Research tools known as

DNA microarrays are

already clarifying the

molecular roots of health

and disease and

speeding drug discovery.

They could also hasten

the day when custom-

tailored treatment plans

replace a one-size-fits-all

approach to health care

DOT PATTERNS EMERGE when DNA microarraysanalyze tissue samples. Individual differences inthose patterns could one day help doctors matchtreatments to the unique needs of each patient.

BY STEPHEN H. FRIEND AND ROLAND B. STOUGHTON

44 S C I E N T I F I C A M E R I C A N

THEMAGICOF MICROARRAYS

Copyright 2002 Scientific American, Inc.

MOS T P EOP LE S T RI CK E N with a cancer called diffuse large B cell lym-phoma initially respond well to standard therapy. Yet in more than half of cases,the cancer soon roars back lethally. Physicians have long assumed that the reasonsome individuals succumb quickly while others do well is that the disease actuallycomes in different forms caused by distinct molecular abnormalities. But until twoyears ago, investigators had no way to spot the patients who had the most viru-lent version and thus needed to consider the riskiest, most intensive treatment.

Then a remarkable tool known as a DNA microarray, or DNA chip, broke theimpasse. It enabled a team of researchers from the National Institutes of Health,Stanford University and elsewhere to distinguish between known long- and short-term survivors based on differences in the overall pattern of activity exhibited by

Copyright 2002 Scientific American, Inc.

46 S C I E N T I F I C A M E R I C A N F E B R U A R Y 2 0 0 2

DE

NN

IS G

ALAN

TE P

HO

TOG

RAP

HY

(pre

ced

ing

pa

ges

)

hundreds of genes in their malignant cellsat the time of diagnosis. That achieve-ment should lead to a diagnostic test ableto identify the patients in greatest danger.

DNA microarrays, first introducedcommercially in 1996, are now mainstaysof drug discovery research, and morethan 20 companies sell them or the in-struments or software needed to interpretthe information they provide. The devicesare also beginning to revolutionize howscientists explore the operation of normalcells in the body and the molecular aber-rations that underlie medical disorders.The tools promise as well to pave the wayfor faster, more accurate diagnoses ofmany conditions and to help doctors per-sonalize medical care—that is, tailor ther-apies to the exact form of disease in eachperson and select the drugs likely to workbest, with the mildest side effects, in thoseindividuals.

Tiny TroupersTHE ARRAYS COME IN several vari-eties, but all assess the composition ofgenetic material in a tissue sample, andall consist of a lawn of single-strandedDNA molecules (probes) that are teth-ered to a wafer often no bigger than athumbprint. These chips also capitalizeon a very handy property of DNA: com-plementary base pairing.

DNA is the material that forms themore than 30,000 genes in human cells—

the sequences of code that constitute theblueprints for proteins. It is built fromfour building blocks, usually referred toby the first letter of their distinguishing

chemical bases: A, C, G and T. The baseA in one strand of DNA will pair onlywith T (A’s complement) on anotherstrand, and C will pair only with G.

Hence, if a DNA molecule from a tis-sue sample binds to a probe having thesequence ATCGGC, an observer will beable to infer that the molecule from thesample has the complementary sequence:TAGCCG. RNA, which is DNA’s chem-ical cousin, also follows a strict base-pair-ing rule when binding to DNA, so the se-quence of any RNA strand that pairs upwith DNA on a microarray can be in-ferred as well.

Complementary base-pairing reac-tions have been integral to many biologi-cal tests for years. But amazingly, DNAmicroarrays can track tens of thousandsof those reactions in parallel on a singlechip. Such tracking is possible becauseeach kind of probe—be it a gene or ashorter sequence of code—sits at an as-signed spot within a checkerboardlikegrid on the chip and because the DNA orRNA molecules that get poured over thearray carry a fluorescent tag or other la-bel that can be detected by a scanner.Once a chip has been scanned, a comput-er converts the raw data into a color-cod-ed readout.

Scientists rely on DNA microarraysfor two very different purposes. So-calledgenotype applications compare the DNAon a chip with DNA in a tissue sample todetermine which genes are in the sampleor to decipher the order of code letters inas yet unsequenced strings of DNA. Fre-quently, however, investigators these days

use the devices to assess not merely thepresence or sequence of genes in a samplebut the expression, or activity level, ofthose genes. A gene is said to be expressedwhen it is transcribed into messengerRNA (mRNA) and translated into pro-tein. Messenger RNA molecules are themobile transcripts of genes and serve asthe templates for protein synthesis.

