Detection of pathogenic copy number variants in children with idiopathic intellectual disability using 500 K SNP array genomic hybridization

Chromosomal imbalance has been recognized as the most frequent cause of intellectual disability (ID) for 50 years [1–3]. Until recently, most of this genomic imbalance was diagnosed by cytogenetic analysis, but studies over the past few years have found that ID is caused by constitutional gains or losses of submicroscopic genomic segments even more often than by microscopically-apparent chromosomal aberrations [4–6]. Our ability to recognize these submicroscopic genomic changes, which are usually called copy number variants (CNVs), as the most frequent cause of ID depends on the use of Array Genomic Hybridization (AGH) (also known as array-comparative genomic hybridization, chromosomal microarray analysis, or copy number analysis). AGH can survey the entire genome for imbalance that is 1/100th the size of that detectable by conventional cytogenetic analysis.

Although AGH is now being used routinely as a clinical test for the identification of chromosomal imbalance in people with ID and other birth defects, controversy still exists regarding the most appropriate platform to use for this purpose. Initial clinical studies were done with arrays having a few thousand BACs distributed at 1-3 Mb intervals across the genome or with BACs targeted to regions where pathogenic submicroscopic deletions or duplications were known to occur. More recent studies have shown that arrays with higher resolution and genome-wide coverage provide better detection rates for pathogenic CNVs in children with ID and normal cytogenetic analysis [7–12]. Other methods have identified pathogenic CNVs that are too small to have been detected by the array platforms used in most AGH studies [13–15], so analysis at even higher resolution may be necessary to detect all pathogenic CNVs in children with ID.

Unfortunately, use of higher resolution AGH for detection of genomic imbalance is confounded because most CNVs are not pathogenic. Estimates of the mean number of apparently benign CNVs per person range from 6-824 in various studies, depending on the technology used to identify the variants and the size range used to define a CNV [16–25]. Sequencing of the complete diploid genomes of normal individuals has shown that the number of CNVs per person is actually even greater [26–29]. Distinguishing these benign CNVs from those that cause ID and other birth defects is the most serious challenge to routine clinical use of AGH.

We previously reported our findings using 100 K Affymetrix GeneChip^® AGH [10, 16] to perform a genome-wide survey of benign and pathogenic CNVs in 100 idiopathic ID trios, each comprised of an affected child and both unaffected parents. Here we describe the results of a study of 100 new idiopathic ID trios, as well as 54 of the trios tested previously, using 500 K Affymetrix GeneChip^® AGH. We found that higher resolution AGH detected a larger number of apparently pathogenic CNVs in both groups. Many benign CNVs, including at least one that was de novo, were also detected with the 500 K AGH, but it was possible to distinguish benign and pathogenic CNVs with confidence in almost all cases.

We performed AGH with 500 K Affymetrix GeneChip^® arrays on 154 children with idiopathic ID and on both parents of each affected child. Fifty-four of these trios (called "the 100 K cohort") had previously been studied with lower-resolution 100 K GeneChip^® AGH [10, 16]; the other 100 trios (called "the new cohort") were studied by AGH for the first time with the 500 K GeneChip^® arrays. Data were analyzed to determine copy number along the length of all chromosomes except the Y (for which there are no probes on either the 100 K or 500 K GeneChip^® arrays), and the findings for each child were compared with those for his or her parents. Autosomal CNVs seen in the child and in at least one parent were considered likely to be benign polymorphisms. Autosomal CNVs found in the child but not in either parent were evaluated by an independent method to confirm the presence of the CNV and its de novo occurrence. CNV calls on the X-chromosome in a female child were validated by an independent method if they appeared to have occurred de novo; CNV calls on the X-chromosome in a male child were validated by an independent method whether they appeared to be de novo or to have been inherited from the mother.

