BAC and oligo array platforms each have unique advantages and disadvantages in a diagnostic setting; these may include turnaround times, genomic coverage, and costs. One of the most important characteristics of each platform is the detection rate of clinically significant alterations. Our results demonstrate that our whole-genome oligo array was able to detect such alterations in 15.4% of patients tested, compared to the BAC array detection rate of 10.6%, a statistically significant difference (Fisher's Exact Test, p < 0.0001). The alterations that constitute the 4.8% difference in detection rate between the BAC and oligo arrays are either too small to be detected by the BAC array but are not below the resolution of the oligo platform (Figure ) or fall within gaps in the BAC array coverage (Figure ). Figure shows a 44 kb deletion of 17p13.3 detected in a patient referred to our laboratory for convulsions. This deletion encompasses the first exon of PAFAH1B1
). While it is not known whether this deletion results in a null allele or simply a truncated gene product, hemizygous deletions and mutations of this gene are found in patients with isolated lissencephaly type 1 (OMIM 607432) and have been linked to epileptic seizures and convulsions [20
]. Although RP11-135N5 provides coverage of this region on the BAC array, FISH analysis using this clone could not confirm the deletion in any cells because of the deletion's small size compared to the FISH probe used. Thus, this clinically significant deletion could only have been reliably detected using the oligo platform. Although oligo-based aCGH has the power to detect alterations smaller than the size of a BAC probe, BAC-based aCGH has an advantage in that the analysis makes evident the appropriate probe to be used for FISH confirmation. In addition, this probe is usually readily available because of its inclusion on the microarray platform and will have a high rate of successful confirmation. When oligonucleotide-based aCGH is performed, BAC probes must be specifically selected for the FISH confirmation of each small abnormality that is detected. Once a probe has been selected, it must also be specially prepared or ordered before FISH can be performed. This process increases both the time it takes to perform FISH confirmation of oligo aCGH results and the cost associated with the analysis.
Figure 4 Oligonucleotide microarray characterization of an interstitial deletion at 17p13.3. The zoomed-in microarray plot shows a single-copy loss of six probes from the short arm of chromosome 17 at 17p13.3, approximately 44.0 kb in size (chr17: 2,415,074-2,459,051, (more ...)
Figure 5 Oligonucleotide microarray characterization of an interstitial deletion at 6q14.1. The zoomed-in microarray plot shows a single-copy loss of 43 oligonucleotide probes from the long arm of chromosome 6 at 6q14.1, approximately 2.9 Mb in size (chr6: 79,838,518-82,730,466, (more ...)
Figure shows a 2.9 Mb deletion of 6q14.1 detected in a patient referred to our laboratory for developmental delay and dysmorphic features. This deletion encompasses eight genes: PHIP
, two of which are known to be associated with human disease [22
]. Although this 2.9 Mb deletion is likely to be clinically significant, it lies within a gap in the coverage of our BAC array and could only be detected using the oligo platform because of its more uniform backbone coverage.
The detection rate of alterations of unclear clinical significance is also a concern during the selection of a microarray platform in a clinical diagnostic setting. Our data suggest that both the oligo and BAC platforms detect similar numbers of abnormalities of unclear significance (7.0% by BAC and 7.1% by oligo), although the circumstances leading to an unclear clinical interpretation may vary between the platforms. On the BAC platform, unclear results are often associated with gaps in coverage which prevent the precise determination of the breakpoints and gene content of an abnormal region. This lack of information prohibits definitive interpretation of the clinical significance of the alteration. Figure presents a 262 kb deletion of 9q33.1 detected by BAC array in a patient referred for developmental delay, dysmorphic features, and multiple congenital anomalies. The boundaries of this alteration as defined by BAC array include only one gene, TLR4
]. However, gaps in BAC coverage on both sides of the alteration span 4.5 Mb proximally and 4.0 Mb distally. As a result of these coverage gaps, this alteration, though estimated to be just 262 kb, may be as large as 8.7 Mb and include up to 48 additional genes. The design of BAC arrays with dense clone coverage is possible; however, probe density is limited by the availability of BAC clones and the presence of potentially interfering genomic architecture such as segmental duplications. In addition, BAC-based microarrays will not reliably detect abnormalities smaller than the size of an individual probe--80-200 kb, on average, for BAC clones.
Figure 6 BAC microarray characterization of a 9q33.1 deletion. The zoomed-in microarray plot shows a single-copy loss of three BAC clones from the long arm of chromosome 9 at 9q33.1, approximately 262 kb in size (chr9: 119,452,279-119,714,054 based on UCSC 2006 (more ...)
Although gaps in coverage and limited breakpoint-resolving power are primarily a concern for BAC platforms, both oligo and BAC platforms produce results that are unclear because a lack of published evidence prevents a conclusive association between the gene content of an alteration and the clinical features of the patient from being made. Figure presents a 160 kb deletion of 4q25 detected by oligo array in a patient referred for developmental delay. Follow-up analysis performed on this patient's parents revealed that this alteration was de novo
in origin. This alteration deletes two genes, PAPSS1
. While mutations or alterations of these genes have not been associated with disease in humans, it has been shown that PAPSS1 plays a key role in post-translational modification and SGMS2 mediates the production of sphingomyelin [26
]. Thus, although the gene content and inheritance pattern of this deletion suggest a causative role in the patient's clinical features, a lack of published information linking the genes affected by this alteration with a distinct phenotype prevents a clear interpretation from being made based on only aCGH results. This type of unclear result, although more prominent with oligo platforms (4.2% by BAC vs. 7.1% by oligo), is an element of all aCGH analysis regardless of platform and accentuates the need for databases containing aCGH results in combination with phenotypic information. Although the number of characterized genetic disorders and genomic regions is rapidly increasing, the clinical consequences of alterations involving much of the genome still remain unclear.