Gene HuntersRESEARCHERS have employed the ge-notype approach to compare the genesin different organisms (to find clues tothe evolutionary history of the organ-isms, for example) and to compare thegenes in tumors with those in normaltissues (to uncover subtle differences ingene composition or number). One daygene comparisons performed on DNAchips could prove valuable in medicalpractice as well.

Carefully designed arrays could, forinstance, announce the precise cause ofinfection in a patient whose flulike symp-toms (such as aches, high fever andbreathing difficulty) do not point to oneclear culprit. A surface could be arrayedwith DNA representing genes that occuronly in selected disease-causing agents,and a medical laboratory could extractand label DNA from a sample of infect-ed tissue (perhaps drawn from the per-son’s nasal passages). Binding of the pa-tient’s DNA to some gene sequence on thechip would indicate which of the agentswas at fault. Similarly, chips now beingdeveloped could signal that bioterroristshave released specific types of anthrax orother exotic germs into a community.

For better or worse, gene-detectingmicroarrays could also identify an indi-vidual’s genetic propensity to a host ofdisorders. Most genetic differences inpeople probably take the form of singlenucleotide polymorphisms, or SNPs (pro-nounced “snips”), in which a single DNAletter substitutes for another. A chip bear-ing illness-linked gene variants could beconstructed to reveal an individual’s SNPsand thus predict the person’s likelihood ofacquiring Alzheimer’s disease, diabetes,specific cancers and so on. Those peopleat greatest risk could then receive close

� DNA microarrays, also known as DNA or gene chips, can track tens of thousandsof molecular reactions in parallel on a wafer smaller than a microscope slide. Thechips can be designed to detect specific genes or to measure gene activity intissue samples.

� These properties are proving immensely valuable to cell biologists, to scientistswho study the roots of cancer and other complex diseases, and to drug researchers.Microarrays are also under study as quick diagnostic and prognostic tools.

� Protein arrays, which have great promise as diagnostic devices and as aids tobiological research, are being developed as well.

� The research and diagnostic information provided by DNA chips and proteinarrays should eventually help physicians provide highly individualized therapies.

Overview/Microarrays

Copyright 2002 Scientific American, Inc.

AGG

AC

TG

TCC

TG

AC

GGG

CC

AT

CCC

GG

TA

AAT

TC

CG

TTA

AG

GC

AAT

TC

CG

TTA

AG

GC

GENE THAT STRONGLY INCREASED ACTIVITY IN TREATED CELLS

GENE THAT STRONGLY DECREASED ACTIVITY IN TREATED CELLS

GENE THAT WAS EQUALLY ACTIVE IN TREATED AND UNTREATED CELLS

GENE THAT WAS INACTIVE IN BOTH GROUPS

AGG

AC

TG

JAR

ED

SC

HN

EID

MAN

DE

SIG

N

1Construct or buy a microarray, or chip, containing single-stranded

DNA representing thousands of differentgenes, each assigned to a specified spoton the one-by-three-inch or smallerdevice. Have every spot include thousandsto millions of copies of a DNA strand.

2Obtain two samples of liver cells;apply the drug to one sample. Then,

from each sample, collect molecules of messenger RNA (mRNA)—the mobilecopies of genes and the templates forprotein synthesis in cells.

5Put the chip in a scanner. Have a computer calculate the ratio of red to

green at each spot (to quantify anychanges in gene activity induced by thedrug) and generate a color-coded readout.

6Determine whether any genes responded strongly to the drug in ways known to promote or reflect liver damage.

Or compare the overall expression pattern produced by strong responders with the patterns produced when those genes react toknown liver toxins (right). Close similarity would indicate that thenew candidate was probably toxic as well. In the diagram, each boxrepresents a single gene’s response to a compound.

