K. COHEN*, A. TZIKA, H. WOOD, S. BERRI, P. ROBERTS, G. MASON* and E. SHERIDAN

Transcription

1 Ultrasound Obstet Gynecol 05; 45: Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 0.00/uog.4767 Diagnosis of fetal submicroscopic chromosomal abnormalities in failed array CGH samples: copy number by sequencing as an alternative to microarrays for invasive fetal testing K. COHEN*, A. TZIKA, H. WOOD, S. BERRI, P. ROBERTS, G. MASON* and E. SHERIDAN *Department of Fetal Medicine, Leeds General Infirmary, Leeds, UK; Leeds Institute of Cancer and Pathology, Leeds, UK; Department of Cytogenetics, St James s University Hospital, Leeds, UK; Illumina UK Ltd, Chesterford Research Park, Little Chesterford, Essex, UK KEYWORDS: array CGH limitations; molecular karyotyping; prenatal CNV-Seq ABSTRACT Objectives Array comparative genomic hybridization (CGH) has become the technology of choice for high-resolution prenatal whole genome analysis. Limitations of microarrays are mainly related to the analog nature of the analysis, and poor-quality DNA can result in failed quality metrics with these platforms. We examined a cohort of abnormal fetuses with failed array CGH results using a next-generation sequencing algorithm, CNV-Seq. We assessed the ability of the platform to handle suboptimal prenatal samples and generate interpretable molecular karyotypes. Methods Nine samples obtained from abnormal fetuses and one from a normal control fetus were sequenced using an Illumina GAIIx. A segmentation algorithm for sequencing data was used to determine regional copy number data on the sequencing datasets. Results Phred quality scores were satisfactory for analysis of all samples. CNV-Seq identified both large- and small-scale abnormalities in the cohort, and normal results were obtained for fetuses for which microarray data were previously uninterpretable. No variants of uncertain significance were detected. Analysis of the digital sequencing datasets offered some advantages over array CGH output. Conclusions Using next-generation sequencing for the detection of genomic copy number variants may be advantageous for poor-quality, invasively-acquired prenatal samples. CNV-Seq could become a potential alternative to array CGH in this setting. Copyright 04 ISUOG. Published by John Wiley & Sons Ltd. INTRODUCTION Array comparative genomic hybridization (CGH) has become the gold standard investigation for the postnatal diagnosis of genomic copy number variants (CNVs). Challenges specific to prenatal diagnosis have resulted in cautious adoption of this technology, and the implementation of arrays into fetal diagnostics in the UK National Health Service is still not fully established. However, it is clear that microarray analysis offers a significant improvement in the diagnostic yield over standard antenatal karyotyping, especially in cohorts of structurally-abnormal fetuses or stillbirths,. Although the advantages of array CGH are clear, the relatively short lifespan of the technology to date means that potential disadvantages are still emerging. While the detection of variants of uncertain significance (VOUS) is the most challenging drawback of the technology so far encountered, other difficulties have been acknowledged. Technical challenges when using prenatal tissue have been reported with array CGH platforms, primarily issues with low volumes of, and poor-quality, DNA 4. Although culturing of tissue may overcome the problem of DNA quantity, the impact on turnaround time is undesirable in the prenatal setting. Poor-quality, contaminated or fragmented DNA can perform poorly on array CGH platforms, owing mainly to suboptimal hybridization or an increased signal generated by contaminants 5. The analog nature of array technology could be considered an increasing disadvantage in the era of next-generation sequencing. Initially described in 009 6, CNV mapping by sequencing is a digital approach, relying on numerical differences between aligned reads from different individuals. Correspondence to: Dr K. Cohen, Department of Fetal Medicine, Clarendon Wing, Leeds General Infirmary, Leeds, LS EX, UK ( k.cohen@nhs.net) Accepted: 0 December 04 Copyright 04 ISUOG. Published by John Wiley & Sons Ltd. ORIGINAL PAPER

