Oncogene-Mediated Alterations in Chromatin Conformation

Oncogene-mediated alterations in chromatin conformation

David S. Rickmana,1, T. David Soongb,1, Benjamin Mossa, Juan Miguel Mosqueraa, Jan Dlabalb, Stéphane Terrya, Theresa Y. MacDonalda, Joseph Tripodic, Karen Buntingd, Vesna Najfeldc, Francesca Demichelisb,e, Ari M. Melnickd,f, Olivier Elementob,2, and Mark A. Rubina,f,2,3

Departments of aPathology and Laboratory Medicine and dMedicine, bDepartment of Physiology and Biophysics, HRH Prince Alwaleed Bin Talal Bin Abdulaziz Alsaud Institute for Computational Biomedicine, and fWeill Cornell Cancer Center, Weill Cornell Medical College, New York, NY 10065; cTumor Cytogenetics Laboratory, Department of Pathology, Tisch Cancer Center, Mount Sinai School of Medicine, New York, NY 10029; and eCentre for Integrative Biology, University of Trento, 38122 Trento, Italy

Edited* by Eric S. Lander, The Broad Institute of MIT and Harvard, Cambridge, MA, and approved April 24, 2012 (received for review August 3, 2011) Emerging evidence suggests that chromatin adopts a nonrandom Results 3D topology and that the organization of genes into structural ERG Overexpression Is Associated with Chromatin Topology. To test hubs and domains affects their transcriptional status. How chroma- our hypothesis, we used stable isogenic, normal benign prostate tin conformation changes in diseases such as cancer is poorly un- epithelial cell lines (RWPE1) (21) that differ with respect to derstood. Moreover, how oncogenic transcription factors, which ERG overexpression (17) (Fig. S1A). To test whether ERG bind to thousands of sites across the genome, influence gene reg- overexpression is associated with global changes in chromatin ulation by globally altering the topology of chromatin requires structure, we performed unbiased chromatin interaction mapping further investigation. To address these questions, we performed using the Hi-C technique (22) from both RWPE1-ERG and unbiased high-resolution mapping of intra- and interchromosome RWPE1-GFP cells, with biological replicates (Fig. 1A and Fig. interactions upon overexpression of ERG, an oncogenic transcrip- S1A). Successful fill-in and ligation were determined as previously tion factor frequently overexpressed in prostate cancer as a result of reported (22) by testing for a known interaction between two a gene fusion. By integrating data from genome-wide chromosome distant genomic loci located on chromosome 6 (23) (Fig. S1B). conformation capture (Hi-C), ERG binding, and gene expression, we The Hi-C libraries were paired-end sequenced using an Illumina demonstrate that oncogenic transcription factor overexpression is GAIIx platform. Following alignment to the human genome associated with global, reproducible, and functionally coherent (hg18) and filtering to remove unligated and self-ligated DNA, changes in chromatin organization. The results presented here have we identified intrachromosomal (or cis-) and interchromosomal broader implications, as genomic alterations in other cancer types (or trans-) interactions in both RWPE1 cell lines. Correlation frequently give rise to aberrant transcription factor expression, e.g., α matrices obtained from independent biological replicates were EWS-FLI1, c-Myc, n-Myc, and PML-RAR . highly similar in both cell lines (Pearson’s correlation coefficient r = 0.99), indicating that the interaction patterns are highly re- ounting evidence suggests that many genes dynamically producible. Biological replicate interactions were therefore com- Mcolocalize to shared nuclear compartments that favor gene bined for further analyses. In total, we identified 18.4 million activation or silencing (1–3). As demonstrated by chromosome intrachromosomal (or cis-) and 18.3 million interchromosomal (or conformation capture (3C) (4), ligand-bound androgen recep- trans-) interactions in RWPE1-ERG cells (Dataset S1). We also tors (AR) and estrogen receptors mediate looped chromatin identified 16.9 million cis- and 18.6 million trans-interactions in structures resulting in coordinated transcription of target genes RWPE1-GFP cells. To visualize global patterns of cis-interactions, we binned the (5, 6). In embryonic carcinoma cells, the PolyComb complex – subunit EZH2 represses some of its target genes via the for- genome (chromosomes 1 22, X) into 1-Mb intervals and calcu- lated the ratio between observed and expected number of mation of similar looped chromatin structures (7). Trans-inter- interactions connecting each interval pair. We then generated actions that regulate gene expression have also been reported correlation heat maps depicting the extent to which pairs of – (8 10). These data suggest that oncogenic transcriptional regu- genomic intervals interact with the same intervals, on the basis of lators are capable of inducing changes in chromatin structures. the assumption that if two loci are close in nuclear space, their These studies have mainly focused on local chromatin structures, patterns of interaction with other loci should be highly corre- and it is still unclear whether more global changes occur in the lated. Confirming prior observations (22), the heat maps showed process of oncogene-mediated transformation. A broader im- plication of these observations is that global chromatin organization changes could impact functional and phenotypic aspects Author contributions: D.S.R., T.D.S., O.E., and M.A.R. designed research; D.S.R., T.D.S., B.M., J.D., S.T., T.Y.M., J.T., V.N., and O.E. performed research; T.D.S., K.B., F.D., A.M.M., of cancer. and O.E. contributed new reagents/analytic tools; D.S.R., T.D.S., B.M., J.M.M., J.D., S.T., To globally investigate oncogene-mediated chromatin struc- J.T., V.N., O.E., and M.A.R. analyzed data; and D.S.R., T.D.S., J.D., S.T., O.E., and M.A.R. ture changes we focused on ERG, the ETS-family transcription wrote the paper. factor most frequently rearranged and overexpressed in prostate The authors declare no conflict of interest. cancer through the TMPRSS2-ERG and other gene fusions in- *This Direct Submission article had a prearranged editor. volving androgen-responsive promoters (11–13). ERG interacts Freely available online through the PNAS open access option. MEDICAL SCIENCES with several cofactors (14) and other transcription factors in- Data deposition: Hi-C and ChIPseq data reported in this paper have been deposited in the Gene Expression Omnibus (GEO) database, www.ncbi.nlm.nih.gov/geo (accession no. cluding AR to regulate the expression of thousands of genes that GSE37752). favor dedifferentiation, cell invasion, and neoplastic transforma- 1D.S.R. and T.D.S. contributed equally to this work. – tion of prostate epithelium when overexpressed (15 20). We 2O.E. and M.A.R. contributed equally to this work. therefore hypothesized that changes in global gene expression 3To whom correspondence should be addressed. E-mail: [email protected]. induced by ERG overexpression could be associated with global This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. changes in the 3D structure of chromosomes. 