ARTICLE

DOI: 10.1038/s41467-018-06366-z OPEN Metastatic displays higher mutation rate and tumor heterogeneity than primary tumors

Sudheer Kumar Gara1, Justin Lack2, Lisa Zhang1, Emerson Harris1, Margaret Cam2 & Electron Kebebew1,3

Adrenocortical (ACC) is a rare cancer with poor prognosis and high mortality due to metastatic disease. All reported genetic alterations have been in primary ACC, and it is 1234567890():,; unknown if there is molecular heterogeneity in ACC. Here, we report the genetic changes associated with metastatic ACC compared to primary ACCs and tumor heterogeneity. We performed whole-exome sequencing of 33 metastatic tumors. The overall mutation rate (per megabase) in metastatic tumors was 2.8-fold higher than primary ACC tumor samples. We found tumor heterogeneity among different metastatic sites in ACC and discovered recurrent mutations in several novel . We observed 37–57% overlap in genes that are mutated among different metastatic sites within the same patient. We also identified new therapeutic targets in recurrent and metastatic ACC not previously described in primary ACCs.

1 Endocrine Oncology Branch, National Cancer Institute, National Institutes of Health, Bethesda, MD 20892, USA. 2 Center for , Collaborative Resource, National Cancer Institute, National Institutes of Health, Bethesda, MD 20892, USA. 3 Department of Surgery and Stanford Cancer Institute, Stanford University, Stanford, CA 94305, USA. Correspondence and requests for materials should be addressed to E.K. (email: [email protected])

NATURE COMMUNICATIONS | (2018) 9:4172 | DOI: 10.1038/s41467-018-06366-z | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-06366-z

drenocortical carcinoma (ACC) is a rare malignancy with types including primary ACC from the TCGA to understand our A0.7–2 cases per million per year1,2. The five-year survival findings in the context of not only primary ACC but also other rate for patients with resectable tumors is less than 30%. primary cancer types (Fig. 1c). Next, we tested the idea whether Particularly, patients with metastatic disease have a median sur- primary ACC tumors of the TCGA from patients with metastasis vival of less than 1 year. Unfortunately, most patients with ACC has any effect on mutation burden and found no significant have locally advanced cancer or metastasis at the time of diag- difference in the mutation rate by tumor stage supporting our nosis. Deaths due to ACC are associated with metastatic disease. findings that the mutation burden is higher in metastatic ACC as Our understanding of the pathogenesis of ACC has been compared to primary ACC (Fig. 1d). We also analyzed the rela- greatly improved over the past decade. Molecular studies have tive proportions of the six possible base-pair substitutions in the demonstrated that TP53 inactivating mutations, CTNNB1 acti- transcribed and non-transcribed strands among all the metastatic vating mutations, IGF2 overexpression, damaging mutations in ACC tumors to understand their relevance with other solid ZNRF3, and high-level amplification of TERT are common and tumors including primary ACC tumors. Similar to primary ACC key drivers of ACC3–6. Furthermore, the international con- from the TCGA data and most other solid tumors13,14, metastatic sortium of The Cancer Genome Atlas (TCGA) has identified ACC were characterized by a predominance of C > T transitions additional ACC driver genes including PRKAR1A, RPL22, TERF2, (Fig. 1e). However, we did not notice any significant difference of CCNE1, and NF17. This study further showed that a whole- T > C mutations between transcribed and non-transcribed genome doubling event is a marker for ACC progression and strands in metastatic ACC, unlike in primary ACC tumors prognosis7. Multiple studies have reported that activating and (Fig. 1e, f). inactivating alterations in the TP53/Rb and WNT pathways are We next analyzed genes recurrently mutated in multiple key molecular events in the pathogenesis of ACC4,7. However, all metastatic ACC tumor samples and compared their status in of the above studies are confined to primary ACC, and death primary ACC from the TCGA (Fig. 1g). Similar to primary ACC, from ACC is primarily due to recurrent or metastatic disease. we observed that CTNNB1, DNHD1, and TTN were frequently Despite recent advances in our understanding of ACC, the mutated in metastatic ACC. Nevertheless, we have also identified therapeutic options for ACC are still limited, and treatment with genes such as ENTHD1, HELZ2, PCDH12, SHANK1, and WDR66 combination mitotane-etoposide, doxorubicin, and cisplatin that were more frequently mutated in metastatic ACC but not in results in low response rates8,9. Studies on targeted therapeutic primary tumors (Fig. 1g). Next, we selected the frequently options using IGF-targeting agents or tyrosine kinase inhibitors mutated genes in primary ACC and compared their mutation have not shown good efficacy10–12. It is evident that most of the status in metastatic ACC (Fig. 1h). As expected, the majority of advanced or late-stage or treatment-resistant tumors the genes that are mutated in primary ACC were also mutated in adapt differently and gain additional mutations. Mutations ana- metastatic ACC (Fig. 1h). We also validated 5/5 tested mutations lysis of late-stage cancer tumor sites and treatment-resistant using droplet digital polymerase chain reaction (ddPCR) and 72 tumors has shown biologically informative genomic alterations. of 77 mutations using Sanger sequencing (Supplementary Figs. 1, Given this, it is important to understand the nature of recurrent 2). In addition, we noticed that two metastatic ACCs from the and metastatic ACC to inform candidate genetic changes same patient (Case 1) had a strong hypermutation phenotype involved in cancer progression and to identify therapeutic targets. (Supplementary Table 1), and two additional patients had > /10 Therefore, our study objectives were to analyze the genomic single-nucleotide polymorphisms (SNPs) per megabase compared landscape of metastatic ACC, the nature of different metastatic to a median of 2.63 SNPs per megabase for all metastatic ACCs in tumor sites (tumor heterogeneity), and their commonality and our data set. To determine if these hypermutation phenotypes are differences compared to primary ACC. driven by DNA repair defects, we examined copy number variations (CNVs), somatic SNPs and insertions and deletions (INDELs), and germline SNPs and INDELs for putative loss-of- Results function (LoF) mutations (homozygous deletions; nonsense, non- Metastatic ACC has a higher mutation rate. We performed frameshift INDELs > = 3 amino acids; and frameshift mutations whole-exome sequencing on 33 histologically confirmed meta- in the Wood DNA repair genes)15. For the two tumor samples static ACCs (, liver, pancreas, , peritoneum, and other with the highest mutation rates, the only LoF mutation shared tissue sites with matched peripheral blood sample DNA) collected between both tumors is a heterozygous 9-base germline insertion from 14 patients with metastatic ACC. We identified in 1of MSH3 (Supplementary Table 2 and Supplementary 15,321 somatic mutations in the coding region of the genome in Fig. 3), which was amplified in copy number and had (allele 33 tumors; 5928 nonsynonymous mutations and 9393 silent frequency > 0.9 in both samples) complete loss of heterozygosity mutations (Supplementary Table 1). The mean somatic mutation (LOH) in both tumor samples. Somatic INDELs overlapping this rate in the coding region of metastatic ACC was 10.17 mutations mutation were also detected for three primary ACCs, and while it per megabase, with 3.93 and 6.24 mutations per megabase for is unclear if any of these samples have LOH and fixation of these nonsynonymous and silent mutations, respectively, (Supple- alterations, one of the three samples (TCGA-PK-A5HB-01) has a mentary Table 1). We compared the overall somatic mutation hypermutation phenotype, possessing the second-highest muta- burden between our cohort of metastatic ACC and the primary tion rate of the 92 primary ACCs available (24). For the ACC ACC tumors from the TCGA, reprocessed through our somatic metastasis with the second-highest mutation rate (Tumor 20; variant calling pipeline (see methods) to eliminate pipeline-driven Supplementary Table 1), no germline LoF events were detected, difference in variant call rates. The ACC metastatic tumors had but multiple somatic LoF events were detected that may be 2.8-fold higher median mutation rate compared to primary ACC contributing to its hypermutation phenotype. We detected a focal (Fig. 1a). Since the mutation rate was compared between primary homozygous deletion of MSH6 (Supplementary Table 2 and ACC from the TCGA cohort to the metastatic cohort, we per- Supplementary Fig. 4), which was not detected in any of the 92 formed the same analysis in one matched primary ACC with a primary ACC tumors available in cBioPortal (Supplementary metastatic lung ACC tumor from the same patient. We found Table 2 and Supplementary Fig. 3). In addition, we detected a that metastatic lung ACC had a threefold higher mutation rate frameshift mutation in ATM that has become homozygous compared to the primary ACC (Fig. 1b). In addition, we com- through an LoH event (Supplementary Table 2 and Supplemen- pared the mutation rate of metastatic ACC with other cancer tary Fig. 4B) and is fixed in the tumor sample (allele frequency =

