a transcriptional and metabolic signature of primary...

Integrated Systems and Technologies

A Transcriptional and Metabolic Signature of PrimaryAneuploidy Is Present in Chromosomally UnstableCancer Cells and Informs Clinical Prognosis

Jason M. Sheltzer

AbstractAneuploidy is invariably associated with poor proliferation of primary cells, but the specific contributions of

abnormal karyotypes to cancer, a disease characterized by aneuploidy and dysregulated proliferation, remainunclear. In this study, I demonstrate that the transcriptional alterations caused by aneuploidy in primary cells arealso present in chromosomally unstable cancer cell lines, but the same alterations are not common to allaneuploid cancers. Chromosomally unstable cancer lines and aneuploid primary cells also share an increase inglycolytic and TCA cycle flux. The biological response to aneuploidy is associated with cellular stress and slowproliferation, and a 70-gene signature derived from primary aneuploid cells was defined as a strong predictor ofincreased survival in several cancers. Inversely, a transcriptional signature derived from clonal aneuploidy intumors correlatedwith highmitotic activity and poor prognosis. Together, these findings suggested that there aretwo types of aneuploidy in cancer: one is clonal aneuploidy, which is selected during tumor evolution andassociated with robust growth, and the other is subclonal aneuploidy caused by chromosomal instability (CIN).Subclonal aneuploidy more closely resembles the stressed state of primary aneuploid cells, yet CIN is not benign;a subset of genes upregulated in high-CIN cancers predict aggressive disease in human patients in a proliferation-independent manner. Cancer Res; 73(21); 6401–12. �2013 AACR.

IntroductionAneuploidy and chromosomal instability (CIN) are inter-

related but not identical. Aneuploidy is a description of acellular state; it specifically describes a cell whose karyotypeis not a whole-number multiple of the haploid complement.CIN is most accurately characterized as a rate; it refers to acell that missegregates chromosomes more frequently thana wild-type cell does. CIN can be caused by mutations ordrug treatments that impair the various cellular processesrequired for accurate chromosome segregation (1). In certaincircumstances, aneuploidy itself may cause CIN (2–5). But acell can become aneuploid without exhibiting CIN, as eventssuch as chromosome nondisjunction, micronuclei forma-tion, and cytokinesis failure occur sporadically in normalcells.Aneuploidy and CIN are of particular interest due to their

paradoxical role in cancer development. Nearly all solid tumorsare aneuploid (6), and many types of cancers display elevated

levels of chromosome missegregation (7). However, aneuploidprimary cells generally exhibit poor proliferative capacity, instark contrast to the robust growth displayed by aneuploidcancer cells (8–12). Individuals with trisomy 21 (Down syn-drome) are at elevated risk for the development of childhoodleukemias, but have significantly lower levels of solid tumorformation throughout life (13). Furthermore, in mouse modelsof CIN, CINmice are typically tumor prone (14–16), but in somecontexts CIN seems to protect against cancer development(17–19). Thus, a complete understanding of the role of CIN andaneuploidy in cancer must account for both its apparentlypositive and negative roles in tumorigenesis.

Various strategies to detect aneuploidy and CIN, includingFISH, flow cytometry, and the direct analysis of mitotic figures,have been used in an attempt to associate a tumor's chromo-somal content with patient risk (7). Most studies have iden-tified a link between aneuploidy and/or CIN and poor clinicaloutcome (7), though in some contexts, excessive CIN mayactually promote survival (20). Moreover, interpretation ofsome assays may be complicated by the conflation of aneu-ploidy (a state), CIN (a rate), and other aspects of cancerbiology, including heightened proliferation (21). Nonetheless,the biological changes associated with aneuploidy and CINcould potentially be useful for stratifying patient risk.

We have recently analyzed gene expression data from pri-mary aneuploid cells in various species, including yeast, plants,mice, and humans (22). We reported that aneuploidy elicits atranscriptional response indicative of stress and slow growth,and this signaturewas conserved across eukaryotes. In cancers,

Author'sAffiliation:DavidH. Koch Institute for IntegrativeCancer Research,Massachusetts Institute of Technology, Cambridge, Massachusetts

Note: Supplementary data for this article are available at Cancer ResearchOnline (http://cancerres.aacrjournals.org/).

Corresponding Author: Jason M. Sheltzer, Massachusetts Institute ofTechnology, 76-543, 500Main Street, Cambridge, MA 02139. Phone: 617-253-3045; Fax: 617-258-6558; E-mail: [email protected]

doi: 10.1158/0008-5472.CAN-13-0749

�2013 American Association for Cancer Research.

CancerResearch

www.aacrjournals.org 6401

on September 8, 2020. © 2013 American Association for Cancer Research. cancerres.aacrjournals.org Downloaded from

Published OnlineFirst September 16, 2013; DOI: 10.1158/0008-5472.CAN-13-0749

http://cancerres.aacrjournals.org/

gene expression levels generally scale with chromosome copynumber (23), but the wider effects of aneuploidy and CIN oncancer transcriptomes are unknown. Here, I demonstrate thatchromosomally unstable cancer cell lines are transcriptionallyand metabolically similar to aneuploid primary cells. Althoughthe gene expression changes caused by primary aneuploidy areassociated with poor proliferation in vitro and improvedpatient survival in vivo, a subset of genes upregulated by CINare also associated with aggressive tumors and increasedpatient risk. I discuss the relevance of these findings for ourunderstanding of the role of aneuploidy in tumorigenesis.

Materials and MethodsData sources

Gene expression data from trisomic mouse embryonicfibroblasts (MEF) were acquired from ref. 9, and the set ofprobes classified as "expressed" were used in this study. Geneexpression data from the NCI60 cell lines analyzed on HG-U133A arrays were downloaded from CellMiner (24). Probeswere updated using Release 32 of the NetAffx probe annota-tions (25). Karyotype heterogeneity values were acquired fromref. 26, and missing values were kindly provided by A. Roschke(NCI, Bethesda, MD). Doubling times of the NCI60 panel wereacquired from ref. 27. Gene expression values from cancer/normal tissue pairs were acquired from the Gene ExpressionOmnibus (GEO; GSE15641, GSE27943, GSE3167, GSE3189, andGSE5364). Gene expression values from breast tumors ofknown ploidy were acquired fromGEO (GSE12071). Consump-tion/release rates of metabolites in the NCI60 panel wereacquired from ref. 28. Gene expression values from tetrasomicHCT116 cells were acquired from ref. 29. Gene expressionvalues from normal human tissues were acquired from GEO(GSE1133). Gene expression values from clinical cancer speci-mens were acquired from the sources cited in SupplementaryTables S18 and S20.

Mouse husbandry and MEF cultureAll animal studies and procedures were approved by the MIT

Institutional Animal Care and Use Committee. Cdc20AAA micewere kindly provided by Dr. Pumin Zhang (Baylor College ofMedicine, Houston, Texas). BubR1H mice were kindly providedby Dr. Jan van Deursen (Mayo Clinic). Embryos were derived viatimed matings between heterozygous animals. Embryos wereharvested ondayE12.5 for Cdc20AAA crosses andE13.5–E14.5 forBubR1H crosses. MEFs were isolated and cultured as describedin ref. 9.

MEF microarraysRNA was isolated from early-passage MEFs using TRizol

(Invitrogen) followed by purification with RNeasy Mini Kits(Qiagen). cDNAwas constructed and labeled using the OvationRNAAmplification SystemV2 and the FL-Ovation cDNABiotinModule V2 (NuGEN) according to the manufacturer's specifi-cations. cDNA was hybridized to Mouse 430A 2.0 Arrays(Affymetrix) for 16 hours and then scanned on an AffymetrixGeneChip scanner according to the manufacturer's specifica-tions. Data were summarized and normalized using gcRMAand have been deposited in GEO (GSE49894).

