Meeting the challenge of interpreting high-resolution single nucleotide polymorphism array data in prenatal diagnosis: does increased diagnostic power outweigh the dilemma of rare variants?
Correspondence: Dr Howard R Slater, VCGS, Murdoch Childrens Research Institute, Royal Children's Hospital, Parkville, 3052, Victoria, Australia. Email email@example.com
Several studies have already shown the superiority of chromosomal microarray analysis (CMA) compared with conventional karyotyping for prenatal investigation of fetal ultrasound abnormality. This study used very high-resolution single nucleotide polymorphism (SNP) arrays to determine the impact on detection rates of all clinical categories of copy number variations (CNVs), and address the issue of interpreting and communicating findings of uncertain or unknown clinical significance, which are to be expected at higher frequency when using very high-resolution CMA.
Prospective validation study.
Tertiary clinical genetics centre.
Women referred for further investigation of fetal ultrasound anomaly.
We prospectively tested 104 prenatal samples using both conventional karyotyping and high-resolution arrays.
Main outcome measures
The detection rates for each clinical category of CNV.
Unequivocal pathogenic CNVs were found in six cases, including one with uniparental disomy (paternal UPD 14). A further four cases had a ‘likely pathogenic’ finding. Overall, CMA improved the detection of ‘pathogenic’ and ‘likely pathogenic’ abnormalities from 2.9% (3/104) to 9.6% (10/104). CNVs of ‘unknown’ clinical significance that presented interpretational difficulties beyond results from parental investigations were detected in 6.7% (7/104) of samples.
Increased detection sensitivity appears to be the main benefit of high-resolution CMA. Despite this, in this cohort there was no significant benefit in terms of improving detection of small pathogenic CNVs. A potential disadvantage is the high detection rate of CNVs of ‘unknown’ clinical significance. These findings emphasise the importance of establishing an evidence-based policy for the interpretation and reporting of CNVs, and the need to provide appropriate pre- and post-test counselling.
Conventional karyotyping has been the ‘gold standard’ for prenatal chromosome analysis for 40 years, typically prompted by advanced maternal age, a maternal serum screening test result indicating increased risk of fetal aneuploidy, or by an abnormal finding on ultrasound examination.
Molecular karyotyping by chromosomal microarray analysis (CMA) for investigation of intellectual impairment, congenital abnormality, and developmental delay in children has afforded very significant improvements in diagnostic success, of the order of 10–15%. Despite clear recommendations to replace conventional karyotyping using microscopy by molecular karyotyping, and the rapid transition to this new methodology, there has been much less enthusiasm for prenatal investigation of chromosome abnormality. The great advance afforded by microarrays is the one to two orders of magnitude increase in resolution compared with conventional microscopy. This has led to the discovery of a multitude of hitherto unrecognised recurrent and novel submicroscopic pathogenic copy number variations (CNVs),[3, 4] even at the gene or exon level,[5-7] and includes the identification of uniparental disomy (UPD) and cryptic mosaicism (i.e. mosaic events that are difficult to investigate because of genomic size, level of mosaicism, or involvement of allelic imbalance) if single nucleotide polymorphism (SNP) genotyping is used together with copy number data. Microarray analysis also very efficiently detects common (polymorphic) and rare copy number variation that, collectively, is ubiquitous in the human genome, and the frequency of which is inversely correlated with CNV length. Because of the incomplete description of the variome, a catalogue of all genetic variation in the human genome and its associated phenotypes, CMA also detects novel and rare CNVs that are of unknown or uncertain clinical significance: it is the interpretation of these and the resulting counselling dilemmas that present the main impediment to the adoption of prenatal CMA as a ‘stand alone’ first-tier test.
Several prenatal studies have been published using relatively low-resolution microarrays for the investigation of ultrasound anomaly.[9-13] The detection rate of CNVs of ‘uncertain’ and ‘unknown’ significance has not presented significant challenges for the clinical application of CMA in this group of women. More recently, data have emerged from several studies about detection rates of ‘clinically significant’, ‘unknown’, and ‘uncertain’ CNVs in pregnancies with a low a priori risk of fetal chromosomal abnormality.[11, 13, 14] This has prompted a discussion of the efficacy of CMA as a ‘stand alone’ first-tier prenatal test for all indication groups.
Logically, the use of very high-resolution genome-wide microarrays, that is, with resolution down to at least 50 Kb, is expected to increase the detection rate of smaller CNVs, and correspondingly the number of novel and rare CNVs of unknown or uncertain significance. We report the findings of a prospective pilot study using very high-resolution SNP microarrays, in parallel with conventional karyotyping, where the resolution is no higher than 5 Mb, for the investigation of 104 fetuses showing ultrasound anomaly.
This study aimed to extend the analytical sensitivity to the smallest pathogenic CNVs detectable using very high-resolution SNP arrays, and to determine the impact on detection rates of clinically significant CNVs relative to conventional karyotyping. A second objective was to establish the detection rate of CNVs of uncertain and unknown significance, and to develop policies to address the issue of interpreting and communicating these classes of CNVs.