2 NATURE COMMUNICATIONS | (2018) 9:4172 | DOI: 10.1038/s41467-018-06366-z | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-06366-z ARTICLE

a 120 100 Metastasis 80 Primary tumor 80 40 20 10

5 Total mutations/Mb 0

b c Lung metastatic tumor 10,000 Primary ACC 250

200

150 1000

100

50

# of non-silent mutations 0 100 d 40 ANOVA: P = 0.71 Log10 (Mutations per sample)

30 10

20

10 1 Mutations per megabase 0

OV LIHC KIRP KIRCUCS GBM ACCLGG KICH UVM Stage.I Stage.II Stage.III Stage.IV SKCMLUSCLUADBLCAESCAHNSCSTADDLBCUCECCOAD CESCREAD BRCA SARCCHOLMESOPAAD PRAD TGCTTHYMLAML THCAPCPG ACC_Mets

e 0.4 f Transcribed 0.4 Transcribed Non-transcribed Non-transcribed 0.3 0.3

0.2 0.2

0.1 0.1 Percentage of mutations Percentage of mutations 0.0 0.0

C > T C > A C > G T > C T > A T > G C > T C > A C > G T > C T > A T > G

g 0.4 Metastasis Primary tumor 0.3

0.2

0.1 Proportion of tumors 0.0

TTN DST GK2 INF2 PPL VWF ANK2 FAT4 ANO6 ASPM CD93 GJD2 HCN4 NAV3 DNHD1 DNAH6 HELZ2HYDIN WDR66 ABCA4 CNTN1 CPNE4CSMD2CWH43 DNAH2EHMT1 ITGAX LRP1BMAP1S MUC16MUC5B MYO9B PKHD1 PTPRDRLTPR SYCP2TENM2TENM4 WDR52 CTNNB1 ENTHD1 PCDH12SHANK1 ZNF491ABCA13 C12orf55 COL3A1 DNAH11DNAH17 EOMES IQSEC3 MAP3K2 MYO1G SAMD9L CNTNAP4 CACNA1ACACNA1E CNTNAP5 D2HGDH KIAA0100KIAA1033 PDE4DIP ARHGEF28

h 0.4 Metastasis Primary tumor 0.3

0.2

0.1 Proportion of tumors

0.0 SI TTN TP53 DSTEYS GNAS LRP1 ANK2 APOB FAT4FBN2 OTOG ANK3 ATRX CDH4 FAT1FAT3FBN3 GPX1 GRM3 IGSF3ITPR3 MEN1NOS3 PDPRPEG3 SMC3 UBR4 USP6VCAN ZFAT MUC16 CDC27 HYDINPKHD1ASXL3HMCN1 GPR98OBSCN SVEP1 CMYA5 DNAH6DNHD1 FRAS1HIVEP1HUWE1 SRRM2STAB1SYNE2TENM2 ATF7IP CDH23 CNTN5 CRIM1DNAH5DOCK2EPPK1 KCNH1KCNH7 LAMA1LAMA2MACF1 NPY4RPAPPA PTPRDRHPN2ROBO2 SPTA1 ZFPM2 CTNNB1 PCDH15 RBMXL3AHNAK2 ADAM21 ATP2B3 COL5A1 FRMPD4GOLGA4GRID2IPHS6ST1 PAPPA2PCNXL2 PLXNB3 UNC13C VWA5B2 CNTNAP5 ZNF804B ANKRD36C ADAMTS16ADAMTS17 KRTAP10-4 SPATA31E1 Fig. 1 Metastatic ACC has a higher mutation burden compared to primary ACC. Total number of mutations per megabase in each tumor sample of both metastatic and primary ACC (TCGA). Each bar on x-axis represents a patient or a tumor sample (a). Total mutations per megabase in metastatic lung ACC and its counterpart primary ACC tumor from the same patient (b). Metastatic ACC mutation rate compared to other cancer types including primary ACC (TCGA) (c). Mutation rate in the primary ACC (TCGA) based on tumor stage (d). Relative proportions of the six possible base-pair mutations in 33 metastatic ACC (e) and primary ACC (f). The top genes that are frequently mutated in metastatic (g) and primary ACC (h)

NATURE COMMUNICATIONS | (2018) 9:4172 | DOI: 10.1038/s41467-018-06366-z | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-06366-z

