Four CRC patients and two healthy donors, admitted at the Shandong Cancer Hospital between september 2017 and July 2018, were selected for this study. Written informed consent was obtained from all participants. The clinical stage of the CRC patients was determined, and one patient was classified as stage I, one as stage II, and two patients as stage IV according to the 8 AJCC Cancer Staging Handbook of the American Joint Committee on Cancer. The protocol was approved by the Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences of committee. All subjects gave written informed consent in accordance with the Declaration of Helsinki.

CRC patients were not received any tumor treatment or suffered from any other immune, metabolic diseases or endocrine before serum collection, and all patients had clearly pathologically diagnosed as colorectal cancer, four of which were adenocarcinoma. Healthy donors do not have any underlying diseases, especially gastrointestinal diseases. Four CRC patients and two healthy donors were from 18 to 65 years old.

The exosomes were isolated by ultracentrifugation, as previously described [20]. Briefly, the serum was centrifuged at 10000 ×g for 30 min at 4°C to remove the cellular debris, and then ultracentrifuged at 100,000 ×g (Beckman Coulter, Brea, CA, United States) at 4°C for 2 h to precipitate the exosomes. Trizol was added to the precipitate, and the samples were stored at −80°C.

The exosome pellets were placed to the copper grids in a 50 µL drop of 1% glutaraldehyde for 5 min and transferred to a 100-µL drop of distilled water, after 2 min, the grids were stained with to a 50-µL drop of uranyl-oxalate solution (pH 7) for 5 min and placed a parafilm-covered glass dish covered anon ice. Subsequently, the exosome grids were cleaned using distilled water seven times for 2 min each. The TEM images were obtained using a JEM-1200EX transmission electron microscope (JEOL, Japan) operated at 100 kV.

The size of the vesicles and the particle concentration of exosomes were measured by using TRPS (qNano; Izon Science Ltd.). The tunable nanopores with proprietary data capture and analysis using Izon Control Suite software v.3.3.2.2000 (Izon Science Ltd.).

The extracted protein was resolved by 10% SDS-PAGE, transferred onto the PVDF membranes and placed in blocking buffer for 2 h, the membranes were incubated overnight with anti-CD63, anti-GM130, anti-TSG101 (CST, Danvers, United State) at 4°C, followed by incubation with HRP-coupled secondary Ab at room temperature for 1 h. The protein bands were detected through an ECL blot detection reagent.

The input material for the small RNA library was generated from 3 μg RNA in each sample. After the cluster was generated, the libraries were sequenced and 50 bp single-end reads were generated on the Illumina HiSeq 2500/2000 platform (Illumina, United States). After sequencing, coding potential analysis, quality control analysis, gene expression level quantification, transcriptome assembly, conservative analysis, target gene prediction, read mapping to the reference genome and differential expression analysis were carried out by Novogene Corporation.

Global analysis of human lncRNA and protein-coding transcripts was conducted using human lncRNA microarray V4.0. Briefly, after the rRNA was removed using the mRNA-ONLY™ eukaryotic mRNA isolation kit (Epicentre), the mRNA was purified from the total RNA. Then, a random primer method (Arraystar flash RNA labeling kit) was used to amplify each sample and transcribe it into fluorescent cRNA along the entire length of the transcript without 3’bias. RNeasy mini kit (Qiagen) was used to purify the labeled cRNAs. The concentration and specific activity of the labeled cRNAs were measured by NanoDrop ND-1000. The labeled cRNA was diluted using 50 μl of hybridization solution, distributed on the spacer slides, and placed on the lncRNA expression microarray slides. The slides were incubated at 65°C for 17 h in an Agilent hybridization oven. After washing, the slides were fixed and scanned using Agilent DNA Microarray Scanner (Part No. G2505C).

To identify potential pairs among exosomal DEmRNAs, DElncRNAs, and DEmiRNAs between the CRC patients and healthy donors, data from two databases were retrieved. Arraystar’s in-house miRNA target prediction software was applied based on TargetScan and miRanda algorithms. We intersected the predicted targets obtained from the database with miRNA sequencing and lncRNA, mRNA microarray. Fold-change of >1.5 and a p-value of <0.05 were set as the criteria.

Thereafter, the miRanda and TargetScan databases were used to predict potential miRNA targets, and Cytoscape software was used to delineate the ceRNA network of lncRNA-miRNA-mRNA.

The Gene Ontology (GO) project describes the attributes of genes and gene products in any organism (http://www.geneontology.org). The ontology includes three domains: Cellular Component, Molecular Function and Biological Process. The p-value indicates the importance of GO enrichment in the DE gene. The lower the p-value, the more significant is the GO term (p-value ≤0.05). The functional analysis of gene mapping to KEGG pathway is called pathway analysis. The p-value indicates the importance of disease-related pathways. Lower the p-value, more significant is the pathway. The recommended p-value cutoff is 0.05.

GraphPad Prism 6.0 (San Diego, CA, United States) and SPSS 22.0 (IBM, Ehningen, Germany) were used for the statistical analyses. Mann–Whitney U or t-test were used to perform the comparisons. Receiver operating characteristic (ROC) curves with area under the curve (AUC) were used to determine the corresponding cutoff values, with pathological diagnosis as the “gold standard”. A p-value <0.05 was considered to be statistically significant. In addition, each combination of miRNA and mRNA or mRNA was analyzed.

Exosomes isolated from CRC patients and healthy donors were characterized by qNano, TNM and western blotting analysis. The results suggested the typical exosome-like round morphology with 50–150 nm in diameter under TEM (Figure 1A), and the size range was in agreement with qNano (Figure 1B). Consistently, the characteristic protein markers of exosomes, CD63 and TSG101 were found in exosomes, but not in the whole cell lysates. whereas GM130 (the negative control) was not detected in the exosomal protein lysates (Figure 1C).

