@ © 1996 Nature Publishing Group http://www.nature.com/naturegenetics
`
`article
`
`Detection of heterozygous
`mutations in BECAf using high
`density oligonucleotide arrays and
`two—colour fluorescence analysis
`
`Joseph G. Hacial, Lawrence C. Brodyl, Mark S. Cheez, Stephen P. A. Fodor2 8:
`Francis S. Collins1
`
`The ability to scan a large gene rapidly and accurately for all possible heterozygous
`mutations in large numbers of patient samples will be critical for the future of medicine.
`We have designed high-density arrays consisting of over 96,600 oligonucleotides 20-
`nucleotides (nt) in length to screen for a wide range of heterozygous mutations in the
`3.45-kilobases (kb) exon 11 of the hereditary breast and ovarian cancer gene BRCA1.
`Reference and test samples were co-hybridized to these arrays and differences in
`hybridization patterns quantitated by two-colour analysis. Fourteen of fifteen patient
`samples with known mutations were accurately diagnosed, and no false positive
`mutations were identified in 20 control samples. Eight single nucleotide polymorphisms
`were also readily detected. DNA chip—based assays may provide a valuable new
`technology for high-throughput cost-efficient detection of genetic alterations.
`
`(ASO) hybridization} which detect only a finite set of
`While some genetic conditions such as sickle cell dis-
`ease, achondroplasia, or the triplet repeat disorders are previously described mutations.
`A variety of protocols have been used to screen for all
`mutationally monomorphic, extensive heterogeneity of
`mutations amongst affected individuals is the more
`usual observation. This allelic heterogeneity pre—
`sents a considerable challenge to the development
`of high-throughput cost-effective analytical meth—
`ods for mutation detection. All possible mutations
`must be detectable, and in diploid organisms a
`mutation in an autosomal gene must routinely be
`identifiable in the context of wild—type sequence
`from the other allele.
`
`+A,+C,+G, or +T
`5'$ 3'
`
`A case in point is the familial early onset breast
`cancer gene, BRCAI. Germline mutations in
`BRCAI are present in 50—60% of kindreds with
`breast and ovarian cancer, and may account for
`approximately 26% of all breast cancer cases in
`the general population”. Heterozygotes are
`markedly predisposed to early onset breast and
`ovarian cancer, and are also at moderately increased
`risk of developing colon and prostate cancers. The
`protein coding region of BRCA1 contains 5,592
`basepairs (bp) in 22 coding exons spread over 100
`kb of genomic DNA". Over 111 unique BRCAI
`mutations distributed throughout the gene have
`been described7x8. Most of these are frameshift,
`nonsense, or splice mutations resulting in a dis—
`ruption of the normal reading frame Except for
`the Ashkenazi Jewish population, where two muta»
`tions account for the majority of BRCA1 alter-
`ationsg‘”, allelic heterogeneity confounds the
`ability to identify BRCAI mutation carriers by
`methods (such as allele—specific oligonucleotide
`
`‘National Center
`for Human Genome
`Research, Building
`49/3/11 4, National
`Institutes ofHealth.
`Bethesda,
`Maryland 20892,
`USA
`
`ZAffymetrix, 3380
`Central Expressway
`Santa Clam,
`California 95051,
`USA
`
`Correspondence
`should be addressed
`to F. S. C.
`
`nature genetics volume 14 december 1996
`
`
`
`A1, A2, A3, A4, or A5
`
`5' _—I_3'
`
`Fig. 1 Classes of probe array oligonucleotides. Each
`position is interrogated with a total of 28 separate
`oligonucleotides, 14 (two wild type, three base sub-
`stitution, four insertion and five deletion) each for the
`sense and antisense strands. All probes are 20 nt in
`length. a, Sequencing probes contain each of the four
`nucleotide substitutions, nine bases from the 3' end
`of the oligonucleotide (one of these will represent the
`wild-type sequence). b, insertion probes contain each
`of the four possible single nucleotide insertions, nine
`bases from the 3‘ end of the oligonucleotide. c, Dele-
`tion probes have 1-5 nt deleted, nine bases from the
`3‘—end of the oligonucleotide. d, Sequencing and insertion probes are
`tiled adjacent to one another in the upper portion of the array. e, Deletion
`probes along with the corresponding wild-type probe are tiled in the
`lower portion of the array.