0.955). Interestingly, a single sample from the TCGA ACC refs. 4,7 for primary tumor CNV patterns) and is essentially primary tumors (TCGA-OR-A5K4-01) also possessed LoF identical to that of Assie et al.4. In general, large-scale LoH is mutations in MSH6 and ATM (although both were nonsense typical of primary ACC4,16 and was readily apparent in our mutations, Supplementary Fig. 5), and had the eighth-highest metastatic samples, with an average of 49.6% LoH for our samples mutation rate of that sample of 92 primary tumors (24), (Fig. 4, Supplementary Table 1). However, in contrast to primary suggesting compound LoF in these two genes may be driving tumors16, we detected no whole-genome doubling (WGD) events hypermutation. The final patient with hypermutating ACC in the ABSOLUTE analysis of our ACC metastases, in spite of the metastases had no detected LoF mutations in any DNA repair fact that the proportion of CNV-altered genomes ranged from genes that were present in all three of the metastases sampled for 0.17 to 0.96 (Supplementary Table 1). Zheng et al.7 pointed out that patient. However, there were two rare germline ATM that these “noisy” CNV hypervariable samples consistently had missense mutations that rose to fixation in all three of the tumor WGD events, but we failed to detect this signal in our metastatic samples from that patient through complete LoH (Supplementary ACC samples. Furthermore, clonal diversity was extremely low, Table 2 and Supplementary Fig. 6). with the subclonal genome fraction ≤ 0.02 across all samples. At the level, the TERT amplification, CHEK2/ZNRF3 deletion, and CDKN2A deletion previously identified as being common in Metastatic ACCs are highly heterogeneous. To understand the fi nature of metastatic ACC, we selected all the somatic mutations primary ACC were also identi ed in our metastatic sample. within the coding regions of each tumor sample site and per- However, none of our samples were positive for deletions in RB1, 3q13.31, or 4q34.3, all of which were identified as recurrent in formed principal component analysis (PCA) on the entire patient 4,16 cohort. We found that metastatic ACC tumor samples clustered primary ACC . We next examined variation among metastatic sites in terms of by patients and not by location of the tumor metastases (Fig. 2a), fi fi as is expected given their origin from the same primary tumor. homozygous deletions and high-level ampli cations ( ve or more Next, we analyzed each metastatic tumor site to identify the genes copies). Only one homozygous deletion on 9 that are frequently mutated in each tumor site. Since one of the (CDKN2A) is common in metastatic lung and other tissue (three patient (Case 1) had a hypermutation phenotype in both the out of six patients) when compared across different patients metastatic sites (lung and other tissue site), we have performed (Supplementary Fig. 8A, B and Supplementary Fig. 9). On the other hand, we did not observe any common high-level the heatmaps and phylogenies to identify the overlapping and fi non-overlapping variants with and without this patient (Supple- ampli cations in both metastatic lung and other tumor tissues mentary Fig. 7 and Fig. 2b, c). Consistent with PCA, the majority across different patients (Supplementary Fig. 8C, D). However, of mutated genes were not common across metastatic tumor sites we found common homozygous deletions in CDK11A/B, STK4/ (Fig. 2b–e). Nevertheless, we observed that genes such as DNAH6, TOMM34, and CDKN2A/MTAP (Supplementary Figs. 10, 11), and in lung and other tissue of the same patient. Meanwhile, MUC5B, HELZ2, and KIAA0100 were frequently mutated in three fi of six lung ACC metastases (Supplementary Table 3). In addition, common ampli cations were observed in 15 genes in two CTNNB1 was mutated in both ACC liver metastases (Fig. 2d). different metastatic tumors of the same patient (Supplementary Genes such as ARHGEF28 and PPL were mutated in three of six Fig. 8E, F). In agreement with the patterns observed for SNPs, ACC samples classified as other tumor sites (Fig. 2c and Sup- copy number variation in metastatic tumors was also highly plementary Table 4). We did not see any common mutations in homogenous within patients but divergent between metastases in metastatic peritoneal tumor sites (Fig. 2e). the same site (or different sites) from different patients.

Metastatic ACC tumor sites within a patient are homogenous. Molecular pathways associated with metastatic ACC. We ana- As we did not find significant commonality among metastatic lyzed the entire gene list of recurrent mutations in metastatic tumor sites across patients, we decided to analyze the status of the ACC through ingenuity pathway analysis (IPA) to identify different metastatic ACCs within a given patient. We selected molecular pathways associated with these tumors. Summarized in four patients with metastases to more than one tumor site and Fig. 5a are the key networks of the altered pathways in these found that most mutated genes were common within a patient tumors. Some of the top altered pathways—with the total number regardless of the site of metastases (Fig. 3a–d). For example, we of overlapping genes in our cohort and the total number of genes observed that 57%, 44%, and 37% of the genes that were mutated associated in that pathway are also summarized in Table 1.In in Case 1, Case 2, and Case 3, respectively, were common particular, four signaling pathways, including ERBB4, retinoic regardless sites of tumor metastases (Fig. 3a–c). Particularly, acid receptor (RAR), G--coupled receptor (GPCR), and among the 16 mutations that are shared between the three tissues, platelet-derived growth factor receptor (PDGFR), were frequently genes such as CTNNB1, IGF2R, and SF1 that were known to play altered in metastatic ACC (Fig. 5b–d). an important role in the pathogenesis of adrenocortical tumor were present (Fig. 3a). Higher number of additional shared fi Novel drug candidates for metastatic ACC. We analyzed tar- mutations in genes ( ve) between kidney and liver are present getable recurrent mutations in ACC metastases using commer- when compared to only one mutation in gene between kidney cially available and FDA-approved drugs through the IPA and peritoneum (Fig. 3a). Although there are 20 shared mutations database. The majority of the drug targets (52%) were membrane in genes between lung and other tissue site, CTNNB1 is the only comprising transmembrane, GPCRs, transporters, and known common gene that is mutated between these tumor tissues ion channels. Twenty-two percent of the drug targets were (Fig. 3b). However, we noticed that only 14% of the genes cytoplasmic, including kinases, phosphatases, and enzymes, mutated in the lung metastases and other tumor sites were whereas 11% of them were nuclear (Fig. 6a, b). About 15% of the common in Case 4, possibly because this patient had a hyper- drug targets were extracellular, mainly cytokines and growth mutation phenotype (Fig. 3d). factors. We selected all the drugs that can be used to treat mul- tiple patients with metastasis in our cohort based on the muta- Copy number variation in metastatic ACC tumors. At the tional spectrum (Fig. 6c). We identified drugs such as Afatinib, genome-wide scale, copy number variation in metastatic tumors Cabozantinib, and Sunitinib that could target receptor tyrosine (Fig. 4) appears to be very similar to that of primary tumors (see kinases ERBB4, AXL, and FLT1/3, respectively. Drugs targeting

4 NATURE COMMUNICATIONS | (2018) 9:4172 | DOI: 10.1038/s41467-018-06366-z | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-06366-z ARTICLE

a 0.15

0.1 Case 1 Case 2 Case 3 0.05 Case 4 Case 5 Case 6 Case 7 0 Case 8 –0.05 0 0.05 0.1 0.15 0.2 Case 9 Case 10 –0.05 Case 11 Case 12 Case 13 –0.1 Case 14

–0.15

bcd CTNNB1 CSMD2 VANGL1 PPFIA4 TRIP12 SETD2 TCF21 TFB1M Case 2 IGF2R Case 13 ZPBP ZSCAN25 SLC13A1 CSGALNACT1 C9orf40 COL13A1 CYP2C18 SF1 C2CD3 PPM1H RNF6 TTC6 NPAP1 DAPK3 PGLS BPI ZMYND8 SIK1 Case 8 Case 4 DFFA EXTL2 DNAH6 CNTNAP5 LRP1B CHRNA1 TTN ZNF804A ACSL3 SEC22C MFSD10 ARAP2 GABRA4 GK2 ANK2 ZNF827 PCDH12 Case 2 SLC6A7 Case 9 SLC34A1 TRIM15 HLA−DRB1 PNPLA1 PKHD1 TBX18 DNAH11 FIGNL1 ZNF716 DNAJC2 ADAMDEC1 TRPA1 CCDC166 MFSD3 DRGX HPS6 OR4C5 Case 14 TYR IQSEC3 Case 3 C2CD5 CAPRIN2 CNTN1 AKAP11 EXOC3L4 SCG3 HCN4 CLEC16A NOD2 PDP2 GPR179 RAPGEFL1 POLI TNFRSF11A PLIN3 ECSIT Case 10 ZNF99 ZNF574 Case 7 LENG9 B4GALT5 C20orf195 C21orf59 USP11 MAGED1 RPL10 Case 6 Case 12

