doi:10.1016/S1470-2045(04)01382-8

THE LANCET Oncology Vol 5 February 2004 http://oncology.thelancet.com 89

No currently available chemotherapy seems likely tosubstantially improve outcome in most patients with braintumours. Several resistance-associated cellular factors,which were discovered in other cancer models, have alsobeen identified in brain tumours. Although thesemechanisms play some part in resistance in brain tumours,they are not sufficient to explain the poor clinical response tochemotherapy. There could be other brain-tumour-specificgenetic profiles that are associated with tumour sensitivity tochemotherapy. There is increasing awareness that drugresistance in brain tumours is not a result of changes insingle molecular pathways but is likely to involve a complexnetwork of regulatory dynamics. Further insights intochemoresistance in brain tumours could come withcomprehensive characterisation of their gene expression, aswell as the genetic changes occurring in response tochemotherapy. Recent progress in high-throughputbioanalytical methods for genome-wide studies has madepossible a novel research model of initial hypothesisgeneration followed by functional testing of the generatedhypothesis.

Lancet Oncol 2004; 5: 89–100

Despite the well-established use of chemotherapeuticapproaches in the treatment of malignant brain tumours,prognosis has not improved much over the past 20 yearsand continues to be dismal for the majority of patients.Although chemotherapy prolongs survival for some typesof brain tumours, such as medulloblastoma, primitiveneuroectodermal tumour, oligodendroglioma, germ-celltumour, and primary central-nervous-system lymphoma,for most histological types chemotherapy is applied as a lastresort rather than as an established beneficial componentof a multimodality treatment regimen. The most commontype of brain tumour, the large group of high-gradegliomas, tends to be resistant to chemotherapy, and long-term tumour control is rarely achieved.

Chemotherapy for brain tumours poses a specialchallenge owing to the existence of a blood–tumour barrier,which is intact in those tumour regions that are biologicallyand clinically most important—namely in areas of the braininfiltrated with tumour cells. However, there is growingconsensus that, in addition to difficulties related to theblood–tumour barrier and pharmacokinetic issues, themodest effect of chemotherapy in brain tumours is mainlylinked to tumour-cell resistance as shown by certain geneticand epigenetic factors.1,2 This review discusses the potential

usefulness of recent progress in high-throughputbioanalytical methods for genome studies as tools to identifydeterminants of response and to characterise the geneticchanges associated with chemotherapeutic perturbation inbrain neoplasms (figure 1).

Drug resistance in human cancerMost cancers have heterogeneous cell populations. Goldieand Coldman3 postulated a mathematical model forrelating drug sensitivity of human tumours to genemutation rate and hypothesised that tumours undergospontaneous genetic changes that enable development ofresistance to cytotoxic agents to which the tumours havenever been exposed. Selection of mammalian cells in vitrofor resistance to cytotoxic agents via exposure toincrementally increased sublethal drug concentrationscommonly results in cross-resistance to many other drugs,which share little structural similarity with the primaryselective agent and act at different intracellular targets. Thispleiotropic process, which is a major impediment to

ReviewGenomics and brain-tumour drug resistance

MB is Assistant Professor of Experimental Neurooncology,Department of General Neurosurgery at the Neurocenter, Universityof Freiburg, Germany; and visiting Assistant Professor, Division ofOncology, Stanford University School of Medicine, CA, USA. CB isa molecular biologist in the Department of General Neurosurgery atthe Neurocenter, University of Freiburg, Germany. BIS is Professorof Medicine (Oncology and Clinical Pharmacology) at StanfordUniversity School of Medicine, CA, USA.

Correspondence: Dr Markus Bredel, Division of Oncology,Stanford University School of Medicine, 269 Campus Drive, CCSR-1110, Stanford, CA 94305-5151, USA. Tel: +1 650 498 6949. Fax: +1 510 438 8830. Email: [email protected]

Genomics-based hypothesis generation: a novelapproach to unravelling drug resistance in braintumours?

Markus Bredel, Claudia Bredel, Branimir I Sikic

Figure 1. Images of a cDNA microarray slide profiling gene expression.

For personal use. Only reproduce with permission from The Lancet.

THE LANCET Oncology Vol 5 February 2004 http://oncology.thelancet.com90

treating patients with cancer, has become known asmultiple-drug (or multidrug) resistance (MDR).

Resistance mechanisms are expressed constitutivelyeither as genes normally expressed by the tissue of origin ofthe tumour (eg, MDR1 in colon cancer) or as geneticalterations during tumorigenesis (eg, p53 mutations). Suchintrinsic (or de novo or constitutive) resistance results inlittle treatment response or failure of initial treatment(figure 2). In many tumours, early drug treatment canachieve substantial cell killing, though there might beselection of a clonal variant of cells that confers acquiredresistance to subsequent treatment, leading via clonalexpansion, to repopulation and tumour recurrence (drugselection; figure 2). Two principles help to explain theacquired resistance phenotype (figure 2). First, the inherentgenetic instability of tumour cells can lead pre-existingresistant cell clones that are present before initial therapy toexpand under long-term chemotherapy. Second, drugresistance can be acquired through induction of resistancepathways in cancer cells during chemotherapy. Anticancer

agents have mutagenic potential and could cause mutationsin key cellular target genes (genetic level); and chemo-therapy can cause surviving cells to induce the coordinatedexpression of protective stress response genes (epigeneticlevel). The resistant clones formed could be selected bysubsequent therapy and expand, eventually leading todisease relapse (figure 2).

Drug-resistance in brain tumoursSeveral mechanisms of drug resistance discovered in othertumour models have also been implicated in brain tumours(figure 3). They include ATP-dependent efflux of cytotoxicagents by transmembrane transporter proteins encoded by thegenes multiple-drug resistance 1 (MDR1, ABCB1) andmultidrug-resistance-associated protein (MRP1, ABCC1);DNA damage caused by quantitative changes in expression of DNA topoisomerase II�; increased detoxification ofalkylating agents by glutathione and the glutathione-linkedenzyme system, particularly glutathione-S-transferases; andincreased activity and expression of members of the protein

kinase C family, causing changes incell-cycle transition and in theexpression status of various resistancemarkers that are relevant to drugaction.1,2 In addition, a deficiency inpathways of DNA mismatch repair(which render tumour cells tolerant tomethylation) and increased nucleotideexcision repair of DNA adducts owingto altered activity of poly(ADP)-ribosepolymerase and the product of theexcision-repair cross-complementingrodent repair deficiency gene 2(ERCC2) could be involved in theMDR of brain-tumour cells.1,2 Sincemost chemotherapeutic agents killtumour cells via apoptosis, dysfunctionof genes involved in apoptoticpathways and resultant impaired abilityto commit to apoptosis can alsocontribute to chemotherapy resistanceof brain-tumour cells (figure 3).