`
`441
`
`GeneDX 1007, pg. 1
`
`GeneDX 1007, pg. 1
`
`

`

`© 1996 Nature Publishing Group http://www.nature.com/naturegenetics
`
`article
`
`
`
`
`
`--
`
`d‘r-— '
`
`I
`
`11-";-
`
`.-
`
`'3'“‘35:...wzz—523 :‘Trw'
`
`fifi‘u If'I’ .F'vl-I-:'
`_
`1-.-
`
`f- "- no. ”-
`him:... 4:2:IL’T. .r..r3F""‘
`
`.
`- 1‘.-
`..
`'-"— ..-;---I -
`
`
`’1'.
`h
`
`
`
`I...“9.1 ,1 is! I".r -- ;'”-r unv-.‘—
`’3
`m
`"-1-. .I.
`
`
`
`
`1.517%
`
`99173:
`
`fig. 2 Chip image comparisons. a, Hybridization pattern of fluo-
`rescein reference target to an 1.28 x 1.28 cm array of 48,300
`oligonucleotides (50 micron feature size) false coloured in green. b,
`Hybridization pattern of phycoerythrin stained biotinylated RUL47
`target false coloured in red. c, Composite image of the false
`coloured green and red images with areas of identical signal given
`in yellow. d, Magnification of the region surmunding the 2457 C—>T
`mutation found in the upper left hand quadrant of (c). e, Close—up
`of the probe sets surrounding the 2457 C~>T mutation. The con—
`trast and threshold values have been set differently relative to (d) to
`increase clarity. BRCAl cDNA nucleotide positions and identéty of
`probes are labelled. The intervening columns represent different
`insertion probes at these positions as shown in Fig. 1d. The normal
`sequence is T at 2456. C at 2457, and A at 2458. (Details for the
`method of chip synthesis using light-directed oligonucleotide syrr
`thesis can be found in refs 2122). Briefly. DNA phosphoramidites
`bearing S'Aphotolabile protecting groups are coupled to a solid
`silica substrate utilizing modified DNA synthesis protocols. Spa—
`tially addressable synthesis of oligonucleotide species is obtained through photolithographic techniques where selected oligonu~
`cleotides are photodeprotected on the chip surface for each coupling cycle. Combinatorial synthesis strategies may yield up to 2”
`differeat oligonucleotide species in n synthesis cycles allowing the described array to be manufactured in 80 such cycles. In the pre-
`sent scheme, 30 identical high density array chips were simultaneously produced in a single 3 h synthesis.
`
`
`
`possible BRCAI germline mutations, virtually all of
`which begin with amplification of individual exons by
`DNA PCR or of the transcript by RT-PCR. These
`include the single-strand conformation polymorphism
`assay (SSCP), manual or automated direct DNA
`sequencing, clamped denaturing gel electrophoresis
`
`(CDGE), heteroduplex analysis, and the protein trun—
`cation assay‘HO. All of these require gel electrophoresis,
`seriously complicating the challenge of scale-up,
`automation, and reduction in cost.
`Advances in light-directed combinatorial chemical
`synthesis have made manufacturing of high-density
`
`442
`
`nature genetics volume 14 december 1996
`
`GeneDX 1007, pg. 2
`
`GeneDX 1007, pg. 2
`
`

`

`13% © 1996 Nature Publishing Group http://www.nature.com/naturegenetics
`article
`
`Fig. 3 Two—colour loss of signal assay for a h
`deletion. Fluorescein-iabelied reference and
`biotinylated targets were co~hybridized to
`the array. To correct for reproducible differ‘
`ences in the hybridization efficiencies of ref-
`erence and test targets. the ratio of
`fiuorescein to phycoerythrin signal at each
`Wald-type posétion was normalized against
`ratios derived from a separate chip co—
`hybridization experiment. Five data point
`moving averages of sense and antisense
`strand—corrected ratios are plotted against
`nucleotide posrtion. a, Sense strand ratios
`from 185—F15 (3875dal4). b, Antisense
`strand ratios from 185—F15 (3875del4) A
`peak at the position of mutation is present
`on both strands.
