Exon Array CGH: Detection of Copy-Number Changes at the Resolution of Individual Exons in the Human Genome

Abstract

The development of high-throughput screening methods such as array-based comparative genome hybridization (array CGH) allows screening of the human genome for copy-number changes. Current array CGH strategies have limits of resolution that make detection of small (less than a few tens of kilobases) gains or losses of genomic DNA difficult to identify. We report here a significant improvement in the resolution of array CGH, with the development of an array platform that utilizes single-stranded DNA array elements to accurately measure copy-number changes of individual exons in the human genome. Using this technology, we screened 31 patient samples across an array containing a total of 162 exons for five disease genes and detected copy-number changes, ranging from whole-gene deletions and duplications to single-exon deletions and duplications, in 100% of the cases. Our data demonstrate that it is possible to screen the human genome for copy-number changes with array CGH at a resolution that is 2 orders of magnitude higher than that previously reported.

Introduction

The extent to which genomic copy-number polymorphisms (CNPs) contribute to human genetic diversity is not known. Recent studies have demonstrated the presence of CNPs, a proportion of which encompass genes, in the genomes of normal individuals (Iafrate et al. 2004; Sebat et al. 2004). This suggests that these variants may be important in our understanding of phenotypic variation or may predispose to or directly cause disease. On the basis of current knowledge, ∼5%–6% of gene mutations that are causative of inherited disorders are copy-number changes defined as gross deletions or duplications (Armour et al. 2002), although this frequency may represent an underestimate. These disease-causing copy-number changes can range in size from 100 bp to several megabases and can encompass as little as a single exon and as much as entire genes or several genes.

The ability to detect CNPs and pathogenic chromosomal imbalances has been dramatically improved through the use of array-based comparative genome hybridization (array CGH) (Solinas-Toldo et al. 1997; Pinkel et al. 1998). Array CGH offers high sensitivity and dynamic range to quantitatively measure from single copy-number losses or gains to high copy-number amplifications, as well as sufficient resolution and scalability for complete genomewide scans (Snijders et al. 2001; Fiegler et al. 2003; Vissers et al. 2003; Iafrate et al. 2004; Inazawa et al. 2004; Ishkanian et al. 2004). Currently, the most robust array CGH approaches utilize large genomic clone inserts as array elements but have a maximum resolution of ∼40–50 kb (Albertson and Pinkel 2003; Snijders et al. 2003; Mantripragada et al. 2004a). The use of oligonucleotides (Lucito et al. 2000, 2003; Bignell et al. 2004; Carvalho et al. 2004; Sebat et al. 2004), cDNA clones (Pollack et al. 1999, 2002; Heiskanen et al. 2000), or sequence-defined PCR products (Mantripragada et al. 2003, 2004b) for array CGH allow this resolution to be theoretically increased solely on the basis of the size of the genomic region covered by the array element. However, because of limitations of the technologies, the effective resolution for these methods is no greater than 15–30 kb, since they rely on averaging measurements taken from multiple array elements, pooling PCR products to be spotted as single array elements, or reducing the complexity of the human genome to detect and quantitate copy-number changes. Therefore, it has not yet been demonstrated that array elements covering small regions of the genome can deliver substantially higher resolution than that of large genomic clone array CGH to identify small genomic deletions and duplications (100 bp to 15 kb).

The ability to detect copy-number changes encompassing such small regions of the genome has thus far been achievable only by using other types of molecular assays. Apart from the more traditional approaches of Southern analysis, FISH, and quantitative PCR (Armour et al. 2002), more-recent approaches have included the development of multiplex amplifiable probe hybridization (MAPH) (Armour et al. 2000; Sellner and Taylor 2004) and the multiplex ligation-dependent probe amplification (MLPA) (Schouten et al. 2002; Sellner and Taylor 2004). For MAPH and MLPA, the resolution is at the level of individual exons, and they are becoming widely used as research and diagnostic tools, with very high reliability for identifying and quantitating copy-number changes in human disease genes (White et al. 2002, 2003, 2004; Akrami et al. 2003; Sellner and Taylor 2004). However, because both methods rely on multiplexing, the number of measurements obtainable from a single assay is limited. Therefore, the development of approaches that provide robust measurement precision of copy-number changes, scalability, and very high resolution have thus far been unavailable in the field of human molecular genetics.

We describe here an array CGH–based approach that addresses all of these issues for the analysis of copy-number changes in the human genome. We have developed an array platform that allows single strands of DNA derived from double-stranded PCR products to be retained on the surface of a slide through the use of 5′-aminolink chemistry. This platform improves the signal:noise ratio, such that it is possible to detect individual exons in the human genome and to quantify their copy number accurately. We refer to this method as “exon array CGH.” To demonstrate the utility of the approach, we constructed an array containing 162 exons that collectively span five human genes (COL4A5 [MIM 303630], DMD [MIM 300377], NF2 [MIM 607379], PLP1 [MIM 300401], and PMP22 [MIM 601097]) involved in inherited genetic disorders (see table 1). We have analyzed a series of 31 DNA samples from patients affected with these disorders and have characterized copy-number changes that have been validated with other molecular methods for all 31 samples. Our method is 2 orders of magnitude more sensitive than other forms of array CGH and provides resolution and accuracy that are similar to other current methods of screening genes for copy-number changes at the level of the exon. Since exon array CGH is completely scalable, this new molecular tool will provide the means for researchers to screen the human genome at high resolution to identify and annotate novel CNPs and causative mutations in normal and disease states, respectively, as well as to facilitate array-based applications to study other aspects of genome biology.

Table 1.

Human Disease Genes Represented on the Exon Array^[Note]

Gene	ChromosomalLocation	GenomicSize(kb)	No. ofExonsa	Disease	Frequencyin Population	Pathogenic Copy-Number Changesb	Technical Consideration
DMD	Xp21.1	2,400	79 (+1)	Duchenne muscular dystrophy (DMD [MIM 310200])/Becker muscular dystrophy (MIM 300376)	1/3,500c	60–70d	Complex gene for diagnostic screening
NF2	22q12	95	17	Neurofibromatosis type 2 (NF2 [MIM 101000])	1/40,000e	20–30e	High repeat content results in quantitation issues for array CGHf
COL4A5	Xq22.3	257.6	51 (+2)	Alports syndrome (AS [MIM 301050])	1/5,000g	10–15h	Specificity issues, since it is a member of a six-gene triple helical collagen family
PMP22	17p12	35.6	5	Charcot-Marie Tooth type 1 disease (CMT1A [MIM 118220]); hereditary neuropathy with liability to pressure palsy (HNPP [MIM 162500])	10/100,000–40/100,000i	70 dupl (for CMT1A);84 del (for HNPP)j
PLP1	Xq22.2	15.8	7	Pelizaeus-Merzbacher disease (PMD [MIM 312080])	Extremely rarek	60–70l

Note.— Genomic characteristics and clinical/technical relevance for inclusion on the array of the five human disease genes are shown.

^aThe number of exons shown in parentheses represents alternative exons found in some transcripts that were included as array elements.

^bAs a percentage of all mutations.

^cDMD is the most common inherited neuromuscular disorder (Worton and Thompson 1988).

^dDen Dunnen et al. 1989; Gillard et al. 1989.

^eBruder et al. 2001.

^fMantripragada et al. 2003.

^gAtkin et al. 1988.

^hLemmink et al. 1997; Plant et al. 1999.

ⁱRautenstrauss et al. 2000.

^jNelis et al. 1996.

^kHeim et al. 1997.

^lSistermans et al. 1998; Mimault et al. 1999.

Material and Methods

PCR Amplification and Microarray Fabrication

Primers pairs for exons of the COL4A5, DMD, NF2, PLP1, and PMP22 genes (table 2); for amplicons covering the 3′ UTRs of genes on chromosome 22 (normalization controls); and for other amplicons used here were obtained from primer sets published elsewhere (Strautnieks et al. 1992; Plant et al. 1999; Leiden University Medical Center Web site) or were designed using the relevant genomic sequence and the Primer3 software and Web site (Rozen and Skaletsky 2000; Whitehead Institute). All amplicons used in the present study were repeat free. When possible, primer pairs were designed or obtained for each exon and, ideally, were located in the upstream and downstream intronic sequences flanking the exon. In a few instances, however, primers were designed within the exonic sequence. When two adjacent exons were separated only by a very small intron, a single pair of primers was designed to contain both exons. Primer pairs and all amplicons sequences were compared with the entire human genome sequence by use of e-PCR (Schuler 1997) and BLASTN to identify any potential cross-reacting DNA sequences. In cases in which the PCR product for a given exon did not accurately or reproducibly report copy-number values, additional PCR assays were designed and tested on the array.