Data analysisAnalysis was performed in Excel, Python, and R. For each

microarray experiment, probesets were log2-transformed andthen normalized by subtracting the average expression level ofthat probeset. Probesets annotated to the same gene werecollapsed by averaging. For primary MEFs and tetrasomicHCT116 cells, aneuploid chromosomes were excluded fromanalysis. For most samples, genes annotated to sex chromo-somes were excluded from analysis, and, on the Affymetrixplatform, nonspecific probesets (those annotated with an "x")were excluded from analysis. The exception to this is the CIN70gene signature in Supplementary Fig. S4, in which some of thegenes are present on sex chromosomes or are onlymeasured bynonspecific probesets.

Gene Ontology (GO) term enrichment analysis was doneusing GProfiler with a Benjamini–Hochberg-corrected P valueof 0.05 and a maximum P value of 10�2 (30). Enrichments wereperformed against the relevant background gene set, that is,against all genes not present on an aneuploid or sex chromo-some. Mouse and human orthologs were identified usingGProfiler. Only genes that displayed a one-to-one orthologyrelationship betweenmouse andhumanwere used for analysis.Gene Set Enrichment Analysis (GSEA) was performed using aPearson ranking metric, a gene set size range of 5 to 5,000, and1,000 permutations of gene set labels (31). Transcription factormotif enrichments were identified using MSigDB (32). Meta-bolic pathway enrichments were identified with MetaboAna-lyst using a maximum P value of 10�2 (33). Gene sets wereclustered in Cluster 3.0 (34) and visualized in Java Treeview(35). For single-gene correlation analysis, a cutoff of 0.3 ormoreor �0.3 or less was uniformly applied. For two-sample com-parisons, a combined Student t test (P < 0.05) and mean foldchange (>1.3 or <�1.3) threshold was used to identify differ-entially expressed genes. As only the mean mRNA expressionvalues were reported for the HCT116þ5 cells, genes expressed�1 SD from the overall mean were considered differentiallyexpressed.

Analysis of the NCI60 panel of cancer cell linesData from57NCI60 cell lines were used for this analysis. Lung

cancer line H23 was excluded as it was not profiled on HG-U133A arrays, MDA-N was excluded as it lacked karyotypeinformation, andNCI-ADR-RESwas excluded as it is a derivativeofOVCAR-8.To identify transcriptswhoseexpression correlatedwith CIN, I used a strategy similar to that described in ref. 36.Levels of CIN were determined using the index of numericalheterogeneity (INH) of each cell line, as reported in ref. 26. INHvalues were derived from the cell-to-cell variation in the copynumber of each chromosome, as determined by spectral kar-yotype analysis of metaphase spreads. A chromosome whosecopy number varied in different cells was judged to be hetero-geneous in that cell line. (Chromosomes that seemed to be lostin two or fewer metaphase spreads or gained in a singlemetaphase spread were not counted as heterogeneous, due tothe possibility that these variationswere experimental artifacts.)Fractional INH values ranged from 0 (no chromosomes exhib-ited varying copy numbers) to 1 (every chromosome exhibitedvarying copy numbers), in multiples of 1/23, as the X and Y

Sheltzer

Cancer Res; 73(21) November 1, 2013 Cancer Research6402




chromosomeswere grouped together for this analysis. Finally, toexclude genes that showed tissue-of-origin–dependent expres-sion patterns, I performed a 9-way ANOVA in TM4 MeV (37). ABonferroni-corrected P value of 0.05 was used to exclude genesthat showed tissue-specific expression patterns.

Analysis of clinical gene expression dataGene expression and survival data were downloaded from

GEOand normalized as described above. Probeset annotationswere acquired from GEO. PCNA25, CIN70, and HET70 genescores were calculated by averaging the expression levels ofthese genes for each patient. TRI70 gene scores were calculatedby subtracting the average of the genes that are downregulatedby primary aneuploidy from the average of the genes that areupregulated by primary aneuploidy. For each patient cohort,samples were divided into high expression and low expressionclasses depending onwhether the gene set scores weremore orless than themean, respectively, for each gene set. P values andHRs were determined using the survival package in R or inPrism. Cox proportional hazard models were constructedusing the coxph function in R.

ResultsTranscriptional similarities between aneuploid primarycells and chromosomally unstable cancer cellsI first developed a transcriptional signature of primary

aneuploidy using gene expression data from 18 lines of MEFsthat were either euploid or trisomic for one of four chromo-somes (9). I sorted these cell lines according to their degree ofaneuploidy, as determined by the number of protein-codinggenes that were present on the trisomic chromosomes. I thencalculated the Spearman rank-order correlation coefficient(SCC) between each gene expression vector and the degreeof aneuploidy across the panel of MEFs. I extracted geneswhose expression tended to either increase or decrease accord-ing to the degree of aneuploidy (r > 0.3 or r < �0.3), thenidentified GO terms enriched among these gene sets. Amongupregulated genes (those that were positively correlated withaneuploidy), GO term analysis revealed an enrichment of genesannotated to the membrane (P < 10�12), the vesicle (P < 10�4),and the extracellular region (P < 10�3), among other processes(Supplementary Table S1). Downregulated genes wereenriched for those annotated to nucleic acid metabolism (P <10�17), DNA repair (P < 10�12), and RNA processing (P < 10�9).These GO terms are consistent with our hypothesis thatprimary aneuploidy is an antiproliferative cellular stress (22).Next, I analyzed gene expression data from the NCI60 panel

of cancer cell lines, which comprise 59 independent cell linesderived from 9 types of cancers. Importantly, several karyo-typic parameters have been determined for each of these lines,including their INH (26). The INH is a measure of cell-to-cellvariability in chromosome content: cancers that exhibit CINhave large INH values, whereas karyotypically stable cancershave INH values that are close to 0. Aneuploidies that arepresent in all cells do not contribute to INH (see Materials andMethods). I hypothesized that if aneuploidy is a cellular stress,then the continuous generation of new aneuploidies in celllines that displayed CIN could cause an analogous stress

response. This would be reflected in similar gene expressionpatterns betweenprimary aneuploid cells andCIN cancer lines.To test this, I calculated SCCs between gene expression vectorsin the NCI60 lines and their respective indexes of numericalheterogeneity (36). As INH values varied depending on thetissue of origin of the cell lines, I removed genes that displayedtissue-specific expression patterns via an ANOVA. Represen-tative genes are displayed in Fig. 1A and B: transcript levels ofthe DNA replication factorMCM6 tend to decrease in cell lineswith high CIN, whereas transcript levels of the ER-associatedgene CALU are correlated with increasing CIN. Surprisingly,genes correlated or anticorrelated with CIN were enriched formany of the same functional categories as observed in aneu-ploid primary cells (Fig. 1C and Supplementary Table S2).Nucleic acid metabolism (P < 10�34) and the chromosome(P < 10�13) were highly enriched among genes that decreasedwith increasing CIN, whereas themembrane (P < 10�6) and theextracellular matrix (P < 10�6) were positively correlated withCIN. In addition, gene expression (P < 10�21) and RNA splicing(P < 10�25) transcripts were strongly downregulated by CIN,but were more modestly affected by aneuploidy (P < 10�10 andP< 10�6, respectively). In total, 26%ofGO termsupregulated byaneuploidy in MEFs were also upregulated by CIN in cancercells, and 55% of GO terms downregulated by aneuploidy inMEFs were also downregulated by CIN in cancer cells. Amongthe 35 GO terms that were most strongly anticorrelated withprimary aneuploidy, all 35 were also anticorrelated with CIN.

To confirm that these GO term enrichments were not due tothe specific correlation cutoff used, I also analyzed the primaryaneuploidy and NCI60 datasets with GSEA, a threshold-inde-pendent method for correlating phenotypes with gene sets(31). GSEA revealed similar GO term enrichments as describedearlier, particularly among downregulated genes. DNA repli-cation, RNA splicing, and the cell cycle were significantlyanticorrelated with both increasing aneuploidy and increasingCIN (Supplementary Fig. S1 and Supplementary Tables S3 andS4). I conclude that aneuploidy in primary cells and CIN incancer cell lines affect the expression of genes involved in alimited and highly similar set of biologic functions.

