The present study involved 68 patients with CRC (clinicopathological stages I–IV) and 27 healthy control (HC) individuals (15). The patients were admitted to the Department of Gastrointestinal Surgery, The Third Affiliated Hospital of Guangzhou Medical University (Guangzhou, China) between March 2017 and June 2019. The present study was prospective and was approved by Ethics Committee of the Third Affiliated Hospital of Guangzhou Medical University (Guangzhou, China; approval no. 2017.117), including the use of all datasets. Written informed consent was obtained from all participants.

Among the patients with CRC, 41 were males and 27 were females, with an age range of 38–77 years. The 27 HC individuals were selected from healthy subjects who visited the Third Affiliated Hospital of Guangzhou Medical University during the same period of time, of whom 12 were males and 15 were females, with an age range of 58–68 years. A total of 16 patients were enrolled in the screening period, 12 patients with CRC and 4 HC individuals; 79 patients were enrolled in the validation period, 56 patients with CRC and 27 HC individuals. Differences in sex, age and other general data between the two groups were not statistically significant (Table I).

The inclusion criteria were as follows: i) Age >18 years with no mental impairments; ii) histopathologically or cytologically confirmed CRC; and, for HC individuals, iii) no CRC or adenoma found by electronic colonoscopy. The exclusion criteria were as follows: Patients who i) exhibited ≥2 types of primary solid tumor; ii) had received chemoradiotherapy, surgery or systemic steroid therapy within the last 3 months; iv) were diagnosed with other serious diseases, or suffered from systemic active infection requiring treatment, immunodeficiency, autoimmune diseases and/or severe allergic diseases; and v) were pregnant or lactating women. All patients with CRC were diagnosed by histopathological analysis of surgically resected tissues or by electronic colonoscopy combined with histopathological biopsy. Pathological diagnosis and confirmation in tissue specimens were performed by ≥2 pathologists. Briefly, 5 ml fasting peripheral blood (collected after fasting for ≥8 h) was collected from patients with CRC and healthy subjects, and blood samples were processed within 2 h of collection. Immediately after blood collection, the samples were inverted and mixed well 5–8 times, before being placed at 25°C, followed by centrifugation at 3,500 × g for 10 min at 25°C within 2 h. The supernatants, i.e. plasma, were then carefully collected in cryovials and frozen at −80°C until further use.

Plasma exosomes were extracted using exoRNeasy Midi Kit (cat. no. 77144; Qiagen GmbH), according to the manufacturer's instructions. The plasma samples frozen at −80°C were thawed at room temperature and centrifuged at 3,000 × g for 15 min at 4°C. In total, 1 ml supernatant was collected and decanted into a new Eppendorf tube (Eppendorf SE), to which 400 µl buffer XE was added. The tube was then thoroughly mixed and incubated at 25°C for 1 min, followed by centrifugation at 3,000 × g for 5 min at 25°C. Next, the supernatant was discarded and the sample was again subjected to centrifugation at 3,000 × g for 5 min at 25°C. Finally, the remaining pellet was considered to be the exosome fraction. The obtained samples were then aliquoted and frozen for storage at −80°C.

Plasma exosomes were observed under transmission electron microscopy (TEM). Briefly, 5 µl exosome suspension was added to 200 mesh copper grids for 20 min at 25°C. Excess suspension was removed with filter paper and then fixed with 2% paraformaldehyde for 20 min at 25°C. The grids were then washed 10 times with PBS and fixed with 1% glutaraldehyde at 4°C for 2 h before being washed 10 times with water. The exosomes were negatively stained with 4% uranyl acetate for 10 min at 25°C, air dried and then visualized using TEM (JEM-1400PLUS; Hitachi, Ltd.) at 80 kV. Plasma EV nanoparticle tracking analysis (NTA) was performed to measure the exosome sizes in all samples. Protein concentration was determined using the Pierce™ BCA Protein Assay Kit (Thermo Fisher Scientific, Inc.) at a concentration range of 125–2,000 µg/ml.