e

H6PD ABCA4 ACTG2 ARHGEF26 SLCO6A1 RAPGEF6 KIAA1919 RP1 PTPRD OR13J1 WDR66 EXOC6 DST DNHD1 TTN VWF PDZRN4 MUC5B SCAF11 CNOT1 ATP7B TENM4 TEP1 GJD2 RAB26 ACSM2B ITGAX CTNNB1 ILVBL ZNF681 MUC16 IFNGR2 EOMES SHANK1 DAPK1 PCMTD1 HELZ2 PAX2 TRIM49 MYOM3 CCDC122 LILRA4 TSHZ2 KEL HHIPL2 ANKRD36 GALNT9 ZNF717 KIF19 ARHGEF28 TMEM260 CNTNAP4 KRTAP10−8 ZNF491 ABCC5 ASPM PDE4DIP HAUS6 DNHD1 SLC25A6 FCGR3A FAT4 F5 PPL PPFIA4 SLC6A5 KLHL29 CWH43 CTNNB1 GAL3ST3 SETD2 NAV3 MYLK GRIA1 KIAA0100 IGF2R ABCA13 ZSCAN25 PDZRN4 SLC13A1 TEK CSGALNACT1 C9orf40 DNAH6 NOVA1 SF1 C2CD3 CTNNB1 PFDN5 CPNE4 RNF6 TTC6 RLTPR ENTPD5 NPAP1 NARF HCN4 Case 2 Case 4 DAPK3 Case 14 SMTN Case 10 Case 13 PGLS SAMD9L TRIM28 BPI MN1 SIK1 SOX11 EME1 KIAA1033 DMD HYDIN Case 12 Case 13 Case 3 Case 9 Case 8 Case 2 Case 7 Fig. 2 Genotyping of metastatic ACC shows tumor heterogeneity among different metastatic tumor sites. The principal component analysis (PCA) of all the missense, nonsense, and splice site mutations in 33 metastatic adrenocortical tumors (a). Heatmap and clustering of the overlapping variants within the genes in metastatic lung (b) and other tumor sites (c) from five different patients (Case 1 with hypermutation phenotype was excluded). The region containing maximum similarity was reanalyzed to better represent the shared genes that are mutated. Heatmap of the total number of overlapping and non-overlapping variants within the genes that are mutated in multiple patients with metastasis in liver (d), and peritoneum (e). Red bar in the heatmap indicates the gene is mutated, whereas the blue represents that the corresponding gene is wild-type in each sample

GPCRs and ion channels were also among the list that can be Discussion explored in ACC. Collectively, based on whole-exome sequencing In this study, we provided data on the types of mutations present mutations identified in metastatic ACC, 9 of 14 patients could in metastatic ACC and the higher mutation rate in metastatic have treatment using commercially available and approved drugs ACC relative to primary ACC. In addition, we demonstrate that that target alterations in their tumors. metastatic ACCs are highly heterogenous across patients but also

NATURE COMMUNICATIONS | (2018) 9:4172 | DOI: 10.1038/s41467-018-06366-z | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-06366-z

ab

PPFIA4 EVX2 KCNH8 CTNNB1 CTNNB1 SETD2 UROC1 IGF2R CPNE4 ZSCAN25 FHDC1 RASGRF2 CSGALNACT1 RPS6KA2 C9orf40 TAS2R41 SF1 SFTPD C2CD3 EML3 RNF6 CACNA2D4 TTC6 ZNF646 SPOP NPAP1 ZNF808 DAPK3 LILRA4 PGLS GRIK1 BPI B3GALT5 RME1 SIK1 LONRF3 SLC13A1 RECOL TRIP12 ATP5B ZPBP SEPT6 COL13A1 DDX43 CYP2C18 HEPH IGSF8 PPM1H DNAH6 HYDIN SNRNP200 CSMD2 NRROS VANGL1 FSTL4 TCF21 CKAP5 RLTPR TFB1M VSTM2B ZMYND8 PLXNA3 MYLK ARHGP25 ENTPD5 ANO3 TRIM28 DMD SOX11 NRP2 GIMAP1 KIAA1033 USP35 FCGR3A OR2T3 ETV3L C2orf50 MAP3K2 ALPPL2 DMGDH FAT4 C5orf60 JAKMIP2 DST MC5R KIAA1033 DACT3 NCOR2 ZNF419 CNOT1 FKBP1A ZNF491 CYP4F2 ENTHD1 LIPE Liver Lung Other tissue Kidney Peritoneum Shared mutations CHD5 CEP120 SVIL CHD9 cd KAZN ZNF608 WDFY4 RBL2 RAP1GAP GRAMD3 ERCC6 CCDC135 PIGV FBN2 SYNPO2L CES2 H6PD PTPRU ADAMTS19 PLCE1 PLEKHG4 MACF1 IL13 DCLRE1A PMFBP1 ABCA4 CCDC18 MATR3 PHRF1 MON1B ANKRD35 PCDHA6 MUC5B MAF NPR1 PCDHB16 NUP98 ZAF469 ACTG2 ADAMTS4 PCDHB14 DNHD1 ANKRD11 ASPM PCDHGB1 TTC17 PELP1 SLCO6A1 KIF14 ARAP3 NR1H3 GP1BA EIF2D FAT2 GIF MYH3 DEGS1 TENM2 ARHGEF17 DNAH9 RAPGEF6 PRSS38 RARS CACNA1C NCOR1 LYST SPDL1 VWF RAI1 KIAA1919 NLRP3 CDHR2 GRIN2B GPR179 APOB HIVEP1 LDHB IGFBP4 PTPRD ITSN2 NUP153 DENND5B CACNA1G ADCY3 ZNF311 SLC2A13 DCAF7 GAREML EHMT2 ADAMTS20 SCN4A EX0C6 PLB1 TNXB ANO6 ZNF750 ADD2 CUL9 ARID2 EPB41L3 VWF DNAH6 EFHC1 HDAC7 GRIN3B KCMF1 MDN1 KMT2D REXO1 FIF2AK3 ARMC2 FAM186A CACTIN PDZRN4 CNOT11 SLC2A12 SLC11A2 ZFR2 CLASP1 AHI1 SLC4A8 TICAM1 SCAF11 NCKAP5 SYNJ2 CSAD PRAM1 TTN MAP3K4 ESPL1 MUC16 ATP7B FSIP2 AGPAT4 TBK1 PALM3 CASP10 SDK1 LRRIQ1 NOTCH3 TRAK2 COL28A1 SETD18 MYO9B TEP1 PIKFYVE DNAH11 WDR66 FCHO1 UNC80 ZMIZ2 BRCA2 KIAA0355 RAB26 TTLL4 SRCRB4D SETDB2 LSR WNT6 GRM3 DCT KMT2B SPEG ABCB4 COL4A1 WDR62 ILVBL HDAC4 C7orf63 ZFHX2 MAP4K1 KIF1A MUC3A NYNRIN CIC ZNF681 FANCD2 MUC17 PRKD1 SHANK1 MLH1 RELN SPTB ZNF324B IFNGR2 VILL LAMB1 ZEYVE26 TMEM74B COL7A1 NRCAM MAP3K9 SEC23B OR5H14 ZNF425 OCA2 PAX1 ARHGEF26 GOLGB1 GALNT11 HERC2 ASXL1 COL6A6 RAB11FIP1 APBA2 GDF5 NOVA1 PLCH1 CHD7 TJP1 DLGAP4 CHRD TMEM67 PGBD4 PLCG1 SLC34A2 CSMD3 DISP2 JPH2 0R13J1 SLC4A4 EXT1 VPS39 P13 ANKRD17 COL22A1 ZNF106 SALL4 DNHD1 MMRN1 PARP10 STARD9 ZNF831 C4orf21 KANK1 MAP1A SYCP2 RP1 ANK2 SMARCA2 WDR76 LAMA5 KIAA1109 FOCAD SLC12A1 DIDO1 SCHS2 IGFBPL1 FBN1 HELZ2 ACSM2B ACSL1 TMC1 DMXL2 MMP11 FAT1 RORB CILP TFIP11 BRDT TPPP VPS13A DPP8 C22orf23 KIAA0947 WNK2 BNC1 EP300 NSUN2 FANCC ALPK3 SHANK3 KLHL8 TRIO KIAA0368 CHD2 FANCB PDZD2 TNC FAM173A MED14 PRDM6 TARS CNTRL TSC2 ATRX C5orf42 NR6A1 NTN3 DIAPH2 PIK3R1 GOLGA1 CREBBP IL1RAPL2 BDP1 LAMC3 HMOX2 MUM1L1 ST8SIA4 MLLT10 PRKCB THOC2 TRIM36 ANKRD26 SALL1 Peritoneum Other tissue Lung Other tissue

