Mesenchymal Colorectal Cancers Secrete Vesicles With Unique Cargo That Can Be Used for Liquid Biopsy Based Diagnostics

ABSTRACT

Tumour‐derived extracellular vesicles (TEVs) play a crucial role in cancer progression, metastasis and therapy resistance but their distinct profiles across different cancer stages and molecular subtypes remain underexplored. This study initially analysed TEVs from all CMS subtypes in colorectal cancer (CRC) cells and continued focusing on the epithelial (CMS2) and mesenchymal (CMS4) subtypes using six cell lines and clinical samples. Investigation of the cargo of vesicles secreted by the two subtypes revealed significant differences in mRNA, miRNA, and protein profiles between the two subtypes. Notably, CMS2 predominantly secreted smaller, Tetraspanin‐8 (TSPAN8) enriched EVs, while CMS4 produced both larger and smaller EVs, enriched in TSPAN4. This underscores the complexity of vesicle heterogeneity between these subtypes. Additionally, we assessed miRNA profiles from plasma‐derived bulk TEVs in CRC patients. Our integrative analysis identified a subtype‐specific miRNA signature, indicating that TEVs from CMS2 and CMS4 cells can be detected in circulation and may serve as potential diagnostic tool for CRC.

1. Introduction

Emerging evidence has shown that tumour cells produce and secrete tumour‐derived extracellular vesicles (TEVs). Although TEVs range in diameter from 30 to 1000 nm, exosome sizes typically range between 40 and 160 nm in diameter (Kalluri and LeBleu 2020; Xu et al. 2018; Brennan et al. 2020). They carry DNA, RNA, lipids, protein and microRNA (miRNA) cargo and play a critical role in intercellular communication, influencing several cellular processes (Kalluri and LeBleu 2020; Xu et al. 2018). Notably, cancer cells secrete more extracellular vesicles (EVs) than surrounding normal cells, which can be modulated by both cell‐intrinsic (e.g., oncogenic signalling pathways) and environmental signals like hypoxia (Bebelman et al. 2018). TEVs have been shown to regulate the dynamic multistep progress of cancer by inducing autocrine/paracrine oncogenesis, which subsequently promotes cancer proliferation (Dai et al. 2020; Tai et al. 2018). Additionally, they play a role in regulating tumour growth, invasion, metastasis, angiogenesis and drug resistance in tumour cells (Xu et al. 2018; Dai et al. 2020).

The lipid bilayer surface of EVs is enriched in tetraspanins, which are integral membrane proteins that play a crucial role in biogenesis and composition of these vesicles. Particularly, CD9, CD63 and CD81 are major members of the tetraspanin family present in exosomes. Members of the tetraspanin family of proteins have crucial functions in regulating cancer cell migration and interactions between cancer‐ and endothelial cells. These interactions are pivotal for facilitating cancer invasion and metastasis (Detchokul et al. 2014). Furthermore, these tetraspanins are reported to be involved in the uptake of EV in recipient cells and frequently used as EV markers (Jankovičová et al. 2020; Okada‐Tsuchioka et al. 2022). Other important biomarkers of EVs are Syntenin and Alix, these markers were found in high abundance in EVs and are both important for the biogenesis of EVs and the endosomal trafficking (Kugeratski et al. 2021).

Several studies have determined that crosstalk among cells via TEVs facilitates cancer metastasis (McCready et al. 2010; Hayashido et al. 2005; Sakha et al. 2016; Guo et al. 2019). In addition, crucial steps in the metastatic cascade are reportedly affected by TEVs. For instance, TEVs enhance angiogenesis, which not only allows for more blood flow into the tumour, but also provides means for cancer cells to escape the primary site (Mashouri et al. 2019). In addition, TEVs can enhance tumour cell migration (Carvalho et al. 2020). Intriguingly, migrating tumour cells themselves secrete a specific type of EV called migrasomes, which, in turn, promote directional migration (Jiang et al. 2023). These are typically bigger than exosomes and display tetraspanins (TSPANs) 4 and 7 on their surface (Tan et al. 2023).

As the role of TEVs in promoting cancer metastasis becomes increasingly clear, it is essential to explore their molecular content and the implications for tumour biology. In addition to proteins and mRNA, TEVs are known to carry substantial amounts of miRNAs. In recent years there has been an increasing interest in miRNAs derived from TEVs, both from a functional and diagnostic perspective. These miRNAs have been suggested to regulate cellular processes such as proliferation, apoptosis and metabolism in the cells that take up TEVs (Dilsiz 2020). Moreover, the stability of miRNAs especially in TEVs in the circulation makes them attractive as diagnostic or predictive biomarkers in liquid biopsies. Indeed, studies revealed a number of miRNAs in the circulation to be useful in diagnostics (Neerincx et al. 2015). Similarly, our previous work showed specific miRNAs in circulation to be prognostic for outcome and survival and predictive for response to systemic therapy in metastatic CRC (Neerincx et al. 2018; Drees et al. 2024).

Colorectal cancer (CRC) is the second most common cause of cancer‐related death worldwide and TEVs have been found to confer therapy resistance in CRC (Mannavola et al. 2019; Rahmati et al. 2024). Several studies show release of TEVs after drug treatment and thereby infer drug resistance (Chen et al. 2014; Aung et al. 2011; Aubertin et al. 2016). Other studies have shown that TEVs are essential for local and long‐distance communication in cancer cells; they carry molecules which can cause changes in cell behaviour after TEV uptake (Zomer et al. 2015; Li and Nabet 2019). The contents of TEVs may serve as prognostic markers or as a grading basis for assessing cancer progression (Carvalho et al. 2020). However, to better understand the role of TEVs in CRC it is important to understand the heterogeneity of TEVs throughout CRC, especially in the context of the four (gene expression‐based) consensus molecular subtypes (CMSs) (Schlicker et al. 2012; De Sousa et al. 2013; Sadanandam et al. 2013; Marisa et al. 2013; Budinska et al. 2013; Perez‐Villamil et al. 2012; Roepman et al. 2014; Guinney et al. 2015; Song et al. 2016). These subtypes differ in both biological as well as clinical properties and treatment response. CMS1 is distinguished by microsatellite instability (MSI) and a high immune infiltration, CMS2 is often regarded to be the epithelial canonical subtype, whereas CMS3 is a metabolic dysregulated subtype with a mixed MSI status and CMS4 cancers represent a poor‐prognosis mesenchymal subtype that are typified by a high stromal gene signature (Guinney et al. 2015). Both Linnekamp and coworkers (Linnekamp et al. 2018) and Sveen and coworkers (Sveen et al. 2018) identified the presence of CMS subtypes in a large panel of CRC cell lines and showed a clear difference in the sensitivity to chemotherapy and selective drugs, specifically between the epithelial and mesenchymal CMS2 and CMS4 (Brennan et al. 2020).

In this study, we have examined the TEV profiles of the subtypes of CRC. Proteomic and transcriptomic comparison of CMS2 versus CMS4 TEVs revealed subtype‐specific profiles, which relate back to gene expression differences observed between these distinct CMSs. Additionally, TEVs of different subtypes differ in size with CMS1/CMS4‐derived TEVs exhibiting heterogeneity and TEVs with larger diameters. miRNA expression profiles of TEVs from plasma of CRC patients revealed subtype‐specific signatures, indicating potential utility as diagnostic and prognostic tools.

