Chromosome Instability, Chromosome Transcriptome, and Clonal Evolution of Tumor Cell Populations

Chromosome instability, chromosome transcriptome, and clonal evolution of tumor cell populations

ChongFeng Gao*, Kyle Furge†, Julie Koeman‡, Karl Dykema†, Yanli Su*, Mary Lou Cutler§, Adam Werts*, Pete Haak¶, and George F. Vande Woude*ʈ

Laboratories of *Molecular Oncology, †Computational Biology, ‡Germline Modiﬁcation, and ¶Microarray Technology, Van Andel Research Institute, 333 Bostwick Avenue, N.E., Grand Rapids, MI 49503; and §University Services University of the Health Sciences, 4301 Jones Bridge Road, Bethesda, MD 20814

Edited by Janet D. Rowley, University of Chicago Medical Center, Chicago, IL, and approved March 29, 2007 (received for review January 23, 2007) Chromosome instability and aneuploidy are hallmarks of cancer, Phenotypic switching is fundamental for malignant progression but it is not clear how changes in the chromosomal content of a cell (6), and therefore it is important to understand the responsible contribute to the malignant phenotype. Previously we have shown mechanisms. that we can readily isolate highly proliferative tumor cells and their Glioblastomas characteristically show extensive regional cytoge- revertants from highly invasive tumor cell populations, indicating netic heterogeneity (10, 11), and this diversity may be responsible how phenotypic shifting can contribute to malignant progression. for tumor evolution and progression (10, 12). Here we show that Here we show that chromosome instability and changes in chro- distinct changes in karyotype from chromosome instability accom- mosome content occur with phenotypic switching. Further, we pany phenotypic switching. These changes, in turn, dictate changes show that changes in the copy number of each chromosome in the chromosome transcriptome that provide the expression of quantitatively impose a proportional change in the chromosome individual genes that are necessary for the conversion between the transcriptome ratio. This correlation also applies to subchromo- invasive and proliferative phenotypes. somal regions of derivative chromosomes. Importantly, we show that the changes in chromosome content and the transcriptome Results and Discussion favor the expression of a large number of genes appropriate for Karyotype Differences Accompany Switching of Glioblastoma Tumor the speciﬁc tumor phenotype. We conclude that chromosome Cell Phenotypes. To determine whether chromosome instability is instability generates the necessary chromosome diversity in the responsible for tumor cell phenotypic switching, we examined tumor cell populations and, therefore, the transcriptome diversity DB-P, DB-A2, DB-A6, and A2-BH7 cells (SI Table 4) by using to allow for environment-facilitated clonal expansion and clonal spectral karyotyping (SKY) (SI Fig. 4). For each cell type, we evolution of tumor cell populations. determined the total number of copies of each chromosome [or derivative (der) chromosomes] on the basis of 10 metaphase cells aneuploidy ͉ glioma ͉ invasion ͉ proliferation ͉ HGF/SF (Tables 1 and 2, respectively). Each cell population had near- tetraploid karyotypes, but karyotypes were particularly different from the parental DB-P cells as well as distinct from one another owell first proposed the clonal evolution of tumor cell popu- (Tables 1 and 2). Nlations to explain how malignant tumors arise over time (1). Tumor progression results from genetic variability within the tumor The Differences in DB-A2 and DB-P Cell Karyotypes Are Reflected in cell population that allows for clonal expansion of more aggressive Their Transcriptome Ratios. The significant differences in the karyo- tumor phenotypes (1, 2). Although chromosome instability and the types of each subclone led us to ask whether the changes in resulting cytogenetic heterogeneity are the most readily recognized chromosome content mediated changes in chromosome transcrip- genetic events associated with tumor progression (3–5) and may be tome that could influence phenotype determination. Recent gene responsible for tumor evolution and progression, precisely how they expression profiling studies have been used to assess the influence contribute to the malignant phenotype is not clear. Invasion and of chromosomal imbalance on overall gene expression (13–17). We proliferation are crucial requirements for tumor progression, and used this approach to determine whether the ratio of the changes we have chosen to study these steps in glioblastoma tumor cells (6). in chromosome content between parental DB-P cells and the Glioblastoma cells invade normal brain tissue (7); after surgical subclones influenced the chromosome transcriptome ratios. Begin- resection, residual invasive cells can quickly regain a proliferative ning with cDNA microarrays, individual gene expression differ- phenotype, progressing to a more aggressive tumor (8). Because ences between the DB-A2 subclone and the parental line, DB-P, glioblastomas rarely metastasize from the CNS, the sequential were calculated as described (15). The transcriptome ratios were selection of invasive and proliferative tumor cells constitutes the determined by averaging the relative expression ratios of each gene main theme for this tumor’s progression (7). It is, therefore, in a given chromosome or der subchromosomal region. These data MEDICAL SCIENCES critically important to understand the molecular mechanisms that are presented as log2-transformed to show the direction (ϩ/Ϫ) and permit high-frequency phenotypic switching and control tumor cell magnitude or are in linear scale to compare to the chromosome proliferation and invasion (6). content ratio. The log2-transformed data were displayed as scatter c-Met and its ligand, hepatocyte growth factor/scatter factor plots on the respective chromosome (Fig. 1). In this way, the (HGF/SF), can regulate both proliferative and invasive phenotypes of glioblastoma tumor cells (9). We have previously shown that alternating between proliferative and invasive phenotypes was Author contributions: C.G. and G.F.V.W. designed research; C.G., J.K., Y.S., A.W., and P.H. critically linked with switching between the Myc and Ras/MAPK performed research; C.G., K.F., K.D., and M.L.C. analyzed data; and C.G. wrote the paper. pathways, respectively (6). Starting with a highly invasive popula- The authors declare no conﬂict of interest. tion, DB-P, we were able to select two subclones, DB-A2 and This article is a PNAS Direct Submission. DB-A6, that are highly proliferative or both invasive and prolifer- Abbreviations: der, derivative; HGF/SF, hepatocyte growth factor/scatter factor; SKY, spec- ative, respectively. The parental DB-P cells and each subclone tral karyotyping; uPA, urokinase-type plasminogen activator. showed distinct in vitro and in vivo phenotypes and signaling ʈTo whom correspondence should be addressed. E-mail: [email protected]. pathways that correlated with their invasive or proliferative phe- This article contains supporting information online at www.pnas.org/cgi/content/full/ notypes [supporting information (SI) Table 4]. From the DB-A2 0700631104/DC1. subclone, we further selected a highly invasive revertant, A2-BH7. © 2007 by The National Academy of Sciences of the USA

