RNA expression profiling of peritoneal metastasis from pancreatic cancer treated with Pressurized Intraperitoneal Aerosol Chemotherapy (PIPAC)

Clinical data and inclusion of specimens

We included peritoneal biopsies (PBs) from patients with treatment-naïve PM-PC, using the following inclusion criteria: The diagnosis (PM from pancreatic ductal adenocarcinoma) was established on a PB, typically a frozen section as part of a laparoscopic evaluation, performed at Odense University Hospital (OUH), Denmark, in the period from 01.03.2015 to 28.02.2019. The patient had to be treatment-naïve. Region of interest (ROI) had a size of ≥20 mm². This resulted in the inclusion of 10 patients with PM-PC (Figure 1A).

Figure 1:

Flow chart diagrams of patient inclusion. (A) Inclusion of patients with chemotherapy-naïve peritoneal metastasis from pancreatic cancer (PM-PC). (B) Inclusion of patients with PM-PC and peritoneal biopsies showing complete regression after systemic chemotherapy and PIPAC (Regression). OUH, Odense University Hospital; PDAC, pancreatic ductal adenocarcinoma; PRGS, Peritoneal Regression Grading Score.

We included PBs from patients with regression of PM-PC after systemic chemotherapy and PIPAC treatment (Regression group), who had received their first PIPAC treatment in the period 01.01.2016–30.06.2020, using the following criteria: One PB with complete regression (PRGS-1) was included, the patient had at least one PIPAC treatment prior to the included PB, and the ROI had a size of ≥7 mm². First, PBs taken prior to PIPAC-3 were evaluated, followed by those obtained prior to PIPAC-2, PIPAC-4, and, lastly, PIPAC-5. The first PB fulfilling these criteria was included in the Regression group. Of note, the entire quadrant PB set did not have to show complete regression (PRGS-1). This resulted in the inclusion of 11 patients with Regression (Figure 1B).

To be able to compare possible transcriptomic differences between Regression and chronic peritonitis–related fibrosis (Controls), we included 10 cholecystectomy specimens with chronic cholecystitis and severe subperitoneal fibrosis. These were included consecutively, starting with patients who had undergone cholecystectomy at OUH from 01.01.2016 onward. The ROI had a size ≥20 mm².

This study complies with the World Medical Association Declaration of Helsinki regarding ethical conduct of research involving human subjects. It was approved by the Scientific Ethics Committee of the Region of Southern Denmark (project-ID S-20200122) and by the Data Protection Agency of Region of Southern Denmark (journal-no 20/51469). We ensured that patients had not advocated against the use of their tissue in the Danish registry for the use of tissue in research (“Vævsanvendelsesregisteret”).

The Peritoneal Regression Grading Score (PRGS)

PBs were assessed by the four-tiered PRGS to evaluate the histological response to therapy in patients with PM [14]. Scoring was in all cases performed by the same pathologist (SD) who had an interest in peritoneal pathology. PRGS is based on the presence of residual tumor cells in relation to the extent of regressive features. Major histological features of regression are fibrosis, inflammation, hyalinosis, acellular mucin pools, ischemic necrosis, accumulation of macrophages, multinucleated giant cells, and granulomas. PRGS-1 corresponds to a complete regression with absence of tumor cells; PRGS-2 to a major histological response; PRGS-3 to a minor histological response; and PRGS-4 to metastatic tumor only, with a lack of histological response to therapy [13]. PRGS is given for each PB separately, in addition to a mean score for all biopsies belonging to a given quadrant PB set.

RNA extraction

We mounted 10 µm-thick FFPE sections from the PM-PC, Regression, and Control cases on Superfrost PlusTM Adhesion Microscope Slides (ThermoFisher Scientific, Braunschweig, Germany, J1800AMNZ). We aimed to include a total area of 100 mm², and the number of sections was adjusted accordingly. Macrodissection was done when appropriate. Sections were dried overnight prior to deparaffinization. Then, RNA extraction was performed using High Pure FFPE RNA isolation kit (Roche diagnostics GmbH, Mannheim, Germany, 06650775001) according to the Prosigna protocol. RNA concentration and A260/280 were determined using NanoDrop One (Thermo Fischer Scientific, Wilmington, DE, USA). All samples were immediately stored at −80 °C.

mRNA expression profiling and data processing

For the 31 specimens, mRNA gene expression levels were assessed using the PanCancer immune-profiling panel, Immuno-Oncology 360 (IO-360, NanoString Technologies, Seattle, WA), designed to give a unique 360° view of gene expression in the tumors. The panel consists of 750 genes (and in addition 20 housekeeping genes), falling into eight functional categories: (1) tumor immunogenicity, (2) tumor sensitivity to immune attack, (3) inhibitory immune mechanisms, (4) stromal factors, (5) inhibitory metabolism, (6) antitumor immune activity, (7) inhibitory immune signaling, and (8) immune cell population abundance. A total of 30 custom genes were added, related to fibroblasts, endothelial cells, and extracellular matrix (ECM), as described previously [22]: ACTA2, ANO1, CALD1, CD34, CEACAM5, COL3A1, COL4A1, CYGB, FN1, GPC1, HAS2, INS, KCNH2, KRT7, KRT8, LGALS1, MME, MUC1, NES, PDPN, POSTN, PRSS1, S100A4, SLC16A3, SMAD4, SPARC, SYP, TNC, VCL, and VIM [23].

