Revisiting the diagnostic performance of exosomes: harnessing the feasibility of combinatorial exosomal miRNA profiles for colorectal cancer diagnosis

Abstract

The challenges associated with liquid biopsy of colorectal cancer (CRC) are closely linked to the substantial variations observed in gene expression profiles among patients. This variability complicates the selection of an ideal biomarker for accurate diagnosis. In this report, we propose that employing a combination of miRNAs offers a better change for enhancing the accuracy of CRC diagnosis compared to solely relying on single miRNAs. As an illustrative example, we measured 9 miRNAs from 45 patient samples (comprising 31 CRC cases and 14 healthy controls) via RT-qPCR. We then utilized two methods: (1) LASSO regression for marker ranking and (2) linear discriminant analysis (LDA) to identify the optimal weighted combination of multiple markers. Our data indicates that combination of triple markers, selected based on their ranking, exhibited the highest diagnostic performance, including a sensitivity of 93.6% (95% confidence interval, CI 79.3–98.9%), specificity of 100% (CI 78.5–100.0%), positive predictive value (PPV) of 100%, negative predictive value (NPV) of 87.5%, and an overall accuracy of 95.6%. In contrast, the diagnostic performance of each individual miRNA used in the triple marker combination ranged from 53.3 to 80.0% in accuracy. While we acknowledge the need for further extensive studies involving larger patient cohorts and the consideration of additional miRNA candidates, our research undeniably highlights the potential of combining multiple markers as a robust methodology for identifying biomarkers among heterogeneous patient profiles.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-024-01481-4.

Introduction

Colorectal cancer (CRC) stands out as one of the most prevalent cancer types globally and continues to rank among the leading causes of cancer-related deaths worldwide. According to the report based on cases from the United States [1], the overall 5-year survival rate for colon cancer remains at approximately 65%. However, when detected early at the localized stage, the 5-year survival rate significantly climbs to 91%. Unfortunately, only 37% of patients receive early diagnoses. Unfortunately, a considerable 21% of cases are identified at the late stage, with cancer having already spread to distant parts of the body, resulting in a 5-year relative survival rate of only 14%. Notably, early-stage CRC typically manifests no symptoms, underscoring the critical importance of early detection through medical screening.

CRC diagnosis traditionally relies on methods such as the fecal occult blood test (FOBT) or colonoscopy. FOBT is employed to detect blood in the stool, often stemming from colon tumors that cause bleeding. The fecal immunochemical test (FIT) can also be performed for the same manner. The diagnostic performance of FOBT, however, is relatively low, with a reported sensitivity of 31% and specificity of 87% in a meta-analysis of six reported studies [2]. The effectiveness of FOBT also heavily relies on specimen conditions, such as stool rehydration, hemoglobin degradation, and the presence of interfering compounds [3]. Rehydrated samples can increase false-positive results by 8–16%, posing a challenge in the precise detection of cancer [3]. Furthermore, FOBT is not well-suited for early diagnosis as it does not reliably detect cancer precursors such as adenomas and serrated neoplasms [4]. On the other hand, colonoscopy not only enables the identification of precancerous lesions but also facilitates their removal, significantly lowering the risk of cancer development [5]. However, the effectiveness of colonoscopy heavily relies on the cleanliness of the colon, achieved through a carefully planned dietary regimen and meticulous bowel preparation. Failure to adhere to the prescribed bowel preparation regimen can significantly reduce the detection rate of polyps, making early treatment and prevention unattainable [6]. Ongoing research endeavors aim to develop more palatable bowel preparation agents, but these may still induce abdominal discomfort, bloating, vomiting, and nausea due to their distinct taste and odor [7, 8]. Furthermore, colonoscopy, as an invasive procedure, carries inherent risks, including intestinal perforation, bleeding, and post-examination abdominal pain [4]. Consequently, the demanding and intricate nature of bowel preparation for colonoscopy dissuades many patients from pursuing the procedure, potentially leading to exacerbated health complications.

