Molecular alterations associated with metastases of solid pseudopapillary neoplasms of the pancreas

Abstract Solid pseudopapillary neoplasms (SPN) of the pancreas are rare, low‐grade malignant neoplasms that metastasise to the liver or peritoneum in 10–15% of cases. They almost invariably present somatic activating mutations of CTNNB1. No comprehensive molecular characterisation of metastatic disease has been conducted to date. We performed whole‐exome sequencing and copy‐number variation (CNV) analysis of 10 primary SPN and comparative sequencing of five matched primary/metastatic tumour specimens by high‐coverage targeted sequencing of 409 genes. In addition to CTNNB1‐activating mutations, we found inactivating mutations of epigenetic regulators (KDM6A, TET1, BAP1) associated with metastatic disease. Most of these alterations were shared between primary and metastatic lesions, suggesting that they occurred before dissemination. Differently from mutations, the majority of CNVs were not shared among lesions from the same patients and affected genes involved in metabolic and pro‐proliferative pathways. Immunostaining of 27 SPNs showed that loss or reduction of KDM6A and BAP1 expression was significantly enriched in metastatic SPNs. Consistent with an increased transcriptional response to hypoxia in pancreatic adenocarcinomas bearing KDM6A inactivation, we showed that mutation or reduced KDM6A expression in SPNs is associated with increased expression of the HIF1α‐regulated protein GLUT1 at both primary and metastatic sites. Our results suggest that BAP1 and KDM6A function is a barrier to the development of metastasis in a subset of SPNs, which might open novel avenues for the treatment of this disease. © 2018 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of Pathological Society of Great Britain and Ireland.


Introduction
Solid pseudopapillary neoplasm (SPN) of the pancreas is a rare low-grade malignant neoplasm that usually occurs in young women [1]. Histologically, SPN is composed by polygonal cells that form a solid and pseudopapillary pattern, where they become discohesive. SPN has a peculiar immunophenotype among primary pancreatic neoplasms, which encompasses the expression of the mesenchymal marker vimentin, the 124 E Amato, A Mafficini, K Hirabayashi and RT Lawlor et al protease inhibitor α1-anti-trypsin, the neuroendocrine marker neuron-specific enolase [1], progesterone receptors, CD10, CD56, cyclin D1 and claudins 5 and 7 [2]. There may be some variable expression of synapthophysin and focal, faint positivity for cytokeratins [1].
The majority of SPNs are confined to the pancreas, with metastases to the liver and peritoneum in up to 15% of cases [1]. SPN has an indolent clinical behaviour even for cases of large tumour size, and long-term prognosis following surgical resection is generally excellent for both localised and distant disease [3].
Transcriptional activation of the WNT signalling pathway through the oncogenic mutation of CTNNB1 (β-catenin gene) is the major driver of SPN tumourigenesis [1]. Mutations of CTNNB1 clustering into exon 3 of the gene prevent degradation of the encoded protein β-catenin, which accumulates in the nucleus to form transcriptionally active complexes with the DNA-binding proteins TCF and LEF1 [4].
In keeping with this, abnormal cytoplasmic and nuclear β-catenin accumulation is evidenced by immunohistochemistry (IHC) in virtually all SPNs [5]. In the nucleus, β-catenin activates transcription of CCND1 that is overexpressed in about 70% of SPNs [6]. Whole-exome sequencing (WES) of eight primary SPNs confirmed that they have nearly universal mutation of CTNNB1 and demonstrated that these tumours have a low mutational burden and infrequent copy-number changes compared with pancreatic adenocarcinoma [7].
Given that no molecular characterisation of metastatic SPN has been conducted to date, the genetic determinants of progression and metastasis of this tumour type remain undetermined. In this study, we undertook a comprehensive molecular characterisation of 10 primary SPNs for mutations and copy-number variations (CNVs) of coding genes. In addition, we analysed five matched primary/metastatic lesions, performing high-coverage targeted sequencing (HCTS) of the relevant regions of 409 cancer-related genes to uncover genetic alterations involved in the progression of the disease.