1073/pnas.1112570109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1112570109 PNAS | June 5, 2012 | vol. 109 | no. 23 | 9083–9088 Downloaded by guest on October 2, 2021 A + ERG abrogated in the presence of ERG (P = 0.01, Fisher’s exact test; Fig. S3D). To test this result, we performed targeted PCR from Remove x-link 3C libraries (3C-PCR) generated from asynchronous cells from Fill (biotin-dCTP) Shear/ Paired-end X-link Hind III Digested Chromatin Ligate Ligated DNA/ pull down Hi-C sequence Hi-C the two cell lines. Consistent with the Hi-C data, the interaction chromatin prep protein library Dataset prep between the promoters of TMPRSS2 and TFF3 was enriched in - ERG control cells (RWPE1-GFP) and depleted in the ERG overexpressing cells (Fig. S3D). One of the most statistically significant cis-interaction changes involved two distant regions on chromosome 6 (regions between B p22.3–12.3 and q22.31–25.32); these regions interact significantly more in cells overexpressing ERG compared with control cells (Fig. 2A and Fig. S3A). We also observed loss of interactions in RWPE1-ERG cells [e.g., chromosome 10 (regions between p15.3– p13 and q21.3–25.1, Fig. S3B) and chromosome 13 (regions between q12.11–13.1 and q13.1–21.1, Fig. S3C)]. Consistent with previous findings (22), we detected more trans-interactions between small, gene-rich chromosomes (e.g., chromosomes 16–22) than between larger chromosomes in both RWPE1-ERG and − − RWPE1-GFP cells (P =1.4× 10 9 and P =4.1× 10 6,respectively, Mann–Whitney tests; Fig. S4A). Moreover, specific regions of chromosomes tended to interact preferentially on the basis of high-resolution heat maps of normalized interactions between all cytogenetic bands (Fig. S4B). RWPE-1 cells have been reported to be primarily diploid and to harbor specific translocations involving chromosomes 9, 12, 18, and 21 in at least 90% of the cells (21). To determine whether additional, unreported translocations accounted for the enriched Hi-C interactions, we performed spectral karyotyping of both cell lines (24). The kar- yotypes observed in the two cell lines were identical, barring a der (13)t(13;15)(q12;q12) unique to ERG overexpressing cells (Fig. S5A), and also detectable using chromosome painting (Fig. S5B). Accordingly, our Hi-C data show a specific increase in interaction between chromosome 13 and chromosome 15 in RWPE1-ERG Fig. 1. Specific cis-interactions that depend on ERG expression. (A) Sche- cells (Figs. S4B and S5C). The spectral karyotyping results thus matic representation of the experimental flow of the Hi-C experiments. Red indicate that changes in interchromosomal interactions detected symbols refer to ERG protein and white symbols refer to other DNA-inter- by Hi-C sometimes represent translocations. However, further acting proteins. (B) Circos plot depicting chromosomes 1–22 and X anno- inspection of the Hi-C and spectral karyotype data shows that tated for ERG binding on the basis of ChIP-seq data (normalized reads shown several ERG- or GFP-specific trans-interactions revealed by Hi-C in orange) and 1,266 differentially expressed genes annotated on the basis are unlikely to be translocations. This result includes Hi-C–pre- of RNA-seq data, indicating over- (red) and underexpressed genes in RWPE1- ERG relative to RWPE1-GFP cells. Lines indicate significantly different cis- and dicted interactions between chromosomes 4 and 21 and between trans-interactions in which line color represents interactions enriched in RWPE- chromosomes 4 and 13 in ERG cells, both of which involve ERG cells (orange) or RWPE-GFP cells (blue). regions that are ∼3 Mb long, and between chromosomes 5 and 9 in both cells, which involve much larger regions of each chromosome (Fig. 1B and Figs. S4B and S5A). plaid patterns, indicating that each chromosome is composed of domains that are either enriched (red to yellow) or depleted ERG Binding Is Enriched at Loci Whose Interactions Differ Between the (green) for cis-interactions (Fig. S2). Cell Types. To characterize ERG binding and ERG-mediated gene The patterns of enriched and depleted cis-interactions were expression changes in these cells, we performed chromatin im- predominantly similar between the two cell lines for each chro- munoprecipitation combined with high-throughput sequencing mosome but some differences were visible (Fig. S3 A–C). Be- (ChIP-seq) and RNA sequencing (RNA-seq). ERG was bound to cause gains and losses of a correlation can be due to gains and 6,398 sites in RPWE1-ERG cells, of which 23.4% were associated ± losses of other interactions shared by the two loci, we wanted to with promoters ( 2 kb from transcription start sites), 30.8% of > quantitatively identify differences in cis- and trans-interactions ERG binding sites were intronic, and 42.9% occurred 2 kb away B between the two cell lines. For this, we evaluated the statistical from known genes (Fig. 2 ). Unbiased DNA motif analysis (25) significance of the differences in interaction counts, using Fish- detected the ETS response element as the most represented element within the ERG peaks (Fig. S5D and Dataset S2). ERG was er’s exact tests coupled with Benjamini–Hochberg corrections bound to its own promoter as well as to known ERG target genes (SI Computational Analysis). Using an adjusted P-value threshold in prostate, e.g., ADRB2, CDH1, DAB2IP, FKBP5, SLC45A3, <0.05 (5% FDR), we identified 8,611 cis- and 86 trans-inter- LAMC2, KCNS3,andSERPINE1 (18, 20), and in nonprostate actions with significant differences between the two cell lines ICAM2 CDH5 B cis tissue, e.g., and (26). Importantly, our results (Fig. 1 ). Among the 8,611 -interactions, 2,501 and 6,110 indicate that hotspots of differential chromatin interaction, i.e., increased and decreased in proximity in RWPE1-ERG vs. loci that have the most differential interactions with other loci, are − RWPE1-GFP cells, respectively. Interestingly, we found a dif- highly enriched in ERG binding (P < 10 11;Mann–Whitney test; ferential interaction between loci containing two genes highly Fig. 2C and Fig. S5E). relevant for prostate cancer, TMPRSS2 (the most common ERG fusion partner and itself an ERG-target gene) (20) and TFF3 ERG-Associated Chromatin Topological Changes Are Associated with (a gene that is ∼800 kb away from TMPRSS2 and also an ERG Coordinated Gene Expression. We next investigated the association target gene) (17) in RWPE1-GFP cells. This interaction was between gene expression and chromatin interactions. On the basis

