AFFYMETRIX UK Ltd

Voyager, Mercury Park,Wycombe Lane,Wooburn Green,HighWycombe HP10 0HHUnited KingdomUK and OthersTel: +44 (0) 1628 552550FranceTel: 0800919505GermanyTel: 01803001334Fax: +44 (0) 1628 552585saleseurope@affymetrix.comsupporteurope@affymetrix.com

AFFYMETRIX JAPAN K.K.

Mita NN Bldg., 16 F4-1-23 Shiba, Minato-ku,Tokyo 108-0014 JapanTel: +81-(0)3-5730-8200Fax: +81-(0)3-5730-8201salesjapan@affymetrix.comsupportjapan@affymetrix.com

Part Number 701009 Rev. 4

©2001-2007 Affymetrix, Inc. All rights reserved.Affymetrix®, ®, GeneChip®, HuSNP®, GenFlex®, CustomExpress®, CustomSeq®, NetAffx®,ToolsToTakeYouAs Far AsYourVision®,TheWayAhead™,Powered by Affymetrix™, GeneChip-compatible™, Flying Objective™, and Command Console™ are trademarks of Affymetrix, Inc. Array products may be covered by one or more of the following patentsand/or sold under license from Oxford GeneTechnology: U.S. Patent Nos. 5,445,934; 5,700,637; 5,744,305; 5,945,334; 6,054,270; 6,140,044; 6,261,776; 6,291,183; 6,346,413; 6,399,365; 6,420,169; 6,551,817;6,610,482; 6,733,977; and EP 619 321; 373 203 and other U.S. or foreign patents.

AFFYMETRIX, INC.

3420 Central ExpresswaySanta Clara, CA 95051 USATel: 1-888-DNA-CHIP (1-888-362-2447)Fax: 1-408-731-5441sales@affymetrix.comsupport@affymetrix.com

www.affymetrix.com Please visit our web site for international distributor contact information.

For research use only. Not for use in diagnostic procedures.

AFFYMETR IX® PRODUCT FAMILY > RNA ARRAYS AND REAGENTS >

GeneChip® Arrays Provide OptimalSensitivity and Specificity forMicroarray Expression Analysis

The rapid development in microarraytechnology has launched a revolutionin biology, enabling researchers tosimultaneously monitor nearly eachand every mRNA transcript in anentire cellular sample. Unprecedented,highly parallel discoveries have beenmade possible, including the ability:to follow changes in the expressionlevels of many genes simultaneously1,or to identify distinctive expressionpatterns characteristic of physiologicaland pathological states2, and to screenfor very subtle changes in response toa particular therapeutic treatment3, toname a few.

����

25-mer vs. 60-mer

There are varieties of microarraytechnologies available that use differentattachment chemistries, different strategiesin choosing immobilized molecules,different assay conditions and differentdata analysis tools.The most effective microarray platformshould offer the following qualities:

� Sensitivity to identify the rarelyexpressed transcripts in a complexbackground.

� Specificity to discern betweendifferent family members.

� Absolute quantitation to enablerobust databases to be built with aminimum number of samples.

The hybridization efficiency of twonucleic acid strands results from thecombination of many factors; theircumulative effect determines thesensitivity and specificity of a microarrayexperiment. These effects have beeninvestigated in extensive detail atAffymetrix4 and include:

� Sequence-dependent factors such aslength, extent of complementarityand the overall base composition.

� Sequence-independent factors suchas concentration of the probe andtarget, time, temperature, cationconcentration, valency and character,pH, dielectric and chaotropic media,surface characteristics of the solidsupport, and density spacing of theprobe molecules synthesized on thesurface.

� Sample-dependent complexbackground signal. Any probesequence, even under conditionsoptimized for the above factors, has

the potential to interact with othercomplementary sequences present inthe sample, even if thecomplementarity is short and partial.This may result in increasednonspecific background signal andreduced specificity and sensitivity.Furthermore, samples in which thecomposition and concentrations ofthe transcripts differ may result indifferent background signals that arenot easy to extrapolate.

The Affymetrix Approach

Affymetrix has designed the followingstrategies to specifically address the issues,and to ensure that GeneChip® arraysfeature the optimal solution for mRNAexpression analysis.