Women and samples
Molecular karyotyping by microarray was performed on 104 amniotic fluid, chorionic villus, or fetal blood samples for the investigation of fetal anomalies detected by ultrasound examination. There were 28 cases showing multiple fetal abnormalities, 24 with a brain abnormality, 16 with a skeletal abnormality, six with intrauterine growth retardation, five with hydrops, three with an abdominal wall defect, two with a neural tube defect, and ten with isolated miscellaneous abnormalities. Of the 104 cases, there were four with sonographic markers only, that is, two with nuchal thickening, one with echogenic kidneys, and one with multiple, undisclosed, soft markers. All cases were referred for clinical genetic testing from two tertiary level obstetrics centres to the VCGS Cytogenetics Laboratory, Melbourne, Australia. Women received pre-test genetic counselling about CMA from the sonologist performing the amniocentesis/chorionic villus sampling: this included written information about chromosome microarray analysis and the range of chromosome changes detectable by the arrays and reported by the laboratory. All gave written informed consent for the testing. The study was limited by the lack of postmortem assessments on the pregnancy terminations and the absence of postnatal follow-up on the live-births.
Cell culture and DNA extraction
All amniotic fluid and chorionic villus samples were cultured using standard techniques, and were prepared for G-band karyotyping according to the laboratory protocols. DNA was isolated directly from 5–10 mL of amniotic fluid, 1–5 mg of chorionic villus, or fetal blood samples using NucleoSpin Blood extraction kits (Macherey–Nagel, Düren, Germany). DNA was obtained from cultured cells for cases with insufficient DNA from the direct extraction method.
The DNA samples were run on either Affymetrix 2.7M arrays (93 samples; Affymetrix, Santa Clara, CA, USA) or Illumina HumanCytoSNP-12 v2.1 arrays (11 samples; Illumina, San Diego, CA, USA) using the manufacturers' protocols. These provide an effective analytical resolution of at least 200 Kb when a ‘call threshold’ of 20 (i.e. one probe per 10 Kb backbone coverage) and 50 (i.e. one probe per 1 Kb backbone coverage) consecutive, aberrant probes are applied to Illumina and Affymetrix array data, respectively. Our settings are based on substantial experience in analysing the same type of data from postnatal investigations (Bruno DL unpublished data; Bruno et al.). The resolution of these analyses is comparable with that recommended for postnatal microarray testing, allowing the direct comparison of findings from this prenatal cohort with those detected in patients with neurodevelopmental and/or congenital abnormalities.[2, 16] The Affymetrix 2.7M platform was preferred because of the single sample processing format.
Image capture, initial analysis, and quality assessment of the array data were performed with the Affymetrix GeneChip Scanner 3000 and command console.
Illumina HumanCytoSNP-12 analysis was performed using karyostudio software (Illumina). All abnormalities identified were further characterised by visual inspection of the chromosomal plots displaying logR and BAF data.
For each of the arrays, the quality control of the probe intensity data was exercised by the arbitrary definition of thresholds for the number of consecutive probes necessary for accurate CNV calling. These were benchmarked to the quality of individual array data (i.e. the variance of logR data for all autosomal probes), and reflected an average practical detection resolution of 50 Kb (Affymetrix) and 200 Kb (Illumina).
Interpretation of the clinical significance of CNVs and long continuous stretches of homozygosity (LCSH) was primarily based on gene content. This was determined using the University of California, Santa Cruz (UCSC) browser based on NCBI build 36. Where there was information available in the literature and/or public databases of CNVs, such as Decipher (http://decipher.sanger.ac.uk), CHOP (http://cnv.chop.edu), ISCA (http://www.iscaconsortium.org), and DGV (http://projects.tcag.ca/variation), they were classified within a spectrum from ‘benign’ to ‘pathogenic’. The significance of some CNVs could not be determined, i.e. ‘CNVs of unknown significance’; these were completely novel or rare CNVs that involved at least one gene, but for which there were no prior literature or public database reports to suggest pathogenicity.
Briefly, we defined ‘pathogenic’ changes as those that have well-established associations in the peer-reviewed literature with defined syndromes or disorders. ‘Likely pathogenic’ changes are usually large (>500 Kb) deletions or duplications containing multiple genes, at least one of which has some evidence in the literature indicating relevance to the clinical disorder. Changes where there is insufficient or unconvincing evidence for assignment of pathogenicity are classified as ‘of uncertain clinical significance’: these include recurrent CNVs that are known risk factors for neurocognitive disorders (examples are: 16p13.1 deletion/duplication; 15q11.2 deletion; 15q13.3 deletion). Changes that are ‘likely benign’ are usually rare, are often (but not always) inherited from a parent without a comparable phenotype, and either contain no genes or genes in which the known function has no apparent relevance to the phenotypic abnormality under investigation. ‘Benign’ changes are those that have been reported in the literature and curated databases (i.e. CHOP [http://cnv.chop.edu], ISCA [https://www.iscaconsortium.org], and DGV (http://projects.tcag.ca/variation]) as benign variants with no clinical significance: these include polymorphisms (i.e. with minor allele frequency >1%).
