The construction and analysis of tumor-infiltrating immune cell and ceRNA networks in recurrent soft tissue sarcoma

Abstract

Soft tissue sarcoma (STS) is one of the most challenging tumors for medical oncologists, with a high rate of recurrence after initial resection. In this study, a recurrent STS-specific competitive endogenous RNA (ceRNA) network including seven recurrence and overall survival (OS)-associated genes (LPP-AS2, MUC1, GAB2, hsa-let-7i-5p, hsa-let-7f-5p, hsa-miR-101-3p and hsa-miR-1226-3p) was established based on the gene expression profiling of 259 primary sarcomas and 3 local recurrence samples from the TCGA database. The algorithm "cell type identification by estimating relative subsets of RNA transcripts (CIBERSORT)" was applied to estimate the fraction of immune cells in sarcomas. Based on 5 recurrence and OS-associated immune cells (NK cells activated, dendritic cells resting, mast cells resting, mast cells activated and macrophages M1), we constructed a recurrent STS-specific immune cells network. Both nomograms were identified to have good reliabilities (Area Under Curve (AUC) of 5-year survival is 0.724 and 0.773, respectively). Then the co-expression analysis was performed to identify the potential regulation network among recurrent STS-specific immune cells and ceRNAs. Hsa-miR-1226-3p and MUC1 were significantly correlated and dendritic cells resting was related to hsa-miR-1226-3p. Additionally, the expression of MUC1 and dendritic cell marker CD11c were also verified by immunohistochemistry (IHC) assay and multidimensional databases. In conclusion, this study illustrated the potential mechanism of hsa-miR-1226-3p regulating MUC1 and dendritic cells resting might play an important role in STS recurrence. These findings might provide potential prognostic biomarkers and therapeutic targets for recurrent STS.

Keywords: bone tumor; ceRNA; immune cell; prognosis; recurrence; soft tissue sarcoma.

Figure 1

The flow chart of the analysis process. Abbreviations: TCGA: The Cancer Genome Atlas; STS: Soft tissue sarcoma; GEO: Gene Expression Omnibus; CCLE: Cancer Cell Line Encyclopedia; GTEx: Genotype-Tissue Expression; UCSC: University of California, Santa Cruz.

Figure 2

The differentially expressed genes between primary and recurrent STSs. (A) The heatmap and the volcano plot (B) of 178 differentially expressed genes between 259 primary and 3 recurrent STSs; (C) The volcano plot of 148 differentially expressed protein-coding genes between 259 primary and 3 recurrent STSs; The volcano Plot (D) of 21 differentially lncRNAs between 259 primary and 3 recurrent STSs; (E) The composition of differentially expressed genes. The log(fold-change) > 1.0 or < -1.0 and FDR < 0.05. Abbreviations: ceRNAs: competing endogenous RNAs; STSs: soft tissue sarcomas; LncRNA: long non-coding RNA.

Figure 3

(A) The STS-recurrence related ceRNA network; The Kaplan-Meier survival curves of LPP-AS2 (B), MUC1 (C), hsa-let-7i-5p (D), hsa-let-7f-5p (E), hsa-miR-101-3p (F) and hsa-miR-1226-3p (G). Abbreviations: STSs: soft tissue sarcomas; ceRNAs: competing endogenous RNAs

Figure 4

The results of the multivariate Cox regression, nomogram (E) and model diagnosis process (B, C, D, F) based on the key members in the ceRNA network. Seven potential prognosis-related ceRNAs were integrated into a new multivariable model. The results of the Lasso regression suggested that all seven genes were essential for modeling (A, B). The nomogram was constructed based on the model (D). The ROC and the calibration curves indicated acceptable accuracy (Area Under Curve (AUC) of 3-year survival: 0.731; AUC of 5-year survival: 0.724) and discrimination of the nomogram (C, E).

Figure 5

The composition (A) and heatmap (B) of immune cells estimated by CIBERSORT algorithm in sarcomas. (C) The violin plot of immune cells (The blue and red bar stand for recurrent tumor group and primary tumor group, respectively). Abbreviations: CIBERSORT: Cell type identification by estimating relative subsets of RNA transcripts.

Figure 6

The results of the multivariate Cox regression, Lasso regression (A, B), Kaplan–Meier survival curve of (D), nomogram (E) and model diagnosis process (C, F) based on prognosis related immune cells. All immune cells were integrated into an initial Cox regression model. After the screening process of the Lasso regression, the results suggested that the model was not overfitting (A, B). The nomogram based on the multivariable model (E). The calibration curve and the ROC demonstrated good discrimination and concordance of the nomogram (AUC of 3-year survival: 0.709; AUC of 5-year survival: 0.773) (C, F).

Figure 7

The co-expression patterns among fractions of immune cells and key members in the ceRNA network. (A) co-expression heatmap of all immune cells; (B) co-expression heatmap of prognostic immune cells and key members of ceRNA network; (C) has-let-7i-5p was significantly associated with dendritic cells resting (R = 0.200, P = 0.003); (D) hsa-miR-1226-3p was significantly associated with dendritic cells resting (R = -0.190, P = 0.004).

Figure 8

The expressions of MUC1 and CD11c proteins in primary/ recurrent leiomyosarcoma (LMS) (A, B) and liposarcoma (LPS) (C, D) specimens examined by immunohistochemistry (IHC) assay. The upper one is primary STS and under one is recurrent STS.