In addition to MDR phenotypes,resistance to chemotherapeutic drugscan affect single agents or a class ofrelated drugs that share structuralsimilarity. This type of resistance,which is broadly referred to asindividual drug resistance and has also been associated with resistance to some chemotherapies in braintumours (figure 3), might be caused by raised concentrations of enzymesinvolved in intracellular drug metabolism, for example O6-methyl-guanine-DNA methyltransferase(MGMT), thymidylate synthase, anddihydrofolate reductase.1,2 Further-more, metallothioneins can function to

Review Genomics and brain-tumour drug resistance

Anticancer drug treatment

Initial response

Anticancer drug resistance

No response

Remission Relapse

First drug exposure

Clonal selection and expansion by prolonged therapy

Resistant clones

Inherent genetic instability

Mutagenesis

Drug sensitivity

De novo

Intrinsic (epi-)genetic

profile

Pre-existingresistant clones

Acquired

Epigeneticstress

response

Figure 2. Relation between drug effect and drug resistance/sensitivity of human cancers.



inactivate anticancer drugs in brain-tumour cells. Inoligodendroglioma, loss of heterozygosity (LOH) onchromosomes 1p and 19q, a unique cytogenetic profile, isrelated to responsiveness to chemotherapy.

Importance of the hypothesis-driven researchmodel The presence of resistance factors in a subset of braintumours has generally been associated with the prevalenceof clinical resistance to chemotherapy. However,substantial uncertainty remains about whether theexpression of these factors in brain tumours is the cause ofthe low response to chemotherapy in patients with braintumours. Apart from the DNA-repair gene MGMT—forwhich expression and hypermethylation-induced geneinactivation have been related to response of brain tumoursto nitrosoureas1,2,4—no causal link between these markers inbrain tumours and failure of chemotherapy has been found.

The traditional research model has dictated that when agene is newly found to be related to drug resistance in onetype of tumour, the hypothesis that the same gene isimportant in drug resistance in brain tumours isformulated and experiments are done to test thishypothesis. Although some advances have been made withthis approach, others are needed to elucidate thepolygenetic basis of resistance to chemotherapy in braintumours (figure 4).

Many known mechanisms of drug resistance areexpressed constitutively in brain tumours and could be thecause of intrinsic resistance to therapy.1,5 However, othermechanisms are likely to be either induced by drugexposure or selected as mutations that occur during theevolution of a tumour-cell population. The observation ofintrinsic expression in brain tumours could also reflect theknown physiological role of some of these genes in non-neoplastic brain, for example the contribution of the MDR1


Drug Brain tumourblood vessel lumen

Brain tumour cell

Basal membraneNucleus

Pgp

GSTGSH

GSH

MRP

Pgp

PKC

M

SG1 G2

Lethal mitosis

DNA damage

DNA repair

Apoptosis

Anti-apoptopic Pro-apoptopic

Apoptopic Components

MGMT

Alkyl G

T

TII�

DHFR

dUMP

dTMP

DNA

TS

MT

Figure 3. Subcellular location and possible mechanism of action of previously identified resistance markers in brain tumours. The MDR1-encoded P-glycoprotein (Pgp) and members of the multidrug-resistance-associated protein (MRP) family act at the cell membrane as ATP-driven drug-effluxpumps. Glutathione-S-transferase (GST) mediates conjugation of anticancer agents with reduced glutathione (GSH). MRP1 mediates excretion ofglutathione–drug conjugates. Quantitative changes in topoisomerase II� (TII�) can withdraw the cellular target of agents targeting this enzyme. NuclearO6-methylguanine-DNA methyltransferase (MGMT) may act to remove drug-induced O6-alkylguanine-DNA. Increased expression of nuclear thymidylatesynthase (TS) and dihydrofolate reductase (DHFR), which are involved in the conversion of deoxyuridine monophosphate (dUMP) to deoxythymidinemonophosphate (dTMP), can result in maintenance of free enzyme when challenged with drugs that deplete these enzymes. Metallothioneins (MT) caninactivate drugs either by chelation or by intracellular sequestration. Protein kinase C (PKC) induces a cell-cycle arrest to allow repair of drug-inducedDNA damage before the cell enters lethal mitosis, can modulate the expression of a wide variety of resistance markers, and influences apoptoticpathways. Changes in the balance of proapoptotic and antiapoptotic factors prevent drug-induced apoptosis. The yellow arrows indicate increased ordecreased marker expression associated with drug resistance.



(ABCB1) gene encoding P-glycoprotein to the formation ofa blood–brain barrier. The dual roles of other genes, such asthose for thymidylate synthase, dihydrofolate reductase,and topoisomerase II�, in drug resistance and cellproliferation raise uncertainty over whether changes inexpression of some of these genes in brain tumours arerelated to drug resistance, cell-cycle turnover, or both.Changes in the expression state of some of these genesmight be completely independent of any role in drugresistance. For example, the locus for topoisomerase II� isclose to that of the oncogene HER2/neu on chromosome17q. Overexpression of HER2/neu has been shown in asubset of brain tumours,6 and the accompanying increasedexpression of topoisomerase II� might be a consequence ofmalignant transformation in these neoplasms.

The search for brain-tumour type-specificresistance fingerprintsEvidence for the existence of a tumour type-specificresistance profile to a certain anticancer agent in the formof a resistance “fingerprint” comes from recentinvestigations that have used large-scale gene expressionprofiling and have shown substantially differing geneticresponse patterns to a defined chemotherapeutic agent fordistinct histological types of cancer cells.7–10 There isgrowing awareness that drug resistance in human cancer isprobably dictated in a combined way by the abnormalexpression of groups of coregulated genes; this associationsuggests that many more genes are involved in thedevelopment of resistance phenotypes in brain tumoursthan the gene expression changes described thus far for alimited number of known resistance genes (figure 5).

The phenotypes of benign and malignant cells andtissues ultimately depend on the types and amounts ofproteins present at any given time. Translational and post-translational biochemical modifications exert decisiveeffects in regulating the amounts of active forms of proteinsinvolved in pathological and pharmacological conditions.However, a primary mechanism by which protein

expression is regulated and by which cells adjust to varyingconditions is through variation of the abundance of mRNApresent in the cell. Accordingly, a primary event in thedevelopment of tumour-cell resistance could be a change indegree of gene transcription. Analysis of the pattern ofvariation in expression of genes could therefore be useful inassessment of variation in the resistance characteristics oftumour cells and tissues (figure 4). Pharmacogenomics ispredicated on the concept that cellular responses toanticancer drugs are analogous to a higher gated logiccircuit, in which, in some cases, many contradictory inputsmodulated by features such as feedback, feed-forward,error checking, and redundancy, are summed to produce aresponse (figure 5).