Given the complexities associated withany microarray system, Affymetrix usesstrategies that specifically address criticalissues to discern between specific and non-specific binding with the PM-MM probestrategy. Based on extensive experiment-ation, we have found that 25-mers are idealfor utilization of the PM-MM strategy.

� PM-MM probe pairs offersignificantly higher sensitivity atlow target concentrations thanusing PM probes alone.

Technical Note

platform upon which to build a geneexpression database.”

In fact, Aach et al8 has reported suchan effort in systematically integ-ratingand managing expression analysisresults from several different studiesinto one database. They have foundthat, “Generation of ERAs (mRNArelative abundance estimates) isstraightforward for data derived fromAffymetrix oligonucleotide arrays andSAGE, but (two-color) microarray-derived data present a significantissue.”

Interestingly, Hughes et al6

indicated that the ‘one probe persequence’ approach is only suitablefor obtaining comparison results intwo-color assays, but not for absoluteexpression quantitation. “Thissuggests that when the experimentaldesign is focused on determiningchanges in transcript abundancebetween two samples, rather thanabsolute transcript abundance,multiple oligonucleotides are notneeded in most cases.”

Conclusion

In conclusion, Affymetrix has employedthe unique PM-MM probe pair probedesign approach to obtain the optimalbalance of highest sensitivity andspecificity in the presence of a complexsample background. The global assayoptimization strategy involving investig-ating the hybridization results of all probeson an array as a single data set, allows thederivation of assay conditions thatmaximize the number of genes detected onan array. In addition, redundant samplingof each sequence with multiple probe pairsin a probe set provides robustness andreliability in the data obtained. Incombination, the GeneChip® platformprovides robust, reliable and reproduciblequantitative expression analysis resultswith statistical significance for effectivelybuilding large expression databases.

REFERENCES

1Cho, R.J. et al., (1998) A Genome-WideTranscriptional Analysis of the Mitotic Cell Cycle.Molecular Cell, 2:65-73

2Golub, T.R. et al., (1999) Molecular Classification ofCancer: Class Discovery and Class Prediction byGene Expression Monitoring. Science, 286:531-537

3Jin et al., (2001) Effects of Early Angiotensin-Converting Enzyme Inhibition on Cardiac GeneExpression After Acute Myocardial Infarction.Circulation, 103:736-742

4Forman, J.E. et al., (1997) Thermodynamics ofDuplex Formation and Mismatch Discrimination onPhotolithographically Synthesized OligonucleotideArrays in Molecular Modeling of Nucleic Acids.(N.B. Leontis and J. Santa Lucia Jr., eds.) AmericanChemical Society Symposium Series 682:208-228

5Lockhart, D.J. et al., (1996) Expression Monitoringby Hybridization to High-Density OligonucleotideArrays. Nature Biotechnology, 14:1675-1680

6Hughes, T.R. et al., (2001) Expression ProfilingUsing Microarrays Fabricated by an Ink-JetOligonucleotide Synthesizer. Nature Biotechnology,19: 342-347

7Ishii, M. et al., (2000) Direct Comparison ofGeneChip and SAGE on the Quantitative Accuracy inTranscript Profiling Analysis. Genomics, 68:136-143

8Aach, J. et al., (2000) Systematic Management andAnalysis of Yeast Gene Expression Data. GenomeResearch, 10:431-445 Using the combination of a pair of

25-mer probes that are designed to beperfect match (PM) and mismatch (MM)to the targets offers the balance ofhighest sensitivity and specificity in thepresence of a complex background—especially for low abundance transcripts.

MM probes have the ability toquantify background signals that arepresent in the complex sample.Background discrimination becomeseven more important when thespecific signal intensity is relativelylow and close to the backgroundsignal intensity.

In Figure 1, mixtures of clonedtranscripts were spiked at variousconcentrations into tissue sampleswhere the transcripts were originallyabsent. Labeled samples were thenhybridized on the GeneChip® HumanGenome U95 Array. The value of MMprobes becomes apparent below 8 pMwhere the PM-only probe sets losesensitivity and do not respond tochanges in the transcriptconcentration. The improved assaysensitivity with PM-MM probe pairsallows detection of transcriptsrepresenting 1:100,000 – 1:300,000of total transcripts in a sample.