Table 2.

Primer Pairs Used to Amplify Exons for Array CGH

	Primer(5′→3′)
Amplicon Namea	1(Forward)	2(Reverse)	Sizeb(bp)	Genome Location
COL4A5-01	CTTCCACTCTTAAAAAGCTTC	GGAGGGGAGGAAGGACTTAC	258	X:106455055–106455313
COL4A5-01ver2	CAAGCCTCACTGTCCCTCTC	AAGGACTTACCGCAGCCTCT	215	X:106455088–106455303
COL4A5-02	TGGTGTAGCTTTCCATCTTTAGC	TGCCAATCTAAGACCTCACC	391	X:106554520–106554911
COL4A5-03	TCTCAACCATGCCTGTGCTT	GATGTGACACCTAGTCCCAC	228	X:106574085–106574313
COL4A5-03ver2	AGGGAGAGAGAGGGTTTCCA	CCCGTCATTCCATTGTTTCA	228	X:106574148–106574376
COL4A5-04	TCACAGATGTTTACAGTAGT	GGTCTTTTCCAATTGTCTCA	237	X:106578864–106579101
COL4A5-04ver2	TGTCAGCGATTTCCTTCTCA	GAATTCCATCATCACCCTGAA	244	X:106578740–106578984
COL4A5-05/06	TGGGTTGTCATTTAGTTTAAGGAT	ATGGATGGATCTCACCTTGG	249	X:106583674–106583923
COL4A5-07	GGAAAGTGAAGGCTAATGAA	GCATTGGGCTCTCTCACTAC	278	X:106586353–106586631
COL4A5-08	GAAGTTGGTAATATAGCTTTCTCC	AAGACAGAAGTTTGAGTCATTATCC	323	X:106586692–106587015
COL4A5-09	TTTATTACTGAGATGGCCTAATC	GAGGGATTGTTGTAATCTTCTGG	278	X:106588528–106588806
COL4A5-10	CTGGGCGACACAAGTGAGA	TTAAAGAAGAGAGCTTACCATC	181	X:106590896–106591077
COL4A5-10ver2	CCTACCTGGTCCCACTGGTAT	TCTGAAATGGCCAGAATTGA	352	X:106590998–106591350
COL4A5-11/12	ACCAGGACTTCCAGGACCTAA	AAGAGAAAGCTGAGGGCTGA	200	X:106593052–106593252
COL4A5-13	TTCAGCCCTCAGCTTTCTC	TCCCTACTTACTCACCTGATCTCC	254	X:106593231–106593485
COL4A5-14/15	GATTTCCTTTCCCCTACTACTG	GAATATTAGCAGTTACATCAC	258	X:106595592–106595850
COL4A5-16	GTCCAAGCCAGGAATAAAGC	GCTTCTCTGCCACCTTATGTTATAC	219	X:106595972–106596191
COL4A5-17	TGTGCTGATGTCACCCTATCC	TTTCTGCAACATGGACTGTG	302	X:106597935–106598237
COL4A5-18	AGACACCATCACAGGTTAGGC	AAGTAGGGAGTATGCGATTAAGG	304	X:106599391–106599695
COL4A5-18ma	CTCTTATATTCTTGAGGAAATTG	GCGATTAAGGCACAAAAATG	184	X:106599498–106599682
COL4A5-19	TTTTCTTTGGTAATAAAGG	AAGGCCATAAATGCAATCTC	177	X:106601683–106601860
COL4A5-20	TCTTGCTAATGACTTTCCGGTAT	CAACAAATGAGGCCAACCTT	567	X:106605843–106606410
COL4A5-21	TCTGGCTTGTCAGGCTTTC	GTTTGGAGATCAAACCTTTCAC	179	X:106606567–106606746
COL4A5-22	GTTCTTGGCACACAGTAAGC	GAGGCATAATTTACCTGGAGGAC	290	X:106610412–106610702
COL4A5-23	TGTGTTAGGATCTCTTGGTTTCC	CCCTTCAACCCATCCAATAAC	286	X:106612076–106612362
COL4A5-24	CTTTTTTTCCTTACTCATTTC	AACCAAAAATATCAAACCAAC	240	X:106612440–106612680
COL4A5-24ver2	TTCCAGGAATGAAGGGTGAC	AGCATCAGTCCCATCCTTTG	211	X:106612533–106612744
COL4A5-24ver3	CTCATTTCAGGGCATTCCAG	CAGCATCAGTCCCATCCTTT	292	X:106612453–106612745
COL4A5-25	ATATGTTTCTGTATTAAAC	TAAGCACCAAGTTTAAAAC	239	X:106613755–106613994
COL4A5-26	CATCTTGAATCATGGAAGTCCT	ACCATCTCTGCCTGGGTTAC	311	X:106616245–106616556
COL4A5-27	TGGCCCTGATGGCTTCTTTC	TGCTACCCATTCCAACATACC	185	X:106616911–106617096
COL4A5-27ver2	TCCATAGGTGACCCTGGACT	CATGCTAAACTTCTAACCTACTCTGC	159	X:106616964–106617123
COL4A5-28	AATGTGGGTTTGGGAATTTG	GGCTTGGGATGATATCTGAGT	273	X:106617971–106618244
COL4A5-29	TTTGTCATGTGTATGCTC	AGCAAGTTGAGATGCAGTGACAG	204	X:106621807–106622011
COL4A5-29ver2	CCAGGACTTCCAGGTTTCAA	AACAGAATGGCAGCATAGGG	230	X:106621867–106622097
COL4A5-30	TTTTCTTGCTGAATGAATGC	CACTTTATTGATGAGCTAAC	301	X:106629880–106630181
COL4A5-31	ATTGATATTGTATTAACTAG	GAAAACTTTAAACAAATTAC	208	X:106635325–106635533
COL4A5-32(ma)	ATAGTTTTCTGGTTGACATC	ATTCTGTACTGACATAAAGC	250	X:106636801–106637051
COL4A5-33	TTGTGTTTTCACACACATTG	CATAAATAAATTCATTCACTC	200	X:106637735–106637935
COL4A5-33ver2	TAGGTGATGATGGCTTGCAG	GCCTTTGTTGGTGAATTCTG	258	X:106637759–106638017
COL4A5-34	TTCGTGTGAGTCCAGTGCTAA	TACCTGGTGGTCCTGGTAATC	260	X:106639164–106639424
COL4A5-35	CCATTCCCATGAAACCAGAC	TGATTCTATCAACACCAGCCTC	258	X:106640721–106640979
COL4A5-36	ATATATCACATATTTTCAACAG	CCTAAAACTATATGCCAAAG	187	X:106641274–106641461
COL4A5-37	TTCAGGGTGAGCCTGGTCT	AGCTGCTCTGTGATACTGGTTG	203	X:106670412–106670615
COL4A5-38	CCAACAAACACCAGCATCTTT	GCTGAACATGATTTGACTTTCC	303	X:106680454–106680757
COL4A5-39	AAACTGGGTGTAACCTGCTG	GGGAAAGTGTGTGGTAGCTTAG	255	X:106681547–106681802
COL4A5-40	TTTGGCCTTGTTTCAGTTTG	ACACTAGGTTTGAGTTTTCTGGGTTC	250	X:106682152–106682402
COL4A5-41	ATCTTCTAATTATACTTTACTTTC	CATTCTCCTACCACTCACGG	240	X:106683377–106683617
COL4A5-41A	AATGATGACATTGGGCCTTG	CTTTCCAATAGACAGAAACACTGG	234	X:106685272–106685506
COL4A5-41B	AAGAAATGCTGGCAAAGAAC	TTGAGGATAACAGCACTCTCC	273	X:106689688–106689961
COL4A5-41ver2	TGGGTAGATTTGGGATTTGG	GGTTGCCTTTAGGTCCTTCC	298	X:106683263–106683561
COL4A5-42	AATGTCGTCATTTGCTGTGGATTA	CATCAGATATCTACTTCCATTTCC	190	X:106692557–106692747
COL4A5-43/44	GTAACATTAATGATTTTATTTATT	TATAACTATCTTCAGGAATAAGTC	333	X:106695738–106696071
COL4A5-45	GGTCACTTTGGCTTCCATTTC	GGTAAGCCTTGAGGTCCTTG	195	X:106696778–106696973
COL4A5-46	GCATCCCAGAAGCAAATGGA	AGACTTTCCCTCCACCAACAG	347	X:106700909–106701256
COL4A5-47	GTATACTGATTATTTCGTGG	GGAAATTAGATATTGATTATC	264	X:106702542–106702806
COL4A5-48	CTTTACTGTTTTCTCTCCAAATCT	AAAAGTCACAGCTAAATCAATGCC	236	X:106707804–106708040
COL4A5-49	ATTATGTTCCTTCTCCTTTTCCTT	GACAAATGCAAGGAAGAGTG	173	X:106709863–106710036
COL4A5-50	GCGGCACATTTTTCCTTGTC	CTGAATTAAAGCTATAAGCAC	227	X:106710325–106710552
COL4A5-51	GATCTGATTGTCTTATTTCTTATT	TAAAACACAAAAGGAATTCTTCAA	140	X:106711353–106711493
COL4A5-51ver2	TGACCACTAATGGCTCTTAAACC	CGGCAGCAGTAGTAAAGTTGG	367	X:106711200–106711566
DMD-01A	TCTGGCTCATGTGTTTGCTCCGAGGTATAG	CTTCCATGCCAGCTGTTTTTCCTGTCACTC	332	X:32718801-32719133
DMD-01B	TAGACAGTGGATACATAACAAATGCATG	TTCTCCGAAGGTAATTGCCTCCCAGATCTGAGTCC	529	X:32591003–32591532
DMD-01Zma	GAAGGCGGGTCACTTGCTTGTGCGCAG	CAATCTACCTAATTAGTGAGCTTG	419	X:32590744-32591163
DMD-02	CACTAACACATCATAATGG	GATACACAGGTACATAGTC	233	X:32399606–32399839
