Pancreatic cancer genomes reveal aberrations in axon guidance pathway genes

Abstract

Pancreatic cancer is a highly lethal malignancy with few effective therapies. We performed exome sequencing and copy number analysis to define genomic aberrations in a prospectively accrued clinical cohort (n = 142) of early (stage I and II) sporadic pancreatic ductal adenocarcinoma. Detailed analysis of 99 informative tumours identified substantial heterogeneity with 2,016 non-silent mutations and 1,628 copy-number variations. We define 16 significantly mutated genes, reaffirming known mutations (KRAS, TP53, CDKN2A, SMAD4, MLL3, TGFBR2, ARID1A and SF3B1), and uncover novel mutated genes including additional genes involved in chromatin modification (EPC1 and ARID2), DNA damage repair (ATM) and other mechanisms (ZIM2, MAP2K4, NALCN, SLC16A4 and MAGEA6). Integrative analysis with in vitro functional data and animal models provided supportive evidence for potential roles for these genetic aberrations in carcinogenesis. Pathway-based analysis of recurrently mutated genes recapitulated clustering in core signalling pathways in pancreatic ductal adenocarcinoma, and identified new mutated genes in each pathway. We also identified frequent and diverse somatic aberrations in genes described traditionally as embryonic regulators of axon guidance, particularly SLIT/ROBO signalling, which was also evident in murine Sleeping Beauty transposon-mediated somatic mutagenesis models of pancreatic cancer, providing further supportive evidence for the potential involvement of axon guidance genes in pancreatic carcinogenesis.

Pancreatic cancer is the fourth leading cause of cancer death, with an overall 5-year survival rate of <5%, statistics that have not changed in almost 50 years¹. Advances in neoadjuvant and adjuvant chemotherapeutic regimens have resulted in some improvement in outcome, but pancreatectomy remains the single most effective treatment modality for pancreatic cancer, and offers the only potential for cure. Only 20% of patients present with localized, non-metastatic disease which is suitable for resection². Those who undergo resection and receive adjuvant therapy have a median survival of 12–22 months and a 5-year survival of 20–25%³. Existing systemic therapies are only modestly effective and the median survival for patients with metastatic disease remains 6 months. Genomic characterization of pancreatic ductal adenocarcinoma (PDAC), which accounts for over 90% of pancreatic cancer, has so far focused on targeted polymerase chain reaction (PCR)-based exome sequencing of primary and metastatic lesions propagated as xenografts or cell lines⁴. A deeper understanding of the underlying molecular pathophysiology of the clinical disease is needed to advance the development of effective therapeutic and early detection strategies.

Clinical cohort

A cohort of 142 consecutive patients with primary operable, untreated PDAC who underwent pancreatectomy with curative intent (pre-operative clinical stages I and II) were recruited, and consent was obtained for genomic sequencing through the Australian Pancreatic Cancer Genome Initiative (APGI), the Baylor College of Medicine Pancreatic Cancer Genome Project and the Ontario Institute for Cancer Research Pancreatic Cancer Genome Study (ABO collaboration) between June 2005 and June 2011 as part of the International Cancer Genome Consortium (ICGC)⁵. Detailed clinico-pathological characteristics of the cohort demonstrated features typical of resected PDAC with regard to tumour size, grade, lymph node metastasis and survival when compared to multiple retrospectively acquired cohorts^6–8, defining the accrued population as representative of the clinical disease in the community (Supplementary Table 1 and Supplementary Fig. 1).

Cellularity and mutation detection

A major challenge in genomic sequencing is the low malignant epithelial cell content of many cancers, which can adversely impact on the sensitivity of mutation detection. Most sequencing studies so far have used samples with >70% tumour cellularity, or cell lines/xenografts^4,9. To implement genomic sequencing approaches in clinical practice, it is imperative to efficiently and accurately detect actionable mutations in diagnostic clinical samples. We devised methodologies to overcome the challenges associated with extensive desmoplastic stroma that is characteristic of the majority of PDAC, and these strategies facilitated the discovery of novel molecular mechanisms in the pathophysiology of this disease. The cellularity of each primary sample was estimated through pathological review, deep amplicon-based sequencing of exons 2 and 3 of KRAS (average depth of 1,000×), and single nucleotide polymorphism (SNP) array-based cellularity estimates using a novel algorithm (qpure)¹⁰. KRAS mutations were identified in 93% of 142 cases and tumour cellularity ranged from 5% to 85% with a mean of 38% (Supplementary Table 2, Supplementary Figs 2 and 3, and Supplementary Methods).