Genomics-based generation of drug-resistancehypothesesThere is increasing recognition of the value ofcomprehensive approaches to the molecular characteri-sation of biological phenotypes such as drug resistance.This appreciation has highlighted the need forbiotechnology that allows parallel, large-scale assessment ofmany genes. The emergence of high-throughput genomicsplatforms has enabled genome-wide studies of geneexpression. The key feature of successful genomiccharacterisation of a drug-resistance state is the ability tomeasure changes in mRNA abundance accompanying theformation of this state, or differences between sensitive andresistant phenotypes.

Various strategies have emerged for gene expressionprofiling, including serial analysis of gene expression(SAGE), differential display, subtractive hybridisationapproaches, and massively parallel signature sequencing.Microarray technology that uses densely spotted cDNAs oroligonucleotides allows the large-scale analysis of geneexpression in a high-throughput way (figure 1). There aretwo main types of microarray systems. In the single-colourAffymetrix system, short oligonucleotide probes aresynthesised in situ on a chip. Each gene is represented by a


Research paradigm

Resistance mechanism

Brain tumour resistance phenotype

Traditional Genomics-based

Observation inother tumours

Hypothesistesting

Hypothesisgeneration Genomics

?

?

Hypothesisformulation

Empirical reductionism Functional genomics

Figure 4. Traditional versus genomics-based research approaches.



set of probe pairs that correspond to different gene regions.Each probe pair consists of an exact-match base-pairsequence probe and a mismatch sequence probe. Geneexpression is measured absolutely by the brightness ofhybridisation to the multiple probes representing a gene orrelatively to the hybridisation signal of a reference RNA ona separate chip. By contrast, cDNA microarrays use a dual-colour hybridisation scheme for the comparison of geneexpression in two or more samples. Typically, mixtures of(m)RNA reverse-transcription-generated Cy5-labelled“red” cDNA from one condition (eg, resistant or sensitivecell line) and mixtures of Cy3-labelled “green” cDNA fromanother condition (eg, reference sample) are combined andallowed to hybridise to the glass microarray slide (figure 1).The common reference control—for example, a pooledsample of (m)RNAs from a set of cancer cell-lines—generally has no biological significance and simplycontributes a consistently measurable signal in thedenominator of the assayed ratios. The amount of red andgreen fluorescence (R/G ratio) at each spot allows theamount of (m)RNA for each gene in the test sample to becompared with that in the common reference sample,avoiding the need for absolute measurements of (m)RNA(figure 1).

An increasing number of studies have used microarraymethods to examine gene expression patterns characteristicof sensitive or drug-resistance phenotypes in varioushuman cancers.7–25 Many of these studies have used the invitro resistance selection/induction model, which meansthat a more or less sensitive parental cell line is exposed to(sub)lethal drug concentrations until a sufficient degree ofresistance is achieved (figure 6). These studies haveprovided preliminary insights into how changes in thesensitivity state of a cancer cell are reflected in changes inoverall gene expression. Other genomic studies have

examined the sensitivity state of various established andlow-passage parental cancer cell lines.26–28 Scherf andcolleagues29 first integrated large databases on geneexpression and chemosensitivity. They linked theexpression profiles of about 8000 unique genes assessed in aset of 60 human tumour cell-lines, termed the NCI60, withthe activity patterns of more than 70 000 anticancer drugsobtained from a database of the DevelopmentalTherapeutics Program (DTP) of the US National CancerInstitute. Many gene–drug relations were identified. Similaranalysis in the NCI60 by Staunton and co-workers30 with6817-gene microarrays revealed gene-expression-basedclassifiers of sensitivity or resistance for 232 compounds.Butte and colleagues31 identified gene regulatory networksdetermining chemotherapeutic susceptibility of the NCI60to 5084 agents by use of 7245-gene oligonucleotide arrays.

Dan and associates24 have used 9200-gene cDNAmicroarrays to compare the sensitivity state of 39 humancancer cell-lines to 55 anticancer agents. Pearsoncorrelation and clustering analysis identified genes withexpression patterns that showed significant association withpatterns of drug responsiveness. Some genes werecommonly associated with response to various drug classes,whereas others were associated only with response to drugswith a similar mechanism of action. Similarly, Zembutsuand colleagues20 analysed the expression profile of morethan 23 000 genes in a panel of 85 cancer xenografts,derived from nine human organs and implanted into nudemice, along with chemosensitivity to nine differinganticancer drugs. This approach identified various genesthat were significantly associated with sensitivity to one ormore drugs examined.

In addition, gene expression profiles have been linkedto drug response in clinical tumour samples (figure 6).Kihara and co-workers15 used a cDNA microarray


Response to chemotherapy

Genomic reponse fingerprint

Gene expression

Microarray

Gene copy number

Array CGH

Polygenetic network(including contradictory

inputs)

genome-wide mRNA abundance genome-wide DNA copy numbers

Tumour type-specific

co-regulation co-alteration

Drug-specific

Clinicaltumour response

Figure 5. The molecular determination of tumour response to chemotherapy, which includes in some cases many contradictory inputs, is a feature ofthe genome that might be determined by the expression and copy number of many known and unknown genes. Two complementary strategies couldbe useful in assessing molecular response fingerprints. First, gene expression profiling with microarrays allows characterisation of genome-wide mRNAabundance. Second, microarray-based comparative genomic hybridisation (array CGH) enables assessment of DNA copy numbers throughout thegenome.



containing 9216 genes to predict the sensitivity ofoesophageal tumours to adjuvant chemotherapy. Theyidentified 52 genes that were linked to prognosis andpossibly to chemosensitivity and/or chemoresistance, and a

drug response score based on differential expression ofthese genes was predictive of outlook for individualpatients. Sotiriou and associates17 investigated the responseto systemic chemotherapy of fine-needle aspirates of breast


Established cell line

Stablestep-wisetreatment

Transientsingle-steptreatment

cDNA microarray

Mocktreatment

Surgical tumour sample

Outlier genes

Gene(s) of clinical interest

PatientHistological tumour type

In vitro system In vivo system

sample pair

Resistantdaughtercell line

Stresseddaughtercell line

Distinct geneexpression profiles

Candidate markervalidation

Sensitiveparentalcell line

Purifiedresistanttumourcells

Purifiedsensitivetumourcells

Biostatistic algorithms

Post-chemotherapysample

Pre-chemotherapysample

Laser microbeammicrodissection

Real-time quantitative PCR

Pharmacogenomics

Candidate markers

Laser pressurecatapulting

Low passagecell line

Magnetic cellsorting

drug

Resistance induction/selection

Clinical tumour response data

Figure 6. Genomics-based resistance research can pursue a dual strategy of assessing tumour (cell) responsiveness in vitro and in vivo. In vitro resistance research commonly involves the selection/induction model in which a parental cell line is selected/induced for resistance viachronic exposure to sublethal drug concentrations until a sufficient degree of resistance is achieved. Comparison of the gene expression of theparental line and the resistant subline can identify subsets of genes that are differentially expressed. Inclusion of stress response data by transientsingle-step drug exposure can help in dissecting gene expression changes primarily linked to drug resistance from secondary genetic changes thatmay be only downstream and marginally relevant to resistance formation. Ideally, genomics-based in vivo approaches to resistance research arecombined with purification techniques that allow for enrichment of tumour cells. Laser-assisted tissue microdissection has emerged as the methodof choice to limit tissue heterogeneity. Gene expression profiling in surgical tumour samples can use a dual strategy. On the one hand, theexpression pattern of prechemotherapy samples is directly compared with that of postchemotherapy samples; on the other hand, with apharmacogenomics strategy, expression data are directly correlated with clinical tumour response data to identify response-associated patterns ofgene expression.



cancers with expression profiles from 7600-feature cDNAmicroarrays. Candidate gene expression profiles wereidentified that distinguished tumours with completeresponse to chemotherapy from those with no response.