Identification of protein expression in plasma exosomes was performed by western blot analysis. Isolated plasma exosomes were resuspended in RIPA buffer [0.05 M Tris-HCl (pH 8), 0.15 M NaCl, 1% Triton X-100, 0.1% sodium deoxycholate and 0.1% sodium dodecyl sulfate] with a protease inhibitor (Roche Diagnostics), and then lysed on ice for 15 min and boiled in 1X SDS sample buffer for 15 min. Protein samples (100 µg) were loaded, separated by SDS-PAGE on 12% gels and transferred to polyvinylidene difluoride membranes, followed by blocking with 5% non-fat milk in TBS-0.05% Tween-20 (TBST) for 1 h at 25°C. Subsequently, the membranes were incubated overnight with primary antibodies at 4°C, before being washed with TBST and incubated with secondary antibodies for 1 h at 25°C. The membranes were then washed again and visualized using enhanced chemiluminescence (ECL) reagent (Pierce ECL Western Blotting Substrate; Thermo Fisher Scientific, Inc.). The antibodies used for western blot analysis are listed in Table II.

Total RNA from exosomes was isolated using exoRNeasy Midi Kit (Qiagen GmbH) following the manufacturers' instructions. The resulting total RNA samples were used for reverse transcription-quantitative PCR (RT-qPCR).

The concentration and RNA integrity of the total RNA from plasma EVs were respectively determined using Qubit (Guangzhou IGE Biotechnology Ltd.) and Agilent 2200 TapeStation (Agilent Technologies, Inc.), respectively, prior to library construction. Small RNA libraries were prepared using the NEBNext Small RNA Library Prep Set for Illumina (New England BioLabs, Inc.) according to the manufacturer's protocol. Briefly, 3′ adapters were ligated to the input RNA followed by hybridization of reverse transcription primers and ligation of 5′ adapters. RNA was reverse transcribed and amplified by PCR (15 cycles). The PCR amplification products were purified by agarose gel electrophoresis and were then assessed using the Agilent 2200 for quality control. Next-generation sequencing (50 bp, single end) of the small RNA libraries was then performed on an Illumina HiSeq 2500 sequencer (Illumina, Inc.), and the raw data were subjected to a quality assessment to determine their suitability for bioinformatics analysis. Specifically, the quality analysis mainly included sequencing quality and base quality analyses. After Illumina HiSeq 2500 sequencing yielded raw reads, the raw reads were filtered: Adapters at both ends of reads were removed, reads with fragment lengths <17 nt and low-quality reads were removed, and preliminary filtering of the data was completed to obtain filtered clean data. This step was performed using Trimmomatic (Version 0.36; http://www.usadellab.org/cms/?page=trimmomatic). The filtered clean data were then aligned and annotated to known non-coding RNA (ncRNA) databases: miRBase version 22 was used as the miRNA database (16), while Rfam12.1 (17) was the database used for transfer RNA, ribosomal RNA (rRNA), small nuclear RNA and small nucleolar RNA, and piRNABank (18) was the database used for P-element induced wimpy testis in Drosophila-interacting RNAs. The annotation results were then statistically compared and graphically presented. Calculations of expression levels were performed on the identified miRNAs. Based on all miRNA expression data, correlation analysis between samples was performed on the sequenced samples. Pearson's correlation coefficient (r) was used to measure the correlation between two variables. The closer the absolute r-value is to 1, the higher the correlation in expression levels between the two samples. A positive r-value was considered to represent positive correlation in the expression patterns between two samples, whilst a negative r-value was considered to represent negative correlation between two samples. Through principal component analysis, the data were analyzed to evaluate the association among the samples, whilst assessing if a specific role was mediated by samples found in different groups. Differential miRNA expression analysis was performed using the ‘edgeR’ R language package (19).

