Use of prenatal chromosomal microarray: prospective cohort study and systematic review and meta-analysis

Transcription

1 Ultrasound Obstet Gynecol 2013; 41: Published online 7 May 2013 in Wiley Online Library (wileyonlinelibrary.com). DOI: /uog Use of prenatal chromosomal microarray: prospective cohort study and systematic review and meta-analysis S. C. HILLMAN*, D. J. McMULLAN, G. HALL, F. S. TOGNERI, N. JAMES, E. J. MAHER, C. H. MELLER, D. WILLIAMS, R. J. WAPNER**, E. R. MAHER* and M. D. KILBY* *School of Clinical and Experimental Medicine, College of Medicine and Dentistry, University of Birmingham, Edgbaston, Birmingham, UK; West Midlands Regional Genetics Laboratories and Department of Clinical Genetics, Birmingham Women s Foundation Trust, Edgbaston, Birmingham, UK; South East Scotland Cytogenetics Service, Western General Hospital, Edinburgh, UK; Fetal Medicine Centre, Birmingham Women s Foundation Trust, Edgbaston, Birmingham, UK; Obstetric Division, Hospital Italiano de Buenos Aires, Buenos Aires, Argentina; **Department of Obstetrics and Gynecology, Columbia University Medical Center, New York, NY, USA KEYWORDS: abnormal ultrasound findings; chromosomal microarray; G-band karyotyping; prenatal diagnosis ABSTRACT Objectives Chromosomal microarray analysis (CMA) is utilized in prenatal diagnosis to detect chromosomal abnormalities not visible by conventional karyotyping. A prospective cohort of women undergoing fetal CMA and karyotyping following abnormal prenatal ultrasound findings is presented in the context of a systematic review and meta-analysis of the literature describing detection rates by CMA and karyotyping. Methods We performed a prospective cohort study of 243 women undergoing CMA alongside karyotyping when a structural abnormality was detected on prenatal ultrasound. A systematic review of the literature was also performed. MEDLINE (1970 Dec 2012), EMBASE (1980 Dec 2012) and CINAHL (1982 June 2012) databases were searched electronically. Selected studies included > 10 cases and prenatal CMA in addition to karyotyping. The search yielded 560 citations. Full papers were retrieved for 86, and 25 primary studies were included in the systematic review. Results Our cohort study found an excess detection rate of abnormalities by CMA of 4.1% over conventional karyotyping when the clinical indication for testing was an abnormal fetal ultrasound finding; this was lower than the detection rate of 10% (95% CI, 8 13%) by metaanalysis. The rate of detection for variants of unknown significance (VOUS) was 2.1% (95% CI, %) when the indication for CMA was an abnormal scan finding. The VOUS detection rate was lower (1.4%; 95% CI, %) when any indication for prenatal CMA was meta-analyzed. Conclusion We present evidence for a higher detection rate by CMA than by karyotyping not just in the case of abnormal ultrasound findings but also in cases of other indications for invasive testing. It is likely that CMA will replace karyotyping in high-risk pregnancies. Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd. INTRODUCTION Standard G-band karyotyping has classically been used to detect chromosomal anomalies at a resolution of 5 10 Mb. This technology has been used since the 1960s and is now being supplemented, and in some instances replaced, by chromosomal microarray analysis (CMA), which allows examination of chromosomes to a resolution of 1 kb, smaller than the average gene. The advantages of CMA lie in the fact that it allows detection of smaller pathogenic chromosomal variants that are undetectable using standard cytogenetic analyses, it can be customized and it is amenable to high throughput. A potential drawback of CMA is that it does not allow detection of balanced chromosomal rearrangements, triploidy and some instances of mosaicism. The biggest challenge presented by CMA is the detection of chromosomal variants of unknown clinical significance (VOUS). Chromosomal copy number variation (CNV), detectable by CMA, is variation from the expected number of copies of a segment of DNA when compared to a reference genome 1.Itsaraet al. reported an average of three to seven variants in control data sets with a global average of 540 kb (c. 0.02% of the genome) of CNV DNA per individual 2. With such prevalent, largely benign CNV density, application and interpretation in a clinical setting can be challenging in terms of classifying variants as pathogenic, benign or VOUS. Prenatal cytogenetic testing is currently most often offered to couples at high risk of having a child Correspondence to: Prof. M. D. Kilby, School of Clinical and Experimental Medicine, College of Medical and Dental Sciences, University of Birmingham, B15 2TT, UK ( m.d.kilby@bham.ac.uk) Accepted: 1 March 2013 Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd. SYSTEMATIC REVIEW

2 Prenatal CMA: cohort study and systematic review 611 with chromosomal abnormality, either suspected after screening or because of family history of chromosomal abnormality or detection of a structural anomaly on prenatal ultrasound. Congenital anomalies are highly correlated with chromosomal abnormalities and vary depending on the number and type of scan anomalies. Knowing the presence (or absence) of a chromosomal abnormality can be useful in determining the overall prognosis for the child. In 2011 we published a systematic review and metaanalysis on the use of prenatal CMA 3. Since then, a significant number of studies involving important and large prenatal cohorts have been published. Here we present our prospective cohort of 243 pregnant women undergoing prenatal CMA following visualization of a structural anomaly on ultrasound. The results of our cohort case study are reported in the context of a contemporary systematic review of the literature and meta-analysis. METHODS Birmingham bacterial artificial chromosome cohort study We prospectively recruited 328 pregnant women seen at the Fetal Medicine Centre at Birmingham Women s Foundation Trust between November 2009 and April The subjects gave informed written consent for CMA when a fetal anomaly was detected on ultrasound and invasive testing was being considered. Single soft markers (echogenic bowel, choroid plexus cyst, echogenic cardiac foci and single umbilical artery) were excluded, although multiple markers were included. Increased nuchal translucency > 3.5 mm was included. Following invasive testing by chorionic villus sampling (CVS), fetal blood sampling or amniocentesis, samples were sent to the West Midlands Regional Genetics Laboratory (a single reference laboratory). Parental (maternal and paternal) venous blood samples were obtained at the time of invasive fetal sampling to exclude maternal cell contamination and to establish inheritance of CNVs where necessary. Sample types included are described in Table 1. This study was approved by the Staffordshire Research Ethics Committee (UK Reference no. 09/H1203/74). CMA was not initiated until results of the quantitative fluorescent polymerase chain reaction (QF-PCR) were available, and samples were not processed further if Table 1 Classification of sample types included in our prospective Birmingham bacterial artificial chromosome cohort Sample n (%) Uncultured