Solid pseudopapillary neoplasms of the pancreas are dependent on the Wnt pathway

Solid pseudopapillary neoplasms (SPNs) are rare and relatively indolent tumors of the pancreas. While primary SPNs can be surgically resected, there are currently no therapies available for patients with advanced stage disease. Given that these tumors frequently carry CTNNB1 hotspot (recurrently mutated loci in a gene) mutations resulting in β‐catenin nuclear accumulation, it has been speculated that the Wnt pathway may be a driver in this disease. Here, we present a comprehensive “multi‐omics” study where the genome, transcriptome, and methylome of SPNs were analyzed. We found that SPNs are characterized by a low‐complexity genome where somatic mutations in CTNNB1, present in 100% of the cases, are the only actionable genomic lesions. Compared to more common subtypes of pancreatic tumors (adenocarcinomas and pancreatic neuroendocrine tumors), SPNs show high expression levels of genes belonging to the Wnt pathway. Their methylome was consistent with an epithelial cell origin and a general upregulation of Wnt pathway genes. Clinical studies to evaluate the exquisite sensitivity of SPNs to inhibitors of the Wnt pathway are warranted.


Introduction
Solid pseudopapillary neoplasms (SPNs) are rare pancreatic tumors of unknown etiology, accounting for 1% of all pancreatic exocrine neoplasms. SPNs typically present as large, solitary, well-circumscribed lesions with indolent clinical course and a low propensity to metastasize (Klimstra et al., 2000;Ren et al., 2014). Although they are commonly managed surgically, some SPNs can exhibit a more aggressive behavior and metastasize. Currently, there is no consensus on effective systemic treatments for malignant SPNs, and given the rarity of advanced disease, prospective clinical trials testing novel therapies have not been performed as yet.
Solid pseudopapillary neoplasms are characterized by the presence of somatic CTNNB1 exon 3 hotspot mutations (Abraham et al., 2002;Guo et al., 2017;Tanaka et al., 2001;Wu et al., 2011), leading to stabilization and nuclear localization of b-catenin, which can be detected by immunohistochemistry (Tanaka et al., 2001). Given these premises, the role of the Wnt pathway in SPN development has been suggested. The Wnt pathway is a signal transduction cascade that under normal physiological conditions regulates development and stemness but has also been tightly associated with multiple growth-related pathologies and cancer (Nusse and Clevers, 2017). A key component of the Wnt pathway is the protein b-catenin encoded by the gene CTNNB1, which regulates transcription of downstream target genes (e.g., LEF1, AXIN2). Hotspot mutations in CTNNB1 can cause abnormal accumulation of nuclear b-catenin through stabilization of the protein and eventually constitutive activation of the pathway.
Here, we assessed whether SPNs are dependent on the Wnt signaling pathway and performed a comprehensive "multi-omics" profiling of SPNs collected at our institution during the last two decades to define their repertoire of somatic mutations and copy number alterations, the transcriptional activation of the Wnt pathway and their methylation profiles.

Cases
Patients with written informed consent for research that had a diagnosis of SPN treated at Memorial Sloan Kettering Cancer Center (MSKCC) were identified from our institutional database. Clinicopathological characteristics of the SPN patients are shown in Table 1. Electronic medical records were retrospectively reviewed for data on patient demographics, pathology, presentation (benign versus malignant SPN), treatment (surgical and systemic therapies), and outcomes. In addition, pancreatic adenocarcinomas (ADCs) and pancreatic neuroendocrine tumors (PNETs) were identified as controls. All pathology was reviewed by gastrointestinal pathologists at MSKCC experienced in the diagnosis of SPN. Approval for data collection and analysis was obtained from the MSKCC Institutional Review Board. The study conforms to the guidelines set by the Declaration of Helsinki.

Immunohistochemistry
Immunohistochemical analysis for b-catenin (Clone 14, 1 : 200; BD Transduction, San Jose, CA) was performed on 4-lm-thick sections from representative formalin-fixed, paraffin-embedded SPN tissue blocks, as previously described (Basturk et al., 2016). Positive and negative controls were included in each slide run. The expression of b-catenin was assessed in the membrane and cytoplasmic and nuclear subcellular compartments, and considered abnormal if cytoplasmic and nuclear accumulation were present.