Liquid biopsy refers to a diagnostic method developed as an alternative to tissue biopsy, involving the sampling and molecular analysis of biomarkers in biofluids such as blood, urine, and saliva. This approach is gaining prominence due to its minimally invasive nature, convenience for repeated testing, and suitability for long-term monitoring [9]. In current clinical practice, blood-based tumor markers like CA 19-9 and CEA are already being utilized for cancer diagnosis and prognosis prediction. The high C-reactive protein to albumin ratio (CAR) has emerged as novel a marker, not only for the diagnosis but also for the prognosis after surgery of CRC patients [10]. However, the diagnostic performance of these markers is not sufficiently high. For CEA, sensitivity ranges between 50 and 80%, with specificity and negative predictive value above 80%, and positive predictive values varied from 45.8 to 95.2% and showed low reliability [11]. Consequently, their role in diagnostics remains primarily supplementary, serving to complement rather than replace conventional methods like FOBT and colonoscopy. This indicates the current unmet need for the development of molecular signatures characterized by both high sensitivity and specificity, which is crucial for the successful translation of liquid biopsy from laboratory research to clinical application.

There are two issues related to the heterogeneous nature of tumors that raise challenges in identifying novel biomarkers. The first issue is the heterogeneity within solid tumors, which can potentially lead to misreading in tissue biopsies due to the small portion of tissue taken from a specific area of the tumor [12]. In addressing this concern, molecular analysis of extracellular vesicles, particularly exosomes, has emerged. Exosomes are nano-vesicles that contain a variety of molecular marker candidates, including proteins, mRNAs, and miRNAs, all of which provide insights into the status of the parent cells. By transporting these molecules, exosomes are recognized as playing essential roles in intercellular communications [13, 14]. These vesicles offer not only exceptional stability for dependable storage and analysis but also the capability of collectively monitoring multiple molecular signatures secreted from the entire tumors, which holds the potential to significantly enhance overall diagnostic accuracy [12, 15, 16]. However, there are still technical issues that require concerted efforts. A significant challenge still lies in the specific identification of disease-related exosomes among the multitude of vesicles originating from various sources in the bloodstream [17, 18]. Various isolation methods, including ultracentrifugation, size-exclusion chromatography, and polymer-based precipitation, have been popular. However, due to their low purity and yield, more advanced techniques have gained attention [19, 20]. Furthermore, owing to their small size (with diameters ranging from 50 to 200 nm), current molecular analysis methods like flow cytometry, western blotting, and PCR, which are optimized in single cell levels, still need further advancement [21]. The second issue related to heterogeneity is the variations between patients. This can be a main reason why the aforementioned current blood-based tumor markers in clinical practice still have low accuracy. In conjugation with these ongoing challenges, our primary goal in this study is to address the importance of simultaneous monitoring multiple miRNA markers as a CRC signature to enhance diagnostic accuracy (Fig. 1A). To achieve this, we have employed a statistical method, linear discriminant analysis (LDA), which is well-suited for identifying the weighting factors of each marker during their combination, maximizing the discriminative power between two independent groups, in this case, CRC versus healthy controls (Fig. 1B). We expect that the combination of exosomal miRNAs will offer greater diagnostic potential than individual markers. By integrating these miRNAs with advanced molecular profiling technologies, exosomal miRNAs hold promise for clinical diagnostic applications.

Fig. 1.

An illustration of a promising liquid biopsy using exosomal miRNA profiling for successful colorectal cancer diagnosis (A) and the workflow in this study for establishing an exosomal miRNA signature (B)

Results and discussion

Discovery of exosomal miRNAs candidates

miRNA expression profiles have been extensively investigated in clinical biofluids, revealing several miRNAs with the potential to serve as biomarkers in liquid biopsy [22]. Yet, in most studies, it remains challenging to identify the exact origin of these miRNAs, whether they are associated with exosomes, lipoproteins, or other compartments. By choosing exosomal miRNAs, we aimed to alleviate any uncertainty issues in our demonstration. After conducting an extensive literature review, we selected 9 exosomal miRNAs that had previously been reported as significant differentiators between CRC patients and normal controls [23–27] (detailed information on the 9 analyzed miRNAs are shown in Table S1). Next, the association of these 9 miRNAs and exosomes was investigated by measuring their expression levels using RT-qPCR after isolating exosomes from the blood plasma of a cohort of 31 CRC patients and 14 healthy controls (see clinicopathologic characteristics in Table 1).

