ADP-ribosylation is an important posttranslational protein modification that regulates diverse biological processes, controlled by dedicated transferases and hydrolases. Here, we show that frequent deletions (∼30%) of the MACROD2 mono-ADP-ribosylhydrolase locus in human colorectal cancer cause impaired PARP1 transferase activity in a gene dosage–dependent manner. MACROD2 haploinsufficiency alters DNA repair and sensitivity to DNA damage and results in chromosome instability. Heterozygous and homozygous depletion of Macrod2 enhances intestinal tumorigenesis in ApcMin/+ mice and the growth of human colorectal cancer xenografts. MACROD2 deletion in sporadic colorectal cancer is associated with the extent of chromosome instability, independent of clinical parameters and other known genetic drivers. We conclude that MACROD2 acts as a haploinsufficient tumor suppressor, with loss of function promoting chromosome instability, thereby driving cancer evolution.
Significance: Chromosome instability (CIN) is a hallmark of cancer. We identify MACROD2 deletion as a cause of CIN in human colorectal cancer. MACROD2 loss causes repression of PARP1 activity, impairing DNA repair. MACROD2 haploinsufficiency promotes CIN and intestinal tumor growth. Our results reveal MACROD2 as a major caretaker tumor suppressor gene. Cancer Discov; 8(8); 988–1005. ©2018 AACR.
See related commentary by Jin and Burkard, p. 921.
This article is highlighted in the In This Issue feature, p. 899
ADP-ribosylation is a widespread posttranslational protein modification at DNA lesions, which is governed by the activities of specific transferases and hydrolases. This modification regulates various biological processes, including DNA-damage response, chromatin reorganization, transcriptional regulation, apoptosis, and mitosis (1–3). Genome-wide DNA copy-number analyses across human cancers have revealed common focal deletions of the MACROD2 mono-ADP-ribosylhydrolase locus on chromosome 20p12.1 in multiple malignancies, including gastric and colorectal cancers (4, 5). However, the locus is considered a tissue-specific fragile site (6), and whether MACROD2 aberrations contribute to carcinogenesis is unknown.
Recent studies have identified MACROD2 as a regulator of PARP1, a principal sensor of DNA single-strand breaks (SSB) and double-strand breaks (DSB; ref. 7). Following binding to sites of DNA nicks or breaks, PARP1 polymerizes PAR chains onto histones and other proteins, including itself. This auto- and substrate-PARylation by PARP1 establishes and amplifies the DNA-damage signal, recruiting repair factors and activating effector proteins involved in the DNA-damage response, including master regulators such as ATM, ATR, and DNA-dependent protein kinase (8). PAR synthesis is reversed by degradation by poly(ADP-ribose) glycohydrolase; however, removal of the terminal autoinhibitory mono-ADP-ribose from PARP1 to cause reactivation requires MACROD2 recruitment and enzymatic activity (7). MACROD2 phosphorylation by ATM acts as a negative feedback loop, triggering MACROD2 nuclear export upon DNA damage, thus temporally restricting its recruitment to DNA lesions (9).
MACROD2 has further been implicated as a regulator of glycogen synthase kinase 3-beta (GSK3β), indicating a role in the modulation of WNT signaling (10). MACROD2 reverses PARP10-catalyzed mono-ADP-ribose-mediated inhibition of GSK3β, activating GSK3β to phosphorylate β-catenin in the context of a protein complex with adenomatous polyposis coli (APC), axin, and other components. Phosphorylation targets β-catenin for ubiquitination and proteasomal degradation, preventing its translocation to the nucleus and interaction with members of the T-cell factor/lymphoid enhancer factor (TCF/LEF) transcription factor family to activate expression of WNT target genes (11).
These observations raise the possibility that somatic genomic aberrations in MACROD2 contribute to cancer development through impairing the DNA-damage response or promoting aberrant WNT signaling. Here, we present genetic, biochemical, and functional data that reveal MACROD2 as a caretaker tumor suppressor gene essential for the maintenance of cancer genome integrity.
Focal Deletions at Chromosome 20p12.1 Target MACROD2 in Human Colorectal Cancer
To comprehensively characterize the human cancer types in which the MACROD2 locus is subject to focal deletions at chromosome 20p12.1, we analyzed The Cancer Genome Atlas (TCGA)–derived DNA copy-number data from 10,575 tumors representing 32 malignancies using Genomic Identification of Significant Targets in Cancer (GISTIC; ref. 12). Colorectal adenocarcinoma (COAD/READ) exhibited the strongest evidence for recurrent focal DNA copy-number loss at MACROD2 (q = 5.89E−78), but targeting was also observed in stomach adenocarcinoma (STAD), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), esophageal carcinoma (ESCA), uterine corpus endometrial carcinoma (UCEC), uterine carcinosarcoma (UCS), lung adenocarcinoma (LUAD), liver hepatocellular carcinoma (LIHC), and thyroid carcinoma (THCA; Fig. 1A; Supplementary Table S1).
To fine-map the spectrum of somatic MACROD2 deletions in human colorectal cancer, we estimated absolute DNA copy-number states at chromosome 20p12.1 for 616 TCGA tumors and an in-house cohort of 651 tumors using single-nucleotide polymorphism (SNP) array data and OncoSNP (top, Fig. 1B; Supplementary Fig. S1; Supplementary Table S2). We detected frequent heterozygous and homozygous loss of variable size on 20p12.1, involving the MACROD2 locus in 27.9% (172/616) of TCGA and 32.3% (210/651) of in-house colorectal cancers. Deletion frequencies were similar across tumor stages (P = 0.084). The majority of loss events (71.5%; 273 of 382) were intragenic microdeletions in MACROD2 excluding any surrounding genes, with most mapping to a region from exons 4 to 5. A similar deletion spectrum was also observed for a panel of 53 human colorectal cancer cell lines (bottom, Fig. 1B; Supplementary Tables S3 and S4), with validation of detected homozygous exonic MACROD2 deletions by the absence of reads in whole-exome sequencing (WES) data (Fig. 1C).