To inform cellularity thresholds for subsequent analyses, we defined the impact of stromal DNA content on mutation detection by exome capturing and sequencing different mixtures of cancer cell line and matched germline DNA (100%, 80%, 60%, 40%, 20% and 10% cell line DNA) when sequenced to a depth of 70× coverage. Using these data as a standard, the median sensitivity to detect true positives across all samples in the cohort with greater than 20% epithelial cellularity was estimated at 45% (Supplementary Table 3). An informative cohort of 99 patients who had greater than 20% cellularity and/or ≥10 validated somatic mutations was taken forward for further analysis.

Mutation detection and CNV analysis

We performed hybrid-selection-based capture and sequencing of the entire exomes of tumour and matched normal DNA derived from all 142 patients using a combination of capture systems and next-generation sequencing platforms (see Supplementary Methods). The sequence depths at each site (APGI 65×, BCM 104× and OICR 205×) were adopted to ensure suitable sensitivity across their respective cohorts (Supplementary Table 3). In the informative 99 samples, we detected 2,627 high-confidence mutations, 2,016 of which were non-silent (Table 1). A total of 1,502 of these events (1,350 non-silent) were independently validated via an orthogonal sequencing method (see Supplementary Methods). The average number of mutations detected per patient was 26 (range 1–116), consistent with the expected sensitivity based on cellularity estimates and previous studies^4,11 (Supplementary Table 2). We confirmed the high prevalence of genetic aberrations known to be important in PDAC and observed mutations in 38 of the 79 genes (48% overlap) that occurred more than once previously reported by ref. 4, and 186 of all 998 mutated genes (19% overlap) in that study. We also defined a large number of novel mutations (1,456 genes), most of which occurred at low frequency (see Supplementary Tables 4– 6 and Supplementary Fig. 4 for detailed comparisons). The observed transversion/transition rates in the cohort correlated closely with those previously reported in PDAC cell lines and xenografts (Supplementary Table 7).

Table 1.

Mutations in pancreatic ductal adenocarcinoma (n = 99)

Mutation class	Total
Missense	1,684
Nonsense	99
Splice site	89
Insertion/deletion	144
Non-silent	2,016
Silent	611

Significant mutated gene analysis¹² of genes with non-silent mutations that occurred in 2 or more individual cancers identified 16 genes in the top 20 mutated genes in 2 of 3 stringent analytical approaches (Table 2, Supplementary Table 8 and Supplementary Methods) and reaffirmed the importance of mutations known to occur in PDAC: KRAS, TP53, CDKN2A, SMAD4, MLL3, TGFBR2, ARID1A and SF3B1. Novel significantly mutated genes included additional genes involved in chromatin modification (EPC1 and ARID2) and ATM, recently implicated as a PDAC susceptibility gene through bi-allelic inactivation in a case of familial PDAC (germline mutation and loss of heterozygosity (LOH) in the tumour)¹³. Aberrations of ATM occurred in 8% of our cohort (mutated in 5%, LOH or loss in 5%, with two patients exhibiting both mutation and LOH or loss) and mutations detected in other genes not previously reported: ZIM2, MAP2K4, NALCN, SLC16A4 and MAGEA6 (Table 2). GISTIC2.0¹⁴ identified 30 genes affected by copy-number alterations (Q value <0.0001) and included losses of CDKN2A and SMAD4 (Supplementary Table 4).

Table 2.