www.pnas.org͞cgi͞doi͞10.1073͞pnas.0700631104 PNAS ͉ May 22, 2007 ͉ vol. 104 ͉ no. 21 ͉ 8995–9000 Downloaded by guest on September 30, 2021 Table 1. Full chromosomes in DB-P and its subclones via SKY subclone and the parental DB-P (Table 1 and Fig. 1A). Concor- Chromosome DB-P DB-A2 DB-A6 A2-BH7 dantly, a comparison of the chromosome transcriptomes shows that the average of the number of up- and down-regulated genes on 1 2(10) 20 2(10) 20 2(10) 20 2(10) 20 these chromosomes was largely unchanged (0.08, Ϫ0.02, 0.01, and 2 3(10) 30 4(5) 3(5) 35 3(9) 2(1) 29 4(7) 3(3) 37 0.02 in log2 ratio units, respectively) (Fig. 1A). However, when the 3 3(10) 30 4(1) 3(9) 31 3(10) 30 2(10) 20 chromosome copy number changed between DB-A2 and DB-P, the 4* 0 000 5 2(10) 20 2(9) 1(1) 19 2(7) 1(3) 17 2(10) 20 chromosome transcriptome also changed in the same direction. 6 3(10) 30 3(10) 30 3(6) 2(4) 26 2(10) 20 Thus, for chromosomes 2, 7, and 8, which were higher in copy 7 3(10) 30 4(10) 40 5(1)3(4) 2(5) 27 3(10) 30 number in DB-A2 cells, the transcriptome ratios were 0.18, 0.16, 8 3(10) 30 4(10) 40 4(7) 3(3) 37 4(10) 40 and 0.30 in log2 ratio units, respectively (Fig. 1B). For chromosomes 9 3(9) 2(1) 29 2(10) 20 3(1)2(7) 1(2) 19 2(10) 20 9, 14, 15, 20, and 21, which were fewer in copy number in DB-A2, 10 3(10) 30 2(10) 20 2(10) 20 2(10) 20 the corresponding ratios were Ϫ0.21, Ϫ0.33, Ϫ0.25, Ϫ0.32, and 11* 4(2) 3(6) 2(2) 30000Ϫ0.54, respectively (Fig. 1B); chromosome 21 showed the greatest 12 4(10) 40 4(9) 3(1) 39 4(8) 3(2) 38 4(10) 40 difference in copy number and the largest change in transcriptome 13 3(9) 2(1) 29 2(8) 1(2) 18 2(10) 20 2(10) 20 14 3(7) 2(3) 27 2(10) 20 2(10) 20 2(10) 20 ratio. These results show remarkable concordance between the 15 5(5) 4(3) 3(2) 43 4(8) 3(2) 38 3(9) 2(1) 29 4(10) 40 chromosome copy number and the change in chromosome tran- 16 1(10) 10000scriptome, suggesting that the change in chromosome content can 17 4(10) 40 4(9) 3(1) 39 4(7)3(2)2(1) 36 4(10) 40 be responsible for the changes in expression of each gene as part of 18 3(9) 2(1) 29 3(10) 30 3(10) 30 3(10) 30 the transcriptome. 19 4(3) 3(7) 33 4(10) 40 4(10) 40 4(10) 40 20 5(8) 4(2) 48 4(9) 3(1) 39 4(10) 40 4(10) 40 Subchromosomal Regions of Derivative Chromosomes Contribute to 21 8(1) 7(7) 6(2) 69 4(9) 3(1) 39 4(10) 40 4(4) 3(6) 34 the Chromosome Transcriptome (DB-A2 vs. DB-P). Many differences 22† 4(6) 3(4) 36 4(10) 40 4(9) 3(1) 39 4(10) 40 X 4(9) 3(1) 39 4(9) 3(1) 39 4(6) 3(4) 36 4(10) 40 exist in derivative chromosomes between DB-P and the subclones (Table 2). Some derivatives were common to all cell lines, whereas Bold, number of copies of chromosome (n) observed out of 10 metaphases. others were present in a fraction of the subclones. Particularly Italics, total number of copies of chromosome in 10 metaphases; this is the interesting are the derivative chromosomes that are unique to the number used to determine the chromosome content ratio. subclones, indicating that they were present in Ͻ1 in 10 metaphases *There are no full copies of chromosomes 4 and 11 in any of the cells. However, portions of chromosome 4 and 11 are represented in der chromosomes (see in the parental DB-P cells (or perhaps new translocation events). Table 2). The surprising concordance between chromosome copy number †In some metaphases, chromosome 22 is indistinguishable from marker chromo- and transcriptome changes led us to test whether subchromosomal somes by SKY. Detailed examination with FISH showed a DB-A2/DB-P ratio of regions in derivative chromosomes would also influence the chro- 1.02 (240/236 signals for chromosome 22 in 60 cells). Their value is used as mosome transcriptome. Thus, the chromosome copy number (Ta- chromosome content ratio. ble 1) is added to the subchromosomal region provided by the derivative copy (Table 2) to yield the chromosome content for that consequences of the chromosomal changes can be directly com- region. Such comparisons for DB-A2 and DB-P are shown (Fig. pared with changes in the transcriptome. 1C). Thus, a net gain in chromosome 1q31-qter in DB-A2 from der (11)t (1:11) (Table 2) compared with DB-P increases the chromo- Comparison of Chromosome Content and Transcriptome for DB-A2 some transcriptome ratio in this region to 0.32 in log2 units (Fig. and DB-P. For chromosomes 6, 12, 17, and 22, no changes in 1C). Interestingly, the contributions of der (3)t (3, 13) and der (3)t chromosome copy number were observed between the DB-A2 (3, 18) to the DB-A2 and DB-P chromosome 3 transcriptomes have