Human research ethical approvals
Ethics committee approval was obtained at the three institutions involved in the collection of SPN cases. ARC-Net, University of Verona, Italy: approval number 1885 from the Integrated University Hospital Trust Ethics Committee (Comitato Etico Azienda Ospedaliera Universitaria Integrata). Tokai University, Japan: The Research Ethics Committee of Tokai University School of Medicine, Japan: approval number 17R275. University of Munich, Germany: approval number 503/16s from the Research Ethics Committee of Technical University of Munich. A cohort of 27 treatment-naive and surgically resected  SPNs was included in this study; details of molecular  analyses conducted on each case are reported in supplementary material, Supplementary materials and methods  and Table S1.

SPN cohort
All cases were classified according to World Health Organization 2010 criteria [1]. The sequencing cohort included 15 cases whose clinicopathological information is provided in Table 1, comprising 11 with fresh-frozen and four with formalin-fixed paraffin-embedded (FFPE) tissue available. Five of these SPNs were metastatic (one fresh-frozen and four FFPE) and their matched liver metastases were available.

Sequencing analysis
WES was performed on 10 primary tumour/normal DNA pairs using the SOLID 4 platform (Thermo Fisher   Genome-wide copy-number analysis Genome-wide SNP genotyping was performed on 10 primary tumours with the Illumina GoldenGate assay (Illumina). DNA copy-number changes were estimated using the ASCAT software release 2.1 [14].
Targeted sequencing and copy-number analysis

Computational and statistical analysis
Statistical analyses were conducted using Prism5 (GraphPad Software, La Jolla, CA, USA), and p < 0.05 was considered statistically significant. Fisher's exact test, chi-squared test and chi-squared test for trend were used as appropriate to assess the significance of BAP1, KDM6A and GLUT1 status and disease progression in the IHC analysis of SPN tissues. For data mining, we used the National Cancer Institute's Genomic Data Commons portal (https://portal.gdc.cancer.gov/) and the International Cancer Genome Consortium portal (http://icgc.org/). To determine the enrichment of genes involved in response to hypoxia in pancreatic ductal adenocarcinoma (PDAC) with inactivation of KDM6A, we used the R/Bioconductor package GSVA [15] and the MSigDB_HALLMARK_HYPOXIA dataset containing 200 genes. GSVA enrichment scores were generated using transformed count data and stratified based on KDM6A status. The Wilcoxon rank-sum test was applied to the stratified scores to compare normal and mutated samples.

Results
The study workflow is illustrated in supplementary material, Figure S1. For sequencing analysis, 15 SPN cases were collected. Of these 15 cases, five had metastases to the liver that were also available for sequencing analysis. Clinicopathological characteristics of the SPNs of the sequencing cohort are provided in Table 1. IHC analyses were conducted on the specimens from the sequencing cohort and on 12 additional SPNs, comprising 10 non-metastatic and two metastatic cases (see supplementary material, Table S1). Matched tumour/normal DNA of 10 non-metastatic SPNs underwent WES and high-density SNP array (see supplementary material, Tables S2  and S3). Comparative lesion sequencing [16,17] was performed by WES on one metastatic case with a single metastasis to the liver (see supplementary material, Tables S2 and S3). HCTS of the relevant region of 409 cancer-related genes was performed on the above case and four additional matched primary/metastatic specimens (see supplementary material, Table S4).

Clinicopathological characteristics of primary and metastatic SPNs
Most primary tumours showed the classical SPN microscopic features characterised by a heterogeneous combination of solid and cystic components. The solid components consisted of pseudopapillae with vascular stalks and hyalinised stroma within tumour areas and were intermingled by cystic haemorrhagic areas (see supplementary material, Figure S2A,B). Two of the metastatic tumours showed a diffuse solid growth pattern with minimal supporting fibrovascular stroma (see supplementary material, Figure S2C). This microscopic appearance was also maintained in the metastatic disease. Both of these tumours showed morphological features of aggressive behaviour, including mitoses, increased nuclear to cytoplasmic ratio, nuclear hyperchromasia and mild cellular atypia. In one of these two cases (SPN13), one of three hepatic lesions also presented with a distinct morphological component having sarcomatoid/dedifferentiated aspects (see supplementary material, Figure S3).

