Subgroup analysis reveals molecular heterogeneity and provides potential precise treatment for pancreatic cancers

Abstract

Background: The relationship between molecular heterogeneity and clinical features of pancreatic cancer remains unclear. In this study, pancreatic cancer was divided into different subgroups to explore its specific molecular characteristics and potential therapeutic targets.

Patients and methods: Expression profiling data were downloaded from The Cancer Genome Atlas database and standardized. Bioinformatics techniques such as unsupervised hierarchical clustering was used to explore the optimal molecular subgroups in pancreatic cancer. Clinical pathological features and pathways in each subgroup were also analyzed to find out the potential clinical applications and initial promotive mechanisms of pancreatic cancer.

Results: Pancreatic cancer was divided into three subgroups based on different gene expression features. Patients included in each subgroup had specific biological features and responded significantly different to chemotherapy.

Conclusion: Three distinct subgroups of pancreatic cancer were identified, which means that patients in each subgroup might benefit from targeted individual management.

Keywords: TCGA; bioinformatics; pancreatic cancer; therapeutic target.

Figure 1

Unsupervised hierarchical clustering analysis of pancreatic cancer samples. Notes: (A) Subgroups were distinguished with a red border. (B) Distribution of samples in the three subgroups. Subgroup 1, 2, and 3 clusters were marked with their respective numbers.

Figure 2

Characterization of subgroup information. Notes: (A) Sex distribution. (B) Smoking history. (C) Drinking history. (D) Pancreatitis history. (E) Other malignant tumor history. (F) Tumor grade. (G) Tumor diameter. (H) Treatment response. The y-axis for all graphs represents the percentage of each subgroup in all samples. There were relatively more men in subgroup 1, while the number of females was more in subgroup 2 and subgroup 3. Most patients in the three subgroups had not smoked for more than 2 years. There were fewer patients who drank for more than 5 years in subgroup 1 versus the other two groups, and most patients in the three subgroups had no history of pancreatitis. Most samples had no history of other malignant tumors. The proportion of patients with G3 and G4 tumors in subgroup 2 was greater than that in the other two subgroups, as was the proportion of patients with a tumor mass greater than 3 cm and PD patients after treatment, suggesting that samples in subgroup 2 tended to be more malignant. Abbreviations: CR, complete remission; PR, partial remission; SD, stable disease; PD, progressive disease.

Figure 2

Characterization of subgroup information. Notes: (A) Sex distribution. (B) Smoking history. (C) Drinking history. (D) Pancreatitis history. (E) Other malignant tumor history. (F) Tumor grade. (G) Tumor diameter. (H) Treatment response. The y-axis for all graphs represents the percentage of each subgroup in all samples. There were relatively more men in subgroup 1, while the number of females was more in subgroup 2 and subgroup 3. Most patients in the three subgroups had not smoked for more than 2 years. There were fewer patients who drank for more than 5 years in subgroup 1 versus the other two groups, and most patients in the three subgroups had no history of pancreatitis. Most samples had no history of other malignant tumors. The proportion of patients with G3 and G4 tumors in subgroup 2 was greater than that in the other two subgroups, as was the proportion of patients with a tumor mass greater than 3 cm and PD patients after treatment, suggesting that samples in subgroup 2 tended to be more malignant. Abbreviations: CR, complete remission; PR, partial remission; SD, stable disease; PD, progressive disease.

Figure 3

The chemotherapeutic agents used in these three subgroups. Notes: Usage percentages of 5-Fu were obviously higher in subgroup 1 (58.3%). Gemcitabine was more widely used in subgroup 2 (48.0%). In addition, oxaliplatin was another common drug used in subgroup 2 (60%), followed by subgroup 1 (30.0%) and subgroup 2 (10.0%).

Figure 4

Comparison among different subgroup pathway scores. Notes: The scores of 16 pathways were distributed from 0 to 1.6. Subgroup 1 was significantly abnormal, mainly in purine metabolism, Jak-signal transducer and activator of transcription (STAT), and gonadotropin-releasing hormone (GnRH) signal pathways. Subgroup 2 was significantly abnormal, mainly in aminoacyl-tRNA biosynthesis, glutamate metabolism, sphingomyelin biosynthesis, non-homologous end joining, and Jak-STAT signaling. Subgroup 3 was significantly abnormal, mainly in sphingomyelin biosynthesis, non-homologous end joining, adhesion spots, and other signal pathways.

Figure 5

The composite regulation network. Notes: Green nodes, mRNA; purple nodes, miRNA; brown nodes, lncRNA; red sides mean that the correlation coefficient was >0.8; orange sides mean that the correlation coefficient was >0.6; yellow sides mean that the correlation coefficient was >0.5. Some mRNAs were associated with multiple lncRNAs or miRNAs simultaneously, indicating that these genes were regulated by multiple complex effects and likely to be involved in important disease-related functions. Some mRNAs were correlated with each other, indicating that these mRNAs had a high synergistic effect and were involved in regulating downstream biological functions together. Abbreviations: miRNA, microRNA; lncRNA, long non-coding RNA.