`
`
`
`IntensityRatio
`
`Q"
`
`
`
`IntensityRatio
`
`
`
`Nucleotide
`
`in the 3.45—kb BRCAI exon 11, which contaires approx-
`oligonucleotide probe arrays on solid surfaces possi—
`imately 60% of the BRCA1 coding region, including 10-
`bleZHZ. Oligonucleotides are generated in sin: on a sil—
`bp of flanking intronic sequence. Families of over 96,600
`icon surface by combining standard DNA synthesis
`oligonucleotides were designed to detect all possible sin-
`protocols with phosphoramidite reagents modified with
`gle base substitutions, single base insertions, and I—S-bp
`photolabile 5'~protecting groups. Spatially addressable
`deletions on both strands. Four 20=nt sequencing
`synthesis is accomplished through selective photodec
`probes, substituted with one of the four nucleotides in
`protection of chip areas, utilizing a photolithographic
`the central position. interrogate the identity of each
`mask set in a process similar in principle to that utilized
`nucleotide (Fig. la,d). Four ZO-nt insertion probes con-
`in computer microchip manufacture. These deprotect‘
`taining the possible single base insertions at the central
`ed areas are activated for chemical coupling. Selective
`position query for the presence and identity of an inser-
`deprotection of multiple areas containing a distinct
`tion (Fig. 111,51). Likewise five 20-nt long deletion probes
`oligonucleotide sequence (such as all those having an
`adenine as the next residue) allows for the simultane-
`query for the presence and identity of all possible l—S-
`bp deletions (Fig. lc,e). All probes were selected to be
`ous stepwise synthesis of numerous different oligonu—
`ZO-nt long as this length gave the optimal signal inten—
`cleotide species. In one manufacturing protocol, there
`sity and specificity under the tested experimental con—
`are four reactions performed for each coupling step cor-
`ditions
`(data
`not
`shown). Allele—specific
`responding to the separate photodeprotection and cou-
`oligonucleotides complementary to other described
`pling to those oligonucleotides in areas needing
`mutations not included in the above classes may be eas—
`incorporation of an A, C, G or T residue. The remainder
`ily incorporated.
`of each synthesis cycle, including oxidation and cap-
`
`a
`ping steps, are performed simultaneously for
`all oligonucleotides in the array. Multiple
`arrays consisting of many thousands distinct
`oligonucleotides can be reproducibly man—
`ufactured in several hours’ time. They can
`be designed to provide sequence informa—
`tion of any known gene. By analysing the
`hybridization pattern of fluorescent —labelled
`nucleic acid target to such arrays, sensitive
`
`high-throughput assays have been developed
`to screen for mutations in the cystic fibrosis
`(CFTR) gene”, the HIV—1 reverse transcrip-
`tase and protease gene524‘25, the fi—globin
`gene”, and the mitochondrial genome".
`None of these prior applications, however,
`have tested the ability of the method to
`detect all possible heterozygous mutations
`in a large gene at high sensitivity and speci-
`ficity.
`
`
`
`Design of oligonucieotide array
`We examined the ability ofa DNA-chip
`based assay to detect heterozygous mutations
`
`Fig. 4 Twocolour loss of signal assay for a nonsense mutation. a, Sense strand
`ratios from RUL47 (2457 C—aT). b, Antisense strand ratios from RUL47 (2457
`0—91). The heterozygous nonsense mutation and five homozygous polymor—
`phisms appear as distinct peaks on both strands.
`
`nature genetics volume 14 december 1996
`
`443
`
`GeneDX 1007, pg. 3
`
`Nucleotide
`
`
`
`intensityRatio
`
`
`
`IntensityRatio
`
`
`
`
`
`
`GeneDX 1007, pg. 3
`
`

`

`@ © 1996 Nature Publishing Group http://www.nature.com/naturegenetics
`article
`
`
`
`
`
`_1'abie 1 Sensitivity of mutation-detection in patient samples with germline mutations in
` BRCA1 exon 1 1
`
`Mutation
`Loss of Signalb
`Gain of Signal3
`Sample
`Mutation
`identificationC
`Coding Noncoding
`Coding Noncoding
`No
`——
`—
`-
`—
`ST750
`1128insA
`Yes
`+
`+
`na
`na
`624~F32
`1294de|40
`Yes
`+
`—
`—
`+
`3295
`1323delG
`Yes
`+
`+
`-
`—
`ST755
`2294delG
`Yes
`+
`+
`+
`+
`RUL57
`2314d615
`Yes
`+
`+
`+
`+
`RUL47
`2457C—>T
`Yes
`+
`+
`+
`+
`M00 52
`2804delAA
`Yes
`—-
`+
`+
`+
`ENGQ
`31 21 delA
`Yes
`+
`+
`+
`+
`ENG?