Somatic mutations and copy-number changes in non-metastatic SPNs
WES was performed on 10 primary SPNs and achieved a mean coverage of 44× in the tumour and 34× in matched normal samples. Detailed coverage information and the percentage of targeted bases are reported in supplementary material, Table S2. Activating mutations of CTNNB1 were present in all SPNs and were the only non-synonymous coding mutations identified in our cohort (see supplementary material, Table S3). All 10 CTNNB1 mutations were 1-bp missense mutations affecting exon 3 of the gene, distributed as follows: three mutations at codon 32, three at codon 37, two at codon 33 and two at codon 41 ( Figure 1A). The allele frequency of CTNNB1 mutations was in the range of 21-44% (see supplementary material, Table S3). Copy-number analysis revealed that all cases were diploid except for one tumour (SPN3) showing loss of heterozygosity (LOH) on chromosome 21 ( Figure 1B). Beyond mutations of CTNNB1, no recurrent genetic events were found in non-metastatic SPNs. . The genomic copy number is shown along the y-axis, while the chromosomal copy number is shown along the x-axis. In SPN3, ASCAT estimated that one copy of chromosome 21 (the 'green' copy) is completely deleted (copy number 0), leading to LOH (black arrow).

Somatic mutations in metastatic SPN
To address whether alterations other than CTNNB1 mutations are associated with the metastasis of SPN, we performed WES of an index case (SPN11), for which frozen tissue of a liver metastatic lesion was also available. WES achieved coverage of 60× in the tumour, 71× in the metastatic lesion and 21× in the normal sample (see supplementary material, Table S2). A total of four and two non-synonymous mutations were found in the tumour and the metastatic sample, respectively ( Figure 2A). Mutations in CTNNB1 (p.D32Y) and in the histones biotinylation gene BTD [18] (p.L297Q) were shared between tumour and metastasis, whereas mutations in the apyrase gene ENTPD4 (p.R340NfsTer14) and the translocase of outer mitochondrial membrane gene TOMM22 (p.V99I) were identified exclusively in the primary. Mutations in BTD and ENTPD4 were predicted to be deleterious for gene function, whereas mutation in TOMM22 (p.V99I) was predicted to be benign (see supplementary material, Table S3). To further confirm that progression of SPN is associated with accumulation of genetic alterations and to explore the clonal relationship between primaries and metastases, we performed HCTS of the relevant regions of 409 cancer-related genes in four additional primary/metastatic specimens for which only FFPE tissues were available. The index case was also included in this analysis, leading to a total of 13 specimens analysed (five primaries and eight metastatic lesions). HCTS yielded an average coverage of 100× for at least 70% of targeted bases in both neoplastic and non-neoplastic samples. In total, 27 mutations affecting eight genes, which included CTNNB1, KDM6A, BAP1, TET1, SMAD4, TP53, FLT1 and FGFR3 ( Figure 2B), were identified. Almost all mutations were missense, except for frameshift deletions affecting the genes encoding for the deubiquitinating enzyme BAP1 and the tyrosine kinase receptor FGFR3 (see supplementary material, Table S4). Mutations identified through WES in the index case could not be confirmed by HCTS, as the affected genes (BTD, ENTPD4, TOMM22) were not available in the targeted panel. Germline mutations of BTD cause an autosomal recessive disorder that can lead to delayed neurological development [19]. However, the role of this gene in cancer as well as the expression of its protein product in normal and neoplastic pancreatic tissues are neglected to date. Therefore, we first assessed the expression of BTD by IHC and found no immunoreactivity in both normal pancreas and in the 27 SPNs analysed. BTD staining was instead prominent in endothelial cells and normal hepatocytes (see supplementary material, Figure S4A). We concluded that the mutation of BTD was probably a passenger event. A low allele frequency (4%) missense mutation of TP53 was identified through HCTS of the metastatic specimen from the index case. Consistently, p53 immunoreactivity in this case was limited to a few nuclei (see supplementary material, Figure S4B), whereas none of the other metastatic or non-metastatic specimens showed any immunostaining.
In keeping with our observation, mutation of TP53 has been previously reported as a rare event in SPNs [11]. As expected for a founder mutation, CTNNB1 alterations were shared among primary and matched metastatic lesions. Additional events that were present in all samples from a given patient included KDM6A mutations in SPN13, TET1 mutations in SPN14 and FLT1 mutations in SPN56. Mutations present in one  Table S4 for details.
or more but not all specimens from a given patient, and as such defined as progressor mutations, affected BAP1 and SMAD4 in SPN13 and FGFR3 in SPN56. In particular, BAP1 mutation was detected in the primary tumour and only in one of the three liver lesions, whereas SMAD4 mutation was detected exclusively in one metastatic lesion but not in the primary tumour. These data suggest that the majority of mutations occur before metastatic spread as they are shared among the different lesions from a given patient ( Figure 2C). However, the identification of lesion-specific mutations indicates that subclones emerged either independently at primary or at metastatic sites as a consequence of ongoing clonal evolution.