4Apply the labeled cDNAs to the chip. Binding occurswhen cDNA from a sample finds its complementary

sequence of bases on the chip (detail at right). Suchbinding means that the gene represented by the chip DNAwas active, or expressed, in the sample.

mRNA

INACTIVEGENESACTIVEGENE

PROTEIN

mRNA

cDNA

mRNA

cDNA

DRUG

cDNA FROM UNTREATED

CELLS

PAIR OF COMPLEMENTARY

BASEScDNAFROMTREATEDCELLSCHIP

DNA

SCANNER

NONTOXICSUBSTANCES

KNOWN LIVER TOXINS

NEW DRUG CANDIDATE

SEGMENTOF A CHIP

SPOT CONTAINING COPIESOF A SINGLE DNA MOLECULE

PART OF ONEDNA STRAND

TREATED CELL

MICROARRAY(CHIP)

EXAMPLES OF REACTIONS

UNTREATED CELL

READOUT

HYPOTHETICAL PROFILES OF GENE ACTIVITY IN CELLS TREATED WITH VARIOUS COMPOUNDS

DNA BASES

TO DETERMINE QUICKLY whether a potential new drug is likely toharm the liver, a researcher could follow the steps below, asking

this question: Does the drug cause genes

(the blueprints for proteins) in liver cells to alter their activity inways that are known to cause or reflect liver damage? A “yes”answer would be a sign of trouble.

HOW ARRAYS WORK

GENES

3Transcribe the mRNA into more stablecomplementary DNA (cDNA) and add

fluorescent labels—green to cDNAs derived fromuntreated cells, red to those from treated cells.

w w w . s c i a m . c o m S C I E N T I F I C A M E R I C A N 47Copyright 2002 Scientific American, Inc.

monitoring, intensive preventive care andearly intervention. Whether these kinds oftests would appeal to the public is an openquestion, though; the downside of suchknowledge can be increased anxiety andthe potential for discrimination by em-ployers and insurers.

Other valuable information providedby SNP chips would pose no threat topeople’s mental state, employability or in-surability. The gene variants we possessinfluence how our bodies process themedicines we take, which in turn influ-ences the effectiveness of the drugs andthe intensity of their side effects. Chipsthat highlighted our unique genetic sen-sitivities would help physicians choose thedrugs that work best and pose the fewestdangers in each of us. SNP chips display-ing genetic mutations that increase the ag-gressiveness of tumors might also helppathologists determine whether benign-looking tumors are actually fiercer thanthey seem based on microscopic analyses.Both types of arrays are already being in-vestigated for use in medical care.

Choice ExpressionsAS EXCIT ING AS such applicationsare, it is the other major use of arrays—

expression profiling—that has increas-ingly captivated researchers over thepast few years. Laboratory workers pro-duce these profiles by measuring theamounts of different mRNAs in a tissuesample. Generally, the more copies ofmRNA a cell makes, the more copies ofprotein it will make, so the quantities ofthe various mRNAs in a sample can in-directly indicate the types and amountsof proteins present. Proteins are often ofinterest because they control and carryout most activities in our bodies’ cellsand tissues. Chips that directly measureprotein levels are being developed [seebox on page 52], but constructing themremains challenging.

By using the genome as a sensor padto detect activity changes in a cell’s vari-ous genes, scientists can gain exquisitelydetailed “snapshots” of how a cell’s func-tions have been altered by drugs or dis-ease states. At times, knowing the over-all on-off pattern of gene activity in a sam-

ple can actually be more useful thanknowing which particular genes turn onand off in response to some influence. Inthose cases, as will be seen, the patternserves as a shorthand “signature” reflect-ing the molecular state of a sample undersome specific condition.

Expression profiling has proved in-valuable on many fronts. Cell biologistslike it because knowledge of the proteinsthat predominate after a tissue is exposedto different conditions can provide insightinto how the tissue normally compensatesfor disruptions and what goes wrongwhen diseases develop.

These scientists are also using expres-sion arrays to learn the functions of genesthat have been discovered as a result ofthe recent sequencing of nearly all theDNA in the nucleus of the human cell.Several techniques that do not involve mi-croarrays can reveal the jobs performedby newly discovered genes (or, more

properly, by the proteins those genes en-code), but those approaches do not alwayswork well or quickly. In what has cometo be called the guilt-by-association ap-plication, expression arrays can help fill inthe blanks, even in the absence of any pri-or clues to a gene’s role in the body.