amniotic fluid 146 (60) Cultured amniotic fluid 8 (3.5) Uncultured chorionic villus sample 50 (20.5) Cultured chorionic villus sample 3 (1) Fetal blood 29 (12) Fetal tissue 6 (2.5) Pulmonary effusion 1 (0.5) trisomy 13, 18, 21 or monosomy X were detected. If the QF-PCR result was normal, karyotyping and CMA were done in parallel. Typically, >10 mg of CVS or >12 ml of amniotic fluid were required for CMA. Differentially fluorescently labeled test and reference DNAs of the same gender were competitively hybridized to whole genome BAC (Bacterial Artificial Chromosome) microarrays (CytoChip Focus constitutional, BlueGnome, Cambridge, UK ). These arrays have BAC data points at every 1 Mb in the genome as a whole and tiled, overlapping BACs in regions associated with recurrent constitutional syndromes and within subtelomeric and centromeric regions. Each data point is replicated within the array design (triplicate in the genome backbone and quadruplicate in targeted regions) eliminating the need for a dye-swap experiment. This relatively focused microarray was chosen to minimize detection of VOUS. In essence, we used a > 2 BAC threshold for calling, equivalent to a working resolution of > 2Mb genomic backbone and > 200 kb in targeted regions. Detected gains or losses in copy number were compared with known CNVs in publically available databases (Database of Genomic Variants Decipher, decipher.sanger.ac.uk/, International Standards for Cytogenomic Arrays Consortium database, iscaconsortium.org/index.php/search) and against our own internal database to ascertain the clinical significance of the variation. CNVs of clinical significance or unknown significance were confirmed by fluorescence in situ hybridization (FISH) on metaphase spreads using one or more BAC clones within the abnormal region, and in cases of possible VOUS, were repeated on higher resolution microarrays (Affymetrix 2.7 M Cytogenetics Research array or ISCA 60 K CytoChip oligonucleotide array). The same BAC microarray was also performed on parental samples to evaluate whether the CNV was inherited or had occurred de novo. Classification of results and blinding CNVs were classified as benign, VOUS or pathogenic in accordance with the American College of Medical Genetics (ACMG) guideline 4. Pathogenic CNVs are documented as clinically significant in multiple peerreviewed publications. Penetrance and expressivity of the CNV is well defined, even if known to be variable. They may include large CNVs not described in the medical literature at the size observed in the fetus, but which overlap a smaller interval with clearly established clinical significance. A benign CNV will have been reported in multiple peer-reviewed publications or curated databases as being a benign variant, particularly if the nature of the copy number variation has been well characterized, and will typically represent a common polymorphism (documented in > 1% of the population). The category VOUS will have included findings later demonstrated to be either clearly pathogenic or clearly benign, but with insufficient evidence available for unequivocal determination of clinical significance at the time of reporting 4.

3 612 Hillman et al. Results of a pathological nature or the presence of VOUS were revealed to patients. When a chromosome anomaly was detected by either karyotyping or CMA, a consultant clinical geneticist was contacted for detailed discussion and counseling the next working day. Initial analysis of CMA and karyotyping results was performed in the West Midlands Regional Genetics Laboratory. Slides for conventional G-band karyotyping were sent to a second laboratory (where CMA results were unknown and therefore could not influence the results of G-band karyotyping) for prospective and independent interpretation (South East Scotland Cytogenetics Service, Western General Hospital). Analysis of the slides was added to the usual workflow of the analysts of this second laboratory to ensure that they were blinded. Systematic review and meta-analysis A prospective protocol was developed using widely recommended and comprehensive methodology 5. Data sources The search focused on prenatal studies using microarray technology. A research strategy was developed based on existing advice for prevalence searches 6. MEDLINE (1970 Dec 2012), EMBASE (1980 Dec 2012), CINAHL (1982 Dec 2012) and clinicaltrials.gov databases were searched electronically. The search of MEDLINE and EMBASE captured citations containing the relevant MeSH keywords and word variants for microarray and prenatal. The following terms were used to describe microarrays: microarray, DNA microarrays, array comparative genomic hybridisation, array CGH. Similarly, antenatal diagnosis, prenatal and fetal were used to capture prenatal. Bibliographies of relevant articles were searched manually to identify papers not captured by electronic searches. Experts were also contacted to ensure completeness 7 9. There were no language restrictions in the search for or selection of papers. Eligibility criteria for study selection Studies were selected in a two-stage process. Initially, all abstracts or titles were scrutinized by two reviewers (S.C.H. and C.H.M.) and full papers potentially eligible for citation were obtained. Studies were included if CMA had been used on prenatal specimens (analyzed during pregnancy or after delivery). CMA may have been performed for any indication and was not limited to cases referred because of abnormal findings on fetal ultrasound. Papers were excluded if CMA was not performed on acquired prenatal samples and if it was performed for pre-implantation diagnostics or for recurrent miscarriage. Finally, papers were excluded if the comparative genomic hybridization (CGH) technique was used rather than array CGH/CMA. Non-English papers were assessed by someone with a command of the relevant language if the title or abstract appeared to fit the criteria. Only papers that allowed generation of a 2 2 table comparing outcomes of CMA against those of conventional karyotyping were included. Data were extracted by two reviewers (S.C.H. and C.H.M.). Differences were resolved by discussion with a third reviewer (M.D.K.). R.J.W. also provided information on the NIH cohort prior to publication in the public domain 1. For each outcome, data were extracted into tables giving descriptive and numerical information for each study. Data were extracted on study characteristics and data quality. Data were used to construct 2 2 tables of test accuracy comparing normal and abnormal CMA results. Studies of fewer than 10 cases were excluded from the meta-analysis. Quality assessment and data synthesis Study quality was assessed using STROBE 10. CMA results were considered positive if they indicated VOUS (and therefore potential pathogenicity) or pathogenicity. Benign results were included in the CMA-negative group as they are not clinically relevant and would not be reported to clinicians or patients. These data were further broken down and the analysis was repeated for cases in which the clinical indication for CMA was a structural abnormality seen on ultrasound. Using 2 2 tables, we computed and pooled the percentage agreement between the