Although 50.5% (138 of 273) of MACROD2 microdeletions involved exonic regions, 49.5% (135 of 273) were limited to introns. To test whether intronic deletions influenced MACROD2 gene expression, we sequenced MACROD2 mRNA transcripts in our colorectal cancer cell lines, thereby avoiding contamination from nontumor cells that would otherwise be present in patient samples. Cell lines wild-type for MACROD2 (n = 20) or with exonic microdeletions (n = 19) were verified to express full-length or deletion transcripts, respectively (Fig. 1C; Supplementary Fig. S2; Supplementary Table S4). Moreover, 66.7% (4 of 6) of cell lines with intronic microdeletions also exhibited aberrant MACROD2 transcripts lacking one or more exons, indicating pathogenicity in these cases. In five cell lines with multiple heterozygous exonic or intronic MACROD2 aberrations, no wild-type transcript was detected consistent with deletions affecting both alleles. Aberrant transcripts were predicted to result in premature MACROD2 protein truncation in 28% (7 of 26) of cases, or in-frame exonic deletions involving the catalytic macrodomain of the protein in 72% (19 of 26) of cases (Supplementary Table S4).
MACROD2 Somatic Mutation Burden in Human Colorectal Cancer
We next examined the contribution of somatic mutations to the burden of MACROD2 aberrations in colorectal cancer. We sequenced all exons of MACROD2 in 102 sporadic colorectal cancers and 53 colorectal cancer cell lines. Detected variants were verified to not correspond to known SNPs, and for primary tumors were confirmed to be somatically acquired by sequencing of matched normal tissue. Results were combined with TCGA-derived mutation data for colorectal cancer (n = 536; Supplementary Table S5).
Somatic MACROD2 mutations were found to be of low prevalence, with 14 missense mutations detected in a total of 691 cases. MACROD2 missense mutations were localized throughout the protein with no apparent clustering (Fig. 1D). Fifty percent (7 of 14) of the mutations were assigned as pathogenic based on the consensus from two in silico prediction algorithms (PolyPhen and SIFT; refs. 13, 14; Supplementary Table S5). Of the seven missense variants that mapped to the available crystal structure of the MACROD2 macrodomain (PDB: 4IQY; ref. 7), three variants (Ala91Ser, Gly100Ser, Thr187Ala) were predicted to interfere with binding of the cocrystalized ADP-ribose moiety to the catalytic pocket (Fig. 1E).
Considering the combined somatic copy-number alteration (SCNA), transcript sequencing, and mutation data for our colorectal cancer cell line, 61.3% (19 of 31) of cases with MACROD2 aberrations showed evidence of “two hits,” consistent with a tumor suppressor role. Nevertheless, 38.7% (12 of 31) of cell lines with MACROD2 aberrations exhibited only “one hit,” suggesting that inactivation of a single copy of MACROD2 may be sufficient to impart a clonal growth advantage during tumor development.
MACROD2 Deficiency Enhances Growth of Intestinal Tumors in a Haploinsufficient Manner
The majority (∼70%) of human sporadic colorectal cancers are initiated by biallelic mutations in the APC tumor suppressor gene (15, 16). To test whether MACROD2 deficiency could promote intestinal tumorigenesis, we crossed Macrod2 knockout mice obtained from the Knockout Mouse Project Repository (The Jackson Laboratory; Fig. 2A), which develop normally into adulthood (see Supplementary Data), with the ApcMin/+ mouse model (17). ApcMin/+ mice harbor a truncating germline mutation in Apc and intestinal tumors arise spontaneously from loss of heterozygosity of the wild-type Apc allele (17), a mechanism found in human sporadic colorectal cancer and familial adenomatous polyposis (18, 19).
We observed a significant increase in the multiplicity of intestinal adenomas in F2 ApcMin/+/Macrod2−/− and ApcMin/+/Macrod2−/+ mice as compared with ApcMin/+/Macrod2+/+ mice (35 animals per group), with an average number of 81.8, 81.3, and 65.6 polyps, respectively (P < 0.012 and P < 0.007, respectively; Fig. 2B and C). Concomitantly, total adenoma burden (indicated by the overall area of tumors) was significantly increased in both the ApcMin/+/Macrod2−/− and the ApcMin/+/Macrod2−/+ mice (P < 0.033 and P < 0.044, respectively), with evidence for a significant gene dosage–dependent enhancement when considering large (>3 mm) adenomas (2.5-fold and 2-fold increase, P < 0.001 and P < 0.019, respectively; Fig. 2C). Accordingly, the median number of neoplastic cells per adenoma as assessed by image analysis of hematoxylin and eosin (H&E)–stained intestinal sections increased from 1,212.5 in ApcMin/+/Macrod2+/+ mice to 1,808 and 1,865.5 in ApcMin/+/Macrod2−/+ and ApcMin/+/Macrod2−/− mice, respectively (P < 0.001 for both comparisons; Supplementary Fig. S3A–S3B). Overall survival of aging ApcMin/+/Macrod2−/− and ApcMin/+/Macrod2−/+ mice as compared with ApcMin/+/Macrod2+/+ mice was significantly decreased, with a median survival time of 109.5, 126, and 142 days, respectively (P < 0.006 and P < 0.013, respectively; Fig. 2D). No tumors were found in the intestines of animals from each of the three control groups (Apc+/+/Macrod2+/+, Apc+/+/Macrod2−/+, and Apc+/+/Macrod2−/−; 15 animals per group).
To determine whether MACROD2 deficiency could also enhance growth of human colorectal cancer cells in vivo, we generated xenografts of MACROD2 wild-type, heterozygous, and homozygous knockout HCT116 cells (β-catenin mutated) using CRISPR/Cas9 technology (Supplementary Figs. S4A–S4B and S5A–S5B). Isogenic HCT116-MACROD2−/− and HCT116-MACROD2−/+ cells exhibited significantly increased establishment rates in CBA athymic nude mice as compared with HCT116-MACROD2+/+ cells (P < 0.002 for both comparisons, Fig. 2E), and HCT116-MACROD2−/− cells further displayed increased tumor growth over a 20-day period (P = 0.007). Increased tumor growth was similarly observed for shRNA-mediated MACROD2 knockdown in HCT116-MACROD2+/+ cells (Supplementary Fig. S6A–S6B). Conversely, reconstitution of GFP-tagged wild-type MACROD2 in homozygous deleted LIM2405 cells (APC mutated) resulted in a significant reduction in tumor growth as compared with GFP-control transfected cells (P < 0.001; Fig. 2E; Supplementary Fig. S5A–S5B). Collectively, these results show that MACROD2 deficiency promotes the growth of intestinal tumors in a haploinsufficient manner.