Fig. 3 Different metastatic ACC sites within the same patient are predominantly homogenous. Heatmap and clustering of the total number of overlapping and non-overlapping variants within the genes that are mutated in multiple tumor sites from each patient (a–d). The list of genes with shared mutations between lung and other tissue in the hypermutated patient (d) were shown in a box. Red bar in the heatmap indicates the gene is mutated, whereas the blue represents that the corresponding gene is wild-type in each sample share many similarities between different metastatic tumors cancers as compared to their corresponding primary tumors19–22. within the same patient. Furthermore, we identified novel path- For the same reason, one would expect the candidate driver genes ways that have not been reported in association with ACC, such of metastasis will be different compared to primary tumors. as ERBB4, , GPCR, and PDGFR signaling, Therefore, understanding candidate driver genes in primary that are genetically altered and are potentially targetable with tumors alone may not suffice when treating patients with meta- currently available drugs. static disease. Although it is still unclear, some studies suggest that ACC is a Multiple genes such as CSMD2 (CUB and Sushi Multiple multistep process in which several genome-wide alterations are Domains 2), LRP1b (LDL Receptor related Protein 1B) and accumulated over time17,18. Moreover, it is evident that meta- KIAA0100, which have been suggested to have tumor suppressor static tumors often undergo genomic evolution during progres- function in the literature were also found to be mutated in sion and drug resistance, thereby accumulating additional metastatic ACC tumors16,23–25. Particularly, CSMD2, a candidate mutations. Therefore, the higher mutation rate we observed in tumor suppressor gene in and breast cancer metastatic ACC is not surprising and has been observed in patients was mutated in three metastatic ACC samples (Fig. 1g). metastatic tumors from prostate, ovarian, colorectal, and breast LRP1b, a candidate tumor suppressor gene that is frequently

6 NATURE COMMUNICATIONS | (2018) 9:4172 | DOI: 10.1038/s41467-018-06366-z | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-06366-z ARTICLE

TERT

Gain High-level amplification 60% 20% 40% 10% 20%

0% 0%

20% 10% 40% 20% 60% Loss Deletion LOH

CDKN2A CHEK2/ZNRF3 12345678910 11 12 13 14 15 16 17 18 19 2021 22

Fig. 4 Copy number variation in metastatic tumors. High-level amplifications and deletions in metastatic ACC tumors. Regions of TERT, CDKN2A, and CHEK2/ZNF3 that are common in primary tumors and observed in metastatic tumors are highlighted. Red and blue represent chromosomal gain and loss, respectively inactivated in non-small cells was also mutated in report suggested WDR66 as an oncogene and a marker for risk three metastatic ACC samples (Fig. 1g). Finally, KIAA0100,a stratification in esophageal squamous cell carcinoma30.Itis candidate tumor suppressor gene in acute monocytic leukemia tempting to speculate that some of the frequently mutated genes was found to be mutated in three metastatic ACC samples in primary ACCs, including TP53, ANKRD36C, ADAM21, and (Fig. 1d). However, these mutations in metastatic ACC were CDH23, were not mutated in metastatic ACC, suggesting that missense mutations. We did not find any mutations in known or these tumors have lower metastatic potential. Nevertheless, our suspected oncogenes in the literature to be frequently mutated in results provide evidence that metastatic ACCs do evolve and gain metastatic ACC. Primary ACC are enriched for mutations in mutations that are different than primary ACCs. This makes it CTNNB1, TP53, ZNRF3, DAXX, and MEN1 genes, including challenging for treatments, as most of the targeted therapeutic homozygous deletions or high-level amplifications4,7. Although options are based on the mutational status of the primary ACCs we observed a higher frequency of mutations in the CTNNB1 (p. and may explain the lack of response to treatment observed in D32G, p.G34R, p.S45P, and p.S45F) in metastatic ACC, there ACC clinical trials. Although we do not provide any evidence that were no mutations observed in TP53, ZNRF3, MEN1, and DAXX. chemotherapy resistance driver genes are common in metastatic However, it should be noted that only 7 of 91, 4 of 91, and 2 of 91 tumors and moreover it is very difficult to make a connection primary ACCs in the TCGA study were found to have damaging between heterogeneity and primary resistance, it is tempting to mutations in TP53, MEN1, and DAXX, respectively. It appears speculate high heterogeneity in tumors could lead to rapid that metastatic ACCs do adapt and gain mutations in other genes acquired resistance because of the higher probability of pre- that are different than primary ACCs. The majority of the existing drug-resistant subclones, which may explain resistance to recurrent mutations in the metastatic ACC samples were mis- systemic chemotherapy that are common in ACC. sense mutations. Although, we found frameshift-, nonsense- or Multiple studies including TCGA have revealed key molecular splice site mutations in metastatic ACC tumors, the majority of pathways in different cancer types based on whole-exome them were not recurrent (present in multiple metastatic samples) sequencing, copy number and/or genome-wide gene expression and are, therefore, not discussed here. Many of the genes that are data31–36. Molecular characterization of primary ACCs suggests frequently mutated in metastatic ACCs alone are known to be that alterations in Wnt/beta-catenin and TP53/Rb signaling and associated with other types of cancers. For instance, somatic chromatin remodeling are primary events4,37. Although many mutations in CSMD2 was reported in non-small cell lung can- metastatic ACC tumor samples harbored mutations in the beta- cers26. Particularly, loss of CSMD3 has been shown to increase the catenin gene like primary ACC, other genes involving molecular proliferation of airway epithelial cells26. The extracellular matrix pathways such as ERBB4, GPCR, RAR, and PDGFR signaling gene, COL3A1, which was originally discovered to cause were also frequently mutated in metastatic ACC. Activating autosomal-dominant Ehlers-Danlos syndrome, was also recently mutations in ERBB4 have recently been reported in non-small cell found to be significantly altered in patients with melanoma27. lung cancer38. In , ERBB4 mutations have been shown Homozygous LoF mutations in RLTPR (CARMIL2) has been to be the major oncogenic driver with both aberrant ERBB4 and associated with disseminated EBV + smooth muscle tumors28.In PI3K-AKT signaling39. Likewise, we observed mutations in contrast, some of the genes, including ENTHD1, HELZ2, ERBB4 and PI3K-AKT signaling genes in metastatic ACC. PCDH12, CPNE4, and SHANK1, that are mutated in metastatic Conversely, RAR signaling is dysregulated in many cancers, but ACCs alone are completely novel and have not been functionally mutations in this receptor family are not a commonly observed characterized in any cancer type until now. However, some of phenomenon40–42. Activating mutations of G-protein-coupled these variants are present in the TCGA data set of other cancer receptor are mutated in approximately 20% of all cancers types. For example, the P1772S variant in the HELZ2 gene is including skin, prostate, breast, thyroid, liver, kidney, pancreas, present in cutaneous melanoma and colorectal adenocarcinoma, skin, , and large intestine43. Particularly, the glutamate the C1183S variant in PCDH12 is present in family of G-protein-linked receptors (GRM1-8) that were seen in and colorectal adenocarcinoma. Somatic mutations in ENTHD1 our cohort was mutated in 8% of non-small cell lung cancers43. has also been identified in multiple breast cancer cases29. A recent This family of proteins was also frequently altered in metastatic