Long continuous stretches of homozygosity (LCSH) were interpreted as clinically significant if they were subsequently shown by follow-up testing to be a segment of uniparental isodisomy mapping to a clinically significant chromosome/locus, or if a pathogenic mutation in a gene of interest was found by clinically directed Sanger sequencing. A LCSH was considered of possible clinical relevance if it contained a gene of known relevance, but where Sanger sequencing had not yet been performed. All other LCSHs were designated of unknown significance, most likely being regions of identity by descent, but which nonetheless may be associated with an increased risk of recessive disease.
These criteria are consistent with those recommended by the American College of Medical Genetics for reporting of postnatal constitutional copy number variants, with the notable addition of the ‘unknown significance’ category that recognises the different responsibilities and consequences of reporting in the prenatal diagnosis area.
Equivocal findings, comprising primarily novel and rare changes, were interpreted by a team of clinical and laboratory geneticists with substantial collective experience of prenatal diagnosis and reporting of postnatal CMA results. CNVs classified as being ‘of uncertain clinical significance’ were reported. ‘Benign’ and ‘likely benign’ CNVs were not reported. This is consistent with our postnatal reporting policy. LCSH >5 Mb in length that involved a chromosome with a clinically relevant imprinted gene/locus (i.e. chromosomes 7, 11, 14, and 15) were reported. Samples showing extensive LCSH or an isolated LCSH on a chromosome other than the ones described above were assessed in light of any autosomal recessive disorders that may have been of relevance to the fetal anomaly.
Reports identifying CNVs that were ‘pathogenic’, ‘likely pathogenic’, or ‘of uncertain’ clinical significance were reported with recommendations for genetic counselling by clinical geneticists and genetic counsellors in the referring hospitals. Information regarding the gene content of LCSH regions was made available on request to the geneticists counselling the patients.
Follow-up parental testing
Where it was deemed to be potentially useful as part of the interpretation process, and where consent was given, follow-up parental sampling was undertaken. The origin of ‘pathogenic’ and ‘likely pathogenic’ CNVs was investigated to determine the probability of recurrence.
Concordant ‘pathogenic’ or ‘likely pathogenic’ CMAs and conventional karyotype results were found in three of the 104 fetuses with ultrasound anomalies (3%; Table 1). The two possible types of discordance were seen. Firstly, discordant results, where the conventional karyotype was normal but the CMA result was ‘pathogenic’ or ‘likely pathogenic’, were found in seven of the 104 fetuses (7%; Table 1). Secondly, discordant results, where the CMA result was normal but the conventional karyotype showed an abnormality, were found in four of the 104 fetuses (4%); all four, however, were interpreted as either ‘benign’ or ‘likely benign’ imbalances.
Table 1. ‘Pathogenic’ and ‘likely pathogenic’ findings
|1||Fetal blood||Congenital diaphragmatic hernia, unilateral cleft lip/palate, ↑ NT (1:3 risk for T21 on first-trimester screen)||46,XY||arr Xq26.2(131,336,145-132,612,743) × 0 mat||Maternal||Pathogenic. Hemizygous GPC3 loss in males is associated with Simpson–Golabi–Behmel syndrome (MIM:312870)||Termination|
|2||Cultured AF||Cardiac and abdominal heterotaxy, complex congenital heart disease, severe cerebral ventriculomegaly (NT 9mm)||46,XX||arr 1q43q44(238,168,460-247,169,378) × 1 dn|| de novo ||Pathogenic. Terminal 9–Mb deletion containing the described 1q43q44 microdeletion syndrome region (MIM:612337)||Termination|
|3||Cultured AF||Radial aplasia, ulnar hypoplasia, short bowed humeri, flexed wrists||46,XY||arr 1q21.1(144,095,255-144,544,340) × 1 pat||Paternal||Pathogenic. 1q21.1 deletion is typical of that observed in cases of TAR syndrome (MIM:274000) Chr1: 145,507,646 (rs139428292) 5′ UTR G/A subsequently detected in mother||Termination|
|4||Uncultured AF||Truncus arteriosis||46,XY||arr 22q11.21(17,370,127-19,790,008) × 1 dn|| de novo ||Pathogenic. 22q11.21 deletion associated with Di George/VCF syndrome (MIM:188400)||Termination|