`3286ClelG
`Yes
`+
`+
`—
`—
`ENGS
`3452del4
`Yes
`+
`+
`—
`-
`3265
`3600del1 1
`Yes
`+
`+
`+
`+
`808—F1 61
`3867G—>T
`Yes
`+
`+
`—
`—
`185—F1 5
`3875del4
`Yes
`+
`+
`+
`+
`RUL77
`3937insG
`
`
`
`
`+ + + +ENG33986delAA Yes
`
`
`a"+’indicatt=,-s that the mutant probe has an intensity 1.2x or greater than the corresponding wild-type
`probe. ‘—’ indicates that the mutant probe has an intensity less than a factor of 1.2x to the corre-
`sponding wild-type probe; na indicates data not available. b‘+’ indicates a distinct peak at the mutant
`position. ‘—’ indicates the absence of a distinct peak at the mutant position. cmutation detection algo—
`rithm defined in text and Fig. 5,
`
`based targets as they showed superior hybridization
`fidelity and signal strength in this system (data not
`shown).
`
`Gain of signal analysis
`Mutant substitution, insertion, and deletion probes
`should detect sequence changes through a gain of
`hybridization signal in the test target, as the wild-type
`target generally should not hybridize strongly to them.
`We used this analysis to detect heterozygous base sub—
`stitutions (Fig. 2). The hybridization pattern of the ref-
`erence is shown in green (Fig. 2a) while the
`hybridization pattern of sample RUL47 containing a
`2457 C—>T nonsense mutation is shown in red (Fig. 2b).
`These images were superimposed with the areas of iden—
`tical signal given in yellow (Fig. 2c—e). Only the wild,
`type allele 2457 ‘C’ probe hybridized with the reference
`target. In contrast, the heterozygous mutant target
`
`Test Sample
`
`This chip design provides redundant information
`which contributes to sensitivity and specificity. Ideally,
`a heterozygous mutation in a patient sample should
`result in (i) a ‘gain of signal’ increase in hybridization
`to an oligonucleotide representing a perfect match to
`the mutant sequence, provided it is represented on the
`chip and (ii) a 50% ‘loss of signal’ intensity (relative to
`a normal control) for the family of wild-type oligonu~
`clcotide probes that query the position of the mutation.
`Under ideal circumstances, a true positive should appear
`on both stands. Due to the complexity of the hybridiza—
`tion reaction, specific mutations may only fulfill a sub-
`set of these ideal criteria. One of our goals was to define
`an algorithm which maximizes sensitivity and speci—
`ficity in the analysis of this intentionally redundant data.
`
`Advantages of a two-colour analysis system
`Two—colour assays have been used to great advantage in
`comparative genomic hybridization (CGH) as a means
`of measuring two-fold or greater differences in copy
`number between samples such as occur with chromo—
`somal deletions, duplications, and amplifications”. This
`approach is based upon measuring the relative ratios of
`fluorescence from reference and test targets, labelled
`with different fluorophores and bound to normal chroe
`mosomal spreads after a competitive hybridization step.
`Based on the success of this approach, we investigated
`the use of two-colour assay systems to provide an ana-
`lytical means of detecting heterozygous sequence dif-
`ferences between samples at the nucleotide level. This
`allows a locus on the differences between a wild—type
`reference standard and the unknown patient sample,
`rather than requiring de navo determination of the
`sequence with each patient sample. In this approach,
`wild-type fluorescein-labelled (‘green’) reference and
`biotinylated (stained with a phycoerythrin—streptavidin
`‘rcd’ conjugate after hybridization) in vitro transcribed
`RNA test targets were competitively co~hybridized to
`the array with the relative binding to all probes mea-
`sured”. The ratio of reference and test targets occu~
`pancy to each ofthe 96,600 oligonucleotides in the array
`was used to detect sequence differences between the two
`samples. RNA-based targets were used in favor of DNA
`
`
`
`Both Strands One Strand
`
`Not Present
`
`Mutation
`Detected
`
`Undetermined Mutation
`2/1 5
`Not Detected
`
`12/15
`
`1
`
`1/15
`
` Loss of Signal Loss of Signal Loss of Signal
`
`
`of Signal
`G
`Either Strand
`
`Mutation
`Detected
`2/15
`_ _
`‘
`Fig. 5 Data analysis algorithm flowchart. The number of analysed
`patient sampr containing exon 11 mutations which fulfill the indi—
`cated criteria of Ioss-of—signal or gain—of—signal assays is given.