Application of genomics methodsGene expression profiling has been used for moleculargenetic characterisation, classification, and subclassificationof intracranial neoplasms, including meningiomas32 andvarious histological types of glioma.33–46 Such strategies haveprovided initial evidence to offer the promise of refinedprognosis prediction in brain tumours such as diffuseastrocytoma,42 malignant glioma,37,47 and medulloblastoma.48,49

Microarray analysis has also identified a potential serummarker for glioblastoma multiforme.50 The potential value ofmicroarrays as a tool for assessing variation in the responsecharacteristics and resistance state of brain tumours is justbeginning to be appreciated.

Rhee and colleagues51 characterised cellular pathwaysinvolved in the response of glioblastoma multiforme cells tothe nitrosourea carmustine. Their array contained only 588genes, of which 17 had differential expression between acarmustine-sensitive glioblastoma multiforme subline and a resistant parental subline. Most (13) of these genesshowed lower expression in the resistant variant.Expression of MGMT, which was previously thought to bethe primary mechanism of resistance of glioblastomamultiforme cells to carmustine, was similar in both sublinesand not altered by carmustine treatment, which suggeststhat the MGMT pathway did not contribute to thecarmustine resistance phenotype. By contrast, six otherDNA repair genes, including ERCC2, were downregulatedin the sensitive subclone.

Bacolod and co-workers52 used microarrays with 12 000elements to study changes in gene expression accompanyingcarmustine resistance selection in medulloblastoma cells.

Gene expression profiling identified 89 genes withupregulated or downregulated expression. These includedthe changed expression of genes related to various biologicalfunctions, including increased expression of severalmetallothionein genes and reduced expression of several proapoptotic genes. Although MGMT activity of oneresistant clone was twice that of the sensitive parental cells, a second clone selected with carmustine and O6-benzylguanine (an irreversible inhibitor of MGMT) didnot show changes in MGMT, which implies that othermechanisms must have had a role in the resistancephenotype of these cells.

Baseline information about differences in chemo-sensitivity between genetic subsets of oligodendrogliomahas been provided by gene expression profiling. Mukasaand colleagues53 used oligonucleotide microarrays withmore than 12 600 human genes and expressed sequencetags to examine 11 tumour specimens of two cytogeneticsubsets of oligodendroglioma, (characterised by thepresence or absence of loss of heterozygosity [LOH] ofchromosome 1p), which show profoundly differentresponse rates to chemotherapy.1,2 The researchersidentified 209 genes expressed differentially between thetumour subsets, with 86 genes showing higher expressionand 123 lower expression in tumours with 1p LOH.Notably, only 60% of the 123 genes with reducedexpression in tumours with 1p LOH were located onchromosomes 1 and 19 (oligodendrogliomas with 1p LOHcommonly also show 19q deletion).

Potential pitfalls of comprehensive geneexpression profiling strategiesThe strength of microarray technology in research on drugresistance is the ability to assess the expression of manygenes at the same time and subsequently, by use ofsophisticated computer-driven pattern recognition and


Functional testing via conventionalbiochemistry for logical coherence

Prioritisation of probable hypotheses Many hypotheses

Raw microarray data

BioinformaticsIdentification of co-regulated

genes or gene families

Data mining for meaningful information

Multiple upregulated and downregulated genes

hiearchicalclustering

relevancenetworks

self-organising

maps

geneshaving

principalcomponent

analysis

terrainmaps

qualitythresholdclustering

Figure 7. Hypotheses generated from a microarray experiment require sophisticated data mining for meaningful information to be obtained. In general,coregulated genes and families of genes are identified via biostatistical approaches with the assumption that genes that are similarly altered inexpression are likely to be functionally related. Functional testing by conventional biochemical approaches is essential for logical coherence and toascertain a causal link between a resistance phenotype and altered gene expression.



map building, to cluster those that are simultaneouslyupregulated or downregulated when a tumour becomeschemoresistant (figure 7). The studies on drug resistancethat have used microarray methods provide evidence thatgenomic responses of a tumour cell to a chemotherapeuticagent can include altered expression of a substantialnumber of genes that might well be in the range of 5–10%of the human genome; a large proportion of these changeshave been confirmed by traditional methods of measuringmRNA.7–9 In addition, the intrinsic sensitivity of a tumourmight be reflected by the expression of several genes or agroup rather than by single genes. For example, in thestudy by Zembutsu and colleagues,20 comparison of thegene expression profiles of 85 human cancer xenograftswith sensitivities to various drugs identified 1578 genes forwhich degree of expression was significantly related tochemosensitivity; 333 of these genes showed significantassociations with two or more drugs, and 32 genes werelinked with sensitivity to six or seven drugs.

Microarray technology provides a rapid way ofcollecting large amounts of data and can generate newhypotheses on mechanisms leading to drug resistance.However, potential problems with the application of thisapproach to drug resistance research have to be recognised.In particular, biostatistical methods have to be available toanalyse the substantial number of hypotheses that might begenerated from a microarray experiment (figure 7). Thisapproach does not differentiate between genes that mightcause a certain resistance phenotype and secondary geneticchanges that might be downstream, marginally relevant, oreven irrelevant to the resistance phenotype. In fact, asubstantial proportion of transcriptional changes observedduring resistance formation probably do not immediatelyrelate to the drug-resistant phenotype. The early changes inresistance selection are likely to be broader than mereresistance formation, representing a stress response of atumour cell when exposed to a chemotherapeutic agentrather than a targeted tumour-cell response to reduce itsvulnerability to this drug.