Copy-number changes in metastatic SPN
CNV analysis at the chromosomal level was feasible only for the index case (SPN11) using WES data (supplementary material, Table S3). LOH of chromosome 22 detected through BAF analysis in the primary tumour was shared by the matched metastatic sample (see supplementary material, Table S3). Gene-level CNV analysis of the five metastatic SPN cases was conducted using HCTS data and revealed alterations in all the specimens analysed (see supplementary material, Figure S5). Figure 3A shows the detail of CNV differences detected among the four lesions analysed for the case SPN13. Differently from mutations, many CNVs were not shared among different specimens from a given case, thereby suggesting that they were independently acquired in each lesion as contributors to the progression of the disease ( Figure 3B). Pathway enrichment analysis of genes with losses and gains among the SPN cases (SPN11, SPN13, SPN14 and SPN56) showed that altered genes are involved in metabolic and pro-proliferative pathways, such as central carbon metabolism, Rap1, Ras, Pi3K/Akt, mTOR and focal adhesion, all of which are known to be involved in tumour progression [20][21][22][23][24][25] (see supplementary material, Tables S6A,B).

Inactivation of epigenetic regulators in metastatic SPNs
Two of the five metastatic cases in our cohort showed mutations that were predicted to inactivate genes involved in epigenetic regulation, namely BAP1, KDM6A and TET1. Mutations of BAP1 and KDM6A were concurrent in the primary and one metastatic lesion of case SPN13, and co-occurring mutations of these genes have already been reported in bladder cancer [26].  Figure S5 and Table S5 for details.
KDM6A (also known as UTX) is a core component of the 'complex of proteins associated with SET1' (COM-PASS) [10,27] and it has been shown to require BAP1 for its recruitment at chromatin enhancers [28]. Inactivation of those genes has been reported to occur through different mechanisms, ranging from DNA methylation to complex DNA rearrangements [25,29], which can be missed by targeted DNA sequencing. IHC is a reliable method to assess the status of BAP1 [30], while IHC for KDM6A has been recently used to assess the status of KDM6A in PDAC [10]. Therefore, we expanded the analysis of BAP1 and KDM6A status to the entire cohort of 27 SPNs (20 primaries and seven metastatic cases) by performing IHC on FFPE tissues. BAP1 was expressed in both islets and exocrine pancreatic cells (see supplementary material, Figure S4C). The primary tumour and the matched liver metastasis of case SPN13 sharing a deleterious mutation of BAP1 featured heterogeneous and homogenously negative staining for the gene product, respectively. Conversely, the other two hepatic  metastases from SPN13 presented a homogeneously positive staining consistent with the absence of BAP1 mutation ( Figure 4A). BAP1 was always expressed in the 20 non-metastatic SPNs, whereas staining was heterogeneous in three metastatic SPNs (one primary tissue and two metastases) and uniformly negative in one metastatic lesion ( Figure 4B, Table 2). An inverse correlation was observed between metastatic SPNs and BAP1 staining ( Figure 4B). In agreement with a previous study [10], KDM6A was expressed in islets and presented a mosaic expression in exocrine pancreas (see supplementary material, Figure S4C). Specimens from the case SPN13 bearing a missense mutation of KDM6A showed moderate staining, suggesting that the genetic alteration, which occurred proximal to the enzymatic domain, should not affect protein expression, but rather its function (Table 2). Strong expression of KDM6A was only detected in non-metastatic SPNs ( Figure 4C, Table 2), whereas staining was absent in one hepatic lesion from a metastatic SPN. Overall, there was a statistically significant inverse correlation between metastatic SPNs and KDM6A staining ( Figure 4C).