I next sought to determine whether the transcriptionalsimilarities between aneuploid and CIN cell lines extended tothe level of individual genes. To test this, I identified one-to-oneorthologs between mouse and human genes (see Materials andMethods) and then compared the SCC values for each trans-cript. I found highly significant overlap among the sets oforthologous genes that were upregulated or downregulated byaneuploidy and CIN (Fig. 1D and E). For instance, 19% of allgeneswere anticorrelatedwith aneuploidy inMEFs at ar <�0.3cutoff. Among orthologous genes anticorrelated with CIN incancer cells, a significantly greater percentage (36%) were alsoanticorrelated with aneuploidy in MEFs (P < 10�16, hypergeo-metric test). In contrast, among genes that were positivelycorrelated with CIN, significantly fewer (10%) were anticorre-lated with aneuploidy (P < 10�3, hypergeometric test). Corre-spondingly, SCC values in bothMEFs and cancer cells displayeda small but highly significant genome-wide correlation (r¼ 0.14,P < 10�23). When the analysis was restricted to only those genesaffectedbyCIN incancer cells (r<�0.3 andr> 0.3), the strength

Transcriptional Consequences of Aneuploidy and CIN

www.aacrjournals.org Cancer Res; 73(21) November 1, 2013 6403




of the correlation increased to r ¼ 0.28 (P < 10�13). I concludethat aneuploid primary cells and chromosomally unstable can-cer cells share a high degree of transcriptional similarity at thelevel of both functional processes and individual genes.

Aneuploid primary cells and human tumors displaydistinct transcriptional programs

As the aneuploid transcriptional program was apparent inchromosomally unstable cancer cells, and as the vast majorityof human tumors are aneuploid, I next sought to determinewhether the transcriptomes of human tumors resemble thoseof primary aneuploid cells. I examined gene expression datafrom 9 common cancer types that were normalized to disease-free tissue, and I compared ranks between the mean foldchange expression levels in the tumors and the SCC valuesfrom trisomic MEFs (Supplementary Fig. S2). Interestingly, theaverage expression levels from tumors were anticorrelatedwith the aneuploidy responsiveness of genes in MEFs (averager¼�0.16, P < 10�29; Supplementary Fig. S2A). I then identifieddifferentially expressed genes between tumors and normaltissue using a combined Student t test (P < 0.05) and meanfold change (�1.3) threshold. Transcripts upregulated in aneu-ploid primary cells were significantly more likely to be down-regulated in tumors, whereas transcripts downregulated inaneuploid primary cells were likely to be upregulated in tumors(Supplementary Fig. S2B and S2C). Few GO terms were sig-nificantly enriched among the small number of genes thatweresimilarly affected by aneuploidy and cancer (Supplementary

Fig. S2D and S2E and Supplementary Table S5). Conversely, GOterms related to cell-cycle progression were strongly enrichedamong genes downregulated by aneuploidy but upregulated intumors (Supplementary Fig. S2E and Supplementary Table S6).To address the possibility that these results were due tocomparing tumors with untransformed tissue, I next analyzeda cohort of breast tumors that had either a diploid or aneuploidbasal ploidy (38). After normalizing gene expression in aneu-ploid tumors to the mean levels in diploid tumors, I found thatgene expression ranks in aneuploid tumors were also antic-orrelated with the expression levels from trisomic MEFs (r ¼�0.12, P < 10�13; Supplementary Fig. S2F). In addition, the GOcategories perturbed in aneuploid tumors were dissimilar tothose affected by aneuploidy in primary cells (SupplementaryFig. S2G and Table S7). In particular, transcripts upregulated inaneuploid tumors were highly enriched for those involved incell-cycle progression. Thus, unlike in primary cells, whereaneuploidy induces a chromosome- and species-independentgene expression response (22), aneuploid human tumors aretranscriptionally distinct from aneuploid primary cells.

The CIN transcriptional signature resembles a slowproliferation response

Whywouldprimaryaneuploid cells resemblehigh-CINcancercell lines but not all aneuploid cancers? We have proposed thatthe transcriptional similarities between aneuploid cells resultfrom their shared stress and/or slow-growth phenotype (22). Ihypothesized that a similar effect could drive the transcriptional

Figure 1. Transcriptional similarities between trisomicMEFs and chromosomally unstable cancer cell lines. A andB, representative examples of genes that arenegatively (MCM6) or positively (CALU) correlated with karyotype heterogeneity in the NCI60 panel. Each square represents the expression level of theindicated gene in one cell line, and a linear regression (gray line) is plotted against the data. C, gene ontology categories that are enriched among genesnegatively or positively correlatedwith aneuploidy inMEFs (black bars) or karyotype heterogeneity in theNCI60 panel (white bars). Complete GO term lists arelisted in Supplementary Tables S1 and S2. D, genes that are negatively correlated with aneuploidy in MEFs are significantly enriched for those thatare also negatively correlated with karyotype heterogeneity in the NCI60 panel. The black bar depicts the percentage of genes that are negatively correlatedwith aneuploidy in MEFs. White bars represent the percentage of genes negatively correlated with aneuploidy in MEFs that are also positively correlatedwith karyotypeheterogeneity in theNCI60panel. Hashedbars represent thepercentageof genesnegatively correlatedwith aneuploidy that are alsonegativelycorrelated with karyotype heterogeneity. The asterisks represent a degree of overlap that is greater than or less than the overlap expected by chance(P < 0.05; hypergeometric test). E, the same analysis as in D for genes positively correlated with aneuploidy in MEFs.

Sheltzer





program observed in karyotypically unstable cancer cells: thestresses associated with a continuously varying karyotype couldlead to the downregulation of cell-cycle genes and the upregula-tion of stress- and signaling-related genes. Consistent with thishypothesis, I found that the INH in the NCI60 panel wassignificantly correlated with increasing doubling times: cellswith high CIN typically divided more slowly than did cells withlow CIN (r ¼ 0.40, P < 0.003; Fig. 2A). This effect remained truewhencell linesderived fromhematopoietic cancerswereexclud-ed (r ¼ 0.36, P < 0.02). Consistent with a relationship betweenCIN and cell-cycle progression, genes downregulated by CINwere highly enriched for targets of the canonical cell-cycletranscription factor E2F1 as well as other E2F family members(P < 10�8; Supplementary Table S8). I recalculated SCC valuesaccording to how strongly a gene expression vector correlatedwith doubling time, and I observed that GO terms enrichedamong genes that were correlated or anticorrelated with dou-bling time were very similar to those enriched among CIN-correlated genes. Among both gene sets, membrane and signal-ing-related transcripts were upregulated and RNA processingand cell-cycle–related transcripts were downregulated (Fig. 2Band Supplementary Table S9). However, the transcriptional

responses to CIN and slow growth were not identical: I sub-tracted doubling time-SCC values from CIN-SCC values andanalyzed the sets of genes that weremore strongly correlated oranticorrelatedwithCIN thanwith proliferation (rCIN� rDT> 0.3or <�0.3). Genes annotated to the mitochondrion (P < 10�15),the mitochondrial matrix (P < 10�6), and cellular metabolicprocesses (P < 10�3) displayed a stronger correlation with CINthan with doubling time (Fig 2C; Supplementary Table S10). NoGO terms were significantly enriched among genes that weremore strongly correlated with doubling time than with CIN.These data are intriguing as primary aneuploid cells displayvarious metabolic alterations (9, 12, 16), and they suggest thatthe upregulation of certain metabolic processes may be agrowth-rate–independent consequence of CIN in cancer. I con-clude that, aside frommetabolic gene expression, the transcrip-tional programsobserved in chromosomally unstable and slowlydividing cancers are highly similar.