Identification of sets of genes that concomitantly showaltered expression during development of drug resistancecould directly relate to regulation and coregulation forfamilies of genes. Therefore, the most common approach toorganisation of microarray data is hierarchal clustering54

(figure 7). It relies on the basic premise that genes withsimilarly altered expression profiles are likely to becoregulated and functionally related and uses a “guilt byassociation” strategy to identify functional clusters.Hypotheses generated by such associations requireconfirmation by conventional biochemical approaches, toestablish logical coherence between the resistance state of abrain-tumour cell and an altered state of expression ofsingle genes or a group of genes (figures 4 and 7).Alternative statistical approaches for data mining includethe application of self-organising maps,55 the constructionof relevance networks,27,31 principal component analysis,56

terrain maps,57 quality threshold clustering,58 and amathematical strategy called “gene shaving” that differsfrom hierarchical clustering in that it takes into account

that genes may belong to more than one cluster59 (figure 7).In addition, computational algorithms allow organisationof upregulated and downregulated genes on the basis oftheir ontology. For example, sophisticated software can beused to view and analyse gene expression data according tobiological pathways and gene relations can be directlyexplored and annotated.60 Ontology-based arrangement ofdifferential gene expression can provide immediate cluesabout the potential involvement of certain signalling andbiochemical pathways in drug resistance states of tumourcells.

A further problem of microarray technology is its lowsensitivity to transcriptional changes in the baseline range,which complicates the identification of modest changes ingene expression that might be relevant to drug resistance.In addition, normal variability in gene expression couldlead to differential expression patterns of a gene in asensitive and resistant sample without reflecting the trueeffect of the resistance formation process.

In general, measurements of gene expression providedirect information on the abundance of the mRNAtemplate only, rather than on the expression of the finalgene product, the protein. This point is particularlyimportant, because concentrations of proteins can varysignificantly among genes with similar abundance ofmRNA. Conversely, there can be substantial variation inthe amounts of mRNAs encoding proteins expressed withsimilar abundance.61 At present, the efficiency of obtainingan overall view of gene expression by measurement ofmRNA is better than that for similar proteomic methods, interms of throughput, reproducibility, and compatibilitywith clinical material. The degree of protein expression ofsmall numbers of genes potentially associated with responseto therapy can be examined in a high-throughput way bytissue-microarray-based immunohistochemistry, whichallows rapid screening of large numbers of tumourspecimens for candidate marker validation on paraffin-embedded tissue. The usefulness of tissue microarrays forneuropathology research has been readily demonstrated.62

Recent progress in the application of microarrays to cytogenetics—particularly comparative genomichybridisation—has led to chip-based genome-widescreening for changes in gene copy numbers.63 Since genedose effects are also relevant to chemosensitivity andchanges in gene copy number are frequently observed intumour cells exposed to chemotherapeutic agents, theapplication of array comparative genomic hybridisation toresearch on brain-tumour drug resistance might providevaluable information on codeleted and coamplified genesthat are potentially important to resistance formation(figure 5).

In vitro drug resistance versus in vivochemoresistanceExperiments in tumour cell-lines are an important first stepin characterising genomic changes linked to resistancedevelopment in vitro. However, culture-derived patterns ofgene expression changes accompanying resistanceformation may not correlate with resistance-associated




gene expression in solid brain tumours. Long-term brain-tumour cell lines differ to some extent from brain-tumourtissue samples in their basic gene expression profile. Hessand colleagues64 showed that in hierarchical cluster analysisof microarray data, three glioblastoma multiforme cell linesformed one main cluster and ten glioblastoma multiformetissue samples formed a separate cluster. Multidimensionalscaling and principal component analysis provided furtherevidence that the cell lines clearly differed from the tissuesamples and that the cell lines differed genetically morebetween themselves than the tissues did. Sasaki andassociates65 showed that gene expression of tissue-culturedmeningiomas and corresponding in situ meningiomasdiffered significantly for a portion of genes. Accordingly, anintegrative look at resistance-related changes in geneexpression in response to drug application necessarily hasto include actual brain tumour specimens (figure 6).

One of the confounding factors in studying primaryclinical specimens by microarrays is the mixture ofneoplastic and various normal cell-types in the tumours.Tissue heterogeneity in brain tumours—such as thepresence of non-neoplastic fibroblasts, endothelial cells,immune cells, and necrosis—could substantially interferewith drug-resistance analyses in brain tumour specimensobtained by surgery. Much attention has lately focused onstrategies designed to enrich selected cell types from tissuesamples to facilitate the use of material from patients in thestudy of markers or target discovery. In this regard, laser-assisted tissue microdissection has emerged as a method ofchoice to reduce tissue heterogeneity (figures 6 and 8).Several distinct laser-based techniques for tissue

microdissection have been developed.For example, laser capture micro-dissection66 uses a cap coated with athermolabile film of ethylene vinylacetate, which is placed in contactwith a tissue section stained to enableselection of the relevant cell type.Visualisation is achieved with aninverted microscope. An appliedfocused laser beam of variablediameter causes localised melting ofthe film over selected cells and thesecells fuse to the cap. The selectedtissue is selectively removed when thecap is lifted. Several thousand targetedlaser “shots” of material can be madeon a single cap, thereby enablingconcentration of the cell type ofinterest. A second technique67 avoidsmechanical contact to capturesamples of between 1 µm and severalhundred micrometers in diameter. Inthis method, the laser is used not onlyfor microbeam microdissection butalso for transfer of the selectedclusters of cells or single cells from theobject plane directly into the cap of aroutine microfuge tube via high

photonic pressure force—laser pressure catapulting (figure8). The feasibility and reproducibility of this approach forsubsequent genetic analysis with no adverse effect onmRNA have been clearly demonstrated.68

Laser microdissection has been used for reduction oftissue heterogeneity in gene expression profiling inmammalian brain69 and brain disease, including temporal-lobe epilepsy70 and brain tumours.71 Many of the studiesthat have combined microdissection and overall geneexpression analysis have used mRNA amplification toachieve sufficient amounts of mRNA.72,73 Since amplifi-cation protocols are prone to representative biases, otherstudies have shown that expression profiles for thousandsof genes can be successfully generated with non-amplifiedmRNA derived from clinical cancer specimens procured bylaser microdissection.74–76

PerspectiveWith the perception that gene networks rather thanindividual genes determine chemoresistance, genomics willhave an important role in efforts to unravel how thetranscriptome and genome of a brain-tumour cellinfluence its sensitivity to chemotherapy. Ideally, thegenome-wide analysis of gene expression and copynumbers in samples obtained by surgery will be combinedwith laser microdissection techniques to ascertain that theassessed profiles truly reflect the signature characteristics ofthe tumour-cell component. Such approaches will becomeincreasingly feasible with the development of microarraytechniques that depend on lower amounts of study mRNA and DNA. In addition, further insights into


10 �m

Figure 8. Histological section of a high-grade glioma from a child, imaged by differential contrastand analysed by immunocytochemistry for expression of DNA topoisomerase II� protein. The whitearrow points to a nucleus positive for expression of the protein. The black arrow points to amicrodissected area where a single positive nucleus has been captured by laser pressurecatapulting.



chemoresistance in brain tumours will come with acomprehensive analysis of the epigenetic regulation of geneexpression in these neoplasms. Aberrant methylation ofcytosine guanine dinucleotide (CpG) islands is apredominant epigenetic mechanism of gene inactivation.Microarrays that allow analysis of CpG island methylationthroughout the genome (epigenomics) have recently beendeveloped.