Table 1.

Clinicopathologic characteristics of CRC patients and healthy controls

Variables	Healthy (n = 14)	CRC (n = 31)
Sex
M	8	22
F	6	9
Age (years)
< 55	3	2
55–70	4	15
> 70	7	14
Lymph node metastasis
Absent	–	5
Present	–	26
T stage
Tis	–	1
T1	–	0
T2	–	2
T3	–	18
T4	–	10
N stage
N0	–	6
N1	–	13
N2	–	12
TNM stage
0	–	1
I	–	1
II	–	2
III	–	6
IV	–	21
Grade of differentiation
Well	–	4
Moderate	–	19
Poor	–	7
Mucinous adenoca	–	1
KRAS mutation
Wild type	–	13
Mutation	–	7
NRAS mutation
Wild type	–	16
Mutation	–	0
BRAF mutation
Wild type	–	5
Mutation	–	3
P53
Negative	–	5
Positive	–	8
Elastic fiber
Negative	–	0
Positive	–	12
EGFR
Negative	–	1
Positive	–	13
EMVI
Negative	–	5
Positive	–	8

Figure 2A shows the size distribution of particles isolated from a representative CRC patient, with the majority falling within the 50 to 200 nm range, consistent with previous studies [16]. We also compared the particle concentrations between CRC patients and healthy controls, revealing no significant differences (Fig. 2B). In addition, we evaluated exosome concentrations by measuring the expression levels of two commonly used internal standards. While GAPDH mRNA showed no significant difference between the groups, miR-26a-5p expression was slightly elevated in CRC patients (Fig. 2C, D).

Fig. 2.

Characterization of plasma-derived exosomes from patients with colorectal cancer (CRC) and healthy controls. A Size distribution of particles isolated from a representative CRC patient. B Comparison of particle concentration between the CRC and healthy control groups, as analyzed by nanoparticle tracking analysis (NTA). C, D Expression levels of two internal standards, GAPDH mRNA and miR-26a-5p, in exosomes isolated from CRC patients and healthy controls

As shown in Fig. 3A, all 9 miRNAs were detected in exosomes from blood plasma, but only 2 miRNAs (miR-23a and miR-486) exhibited statistically significant differences between the CRC patients and healthy control groups. We assessed these 9 exosomal miRNAs using LASSO regression, which ranked them accordingly (Fig. 3B). Figure 3C displays the relative expression levels of the top 4 high-ranked miRNAs in CRC patients, normalized against the average expression levels in healthy controls. This reveals that most of these miRNAs were down-regulated in CRC patients, except for miR-320a. The receiver operating characteristic (ROC) curves shown in Fig. 3D indicate that the area under the curve (AUC) values for the two individual miRNAs (miR-23a and miR-486) among the top 4 high-ranked ones were found to be fair (0.8–0.9) but not outstanding (> 0.9) for precise CRC diagnosis.

Fig. 3.

Discovery of candidate miRNAs in extracellular vesicles for discriminating CRC patients from healthy controls. A The expression level of miRNAs in extracellular vesicles isolated from the blood plasma of CRC patients and healthy controls. B LASSO coefficient profiles of the 9 tested miRNAs. The top 4 high-ranked miRNAs for CRC diagnosis were highlighted. C Fold change in expression for the 4 selected miRNAs in CRC patients compared to the average expression in healthy controls. D ROC curve analysis of the 4 miRNA candidates

Combinations of miRNAs using linear discriminant analysis (LDA)