Metabolic similarities between aneuploid primary cellsand chromosomally unstable cancer cells

Aneuploid MEFs display various metabolic alterations: ingeneral, trisomic cells use more glutamine, and produce more

Figure 2. Transcriptional similaritiesbetween chromosomally unstablecancer cell lines and cancer celllines with slow doubling times.A, karyotype heterogeneity anddoubling time are significantlycorrelated in the NCI60 panel.A linear regression (gray line) isplotted against the data. B,gene ontology categoriesthat are enriched among genesnegatively or positively correlatedwith doubling time (black bars) orkaryotype heterogeneity (whitebars) in the NCI60 panel. CompleteGO term lists are listed inSupplementary Tables S2 and S9.C, Spearman rank correlations foreach gene expression vector andkaryotype heterogeneity (x-axis)or doubling time (y-axis) wereplotted. Genes that display astronger positive correlation withkaryotype heterogeneity aredisplayed in blue, whereas genesthat display a stronger positivecorrelation with doubling time aredisplayed in red.






lactate, glutamate, and ammonium than euploid cells do. Inaddition, some aneuploid cells use more glucose (9, 16). Dochromosomally unstable cancer cells exhibit the same meta-bolic patterns? To address this question, I calculated SCCvalues between numerical heterogeneity and the consump-tion/release rates of 140 different metabolites from a recentlypublished analysis of the NCI60 panel (28). High-CIN cellstended to use more glutamine and glucose, and produce morelactate and glutamate, than low-CIN cells did (Fig. 3A; ammo-nium was not measured). Indeed, the released metabolite thatwas most strongly correlated with increasing CIN was gluta-mate, whereas the second strongest correlation among con-sumedmetabolites waswith glutamine (Fig. 3B).More broadly,metabolites whose production tended to increase with CINwere associated with the citric acid cycle (e.g., fumarate andmalate; P < 10�3) and glycolysis (e.g., pyruvate and 2-phos-phoglycerate; P < 10�3; Supplementary Table S11). Consistentwith the microarray data (Fig. 2C), some of these metaboliteswere also correlated with doubling time, though generally less

strongly than they were correlated with CIN (Fig. 3B). More-over, the citric acid cycle and glycolysis were not enrichedamong metabolites that were correlated with doubling time(Supplementary Table S12). Thus, in addition to their similar-ities at the transcriptional level, aneuploid primary cells andchromosomally unstable cancers share several metabolicalterations. The cause of these alterations is not currentlyclear (see Discussion).

The CIN transcriptional signature is apparent in spindlecheckpoint–defective MEFs and after chromosometransfer into karyotypically stable cancer cells

Thus far, I have established a correlative relationshipbetween the metabolic and transcriptional programs of aneu-ploid primary cells and chromosomally unstable cancer lines.However, it remains possible that other aspects of the cancerlines cause the similarity with aneuploid primary cells, and CINis an unrelated covariant. To demonstrate a direct link betweenchromosome missegregation and the CIN transcriptional

Figure 3. Metabolic similaritiesbetween trisomic MEFs andchromosomally unstable cancercell lines. A, glucose and glutamineconsumption tend to increase, andglutamate and lactate productiontend to increase, with karyotypeheterogeneity in the NCI60 panel.Gray lines represent linearregressions plotted against thedata. B, comparison betweenthe metabolic profiles ofchromosomally unstable and slowgrowing cancer lines. Metaboliteswhose consumption or productionincreases with doubling time(white bars) tend to show similar butsmaller trends than when thosemetabolites are correlated withkaryotype heterogeneity (blackbars; P < 0.01 for both producedand consumed metabolites,paired t test).

Sheltzer





signature, I examined the transcriptomes in chromosomallyunstable primary cells and in chromosomally stable cancer celllines in which aneuploidy was ectopically induced.CIN can be genetically encoded by introducing mutations

that affect the function of the Spindle Assembly Checkpoint(SAC). Hypomorphic alleles of two crucial SAC components,Cdc20 and BubR1, have been constructed, and MEFs contain-ing these mutations exhibit significant CIN (39, 40). I derivedMEFs from embryos that were homozygous for either hypo-morphic allele (Cdc20AAA/AAA or BubR1H/H) as well as fromtheir wild-type littermates (Cdc20þ/þ and BubR1þ/þ). I thenperformed microarray analysis on two early-passage BubR1sibling pairs and one early-passage Cdc20 sibling pair, and ineach case normalized gene expression in the mutant MEF lineto its wild-type littermate. Transcript levels were similarbetween the two CIN alleles (data not shown) and, therefore,the following comparisons were made using the averageexpression levels in the CIN MEFs. As expected, gene expres-sion levels in the chromosomally unstable MEFs were mod-erately but significantly correlated with SCC-CIN values fromthe NCI60 panel and with expression levels in the trisomicMEFs (r ¼ 0.15, P < 10�26, and r ¼ 0.26, P < 10�106, respec-tively). Genes upregulated in CIN MEFs tended to be upregu-lated in high-CIN cancer cells and in trisomic MEFs, and viceversa (Supplementary Fig. S3A–S3D). GO term enrichmentanalysis revealed the upregulation of extracellular and cellmotility genes, and the downregulation of cell-cycle tran-scripts, as is also observed in the trisomic MEFs and NCI60panel (Supplementary Table S13 and Supplementary Fig. S3E).Similar results were obtained when either Cdc20AAA/AAA orBubR1H/H MEFs were analyzed independently (data notshown). I conclude that the transcriptional patterns presentin primary cells that exhibit CIN are highly similar to bothprimary aneuploid cells and cancer cells that are chromosom-ally unstable.To further assess the link between the CIN transcriptional

signature and chromosome missegregation, I examined theeffects of ectopic aneuploidy in cancer cells. A recent reporthas analyzed the consequences of transferring additionalchromosomes into HCT116 cells, a karyotypically stable(INH ¼ 0) colon cancer line (29). I acquired microarraydata from HCT116 cells that had two extra copies of chro-mosome 5 introduced via microcell-mediated chromosometransfer (MMCT). I then repeated the above analysis com-paring gene expression in the newly tetrasomic cancer cellswith the SCC-CIN values from the NCI60 panel and with thetrisomic MEFs. The transcriptional changes induced byaneuploidy in the HCT116 cells were moderately similar tothose correlated with CIN across the NCI60 panel and withaneuploidy in primary MEFs (r ¼ 0.26, P < 10�47, and r ¼0.22, P < 10�43, respectively; Supplementary Fig. S3F–S3I).The induction of aneuploidy led to the downregulation ofcell-cycle transcripts and the upregulation of stress andmembrane-related transcripts (Supplementary Table S14and Supplementary Fig. S3E). Thus, inducing aneuploidy inan otherwise stable cancer cell line mimics the CIN tran-scriptional signature. Among all four microarray datasetsexamined, there was significant overlap among GO terms,

particularly among downregulated transcripts and thoseassociated with the extracellular matrix. However, tran-scripts involved in RNA processing were enriched amongdownregulated genes in the NCI60 panel and in aneuploidMEFs, but not in CIN MEFs or HCT116þ5 cells (Supple-mentary Fig. S3E). The cause of the minor transcriptionaldifferences that are apparent between these cell types is atpresent unknown.

A previous transcriptional signature of CIN isanticorrelated with karyotype heterogeneity anddoubling time

Carter and colleagues (41) have described a transcriptionalsignature for chromosomally unstable cancers that has beenwidely used as a marker for genomic instability (20, 42–44).They determined the degree of aneuploidy across severalcancer datasets by summing the total number of chromosomalregions that showed consistent alterations in expression levels.Genes were then ranked according to how strongly theycorrelated with the total aneuploidy in each tumor, and thetop 70 genes were used to create the CIN70 gene signature.Finally, they showed that in various cancer types, a high CIN70score was associated with poor clinical prognosis. It is impor-tant to note that their expression signature most likely detectsclonal aneuploidy, i.e., aneuploidy that is present in the bulk ofa tumor, as rare aneuploidies would not cause detectabledifferences in gene expression levels in cis. Furthermore, agenome with many chromosomal aneuploidies is not neces-sarily indicative of a current state of CIN: genetic alterationsmay accumulate at a low rate over a long period of time, withevolutionary pressure selecting the spread of cells with themost growth-advantageous mutations (45). As my analysis ofCIN suggested that it was associated with poor proliferation, Idecided to further explore the relationship between the CIN70gene set and cancer.