MACROD2 Deficiency Does Not Affect WNT Signaling in APC or CTNNB1 Mutated Intestinal Tumor Cells
MACROD2 is a positive regulator of GSK3β, which in turn controls cellular β-catenin levels, nuclear relocalization, and engagement of downstream WNT signaling (10). However, the majority of human colorectal cancers carry inactivating mutations in APC, a central member of the β-catenin destruction complex, or constitutively activating mutations in CTNNB1 (encoding β-catenin; ref. 20). Whether the regulatory role of GSK3β is maintained in this context is uncertain. To assess the impact of MACROD2 deficiency in human colorectal cancer cells with WNT pathway mutations, we evaluated the expression of WNT target genes (CD44, AXIN2, and DVL1; TCF/LEF reporter assay) and β-catenin relocalization in representative CTNNB1–mutated (HCT116-MACROD2−/−, HCT116-MACROD2−/+ and HCT116-MACROD2+/+, LIM1215-MACROD2−/+) and APC-mutated cell lines (LIM2405-MACROD2−/−, COLO320-MACROD2−/+, and LOVO-MACROD2+/+). For LIM1215, LIM2405, COLO320, and LOVO cell lines, MACROD2 was either reconstituted or suppressed depending on their MACROD2 deletion status (Supplementary Fig. S5A–S5B). HEK293T cells were used as APC and CTNNB1 wild-type control cells.
Consistent with previous reports, MACROD2 suppression in HEK293T cells resulted in significant induction of WNT target gene expression, increased TCF/LEF reporter activity, and β-catenin nuclear relocalization in response to WNT3A stimulation (Supplementary Fig. S7A–S7F). However, in colorectal cancer cells with APC or CTNNB1 mutations, MACROD2 suppression or reconstitution did not affect the expression of WNT target genes (Fig. 3A), TCF/LEF reporter activity (Fig. 3B), or β-catenin subcellular localization (Fig. 3C and D). In addition, no association between MACROD2 deletion status and WNT target gene expression was observed when integrating SNP array and RNA sequencing (RNA-seq) data from 208 TCGA-analyzed COAD/READ samples (Fig. 3E). Consistent with the findings in human colorectal cancer cells, evaluation of intestinal polyps from ApcMin/+/Macrod2−/−, ApcMin/+/Macrod2−/+, and ApcMin/+/Macrod2+/+ mice showed similar expression of WNT target genes by RNA-seq analysis and qRT-PCR (Fig. 3F and G), and similar patterns of subcellular β-catenin localization (Fig. 3H and I). Taken together, these data indicate that MACROD2 deficiency has no major impact on WNT signaling—within the level detectable by the assays used—in the context of APC- or CTNNB1–mutated intestinal tumor cells.
MACROD2 Haploinsufficiency Causes Repression of PARP1 Transferase Activity
The macrodomain of MACROD2 has previously been shown to be recruited to sites of laser-induced DNA damage in a PARP1-dependent manner, and to regulate PARP1 activity through the removal of terminal autoinhibitory mono-ADP-ribose (7, 21).
Given that MACROD2 microdeletions in human colorectal cancer often produce in-frame exonic deletions involving the macrodomain, we tested whether these resulted in a loss of protein recruitment to sites of DNA damage induced by near-infrared (NIR) laser microirradiation. Corresponding deletion transcripts were cloned from colorectal cancer cell lines and expressed in HeLa cells and shown to produce truncated proteins (MACROD2Δex1-8, MACROD2Δex4, MACROD2Δex5, and MACROD2Δex4-5; Supplementary Fig. S8). As anticipated, GFP-tagged MACROD2 deletion proteins lacking an intact macrodomain were not recruited to DNA lesions, whereas full-length protein (MACROD2-GFP) and a control truncated protein with an intact macrodomain (MACROD2Δex11-17) exhibited rapid accumulation (Fig. 4A).
To determine whether MACROD2 loss results in an increase of PARP1 autoinhibitory mono-ADP-ribosylation following induction of DNA damage, we treated our isogenic HCT116 MACROD2+/+, MACROD2−/+, and MACROD−/− cells with the DNA topoisomerase II inhibitor doxorubicin and probed immunoprecipitated PARP1 with anti-mono(ADP-ribose) antibody. As anticipated, PARP1 mono-ADP-ribosylation levels (relative to total PARP1) were elevated with MACROD2 loss posttreatment with doxorubicin in a gene dosage–dependent manner (Fig. 4B). Conversely, rescue of homozygous or heterozygous MACROD2-deleted cell lines (LIM2405-MACROD2−/− and LIM1215-MACROD2−/+) by expression of wild-type MACROD2 resulted in a decrease in mono-ADP-ribosylated PARP1 (Fig. 4B).
To quantify the impact of homozygous and heterozygous MACROD2 deficiency on PARP1 transferase activity in human colorectal cancer cells, we measured global protein PARylation levels (Fig. 4C) and incorporation of biotinylated PAR onto histone proteins (Fig. 4D) following DNA damage with doxorubicin. Isogenic HCT116-MACROD2−/− and HCT116-MACROD2−/+ cells showed significant attenuation of both overall protein PARylation and PAR incorporation onto histones as compared with HCT116-MACROD2+/+ cells (P < 0.010). Conversely, rescue of homozygous or heterozygous MACROD2-deleted cell lines (LIM2405-MACROD2−/− and LIM1215-MACROD2−/+) by expression of wild-type MACROD2 resulted in a significant increase in global protein and histone PARylation (P < 0.005 for all comparisons). Similar results for PAR incorporation onto histones were observed when cells were exposed to γ-irradiation (Supplementary Fig. S9). Protein PARylation was reflective of PARP1 activity, as the addition of PARP1 inhibitor (olaparib) abrogated PAR incorporation in MACROD2 wild-type cells and MACROD2 rescued cells (Fig. 4C).
Together, these findings indicate that MACROD2 haploinsufficiency, due to whole-gene or catalytic macrodomain deletions, causes impaired PARP1 transferase activity in the context of human colorectal cancer cells.