DMD-02ma	TAATTTGGATGCCCCAAACCAGC	GTAACAAACCATTCTTACCTTAGA	253	X:32399664-32399916
DMD-03(ma)	TCATCCGTCATCTTCGGCAGATTAA	CAGGCGGTAGAGTATGCCAAATGAAAATCA	408	X:32229094–32229502
DMD-04	AGTAGATTGTCGGTCTCTCTG	CCAAAGCCCTCACTCAAAC	203	X:32224255-32224458
DMD-04ma	TTGTCGGTCTCTCTGCTGGT	CAAAGCCCTCACTCAAACATGAAGC	196	X:32224256–32224451
DMD-04ver2	GGAGCAGCCTATCAGGTCAG	ATCCACAAGAGTTCATGCCC	307	X:32224082-32224389
DMD-05	CAACTAGGCATTTGGTCTC	TTGTTTCACACGTCAAGGG	225	X:32202758–32202983
DMD-05ma	GCAACTAGGCATTTGGTCTCTTACCT	CCATTCATCAGGATTCTTACCTGCC	167	X:32202817-32202984
DMD-06(ma)	GTTTGCATGGTTCTTGCTCA	CTTTCACGCTCCGCTAATGT	466	X:32195777–32196243
DMD-06ver3	CCACATGTAGGTCAAAAATGTAATGAA	GTCTCAGTAATCTTCTTACCTATGACTATGG	202	X:32195991–32196193
DMD-07	GCATGGAAGTAAATCTCATGGAAC	GTGTAGAAATGACAAGTCTCAGATG	278	X:32188956–32189234
DMD-07ma	AAGGACTATGGGCATTGGTTGTCA	TGTGTAGAAATGACAAGTCTCAGA	314	X:32188955–32189269
DMD-08(ma)	GTCCTTTACACACTTTACCTGTTGAG	GGCCTCATTCTCATGTTCTAATTAG	360	X:32078636–32078996
DMD-09(Zma)	CTTCTCTGCAGATCACG	CGAGGAGATAAAAGGCAC	245	X:32077307–32077552
DMD-09ver2	GTAGTCCTTTCGGGTTACTTATGG	AACAAACCAGCTCTTCACGAGGAGA	321	X:32077290–32077611
DMD-10	ATTGTGCAGCATTTGGAAGC	GTTGGAATCCCAAGCACATC	332	X:32024372–32024704
DMD-11	CAAATAAAACTCAAAACCACACC	CTTCCAAAACTTGTTAGTCTTC	301	X:32023626–32023927
DMD-12	GATAGTGGGCTTTACTTACATCCTTC	GAAAGCACGCAACATAAGATACACCT	332	X:31993817–31994149
DMD-12ma	GATAGTGGGCTTTACTTACATCCTTC	TAATTCTCTCCCATCAACCATGTCATCT	405	X:31993744–31994149
DMD-13(ma)	AATGGGAGTACCTGAGATGTAGCAGAAAT	CTGACCTTAAGTTGTTCTTCCAAAGCAG	238	X:31975294–31975532
DMD-14/15	TTTGCTGATGCTGTGCTTGATTGTC	CAATAGCATAGAAGAGACTAA	436	X:31952988–31953424
DMD-16	TCTATGCAAATGAGCAAATACACGC	GGTATCACTAACCTGTGCTGTACTC	290	X:31945232–31945522
DMD-17	TTTCGATGTTGAGATTACTTTCCC	AAGCTTAAGATGCTCTCACCTTTTCC	414	X:31924683:31925096
DMD-17ma	GCTATTTTGATCTGAAGGTCAATCTACC	AAGCTTGAGATGCTCTCACCTTTTCC	326	X:31924683:31925008
DMD-18	GGAGTCTCAGATTGAGAAAAGAATG	CAAGCAGCACAAAATGAGTACAG	279	X:31897502–31897781
DMD-18ver2	GGCATCCCTAGTCAGTCACAG	TTCACAGCTGGATTACTCGC	314	X:31897337–31897651
DMD-19(ma)	GATGGCAAAAGTGTTGAGAAAAAGTC	TTCTACCACATCCCATTTTCTTCCA	459	X:31881080–31881539
DMD-20	TGGCTTTCAGATCATTTCTTTC	AAATACCTATTGATTATGCTCC	357	X:31870772–31871129
DMD-20ver2	ATGCTCCAAATGGAAGGAGA	GTGGATCGAATTCTGCCAGT	222	X:31870787–31871009
DMD-21	GCAAAATGTAATGTATGCAAAG	ATGTTAGTACCTTCTGGATTTC	319	X:31864382–31864701
DMD-22	TTGACACTTTGCCACCAATG	GATAAGCGTGCTTTATTGTTTTGAC	186	X:31851663–31851849
DMD-23	GTTTATAACTGATAGAAGATCATC	TTTACAGTTTACAGTGTATCGTTAG	426	X:31847964–31848390
DMD-24	TTGGGCCTGTGTTTAGACATA	AAATCCACCCCAGCTGTAAAA	291	X:31844040–31844331
DMD-25	TGTGGCAGTAATTTTTTTCAG	AGGAAATCTTAGTTAAGTACG	260	X:31842925–31843185
DMD-26	TGTGGAAGGTCTATGCCAGA	CTTCATCTCTTCAACTGCTTTCTGT	179	X:31834204–31834383
DMD-27	AGAGCTAAAGAAGAGGCCCA	CACATATGACCATGTATTGACAT	238	X:31827943–31828181
DMD-28	GAAGTTTTAATAATGAAATGGCAAAA	TGACCTCTTTTAATACTGCATAT	275	X:31820665–31820940
DMD-29	CCAATGTATTTAGAAAAAAAAGGAG	GCAAATTAGATTAAAGAGATTTTTCAC	243	X:31817745–31817987
DMD-30	TACAGAAAAGCTATCAAGAG	AAAAACAAAAGAATGGAAGC	261	X:31791254–31791515
DMD-30ma	AGGCTGTAAGGAGGCAAAAGTTGC	GATGTACTTGCCTGGGCTTCCTGAGGC	175	X:31791283–31791458
DMD-31	ACAGGTGTGAACCACCACTC	ACTGGAGTATAATGCCCAACG	321	X:31769557–31769878
DMD-32	GACCAGTTATTGTTTGAAAGGCAAA	TTGCCACCAGAAATACATACCACACAATG	253	X:31769023–31769276
DMD-33	TAGATATTGACCACCGCTGC	TTGCTAAGACTCTAATCATAC	402	X:31765756–31766158
DMD-34	CAGAAATATAAAAGTTCCAAATAAGTG	CATGTTAATACTTCCTTACAAAATC	338	X:31759973–31760311
DMD-35	CCGTTTCATAAGCATTAAATC	AGCTTCTAGCCTTTTCTC	271	X:31744540–31744811
DMD-36	TGTCTAACCAATAATGCCATG	CTGGTGTACAATTTGGACA	215	X:31744080–31744295
DMD-37	CTTTCTCACTCTTCTCGCTCAC	TTCGCAAGAGACCATTTAGCAC	335	X:31742278–31742613
DMD-38	GGTTTATGTTTCTAATAAAAAGTAA	ATTTATTTCCACTCCTAGTT	231	X:31727899–31728130
DMD-39	AAAGAAAGGCTATGAGCACAGT	TTTCTGATGACTAAGTCTGAAGCAG	404	X:31725377–31725781
DMD-40	CTGCAGCCAGAAGTGCACTA	GGAAATGCATCAAATCAAAGA	340	X:31722552–31722892
DMD-40ver2	AATAACTGCAGCCAGAAGTGC	ATCTCTGGGCTCAGGTAGGC	206	X:31722691–31722897
DMD-41	GTTAGCTAACTGCCCTGGGCCCTGTATTG	TAGAGTAGTAGTTGCAAACACATACGTGG	275	X:31721593–31721868
DMD-42	CACACTGTCCGTGAAGAAACGATGATGG	CTTCAGAGACTCCTCTTGCTTAAAGAGAT	195	X:31689624–31689819
DMD-42ma	CACACTGTCCGTGAAGAAACGATGATG	TTAGCACAGAGGTCAGGAGCATTGAG	155	X:31689664–31689819
DMD-43(ma)	GAACATGTCAAAGTCACTGGACTTCATGG	ATATATGTGTTACCTACCCTTGTCGGTCC	357	X:31667054–31667411
DMD-44	CTTGATCCATATGCTTTTACCTGCA	TCCATCACCCTTCAGAACCTGATCT	268	X:31596364–31596632
DMD-44ma	GTTGTGTGTACATGCTAGGTGTGTA	TCCATCACCCTTCAGAACCTGATCT	426	X:31596364–31596790
DMD-45	AAACATGGAACATCCTTGTGGGGAC	CATTCCTATTAGATCTGTCGCCCTAC	547	X:31347855–31348402
DMD-45ma	CTTTCTTTGCCAGTACAACTGCATGTG	CATTCCTATTAGATCTGTCGCCCTAC	307	X:31347855–31348162
DMD-46	CCAGTTTGCATTAACAAATAGTTTGAG	AGGGTTAAGAAGAAATAAAGTTGTGAG	373	X:31311470–31311843
DMD-46ma	GCTAGAAGAACAAAAGAATATCTTGTC	CTTGACTTGCTCAAGCTTTTCTTTTAG	148	X:31311622–31311770
DMD-47	CGTTGTTGCATTTGTCTGTTTCAGTTAC	GTCTAACCTTTATCCACTGGAGATTTG	181	X:31309131–31309312
DMD-48	TTTGGCTTATGCCTTGAGAA	CGTCAAATGGTCCTTCTTGG	235	X:31254735–31254970
DMD-49(ma)	GTGCCCTTATGTACCAGGCAGAAATTG	GCAATGACTCGTTAATAGCCTTAAGATC	439	X:31216049–31216488
DMD-50(ma)	CACCAAATGGATTAAGATGTTCATGAAT	TCTCTCTCACCCAGTCATCACTTCATAG	271	X:31199434–31199705
DMD-51(ma)	GAAATTGGCTCTTTAGCTTGTGTTTC	GGAGAGTAAAGTGATTGGTGGAAAATC	388	X:31153426–31153814
DMD-52	ATGCATCTTGCTTTGTGTGTC	GGCTGGTCTCACAATTGTAC	254	X:31109130–31109384