Surprisingly, I found that the CIN70 score was anticorre-lated with karyotype heterogeneity in the NCI60 panel (r ¼�0.35, P < 0.01; Supplementary Fig. S4A). Chromosomallystable cancer cell lines (HCT116 and CCRF-CEM) generallyhad higher CIN70 scores than unstable cell lines (HOP-92and TK-10). Upon inspection of the CIN70 gene list, I notedthat many transcripts were well-characterized markers ofproliferation (e.g., PCNA, CDK1, and MCM2). It has recentlybeen suggested that many prognostic gene signatures,including CIN70, function in part by capturing informationabout cell proliferation rates (46). I hypothesized that clonalaneuploidy in a tumor was not a strong indicator of CIN, butinstead reflected a highly evolved cancer state, in which atumor had acquired numerous growth-promoting geneticalterations. Indeed, the CIN70 score was strongly correlatedwith decreased doubling times in the NCI60 panel (r ¼�0.46, P < 0.001; Supplementary Fig. S4B). To further test thelink between CIN70 and proliferation rates, I examined amicroarray dataset of 81 nondiseased human tissues and cellpopulations. As a proxy for their proliferative index, I sortedthe tissues by their level of proliferating cell nuclear antigen(PCNA) expression. While 30% of all genes were positivelycorrelated with PCNA in this dataset, I found that 100% of






CIN70 genes were positively correlated with PCNA (averager ¼ 0.45, P < 10�7; Supplementary Fig. S4C). Similar correla-tions were observed between CIN70 and other commonproliferative markers (e.g., rMKI67 ¼ 0.57, P <10�13, rTOP2A¼ 0.50, P < 10�9). It has been argued that CIN700s prognosticutility is not derived from its ability to detect proliferationrates, as all cell-cycle–regulated genes can be removed fromCIN70 without abolishing its power to stratify tumors (41).However, the remaining genes (referred to as the CIN27wpsignature; ref. 42) were still highly correlated with PCNA(average r¼ 0.40, P < 10�6), demonstrating that this methodis insufficient to control for the widespread effects of pro-liferation on gene expression. I conclude that the CIN70score is not an accurate predictor of CIN in the NCI60 panel,but instead captures information about cell proliferationrates in both the NCI60 lines and chromosomally stablenormal tissue.

Construction of gene signatures associated withaneuploidy and CIN

As aneuploidy and CIN have been hypothesized to beintegral drivers of cancer evolution, I sought to determinewhether the aneuploidy and CIN-associated gene expressionpatterns identified herein could serve as useful in vivomarkersfor cancer progression. Because our analysis of the CIN70 geneset suggested that it largely reflected proliferative capacity, Ialso sought to clarify whether the aneuploidy-induced tran-scriptional programs were identical to or distinct from thetranscriptional response to slow growth.

To explore the prognostic relevance of the gene expressionchanges associated with aneuploidy and CIN, I constructedthree additional univariate gene signatures that are analogousto CIN70. First, I defined PCNA25, which consists of the 25genes that displayed the strongest correlation with PCNAexpression in normal human tissue, and which I used as anonspecific marker of cell proliferation (Supplementary TableS15; ref. 46). Second, I defined TRI70, which consists of the 70genes that displayed the strongest absolute correlation withaneuploidy in trisomic MEFs (Supplementary Table S16).Third, I defined HET70, which consists of the 70 genes thatdisplayed the strongest correlation with karyotype heteroge-neity in the NCI60 panel (Supplementary Table S17). I hypoth-esized that the most prominent gene expression changesassociated with clonal aneuploidy in cancer and primaryaneuploidy in MEFs were related to proliferation, and thattherefore CIN70 could be used to identify rapidly dividing cellswhereas TRI70 would identify slowly dividing cells. Indeed,CIN70 and TRI70 were able to sort human tissues by their levelof PCNA expression, demonstrating that in the absence ofaneuploidy or CIN, these gene signatures reflect cell prolifer-ation rates (r ¼ 0.61, P < 10�15, and r ¼ �0.29, P < 10�3,respectively; Supplementary Fig. S5A and S5B). Interestingly,theHET70 scorewas neither correlatednor anticorrelatedwithPCNA in human tissue (r ¼ 0.11, P > 0.05; Supplementary Fig.S5C and S5D). Although the overall gene expression pattern ofcells with heterogeneous karyotypes resembled that of a slowgrowth response (Fig. 2), the 70 genes most associated withkaryotype heterogeneity were not enriched for proliferation

markers and only 30% of the transcripts were individuallycorrelated with PCNA. I conclude that, unlike the CIN70 andTRI70 gene sets, HET70 scores are uncorrelated with prolifer-ative index.

Prognostic relevance of aneuploidy and karyotypeheterogeneity gene sets in cancer

I next assembled 27 cancer gene expression datasets frompublished clinical cohorts. Datasets were chosen to repre-sent a variety of solid tumor types, microarray designs, andoutcome measurements (i.e., overall survival, recurrence-free survival, etc.). The PCNA25, CIN70, TRI70, and HET70gene sets were used to stratify patients into above mean andbelow mean groups in each clinical cohort, and Kaplan–Meier survival curves were calculated for each gene set (Fig.4; Supplementary Table S18 and Supplementary Fig. S6).I found that high PCNA25 was significantly associated withpoor survival in 15 of 27 cohorts. PCNA25 was particularlyinformative in breast cancers (significant in six of sixcohorts), brain cancers (significant in four of four cohorts),and bladder cancers (significant in two of three cohorts).PCNA25 did not provide significant patient stratification inlung cancer (one of five cohorts), ovarian cancer (zero ofthree cohorts), or colorectal cancer (zero of three cohorts).These results are consistent with the known value of pro-liferation markers in breast cancer (47, 48) and suggest thatsimilar gene sets may be useful for other cancer types as well.In cancer types in which PCNA25 was not predictive, otherfacets of tumor biology, including immune evasion andmetastatic potential, may be of greater importance forpatient survival than proliferation rates.

Subsequently, I found that CIN70 provided significant strat-ification of patient risk in 15 of 27 cohorts (Supplementary Fig.S6). Importantly, every cohort in which CIN70 was a significantpredictor of survival was one in which PCNA25 was predictiveas well. Across all datasets, the HRs generated by CIN70 andPCNA25 were highly correlated (r ¼ 0.98, P < 10�18; Supple-mentary Fig. S7A). Despite having only 12 genes in common,the individual patient scores generated by CIN70 and PCNA25were nearly identical (r ¼ 0.97–0.99; Supplementary Fig. S7). Iconclude that CIN70 functions primarily by reporting prolif-erative capacity, and that, in general, CIN70 does not provideany information beyond that which is provided by a smallernumber of PCNA-associated markers.

TRI70was found to be a significant predictor of survival in 12of 27 patient cohorts. Unlike PCNA25 and CIN70, high TRI70was associated with favorable outcomes and low TRI70 wasassociated with disease progression and death. All cohorts inwhich TRI70 was predictive were also stratified by PCNA25,and HRs calculated using the two gene signatures were highlycorrelated (r ¼ �0.90, P < 10�10; Supplementary Fig. S7B).Thus, consistent with our above analysis, the most prominentgene expression changes caused by aneuploidy in primary cellsare directly related to slow proliferation, which can predictsurvival in various cancer types.

High HET70 scores were significantly associated with poorprognosis in 12 of 27 patient cohorts. Surprisingly, five of thesecohorts were not stratified by PCNA25: HET70 was able to

Sheltzer





stratify colorectal cancer (three of three patient cohorts),which PCNA25 failed to do. Across all datasets, the HRsgenerated by HET70 and PCNA25 were uncorrelated (Supple-mentary Fig. S7C), and individual patient scores generated byHET70 and PCNA25 were poorly correlated or uncorrelated(Supplementary Fig. S7D–S7F). I conclude that the geneexpression changes associated with CIN in vitro can serve asproliferation-independent markers for poor prognosis in vivo.Though I cannot ascertain at this time whether HET70 scoresstratify patients according to the CIN of their tumors, Ihypothesize that CIN is onemechanismbywhich the increasedexpression of these genes can arise.