2 Copy number by sequencing for invasive fetal testing 95 To compare test and control samples, segmentation algorithms create consecutive read windows, positioned equally in both genomes 7. Read counts, or the number of sequenced fragments per segment, can then be compared. A reduction in read count in a particular test window suggests a deletion in that region, and an increase in reads suggests a duplication. This segmental approach has allowed the resolution of 5 kb CNVs using ultra-low read depths with costs comparable to those of array CGH 8,9. With sequencing, successful detection of CNVs from archived picoquantities of DNA has been reported 8, and the ability to add oligonucleotide barcodes to individual samples means that multiple genomes can be processed simultaneously, reducing costs significantly. Using CNV-Seq, we analyzed a cohort of prenatal samples obtained from abnormal fetuses that had previously failed array CGH quality-control parameters. METHODS Nine fetal samples were selected for CNV-Seq (Table ), all of which had been analyzed previously using the BlueGnome Cytochip Focus (Illumina, San Diego, CA, USA) microarray (version ). This bacterial artificial chromosome (BAC)-based array is optimized specially for use with challenging tissue samples by applying increased probe replication and data smoothing algorithms. For the samples analyzed by CNV-Seq, array CGH experiments had failed owing to inadequate probe-inclusion ratios and increased signal-to-noise ratios, as measured by the standard deviation of each autosome on the array. Two samples used were known to be abnormal; a sample with aneuploidy was included to assess the effect of multiplexing on the detection of large-scale abnormalities and another sample appeared to contain a submicroscopic rearrangement on array CGH, but the quality scores were too poor to report. Genomic DNA was extracted from uncultured chorionic villi and amniotic fluid using the Gentra Puregene Micro-DNA kit (Qiagen, Venlo, Limburg, The Netherlands). DNA concentration and purity were determined using the Quant-iT PicoGreen dsdna BR (Life Technologies, Inchinnan Business Park, Paisley, UK) assay and the Agilent Bioanalyser Genechip (Agilent Technologies, Santa Clara, CA, USA). For each sample, μg of genomic DNA was used to prepare the DNA libraries for sequencing on an Illumina GAIIx (Illumina) sequencing platform. Pooled DNA, obtained from 50 normal individuals, was used as a control. Fetal DNA was sheared to approximately base pairs (bp) using adaptive focused acoustics. Fragments were purified using MinElute (Qiagen) columns, and overhanging ends repaired using the End-It DNA End-Repair kit (Epicentre, Cambridge, UK). Using previously described methods, 6-nucleotide barcodes were used to index the samples 8. Fragments were selected by size to 00 bp using gel-cut electrophoresis. Following polymerase chain reaction amplification, the libraries were purified on a QiaQuick (Qiagen) column and quality-checked using an Agilent Bioanalyser DNA 000 LabChip (Agilent), with Table Phenotype and array comparative genomic hybridization (CGH) quality-control (QC) metrics of nine DNA samples obtained from abnormal fetuses Sample Abnormality Array CGH result SD autosome* Probe inclusion (%) Increased NT Failed QC Increased NT 47,XY Isolated cerebral ventriculomegaly at weeks Failed QC but del 7p; dup 0q Increased NT Failed QC Exomphalos at weeks Failed QC Bilateral talipes and cardiac defect Failed QC Increased NT Failed QC Exomphalos Failed QC Increased NT Failed QC *SD < 0.4 is acceptable. > 95% is acceptable. del, deletion; dup, duplication; NT, nuchal translucency. Table CNV-Seq data of nine DNA samples obtained from abnormal fetuses Sample Total read count Mean window size (kb)* Mean resolution (kb) CNV-Seq result Trisomy Normal Normal Normal Normal Normal Normal del 7p; dup 0q Normal *Genomic length containing 00 aligned reads. Span of two read windows. del, deletion; dup, duplication.

3 96 Cohen et al. (a) X Y Chromosome (b) Log ratio Ch/Ch X Y Chromosomal position Figure Sequencing karyogram from CNV-Seq (a) and corresponding array comparative genomic hybridization result (b) across the whole genome of a normal fetus. For CNV-Seq, each data point represents a single read window, each containing 00 reads. The x-axis displays the chromosomal location of the windows. The y-axis shows the haploid equivalents present in the test sample, when compared with the normal control. As expected in a normal fetus, the haploid equivalent is for the entire genome. quantification performed by a Quant-iT Picogreen double-stranded DNA assay (Life Technologies). Equal amounts of each tagged library were then pooled for cluster generation and sequencing using the standard Illumina single-read 76-cycle operating protocol ( instruments/genome_analyzer_iix.html). Data analysis Image analysis, base calling quality scores and alignment were performed using the Illumina CASAVA pipeline (UCSC build hg9). Initially, samples were separated according to their tags. The control genome was then segmented into non-overlapping windows, each containing 00 reads, and the same windows were then applied to each fetal sample. The number of reads falling into each corresponding window was then counted. Subsequent analysis performed in R software (Revolution Analytics, Mountain View, CA, USA ( normalized the read counts across the genome and calculated a log ratio of normalized sample (control read count for each window). A mean log ratio was then generated for consecutive windows and used to produce graphical representations of copy number variation for each patient. RESULTS The fetal samples and one normal control were sequenced in a 0-plex experiment producing a total of reads. The depth of coverage obtained was 8,