Significantly mutated genes in pancreatic ductal adenocarcinoma

Gene symbol	Gene name and protein function	SB mutagenesis*	shRNA†
KRAS	Oncogene; GTPase; activation of MAPK activity	Yes	Yes
TP53	Tumour suppressor p53; DNA damage response	–	Yes
CDKN2A	Cyclin-dependent kinase inhibitor 2A; G1/S transition of mitotic cell cycle; tumour suppressor	Yes	–
SMAD4	Mothers against decapentaplegic homologue 4; BMP signalling pathway	Yes	Yes
MLL3	Myeloid/lymphoid or mixed-lineage leukaemia protein 3; DNA binding; regulation of transcription	Yes	Yes
TGFBR2	Transforming growth factor-β receptor type II; regulation of growth	Yes	–
ARID1A	AT-rich interactive domain-containing protein 1A; SWI/SNF complex; chromatin modification	Yes	Yes
ARID2	AT-rich interactive domain-containing protein 2; chromatin modification	Yes	–
EPC1	Enhancer of polycomb homologue 1; histone acetylation	Yes	–
ATM	Ataxia telangiectasia mutated; DNA damage response	–	Yes
SF3B1	Splicing factor 3B subunit 1; nuclear mRNA splicing	–	Yes
ZIM2	Zinc finger imprinted 2; regulation of transcription	–	Yes
MAP2K4	Dual specificity mitogen-activated protein kinase kinase 4; Toll-like receptor signalling pathway	Yes	Yes
NALCN	Sodium leak channel non-selective protein; sodium channel activity	–	Yes
SLC16A4	Solute carrier family 16 member 4; monocarboxylate transporter	–	Yes
MAGEA6	Melanoma-associated antigen 6; protein binding	–	ND

ND, not determined.

^*Significant insertion sites in two independent Sleeping Beauty mutagenesis screens^17,18.

^†In vitro shRNA screens in 102 cancer cell lines with effect on cell survival¹⁹.

Pathways in pancreatic cancer

To better understand potential underlying mechanisms of importance in PDAC, we performed a series of pathway analyses using genes that were recurrently mutated in two or more individuals using GeneGO¹⁵, and identified mechanisms known to be importantin cancer: G1/S checkpoint machinery (P = 1.49 × 10^–3), apoptosis (P = 1.32 × 10^–4), regulation of angiogenesis (P = 7.72 × 10^–4) and TGF-β signalling (P = 9.50 × 10^–4). Interestingly, novel gene signatures were enriched in our cohort, including axon guidance (P = 5.30 × 10^–5) (Supplementary Table 9). The inclusion of mutation data for 24 cases from ref. 4 strengthened the association of axon guidance (P = 3.3 × 10^–7), and was more evident still when all mutated genes in our data set were used as input (P = 4.67 × 10^–8).

Functional relevance of genomic events

Differentiating somatic driving events of carcinogenesis from passenger mutations is a major challenge in cancer genomics¹⁶. Despite significant advances in computational algorithms, experimental evidence of functional relevance is paramount. We used data from three published experimental biological screens to infer functional consequences for the individual genomic events and the pathways we identified. These included data from two independent Sleeping Beauty transposon (SB) mutagenesis screens in Kras transgenic mouse models of PDAC^17,18 and an in vitro short hairpin RNA (shRNA) screen which examined the consequences of downregulating 11,194 putative cancer genes on survival in a panel of 102 cell lines (13 pancreatic)¹⁹ (Supplementary Methods and Supplementary Figs 5 and 6). Data from these screens confirmed the functional importance of KRAS, TP53, CDKN2A and SMAD4 mutations and attributed potential functional relevance to most significantly mutated genes—MLL3, TGFBR2, SF3B1, EPC1, ARID1A, ARID2, MAP2K4, ATM, NALCN, ZIM2, SLC16A4 (Table 2)—and many genes mutated at low frequency (Supplementary Table 4).

Pathway analysis of high confidence insertions in SB transposon mutagenesis screens demonstrated enrichment for axon guidance genes (P = 1.6 × 10^–3), providing independent supportive evidence for a potential role in the pathogenesis of PDAC. In these screens, 14 genes involved in axon guidance pathways were detected (5 genes common to both). In addition, a further 32 genes were mutated in at least one SB pancreatic tumour (out of 21) but did not meet the significance threshold with the stringent analyses that were applied¹⁷ (Supplementary Tables 10 and 11).