2. Material and Methods

2.1. Resources Table

Reagent or resource	Source	Identifier
Antibodies
Monoclonal anti‐CD9	Merck	Cat. #SAB470
Purified Mouse anti‐human CD63	BD Biosciences	Cat. #556019; RRID: AB_396297
CD81	BD Biosciences	Cat. #555675; RRID: AB_396028
TSPAN4	ThermoFisher	PA5‐69344; RRID: AB_2688603
TSPAN8	ThermoFisher	MA5‐24179; RRID: AB_2609273
Purified mouse IgG1, k isotype control	BD Biosciences	Cat. #555746; RRID: AB_396088
Rabbit bridging anti‐mouse antibody	DAKO	Cat. #Z0259
Bacterial and virus strains
CD63 NanoLuciferase plasmid	Generated as described in bioRxihttps://doi.org/10.1101/2023.02.23.529257v1	N/A
Biological samples
RNA samples patients	Jeroen Bosch ziekenhuis and UMC Utrecht	N/A
Plasma samples patients	Jeroen Bosch ziekenhuis and UMC Utrecht	N/A
Chemicals, peptides and recombinant proteins
DMEM/F12	ThermoFisher	Cat. #31330038
Pen/Strep	Gibco	Cat. #15140122
FCS	Serana	Cat. #s‐FBS‐SA‐025
Culture‐Inserts 3 Well self‐insertion	Ibidi	Cat. #80369
Lipofectamine 2000	ThermoFisher	Cat. #11668027
Cross‐linked Sepharose 2B beads	GE Healthcare	Cat. #GE17‐0140‐01
BSA	Sigma–Aldrich	Cat. #A4503‐50G; CAS9048‐46‐8
PBS	Gibco	Cat. #18912‐014
Glycine	Merck	Cat. #K27662101
Colloidal gold particles	Utrecht University	N/A
Uranyl acetate	EMS	Cat. #22400
QIAzol Lysis Reagent	QIAgen	Cat. #79306
Centricon Plus‐70 ultrafiltration units	Milipore	Cat. #UFC710008
qEVoriginal/70 nm gen2	Izon Science Limited	Cat. #ICO‐70
Amicon Ultra‐2 Centrifugal Filter Unit	Milipore	Cat. #UFC201024
LDS sample buffer	ThermoFisher	Cat. #NP0007
DTT	Sigma–Aldrich	Cat. #d9163
IAA	Merck	Cat. #1149
Urea	Sigma–Aldrich	Cat. #51456
Tris‐buffer	Merck	Cat. #1.08382.0500
Eppendorf concentrator plus	Eppendorf	Cat. #EP5305000509
Nanosep 30k Omega centrifugal devices	Pall Life Sciences	Cat #DO010C34 / Cat #OD010C33
Oasis columns (10 mg)	Waters	Cat. #186000383
ReproSil C18 aqua	Dr. Maisch	Cat. #R119 aq
Trypsin	Promega	Cat. #V5111
Fused silica emitter	New objective	N/A
Critical commercial assays
Nano‐Glo(R) Luciferase Assay	Promega	N1110
miRNeasy serum/plasma kit	QIAgen	Cat. #217184
TaqManMicroRNA Reverse Transcription kit	ThermoFisher	Cat. #4366596
SMARTer Stranded Total RNA‐Seq Kit v3 – Pico Input Mammalian	Takara Bio USA	Cat. #634488
NucleoMag NGS Clean‐up and Size Select	Takara Bio USA	Cat. #744970.50
Deposited data
mRNA and miRNA seq data	GEO Accession viewer (nih.gov)	GEO: GSE280385
Proteomics data	ProteomeXchange	Data are available via ProteomeXchange with identifier PXD057149
Experimental models: Cell lines
HuTu 80	ATCC	CVCL_1301
CaR‐1	ATCC	CVCL_1116
HT55	ATCC	CVCL_1294
MDST8	ATCC	CVCL_2588
SW1116	ATCC	CVCL_0544
RCM‐1	ATCC	CVCL_1648
Software and algorithms
CMS classifier	Guinney et al. (Guinney et al. 2015)	GitHub—Sage‐Bionetworks/CMSclassifier
R2 bioinformatics platform	AMC	R2: Genomics Analysis and Visualization Platform (amc.nl)
ImageJ	Schneider et al. (Schneider et al. 2012)	https://imagej.nih.gov/ij/
Prism 9.5.1	Graphpad	www.graphpad.com
R 4.3.2	The R project for Statistical Computing	https://www.r‐project.org/
MaxQuant 1.6.10.43	N/A	https://www.maxquant.org/
Other
Plate spectrophotometer	BioTek Instruments	Biotek HT
Transmission electron microscope	ThermoFisher	FEI Technai T12
PowerPlex 16 system	Promega	Cat. #DC6531
U3000 RSLC high pressure nanoLC	Dionex	N/A
Q Exactive mass spectrometer	ThermoFisher	N/A
EVOS FL Cell Imaging System	ThermoFisher	N/A

2.2. Resource Availability

2.2.1. Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Jan Paul Medema (j.p.medema@amsterdamumc.nl).

2.2.2. Materials Availability

All unique/stable reagents generated in this study are available from the lead contact with a completed materials transfer agreement.

2.2.3. Data and Code Availability

RNA‐Seq data have been deposited at Gene Expression Omnibus (GEO) and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. No new software was developed as part of the research presented in this paper. The analyses were conducted using existing, publicly available software and tools, which are cited appropriately in the key resources table. For reproducibility and transparency, all the R scripts used in this study are available upon request and without restriction to the lead contact. Any additional information required to reanalyse the data reported in this paper is available from the lead contact upon request.

2.2.4. Patient Samples

Plasma samples of 16 patients with either a CMS2 or CMS4 subtyped tumour were used and derived from Jeroen Bosch Hospital in the Netherlands. All patients provided informed consent after approval from the institutional board at the Jeroen Bosch hospital. For validation, a prospectively collected biobank was used, consisting of frozen tumour tissue from liver metastases obtained from resection specimens and corresponding plasma samples obtained before and after surgery between 2012 and 2015 in the UMC Utrecht. The protocol was approved by the medical ethical committee of the UMC Utrecht (METC M09‐145, 2012: Collection of blood and tissue samples from patients subjected to liver surgery for liver malignancies). RNA sequencing and single sample prediction was used to classify the samples in the different CMS and the properly predicted (>0.5) CMS2 and CMS4 samples were used for further analysis.

2.2.5. Cell Culture

Six Sanger CRC lines: CaR‐1, HuTu 80, MDST‐8, RCM‐1, HT‐55 and SW‐1116 cells were cultured in DMEM/F12 (Thermo Fisher Scientific) supplemented with 10% v/v foetal bovine serum (FBS) (Serana) and 50 units/mL of penicillin/streptomycin (Gibco). These cells were obtained from Sanger Institute (Cambridge, UK). Cells were maintained at 37 °C in humidified air containing 5% CO2. All cell lines were routinely checked for mycoplasma contamination. Moreover, cell line authentication was performed every 2 months by Short Tandem Repeat (STR) genotyping using the PowerPlex 16 system (Promega) according to the manufacturer's protocol. Results were analysed using GeneScan software (Applied Biosystems).