9084 | www.pnas.org/cgi/doi/10.1073/pnas.1112570109 Rickman et al. Downloaded by guest on October 2, 2021 ERG GFP ERG - GFP A Chr. 6 Chr. 6 Chr. 6

Fig. 2. (A) Correlation matrices for chromosome 6 obtained from RWPE1- ERG cells (Left)andRWPE1-GFPcells (Center) and the difference between the − Chr. 6 Chr. 6 Chr. 6 Chr. data from two cell lines [(RWPE1_ERG) (RWPE1_GFP)] (Right). A schematic of chromosome 6 is given (Above and Left 1.0 of each matrix). (B) Pie chart of the 6,398 1.0 ERG ChIP-seq peaks corresponding to the distribution of ERG binding relative to -1.0 -1.0 the nearest gene in RPWE1-ERG cells. Binding was broken down into the fol- B C 2.0 lowing four categories: within promoters 1.9 (±2 kb from transcription start sites), ′ 1.8 intronic, 3 from the end of the last exon (3′ End), and distal (>2kbawayfrom Promoter (<= 2kb) 1.7 (23.4%) 3’ End known genes). (C) Graph showing the Distal (> 2Kb) (1.2%) 1.6 (42.9%) amount of ERG binding (kb) at each of Exon 1.5 the ranked (N) differential interacting (1.7%) ERG binding (Kb) Intron 1.4 chromatin loci of 1-Mb regions plus (30.8%) 1.3 a 1-Mb region ﬂanking either side (3 Mb 05001000 1500 2000 total window size) showing the top in- Top N hubs of differential interactions teraction loci are strongly enriched with ERG binding sites.