Axon guidance pathway genes

The class of genes traditionally described for their roles in axon guidance (semaphorins, slits, netrins and ephrins) are important regulators of normal neuronal migration and positioning during embryonic development. More recently, they have been implicated in cancer cell growth, survival, invasion and angiogenesis²⁰; however, the incidence of aberrations in these genes in cancer is largely unknown. We identified recurrent mutations and copy-number variations (CNVs) of axon guidance pathway genes in this cohort (Fig. 1 and Supplementary Table 4): SLIT2 and ROBO2 mutations were present in 5% of patients, with focal copy-number losses of ROBO1, and SLIT2 detected by GISTIC2.0 analysis and confirmed by manual review, potentially having an impact on a further 15% of the cohort, suggesting that aberrant SLIT/ROBO signalling is potentially a common feature of PDAC (Figs 1 and 2). In addition, we used targeted PCR-based sequencing of an additional 30 cases of PDAC for axon guidance genes and identified mutations in ROBO1 in two patients and additional mutations in SLIT2 and ROBO2 (one patient each). Low mRNA expression of the ROBO2 receptor was associated with poor patient survival (P = 0.04). Furthermore, high mRNA expression of ROBO3, a known inhibitor of ROBO2 signalling²¹, demonstrated an appropriate reciprocal inverse association with poor survival (P < 0.006) (Fig. 2).

Figure 1. Mutations and copy number variation in axon guidance genes.

Axon guidance pathway genes with recurrent mutations and/or copy-number changes defined by GISTIC2.0 analysis (Q < 0.2), and manually reviewed for focal alterations. a, SNV and CNV frequency per patient with gene-centric summary (left) and patient-centric summary (top); numbers of patients with mutations and proportion of each event are presented. Please see Supplementary Table 4 for further details. b, Clinico-pathological variables for individual patients. APGI, Australian Pancreatic Cancer Genome Initiative; BCM, Baylor College of Medicine; IPMN, intraductal papillary mucinous neoplasm; Mod, moderately differentiated; OICR, Ontario Institute for Cancer Research; PDAC, pancreatic ductal adenocarcinoma; Undiff, undifferentiated.

Figure 2. SLIT/ROBO signalling in pancreatic ductal adenocarcinoma.

a, SLIT/ROBO signalling normally enhances β-catenin complex formation with E-cadherin and suppresses WNT signalling activity. Loss of ROBO1/2 signalling promotes stabilization of β-catenin, which decreases E-cadherin complex formation and cell adhesion and augments WNT signalling activity through increased nuclear translocation of β-catenin. In addition, SLIT/ROBO signalling can downregulate MET signalling activity; loss of ROBO signalling activity promotes MET signalling downstream and may have an impact on therapeutic strategies aimed at inhibiting MET activity at the receptor level. (Adapted from ref. 20.) Aberrations in SLIT2 and/or ROBO1/2 affected 23% of patients (6% mutated with 1 patient showing mutations in both SLIT2 and ROBO2), with 18% demonstrating CNV corresponding to loss of the gene. b, c, High expression of SLIT receptor ROBO2 was associated with a better prognosis (b), and high expression of ROBO3, an inhibitor of ROBO2, showed an inverse relationship, with high levels associated with poor survival (c). HR, hazard ratio.

Class 3 semaphorins (SEMA3A and SEMA3E) exhibited significant amplification in 18% of patients and an additional 3% harboured mutations (Fig. 1). Semaphorins signal through neuropilin and plexin receptors to elicit their effects²². SEMA3A amplification correlated with high mRNA expression on microarray (P = 0.03), and high mRNA expression of SEMA3A and PLXNA1, another molecule central to semaphorin signalling, were both associated with poor patient survival on univariate analysis (Fig. 3a), and were independently prognostic on multivariate analyses with clinico-pathological variables (Supplementary Table 12).

Figure 3. Axon guidance genes in human and murine pancreatic ductal adenocarcinoma.

a, Kaplan–Meier survival curves showing co-segregation of aberrant expression of components of semaphorin signalling with outcome. Amplification at SEMA3A and PLXNA1 loci was associated with high mRNA expression and both are independent poor prognostic factors. b, Quantitative RT–PCR for components of semaphorin and SLIT/ROBO signalling in murine models of early (acinar-to-ductal metaplasia (ADM) and pancreatic injury) and established PDAC in genetically engineered mice with a Pdx1-promoter-driven activating mutation of Kras and mutant Tp53 allele (Pdx1-Cre; LSL-Kras^G12D; LSL-Trp53^R172H). Error bars represent standard error of the mean (see Supplementary Table 15 for details).

To elucidate further the significance of the observed CNV events, we reviewed copy number, CNV segment size and changes in heterozygosity of axon guidance genes in a recent independent CNV analysis of 39 fine-needle aspiration biopsies²³ and the 16 PDAC cell lines in the CONAN database (http://www.sanger.ac.uk/cosmic)²⁴. Overall, the predominant changes recapitulated our studies, showing frequent focal losses within genes involved in SLIT/ROBO signalling, and gains in genes involved in canonical semaphorin signalling (Supplementary Tables 4, 13 and 14).