MACROD2 Haploinsufficiency Alters DNA Repair and Sensitivity to DNA Damage
Impairment of PARP1 activity caused by loss of MACROD2 regulatory function is predicted to result in altered DNA repair and sensitivity of cells to genotoxic stress–induced DNA damage. Accordingly, comet assays on isogenic HCT116-MACROD2−/− and HCT116-MACROD2−/+ cells identified a significant, gene dosage–dependent increase for both DSBs (neutral comet assay) and SSBs (alkaline comet assay) as compared with HCT116-MACROD2+/+ when challenged with either IR or doxorubicin (P < 0.010 for all comparisons; Fig. 4E–F; Supplementary Fig. S10A–S10B). Equivalent results were obtained when analyzing mouse embryonic fibroblasts (MEF) generated from Macrod2−/−, Macrod2−/+, and MACROD2+/+ mice (Supplementary Fig. S11A–S11B). Conversely, reconstitution of wild-type MACROD2 in colorectal cancer cell lines with homozygous MACROD2 deletions (LIM2405 and SW480) or heterozygous MACROD2 deletions (COLO320 and LIM1215) resulted in a significant reduction of DSBs and SSBs (P < 0.001; Fig. 4G and H; Supplementary Fig. S12A and S10B). Reconstitution with MACROD2Δex4 protein lacking the catalytic macrodomain did not result in rescue of MACROD2-deleted cell lines (Supplementary Fig. S12C).
Pharmacologic inhibition of PARP1 (olaparib) increased DSBs and SSBs in HCT116-MACROD2+/+ and, to a lesser extent, HCT116-MACROD2+/− cells, but not in HCT116-MACROD2−/− cells (Fig. 4I). Conversely, rescue of MACROD2-deleted cell lines with wild-type MACROD2 was blocked with PARP1 inhibition (olaparib), and treatment did not increase DSBs and SSBs in MACROD2-deleted GFP-control cells (Fig. 4J). Similar comet assay results for rescue and nonadditivity were observed for an alternative PARP1 inhibitor (veliparib; Supplementary Fig. S13).
The impact of MACROD2 haploinsufficiency on DNA repair was further demonstrated through differential formation of DNA damage–induced foci of phosphorylated H2AX (γ-H2AX) and phosphorylated ATM (pATM), markers of DSBs. Human colorectal cancer or MEF cells with homozygous or heterozygous MACROD2 deletions developed significantly more γ-H2AX foci as compared with MACROD2 wild-type cells when challenged with IR or doxorubicin, whereas the inverse was observed upon MACROD2 reconstitution in MACROD2−/− or MACROD2−/+ cells (Fig. 5A–B; Supplementary Fig. S14A and S14B). Similar results were obtained for foci of pATM (Supplementary Fig. S15A and S15B). Reconstitution with MACROD2Δex4 protein again did not rescue γ-H2AX foci formation in MACROD2-deleted cell lines (Supplementary Fig. S16A and S16B). Moreover, a significant dose-dependent increase in γ-H2AX staining with MACROD2 loss was observed for intestinal polyps from ApcMin/+/Macrod2−/+ and ApcMin/+/Macrod2−/− mice as compared with ApcMin/+/Macrod2+/+ mice (Fig. 5C and D).
Impairment of DNA repair was reflected by increased tumor cell sensitivity to DNA damage. In clonogenic assays, isogenic HCT116-MACROD2−/− cells showed significantly reduced viability as compared with HCT116-MACROD2+/+ cells when treated with IR or doxorubicin (P < 0.010 for all comparisons; Supplementary Fig. S17A–A17B). HCT116-MACROD2−/+ cells showed reduced viability for IR (P < 0.05) but not doxorubicin treatment. Accordingly, suppression or deletion of wild-type MACROD2 resulted in increased apoptosis posttreatment as measured by the luminescent assay for caspase 3/7 activity or by flow cytometric analysis of Annexin V staining, whereas rescue of MACROD2-deleted cell lines resulted in reduced apoptosis (P < 0.05 for all comparisons; Supplementary Fig. S18A–S18B).
A well-established consequence of loss of PARP1 function is impaired homologous recombination (HR)–mediated DSB repair (22). To test for the impact of MACROD2 deficiency on HR-mediated DSB repair, we performed DR-GFP reporter assays on MACROD2−/− LIM2405 and SW480 cells with and without MACROD2 reconstitution. In this system, transient expression of an I-SceI endonuclease generates a DSB at integrated GFP-deletion gene sequences in which error-free repair leads to a full-length GFP product detectable by flow cytometry (23). As anticipated, reintroduction of RFP-tagged full-length MACROD2 into LIM2405-DR and SW480-DR cells resulted in significant enhancement of HR-mediated repair as compared with RFP control cells (P < 0.001 for both comparisons; Fig. 5E). To further evaluate HR-mediated DSB repair, we performed assessment of BRCA1 foci posttreatment with IR or doxorubicin. HCT116-MACROD2−/− and HCT116-MACROD−/+ cells developed significantly fewer BRCA1 foci as compared with HCT116-MACROD2+/+ cells, whereas the inverse was observed upon MACROD2 reconstitution in LIM2405-MACROD2−/− or LIM1215-MACROD2−/+ cells (Fig. 5F; Supplementary Fig. S19). Comparable results were observed in isogenic HCT116 MACROD2 knockout cells for RAD51 foci (Supplementary Fig. S20).
Together, these data indicate that impairment of PARP1 activity caused by MACROD2 haploinsufficiency results in altered DNA repair and increased sensitivity to DNA damage.
MACROD2 Haploinsufficiency Promotes the Development of CIN
Increased susceptibility to DNA damage is a major pathway to chromosome instability (CIN) and aneuploidy in human cancer (24, 25). To examine whether MACROD2 haploinsufficiency compromised maintenance of genome integrity, we examined karyotype stability of MEFs from Macrod2 knockout mice during long-term cell culture. Karyotype variability was strikingly increased by passage 60 in MACROD2−/− and Macrod2−/+ MEFs as compared with MACROD2+/+ MEFs (P < 0.001 for both comparisons), with 82% (86 of 105), 47% (59 of 126), and 15% (17 of 116) of metaphases exhibiting >55 chromosomes, respectively (Supplementary Fig. S21A). Gross aneuploidy in MACROD2−/− and MACROD2−/+ MEFs was further evident by flow cytometry analysis of single-cell clones. Ninety-one percent (21 of 23) of MACROD2−/− MEF clones and 53% (8 of 15) of MACROD2−/+ MEF clones were aneuploid, as compared with 16% (6 of 38) of MACROD2+/+ MEF clones (P < 0.005 for both comparisons; Supplementary Fig. S21B).
