Figure 1.
Breakpoint analysis for discovering novel cancer gene rearrangements.
Schematic depiction of the approach and workflow, demonstrated by example of the rediscovery of a known gene fusion, SET/NUP214, in the T-ALL cell line LOUCY. Various publicly-available and in-house exon microarray and high-density CGH/SNP array experiments were analyzed. RNA breakpoint analysis (RBA) identifies significant transitions in exon expression level, which may reflect elevated expression of exons distal (3′ partner) or proximal (5′ partner) to a gene fusion junction. To identify such transitions a “walking” Student's t-test was applied, comparing expression levels of proximal and distal exons. Candidate rearrangements were subsequently filtered for those disrupting genes of the Cancer Gene Census, with directional orientation (i.e. being the 5′ or 3′ partner) consistent with known rearrangements of that gene. RBA candidates were further filtered using a Bonferroni correction to adjust for multiple t-tests. DNA breakpoint analysis (DBA) screens for intragenic DNA copy number transitions, which may reflect unbalanced chromosomal rearrangements leading to the formation of gene fusions. The fused lasso method (FDR 1%) followed by a copy number smoothing algorithm was applied to identify CNAs. Copy number transitions were filtered for those disrupting any annotated gene and then further filtered for those disrupting genes of the Cancer Gene Census. We included only candidate breakpoints where the directional orientation of the copy number transition was consistent with known rearrangements of that gene. Several candidates were then validated using molecular and cytogenetic approaches. The average numbers of candidate rearrangements per cancer sample are depicted along the left and right panels at various stages of the workflow.
Table 1.
Validated gene fusions and rearrangements.
Figure 2.
Discovery and characterization of CEP85L/ROS1 in angiosarcoma.
(A) RBA of angiosarcoma specimen AS1 reveals an expression breakpoint between exons 34 and 35 of ROS1, suggesting rearrangement. (B) Experimental validation of CEP85L/ROS1 in AS1 by RT-PCR, using primers flanking the gene fusion junction. (C) Predicted structure of CEP85L/ROS1. CEP85L and ROS1 are oriented in the same direction and located ∼1 MB apart within cytoband 6q22. The gene fusion preserves a coiled-coil (CC) domain from CEP85L and the tyrosine kinase (TK) domain of ROS1. Exons are numbered, with untranslated regions depicted in corresponding lighter shades. (D) Break-apart FISH demonstrates rearrangement of ROS1 in angiosarcoma and epithelioid hemangioendothelioma. Co-localizing red and green signals are indicative of normal chr 6 (left panel). AS1 exhibits loss of red signal with multiple green signals indicative of amplification of rearranged ROS1. An epithelioid hemangioendothelioma specimen (EHE10) also exhibits loss of red signal, indicative of unbalanced rearrangement of ROS1. (E) Increased ROS1 expression in angiosarcoma compared to other sarcoma subtypes. Heatmap depicts genes selectively overexpressed in angiosarcoma, identified by supervised analysis. Genes are ordered by rank value of their t-statistic scores. Mean-centered gene expression ratios are depicted by a log2 pseudocolor scale (ratio-fold change indicated). AS: angiosarcoma, DTF: desmoid-type fibromatosis, GCTTS: giant cell tumor-tendon sheath, HPC: hemangiopericytoma, PVNS: pigmented villonodular synovitis, SFT: solitary fibrous tumor, SS: synovial sarcoma, LMS: leiomyosarcoma. *** P = 4.26×10−28 (Student's t-test).
Figure 3.
Discovery of APIP/SLC1A2 in colon cancer.
(A) Array CGH heatmap displaying genomic breakpoints disrupting SLC1A2 in the SNU-C1 colon cancer cell line and the SNU-16 gastric cancer cell line. SNU-16 is known to harbor CD44/SLC1A2 and its array CGH profile is depicted for comparison. Unsmoothed log2 ratios are displayed. (B) Paired-end RNA seq uncovers APIP/SLC1A2 in SNU-C1. A subset of paired-end reads mapping to APIP/SLC1A2 as well as the gene fusion structure are displayed (left panel). The structure of the known gastric cancer gene fusion CD44/SLC1A2 is depicted for comparison (right panel). An internal start codon within exon 2 of SLC1A2 is predicted to initiate translation in both rearrangements. Inset: experimental validation of APIP/SLC1A2 by RT-PCR with primers flanking the gene fusion junction. (C, D) Gene expression profiling depicts high-level expression of APIP in normal colon (C) and overexpression of SLC1A2 in SNU-C1 (D). Mean-centered gene expression ratios are depicted by a log2 pseudocolor scale and ranked in descending order from left to right.
Figure 4.
Identification and characterization of novel RAF1 gene fusions in pancreatic cancer and anaplastic astrocytoma.