two technologies (for both any clinical indication and abnormal ultrasound result) with 95% CI. The percentage of extra cases identified by CMA in those with a negative karyotype (for both any clinical indication and abnormal ultrasound result) with 95% CIs was calculated and pooled. Conversely, we then calculated and pooled the percentage of extra cases identified by karyotyping in those with a negative CMA result (for both any clinical indication and abnormal ultrasound result) with 95% CIs. Finally, we calculated and pooled the percentage of cases in which VOUS was reported. Heterogeneity in rates was examined graphically and statistically. For graphic assessment, forest plots of point estimate of rates and their 95% CIs were used. For exploration of reasons for heterogeneity, stratified analysis was performed according to the year of publication. A random effects model was used in light of heterogeneity. All statistical analyses were performed using Stata 11.0 statistical software (Stata Corp., College Station, Texas, USA). RESULTS Results of the Birmingham bacterial artificial chromosome cohort study Agreement to participate in this study was 90% of those approached. A total of 328 women were prospectively recruited for both CMA and G-band karyotyping. After exclusions, CMA was performed on 243 samples (from 243 patients) and compared to G-band karyotyping in the same samples (Figure 1). A total of 228 (94%) samples from fetuses with abnormalities in a single system were included along

4 Prenatal CMA: cohort study and systematic review 613 Studies recruited with fetal anomaly on ultrasound (n = 328) Both CMA and G-band karyotyping performed (n = 243; 74%) Excluded studies (n = 85; 26%) Soft marker only (n = 1) Insufficient DNA for both tests (n = 11) Performed on 2.7 M array only (n = 1) Trisomy 21, 13, 18 or monosomy X on QF-PCR (n = 66) CMA failed (n = 5) G-band karyotyping failed (n = 1) Abnormal on both CMA and G-band karyotyping (n = 12; 4.9%) Abnormal on CMA only (n = 10; 4.1%) Abnormal on G-band karyotyping only (n = 4; 1.6%) Normal on both CMA and G-band karyotyping (n = 217; 89.4%) Figure 1 Flow diagram of samples for the Birmingham bacterial artificial chromosome chromosomal microarray analysis (CMA) cohort. *Blinded results. Including nine pathogenic copy number variants and one variant of unknown significance reported to patients. Two cases of maternal cell contamination and three cases of failed clean-up. QF-PCR, quantitative fluorescent polymerase chain reaction. Table 2 Classification of structural anomalies in the 243 patients included in the Birmingham bacterial artificial chromosome prospective cohort using the Human Phenotype Ontology System Structural anomaly n (%) Single system 228 (93.8) Central nervous system 51 (21.0) Cardiovascular system 41 (16.9) NT > 3.5 mm/cystic hygroma 40 (16.5) Musculoskeletal system 25 (10.3) Genitourinary system 20 (8.2) Congenital diaphragmatic hernia 15 (6.2) Abdominal wall defect 11 (4.5) Head/face/neck 7 (2.9) Respiratory system 5 (2.1) Spina bifida/encephalocele 5 (2.1) Hydrops 4 (1.6) Gastrointestinal tract 3 (1.2) Tracheal/esophageal fistulae 1 (0.4) Multiple systems 15 (6.2) NT, nuchal translucency. with 15 (6%) samples from fetuses with abnormalities within two or more systems as seen on scan. Details of abnormalities are given in Table 2. In two fetal samples the original chromosomal abnormality seen at the first testing center was not noted on the blinded review at a second laboratory where the cytogeneticists were unaware of the CMA findings (Supplementary Table S1 (fetal samples 16 and 17)). The G-band karyotyping results presented for these two cases are those of the blinded data series. A total of 243 BAC microarrays were processed; 156 (64%) had no CNVs and 87 (36%) had at least one CNV. In total, 121 CNVs were found within these 87 samples and these were divided into benign CNVs, VOUS and pathogenic CNVs. Ninety of 121 samples (74%) were common benign CNVs and two (1.6%) were uncommon benign CNVs requiring further testing with FISH (in two separate samples). These results were not given to patients. Three VOUS were detected in three separate samples (2.5%) (Table S1 (fetal samples 1 3)) although this was reduced to a single sample (fetal sample 1) (0.8%) after further testing with a higher resolution microarray, resulting in two VOUS being reclassified as either an artifact or an uncommon benign CNV. The single true result of VOUS was relayed to the patient in a difficult counseling session. It consisted of a Mb duplication between Xp22.32 and Xp22.31 in a male fetus with a cardiac anomaly (truncus arteriosus). This had led to partial duplication and possible disruption of the gene NLGN4. The gene has no known cardiac link but has been linked to neurodevelopmental delay and autism. This duplication was maternally inherited (although the mother had no learning difficulties). Only one phenotypically normal male relative was available for testing, but he did not have the CNV. X-inactivation studies carried out on maternal DNA were inconclusive. Seventeen CNVs (14.0%) were classified as pathogenic and were also detected by karyotyping (in 12 separate fetal samples, 4.9% of the cohort; Table S1 (fetal samples 4 15)). These included four cases of aneuploidy (two cases of triple X also detected by QF-PCR and two cases of trisomy 9). In seven cases there was a large structural chromosome anomaly present, visible on both G-banding and CMA. In one case the structural chromosome abnormality (resulting in Smith Magenis syndrome) was at the limit of cytogenetic resolution and was visible only because of high-quality chromosome preparation. Nine (7.4%) CNVs in nine fetuses were considered pathogenic and were not detectable by karyotyping (Table S1 (fetal samples 16 24)). These included a mosaic isochromosome 12p (consistent with Pallister Killian syndrome), 1p36 microdeletion, four cases of 22q11.2 microdeletion (DiGeorge syndrome) and deletion of

5 614 Hillman et al. PMP22 (associated with hereditary neuropathy with liability to pressure palsies). One fetus (Table S1 (fetal sample 17)) had a de novo 6 8-Mb duplication between 11q24.2 and 11q25, including c. 50 HGNC genes. Finally, one fetus (Table S1 (fetal sample 24)) had a c. 3.2 Mb deletion at 5q35.3. More detailed analysis using an ISCA 60 K oligoarray further defined the deletion as being c. 1.9 Mb in size with additional c. 1.1 Mb duplication at 17q25.3, both of which were inherited from the mother who had an unbalanced translocation between 5q and 17q. The mother has been diagnosed with dyspraxia and has mild facial dysmorphia. The unbalanced translocation was inherited from the proband s maternal grandmother who had the translocation in a balanced form and does not have any learning difficulties. Prenatal ultrasound findings demonstrated absent corpus callosum and a meningocele. As the mother shows a phenotype consistent with the chromosomal abnormality it was deemed likely to be pathogenic. In five fetal samples (2% of the cohort) karyotyping revealed a chromosomal anomaly when CMA results were reported as normal; one was false-positive due to maternal cell contamination and three were balanced inherited