of paired-end RNA-seq data from both cell lines, 1,266 genes were gene cluster members PYGO1 and NKX3.1) were also signifi- differentially expressed (more than twofold change), indicating cantly enriched (adjusted P = 0.004). This result is consistent with that ERG is highly active in RWPE1-ERG cells and induces global ERG expression disrupting the normal lineage-specific prostate gene expression changes. Similar to Lieberman-Aiden et al. (22), cell differentiation and promoting an EZH2-mediated dedif- we observed that the most highly expressed genes are in closer ferentiation program (20). Cluster analysis using RNA-seq ex- nuclear proximity than expected by chance, comparing randomly pression of the interacting gene set partitioned 50 human prostate − selected genes from both cell lines (P < 10 299 in GFP and ERG; cancer samples into subgroups that are associated with ERG Fig. S6A). This observation would support the view that active rearrangement status (Fig. S7, P = 0.0195, Fisher’s exact test), genes reside in transcription-prone nuclear areas, i.e., transcrip- suggesting that gene expression changes associated with ERG- tion factories or hubs. We also found that the 1,266 differentially associated cis- and trans-interactions are highly relevant in the expressed genes between ERG and GFP cells were in close context of aggressive prostate cancer. − proximity in ERG cells but also in GFP cells (P < 10 11, Fig. S6 B Many of the genomic loci containing differentially expressed and C). However, we found that 777 (65%) of the 1,266 ERG- genes are spatially interconnected, as shown using a network regulated genes are mapped to differentially interacting genomic representation (Fig. S8). In Fig. S8, nodes correspond to 1-Mb regions between ERG and GFP cells (Fig. 3A). This set of 777 genomic regions containing the differentially expressed gene(s) genes is hereafter denoted as the “interacting gene set.” The other and the lines connecting the regions (edges) indicate interactions 489 (35%) genes were not found to be significantly interacting or either enriched in RWPE1-ERG (orange edges) or depleted showed no difference in interaction between ERG and GFP cells (blue edges) compared with RWPE1-GFP. Node color and (Fig. 3A). On the basis of Gene Ontology analyses (Fig. 3B and shape indicate expression status of the indicated gene(s) in Dataset S3) the interacting gene set was significantly enriched RWPE1-ERG relative to RWPE1-GFP cells and ERG binding, with genes encoding for proteins that promote cell motility and respectively. As shown in Fig S8A, the network of differentially invasion (adjusted P = 0.017, Fisher’s exact test), which are interacting genes associated with ERG overexpression consists of phenotypes induced by ERG overexpression (16, 18). Genes as- several unconnected smaller subnetworks. One subnetwork (Fig. sociated with urogenital development (e.g., HOXA, -B, and -C 4 A and B) linked several genes located throughout chromosome

1,266 ERG-regulated genes Enriched categories of ERG-mediated interacting genes Fig. 3. (A) Pie chart of the 1,266 ERG-differ- A B Adjusted P value 0.03 0.02 0.01 < 0.0001 entially expressed genes as a function of sig- ﬁ Cell adhesion ni cantly interacting on the basis of the Hi-C data. These genes are divided into two groups: Skeletal syst. developement the interacting genes (housed in loci that are

significantly interacting differently in the ERG MEDICAL SCIENCES Urogenital syst. developement Non- compared with the control cell lines) and the interacting Cell migration noninteracting genes (housed in loci that are (35%) not significantly interacting differently in the Regionalization Interacting ERG compared with the control cell lines). (B) (65%) Extracell. struct. organization Gene Ontology (GO) analysis results showing a subset of the significant categories (adjusted P Localization of cell value <0.05), using the interacting gene set. There were no significant categories based on Embryonic morphogenesis the noninteracting gene set.

Rickman et al. PNAS | June 5, 2012 | vol. 109 | no. 23 | 9085 Downloaded by guest on October 2, 2021 A RWPE1-GFP chrom. 6 cis-interacting loci RWPE1-ERG chrom. 6 cis-interacting loci

* Anchor B Loci 7 645 3-1

SERPINB9 Chrom. 6 FYN HEY2 MOXD1/OR2A4

C 3C Loci targeted by PCR 1:7 1:61:5 1:4 1:3 1:2 12 ERG 10 GFP Overlap 8

4 Loci 1 Loci 7 Loci 6 Loci 5 Relative Frequency of interaction to GFP 2

0 129.96 20.60 6.65 5.50 1.24 0.0035 Distance from anchor (Mb) Distant

mRNA by qRT-PCR 16.0

14.0 ERG GFP Loci 1 Loci 7 Loci 6 Loci 5 12.0 P value < 0.0001 P value = 0.014 10.0 100 100

8.0 80 80

6.0 60 60

4.0 40 40

2.0 20 Fold change relative to GFP to change relative Fold 20 0.0 % nuclei with overlap 0 % nuclei with overlap 0 SERPINB9 FYN HEY2 MOXD1 ERG GFP ERG GFP