HET5: a five-gene signature with prognostic relevance inmultiple cancer typesTo narrow the HET70 gene signature into a smaller set of

genes that may have clinical relevance, I calculated univariateCox proportional hazard scores for each HET70 gene in eachcancer dataset. I then sorted the HET70 genes by their averageZ score across the bladder, brain, colon, lung, and ovariancancer datasets. The five genes with the highest average Zscores were LGALS1, LEPRE1, FN1, CALU, and PLOD2, which Idenote as the HET5 gene set. Among the cohorts examinedabove, these five genes were able to stratify patients into low-risk and high-risk subgroups in two of three bladder cancercohorts, four of four brain cancer cohorts, three of threecolorectal cancer cohorts, one of one liposarcoma cohort, oneof one liver cancer cohort, three offive lung cancer cohorts, andone of three ovarian cancer cohorts (Supplementary TableS19). I then assembled a second set of published clinicalexpression arrays that were not used in the initial derivationof the HET5 signature. Within this test set, HET5 providedsignificant stratification in eight of 15 cohorts, including one of

one bladder cancer cohort, two of two brain cancer cohorts,and twoof four lung cancer cohorts (Supplementary Fig. S8 andSupplementary Table S20). Remarkably, HET5 was also able tostratify patient risk in three of four melanoma cohorts, despitethe fact that nomelanomadatasets were present in the originaltraining sets. Thus, the HET5 genes in particular may warrantfurther investigation as proliferation-independent markers ofcancer progression and patient survival.

DiscussionMany cancers display CIN, defined as an increased rate of

chromosome missegregation (45). Our results suggest thataneuploidies resulting from chromosome segregation errorsimpose afitness cost on cancer, in vitro if not also in vivo (Figs. 1and 2; ref. 22). CIN cancer cell lines divide more slowly thanchromosomally stable lines, downregulate transcripts associ-ated with progression through the cell cycle, and upregulatestress- and signaling-related genes (Fig. 2A; SupplementaryTable S2; ref. 36). A similar transcriptional response is evidentin CIN primary cells and in a stable cancer cell line that hasbeen induced to gain extra chromosomes (Supplementary Fig.S3). In addition, CIN cancers seem to have increasedmetabolicrequirements: they consume more glutamine and glucose percell, and producemore citric acid cycle byproducts, than stablecancer cell lines (Fig. 3). Thus, in these respects, chromosom-ally unstable cancer cells resemble the stressed state of primaryaneuploid cells.

Yet, these stresses are not apparent in all aneuploid cancers.When normalized to diploid tumors, aneuploid tumors displaytranscriptional patterns that are opposite to those observed inuntransformed aneuploid cells. Furthermore, increasing levelsof clonal aneuploidy within specific tumor types are tightly

Figure 4. Stratification of patient risk based on proliferation and aneuploidy gene signatures. Patients from each clinical cohort were divided into below-meanand above-mean subsets for each gene signature, and significant differences in survival time were identified using a log-rank test. Representative Kaplan–Meier curves for GSE14520 and GSE8894 are displayed. In general, PCNA25, CIN70, and TRI70 were able to classify the same patient cohorts,whereas HET70 classified some cohorts in which proliferation-related markers were not prognostic. The x-axes indicate survival time in months.Complete results are presented in Supplementary Table S18 and Supplementary Figure S6.






correlated with markers of rapid proliferation (SupplementaryFig. S5). To reconcile these results, I hypothesize that there aretwo "types" of aneuploidies that are common to cancer cells:clonal aneuploidy, which is present in the bulk of a tumor andarises due to the selective advantages that it confers, andspontaneous aneuploidy, which results from chromosomemissegregation and generally induces fitness costs on acell.

The CIN70 gene signature was initially constructed byidentifying genes that correlated most strongly with clonalaneuploidy in cancer: tumors with high CIN70 scores were themost structurally complex, whereas tumors with low CIN70scores displayed the fewest alterations from the diploid state(41). However, our findings demonstrate that CIN70 is amarker of cell proliferation, rather than CIN: 100% of CIN70genes are significantly correlated with PCNA expression innormal tissue, and CIN70 scores accurately differentiatebetween rapidly dividing and slowly dividing karyotypicallynormal human tissue (Supplementary Fig. S5). These resultscan be explained by hypothesizing that structural complexityin tumors results from an ongoing evolutionary process.Genetic alterations, including mutations, deletions, and chro-mosomal duplications, arise during a cancer's growth. Thosechanges that induce a proliferative advantage are selectedwithin the tumor and increase to clonal levels. Althoughaneuploidy is generally associated with a fitness cost, thefew chromosomal alterations that do confer proliferativeadvantages, such as the gain of a chromosome arm carryingan oncogene, are naturally selected and become the dominanttumor cell population over time. Thus, tumors with moreaneuploid regions are likely to have acquired more growth-promoting genetic alterations, explaining the tight linkbetween CIN70 and cell proliferation.

In contrast, aneuploidies resulting from random segregationerrors have not been directly selected for. Whole-chromosomeaneuploidy alters the copy numbers of hundreds or thousandsof genes after a single mitotic event. Human cells lack a globaldosage compensation system for autosomal aneuploidy; mostcopy number variation has proportional effects on the levels ofmRNA's and proteins encoded on the aneuploid chromosome(29). Thus, it has been hypothesized that aneuploidy induces afitness cost bywasting energy on the transcription, translation,and degradation of unnecessary proteins and by interferingwith the formation or function of stoichiometry-sensitiveprotein complexes (49). Consistent with the results presentedhere, we have shown that both primary aneuploid cells andchromosomally unstable (but not chromosomally stable) can-cers are sensitive to energy stress- and proteotoxic stress-inducing compounds (12). Note that I do not intend to suggestthat CIN is always antagonistic to robust proliferation. Somemutations, such as loss of p53, can increase both proliferationand CIN (50). As p53 loss is a late event in many cancers (51–53), aggressive stage III or stage IV tumors may display bothCIN and heightened mitotic activity. Nonetheless, the overalltranscription patterns observed in CIN cells, or stable cells thathave been induced to gain chromosomes, are consistent withunselected aneuploidy, placing a substantial burden on cellularhomeostasis.

The stressed state of CIN cells may have further relevancefor its role in tumor development and evolution. CINhas been linked with drug resistance and cancer relapse(54, 55). The ability of CIN cells to generate genomicallyheterogeneous progeny can likely contribute to a tumor'sability to evolve and metastasize despite treatment. How-ever, as genotoxic chemotherapy selectively kills rapidlydividing cells, it may also be the case that CIN cells escapeelimination due to their inherently stressed state and slowgrowth. The physiologic differences between chromosom-ally stable and unstable cells could be exploited to developdrugs that selectively kill high-CIN cells, which may beadded to existing chemotherapy regimens to minimizerelapse (12).

More broadly, there is growing interest in the use ofgenomic technology to stratify patient risk and identifyappropriate treatments for cancer (56). Although manypublished gene signatures have demonstrated the abilityto identify aggressive tumors, further analysis has estab-lished that some of these signatures are prognostic only dueto their ability to detect proliferation rates (46). Indeed, Ihave confirmed here that proliferation, measured byPCNA25, is a significant but not universal risk factor indiverse cancers, and that one previously published signa-ture (CIN70) is tightly coupled to proliferative index. How-ever, I demonstrate that HET70, the 70 genes most stronglyupregulated in karyotypically heterogeneous cancers, func-tion as a proliferation-independent risk factor in severaldifferent patient cohorts. How strongly these genes corre-late with CIN in vivo is at present unknown, and I suspectthat CIN is one of many factors that can drive theirexpression. Furthermore, the HET70 signature can be nar-rowed to five genes that maintain significant prognosticutility in multiple cancer types. The HET5 gene set includesFN1, encoding fibronectin, and LGALS1, encoding Galectin-1, which have extensive roles in cancer growth, metastasis,and immune evasion (57–60). Three other genes that com-prise HET5 (CALU, PLOD2, and LEPRE1) have functions thatare less well understood. Further analysis of these genesmay shed light on the clinical progression of diverse can-cers. Moreover, these findings provide a mechanistic linkbetween CIN and aggressive disease and suggest genes andbiological processes that could be targeted to contraveneCIN-induced cancer progression.