To confirm that MACROD2 haploinsufficiency could also promote the development of aneuploidy in human colorectal cancer cells, we repeated the long-term culture experiments using our isogenic HCT116 MACROD2 knockout cells. Consistent with our results in MEFs, karyotype variability was increased by passage 50 in HCT116-MACROD2−/− and HCT116-MACROD2−/+ cells as compared with HCT116-MACROD2+/+ cells (P < 0.011 for both comparisons), with 11.5% (8 of 61), 4% (3 of 74), and 2% (1 of 84) of metaphases exhibiting >50 chromosomes, respectively (Fig. 6A). Aneuploidy in HCT116-MACROD2−/− and HCT116-MACROD2−/+ cells was again confirmed by flow cytometry, with 9.6% (8 of 83) of HCT116-MACROD2−/− clones and 2.4% (4 of 104) of HCT116-MACROD2−/+ clones showing aneuploidy, as compared with 0% (0 of 43) of HCT116-MACROD2+/+ clones (P = 0.035 and P = 0.196, respectively, Z-test; Fig. 6B).
Further, G-banding analysis for isogenic HCT116 MACROD2 knockout cells demonstrated increased structural chromosome abnormalities in HCT116-MACROD2−/− and HCT116-MACROD2−/+ cells as compared with HCT116-MACROD2+/+ cells, with 20, 12, and 5 aberrations in 30 metaphases, respectively (Fig. 6C; Supplementary Table S6). Structural abnormalities included abundant interstitial deletions and translocations (derivative chromosomes). Cytogenetic analysis further confirmed our initial observation of increased numerical chromosome abnormalities (20, 12, and 7 aberrations, respectively), but clarified that many of these represented marker chromosomes (unidentifiable chromosomes). Notably, short-term treatment of HCT116-MACROD2+/+ cells with the PARP1 inhibitor olaparib produced a similar spectrum of chromosome aberrations, with 19 structural and 22 numerical abnormalities, respectively (Fig. 6C; Supplementary Table S6).
Our karyotype and flow cytometry analyses of long-term cell cultures suggest that MACROD2 haploinsufficiency promotes CIN. To measure the impact of MACROD2 loss on mitotic chromosome segregation, we quantified the level of mitotic defects—including anaphase bridges, lagging chromosomes, and micronuclei—in our isogenic HCT116 MACROD2 knockout cells by immunostaining for kinetochores and microtubules (Fig. 7A). As anticipated, HCT116-MACROD2−/+ and HCT116-MACROD2−/− cells displayed a significant gene dosage–dependent increase in chromosome missegregation errors as compared with HCT116-MACROD2+/+ cells (P < 0.001 and P < 0.001, respectively; Fig. 7B). A gene dosage–dependent increase in mitotic defects was further confirmed for tumors from ApcMin/+/Macrod2−/+ and ApcMin/+/Macrod2−/− mice as compared with ApcMin/+/Macrod2+/+ mice (P < 0.001 and P < 0.001, respectively; Fig. 7C and D). We further noted evidence of centrosome amplification associated with MACROD2 loss, with ∼10% of HCT116-MACROD2−/− and HCT116-MACROD2−/+ cells harboring more than two centrosomes, whereas HCT116-MACROD2+/+ cells showed no abnormal centrosome numbers (Fig. 7E and F). The chromosome missegregation phenotype associated with MACROD2 haploinsufficiency was phenocopied by treatment of HCT116-MACROD2+/+ cells with PARP1 inhibitors (olaparib and veliparib; Fig. 7B; Supplementary Fig. S22).
To examine the relationship between MACROD2 loss and the extent of aneuploidy in human primary tumors, we calculated the estimated chromosome segment number (eCSN; Supplementary Methods) from SNP array data for the 616 TCGA and 651 in-house colorectal cancers. eCSN increased by ∼1.3-fold in MACROD2−/+ tumors and ∼1.8-fold in MACROD2−/− tumors as compared with the respective MACROD2+/+ tumors (P < 0.001 for both comparisons in both cohorts; Fig. 7G). Concomitantly, the proportion of aneuploid tumors increased from 72.7% in MACROD2+/+ to 83.0% and 92.3% in MACROD2−/+ and MACROD2−/− cases, respectively (Fig. 7H). In multivariate analysis, adjusting for age at diagnosis, gender, tumor location, stage, differentiation, and MSI status, eCSN remained independently associated with MACROD2 deletion status in both cohorts (Supplementary Table S7). Considering other proposed drivers of CIN in colorectal cancer, including APC (26, 27), TP53 (28), FBXW7 (29), and BCL9L (30), MACROD2 status remained an independent predictor of eCSN in both cohorts (Supplementary Table S8; BCL9L mutation status only available for TCGA cohort). Our results identify MACROD2 as a critical caretaker gene in human colorectal cancer, haploinsufficiency of which leads to CIN and aneuploidy.
In summary, we propose a model in which partial or complete loss of MACROD2 mono-ADP-ribosylhydrolase function, because of whole-gene or catalytic domain deletions, causes impaired PARP1 transferase activity in human colorectal cancer by abrogating removal of PARP1 terminal autoinhibitory mono-ADP-ribose (Fig. 7I). Repression of PARP1 activity results in altered DNA repair and sensitivity to DNA damage. DNA repair deficiency, potentially enhanced by centrosome amplification, appears to culminate in CIN, thereby promoting intratumor heterogeneity and cancer progression.
CIN is a principal driver of cancer evolution (31), has prognostic relevance in multiple cancer types (32–34), and is associated with tumor multidrug resistance (35, 36). Our studies identify MACROD2 as a major caretaker tumor suppressor gene in human colorectal cancer, with haploinsufficiency causing repression of PARP1 activity, altered DNA repair, and CIN.
Our DNA copy-number analysis on 651 sporadic colorectal cancers validates the original observation of recurrent focal MACROD2 deletions in TCGA-analyzed colorectal cancers, confirming a prevalence of ∼30% (4). Integrating both datasets and expanding the analysis to 53 human colorectal cancer cell lines, our report provides the first comprehensive fine-mapping data for MACROD2 deletions, establishing deletion extent and status (heterozygous vs. homozygous), and demonstrating the impact of both exonic and intronic deletions at the MACROD2 transcript level. MACROD2 was almost exclusively subject to whole-gene deletions or microdeletions involving the catalytic macrodomain, with only rare somatic point mutations. About two thirds of tumors with MACROD2 aberrations showed evidence of two hits. A similar pattern of targeting with predominant focal deletions has previously been described for the PARK2 tumor suppressor in colorectal cancer (37). The high frequency of MACROD2 deletions in colorectal cancer cells may be a consequence of the reported susceptibility of this locus to breakage during replicative stress (6).