Genomics investigations cannot be considered theendpoint in research on brain-tumour drug resistance.Such strategies allow the formulation of hypotheses aboutthe relation between drug sensitivity and transcriptomicand genomic variation at the level of correlation ratherthan cause and effect. Ascertainment of biological functionrequires candidate gene validation via conventionalmolecular biological approaches. The challenge offunctional genomics will be to elucidate how validatedcandidates act and interact as components of complex generegulatory networks that determine drug sensitivity inbrain tumours. Better understanding will ultimatelysupport the search for refined therapeutic strategies thatimprove the response to cytotoxic agents and thus themostly poor outlook for patients with brain tumours.

Conflict of interestNone declared.

References1 Bredel M. Anticancer drug resistance in primary human brain

tumors. Brain Res Brain Res Rev 2001; 35: 161–204.2 Bredel M, Zentner J. Brain-tumour drug resistance: the bare

essentials. Lancet Oncol 2002; 3: 397–406.3 Goldie JH, Coldman AJ. A mathematic model for relating the

drug sensitivity of tumors to their spontaneous mutation rate.Cancer Treat Rep 1979; 63: 1727–33.

4 Esteller M, Garcia-Foncillas J, Andion E, et al. Inactivation of theDNA-repair gene MGMT and the clinical response of gliomas toalkylating agents. N Engl J Med 2000; 343: 1350–54.

5 Rieger L, Rieger J, Winter S, et al. Evidence for a constitutive,verapamil-sensitive, non-P-glycoprotein multidrug resistancephenotype in malignant glioma that is unaltered byradiochemotherapy in vivo. Acta Neuropathol (Berl) 2000; 99:555–62.

6 Schwechheimer K, Laufle RM, Schmahl W, et al. Expression ofneu/c-erbB-2 in human brain tumors. Hum Pathol 1994; 25:772–80.

7 Duan Z, Feller AJ, Penson RT, et al. Discovery of differentiallyexpressed genes associated with paclitaxel resistance using cDNA

array technology: analysis of interleukin (IL) 6, IL-8, andmonocyte chemotactic protein 1 in the paclitaxel-resistantphenotype. Clin Cancer Res 1999; 5: 3445–53.

8 Kudoh K, Ramanna M, Ravatn R, et al. Monitoring the expressionprofiles of doxorubicin-induced and doxorubicin-resistant cancercells by cDNA microarray. Cancer Res 2000; 60: 4161–66.

9 Watts GS, Futscher BW, Isett R, et al. cDNA microarray analysis ofmultidrug resistance: doxorubicin selection produces multipledefects in apoptosis signaling pathways. J Pharmacol Exp Ther2001; 299: 434–41.

10 Turton NJ, Judah DJ, Riley J, et al. Gene expression andamplification in breast carcinoma cells with intrinsic and acquireddoxorubicin resistance. Oncogene 2001; 20: 1300–06.

11 Robert J. Resistance to cytotoxic agents. Curr Opin Pharmacol2001; 1: 353–57.

12 Crowley-Weber CL, Payne CM, Gleason-Guzman M, et al.Development and molecular characterization of HCT-116 celllines resistant to the tumor promoter and multiple stress-inducer,deoxycholate. Carcinogenesis 2002; 23: 2063–80.

13 Dvorakova K, Payne CM, Tome ME, et al. Molecular and cellularcharacterization of imexon-resistant RPMI8226/I myeloma cells.Mol Cancer Ther 2002; 1: 185–95.

14 Hoshida Y, Moriyama M, Otsuka M, et al. Identification of genesassociated with sensitivity to 5-fluorouracil and cisplatin inhepatoma cells. J Gastroenterol 2002; 37 (suppl 14): 92–95.

15 Kihara C, Tsunoda T, Tanaka T, et al. Prediction of sensitivity ofesophageal tumors to adjuvant chemotherapy by cDNAmicroarray analysis of gene-expression profiles. Cancer Res 2001;61: 6474–79.

16 Sakamoto M, Kondo A, Kawasaki K, et al. Analysis of geneexpression profiles associated with cisplatin resistance in humanovarian cancer cell lines and tissues using cDNA microarray. Hum Cell 2001; 14: 305–15.

17 Sotiriou C, Powles TJ, Dowsett M, et al. Gene expression profilesderived from fine needle aspiration correlate with response tosystemic chemotherapy in breast cancer. Breast Cancer Res 2002; 4: R3.

18 Tracey L, Villuendas R, Ortiz P, et al. Identification of genesinvolved in resistance to interferon-alpha in cutaneous T-celllymphoma. Am J Pathol 2002; 161: 1825–37.

19 Weldon CB, Scandurro AB, Rolfe KW, et al. Identification ofmitogen-activated protein kinase kinase as a chemoresistantpathway in MCF-7 cells by using gene expression microarray.Surgery 2002; 132: 293–301.

20 Zembutsu H, Ohnishi Y, Tsunoda T, et al. Genome-wide cDNAmicroarray screening to correlate gene expression profiles withsensitivity of 85 human cancer xenografts to anticancer drugs.Cancer Res 2002; 62: 518–27.

21 Vikhanskaya F, Marchini S, Marabese M, et al. P73aoverexpression is associated with resistance to treatment withDNA-damaging agents in a human ovarian cancer cell line. Cancer Res 2001; 61: 935–38.

22 Komatani H, Kotani H, Hara Y, et al. Identification of breastcancer resistant protein/mitoxantrone resistance/placenta-specific,ATP-binding cassette transporter as a transporter of NB-506 andJ-107088, topoisomerase I inhibitors with an indolocarbazolestructure. Cancer Res 2001; 61: 2827–32.

23 Levenson VV, Davidovich IA, Roninson IB. Pleiotropic resistanceto DNA-interactive drugs is associated with increased expressionof genes involved in DNA replication, repair, and stress response.Cancer Res 2000; 60: 5027–30.

24 Dan S, Tsunoda T, Kitahara O, et al. An integrated database ofchemosensitivity to 55 anticancer drugs and gene expressionprofiles of 39 human cancer cell lines. Cancer Res 2002; 62:1139–47.

25 Wittig R, Nessling M, Will RD, et al. Candidate genes for cross-resistance against DNA-damaging drugs. Cancer Res 2002; 62:6698–705.

26 Bao L, Guo T, Sun Z. Mining functional relationships in featuresubspaces from gene expression profiles and drug activity profiles.FEBS Lett 2002; 516: 113–18.

27 Moriyama M, Hoshida Y, Otsuka M, et al. Relevance networkbetween chemosensitivity and transcriptome in human hepatomacells. Mol Cancer Ther 2003; 2: 199–205.

28 Wallqvist A, Rabow AA, Shoemaker et al. Establishing connectionsbetween microarray expression data and chemotherapeutic cancerpharmacology. Mol Cancer Ther 2002; 1: 311–20.