Fig. 4. ChIP-seq, RNA-seq, and Hi-C data integration. (A) Network of genes (visualized using Cytoscape) (35) that are represented in independent cis-interactions between 1-Mb chromosome loci on chromosome 6 and that are signiﬁcantly differentially expressed between RWPE-GFP (Left) and RWPE-ERG (Right) cells. Node color represents over- (red) and underexpressed (green) genes in RWPE-ERG cells relative to RWPE-GFP cells. Pink indicates that the genes in the node are both over- and underexpressed in RWPE-ERG cells. Node shapes are deﬁned by ERG binding peak calls from ChIP-seq data [triangle, bound to promoter; diamond, bound to gene body; square, bound to distal region (between 2 kb and 50 kb); circle, bound to far distal region (> 50 kb)]. Edge (lines connecting the nodes) color represents interaction (from Hi-C) enriched in RWPE-ERG cells (orange) or RWPE-GFP cells (blue). (B and C) Validation of a subset of cis-interactions on chromosome 6 that involved ERG-regulated genes (MOXD1, HEY2, FYN, and SERPINB9). (B) Schematic map of chromosome 6 showing a subset of interactions and deregulated genes from A is shown with loci and anchor regions indicated. (C)(Left) The mean frequencies of interactions for the indicated loci pair based on PCR products obtained from two independent 3C libraries (above) and mRNA levels of the indicated genes (below) generated from either RWPE1-ERG (orange) or RPWE1-GFP (blue). SDs are shown as error bars. (Upper Right) Representative FISH images of G1-enriched nuclei based on analyses using probes made from BACs housing, to the left, MOXD1 (loci 1, red) and SERPINB9 (loci 6, green) or, to the right, FYN (loci 6, red) and HEY2 (loci 5, green), show close or distant proximity of the two loci. (Lower Right) Graphical representation of the percentage of nuclei having overlapping signals from RWPE1-ERG (orange bars) or RWPE1-GFP (blue bars) cells based on 251 and 304 (MOXD1-SERPINB9) nuclei or 150 and 150 nuclei (HEY2-FYN) evaluated, respectively (P value based on a two-tailed Fisher’sexact test is shown above).

6 and includes FYN, a known prostate cancer oncogene with network analysis revealed a variety of patterns of interaction and a role in cell invasion (27, 28), and MOXD1, a gene up-regulated expression changes that are consistent with ERG-associated in primary compared with metastatic prostate cancer (29). These formation or disappearance of transcription or repression fac- genes are bound by ERG proximally or distally (Fig. S8B). This tories. For some genes, increase in proximity correlated with