Exosomal RNA was extracted using the miRNeasy advanced kit (Qiagen GmbH) and miRNA levels in plasma EVs were quantified using RT-qPCR. The poly-(A)-tailing method was used for first-strand cDNA synthesis using a miRcute Plus miRNA First-Strand cDNA kit (cat. no. KR211; Tiangen Biotech Co., Ltd.) according to the manufacturer's instruction. RT-qPCR for miRNA quantification was performed using a miRcute Plus miRNA qPCR kit (SYBR Green) (cat. no. FP411; Tiangen Biotech Co., Ltd.) following the manufacturer's protocol. The qPCR thermocycling conditions were as follows: Pre-denaturation at 95°C for 15 min, followed by five cycles at 94°C for 20 sec, 65°C for 30 sec and 72°C for 34 sec for the enrichment of low-abundance target miRNAs, and 40 cycles at 94°C for 20 sec and 60°C for 34 sec for amplification of target miRNAs. Melting curve analysis was performed by heating samples from 65 to 97°C at a ramp rate of 0.11°C/sec. The sequences of the primers used in qPCR are shown in Table III. The quantitative cycle (Cq) value was used to perform the relative quantification of miRNA expression as follows: ΔCq=Cq (candidate miRNA)-Cq (reference miRNA). Subsequently, the 2^−ΔΔCq method was used to describe the differences in the relative expression levels of miRNA among the different groups (20). The data in the HC group were then normalized to 1, before fold-change (FC) was used to describe the fold-difference in the relative expression of miRNAs in the CRC and HC groups as follows: FC=2^−ΔΔCq=2^−ΔCq (reference group)/2^−ΔCq (analyzed group). miR-16-5p was used as an internal reference miRNA (21). The expression stability of miR-16-5p in plasma EV-RNA samples was confirmed by analyzing the relative expression of miR-16-5p in each sample. Namely, miR-16-5p was consistently expressed in each sample.

GraphPad Prism7 software (GraphPad; Dotmatics) and SPSS 25.0 (IBM Corp) were used for statistical analysis and mapping of relative miRNA expression. Preliminary bar graphs were prepared using Excel 2019 (Microsoft Corporation). Measurement data are expressed as the mean ± SD. Independent samples Student's t-test was used for the statistical analysis of two-group unpaired data with equal variance. Corrected Welch's t-test would be used if the variance was found to be unequal. Logistic regression analysis was used for risk factor analysis, where odds ratio (OR) was obtained. OR >1 would be considered to indicate that a factor is a risk factor for the tested outcome events, whereas OR <1 would be considered to indicate that a factor is a protective factor against the tested outcome events. Receiver operating characteristic (ROC) curves were used to assess the diagnostic potential of miRNAs for CRC. Youden's index maximum value was used to determine the sensitivity and specificity corresponding to the cut-off value of the CRC curve as follows: Youden index=(Sensitivity + specificity)-1. A miRNA combination model was developed using multivariate logistic regression analysis. The diagnostic potential of miRNA combinations for CRC was assessed in combination with the ROC curves. P<0.05 was considered to indicate a statistically significant difference.

Plasma exosomes were extracted from patients with CRC using the aforementioned exoRNeasy Midi Kit and were observed using TEM. The particles were evenly distributed and displayed several micromembrane-coated vesicular structures, with a single distribution and a double-layered membrane structure, suggesting that exosomes were successfully obtained (Fig. 1A and B).

The particle size distribution and concentration of EV sample particles were determined using NTA. The results revealed that 97.6% of EV particles were distributed around 163.9 nm, with an average EV particle size of 168 nm (Fig. 1C). EV marker proteins were next assessed using western blotting, revealing that common marker proteins of EVs, namely CD63, CD81, TSG101 and Alix (22), were present on EV membranes (Fig. 1D). In addition, tumor susceptibility gene 101 and Alix proteins, which are commonly found EV markers (22), were found to be expressed (Fig. 1D). These results suggested that plasma exosomes were successfully extracted.