Figure 7 Oligonucleotide microarray characterization of an interstitial deletion at 4q25. The zoomed-in microarray plots shows a single-copy loss of 15 oligonucleotide probes from the long arm of chromosome 4 at 4q25, approximately 159.6 kb in size (chr4: 108,834,399-108,994,048, (more ...)
The increase in the number of copy number alterations identified by higher-resolution whole-genome arrays underscores the need for a variety of tools to facilitate the interpretation of array results in a clinical diagnostic setting. We propose the use of an algorithm such as the one outlined here in conjunction with databases of normal population variants, clinically significant alterations, and those of unclear significance. Although such databases can provide invaluable context for the analysis of aCGH data, care must be taken by the diagnostician when comparing their data to pre-existing databases of copy-number variations. For example, data in the DGV are pooled from a variety of sources, platforms, and populations using a variety of different controls and without independent verification, and thus may not be appropriate for comparison in all situations. Furthermore, recent evidence suggests that most data in the DGV overestimate the size of the regions involved because they are dependent primarily on BAC array data, which has a tendency to overestimate the true size of small aberrations [28
]. Thus, the most useful CNV databases may be those generated by individual laboratories using identical reference controls and array platforms. Based on our experience, we have constructed a database of abnormal copy number aberrations identified by BAC and oligo aCGH in our laboratory and a database of copy-number variations thought to have no significance. Such databases are essential for understanding the various copy number aberrations identified by microarray analysis.
Genotype-phenotype correlations in a diagnostic setting must address a variety of factors including gene content, potential position effects, aberration size, and inheritance patterns. These factors often present conflicting evidence about the potential clinical significance of a rare alteration. For instance, the size of an abnormality is commonly used as justification for its proposed clinical consequences; however, this association is not always straightforward. High-resolution microarray analysis routinely detects abnormalities smaller than 500 kb that disrupt clinically significant genes and have clear phenotypic impact (Figure ); conversely, numerous examples of common copy-number variants have been observed that are relatively large but lie in regions with sparse gene content. In addition, although it is generally assumed that de novo
abnormalities are causative and inherited abnormalities are not, this is not always the case. There are a number of regions of the genome where both inherited and de novo
copy number alterations have been identified, some of which result in mild phenotypes that may be inherited from parents who have a milder or subclinical, presentation. For example, deletions of distinct regions of 1q21 have been associated with both thrombocytopenia absent radius (TAR) syndrome and a variable phenotype including microcephaly/macrocephaly, developmental delay, cardiac abnormalities, and schizophrenia [29
], but in many instances aberrations of these regions are inherited from phenotypically normal parents [32
]. Another example is the 16p11.2 region associated with a range of cognitive, developmental, and speech delays, behavioral issues, and autism, deletions and duplications of which can be inherited or de novo
]. In regions such as these, copy number changes may unmask recessive alleles or work in conjunction with various genetic modifiers, perhaps even other CNVs, to produce a clinical phenotype. Potentially, non-paternity may also confound genotype-phenotype correlation for copy number alterations in these complex regions of the genome. These reasons underscore the need for thorough databases of normal population variants and clinically significant alterations complete with genotype-phenotype correlations. Such databases expedite the process of determining the potential significance of copy-number alterations in a diagnostic setting; aid in the elucidation of new microdeletion/duplication syndromes and new regions of benign copy-number variation; and help reduce the burden of expensive, time-consuming, and difficult follow-up necessitated by the increased number of alterations of unclear clinical significance detected by microarray analysis.
] and others [37
] have shown that mosaicism can be detected at low frequencies of chromosomally abnormal cells using BAC-based aCGH; however, the ability of oligo platforms to reliably detect mosaic abnormalities has not yet been well established. Our current assessments demonstrate that aCGH using either BAC or oligo platforms can easily detect mosaicism of 30% or greater for a variety of alterations and that levels as low as 10% can be detected with both platforms under optimal conditions. In addition, our retrospective analysis showed that there is no significant difference between the two types of platforms in the number of mosaic abnormalities detected in a clinical diagnostic setting (p = 0.7066). However, BAC-based arrays may still have a greater ability to detect mosaic abnormalities present at very low levels (less than 20%), perhaps due to the routine use of dye-swap experiments which can be cost-prohibitive with oligo arrays but promote the visual identification of mosaic abnormalities. The sensitivity of the BAC array is demonstrated by the detection in three cases of abnormalities in only 10% of cells during the retrospective study, whereas the lowest level of mosaicism detected by our oligo array was 21% (Table ). The ability of an aCGH platform to detect mosaic abnormalities also depends largely on the effectiveness of the software used to analyze the data, as low-level mosaic alterations are difficult to identify using only visual inspection (Figure ). For this reason, it is important to select analysis software which facilitates the identification of mosaic alterations.
These data suggest high-resolution oligo-based aCGH detects a higher proportion of clinically significant abnormalities than BAC-based aCGH. Our results also demonstrate the ability of microarray-based CGH to reliably produce high-yield results in a clinical setting using differing platforms, array designs, and analysis algorithms, supporting the validity of array CGH as a first-tier diagnostic screening tool [38
]. Finally, the prevalence of copy number variants of unclear clinical significance detected on both platforms underscores the need for the development of readily accessible diagnostic tools in the form of databases of documented chromosome abnormalities to aid in the interpretation of microarray data.