Table 2. Derivative chromosomes partitioning in DB-P cells and subclones Derivative chromosomes DB-P DB-A2 DB-A6 A2-BH7

der(1)t(11;1;4)(?;p31-q43?;q?) 1(10) 10 1(10) 10 1(9) 9 1(10) 10 der(4)t(1;4)(?;q12) 2(9) 1(1) 19 2(10) 20 2(8) 1(2) 18 2(10) 20 der(4)t(4;5)(p14;q?) 2(10) 20 2(10) 20 2(10) 20 2(10) 20 der(5)t(16;5;4)(q12;p14-q23.3;p15) 1(10) 10 2(10) 20 2(10) 20 2(10) 20 der(7)t(6;7)(?;p22) 2(10) 20 2(10) 20 2(9) 18 2(10) 20 ins(9;13)(p12;?) 2(9) 1(1) 19 2(10) 20 2(8) 1(2) 18 2(10) 20 der(11)t(1;11)(q31;q13) 1(9) 9 2(9) 1(1) 19 2(8) 1(2) 18 2(10) 20 del(13)(q13) 2(8) 1(2) 18 4(1) 2(9) 22 2(9) 1(1) 19 2(10) 20 der(16)t(5;16)(p13;q12) 2(10) 20 2(10) 20 2(8) 1(2) 18 1(10) 10 der(3)t(3;13)(p21;q12) 1(10) 10 0 0 0 der(5)t(19;5;4)(p13?;p11-q23.3;p15) 1(10) 10 0 0 0 der(10)t(X;10)(p21;p11) 1(10) 10 0 0 0 der(13)t(4;13)(p10;q10) 1(8) 80 0 0 der(19)t(16;19)(q12;p13) 1(8) 80 0 0 del(15)(?) 1(3) 3 1(1) 10 0 der(11)t(4;5;11)(p15;q13-q23.3;p15) 0 2(10) 20 2(10) 20 2(10) 20 der(16)t(10;16)(q22;q22) 0 1(10) 10 2(9) 1(1) 19 2(10) 20 del(16)(p11.2)der(16)t(10;16)(q2;q22) 0 1(10) 10 0 1(10) 10 der(3)t(3;17)(p10;q11) 001(10) 10 0 der(3)t(3;18)(q21;p11.2) 0 1(9) 901(10) 10 der(5)t(5;15)(q21;q22) 001(3) 30 der(6)t(3;6)(q26?;q23) 001(4) 40 der(15)t(10;15)(?;q22) 001(9) 90

Bold, number of copies of chromosome (n) observed out of 10 metaphases. Italics, total number of copies of chromosome in10 metaphases and the value used to determine the chromosome content ratio.