We found a total of 4577 hits (putative CNVs including at least 10 contiguous SNPs called by the SMD software with a p-value below 1 × 10^-8) in the 462 samples (154 trios) analyzed by 500 K GeneChip^® AGH. This is an average of about 10 hits per sample, which is an underestimate of the total number of benign CNVs present because of the stringent cutoff used to obtain a false discovery rate of less than 5% (see Methods).

The 4577 putative CNVs called in all 154 trios were subjected to further bioinformatic analysis (see Methods) to produce a final annotated list of 58 apparently de novo CNVs and two cases of mosaic trisomy that were called in 50 patients. These apparent genomic imbalances were subjected to validation by independent methods. Thus, an apparent de novo CNV call that required independent validation was made in about 1 child in 3. Thirty-three of these CNVs in 30 patients and both cases of mosaic trisomy were confirmed to be de novo by an independent method and are described in detail below. The other putative CNVs were found to be present in both of the parents as well as in the child (false negative AGH calls in the parents) in two instances or could not be confirmed to be present in the child (false positive AGH calls) in 22 instances. Altogether, false positive CNV calls were made in 21 (13.6%) of the 154 trios studied and false negative CNV calls were made in 2 (1.3%) of the 154 trios studied. In one other trio (Family 5202), an apparent de novo deletion of chromosome 14 called on AGH in the child was found by FISH to be a duplication of the region in both parents instead.

Using 500 K Affymetrix GeneChip^® AGH, we also confirmed the genomic imbalance that had previously been identified in 10 of the 54 ID trios from the 100 K cohort (Table 3 and Additional File 1: Supplemental Table S1). In addition, we identified and confirmed by FISH two de novo CNVs that were not called on the 100 K assay - a 1.6 Mb duplication of 8q23.2-23.3 in Patient 3890 and a 1.5 Mb deletion of 4p16.3 in Patient 4840 (Table 3 and Figure 1).

We found two instances of uniparental disomy (UPD), diagnosed by the occurrence of mendelian inconsistency in a region with a normal copy number of 2 [30], among the 100 ID trios in the new cohort studied by 500 K Affymetrix GeneChip^® AGH (Table 4 and Figures 2A and 2B). Patient 6904 has mosaic paternal isodisomy of most of the short arm of chromosome 11. Patient 1658 has maternal UPD 16, being heterodisomic for the central portion of the chromosome and isodisomic for both ends. Both cases were confirmed to be disomic with microsatellite markers (Figures 2C and 2D).

We judged the mosaic trisomy 9, both unbalanced reciprocal translocations, 11 of the other de novo deletions, and two of the other de novo duplications found in the new cohort to be pathogenic (Table 2). A mosaic 107 kb de novo deletion of chromosome 14 q11.2 (Patient 818) and a homozygous 89 kb deletion of the HLA-G region that resulted from transmission by parents who both carried heterozygous deletions of the same region (Patient 216) were judged to be benign variants. A 186 kb deletion of chromosome 21q22.11 (Patient 8619), a 362 kb duplication of chromosome 22 q11.21 (Patient 9979) and both cases of UPD (Patients 1658 and 6904) were of uncertain clinical significance.

Of the two de novo CNVs identified by 500 K Affymetrix GeneChip^® AGH in the 44 children with ID whose 100 K GeneChip^® AGH studies had been interpreted as normal (Table 3), one is likely to be pathogenic and the other is of uncertain clinical significance. A 1.5 Mb deletion of 4p16.3 in Patient 4840 is pathogenic because of its size, the inclusion of two genes that have been implicated in the Wolf-Hirschhorn syndrome (WHSC1 and WHSC2), overlap with known pathogenic CNVs, and a compatible clinical phenotype. The 1.6 Mb duplication of 8q23.2-23.3 in Patient 3890 is of unknown clinical significance. It includes no RefSeq genes but involves a large (1.6 Mb) genomic region, most of which has never been reported to be polymorphic in normal individuals.