(A) Array CGH heatmaps displaying intragenic RAF1 genomic breakpoints identified in the PL5 pancreatic cancer cell line (left panel) and the D-538MG anaplastic astrocytoma cell line (right panel). Unsmoothed log2 ratios are displayed. (B) Identification of ATG7/RAF1 (left) and BCL6/RAF1 (right) in PL5 and D-538MG cells, respectively, by paired-end RNA-seq. A subset of the paired-end reads supporting each gene fusion is displayed. Both gene fusions are in-frame and the RAF1 serine threonine kinase domain (STK) is retained in both fusions. (C) Experimental validation of gene fusions by RT-PCR, using primers flanking the respective gene fusion junction. (D) Western blotting verifies knockdown of ATG7/RAF1 in PL5 following transfection of a RAF1-targeting siRNA pool. ATG7/RAF1 protein levels were monitored using an anti-RAF1 antibody, with anti-GAPDH providing a loading control. (E) Decreased cell proliferation and (F) invasion rates of PL5 following transfection of a RAF1-targeting siRNA pool, compared to transfection of a non-targeting control (NTC) siRNA pool. ** P<0.01 (two-sided Student's t-test). (G) Break-apart FISH demonstrates rearrangement of BRAF in a pancreatic cancer case from the TMA, as evidenced by physical separation of the red and green probes (arrows) flanking BRAF (single interphase nucleus shown).
Figure 5.
Discovery and characterization of EWSR1/CREM in melanoma.
(A) Array CGH heatmap displaying intragenic EWSR1 breakpoints identified in the SH-4 and CHL-1 melanoma cell lines. (B) Paired-end RNA-seq identification of EWSR1/CREM in CHL-1. Paired-end reads supporting the rearrangement are depicted along with the predicted gene fusion structure. CREM contributes a basic leucine zipper motif (ZIP), while EWSR1 contributes the EWS Activation Domain (EAD). (C) RT-PCR verification of EWSR1/CREM in CHL-1. (D) Quantitative RT-PCR using primers flanking the gene fusion junction verifies EWSR1/CREM knockdown following transfection of an siRNA pool targeting the 3′ end of CREM. (E, F, G) Transfection of CHL-1 with CREM-targeting siRNA pool results in (E) decreased cell proliferation, (F) decreased invasion, and (G) a higher fraction of senescent cells, compared to non-targeting control (NTC). **P<0.01 (two-sided Student's t-test).
Figure 6.
Identification and characterization of FAM133B/CDK6 in J.RT3-T3.5.
(A) Heatmap depicting rearrangement of CDK6 in J.RT3-T3.5 (Jurkat derivative). (B) Discovery of the FAM133B/CDK6 rearrangement by paired-end RNA-seq. The fusion junction was confirmed by RT-PCR (not shown) and Sanger sequencing. (C) Gene expression profiling reveals high-level expression of CDK6 in J.RT3-T3.5 compared to other leukemia cell lines. Note that array probes mapped to the portion of CDK6 retained in the fusion. (D) Jurkat demonstrates marked sensitivity to the CDK4/6 inhibitor PD0332991 (IC50 = 0.27 µM). K562, which expresses only wildtype CDK6, is used as a negative control cell line and shows minimal sensitivity to PD0332991 (IC50 = 5.9 µM).
Figure 7.
DBA discovery of recurrent rearrangements of CLTC and VMP1 across diverse cancer types.
(A) Heatmap depicting focal deletions between CLTC and VMP1 in the breast cancer cell lines BT-549 and HCC1954. (B) Discovery of the recurrent CLTC/VMP1 rearrangement in BT-549 (left panel) and HCC1954 (right panel) by paired-end RNA-seq. (C) RT-PCR verification of CLTC/VMP1 fusion in BT-549 and HCC1954. (D) Heatmap depicting focal deletions disrupting CLTC, PTRH2 and/or VMP1 in various cancer types (see legend). (E) A renal cell carcinoma line, RXF393, was also profiled by exon microarray where an expression breakpoint was evident within CLTC. ***P<10−9 (Student's t-test).
Figure 8.
Discovery of new cell line models for the known rearrangements, EGFRvIII and FIP1L1/PDGFRA.
(A) Heatmap depicting genomic breakpoints within EGFR in the glioblastoma cell lines, CAS-1 and DKMG. (B) Identification of EGFRvIII in DKMG cells by paired-end RNA-seq. Paired-end reads supporting the rearrangement are depicted. (C) Verification of EGFRvIII expression by RT-PCR (top panel) and Western blotting (bottom panel) in DKMG. RT-PCR was done using primers flanking the exon 1/exon 8 junction of EGFRvIII, and Western blotting was done using an antibody specific to the EGFRvIII isoform. Control samples include U87 glioblastoma cells without EGFR rearrangement, U87-vIII cells engineered to express exogenous EGFRvIII, and A431 epidermoid carcinoma cells with EGFR amplification. (D) RBA identification of expression-level breakpoint within PDGFRA in SUPT13 T-ALL cells. ***P<10−11 (Student's t-test). (E) RNA-seq identification of FIP1L1/PDGFRA. (F) RT-PCR validation of FIP1L1/PDGFRA expression in SUPT13. (G) SUPT13 cells are sensitive to imatinib (IC50 = 0.036 µM). K562 is a positive control CML cell line harboring BCR/ABL1 with known sensitivity to imatinib (IC50 = 0.18 µM).