`Fourteen of fifteen patient samples fulfill the requirements of muta-
`tion detection given in the text,
`
`444
`
`nature genetics volume 14 december 1996
`
`GeneDX 1007, pg. 4
`
`GeneDX 1007, pg. 4
`
`

`

`@ © 1996 Nature Publishing Group http://www.nature.com/naturegenetics
`article
`
`hybridized to both the wild—type allele 2457 ‘C’ probe
`as well as the mutant allele 2457 ‘T’ probe, resulting in
`a red signal at the position of the mutation.
`
`Loss of signal analysis
`Two—colour loss of signal assays are a second analytical
`means of detecting sequence differences between sam-
`ples. Localized changes in hybridization signal ratios
`may reflect different stoichiometries of wild-type alleles
`in the hybridization solution, axed thus potential het—
`erozygous sequence differences between the two samv
`pies. A corrected ratio (see Fig. 3 legend) of reference
`sample and test sample binding to wild-type sequencing
`probes for each strand can be plotted against nucleotide
`position. Regions of identical sequence should be close
`to a value of 1.0, while regions with sequence differ-
`ences should show a peak centered near the point of
`mutation. For a point mutation, this Width should be
`about 20 bp wide; for a deletion of 11 bp it is expected
`to be approximately (n+20)-bp in width. This ‘width
`propertfv’ of true peaks helps to distinguish them from
`noise (Fig. 3a,b). Ideal heterozygous mutations should
`produce peaks with a value of about 2.0, reflecting the
`two wild-type alleles in the reference compared to the
`single wild-type allele in the mutant heterozygote sam—
`ple. Any cross—hybridization of the mutant allele to the
`wild-type probe will reduce this ratio closer to 1.0. In
`practice, a cut-off of 1.2 was found to represent a good
`threshold.
`
`Relative to the heterozygous state, homozygous
`sequence differences will produce larger peaks due to
`the absence of signal from a wild—type allele. In these
`cases the theoretical peak height is infinite although in
`practice cross-hybridization usually reduces these sig-
`nals to be within a value of 10. For example, six sepa~
`rate peaks were observed for each strand of sample
`RUL47 (Fig. 4a,b). One of these (the smallest) detects
`the heterozygous 2457 C—aT nonsense mutation. The
`other five strong peaks correspond to five polymor-
`phisms 2201 T/T, 2430 C/C, 2731 T/T, 3232 G/G, and
`3667 G/G found in sample RUL47 in the homozygous
`state. These variants have been described and are in
`
`strong disequilibrium with each otherléiz‘). Polymor—
`phic signals will not be observed when the test sample
`has the same genotype as the reference.
`
`Application to multiple BRCA1 mutations
`A summary of the results of two-colour gain and loss
`of signal analysis experiments for 15 known exon 11
`BRCAI mutations is given in Table 1, representing all
`of the genomic DNA samples available to us which con-
`tain known alterations in this exon. In addition, twen-
`ty control samples from individuals without a known
`family history of breast cancer were evaluated to ascer-
`tain the specificity of the assay. Seven reported poly—
`morphisms were detected along with a previously
`unreported heterozygous base substitution 1606 G-aA
`(Arg496His) found in sample RUL57, which was con-
`firmed by dideoxysequencing. Seven of the pathologic
`mutations (2314de15, 2457C—>T, 2804delAA, 3286delG,
`3867G—>T 3937insG, and 3986delAA) were detected in
`completely optimal fashion with clear gain of signal and
`loss of signal results on each strand. For the other muta
`tions, the sensitivity of any particular assay on one or
`both strands was imperfect. For example, while the
`
`3875del4 mutation was readily detected with the loss of
`signal assay (Fig. 3a,b), the gain of signal assay failed to
`give a distinct signal, presumably because of strong
`wild-type target cross»hybridization at this location.
`Gain of signal assays based upon insertion and deletion
`probes were also capable of generating a significant
`number of false positive signals using this criteria (data
`not shown), whereas the loss of signal assay was much
`more robust.