|5||Uncultured AF||Radial club hands, talipes equinovarus, polyhydramnios, micrognathia (1:20 risk for T21 on first-trimester screen)||46,XX||arr 14q11.2q32.33(19,432,659-106,313,072) × 2 hmz||Paternal (UPD)||Pathogenic. Confirmed as UPD 14 of paternal origin by parental–fetal microsatellite studies (MIM:608149)||Termination|
|6||Uncultured AF||Microcephaly, IUGR, EIF, SUA, outflow tract abnormality||45,XX,-13/46,XX,r(13)(p10q32)||arr 13q32.3q34(99,597,424-114,121,221) × 1 dn|| de novo ||Pathogenic. 14.5–Mb terminal deletion in region 13q32.3q34 containing 71 RefSeq (but NOT including ZIC2) and overlapping the 13q deletion syndrome (PMIDs:19022413, 22366306)||Termination|
|7||Uncultured AF||Dandy walker malformation, hydrocephalus||46,XY||arr 18p11.32(512,467-668,848) × 3 mat, 18q23(74,636,874-74,860,903) × 3 mat||Maternal||Likely pathogenic. Extended familial studies (phenotype segregates with a complex chr18 structural rearrangement)||Termination|
|8||Cultured AF||NT 6.7 mm (1:5 risk for T21 on first-trimester screen), VSD, pelvic kidney||46,XX||arr 3p26.3p26.1(56,668-8,616,087) × 3 dn, 18p11.32p11.31(50,739-4,990,626) × 1 dn|| de novo ||Likely pathogenic. 8.6–Mb duplication from region 3p26.3p26.1 involving 20 RefSeq genes; and 5.0–Mb deletion in 18p11.32p11.31 involving 27 RefSeq genes||Termination|
|9||Uncultured AF||Cerebellar hypoplasia, mild cerebral ventriculomegaly, hypoplasia aortic arch and left ventricle||46,XX,del(5)(q?11.2q?13),ins(11;5) (q?21;q?13q23.1)||arr 5q11.2(54,740,031-56,650,248) × 1, 5q12.1(59,712,894-61,390,361) × 1, 5q12.1q13.1(62,391,430-66,979,840) × 1, 5q13.1(68,018,541-68,386,287) × 1, 5q13.2(70,920,664-72,243,273) × 1, 5q13.2(72,535,116-72,778,206) × 1, 5q13.3q14.1(76,151,757-76,421,209) × 1, 5q14.1(76,468,591-76,738,925) × 1, 5q14.1(77,550,070-77,713,885) × 1, 5q23.1(115,786,779-115,956,712) × 1, 11q22.1(98,589,690-100,823,748) × 1|| de novo ||Likely pathogenic. Nine interstitial deletions in 5q11.2-14.1 involving a total of 10.8 Mb of DNA sequence, a 0.2-Mb deletion in 5q23.1, and a 2.2-Mb deletion in 11q22.1.||Termination|
|10||Cultured AF||Cardiac and renal abnormalities||46,XX,(add)(15)(q?25)||arr 15q25.3q26.3(86,242,719-98,492,328) × 3 dn, 15q26.3(98,492,327-100,208,896) × 1 dn, 18p11.32p11.22(50,739-9,515,286) × 3 dn|| de novo ||Likely pathogenic. 12–Mb duplication from 15q25.3q26.3 (includes IGF1R gene), 1.5–Mb terminal deletion in 15q26.3, and a 9.5-Mb duplication from 18p11.32p11.22||No information|
The genome imbalances found by CMA, conventional karyotyping, or both were classified as ‘pathogenic’, ‘likely pathogenic’, ‘benign’, ‘likely benign’, or ‘unknown’. With the exception of ‘benign’ CNVs, these are described as in Tables 1,2, and 3.
Table 2. Copy number changes of ‘unknown clinical significance’ for which parental testing was performed or recommended
|7||Fetal skin||Dandy Walker malformation, hydrocephalus||dup||19||39 636 223||40 178 213||542||UBA2, WTIP, SCGB1B2P, SCGB2B2, SCGB2B3P, ZNF302, ZNF181, ZNF599, LOC400685, LOC100652909, ZNF30, ZNF792||12||Novela||Maternal||Termination|
|15||Fetal blood||Micrencephaly Global developmental brain anomaly||dup||1||57 680 374||57 890 732||210||DAB1||Intragenic (exon 3)||Novela||Maternal||Termination|
|dup||11||70 991 420||71 651 998||661||FAM86C1, FAM86C1, DEFB108B, LOC100133315, LOC100129216, RNF121, IL18BP, NUMA1, MIR3165, LRTOMT, LAMTOR1, C11orf51, FOLR3, FOLR1, FOLR2, INPPL1, PHOX2A||17||Novela||Paternal (father is a mosaic carrier)|
|16||Cultured amniocytes||Inferior vermian hypoplasia, Blake's pouch cyst||del||4||77 213 524||77 999 119||786||ART3, ART3, NUP54, SCARB2, FAM47E, FAM47E-STBD1, STBD1, CCDC158, SHROOM3, MIR4450||10||Novela||Paternal (father had isolated congenital diaphragmatic hernia)||Liveborn infant, brain changes confirmed postnatally. Normal development at 18 months|
|17||Uncultured AF||Persistent nuchal oedema/cystic hygroma||trip||4||170 298 544||170 737 045||439||SH3RF1, NEK1||2||Novela||Unknown (parental consent not given)||Termination|
|18||Cultured amniocytes||Obstructive hydrocephalus (female)||dup||X||46 047 921||46 728 580||681||ZNF673, ZNF674, LOC401588, CHST7, SLC9A7, RP2, CXorf31, PHF16||8||Three smaller, overlapping cases in ISCA||Maternal||Termination|