� PM-MM probe pairs offerspecificity in the presence ofcomplex background signals,

allowing the optimal balancebetween sensitivity and specificity.MM probes are effective internalcontrols, since they will hybridize tononspecific sequences about aseffectively as their counterpart PMprobes. As a result, unpredictablebackground signal variationsassociated with samples from differentsources as well as from cross-hybridization can now be efficientlyquantified and subtracted with theutilization of the MM probes.

Because of the targeted design ofshort probes and multiple probes persequence (Figure 1), it is possible tounambiguously detect closely relatedsequences. For example, the yeasthistone genes HTA1 and HTA2 thatshare 93% identity at the DNA levelcan be detected with GeneChip®

arrays (L. Wodicka and D. Lockhart,unpublished data).

Sensitivity/discrimination experimentsshowed that probes of approximately 25nucleotides long provide a very effectivebalance between signal intensity and

related sequence discrimination. Thisenables expression monitoring of thousandsof targets in complex samples5. Such apowerful strategy, however, is mosteffective with shorter (e.g., 25-mer) probes,since a single base mismatch is sufficient todestabilize the hybridization. Conversely, alonger probe such as a 60-meroligonucleotide, cannot efficiently takeadvantage of a similar strategy since itrequires much longer stretches ofmutations or deletions to achieve a similardiscrimination effect6. In addition, 60-merprobes will often not be able todiscriminate between closely relatedsequences, especially with relativelyabundant transcripts.

While it is often feasible to achieve veryhigh discrimination for a particular targetsequence by optimizing hybridizationconditions, there can be no assurance thatthe same conditions are sufficientlystringent for the other thousands of probesrepresented on the same array. Instead ofexamining the hybridization behavior ofonly a small subset of sequences asdescribed by Hughes et al6, Affymetrix hastaken an approach to optimize hybrid-ization and detection conditions bysurveying the hybridization results overessentially all probe pairs on an array.

We have used population distributionsto monitor the global hybridizationbehavior of an array. It is assumed that atrue stray signal has equal likelihood tohybridize more efficiently to either the PMor the MM probe in a pair. In contrast,under conditions optimized for globalhybridization efficiency, more PM probeswill have a higher intensity value than theircorresponding MM probes when a complextarget sample is hybridized to the array.

As shown in the histogram in Figure 2,the x-axis represents bins of thediscrimination score metric used, log(PM/MM). The negative values at the far

G E N E E X P R E S S I O N M O N I T O R I N G���� 2

The Value of Mismatch Probes

0

2

4

6

8

10

12

14

0.125 0.25 0.5 1 2 4 8 16 32 64 128 256 512 1024

Concentration pM

Sig

nal

PM-MM PM

Figure 1. Comparison of the assay sensitivity using PM probes only orPM-MM probe pairs.

���� 3

left of the graph correspond to probe pairswhere virtually all signal is in the MMprobe. Conversely, the probe pairs forwhich all the signal is in the PM probe arereflected by positive values at the right endof the graph. The data in green correspondto complete stray signal distribution and asexpected, the mode is centered onPM=MM (i.e. log (PM/MM) = 0). Thepopulation shown in red estimates thenumber of probe pairs bound to true targetsequences in the mixture and is thereforeskewed to the right. Assay conditions havebeen optimized in this experiment tomaximize the size and displacementof this population (red) relative tothe stray signal population (green).

Improving global discrimination scoresthrough the optimization of assayconditions as described above hastranslated directly into assay performanceimprovements in the form of increasedsensitivity for “present” calls andincrease/decrease change calls, as well asreduction in false positives for bothmetrics. It ensures that the optimized assayconditions will accommodate themaximum number of probes representedon the array. This strategy cannot be

accomplished by observing only a limitednumber of selected sequences.

Cross-hybridization derived signal is anintrinsic and possibly significant com-ponent of the total signal for many probesinterrogating a complex sample, resultingin inaccuracy in RNA quantitation andincreased rate of false positives andmiscalls. This was observed in the initialwork using high density arrays to monitorexpression5, and is abundantly evident inthe system presented by Hughes et al6.

If only one probe is used to interrogateeach sequence, it is absolutely essential thateach selected probe meets all requirementsto be a highly functioning probe. Forexample, the selected probe: a) does notexhibit cross-hybridization to otherbackground sequences, b) does not fold onitself, and c) needs to hybridize efficientlyto the target under the assay conditions. Tocompensate for many of these sources ofnoise, error and variance, Affymetrix uses

the multiple probe pair sampling strategyin the GeneChip® array design.