9086 | www.pnas.org/cgi/doi/10.1073/pnas.1112570109 Rickman et al. Downloaded by guest on October 2, 2021 increased expression, consistent with the formation of an active with ERG overexpression (Fig. 5C), we did not find any difference transcription factory. For example, the MOXD1 gene specifically in interaction between MOXD1 and FYN in DU145 cells as we interacts with FYN, SERPINB9, and ARMC2 in RWPE1-ERG found in RWPE1-ERG cells. This result indicates that a subset of cells, and all of these genes are overexpressed in these same cells. ERG-associated interactions is also dependent on cellular context In other cases, a decrease in interaction in RWPE1-ERG cells and genetic background and does not always form upon ERG was associated with increased expression, consistent with po- overexpression. tential escape from a repression factory. For example, MOXD1 and HEY2, SYNE1 are farther apart in RWPE1-ERG cells than in Discussion RWPE1-GFP cells and these genes are up-regulated in RWPE1- This study provides the characterization of the correlation between ERG cells, similar to what we observed for TMPRSS2 and TFF3. the overexpression of an oncogenic transcription factor (ERG) We then sought to validate a subset of these interactions using and changes to the 3D chromosome structure. It has been pro- fluorescent in situ hybridization (FISH) and 3C-PCR. We also posed that ETS-rearranged prostate cancer, similar to other validated expression changes of genes located within these regions, translocation-associated tumors, represents a distinct molecular using quantitative RT-PCR. Using the MOXD1 genomic locus as subclass of prostate cancer (12). Our data show that overexpres- the anchor point, 3C-PCR data revealed an increased number of sion of ERG is associated with broad chromatin topology changes interactions involving distant loci enriched in RWPE1-ERG cells and, interestingly, that ERG binding is significantly associated with (20.6 Mb, MOXD1-FYN interactions; and 129.69 Mb, MOXD1- hotspots of differential chromatin interactions. The shift in chro- SERPINB9 interactions; Fig. 4 B and C) or in the RWPE1-GFP matin territories that may be associated with differential chromatin cells (6.65 Mb, MOXD1-HEY2 interactions). FISH analysis also interactions appears to correlate with changes in the expression of confirmed the enriched interaction between loci containing a subset of genes relevant to aggressive prostate cancer. However, MOXD1 and SERPINB9 or FYN and HEY2 in the RWPE1-ERG this correlation does not directly imply causality as the associations cells compared with RWPE1-GFP cells. We then sought to de- observed could be resultant of clonal selection of the cell lines termine whether certain chromosomal interactions are dynamically assayed. Further experiments will be required to characterize more orchestrated during the cell cycle. FISH analyses on synchronized precisely the subset of ERG-associated interactions that are di- cell lines showed significantly enriched interactions between regions rectly caused by ERG binding. Such characterization could be associated with MOXD1 and SERPINB9 in cells enriched during the achieved, for example, using an inducible system in which the G1 phase (P < 0.0001, two-tailed Fisher’s exact test) compared with timing of ERG expression can be tightly controlled and alterations fi the G2/M phase (P = 0.5004, Fig. S9). These results confirm the cell- in proximity and DNA binding are quanti ed concurrently. cycle dependency of some of the interactions initially identified by Of note, combining our Hi-C data with spectral karyotyping Hi-C from unsynchronized cells. We then wanted to know whether and chromosome painting led to the detection of a translocation ERG-associated interactions could be observed in another isogenic between chromosomes 13 and 15 that is specific to ERG-over- cell type overexpressing ERG. For this analysis we generated two expressed RWPE-1 cells. Moreover, we found that ERG binding independent 3C libraries from DU145 prostate cancer cells, with is enriched at several loci in chromosomes 13 and 15 whose in- (DU145-ERG) or without (DU145-GFP) stable ERG over- teraction status changed as a function of ERG overexpression expression (Fig. 5A). As a control, we generated a DU145-ERG compared with other genomic regions, e.g., two regions in q14.12 cell line stably expressing an shRNA that targets ERG mRNA and q22.1 on chromosome 13 (P = 0.024, Mann–Whitney test; (DU145-ERG shRNA-ERG). As observed in the RWPE1 cells, we Fig. S5C). It has been shown that ERG induces DNA damage found that ERG expression is associated with increased interaction (30) and that there is a positive correlation between the location of between the chromosome 6 loci containing MOXD1 and SER- breakpoints in ETS-rearranged tumor cells and transcriptional or PINB9 and with the overexpression of both genes (Fig. 5B). chromatin compartments of active genes (31). This observation is Knocking down the ERG protein in the DU145-ERG shRNA- intriguing given that others have shown that active genes (1, 22) ERG cells led to decreased interaction between these loci and and translocated genes (2) are in close spatial proximity. Data abrogated the overexpression of both genes, providing further ev- from a recent study combining Hi-C and high-throughput genome- idence of the dynamic nature of ERG-associated chromosome wide translocation sequencing in G1-arrested, mouse pro-B cells interactions. Although FYN was overexpressed in conjunction suggest that spatial proximity may indeed guide translocations

A B 3C Chrom. 6 Loci targeted by PCR C mRNA by qRT-PCR 50 1:6 1:5 1:2 18.0 40 10 DU145-ERG DU145-ERG 30 DU145-ERG shRNA-ERG DU145-ERG shRNA-ERG DU145-GFP 14.0 20 8 DU145-GFP 10 ERG mRNA fold change ERG mRNA fold 0 6 DU145 ERG ERG ERG GFP 10.0 shRNA-ERG -++- 4 ERG protein 6.0 Relative Frequency of interaction to GFP 2 tubulin protein 2.0 Fold change relative to DU145-GFP to change relative Fold

0 MEDICAL SCIENCES 129.96 20.60 0.0035 SERPINB9 FYN MOXD1 Distance from anchor (Mb)

Fig. 5. Validation of speciﬁc cis-interactions in DU145 prostate cancer cell lines overexpressing ERG alone or in the presence of shRNA targeting ERG mRNA. (A) ERG mRNA (Upper) and protein (Lower) levels measured by quantitative RT-PCR (17) and Western blot analysis in DU145 cells stably overexpressing ERG alone or in the presence of stable expression of shRNA targeting the ERG mRNA (two independent cell line clones are shown) or GFP as a control. shRNA expression led to a drastic reduction in transcript and protein levels. (B) Validation of a subset of cis-interactions on chromosome 6 that involved ERG-regulated genes (MOXD1, FYN, and SERPINB9) in the DU145 cell lines based on the primers described in the Fig. 3 legend (MOXD1 region is the anchor, and FYN regions are located 20.6 Mb and 129.96 Mb from the anchor region, respectively). (C) Validation of the overexpression of ERG-regulated genes (MOXD1, FYN, and SERPINB9) in the DU145 cell lines.