29 Scherf U, Ross DT, Waltham M, et al. A gene expression


Search strategy and selection criteriaData for this review were identified by searches of PubMed. Allpapers related to gene expression profiling, drug resistance,and laser microdissection in primary human brain tumoursdiscussed in the article were reviewed. A subset of articleschosen for their importance and the further readingopportunities they provide was included in the review. Searchterms included “array CGH”, “astrocytoma”, “brain tumo(u)r”,“cDNA array”, “chemoresistance”, “comparative genomichybridization”, “drug resistance”, “ependymoma”, “expressionarray”, “gene expression profiling”, “genomics”, “glioma”,“glioblastoma”, “laser microdissection”, “medulloblastoma”,“microarray”, “microdissection”, “multidrug resistance”,“oligodendroglioma”, “oligonucleotide array”, “PNET”, and“tissue microarray”.



database for the molecular pharmacology of cancer. Nat Genet2000; 24: 236–44.

30 Staunton JE, Slonim DK, Coller HA, et al. Chemosensitivityprediction by transcriptional profiling. Proc Natl Acad Sci USA2001; 98: 10787–92.

31 Butte AJ, Tamayo P, Slonim D, et al. Discovering functionalrelationships between RNA expression and chemotherapeuticsusceptibility using relevance networks. Proc Natl Acad Sci USA2000; 97: 12182–86.

32 Watson MA, Gutmann DH, Peterson K, et al. Molecularcharacterization of human meningiomas by gene expressionprofiling using high-density oligonucleotide microarrays. Am J Pathol 2002; 161: 665–72.

33 Watson MA, Perry A, Budhjara V, et al. Gene expression profilingwith oligonucleotide microarrays distinguishes World HealthOrganization grade of oligodendrogliomas. Cancer Res 2001; 61:1825–29.

34 Hunter S, Young A, Olson J, et al. Differential expression betweenpilocytic and anaplastic astrocytomas: identification ofapolipoprotein D as a marker for low-grade, non-infiltratingprimary CNS neoplasms. J Neuropathol Exp Neurol 2002; 61:275–81.

35 Kim S, Dougherty ER, Shmulevich L, et al. Identification ofcombination gene sets for glioma classification. Mol Cancer Ther2002; 1: 1229–36.

36 Fathallah-Shaykh HM, Rigen M, Zhao LJ, et al. Mathematicalmodeling of noise and discovery of genetic expression classes ingliomas. Oncogene 2002; 21: 7164–74.

37 Fuller GN, Hess KR, Rhee CH, et al. Molecular classification ofhuman diffuse gliomas by multidimensional scaling analysis ofgene expression profiles parallels morphology-based classification,correlates with survival, and reveals clinically-relevant novelglioma subsets. Brain Pathol 2002; 12: 108–16.

38 Gutmann DH, Hedrick NM, Li J, et al. Comparative geneexpression profile analysis of neurofibromatosis 1-associated andsporadic pilocytic astrocytomas. Cancer Res 2002; 62: 2085–91.

39 Ljubimova JY, Khazenzon NM, Chen Z, et al. Gene expressionabnormalities in human glial tumors identified by gene array. Int J Oncol 2001; 18: 287–95.

40 Qi ZY, Li Y, Ying K, et al. Isolation of novel differentiallyexpressed genes related to human glioma using cDNA microarrayand characterizations of two novel full-length genes. J Neurooncol2002; 56: 197–208.

41 Rickman DS, Bobek MP, Misek DE, et al. Distinctive molecularprofiles of high-grade and low-grade gliomas based onoligonucleotide microarray analysis. Cancer Res 2001; 61: 6885–91.

42 Sallinen SL, Sallinen PK, Haapasalo HK, et al. Identification ofdifferentially expressed genes in human gliomas by DNAmicroarray and tissue chip techniques. Cancer Res 2000; 60:6617–22.

43 Sehgal A, Boynton AL, Young RF, et al. Application of thedifferential hybridization of Atlas Human expression arraystechnique in the identification of differentially expressed genes inhuman glioblastoma multiforme tumor tissue. J Surg Oncol 1998;67: 234–41.

44 Huang H, Colella S, Kurrer M, et al. Gene expression profiling oflow-grade diffuse astrocytomas by cDNA arrays. Cancer Res 2000;60: 6868–74.

45 Markert JM, Fuller CM, Gillespie GY, et al. Differential geneexpression profiling in human brain tumors. Physiol Genomics2001; 5: 21–33.

46 Mischel PS, Shai R, Shi T, et al. Identification of molecularsubtypes of glioblastoma by gene expression profiling. Oncogene2003; 22: 2361–73.

47 Nutt CL, Mani DR, Betensky RA, et al. Gene expression-basedclassification of malignant gliomas correlates better with survivalthan histological classification. Cancer Res 2003; 63: 1602–07.

48 Korenberg MJ. Gene expression monitoring accurately predictsmedulloblastoma positive and negative clinical outcomes. FEBSLett 2003; 533: 110–14.

49 Pomeroy SL, Tamayo P, Gaasenbeek M, et al. Prediction of centralnervous system embryonal tumour outcome based on geneexpression. Nature 2002; 415: 436–42.

50 Tanwar MK, Gilbert MR, Holland EC. Gene expressionmicroarray analysis reveals YKL-40 to be a potential serum markerfor malignant character in human glioma. Cancer Res 2002; 62:4364–68.

51 Rhee CH, Ruan S, Chen S, et al. Characterization of cellular

pathways involved in glioblastoma response to thechemotherapeutic agent 1, 3-bis(2-chloroethyl)-1-nitrosourea(BCNU) by gene expression profiling. Oncol Rep 1999; 6: 393–401.

52 Bacolod MD, Johnson SP, Ali-Osman F, et al. Mechanisms ofresistance to 1,3-bis(2-chloroethyl)-1-nitrosourea in humanmedulloblastoma and rhabdomyosarcoma. Mol Cancer Ther 2002;1: 727–36.

53 Mukasa A, Ueki K, Matsumoto S, et al. Distinction in geneexpression profiles of oligodendrogliomas with and without allelicloss of 1p. Oncogene 2002; 21: 3961–68.

54 Eisen MB, Spellman PT, Brown PO, Botstein D. Cluster analysisand display of genome-wide expression patterns. Proc Natl AcadSci USA 1998; 95: 14863–68.

55 Tamayo P, Slonim D, Mesirov J, et al. Interpreting patterns ofgene expression with self-organizing maps: methods andapplication to hematopoietic differentiation. Proc Natl Acad SciUSA 1999; 96: 2907–12.

56 Hilsenbeck SG, Friedrichs WE, Schiff R, et al. Statistical analysis ofarray expression data as applied to the problem of tamoxifenresistance. J Natl Cancer Inst 1999; 91: 453–59.