inversions unlikely to have a phenotypic effect on the fetus. One fetus with a univentricular heart and hydrops fetalis was mosaic for monosomy X on amniocentesis and triple X on CVS (undetected by QF-PCR). Blood taken from the baby postnatally showed mosaicism for monosomy X and triple X (47,XXX[45]/45,X[15]). This chromosomal anomaly was consistent with the phenotype on scan (Table S1 (fetal samples 25 28)). Of note was that in seven of the 22 cases with a pathogenic finding (n = 21) or true VOUS (n = 1), identified by CMA and/or karyotyping, increased nuchal translucency (> 3.5 mm) was either part or the sole reason for testing. Results of systematic review and meta-analysis The process of literature identification and selection is summarized in Figure 2. In addition to the data from our own cohort, there were 25 primary articles that met the selection criteria 1,8,9,11 32 (17 not included in our previous meta-analysis 1,8,9,11 17,20 23,28,29,31 ). Twentythree cohorts (including the Birmingham BAC cohort) were included in the meta-analysis and three were excluded either because CMA was performed only when an abnormal karyotype had been detected 30,32 or because we could not extract data required for a 2 2table 8. In all 26 cohorts the collective number of samples analyzed was In the case of Fiorentino et al. their publication in contains data from their publication in We were able to extract data from the 2012 paper to look at detection by CMA and karyotyping; however, the 2011 publication was used to investigate overall agreement between the two tests, VOUS prevalence and the karyotyping vs CMA detection rate, as these were not extractable from the 2012 publication. Table S2 summarizes the study characteristics including study design, microarray type, sample type, indication for sampling and sample size. Figure 3 shows the quality assessment of papers included. Chromosomal testing for any clinical indication When CMA and conventional karyotyping were performed for any clinical indication (Table S2) overall agreement between the two tests was good at 93.4% (95% CI, %). These data were heterogeneous (P = 0.017; chi-square test). Twenty-two cohorts out of the 23 included in the meta-analysis were used here. One was excluded as the information could not be used to extract a2 2Table 11. In the present cohort study the G-band karyotyping analysts were blinded to the CMA results. We attempted to meta-analyze the excess rate of detection by CMA compared to karyotyping when the referral indication was varied; however, the results were highly heterogeneous (Figure 4) (P < 0.01; chisquare test), with excess detection by CMA ranging from 0.4 to 50%. Four papers seemed to contribute disproportionately to the heterogeneity of these data. In three cases this can be explained by small sample size and is likely not to be representative of the true detection rates by CMA 13,16,23. The fourth study relied on a highresolution CMA platform (Affymetrix SNP 6.0) but did not use parental samples to follow up on VOUS, therefore yielding a high detection rate by CMA 27. However, even when we performed a sensitivity analysis by removing these four papers, the variation in the excess rate of chromosomal imbalances detected by CMA was still unexplained (P < 0.01; chi-square test). We therefore felt a pooled result would be misleading. Although heterogeneous, all cohorts showed a positive result, with CMA revealing chromosomal abnormalities not detected by standard G-band karyotyping. Conventional karyotyping revealed an extra 0.6% (95% CI, %) abnormality rate when CMA results were normal. These data are also heterogeneous (P < 0.01; chi-square test). Eight papers were excluded from this meta-analysis 11,13,18,21 23,26,27 as CMA was performed only on samples that had a known normal karyotype by G-band analysis. The rate of VOUS was 1.4% (95% CI, %) when samples were analyzed for any indication. The meta-analysis was performed using VOUS rates from 17 cohorts. In the cohort of Wapner et al. the reclassified VOUS rate from 2012 was meta-analyzed 1. In four papers the VOUS rate could not be extracted 11,17,20,31. The paper by Tyreman et al. was excluded from this analysis as parental samples were not tested in order to reclassify results as benign or pathogenic, leaving a disproportionate amount of VOUS 27. Exclusion of the present cohort did not change the meta-analysis significantly (1.5%; 95% CI, %). Subgroup analysis by date of publication for any clinical indication To determine whether we could account for the heterogeneity within the meta-analysis, publications were

6 Prenatal CMA: cohort study and systematic review 615 Total citations from electronic searches (n = 560) MEDLINE (n = 116) CINAHL (n = 32) EMBASE (n = 404) Clinical trials.gov (n = 1) Experts (n = 6) Own study (n = 1) Excluded after screening abstract (n = 460) Duplications removed (n = 14) Studies retrieved for detailed evaluation (n = 86) Manual searching of relevant lists (n = 8) Studies included in systematic review (n = 25) Studies excluded (n = 69) Used cell lines (n = 6) Unable to distinguish if prenatal or abortion material (n = 1) Unable to construct 2 2 tables (n = 7) Postnatal samples (n = 4) Assessing only trisomy 13, 18 and 21 (n = 3) Detecting only mosaicism (n = 1) No microarray used (n = 13) No data (n = 9) Not relevant (n = 10) Full data unavailable (n = 15) Figure 2 Flow diagram showing study selection process for the systematic review. Title and abstract appropriate Introduction states background and objectives Study design prospective Setting and recruitment period described Eligibility and matching criteria given All variables defined Data sources and methods of assessment (commercial array used) given Efforts to address bias (attempts to look into parental samples) described Study size explained Quantitative variables and their analysis explained All statistical methods described Number of individuals reported in flow diagram Participant characteristics and follow-up time reported Number of outcome events reported All results reported with precision Key results summarized Limitations of study stated Interpretation of results given Generalizability (normal and abnormal karyotypes) addressed Source of funding given Number of studies Figure 3 Quality assessment of the studies included in the systematic review using the STROBE 10 checklist.,yes;, no;, unclear. split into those published in or before 2009 and those published subsequently. Publications from 2009 or before reported a lower excess detection rate by CMA over karyotyping compared with publications from but the results remained heterogeneous in both cases (P < 0.01; chi-square test). We noted an increase in the number of VOUS as the detection rate by CMA increased (i.e. as resolution of the array increases). For all papers published from 2010 onwards we performed a Spearman s rank correlation, looking at the association between overall detection rate by CMA and VOUS rate. For these 12 cohorts 1,9,13 15,16,21 23,28,29 the Spearman rho coefficient was 0.81 (P = (positive correlation)). This shows a significant positive relationship between the increase in VOUS rate and the overall detection rate by CMA.