In total, 16 plasma samples, collected from 12 patients with CRC and 4 HC individuals, were subjected to total RNA extraction from the isolated EVs, followed by high-throughput sRNA sequencing. The 16 samples were grouped according to the clinicopathological characteristics of the participants (Table IV). As shown in Table V, all samples had >11 million raw reads, >10 million filtered clean reads and Q30 of raw reads >85% for all samples, suggesting that the raw and filtered data obtained by sequencing were of sufficient quality for subsequent bioinformatics analysis (16). The clean reads obtained by sequencing were analyzed using miRBase. Alignment with all human miRNA mature sequences in the miRBase version 22 database (12) provided information on the structure, length and expression of the miRNAs. In total, 458–698 mature miRNAs were aligned in each of the 16 samples, as shown in Table VI.

Next, differential miRNA expression analysis was performed using the ‘edgeR’ software package using the following two criteria: |log₂ (FC)|>1 and P<0.05. By using this method, miRNAs that were differentially expressed among the different groups were found. Grouping was performed based on the clinicopathological parameters of the 16 sequenced samples, and multiple pairwise comparisons were performed to obtain miRNAs that were differentially expressed among different groups and had a mean read per million (RPM) value of >100 in 16 samples (Table VII). The mean RPM expression values of the aberrantly expressed miRNAs in the 16 samples was log₂-transformed, before the relative mean expression value of the miRNAs of interest in the 16 plasma EV-RNA samples was plotted as a bar chart (Fig. 2). Since the plasma EV-miRNA abundance is of particular importance for the quantitative detection of miRNAs, we combined the two factors: miRNA expression difference results between groups and expression abundance, to select five miRNAs with significant differential expression between groups from the differential analysis results, and all five miRNAs were expressed in high abundance in samples for subsequent clinical plasma sample validation. Five miRNAs, namely miR-122-5p, miR-409-3p, miR-99b-5p, miR-625-3p and miR-143-3p, were selected for validation, since they exhibited significant differential expression and high expression levels in the analyzed samples (Fig. 2).

The expression levels of the aforementioned five candidate miRNAs (miR-122-5p, miR-409-3p, miR-99b-5p, miR-625-3p and miR-143-5p) in the 79 plasma EV samples from patients with CRC and healthy individuals were determined using RT-qPCR. There were 56 samples in the CRC group and 23 samples in the HC group. The FC in miRNA expression levels in the CRC group compared with the HC group was then calculated. As shown in Fig. 3A and B, miR-99b-5p and miR-409-3p were respectively found to be significantly increased in the CRC group compared with the HC group (P<0.05). Differences in expression were also observed between the CRC and HC groups for miR-122-5p, miR-143-3p and miR-625-3p (Fig. 3C-E), but without statistical significance. These results suggest that miR-99b-5p may be a useful marker for diagnosis of CRC.

To further evaluate the significance of miRNAs expression profiles, the miRNA expression levels in CRC samples of patients with different pathological features were compared. As shown in Table VIII, the expression levels of miR-99b-5p were found to be different among different clinical stages and lesion sites of CRC. As shown in Fig. 4, among the various clinical stages of CRC, the expression levels of miR-99b-5p were significantly higher during early CRC compared with advanced CRC (P=0.03). And the expression levels of miR-409-3p were higher during early CRC compared with in the HC group (P<0.05; Fig. 5). Furthermore, exosomal miR-99b-5p expression levels were significantly higher in patients with CRC originating from the colon compared with those in CRC originating from the rectum (P=0.028; Fig. 6A), which suggests that exosomal miR-99b-5p may be an indicator of early CRC pathogenesis or progression. CRC located in the proximal colon (right) and distal colon (left) were previously found to exhibit different molecular characteristics and histology, suggesting that the pathogenesis of CRC depends on tumor location (23). In the present study, exosomal miR-625-3p expression levels were significantly higher in right-sided CRC compared left-sided CRC (P=0.007; Fig. 6B), suggesting that exosomal miR-625-3p may also serve as a biomarker for CRC pathogenesis. Using logistic regression analysis for assessing the potential risk factors of early CRC, it was found that high miR-99b-5p expression may be a risk factor of early CRC, with an odds ratio of 2.725 and a 95% confidence interval (CI) of 1.236–6.008 (P=0.013) (data not shown).