In line with recent studies suggesting the combination of miRNAs as a potential strategy for discovering improved diagnostic markers [28–37], we designed combinations of 2–4 miRNAs based on the ranking from LASSO regression (Table S2). As shown in Table 2, two methods, logistic regression analysis and LDA, have been employed to establish these novel marker combinations, both ensuring higher accuracy than individual markers in each study. While both methods are well-suited for evaluating marker combinations, we chose LDA because it maximizes the discriminative power between the two independent groups, CRC versus healthy controls. In our study, LDA provided a clear and interpretable decision boundary by identifying the weighting factors of each marker in their linear combination (Table S3). As shown in Fig. 4A, all the dual, triple, and quad marker signatures prepared by LDA exhibited high statistical significance between the CRC and healthy control groups (****P < 0.0001). In further ROC curve analysis, the triple marker (combination of miR-23a, miR-486, and miR-320a) displayed the highest AUC of 0.9885 (Fig. 4B). When compared to these miRNA combinations, the currently available clinical markers, CA19-9 and CEA, exhibited relatively poor diagnostic performance; while CEA performed better than CA19-9, it still differentiated between CRC and healthy controls with *P < 0.05 and AUC of 0.8133 (Fig. 4C, D).

Table 2.

Comparison of our study with literature [28–37]

miRNAs panel	Biofluids	Combination method	Diagnostic performance	References
miR-23a-3p, miR-486-5p, miR-320a-5p	Plasma exosome	Linear discriminant analysis	AUC 0.9885 (healthy vs CRC); sensitivity 93.55%; specificity 100%	This study
miR-1246, miR-1268b, miR-4648	Serum	Linear discriminant analysis	AUC 0.821 (healthy vs CRC)	[28]
miR-622, miR-362-5p, miR-486-5p	Tissue	Linear discriminant analysis	AUC 0.77 (differentiation of 2 types of microsatellite instability patients)	[29]
miR-17, miR-10a, let-7g, mir-152, miR-141		Linear discriminant analysis	Mean accuracy 0.8478	[30]
D1: miR-182, miR-183, miR-30a-5p, miR-378 D2: miR-147, miR-182*, miR-30a-3p D3: miR-137, miR-182, miR-224, miR-30a-3p	Tissue	Linear discriminant analysis	Overall sensitivity = 1.000; overall specificity = 1.000	[31]
miR-125b-2-3p, miR-933	Tissue	LASSO Cox regression	AUC 0.663 (prognostic accuracy for first-line chemotherapy response)	[32]
miR-223, miR-92a	Stool, plasma	Logistic regression	AUC 0.907 (healthy vs CRC); sensitivity 96.8%; specificity 75%	[33]
miR-1246, miR-202-3p, miR-21-3p, miR-1229-3p, miR-532-3p	Serum	Binary logistic regression	AUC 0.960 (healthy vs CRC); sensitivity 91.6%; specificity 91.7%	[34]
miR-30e-3p, miR-146a-5p, miR-148a-3p	Serum	Multiple logistic regression	AUC 0.883 (healthy vs CRC); sensitivity 80%; specificity 78.7%	[35]
miR-144-3p, miR-425-5p, miR-1260b	Serum	Logistic regression	AUC 0.954 (healthy vs CRC); sensitivity 93.8%; specificity 91.3%	[36]
miR-409-3p, miR-7, miR-93	Plasma	Logistic regression	AUC 0.866 (healthy vs CRC); sensitivity 91%; specificity 88%	[37]

Fig. 4.

Linear discriminant analysis for miRNA combinations. A Discriminant scores of 2 to 4 miRNA combinations. B ROC curve analysis of the 3 combinations. C Serum levels of the two clinical markers, CA19-9 and CEA, in tested CRC patients and healthy controls. ROC curve analysis of these clinical markers are shown in D. P values were calculated using the Mann–Whitney U test. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

We next compared the diagnostic values of these combined marker signatures with individual miRNAs and clinical markers (Table 3). The cut-off for each marker was determined using the Youden index method, where the sum of sensitivity and specificity is maximized. In the case of CA19-9, the clinical threshold typically exceeds 37 U/mL, but in our cohort, the determined cut-off was < 22.67 U/mL. This contrary trend could be attributed to either the small batch size or the inherent poor diagnostic performance. The cut-off values for CEA established in this study aligned with clinical criteria (from our cohort: > 3.270 ng/mL, clinical criteria: > 3–5 ng/mL). Alongside the highest AUC, the triple marker signature exhibited the highest diagnostic performance (sensitivity: 93.55%, specificity: 100%, PPV: 100%, NPV: 87.5%, accuracy: 95.56%) based on the cut-off established from our cohort.