2.2.6. Migration Assay

CMS2 and CMS4 cell lines were added to the outer wells of an Ibidi culture insert in a 24 well plate and incubated for 36 h at 37 °C in humidified air containing 5% CO2. After 36 h, the culture insert was removed carefully to avoid disturbing the cellular monolayer. Fresh medium (DMEM/F12, supplemented with FBS and penicillin/streptomycin) was added and the plates were imaged with the EVOS FL Cell Imaging system (Thermo Scientific) at timepoint 0 and 72 h.

2.2.7. Ultrafiltration (UF)—Size Exclusion Chromatography (SEC)

2.2.7.1. TEV Isolation

TEVs were isolated via ultrafiltration combined with SEC. Cells were cultured to 70%–80% confluency in normal culture medium, after which they were deprived of FBS for 24 h. Supernatant containing the TEVs was collected and processed for sequential centrifugation: 2 × 10 min at 500 × g, and 2 × 15 min at 2000 × g.

2.2.7.2. Ultrafiltration (UF)

After isolation, supernatant was loaded on centricon plus‐70 ultrafiltration units (Merck‐Milipore) and concentrated to 1.5 mL by repeated centrifugation at 3500 × g. To collect the exomes, the filter was inverted and centrifuged at 1000 × g for 2 min until 1.5 mL was eluted, resulting in 1.5 mL of concentrated medium.

2.2.7.3. Size Exclusion Chromatography (SEC)

This concentrated medium was then purified by SEC using Sepharose, separating the medium based on the molecular size. Cross‐linked Sepharose 2B beads (Sepharose CL‐2B) (GE Healthcare) were washed three times with filtered Phosphate‐buffered Saline (PBS) (Gibco). For filtration nylon stockings (20 denier panty, Hema) were cut and used inside of the tip of a 10 mL plastic syringe and the syringe was filled with 10 mL washed Sepharose CL‐2B. The concentrated medium was loaded on the column and collected in 24 fractions of 500 µL. Exosome fractions were determined by a nanoluciferase assay.

2.2.7.4. Nanoluciferase Assay

HT‐55, RCM‐1, SW‐1116 (CMS2), CaR‐1, HuTu 80 and MDST‐8 (CMS4) cell lines were transduced with HA‐NanoLuciferase (NL)‐tagged CD63 and selected with blasticidin. CD63 transduced cell lines were fractionated by UF‐SEC as described above and fractions were analysed using the NanaGlo luciferase assay (Promega). In short, fractions were collected and centrifuged at 500 × g. In total, 50 µL supernatant of each fraction was transferred to a new plate and 25 µL diluted furimazine was added (dilution furimazine 1:1000 in buffer). Furimazine is converted to furimamide which produces high intensity, glow‐type luminescence and was measured on a SynergyTM HT multi‐detection microplate reader (Agilent). The intensity of luciferase luminescence is correlated with the levels of luciferase activity measured directly in the cell culture medium or cells.

2.2.8. Electron Microscopy

To visualize TEVs, EM, the most common EV imaging method, was used. TEVs were isolated, UF‐SEC was performed as described in the previous methods section and fractions 9 and 10 were pooled. To perform EM, formvar‐carbon‐coated grids were used. A 10 µL sample was placed on a sheet of parafilm. The grids were incubated with the sample on the carbon/formvar side for 7 min. Grids were washed with water by placing the grid in water and afterward incubated with 3.5% UA for 7 min. After incubation, the excess UA was discarded with filter paper and thereafter left to dry. Stained grids were visualized with electron microscopy operated at 120 kV (FEI, Thermo Fisher Scientific) using a Veleta 2000 × 2000 side‐mounted CCD camera and Imaging Solutions software (Radius, EMSIS, Germany). Captured images were analysed and quantified using ImageJ by calculating the area in nm² (ImageJ v1.50i, National institute of Health).

2.2.9. Immunogold Labelling

To visualize the classical exosome markers CD9, CD63 and CD81 and later also TSPAN4 and TSPAN8 on TEVs, we performed immunogold labelling. First, TEVs were isolated and purified by UF‐SEC as described above and fractions 9 and 10 were pooled. Formvar‐carbon‐coated grids were incubated with 10 µL purified TEVs for 7 min. Grids were washed with PBS (Gibco)/glycine (Merck) by placing the grid on a droplet of PBS/Glycine to another (5 × 2 min) followed by a blocking step with 1% Bovine serum albumin (BSA) (Sigma–Aldrich) in PBS/Glycine for 3 min. After blocking, the grids were incubated with monoclonal antibodies diluted in 1% PBS/BSA against anti‐CD9, clone MEM‐61 (1:100, Merck), purified mouse anti‐human CD63 clone H5C6 (1:100, BD Biosciences), purified mouse anti‐human CD81, clone JS‐81 (1:100, BD Biosciences) or purified mouse IgG1, k isotype control, clone MOPC‐21 (1:1000, BD Biosciences) for 45 min. The grids were washed again in PBS/Glycine (5 × 2 min) and incubated with the rabbit bridging anti‐mouse antibody (1:200 DAKO) diluted in 1% BSA/PBS for 3 min. Grids were blocked with 0.1% BSA in PBS/glycine and incubated with 10 µL protein A conjugated with colloidal gold particles (1:25, Utrecht University) for 20 min and washed in PBS (6 × 3 min). The grids were incubated with 1% glutaraldehyde, diluted in PBS for 5 min. The grids were washed in PBS and incubated for 5 min with 3.5% UA (EMS) for the negative staining. After incubation, the excess uranyl acetate was discarded with filter paper and thereafter left to dry. Stained grids were visualized with electron microscopy operated at 120 kV (Thermo Fisher Scientific) using a Veleta 2000 × 2000 side‐mounted CCD camera and Imaging Solutions software (Radius, EMSIS). Captured images were analysed and quantified using ImageJ by calculating the area in nm² and the number of immunogold labels were counted (ImageJ v1.50i, National institute of Health).