Whole-exome sequencing analysis
DNA samples extracted from 14 flash-frozen SPNs and matched normal tissues were subjected to wholeexome sequencing at the MSKCC Integrated Genomics Operation as previously described (Ng et al., 2017). Briefly, raw sequence reads were aligned to the reference human genome GRCh37 using the BURROWS-WHEELER ALIGNER (BWA 0.7.15) (Li and Durbin, 2009). Local realignment, duplicate read removal, and base quality score recalibration were performed using the GENOME ANALYSIS TOOLKIT (GATK 3.7) (McKenna et al., 2010). Somatic single nucleotide variants (SNVs) were called using MUTECT (1.1.7) (Cibulskis et al., 2013), and small insertions and deletions (indels) were identified using STRELKA (1.0.15) (Saunders et al., 2012), VARSCAN2 (2.3.7) (Koboldt et al., 2012), LANCET (1.0.0) (Narzisi et al., 2018), and SCALPEL (0.5.3) (Narzisi et al., 2014) and further curated by manual inspection. SNVs and indels outside of target regions were filtered out, as were SNVs and indels for which the variant allele fraction (VAF) in the tumor sample was < 5 times that of the paired normal VAF as previously described (Ng et al., 2017;Weigelt et al., 2018). Finally, SNVs and indels found at > 5% global minor allele frequency in dbSNP (build 137) and > 5% global allele frequency in EXAC (0.3.1) were discarded. Somatic copy number alterations and loss of heterozygosity were obtained using FACETS (Shen and Seshan, 2016). The cancer cell fractions (CCF) of all mutations were computed using ABSOLUTE (1.0.6) (Carter et al., 2012). A mutation was classified as clonal if its probability of being clonal was > 50% ( Landau et al., 2013) or if the lower bound of the 95% confidence interval of its CCF was > 90% (Ng et al., 2017;Weigelt et al., 2018). Mutations that did not meet the above criteria were considered subclonal. A combination of in silico functional predictors was used to define the potential functional impact of each missense SNV as previously described (Martelotto et al., 2014;Ng et al., 2017). Mutation hotspots were assigned according to Chang et al. (2016). CTNNB1 mutation frequencies were assessed in 10 100 common cancers from The Cancer Genome Atlas (TCGA) (Bailey et al., 2018;Gao et al., 2013), including 175 pancreatic ADCs. In addition, mutational data from 8 SPNs from Wu et al. (2011) and 344 non-SPN pancreatic tumors, including 23 acinar cell carcinomas from Jiao et al. (2014), 98 PNETs from Scarpa et al. (2017), and 24 cystic tumors of the pancreas Wu et al. (2011). The mutational data were retrieved from cBioPortal (Gao et al., 2013). CTNNB1 mutation and hotspot mutation frequencies were assessed following exclusion of hypermutated cases, defined as cancers harboring more than 1000 nonsynonymous mutations, microsatellite-unstable, or harboring POLE or POLD1 exonuclease domain mutations, as previously described . Mutation diagrams ('lollipop' plots) were generated using Muta-tionMapper on CBIOPORTAL (Gao et al., 2013) and manually curated.

Gene expression analysis
Total RNA extracted from nine SPNs, seven pancreatic ADCs, and 8 PNETs was subjected to expression analysis of 240 selected genes using the nCounter platform from Nanostring Technologies, Seattle, WA (Table S3). The raw counts were normalized using the nSolver analysis software. We performed background subtraction using the negative control and normalized the background corrected counts using the housekeeping genes ACTB, MRPL19, PSMC4, RPLP0, and SF3A1. Comparison between SPN samples and ADCs or PNETs was done using the R/BIOCONDUCTOR LIMMA package, and genes with P values < 0.05 were considered as significantly differentially expressed.