RNA samples were aligned with a synthetic panel standard and hybridized in a 12-strip tube well with gene-specific probes (reporter and capture) for 16–21 h at 65 °C, following the manufacturer’s protocol. Subsequently, the target-probe complexes were read and counted by scanning of 550 fields of view (FOV) using the nCounter Digital Analyzer (NanoString Technologies, Seattle, WA) [24]. Raw digital counts of expression were exported to the nSolver v4.0 software (NanoString) for downstream analysis, following the manufacturer’s protocol.

In nSolver v. 4.0, the data were investigated to check quality of reads according to manufacturer’s instructions. Genes with an expression level below the average count of negative controls plus two standard deviations were considered undetected. After normalization against chemical positive controls, data were exported from nSOLVER 4.0 software. The subsequent calculations and analyses were performed using the open source R-environment (v. 4.3.2).

Statistics

Values were given as medians with range, means with standard deviation (SD), or percentages where appropriate. Comparisons between gene counts for Regression vs. PM-PC and Regression vs. Controls were performed using the open source R-environment (R version 4.3.2) (http://cran.r-project.org/). Gene expression data were converted to counts-per-million (CPM), normalized using the trimmed mean of M-values (TMM) method and converted to log2 scale, and based on all genes, the samples were analyzed by unsupervised hieararchial clustering, and expression levels were visualized using a heatmap. The heatmap was created using the heatmap function embedded in the ComplexHeatmap R-package [25, 26].

Differential gene expression analysis between Regression vs. PM-PC and Regression vs. Controls was performed using an un-paired limma t-test embedded in the limma R-package [27]. All comparisons were adjusted for multiple testing using the false discovery rate (FDR), and genes with FDR≤0.05 were considered as being significantly differentially expressed. The results of the differential gene expression analysis were visualized by volcano plots, using the EnhancedVolcano function embedded in the EnhancedVolcano R-package [28]. Waterfall plots were created by application of the boxplot R function and using the output from the differential gene expression analysis tests, comparing Regression with PM-PC and Regression with Controls. Specifically, the genes were sorted by the log2 fold change in descending order, and this list was visualized in a barplot, where significantly differentially up- and down-regulated genes were colored red and blue, respectively.

Gene set enrichment analysis (GSEA) was performed using the fgsea (v.1.2.8) R-package [29, 30]. GSEA was run on a list of preranked individual expressed genes using the Log2 fold change values derived from the differential gene expression analysis as ranking metric. The collection of human hallmark gene sets (https://www.gsea-msigdb.org/gsea/msigdb/human/genesets.jsp?collection=H) was used as input. The GSEA was conducted using 1,000 permutations, eps set to zero, and minimum and maximum gene set sizes were set to 15 and 500, respectively. Gene sets with FDR≤0.05 were considered as being statistically significantly enriched. A selection of significantly enriched gene sets were visualized by an enrichment plot using the plotEnrichmentData function embedded in the fgsea R-package.

Histology and immunohistochemistry

PBs from the Regression group, PM-PC, and Controls were fixed in formalin and embedded in paraffin (FFPE). Three to four-micron sections were cut and mounted on FLEX IHC microscope slides and stained with hematoxylin & eosin (H&E). Sections were dried at room temperature and baked at 60 °C for 60 min before immunostaining. Antibodies used are shown in Supplementary Table 1. Nuclear counter staining was performed using Hematoxylin FLEX at the Dako Omnis platform. Slides were washed, dehydrated, and cover slipped using an automated Dako cover slipper (Dako/Agilent, Glostrup, Denmark).

Patients’ characteristics

Clinical baseline data are shown in Table 1. Patients with PM-PC had a median age of 71 years (range 57–82). The Regression group had a median age of 56 (range 49–71). Median number of PIPAC treatments prior to the PB included in the Regression group was 2 (range 1–3). All patients in the Regression group had received first-line palliative systemic chemotherapy, and two patients had undergone second-line systemic chemotherapy. Patients in the Control group (chronic peritonitis due to chronic cholecystitis) had a median age of 60 years (range 19–72 years), seven females and three males, none with any known malignant disease.

Unsupervised clustering analysis of mRNA expression data

Based on unsupervised clustering of all mRNAs included in the mRNA profiling (n=800), we generated a principal component analysis (PCA) plot (Figure 2A). Clear separation between PM-PC and Controls was observed, with accumulation of Regression cases in between. A heat map visualized three clusters, where all PM-PCs accumulated in one and all Controls in another (Figure 2B). Regression cases accumulated in a separate cluster, with a few Regression cases clustering together with Controls. Venn diagrams, illustrating up- and down-regulated genes in Regression vs. PM-PC and in Regression vs. Controls, are shown in Figure 2C and D.