Table 3.

Comparison of the diagnostic values of the individual 4 miRNAs, their combinations through LDA, and the 2 clinical markers

Marker	Cut-off	Sensitivity%	95% CI	Specificity%	95% CI	Youden index	PPV%	NPV%	Accuracy%
Single
miR-23a ①	< 0.9882	74.19	56.75% to 86.30%	85.71	60.06% to 97.46%	0.5990	92	60	77.78
miR-486 ②	< 0.4845	80.65	63.72% to 90.81%	78.57	52.41% to 92.43%	0.5922	89.29	64.71	80
miR-320a ③	> 0.6171	32.26	18.57% to 49.86%	100	78.47% to 100.0%	0.3226	100	40	53.33
miR-125b ④	< 0.007413	58.06	40.77% to 73.58%	85.71	60.06% to 97.46%	0.4377	90	48	66.67
Combination
① + ②	< 0.7098	80.65	63.72% to 90.81%	78.57	52.41% to 92.43%	0.5922	89.29	64.71	80
① + ② + ③	> − 0.1426	93.55	79.28% to 98.85%	100	78.47% to 100.0%	0.9355	100	87.5	95.56
① + ② + ③ + ④	> − 0.1018	87.1	71.15% to 94.87%	92.86	68.53% to 99.63%	0.7996	96.43	76.47	88.89
Clinical
CA19-9	< 22.67	56.67	39.20% to 72.62%	75	30.06% to 98.72%	0.3167	94.44	18.75	58.82
CEA	> 3.270	73.33	55.55% to 85.82%	100	56.55% to 100.0%	0.7333	100	38.46	77.14

The cut-off was obtained by Youden index

As a secondary validation, we conducted LDA of the triple markers sourced from The Cancer Genome Atlas (TCGA) database (Figure S1). As expected, the triple marker combination exhibited superior diagnostic performance compared to individual miRNAs. However, it is crucial to approach this data with caution due to the dataset’s unbalanced distribution, with only 8 healthy controls compared to 451 CRC patients.

Figure 5 presents the simultaneous assessment of the triple marker and CEA levels for accurate CRC diagnosis. Despite the small batch size, all 5 healthy controls tested negative according to both the triple marker and CEA criteria. Furthermore, 6 CRC patients who were falsely classified as negative based on the CEA criteria were identified as positive based on the triple marker. These imply that our triple marker can serve as a supplementary diagnostic indicator alongside the currently available markers in clinical practice.

Fig. 5.

Simultaneous evaluation of the established triple marker and CEA levels for precise CRC diagnosis (green dot: healthy, red dot: CRC in the bottom graph)

Target gene prediction and functional annotation

We further elucidated the biological functions associated with the three miRNAs selected for the triple marker combination. In the GO annotation analysis conducted using DAVID, 33 biological process terms, 5 cellular component terms, and 3 molecular function terms were significantly enriched (P < 0.05) (Fig. 6). Notably, the top 10 significantly enriched GO terms suggest a strong association between the three miRNAs and gene silencing, as well as the negative regulation of translation. Additionally, in the KEGG analysis, it was observed that one pathway, MicroRNAs in cancer, was linked to the single marker, miR-23a. The mapping of the predicted target genes for the three miRNAs is also shown in Figure S2.

Fig. 6.

GO analysis of miRNAs combined as a triple marker (miR-23a, miR-486, miR-320a). BP biological process, CC cellular component, MF molecular function. In the BP analysis, the top 10 GO terms are presented

Limitations of the triple marker in subclassification based on clinicopathologic characteristics

Finally, we applied the established triple marker signature to subclassify CRC patients based on various clinicopathological characteristics. As shown in Fig. 7A, the triple marker effectively discriminated between healthy controls with patients diagnosed with either stage I–III or stage IV, demonstrating high statistical significance. Although the signature value was higher among patients with stage IV compared to those in stage I–III, there was no statistically significant difference observed between patients in stage I–III and those in stage IV. Furthermore, the triple marker combination revealed greater deviations in the group of patients diagnosed with distant metastases (M1), but this difference was not statistically significant when compared to the group of patients diagnosed with no distant metastases (Mx + M0) (Fig. 7B).