|19||Uncultured AF||Hypoplastic left heart||dup||18||3 362 404||4 290 761||928||TGIF1, DLGAP1, FLJ35776, LOC201477, LOC284215||5||Novela||Paternal||Termination|
|20||Uncultured AF||Bilateral echogenic kidneys||del||22||21 328 049||21 976 721||649||MIR650, IGLL5, RTDR1, GNAZ, RAB36, BCR, FBXW4P1||7||Recurrent with several cases in local and public databases||Paternal (father is healthy and had no congenital anomalies)||Liveborn infant.cystic renal dysplasia. Otherwise healthy|
Table 3. Summary of copy number changes of unknown clinical significance that were not reported, on the basis of rigorous assessment of the available evidence
Six CNVs have a well-established association with a clinical syndrome or disorder: a maternally inherited 1.3–Mb deletion in chromosome region Xq26.2 removed the gene GPC3 associated with Simpson–Golabi–Behmel syndrome (MIM 312870; case 1); a 9-Mb terminal deletion spanning the 1q43-q44 deletion syndrome region (MIM 612337; case 2); a paternally inherited 0.5-Mb deletion in region 1q21.1 typical of that observed in thrombocytopenia with absent radius (TAR) syndrome (MIM 274000; case 3); a 2.4-Mb deletion in region 22q11.2 is indicative of Di George/velo-cardio-facial (VCF) syndrome (MIM 188400; case 4); an LCSH involving the whole of chromosome 14, which was shown by follow-up studies to be uniparental isodisomy 14 of paternal origin (case 5); and a 14.5–Mb terminal deletion in region 13q32.3q34 containing 71 RefSeq genes (but not including ZIC2), showed by karyotyping to be caused by a ring chromosome 13, and overlapping the 13q deletion syndrome (case 6).[21, 22] With the exception of case 6, none of these abnormalities was detected by the conventional karyotype; the relatively large (9–Mb) terminal deletion in case 2 might have been expected to be apparent on the karyotype, but was only seen on review, directed by the array result (Table 1).
‘Likely pathogenic’ findings
Cases 8, 9, and 10 showed multiple CNVs, which were classified as ‘likely pathogenic’ because of their large size and involvement of multiple genes, of which at least one has some evidence in the literature indicating relevance to the clinical disorder. Family study information was an additional criterion for classification into this clinical category: the abnormalities detected in case 7 were considered ‘likely pathogenic’ on this basis. The abnormalities in question were also detected by conventional karyotyping in cases 9 and 10.
Case 8 showed a 8.6-Mb duplication in region 3p26.1p26.3 containing 20 RefSeq genes and a 4.9-Mb deletion in region 18p11.31p11.32 containing 27 RefSeq genes. The karyotype in case 9 showed a complex interchromosomal rearrangement involving chromosomes 5 and 11, shown by array analysis to be unbalanced because of nine interstitial deletions in chromosome region 5q11.2-14.1 (total 10.8 Mb), a 0.2-Mb deletion in 5q23.1, and a 2.2-Mb deletion in 11q22.1. The karyotype in case 10 showed additional unidentified material added to the long arm of chromosome 15: array analysis revealed a complex rearrangement between chromosomes 15 and 18, with an interstitial duplication of 12 Mb from 15q25.3q26.3, a terminal deletion of 1.5 Mb in 15q26.3, and a 9.5-Mb duplication from 18p11.32p11.22. The karyotype did not demonstrate the spatial rearrangements. All these changes (cases 7, 8, 9, and 10) were shown to be de novo.
Case 7 showed three maternally inherited copy number changes: a 0.2-Mb duplication containing five RefSeq genes from region 18p11.32, shown by fluorescent in situ hybridization (FISH) to be translocated adjacent to another duplication of 0.2 Mb in region 18q23 containing two genes on the same homologue (bacterial artificial chromosomes [BACs] RP11-145B19 and RP11-91C19), and a 0.5-Mb duplication from region 19q13.11 containing 14 genes. The copy number changes in chromosome 18 were also found in two of the mother's children, two of the mother's siblings, and her own mother, all diagnosed with Dandy Walker malformation spectrum with hydrocephalus, clinically and by imaging (both by magnetic resonance imaging [MRI] and computed tomography [CT]), suggesting a causative association between the genetic and phenotypic findings. The duplication from region 19q13.11 was not found to segregate with the clinical abnormality in this family.