DMD-52ma	AATGCAGGATTTGGAACAGAGGCGTCC	CATTATGGACTGAAAATCTCAGCAC	265	X:31109021–31109286
DMD-53(ma)	TTGAAAGAATTCAGAATCAGTGGGATG	CTTGGTTTCTGTGATTTTCTTTTGGATTG	212	X:31058917–31059129
DMD-54	GTTTGTCCTGAAAGGTGGGTTAC	TTATCGTCTTGAACCCTCCCAAG	471	X:31037406–31037877
DMD-55	AATTTAGTTCCTCCATCTTTCTCT	AAATACATCAGGCTGTATAAAAGC	409	X:31007090–31007499
DMD-56	TTTGTTTGGTAATTCTGC	CTGAAATTGGATGATTTAC	336	X:30886706–30887042
DMD-57	TCTGACATGGTACGCTGCTG	TGACCCTTGGGTGAGAAGAG	317	X:30876201–30876518
DMD-58	TTTTGAGAAGAATGCCACAAGCC	AAAATATGAGAGCTATCCAGACC	279	X:30858422–30858701
DMD-59	CCTGAGGAGAGAGCCCAGA	TCGAGGTGATCTTGGAGA	251	X:30857658–30857909
DMD-59ver2	TTGTGGGAAGATAACACTGCAC	ACTCGGCTTCTACGAAAGCA	325	X:30857560–30857885
DMD-60	TAAATATTCTCATCTTCCAATTTGC	TTACTGTAACAAAGGACAACAATG	231	X:30823983–30824214
DMD-60ma	AGGAGAAATTGCGCCTCTGAAAGAGAACG	CTGCAGAAGCTTCCATCTGGTGTTCAGG	139	X:30824023–30824162
DMD-61	ATCCTTTGTGTTTGGCCTTG	AGAAAGTGCTGAGATGCTGG	279	X:30728103–30728382
DMD-62	TAATGTTGTCTTTCCTGTTTGCG	ATACAGGTTAGTCACAATAAATGC	185	X:30703086–30703271
DMD-62ma	GTCTTTCCTGTTTGCGATGAATTTGACC	CTCACTTGTGAATATACAGGTTAGTCAC	191	X:30703073–30703264
DMD-63	TACTCATTGTAAATGCTAAAGTC	TAGCAAGTAACTTTCACACTGC	193	X:30640412–30640605
DMD-64	TTCTGATGGAATAACAAATGCT	CATTCTAGGCAAACTCTAGGC	286	X:30602421–30602707
DMD-65	TATGAGAGAGTCCTAGCTAGG	TAAGCCTCCTGTGACAGAGC	347	X:30588992–30589339
DMD-66	GTCAGTAATTGTTTTCTGCTTTG	ATAAGAACAGTCTGTCATTTCCC	210	X:30586037–30586247
DMD-66ver2	GAGGATCCGTGTCCTGTCTT	TTTGACAAGGAATGGCACAA	241	X:30585957–30586198
DMD-66ver4	AAACTGGCATCATTTCCCTG	TTCCGACTTACAGAAAGAGGTT	364	X:30585811–30586175
DMD-67	AATTGCTACTGGAATTGAGTTGG	AAGAATAAATATGTTACCTAGAAGG	286	X:30583466–30583752
DMD-68	TAATCGAACTGATATACACCTCC	ACTAACAGCAACTGGCACAGG	351	X:30562185–30562536
DMD-69	GAACGTGGTAGAAGGTTTATTAAA	CTAACTCTCACGTCAGGCTG	231	X:30559879–30560110
DMD-70	TGGTCATTAGTTTTGAAATCATC	CATCAAACAAGAGTGTGTTCTG	237	X:30558172–30558409
DMD-71	TGATGGCAGTCACACAACTG	GAGCGAATGTGTTGGTGGTAG	268	X:30557424–30557692
DMD-72	AAAGCATTCTAGGCCATGTGT	ATGCACGTTAGAGGGCAAGT	307	X:30552937–30553244
DMD-73	ACGTCACATAAGTTTTAATGAGC	ATGCTAATTCCTATATCCTGTGC	202	X:30551795–30551997
DMD-74	ACCAAAACCTTTGATTTTATTTTCC	TTTCTATGTGTGCAAGTGTATGC	248	X:30548923–30549171
DMD-75	TCTTTTTTACTTTTTTGATGC	AGTGCTCTCTGAGGTTTAG	344	X:30526744–30527088
DMD-76	GTAATTCTGTTTTCTTTTGGATGAC	CTACCTTTCTTCAGACAACAAAAT	200	X:30525786–30525986
DMD-77	TAATCATGGCCCTTTAATATCTG	GATACTGCGTGTTGGCTTCC	270	X:30513537–30513807
DMD-78	TTCTGATATCTCTGCCTCTTCC	CATGAGCTGCAAGTGGAGAGG	231	X:30506024–30506255
DMD-79	AGAGTGATGCTATCTATCTGCAC	TGCATAGACGTGTAAAACCTGCC	349	X:30501186–30501535
NF2-01	GGGCTAAAGGGCTCAGAGTG	AACCTCTCGAGCTTCCACCT	235	22:28324479–28324714
NF2-02	AAGGAACTGTCCAAGGGATGT	CCCACCAGTTTCATCGAGTT	398	22:28357051–28357449
NF2-03	TTGCAAAGGCTTCTTTGAGG	TGGTCAACTCTGAGGCCAAC	280	22:28359554–28359834
NF2-04	AGCCTACACACCTCACTTCCA	GAGTGATCCCATGACCCAAA	204	22:28362685–28362889
NF2-04ver2	CTCTCCACCTGTCTGCATCA	TCCCATGACCCAAATTAACG	310	22:28362573–28362883
NF2-05	AATCTCAATCGCCTGCTCTC	TCCTTCAAGTCCTTTGGTTAGC	184	22:28375163–28375347
NF2-06	ACACGCCCTCTCTCTGTGTG	GCCCATAAAGGAATGTAAACC	276	22:28375983–28376259
NF2-06ver2	TCCTCCAATAGAATGGTGCC	GCCCATAAAGGAATGTAAACCA	302	22:28375957–28376259
NF2-07	TAGAGGAATGGCAGGGTCAG	TCGCCTTGGAATGAAGAAGT	267	22:28378633–28378900
NF2-07ver2	GGATGGGAAATTCTGCTTGA	CACCCGGATTGCAAAGTAGT	282	22:28378528–28378810
NF2-08	CTTCTACCTGCCCCAATTCAG	CAGGGAAAGATCTGCTGGAC	361	22:28381567–28381928
NF2-08ver2	TGTGTGCCAGATTCTTTGGA	AACAACCACACCCTCAAAGC	355	22:28381664–28382019
NF2-09	GATACTGGGAAGCCAGGACA	GAAAGTATGCGCCAAGTGAG	339	22:28385397–28385736
NF2-10	CTACCTGCAAGAGCTCAAAC	GTAGGCATCGGCAAATGAAG	366	22:28388776–28389142
NF2-11	CATCTTTGGGCCCTTGTG	CCCAAGTAGCCTCCTGGAAC	237	22:28392294–28392531
NF2-12	GGGAATGTGGCTTGTCATTTC	CTTCCAGCACCTTCTGCTC	339	22:28393647–28393986
NF2-13	TTTCCTGCTACCTGCCCTCT	CTCTCTGCACCTCTCATCCA	297	22:28395324–28395621
NF2-14	AGTTGTGCCCATTGCCTCTG	CCCCAATCACTCAGTCTAGTTC	294	22:28398630–28398924
NF2-15	TGTCTCACTGTCTGCCCAAG	AGGAAACCAGATGCCAACC	290	22:28401931–28402221
NF2-16	CCCAGGAACAGTCTCACCAT	GCAGCACCATCACCACATAG	314	22:28403335–28403649
NF2-17A	GGGACTGACAGCCAACTTCT	CCCTATGGATGGCTCTCTTG	291	22:28415142–28415433
NF2-17B	GTCTCAAGCCCAAAGAGCAG	GGCCCTACAGGAAGAAGTGA	571	22:28418465–28419036
PLP-01	AAAGCAGGCCTGTCCCTTTA	TGTGAATTCCTGTGTCCTCTTG	260	X:101803577–101803837
PLP-02	TGGATGTGCCTGACTGTTTC	TCTGAATGGGCTTCGAATGT	399	X:101812330–101812729
PLP-03	ATTCCCTGGTCTCGTTTGTC	CACCTTGTCGGGATGTCCTAG	300	X:101813215–101813515
PLP-04	CTCCAGGATCTCCCAGTTTG	GCACCCGTACCCTAACTCAC	267	X:101814505–101814772
PLP-05	AGAGATGGAAGAAGGGCTCT	ATTGAAGGCCATGGGTGTAA	236	X:101815132–101815368
PLP-06	AAAGATATCAACACATTCAG	TTGCCTTTCAGAATAGCTGT	239	X:101816019–101816258
PLP-07	CCTCATTCCCAAAGGGATTT	CAGCATTGTAGGCTGTGTGG	212	X:101817233–101817445
PMP22-01	TGGCTTCAGTTACAGGGAGC	TAGGCACACATCACCCAGAG	224	17:15368992–15369216
PMP22-01pma	GTCTTGGCATCACAGGCTTCAGGCA	ACCAGGCTCCCCGAGATGTTCCCTG	268	17:15369166–15369434
PMP22-02	TCCTCGCAGGCAGAAACT	CAGTCCTGAACCAGCAGGAG	206	17:15364447–15364653
PMP22-02ver4	ACTCCGCTGAGCAGAACTTG	GCAGATTGCCAGAAACTTCC	249	17:15364389–15364638
PMP22-03	CTGGCAGAACTGTAGCACCT	CATGGCTCCCTGTCACATC	205	17:15362836-15363041
PMP22-04	TTCTGCTGCCTGTGAGGAC	TTCTGAGGCCACATCCTTCT	349	17:15343264–15343613
PMP22-05	CAGGTCTGTGCGTGATGAGT	CCACCTCCACTGCTTTCTGT	304	17:15334662–15334966
PMP22-05uma	CTGCTTTTGTACCTAGCTAGGCTGC	ATGCATCTTAGTCCACACAGTTGG	306	17:15333685–15333991