MACROD2 is a mono-ADP-ribosylhydrolase with a critical role in the regulation of the DNA-damage sensor PARP1 through controlling the removal of the terminal autoinhibitory mono-ADP-ribose (7, 21). Our results indicate that MACROD2 deficiency in colorectal cancer causes increased PARP1 mono-ADP-ribosylation and repression of PARP1 transferase activity, leading to impaired DNA repair and sensitivity to genotoxic stress-induced DNA damage in a gene dosage–dependent manner. The MACROD2 macrodomain is required for PARP1-dependent recruitment to sites of DNA damage (7), and accordingly MACROD2 proteins with in-frame exonic deletions disrupting this region exhibited loss of accumulation. Our findings are supported by studies in PARP1-deficient cells that similarly show altered DNA repair and sensitivity to DNA damage (38, 39). PARP1 inhibitor use in cells deficient in HR repair is a prime example of the therapeutic paradigm of synthetic lethality in cancer (38, 40, 41), and MACROD2 haploinsufficiency may constitute a potential vulnerability in tumors that could be therapeutically exploited.
Our data reveal MACROD2 as a caretaker gene in colorectal cancer, haploinsufficiency of which results in CIN. MACROD2-deficient colorectal cancer cells displayed a gene dosage–dependent increase in structural and numerical chromosome abnormalities and were prone to chromosome missegregation errors. Similar to our findings for MACROD2 deficiency, Parp1 knockout MEFs have been shown to accumulate structural chromosome abnormalities (38, 39). Cytogenetic analyses of Parp1 knockout MEFs have further reported an increase in aneuploidy from ∼1% in Parp1+/+ cells to ∼5% and ∼20% in Parp1−/+ and Parp1−/− cells, respectively (42). Accordingly, PARP1 inhibitor treatment of MACROD2 wild-type colorectal cancer cells recapitulated the spectrum of chromosome abnormalities and missegregation phenotype associated with MACROD2 loss. MACROD2 deletion in primary tumors was associated with increased eCSN and aneuploidy, independent of clinical features and other known drivers of CIN. Interestingly, MACROD2-deficient colorectal cancer cells also showed a tendency to centrosome amplification, a phenotype linked to PARP1 deficiency (43), which may contribute to the CIN phenotype. Together, these data are consistent with a model in which MACROD2 loss promotes cancer CIN at least in part mediated through the impairment of PARP1.
Our in vivo studies support a role of MACROD2 as a tumor suppressor acting in a haploinsufficient manner. Homozygous and heterozygous MACROD2 deficiency enhanced intestinal tumorigenesis in ApcMin/+ mice, as evident for both tumor multiplicity and burden, and promoted the growth of human colorectal cancer xenografts. Previous evaluation of PARP1-deficient mice for an alternative model of sporadic colon tumorigenesis (azoxymethane treatment) has found increased colon tumorigenesis (44), demonstrating that when combined with relevant drivers, PARP1 deficiency promotes intestinal tumor development, supporting our observations for MACROD2 deficiency. Notably, functions other than impairment of PARP1 may contribute to the protumorigenic impact of MACROD2 loss. Previous studies have implicated MACROD2 as a regulator of GSK3β and WNT signaling that could provide such an additional protumorigenic drive (10). However, we found no supporting evidence for this in APC or CTNNB1 mutated intestinal tumor cells. Studies to expand our knowledge of MACROD2 protein targets and related pathways are clearly warranted.
In summary, our findings reveal MACROD2 as a haploinsufficient caretaker tumor suppressor gene, essential for the maintenance of genome integrity. Although MACROD2 aberrations were particularly frequent in human colorectal cancer, recurrent focal loss was also noted in cancers of the stomach, esophagus, cervix, uterus, lung, liver, and thyroid, supporting a broader role of MACROD2 haploinsufficiency in the development and evolution of human malignancies.
Additional details are provided in the Supplementary Methods.
A total of 651 patients with stages 1–4 colorectal cancer were recruited from the Royal Melbourne Hospital (Parkville, VIC, Australia), Western Hospital Footscray (Footscray, VIC, Australia), and St Vincent's Hospital Sydney (Darlinghurst, NSW, Australia). Patients with familial polyposis syndromes, ulcerative colitis, or Crohn disease–associated colorectal cancer were excluded. Written informed consent was obtained from all patients; the studies were conducted in accordance with recognized ethical guidelines (National Statement on Ethical Conduct in Human Research, Commonwealth of Australia) and approved by the Human Research Ethics Committee of the Walter and Eliza Hall Institute of Medical Research (Parkville, VIC, Australia; WEHI HREC 12/19). Fresh-frozen tumor and matched normal tissues were retrieved from hospital-associated tissue banks, including 60 stage 1, 208 stage 2, 297 stage 3, and 86 stage 4 cases. Colon tumors from caecum to transverse colon were defined as proximal, and those from splenic flexure to rectum as distal. Patients’ clinicopathologic characteristics were collected using a multisite database and are summarized in Supplementary Table S2.
A total of 53 colorectal cancer cell lines were studied: C10, C32, C80, C99, C106, C135, CCK81, CoCM-1, COLO320, COLO678, CX-1, Gp2D, HCA7, HCC2998, HCT15, HCT116, HDC54, HDC82, HDC87, HDC90, HRA19, HT115, HT55, KM12, LIM1215, LIM1899, LIM2099, LIM2405, LIM2551, LOVO, LS180, LS411, LS513, NCI-H716, NCI-H747, RKO, RW2982, RW7213, SKCO-1, SNU-175, SNU-283, SNU-C1, SNU-C2B, SW1222, SW1417, SW1463, SW48, SW480, SW837, T84, VACO4S, VACO5, and VACO10. HEK293T cells were used as APC and CTNNB1 wild-type controls for WNT stimulation assays. Cells were cultured with DMEM and 10% FBS at 37°C and 5% CO2. Cell lines were authenticated by short tandem repeat analysis using the GenePrint 10 System (Promega) at the Australian Genome Research Facility (AGRF; Parkville, VIC, Australia, February 2016) and confirmed to be Mycoplasma free using the Lookout Mycoplasma PCR Detection Kit (Sigma-Aldrich, January 2017). Details of colorectal cancer cell lines are summarized in Supplementary Table S3. Treatments were performed with doxorubicin (topoisomerase II inhibitor, SelleckChem, S1208), olaparib (an effective catalytic PARP1 inhibitor/effective PARP1 trapper, SelleckChem, S1060), veliparib (an effective catalytic PARP1 inhibitor/poor PARP1 trapper, SelleckChem, S1004), or γ-irradiation using a Cobalt-60 source as indicated.