NATURE COMMUNICATIONS | (2018) 9:4172 | DOI: 10.1038/s41467-018-06366-z | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-06366-z

a

b c SMAD 9% PI3K 9% ERBB4 RBP3 RARG NCOR1/2 12% γ PLC PKC 6% 6% 9% 6% 6%

NCOA1 SMARCA2 12% 6%

CREBBP EP300 6% 6%

d e PI3K PI3K 9% 9%

γ PKCβ PLC PKC 6% 6% GPCR 6% PDGFR (Gq, Gs, Gi) 6% 15% ADCY3/9 ABL1 9% 6%

STAT3 EIF2AK2 STAT3 9% 6% 9%

Fig. 5 Key molecular pathways altered in metastatic ACC. Network of canonical pathways that are significantly altered in metastatic ACC based on ingenuity pathway analysis (IPA) (a). Pathways that are predominantly altered in metastatic ACC including ERBB4 signaling (b), retinoic acid receptor signaling (c), G-protein-coupled receptor signaling (d), and platelet-derived growth factor receptor signaling (e). The number within each box represents the percentage of gene mutations in these tumors tumors arising from melanoma, prostate, large intestine, and lung critical in a complete assessment of the drug potential in addition cancers44. to the mutation status of metastatic ACC site(s)47. As mentioned before, targeted therapeutics for ACC are still Most importantly, our study raises the question whether pre- under investigation. We found the druggable target ERBB4 to be cision medicine is a better approach to treat patients with frequently mutated in metastatic ACC and we propose exploring metastatic ACC. With rapidly advancing technologies and the possibility of targeting this class of molecules. For instance, decreasing costs of genome profiling, precision medicine could lapatinib is already under clinical trial in patients with stage IV certainly be one of the most cost-effective and therapeutically melanoma with ERBB4 mutations. However, it has not been successful options in the control of aggressive ACC based on the tested in patients with advanced metastatic ACC. On the other tumor mutation status. hand, sunitinib has been tested in a phase II study in patients with ACC and is reported to have no response but prolonged tumor 45 Methods stabilization in 62% of the assessed patients . Another case Tissue samples. Metastatic adrenocortical tissue and blood samples were collected report of metastatic ACC with aberrant vascular endothelial on an institutional review board-approved clinical protocol (NCT01005654 and growth factor (VEGF) expression showed significant antitumor NCT01348698, National Cancer Institute at the National Institutes of Health). activity following sunitinib treatment46. It appears that sunitinib Written informed consent was obtained from the patients. We included 33 metastatic adrenocortical tissues collected from lung, liver, kidney, pancreas, interaction with mitotane drastically reduces its antitumor peritoneum, and other tissues (retroperitoneal perinephric/perisplenic/peripan- activity, implying that prior treatments of the patients are very creatic/para-aortic tissue or metastatic nodule on gallbladder) for 14 patients with

8 NATURE COMMUNICATIONS | (2018) 9:4172 | DOI: 10.1038/s41467-018-06366-z | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-06366-z ARTICLE metastatic disease. All diagnosis was confirmed by an endocrine pathologist based DNA extraction. Total genomic DNA was extracted from freshly frozen tumor on the Weiss criteria, and tumor samples were confirmed to contain ≥ 90% tumor tissues using a prep kit from Qiagen. The germline DNA was extracted from blood cells/nuclei. Tumor samples were obtained at surgical resection and immediately using a Qiagen Blood DNA Prep kit. All the procedures were done based on snap frozen and stored at −80 °C. manufacturer’s recommendations.

Whole-exome sequencing and variant calling. Tumor and germline blood DNA was used for whole-exome sequencing using the Agilent SureSelect v5 all exon + UTR (Agilent Technologies UK). Ten micrograms of genomic DNA were isolated Table 1 Key molecular pathways and the total number of from blood samples, and 125 base-pair paired-end reads were generated on the Illumina HiSeq 2000 (Illumina, Inc., San Diego, CA). All NGS data processing was overlapping genes in each pathway in metastatic ACC tumors done using our in-house developed pipeline [https://github.com/CCBR/Pipeliner)]. To mitigate the impact of pipeline-specific differences in somatic variant call rates, fi Pathway Number of Total number all of the TCGA primary ACC BAM les were downloaded from the Genomic Data overlapping genes of genes Commons (GDC) data portal and passed the exact same pipeline as our metastatic tumor samples. Short read data was trimmed for the presence of adaptors and low Hepatic stellate cell 24 183 quality using Trimmomatic v0.3648 and the following parameter settings: Lead- activation ing:10; Trailing:10; Sliding window:4:20; Minlen:20). Reads were then mapped to ERBB4 signaling 7 72 the hs37d5 (with decoys; ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/technical/refer- AMPK signaling 19 189 ence/phase2_reference_assembly_sequence/hs37d5.fa.gz) reference genome using RAR activation 18 190 BWA-mem v0.7.15 with default parameter settings (https://arxiv.org/abs/ fi GPCR signaling 22 270 1303.3997). The resulting BAM les were sorted using SAMtools v1.317 and PCR duplicates were marked using Picard v2.1.1 (https://broadinstitute.github.io/picard/ DNA double strand break 414 ). Realignment around INDELs and base recalibration was performed using the signaling Genome Analysis Toolkit v.3.4 (GATK, , Cambridge, MA), fol- Cancer Metastasis 19 247 lowing the GATK Best Practices49,50. For somatic SNP detection, we used signaling MuTect51 with paired tumor/normal and run in high confidence mode. For PDGF signaling 9 90 somatic INDEL detection, we used MuTect2 [https://software.broadinstitute.org/ PPAR signaling 9 93 gatk/documentation/tooldocs/current/ org_broadinstitute_gatk_tools_walkers_cancer_m2_MuTect2.php)]. For germline SNP and small INDEL calling, we used the HaplotypeCaller from the GATK package50. For copy number analysis, we used PSCBS segmentation52 implemented

a b Ligand dependent Cytokine Growth factors Extracellular nuclear receptors 1% 1% space 3% 15% regulators Phosphatases 3% 1%

Peptidases Enzymes 5% 21% Nucleus Transporters 11% 6%

Plasma Transmembrane membrance receptors 52% 7%

Ion channels 16% GPCR 9% Cytoplasm 22%

Other Kinases Drug targets by location 14% 13% Drug targets by type c Drug # of Target class Target candidates patients BMS-599626/Afatinib 4 Kinase ERBB4

Bumetanide 3 Transporter SLC12A5, SLC12A2

Cabozantinib 3 Kinase NTRK2, AXL, TEK

Dalfampridine 3 KCNA1, KCNA10, KCNA3, KCNB2, KCNC2, KCNC3, KCND2

Icosapent 3 GPCR and transporter FFAR1, SLC8A1

Amlodipine 2 Ion channel CACNB2, CACNA2D1

Carbidopa/levodopa 2 GPCR DRD1, DRD5

Fasorectam 2 GPCR GRM3. GRM6, GRM8

Nicorandil 2 Ion channel KCNJ2, KCNJ12

Paliperidone 2 GPCR ADRA1A, ADRA1D, ADRA2B

Pregabalin 2 Ion channel CACNA2D3, CACNA2D4

Sunitnib 2 Kinase FLT1, FLT3, RET

Fig. 6 Potential drug candidates for metastatic ACC. Recurrent gene mutations in metastatic ACC and the druggable targets based on location (a) and type (b). List of selected drugs and number of patients with mutations that can be targeted with the agent (c)

