TCGA data and patient-derived orthotopic xenografts highlight pancreatic cancer-associated angiogenesis

Pancreatic ductal adenocarcinomas (PDACs) overexpress pro-angiogenic factors but are not viewed as vascular. Using data from The Cancer Genome Atlas we demonstrate that a subset of PDACs exhibits a strong pro-angiogenic signature that includes 37 genes, such as HDAC9, that are overexpressed in PDAC arising in KRC mice, which express mutated Kras and lack RB. Moreover, patient-derived orthotopic xenografts can exhibit tumor angiogenesis, whereas conditioned media (CM) from KRC-derived pancreatic cancer cells (PCCs) enhance endothelial cell (EC) growth and migration, and activate canonical TGF-β signaling and STAT3. Inhibition of the type I TGF-β receptor with SB505124 does not alter endothelial activation in vitro, but decreases pro-angiogenic gene expression and suppresses angiogenesis in vivo. Conversely, STAT3 silencing or JAK1–2 inhibition with ruxolitinib blocks CM-enhanced EC proliferation. STAT3 disruption also suppresses endothelial HDAC9 and blocks CM-induced HDAC9 expression, whereas HDAC9 re-expression restores CM-enhanced endothelial proliferation. Moreover, ruxolitinib blocks mitogenic EC/PCC cross-talk, and suppresses endothelial p-STAT3 and HDAC9, and PDAC progression and angiogenesis in vivo, while markedly prolonging survival of KRC mice. Thus, targeting JAK1–2 with ruxolitinib blocks a final pathway that is common to multiple pro-angiogenic factors, suppresses EC-mediated PCC proliferation, and may be useful in PDACs with a strong pro-angiogenic signature.


Immunohistochemistry
Tissues were harvested, fixed in 10% formalin overnight and embedded in paraffin. 4 μm sections were prepared using a HM355S microtome (Thermo Scientific). After deparaffinization and tissue rehydration, antigen retrieval was performed with antigen unmasking solution (Vector Labs, Burlingame, CA) or EDTA. Immunohistochemical detection was using biotinylated secondary antibodies, and a NOVA RED detection kit (Vector Labs). Masson's Trichrome (Richard-Allan Scientific) and Picosirius Red (Polysciences, Inc.) were performed according to manufacturer's recommendations.

Microarrays
Murine array intensity values were extracted, converted to log2-scale, and LOESS normalization was performed. Unpaired t-tests with equal variance were used to test log2-normalized data for significant differences. P-values were subjected to multiple testing (Benjamini-Hochberg) correction to reduce false discovery rate (FDR). Differentially expressed genes were considered statistically significant using fold change (FC: -1.5-fold and 1.5-fold), P-value (P < .001) and FDR (FDR < 0.05) cutoffs.
For gene ontology (GO) analysis, each probeset on the array (55,681 probesets), gene annotation information, including the Entrez Gene ID, RefSeq ID or gene symbol if available, were used to identify the associated Mouse Genome Informatics (MGI) ID. The final MGI ID gene list and the background gene list were uploaded on the Database for Annotation, Visualization and Integrated Discovery site (DAVID, version 6.7), and an analysis was run using the MGI ID as the identifier. The enriched GO terms in the biological process FAT ontology were focused on, and the visualization tool in AmiGO was used to generate the graphical GO graphs. For ingenuity pathway analysis (IPA), Agilent probe IDs, fold changes and P-values were uploaded into IPA, and a core analysis was performed using FC (-1.5-fold and 1.5-fold) and P-value (P < .01) cutoffs.

TCGA analysis
From the PAAD TCGA data, 580 Human UniProt IDs directly or indirectly annotated to the gene ontology [1] [2][3][4][5], and 384 had expression values in the TCGA dataset. Hierarchical clustering was performed in R by applying a Pearson correlation distance and average linkage function to the normalized RSEM values of the 384 genes for the 85 tumor samples, and then scaling and graphing the result using the heatmap.2 function of the gplots R package. Because the resulting dendrograms indicated there was a subset of patients with up-regulated angiogenesis genes, we zoomed in on a dendrogram leaf of 129 genes, and then reclustered the data as follows.
The normalized RSEM values for the 129 genes for the 85 tumor samples were first centered and scaled in R, and then hierarchical clustering on the rows was done using the Pearson correlation distance and average linkage function while column clustering was done using the Euclidian distance and complete linkage function. Differential expression analysis between the strong and weak groups was carried out using DESeq (7) on the raw count data with 77 significantly changed genes meeting the following cutoff: FC ≥ 1.5; P < 0.01; FDR < 0.05. For comparison to KRC tumors, the 129 Entrez gene IDs were mapped to 127 Mouse Genome Informatics (MGI) ids by using the Vertebrate Homology file available at MGI. Of the 77 differentially expressed genes in the human dataset, 73 had mouse homologs, and 37 of those were also differentially expressed in the same direction according to the same cutoffs.

Gene set enrichment analysis (GSEA)
For KRC tumors, normalized, log2-transformed data from the were prepared in GSEA format [6]. A custom chip file that mapped probesets from this array to HUGO gene symbols was generated. Using version 2.1.0 of the command line jar application, this dataset was then compared against the 77 TCGA strong angiogenesis gene signature ( Figure 3E) and a TGF-β gene set [7] ( Figure 5B) from the Molecular Signature Database (MSigDB), version 4.0. For the TCGA data, modified log2 fold changes generated from the strong vs. weak DESeq analysis were used to rank the genes.  Table 1: Genes annotated to angiogenesis GO terms are differentially expressed in a subset of PDACs. Shown is the differential expression analysis of PDACs with strong or weak angiogenesis signatures as determined by TCGA analysis. 77 genes are significantly (FC ≥ 1.5; P < 0.01; FDR < 0.05) up-regulated in tumors with a strong signature, and of these, 63 are pro-angiogenic whereas 14 are anti-angiogenic. Genes are ranked by P-value and FDR.

Number
Gene