This method derives from the aware-ness that no gene is an island. If genes ina tissue switch on and off together in re-sponse to some influence—say, a drug, aninfection or an induced gene mutation—

workers can surmise that those like-act-ing genes operate in the same regulatorypathway; that is, the genes work togeth-er or in series to induce a cellular re-sponse. Investigators can reasonablyguess, then, that the jobs of any original-ly mysterious genes in the group resemblethose of genes whose responsibilities arealready understood.

Drug Discovery ToolsDRUG RESEARCHERS, too, take ad-vantage of the guilt-by-association meth-od—to discover proteins not previouslyknown to operate in biological path-ways involved in diseases. Once thoseproteins are found, they can be enlistedas targets for the development of newand better medicines.

In one example, Peter S. Linsley, ourcolleague at Rosetta Inpharmatics, want-ed to identify fresh targets for drugs thatmight combat inflammatory illnesses, inwhich the immune system perverselydamages parts of the body. He thereforeasked which genes in white blood cells ofthe immune system increase and decreasetheir protein production in parallel withthe gene for a protein called interleukin-2(IL-2), which is strongly implicated in in-flammatory disorders.

He got the answer by producing ex-pression profiles for white blood cells ex-posed to various chemicals and then hav-ing a computer run a sophisticated pat-tern-matching program to pinpoint a setof genes that consistently switched on oroff when the IL-2 gene was activated.This set included a gene whose functionin the body had not been determined byother means. At about the same time, in-vestigators at the Pasteur Institute in

48 S C I E N T I F I C A M E R I C A N F E B R U A R Y 2 0 0 2

AN ARRAYOF COMPANIES

The following are just some of thecompanies that sell or are developingarray-related products and services:

DNA MICROARRAYSAffymetrix, Santa Clara, Calif.

Agilent Technologies, Palo Alto, Calif.

Amersham Biosciences, Piscataway, N.J.Axon Instruments, Union City, Calif.

BioDiscovery, Marina del Rey, Calif.

Clontech, Palo Alto, Calif.

Genomic Solutions, Ann Arbor, Mich.Mergen, San Leandro, Calif.

Motorola Life Sciences, Northbrook, Ill.

Nanogen, San Diego, Calif.

Partek, St. Peters, Mo.

PerkinElmer, Boston, Mass.

Rosetta Inpharmatics, Kirkland, Wash.

Spotfire, Cambridge, Mass.

Virtek Vision International, Ontario, Canada

PROTEIN ARRAYSBiacore International, Uppsala, Sweden

Biosite Diagnostics, San Diego, Calif.

Ciphergen, Fremont, Calif.

Large Scale Biology, Germantown, Md.

Copyright 2002 Scientific American, Inc.

Paris independently confirmed, with oth-er methods, that this gene operates in theIL-2 pathway. Together the findings sug-gest that the protein encoded by the genecould be a good target for anti-inflamma-tory drugs.

Pharmaceutical scientists use expres-sion profiling in a different way: to pickout—and eliminate—drug candidates thatare likely to produce unacceptable side ef-fects. Workers who want to determinewhether a given compound could damagethe heart, for example, can compile a com-pendium of expression profiles for heartcells exposed to existing drugs and otherchemicals. If they also treat heart cells withthe drug candidate under study, they canask a computer to compare the resultingsignature with those in the compendium.A signature matching those produced bysubstances already known to disrupt car-diac cells would raise a red flag.

A compendium of expression profilescan also help explain why a drug pro-

duces particular side effects. A pressingquestion today, for instance, is why pro-tease inhibitors, which are lifesavers topeople infected with HIV (the virus thatcauses AIDS), can lead to high cholesteroland triglyceride levels in the blood,strange redistributions of body fat, andinsulin resistance. Aware that the liver in-fluences the production and breakdownof lipids (the group that includes choles-terol and triglycerides) and of lipid-con-taining proteins, we and others at Roset-ta, in collaboration with Roger G. Ulrichand his team at Abbott Laboratories, de-

cided to see whether one protease in-hibitor—ritonavir—produced some of itsside effects by acting on the liver.