NATURE COMMUNICATIONS | (2018) 9:4172 | DOI: 10.1038/s41467-018-06366-z | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-06366-z in a CNVkit v0.8.553. ABSOLUTE v1.454 was used to estimate tumor purity and 11. Naing, A. et al. Insulin growth factor receptor (IGF-1R) antibody assess the subclonal genome fraction for each tumor. cixutumumab combined with the mTOR inhibitor temsirolimus in patients The transcribed and non-transcribed regions were defined based on widely used with metastatic adrenocortical carcinoma. Br. J. Cancer 108, 826–830 (2013). annotation tool (Ensembl v91) of and transcriptome. Since the 12. O’Sullivan, C. et al. The VEGF inhibitor axitinib has limited effectiveness as a libraries are stranded from the antisense strand, all the library effectively comes therapy for adrenocortical cancer. J. Clin. Endocrinol. Metab. 99, 1291–1297 from (-) strand by which we calculated the mutations of genes on each strand. (2014). 13. Lawrence, M. S. et al. Mutational heterogeneity in cancer and the search for – Droplet digital PCR. Each 20 μl PCR reaction consisted of 10 μl of 2x ddPCR new cancer-associated genes. Nature 499, 214 218 (2013). Supermix for probes (no dUTP) and 1 μl of 20x mutant target (FAM) and wild- 14. Greenman, C. et al. Patterns of somatic mutation in human cancer genomes. type (HEX) primers/probe and 60 ng of genomic DNA, to a final sample volume Nature 446, 153–158 (2007). adjusted with nuclease-free water. This was loaded in to a DG8™ Cartridge with 15. Williams, K. & Sobol, R. W. Mutation research/fundamental and molecular accompanying DG8™ Gasket and 70 μl of QX200™ Droplet Generation Oil for mechanisms of mutagenesis: special issue: DNA repair and genetic instability. Evagreen for droplet generation using a QX200™ Droplet Generator. Ninety-six- Mutat. Res. 743-744,1–3 (2013). well plates were then sealed using pierceable foil plate seals with a PX1™ PCR plate 16. Liu, C. X. et al. LRP-DIT, a putative endocytic receptor gene, is frequently sealer. A T100™ Thermal Cycler was used with the following cycling conditions: inactivated in non-small cell lung cancer cell lines. Cancer Res. 60, 1961–1967 enzyme activation for 5 min at 95 °C, followed by 40 cycles of denaturation at 94 °C (2000). for 30 s and annealing/extension at 55 °C for 1 min. Signal stabilization was 17. Gara, S. K. et al. Integrated genome-wide analysis of genomic changes and achieved by cooling to 4 °C for 5 min, and heating to 98 °C for 5 min. A ramp rate gene regulation in human adrenocortical tissue samples. Nucl. Acids Res. 43, of 2 °C per second was required for each step in the PCR. Data was then obtained 9327–9339 (2015). using a QX200™ Droplet Reader with ddPCR™ Droplet Reader Oil and QuantaSoft™ 18. Bernard, M. H. et al. A case report in favor of a multistep adrenocortical Software, version 1.7. tumorigenesis. J. Clin. Endocrinol. Metab. 88, 998–1001 (2003). 19. Beltran, H. et al. Whole-exome sequencing of metastatic cancer and – PCR and Sanger sequencing. Polymerase chain reaction (PCR) was performed biomarkers of treatment response. JAMA Oncol. 1, 466 474 (2015). with 100 ng of template genomic DNA and gene specific primers using High- 20. Zhang, J. et al. Exome profiling of primary, metastatic and recurrent ovarian Fidelity PCR master mix (Qiagen) according to the manufacturer’s recommen- carcinomas in a BRCA1-positive patient. BMC Cancer 13, 146 (2013). dation. The list of gene variants tested along with the primer sequences are pre- 21. Vignot, S. et al. Comparative analysis of primary tumour and matched sented in S5 Table. The PCR products were analyzed in 2% agarose gel and purified metastases in colorectal cancer patients: evaluation of concordance between using PCR purification kit (Qiagen). The purified products were subjected to genomic and transcriptional profiles. Eur. J. Cancer 51, 791–799 (2015). Sanger sequencing and the results were analyzed using Chromas software (Tech- 22. Lefebvre, C. et al. Mutational profile of metastatic breast cancers: A nelysium Pty. Ltd., South Brisbane, Australia). Retrospective Analysis. PLoS Med. 13, e1002201 (2016). 23. Zhang, R. & Song, C. Loss of CSMD1 or 2 may contribute to the poor – Pathway analysis and selection of drug candidates. The list of genes that carry prognosis of colorectal cancer patients. Tumor Biol. 35, 4419 4423 (2014). missense, nonsense and splice site mutations and also mutated in at least two 24. Shull, A. Y. et al. Somatic mutations, allele loss, and DNA methylation of the samples from the pairwise tumor and germline data of metastatic adrenocortical cub and sushi multiple domains 1 (CSMD1) gene reveals association with early tumors was uploaded into the IPA software (Qiagen). The “core analysis” function age of diagnosis in colorectal cancer patients. PLoS ONE 8, e58731 (2013). included in the software was used to interpret the mutation data, which included 25. Cui, H., Lan, X., Lu, S. M., Zhang, F. J. & Zhang, W. G. Bioinformatic the biological processes, canonical pathways, upstream transcriptional regulators, prediction and functional characterization of human KIAA0100 gene. J. – and gene networks. The drug candidates that are presented and can be targeted in Pharm. Ana 7,10 18 (2017). at least two patients were only selected. 26. Liu, P. et al. Identification of somatic mutations in non-small cell lung carcinomas using whole-exome sequencing. Carcinogenesis 33,1270–1276 (2012). 27. Zhao, B. & Pritchard, J. R. Inherited disease genetics improves the Data availability identification of cancer-associated genes. PLoS Genet. 12, e1006081 (2016). The datasets generated in this study is available in the dbGAP public repository with the 28. Schober, T. et al. A human immunodeficiency syndrome caused by mutations accession ID: phs001658.v1.p1. in CARMIL2. Nat. Commun. 8, 14209 (2017). 29. Zhang, Y. et al. Whole-exome sequencing identifies novel somatic mutations in chinese breast cancer patients. J Mol Genet Med 9, https://doi.org/10.4172/ Received: 9 November 2017 Accepted: 15 August 2018 1747-0862.1000183 (2015). 30. Wang, Q., Ma, C. & Kemmner, W. Wdr66 is a novel marker for risk stratification and involved in epithelial-mesenchymal transition of esophageal squamous cell carcinoma. BMC Cancer 13, 137 (2013). 31. Sanchez-Vega, F. et al. Oncogenic signaling pathways in the cancer genome atlas. Cell 173, 321 (2018). References 32. Kan, Z. Y. et al. Diverse somatic mutation patterns and pathway alterations in 1. Billimoria, C. P., Kraus, B. J., Narayan, R., Maddox, R. K. & Sen, K. Invariance human cancers. Nature 466, 869–U103 (2010). and sensitivity to intensity in neural discrimination of natural sounds. J. 33. Green, H. et al. Using Whole-Exome Sequencing to Identify Genetic Markers Neurosci. 28, 6304–6308 (2008). for Carboplatin and Gemcitabine-Induced Toxicities. Clin. Cancer Res. 22, 2. Else, T. et al. Adrenocortical carcinoma. Endocr. Rev. 35, 282–326 (2014). 366–373 (2016). 3. Giordano, T. J. et al. Distinct transcriptional profiles of adrenocortical tumors 34. Taylor, B. D. et al. Whole-exome sequencing to identify novel biological uncovered by DNA microarray analysis. Am. J. Pathol. 162, 521–531 (2003). pathways associated with infertility after pelvic inflammatory disease. Sex 4. Assie, G. et al. Integrated genomic characterization of adrenocortical Transm. Dis. 44,35–41 (2017). carcinoma. Nat. Genet. 46, 607–612 (2014). 35. Cox, S. N. et al. Multiple rare genetic variants co-segregating with familial IgA 5. Tissier, F. et al. Mutations of beta-catenin in adrenocortical tumors: activation nephropathy all act within a single immune-related network. J. Intern. Med. of the Wnt signaling pathway is a frequent event in both benign and 281, 189–205 (2017). malignant adrenocortical tumors. Cancer Res. 65, 7622–7627 (2005). 36. Esteban-Jurado, C. et al. Whole-exome sequencing identifies rare pathogenic 6. Pinto, E. M. et al. Genomic landscape of paediatric adrenocortical tumours. variants in new predisposition genes for familial colorectal cancer. Genet. Med. Nat. Commun. 6, 6302 (2015). 17, 131–142 (2015). 7. Zheng, S. et al. Comprehensive pan-genomic characterization of 37. Zheng, S. et al. Comprehensive pan-genomic characterization of adrenocortical carcinoma. Cancer Cell. 29, 723–736 (2016); erratum 30, 363 adrenocortical carcinoma. Cancer Cell. 29, 723–736 (2016). (2016). 38. Kurppa, K. J., Denessiouk, K., Johnson, M. S. & Elenius, K. Activating ERBB4 8. Schteingart, D. E. et al. Management of patients with adrenal cancer: mutations in non-small cell lung cancer. Oncogene 35, 1283–1291 (2016). recommendations of an international consensus conference. Endocr. Relat. 39. Lau, C., Killian, K. J., Samuels, Y. & Rudloff, U. ERBB4 mutation analysis: Cancer 12, 667–680 (2005). emerging molecular target for melanoma treatment. Methods Mol. Biol. 1102, 9. Fassnacht, M. et al. Limited prognostic value of the 2004 International Union 461–480 (2014). Against Cancer staging classification for adrenocortical carcinoma: proposal 40. Schenk, T., Stengel, S. & Zelent, A. Unlocking the potential of retinoic acid in for a Revised TNM Classification. Cancer 115, 243–250 (2009). anticancer therapy. Br. J. Cancer 111, 2039–2045 (2014). 10. Fassnacht, M. et al. Linsitinib (OSI-906) versus placebo for patients with 41. Minna, J. D. & Mangelsdorf, D. J. Retinoic acid receptor expression locally advanced or metastatic adrenocortical carcinoma: a double-blind, abnormalities in lung cancer: important clues or major obstacles? J. Natl. randomised, phase 3 study. Lancet Oncol. 16, 426–435 (2015). Cancer Inst. 89, 602–604 (1997).