7 616 Hillman et al. Reference % (95% CI) Le Caignec et al. (2005) ( ) Sahoo et al. (2006) ( ) Bi et al. (2008) ( ) Kleeman et al. (2009) ( ) Tyreman et al. (2009) ( ) Vialard et al. (2009) ( ) Evangelidou et al. (2010) ( ) Faas et al. (2010) ( ) Maya et al. (2010) ( ) Park et al. (2010) ( ) Valduga et al. (2010) ( ) D'amours et al. (2011) ( ) Leung et al. (2011) ( ) Park et al. (2011) ( ) Schmid et al. (2012) ( ) Armengol et al. (2012) ( ) Fiorentino et al. (2012) ( ) Hillman et al ( ) Shaffer et al. (2012) ( ) Srebniak et al. (2012) ( ) Wapner et al. (2012) ( ) Percentage of positive findings by CMA with normal karyotype Figure 4 Forest plot for meta-analysis of detection rate of chromosomal imbalances by chromosomal microarray analysis when karyotype is normal and chromosomal testing is performed for any clinical indication. *Present study. Results for abnormal ultrasound scan Data from 16 cohorts in addition to our own were used to compare conventional karyotyping to CMA when the indication was structural abnormality on ultrasound 1,9,13,14,16,18 24, Here the excess rate of detection by CMA over karyotyping is somewhat increased to 10% (95% CI, 8 13%) (Figure 5) (P < 0.01; chi-square test). Exclusion of the present cohort did not change the meta-analysis significantly (10.5%; 95% CI, %). Karyotyping revealed only 0.8% (95% CI, %) more abnormalities than did CMA using data from nine cohorts in addition to our own 1,9,14,16,19,20,24,28,29, the main examples being balanced rearrangements and triploidy. Both analyses were heterogeneous (P < 0.01; chi-square test). The VOUS rate was meta-analyzed using data from 16 cohorts 1,9,13,16,18 24,26 29 (including our own). When indication for testing is an abnormal scan the VOUS rate is 2.1% (95% CI, %), higher than the metaanalyzed rate of 1.4% for any indication. Exclusion of the present cohort did not change the meta-analysis significantly (2.3%; 95% CI, %). Subgroup analysis by date of publication for abnormal ultrasound findings Given the increasing resolution of CMA over time and the heterogeneity present in the meta-analysis we performed subgroup analysis by date of publication when the indication for CMA was abnormal ultrasound findings saw the publication of many larger cohorts of fetuses who underwent CMA because of abnormal ultrasound findings 19,11,14,22,28,29. In this subanalysis two papers were removed. Schmid et al. 13 had a 50% detection rate by CMA over karyotyping, partly due to a very small sample size (n = 12), a high-resolution array (SNP 6.0 Affymetrix) and no evidence that parental samples were analyzed to determine whether the CNV was de novo. This paper was therefore excluded. The second paper excluded was that of D Amours et al. 23, as it also was an outlier, with a detection rate by CMA of 20.4% over karyotyping. This was in part due to a smaller cohort size (n = 49) and the use of four custom-designed arrays. Subanalysis of the remaining seven cohorts (including our own) still yielded heterogeneous statistical results

8 Prenatal CMA: cohort study and systematic review 617 Reference Le Caignec et al. (2004) 26 % (95% CI) ( ) Bi et al. (2008) ( ) Tyreman et al. (2009) ( ) Kleeman et al. (2009) ( ) Vialard et al. (2009) ( ) Faas et al. (2010) ( ) Evangelidou et al. (2010) ( ) Valduga et al. (2010) ( ) Leung et al. (2011) ( ) D'amours et al. (2011) ( ) Schmid et al. (2012) ( ) Armengol et al. (2012) ( ) Fiorentino et al. (2012) ( ) Hillman et al ( ) Shaffer et al. (2012) ( ) Srebniak et al. (2012) ( ) Wapner et al. (2012) ( ) Overall ( ) Percentage of positive findings by CMA with normal karyotype Figure 5 Forest plot for meta-analysis of detection rate of chromosomal imbalances by chromosomal microarray analysis when karyotype is normal and chromosomal testing is performed for structural abnormality on ultrasound. *Present study. Reference % (95% CI) Leung et al. (2011) ( ) Armengol et al. (2012) ( ) Fiorentino et al. (2012) ( ) Hillman et al ( ) Shaffer et al. (2012) ( ) Srebniak et al. (2012) ( ) Wapner et al. (2012) ( ) Overall 7.20 ( ) Percentage of positive findings by CMA with normal karyotype Figure 6 Forest plot for meta-analysis of detection rate of chromosomal imbalances by chromosomal microarray analysis when karyotype is normal and chromosomal testing is performed for structural abnormality on ultrasound (subgroup analysis of publications from 2011 to 2012). *Present study.

