There is currently insufficient information about this tumor to either predict prognosis of patients diagnosed with PanNETs or to develop companion diagnostics and personalized treatments to improve disease management. Biallelic inactivation of the MEN1 gene, usually through a mutation in one allele coupled with loss of the remaining wild-type allele, occurs in 25-30% of PanNETs (5, 6). MEN1 is a tumor suppressor gene which, when mutated in the germline, predisposes to multiple endocrine neoplasia type 1 syndrome. Chromosomal gains and losses and expression analyses have revealed candidate loci for genes involved in the development of PanNETs, but these have not been substantiated by genetic or functional analyses (7-9).

To gain insights into the genetic basis of this tumor type, we determined the exomic sequence of ~18,000 protein-coding genes in a Discovery set of ten well-characterized sporadic PanNETs. A clinically homogeneous set of tumors of high neoplastic cellularity is essential for the successful identification of genes and pathways involved in any tumor type. Thus, we excluded small cell and large cell neuroendocrine carcinomas and studied only samples that were not part of a familial syndrome associated with PanNETs (table S1) (1). We microdisected tumor samples to achieve a neoplastic cellularity of >80%. DNA from the enriched neoplastic samples and from matched non-neoplastic tissue from ten patients was used to prepare fragment libraries suitable for massively parallel sequencing. The coding sequences were enriched by capture with the SureSelect Enrichment System and sequenced using an Illumina GAIIx platform (10). The average coverage of each base in the targeted regions was 101-fold and 94.8 % of the bases were represented by at least 10 reads (table S2).

We identified 157 somatic mutations in 149 genes among the ten tumors used in the Discovery set. The mutations per tumor ranged from 8 to 23, with a mean of 16 (table S3). Of these mutations, 91 % were validated by Sanger sequencing. There were some obvious differences between the genetic landscapes of PanNETs and those of pancreatic ductal adenocarcinomas (PDAC, ref. 11). First, there were 60% fewer genes mutated per tumor in PanNETs than in PDACs. Second, the genes most commonly affected by mutation in PDACs (KRAS, TGF-β pathway, CDKN2A, TP53) were rarely altered in PanNETs and vice versa (Table 1). Third, the spectrum of mutations in PDAC and PanNET were different, with C to T transitions more common in PDACs than in PanNETs, and C to G transversions more common in PanNETs than in PDACs (table S4). This suggests that mutations in PanNETs and PDAC arise through different mechanisms, perhaps due to exposure to different environmental carcinogens or through the action of different DNA repair pathways.

We next selected genes for further analysis that were well-documented components of a pathway that was genetically altered in more than one tumor, because alterations in these genes are most likely to be clinically relevant. Four genes were mutated in at least two tumors in the Discovery set: MEN1 in five, DAXX in three, PTEN in two, and TSC2 in two. ATRX was mutated in only one sample in the Discovery set, but its product forms a heterodimer with DAXX and therefore is part of the same pathway, so it was also evaluated in the Validation set. Similarly, PIK3CA was included because its product is part of the mTOR pathway that includes PTEN and TSC2 (12-14). The sequences of these genes were then determined by Sanger sequencing in a Validation set consisting of 58 additional PanNETs and their corresponding normal tissues (Fig. 1, A and B). In total, somatic mutations in MEN1, DAXX, ATRX, PTEN, TSC2, and PIK3CA were identified in 44.1%, 25%, 17.6%, 7.3%, 8.8%, and 1.4% PanNETs, respectively (Table 2).

Of the 30 mutations in MEN1, 25 were inactivating mutations (18 insertions or deletions (indels), 5 nonsense and 2 splice-site mutations), while five were missense. At least 11 were homozygous; in the others, the presence of “contaminating” DNA from normal cells made it difficult to reliably distinguish heterozygous from homozygous changes. MEN1 encodes menin, a nuclear protein that acts as a scaffold to regulate gene transcription by coordinating chromatin remodeling. It is an essential component of the MLL SET1-like histone methyltransferase (HMT) complex (15-19). Overall, MEN1 was mutated in 30 of the 68 PanNETs used in the Discovery and Validation sets combined.