^aAmplicon names reflect the gene and exons amplified in each assay. Amplicon names ending in the “ma” or “ver2, -3, -4, etc.” designations represent alternative amplicons for the same exon but that differ in sequence. Amplicon names followed by characters in parentheses refer to amplicons that were spotted in two different locations on the array.

^bAmplicon sizes are shown and genomic locations are given with respect to build NCBI 34 and ENSEMBL version 21.34d.1.

To generate arrays containing single-stranded array elements, all PCR products used in the present study were prepared as follows. To the 5′ end of the forward primer of each pair was added an 8-bp universal sequence (5′-TGACCATG-3′). These primer pairs (final concentration 0.5 μM) were used to amplify exon-containing PCR products in a 20-μl final volume first-round PCR containing 50 mM KCl, 5 mM Tris HCl (pH 8.5), 2.5 mM MgCl₂, 10 mM dNTPs (Pharmacia), 0.625 U Taq polymerase (Perkin Elmer–Cetus), and 50 ng of human genomic DNA. Thermocycling was optimized for each primer pair by use of an annealing temperature of 50°C–60°C (cycling conditions available on request). PCR products from these reactions were diluted 1:1000 and were used in second-round reactions containing the 5′-(C6) amino-modified universal primer 5′-GCTGAACAGCTATGACCATG-3′ (Eurogentec) and the reverse gene-specific primer of each set. Reaction conditions were the same as those in the first round, except the reaction volume was 60 μl. For the comparison of single- versus double-stranded DNA array elements, the universal linker was also added to the 5′ end of the reverse gene-specific primer. This allowed a 5′-aminolink to be incorporated on either the forward or the reverse strand in second-round reactions.