� Multiple probes offer the potentialto calculate statistics that providemore confidence in microarrayresults.Information from multiple probes persequence aggregates in statisticallyrobust ways, especially when MMprobe results are incorporated into theanalysis. Because of this, it is possibleto use statistically based algorithms toprovide confidence and probabilityinformation for absolute call results aswell as comparison ratios. Interrogat-ing sequences that correspond tomultiple 25-mers on GeneChip®

arrays vs. a single 60-mer6 results inmore reliable data that are robust andresistant to target sequence mutations.

� GeneChip® arrays offer absoluteexpression analysis in addition tocomparative analysis. Therefore,the results obtained are conduciveto building large databases.Redundancy in the number of probesused for each sequence in combinationwith statistical analysis softwareallows GeneChip® arrays to be usednot only for comparative analysis, butalso for absolute quantitativeexpression analysis. The use ofmultiple independent probe pairs todetect the same RNA moleculegreatly improves the RNAquantitation accuracy by averagingand outlierrejection. This eliminatesthe necessity to include controlsamples in each experiment as used inthe two-color assay.

Ishii et al7 compared the GeneChip®

technology with SAGE and conclud-ed, “These two methods correlatedquite well in both absolute expressionanalysis and comparative analysisduring differen-tiation.” In addition,they commented, “This findingsuggests that GeneChip® technologyis reasonably reliable for quantitativeanalysis of expression profiling andwould be appropriate as a common

Global performance evaluation onGeneChip® arrays vs. examination of asmall subset of genes maximizes thenumber of genes detected on the arrays.

Using multiple probes per sequenceallows both absolute and comparativeanalysis that is essential for building aquality database with the minimumnumber of samples.

Enhanced Discrimination in Optimized Assay System

0

2000

4000

6000

8000

10000

-5 -4 -3 -2 -1 -0 1 2 3 4 5

ln(PM/MM)

Nu

mb

ero

fP

rob

eP

airs

/B

in

Stray Signal Population Target Signal Population

Figure 2. Global evaluation of hybridization results on a GeneChip® array.

MM probes have the ability toquantify background signals that arepresent in the complex sample.Background discrimination becomeseven more important when thespecific signal intensity is relativelylow and close to the backgroundsignal intensity.

In Figure 1, mixtures of clonedtranscripts were spiked at variousconcentrations into tissue sampleswhere the transcripts were originallyabsent. Labeled samples were thenhybridized on the GeneChip® HumanGenome U95 Array. The value of MMprobes becomes apparent below 8 pMwhere the PM-only probe sets losesensitivity and do not respond tochanges in the transcriptconcentration. The improved assaysensitivity with PM-MM probe pairsallows detection of transcriptsrepresenting 1:100,000 – 1:300,000of total transcripts in a sample.

� PM-MM probe pairs offerspecificity in the presence ofcomplex background signals,

allowing the optimal balancebetween sensitivity and specificity.MM probes are effective internalcontrols, since they will hybridize tononspecific sequences about aseffectively as their counterpart PMprobes. As a result, unpredictablebackground signal variationsassociated with samples from differentsources as well as from cross-hybridization can now be efficientlyquantified and subtracted with theutilization of the MM probes.

Because of the targeted design ofshort probes and multiple probes persequence (Figure 1), it is possible tounambiguously detect closely relatedsequences. For example, the yeasthistone genes HTA1 and HTA2 thatshare 93% identity at the DNA levelcan be detected with GeneChip®

arrays (L. Wodicka and D. Lockhart,unpublished data).

Sensitivity/discrimination experimentsshowed that probes of approximately 25nucleotides long provide a very effectivebalance between signal intensity and

related sequence discrimination. Thisenables expression monitoring of thousandsof targets in complex samples5. Such apowerful strategy, however, is mosteffective with shorter (e.g., 25-mer) probes,since a single base mismatch is sufficient todestabilize the hybridization. Conversely, alonger probe such as a 60-meroligonucleotide, cannot efficiently takeadvantage of a similar strategy since itrequires much longer stretches ofmutations or deletions to achieve a similardiscrimination effect6. In addition, 60-merprobes will often not be able todiscriminate between closely relatedsequences, especially with relativelyabundant transcripts.