Rickman et al. PNAS | June 5, 2012 | vol. 109 | no. 23 | 9087 Downloaded by guest on October 2, 2021 (32). Our data support the notion that conformational proximity RARα. We hypothesize that overexpression of other oncogenic associated with overexpression of an oncogenic transcription fac- transcription factors also alters 3D chromatin structure in dif- tor may play an important role in translocation partner choice. ferent cellular contexts. Taken together, these data—although currently correlative—lead us to speculate that ERG overexpression may favor the formation Methods of secondary genomic alterations. Supporting this hypothesis, the Human Cell Lines. RWPE1 and DU145 cells were obtained from ATCC and TMPRSS2-ERG gene fusion is an early event in prostate cancer maintained according to the manufacturer’s protocol. Isogenic cell lines and is almost always associated with other genomic alterations. generated to overexpress either truncated ERG (most commonly encoded Nonetheless, it is important to remember that the observed cor- isoform based on TMPRSS2-ERG fusion) have been previously described (17). relation between ERG overexpression and a novel translocation does not imply causality. Moreover, although this translocation Hi-C Library Generation. Fifty million RWPE1-ERG or RWPE1-GFP cells were fi was observed in independent clonal populations of RWPE1-ERG xed and processed to generate Hi-C and 3C libraries as previously reported (22). Briefly, cells were cross-linked and the chromatin was digested with cells, we cannot entirely exclude the possibility that this trans- fi fi location did not arise through an incidental clonal selection of a HindIII, ligated after ll-in with biotin-conjugated dCTP, and puri ed using streptavidivin-conjugated magnetic beads. The Hi-C libraries were then nonregulated process. Further experiments, e.g., using an in- paired-end sequenced using an Illumina GAIIx platform, resulting in repli- ducible ERG expression system as described above, are required cate-combined 158.5 million and 159.2 million paired-end DNA sequence to determine whether ERG binding is directly involved in medi- reads from RWPE1-ERG and RWPE1-GFP, respectively. Sequences for all ating translocations such as the one observed here. We also note primers used to validate Hi-C and RNAseq data are given in Dataset S4. that, whereas the resolution for detecting rearrangements using our approach (multicolor FISH and G-banding) is ∼0.5–1.5 Mb ChIP-seq. ChIP-seq data from 50 million RWPE1-ERG cells were generated as (24, 33), we cannot entirely rule out translocations accounting for previously described (17). Eight separate chromatin samples were immuno- the trans-interactions involving smaller chromosomal regions that precipitated with rabbit anti-ERG (Epitomics; 2849-1) that we have shown to were enriched in the ERG overexpressing cells. Moving forward, be highly specific for ERG protein expression in prostate cancer (34). high-resolution assessment of structural changes, e.g., using Details regarding all computational approaches and other methods are whole-genome sequencing, should accompany large-scale chro- found in SI Methods and SI Computational Analysis. mosome capture data interpretation to better resolve differences between proximity and genomic rearrangements. ACKNOWLEDGMENTS. The authors thank Naoki Kitabayashi, Kyung Park, Finally, our findings also extend beyond the context of the Aparna Nigam, and Wasay Hussain for their technical support. This work was supported by funding from the Department of Defense New Investigator Award prostate as many driving genetic lesions in other cancer types PC081337 (to D.S.R.), by National Science Foundation CAREER Grant 1054964 (to involve aberrant expression of transcription factors due to ge- O.E.), by National Institutes of Health National Cancer Institute Grant CA125612 nomic alterations, e.g., EWS-FLI1, c-Myc, n-Myc, and PML- (to M.A.R.), and by the Starr Cancer Consortium (F.D. and M.A.R.).