Because its detection rate for pathogenic genomic imbalance is much higher than that of conventional cytogenetic analysis, a consensus has developed that AGH should be used clinically for the evaluation of patients with ID and other birth defects [31–38]. It is clear that AGH using "targeted" arrays that only include probes for genomic regions known to be involved in microdeletion or microduplication syndromes has substantially lower detection rates for CNVs that cause ID than AGH using arrays that provide genome-wide coverage [37, 39–41]. Beyond this, however, there is no agreement regarding the kind of array, the distribution of probes across the genome, or the resolution that is most appropriate for clinical use. Although BAC arrays were initially used, most clinical laboratories now prefer oligonucleotide arrays because high-quality platforms that produce consistent results are reliably available from commercial sources. In addition, the use of larger numbers of smaller probes on oligonucleotide arrays permits more precise delineation of the breakpoints of CNVs that are detected, which facilitates genotype-phenotype correlation and clinical interpretation. AGH with a SNP array provides the additional advantage of generating genotypes that can be used to verify family relationships and find uniparental disomy as well as a second method (in addition to hybridization intensity) for identifying genomic imbalance [10, 42–44].

We previously reported that 100 K Affymetrix GeneChip^® AGH is a robust platform for the detection of pathogenic CNVs in patients with ID [10]. Here we show that the detection rate of CNVs among such patients is higher with 500 K GeneChip^® AGH than with 100 K GeneChip^® AGH. We made about four times as many CNV calls overall with the 500 K platform as with the 100 K platform when using the same method of bioinformatic analysis in 54 trios studied with both technologies (Table 1). We also found 18 instances of pathogenic genomic imbalance in 16 of 100 children with ID and normal cytogenetic analysis studied by 500 K GeneChip^® AGH (Table 2), compared to 11 instances of pathogenic genomic imbalance in 11 of 100 similarly-ascertained children tested by 100 K GeneChip^® AGH in our previous study. Although the higher detection rate we observed with the 500 K platform may have occurred by chance because we just happened to include a few more patients with such genomic changes in the new cohort than in the 100 K cohort, we also found two additional de novo CNVs by 500 K GeneChip^® AGH among the 44 children whose 100 K AGH was interpreted as normal in our earlier studies (Figure 1 and Table 3).

We detected three apparently de novo CNVs smaller than 200 kb among the 100 trios tested with 500 K GeneChip^® AGH in the new cohort (a 107 kb deletion of chromosome 14 in Patient 818, a 186 kb deletion of chromosome 21 in Patient 8619, and an 89 kb deletion of chromosome 6 in Patient 216 that was actually a homozygous loss inherited from two heterozygous parents), but none of these CNVs was clearly pathogenic. The overall size distribution of pathogenic CNVs detected by 500 K GeneChip^® AGH among 154 children with idiopathic ID in the present study is similar to that observed by 100 K GeneChip^® AGH among 100 children with idiopathic ID whom we studied previously [10] (see Additional File 2: Supplemental Figure S1). The higher detection rate on the 500 K array therefore appears to be related more to better probe coverage in relevant genomic regions and an improved ability to distinguish CNVs from background noise, rather than to a capacity to identify much smaller pathogenic CNVs. This is illustrated in Families 3890 and 4840 (Figure 1), in which a 1.6 Mb duplication of 8q23.2q23.3 and a 1.5 Mb deletion of 4p16.3, respectively, are obvious on the 500 K AGH but were not called on the 100 K analysis. In retrospect, the 4p16.3 deletion in Patient 4840 can be seen on the 100 K AGH copy number plot despite the noisy data, but it was not called by either the automated analysis or visual inspection of these plots when the initial study was done. Our failure to detect the de novo duplication of 8q23.2q23.3 in Patient 3890 was probably caused by the noisy data in the father's study.

Distinguishing benign CNVs from those that cause ID and other birth defects is a critical issue in routine clinical use of AGH. Benign CNVs occur in all people and are a major source of genetic variation in the normal population [21, 27, 29]. Most apparently benign CNVs over 2 kb in size occur as polymorphisms with minor allele frequencies of at least 5% [21] and are inherited from a parent [21, 23, 45, 46].