2.2.10. Sample Processing and Nano‐LC‐MS/MS Measurement and Database Searching

TEVs from the CMS2 and CMS4 cell lines were processed using Filter Aided Sample preparation (FASP), using Pall Nanosep 30k Omega centrifugal devices (Pall Life Sciences). Purified TEV samples were concentrated to about 30 µL on the filters, lysed in LDS‐sample buffer (Thermo Fisher Scientific) (with 10% dithiothreitol (DTT)) (Sigma–Aldrich), and incubated for 2 min at RT. Subsequently 200 µL Urea (8.88 M) was added, and the samples were washed twice for at least 25 min at 10,000 × g. One hundred microliters of iodoacetamide (IAA) (Merck) (0.05 M in urea) was added and incubated for 1 min at RT. Samples were centrifuged for 25 min at 10,000 × g and proteins were washed twice in 100 µL Tris‐buffer (Merck). Subsequently, trypsin (Promega) was added in 40 µL Tris‐buffer and proteins were incubated O/N at RT. Peptides were collected after centrifuging 10,000 × g for 10 min, and subsequent washing of the filter by 50 µL Tris‐buffer. Extracted peptides were concentrated in a vacuum centrifuge (Eppendorf) and concentrated to 50 µL. Samples were desalted using OASIS columns (10 mg, Waters). Cleaned up‐peptides (5 µL) were separated on a 75 µm × g42 cm custom packed Reprosil C18 aqua column (1.9 µm, 120 Å, Dr. Maisch) in 90 min. gradient (2%–32% Acetonitrile + 0.5% Acetic acid at 300 nL/min) using a U3000 RSLC high pressure nanoLC (Dionex). Eluted peptides were measured on‐line by a Q Exactive mass spectrometer (Thermo Fisher Scientific) operating in data‐dependent acquisition mode. Peptides were ionized using a fused silica emitter (New Objective) with a distal high voltage of +2 kV. Intact peptide ions were detected at a resolution of 35,000 (at m/z 200) and fragment ions at a resolution of 17,500 (at m/z 200); the MS mass range was 350–1400 Da. AGC Target settings for MS were 3E6 charges and for MS/MS 2E5 charges. Peptides were selected for higher‐energy dissociation fragmentation at an underfill ratio of 1% and a quadrupole isolation window of 1.5 Da, peptides were fragmented at a normalized collision energy of 25. Raw files from MS analysis were processed using the MaxQuant (2.0.3.0). MS/MS spectra were searched against the Swissprot human database (download jan. 2021, canonical and isoforms; 42383 entries) with a precursor tolerance of 4.5 ppm and an MS/MS tolerance of 20 ppm. Peptides with a minimum of seven amino‐acid lengths were considered with both the peptide and protein false discovery rate (FDR) set to 1%. Enzyme specificity was set to trypsin and up to two missed cleavage sites were allowed. Cysteine carbamidomethylation (Cys) was searched as a fixed modification, whereas N‐acetylation of proteins and oxidized methionine (Met) were searched as variable modifications (default MaxQuant settings). Proteins were quantified label‐free by spectral counting. Normalization was performed by dividing the spectral counts of each protein by the total spectral counts of all proteins within a sample. This number was multiplied with a constant equal to the average of total spectral counts of all samples. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (Perez‐Riverol et al. 2022; Deutsch et al. 2023) partner repository with the dataset identifier PXD057149.

2.2.11. EV Isolation of Cell Lines and Patients for miRNA and mRNA Sequencing

For TEVs isolated from the cell lines, UF was performed prior to SEC. For both cell lines and patients, SEC with the Automated Fraction Collector (AFC‐V1, Izon) was performed to isolate vesicles from 2 to 3 mL of plasma of each patient or isolate vesicles from cell lines, using 2 qEV original/70 nm gen 2 columns (IZON) per sample. Using a buffer volume of 2.70 mL, and a fraction volume of 0.5 mL, the particles of interest were collected in fractions 3 and 4 (between 1 and 2 mL of the purified collection volume [PCV]). Both fractions of the two isolations were pooled, resulting in a volume of 2 mL per plasma. One millilitre of the vesicle isolate was concentrated with an Amicon Ultra‐2 Centrifugal Filter Unit (Millipore) to 250 µL.

2.2.12. RNA Isolation and Quality Control

Total RNA from the EVs was isolated using the miRNeasy serum/plasma kit (QIAgen) according to the manufacturers' protocol with minor adjustments. One thousand two hundred and fifty microliters of QIAzol Lysis Reagent (Qiagen) was added to each sample after the concentration of the EVs, incubated at room temperature for 10–20 min and stored at −80°C until isolation. To quantify the RNA from EVs, a QC‐PCR was performed according to this STAR protocol (van Eijndhoven et al. 2023).

2.2.13. Small RNA Library Preparation and miRNA Sequencing

Small RNA libraries were prepared using the costum IsoSeek protocol described in the STAR protocol (van Eijndhoven et al. 2023). Custom designed 5'‐ and 3'‐5N‐adapters are used to prevent ligation and amplification bias. For the patient samples, the gel size selection was done by Bluepipping (according to the procedures of the manufacturer, using a 3% agarose gel‐cassette and a size range of 140–165 base pairs. Equimolar libraries (1.5 nM each) were pooled and sequenced on a Nova‐Seq6000 platform (PE50, with the addition of 5% PhiX) (van Eijndhoven et al. 2023). Data processing and UMI correction was done using sRNAbench (Aparicio‐Puerta et al. 2022).

2.2.14. mRNA Sequencing

mRNA libraries were prepared using the SMARTer Stranded Total RNA‐Seq Kit v3 – Pico Input Mammalian (Takara Bio Inc.) according to the manufacturers' protocol, including ribosomal cDNA depletion. Eight microliters of RNA was used as input, followed a fragmentation step of 3 min at 94°C. The samples were subjected to five rounds of amplification in the initial PCR, the final RNA‐seq library amplification PCR consisted of 16 PCR cycles. Sequencing was performed on a NovaSeq6000 platform, PE150. Adapter sequences were removed using cutadapt (Kechin et al. 2017) after which sequence reads were mapped to the human genome (GRCh38) with STAR 2.7.10a (Dobin et al. 2013) using GENCODE v43. Gene counts were determined during the STAR mapping procedure with the “–quantMode GeneCounts” setting. Gene counts were normalised to log2 transformed TPM values.

2.2.15. Validation Plasma Set

For the validation set, TEVs were isolated with SEC using CL/2B qEV columns (IZON Science). After applying the samples onto the column, fractions of 0.5 mL were collected. Fractions 9 and 10 are considered as vesicle‐enriched fractions. Subsequently, total RNA was isolated using TRIzol (Thermo Fisher Scientific) according to the manufacturer's instructions, with some modifications. 0.75 mL TRIzol was added to 0.25 mL SEC fractions, mixed properly, and incubated at room temperature for 15 min. Samples were stored at –80°C for at least 3 h to increase the RNA yield from vesicle fractions specifically. Prior to isopropyl precipitation 50 µg glycogen (Roche) was added. The final RNA pellet was dissolved in 10 µL nuclease‐free water. Subsequently, small RNA libraries were prepared using the Illumina TruSeq small RNA Preparation Kit according to the manufacturer's instructions. RNAseq was performed on a HiSeq4000 (Illumina, paired‐end 150).

2.2.16. Data Integration

miRNA data for the cell lines and patient samples were integrated by selecting miRNAs with >12 detected reads in the cell lines (n = 6) and >32 detected reads in the set of patient samples (n = 16). Differential expression between CMS2 and CMS4 samples was analysed using the edgeR R package (quasi‐likelihood method). The design included terms for the CMS subtype and the sample origin (cell line or patient).

2.2.17. Classification of Patient Samples

For the validation of the CMS2/CMS4 specific expression of TEV miRNAs in plasma all miRNAs that were expressed in both sets (defined as >32 reads in the training set [n = 16] and >28 reads in the validation set [n = 14]) were defined, resulting in 281 shared expressed miRNAs. Using the edgeR R package we determined the miRNA log fold changes (CMS4 vs. CMS2) in the training set. We created a log fold change vector or classifier of the 75 most differential expressed miRNAs and calculated the Pearson correlation coefficients of this classifier with all the samples in the validation set, using the mean centred (gene wise) expression.

