Clinical annotation and genotyping for ATRX, DAXX, and MEN1
We initially performed Sanger sequencing to genotype the ATRX, DAXX, and MEN1 genes in 64 individual PanNETs. All cases were histologically confirmed to be well-differentiated PanNETs of WHO G1/G2 grade, and cases of poorly differentiated neuroendocrine carcinoma were excluded. The mean patient age was 52 ± 1.5 years (mean ± standard error, ranging from 26–73) with a 59% male population. The locations of the tumors were 38% proximal/mid body, and 62% distal pancreas. Eighty-one percent of the cases were clinically non-functional and the remaining cases included insulinomas, glucagonomas, gastrinomas, and VIPomas. The median size of tumor was 3.6 ± 0.4 cm (ranging from 1.0–14.5 cm). Sixty-eight percent of patients had localized disease without distant metastasis at the time of initial diagnosis (Supplementary data 1).
An A-D-M mutant genotype was identified in 58% (37/64) of cases with ATRX, DAXX, MEN1, MEN1/ATRX, and MEN1/DAXX mutations in 8, 16, 20, 3, and 11% cases, respectively (Fig. 1a). The majority of mutations in ATRX, DAXX, and MEN1 were truncation mutations (stopgain or frameshift) and loss of function consistent with their role as tumor suppressors (Supplementary Data 2). Similar to the observations in our previously published datap-value
Fig. 1
Mutational landscape of ATRX, DAXX, and MEN1 in PanNETs. a Oncoprint mutational profile for PanNETs samples. ATRX/DAXX/MEN1 mutations were identified in 37/64 (58 %) of PanNETs using Sanger sequencing. b Among 44 patients who initially presented with localized PanNETs (without distant metastasis), those with A-D-M mutant genotype had a worse recurrence free survival outcome than those A-D-M WT genotype in their primary tumors. A-D-M mutated samples are annotated as any mutation (n = 20) and A-D-M WT samples annotated as WT (n = 24)
Gene expression and DNA methylation reveal two subtypes of PanNETs
We performed RNA sequencing on 33 randomly selected tumors (19 A-D-M mutant, and 14 A-D-M WT). Unsupervised hierarchical clustering of the top 3000 variable genes across the PanNETs revealed two distinct clusters where almost all A-D-M mutant PanNETs were found in one cluster (Fig. 2a). The grouping of A-D-M mutant PanNETs into one distinct cluster by gene expression was robust to the number of most variable genes used for clustering (Supplementary Figure 1). Principal component analysis (PCA) separated the A-D-M mutant PanNETs from the A-D-M WT PanNETs along the first principal component (corresponding to the component comprising the largest variation in gene expression) (Fig. 2b). The separation of A-D-M mutant PanNETs from A-D-M WT PanNETs by PCA was robust to the number of top variable genes used (Supplementary Figure 2). These data show that A-D-M mutant tumors have a distinct gene expression pattern from that of A-D-M WT PanNETs. Neither hierarchical clustering nor PCA from gene expression revealed further subgrouping of the tumors with single mutations in ATRX, DAXX, or MEN1 or double mutations in ATRX/MEN1 or DAXX/MEN1.
Fig. 2
A-D-M mutant and WT PanNETs as two distinct gene expression groups. a Unsupervised clustering of PanNETs using top 3000 variant genes across all samples revealed two distinct robust clusters. The two clusters almost perfectly separate A-D-M WT panNETs from A-D-M mutant panNETs. b Principal component analysis using top 3000 variant genes separated the A-D-M mutant from A-D-M WT PanNETs along the first principal component (PC1). A-D-M mutant panNETs were more homogeneous in gene expression than A-D-M WT as shown by smaller variation along PC1. c Heatmap of pair-wise Pearson correlation of panNETs using top 3000 variant genes across all samples revealed a higher correlation among A-D-M mutants as compared to A-D-M WT panNETs. Red color represents higher correlation and blue represents lower correlation. d Heatmap of top variants genes showing liver, complement, and coagulation genes highly expressed in A-D-M mutant panNETs. Star (*) below sample names represent liver metastatic samples (except for A_mk12 which is a lymph node)
In hierarchical clustering, the A-D-M mutant PanNETs formed a tighter cluster than the A-D-M WT PanNETs. In PCA, the A-D-M mutant PanNETs had smaller variance along PC1 than A-D-M WT PanNETs. Pair-wise correlation of gene expression between all PanNETs, showed a higher correlation among A-D-M mutant PanNETs as compared to A-D-M WT PanNETs (Fig. 2c). Among A-D-M mutant PanNETs, mutants with the same genotype (mutations in ATRX/DAXX/MEN1) did not show greater gene expression correlation. These data suggest that A-D-M mutant PanNETs are a more homogeneous group compared to A-D-M WT PanNETs.