Fig. 7.

Feasibility of the established triple marker in discriminating clinicopathologic characteristics of CRC patients. A Discriminant scores of healthy controls compared to patients in different stages (I-III vs IV). B–D Discriminant score comparisons of CRC patients with different histological types, the presence of distant metastases, and mutations occurrence. P values were calculated using the Mann–Whitney U test (for 2 groups) and Kruskal–Wallis test (for 3 groups). (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)

KRAS mutations are among the most dominant mutations in CRC, serving as a key deterministic carcinogenic factor, with approximately 40–52% of CRC cases carrying these mutations [38]. KRAS mutations in CRC are associated with abnormal activation of signaling pathways, promotion of angiogenesis, alternations in the tumor microenvironment, and metabolic enrichment, and their impact on prognosis and patient survival has also drawn significant attention [38]. Similarly, p53 mutations are frequently detected, occurring in about 40–50% of sporadic CRC cases, and are closely linked to disease progression and outcomes [39]. Therefore, we next subcategorized patients based on their KRAS and p53 mutation status to explore any potential meaningful correlations (Fig. 7C, D). However, out results showed no significant differences in the triple marker combination were observed among the subcategorized groups. These results align with the notion that genetic mutations do not significantly influence the expression levels of individual miRNAs (Figure S3).

The improved accuracy demonstrated through marker combinations in our study, along with many others, strongly supports the notion of employing multiple markers in liquid biopsy. However, the choice of marker candidates remains a subject of debate. In the studies shown in Table 2, there were some proposed miRNAs that were rejected in our study due to low significant expression between CRC and healthy patients, although some others were also included in our triple marker set. Alongside this, the expression profile of individual miRNAs in our study, specifically miR-23a, miR-486-5p, and miR-320a, both aligns with and contradicts other previous studies. For example, in our study, miR-23a was shown to be down-regulated in CRC compared to healthy controls, while a study by Karimi et al. reported significantly higher expression of miR-23a in CRC patients compared to healthy controls [40]. Similarly, A. Pisano et al. confirmed that miR-486-5p was up-regulated in the serum and stool of patients with low-stage, non-metastatic CRC (stage I–II) but was downgraded in a cancer stem cell model in vitro [41]. One noteworthy observation from these previous studies is that miRNA profiles can vary depending on biofluids. J. Dohmen et al. reported that miR-16, miR-23, and let-7, when collected from exosomes in serum, displayed distinctions between CRC and healthy controls, but these differences were not significant when collected directly from serum [42]. Therefore, advancements in the isolation and profiling techniques of biomarkers particularly those originating from exosomal compartments, depending on their biological origin, are essential for the reliable identification of biomarkers in successful clinical diagnostic applications.

In addition, our study has limitations, notably the insufficient number of patients and healthy individuals involved in the experiment. The sample size of both patients and healthy controls was relatively small, and there was an imbalance between the two groups. It is important to note that our study failed to detect significant differences in miRNA expression based on CRC progression. Although we applied the triple marker established to distinguish various clinicopathological characteristics such as KRAS mutations, our results did not yield meaningful insights. While the triple marker combination did reveal statistically significant differences in miRNA expression between these groups, it is crucial to acknowledge that we do not claim the marker combination optimized as the definitive choice for CRC diagnosis. Indeed, we only tested 9 exosomal miRNAs as a simple demonstration, indicating that there is still a chance we are overlooking clinically meaningful miRNAs. In expanded studies with larger group sizes and more miRNA candidates, different marker combinations may emerge. Utilizing meta-analysis could offer a better option for selecting miRNA candidates from the literature for an expanded RT-qPCR study in the future [43]. The validation of employing LDA in combining exosomal miRNAs in this study now paves the way for expanded studies.