3. Results

3.1. CRC TEVs of Different Subtypes Differ in Size

To identify TEVs from distinct CRC subtypes, cell lines were used that were previously CMS classified (Linnekamp et al. 2018; Sveen et al. 2018). Through the application of ultrafiltration and size exclusion chromatography (UF‐SEC), vesicles were separated from cellular debris and soluble protein (Figure 1A). To confirm the validity of this approach, two independent CRC cell lines were transduced with a construct expressing the fusion protein HA‐Nanoluciferase (NL)‐CD63. As CD63 is a marker for exosomes, this fusion protein provides a useful biomarker to follow the enrichment of TEVs and was shown to be strongest in elution fractions 8–11 of the size exclusion column (Figure 1B) (Andreu and Yáñez‐Mó 2014).

FIGURE 1.

Subtype differences in TEV composition. (A) Illustrated overview of vesicle separation through ultracentrifugation and size exclusion chromatography (UF‐SEC). Vesicles were separated from cellular debris and soluble protein. (B) NanoLuc activity measurements on supernatant from HT‐55 (CMS2 in blue) and CaR‐1 (CMS4 in green) cells endogenously expressing HA‐NL‐CD63 after UF‐SEC. Representative graph from three independent experiments. Data is presented as mean ± SD. NanoLuciferase signal was measured with a SynergyTM HT multi‐detection microplate reader. (C) Electron microscopy images of the combined fractions 9–10 (UF‐SEC), CMS1 (RKO, magnification 30000x) in orange, CMS2 (HT‐55, RCM‐1, SW1116, magnification 9300x) in blue, CMS3 (LS180, magnification 30000x and LS513, magnification 49000x) in pink and CMS4 (CaR‐1, HuTu 80 andMDST‐8, magnification 9300x) in green, scale bar: 500 nm. (D) Boxplot showing the number of TEVs per image (n = 5) per cell type. (E) Ridgeline plot showing the distribution of the different sizes of the TEVs per cell line based on exosome diameters.

The elution fractions 8–11 were pooled per cell line from a total of nine different CRC cell lines representing all CMS subtypes and analysed plus quantified using electron microscopy. Strikingly, this analysis revealed strong cell line and CMS differences in both quantity and size of TEVs (Figure 1C–E).

Comparison between CMS2, 3 and CMS1, 4 cell lines indicated a clear difference in the size distribution of the TEVs (Figure 1E). Where CMS2 and CMS3 lines secreted TEVs with a relatively uniform distribution (majority < 100 nm), CMS1 and CMS4 cell lines showed a more heterogeneous population with both large and small TEVs (Figure 1E). Quantification of the number of vesicles indicated that on average CMS2 cell lines secreted the highest level of TEVs followed by CMS4 cell lines, while CMS1 and CMS3 cell lines secreted clearly lower amounts of TEVs (Figure 1D). Importantly, CMS2 cells secreted even higher levels of (NL)‐CD63 when cultured as spheroids, confirming that this subtype is secreting high levels of TEVs also under 3D conditions (Figure S1A,B).

3.2. Distinct Tetraspanin Expression Profiles and Cargo Composition Define CMS2 and CMS4 TEV Landscapes

Previous data suggest that CRC cell lines classified using the CMS stratification display more undifferentiated/mesenchymal features in CMS1 and CMS4, while CMS2 and CMS3 represent the more epithelial subtypes (Linnekamp et al. 2018; Sveen et al. 2018). Interestingly, the TEV heterogeneity observed in the CRC cell lines followed a similar mesenchymal/epithelial patterning, with the epithelial cells showing smaller TEVs and the mesenchymal lines showing both small and large TEVs. To understand the nature of this difference, a more detailed analysis was conducted on the two most extreme subtypes CMS2 and CMS4. First, the epithelial versus mesenchymal nature of these cell lines was validated. A direct comparison of the morphology of the six cell lines used revealed epithelial features in the CMS2 types and more elongated mesenchymal features in the CMS4 typed cells (Figure S2A). Moreover, as expected CMS4 typed cells displayed greater migration potential than CMS2 cells (Figure S2B), while comparison of gene expression using the Broad Hallmark EMT gene signature showed a significant enrichment of the epithelial genes in the CMS2 cell lines and the EMT genes in the CMS4 cell lines (Figure S2C).

Next, an in‐depth proteomic analysis was conducted on the TEVs isolated from three cell lines per subtype. Tetraspanins, such as CD9, CD63, and CD81, have emerged as critical components of TEVs, orchestrating the biogenesis, cargo sorting, and fusion processes that underlie exosomal function (Berditchevski and Odintsova 2007; Khushman et al. 2017). These tetraspanin proteins, characterized by their four transmembrane domains, facilitate stability and regulation of vesicle fusion with target cells (Andreu and Yáñez‐Mó 2014). In order to validate that our isolates contain bona fide TEVs, the expression levels of established exosome biomarkers, including tetraspanins CD9 (Théry et al. 1999), CD63 (Escola et al. 1998), and CD81 (Escola et al. 1998), CD82 (van der Pol et al. 2012), Heat Shock Protein (HSP90) (van der Pol et al. 2012), tumour susceptibility gene 101 (TSG101) (van der Pol et al. 2012), Alix (PDCD6IP) (van der Pol et al. 2012), Flotillin (FLOT1) (van der Pol et al. 2012), syntenin‐1 (SDCBP) (Kugeratski et al. 2021) and Annexin A1 (ANXA1) (Pessolano et al. 2019) were assessed (Figure 2A). Vesicles from all cell lines showed protein expression of the expected exosome markers, which validated the isolation procedure. Next to these general exosomal proteins, a highly variable protein composition of the TEVs was detected, with 106 proteins significantly higher expressed in CMS2 TEVs and 45 proteins significantly higher in CMS4 TEVs (Tables S1 and S2, p < 0.05, FDR corrected). Figure 2B shows the differences in protein expression between the two subtypes with the 10 most significantly expressed proteins highlighted in both subtypes. Next, proteins enriched per subtype (p < 0.05, FDR corrected) were used to construct a protein interaction network (Szklarczyk et al. 2023). The CMS2 network exhibited an epithelial core featuring EpCam, Cdh17 and several classical keratins (Figure 2C). In contrast, the CMS4 interactome displayed a more mesenchymal nature characterized by the presence of collagens, LOXL2 and FLNC (Figure 2D). These findings underscore the heterogeneity in TEV cargo between CMS2 and CMS4 CRC subtypes.

FIGURE 2.