`
`Discussion
`
`Our observations suggest the following procedure for
`interpreting chip hybridization data (Fig. 5). First exam-
`ine the loss of signal data. If a peak of width >20 bp is
`found in the same position on both strands, a sequence
`alteration is almost certainly present. We encountered
`no false positives of this sort in examining 3.45 kb of
`BRCA1 exon 11 sequence in 15 patients and 20 control
`samples (a total onlZO kb of BRCA1 sequence).
`Twelve of the 15 patient mutations in Table 1 were
`immediately detectable by this strategy, and the pre-
`cise mutation could then be identified in seven of them
`
`by examining the gain of signal data. In ambiguous
`cases where there is a loss of signal on one strand but
`not the other, the gain of signal data can still lead to
`accurate mutation detection. Two samples show loss
`of signal (of 220 bp) on one strand but not the other.
`In one of these (ENG9, 3121delA) there is a specific
`gain of signal for the appropriate oligonucleotide on
`both strands, whereas the other (3295, 1323delG)
`reveals this gain ofsignal above threshold on one
`strand only. In 20 control samples, a specific loss of
`signal on one strand was never accompanied by a con—
`firmatory gain of signal on either strand. As this cri-
`terion does not appear to result in false positives, we
`scored both ENG9 and 3295 as having been correctly
`identified as mutation-bearing. Only one of the fifteen
`samples (ST750, 1128insA), which showed neither a
`specific gain or loss of signal on either strand, would
`be scored as a false negative in the current assay. That
`mutation results from the expansion of a poly (dA) -
`(dT) tract from 7—8 nts in length, and would be pre—
`dicted to be particularly difficult to detect. Interest-
`ingly, two other samples that putatively contained the
`mutations 2086insG and 2035 T—aA did not generate
`specific loss or gain of signal on either strand. Dideoxy—
`sequencing analysis confirmed that these samples were
`of wild~type sequence in this region. Thus with the
`current algorithm the sensitivity of the method is 93%
`and the specificity is 100%.
`There are a number of possible mechanisms for false
`negative and false positive mutation detection results.
`Because the RNA targets are fragmented to an average
`length of 50~100 nt in order to minimize secondary
`structure formation prior to hybridization, the differ—
`ing lengths of RNA species between reference and test
`targets may affect the result. Sequence changes that
`enhance or decrease hybridization due to intramolec~
`ular or intermolecular target structure may also con-
`found analysis. Certain 20—mer probes may adopt
`secondary structures that could inhibit
`target
`hybridization. Short repetitive sequences (as is the
`1128insA case) and duplications will pose a serious
`challenge to any hybridization based assay. In these
`cases, there is an increased potential for cross-
`
`nature genetics volume 14 december 1996
`
`445
`
`GeneDX 1007, pg. 5
`
`GeneDX 1007, pg. 5
`
`

`

`article
`
`@5 © 1996 Nature Publishing Group http://www.nature.com/naturegenetics
`
`hybridization due in part to probe slippage that
`induces the formation of stable partial duplexes. Iden—
`tification of mutations within 20 bp of polymorphisms
`may prove challenging as the loss of signal from poly-
`morphic sequence changes may partially obscure the
`loss of signal due to mutations. Modifying the array to
`contain additional wild-type probes based on common
`polymorphic sequences would optimize mutation
`detection by eliminating interference from these poly—
`morphisms.
`A two—tiered strategy for mutation detection by DNA
`chips is suggested by our results. The first level would
`consist of an array containing only wild—type probes
`for both strands. Hybridization of reference and patient
`target to such an array and analysis for loss of signal
`peaks on one or both strands should reveal candidate
`regions where sequence changes may be present.
`Including redundant alternative probe lengths (such
`as 17, 20 and 25 nts) on the chip should further assist
`in both sensitivity and specificity not only by repro—
`ducing the peak signal but also by causing predictable
`changes in peak signal widths. Data from longer probes
`would be expected to generate broader peak signals
`while data from shorter probes are expected to gener-
`ate sharper peak signals. Gain-of-signal probes for
`common polymorphisms would also be included on
`the first tier chip. If a loss of signal peak is centered at
`a known polymorphic site, it can be checked for gain of
`signal at those polymorphism probes. For those sam-
`ples where loss—of-signal peaks remain unexplained,
`both strands of the sample could be analysed on a sec-
`ond tier chip containing a more complex array of base
`substitution, deletion, and insertion gain of signal
`probes. Information from gain—of—signal assays can be
`used to confirm the loss—of-signal assay and in many
`cases will precisely identify the exact nucleotides mutat—
`ed. This two»level approach provides a way of econo—
`mizing the probe content on a single chip. One could
`imagine screening for mutations in several candidate
`genes at once with the less complex screening chip, and
`only a fraction of samples would need to be analysed
`by the more complex gain of signal mutation detection
`chip. All identified nonpolymorphic sequence changes
`should be verified by directed dideoxysequencing
`analysis of the indicated region. As with any mutation
`analysis method, determination of the phenotypic con-
`sequences of a missense alteration not found in nor—
`mal chromosomes can prove vexing. Development of
`a functional assay for BRCAI to distinguish benign
`variants from pathological missense alterations will be
`an important step”.