8996 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0700631104 Gao et al. Downloaded by guest on September 30, 2021 A opposing effects in the regions of 3pter-p21 and 3q21-qter. Here the 6 transcriptome ratio changes in opposite directions to 0.30 and DB-A2 7 DB-A2 Ϫ 6 6 0.08 17 0.01 0.28 log2 units, respectively, but remains unchanged at 0.08 in the DB-P 7 DB-P region between (Fig. 1C). Next, in chromosome 4, a gain in the 4p15 region from der (11)t (4, 5, 11) changes the chromosome 4 DB-A2 DB-A2 12 -0.02 22 0.02 transcriptome ratio in this region for DB-A2 to 0.31. However, the DB-P DB-P same der chromosomes, der (11)t (4, 5, 11), also contribute to an increase in the chromosome 5 transcriptome in the 5q13-q23.3 B region to 0.25. The chromosome transcriptome can also be influenced by mul- DB-A2 DB-A2 2 0.18 14 -0.33 tiple derivative chromosomes. Thus, in DB-A2, the absence of DB-P DB-P 10pter-q22 in der (16)t (10, 16) and der (16)(p11.2)der (16)t (10, 16) results in the chromosome 10 transcriptome ratio in this region of 6 DB-A2 DB-A2 Ϫ 7 -0.25 0.56 compared with 0.04 in the remainder of the chromosome, 7 6 0.16 15 DB-P 7 DB-P where the content of DB-A2 and DB-P are the same (Fig. 1C). Likewise, gene expression on chromosome 11 is influenced by three DB-A2 DB-A2 chromosomes, der (1)t (11, 1, 4), der (11)t (1, 11), and der (11)t (4, 8 0.30 20 -0.32 DB-P DB-P 5, 11), resulting in a decrease of two different regions (11pter-11p15 and 11q13-qter) of Ϫ0.69 and Ϫ0.37, respectively. Chromosomes 13 DB-A2 Ins13 DB-A2 and 16 changes are complex, consisting entirely of der chromo- 9 -0.21 21 -0.54 DB-P Ins13 DB-P somes or a combination of whole chromosomes and derivatives (Fig. 1C). The chromosome 19 transcriptome ratio does not change with der (5)t (19;5;4;) and der (19)t (16, 19) (Table 2), but these two C derivatives may compensate for the lower copy numbers of chro- 11 11 1 4 1q31 mosome 19 in DB-P cells (Tables 1 and 2). The major exception we DB-A2 4 1 1 -0.02 1 found in associating chromosome copy number changes with tran- 11 11 1q31 4 1q31 DB-P 1 1 1 0.32 4 scriptome changes is the significant decrease in the X transcriptome in DB-A2 cells, although the only region absent is pter-p21 (Fig. 3 0.30 DB-A2 3q21 3p21 1C), which can possibly be explained by X-inactivation. 3 18 0.08 13 3q21 DB-P 3p21 -0.28 We conclude that, whether comparing full chromosome content 3 or segments of translocated regions of der chromosome(s), the 16 4p15 0.31 5 11 4p15 1 5 5 chromosome transcriptomes’ ratios reflect chromosome content. DB-A2 4 4 1 4 4 11 4 11 5 19 4 0.09 DB-P 4 1 1 4 4 13 Comparing Chromosome and Transcriptome Ratios of DB-A6 to DB-A2 (vs. DB-P). We next compared the chromosome content changes and 5 4 16 5q13-q23 5 0.05 transcriptome ratios between DB-A6 and DB-P to see whether the DB-A2 4 11 5 4 5q13 0.25 5 5q23 results obtained with DB-A2 versus DB-P extended to other cell DB-P 19 0.10 5 4 4 clonal isolates. The DB-A6 and DB-A2 karyotypes are more similar del16p11.2 to each other than either is to the parental DB-P cells (Tables 1 and DB-A2 16 -0.56 10q22 10q22 10 10 2). The chromosome content ratios of DB-A6 to DB-P are the same 0.04 x DB-P 10p11 as DB-A2 to DB-P with chromosomes 8, 9, 14, 20, and 21 (cf. Fig. 10 1B with SI Fig. 5B), and the chromosome transcriptome ratios are 4 11 -0.69 5 11p15 all virtually the same [(0.30/0.33), (Ϫ0.21/Ϫ0.22), (Ϫ0.33/Ϫ0.35), DB-A2 11 11p15 1 1 11 4 -0.02 11 11q13 Ϫ Ϫ Ϫ Ϫ 11 11 ( 0.32/ 0.28), and ( 0.54/ 0.55), respectively]. By contrast, DB-P 11q13 1 -0.37 1 4 DB-A6 and DB-A2 differ significantly in chromosomes 2, 7, and 15 content (Tables 1 and 2), and we observe very different transcrip- Ins:13 DB-A2 tome ratios [(0.18/0.11), (0.16/0.02), and (Ϫ0.25/-0.40), respectively; 9 -0.50 13 4 3 13q10 Ins:13 cf.Fig.1B and SI Fig. 5 A and B]. For the six chromosomes with no DB-P 13q12 13 9 obvious changes in chromosome copy number, three chromosomes del16p11.2 4 (7, 12, and 22) show little change in chromosome transcriptome 16 16 -0.06 DB-A2 16q22 5 5 16q12 16p11 Ϫ ratios (0.02, 0.05, and 0.07 respectively), two of them (chromo- MEDICAL SCIENCES 16 10 16 0.14 4 16 19 16q22 DB-P 16q12 5 16q12 -0.07 somes 2 and 6) show a marginal increase (0.11), and chromosome 5 16 16 18 shows a substantial increase (0.20) (SI Fig. 5A). These variations 18 18p11 18p11 0.24 in the chromosome transcriptome ratio may result from unde- DB-A2 3 18 -0.06 tectable cytogenetic changes or, for chromosome 18, be due to the DB-P

DB-A2 -0.01 the relative gain or loss of chromosome copies. Expression data for genes that 19 16 19p13 19 DB-P 19p13 5 map to each chromosome are displayed as scatter plots. Blue dots represent 19 4 down-regulated genes and red dots represent up-regulated genes. The change in the transcriptome for each chromosome is in log2 units and is shown by the DB-A2 X -0.31 number to the right of each scatter plot. (A) Chromosomes without copy number DB-P X Xp21 changes. (B) Chromosomes with full copy gains or losses. (C) Chromosomes with 10 derivative regions. Each scatter plot was subdivided according to gain or loss of the subchromosomal regions. The chromosome transcriptome ratio for each Fig. 1. Comparisons of chromosome copy number and the chromosome region is indicated by the mean values (in log2 units) to the right of the scatter transcriptome for DB-A2 and DB-P. Representative chromosomes (indicated plot. A correlation is observed between each derivative subchromosomal region by the numeral at the left) from DB-A2 and DB-P cells were aligned to compare and the chromosome transcriptome proﬁle.

Gao et al. PNAS ͉ May 22, 2007 ͉ vol. 104 ͉ no. 21 ͉ 8997 Downloaded by guest on September 30, 2021 1.6

1.4

1.2

Chr. 0.8 Gene Ratio 0.6

0.4

0.2

1 r 2 1 1 5 r 3 r 6 7 8 9 2 5 2 3 4 7 9 0 1 2 3 2 2 1 e 13 2 e er 1 1 1 1 15 1 1 2 2 2 ter p t t t p ter ter -qte -p -q -q -q -q -q2 -q q -p11 -1q 1 r 1-q 1 r- r-q 3 3 r 2-q 3 1- r r 3 te 2 2 15 te 2 te 2 1 te q p p pte p q1 q p pter- p pte 1 3 3 3q 4 4p 5 5 5 0 0q 1 8 1 1 1 1 11q1 1 18p