Reduced expression of KDM6A is associated with upregulation of GLUT1 in SPN
Direct and indirect evidence suggests that KDM6A is sensitive to changes in oxygen levels within the cells P, positive nuclear staining in 100% of tumour cells; N, negative for nuclear staining in 100% of tumours cells; H, heterogeneous nuclear staining with at least 25% of negative tumour cells; NE, not evaluable due to the absence of a positive internal control. 0, no tumour nuclear staining; 1, weak tumour nuclear staining; 2, moderate tumour nuclear staining; 3, strong tumour nuclear staining. [31]. Inactivation of KDM6A is enriched in aggressive tumours, including PDAC [32]. Interrogating the International Cancer Genome Consortium (ICGC) cohort, we found that a hypoxia transcriptional signature is enriched in PDAC characterised by KDM6A inactivation (see supplementary material, Figure S6A). Although mechanisms are not always shared among tumour types, we sought to investigate whether mutation or reduced expression of KDM6A was also associated with a hypoxic signature in SPNs. To identify a reliable marker of hypoxia linked to KDM6A status, we retrieved PDAC data from the ICGC and The Cancer Genome Atlas (TCGA) databases and found that in both cohorts the hypoxia-related gene SCL2A1 (encoding for GLUT1) was significantly upregulated when KDM6A was inactive (see supplementary material, Figure S6B,C). Based on this information, we immunostained 18 samples from our cohort for GLUT1. GLUT1 expression was strongly upregulated in KDM6A-mutated tumours (both primary and metastatic) compared with KDM6A-proficient tumours ( Figure 5A). Moreover, we found a trend to increased expression of GLUT1 in advanced SPNs (p = 0.1) and an inverse correlation between GLUT1 and KDM6A staining (p = 0.003, Figure 5B).