Figure 2:

Unsupervised clustering and differential expression of RNA profiling data (800 mRNAs) from peritoneal biopsies (PBs) with chemotherapy-naïve peritoneal metastasis from pancreatic cancer (PM-PC), complete regression after systemic chemotherapy and PIPAC (Regression), and Controls. (A) Principal component analysis (PCA) plot. Clear separation between PM-PC and Controls, with accumulation of Regression cases in between. (B) Heat map based on unsupervised hierarchical clustering of all mRNAs, visualizing three clusters, where all PM-PCs accumulate in one and all Controls in another. Regression cases accumulate in a separate cluster, with a few Regression cases clustering together with Controls. (C) Venn diagram illustrating down-regulated genes in Regression (n=255). There is overlap in 41 of these genes, found when comparing Regression vs. PM-PC and Regression vs. Controls. (D) Venn diagram illustrating up-regulated genes in Regression (n=47). There is overlap in two of these genes, found when comparing Regression vs. PM-PC and Regression vs. Controls.

Differentially expressed mRNAs in Regression compared to PM-PC

When comparing the mRNA expression in the Regression group with PM-PC, we found six mRNAs that were significantly up-regulated: NCAM1, IL-33, ANGPT1, DPP4, CD209, and ACVR1C (FDR≤0.05). In total, 197 mRNAs were significantly down-regulated (Table 2 and Supplementary Table 2). Figure 3A and B show a volcano plot and waterfall plot, respectively, illustrating the significantly differentially expressed mRNAs when comparing the Regression group with chemotherapy-naïve PM-PC. The significantly differently expressed genes were evaluated for over-representation of gene sets or pathways in MSigDB. When comparing with the hallmark gene sets as per gene set enrichment analysis (GSEA), TNF-α signaling via NF-kB, G2M checkpoint, epithelial-mesenchymal transition, late and early estrogen response, and coagulation were significantly down-regulated gene sets in Regression (Table 3 and Supplementary Figure 1).

Figure 3:

Differential expression of 800 mRNAs from peritoneal biopsies (PBs) with complete regression after systemic chemotherapy and PIPAC (Regression) compared to chemotherapy-naïve peritoneal metastasis from pancreatic cancer (PM-PC) and Controls. (A) Waterfall plot illustrating up- and down-regulated genes when comparing Regression vs. PM-PC. Significantly up- and down-regulated genes (FDR≤0.05) are shown in red and blue, respectively. Names of up to eight most significantly differentially expressed genes are given in the graph. (B) Volcano plot illustrating results of differential gene expression analysis of Regression vs. PM-PC. (C) Waterfall plot illustrating up- and down-regulated genes when comparing Regression vs. Controls. Significantly up- and down-regulated genes (FDR≤0.05) are shown in red and blue, respectively. Names of the eight most significantly differentially expressed genes are given in the graph. (D) Volcano plot illustrating results of differential gene expression analysis of Regression vs. Controls.

Differentially expressed mRNAs in Regression compared to Controls

When comparing the mRNA expression in Regression with Controls, we found 43 mRNAs that were significantly up-regulated (FDR≤0.05) (Table 4 and Supplementary Table 3). In total, 99 mRNAs were significantly down-regulated (FDR≤0.05) (Table 4 and Supplementary Table 3). Figure 3C and D show a volcano plot and a waterfall plot, respectively, illustrating the significantly differentially expressed mRNAs when comparing Regression with Controls, which were evaluated for over-representation of gene sets or pathways in MSigDB. When comparing with the hallmark gene sets as per GSEA, interferon-α response was significantly up-regulated (Supplementary Table 4). TNF-α signaling via NF-kB, inflammatory response, epithelial-mesenchymal transition, KRAS signaling, and hypoxia were significantly down-regulated (Supplementary Table 4 and Figure 2).

Immunohistochemistry

To get an impression whether mRNA expression was related to expression of the encoded proteins, we performed immunohistochemistry for four proteins translated from four of the differentially expressed genes: CEA, EpCAM, maspin (encoded by Serpin B5), and vimentin. Representative images of histology of PBs with Regression, PM-PC, and Controls are shown in Figure 4A and B. EpCAM was strongly expressed in PM-PC but negative in Regression and Controls (Figure 4C). Vimentin was more strongly expressed in Regression and Controls, compared to PM-PC (Figure 4D). CEA (Figure 4E) and maspin (Figure 4F) were strongly expressed in PM-PC but negative in Regression and Controls.

Figure 4:

Histology and immunohistochemistry of regression after systemic chemotherapy and PIPAC (Regression), therapy-naïve peritoneal metastasis from pancreatic cancer (PM-PC), and chronic peritonitis–related fibrosis (Controls). (A) Panel showing representative specimens at low magnification (H&E, x210). (B) Panel showing representative specimens at high magnification (H&E, x710). (C) EpCAM immunostaining (x710). (D) Vimentin immunostaining (x710). (E) CEA immunostaining (x710). (F) Maspin (encoded by Serpin B5) immunostaining (x710).