Fig. 2. Comparing chromosomal copy number ratio to the chromosome transcriptome ratio. The copy numbers of each chromosome in DB-P and DB-A2 were determined by SKY and FISH and are presented in Tables 1 and 2. The ratios of chromosome numbers for DB-A2 versus DB-P were determined by comparing the total number of chromosome or subchromosomal region in 10 metaphases. The chromosome transcriptome ratios were obtained by averaging the expression of all available genes on each chromosome or subchromosome region, and values determined in log2 units were converted to numerical ratios.

low number of genes expressed from the chromosome (chromo- Do Changes in Chromosome Content Provide the Transcriptome some 18). Changes Required for Phenotype Determination? The major question As with the DB-A2 comparison with DB-P cells, we also observe of these analyses is whether changes in the chromosome transcrip- a strong influence of der chromosomes on the transcriptomes. The tome result in changes in the expression of specific genes that favor chromosome 15 transcriptome ratio in DB-A6 is influenced by der the invasive or proliferative phenotypes. Discriminant gene analysis (15)t (10:15) (SI Fig. 5C). Likewise, the transcriptome ratios of was performed to identify specific genes/gene sets that had signif- chromosome 3 [(0.30, 0.08, Ϫ0.28) (Fig. 1C) and (0.15, 0.22, Ϫ0.24) icant expression differences between DB-A2 and DB-P (SI Table (SI Fig. 5C) ] and chromosome 16 [(Ϫ0.06, 0.14, Ϫ0.07) (Fig. 1C) 7). Expression of 89 genes that reside in autosomes are altered Ϫ and (0.30, 0.29, Ϫ0.13) (SI Fig. 5C)] are dramatically different Ͼ2-fold in DB-A2 (all P values Ͻ10 4; SI Table 7). Twenty-seven among DB-A2, DB-A6, and DB-P due to derivative chromosomes of these genes have been implicated in the regulation of tumor (Table 2). These analyses further show that the content of chro- growth or apoptosis (Table 3 and SI Table 7). Consistent with the mosome governs the differences in the chromosome transcriptome rapid tumor growth and low invasion/migration phenotypes of ratio. DB-A2 cells (SI Table 4), all 22 genes down-regulated in DB-A2 are related to proinvasion, proapoptosis, or growth inhibition, whereas Global Comparison of Chromosome Content and Transcriptome Dif- all five up-regulated genes are proproliferation or antiinvasion ferences. Up to this point, we have been comparing the numerical genes. Interestingly, 17 of the 22 down-regulated genes reside in values of the chromosome content ratio to the transcriptome ratio chromosomes or subchromosomal regions that decrease in copy number in DB-A2, only three reside in chromosomes that increase, in log2 units. We, therefore, converted the log2-transformed transcriptome ratios back to a linear scale and made the comparisons and two are located in chromosomes that do not change (Table 3). These data show that karyotypic changes are consistent with having for all combinations of tumor cell populations for full chromosomes a role in phenotypic conversion of DB-A2 cells. (SI Table 5) or for the DB-A2/DB-P comparison of subchromo- Whereas the expression of a small number of genes (89 of 19,552) somal regions of der chromosomes (SI Table 6). In Fig. 2, we alters significantly (SI Table 7), the expression changes of the present the ratios of the total number of chromosomes counted in majority of genes track closely with chromosome content (SI Fig. 7 10 metaphase spreads (Tables 1 and 2) as histogram plots against A–C). The significant changes in transcription of a small subset of the numerical ratios of respective chromosome transcriptomes genes may result from mutations or epigenetics that alter the converted from the log2 ratios. Further, the comparisons between transcriptome, but it is also possible that they result from changes chromosome content ratios and chromosome transcriptome ratios in the chromosome content of other chromosomes that bear for DB-P parental and DB-A2, DB-A6, and A2-BH7 comparisons transcription regulator genes, and it would not be too surprising for are presented in SI Fig. 6 and SI Table 5. These analyses show gene expression changes to occur, surrounding the translocated dramatically that the fold increase or decrease in the chromosome regions of the der breakpoints. However, it is likely that these content ratio is virtually the same as the transcriptome ratio for all changes already exist and are delivered by the changes in chromo- comparisons, and the chromosome content transcriptome ratios some content. over all chromosomes average Ϸ1.0 (SI Table 5). Even the sub- The gain of chromosome 7 and the loss of chromosome 10 are chromosomal regions of derivative chromosomes, when considered common cytogenetic alterations in glioblastoma (11) and may be as part of the copy number of a specific chromosome (Fig. 2 and SI directly related to malignant phenotype (18). In this study, we Table 6), markedly influence the transcriptome ratio of that chro- observed changes in chromosomes 7 and 10 in the subclones (Table mosome, and, dramatically, the numerical ratios of the chromo- 1). In the highly proliferative DB-A2 subclone, relative to DB-P, an some content are virtually the same as the transcriptome of the increase in the copy number of chromosome 7 results in an increase specific chromosome region. We conclude that there is a direct in the transcriptome (0.16) (Fig. 1B). Further, chromosome 7 quantitative correlation between chromosome content and a pro- content in the invasive revertant of DB-A2, A2-BH7 (SI Fig. 8), was portional change in the transcriptome. restored to the DB-P level, as was the transcriptome ratio of Ϫ0.17,