To assess whether dysregulation of axon guidance genes is associated with early neoplastic transformation, as are many developmental signalling pathways, we examined mRNA expression in murine models of early pancreatic carcinogenesis (in vitro acinar-to-ductal metaplasia and in vivo pancreatic injury). Expression levels of components of SLIT/ROBO and semaphorin signalling changed progressively from normal pancreas, through acinar-to-ductal metaplasia and pancreatic injury to genetically engineered murine PDAC, indicating a role for the dysregulation of these axon guidance genes in tumour initiation and progression (Fig. 3b and Supplementary Table 15).

Discussion

We devised methodologies to optimize mutation detection for clinical samples in a large cohort of patients and reaffirm known mutations in PDAC, better define their prevalence in a large cohort of early PDAC, and identify potential novel drivers in this disease. Somatic mutations in ATM were identified in a significant proportion of patients (8%), highlighting the importance of BRCA-mediated DNA damage repair mechanisms in sporadic PDAC as well as familial disease¹³. Previously, mutations in individual genes involved in chromatin remodelling such as ARID1A²⁵ have been described and additional genes identified here (EPC1 and ARID2) infer that chromatin remodelling may have an important role in PDAC, along with other cancer types²⁶.

Novel mutations in genes traditionally described for their roles in axon guidance were also observed by a combination of genomic data and supportive experimental evidence from independent murine SB mutagenesis screens. Axon guidance is integral to organogenesis, regeneration, wound healing and other basic cellular processes^22,27. The widespread genomic aberrations observed here in axon guidance genes suggests that they may have a role in PDAC, joining mounting evidence in other cancers^20,28, including a recent report demonstrating ROBO2 mutations in liver-fluke-associated cholangiocarcinoma²⁹. In addition, evidence from cancers of the lung, breast, kidney and cervix implicate aberrant SLIT/ROBO signalling in carcinogenesis²⁰; Robo1 knockout mice develop bronchial hyperplasia and focal dysplasia, and inactivation of Slit2 and Slit3 leads to the development of hyperplastic disorganized lesions in the breast²⁰. Upregulation of MET and WNT signalling have important roles in PDAC, and recent data indicate that SLIT/ROBO signalling modulates MET and WNT signalling activity through CDC42 and β-catenin, respectively²⁰. Loss of SLIT/ROBO signalling can potentially be an alternative mechanism for deregulating these pathways downstream of their receptors, and in addition could influence the activity of inhibitors that target these upstream components, for example, MET inhibitors (Fig. 2).

Class 3 semaphorins are the only secreted semaphorins in vertebrates. They regulate cell growth, invasiveness and angiogenesis, and are highly expressed in metastatic cells in many cancer types^30,31. Although aberrant semaphorin signalling in cancer seems to be organ specific³², our finding that high expression of SEMA3A and its receptor PLXNA1 co-segregates with poor patient survival is supported by a previous study that reported this association and also demonstrated promotion of invasiveness of PDAC cell lines by SEMA3A³¹. Therapeutics targeting molecules involved in axon guidance have been developed as potential strategies to facilitate neuronal regeneration after injury³³, but are yet to be assessed for their role in cancer treatment.

As illustrated here, global genomic analysis of large, well-annotated and clinically homogeneous cohorts of patients can identify mechanisms that are common among genomically diverse cancers, and will be pivotal in the development of novel therapeutic strategies that are guided by the determination of the molecular phenotype of individual patients³⁴. Future work will be required to determine which key components, when damaged, drive the disease, and these mechanisms will need to be assessed in molecularly well-characterized preclinical models³⁵. The potential therapeutic strategies identified will then require testing in appropriate clinical trials that are specifically designed to target subsets of patients stratified according to well-defined molecular markers^36,37.

METHODS SUMMARY

Sample acquisition and processing

Samples used were prospectively acquired and restricted to primary operable, non-pretreated pancreatic ductal adenocarcinoma. Representative sections were reviewed independently by at least one additional pathologist with specific expertise in pancreatic diseases. Samples either had full face frozen sectioning performed in optimal cutting temperature (OCT) medium, or the ends excised and processed in formalin to verify the presence of carcinoma in the sample to be sequenced and to estimate the percentage of malignant epithelial nuclei in the sample relative to stromal nuclei. Macrodissection was performed if required to excise areas that did not contain malignant epithelium.