While it is often feasible to achieve veryhigh discrimination for a particular targetsequence by optimizing hybridizationconditions, there can be no assurance thatthe same conditions are sufficientlystringent for the other thousands of probesrepresented on the same array. Instead ofexamining the hybridization behavior ofonly a small subset of sequences asdescribed by Hughes et al6, Affymetrix hastaken an approach to optimize hybrid-ization and detection conditions bysurveying the hybridization results overessentially all probe pairs on an array.

We have used population distributionsto monitor the global hybridizationbehavior of an array. It is assumed that atrue stray signal has equal likelihood tohybridize more efficiently to either the PMor the MM probe in a pair. In contrast,under conditions optimized for globalhybridization efficiency, more PM probeswill have a higher intensity value than theircorresponding MM probes when a complextarget sample is hybridized to the array.

As shown in the histogram in Figure 2,the x-axis represents bins of thediscrimination score metric used, log(PM/MM). The negative values at the far

G E N E E X P R E S S I O N M O N I T O R I N G���� 2

The Value of Mismatch Probes

0

2

4

6

8

10

12

14

0.125 0.25 0.5 1 2 4 8 16 32 64 128 256 512 1024

Concentration pM

Sig

nal

PM-MM PM

Figure 1. Comparison of the assay sensitivity using PM probes only orPM-MM probe pairs.

���� 3

left of the graph correspond to probe pairswhere virtually all signal is in the MMprobe. Conversely, the probe pairs forwhich all the signal is in the PM probe arereflected by positive values at the right endof the graph. The data in green correspondto complete stray signal distribution and asexpected, the mode is centered onPM=MM (i.e. log (PM/MM) = 0). Thepopulation shown in red estimates thenumber of probe pairs bound to true targetsequences in the mixture and is thereforeskewed to the right. Assay conditions havebeen optimized in this experiment tomaximize the size and displacementof this population (red) relative tothe stray signal population (green).

Improving global discrimination scoresthrough the optimization of assayconditions as described above hastranslated directly into assay performanceimprovements in the form of increasedsensitivity for “present” calls andincrease/decrease change calls, as well asreduction in false positives for bothmetrics. It ensures that the optimized assayconditions will accommodate themaximum number of probes representedon the array. This strategy cannot be

accomplished by observing only a limitednumber of selected sequences.

Cross-hybridization derived signal is anintrinsic and possibly significant com-ponent of the total signal for many probesinterrogating a complex sample, resultingin inaccuracy in RNA quantitation andincreased rate of false positives andmiscalls. This was observed in the initialwork using high density arrays to monitorexpression5, and is abundantly evident inthe system presented by Hughes et al6.

If only one probe is used to interrogateeach sequence, it is absolutely essential thateach selected probe meets all requirementsto be a highly functioning probe. Forexample, the selected probe: a) does notexhibit cross-hybridization to otherbackground sequences, b) does not fold onitself, and c) needs to hybridize efficientlyto the target under the assay conditions. Tocompensate for many of these sources ofnoise, error and variance, Affymetrix uses

the multiple probe pair sampling strategyin the GeneChip® array design.

� Multiple probes offer the potentialto calculate statistics that providemore confidence in microarrayresults.Information from multiple probes persequence aggregates in statisticallyrobust ways, especially when MMprobe results are incorporated into theanalysis. Because of this, it is possibleto use statistically based algorithms toprovide confidence and probabilityinformation for absolute call results aswell as comparison ratios. Interrogat-ing sequences that correspond tomultiple 25-mers on GeneChip®

arrays vs. a single 60-mer6 results inmore reliable data that are robust andresistant to target sequence mutations.

� GeneChip® arrays offer absoluteexpression analysis in addition tocomparative analysis. Therefore,the results obtained are conduciveto building large databases.Redundancy in the number of probesused for each sequence in combinationwith statistical analysis softwareallows GeneChip® arrays to be usednot only for comparative analysis, butalso for absolute quantitativeexpression analysis. The use ofmultiple independent probe pairs todetect the same RNA moleculegreatly improves the RNAquantitation accuracy by averagingand outlierrejection. This eliminatesthe necessity to include controlsamples in each experiment as used inthe two-color assay.