Benign and pathogenic CNVs can usually be distinguished in patients with ID and other birth defects by inheritance and genotype-phenotype correlation [5, 33, 47]. In this study, we identified a mean of about 10 CNVs per subject in the 154 ID trios tested by 500 K AGH. The vast majority of these CNVs were characterized as benign because they were inherited from a normal parent. Genomic imbalance that occurs de novo in a patient with ID whose parents are normal is more likely to be pathogenic than genomic imbalance that was inherited unchanged from a normal parent. We performed AGH on both parents of every child with ID to determine the inheritance of the CNVs found in the child, but this is sometimes not possible in clinical practice. In such instances it is necessary to infer likely de novo occurrence by information obtained from populations that have previously been studied [19, 22–24, 48, 49]. Great care must be taken to avoid misinterpretation when this is done, especially if the available data were obtained with lower resolution AGH, the phenotypic characteristics of the comparison population are uncertain, reported polymorphic CNVs do not have exactly the same breakpoints as the CNV of interest, or the population frequency of a previously-reported CNV is unknown.

Compelling evidence that a CNV in a person with ID is pathogenic exists if the genomic imbalance is known to cause the patient's phenotype in other individuals, e.g., if a child with del 9q34.3 has features of the 9q subtelomeric deletion syndrome (Patient 523), a child with del 1p36.32p36.33 has features of the 1p36 deletion syndrome (Patient 9133), or a child with del 17q21.31 has features of the syndrome associated with this deletion (Patient 2106). Pathogenicity is also supported when a CNV includes a gene that is known to cause the patient's phenotype when inactivated (if the CNV is a deletion) or over-expressed (if the CNV is a duplication). On the other hand, a CNV is unlikely to be pathogenic if it involves a highly polymorphic region in which genomic loss (or gain, whichever is present in the patient) of the entire involved segment is known to occur in normal people.

If a direct genotype-phenotype correlation of this kind cannot be made in a particular case, certain genetic features of the CNV may provide clues to its pathogenicity. CNVs that are larger and those that involve gene-rich regions are more likely to be pathogenic than CNVs that are smaller and involve only gene-poor regions [5, 47]. In addition, clinical experience suggests that deletions are more likely to be pathogenic than duplications [47]. The genetic content of a CNV may also make pathogenicity more or less likely. For example, involvement of a gene that lies within a pathway that is known to contain other dosage-sensitive genes associated with a similar phenotype strengthens the possibility of pathogenicity, while a CNV that does not contain any genes that are expressed in relevant tissues during embryogenesis is unlikely to be pathogenic for ID.

There are, of course, exceptions to each of these "rules". Some benign CNVs arise de novo[21, 23, 45, 46, 50], as appears to have occurred in the de novo deletion of chromosome 14q11.2 we found in Patient 818. The 107 kb region involved is highly polymorphic and contains several T-cell receptor variable region genes. On the other hand, some CNVs that are inherited from a normal parent are pathogenic for ID. Examples include maternal transmission of a UBE3A deletion to a child with Angelman syndrome [51], maternal transmission of a MECP2 duplication to a son [52], and CNVs such as dup 22q11.2 [53, 54] or del 1q21.2 [55] that can cause ID but exhibit incomplete penetrance.

Although large (> 250 kb) CNVs are often pathogenic, they may be benign [19, 29]. Most benign CNVs are small (< 250 kb) [19, 27, 29], and it seems probable that the smaller a CNV, the more likely it is to be benign. Nevertheless, no clear size distinction exists between benign and pathogenic CNVs. We found pathogenic CNVs as small as the 298 kb deletion of 9q34.3 in Patient 523 in this study, and others have reported even smaller pathogenic CNVs [13–15, 56–58]. Expression patterns, functional annotation and animal models can provide important clues to pathogenicity in some cases, but without knowledge of the phenotypic effects of a copy number alteration in humans, one can rarely, if ever, be certain whether a novel gain or loss of a particular genomic region can produce ID or other birth defects.