Within A-D-M mutant or A-D-M WT PanNETs groups, unsupervised clustering and PCA did not reveal differences between primary and metastatic tumors. Top 100 genes with highest variance across all samples separates mutant from A-D-M WT PanNETs and showed relatively high expression of “liver-specific” genes (APOH, ALDH1A1, FGB, APOC3 etc.) as well as complement and coagulation pathway genes (SERPINA1, FGA, F10, CP, MT3 etc.) in A-D-M mutant PanNETs (Fig. 2d; Supplementary Figure 3), both in primary (collected in absence of liver tissue) and metastatic tumors. Moreover, the pathological estimate of tumor purity was over 80% for all samples of PanNETs consistent with inference from ESTIMATE3) showing high tumor purity characteristic of well differentiated PanNETs. In addition, seven A-D-M mutants and one A-D-M WT PanNETs were from the tissue of liver metastases and they had gene expression profile most similar to the genotype group of their primary PanNET counterpart (Fig. 2d). We confirmed the distinct gene expression signature of A-D-M mutant PanNETs in a larger tumor set (47 PanNETs including the 33 PanNETs where RNA sequencing was performed) using gene expression microarray technology. The 14 additional samples are comprised of 3 A-D-M WT PanNETs and 11 A-D-M mutant PanNETs (7 MEN1 mutant, 2 DAXX mutant, and 2 DAXX/MEN1 mutant)(Supplementary Figure 4).
To investigate epigenetic differences between PanNETs, we used the Illumina 450 K chip to assay the DNA methylation at 411,549 CpG sites in 32 PanNETs. Unsupervised hierarchical clustering of the top 2000 variable DNA methylation sites across the PanNETs revealed two distinct clusters where almost all A-D-M mutant PanNETs were found in one cluster (Fig. 3a). Principal component analysis (PCA) separated the A-D-M mutant PanNETs from the A-D-M WT PanNETs along the first principal component (corresponding to the component comprising the largest variation in DNA methylation) (Fig. 3b). The separation of A-D-M mutant PanNETs from A-D-M WT PanNETs by PCA was robust to the number of top variable DNA methylation sites used (Supplementary Figure 5). These data reveal that A-D-M mutants PanNETs have a distinct DNA methylation pattern from that of A-D-M WT PanNETs. Neither hierarchical clustering nor PCA revealed differences in DNA methylation sites between the different combinations of genes mutated among the A-D-M mutant PanNETs. Within A-D-M mutant or A-D-M WT PanNETs groups, unsupervised clustering and PCA of DNA methylation did not reveal differences between primary and metastatic tumors.
Fig. 3
Distinct DNA methylation pattern between A-D-M mutant and A-D-M WT PanNETs. a Unsupervised clustering of PanNETs using top 2000 variant CpG sites across all samples revealed two clusters. The two clusters separate A-D-M mutant from A-D-M WT PanNETs. b Principal component analysis using top 2000 variant CpG sites separated A-D-M mutant from A-D-M WT PanNETs along PC1
To investigate the global histone methylation level in PanNETs with and without A-D-M mutations, we performed immunohistochemistry on H3K4me3, H3K9me3, H3K27me3, and H3K36me3 on 36 PanNETs. There was a general trend of lower histone methylation level for MEN1 mutated PanNETs when compared to WT PanNETs (Supplementary Figure 6 and Supplementary Table 1).
A-D-M mutant PanNET gene expression resembles that of alpha cells
There are multiple neuroendocrine cell types in the pancreas including alpha, beta, gamma, delta, and epsilon. We used gene expression data for these various pancreatic neuroendocrine and exocrine cell types from a single cell sequencing study2) to identify gene-set signatures representing highly expressed cell-type-specific genes (Supplementary data 4). The A-D-M mutant PanNETs uniformly exhibited a gene expression signature that was very similar to that of alpha cells (Fig. 4a). The A-D-M WT PanNETs were more heterogeneous in their expression of the genes among the gene set signatures for the different pancreatic neuroendocrine cell types. Greater heterogeneity of gene expression signature in A-D-M WT PanNETs was consistent with the greater heterogeneity found in global gene expression.