10 NATURE COMMUNICATIONS | (2018) 9:4172 | DOI: 10.1038/s41467-018-06366-z | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-06366-z ARTICLE

42. Widschwendter, M. et al. Methylation and silencing of the retinoic acid Author contributions receptor-beta2 gene in breast cancer. J. Natl. Cancer Inst. 92, 826–832 (2000). Concept and design: S.K.G. and E.K. Development of methodology: S.K.G. and J.L. 43. Kan, Z. et al. Diverse somatic mutation patterns and pathway alterations in Generating and acquisition of data: S.K.G. Analysis and interpretation of data: S.K.G., J. human cancers. Nature 466, 869–873 (2010). L., and E.K. Manuscript writing and review: S.K.G., J.L., L.Z., E.H., M.C., and E.K. Study 44. O’Hayre, M. et al. The emerging mutational landscape of G proteins and G- supervision: E.K. protein-coupled receptors in cancer. Nat. Rev. Cancer 13, 412–424 (2013). 45. Chau, N. G. et al. A phase II study of sunitinib in recurrent and/or metastatic Additional information adenoid cystic carcinoma (ACC) of the salivary glands: current progress and challenges in evaluating molecularly targeted agents in ACC. Ann. Oncol. 23, Supplementary Information accompanies this paper at https://doi.org/10.1038/s41467- 1562–1570 (2012). 018-06366-z. 46. Lee, J. O. et al. Metastatic adrenocortical carcinoma treated with sunitinib: a case report. Jpn. J. Clin. Oncol. 39, 183–185 (2009). Competing interests: The authors declare no competing interests. 47. Kroiss, M. et al. Sunitinib in refractory adrenocortical carcinoma: a phase II, single-arm, open-label trial. J. Clin. Endocrinol. Metab. 97, 3495–3503 (2012). Reprints and permission information is available online at http://npg.nature.com/ 48. Bolger, A. M., Lohse, M. & Usadel, B. Trimmomatic: a flexible trimmer for reprintsandpermissions/ Illumina sequence data. Bioinformatics 30, 2114–2120 (2014). Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in 49. McKenna, A. et al. The Genome Analysis Toolkit: a MapReduce framework fi for analyzing next-generation DNA sequencing data. Genome Res. 20, published maps and institutional af liations. 1297–1303 (2010). 50. Van der Auwera, G. A. et al. From FastQ data to high confidence variant calls: the Genome Analysis Toolkit best practices pipeline. Curr. Protoc. Bioinforma. 43,11–33 (2013). 11 10. Open Access This article is licensed under a Creative Commons 51. Cibulskis, K. et al. Sensitive detection of somatic point mutations in impure Attribution 4.0 International License, which permits use, sharing, and heterogeneous cancer samples. Nat. Biotechnol. 31, 213–219 (2013). adaptation, distribution and reproduction in any medium or format, as long as you give 52. Olshen, A. B. et al. Parent-specific copy number in paired tumor-normal appropriate credit to the original author(s) and the source, provide a link to the Creative studies using circular binary segmentation. Bioinformatics 27, 2038–2046 Commons license, and indicate if changes were made. The images or other third party ’ (2011). material in this article are included in the article s Creative Commons license, unless 53. Talevich, E., Shain, A. H., Botton, T. & Bastian, B. C. CNVkit: Genome-wide indicated otherwise in a credit line to the material. If material is not included in the copy number detection and visualization from targeted DNA sequencing. article’s Creative Commons license and your intended use is not permitted by statutory PLoS Comput. Biol. 12, e1004873 (2016). regulation or exceeds the permitted use, you will need to obtain permission directly from 54. Carter, S. L. et al. Absolute quantification of somatic DNA alterations in the copyright holder. To view a copy of this license, visit http://creativecommons.org/ human cancer. Nat. Biotechnol. 30, 413–421 (2012). licenses/by/4.0/.

Acknowledgements © The Author(s) 2018 This work was supported by the intramural research program of the Center for Cancer Research, National Cancer Institute, National Institutes of Health.

NATURE COMMUNICATIONS | (2018) 9:4172 | DOI: 10.1038/s41467-018-06366-z | www.nature.com/naturecommunications 11