Sequencing

Cellularity of each tumour sample was estimated with pathology review, deep sequencing of KRAS and a method developed using genome-wide SNP array data (qpure¹⁰). Exon capture was performed using the SureSelect II or Nimblegen capture methods and paired-end sequenced on the SOLiD (v4) or GAII/HiSeq platforms. Somatic mutations were called and then verified on the Ion Torrent Personal Genome Machine (Life Technologies Corporation) and 454 (Hoffman–La Roche Limited).

Analysis

Significantly mutated genes were identified using the Genome MuSiC package¹². DNA copy number analyses were performed using the Illumina HumanOmni1 Quad genotyping arrays and GenoCN software. Recurrent and significant copy number changes were identified using GISTIC2.0¹⁴. Functional enrichment of gene categories was assessed using the Metacore package (Thomson-Reuters Corporation) and the MSigDB v3.0 database³⁸. All sample information and data for mutation, copy number and expression analyses were submitted to the ICGC DCC at http://dcc.icgc.org/. A complete description of the materials and methods including approvals for human research and animal experimentation is provided in Supplementary Information.

Australian Pancreatic Cancer Genome Initiative

The Kinghorn Cancer Centre, Garvan Institute of Medical Research Andrew V. Biankin¹, Amber L. Johns¹, Amanda Mawson¹, David K. Chang¹, Christopher J. Scarlett¹, Mary-Anne L. Brancato¹, Sarah J. Rowe¹, Skye L. Simpson¹, Mona Martyn-Smith¹, Michelle T. Thomas¹, Lorraine A. Chantrill¹, Venessa T. Chin¹, Angela Chou¹, Mark J. Cowley¹, Jeremy L. Humphris¹, Marc D. Jones¹, R. Scott Mead¹, Adnan M. Nagrial¹, Marina Pajic¹, Jessica Pettit¹, Mark Pinese¹, Ilse Rooman¹, Jianmin Wu¹, Jiang Tao¹, Renee DiPietro¹, Clare Watson¹, Rachel Wong¹, Andreia V. Pinho¹, Marc Giry-Laterriere¹, Roger J. Daly¹, Elizabeth A. Musgrove¹, Robert L. Sutherland¹; Queensland Center for Medical Genomics, Institute for Molecular Bioscience Sean M. Grimmond², Nicola Waddell², Karin S. Kassahn², David K. Miller², Peter J. Wilson², Ann-Marie Patch², Sarah Song², Ivon Harliwong², Senel Idrisoglu², Craig Nourse², Ehsan Nourbakhsh², Suzanne Manning², Shivangi Wani², Milena Gongora², Matthew Anderson², Oliver Holmes², Conrad Leonard², Darrin Taylor², Scott Wood², Qinying Xu², Katia Nones², J. Lynn Fink², Angelika Christ², Tim Bruxner², Nicole Cloonan², Felicity Newell², John V. Pearson²; Royal North Shore Hospital Jaswinder S. Samra³, Anthony J. Gill³, Nick Pavlakis³, Alex Guminski³, Christopher Toon³; Bankstown Hospital Andrew V. Biankin⁴, Ray Asghari⁴, Neil D. Merrett⁴, David K. Chang⁴, Darren A. Pavey⁴, Amitabha Das⁴; Liverpool Hospital Peter H. Cosman⁵, Kasim Ismail⁵, Chelsie O'Connor⁵; Westmead Hospital Vincent W. Lam⁶, Duncan McLeod⁶, Henry C. Pleass⁶, Arthur Richardson⁶, Virginia James⁶; Royal Prince Alfred Hospital James G. Kench⁷, Caroline L. Cooper⁷, David Joseph⁷, Charbel Sandroussi⁷, Michael Crawford⁷, James Gallagher⁷; Fremantle Hospital Michael Texler⁸, Cindy Forrest⁸, Andrew Laycock⁸, Krishna P. Epari⁸, Mo Ballal⁸, David R. Fletcher⁸, Sanjay Mukhedkar⁸; Sir Charles Gairdner Hospital Nigel A. Spry⁹, Bastiaan DeBoer⁹, Ming Chai⁹; St John of God Healthcare Nikolajs Zeps¹⁰, Maria Beilin¹⁰, Kynan Feeney¹⁰; Royal Adelaide Hospital Nam Q. Nguyen¹¹, Andrew R. Ruszkiewicz¹¹, Chris Worthley¹¹, Chuan P. Tan¹¹, Tamara Debrencini¹¹; Flinders Medical Centre John Chen¹², Mark E. Brooke-Smith¹², Virginia Papangelis¹²; Greenslopes Private Hospital Henry Tang¹³, Andrew P. Barbour¹³; Envoi Pathology Andrew D. Clouston¹⁴, Patrick Martin¹⁴; Princess Alexandria Hospital Thomas J. O'Rourke¹⁵, Amy Chiang¹⁵, Jonathan W. Fawcett¹⁵, Kellee Slater¹⁵, Shinn Yeung¹⁵, Michael Hatzifotis¹⁵, Peter Hodgkinson¹⁵; Austin Hospital Christopher Christophi¹⁶, Mehrdad Nikfarjam¹⁶, Angela Mountain¹⁶; Victorian Cancer Biobank¹⁷; Johns Hopkins Medical Institutes James R. Eshleman¹⁸, Ralph H. Hruban¹⁸, Anirban Maitra¹⁸, Christine A. Iacobuzio-Donahue¹⁸, Richard D. Schulick¹⁸, Christopher L. Wolfgang¹⁸, Richard A. Morgan¹⁸, Mary B. Hodgin¹⁸; ARC-NET Center for Applied Research on Cancer Aldo Scarpa¹⁹, Rita T. Lawlor¹⁹, Paola Capelli¹⁹, Stefania Beghelli¹⁹, Vincenzo Corbo¹⁹, Maria Scardoni¹⁹, Paolo Pederzoli¹⁹, Giampaolo Tortora¹⁹, Claudio Bassi¹⁹; University of California, San Francisco Margaret A. Tempero²⁰