Proteomic composition of CMS2 and CMS4 cell line TEVs. (A) Heatmap of proteomic expression levels of classical exosome biomarkers in cell line TEVs, with CMS2 cells in blue and CMS4 cells in green. (B) Volcano plot with differential expression of subtype specific proteins resulting from proteomic analysis of CMS2 and CMS4 cell line EVs in red. In purple, the differentially expressed TSPANs of CMS2 and CMS4 cell line EVs. (C) Known and predicted protein–protein interactions of the most significantly (p < 0.05, FDR corrected) differentially expressed proteins of CMS2 cell line TEVs. (D) Known and predicted protein‐protein interactions of the most significant (p < 0.05, FDR corrected) differentially expressed proteins of CMS4 cell line TEVs. (E) Boxplot of gene expression of the top 10 most significant (p < 0.05, FDR corrected) differentially expressed genes of CMS2 cell line TEVs from the in‐depth proteomic analysis compared to the mRNA expression of different CMS2 cell lines. The black dots represent the lines used for proteomic analysis. (F) Boxplot of gene expression level of the top 10 most significant (p < 0.05, FDR corrected) differentially expressed genes of CMS4 cell line TEVs from the in‐depth proteomic analysis compared to the mRNA expression of different CMS4 cell lines. The black dots represent the lines used for proteomic analysis.

The protein profiles detected in TEVs were also selectively enriched with molecules located in membrane, such as tetraspanin‐enriched microdomains (Figure 2B) (Zöller 2009; Raimondo et al. 2011). To gain a more comprehensive understanding of the TEV content and to determine whether the cargo is a reflection of the parental cell lines, a comparative analysis was conducted between the level of mRNA in the cells of a large set of CMS2 and CMS4‐typed CRC cell lines and the proteins detected in the TEVs. For the majority of the 10 most up‐regulated CMS2 and CMS4 proteins, the TEV content reflected mRNA patterns across CMS2/4 cell lines (Figure 2E,F). This was even more evident when focusing on the mRNA levels of these genes in the six cell lines used in this study (cell lines represented by dots in Figure 2E,F). This implies that the TEV protein cargo was in part dictated by subtype specific expression patterns of the cells.

3.3. Transcriptomic Analysis Revealed Distinct mRNA Cargo in Vesicles

Vesicle mRNAs are important mediators of intercellular communication that can exert their function in recipient cells as they can be translated into functional proteins upon internalization (O'Brien et al. 2020). As distinct subtype‐specific proteomic patterns were evident in TEVs of CMS2 and CMS4 CRC cell lines, in a next step the mRNA contents of two CMS2 and two CMS4 cell line TEVs were analysed. Similar to the protein expression profiles, a clear and consistent subtype‐specific exosomal mRNA expression pattern was detected (Figure 3A,B), pointing to subtype specific cargo also at the mRNA level.

FIGURE 3.

mRNA composition of CMS2 and CMS4 cell line TEVs. (A) Volcano plot of the differential mRNA expression of TEVs derived from CMS4 and CMS2 cell lines. (B) Heatmap showing the 10 most significantly up‐regulated and 10 most down‐regulated genes (CMS4 vs. CMS2) in our cell line panel. (C) Boxplot of the top 10 most significant (p < 0.05, FDR corrected) differentially expressed genes based on mRNA analysis in TEVs derived from CMS2 cell lines, compared to the mRNA expression of different CMS2 cell lines. (D) Boxplot of the top 10 most significant p < 0.05, FDR corrected) differentially expressed genes based on mRNA analysis in TEVs derived from CMS4 cell lines, compared to the mRNA expression of different CMS4 cell lines.

Usually, mRNAs transported by TEVs mirror those expressed in the originating cell, implying representativeness of their cellular source (Prieto‐Vila et al. 2021). Although this direct link was evident for some of the top 10 subtype specific mRNAs, several of these top differentially expressed mRNAs in TEVs were not mirrored by the parental cell mRNA patterns (Figures 3C,D). For instance, while KRT20 and MSN showed clear differential mRNA expression in cells and TEVs, other genes like MAGIX and MRPL33 only displayed differential expression in the TEVs. These observations underscore that TEV mRNA cargo is not solely a passive reflection of the cellular transcriptome but may be shaped by selective loading mechanisms. This selective enrichment suggests a level of regulatory control that could influence vesicle function and intercellular signalling. To further explore the molecular features that distinguish TEVs across CMS subtypes, we next examined subtype‐specific differences in vesicle‐associated membrane proteins.

3.4. Differential Expression of TSPAN4 and TSPAN8 Distinguishes CMS4 and CMS3 TEVs Independent of Vesicle Size

Among the membrane‐associated proteins, members of the TSPAN family are of particular interest due to their roles in vesicle biogenesis, cargo sorting, and uptake. We therefore assessed the expression profiles of TSPAN family members in TEVs derived from CMS2 and CMS4 cell lines, with a focus on identifying subtype‐specific patterns that may reflect functional divergence. Although protein profiles showed differential expression of various proteins with distinct properties, the most significantly, differentially expressed proteins were TSPAN8 and TSPAN4 as seen in Figure 2B. As they are of particular significance in the context of TEVs, an investigation was conducted into the differences of expression of these and other TSPAN family members in CMS2 and CMS4 TEVs.TSPAN1 and TSPAN15 were also found to be significantly higher expressed in CMS2 TEVs. However, the log fold change observed for TSPAN8 (logFC −5.13) was much more pronounced than those for TSPAN1 (logFC −2.45) and TSPAN15 (logFC −1.81) (Figure 2B). The analysis further showed that TSPAN4 was the only observed TSPAN to be selectively expressed in CMS4 TEVs. Importantly, TSPAN7, which is often co‐expressed with TSPAN4 on migrasomes, was not detected, suggesting these vesicles were not migrasomes but TSPAN4 expressing TEVs. As shown in Figure 1E, CMS4 cell lines secreted both large (≥100 nm in diameter) and small (24–100 nm in diameter) TEVs. In contrast, CMS2 cells primarily produced small vesicles (Figure S3A). To compare the sizes of the TEVs from each subtype, the median sizes for the three CMS2 and CMS4 cell lines were calculated. Most vesicles from CMS2, as observed by EM, had median diameters of 55, 60 and 56 nm for HT‐55, RCM‐1 and SW1116, respectively (Figure S3B). In contrast, the small TEVs from the three CMS4 cell lines were on average slightly larger, with median diameters of 70, 64 and 75 nm for CaR‐1, HuTu 80 and MDST‐8, respectively. Importantly, approximately 60% of the vesicles from the CMS4 cell lines fell into the large TEV category. These larger vesicles had median diameters of 144, 151 and 162 nm for CaR‐1, HuTu 80 and MDST‐8, respectively (Figure S3C).

To ascertain that both vesicle size categories represent EVs, an immunogold analysis was performed on the two subtypes. First, an analysis was performed on the mRNA expression of CD9, CD63 and CD81 in CMS2 and CMS4 cell lines, using publicly available gene expression datasets on R2, a genomics analysis and visualization platform. Interestingly, no distinguishable differences on mRNA level in the cell lines was detected between the two subtypes for CD9, CD63 and CD81 (Figure S3D). To determine whether this homogeneous expression pattern is translated into a similar representation of these markers on the secreted vesicles, TEVs from CMS2 and CMS4 cells were labelled with immunogold using antibodies against these three marker tetraspanins (Figure 4A). In line with the relatively uniform mRNA expression profiles and the lack of significant differential protein expression for these tetraspanins (Figure S2D), no clear differences in immunogold labelling of CD9, CD63 and CD81 was detected between the two subtypes on the vesicles either. As CMS4 had both small and large vesicles, our analysis further explored whether size correlated with variations in marker expression. Importantly, differences in expression of these markers were also not evident across the distinct CMS4 vesicle sizes (Figure 4B). CD9, CD63 and CD81 were all heterogeneously expressed on both small and large vesicles derived from CMS4, indicating that both vesicle types indeed represented TEVs.