4 Copy number by sequencing for invasive fetal testing 97 (a) X Y Chromosome (b) Log ratio Ch/Ch X Y Chromosomal position Figure (a) Sequencing karyogram of a sample from a fetus with trisomy. (b) Noisy array comparative genomic hybridization karyogram from same fetus. calculated by the number of reads/total number of genome base pairs ( 0 9 ) ratio. In Table, the overall read count, mean genomic resolution and variant breakpoints are shown for each sample. A window size of 00 reads was used to analyze the data (Figures and ). Base calling quality scores were satisfactory, with a Phred quality score of 0, indicating base call accuracy of 99.9%. Figure shows the submicroscopic rearrangement detected by both platforms, and the breakpoint positions of these rearrangements are seen in Table. The corresponding array CGH chromosomal ideogram for the deletion on chromosome 7 is also shown (Figure b). In this sample, a.89-mb deletion of nine consecutive probes was detected on chromosome 7p by the array CGH platform. The duplication identified on chromosome 0 was a large rearrangement, with start and end clone positions suggestive of up to 6.9 MB in size. Definitive sizes of the variants were not ascertainable, as the wide tiling of probes in these regions meant that the exact breakpoints lay within a genomic range, somewhere between the span of the start and end clones. In the CNV-Seq karyograms, the terminal deletion of 7p is indicated by the presence of haploid equivalents. On chromosome 0, the duplicated segments are clearly evident as a region of haploid equivalents (Figure ). A small number of unexpected calls were made by the software and required further analysis (Table 4). Interrogation of the genomic positions of these data points revealed that all of these regions were contained within the centromeres of the chromosomes (Figure 4), and were excluded from further study owing to the known problems with read alignment within these areas.

5 98 Cohen et al. (a) (c) (b) Log Ratio Ch Chromosome Log ratio Ch/Ch X Figure Submicroscopic chromosomal abnormality detected by CNV-Seq and array comparative genomic hybridization (acgh) of a sample obtained from a structurally abnormal fetus. (a) CNV-Seq karyogram reveals a deletion of 7p. (b) acgh ideogram for chromosome 7 shows a limited number of probes included in the analysis but the terminal deletion is evident. (c) The duplication of 0p was detected by the CNV-Seq platform. DISCUSSION The findings from our study indicate that poor-quality DNA did not appear to have a deleterious effect on the next-generation sequencing data output. All samples had been analyzed previously by array CGH, but interpretation was not possible owing to the failure of quality-control metrics. Subsequent sequencing of the same degraded DNA by CNV-Seq produced sequencing datasets with acceptable quality scores. However, the main difference between the outputs was the absence of the multiple single-clone calls that had characterized the arrays of these samples. This noise was not evident when using CNV-Seq. As the DNA samples were randomly