Conclusions

In this study, we have successfully validated the feasibility of utilizing a combination of exosomal miRNAs for diagnosing CRC. Our methodology, involving the application of LASSO regression for marker ranking and LDA for searching proper combinations, resulted in the establishment of a triple marker with remarkable diagnostic performance compared to individual markers. While recognizing the imperative need to expand our investigation to encompass larger patient cohorts and a broader array of miRNA candidates in screening of clinically meaningful marker, our demonstration undeniably illustrates the substantial potential of utilizing marker combinations. Our methodology demonstrated in this study not only provide a complementary tool but also hold the promise of replacing current diagnostic tools.

Molecular markers detectable in biofluids, such as exosomal miRNAs, can serve not only in primary diagnosis but also in predicting prognosis and monitoring drug responses owing to their advantages in convenience for repeat testing. In the current standard treatment protocol for high-risk stage 2 and stage 3, or advanced CRC patients who have undergone radical surgery, adjuvant chemotherapy is typically administered. However, deciding which patients should undergo cancer treatment requires careful consideration due to the unpredictable toxicity associated with chemotherapy and the suboptimal clinical outcomes it may entail. Toxicities such as hair loss, neuropathy, nausea, and sleep disturbances are well-documented to have an adverse impact on the quality of life (QOL) of patients, both before and after surgery, which can subsequently influence the outcomes of future treatments. We anticipate that the analysis of exosomal miRNAs in biofluids has the potential to play a significant role in guiding the establishment of personalized treatment plans that are tailored to the unique characteristics of individual patients, thereby enhancing the overall effectiveness of treatment.

Methods

Informed consent

All methods were conducted in accordance with the Declaration of Helsinki and the ethics regulations of the Ethics Committee, with approval from the Institutional Review Board (IRB) of Kyungpook National University Chilgok Hospital (KNUCH) (approval code: KNUCH 2015-11-005-011). 10 mL of blood was collected in BD vacutainer® EDTA tubes (BD, USA) from 31 CRC patients and 14 healthy control individuals in KNUCH between 11/2018 and 9/2022. Informed consent was obtained from all participants/patients. Whole blood was centrifuged at 400 g for 15 min and the supernatants were stored at − 80 °C until use.

Exosome isolation

Exosomes were isolated from conditioned media or human plasma using the Total Exosome Isolation Reagents (Thermo Fisher Scientific, Inc.), following the manufacturer’s instructions. Briefly, conditioned media collected from cell culture were first centrifuged at 2000g for 30 min to remove debris. The supernatant was then mixed with the reagent at a 2:1 ratio and incubated at 4 °C overnight. After incubation, the exosomes were pelleted by centrifugation at 10,000×g for 1 h at 4 °C, redispersed in PBS, and stored at − 80 °C until further use. For human plasma, 500 µL of plasma, diluted with an equal volume of PBS, was centrifuged at 10,000g for 20 min. The supernatant was filtered through 0.22 µm syringe filters, mixed with 0.2 equivalent volumes of the reagent, and incubated at room temp for 10 min. The exosomes were then pelleted by centrifugation at 10,000×g for 5 min, redispersed in PBS, and stored at − 80 °C until use. The isolated particles in PBS were diluted 10 to 1000 times with PBS and analyzed using a Nanoparticle Tracking Analysis (NTA) System (Malvern Instruments, Nanosight LM10) equipped with a 642 nm laser and an sCMOS video camera.

Exosomal miRNA extraction

Exosomal miRNAs were extracted by Total Exosome RNA & Protein Isolation Kit (Thermo Fisher Scientific, Inc.) following the manufacturer’s protocols. Briefly, 200 µL of isolated exosomes dispersed in PBS was mixed with the same volume of 2 × denaturing solution of the kit and further processed to enrich small RNAs using contained glass-fiber filters in the kit. The concentration and quality of eluted RNAs were measured with Nanodrop spectrophotometer (Thermo Fisher Scientific, Inc.) and total small RNAs were stored at − 80 °C until use.