FIGURE 4.

Tetraspanin expression on TEVs. (A) Immunogold labeling of CD9, CD63 and CD81 tetraspanins on HT‐55 (CMS2) and CaR‐1 (CMS4), visualized using Uranyl Acetate (UA) staining and EM, scale bar: 200 nm. (B) Barplots showing the differences in percentage of positive CD9, CD63 and CD81 CMS4 cell lines in small and large TEVs. A cross‐comparison between markers and small and large TEVs showed no significance. (C) Percentage of immunogold labeled TSPAN4 and TSPAN8 positive TEVs. TSPAN4 is significantly higher in CMS4 TEVs (t‐test, p < 0.001) and TSPAN8 is significantly higher in CMS2 TEVs (t‐test, p < 0.001). (D) Histogram showing small and large TSPAN4 and TSPAN8 positive CMS4 cell line TEVs (CaR‐1) (not significant). (E) Immunogold labelling of TSPAN4 on CaR‐1, visualized using UA staining and EM, scale bar: 200 nm. (F). Immunogold labelling of TSPAN8 on HT‐55 visualized using UA staining and EM, scale bar: 200 nm.

Next, the differential TSPAN4 and TSPAN8 was explored. To determine whether this differential expression was related to the size of TEVs, CMS2 and CMS4 cell line TEVs were labelled with immunogold using antibodies against TSPAN4 and TSPAN8. In line with the differential protein expression TSPAN4 was solely detected with this technique on CMS4 TEVs (Figure 4C,E). Vice versa, TSPAN8 expression was much more pronounced on CMS2 TEVs (Figure 4C,F). Moreover, even though TSPAN8 was only detected on a small subset of CMS4 TEVs, this included both small and large vesicles (Figure 4D). Similarly, TSPAN4 was detected on both small and large TEVs in CMS4 (Figure 4D,E). This also confirmed that TSPAN4 expression on TEVs secreted by CMS4 cell lines was not related to migrasomes, which are much larger vesicles, but was likely an effect of TSPAN4 expression in the cells from which the TEVs were derived.

3.5. miRNA Expression in TEVs and Patient Plasma Samples

EVs contain significant amounts of miRNAs, which regulate key cellular processes and are stable in circulation, making them valuable diagnostic and predictive biomarkers. Therefore, TEVs from CMS2 and CMS4 cell lines were isolated and differential miRNA expression was determined. This further confirmed the strongly divergent nature of the TEVs derived from these two subtypes with significantly different miRNAs in the CMS2 and CMS4 cell line‐derived TEVs (Figure 5A). Importantly, as for the protein expression, the differential miRNA expression was relatively high and strongly significant. To explore whether this subtype‐specific miRNA content could be used for diagnostic purposes, blood plasma from patients with CRC was used to isolate vesicles. Importantly, the subtype of each patient's tumour was determined using bulk RNA sequencing on the primary tumour material and a series of CMS2 and CMS4‐typed patients were selected for analysis (Figure 5B). Subsequently, the miRNA content of these plasma‐derived vesicles was sequenced and compared with the cell line miRNA expression data and used to identify miRNAs differentially expressed between CMS2 and CMS4 samples (Figure 5C). Intriguingly, a clear separation into two subclusters was apparent in which all cell lines and all but two patient samples clustered within their respective CMS2 or CMS4 subtypes (Figure 5C), suggesting vesicles from plasma can be used to identify the patient CMS subtype.

FIGURE 5.

miRNA differences between CMS2 and CMS4 cell lines and patients. (A) Volcano plot of the differential miRNA expression of CMS4 versus CMS2 cell line TEVs. (B) Volcano plot of the differential miRNA expression of CMS4 versus CMS2 patient plasma samples. (C) Heatmap showing miRNAs differentially expressed (p < 0.1, FDR corrected) between CMS4 and CMS2 cell lines and patient samples. (D) Boxplots showing the classifier correlation scores for the distinct patient samples grouped per CMS2 and CMS4 (p = 0.012). p value was calculated using the Wilcoxon statistical test.

To directly provide evidence for the use of vesicle‐derived miRNAs in defining CMS2 and CMS4 subtypes, a separate set of plasma samples from CRC patients was used in order to allow for a training and validation set‐up using the two independent patient sets. First, a classifier was built with the top differentially expressed miRNA between CMS2 and CMS4 plasma samples in the training set. This resulted in a classifier composed of 75 miRNAs, which was then used to classify the independent validation set of 14 samples (9× CMS2, 5× CMS4). Importantly, this revealed a significant difference in classifier correlation scores between CMS2 and CMS4 samples (p = 0.012) and thus a validation of the differential expression of the miRNAs between CMS2 and CMS4 samples (Figure 5D).

This indicated that TEVs derived from patient plasma exhibited distinct subtype‐specific miRNA expression patterns, which can be used as subtype biomarkers for the primary tumour.

4. Discussion

TEVs have emerged as intricate messengers involved in the complex communication between cancer cells and their surroundings. Our investigation into the extracellular vesicle landscape in CRC reveals significant differences between the epithelial and mesenchymal subtypes of CRC. Prior research has established the significance of TEVs in cancer progression, metastasis, and therapy resistance (Dai et al. 2020; Tai et al. 2018; Chang et al. 2021; Wandrey et al. 2023). Although previous studies have explored EV cargo and its implications in CRC, none have delved into the differences between CRC subtypes at the vesicle level. Our findings build upon this by specifically examining the distinct TEV profiles of all distinct CMS types and a detailed comparison between epithelial (CMS2) and mesenchymal (CMS4) subtypes of CRC.

Although all four CMS subtypes were used to identify subtype‐specific patterns, we further focused on CMS2 and CMS4 due to their distinct biological features and suitability for in vitro modelling. CMS1 tumours, driven by MSI and immune‐related features, follow a distinct oncogenic trajectory that may strongly impact tumour‐intrinsic TEV characteristics. CMS3, while molecularly distinct, is underrepresented in cell line models and closely resembles CMS2 in vitro, limiting its added value for comparative analysis. Moreover, our data also showed that CMS1 and CMS3 cell lines secreted a more limited amount of vesicles. Focusing on CMS2 and CMS4 allowed us to capture the most robust and biologically meaningful differences in TEV composition.

The characterization of TEV landscapes sheds light on the striking differences between CMS2 and CMS4 subtypes. Our findings highlight substantial differences in cargo composition, tetraspanin expression and vesicle morphology between these subtypes. This variance suggested distinct mechanisms of TEV biogenesis, cargo sorting and cellular uptake. Whether this contributes to the diverse biological and clinical behaviour observed for CMS2 and CMS4 CRC is not known, but appears likely due to the strongly divergent TEV content available for intercellular communication. These differences are interesting from a biological perspective and our miRNA results show that they can also be used as a diagnostic tool to identify the subtype of cancer using liquid biopsies.

