less likely to contribute to the etiology of the genomic disorder. Of course, this assumes that combinations of multiple CNVs may not be pathogenic—which, although not yet refuted, may not actually be true.
Publicly available databases that accumulate CNV data on hundreds of healthy individuals are also now available. One such initiative is the Database of Genomic Variants (http://projects.tcag.ca/variation). As of March 18, 2007, this database contained 6,482 CNV entries from 40 different published studies; many of the CNVs are 40–100 kb. The catalogued CNVs account for over 500 Mb of genomic DNA, representing nearly 20% of the human reference genome. To determine which genomic imbalances are pathogenic, a clinical cytogeneticist could simply tabulate all genomic imbalances observed in an individual's array CGH test, disregard CNVs found in normal individuals and consider the remaining copy number changes as potentially pathogenic in nature. However, currently, much of the information in the Database of Genomic Variants is derived from CNV discovery projects that use a single platform and/or technology to identify CNVs, and only a fraction of those identified CNVs are ultimately validated. This raises the possibility that a substantial amount of the catalogued data may be erroneous and, therefore, that caution should be exercised when relying heavily on such databases to determine if a CNV is clinically significant. This holds true especially for CNVs that have been reported only in a single individual. Although such a CNV could be a rare variant (accounting for its low frequency), it is more likely to be a false positive and thus could lead to incorrect interpretation of the clinical significance of that CNV. Indeed, with the increasing number of CNV entries now being routinely incorporated into such public resources, careful examination and standardization of CNV data is now required8.
In addition, databases are now becoming available that specifically annotate copy number variation information from individuals seen in genetic clinics. De novo CNVs that are not found in the Database of Genomic Variants (or similar databases) should then be cross referenced to recurrent pathogenic CNVs found in specimens from affected individuals. Such a comparative analysis comes with an understanding that any matched CNV increases the probability of the given CNV being pathogenic in nature, especially if it is observed recurrently in individuals with similar or overlapping phenotypes. One such database is the Database of Chromosomal Imbalance and Phenotype in Humans using Ensembl Resources (DECIPHER, http://www.sanger.ac.uk/PostGenomics/decipher/). Clinical information, along with chromosomal aberration data (primarily in the form of array CGH information), is collected for each individual included in the database. The aim of this effort is to facilitate accurate interpretation of diagnostic array CGH results for clinical cytogeneticists and clinical geneticists who provide constitutional genetic (and in some cases prenatal) testing. As of May 23, 2007, this database contained information on 803 individuals with 48 defined genetic syndromes. Other similar initiatives include the Chromosome Abnormality Database (CAD; http://www.ukcad.org.uk/cocoon/ukcad/), the Mendelian Cytogenetics Network Online Database and the European Cytogeneticists Association Register of Unbalanced Chromosome Aberrations (ECARUCA, http://www.ecaruca.net). The entry of data from different studies into these public databases will facilitate comparison of array CGH results from individuals with rare genetic diseases and disorders, leading to rapid delineation of new genomic syndromes (for example, refs. 20–22). If these databases minimize biases in CNV entries, significant correlations may eventually identify specific combinations of CNVs that lead to pathogenicity but that in isolation may be benign.
For de novo CNVs (especially those that have not been observed among healthy individuals and do not coincide with recurrent CNVs observed in individuals with similar clinical presentations), several factors can be used to guide further assessment of the clinical relevance of the CNVs.
Gene content. The potential clinical relevance of a CNV increases with respect to the number of genes within the region of genomic imbalance. This is especially true if the genes include morbid OMIM genes, if they are known to be important in development and/or if they are known to have a dosage phenotypic effect.
Nature of the CNV. As it is generally thought that duplications are genetic alterations that are better tolerated in the genome than deletions23, 24, deletion CNVs have a higher likelihood of being pathogenic. However, single-copy deletion CNVs can be commonly observed in healthy individuals. By current estimates, a given person's genome may contain as much as 740 kb of DNA sequences that are present in only one copy25. Homozygous deletions have also been noted in the genomes of healthy individuals, although fewer of these have been described18, 25, 26, 27. Clinical cytogeneticists should pay particular attention to whether a deletion CNV has been found as a single-copy deletion or a two-copy deletion in healthy control individuals and should also note that current data indicate that most one-copy deletions, found in healthy individuals, have an average size of 15–20 kb25. This suggests that larger deletion CNVs may be more prevalent in individuals with genomic disorders and therefore confer a higher risk factor for pathogenicity when observed as a de novo alteration in an affected individual.