Disclosure of Potential Conflicts of InterestNo potential conflicts of interest were disclosed.

AcknowledgmentsThe author thanks Joan Clara Smith for coding assistance, Pumin Zhang and

Jan van Deursen for providing mice, and Erica Chung for help with MEF culture.The author also thanks Angelika Amon, Maitreya Dunham, Mike Laub, AvivRegev, Yitzhak Pilpel, and members of the Amon Laboratory for comments onthe article.

Grant SupportThis work was supported by NIH grant GM056800 (Angelika Amon). Angelika

Amon is an investigator of the Howard Hughes Medical Institute, whose fundingalso supported this research.

The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby marked

Sheltzer





advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate thisfact.

ReceivedMarch 12, 2013; revised August 19, 2013; accepted September 5, 2013;published OnlineFirst September 16, 2013.

References1. Holland AJ, Cleveland DW. Boveri revisited: chromosomal instabil-

ity, aneuploidy and tumorigenesis. Nat Rev Mol Cell Biol 2009;10:478–87.

2. Sheltzer JM, Blank HM, Pfau SJ, Tange Y, George BM, Humpton TJ,et al. Aneuploidy drives genomic instability in yeast. Science 2011;333:1026–30.

3. Thompson SL, Compton DA. Proliferation of aneuploid human cells islimited by a p53-dependent mechanism. J Cell Biol 2010;188:369–81.

4. Duesberg P, Rausch C, Rasnick D, Hehlmann R. Genetic instability ofcancer cells is proportional to their degree of aneuploidy. Proc NatlAcad Sci U S A 1998;95:13692–7.

5. Zhu J, Pavelka N, Bradford WD, Rancati G, Li R. Karyotypic determi-nants of chromosome instability in aneuploid budding yeast. PLoSGenet 2012;8:e1002719.

6. Weaver BA, Cleveland DW. Does aneuploidy cause cancer? Curr OpinCell Biol 2006;18:658–67.

7. McGranahan N, Burrell RA, Endesfelder D, Novelli MR, Swanton C.Cancer chromosomal instability: therapeutic and diagnostic chal-lenges. EMBO Rep 2012;13:528–38.

8. Torres EM, Sokolsky T, Tucker CM, Chan LY, Boselli M, Dunham MJ,et al. Effects of aneuploidy on cellular physiology and cell division inhaploid yeast. Science 2007;317:916–24.

9. Williams BR, Prabhu VR, Hunter KE, Glazier CM, Whittaker CA, Hous-man DE, et al. Aneuploidy affects proliferation and spontaneousimmortalization in mammalian cells. Science 2008;322:703–9.

10. Thompson SL, Compton DA. Examining the link between chromo-somal instability and aneuploidy in human cells. J Cell Biol 2008;180:665–72.

11. Pavelka N, Rancati G, Zhu J, Bradford WD, Saraf A, Florens L, et al.Aneuploidy confers quantitative proteome changes and phenotypicvariation in budding yeast. Nature 2010;468:321–5.

12. Tang Y-C, Williams BR, Siegel JJ, Amon A. Identification of aneuploi-dy-selective antiproliferation compounds. Cell 2011;144:499–512.

13. Ni�zeti�c D, Groet J. Tumorigenesis in Down's syndrome: big lessonsfrom a small chromosome. Nat Rev Cancer 2012;12:721–32.

14. Sotillo R, Hernando E, Díaz-Rodríguez E, Teruya-Feldstein J, Cord�on-Cardo C, Lowe SW, et al. Mad2 overexpression promotes aneuploidyand tumorigenesis in mice. Cancer Cell 2007;11:9–23.

15. Sotillo R, Schvartzman J-M, Socci ND, Benezra R. Mad2-inducedchromosome instability leads to lung tumour relapse after oncogenewithdrawal. Nature 2010;464:436–40.

16. Li M, Fang X, Baker DJ, Guo L, Gao X, Wei Z, et al. The ATM-p53pathway suppresses aneuploidy-induced tumorigenesis. Proc NatlAcad Sci U S A 2010;107:14188–93.

17. Weaver BAA, Silk AD, Montagna C, Verdier-Pinard P, Cleveland DW.Aneuploidy acts both oncogenically and as a tumor suppressor.Cancer Cell 2007;11:25–36.

18. Baker DJ, Jin F, Jeganathan KB, vanDeursen JM.Whole chromosomeinstability caused by Bub1 insufficiency drives tumorigenesis throughtumor suppressor gene loss of heterozygosity. Cancer Cell2009;16:475–86.

19. Remeseiro S, Cuadrado A, CarreteroM,Martínez P, DrosopoulosWC,Ca~namero M, et al. Cohesin-SA1 deficiency drives aneuploidy andtumourigenesis in mice due to impaired replication of telomeres.EMBO J 2012;31:2076–89.

20. Birkbak NJ, Eklund AC, Li Q, McClelland SE, Endesfelder D, Tan P,et al. Paradoxical relationship between chromosomal instability andsurvival outcome in cancer. Cancer Res 2011;71:3447–52.

21. Zasadil LM, Britigan EMC, Weaver BA. 2n or not 2n: aneuploidy,polyploidy, and chromosomal instability in primary and tumor cells.Semin Cell Dev Biol 2013;24:370–9.

22. Sheltzer JM, Torres EM, Dunham MJ, Amon A. Transcriptionalconsequences of aneuploidy. Proc Natl Acad Sci U S A 2012;109:12644–9.

23. Ried T, Hu Y, Difilippantonio MJ, Ghadimi BM, GradeM, Camps J. Theconsequences of chromosomal aneuploidy on the transcriptome ofcancer cells. Biochim Biophys Acta 2012;1819:784–93.

24. Shankavaram UT, Varma S, Kane D, Sunshine M, Chary KK, ReinholdWC, et al. CellMiner: a relational databaseandquery tool for theNCI-60cancer cell lines. BMC Genomics 2009;10:277.

25. Liu G, Loraine AE, Shigeta R, Cline M, Cheng J, Valmeekam V, et al.NetAffx: Affymetrix probesets and annotations. Nucleic Acids Res2003;31:82–6.

26. Roschke AV, Tonon G, Gehlhaus KS, McTyre N, Bussey KJ, LababidiS, et al. Karyotypic complexity of the NCI-60 drug-screening panel.Cancer Res 2003;63:8634–47.

27. NCI60 Discovery Services [Internet]. Available from: http://dtp.nci.nih.gov/docs/misc/common_files/cell_list.html.

28. Jain M, Nilsson R, Sharma S, Madhusudhan N, Kitami T, Souza AL,et al. Metabolite profiling identifies a key role for glycine in rapid cancercell proliferation. Science 2012;336:1040–4.

29. Stingele S, Stoehr G, Peplowska K, Cox J, Mann M, Storchova Z.Global analysis of genome, transcriptome and proteome revealsthe response to aneuploidy in human cells. Mol Syst Biol 2012;8:608.

30. Reimand J, Arak T, Vilo J. g:Profiler—a web server for functionalinterpretation of gene lists (2011 update). Nucleic Acids Res 2011;39:W307–15.

31. SubramanianA, TamayoP,Mootha VK,Mukherjee S, Ebert BL,GilletteMA, et al. Gene set enrichment analysis: a knowledge-based approachfor interpreting genome-wide expression profiles. Proc Natl Acad SciU S A 2005;102:15545–50.

