Characterizing the role of exosomal miRNAs in metastasis

Abstract

Background: Exosomal microRNAs (exomiRs), transported via exosomes, play a pivotal role in intercellular communication. In cancer, exomiRs influence tumor progression by regulating key cellular processes such as proliferation, angiogenesis, and metastasis. Their role in mediating communication between cancer cells and the tumor microenvironment highlights their significance as potential diagnostic and therapeutic targets.

Methodology: In this study, we aimed to characterize the role of exomiRs in influencing the pre-metastatic niche (PMN). Across 7 tumor types, including 4 cell lines and three tumors, we extracted high confidence exomiRs (Log FC >= 2 in exosomes relative to control) and their targets (experimentally identified and targeted by at least 2 exomiRs). Subsequently, we identified enriched pathways and selected the top 100 high-confidence exomiR targets based on the frequency of their appearance in the enriched pathways. These top 100 targets were consistently used throughout the analysis.

Results: Cancer cell line and tumor derived ExomiRs have significantly higher GC content relative to genomic background. Pathway enriched among the top exomiR targets included general cancer-associated processes such as "wound healing" and "regulation of epithelial cell proliferation", as well as cancer-specific processes, such as "regulation of angiogenesis in kidney" (KIRC), "ossification" in lung (LUAD), and "positive regulation of cytokine production" in pancreatic cancer (PAAD). Similarly, 'Pathways in cancer' and 'MicroRNAs in cancer' ranked among the top 10 enriched KEGG pathways in all cancer types. ExomiR targets were not only enriched for cancer-specific tumor suppressor genes (TSG) but are also downregulated in pre-metastatic niche formed in lungs compared to normal lung. Motif analysis shows high similarity among motifs identified from exomiRs across cancer types. Our analysis recapitulates exomiRs associated with M2 macrophage differentiation and chemoresistance such as miR-21 and miR-222-3p, regulating signaling pathways such as PTEN/PI3/Akt, NF-κB, etc. Cox regression indicated that exomiR targets are significantly associated with overall survival of patients in TCGA. Lastly, a Support Vector Machine (SVM) model using exomiR target gene expression classified responders and non-responders to neoadjuvant chemotherapy with an AUROC of 0.96 (in LUAD), higher than other previously reported gene signatures.

Conclusion: Our study characterizes the pivotal role of exomiRs in shaping the PMN in diverse cancers, underscoring their diagnostic and therapeutic potential.

Keywords: Cancer metastasis; Exosomes; Machine Learning; Survival Analysis; miRNA.

Figure 1.. GC content and motif analysis.

(A) GC content analysis of exomiRs characterized from various cancer types and complete mature human miRNAs. (B) TOMTOM alignment results for the paired motifs between characterized from LIHC-LUAD exomiRs and (C) KIRC – ESCC exomiRs. (D) GC content analysis of 3’ UTR region of downregulated genes in TCGA-NATs compared to GTEx with upregulated genes.

Figure 2.. exomiR targets are enriched with essential biological processes.

Top 10 enriched biological processes associated with common exomiR targets (A); unique exomiR targets in BRCA (B); unique exomiR targets in COAD (C); unique exomiR targets in ESCA (D); unique exomiR targets in KIRC (E); unique exomiR targets in LIHC (F); unique exomiR targets in LUAD (G); and unique exomiR targets in PAAD (H).

Figure 3.. exomiR targets are enriched for Tumor Suppressor Genes.

Fisher’s exact test Odd’s ratio shows statistically significant enrichment of TSG among exomiR targets in various cancer types.

Figure 4.. exomiR targets are downregulated in the pre-metastatic niche compared to normal.

Violin plot showing median Log fold change of gene expression in PMN which formed in lung tissue, compared to normal in the mouse data with rhabdosarcoma. Median Log2FC value is printed on the plot.

Figure 5A:. exomiR targets are associated with overall survival.

Cox proportional hazard model shows that exomiR targets are more statistically significantly enriched for survival association compared to in general expectation in each cancer type. Log2FC (Observed/Expected) was computed where ‘Observed’ represents percentage of exomiR targets with negative HRs and p-value < 0.05, and ‘Expected’ represents percentage of any target (coding & non-coding) with negative HRs and p-value < 0.05.

Figure 5B.. Kaplan Meier Survival Analysis.

Kaplan Meier curve of the top genes selected from cox-regression analysis across cancer types. For each gene, patients were stratified into high and low category based on median expression and p-values were estimated using log-rank test.

Figure 6.. Performance of SVM model on independent dataset.

Average gene expression of various gene signatures was used as a feature to build SVM models. exomiR targets were able to discriminate responder and non-responder significantly with high AUROC compared to other signatures in all cancer types except COAD where CD8+ signature shows highest AUROC followed by exomiR targets.