Discussion
Here we describe for the first time the genetic events associated with progression and metastasis of SPN. In keeping with previous reports [4,5,7], we showed that activating mutation of CTNNB1 is a universal feature of primary SPN and that no other recurrent alteration that affects protein-coding genes could be identified. Moreover, we found that clinically benign SPNs have diploid genomes and no relevant CNV, whereas one case with features of malignancy, including cellular atypia and high proliferative index associated with necrosis, showed LOH on chromosome 21. Overall, our data suggest that transcriptional activation of the WNT pathway is sufficient to drive tumourigenesis of SPN. Metastasis is the most common cause of cancer-related death showing an inverse correlation between expression of KDM6A and GLUT1 as assessed by IHC staining. For both GLUT1 and KDM6A, the category 'low' includes cases with absent or weak immunostaining, whereas 'high' denotes moderate or strong staining. Significance was determined by Fisher's exact test. [33]. Although metastasis can occur in up to 15% of SPN patients [1], only a few patients harbouring an undifferentiated carcinoma component have died of a metastasising SPN [34]. How cancer cells acquire the ability to colonise distant organs remains an unsolved question for SPN, mainly due to the paucity of metastatic specimens for molecular analyses. We used HCTS to analyse the clonal relationship between primary and hepatic lesions in five patients and found that the majority of mutations were shared among primaries and metastases, thereby suggesting that these events precede dissemination of cancer cells. Differently from mutations, the majority of CNVs were not shared among primaries and metastases. Although the reasons for this difference are difficult to infer from our data, we can speculate that differences in gene dosage reflect the selective pressure exerted by the different microenvironments or that subclones exist within primary tumours that could not be identified by our approach. Pathway enrichment analysis of genes affected by CNV in metastatic SPNs coherently identified biological processes known to sustain proliferation of cancer cells and metastatic behaviour. The majority of genetic alterations identified in metastatic SPNs affected bona fide tumour suppressor genes involved in epigenetic regulation of gene expression (KDM6A, TET1 and BAP1). Of those, BAP1 [35][36][37] and KDM6A [25,32] are genes commonly inactivated in aggressive and metastatic tumours. BAP1 and KDM6A have been shown to interact at chromatin enhancers [28] and disruption of this functional interaction in cancers drives malignancy through a mechanism that can be pharmacologically targeted [28]. Inactivation of those genes has been reported to occur through different genetic and epigenetic mechanisms [25,29], which can be missed by targeted DNA sequencing approaches and often lead to the loss of functional proteins. In our cohort of SPNs, we found that a lack of or reduced expression of both KDM6A and BAP1 is enriched in metastatic cases, suggesting that their function is a barrier to the development of metastatic disease at least in a subset of SPNs.
KDM6A inactivation is also observed in PDAC [25], where it associates with an aggressive and metastatic molecular subtype driven by 'squamous-like' transcriptional programmes, including response to hypoxia [38].
In keeping with this, our interrogation of ICGC and TCGA databases revealed that a hypoxia transcriptional signature is upregulated in PDAC tumours with inactivation of KDM6A. We explored this signature and identified GLUT1 as a biomarker that links hypoxia to KDM6A status in PDAC, and leveraged this information to verify if similar mechanisms were operating in SPNs. Although expression of GLUT1 did not significantly discriminate between metastatic and non-metastatic SPNs, we show here that tumours bearing inactivation of KDM6A or showing reduced protein expression strongly upregulate the expression of the hypoxia-marker GLUT1 in both primary and metastatic tissues. Although tumour hypoxia is strongly associated with cancer progression and resistance to therapy, we cannot conclude whether in SPN mutation or reduced expression of KDM6A is an epiphenomenon of oxygen shortage or instead precedes the formation of a hypoxic microenvironment. Despite our sequencing cohort representing the largest analysed to date, we could not identify recurrent genetic alterations in metastatic SPNs. This may be due to either the number of cases analysed or the targeted sequencing approach, which have possibly precluded us from identifying common drivers of SPN progression. Nevertheless, this study provides the most comprehensive description, to date, of the molecular events that characterise the progression of malignancy in SPN and indicates alterations of BAP1 and KDM6A as potential drivers of metastasis in SPNs. This might have important therapeutic implication as, in other tumour types, BAP1 loss and reduced expression of KDM6A have been shown to drive malignancy through epigenetic alterations that can be pharmacologically reverted [28]. Although we have provided evidence that mechanisms can be shared between different malignancies, it remains to be determined whether BAP1 and KDM6A alterations are promoting tumour growth in SPNs through a similar epigenetic mechanism. In the absence of preclinical models, this might only happen through multidimensional tumour profiling that will necessarily involve the integration of genetic and epigenetic analyses of a larger cohort of SPNs.  Table S1. Details of molecular analyses conducted on 27 SPNs Table S2. WES summary data for 11 SPN specimens Table S3. Somatic alterations identified in 11 SPN specimens by WES and SNP array analysis Table S4. Somatic alterations identified in matched tumour/metastatic specimens by targeted sequencing of 409 genes Table S5. Somatic gene copy-number alterations identified in matched tumour/metastatic specimens by targeted sequencing of 409 genes Table S6A. Enrichment analysis of main altered pathways according to copy-number alterations in four metastatic SPNs Table S6B. Detailed list of genes in main affected pathways according to copy-number alterations analysis in four metastatic SPNs