Exosomal RNAs from four CRC patients and two healthy donors were isolated for RNA sequencing and microarray. A total of 28189 downregulated genes (10808 mRNAs, 16840 lncRNAs, and 541 miRNAs) and 28364 upregulated genes (8428 mRNAs, 19332 lncRNAs, and 604 miRNAs) were screened in CRC patients compared to healthy donors, of which 1182 DE mRNAs and 2092 DE lncRNAs were selected with the thresholds of p-value <0.05 and fold-change >1.5 (Figure 2A). Exosomal miRNAs have been reported in our previous study [21].

The GO and KEGG enrichment analyses of these 1182 exosomal DE mRNAs were further conducted to investigate the possible signaling mechanisms of CRC. Interestingly, exosomal DE mRNAs were mainly enriched in negative regulation of ATPase activity in the biological process classification, and nucleic acid binding in the molecular function classification (Figure 2B). Additionally, KEGG pathway analysis showed that 207 pathways were significantly enriched. The top 10 significantly enriched pathways are shown in Figure 2C, including ECM-receptor interaction, adenosine monophosphate-activated protein kinase (AMPK) signaling pathway and mRNA surveillance pathway, which are closely correlated with the carcinogenesis and development of CRC. Therefore, these findings provided a resource for additional molecular participants in CRC.

To explore the relationship between exosomal DE lncRNAs and DE miRNAs in CRC, we predicted the target miRNAs of top 40 DE lncRNAs in the microarray expression profiles using TargetScan and miRanda databases. Finally, miRNA-150-5p and miRNA-10b-5p were selected since they could be targeted by the majority of exosomal DE lncRNAs (Figure 3A).

The two exosomal DE miRNAs were also analyzed using the GEO database, which includes exosomal RNA-sequence profiles from 100 CRC patients and 50 healthy donors (GSE71008). In the database, the exosomal miR-150-5p and miR-10b-5p expression levels were markedly downregulated in CRC patients (p < 0.0125 and p < 0.0014, respectively) compared to healthy donors (Figures 3B,C). In order to analyze the diagnostic performance of exosomal miR-150-5p and miR-10b-5p for CRC, a ROC curve was calculated. The AUC of miR-10b-5p was 0.625 (95% CI: 0.567–0.689), with specificity of 82% and sensitivity of 46%, while the AUC of miR-150-5p was 0.658 (95% CI: 0.571–0.745), with sensitivity of 56% and specificity of 58.8% (Figures 3D,E). Taken together, these data suggested that exosomal miR-150-5p and miR-10b-5p were significantly associated with CRC.

The study flowchart was shown in Figure 4A. To identify targeted DE mRNA pairs, we predicted targeted mRNAs from the TargetScan and miRanda databases, followed by intersecting the predicted mRNAs with the exosomal DE mRNAs in the microarray, and finally constructed 27 mRNAs targeted by miR-10b-5p and miR-150-5p in the CRC regulatory network (Figure 4B). Due to lower complexity and more reliability, five significantly upregulated mRNAs (TOMM70A, RBM48, BEND3, RHOBTB1 and ADAMTS2) in CRC patients (n = 623) compared to healthy donors (n = 51) were identified and analyzed using data from TCGA. As shown in Figures 4C–G, ROC curves were also calculated, and the AUCs of BEND3, ADAMTS2, TOMM70A, RBM48 and RHOBTB1 were 0.887, 0.695, 0.73, 0.66 and 0.692, respectively. The study found that BEND3, ADAMTS2, TOMM70A, RBM48 and RHOBTB1 were significantly up-regulated in CRC tissues compared to the precancerous in the TCGA database (Figure 4H). Notably, the AUC of the combination of these five mRNAs was 0.925 (95% CI: 0.899-0.951), with specificity and sensitivity of 90.2% and 82.8%, respectively (Figure 4I).

To elucidate the regulatory mechanism of CRC, an exosomal lncRNA-related ceRNA network of CRC was developed based on the above results. The ceRNA hypothesis states that exosomal lncRNAs harboring MREs can competitively bind to certain miRNAs, thus regulating miRNA-mediated downstream target gene silencing at the post-transcriptional level. We found that 40 exosomal DE lncRNAs might competitively bind to miR-10b-5p and miR-150-5p, and the latter might regulate 27 downstream targeted mRNAs (Figure 4B), which explained how mRNA expression was regulated by exosomal lncRNAs by combining with miRNAs. To confirm the possibility of indirect interactions between DE lncRNAs and DE mRNAs, a Pearson’s correlation analysis was performed with R2 ≥0.95 and p < 0.001 as criteria, which revealed a positive correlation between the expression levels of ten DE lncRNA-DE mRNA pairs (Figure 5A). Among which exosomal lncRNA G016261/RBM48 and lncRNA XLOC_011677/BEND3 were better matched than others in binding sites, thus lncRNA G016261/RBM48 and lncRNA XLOC_011677/BEND3 were selected (Tables 1, 2).

Finally, we predicted the exosomal ceRNA network in CRC. As shown in Figure 5B, lncRNA G016261 would function as ceRNA that interact with RBM48, serving as miRNA sponges to restrain miR-150-5p function, then inhibit its regulation of target gene RBM48 expression; Meanwhile, lncRNA XLOC_011677 may also act as ceRNA to compete with BEND3 transcripts for the miR-150-5p, thus mediating the interaction and regulation between miR-10b-5p and BEND3 (Figure 5C).