Single-copy deletion CNVs can be phenotypically benign, with a corresponding homozygous deletion leading to a recognizable clinical phenotype (Fig. 3). For example, a single-copy deletion of the NPHP1 gene may be found in otherwise healthy individuals, but a two-copy (homozygous) deletion of this gene results in juvenile nephronophthisis28, 29. Moreover, caution should be maintained when suggesting that a single-copy deletion CNV is benign simply because it has been previously observed among healthy individuals. A limitation of current array CGH assays is that they are unable to detect small or point-recessive mutations. This means that a given single-copy deletion CNV could be detected by array CGH in one or more healthy individuals but that the same single-copy deletion may also be associated with pathogenicity if it uncovers a recessive mutation on the other chromosome in an affected individual (Fig. 4).
Figure 3: An example of a homozygous deletion in a clinically affected individual.
In a 17-year-old male with developmental delay and multiple dysmorphic features, a single BAC clone detects a homozygous deletion in this individual. This BAC clone, RP11-M17M8, overlaps entirely with variation locus ID#1866 in the Database of Genomic Variants, and therefore a single-copy deletion of this genomic region may be interpreted as a benign variant. However, it is unclear if a homozygous deletion of this same region is also clinically insignificant. (a) Chromosome 8–specific array CGH profile showing the homozygous deletion (arrow). (b) FISH results of the same BAC probe on a healthy control specimen showing the chromosome 8 centromere and the RP11-17M8 clone (red signals). (c) FISH of the affected individual confirming the deletion of this BAC probe on both chromosome 8 homologs (only the centromere signal, in red, remains). At least one of the healthy parents had a single-copy deletion detected by this same BAC probe. The other parent was unavailable for follow-up testing, and uniparental disomy was ruled out.
Full size image (127 KB)Figure 4: Individuals with a single-copy deletion CNV or a single recessive mutation can be healthy.
However, the same single-copy deletion CNV could uncover a recessive mutation in certain individuals, leading to a clinically recognizable phenotype.
Full size image (90 KB)CNVs that exist as both deletions and duplications in a population are even more difficult to interpret in the absence of further correlative data. For example, CNVs that are duplicated in healthy individuals may be haploinsufficient in affected individuals. Other CNVs that are found in zero, one or two copies in healthy individuals may be less tolerable when increased to more than two copies per cell (J.R. Vermeesch, personal communication).
Size of the CNV. The size of a given CNV should also be taken into account when considering its significance in the etiology of a disease or disorder. Clearly, larger, contiguous CNVs are more likely to encompass genomic regions that subsequently alter the RNA and protein levels of genes critical for maintaining proper development and immunologic welfare. However, many apparently benign CNVs have been found to be on the order of 2 Mb in size18, and in some case studies, benign CNVs as large as 10 Mb in size have been reported30. Compared with CNV size, gene content would be a more reliable indicator for clinical significance such that small, gene-rich CNVs are more likely to be pathogenic than larger, gene-poor CNVs (for example, CNVs encompassing gene deserts and devoid of known regulatory elements). (It should be noted that CNVs of some potential regulatory elements have recently been shown to affect transcription of genes as far as 6 Mb away31.)
When considering all of these factors (Table 1), one should remember that most currently available CNV data lack precise breakpoint information. This includes CNV data in public databases that collate genomic imbalances found in normal individuals as well as in affected individuals, as many of these data originated from use of array CGH platforms that contain large-insert clones (for example, BAC and PAC clones, which are on the order of 150 kb in size). Empirically, a significant copy number ratio change can be noted when a clone has gained or lost a copy of even 25%–30% of its unique DNA sequences. This limited resolution for currently identified CNVs sometimes leads to the potentially erroneous assumption that a CNV identified by a given clone on an array is the same CNV in another individual when identified by the same clone (Fig. 5). This will eventually become less of an issue as the molecular structures of individual CNVs are better defined and clinical array CGH assays employ higher resolution platforms to more accurately genotype each genomic imbalance being detected.
Figure 5: Two affected individuals with different CNVs may produce an indistinguishable array CGH result.