6 Copy number by sequencing for invasive fetal testing 99 Table Breakpoints of the submicroscopic chromosomal abnormalities detected in a DNA sample obtained from a structurally abnormal fetus Window position (genomic loci) Submicroscopic abnormality Left flanking Start End Right flanking Minimum size (kb) Maximum size (kb) del 7p dup 0q del, deletion; dup, duplication. Table 4 Calls made by the CNV-Seq algorithm that required further analysis after sequencing of nine DNA samples obtained from abnormal fetuses Sample Chromosome Genomic start position Genomic end position Log ratio Figure 4 A centromeric call in CNV-Seq. Alignment is poor in highly repetitive areas of the genome and a disproportionate number of reads may align in these regions. cut prior to sequencing, the fragmenting effects of DNA degradation may have been attenuated. The digital nature of the analysis, not relying on the assessment of fluorescence intensities, may have resulted in a less spurious signal from experimental contaminants. It is possible that technical errors with in-house DNA extraction or array processing impacted upon the DNA quality. The platform used in this study was a BAC-based array, technology that has been superseded mostly by oligoarrays. Certainly, the problem of single-clone calls has largely disappeared since this transition 0. Recent meta-analysis data would suggest that most prenatal array experiments do not fail. However, in many cohorts, problems with using poor-quality DNA have been overcome by the use of culturing, with up to one in five results relying on cultured samples. Whole genome amplification has been utilized but raises the possible difficulties of amplification bias, and, overall, the prevalence of failed direct DNA in prenatal arrays is not clear. To our knowledge, no failed array CGH samples have been analyzed previously using this particular algorithm, although sequencing of archived tumor and constitutional samples has shown similar advantages 8. Similar work has been reported for postnatal samples, using the same platform, with the detection of genomic variants comparable with that of an oligoarray platform 9. Challenging samples, such as those obtained from pregnancy loss or stillbirth, may be especially suited to analysis by CNV-Seq. For this method of analysis, the resolution was defined as being equal to the length of two genomic read windows. The range of resolution obtained across the samples was between 94.8 and 89. kb. Abnormalities less than the span of two consecutive windows were assumed to remain undetected with this platform, similar to gaps between probes on an array. Low read depths were produced by multiplexing, an approach widely adopted in constitutional genomics. As each lane of the sequencer produces a finite number of reads, multiplexing reduces proportionally the read depth per sample. Processing of up to 96 samples per lane is possible with newer sequencing platforms, allowing a choice of increased depth or higher throughput. In CNV-Seq, high depth of coverage appears to increase sensitivity and detection rates. Ultra-low coverage, as in this project, has the advantage of reduced costs but is adequate only for detecting relatively large-scale variants when compared with deep sequencing. In this small project, the control genome was segmented into non-overlapping windows containing approximately 00 test reads. These windows were then applied to the test genomes. Different sized windows may be selected and the data analysis repeated in silico without the requirement for further experiments, and the ability to manipulate the dataset digitally is another potential advantage over arrays. Figure 5 shows the same chromosomal abnormality as in Figure, analyzed by both a 400 read window and a 50 read window. It is

7 400 Cohen et al. K5F chromosome 7 K5 chromosome 7 (a) (b) Figure 5 Different window sizes used for analyzing chromosome 7. Altering the window size on the data points available for analysis has an effect on results: (a) 400 read window; (b) 50 read window. evident that when the window size is very small, the increased number of data points in the karyograms could be a disadvantage, analogous to increased signal/noise ratios in array CGH experiments. Large window sizes resulted in very few datapoints available for analysis, and a reduced resolution. Digital approaches such as CNV-Seq offer the possibility of flexibly manipulating the resolution using alterations in read depth and window size, rather than changing platforms. For the submicroscopic rearrangement, CNV-Seq allowed the improved assessment of genomic breakpoints. The windows were smaller than the BAC probes, and the genomic positions of the start and end windows were available. Although the read count was not sufficient to determine exact breakpoints, this compared favorably with the array CGH platform. Rare variants in any reference genome may manifest as dosage change in a test sample 9. A potential advantage of next-generation sequencing is that in-silico controls can be utilized for each run, reducing costs. A control sample can be prepared by merging sequencing data from a large number of normal individuals, offering an advantage over array CGH, in which each experiment requires a fresh control to be processed. The software called a number of possible variants that mapped to genomic regions of high repeat sequences. Alignment in these regions is known to be reduced, particularly at the centromeres and telomeres 4. Calls were excluded based on either their small size or their centromeric or telomeric position, and no VOUS that required follow-up were detected. The small number of samples analyzed precludes any conclusions regarding detection rates and VOUS. With a resolution similar to that of oligoarray platforms, it is almost certain that using CNV-Seq on larger cohorts would reveal VOUS. Larger studies are needed before the apparent improvement in false-positive calls can be verified. Laboratory time for a sequencing reaction was similar to that required by array CGH, with data output ready for analysis within days. In spite of the large digital datasets, the time taken to produce a molecular karyogram was 4 h. Microarray platforms target coverage to regions of interest, using increased probe density. In CNV-Seq, genome coverage is proportional to read depth and cannot be targeted to specific regions. Array platforms that interrogate single nucleotide polymorphisms (SNPs) alongside oligoarray data have been shown to increase the detection of pathogenic variation 5 ; CNV-Seq is a low-read depth platform and will not detect SNPs. SNP platforms also utilize digital controls, minimizing the effects of control variation, and are able to detect loss of heterozygosity indicating potential recessive conditions. CONCLUSIONS Genomic diagnosis of fetal abnormalities facilitates both reproductive choice and an accurate assessment of the risk of recurrence for families. Enormous transitions are underway in genomic testing, and the availability of increasingly high-resolution array platforms has been coupled with dramatic falls in sequencing costs. There is now the very real possibility that convergence of technology will result in a single platform for the detection of genomic variation, from SNP to aneuploidy. At present, CNV-Seq and other sequencing-based copy number detection platforms may offer advantages when tissue samples are of poor quality or when array CGH fails. Further assessment is needed to determine the optimal operating parameters for prenatal diagnosis.