Ishii et al7 compared the GeneChip®

technology with SAGE and conclud-ed, “These two methods correlatedquite well in both absolute expressionanalysis and comparative analysisduring differen-tiation.” In addition,they commented, “This findingsuggests that GeneChip® technologyis reasonably reliable for quantitativeanalysis of expression profiling andwould be appropriate as a common

Global performance evaluation onGeneChip® arrays vs. examination of asmall subset of genes maximizes thenumber of genes detected on the arrays.

Using multiple probes per sequenceallows both absolute and comparativeanalysis that is essential for building aquality database with the minimumnumber of samples.

Enhanced Discrimination in Optimized Assay System

0

2000

4000

6000

8000

10000

-5 -4 -3 -2 -1 -0 1 2 3 4 5

ln(PM/MM)

Nu

mb

ero

fP

rob

eP

airs

/B

in

Stray Signal Population Target Signal Population

Figure 2. Global evaluation of hybridization results on a GeneChip® array.

AFFYMETRIX UK Ltd

Voyager, Mercury Park,Wycombe Lane,Wooburn Green,HighWycombe HP10 0HHUnited KingdomUK and OthersTel: +44 (0) 1628 552550FranceTel: 0800919505GermanyTel: 01803001334Fax: +44 (0) 1628 552585saleseurope@affymetrix.comsupporteurope@affymetrix.com

AFFYMETRIX JAPAN K.K.

Mita NN Bldg., 16 F4-1-23 Shiba, Minato-ku,Tokyo 108-0014 JapanTel: +81-(0)3-5730-8200Fax: +81-(0)3-5730-8201salesjapan@affymetrix.comsupportjapan@affymetrix.com

Part Number 701009 Rev. 4

©2001-2007 Affymetrix, Inc. All rights reserved.Affymetrix®, ®, GeneChip®, HuSNP®, GenFlex®, CustomExpress®, CustomSeq®, NetAffx®,ToolsToTakeYouAs Far AsYourVision®,TheWayAhead™,Powered by Affymetrix™, GeneChip-compatible™, Flying Objective™, and Command Console™ are trademarks of Affymetrix, Inc. Array products may be covered by one or more of the following patentsand/or sold under license from Oxford GeneTechnology: U.S. Patent Nos. 5,445,934; 5,700,637; 5,744,305; 5,945,334; 6,054,270; 6,140,044; 6,261,776; 6,291,183; 6,346,413; 6,399,365; 6,420,169; 6,551,817;6,610,482; 6,733,977; and EP 619 321; 373 203 and other U.S. or foreign patents.

AFFYMETRIX, INC.

3420 Central ExpresswaySanta Clara, CA 95051 USATel: 1-888-DNA-CHIP (1-888-362-2447)Fax: 1-408-731-5441sales@affymetrix.comsupport@affymetrix.com

www.affymetrix.com Please visit our web site for international distributor contact information.

For research use only. Not for use in diagnostic procedures.

AFFYMETR IX® PRODUCT FAMILY > RNA ARRAYS AND REAGENTS >

GeneChip® Arrays Provide OptimalSensitivity and Specificity forMicroarray Expression Analysis

The rapid development in microarraytechnology has launched a revolutionin biology, enabling researchers tosimultaneously monitor nearly eachand every mRNA transcript in anentire cellular sample. Unprecedented,highly parallel discoveries have beenmade possible, including the ability:to follow changes in the expressionlevels of many genes simultaneously1,or to identify distinctive expressionpatterns characteristic of physiologicaland pathological states2, and to screenfor very subtle changes in response toa particular therapeutic treatment3, toname a few.

����

25-mer vs. 60-mer

There are varieties of microarraytechnologies available that use differentattachment chemistries, different strategiesin choosing immobilized molecules,different assay conditions and differentdata analysis tools.The most effective microarray platformshould offer the following qualities:

� Sensitivity to identify the rarelyexpressed transcripts in a complexbackground.

� Specificity to discern betweendifferent family members.

� Absolute quantitation to enablerobust databases to be built with aminimum number of samples.

The hybridization efficiency of twonucleic acid strands results from thecombination of many factors; theircumulative effect determines thesensitivity and specificity of a microarrayexperiment. These effects have beeninvestigated in extensive detail atAffymetrix4 and include:

� Sequence-dependent factors such aslength, extent of complementarityand the overall base composition.