`We have demonstrated the feasibility of using
`oligonucleotide arrays in screening for a variety of
`nucleotide changes in the heterozygous state for
`BRCAI. These assays may provide a sensitive screen for
`detecting a multitude of different mutations. Oligonu—
`cleotide arrays for other applications such as analysing
`specific CFTR gene mutations or the HIV reverse tran—
`scriptase and protease genes have been customized by
`modifying probe positioning and length to enhance the
`fidelity of the assay23'25. Such modifications can be
`readily applied to this system also. The use of DNA
`chips for mutation detection should be extensible to
`virtually any gene by synthesis of the appropriate chip.
`Advances in chip manufacturing and signal detection
`
`have made analysis of hybridization patterns to arrays
`consisting ofgreater than 250,000 oligonucleotide
`probes possible (data not shown). This will facilitate
`screening of the entire BRCAI coding region includ-
`ing splice site junctions using multiplexed amplifica—
`tion of target exons and the array design strategies
`proposed above. As the photolithographic mask set
`used in chip manufacture is the major expense (cur—
`rently costing a few thousand dollars) and is reusable,
`there are major economics of scale associated with high
`volume of testing. This technology therefore especial—
`ly lends itself towards making large scale genetic stud-
`ies economically feasible. Given the attractive features
`of high—throughput, automation, and modest cost, the
`DNA chip approach has the potential to become a valu—
`able method in future applications of mutation detec—
`tion to medicine.
`
`Methods
`PCR from genomic DNA and RNA target preparation. PCR
`reactions were performed on genomic samples using the
`EXPANDTM Long Range PCR Kit (Boehringer) with intronic
`primers 11FT3 S‘~ATTAACCCTCACTAAAGGGAATTAAAT~
`GAAAGAGTATGAGC-3' and 11RT7 S'MTAATACGACTCAC»
`TATAGGGAGTGCTCCCAAAAGCATAAA~3' containing T3
`and T7 RNA polymerase promoter sequences respectively. In
`vitro transcription reactions from these exon 11 amplicon tem-
`plates were performed in 10 pl reaction volumes using T3 RNA
`polymerase transcription buffer (Promega), 0.7 mM of ATP,
`CTP, GTP, and UTP, 10 mM DTT, 0.7 mM fluorescein—lZ-UTP
`or 0.15 mM biotin-itS-UTP (Boehringer Mannheim) for refer—
`ence and test samples respectively, and lOU T3 or T7 RNA poly-
`merase as indicated.
`
`Target Preparation and Analysis. Reference template was gen—
`erated from PCR amplification product of exon 11 from a
`BRCAI cDNA clone modified to contain I7l—nt and 70-nt of
`exon 11 upstream and downstream intronic splice site
`sequences respectively. Reference and test sample transcription
`products were diluted to a final concentration of 100 nM in a 25
`til solution of 30 mM MgC12. The reaction was incubated at 94
`“C for 70 min to fragment targetsl‘H5. Co-fragmented targets
`were diluted 1/100 into a 300 pl volume of hybridization buffer
`(3 M TMACC] (tetramethylammonium chloride), leE pH
`7.4, 0.001% Triton X-100, 1 nM S‘-—fluorescein—labelled control
`oligonucleotide 5'—CGGTAGCATCTTGAC—3'). This control
`oligonucleotide is designed to hybridize to specific surface
`probes to aid in image alignment. Target was hybridized with
`the chip in a 250 [.11 volume for 4 h at 35 °C. The chip surface was
`washed with 10 m1 of wash buffer (6X SSPB, 0.001% Triton X»
`100) and stained with phycoerythrin-streptavidjn conjugate
`(Molecular Probes) (2 tig/ml in wash buffer) for 5 min at room
`temperature. The chip was washed with 10 ml of wash buffer
`and scanned for 40 min utilizing a scanning confocal micro-
`scope equipped with a 488 nm argon laser (GeneChip Scanner
`(Affymetrix))“‘25. Fluorescent hybridization signals were
`detected by a photomultiplier tube using 515645 nm bandpass
`and 560 nm longpass emission filters for fluorescein reference
`(green) and biotin test (red) samples respectively23'27.