Here, individual 1 has a 40-kb duplication (upper left; in blue) that is different from a 40-kb duplication in individual 2 (lower left; in red). However, because these two CNVs fall within the same BAC clone on an array CGH platform (that is, BAC clone RP11-258ZF38; lower right), they produce similar array CGH results (upper right) but could lead to different clinical outcomes. The x axis represents fluorescence intensity ratios.
Full size image (97 KB)
'Benign' CNVs that overlap with clinically relevant genomic regions
With the recent establishment of a copy number variation map for the human genome18, there have been several clinically relevant observations. First, 39 CNVs were identified within 500 kb of a chromosome end (that is, involving the subtelomeric regions). As these were identified in healthy individuals, they are presumed to be benign in nature. In a clinical setting, subtelomeric probes have been used routinely to assess the integrity of regions near the chromosome ends32, 33. These chromosome regions are gene rich and are believed to contain many developmentally important genes. De novo genomic imbalances in these chromosome regions are therefore implicated as a cause in 2.6% (ref. 34) to as many as 7.4% (ref. 35) of otherwise idiopathic mental retardation cases, depending on the referral criteria. Genomic imbalances in these chromosome regions that are inherited from a healthy parent are classified as benign and have been detected for at least seven subtelomeric regions (2q, 3q, 4q, 7q, 9p, 17q and Xp/Yp)36. As noted above for CNVs at other genomic regions, a subtelomeric CNV that seems to have been inherited from a clinically normal parent (judging from FISH or MLPA testing) might actually have different breakpoints or might be a different size in the parent and child. This could result in incorrect interpretation of a subtelomeric imbalance identified in an affected individual (and especially concerning in prenatal diagnostic testing). Application of higher-resolution arrays (especially those such as 'subtelomeric molecular rulers' that specifically target the subtelomeric region)37 should be helpful with accurate diagnosis of these chromosomal regions. Nevertheless, the identification (and subsequent characterization) of 39 potentially benign human subtelomeric CNVs will assist in accurate interpretation of future clinical subtelomeric tests, especially when one or both parents are unavailable for testing inheritance.
It has also been noted that apparently benign CNVs18 that overlap with chromosomal regions used as diagnostic probes for at least ten genomic disorders (for examples, see Fig. 6). As these CNVs were found in the HapMap populations, which comprised healthy individuals, it is unlikely that these 'control' individuals had any of these genomic disorders. These CNVs now need to be validated and carefully characterized to determine if the optimal diagnostic probes are being used.
Figure 6: Overlap of CNVs with chromosomal regions associated with genomic disorders.
(a) Smith-Magenis syndrome. (b) Sotos syndrome. The location of a diagnostic probe for each syndrome is shown in red.
Full size image (81 KB)Conclusions
The application of array CGH technology in the clinical diagnostic arena has clearly highlighted the limitations of conventional chromosome banded cytogenetic analysis. Over the last 4–5 years, most clinical and research laboratories that make use of array CGH technology have used array platforms that consist of large-insert DNA clones. These platforms generally provide the highest-quality results owing to high signal-to-noise ratios. However, technology is already shifting such that array CGH platforms that use long isothermic oligonucleotides (such as those manufactured by Agilent, Nimblegen and Combimatrix) provide copy number ratio results that are comparable to those obtained from BAC-based arrays. These arrays have the added advantage of robustly attaching hundreds of thousands of probes to a single glass slide, providing genome-wide analyses of genomic imbalances at unprecedented resolution. Ultimately, genotyping platforms that also reliably detect copy number information (Affymetrix and Illumina) will provide clinical cytogeneticists with added information on genome-wide assessment of uniparental disomy in the form of loss of heterozygosity (for example, refs. 17,38). This added information will undoubtedly serve as another rich source of genetic variability information to help define new genetic syndromes and understand the complex molecular and biochemical interactions within the human genome.
With respect to the interpretation and follow-up studies of CNVs in individuals referred to cytogeneticists because of constitutional abnormalities, further research into the discovery and characterization of human CNVs is needed to develop more comprehensive human genetic variation maps. This in turn will facilitate more accurate interpretations of the clinical impact of specific genomic imbalances. Ultimately, these data (along with the experiences obtained over the years with array CGH testing for constitutional genetic disorders) will help implement widespread use of high-resolution, genome-wide array CGH assays not only for diagnosing genetic disorders in individuals born with dysmorphism and developmental delay but also for prenatal genetic diagnosis.