DAXX and ATRX were mutated in 17 and 12 PanNETs, respectively. No tumor with a mutation in DAXX had a mutation in ATRX, consistent with their presumptive function within the same pathway. Overall 29 of 68 PanNETs (42.6%) had a mutation in this pathway. There were 11 insertions or deletions (indels) and 4 nonsense mutations in DAXX, and six indels and 3 nonsense mutations in ATRX. The three ATRX missense mutations were within the conserved helicase domain while the DAXX missense mutations were non-conserved changes. Five DAXX and four ATRX mutations were homozygous, indicating loss of the other allele. The high ratio of inactivating to missense mutations in both genes establishes them as PanNET tumor suppressor genes. Loss of immunolabelling for DAXX and ATRX correlated with mutation of the respective gene (fig. S1, A and B, and table S5). From these data, we hypothesize that both copies of DAXX are generally inactivated, one by mutation and the other either by loss of the non-mutated allele or by epigenetic silencing. We also hypothesize that both copies of ATRX are inactivated, one by mutation and the other by chromosome X inactivation. Recently, it has been shown that DAXX is an H3.3-specific histone chaperone (20). ATRX encodes for a protein that at the amino-terminus has an ADD (ATRX-DNMTT3-DNMT3L) domain and a carboxy-terminal helicase domain. Almost all missense disease causing mutations are within these two domains (21). DAXX and ATRX interact and both are required for H3.3 incorporation at the telomeres and ATRX is also required for suppression of telomeric repeat-containing RNA expression (22-24). ATRX was recently shown to target CpG islands and G-rich tandem repeats (25), which exist close to telomeric regions.

We identified five PTEN mutations, two indels and three missense; six TSC2 mutations, one indel, one nonsense and four missense; and one PIK3CA missense mutation. Previously published expression analyses have indicated that the expression of genes in the mTOR pathway is altered in most PanNETs (26,27). Our data suggest that, at least at the genetic level, only a subset of PanNETs have alterations of this pathway. This finding may have direct clinical application through prioritization of patients for therapy with mTOR pathway inhibitors. Everolimus (Afinitor, RAD-001, 40-O-(hydroxyethyl)-rapamycin) has been shown to increase progression free survival in a subset of PanNET patients with advanced disease (28). If the mutational status of genes coding for proteins in the mTOR pathway predicts clinical response to mTOR inhibitors, it should be possible to select patients who would benefit most from an mTOR inhibitor through analysis of these genes in patients’ tumors (29, 30).

All 68 tumors evaluated in this study were from patients undergoing aggressive intervention (table S6) and included patients undergoing curative resection as well as those with metastatic disease. Interestingly, mutations in MEN1, DAXX/ATRX or the combination of both MEN1 and DAXX/ATRX were associated with prolonged survival relative to those patients whose tumors lacked these mutations (Fig. 1, C and D and table S7). This was particularly evident in patients with metastatic disease and with mutations in both MEN1 and DAXX/ATRX: 100% of patients with PanNETs that had these mutations survived at least ten years while over 60% of the patients without these mutations died within five years of diagnosis (Fig. 1D). One possible explanation for the difference in survival is that mutations in MEN1 and DAXX/ATRX identify a biologically specific subgroup of PanNETs.

In summary, whole exome sequencing of pancreatic neuroendocrine tumors has led to the identification of novel tumor suppressor genes and illuminated the genetic differences between the two major neoplasms of the pancreas. The mutations may serve to aid prognosis and provide a way to prioritize patients for therapy with mTOR inhibitors.

Sup Figure S1

table S8

Supplement

table S1

table S2

table S3

table S4

table S5

table S6

table S7