To prepare second-round PCR products for arraying, spotting buffer was added at final concentrations of 0.25 M sodium phosphate buffer pH 8.5 and 0.00025% sarkosyl (BDH Laboratory Supplies). The PCR products were then filtered through multiscreen-GV 96-well filter plates (Millipore), were aliquoted into 384-well plates (Genetix), and were arrayed onto Codelink slides (Amersham) in quadruplicate in a 16-block format by use of a Microgrid II arrayer (Biorobotics/Genomic Solutions). Slides were processed to generate single-stranded array elements, as described on the Sanger Institute microarray Web pages, and were stored at room temperature until hybridized.

Patient and Control DNA Samples

Thirty-one patient samples were obtained from a number of laboratories; all samples had been screened for mutations in the relevant disease gene at accredited diagnostic laboratories. Samples from patients with mutations in the COL4A5, DMD, and PMP22 genes were obtained from the Southeast Thames Regional Genetics Centre (at King's, Guy's, and St. Thomas’s Hospitals, London). These samples were all made anonymous for inclusion in this study. DNA samples from patients with NF2 mutations were obtained from the Department of Genetics and Pathology at Uppsala University. DNA samples from patients with mutations in the PLP1 gene were obtained from the Institute of Child Health (London); additional patients with mutations in the PMP22 gene were obtained from the Center for Human Genetics (Leuven). Control DNA was obtained from archives of normal male and female DNA housed at the Sanger Institute (Cambridge). DNA was extracted using a variety of standard methods (salt/chloroform, phenol, commercial kits, etc.) from either peripheral blood or cultured cell lines. Aliquots of patient and control DNA samples were quantitated using a fluorometer (TD-360 [Turner]), were sonicated to an average size of ∼10 kb by use of a water-bath sonicator (VirSonic 300 [Virtis]), and were visualized by agarose-gel electrophoresis. Female and male pool DNA samples that were used as reference DNA controls were derived by combining equal amounts of peripheral blood DNA from five females and five males, respectively. Male (XY) and female (XX) lymphoblastoid cell lines were also used as controls in the validation experiments. None of the control samples had mutations in the genes analyzed in this study.

Fluorescent DNA Labeling, Microarray Hybridization, and Data Analysis

Fluorescent-labeled DNA samples were prepared using a modified Bioprime labeling kit (Invitrogen) in 100-μl reaction volumes containing 600 ng genomic DNA, dNTPs (0.2 mM dATP, 0.2 mM dTTP, 0.2 mM dGTP, and 0.05 mM dCTP), and 0.04 mM Cy5/Cy3 dCTP (GE Healthcare). Reference control samples were labeled with Cy3, and test samples were labeled with Cy5. Labeling reactions were purified using Micro-spin G50 columns (Pharmacia-Amersham) in accordance with the manufacturer's instructions. Reference and test samples were combined and precipitated with 3 M sodium acetate (pH 5.2) in 2.5 volumes of ethanol with 90 μg human C_ot DNA (Invitrogen). The DNA pellet was resuspended in hybridization buffer containing 50% deionized formamide, 10% dextran sulphate, 10 mM Tris-HCl (pH 7.4), 2× SSC, 0.1% Tween-20, and 300 μg yeast tRNA (Invitrogen). Similarly, the prehybridization mixture was prepared by precipitating 400 μg herring sperm DNA (Sigma) and 67.5 μg human C_ot DNA with 3 M sodium acetate (pH 5.2) and 2.5 volumes of ethanol; this DNA pellet was resuspended in the same hybridization buffer but with no yeast tRNA included.

Microarrays were prehybridized, hybridized, and washed using methods described elsewhere (Fiegler et al. 2003). Microarrays were scanned using a ScanArray 4000 confocal laser-based scanner (Perkin Elmer). Mean spot intensities from images were quantified using QuantArray or ScanArray Express (Perkin Elmer) with background subtraction. Mean ratios and SDs for all exon PCR products in quadruplicate were calculated, and the mean exon ratios were normalized to the chromosome 22 control element mean ratios for each of the 16 blocks independently to within 2% of the theoretical value (1.00±0.02). Ratios were obtained for each exon independently, except in the cases in which more than one exon was contained in a single PCR product. Furthermore, there was no averaging of copy-number data for exons found in different PCR products. Although some exons were represented by more than one PCR product, the data from only the array element that most accurately reported copy number (i.e., that behaved closest to the theoretical values) were retained in the final data set for each sample in all but two instances—for the PMP22 gene, the data from two PCR products representing exon 1 were pooled to derive a mean exon 1 ratio for both products. Similarly, this procedure was also performed for PMP22 exon 5. To visualize the data, the final data set of mean ratios was plotted on a histogram for each exon or exon pair of COL4A5, DMD, NF2, PLP1, and PMP22. Microarray experiments with control samples and patient samples were performed—multiple times, in some cases—to determine the reproducibility of the method. However, the derived mean-ratio data were determined for each hybridization experiment rather than from pooling of data from multiple experiments. For single- versus double-stranded DNA comparisons, mean intensities of quadruplicate spots were calculated with background subtraction in a single channel.

Results

Single-Stranded Technology Allows Detection of Exon-Sized Array Elements

One of the main considerations for array CGH is related to the complexity of the genome, as well as how this complexity is reflected in the signal measurements from which the copy-number changes are derived. Given that a typical exon in the human genome (median and mean sizes of 133 bp and 262 bp, respectively) represents a segment of genomic DNA found at a level of 1–2 in 4×10⁷ sequence equivalents in the diploid human genome, it was necessary to address how to detect such a sequence at this level without reducing genome complexity or compromising the accuracy of the measurements. To this end, we developed an array system that allows single strands of DNA derived from double-stranded PCR products to be retained on the surface of the microarray slide. By incorporating a 5′-aminolink modification onto the end of one strand of a double-stranded DNA molecule during PCR, it is possible to covalently attach this strand to the surface of the slide, and the unmodified strand can be removed (fig. 1). Since single-stranded array elements cannot reanneal to form duplex molecules, we anticipated an increase in the signal:noise ratio in array CGH experiments, since there would be more single-stranded DNA on the surface of the slide capable of hybridizing with labeled sample. We demonstrated increased signal:noise measurements for single-stranded array elements (fig. 1). Test arrays were printed with array elements on which 5′-aminolink modifications were incorporated on either one or both strands. For a series of genomic array elements of 200–400 bp in size that we tested in hybridizations with fluorescently labeled human genomic DNA, single-stranded elements provided an increased signal:noise ratio, with a range of 1.16–2.51-fold increase, depending on the array element. On average, this translated into a 1.79-fold increase in the signal:noise ratio. We had previously demonstrated a similar increase in the signal:noise ratio for cDNA expression analysis (C.L. and D.V., unpublished data). Thus, this array platform facilitated increased signal:noise ratios and thereby provided evidence that we could improve the signal sufficiently to detect human genomic DNA sequences at the resolution of single exons.

Figure 1.

Development of single-stranded array platform for exon array CGH. Schematic diagram shows the approach adopted to make single-stranded PCR products with the increased sensitivity on the array platform described in this article. A, Oligonucleotide containing a 5′-C6-aminolink modification (gray ball) used as a primer (short green bar) on a genomic DNA or PCR template strand (red wavy line) in a PCR reaction. B and C, Synthesized double-stranded PCR product containing the 5′-aminolink modification on one strand only. Newly synthesized strand containing the 5′-aminolink modification is shown as a green dotted/solid line. D, PCR product spotted onto the microarray slide. E, Covalent attachment to the surface formed through the 5′-aminolink modification by subjecting the slide to high humidity. F, Removal of reverse strand that is not covalently attached to the surface of the slide by use of physical and chemical denaturation. G, The resultant single-stranded array element. H, Microarray image (gray scale) showing the results of fluorescently labeled genomic DNA hybridized to test arrays containing single-stranded and double-stranded PCR products for a series of genomic array elements (“i–xii”) spotted in duplicate. Single-stranded elements for either one strand (5′-aminolink on forward strand [lanes 1 and 2]) or the other strand (5′-aminolink on the reverse strand [lanes 5 and 6]) or for double-stranded elements (5′-aminolinks on both strands [lanes 3 and 4]) were prepared on the test array. Single-stranded elements gave a higher signal for all genomic products (see the “Results” section).

