High-resolution SNP array analysis of patients with developmental disorder and normal array CGH results

The diagnostic yield of microarray comparative genomic hybridizations (array CGH) has already proven to exceed that of cytogenetic methods, except when it comes to balanced rearrangements. A consensus statement suggests that microarrays should be used as the first line of testing for developmental disorders of unknown cause [1]. However, as our previous study shows, approximately 80 % of patients with a developmental disorder of unknown cause (mental retardation and/or malformations and/or neurological disorders) remain undiagnosed even by array CGH analysis (44 K, 180 K, or 244 K) [2]. Thus, other methods are clearly needed to define the pathogenic mechanisms. It is plausible that small copy number variants (CNVs) may go undetected if the probe coverage is limited, as it may be in low-resolution arrays. By increasing the resolution, one would thus expect to detect increasingly smaller pathogenic CNVs.

The frequency of uniparental disomy (UPD) in newborns is reportedly ~0.029% [3]. Around 1,100 cases of whole chromosome UPD and some 120 reports on segmental UPD are described in the literature [4]. Some recessive diseases are expressed in children who have inherited the mutation form a single carrier parent [5]. The explanation for this is a meiotic or very early mitotic recombination event between parental homologous chromosomes causing segmental UPD of the genomic segment containing the mutation, and thereby causing a reduction to homozygosity.

As the market is flooded with new arrays, most having an increased resolution and a promise of ever higher detection rates, the question remains what the added value is of these ultra-high resolution arrays. To test this, we used an array with 1.8 million probes and an average resolution of 0.7Kb to analyze samples of 35 patients with developmental disorder of unknown cause, normal karyotype, and normal array CGH results by use of Agilent 44 K, 180 K, or 244 K platforms (Agilent Technologies, Santa Clara, CA, USA).

The immense amount of data created by the Genome-wide human SNP array 6.0 warrants filtering for clear interpretation. The Genotyping Console software identified between 200–1000 changes per patient (Figure 1), depending on the technical quality of the result. More changes were detected in samples with lesser quality. After filtering, based on the uniqueness of the CNVs or regions of homozygosity compared to the references, each patient presented 8–20 unique changes (≥90% CNVs) on average. In samples with lesser quality, ≥100 unique changes were detected. Further research on gene content and phenotypes previously mapped to these regions revealed 23 CNVs and 28 regions of homozygosity that putatively correlated with the clinical phenotype in 26 patients. Nine patients had no CNVs or regions of homozygosity spanning known genes or genes known to associate with a disease that correlated with the patient’s phenotype, and their results were thus considered normal. The associated phenotypes related to the aberration found in 26 patients were further evaluated by the patients’ clinicians, and the frequencies of the observed CNVs were monitored in the Finnish population cohort (n = 2,065). As a result, in four patients, a region of allelic homozygosity was considered a potential candidate for causation of their clinical state (Table 2). No CNVs were considered candidates after clinical evaluation.

We further attempted verification of four potential segmental UPDs by microsatellite marker analysis. In all the four cases we observed two distinct alleles with at least some of the more informative multiallelic markers, suggesting that the observed LOHs were most probably caused by the same allelic SNP haplotypes being inherited from both the parents (Table 2).

Further comparison, of all patient SNP array data to the normal data from unaffected individuals, revealed 21 regions of clustering (≥40 % frequency) of allelic homozygosity to specific locations of the genome (Table 3). However, no significant differences were detected between the frequencies of clustered regions in patients and the unaffected relatives in this small set of samples.

In our previous study of 150 patients with developmental disorders of unknown cause and a normal karyotype, we were able to identify a (potential) causative aberration in 18% of the patients, by using a 44 K or 244 K array CGH platform [2].

To determine whether, by increasing the resolution, any additional copy number changes or regions of UPD could be detected, we studied 35 patients with a normal array CGH result.

Allelic homozygosity is typically caused by linkage and co-segregation of certain blocks of DNA, termed haplotypes [8]. In a small founder population, such as the Finnish population, the founder effect increases the likelihood that the parents will have the same haplotype, and is as such not a segmental UPD [10]. True segmental UPD is typically due to a duplication in one chromosome and a reciprocal deletion in the other; or the fertilization of a disomic and monosomic gamete, somatic crossing over and subsequent trisomic rescue [11]. If the genomic segment harbours a recessive mutation, which subsequent to UPD will be present in two copies (reduction to homozygosity), it causes a recessive disease. Equally relevant, the segment can be preferentially imprinted, causing complete silencing, which is the equivalent of a deletion. Such presentations of recessive syndromes that are inherited from one normal parent are known in some 40 patients [12].