Benign copy number variants
As expected, very high-resolution array analysis showed between three and ten small (<500 Kb), benign copy-number variants, including well-described polymorphisms, in almost all cases. Karyotyping showed chromosome rearrangements in three cases, none of which were of relevance to the fetal presentation. A supernumerary marker chromosome was observed in the mosaic karyotype of case 11, and was shown by FISH to be D15S10 positive but SNRPN negative. Microarray analysis showed this to be derived from a 2–Mb duplication from region 15q11.2, which is a recognised polymorphic copy number variable region. Case 12 showed a karyotype with an apparent intrachromosomal rearrangement of region 11q14q25; however, the microarray analysis failed to detect any imbalance in this region, at an effective resolution of 50 Kb. Case 13 showed a karyotype with a Robertsonian translocation between chromosomes 13 and 14 [45,XY, der(13;14)(q10;q10)], which as expected showed no imbalance on microarray analysis. Case 14 showed a mosaic karyotype with a minor 45,X cell line [45,X(2)/46,XX(28)], interpreted as ‘likely benign’. The mosaicism for chromosome X was not detected on microarray analysis.
Copy number variants of unknown significance
A CNV of ‘unknown significance’ was found in 42 samples, with 14 of these showing more than one (between two and three per sample). A total of 61 CNVs of ‘unknown significance’ were detected, of which 53 were not reported as there was insufficient evidence, that is, gene function information, to assess the clinical significance, and it was considered that information gained from parental studies of these would not have altered the clinical interpretation. A summary of the characteristics of these is presented in Table 3.
Eight (seven cases) of the 61 CNVs were selectively followed with parental testing, as they contained one or more genes with a potential relationship to clinical abnormality, based on literature information describing gene function, although in all these cases this information was not derived from human studies. All were reported (Table 2).
Case 15 showed a 210-Kb duplication from region 1p32.2 involving DAB1 and a 661-Kb duplication from region 11q13.4 containing 17 genes. Parental testing showed that the 1p32.2 duplication was maternally inherited, and that the 11q13.4 duplication was paternally inherited, with the father suspected as being a mosaic. Both parents are phenotypically normal. The 1p32.2 duplication involves exon three of DAB1 (NM_021080), which is involved in the laminar organisation of multiple neuronal types in the cerebral cortex, and is required for normal cognitive function. In mice, the disabled–1 gene plays a central role in cerebral and cerebellar development, directing the migration of cortical neurons past previously formed neurons to reach their proper layer.
In case 16, a 786-Kb deletion in region 4q21, involving ten RefSeq genes, including SCARB2 and SHROOM3, was shown to be inherited from a healthy father. The father reported having surgery for congenital diaphragmatic hernia. No association could be found. Case 17 showed a 439-Kb triplication from region 4q32.3q33 involving SH3RF1 and the distal 22 exons of NEK1. The latter is found in a centrosomal complex with FEZ1, a neuronal protein that plays a role in axonal development. Parents declined testing. Case 18 showed a maternally inherited 681-Kb duplication in region Xp11.3, in a female fetus, containing eight RefSeq genes, including ZNF673 and ZNF674, which are both located in a region of the X chromosome thought to be involved in nonsyndromic X-linked mental retardation.[25, 26].
Case 19 showed a paternally inherited 928-Kb duplication from region 18p11.31 involving five RefSeq genes including various transcripts of TGIF1, mutations of which are associated with holoprosencephaly type 4, and of DLGAP1 that codes for a transcription factor identified as a good candidate for the genetic basis of familial psychiatric illness. Case 20 showed a 649-Kb deletion in chromosome region 22q11.22q11.23 containing seven RefSeq genes. This appears to be a recurrent deletion based on the observation of several deletions with similar breakpoints in the public CNV databases, including seven and three separate entries in ISCA and DECIPHER, respectively. Indeed, these recurrent deletions are flanked proximally by LCR22–5 and distally by LCR22–6. Despite this, there are presently no reports in the peer-reviewed literature associating these recurrent deletions with phenotypic abnormality. Parental testing showed the deletion in this case to be paternally inherited. The father is healthy and had no congenital anomalies. He had behavioural problems following a teenage diagnosis of insulin-dependent diabetes. The fetus/patient had postnatal confirmation of bilateral renal dysplasia, and is otherwise healthy with normal infant development.
We have shown that very high-resolution CMA can be successfully incorporated into prenatal diagnostic testing for the investigation of fetal ultrasound anomaly. Because high-resolution CMA, that is, >50 Kb, produces many findings of uncertain or unknown clinical significance, there is an urgent need for research to clarify their importance, and we also need to work out how best to communicate the findings to patients without causing undue alarm.