1. Osborne CS, et al. (2004) Active genes dynamically colocalize to shared sites of on- 19. Tomlins SA, et al. (2007) Integrative molecular concept modeling of prostate cancer going transcription. Nat Genet 36:1065–1071. progression. Nat Genet 39:41–51. 2. Osborne CS, et al. (2007) Myc dynamically and preferentially relocates to a transcrip- 20. Yu J, et al. (2010) An integrated network of androgen receptor, polycomb, and tion factory occupied by Igh. PLoS Biol 5:1763–1772. TMPRSS2-ERG gene fusions in prostate cancer progression. Cancer Cell 17:443–454. 3. Sexton T, Schober H, Fraser P, Gasser SM (2007) Gene regulation through nuclear 21. Roh M, Franco OE, Hayward SW, van der Meer R, Abdulkadir SA (2008) A role for organization. Nat Struct Mol Biol 14:1049–1055. polyploidy in the tumorigenicity of Pim-1-expressing human prostate and mammary 4. Dekker J, Rippe K, Dekker M, Kleckner N (2002) Capturing chromosome conforma- epithelial cells. PLoS ONE 3(7):1–11. tion. Science 295:1306–1311. 22. Lieberman-Aiden E, et al. (2009) Comprehensive mapping of long-range interactions 5. Fullwood MJ, et al. (2009) An oestrogen-receptor-alpha-bound human chromatin reveals folding principles of the human genome. Science 326:289–293. – interactome. Nature 462:58 64. 23. Majumder P, Gomez JA, Chadwick BP, Boss JM (2008) The insulator factor CTCF 6. Wang Q, et al. (2007) A hierarchical network of transcription factors governs an- controls MHC class II gene expression and is required for the formation of long-dis- – drogen receptor-dependent prostate cancer growth. Mol Cell 27:380 392. tance chromatin interactions. J Exp Med 205:785–798. 7. Tiwari VK, et al. (2008) PcG proteins, DNA methylation, and gene repression by 24. Schröck E, et al. (1996) Multicolor spectral karyotyping of human chromosomes. Sci- chromatin looping. PLoS Biol 6:2911–2927. ence 273:494–497. 8. Apostolou E, Thanos D (2008) Virus infection induces NF-kappaB-dependent in- 25. Elemento O, Slonim N, Tavazoie S (2007) A universal framework for regulatory ele- terchromosomal associations mediating monoallelic IFN-beta gene expression. Cell ment discovery across all genomes and data types. Mol Cell 28:337–350. 134:85–96. 26. Birdsey GM, et al. (2008) Transcription factor Erg regulates angiogenesis and endo- 9. Schoenfelder S, et al. (2010) Preferential associations between co-regulated genes thelial apoptosis through VE-cadherin. Blood 111:3498–3506. reveal a transcriptional interactome in erythroid cells. Nat Genet 42:53–61. 27. Cai H, et al. (2011) Differential transformation capacity of Src family kinases during 10. Zhao Z, et al. (2006) Circular chromosome conformation capture (4C) uncovers ex- the initiation of prostate cancer. Proc Natl Acad Sci USA 108:6579–6584. tensive networks of epigenetically regulated intra- and interchromosomal inter- 28. Yang X, Dutta U, Shaw LM (2010) SHP2 mediates the localized activation of Fyn actions. Nat Genet 38:1341–1347. downstream of the α6β4 integrin to promote carcinoma invasion. Mol Cell Biol 30: 11. Pﬂueger D, et al. (2009) N-myc downstream regulated gene 1 (NDRG1) is fused to ERG – in prostate cancer. Neoplasia 11:804–811. 5306 5317. ﬁ 12. Rubin MA, Maher CA, Chinnaiyan AM (2011) Common gene rearrangements in 29. Chandran UR, et al. (2007) Gene expression pro les of prostate cancer reveal in- – prostate cancer. J Clin Oncol 29:3659–3668. volvement of multiple molecular pathways in the metastatic process. BMC Cancer 7:1 21. 13. Tomlins SA, et al. (2005) Recurrent fusion of TMPRSS2 and ETS transcription factor 30. Brenner JC, et al. (2011) Mechanistic rationale for inhibition of poly(ADP-ribose) – genes in prostate cancer. Science 310:644–648. polymerase in ETS gene fusion-positive prostate cancer. Cancer Cell 19:664 678. 14. Li R, Pei H, Watson DK (2000) Regulation of Ets function by protein - protein inter- 31. Berger MF, et al. (2011) The genomic complexity of primary human prostate cancer. actions. Oncogene 19:6514–6523. Nature 470:214–220. 15. Hollenhorst PC, et al. (2011) Oncogenic ETS proteins mimic activated RAS/MAPK sig- 32. Zhang Y, et al. (2012) Spatial organization of the mouse genome and its role in re- naling in prostate cells. Genes Dev 25:2147–2157. current chromosomal translocations. Cell 148:908–921. 16. Klezovitch O, et al. (2008) A causal role for ERG in neoplastic transformation of 33. Fan YS, Siu VM, Jung JH, Xu J (2000) Sensitivity of multiple color spectral karyotyping prostate epithelium. Proc Natl Acad Sci USA 105:2105–2110. in detecting small interchromosomal rearrangements. Genet Test 4:9–14. 17. Rickman DS, et al. (2010) ERG cooperates with androgen receptor in regulating trefoil 34. Park K, et al. (2010) Antibody-based detection of ERG rearrangement-positive pros- factor 3 in prostate cancer disease progression. Neoplasia 12:1031–1040. tate cancer. Neoplasia 12:590–598. 18. Tomlins SA, et al. (2008) Role of the TMPRSS2-ERG gene fusion in prostate cancer. 35. Cline MS, et al. (2007) Integration of biological networks and gene expression data Neoplasia 10:177–188. using Cytoscape. Nat Protoc 2:2366–2382.

9088 | www.pnas.org/cgi/doi/10.1073/pnas.1112570109 Rickman et al. Downloaded by guest on October 2, 2021