9 618 Hillman et al. (P < 0.01; chi-square test) but on graphic representation (Figure 6) the analysis appears more homogeneous than it did previously (Figure 5). This subanalysis, using cohorts published between 2011 and 2012, shows that the excess detection rate by CMA performed because of abnormal ultrasound findings appears to be 7% (95% CI, 5 10%) over conventional karyotyping. We believe this to be a more accurate detection rate when performing CMA because of abnormal ultrasound findings. Targeted fluorescence in situ hybridization (FISH) Using eight cohorts (seven in addition to our own) it was possible to analyze the detection rate of 22q microdeletion (DiGeorge syndrome) if FISH was used rather than CMA, for samples that had been sent for testing because of a fetal cardiac anomaly 8,9,13,16,18,27,28. Meta-analysis revealed a 4% (95% CI, 1 10%) detection rate of 22q microdeletion in those samples for which testing was requested because of fetal cardiac anomaly. This would imply a level of detection by CMA (for any clinical indication) across the genome similar to that using specific FISH probes based on the prenatal phenotype. DISCUSSION Our prospective, case cohort study found that CMA revealed abnormalities not found on conventional karyotyping in 4.1% of cases (10/243) when an abnormality had been seen on ultrasound. This is lower than the increase in detection rate reported in the literature to date, with a meta-analyzed rate of 10%. The reason for this discrepancy is likely to be our choice of a purposefully conservative microarray design at the time of inception of the project, to limit the amount of parental follow-up required and VOUS detected. Our referring region within the UK has a high number of consanguineous families, which may have lowered our detection rate by CMA, as the actual cause of abnormal ultrasound findings may be autozygosity associated with a single-gene autosomal recessive condition, undetectable by CMA platforms not containing single nucleotide polymorphism (SNP) probes. The meta-analysis rate of 10% may also have been higher because some studies with artificially elevated detection rates (due to small cohort numbers, no parental testing and high VOUS rates) were included 13,23,27.The subanalysis using papers between 2011 and 2012 showed a detection rate by CMA of 7%, which we believe is closer to the actual rate detected by CMA over karyotyping using up-to-date array platforms. Our lower detection rate cannot be attributed solely to the targeted nature of the BAC array, as other studies using the same array (Fiorentino et al. 11 and Lee et al. 8, detection rates over karyotyping of 6.3% and 8.2%, respectively) have shown higher detection rates. However, both of these studies report chromosomal size differences > 10 Mb that were missed by karyotyping. We conclude that the good quality of preparations for G-band karyotype analysis available in our laboratory has also contributed to the lower detection rate by CMA over conventional karyotyping. The result of this meta-analysis further strengthens the evidence for use of microarray technology in this particular group. When CMA is performed in the case of an abnormal scan, the VOUS rate is higher (2.1%) than when performed for any indication (1.4%). This is likely due to increased chromosomal pathology within this group. With time, some of the VOUS detected will be redefined as benign but others will be reclassified as pathogenic variants. We found CMA to be robust and accurate, requiring culture of cells in only 4.5% of cases in our cohort. The turnaround time in our laboratory was 10 days; however, this figure would be reduced if uptake of CMA was increased, allowing more effective batch testing. In our cohort study, cases with increased nuchal translucency > 3.5 mm (16.5% of cohort) had a high rate of pathogenic chromosomal differences, even when common trisomies were excluded. This was also true of those anomalies detected by CMA (as has been demonstrated previously) 22. This systematic review and meta-analysis of outcomes was performed to answer important questions regarding overall detection rates by CMA over conventional full karyotyping, the rate of VOUS and how this has changed over time. The concern with increasing resolution of CMA is the potential subsequent increase in VOUS detection rate. Certainly in papers published from 2010 onwards a higher detection rate is positively correlated with a higher VOUS rate. It is thus fundamentally important that, as CMA platforms provide increased resolution, there is national and international guidance regarding reporting of VOUS in the prenatal setting. When performed because of any indication, conventional karyotyping has an additional abnormality detection rate of 0.6% over CMA. However, as we and others have shown, most of this rate can be explained by balanced inversions and translocations. These are unlikely to cause a phenotype in the absence of a phenotype in the parent and, when accompanied by a normal microarray result, provide reassurance that the rearrangement is balanced. Triploidy can also be detected by conventional karyotyping and missed by CMA. However, it is likely to be picked up by other tests, e.g. for maternal cell contamination or, using QF-PCR, for common trisomies. There would have been a 4% detection rate by FISH for 22q microdeletion (DiGeorge syndrome) if the clinical indication had been cardiac anomaly on ultrasound. This is interesting as it shows that intelligent targeting of FISH probes according to the phenotype on ultrasound yields only the same detection rate as that yielded by CMA when performed for any indication (4%). In addition, it examines only a single chromosomal locus whereas CMA interrogates thousands or even millions of loci simultaneously, without requiring prior knowledge on the part of the clinician to request tests for particular chromosomal anomalies.