¹The Kinghorn Cancer Centre, Cancer Research Program, Garvan Institute of Medical Research, 370 Victoria Street, Darlinghurst, Sydney, New South Wales 2010, Australia. ²Queensland Center for Medical Genomics, Institute for Molecular Bioscience, University of Queensland, St Lucia, Queensland 4072, Australia. ³Royal North Shore Hospital, Westbourne Street, St Leonards, New South Wales 2065, Australia. ⁴Bankstown Hospital, Eldridge Road, Bankstown, New South Wales 2200, Australia. ⁵Liverpool Hospital, Elizabeth Street, Liverpool, New South Wales 2170, Australia. ⁶Westmead Hospital, Hawkesbury and Darcy Roads, Westmead, New South Wales 2145, Australia. ⁷Royal Prince Alfred Hospital, Missenden Road, Camperdown, New South Wales 2050, Australia. ⁸Fremantle Hospital, Alma Street, Fremantle, Western Australia 6959, Australia. ⁹Sir Charles Gairdner Hospital, Hospital Avenue, Nedlands, Western Australia 6009, Australia. ¹⁰St John of God Healthcare, 12 Salvado Road, Subiaco, Western Australia 6008, Australia. ¹¹Royal Adelaide Hospital, North Terrace, Adelaide, South Australia 5000, Australia. ¹²Flinders Medical Centre, Flinders Drive, Bedford Park, South Australia 5042, Australia. ¹³Greenslopes Private Hospital, Newdegate Street, Greenslopes, Queensland 4120, Australia. ¹⁴Envoi Pathology, 1/49 Butterfield Street, Herston, Queensland 4006, Australia. ¹⁵Princess Alexandria Hospital, Cornwall Street & Ipswich Road, Woolloongabba, Queensland 4102, Australia. ¹⁶Austin Hospital, 145 Studley Road, Heidelberg, Victoria 3084, Australia. ¹⁷Victorian Cancer Biobank, 1 Rathdowne Street, Carlton, Victoria 3053, Australia. ¹⁸Johns Hopkins Medical Institutes, 600 North Wolfe Street, Baltimore, Maryland 21287, USA. ¹⁹ARC-NET Center for Applied Research on Cancer, University of Verona, Via dell'Artigliere, 19 37129 Verona 37134, Italy. ²⁰University of California, San Francisco, 1600 Divisadero Street, San Francisco, California 94115, USA.