Genome-wide methylation analysis
DNA extracted from 13 SPNs was subjected to Illumina Infinium Methylation 850K array profiling as previously described (Chiang et al., 2016). The R/BIO-CONDUCTOR MINFI package was used to process the raw files and obtain the normalized beta values. The methylation data in the form of beta values of cells of epithelial (n = 22), fibroblast (n = 10), and muscle cell (n = 6) origin were downloaded from encyclopedia of DNA elements (ENCODE) project (ENCODE Project Consortium et al., 2007) and processed as described above. To avoid the confounding effects of gender, probes mapping to sex chromosomes were discarded while batch effect was removed using the COMBAT method (Johnson et al., 2007) implemented in the R/ BIOCONDUCTOR package SVA. Probes which were differentially methylated between ENCODE samples of epithelial, fibroblast, and muscle cell origin were identified using the Kruskal-Wallis test and an FDRadjusted P-value < 0.1 (Table S4). The SPN methylation data were projected onto the first two principal components estimated using the ENCODE samples. Probes of the Wnt pathway found to be differentially methylated between SPNs and ENCODE samples were subjected to hierarchical cluster analysis using Pearson correlation and Ward's distance.

Statistical analysis
Using the transcriptomic data, genes differentially expressed between SPNs and ADCs or PNETs were identified using a linear model fitted to the expression data. We considered genes with a P-value < 0.05 and an absolute fold change value > 2 as statistically significant. In the case of methylation data, the data were filtered by identifying probes with multimodal distributions of beta values in the ENCODE dataset. A mixture of Gaussian distributions was fitted to the pooled data using the R/BIOCONDUCTOR package MCLUST. Probes which were differentially methylated between ENCODE samples of epithelial, fibroblast, and muscle cell origin were identified using the Kruskal-Wallis test and an FDR-adjusted P-value < 0.1. The SPN methylation data were projected onto the first two principal components estimated using the ENCODE samples. All hypothesis tests were unpaired, and Pvalues were two-sided. Unless otherwise stated, all computations were performed in R/BIOCONDUCTOR.

Activation of the Wnt pathway in SPNs
From a genomics standpoint, SPNs appear to be driven solely by CTNNB1 activating somatic mutations, which result in nuclear accumulation of the protein (Polakis, 1999). Accordingly, we detected strong b-catenin protein expression in both cytosol and nuclei of all SPNs analyzed (Fig. 1D). This, in turn, initiates a transcriptional program primed to activate the Wnt pathway (Mosimann et al., 2009). To investigate the components of the Wnt signaling pathway activated in SPNs, we performed a differential gene expression analysis of nine SPNs, four of which were metastatic, seven pancreatic ADCs, and eight PNETs using the Nanostring platform (Table S3). Compared to ADCs and PNETs, SPNs displayed significantly higher expression levels of genes encoding components of the Wnt pathway, including DKK4, which has recently been detected by mass spectrometry in SPNs (Park et al., 2015), and direct TCF/b-catenin target genes such as LEF1 (Aoki et al., 1999) or AXIN2 (Jho et al., 2002) ( Fig. 2A-D, Fig. S2). Other genes whose expression has been described to be induced by Wnt/b-catenin signaling activation such as LGR5, ASCL2, or CCND1 were also expressed at significantly higher levels in SPNs (Fig. S2).

Methylome analysis of SPNs
Given the unknown histogenesis of SPNs, we analyzed the global methylation profiles of 12 of the 14 SPNs analyzed by whole-exome sequencing and compared their methylation patterns with those of normal cells of various anatomical sites from the ENCODE database (ENCODE Project Consortium et al., 2007). The majority of the SPNs profiled in this study clustered with epithelial cells (Fig. 3A). We further performed a differential analysis between ENCODE samples of epithelial origin and SPNs, which identified 2511 differentially methylated probes (Table S3). Probes that were significantly differentially methylated between samples of epithelial, fibroblast, and muscle cell origin in the ENCODE database were mapped to genes in the Wnt pathway and showed differential methylation in the gene bodies (intronic and exonic areas of a gene), which, based on ontology, would have resulted in upregulation of the respective transcript (e.g. WNT5A). Conversely, hypermethylated probe sets corresponded with genes whose transcripts were downregulated in SPNs (e.g., CD44; Fig. 3B). Two exceptions were probes mapping to ZIC2 and WNT5B, which were found to be hypermethylated in SPN compared to ENCODE samples. Unlike CD44, these genes had a significantly higher expression in SPNs compared to ADCs or PNETs. Also, the methylation pattern is not unexpected, given that the hypermethylated probes mapped to the 3 0 UTR of the corresponding gene, a phenomenon not uncommonly associated with overexpression of the target gene.