In our 500 K AGH study of 154 ID trios, 58 de novo CNVs were called by bioinformatic analysis, and 33 of these CNVs were confirmed and shown to be de novo by an independent method. Because we could assess the phenotypes of our patients in detail and correlate the findings with those obtained by AGH, we were able to determine with confidence whether the genomic imbalance we observed was pathogenic or not in every case studied except three (Tables 2 and 3). Such genotype-phenotype correlation is critical to determining the effects of novel CNVs detected by AGH in patients with ID.

A CNV of uncertain clinical significance was encountered in three (1.9%) of the 154 trios analyzed in this series - a 362 kb duplication of 22q11.21 in Patient 9979 (Table 2), a 186 kb deletion of 21q22.11 in Patient 8619 (Table 2) and a 1.6 Mb duplication of chromosome 8q23.2q23.3 in Patient 3890 (Table 3). This rate of CNVs of uncertain clinical significance is similar to that reported in large series of patients with ID and other birth defects studied by AGH with "targeted" chips [31, 35, 59].

We were uncertain of the clinical significance of either case of UPD that we detected. Although only a few liveborn children with UPD 16 have been recognized, the reported experience does not suggest that UPD 16 can cause the abnormalities observed in Patient 1658 [60, 61]. Paternal UPD 11p15 can produce Beckwith-Wiedemann syndrome [62], but this phenotype is very different from that observed in the affected child in Family 6904. However, as both of these cases involved isodisomy of a portion of the chromosome, we cannot rule out the possibility that the abnormal phenotype was produced by homozygosity for a recessive mutant allele [63, 64]. Although the detection of UPD in addition to alterations of copy number is a theoretical benefit of using an array that includes probes for SNPs, the clinical utility of genome-wide screening for UPD in patients with idiopathic ID and other birth defects is uncertain.

We detected more pathogenic CNVs with 500 K AGH than with 100 K AGH, but some CNVs that were present among our patients were not detected using the 500 K assay. For example, our 500 K GeneChip^® analysis failed to identify a pathogenic 83 kb deletion of chromosome 16p13.3 (3,862,993 bp to 3,945,522 bp) involving the CREBBP gene (Patient 5121). This de novo deletion was found by AGH on the Agilent^® 244 K platform and was confirmed by MLPA. The patient is an 8 year-old boy whose clinical features are characteristic of the Rubinstein-Taybi syndrome, which has been associated with deletions and other mutations of CREBBP in other patients [65, 66]. The 83 kb genomic region deleted in our patient is poorly represented on the 500 K GeneChip^® arrays, with a total of only 15 SNPs. SNP arrays have uneven genomic coverage, and the addition of non-polymorphic oligonucleotide probes to the design of arrays like the one used in this study has been shown to provide substantially better detection of CNVs [21].

Affymetrix GeneChip^® 500 K array genomic hybridization performed in individuals with idiopathic intellectual disability detected pathogenic genomic imbalance in 10 of 10 patients in whom 100 K GeneChip^® array genomic hybridization found genomic imbalance, 1 of 44 patients in whom 100 K GeneChip^® array genomic hybridization had found no abnormality, and 16 of 100 patients who had not previously been tested. Further improvements in array design, ongoing improvements in AGH software, and continuing enhancement of resources like DECIPHER https://decipher.sanger.ac.uk/ and the Toronto Database of Genomic Variants http://projects.tcag.ca/variation/ are helping to establish AGH as the primary clinical tool for recognition of genomic imbalance that causes ID and other birth defects [33, 34, 37, 41, 67–70]. It seems likely, however, that no perfect AGH platform for detection of pathogenic CNVs may ever exist and that effective clinical interpretation of these studies will continue to require considerable skill and experience [33, 50, 71].