8998 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0700631104 Gao et al. Downloaded by guest on September 30, 2021 Table 3. Genes with signiﬁcantly altered expression in the A 16 DB-A2/DB-P comparison A2-BH7 10q22 -0.04 10 10 10q22 16 0.32 Function in Chromosome Gene DB-A2 10q22 10 Map Gene tumorigenesis* ratio† ratio† P value‡ –3 –2 –1 0 1 2 3 4p16.3 SOD3 Proliferation 1.28 5.24 1.52eϪ5 Ϫ9 600 17q21.3 HOB7 Proliferation 0.98 2.75 8.75e B 7 C Ϫ5 2 500 2q35 FN1 Invasion 1.17 0.12 4.51e -P -A -BH B B 2 2q37 COL6A3 Invasion 1.17 0.05 4.67eϪ5 D D A 400 Ϫ5 11p15.5 IFITM1 Invasion 0.49 0.17 3.34e HGF - + - + - + 300 11p15.5 IFITM2 Invasion 0.49 0.24 3.79eϪ5 uPA 200 13q33 EFNB2 Invasion 0.49 0.38 2.33eϪ5 100 Ϫ5 14q22 BMP4 Invasion 0.74 0.15 4.70e GAPDH Number of invading cells 0 Ϫ5 16q13 MMP2 Invasion 1.43 0.17 3.56e C B428 20q11.23 MYL9 Invasion 0.81 0.08 7.38eϪ5 21q22.3 COL6A1 Invasion 0.56 0.22 4.01eϪ5 Fig. 3. Chromosome change-associated up-regulation of uPA contributes to 1q21.3 CKIPϪ1 Apoptosis 1.00 0.39 2.60eϪ5 the invasive phenotype of A2-BH7. (A) Changes in A2-BH7 versus DB-A2 cells in 10p13 CUGBP2 Apoptosis 0.75 0.15 4.02eϪ5 chromosome number and chromosome transcriptome. (B) Northern blot analysis 14q22 ZFP36L1 Apoptosis 0.74 0.38 2.04eϪ5 showing up-regulation of uPA in BH7 cells. (C) The uPA inhibitor, B428, blocked 13q31.2 STK24 Apoptosis 0.49 0.47 8.27eϪ6 HGF-induced cell invasion through Matrigel. Cells (10,000 per insert) were loaded 20p13 SMOX Apoptosis 0.81 0.42 1.49eϪ5 into Matrigel inserts and treated with 10 ␮M B428 or left untreated (controls, C) 3q23 RPB1 Growth inhibition 0.76 0.16 3.76eϪ5 for 1 h before adding HGF/SF. 6q24 AKAP12 Growth inhibition 1.00 0.49 5.88eϪ6 9q34.1 AK1 Growth inhibition 0.83 0.33 2.42eϪ5 10p13 RSU1 Growth inhibition 0.75 0.45 9.85eϪ6 bles 5 and 6), the individual gene ratios vary significantly in 10p15 AKR1X3 Growth inhibition 0.75 0.36 2.75eϪ5 contributing to the transcriptome average (Table 3) (P Ͻ 10Ϫ4; SI Ϫ 11p15.2 DKK3 Growth inhibition 0.49 0.45 9.02e 6 Table 7). The chromosome content, therefore, in delivering the Ϫ5 11q23.2 TSLC1 Growth inhibition 0.67 0.44 1.38e transcriptome in direct proportions, delivers a specific level of gene 21q22.1 DSCR1 Anti-invasion 0.56 0.47 6.58eϪ6 Ϫ expression for each gene. 16q24.2 CDH13 Anti-invasion 0.73 6.06 1.89e 17 18p11.3 TGIF AntiϪinvasion 1.34 2.28 1.95eϪ7 RSU-1 is associated with gliomas and is down-regulated in 7q32 ARP3␤ AntiϪmetastasis 1.20 2.84 2.92eϪ9 DB-A2 (Table 3) as part of chromosome 10p. This gene was isolated based on its ability to suppress v-RAS-induced transfor- *See SI Table 7 for references. mation (21). More importantly, ectopic expression of RSU-1 in † ␤ Simple numerical chromosome copy number ratio between D -A2 and DB-P glioma cells inhibits cell proliferation, anchorage-independent cells. growth, and tumorigenic activity in nude mice (21). Therefore, ‡P value for gene expression ratio. RSU-1 is a negative regulator of the proliferative phenotype in gliomas. Consistent with our analysis showing that the MAPK offsetting the gain relative to DB-A2 (see Fig. 1B and SI Fig. 8). pathway is involved in the invasive phenotype (6) (SI Table 4), the This result suggests that genes expressed on chromosome 7 may decreased RSU-1 expression in DB-A2 cells may contribute to their favor proliferation or antagonize invasion. highly proliferative and tumorigenic activity (22). Moreover, RSU-1 The parental clone DB-P has three copies of chromosome 10 plus also enhances Erk-2 activity (21) and is consistent with elevated one copy of der (10)t (X;10), whereas the proliferative DB-A2 Erk-2 activity in response to HGF/SF in DB-P-invasive cells (6). subclone has two full copies of chromosome 10 plus two copies of The der chromosomes that include chromosome 10 regions can der (16)t (10, 16), which contains the 10q22-qter region (Tables 1 also influence the phenotypes of DB-A2 and DB-P. The absence of and 2). Thus, two copies of 10p11-q22 and one copy of 10pter-p11 the 10pter-10q22 region in DB-A2 cells as described above can enhance proliferation and tumorigenic activity, whereas the pres- are absent from the DB-A2. However, in the invasive revertant ence of 10q22-qter in A2-BH7 (Fig. 3, Table 2, and SI Fig. 8) can A2-BH7, one copy of 10q22-qter is regained relative to DB-A2 contribute to its invasive phenotype. cDNA microarray and North- (Table 2 and SI Fig. 8) and the chromosome 10 transcriptome ratio ern blotting indicated that the most significantly up-regulated gene of this region increases to 0.32. In parallel, the expression of genes in 10q22-qter is PLAU (Fig. 3 and data not shown). PLAU encodes on 10pter-q22 (but not other parts of the same chromosome) was urokinase-type plasminogen activator (uPA), a well known medi- significantly down-regulated in DB-A2 relative to DB-P (Fig. 1C), ator of HGF/SF-induced cell invasion (23, 24). The role of uPA whereas expression of genes on 10q22-qter were up-regulated in up-regulation in invasive phenotypic conversion in this study was A2-BH7 relative to DB-A2 (SI Fig. 8). These results are consistent confirmed with the uPA inhibitor, B428 (23), which significantly MEDICAL SCIENCES with the reports showing that the transfer of chromosome 10p blocked HGF/SF-induced invasion in A2-BH7 cells (Fig. 3). In suppresses oncogenic activity of glioma cells (19, 20). These data DB-P cells, uPA is HGF/SF-inducible, whereas, in A2-BH7 cells, it also suggest that the genes in these regions of chromosomes 7 and is constitutive, indicating that convergence to uPA expression has 10 are crucial for glioma tumor proliferation and invasion, as occurred through chromosome instability. reported by others (11, 18). Thus, for these two chromosomes, our The loss of chromosomes 13, 14, 20, and 21 in DB-A2 relative to invasive and proliferative tumor cells (6) (SI Table 4) change in DB-P are also associated with the down-regulation of additional content (Tables 1 and 2). The analyses reveal how changes in genes that have been implicated in glioma invasion (Table 3 and SI karyotype and the transcriptome can mediate the gene expression Fig. 9). These genes encode proteins that include extracellular changes necessary for phenotype determination. matrix proteins COL6A1 (collagen type VI, ␣-1) (25), cytoskeleton- We used literature-based searches and identified genes that are binding protein MYL9 (myosin-light polypeptide 9) (26), as well as associated with glioma development. Gene expression comparisons EFNB2 (ephrin-B2) (27). Down-regulation of BMP4 (14q22), between DB-P and DB-A2 cells indicated that three of the genes which promotes invasion and migration in malignant melanoma on chromosome 10p were significantly down-regulated in DB-A2 (28), may contribute to a low invasive phenotype of DB-A2. By (Table 3 and SI Fig. 9) and none was significantly up-regulated. We contrast, the higher expression of these genes in DB-P cells is show that, regardless of how similar the numerical ratios are for the expected to enhance DB-P cell invasion. respective chromosomes and chromosome transcriptomes (SI Ta- Our studies do not exclude any of the other mechanisms of tumor

