Introduction

Pancreatic cancer is a grievous disease, and the outlook for afflicted patients remains dismal with an estimated 5-year survival of less than 10% [1]. It is the most common tumour among a heterogeneous group of neoplasms arising in the periampullary region, including tumours originating in the distal bile duct, pancreas, ampulla of Vater, and the periampullary duodenum. Unlike many other cancers, targeted therapies including immune checkpoint inhibition have shown little efficacy against pancreatic cancer and, if so, only for a small selection of patients [2-6]. Therefore, chemotherapy remains standard of care, leading to modest survival benefits [7]. Thus, there is a pressing need to identify novel complementary biomarkers to better distinguish patients who are likely to benefit from standard chemotherapy from those who will only suffer from negative side effects and thus fare better with other treatment approaches or best supportive care.

RNA-binding motif protein 3 (RBM3) is an RNA- and DNA-binding protein that has emerged as a promising independent predictive and prognostic biomarker in several solid tumours, including pancreatic cancer [8-14]. In a previous study by our group, silencing of RBM3 was found to render pancreatic cancer cells less sensitive to a variety of chemotherapeutic agents in vitro. Furthermore, in patients with resected pancreatic and other periampullary cancers, high tumour-specific expression of RBM3 was found to be associated with prolonged overall survival (OS) if adjuvant treatment had been given, whereas the opposite was seen if no adjuvant treatment had been given [8]. Similar associations between RBM3 and sensitivity to cisplatin have been described in epithelial ovarian cancer [14], and further mechanistic clues may be derived from another study on ovarian cancer, demonstrating links between RBM3 and cellular processes such as chromatin remodelling, DNA integrity maintenance, and repair [15].

The aim of the present study was to explore RBM3-regulated genes and cellular processes that may influence the biological properties and chemosensitivity of pancreatic adenocarcinoma using RNA interference and next-generation RNA sequencing of transcriptomes in vitro. The top deregulated genes and proteins were further validated in vitro and explored regarding their expression, clinicopathological correlates, and prognostic significance in tumours from a clinically well-characterised cohort of patients with resected pancreatic adenocarcinoma (n = 46).

Materials and methods Cell culture

Human pancreatic cancer cell lines BxPC-3, PANC-1, and MIAPaCa-2 were purchased from Sigma-Aldrich (St. Louis, MO, USA). The cells were maintained in RPMI1640 or DMEM supplemented with 10% foetal bovine serum (FBS) and antibiotics (100 U/ml penicillin and 100 μg/ml streptomycin) in a humified 5% CO2 atmosphere at 37 °C. All in vitro reagents, including cell culture medium and supplements, were purchased from ThermoFisher Scientific (Waltham, MA, USA) unless stated otherwise.

siRNA transfection

For siRNA transfection, pancreatic cancer cells were seeded in T-25 flasks (4–7 × 105 cells) and incubated for 72 h at 37 °C. The cells were then washed twice with phosphate-buffered saline and received growth medium without FBS, together with lipofectamine 2000 and negative control or anti-RBM3 (s11858 + s11860) siRNA in OptiMEM to a final siRNA concentration of 25 nm. The transfection was stopped after 4.5 h, medium was changed to full growth medium, and the cells were left to recover overnight. The following day, cells were harvested and spun down to pellets. The pellets were either fixated, dehydrated, and embedded in paraffin for immunohistochemistry or resuspended in TRIzol and stored at −20 °C for quantitative polymerase chain reaction (qPCR).

RNA sequencing

MIAPaCa-2 cells were transfected with siRNA targeting RBM3 or negative control, as described above, and RNA purification was performed in the same manner as for the qPCR samples. Samples were prepared in triplicate. RNA quantification and quality assessment were performed using Nanodrop 1000 (Mason Technology, Dublin, Ireland) and Bioanalyzer 2100 (Agilent, Santa Clara, CA, USA). cDNA libraries were prepared from the RNA samples using TruSeq Stranded mRNA Library Prep Kit on the NeoPrep instrument (Illumina, San Diego, CA, USA) according to the manufacturer's instructions, and sequenced (paired-end 1 × 75 bp) using the NextSeq 500 platform (Illumina). Fastq files were downloaded from the Illumina BaseSpace using the BaseSpace download tool and the quality of the files was determined using FastQC. Data were trimmed of sequencing adaptors and low-quality base calls using BBDuk tool in the BBMap package. Alignment to the human hg19/GRCh37 genome reference was done using STAR version 2.5.2a [16]. Duplicate reads were marked using Picard MarkDuplicates. Read counts were produced by the featureCounts tool from the SubRead package, combined for all samples and used as input for analysis of differential gene expression. Differential expression (DE) gene analysis was conducted using the R package DESeq2 [17] and genes with adjusted P value of <0.01 were selected for further analyses. The data set is deposited at the NCBI Gene Expression Omnibus database (GSE169758).

Real-time qPCR

The cell samples were thawed and RNA purification was performed using TRIzol with phasemaker tubes according to the manufacturer's instructions. Following this, RNA clean-up was performed with the RNeasy MinElute Cleanup Kit (QIAGEN, Hilden, Germany) and the RNA concentration was determined using Qubit with the RNA HS Assay Kit. Prior to qPCR, cDNA reverse transcription was performed using the High-Capacity cDNA Reverse Transcription Kit and total cDNA concentration was determined using Qubit with the DNA Assay Kit. Ten ng per reaction of each samples was used to run qPCR with RBM3, PDS5A, PRR11, or CCND3 TaqMan gene expression assay (Assay ID Hs00943160_g1, Hs00374857_m1, Hs00383634_m1, or Hs05046059_s1, respectively), with samples run in triplicates. GAPDH was used as endogenous control (Assay ID Hs03929097_g1).

Western immunoblotting

Cells were seeded in 6-well plates (2 × 105) and incubated for 48 h at 37 °C prior to siRNA transfection. The day after transfection, cells were washed, lysed, and stored at −20 °C. Protein determination was performed with Pierce BCA protein assay and 20 μg was used from each sample. Samples were denatured in Laemmli sample buffer (Sigma-Aldrich) and boiled for 5 min at 95 °C. The samples were placed on a 8–16% TGX gradient gel (Bio-Rad, Hercules, CA, USA) with high range rainbow markers at both ends (GE, Chicago, IL, USA). Following electrophoresis, wet tank transfer was performed and proteins were transferred to a 0.45-μm nitrocellulose membrane and dried for 1 h. Total protein stain was then done with Revert 700 (LI-COR, Lincoln, NE, USA) and the membrane was imaged at 700 nm. The membrane was destained and blocked with Intercept TBS blocking buffer (LI-COR). Primary antibody incubation was performed overnight at 4 °C with anti-RBM3 (Atlas Antibodies AB, Stockholm, Sweden), anti-PDS5A (HPA036661, Atlas Antibodies AB), anti-PRR11 (DCS22, Atlas Antibodies AB), anti-cyclin D3 (DCS22, Cell Signaling, Danvers, MA, USA), or anti-Actin (Cell Signaling, Sigma). The membrane was subsequently washed and incubated for 1 h with secondary antibody IRDye 800CW goat anti-mouse or IRDye 680RD goat anti-rabbit (LI-COR). Once the secondary antibody had been thoroughly rinsed off, near-infrared detection was performed using LI-COR Odyssey Fc imager at 700 or 800 nm. Images were analysed using Image studio software and quantification of relative protein expression, normalised to total protein content, was performed with Empiria studio software (LI-COR).

PRR11 expression in TCGA data set

Clinical data and normalised gene-level expression data from the pancreatic cancer cohort TCGA_PAAD were retrieved from The Cancer Genome Atlas (TCGA) project through the Genomic Data Commons (GDC) (https://portal.gdc.cancer.gov, downloaded on 25 May 2021) using the R package TCGAbiolinks. RNA sequencing data were available for 178 patients. Based on the published curation of the data set by Nicolle et al [18], patients with normal pancreas/ampulla/duplicate samples (n = 12), non-pancreatic tumour (n = 4), non-invasive papillary neoplasms (n = 2), tumour origin other than pancreatic ductal adenocarcinoma (n = 9), treated with neoadjuvant treatment (n = 1), in addition to registered follow-up time of less than 30 days (n = 5), were excluded from subsequent analyses. The fragments per kilobase of exon per million mapped reads (FPKM) values were retrieved and the optimal cut-off point for dichotomisation of PRR11 mRNA expression into low versus high was determined using the survminer package, based on maximally selected rank statistics from the maxstat package.

Kaplan–Meier analysis and log-rank test were applied for evaluation of the prognostic impact of PRR11 mRNA expression in TCGA data set using the survminer package.

Study cohort

The study cohort is a previously described retrospective consecutive cohort of primary tumours from 175 incident cases of periampullary adenocarcinoma, including pancreatic cancer (n = 46) [19]. All patients underwent pancreaticoduodenectomy at the University hospitals of Malmö and Lund in the time span of 1 January 2001 to 31 December 2011. Follow-up started at the date of surgery and ended at death or on 31 March 2017, whichever came first. The Swedish National Civil Register was used to obtain information on vital status. Data on adjuvant treatment were obtained from patient charts. All cases underwent thorough histopathological re-evaluation.

The study received approval from the Ethics Committee of Lund University (reference numbers 2007/445, 2008/35, and 2014/748), through which the committee determined no necessity for informed consent other than the option to withdraw.

Immunohistochemistry and staining evaluation

For immunohistochemical analysis of PDS5A, cyclin D3, and PRR11 expression, 4 μm tissue microarray (TMA) sections were automatically pre-treated using the PT Link system and then stained in an Autostainer Plus (DAKO, Glostrup, Denmark) with the rabbit polyclonal anti-PDS5A antibody HPA036661 (diluted 1:100; Atlas Antibodies AB), the mouse monoclonal anti-cyclin D3 antibody DCS22 (diluted 1:1,600; Cell Signaling Technology), and the rabbit polyclonal anti-PRR11 antibody HPA023923 (diluted 1:50; Atlas Antibodies AB).

PDS5A and cyclin D3 were mainly expressed in the tumour cell nuclei, and the intensity of expression was denoted as either 0 (negative), 1 (weak), 2 (moderate), or 3 (strong). As PDS5A was found to be expressed in the majority of tumour cells, but with a varying intensity, the fraction of positive nuclear expression was denoted as 1 (≤50%) or 2 (>50%). For cyclin D3, being more sparsely expressed, the absolute fraction of positive cells was estimated. PRR11 was expressed in the cytoplasm and cell membrane, and the intensity of expression was denoted as either 0 (negative), 1 (weak), 2 (moderate), or 3 (strong), and the fraction of positive cells as 0 (0–10%), 1 (11–25%), 2 (26–50%), 3 (51–75%), or 4 (76–100%). The expression of PRR11 and cyclin D3 was annotated by two independent observers (SOH and KJ, the latter is a senior pathologist), and the expression of PDS5A was annotated by three independent observers (SC, VF, and KJ). A joint re-evaluation was then carried out and discrepant cases were discussed to reach consensus.

Immunocytochemistry

TMAs were constructed from the paraffin-embedded cell pellets in the same manner as the tissue samples, as was the subsequent staining of the cells.

Statistical analyses

For changes in mRNA levels after siRNA transfection in vitro, Student's t-test was performed. Non-parametric Wilcoxon signed-rank, Mann–Whitney U, and Kruskal–Wallis tests were applied for analysis of differences in the distribution of PDS5A, cyclin D3, and PRR11 protein expression in primary tumours and lymph node metastases and in relation to clinicopathological parameters. Spearman's rank correlation test was used to investigate the intercorrelations between the expression of investigative markers and RBM3. Two cases who received neoadjuvant therapy were excluded from all statistical analyses, and one additional case was excluded from the survival analyses due to emigration. Kaplan–Meier analysis and the log-rank test were applied to estimate differences in 5-year OS in relation to expression of PDS5A, cyclin D3, and PRR11. The fraction × intensity across all evaluable cores was calculated for each marker and dichotomised variables of low and high expression were then constructed. For PRR11, classification and regression tree (CRT) analysis established a prognostic cut-off corresponding to the median value. For PDS5A and cyclin D3, no prognostic cut-off could be established by CRT analysis, and the median value was therefore used in the survival analyses.

Cox regression proportional hazards modelling was applied to estimate hazard ratios for death within 5 years in relation to high and low expression of the three investigated markers. All significant variables from the univariable analysis (PRR11 expression, tumour grade, tumour stage, tumour size, involved lymph nodes, growth in lymph vessels, growth in blood vessels, and perineural growth) were entered into the multivariable analysis using a backwards stepwise model with a probability for stepwise entry at 0.05 and removal at 0.10. Statistical analyses were performed using SPSS Statistics version 25.0 (Arnmonk, NY, USA) and R version 4.1.0. A P value of <0.05 was considered statistically significant. Graphs were designed using SPSS, R, and GraphPad Prism version 9 (GraphPad Software, LA Jolla, CA, USA).

Results RBM3-associated cellular processes and genes

As MIAPaCa-2 cells have previously been shown to be the most appropriate model system to study the effects of RBM3 silencing on chemotherapy response [8], this cell line was selected for comparison of the transcriptomes of siRBM3-transfected and control cells by next-generation RNA sequencing.

As shown in Figure 1A, MIAPaCa-2 cells with downregulated RBM3 displayed 19 differentially expressed genes (all p < 0.01), of which 7 were downregulated (PDS5A, NIPSNAP3A, HIF1AN, SLC25A44, PIGN, MORF4L1, and AMBRA1) and 12 were upregulated (SRPR, PRR11, BOD1, CTD-2510F5.6, FAM49B, BANF1, EPB41L1, CIT, PIP4K2A, SMAP1, MCFD2, and CCND3). A summary of the genes and their key functions is provided in supplementary material, Table S1.

In vitro mapping of RBM3-related genes in pancreatic cancer. (A) Bar chart visualising 19 significantly DEGs (adjusted P value < 0.01) identified through RNA sequencing of siRBM3-transfected and control MIAPaCa-2 cells, of which 12 genes were upregulated and 7 downregulated. (B) Volcano plot of top up- and down-regulated genes showing PDS5A as the top downregulated gene and CCND3 as the top upregulated gene. NS, non-significant.

As further shown in the volcano plot in Figure 1B, the top downregulated gene was PDS5A (cohesin associated factor A), encoding the protein PDS5A involved in sister chromatid cohesion [20], and the top upregulated gene was CCND3, encoding the cell cycle regulating protein cyclin D3. Screening in the Human Protein Atlas (HPA) portal (and TCGA) identified three of the genes to be highly prognostic (p < 0.001) in pancreatic cancer (n = 176) at the mRNA level; EPB41L1 (shorter OS) encoding erythrocyte membrane protein band 4.1 like 1, an actin-binding protein, PRR11 (shorter OS) encoding proline rich 11, involved in cell cycle progression, and SLC25A44 (longer OS) encoding solute carrier family 25 member 44, involved in amino acid transport. Given the suggested association of RBM3 with chemosensitivity in pancreatic and other cancers, PRR11 was selected for further study based on its cellular functions, together with PDS5A and CCND3, being the top down- and up-regulated genes, respectively.

Effect of RBM3 silencing on expression levels of PDS5A, PRR11, and cyclin D3 in pancreatic cancer cells in vitro

Expression of the selected genes and corresponding proteins was examined in three siRBM3-treated pancreatic cancer cell lines, BxPC3, PANC-1, and MIAPaCa-2, and compared with control cells. The results demonstrate that knockdown of RBM3 led to reduced levels of PDS5A and increased levels of cyclin D3 and PRR11, both at the mRNA and protein levels, in MIAPaCa-2 cells (Figure 2), whereas no significant differences were seen in PANC-1 or BxPC-3 cells, apart from an upregulation of cyclin D3 in the latter. MIAPaCa-2 cells have a higher level of invasiveness and migration than PANC-1 and BPxPC-3 cells, which might explain why significant differences in protein levels were found only in MIAPaCa-2 cells [21].

Gene and protein expression of cyclin D3, PDS5A, and PRR11 in three siRBM3-treated pancreatic cancer cell lines and controls. (A) Representative images (×20 objective magnification) of the protein expression of PDS5A, cyclin D3, and PRR11 in BxPC3, PANC-1, and MIAPaCa-2 siRBM3-transfected cell lines and controls. (B) Bar charts of the gene expression of PDS5A, CCND3, and PRR11 in BxPC3, PANC-1, and MIAPaCa-2 cell lines compared to controls. ***p < 0.001, **p < 0.01, *p < 0.05. (C) Western blots showing the expression of PDS5A, cyclin D3, and PRR11 in BxPC3, PANC-1, and MIAPaCa-2 cell lines and in controls. ***p < 0.001, **p < 0.01, *p < 0.05.

Protein expression of PDS5A, cyclin D3, and PRR11 in primary tumours and lymph node metastases

Next, the immunohistochemical expression of PDS5A, cyclin D3, and PRR11 was examined in TMAs with matched primary tumours and lymph node metastases from 44 cases of resected pancreatic adenocarcinoma. PDS5A expression could be assessed in 43/44 (97.7%) of the primary tumours and in 24/44 (54.5%) of the lymph node metastases; cyclin D3 expression could be assessed in 41/44 (93.2%) of the primary tumours and in 16/44 (36.4%) of the lymph node metastases; and PRR11 could be assessed in 43/44 (97.7%) of the primary tumours and in 19/44 (43.2%) of the lymph node metastases. Sample immunohistochemical images are demonstrated in Figure 3A. As shown in Figure 3B, the protein expression did not differ significantly between primary tumours and lymph node metastases for PDS5A or cyclin D3, whereas significantly lower expression of PRR11 was found in lymph node metastases compared to primary tumours (p = 0.023).

Expression of PRR11 in primary tumours and in lymph node metastases. (A) Sample immunohistochemical images (×20 objective magnification) of PDS5A, cyclin D3, and PRR11 protein expression in pancreatic cancer. (B) Spaghetti plots visualising the expression of PDS5A, cyclin D3, and PRR1 in paired primary tumours and lymph node metastases.

Clinicopathological correlates of PDS5A, cyclin D3, and PRR11 protein expression

The distribution of patient and tumour characteristics according to PDS5A, cyclin D3, and PRR11 protein expression is shown in Table 1. Cyclin D3 expression was associated with lymph node metastases and PDS5A expression was associated with a high tumour grade and involved resection margins.

Table 1. Associations of PDS5A, cyclin D3, and PRR11 expression in primary tumours with clinicopathological parameters. n PDS5A mean, median (SD) p n Cyclin D3 mean, median (SD) p n PRR11 mean, median (SD) p Age Q1 (38–61) 4 0.58, 0.00 (1.17) 0.11 4 0.04, 0.03 (0.05) 0.62 4 8,75, 9.00 (0.50) 0.27 Q2 (62–67) 13 2.61, 3.25 (1.62) 12 0.27, 0.01 (0.77) 13 4.31, 3.00 (3.82) Q3 (68–72) 13 2.38, 3.00 (1.78) 13 0.35, 0.10 (0.65) 13 5.46, 4.00 (4.22) Q4 (73–84) 13 2.05, 2.00 (1.65) 12 0.27, 0.01 (0.56) 13 5.38, 6.00 (4.01) Gender Female 20 2.23, 2.17 (1.69) 0.96 19 0.43, 0.05 (0.80) 0.25 20 5.05, 5.00 (3.47) 0.63 Male 23 2.15, 2.00 (1.69) 22 0.14, 0.01 (0.38) 23 5.70, 6.00 (4.01) T stage T1-T2 9 1.67, 1.00 (1.34) 0.38 9 0.24, 0.01 (0.59) 0.64 10 6.00, 6.00 (3.62) 0.56 T3-T4 34 2.32, 2.75 (1.71) 32 0.28, 0.03 (0.64) 33 5.21, 6.00 (4.02) Lymph node metastasis N0 9 2.24, 2.50 (1.92) 0.93 9 0.03, 0.00 (0.07) 0.010 10 5,40 6.00 (3.98) 0.81 N1 23 2.24, 2.00 (1.68) 21 0.49, 0.00 (0.80) 22 5.05, 5.00 (3.70) N2 11 2.02, 2.00 (1.48) 11 0.05, 0.00 (0.08) 11 6.09, 8.00 (4.53) Tumour grade Low 15 2.52, 3.00 (1.88) 0.47 14 0.17, 0.01 (0.47) 0.29 15 3.73, 3.00 (3.35) 0.043 High 28 2.01, 2.00 (1.52) 27 0.32, 0.05 (0.68) 28 6.29, 6.00 (3.95) Tumour size (mm) ≤20 5 2.17, 2.00 (1.99) 0.86 5 0.02, 0.00 (0.04) 0.07 6 4.00, 6.00 3.10) 0.30 >20 38 2.19, 2.17 (1.63) 36 0.31, 0.03 (0.65) 37 5.62, 6.00 (4.02) Resection margins R0 1 1.00, 1.00 (–) 0.57 1 0.20, 0.20 (−) 0.27 2 0.50, 0.50 (0.71) 0.041 R1-2 42 2.21, 2.17 (1.66) 40 0.27, 0.01 (0.63) 41 5.63, 6.00 (3.85) Perineural growth No 9 1.87, 1.50 (1.82) 0.51 9 0.21, 0.00 (0.60) 0.17 10 4.00, 3.00 (4.19) 0.15 Yes 34 2.27, 2.42 (1.62) 32 0.29, 0.05 (0.63) 33 5.82, 6.00 (3.79) Lymphatic invasion No 16 2.89, 3.29 (1.79) 0.053 15 0.18, 0.02 (0.47) 0.72 16 4.63, 3.00 (3.81) 0.33 Yes 27 1.77, 1.50 (1.43) 26 0.32, 0.01 (0.70) 27 5.85, 6.00 (3.97) Vascular invasion No 28 2.17, 2.00 (1.78) 0.93 28 0.25, 0.04 (0.61) 0.79 28 4.75, 4.00 (3.80) 0.16 Yes 15 2.21, 2.33 (1.44) 15 0.28, 0.01 (0.62) 15 6.60, 6.00 (3.96) Growth in peripancreatic fat No 11 2.02, 2.00 (1.83) 0.71 11 0.20, 0.02 (0.53) 0.78 12 5.17, 6.00 (3.74) 0.83 Yes 32 2.24, 2.17 (1.61) 30 0.30, 0.01 (0.65) 31 5.48, 6.00 (4.03) Bold values indicate p < 0.05.

The intercorrelations of PDS5A, cyclin D3, PRR11, and RBM3 expression are shown in supplementary material, Table S2. The only significant finding was a weakly positive correlation between PDS5A and RBM3 expression (R = 0.329, p = 0.031).

Prognostic significance of PDS5A, cyclin D3, and PRR11 protein expression

In the in-house cohort, prognostic cut-offs were set at the median for all investigative biomarkers, the median value was 2.0 for PDS5A, 0.01 for cyclin D3, and 5.0 for PRR11. In TCGA, the cut-off for PRR11 was set at the optimal value (1.71 FPKM). Kaplan–Meier analyses showed that both high PRR11 protein expression (Figure 4A) and high PRR11 mRNA expression in the curated TCGA data set (Figure 4B) were associated with a significantly shorter 5-year OS. Expression of PDS5A and cyclin D3 did not show any prognostic value in the study cohort (see supplementary material, Figure S1).