ROC curves were constructed for the differentially expressed miRNAs in plasma EVs to assess their diagnostic potential for CRC. Specifically, 95% CI and area under the curve (AUC) values were obtained. The maximum Youden index value was used to determine the sensitivity and specificity corresponding to the cut-off value of the respective ROC curves. ROC curve analysis was first performed individually for the differentially expressed miRNAs in the CRC group. As shown in Fig. 7A, miR-99b-5p yielded an AUC of 0.660 (95% CI, 0.527–0.793; sensitivity, 55.4%; specificity, 78.3%; P=0.026). This result suggested that miR-99b-5p had potential for the diagnosis of CRC in HC individuals, and its diagnostic specificity could reach 78.3%. However, its diagnostic sensitivity was only 55.4%.

On the other hand, miR-409-3p yielded an AUC of 0.604 (95% CI, 0.471–0.737; sensitivity, 58.9%; specificity, 80.2%; P=0.148), suggesting that this miRNA had limited diagnostic ability for CRC (Fig. 7B). ROC curve analysis of the differentially expressed miR-99b-5p in the early CRC group revealed that it had a higher differential diagnostic potential for early CRC, where its AUC reached 0.735, and its sensitivity and specificity both reached >70% (Fig. 7C). Specifically, miR-99b-5p yielded an AUC of 0.735 (95% CI, 0.587–0.884; sensitivity, 72.7%; specificity, 78.3%; P=0.007; Fig. 7C). ROC analysis of miR-409-3p for early CRC revealed an AUC of 0.611 (95% CI, 0.443–0.778; sensitivity, 63.6%; specificity, 60.9%; P=0.204; Fig. 7D). The performance of CEA, which is used clinically for the diagnosis of early CRC, yielded an AUC of 0.820 (95% CI, 0.695–0.945; sensitivity, 72.7%; specificity, 87%; P<0.001; Fig. 7E).

To explore whether multiple miRNAs in combination could improve the diagnostic potential for CRC, a regression analysis model of various miRNAs was developed. The diagnostic potential of these miRNAs was then analyzed using ROC curves, which revealed that the diagnostic performance for CRC of multiple miRNAs in combination was superior compared with that of individual miRNAs. In particular, the combination of miR-99b-5p and miR-409-3p (Fig. 8A) was evaluated for the diagnosis of early CRC, yielding an AUC of 0.673 (95% CI, 0.542–0.804; sensitivity, 66.1%; specificity, 73.9%; P=0.016). In addition, the combination of CEA and miR-99b-5p for the diagnosis of early CRC was found to exert superior diagnostic efficacy, yielding an AUC of 0.863, (95% CI., 0.774–0.952; sensitivity, 83.6%; specificity, 82.6%; P<0.001; Fig. 8B). When the miR-99b-5p and miR-409-3p combination was applied for the diagnosis of early CRC, an AUC of 0.741 was found (95% CI, 0.592–0.891; sensitivity, 77.3%; specificity, 78.3%; P=0.006; Fig. 9A), while the combination of miR-99b-5p, miR-409-3p and CEA yielded an AUC of 0.812 (95% CI, 0.679–0.945; sensitivity, 90.9%; specificity, 65.2%; P<0.001; Fig. 9B). The potential miRNA markers evaluated for the diagnosis of early CRC are summarized in Table IX. The diagnostic sensitivity of miR-99b-5p for early CRC achieved a sensitivity consistent with that of CEA. Furthermore, the combination of miR-99b-5p and miR-409-3p yielded a higher diagnostic sensitivity for early CRC compared with that of CEA. However. the combination of miR-99b-5p, miR-409-3p and CEA had 90.9% sensitivity in terms of diagnostic power.