We were interested to see whether the regions of allelic homozygosity detected by the SNP array were in fact segmental UPDs and associated with an autosomal recessive disease. We were, however, unable to confirm these results and thus the SNP array did not yield more molecular diagnoses in this study of developmental disorders of unknown cause. This may be due to the fact that all patients had previously been studied by another high-resolution array, with 8.9Kb (244 K), 13Kb (180 K), and 35Kb (44 K) theoretical resolutions. Although several new CNVs, previously undetected by the array CGH platform, as well as regions of homozygosity were detected, the pathogenic relevance of these changes were considered insignificant in correlation to the patient’s phenotype. It is, however, possible that changes dismissed in this study are pathogenic by means of spanning genomic segments that do not directly involve disease genes, but rather their regulatory elements.

Our results differ from previously published studies using similar research settings. Bernardini et al. (2010), using the SNP 6.0 array platform with a 75Kb cut-off value for detected CNVs, reported potentially pathogenic CNVs in 6% of patients with normal array CGH result (44 K) [13]. Mannik et al. (2010), using another SNP array with a 50Kb resolution, reported a 23% detection rate in patients with a normal karyotype [14]. Bernardini et al. and Mannik et al. had higher detection rates than this study; perhaps as their first-line of array analysis was, at least partly, done by a lower-resolution method (35Kb and 50Kb respectively) compared to the first-line of detection in this study (8,9Kb, 13Kb and 35Kb). Also, it is important to note that the statistical power is limited by the small size of our patient cohort, and thus results are not entirely comparable with Bernardini and Mannik’s.

UPDs have not been reported in either of the above mentioned studies. However, in a study of 117 patients with a normal karyotype analysed using the 250 K SNP array (Affymetrix, Santa Clara, CA, USA), pathogenic CNVs were detected in 18 patients, and potentially pathogenic segmental UPDs ≥5 Mb in 5, verified by microsatellite marker analysis [15]. The presence of UPDs was also evaluated in another study of 120 patients, using a 500 K SNP array platform [16]. In that study they were unable to verify UPD in any of 121 detected regions of homozygosity in 72 patients with developmental disorder of unknown cause. In addition, in a study of 100 patients with developmental disorder, using a 500 K SNP array, two patients were found to have UPD, the clinical significance of which remained unclear [17]. Thus, UPDs are detectable using SNP arrays, but their clinical significance is difficult to interpret.

Although there is a clear added value of high-resolution arrays in various fields of genetics, it seems that there is a limit to how much the yield can be increased by increasing the theoretical resolution of the analysis platform. Despite the fact that the SNP array has increased probe spacing compared to the 44 K, 180 K, and 244 K array, and is able to detect more CNVs and regions of homozygosity, interpretation of the vast amount of data and pinpointing of pathogenic changes is difficult. One benefit of using a SNP based platform is the possibility to detect UPDs; however these are relatively rare findings.

This study had a limited number of patients, and so it can only be said that for this study group the optimal yield was conceived when using a resolution of approximately 9Kb [2]. Increasing the resolution beyond that did not confer more diagnoses. It must be emphasized, however, that a larger patient cohort needs to be studied in order to draw final conclusions on the added value of an ultra-high resolution array compared to others. Furthermore, as several reports have shown, some pathogenic aberrations span only a few exons and for detecting such small changes the sensitivity of the SNP 6.0 platform is adequate [18]. It is, however a challenge to filter results correctly and so for diagnostic purposes the choice of platform needs to be carefully considered. Patients with developmental disorders of unknown cause and normal array results may also harbour such small genomic changes (i.e. mutations and unbalanced rearrangements) that are difficult to interpret using microarrays and would require higher resolution methods, such as whole-exome sequencing. Interestingly, a recent study suggests that 80% of patients with a developmental disorder of unknown cause and normal array results can be diagnosed using whole-exome sequencing [19]. Only the future can tell.