Gao et al. PNAS ͉ May 22, 2007 ͉ vol. 104 ͉ no. 21 ͉ 8999 Downloaded by guest on September 30, 2021 progression such as gain of function, loss of tumor suppressor Research Genetics 40K Human Clone Set (Huntsville, AL) were function, or epigenetic changes, but aneuploidy can contribute to prepared at the Van Andel Research Institute. Each hybridization malignant progression by explaining how gene expression changes was scanned by using a confocal fluorescent Scan Array Lite are delivered. Thus, genetic or epigenetic alterations, fueled by scanner (PerkinElmer Life and Analytical Services, Boston, MA) chromosome instability, can produce sufficient chromosome diver- equipped with lasers operating at 532 nm (G, green) and 635 nm sity in the tumor cell population to generate, in the proper (R, red). Array features were identified and assigned background- environment, cells with the appropriate phenotype for malignant corrected red and green fluorescence intensity values by using the progression. Our data show that quantitative changes in the chro- GenePix Pro 5.0 image analysis software (Axon, Union City, CA) mosome transcriptome are largely governed by changes in chro- by using the default settings. Adjustment of gene expression values mosome content, which we refer to as the one-to-one rule. The to compensate for experimental biases (normalization) was per- analysis of changes in specific genes and chromosomes is consistent formed by using the within-print tip group-scaling technique as with these changes being responsible for phenotype determination. implemented in the limma BioConductor package (www. Materials and Methods bioconductor.org) for the R environment (30, 31). Before normalization, spots were excluded if they had a signal lower than three Cell Lines and SKY. The origin of the DB-P (DBTRG parental cells), DB-A2, DB-A6, and A2-BH7 cells and the cell growth conditions times the SD of the global array background in either channel. have been described previously (6). For SKY, probe hybridization and detection were carried out according to the protocol provided Gene Expression Data Analysis and the Chromosome Transcriptome. with the SkyPaintTM kit (Applied Spectral Imaging, Midgal Relative gene expression values between subclones were generated Ha´Emek,Israel). Metaphase images generated from cultured cells by subtraction of the mean log2-transformed expression value from were captured by using a COOL-1300 SpectraCube imaging system the measured value for each subclone. Relative gene expression (Applied Spectral Imaging) connected to an Olympus BX51 fluo- values were visualized by plotting the expression value based on the rescence microscope (Olympus, Tokyo, Japan). Analysis was per- corresponding gene’s chromosome mapping location by using the formed by using SkyView software from Applied Spectral Imaging. BioConductor idiogram package (32). The chromosome transcriptome value is determined by quantification of the overall relative Preparation of RNA Samples for cDNA Array Analysis. Cells at 80% of gene expression in each chromosomal region. For this analysis, confluence were serum starved for 24 h and treated with HGF/SF chromosomes were broken into segments based on visual inspec- (100 ng/ml) or without (control) for 3 h. The cells were solubilized tion of the SKY data. The relative gene expression value for every in TRIzol (Gibco/BRL, Grand Island, NY), and total RNA was gene located within each segment was calculated as described prepared according to the manufacturer’s suggestion. Total RNA above. The average relative gene expression value was used to was recovered by precipitation in 2.5 M LiCl solution (Ambion, quantify the difference in the chromosome transcriptome for each Austin, TX). RNA samples were reverse-transcribed and labeled segment. We also examined the data set for genes that had for microarray analysis as described (www.microarray.vai.org/). significant differential expression between the subclones. Discrimi- Experiments were carried out at least in duplicate. Four arrays with nant genes were identified by using the empirical Bayes method as DB-A2 or DB-A6 cells and two arrays with A2-BH7 HGF/SF- implemented in the limma package by using the default settings. treated cells were processed by using the dye-swap-labeling technique (29). In all cases, RNA from DB-P cells cultured under We thank Tony Hunter, George Klein, Stefan Imreh, and Beatrice Knudsen normal conditions was used as a reference. for critical reading of the manuscript and David Nadziejka and Michelle Reed-Bassett for assistance with preparation of the manuscript. This work cDNA Microarray Data Generation. Corning GAPS2 microarray was supported in part by the Michigan Life Sciences Corridor and the Jay slides (Corning, NY) spotted with 19,552 cDNA clones from the and Betty Van Andel Foundation.