Reverse-transcription and RT-qPCR

cDNA synthesis and qPCR amplification were performed using the TaqMan Advanced miRNA cDNA Synthesis Kit and TaqMan® Advanced miRNA Assays (Thermo Fisher Scientific, Inc.) according to manufacturer’s instructions. Each assay includes small RNA-specific stem-looped RT primers and mixtures of small RNA-specific forward PCR primers, small RNA-specific reverse PCR primers, and small RNA-specific TaqMan™ MGB probes. Briefly, cDNA templates from 3.7 µL of miRNAs were prepared by PCR amplification after universal reverse transcription reaction and the modification with extending the 3ʹ end of the mature transcript through poly(A) addition and lengthening the 5ʹ end by adaptor ligation. 2.5 µL of pre-amplifed cDNA products was combined with 5 µL of Universal PCR Master Mix, 0.5 µL of predesigned miRNA Assay, and 2 µL of nuclease-free water in triplicates for each reaction. TaqMan® Advanced miRNA Assays were used as shown below for analysis of target markers. The hsa-miR-26a-5p assay was used as an internal control of reverse transcription and PCR function [44]. PCR results were analyzed using StepOne™ software, version 2.3. The relative miRNA expression values were calculated as a fold change to the expression level of the internal control, miR-26a-5p.

Statistical analysis

Least absolute shrinkage and selection operator (LASSO) regression analysis was utilized to determine the statistical significance in the differences of miRNAs between CRC and healthy controls. LDA was computed using the eigenvalue decomposition approach, optimizing class separation based on the eigenvalues. This involved calculating the within-class scatter matrix (W) and the between-class scatter matrix (B), and solving the generalized eigenvalue problem for the matrix W⁻¹B. Classification was then achieved by projecting the data onto the subspace defined by the eigenvectors corresponding to the largest eigenvalues. GraphPad Prism 10 (GraphPad Software, San Diego, USA) was used to analyzed and display the data. A receiver operating characteristic (ROC) curve analysis was performed, and the area under the ROC curve (AUC) was calculated to evaluate the diagnostic value. The optimal cut-off value, sensitivity, and specificity were determined by calculating the Youden index. Wilcoxon-Mann–Whitney test was used to determine the statistical significance of each marker between CRC and healthy controls (*P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001).

The Cancer Genome Atlas (TCGA) database

TCGA dataset was downloaded from the University of California Santa Cruz Xena Browser.

Enrichment analysis

GO analysis was performed using the Database for Annotation, Visualization, and Integrated Discovery (DAVID, http://david.ncifcrf.gov/) online tool. Target gene analysis of miRNAs were performed using the online miRTargetLink 2.0 (https://ccb-compute.cs.uni-saarland.de/mirtargetlink2/).

Author contributions

JSP1, JSP2, and SH designed the study. JSP1, DHH, and CB performed the experiments. SGK and SH performed the statistical analysis. JSP2 and SH supervised the project. JSP1, JAC, JSP2, and SH interpreted the data and made significant contributions to manuscript writing. All authors have reviewed and approved the final manuscript (JSP1: Jin Sung Park, JSP2: Jun Seok Park).

Funding

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Government of Korea (MSIT), the Engineering Research Center (ERC) program (NRF-2018R1A5A1025511; to S.H.). S.H. also gratefully acknowledges support by the DGIST R&D programs of the Ministry of Science and ICT (22-SENS2-01 and 24-SENS2-04). This research was also supported by a grant of the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number: HR22C1832; to J.S.P.).

Data availability

Data is provided within the manuscript or supplementary information files.

Declarations

Ethics approval and consent to participate

All methods were conducted in accordance with the Declaration of Helsinki and the ethics regulations of the Ethics Committee, with approval from the Institutional Review Board (IRB) of Kyungpook National University Chilgok Hospital (approval code: KNUCH 2015-11-005-011). 10 mL of blood was collected in BD vacutainer® EDTA tubes (BD, USA) from 31 CRC patients and 14 healthy control individuals in KNUCH between 11/2018 and 9/2022.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.