The differences in TEV protein cargo between CMS2 and CMS4 correlated strongly with the gene expression differences observed in the parental cell lines from which the TEVs were derived. TEV content does not always mirror the protein composition of the parental cells (Anand et al. 2019), but it is commonly observed that mRNAs transported by TEVs largely mirror those expressed in the originating cell (Prieto‐Vila et al. 2021).

Our proteomic analysis delineated differences in the nature of TEV cargo. Importantly, the differentially expressed proteins present in CMS2 TEVs exhibited an epithelial core, contrasting with the mesenchymal nature of the CMS4‐enriched TEV proteins. These distinct TEV protein signatures can potentially serve as biomarkers for differentiating between CMS2 and CMS4 (Mashouri et al. 2019; Samanta et al. 2018). Particularly, the differential loading between subtypes observed in our proteomic, mRNA, and miRNA data presents a compelling avenue for the development of diagnostic tools.

Not only differences in mRNA loading are notable, but also the observation that CMS2 cells predominantly present small TEVs, whereas the mesenchymal CMS4 cells exhibit both small and large vesicles. Further research is needed to understand the consequences for tumour biology. Even more intriguing is that the subtype specific proteins, TSPAN4 and TSPAN8, are independent of these size differences. TSPAN4 can be found on both small and larger TEVs in the mesenchymal CRC subtype. This points to the fact that TSPAN4 expression on CMS4‐derived TEVs is not related to the generation of migrasomes by these cells. In agreement, the lack of TSPAN7 expression, the presence of classical EV markers combined with the size of the vesicles secreted by CMS4 cell lines, indicated that these vesicles are not migrasomes, but small and large TEVs. In agreement, the EM data showed that CMS4 lines secrete two distinct size ranges of TEVs, which both express comparable levels of CD9, CD63 and CD81. Interestingly, as TSPAN4 was detected on both small and large TEVs, this could provide for a promising candidate for subtype specific TEV detection.

Through comprehensive analysis of TEVs from cell lines and patient plasma, our findings not only validate the subtype specificity observed in proteomics and transcriptomics, but also underscore the potential utility of exosome‐derived miRNAs as relevant biomarkers in CRC clinical decision making and prognosis. In a training and validation set‐up CMS2/4 specific miRNA signatures were defined and validated. Although these analyses were performed on relatively small sets, they do provide valuable insights. Our results open up several avenues for future research. Notably, CMS4 cancers have been reported to be resistant to cetuximab in the context of folfox therapy (De Sousa et al. 2013; Ten Hoorn et al. 2021), which emphasizes the importance of determining the subtype of CRC, as it can influence the choice of therapy. Our method offers a straightforward approach to identifying CRC subtypes via circulating TEVs, eliminating the need for primary tumour material which may not always be available. Although tissue samples are typically available for initial diagnosis, TEV analysis may provide valuable insights into differences in metastases, which could influence treatment decisions. This development provides a valuable diagnostic tool for patients, ensuring that even when primary tumour characteristics differ from metastatic sites, effective and personalized treatment plans can still be formulated based on the exosomal profile.

Investigating the specific roles of subtype‐enriched proteins and miRNAs in CRC progression and metastasis could provide mechanistic insights into TEV‐mediated signalling pathways underlying subtype‐specific phenotypes. Functional studies utilizing in vitro and in vivo models of CRC could elucidate the impact of subtype‐specific TEV cargo on tumour growth, invasion, and therapy resistance. Additionally, exploring the potential interactions between CRC cells and stromal cells mediated by subtype‐specific TEVs could uncover novel therapeutic strategies for modulating the tumour microenvironment and overcoming treatment resistance. Furthermore, longitudinal studies assessing changes in exosomal profiles during disease progression and in response to therapy could inform the development of dynamic biomarkers for monitoring treatment efficacy and predicting patient outcomes. Overall, continued research in this area has the potential to transform our understanding of CRC heterogeneity and improve patient stratification and personalized therapeutic approaches.

Author Contributions

Conceptualization, P.J.A., D.M.P. and J.P.M.; data interpretation, analysis, generation, and/or collection, P.J.A., L.H.B., J.d.R., N.J.G., J.P.B., O.K., I.H.M.B.R, O.K., C.G.M., M.v.E, H.F.M.P., A.G., N.v.d.W., S.R.v.H., A.T., T.E.B., D.M.P., J.P.M.; supervision of the project, J.P.M., D.M.P.; writing of the paper, P.J.A., L.H.B., J.P.M.

Conflicts of Interest

D.M.P. holds equity in Y2Y BV and receives funding from Takeda, Amgen, Abbvie and Prediction Biosciences. The other authors declare no conflicts of interest.

GEO Location

This study was conducted in Amsterdam, the Netherlands. The plasma samples were collected from the Jeroen Bosch hospital in Den Bosch, the Netherlands and UMC Utrecht, the Netherlands.

Supporting information

Supporting Fig 1. A: Histogram of the percentage of Nanoluciferase activity from HT‐55 CD63 cultured in 2D and 3D conditions. Representative graph from three independent experiments. Data are presented as mean ± S.D. NanoLuciferase signal was measured with a SynergyTM HT multi‐detection microplate reader. B. EVOS images of HT‐55 CD63 cultured in 2D and 3D.

Supporting Fig 2. A: EVOS images of CMS2 (HT‐55, RCM‐1 and SW1116) and CMS4 (CaR‐1, HuTu 80 and MDST‐8) cell lines, scale‐bar: 100 µm. C. Ibidi images of CMS2 (HT‐55) and CMS4 (CaR‐1), scale‐bar: 100 µm. D. Heatmap of the differential RNA expression comparison between CMS2 and CMS4 cell lines using the Broad Hallmark EMT gene signature.

Supporting Fig 3. A:Barplots showing percentage of small (24‐100 nm) and large TEVs (≥ 100 nm) in CMS2 and CMS4 cell lines, (n = 3 cell lines per subtype, t‐test, p < 0.001). B. Violin graph representing TEV sizes of each measured TEV per cell line of CMS2 cells. C. Violin graph representing TEV sizes of each measured TEV per cell line of CMS4 cells. D. Boxplots of CD9 (p <0.001), CD63 (p = 0.205) and CD81 (p = 0.987) tetraspanin expression on CMS2 and CMS4 cell lines (t‐test).

Supporting Table 1 Top differential expressed proteins in CMS2 vs. CMS4.

Supporting Table 2 Top differential expressed proteins in CMS4 vs. CMS2.

Asif, P. J. , Borghuis L. H., van Hooff S. R., et al. 2025. “Mesenchymal Colorectal Cancers Secrete Vesicles With Unique Cargo That Can Be Used for Liquid Biopsy Based Diagnostics.” Journal of Extracellular Vesicles 14, no. 11: e70171. 10.1002/jev2.70171

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. Proteomics data are available on ProteomeXchange (http://www.ebi.ac.uk/pride), project accession: PXD057149 For token, please contact the corresponding author Sequencing data (miRNA and mRNA) is available on GEO. To review GEO accession GSE280385: Go to https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?&acc=GSE280385 For token, please contact the corresponding author.