10 Prenatal CMA: cohort study and systematic review 619 The strength of our study lies in the rigor of the methodology; it met the quality criteria laid down in the MOOSE statement 33. The systematic review contained data on pregnancies, substantially larger than the 751 pregnancies included in our first systematic review 3 and is likely to be more representative of the true detection rate of CMA over karyotyping. These data are, however, still heterogeneous. This is possibly due to smaller cohorts with an artificially high detection rate and the different platforms used in different chromosomal microarray studies. These articles were published between 2004 and During this period, although microarray design has changed to provide increased resolution, subgroup analysis at the time of publication did not account for the heterogeneity. The authors cannot account for ascertainment bias towards cases that clinicians felt may yield an abnormal CMA result. However, many of the larger more recent cohort studies (including our own) are prospective and recruited consecutive patients which should minimize this confounder. The American College of Obstetrics and Gynecology (ACOG) 34 and the Italian Society of Human Genetics (SIGU) 35 have recommended that karyotyping remain the principal cytogenetic tool in prenatal diagnosis and microarrays should be used as an adjunct when a structural anomaly is seen on ultrasound. We present collective evidence for a higher detection rate by CMA, not just for referrals due to abnormal ultrasound findings but for other indications calling for invasive testing. CMA is robust, accurate and valuable in the prenatal setting, particularly when there are abnormalities on ultrasound. Patients should be counseled and informed that VOUS do occur at a rate of approximately %. Analyses must be performed to assess the economic viability of CMA but, with the falling cost of the test and its potential advantages, it is likely to become costeffective in the future. ACKNOWLEDGMENT S. Hillman was funded by the children s medical research charity, SPARKS, UK. REFERENCES 1. Wapner RJ, Martin CL, Levy B, Ballif BC, Eng CM, Zachary JM Savage M, Platt LD, Saltzman D, Grobman WA, Klugman S, Scholl T, Simpson JL, McCall K, Aggarwal VS, Bunke B, Nahum O, Patel A, Lamb AN, Thom EA, Beaudet AL, Ledbetter DH, Shaffer LG, Jackson L. Chromosomal microarray versus karyotyping for prenatal diagnosis. NEnglJMed2012; 367: Itsara A, Cooper GM, Baker C, Girirajan S, Li J, Absher D, Krauss RM, Myers RM, Ridker PM, Chasman DI, Mefford H, Ying P, Nickerson DA, Eichler EE. Population analysis of large copy number variants and hotspots of human genetic disease. Am J Hum Genet 2009; 84: Hillman SC, Pretlove S, Coomarasamy A, McMullan DJ, Davison EV, Maher ER, Kilby MD. Additional information from array comparative genomic hybridization technology over conventional karyotyping in prenatal diagnosis: a systematic review and meta-analysis. Ultrasound Obstet Gynecol 2011; 37: Kearney HM, Thorland EC, Brown KK, Quintero-Rivera F, South ST, A working group of the american college of medical genetics ACMG laboratory quality assurance commitee. American College of Medical Genetics standards and guidelines for interpretation and reporting of postnatal constitutional copy number variation. Genetics in Med 2011; 13: Khan KS, Terrier G, Glanville J, Sowden AJ, Kleijnen J. Undertaking Systematic Reviews of Research on Effectiveness CRDs Guidance for carrying out or commisioning reviews (2nd edn) CRD report No 4 NHS Centre for Reviews and Dissemination University of York. 6. McKibbon A, Eady A, Marks S. Identifying and selecting studies for inclusion. In PDQ Evidence-Based Principles and Practice. BC Decker Inc: Hamilton, Canada, 1999; Van den Veyver IB, Patel A, Shaw CA, Pursley AN, Kang SH, Simovich MJ, Ward PA, Darilek S, Johnson A, Neill SE, Bi W, White LD, Eng CM, Lupski JR, Cheung SW, Beaudet AL. Clinical use of array comparative genomic hybridization (acgh) for prenatal diagnosis in 300 cases. Prenat Diagn 2009; 29: Lee CN, Lin SY, Lin CH, Shih JC, Lin TH, Su YN. Clinical utility of array comparative genomic hybridisation for prenatal diagnosis: a cohort study of 3171 pregnancies. BJOG 2012; 119: Shaffer LG, Dabell MP, Fisher AJ, Coppinger J, Bandholz AM, Ellison JW, Ravnan JB, Torchia BS, Ballif BC, Rosenfeld JA. Experience with microarray-based comparative genomic hybridization for prenatal diagnosis in over 5000 pregnancies. Prenat Diagn 2012; 32: Von Elm E, Altman DG, Egger M, Pocock SJ, Gotzsche PC, Vandenbroucke JP. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. BMJ 2007; 335: Fiorentino F, Napoletano S, Caiazzo F, Sessa M, Bono S, Spizzichino L, Bono S, Sessa M Nuccitelli A, Biricik A, Gordon A, Rizzo G, Baldi M. Chromosomal microarray analysis as a first-line test in pregnancies with a priori low risk for the detection of submicroscopic chromosomal abnormalities. Eur J Hum Genet DOI: /ejhg [Epub ahead of print] 12. Breman A, Pursley AN, Hixson P, Bi W, Ward P, Bacino CA, Shaw C, Lupski JR, Beaudet A, Patel A, Cheung SW, Van den Veyver I. Prenatal chromosomal microarray analysis in a diagnostic laboratory; experience with >1000 cases and review of the literature. Prenat Diagn 2012; 32: Schmid M, Stary S, Blaicher W, Gollinger M, Husslein P, Streubel B. Prenatal genetic diagnosis using microarray analysis in fetuses with congenital heart defects. Prenat Diagn 2012; 32: Fiorentino F, Caiazzo F, Napolitano S, Spizzichino L, Bono S, Sessa M, Nuccitelli A, Biricik A, Gordon A, Rizzo G, Baldi M. Introducing array comparative genomic hybridization into routine prenatal diagnosis practice: a prospective study on over 1000 consecutive clinical cases. Prenat Diagn 2011; 31: Maya I, Davidov B, Gershovitz L, Zalzstein Y, Taub E, Coppinger J Shaffer LG, Shohat M. Diagnostic utility of arraybased comparative genomic hybridization (acgh) in a prenatal setting. Prenat Diagn 2010; 30: Faas BH, van db, I, Kooper AJ, Pfundt R, Hehir-Kwa JY, Smits AP De Leeuw N. Identification of clinically significant, submicroscopic chromosome alterations and UPD in fetuses with ultrasound anomalies using genome-wide 250 k SNP array analysis. J Med Genet 2010; 47: Park SJ, Jung EH, Ryu RS, Kang HW, Ko JM, Kim HJ, Cheon CK, Hwang SH, Kang HY. Clinical implementation of