57 Kim SK, Lund J, Kiraly M, et al. A gene expression map forCaenorhabditis elegans. Science 2001; 293: 2087–92.

58 Heyer LJ, Kruglyak S, Yooseph S. Exploring expression data:identification and analysis of coexpressed genes. Genome Res 1999;9: 1106–15.

59 Hastie T, Tibshirani R, Eisen MB, et al. ‘Gene shaving’ as a methodfor identifying distinct sets of genes with similar expressionpatterns. Genome Biol 2000; 1: RESEARCH0003.

60 Dahlquist KD, Salomonis N, Vranizan K, et al. GenMAPP, a newtool for viewing and analyzing microarray data on biologicalpathways. Nat Genet 2002; 31: 19–20.

61 Gygi SP, Rochon Y, Franza BR, Aebersold R. Correlation betweenprotein and mRNA abundance in yeast. Mol Cell Biol 1999; 19:1720–30.

62 Wang H, Zhang W, Fuller GN. Tissue microarrays: applications inneuropathology research, diagnosis, and education. Brain Pathol2002; 12: 95–107.

63 Pollack JR, Sorlie T, Perou CM, et al. Microarray analysis reveals amajor direct role of DNA copy number alteration in thetranscriptional program of human breast tumors. Proc Natl AcadSci USA 2002; 99: 12963–68.

64 Hess KR, Fuller GN, Rhee CH, Zhang W. Statistical patternanalysis of gene expression profiles for glioblastoma tissues andcell lines. Int J Mol Med 2001; 8: 183–88.

65 Sasaki T, Hankins GR, Helm GA. Comparison of gene expressionprofiles between frozen original meningiomas and primarycultures of the meningiomas by GeneChip. Neurosurgery 2003; 52:892–99.

66 Emmert-Buck MR, Bonner RF, Smith PD, et al. Laser capturemicrodissection. Science 1996; 274: 998–1001.

67 Schutze K, Lahr G. Identification of expressed genes by laser-mediated manipulation of single cells. Nat Biotechnol 1998; 16:737–42.

68 Fink L, Seeger W, Ermert L, et al. Real-time quantitative RT-PCRafter laser-assisted cell picking. Nat Med 1998; 4: 1329–33.

69 Bonaventure P, Guo H, Tian B, et al. Nuclei and subnuclei geneexpression profiling in mammalian brain. Brain Res 2002; 943:38–47.

70 Becker AJ, Wiestler OD, Blumcke I. Functional genomics inexperimental and human temporal lobe epilepsy: powerful newtools to identify molecular disease mechanisms of hippocampaldamage. Prog Brain Res 2002; 135: 161–73.

71 Mariani L, Beaudry C, McDonough WS, et al. Death-associatedprotein 3 (Dap-3) is overexpressed in invasive glioblastoma cellsin vivo and in glioma cell lines with induced motility phenotype invitro. Clin Cancer Res 2001; 7: 2480–89.

72 Luzzi V, Holtschlag V, Watson MA. Expression profiling ofductal carcinoma in situ by laser capture microdissection and

high-density oligonucleotide arrays. Am J Pathol 2001; 158:2005–10.

73 Kitahara O, Furukawa Y, Tanaka T, et al. Alterations of geneexpression during colorectal carcinogenesis revealed by cDNAmicroarrays after laser-capture microdissection of tumor tissuesand normal epithelia. Cancer Res 2001; 61: 3544–49.

74 Alevizos I, Mahadevappa M, Zhang X, et al. Oral cancer in vivo gene expression profiling assisted by laser capture




microdissection and microarray analysis. Oncogene 2001; 20:6196–204.

75 Leethanakul C, Patel V, Gillespie J, et al. Distinct pattern ofexpression of differentiation and growth-related genes insquamous cell carcinomas of the head and neck revealed by the

use of laser capture microdissection and cDNA arrays. Oncogene2000; 19: 3220–24.

76 Sgroi DC, Teng S, Robinson G, LeVangie R, Hudson JR, Jr.,Elkahloun AG. In vivo gene expression profile analysis of humanbreast cancer progression. Cancer Res 1999; 59: 5656–61.


Call for papers The Lancet Oncology would like to invite all potential authors to consider submitting articles to the Advances in research section of the journal.Further details on this section can be found in the Instructions to authors, and all enquiries and submissions should be sent to the Editor at:[email protected]

Advances in research

We consecutively imaged tumourproliferation and glucose utilisation ina L5178Y lymphoma-bearing mouse.This was done by the use of a highresolution, small-animal PET scanner,and the administration of [18F]FLT and[18F]FDG. Currently, [18F]FDG is themost commonly used tracer for cancerdetection with PET, and reflects theactivity of glucose transporters andhexokinase (panel a). By contrast,tumour detection by [18F]FLT isthought to depend on thymidine kinase 1 (TK1) activity—the keyenzyme of the DNA salvage pathway(panel b). In mice with bilateralvariants of L5178Y tumours (TK1 -/-variant implanted in lower left dorsalregion, TK1 +/- variant implanted inlower right dorsal region), the +/-tumours produced 48% more TK1

enzyme on average, and also grewfaster (27% shorter volume doublingtime), compared with -/- tumours.Uptake of [18F]FLT by PET was higher for the +/- tumour,compared with the -/- tumour. A converse pattern, however,was found in the [18F]FDG image. These findings support theconclusion that the imaging of DNA synthesis is more accu-rate for the assessment of the proliferative potential oftumours, compared with the imaging of glucoseconsumption.

This example highlights the unique possibility of modernPET techniques for non-invasive investigation of differenttumour properties in small laboratory animals with ultra-high spatial resolution (~1 mm). Such techniques should

encourage oncological research because they couldpotentially support drug development and also shorten thetime between translation of preclinical research into a clinicalsetting. With respect to the new tumour proliferation marker,[18F]FLT, our PET imaging experiments provide clarity onthe uptake mechanism of this radiotracer by tumours.Clinical testing of this promising new tumour proliferationmarker for PET is already underway and may provide a newmore sensitive, non-invasive approach for monitoringtreatment response and improving the accuracy of clinicalstaging.

Small-animal imaging of tumour proliferation with PET

Henryk Barthel, Pat Price, and Eric O Aboagye

TK1�/�

TK1�/�

TK1�/�

TK1�/�

KidneyKidney

Brain

A B

Correspondence: Dr Henryk Barthel, Molecular Therapy and PET Oncology Research Group, Faculty of Medicine, Imperial CollegeLondon, Hammersmith Hospital, Du Cane Road, London W12 0NN, UK. Tel: +49 (0)341 9718082. Fax: +49 (0)341 9718009. Email: [email protected]


doi:10.1016/S1470-2045(04)01382-8

Documents

tumourcell resistance

primary type of brain

tumour cells

tumour sensitivity

unravelling drug resistance

neuroectodermal tumour

tumour regions

primitive resistance