Microsatellite Instability Analysis
Tumor and matched normal DNA were PCR-amplified for the Bethesda panel of microsatellite markers (BAT25, BAT26, D2S123, D5S346, and D17S250) using fluorescently labeled primers (45). Reaction products were analyzed on a 3130xl Genetic Analyzer (Applied Biosystems). Microsatellite instability-high was diagnosed if instability was evident at two or more markers.
SNP Array Analysis
SNP assays were performed using Human 610-Quad BeadChips (Illumina) at the AGRF. For patient specimens, matched primary cancers and normal DNA samples were analyzed on the same BeadChip. Raw SNP array data were processed using GenomeStudio software (Illumina), and call rate, genotype, log R ratio (LRR), and B allele frequency data (BAF) were exported. Data have been deposited in the Gene Expression Omnibus (GSE115145). SNP array data for colorectal cancer cell lines were previously deposited (GSE55832). For TCGA-analyzed tumor types, Genome-Wide Human SNP Array 6.0 (Affymetrix) CEL files were retrieved and processed using Affymetrix Power Tools (Affymetrix) and PennCNV (46) to export LRR and BAF data. DNA copy-number aberrations (segments) in tumor and colorectal cancer cell line samples were identified using OncoSNP v2.18 Suite as previously described (47). To account for noise, segmented regions were accepted when the difference between LRR means for adjacent regions (including adjacent chromosomes, i.e., end of chromosome 1 and start of chromosome 2) was greater than their LRR standard deviations. The eCSN is the total number of segments per sample. Aneuploidy was assigned if samples showed three or more nondiploid chromosomes, where individual chromosome copy numbers were estimated based on the modal copy number for all SNPs within that chromosome.
Tumor Burden Assessment
On autopsy of knockout mice, the intestinal tract was opened longitudinally, fixed in methacarn, and stained with methylene blue, and tumors were enumerated according to their size by two observers (A. Sakthianandeswaren and M.J. Parsons) blinded to the genotype of the mice. Representative tumors and adjacent nontumor tissues were harvested for DNA/RNA extraction. Swiss rolls of intestinal segments were prepared and fixed in 10% neutral buffered formalin for paraffin embedding. H&E-stained tissue sections were imaged, and the number of neoplastic cells per adenoma was quantified by image analysis using ImageJ software.
Human Colorectal Cancer Cell Line Xenografts
Xenografts of isogenic HCT116-MACROD2 knockout cells, HCT116-shMACROD2 knockdown cells, LIM2405-MACROD2-GFP reconstituted cells, and respective controls were generated by subcutaneous injection of 1 × 106 cells into the rear flanks of BALB/c athymic nude mice. Knockdown or reexpression of MACROD2 was confirmed using qRT-PCR (see below): forward primer 5′-TGACCTTAGAAGAGAGACGCAAA-3′, reverse primer 5′-TCTTCACCTGGGATGTTTCC-3′. Eight to 12 xenografts were analyzed for each group on two occasions. Tumor growth was measured using digital calipers on alternate days until days 20 to 24 after injection. Length, width, and height of the subcutaneous xenograft nodules were measured, and tumor volume calculated as π/6 × L × W × H. At the experimental endpoint, all mice were euthanized, and tumors were removed and fixed in formalin. All animal procedures were approved and conducted in accordance with the Animal Ethics Committee of the Walter and Eliza Hall Institute of Medical Research (Parkville, VIC, Australia).
NIR Laser Microirradiation Assay
NIR laser microirradiation was performed using a SpectraPhysics MaiTai Ti:Saphhire laser with DeepSee attachment at 800 nm, with a 13-μs laser pulse delivered to the cell nucleus with a total energy of 7.65 mJ. Microirradiated cells were imaged every 20 seconds for 4 minutes using a 60× UPlanApo NA1.2 water immersion objective and an Olympus FV1000 confocal system attached to an Olympus IX-81 microscope (473 nm excitation, 510–550 nm detection). During the experiment, cells were kept at 37°C in a CO2-independent HEPES-based imaging medium (Invitrogen) supplemented with 10% FBS (Invitrogen).
Single-Cell Gel Electrophoresis (Comet) Assay
Colorectal cancer cells (1 × 105 cells/well seeded) and MEF cells (1 × 105 cells/well seeded) were seeded into 24-well plates. Cells were γ-irradiated with 10 Gy using a Cobalt-60 source or treated with 0.5 μmol/L doxorubicin [± olaparib (SelleckChem) or DMSO vehicle control], and the degree of DNA damage analyzed at baseline and 2 hours after treatment under neutral (DSBs) or alkaline (SSBs) conditions using the CometAssay Kit (Trevigen) as per the manufacturer's instructions. After electrophoresis, slides were stained with SYBR Green Reagent (Bio-Rad), and comets were imaged using a SPOT RT3 slide camera attached to a Nikon 90i microscope (Nikon). Comet Olive tail moments [(tail length) × (tail fluorescence/(head + tail fluorescence))] were determined using MetaMorph Microscopy Automation and Image Analysis Software (Molecular Devices). At least 100 randomly chosen comets were analyzed per sample.
Antibodies and Reagents
Immunofluoresence and immunohistochemistry were performed with the following antibodies: anti–β-catenin (BD Transduction Laboratories, #610153, 1:500 dilution), anti-γH2A.X (phospho S140 [3F2]; Abcam, ab22551, 1:1,000 dilution), anti–phospho-histone H2A.X (Ser139; 20E3; Cell Signaling Technology, #9718, 1:200 dilution), anti-BRCA1 (Sigma, SAB2702136, 1:1,000 dilution), anti-RAD51 (Abcam, ab133534, 1:1,000 dilution), or anti-pATM antibody (EP1890Y; Abcam, ab81292, 1:1,000 dilution). Western blotting analyses were performed with the following antibodies: anti-PARP1 antibody (Abcam, ab110915), anti-MACROD2 (in house; Supplementary Fig. S23), anti-(mono)ADP ribose (Merck Millipore, MABE1076), anti-poly(ADP)ribose (Enzo, ALX-804-220-R100), anti-vinculin (Sigma, V9264), anti-GFP (OriGene, TA150052), and anti-PARP1 (Abcam, ab6079).
Primer sequences used for assays detailed in the Supplementary Methods are provided in Supplementary Tables S9, S10, and S11.