11 620 Hillman et al. whole-genome array CGH as a first-tier test in 5080 pre- and postnatal cases. Mol Cytogenet 2011; 4: Kleeman L, Bianchi DW, Shaffer LG, Rorem E, Cowan J, Craigo SD, Tighiouart H, Wilkins-Haug LE. Use of array comparative genomic hybridization for prenatal diagnosis of fetuses with sonographic anomalies and normal metaphase karyotype. Prenat Diagn 2009; 29: Vialard F, Molina GD, Leroy B, Quarello E, Escalona A, Le Sciellour C, Serazin V, Roume J, Ville Y, de Mazancourt P, Selva J. Array comparative genomic hybridization in prenatal diagnosis: another experience. Fetal Diagn Ther 2009; 25: Evangelidou P, Sismani C, Ioannides M, Christodoulou C, Koumbaris G, Kallikas I, Georgiou I, Velissariou V, Patsalis PC. Clinical application of whole-genome array CGH during prenatal diagnosis: Study of 25 selected pregnancies with abnormal ultrasound findings or apparently balanced structural aberrations. Mol Cytogenet 2010; 3: Valduga M, Philippe C, Bach SP, Thiebaugeorges O, Miton A, Beri M Bonnet C, Nemos C, Foliguet B, Jonveaux P. A retrospective study by oligonucleotide array-cgh analysis in 50 fetuses with multiple malformations. Prenat Diagn 2010; 30: Leung TY, Vogel I, Lau TK, Chong W, Hyett JA, Petersen OB, Choy KW. Identification of submicroscopic chromosomal aberrations in fetuses with increased nuchal translucency and apparently normal karyotype. Ultrasound Obstet Gynecol 2011; 38: D Amours G, Kibar Z, Mathonnet G, Fetni R, Tihy F, Desilets V, Nizard S, Michaud JL, Lemyre E. Whole-genome array CGH identifies pathogenic copy number variations in fetuses with major malformations and a normal karyotype. Clin Genet 2012; 81: Bi W, Breman AM, Venable SF, Eng PA, Sahoo T, Lu XY, Beaudet AL, Cheung SW, White LD. Rapid prenatal diagnosis using uncultured amniocytes and oligonucleotide array CGH. Prenat Diagn 2008; 28: Sahoo T, Cheung SW, Ward P, Darilek S, Patel A, Del Gaudio D, Kang SH, Lalani SR, Li J, McAdoo S, Burke A, Shaw CA, Stankiewicz P, Chinault AC, Van den Veyver IB, Roa BB, Beaudet AL, Eng CM. Prenatal diagnosis of chromosomal abnormalities using array-based comparative genomic hybridization. Genet Med 2006; 8: Le Caignec C, Boceno M, Saugier-Veber P, Jacquemont S, Joubert M, David A, Frebourg T, Rival JM. Detection of genomic imbalances by array based comparative genomic hybridisation in fetuses with multiple malformations. JMed Genet 2005; 42: Tyreman M, Abbott KM, Willatt LR, Nash R, Lees C, Whittaker J, Simonic I. High resolution array analysis: diagnosing pregnancies with abnormal ultrasound findings. J Med Genet 2009; 46: Srebniak MI, Boter M, Oudesluijs GO, Cohen-Overbeek T, Govaerts LC, Diderich KE, Oegema R, Knapen MF, van de Laar IM, Joosten M, Van Opstal D, Galjaard RJ. Genomic SNP array as a gold standard for prenatal diagnosis of foetal ultrasound abnormalities. Mol Cytogenet 2012; 5: Armengol L, Nevado J, Serra-Juhe C, Plaja A, Mediano C, Garcia-Santiago FA, Villa O, Mansilla E, Preciado C, Fernández L, ÁngelesMoriM,García-Pérez L, Lapunzina PD, Pérez-Jurado LA. Clinical utility of chromosomal microarray analysis in invasive prenatal diagnosis. Hum Genet 2012; 131: De Gregori M, Ciccone R, Magini P, Pramparo T, Gimelli S, Messa J, Novara F, Vetro A, Rossi E, Maraschio P, Bonaglia MC, Anichini C, Ferrero GB, Silengo M, Fazzi E, Zatterale A, Fischetto R, Previderé C, Belli S, Turci A, Calabrese G, Bernardi F, Meneghelli E, Riegel M, Rocchi M, Guerneri S, Lalatta F, Zelante L, Romano C, Fichera M, Mattina T, Arrigo G, Zollino M, Giglio S, Lonardo F, Bonfante A, Ferlini A, Cifuentes F, Van Esch H, Backx L, Schinzel A, Vermeesch JR, Zuffardi O. Cryptic deletions are a common finding in balanced reciprocal and complex chromosome rearrangements: a study of 59 patients. J Med Genet 2007; 44: Park JH, Woo JH, Shim SH, Yang SJ, Choi YM, Yang KS, Cha DH. Application of a target array comparative genomic hybridization to prenatal diagnosis. BMC Med Genet 2010; 11: Rickman L, Fiegler H, Shaw-Smith C, Nash R, Cirigliano V, Voglino G, Ng BL, Scott C, Whittaker J, Adinolfi M, Carter NP, Bobrow M. Prenatal detection of unbalanced chromosomal rearrangements by array CGH. J Med Genet 2006; 43: Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie D, Moher D, Becker BJ, Sipe TA, Thacker SB. meta-analysis of observational studies in epidemiology: a proposal for reporting. Meta-analysis of Observational studies in epidemiology (MOOSE group). JAMA 2000; 283: ACOG Committee Opinion No. 446: array comparative genomic hybridization in prenatal diagnosis. Obstet Gynecol 2009; 114: Novelli A, Grati FR, Ballarati L, Bernardini L, Bizzoco D, Camurri L, Casalone R, Cardarelli L, Cavalli P, Ciccone R, Clementi M, Dalprà L, Gentile M, Gelli G, Grammatico P, Malacarne M, Nardone AM, Pecile V, Simoni G, Zuffardi O, Giardino D. Microarray application in prenatal diagnosis: a position statement from the cytogenetics working group of the Italian Society of Human Genetics (SIGU), November Ultrasound Obstet Gynecol 2012; 39: SUPPORTING INFORMATION ON THE INTERNET The following supporting information may be found in the online version of this article: Table S1 Results of Birmingham bacterial artificial chromosome prospective cohort study showing details of cases in which chromosomal abnormalities were detected by chromosomal microarray analysis and/or G-band karyotyping in our prospective cohort and concordance between the two centers performing G-band karyotyping. Table S2 Study characteristics of the 26 studies included in the systematic review.