`
`Data analysis. Photomultiplier output signal was converted into
`proportional spatially addressed pixel values using GeneChip
`Software (Aifymetrix) to create a digitized fluorescence image.
`The relative contributions of the reference and test targets to
`each probe signal were extracted from each set of experimental
`green reference and red test
`images using AVl Software
`(Affymetrix) and imported into a Microsoft Excel 7.0a work‘
`sheet. The ratio of reference and test signals for each wild—type
`probe was quantitated and compared to the same ratios mea-
`sured from an identical experiment performed on a different
`
`446
`
`nature genetics volume 14 december 1996
`
`GeneDX 1007, pg. 6
`
`GeneDX 1007, pg. 6
`
`

`

`-@2 © 1996 Nature Publishing Group http://www.nature.com/naturegenetics
`
`article
`
`test sample. The ratios of the referenceltest ratios derived from
`the two separate expesiments were plotted against nucleotide
`position to generate loss of signal data (as in Figs 3,4). it is nec-
`essary to compare reference/test wild—type probe signal ratios
`from separate experiments to generate a stable baseline as the
`baseline generated by plotting reference/test wild—type probe
`signal ratios generated from a single chip against nucleotide
`position was not sufficiently stable to permit data analysis.
`
`Dideoxysequencing analysis. Four pairs of PCR primers
`(PI Ml 3+ 5'~GTTTCCCAGTCACACGGAATTAAATGAAAG
`AGTATGAGC—3’ and P1Mi3- 5'—AGGAAACAGCTATGAC—
`CATGTGAGGGGACGCTCTTG—S', P2M13+ 5'~G'I'I‘TCCCA
`GTCACACGTTGGGAAAACCTATCGGAA» 3‘ and P2Ml3-
`5'—AGGAAACAGCTATGACCATCTTTGGGGTCTTCA
`GCA—S‘, P3M13+ 5'—GTTTCCCAGTCACACGTGTTCAAA
`TACCAGTGAACTTA-3' and P3Ml3— S'—AGGAAACAGC-
`TATGACCATGGAGCCCACTTCATTAGTAC—3‘ and MM 13+
`5"GTTTCCCAGTCACACGCCAAGTACAGTGAGCACAAT-
`TA—3' and P4M13— 5'—AGGAAACAGCTATGACCATGTGC
`TCCCAAAAGCATAAA-3') were designed to generate four
`partially overlapping amplicons which cover
`the entire
`
`."
`
`shoes-in
`
`familial breast cancer to
`1. Hall, J. et 3/. Linkage of early—onset
`chromosome 17q21. Science ”0. 1684-1689 (1990).
`Narod, S. et at. Familial breast-ovarian cancer locus on chromosome
`17q12~q23. Lancet 338. 82—83 (1891).
`Easton. D.F. et al. Genetic analysis In familial breast and ovarian cancer.
`Results from 214 families. Am. J. Hum. Genet. 52, 673-701 (1993).
`Rowell, 5., Newman, 8., Boyd, J. & King, M.C. Inherited predisposition
`to breast and ovarian cancer. Am. J. Hum. Genet. 55. 861—865 (1994).
`Ford, D. and the Breast Cancer Linkage Consortium. Risks of cancer in
`BRCAI-mutation carriers. Lancet 343. 692—695 (1994).
`Miki, Y. et al. A strong candidate for the breast and ovarian cancer
`susceptibility gene BRCA 1. Science 266. 66—71 (1994).
`Shattuck-Eidens, D. et al. A collaborative survey of 80 mutations in the
`BRCA1 breast and ovarian cancer susceptébility gene. implications for
`presymptomatic testing and screening. J. Am. Med. Assoc. 273,
`535-541 (1995).
`8. A comprehensive listing of BRCA1 mutations and mutation screening
`protocols can be found in the Breast Cancer Information Core
`Database
`located
`on
`the
`World Wide Web
`at
`http://ww