Assessing the Performance of DNA Elements on Exon Arrays

We generated an exon array that included a set of array elements designed for every exon of five genes. The five genes included on this array were chosen to assess technical aspects of the technology and to demonstrate its clinical relevance (table 1) (Atkin et al. 1988; Worton and Thompson 1988; Den Dunnen et al. 1989; Gillard et al. 1989; Nelis et al. 1996; Heim et al. 1997; Lemmink et al. 1997; Sistermans et al. 1998; Mimault et al. 1999; Plant et al. 1999; Rautenstrauss et al. 2000; Bruder et al. 2001). These were the X-linked genes COL4A5, DMD, and PLP1 and the autosomal genes NF2 and PMP22. Mutations and/or copy-number changes in these five genes result in human inherited disorders. For the 162 exons of these genes, 158 array elements were initially designed and spotted on our arrays. We included a second array element for each of 18 exons that we chose at random to help assess sequence context on reporting accuracy, which brought the total number of array elements for these genes to 176. From the 3′ UTRs of genes on chromosome 22, 360 repeat-free amplicons were also spotted on these arrays and were used as normalization controls. These controls were, on average, of a size similar to that of the exon elements. We performed a series of validation experiments of male (XY) versus female (XX), and female (XX) versus female (XX) competitive hybridizations across a series of five batches of array printed at different times, with elements printed in quadruplicate, to assess the performance of the exon array in measuring copy-number changes. From the initial minimal set of 158 array elements, 135 (85.4%) reported copy-number measurements that were within 0.15 copy-number units of the theoretical values. The second array element for each of the 18 exons that had two different elements on the array showed a similar success rate (16 [88.9%] of 18). To obtain accurate copy-number measurements for all remaining exons in the five genes, we designed and tested a second element for each of 23 exons, 2 of which required a third iteration of design. These redesigned elements replaced the original elements for those exons in all subsequent analyses. In total, we analyzed 201 array elements, 174 (86.6%) of which reported accurate copy-number measurements and had a size range of 139–571 bp. On the basis of these experiments, we demonstrated that we could obtain accurate copy-number measurements for ∼85% of the exons in our study by designing a single array element and for 100% of all exons by designing a second or, in a small minority of cases, a third array element. Typical examples of the quality of the data that we obtained from these validation experiments for array elements covering all 162 exons of the five genes are shown in figure 2. The three X-linked genes (DMD, PLP1, and COL4A5) showed easily discernible single copy-number differences in XX versus XY validation experiments, when compared with the two autosomal genes (PMP22 and NF2) (fig. 2A). XX versus XX validation experiments also showed the expected copy-number equivalence for all genes (fig. 2B).

Figure 2.

Validation experiments for exon arrays. A, Fluorescently labeled DNA derived from genomic DNA from a single male (Cy5) hybridized on the exon array with that derived from five females (used as a pool) (Cy3). The plot shows the Cy5 channel:Cy3 channel ratios obtained for exons (represented as black dots) for COL4A5, DMD, NF2, PLP1, and PMP22. Each exon is plotted as a function of its position 5′→3′ for each gene. The number of exon array elements assayed for each gene is shown in parentheses below the gene name. Exons for the three X-linked genes and the two autosomal genes showed ratios centered around the theoretical values of 0.5 and 1, respectively. B, Fluorescently labeled DNA derived from genomic DNA from a single female (Cy5) hybridized on the exon array with that derived from the female pool (Cy3). The plot shows the ratios obtained for each exon, as described above (A), centered around the theoretical value of 1. C, The effect of signal:noise ratio on measurement error determined, using exon arrays, in 10 male versus female-pool validation experiments. The plot shows the relationship between mean signal:noise and mean SD. Both means were derived as the mean of all mean quadruplicate element measurements per experiment. The black triangles represent each of the 10 male/female validation experiments. The correlation coefficient (r) between mean SD and mean signal:noise was −0.97.

Across the validation experiments, signal:noise ratios varied from hybridization to hybridization, which we attributed to variations in cyanine dye quality, DNA quality, labeling efficiencies, array batch, and other technical considerations. However, whereas signal:noise ratio did not appear to dramatically affect the range of mean copy-number values for each exon spotted in quadruplicate, the accuracy of the measurements was noticeably improved as the signal:noise ratio increased (fig. 2C). SDs in the quadruplicate measurements made for each exon showed a strong negative correlation with signal:noise ratio (r=-0.97). This would suggest that the single-stranded array platform we have developed not only helps to improve signal:noise ratio, but, as a consequence, also reduces measurement error. More importantly, our data also demonstrate that even exons that exhibit an extremely low signal:noise ratio on our array (i.e., in the range of 1–1.5) could be quantitated accurately.

We studied further the performance of our single-stranded array elements to identify other features that may contribute to reporting accuracy in array CGH experiments. Using regression analysis on data obtained from all 201 exon array elements, we found no strong correlations between the accuracy of copy-number measurements and inherent features of the array elements, such as G+C content, length, melting temperature, secondary structure, similarity to other genomic sequences, and distance from upstream or downstream repeat sequences in genomic DNA. However, given that we could obtain accurate copy-number measurements for some exons only by redesigning array elements, sequence context is likely to contribute in some way to reporting accuracy. We also assessed whether the location of an element on the array affected reporting accuracy, by spotting independent preparations of 12 of the array elements in alternative locations on the array. All 12 of these array elements reported accurate copy-number measurements irrespective of their array coordinate. This evidence, together with the rest of our empirical data, demonstrates that, although it is unclear which factors may influence the performance of DNA elements on our exon array, this platform is robust within a variety of sequence contexts and technical constraints.

Detection of Copy-Number Changes in Patient Material

We further tested the performance of the exon array and its potential application as a research and diagnostic tool by detecting copy-number changes in DNA samples from patients with inherited disorders. Thirty-one patient DNA samples (from 24 males and 7 females) were collected that all had known pathogenic copy-number changes in one of the five disease genes represented on the array. We were able to determine copy-number changes in all 31 DNA samples and with equal efficacy in affected males, affected females, and nonsymptomatic carrier females (for X-linked disorders). All of these results were in agreement with the molecular data determined elsewhere for these patients (Ellis and Malcolm 1994; Harding et al. 1995; Woodward et al. 1998, 2000; Bruder et al. 2001; Buckley et al. 2002; Mantripragada et al. 2003; S.A., unpublished data; J.R.V., unpublished data) or as part of the present study. Our results are summarized in table 3 and figure 3. In total, we detected copy-number changes (either a gain or loss) in patient samples for 121 of the 158 array elements containing exons that were present on the array. The demonstrated copy-number changes included 10 whole-gene deletions (fig. 3A), 8 whole-gene duplications (fig. 3B), 2 whole-gene triplications (fig. 3C and 3D), 3 partial-gene duplications (fig. 3E), and 8 partial-gene deletions (fig. 3F, 3G, and 3H).

Table 3.

Copy-Number Changes Detected in 31 Patient Samples^[Note]

Sample	Patient Reference	Sex	Gene Affected	Mutation
1	JPa,b,c	Male	NF2	Whole-gene deletion
2	p7a,b	Male	NF2	Whole-gene deletion
3	p130b	Male	NF2	Deletion (exons 13–15)
4	EA01(241397)	Male	PMP22	Whole-gene deletion
5	EA02(260055)	Female	PMP22	Whole-gene deletion
6	EA03(218861)	Male	PMP22	Whole-gene deletion
7	EA04(268266)	Male	PMP22	Whole-gene deletion
8	EA05(251650)	Male	PMP22	Whole-gene deletion
9	EA06(263902)	Female	PMP22	Whole-gene duplication
10	EA07(221358)	Male	PMP22	Whole-gene duplication
11	EA08(138116)	Male	PMP22	Whole-gene duplication
12	EA09(231345)	Male	PMP22	Whole-gene duplication
13	EA10(271065)	Male	PMP22	Whole-gene duplication
14	EA11	Male	COL4A5	Whole-gene deletion
15	EA12	Male	DMD	Deletion (exon 51)
16	EA13	Male	DMD	Deletion (exon 44)
17	EA14	Male	DMD	Deletion (exons 3–7)
18	EA15	Female	PMP22	Whole-gene deletion
19	EA16	Male	PMP22	Whole-gene deletion
20	EA17	Male	DMD	Deletion (exons 45–52)
21	EA18	Female	DMD	Deletion (exons 45–48)
22	EA19	Male	DMD	Duplication (exon 2)
23	EA20	Female	DMD	Deletion (exons 17–48)
24	EA21	Male	DMD	Deletion (exons 45–50)
25	EA22	Male	DMD	Duplication (exons 2–7)
26	EA23	Male	DMD	Duplication (exon 3)
27	DH/PMD2-1d,e,f	Male	PLP1	Whole-gene triplication
28	PMD2-2d,e,f	Female	PLP1	Whole-gene triplication
29	NO/PMD9-1d,f	Male	PLP1	Whole-gene duplication
30	PMD4-2f	Female	PLP1	Whole-gene duplication
31	PMD4-1f	Male	PLP1	Whole-gene duplication

Note.— Copy-number changes in a collection of 31 patient samples found using the exon array described in the present study. Where relevant, patient references have been included from other published studies.