With an array representing about25,000 rat genes, we produced expressionprofiles of rat liver tissue exposed to an as-sortment of compounds that can be tox-ic to the liver. After that, we grouped thecompounds according to similarities ofexpression signatures across some 2,400genes that responded strongly to those sub-stances. Next we delivered ritonavir to ratlivers and compared the resulting expres-sion profiles with those generated earlier.

w w w . s c i a m . c o m S C I E N T I F I C A M E R I C A N 49

STEPHEN H. FRIEND and ROLAND B. STOUGHTON are colleagues at Rosetta Inpharmaticsin Kirkland, Wash., which was founded in 1996 to develop molecular profiling methods in-volving computers and DNA microarray technology. Merck & Co. acquired the companylast year. Friend is vice president of basic research at Merck and president of Rosetta. Hewas a pediatric oncologist and molecular biologist at Harvard University before becomingdirector of molecular pharmacology at the Fred Hutchinson Cancer Research Center andco-founding Rosetta. Stoughton, who has a Ph.D. in physics, is senior vice president for in-formatics at Rosetta. Before turning his attention to biotechnology, he worked on devel-oping signal-processing and pattern-recognition tools for geophysics and astrophysics.

THE

AU

THO

RS

0 50 100 150 2000

20

40

60

80

100

Time (months)

Perc

ent F

ree

of M

etas

tase

s

WORK AT ROSETTA INPHARMATICS and the Netherlands CancerInstitute suggests that microarrays can help distinguish cancerpatients with different prognoses. After determining the activity(expression) levels of genes in small, localized breast tumors fromyoung women who were followed for at least five years aftersurgery, the researchers found that the expression profiles—theoverall patterns of activity across a selection of genes in the

tumors—differed among the patients (left). A mathematicalanalysis (right) then revealed that patients whose expressionprofiles resembled a “poor prognosis” signature (the averagepattern in tumors that metastasized) were much more likely tosuffer a quick recurrence than were patients whose profilesresembled a “good prognosis” signature (the typical pattern intumors that did not spread). If such results are confirmed by others,

doctors may one day be able to discernwhich patients need the most intensivetherapy based in part on how closely theirexpression profiles match a standard goodor poor prognosis profile.

PATIENTS’ FATES

EXPRESSION PROFILES

Indi

vidu

al P

atie

nts

(gro

uped

by

sim

ilarit

y of

ove

rall

expr

essi

on p

rofil

es)

Activity Levels of Individual Genes(relative to the levels in normal breast tissue)

� GENE SHOWING INCREASED ACTIVITY

� GENE SHOWING DECREASED ACTIVITY

PREDICTING CANCER’S COURSE

FRACTION OFPATIENTS WITHMETASTASES

0/20(0%)

6/21(29%)

18/24(75%)

10/13(77%)

DAT

A D

ISP

LAY

BY

HO

NG

YUE

DAI

Ros

etta

In

ph

arm

ati

cs;

GR

APH

BY

SAR

A C

HE

N

Women having a“good prognosis”expression profile

Women having a“poor prognosis”expression profile

Copyright 2002 Scientific American, Inc.

by N. Leigh Anderson and Gunars Valkirs

LIKE DNA MICROARRAYS, protein-based chips—which array proteinsinstead of DNA molecules on a small surface—can measure thelevels of proteins in tissues. In fact, they do the job more directlyand, some evidence says, more accurately. Protein arrays alsostand alone in being able to reveal which of thousands of proteinsin a tissue interact with one another.

All these properties make protein arrays quite appealing tobiological researchers. But the average person would most likely beintrigued for a different reason. Hope is high that such chips willdramatically expand the number of conditions that doctors candiagnose quickly in their offices.

These devices should be very useful as diagnostic tools in partbecause, unlike DNA microarrays, they can glean information fromblood plasma, which is easy to obtain. Most medical disorders—

from infectious diseases to heart or kidney damage—leaveidentifiable traces in the blood, in the form of secreted or leakedproteins. Moreover, in a single test, the arrays might measure manyor all of the proteins known to flag the presence of medicalproblems. In contrast, standard diagnostic tests detect only one ora few disease-specific proteins at a time.

The design of protein arrays resembles that of DNA chips.Hundreds to thousands of distinct proteins sit (in millions ofcopies) at specified spots in a grid on a wafer-thin plate. Binding ofproteins from a blood sample to proteins on a chip reveals thenature and quantities of the sample proteins.