32. Liberzon A, Subramanian A, Pinchback R, Thorvaldsd�ottir H, TamayoP, Mesirov JP. Molecular signatures database (MSigDB) 3.0. Bioin-formatics 2011;27:1739–40.

33. Xia J, Mandal R, Sinelnikov IV, Broadhurst D, Wishart DS. MetaboA-nalyst 2.0—a comprehensive server for metabolomic data analysis.Nucl Acids Res 2012;40:W127–33.

34. De Hoon MJL, Imoto S, Nolan J, Miyano S. Open source clusteringsoftware. Bioinformatics 2004;20:1453–4.

35. Saldanha AJ. Java Treeview—extensible visualization of microarraydata. Bioinformatics 2004;20:3246–8.

36. Roschke AV, Glebov OK, Lababidi S, Gehlhaus KS, Weinstein JN,Kirsch IR. Chromosomal instability is associated with higher expres-sion of genes implicated in epithelial–mesenchymal transition, cancerinvasiveness, and metastasis and with lower expression of genesinvolved in cell cycle checkpoints, DNA repair, and chromatin main-tenance. Neoplasia 2008;10:1222–30.

37. Saeed AI, Sharov V, White J, Li J, Liang W, Bhagabati N, et al. TM4: afree, open-source system for microarray data management and anal-ysis. BioTechniques 2003;34:374–8.

38. Karlsson E, Delle U, Danielsson A, Olsson B, Abel F, Karlsson P, et al.Gene expression variation to predict 10-year survival in lymph-node-negative breast cancer. BMC Cancer 2008;8:254.

39. Baker DJ, Jeganathan KB, Cameron JD, Thompson M, Juneja S,Kopecka A, et al. BubR1 insufficiency causes early onset ofaging-associated phenotypes and infertility in mice. Nat Genet2004;36:744–9.

40. Li M, Fang X, Wei Z, York JP, Zhang P. Loss of spindle assemblycheckpoint-mediated inhibition of Cdc20 promotes tumorigenesis inmice. J Cell Biol 2009;185:983–94.

41. Carter SL, Eklund AC, Kohane IS, Harris LN, Szallasi Z. A signature ofchromosomal instability inferred from gene expression profiles pre-dicts clinical outcome in multiple human cancers. Nature Genet 2006;38:1043–8.

42. Swanton C, Nicke B, Schuett M, Eklund AC, Ng C, Li Q, et al.Chromosomal instability determines taxane response. Proc Natl AcadSci U S A 2009;106:8671–6.






43. Salvatore G, Nappi TC, Salerno P, Jiang Y, Garbi C, Ugolini C, et al. Acell proliferation and chromosomal instability signature in anaplasticthyroid carcinoma. Cancer Res 2007;67:10148–58.

44. Decaux O, Lod�e L, Magrangeas F, Charbonnel C, Gouraud W,J�ez�equel P, et al. Prediction of survival in multiple myeloma basedon gene expression profiles reveals cell cycle and chromosomalinstability signatures in high-risk patients and hyperdiploid signaturesin low-risk patients: a study of the Intergroupe Francophone duMy�elome. J Clin Oncol 2008;26:4798–805.

45. Lengauer C, Kinzler KW, Vogelstein B. Genetic instabilities in humancancers. Nature 1998;396:643–9.

46. Venet D, Dumont JE, Detours V. Most random gene expressionsignatures are significantly associated with breast cancer outcome.PLoS Comput Biol 2011;7:e1002240.

47. Dai H, van't Veer L, Lamb J, He YD, Mao M, Fine BM, et al. A cellproliferation signature is a marker of extremely poor outcome in asubpopulation of breast cancer patients. Cancer Res 2005;65:4059–66.

48. Mosley JD, Keri RA. Cell cycle correlated genes dictate the prog-nostic power of breast cancer gene lists. BMC Med Genomics2008;1:11.

49. Sheltzer JM, Amon A. The aneuploidy paradox: costs and benefits ofan incorrect karyotype. Trends Genet 2011;27:446–53.

50. Levine AJ. p53, the cellular gatekeeper for growth and division. Cell1997;88:323–31.

51. Baker SJ, Preisinger AC, Jessup JM, Paraskeva C, Markowitz S,Willson JKV, et al. p53 gene mutations occur in combination with17p allelic deletions as late events in colorectal tumorigenesis. CancerRes 1990;50:7717–22.

52. Lassam NJ, From L, Kahn HJ. Overexpression of p53 is a late eventin the development of malignant melanoma. Cancer Res 1993;53:2235–8.

53. Boyle JO, Hakim J, Koch W, van der Riet P, Hruban RH, Roa RA, et al.The incidence of p53 mutations increases with progression of headand neck cancer. Cancer Res 1993;53:4477–80.

54. Lee AJ, Endesfelder D, Rowan AJ, Walther A, Birkbak NJ, Futreal PA,et al. Chromosomal instability confers intrinsic multidrug resistance.Cancer Res 2011;71:1858–70.

55. Bakhoum SF, Danilova OV, Kaur P, Levy NB, Compton DA. Chromo-somal instability substantiates poor prognosis in patients with diffuselarge B-cell lymphoma. Clin Cancer Res 2011;17:7704–11.

56. Raspe E, Decraene C, Berx G. Gene expression profiling to dissect thecomplexity of cancer biology: pitfalls and promise. Semin Cancer Biol2012;22:250–60.

57. Rabinovich GA. Galectin-1 as a potential cancer target. Br J Cancer2005;92:1188–92.

58. Dalotto-Moreno T, Croci DO, Cerliani JP, Martinez-Allo VC, Dergan-Dylon S, Huergo SPM, et al. Targeting galectin-1 overcomes breastcancer associated immunosuppression and prevents metastatic dis-ease. Cancer Res 2013;73:1107–17.

59. Banh A, Zhang J, Cao H, Bouley DM, Kwok S, Kong C, et al. Tumorgalectin-1 mediates tumor growth and metastasis through regulationof T-cell apoptosis. Cancer Res 2011;71:4423–31.

60. Ruoslahti E. Fibronectin and its integrin receptors in cancer. In: WoudeGFV, Klein G, editors. Advances in cancer research [Internet]. AcademicPress; 1999 [cited 2012 Dec 21], p. 1–20. Available from: http://www.elsevier.com/books/advances-in-cancer-research/vande-woude/978-0-12-006679-7#.

Sheltzer





2013;73:6401-6412. Published OnlineFirst September 16, 2013.Cancer Res Jason M. Sheltzer Clinical PrognosisPresent in Chromosomally Unstable Cancer Cells and Informs A Transcriptional and Metabolic Signature of Primary Aneuploidy Is

Updated version

10.1158/0008-5472.CAN-13-0749doi:

Access the most recent version of this article at:

Material

Supplementary

http://cancerres.aacrjournals.org/content/suppl/2013/09/17/0008-5472.CAN-13-0749.DC1

Access the most recent supplemental material at:

Cited articles

http://cancerres.aacrjournals.org/content/73/21/6401.full#ref-list-1

This article cites 58 articles, 26 of which you can access for free at:

Citing articles

http://cancerres.aacrjournals.org/content/73/21/6401.full#related-urls

This article has been cited by 21 HighWire-hosted articles. Access the articles at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected]

To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at

Permissions

Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://cancerres.aacrjournals.org/content/73/21/6401To request permission to re-use all or part of this article, use this link



http://cancerres.aacrjournals.org/lookup/doi/10.1158/0008-5472.CAN-13-0749

http://cancerres.aacrjournals.org/content/suppl/2013/09/17/0008-5472.CAN-13-0749.DC1

http://cancerres.aacrjournals.org/content/73/21/6401.full#ref-list-1

http://cancerres.aacrjournals.org/content/73/21/6401.full#related-urls

http://cancerres.aacrjournals.org/cgi/alerts

mailto:[email protected]

http://cancerres.aacrjournals.org/content/73/21/6401


a transcriptional and metabolic signature of primary...

Documents