1. Nowell PC (1976) Science 194:23–28. 16. Harding MA, Arden KC, Gildea JW, Gildea JJ, Perlman EJ, Viars C, 2. Chambers AF, Groom AC, MacDonald IC (2002) Nat Rev Cancer 2:563–572. Theodorescu D (2002) Cancer Res 62:6981–6989. 3. Rajagopalan H, Lengauer C (2004) Nature 432:338–341. 17. Upender MB, Habermann JK, McShane LM, Korn EL, Barrett JC, Difilip- 4. Barrett MT, Sanchez CA, Prevo LJ, Wong DJ, Galipeau PC, Paulson TG, pantonio MJ, Ried T (2004) Cancer Res 64:6941–6949. Rabinovitch PS, Reid BJ (1999) Nat Genet 22:106–109. 18. Steilen-Gimbel H, Henn W, Kolles H, Moringlane JR, Feiden W, Steudel WI, 5. Duesberg P (2005) Science 307:41. Zang KD (1996) Genes Chromosomes Cancer 16:180–184. 6. Gao CF, Xie Q, Su YL, Koeman J, Khoo SK, Gustafson M, Knudsen BS, Hay 19. Steck PA, Ligon AH, Cheong P, Yung WK, Pershouse MA (1995) Genes R, Shinomiya N, Vande Woude GF (2005) Proc Natl Acad Sci USA 102:10528– Chromosomes Cancer 12:255–261. 10533. 20. Kon H, Sonoda Y, Kumabe T, Yoshimoto T, Sekiya T, Murakami Y (1998) 7. Maher EA, Furnari FB, Bachoo RM, Rowitch DH, Louis DN, Cavenee WK, Oncogene 16:257–263. DePinho RA (2001) Genes Dev 15:1311–1333. 21. Masuelli L, Cutler ML (1996) Mol Cell Biol 16:5466–5476. 8. Giese A, Bjerkvig R, Berens ME, Westphal M (2003) J Clin Oncol 21:1624– 22. Tsuda T, Marinetti MR, Masuelli L, Cutler ML (1995) Oncogene 11:397–403. 1636. 23. Jeffers M, Rong S, Vande Woude GF (1996) Mol Cell Biol 16:1115–1125. 9. Koochekpour S, Jeffers M, Rulong S, Taylor G, Klineberg E, Hudson EA, 24. Xie Q, Gao CF, Shinomiya N, Sausville E, Hay R, Gustafson M, Shen Y, Resau JH, Vande Woude GF (1997) Cancer Res 57:5391–5398. Wenkert D, Vande Woude GF (2005) Oncogene 24:3697–3707. 10. Coons SW, Johnson PC, Shapiro JR (1995) Cancer Res 55:1569–1577. 25. Chintala SK, Gokaslan ZL, Go Y, Sawaya R, Nicolson GL, Rao JS (1996) Clin 11. Loeper S, Romeike BF, Heckmann N, Jung V, Henn W, Feiden W, Zang KD, Exp Metastasis 14:358–366. Urbschat S (2001) Cytogenet Cell Genet 94:1–8. 26. Murata K, Hirano K, Villa-Moruzzi E, Hartshorne DJ, Brautigan DL (1997) 12. Wolman SR (1986) Cancer Genet Cytogenet 19:129–140. Mol Biol Cell 8:663–673. 13. Phillips JL, Hayward SW, Wang Y, Vasselli J, Pavlovich C, Padilla-Nash H, 27. Nakada M, Niska JA, Miyamori H, McDonough WS, Wu J, Sato H, Berens ME Pezullo JR, Ghadimi BM, Grossfeld GD, Rivera A, et al. (2001) Cancer Res (2004) Cancer Res 64:3179–3185. 61:8143–8149. 28. Rothhammer T, Poser I, Soncin F, Bataille F, Moser M, Bosserhoff AK (2005) 14. Pollack JR, Sorlie T, Perou CM, Rees CA, Jeffrey SS, Lonning PE, Tibshirani Cancer Res 65:448–456. R, Botstein D, Borresen-Dale AL, Brown PO (2002) Proc Natl Acad Sci USA 29. Churchill GA (2002) Nat Genet 32(Suppl):490–495. 99:12963–12968. 30. Yang YH, Dudoit S, Luu P, Lin DM, Peng V, Ngai J, Speed TP (2002) Nucleic 15. Furge KA, Lucas KA, Takahashi M, Sugimura J, Kort EJ, Kanayama HO, Acids Res 30:e15. Kagawa S, Hoekstra P, Curry J, Yang XJ, Teh BT (2004) Cancer Res 31. Smyth GK, Michaud J, Scott HS (2005) Bioinformatics 21:2067–2075. 64:4117–4121. 32. Furge KA, Dykema KJ, Ho C, Chen X (2005) BMC Genomics 6:67.

9000 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0700631104 Gao et al. Downloaded by guest on September 30, 2021