Cells growing in exponential phase were incubated with 0.2 μg/mL colchicine (Sigma, C3915) overnight. Cells were harvested by mitotic shake off, incubated with 0.56% or 0.28% KCl at 37°C for MEFs and human colorectal cancer cells, respectively, fixed with 3:1 methanol/acetic acid (v/v), and dropped onto slides. Slides were air-dried and stained with Leishman's stain (VWR International) for 4 minutes. Chromosome numbers were evaluated using an Axioimager Z2 (Zeiss) a under a 63× objective. At least 50 metaphases were analyzed per sample.
Cells in exponential growth phase (± treatment with 10 μmol/L olaparib for 28 hours) were incubated with 0.2 μg/mL colchicine (Sigma, C3915) for 60 minutes. Metaphases were harvested according to standard procedures (48). Thirty pseudodiploid metaphases with minimal overlaps were selected at 100× magnification, captured, and karyotyped at 1,000× magnification with an Ikaros karyotyping system (Metasystems). Karyotypes were written with reference to the published HCT116 karyotype (45, X, -Y, der(10)dup(10)(q24q26)t(10;16)(q26;q24), der(16)t(8;16)(q13;p13), der(18)t(17;18)(q21;p11.3); ref. 49) as stemline according to International System for Human Cytogenetic Nomenclature (2016; ref. 50). Changes from the stemline karyotype were classified as structural and/or numerical abnormalities. Markers and abnormal chromosomes were each counted as one structural abnormality. Gain or loss of a chromosome from the stemline karyotype was counted as one numerical abnormality.
DSB HR DNA Repair Assays
DSBs in DR-GFP reporter cell lines (106 cells) were induced by transfection with pCBASce plasmid according to the manufacturer's instructions; transfection with empty vector control served as negative control. HR repair activity was assessed by quantification of the percentages of GFP-positive cells using a BD FACSCalibur instrument (BD Biosciences). Data were analyzed using CellQuest 3.2 software (Becton Dickinson). For each experiment, 200,000 cells were scored per treatment group, and the frequency of recombination events was calculated from the number of GFP-positive cells divided by the number of cells analyzed following correction for transfection efficiency.
Chromosome Segregation and Centrosome Assays
Cells were grown overnight on glass cover-slips, fixed in 4% paraformaldehyde for 20 minutes, and permeabilized with 0.2% Triton X-100 for 15 minutes. Samples were blocked in 5% BSA/TBS and labeled with rabbit CENP-A (CST, 2186S, 1:400), rabbit γ-tubulin (Abcam, ab11317, 1:500), and mouse β-tubulin (Sigma, T4026, 1:200) overnight at 4°C. After incubation with Alexa Fluor–conjugated secondary antibodies (Alexa 546 anti-mouse A11003, 1:1,000; Alexa 647 anti-rabbit A21245, 1:1,000) and DAPI (Roche, 1:5,000), the slides were mounted with DPX (Sigma-Aldrich). Immunofluorescence analyses for segregation errors and centrosome number were performed on a Leica SP8 confocal microscope equipped with a 63×/1.8 NA oil immersion objective. At least 100 mitotic events were scored per genotype.
Statistical analyses were performed using the statistical computing software R (R Development Core Team, 2011). For univariate analyses, differences between groups were assessed using the Pearson χ2 test for categorical variables, and the Student t test, Z-proportion test, or Levene test for continuous variables as indicated. Survival data for aging mice were analyzed using the Wald test. Multivariate analysis for the association between MACROD2 deletion status and eCSN, with adjustment for clinicopathologic and molecular features was performed using a quasi-Poisson generalized linear model. The Compare Groups of Growth Curves (http://bioinf.wehi.edu.au/software/compareCurves) permutation test was used to evaluate cell line xenograft assays. Identification of focal MACROD2 deletions in TCGA cohorts was performed using GISTIC v2.022 (12). Masked Copy Number Segment data were retrieved from the Genomic Data Commons Data Portal (https://portal.gdc.cancer.gov/), filtering for tumor samples and removing duplicates. GISTIC was run in “genegistic” mode, with default parameters. Differential gene expression analysis for RNA-seq data was performed with voom/limma using the edgeR R package; results for WNT target genes (http://web.stanford.edu/group/nusselab/cgi-bin/wnt/target_genes) were adjusted for multiple testing using the Benjamini–Hochberg procedure. All comparisons were two sided, and P values of <0.05 were considered statistically significant.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: A. Sakthianandeswaren, D. Mouradov, J. Desai, I. Street, M. Buchert, O.M. Sieber
Development of methodology: A. Sakthianandeswaren, M.J. Parsons, R.N. MacKinnon, C. Tsui, C.J. Nowell, I. Street, O.M. Sieber
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.J. Parsons, R.N. MacKinnon, B. Catimel, S. Liu, M. Palmieri, T.L. Putoczki, A. Preaudet, C. Tsui, R.L. Ward, N.J. Hawkins, J. Desai, P. Gibbs, O.M. Sieber
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Sakthianandeswaren, M.J. Parsons, D. Mouradov, R.N. MacKinnon, C. Love, R.N. Jorissen, L. Whitehead, C.J. Nowell, M. Ernst, M. Buchert, O.M. Sieber
Writing, review, and/or revision of the manuscript: A. Sakthianandeswaren, M.J. Parsons, D. Mouradov, C.J. Nowell, R.L. Ward, J. Desai, P. Gibbs, M. Ernst, M. Buchert, O.M. Sieber
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Sakthianandeswaren, S. Liu, S. Li, C. Tsui, N.J. Hawkins, O.M. Sieber
Study supervision: O.M. Sieber
The authors thank the Victorian Cancer Biobank and BioGrid Australia for provision of patient specimens and data, and Prof. M. Schwab at the DKFZ for access to colorectal cancer cell lines. This work was supported by an NHMRC Project Grant (APP1079364 to O.M. Sieber, M. Buchert, J. Desai, and I. Street), a Cancer Council Victoria Grant-in-Aid (APP1060964 to O.M. Sieber and R.L. Ward), the Ludwig Institute for Cancer Research (O.M. Sieber), the Cancer Therapeutics Cooperative Research Centre (I. Street), an NHMRC Senior Research Fellowship (APP1136119 to O.M. Sieber), a Cancer Therapeutics CRC Top Up PhD Scholarship (M.J. Parsons), and the Victorian Government's Operational Infrastructure Support Program (O.M. Sieber).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.