The first aim was to compare CMA using high-resolution SNP arrays with karyotyping in terms of abnormality detection and inconsistent results. As has been reported in other prenatal studies using a variety of microarray platforms with a range of genome coverage and resolution, the detection rate of pathogenic CNVs is significantly improved compared with conventional karyotyping. These studies showed on average an improvement of 6% (range 4–13%), and here there was an improvement of 6.7% (9.6% or 10/104 versus 2.9% or 3/104). Ten cases with ‘pathogenic’ or ‘likely pathogenic’ findings (including one uniparental disomy) were detected by CMA, seven of which were not detected by karyotyping. Four of these (cases 1, 3, 4, and 7) showed small CNVs that would not be expected to be detected by microscopy, which has a resolution limit of no higher than 5 Mb. The paternal uniparental isodisomy of chromosome 14 detected in case 5, showing the complete loss of heterozygosity, is undetectable by karyotyping or by array comparative genomic hybridisation (aCGH). aCGH without SNP content is the most commonly used array platform in published prenatal studies, but does not provide the genotyping information necessary for UPD detection. The smallest clinically significant CNV detected was 449 Kb (case 3).
Balanced chromosome rearrangements where there is no loss or gain of genetic material are not detectable by CMA (excepting custom arrays designed to detect specific cancer-related translocations). Karyotyping on the other hand has this capacity, and two cases (cases 12 and 13) showed a balanced rearrangement undetected by array analysis. As neither of these karyotype findings was clinically significant in the context of the clinical investigations, the failure to detect them by CMA is not a major shortcoming. The value of CMA to characterise a clinically insignificant karyotype finding was also shown in case 11, where a small marker chromosome was shown to be a 2-Mb duplication from chromosome region 15q11.2, which is a well recognised polymorphic copy number variable region. More significant is the apparent false-negative array result in case 14, which did not detect the low level mosaicism of chromosome X (45,X/46XX) detected on karyotype, although again the finding is unlikely to be related to the brain abnormalities under investigation. Previous array studies have shown the limit of sensitivity for mosaicism to be no higher than 10%,[8, 32] and this is the probable explanation here.
Small pathogenic CNVs detectable only by high-resolution arrays are well documented in the postnatal setting.[5-7] Consequently, there has been some discussion of the merits of high-resolution arrays for prenatal diagnosis. In this cohort of patients, there appeared to be no particular benefit in using very high-resolution microarrays in terms of improving the detection of small pathogenic CNVs. The smallest, clinically significant CNV detected was the 449–Kb genomic region deleted in case 3 (TAR syndrome), and this is covered with increased probe density by most, if not all, microarrays used in diagnostics, irrespective of the average resolution. However, small pathogenic CNVs only detectable by high-resolution arrays are well documented in the postnatal setting.[5-7] As the highest available resolution arrays have the highest diagnostic sensitivity for small, disease-causing CNVs, we don't think that it would be appropriate to try and avoid parental anxiety by limiting the sensitivity of the analysis so that small, potentially significant CNVs are not discovered; it would be better to improve the way that we communicate these results to patients. In addition, we found that the increased probe density afforded by high-resolution CMA does help mitigate any loss in detection sensitivity associated with poor data quality, thus ensuring that the resolution of analysis is not compromised.
Despite the improvement in the detection of clinically significant CNVs, interpretational problems were exemplified in seven (6.7%) of the 104 cases that showed novel or rare, large, genic CNVs, but no relevant literature evidence of pathogenicity (Table 2; cases 7, 15–20). This is a higher rate of detection of CNV of ‘unknown clinical significance’ than is reported in other studies. We believe that this has to do with differences in reporting policies. The CNVs of ‘unknown clinical significance’ in Table 2 were all reported. With the exception of case 17, for which parental samples were unavailable, all were selectively followed-up with parental testing and shown to be inherited. The commonly accepted practice is to interpret these as rare familial variants and to reclassify them as ‘likely benign’; however, the accuracy of this interpretation has become less certain as knowledge has grown of the contribution of rare inherited variants with low effect sizes to the expression of complex genetic disorders and phenotypes. Largely in response to concerns about this class of finding, an approach to systematically assess between 600 and 1000 prospective prenatal cases by paediatric follow-up has recently been described (National Institute of Child Health and Human Development [NICHD] project no. 2U01HD055651–06; http://www.nichd.nih.gov): the hope is that the results from this study will direct clinical decision making in the future.
Particularly difficult in the prenatal setting is the limited time available to investigate these, especially the requirement for parental testing to determine whether the fetal findings are inherited or not, which is a consideration in determining significance. Interpretation of de novo findings in the absence of any other supportive evidence for clinical significance, however, was regarded as inadequate grounds for reclassification, although such information was included in reports and discussed with women.
In addition to these seven cases (eight CNVs), a further 53 CNVs of ‘unknown significance’ were detected in 35 cases (see Table 3), but these were not reported on the basis of a lack of literature evidence regarding gene function or functional effects of intronic CNVs. This is a higher proportion than in most other studies reported, which for the most part used lower resolution arrays (reviewed in Breman et al.), and where inheritance from an apparently healthy parent was the basis for reclassification of novel CNVs of ‘unknown/uncertain’ significance into the ‘likely benign’ category.