Fig. 4
A-D-M mutant PanNETs with alpha-cell signature. a Heatmap of gene expression for top 20 alpha and beta cell-specific genes from Muraro et al.b Gene set enrichment analysis show A-D-M mutant PanNETs to be enriched for expression of alpha cell specific genes. Pancreas cell type (alpha, beta, delta, PP, acinar, ductal) gene signatures were obtained from three different published dataset to access enrichment of cell type signatures in A-D-M mutant vs A-D-M WT PanNETs. Table represents GSEA results where size is the number of genes in gene set. All alpha cell gene sets (from three different sources) are significantly enriched in A-D-M mutant panNETs (highlighted in red). No other cell types were enriched in A-D-M mutant or A-D-M WT panNETs. c GSEA plots of significant alpha cell signatures from Bramswig et al.
To further investigate the gene expression signature of A-D-M mutant PanNETs we performed gene set enrichment analysis2). Our analysis indicates that only the alpha cell gene signature was significantly enriched in A-D-M mutant PanNETs (FDR q-value
Alpha and beta cell lineage specific genes were examined for the A-D-M mutant and WT PanNETs. ARX, IRX2, and TM4SF4 were all highly expressed in A-D-M mutant PanNETs compared to A-D-M WT PanNETs (Supplementary Figure 7). Surprisingly, GCG (glucagon) expression was lower in A-D-M mutants as compared A-D-M WT PanNETs. For beta cell specific genes, PDX1, MAFA, INS, and DLK1, all had lower expression in A-D-M mutant PanNETs than A-D-M WT PanNETs (Supplementary Figure 7). However, these genes had much greater expression heterogeneity in A-D-M WT PanNETs suggesting that some A-D-M WT PanNETs resemble beta cells and others did not (Supplementary Figure 7).
Validation of subtype and alpha cell signature in A-D-M mutant PanNETs
We derived an A-D-M mutant gene expression signature from significant differentially expressed genes between the A-D-M mutant and WT PanNETs from our data set (n = 33). We used two independent panNETn = 29)n = 75)5a) (Supplementary data 5). Additionally, we found alpha cell signatures to be significantly enriched (FDR q
Fig. 5
Validation of A-D-M mutant PanNET and alpha cell signatures. a Pearson correlation boxplot for two independent PanNET datasets show significant positive and negative correlations of A-D-M mutant and WT PanNETs with our A-D-M mutant PanNETs signature respectively (red represent A-D-M mutant and blue represent A-D-M WT with Wilcox p-value; center line is median, bounds of box are first and third quartile, and whiskers are min and max). b GSEA analysis shows A-D-M mutant PanNETs from ICGC PAENc GSEA enrichment plot for significant gene set for A-D-M mutant and alpha cell gene signatures from Sadanandam et al. dataset
HNF1A pathway is up-regulated in A-D-M mutant PanNETs and alpha cells
HNF1A is one of the most significantly differentially expressed genes between A-D-M mutant and WT PanNETs. HNF1A is a homeobox family transcription factor that is highly expressed in the liver and is involved in the regulation of several liver-specific genes. The expression of HNF1A was 2.93 fold higher in A-D-M mutant PanNETs than A-D-M WT PanNETs (corrected p-value
Fig. 6
HNF1A pathway with transcriptionally up-regulation in A-D-M mutant panNETs and alpha cells. a Boxplot of HNF1A gene expression for A-D-M mutant and A-D-M WT PanNETs. HNF1A was homogeneously expressed 2.93 fold higher in A-D-M mutants PanNETs (corrected p-val
Many of the most differentially expressed genes and highly expressed in A-D-M mutant PanNETs are targets of HNF1A and are involved in protein secretion, transport and metabolism (APOH, ALB, AFM, HAO1, UGT1A3, UGT1A1, GC, G6PC, TM4SF4, PKLR etc). APOH is expressed 8.46 fold higher in A-D-M mutant PanNETs (p-value
PDX1 gene is hypermethylated with low expression in A-D-M mutant PanNETs
There is no genome wide hypo or hypermethylation of DNA in A-D-M mutant or WT panNETs. DNA methylation differences between the A-D-M mutant and A-D-M WT PanNETs were found at 378 CpG sites (corrected p-value 0.2, Benjamini–Hochberg), 287 of which were found in genes and 91 in intergenic regions (Supplementary data 7). Of the 287 differentially methylated genic CpG sites, 70 (associated with 59 genes) were found at promoter (transcriptional start site, TSS1500 and TSS200) or within first exon, a region where DNA methylation is associated with transcriptional repressionp-value
Fig. 7
PDX1 has promoter hypermethylated and lower gene expression in A-D-M mutant panNETs. a Four PDX1 promoter CpG sites show strong hypermethylation in A-D-M mutant PanNETs (corrected p-val
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