^aMantripragada et al. 2003.

^bBruder et al. 2001.

^cBuckley et al. 2002.

^dEllis and Malcolm 1994; Woodward et al. 1998.

^eHarding et al. 1995.

^fWoodward et al. 2000.

Figure 3.

Exon copy-number changes in medically relevant disease genes. Representative results of exon array CGH for patient DNA samples hybridized against a female pool that demonstrates the variety of copy-number changes detected in the present study. All patient samples were fluorescently labeled using Cy5, and all control female-pool samples were fluorescently labeled using Cy3. The features of the plots are as described for figure 2. The gray arrows highlight the copy-number changes. A, Female patient—EA02(260055)—with HNPP showing a deletion of the entire PMP22 gene. B, Male patient—EA07(221358)—with CMT1 showing a duplication of the entire PMP22 gene. C, Male patient (DH/PMD2–1) with PMD showing a triplication of the entire PLP1 gene. D, Female PMD carrier (PMD2–2, mother of patient in panel C) showing a triplication of the entire PLP1 gene. E, Male patient (EA19) with DMD showing a duplication of exon 2 in the DMD gene. F, Male patient (p130) with NF2 showing a deletion of exons 13–15 of the NF2 gene. G, Male patient (EA14) with DMD showing a deletion of exons 3–7 of the DMD gene. Panel also shows the confirmatory PCR results; lanes are numbered according to the exons assayed and size markers (“M”), in base pairs, shown at the left of the gel image. H, Male patient (EA12) with DMD showing the deletion of exon 51 of the DMD gene. Confirmatory PCR is also shown with lanes numbered according to the exons assayed and size markers (“M”), in base pairs, shown at the left.

In the DNA sample of one patient (patient EA11), in which the entire COL4A5 gene was deleted, we detected a copy-number measurement in COL4A5 exon 1 (0.26) that did not suggest a complete absence of this exon (theoretical value of 0). The PCR product for this exon showed an 86% sequence similarity across 212 bp to a sequence on the short arm of chromosome X. This was the highest degree of sequence similarity found for any of the elements on the exon array. Therefore, although exon 1 was completely deleted in patient EA11 (as confirmed by PCR), the cross-hybridizing X-linked sequence was being reported by the exon 1 array element.

Of the 11 partial-gene duplications/deletions, 4 involved only a single exon, 1 of which was a single-copy number gain of DMD exon 2 (fig. 3E). The array element for DMD exon 2 displayed the lowest signal:noise ratio across all array CGH experiments performed for the present study. This not only helps demonstrate the sensitivity of the array system and our ability to quantitate weak signals, but it also shows the utility of the array for accurate prediction of copy-number changes involving only DMD exon 2; a single-exon duplication of exon 2 is the most frequently occurring DMD mutation, according to one study that used the MAPH assay (White et al. 2002), but is also one of the most difficult to determine using hybridization-based approaches, because it is highly AT rich (PCR product is 73.4% AT rich) (White et al. 2002).

Discussion

We report here an array CGH platform that can measure copy-number changes accurately at the resolution of single exons. By developing an array system that results in single-stranded DNA elements being bound to the surface of a slide, we have demonstrated an improved signal:noise ratio that facilitates the accuracy in copy-number measurements for array elements in the size range of 139–571 bp. We have further shown that constitutionally inherited copy-number changes, as either deletions or duplications, were detected in 100% of patient samples analyzed on an exon array containing array elements for 162 exons from five disease genes. The results described here represent the first time, to our knowledge, that array CGH has been shown to perform robustly at the resolution of single exons. This represents an increase in array CGH resolution that is 2 orders of magnitude higher than has been reported elsewhere.

Our study provides compelling empirical evidence that it will be possible to obtain accurate copy-number measurements for virtually all exons in the genome, with ∼85% of exons requiring that a single-array element be designed and tested and the remainder requiring that additional array elements be validated in an iterative process. Indeed, in the present study, we were able to design array elements for 161 (99.4%) of 162 exons that faithfully reported copy number in validation experiments and in the analysis of patient samples. For the one remaining exon, similarity with a cross-hybridizing genomic sequence was an issue, which suggests that this and possibly other sequence-based factors may affect the performance of array elements for other exons in the genome. What these factors are remains unclear. In some instances, it may be necessary to use “surrogate” exons—that is, array elements designed for sequences that lie adjacent to exons and that may perform more robustly in array CGH. Averaging measurements from multiple array elements may also provide the means to improve quantitative measurements for other exons.

The applications of this technology are numerous. Since our arrays are both scalable and sensitive at the resolution of single exons, they fill a technological gap between the high-throughput technologies (e.g., array CGH by use of BAC arrays) and the high-resolution but less scalable assays (e.g., MAPH and MLPA). Therefore, the development of this technology opens the way for identifying copy-number changes at high resolution in both normal and disease states, as a research or diagnostic tool, and in both large-scale and small-scale studies.

The construction of a genomewide exon array would represent an important achievement for the analysis of DNA-copy number in the human genome. To maintain single-exon resolution for genomewide screening, such an array would require at least one array element for each of the ∼245,000 exons that have been annotated in the human genome. On the basis of the data presented here, that would require the design and testing of 280,000–300,000 exon-specific primer pairs and their conversion to array elements to obtain a quantitative working assay for every exon. Furthermore, on the basis of current spotting technologies, multiple microarray slides would be required to spot the entire set of exons, resulting in increased costs per genome assayed. Although this may appear to represent an enormous undertaking, it must be balanced against the tremendous value that such an array set could contribute to human genetics—it would allow the identification and annotation of CNPs and mutations, in normal individuals and in disease, that have previously been undiscovered because of limitations in the resolution of other array CGH technologies.

In more focused approaches, exon arrays would also facilitate positional cloning strategies of novel disease genes in well-characterized patient collections. Given that we have shown here a direct application of this technology to identify copy-number changes in patient material for several single-gene disorders, its use as a diagnostic tool for known disease genes is evident. Furthermore, it should be possible to screen many genes in an exon array CGH format to quickly elucidate copy-number changes for some categories of disease phenotypes that are difficult to distinguish clinically. Exon array CGH could also be applied to cancer genetics. Given that there are several hundreds of genes that have been shown to be mutated in cancer (Futreal et al. 2004), this set serves as an excellent starting point for the construction of cancer-specific exon arrays.

It will also be possible to use our array system at this level of resolution to answer other questions about genome biology, with the one obvious application being chromatin immunoprecipitation (ChIP)-on-chip (Ren et al. 2002; Weinmann et al. 2002). Given the quantifiable signal:noise ratios we have shown here, obtaining detailed maps of protein-binding sites, histone modifications, and other features, such as origins of replications and matrix attachment regions, should also be possible with genomic tiling-path arrays or intergenic arrays at high resolution. This type of approach for identifying transcription-factor binding sites onto human chromosomal regions (Horak et al. 2002) and whole chromosomal tiling paths (Martone et al. 2003; Cawley et al. 2004; Euskirchen et al. 2004) has already been described, although the proportions of sites identified that are false-positive results or false-negative results are not yet known. In particular, our array system is well suited to identify low-affinity DNA-protein interactions that give relatively poor enrichments in ChIP and small quantifiable copy-number increases on arrays. We are currently extending our work in this area by generating high-resolution genomic tiling arrays for regions of interest in the human genome.