Discussion
Here, we report a multi-omics analysis of SPNs compared to the other two more common subtypes of pancreatic cancer, ADC and PNET. In agreement with the existing literature, we found that SPNs have low genomic complexity and are characterized by the invariable presence of CTNNB1 exon 3 mutations. Accordingly, we found consistent increased expression of genes encoding components of the Wnt pathway in these tumors. Nuclear b-catenin accumulation promotes the expression of a panel of Wnt target genes codifying for proteins involved in a myriad of cell processes. Among them, some codify components of the Wnt/b-catenin pathway itself such as LEF1, AXIN2, or RNF43, which induce a positive feedback loop that enhances signaling activation (Zhan et al., 2017). Here, we observed that SPNs with mutations in CTNBB1 that provoke constitutive accumulation of nuclear b-catenin express higher levels of some of these Wnt target genes that are also Wnt pathway components such as LEF1, AXIN2, and RNF43 (Fig. 3). We also observed that other genes codifying components of the Wnt pathway are highly expressed in SPN than ADC, such as ligands (WNT5A, WNT2B, or RSPO4), receptor components (LRG6 or FZD10), or other key proteins for signal transduction such as APC. Although these are not formally demonstrated to be Wnt/b-catenin pathway target genes, their higher expression could suggest and enhanced activation of the pathway additional to that promoted by CTNNB1/b-catenin activating mutations.
The lack of somatic genetic alterations other than CTNNB1 mutations and low levels of genomic complexity may provide an explanation for the indolent behavior of SPNs, akin to colorectal polyps associated with familial adenomatous polyposis in patients with germline APC mutations developing multiple KRAS wild-type, benign tumors where a dysfunctional APC can be the only driver (Leoz et al., 2015;Takane et al., 2016). In mice, activation of b-catenin is indeed sufficient to produce large pancreatic tumors that resemble human SPNs both morphologically and by immunohistochemical analysis (Heiser et al., 2008).
Of note, a recent report identified also inactivating mutations of the epigenetic regulators KDM6A, TET1, and BAP1 to be associated with metastatic spread (Amato et al., 2019). Whether these alterations are also playing a role in regulating the expression of Wntdependent genes remains to be elucidated.
While inhibition of the Wnt pathway may be an intuitive therapeutic option for this disease, its rarity and the lack of available preclinical models are clear limitations in the advancement toward an effective treatment for metastatic SPNs. Furthermore, the more clinically advanced Wnt pathway inhibitors target components that are upstream of b-catenin transcriptional activity and would therefore be futile for the treatment of SPN patients.

Conclusions
Taken together, our data suggest that SPNs have simple genomes and appear to be uniformly driven by CTNNB1 exon 3 mutations with consequent activation of the Wnt signaling. Future drugs designed to target directly b-catenin may have therapeutic potential in this disease.  Heatmap of methylation data from SPN samples profiled in this study and ENCODE samples of epithelial origin. Probes found to be differentially methylated (FDR < 0.1) between SPN and ENCODE samples were mapped to the Wnt pathway. With few exceptions, the probes mapped to gene bodies where hypomethylation corresponded to upregulation of the corresponding transcript. The two exceptions were probes mapping to ZIC2 and WNT5B which were located in 3 0 UTR regions where hypermethylation also corresponded to upregulation of the upstream gene.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig. S1. Copy number profiles of SPNs analyzed in this study. Fig. S2. Differential gene expression analysis of SPNs, ADCs and PNETs. Table S1. Whole-exome sequencing statistics. Table S2. Nonsynonymous somatic mutations identified in SPNs using whole-exome sequencing. Table S3. Genes in the Wnt pathway assessed by Nanostring nCounter. Table S4. Probes differentially methylation between SPN samples (this study) and ENCODE samples of epithelial origin.