The kinds of proteins displayed on the chips can varydepending on the questions being asked. But the chips closest tocommercialization (initially for use by researchers) rely on the

remarkable immune system molecules called antibodies—each ofwhich recognizes and binds to one specific protein or, moreprecisely, to a specific segment of a protein. Some of these antibodychips work by what is called the sandwich method: proteinsrecognized by a chip get sandwiched between two differentantibodies, one that grabs the protein and a second that attaches afluorescent label to the snagged molecule (diagram below).

For antibody-based arrays to deliver fully on their potential foradvancing research and diagnostics, scientists will have to toppleat least two major impediments. One is the need for techniques thatmass-produce many different antibodies at once, and not just anyantibodies—those that bind tightly to one target, so as to revealeven small quantities in a sample. This problem is already beingsurmounted. The second obstacle is more fundamental. Medicalscience has so far uncovered only dozens of the perhapsthousands of proteins able to signal the presence or progress of adisease. Until chipmakers know which proteins to look for, they willbe able to seek only a limited number of disease markers in a tissuesample. Fortunately, droves of investigators are now hunting fornew disease-specific proteins. As advances in antibodymanufacture and protein discovery converge, they will yield asecond generation of protein arrays that could well transform bothmedical research and clinical practice.

N. Leigh Anderson and Gunars Valkirs collaborate on protein arrayresearch. Anderson is chief scientific officer at Large Scale BiologyCorporation in Germantown, Md. Valkirs is chief technology officerat Biosite Diagnostics in San Diego, Calif.

JAR

ED

SC

HN

EID

MAN

DE

SIG

N

A PROTEIN CHIP IN ACTION

1Apply blood from a patient to a chip, or array,consisting of antibodies assigned to specific

squares on a grid. Each square includes multiplecopies of an antibody able to bind to a specificprotein from one organism and so represents adistinct disease-causing agent.

ANTIBODY CHIP

READOUTPROTEINFROM BLOOD

ANTHRAXANTIBODY

UNBOUNDANTIBODIES

FLUORESCENTLY LABELED ANTIBODIES

ANTIBODY TOAN ANTHRAXPROTEIN

ANTIBODY TOA SMALLPOXPROTEIN

ANTIBODY TOAN INFLUENZAPROTEIN

PROTEINSIN BLOOD

DOT INDICATINGTHAT THE PATIENT HAS ANTHRAX

SCANNER

2Apply fluorescently labeled antibodies able to attach to a second site on the proteins

recognizable by the antibodies on the chip. If aprotein from the blood has bound to the chip, one of these fluorescent antibodies will bind to thatprotein, enclosing it in an antibody “sandwich.”

3Feed the chip into ascanner to determine

which organism is presentin the patient’s body. In thiscase, the culprit is shown tobe a strain of anthrax.

DOCTORS MIGHT ONE DAY use a “sandwich assay” to identifythe infectious agent responsible for a patient’s illness. Is it a common flu bug or a new, deadly variety? Might the

tuberculosis bacterium be at fault—or even anthrax, smallpoxor Q fever microorganisms unleashed by bioterrorists?Following the steps below would reveal the answer.

Protein Arrays–A New Option

LABELEDANTIBODY

A PROTEIN ARRAY IN ACTION

Copyright 2002 Scientific American, Inc.

Ritonavir, we learned, leads to acti-vation of genes that are usually quieted inresponse to a well-known lipid-loweringagent; ritonavir also decreases the pro-duction of proteins that normally assem-ble into proteosomes, structures thatbreak down no-longer-useful proteins,including lipid-containing types. Thesefindings suggest that ritonavir raises lipidlevels in the liver—and hence in theblood—in part by elevating the liver’ssynthesis of lipids and inhibiting itsbreakdown of lipid-containing proteins.Further study of exactly how ritonavir in-teracts with the lipid- and proteosome-producing pathways will provide ideasfor reducing its side effects.

Treatment TailorsHAVING AN ENLARGED arsenal ofdrugs, and more drugs with fewer sideeffects, would be a great outcome of themolecular profiling made possible byDNA array studies. But many physi-cians are hoping for an even better re-sult: rapid diagnostic tools that woulddivide patients with similar symptomsinto separate groups that would benefitfrom different treatment plans. As thelymphoma study mentioned at the startof this article demonstrated, cancer spe-cialists in particular desperately needways to identify patients who requiremaximally aggressive treatment fromthe beginning.