8 Copy number by sequencing for invasive fetal testing 40 REFERENCES. Reddy UM, Page GP, Saade GR, Silver RM, Thorsten VR, Parker CB, Pinar H, Willinger M, Stoll BJ, Heim-Hall J, Varner MW, Goldenberg RL, Bukowski R, Wapner RJ, Drews-Botsch CD, O Brien BM, Dudley DJ, Levy B; NICHD Stillbirth Collaborative Research Network. Karyotype versus microarray testing for genetic abnormalities after stillbirth. NEnglJMed0; 67: 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. NEnglJMed0; 67: Hillman S, McMullan D, Hall G, Togneri F, James N, Meller C, Williams D, Wapner R, Maher E, Kilby M. Use of prenatal chromosomal microarray: prospective cohort study and systematic review and meta-analysis. Ultrasound Obstet Gynecol 0; 4: Bi W, Breman AM, Venable SF, Eng PA, Sahoo T, Lu XY, Patel A, Beaudet AL, Cheung SW, White LD. Rapid prenatal diagnosis using uncultured amniocytes and oligonucleotide array CGH. Prenat Diagn 008; 8: Hehir-Kwa JY, Egmont-Petersen M, Janssen IM, Smeets D, van Kessel AG, Veltman JA. Genome-wide copy number profiling on high-density bacterial artificial chromosomes, single-nucleotide polymorphisms, and oligonucleotide microarrays: a platform comparison based on statistical power analysis. DNA Res 007; 4:. 6. Xie C, Tammi MT. CNV-seq, a new method to detect copy number variation using high-throughput sequencing. BMC Bioinformatics 009; 0: Koboldt DC, Larson DE, Chen K, Ding L, Wilson RK. Massively parallel sequencing approaches for characterization of structural variation. Methods Mol Biol 0; 88: Wood HM, Belvedere O, Conway C, Daly C, Chalkley R, Bickerdike M, McKinley C, Egan P, Ross L, Hayward B, Morgan J, Davidson L, MacLennan K, Ong TK, Papagiannopoulos K, Cook I, Adams DJ, Taylor GR, Rabbitts P. Using next-generation sequencing for high resolution multiplex analysis of copy number variation from nanogram quantities of DNA from formalin-fixed paraffin-embedded specimens. Nucleic Acids Res 00: 8: e5. 9. Hayes JL, Tzika A, Thygesen H, Berri S, Wood HM, Hewitt S, Pendlebury M, Coates A, Willoughby L, Watson CM, Rabbitts P, Roberts P, Taylor GR. Diagnosis of copy number variation by Illumina next generation sequencing is comparable in performance to oligonucleotide array comparative genomic hybridisation. Genomics 0; 0: Pinto D, Darvishi K, Shi X, Rajan D, Rigler D, Fitzgerald T, Lionel AC, Thiruvahindrapuram B, Macdonald JR, Mills R, Prasad A, Noonan K, Gribble S, Prigmore E, Donahoe PK, Smith RS, Park JH, Hurles ME, Carter NP, Lee C, Scherer SW, Feuk L. Comprehensive assessment of array-based platforms and calling algorithms for detection of copy number variants. Nat Biotechnol 0; 9: Meyer M, Stenzel U, Myles S, Prüfer K, Hofriter M. Targeted high-throughput sequencing of tagged nucleic acid samples. Nucleic Acids Res 007; 5: e97.. Metzker ML. Sequencing technologies the next generation. Nat Rev Genet 00; : 46.. Nord AS, Lee M, King MC, Walsh T. Accurate and exact CNV identification from targeted high-throughput sequence data. BMC Genomics 0; : Zhao M, Wang Q, Wang Q, Jia P, Zhao Z. Computational tools for copy number variation (CNV) detection using next-generation sequencing data: features and perspectives. BMC Informatics 0; 4 (Supp ): S. 5. Ganesamoorthy D, Bruno DL, McGillivray G, Norris F, White SM, Adroub S, Amor DJ, Yeung A, Oertel R, Pertile MD, Ngo C, Arvaj AR, Walker S, Charan P, Palma-Dias R, Woodrow N, Slater HR. 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? BJOG 0;0: