Differentially expressed genes associated with high metabolic tumor volume served as diagnostic markers and potential therapeutic targets for pancreatic cancer

Tumor tissues and their normal counterparts were obtained from 14 patients undergoing preoperative ¹⁸F-FDG PET/CT and potentially curative resection of pancreatic ductal adenocarcinoma at Severance Hospital of the Yonsei University Health System, Korea. The protocol for the study was approved by the Severance Hospital Ethics Committee (IRB number: 4-2017-0657) with a waiver of informed consent. ¹⁸F-FDG PET/CT imaging was performed as previously described [34]. Briefly, ¹⁸F-FDG PET/CT imaging was carried out with Discovery STE (GE Healthcare, Chicago, IL, USA) or Biograph TruePoint PET/CT scanner (SIEMENS Healthineers, Erlangen, Germany). The patients fasted for at least 6 h and were intravenously injected with ~ 5.5 MBq/kg of ¹⁸F-FDG for 60 min before imaging. CT scanning was performed at 30 mA and 130 kVp for the Discovery STE and 36 mA and 120 kVp for Biograph TruePoint without contrast enhancement. After the CT scanning was completed, PET scanning was extended from the neck to the proximal thighs for 3 min per bed in 3D mode. The images acquired from PET scanning were reconstructed with an attenuation correction using ordered subset expectation maximization (OSEM) [35]. The PET/CT images were reviewed by two board-certified nuclear medicine physicians with Advantage Workstation 4.4 (GE Healthcare, Chicago, IL, USA). Maximum standardized uptake value (SUVmax) and MTV2.5 were measured on PET images using Volume Viewer (GE Healthcare, Chicago, IL, USA). The volume of interest (VOI), including the entire lesion in the axial, sagittal, and coronal planes, was examined with a spherical shape. ¹⁸F-FDG uptake of normal organs such as bowel, stomach, and liver was excluded from the VOI. The patient groups with MTV-low and MTV-high were classified based on the MTV2.5 value of 4.5 (MTV-low: MTV2.5 < 4.5, MTV-high: MTV2.5 ≥ 4.5) [34].

Total RNAs were extracted from the tissue samples using TRizol™ Reagent (Invitrogen, Carlsbad, CA, USA), followed by rRNA removal using Lexogen RIBO COP rRNA depletion kit (Lexogen, Vienna, Austria). The RNA quality was analyzed in NanoDrop 2000 (Thermo Fisher Scientific, Waltham, MA, USA) and Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). The RNAs above 1.8 for OD260/280, 1.6 for OD260/230, and 7.0 for RNA integrity number (RIN) were used to construct RNA libraries using SMARTer^® Stranded RNA-Seq Kit (Clontech Laboratories, Inc. Mountain View, CA, USA). Sequencing was performed in Hiseq 2500 System (Illumina, San Diego, CA, USA), and its base calling was carried out by Illumina Casava 1.8 software. Sequenced reads were trimmed for adaptor sequence, masked for low-complexity or low-quality sequence using the FASTX trimmer tool, and then mapped to hg19 whole genome using TopHat. Read counts were extracted and normalized using edgeR.

The MTV RNA-Seq dataset was analyzed in the cancer tissue compared to its normal counterpart. To obtain significant DEGs between low and high MTV groups, we performed t-tests between the two groups and sorted the genes having below 0.05 p-value out using Excel-based Differentially Expressed Gene Analysis tool (ExDEGA, Ebiogen, Seoul, South Korea). The genes sorted out were classified into upregulated and downregulated genes in the high MTV group compared to the low MTV group and then analyzed in the Database for Annotation, Visualization and Integrated Discovery (DAVID) v2021 [36, 37]. Enrichment analysis was performed with lowest, low, medium, high, and highest stringencies. The biological processes with more than 1.3 of enrichment score, equivalent to a non-log scale of 0.05, were considered valid. The correlation of the DEGs with clinical factors was analyzed using the TCGA-PAAD, a large dataset without information about MTV. Genetic alteration analysis was performed in cBioPortal [38, 39] using the genetic alterations of the TCGA, PanCancer Atlas [40] sharing patient samples with the TCGA-PAAD. The putative miRNAs binding to the DEGs were analyzed using the miRNAs commonly predicted in miRWalk [41], TargetScan [42], and miRDB [43] databases.

Protein expression in normal and cancer tissues was evaluated using the human protein atlas [44]. The antibodies selected for immunohistochemical analysis were as follows, HPA008572 for ITGA3, HPA004941 for KIAA1217, HPA010568 for KIF11, HPA027573 for RCC1, HPA074007 for SLC44A1, HPA023535 for SON, CAB018364 for ADCY1, HPA028803 for INPP5B, HPA011392 for SDK1.

Cluster heatmap, bubble plot, balloon plot, Venn diagram, and enrichment bubble were drawn with the genes sorted from the transcriptome analysis using SRPLOT. Bar graph, patient survival curve, pie graph, and heatmap were visualized by GraphPad Prism 9.5.1 (GraphPad Software, La Jolla, CA, USA). The patient survival curve for the DEGs except for SNHG7 was plotted using the best expression cutoff of the Human Protein Atlas. The patient survival curve for SNHG7 was plotted using the z-score of the TCGA-PAAD. Correlation matrices were visualized using the conditional formatting function of Excel (Microsoft, Redmond, WA, USA).

Statistical significance was determined using Student’s t-test (two-tailed, standard deviation), Pearson’s correlation coefficient, and Log-rank test. Results were considered significant at p < 0.05. All statistical analyses were performed using GraphPad Prism 9.5.1 (GraphPad Software, La Jolla, CA, USA). Asterisks were used to indicate p values: one for p ≤ 0.05, two for p ≤ 0.01, and three for p ≤ 0.001.

Pancreatic cancer patient tissues were divided into MTV-low and MTV-high groups, each containing 7 cases, based on PET images presented in Fig. 1A. We followed a five-step process outlined in Additional file 1: Fig. S1 to identify potential markers and therapeutic targets. Initially, we conducted RNA sequencing (RNA-Seq) on both normal and cancerous sections of pancreatic cancer patient tissues. This allowed us to calculate the fold change in gene expression between the cancerous and normal counterparts. The clustered heatmap revealed that, compared to the MTV-low group, the MTV-high group exhibited general upregulation of 211 genes and downregulation of 166 genes (Fig. 1B). Among the 211 upregulated genes, 44 displayed a fold difference of more than 2 and statistical significance. Similarly, among the 166 downregulated genes, 56 exhibited a fold difference of more than 2 and statistical significance. The fold change distribution was centered around twofold for both upregulated and downregulated genes, although some downregulated genes demonstrated a fold difference of 10 or more (Fig. 1C). Functional annotation analysis revealed that the upregulated genes were enriched in processes related to mitosis (KW0498), cell cycle (KW-0131), and cell division (KW-0132), while the downregulated genes were enriched in ATP binding (GO:0005524) and the adenylate cyclase (AC)-activating G-protein coupled receptor(GPCR) signaling pathway (GO:0007189) (Fig. 1D). Notably, among the genes enriched in functional annotations, Kinesin Family Member 11 (KIF11), ATP Binding Cassette Subfamily A Member 13 (ABCA13), and Dicer 1, Ribonuclease III (DICER1) exhibited very significant and dramatic differences in expression between the MTV-low and MTV-high groups (Fig. 1E). Four genes (KIF11, Kinetochore Associated 1/KNTC1, Receptor Accessory Protein 4/REEP4, and Regulator Of Chromosome Condensation 1/RCC1) were commonly associated with cell division, mitosis, and cell cycle, while two genes (Adenylate Cyclase 1/ADCY1 and Adenylate Cyclase 5/ADCY5) were involved in ATP binding and adenylate cyclase(AC)-activating G-protein coupled receptor signaling pathway (Fig. 1F).

To facilitate convenience, the differentially expressed genes (DEGs) between the low and high MTV groups were designated as MTV-associated genes (MAGs), which were further divided into MTV-upregulated genes (MUGs) and MTV-downregulated genes (MDGs). To assess their potential as markers for pancreatic cancer, the expression levels of MUGs and MDGs were compared between the groups of living and deceased patients in the TCGA-PAAD dataset, a large dataset lacking MTV information. As shown in Fig. 2A, out of the 43 MUGs (excluding BST2 Interferon Stimulated Positive Regulator/BISPR, a gene not present in the dataset), 12 genes exhibited a significant positive correlation with patient death, 4 genes displayed a significant inverse correlation, and 27 genes showed no correlation. Among the MDGs, 17 genes demonstrated a significant negative correlation with patient death, 7 genes exhibited a significant inverse correlation, 27 genes showed no correlation, and 2 genes were not expressed. When comparing the living and deceased groups of pancreatic cancer patients, the deceased group displayed a significant upregulation of the MUGs (Cadherin EGF LAG Seven-Pass G-Type Receptor 1/CELSR1, CCR4-NOT Transcription Complex Subunit 1/CNOT1, DNA Cross-Link Repair 1B/DCLRE1B, Integrin Subunit Alpha 3/ITGA3, KIAA1217, KIF11, Membrane Bound O-Acyltransferase Domain Containing 2/MBOAT2, Regulator Of Chromosome Condensation 1/RCC1, Solute Carrier Family 44 Member 1/SLC44A1, SON DNA And RNA Binding Protein/SON, MTOR Associated Protein, Eak-7 Homolog /TLDC1/MEAK7, and ZFP69 Zinc Finger Protein /ZFP69) (Fig. 2B), while showing a significant downregulation of the MDGs (ADCY1, ADP Ribosylation Factor Like GTPase 6 Interacting Protein 4/ARL6IP4, ATPase Phospholipid Transporting 8A1/ATP8A1, Cytochrome C Oxidase Assembly Factor COX14/COX14, Erythropoietin Receptor/EPOR, Family With Sequence Similarity 110 Member D/FAM110D, Growth Arrest And DNA Damage Inducible Gamma/GADD45G, Hematopoietically Expressed Homeobox/HHEX, Inositol Polyphosphate-5-Phosphatase B/INPP5B, Potassium Inwardly Rectifying Channel Subfamily J Member 8/KCNJ8, Lin-7 Homolog B, Crumbs Cell Polarity Complex Component/LIN7B, Musashi RNA Binding Protein 1/MSI1, P21 (RAC1) Activated Kinase 3/PAK3, Retinol Binding Protein 5/RBP5, Sidekick Cell Adhesion Molecule 1/SDK1, Small Nucleolar RNA Host Gene 7/SNHG7, and Synaptogyrin 1/SYNGR1) (Fig. 2C). Additional file 2: Fig. S2 provided information on genes with no or inverse correlation. Groups with high expression of 10 MUGs exhibited significantly poorer patient survival than groups with low expression of MUGs (Fig. 2D). Similarly, groups with low expression of 16 MDGs showed significantly poorer patient survival than groups with high expression of MDGs (Fig. 2E). Furthermore, 10 MUGs and 16 MDGs displayed a significant correlation both in gene expression and patient survival (Fig. 2F). Among the 29 MAGs, 6 genes (ADCY1, ATP8A1, KCNJ8, KIF11, PAK3, and RCC1) were associated with functional annotations (Fig. 2G). Annotation-related genes other than the 6 genes did not show a significant correlation between expression and patient survival (Additional file 3: Fig. S3). In addition to differential gene expression, genetic alteration can affect the survival of patients with pancreatic cancer. Therefore, the genetic alterations of MAGs were analyzed. Of 175 patients in the TCGA-PAAD dataset, genetic alterations were found in 39 (22.3%) but not seen in 136 (77.7%) (Additional file 4: Fig. S4A). Except for MBOAT2, FAM110D, GADD45G, and ARL6IP4, amplification, deletion, and mutation (missense, splice, and truncating) occurred in MUGs and MDGs, but their frequencies were 3% or less (Additional file 4: Fig. S4B). Most of the patients with genetic alteration had less than 3 genetic alterations. However, a patient (TCGA-2J-AAB8) had 16 alterations (Additional file 4: Fig. S4C). Genetic alterations in MAGs were not associated with patient survival. There was no significant difference in patient survival between the group with alteration and non-alteration (Additional file 4: Fig. S4D).

To assess the potential of MAGs as markers for pancreatic cancer, we examined their correlation with TCGA-suggested markers, the top 20 genes associated with unfavorable and favorable prognoses based on TCGA-PAAD data analysis (The human protein atlas) [45]. As shown in Fig. 3A, most MUGs exhibited a positive correlation with unfavorable genes and a negative correlation with favorable genes. Conversely, most MDGs showed a negative correlation with unfavorable genes and a positive correlation with favorable genes. Additionally, we investigated the correlation between MAGs and CEA Cell Adhesion Molecule (CEACAMs), which are blood test markers for pancreatic cancer [46, 47]. Except for CELSR1 and SON, most MAGs correlated with CEACAM1, CEACAM3, CEACAM4, CEACAM5, CEACAM6, CEACAM20, or CEACAM21. CEACAM1, 5, and 6 notably exhibited a significant correlation with many MAGs (Fig. 3B, Additional file 5: Fig. S5A, B). Combinations of tumor markers can potentially improve cancer diagnosis and prognosis prediction compared to individual markers [48]. Therefore, we performed a correlation matrix analysis using MAGs from the MTV RNA-Seq and TCGA-PAAD datasets (deceased patient group) to identify promising combinations. In the MTV RNA-Seq, there was a clear positive correlation between MUGs or MDGs and a negative correlation between MUGs and MDGs. Similarly, the general correlation pattern in the TCGA-PAAD dataset resembled that of the MTV RNA-Seq, although there were some variations in individual correlations (Additional file 6: Fig. S6A). Significant correlations observed in the MTV RNA-Seq, TCGA-PAAD, or between the MTV RNA-Seq and TCGA-PAAD datasets were arranged as a half correlation matrix and further categorized into MTV RNA-Seq only, TCGA-PAAD only, and MTV ∩ PAAD (Additional file 6: Fig. S6B).

The presence of MAGs was observed to be correlated with patient death, as shown in Fig. 2. Therefore, we assumed that the expression of MAGs could potentially be linked to both tumor grade and clinical outcomes. In Fig. 4A, the patient group exhibiting high expression of MUGs demonstrated a greater prevalence of tumor grade (G) 2 and 3 than the group with low expression. Conversely, the patient group with low expression of MDGs exhibited a higher distribution of tumor grade 2 and 3 than the group with high expression. Tumors with higher grades displayed more upregulated MUGs and downregulated MDGs than those with lower grades. The overexpression of KIF11 and RCC1 and the underexpression of ADCY1 and SDK1 were detected in ~ 60% of tumor grade 2 pancreatic cancer patients. Treatment response was analyzed by being categorized into four groups: CR (complete regression), PR (partial regression), PD (partial disease), and SD (stable disease). In Fig. 4B, the expression of MUGs was mainly associated with CR and PD. The patient group with low MUG expression exhibited a higher distribution of CR than the group with high MUG expression, while the patient group with high MUG expression showed a higher distribution of PD than the group with low MUG expression. The expression of MDGs was also mainly associated with CR and PD. The patient group with high MDG expression displayed a higher distribution of CR than the group with low MDG expression, whereas the patient group with low MDG expression exhibited a higher distribution of PD than the group with high MDG expression. Favorable treatment responses (CR and PR) displayed more downregulated MUGs and upregulated MDGs. In contrast, unfavorable treatment responses (PD and SD) exhibited more upregulated MUGs and downregulated MDGs. Post-treatment outcomes were analyzed in DF (disease-free) and RP (recurred or progressed). As shown in Fig. 4C, the patient group with high MUG expression showed a higher distribution of RP and a lower distribution of DF, whereas the group with low MDG expression showed a higher distribution of RP and a lower distribution of DF. Among the MUGs, high expression of KIF11 and RCC1 was commonly correlated with high tumor grades (G2 and G3) and unfavorable clinical outcomes (treatment response-PD, SD, and post-treatment outcome-RP). Conversely, among the MDGs, low expression of ATP8A1 was commonly associated with tumor grades (G2 and G3) and unfavorable clinical outcomes (treatment response-PD, SD, and post-treatment outcome-RP) (Fig. 4D). To assess whether MUGs and MDGs can be used as early detection markers for pancreatic cancer, the mortality rates of stages I and II patients from the TCGA-PAAD dataset were analyzed. In Fig. 5A, ~ 94% of patients in the dataset were in stages I and II. The patients with high MUG expression (marked with a white asterisk) showed ~ 40–60% mortality (Fig. 5B). On the other hand, those with low MDG expression (marked with a white asterisk) showed around 50 to 88 percent mortality (Fig. 5C).

Potential therapeutic options for pancreatic cancer could be found in the MUG-binding miRNAs or inhibitors against the MDG-binding miRNAs (Fig. 6A). Therefore, we first identified the miRNAs that potentially bind to the MAGs using a common group of three databases (miRWalk, TargetScan, and miRDB). 25 MAGs were commonly analyzed in all the databases, but 4 MAGs (LIN7B, SNHG7, SON, and TLDC1) were not (Additional file 7: Fig. S7). Subsequently, we investigated the miRNAs showing a reverse expression correlation with the 25 MAGs in the TCGA-PAAD dataset. Of the 10 MUGs, KIAA1217, MBOAT2, and SLC44A1 showed significant and inverse correlations with 3, 11, and 7 miRNAs, respectively (Fig. 6B-a). In contrast, among the 15 MDGs, INPP5B, PAK3, SDK1, and SYNGR1 exhibited significant and inverse correlations with 2, 5, 3, and 2 miRNAs, respectively (Fig. 6B-b). In Fig. 6C, the patient population with high expression of KIAA1217 demonstrated significant downregulation of miRNA(MIR)130B, MIR301A, and MIR301B compared to those with low expression. The patient population with high expression of MBOAT2 showed significant downregulation of MIR27B, MIR29B-2, MIR30B, MIR32, MIR33B, MIR125A, MIR148B, MIR181D, MIR429, MIR642A, and MIR676 compared to those with low expression. The patient population with high expression of SLC44A1 displayed significant downregulation of MIR7-1, MIR7-2, MIR32, MIR301A, MIR1179, MIR1185-1, and MIR1301 compared to those with low expression. In Fig. 6D, the patient population with low expression of INPP5B demonstrated significant upregulation of MIR654 and MIR1185-1 compared to those with high expression. The patient population with low expression of PAK3 exhibited significant upregulation of MIR18A, MIR23A, MIR138-1, MIR193B, and MIR378A compared to those with high expression. The patient population with low expression of SDK1 showed significant upregulation of MIR7-2, MIR299, and MIR495 compared to those with high expression. The patient population with low expression of SYNGR1 displayed significant upregulation of MIR485 and MIR654 compared to those with high expression. To assess the risk of the therapeutic manipulation of corresponding miRNAs, we analyzed patient survival based on the low and high expression of the miRNAs. Patient survival was classified into three groups: MUG (16 miRNAs), MUG ∩ MDG (2 miRNAs), and MDG (9 miRNAs). As shown in Fig. 6E, the patient group with high expression of miRNAs targeting the MUGs did not show a significantly poorer survival than those with low expression. The expression of miRNAs commonly involved in MUG and MDG did not significantly affect patient survival. Except for one miRNA, MIR193B, the patient group with low expression of miRNAs targeting the MDG did not show a significantly poorer survival than those with high expression. The Patients with high expression of MIR193B displayed better survival than those with low expression. Therefore, the inhibition of MIR193B may not be helpful for patient survival.

The MAGs validated in the TCGA-PAAD dataset were protein-coding genes, except for SHNG7 (long non-coding RNA). Therefore, their protein expression was analyzed between pancreatic normal and cancer tissues using the human protein atlas. In Fig. 7A, KIF11, RCC1, KIAA1217, SLC44A1, ITGA3, and SON were overexpressed in cancer tissues compared to normal ones. Among those MUG proteins, KIF11 and RCC1 were the most validated markers according to the number of analyzed subjects. In Fig. 7B, compared to normal tissues, ATP8A1, ADCY1, INPP5B, and SDK1 were underexpressed in cancer tissues. Among those MDG proteins, ATP8A1 and INPP5B were the most validated markers according to the number of analyzed subjects. Like the correlation matrix analysis with MAG transcripts, the possible combinations of MAG proteins were presented. ITGA3 was combinable with ADCY1, ATP8A1, and SDK1. KIAA1217 was combinable with SLC44A1 and SON. KIF11 were combinable with RCC1, ADCY1, and SDK1. SLC44A1 was combinable with SON and SDK1. ADCY1 was combinable with ATP8A1 and SDK1. ATP8A1 was combinable with SDK1 (Fig. 7C). To evaluate the potential of the MAG combinations in the early detection of pancreatic cancer, patient mortality rates were analyzed using stages I and stage II patients from the TCGA-PAAD dataset. In Additional file 8: Fig. S8, the MUG x MUG combination showed ~ 60–70% mortality, the MDG × MUG combination showed ~ 66–70% mortality, and the MDG × MDG combination showed ~ 63–66% mortality.