Research into breast cancer by ourgroup at Rosetta, working with collabo-rators from the Netherlands Cancer In-stitute in Amsterdam, demonstrates howexpression arrays can help [see box onpage 49]. In this case, we wanted to inventa test able to determine which young pa-tients with early-stage breast cancer (withno evidence of cancer in the lymph nodes)need systemic drug therapy to prevent tu-mor spread (metastasis) after surgery andwhich do not. Although current guide-lines recommend systemic treatment forabout 90 percent of these women, a goodmany of them would probably avoid dis-tant metastases even if they did not havesuch treatment. Unfortunately, standardtools cannot single out the women atgreatest risk.

We began by generating expressionprofiles for tumors from close to 100women under age 55 whose clinical coursehad been followed for more than fiveyears after surgery. We initially workedwith a microarray representing 25,000human genes. In the end, we found thatone particular signature produced byabout 70 genes strongly indicated thatmetastases would soon appear. In addi-tion, the opposite pattern was stronglyindicative of a good prognosis. Clearly,some tumors are programmed to metas-tasize before they grow to a size smallerthan half a dime, whereas other, largermasses are programmed not to spread.

Our results have to be confirmed byothers before expression profiling can be-come a routine part of breast cancerworkups. Within two years, many med-ical centers will probably begin to test ex-pression profiling as a guide to therapy,not just for breast cancer but for othertypes as well. Other diseases need im-proved diagnostic tools, too. Expressionprofiling might help distinguish sub-groups of patients with such disorders asasthma, diabetes or obesity who have spe-cial treatment needs. Those applicationsare now under study.

Before microarrays can live up totheir full potential as research and diag-nostic tools, several roadblocks have tobe toppled. The chips, scanners and oth-er accoutrements remain expensive (en-gendering “array envy” in many under-funded academics). Presumably, how-ever, costs will drop with time.

Yet even if prices fall, the technolo-gies may prove infeasible, at least initial-ly, for doctors’ offices or standard med-ical laboratories. Few physicians or tech-

nicians have the equipment and the skillto prepare tissue samples properly for usewith arrays. What is more, to diagnose,say, liver disease based on changes ingene expression in liver cells, a doctorwould ideally need to obtain tissue fromthe liver. But that organ is not readily accessible.

These problems loom large right nowbut are probably surmountable with in-genuity. At times, for instance, accessibletissues might function as acceptablestand-ins for inaccessible ones. More-over, in some instances, microarraysthemselves may not have to be used; theymight provide the research informationneeded for devising new diagnostic tests,which can then take other forms.

As the operations of cells and the en-tire body become better understood,physicians will be able to make more pre-cise diagnoses, to offer patients more so-phisticated therapies (possibly includinggene therapies), and to tailor these inter-ventions to an individual’s genetic back-ground and current state of physiologicalfunctioning. By the year 2020, healthmaintenance organizations and their ilkcould conceivably keep in silico modelsof the personal molecular states of theirsubscribers—virtual simulations thatcould be updated constantly with mi-croarray and other data from doctor vis-its and with new scientific informationabout cell biology. Perhaps some sub-scribers won’t like that idea and will for-go a rate discount—and quite possiblythe best care—in return for a feeling ofprivacy. Those who go along with theprogram, though, will probably delay theeffects of aging more successfully andlead healthier lives.

w w w . s c i a m . c o m S C I E N T I F I C A M E R I C A N 53

The Chipping Forecast. Supplement to Nature Genetics, Vol. 21, pages 1–60; January 1999.Genomics, Gene Expression and DNA Arrays. David Lockhart and Elizabeth Winzeler in Nature, Vol. 405, pages 827–836; June 15, 2000.Experimental Annotation of the Human Genome using Microarray Technology. D. D. Shoemaker et al. in Nature, Vol. 409, pages 922–927; February 15, 2001.Web sites listing links and publications on microarrays can be found at:http://bioinformatics.phrma.org/microarrays.htmlhttp://industry.ebi.ac.uk/~alan/MicroArray/www.rii.com/publications/default.htmhttp://ihome.cuhk.edu.hk/~b400559/2001j_mray.htmlwww.biologie.ens.fr/en/genetiqu/puces/links.html#news

M O R E T O E X P L O R E

Copyright 2002 Scientific American, Inc.