We did not identify any ‘susceptibility’ CNVs: these have been shown from association studies to be more common in patients with neurocognitive disorders (autism, epilepsy, schizophrenia, and cognitive impairment) than in control groups. Collectively, they are found in approximately 1–3% of individuals referred for investigation of these disorders, but would be expected to be much less common in prenatal diagnosis where the ascertainment is quite different. Their significance is uncertain as they are often inherited from an asymptomatic parent, are found in unaffected individuals in control populations, and show variable expressivity. In the absence of reliable data regarding the risk of associated postnatal phenotypes, the counselling of prospective parents after the prenatal ascertainment of these CNVs is challenging, and furthermore has significant cost implications. The discovery of a clinically significant CNV that is incidental to the clinical problem under investigation, but which may be relevant to other family members, is equally likely in the prenatal and postnatal settings. The identification of high levels of autozygosity on SNP microarrays indicative of an undisclosed incestuous parental relationship can also be considered an incidental finding, and management strategies have been proposed by other authors.[37, 38].
In the context of this study where there was an existing fetal anomaly in every case, the need to interpret these rare findings was overtaken by the importance and prognostic implications of the fetal anomaly. In five of the seven cases, a request for termination of the pregnancy was based on the imaging findings. The appropriateness of using very high-resolution CMA for prenatal genetic testing outside of the group with a known fetal anomaly (i.e. in those with increased maternal age, maternal serum screening risk, or anxiety) has yet to be determined, and may be less attractive because of the interpretational and counselling issues raised by all the above types of finding. Because of increased diagnostic sensitivity, it is becoming accepted that CMA will become the ‘first-tier’ test for prenatal investigation of fetal anomalies. In general, CMA is currently more expensive than conventional karyotyping, although this is dependant largely on throughput and array design. Significant cost efficiencies can be introduced if automated processing is used.
If CMA is to be offered in a wider context, that is, for non-fetal indications, such as maternal age, the provision of clear pre-test counselling and information on the advantages and disadvantages of this type of testing is a very important aspect of the consenting process. A pragmatic approach would be to offer parents the option of a more limited test, for example, a low-resolution test targeting a battery of well-established pathogenic CNVs with a genome-wide backbone coverage that will detect large pathogenic changes containing many genes, but minimise the detection of small novel changes. This allows the parents to make an informed choice of whether to opt for the highest resolution test to maximise the chances of detecting a pathogenic change, with the drawback of potentially difficult interpretations, or of a lower resolution test that avoids these.
The key feature of our reporting policy was the use of a clinical review committee for the case-by-case consideration of the significance of novel and rare CNVs, and their disclosure to clinicians. This committee assessed the relevant current peer-reviewed literature evidence and, where appropriate, commented on the expected clinical significance of a CNV for counselling purposes. It was also decided that it was advisable to align our prenatal and postnatal reporting policies. An unfortunate scenario would be the detection and reporting of a ‘susceptibility’ CNV in a child with autism, for instance, irrespective of its contribution to the aetiology, which was also detected prenatally but not reported.
In summary, this study has confirmed the higher rate of detecting pathogenic CNVs compared with conventional karyotyping reported in other studies. In this case, the use of very high-resolution SNP arrays has demonstrated the detection of pathogenic CNVs as small as 0.45 Mb, and also uniparental disomy. The study has also detected a relatively high rate (6.7%) of CNVs of ‘unknown significance’ that present interpretational difficulties beyond results from parental investigations. Our suggested policy for interpreting and reporting high-resolution CMA is based largely on a critical assessment of the evidence base current at the time of reporting, which for equivocal findings was carried out by an expert committee of clinical and laboratory geneticists on a case-by-case basis. The importance of pre-test counselling is emphasised to alert parents to the possibility of uninterpretable findings, non-disclosure, incompletely penetrant CNVs that confer disease susceptibility, and unexpected findings unrelated to the reason for investigation. Finally, the issue of balancing the use of the highest resolution arrays to maximise detection against the drawbacks of novel or uncertain findings may best be managed by giving parents the option of high-resolution or low-resolution, targeted, prenatal analysis. We envisage that reporting policies such as that suggested here for high-resolution prenatal microarray analysis will prove useful with respect to future attempts to analyse the fetal genome at even greater resolution, such as that afforded by whole-exome or whole-genome massively parallel sequencing.
This work was supported by the Victorian Government's Operational Infrastructure Support.
Disclosure of interests
None to declare.
Contribution to authorship
DG drove the project. DLB and HRS drafted the article. DG, DLB, FN, RO, MP, and CN analysed and interpreted the genomic data. SA and ARA processed the microarrays. GM, SMW, DA, PC, AY, SW, RPD, and NW performed the clinical evaluations. GM, SMW, and DA contributed to the interpretation of equivocal results. GM performed detailed genotype–phenotype analyses and summarised the clinical information for the cases presented in the paper. HRS initiated and coordinated the study. All authors contributed to the final version of the article.
Details of ethics approval
All women gave written informed consent for microarray testing. This work was carried out in accordance with the local institutional policy for prospective validation studies of diagnostic tests. Permission was obtained from the Royal Women's Hospital Research Committee endorsed by the Royal Women's Hospital Human Research Ethics Committee to obtain clinical information from the patient's records.