� Sequence-independent factors suchas concentration of the probe andtarget, time, temperature, cationconcentration, valency and character,pH, dielectric and chaotropic media,surface characteristics of the solidsupport, and density spacing of theprobe molecules synthesized on thesurface.

� Sample-dependent complexbackground signal. Any probesequence, even under conditionsoptimized for the above factors, has

the potential to interact with othercomplementary sequences present inthe sample, even if thecomplementarity is short and partial.This may result in increasednonspecific background signal andreduced specificity and sensitivity.Furthermore, samples in which thecomposition and concentrations ofthe transcripts differ may result indifferent background signals that arenot easy to extrapolate.

The Affymetrix Approach

Affymetrix has designed the followingstrategies to specifically address the issues,and to ensure that GeneChip® arraysfeature the optimal solution for mRNAexpression analysis.

Given the complexities associated withany microarray system, Affymetrix usesstrategies that specifically address criticalissues to discern between specific and non-specific binding with the PM-MM probestrategy. Based on extensive experiment-ation, we have found that 25-mers are idealfor utilization of the PM-MM strategy.

� PM-MM probe pairs offersignificantly higher sensitivity atlow target concentrations thanusing PM probes alone.

Technical Note

platform upon which to build a geneexpression database.”

In fact, Aach et al8 has reported suchan effort in systematically integ-ratingand managing expression analysisresults from several different studiesinto one database. They have foundthat, “Generation of ERAs (mRNArelative abundance estimates) isstraightforward for data derived fromAffymetrix oligonucleotide arrays andSAGE, but (two-color) microarray-derived data present a significantissue.”

Interestingly, Hughes et al6

indicated that the ‘one probe persequence’ approach is only suitablefor obtaining comparison results intwo-color assays, but not for absoluteexpression quantitation. “Thissuggests that when the experimentaldesign is focused on determiningchanges in transcript abundancebetween two samples, rather thanabsolute transcript abundance,multiple oligonucleotides are notneeded in most cases.”

Conclusion

In conclusion, Affymetrix has employedthe unique PM-MM probe pair probedesign approach to obtain the optimalbalance of highest sensitivity andspecificity in the presence of a complexsample background. The global assayoptimization strategy involving investig-ating the hybridization results of all probeson an array as a single data set, allows thederivation of assay conditions thatmaximize the number of genes detected onan array. In addition, redundant samplingof each sequence with multiple probe pairsin a probe set provides robustness andreliability in the data obtained. Incombination, the GeneChip® platformprovides robust, reliable and reproduciblequantitative expression analysis resultswith statistical significance for effectivelybuilding large expression databases.

REFERENCES

1Cho, R.J. et al., (1998) A Genome-WideTranscriptional Analysis of the Mitotic Cell Cycle.Molecular Cell, 2:65-73

2Golub, T.R. et al., (1999) Molecular Classification ofCancer: Class Discovery and Class Prediction byGene Expression Monitoring. Science, 286:531-537

3Jin et al., (2001) Effects of Early Angiotensin-Converting Enzyme Inhibition on Cardiac GeneExpression After Acute Myocardial Infarction.Circulation, 103:736-742

4Forman, J.E. et al., (1997) Thermodynamics ofDuplex Formation and Mismatch Discrimination onPhotolithographically Synthesized OligonucleotideArrays in Molecular Modeling of Nucleic Acids.(N.B. Leontis and J. Santa Lucia Jr., eds.) AmericanChemical Society Symposium Series 682:208-228

5Lockhart, D.J. et al., (1996) Expression Monitoringby Hybridization to High-Density OligonucleotideArrays. Nature Biotechnology, 14:1675-1680

6Hughes, T.R. et al., (2001) Expression ProfilingUsing Microarrays Fabricated by an Ink-JetOligonucleotide Synthesizer. Nature Biotechnology,19: 342-347

7Ishii, M. et al., (2000) Direct Comparison ofGeneChip and SAGE on the Quantitative Accuracy inTranscript Profiling Analysis. Genomics, 68:136-143

8Aach, J. et al., (2000) Systematic Management andAnalysis of Yeast Gene Expression Data. GenomeResearch, 10:431-445 Using the combination of a pair of

25-mer probes that are designed to beperfect match (PM) and mismatch (MM)to the targets offers the balance ofhighest sensitivity and specificity in thepresence of a complex background—especially for low abundance transcripts.